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/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end >= PAGE_CACHE_SIZE ||
 253             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 /*
 352  * we create compressed extents in two phases.  The first
 353  * phase compresses a range of pages that have already been
 354  * locked (both pages and state bits are locked).
 355  *
 356  * This is done inside an ordered work queue, and the compression
 357  * is spread across many cpus.  The actual IO submission is step
 358  * two, and the ordered work queue takes care of making sure that
 359  * happens in the same order things were put onto the queue by
 360  * writepages and friends.
 361  *
 362  * If this code finds it can't get good compression, it puts an
 363  * entry onto the work queue to write the uncompressed bytes.  This
 364  * makes sure that both compressed inodes and uncompressed inodes
 365  * are written in the same order that the flusher thread sent them
 366  * down.
 367  */
 368 static noinline int compress_file_range(struct inode *inode,
 369                                         struct page *locked_page,
 370                                         u64 start, u64 end,
 371                                         struct async_cow *async_cow,
 372                                         int *num_added)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         u64 num_bytes;
 376         u64 blocksize = root->sectorsize;
 377         u64 actual_end;
 378         u64 isize = i_size_read(inode);
 379         int ret = 0;
 380         struct page **pages = NULL;
 381         unsigned long nr_pages;
 382         unsigned long nr_pages_ret = 0;
 383         unsigned long total_compressed = 0;
 384         unsigned long total_in = 0;
 385         unsigned long max_compressed = 128 * 1024;
 386         unsigned long max_uncompressed = 128 * 1024;
 387         int i;
 388         int will_compress;
 389         int compress_type = root->fs_info->compress_type;
 390         int redirty = 0;
 391
 392         /* if this is a small write inside eof, kick off a defrag */
 393         if ((end - start + 1) < 16 * 1024 &&
 394             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 395                 btrfs_add_inode_defrag(NULL, inode);
 396
 397         /*
 398          * skip compression for a small file range(<=blocksize) that
 399          * isn't an inline extent, since it dosen't save disk space at all.
 400          */
 401         if ((end - start + 1) <= blocksize &&
 402             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 403                 goto cleanup_and_bail_uncompressed;
 404
 405         actual_end = min_t(u64, isize, end + 1);
 406 again:
 407         will_compress = 0;
 408         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 409         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 410
 411         /*
 412          * we don't want to send crud past the end of i_size through
 413          * compression, that's just a waste of CPU time.  So, if the
 414          * end of the file is before the start of our current
 415          * requested range of bytes, we bail out to the uncompressed
 416          * cleanup code that can deal with all of this.
 417          *
 418          * It isn't really the fastest way to fix things, but this is a
 419          * very uncommon corner.
 420          */
 421         if (actual_end <= start)
 422                 goto cleanup_and_bail_uncompressed;
 423
 424         total_compressed = actual_end - start;
 425
 426         /* we want to make sure that amount of ram required to uncompress
 427          * an extent is reasonable, so we limit the total size in ram
 428          * of a compressed extent to 128k.  This is a crucial number
 429          * because it also controls how easily we can spread reads across
 430          * cpus for decompression.
 431          *
 432          * We also want to make sure the amount of IO required to do
 433          * a random read is reasonably small, so we limit the size of
 434          * a compressed extent to 128k.
 435          */
 436         total_compressed = min(total_compressed, max_uncompressed);
 437         num_bytes = ALIGN(end - start + 1, blocksize);
 438         num_bytes = max(blocksize,  num_bytes);
 439         total_in = 0;
 440         ret = 0;
 441
 442         /*
 443          * we do compression for mount -o compress and when the
 444          * inode has not been flagged as nocompress.  This flag can
 445          * change at any time if we discover bad compression ratios.
 446          */
 447         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 448             (btrfs_test_opt(root, COMPRESS) ||
 449              (BTRFS_I(inode)->force_compress) ||
 450              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 451                 WARN_ON(pages);
 452                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 453                 if (!pages) {
 454                         /* just bail out to the uncompressed code */
 455                         goto cont;
 456                 }
 457
 458                 if (BTRFS_I(inode)->force_compress)
 459                         compress_type = BTRFS_I(inode)->force_compress;
 460
 461                 /*
 462                  * we need to call clear_page_dirty_for_io on each
 463                  * page in the range.  Otherwise applications with the file
 464                  * mmap'd can wander in and change the page contents while
 465                  * we are compressing them.
 466                  *
 467                  * If the compression fails for any reason, we set the pages
 468                  * dirty again later on.
 469                  */
 470                 extent_range_clear_dirty_for_io(inode, start, end);
 471                 redirty = 1;
 472                 ret = btrfs_compress_pages(compress_type,
 473                                            inode->i_mapping, start,
 474                                            total_compressed, pages,
 475                                            nr_pages, &nr_pages_ret,
 476                                            &total_in,
 477                                            &total_compressed,
 478                                            max_compressed);
 479
 480                 if (!ret) {
 481                         unsigned long offset = total_compressed &
 482                                 (PAGE_CACHE_SIZE - 1);
 483                         struct page *page = pages[nr_pages_ret - 1];
 484                         char *kaddr;
 485
 486                         /* zero the tail end of the last page, we might be
 487                          * sending it down to disk
 488                          */
 489                         if (offset) {
 490                                 kaddr = kmap_atomic(page);
 491                                 memset(kaddr + offset, 0,
 492                                        PAGE_CACHE_SIZE - offset);
 493                                 kunmap_atomic(kaddr);
 494                         }
 495                         will_compress = 1;
 496                 }
 497         }
 498 cont:
 499         if (start == 0) {
 500                 /* lets try to make an inline extent */
 501                 if (ret || total_in < (actual_end - start)) {
 502                         /* we didn't compress the entire range, try
 503                          * to make an uncompressed inline extent.
 504                          */
 505                         ret = cow_file_range_inline(root, inode, start, end,
 506                                                     0, 0, NULL);
 507                 } else {
 508                         /* try making a compressed inline extent */
 509                         ret = cow_file_range_inline(root, inode, start, end,
 510                                                     total_compressed,
 511                                                     compress_type, pages);
 512                 }
 513                 if (ret <= 0) {
 514                         unsigned long clear_flags = EXTENT_DELALLOC |
 515                                 EXTENT_DEFRAG;
 516                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 517
 518                         /*
 519                          * inline extent creation worked or returned error,
 520                          * we don't need to create any more async work items.
 521                          * Unlock and free up our temp pages.
 522                          */
 523                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 524                                                      clear_flags, PAGE_UNLOCK |
 525                                                      PAGE_CLEAR_DIRTY |
 526                                                      PAGE_SET_WRITEBACK |
 527                                                      PAGE_END_WRITEBACK);
 528                         goto free_pages_out;
 529                 }
 530         }
 531
 532         if (will_compress) {
 533                 /*
 534                  * we aren't doing an inline extent round the compressed size
 535                  * up to a block size boundary so the allocator does sane
 536                  * things
 537                  */
 538                 total_compressed = ALIGN(total_compressed, blocksize);
 539
 540                 /*
 541                  * one last check to make sure the compression is really a
 542                  * win, compare the page count read with the blocks on disk
 543                  */
 544                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 545                 if (total_compressed >= total_in) {
 546                         will_compress = 0;
 547                 } else {
 548                         num_bytes = total_in;
 549                 }
 550         }
 551         if (!will_compress && pages) {
 552                 /*
 553                  * the compression code ran but failed to make things smaller,
 554                  * free any pages it allocated and our page pointer array
 555                  */
 556                 for (i = 0; i < nr_pages_ret; i++) {
 557                         WARN_ON(pages[i]->mapping);
 558                         page_cache_release(pages[i]);
 559                 }
 560                 kfree(pages);
 561                 pages = NULL;
 562                 total_compressed = 0;
 563                 nr_pages_ret = 0;
 564
 565                 /* flag the file so we don't compress in the future */
 566                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 567                     !(BTRFS_I(inode)->force_compress)) {
 568                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 569                 }
 570         }
 571         if (will_compress) {
 572                 *num_added += 1;
 573
 574                 /* the async work queues will take care of doing actual
 575                  * allocation on disk for these compressed pages,
 576                  * and will submit them to the elevator.
 577                  */
 578                 add_async_extent(async_cow, start, num_bytes,
 579                                  total_compressed, pages, nr_pages_ret,
 580                                  compress_type);
 581
 582                 if (start + num_bytes < end) {
 583                         start += num_bytes;
 584                         pages = NULL;
 585                         cond_resched();
 586                         goto again;
 587                 }
 588         } else {
 589 cleanup_and_bail_uncompressed:
 590                 /*
 591                  * No compression, but we still need to write the pages in
 592                  * the file we've been given so far.  redirty the locked
 593                  * page if it corresponds to our extent and set things up
 594                  * for the async work queue to run cow_file_range to do
 595                  * the normal delalloc dance
 596                  */
 597                 if (page_offset(locked_page) >= start &&
 598                     page_offset(locked_page) <= end) {
 599                         __set_page_dirty_nobuffers(locked_page);
 600                         /* unlocked later on in the async handlers */
 601                 }
 602                 if (redirty)
 603                         extent_range_redirty_for_io(inode, start, end);
 604                 add_async_extent(async_cow, start, end - start + 1,
 605                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 606                 *num_added += 1;
 607         }
 608
 609 out:
 610         return ret;
 611
 612 free_pages_out:
 613         for (i = 0; i < nr_pages_ret; i++) {
 614                 WARN_ON(pages[i]->mapping);
 615                 page_cache_release(pages[i]);
 616         }
 617         kfree(pages);
 618
 619         goto out;
 620 }
 621
 622 /*
 623  * phase two of compressed writeback.  This is the ordered portion
 624  * of the code, which only gets called in the order the work was
 625  * queued.  We walk all the async extents created by compress_file_range
 626  * and send them down to the disk.
 627  */
 628 static noinline int submit_compressed_extents(struct inode *inode,
 629                                               struct async_cow *async_cow)
 630 {
 631         struct async_extent *async_extent;
 632         u64 alloc_hint = 0;
 633         struct btrfs_key ins;
 634         struct extent_map *em;
 635         struct btrfs_root *root = BTRFS_I(inode)->root;
 636         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 637         struct extent_io_tree *io_tree;
 638         int ret = 0;
 639
 640         if (list_empty(&async_cow->extents))
 641                 return 0;
 642
 643 again:
 644         while (!list_empty(&async_cow->extents)) {
 645                 async_extent = list_entry(async_cow->extents.next,
 646                                           struct async_extent, list);
 647                 list_del(&async_extent->list);
 648
 649                 io_tree = &BTRFS_I(inode)->io_tree;
 650
 651 retry:
 652                 /* did the compression code fall back to uncompressed IO? */
 653                 if (!async_extent->pages) {
 654                         int page_started = 0;
 655                         unsigned long nr_written = 0;
 656
 657                         lock_extent(io_tree, async_extent->start,
 658                                          async_extent->start +
 659                                          async_extent->ram_size - 1);
 660
 661                         /* allocate blocks */
 662                         ret = cow_file_range(inode, async_cow->locked_page,
 663                                              async_extent->start,
 664                                              async_extent->start +
 665                                              async_extent->ram_size - 1,
 666                                              &page_started, &nr_written, 0);
 667
 668                         /* JDM XXX */
 669
 670                         /*
 671                          * if page_started, cow_file_range inserted an
 672                          * inline extent and took care of all the unlocking
 673                          * and IO for us.  Otherwise, we need to submit
 674                          * all those pages down to the drive.
 675                          */
 676                         if (!page_started && !ret)
 677                                 extent_write_locked_range(io_tree,
 678                                                   inode, async_extent->start,
 679                                                   async_extent->start +
 680                                                   async_extent->ram_size - 1,
 681                                                   btrfs_get_extent,
 682                                                   WB_SYNC_ALL);
 683                         else if (ret)
 684                                 unlock_page(async_cow->locked_page);
 685                         kfree(async_extent);
 686                         cond_resched();
 687                         continue;
 688                 }
 689
 690                 lock_extent(io_tree, async_extent->start,
 691                             async_extent->start + async_extent->ram_size - 1);
 692
 693                 ret = btrfs_reserve_extent(root,
 694                                            async_extent->compressed_size,
 695                                            async_extent->compressed_size,
 696                                            0, alloc_hint, &ins, 1, 1);
 697                 if (ret) {
 698                         int i;
 699
 700                         for (i = 0; i < async_extent->nr_pages; i++) {
 701                                 WARN_ON(async_extent->pages[i]->mapping);
 702                                 page_cache_release(async_extent->pages[i]);
 703                         }
 704                         kfree(async_extent->pages);
 705                         async_extent->nr_pages = 0;
 706                         async_extent->pages = NULL;
 707
 708                         if (ret == -ENOSPC) {
 709                                 unlock_extent(io_tree, async_extent->start,
 710                                               async_extent->start +
 711                                               async_extent->ram_size - 1);
 712
 713                                 /*
 714                                  * we need to redirty the pages if we decide to
 715                                  * fallback to uncompressed IO, otherwise we
 716                                  * will not submit these pages down to lower
 717                                  * layers.
 718                                  */
 719                                 extent_range_redirty_for_io(inode,
 720                                                 async_extent->start,
 721                                                 async_extent->start +
 722                                                 async_extent->ram_size - 1);
 723
 724                                 goto retry;
 725                         }
 726                         goto out_free;
 727                 }
 728
 729                 /*
 730                  * here we're doing allocation and writeback of the
 731                  * compressed pages
 732                  */
 733                 btrfs_drop_extent_cache(inode, async_extent->start,
 734                                         async_extent->start +
 735                                         async_extent->ram_size - 1, 0);
 736
 737                 em = alloc_extent_map();
 738                 if (!em) {
 739                         ret = -ENOMEM;
 740                         goto out_free_reserve;
 741                 }
 742                 em->start = async_extent->start;
 743                 em->len = async_extent->ram_size;
 744                 em->orig_start = em->start;
 745                 em->mod_start = em->start;
 746                 em->mod_len = em->len;
 747
 748                 em->block_start = ins.objectid;
 749                 em->block_len = ins.offset;
 750                 em->orig_block_len = ins.offset;
 751                 em->ram_bytes = async_extent->ram_size;
 752                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 753                 em->compress_type = async_extent->compress_type;
 754                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 755                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 756                 em->generation = -1;
 757
 758                 while (1) {
 759                         write_lock(&em_tree->lock);
 760                         ret = add_extent_mapping(em_tree, em, 1);
 761                         write_unlock(&em_tree->lock);
 762                         if (ret != -EEXIST) {
 763                                 free_extent_map(em);
 764                                 break;
 765                         }
 766                         btrfs_drop_extent_cache(inode, async_extent->start,
 767                                                 async_extent->start +
 768                                                 async_extent->ram_size - 1, 0);
 769                 }
 770
 771                 if (ret)
 772                         goto out_free_reserve;
 773
 774                 ret = btrfs_add_ordered_extent_compress(inode,
 775                                                 async_extent->start,
 776                                                 ins.objectid,
 777                                                 async_extent->ram_size,
 778                                                 ins.offset,
 779                                                 BTRFS_ORDERED_COMPRESSED,
 780                                                 async_extent->compress_type);
 781                 if (ret) {
 782                         btrfs_drop_extent_cache(inode, async_extent->start,
 783                                                 async_extent->start +
 784                                                 async_extent->ram_size - 1, 0);
 785                         goto out_free_reserve;
 786                 }
 787
 788                 /*
 789                  * clear dirty, set writeback and unlock the pages.
 790                  */
 791                 extent_clear_unlock_delalloc(inode, async_extent->start,
 792                                 async_extent->start +
 793                                 async_extent->ram_size - 1,
 794                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 795                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 796                                 PAGE_SET_WRITEBACK);
 797                 ret = btrfs_submit_compressed_write(inode,
 798                                     async_extent->start,
 799                                     async_extent->ram_size,
 800                                     ins.objectid,
 801                                     ins.offset, async_extent->pages,
 802                                     async_extent->nr_pages);
 803                 alloc_hint = ins.objectid + ins.offset;
 804                 kfree(async_extent);
 805                 if (ret)
 806                         goto out;
 807                 cond_resched();
 808         }
 809         ret = 0;
 810 out:
 811         return ret;
 812 out_free_reserve:
 813         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 814 out_free:
 815         extent_clear_unlock_delalloc(inode, async_extent->start,
 816                                      async_extent->start +
 817                                      async_extent->ram_size - 1,
 818                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 819                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 820                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 821                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 822         kfree(async_extent);
 823         goto again;
 824 }
 825
 826 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 827                                       u64 num_bytes)
 828 {
 829         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 830         struct extent_map *em;
 831         u64 alloc_hint = 0;
 832
 833         read_lock(&em_tree->lock);
 834         em = search_extent_mapping(em_tree, start, num_bytes);
 835         if (em) {
 836                 /*
 837                  * if block start isn't an actual block number then find the
 838                  * first block in this inode and use that as a hint.  If that
 839                  * block is also bogus then just don't worry about it.
 840                  */
 841                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 842                         free_extent_map(em);
 843                         em = search_extent_mapping(em_tree, 0, 0);
 844                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 845                                 alloc_hint = em->block_start;
 846                         if (em)
 847                                 free_extent_map(em);
 848                 } else {
 849                         alloc_hint = em->block_start;
 850                         free_extent_map(em);
 851                 }
 852         }
 853         read_unlock(&em_tree->lock);
 854
 855         return alloc_hint;
 856 }
 857
 858 /*
 859  * when extent_io.c finds a delayed allocation range in the file,
 860  * the call backs end up in this code.  The basic idea is to
 861  * allocate extents on disk for the range, and create ordered data structs
 862  * in ram to track those extents.
 863  *
 864  * locked_page is the page that writepage had locked already.  We use
 865  * it to make sure we don't do extra locks or unlocks.
 866  *
 867  * *page_started is set to one if we unlock locked_page and do everything
 868  * required to start IO on it.  It may be clean and already done with
 869  * IO when we return.
 870  */
 871 static noinline int cow_file_range(struct inode *inode,
 872                                    struct page *locked_page,
 873                                    u64 start, u64 end, int *page_started,
 874                                    unsigned long *nr_written,
 875                                    int unlock)
 876 {
 877         struct btrfs_root *root = BTRFS_I(inode)->root;
 878         u64 alloc_hint = 0;
 879         u64 num_bytes;
 880         unsigned long ram_size;
 881         u64 disk_num_bytes;
 882         u64 cur_alloc_size;
 883         u64 blocksize = root->sectorsize;
 884         struct btrfs_key ins;
 885         struct extent_map *em;
 886         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 887         int ret = 0;
 888
 889         if (btrfs_is_free_space_inode(inode)) {
 890                 WARN_ON_ONCE(1);
 891                 ret = -EINVAL;
 892                 goto out_unlock;
 893         }
 894
 895         num_bytes = ALIGN(end - start + 1, blocksize);
 896         num_bytes = max(blocksize,  num_bytes);
 897         disk_num_bytes = num_bytes;
 898
 899         /* if this is a small write inside eof, kick off defrag */
 900         if (num_bytes < 64 * 1024 &&
 901             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 902                 btrfs_add_inode_defrag(NULL, inode);
 903
 904         if (start == 0) {
 905                 /* lets try to make an inline extent */
 906                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 907                                             NULL);
 908                 if (ret == 0) {
 909                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 910                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 911                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 912                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 913                                      PAGE_END_WRITEBACK);
 914
 915                         *nr_written = *nr_written +
 916                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 917                         *page_started = 1;
 918                         goto out;
 919                 } else if (ret < 0) {
 920                         goto out_unlock;
 921                 }
 922         }
 923
 924         BUG_ON(disk_num_bytes >
 925                btrfs_super_total_bytes(root->fs_info->super_copy));
 926
 927         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 928         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 929
 930         while (disk_num_bytes > 0) {
 931                 unsigned long op;
 932
 933                 cur_alloc_size = disk_num_bytes;
 934                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 935                                            root->sectorsize, 0, alloc_hint,
 936                                            &ins, 1, 1);
 937                 if (ret < 0)
 938                         goto out_unlock;
 939
 940                 em = alloc_extent_map();
 941                 if (!em) {
 942                         ret = -ENOMEM;
 943                         goto out_reserve;
 944                 }
 945                 em->start = start;
 946                 em->orig_start = em->start;
 947                 ram_size = ins.offset;
 948                 em->len = ins.offset;
 949                 em->mod_start = em->start;
 950                 em->mod_len = em->len;
 951
 952                 em->block_start = ins.objectid;
 953                 em->block_len = ins.offset;
 954                 em->orig_block_len = ins.offset;
 955                 em->ram_bytes = ram_size;
 956                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 957                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 958                 em->generation = -1;
 959
 960                 while (1) {
 961                         write_lock(&em_tree->lock);
 962                         ret = add_extent_mapping(em_tree, em, 1);
 963                         write_unlock(&em_tree->lock);
 964                         if (ret != -EEXIST) {
 965                                 free_extent_map(em);
 966                                 break;
 967                         }
 968                         btrfs_drop_extent_cache(inode, start,
 969                                                 start + ram_size - 1, 0);
 970                 }
 971                 if (ret)
 972                         goto out_reserve;
 973
 974                 cur_alloc_size = ins.offset;
 975                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 976                                                ram_size, cur_alloc_size, 0);
 977                 if (ret)
 978                         goto out_drop_extent_cache;
 979
 980                 if (root->root_key.objectid ==
 981                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 982                         ret = btrfs_reloc_clone_csums(inode, start,
 983                                                       cur_alloc_size);
 984                         if (ret)
 985                                 goto out_drop_extent_cache;
 986                 }
 987
 988                 if (disk_num_bytes < cur_alloc_size)
 989                         break;
 990
 991                 /* we're not doing compressed IO, don't unlock the first
 992                  * page (which the caller expects to stay locked), don't
 993                  * clear any dirty bits and don't set any writeback bits
 994                  *
 995                  * Do set the Private2 bit so we know this page was properly
 996                  * setup for writepage
 997                  */
 998                 op = unlock ? PAGE_UNLOCK : 0;
 999                 op |= PAGE_SET_PRIVATE2;
1000
1001                 extent_clear_unlock_delalloc(inode, start,
1002                                              start + ram_size - 1, locked_page,
1003                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1004                                              op);
1005                 disk_num_bytes -= cur_alloc_size;
1006                 num_bytes -= cur_alloc_size;
1007                 alloc_hint = ins.objectid + ins.offset;
1008                 start += cur_alloc_size;
1009         }
1010 out:
1011         return ret;
1012
1013 out_drop_extent_cache:
1014         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1015 out_reserve:
1016         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1017 out_unlock:
1018         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1019                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1020                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1021                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1022                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1023         goto out;
1024 }
1025
1026 /*
1027  * work queue call back to started compression on a file and pages
1028  */
1029 static noinline void async_cow_start(struct btrfs_work *work)
1030 {
1031         struct async_cow *async_cow;
1032         int num_added = 0;
1033         async_cow = container_of(work, struct async_cow, work);
1034
1035         compress_file_range(async_cow->inode, async_cow->locked_page,
1036                             async_cow->start, async_cow->end, async_cow,
1037                             &num_added);
1038         if (num_added == 0) {
1039                 btrfs_add_delayed_iput(async_cow->inode);
1040                 async_cow->inode = NULL;
1041         }
1042 }
1043
1044 /*
1045  * work queue call back to submit previously compressed pages
1046  */
1047 static noinline void async_cow_submit(struct btrfs_work *work)
1048 {
1049         struct async_cow *async_cow;
1050         struct btrfs_root *root;
1051         unsigned long nr_pages;
1052
1053         async_cow = container_of(work, struct async_cow, work);
1054
1055         root = async_cow->root;
1056         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1057                 PAGE_CACHE_SHIFT;
1058
1059         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1060             5 * 1024 * 1024 &&
1061             waitqueue_active(&root->fs_info->async_submit_wait))
1062                 wake_up(&root->fs_info->async_submit_wait);
1063
1064         if (async_cow->inode)
1065                 submit_compressed_extents(async_cow->inode, async_cow);
1066 }
1067
1068 static noinline void async_cow_free(struct btrfs_work *work)
1069 {
1070         struct async_cow *async_cow;
1071         async_cow = container_of(work, struct async_cow, work);
1072         if (async_cow->inode)
1073                 btrfs_add_delayed_iput(async_cow->inode);
1074         kfree(async_cow);
1075 }
1076
1077 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1078                                 u64 start, u64 end, int *page_started,
1079                                 unsigned long *nr_written)
1080 {
1081         struct async_cow *async_cow;
1082         struct btrfs_root *root = BTRFS_I(inode)->root;
1083         unsigned long nr_pages;
1084         u64 cur_end;
1085         int limit = 10 * 1024 * 1024;
1086
1087         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1088                          1, 0, NULL, GFP_NOFS);
1089         while (start < end) {
1090                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1091                 BUG_ON(!async_cow); /* -ENOMEM */
1092                 async_cow->inode = igrab(inode);
1093                 async_cow->root = root;
1094                 async_cow->locked_page = locked_page;
1095                 async_cow->start = start;
1096
1097                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1098                         cur_end = end;
1099                 else
1100                         cur_end = min(end, start + 512 * 1024 - 1);
1101
1102                 async_cow->end = cur_end;
1103                 INIT_LIST_HEAD(&async_cow->extents);
1104
1105                 btrfs_init_work(&async_cow->work,
1106                                 btrfs_delalloc_helper,
1107                                 async_cow_start, async_cow_submit,
1108                                 async_cow_free);
1109
1110                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1111                         PAGE_CACHE_SHIFT;
1112                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1113
1114                 btrfs_queue_work(root->fs_info->delalloc_workers,
1115                                  &async_cow->work);
1116
1117                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1118                         wait_event(root->fs_info->async_submit_wait,
1119                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1120                             limit));
1121                 }
1122
1123                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1124                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1125                         wait_event(root->fs_info->async_submit_wait,
1126                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1127                            0));
1128                 }
1129
1130                 *nr_written += nr_pages;
1131                 start = cur_end + 1;
1132         }
1133         *page_started = 1;
1134         return 0;
1135 }
1136
1137 static noinline int csum_exist_in_range(struct btrfs_root *root,
1138                                         u64 bytenr, u64 num_bytes)
1139 {
1140         int ret;
1141         struct btrfs_ordered_sum *sums;
1142         LIST_HEAD(list);
1143
1144         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1145                                        bytenr + num_bytes - 1, &list, 0);
1146         if (ret == 0 && list_empty(&list))
1147                 return 0;
1148
1149         while (!list_empty(&list)) {
1150                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1151                 list_del(&sums->list);
1152                 kfree(sums);
1153         }
1154         return 1;
1155 }
1156
1157 /*
1158  * when nowcow writeback call back.  This checks for snapshots or COW copies
1159  * of the extents that exist in the file, and COWs the file as required.
1160  *
1161  * If no cow copies or snapshots exist, we write directly to the existing
1162  * blocks on disk
1163  */
1164 static noinline int run_delalloc_nocow(struct inode *inode,
1165                                        struct page *locked_page,
1166                               u64 start, u64 end, int *page_started, int force,
1167                               unsigned long *nr_written)
1168 {
1169         struct btrfs_root *root = BTRFS_I(inode)->root;
1170         struct btrfs_trans_handle *trans;
1171         struct extent_buffer *leaf;
1172         struct btrfs_path *path;
1173         struct btrfs_file_extent_item *fi;
1174         struct btrfs_key found_key;
1175         u64 cow_start;
1176         u64 cur_offset;
1177         u64 extent_end;
1178         u64 extent_offset;
1179         u64 disk_bytenr;
1180         u64 num_bytes;
1181         u64 disk_num_bytes;
1182         u64 ram_bytes;
1183         int extent_type;
1184         int ret, err;
1185         int type;
1186         int nocow;
1187         int check_prev = 1;
1188         bool nolock;
1189         u64 ino = btrfs_ino(inode);
1190
1191         path = btrfs_alloc_path();
1192         if (!path) {
1193                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1194                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1195                                              EXTENT_DO_ACCOUNTING |
1196                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1197                                              PAGE_CLEAR_DIRTY |
1198                                              PAGE_SET_WRITEBACK |
1199                                              PAGE_END_WRITEBACK);
1200                 return -ENOMEM;
1201         }
1202
1203         nolock = btrfs_is_free_space_inode(inode);
1204
1205         if (nolock)
1206                 trans = btrfs_join_transaction_nolock(root);
1207         else
1208                 trans = btrfs_join_transaction(root);
1209
1210         if (IS_ERR(trans)) {
1211                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1212                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1213                                              EXTENT_DO_ACCOUNTING |
1214                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1215                                              PAGE_CLEAR_DIRTY |
1216                                              PAGE_SET_WRITEBACK |
1217                                              PAGE_END_WRITEBACK);
1218                 btrfs_free_path(path);
1219                 return PTR_ERR(trans);
1220         }
1221
1222         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1223
1224         cow_start = (u64)-1;
1225         cur_offset = start;
1226         while (1) {
1227                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1228                                                cur_offset, 0);
1229                 if (ret < 0)
1230                         goto error;
1231                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1232                         leaf = path->nodes[0];
1233                         btrfs_item_key_to_cpu(leaf, &found_key,
1234                                               path->slots[0] - 1);
1235                         if (found_key.objectid == ino &&
1236                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1237                                 path->slots[0]--;
1238                 }
1239                 check_prev = 0;
1240 next_slot:
1241                 leaf = path->nodes[0];
1242                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1243                         ret = btrfs_next_leaf(root, path);
1244                         if (ret < 0)
1245                                 goto error;
1246                         if (ret > 0)
1247                                 break;
1248                         leaf = path->nodes[0];
1249                 }
1250
1251                 nocow = 0;
1252                 disk_bytenr = 0;
1253                 num_bytes = 0;
1254                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1255
1256                 if (found_key.objectid > ino ||
1257                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1258                     found_key.offset > end)
1259                         break;
1260
1261                 if (found_key.offset > cur_offset) {
1262                         extent_end = found_key.offset;
1263                         extent_type = 0;
1264                         goto out_check;
1265                 }
1266
1267                 fi = btrfs_item_ptr(leaf, path->slots[0],
1268                                     struct btrfs_file_extent_item);
1269                 extent_type = btrfs_file_extent_type(leaf, fi);
1270
1271                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1272                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1273                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1274                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1275                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1276                         extent_end = found_key.offset +
1277                                 btrfs_file_extent_num_bytes(leaf, fi);
1278                         disk_num_bytes =
1279                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1280                         if (extent_end <= start) {
1281                                 path->slots[0]++;
1282                                 goto next_slot;
1283                         }
1284                         if (disk_bytenr == 0)
1285                                 goto out_check;
1286                         if (btrfs_file_extent_compression(leaf, fi) ||
1287                             btrfs_file_extent_encryption(leaf, fi) ||
1288                             btrfs_file_extent_other_encoding(leaf, fi))
1289                                 goto out_check;
1290                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1291                                 goto out_check;
1292                         if (btrfs_extent_readonly(root, disk_bytenr))
1293                                 goto out_check;
1294                         if (btrfs_cross_ref_exist(trans, root, ino,
1295                                                   found_key.offset -
1296                                                   extent_offset, disk_bytenr))
1297                                 goto out_check;
1298                         disk_bytenr += extent_offset;
1299                         disk_bytenr += cur_offset - found_key.offset;
1300                         num_bytes = min(end + 1, extent_end) - cur_offset;
1301                         /*
1302                          * if there are pending snapshots for this root,
1303                          * we fall into common COW way.
1304                          */
1305                         if (!nolock) {
1306                                 err = btrfs_start_nocow_write(root);
1307                                 if (!err)
1308                                         goto out_check;
1309                         }
1310                         /*
1311                          * force cow if csum exists in the range.
1312                          * this ensure that csum for a given extent are
1313                          * either valid or do not exist.
