fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/compat.h>
  34 #include <linux/bit_spinlock.h>
  35 #include <linux/xattr.h>
  36 #include <linux/posix_acl.h>
  37 #include <linux/falloc.h>
  38 #include <linux/slab.h>
  39 #include <linux/ratelimit.h>
  40 #include <linux/mount.h>
  41 #include <linux/btrfs.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/posix_acl_xattr.h>
  44 #include <linux/uio.h>
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "print-tree.h"
  50 #include "ordered-data.h"
  51 #include "xattr.h"
  52 #include "tree-log.h"
  53 #include "volumes.h"
  54 #include "compression.h"
  55 #include "locking.h"
  56 #include "free-space-cache.h"
  57 #include "inode-map.h"
  58 #include "backref.h"
  59 #include "hash.h"
  60 #include "props.h"
  61 #include "qgroup.h"
  62 #include "dedupe.h"
  63
  64 struct btrfs_iget_args {
  65         struct btrfs_key *location;
  66         struct btrfs_root *root;
  67 };
  68
  69 struct btrfs_dio_data {
  70         u64 outstanding_extents;
  71         u64 reserve;
  72         u64 unsubmitted_oe_range_start;
  73         u64 unsubmitted_oe_range_end;
  74         int overwrite;
  75 };
  76
  77 static const struct inode_operations btrfs_dir_inode_operations;
  78 static const struct inode_operations btrfs_symlink_inode_operations;
  79 static const struct inode_operations btrfs_dir_ro_inode_operations;
  80 static const struct inode_operations btrfs_special_inode_operations;
  81 static const struct inode_operations btrfs_file_inode_operations;
  82 static const struct address_space_operations btrfs_aops;
  83 static const struct address_space_operations btrfs_symlink_aops;
  84 static const struct file_operations btrfs_dir_file_operations;
  85 static const struct extent_io_ops btrfs_extent_io_ops;
  86
  87 static struct kmem_cache *btrfs_inode_cachep;
  88 struct kmem_cache *btrfs_trans_handle_cachep;
  89 struct kmem_cache *btrfs_path_cachep;
  90 struct kmem_cache *btrfs_free_space_cachep;
  91
  92 #define S_SHIFT 12
  93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  94         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  95         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  96         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  97         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  98         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  99         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 100         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 101 };
 102
 103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
 104 static int btrfs_truncate(struct inode *inode);
 105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 106 static noinline int cow_file_range(struct inode *inode,
 107                                    struct page *locked_page,
 108                                    u64 start, u64 end, u64 delalloc_end,
 109                                    int *page_started, unsigned long *nr_written,
 110                                    int unlock, struct btrfs_dedupe_hash *hash);
 111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
 112                                        u64 orig_start, u64 block_start,
 113                                        u64 block_len, u64 orig_block_len,
 114                                        u64 ram_bytes, int compress_type,
 115                                        int type);
 116
 117 static void __endio_write_update_ordered(struct inode *inode,
 118                                          const u64 offset, const u64 bytes,
 119                                          const bool uptodate);
 120
 121 /*
 122  * Cleanup all submitted ordered extents in specified range to handle errors
 123  * from the fill_dellaloc() callback.
 124  *
 125  * NOTE: caller must ensure that when an error happens, it can not call
 126  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
 127  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
 128  * to be released, which we want to happen only when finishing the ordered
 129  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
 130  * fill_delalloc() callback already does proper cleanup for the first page of
 131  * the range, that is, it invokes the callback writepage_end_io_hook() for the
 132  * range of the first page.
 133  */
 134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
 135                                                  const u64 offset,
 136                                                  const u64 bytes)
 137 {
 138         return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
 139                                             bytes - PAGE_SIZE, false);
 140 }
 141
 142 static int btrfs_dirty_inode(struct inode *inode);
 143
 144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 145 void btrfs_test_inode_set_ops(struct inode *inode)
 146 {
 147         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 148 }
 149 #endif
 150
 151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 152                                      struct inode *inode,  struct inode *dir,
 153                                      const struct qstr *qstr)
 154 {
 155         int err;
 156
 157         err = btrfs_init_acl(trans, inode, dir);
 158         if (!err)
 159                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 160         return err;
 161 }
 162
 163 /*
 164  * this does all the hard work for inserting an inline extent into
 165  * the btree.  The caller should have done a btrfs_drop_extents so that
 166  * no overlapping inline items exist in the btree
 167  */
 168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 169                                 struct btrfs_path *path, int extent_inserted,
 170                                 struct btrfs_root *root, struct inode *inode,
 171                                 u64 start, size_t size, size_t compressed_size,
 172                                 int compress_type,
 173                                 struct page **compressed_pages)
 174 {
 175         struct extent_buffer *leaf;
 176         struct page *page = NULL;
 177         char *kaddr;
 178         unsigned long ptr;
 179         struct btrfs_file_extent_item *ei;
 180         int ret;
 181         size_t cur_size = size;
 182         unsigned long offset;
 183
 184         if (compressed_size && compressed_pages)
 185                 cur_size = compressed_size;
 186
 187         inode_add_bytes(inode, size);
 188
 189         if (!extent_inserted) {
 190                 struct btrfs_key key;
 191                 size_t datasize;
 192
 193                 key.objectid = btrfs_ino(BTRFS_I(inode));
 194                 key.offset = start;
 195                 key.type = BTRFS_EXTENT_DATA_KEY;
 196
 197                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 198                 path->leave_spinning = 1;
 199                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 200                                               datasize);
 201                 if (ret)
 202                         goto fail;
 203         }
 204         leaf = path->nodes[0];
 205         ei = btrfs_item_ptr(leaf, path->slots[0],
 206                             struct btrfs_file_extent_item);
 207         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 208         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 209         btrfs_set_file_extent_encryption(leaf, ei, 0);
 210         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 211         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 212         ptr = btrfs_file_extent_inline_start(ei);
 213
 214         if (compress_type != BTRFS_COMPRESS_NONE) {
 215                 struct page *cpage;
 216                 int i = 0;
 217                 while (compressed_size > 0) {
 218                         cpage = compressed_pages[i];
 219                         cur_size = min_t(unsigned long, compressed_size,
 220                                        PAGE_SIZE);
 221
 222                         kaddr = kmap_atomic(cpage);
 223                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 224                         kunmap_atomic(kaddr);
 225
 226                         i++;
 227                         ptr += cur_size;
 228                         compressed_size -= cur_size;
 229                 }
 230                 btrfs_set_file_extent_compression(leaf, ei,
 231                                                   compress_type);
 232         } else {
 233                 page = find_get_page(inode->i_mapping,
 234                                      start >> PAGE_SHIFT);
 235                 btrfs_set_file_extent_compression(leaf, ei, 0);
 236                 kaddr = kmap_atomic(page);
 237                 offset = start & (PAGE_SIZE - 1);
 238                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 239                 kunmap_atomic(kaddr);
 240                 put_page(page);
 241         }
 242         btrfs_mark_buffer_dirty(leaf);
 243         btrfs_release_path(path);
 244
 245         /*
 246          * we're an inline extent, so nobody can
 247          * extend the file past i_size without locking
 248          * a page we already have locked.
 249          *
 250          * We must do any isize and inode updates
 251          * before we unlock the pages.  Otherwise we
 252          * could end up racing with unlink.
 253          */
 254         BTRFS_I(inode)->disk_i_size = inode->i_size;
 255         ret = btrfs_update_inode(trans, root, inode);
 256
 257 fail:
 258         return ret;
 259 }
 260
 261
 262 /*
 263  * conditionally insert an inline extent into the file.  This
 264  * does the checks required to make sure the data is small enough
 265  * to fit as an inline extent.
 266  */
 267 static noinline int cow_file_range_inline(struct btrfs_root *root,
 268                                           struct inode *inode, u64 start,
 269                                           u64 end, size_t compressed_size,
 270                                           int compress_type,
 271                                           struct page **compressed_pages)
 272 {
 273         struct btrfs_fs_info *fs_info = root->fs_info;
 274         struct btrfs_trans_handle *trans;
 275         u64 isize = i_size_read(inode);
 276         u64 actual_end = min(end + 1, isize);
 277         u64 inline_len = actual_end - start;
 278         u64 aligned_end = ALIGN(end, fs_info->sectorsize);
 279         u64 data_len = inline_len;
 280         int ret;
 281         struct btrfs_path *path;
 282         int extent_inserted = 0;
 283         u32 extent_item_size;
 284
 285         if (compressed_size)
 286                 data_len = compressed_size;
 287
 288         if (start > 0 ||
 289             actual_end > fs_info->sectorsize ||
 290             data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
 291             (!compressed_size &&
 292             (actual_end & (fs_info->sectorsize - 1)) == 0) ||
 293             end + 1 < isize ||
 294             data_len > fs_info->max_inline) {
 295                 return 1;
 296         }
 297
 298         path = btrfs_alloc_path();
 299         if (!path)
 300                 return -ENOMEM;
 301
 302         trans = btrfs_join_transaction(root);
 303         if (IS_ERR(trans)) {
 304                 btrfs_free_path(path);
 305                 return PTR_ERR(trans);
 306         }
 307         trans->block_rsv = &fs_info->delalloc_block_rsv;
 308
 309         if (compressed_size && compressed_pages)
 310                 extent_item_size = btrfs_file_extent_calc_inline_size(
 311                    compressed_size);
 312         else
 313                 extent_item_size = btrfs_file_extent_calc_inline_size(
 314                     inline_len);
 315
 316         ret = __btrfs_drop_extents(trans, root, inode, path,
 317                                    start, aligned_end, NULL,
 318                                    1, 1, extent_item_size, &extent_inserted);
 319         if (ret) {
 320                 btrfs_abort_transaction(trans, ret);
 321                 goto out;
 322         }
 323
 324         if (isize > actual_end)
 325                 inline_len = min_t(u64, isize, actual_end);
 326         ret = insert_inline_extent(trans, path, extent_inserted,
 327                                    root, inode, start,
 328                                    inline_len, compressed_size,
 329                                    compress_type, compressed_pages);
 330         if (ret && ret != -ENOSPC) {
 331                 btrfs_abort_transaction(trans, ret);
 332                 goto out;
 333         } else if (ret == -ENOSPC) {
 334                 ret = 1;
 335                 goto out;
 336         }
 337
 338         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 339         btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
 340         btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
 341 out:
 342         /*
 343          * Don't forget to free the reserved space, as for inlined extent
 344          * it won't count as data extent, free them directly here.
 345          * And at reserve time, it's always aligned to page size, so
 346          * just free one page here.
 347          */
 348         btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
 349         btrfs_free_path(path);
 350         btrfs_end_transaction(trans);
 351         return ret;
 352 }
 353
 354 struct async_extent {
 355         u64 start;
 356         u64 ram_size;
 357         u64 compressed_size;
 358         struct page **pages;
 359         unsigned long nr_pages;
 360         int compress_type;
 361         struct list_head list;
 362 };
 363
 364 struct async_cow {
 365         struct inode *inode;
 366         struct btrfs_root *root;
 367         struct page *locked_page;
 368         u64 start;
 369         u64 end;
 370         struct list_head extents;
 371         struct btrfs_work work;
 372 };
 373
 374 static noinline int add_async_extent(struct async_cow *cow,
 375                                      u64 start, u64 ram_size,
 376                                      u64 compressed_size,
 377                                      struct page **pages,
 378                                      unsigned long nr_pages,
 379                                      int compress_type)
 380 {
 381         struct async_extent *async_extent;
 382
 383         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 384         BUG_ON(!async_extent); /* -ENOMEM */
 385         async_extent->start = start;
 386         async_extent->ram_size = ram_size;
 387         async_extent->compressed_size = compressed_size;
 388         async_extent->pages = pages;
 389         async_extent->nr_pages = nr_pages;
 390         async_extent->compress_type = compress_type;
 391         list_add_tail(&async_extent->list, &cow->extents);
 392         return 0;
 393 }
 394
 395 static inline int inode_need_compress(struct inode *inode)
 396 {
 397         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 398
 399         /* force compress */
 400         if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
 401                 return 1;
 402         /* bad compression ratios */
 403         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 404                 return 0;
 405         if (btrfs_test_opt(fs_info, COMPRESS) ||
 406             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 407             BTRFS_I(inode)->force_compress)
 408                 return 1;
 409         return 0;
 410 }
 411
 412 static inline void inode_should_defrag(struct btrfs_inode *inode,
 413                 u64 start, u64 end, u64 num_bytes, u64 small_write)
 414 {
 415         /* If this is a small write inside eof, kick off a defrag */
 416         if (num_bytes < small_write &&
 417             (start > 0 || end + 1 < inode->disk_i_size))
 418                 btrfs_add_inode_defrag(NULL, inode);
 419 }
 420
 421 /*
 422  * we create compressed extents in two phases.  The first
 423  * phase compresses a range of pages that have already been
 424  * locked (both pages and state bits are locked).
 425  *
 426  * This is done inside an ordered work queue, and the compression
 427  * is spread across many cpus.  The actual IO submission is step
 428  * two, and the ordered work queue takes care of making sure that
 429  * happens in the same order things were put onto the queue by
 430  * writepages and friends.
 431  *
 432  * If this code finds it can't get good compression, it puts an
 433  * entry onto the work queue to write the uncompressed bytes.  This
 434  * makes sure that both compressed inodes and uncompressed inodes
 435  * are written in the same order that the flusher thread sent them
 436  * down.
 437  */
 438 static noinline void compress_file_range(struct inode *inode,
 439                                         struct page *locked_page,
 440                                         u64 start, u64 end,
 441                                         struct async_cow *async_cow,
 442                                         int *num_added)
 443 {
 444         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 445         struct btrfs_root *root = BTRFS_I(inode)->root;
 446         u64 num_bytes;
 447         u64 blocksize = fs_info->sectorsize;
 448         u64 actual_end;
 449         u64 isize = i_size_read(inode);
 450         int ret = 0;
 451         struct page **pages = NULL;
 452         unsigned long nr_pages;
 453         unsigned long total_compressed = 0;
 454         unsigned long total_in = 0;
 455         int i;
 456         int will_compress;
 457         int compress_type = fs_info->compress_type;
 458         int redirty = 0;
 459
 460         inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
 461                         SZ_16K);
 462
 463         actual_end = min_t(u64, isize, end + 1);
 464 again:
 465         will_compress = 0;
 466         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 467         BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
 468         nr_pages = min_t(unsigned long, nr_pages,
 469                         BTRFS_MAX_COMPRESSED / PAGE_SIZE);
 470
 471         /*
 472          * we don't want to send crud past the end of i_size through
 473          * compression, that's just a waste of CPU time.  So, if the
 474          * end of the file is before the start of our current
 475          * requested range of bytes, we bail out to the uncompressed
 476          * cleanup code that can deal with all of this.
 477          *
 478          * It isn't really the fastest way to fix things, but this is a
 479          * very uncommon corner.
 480          */
 481         if (actual_end <= start)
 482                 goto cleanup_and_bail_uncompressed;
 483
 484         total_compressed = actual_end - start;
 485
 486         /*
 487          * skip compression for a small file range(<=blocksize) that
 488          * isn't an inline extent, since it doesn't save disk space at all.
 489          */
 490         if (total_compressed <= blocksize &&
 491            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 492                 goto cleanup_and_bail_uncompressed;
 493
 494         total_compressed = min_t(unsigned long, total_compressed,
 495                         BTRFS_MAX_UNCOMPRESSED);
 496         num_bytes = ALIGN(end - start + 1, blocksize);
 497         num_bytes = max(blocksize,  num_bytes);
 498         total_in = 0;
 499         ret = 0;
 500
 501         /*
 502          * we do compression for mount -o compress and when the
 503          * inode has not been flagged as nocompress.  This flag can
 504          * change at any time if we discover bad compression ratios.
 505          */
 506         if (inode_need_compress(inode)) {
 507                 WARN_ON(pages);
 508                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 509                 if (!pages) {
 510                         /* just bail out to the uncompressed code */
 511                         goto cont;
 512                 }
 513
 514                 if (BTRFS_I(inode)->force_compress)
 515                         compress_type = BTRFS_I(inode)->force_compress;
 516
 517                 /*
 518                  * we need to call clear_page_dirty_for_io on each
 519                  * page in the range.  Otherwise applications with the file
 520                  * mmap'd can wander in and change the page contents while
 521                  * we are compressing them.
 522                  *
 523                  * If the compression fails for any reason, we set the pages
 524                  * dirty again later on.
 525                  */
 526                 extent_range_clear_dirty_for_io(inode, start, end);
 527                 redirty = 1;
 528                 ret = btrfs_compress_pages(compress_type,
 529                                            inode->i_mapping, start,
 530                                            pages,
 531                                            &nr_pages,
 532                                            &total_in,
 533                                            &total_compressed);
 534
 535                 if (!ret) {
 536                         unsigned long offset = total_compressed &
 537                                 (PAGE_SIZE - 1);
 538                         struct page *page = pages[nr_pages - 1];
 539                         char *kaddr;
 540
 541                         /* zero the tail end of the last page, we might be
 542                          * sending it down to disk
 543                          */
 544                         if (offset) {
 545                                 kaddr = kmap_atomic(page);
 546                                 memset(kaddr + offset, 0,
 547                                        PAGE_SIZE - offset);
 548                                 kunmap_atomic(kaddr);
 549                         }
 550                         will_compress = 1;
 551                 }
 552         }
 553 cont:
 554         if (start == 0) {
 555                 /* lets try to make an inline extent */
 556                 if (ret || total_in < (actual_end - start)) {
 557                         /* we didn't compress the entire range, try
 558                          * to make an uncompressed inline extent.
 559                          */
 560                         ret = cow_file_range_inline(root, inode, start, end,
 561                                             0, BTRFS_COMPRESS_NONE, NULL);
 562                 } else {
 563                         /* try making a compressed inline extent */
 564                         ret = cow_file_range_inline(root, inode, start, end,
 565                                                     total_compressed,
 566                                                     compress_type, pages);
 567                 }
 568                 if (ret <= 0) {
 569                         unsigned long clear_flags = EXTENT_DELALLOC |
 570                                 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
 571                         unsigned long page_error_op;
 572
 573                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 574                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 575
 576                         /*
 577                          * inline extent creation worked or returned error,
 578                          * we don't need to create any more async work items.
 579                          * Unlock and free up our temp pages.
 580                          */
 581                         extent_clear_unlock_delalloc(inode, start, end, end,
 582                                                      NULL, clear_flags,
 583                                                      PAGE_UNLOCK |
 584                                                      PAGE_CLEAR_DIRTY |
 585                                                      PAGE_SET_WRITEBACK |
 586                                                      page_error_op |
 587                                                      PAGE_END_WRITEBACK);
 588                         if (ret == 0)
 589                                 btrfs_free_reserved_data_space_noquota(inode,
 590                                                                start,
 591                                                                end - start + 1);
 592                         goto free_pages_out;
 593                 }
 594         }
 595
 596         if (will_compress) {
 597                 /*
 598                  * we aren't doing an inline extent round the compressed size
 599                  * up to a block size boundary so the allocator does sane
 600                  * things
 601                  */
 602                 total_compressed = ALIGN(total_compressed, blocksize);
 603
 604                 /*
 605                  * one last check to make sure the compression is really a
 606                  * win, compare the page count read with the blocks on disk,
 607                  * compression must free at least one sector size
 608                  */
 609                 total_in = ALIGN(total_in, PAGE_SIZE);
 610                 if (total_compressed + blocksize <= total_in) {
 611                         num_bytes = total_in;
 612                         *num_added += 1;
 613
 614                         /*
 615                          * The async work queues will take care of doing actual
 616                          * allocation on disk for these compressed pages, and
 617                          * will submit them to the elevator.
 618                          */
 619                         add_async_extent(async_cow, start, num_bytes,
 620                                         total_compressed, pages, nr_pages,
 621                                         compress_type);
 622
 623                         if (start + num_bytes < end) {
 624                                 start += num_bytes;
 625                                 pages = NULL;
 626                                 cond_resched();
 627                                 goto again;
 628                         }
 629                         return;
 630                 }
 631         }
 632         if (pages) {
 633                 /*
 634                  * the compression code ran but failed to make things smaller,
 635                  * free any pages it allocated and our page pointer array
 636                  */
 637                 for (i = 0; i < nr_pages; i++) {
 638                         WARN_ON(pages[i]->mapping);
 639                         put_page(pages[i]);
 640                 }
 641                 kfree(pages);
 642                 pages = NULL;
 643                 total_compressed = 0;
 644                 nr_pages = 0;
 645
 646                 /* flag the file so we don't compress in the future */
 647                 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
 648                     !(BTRFS_I(inode)->force_compress)) {
 649                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 650                 }
 651         }
 652 cleanup_and_bail_uncompressed:
 653         /*
 654          * No compression, but we still need to write the pages in the file
 655          * we've been given so far.  redirty the locked page if it corresponds
 656          * to our extent and set things up for the async work queue to run
 657          * cow_file_range to do the normal delalloc dance.
 658          */
 659         if (page_offset(locked_page) >= start &&
 660             page_offset(locked_page) <= end)
 661                 __set_page_dirty_nobuffers(locked_page);
 662                 /* unlocked later on in the async handlers */
 663
 664         if (redirty)
 665                 extent_range_redirty_for_io(inode, start, end);
 666         add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
 667                          BTRFS_COMPRESS_NONE);
 668         *num_added += 1;
 669
 670         return;
 671
 672 free_pages_out:
 673         for (i = 0; i < nr_pages; i++) {
 674                 WARN_ON(pages[i]->mapping);
 675                 put_page(pages[i]);
 676         }
 677         kfree(pages);
 678 }
 679
 680 static void free_async_extent_pages(struct async_extent *async_extent)
 681 {
 682         int i;
 683
 684         if (!async_extent->pages)
 685                 return;
 686
 687         for (i = 0; i < async_extent->nr_pages; i++) {
 688                 WARN_ON(async_extent->pages[i]->mapping);
 689                 put_page(async_extent->pages[i]);
 690         }
 691         kfree(async_extent->pages);
 692         async_extent->nr_pages = 0;
 693         async_extent->pages = NULL;
 694 }
 695
 696 /*
 697  * phase two of compressed writeback.  This is the ordered portion
 698  * of the code, which only gets called in the order the work was
 699  * queued.  We walk all the async extents created by compress_file_range
 700  * and send them down to the disk.
 701  */
 702 static noinline void submit_compressed_extents(struct inode *inode,
 703                                               struct async_cow *async_cow)
 704 {
 705         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 706         struct async_extent *async_extent;
 707         u64 alloc_hint = 0;
 708         struct btrfs_key ins;
 709         struct extent_map *em;
 710         struct btrfs_root *root = BTRFS_I(inode)->root;
 711         struct extent_io_tree *io_tree;
 712         int ret = 0;
 713
 714 again:
 715         while (!list_empty(&async_cow->extents)) {
 716                 async_extent = list_entry(async_cow->extents.next,
 717                                           struct async_extent, list);
 718                 list_del(&async_extent->list);
 719
 720                 io_tree = &BTRFS_I(inode)->io_tree;
 721
 722 retry:
 723                 /* did the compression code fall back to uncompressed IO? */
 724                 if (!async_extent->pages) {
 725                         int page_started = 0;
 726                         unsigned long nr_written = 0;
 727
 728                         lock_extent(io_tree, async_extent->start,
 729                                          async_extent->start +
 730                                          async_extent->ram_size - 1);
 731
 732                         /* allocate blocks */
 733                         ret = cow_file_range(inode, async_cow->locked_page,
 734                                              async_extent->start,
 735                                              async_extent->start +
 736                                              async_extent->ram_size - 1,
 737                                              async_extent->start +
 738                                              async_extent->ram_size - 1,
 739                                              &page_started, &nr_written, 0,
 740                                              NULL);
 741
 742                         /* JDM XXX */
 743
 744                         /*
 745                          * if page_started, cow_file_range inserted an
 746                          * inline extent and took care of all the unlocking
 747                          * and IO for us.  Otherwise, we need to submit
 748                          * all those pages down to the drive.
 749                          */
 750                         if (!page_started && !ret)
 751                                 extent_write_locked_range(io_tree,
 752                                                   inode, async_extent->start,
 753                                                   async_extent->start +
 754                                                   async_extent->ram_size - 1,
 755                                                   btrfs_get_extent,
 756                                                   WB_SYNC_ALL);
 757                         else if (ret)
 758                                 unlock_page(async_cow->locked_page);
 759                         kfree(async_extent);
 760                         cond_resched();
 761                         continue;
 762                 }
 763
 764                 lock_extent(io_tree, async_extent->start,
 765                             async_extent->start + async_extent->ram_size - 1);
 766
 767                 ret = btrfs_reserve_extent(root, async_extent->ram_size,
 768                                            async_extent->compressed_size,
 769                                            async_extent->compressed_size,
 770                                            0, alloc_hint, &ins, 1, 1);
 771                 if (ret) {
 772                         free_async_extent_pages(async_extent);
 773
 774                         if (ret == -ENOSPC) {
 775                                 unlock_extent(io_tree, async_extent->start,
 776                                               async_extent->start +
 777                                               async_extent->ram_size - 1);
 778
 779                                 /*
 780                                  * we need to redirty the pages if we decide to
 781                                  * fallback to uncompressed IO, otherwise we
 782                                  * will not submit these pages down to lower
 783                                  * layers.
 784                                  */
 785                                 extent_range_redirty_for_io(inode,
 786                                                 async_extent->start,
 787                                                 async_extent->start +
 788                                                 async_extent->ram_size - 1);
 789
 790                                 goto retry;
 791                         }
 792                         goto out_free;
 793                 }
 794                 /*
 795                  * here we're doing allocation and writeback of the
 796                  * compressed pages
 797                  */
 798                 em = create_io_em(inode, async_extent->start,
 799                                   async_extent->ram_size, /* len */
 800                                   async_extent->start, /* orig_start */
 801                                   ins.objectid, /* block_start */
 802                                   ins.offset, /* block_len */
 803                                   ins.offset, /* orig_block_len */
 804                                   async_extent->ram_size, /* ram_bytes */
 805                                   async_extent->compress_type,
 806                                   BTRFS_ORDERED_COMPRESSED);
 807                 if (IS_ERR(em))
 808                         /* ret value is not necessary due to void function */
 809                         goto out_free_reserve;
 810                 free_extent_map(em);
 811
 812                 ret = btrfs_add_ordered_extent_compress(inode,
 813                                                 async_extent->start,
 814                                                 ins.objectid,
 815                                                 async_extent->ram_size,
 816                                                 ins.offset,
 817                                                 BTRFS_ORDERED_COMPRESSED,
 818                                                 async_extent->compress_type);
 819                 if (ret) {
 820                         btrfs_drop_extent_cache(BTRFS_I(inode),
 821                                                 async_extent->start,
 822                                                 async_extent->start +
 823                                                 async_extent->ram_size - 1, 0);
 824                         goto out_free_reserve;
 825                 }
 826                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 827
 828                 /*
 829                  * clear dirty, set writeback and unlock the pages.
 830                  */
 831                 extent_clear_unlock_delalloc(inode, async_extent->start,
 832                                 async_extent->start +
 833                                 async_extent->ram_size - 1,
 834                                 async_extent->start +
 835                                 async_extent->ram_size - 1,
 836                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 837                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 838                                 PAGE_SET_WRITEBACK);
 839                 if (btrfs_submit_compressed_write(inode,
 840                                     async_extent->start,
 841                                     async_extent->ram_size,
 842                                     ins.objectid,
 843                                     ins.offset, async_extent->pages,
 844                                     async_extent->nr_pages)) {
 845                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 846                         struct page *p = async_extent->pages[0];
 847                         const u64 start = async_extent->start;
 848                         const u64 end = start + async_extent->ram_size - 1;
 849
 850                         p->mapping = inode->i_mapping;
 851                         tree->ops->writepage_end_io_hook(p, start, end,
 852                                                          NULL, 0);
 853                         p->mapping = NULL;
 854                         extent_clear_unlock_delalloc(inode, start, end, end,
 855                                                      NULL, 0,
 856                                                      PAGE_END_WRITEBACK |
 857                                                      PAGE_SET_ERROR);
 858                         free_async_extent_pages(async_extent);
 859                 }
 860                 alloc_hint = ins.objectid + ins.offset;
 861                 kfree(async_extent);
 862                 cond_resched();
 863         }
 864         return;
 865 out_free_reserve:
 866         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 867         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
 868 out_free:
 869         extent_clear_unlock_delalloc(inode, async_extent->start,
 870                                      async_extent->start +
 871                                      async_extent->ram_size - 1,
 872                                      async_extent->start +
 873                                      async_extent->ram_size - 1,
 874                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 875                                      EXTENT_DELALLOC_NEW |
 876                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 877                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 878                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 879                                      PAGE_SET_ERROR);
 880         free_async_extent_pages(async_extent);
 881         kfree(async_extent);
 882         goto again;
 883 }
 884
 885 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 886                                       u64 num_bytes)
 887 {
 888         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 889         struct extent_map *em;
 890         u64 alloc_hint = 0;
 891
 892         read_lock(&em_tree->lock);
 893         em = search_extent_mapping(em_tree, start, num_bytes);
 894         if (em) {
 895                 /*
 896                  * if block start isn't an actual block number then find the
 897                  * first block in this inode and use that as a hint.  If that
 898                  * block is also bogus then just don't worry about it.
 899                  */
 900                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 901                         free_extent_map(em);
 902                         em = search_extent_mapping(em_tree, 0, 0);
 903                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 904                                 alloc_hint = em->block_start;
 905                         if (em)
 906                                 free_extent_map(em);
 907                 } else {
 908                         alloc_hint = em->block_start;
 909                         free_extent_map(em);
 910                 }
 911         }
 912         read_unlock(&em_tree->lock);
 913
 914         return alloc_hint;
 915 }
 916
 917 /*
 918  * when extent_io.c finds a delayed allocation range in the file,
 919  * the call backs end up in this code.  The basic idea is to
 920  * allocate extents on disk for the range, and create ordered data structs
 921  * in ram to track those extents.
 922  *
 923  * locked_page is the page that writepage had locked already.  We use
 924  * it to make sure we don't do extra locks or unlocks.
 925  *
 926  * *page_started is set to one if we unlock locked_page and do everything
 927  * required to start IO on it.  It may be clean and already done with
 928  * IO when we return.
 929  */
 930 static noinline int cow_file_range(struct inode *inode,
 931                                    struct page *locked_page,
 932                                    u64 start, u64 end, u64 delalloc_end,
 933                                    int *page_started, unsigned long *nr_written,
 934                                    int unlock, struct btrfs_dedupe_hash *hash)
 935 {
 936         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 937         struct btrfs_root *root = BTRFS_I(inode)->root;
 938         u64 alloc_hint = 0;
 939         u64 num_bytes;
 940         unsigned long ram_size;
 941         u64 disk_num_bytes;
 942         u64 cur_alloc_size = 0;
 943         u64 blocksize = fs_info->sectorsize;
 944         struct btrfs_key ins;
 945         struct extent_map *em;
 946         unsigned clear_bits;
 947         unsigned long page_ops;
 948         bool extent_reserved = false;
 949         int ret = 0;
 950
 951         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 952                 WARN_ON_ONCE(1);
 953                 ret = -EINVAL;
 954                 goto out_unlock;
 955         }
 956
 957         num_bytes = ALIGN(end - start + 1, blocksize);
 958         num_bytes = max(blocksize,  num_bytes);
 959         disk_num_bytes = num_bytes;
 960
 961         inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
 962
 963         if (start == 0) {
 964                 /* lets try to make an inline extent */
 965                 ret = cow_file_range_inline(root, inode, start, end, 0,
 966                                         BTRFS_COMPRESS_NONE, NULL);
 967                 if (ret == 0) {
 968                         extent_clear_unlock_delalloc(inode, start, end,
 969                                      delalloc_end, NULL,
 970                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 971                                      EXTENT_DELALLOC_NEW |
 972                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 973                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 974                                      PAGE_END_WRITEBACK);
 975                         btrfs_free_reserved_data_space_noquota(inode, start,
 976                                                 end - start + 1);
 977                         *nr_written = *nr_written +
 978                              (end - start + PAGE_SIZE) / PAGE_SIZE;
 979                         *page_started = 1;
 980                         goto out;
 981                 } else if (ret < 0) {
 982                         goto out_unlock;
 983                 }
 984         }
 985
 986         BUG_ON(disk_num_bytes >
 987                btrfs_super_total_bytes(fs_info->super_copy));
 988
 989         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 990         btrfs_drop_extent_cache(BTRFS_I(inode), start,
 991                         start + num_bytes - 1, 0);
 992
 993         while (disk_num_bytes > 0) {
 994                 cur_alloc_size = disk_num_bytes;
 995                 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
 996                                            fs_info->sectorsize, 0, alloc_hint,
 997                                            &ins, 1, 1);
 998                 if (ret < 0)
 999                         goto out_unlock;
1000                 cur_alloc_size = ins.offset;
1001                 extent_reserved = true;
1002
1003                 ram_size = ins.offset;
1004                 em = create_io_em(inode, start, ins.offset, /* len */
1005                                   start, /* orig_start */
1006                                   ins.objectid, /* block_start */
1007                                   ins.offset, /* block_len */
1008                                   ins.offset, /* orig_block_len */
1009                                   ram_size, /* ram_bytes */
1010                                   BTRFS_COMPRESS_NONE, /* compress_type */
1011                                   BTRFS_ORDERED_REGULAR /* type */);
1012                 if (IS_ERR(em))
1013                         goto out_reserve;
1014                 free_extent_map(em);
1015
1016                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1017                                                ram_size, cur_alloc_size, 0);
1018                 if (ret)
1019                         goto out_drop_extent_cache;
1020
1021                 if (root->root_key.objectid ==
1022                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1023                         ret = btrfs_reloc_clone_csums(inode, start,
1024                                                       cur_alloc_size);
1025                         /*
1026                          * Only drop cache here, and process as normal.
1027                          *
1028                          * We must not allow extent_clear_unlock_delalloc()
1029                          * at out_unlock label to free meta of this ordered
1030                          * extent, as its meta should be freed by
1031                          * btrfs_finish_ordered_io().
1032                          *
1033                          * So we must continue until @start is increased to
1034                          * skip current ordered extent.
1035                          */
1036                         if (ret)
1037                                 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1038                                                 start + ram_size - 1, 0);
1039                 }
1040
1041                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1042
1043                 /* we're not doing compressed IO, don't unlock the first
1044                  * page (which the caller expects to stay locked), don't
1045                  * clear any dirty bits and don't set any writeback bits
1046                  *
1047                  * Do set the Private2 bit so we know this page was properly
1048                  * setup for writepage
1049                  */
1050                 page_ops = unlock ? PAGE_UNLOCK : 0;
1051                 page_ops |= PAGE_SET_PRIVATE2;
1052
1053                 extent_clear_unlock_delalloc(inode, start,
1054                                              start + ram_size - 1,
1055                                              delalloc_end, locked_page,
1056                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1057                                              page_ops);
1058                 if (disk_num_bytes < cur_alloc_size)
1059                         disk_num_bytes = 0;
1060                 else
1061                         disk_num_bytes -= cur_alloc_size;
1062                 num_bytes -= cur_alloc_size;
1063                 alloc_hint = ins.objectid + ins.offset;
1064                 start += cur_alloc_size;
1065                 extent_reserved = false;
1066
1067                 /*
1068                  * btrfs_reloc_clone_csums() error, since start is increased
1069                  * extent_clear_unlock_delalloc() at out_unlock label won't
1070                  * free metadata of current ordered extent, we're OK to exit.
1071                  */
1072                 if (ret)
1073                         goto out_unlock;
1074         }
1075 out:
1076         return ret;
1077
1078 out_drop_extent_cache:
1079         btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1080 out_reserve:
1081         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1082         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1083 out_unlock:
1084         clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1085                 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1086         page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1087                 PAGE_END_WRITEBACK;
1088         /*
1089          * If we reserved an extent for our delalloc range (or a subrange) and
1090          * failed to create the respective ordered extent, then it means that
1091          * when we reserved the extent we decremented the extent's size from
1092          * the data space_info's bytes_may_use counter and incremented the
1093          * space_info's bytes_reserved counter by the same amount. We must make
1094          * sure extent_clear_unlock_delalloc() does not try to decrement again
1095          * the data space_info's bytes_may_use counter, therefore we do not pass
1096          * it the flag EXTENT_CLEAR_DATA_RESV.
1097          */
1098         if (extent_reserved) {
1099                 extent_clear_unlock_delalloc(inode, start,
1100                                              start + cur_alloc_size,
1101                                              start + cur_alloc_size,
1102                                              locked_page,
1103                                              clear_bits,
1104                                              page_ops);
1105                 start += cur_alloc_size;
1106                 if (start >= end)
1107                         goto out;
1108         }
1109         extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1110                                      locked_page,
1111                                      clear_bits | EXTENT_CLEAR_DATA_RESV,
1112                                      page_ops);
1113         goto out;
1114 }
1115
1116 /*
1117  * work queue call back to started compression on a file and pages
1118  */
1119 static noinline void async_cow_start(struct btrfs_work *work)
1120 {
1121         struct async_cow *async_cow;
1122         int num_added = 0;
1123         async_cow = container_of(work, struct async_cow, work);
1124
1125         compress_file_range(async_cow->inode, async_cow->locked_page,
1126                             async_cow->start, async_cow->end, async_cow,
1127                             &num_added);
1128         if (num_added == 0) {
1129                 btrfs_add_delayed_iput(async_cow->inode);
1130                 async_cow->inode = NULL;
1131         }
1132 }
1133
1134 /*
1135  * work queue call back to submit previously compressed pages
1136  */
1137 static noinline void async_cow_submit(struct btrfs_work *work)
1138 {
1139         struct btrfs_fs_info *fs_info;
1140         struct async_cow *async_cow;
1141         struct btrfs_root *root;
1142         unsigned long nr_pages;
1143
1144         async_cow = container_of(work, struct async_cow, work);
1145
1146         root = async_cow->root;
1147         fs_info = root->fs_info;
1148         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1149                 PAGE_SHIFT;
1150
1151         /*
1152          * atomic_sub_return implies a barrier for waitqueue_active
1153          */
1154         if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1155             5 * SZ_1M &&
1156             waitqueue_active(&fs_info->async_submit_wait))
1157                 wake_up(&fs_info->async_submit_wait);
1158
1159         if (async_cow->inode)
1160                 submit_compressed_extents(async_cow->inode, async_cow);
1161 }
1162
1163 static noinline void async_cow_free(struct btrfs_work *work)
1164 {
1165         struct async_cow *async_cow;
1166         async_cow = container_of(work, struct async_cow, work);
1167         if (async_cow->inode)
1168                 btrfs_add_delayed_iput(async_cow->inode);
1169         kfree(async_cow);
1170 }
1171
1172 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1173                                 u64 start, u64 end, int *page_started,
1174                                 unsigned long *nr_written)
1175 {
1176         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1177         struct async_cow *async_cow;
1178         struct btrfs_root *root = BTRFS_I(inode)->root;
1179         unsigned long nr_pages;
1180         u64 cur_end;
1181
1182         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1183                          1, 0, NULL, GFP_NOFS);
1184         while (start < end) {
1185                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1186                 BUG_ON(!async_cow); /* -ENOMEM */
1187                 async_cow->inode = igrab(inode);
1188                 async_cow->root = root;
1189                 async_cow->locked_page = locked_page;
1190                 async_cow->start = start;
1191
1192                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1193                     !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1194                         cur_end = end;
1195                 else
1196                         cur_end = min(end, start + SZ_512K - 1);
1197
1198                 async_cow->end = cur_end;
1199                 INIT_LIST_HEAD(&async_cow->extents);
1200
1201                 btrfs_init_work(&async_cow->work,
1202                                 btrfs_delalloc_helper,
1203                                 async_cow_start, async_cow_submit,
1204                                 async_cow_free);
1205
1206                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1207                         PAGE_SHIFT;
1208                 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1209
1210                 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1211
1212                 while (atomic_read(&fs_info->async_submit_draining) &&
1213                        atomic_read(&fs_info->async_delalloc_pages)) {
1214                         wait_event(fs_info->async_submit_wait,
1215                                    (atomic_read(&fs_info->async_delalloc_pages) ==
1216                                     0));
1217                 }
1218
1219                 *nr_written += nr_pages;
1220                 start = cur_end + 1;
1221         }
1222         *page_started = 1;
1223         return 0;
1224 }
1225
1226 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1227                                         u64 bytenr, u64 num_bytes)
1228 {
1229         int ret;
1230         struct btrfs_ordered_sum *sums;
1231         LIST_HEAD(list);
1232
1233         ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1234                                        bytenr + num_bytes - 1, &list, 0);
1235         if (ret == 0 && list_empty(&list))
1236                 return 0;
1237
1238         while (!list_empty(&list)) {
1239                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1240                 list_del(&sums->list);
1241                 kfree(sums);
1242         }
1243         return 1;
1244 }
1245
1246 /*
1247  * when nowcow writeback call back.  This checks for snapshots or COW copies
1248  * of the extents that exist in the file, and COWs the file as required.
