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