1314                          */
1315                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1316                                 goto out_check;
1317                         nocow = 1;
1318                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1319                         extent_end = found_key.offset +
1320                                 btrfs_file_extent_inline_len(leaf,
1321                                                      path->slots[0], fi);
1322                         extent_end = ALIGN(extent_end, root->sectorsize);
1323                 } else {
1324                         BUG_ON(1);
1325                 }
1326 out_check:
1327                 if (extent_end <= start) {
1328                         path->slots[0]++;
1329                         if (!nolock && nocow)
1330                                 btrfs_end_nocow_write(root);
1331                         goto next_slot;
1332                 }
1333                 if (!nocow) {
1334                         if (cow_start == (u64)-1)
1335                                 cow_start = cur_offset;
1336                         cur_offset = extent_end;
1337                         if (cur_offset > end)
1338                                 break;
1339                         path->slots[0]++;
1340                         goto next_slot;
1341                 }
1342
1343                 btrfs_release_path(path);
1344                 if (cow_start != (u64)-1) {
1345                         ret = cow_file_range(inode, locked_page,
1346                                              cow_start, found_key.offset - 1,
1347                                              page_started, nr_written, 1);
1348                         if (ret) {
1349                                 if (!nolock && nocow)
1350                                         btrfs_end_nocow_write(root);
1351                                 goto error;
1352                         }
1353                         cow_start = (u64)-1;
1354                 }
1355
1356                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1357                         struct extent_map *em;
1358                         struct extent_map_tree *em_tree;
1359                         em_tree = &BTRFS_I(inode)->extent_tree;
1360                         em = alloc_extent_map();
1361                         BUG_ON(!em); /* -ENOMEM */
1362                         em->start = cur_offset;
1363                         em->orig_start = found_key.offset - extent_offset;
1364                         em->len = num_bytes;
1365                         em->block_len = num_bytes;
1366                         em->block_start = disk_bytenr;
1367                         em->orig_block_len = disk_num_bytes;
1368                         em->ram_bytes = ram_bytes;
1369                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1370                         em->mod_start = em->start;
1371                         em->mod_len = em->len;
1372                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1373                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1374                         em->generation = -1;
1375                         while (1) {
1376                                 write_lock(&em_tree->lock);
1377                                 ret = add_extent_mapping(em_tree, em, 1);
1378                                 write_unlock(&em_tree->lock);
1379                                 if (ret != -EEXIST) {
1380                                         free_extent_map(em);
1381                                         break;
1382                                 }
1383                                 btrfs_drop_extent_cache(inode, em->start,
1384                                                 em->start + em->len - 1, 0);
1385                         }
1386                         type = BTRFS_ORDERED_PREALLOC;
1387                 } else {
1388                         type = BTRFS_ORDERED_NOCOW;
1389                 }
1390
1391                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1392                                                num_bytes, num_bytes, type);
1393                 BUG_ON(ret); /* -ENOMEM */
1394
1395                 if (root->root_key.objectid ==
1396                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1397                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1398                                                       num_bytes);
1399                         if (ret) {
1400                                 if (!nolock && nocow)
1401                                         btrfs_end_nocow_write(root);
1402                                 goto error;
1403                         }
1404                 }
1405
1406                 extent_clear_unlock_delalloc(inode, cur_offset,
1407                                              cur_offset + num_bytes - 1,
1408                                              locked_page, EXTENT_LOCKED |
1409                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1410                                              PAGE_SET_PRIVATE2);
1411                 if (!nolock && nocow)
1412                         btrfs_end_nocow_write(root);
1413                 cur_offset = extent_end;
1414                 if (cur_offset > end)
1415                         break;
1416         }
1417         btrfs_release_path(path);
1418
1419         if (cur_offset <= end && cow_start == (u64)-1) {
1420                 cow_start = cur_offset;
1421                 cur_offset = end;
1422         }
1423
1424         if (cow_start != (u64)-1) {
1425                 ret = cow_file_range(inode, locked_page, cow_start, end,
1426                                      page_started, nr_written, 1);
1427                 if (ret)
1428                         goto error;
1429         }
1430
1431 error:
1432         err = btrfs_end_transaction(trans, root);
1433         if (!ret)
1434                 ret = err;
1435
1436         if (ret && cur_offset < end)
1437                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1438                                              locked_page, EXTENT_LOCKED |
1439                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1440                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1441                                              PAGE_CLEAR_DIRTY |
1442                                              PAGE_SET_WRITEBACK |
1443                                              PAGE_END_WRITEBACK);
1444         btrfs_free_path(path);
1445         return ret;
1446 }
1447
1448 /*
1449  * extent_io.c call back to do delayed allocation processing
1450  */
1451 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1452                               u64 start, u64 end, int *page_started,
1453                               unsigned long *nr_written)
1454 {
1455         int ret;
1456         struct btrfs_root *root = BTRFS_I(inode)->root;
1457
1458         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1459                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1460                                          page_started, 1, nr_written);
1461         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1462                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1463                                          page_started, 0, nr_written);
1464         } else if (!btrfs_test_opt(root, COMPRESS) &&
1465                    !(BTRFS_I(inode)->force_compress) &&
1466                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1467                 ret = cow_file_range(inode, locked_page, start, end,
1468                                       page_started, nr_written, 1);
1469         } else {
1470                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1471                         &BTRFS_I(inode)->runtime_flags);
1472                 ret = cow_file_range_async(inode, locked_page, start, end,
1473                                            page_started, nr_written);
1474         }
1475         return ret;
1476 }
1477
1478 static void btrfs_split_extent_hook(struct inode *inode,
1479                                     struct extent_state *orig, u64 split)
1480 {
1481         /* not delalloc, ignore it */
1482         if (!(orig->state & EXTENT_DELALLOC))
1483                 return;
1484
1485         spin_lock(&BTRFS_I(inode)->lock);
1486         BTRFS_I(inode)->outstanding_extents++;
1487         spin_unlock(&BTRFS_I(inode)->lock);
1488 }
1489
1490 /*
1491  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1492  * extents so we can keep track of new extents that are just merged onto old
1493  * extents, such as when we are doing sequential writes, so we can properly
1494  * account for the metadata space we'll need.
1495  */
1496 static void btrfs_merge_extent_hook(struct inode *inode,
1497                                     struct extent_state *new,
1498                                     struct extent_state *other)
1499 {
1500         /* not delalloc, ignore it */
1501         if (!(other->state & EXTENT_DELALLOC))
1502                 return;
1503
1504         spin_lock(&BTRFS_I(inode)->lock);
1505         BTRFS_I(inode)->outstanding_extents--;
1506         spin_unlock(&BTRFS_I(inode)->lock);
1507 }
1508
1509 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1510                                       struct inode *inode)
1511 {
1512         spin_lock(&root->delalloc_lock);
1513         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1514                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1515                               &root->delalloc_inodes);
1516                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1517                         &BTRFS_I(inode)->runtime_flags);
1518                 root->nr_delalloc_inodes++;
1519                 if (root->nr_delalloc_inodes == 1) {
1520                         spin_lock(&root->fs_info->delalloc_root_lock);
1521                         BUG_ON(!list_empty(&root->delalloc_root));
1522                         list_add_tail(&root->delalloc_root,
1523                                       &root->fs_info->delalloc_roots);
1524                         spin_unlock(&root->fs_info->delalloc_root_lock);
1525                 }
1526         }
1527         spin_unlock(&root->delalloc_lock);
1528 }
1529
1530 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1531                                      struct inode *inode)
1532 {
1533         spin_lock(&root->delalloc_lock);
1534         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1535                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1536                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1537                           &BTRFS_I(inode)->runtime_flags);
1538                 root->nr_delalloc_inodes--;
1539                 if (!root->nr_delalloc_inodes) {
1540                         spin_lock(&root->fs_info->delalloc_root_lock);
1541                         BUG_ON(list_empty(&root->delalloc_root));
1542                         list_del_init(&root->delalloc_root);
1543                         spin_unlock(&root->fs_info->delalloc_root_lock);
1544                 }
1545         }
1546         spin_unlock(&root->delalloc_lock);
1547 }
1548
1549 /*
1550  * extent_io.c set_bit_hook, used to track delayed allocation
1551  * bytes in this file, and to maintain the list of inodes that
1552  * have pending delalloc work to be done.
1553  */
1554 static void btrfs_set_bit_hook(struct inode *inode,
1555                                struct extent_state *state, unsigned long *bits)
1556 {
1557
1558         /*
1559          * set_bit and clear bit hooks normally require _irqsave/restore
1560          * but in this case, we are only testing for the DELALLOC
1561          * bit, which is only set or cleared with irqs on
1562          */
1563         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1564                 struct btrfs_root *root = BTRFS_I(inode)->root;
1565                 u64 len = state->end + 1 - state->start;
1566                 bool do_list = !btrfs_is_free_space_inode(inode);
1567
1568                 if (*bits & EXTENT_FIRST_DELALLOC) {
1569                         *bits &= ~EXTENT_FIRST_DELALLOC;
1570                 } else {
1571                         spin_lock(&BTRFS_I(inode)->lock);
1572                         BTRFS_I(inode)->outstanding_extents++;
1573                         spin_unlock(&BTRFS_I(inode)->lock);
1574                 }
1575
1576                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1577                                      root->fs_info->delalloc_batch);
1578                 spin_lock(&BTRFS_I(inode)->lock);
1579                 BTRFS_I(inode)->delalloc_bytes += len;
1580                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1581                                          &BTRFS_I(inode)->runtime_flags))
1582                         btrfs_add_delalloc_inodes(root, inode);
1583                 spin_unlock(&BTRFS_I(inode)->lock);
1584         }
1585 }
1586
1587 /*
1588  * extent_io.c clear_bit_hook, see set_bit_hook for why
1589  */
1590 static void btrfs_clear_bit_hook(struct inode *inode,
1591                                  struct extent_state *state,
1592                                  unsigned long *bits)
1593 {
1594         /*
1595          * set_bit and clear bit hooks normally require _irqsave/restore
1596          * but in this case, we are only testing for the DELALLOC
1597          * bit, which is only set or cleared with irqs on
1598          */
1599         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1600                 struct btrfs_root *root = BTRFS_I(inode)->root;
1601                 u64 len = state->end + 1 - state->start;
1602                 bool do_list = !btrfs_is_free_space_inode(inode);
1603
1604                 if (*bits & EXTENT_FIRST_DELALLOC) {
1605                         *bits &= ~EXTENT_FIRST_DELALLOC;
1606                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1607                         spin_lock(&BTRFS_I(inode)->lock);
1608                         BTRFS_I(inode)->outstanding_extents--;
1609                         spin_unlock(&BTRFS_I(inode)->lock);
1610                 }
1611
1612                 /*
1613                  * We don't reserve metadata space for space cache inodes so we
1614                  * don't need to call dellalloc_release_metadata if there is an
1615                  * error.
1616                  */
1617                 if (*bits & EXTENT_DO_ACCOUNTING &&
1618                     root != root->fs_info->tree_root)
1619                         btrfs_delalloc_release_metadata(inode, len);
1620
1621                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1622                     && do_list && !(state->state & EXTENT_NORESERVE))
1623                         btrfs_free_reserved_data_space(inode, len);
1624
1625                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1626                                      root->fs_info->delalloc_batch);
1627                 spin_lock(&BTRFS_I(inode)->lock);
1628                 BTRFS_I(inode)->delalloc_bytes -= len;
1629                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1630                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1631                              &BTRFS_I(inode)->runtime_flags))
1632                         btrfs_del_delalloc_inode(root, inode);
1633                 spin_unlock(&BTRFS_I(inode)->lock);
1634         }
1635 }
1636
1637 /*
1638  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1639  * we don't create bios that span stripes or chunks
1640  */
1641 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1642                          size_t size, struct bio *bio,
1643                          unsigned long bio_flags)
1644 {
1645         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1646         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1647         u64 length = 0;
1648         u64 map_length;
1649         int ret;
1650
1651         if (bio_flags & EXTENT_BIO_COMPRESSED)
1652                 return 0;
1653
1654         length = bio->bi_iter.bi_size;
1655         map_length = length;
1656         ret = btrfs_map_block(root->fs_info, rw, logical,
1657                               &map_length, NULL, 0);
1658         /* Will always return 0 with map_multi == NULL */
1659         BUG_ON(ret < 0);
1660         if (map_length < length + size)
1661                 return 1;
1662         return 0;
1663 }
1664
1665 /*
1666  * in order to insert checksums into the metadata in large chunks,
1667  * we wait until bio submission time.   All the pages in the bio are
1668  * checksummed and sums are attached onto the ordered extent record.
1669  *
1670  * At IO completion time the cums attached on the ordered extent record
1671  * are inserted into the btree
1672  */
1673 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1674                                     struct bio *bio, int mirror_num,
1675                                     unsigned long bio_flags,
1676                                     u64 bio_offset)
1677 {
1678         struct btrfs_root *root = BTRFS_I(inode)->root;
1679         int ret = 0;
1680
1681         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1682         BUG_ON(ret); /* -ENOMEM */
1683         return 0;
1684 }
1685
1686 /*
1687  * in order to insert checksums into the metadata in large chunks,
1688  * we wait until bio submission time.   All the pages in the bio are
1689  * checksummed and sums are attached onto the ordered extent record.
1690  *
1691  * At IO completion time the cums attached on the ordered extent record
1692  * are inserted into the btree
1693  */
1694 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1695                           int mirror_num, unsigned long bio_flags,
1696                           u64 bio_offset)
1697 {
1698         struct btrfs_root *root = BTRFS_I(inode)->root;
1699         int ret;
1700
1701         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1702         if (ret)
1703                 bio_endio(bio, ret);
1704         return ret;
1705 }
1706
1707 /*
1708  * extent_io.c submission hook. This does the right thing for csum calculation
1709  * on write, or reading the csums from the tree before a read
1710  */
1711 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1712                           int mirror_num, unsigned long bio_flags,
1713                           u64 bio_offset)
1714 {
1715         struct btrfs_root *root = BTRFS_I(inode)->root;
1716         int ret = 0;
1717         int skip_sum;
1718         int metadata = 0;
1719         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1720
1721         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1722
1723         if (btrfs_is_free_space_inode(inode))
1724                 metadata = 2;
1725
1726         if (!(rw & REQ_WRITE)) {
1727                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1728                 if (ret)
1729                         goto out;
1730
1731                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1732                         ret = btrfs_submit_compressed_read(inode, bio,
1733                                                            mirror_num,
1734                                                            bio_flags);
1735                         goto out;
1736                 } else if (!skip_sum) {
1737                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1738                         if (ret)
1739                                 goto out;
1740                 }
1741                 goto mapit;
1742         } else if (async && !skip_sum) {
1743                 /* csum items have already been cloned */
1744                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1745                         goto mapit;
1746                 /* we're doing a write, do the async checksumming */
1747                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1748                                    inode, rw, bio, mirror_num,
1749                                    bio_flags, bio_offset,
1750                                    __btrfs_submit_bio_start,
1751                                    __btrfs_submit_bio_done);
1752                 goto out;
1753         } else if (!skip_sum) {
1754                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1755                 if (ret)
1756                         goto out;
1757         }
1758
1759 mapit:
1760         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1761
1762 out:
1763         if (ret < 0)
1764                 bio_endio(bio, ret);
1765         return ret;
1766 }
1767
1768 /*
1769  * given a list of ordered sums record them in the inode.  This happens
1770  * at IO completion time based on sums calculated at bio submission time.
1771  */
1772 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1773                              struct inode *inode, u64 file_offset,
1774                              struct list_head *list)
1775 {
1776         struct btrfs_ordered_sum *sum;
1777
1778         list_for_each_entry(sum, list, list) {
1779                 trans->adding_csums = 1;
1780                 btrfs_csum_file_blocks(trans,
1781                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1782                 trans->adding_csums = 0;
1783         }
1784         return 0;
1785 }
1786
1787 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1788                               struct extent_state **cached_state)
1789 {
1790         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1791         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1792                                    cached_state, GFP_NOFS);
1793 }
1794
1795 /* see btrfs_writepage_start_hook for details on why this is required */
1796 struct btrfs_writepage_fixup {
1797         struct page *page;
1798         struct btrfs_work work;
1799 };
1800
1801 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1802 {
1803         struct btrfs_writepage_fixup *fixup;
1804         struct btrfs_ordered_extent *ordered;
1805         struct extent_state *cached_state = NULL;
1806         struct page *page;
1807         struct inode *inode;
1808         u64 page_start;
1809         u64 page_end;
1810         int ret;
1811
1812         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1813         page = fixup->page;
1814 again:
1815         lock_page(page);
1816         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1817                 ClearPageChecked(page);
1818                 goto out_page;
1819         }
1820
1821         inode = page->mapping->host;
1822         page_start = page_offset(page);
1823         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1824
1825         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1826                          &cached_state);
1827
1828         /* already ordered? We're done */
1829         if (PagePrivate2(page))
1830                 goto out;
1831
1832         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1833         if (ordered) {
1834                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1835                                      page_end, &cached_state, GFP_NOFS);
1836                 unlock_page(page);
1837                 btrfs_start_ordered_extent(inode, ordered, 1);
1838                 btrfs_put_ordered_extent(ordered);
1839                 goto again;
1840         }
1841
1842         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1843         if (ret) {
1844                 mapping_set_error(page->mapping, ret);
1845                 end_extent_writepage(page, ret, page_start, page_end);
1846                 ClearPageChecked(page);
1847                 goto out;
1848          }
1849
1850         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1851         ClearPageChecked(page);
1852         set_page_dirty(page);
1853 out:
1854         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1855                              &cached_state, GFP_NOFS);
1856 out_page:
1857         unlock_page(page);
1858         page_cache_release(page);
1859         kfree(fixup);
1860 }
1861
1862 /*
1863  * There are a few paths in the higher layers of the kernel that directly
1864  * set the page dirty bit without asking the filesystem if it is a
1865  * good idea.  This causes problems because we want to make sure COW
1866  * properly happens and the data=ordered rules are followed.
1867  *
1868  * In our case any range that doesn't have the ORDERED bit set
1869  * hasn't been properly setup for IO.  We kick off an async process
1870  * to fix it up.  The async helper will wait for ordered extents, set
1871  * the delalloc bit and make it safe to write the page.
1872  */
1873 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1874 {
1875         struct inode *inode = page->mapping->host;
1876         struct btrfs_writepage_fixup *fixup;
1877         struct btrfs_root *root = BTRFS_I(inode)->root;
1878
1879         /* this page is properly in the ordered list */
1880         if (TestClearPagePrivate2(page))
1881                 return 0;
1882
1883         if (PageChecked(page))
1884                 return -EAGAIN;
1885
1886         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1887         if (!fixup)
1888                 return -EAGAIN;
1889
1890         SetPageChecked(page);
1891         page_cache_get(page);
1892         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
1893                         btrfs_writepage_fixup_worker, NULL, NULL);
1894         fixup->page = page;
1895         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
1896         return -EBUSY;
1897 }
1898
1899 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1900                                        struct inode *inode, u64 file_pos,
1901                                        u64 disk_bytenr, u64 disk_num_bytes,
1902                                        u64 num_bytes, u64 ram_bytes,
1903                                        u8 compression, u8 encryption,
1904                                        u16 other_encoding, int extent_type)
1905 {
1906         struct btrfs_root *root = BTRFS_I(inode)->root;
1907         struct btrfs_file_extent_item *fi;
1908         struct btrfs_path *path;
1909         struct extent_buffer *leaf;
1910         struct btrfs_key ins;
1911         int extent_inserted = 0;
1912         int ret;
1913
1914         path = btrfs_alloc_path();
1915         if (!path)
1916                 return -ENOMEM;
1917
1918         /*
1919          * we may be replacing one extent in the tree with another.
1920          * The new extent is pinned in the extent map, and we don't want
1921          * to drop it from the cache until it is completely in the btree.
1922          *
1923          * So, tell btrfs_drop_extents to leave this extent in the cache.
1924          * the caller is expected to unpin it and allow it to be merged
1925          * with the others.
1926          */
1927         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1928                                    file_pos + num_bytes, NULL, 0,
1929                                    1, sizeof(*fi), &extent_inserted);
1930         if (ret)
1931                 goto out;
1932
1933         if (!extent_inserted) {
1934                 ins.objectid = btrfs_ino(inode);
1935                 ins.offset = file_pos;
1936                 ins.type = BTRFS_EXTENT_DATA_KEY;
1937
1938                 path->leave_spinning = 1;
1939                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1940                                               sizeof(*fi));
1941                 if (ret)
1942                         goto out;
1943         }
1944         leaf = path->nodes[0];
1945         fi = btrfs_item_ptr(leaf, path->slots[0],
1946                             struct btrfs_file_extent_item);
1947         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1948         btrfs_set_file_extent_type(leaf, fi, extent_type);
1949         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1950         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1951         btrfs_set_file_extent_offset(leaf, fi, 0);
1952         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1953         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1954         btrfs_set_file_extent_compression(leaf, fi, compression);
1955         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1956         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1957
1958         btrfs_mark_buffer_dirty(leaf);
1959         btrfs_release_path(path);
1960
1961         inode_add_bytes(inode, num_bytes);
1962
1963         ins.objectid = disk_bytenr;
1964         ins.offset = disk_num_bytes;
1965         ins.type = BTRFS_EXTENT_ITEM_KEY;
1966         ret = btrfs_alloc_reserved_file_extent(trans, root,
1967                                         root->root_key.objectid,
1968                                         btrfs_ino(inode), file_pos, &ins);
1969 out:
1970         btrfs_free_path(path);
1971
1972         return ret;
1973 }
1974
1975 /* snapshot-aware defrag */
1976 struct sa_defrag_extent_backref {
1977         struct rb_node node;
1978         struct old_sa_defrag_extent *old;
1979         u64 root_id;
1980         u64 inum;
1981         u64 file_pos;
1982         u64 extent_offset;
1983         u64 num_bytes;
1984         u64 generation;
1985 };
1986
1987 struct old_sa_defrag_extent {
1988         struct list_head list;
1989         struct new_sa_defrag_extent *new;
1990
1991         u64 extent_offset;
1992         u64 bytenr;
1993         u64 offset;
1994         u64 len;
1995         int count;
1996 };
1997
1998 struct new_sa_defrag_extent {
1999         struct rb_root root;
2000         struct list_head head;
2001         struct btrfs_path *path;
2002         struct inode *inode;
2003         u64 file_pos;
2004         u64 len;
2005         u64 bytenr;
2006         u64 disk_len;
2007         u8 compress_type;
2008 };
2009
2010 static int backref_comp(struct sa_defrag_extent_backref *b1,
2011                         struct sa_defrag_extent_backref *b2)
2012 {
2013         if (b1->root_id < b2->root_id)
2014                 return -1;
2015         else if (b1->root_id > b2->root_id)
2016                 return 1;
2017
2018         if (b1->inum < b2->inum)
2019                 return -1;
2020         else if (b1->inum > b2->inum)
2021                 return 1;
2022
2023         if (b1->file_pos < b2->file_pos)
2024                 return -1;
2025         else if (b1->file_pos > b2->file_pos)
2026                 return 1;
2027
2028         /*
2029          * [------------------------------] ===> (a range of space)
2030          *     |<--->|   |<---->| =============> (fs/file tree A)
2031          * |<---------------------------->| ===> (fs/file tree B)
2032          *
2033          * A range of space can refer to two file extents in one tree while
2034          * refer to only one file extent in another tree.
2035          *
2036          * So we may process a disk offset more than one time(two extents in A)
2037          * and locate at the same extent(one extent in B), then insert two same
2038          * backrefs(both refer to the extent in B).
2039          */
2040         return 0;
2041 }
2042
2043 static void backref_insert(struct rb_root *root,
2044                            struct sa_defrag_extent_backref *backref)
2045 {
2046         struct rb_node **p = &root->rb_node;
2047         struct rb_node *parent = NULL;
2048         struct sa_defrag_extent_backref *entry;
2049         int ret;
2050
2051         while (*p) {
2052                 parent = *p;
2053                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2054
2055                 ret = backref_comp(backref, entry);
2056                 if (ret < 0)
2057                         p = &(*p)->rb_left;
2058                 else
2059                         p = &(*p)->rb_right;
2060         }
2061
2062         rb_link_node(&backref->node, parent, p);
2063         rb_insert_color(&backref->node, root);
2064 }
2065
2066 /*
2067  * Note the backref might has changed, and in this case we just return 0.
2068  */
2069 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2070                                        void *ctx)
2071 {
2072         struct btrfs_file_extent_item *extent;
2073         struct btrfs_fs_info *fs_info;
2074         struct old_sa_defrag_extent *old = ctx;
2075         struct new_sa_defrag_extent *new = old->new;
2076         struct btrfs_path *path = new->path;
2077         struct btrfs_key key;
2078         struct btrfs_root *root;
2079         struct sa_defrag_extent_backref *backref;
2080         struct extent_buffer *leaf;
2081         struct inode *inode = new->inode;
2082         int slot;
2083         int ret;
2084         u64 extent_offset;
2085         u64 num_bytes;
2086
2087         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2088             inum == btrfs_ino(inode))
2089                 return 0;
2090
2091         key.objectid = root_id;
2092         key.type = BTRFS_ROOT_ITEM_KEY;
2093         key.offset = (u64)-1;
2094
2095         fs_info = BTRFS_I(inode)->root->fs_info;
2096         root = btrfs_read_fs_root_no_name(fs_info, &key);
2097         if (IS_ERR(root)) {
2098                 if (PTR_ERR(root) == -ENOENT)
2099                         return 0;
2100                 WARN_ON(1);
2101                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2102                          inum, offset, root_id);
2103                 return PTR_ERR(root);
2104         }
2105
2106         key.objectid = inum;
2107         key.type = BTRFS_EXTENT_DATA_KEY;
2108         if (offset > (u64)-1 << 32)
2109                 key.offset = 0;
2110         else
2111                 key.offset = offset;
2112
2113         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2114         if (WARN_ON(ret < 0))
2115                 return ret;
2116         ret = 0;
2117
2118         while (1) {
2119                 cond_resched();
2120
2121                 leaf = path->nodes[0];
2122                 slot = path->slots[0];
2123
2124                 if (slot >= btrfs_header_nritems(leaf)) {
2125                         ret = btrfs_next_leaf(root, path);
2126                         if (ret < 0) {
2127                                 goto out;
2128                         } else if (ret > 0) {
2129                                 ret = 0;
2130                                 goto out;
2131                         }
2132                         continue;
2133                 }
2134
2135                 path->slots[0]++;
2136
2137                 btrfs_item_key_to_cpu(leaf, &key, slot);
2138
2139                 if (key.objectid > inum)
2140                         goto out;
2141
2142                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2143                         continue;
2144
2145                 extent = btrfs_item_ptr(leaf, slot,
2146                                         struct btrfs_file_extent_item);
2147
2148                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2149                         continue;
2150
2151                 /*
2152                  * 'offset' refers to the exact key.offset,
2153                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2154                  * (key.offset - extent_offset).
2155                  */
2156                 if (key.offset != offset)
2157                         continue;
2158
2159                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2160                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2161
2162                 if (extent_offset >= old->extent_offset + old->offset +
2163                     old->len || extent_offset + num_bytes <=
2164                     old->extent_offset + old->offset)
2165                         continue;
2166                 break;
2167         }
2168
2169         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2170         if (!backref) {
2171                 ret = -ENOENT;
2172                 goto out;
2173         }
2174
2175         backref->root_id = root_id;
2176         backref->inum = inum;
2177         backref->file_pos = offset;
2178         backref->num_bytes = num_bytes;
2179         backref->extent_offset = extent_offset;
2180         backref->generation = btrfs_file_extent_generation(leaf, extent);
2181         backref->old = old;
2182         backref_insert(&new->root, backref);
2183         old->count++;
2184 out:
2185         btrfs_release_path(path);
2186         WARN_ON(ret);
2187         return ret;
2188 }
2189
2190 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2191                                    struct new_sa_defrag_extent *new)
2192 {
2193         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2194         struct old_sa_defrag_extent *old, *tmp;
2195         int ret;
2196
2197         new->path = path;
2198
2199         list_for_each_entry_safe(old, tmp, &new->head, list) {
2200                 ret = iterate_inodes_from_logical(old->bytenr +
2201                                                   old->extent_offset, fs_info,
2202                                                   path, record_one_backref,
2203                                                   old);
2204                 if (ret < 0 && ret != -ENOENT)
2205                         return false;
2206
2207                 /* no backref to be processed for this extent */
2208                 if (!old->count) {
2209                         list_del(&old->list);
2210                         kfree(old);
2211                 }
2212         }
2213
2214         if (list_empty(&new->head))
2215                 return false;
2216
2217         return true;
2218 }
2219
2220 static int relink_is_mergable(struct extent_buffer *leaf,
2221                               struct btrfs_file_extent_item *fi,
2222                               struct new_sa_defrag_extent *new)
2223 {
2224         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2225                 return 0;
2226
2227         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2228                 return 0;
2229
2230         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2231                 return 0;
2232
2233         if (btrfs_file_extent_encryption(leaf, fi) ||
2234             btrfs_file_extent_other_encoding(leaf, fi))
2235                 return 0;
2236
2237         return 1;
2238 }
2239
2240 /*
2241  * Note the backref might has changed, and in this case we just return 0.