1249  *
1250  * If no cow copies or snapshots exist, we write directly to the existing
1251  * blocks on disk
1252  */
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254                                        struct page *locked_page,
1255                               u64 start, u64 end, int *page_started, int force,
1256                               unsigned long *nr_written)
1257 {
1258         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259         struct btrfs_root *root = BTRFS_I(inode)->root;
1260         struct extent_buffer *leaf;
1261         struct btrfs_path *path;
1262         struct btrfs_file_extent_item *fi;
1263         struct btrfs_key found_key;
1264         struct extent_map *em;
1265         u64 cow_start;
1266         u64 cur_offset;
1267         u64 extent_end;
1268         u64 extent_offset;
1269         u64 disk_bytenr;
1270         u64 num_bytes;
1271         u64 disk_num_bytes;
1272         u64 ram_bytes;
1273         int extent_type;
1274         int ret, err;
1275         int type;
1276         int nocow;
1277         int check_prev = 1;
1278         bool nolock;
1279         u64 ino = btrfs_ino(BTRFS_I(inode));
1280
1281         path = btrfs_alloc_path();
1282         if (!path) {
1283                 extent_clear_unlock_delalloc(inode, start, end, end,
1284                                              locked_page,
1285                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1286                                              EXTENT_DO_ACCOUNTING |
1287                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1288                                              PAGE_CLEAR_DIRTY |
1289                                              PAGE_SET_WRITEBACK |
1290                                              PAGE_END_WRITEBACK);
1291                 return -ENOMEM;
1292         }
1293
1294         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1295
1296         cow_start = (u64)-1;
1297         cur_offset = start;
1298         while (1) {
1299                 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1300                                                cur_offset, 0);
1301                 if (ret < 0)
1302                         goto error;
1303                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304                         leaf = path->nodes[0];
1305                         btrfs_item_key_to_cpu(leaf, &found_key,
1306                                               path->slots[0] - 1);
1307                         if (found_key.objectid == ino &&
1308                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1309                                 path->slots[0]--;
1310                 }
1311                 check_prev = 0;
1312 next_slot:
1313                 leaf = path->nodes[0];
1314                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315                         ret = btrfs_next_leaf(root, path);
1316                         if (ret < 0)
1317                                 goto error;
1318                         if (ret > 0)
1319                                 break;
1320                         leaf = path->nodes[0];
1321                 }
1322
1323                 nocow = 0;
1324                 disk_bytenr = 0;
1325                 num_bytes = 0;
1326                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1327
1328                 if (found_key.objectid > ino)
1329                         break;
1330                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1331                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1332                         path->slots[0]++;
1333                         goto next_slot;
1334                 }
1335                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1336                     found_key.offset > end)
1337                         break;
1338
1339                 if (found_key.offset > cur_offset) {
1340                         extent_end = found_key.offset;
1341                         extent_type = 0;
1342                         goto out_check;
1343                 }
1344
1345                 fi = btrfs_item_ptr(leaf, path->slots[0],
1346                                     struct btrfs_file_extent_item);
1347                 extent_type = btrfs_file_extent_type(leaf, fi);
1348
1349                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1350                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1351                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1352                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1353                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1354                         extent_end = found_key.offset +
1355                                 btrfs_file_extent_num_bytes(leaf, fi);
1356                         disk_num_bytes =
1357                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1358                         if (extent_end <= start) {
1359                                 path->slots[0]++;
1360                                 goto next_slot;
1361                         }
1362                         if (disk_bytenr == 0)
1363                                 goto out_check;
1364                         if (btrfs_file_extent_compression(leaf, fi) ||
1365                             btrfs_file_extent_encryption(leaf, fi) ||
1366                             btrfs_file_extent_other_encoding(leaf, fi))
1367                                 goto out_check;
1368                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1369                                 goto out_check;
1370                         if (btrfs_extent_readonly(fs_info, disk_bytenr))
1371                                 goto out_check;
1372                         if (btrfs_cross_ref_exist(root, ino,
1373                                                   found_key.offset -
1374                                                   extent_offset, disk_bytenr))
1375                                 goto out_check;
1376                         disk_bytenr += extent_offset;
1377                         disk_bytenr += cur_offset - found_key.offset;
1378                         num_bytes = min(end + 1, extent_end) - cur_offset;
1379                         /*
1380                          * if there are pending snapshots for this root,
1381                          * we fall into common COW way.
1382                          */
1383                         if (!nolock) {
1384                                 err = btrfs_start_write_no_snapshoting(root);
1385                                 if (!err)
1386                                         goto out_check;
1387                         }
1388                         /*
1389                          * force cow if csum exists in the range.
1390                          * this ensure that csum for a given extent are
1391                          * either valid or do not exist.
1392                          */
1393                         if (csum_exist_in_range(fs_info, disk_bytenr,
1394                                                 num_bytes)) {
1395                                 if (!nolock)
1396                                         btrfs_end_write_no_snapshoting(root);
1397                                 goto out_check;
1398                         }
1399                         if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1400                                 if (!nolock)
1401                                         btrfs_end_write_no_snapshoting(root);
1402                                 goto out_check;
1403                         }
1404                         nocow = 1;
1405                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1406                         extent_end = found_key.offset +
1407                                 btrfs_file_extent_inline_len(leaf,
1408                                                      path->slots[0], fi);
1409                         extent_end = ALIGN(extent_end,
1410                                            fs_info->sectorsize);
1411                 } else {
1412                         BUG_ON(1);
1413                 }
1414 out_check:
1415                 if (extent_end <= start) {
1416                         path->slots[0]++;
1417                         if (!nolock && nocow)
1418                                 btrfs_end_write_no_snapshoting(root);
1419                         if (nocow)
1420                                 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1421                         goto next_slot;
1422                 }
1423                 if (!nocow) {
1424                         if (cow_start == (u64)-1)
1425                                 cow_start = cur_offset;
1426                         cur_offset = extent_end;
1427                         if (cur_offset > end)
1428                                 break;
1429                         path->slots[0]++;
1430                         goto next_slot;
1431                 }
1432
1433                 btrfs_release_path(path);
1434                 if (cow_start != (u64)-1) {
1435                         ret = cow_file_range(inode, locked_page,
1436                                              cow_start, found_key.offset - 1,
1437                                              end, page_started, nr_written, 1,
1438                                              NULL);
1439                         if (ret) {
1440                                 if (!nolock && nocow)
1441                                         btrfs_end_write_no_snapshoting(root);
1442                                 if (nocow)
1443                                         btrfs_dec_nocow_writers(fs_info,
1444                                                                 disk_bytenr);
1445                                 goto error;
1446                         }
1447                         cow_start = (u64)-1;
1448                 }
1449
1450                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1451                         u64 orig_start = found_key.offset - extent_offset;
1452
1453                         em = create_io_em(inode, cur_offset, num_bytes,
1454                                           orig_start,
1455                                           disk_bytenr, /* block_start */
1456                                           num_bytes, /* block_len */
1457                                           disk_num_bytes, /* orig_block_len */
1458                                           ram_bytes, BTRFS_COMPRESS_NONE,
1459                                           BTRFS_ORDERED_PREALLOC);
1460                         if (IS_ERR(em)) {
1461                                 if (!nolock && nocow)
1462                                         btrfs_end_write_no_snapshoting(root);
1463                                 if (nocow)
1464                                         btrfs_dec_nocow_writers(fs_info,
1465                                                                 disk_bytenr);
1466                                 ret = PTR_ERR(em);
1467                                 goto error;
1468                         }
1469                         free_extent_map(em);
1470                 }
1471
1472                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1473                         type = BTRFS_ORDERED_PREALLOC;
1474                 } else {
1475                         type = BTRFS_ORDERED_NOCOW;
1476                 }
1477
1478                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1479                                                num_bytes, num_bytes, type);
1480                 if (nocow)
1481                         btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1482                 BUG_ON(ret); /* -ENOMEM */
1483
1484                 if (root->root_key.objectid ==
1485                     BTRFS_DATA_RELOC_TREE_OBJECTID)
1486                         /*
1487                          * Error handled later, as we must prevent
1488                          * extent_clear_unlock_delalloc() in error handler
1489                          * from freeing metadata of created ordered extent.
1490                          */
1491                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1492                                                       num_bytes);
1493
1494                 extent_clear_unlock_delalloc(inode, cur_offset,
1495                                              cur_offset + num_bytes - 1, end,
1496                                              locked_page, EXTENT_LOCKED |
1497                                              EXTENT_DELALLOC |
1498                                              EXTENT_CLEAR_DATA_RESV,
1499                                              PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1500
1501                 if (!nolock && nocow)
1502                         btrfs_end_write_no_snapshoting(root);
1503                 cur_offset = extent_end;
1504
1505                 /*
1506                  * btrfs_reloc_clone_csums() error, now we're OK to call error
1507                  * handler, as metadata for created ordered extent will only
1508                  * be freed by btrfs_finish_ordered_io().
1509                  */
1510                 if (ret)
1511                         goto error;
1512                 if (cur_offset > end)
1513                         break;
1514         }
1515         btrfs_release_path(path);
1516
1517         if (cur_offset <= end && cow_start == (u64)-1) {
1518                 cow_start = cur_offset;
1519                 cur_offset = end;
1520         }
1521
1522         if (cow_start != (u64)-1) {
1523                 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1524                                      page_started, nr_written, 1, NULL);
1525                 if (ret)
1526                         goto error;
1527         }
1528
1529 error:
1530         if (ret && cur_offset < end)
1531                 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1532                                              locked_page, EXTENT_LOCKED |
1533                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1534                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1535                                              PAGE_CLEAR_DIRTY |
1536                                              PAGE_SET_WRITEBACK |
1537                                              PAGE_END_WRITEBACK);
1538         btrfs_free_path(path);
1539         return ret;
1540 }
1541
1542 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1543 {
1544
1545         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1546             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1547                 return 0;
1548
1549         /*
1550          * @defrag_bytes is a hint value, no spinlock held here,
1551          * if is not zero, it means the file is defragging.
1552          * Force cow if given extent needs to be defragged.
1553          */
1554         if (BTRFS_I(inode)->defrag_bytes &&
1555             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1556                            EXTENT_DEFRAG, 0, NULL))
1557                 return 1;
1558
1559         return 0;
1560 }
1561
1562 /*
1563  * extent_io.c call back to do delayed allocation processing
1564  */
1565 static int run_delalloc_range(void *private_data, struct page *locked_page,
1566                               u64 start, u64 end, int *page_started,
1567                               unsigned long *nr_written)
1568 {
1569         struct inode *inode = private_data;
1570         int ret;
1571         int force_cow = need_force_cow(inode, start, end);
1572
1573         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1574                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1575                                          page_started, 1, nr_written);
1576         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1577                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1578                                          page_started, 0, nr_written);
1579         } else if (!inode_need_compress(inode)) {
1580                 ret = cow_file_range(inode, locked_page, start, end, end,
1581                                       page_started, nr_written, 1, NULL);
1582         } else {
1583                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1584                         &BTRFS_I(inode)->runtime_flags);
1585                 ret = cow_file_range_async(inode, locked_page, start, end,
1586                                            page_started, nr_written);
1587         }
1588         if (ret)
1589                 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1590         return ret;
1591 }
1592
1593 static void btrfs_split_extent_hook(void *private_data,
1594                                     struct extent_state *orig, u64 split)
1595 {
1596         struct inode *inode = private_data;
1597         u64 size;
1598
1599         /* not delalloc, ignore it */
1600         if (!(orig->state & EXTENT_DELALLOC))
1601                 return;
1602
1603         size = orig->end - orig->start + 1;
1604         if (size > BTRFS_MAX_EXTENT_SIZE) {
1605                 u32 num_extents;
1606                 u64 new_size;
1607
1608                 /*
1609                  * See the explanation in btrfs_merge_extent_hook, the same
1610                  * applies here, just in reverse.
1611                  */
1612                 new_size = orig->end - split + 1;
1613                 num_extents = count_max_extents(new_size);
1614                 new_size = split - orig->start;
1615                 num_extents += count_max_extents(new_size);
1616                 if (count_max_extents(size) >= num_extents)
1617                         return;
1618         }
1619
1620         spin_lock(&BTRFS_I(inode)->lock);
1621         BTRFS_I(inode)->outstanding_extents++;
1622         spin_unlock(&BTRFS_I(inode)->lock);
1623 }
1624
1625 /*
1626  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1627  * extents so we can keep track of new extents that are just merged onto old
1628  * extents, such as when we are doing sequential writes, so we can properly
1629  * account for the metadata space we'll need.
1630  */
1631 static void btrfs_merge_extent_hook(void *private_data,
1632                                     struct extent_state *new,
1633                                     struct extent_state *other)
1634 {
1635         struct inode *inode = private_data;
1636         u64 new_size, old_size;
1637         u32 num_extents;
1638
1639         /* not delalloc, ignore it */
1640         if (!(other->state & EXTENT_DELALLOC))
1641                 return;
1642
1643         if (new->start > other->start)
1644                 new_size = new->end - other->start + 1;
1645         else
1646                 new_size = other->end - new->start + 1;
1647
1648         /* we're not bigger than the max, unreserve the space and go */
1649         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1650                 spin_lock(&BTRFS_I(inode)->lock);
1651                 BTRFS_I(inode)->outstanding_extents--;
1652                 spin_unlock(&BTRFS_I(inode)->lock);
1653                 return;
1654         }
1655
1656         /*
1657          * We have to add up either side to figure out how many extents were
1658          * accounted for before we merged into one big extent.  If the number of
1659          * extents we accounted for is <= the amount we need for the new range
1660          * then we can return, otherwise drop.  Think of it like this
1661          *
1662          * [ 4k][MAX_SIZE]
1663          *
1664          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1665          * need 2 outstanding extents, on one side we have 1 and the other side
1666          * we have 1 so they are == and we can return.  But in this case
1667          *
1668          * [MAX_SIZE+4k][MAX_SIZE+4k]
1669          *
1670          * Each range on their own accounts for 2 extents, but merged together
1671          * they are only 3 extents worth of accounting, so we need to drop in
1672          * this case.
1673          */
1674         old_size = other->end - other->start + 1;
1675         num_extents = count_max_extents(old_size);
1676         old_size = new->end - new->start + 1;
1677         num_extents += count_max_extents(old_size);
1678         if (count_max_extents(new_size) >= num_extents)
1679                 return;
1680
1681         spin_lock(&BTRFS_I(inode)->lock);
1682         BTRFS_I(inode)->outstanding_extents--;
1683         spin_unlock(&BTRFS_I(inode)->lock);
1684 }
1685
1686 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1687                                       struct inode *inode)
1688 {
1689         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1690
1691         spin_lock(&root->delalloc_lock);
1692         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1693                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1694                               &root->delalloc_inodes);
1695                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1696                         &BTRFS_I(inode)->runtime_flags);
1697                 root->nr_delalloc_inodes++;
1698                 if (root->nr_delalloc_inodes == 1) {
1699                         spin_lock(&fs_info->delalloc_root_lock);
1700                         BUG_ON(!list_empty(&root->delalloc_root));
1701                         list_add_tail(&root->delalloc_root,
1702                                       &fs_info->delalloc_roots);
1703                         spin_unlock(&fs_info->delalloc_root_lock);
1704                 }
1705         }
1706         spin_unlock(&root->delalloc_lock);
1707 }
1708
1709 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1710                                      struct btrfs_inode *inode)
1711 {
1712         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1713
1714         spin_lock(&root->delalloc_lock);
1715         if (!list_empty(&inode->delalloc_inodes)) {
1716                 list_del_init(&inode->delalloc_inodes);
1717                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1718                           &inode->runtime_flags);
1719                 root->nr_delalloc_inodes--;
1720                 if (!root->nr_delalloc_inodes) {
1721                         spin_lock(&fs_info->delalloc_root_lock);
1722                         BUG_ON(list_empty(&root->delalloc_root));
1723                         list_del_init(&root->delalloc_root);
1724                         spin_unlock(&fs_info->delalloc_root_lock);
1725                 }
1726         }
1727         spin_unlock(&root->delalloc_lock);
1728 }
1729
1730 /*
1731  * extent_io.c set_bit_hook, used to track delayed allocation
1732  * bytes in this file, and to maintain the list of inodes that
1733  * have pending delalloc work to be done.
1734  */
1735 static void btrfs_set_bit_hook(void *private_data,
1736                                struct extent_state *state, unsigned *bits)
1737 {
1738         struct inode *inode = private_data;
1739
1740         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1741
1742         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1743                 WARN_ON(1);
1744         /*
1745          * set_bit and clear bit hooks normally require _irqsave/restore
1746          * but in this case, we are only testing for the DELALLOC
1747          * bit, which is only set or cleared with irqs on
1748          */
1749         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1750                 struct btrfs_root *root = BTRFS_I(inode)->root;
1751                 u64 len = state->end + 1 - state->start;
1752                 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1753
1754                 if (*bits & EXTENT_FIRST_DELALLOC) {
1755                         *bits &= ~EXTENT_FIRST_DELALLOC;
1756                 } else {
1757                         spin_lock(&BTRFS_I(inode)->lock);
1758                         BTRFS_I(inode)->outstanding_extents++;
1759                         spin_unlock(&BTRFS_I(inode)->lock);
1760                 }
1761
1762                 /* For sanity tests */
1763                 if (btrfs_is_testing(fs_info))
1764                         return;
1765
1766                 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1767                                          fs_info->delalloc_batch);
1768                 spin_lock(&BTRFS_I(inode)->lock);
1769                 BTRFS_I(inode)->delalloc_bytes += len;
1770                 if (*bits & EXTENT_DEFRAG)
1771                         BTRFS_I(inode)->defrag_bytes += len;
1772                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1773                                          &BTRFS_I(inode)->runtime_flags))
1774                         btrfs_add_delalloc_inodes(root, inode);
1775                 spin_unlock(&BTRFS_I(inode)->lock);
1776         }
1777
1778         if (!(state->state & EXTENT_DELALLOC_NEW) &&
1779             (*bits & EXTENT_DELALLOC_NEW)) {
1780                 spin_lock(&BTRFS_I(inode)->lock);
1781                 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1782                         state->start;
1783                 spin_unlock(&BTRFS_I(inode)->lock);
1784         }
1785 }
1786
1787 /*
1788  * extent_io.c clear_bit_hook, see set_bit_hook for why
1789  */
1790 static void btrfs_clear_bit_hook(void *private_data,
1791                                  struct extent_state *state,
1792                                  unsigned *bits)
1793 {
1794         struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1795         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1796         u64 len = state->end + 1 - state->start;
1797         u32 num_extents = count_max_extents(len);
1798
1799         spin_lock(&inode->lock);
1800         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1801                 inode->defrag_bytes -= len;
1802         spin_unlock(&inode->lock);
1803
1804         /*
1805          * set_bit and clear bit hooks normally require _irqsave/restore
1806          * but in this case, we are only testing for the DELALLOC
1807          * bit, which is only set or cleared with irqs on
1808          */
1809         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1810                 struct btrfs_root *root = inode->root;
1811                 bool do_list = !btrfs_is_free_space_inode(inode);
1812
1813                 if (*bits & EXTENT_FIRST_DELALLOC) {
1814                         *bits &= ~EXTENT_FIRST_DELALLOC;
1815                 } else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1816                         spin_lock(&inode->lock);
1817                         inode->outstanding_extents -= num_extents;
1818                         spin_unlock(&inode->lock);
1819                 }
1820
1821                 /*
1822                  * We don't reserve metadata space for space cache inodes so we
1823                  * don't need to call dellalloc_release_metadata if there is an
1824                  * error.
1825                  */
1826                 if (*bits & EXTENT_CLEAR_META_RESV &&
1827                     root != fs_info->tree_root)
1828                         btrfs_delalloc_release_metadata(inode, len);
1829
1830                 /* For sanity tests. */
1831                 if (btrfs_is_testing(fs_info))
1832                         return;
1833
1834                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1835                     do_list && !(state->state & EXTENT_NORESERVE) &&
1836                     (*bits & EXTENT_CLEAR_DATA_RESV))
1837                         btrfs_free_reserved_data_space_noquota(
1838                                         &inode->vfs_inode,
1839                                         state->start, len);
1840
1841                 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1842                                          fs_info->delalloc_batch);
1843                 spin_lock(&inode->lock);
1844                 inode->delalloc_bytes -= len;
1845                 if (do_list && inode->delalloc_bytes == 0 &&
1846                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1847                                         &inode->runtime_flags))
1848                         btrfs_del_delalloc_inode(root, inode);
1849                 spin_unlock(&inode->lock);
1850         }
1851
1852         if ((state->state & EXTENT_DELALLOC_NEW) &&
1853             (*bits & EXTENT_DELALLOC_NEW)) {
1854                 spin_lock(&inode->lock);
1855                 ASSERT(inode->new_delalloc_bytes >= len);
1856                 inode->new_delalloc_bytes -= len;
1857                 spin_unlock(&inode->lock);
1858         }
1859 }
1860
1861 /*
1862  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1863  * we don't create bios that span stripes or chunks
1864  *
1865  * return 1 if page cannot be merged to bio
1866  * return 0 if page can be merged to bio
1867  * return error otherwise
1868  */
1869 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1870                          size_t size, struct bio *bio,
1871                          unsigned long bio_flags)
1872 {
1873         struct inode *inode = page->mapping->host;
1874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1875         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1876         u64 length = 0;
1877         u64 map_length;
1878         int ret;
1879
1880         if (bio_flags & EXTENT_BIO_COMPRESSED)
1881                 return 0;
1882
1883         length = bio->bi_iter.bi_size;
1884         map_length = length;
1885         ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1886                               NULL, 0);
1887         if (ret < 0)
1888                 return ret;
1889         if (map_length < length + size)
1890                 return 1;
1891         return 0;
1892 }
1893
1894 /*
1895  * in order to insert checksums into the metadata in large chunks,
1896  * we wait until bio submission time.   All the pages in the bio are
1897  * checksummed and sums are attached onto the ordered extent record.
1898  *
1899  * At IO completion time the cums attached on the ordered extent record
1900  * are inserted into the btree
1901  */
1902 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1903                                     int mirror_num, unsigned long bio_flags,
1904                                     u64 bio_offset)
1905 {
1906         struct inode *inode = private_data;
1907         blk_status_t ret = 0;
1908
1909         ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1910         BUG_ON(ret); /* -ENOMEM */
1911         return 0;
1912 }
1913
1914 /*
1915  * in order to insert checksums into the metadata in large chunks,
1916  * we wait until bio submission time.   All the pages in the bio are
1917  * checksummed and sums are attached onto the ordered extent record.
1918  *
1919  * At IO completion time the cums attached on the ordered extent record
1920  * are inserted into the btree
1921  */
1922 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1923                           int mirror_num, unsigned long bio_flags,
1924                           u64 bio_offset)
1925 {
1926         struct inode *inode = private_data;
1927         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1928         blk_status_t ret;
1929
1930         ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1931         if (ret) {
1932                 bio->bi_status = ret;
1933                 bio_endio(bio);
1934         }
1935         return ret;
1936 }
1937
1938 /*
1939  * extent_io.c submission hook. This does the right thing for csum calculation
1940  * on write, or reading the csums from the tree before a read
1941  */
1942 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1943                                  int mirror_num, unsigned long bio_flags,
1944                                  u64 bio_offset)
1945 {
1946         struct inode *inode = private_data;
1947         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1948         struct btrfs_root *root = BTRFS_I(inode)->root;
1949         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1950         blk_status_t ret = 0;
1951         int skip_sum;
1952         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1953
1954         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1955
1956         if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1957                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1958
1959         if (bio_op(bio) != REQ_OP_WRITE) {
1960                 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1961                 if (ret)
1962                         goto out;
1963
1964                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1965                         ret = btrfs_submit_compressed_read(inode, bio,
1966                                                            mirror_num,
1967                                                            bio_flags);
1968                         goto out;
1969                 } else if (!skip_sum) {
1970                         ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1971                         if (ret)
1972                                 goto out;
1973                 }
1974                 goto mapit;
1975         } else if (async && !skip_sum) {
1976                 /* csum items have already been cloned */
1977                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1978                         goto mapit;
1979                 /* we're doing a write, do the async checksumming */
1980                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1981                                           bio_offset, inode,
1982                                           __btrfs_submit_bio_start,
1983                                           __btrfs_submit_bio_done);
1984                 goto out;
1985         } else if (!skip_sum) {
1986                 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1987                 if (ret)
1988                         goto out;
1989         }
1990
1991 mapit:
1992         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1993
1994 out:
1995         if (ret) {
1996                 bio->bi_status = ret;
1997                 bio_endio(bio);
1998         }
1999         return ret;
2000 }
2001
2002 /*
2003  * given a list of ordered sums record them in the inode.  This happens
2004  * at IO completion time based on sums calculated at bio submission time.
2005  */
2006 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2007                              struct inode *inode, struct list_head *list)
2008 {
2009         struct btrfs_ordered_sum *sum;
2010
2011         list_for_each_entry(sum, list, list) {
2012                 trans->adding_csums = 1;
2013                 btrfs_csum_file_blocks(trans,
2014                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
2015                 trans->adding_csums = 0;
2016         }
2017         return 0;
2018 }
2019
2020 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2021                               struct extent_state **cached_state, int dedupe)
2022 {
2023         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2024         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2025                                    cached_state);
2026 }
2027
2028 /* see btrfs_writepage_start_hook for details on why this is required */
2029 struct btrfs_writepage_fixup {
2030         struct page *page;
2031         struct btrfs_work work;
2032 };
2033
2034 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2035 {
2036         struct btrfs_writepage_fixup *fixup;
2037         struct btrfs_ordered_extent *ordered;
2038         struct extent_state *cached_state = NULL;
2039         struct extent_changeset *data_reserved = NULL;
2040         struct page *page;
2041         struct inode *inode;
2042         u64 page_start;
2043         u64 page_end;
2044         int ret;
2045
2046         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2047         page = fixup->page;
2048 again:
2049         lock_page(page);
2050         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2051                 ClearPageChecked(page);
2052                 goto out_page;
2053         }
2054
2055         inode = page->mapping->host;
2056         page_start = page_offset(page);
2057         page_end = page_offset(page) + PAGE_SIZE - 1;
2058
2059         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2060                          &cached_state);
2061
2062         /* already ordered? We're done */
2063         if (PagePrivate2(page))
2064                 goto out;
2065
2066         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2067                                         PAGE_SIZE);
2068         if (ordered) {
2069                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2070                                      page_end, &cached_state, GFP_NOFS);
2071                 unlock_page(page);
2072                 btrfs_start_ordered_extent(inode, ordered, 1);
2073                 btrfs_put_ordered_extent(ordered);
2074                 goto again;
2075         }
2076
2077         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2078                                            PAGE_SIZE);
2079         if (ret) {
2080                 mapping_set_error(page->mapping, ret);
2081                 end_extent_writepage(page, ret, page_start, page_end);
2082                 ClearPageChecked(page);
2083                 goto out;
2084          }
2085
2086         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2087                                   0);
2088         ClearPageChecked(page);
2089         set_page_dirty(page);
2090 out:
2091         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2092                              &cached_state, GFP_NOFS);
2093 out_page:
2094         unlock_page(page);
2095         put_page(page);
2096         kfree(fixup);
2097         extent_changeset_free(data_reserved);
2098 }
2099
2100 /*
2101  * There are a few paths in the higher layers of the kernel that directly
2102  * set the page dirty bit without asking the filesystem if it is a
2103  * good idea.  This causes problems because we want to make sure COW
2104  * properly happens and the data=ordered rules are followed.
2105  *
2106  * In our case any range that doesn't have the ORDERED bit set
2107  * hasn't been properly setup for IO.  We kick off an async process
2108  * to fix it up.  The async helper will wait for ordered extents, set
2109  * the delalloc bit and make it safe to write the page.
2110  */
2111 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2112 {
2113         struct inode *inode = page->mapping->host;
2114         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2115         struct btrfs_writepage_fixup *fixup;
2116
2117         /* this page is properly in the ordered list */
2118         if (TestClearPagePrivate2(page))
2119                 return 0;
2120
2121         if (PageChecked(page))
2122                 return -EAGAIN;
2123
2124         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2125         if (!fixup)
2126                 return -EAGAIN;
2127
2128         SetPageChecked(page);
2129         get_page(page);
2130         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2131                         btrfs_writepage_fixup_worker, NULL, NULL);
2132         fixup->page = page;
2133         btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2134         return -EBUSY;
2135 }
2136
2137 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2138                                        struct inode *inode, u64 file_pos,
2139                                        u64 disk_bytenr, u64 disk_num_bytes,
2140                                        u64 num_bytes, u64 ram_bytes,
2141                                        u8 compression, u8 encryption,
2142                                        u16 other_encoding, int extent_type)
2143 {
2144         struct btrfs_root *root = BTRFS_I(inode)->root;
2145         struct btrfs_file_extent_item *fi;
2146         struct btrfs_path *path;
2147         struct extent_buffer *leaf;
2148         struct btrfs_key ins;
2149         u64 qg_released;
2150         int extent_inserted = 0;
2151         int ret;
2152
2153         path = btrfs_alloc_path();
2154         if (!path)
2155                 return -ENOMEM;
2156
2157         /*
2158          * we may be replacing one extent in the tree with another.
2159          * The new extent is pinned in the extent map, and we don't want
2160          * to drop it from the cache until it is completely in the btree.
2161          *
2162          * So, tell btrfs_drop_extents to leave this extent in the cache.
2163          * the caller is expected to unpin it and allow it to be merged
2164          * with the others.
2165          */
2166         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2167                                    file_pos + num_bytes, NULL, 0,
2168                                    1, sizeof(*fi), &extent_inserted);
2169         if (ret)
2170                 goto out;
2171
2172         if (!extent_inserted) {
2173                 ins.objectid = btrfs_ino(BTRFS_I(inode));
2174                 ins.offset = file_pos;
2175                 ins.type = BTRFS_EXTENT_DATA_KEY;
2176
2177                 path->leave_spinning = 1;
2178                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2179                                               sizeof(*fi));
2180                 if (ret)
2181                         goto out;
2182         }
2183         leaf = path->nodes[0];
2184         fi = btrfs_item_ptr(leaf, path->slots[0],
2185                             struct btrfs_file_extent_item);
2186         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2187         btrfs_set_file_extent_type(leaf, fi, extent_type);
2188         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2189         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2190         btrfs_set_file_extent_offset(leaf, fi, 0);
2191         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2192         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2193         btrfs_set_file_extent_compression(leaf, fi, compression);
2194         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2195         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2196
2197         btrfs_mark_buffer_dirty(leaf);
2198         btrfs_release_path(path);
2199
2200         inode_add_bytes(inode, num_bytes);
2201
2202         ins.objectid = disk_bytenr;
2203         ins.offset = disk_num_bytes;
2204         ins.type = BTRFS_EXTENT_ITEM_KEY;
2205
2206         /*
2207          * Release the reserved range from inode dirty range map, as it is
2208          * already moved into delayed_ref_head
2209          */
2210         ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2211         if (ret < 0)
2212                 goto out;
2213         qg_released = ret;
2214         ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2215                         btrfs_ino(BTRFS_I(inode)), file_pos, qg_released, &ins);
2216 out:
2217         btrfs_free_path(path);
2218
2219         return ret;
2220 }
2221
2222 /* snapshot-aware defrag */
2223 struct sa_defrag_extent_backref {
2224         struct rb_node node;
2225         struct old_sa_defrag_extent *old;
2226         u64 root_id;
2227         u64 inum;
2228         u64 file_pos;
2229         u64 extent_offset;
2230         u64 num_bytes;
2231         u64 generation;
2232 };
2233
2234 struct old_sa_defrag_extent {
2235         struct list_head list;
2236         struct new_sa_defrag_extent *new;
2237
2238         u64 extent_offset;
2239         u64 bytenr;
2240         u64 offset;
2241         u64 len;
2242         int count;
2243 };
2244
2245 struct new_sa_defrag_extent {
2246         struct rb_root root;
2247         struct list_head head;
2248         struct btrfs_path *path;
2249         struct inode *inode;
2250         u64 file_pos;
2251         u64 len;
2252         u64 bytenr;
2253         u64 disk_len;
2254         u8 compress_type;
2255 };
2256
2257 static int backref_comp(struct sa_defrag_extent_backref *b1,
2258                         struct sa_defrag_extent_backref *b2)
2259 {
2260         if (b1->root_id < b2->root_id)
2261                 return -1;
2262         else if (b1->root_id > b2->root_id)
2263                 return 1;
2264
2265         if (b1->inum < b2->inum)
2266                 return -1;
2267         else if (b1->inum > b2->inum)
2268                 return 1;
2269
2270         if (b1->file_pos < b2->file_pos)
2271                 return -1;
2272         else if (b1->file_pos > b2->file_pos)
2273                 return 1;
2274
2275         /*
2276          * [------------------------------] ===> (a range of space)
2277          *     |<--->|   |<---->| =============> (fs/file tree A)
2278          * |<---------------------------->| ===> (fs/file tree B)
2279          *
2280          * A range of space can refer to two file extents in one tree while
2281          * refer to only one file extent in another tree.
2282          *
2283          * So we may process a disk offset more than one time(two extents in A)
2284          * and locate at the same extent(one extent in B), then insert two same
2285          * backrefs(both refer to the extent in B).
2286          */
2287         return 0;
2288 }
2289
2290 static void backref_insert(struct rb_root *root,
2291                            struct sa_defrag_extent_backref *backref)
2292 {
2293         struct rb_node **p = &root->rb_node;
2294         struct rb_node *parent = NULL;
2295         struct sa_defrag_extent_backref *entry;
2296         int ret;
2297
2298         while (*p) {
2299                 parent = *p;
2300                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2301
2302                 ret = backref_comp(backref, entry);
2303                 if (ret < 0)
2304                         p = &(*p)->rb_left;
2305                 else
2306                         p = &(*p)->rb_right;
2307         }
2308
2309         rb_link_node(&backref->node, parent, p);
2310         rb_insert_color(&backref->node, root);
2311 }
2312
2313 /*
2314  * Note the backref might has changed, and in this case we just return 0.
2315  */
2316 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2317                                        void *ctx)
2318 {
2319         struct btrfs_file_extent_item *extent;
2320         struct old_sa_defrag_extent *old = ctx;
2321         struct new_sa_defrag_extent *new = old->new;
2322         struct btrfs_path *path = new->path;
2323         struct btrfs_key key;
2324         struct btrfs_root *root;
2325         struct sa_defrag_extent_backref *backref;
2326         struct extent_buffer *leaf;
2327         struct inode *inode = new->inode;
2328         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2329         int slot;
2330         int ret;
2331         u64 extent_offset;
2332         u64 num_bytes;
2333
2334         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2335             inum == btrfs_ino(BTRFS_I(inode)))
2336                 return 0;
2337
2338         key.objectid = root_id;
2339         key.type = BTRFS_ROOT_ITEM_KEY;
2340         key.offset = (u64)-1;
2341
2342         root = btrfs_read_fs_root_no_name(fs_info, &key);
2343         if (IS_ERR(root)) {
2344                 if (PTR_ERR(root) == -ENOENT)
2345                         return 0;
2346                 WARN_ON(1);
2347                 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2348                          inum, offset, root_id);
2349                 return PTR_ERR(root);
2350         }
2351
2352         key.objectid = inum;
2353         key.type = BTRFS_EXTENT_DATA_KEY;
2354         if (offset > (u64)-1 << 32)
2355                 key.offset = 0;
2356         else
2357                 key.offset = offset;
2358
2359         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2360         if (WARN_ON(ret < 0))
2361                 return ret;
2362         ret = 0;
2363
2364         while (1) {
2365                 cond_resched();
2366
2367                 leaf = path->nodes[0];
2368                 slot = path->slots[0];
2369
2370                 if (slot >= btrfs_header_nritems(leaf)) {
2371                         ret = btrfs_next_leaf(root, path);
2372                         if (ret < 0) {
2373                                 goto out;
2374                         } else if (ret > 0) {
2375                                 ret = 0;
2376                                 goto out;
2377                         }
2378                         continue;
2379                 }
2380
2381                 path->slots[0]++;
2382
2383                 btrfs_item_key_to_cpu(leaf, &key, slot);
2384
2385                 if (key.objectid > inum)
2386                         goto out;
2387
2388                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2389                         continue;
2390
2391                 extent = btrfs_item_ptr(leaf, slot,
2392                                         struct btrfs_file_extent_item);
2393
2394                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2395                         continue;
2396
2397                 /*
2398                  * 'offset' refers to the exact key.offset,
2399                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2400                  * (key.offset - extent_offset).
2401                  */
2402                 if (key.offset != offset)
2403                         continue;
2404
2405                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2406                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2407
2408                 if (extent_offset >= old->extent_offset + old->offset +
2409                     old->len || extent_offset + num_bytes <=
2410                     old->extent_offset + old->offset)
2411                         continue;
2412                 break;
2413         }
2414
2415         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2416         if (!backref) {
2417                 ret = -ENOENT;
2418                 goto out;
2419         }
2420
2421         backref->root_id = root_id;
2422         backref->inum = inum;
2423         backref->file_pos = offset;
2424         backref->num_bytes = num_bytes;
2425         backref->extent_offset = extent_offset;
2426         backref->generation = btrfs_file_extent_generation(leaf, extent);
2427         backref->old = old;
2428         backref_insert(&new->root, backref);
2429         old->count++;
2430 out:
2431         btrfs_release_path(path);
2432         WARN_ON(ret);
2433         return ret;
2434 }
2435
2436 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2437                                    struct new_sa_defrag_extent *new)
2438 {
2439         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2440         struct old_sa_defrag_extent *old, *tmp;
2441         int ret;
2442
2443         new->path = path;
2444
2445         list_for_each_entry_safe(old, tmp, &new->head, list) {
2446                 ret = iterate_inodes_from_logical(old->bytenr +
2447                                                   old->extent_offset, fs_info,
2448                                                   path, record_one_backref,
2449                                                   old);
2450                 if (ret < 0 && ret != -ENOENT)
2451                         return false;
2452
2453                 /* no backref to be processed for this extent */
2454                 if (!old->count) {
2455                         list_del(&old->list);
2456                         kfree(old);
2457                 }
2458         }
2459
2460         if (list_empty(&new->head))
2461                 return false;
2462
2463         return true;
2464 }
2465
2466 static int relink_is_mergable(struct extent_buffer *leaf,
2467                               struct btrfs_file_extent_item *fi,
2468                               struct new_sa_defrag_extent *new)
2469 {
2470         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2471                 return 0;
2472
2473         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2474                 return 0;
2475
2476         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2477                 return 0;
2478
2479         if (btrfs_file_extent_encryption(leaf, fi) ||
2480             btrfs_file_extent_other_encoding(leaf, fi))
2481                 return 0;
2482
2483         return 1;
2484 }
2485
2486 /*
2487  * Note the backref might has changed, and in this case we just return 0.