2242  */
2243 static noinline int relink_extent_backref(struct btrfs_path *path,
2244                                  struct sa_defrag_extent_backref *prev,
2245                                  struct sa_defrag_extent_backref *backref)
2246 {
2247         struct btrfs_file_extent_item *extent;
2248         struct btrfs_file_extent_item *item;
2249         struct btrfs_ordered_extent *ordered;
2250         struct btrfs_trans_handle *trans;
2251         struct btrfs_fs_info *fs_info;
2252         struct btrfs_root *root;
2253         struct btrfs_key key;
2254         struct extent_buffer *leaf;
2255         struct old_sa_defrag_extent *old = backref->old;
2256         struct new_sa_defrag_extent *new = old->new;
2257         struct inode *src_inode = new->inode;
2258         struct inode *inode;
2259         struct extent_state *cached = NULL;
2260         int ret = 0;
2261         u64 start;
2262         u64 len;
2263         u64 lock_start;
2264         u64 lock_end;
2265         bool merge = false;
2266         int index;
2267
2268         if (prev && prev->root_id == backref->root_id &&
2269             prev->inum == backref->inum &&
2270             prev->file_pos + prev->num_bytes == backref->file_pos)
2271                 merge = true;
2272
2273         /* step 1: get root */
2274         key.objectid = backref->root_id;
2275         key.type = BTRFS_ROOT_ITEM_KEY;
2276         key.offset = (u64)-1;
2277
2278         fs_info = BTRFS_I(src_inode)->root->fs_info;
2279         index = srcu_read_lock(&fs_info->subvol_srcu);
2280
2281         root = btrfs_read_fs_root_no_name(fs_info, &key);
2282         if (IS_ERR(root)) {
2283                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2284                 if (PTR_ERR(root) == -ENOENT)
2285                         return 0;
2286                 return PTR_ERR(root);
2287         }
2288
2289         if (btrfs_root_readonly(root)) {
2290                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2291                 return 0;
2292         }
2293
2294         /* step 2: get inode */
2295         key.objectid = backref->inum;
2296         key.type = BTRFS_INODE_ITEM_KEY;
2297         key.offset = 0;
2298
2299         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2300         if (IS_ERR(inode)) {
2301                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2302                 return 0;
2303         }
2304
2305         srcu_read_unlock(&fs_info->subvol_srcu, index);
2306
2307         /* step 3: relink backref */
2308         lock_start = backref->file_pos;
2309         lock_end = backref->file_pos + backref->num_bytes - 1;
2310         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2311                          0, &cached);
2312
2313         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2314         if (ordered) {
2315                 btrfs_put_ordered_extent(ordered);
2316                 goto out_unlock;
2317         }
2318
2319         trans = btrfs_join_transaction(root);
2320         if (IS_ERR(trans)) {
2321                 ret = PTR_ERR(trans);
2322                 goto out_unlock;
2323         }
2324
2325         key.objectid = backref->inum;
2326         key.type = BTRFS_EXTENT_DATA_KEY;
2327         key.offset = backref->file_pos;
2328
2329         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2330         if (ret < 0) {
2331                 goto out_free_path;
2332         } else if (ret > 0) {
2333                 ret = 0;
2334                 goto out_free_path;
2335         }
2336
2337         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2338                                 struct btrfs_file_extent_item);
2339
2340         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2341             backref->generation)
2342                 goto out_free_path;
2343
2344         btrfs_release_path(path);
2345
2346         start = backref->file_pos;
2347         if (backref->extent_offset < old->extent_offset + old->offset)
2348                 start += old->extent_offset + old->offset -
2349                          backref->extent_offset;
2350
2351         len = min(backref->extent_offset + backref->num_bytes,
2352                   old->extent_offset + old->offset + old->len);
2353         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2354
2355         ret = btrfs_drop_extents(trans, root, inode, start,
2356                                  start + len, 1);
2357         if (ret)
2358                 goto out_free_path;
2359 again:
2360         key.objectid = btrfs_ino(inode);
2361         key.type = BTRFS_EXTENT_DATA_KEY;
2362         key.offset = start;
2363
2364         path->leave_spinning = 1;
2365         if (merge) {
2366                 struct btrfs_file_extent_item *fi;
2367                 u64 extent_len;
2368                 struct btrfs_key found_key;
2369
2370                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2371                 if (ret < 0)
2372                         goto out_free_path;
2373
2374                 path->slots[0]--;
2375                 leaf = path->nodes[0];
2376                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2377
2378                 fi = btrfs_item_ptr(leaf, path->slots[0],
2379                                     struct btrfs_file_extent_item);
2380                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2381
2382                 if (extent_len + found_key.offset == start &&
2383                     relink_is_mergable(leaf, fi, new)) {
2384                         btrfs_set_file_extent_num_bytes(leaf, fi,
2385                                                         extent_len + len);
2386                         btrfs_mark_buffer_dirty(leaf);
2387                         inode_add_bytes(inode, len);
2388
2389                         ret = 1;
2390                         goto out_free_path;
2391                 } else {
2392                         merge = false;
2393                         btrfs_release_path(path);
2394                         goto again;
2395                 }
2396         }
2397
2398         ret = btrfs_insert_empty_item(trans, root, path, &key,
2399                                         sizeof(*extent));
2400         if (ret) {
2401                 btrfs_abort_transaction(trans, root, ret);
2402                 goto out_free_path;
2403         }
2404
2405         leaf = path->nodes[0];
2406         item = btrfs_item_ptr(leaf, path->slots[0],
2407                                 struct btrfs_file_extent_item);
2408         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2409         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2410         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2411         btrfs_set_file_extent_num_bytes(leaf, item, len);
2412         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2413         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2414         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2415         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2416         btrfs_set_file_extent_encryption(leaf, item, 0);
2417         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2418
2419         btrfs_mark_buffer_dirty(leaf);
2420         inode_add_bytes(inode, len);
2421         btrfs_release_path(path);
2422
2423         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2424                         new->disk_len, 0,
2425                         backref->root_id, backref->inum,
2426                         new->file_pos, 0);      /* start - extent_offset */
2427         if (ret) {
2428                 btrfs_abort_transaction(trans, root, ret);
2429                 goto out_free_path;
2430         }
2431
2432         ret = 1;
2433 out_free_path:
2434         btrfs_release_path(path);
2435         path->leave_spinning = 0;
2436         btrfs_end_transaction(trans, root);
2437 out_unlock:
2438         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2439                              &cached, GFP_NOFS);
2440         iput(inode);
2441         return ret;
2442 }
2443
2444 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2445 {
2446         struct old_sa_defrag_extent *old, *tmp;
2447
2448         if (!new)
2449                 return;
2450
2451         list_for_each_entry_safe(old, tmp, &new->head, list) {
2452                 list_del(&old->list);
2453                 kfree(old);
2454         }
2455         kfree(new);
2456 }
2457
2458 static void relink_file_extents(struct new_sa_defrag_extent *new)
2459 {
2460         struct btrfs_path *path;
2461         struct sa_defrag_extent_backref *backref;
2462         struct sa_defrag_extent_backref *prev = NULL;
2463         struct inode *inode;
2464         struct btrfs_root *root;
2465         struct rb_node *node;
2466         int ret;
2467
2468         inode = new->inode;
2469         root = BTRFS_I(inode)->root;
2470
2471         path = btrfs_alloc_path();
2472         if (!path)
2473                 return;
2474
2475         if (!record_extent_backrefs(path, new)) {
2476                 btrfs_free_path(path);
2477                 goto out;
2478         }
2479         btrfs_release_path(path);
2480
2481         while (1) {
2482                 node = rb_first(&new->root);
2483                 if (!node)
2484                         break;
2485                 rb_erase(node, &new->root);
2486
2487                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2488
2489                 ret = relink_extent_backref(path, prev, backref);
2490                 WARN_ON(ret < 0);
2491
2492                 kfree(prev);
2493
2494                 if (ret == 1)
2495                         prev = backref;
2496                 else
2497                         prev = NULL;
2498                 cond_resched();
2499         }
2500         kfree(prev);
2501
2502         btrfs_free_path(path);
2503 out:
2504         free_sa_defrag_extent(new);
2505
2506         atomic_dec(&root->fs_info->defrag_running);
2507         wake_up(&root->fs_info->transaction_wait);
2508 }
2509
2510 static struct new_sa_defrag_extent *
2511 record_old_file_extents(struct inode *inode,
2512                         struct btrfs_ordered_extent *ordered)
2513 {
2514         struct btrfs_root *root = BTRFS_I(inode)->root;
2515         struct btrfs_path *path;
2516         struct btrfs_key key;
2517         struct old_sa_defrag_extent *old;
2518         struct new_sa_defrag_extent *new;
2519         int ret;
2520
2521         new = kmalloc(sizeof(*new), GFP_NOFS);
2522         if (!new)
2523                 return NULL;
2524
2525         new->inode = inode;
2526         new->file_pos = ordered->file_offset;
2527         new->len = ordered->len;
2528         new->bytenr = ordered->start;
2529         new->disk_len = ordered->disk_len;
2530         new->compress_type = ordered->compress_type;
2531         new->root = RB_ROOT;
2532         INIT_LIST_HEAD(&new->head);
2533
2534         path = btrfs_alloc_path();
2535         if (!path)
2536                 goto out_kfree;
2537
2538         key.objectid = btrfs_ino(inode);
2539         key.type = BTRFS_EXTENT_DATA_KEY;
2540         key.offset = new->file_pos;
2541
2542         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2543         if (ret < 0)
2544                 goto out_free_path;
2545         if (ret > 0 && path->slots[0] > 0)
2546                 path->slots[0]--;
2547
2548         /* find out all the old extents for the file range */
2549         while (1) {
2550                 struct btrfs_file_extent_item *extent;
2551                 struct extent_buffer *l;
2552                 int slot;
2553                 u64 num_bytes;
2554                 u64 offset;
2555                 u64 end;
2556                 u64 disk_bytenr;
2557                 u64 extent_offset;
2558
2559                 l = path->nodes[0];
2560                 slot = path->slots[0];
2561
2562                 if (slot >= btrfs_header_nritems(l)) {
2563                         ret = btrfs_next_leaf(root, path);
2564                         if (ret < 0)
2565                                 goto out_free_path;
2566                         else if (ret > 0)
2567                                 break;
2568                         continue;
2569                 }
2570
2571                 btrfs_item_key_to_cpu(l, &key, slot);
2572
2573                 if (key.objectid != btrfs_ino(inode))
2574                         break;
2575                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2576                         break;
2577                 if (key.offset >= new->file_pos + new->len)
2578                         break;
2579
2580                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2581
2582                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2583                 if (key.offset + num_bytes < new->file_pos)
2584                         goto next;
2585
2586                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2587                 if (!disk_bytenr)
2588                         goto next;
2589
2590                 extent_offset = btrfs_file_extent_offset(l, extent);
2591
2592                 old = kmalloc(sizeof(*old), GFP_NOFS);
2593                 if (!old)
2594                         goto out_free_path;
2595
2596                 offset = max(new->file_pos, key.offset);
2597                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2598
2599                 old->bytenr = disk_bytenr;
2600                 old->extent_offset = extent_offset;
2601                 old->offset = offset - key.offset;
2602                 old->len = end - offset;
2603                 old->new = new;
2604                 old->count = 0;
2605                 list_add_tail(&old->list, &new->head);
2606 next:
2607                 path->slots[0]++;
2608                 cond_resched();
2609         }
2610
2611         btrfs_free_path(path);
2612         atomic_inc(&root->fs_info->defrag_running);
2613
2614         return new;
2615
2616 out_free_path:
2617         btrfs_free_path(path);
2618 out_kfree:
2619         free_sa_defrag_extent(new);
2620         return NULL;
2621 }
2622
2623 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2624                                          u64 start, u64 len)
2625 {
2626         struct btrfs_block_group_cache *cache;
2627
2628         cache = btrfs_lookup_block_group(root->fs_info, start);
2629         ASSERT(cache);
2630
2631         spin_lock(&cache->lock);
2632         cache->delalloc_bytes -= len;
2633         spin_unlock(&cache->lock);
2634
2635         btrfs_put_block_group(cache);
2636 }
2637
2638 /* as ordered data IO finishes, this gets called so we can finish
2639  * an ordered extent if the range of bytes in the file it covers are
2640  * fully written.
2641  */
2642 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2643 {
2644         struct inode *inode = ordered_extent->inode;
2645         struct btrfs_root *root = BTRFS_I(inode)->root;
2646         struct btrfs_trans_handle *trans = NULL;
2647         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2648         struct extent_state *cached_state = NULL;
2649         struct new_sa_defrag_extent *new = NULL;
2650         int compress_type = 0;
2651         int ret = 0;
2652         u64 logical_len = ordered_extent->len;
2653         bool nolock;
2654         bool truncated = false;
2655
2656         nolock = btrfs_is_free_space_inode(inode);
2657
2658         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2659                 ret = -EIO;
2660                 goto out;
2661         }
2662
2663         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2664                 truncated = true;
2665                 logical_len = ordered_extent->truncated_len;
2666                 /* Truncated the entire extent, don't bother adding */
2667                 if (!logical_len)
2668                         goto out;
2669         }
2670
2671         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2672                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2673                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2674                 if (nolock)
2675                         trans = btrfs_join_transaction_nolock(root);
2676                 else
2677                         trans = btrfs_join_transaction(root);
2678                 if (IS_ERR(trans)) {
2679                         ret = PTR_ERR(trans);
2680                         trans = NULL;
2681                         goto out;
2682                 }
2683                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2684                 ret = btrfs_update_inode_fallback(trans, root, inode);
2685                 if (ret) /* -ENOMEM or corruption */
2686                         btrfs_abort_transaction(trans, root, ret);
2687                 goto out;
2688         }
2689
2690         lock_extent_bits(io_tree, ordered_extent->file_offset,
2691                          ordered_extent->file_offset + ordered_extent->len - 1,
2692                          0, &cached_state);
2693
2694         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2695                         ordered_extent->file_offset + ordered_extent->len - 1,
2696                         EXTENT_DEFRAG, 1, cached_state);
2697         if (ret) {
2698                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2699                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2700                         /* the inode is shared */
2701                         new = record_old_file_extents(inode, ordered_extent);
2702
2703                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2704                         ordered_extent->file_offset + ordered_extent->len - 1,
2705                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2706         }
2707
2708         if (nolock)
2709                 trans = btrfs_join_transaction_nolock(root);
2710         else
2711                 trans = btrfs_join_transaction(root);
2712         if (IS_ERR(trans)) {
2713                 ret = PTR_ERR(trans);
2714                 trans = NULL;
2715                 goto out_unlock;
2716         }
2717
2718         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2719
2720         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2721                 compress_type = ordered_extent->compress_type;
2722         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2723                 BUG_ON(compress_type);
2724                 ret = btrfs_mark_extent_written(trans, inode,
2725                                                 ordered_extent->file_offset,
2726                                                 ordered_extent->file_offset +
2727                                                 logical_len);
2728         } else {
2729                 BUG_ON(root == root->fs_info->tree_root);
2730                 ret = insert_reserved_file_extent(trans, inode,
2731                                                 ordered_extent->file_offset,
2732                                                 ordered_extent->start,
2733                                                 ordered_extent->disk_len,
2734                                                 logical_len, logical_len,
2735                                                 compress_type, 0, 0,
2736                                                 BTRFS_FILE_EXTENT_REG);
2737                 if (!ret)
2738                         btrfs_release_delalloc_bytes(root,
2739                                                      ordered_extent->start,
2740                                                      ordered_extent->disk_len);
2741         }
2742         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2743                            ordered_extent->file_offset, ordered_extent->len,
2744                            trans->transid);
2745         if (ret < 0) {
2746                 btrfs_abort_transaction(trans, root, ret);
2747                 goto out_unlock;
2748         }
2749
2750         add_pending_csums(trans, inode, ordered_extent->file_offset,
2751                           &ordered_extent->list);
2752
2753         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2754         ret = btrfs_update_inode_fallback(trans, root, inode);
2755         if (ret) { /* -ENOMEM or corruption */
2756                 btrfs_abort_transaction(trans, root, ret);
2757                 goto out_unlock;
2758         }
2759         ret = 0;
2760 out_unlock:
2761         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2762                              ordered_extent->file_offset +
2763                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2764 out:
2765         if (root != root->fs_info->tree_root)
2766                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2767         if (trans)
2768                 btrfs_end_transaction(trans, root);
2769
2770         if (ret || truncated) {
2771                 u64 start, end;
2772
2773                 if (truncated)
2774                         start = ordered_extent->file_offset + logical_len;
2775                 else
2776                         start = ordered_extent->file_offset;
2777                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2778                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2779
2780                 /* Drop the cache for the part of the extent we didn't write. */
2781                 btrfs_drop_extent_cache(inode, start, end, 0);
2782
2783                 /*
2784                  * If the ordered extent had an IOERR or something else went
2785                  * wrong we need to return the space for this ordered extent
2786                  * back to the allocator.  We only free the extent in the
2787                  * truncated case if we didn't write out the extent at all.
2788                  */
2789                 if ((ret || !logical_len) &&
2790                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2791                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2792                         btrfs_free_reserved_extent(root, ordered_extent->start,
2793                                                    ordered_extent->disk_len, 1);
2794         }
2795
2796
2797         /*
2798          * This needs to be done to make sure anybody waiting knows we are done
2799          * updating everything for this ordered extent.
2800          */
2801         btrfs_remove_ordered_extent(inode, ordered_extent);
2802
2803         /* for snapshot-aware defrag */
2804         if (new) {
2805                 if (ret) {
2806                         free_sa_defrag_extent(new);
2807                         atomic_dec(&root->fs_info->defrag_running);
2808                 } else {
2809                         relink_file_extents(new);
2810                 }
2811         }
2812
2813         /* once for us */
2814         btrfs_put_ordered_extent(ordered_extent);
2815         /* once for the tree */
2816         btrfs_put_ordered_extent(ordered_extent);
2817
2818         return ret;
2819 }
2820
2821 static void finish_ordered_fn(struct btrfs_work *work)
2822 {
2823         struct btrfs_ordered_extent *ordered_extent;
2824         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2825         btrfs_finish_ordered_io(ordered_extent);
2826 }
2827
2828 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2829                                 struct extent_state *state, int uptodate)
2830 {
2831         struct inode *inode = page->mapping->host;
2832         struct btrfs_root *root = BTRFS_I(inode)->root;
2833         struct btrfs_ordered_extent *ordered_extent = NULL;
2834         struct btrfs_workqueue *wq;
2835         btrfs_work_func_t func;
2836
2837         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2838
2839         ClearPagePrivate2(page);
2840         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2841                                             end - start + 1, uptodate))
2842                 return 0;
2843
2844         if (btrfs_is_free_space_inode(inode)) {
2845                 wq = root->fs_info->endio_freespace_worker;
2846                 func = btrfs_freespace_write_helper;
2847         } else {
2848                 wq = root->fs_info->endio_write_workers;
2849                 func = btrfs_endio_write_helper;
2850         }
2851
2852         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2853                         NULL);
2854         btrfs_queue_work(wq, &ordered_extent->work);
2855
2856         return 0;
2857 }
2858
2859 /*
2860  * when reads are done, we need to check csums to verify the data is correct
2861  * if there's a match, we allow the bio to finish.  If not, the code in
2862  * extent_io.c will try to find good copies for us.
2863  */
2864 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2865                                       u64 phy_offset, struct page *page,
2866                                       u64 start, u64 end, int mirror)
2867 {
2868         size_t offset = start - page_offset(page);
2869         struct inode *inode = page->mapping->host;
2870         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2871         char *kaddr;
2872         struct btrfs_root *root = BTRFS_I(inode)->root;
2873         u32 csum_expected;
2874         u32 csum = ~(u32)0;
2875         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2876                                       DEFAULT_RATELIMIT_BURST);
2877
2878         if (PageChecked(page)) {
2879                 ClearPageChecked(page);
2880                 goto good;
2881         }
2882
2883         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2884                 goto good;
2885
2886         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2887             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2888                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2889                                   GFP_NOFS);
2890                 return 0;
2891         }
2892
2893         phy_offset >>= inode->i_sb->s_blocksize_bits;
2894         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2895
2896         kaddr = kmap_atomic(page);
2897         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2898         btrfs_csum_final(csum, (char *)&csum);
2899         if (csum != csum_expected)
2900                 goto zeroit;
2901
2902         kunmap_atomic(kaddr);
2903 good:
2904         return 0;
2905
2906 zeroit:
2907         if (__ratelimit(&_rs))
2908                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2909                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2910         memset(kaddr + offset, 1, end - start + 1);
2911         flush_dcache_page(page);
2912         kunmap_atomic(kaddr);
2913         if (csum_expected == 0)
2914                 return 0;
2915         return -EIO;
2916 }
2917
2918 struct delayed_iput {
2919         struct list_head list;
2920         struct inode *inode;
2921 };
2922
2923 /* JDM: If this is fs-wide, why can't we add a pointer to
2924  * btrfs_inode instead and avoid the allocation? */
2925 void btrfs_add_delayed_iput(struct inode *inode)
2926 {
2927         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2928         struct delayed_iput *delayed;
2929
2930         if (atomic_add_unless(&inode->i_count, -1, 1))
2931                 return;
2932
2933         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2934         delayed->inode = inode;
2935
2936         spin_lock(&fs_info->delayed_iput_lock);
2937         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2938         spin_unlock(&fs_info->delayed_iput_lock);
2939 }
2940
2941 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2942 {
2943         LIST_HEAD(list);
2944         struct btrfs_fs_info *fs_info = root->fs_info;
2945         struct delayed_iput *delayed;
2946         int empty;
2947
2948         spin_lock(&fs_info->delayed_iput_lock);
2949         empty = list_empty(&fs_info->delayed_iputs);
2950         spin_unlock(&fs_info->delayed_iput_lock);
2951         if (empty)
2952                 return;
2953
2954         spin_lock(&fs_info->delayed_iput_lock);
2955         list_splice_init(&fs_info->delayed_iputs, &list);
2956         spin_unlock(&fs_info->delayed_iput_lock);
2957
2958         while (!list_empty(&list)) {
2959                 delayed = list_entry(list.next, struct delayed_iput, list);
2960                 list_del(&delayed->list);
2961                 iput(delayed->inode);
2962                 kfree(delayed);
2963         }
2964 }
2965
2966 /*
2967  * This is called in transaction commit time. If there are no orphan
2968  * files in the subvolume, it removes orphan item and frees block_rsv
2969  * structure.
2970  */
2971 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2972                               struct btrfs_root *root)
2973 {
2974         struct btrfs_block_rsv *block_rsv;
2975         int ret;
2976
2977         if (atomic_read(&root->orphan_inodes) ||
2978             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2979                 return;
2980
2981         spin_lock(&root->orphan_lock);
2982         if (atomic_read(&root->orphan_inodes)) {
2983                 spin_unlock(&root->orphan_lock);
2984                 return;
2985         }
2986
2987         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2988                 spin_unlock(&root->orphan_lock);
2989                 return;
2990         }
2991
2992         block_rsv = root->orphan_block_rsv;
2993         root->orphan_block_rsv = NULL;
2994         spin_unlock(&root->orphan_lock);
2995
2996         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
2997             btrfs_root_refs(&root->root_item) > 0) {
2998                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2999                                             root->root_key.objectid);
3000                 if (ret)
3001                         btrfs_abort_transaction(trans, root, ret);
3002                 else
3003                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3004                                   &root->state);
3005         }
3006
3007         if (block_rsv) {
3008                 WARN_ON(block_rsv->size > 0);
3009                 btrfs_free_block_rsv(root, block_rsv);
3010         }
3011 }
3012
3013 /*
3014  * This creates an orphan entry for the given inode in case something goes
3015  * wrong in the middle of an unlink/truncate.
3016  *
3017  * NOTE: caller of this function should reserve 5 units of metadata for
3018  *       this function.
3019  */
3020 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3021 {
3022         struct btrfs_root *root = BTRFS_I(inode)->root;
3023         struct btrfs_block_rsv *block_rsv = NULL;
3024         int reserve = 0;
3025         int insert = 0;
3026         int ret;
3027
3028         if (!root->orphan_block_rsv) {
3029                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3030                 if (!block_rsv)
3031                         return -ENOMEM;
3032         }
3033
3034         spin_lock(&root->orphan_lock);
3035         if (!root->orphan_block_rsv) {
3036                 root->orphan_block_rsv = block_rsv;
3037         } else if (block_rsv) {
3038                 btrfs_free_block_rsv(root, block_rsv);
3039                 block_rsv = NULL;
3040         }
3041
3042         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3043                               &BTRFS_I(inode)->runtime_flags)) {
3044 #if 0
3045                 /*
3046                  * For proper ENOSPC handling, we should do orphan
3047                  * cleanup when mounting. But this introduces backward
3048                  * compatibility issue.
3049                  */
3050                 if (!xchg(&root->orphan_item_inserted, 1))
3051                         insert = 2;
3052                 else
3053                         insert = 1;
3054 #endif
3055                 insert = 1;
3056                 atomic_inc(&root->orphan_inodes);
3057         }
3058
3059         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3060                               &BTRFS_I(inode)->runtime_flags))
3061                 reserve = 1;
3062         spin_unlock(&root->orphan_lock);
3063
3064         /* grab metadata reservation from transaction handle */
3065         if (reserve) {
3066                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3067                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3068         }
3069
3070         /* insert an orphan item to track this unlinked/truncated file */
3071         if (insert >= 1) {
3072                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3073                 if (ret) {
3074                         atomic_dec(&root->orphan_inodes);
3075                         if (reserve) {
3076                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3077                                           &BTRFS_I(inode)->runtime_flags);
3078                                 btrfs_orphan_release_metadata(inode);
3079                         }
3080                         if (ret != -EEXIST) {
3081                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3082                                           &BTRFS_I(inode)->runtime_flags);
3083                                 btrfs_abort_transaction(trans, root, ret);
3084                                 return ret;
3085                         }
3086                 }
3087                 ret = 0;
3088         }
3089
3090         /* insert an orphan item to track subvolume contains orphan files */
3091         if (insert >= 2) {
3092                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3093                                                root->root_key.objectid);
3094                 if (ret && ret != -EEXIST) {
3095                         btrfs_abort_transaction(trans, root, ret);
3096                         return ret;
3097                 }
3098         }
3099         return 0;
3100 }
3101
3102 /*
3103  * We have done the truncate/delete so we can go ahead and remove the orphan
3104  * item for this particular inode.
3105  */
3106 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3107                             struct inode *inode)
3108 {
3109         struct btrfs_root *root = BTRFS_I(inode)->root;
3110         int delete_item = 0;
3111         int release_rsv = 0;
3112         int ret = 0;
3113
3114         spin_lock(&root->orphan_lock);
3115         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3116                                &BTRFS_I(inode)->runtime_flags))
3117                 delete_item = 1;
3118
3119         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3120                                &BTRFS_I(inode)->runtime_flags))
3121                 release_rsv = 1;
3122         spin_unlock(&root->orphan_lock);
3123
3124         if (delete_item) {
3125                 atomic_dec(&root->orphan_inodes);
3126                 if (trans)
3127                         ret = btrfs_del_orphan_item(trans, root,
3128                                                     btrfs_ino(inode));
3129         }
3130
3131         if (release_rsv)
3132                 btrfs_orphan_release_metadata(inode);
3133
3134         return ret;
3135 }
3136
3137 /*
3138  * this cleans up any orphans that may be left on the list from the last use
3139  * of this root.
3140  */
3141 int btrfs_orphan_cleanup(struct btrfs_root *root)
3142 {
3143         struct btrfs_path *path;
3144         struct extent_buffer *leaf;
3145         struct btrfs_key key, found_key;
3146         struct btrfs_trans_handle *trans;
3147         struct inode *inode;
3148         u64 last_objectid = 0;
3149         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3150
3151         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3152                 return 0;
3153
3154         path = btrfs_alloc_path();
3155         if (!path) {
3156                 ret = -ENOMEM;
3157                 goto out;
3158         }
3159         path->reada = -1;
3160
3161         key.objectid = BTRFS_ORPHAN_OBJECTID;
3162         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3163         key.offset = (u64)-1;
3164
3165         while (1) {
3166                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3167                 if (ret < 0)
3168                         goto out;
3169
3170                 /*
3171                  * if ret == 0 means we found what we were searching for, which
3172                  * is weird, but possible, so only screw with path if we didn't
3173                  * find the key and see if we have stuff that matches
3174                  */
3175                 if (ret > 0) {
3176                         ret = 0;
3177                         if (path->slots[0] == 0)
3178                                 break;
3179                         path->slots[0]--;
3180                 }
3181
3182                 /* pull out the item */
3183                 leaf = path->nodes[0];
3184                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3185
3186                 /* make sure the item matches what we want */
3187                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3188                         break;
3189                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3190                         break;
3191
3192                 /* release the path since we're done with it */
3193                 btrfs_release_path(path);
3194
3195                 /*
3196                  * this is where we are basically btrfs_lookup, without the
3197                  * crossing root thing.  we store the inode number in the
3198                  * offset of the orphan item.
3199                  */
3200
3201                 if (found_key.offset == last_objectid) {
3202                         btrfs_err(root->fs_info,
3203                                 "Error removing orphan entry, stopping orphan cleanup");
3204                         ret = -EINVAL;
3205                         goto out;
3206                 }
3207
3208                 last_objectid = found_key.offset;
3209
3210                 found_key.objectid = found_key.offset;
3211                 found_key.type = BTRFS_INODE_ITEM_KEY;
3212                 found_key.offset = 0;
3213                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3214                 ret = PTR_ERR_OR_ZERO(inode);
3215                 if (ret && ret != -ESTALE)
3216                         goto out;
3217
3218                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3219                         struct btrfs_root *dead_root;
3220                         struct btrfs_fs_info *fs_info = root->fs_info;
3221                         int is_dead_root = 0;
3222
3223                         /*
3224                          * this is an orphan in the tree root. Currently these
3225                          * could come from 2 sources:
3226                          *  a) a snapshot deletion in progress
3227                          *  b) a free space cache inode
3228                          * We need to distinguish those two, as the snapshot
3229                          * orphan must not get deleted.
3230                          * find_dead_roots already ran before us, so if this
3231                          * is a snapshot deletion, we should find the root
3232                          * in the dead_roots list
3233                          */
3234                         spin_lock(&fs_info->trans_lock);
3235                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3236                                             root_list) {
3237                                 if (dead_root->root_key.objectid ==
3238                                     found_key.objectid) {
3239                                         is_dead_root = 1;
3240                                         break;
3241                                 }
3242                         }
3243                         spin_unlock(&fs_info->trans_lock);
3244                         if (is_dead_root) {
3245                                 /* prevent this orphan from being found again */
3246                                 key.offset = found_key.objectid - 1;
3247                                 continue;
3248                         }
3249                 }
3250                 /*
3251                  * Inode is already gone but the orphan item is still there,
3252                  * kill the orphan item.
3253                  */
3254                 if (ret == -ESTALE) {
3255                         trans = btrfs_start_transaction(root, 1);
3256                         if (IS_ERR(trans)) {
3257                                 ret = PTR_ERR(trans);
3258                                 goto out;
3259                         }
3260                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3261                                 found_key.objectid);
3262                         ret = btrfs_del_orphan_item(trans, root,
3263                                                     found_key.objectid);
3264                         btrfs_end_transaction(trans, root);
3265                         if (ret)
3266                                 goto out;
3267                         continue;
3268                 }
3269
3270                 /*
3271                  * add this inode to the orphan list so btrfs_orphan_del does
3272                  * the proper thing when we hit it
3273                  */
3274                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3275                         &BTRFS_I(inode)->runtime_flags);
3276                 atomic_inc(&root->orphan_inodes);
3277
3278                 /* if we have links, this was a truncate, lets do that */
3279                 if (inode->i_nlink) {
3280                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3281                                 iput(inode);
3282                                 continue;
3283                         }
3284                         nr_truncate++;
3285
3286                         /* 1 for the orphan item deletion. */
3287                         trans = btrfs_start_transaction(root, 1);
3288                         if (IS_ERR(trans)) {
3289                                 iput(inode);
3290                                 ret = PTR_ERR(trans);
3291                                 goto out;
3292                         }
3293                         ret = btrfs_orphan_add(trans, inode);
3294                         btrfs_end_transaction(trans, root);
3295                         if (ret) {
3296                                 iput(inode);
3297                                 goto out;
3298                         }
3299
3300                         ret = btrfs_truncate(inode);
3301                         if (ret)
3302                                 btrfs_orphan_del(NULL, inode);
3303                 } else {
3304                         nr_unlink++;
3305                 }
3306
3307                 /* this will do delete_inode and everything for us */
3308                 iput(inode);
3309                 if (ret)
3310                         goto out;
3311         }
3312         /* release the path since we're done with it */
3313         btrfs_release_path(path);
3314
3315         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3316
3317         if (root->orphan_block_rsv)
3318                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3319                                         (u64)-1);
3320
3321         if (root->orphan_block_rsv ||
3322             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3323                 trans = btrfs_join_transaction(root);
3324                 if (!IS_ERR(trans))
3325                         btrfs_end_transaction(trans, root);
3326         }
3327
3328         if (nr_unlink)
3329                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3330         if (nr_truncate)
3331                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3332
3333 out:
3334         if (ret)
3335                 btrfs_crit(root->fs_info,
3336                         "could not do orphan cleanup %d", ret);
3337         btrfs_free_path(path);
3338         return ret;
3339 }
3340
3341 /*
3342  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3343  * don't find any xattrs, we know there can't be any acls.
3344  *
3345  * slot is the slot the inode is in, objectid is the objectid of the inode
3346  */
3347 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3348                                           int slot, u64 objectid,
3349                                           int *first_xattr_slot)
3350 {
3351         u32 nritems = btrfs_header_nritems(leaf);
3352         struct btrfs_key found_key;
3353         static u64 xattr_access = 0;
3354         static u64 xattr_default = 0;
3355         int scanned = 0;
3356
3357         if (!xattr_access) {
3358                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3359                                         strlen(POSIX_ACL_XATTR_ACCESS));
3360                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3361                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3362         }
3363
3364         slot++;
3365         *first_xattr_slot = -1;
3366         while (slot < nritems) {
3367                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3368
3369                 /* we found a different objectid, there must not be acls */
3370                 if (found_key.objectid != objectid)
3371                         return 0;
3372
3373                 /* we found an xattr, assume we've got an acl */
3374                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3375                         if (*first_xattr_slot == -1)
3376                                 *first_xattr_slot = slot;
3377                         if (found_key.offset == xattr_access ||
3378                             found_key.offset == xattr_default)
3379                                 return 1;
3380                 }
3381
3382                 /*
3383                  * we found a key greater than an xattr key, there can't
3384                  * be any acls later on
3385                  */
3386                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3387                         return 0;
3388
3389                 slot++;
3390                 scanned++;
3391
3392                 /*
3393                  * it goes inode, inode backrefs, xattrs, extents,
3394                  * so if there are a ton of hard links to an inode there can
3395                  * be a lot of backrefs.  Don't waste time searching too hard,
3396                  * this is just an optimization
3397                  */
3398                 if (scanned >= 8)
3399                         break;
3400         }
3401         /* we hit the end of the leaf before we found an xattr or
3402          * something larger than an xattr.  We have to assume the inode
3403          * has acls
3404          */
3405         if (*first_xattr_slot == -1)
3406                 *first_xattr_slot = slot;
3407         return 1;
3408 }
3409
3410 /*
3411  * read an inode from the btree into the in-memory inode
3412  */
3413 static void btrfs_read_locked_inode(struct inode *inode)
3414 {
3415         struct btrfs_path *path;
3416         struct extent_buffer *leaf;
3417         struct btrfs_inode_item *inode_item;
3418         struct btrfs_timespec *tspec;
3419         struct btrfs_root *root = BTRFS_I(inode)->root;
3420         struct btrfs_key location;
3421         unsigned long ptr;
3422         int maybe_acls;
3423         u32 rdev;
3424         int ret;
3425         bool filled = false;
3426         int first_xattr_slot;
3427
3428         ret = btrfs_fill_inode(inode, &rdev);
3429         if (!ret)
3430                 filled = true;
3431
3432         path = btrfs_alloc_path();
3433         if (!path)
3434                 goto make_bad;
3435
3436         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3437
3438         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3439         if (ret)
3440                 goto make_bad;
3441
3442         leaf = path->nodes[0];
3443
3444         if (filled)
3445                 goto cache_index;
3446
3447         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3448                                     struct btrfs_inode_item);
3449         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3450         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3451         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3452         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3453         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3454
3455         tspec = btrfs_inode_atime(inode_item);
3456         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3457         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3458
3459         tspec = btrfs_inode_mtime(inode_item);
3460         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3461         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3462
3463         tspec = btrfs_inode_ctime(inode_item);
3464         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3465         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3466
3467         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3468         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3469         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3470
3471         /*
3472          * If we were modified in the current generation and evicted from memory
3473          * and then re-read we need to do a full sync since we don't have any
3474          * idea about which extents were modified before we were evicted from
3475          * cache.
3476          */
3477         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3478                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3479                         &BTRFS_I(inode)->runtime_flags);
3480
3481         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3482         inode->i_generation = BTRFS_I(inode)->generation;
3483         inode->i_rdev = 0;
3484         rdev = btrfs_inode_rdev(leaf, inode_item);
3485
3486         BTRFS_I(inode)->index_cnt = (u64)-1;
3487         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3488
3489 cache_index:
3490         path->slots[0]++;
3491         if (inode->i_nlink != 1 ||
3492             path->slots[0] >= btrfs_header_nritems(leaf))
3493                 goto cache_acl;
3494
3495         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3496         if (location.objectid != btrfs_ino(inode))
3497                 goto cache_acl;
3498
3499         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3500         if (location.type == BTRFS_INODE_REF_KEY) {
3501                 struct btrfs_inode_ref *ref;
3502
3503                 ref = (struct btrfs_inode_ref *)ptr;
3504                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3505         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3506                 struct btrfs_inode_extref *extref;
3507
3508                 extref = (struct btrfs_inode_extref *)ptr;
3509                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3510                                                                      extref);
3511         }
3512 cache_acl:
3513         /*
3514          * try to precache a NULL acl entry for files that don't have
3515          * any xattrs or acls
3516          */
3517         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3518                                            btrfs_ino(inode), &first_xattr_slot);
3519         if (first_xattr_slot != -1) {
3520                 path->slots[0] = first_xattr_slot;
3521                 ret = btrfs_load_inode_props(inode, path);
3522                 if (ret)
3523                         btrfs_err(root->fs_info,
3524                                   "error loading props for ino %llu (root %llu): %d",
3525                                   btrfs_ino(inode),
3526                                   root->root_key.objectid, ret);
3527         }
3528         btrfs_free_path(path);
3529
3530         if (!maybe_acls)
3531                 cache_no_acl(inode);
3532
3533         switch (inode->i_mode & S_IFMT) {
3534         case S_IFREG:
3535                 inode->i_mapping->a_ops = &btrfs_aops;
3536                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3537                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3538                 inode->i_fop = &btrfs_file_operations;
3539                 inode->i_op = &btrfs_file_inode_operations;
3540                 break;
3541         case S_IFDIR:
3542                 inode->i_fop = &btrfs_dir_file_operations;
3543                 if (root == root->fs_info->tree_root)
3544                         inode->i_op = &btrfs_dir_ro_inode_operations;
3545                 else
3546                         inode->i_op = &btrfs_dir_inode_operations;
3547                 break;
3548         case S_IFLNK:
3549                 inode->i_op = &btrfs_symlink_inode_operations;
3550                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3551                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3552                 break;
3553         default:
3554                 inode->i_op = &btrfs_special_inode_operations;
3555                 init_special_inode(inode, inode->i_mode, rdev);
3556                 break;
3557         }
3558
3559         btrfs_update_iflags(inode);
3560         return;
3561
3562 make_bad:
3563         btrfs_free_path(path);
3564         make_bad_inode(inode);
3565 }
3566
3567 /*
3568  * given a leaf and an inode, copy the inode fields into the leaf
3569  */
3570 static void fill_inode_item(struct btrfs_trans_handle *trans,
3571                             struct extent_buffer *leaf,
3572                             struct btrfs_inode_item *item,
3573                             struct inode *inode)
3574 {
3575         struct btrfs_map_token token;
3576
3577         btrfs_init_map_token(&token);
3578
3579         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3580         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3581         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3582                                    &token);
3583         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3584         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3585
3586         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3587                                      inode->i_atime.tv_sec, &token);
3588         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3589                                       inode->i_atime.tv_nsec, &token);
3590
3591         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3592                                      inode->i_mtime.tv_sec, &token);
3593         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3594                                       inode->i_mtime.tv_nsec, &token);
3595
3596         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3597                                      inode->i_ctime.tv_sec, &token);
3598         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3599                                       inode->i_ctime.tv_nsec, &token);
3600
3601         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3602                                      &token);
3603         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3604                                          &token);
3605         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3606         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3607         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3608         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3609         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3610 }
3611
3612 /*
3613  * copy everything in the in-memory inode into the btree.
3614  */
3615 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3616                                 struct btrfs_root *root, struct inode *inode)
3617 {
3618         struct btrfs_inode_item *inode_item;
3619         struct btrfs_path *path;
3620         struct extent_buffer *leaf;
3621         int ret;
3622
3623         path = btrfs_alloc_path();
3624         if (!path)
3625                 return -ENOMEM;
3626
3627         path->leave_spinning = 1;
3628         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3629                                  1);
3630         if (ret) {
3631                 if (ret > 0)
3632                         ret = -ENOENT;
3633                 goto failed;
3634         }
3635
3636         leaf = path->nodes[0];
3637         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638                                     struct btrfs_inode_item);
3639
3640         fill_inode_item(trans, leaf, inode_item, inode);
3641         btrfs_mark_buffer_dirty(leaf);
3642         btrfs_set_inode_last_trans(trans, inode);
3643         ret = 0;
3644 failed:
3645         btrfs_free_path(path);
3646         return ret;
3647 }
3648
3649 /*
3650  * copy everything in the in-memory inode into the btree.
3651  */
3652 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3653                                 struct btrfs_root *root, struct inode *inode)
3654 {
3655         int ret;
3656
3657         /*
3658          * If the inode is a free space inode, we can deadlock during commit
3659          * if we put it into the delayed code.
3660          *
3661          * The data relocation inode should also be directly updated
3662          * without delay
3663          */
3664         if (!btrfs_is_free_space_inode(inode)
3665             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3666                 btrfs_update_root_times(trans, root);
3667
3668                 ret = btrfs_delayed_update_inode(trans, root, inode);
3669                 if (!ret)
3670                         btrfs_set_inode_last_trans(trans, inode);
3671                 return ret;
3672         }
3673
3674         return btrfs_update_inode_item(trans, root, inode);
3675 }
3676
3677 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3678                                          struct btrfs_root *root,
3679                                          struct inode *inode)
3680 {
3681         int ret;
3682
3683         ret = btrfs_update_inode(trans, root, inode);
3684         if (ret == -ENOSPC)
3685                 return btrfs_update_inode_item(trans, root, inode);
3686         return ret;
3687 }
3688
3689 /*
3690  * unlink helper that gets used here in inode.c and in the tree logging
3691  * recovery code.  It remove a link in a directory with a given name, and
3692  * also drops the back refs in the inode to the directory
3693  */
3694 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3695                                 struct btrfs_root *root,
3696                                 struct inode *dir, struct inode *inode,
3697                                 const char *name, int name_len)
3698 {
3699         struct btrfs_path *path;
3700         int ret = 0;
3701         struct extent_buffer *leaf;
3702         struct btrfs_dir_item *di;
3703         struct btrfs_key key;
3704         u64 index;
3705         u64 ino = btrfs_ino(inode);
3706         u64 dir_ino = btrfs_ino(dir);
3707
3708         path = btrfs_alloc_path();
3709         if (!path) {
3710                 ret = -ENOMEM;
3711                 goto out;
3712         }
3713
3714         path->leave_spinning = 1;
3715         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3716                                     name, name_len, -1);
3717         if (IS_ERR(di)) {
3718                 ret = PTR_ERR(di);
3719                 goto err;
3720         }
3721         if (!di) {
3722                 ret = -ENOENT;
3723                 goto err;
3724         }
3725         leaf = path->nodes[0];
3726         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3727         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3728         if (ret)
3729                 goto err;
3730         btrfs_release_path(path);
3731
3732         /*
3733          * If we don't have dir index, we have to get it by looking up
3734          * the inode ref, since we get the inode ref, remove it directly,
3735          * it is unnecessary to do delayed deletion.
3736          *
3737          * But if we have dir index, needn't search inode ref to get it.
3738          * Since the inode ref is close to the inode item, it is better
3739          * that we delay to delete it, and just do this deletion when
3740          * we update the inode item.
3741          */
3742         if (BTRFS_I(inode)->dir_index) {
3743                 ret = btrfs_delayed_delete_inode_ref(inode);
3744                 if (!ret) {
3745                         index = BTRFS_I(inode)->dir_index;
3746                         goto skip_backref;
3747                 }
3748         }
3749
3750         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3751                                   dir_ino, &index);
3752         if (ret) {
3753                 btrfs_info(root->fs_info,
3754                         "failed to delete reference to %.*s, inode %llu parent %llu",
3755                         name_len, name, ino, dir_ino);
3756                 btrfs_abort_transaction(trans, root, ret);
3757                 goto err;
3758         }
3759 skip_backref:
3760         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3761         if (ret) {
3762                 btrfs_abort_transaction(trans, root, ret);
3763                 goto err;
3764         }
3765
3766         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3767                                          inode, dir_ino);
3768         if (ret != 0 && ret != -ENOENT) {
3769                 btrfs_abort_transaction(trans, root, ret);
3770                 goto err;
3771         }
3772
3773         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3774                                            dir, index);
3775         if (ret == -ENOENT)
3776                 ret = 0;
3777         else if (ret)
3778                 btrfs_abort_transaction(trans, root, ret);
3779 err:
3780         btrfs_free_path(path);
3781         if (ret)
3782                 goto out;
3783
3784         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3785         inode_inc_iversion(inode);
3786         inode_inc_iversion(dir);
3787         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3788         ret = btrfs_update_inode(trans, root, dir);
3789 out:
3790         return ret;
3791 }
3792
3793 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3794                        struct btrfs_root *root,
3795                        struct inode *dir, struct inode *inode,
3796                        const char *name, int name_len)
3797 {
3798         int ret;
3799         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3800         if (!ret) {
3801                 drop_nlink(inode);
3802                 ret = btrfs_update_inode(trans, root, inode);
3803         }
3804         return ret;
3805 }
3806
3807 /*
3808  * helper to start transaction for unlink and rmdir.
3809  *
3810  * unlink and rmdir are special in btrfs, they do not always free space, so
3811  * if we cannot make our reservations the normal way try and see if there is
3812  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3813  * allow the unlink to occur.
3814  */
3815 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3816 {
3817         struct btrfs_trans_handle *trans;
3818         struct btrfs_root *root = BTRFS_I(dir)->root;
3819         int ret;
3820
3821         /*
3822          * 1 for the possible orphan item
3823          * 1 for the dir item
3824          * 1 for the dir index
3825          * 1 for the inode ref
3826          * 1 for the inode
3827          */
3828         trans = btrfs_start_transaction(root, 5);
3829         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3830                 return trans;
3831
3832         if (PTR_ERR(trans) == -ENOSPC) {
3833                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3834
3835                 trans = btrfs_start_transaction(root, 0);
3836                 if (IS_ERR(trans))
3837                         return trans;
3838                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3839                                                &root->fs_info->trans_block_rsv,
3840                                                num_bytes, 5);
3841                 if (ret) {
3842                         btrfs_end_transaction(trans, root);
3843                         return ERR_PTR(ret);
3844                 }
3845                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3846                 trans->bytes_reserved = num_bytes;
3847         }
3848         return trans;
3849 }
3850
3851 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3852 {
3853         struct btrfs_root *root = BTRFS_I(dir)->root;
3854         struct btrfs_trans_handle *trans;
3855         struct inode *inode = dentry->d_inode;
3856         int ret;
3857
3858         trans = __unlink_start_trans(dir);
3859         if (IS_ERR(trans))
3860                 return PTR_ERR(trans);
3861
3862         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3863
3864         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3865                                  dentry->d_name.name, dentry->d_name.len);
3866         if (ret)
3867                 goto out;
3868
3869         if (inode->i_nlink == 0) {
3870                 ret = btrfs_orphan_add(trans, inode);
3871                 if (ret)
3872                         goto out;
3873         }
3874
3875 out:
3876         btrfs_end_transaction(trans, root);
3877         btrfs_btree_balance_dirty(root);
3878         return ret;
3879 }
3880
3881 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3882                         struct btrfs_root *root,
3883                         struct inode *dir, u64 objectid,
3884                         const char *name, int name_len)
3885 {
3886         struct btrfs_path *path;
3887         struct extent_buffer *leaf;
3888         struct btrfs_dir_item *di;
3889         struct btrfs_key key;
3890         u64 index;
3891         int ret;
3892         u64 dir_ino = btrfs_ino(dir);
3893
3894         path = btrfs_alloc_path();
3895         if (!path)
3896                 return -ENOMEM;
3897
3898         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3899                                    name, name_len, -1);
3900         if (IS_ERR_OR_NULL(di)) {
3901                 if (!di)
3902                         ret = -ENOENT;
3903                 else
3904                         ret = PTR_ERR(di);
3905                 goto out;
3906         }
3907
3908         leaf = path->nodes[0];
3909         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3910         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3911         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3912         if (ret) {
3913                 btrfs_abort_transaction(trans, root, ret);
3914                 goto out;
3915         }
3916         btrfs_release_path(path);
3917
3918         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3919                                  objectid, root->root_key.objectid,
3920                                  dir_ino, &index, name, name_len);
3921         if (ret < 0) {
3922                 if (ret != -ENOENT) {
3923                         btrfs_abort_transaction(trans, root, ret);
3924                         goto out;
3925                 }
3926                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3927                                                  name, name_len);
3928                 if (IS_ERR_OR_NULL(di)) {
3929                         if (!di)
3930                                 ret = -ENOENT;
3931                         else
3932                                 ret = PTR_ERR(di);
3933                         btrfs_abort_transaction(trans, root, ret);
3934                         goto out;
3935                 }
3936
3937                 leaf = path->nodes[0];
3938                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3939                 btrfs_release_path(path);
3940                 index = key.offset;
3941         }
3942         btrfs_release_path(path);
3943
3944         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3945         if (ret) {
3946                 btrfs_abort_transaction(trans, root, ret);
3947                 goto out;
3948         }
3949
3950         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3951         inode_inc_iversion(dir);
3952         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3953         ret = btrfs_update_inode_fallback(trans, root, dir);
3954         if (ret)
3955                 btrfs_abort_transaction(trans, root, ret);
3956 out:
3957         btrfs_free_path(path);
3958         return ret;
3959 }
3960
3961 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3962 {
3963         struct inode *inode = dentry->d_inode;
3964         int err = 0;
3965         struct btrfs_root *root = BTRFS_I(dir)->root;
3966         struct btrfs_trans_handle *trans;
3967
3968         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3969                 return -ENOTEMPTY;
3970         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3971                 return -EPERM;
3972
3973         trans = __unlink_start_trans(dir);
3974         if (IS_ERR(trans))
3975                 return PTR_ERR(trans);
3976
3977         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3978                 err = btrfs_unlink_subvol(trans, root, dir,
3979                                           BTRFS_I(inode)->location.objectid,
3980                                           dentry->d_name.name,
3981                                           dentry->d_name.len);
3982                 goto out;
3983         }
3984
3985         err = btrfs_orphan_add(trans, inode);
3986         if (err)
3987                 goto out;
3988
3989         /* now the directory is empty */
3990         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3991                                  dentry->d_name.name, dentry->d_name.len);
3992         if (!err)
3993                 btrfs_i_size_write(inode, 0);
3994 out:
3995         btrfs_end_transaction(trans, root);
3996         btrfs_btree_balance_dirty(root);
3997
3998         return err;
3999 }
4000
4001 /*
4002  * this can truncate away extent items, csum items and directory items.
4003  * It starts at a high offset and removes keys until it can't find
4004  * any higher than new_size
4005  *
4006  * csum items that cross the new i_size are truncated to the new size
4007  * as well.
4008  *
4009  * min_type is the minimum key type to truncate down to.  If set to 0, this
4010  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4011  */
4012 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4013                                struct btrfs_root *root,
4014                                struct inode *inode,
4015                                u64 new_size, u32 min_type)
4016 {
4017         struct btrfs_path *path;
4018         struct extent_buffer *leaf;
4019         struct btrfs_file_extent_item *fi;
4020         struct btrfs_key key;
4021         struct btrfs_key found_key;
4022         u64 extent_start = 0;
4023         u64 extent_num_bytes = 0;
4024         u64 extent_offset = 0;
4025         u64 item_end = 0;
4026         u64 last_size = (u64)-1;
4027         u32 found_type = (u8)-1;
4028         int found_extent;
4029         int del_item;
4030         int pending_del_nr = 0;
4031         int pending_del_slot = 0;
4032         int extent_type = -1;
4033         int ret;
4034         int err = 0;
4035         u64 ino = btrfs_ino(inode);
4036
4037         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4038
4039         path = btrfs_alloc_path();
4040         if (!path)
4041                 return -ENOMEM;
4042         path->reada = -1;
4043
4044         /*
4045          * We want to drop from the next block forward in case this new size is
4046          * not block aligned since we will be keeping the last block of the
4047          * extent just the way it is.
4048          */
4049         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4050             root == root->fs_info->tree_root)
4051                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4052                                         root->sectorsize), (u64)-1, 0);
4053
4054         /*
4055          * This function is also used to drop the items in the log tree before
4056          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4057          * it is used to drop the loged items. So we shouldn't kill the delayed
4058          * items.
4059          */
4060         if (min_type == 0 && root == BTRFS_I(inode)->root)
4061                 btrfs_kill_delayed_inode_items(inode);
4062
4063         key.objectid = ino;
4064         key.offset = (u64)-1;
4065         key.type = (u8)-1;
4066
4067 search_again:
4068         path->leave_spinning = 1;
4069         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4070         if (ret < 0) {
4071                 err = ret;
4072                 goto out;
4073         }
4074
4075         if (ret > 0) {
4076                 /* there are no items in the tree for us to truncate, we're
4077                  * done
4078                  */
4079                 if (path->slots[0] == 0)
4080                         goto out;
4081                 path->slots[0]--;
4082         }
4083
4084         while (1) {
4085                 fi = NULL;
4086                 leaf = path->nodes[0];
4087                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4088                 found_type = btrfs_key_type(&found_key);
4089
4090                 if (found_key.objectid != ino)
4091                         break;
4092
4093                 if (found_type < min_type)
4094                         break;
4095
4096                 item_end = found_key.offset;
4097                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4098                         fi = btrfs_item_ptr(leaf, path->slots[0],
4099                                             struct btrfs_file_extent_item);
4100                         extent_type = btrfs_file_extent_type(leaf, fi);
4101                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4102                                 item_end +=
4103                                     btrfs_file_extent_num_bytes(leaf, fi);
4104                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4105                                 item_end += btrfs_file_extent_inline_len(leaf,
4106                                                          path->slots[0], fi);
4107                         }
4108                         item_end--;
4109                 }
4110                 if (found_type > min_type) {
4111                         del_item = 1;
4112                 } else {
4113                         if (item_end < new_size)
4114                                 break;
4115                         if (found_key.offset >= new_size)
4116                                 del_item = 1;
4117                         else
4118                                 del_item = 0;
4119                 }
4120                 found_extent = 0;
4121                 /* FIXME, shrink the extent if the ref count is only 1 */
4122                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4123                         goto delete;
4124
4125                 if (del_item)
4126                         last_size = found_key.offset;
4127                 else
4128                         last_size = new_size;
4129
4130                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4131                         u64 num_dec;
4132                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4133                         if (!del_item) {
4134                                 u64 orig_num_bytes =
4135                                         btrfs_file_extent_num_bytes(leaf, fi);
4136                                 extent_num_bytes = ALIGN(new_size -
4137                                                 found_key.offset,
4138                                                 root->sectorsize);
4139                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4140                                                          extent_num_bytes);
4141                                 num_dec = (orig_num_bytes -
4142                                            extent_num_bytes);
4143                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4144                                              &root->state) &&
4145                                     extent_start != 0)
4146                                         inode_sub_bytes(inode, num_dec);
4147                                 btrfs_mark_buffer_dirty(leaf);
4148                         } else {
4149                                 extent_num_bytes =
4150                                         btrfs_file_extent_disk_num_bytes(leaf,
4151                                                                          fi);
4152                                 extent_offset = found_key.offset -
4153                                         btrfs_file_extent_offset(leaf, fi);
4154
4155                                 /* FIXME blocksize != 4096 */
4156                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4157                                 if (extent_start != 0) {
4158                                         found_extent = 1;
4159                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4160                                                      &root->state))
4161                                                 inode_sub_bytes(inode, num_dec);
4162                                 }
4163                         }
4164                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4165                         /*
4166                          * we can't truncate inline items that have had
4167                          * special encodings
4168                          */
4169                         if (!del_item &&
4170                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4171                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4172                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4173                                 u32 size = new_size - found_key.offset;
4174
4175                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4176                                         inode_sub_bytes(inode, item_end + 1 -
4177                                                         new_size);
4178
4179                                 /*
4180                                  * update the ram bytes to properly reflect
4181                                  * the new size of our item
4182                                  */
4183                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4184                                 size =
4185                                     btrfs_file_extent_calc_inline_size(size);
4186                                 btrfs_truncate_item(root, path, size, 1);
4187                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4188                                             &root->state)) {
4189                                 inode_sub_bytes(inode, item_end + 1 -
4190                                                 found_key.offset);
4191                         }
4192                 }
4193 delete:
4194                 if (del_item) {
4195                         if (!pending_del_nr) {
4196                                 /* no pending yet, add ourselves */
4197                                 pending_del_slot = path->slots[0];
4198                                 pending_del_nr = 1;
4199                         } else if (pending_del_nr &&
4200                                    path->slots[0] + 1 == pending_del_slot) {
4201                                 /* hop on the pending chunk */
4202                                 pending_del_nr++;
4203                                 pending_del_slot = path->slots[0];
4204                         } else {
4205                                 BUG();
4206                         }
4207                 } else {
4208                         break;
4209                 }
4210                 if (found_extent &&
4211                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4212                      root == root->fs_info->tree_root)) {
4213                         btrfs_set_path_blocking(path);
4214                         ret = btrfs_free_extent(trans, root, extent_start,
4215                                                 extent_num_bytes, 0,
4216                                                 btrfs_header_owner(leaf),
4217                                                 ino, extent_offset, 0);
4218                         BUG_ON(ret);
4219                 }
4220
4221                 if (found_type == BTRFS_INODE_ITEM_KEY)
4222                         break;
4223
4224                 if (path->slots[0] == 0 ||
4225                     path->slots[0] != pending_del_slot) {
4226                         if (pending_del_nr) {
4227                                 ret = btrfs_del_items(trans, root, path,
4228                                                 pending_del_slot,
4229                                                 pending_del_nr);
4230                                 if (ret) {
4231                                         btrfs_abort_transaction(trans,
4232                                                                 root, ret);
4233                                         goto error;
4234                                 }
4235                                 pending_del_nr = 0;
4236                         }
4237                         btrfs_release_path(path);
4238                         goto search_again;
4239                 } else {
4240                         path->slots[0]--;
4241                 }
4242         }
4243 out:
4244         if (pending_del_nr) {
4245                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4246                                       pending_del_nr);
4247                 if (ret)
4248                         btrfs_abort_transaction(trans, root, ret);
4249         }
4250 error:
4251         if (last_size != (u64)-1 &&
4252             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4253                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4254         btrfs_free_path(path);
4255         return err;
4256 }
4257
4258 /*
4259  * btrfs_truncate_page - read, zero a chunk and write a page
4260  * @inode - inode that we're zeroing
4261  * @from - the offset to start zeroing
4262  * @len - the length to zero, 0 to zero the entire range respective to the
4263  *      offset
4264  * @front - zero up to the offset instead of from the offset on
4265  *
4266  * This will find the page for the "from" offset and cow the page and zero the
4267  * part we want to zero.  This is used with truncate and hole punching.
4268  */
4269 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4270                         int front)
4271 {
4272         struct address_space *mapping = inode->i_mapping;
4273         struct btrfs_root *root = BTRFS_I(inode)->root;
4274         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4275         struct btrfs_ordered_extent *ordered;
4276         struct extent_state *cached_state = NULL;
4277         char *kaddr;
4278         u32 blocksize = root->sectorsize;
4279         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4280         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4281         struct page *page;
4282         gfp_t mask = btrfs_alloc_write_mask(mapping);
4283         int ret = 0;
4284         u64 page_start;
4285         u64 page_end;
4286
4287         if ((offset & (blocksize - 1)) == 0 &&
4288             (!len || ((len & (blocksize - 1)) == 0)))
4289                 goto out;
4290         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4291         if (ret)
4292                 goto out;
4293
4294 again:
4295         page = find_or_create_page(mapping, index, mask);
4296         if (!page) {
4297                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4298                 ret = -ENOMEM;
4299                 goto out;
4300         }
4301
4302         page_start = page_offset(page);
4303         page_end = page_start + PAGE_CACHE_SIZE - 1;
4304
4305         if (!PageUptodate(page)) {
4306                 ret = btrfs_readpage(NULL, page);
4307                 lock_page(page);
4308                 if (page->mapping != mapping) {
4309                         unlock_page(page);
4310                         page_cache_release(page);
4311                         goto again;
4312                 }
4313                 if (!PageUptodate(page)) {
4314                         ret = -EIO;
4315                         goto out_unlock;
4316                 }
4317         }
4318         wait_on_page_writeback(page);
4319
4320         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4321         set_page_extent_mapped(page);
4322
4323         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4324         if (ordered) {
4325                 unlock_extent_cached(io_tree, page_start, page_end,
4326                                      &cached_state, GFP_NOFS);
4327                 unlock_page(page);
4328                 page_cache_release(page);
4329                 btrfs_start_ordered_extent(inode, ordered, 1);
4330                 btrfs_put_ordered_extent(ordered);
4331                 goto again;
4332         }
4333
4334         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4335                           EXTENT_DIRTY | EXTENT_DELALLOC |
4336                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4337                           0, 0, &cached_state, GFP_NOFS);
4338
4339         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4340                                         &cached_state);
4341         if (ret) {
4342                 unlock_extent_cached(io_tree, page_start, page_end,
4343                                      &cached_state, GFP_NOFS);
4344                 goto out_unlock;
4345         }
4346
4347         if (offset != PAGE_CACHE_SIZE) {
4348                 if (!len)
4349                         len = PAGE_CACHE_SIZE - offset;
4350                 kaddr = kmap(page);
4351                 if (front)
4352                         memset(kaddr, 0, offset);
4353                 else
4354                         memset(kaddr + offset, 0, len);
4355                 flush_dcache_page(page);
4356                 kunmap(page);
4357         }
4358         ClearPageChecked(page);
4359         set_page_dirty(page);
4360         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4361                              GFP_NOFS);
4362
4363 out_unlock:
4364         if (ret)
4365                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4366         unlock_page(page);
4367         page_cache_release(page);
4368 out:
4369         return ret;
4370 }
4371
4372 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4373                              u64 offset, u64 len)
4374 {
4375         struct btrfs_trans_handle *trans;
4376         int ret;
4377
4378         /*
4379          * Still need to make sure the inode looks like it's been updated so
4380          * that any holes get logged if we fsync.
4381          */
4382         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4383                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4384                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4385                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4386                 return 0;
4387         }
4388
4389         /*
4390          * 1 - for the one we're dropping
4391          * 1 - for the one we're adding
4392          * 1 - for updating the inode.
4393          */
4394         trans = btrfs_start_transaction(root, 3);
4395         if (IS_ERR(trans))
4396                 return PTR_ERR(trans);
4397
4398         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4399         if (ret) {
4400                 btrfs_abort_transaction(trans, root, ret);
4401                 btrfs_end_transaction(trans, root);
4402                 return ret;
4403         }
4404
4405         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4406                                        0, 0, len, 0, len, 0, 0, 0);
4407         if (ret)
4408                 btrfs_abort_transaction(trans, root, ret);
4409         else
4410                 btrfs_update_inode(trans, root, inode);
4411         btrfs_end_transaction(trans, root);
4412         return ret;
4413 }
4414
4415 /*
4416  * This function puts in dummy file extents for the area we're creating a hole
4417  * for.  So if we are truncating this file to a larger size we need to insert
4418  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4419  * the range between oldsize and size
4420  */
4421 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4422 {
4423         struct btrfs_root *root = BTRFS_I(inode)->root;
4424         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4425         struct extent_map *em = NULL;
4426         struct extent_state *cached_state = NULL;
4427         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4428         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4429         u64 block_end = ALIGN(size, root->sectorsize);
4430         u64 last_byte;
4431         u64 cur_offset;
4432         u64 hole_size;
4433         int err = 0;
4434
4435         /*
4436          * If our size started in the middle of a page we need to zero out the
4437          * rest of the page before we expand the i_size, otherwise we could
4438          * expose stale data.
4439          */
4440         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4441         if (err)
4442                 return err;
4443
4444         if (size <= hole_start)
4445                 return 0;
4446
4447         while (1) {
4448                 struct btrfs_ordered_extent *ordered;
4449
4450                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4451                                  &cached_state);
4452                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4453                                                      block_end - hole_start);
4454                 if (!ordered)
4455                         break;
4456                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4457                                      &cached_state, GFP_NOFS);
4458                 btrfs_start_ordered_extent(inode, ordered, 1);
4459                 btrfs_put_ordered_extent(ordered);
4460         }
4461
4462         cur_offset = hole_start;
4463         while (1) {
4464                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4465                                 block_end - cur_offset, 0);
4466                 if (IS_ERR(em)) {
4467                         err = PTR_ERR(em);
4468                         em = NULL;
4469                         break;
4470                 }
4471                 last_byte = min(extent_map_end(em), block_end);
4472                 last_byte = ALIGN(last_byte , root->sectorsize);
4473                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4474                         struct extent_map *hole_em;
4475                         hole_size = last_byte - cur_offset;
4476
4477                         err = maybe_insert_hole(root, inode, cur_offset,
4478                                                 hole_size);
4479                         if (err)
4480                                 break;
4481                         btrfs_drop_extent_cache(inode, cur_offset,
4482                                                 cur_offset + hole_size - 1, 0);
4483                         hole_em = alloc_extent_map();
4484                         if (!hole_em) {
4485                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4486                                         &BTRFS_I(inode)->runtime_flags);
4487                                 goto next;
4488                         }
4489                         hole_em->start = cur_offset;
4490                         hole_em->len = hole_size;
4491                         hole_em->orig_start = cur_offset;
4492
4493                         hole_em->block_start = EXTENT_MAP_HOLE;
4494                         hole_em->block_len = 0;
4495                         hole_em->orig_block_len = 0;
4496                         hole_em->ram_bytes = hole_size;
4497                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4498                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4499                         hole_em->generation = root->fs_info->generation;
4500
4501                         while (1) {
4502                                 write_lock(&em_tree->lock);
4503                                 err = add_extent_mapping(em_tree, hole_em, 1);
4504                                 write_unlock(&em_tree->lock);
4505                                 if (err != -EEXIST)
4506                                         break;
4507                                 btrfs_drop_extent_cache(inode, cur_offset,
4508                                                         cur_offset +
4509                                                         hole_size - 1, 0);
4510                         }
4511                         free_extent_map(hole_em);
4512                 }
4513 next:
4514                 free_extent_map(em);
4515                 em = NULL;
4516                 cur_offset = last_byte;
4517                 if (cur_offset >= block_end)
4518                         break;
4519         }
4520         free_extent_map(em);
4521         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4522                              GFP_NOFS);
4523         return err;
4524 }
4525
4526 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4527 {
4528         struct btrfs_root *root = BTRFS_I(inode)->root;
4529         struct btrfs_trans_handle *trans;
4530         loff_t oldsize = i_size_read(inode);
4531         loff_t newsize = attr->ia_size;
4532         int mask = attr->ia_valid;
4533         int ret;
4534
4535         /*
4536          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4537          * special case where we need to update the times despite not having
4538          * these flags set.  For all other operations the VFS set these flags
4539          * explicitly if it wants a timestamp update.
4540          */
4541         if (newsize != oldsize) {
4542                 inode_inc_iversion(inode);
4543                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4544                         inode->i_ctime = inode->i_mtime =
4545                                 current_fs_time(inode->i_sb);
4546         }
4547
4548         if (newsize > oldsize) {
4549                 truncate_pagecache(inode, newsize);
4550                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4551                 if (ret)
4552                         return ret;
4553
4554                 trans = btrfs_start_transaction(root, 1);
4555                 if (IS_ERR(trans))
4556                         return PTR_ERR(trans);
4557
4558                 i_size_write(inode, newsize);
4559                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4560                 ret = btrfs_update_inode(trans, root, inode);
4561                 btrfs_end_transaction(trans, root);
4562         } else {
4563
4564                 /*
4565                  * We're truncating a file that used to have good data down to
4566                  * zero. Make sure it gets into the ordered flush list so that
4567                  * any new writes get down to disk quickly.
4568                  */
4569                 if (newsize == 0)
4570                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4571                                 &BTRFS_I(inode)->runtime_flags);
4572
4573                 /*
4574                  * 1 for the orphan item we're going to add
4575                  * 1 for the orphan item deletion.
4576                  */
4577                 trans = btrfs_start_transaction(root, 2);
4578                 if (IS_ERR(trans))
4579                         return PTR_ERR(trans);
4580
4581                 /*
4582                  * We need to do this in case we fail at _any_ point during the
4583                  * actual truncate.  Once we do the truncate_setsize we could
4584                  * invalidate pages which forces any outstanding ordered io to
4585                  * be instantly completed which will give us extents that need
4586                  * to be truncated.  If we fail to get an orphan inode down we
4587                  * could have left over extents that were never meant to live,
4588                  * so we need to garuntee from this point on that everything
4589                  * will be consistent.
4590                  */
4591                 ret = btrfs_orphan_add(trans, inode);
4592                 btrfs_end_transaction(trans, root);
4593                 if (ret)
4594                         return ret;
4595
4596                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4597                 truncate_setsize(inode, newsize);
4598
4599                 /* Disable nonlocked read DIO to avoid the end less truncate */
4600                 btrfs_inode_block_unlocked_dio(inode);
4601                 inode_dio_wait(inode);
4602                 btrfs_inode_resume_unlocked_dio(inode);
4603
4604                 ret = btrfs_truncate(inode);
4605                 if (ret && inode->i_nlink) {
4606                         int err;
4607
4608                         /*
4609                          * failed to truncate, disk_i_size is only adjusted down
4610                          * as we remove extents, so it should represent the true
4611                          * size of the inode, so reset the in memory size and
4612                          * delete our orphan entry.