2488  */
2489 static noinline int relink_extent_backref(struct btrfs_path *path,
2490                                  struct sa_defrag_extent_backref *prev,
2491                                  struct sa_defrag_extent_backref *backref)
2492 {
2493         struct btrfs_file_extent_item *extent;
2494         struct btrfs_file_extent_item *item;
2495         struct btrfs_ordered_extent *ordered;
2496         struct btrfs_trans_handle *trans;
2497         struct btrfs_root *root;
2498         struct btrfs_key key;
2499         struct extent_buffer *leaf;
2500         struct old_sa_defrag_extent *old = backref->old;
2501         struct new_sa_defrag_extent *new = old->new;
2502         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2503         struct inode *inode;
2504         struct extent_state *cached = NULL;
2505         int ret = 0;
2506         u64 start;
2507         u64 len;
2508         u64 lock_start;
2509         u64 lock_end;
2510         bool merge = false;
2511         int index;
2512
2513         if (prev && prev->root_id == backref->root_id &&
2514             prev->inum == backref->inum &&
2515             prev->file_pos + prev->num_bytes == backref->file_pos)
2516                 merge = true;
2517
2518         /* step 1: get root */
2519         key.objectid = backref->root_id;
2520         key.type = BTRFS_ROOT_ITEM_KEY;
2521         key.offset = (u64)-1;
2522
2523         index = srcu_read_lock(&fs_info->subvol_srcu);
2524
2525         root = btrfs_read_fs_root_no_name(fs_info, &key);
2526         if (IS_ERR(root)) {
2527                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2528                 if (PTR_ERR(root) == -ENOENT)
2529                         return 0;
2530                 return PTR_ERR(root);
2531         }
2532
2533         if (btrfs_root_readonly(root)) {
2534                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2535                 return 0;
2536         }
2537
2538         /* step 2: get inode */
2539         key.objectid = backref->inum;
2540         key.type = BTRFS_INODE_ITEM_KEY;
2541         key.offset = 0;
2542
2543         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2544         if (IS_ERR(inode)) {
2545                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2546                 return 0;
2547         }
2548
2549         srcu_read_unlock(&fs_info->subvol_srcu, index);
2550
2551         /* step 3: relink backref */
2552         lock_start = backref->file_pos;
2553         lock_end = backref->file_pos + backref->num_bytes - 1;
2554         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2555                          &cached);
2556
2557         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2558         if (ordered) {
2559                 btrfs_put_ordered_extent(ordered);
2560                 goto out_unlock;
2561         }
2562
2563         trans = btrfs_join_transaction(root);
2564         if (IS_ERR(trans)) {
2565                 ret = PTR_ERR(trans);
2566                 goto out_unlock;
2567         }
2568
2569         key.objectid = backref->inum;
2570         key.type = BTRFS_EXTENT_DATA_KEY;
2571         key.offset = backref->file_pos;
2572
2573         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2574         if (ret < 0) {
2575                 goto out_free_path;
2576         } else if (ret > 0) {
2577                 ret = 0;
2578                 goto out_free_path;
2579         }
2580
2581         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2582                                 struct btrfs_file_extent_item);
2583
2584         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2585             backref->generation)
2586                 goto out_free_path;
2587
2588         btrfs_release_path(path);
2589
2590         start = backref->file_pos;
2591         if (backref->extent_offset < old->extent_offset + old->offset)
2592                 start += old->extent_offset + old->offset -
2593                          backref->extent_offset;
2594
2595         len = min(backref->extent_offset + backref->num_bytes,
2596                   old->extent_offset + old->offset + old->len);
2597         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2598
2599         ret = btrfs_drop_extents(trans, root, inode, start,
2600                                  start + len, 1);
2601         if (ret)
2602                 goto out_free_path;
2603 again:
2604         key.objectid = btrfs_ino(BTRFS_I(inode));
2605         key.type = BTRFS_EXTENT_DATA_KEY;
2606         key.offset = start;
2607
2608         path->leave_spinning = 1;
2609         if (merge) {
2610                 struct btrfs_file_extent_item *fi;
2611                 u64 extent_len;
2612                 struct btrfs_key found_key;
2613
2614                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2615                 if (ret < 0)
2616                         goto out_free_path;
2617
2618                 path->slots[0]--;
2619                 leaf = path->nodes[0];
2620                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2621
2622                 fi = btrfs_item_ptr(leaf, path->slots[0],
2623                                     struct btrfs_file_extent_item);
2624                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2625
2626                 if (extent_len + found_key.offset == start &&
2627                     relink_is_mergable(leaf, fi, new)) {
2628                         btrfs_set_file_extent_num_bytes(leaf, fi,
2629                                                         extent_len + len);
2630                         btrfs_mark_buffer_dirty(leaf);
2631                         inode_add_bytes(inode, len);
2632
2633                         ret = 1;
2634                         goto out_free_path;
2635                 } else {
2636                         merge = false;
2637                         btrfs_release_path(path);
2638                         goto again;
2639                 }
2640         }
2641
2642         ret = btrfs_insert_empty_item(trans, root, path, &key,
2643                                         sizeof(*extent));
2644         if (ret) {
2645                 btrfs_abort_transaction(trans, ret);
2646                 goto out_free_path;
2647         }
2648
2649         leaf = path->nodes[0];
2650         item = btrfs_item_ptr(leaf, path->slots[0],
2651                                 struct btrfs_file_extent_item);
2652         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2653         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2654         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2655         btrfs_set_file_extent_num_bytes(leaf, item, len);
2656         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2657         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2658         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2659         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2660         btrfs_set_file_extent_encryption(leaf, item, 0);
2661         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2662
2663         btrfs_mark_buffer_dirty(leaf);
2664         inode_add_bytes(inode, len);
2665         btrfs_release_path(path);
2666
2667         ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2668                         new->disk_len, 0,
2669                         backref->root_id, backref->inum,
2670                         new->file_pos); /* start - extent_offset */
2671         if (ret) {
2672                 btrfs_abort_transaction(trans, ret);
2673                 goto out_free_path;
2674         }
2675
2676         ret = 1;
2677 out_free_path:
2678         btrfs_release_path(path);
2679         path->leave_spinning = 0;
2680         btrfs_end_transaction(trans);
2681 out_unlock:
2682         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2683                              &cached, GFP_NOFS);
2684         iput(inode);
2685         return ret;
2686 }
2687
2688 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2689 {
2690         struct old_sa_defrag_extent *old, *tmp;
2691
2692         if (!new)
2693                 return;
2694
2695         list_for_each_entry_safe(old, tmp, &new->head, list) {
2696                 kfree(old);
2697         }
2698         kfree(new);
2699 }
2700
2701 static void relink_file_extents(struct new_sa_defrag_extent *new)
2702 {
2703         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2704         struct btrfs_path *path;
2705         struct sa_defrag_extent_backref *backref;
2706         struct sa_defrag_extent_backref *prev = NULL;
2707         struct inode *inode;
2708         struct btrfs_root *root;
2709         struct rb_node *node;
2710         int ret;
2711
2712         inode = new->inode;
2713         root = BTRFS_I(inode)->root;
2714
2715         path = btrfs_alloc_path();
2716         if (!path)
2717                 return;
2718
2719         if (!record_extent_backrefs(path, new)) {
2720                 btrfs_free_path(path);
2721                 goto out;
2722         }
2723         btrfs_release_path(path);
2724
2725         while (1) {
2726                 node = rb_first(&new->root);
2727                 if (!node)
2728                         break;
2729                 rb_erase(node, &new->root);
2730
2731                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2732
2733                 ret = relink_extent_backref(path, prev, backref);
2734                 WARN_ON(ret < 0);
2735
2736                 kfree(prev);
2737
2738                 if (ret == 1)
2739                         prev = backref;
2740                 else
2741                         prev = NULL;
2742                 cond_resched();
2743         }
2744         kfree(prev);
2745
2746         btrfs_free_path(path);
2747 out:
2748         free_sa_defrag_extent(new);
2749
2750         atomic_dec(&fs_info->defrag_running);
2751         wake_up(&fs_info->transaction_wait);
2752 }
2753
2754 static struct new_sa_defrag_extent *
2755 record_old_file_extents(struct inode *inode,
2756                         struct btrfs_ordered_extent *ordered)
2757 {
2758         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2759         struct btrfs_root *root = BTRFS_I(inode)->root;
2760         struct btrfs_path *path;
2761         struct btrfs_key key;
2762         struct old_sa_defrag_extent *old;
2763         struct new_sa_defrag_extent *new;
2764         int ret;
2765
2766         new = kmalloc(sizeof(*new), GFP_NOFS);
2767         if (!new)
2768                 return NULL;
2769
2770         new->inode = inode;
2771         new->file_pos = ordered->file_offset;
2772         new->len = ordered->len;
2773         new->bytenr = ordered->start;
2774         new->disk_len = ordered->disk_len;
2775         new->compress_type = ordered->compress_type;
2776         new->root = RB_ROOT;
2777         INIT_LIST_HEAD(&new->head);
2778
2779         path = btrfs_alloc_path();
2780         if (!path)
2781                 goto out_kfree;
2782
2783         key.objectid = btrfs_ino(BTRFS_I(inode));
2784         key.type = BTRFS_EXTENT_DATA_KEY;
2785         key.offset = new->file_pos;
2786
2787         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2788         if (ret < 0)
2789                 goto out_free_path;
2790         if (ret > 0 && path->slots[0] > 0)
2791                 path->slots[0]--;
2792
2793         /* find out all the old extents for the file range */
2794         while (1) {
2795                 struct btrfs_file_extent_item *extent;
2796                 struct extent_buffer *l;
2797                 int slot;
2798                 u64 num_bytes;
2799                 u64 offset;
2800                 u64 end;
2801                 u64 disk_bytenr;
2802                 u64 extent_offset;
2803
2804                 l = path->nodes[0];
2805                 slot = path->slots[0];
2806
2807                 if (slot >= btrfs_header_nritems(l)) {
2808                         ret = btrfs_next_leaf(root, path);
2809                         if (ret < 0)
2810                                 goto out_free_path;
2811                         else if (ret > 0)
2812                                 break;
2813                         continue;
2814                 }
2815
2816                 btrfs_item_key_to_cpu(l, &key, slot);
2817
2818                 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2819                         break;
2820                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2821                         break;
2822                 if (key.offset >= new->file_pos + new->len)
2823                         break;
2824
2825                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2826
2827                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2828                 if (key.offset + num_bytes < new->file_pos)
2829                         goto next;
2830
2831                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2832                 if (!disk_bytenr)
2833                         goto next;
2834
2835                 extent_offset = btrfs_file_extent_offset(l, extent);
2836
2837                 old = kmalloc(sizeof(*old), GFP_NOFS);
2838                 if (!old)
2839                         goto out_free_path;
2840
2841                 offset = max(new->file_pos, key.offset);
2842                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2843
2844                 old->bytenr = disk_bytenr;
2845                 old->extent_offset = extent_offset;
2846                 old->offset = offset - key.offset;
2847                 old->len = end - offset;
2848                 old->new = new;
2849                 old->count = 0;
2850                 list_add_tail(&old->list, &new->head);
2851 next:
2852                 path->slots[0]++;
2853                 cond_resched();
2854         }
2855
2856         btrfs_free_path(path);
2857         atomic_inc(&fs_info->defrag_running);
2858
2859         return new;
2860
2861 out_free_path:
2862         btrfs_free_path(path);
2863 out_kfree:
2864         free_sa_defrag_extent(new);
2865         return NULL;
2866 }
2867
2868 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2869                                          u64 start, u64 len)
2870 {
2871         struct btrfs_block_group_cache *cache;
2872
2873         cache = btrfs_lookup_block_group(fs_info, start);
2874         ASSERT(cache);
2875
2876         spin_lock(&cache->lock);
2877         cache->delalloc_bytes -= len;
2878         spin_unlock(&cache->lock);
2879
2880         btrfs_put_block_group(cache);
2881 }
2882
2883 /* as ordered data IO finishes, this gets called so we can finish
2884  * an ordered extent if the range of bytes in the file it covers are
2885  * fully written.
2886  */
2887 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2888 {
2889         struct inode *inode = ordered_extent->inode;
2890         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2891         struct btrfs_root *root = BTRFS_I(inode)->root;
2892         struct btrfs_trans_handle *trans = NULL;
2893         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2894         struct extent_state *cached_state = NULL;
2895         struct new_sa_defrag_extent *new = NULL;
2896         int compress_type = 0;
2897         int ret = 0;
2898         u64 logical_len = ordered_extent->len;
2899         bool nolock;
2900         bool truncated = false;
2901         bool range_locked = false;
2902         bool clear_new_delalloc_bytes = false;
2903
2904         if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2905             !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2906             !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2907                 clear_new_delalloc_bytes = true;
2908
2909         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2910
2911         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2912                 ret = -EIO;
2913                 goto out;
2914         }
2915
2916         btrfs_free_io_failure_record(BTRFS_I(inode),
2917                         ordered_extent->file_offset,
2918                         ordered_extent->file_offset +
2919                         ordered_extent->len - 1);
2920
2921         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2922                 truncated = true;
2923                 logical_len = ordered_extent->truncated_len;
2924                 /* Truncated the entire extent, don't bother adding */
2925                 if (!logical_len)
2926                         goto out;
2927         }
2928
2929         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2930                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2931
2932                 /*
2933                  * For mwrite(mmap + memset to write) case, we still reserve
2934                  * space for NOCOW range.
2935                  * As NOCOW won't cause a new delayed ref, just free the space
2936                  */
2937                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2938                                        ordered_extent->len);
2939                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2940                 if (nolock)
2941                         trans = btrfs_join_transaction_nolock(root);
2942                 else
2943                         trans = btrfs_join_transaction(root);
2944                 if (IS_ERR(trans)) {
2945                         ret = PTR_ERR(trans);
2946                         trans = NULL;
2947                         goto out;
2948                 }
2949                 trans->block_rsv = &fs_info->delalloc_block_rsv;
2950                 ret = btrfs_update_inode_fallback(trans, root, inode);
2951                 if (ret) /* -ENOMEM or corruption */
2952                         btrfs_abort_transaction(trans, ret);
2953                 goto out;
2954         }
2955
2956         range_locked = true;
2957         lock_extent_bits(io_tree, ordered_extent->file_offset,
2958                          ordered_extent->file_offset + ordered_extent->len - 1,
2959                          &cached_state);
2960
2961         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2962                         ordered_extent->file_offset + ordered_extent->len - 1,
2963                         EXTENT_DEFRAG, 0, cached_state);
2964         if (ret) {
2965                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2966                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2967                         /* the inode is shared */
2968                         new = record_old_file_extents(inode, ordered_extent);
2969
2970                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2971                         ordered_extent->file_offset + ordered_extent->len - 1,
2972                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2973         }
2974
2975         if (nolock)
2976                 trans = btrfs_join_transaction_nolock(root);
2977         else
2978                 trans = btrfs_join_transaction(root);
2979         if (IS_ERR(trans)) {
2980                 ret = PTR_ERR(trans);
2981                 trans = NULL;
2982                 goto out;
2983         }
2984
2985         trans->block_rsv = &fs_info->delalloc_block_rsv;
2986
2987         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2988                 compress_type = ordered_extent->compress_type;
2989         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2990                 BUG_ON(compress_type);
2991                 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2992                                                 ordered_extent->file_offset,
2993                                                 ordered_extent->file_offset +
2994                                                 logical_len);
2995         } else {
2996                 BUG_ON(root == fs_info->tree_root);
2997                 ret = insert_reserved_file_extent(trans, inode,
2998                                                 ordered_extent->file_offset,
2999                                                 ordered_extent->start,
3000                                                 ordered_extent->disk_len,
3001                                                 logical_len, logical_len,
3002                                                 compress_type, 0, 0,
3003                                                 BTRFS_FILE_EXTENT_REG);
3004                 if (!ret)
3005                         btrfs_release_delalloc_bytes(fs_info,
3006                                                      ordered_extent->start,
3007                                                      ordered_extent->disk_len);
3008         }
3009         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3010                            ordered_extent->file_offset, ordered_extent->len,
3011                            trans->transid);
3012         if (ret < 0) {
3013                 btrfs_abort_transaction(trans, ret);
3014                 goto out;
3015         }
3016
3017         add_pending_csums(trans, inode, &ordered_extent->list);
3018
3019         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3020         ret = btrfs_update_inode_fallback(trans, root, inode);
3021         if (ret) { /* -ENOMEM or corruption */
3022                 btrfs_abort_transaction(trans, ret);
3023                 goto out;
3024         }
3025         ret = 0;
3026 out:
3027         if (range_locked || clear_new_delalloc_bytes) {
3028                 unsigned int clear_bits = 0;
3029
3030                 if (range_locked)
3031                         clear_bits |= EXTENT_LOCKED;
3032                 if (clear_new_delalloc_bytes)
3033                         clear_bits |= EXTENT_DELALLOC_NEW;
3034                 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3035                                  ordered_extent->file_offset,
3036                                  ordered_extent->file_offset +
3037                                  ordered_extent->len - 1,
3038                                  clear_bits,
3039                                  (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3040                                  0, &cached_state, GFP_NOFS);
3041         }
3042
3043         if (root != fs_info->tree_root)
3044                 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3045                                 ordered_extent->len);
3046         if (trans)
3047                 btrfs_end_transaction(trans);
3048
3049         if (ret || truncated) {
3050                 u64 start, end;
3051
3052                 if (truncated)
3053                         start = ordered_extent->file_offset + logical_len;
3054                 else
3055                         start = ordered_extent->file_offset;
3056                 end = ordered_extent->file_offset + ordered_extent->len - 1;
3057                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3058
3059                 /* Drop the cache for the part of the extent we didn't write. */
3060                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3061
3062                 /*
3063                  * If the ordered extent had an IOERR or something else went
3064                  * wrong we need to return the space for this ordered extent
3065                  * back to the allocator.  We only free the extent in the
3066                  * truncated case if we didn't write out the extent at all.
3067                  */
3068                 if ((ret || !logical_len) &&
3069                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3070                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3071                         btrfs_free_reserved_extent(fs_info,
3072                                                    ordered_extent->start,
3073                                                    ordered_extent->disk_len, 1);
3074         }
3075
3076
3077         /*
3078          * This needs to be done to make sure anybody waiting knows we are done
3079          * updating everything for this ordered extent.
3080          */
3081         btrfs_remove_ordered_extent(inode, ordered_extent);
3082
3083         /* for snapshot-aware defrag */
3084         if (new) {
3085                 if (ret) {
3086                         free_sa_defrag_extent(new);
3087                         atomic_dec(&fs_info->defrag_running);
3088                 } else {
3089                         relink_file_extents(new);
3090                 }
3091         }
3092
3093         /* once for us */
3094         btrfs_put_ordered_extent(ordered_extent);
3095         /* once for the tree */
3096         btrfs_put_ordered_extent(ordered_extent);
3097
3098         return ret;
3099 }
3100
3101 static void finish_ordered_fn(struct btrfs_work *work)
3102 {
3103         struct btrfs_ordered_extent *ordered_extent;
3104         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3105         btrfs_finish_ordered_io(ordered_extent);
3106 }
3107
3108 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3109                                 struct extent_state *state, int uptodate)
3110 {
3111         struct inode *inode = page->mapping->host;
3112         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3113         struct btrfs_ordered_extent *ordered_extent = NULL;
3114         struct btrfs_workqueue *wq;
3115         btrfs_work_func_t func;
3116
3117         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3118
3119         ClearPagePrivate2(page);
3120         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3121                                             end - start + 1, uptodate))
3122                 return;
3123
3124         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3125                 wq = fs_info->endio_freespace_worker;
3126                 func = btrfs_freespace_write_helper;
3127         } else {
3128                 wq = fs_info->endio_write_workers;
3129                 func = btrfs_endio_write_helper;
3130         }
3131
3132         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3133                         NULL);
3134         btrfs_queue_work(wq, &ordered_extent->work);
3135 }
3136
3137 static int __readpage_endio_check(struct inode *inode,
3138                                   struct btrfs_io_bio *io_bio,
3139                                   int icsum, struct page *page,
3140                                   int pgoff, u64 start, size_t len)
3141 {
3142         char *kaddr;
3143         u32 csum_expected;
3144         u32 csum = ~(u32)0;
3145
3146         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3147
3148         kaddr = kmap_atomic(page);
3149         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3150         btrfs_csum_final(csum, (u8 *)&csum);
3151         if (csum != csum_expected)
3152                 goto zeroit;
3153
3154         kunmap_atomic(kaddr);
3155         return 0;
3156 zeroit:
3157         btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3158                                     io_bio->mirror_num);
3159         memset(kaddr + pgoff, 1, len);
3160         flush_dcache_page(page);
3161         kunmap_atomic(kaddr);
3162         if (csum_expected == 0)
3163                 return 0;
3164         return -EIO;
3165 }
3166
3167 /*
3168  * when reads are done, we need to check csums to verify the data is correct
3169  * if there's a match, we allow the bio to finish.  If not, the code in
3170  * extent_io.c will try to find good copies for us.
3171  */
3172 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3173                                       u64 phy_offset, struct page *page,
3174                                       u64 start, u64 end, int mirror)
3175 {
3176         size_t offset = start - page_offset(page);
3177         struct inode *inode = page->mapping->host;
3178         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3179         struct btrfs_root *root = BTRFS_I(inode)->root;
3180
3181         if (PageChecked(page)) {
3182                 ClearPageChecked(page);
3183                 return 0;
3184         }
3185
3186         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3187                 return 0;
3188
3189         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3190             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3191                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3192                 return 0;
3193         }
3194
3195         phy_offset >>= inode->i_sb->s_blocksize_bits;
3196         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3197                                       start, (size_t)(end - start + 1));
3198 }
3199
3200 void btrfs_add_delayed_iput(struct inode *inode)
3201 {
3202         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3203         struct btrfs_inode *binode = BTRFS_I(inode);
3204
3205         if (atomic_add_unless(&inode->i_count, -1, 1))
3206                 return;
3207
3208         spin_lock(&fs_info->delayed_iput_lock);
3209         if (binode->delayed_iput_count == 0) {
3210                 ASSERT(list_empty(&binode->delayed_iput));
3211                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3212         } else {
3213                 binode->delayed_iput_count++;
3214         }
3215         spin_unlock(&fs_info->delayed_iput_lock);
3216 }
3217
3218 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3219 {
3220
3221         spin_lock(&fs_info->delayed_iput_lock);
3222         while (!list_empty(&fs_info->delayed_iputs)) {
3223                 struct btrfs_inode *inode;
3224
3225                 inode = list_first_entry(&fs_info->delayed_iputs,
3226                                 struct btrfs_inode, delayed_iput);
3227                 if (inode->delayed_iput_count) {
3228                         inode->delayed_iput_count--;
3229                         list_move_tail(&inode->delayed_iput,
3230                                         &fs_info->delayed_iputs);
3231                 } else {
3232                         list_del_init(&inode->delayed_iput);
3233                 }
3234                 spin_unlock(&fs_info->delayed_iput_lock);
3235                 iput(&inode->vfs_inode);
3236                 spin_lock(&fs_info->delayed_iput_lock);
3237         }
3238         spin_unlock(&fs_info->delayed_iput_lock);
3239 }
3240
3241 /*
3242  * This is called in transaction commit time. If there are no orphan
3243  * files in the subvolume, it removes orphan item and frees block_rsv
3244  * structure.
3245  */
3246 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3247                               struct btrfs_root *root)
3248 {
3249         struct btrfs_fs_info *fs_info = root->fs_info;
3250         struct btrfs_block_rsv *block_rsv;
3251         int ret;
3252
3253         if (atomic_read(&root->orphan_inodes) ||
3254             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3255                 return;
3256
3257         spin_lock(&root->orphan_lock);
3258         if (atomic_read(&root->orphan_inodes)) {
3259                 spin_unlock(&root->orphan_lock);
3260                 return;
3261         }
3262
3263         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3264                 spin_unlock(&root->orphan_lock);
3265                 return;
3266         }
3267
3268         block_rsv = root->orphan_block_rsv;
3269         root->orphan_block_rsv = NULL;
3270         spin_unlock(&root->orphan_lock);
3271
3272         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3273             btrfs_root_refs(&root->root_item) > 0) {
3274                 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3275                                             root->root_key.objectid);
3276                 if (ret)
3277                         btrfs_abort_transaction(trans, ret);
3278                 else
3279                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3280                                   &root->state);
3281         }
3282
3283         if (block_rsv) {
3284                 WARN_ON(block_rsv->size > 0);
3285                 btrfs_free_block_rsv(fs_info, block_rsv);
3286         }
3287 }
3288
3289 /*
3290  * This creates an orphan entry for the given inode in case something goes
3291  * wrong in the middle of an unlink/truncate.
3292  *
3293  * NOTE: caller of this function should reserve 5 units of metadata for
3294  *       this function.
3295  */
3296 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3297                 struct btrfs_inode *inode)
3298 {
3299         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3300         struct btrfs_root *root = inode->root;
3301         struct btrfs_block_rsv *block_rsv = NULL;
3302         int reserve = 0;
3303         int insert = 0;
3304         int ret;
3305
3306         if (!root->orphan_block_rsv) {
3307                 block_rsv = btrfs_alloc_block_rsv(fs_info,
3308                                                   BTRFS_BLOCK_RSV_TEMP);
3309                 if (!block_rsv)
3310                         return -ENOMEM;
3311         }
3312
3313         spin_lock(&root->orphan_lock);
3314         if (!root->orphan_block_rsv) {
3315                 root->orphan_block_rsv = block_rsv;
3316         } else if (block_rsv) {
3317                 btrfs_free_block_rsv(fs_info, block_rsv);
3318                 block_rsv = NULL;
3319         }
3320
3321         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3322                               &inode->runtime_flags)) {
3323 #if 0
3324                 /*
3325                  * For proper ENOSPC handling, we should do orphan
3326                  * cleanup when mounting. But this introduces backward
3327                  * compatibility issue.
3328                  */
3329                 if (!xchg(&root->orphan_item_inserted, 1))
3330                         insert = 2;
3331                 else
3332                         insert = 1;
3333 #endif
3334                 insert = 1;
3335                 atomic_inc(&root->orphan_inodes);
3336         }
3337
3338         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3339                               &inode->runtime_flags))
3340                 reserve = 1;
3341         spin_unlock(&root->orphan_lock);
3342
3343         /* grab metadata reservation from transaction handle */
3344         if (reserve) {
3345                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3346                 ASSERT(!ret);
3347                 if (ret) {
3348                         atomic_dec(&root->orphan_inodes);
3349                         clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3350                                   &inode->runtime_flags);
3351                         if (insert)
3352                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3353                                           &inode->runtime_flags);
3354                         return ret;
3355                 }
3356         }
3357
3358         /* insert an orphan item to track this unlinked/truncated file */
3359         if (insert >= 1) {
3360                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3361                 if (ret) {
3362                         atomic_dec(&root->orphan_inodes);
3363                         if (reserve) {
3364                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3365                                           &inode->runtime_flags);
3366                                 btrfs_orphan_release_metadata(inode);
3367                         }
3368                         if (ret != -EEXIST) {
3369                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3370                                           &inode->runtime_flags);
3371                                 btrfs_abort_transaction(trans, ret);
3372                                 return ret;
3373                         }
3374                 }
3375                 ret = 0;
3376         }
3377
3378         /* insert an orphan item to track subvolume contains orphan files */
3379         if (insert >= 2) {
3380                 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3381                                                root->root_key.objectid);
3382                 if (ret && ret != -EEXIST) {
3383                         btrfs_abort_transaction(trans, ret);
3384                         return ret;
3385                 }
3386         }
3387         return 0;
3388 }
3389
3390 /*
3391  * We have done the truncate/delete so we can go ahead and remove the orphan
3392  * item for this particular inode.
3393  */
3394 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3395                             struct btrfs_inode *inode)
3396 {
3397         struct btrfs_root *root = inode->root;
3398         int delete_item = 0;
3399         int release_rsv = 0;
3400         int ret = 0;
3401
3402         spin_lock(&root->orphan_lock);
3403         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3404                                &inode->runtime_flags))
3405                 delete_item = 1;
3406
3407         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3408                                &inode->runtime_flags))
3409                 release_rsv = 1;
3410         spin_unlock(&root->orphan_lock);
3411
3412         if (delete_item) {
3413                 atomic_dec(&root->orphan_inodes);
3414                 if (trans)
3415                         ret = btrfs_del_orphan_item(trans, root,
3416                                                     btrfs_ino(inode));
3417         }
3418
3419         if (release_rsv)
3420                 btrfs_orphan_release_metadata(inode);
3421
3422         return ret;
3423 }
3424
3425 /*
3426  * this cleans up any orphans that may be left on the list from the last use
3427  * of this root.
3428  */
3429 int btrfs_orphan_cleanup(struct btrfs_root *root)
3430 {
3431         struct btrfs_fs_info *fs_info = root->fs_info;
3432         struct btrfs_path *path;
3433         struct extent_buffer *leaf;
3434         struct btrfs_key key, found_key;
3435         struct btrfs_trans_handle *trans;
3436         struct inode *inode;
3437         u64 last_objectid = 0;
3438         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3439
3440         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3441                 return 0;
3442
3443         path = btrfs_alloc_path();
3444         if (!path) {
3445                 ret = -ENOMEM;
3446                 goto out;
3447         }
3448         path->reada = READA_BACK;
3449
3450         key.objectid = BTRFS_ORPHAN_OBJECTID;
3451         key.type = BTRFS_ORPHAN_ITEM_KEY;
3452         key.offset = (u64)-1;
3453
3454         while (1) {
3455                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3456                 if (ret < 0)
3457                         goto out;
3458
3459                 /*
3460                  * if ret == 0 means we found what we were searching for, which
3461                  * is weird, but possible, so only screw with path if we didn't
3462                  * find the key and see if we have stuff that matches
3463                  */
3464                 if (ret > 0) {
3465                         ret = 0;
3466                         if (path->slots[0] == 0)
3467                                 break;
3468                         path->slots[0]--;
3469                 }
3470
3471                 /* pull out the item */
3472                 leaf = path->nodes[0];
3473                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3474
3475                 /* make sure the item matches what we want */
3476                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3477                         break;
3478                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3479                         break;
3480
3481                 /* release the path since we're done with it */
3482                 btrfs_release_path(path);
3483
3484                 /*
3485                  * this is where we are basically btrfs_lookup, without the
3486                  * crossing root thing.  we store the inode number in the
3487                  * offset of the orphan item.
3488                  */
3489
3490                 if (found_key.offset == last_objectid) {
3491                         btrfs_err(fs_info,
3492                                   "Error removing orphan entry, stopping orphan cleanup");
3493                         ret = -EINVAL;
3494                         goto out;
3495                 }
3496
3497                 last_objectid = found_key.offset;
3498
3499                 found_key.objectid = found_key.offset;
3500                 found_key.type = BTRFS_INODE_ITEM_KEY;
3501                 found_key.offset = 0;
3502                 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3503                 ret = PTR_ERR_OR_ZERO(inode);
3504                 if (ret && ret != -ENOENT)
3505                         goto out;
3506
3507                 if (ret == -ENOENT && root == fs_info->tree_root) {
3508                         struct btrfs_root *dead_root;
3509                         struct btrfs_fs_info *fs_info = root->fs_info;
3510                         int is_dead_root = 0;
3511
3512                         /*
3513                          * this is an orphan in the tree root. Currently these
3514                          * could come from 2 sources:
3515                          *  a) a snapshot deletion in progress
3516                          *  b) a free space cache inode
3517                          * We need to distinguish those two, as the snapshot
3518                          * orphan must not get deleted.
3519                          * find_dead_roots already ran before us, so if this
3520                          * is a snapshot deletion, we should find the root
3521                          * in the dead_roots list
3522                          */
3523                         spin_lock(&fs_info->trans_lock);
3524                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3525                                             root_list) {
3526                                 if (dead_root->root_key.objectid ==
3527                                     found_key.objectid) {
3528                                         is_dead_root = 1;
3529                                         break;
3530                                 }
3531                         }
3532                         spin_unlock(&fs_info->trans_lock);
3533                         if (is_dead_root) {
3534                                 /* prevent this orphan from being found again */
3535                                 key.offset = found_key.objectid - 1;
3536                                 continue;
3537                         }
3538                 }
3539                 /*
3540                  * Inode is already gone but the orphan item is still there,
3541                  * kill the orphan item.
3542                  */
3543                 if (ret == -ENOENT) {
3544                         trans = btrfs_start_transaction(root, 1);
3545                         if (IS_ERR(trans)) {
3546                                 ret = PTR_ERR(trans);
3547                                 goto out;
3548                         }
3549                         btrfs_debug(fs_info, "auto deleting %Lu",
3550                                     found_key.objectid);
3551                         ret = btrfs_del_orphan_item(trans, root,
3552                                                     found_key.objectid);
3553                         btrfs_end_transaction(trans);
3554                         if (ret)
3555                                 goto out;
3556                         continue;
3557                 }
3558
3559                 /*
3560                  * add this inode to the orphan list so btrfs_orphan_del does
3561                  * the proper thing when we hit it
3562                  */
3563                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3564                         &BTRFS_I(inode)->runtime_flags);
3565                 atomic_inc(&root->orphan_inodes);
3566
3567                 /* if we have links, this was a truncate, lets do that */
3568                 if (inode->i_nlink) {
3569                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3570                                 iput(inode);
3571                                 continue;
3572                         }
3573                         nr_truncate++;
3574
3575                         /* 1 for the orphan item deletion. */
3576                         trans = btrfs_start_transaction(root, 1);
3577                         if (IS_ERR(trans)) {
3578                                 iput(inode);
3579                                 ret = PTR_ERR(trans);
3580                                 goto out;
3581                         }
3582                         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3583                         btrfs_end_transaction(trans);
3584                         if (ret) {
3585                                 iput(inode);
3586                                 goto out;
3587                         }
3588
3589                         ret = btrfs_truncate(inode);
3590                         if (ret)
3591                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
3592                 } else {
3593                         nr_unlink++;
3594                 }
3595
3596                 /* this will do delete_inode and everything for us */
3597                 iput(inode);
3598                 if (ret)
3599                         goto out;
3600         }
3601         /* release the path since we're done with it */
3602         btrfs_release_path(path);
3603
3604         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3605
3606         if (root->orphan_block_rsv)
3607                 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3608                                         (u64)-1);
3609
3610         if (root->orphan_block_rsv ||
3611             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3612                 trans = btrfs_join_transaction(root);
3613                 if (!IS_ERR(trans))
3614                         btrfs_end_transaction(trans);
3615         }
3616
3617         if (nr_unlink)
3618                 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3619         if (nr_truncate)
3620                 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3621
3622 out:
3623         if (ret)
3624                 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3625         btrfs_free_path(path);
3626         return ret;
3627 }
3628
3629 /*
3630  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3631  * don't find any xattrs, we know there can't be any acls.
3632  *
3633  * slot is the slot the inode is in, objectid is the objectid of the inode
3634  */
3635 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3636                                           int slot, u64 objectid,
3637                                           int *first_xattr_slot)
3638 {
3639         u32 nritems = btrfs_header_nritems(leaf);
3640         struct btrfs_key found_key;
3641         static u64 xattr_access = 0;
3642         static u64 xattr_default = 0;
3643         int scanned = 0;
3644
3645         if (!xattr_access) {
3646                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3647                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3648                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3649                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3650         }
3651
3652         slot++;
3653         *first_xattr_slot = -1;
3654         while (slot < nritems) {
3655                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3656
3657                 /* we found a different objectid, there must not be acls */
3658                 if (found_key.objectid != objectid)
3659                         return 0;
3660
3661                 /* we found an xattr, assume we've got an acl */
3662                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3663                         if (*first_xattr_slot == -1)
3664                                 *first_xattr_slot = slot;
3665                         if (found_key.offset == xattr_access ||
3666                             found_key.offset == xattr_default)
3667                                 return 1;
3668                 }
3669
3670                 /*
3671                  * we found a key greater than an xattr key, there can't
3672                  * be any acls later on
3673                  */
3674                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3675                         return 0;
3676
3677                 slot++;
3678                 scanned++;
3679
3680                 /*
3681                  * it goes inode, inode backrefs, xattrs, extents,
3682                  * so if there are a ton of hard links to an inode there can
3683                  * be a lot of backrefs.  Don't waste time searching too hard,
3684                  * this is just an optimization
3685                  */
3686                 if (scanned >= 8)
3687                         break;
3688         }
3689         /* we hit the end of the leaf before we found an xattr or
3690          * something larger than an xattr.  We have to assume the inode
3691          * has acls
3692          */
3693         if (*first_xattr_slot == -1)
3694                 *first_xattr_slot = slot;
3695         return 1;
3696 }
3697
3698 /*
3699  * read an inode from the btree into the in-memory inode
3700  */
3701 static int btrfs_read_locked_inode(struct inode *inode)
3702 {
3703         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3704         struct btrfs_path *path;
3705         struct extent_buffer *leaf;
3706         struct btrfs_inode_item *inode_item;
3707         struct btrfs_root *root = BTRFS_I(inode)->root;
3708         struct btrfs_key location;
3709         unsigned long ptr;
3710         int maybe_acls;
3711         u32 rdev;
3712         int ret;
3713         bool filled = false;
3714         int first_xattr_slot;
3715
3716         ret = btrfs_fill_inode(inode, &rdev);
3717         if (!ret)
3718                 filled = true;
3719
3720         path = btrfs_alloc_path();
3721         if (!path) {
3722                 ret = -ENOMEM;
3723                 goto make_bad;
3724         }
3725
3726         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3727
3728         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3729         if (ret) {
3730                 if (ret > 0)
3731                         ret = -ENOENT;
3732                 goto make_bad;
3733         }
3734
3735         leaf = path->nodes[0];
3736
3737         if (filled)
3738                 goto cache_index;
3739
3740         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3741                                     struct btrfs_inode_item);
3742         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3743         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3744         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3745         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3746         btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3747
3748         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3749         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3750
3751         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3752         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3753
3754         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3755         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3756
3757         BTRFS_I(inode)->i_otime.tv_sec =
3758                 btrfs_timespec_sec(leaf, &inode_item->otime);
3759         BTRFS_I(inode)->i_otime.tv_nsec =
3760                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3761
3762         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3763         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3764         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3765
3766         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3767         inode->i_generation = BTRFS_I(inode)->generation;
3768         inode->i_rdev = 0;
3769         rdev = btrfs_inode_rdev(leaf, inode_item);
3770
3771         BTRFS_I(inode)->index_cnt = (u64)-1;
3772         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3773
3774 cache_index:
3775         /*
3776          * If we were modified in the current generation and evicted from memory
3777          * and then re-read we need to do a full sync since we don't have any
3778          * idea about which extents were modified before we were evicted from
3779          * cache.
3780          *
3781          * This is required for both inode re-read from disk and delayed inode
3782          * in delayed_nodes_tree.
3783          */
3784         if (BTRFS_I(inode)->last_trans == fs_info->generation)
3785                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3786                         &BTRFS_I(inode)->runtime_flags);
3787
3788         /*
3789          * We don't persist the id of the transaction where an unlink operation
3790          * against the inode was last made. So here we assume the inode might
3791          * have been evicted, and therefore the exact value of last_unlink_trans
3792          * lost, and set it to last_trans to avoid metadata inconsistencies
3793          * between the inode and its parent if the inode is fsync'ed and the log
3794          * replayed. For example, in the scenario:
3795          *
3796          * touch mydir/foo
3797          * ln mydir/foo mydir/bar
3798          * sync
3799          * unlink mydir/bar
3800          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3801          * xfs_io -c fsync mydir/foo
3802          * <power failure>
3803          * mount fs, triggers fsync log replay
3804          *
3805          * We must make sure that when we fsync our inode foo we also log its
3806          * parent inode, otherwise after log replay the parent still has the
3807          * dentry with the "bar" name but our inode foo has a link count of 1
3808          * and doesn't have an inode ref with the name "bar" anymore.
3809          *
3810          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3811          * but it guarantees correctness at the expense of occasional full
3812          * transaction commits on fsync if our inode is a directory, or if our
3813          * inode is not a directory, logging its parent unnecessarily.
3814          */
3815         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3816
3817         path->slots[0]++;
3818         if (inode->i_nlink != 1 ||
3819             path->slots[0] >= btrfs_header_nritems(leaf))
3820                 goto cache_acl;
3821
3822         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3823         if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3824                 goto cache_acl;
3825
3826         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3827         if (location.type == BTRFS_INODE_REF_KEY) {
3828                 struct btrfs_inode_ref *ref;
3829
3830                 ref = (struct btrfs_inode_ref *)ptr;
3831                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3832         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3833                 struct btrfs_inode_extref *extref;
3834
3835                 extref = (struct btrfs_inode_extref *)ptr;
3836                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3837                                                                      extref);
3838         }
3839 cache_acl:
3840         /*
3841          * try to precache a NULL acl entry for files that don't have
3842          * any xattrs or acls
3843          */
3844         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3845                         btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3846         if (first_xattr_slot != -1) {
3847                 path->slots[0] = first_xattr_slot;
3848                 ret = btrfs_load_inode_props(inode, path);
3849                 if (ret)
3850                         btrfs_err(fs_info,
3851                                   "error loading props for ino %llu (root %llu): %d",
3852                                   btrfs_ino(BTRFS_I(inode)),
3853                                   root->root_key.objectid, ret);
3854         }
3855         btrfs_free_path(path);
3856
3857         if (!maybe_acls)
3858                 cache_no_acl(inode);
3859
3860         switch (inode->i_mode & S_IFMT) {
3861         case S_IFREG:
3862                 inode->i_mapping->a_ops = &btrfs_aops;
3863                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3864                 inode->i_fop = &btrfs_file_operations;
3865                 inode->i_op = &btrfs_file_inode_operations;
3866                 break;
3867         case S_IFDIR:
3868                 inode->i_fop = &btrfs_dir_file_operations;
3869                 inode->i_op = &btrfs_dir_inode_operations;
3870                 break;
3871         case S_IFLNK:
3872                 inode->i_op = &btrfs_symlink_inode_operations;
3873                 inode_nohighmem(inode);
3874                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3875                 break;
3876         default:
3877                 inode->i_op = &btrfs_special_inode_operations;
3878                 init_special_inode(inode, inode->i_mode, rdev);
3879                 break;
3880         }
3881
3882         btrfs_update_iflags(inode);
3883         return 0;
3884
3885 make_bad:
3886         btrfs_free_path(path);
3887         make_bad_inode(inode);
3888         return ret;
3889 }
3890
3891 /*
3892  * given a leaf and an inode, copy the inode fields into the leaf
3893  */
3894 static void fill_inode_item(struct btrfs_trans_handle *trans,
3895                             struct extent_buffer *leaf,
3896                             struct btrfs_inode_item *item,
3897                             struct inode *inode)
3898 {
3899         struct btrfs_map_token token;
3900
3901         btrfs_init_map_token(&token);
3902
3903         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3904         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3905         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3906                                    &token);
3907         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3908         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3909
3910         btrfs_set_token_timespec_sec(leaf, &item->atime,
3911                                      inode->i_atime.tv_sec, &token);
3912         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3913                                       inode->i_atime.tv_nsec, &token);
3914
3915         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3916                                      inode->i_mtime.tv_sec, &token);
3917         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3918                                       inode->i_mtime.tv_nsec, &token);
3919
3920         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3921                                      inode->i_ctime.tv_sec, &token);
3922         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3923                                       inode->i_ctime.tv_nsec, &token);
3924
3925         btrfs_set_token_timespec_sec(leaf, &item->otime,
3926                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3927         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3928                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3929
3930         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3931                                      &token);
3932         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3933                                          &token);
3934         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3935         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3936         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3937         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3938         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3939 }
3940
3941 /*
3942  * copy everything in the in-memory inode into the btree.
3943  */
3944 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3945                                 struct btrfs_root *root, struct inode *inode)
3946 {
3947         struct btrfs_inode_item *inode_item;
3948         struct btrfs_path *path;
3949         struct extent_buffer *leaf;
3950         int ret;
3951
3952         path = btrfs_alloc_path();
3953         if (!path)
3954                 return -ENOMEM;
3955
3956         path->leave_spinning = 1;
3957         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3958                                  1);
3959         if (ret) {
3960                 if (ret > 0)
3961                         ret = -ENOENT;
3962                 goto failed;
3963         }
3964
3965         leaf = path->nodes[0];
3966         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3967                                     struct btrfs_inode_item);
3968
3969         fill_inode_item(trans, leaf, inode_item, inode);
3970         btrfs_mark_buffer_dirty(leaf);
3971         btrfs_set_inode_last_trans(trans, inode);
3972         ret = 0;
3973 failed:
3974         btrfs_free_path(path);
3975         return ret;
3976 }
3977
3978 /*
3979  * copy everything in the in-memory inode into the btree.
3980  */
3981 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3982                                 struct btrfs_root *root, struct inode *inode)
3983 {
3984         struct btrfs_fs_info *fs_info = root->fs_info;
3985         int ret;
3986
3987         /*
3988          * If the inode is a free space inode, we can deadlock during commit
3989          * if we put it into the delayed code.
3990          *
3991          * The data relocation inode should also be directly updated
3992          * without delay
3993          */
3994         if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3995             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3996             && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3997                 btrfs_update_root_times(trans, root);
3998
3999                 ret = btrfs_delayed_update_inode(trans, root, inode);
4000                 if (!ret)
4001                         btrfs_set_inode_last_trans(trans, inode);
4002                 return ret;
4003         }
4004
4005         return btrfs_update_inode_item(trans, root, inode);
4006 }
4007
4008 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4009                                          struct btrfs_root *root,
4010                                          struct inode *inode)
4011 {
4012         int ret;
4013
4014         ret = btrfs_update_inode(trans, root, inode);
4015         if (ret == -ENOSPC)
4016                 return btrfs_update_inode_item(trans, root, inode);
4017         return ret;
4018 }
4019
4020 /*
4021  * unlink helper that gets used here in inode.c and in the tree logging
4022  * recovery code.  It remove a link in a directory with a given name, and
4023  * also drops the back refs in the inode to the directory
4024  */
4025 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4026                                 struct btrfs_root *root,
4027                                 struct btrfs_inode *dir,
4028                                 struct btrfs_inode *inode,
4029                                 const char *name, int name_len)
4030 {
4031         struct btrfs_fs_info *fs_info = root->fs_info;
4032         struct btrfs_path *path;
4033         int ret = 0;
4034         struct extent_buffer *leaf;
4035         struct btrfs_dir_item *di;
4036         struct btrfs_key key;
4037         u64 index;
4038         u64 ino = btrfs_ino(inode);
4039         u64 dir_ino = btrfs_ino(dir);
4040
4041         path = btrfs_alloc_path();
4042         if (!path) {
4043                 ret = -ENOMEM;
4044                 goto out;
4045         }
4046
4047         path->leave_spinning = 1;
4048         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4049                                     name, name_len, -1);
4050         if (IS_ERR(di)) {
4051                 ret = PTR_ERR(di);
4052                 goto err;
4053         }
4054         if (!di) {
4055                 ret = -ENOENT;
4056                 goto err;
4057         }
4058         leaf = path->nodes[0];
4059         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4060         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4061         if (ret)
4062                 goto err;
4063         btrfs_release_path(path);
4064
4065         /*
4066          * If we don't have dir index, we have to get it by looking up
4067          * the inode ref, since we get the inode ref, remove it directly,
4068          * it is unnecessary to do delayed deletion.
4069          *
4070          * But if we have dir index, needn't search inode ref to get it.
4071          * Since the inode ref is close to the inode item, it is better
4072          * that we delay to delete it, and just do this deletion when
4073          * we update the inode item.