4613                          */
4614                         trans = btrfs_join_transaction(root);
4615                         if (IS_ERR(trans)) {
4616                                 btrfs_orphan_del(NULL, inode);
4617                                 return ret;
4618                         }
4619                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4620                         err = btrfs_orphan_del(trans, inode);
4621                         if (err)
4622                                 btrfs_abort_transaction(trans, root, err);
4623                         btrfs_end_transaction(trans, root);
4624                 }
4625         }
4626
4627         return ret;
4628 }
4629
4630 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4631 {
4632         struct inode *inode = dentry->d_inode;
4633         struct btrfs_root *root = BTRFS_I(inode)->root;
4634         int err;
4635
4636         if (btrfs_root_readonly(root))
4637                 return -EROFS;
4638
4639         err = inode_change_ok(inode, attr);
4640         if (err)
4641                 return err;
4642
4643         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4644                 err = btrfs_setsize(inode, attr);
4645                 if (err)
4646                         return err;
4647         }
4648
4649         if (attr->ia_valid) {
4650                 setattr_copy(inode, attr);
4651                 inode_inc_iversion(inode);
4652                 err = btrfs_dirty_inode(inode);
4653
4654                 if (!err && attr->ia_valid & ATTR_MODE)
4655                         err = posix_acl_chmod(inode, inode->i_mode);
4656         }
4657
4658         return err;
4659 }
4660
4661 /*
4662  * While truncating the inode pages during eviction, we get the VFS calling
4663  * btrfs_invalidatepage() against each page of the inode. This is slow because
4664  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4665  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4666  * extent_state structures over and over, wasting lots of time.
4667  *
4668  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4669  * those expensive operations on a per page basis and do only the ordered io
4670  * finishing, while we release here the extent_map and extent_state structures,
4671  * without the excessive merging and splitting.
4672  */
4673 static void evict_inode_truncate_pages(struct inode *inode)
4674 {
4675         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4676         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4677         struct rb_node *node;
4678
4679         ASSERT(inode->i_state & I_FREEING);
4680         truncate_inode_pages_final(&inode->i_data);
4681
4682         write_lock(&map_tree->lock);
4683         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4684                 struct extent_map *em;
4685
4686                 node = rb_first(&map_tree->map);
4687                 em = rb_entry(node, struct extent_map, rb_node);
4688                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4689                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4690                 remove_extent_mapping(map_tree, em);
4691                 free_extent_map(em);
4692                 if (need_resched()) {
4693                         write_unlock(&map_tree->lock);
4694                         cond_resched();
4695                         write_lock(&map_tree->lock);
4696                 }
4697         }
4698         write_unlock(&map_tree->lock);
4699
4700         spin_lock(&io_tree->lock);
4701         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4702                 struct extent_state *state;
4703                 struct extent_state *cached_state = NULL;
4704
4705                 node = rb_first(&io_tree->state);
4706                 state = rb_entry(node, struct extent_state, rb_node);
4707                 atomic_inc(&state->refs);
4708                 spin_unlock(&io_tree->lock);
4709
4710                 lock_extent_bits(io_tree, state->start, state->end,
4711                                  0, &cached_state);
4712                 clear_extent_bit(io_tree, state->start, state->end,
4713                                  EXTENT_LOCKED | EXTENT_DIRTY |
4714                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4715                                  EXTENT_DEFRAG, 1, 1,
4716                                  &cached_state, GFP_NOFS);
4717                 free_extent_state(state);
4718
4719                 cond_resched();
4720                 spin_lock(&io_tree->lock);
4721         }
4722         spin_unlock(&io_tree->lock);
4723 }
4724
4725 void btrfs_evict_inode(struct inode *inode)
4726 {
4727         struct btrfs_trans_handle *trans;
4728         struct btrfs_root *root = BTRFS_I(inode)->root;
4729         struct btrfs_block_rsv *rsv, *global_rsv;
4730         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4731         int ret;
4732
4733         trace_btrfs_inode_evict(inode);
4734
4735         evict_inode_truncate_pages(inode);
4736
4737         if (inode->i_nlink &&
4738             ((btrfs_root_refs(&root->root_item) != 0 &&
4739               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4740              btrfs_is_free_space_inode(inode)))
4741                 goto no_delete;
4742
4743         if (is_bad_inode(inode)) {
4744                 btrfs_orphan_del(NULL, inode);
4745                 goto no_delete;
4746         }
4747         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4748         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4749
4750         if (root->fs_info->log_root_recovering) {
4751                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4752                                  &BTRFS_I(inode)->runtime_flags));
4753                 goto no_delete;
4754         }
4755
4756         if (inode->i_nlink > 0) {
4757                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4758                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4759                 goto no_delete;
4760         }
4761
4762         ret = btrfs_commit_inode_delayed_inode(inode);
4763         if (ret) {
4764                 btrfs_orphan_del(NULL, inode);
4765                 goto no_delete;
4766         }
4767
4768         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4769         if (!rsv) {
4770                 btrfs_orphan_del(NULL, inode);
4771                 goto no_delete;
4772         }
4773         rsv->size = min_size;
4774         rsv->failfast = 1;
4775         global_rsv = &root->fs_info->global_block_rsv;
4776
4777         btrfs_i_size_write(inode, 0);
4778
4779         /*
4780          * This is a bit simpler than btrfs_truncate since we've already
4781          * reserved our space for our orphan item in the unlink, so we just
4782          * need to reserve some slack space in case we add bytes and update
4783          * inode item when doing the truncate.
4784          */
4785         while (1) {
4786                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4787                                              BTRFS_RESERVE_FLUSH_LIMIT);
4788
4789                 /*
4790                  * Try and steal from the global reserve since we will
4791                  * likely not use this space anyway, we want to try as
4792                  * hard as possible to get this to work.
4793                  */
4794                 if (ret)
4795                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4796
4797                 if (ret) {
4798                         btrfs_warn(root->fs_info,
4799                                 "Could not get space for a delete, will truncate on mount %d",
4800                                 ret);
4801                         btrfs_orphan_del(NULL, inode);
4802                         btrfs_free_block_rsv(root, rsv);
4803                         goto no_delete;
4804                 }
4805
4806                 trans = btrfs_join_transaction(root);
4807                 if (IS_ERR(trans)) {
4808                         btrfs_orphan_del(NULL, inode);
4809                         btrfs_free_block_rsv(root, rsv);
4810                         goto no_delete;
4811                 }
4812
4813                 trans->block_rsv = rsv;
4814
4815                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4816                 if (ret != -ENOSPC)
4817                         break;
4818
4819                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4820                 btrfs_end_transaction(trans, root);
4821                 trans = NULL;
4822                 btrfs_btree_balance_dirty(root);
4823         }
4824
4825         btrfs_free_block_rsv(root, rsv);
4826
4827         /*
4828          * Errors here aren't a big deal, it just means we leave orphan items
4829          * in the tree.  They will be cleaned up on the next mount.
4830          */
4831         if (ret == 0) {
4832                 trans->block_rsv = root->orphan_block_rsv;
4833                 btrfs_orphan_del(trans, inode);
4834         } else {
4835                 btrfs_orphan_del(NULL, inode);
4836         }
4837
4838         trans->block_rsv = &root->fs_info->trans_block_rsv;
4839         if (!(root == root->fs_info->tree_root ||
4840               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4841                 btrfs_return_ino(root, btrfs_ino(inode));
4842
4843         btrfs_end_transaction(trans, root);
4844         btrfs_btree_balance_dirty(root);
4845 no_delete:
4846         btrfs_remove_delayed_node(inode);
4847         clear_inode(inode);
4848         return;
4849 }
4850
4851 /*
4852  * this returns the key found in the dir entry in the location pointer.
4853  * If no dir entries were found, location->objectid is 0.
4854  */
4855 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4856                                struct btrfs_key *location)
4857 {
4858         const char *name = dentry->d_name.name;
4859         int namelen = dentry->d_name.len;
4860         struct btrfs_dir_item *di;
4861         struct btrfs_path *path;
4862         struct btrfs_root *root = BTRFS_I(dir)->root;
4863         int ret = 0;
4864
4865         path = btrfs_alloc_path();
4866         if (!path)
4867                 return -ENOMEM;
4868
4869         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4870                                     namelen, 0);
4871         if (IS_ERR(di))
4872                 ret = PTR_ERR(di);
4873
4874         if (IS_ERR_OR_NULL(di))
4875                 goto out_err;
4876
4877         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4878 out:
4879         btrfs_free_path(path);
4880         return ret;
4881 out_err:
4882         location->objectid = 0;
4883         goto out;
4884 }
4885
4886 /*
4887  * when we hit a tree root in a directory, the btrfs part of the inode
4888  * needs to be changed to reflect the root directory of the tree root.  This
4889  * is kind of like crossing a mount point.
4890  */
4891 static int fixup_tree_root_location(struct btrfs_root *root,
4892                                     struct inode *dir,
4893                                     struct dentry *dentry,
4894                                     struct btrfs_key *location,
4895                                     struct btrfs_root **sub_root)
4896 {
4897         struct btrfs_path *path;
4898         struct btrfs_root *new_root;
4899         struct btrfs_root_ref *ref;
4900         struct extent_buffer *leaf;
4901         int ret;
4902         int err = 0;
4903
4904         path = btrfs_alloc_path();
4905         if (!path) {
4906                 err = -ENOMEM;
4907                 goto out;
4908         }
4909
4910         err = -ENOENT;
4911         ret = btrfs_find_item(root->fs_info->tree_root, path,
4912                                 BTRFS_I(dir)->root->root_key.objectid,
4913                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4914         if (ret) {
4915                 if (ret < 0)
4916                         err = ret;
4917                 goto out;
4918         }
4919
4920         leaf = path->nodes[0];
4921         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4922         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4923             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4924                 goto out;
4925
4926         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4927                                    (unsigned long)(ref + 1),
4928                                    dentry->d_name.len);
4929         if (ret)
4930                 goto out;
4931
4932         btrfs_release_path(path);
4933
4934         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4935         if (IS_ERR(new_root)) {
4936                 err = PTR_ERR(new_root);
4937                 goto out;
4938         }
4939
4940         *sub_root = new_root;
4941         location->objectid = btrfs_root_dirid(&new_root->root_item);
4942         location->type = BTRFS_INODE_ITEM_KEY;
4943         location->offset = 0;
4944         err = 0;
4945 out:
4946         btrfs_free_path(path);
4947         return err;
4948 }
4949
4950 static void inode_tree_add(struct inode *inode)
4951 {
4952         struct btrfs_root *root = BTRFS_I(inode)->root;
4953         struct btrfs_inode *entry;
4954         struct rb_node **p;
4955         struct rb_node *parent;
4956         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4957         u64 ino = btrfs_ino(inode);
4958
4959         if (inode_unhashed(inode))
4960                 return;
4961         parent = NULL;
4962         spin_lock(&root->inode_lock);
4963         p = &root->inode_tree.rb_node;
4964         while (*p) {
4965                 parent = *p;
4966                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4967
4968                 if (ino < btrfs_ino(&entry->vfs_inode))
4969                         p = &parent->rb_left;
4970                 else if (ino > btrfs_ino(&entry->vfs_inode))
4971                         p = &parent->rb_right;
4972                 else {
4973                         WARN_ON(!(entry->vfs_inode.i_state &
4974                                   (I_WILL_FREE | I_FREEING)));
4975                         rb_replace_node(parent, new, &root->inode_tree);
4976                         RB_CLEAR_NODE(parent);
4977                         spin_unlock(&root->inode_lock);
4978                         return;
4979                 }
4980         }
4981         rb_link_node(new, parent, p);
4982         rb_insert_color(new, &root->inode_tree);
4983         spin_unlock(&root->inode_lock);
4984 }
4985
4986 static void inode_tree_del(struct inode *inode)
4987 {
4988         struct btrfs_root *root = BTRFS_I(inode)->root;
4989         int empty = 0;
4990
4991         spin_lock(&root->inode_lock);
4992         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4993                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4994                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4995                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4996         }
4997         spin_unlock(&root->inode_lock);
4998
4999         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5000                 synchronize_srcu(&root->fs_info->subvol_srcu);
5001                 spin_lock(&root->inode_lock);
5002                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5003                 spin_unlock(&root->inode_lock);
5004                 if (empty)
5005                         btrfs_add_dead_root(root);
5006         }
5007 }
5008
5009 void btrfs_invalidate_inodes(struct btrfs_root *root)
5010 {
5011         struct rb_node *node;
5012         struct rb_node *prev;
5013         struct btrfs_inode *entry;
5014         struct inode *inode;
5015         u64 objectid = 0;
5016
5017         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5018                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5019
5020         spin_lock(&root->inode_lock);
5021 again:
5022         node = root->inode_tree.rb_node;
5023         prev = NULL;
5024         while (node) {
5025                 prev = node;
5026                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5027
5028                 if (objectid < btrfs_ino(&entry->vfs_inode))
5029                         node = node->rb_left;
5030                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5031                         node = node->rb_right;
5032                 else
5033                         break;
5034         }
5035         if (!node) {
5036                 while (prev) {
5037                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5038                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5039                                 node = prev;
5040                                 break;
5041                         }
5042                         prev = rb_next(prev);
5043                 }
5044         }
5045         while (node) {
5046                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5047                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5048                 inode = igrab(&entry->vfs_inode);
5049                 if (inode) {
5050                         spin_unlock(&root->inode_lock);
5051                         if (atomic_read(&inode->i_count) > 1)
5052                                 d_prune_aliases(inode);
5053                         /*
5054                          * btrfs_drop_inode will have it removed from
5055                          * the inode cache when its usage count
5056                          * hits zero.
5057                          */
5058                         iput(inode);
5059                         cond_resched();
5060                         spin_lock(&root->inode_lock);
5061                         goto again;
5062                 }
5063
5064                 if (cond_resched_lock(&root->inode_lock))
5065                         goto again;
5066
5067                 node = rb_next(node);
5068         }
5069         spin_unlock(&root->inode_lock);
5070 }
5071
5072 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5073 {
5074         struct btrfs_iget_args *args = p;
5075         inode->i_ino = args->location->objectid;
5076         memcpy(&BTRFS_I(inode)->location, args->location,
5077                sizeof(*args->location));
5078         BTRFS_I(inode)->root = args->root;
5079         return 0;
5080 }
5081
5082 static int btrfs_find_actor(struct inode *inode, void *opaque)
5083 {
5084         struct btrfs_iget_args *args = opaque;
5085         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5086                 args->root == BTRFS_I(inode)->root;
5087 }
5088
5089 static struct inode *btrfs_iget_locked(struct super_block *s,
5090                                        struct btrfs_key *location,
5091                                        struct btrfs_root *root)
5092 {
5093         struct inode *inode;
5094         struct btrfs_iget_args args;
5095         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5096
5097         args.location = location;
5098         args.root = root;
5099
5100         inode = iget5_locked(s, hashval, btrfs_find_actor,
5101                              btrfs_init_locked_inode,
5102                              (void *)&args);
5103         return inode;
5104 }
5105
5106 /* Get an inode object given its location and corresponding root.
5107  * Returns in *is_new if the inode was read from disk
5108  */
5109 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5110                          struct btrfs_root *root, int *new)
5111 {
5112         struct inode *inode;
5113
5114         inode = btrfs_iget_locked(s, location, root);
5115         if (!inode)
5116                 return ERR_PTR(-ENOMEM);
5117
5118         if (inode->i_state & I_NEW) {
5119                 btrfs_read_locked_inode(inode);
5120                 if (!is_bad_inode(inode)) {
5121                         inode_tree_add(inode);
5122                         unlock_new_inode(inode);
5123                         if (new)
5124                                 *new = 1;
5125                 } else {
5126                         unlock_new_inode(inode);
5127                         iput(inode);
5128                         inode = ERR_PTR(-ESTALE);
5129                 }
5130         }
5131
5132         return inode;
5133 }
5134
5135 static struct inode *new_simple_dir(struct super_block *s,
5136                                     struct btrfs_key *key,
5137                                     struct btrfs_root *root)
5138 {
5139         struct inode *inode = new_inode(s);
5140
5141         if (!inode)
5142                 return ERR_PTR(-ENOMEM);
5143
5144         BTRFS_I(inode)->root = root;
5145         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5146         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5147
5148         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5149         inode->i_op = &btrfs_dir_ro_inode_operations;
5150         inode->i_fop = &simple_dir_operations;
5151         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5152         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5153
5154         return inode;
5155 }
5156
5157 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5158 {
5159         struct inode *inode;
5160         struct btrfs_root *root = BTRFS_I(dir)->root;
5161         struct btrfs_root *sub_root = root;
5162         struct btrfs_key location;
5163         int index;
5164         int ret = 0;
5165
5166         if (dentry->d_name.len > BTRFS_NAME_LEN)
5167                 return ERR_PTR(-ENAMETOOLONG);
5168
5169         ret = btrfs_inode_by_name(dir, dentry, &location);
5170         if (ret < 0)
5171                 return ERR_PTR(ret);
5172
5173         if (location.objectid == 0)
5174                 return ERR_PTR(-ENOENT);
5175
5176         if (location.type == BTRFS_INODE_ITEM_KEY) {
5177                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5178                 return inode;
5179         }
5180
5181         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5182
5183         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5184         ret = fixup_tree_root_location(root, dir, dentry,
5185                                        &location, &sub_root);
5186         if (ret < 0) {
5187                 if (ret != -ENOENT)
5188                         inode = ERR_PTR(ret);
5189                 else
5190                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5191         } else {
5192                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5193         }
5194         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5195
5196         if (!IS_ERR(inode) && root != sub_root) {
5197                 down_read(&root->fs_info->cleanup_work_sem);
5198                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5199                         ret = btrfs_orphan_cleanup(sub_root);
5200                 up_read(&root->fs_info->cleanup_work_sem);
5201                 if (ret) {
5202                         iput(inode);
5203                         inode = ERR_PTR(ret);
5204                 }
5205                 /*
5206                  * If orphan cleanup did remove any orphans, it means the tree
5207                  * was modified and therefore the commit root is not the same as
5208                  * the current root anymore. This is a problem, because send
5209                  * uses the commit root and therefore can see inode items that
5210                  * don't exist in the current root anymore, and for example make
5211                  * calls to btrfs_iget, which will do tree lookups based on the
5212                  * current root and not on the commit root. Those lookups will
5213                  * fail, returning a -ESTALE error, and making send fail with
5214                  * that error. So make sure a send does not see any orphans we
5215                  * have just removed, and that it will see the same inodes
5216                  * regardless of whether a transaction commit happened before
5217                  * it started (meaning that the commit root will be the same as
5218                  * the current root) or not.
5219                  */
5220                 if (sub_root->node != sub_root->commit_root) {
5221                         u64 sub_flags = btrfs_root_flags(&sub_root->root_item);
5222
5223                         if (sub_flags & BTRFS_ROOT_SUBVOL_RDONLY) {
5224                                 struct extent_buffer *eb;
5225
5226                                 /*
5227                                  * Assert we can't have races between dentry
5228                                  * lookup called through the snapshot creation
5229                                  * ioctl and the VFS.
5230                                  */
5231                                 ASSERT(mutex_is_locked(&dir->i_mutex));
5232
5233                                 down_write(&root->fs_info->commit_root_sem);
5234                                 eb = sub_root->commit_root;
5235                                 sub_root->commit_root =
5236                                         btrfs_root_node(sub_root);
5237                                 up_write(&root->fs_info->commit_root_sem);
5238                                 free_extent_buffer(eb);
5239                         }
5240                 }
5241         }
5242
5243         return inode;
5244 }
5245
5246 static int btrfs_dentry_delete(const struct dentry *dentry)
5247 {
5248         struct btrfs_root *root;
5249         struct inode *inode = dentry->d_inode;
5250
5251         if (!inode && !IS_ROOT(dentry))
5252                 inode = dentry->d_parent->d_inode;
5253
5254         if (inode) {
5255                 root = BTRFS_I(inode)->root;
5256                 if (btrfs_root_refs(&root->root_item) == 0)
5257                         return 1;
5258
5259                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5260                         return 1;
5261         }
5262         return 0;
5263 }
5264
5265 static void btrfs_dentry_release(struct dentry *dentry)
5266 {
5267         kfree(dentry->d_fsdata);
5268 }
5269
5270 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5271                                    unsigned int flags)
5272 {
5273         struct inode *inode;
5274
5275         inode = btrfs_lookup_dentry(dir, dentry);
5276         if (IS_ERR(inode)) {
5277                 if (PTR_ERR(inode) == -ENOENT)
5278                         inode = NULL;
5279                 else
5280                         return ERR_CAST(inode);
5281         }
5282
5283         return d_materialise_unique(dentry, inode);
5284 }
5285
5286 unsigned char btrfs_filetype_table[] = {
5287         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5288 };
5289
5290 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5291 {
5292         struct inode *inode = file_inode(file);
5293         struct btrfs_root *root = BTRFS_I(inode)->root;
5294         struct btrfs_item *item;
5295         struct btrfs_dir_item *di;
5296         struct btrfs_key key;
5297         struct btrfs_key found_key;
5298         struct btrfs_path *path;
5299         struct list_head ins_list;
5300         struct list_head del_list;
5301         int ret;
5302         struct extent_buffer *leaf;
5303         int slot;
5304         unsigned char d_type;
5305         int over = 0;
5306         u32 di_cur;
5307         u32 di_total;
5308         u32 di_len;
5309         int key_type = BTRFS_DIR_INDEX_KEY;
5310         char tmp_name[32];
5311         char *name_ptr;
5312         int name_len;
5313         int is_curr = 0;        /* ctx->pos points to the current index? */
5314
5315         /* FIXME, use a real flag for deciding about the key type */
5316         if (root->fs_info->tree_root == root)
5317                 key_type = BTRFS_DIR_ITEM_KEY;
5318
5319         if (!dir_emit_dots(file, ctx))
5320                 return 0;
5321
5322         path = btrfs_alloc_path();
5323         if (!path)
5324                 return -ENOMEM;
5325
5326         path->reada = 1;
5327
5328         if (key_type == BTRFS_DIR_INDEX_KEY) {
5329                 INIT_LIST_HEAD(&ins_list);
5330                 INIT_LIST_HEAD(&del_list);
5331                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5332         }
5333
5334         btrfs_set_key_type(&key, key_type);
5335         key.offset = ctx->pos;
5336         key.objectid = btrfs_ino(inode);
5337
5338         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5339         if (ret < 0)
5340                 goto err;
5341
5342         while (1) {
5343                 leaf = path->nodes[0];
5344                 slot = path->slots[0];
5345                 if (slot >= btrfs_header_nritems(leaf)) {
5346                         ret = btrfs_next_leaf(root, path);
5347                         if (ret < 0)
5348                                 goto err;
5349                         else if (ret > 0)
5350                                 break;
5351                         continue;
5352                 }
5353
5354                 item = btrfs_item_nr(slot);
5355                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5356
5357                 if (found_key.objectid != key.objectid)
5358                         break;
5359                 if (btrfs_key_type(&found_key) != key_type)
5360                         break;
5361                 if (found_key.offset < ctx->pos)
5362                         goto next;
5363                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5364                     btrfs_should_delete_dir_index(&del_list,
5365                                                   found_key.offset))
5366                         goto next;
5367
5368                 ctx->pos = found_key.offset;
5369                 is_curr = 1;
5370
5371                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5372                 di_cur = 0;
5373                 di_total = btrfs_item_size(leaf, item);
5374
5375                 while (di_cur < di_total) {
5376                         struct btrfs_key location;
5377
5378                         if (verify_dir_item(root, leaf, di))
5379                                 break;
5380
5381                         name_len = btrfs_dir_name_len(leaf, di);
5382                         if (name_len <= sizeof(tmp_name)) {
5383                                 name_ptr = tmp_name;
5384                         } else {
5385                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5386                                 if (!name_ptr) {
5387                                         ret = -ENOMEM;
5388                                         goto err;
5389                                 }
5390                         }
5391                         read_extent_buffer(leaf, name_ptr,
5392                                            (unsigned long)(di + 1), name_len);
5393
5394                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5395                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5396
5397
5398                         /* is this a reference to our own snapshot? If so
5399                          * skip it.
5400                          *
5401                          * In contrast to old kernels, we insert the snapshot's
5402                          * dir item and dir index after it has been created, so
5403                          * we won't find a reference to our own snapshot. We
5404                          * still keep the following code for backward
5405                          * compatibility.
5406                          */
5407                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5408                             location.objectid == root->root_key.objectid) {
5409                                 over = 0;
5410                                 goto skip;
5411                         }
5412                         over = !dir_emit(ctx, name_ptr, name_len,
5413                                        location.objectid, d_type);
5414
5415 skip:
5416                         if (name_ptr != tmp_name)
5417                                 kfree(name_ptr);
5418
5419                         if (over)
5420                                 goto nopos;
5421                         di_len = btrfs_dir_name_len(leaf, di) +
5422                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5423                         di_cur += di_len;
5424                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5425                 }
5426 next:
5427                 path->slots[0]++;
5428         }
5429
5430         if (key_type == BTRFS_DIR_INDEX_KEY) {
5431                 if (is_curr)
5432                         ctx->pos++;
5433                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5434                 if (ret)
5435                         goto nopos;
5436         }
5437
5438         /* Reached end of directory/root. Bump pos past the last item. */
5439         ctx->pos++;
5440
5441         /*
5442          * Stop new entries from being returned after we return the last
5443          * entry.
5444          *
5445          * New directory entries are assigned a strictly increasing
5446          * offset.  This means that new entries created during readdir
5447          * are *guaranteed* to be seen in the future by that readdir.
5448          * This has broken buggy programs which operate on names as
5449          * they're returned by readdir.  Until we re-use freed offsets
5450          * we have this hack to stop new entries from being returned
5451          * under the assumption that they'll never reach this huge
5452          * offset.
5453          *
5454          * This is being careful not to overflow 32bit loff_t unless the
5455          * last entry requires it because doing so has broken 32bit apps
5456          * in the past.
5457          */
5458         if (key_type == BTRFS_DIR_INDEX_KEY) {
5459                 if (ctx->pos >= INT_MAX)
5460                         ctx->pos = LLONG_MAX;
5461                 else
5462                         ctx->pos = INT_MAX;
5463         }
5464 nopos:
5465         ret = 0;
5466 err:
5467         if (key_type == BTRFS_DIR_INDEX_KEY)
5468                 btrfs_put_delayed_items(&ins_list, &del_list);
5469         btrfs_free_path(path);
5470         return ret;
5471 }
5472
5473 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5474 {
5475         struct btrfs_root *root = BTRFS_I(inode)->root;
5476         struct btrfs_trans_handle *trans;
5477         int ret = 0;
5478         bool nolock = false;
5479
5480         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5481                 return 0;
5482
5483         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5484                 nolock = true;
5485
5486         if (wbc->sync_mode == WB_SYNC_ALL) {
5487                 if (nolock)
5488                         trans = btrfs_join_transaction_nolock(root);
5489                 else
5490                         trans = btrfs_join_transaction(root);
5491                 if (IS_ERR(trans))
5492                         return PTR_ERR(trans);
5493                 ret = btrfs_commit_transaction(trans, root);
5494         }
5495         return ret;
5496 }
5497
5498 /*
5499  * This is somewhat expensive, updating the tree every time the
5500  * inode changes.  But, it is most likely to find the inode in cache.
5501  * FIXME, needs more benchmarking...there are no reasons other than performance
5502  * to keep or drop this code.
5503  */
5504 static int btrfs_dirty_inode(struct inode *inode)
5505 {
5506         struct btrfs_root *root = BTRFS_I(inode)->root;
5507         struct btrfs_trans_handle *trans;
5508         int ret;
5509
5510         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5511                 return 0;
5512
5513         trans = btrfs_join_transaction(root);
5514         if (IS_ERR(trans))
5515                 return PTR_ERR(trans);
5516
5517         ret = btrfs_update_inode(trans, root, inode);
5518         if (ret && ret == -ENOSPC) {
5519                 /* whoops, lets try again with the full transaction */
5520                 btrfs_end_transaction(trans, root);
5521                 trans = btrfs_start_transaction(root, 1);
5522                 if (IS_ERR(trans))
5523                         return PTR_ERR(trans);
5524
5525                 ret = btrfs_update_inode(trans, root, inode);
5526         }
5527         btrfs_end_transaction(trans, root);
5528         if (BTRFS_I(inode)->delayed_node)
5529                 btrfs_balance_delayed_items(root);
5530
5531         return ret;
5532 }
5533
5534 /*
5535  * This is a copy of file_update_time.  We need this so we can return error on
5536  * ENOSPC for updating the inode in the case of file write and mmap writes.
5537  */
5538 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5539                              int flags)
5540 {
5541         struct btrfs_root *root = BTRFS_I(inode)->root;
5542
5543         if (btrfs_root_readonly(root))
5544                 return -EROFS;
5545
5546         if (flags & S_VERSION)
5547                 inode_inc_iversion(inode);
5548         if (flags & S_CTIME)
5549                 inode->i_ctime = *now;
5550         if (flags & S_MTIME)
5551                 inode->i_mtime = *now;
5552         if (flags & S_ATIME)
5553                 inode->i_atime = *now;
5554         return btrfs_dirty_inode(inode);
5555 }
5556
5557 /*
5558  * find the highest existing sequence number in a directory
5559  * and then set the in-memory index_cnt variable to reflect
5560  * free sequence numbers
5561  */
5562 static int btrfs_set_inode_index_count(struct inode *inode)
5563 {
5564         struct btrfs_root *root = BTRFS_I(inode)->root;
5565         struct btrfs_key key, found_key;
5566         struct btrfs_path *path;
5567         struct extent_buffer *leaf;
5568         int ret;
5569
5570         key.objectid = btrfs_ino(inode);
5571         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5572         key.offset = (u64)-1;
5573
5574         path = btrfs_alloc_path();
5575         if (!path)
5576                 return -ENOMEM;
5577
5578         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5579         if (ret < 0)
5580                 goto out;
5581         /* FIXME: we should be able to handle this */
5582         if (ret == 0)
5583                 goto out;
5584         ret = 0;
5585
5586         /*
5587          * MAGIC NUMBER EXPLANATION:
5588          * since we search a directory based on f_pos we have to start at 2
5589          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5590          * else has to start at 2
5591          */
5592         if (path->slots[0] == 0) {
5593                 BTRFS_I(inode)->index_cnt = 2;
5594                 goto out;
5595         }
5596
5597         path->slots[0]--;
5598
5599         leaf = path->nodes[0];
5600         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5601
5602         if (found_key.objectid != btrfs_ino(inode) ||
5603             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5604                 BTRFS_I(inode)->index_cnt = 2;
5605                 goto out;
5606         }
5607
5608         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5609 out:
5610         btrfs_free_path(path);
5611         return ret;
5612 }
5613
5614 /*
5615  * helper to find a free sequence number in a given directory.  This current
5616  * code is very simple, later versions will do smarter things in the btree
5617  */
5618 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5619 {
5620         int ret = 0;
5621
5622         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5623                 ret = btrfs_inode_delayed_dir_index_count(dir);
5624                 if (ret) {
5625                         ret = btrfs_set_inode_index_count(dir);
5626                         if (ret)
5627                                 return ret;
5628                 }
5629         }
5630
5631         *index = BTRFS_I(dir)->index_cnt;
5632         BTRFS_I(dir)->index_cnt++;
5633
5634         return ret;
5635 }
5636
5637 static int btrfs_insert_inode_locked(struct inode *inode)
5638 {
5639         struct btrfs_iget_args args;
5640         args.location = &BTRFS_I(inode)->location;
5641         args.root = BTRFS_I(inode)->root;
5642
5643         return insert_inode_locked4(inode,
5644                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5645                    btrfs_find_actor, &args);
5646 }
5647
5648 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5649                                      struct btrfs_root *root,
5650                                      struct inode *dir,
5651                                      const char *name, int name_len,
5652                                      u64 ref_objectid, u64 objectid,
5653                                      umode_t mode, u64 *index)
5654 {
5655         struct inode *inode;
5656         struct btrfs_inode_item *inode_item;
5657         struct btrfs_key *location;
5658         struct btrfs_path *path;
5659         struct btrfs_inode_ref *ref;
5660         struct btrfs_key key[2];
5661         u32 sizes[2];
5662         int nitems = name ? 2 : 1;
5663         unsigned long ptr;
5664         int ret;
5665
5666         path = btrfs_alloc_path();
5667         if (!path)
5668                 return ERR_PTR(-ENOMEM);
5669
5670         inode = new_inode(root->fs_info->sb);
5671         if (!inode) {
5672                 btrfs_free_path(path);
5673                 return ERR_PTR(-ENOMEM);
5674         }
5675
5676         /*
5677          * O_TMPFILE, set link count to 0, so that after this point,
5678          * we fill in an inode item with the correct link count.
5679          */
5680         if (!name)
5681                 set_nlink(inode, 0);
5682
5683         /*
5684          * we have to initialize this early, so we can reclaim the inode
5685          * number if we fail afterwards in this function.
5686          */
5687         inode->i_ino = objectid;
5688
5689         if (dir && name) {
5690                 trace_btrfs_inode_request(dir);
5691
5692                 ret = btrfs_set_inode_index(dir, index);
5693                 if (ret) {
5694                         btrfs_free_path(path);
5695                         iput(inode);
5696                         return ERR_PTR(ret);
5697                 }
5698         } else if (dir) {
5699                 *index = 0;
5700         }
5701         /*
5702          * index_cnt is ignored for everything but a dir,
5703          * btrfs_get_inode_index_count has an explanation for the magic
5704          * number
5705          */
5706         BTRFS_I(inode)->index_cnt = 2;
5707         BTRFS_I(inode)->dir_index = *index;
5708         BTRFS_I(inode)->root = root;
5709         BTRFS_I(inode)->generation = trans->transid;
5710         inode->i_generation = BTRFS_I(inode)->generation;
5711
5712         /*
5713          * We could have gotten an inode number from somebody who was fsynced
5714          * and then removed in this same transaction, so let's just set full
5715          * sync since it will be a full sync anyway and this will blow away the
5716          * old info in the log.