4074          */
4075         if (inode->dir_index) {
4076                 ret = btrfs_delayed_delete_inode_ref(inode);
4077                 if (!ret) {
4078                         index = inode->dir_index;
4079                         goto skip_backref;
4080                 }
4081         }
4082
4083         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4084                                   dir_ino, &index);
4085         if (ret) {
4086                 btrfs_info(fs_info,
4087                         "failed to delete reference to %.*s, inode %llu parent %llu",
4088                         name_len, name, ino, dir_ino);
4089                 btrfs_abort_transaction(trans, ret);
4090                 goto err;
4091         }
4092 skip_backref:
4093         ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4094         if (ret) {
4095                 btrfs_abort_transaction(trans, ret);
4096                 goto err;
4097         }
4098
4099         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4100                         dir_ino);
4101         if (ret != 0 && ret != -ENOENT) {
4102                 btrfs_abort_transaction(trans, ret);
4103                 goto err;
4104         }
4105
4106         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4107                         index);
4108         if (ret == -ENOENT)
4109                 ret = 0;
4110         else if (ret)
4111                 btrfs_abort_transaction(trans, ret);
4112 err:
4113         btrfs_free_path(path);
4114         if (ret)
4115                 goto out;
4116
4117         btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4118         inode_inc_iversion(&inode->vfs_inode);
4119         inode_inc_iversion(&dir->vfs_inode);
4120         inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4121                 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4122         ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4123 out:
4124         return ret;
4125 }
4126
4127 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4128                        struct btrfs_root *root,
4129                        struct btrfs_inode *dir, struct btrfs_inode *inode,
4130                        const char *name, int name_len)
4131 {
4132         int ret;
4133         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4134         if (!ret) {
4135                 drop_nlink(&inode->vfs_inode);
4136                 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4137         }
4138         return ret;
4139 }
4140
4141 /*
4142  * helper to start transaction for unlink and rmdir.
4143  *
4144  * unlink and rmdir are special in btrfs, they do not always free space, so
4145  * if we cannot make our reservations the normal way try and see if there is
4146  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4147  * allow the unlink to occur.
4148  */
4149 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4150 {
4151         struct btrfs_root *root = BTRFS_I(dir)->root;
4152
4153         /*
4154          * 1 for the possible orphan item
4155          * 1 for the dir item
4156          * 1 for the dir index
4157          * 1 for the inode ref
4158          * 1 for the inode
4159          */
4160         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4161 }
4162
4163 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4164 {
4165         struct btrfs_root *root = BTRFS_I(dir)->root;
4166         struct btrfs_trans_handle *trans;
4167         struct inode *inode = d_inode(dentry);
4168         int ret;
4169
4170         trans = __unlink_start_trans(dir);
4171         if (IS_ERR(trans))
4172                 return PTR_ERR(trans);
4173
4174         btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4175                         0);
4176
4177         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4178                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4179                         dentry->d_name.len);
4180         if (ret)
4181                 goto out;
4182
4183         if (inode->i_nlink == 0) {
4184                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4185                 if (ret)
4186                         goto out;
4187         }
4188
4189 out:
4190         btrfs_end_transaction(trans);
4191         btrfs_btree_balance_dirty(root->fs_info);
4192         return ret;
4193 }
4194
4195 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4196                         struct btrfs_root *root,
4197                         struct inode *dir, u64 objectid,
4198                         const char *name, int name_len)
4199 {
4200         struct btrfs_fs_info *fs_info = root->fs_info;
4201         struct btrfs_path *path;
4202         struct extent_buffer *leaf;
4203         struct btrfs_dir_item *di;
4204         struct btrfs_key key;
4205         u64 index;
4206         int ret;
4207         u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4208
4209         path = btrfs_alloc_path();
4210         if (!path)
4211                 return -ENOMEM;
4212
4213         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4214                                    name, name_len, -1);
4215         if (IS_ERR_OR_NULL(di)) {
4216                 if (!di)
4217                         ret = -ENOENT;
4218                 else
4219                         ret = PTR_ERR(di);
4220                 goto out;
4221         }
4222
4223         leaf = path->nodes[0];
4224         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4225         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4226         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4227         if (ret) {
4228                 btrfs_abort_transaction(trans, ret);
4229                 goto out;
4230         }
4231         btrfs_release_path(path);
4232
4233         ret = btrfs_del_root_ref(trans, fs_info, objectid,
4234                                  root->root_key.objectid, dir_ino,
4235                                  &index, name, name_len);
4236         if (ret < 0) {
4237                 if (ret != -ENOENT) {
4238                         btrfs_abort_transaction(trans, ret);
4239                         goto out;
4240                 }
4241                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4242                                                  name, name_len);
4243                 if (IS_ERR_OR_NULL(di)) {
4244                         if (!di)
4245                                 ret = -ENOENT;
4246                         else
4247                                 ret = PTR_ERR(di);
4248                         btrfs_abort_transaction(trans, ret);
4249                         goto out;
4250                 }
4251
4252                 leaf = path->nodes[0];
4253                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4254                 btrfs_release_path(path);
4255                 index = key.offset;
4256         }
4257         btrfs_release_path(path);
4258
4259         ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4260         if (ret) {
4261                 btrfs_abort_transaction(trans, ret);
4262                 goto out;
4263         }
4264
4265         btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4266         inode_inc_iversion(dir);
4267         dir->i_mtime = dir->i_ctime = current_time(dir);
4268         ret = btrfs_update_inode_fallback(trans, root, dir);
4269         if (ret)
4270                 btrfs_abort_transaction(trans, ret);
4271 out:
4272         btrfs_free_path(path);
4273         return ret;
4274 }
4275
4276 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4277 {
4278         struct inode *inode = d_inode(dentry);
4279         int err = 0;
4280         struct btrfs_root *root = BTRFS_I(dir)->root;
4281         struct btrfs_trans_handle *trans;
4282         u64 last_unlink_trans;
4283
4284         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4285                 return -ENOTEMPTY;
4286         if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4287                 return -EPERM;
4288
4289         trans = __unlink_start_trans(dir);
4290         if (IS_ERR(trans))
4291                 return PTR_ERR(trans);
4292
4293         if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4294                 err = btrfs_unlink_subvol(trans, root, dir,
4295                                           BTRFS_I(inode)->location.objectid,
4296                                           dentry->d_name.name,
4297                                           dentry->d_name.len);
4298                 goto out;
4299         }
4300
4301         err = btrfs_orphan_add(trans, BTRFS_I(inode));
4302         if (err)
4303                 goto out;
4304
4305         last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4306
4307         /* now the directory is empty */
4308         err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4309                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4310                         dentry->d_name.len);
4311         if (!err) {
4312                 btrfs_i_size_write(BTRFS_I(inode), 0);
4313                 /*
4314                  * Propagate the last_unlink_trans value of the deleted dir to
4315                  * its parent directory. This is to prevent an unrecoverable
4316                  * log tree in the case we do something like this:
4317                  * 1) create dir foo
4318                  * 2) create snapshot under dir foo
4319                  * 3) delete the snapshot
4320                  * 4) rmdir foo
4321                  * 5) mkdir foo
4322                  * 6) fsync foo or some file inside foo
4323                  */
4324                 if (last_unlink_trans >= trans->transid)
4325                         BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4326         }
4327 out:
4328         btrfs_end_transaction(trans);
4329         btrfs_btree_balance_dirty(root->fs_info);
4330
4331         return err;
4332 }
4333
4334 static int truncate_space_check(struct btrfs_trans_handle *trans,
4335                                 struct btrfs_root *root,
4336                                 u64 bytes_deleted)
4337 {
4338         struct btrfs_fs_info *fs_info = root->fs_info;
4339         int ret;
4340
4341         /*
4342          * This is only used to apply pressure to the enospc system, we don't
4343          * intend to use this reservation at all.
4344          */
4345         bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4346         bytes_deleted *= fs_info->nodesize;
4347         ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4348                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4349         if (!ret) {
4350                 trace_btrfs_space_reservation(fs_info, "transaction",
4351                                               trans->transid,
4352                                               bytes_deleted, 1);
4353                 trans->bytes_reserved += bytes_deleted;
4354         }
4355         return ret;
4356
4357 }
4358
4359 static int truncate_inline_extent(struct inode *inode,
4360                                   struct btrfs_path *path,
4361                                   struct btrfs_key *found_key,
4362                                   const u64 item_end,
4363                                   const u64 new_size)
4364 {
4365         struct extent_buffer *leaf = path->nodes[0];
4366         int slot = path->slots[0];
4367         struct btrfs_file_extent_item *fi;
4368         u32 size = (u32)(new_size - found_key->offset);
4369         struct btrfs_root *root = BTRFS_I(inode)->root;
4370
4371         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4372
4373         if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4374                 loff_t offset = new_size;
4375                 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4376
4377                 /*
4378                  * Zero out the remaining of the last page of our inline extent,
4379                  * instead of directly truncating our inline extent here - that
4380                  * would be much more complex (decompressing all the data, then
4381                  * compressing the truncated data, which might be bigger than
4382                  * the size of the inline extent, resize the extent, etc).
4383                  * We release the path because to get the page we might need to
4384                  * read the extent item from disk (data not in the page cache).
4385                  */
4386                 btrfs_release_path(path);
4387                 return btrfs_truncate_block(inode, offset, page_end - offset,
4388                                         0);
4389         }
4390
4391         btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4392         size = btrfs_file_extent_calc_inline_size(size);
4393         btrfs_truncate_item(root->fs_info, path, size, 1);
4394
4395         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4396                 inode_sub_bytes(inode, item_end + 1 - new_size);
4397
4398         return 0;
4399 }
4400
4401 /*
4402  * this can truncate away extent items, csum items and directory items.
4403  * It starts at a high offset and removes keys until it can't find
4404  * any higher than new_size
4405  *
4406  * csum items that cross the new i_size are truncated to the new size
4407  * as well.
4408  *
4409  * min_type is the minimum key type to truncate down to.  If set to 0, this
4410  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4411  */
4412 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4413                                struct btrfs_root *root,
4414                                struct inode *inode,
4415                                u64 new_size, u32 min_type)
4416 {
4417         struct btrfs_fs_info *fs_info = root->fs_info;
4418         struct btrfs_path *path;
4419         struct extent_buffer *leaf;
4420         struct btrfs_file_extent_item *fi;
4421         struct btrfs_key key;
4422         struct btrfs_key found_key;
4423         u64 extent_start = 0;
4424         u64 extent_num_bytes = 0;
4425         u64 extent_offset = 0;
4426         u64 item_end = 0;
4427         u64 last_size = new_size;
4428         u32 found_type = (u8)-1;
4429         int found_extent;
4430         int del_item;
4431         int pending_del_nr = 0;
4432         int pending_del_slot = 0;
4433         int extent_type = -1;
4434         int ret;
4435         int err = 0;
4436         u64 ino = btrfs_ino(BTRFS_I(inode));
4437         u64 bytes_deleted = 0;
4438         bool be_nice = 0;
4439         bool should_throttle = 0;
4440         bool should_end = 0;
4441
4442         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4443
4444         /*
4445          * for non-free space inodes and ref cows, we want to back off from
4446          * time to time
4447          */
4448         if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4449             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4450                 be_nice = 1;
4451
4452         path = btrfs_alloc_path();
4453         if (!path)
4454                 return -ENOMEM;
4455         path->reada = READA_BACK;
4456
4457         /*
4458          * We want to drop from the next block forward in case this new size is
4459          * not block aligned since we will be keeping the last block of the
4460          * extent just the way it is.
4461          */
4462         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4463             root == fs_info->tree_root)
4464                 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4465                                         fs_info->sectorsize),
4466                                         (u64)-1, 0);
4467
4468         /*
4469          * This function is also used to drop the items in the log tree before
4470          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4471          * it is used to drop the loged items. So we shouldn't kill the delayed
4472          * items.
4473          */
4474         if (min_type == 0 && root == BTRFS_I(inode)->root)
4475                 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4476
4477         key.objectid = ino;
4478         key.offset = (u64)-1;
4479         key.type = (u8)-1;
4480
4481 search_again:
4482         /*
4483          * with a 16K leaf size and 128MB extents, you can actually queue
4484          * up a huge file in a single leaf.  Most of the time that
4485          * bytes_deleted is > 0, it will be huge by the time we get here
4486          */
4487         if (be_nice && bytes_deleted > SZ_32M) {
4488                 if (btrfs_should_end_transaction(trans)) {
4489                         err = -EAGAIN;
4490                         goto error;
4491                 }
4492         }
4493
4494
4495         path->leave_spinning = 1;
4496         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4497         if (ret < 0) {
4498                 err = ret;
4499                 goto out;
4500         }
4501
4502         if (ret > 0) {
4503                 /* there are no items in the tree for us to truncate, we're
4504                  * done
4505                  */
4506                 if (path->slots[0] == 0)
4507                         goto out;
4508                 path->slots[0]--;
4509         }
4510
4511         while (1) {
4512                 fi = NULL;
4513                 leaf = path->nodes[0];
4514                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4515                 found_type = found_key.type;
4516
4517                 if (found_key.objectid != ino)
4518                         break;
4519
4520                 if (found_type < min_type)
4521                         break;
4522
4523                 item_end = found_key.offset;
4524                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4525                         fi = btrfs_item_ptr(leaf, path->slots[0],
4526                                             struct btrfs_file_extent_item);
4527                         extent_type = btrfs_file_extent_type(leaf, fi);
4528                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4529                                 item_end +=
4530                                     btrfs_file_extent_num_bytes(leaf, fi);
4531
4532                                 trace_btrfs_truncate_show_fi_regular(
4533                                         BTRFS_I(inode), leaf, fi,
4534                                         found_key.offset);
4535                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4536                                 item_end += btrfs_file_extent_inline_len(leaf,
4537                                                          path->slots[0], fi);
4538
4539                                 trace_btrfs_truncate_show_fi_inline(
4540                                         BTRFS_I(inode), leaf, fi, path->slots[0],
4541                                         found_key.offset);
4542                         }
4543                         item_end--;
4544                 }
4545                 if (found_type > min_type) {
4546                         del_item = 1;
4547                 } else {
4548                         if (item_end < new_size)
4549                                 break;
4550                         if (found_key.offset >= new_size)
4551                                 del_item = 1;
4552                         else
4553                                 del_item = 0;
4554                 }
4555                 found_extent = 0;
4556                 /* FIXME, shrink the extent if the ref count is only 1 */
4557                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4558                         goto delete;
4559
4560                 if (del_item)
4561                         last_size = found_key.offset;
4562                 else
4563                         last_size = new_size;
4564
4565                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4566                         u64 num_dec;
4567                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4568                         if (!del_item) {
4569                                 u64 orig_num_bytes =
4570                                         btrfs_file_extent_num_bytes(leaf, fi);
4571                                 extent_num_bytes = ALIGN(new_size -
4572                                                 found_key.offset,
4573                                                 fs_info->sectorsize);
4574                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4575                                                          extent_num_bytes);
4576                                 num_dec = (orig_num_bytes -
4577                                            extent_num_bytes);
4578                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4579                                              &root->state) &&
4580                                     extent_start != 0)
4581                                         inode_sub_bytes(inode, num_dec);
4582                                 btrfs_mark_buffer_dirty(leaf);
4583                         } else {
4584                                 extent_num_bytes =
4585                                         btrfs_file_extent_disk_num_bytes(leaf,
4586                                                                          fi);
4587                                 extent_offset = found_key.offset -
4588                                         btrfs_file_extent_offset(leaf, fi);
4589
4590                                 /* FIXME blocksize != 4096 */
4591                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4592                                 if (extent_start != 0) {
4593                                         found_extent = 1;
4594                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4595                                                      &root->state))
4596                                                 inode_sub_bytes(inode, num_dec);
4597                                 }
4598                         }
4599                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4600                         /*
4601                          * we can't truncate inline items that have had
4602                          * special encodings
4603                          */
4604                         if (!del_item &&
4605                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4606                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4607
4608                                 /*
4609                                  * Need to release path in order to truncate a
4610                                  * compressed extent. So delete any accumulated
4611                                  * extent items so far.
4612                                  */
4613                                 if (btrfs_file_extent_compression(leaf, fi) !=
4614                                     BTRFS_COMPRESS_NONE && pending_del_nr) {
4615                                         err = btrfs_del_items(trans, root, path,
4616                                                               pending_del_slot,
4617                                                               pending_del_nr);
4618                                         if (err) {
4619                                                 btrfs_abort_transaction(trans,
4620                                                                         err);
4621                                                 goto error;
4622                                         }
4623                                         pending_del_nr = 0;
4624                                 }
4625
4626                                 err = truncate_inline_extent(inode, path,
4627                                                              &found_key,
4628                                                              item_end,
4629                                                              new_size);
4630                                 if (err) {
4631                                         btrfs_abort_transaction(trans, err);
4632                                         goto error;
4633                                 }
4634                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4635                                             &root->state)) {
4636                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4637                         }
4638                 }
4639 delete:
4640                 if (del_item) {
4641                         if (!pending_del_nr) {
4642                                 /* no pending yet, add ourselves */
4643                                 pending_del_slot = path->slots[0];
4644                                 pending_del_nr = 1;
4645                         } else if (pending_del_nr &&
4646                                    path->slots[0] + 1 == pending_del_slot) {
4647                                 /* hop on the pending chunk */
4648                                 pending_del_nr++;
4649                                 pending_del_slot = path->slots[0];
4650                         } else {
4651                                 BUG();
4652                         }
4653                 } else {
4654                         break;
4655                 }
4656                 should_throttle = 0;
4657
4658                 if (found_extent &&
4659                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4660                      root == fs_info->tree_root)) {
4661                         btrfs_set_path_blocking(path);
4662                         bytes_deleted += extent_num_bytes;
4663                         ret = btrfs_free_extent(trans, fs_info, extent_start,
4664                                                 extent_num_bytes, 0,
4665                                                 btrfs_header_owner(leaf),
4666                                                 ino, extent_offset);
4667                         BUG_ON(ret);
4668                         if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4669                                 btrfs_async_run_delayed_refs(fs_info,
4670                                         trans->delayed_ref_updates * 2,
4671                                         trans->transid, 0);
4672                         if (be_nice) {
4673                                 if (truncate_space_check(trans, root,
4674                                                          extent_num_bytes)) {
4675                                         should_end = 1;
4676                                 }
4677                                 if (btrfs_should_throttle_delayed_refs(trans,
4678                                                                        fs_info))
4679                                         should_throttle = 1;
4680                         }
4681                 }
4682
4683                 if (found_type == BTRFS_INODE_ITEM_KEY)
4684                         break;
4685
4686                 if (path->slots[0] == 0 ||
4687                     path->slots[0] != pending_del_slot ||
4688                     should_throttle || should_end) {
4689                         if (pending_del_nr) {
4690                                 ret = btrfs_del_items(trans, root, path,
4691                                                 pending_del_slot,
4692                                                 pending_del_nr);
4693                                 if (ret) {
4694                                         btrfs_abort_transaction(trans, ret);
4695                                         goto error;
4696                                 }
4697                                 pending_del_nr = 0;
4698                         }
4699                         btrfs_release_path(path);
4700                         if (should_throttle) {
4701                                 unsigned long updates = trans->delayed_ref_updates;
4702                                 if (updates) {
4703                                         trans->delayed_ref_updates = 0;
4704                                         ret = btrfs_run_delayed_refs(trans,
4705                                                                    fs_info,
4706                                                                    updates * 2);
4707                                         if (ret && !err)
4708                                                 err = ret;
4709                                 }
4710                         }
4711                         /*
4712                          * if we failed to refill our space rsv, bail out
4713                          * and let the transaction restart
4714                          */
4715                         if (should_end) {
4716                                 err = -EAGAIN;
4717                                 goto error;
4718                         }
4719                         goto search_again;
4720                 } else {
4721                         path->slots[0]--;
4722                 }
4723         }
4724 out:
4725         if (pending_del_nr) {
4726                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4727                                       pending_del_nr);
4728                 if (ret)
4729                         btrfs_abort_transaction(trans, ret);
4730         }
4731 error:
4732         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4733                 ASSERT(last_size >= new_size);
4734                 if (!err && last_size > new_size)
4735                         last_size = new_size;
4736                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4737         }
4738
4739         btrfs_free_path(path);
4740
4741         if (be_nice && bytes_deleted > SZ_32M) {
4742                 unsigned long updates = trans->delayed_ref_updates;
4743                 if (updates) {
4744                         trans->delayed_ref_updates = 0;
4745                         ret = btrfs_run_delayed_refs(trans, fs_info,
4746                                                      updates * 2);
4747                         if (ret && !err)
4748                                 err = ret;
4749                 }
4750         }
4751         return err;
4752 }
4753
4754 /*
4755  * btrfs_truncate_block - read, zero a chunk and write a block
4756  * @inode - inode that we're zeroing
4757  * @from - the offset to start zeroing
4758  * @len - the length to zero, 0 to zero the entire range respective to the
4759  *      offset
4760  * @front - zero up to the offset instead of from the offset on
4761  *
4762  * This will find the block for the "from" offset and cow the block and zero the
4763  * part we want to zero.  This is used with truncate and hole punching.
4764  */
4765 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4766                         int front)
4767 {
4768         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4769         struct address_space *mapping = inode->i_mapping;
4770         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4771         struct btrfs_ordered_extent *ordered;
4772         struct extent_state *cached_state = NULL;
4773         struct extent_changeset *data_reserved = NULL;
4774         char *kaddr;
4775         u32 blocksize = fs_info->sectorsize;
4776         pgoff_t index = from >> PAGE_SHIFT;
4777         unsigned offset = from & (blocksize - 1);
4778         struct page *page;
4779         gfp_t mask = btrfs_alloc_write_mask(mapping);
4780         int ret = 0;
4781         u64 block_start;
4782         u64 block_end;
4783
4784         if ((offset & (blocksize - 1)) == 0 &&
4785             (!len || ((len & (blocksize - 1)) == 0)))
4786                 goto out;
4787
4788         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4789                         round_down(from, blocksize), blocksize);
4790         if (ret)
4791                 goto out;
4792
4793 again:
4794         page = find_or_create_page(mapping, index, mask);
4795         if (!page) {
4796                 btrfs_delalloc_release_space(inode, data_reserved,
4797                                 round_down(from, blocksize),
4798                                 blocksize);
4799                 ret = -ENOMEM;
4800                 goto out;
4801         }
4802
4803         block_start = round_down(from, blocksize);
4804         block_end = block_start + blocksize - 1;
4805
4806         if (!PageUptodate(page)) {
4807                 ret = btrfs_readpage(NULL, page);
4808                 lock_page(page);
4809                 if (page->mapping != mapping) {
4810                         unlock_page(page);
4811                         put_page(page);
4812                         goto again;
4813                 }
4814                 if (!PageUptodate(page)) {
4815                         ret = -EIO;
4816                         goto out_unlock;
4817                 }
4818         }
4819         wait_on_page_writeback(page);
4820
4821         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4822         set_page_extent_mapped(page);
4823
4824         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4825         if (ordered) {
4826                 unlock_extent_cached(io_tree, block_start, block_end,
4827                                      &cached_state, GFP_NOFS);
4828                 unlock_page(page);
4829                 put_page(page);
4830                 btrfs_start_ordered_extent(inode, ordered, 1);
4831                 btrfs_put_ordered_extent(ordered);
4832                 goto again;
4833         }
4834
4835         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4836                           EXTENT_DIRTY | EXTENT_DELALLOC |
4837                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4838                           0, 0, &cached_state, GFP_NOFS);
4839
4840         ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4841                                         &cached_state, 0);
4842         if (ret) {
4843                 unlock_extent_cached(io_tree, block_start, block_end,
4844                                      &cached_state, GFP_NOFS);
4845                 goto out_unlock;
4846         }
4847
4848         if (offset != blocksize) {
4849                 if (!len)
4850                         len = blocksize - offset;
4851                 kaddr = kmap(page);
4852                 if (front)
4853                         memset(kaddr + (block_start - page_offset(page)),
4854                                 0, offset);
4855                 else
4856                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4857                                 0, len);
4858                 flush_dcache_page(page);
4859                 kunmap(page);
4860         }
4861         ClearPageChecked(page);
4862         set_page_dirty(page);
4863         unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4864                              GFP_NOFS);
4865
4866 out_unlock:
4867         if (ret)
4868                 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4869                                              blocksize);
4870         unlock_page(page);
4871         put_page(page);
4872 out:
4873         extent_changeset_free(data_reserved);
4874         return ret;
4875 }
4876
4877 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4878                              u64 offset, u64 len)
4879 {
4880         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4881         struct btrfs_trans_handle *trans;
4882         int ret;
4883
4884         /*
4885          * Still need to make sure the inode looks like it's been updated so
4886          * that any holes get logged if we fsync.
4887          */
4888         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4889                 BTRFS_I(inode)->last_trans = fs_info->generation;
4890                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4891                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4892                 return 0;
4893         }
4894
4895         /*
4896          * 1 - for the one we're dropping
4897          * 1 - for the one we're adding
4898          * 1 - for updating the inode.
4899          */
4900         trans = btrfs_start_transaction(root, 3);
4901         if (IS_ERR(trans))
4902                 return PTR_ERR(trans);
4903
4904         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4905         if (ret) {
4906                 btrfs_abort_transaction(trans, ret);
4907                 btrfs_end_transaction(trans);
4908                 return ret;
4909         }
4910
4911         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4912                         offset, 0, 0, len, 0, len, 0, 0, 0);
4913         if (ret)
4914                 btrfs_abort_transaction(trans, ret);
4915         else
4916                 btrfs_update_inode(trans, root, inode);
4917         btrfs_end_transaction(trans);
4918         return ret;
4919 }
4920
4921 /*
4922  * This function puts in dummy file extents for the area we're creating a hole
4923  * for.  So if we are truncating this file to a larger size we need to insert
4924  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4925  * the range between oldsize and size
4926  */
4927 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4928 {
4929         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4930         struct btrfs_root *root = BTRFS_I(inode)->root;
4931         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4932         struct extent_map *em = NULL;
4933         struct extent_state *cached_state = NULL;
4934         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4935         u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4936         u64 block_end = ALIGN(size, fs_info->sectorsize);
4937         u64 last_byte;
4938         u64 cur_offset;
4939         u64 hole_size;
4940         int err = 0;
4941
4942         /*
4943          * If our size started in the middle of a block we need to zero out the
4944          * rest of the block before we expand the i_size, otherwise we could
4945          * expose stale data.
4946          */
4947         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4948         if (err)
4949                 return err;
4950
4951         if (size <= hole_start)
4952                 return 0;
4953
4954         while (1) {
4955                 struct btrfs_ordered_extent *ordered;
4956
4957                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4958                                  &cached_state);
4959                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4960                                                      block_end - hole_start);
4961                 if (!ordered)
4962                         break;
4963                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4964                                      &cached_state, GFP_NOFS);
4965                 btrfs_start_ordered_extent(inode, ordered, 1);
4966                 btrfs_put_ordered_extent(ordered);
4967         }
4968
4969         cur_offset = hole_start;
4970         while (1) {
4971                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4972                                 block_end - cur_offset, 0);
4973                 if (IS_ERR(em)) {
4974                         err = PTR_ERR(em);
4975                         em = NULL;
4976                         break;
4977                 }
4978                 last_byte = min(extent_map_end(em), block_end);
4979                 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4980                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4981                         struct extent_map *hole_em;
4982                         hole_size = last_byte - cur_offset;
4983
4984                         err = maybe_insert_hole(root, inode, cur_offset,
4985                                                 hole_size);
4986                         if (err)
4987                                 break;
4988                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4989                                                 cur_offset + hole_size - 1, 0);
4990                         hole_em = alloc_extent_map();
4991                         if (!hole_em) {
4992                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4993                                         &BTRFS_I(inode)->runtime_flags);
4994                                 goto next;
4995                         }
4996                         hole_em->start = cur_offset;
4997                         hole_em->len = hole_size;
4998                         hole_em->orig_start = cur_offset;
4999
5000                         hole_em->block_start = EXTENT_MAP_HOLE;
5001                         hole_em->block_len = 0;
5002                         hole_em->orig_block_len = 0;
5003                         hole_em->ram_bytes = hole_size;
5004                         hole_em->bdev = fs_info->fs_devices->latest_bdev;
5005                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
5006                         hole_em->generation = fs_info->generation;
5007
5008                         while (1) {
5009                                 write_lock(&em_tree->lock);
5010                                 err = add_extent_mapping(em_tree, hole_em, 1);
5011                                 write_unlock(&em_tree->lock);
5012                                 if (err != -EEXIST)
5013                                         break;
5014                                 btrfs_drop_extent_cache(BTRFS_I(inode),
5015                                                         cur_offset,
5016                                                         cur_offset +
5017                                                         hole_size - 1, 0);
5018                         }
5019                         free_extent_map(hole_em);
5020                 }
5021 next:
5022                 free_extent_map(em);
5023                 em = NULL;
5024                 cur_offset = last_byte;
5025                 if (cur_offset >= block_end)
5026                         break;
5027         }
5028         free_extent_map(em);
5029         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5030                              GFP_NOFS);
5031         return err;
5032 }
5033
5034 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5035 {
5036         struct btrfs_root *root = BTRFS_I(inode)->root;
5037         struct btrfs_trans_handle *trans;
5038         loff_t oldsize = i_size_read(inode);
5039         loff_t newsize = attr->ia_size;
5040         int mask = attr->ia_valid;
5041         int ret;
5042
5043         /*
5044          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5045          * special case where we need to update the times despite not having
5046          * these flags set.  For all other operations the VFS set these flags
5047          * explicitly if it wants a timestamp update.
5048          */
5049         if (newsize != oldsize) {
5050                 inode_inc_iversion(inode);
5051                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5052                         inode->i_ctime = inode->i_mtime =
5053                                 current_time(inode);
5054         }
5055
5056         if (newsize > oldsize) {
5057                 /*
5058                  * Don't do an expanding truncate while snapshoting is ongoing.
5059                  * This is to ensure the snapshot captures a fully consistent
5060                  * state of this file - if the snapshot captures this expanding
5061                  * truncation, it must capture all writes that happened before
5062                  * this truncation.
5063                  */
5064                 btrfs_wait_for_snapshot_creation(root);
5065                 ret = btrfs_cont_expand(inode, oldsize, newsize);
5066                 if (ret) {
5067                         btrfs_end_write_no_snapshoting(root);
5068                         return ret;
5069                 }
5070
5071                 trans = btrfs_start_transaction(root, 1);
5072                 if (IS_ERR(trans)) {
5073                         btrfs_end_write_no_snapshoting(root);
5074                         return PTR_ERR(trans);
5075                 }
5076
5077                 i_size_write(inode, newsize);
5078                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5079                 pagecache_isize_extended(inode, oldsize, newsize);
5080                 ret = btrfs_update_inode(trans, root, inode);
5081                 btrfs_end_write_no_snapshoting(root);
5082                 btrfs_end_transaction(trans);
5083         } else {
5084
5085                 /*
5086                  * We're truncating a file that used to have good data down to
5087                  * zero. Make sure it gets into the ordered flush list so that
5088                  * any new writes get down to disk quickly.
5089                  */
5090                 if (newsize == 0)
5091                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5092                                 &BTRFS_I(inode)->runtime_flags);
5093
5094                 /*
5095                  * 1 for the orphan item we're going to add
5096                  * 1 for the orphan item deletion.
5097                  */
5098                 trans = btrfs_start_transaction(root, 2);
5099                 if (IS_ERR(trans))
5100                         return PTR_ERR(trans);
5101
5102                 /*
5103                  * We need to do this in case we fail at _any_ point during the
5104                  * actual truncate.  Once we do the truncate_setsize we could
5105                  * invalidate pages which forces any outstanding ordered io to
5106                  * be instantly completed which will give us extents that need
5107                  * to be truncated.  If we fail to get an orphan inode down we
5108                  * could have left over extents that were never meant to live,
5109                  * so we need to guarantee from this point on that everything
5110                  * will be consistent.
5111                  */
5112                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5113                 btrfs_end_transaction(trans);
5114                 if (ret)
5115                         return ret;
5116
5117                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5118                 truncate_setsize(inode, newsize);
5119
5120                 /* Disable nonlocked read DIO to avoid the end less truncate */
5121                 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5122                 inode_dio_wait(inode);
5123                 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5124
5125                 ret = btrfs_truncate(inode);
5126                 if (ret && inode->i_nlink) {
5127                         int err;
5128
5129                         /* To get a stable disk_i_size */
5130                         err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5131                         if (err) {
5132                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5133                                 return err;
5134                         }
5135
5136                         /*
5137                          * failed to truncate, disk_i_size is only adjusted down
5138                          * as we remove extents, so it should represent the true
5139                          * size of the inode, so reset the in memory size and
5140                          * delete our orphan entry.
5141                          */
5142                         trans = btrfs_join_transaction(root);
5143                         if (IS_ERR(trans)) {
5144                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5145                                 return ret;
5146                         }
5147                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5148                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
5149                         if (err)
5150                                 btrfs_abort_transaction(trans, err);
5151                         btrfs_end_transaction(trans);
5152                 }
5153         }
5154
5155         return ret;
5156 }
5157
5158 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5159 {
5160         struct inode *inode = d_inode(dentry);
5161         struct btrfs_root *root = BTRFS_I(inode)->root;
5162         int err;
5163
5164         if (btrfs_root_readonly(root))
5165                 return -EROFS;
5166
5167         err = setattr_prepare(dentry, attr);
5168         if (err)
5169                 return err;
5170
5171         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5172                 err = btrfs_setsize(inode, attr);
5173                 if (err)
5174                         return err;
5175         }
5176
5177         if (attr->ia_valid) {
5178                 setattr_copy(inode, attr);
5179                 inode_inc_iversion(inode);
5180                 err = btrfs_dirty_inode(inode);
5181
5182                 if (!err && attr->ia_valid & ATTR_MODE)
5183                         err = posix_acl_chmod(inode, inode->i_mode);
5184         }
5185
5186         return err;
5187 }
5188
5189 /*
5190  * While truncating the inode pages during eviction, we get the VFS calling
5191  * btrfs_invalidatepage() against each page of the inode. This is slow because
5192  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5193  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5194  * extent_state structures over and over, wasting lots of time.
5195  *
5196  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5197  * those expensive operations on a per page basis and do only the ordered io
5198  * finishing, while we release here the extent_map and extent_state structures,
5199  * without the excessive merging and splitting.
5200  */
5201 static void evict_inode_truncate_pages(struct inode *inode)
5202 {
5203         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5204         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5205         struct rb_node *node;
5206
5207         ASSERT(inode->i_state & I_FREEING);
5208         truncate_inode_pages_final(&inode->i_data);
5209
5210         write_lock(&map_tree->lock);
5211         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5212                 struct extent_map *em;
5213
5214                 node = rb_first(&map_tree->map);
5215                 em = rb_entry(node, struct extent_map, rb_node);
5216                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5217                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5218                 remove_extent_mapping(map_tree, em);
5219                 free_extent_map(em);
5220                 if (need_resched()) {
5221                         write_unlock(&map_tree->lock);
5222                         cond_resched();
5223                         write_lock(&map_tree->lock);
5224                 }
5225         }
5226         write_unlock(&map_tree->lock);
5227
5228         /*
5229          * Keep looping until we have no more ranges in the io tree.
5230          * We can have ongoing bios started by readpages (called from readahead)
5231          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5232          * still in progress (unlocked the pages in the bio but did not yet
5233          * unlocked the ranges in the io tree). Therefore this means some
5234          * ranges can still be locked and eviction started because before
5235          * submitting those bios, which are executed by a separate task (work
5236          * queue kthread), inode references (inode->i_count) were not taken
5237          * (which would be dropped in the end io callback of each bio).
5238          * Therefore here we effectively end up waiting for those bios and
5239          * anyone else holding locked ranges without having bumped the inode's
5240          * reference count - if we don't do it, when they access the inode's
5241          * io_tree to unlock a range it may be too late, leading to an
5242          * use-after-free issue.
5243          */
5244         spin_lock(&io_tree->lock);
5245         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5246                 struct extent_state *state;
5247                 struct extent_state *cached_state = NULL;
5248                 u64 start;
5249                 u64 end;
5250
5251                 node = rb_first(&io_tree->state);
5252                 state = rb_entry(node, struct extent_state, rb_node);
5253                 start = state->start;
5254                 end = state->end;
5255                 spin_unlock(&io_tree->lock);
5256
5257                 lock_extent_bits(io_tree, start, end, &cached_state);
5258
5259                 /*
5260                  * If still has DELALLOC flag, the extent didn't reach disk,
5261                  * and its reserved space won't be freed by delayed_ref.
5262                  * So we need to free its reserved space here.
5263                  * (Refer to comment in btrfs_invalidatepage, case 2)
5264                  *
5265                  * Note, end is the bytenr of last byte, so we need + 1 here.
5266                  */
5267                 if (state->state & EXTENT_DELALLOC)
5268                         btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5269
5270                 clear_extent_bit(io_tree, start, end,
5271                                  EXTENT_LOCKED | EXTENT_DIRTY |
5272                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5273                                  EXTENT_DEFRAG, 1, 1,
5274                                  &cached_state, GFP_NOFS);
5275
5276                 cond_resched();
5277                 spin_lock(&io_tree->lock);
5278         }
5279         spin_unlock(&io_tree->lock);
5280 }
5281
5282 void btrfs_evict_inode(struct inode *inode)
5283 {
5284         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5285         struct btrfs_trans_handle *trans;
5286         struct btrfs_root *root = BTRFS_I(inode)->root;
5287         struct btrfs_block_rsv *rsv, *global_rsv;
5288         int steal_from_global = 0;
5289         u64 min_size;
5290         int ret;
5291
5292         trace_btrfs_inode_evict(inode);
5293
5294         if (!root) {
5295                 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5296                 return;
5297         }
5298
5299         min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5300
5301         evict_inode_truncate_pages(inode);
5302
5303         if (inode->i_nlink &&
5304             ((btrfs_root_refs(&root->root_item) != 0 &&
5305               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5306              btrfs_is_free_space_inode(BTRFS_I(inode))))
5307                 goto no_delete;
5308
5309         if (is_bad_inode(inode)) {
5310                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5311                 goto no_delete;
5312         }
5313         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5314         if (!special_file(inode->i_mode))
5315                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5316
5317         btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5318
5319         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5320                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5321                                  &BTRFS_I(inode)->runtime_flags));
5322                 goto no_delete;
5323         }
5324
5325         if (inode->i_nlink > 0) {
5326                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5327                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5328                 goto no_delete;
5329         }
5330
5331         ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5332         if (ret) {
5333                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5334                 goto no_delete;
5335         }
5336
5337         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5338         if (!rsv) {
5339                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5340                 goto no_delete;
5341         }
5342         rsv->size = min_size;
5343         rsv->failfast = 1;
5344         global_rsv = &fs_info->global_block_rsv;
5345
5346         btrfs_i_size_write(BTRFS_I(inode), 0);
5347
5348         /*
5349          * This is a bit simpler than btrfs_truncate since we've already
5350          * reserved our space for our orphan item in the unlink, so we just
5351          * need to reserve some slack space in case we add bytes and update
5352          * inode item when doing the truncate.
5353          */
5354         while (1) {
5355                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5356                                              BTRFS_RESERVE_FLUSH_LIMIT);
5357
5358                 /*
5359                  * Try and steal from the global reserve since we will
5360                  * likely not use this space anyway, we want to try as
5361                  * hard as possible to get this to work.
5362                  */
5363                 if (ret)
5364                         steal_from_global++;
5365                 else
5366                         steal_from_global = 0;
5367                 ret = 0;
5368
5369                 /*
5370                  * steal_from_global == 0: we reserved stuff, hooray!
5371                  * steal_from_global == 1: we didn't reserve stuff, boo!
5372                  * steal_from_global == 2: we've committed, still not a lot of
5373                  * room but maybe we'll have room in the global reserve this
5374                  * time.
5375                  * steal_from_global == 3: abandon all hope!
5376                  */
5377                 if (steal_from_global > 2) {
5378                         btrfs_warn(fs_info,
5379                                    "Could not get space for a delete, will truncate on mount %d",
5380                                    ret);
5381                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5382                         btrfs_free_block_rsv(fs_info, rsv);
5383                         goto no_delete;
5384                 }
5385
5386                 trans = btrfs_join_transaction(root);
5387                 if (IS_ERR(trans)) {
5388                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5389                         btrfs_free_block_rsv(fs_info, rsv);
5390                         goto no_delete;
5391                 }
5392
5393                 /*
5394                  * We can't just steal from the global reserve, we need to make
5395                  * sure there is room to do it, if not we need to commit and try
5396                  * again.
5397                  */
5398                 if (steal_from_global) {
5399                         if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5400                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5401                                                               min_size, 0);
5402                         else
5403                                 ret = -ENOSPC;
5404                 }
5405
5406                 /*
5407                  * Couldn't steal from the global reserve, we have too much
5408                  * pending stuff built up, commit the transaction and try it
5409                  * again.
5410                  */
5411                 if (ret) {
5412                         ret = btrfs_commit_transaction(trans);
5413                         if (ret) {
5414                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5415                                 btrfs_free_block_rsv(fs_info, rsv);
5416                                 goto no_delete;
5417                         }
5418                         continue;
5419                 } else {
5420                         steal_from_global = 0;
5421                 }
5422
5423                 trans->block_rsv = rsv;
5424
5425                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5426                 if (ret != -ENOSPC && ret != -EAGAIN)
5427                         break;
5428
5429                 trans->block_rsv = &fs_info->trans_block_rsv;
5430                 btrfs_end_transaction(trans);
5431                 trans = NULL;
5432                 btrfs_btree_balance_dirty(fs_info);
5433         }
5434
5435         btrfs_free_block_rsv(fs_info, rsv);
5436
5437         /*
5438          * Errors here aren't a big deal, it just means we leave orphan items
5439          * in the tree.  They will be cleaned up on the next mount.
5440          */
5441         if (ret == 0) {
5442                 trans->block_rsv = root->orphan_block_rsv;
5443                 btrfs_orphan_del(trans, BTRFS_I(inode));
5444         } else {
5445                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5446         }
5447
5448         trans->block_rsv = &fs_info->trans_block_rsv;
5449         if (!(root == fs_info->tree_root ||
5450               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5451                 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5452
5453         btrfs_end_transaction(trans);
5454         btrfs_btree_balance_dirty(fs_info);
5455 no_delete:
5456         btrfs_remove_delayed_node(BTRFS_I(inode));
5457         clear_inode(inode);
5458 }
5459
5460 /*
5461  * this returns the key found in the dir entry in the location pointer.
5462  * If no dir entries were found, location->objectid is 0.
5463  */
5464 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5465                                struct btrfs_key *location)
5466 {
5467         const char *name = dentry->d_name.name;
5468         int namelen = dentry->d_name.len;
5469         struct btrfs_dir_item *di;
5470         struct btrfs_path *path;
5471         struct btrfs_root *root = BTRFS_I(dir)->root;
5472         int ret = 0;
5473
5474         path = btrfs_alloc_path();
5475         if (!path)
5476                 return -ENOMEM;
5477
5478         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5479                         name, namelen, 0);
5480         if (IS_ERR(di))
5481                 ret = PTR_ERR(di);
5482
5483         if (IS_ERR_OR_NULL(di))
5484                 goto out_err;
5485
5486         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5487 out:
5488         btrfs_free_path(path);
5489         return ret;
5490 out_err:
5491         location->objectid = 0;
5492         goto out;
5493 }
5494
5495 /*
5496  * when we hit a tree root in a directory, the btrfs part of the inode
5497  * needs to be changed to reflect the root directory of the tree root.  This
5498  * is kind of like crossing a mount point.
5499  */
5500 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5501                                     struct inode *dir,
5502                                     struct dentry *dentry,
5503                                     struct btrfs_key *location,
5504                                     struct btrfs_root **sub_root)
5505 {
5506         struct btrfs_path *path;
5507         struct btrfs_root *new_root;
5508         struct btrfs_root_ref *ref;
5509         struct extent_buffer *leaf;
5510         struct btrfs_key key;
5511         int ret;
5512         int err = 0;
5513
5514         path = btrfs_alloc_path();
5515         if (!path) {
5516                 err = -ENOMEM;
5517                 goto out;
5518         }
5519
5520         err = -ENOENT;
5521         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5522         key.type = BTRFS_ROOT_REF_KEY;
5523         key.offset = location->objectid;
5524
5525         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5526         if (ret) {
5527                 if (ret < 0)
5528                         err = ret;
5529                 goto out;
5530         }
5531
5532         leaf = path->nodes[0];
5533         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5534         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5535             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5536                 goto out;
5537
5538         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5539                                    (unsigned long)(ref + 1),
5540                                    dentry->d_name.len);
5541         if (ret)
5542                 goto out;
5543
5544         btrfs_release_path(path);
5545
5546         new_root = btrfs_read_fs_root_no_name(fs_info, location);
5547         if (IS_ERR(new_root)) {
5548                 err = PTR_ERR(new_root);
5549                 goto out;
5550         }
5551
5552         *sub_root = new_root;
5553         location->objectid = btrfs_root_dirid(&new_root->root_item);
5554         location->type = BTRFS_INODE_ITEM_KEY;
5555         location->offset = 0;
5556         err = 0;
5557 out:
5558         btrfs_free_path(path);
5559         return err;
5560 }
5561
5562 static void inode_tree_add(struct inode *inode)
5563 {
5564         struct btrfs_root *root = BTRFS_I(inode)->root;
5565         struct btrfs_inode *entry;
5566         struct rb_node **p;
5567         struct rb_node *parent;
5568         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5569         u64 ino = btrfs_ino(BTRFS_I(inode));
5570
5571         if (inode_unhashed(inode))
5572                 return;
5573         parent = NULL;
5574         spin_lock(&root->inode_lock);
5575         p = &root->inode_tree.rb_node;
5576         while (*p) {
5577                 parent = *p;
5578                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5579
5580                 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5581                         p = &parent->rb_left;
5582                 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5583                         p = &parent->rb_right;
5584                 else {
5585                         WARN_ON(!(entry->vfs_inode.i_state &
5586                                   (I_WILL_FREE | I_FREEING)));
5587                         rb_replace_node(parent, new, &root->inode_tree);
5588                         RB_CLEAR_NODE(parent);
5589                         spin_unlock(&root->inode_lock);
5590                         return;
5591                 }
5592         }
5593         rb_link_node(new, parent, p);
5594         rb_insert_color(new, &root->inode_tree);
5595         spin_unlock(&root->inode_lock);
5596 }
5597
5598 static void inode_tree_del(struct inode *inode)
5599 {
5600         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5601         struct btrfs_root *root = BTRFS_I(inode)->root;
5602         int empty = 0;
5603
5604         spin_lock(&root->inode_lock);
5605         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5606                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5607                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5608                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5609         }
5610         spin_unlock(&root->inode_lock);
5611
5612         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5613                 synchronize_srcu(&fs_info->subvol_srcu);
5614                 spin_lock(&root->inode_lock);
5615                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5616                 spin_unlock(&root->inode_lock);
5617                 if (empty)
5618                         btrfs_add_dead_root(root);
5619         }
5620 }
5621
5622 void btrfs_invalidate_inodes(struct btrfs_root *root)
5623 {
5624         struct btrfs_fs_info *fs_info = root->fs_info;
5625         struct rb_node *node;
5626         struct rb_node *prev;
5627         struct btrfs_inode *entry;
5628         struct inode *inode;
5629         u64 objectid = 0;
5630
5631         if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5632                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5633
5634         spin_lock(&root->inode_lock);
5635 again:
5636         node = root->inode_tree.rb_node;
5637         prev = NULL;
5638         while (node) {
5639                 prev = node;
5640                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5641
5642                 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5643                         node = node->rb_left;
5644                 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5645                         node = node->rb_right;
5646                 else
5647                         break;
5648         }
5649         if (!node) {
5650                 while (prev) {
5651                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5652                         if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5653                                 node = prev;
5654                                 break;
5655                         }
5656                         prev = rb_next(prev);
5657                 }
5658         }
5659         while (node) {
5660                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5661                 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5662                 inode = igrab(&entry->vfs_inode);
5663                 if (inode) {
5664                         spin_unlock(&root->inode_lock);
5665                         if (atomic_read(&inode->i_count) > 1)
5666                                 d_prune_aliases(inode);
5667                         /*
5668                          * btrfs_drop_inode will have it removed from
5669                          * the inode cache when its usage count
5670                          * hits zero.
5671                          */
5672                         iput(inode);
5673                         cond_resched();
5674                         spin_lock(&root->inode_lock);
5675                         goto again;
5676                 }
5677
5678                 if (cond_resched_lock(&root->inode_lock))
5679                         goto again;
5680
5681                 node = rb_next(node);
5682         }
5683         spin_unlock(&root->inode_lock);
5684 }
5685
5686 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5687 {
5688         struct btrfs_iget_args *args = p;
5689         inode->i_ino = args->location->objectid;
5690         memcpy(&BTRFS_I(inode)->location, args->location,
5691                sizeof(*args->location));
5692         BTRFS_I(inode)->root = args->root;
5693         return 0;
5694 }
5695
5696 static int btrfs_find_actor(struct inode *inode, void *opaque)
5697 {
5698         struct btrfs_iget_args *args = opaque;
5699         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5700                 args->root == BTRFS_I(inode)->root;
5701 }
5702
5703 static struct inode *btrfs_iget_locked(struct super_block *s,
5704                                        struct btrfs_key *location,
5705                                        struct btrfs_root *root)
5706 {
5707         struct inode *inode;
5708         struct btrfs_iget_args args;
5709         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5710
5711         args.location = location;
5712         args.root = root;
5713
5714         inode = iget5_locked(s, hashval, btrfs_find_actor,
5715                              btrfs_init_locked_inode,
5716                              (void *)&args);
5717         return inode;
5718 }
5719
5720 /* Get an inode object given its location and corresponding root.
5721  * Returns in *is_new if the inode was read from disk
5722  */
5723 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5724                          struct btrfs_root *root, int *new)
5725 {
5726         struct inode *inode;
5727
5728         inode = btrfs_iget_locked(s, location, root);
5729         if (!inode)
5730                 return ERR_PTR(-ENOMEM);
5731
5732         if (inode->i_state & I_NEW) {
5733                 int ret;
5734
5735                 ret = btrfs_read_locked_inode(inode);
5736                 if (!is_bad_inode(inode)) {
5737                         inode_tree_add(inode);
5738                         unlock_new_inode(inode);
5739                         if (new)
5740                                 *new = 1;
5741                 } else {
5742                         unlock_new_inode(inode);
5743                         iput(inode);
5744                         ASSERT(ret < 0);
5745                         inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5746                 }
5747         }
5748
5749         return inode;
5750 }
5751
5752 static struct inode *new_simple_dir(struct super_block *s,
5753                                     struct btrfs_key *key,
5754                                     struct btrfs_root *root)
5755 {
5756         struct inode *inode = new_inode(s);
5757
5758         if (!inode)
5759                 return ERR_PTR(-ENOMEM);
5760
5761         BTRFS_I(inode)->root = root;
5762         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5763         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5764
5765         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5766         inode->i_op = &btrfs_dir_ro_inode_operations;
5767         inode->i_opflags &= ~IOP_XATTR;
5768         inode->i_fop = &simple_dir_operations;
5769         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5770         inode->i_mtime = current_time(inode);
5771         inode->i_atime = inode->i_mtime;
5772         inode->i_ctime = inode->i_mtime;
5773         BTRFS_I(inode)->i_otime = inode->i_mtime;
5774
5775         return inode;
5776 }
5777
5778 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5779 {
5780         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5781         struct inode *inode;
5782         struct btrfs_root *root = BTRFS_I(dir)->root;
5783         struct btrfs_root *sub_root = root;
5784         struct btrfs_key location;
5785         int index;
5786         int ret = 0;
5787
5788         if (dentry->d_name.len > BTRFS_NAME_LEN)
5789                 return ERR_PTR(-ENAMETOOLONG);
5790
5791         ret = btrfs_inode_by_name(dir, dentry, &location);
5792         if (ret < 0)
5793                 return ERR_PTR(ret);
5794
5795         if (location.objectid == 0)
5796                 return ERR_PTR(-ENOENT);
5797
5798         if (location.type == BTRFS_INODE_ITEM_KEY) {
5799                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5800                 return inode;
5801         }
5802
5803         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5804
5805         index = srcu_read_lock(&fs_info->subvol_srcu);
5806         ret = fixup_tree_root_location(fs_info, dir, dentry,
5807                                        &location, &sub_root);
5808         if (ret < 0) {
5809                 if (ret != -ENOENT)
5810                         inode = ERR_PTR(ret);
5811                 else
5812                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5813         } else {
5814                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5815         }
5816         srcu_read_unlock(&fs_info->subvol_srcu, index);
5817
5818         if (!IS_ERR(inode) && root != sub_root) {
5819                 down_read(&fs_info->cleanup_work_sem);
5820                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5821                         ret = btrfs_orphan_cleanup(sub_root);
5822                 up_read(&fs_info->cleanup_work_sem);
5823                 if (ret) {
5824                         iput(inode);
5825                         inode = ERR_PTR(ret);
5826                 }
5827         }
5828
5829         return inode;
5830 }
5831
5832 static int btrfs_dentry_delete(const struct dentry *dentry)
5833 {
5834         struct btrfs_root *root;
5835         struct inode *inode = d_inode(dentry);
5836
5837         if (!inode && !IS_ROOT(dentry))
5838                 inode = d_inode(dentry->d_parent);
5839
5840         if (inode) {
5841                 root = BTRFS_I(inode)->root;
5842                 if (btrfs_root_refs(&root->root_item) == 0)
5843                         return 1;
5844
5845                 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5846                         return 1;
5847         }
5848         return 0;
5849 }
5850
5851 static void btrfs_dentry_release(struct dentry *dentry)
5852 {
5853         kfree(dentry->d_fsdata);
5854 }
5855
5856 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5857                                    unsigned int flags)
5858 {
5859         struct inode *inode;
5860
5861         inode = btrfs_lookup_dentry(dir, dentry);
5862         if (IS_ERR(inode)) {
5863                 if (PTR_ERR(inode) == -ENOENT)
5864                         inode = NULL;
5865                 else
5866                         return ERR_CAST(inode);
5867         }
5868
5869         return d_splice_alias(inode, dentry);
5870 }
5871
5872 unsigned char btrfs_filetype_table[] = {
5873         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5874 };
5875
5876 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5877 {
5878         struct inode *inode = file_inode(file);
5879         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5880         struct btrfs_root *root = BTRFS_I(inode)->root;
5881         struct btrfs_dir_item *di;
5882         struct btrfs_key key;
5883         struct btrfs_key found_key;
5884         struct btrfs_path *path;
5885         struct list_head ins_list;
5886         struct list_head del_list;
5887         int ret;
5888         struct extent_buffer *leaf;
5889         int slot;
5890         unsigned char d_type;
5891         int over = 0;
5892         char tmp_name[32];
5893         char *name_ptr;
5894         int name_len;
5895         bool put = false;
5896         struct btrfs_key location;
5897
5898         if (!dir_emit_dots(file, ctx))
5899                 return 0;
5900
5901         path = btrfs_alloc_path();
5902         if (!path)
5903                 return -ENOMEM;
5904
5905         path->reada = READA_FORWARD;
5906
5907         INIT_LIST_HEAD(&ins_list);
5908         INIT_LIST_HEAD(&del_list);
5909         put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5910
5911         key.type = BTRFS_DIR_INDEX_KEY;
5912         key.offset = ctx->pos;
5913         key.objectid = btrfs_ino(BTRFS_I(inode));
5914
5915         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5916         if (ret < 0)
5917                 goto err;
5918
5919         while (1) {
5920                 leaf = path->nodes[0];
5921                 slot = path->slots[0];
5922                 if (slot >= btrfs_header_nritems(leaf)) {
5923                         ret = btrfs_next_leaf(root, path);
5924                         if (ret < 0)
5925                                 goto err;
5926                         else if (ret > 0)
5927                                 break;
5928                         continue;
5929                 }
5930
5931                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5932
5933                 if (found_key.objectid != key.objectid)
5934                         break;
5935                 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5936                         break;
5937                 if (found_key.offset < ctx->pos)
5938                         goto next;
5939                 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5940                         goto next;
5941
5942                 ctx->pos = found_key.offset;
5943
5944                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5945                 if (verify_dir_item(fs_info, leaf, slot, di))
5946                         goto next;
5947
5948                 name_len = btrfs_dir_name_len(leaf, di);
5949                 if (name_len <= sizeof(tmp_name)) {
5950                         name_ptr = tmp_name;
5951                 } else {
5952                         name_ptr = kmalloc(name_len, GFP_KERNEL);
5953                         if (!name_ptr) {
5954                                 ret = -ENOMEM;
5955                                 goto err;
5956                         }
5957                 }
5958                 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5959                                    name_len);
5960
5961                 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5962                 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5963
5964                 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5965                                  d_type);
5966
5967                 if (name_ptr != tmp_name)
5968                         kfree(name_ptr);
5969
5970                 if (over)
5971                         goto nopos;
5972                 ctx->pos++;
5973 next:
5974                 path->slots[0]++;
5975         }
5976
5977         ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5978         if (ret)
5979                 goto nopos;
5980
5981         /*
5982          * Stop new entries from being returned after we return the last
5983          * entry.
5984          *
5985          * New directory entries are assigned a strictly increasing
5986          * offset.  This means that new entries created during readdir
5987          * are *guaranteed* to be seen in the future by that readdir.
5988          * This has broken buggy programs which operate on names as
5989          * they're returned by readdir.  Until we re-use freed offsets
5990          * we have this hack to stop new entries from being returned
5991          * under the assumption that they'll never reach this huge
5992          * offset.
5993          *
5994          * This is being careful not to overflow 32bit loff_t unless the
5995          * last entry requires it because doing so has broken 32bit apps
5996          * in the past.
5997          */
5998         if (ctx->pos >= INT_MAX)
5999                 ctx->pos = LLONG_MAX;
6000         else
6001                 ctx->pos = INT_MAX;
6002 nopos:
6003         ret = 0;
6004 err:
6005         if (put)
6006                 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6007         btrfs_free_path(path);
6008         return ret;
6009 }
6010
6011 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6012 {
6013         struct btrfs_root *root = BTRFS_I(inode)->root;
6014         struct btrfs_trans_handle *trans;
6015         int ret = 0;
6016         bool nolock = false;
6017
6018         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6019                 return 0;
6020
6021         if (btrfs_fs_closing(root->fs_info) &&
6022                         btrfs_is_free_space_inode(BTRFS_I(inode)))
6023                 nolock = true;
6024
6025         if (wbc->sync_mode == WB_SYNC_ALL) {
6026                 if (nolock)
6027                         trans = btrfs_join_transaction_nolock(root);
6028                 else
6029                         trans = btrfs_join_transaction(root);
6030                 if (IS_ERR(trans))
6031                         return PTR_ERR(trans);
6032                 ret = btrfs_commit_transaction(trans);
6033         }
6034         return ret;
6035 }
6036
6037 /*
6038  * This is somewhat expensive, updating the tree every time the
6039  * inode changes.  But, it is most likely to find the inode in cache.
6040  * FIXME, needs more benchmarking...there are no reasons other than performance
6041  * to keep or drop this code.
6042  */
6043 static int btrfs_dirty_inode(struct inode *inode)
6044 {
6045         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6046         struct btrfs_root *root = BTRFS_I(inode)->root;
6047         struct btrfs_trans_handle *trans;
6048         int ret;
6049
6050         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6051                 return 0;
6052
6053         trans = btrfs_join_transaction(root);
6054         if (IS_ERR(trans))
6055                 return PTR_ERR(trans);
6056
6057         ret = btrfs_update_inode(trans, root, inode);
6058         if (ret && ret == -ENOSPC) {
6059                 /* whoops, lets try again with the full transaction */
6060                 btrfs_end_transaction(trans);
6061                 trans = btrfs_start_transaction(root, 1);
6062                 if (IS_ERR(trans))
6063                         return PTR_ERR(trans);
6064
6065                 ret = btrfs_update_inode(trans, root, inode);
6066         }
6067         btrfs_end_transaction(trans);
6068         if (BTRFS_I(inode)->delayed_node)
6069                 btrfs_balance_delayed_items(fs_info);
6070
6071         return ret;
6072 }
6073
6074 /*
6075  * This is a copy of file_update_time.  We need this so we can return error on
6076  * ENOSPC for updating the inode in the case of file write and mmap writes.
6077  */
6078 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6079                              int flags)
6080 {
6081         struct btrfs_root *root = BTRFS_I(inode)->root;
6082
6083         if (btrfs_root_readonly(root))
6084                 return -EROFS;
6085
6086         if (flags & S_VERSION)
6087                 inode_inc_iversion(inode);
6088         if (flags & S_CTIME)
6089                 inode->i_ctime = *now;
6090         if (flags & S_MTIME)
6091                 inode->i_mtime = *now;
6092         if (flags & S_ATIME)
6093                 inode->i_atime = *now;
6094         return btrfs_dirty_inode(inode);
6095 }
6096
6097 /*
6098  * find the highest existing sequence number in a directory
6099  * and then set the in-memory index_cnt variable to reflect
6100  * free sequence numbers
6101  */
6102 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6103 {
6104         struct btrfs_root *root = inode->root;
6105         struct btrfs_key key, found_key;
6106         struct btrfs_path *path;
6107         struct extent_buffer *leaf;
6108         int ret;
6109
6110         key.objectid = btrfs_ino(inode);
6111         key.type = BTRFS_DIR_INDEX_KEY;
6112         key.offset = (u64)-1;
6113
6114         path = btrfs_alloc_path();
6115         if (!path)
6116                 return -ENOMEM;
6117
6118         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6119         if (ret < 0)
6120                 goto out;
6121         /* FIXME: we should be able to handle this */
6122         if (ret == 0)
6123                 goto out;
6124         ret = 0;
6125
6126         /*
6127          * MAGIC NUMBER EXPLANATION:
6128          * since we search a directory based on f_pos we have to start at 2
6129          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6130          * else has to start at 2
6131          */
6132         if (path->slots[0] == 0) {
6133                 inode->index_cnt = 2;
6134                 goto out;
6135         }
6136
6137         path->slots[0]--;
6138
6139         leaf = path->nodes[0];
6140         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6141
6142         if (found_key.objectid != btrfs_ino(inode) ||
6143             found_key.type != BTRFS_DIR_INDEX_KEY) {
6144                 inode->index_cnt = 2;
6145                 goto out;
6146         }
6147
6148         inode->index_cnt = found_key.offset + 1;
6149 out:
6150         btrfs_free_path(path);
6151         return ret;
6152 }
6153
6154 /*
6155  * helper to find a free sequence number in a given directory.  This current
6156  * code is very simple, later versions will do smarter things in the btree
6157  */
6158 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6159 {
6160         int ret = 0;
6161
6162         if (dir->index_cnt == (u64)-1) {
6163                 ret = btrfs_inode_delayed_dir_index_count(dir);
6164                 if (ret) {
6165                         ret = btrfs_set_inode_index_count(dir);
6166                         if (ret)
6167                                 return ret;
6168                 }
6169         }
6170
6171         *index = dir->index_cnt;
6172         dir->index_cnt++;
6173
6174         return ret;
6175 }
6176
6177 static int btrfs_insert_inode_locked(struct inode *inode)
6178 {
6179         struct btrfs_iget_args args;
6180         args.location = &BTRFS_I(inode)->location;
6181         args.root = BTRFS_I(inode)->root;
6182
6183         return insert_inode_locked4(inode,
6184                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6185                    btrfs_find_actor, &args);
6186 }
6187
6188 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6189                                      struct btrfs_root *root,
6190                                      struct inode *dir,
6191                                      const char *name, int name_len,
6192                                      u64 ref_objectid, u64 objectid,
6193                                      umode_t mode, u64 *index)
6194 {
6195         struct btrfs_fs_info *fs_info = root->fs_info;
6196         struct inode *inode;
6197         struct btrfs_inode_item *inode_item;
6198         struct btrfs_key *location;
6199         struct btrfs_path *path;
6200         struct btrfs_inode_ref *ref;
6201         struct btrfs_key key[2];
6202         u32 sizes[2];
6203         int nitems = name ? 2 : 1;
6204         unsigned long ptr;
6205         int ret;
6206
6207         path = btrfs_alloc_path();
6208         if (!path)
6209                 return ERR_PTR(-ENOMEM);
6210
6211         inode = new_inode(fs_info->sb);
6212         if (!inode) {
6213                 btrfs_free_path(path);
6214                 return ERR_PTR(-ENOMEM);
6215         }
6216
6217         /*
6218          * O_TMPFILE, set link count to 0, so that after this point,
6219          * we fill in an inode item with the correct link count.
6220          */
6221         if (!name)
6222                 set_nlink(inode, 0);
6223
6224         /*
6225          * we have to initialize this early, so we can reclaim the inode
6226          * number if we fail afterwards in this function.
6227          */
6228         inode->i_ino = objectid;
6229
6230         if (dir && name) {
6231                 trace_btrfs_inode_request(dir);
6232
6233                 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6234                 if (ret) {
6235                         btrfs_free_path(path);
6236                         iput(inode);
6237                         return ERR_PTR(ret);
6238                 }
6239         } else if (dir) {
6240                 *index = 0;
6241         }
6242         /*
6243          * index_cnt is ignored for everything but a dir,
6244          * btrfs_get_inode_index_count has an explanation for the magic
6245          * number
6246          */
6247         BTRFS_I(inode)->index_cnt = 2;
6248         BTRFS_I(inode)->dir_index = *index;
6249         BTRFS_I(inode)->root = root;
6250         BTRFS_I(inode)->generation = trans->transid;
6251         inode->i_generation = BTRFS_I(inode)->generation;
6252
6253         /*
6254          * We could have gotten an inode number from somebody who was fsynced
6255          * and then removed in this same transaction, so let's just set full
6256          * sync since it will be a full sync anyway and this will blow away the
6257          * old info in the log.
6258          */
6259         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6260
6261         key[0].objectid = objectid;
6262         key[0].type = BTRFS_INODE_ITEM_KEY;
6263         key[0].offset = 0;
6264
6265         sizes[0] = sizeof(struct btrfs_inode_item);
6266
6267         if (name) {
6268                 /*
6269                  * Start new inodes with an inode_ref. This is slightly more
6270                  * efficient for small numbers of hard links since they will
6271                  * be packed into one item. Extended refs will kick in if we
6272                  * add more hard links than can fit in the ref item.
6273                  */
6274                 key[1].objectid = objectid;
6275                 key[1].type = BTRFS_INODE_REF_KEY;
6276                 key[1].offset = ref_objectid;
6277
6278                 sizes[1] = name_len + sizeof(*ref);
6279         }
6280
6281         location = &BTRFS_I(inode)->location;
6282         location->objectid = objectid;
6283         location->offset = 0;
6284         location->type = BTRFS_INODE_ITEM_KEY;
6285
6286         ret = btrfs_insert_inode_locked(inode);
6287         if (ret < 0)
6288                 goto fail;
6289
6290         path->leave_spinning = 1;
6291         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6292         if (ret != 0)
6293                 goto fail_unlock;
6294
6295         inode_init_owner(inode, dir, mode);
6296         inode_set_bytes(inode, 0);
6297
6298         inode->i_mtime = current_time(inode);
6299         inode->i_atime = inode->i_mtime;
6300         inode->i_ctime = inode->i_mtime;
6301         BTRFS_I(inode)->i_otime = inode->i_mtime;
6302
6303         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6304                                   struct btrfs_inode_item);
6305         memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6306                              sizeof(*inode_item));
6307         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6308
6309         if (name) {
6310                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6311                                      struct btrfs_inode_ref);
6312                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6313                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6314                 ptr = (unsigned long)(ref + 1);
6315                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6316         }
6317
6318         btrfs_mark_buffer_dirty(path->nodes[0]);
6319         btrfs_free_path(path);
6320
6321         btrfs_inherit_iflags(inode, dir);
6322
6323         if (S_ISREG(mode)) {
6324                 if (btrfs_test_opt(fs_info, NODATASUM))
6325                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6326                 if (btrfs_test_opt(fs_info, NODATACOW))
6327                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6328                                 BTRFS_INODE_NODATASUM;
6329         }
6330
6331         inode_tree_add(inode);
6332
6333         trace_btrfs_inode_new(inode);
6334         btrfs_set_inode_last_trans(trans, inode);
6335
6336         btrfs_update_root_times(trans, root);
6337
6338         ret = btrfs_inode_inherit_props(trans, inode, dir);
6339         if (ret)
6340                 btrfs_err(fs_info,
6341                           "error inheriting props for ino %llu (root %llu): %d",
6342                         btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6343
6344         return inode;
6345
6346 fail_unlock:
6347         unlock_new_inode(inode);
6348 fail:
6349         if (dir && name)
6350                 BTRFS_I(dir)->index_cnt--;
6351         btrfs_free_path(path);
6352         iput(inode);
6353         return ERR_PTR(ret);
6354 }
6355
6356 static inline u8 btrfs_inode_type(struct inode *inode)
6357 {
6358         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6359 }
6360
6361 /*
6362  * utility function to add 'inode' into 'parent_inode' with
6363  * a give name and a given sequence number.
6364  * if 'add_backref' is true, also insert a backref from the
6365  * inode to the parent directory.
6366  */
6367 int btrfs_add_link(struct btrfs_trans_handle *trans,
6368                    struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6369                    const char *name, int name_len, int add_backref, u64 index)
6370 {
6371         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6372         int ret = 0;
6373         struct btrfs_key key;
6374         struct btrfs_root *root = parent_inode->root;
6375         u64 ino = btrfs_ino(inode);
6376         u64 parent_ino = btrfs_ino(parent_inode);
6377
6378         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6379                 memcpy(&key, &inode->root->root_key, sizeof(key));
6380         } else {
6381                 key.objectid = ino;
6382                 key.type = BTRFS_INODE_ITEM_KEY;
6383                 key.offset = 0;
6384         }
6385
6386         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6387                 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6388                                          root->root_key.objectid, parent_ino,
6389                                          index, name, name_len);
6390         } else if (add_backref) {
6391                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6392                                              parent_ino, index);
6393         }
6394
6395         /* Nothing to clean up yet */
6396         if (ret)
6397                 return ret;
6398
6399         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6400                                     parent_inode, &key,
6401                                     btrfs_inode_type(&inode->vfs_inode), index);
6402         if (ret == -EEXIST || ret == -EOVERFLOW)
6403                 goto fail_dir_item;
6404         else if (ret) {
6405                 btrfs_abort_transaction(trans, ret);
6406                 return ret;
6407         }
6408
6409         btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6410                            name_len * 2);
6411         inode_inc_iversion(&parent_inode->vfs_inode);
6412         parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6413                 current_time(&parent_inode->vfs_inode);
6414         ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6415         if (ret)
6416                 btrfs_abort_transaction(trans, ret);
6417         return ret;
6418
6419 fail_dir_item:
6420         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6421                 u64 local_index;
6422                 int err;
6423                 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6424                                          root->root_key.objectid, parent_ino,
6425                                          &local_index, name, name_len);
6426
6427         } else if (add_backref) {
6428                 u64 local_index;
6429                 int err;
6430
6431                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6432                                           ino, parent_ino, &local_index);
6433         }
6434         return ret;
6435 }
6436
6437 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6438                             struct btrfs_inode *dir, struct dentry *dentry,
6439                             struct btrfs_inode *inode, int backref, u64 index)
6440 {
6441         int err = btrfs_add_link(trans, dir, inode,
6442                                  dentry->d_name.name, dentry->d_name.len,
6443                                  backref, index);
6444         if (err > 0)
6445                 err = -EEXIST;
6446         return err;
6447 }
6448
6449 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6450                         umode_t mode, dev_t rdev)
6451 {
6452         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6453         struct btrfs_trans_handle *trans;
6454         struct btrfs_root *root = BTRFS_I(dir)->root;
6455         struct inode *inode = NULL;
6456         int err;
6457         int drop_inode = 0;
6458         u64 objectid;
6459         u64 index = 0;
6460
6461         /*
6462          * 2 for inode item and ref
6463          * 2 for dir items
6464          * 1 for xattr if selinux is on
6465          */
6466         trans = btrfs_start_transaction(root, 5);
6467         if (IS_ERR(trans))
6468                 return PTR_ERR(trans);
6469
6470         err = btrfs_find_free_ino(root, &objectid);
6471         if (err)
6472                 goto out_unlock;
6473
6474         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6475                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6476                         mode, &index);
6477         if (IS_ERR(inode)) {
6478                 err = PTR_ERR(inode);
6479                 goto out_unlock;
6480         }
6481
6482         /*
6483         * If the active LSM wants to access the inode during
6484         * d_instantiate it needs these. Smack checks to see
6485         * if the filesystem supports xattrs by looking at the
6486         * ops vector.
6487         */
6488         inode->i_op = &btrfs_special_inode_operations;
6489         init_special_inode(inode, inode->i_mode, rdev);
6490
6491         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6492         if (err)
6493                 goto out_unlock_inode;
6494
6495         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6496                         0, index);
6497         if (err) {
6498                 goto out_unlock_inode;
6499         } else {
6500                 btrfs_update_inode(trans, root, inode);
6501                 unlock_new_inode(inode);
6502                 d_instantiate(dentry, inode);
6503         }
6504
6505 out_unlock:
6506         btrfs_end_transaction(trans);
6507         btrfs_balance_delayed_items(fs_info);
6508         btrfs_btree_balance_dirty(fs_info);
6509         if (drop_inode) {
6510                 inode_dec_link_count(inode);
6511                 iput(inode);
6512         }
6513         return err;
6514
6515 out_unlock_inode:
6516         drop_inode = 1;
6517         unlock_new_inode(inode);
6518         goto out_unlock;
6519
6520 }
6521
6522 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6523                         umode_t mode, bool excl)
6524 {
6525         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6526         struct btrfs_trans_handle *trans;
6527         struct btrfs_root *root = BTRFS_I(dir)->root;
6528         struct inode *inode = NULL;
6529         int drop_inode_on_err = 0;
6530         int err;
6531         u64 objectid;
6532         u64 index = 0;
6533
6534         /*
6535          * 2 for inode item and ref
6536          * 2 for dir items
6537          * 1 for xattr if selinux is on
6538          */
6539         trans = btrfs_start_transaction(root, 5);
6540         if (IS_ERR(trans))
6541                 return PTR_ERR(trans);
6542
6543         err = btrfs_find_free_ino(root, &objectid);
6544         if (err)
6545                 goto out_unlock;
6546
6547         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6548                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6549                         mode, &index);
6550         if (IS_ERR(inode)) {
6551                 err = PTR_ERR(inode);
6552                 goto out_unlock;
6553         }
6554         drop_inode_on_err = 1;
6555         /*
6556         * If the active LSM wants to access the inode during
6557         * d_instantiate it needs these. Smack checks to see
6558         * if the filesystem supports xattrs by looking at the
6559         * ops vector.
6560         */
6561         inode->i_fop = &btrfs_file_operations;
6562         inode->i_op = &btrfs_file_inode_operations;
6563         inode->i_mapping->a_ops = &btrfs_aops;
6564
6565         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6566         if (err)
6567                 goto out_unlock_inode;
6568
6569         err = btrfs_update_inode(trans, root, inode);
6570         if (err)
6571                 goto out_unlock_inode;
6572
6573         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6574                         0, index);
6575         if (err)
6576                 goto out_unlock_inode;
6577
6578         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6579         unlock_new_inode(inode);
6580         d_instantiate(dentry, inode);
6581
6582 out_unlock:
6583         btrfs_end_transaction(trans);
6584         if (err && drop_inode_on_err) {
6585                 inode_dec_link_count(inode);
6586                 iput(inode);
6587         }
6588         btrfs_balance_delayed_items(fs_info);
6589         btrfs_btree_balance_dirty(fs_info);
6590         return err;
6591
6592 out_unlock_inode:
6593         unlock_new_inode(inode);
6594         goto out_unlock;
6595
6596 }
6597
6598 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6599                       struct dentry *dentry)
6600 {
6601         struct btrfs_trans_handle *trans = NULL;
6602         struct btrfs_root *root = BTRFS_I(dir)->root;
6603         struct inode *inode = d_inode(old_dentry);
6604         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6605         u64 index;
6606         int err;
6607         int drop_inode = 0;
6608
6609         /* do not allow sys_link's with other subvols of the same device */
6610         if (root->objectid != BTRFS_I(inode)->root->objectid)
6611                 return -EXDEV;
6612
6613         if (inode->i_nlink >= BTRFS_LINK_MAX)
6614                 return -EMLINK;
6615
6616         err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6617         if (err)
6618                 goto fail;
6619
6620         /*
6621          * 2 items for inode and inode ref
6622          * 2 items for dir items
6623          * 1 item for parent inode
6624          */
6625         trans = btrfs_start_transaction(root, 5);
6626         if (IS_ERR(trans)) {
6627                 err = PTR_ERR(trans);
6628                 trans = NULL;
6629                 goto fail;
6630         }
6631
6632         /* There are several dir indexes for this inode, clear the cache. */
6633         BTRFS_I(inode)->dir_index = 0ULL;
6634         inc_nlink(inode);
6635         inode_inc_iversion(inode);
6636         inode->i_ctime = current_time(inode);
6637         ihold(inode);
6638         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6639
6640         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6641                         1, index);
6642
6643         if (err) {
6644                 drop_inode = 1;
6645         } else {
6646                 struct dentry *parent = dentry->d_parent;
6647                 err = btrfs_update_inode(trans, root, inode);
6648                 if (err)
6649                         goto fail;
6650                 if (inode->i_nlink == 1) {
6651                         /*
6652                          * If new hard link count is 1, it's a file created
6653                          * with open(2) O_TMPFILE flag.
6654                          */
6655                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
6656                         if (err)
6657                                 goto fail;
6658                 }
6659                 d_instantiate(dentry, inode);
6660                 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6661         }
6662
6663         btrfs_balance_delayed_items(fs_info);
6664 fail:
6665         if (trans)
6666                 btrfs_end_transaction(trans);
6667         if (drop_inode) {
6668                 inode_dec_link_count(inode);
6669                 iput(inode);
6670         }
6671         btrfs_btree_balance_dirty(fs_info);
6672         return err;
6673 }
6674
6675 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6676 {
6677         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6678         struct inode *inode = NULL;
6679         struct btrfs_trans_handle *trans;
6680         struct btrfs_root *root = BTRFS_I(dir)->root;
6681         int err = 0;
6682         int drop_on_err = 0;
6683         u64 objectid = 0;
6684         u64 index = 0;
6685
6686         /*
6687          * 2 items for inode and ref
6688          * 2 items for dir items
6689          * 1 for xattr if selinux is on
6690          */
6691         trans = btrfs_start_transaction(root, 5);
6692         if (IS_ERR(trans))
6693                 return PTR_ERR(trans);
6694
6695         err = btrfs_find_free_ino(root, &objectid);
6696         if (err)
6697                 goto out_fail;
6698
6699         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6700                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6701                         S_IFDIR | mode, &index);
6702         if (IS_ERR(inode)) {
6703                 err = PTR_ERR(inode);
6704                 goto out_fail;
6705         }
6706
6707         drop_on_err = 1;
6708         /* these must be set before we unlock the inode */
6709         inode->i_op = &btrfs_dir_inode_operations;
6710         inode->i_fop = &btrfs_dir_file_operations;
6711
6712         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6713         if (err)
6714                 goto out_fail_inode;
6715
6716         btrfs_i_size_write(BTRFS_I(inode), 0);
6717         err = btrfs_update_inode(trans, root, inode);
6718         if (err)
6719                 goto out_fail_inode;
6720
6721         err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6722                         dentry->d_name.name,
6723                         dentry->d_name.len, 0, index);
6724         if (err)
6725                 goto out_fail_inode;
6726
6727         d_instantiate(dentry, inode);
6728         /*
6729          * mkdir is special.  We're unlocking after we call d_instantiate
6730          * to avoid a race with nfsd calling d_instantiate.
6731          */
6732         unlock_new_inode(inode);
6733         drop_on_err = 0;
6734
6735 out_fail:
6736         btrfs_end_transaction(trans);
6737         if (drop_on_err) {
6738                 inode_dec_link_count(inode);
6739                 iput(inode);
6740         }
6741         btrfs_balance_delayed_items(fs_info);
6742         btrfs_btree_balance_dirty(fs_info);
6743         return err;
6744
6745 out_fail_inode:
6746         unlock_new_inode(inode);
6747         goto out_fail;
6748 }
6749
6750 /* Find next extent map of a given extent map, caller needs to ensure locks */
6751 static struct extent_map *next_extent_map(struct extent_map *em)
6752 {
6753         struct rb_node *next;
6754
6755         next = rb_next(&em->rb_node);
6756         if (!next)
6757                 return NULL;
6758         return container_of(next, struct extent_map, rb_node);
6759 }
6760
6761 static struct extent_map *prev_extent_map(struct extent_map *em)
6762 {
6763         struct rb_node *prev;
6764
6765         prev = rb_prev(&em->rb_node);
6766         if (!prev)
6767                 return NULL;
6768         return container_of(prev, struct extent_map, rb_node);
6769 }
6770
6771 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6772  * the existing extent is the nearest extent to map_start,
6773  * and an extent that you want to insert, deal with overlap and insert
6774  * the best fitted new extent into the tree.
6775  */
6776 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6777                                 struct extent_map *existing,
6778                                 struct extent_map *em,
6779                                 u64 map_start)
6780 {
6781         struct extent_map *prev;
6782         struct extent_map *next;
6783         u64 start;
6784         u64 end;
6785         u64 start_diff;
6786
6787         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6788
6789         if (existing->start > map_start) {
6790                 next = existing;
6791                 prev = prev_extent_map(next);
6792         } else {
6793                 prev = existing;
6794                 next = next_extent_map(prev);
6795         }
6796
6797         start = prev ? extent_map_end(prev) : em->start;
6798         start = max_t(u64, start, em->start);
6799         end = next ? next->start : extent_map_end(em);
6800         end = min_t(u64, end, extent_map_end(em));
6801         start_diff = start - em->start;
6802         em->start = start;
6803         em->len = end - start;
6804         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6805             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6806                 em->block_start += start_diff;
6807                 em->block_len -= start_diff;
6808         }
6809         return add_extent_mapping(em_tree, em, 0);
6810 }
6811
6812 static noinline int uncompress_inline(struct btrfs_path *path,
6813                                       struct page *page,
6814                                       size_t pg_offset, u64 extent_offset,
6815                                       struct btrfs_file_extent_item *item)
6816 {
6817         int ret;
6818         struct extent_buffer *leaf = path->nodes[0];
6819         char *tmp;
6820         size_t max_size;
6821         unsigned long inline_size;
6822         unsigned long ptr;
6823         int compress_type;
6824
6825         WARN_ON(pg_offset != 0);
6826         compress_type = btrfs_file_extent_compression(leaf, item);
6827         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6828         inline_size = btrfs_file_extent_inline_item_len(leaf,
6829                                         btrfs_item_nr(path->slots[0]));
6830         tmp = kmalloc(inline_size, GFP_NOFS);
6831         if (!tmp)
6832                 return -ENOMEM;
6833         ptr = btrfs_file_extent_inline_start(item);
6834
6835         read_extent_buffer(leaf, tmp, ptr, inline_size);
6836
6837         max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6838         ret = btrfs_decompress(compress_type, tmp, page,
6839                                extent_offset, inline_size, max_size);
6840
6841         /*
6842          * decompression code contains a memset to fill in any space between the end
6843          * of the uncompressed data and the end of max_size in case the decompressed
6844          * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6845          * the end of an inline extent and the beginning of the next block, so we
6846          * cover that region here.
6847          */
6848
6849         if (max_size + pg_offset < PAGE_SIZE) {
6850                 char *map = kmap(page);
6851                 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6852                 kunmap(page);
6853         }
6854         kfree(tmp);
6855         return ret;
6856 }
6857
6858 /*
6859  * a bit scary, this does extent mapping from logical file offset to the disk.
6860  * the ugly parts come from merging extents from the disk with the in-ram
6861  * representation.  This gets more complex because of the data=ordered code,
6862  * where the in-ram extents might be locked pending data=ordered completion.
6863  *
6864  * This also copies inline extents directly into the page.