5717          */
5718         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5719
5720         key[0].objectid = objectid;
5721         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5722         key[0].offset = 0;
5723
5724         sizes[0] = sizeof(struct btrfs_inode_item);
5725
5726         if (name) {
5727                 /*
5728                  * Start new inodes with an inode_ref. This is slightly more
5729                  * efficient for small numbers of hard links since they will
5730                  * be packed into one item. Extended refs will kick in if we
5731                  * add more hard links than can fit in the ref item.
5732                  */
5733                 key[1].objectid = objectid;
5734                 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5735                 key[1].offset = ref_objectid;
5736
5737                 sizes[1] = name_len + sizeof(*ref);
5738         }
5739
5740         location = &BTRFS_I(inode)->location;
5741         location->objectid = objectid;
5742         location->offset = 0;
5743         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5744
5745         ret = btrfs_insert_inode_locked(inode);
5746         if (ret < 0)
5747                 goto fail;
5748
5749         path->leave_spinning = 1;
5750         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5751         if (ret != 0)
5752                 goto fail_unlock;
5753
5754         inode_init_owner(inode, dir, mode);
5755         inode_set_bytes(inode, 0);
5756         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5757         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5758                                   struct btrfs_inode_item);
5759         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5760                              sizeof(*inode_item));
5761         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5762
5763         if (name) {
5764                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5765                                      struct btrfs_inode_ref);
5766                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5767                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5768                 ptr = (unsigned long)(ref + 1);
5769                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5770         }
5771
5772         btrfs_mark_buffer_dirty(path->nodes[0]);
5773         btrfs_free_path(path);
5774
5775         btrfs_inherit_iflags(inode, dir);
5776
5777         if (S_ISREG(mode)) {
5778                 if (btrfs_test_opt(root, NODATASUM))
5779                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5780                 if (btrfs_test_opt(root, NODATACOW))
5781                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5782                                 BTRFS_INODE_NODATASUM;
5783         }
5784
5785         inode_tree_add(inode);
5786
5787         trace_btrfs_inode_new(inode);
5788         btrfs_set_inode_last_trans(trans, inode);
5789
5790         btrfs_update_root_times(trans, root);
5791
5792         ret = btrfs_inode_inherit_props(trans, inode, dir);
5793         if (ret)
5794                 btrfs_err(root->fs_info,
5795                           "error inheriting props for ino %llu (root %llu): %d",
5796                           btrfs_ino(inode), root->root_key.objectid, ret);
5797
5798         return inode;
5799
5800 fail_unlock:
5801         unlock_new_inode(inode);
5802 fail:
5803         if (dir && name)
5804                 BTRFS_I(dir)->index_cnt--;
5805         btrfs_free_path(path);
5806         iput(inode);
5807         return ERR_PTR(ret);
5808 }
5809
5810 static inline u8 btrfs_inode_type(struct inode *inode)
5811 {
5812         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5813 }
5814
5815 /*
5816  * utility function to add 'inode' into 'parent_inode' with
5817  * a give name and a given sequence number.
5818  * if 'add_backref' is true, also insert a backref from the
5819  * inode to the parent directory.
5820  */
5821 int btrfs_add_link(struct btrfs_trans_handle *trans,
5822                    struct inode *parent_inode, struct inode *inode,
5823                    const char *name, int name_len, int add_backref, u64 index)
5824 {
5825         int ret = 0;
5826         struct btrfs_key key;
5827         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5828         u64 ino = btrfs_ino(inode);
5829         u64 parent_ino = btrfs_ino(parent_inode);
5830
5831         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5832                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5833         } else {
5834                 key.objectid = ino;
5835                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5836                 key.offset = 0;
5837         }
5838
5839         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5840                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5841                                          key.objectid, root->root_key.objectid,
5842                                          parent_ino, index, name, name_len);
5843         } else if (add_backref) {
5844                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5845                                              parent_ino, index);
5846         }
5847
5848         /* Nothing to clean up yet */
5849         if (ret)
5850                 return ret;
5851
5852         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5853                                     parent_inode, &key,
5854                                     btrfs_inode_type(inode), index);
5855         if (ret == -EEXIST || ret == -EOVERFLOW)
5856                 goto fail_dir_item;
5857         else if (ret) {
5858                 btrfs_abort_transaction(trans, root, ret);
5859                 return ret;
5860         }
5861
5862         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5863                            name_len * 2);
5864         inode_inc_iversion(parent_inode);
5865         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5866         ret = btrfs_update_inode(trans, root, parent_inode);
5867         if (ret)
5868                 btrfs_abort_transaction(trans, root, ret);
5869         return ret;
5870
5871 fail_dir_item:
5872         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5873                 u64 local_index;
5874                 int err;
5875                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5876                                  key.objectid, root->root_key.objectid,
5877                                  parent_ino, &local_index, name, name_len);
5878
5879         } else if (add_backref) {
5880                 u64 local_index;
5881                 int err;
5882
5883                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5884                                           ino, parent_ino, &local_index);
5885         }
5886         return ret;
5887 }
5888
5889 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5890                             struct inode *dir, struct dentry *dentry,
5891                             struct inode *inode, int backref, u64 index)
5892 {
5893         int err = btrfs_add_link(trans, dir, inode,
5894                                  dentry->d_name.name, dentry->d_name.len,
5895                                  backref, index);
5896         if (err > 0)
5897                 err = -EEXIST;
5898         return err;
5899 }
5900
5901 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5902                         umode_t mode, dev_t rdev)
5903 {
5904         struct btrfs_trans_handle *trans;
5905         struct btrfs_root *root = BTRFS_I(dir)->root;
5906         struct inode *inode = NULL;
5907         int err;
5908         int drop_inode = 0;
5909         u64 objectid;
5910         u64 index = 0;
5911
5912         if (!new_valid_dev(rdev))
5913                 return -EINVAL;
5914
5915         /*
5916          * 2 for inode item and ref
5917          * 2 for dir items
5918          * 1 for xattr if selinux is on
5919          */
5920         trans = btrfs_start_transaction(root, 5);
5921         if (IS_ERR(trans))
5922                 return PTR_ERR(trans);
5923
5924         err = btrfs_find_free_ino(root, &objectid);
5925         if (err)
5926                 goto out_unlock;
5927
5928         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5929                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5930                                 mode, &index);
5931         if (IS_ERR(inode)) {
5932                 err = PTR_ERR(inode);
5933                 goto out_unlock;
5934         }
5935
5936         /*
5937         * If the active LSM wants to access the inode during
5938         * d_instantiate it needs these. Smack checks to see
5939         * if the filesystem supports xattrs by looking at the
5940         * ops vector.
5941         */
5942         inode->i_op = &btrfs_special_inode_operations;
5943         init_special_inode(inode, inode->i_mode, rdev);
5944
5945         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5946         if (err)
5947                 goto out_unlock_inode;
5948
5949         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5950         if (err) {
5951                 goto out_unlock_inode;
5952         } else {
5953                 btrfs_update_inode(trans, root, inode);
5954                 unlock_new_inode(inode);
5955                 d_instantiate(dentry, inode);
5956         }
5957
5958 out_unlock:
5959         btrfs_end_transaction(trans, root);
5960         btrfs_balance_delayed_items(root);
5961         btrfs_btree_balance_dirty(root);
5962         if (drop_inode) {
5963                 inode_dec_link_count(inode);
5964                 iput(inode);
5965         }
5966         return err;
5967
5968 out_unlock_inode:
5969         drop_inode = 1;
5970         unlock_new_inode(inode);
5971         goto out_unlock;
5972
5973 }
5974
5975 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5976                         umode_t mode, bool excl)
5977 {
5978         struct btrfs_trans_handle *trans;
5979         struct btrfs_root *root = BTRFS_I(dir)->root;
5980         struct inode *inode = NULL;
5981         int drop_inode_on_err = 0;
5982         int err;
5983         u64 objectid;
5984         u64 index = 0;
5985
5986         /*
5987          * 2 for inode item and ref
5988          * 2 for dir items
5989          * 1 for xattr if selinux is on
5990          */
5991         trans = btrfs_start_transaction(root, 5);
5992         if (IS_ERR(trans))
5993                 return PTR_ERR(trans);
5994
5995         err = btrfs_find_free_ino(root, &objectid);
5996         if (err)
5997                 goto out_unlock;
5998
5999         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6000                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6001                                 mode, &index);
6002         if (IS_ERR(inode)) {
6003                 err = PTR_ERR(inode);
6004                 goto out_unlock;
6005         }
6006         drop_inode_on_err = 1;
6007         /*
6008         * If the active LSM wants to access the inode during
6009         * d_instantiate it needs these. Smack checks to see
6010         * if the filesystem supports xattrs by looking at the
6011         * ops vector.
6012         */
6013         inode->i_fop = &btrfs_file_operations;
6014         inode->i_op = &btrfs_file_inode_operations;
6015         inode->i_mapping->a_ops = &btrfs_aops;
6016         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6017
6018         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6019         if (err)
6020                 goto out_unlock_inode;
6021
6022         err = btrfs_update_inode(trans, root, inode);
6023         if (err)
6024                 goto out_unlock_inode;
6025
6026         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6027         if (err)
6028                 goto out_unlock_inode;
6029
6030         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6031         unlock_new_inode(inode);
6032         d_instantiate(dentry, inode);
6033
6034 out_unlock:
6035         btrfs_end_transaction(trans, root);
6036         if (err && drop_inode_on_err) {
6037                 inode_dec_link_count(inode);
6038                 iput(inode);
6039         }
6040         btrfs_balance_delayed_items(root);
6041         btrfs_btree_balance_dirty(root);
6042         return err;
6043
6044 out_unlock_inode:
6045         unlock_new_inode(inode);
6046         goto out_unlock;
6047
6048 }
6049
6050 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6051                       struct dentry *dentry)
6052 {
6053         struct btrfs_trans_handle *trans;
6054         struct btrfs_root *root = BTRFS_I(dir)->root;
6055         struct inode *inode = old_dentry->d_inode;
6056         u64 index;
6057         int err;
6058         int drop_inode = 0;
6059
6060         /* do not allow sys_link's with other subvols of the same device */
6061         if (root->objectid != BTRFS_I(inode)->root->objectid)
6062                 return -EXDEV;
6063
6064         if (inode->i_nlink >= BTRFS_LINK_MAX)
6065                 return -EMLINK;
6066
6067         err = btrfs_set_inode_index(dir, &index);
6068         if (err)
6069                 goto fail;
6070
6071         /*
6072          * 2 items for inode and inode ref
6073          * 2 items for dir items
6074          * 1 item for parent inode
6075          */
6076         trans = btrfs_start_transaction(root, 5);
6077         if (IS_ERR(trans)) {
6078                 err = PTR_ERR(trans);
6079                 goto fail;
6080         }
6081
6082         /* There are several dir indexes for this inode, clear the cache. */
6083         BTRFS_I(inode)->dir_index = 0ULL;
6084         inc_nlink(inode);
6085         inode_inc_iversion(inode);
6086         inode->i_ctime = CURRENT_TIME;
6087         ihold(inode);
6088         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6089
6090         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6091
6092         if (err) {
6093                 drop_inode = 1;
6094         } else {
6095                 struct dentry *parent = dentry->d_parent;
6096                 err = btrfs_update_inode(trans, root, inode);
6097                 if (err)
6098                         goto fail;
6099                 if (inode->i_nlink == 1) {
6100                         /*
6101                          * If new hard link count is 1, it's a file created
6102                          * with open(2) O_TMPFILE flag.
6103                          */
6104                         err = btrfs_orphan_del(trans, inode);
6105                         if (err)
6106                                 goto fail;
6107                 }
6108                 d_instantiate(dentry, inode);
6109                 btrfs_log_new_name(trans, inode, NULL, parent);
6110         }
6111
6112         btrfs_end_transaction(trans, root);
6113         btrfs_balance_delayed_items(root);
6114 fail:
6115         if (drop_inode) {
6116                 inode_dec_link_count(inode);
6117                 iput(inode);
6118         }
6119         btrfs_btree_balance_dirty(root);
6120         return err;
6121 }
6122
6123 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6124 {
6125         struct inode *inode = NULL;
6126         struct btrfs_trans_handle *trans;
6127         struct btrfs_root *root = BTRFS_I(dir)->root;
6128         int err = 0;
6129         int drop_on_err = 0;
6130         u64 objectid = 0;
6131         u64 index = 0;
6132
6133         /*
6134          * 2 items for inode and ref
6135          * 2 items for dir items
6136          * 1 for xattr if selinux is on
6137          */
6138         trans = btrfs_start_transaction(root, 5);
6139         if (IS_ERR(trans))
6140                 return PTR_ERR(trans);
6141
6142         err = btrfs_find_free_ino(root, &objectid);
6143         if (err)
6144                 goto out_fail;
6145
6146         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6147                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6148                                 S_IFDIR | mode, &index);
6149         if (IS_ERR(inode)) {
6150                 err = PTR_ERR(inode);
6151                 goto out_fail;
6152         }
6153
6154         drop_on_err = 1;
6155         /* these must be set before we unlock the inode */
6156         inode->i_op = &btrfs_dir_inode_operations;
6157         inode->i_fop = &btrfs_dir_file_operations;
6158
6159         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6160         if (err)
6161                 goto out_fail_inode;
6162
6163         btrfs_i_size_write(inode, 0);
6164         err = btrfs_update_inode(trans, root, inode);
6165         if (err)
6166                 goto out_fail_inode;
6167
6168         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6169                              dentry->d_name.len, 0, index);
6170         if (err)
6171                 goto out_fail_inode;
6172
6173         d_instantiate(dentry, inode);
6174         /*
6175          * mkdir is special.  We're unlocking after we call d_instantiate
6176          * to avoid a race with nfsd calling d_instantiate.
6177          */
6178         unlock_new_inode(inode);
6179         drop_on_err = 0;
6180
6181 out_fail:
6182         btrfs_end_transaction(trans, root);
6183         if (drop_on_err)
6184                 iput(inode);
6185         btrfs_balance_delayed_items(root);
6186         btrfs_btree_balance_dirty(root);
6187         return err;
6188
6189 out_fail_inode:
6190         unlock_new_inode(inode);
6191         goto out_fail;
6192 }
6193
6194 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6195  * and an extent that you want to insert, deal with overlap and insert
6196  * the new extent into the tree.
6197  */
6198 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6199                                 struct extent_map *existing,
6200                                 struct extent_map *em,
6201                                 u64 map_start)
6202 {
6203         u64 start_diff;
6204
6205         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6206         start_diff = map_start - em->start;
6207         em->start = map_start;
6208         em->len = existing->start - em->start;
6209         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6210             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6211                 em->block_start += start_diff;
6212                 em->block_len -= start_diff;
6213         }
6214         return add_extent_mapping(em_tree, em, 0);
6215 }
6216
6217 static noinline int uncompress_inline(struct btrfs_path *path,
6218                                       struct inode *inode, struct page *page,
6219                                       size_t pg_offset, u64 extent_offset,
6220                                       struct btrfs_file_extent_item *item)
6221 {
6222         int ret;
6223         struct extent_buffer *leaf = path->nodes[0];
6224         char *tmp;
6225         size_t max_size;
6226         unsigned long inline_size;
6227         unsigned long ptr;
6228         int compress_type;
6229
6230         WARN_ON(pg_offset != 0);
6231         compress_type = btrfs_file_extent_compression(leaf, item);
6232         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6233         inline_size = btrfs_file_extent_inline_item_len(leaf,
6234                                         btrfs_item_nr(path->slots[0]));
6235         tmp = kmalloc(inline_size, GFP_NOFS);
6236         if (!tmp)
6237                 return -ENOMEM;
6238         ptr = btrfs_file_extent_inline_start(item);
6239
6240         read_extent_buffer(leaf, tmp, ptr, inline_size);
6241
6242         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6243         ret = btrfs_decompress(compress_type, tmp, page,
6244                                extent_offset, inline_size, max_size);
6245         kfree(tmp);
6246         return ret;
6247 }
6248
6249 /*
6250  * a bit scary, this does extent mapping from logical file offset to the disk.
6251  * the ugly parts come from merging extents from the disk with the in-ram
6252  * representation.  This gets more complex because of the data=ordered code,
6253  * where the in-ram extents might be locked pending data=ordered completion.
6254  *
6255  * This also copies inline extents directly into the page.
6256  */
6257
6258 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6259                                     size_t pg_offset, u64 start, u64 len,
6260                                     int create)
6261 {
6262         int ret;
6263         int err = 0;
6264         u64 extent_start = 0;
6265         u64 extent_end = 0;
6266         u64 objectid = btrfs_ino(inode);
6267         u32 found_type;
6268         struct btrfs_path *path = NULL;
6269         struct btrfs_root *root = BTRFS_I(inode)->root;
6270         struct btrfs_file_extent_item *item;
6271         struct extent_buffer *leaf;
6272         struct btrfs_key found_key;
6273         struct extent_map *em = NULL;
6274         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6275         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6276         struct btrfs_trans_handle *trans = NULL;
6277         const bool new_inline = !page || create;
6278
6279 again:
6280         read_lock(&em_tree->lock);
6281         em = lookup_extent_mapping(em_tree, start, len);
6282         if (em)
6283                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6284         read_unlock(&em_tree->lock);
6285
6286         if (em) {
6287                 if (em->start > start || em->start + em->len <= start)
6288                         free_extent_map(em);
6289                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6290                         free_extent_map(em);
6291                 else
6292                         goto out;
6293         }
6294         em = alloc_extent_map();
6295         if (!em) {
6296                 err = -ENOMEM;
6297                 goto out;
6298         }
6299         em->bdev = root->fs_info->fs_devices->latest_bdev;
6300         em->start = EXTENT_MAP_HOLE;
6301         em->orig_start = EXTENT_MAP_HOLE;
6302         em->len = (u64)-1;
6303         em->block_len = (u64)-1;
6304
6305         if (!path) {
6306                 path = btrfs_alloc_path();
6307                 if (!path) {
6308                         err = -ENOMEM;
6309                         goto out;
6310                 }
6311                 /*
6312                  * Chances are we'll be called again, so go ahead and do
6313                  * readahead
6314                  */
6315                 path->reada = 1;
6316         }
6317
6318         ret = btrfs_lookup_file_extent(trans, root, path,
6319                                        objectid, start, trans != NULL);
6320         if (ret < 0) {
6321                 err = ret;
6322                 goto out;
6323         }
6324
6325         if (ret != 0) {
6326                 if (path->slots[0] == 0)
6327                         goto not_found;
6328                 path->slots[0]--;
6329         }
6330
6331         leaf = path->nodes[0];
6332         item = btrfs_item_ptr(leaf, path->slots[0],
6333                               struct btrfs_file_extent_item);
6334         /* are we inside the extent that was found? */
6335         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6336         found_type = btrfs_key_type(&found_key);
6337         if (found_key.objectid != objectid ||
6338             found_type != BTRFS_EXTENT_DATA_KEY) {
6339                 /*
6340                  * If we backup past the first extent we want to move forward
6341                  * and see if there is an extent in front of us, otherwise we'll
6342                  * say there is a hole for our whole search range which can
6343                  * cause problems.
6344                  */
6345                 extent_end = start;
6346                 goto next;
6347         }
6348
6349         found_type = btrfs_file_extent_type(leaf, item);
6350         extent_start = found_key.offset;
6351         if (found_type == BTRFS_FILE_EXTENT_REG ||
6352             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6353                 extent_end = extent_start +
6354                        btrfs_file_extent_num_bytes(leaf, item);
6355         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6356                 size_t size;
6357                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6358                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6359         }
6360 next:
6361         if (start >= extent_end) {
6362                 path->slots[0]++;
6363                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6364                         ret = btrfs_next_leaf(root, path);
6365                         if (ret < 0) {
6366                                 err = ret;
6367                                 goto out;
6368                         }
6369                         if (ret > 0)
6370                                 goto not_found;
6371                         leaf = path->nodes[0];
6372                 }
6373                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6374                 if (found_key.objectid != objectid ||
6375                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6376                         goto not_found;
6377                 if (start + len <= found_key.offset)
6378                         goto not_found;
6379                 if (start > found_key.offset)
6380                         goto next;
6381                 em->start = start;
6382                 em->orig_start = start;
6383                 em->len = found_key.offset - start;
6384                 goto not_found_em;
6385         }
6386
6387         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6388
6389         if (found_type == BTRFS_FILE_EXTENT_REG ||
6390             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6391                 goto insert;
6392         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6393                 unsigned long ptr;
6394                 char *map;
6395                 size_t size;
6396                 size_t extent_offset;
6397                 size_t copy_size;
6398
6399                 if (new_inline)
6400                         goto out;
6401
6402                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6403                 extent_offset = page_offset(page) + pg_offset - extent_start;
6404                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6405                                 size - extent_offset);
6406                 em->start = extent_start + extent_offset;
6407                 em->len = ALIGN(copy_size, root->sectorsize);
6408                 em->orig_block_len = em->len;
6409                 em->orig_start = em->start;
6410                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6411                 if (create == 0 && !PageUptodate(page)) {
6412                         if (btrfs_file_extent_compression(leaf, item) !=
6413                             BTRFS_COMPRESS_NONE) {
6414                                 ret = uncompress_inline(path, inode, page,
6415                                                         pg_offset,
6416                                                         extent_offset, item);
6417                                 if (ret) {
6418                                         err = ret;
6419                                         goto out;
6420                                 }
6421                         } else {
6422                                 map = kmap(page);
6423                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6424                                                    copy_size);
6425                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6426                                         memset(map + pg_offset + copy_size, 0,
6427                                                PAGE_CACHE_SIZE - pg_offset -
6428                                                copy_size);
6429                                 }
6430                                 kunmap(page);
6431                         }
6432                         flush_dcache_page(page);
6433                 } else if (create && PageUptodate(page)) {
6434                         BUG();
6435                         if (!trans) {
6436                                 kunmap(page);
6437                                 free_extent_map(em);
6438                                 em = NULL;
6439
6440                                 btrfs_release_path(path);
6441                                 trans = btrfs_join_transaction(root);
6442
6443                                 if (IS_ERR(trans))
6444                                         return ERR_CAST(trans);
6445                                 goto again;
6446                         }
6447                         map = kmap(page);
6448                         write_extent_buffer(leaf, map + pg_offset, ptr,
6449                                             copy_size);
6450                         kunmap(page);
6451                         btrfs_mark_buffer_dirty(leaf);
6452                 }
6453                 set_extent_uptodate(io_tree, em->start,
6454                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6455                 goto insert;
6456         }
6457 not_found:
6458         em->start = start;
6459         em->orig_start = start;
6460         em->len = len;
6461 not_found_em:
6462         em->block_start = EXTENT_MAP_HOLE;
6463         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6464 insert:
6465         btrfs_release_path(path);
6466         if (em->start > start || extent_map_end(em) <= start) {
6467                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6468                         em->start, em->len, start, len);
6469                 err = -EIO;
6470                 goto out;
6471         }
6472
6473         err = 0;
6474         write_lock(&em_tree->lock);
6475         ret = add_extent_mapping(em_tree, em, 0);
6476         /* it is possible that someone inserted the extent into the tree
6477          * while we had the lock dropped.  It is also possible that
6478          * an overlapping map exists in the tree
6479          */
6480         if (ret == -EEXIST) {
6481                 struct extent_map *existing;
6482
6483                 ret = 0;
6484
6485                 existing = lookup_extent_mapping(em_tree, start, len);
6486                 if (existing && (existing->start > start ||
6487                     existing->start + existing->len <= start)) {
6488                         free_extent_map(existing);
6489                         existing = NULL;
6490                 }
6491                 if (!existing) {
6492                         existing = lookup_extent_mapping(em_tree, em->start,
6493                                                          em->len);
6494                         if (existing) {
6495                                 err = merge_extent_mapping(em_tree, existing,
6496                                                            em, start);
6497                                 free_extent_map(existing);
6498                                 if (err) {
6499                                         free_extent_map(em);
6500                                         em = NULL;
6501                                 }
6502                         } else {
6503                                 err = -EIO;
6504                                 free_extent_map(em);
6505                                 em = NULL;
6506                         }
6507                 } else {
6508                         free_extent_map(em);
6509                         em = existing;
6510                         err = 0;
6511                 }
6512         }
6513         write_unlock(&em_tree->lock);
6514 out:
6515
6516         trace_btrfs_get_extent(root, em);
6517
6518         if (path)
6519                 btrfs_free_path(path);
6520         if (trans) {
6521                 ret = btrfs_end_transaction(trans, root);
6522                 if (!err)
6523                         err = ret;
6524         }
6525         if (err) {
6526                 free_extent_map(em);
6527                 return ERR_PTR(err);
6528         }
6529         BUG_ON(!em); /* Error is always set */
6530         return em;
6531 }
6532
6533 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6534                                            size_t pg_offset, u64 start, u64 len,
6535                                            int create)
6536 {
6537         struct extent_map *em;
6538         struct extent_map *hole_em = NULL;
6539         u64 range_start = start;
6540         u64 end;
6541         u64 found;
6542         u64 found_end;
6543         int err = 0;
6544
6545         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6546         if (IS_ERR(em))
6547                 return em;
6548         if (em) {
6549                 /*
6550                  * if our em maps to
6551                  * -  a hole or
6552                  * -  a pre-alloc extent,
6553                  * there might actually be delalloc bytes behind it.
6554                  */
6555                 if (em->block_start != EXTENT_MAP_HOLE &&
6556                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6557                         return em;
6558                 else
6559                         hole_em = em;
6560         }
6561
6562         /* check to see if we've wrapped (len == -1 or similar) */
6563         end = start + len;
6564         if (end < start)
6565                 end = (u64)-1;
6566         else
6567                 end -= 1;
6568
6569         em = NULL;
6570
6571         /* ok, we didn't find anything, lets look for delalloc */
6572         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6573                                  end, len, EXTENT_DELALLOC, 1);
6574         found_end = range_start + found;
6575         if (found_end < range_start)
6576                 found_end = (u64)-1;
6577
6578         /*
6579          * we didn't find anything useful, return
6580          * the original results from get_extent()
6581          */
6582         if (range_start > end || found_end <= start) {
6583                 em = hole_em;
6584                 hole_em = NULL;
6585                 goto out;
6586         }
6587
6588         /* adjust the range_start to make sure it doesn't
6589          * go backwards from the start they passed in
6590          */
6591         range_start = max(start, range_start);
6592         found = found_end - range_start;
6593
6594         if (found > 0) {
6595                 u64 hole_start = start;
6596                 u64 hole_len = len;
6597
6598                 em = alloc_extent_map();
6599                 if (!em) {
6600                         err = -ENOMEM;
6601                         goto out;
6602                 }
6603                 /*
6604                  * when btrfs_get_extent can't find anything it
6605                  * returns one huge hole
6606                  *
6607                  * make sure what it found really fits our range, and
6608                  * adjust to make sure it is based on the start from
6609                  * the caller
6610                  */
6611                 if (hole_em) {
6612                         u64 calc_end = extent_map_end(hole_em);
6613
6614                         if (calc_end <= start || (hole_em->start > end)) {
6615                                 free_extent_map(hole_em);
6616                                 hole_em = NULL;
6617                         } else {
6618                                 hole_start = max(hole_em->start, start);
6619                                 hole_len = calc_end - hole_start;
6620                         }
6621                 }
6622                 em->bdev = NULL;
6623                 if (hole_em && range_start > hole_start) {
6624                         /* our hole starts before our delalloc, so we
6625                          * have to return just the parts of the hole
6626                          * that go until  the delalloc starts
6627                          */
6628                         em->len = min(hole_len,
6629                                       range_start - hole_start);
6630                         em->start = hole_start;
6631                         em->orig_start = hole_start;
6632                         /*
6633                          * don't adjust block start at all,
6634                          * it is fixed at EXTENT_MAP_HOLE
6635                          */
6636                         em->block_start = hole_em->block_start;
6637                         em->block_len = hole_len;
6638                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6639                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6640                 } else {
6641                         em->start = range_start;
6642                         em->len = found;
6643                         em->orig_start = range_start;
6644                         em->block_start = EXTENT_MAP_DELALLOC;
6645                         em->block_len = found;
6646                 }
6647         } else if (hole_em) {
6648                 return hole_em;
6649         }
6650 out:
6651
6652         free_extent_map(hole_em);
6653         if (err) {
6654                 free_extent_map(em);
6655                 return ERR_PTR(err);
6656         }
6657         return em;
6658 }
6659
6660 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6661                                                   u64 start, u64 len)
6662 {
6663         struct btrfs_root *root = BTRFS_I(inode)->root;
6664         struct extent_map *em;
6665         struct btrfs_key ins;
6666         u64 alloc_hint;
6667         int ret;
6668
6669         alloc_hint = get_extent_allocation_hint(inode, start, len);
6670         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6671                                    alloc_hint, &ins, 1, 1);
6672         if (ret)
6673                 return ERR_PTR(ret);
6674
6675         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6676                               ins.offset, ins.offset, ins.offset, 0);
6677         if (IS_ERR(em)) {
6678                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6679                 return em;
6680         }
6681
6682         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6683                                            ins.offset, ins.offset, 0);
6684         if (ret) {
6685                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6686                 free_extent_map(em);
6687                 return ERR_PTR(ret);
6688         }
6689
6690         return em;
6691 }
6692
6693 /*
6694  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6695  * block must be cow'd
6696  */
6697 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6698                               u64 *orig_start, u64 *orig_block_len,
6699                               u64 *ram_bytes)
6700 {
6701         struct btrfs_trans_handle *trans;
6702         struct btrfs_path *path;
6703         int ret;
6704         struct extent_buffer *leaf;
6705         struct btrfs_root *root = BTRFS_I(inode)->root;
6706         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6707         struct btrfs_file_extent_item *fi;
6708         struct btrfs_key key;
6709         u64 disk_bytenr;
6710         u64 backref_offset;
6711         u64 extent_end;
6712         u64 num_bytes;
6713         int slot;
6714         int found_type;
6715         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6716
6717         path = btrfs_alloc_path();
6718         if (!path)
6719                 return -ENOMEM;
6720
6721         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6722                                        offset, 0);
6723         if (ret < 0)
6724                 goto out;
6725
6726         slot = path->slots[0];
6727         if (ret == 1) {
6728                 if (slot == 0) {
6729                         /* can't find the item, must cow */
6730                         ret = 0;
6731                         goto out;
6732                 }
6733                 slot--;
6734         }
6735         ret = 0;
6736         leaf = path->nodes[0];
6737         btrfs_item_key_to_cpu(leaf, &key, slot);
6738         if (key.objectid != btrfs_ino(inode) ||
6739             key.type != BTRFS_EXTENT_DATA_KEY) {
6740                 /* not our file or wrong item type, must cow */
6741                 goto out;
6742         }
6743
6744         if (key.offset > offset) {
6745                 /* Wrong offset, must cow */
6746                 goto out;
6747         }
6748
6749         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6750         found_type = btrfs_file_extent_type(leaf, fi);
6751         if (found_type != BTRFS_FILE_EXTENT_REG &&
6752             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6753                 /* not a regular extent, must cow */
6754                 goto out;
6755         }
6756
6757         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6758                 goto out;
6759
6760         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6761         if (extent_end <= offset)
6762                 goto out;
6763
6764         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6765         if (disk_bytenr == 0)
6766                 goto out;
6767
6768         if (btrfs_file_extent_compression(leaf, fi) ||
6769             btrfs_file_extent_encryption(leaf, fi) ||
6770             btrfs_file_extent_other_encoding(leaf, fi))
6771                 goto out;
6772
6773         backref_offset = btrfs_file_extent_offset(leaf, fi);
6774
6775         if (orig_start) {
6776                 *orig_start = key.offset - backref_offset;
6777                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6778                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6779         }
6780
6781         if (btrfs_extent_readonly(root, disk_bytenr))
6782                 goto out;
6783
6784         num_bytes = min(offset + *len, extent_end) - offset;
6785         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6786                 u64 range_end;
6787
6788                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6789                 ret = test_range_bit(io_tree, offset, range_end,
6790                                      EXTENT_DELALLOC, 0, NULL);
6791                 if (ret) {
6792                         ret = -EAGAIN;
6793                         goto out;
6794                 }
6795         }
6796
6797         btrfs_release_path(path);
6798
6799         /*
6800          * look for other files referencing this extent, if we
6801          * find any we must cow
6802          */
6803         trans = btrfs_join_transaction(root);
6804         if (IS_ERR(trans)) {
6805                 ret = 0;
6806                 goto out;
6807         }
6808
6809         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6810                                     key.offset - backref_offset, disk_bytenr);
6811         btrfs_end_transaction(trans, root);
6812         if (ret) {
6813                 ret = 0;
6814                 goto out;
6815         }
6816
6817         /*
6818          * adjust disk_bytenr and num_bytes to cover just the bytes
6819          * in this extent we are about to write.  If there
6820          * are any csums in that range we have to cow in order
6821          * to keep the csums correct
6822          */
6823         disk_bytenr += backref_offset;
6824         disk_bytenr += offset - key.offset;
6825         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6826                                 goto out;
6827         /*
6828          * all of the above have passed, it is safe to overwrite this extent
6829          * without cow
6830          */
6831         *len = num_bytes;
6832         ret = 1;
6833 out:
6834         btrfs_free_path(path);
6835         return ret;
6836 }
6837
6838 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
6839 {
6840         struct radix_tree_root *root = &inode->i_mapping->page_tree;
6841         int found = false;
6842         void **pagep = NULL;
6843         struct page *page = NULL;
6844         int start_idx;
6845         int end_idx;
6846
6847         start_idx = start >> PAGE_CACHE_SHIFT;
6848
6849         /*
6850          * end is the last byte in the last page.  end == start is legal
6851          */
6852         end_idx = end >> PAGE_CACHE_SHIFT;
6853
6854         rcu_read_lock();
6855
6856         /* Most of the code in this while loop is lifted from
6857          * find_get_page.  It's been modified to begin searching from a
6858          * page and return just the first page found in that range.  If the
6859          * found idx is less than or equal to the end idx then we know that
6860          * a page exists.  If no pages are found or if those pages are
6861          * outside of the range then we're fine (yay!) */
6862         while (page == NULL &&
6863                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
6864                 page = radix_tree_deref_slot(pagep);
6865                 if (unlikely(!page))
6866                         break;
6867
6868                 if (radix_tree_exception(page)) {
6869                         if (radix_tree_deref_retry(page)) {
6870                                 page = NULL;
6871                                 continue;
6872                         }
6873                         /*
6874                          * Otherwise, shmem/tmpfs must be storing a swap entry
6875                          * here as an exceptional entry: so return it without
6876                          * attempting to raise page count.