6865  */
6866 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6867                 struct page *page,
6868             size_t pg_offset, u64 start, u64 len,
6869                 int create)
6870 {
6871         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6872         int ret;
6873         int err = 0;
6874         u64 extent_start = 0;
6875         u64 extent_end = 0;
6876         u64 objectid = btrfs_ino(inode);
6877         u32 found_type;
6878         struct btrfs_path *path = NULL;
6879         struct btrfs_root *root = inode->root;
6880         struct btrfs_file_extent_item *item;
6881         struct extent_buffer *leaf;
6882         struct btrfs_key found_key;
6883         struct extent_map *em = NULL;
6884         struct extent_map_tree *em_tree = &inode->extent_tree;
6885         struct extent_io_tree *io_tree = &inode->io_tree;
6886         struct btrfs_trans_handle *trans = NULL;
6887         const bool new_inline = !page || create;
6888
6889 again:
6890         read_lock(&em_tree->lock);
6891         em = lookup_extent_mapping(em_tree, start, len);
6892         if (em)
6893                 em->bdev = fs_info->fs_devices->latest_bdev;
6894         read_unlock(&em_tree->lock);
6895
6896         if (em) {
6897                 if (em->start > start || em->start + em->len <= start)
6898                         free_extent_map(em);
6899                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6900                         free_extent_map(em);
6901                 else
6902                         goto out;
6903         }
6904         em = alloc_extent_map();
6905         if (!em) {
6906                 err = -ENOMEM;
6907                 goto out;
6908         }
6909         em->bdev = fs_info->fs_devices->latest_bdev;
6910         em->start = EXTENT_MAP_HOLE;
6911         em->orig_start = EXTENT_MAP_HOLE;
6912         em->len = (u64)-1;
6913         em->block_len = (u64)-1;
6914
6915         if (!path) {
6916                 path = btrfs_alloc_path();
6917                 if (!path) {
6918                         err = -ENOMEM;
6919                         goto out;
6920                 }
6921                 /*
6922                  * Chances are we'll be called again, so go ahead and do
6923                  * readahead
6924                  */
6925                 path->reada = READA_FORWARD;
6926         }
6927
6928         ret = btrfs_lookup_file_extent(trans, root, path,
6929                                        objectid, start, trans != NULL);
6930         if (ret < 0) {
6931                 err = ret;
6932                 goto out;
6933         }
6934
6935         if (ret != 0) {
6936                 if (path->slots[0] == 0)
6937                         goto not_found;
6938                 path->slots[0]--;
6939         }
6940
6941         leaf = path->nodes[0];
6942         item = btrfs_item_ptr(leaf, path->slots[0],
6943                               struct btrfs_file_extent_item);
6944         /* are we inside the extent that was found? */
6945         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6946         found_type = found_key.type;
6947         if (found_key.objectid != objectid ||
6948             found_type != BTRFS_EXTENT_DATA_KEY) {
6949                 /*
6950                  * If we backup past the first extent we want to move forward
6951                  * and see if there is an extent in front of us, otherwise we'll
6952                  * say there is a hole for our whole search range which can
6953                  * cause problems.
6954                  */
6955                 extent_end = start;
6956                 goto next;
6957         }
6958
6959         found_type = btrfs_file_extent_type(leaf, item);
6960         extent_start = found_key.offset;
6961         if (found_type == BTRFS_FILE_EXTENT_REG ||
6962             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6963                 extent_end = extent_start +
6964                        btrfs_file_extent_num_bytes(leaf, item);
6965
6966                 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6967                                                        extent_start);
6968         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6969                 size_t size;
6970                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6971                 extent_end = ALIGN(extent_start + size,
6972                                    fs_info->sectorsize);
6973
6974                 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6975                                                       path->slots[0],
6976                                                       extent_start);
6977         }
6978 next:
6979         if (start >= extent_end) {
6980                 path->slots[0]++;
6981                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6982                         ret = btrfs_next_leaf(root, path);
6983                         if (ret < 0) {
6984                                 err = ret;
6985                                 goto out;
6986                         }
6987                         if (ret > 0)
6988                                 goto not_found;
6989                         leaf = path->nodes[0];
6990                 }
6991                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6992                 if (found_key.objectid != objectid ||
6993                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6994                         goto not_found;
6995                 if (start + len <= found_key.offset)
6996                         goto not_found;
6997                 if (start > found_key.offset)
6998                         goto next;
6999                 em->start = start;
7000                 em->orig_start = start;
7001                 em->len = found_key.offset - start;
7002                 goto not_found_em;
7003         }
7004
7005         btrfs_extent_item_to_extent_map(inode, path, item,
7006                         new_inline, em);
7007
7008         if (found_type == BTRFS_FILE_EXTENT_REG ||
7009             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7010                 goto insert;
7011         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7012                 unsigned long ptr;
7013                 char *map;
7014                 size_t size;
7015                 size_t extent_offset;
7016                 size_t copy_size;
7017
7018                 if (new_inline)
7019                         goto out;
7020
7021                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7022                 extent_offset = page_offset(page) + pg_offset - extent_start;
7023                 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7024                                   size - extent_offset);
7025                 em->start = extent_start + extent_offset;
7026                 em->len = ALIGN(copy_size, fs_info->sectorsize);
7027                 em->orig_block_len = em->len;
7028                 em->orig_start = em->start;
7029                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7030                 if (create == 0 && !PageUptodate(page)) {
7031                         if (btrfs_file_extent_compression(leaf, item) !=
7032                             BTRFS_COMPRESS_NONE) {
7033                                 ret = uncompress_inline(path, page, pg_offset,
7034                                                         extent_offset, item);
7035                                 if (ret) {
7036                                         err = ret;
7037                                         goto out;
7038                                 }
7039                         } else {
7040                                 map = kmap(page);
7041                                 read_extent_buffer(leaf, map + pg_offset, ptr,
7042                                                    copy_size);
7043                                 if (pg_offset + copy_size < PAGE_SIZE) {
7044                                         memset(map + pg_offset + copy_size, 0,
7045                                                PAGE_SIZE - pg_offset -
7046                                                copy_size);
7047                                 }
7048                                 kunmap(page);
7049                         }
7050                         flush_dcache_page(page);
7051                 } else if (create && PageUptodate(page)) {
7052                         BUG();
7053                         if (!trans) {
7054                                 kunmap(page);
7055                                 free_extent_map(em);
7056                                 em = NULL;
7057
7058                                 btrfs_release_path(path);
7059                                 trans = btrfs_join_transaction(root);
7060
7061                                 if (IS_ERR(trans))
7062                                         return ERR_CAST(trans);
7063                                 goto again;
7064                         }
7065                         map = kmap(page);
7066                         write_extent_buffer(leaf, map + pg_offset, ptr,
7067                                             copy_size);
7068                         kunmap(page);
7069                         btrfs_mark_buffer_dirty(leaf);
7070                 }
7071                 set_extent_uptodate(io_tree, em->start,
7072                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
7073                 goto insert;
7074         }
7075 not_found:
7076         em->start = start;
7077         em->orig_start = start;
7078         em->len = len;
7079 not_found_em:
7080         em->block_start = EXTENT_MAP_HOLE;
7081         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7082 insert:
7083         btrfs_release_path(path);
7084         if (em->start > start || extent_map_end(em) <= start) {
7085                 btrfs_err(fs_info,
7086                           "bad extent! em: [%llu %llu] passed [%llu %llu]",
7087                           em->start, em->len, start, len);
7088                 err = -EIO;
7089                 goto out;
7090         }
7091
7092         err = 0;
7093         write_lock(&em_tree->lock);
7094         ret = add_extent_mapping(em_tree, em, 0);
7095         /* it is possible that someone inserted the extent into the tree
7096          * while we had the lock dropped.  It is also possible that
7097          * an overlapping map exists in the tree
7098          */
7099         if (ret == -EEXIST) {
7100                 struct extent_map *existing;
7101
7102                 ret = 0;
7103
7104                 existing = search_extent_mapping(em_tree, start, len);
7105                 /*
7106                  * existing will always be non-NULL, since there must be
7107                  * extent causing the -EEXIST.
7108                  */
7109                 if (existing->start == em->start &&
7110                     extent_map_end(existing) >= extent_map_end(em) &&
7111                     em->block_start == existing->block_start) {
7112                         /*
7113                          * The existing extent map already encompasses the
7114                          * entire extent map we tried to add.
7115                          */
7116                         free_extent_map(em);
7117                         em = existing;
7118                         err = 0;
7119
7120                 } else if (start >= extent_map_end(existing) ||
7121                     start <= existing->start) {
7122                         /*
7123                          * The existing extent map is the one nearest to
7124                          * the [start, start + len) range which overlaps
7125                          */
7126                         err = merge_extent_mapping(em_tree, existing,
7127                                                    em, start);
7128                         free_extent_map(existing);
7129                         if (err) {
7130                                 free_extent_map(em);
7131                                 em = NULL;
7132                         }
7133                 } else {
7134                         free_extent_map(em);
7135                         em = existing;
7136                         err = 0;
7137                 }
7138         }
7139         write_unlock(&em_tree->lock);
7140 out:
7141
7142         trace_btrfs_get_extent(root, inode, em);
7143
7144         btrfs_free_path(path);
7145         if (trans) {
7146                 ret = btrfs_end_transaction(trans);
7147                 if (!err)
7148                         err = ret;
7149         }
7150         if (err) {
7151                 free_extent_map(em);
7152                 return ERR_PTR(err);
7153         }
7154         BUG_ON(!em); /* Error is always set */
7155         return em;
7156 }
7157
7158 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7159                 struct page *page,
7160                 size_t pg_offset, u64 start, u64 len,
7161                 int create)
7162 {
7163         struct extent_map *em;
7164         struct extent_map *hole_em = NULL;
7165         u64 range_start = start;
7166         u64 end;
7167         u64 found;
7168         u64 found_end;
7169         int err = 0;
7170
7171         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7172         if (IS_ERR(em))
7173                 return em;
7174         /*
7175          * If our em maps to:
7176          * - a hole or
7177          * - a pre-alloc extent,
7178          * there might actually be delalloc bytes behind it.
7179          */
7180         if (em->block_start != EXTENT_MAP_HOLE &&
7181             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7182                 return em;
7183         else
7184                 hole_em = em;
7185
7186         /* check to see if we've wrapped (len == -1 or similar) */
7187         end = start + len;
7188         if (end < start)
7189                 end = (u64)-1;
7190         else
7191                 end -= 1;
7192
7193         em = NULL;
7194
7195         /* ok, we didn't find anything, lets look for delalloc */
7196         found = count_range_bits(&inode->io_tree, &range_start,
7197                                  end, len, EXTENT_DELALLOC, 1);
7198         found_end = range_start + found;
7199         if (found_end < range_start)
7200                 found_end = (u64)-1;
7201
7202         /*
7203          * we didn't find anything useful, return
7204          * the original results from get_extent()
7205          */
7206         if (range_start > end || found_end <= start) {
7207                 em = hole_em;
7208                 hole_em = NULL;
7209                 goto out;
7210         }
7211
7212         /* adjust the range_start to make sure it doesn't
7213          * go backwards from the start they passed in
7214          */
7215         range_start = max(start, range_start);
7216         found = found_end - range_start;
7217
7218         if (found > 0) {
7219                 u64 hole_start = start;
7220                 u64 hole_len = len;
7221
7222                 em = alloc_extent_map();
7223                 if (!em) {
7224                         err = -ENOMEM;
7225                         goto out;
7226                 }
7227                 /*
7228                  * when btrfs_get_extent can't find anything it
7229                  * returns one huge hole
7230                  *
7231                  * make sure what it found really fits our range, and
7232                  * adjust to make sure it is based on the start from
7233                  * the caller
7234                  */
7235                 if (hole_em) {
7236                         u64 calc_end = extent_map_end(hole_em);
7237
7238                         if (calc_end <= start || (hole_em->start > end)) {
7239                                 free_extent_map(hole_em);
7240                                 hole_em = NULL;
7241                         } else {
7242                                 hole_start = max(hole_em->start, start);
7243                                 hole_len = calc_end - hole_start;
7244                         }
7245                 }
7246                 em->bdev = NULL;
7247                 if (hole_em && range_start > hole_start) {
7248                         /* our hole starts before our delalloc, so we
7249                          * have to return just the parts of the hole
7250                          * that go until  the delalloc starts
7251                          */
7252                         em->len = min(hole_len,
7253                                       range_start - hole_start);
7254                         em->start = hole_start;
7255                         em->orig_start = hole_start;
7256                         /*
7257                          * don't adjust block start at all,
7258                          * it is fixed at EXTENT_MAP_HOLE
7259                          */
7260                         em->block_start = hole_em->block_start;
7261                         em->block_len = hole_len;
7262                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7263                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7264                 } else {
7265                         em->start = range_start;
7266                         em->len = found;
7267                         em->orig_start = range_start;
7268                         em->block_start = EXTENT_MAP_DELALLOC;
7269                         em->block_len = found;
7270                 }
7271         } else if (hole_em) {
7272                 return hole_em;
7273         }
7274 out:
7275
7276         free_extent_map(hole_em);
7277         if (err) {
7278                 free_extent_map(em);
7279                 return ERR_PTR(err);
7280         }
7281         return em;
7282 }
7283
7284 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7285                                                   const u64 start,
7286                                                   const u64 len,
7287                                                   const u64 orig_start,
7288                                                   const u64 block_start,
7289                                                   const u64 block_len,
7290                                                   const u64 orig_block_len,
7291                                                   const u64 ram_bytes,
7292                                                   const int type)
7293 {
7294         struct extent_map *em = NULL;
7295         int ret;
7296
7297         if (type != BTRFS_ORDERED_NOCOW) {
7298                 em = create_io_em(inode, start, len, orig_start,
7299                                   block_start, block_len, orig_block_len,
7300                                   ram_bytes,
7301                                   BTRFS_COMPRESS_NONE, /* compress_type */
7302                                   type);
7303                 if (IS_ERR(em))
7304                         goto out;
7305         }
7306         ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7307                                            len, block_len, type);
7308         if (ret) {
7309                 if (em) {
7310                         free_extent_map(em);
7311                         btrfs_drop_extent_cache(BTRFS_I(inode), start,
7312                                                 start + len - 1, 0);
7313                 }
7314                 em = ERR_PTR(ret);
7315         }
7316  out:
7317
7318         return em;
7319 }
7320
7321 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7322                                                   u64 start, u64 len)
7323 {
7324         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7325         struct btrfs_root *root = BTRFS_I(inode)->root;
7326         struct extent_map *em;
7327         struct btrfs_key ins;
7328         u64 alloc_hint;
7329         int ret;
7330
7331         alloc_hint = get_extent_allocation_hint(inode, start, len);
7332         ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7333                                    0, alloc_hint, &ins, 1, 1);
7334         if (ret)
7335                 return ERR_PTR(ret);
7336
7337         em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7338                                      ins.objectid, ins.offset, ins.offset,
7339                                      ins.offset, BTRFS_ORDERED_REGULAR);
7340         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7341         if (IS_ERR(em))
7342                 btrfs_free_reserved_extent(fs_info, ins.objectid,
7343                                            ins.offset, 1);
7344
7345         return em;
7346 }
7347
7348 /*
7349  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7350  * block must be cow'd
7351  */
7352 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7353                               u64 *orig_start, u64 *orig_block_len,
7354                               u64 *ram_bytes)
7355 {
7356         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7357         struct btrfs_path *path;
7358         int ret;
7359         struct extent_buffer *leaf;
7360         struct btrfs_root *root = BTRFS_I(inode)->root;
7361         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7362         struct btrfs_file_extent_item *fi;
7363         struct btrfs_key key;
7364         u64 disk_bytenr;
7365         u64 backref_offset;
7366         u64 extent_end;
7367         u64 num_bytes;
7368         int slot;
7369         int found_type;
7370         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7371
7372         path = btrfs_alloc_path();
7373         if (!path)
7374                 return -ENOMEM;
7375
7376         ret = btrfs_lookup_file_extent(NULL, root, path,
7377                         btrfs_ino(BTRFS_I(inode)), offset, 0);
7378         if (ret < 0)
7379                 goto out;
7380
7381         slot = path->slots[0];
7382         if (ret == 1) {
7383                 if (slot == 0) {
7384                         /* can't find the item, must cow */
7385                         ret = 0;
7386                         goto out;
7387                 }
7388                 slot--;
7389         }
7390         ret = 0;
7391         leaf = path->nodes[0];
7392         btrfs_item_key_to_cpu(leaf, &key, slot);
7393         if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7394             key.type != BTRFS_EXTENT_DATA_KEY) {
7395                 /* not our file or wrong item type, must cow */
7396                 goto out;
7397         }
7398
7399         if (key.offset > offset) {
7400                 /* Wrong offset, must cow */
7401                 goto out;
7402         }
7403
7404         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7405         found_type = btrfs_file_extent_type(leaf, fi);
7406         if (found_type != BTRFS_FILE_EXTENT_REG &&
7407             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7408                 /* not a regular extent, must cow */
7409                 goto out;
7410         }
7411
7412         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7413                 goto out;
7414
7415         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7416         if (extent_end <= offset)
7417                 goto out;
7418
7419         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7420         if (disk_bytenr == 0)
7421                 goto out;
7422
7423         if (btrfs_file_extent_compression(leaf, fi) ||
7424             btrfs_file_extent_encryption(leaf, fi) ||
7425             btrfs_file_extent_other_encoding(leaf, fi))
7426                 goto out;
7427
7428         backref_offset = btrfs_file_extent_offset(leaf, fi);
7429
7430         if (orig_start) {
7431                 *orig_start = key.offset - backref_offset;
7432                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7433                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7434         }
7435
7436         if (btrfs_extent_readonly(fs_info, disk_bytenr))
7437                 goto out;
7438
7439         num_bytes = min(offset + *len, extent_end) - offset;
7440         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7441                 u64 range_end;
7442
7443                 range_end = round_up(offset + num_bytes,
7444                                      root->fs_info->sectorsize) - 1;
7445                 ret = test_range_bit(io_tree, offset, range_end,
7446                                      EXTENT_DELALLOC, 0, NULL);
7447                 if (ret) {
7448                         ret = -EAGAIN;
7449                         goto out;
7450                 }
7451         }
7452
7453         btrfs_release_path(path);
7454
7455         /*
7456          * look for other files referencing this extent, if we
7457          * find any we must cow
7458          */
7459
7460         ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7461                                     key.offset - backref_offset, disk_bytenr);
7462         if (ret) {
7463                 ret = 0;
7464                 goto out;
7465         }
7466
7467         /*
7468          * adjust disk_bytenr and num_bytes to cover just the bytes
7469          * in this extent we are about to write.  If there
7470          * are any csums in that range we have to cow in order
7471          * to keep the csums correct
7472          */
7473         disk_bytenr += backref_offset;
7474         disk_bytenr += offset - key.offset;
7475         if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7476                 goto out;
7477         /*
7478          * all of the above have passed, it is safe to overwrite this extent
7479          * without cow
7480          */
7481         *len = num_bytes;
7482         ret = 1;
7483 out:
7484         btrfs_free_path(path);
7485         return ret;
7486 }
7487
7488 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7489 {
7490         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7491         bool found = false;
7492         void **pagep = NULL;
7493         struct page *page = NULL;
7494         unsigned long start_idx;
7495         unsigned long end_idx;
7496
7497         start_idx = start >> PAGE_SHIFT;
7498
7499         /*
7500          * end is the last byte in the last page.  end == start is legal
7501          */
7502         end_idx = end >> PAGE_SHIFT;
7503
7504         rcu_read_lock();
7505
7506         /* Most of the code in this while loop is lifted from
7507          * find_get_page.  It's been modified to begin searching from a
7508          * page and return just the first page found in that range.  If the
7509          * found idx is less than or equal to the end idx then we know that
7510          * a page exists.  If no pages are found or if those pages are
7511          * outside of the range then we're fine (yay!) */
7512         while (page == NULL &&
7513                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7514                 page = radix_tree_deref_slot(pagep);
7515                 if (unlikely(!page))
7516                         break;
7517
7518                 if (radix_tree_exception(page)) {
7519                         if (radix_tree_deref_retry(page)) {
7520                                 page = NULL;
7521                                 continue;
7522                         }
7523                         /*
7524                          * Otherwise, shmem/tmpfs must be storing a swap entry
7525                          * here as an exceptional entry: so return it without
7526                          * attempting to raise page count.
7527                          */
7528                         page = NULL;
7529                         break; /* TODO: Is this relevant for this use case? */
7530                 }
7531
7532                 if (!page_cache_get_speculative(page)) {
7533                         page = NULL;
7534                         continue;
7535                 }
7536
7537                 /*
7538                  * Has the page moved?
7539                  * This is part of the lockless pagecache protocol. See
7540                  * include/linux/pagemap.h for details.
7541                  */
7542                 if (unlikely(page != *pagep)) {
7543                         put_page(page);
7544                         page = NULL;
7545                 }
7546         }
7547
7548         if (page) {
7549                 if (page->index <= end_idx)
7550                         found = true;
7551                 put_page(page);
7552         }
7553
7554         rcu_read_unlock();
7555         return found;
7556 }
7557
7558 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7559                               struct extent_state **cached_state, int writing)
7560 {
7561         struct btrfs_ordered_extent *ordered;
7562         int ret = 0;
7563
7564         while (1) {
7565                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7566                                  cached_state);
7567                 /*
7568                  * We're concerned with the entire range that we're going to be
7569                  * doing DIO to, so we need to make sure there's no ordered
7570                  * extents in this range.
7571                  */
7572                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7573                                                      lockend - lockstart + 1);
7574
7575                 /*
7576                  * We need to make sure there are no buffered pages in this
7577                  * range either, we could have raced between the invalidate in
7578                  * generic_file_direct_write and locking the extent.  The
7579                  * invalidate needs to happen so that reads after a write do not
7580                  * get stale data.
7581                  */
7582                 if (!ordered &&
7583                     (!writing ||
7584                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7585                         break;
7586
7587                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7588                                      cached_state, GFP_NOFS);
7589
7590                 if (ordered) {
7591                         /*
7592                          * If we are doing a DIO read and the ordered extent we
7593                          * found is for a buffered write, we can not wait for it
7594                          * to complete and retry, because if we do so we can
7595                          * deadlock with concurrent buffered writes on page
7596                          * locks. This happens only if our DIO read covers more
7597                          * than one extent map, if at this point has already
7598                          * created an ordered extent for a previous extent map
7599                          * and locked its range in the inode's io tree, and a
7600                          * concurrent write against that previous extent map's
7601                          * range and this range started (we unlock the ranges
7602                          * in the io tree only when the bios complete and
7603                          * buffered writes always lock pages before attempting
7604                          * to lock range in the io tree).
7605                          */
7606                         if (writing ||
7607                             test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7608                                 btrfs_start_ordered_extent(inode, ordered, 1);
7609                         else
7610                                 ret = -ENOTBLK;
7611                         btrfs_put_ordered_extent(ordered);
7612                 } else {
7613                         /*
7614                          * We could trigger writeback for this range (and wait
7615                          * for it to complete) and then invalidate the pages for
7616                          * this range (through invalidate_inode_pages2_range()),
7617                          * but that can lead us to a deadlock with a concurrent
7618                          * call to readpages() (a buffered read or a defrag call
7619                          * triggered a readahead) on a page lock due to an
7620                          * ordered dio extent we created before but did not have
7621                          * yet a corresponding bio submitted (whence it can not
7622                          * complete), which makes readpages() wait for that
7623                          * ordered extent to complete while holding a lock on
7624                          * that page.
7625                          */
7626                         ret = -ENOTBLK;
7627                 }
7628
7629                 if (ret)
7630                         break;
7631
7632                 cond_resched();
7633         }
7634
7635         return ret;
7636 }
7637
7638 /* The callers of this must take lock_extent() */
7639 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7640                                        u64 orig_start, u64 block_start,
7641                                        u64 block_len, u64 orig_block_len,
7642                                        u64 ram_bytes, int compress_type,
7643                                        int type)
7644 {
7645         struct extent_map_tree *em_tree;
7646         struct extent_map *em;
7647         struct btrfs_root *root = BTRFS_I(inode)->root;
7648         int ret;
7649
7650         ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7651                type == BTRFS_ORDERED_COMPRESSED ||
7652                type == BTRFS_ORDERED_NOCOW ||
7653                type == BTRFS_ORDERED_REGULAR);
7654
7655         em_tree = &BTRFS_I(inode)->extent_tree;
7656         em = alloc_extent_map();
7657         if (!em)
7658                 return ERR_PTR(-ENOMEM);
7659
7660         em->start = start;
7661         em->orig_start = orig_start;
7662         em->len = len;
7663         em->block_len = block_len;
7664         em->block_start = block_start;
7665         em->bdev = root->fs_info->fs_devices->latest_bdev;
7666         em->orig_block_len = orig_block_len;
7667         em->ram_bytes = ram_bytes;
7668         em->generation = -1;
7669         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7670         if (type == BTRFS_ORDERED_PREALLOC) {
7671                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7672         } else if (type == BTRFS_ORDERED_COMPRESSED) {
7673                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7674                 em->compress_type = compress_type;
7675         }
7676
7677         do {
7678                 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7679                                 em->start + em->len - 1, 0);
7680                 write_lock(&em_tree->lock);
7681                 ret = add_extent_mapping(em_tree, em, 1);
7682                 write_unlock(&em_tree->lock);
7683                 /*
7684                  * The caller has taken lock_extent(), who could race with us
7685                  * to add em?
7686                  */
7687         } while (ret == -EEXIST);
7688
7689         if (ret) {
7690                 free_extent_map(em);
7691                 return ERR_PTR(ret);
7692         }
7693
7694         /* em got 2 refs now, callers needs to do free_extent_map once. */
7695         return em;
7696 }
7697
7698 static void adjust_dio_outstanding_extents(struct inode *inode,
7699                                            struct btrfs_dio_data *dio_data,
7700                                            const u64 len)
7701 {
7702         unsigned num_extents = count_max_extents(len);
7703
7704         /*
7705          * If we have an outstanding_extents count still set then we're
7706          * within our reservation, otherwise we need to adjust our inode
7707          * counter appropriately.
7708          */
7709         if (dio_data->outstanding_extents >= num_extents) {
7710                 dio_data->outstanding_extents -= num_extents;
7711         } else {
7712                 /*
7713                  * If dio write length has been split due to no large enough
7714                  * contiguous space, we need to compensate our inode counter
7715                  * appropriately.
7716                  */
7717                 u64 num_needed = num_extents - dio_data->outstanding_extents;
7718
7719                 spin_lock(&BTRFS_I(inode)->lock);
7720                 BTRFS_I(inode)->outstanding_extents += num_needed;
7721                 spin_unlock(&BTRFS_I(inode)->lock);
7722         }
7723 }
7724
7725 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7726                                    struct buffer_head *bh_result, int create)
7727 {
7728         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7729         struct extent_map *em;
7730         struct extent_state *cached_state = NULL;
7731         struct btrfs_dio_data *dio_data = NULL;
7732         u64 start = iblock << inode->i_blkbits;
7733         u64 lockstart, lockend;
7734         u64 len = bh_result->b_size;
7735         int unlock_bits = EXTENT_LOCKED;
7736         int ret = 0;
7737
7738         if (create)
7739                 unlock_bits |= EXTENT_DIRTY;
7740         else
7741                 len = min_t(u64, len, fs_info->sectorsize);
7742
7743         lockstart = start;
7744         lockend = start + len - 1;
7745
7746         if (current->journal_info) {
7747                 /*
7748                  * Need to pull our outstanding extents and set journal_info to NULL so
7749                  * that anything that needs to check if there's a transaction doesn't get
7750                  * confused.
7751                  */
7752                 dio_data = current->journal_info;
7753                 current->journal_info = NULL;
7754         }
7755
7756         /*
7757          * If this errors out it's because we couldn't invalidate pagecache for
7758          * this range and we need to fallback to buffered.
7759          */
7760         if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7761                                create)) {
7762                 ret = -ENOTBLK;
7763                 goto err;
7764         }
7765
7766         em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7767         if (IS_ERR(em)) {
7768                 ret = PTR_ERR(em);
7769                 goto unlock_err;
7770         }
7771
7772         /*
7773          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7774          * io.  INLINE is special, and we could probably kludge it in here, but
7775          * it's still buffered so for safety lets just fall back to the generic
7776          * buffered path.
7777          *
7778          * For COMPRESSED we _have_ to read the entire extent in so we can
7779          * decompress it, so there will be buffering required no matter what we
7780          * do, so go ahead and fallback to buffered.
7781          *
7782          * We return -ENOTBLK because that's what makes DIO go ahead and go back
7783          * to buffered IO.  Don't blame me, this is the price we pay for using
7784          * the generic code.
7785          */
7786         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7787             em->block_start == EXTENT_MAP_INLINE) {
7788                 free_extent_map(em);
7789                 ret = -ENOTBLK;
7790                 goto unlock_err;
7791         }
7792
7793         /* Just a good old fashioned hole, return */
7794         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7795                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7796                 free_extent_map(em);
7797                 goto unlock_err;
7798         }
7799
7800         /*
7801          * We don't allocate a new extent in the following cases
7802          *
7803          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7804          * existing extent.
7805          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7806          * just use the extent.
7807          *
7808          */
7809         if (!create) {
7810                 len = min(len, em->len - (start - em->start));
7811                 lockstart = start + len;
7812                 goto unlock;
7813         }
7814
7815         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7816             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7817              em->block_start != EXTENT_MAP_HOLE)) {
7818                 int type;
7819                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7820
7821                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7822                         type = BTRFS_ORDERED_PREALLOC;
7823                 else
7824                         type = BTRFS_ORDERED_NOCOW;
7825                 len = min(len, em->len - (start - em->start));
7826                 block_start = em->block_start + (start - em->start);
7827
7828                 if (can_nocow_extent(inode, start, &len, &orig_start,
7829                                      &orig_block_len, &ram_bytes) == 1 &&
7830                     btrfs_inc_nocow_writers(fs_info, block_start)) {
7831                         struct extent_map *em2;
7832
7833                         em2 = btrfs_create_dio_extent(inode, start, len,
7834                                                       orig_start, block_start,
7835                                                       len, orig_block_len,
7836                                                       ram_bytes, type);
7837                         btrfs_dec_nocow_writers(fs_info, block_start);
7838                         if (type == BTRFS_ORDERED_PREALLOC) {
7839                                 free_extent_map(em);
7840                                 em = em2;
7841                         }
7842                         if (em2 && IS_ERR(em2)) {
7843                                 ret = PTR_ERR(em2);
7844                                 goto unlock_err;
7845                         }
7846                         /*
7847                          * For inode marked NODATACOW or extent marked PREALLOC,
7848                          * use the existing or preallocated extent, so does not
7849                          * need to adjust btrfs_space_info's bytes_may_use.
7850                          */
7851                         btrfs_free_reserved_data_space_noquota(inode,
7852                                         start, len);
7853                         goto unlock;
7854                 }
7855         }
7856
7857         /*
7858          * this will cow the extent, reset the len in case we changed
7859          * it above
7860          */
7861         len = bh_result->b_size;
7862         free_extent_map(em);
7863         em = btrfs_new_extent_direct(inode, start, len);
7864         if (IS_ERR(em)) {
7865                 ret = PTR_ERR(em);
7866                 goto unlock_err;
7867         }
7868         len = min(len, em->len - (start - em->start));
7869 unlock:
7870         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7871                 inode->i_blkbits;
7872         bh_result->b_size = len;
7873         bh_result->b_bdev = em->bdev;
7874         set_buffer_mapped(bh_result);
7875         if (create) {
7876                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7877                         set_buffer_new(bh_result);
7878
7879                 /*
7880                  * Need to update the i_size under the extent lock so buffered
7881                  * readers will get the updated i_size when we unlock.
7882                  */
7883                 if (!dio_data->overwrite && start + len > i_size_read(inode))
7884                         i_size_write(inode, start + len);
7885
7886                 adjust_dio_outstanding_extents(inode, dio_data, len);
7887                 WARN_ON(dio_data->reserve < len);
7888                 dio_data->reserve -= len;
7889                 dio_data->unsubmitted_oe_range_end = start + len;
7890                 current->journal_info = dio_data;
7891         }
7892
7893         /*
7894          * In the case of write we need to clear and unlock the entire range,
7895          * in the case of read we need to unlock only the end area that we
7896          * aren't using if there is any left over space.
7897          */
7898         if (lockstart < lockend) {
7899                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7900                                  lockend, unlock_bits, 1, 0,
7901                                  &cached_state, GFP_NOFS);
7902         } else {
7903                 free_extent_state(cached_state);
7904         }
7905
7906         free_extent_map(em);
7907
7908         return 0;
7909
7910 unlock_err:
7911         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7912                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7913 err:
7914         if (dio_data)
7915                 current->journal_info = dio_data;
7916         /*
7917          * Compensate the delalloc release we do in btrfs_direct_IO() when we
7918          * write less data then expected, so that we don't underflow our inode's
7919          * outstanding extents counter.
7920          */
7921         if (create && dio_data)
7922                 adjust_dio_outstanding_extents(inode, dio_data, len);
7923
7924         return ret;
7925 }
7926
7927 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7928                                                  struct bio *bio,
7929                                                  int mirror_num)
7930 {
7931         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7932         blk_status_t ret;
7933
7934         BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7935
7936         bio_get(bio);
7937
7938         ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7939         if (ret)
7940                 goto err;
7941
7942         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7943 err:
7944         bio_put(bio);
7945         return ret;
7946 }
7947
7948 static int btrfs_check_dio_repairable(struct inode *inode,
7949                                       struct bio *failed_bio,
7950                                       struct io_failure_record *failrec,
7951                                       int failed_mirror)
7952 {
7953         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7954         int num_copies;
7955
7956         num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7957         if (num_copies == 1) {
7958                 /*
7959                  * we only have a single copy of the data, so don't bother with
7960                  * all the retry and error correction code that follows. no
7961                  * matter what the error is, it is very likely to persist.
7962                  */
7963                 btrfs_debug(fs_info,
7964                         "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7965                         num_copies, failrec->this_mirror, failed_mirror);
7966                 return 0;
7967         }
7968
7969         failrec->failed_mirror = failed_mirror;
7970         failrec->this_mirror++;
7971         if (failrec->this_mirror == failed_mirror)
7972                 failrec->this_mirror++;
7973
7974         if (failrec->this_mirror > num_copies) {
7975                 btrfs_debug(fs_info,
7976                         "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7977                         num_copies, failrec->this_mirror, failed_mirror);
7978                 return 0;
7979         }
7980
7981         return 1;
7982 }
7983
7984 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7985                                    struct page *page, unsigned int pgoff,
7986                                    u64 start, u64 end, int failed_mirror,
7987                                    bio_end_io_t *repair_endio, void *repair_arg)
7988 {
7989         struct io_failure_record *failrec;
7990         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7991         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7992         struct bio *bio;
7993         int isector;
7994         int read_mode = 0;
7995         int segs;
7996         int ret;
7997         blk_status_t status;
7998
7999         BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
8000
8001         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8002         if (ret)
8003                 return errno_to_blk_status(ret);
8004
8005         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8006                                          failed_mirror);
8007         if (!ret) {
8008                 free_io_failure(failure_tree, io_tree, failrec);
8009                 return BLK_STS_IOERR;
8010         }
8011
8012         segs = bio_segments(failed_bio);
8013         if (segs > 1 ||
8014             (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8015                 read_mode |= REQ_FAILFAST_DEV;
8016
8017         isector = start - btrfs_io_bio(failed_bio)->logical;
8018         isector >>= inode->i_sb->s_blocksize_bits;
8019         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8020                                 pgoff, isector, repair_endio, repair_arg);
8021         bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8022
8023         btrfs_debug(BTRFS_I(inode)->root->fs_info,
8024                     "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
8025                     read_mode, failrec->this_mirror, failrec->in_validation);
8026
8027         status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8028         if (status) {
8029                 free_io_failure(failure_tree, io_tree, failrec);
8030                 bio_put(bio);
8031         }
8032
8033         return status;
8034 }
8035
8036 struct btrfs_retry_complete {
8037         struct completion done;
8038         struct inode *inode;
8039         u64 start;
8040         int uptodate;
8041 };
8042
8043 static void btrfs_retry_endio_nocsum(struct bio *bio)
8044 {
8045         struct btrfs_retry_complete *done = bio->bi_private;
8046         struct inode *inode = done->inode;
8047         struct bio_vec *bvec;
8048         struct extent_io_tree *io_tree, *failure_tree;
8049         int i;
8050
8051         if (bio->bi_status)
8052                 goto end;
8053
8054         ASSERT(bio->bi_vcnt == 1);
8055         io_tree = &BTRFS_I(inode)->io_tree;
8056         failure_tree = &BTRFS_I(inode)->io_failure_tree;
8057         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8058
8059         done->uptodate = 1;
8060         ASSERT(!bio_flagged(bio, BIO_CLONED));
8061         bio_for_each_segment_all(bvec, bio, i)
8062                 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8063                                  io_tree, done->start, bvec->bv_page,
8064                                  btrfs_ino(BTRFS_I(inode)), 0);
8065 end:
8066         complete(&done->done);
8067         bio_put(bio);
8068 }
8069
8070 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
8071                                                 struct btrfs_io_bio *io_bio)
8072 {
8073         struct btrfs_fs_info *fs_info;
8074         struct bio_vec bvec;
8075         struct bvec_iter iter;
8076         struct btrfs_retry_complete done;
8077         u64 start;
8078         unsigned int pgoff;
8079         u32 sectorsize;
8080         int nr_sectors;
8081         blk_status_t ret;
8082         blk_status_t err = BLK_STS_OK;
8083
8084         fs_info = BTRFS_I(inode)->root->fs_info;
8085         sectorsize = fs_info->sectorsize;
8086
8087         start = io_bio->logical;
8088         done.inode = inode;
8089         io_bio->bio.bi_iter = io_bio->iter;
8090
8091         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8092                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8093                 pgoff = bvec.bv_offset;
8094
8095 next_block_or_try_again:
8096                 done.uptodate = 0;
8097                 done.start = start;
8098                 init_completion(&done.done);
8099
8100                 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8101                                 pgoff, start, start + sectorsize - 1,
8102                                 io_bio->mirror_num,
8103                                 btrfs_retry_endio_nocsum, &done);
8104                 if (ret) {
8105                         err = ret;
8106                         goto next;
8107                 }
8108
8109                 wait_for_completion(&done.done);
8110
8111                 if (!done.uptodate) {
8112                         /* We might have another mirror, so try again */
8113                         goto next_block_or_try_again;
8114                 }
8115
8116 next:
8117                 start += sectorsize;
8118
8119                 nr_sectors--;
8120                 if (nr_sectors) {
8121                         pgoff += sectorsize;
8122                         ASSERT(pgoff < PAGE_SIZE);
8123                         goto next_block_or_try_again;
8124                 }
8125         }
8126
8127         return err;
8128 }
8129
8130 static void btrfs_retry_endio(struct bio *bio)
8131 {
8132         struct btrfs_retry_complete *done = bio->bi_private;
8133         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8134         struct extent_io_tree *io_tree, *failure_tree;
8135         struct inode *inode = done->inode;
8136         struct bio_vec *bvec;
8137         int uptodate;
8138         int ret;
8139         int i;
8140
8141         if (bio->bi_status)
8142                 goto end;
8143
8144         uptodate = 1;
8145
8146         ASSERT(bio->bi_vcnt == 1);
8147         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8148
8149         io_tree = &BTRFS_I(inode)->io_tree;
8150         failure_tree = &BTRFS_I(inode)->io_failure_tree;
8151
8152         ASSERT(!bio_flagged(bio, BIO_CLONED));
8153         bio_for_each_segment_all(bvec, bio, i) {
8154                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8155                                              bvec->bv_offset, done->start,
8156                                              bvec->bv_len);
8157                 if (!ret)
8158                         clean_io_failure(BTRFS_I(inode)->root->fs_info,
8159                                          failure_tree, io_tree, done->start,
8160                                          bvec->bv_page,
8161                                          btrfs_ino(BTRFS_I(inode)),
8162                                          bvec->bv_offset);
8163                 else
8164                         uptodate = 0;
8165         }
8166
8167         done->uptodate = uptodate;
8168 end:
8169         complete(&done->done);
8170         bio_put(bio);
8171 }
8172
8173 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8174                 struct btrfs_io_bio *io_bio, blk_status_t err)
8175 {
8176         struct btrfs_fs_info *fs_info;
8177         struct bio_vec bvec;
8178         struct bvec_iter iter;
8179         struct btrfs_retry_complete done;
8180         u64 start;
8181         u64 offset = 0;
8182         u32 sectorsize;
8183         int nr_sectors;
8184         unsigned int pgoff;
8185         int csum_pos;
8186         bool uptodate = (err == 0);
8187         int ret;
8188         blk_status_t status;
8189
8190         fs_info = BTRFS_I(inode)->root->fs_info;
8191         sectorsize = fs_info->sectorsize;
8192
8193         err = BLK_STS_OK;
8194         start = io_bio->logical;
8195         done.inode = inode;
8196         io_bio->bio.bi_iter = io_bio->iter;
8197
8198         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8199                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8200
8201                 pgoff = bvec.bv_offset;
8202 next_block:
8203                 if (uptodate) {
8204                         csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8205                         ret = __readpage_endio_check(inode, io_bio, csum_pos,
8206                                         bvec.bv_page, pgoff, start, sectorsize);
8207                         if (likely(!ret))
8208                                 goto next;
8209                 }
8210 try_again:
8211                 done.uptodate = 0;
8212                 done.start = start;
8213                 init_completion(&done.done);
8214
8215                 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8216                                         pgoff, start, start + sectorsize - 1,
8217                                         io_bio->mirror_num, btrfs_retry_endio,
8218                                         &done);
8219                 if (status) {
8220                         err = status;
8221                         goto next;
8222                 }
8223
8224                 wait_for_completion(&done.done);
8225
8226                 if (!done.uptodate) {
8227                         /* We might have another mirror, so try again */
8228                         goto try_again;
8229                 }
8230 next:
8231                 offset += sectorsize;
8232                 start += sectorsize;
8233
8234                 ASSERT(nr_sectors);
8235
8236                 nr_sectors--;
8237                 if (nr_sectors) {
8238                         pgoff += sectorsize;
8239                         ASSERT(pgoff < PAGE_SIZE);
8240                         goto next_block;
8241                 }
8242         }
8243
8244         return err;
8245 }
8246
8247 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8248                 struct btrfs_io_bio *io_bio, blk_status_t err)
8249 {
8250         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8251
8252         if (skip_csum) {
8253                 if (unlikely(err))
8254                         return __btrfs_correct_data_nocsum(inode, io_bio);
8255                 else
8256                         return BLK_STS_OK;
8257         } else {
8258                 return __btrfs_subio_endio_read(inode, io_bio, err);
8259         }
8260 }
8261
8262 static void btrfs_endio_direct_read(struct bio *bio)
8263 {
8264         struct btrfs_dio_private *dip = bio->bi_private;
8265         struct inode *inode = dip->inode;
8266         struct bio *dio_bio;
8267         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8268         blk_status_t err = bio->bi_status;
8269
8270         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED) {
8271                 err = btrfs_subio_endio_read(inode, io_bio, err);
8272                 if (!err)
8273                         bio->bi_status = 0;
8274         }
8275
8276         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8277                       dip->logical_offset + dip->bytes - 1);
8278         dio_bio = dip->dio_bio;
8279
8280         kfree(dip);
8281
8282         dio_bio->bi_status = bio->bi_status;
8283         dio_end_io(dio_bio);
8284
8285         if (io_bio->end_io)
8286                 io_bio->end_io(io_bio, blk_status_to_errno(err));
8287         bio_put(bio);
8288 }
8289
8290 static void __endio_write_update_ordered(struct inode *inode,
8291                                          const u64 offset, const u64 bytes,
8292                                          const bool uptodate)
8293 {
8294         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8295         struct btrfs_ordered_extent *ordered = NULL;
8296         struct btrfs_workqueue *wq;
8297         btrfs_work_func_t func;
8298         u64 ordered_offset = offset;
8299         u64 ordered_bytes = bytes;
8300         int ret;
8301
8302         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8303                 wq = fs_info->endio_freespace_worker;
8304                 func = btrfs_freespace_write_helper;
8305         } else {
8306                 wq = fs_info->endio_write_workers;
8307                 func = btrfs_endio_write_helper;
8308         }
8309
8310 again:
8311         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8312                                                    &ordered_offset,
8313                                                    ordered_bytes,
8314                                                    uptodate);
8315         if (!ret)
8316                 goto out_test;
8317
8318         btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8319         btrfs_queue_work(wq, &ordered->work);
8320 out_test:
8321         /*
8322          * our bio might span multiple ordered extents.  If we haven't
8323          * completed the accounting for the whole dio, go back and try again
8324          */
8325         if (ordered_offset < offset + bytes) {
8326                 ordered_bytes = offset + bytes - ordered_offset;
8327                 ordered = NULL;
8328                 goto again;
8329         }
8330 }
8331
8332 static void btrfs_endio_direct_write(struct bio *bio)
8333 {
8334         struct btrfs_dio_private *dip = bio->bi_private;
8335         struct bio *dio_bio = dip->dio_bio;
8336
8337         __endio_write_update_ordered(dip->inode, dip->logical_offset,
8338                                      dip->bytes, !bio->bi_status);
8339
8340         kfree(dip);
8341
8342         dio_bio->bi_status = bio->bi_status;
8343         dio_end_io(dio_bio);
8344         bio_put(bio);
8345 }
8346
8347 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8348                                     struct bio *bio, int mirror_num,
8349                                     unsigned long bio_flags, u64 offset)
8350 {
8351         struct inode *inode = private_data;
8352         blk_status_t ret;
8353         ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8354         BUG_ON(ret); /* -ENOMEM */
8355         return 0;
8356 }
8357
8358 static void btrfs_end_dio_bio(struct bio *bio)
8359 {
8360         struct btrfs_dio_private *dip = bio->bi_private;
8361         blk_status_t err = bio->bi_status;
8362
8363         if (err)
8364                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8365                            "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8366                            btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8367                            bio->bi_opf,
8368                            (unsigned long long)bio->bi_iter.bi_sector,
8369                            bio->bi_iter.bi_size, err);
8370
8371         if (dip->subio_endio)
8372                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8373
8374         if (err) {
8375                 dip->errors = 1;
8376
8377                 /*
8378                  * before atomic variable goto zero, we must make sure
8379                  * dip->errors is perceived to be set.