6877                          */
6878                         page = NULL;
6879                         break; /* TODO: Is this relevant for this use case? */
6880                 }
6881
6882                 if (!page_cache_get_speculative(page)) {
6883                         page = NULL;
6884                         continue;
6885                 }
6886
6887                 /*
6888                  * Has the page moved?
6889                  * This is part of the lockless pagecache protocol. See
6890                  * include/linux/pagemap.h for details.
6891                  */
6892                 if (unlikely(page != *pagep)) {
6893                         page_cache_release(page);
6894                         page = NULL;
6895                 }
6896         }
6897
6898         if (page) {
6899                 if (page->index <= end_idx)
6900                         found = true;
6901                 page_cache_release(page);
6902         }
6903
6904         rcu_read_unlock();
6905         return found;
6906 }
6907
6908 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6909                               struct extent_state **cached_state, int writing)
6910 {
6911         struct btrfs_ordered_extent *ordered;
6912         int ret = 0;
6913
6914         while (1) {
6915                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6916                                  0, cached_state);
6917                 /*
6918                  * We're concerned with the entire range that we're going to be
6919                  * doing DIO to, so we need to make sure theres no ordered
6920                  * extents in this range.
6921                  */
6922                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6923                                                      lockend - lockstart + 1);
6924
6925                 /*
6926                  * We need to make sure there are no buffered pages in this
6927                  * range either, we could have raced between the invalidate in
6928                  * generic_file_direct_write and locking the extent.  The
6929                  * invalidate needs to happen so that reads after a write do not
6930                  * get stale data.
6931                  */
6932                 if (!ordered &&
6933                     (!writing ||
6934                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
6935                         break;
6936
6937                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6938                                      cached_state, GFP_NOFS);
6939
6940                 if (ordered) {
6941                         btrfs_start_ordered_extent(inode, ordered, 1);
6942                         btrfs_put_ordered_extent(ordered);
6943                 } else {
6944                         /* Screw you mmap */
6945                         ret = filemap_write_and_wait_range(inode->i_mapping,
6946                                                            lockstart,
6947                                                            lockend);
6948                         if (ret)
6949                                 break;
6950
6951                         /*
6952                          * If we found a page that couldn't be invalidated just
6953                          * fall back to buffered.
6954                          */
6955                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6956                                         lockstart >> PAGE_CACHE_SHIFT,
6957                                         lockend >> PAGE_CACHE_SHIFT);
6958                         if (ret)
6959                                 break;
6960                 }
6961
6962                 cond_resched();
6963         }
6964
6965         return ret;
6966 }
6967
6968 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6969                                            u64 len, u64 orig_start,
6970                                            u64 block_start, u64 block_len,
6971                                            u64 orig_block_len, u64 ram_bytes,
6972                                            int type)
6973 {
6974         struct extent_map_tree *em_tree;
6975         struct extent_map *em;
6976         struct btrfs_root *root = BTRFS_I(inode)->root;
6977         int ret;
6978
6979         em_tree = &BTRFS_I(inode)->extent_tree;
6980         em = alloc_extent_map();
6981         if (!em)
6982                 return ERR_PTR(-ENOMEM);
6983
6984         em->start = start;
6985         em->orig_start = orig_start;
6986         em->mod_start = start;
6987         em->mod_len = len;
6988         em->len = len;
6989         em->block_len = block_len;
6990         em->block_start = block_start;
6991         em->bdev = root->fs_info->fs_devices->latest_bdev;
6992         em->orig_block_len = orig_block_len;
6993         em->ram_bytes = ram_bytes;
6994         em->generation = -1;
6995         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6996         if (type == BTRFS_ORDERED_PREALLOC)
6997                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6998
6999         do {
7000                 btrfs_drop_extent_cache(inode, em->start,
7001                                 em->start + em->len - 1, 0);
7002                 write_lock(&em_tree->lock);
7003                 ret = add_extent_mapping(em_tree, em, 1);
7004                 write_unlock(&em_tree->lock);
7005         } while (ret == -EEXIST);
7006
7007         if (ret) {
7008                 free_extent_map(em);
7009                 return ERR_PTR(ret);
7010         }
7011
7012         return em;
7013 }
7014
7015
7016 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7017                                    struct buffer_head *bh_result, int create)
7018 {
7019         struct extent_map *em;
7020         struct btrfs_root *root = BTRFS_I(inode)->root;
7021         struct extent_state *cached_state = NULL;
7022         u64 start = iblock << inode->i_blkbits;
7023         u64 lockstart, lockend;
7024         u64 len = bh_result->b_size;
7025         int unlock_bits = EXTENT_LOCKED;
7026         int ret = 0;
7027
7028         if (create)
7029                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
7030         else
7031                 len = min_t(u64, len, root->sectorsize);
7032
7033         lockstart = start;
7034         lockend = start + len - 1;
7035
7036         /*
7037          * If this errors out it's because we couldn't invalidate pagecache for
7038          * this range and we need to fallback to buffered.
7039          */
7040         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7041                 return -ENOTBLK;
7042
7043         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7044         if (IS_ERR(em)) {
7045                 ret = PTR_ERR(em);
7046                 goto unlock_err;
7047         }
7048
7049         /*
7050          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7051          * io.  INLINE is special, and we could probably kludge it in here, but
7052          * it's still buffered so for safety lets just fall back to the generic
7053          * buffered path.
7054          *
7055          * For COMPRESSED we _have_ to read the entire extent in so we can
7056          * decompress it, so there will be buffering required no matter what we
7057          * do, so go ahead and fallback to buffered.
7058          *
7059          * We return -ENOTBLK because thats what makes DIO go ahead and go back
7060          * to buffered IO.  Don't blame me, this is the price we pay for using
7061          * the generic code.
7062          */
7063         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7064             em->block_start == EXTENT_MAP_INLINE) {
7065                 free_extent_map(em);
7066                 ret = -ENOTBLK;
7067                 goto unlock_err;
7068         }
7069
7070         /* Just a good old fashioned hole, return */
7071         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7072                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7073                 free_extent_map(em);
7074                 goto unlock_err;
7075         }
7076
7077         /*
7078          * We don't allocate a new extent in the following cases
7079          *
7080          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7081          * existing extent.
7082          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7083          * just use the extent.
7084          *
7085          */
7086         if (!create) {
7087                 len = min(len, em->len - (start - em->start));
7088                 lockstart = start + len;
7089                 goto unlock;
7090         }
7091
7092         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7093             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7094              em->block_start != EXTENT_MAP_HOLE)) {
7095                 int type;
7096                 int ret;
7097                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7098
7099                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7100                         type = BTRFS_ORDERED_PREALLOC;
7101                 else
7102                         type = BTRFS_ORDERED_NOCOW;
7103                 len = min(len, em->len - (start - em->start));
7104                 block_start = em->block_start + (start - em->start);
7105
7106                 if (can_nocow_extent(inode, start, &len, &orig_start,
7107                                      &orig_block_len, &ram_bytes) == 1) {
7108                         if (type == BTRFS_ORDERED_PREALLOC) {
7109                                 free_extent_map(em);
7110                                 em = create_pinned_em(inode, start, len,
7111                                                        orig_start,
7112                                                        block_start, len,
7113                                                        orig_block_len,
7114                                                        ram_bytes, type);
7115                                 if (IS_ERR(em))
7116                                         goto unlock_err;
7117                         }
7118
7119                         ret = btrfs_add_ordered_extent_dio(inode, start,
7120                                            block_start, len, len, type);
7121                         if (ret) {
7122                                 free_extent_map(em);
7123                                 goto unlock_err;
7124                         }
7125                         goto unlock;
7126                 }
7127         }
7128
7129         /*
7130          * this will cow the extent, reset the len in case we changed
7131          * it above
7132          */
7133         len = bh_result->b_size;
7134         free_extent_map(em);
7135         em = btrfs_new_extent_direct(inode, start, len);
7136         if (IS_ERR(em)) {
7137                 ret = PTR_ERR(em);
7138                 goto unlock_err;
7139         }
7140         len = min(len, em->len - (start - em->start));
7141 unlock:
7142         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7143                 inode->i_blkbits;
7144         bh_result->b_size = len;
7145         bh_result->b_bdev = em->bdev;
7146         set_buffer_mapped(bh_result);
7147         if (create) {
7148                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7149                         set_buffer_new(bh_result);
7150
7151                 /*
7152                  * Need to update the i_size under the extent lock so buffered
7153                  * readers will get the updated i_size when we unlock.
7154                  */
7155                 if (start + len > i_size_read(inode))
7156                         i_size_write(inode, start + len);
7157
7158                 spin_lock(&BTRFS_I(inode)->lock);
7159                 BTRFS_I(inode)->outstanding_extents++;
7160                 spin_unlock(&BTRFS_I(inode)->lock);
7161
7162                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7163                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
7164                                      &cached_state, GFP_NOFS);
7165                 BUG_ON(ret);
7166         }
7167
7168         /*
7169          * In the case of write we need to clear and unlock the entire range,
7170          * in the case of read we need to unlock only the end area that we
7171          * aren't using if there is any left over space.
7172          */
7173         if (lockstart < lockend) {
7174                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7175                                  lockend, unlock_bits, 1, 0,
7176                                  &cached_state, GFP_NOFS);
7177         } else {
7178                 free_extent_state(cached_state);
7179         }
7180
7181         free_extent_map(em);
7182
7183         return 0;
7184
7185 unlock_err:
7186         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7187                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7188         return ret;
7189 }
7190
7191 static void btrfs_endio_direct_read(struct bio *bio, int err)
7192 {
7193         struct btrfs_dio_private *dip = bio->bi_private;
7194         struct bio_vec *bvec;
7195         struct inode *inode = dip->inode;
7196         struct btrfs_root *root = BTRFS_I(inode)->root;
7197         struct bio *dio_bio;
7198         u32 *csums = (u32 *)dip->csum;
7199         u64 start;
7200         int i;
7201
7202         start = dip->logical_offset;
7203         bio_for_each_segment_all(bvec, bio, i) {
7204                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
7205                         struct page *page = bvec->bv_page;
7206                         char *kaddr;
7207                         u32 csum = ~(u32)0;
7208                         unsigned long flags;
7209
7210                         local_irq_save(flags);
7211                         kaddr = kmap_atomic(page);
7212                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
7213                                                csum, bvec->bv_len);
7214                         btrfs_csum_final(csum, (char *)&csum);
7215                         kunmap_atomic(kaddr);
7216                         local_irq_restore(flags);
7217
7218                         flush_dcache_page(bvec->bv_page);
7219                         if (csum != csums[i]) {
7220                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
7221                                           btrfs_ino(inode), start, csum,
7222                                           csums[i]);
7223                                 err = -EIO;
7224                         }
7225                 }
7226
7227                 start += bvec->bv_len;
7228         }
7229
7230         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7231                       dip->logical_offset + dip->bytes - 1);
7232         dio_bio = dip->dio_bio;
7233
7234         kfree(dip);
7235
7236         /* If we had a csum failure make sure to clear the uptodate flag */
7237         if (err)
7238                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7239         dio_end_io(dio_bio, err);
7240         bio_put(bio);
7241 }
7242
7243 static void btrfs_endio_direct_write(struct bio *bio, int err)
7244 {
7245         struct btrfs_dio_private *dip = bio->bi_private;
7246         struct inode *inode = dip->inode;
7247         struct btrfs_root *root = BTRFS_I(inode)->root;
7248         struct btrfs_ordered_extent *ordered = NULL;
7249         u64 ordered_offset = dip->logical_offset;
7250         u64 ordered_bytes = dip->bytes;
7251         struct bio *dio_bio;
7252         int ret;
7253
7254         if (err)
7255                 goto out_done;
7256 again:
7257         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7258                                                    &ordered_offset,
7259                                                    ordered_bytes, !err);
7260         if (!ret)
7261                 goto out_test;
7262
7263         btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7264                         finish_ordered_fn, NULL, NULL);
7265         btrfs_queue_work(root->fs_info->endio_write_workers,
7266                          &ordered->work);
7267 out_test:
7268         /*
7269          * our bio might span multiple ordered extents.  If we haven't
7270          * completed the accounting for the whole dio, go back and try again
7271          */
7272         if (ordered_offset < dip->logical_offset + dip->bytes) {
7273                 ordered_bytes = dip->logical_offset + dip->bytes -
7274                         ordered_offset;
7275                 ordered = NULL;
7276                 goto again;
7277         }
7278 out_done:
7279         dio_bio = dip->dio_bio;
7280
7281         kfree(dip);
7282
7283         /* If we had an error make sure to clear the uptodate flag */
7284         if (err)
7285                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7286         dio_end_io(dio_bio, err);
7287         bio_put(bio);
7288 }
7289
7290 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7291                                     struct bio *bio, int mirror_num,
7292                                     unsigned long bio_flags, u64 offset)
7293 {
7294         int ret;
7295         struct btrfs_root *root = BTRFS_I(inode)->root;
7296         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7297         BUG_ON(ret); /* -ENOMEM */
7298         return 0;
7299 }
7300
7301 static void btrfs_end_dio_bio(struct bio *bio, int err)
7302 {
7303         struct btrfs_dio_private *dip = bio->bi_private;
7304
7305         if (err) {
7306                 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7307                           "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7308                       btrfs_ino(dip->inode), bio->bi_rw,
7309                       (unsigned long long)bio->bi_iter.bi_sector,
7310                       bio->bi_iter.bi_size, err);
7311                 dip->errors = 1;
7312
7313                 /*
7314                  * before atomic variable goto zero, we must make sure
7315                  * dip->errors is perceived to be set.
7316                  */
7317                 smp_mb__before_atomic();
7318         }
7319
7320         /* if there are more bios still pending for this dio, just exit */
7321         if (!atomic_dec_and_test(&dip->pending_bios))
7322                 goto out;
7323
7324         if (dip->errors) {
7325                 bio_io_error(dip->orig_bio);
7326         } else {
7327                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7328                 bio_endio(dip->orig_bio, 0);
7329         }
7330 out:
7331         bio_put(bio);
7332 }
7333
7334 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7335                                        u64 first_sector, gfp_t gfp_flags)
7336 {
7337         int nr_vecs = bio_get_nr_vecs(bdev);
7338         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7339 }
7340
7341 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7342                                          int rw, u64 file_offset, int skip_sum,
7343                                          int async_submit)
7344 {
7345         struct btrfs_dio_private *dip = bio->bi_private;
7346         int write = rw & REQ_WRITE;
7347         struct btrfs_root *root = BTRFS_I(inode)->root;
7348         int ret;
7349
7350         if (async_submit)
7351                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7352
7353         bio_get(bio);
7354
7355         if (!write) {
7356                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7357                 if (ret)
7358                         goto err;
7359         }
7360
7361         if (skip_sum)
7362                 goto map;
7363
7364         if (write && async_submit) {
7365                 ret = btrfs_wq_submit_bio(root->fs_info,
7366                                    inode, rw, bio, 0, 0,
7367                                    file_offset,
7368                                    __btrfs_submit_bio_start_direct_io,
7369                                    __btrfs_submit_bio_done);
7370                 goto err;
7371         } else if (write) {
7372                 /*
7373                  * If we aren't doing async submit, calculate the csum of the
7374                  * bio now.
7375                  */
7376                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7377                 if (ret)
7378                         goto err;
7379         } else if (!skip_sum) {
7380                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7381                                                 file_offset);
7382                 if (ret)
7383                         goto err;
7384         }
7385
7386 map:
7387         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7388 err:
7389         bio_put(bio);
7390         return ret;
7391 }
7392
7393 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7394                                     int skip_sum)
7395 {
7396         struct inode *inode = dip->inode;
7397         struct btrfs_root *root = BTRFS_I(inode)->root;
7398         struct bio *bio;
7399         struct bio *orig_bio = dip->orig_bio;
7400         struct bio_vec *bvec = orig_bio->bi_io_vec;
7401         u64 start_sector = orig_bio->bi_iter.bi_sector;
7402         u64 file_offset = dip->logical_offset;
7403         u64 submit_len = 0;
7404         u64 map_length;
7405         int nr_pages = 0;
7406         int ret = 0;
7407         int async_submit = 0;
7408
7409         map_length = orig_bio->bi_iter.bi_size;
7410         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7411                               &map_length, NULL, 0);
7412         if (ret)
7413                 return -EIO;
7414
7415         if (map_length >= orig_bio->bi_iter.bi_size) {
7416                 bio = orig_bio;
7417                 goto submit;
7418         }
7419
7420         /* async crcs make it difficult to collect full stripe writes. */
7421         if (btrfs_get_alloc_profile(root, 1) &
7422             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7423                 async_submit = 0;
7424         else
7425                 async_submit = 1;
7426
7427         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7428         if (!bio)
7429                 return -ENOMEM;
7430
7431         bio->bi_private = dip;
7432         bio->bi_end_io = btrfs_end_dio_bio;
7433         atomic_inc(&dip->pending_bios);
7434
7435         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7436                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7437                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7438                                  bvec->bv_offset) < bvec->bv_len)) {
7439                         /*
7440                          * inc the count before we submit the bio so
7441                          * we know the end IO handler won't happen before
7442                          * we inc the count. Otherwise, the dip might get freed
7443                          * before we're done setting it up
7444                          */
7445                         atomic_inc(&dip->pending_bios);
7446                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7447                                                      file_offset, skip_sum,
7448                                                      async_submit);
7449                         if (ret) {
7450                                 bio_put(bio);
7451                                 atomic_dec(&dip->pending_bios);
7452                                 goto out_err;
7453                         }
7454
7455                         start_sector += submit_len >> 9;
7456                         file_offset += submit_len;
7457
7458                         submit_len = 0;
7459                         nr_pages = 0;
7460
7461                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7462                                                   start_sector, GFP_NOFS);
7463                         if (!bio)
7464                                 goto out_err;
7465                         bio->bi_private = dip;
7466                         bio->bi_end_io = btrfs_end_dio_bio;
7467
7468                         map_length = orig_bio->bi_iter.bi_size;
7469                         ret = btrfs_map_block(root->fs_info, rw,
7470                                               start_sector << 9,
7471                                               &map_length, NULL, 0);
7472                         if (ret) {
7473                                 bio_put(bio);
7474                                 goto out_err;
7475                         }
7476                 } else {
7477                         submit_len += bvec->bv_len;
7478                         nr_pages++;
7479                         bvec++;
7480                 }
7481         }
7482
7483 submit:
7484         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7485                                      async_submit);
7486         if (!ret)
7487                 return 0;
7488
7489         bio_put(bio);
7490 out_err:
7491         dip->errors = 1;
7492         /*
7493          * before atomic variable goto zero, we must
7494          * make sure dip->errors is perceived to be set.
7495          */
7496         smp_mb__before_atomic();
7497         if (atomic_dec_and_test(&dip->pending_bios))
7498                 bio_io_error(dip->orig_bio);
7499
7500         /* bio_end_io() will handle error, so we needn't return it */
7501         return 0;
7502 }
7503
7504 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7505                                 struct inode *inode, loff_t file_offset)
7506 {
7507         struct btrfs_root *root = BTRFS_I(inode)->root;
7508         struct btrfs_dio_private *dip;
7509         struct bio *io_bio;
7510         int skip_sum;
7511         int sum_len;
7512         int write = rw & REQ_WRITE;
7513         int ret = 0;
7514         u16 csum_size;
7515
7516         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7517
7518         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7519         if (!io_bio) {
7520                 ret = -ENOMEM;
7521                 goto free_ordered;
7522         }
7523
7524         if (!skip_sum && !write) {
7525                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7526                 sum_len = dio_bio->bi_iter.bi_size >>
7527                         inode->i_sb->s_blocksize_bits;
7528                 sum_len *= csum_size;
7529         } else {
7530                 sum_len = 0;
7531         }
7532
7533         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7534         if (!dip) {
7535                 ret = -ENOMEM;
7536                 goto free_io_bio;
7537         }
7538
7539         dip->private = dio_bio->bi_private;
7540         dip->inode = inode;
7541         dip->logical_offset = file_offset;
7542         dip->bytes = dio_bio->bi_iter.bi_size;
7543         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7544         io_bio->bi_private = dip;
7545         dip->errors = 0;
7546         dip->orig_bio = io_bio;
7547         dip->dio_bio = dio_bio;
7548         atomic_set(&dip->pending_bios, 0);
7549
7550         if (write)
7551                 io_bio->bi_end_io = btrfs_endio_direct_write;
7552         else
7553                 io_bio->bi_end_io = btrfs_endio_direct_read;
7554
7555         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7556         if (!ret)
7557                 return;
7558
7559 free_io_bio:
7560         bio_put(io_bio);
7561
7562 free_ordered:
7563         /*
7564          * If this is a write, we need to clean up the reserved space and kill
7565          * the ordered extent.
7566          */
7567         if (write) {
7568                 struct btrfs_ordered_extent *ordered;
7569                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7570                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7571                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7572                         btrfs_free_reserved_extent(root, ordered->start,
7573                                                    ordered->disk_len, 1);
7574                 btrfs_put_ordered_extent(ordered);
7575                 btrfs_put_ordered_extent(ordered);
7576         }
7577         bio_endio(dio_bio, ret);
7578 }
7579
7580 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7581                         const struct iov_iter *iter, loff_t offset)
7582 {
7583         int seg;
7584         int i;
7585         unsigned blocksize_mask = root->sectorsize - 1;
7586         ssize_t retval = -EINVAL;
7587
7588         if (offset & blocksize_mask)
7589                 goto out;
7590
7591         if (iov_iter_alignment(iter) & blocksize_mask)
7592                 goto out;
7593
7594         /* If this is a write we don't need to check anymore */
7595         if (rw & WRITE)
7596                 return 0;
7597         /*
7598          * Check to make sure we don't have duplicate iov_base's in this
7599          * iovec, if so return EINVAL, otherwise we'll get csum errors
7600          * when reading back.
7601          */
7602         for (seg = 0; seg < iter->nr_segs; seg++) {
7603                 for (i = seg + 1; i < iter->nr_segs; i++) {
7604                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
7605                                 goto out;
7606                 }
7607         }
7608         retval = 0;
7609 out:
7610         return retval;
7611 }
7612
7613 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7614                         struct iov_iter *iter, loff_t offset)
7615 {
7616         struct file *file = iocb->ki_filp;
7617         struct inode *inode = file->f_mapping->host;
7618         size_t count = 0;
7619         int flags = 0;
7620         bool wakeup = true;
7621         bool relock = false;
7622         ssize_t ret;
7623
7624         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
7625                 return 0;
7626
7627         atomic_inc(&inode->i_dio_count);
7628         smp_mb__after_atomic();
7629
7630         /*
7631          * The generic stuff only does filemap_write_and_wait_range, which
7632          * isn't enough if we've written compressed pages to this area, so
7633          * we need to flush the dirty pages again to make absolutely sure
7634          * that any outstanding dirty pages are on disk.
7635          */
7636         count = iov_iter_count(iter);
7637         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7638                      &BTRFS_I(inode)->runtime_flags))
7639                 filemap_fdatawrite_range(inode->i_mapping, offset,
7640                                          offset + count - 1);
7641
7642         if (rw & WRITE) {
7643                 /*
7644                  * If the write DIO is beyond the EOF, we need update
7645                  * the isize, but it is protected by i_mutex. So we can
7646                  * not unlock the i_mutex at this case.
7647                  */
7648                 if (offset + count <= inode->i_size) {
7649                         mutex_unlock(&inode->i_mutex);
7650                         relock = true;
7651                 }
7652                 ret = btrfs_delalloc_reserve_space(inode, count);
7653                 if (ret)
7654                         goto out;
7655         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7656                                      &BTRFS_I(inode)->runtime_flags))) {
7657                 inode_dio_done(inode);
7658                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7659                 wakeup = false;
7660         }
7661
7662         ret = __blockdev_direct_IO(rw, iocb, inode,
7663                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7664                         iter, offset, btrfs_get_blocks_direct, NULL,
7665                         btrfs_submit_direct, flags);
7666         if (rw & WRITE) {
7667                 if (ret < 0 && ret != -EIOCBQUEUED)
7668                         btrfs_delalloc_release_space(inode, count);
7669                 else if (ret >= 0 && (size_t)ret < count)
7670                         btrfs_delalloc_release_space(inode,
7671                                                      count - (size_t)ret);
7672                 else
7673                         btrfs_delalloc_release_metadata(inode, 0);
7674         }
7675 out:
7676         if (wakeup)
7677                 inode_dio_done(inode);
7678         if (relock)
7679                 mutex_lock(&inode->i_mutex);
7680
7681         return ret;
7682 }
7683
7684 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7685
7686 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7687                 __u64 start, __u64 len)
7688 {
7689         int     ret;
7690
7691         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7692         if (ret)
7693                 return ret;
7694
7695         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7696 }
7697
7698 int btrfs_readpage(struct file *file, struct page *page)
7699 {
7700         struct extent_io_tree *tree;
7701         tree = &BTRFS_I(page->mapping->host)->io_tree;
7702         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7703 }
7704
7705 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7706 {
7707         struct extent_io_tree *tree;
7708
7709
7710         if (current->flags & PF_MEMALLOC) {
7711                 redirty_page_for_writepage(wbc, page);
7712                 unlock_page(page);
7713                 return 0;
7714         }
7715         tree = &BTRFS_I(page->mapping->host)->io_tree;
7716         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7717 }
7718
7719 static int btrfs_writepages(struct address_space *mapping,
7720                             struct writeback_control *wbc)
7721 {
7722         struct extent_io_tree *tree;
7723
7724         tree = &BTRFS_I(mapping->host)->io_tree;
7725         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7726 }
7727
7728 static int
7729 btrfs_readpages(struct file *file, struct address_space *mapping,
7730                 struct list_head *pages, unsigned nr_pages)
7731 {
7732         struct extent_io_tree *tree;
7733         tree = &BTRFS_I(mapping->host)->io_tree;
7734         return extent_readpages(tree, mapping, pages, nr_pages,
7735                                 btrfs_get_extent);
7736 }
7737 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7738 {
7739         struct extent_io_tree *tree;
7740         struct extent_map_tree *map;
7741         int ret;
7742
7743         tree = &BTRFS_I(page->mapping->host)->io_tree;
7744         map = &BTRFS_I(page->mapping->host)->extent_tree;
7745         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7746         if (ret == 1) {
7747                 ClearPagePrivate(page);
7748                 set_page_private(page, 0);
7749                 page_cache_release(page);
7750         }
7751         return ret;
7752 }
7753
7754 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7755 {
7756         if (PageWriteback(page) || PageDirty(page))
7757                 return 0;
7758         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7759 }
7760
7761 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7762                                  unsigned int length)
7763 {
7764         struct inode *inode = page->mapping->host;
7765         struct extent_io_tree *tree;
7766         struct btrfs_ordered_extent *ordered;
7767         struct extent_state *cached_state = NULL;
7768         u64 page_start = page_offset(page);
7769         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7770         int inode_evicting = inode->i_state & I_FREEING;
7771
7772         /*
7773          * we have the page locked, so new writeback can't start,
7774          * and the dirty bit won't be cleared while we are here.
7775          *
7776          * Wait for IO on this page so that we can safely clear
7777          * the PagePrivate2 bit and do ordered accounting
7778          */
7779         wait_on_page_writeback(page);
7780
7781         tree = &BTRFS_I(inode)->io_tree;
7782         if (offset) {
7783                 btrfs_releasepage(page, GFP_NOFS);
7784                 return;
7785         }
7786
7787         if (!inode_evicting)
7788                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7789         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7790         if (ordered) {
7791                 /*
7792                  * IO on this page will never be started, so we need
7793                  * to account for any ordered extents now
7794                  */
7795                 if (!inode_evicting)
7796                         clear_extent_bit(tree, page_start, page_end,
7797                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7798                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7799                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7800                                          GFP_NOFS);
7801                 /*
7802                  * whoever cleared the private bit is responsible
7803                  * for the finish_ordered_io
7804                  */
7805                 if (TestClearPagePrivate2(page)) {
7806                         struct btrfs_ordered_inode_tree *tree;
7807                         u64 new_len;
7808
7809                         tree = &BTRFS_I(inode)->ordered_tree;
7810
7811                         spin_lock_irq(&tree->lock);
7812                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7813                         new_len = page_start - ordered->file_offset;
7814                         if (new_len < ordered->truncated_len)
7815                                 ordered->truncated_len = new_len;
7816                         spin_unlock_irq(&tree->lock);
7817
7818                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7819                                                            page_start,
7820                                                            PAGE_CACHE_SIZE, 1))
7821                                 btrfs_finish_ordered_io(ordered);
7822                 }
7823                 btrfs_put_ordered_extent(ordered);
7824                 if (!inode_evicting) {
7825                         cached_state = NULL;
7826                         lock_extent_bits(tree, page_start, page_end, 0,
7827                                          &cached_state);
7828                 }
7829         }
7830
7831         if (!inode_evicting) {
7832                 clear_extent_bit(tree, page_start, page_end,
7833                                  EXTENT_LOCKED | EXTENT_DIRTY |
7834                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7835                                  EXTENT_DEFRAG, 1, 1,
7836                                  &cached_state, GFP_NOFS);
7837
7838                 __btrfs_releasepage(page, GFP_NOFS);
7839         }
7840
7841         ClearPageChecked(page);
7842         if (PagePrivate(page)) {
7843                 ClearPagePrivate(page);
7844                 set_page_private(page, 0);
7845                 page_cache_release(page);
7846         }
7847 }
7848
7849 /*
7850  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7851  * called from a page fault handler when a page is first dirtied. Hence we must
7852  * be careful to check for EOF conditions here. We set the page up correctly
7853  * for a written page which means we get ENOSPC checking when writing into
7854  * holes and correct delalloc and unwritten extent mapping on filesystems that
7855  * support these features.
7856  *
7857  * We are not allowed to take the i_mutex here so we have to play games to
7858  * protect against truncate races as the page could now be beyond EOF.  Because
7859  * vmtruncate() writes the inode size before removing pages, once we have the
7860  * page lock we can determine safely if the page is beyond EOF. If it is not
7861  * beyond EOF, then the page is guaranteed safe against truncation until we
7862  * unlock the page.
7863  */
7864 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7865 {
7866         struct page *page = vmf->page;
7867         struct inode *inode = file_inode(vma->vm_file);
7868         struct btrfs_root *root = BTRFS_I(inode)->root;
7869         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7870         struct btrfs_ordered_extent *ordered;
7871         struct extent_state *cached_state = NULL;
7872         char *kaddr;
7873         unsigned long zero_start;
7874         loff_t size;
7875         int ret;
7876         int reserved = 0;
7877         u64 page_start;
7878         u64 page_end;
7879
7880         sb_start_pagefault(inode->i_sb);
7881         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7882         if (!ret) {
7883                 ret = file_update_time(vma->vm_file);
7884                 reserved = 1;
7885         }
7886         if (ret) {
7887                 if (ret == -ENOMEM)
7888                         ret = VM_FAULT_OOM;
7889                 else /* -ENOSPC, -EIO, etc */
7890                         ret = VM_FAULT_SIGBUS;
7891                 if (reserved)
7892                         goto out;
7893                 goto out_noreserve;
7894         }
7895
7896         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7897 again:
7898         lock_page(page);
7899         size = i_size_read(inode);
7900         page_start = page_offset(page);
7901         page_end = page_start + PAGE_CACHE_SIZE - 1;
7902
7903         if ((page->mapping != inode->i_mapping) ||
7904             (page_start >= size)) {
7905                 /* page got truncated out from underneath us */
7906                 goto out_unlock;
7907         }
7908         wait_on_page_writeback(page);
7909
7910         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7911         set_page_extent_mapped(page);
7912
7913         /*
7914          * we can't set the delalloc bits if there are pending ordered
7915          * extents.  Drop our locks and wait for them to finish
7916          */
7917         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7918         if (ordered) {
7919                 unlock_extent_cached(io_tree, page_start, page_end,
7920                                      &cached_state, GFP_NOFS);
7921                 unlock_page(page);
7922                 btrfs_start_ordered_extent(inode, ordered, 1);
7923                 btrfs_put_ordered_extent(ordered);
7924                 goto again;
7925         }
7926
7927         /*
7928          * XXX - page_mkwrite gets called every time the page is dirtied, even
7929          * if it was already dirty, so for space accounting reasons we need to
7930          * clear any delalloc bits for the range we are fixing to save.  There
7931          * is probably a better way to do this, but for now keep consistent with
7932          * prepare_pages in the normal write path.