8380                  */
8381                 smp_mb__before_atomic();
8382         }
8383
8384         /* if there are more bios still pending for this dio, just exit */
8385         if (!atomic_dec_and_test(&dip->pending_bios))
8386                 goto out;
8387
8388         if (dip->errors) {
8389                 bio_io_error(dip->orig_bio);
8390         } else {
8391                 dip->dio_bio->bi_status = 0;
8392                 bio_endio(dip->orig_bio);
8393         }
8394 out:
8395         bio_put(bio);
8396 }
8397
8398 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8399                                                  struct btrfs_dio_private *dip,
8400                                                  struct bio *bio,
8401                                                  u64 file_offset)
8402 {
8403         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8404         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8405         blk_status_t ret;
8406
8407         /*
8408          * We load all the csum data we need when we submit
8409          * the first bio to reduce the csum tree search and
8410          * contention.
8411          */
8412         if (dip->logical_offset == file_offset) {
8413                 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8414                                                 file_offset);
8415                 if (ret)
8416                         return ret;
8417         }
8418
8419         if (bio == dip->orig_bio)
8420                 return 0;
8421
8422         file_offset -= dip->logical_offset;
8423         file_offset >>= inode->i_sb->s_blocksize_bits;
8424         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8425
8426         return 0;
8427 }
8428
8429 static inline blk_status_t
8430 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8431                        int skip_sum, int async_submit)
8432 {
8433         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8434         struct btrfs_dio_private *dip = bio->bi_private;
8435         bool write = bio_op(bio) == REQ_OP_WRITE;
8436         blk_status_t ret;
8437
8438         if (async_submit)
8439                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8440
8441         bio_get(bio);
8442
8443         if (!write) {
8444                 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8445                 if (ret)
8446                         goto err;
8447         }
8448
8449         if (skip_sum)
8450                 goto map;
8451
8452         if (write && async_submit) {
8453                 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8454                                           file_offset, inode,
8455                                           __btrfs_submit_bio_start_direct_io,
8456                                           __btrfs_submit_bio_done);
8457                 goto err;
8458         } else if (write) {
8459                 /*
8460                  * If we aren't doing async submit, calculate the csum of the
8461                  * bio now.
8462                  */
8463                 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8464                 if (ret)
8465                         goto err;
8466         } else {
8467                 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8468                                                      file_offset);
8469                 if (ret)
8470                         goto err;
8471         }
8472 map:
8473         ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8474 err:
8475         bio_put(bio);
8476         return ret;
8477 }
8478
8479 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8480                                     int skip_sum)
8481 {
8482         struct inode *inode = dip->inode;
8483         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8484         struct bio *bio;
8485         struct bio *orig_bio = dip->orig_bio;
8486         u64 start_sector = orig_bio->bi_iter.bi_sector;
8487         u64 file_offset = dip->logical_offset;
8488         u64 map_length;
8489         int async_submit = 0;
8490         u64 submit_len;
8491         int clone_offset = 0;
8492         int clone_len;
8493         int ret;
8494         blk_status_t status;
8495
8496         map_length = orig_bio->bi_iter.bi_size;
8497         submit_len = map_length;
8498         ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8499                               &map_length, NULL, 0);
8500         if (ret)
8501                 return -EIO;
8502
8503         if (map_length >= submit_len) {
8504                 bio = orig_bio;
8505                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8506                 goto submit;
8507         }
8508
8509         /* async crcs make it difficult to collect full stripe writes. */
8510         if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8511                 async_submit = 0;
8512         else
8513                 async_submit = 1;
8514
8515         /* bio split */
8516         ASSERT(map_length <= INT_MAX);
8517         atomic_inc(&dip->pending_bios);
8518         do {
8519                 clone_len = min_t(int, submit_len, map_length);
8520
8521                 /*
8522                  * This will never fail as it's passing GPF_NOFS and
8523                  * the allocation is backed by btrfs_bioset.
8524                  */
8525                 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8526                                               clone_len);
8527                 bio->bi_private = dip;
8528                 bio->bi_end_io = btrfs_end_dio_bio;
8529                 btrfs_io_bio(bio)->logical = file_offset;
8530
8531                 ASSERT(submit_len >= clone_len);
8532                 submit_len -= clone_len;
8533                 if (submit_len == 0)
8534                         break;
8535
8536                 /*
8537                  * Increase the count before we submit the bio so we know
8538                  * the end IO handler won't happen before we increase the
8539                  * count. Otherwise, the dip might get freed before we're
8540                  * done setting it up.
8541                  */
8542                 atomic_inc(&dip->pending_bios);
8543
8544                 status = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8545                                                 async_submit);
8546                 if (status) {
8547                         bio_put(bio);
8548                         atomic_dec(&dip->pending_bios);
8549                         goto out_err;
8550                 }
8551
8552                 clone_offset += clone_len;
8553                 start_sector += clone_len >> 9;
8554                 file_offset += clone_len;
8555
8556                 map_length = submit_len;
8557                 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8558                                       start_sector << 9, &map_length, NULL, 0);
8559                 if (ret)
8560                         goto out_err;
8561         } while (submit_len > 0);
8562
8563 submit:
8564         status = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8565                                         async_submit);
8566         if (!status)
8567                 return 0;
8568
8569         bio_put(bio);
8570 out_err:
8571         dip->errors = 1;
8572         /*
8573          * before atomic variable goto zero, we must
8574          * make sure dip->errors is perceived to be set.
8575          */
8576         smp_mb__before_atomic();
8577         if (atomic_dec_and_test(&dip->pending_bios))
8578                 bio_io_error(dip->orig_bio);
8579
8580         /* bio_end_io() will handle error, so we needn't return it */
8581         return 0;
8582 }
8583
8584 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8585                                 loff_t file_offset)
8586 {
8587         struct btrfs_dio_private *dip = NULL;
8588         struct bio *bio = NULL;
8589         struct btrfs_io_bio *io_bio;
8590         int skip_sum;
8591         bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8592         int ret = 0;
8593
8594         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8595
8596         bio = btrfs_bio_clone(dio_bio);
8597
8598         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8599         if (!dip) {
8600                 ret = -ENOMEM;
8601                 goto free_ordered;
8602         }
8603
8604         dip->private = dio_bio->bi_private;
8605         dip->inode = inode;
8606         dip->logical_offset = file_offset;
8607         dip->bytes = dio_bio->bi_iter.bi_size;
8608         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8609         bio->bi_private = dip;
8610         dip->orig_bio = bio;
8611         dip->dio_bio = dio_bio;
8612         atomic_set(&dip->pending_bios, 0);
8613         io_bio = btrfs_io_bio(bio);
8614         io_bio->logical = file_offset;
8615
8616         if (write) {
8617                 bio->bi_end_io = btrfs_endio_direct_write;
8618         } else {
8619                 bio->bi_end_io = btrfs_endio_direct_read;
8620                 dip->subio_endio = btrfs_subio_endio_read;
8621         }
8622
8623         /*
8624          * Reset the range for unsubmitted ordered extents (to a 0 length range)
8625          * even if we fail to submit a bio, because in such case we do the
8626          * corresponding error handling below and it must not be done a second
8627          * time by btrfs_direct_IO().
8628          */
8629         if (write) {
8630                 struct btrfs_dio_data *dio_data = current->journal_info;
8631
8632                 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8633                         dip->bytes;
8634                 dio_data->unsubmitted_oe_range_start =
8635                         dio_data->unsubmitted_oe_range_end;
8636         }
8637
8638         ret = btrfs_submit_direct_hook(dip, skip_sum);
8639         if (!ret)
8640                 return;
8641
8642         if (io_bio->end_io)
8643                 io_bio->end_io(io_bio, ret);
8644
8645 free_ordered:
8646         /*
8647          * If we arrived here it means either we failed to submit the dip
8648          * or we either failed to clone the dio_bio or failed to allocate the
8649          * dip. If we cloned the dio_bio and allocated the dip, we can just
8650          * call bio_endio against our io_bio so that we get proper resource
8651          * cleanup if we fail to submit the dip, otherwise, we must do the
8652          * same as btrfs_endio_direct_[write|read] because we can't call these
8653          * callbacks - they require an allocated dip and a clone of dio_bio.
8654          */
8655         if (bio && dip) {
8656                 bio_io_error(bio);
8657                 /*
8658                  * The end io callbacks free our dip, do the final put on bio
8659                  * and all the cleanup and final put for dio_bio (through
8660                  * dio_end_io()).
8661                  */
8662                 dip = NULL;
8663                 bio = NULL;
8664         } else {
8665                 if (write)
8666                         __endio_write_update_ordered(inode,
8667                                                 file_offset,
8668                                                 dio_bio->bi_iter.bi_size,
8669                                                 false);
8670                 else
8671                         unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8672                               file_offset + dio_bio->bi_iter.bi_size - 1);
8673
8674                 dio_bio->bi_status = BLK_STS_IOERR;
8675                 /*
8676                  * Releases and cleans up our dio_bio, no need to bio_put()
8677                  * nor bio_endio()/bio_io_error() against dio_bio.
8678                  */
8679                 dio_end_io(dio_bio);
8680         }
8681         if (bio)
8682                 bio_put(bio);
8683         kfree(dip);
8684 }
8685
8686 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8687                                struct kiocb *iocb,
8688                                const struct iov_iter *iter, loff_t offset)
8689 {
8690         int seg;
8691         int i;
8692         unsigned int blocksize_mask = fs_info->sectorsize - 1;
8693         ssize_t retval = -EINVAL;
8694
8695         if (offset & blocksize_mask)
8696                 goto out;
8697
8698         if (iov_iter_alignment(iter) & blocksize_mask)
8699                 goto out;
8700
8701         /* If this is a write we don't need to check anymore */
8702         if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8703                 return 0;
8704         /*
8705          * Check to make sure we don't have duplicate iov_base's in this
8706          * iovec, if so return EINVAL, otherwise we'll get csum errors
8707          * when reading back.
8708          */
8709         for (seg = 0; seg < iter->nr_segs; seg++) {
8710                 for (i = seg + 1; i < iter->nr_segs; i++) {
8711                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8712                                 goto out;
8713                 }
8714         }
8715         retval = 0;
8716 out:
8717         return retval;
8718 }
8719
8720 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8721 {
8722         struct file *file = iocb->ki_filp;
8723         struct inode *inode = file->f_mapping->host;
8724         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8725         struct btrfs_dio_data dio_data = { 0 };
8726         struct extent_changeset *data_reserved = NULL;
8727         loff_t offset = iocb->ki_pos;
8728         size_t count = 0;
8729         int flags = 0;
8730         bool wakeup = true;
8731         bool relock = false;
8732         ssize_t ret;
8733
8734         if (check_direct_IO(fs_info, iocb, iter, offset))
8735                 return 0;
8736
8737         inode_dio_begin(inode);
8738         smp_mb__after_atomic();
8739
8740         /*
8741          * The generic stuff only does filemap_write_and_wait_range, which
8742          * isn't enough if we've written compressed pages to this area, so
8743          * we need to flush the dirty pages again to make absolutely sure
8744          * that any outstanding dirty pages are on disk.
8745          */
8746         count = iov_iter_count(iter);
8747         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8748                      &BTRFS_I(inode)->runtime_flags))
8749                 filemap_fdatawrite_range(inode->i_mapping, offset,
8750                                          offset + count - 1);
8751
8752         if (iov_iter_rw(iter) == WRITE) {
8753                 /*
8754                  * If the write DIO is beyond the EOF, we need update
8755                  * the isize, but it is protected by i_mutex. So we can
8756                  * not unlock the i_mutex at this case.
8757                  */
8758                 if (offset + count <= inode->i_size) {
8759                         dio_data.overwrite = 1;
8760                         inode_unlock(inode);
8761                         relock = true;
8762                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8763                         ret = -EAGAIN;
8764                         goto out;
8765                 }
8766                 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8767                                                    offset, count);
8768                 if (ret)
8769                         goto out;
8770                 dio_data.outstanding_extents = count_max_extents(count);
8771
8772                 /*
8773                  * We need to know how many extents we reserved so that we can
8774                  * do the accounting properly if we go over the number we
8775                  * originally calculated.  Abuse current->journal_info for this.
8776                  */
8777                 dio_data.reserve = round_up(count,
8778                                             fs_info->sectorsize);
8779                 dio_data.unsubmitted_oe_range_start = (u64)offset;
8780                 dio_data.unsubmitted_oe_range_end = (u64)offset;
8781                 current->journal_info = &dio_data;
8782                 down_read(&BTRFS_I(inode)->dio_sem);
8783         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8784                                      &BTRFS_I(inode)->runtime_flags)) {
8785                 inode_dio_end(inode);
8786                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8787                 wakeup = false;
8788         }
8789
8790         ret = __blockdev_direct_IO(iocb, inode,
8791                                    fs_info->fs_devices->latest_bdev,
8792                                    iter, btrfs_get_blocks_direct, NULL,
8793                                    btrfs_submit_direct, flags);
8794         if (iov_iter_rw(iter) == WRITE) {
8795                 up_read(&BTRFS_I(inode)->dio_sem);
8796                 current->journal_info = NULL;
8797                 if (ret < 0 && ret != -EIOCBQUEUED) {
8798                         if (dio_data.reserve)
8799                                 btrfs_delalloc_release_space(inode, data_reserved,
8800                                         offset, dio_data.reserve);
8801                         /*
8802                          * On error we might have left some ordered extents
8803                          * without submitting corresponding bios for them, so
8804                          * cleanup them up to avoid other tasks getting them
8805                          * and waiting for them to complete forever.
8806                          */
8807                         if (dio_data.unsubmitted_oe_range_start <
8808                             dio_data.unsubmitted_oe_range_end)
8809                                 __endio_write_update_ordered(inode,
8810                                         dio_data.unsubmitted_oe_range_start,
8811                                         dio_data.unsubmitted_oe_range_end -
8812                                         dio_data.unsubmitted_oe_range_start,
8813                                         false);
8814                 } else if (ret >= 0 && (size_t)ret < count)
8815                         btrfs_delalloc_release_space(inode, data_reserved,
8816                                         offset, count - (size_t)ret);
8817         }
8818 out:
8819         if (wakeup)
8820                 inode_dio_end(inode);
8821         if (relock)
8822                 inode_lock(inode);
8823
8824         extent_changeset_free(data_reserved);
8825         return ret;
8826 }
8827
8828 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8829
8830 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8831                 __u64 start, __u64 len)
8832 {
8833         int     ret;
8834
8835         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8836         if (ret)
8837                 return ret;
8838
8839         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8840 }
8841
8842 int btrfs_readpage(struct file *file, struct page *page)
8843 {
8844         struct extent_io_tree *tree;
8845         tree = &BTRFS_I(page->mapping->host)->io_tree;
8846         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8847 }
8848
8849 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8850 {
8851         struct extent_io_tree *tree;
8852         struct inode *inode = page->mapping->host;
8853         int ret;
8854
8855         if (current->flags & PF_MEMALLOC) {
8856                 redirty_page_for_writepage(wbc, page);
8857                 unlock_page(page);
8858                 return 0;
8859         }
8860
8861         /*
8862          * If we are under memory pressure we will call this directly from the
8863          * VM, we need to make sure we have the inode referenced for the ordered
8864          * extent.  If not just return like we didn't do anything.
8865          */
8866         if (!igrab(inode)) {
8867                 redirty_page_for_writepage(wbc, page);
8868                 return AOP_WRITEPAGE_ACTIVATE;
8869         }
8870         tree = &BTRFS_I(page->mapping->host)->io_tree;
8871         ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8872         btrfs_add_delayed_iput(inode);
8873         return ret;
8874 }
8875
8876 static int btrfs_writepages(struct address_space *mapping,
8877                             struct writeback_control *wbc)
8878 {
8879         struct extent_io_tree *tree;
8880
8881         tree = &BTRFS_I(mapping->host)->io_tree;
8882         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8883 }
8884
8885 static int
8886 btrfs_readpages(struct file *file, struct address_space *mapping,
8887                 struct list_head *pages, unsigned nr_pages)
8888 {
8889         struct extent_io_tree *tree;
8890         tree = &BTRFS_I(mapping->host)->io_tree;
8891         return extent_readpages(tree, mapping, pages, nr_pages,
8892                                 btrfs_get_extent);
8893 }
8894 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8895 {
8896         struct extent_io_tree *tree;
8897         struct extent_map_tree *map;
8898         int ret;
8899
8900         tree = &BTRFS_I(page->mapping->host)->io_tree;
8901         map = &BTRFS_I(page->mapping->host)->extent_tree;
8902         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8903         if (ret == 1) {
8904                 ClearPagePrivate(page);
8905                 set_page_private(page, 0);
8906                 put_page(page);
8907         }
8908         return ret;
8909 }
8910
8911 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8912 {
8913         if (PageWriteback(page) || PageDirty(page))
8914                 return 0;
8915         return __btrfs_releasepage(page, gfp_flags);
8916 }
8917
8918 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8919                                  unsigned int length)
8920 {
8921         struct inode *inode = page->mapping->host;
8922         struct extent_io_tree *tree;
8923         struct btrfs_ordered_extent *ordered;
8924         struct extent_state *cached_state = NULL;
8925         u64 page_start = page_offset(page);
8926         u64 page_end = page_start + PAGE_SIZE - 1;
8927         u64 start;
8928         u64 end;
8929         int inode_evicting = inode->i_state & I_FREEING;
8930
8931         /*
8932          * we have the page locked, so new writeback can't start,
8933          * and the dirty bit won't be cleared while we are here.
8934          *
8935          * Wait for IO on this page so that we can safely clear
8936          * the PagePrivate2 bit and do ordered accounting
8937          */
8938         wait_on_page_writeback(page);
8939
8940         tree = &BTRFS_I(inode)->io_tree;
8941         if (offset) {
8942                 btrfs_releasepage(page, GFP_NOFS);
8943                 return;
8944         }
8945
8946         if (!inode_evicting)
8947                 lock_extent_bits(tree, page_start, page_end, &cached_state);
8948 again:
8949         start = page_start;
8950         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8951                                         page_end - start + 1);
8952         if (ordered) {
8953                 end = min(page_end, ordered->file_offset + ordered->len - 1);
8954                 /*
8955                  * IO on this page will never be started, so we need
8956                  * to account for any ordered extents now
8957                  */
8958                 if (!inode_evicting)
8959                         clear_extent_bit(tree, start, end,
8960                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8961                                          EXTENT_DELALLOC_NEW |
8962                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8963                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8964                                          GFP_NOFS);
8965                 /*
8966                  * whoever cleared the private bit is responsible
8967                  * for the finish_ordered_io
8968                  */
8969                 if (TestClearPagePrivate2(page)) {
8970                         struct btrfs_ordered_inode_tree *tree;
8971                         u64 new_len;
8972
8973                         tree = &BTRFS_I(inode)->ordered_tree;
8974
8975                         spin_lock_irq(&tree->lock);
8976                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8977                         new_len = start - ordered->file_offset;
8978                         if (new_len < ordered->truncated_len)
8979                                 ordered->truncated_len = new_len;
8980                         spin_unlock_irq(&tree->lock);
8981
8982                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8983                                                            start,
8984                                                            end - start + 1, 1))
8985                                 btrfs_finish_ordered_io(ordered);
8986                 }
8987                 btrfs_put_ordered_extent(ordered);
8988                 if (!inode_evicting) {
8989                         cached_state = NULL;
8990                         lock_extent_bits(tree, start, end,
8991                                          &cached_state);
8992                 }
8993
8994                 start = end + 1;
8995                 if (start < page_end)
8996                         goto again;
8997         }
8998
8999         /*
9000          * Qgroup reserved space handler
9001          * Page here will be either
9002          * 1) Already written to disk
9003          *    In this case, its reserved space is released from data rsv map
9004          *    and will be freed by delayed_ref handler finally.
9005          *    So even we call qgroup_free_data(), it won't decrease reserved
9006          *    space.
9007          * 2) Not written to disk
9008          *    This means the reserved space should be freed here. However,
9009          *    if a truncate invalidates the page (by clearing PageDirty)
9010          *    and the page is accounted for while allocating extent
9011          *    in btrfs_check_data_free_space() we let delayed_ref to
9012          *    free the entire extent.
9013          */
9014         if (PageDirty(page))
9015                 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
9016         if (!inode_evicting) {
9017                 clear_extent_bit(tree, page_start, page_end,
9018                                  EXTENT_LOCKED | EXTENT_DIRTY |
9019                                  EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9020                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9021                                  &cached_state, GFP_NOFS);
9022
9023                 __btrfs_releasepage(page, GFP_NOFS);
9024         }
9025
9026         ClearPageChecked(page);
9027         if (PagePrivate(page)) {
9028                 ClearPagePrivate(page);
9029                 set_page_private(page, 0);
9030                 put_page(page);
9031         }
9032 }
9033
9034 /*
9035  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9036  * called from a page fault handler when a page is first dirtied. Hence we must
9037  * be careful to check for EOF conditions here. We set the page up correctly
9038  * for a written page which means we get ENOSPC checking when writing into
9039  * holes and correct delalloc and unwritten extent mapping on filesystems that
9040  * support these features.
9041  *
9042  * We are not allowed to take the i_mutex here so we have to play games to
9043  * protect against truncate races as the page could now be beyond EOF.  Because
9044  * vmtruncate() writes the inode size before removing pages, once we have the
9045  * page lock we can determine safely if the page is beyond EOF. If it is not
9046  * beyond EOF, then the page is guaranteed safe against truncation until we
9047  * unlock the page.
9048  */
9049 int btrfs_page_mkwrite(struct vm_fault *vmf)
9050 {
9051         struct page *page = vmf->page;
9052         struct inode *inode = file_inode(vmf->vma->vm_file);
9053         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9054         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9055         struct btrfs_ordered_extent *ordered;
9056         struct extent_state *cached_state = NULL;
9057         struct extent_changeset *data_reserved = NULL;
9058         char *kaddr;
9059         unsigned long zero_start;
9060         loff_t size;
9061         int ret;
9062         int reserved = 0;
9063         u64 reserved_space;
9064         u64 page_start;
9065         u64 page_end;
9066         u64 end;
9067
9068         reserved_space = PAGE_SIZE;
9069
9070         sb_start_pagefault(inode->i_sb);
9071         page_start = page_offset(page);
9072         page_end = page_start + PAGE_SIZE - 1;
9073         end = page_end;
9074
9075         /*
9076          * Reserving delalloc space after obtaining the page lock can lead to
9077          * deadlock. For example, if a dirty page is locked by this function
9078          * and the call to btrfs_delalloc_reserve_space() ends up triggering
9079          * dirty page write out, then the btrfs_writepage() function could
9080          * end up waiting indefinitely to get a lock on the page currently
9081          * being processed by btrfs_page_mkwrite() function.
9082          */
9083         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
9084                                            reserved_space);
9085         if (!ret) {
9086                 ret = file_update_time(vmf->vma->vm_file);
9087                 reserved = 1;
9088         }
9089         if (ret) {
9090                 if (ret == -ENOMEM)
9091                         ret = VM_FAULT_OOM;
9092                 else /* -ENOSPC, -EIO, etc */
9093                         ret = VM_FAULT_SIGBUS;
9094                 if (reserved)
9095                         goto out;
9096                 goto out_noreserve;
9097         }
9098
9099         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9100 again:
9101         lock_page(page);
9102         size = i_size_read(inode);
9103
9104         if ((page->mapping != inode->i_mapping) ||
9105             (page_start >= size)) {
9106                 /* page got truncated out from underneath us */
9107                 goto out_unlock;
9108         }
9109         wait_on_page_writeback(page);
9110
9111         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9112         set_page_extent_mapped(page);
9113
9114         /*
9115          * we can't set the delalloc bits if there are pending ordered
9116          * extents.  Drop our locks and wait for them to finish
9117          */
9118         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9119                         PAGE_SIZE);
9120         if (ordered) {
9121                 unlock_extent_cached(io_tree, page_start, page_end,
9122                                      &cached_state, GFP_NOFS);
9123                 unlock_page(page);
9124                 btrfs_start_ordered_extent(inode, ordered, 1);
9125                 btrfs_put_ordered_extent(ordered);
9126                 goto again;
9127         }
9128
9129         if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9130                 reserved_space = round_up(size - page_start,
9131                                           fs_info->sectorsize);
9132                 if (reserved_space < PAGE_SIZE) {
9133                         end = page_start + reserved_space - 1;
9134                         spin_lock(&BTRFS_I(inode)->lock);
9135                         BTRFS_I(inode)->outstanding_extents++;
9136                         spin_unlock(&BTRFS_I(inode)->lock);
9137                         btrfs_delalloc_release_space(inode, data_reserved,
9138                                         page_start, PAGE_SIZE - reserved_space);
9139                 }
9140         }
9141
9142         /*
9143          * page_mkwrite gets called when the page is firstly dirtied after it's
9144          * faulted in, but write(2) could also dirty a page and set delalloc
9145          * bits, thus in this case for space account reason, we still need to
9146          * clear any delalloc bits within this page range since we have to
9147          * reserve data&meta space before lock_page() (see above comments).
9148          */
9149         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9150                           EXTENT_DIRTY | EXTENT_DELALLOC |
9151                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9152                           0, 0, &cached_state, GFP_NOFS);
9153
9154         ret = btrfs_set_extent_delalloc(inode, page_start, end,
9155                                         &cached_state, 0);
9156         if (ret) {
9157                 unlock_extent_cached(io_tree, page_start, page_end,
9158                                      &cached_state, GFP_NOFS);
9159                 ret = VM_FAULT_SIGBUS;
9160                 goto out_unlock;
9161         }
9162         ret = 0;
9163
9164         /* page is wholly or partially inside EOF */
9165         if (page_start + PAGE_SIZE > size)
9166                 zero_start = size & ~PAGE_MASK;
9167         else
9168                 zero_start = PAGE_SIZE;
9169
9170         if (zero_start != PAGE_SIZE) {
9171                 kaddr = kmap(page);
9172                 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9173                 flush_dcache_page(page);
9174                 kunmap(page);
9175         }
9176         ClearPageChecked(page);
9177         set_page_dirty(page);
9178         SetPageUptodate(page);
9179
9180         BTRFS_I(inode)->last_trans = fs_info->generation;
9181         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9182         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9183
9184         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9185
9186 out_unlock:
9187         if (!ret) {
9188                 sb_end_pagefault(inode->i_sb);
9189                 extent_changeset_free(data_reserved);
9190                 return VM_FAULT_LOCKED;
9191         }
9192         unlock_page(page);
9193 out:
9194         btrfs_delalloc_release_space(inode, data_reserved, page_start,
9195                                      reserved_space);
9196 out_noreserve:
9197         sb_end_pagefault(inode->i_sb);
9198         extent_changeset_free(data_reserved);
9199         return ret;
9200 }
9201
9202 static int btrfs_truncate(struct inode *inode)
9203 {
9204         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9205         struct btrfs_root *root = BTRFS_I(inode)->root;
9206         struct btrfs_block_rsv *rsv;
9207         int ret = 0;
9208         int err = 0;
9209         struct btrfs_trans_handle *trans;
9210         u64 mask = fs_info->sectorsize - 1;
9211         u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9212
9213         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9214                                        (u64)-1);
9215         if (ret)
9216                 return ret;
9217
9218         /*
9219          * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9220          * 3 things going on here
9221          *
9222          * 1) We need to reserve space for our orphan item and the space to
9223          * delete our orphan item.  Lord knows we don't want to have a dangling
9224          * orphan item because we didn't reserve space to remove it.
9225          *
9226          * 2) We need to reserve space to update our inode.
9227          *
9228          * 3) We need to have something to cache all the space that is going to
9229          * be free'd up by the truncate operation, but also have some slack
9230          * space reserved in case it uses space during the truncate (thank you
9231          * very much snapshotting).
9232          *
9233          * And we need these to all be separate.  The fact is we can use a lot of
9234          * space doing the truncate, and we have no earthly idea how much space
9235          * we will use, so we need the truncate reservation to be separate so it
9236          * doesn't end up using space reserved for updating the inode or
9237          * removing the orphan item.  We also need to be able to stop the
9238          * transaction and start a new one, which means we need to be able to
9239          * update the inode several times, and we have no idea of knowing how
9240          * many times that will be, so we can't just reserve 1 item for the
9241          * entirety of the operation, so that has to be done separately as well.
9242          * Then there is the orphan item, which does indeed need to be held on
9243          * to for the whole operation, and we need nobody to touch this reserved
9244          * space except the orphan code.
9245          *
9246          * So that leaves us with
9247          *
9248          * 1) root->orphan_block_rsv - for the orphan deletion.
9249          * 2) rsv - for the truncate reservation, which we will steal from the
9250          * transaction reservation.
9251          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9252          * updating the inode.
9253          */
9254         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9255         if (!rsv)
9256                 return -ENOMEM;
9257         rsv->size = min_size;
9258         rsv->failfast = 1;
9259
9260         /*
9261          * 1 for the truncate slack space
9262          * 1 for updating the inode.
9263          */
9264         trans = btrfs_start_transaction(root, 2);
9265         if (IS_ERR(trans)) {
9266                 err = PTR_ERR(trans);
9267                 goto out;
9268         }
9269
9270         /* Migrate the slack space for the truncate to our reserve */
9271         ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9272                                       min_size, 0);
9273         BUG_ON(ret);
9274
9275         /*
9276          * So if we truncate and then write and fsync we normally would just
9277          * write the extents that changed, which is a problem if we need to
9278          * first truncate that entire inode.  So set this flag so we write out
9279          * all of the extents in the inode to the sync log so we're completely
9280          * safe.
9281          */
9282         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9283         trans->block_rsv = rsv;
9284
9285         while (1) {
9286                 ret = btrfs_truncate_inode_items(trans, root, inode,
9287                                                  inode->i_size,
9288                                                  BTRFS_EXTENT_DATA_KEY);
9289                 if (ret != -ENOSPC && ret != -EAGAIN) {
9290                         err = ret;
9291                         break;
9292                 }
9293
9294                 trans->block_rsv = &fs_info->trans_block_rsv;
9295                 ret = btrfs_update_inode(trans, root, inode);
9296                 if (ret) {
9297                         err = ret;
9298                         break;
9299                 }
9300
9301                 btrfs_end_transaction(trans);
9302                 btrfs_btree_balance_dirty(fs_info);
9303
9304                 trans = btrfs_start_transaction(root, 2);
9305                 if (IS_ERR(trans)) {
9306                         ret = err = PTR_ERR(trans);
9307                         trans = NULL;
9308                         break;
9309                 }
9310
9311                 btrfs_block_rsv_release(fs_info, rsv, -1);
9312                 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9313                                               rsv, min_size, 0);
9314                 BUG_ON(ret);    /* shouldn't happen */
9315                 trans->block_rsv = rsv;
9316         }
9317
9318         if (ret == 0 && inode->i_nlink > 0) {
9319                 trans->block_rsv = root->orphan_block_rsv;
9320                 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9321                 if (ret)
9322                         err = ret;
9323         }
9324
9325         if (trans) {
9326                 trans->block_rsv = &fs_info->trans_block_rsv;
9327                 ret = btrfs_update_inode(trans, root, inode);
9328                 if (ret && !err)
9329                         err = ret;
9330
9331                 ret = btrfs_end_transaction(trans);
9332                 btrfs_btree_balance_dirty(fs_info);
9333         }
9334 out:
9335         btrfs_free_block_rsv(fs_info, rsv);
9336
9337         if (ret && !err)
9338                 err = ret;
9339
9340         return err;
9341 }
9342
9343 /*
9344  * create a new subvolume directory/inode (helper for the ioctl).
9345  */
9346 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9347                              struct btrfs_root *new_root,
9348                              struct btrfs_root *parent_root,
9349                              u64 new_dirid)
9350 {
9351         struct inode *inode;
9352         int err;
9353         u64 index = 0;
9354
9355         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9356                                 new_dirid, new_dirid,
9357                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
9358                                 &index);
9359         if (IS_ERR(inode))
9360                 return PTR_ERR(inode);
9361         inode->i_op = &btrfs_dir_inode_operations;
9362         inode->i_fop = &btrfs_dir_file_operations;
9363
9364         set_nlink(inode, 1);
9365         btrfs_i_size_write(BTRFS_I(inode), 0);
9366         unlock_new_inode(inode);
9367
9368         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9369         if (err)
9370                 btrfs_err(new_root->fs_info,
9371                           "error inheriting subvolume %llu properties: %d",
9372                           new_root->root_key.objectid, err);
9373
9374         err = btrfs_update_inode(trans, new_root, inode);
9375
9376         iput(inode);
9377         return err;
9378 }
9379
9380 struct inode *btrfs_alloc_inode(struct super_block *sb)
9381 {
9382         struct btrfs_inode *ei;
9383         struct inode *inode;
9384
9385         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9386         if (!ei)
9387                 return NULL;
9388
9389         ei->root = NULL;
9390         ei->generation = 0;
9391         ei->last_trans = 0;
9392         ei->last_sub_trans = 0;
9393         ei->logged_trans = 0;
9394         ei->delalloc_bytes = 0;
9395         ei->new_delalloc_bytes = 0;
9396         ei->defrag_bytes = 0;
9397         ei->disk_i_size = 0;
9398         ei->flags = 0;
9399         ei->csum_bytes = 0;
9400         ei->index_cnt = (u64)-1;
9401         ei->dir_index = 0;
9402         ei->last_unlink_trans = 0;
9403         ei->last_log_commit = 0;
9404         ei->delayed_iput_count = 0;
9405
9406         spin_lock_init(&ei->lock);
9407         ei->outstanding_extents = 0;
9408         ei->reserved_extents = 0;
9409
9410         ei->runtime_flags = 0;
9411         ei->force_compress = BTRFS_COMPRESS_NONE;
9412
9413         ei->delayed_node = NULL;
9414
9415         ei->i_otime.tv_sec = 0;
9416         ei->i_otime.tv_nsec = 0;
9417
9418         inode = &ei->vfs_inode;
9419         extent_map_tree_init(&ei->extent_tree);
9420         extent_io_tree_init(&ei->io_tree, inode);
9421         extent_io_tree_init(&ei->io_failure_tree, inode);
9422         ei->io_tree.track_uptodate = 1;
9423         ei->io_failure_tree.track_uptodate = 1;
9424         atomic_set(&ei->sync_writers, 0);
9425         mutex_init(&ei->log_mutex);
9426         mutex_init(&ei->delalloc_mutex);
9427         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9428         INIT_LIST_HEAD(&ei->delalloc_inodes);
9429         INIT_LIST_HEAD(&ei->delayed_iput);
9430         RB_CLEAR_NODE(&ei->rb_node);
9431         init_rwsem(&ei->dio_sem);
9432
9433         return inode;
9434 }
9435
9436 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9437 void btrfs_test_destroy_inode(struct inode *inode)
9438 {
9439         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9440         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9441 }
9442 #endif
9443
9444 static void btrfs_i_callback(struct rcu_head *head)
9445 {
9446         struct inode *inode = container_of(head, struct inode, i_rcu);
9447         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9448 }
9449
9450 void btrfs_destroy_inode(struct inode *inode)
9451 {
9452         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9453         struct btrfs_ordered_extent *ordered;
9454         struct btrfs_root *root = BTRFS_I(inode)->root;
9455
9456         WARN_ON(!hlist_empty(&inode->i_dentry));
9457         WARN_ON(inode->i_data.nrpages);
9458         WARN_ON(BTRFS_I(inode)->outstanding_extents);
9459         WARN_ON(BTRFS_I(inode)->reserved_extents);
9460         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9461         WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9462         WARN_ON(BTRFS_I(inode)->csum_bytes);
9463         WARN_ON(BTRFS_I(inode)->defrag_bytes);
9464
9465         /*
9466          * This can happen where we create an inode, but somebody else also
9467          * created the same inode and we need to destroy the one we already
9468          * created.
9469          */
9470         if (!root)
9471                 goto free;
9472
9473         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9474                      &BTRFS_I(inode)->runtime_flags)) {
9475                 btrfs_info(fs_info, "inode %llu still on the orphan list",
9476                            btrfs_ino(BTRFS_I(inode)));
9477                 atomic_dec(&root->orphan_inodes);
9478         }
9479
9480         while (1) {
9481                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9482                 if (!ordered)
9483                         break;
9484                 else {
9485                         btrfs_err(fs_info,
9486                                   "found ordered extent %llu %llu on inode cleanup",
9487                                   ordered->file_offset, ordered->len);
9488                         btrfs_remove_ordered_extent(inode, ordered);
9489                         btrfs_put_ordered_extent(ordered);
9490                         btrfs_put_ordered_extent(ordered);
9491                 }
9492         }
9493         btrfs_qgroup_check_reserved_leak(inode);
9494         inode_tree_del(inode);
9495         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9496 free:
9497         call_rcu(&inode->i_rcu, btrfs_i_callback);
9498 }
9499
9500 int btrfs_drop_inode(struct inode *inode)
9501 {
9502         struct btrfs_root *root = BTRFS_I(inode)->root;
9503
9504         if (root == NULL)
9505                 return 1;
9506
9507         /* the snap/subvol tree is on deleting */
9508         if (btrfs_root_refs(&root->root_item) == 0)
9509                 return 1;
9510         else
9511                 return generic_drop_inode(inode);
9512 }
9513
9514 static void init_once(void *foo)
9515 {
9516         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9517
9518         inode_init_once(&ei->vfs_inode);
9519 }
9520
9521 void btrfs_destroy_cachep(void)
9522 {
9523         /*
9524          * Make sure all delayed rcu free inodes are flushed before we
9525          * destroy cache.
9526          */
9527         rcu_barrier();
9528         kmem_cache_destroy(btrfs_inode_cachep);
9529         kmem_cache_destroy(btrfs_trans_handle_cachep);
9530         kmem_cache_destroy(btrfs_path_cachep);
9531         kmem_cache_destroy(btrfs_free_space_cachep);
9532 }
9533
9534 int btrfs_init_cachep(void)
9535 {
9536         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9537                         sizeof(struct btrfs_inode), 0,
9538                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9539                         init_once);
9540         if (!btrfs_inode_cachep)
9541                 goto fail;
9542
9543         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9544                         sizeof(struct btrfs_trans_handle), 0,
9545                         SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9546         if (!btrfs_trans_handle_cachep)
9547                 goto fail;
9548
9549         btrfs_path_cachep = kmem_cache_create("btrfs_path",
9550                         sizeof(struct btrfs_path), 0,
9551                         SLAB_MEM_SPREAD, NULL);
9552         if (!btrfs_path_cachep)
9553                 goto fail;
9554
9555         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9556                         sizeof(struct btrfs_free_space), 0,
9557                         SLAB_MEM_SPREAD, NULL);
9558         if (!btrfs_free_space_cachep)
9559                 goto fail;
9560
9561         return 0;
9562 fail:
9563         btrfs_destroy_cachep();
9564         return -ENOMEM;
9565 }
9566
9567 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9568                          u32 request_mask, unsigned int flags)
9569 {
9570         u64 delalloc_bytes;
9571         struct inode *inode = d_inode(path->dentry);
9572         u32 blocksize = inode->i_sb->s_blocksize;
9573         u32 bi_flags = BTRFS_I(inode)->flags;
9574
9575         stat->result_mask |= STATX_BTIME;
9576         stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9577         stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9578         if (bi_flags & BTRFS_INODE_APPEND)
9579                 stat->attributes |= STATX_ATTR_APPEND;
9580         if (bi_flags & BTRFS_INODE_COMPRESS)
9581                 stat->attributes |= STATX_ATTR_COMPRESSED;
9582         if (bi_flags & BTRFS_INODE_IMMUTABLE)
9583                 stat->attributes |= STATX_ATTR_IMMUTABLE;
9584         if (bi_flags & BTRFS_INODE_NODUMP)
9585                 stat->attributes |= STATX_ATTR_NODUMP;
9586
9587         stat->attributes_mask |= (STATX_ATTR_APPEND |
9588                                   STATX_ATTR_COMPRESSED |
9589                                   STATX_ATTR_IMMUTABLE |
9590                                   STATX_ATTR_NODUMP);
9591
9592         generic_fillattr(inode, stat);
9593         stat->dev = BTRFS_I(inode)->root->anon_dev;
9594
9595         spin_lock(&BTRFS_I(inode)->lock);
9596         delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9597         spin_unlock(&BTRFS_I(inode)->lock);
9598         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9599                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9600         return 0;
9601 }
9602
9603 static int btrfs_rename_exchange(struct inode *old_dir,
9604                               struct dentry *old_dentry,
9605                               struct inode *new_dir,
9606                               struct dentry *new_dentry)
9607 {
9608         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9609         struct btrfs_trans_handle *trans;
9610         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9611         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9612         struct inode *new_inode = new_dentry->d_inode;
9613         struct inode *old_inode = old_dentry->d_inode;
9614         struct timespec ctime = current_time(old_inode);
9615         struct dentry *parent;
9616         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9617         u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9618         u64 old_idx = 0;
9619         u64 new_idx = 0;
9620         u64 root_objectid;
9621         int ret;
9622         bool root_log_pinned = false;
9623         bool dest_log_pinned = false;
9624
9625         /* we only allow rename subvolume link between subvolumes */
9626         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9627                 return -EXDEV;
9628
9629         /* close the race window with snapshot create/destroy ioctl */
9630         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9631                 down_read(&fs_info->subvol_sem);
9632         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9633                 down_read(&fs_info->subvol_sem);
9634
9635         /*
9636          * We want to reserve the absolute worst case amount of items.  So if
9637          * both inodes are subvols and we need to unlink them then that would
9638          * require 4 item modifications, but if they are both normal inodes it
9639          * would require 5 item modifications, so we'll assume their normal
9640          * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9641          * should cover the worst case number of items we'll modify.
9642          */
9643         trans = btrfs_start_transaction(root, 12);
9644         if (IS_ERR(trans)) {
9645                 ret = PTR_ERR(trans);
9646                 goto out_notrans;
9647         }
9648
9649         /*
9650          * We need to find a free sequence number both in the source and
9651          * in the destination directory for the exchange.
9652          */
9653         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9654         if (ret)
9655                 goto out_fail;
9656         ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9657         if (ret)
9658                 goto out_fail;
9659
9660         BTRFS_I(old_inode)->dir_index = 0ULL;
9661         BTRFS_I(new_inode)->dir_index = 0ULL;
9662
9663         /* Reference for the source. */
9664         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9665                 /* force full log commit if subvolume involved. */
9666                 btrfs_set_log_full_commit(fs_info, trans);
9667         } else {
9668                 btrfs_pin_log_trans(root);
9669                 root_log_pinned = true;
9670                 ret = btrfs_insert_inode_ref(trans, dest,
9671                                              new_dentry->d_name.name,
9672                                              new_dentry->d_name.len,
9673                                              old_ino,
9674                                              btrfs_ino(BTRFS_I(new_dir)),
9675                                              old_idx);
9676                 if (ret)
9677                         goto out_fail;
9678         }
9679
9680         /* And now for the dest. */
9681         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9682                 /* force full log commit if subvolume involved. */
9683                 btrfs_set_log_full_commit(fs_info, trans);
9684         } else {
9685                 btrfs_pin_log_trans(dest);
9686                 dest_log_pinned = true;
9687                 ret = btrfs_insert_inode_ref(trans, root,
9688                                              old_dentry->d_name.name,
9689                                              old_dentry->d_name.len,
9690                                              new_ino,
9691                                              btrfs_ino(BTRFS_I(old_dir)),
9692                                              new_idx);
9693                 if (ret)
9694                         goto out_fail;
9695         }
9696
9697         /* Update inode version and ctime/mtime. */
9698         inode_inc_iversion(old_dir);
9699         inode_inc_iversion(new_dir);
9700         inode_inc_iversion(old_inode);
9701         inode_inc_iversion(new_inode);
9702         old_dir->i_ctime = old_dir->i_mtime = ctime;
9703         new_dir->i_ctime = new_dir->i_mtime = ctime;
9704         old_inode->i_ctime = ctime;
9705         new_inode->i_ctime = ctime;
9706
9707         if (old_dentry->d_parent != new_dentry->d_parent) {
9708                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9709                                 BTRFS_I(old_inode), 1);
9710                 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9711                                 BTRFS_I(new_inode), 1);
9712         }
9713
9714         /* src is a subvolume */
9715         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9716                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9717                 ret = btrfs_unlink_subvol(trans, root, old_dir,
9718                                           root_objectid,
9719                                           old_dentry->d_name.name,
9720                                           old_dentry->d_name.len);
9721         } else { /* src is an inode */
9722                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9723                                            BTRFS_I(old_dentry->d_inode),
9724                                            old_dentry->d_name.name,
9725                                            old_dentry->d_name.len);
9726                 if (!ret)
9727                         ret = btrfs_update_inode(trans, root, old_inode);
9728         }
9729         if (ret) {
9730                 btrfs_abort_transaction(trans, ret);
9731                 goto out_fail;
9732         }
9733
9734         /* dest is a subvolume */
9735         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9736                 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9737                 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9738                                           root_objectid,
9739                                           new_dentry->d_name.name,
9740                                           new_dentry->d_name.len);
9741         } else { /* dest is an inode */
9742                 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9743                                            BTRFS_I(new_dentry->d_inode),
9744                                            new_dentry->d_name.name,
9745                                            new_dentry->d_name.len);
9746                 if (!ret)
9747                         ret = btrfs_update_inode(trans, dest, new_inode);
9748         }
9749         if (ret) {
9750                 btrfs_abort_transaction(trans, ret);
9751                 goto out_fail;
9752         }
9753
9754         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9755                              new_dentry->d_name.name,
9756                              new_dentry->d_name.len, 0, old_idx);
9757         if (ret) {
9758                 btrfs_abort_transaction(trans, ret);
9759                 goto out_fail;
9760         }
9761
9762         ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9763                              old_dentry->d_name.name,
9764                              old_dentry->d_name.len, 0, new_idx);
9765         if (ret) {
9766                 btrfs_abort_transaction(trans, ret);
9767                 goto out_fail;
9768         }
9769
9770         if (old_inode->i_nlink == 1)
9771                 BTRFS_I(old_inode)->dir_index = old_idx;
9772         if (new_inode->i_nlink == 1)
9773                 BTRFS_I(new_inode)->dir_index = new_idx;
9774
9775         if (root_log_pinned) {
9776                 parent = new_dentry->d_parent;
9777                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9778                                 parent);
9779                 btrfs_end_log_trans(root);
9780                 root_log_pinned = false;
9781         }
9782         if (dest_log_pinned) {
9783                 parent = old_dentry->d_parent;
9784                 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9785                                 parent);
9786                 btrfs_end_log_trans(dest);
9787                 dest_log_pinned = false;
9788         }
9789 out_fail:
9790         /*
9791          * If we have pinned a log and an error happened, we unpin tasks
9792          * trying to sync the log and force them to fallback to a transaction
9793          * commit if the log currently contains any of the inodes involved in
9794          * this rename operation (to ensure we do not persist a log with an
9795          * inconsistent state for any of these inodes or leading to any
9796          * inconsistencies when replayed). If the transaction was aborted, the
9797          * abortion reason is propagated to userspace when attempting to commit
9798          * the transaction. If the log does not contain any of these inodes, we
9799          * allow the tasks to sync it.
9800          */
9801         if (ret && (root_log_pinned || dest_log_pinned)) {
9802                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9803                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9804                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9805                     (new_inode &&
9806                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9807                         btrfs_set_log_full_commit(fs_info, trans);
9808
9809                 if (root_log_pinned) {
9810                         btrfs_end_log_trans(root);
9811                         root_log_pinned = false;
9812                 }
9813                 if (dest_log_pinned) {
9814                         btrfs_end_log_trans(dest);
9815                         dest_log_pinned = false;
9816                 }
9817         }
9818         ret = btrfs_end_transaction(trans);
9819 out_notrans:
9820         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9821                 up_read(&fs_info->subvol_sem);
9822         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9823                 up_read(&fs_info->subvol_sem);
9824
9825         return ret;
9826 }
9827
9828 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9829                                      struct btrfs_root *root,
9830                                      struct inode *dir,
9831                                      struct dentry *dentry)
9832 {
9833         int ret;
9834         struct inode *inode;
9835         u64 objectid;
9836         u64 index;
9837
9838         ret = btrfs_find_free_ino(root, &objectid);
9839         if (ret)
9840                 return ret;
9841
9842         inode = btrfs_new_inode(trans, root, dir,
9843                                 dentry->d_name.name,
9844                                 dentry->d_name.len,
9845                                 btrfs_ino(BTRFS_I(dir)),
9846                                 objectid,
9847                                 S_IFCHR | WHITEOUT_MODE,
9848                                 &index);
9849
9850         if (IS_ERR(inode)) {
9851                 ret = PTR_ERR(inode);
9852                 return ret;
9853         }
9854
9855         inode->i_op = &btrfs_special_inode_operations;
9856         init_special_inode(inode, inode->i_mode,
9857                 WHITEOUT_DEV);
9858
9859         ret = btrfs_init_inode_security(trans, inode, dir,
9860                                 &dentry->d_name);
9861         if (ret)
9862                 goto out;
9863
9864         ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9865                                 BTRFS_I(inode), 0, index);
9866         if (ret)
9867                 goto out;
9868
9869         ret = btrfs_update_inode(trans, root, inode);
9870 out:
9871         unlock_new_inode(inode);
9872         if (ret)
9873                 inode_dec_link_count(inode);
9874         iput(inode);
9875
9876         return ret;
9877 }
9878
9879 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9880                            struct inode *new_dir, struct dentry *new_dentry,
9881                            unsigned int flags)
9882 {
9883         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9884         struct btrfs_trans_handle *trans;
9885         unsigned int trans_num_items;
9886         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9887         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9888         struct inode *new_inode = d_inode(new_dentry);
9889         struct inode *old_inode = d_inode(old_dentry);
9890         u64 index = 0;
9891         u64 root_objectid;
9892         int ret;
9893         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9894         bool log_pinned = false;
9895
9896         if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9897                 return -EPERM;
9898
9899         /* we only allow rename subvolume link between subvolumes */
9900         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9901                 return -EXDEV;
9902
9903         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9904             (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9905                 return -ENOTEMPTY;
9906
9907         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9908             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9909                 return -ENOTEMPTY;
9910
9911
9912         /* check for collisions, even if the  name isn't there */
9913         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9914                              new_dentry->d_name.name,
9915                              new_dentry->d_name.len);
9916
9917         if (ret) {
9918                 if (ret == -EEXIST) {
9919                         /* we shouldn't get
9920                          * eexist without a new_inode */
9921                         if (WARN_ON(!new_inode)) {
9922                                 return ret;
9923                         }
9924                 } else {
9925                         /* maybe -EOVERFLOW */
9926                         return ret;
9927                 }
9928         }
9929         ret = 0;
9930
9931         /*
9932          * we're using rename to replace one file with another.  Start IO on it
9933          * now so  we don't add too much work to the end of the transaction
9934          */
9935         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9936                 filemap_flush(old_inode->i_mapping);
9937
9938         /* close the racy window with snapshot create/destroy ioctl */
9939         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9940                 down_read(&fs_info->subvol_sem);
9941         /*
9942          * We want to reserve the absolute worst case amount of items.  So if
9943          * both inodes are subvols and we need to unlink them then that would
9944          * require 4 item modifications, but if they are both normal inodes it
9945          * would require 5 item modifications, so we'll assume they are normal
9946          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9947          * should cover the worst case number of items we'll modify.
9948          * If our rename has the whiteout flag, we need more 5 units for the
9949          * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9950          * when selinux is enabled).
9951          */
9952         trans_num_items = 11;
9953         if (flags & RENAME_WHITEOUT)
9954                 trans_num_items += 5;
9955         trans = btrfs_start_transaction(root, trans_num_items);
9956         if (IS_ERR(trans)) {
9957                 ret = PTR_ERR(trans);
9958                 goto out_notrans;
9959         }
9960
9961         if (dest != root)
9962                 btrfs_record_root_in_trans(trans, dest);
9963
9964         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9965         if (ret)
9966                 goto out_fail;
9967
9968         BTRFS_I(old_inode)->dir_index = 0ULL;
9969         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9970                 /* force full log commit if subvolume involved. */
9971                 btrfs_set_log_full_commit(fs_info, trans);
9972         } else {
9973                 btrfs_pin_log_trans(root);
9974                 log_pinned = true;
9975                 ret = btrfs_insert_inode_ref(trans, dest,
9976                                              new_dentry->d_name.name,
9977                                              new_dentry->d_name.len,
9978                                              old_ino,
9979                                              btrfs_ino(BTRFS_I(new_dir)), index);
9980                 if (ret)
9981                         goto out_fail;
9982         }
9983
9984         inode_inc_iversion(old_dir);
9985         inode_inc_iversion(new_dir);
9986         inode_inc_iversion(old_inode);
9987         old_dir->i_ctime = old_dir->i_mtime =
9988         new_dir->i_ctime = new_dir->i_mtime =
9989         old_inode->i_ctime = current_time(old_dir);
9990
9991         if (old_dentry->d_parent != new_dentry->d_parent)
9992                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9993                                 BTRFS_I(old_inode), 1);
9994
9995         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9996                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9997                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9998                                         old_dentry->d_name.name,
9999                                         old_dentry->d_name.len);
10000         } else {
10001                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
10002                                         BTRFS_I(d_inode(old_dentry)),
10003                                         old_dentry->d_name.name,
10004                                         old_dentry->d_name.len);
10005                 if (!ret)
10006                         ret = btrfs_update_inode(trans, root, old_inode);
10007         }
10008         if (ret) {
10009                 btrfs_abort_transaction(trans, ret);
10010                 goto out_fail;
10011         }
10012
10013         if (new_inode) {
10014                 inode_inc_iversion(new_inode);
10015                 new_inode->i_ctime = current_time(new_inode);
10016                 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
10017                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
10018                         root_objectid = BTRFS_I(new_inode)->location.objectid;
10019                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
10020                                                 root_objectid,
10021                                                 new_dentry->d_name.name,
10022                                                 new_dentry->d_name.len);
10023                         BUG_ON(new_inode->i_nlink == 0);
10024                 } else {
10025                         ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10026                                                  BTRFS_I(d_inode(new_dentry)),
10027                                                  new_dentry->d_name.name,
10028                                                  new_dentry->d_name.len);
10029                 }
10030                 if (!ret && new_inode->i_nlink == 0)
10031                         ret = btrfs_orphan_add(trans,
10032                                         BTRFS_I(d_inode(new_dentry)));
10033                 if (ret) {
10034                         btrfs_abort_transaction(trans, ret);
10035                         goto out_fail;
10036                 }
10037         }
10038
10039         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10040                              new_dentry->d_name.name,
10041                              new_dentry->d_name.len, 0, index);
10042         if (ret) {
10043                 btrfs_abort_transaction(trans, ret);
10044                 goto out_fail;
10045         }
10046
10047         if (old_inode->i_nlink == 1)
10048                 BTRFS_I(old_inode)->dir_index = index;
10049
10050         if (log_pinned) {
10051                 struct dentry *parent = new_dentry->d_parent;
10052
10053                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10054                                 parent);
10055                 btrfs_end_log_trans(root);
10056                 log_pinned = false;
10057         }
10058
10059         if (flags & RENAME_WHITEOUT) {
10060                 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10061                                                 old_dentry);
10062
10063                 if (ret) {
10064                         btrfs_abort_transaction(trans, ret);
10065                         goto out_fail;
10066                 }
10067         }
10068 out_fail:
10069         /*
10070          * If we have pinned the log and an error happened, we unpin tasks
10071          * trying to sync the log and force them to fallback to a transaction
10072          * commit if the log currently contains any of the inodes involved in
10073          * this rename operation (to ensure we do not persist a log with an
10074          * inconsistent state for any of these inodes or leading to any
10075          * inconsistencies when replayed). If the transaction was aborted, the
10076          * abortion reason is propagated to userspace when attempting to commit
10077          * the transaction. If the log does not contain any of these inodes, we
10078          * allow the tasks to sync it.
10079          */
10080         if (ret && log_pinned) {
10081                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10082                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10083                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10084                     (new_inode &&
10085                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10086                         btrfs_set_log_full_commit(fs_info, trans);
10087
10088                 btrfs_end_log_trans(root);
10089                 log_pinned = false;
10090         }
10091         btrfs_end_transaction(trans);
10092 out_notrans:
10093         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10094                 up_read(&fs_info->subvol_sem);
10095
10096         return ret;
10097 }
10098
10099 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10100                          struct inode *new_dir, struct dentry *new_dentry,
10101                          unsigned int flags)
10102 {
10103         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10104                 return -EINVAL;
10105
10106         if (flags & RENAME_EXCHANGE)
10107                 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10108                                           new_dentry);
10109
10110         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10111 }
10112
10113 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10114 {
10115         struct btrfs_delalloc_work *delalloc_work;
10116         struct inode *inode;
10117
10118         delalloc_work = container_of(work, struct btrfs_delalloc_work,
10119                                      work);
10120         inode = delalloc_work->inode;
10121         filemap_flush(inode->i_mapping);
10122         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10123                                 &BTRFS_I(inode)->runtime_flags))
10124                 filemap_flush(inode->i_mapping);
10125
10126         if (delalloc_work->delay_iput)
10127                 btrfs_add_delayed_iput(inode);
10128         else
10129                 iput(inode);
10130         complete(&delalloc_work->completion);
10131 }
10132
10133 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10134                                                     int delay_iput)
10135 {
10136         struct btrfs_delalloc_work *work;
10137
10138         work = kmalloc(sizeof(*work), GFP_NOFS);
10139         if (!work)
10140                 return NULL;
10141
10142         init_completion(&work->completion);
10143         INIT_LIST_HEAD(&work->list);
10144         work->inode = inode;
10145         work->delay_iput = delay_iput;
10146         WARN_ON_ONCE(!inode);
10147         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10148                         btrfs_run_delalloc_work, NULL, NULL);
10149
10150         return work;
10151 }
10152
10153 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10154 {
10155         wait_for_completion(&work->completion);
10156         kfree(work);
10157 }
10158
10159 /*
10160  * some fairly slow code that needs optimization. This walks the list
10161  * of all the inodes with pending delalloc and forces them to disk.
10162  */
10163 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10164                                    int nr)
10165 {
10166         struct btrfs_inode *binode;
10167         struct inode *inode;
10168         struct btrfs_delalloc_work *work, *next;
10169         struct list_head works;
10170         struct list_head splice;
10171         int ret = 0;
10172
10173         INIT_LIST_HEAD(&works);
10174         INIT_LIST_HEAD(&splice);
10175
10176         mutex_lock(&root->delalloc_mutex);
10177         spin_lock(&root->delalloc_lock);
10178         list_splice_init(&root->delalloc_inodes, &splice);
10179         while (!list_empty(&splice)) {
10180                 binode = list_entry(splice.next, struct btrfs_inode,
10181                                     delalloc_inodes);
10182
10183                 list_move_tail(&binode->delalloc_inodes,
10184                                &root->delalloc_inodes);
10185                 inode = igrab(&binode->vfs_inode);
10186                 if (!inode) {
10187                         cond_resched_lock(&root->delalloc_lock);
10188                         continue;
10189                 }
10190                 spin_unlock(&root->delalloc_lock);
10191
10192                 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10193                 if (!work) {
10194                         if (delay_iput)
10195                                 btrfs_add_delayed_iput(inode);
10196                         else
10197                                 iput(inode);
10198                         ret = -ENOMEM;
10199                         goto out;
10200                 }
10201                 list_add_tail(&work->list, &works);
10202                 btrfs_queue_work(root->fs_info->flush_workers,
10203                                  &work->work);
10204                 ret++;
10205                 if (nr != -1 && ret >= nr)
10206                         goto out;
10207                 cond_resched();
10208                 spin_lock(&root->delalloc_lock);
10209         }
10210         spin_unlock(&root->delalloc_lock);
10211
10212 out:
10213         list_for_each_entry_safe(work, next, &works, list) {
10214                 list_del_init(&work->list);
10215                 btrfs_wait_and_free_delalloc_work(work);
10216         }
10217
10218         if (!list_empty_careful(&splice)) {
10219                 spin_lock(&root->delalloc_lock);
10220                 list_splice_tail(&splice, &root->delalloc_inodes);
10221                 spin_unlock(&root->delalloc_lock);
10222         }
10223         mutex_unlock(&root->delalloc_mutex);
10224         return ret;
10225 }
10226
10227 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10228 {
10229         struct btrfs_fs_info *fs_info = root->fs_info;
10230         int ret;
10231
10232         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10233                 return -EROFS;
10234
10235         ret = __start_delalloc_inodes(root, delay_iput, -1);
10236         if (ret > 0)
10237                 ret = 0;
10238         /*
10239          * the filemap_flush will queue IO into the worker threads, but
10240          * we have to make sure the IO is actually started and that
10241          * ordered extents get created before we return
10242          */
10243         atomic_inc(&fs_info->async_submit_draining);
10244         while (atomic_read(&fs_info->nr_async_submits) ||
10245                atomic_read(&fs_info->async_delalloc_pages)) {
10246                 wait_event(fs_info->async_submit_wait,
10247                            (atomic_read(&fs_info->nr_async_submits) == 0 &&
10248                             atomic_read(&fs_info->async_delalloc_pages) == 0));
10249         }
10250         atomic_dec(&fs_info->async_submit_draining);
10251         return ret;
10252 }
10253
10254 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10255                                int nr)
10256 {
10257         struct btrfs_root *root;
10258         struct list_head splice;
10259         int ret;
10260
10261         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10262                 return -EROFS;
10263
10264         INIT_LIST_HEAD(&splice);
10265
10266         mutex_lock(&fs_info->delalloc_root_mutex);
10267         spin_lock(&fs_info->delalloc_root_lock);
10268         list_splice_init(&fs_info->delalloc_roots, &splice);
10269         while (!list_empty(&splice) && nr) {
10270                 root = list_first_entry(&splice, struct btrfs_root,
10271                                         delalloc_root);
10272                 root = btrfs_grab_fs_root(root);
10273                 BUG_ON(!root);
10274                 list_move_tail(&root->delalloc_root,
10275                                &fs_info->delalloc_roots);
10276                 spin_unlock(&fs_info->delalloc_root_lock);
10277
10278                 ret = __start_delalloc_inodes(root, delay_iput, nr);
10279                 btrfs_put_fs_root(root);
10280                 if (ret < 0)
10281                         goto out;
10282
10283                 if (nr != -1) {
10284                         nr -= ret;
10285                         WARN_ON(nr < 0);
10286                 }
10287                 spin_lock(&fs_info->delalloc_root_lock);
10288         }
10289         spin_unlock(&fs_info->delalloc_root_lock);
10290
10291         ret = 0;
10292         atomic_inc(&fs_info->async_submit_draining);
10293         while (atomic_read(&fs_info->nr_async_submits) ||
10294               atomic_read(&fs_info->async_delalloc_pages)) {
10295                 wait_event(fs_info->async_submit_wait,
10296                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
10297                     atomic_read(&fs_info->async_delalloc_pages) == 0));
10298         }
10299         atomic_dec(&fs_info->async_submit_draining);
10300 out:
10301         if (!list_empty_careful(&splice)) {
10302                 spin_lock(&fs_info->delalloc_root_lock);
10303                 list_splice_tail(&splice, &fs_info->delalloc_roots);
10304                 spin_unlock(&fs_info->delalloc_root_lock);
10305         }
10306         mutex_unlock(&fs_info->delalloc_root_mutex);
10307         return ret;
10308 }
10309
10310 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10311                          const char *symname)
10312 {
10313         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10314         struct btrfs_trans_handle *trans;
10315         struct btrfs_root *root = BTRFS_I(dir)->root;
10316         struct btrfs_path *path;
10317         struct btrfs_key key;
10318         struct inode *inode = NULL;
10319         int err;
10320         int drop_inode = 0;
10321         u64 objectid;
10322         u64 index = 0;
10323         int name_len;
10324         int datasize;
10325         unsigned long ptr;
10326         struct btrfs_file_extent_item *ei;
10327         struct extent_buffer *leaf;
10328
10329         name_len = strlen(symname);
10330         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10331                 return -ENAMETOOLONG;
10332
10333         /*
10334          * 2 items for inode item and ref
10335          * 2 items for dir items
10336          * 1 item for updating parent inode item
10337          * 1 item for the inline extent item
10338          * 1 item for xattr if selinux is on
10339          */
10340         trans = btrfs_start_transaction(root, 7);
10341         if (IS_ERR(trans))
10342                 return PTR_ERR(trans);
10343
10344         err = btrfs_find_free_ino(root, &objectid);
10345         if (err)
10346                 goto out_unlock;
10347
10348         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10349                                 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10350                                 objectid, S_IFLNK|S_IRWXUGO, &index);
10351         if (IS_ERR(inode)) {
10352                 err = PTR_ERR(inode);
10353                 goto out_unlock;
10354         }
10355
10356         /*
10357         * If the active LSM wants to access the inode during
10358         * d_instantiate it needs these. Smack checks to see
10359         * if the filesystem supports xattrs by looking at the
10360         * ops vector.
10361         */
10362         inode->i_fop = &btrfs_file_operations;
10363         inode->i_op = &btrfs_file_inode_operations;
10364         inode->i_mapping->a_ops = &btrfs_aops;
10365         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10366
10367         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10368         if (err)
10369                 goto out_unlock_inode;
10370
10371         path = btrfs_alloc_path();
10372         if (!path) {
10373                 err = -ENOMEM;
10374                 goto out_unlock_inode;
10375         }
10376         key.objectid = btrfs_ino(BTRFS_I(inode));
10377         key.offset = 0;
10378         key.type = BTRFS_EXTENT_DATA_KEY;
10379         datasize = btrfs_file_extent_calc_inline_size(name_len);
10380         err = btrfs_insert_empty_item(trans, root, path, &key,
10381                                       datasize);
10382         if (err) {
10383                 btrfs_free_path(path);
10384                 goto out_unlock_inode;
10385         }
10386         leaf = path->nodes[0];
10387         ei = btrfs_item_ptr(leaf, path->slots[0],
10388                             struct btrfs_file_extent_item);
10389         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10390         btrfs_set_file_extent_type(leaf, ei,
10391                                    BTRFS_FILE_EXTENT_INLINE);
10392         btrfs_set_file_extent_encryption(leaf, ei, 0);
10393         btrfs_set_file_extent_compression(leaf, ei, 0);
10394         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10395         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10396
10397         ptr = btrfs_file_extent_inline_start(ei);
10398         write_extent_buffer(leaf, symname, ptr, name_len);
10399         btrfs_mark_buffer_dirty(leaf);
10400         btrfs_free_path(path);
10401
10402         inode->i_op = &btrfs_symlink_inode_operations;
10403         inode_nohighmem(inode);
10404         inode->i_mapping->a_ops = &btrfs_symlink_aops;
10405         inode_set_bytes(inode, name_len);
10406         btrfs_i_size_write(BTRFS_I(inode), name_len);
10407         err = btrfs_update_inode(trans, root, inode);
10408         /*
10409          * Last step, add directory indexes for our symlink inode. This is the
10410          * last step to avoid extra cleanup of these indexes if an error happens
10411          * elsewhere above.
10412          */
10413         if (!err)
10414                 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10415                                 BTRFS_I(inode), 0, index);
10416         if (err) {
10417                 drop_inode = 1;
10418                 goto out_unlock_inode;
10419         }
10420
10421         unlock_new_inode(inode);
10422         d_instantiate(dentry, inode);
10423
10424 out_unlock:
10425         btrfs_end_transaction(trans);
10426         if (drop_inode) {
10427                 inode_dec_link_count(inode);
10428                 iput(inode);
10429         }
10430         btrfs_btree_balance_dirty(fs_info);
10431         return err;
10432
10433 out_unlock_inode:
10434         drop_inode = 1;
10435         unlock_new_inode(inode);
10436         goto out_unlock;
10437 }
10438
10439 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10440                                        u64 start, u64 num_bytes, u64 min_size,
10441                                        loff_t actual_len, u64 *alloc_hint,
10442                                        struct btrfs_trans_handle *trans)
10443 {
10444         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10445         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10446         struct extent_map *em;
10447         struct btrfs_root *root = BTRFS_I(inode)->root;
10448         struct btrfs_key ins;
10449         u64 cur_offset = start;
10450         u64 i_size;
10451         u64 cur_bytes;
10452         u64 last_alloc = (u64)-1;
10453         int ret = 0;
10454         bool own_trans = true;
10455         u64 end = start + num_bytes - 1;
10456
10457         if (trans)
10458                 own_trans = false;
10459         while (num_bytes > 0) {
10460                 if (own_trans) {
10461                         trans = btrfs_start_transaction(root, 3);
10462                         if (IS_ERR(trans)) {
10463                                 ret = PTR_ERR(trans);
10464                                 break;
10465                         }
10466                 }
10467
10468                 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10469                 cur_bytes = max(cur_bytes, min_size);
10470                 /*
10471                  * If we are severely fragmented we could end up with really
10472                  * small allocations, so if the allocator is returning small
10473                  * chunks lets make its job easier by only searching for those
10474                  * sized chunks.
10475                  */
10476                 cur_bytes = min(cur_bytes, last_alloc);
10477                 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10478                                 min_size, 0, *alloc_hint, &ins, 1, 0);
10479                 if (ret) {
10480                         if (own_trans)
10481                                 btrfs_end_transaction(trans);
10482                         break;
10483                 }
10484                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10485
10486                 last_alloc = ins.offset;
10487                 ret = insert_reserved_file_extent(trans, inode,
10488                                                   cur_offset, ins.objectid,
10489                                                   ins.offset, ins.offset,
10490                                                   ins.offset, 0, 0, 0,
10491                                                   BTRFS_FILE_EXTENT_PREALLOC);
10492                 if (ret) {
10493                         btrfs_free_reserved_extent(fs_info, ins.objectid,
10494                                                    ins.offset, 0);
10495                         btrfs_abort_transaction(trans, ret);
10496                         if (own_trans)
10497                                 btrfs_end_transaction(trans);
10498                         break;
10499                 }
10500
10501                 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10502                                         cur_offset + ins.offset -1, 0);
10503
10504                 em = alloc_extent_map();
10505                 if (!em) {
10506                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10507                                 &BTRFS_I(inode)->runtime_flags);
10508                         goto next;
10509                 }
10510
10511                 em->start = cur_offset;
10512                 em->orig_start = cur_offset;
10513                 em->len = ins.offset;
10514                 em->block_start = ins.objectid;
10515                 em->block_len = ins.offset;
10516                 em->orig_block_len = ins.offset;
10517                 em->ram_bytes = ins.offset;
10518                 em->bdev = fs_info->fs_devices->latest_bdev;
10519                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10520                 em->generation = trans->transid;
10521
10522                 while (1) {
10523                         write_lock(&em_tree->lock);
10524                         ret = add_extent_mapping(em_tree, em, 1);
10525                         write_unlock(&em_tree->lock);
10526                         if (ret != -EEXIST)
10527                                 break;
10528                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10529                                                 cur_offset + ins.offset - 1,
10530                                                 0);
10531                 }
10532                 free_extent_map(em);
10533 next:
10534                 num_bytes -= ins.offset;
10535                 cur_offset += ins.offset;
10536                 *alloc_hint = ins.objectid + ins.offset;
10537
10538                 inode_inc_iversion(inode);
10539                 inode->i_ctime = current_time(inode);
10540                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10541                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10542                     (actual_len > inode->i_size) &&
10543                     (cur_offset > inode->i_size)) {
10544                         if (cur_offset > actual_len)
10545                                 i_size = actual_len;
10546                         else
10547                                 i_size = cur_offset;
10548                         i_size_write(inode, i_size);
10549                         btrfs_ordered_update_i_size(inode, i_size, NULL);
10550                 }
10551
10552                 ret = btrfs_update_inode(trans, root, inode);
10553
10554                 if (ret) {
10555                         btrfs_abort_transaction(trans, ret);
10556                         if (own_trans)
10557                                 btrfs_end_transaction(trans);
10558                         break;
10559                 }
10560
10561                 if (own_trans)
10562                         btrfs_end_transaction(trans);
10563         }
10564         if (cur_offset < end)
10565                 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10566                         end - cur_offset + 1);
10567         return ret;
10568 }
10569
10570 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10571                               u64 start, u64 num_bytes, u64 min_size,
10572                               loff_t actual_len, u64 *alloc_hint)
10573 {
10574         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10575                                            min_size, actual_len, alloc_hint,
10576                                            NULL);
10577 }
10578
10579 int btrfs_prealloc_file_range_trans(struct inode *inode,
10580                                     struct btrfs_trans_handle *trans, int mode,
10581                                     u64 start, u64 num_bytes, u64 min_size,
10582                                     loff_t actual_len, u64 *alloc_hint)
10583 {
10584         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10585                                            min_size, actual_len, alloc_hint, trans);
10586 }
10587
10588 static int btrfs_set_page_dirty(struct page *page)
10589 {
10590         return __set_page_dirty_nobuffers(page);
10591 }
10592
10593 static int btrfs_permission(struct inode *inode, int mask)
10594 {
10595         struct btrfs_root *root = BTRFS_I(inode)->root;
10596         umode_t mode = inode->i_mode;
10597
10598         if (mask & MAY_WRITE &&
10599             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10600                 if (btrfs_root_readonly(root))
10601                         return -EROFS;
10602                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10603                         return -EACCES;
10604         }
10605         return generic_permission(inode, mask);
10606 }
10607
10608 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10609 {
10610         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10611         struct btrfs_trans_handle *trans;
10612         struct btrfs_root *root = BTRFS_I(dir)->root;
10613         struct inode *inode = NULL;
10614         u64 objectid;
10615         u64 index;
10616         int ret = 0;
10617
10618         /*
10619          * 5 units required for adding orphan entry
10620          */
10621         trans = btrfs_start_transaction(root, 5);
10622         if (IS_ERR(trans))
10623                 return PTR_ERR(trans);
10624
10625         ret = btrfs_find_free_ino(root, &objectid);
10626         if (ret)
10627                 goto out;
10628
10629         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10630                         btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10631         if (IS_ERR(inode)) {
10632                 ret = PTR_ERR(inode);
10633                 inode = NULL;
10634                 goto out;
10635         }
10636
10637         inode->i_fop = &btrfs_file_operations;
10638         inode->i_op = &btrfs_file_inode_operations;
10639
10640         inode->i_mapping->a_ops = &btrfs_aops;
10641         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10642
10643         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10644         if (ret)
10645                 goto out_inode;
10646
10647         ret = btrfs_update_inode(trans, root, inode);
10648         if (ret)
10649                 goto out_inode;
10650         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10651         if (ret)
10652                 goto out_inode;
10653
10654         /*
10655          * We set number of links to 0 in btrfs_new_inode(), and here we set
10656          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10657          * through:
10658          *
10659          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10660          */
10661         set_nlink(inode, 1);
10662         unlock_new_inode(inode);
10663         d_tmpfile(dentry, inode);
10664         mark_inode_dirty(inode);
10665
10666 out:
10667         btrfs_end_transaction(trans);
10668         if (ret)
10669                 iput(inode);
10670         btrfs_balance_delayed_items(fs_info);
10671         btrfs_btree_balance_dirty(fs_info);
10672         return ret;
10673
10674 out_inode:
10675         unlock_new_inode(inode);
10676         goto out;
10677
10678 }
10679
10680 __attribute__((const))
10681 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10682 {
10683         return -EAGAIN;
10684 }
10685
10686 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10687 {
10688         struct inode *inode = private_data;
10689         return btrfs_sb(inode->i_sb);
10690 }
10691
10692 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10693                                         u64 start, u64 end)
10694 {
10695         struct inode *inode = private_data;
10696         u64 isize;
10697
10698         isize = i_size_read(inode);
10699         if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10700                 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10701                     "%s: ino %llu isize %llu odd range [%llu,%llu]",
10702                         caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10703         }
10704 }
10705
10706 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10707 {
10708         struct inode *inode = private_data;
10709         unsigned long index = start >> PAGE_SHIFT;
10710         unsigned long end_index = end >> PAGE_SHIFT;
10711         struct page *page;
10712
10713         while (index <= end_index) {
10714                 page = find_get_page(inode->i_mapping, index);
10715                 ASSERT(page); /* Pages should be in the extent_io_tree */
10716                 set_page_writeback(page);
10717                 put_page(page);
10718                 index++;
10719         }
10720 }
10721
10722 static const struct inode_operations btrfs_dir_inode_operations = {
10723         .getattr        = btrfs_getattr,
10724         .lookup         = btrfs_lookup,
10725         .create         = btrfs_create,
10726         .unlink         = btrfs_unlink,
10727         .link           = btrfs_link,
10728         .mkdir          = btrfs_mkdir,
10729         .rmdir          = btrfs_rmdir,
10730         .rename         = btrfs_rename2,
10731         .symlink        = btrfs_symlink,
10732         .setattr        = btrfs_setattr,
10733         .mknod          = btrfs_mknod,
10734         .listxattr      = btrfs_listxattr,
10735         .permission     = btrfs_permission,
10736         .get_acl        = btrfs_get_acl,
10737         .set_acl        = btrfs_set_acl,
10738         .update_time    = btrfs_update_time,
10739         .tmpfile        = btrfs_tmpfile,
10740 };
10741 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10742         .lookup         = btrfs_lookup,
10743         .permission     = btrfs_permission,
10744         .update_time    = btrfs_update_time,
10745 };
10746
10747 static const struct file_operations btrfs_dir_file_operations = {
10748         .llseek         = generic_file_llseek,
10749         .read           = generic_read_dir,
10750         .iterate_shared = btrfs_real_readdir,
10751         .unlocked_ioctl = btrfs_ioctl,
10752 #ifdef CONFIG_COMPAT
10753         .compat_ioctl   = btrfs_compat_ioctl,
10754 #endif
10755         .release        = btrfs_release_file,
10756         .fsync          = btrfs_sync_file,
10757 };
10758
10759 static const struct extent_io_ops btrfs_extent_io_ops = {
10760         /* mandatory callbacks */
10761         .submit_bio_hook = btrfs_submit_bio_hook,
10762         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10763         .merge_bio_hook = btrfs_merge_bio_hook,
10764         .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10765         .tree_fs_info = iotree_fs_info,
10766         .set_range_writeback = btrfs_set_range_writeback,
10767
10768         /* optional callbacks */
10769         .fill_delalloc = run_delalloc_range,
10770         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10771         .writepage_start_hook = btrfs_writepage_start_hook,
10772         .set_bit_hook = btrfs_set_bit_hook,
10773         .clear_bit_hook = btrfs_clear_bit_hook,
10774         .merge_extent_hook = btrfs_merge_extent_hook,
10775         .split_extent_hook = btrfs_split_extent_hook,
10776         .check_extent_io_range = btrfs_check_extent_io_range,
10777 };
10778
10779 /*
10780  * btrfs doesn't support the bmap operation because swapfiles
10781  * use bmap to make a mapping of extents in the file.  They assume
10782  * these extents won't change over the life of the file and they
10783  * use the bmap result to do IO directly to the drive.
10784  *
10785  * the btrfs bmap call would return logical addresses that aren't
10786  * suitable for IO and they also will change frequently as COW
10787  * operations happen.  So, swapfile + btrfs == corruption.
10788  *
10789  * For now we're avoiding this by dropping bmap.
10790  */
10791 static const struct address_space_operations btrfs_aops = {
10792         .readpage       = btrfs_readpage,
10793         .writepage      = btrfs_writepage,
10794         .writepages     = btrfs_writepages,
10795         .readpages      = btrfs_readpages,
10796         .direct_IO      = btrfs_direct_IO,
10797         .invalidatepage = btrfs_invalidatepage,
10798         .releasepage    = btrfs_releasepage,
10799         .set_page_dirty = btrfs_set_page_dirty,
10800         .error_remove_page = generic_error_remove_page,
10801 };
10802
10803 static const struct address_space_operations btrfs_symlink_aops = {
10804         .readpage       = btrfs_readpage,
10805         .writepage      = btrfs_writepage,
10806         .invalidatepage = btrfs_invalidatepage,
10807         .releasepage    = btrfs_releasepage,
10808 };
10809
10810 static const struct inode_operations btrfs_file_inode_operations = {
10811         .getattr        = btrfs_getattr,
10812         .setattr        = btrfs_setattr,
10813         .listxattr      = btrfs_listxattr,
10814         .permission     = btrfs_permission,
10815         .fiemap         = btrfs_fiemap,
10816         .get_acl        = btrfs_get_acl,
10817         .set_acl        = btrfs_set_acl,
10818         .update_time    = btrfs_update_time,
10819 };
10820 static const struct inode_operations btrfs_special_inode_operations = {
10821         .getattr        = btrfs_getattr,
10822         .setattr        = btrfs_setattr,
10823         .permission     = btrfs_permission,
10824         .listxattr      = btrfs_listxattr,
10825         .get_acl        = btrfs_get_acl,
10826         .set_acl        = btrfs_set_acl,
10827         .update_time    = btrfs_update_time,
10828 };
10829 static const struct inode_operations btrfs_symlink_inode_operations = {
10830         .get_link       = page_get_link,
10831         .getattr        = btrfs_getattr,
10832         .setattr        = btrfs_setattr,
10833         .permission     = btrfs_permission,
10834         .listxattr      = btrfs_listxattr,
10835         .update_time    = btrfs_update_time,
10836 };
10837
10838 const struct dentry_operations btrfs_dentry_operations = {
10839         .d_delete       = btrfs_dentry_delete,
10840         .d_release      = btrfs_dentry_release,
10841 };