7933          */
7934         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7935                           EXTENT_DIRTY | EXTENT_DELALLOC |
7936                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7937                           0, 0, &cached_state, GFP_NOFS);
7938
7939         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7940                                         &cached_state);
7941         if (ret) {
7942                 unlock_extent_cached(io_tree, page_start, page_end,
7943                                      &cached_state, GFP_NOFS);
7944                 ret = VM_FAULT_SIGBUS;
7945                 goto out_unlock;
7946         }
7947         ret = 0;
7948
7949         /* page is wholly or partially inside EOF */
7950         if (page_start + PAGE_CACHE_SIZE > size)
7951                 zero_start = size & ~PAGE_CACHE_MASK;
7952         else
7953                 zero_start = PAGE_CACHE_SIZE;
7954
7955         if (zero_start != PAGE_CACHE_SIZE) {
7956                 kaddr = kmap(page);
7957                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7958                 flush_dcache_page(page);
7959                 kunmap(page);
7960         }
7961         ClearPageChecked(page);
7962         set_page_dirty(page);
7963         SetPageUptodate(page);
7964
7965         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7966         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7967         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7968
7969         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7970
7971 out_unlock:
7972         if (!ret) {
7973                 sb_end_pagefault(inode->i_sb);
7974                 return VM_FAULT_LOCKED;
7975         }
7976         unlock_page(page);
7977 out:
7978         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7979 out_noreserve:
7980         sb_end_pagefault(inode->i_sb);
7981         return ret;
7982 }
7983
7984 static int btrfs_truncate(struct inode *inode)
7985 {
7986         struct btrfs_root *root = BTRFS_I(inode)->root;
7987         struct btrfs_block_rsv *rsv;
7988         int ret = 0;
7989         int err = 0;
7990         struct btrfs_trans_handle *trans;
7991         u64 mask = root->sectorsize - 1;
7992         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7993
7994         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7995                                        (u64)-1);
7996         if (ret)
7997                 return ret;
7998
7999         /*
8000          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
8001          * 3 things going on here
8002          *
8003          * 1) We need to reserve space for our orphan item and the space to
8004          * delete our orphan item.  Lord knows we don't want to have a dangling
8005          * orphan item because we didn't reserve space to remove it.
8006          *
8007          * 2) We need to reserve space to update our inode.
8008          *
8009          * 3) We need to have something to cache all the space that is going to
8010          * be free'd up by the truncate operation, but also have some slack
8011          * space reserved in case it uses space during the truncate (thank you
8012          * very much snapshotting).
8013          *
8014          * And we need these to all be seperate.  The fact is we can use alot of
8015          * space doing the truncate, and we have no earthly idea how much space
8016          * we will use, so we need the truncate reservation to be seperate so it
8017          * doesn't end up using space reserved for updating the inode or
8018          * removing the orphan item.  We also need to be able to stop the
8019          * transaction and start a new one, which means we need to be able to
8020          * update the inode several times, and we have no idea of knowing how
8021          * many times that will be, so we can't just reserve 1 item for the
8022          * entirety of the opration, so that has to be done seperately as well.
8023          * Then there is the orphan item, which does indeed need to be held on
8024          * to for the whole operation, and we need nobody to touch this reserved
8025          * space except the orphan code.
8026          *
8027          * So that leaves us with
8028          *
8029          * 1) root->orphan_block_rsv - for the orphan deletion.
8030          * 2) rsv - for the truncate reservation, which we will steal from the
8031          * transaction reservation.
8032          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8033          * updating the inode.
8034          */
8035         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8036         if (!rsv)
8037                 return -ENOMEM;
8038         rsv->size = min_size;
8039         rsv->failfast = 1;
8040
8041         /*
8042          * 1 for the truncate slack space
8043          * 1 for updating the inode.
8044          */
8045         trans = btrfs_start_transaction(root, 2);
8046         if (IS_ERR(trans)) {
8047                 err = PTR_ERR(trans);
8048                 goto out;
8049         }
8050
8051         /* Migrate the slack space for the truncate to our reserve */
8052         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8053                                       min_size);
8054         BUG_ON(ret);
8055
8056         /*
8057          * So if we truncate and then write and fsync we normally would just
8058          * write the extents that changed, which is a problem if we need to
8059          * first truncate that entire inode.  So set this flag so we write out
8060          * all of the extents in the inode to the sync log so we're completely
8061          * safe.
8062          */
8063         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8064         trans->block_rsv = rsv;
8065
8066         while (1) {
8067                 ret = btrfs_truncate_inode_items(trans, root, inode,
8068                                                  inode->i_size,
8069                                                  BTRFS_EXTENT_DATA_KEY);
8070                 if (ret != -ENOSPC) {
8071                         err = ret;
8072                         break;
8073                 }
8074
8075                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8076                 ret = btrfs_update_inode(trans, root, inode);
8077                 if (ret) {
8078                         err = ret;
8079                         break;
8080                 }
8081
8082                 btrfs_end_transaction(trans, root);
8083                 btrfs_btree_balance_dirty(root);
8084
8085                 trans = btrfs_start_transaction(root, 2);
8086                 if (IS_ERR(trans)) {
8087                         ret = err = PTR_ERR(trans);
8088                         trans = NULL;
8089                         break;
8090                 }
8091
8092                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8093                                               rsv, min_size);
8094                 BUG_ON(ret);    /* shouldn't happen */
8095                 trans->block_rsv = rsv;
8096         }
8097
8098         if (ret == 0 && inode->i_nlink > 0) {
8099                 trans->block_rsv = root->orphan_block_rsv;
8100                 ret = btrfs_orphan_del(trans, inode);
8101                 if (ret)
8102                         err = ret;
8103         }
8104
8105         if (trans) {
8106                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8107                 ret = btrfs_update_inode(trans, root, inode);
8108                 if (ret && !err)
8109                         err = ret;
8110
8111                 ret = btrfs_end_transaction(trans, root);
8112                 btrfs_btree_balance_dirty(root);
8113         }
8114
8115 out:
8116         btrfs_free_block_rsv(root, rsv);
8117
8118         if (ret && !err)
8119                 err = ret;
8120
8121         return err;
8122 }
8123
8124 /*
8125  * create a new subvolume directory/inode (helper for the ioctl).
8126  */
8127 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8128                              struct btrfs_root *new_root,
8129                              struct btrfs_root *parent_root,
8130                              u64 new_dirid)
8131 {
8132         struct inode *inode;
8133         int err;
8134         u64 index = 0;
8135
8136         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8137                                 new_dirid, new_dirid,
8138                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8139                                 &index);
8140         if (IS_ERR(inode))
8141                 return PTR_ERR(inode);
8142         inode->i_op = &btrfs_dir_inode_operations;
8143         inode->i_fop = &btrfs_dir_file_operations;
8144
8145         set_nlink(inode, 1);
8146         btrfs_i_size_write(inode, 0);
8147         unlock_new_inode(inode);
8148
8149         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8150         if (err)
8151                 btrfs_err(new_root->fs_info,
8152                           "error inheriting subvolume %llu properties: %d",
8153                           new_root->root_key.objectid, err);
8154
8155         err = btrfs_update_inode(trans, new_root, inode);
8156
8157         iput(inode);
8158         return err;
8159 }
8160
8161 struct inode *btrfs_alloc_inode(struct super_block *sb)
8162 {
8163         struct btrfs_inode *ei;
8164         struct inode *inode;
8165
8166         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8167         if (!ei)
8168                 return NULL;
8169
8170         ei->root = NULL;
8171         ei->generation = 0;
8172         ei->last_trans = 0;
8173         ei->last_sub_trans = 0;
8174         ei->logged_trans = 0;
8175         ei->delalloc_bytes = 0;
8176         ei->disk_i_size = 0;
8177         ei->flags = 0;
8178         ei->csum_bytes = 0;
8179         ei->index_cnt = (u64)-1;
8180         ei->dir_index = 0;
8181         ei->last_unlink_trans = 0;
8182         ei->last_log_commit = 0;
8183
8184         spin_lock_init(&ei->lock);
8185         ei->outstanding_extents = 0;
8186         ei->reserved_extents = 0;
8187
8188         ei->runtime_flags = 0;
8189         ei->force_compress = BTRFS_COMPRESS_NONE;
8190
8191         ei->delayed_node = NULL;
8192
8193         inode = &ei->vfs_inode;
8194         extent_map_tree_init(&ei->extent_tree);
8195         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8196         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8197         ei->io_tree.track_uptodate = 1;
8198         ei->io_failure_tree.track_uptodate = 1;
8199         atomic_set(&ei->sync_writers, 0);
8200         mutex_init(&ei->log_mutex);
8201         mutex_init(&ei->delalloc_mutex);
8202         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8203         INIT_LIST_HEAD(&ei->delalloc_inodes);
8204         RB_CLEAR_NODE(&ei->rb_node);
8205
8206         return inode;
8207 }
8208
8209 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8210 void btrfs_test_destroy_inode(struct inode *inode)
8211 {
8212         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8213         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8214 }
8215 #endif
8216
8217 static void btrfs_i_callback(struct rcu_head *head)
8218 {
8219         struct inode *inode = container_of(head, struct inode, i_rcu);
8220         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8221 }
8222
8223 void btrfs_destroy_inode(struct inode *inode)
8224 {
8225         struct btrfs_ordered_extent *ordered;
8226         struct btrfs_root *root = BTRFS_I(inode)->root;
8227
8228         WARN_ON(!hlist_empty(&inode->i_dentry));
8229         WARN_ON(inode->i_data.nrpages);
8230         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8231         WARN_ON(BTRFS_I(inode)->reserved_extents);
8232         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8233         WARN_ON(BTRFS_I(inode)->csum_bytes);
8234
8235         /*
8236          * This can happen where we create an inode, but somebody else also
8237          * created the same inode and we need to destroy the one we already
8238          * created.
8239          */
8240         if (!root)
8241                 goto free;
8242
8243         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8244                      &BTRFS_I(inode)->runtime_flags)) {
8245                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8246                         btrfs_ino(inode));
8247                 atomic_dec(&root->orphan_inodes);
8248         }
8249
8250         while (1) {
8251                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8252                 if (!ordered)
8253                         break;
8254                 else {
8255                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8256                                 ordered->file_offset, ordered->len);
8257                         btrfs_remove_ordered_extent(inode, ordered);
8258                         btrfs_put_ordered_extent(ordered);
8259                         btrfs_put_ordered_extent(ordered);
8260                 }
8261         }
8262         inode_tree_del(inode);
8263         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8264 free:
8265         call_rcu(&inode->i_rcu, btrfs_i_callback);
8266 }
8267
8268 int btrfs_drop_inode(struct inode *inode)
8269 {
8270         struct btrfs_root *root = BTRFS_I(inode)->root;
8271
8272         if (root == NULL)
8273                 return 1;
8274
8275         /* the snap/subvol tree is on deleting */
8276         if (btrfs_root_refs(&root->root_item) == 0)
8277                 return 1;
8278         else
8279                 return generic_drop_inode(inode);
8280 }
8281
8282 static void init_once(void *foo)
8283 {
8284         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8285
8286         inode_init_once(&ei->vfs_inode);
8287 }
8288
8289 void btrfs_destroy_cachep(void)
8290 {
8291         /*
8292          * Make sure all delayed rcu free inodes are flushed before we
8293          * destroy cache.
8294          */
8295         rcu_barrier();
8296         if (btrfs_inode_cachep)
8297                 kmem_cache_destroy(btrfs_inode_cachep);
8298         if (btrfs_trans_handle_cachep)
8299                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8300         if (btrfs_transaction_cachep)
8301                 kmem_cache_destroy(btrfs_transaction_cachep);
8302         if (btrfs_path_cachep)
8303                 kmem_cache_destroy(btrfs_path_cachep);
8304         if (btrfs_free_space_cachep)
8305                 kmem_cache_destroy(btrfs_free_space_cachep);
8306         if (btrfs_delalloc_work_cachep)
8307                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8308 }
8309
8310 int btrfs_init_cachep(void)
8311 {
8312         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8313                         sizeof(struct btrfs_inode), 0,
8314                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8315         if (!btrfs_inode_cachep)
8316                 goto fail;
8317
8318         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8319                         sizeof(struct btrfs_trans_handle), 0,
8320                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8321         if (!btrfs_trans_handle_cachep)
8322                 goto fail;
8323
8324         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8325                         sizeof(struct btrfs_transaction), 0,
8326                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8327         if (!btrfs_transaction_cachep)
8328                 goto fail;
8329
8330         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8331                         sizeof(struct btrfs_path), 0,
8332                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8333         if (!btrfs_path_cachep)
8334                 goto fail;
8335
8336         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8337                         sizeof(struct btrfs_free_space), 0,
8338                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8339         if (!btrfs_free_space_cachep)
8340                 goto fail;
8341
8342         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8343                         sizeof(struct btrfs_delalloc_work), 0,
8344                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8345                         NULL);
8346         if (!btrfs_delalloc_work_cachep)
8347                 goto fail;
8348
8349         return 0;
8350 fail:
8351         btrfs_destroy_cachep();
8352         return -ENOMEM;
8353 }
8354
8355 static int btrfs_getattr(struct vfsmount *mnt,
8356                          struct dentry *dentry, struct kstat *stat)
8357 {
8358         u64 delalloc_bytes;
8359         struct inode *inode = dentry->d_inode;
8360         u32 blocksize = inode->i_sb->s_blocksize;
8361
8362         generic_fillattr(inode, stat);
8363         stat->dev = BTRFS_I(inode)->root->anon_dev;
8364         stat->blksize = PAGE_CACHE_SIZE;
8365
8366         spin_lock(&BTRFS_I(inode)->lock);
8367         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8368         spin_unlock(&BTRFS_I(inode)->lock);
8369         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8370                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8371         return 0;
8372 }
8373
8374 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8375                            struct inode *new_dir, struct dentry *new_dentry)
8376 {
8377         struct btrfs_trans_handle *trans;
8378         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8379         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8380         struct inode *new_inode = new_dentry->d_inode;
8381         struct inode *old_inode = old_dentry->d_inode;
8382         struct timespec ctime = CURRENT_TIME;
8383         u64 index = 0;
8384         u64 root_objectid;
8385         int ret;
8386         u64 old_ino = btrfs_ino(old_inode);
8387
8388         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8389                 return -EPERM;
8390
8391         /* we only allow rename subvolume link between subvolumes */
8392         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8393                 return -EXDEV;
8394
8395         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8396             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8397                 return -ENOTEMPTY;
8398
8399         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8400             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8401                 return -ENOTEMPTY;
8402
8403
8404         /* check for collisions, even if the  name isn't there */
8405         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8406                              new_dentry->d_name.name,
8407                              new_dentry->d_name.len);
8408
8409         if (ret) {
8410                 if (ret == -EEXIST) {
8411                         /* we shouldn't get
8412                          * eexist without a new_inode */
8413                         if (WARN_ON(!new_inode)) {
8414                                 return ret;
8415                         }
8416                 } else {
8417                         /* maybe -EOVERFLOW */
8418                         return ret;
8419                 }
8420         }
8421         ret = 0;
8422
8423         /*
8424          * we're using rename to replace one file with another.  Start IO on it
8425          * now so  we don't add too much work to the end of the transaction
8426          */
8427         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8428                 filemap_flush(old_inode->i_mapping);
8429
8430         /* close the racy window with snapshot create/destroy ioctl */
8431         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8432                 down_read(&root->fs_info->subvol_sem);
8433         /*
8434          * We want to reserve the absolute worst case amount of items.  So if
8435          * both inodes are subvols and we need to unlink them then that would
8436          * require 4 item modifications, but if they are both normal inodes it
8437          * would require 5 item modifications, so we'll assume their normal
8438          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8439          * should cover the worst case number of items we'll modify.
8440          */
8441         trans = btrfs_start_transaction(root, 11);
8442         if (IS_ERR(trans)) {
8443                 ret = PTR_ERR(trans);
8444                 goto out_notrans;
8445         }
8446
8447         if (dest != root)
8448                 btrfs_record_root_in_trans(trans, dest);
8449
8450         ret = btrfs_set_inode_index(new_dir, &index);
8451         if (ret)
8452                 goto out_fail;
8453
8454         BTRFS_I(old_inode)->dir_index = 0ULL;
8455         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8456                 /* force full log commit if subvolume involved. */
8457                 btrfs_set_log_full_commit(root->fs_info, trans);
8458         } else {
8459                 ret = btrfs_insert_inode_ref(trans, dest,
8460                                              new_dentry->d_name.name,
8461                                              new_dentry->d_name.len,
8462                                              old_ino,
8463                                              btrfs_ino(new_dir), index);
8464                 if (ret)
8465                         goto out_fail;
8466                 /*
8467                  * this is an ugly little race, but the rename is required
8468                  * to make sure that if we crash, the inode is either at the
8469                  * old name or the new one.  pinning the log transaction lets
8470                  * us make sure we don't allow a log commit to come in after
8471                  * we unlink the name but before we add the new name back in.
8472                  */
8473                 btrfs_pin_log_trans(root);
8474         }
8475
8476         inode_inc_iversion(old_dir);
8477         inode_inc_iversion(new_dir);
8478         inode_inc_iversion(old_inode);
8479         old_dir->i_ctime = old_dir->i_mtime = ctime;
8480         new_dir->i_ctime = new_dir->i_mtime = ctime;
8481         old_inode->i_ctime = ctime;
8482
8483         if (old_dentry->d_parent != new_dentry->d_parent)
8484                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8485
8486         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8487                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8488                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8489                                         old_dentry->d_name.name,
8490                                         old_dentry->d_name.len);
8491         } else {
8492                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8493                                         old_dentry->d_inode,
8494                                         old_dentry->d_name.name,
8495                                         old_dentry->d_name.len);
8496                 if (!ret)
8497                         ret = btrfs_update_inode(trans, root, old_inode);
8498         }
8499         if (ret) {
8500                 btrfs_abort_transaction(trans, root, ret);
8501                 goto out_fail;
8502         }
8503
8504         if (new_inode) {
8505                 inode_inc_iversion(new_inode);
8506                 new_inode->i_ctime = CURRENT_TIME;
8507                 if (unlikely(btrfs_ino(new_inode) ==
8508                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8509                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8510                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8511                                                 root_objectid,
8512                                                 new_dentry->d_name.name,
8513                                                 new_dentry->d_name.len);
8514                         BUG_ON(new_inode->i_nlink == 0);
8515                 } else {
8516                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8517                                                  new_dentry->d_inode,
8518                                                  new_dentry->d_name.name,
8519                                                  new_dentry->d_name.len);
8520                 }
8521                 if (!ret && new_inode->i_nlink == 0)
8522                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8523                 if (ret) {
8524                         btrfs_abort_transaction(trans, root, ret);
8525                         goto out_fail;
8526                 }
8527         }
8528
8529         ret = btrfs_add_link(trans, new_dir, old_inode,
8530                              new_dentry->d_name.name,
8531                              new_dentry->d_name.len, 0, index);
8532         if (ret) {
8533                 btrfs_abort_transaction(trans, root, ret);
8534                 goto out_fail;
8535         }
8536
8537         if (old_inode->i_nlink == 1)
8538                 BTRFS_I(old_inode)->dir_index = index;
8539
8540         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8541                 struct dentry *parent = new_dentry->d_parent;
8542                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8543                 btrfs_end_log_trans(root);
8544         }
8545 out_fail:
8546         btrfs_end_transaction(trans, root);
8547 out_notrans:
8548         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8549                 up_read(&root->fs_info->subvol_sem);
8550
8551         return ret;
8552 }
8553
8554 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
8555                          struct inode *new_dir, struct dentry *new_dentry,
8556                          unsigned int flags)
8557 {
8558         if (flags & ~RENAME_NOREPLACE)
8559                 return -EINVAL;
8560
8561         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
8562 }
8563
8564 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8565 {
8566         struct btrfs_delalloc_work *delalloc_work;
8567         struct inode *inode;
8568
8569         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8570                                      work);
8571         inode = delalloc_work->inode;
8572         if (delalloc_work->wait) {
8573                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8574         } else {
8575                 filemap_flush(inode->i_mapping);
8576                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8577                              &BTRFS_I(inode)->runtime_flags))
8578                         filemap_flush(inode->i_mapping);
8579         }
8580
8581         if (delalloc_work->delay_iput)
8582                 btrfs_add_delayed_iput(inode);
8583         else
8584                 iput(inode);
8585         complete(&delalloc_work->completion);
8586 }
8587
8588 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8589                                                     int wait, int delay_iput)
8590 {
8591         struct btrfs_delalloc_work *work;
8592
8593         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8594         if (!work)
8595                 return NULL;
8596
8597         init_completion(&work->completion);
8598         INIT_LIST_HEAD(&work->list);
8599         work->inode = inode;
8600         work->wait = wait;
8601         work->delay_iput = delay_iput;
8602         WARN_ON_ONCE(!inode);
8603         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
8604                         btrfs_run_delalloc_work, NULL, NULL);
8605
8606         return work;
8607 }
8608
8609 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8610 {
8611         wait_for_completion(&work->completion);
8612         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8613 }
8614
8615 /*
8616  * some fairly slow code that needs optimization. This walks the list
8617  * of all the inodes with pending delalloc and forces them to disk.
8618  */
8619 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
8620                                    int nr)
8621 {
8622         struct btrfs_inode *binode;
8623         struct inode *inode;
8624         struct btrfs_delalloc_work *work, *next;
8625         struct list_head works;
8626         struct list_head splice;
8627         int ret = 0;
8628
8629         INIT_LIST_HEAD(&works);
8630         INIT_LIST_HEAD(&splice);
8631
8632         mutex_lock(&root->delalloc_mutex);
8633         spin_lock(&root->delalloc_lock);
8634         list_splice_init(&root->delalloc_inodes, &splice);
8635         while (!list_empty(&splice)) {
8636                 binode = list_entry(splice.next, struct btrfs_inode,
8637                                     delalloc_inodes);
8638
8639                 list_move_tail(&binode->delalloc_inodes,
8640                                &root->delalloc_inodes);
8641                 inode = igrab(&binode->vfs_inode);
8642                 if (!inode) {
8643                         cond_resched_lock(&root->delalloc_lock);
8644                         continue;
8645                 }
8646                 spin_unlock(&root->delalloc_lock);
8647
8648                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8649                 if (unlikely(!work)) {
8650                         if (delay_iput)
8651                                 btrfs_add_delayed_iput(inode);
8652                         else
8653                                 iput(inode);
8654                         ret = -ENOMEM;
8655                         goto out;
8656                 }
8657                 list_add_tail(&work->list, &works);
8658                 btrfs_queue_work(root->fs_info->flush_workers,
8659                                  &work->work);
8660                 ret++;
8661                 if (nr != -1 && ret >= nr)
8662                         goto out;
8663                 cond_resched();
8664                 spin_lock(&root->delalloc_lock);
8665         }
8666         spin_unlock(&root->delalloc_lock);
8667
8668 out:
8669         list_for_each_entry_safe(work, next, &works, list) {
8670                 list_del_init(&work->list);
8671                 btrfs_wait_and_free_delalloc_work(work);
8672         }
8673
8674         if (!list_empty_careful(&splice)) {
8675                 spin_lock(&root->delalloc_lock);
8676                 list_splice_tail(&splice, &root->delalloc_inodes);
8677                 spin_unlock(&root->delalloc_lock);
8678         }
8679         mutex_unlock(&root->delalloc_mutex);
8680         return ret;
8681 }
8682
8683 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8684 {
8685         int ret;
8686
8687         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8688                 return -EROFS;
8689
8690         ret = __start_delalloc_inodes(root, delay_iput, -1);
8691         if (ret > 0)
8692                 ret = 0;
8693         /*
8694          * the filemap_flush will queue IO into the worker threads, but
8695          * we have to make sure the IO is actually started and that
8696          * ordered extents get created before we return
8697          */
8698         atomic_inc(&root->fs_info->async_submit_draining);
8699         while (atomic_read(&root->fs_info->nr_async_submits) ||
8700               atomic_read(&root->fs_info->async_delalloc_pages)) {
8701                 wait_event(root->fs_info->async_submit_wait,
8702                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8703                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8704         }
8705         atomic_dec(&root->fs_info->async_submit_draining);
8706         return ret;
8707 }
8708
8709 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
8710                                int nr)
8711 {
8712         struct btrfs_root *root;
8713         struct list_head splice;
8714         int ret;
8715
8716         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8717                 return -EROFS;
8718
8719         INIT_LIST_HEAD(&splice);
8720
8721         mutex_lock(&fs_info->delalloc_root_mutex);
8722         spin_lock(&fs_info->delalloc_root_lock);
8723         list_splice_init(&fs_info->delalloc_roots, &splice);
8724         while (!list_empty(&splice) && nr) {
8725                 root = list_first_entry(&splice, struct btrfs_root,
8726                                         delalloc_root);
8727                 root = btrfs_grab_fs_root(root);
8728                 BUG_ON(!root);
8729                 list_move_tail(&root->delalloc_root,
8730                                &fs_info->delalloc_roots);
8731                 spin_unlock(&fs_info->delalloc_root_lock);
8732
8733                 ret = __start_delalloc_inodes(root, delay_iput, nr);
8734                 btrfs_put_fs_root(root);
8735                 if (ret < 0)
8736                         goto out;
8737
8738                 if (nr != -1) {
8739                         nr -= ret;
8740                         WARN_ON(nr < 0);
8741                 }
8742                 spin_lock(&fs_info->delalloc_root_lock);
8743         }
8744         spin_unlock(&fs_info->delalloc_root_lock);
8745
8746         ret = 0;
8747         atomic_inc(&fs_info->async_submit_draining);
8748         while (atomic_read(&fs_info->nr_async_submits) ||
8749               atomic_read(&fs_info->async_delalloc_pages)) {
8750                 wait_event(fs_info->async_submit_wait,
8751                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8752                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8753         }
8754         atomic_dec(&fs_info->async_submit_draining);
8755 out:
8756         if (!list_empty_careful(&splice)) {
8757                 spin_lock(&fs_info->delalloc_root_lock);
8758                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8759                 spin_unlock(&fs_info->delalloc_root_lock);
8760         }
8761         mutex_unlock(&fs_info->delalloc_root_mutex);
8762         return ret;
8763 }
8764
8765 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8766                          const char *symname)
8767 {
8768         struct btrfs_trans_handle *trans;
8769         struct btrfs_root *root = BTRFS_I(dir)->root;
8770         struct btrfs_path *path;
8771         struct btrfs_key key;
8772         struct inode *inode = NULL;
8773         int err;
8774         int drop_inode = 0;
8775         u64 objectid;
8776         u64 index = 0;
8777         int name_len;
8778         int datasize;
8779         unsigned long ptr;
8780         struct btrfs_file_extent_item *ei;
8781         struct extent_buffer *leaf;
8782
8783         name_len = strlen(symname);
8784         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8785                 return -ENAMETOOLONG;
8786
8787         /*
8788          * 2 items for inode item and ref
8789          * 2 items for dir items
8790          * 1 item for xattr if selinux is on
8791          */
8792         trans = btrfs_start_transaction(root, 5);
8793         if (IS_ERR(trans))
8794                 return PTR_ERR(trans);
8795
8796         err = btrfs_find_free_ino(root, &objectid);
8797         if (err)
8798                 goto out_unlock;
8799
8800         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8801                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8802                                 S_IFLNK|S_IRWXUGO, &index);
8803         if (IS_ERR(inode)) {
8804                 err = PTR_ERR(inode);
8805                 goto out_unlock;
8806         }
8807
8808         /*
8809         * If the active LSM wants to access the inode during
8810         * d_instantiate it needs these. Smack checks to see
8811         * if the filesystem supports xattrs by looking at the
8812         * ops vector.
8813         */
8814         inode->i_fop = &btrfs_file_operations;
8815         inode->i_op = &btrfs_file_inode_operations;
8816         inode->i_mapping->a_ops = &btrfs_aops;
8817         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8818         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8819
8820         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8821         if (err)
8822                 goto out_unlock_inode;
8823
8824         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8825         if (err)
8826                 goto out_unlock_inode;
8827
8828         path = btrfs_alloc_path();
8829         if (!path) {
8830                 err = -ENOMEM;
8831                 goto out_unlock_inode;
8832         }
8833         key.objectid = btrfs_ino(inode);
8834         key.offset = 0;
8835         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8836         datasize = btrfs_file_extent_calc_inline_size(name_len);
8837         err = btrfs_insert_empty_item(trans, root, path, &key,
8838                                       datasize);
8839         if (err) {
8840                 btrfs_free_path(path);
8841                 goto out_unlock_inode;
8842         }
8843         leaf = path->nodes[0];
8844         ei = btrfs_item_ptr(leaf, path->slots[0],
8845                             struct btrfs_file_extent_item);
8846         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8847         btrfs_set_file_extent_type(leaf, ei,
8848                                    BTRFS_FILE_EXTENT_INLINE);
8849         btrfs_set_file_extent_encryption(leaf, ei, 0);
8850         btrfs_set_file_extent_compression(leaf, ei, 0);
8851         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8852         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8853
8854         ptr = btrfs_file_extent_inline_start(ei);
8855         write_extent_buffer(leaf, symname, ptr, name_len);
8856         btrfs_mark_buffer_dirty(leaf);
8857         btrfs_free_path(path);
8858
8859         inode->i_op = &btrfs_symlink_inode_operations;
8860         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8861         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8862         inode_set_bytes(inode, name_len);
8863         btrfs_i_size_write(inode, name_len);
8864         err = btrfs_update_inode(trans, root, inode);
8865         if (err) {
8866                 drop_inode = 1;
8867                 goto out_unlock_inode;
8868         }
8869
8870         unlock_new_inode(inode);
8871         d_instantiate(dentry, inode);
8872
8873 out_unlock:
8874         btrfs_end_transaction(trans, root);
8875         if (drop_inode) {
8876                 inode_dec_link_count(inode);
8877                 iput(inode);
8878         }
8879         btrfs_btree_balance_dirty(root);
8880         return err;
8881
8882 out_unlock_inode:
8883         drop_inode = 1;
8884         unlock_new_inode(inode);
8885         goto out_unlock;
8886 }
8887
8888 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8889                                        u64 start, u64 num_bytes, u64 min_size,
8890                                        loff_t actual_len, u64 *alloc_hint,
8891                                        struct btrfs_trans_handle *trans)
8892 {
8893         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8894         struct extent_map *em;
8895         struct btrfs_root *root = BTRFS_I(inode)->root;
8896         struct btrfs_key ins;
8897         u64 cur_offset = start;
8898         u64 i_size;
8899         u64 cur_bytes;
8900         int ret = 0;
8901         bool own_trans = true;
8902
8903         if (trans)
8904                 own_trans = false;
8905         while (num_bytes > 0) {
8906                 if (own_trans) {
8907                         trans = btrfs_start_transaction(root, 3);
8908                         if (IS_ERR(trans)) {
8909                                 ret = PTR_ERR(trans);
8910                                 break;
8911                         }
8912                 }
8913
8914                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8915                 cur_bytes = max(cur_bytes, min_size);
8916                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8917                                            *alloc_hint, &ins, 1, 0);
8918                 if (ret) {
8919                         if (own_trans)
8920                                 btrfs_end_transaction(trans, root);
8921                         break;
8922                 }
8923
8924                 ret = insert_reserved_file_extent(trans, inode,
8925                                                   cur_offset, ins.objectid,
8926                                                   ins.offset, ins.offset,
8927                                                   ins.offset, 0, 0, 0,
8928                                                   BTRFS_FILE_EXTENT_PREALLOC);
8929                 if (ret) {
8930                         btrfs_free_reserved_extent(root, ins.objectid,
8931                                                    ins.offset, 0);
8932                         btrfs_abort_transaction(trans, root, ret);
8933                         if (own_trans)
8934                                 btrfs_end_transaction(trans, root);
8935                         break;
8936                 }
8937                 btrfs_drop_extent_cache(inode, cur_offset,
8938                                         cur_offset + ins.offset -1, 0);
8939
8940                 em = alloc_extent_map();
8941                 if (!em) {
8942                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8943                                 &BTRFS_I(inode)->runtime_flags);
8944                         goto next;
8945                 }
8946
8947                 em->start = cur_offset;
8948                 em->orig_start = cur_offset;
8949                 em->len = ins.offset;
8950                 em->block_start = ins.objectid;
8951                 em->block_len = ins.offset;
8952                 em->orig_block_len = ins.offset;
8953                 em->ram_bytes = ins.offset;
8954                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8955                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8956                 em->generation = trans->transid;
8957
8958                 while (1) {
8959                         write_lock(&em_tree->lock);
8960                         ret = add_extent_mapping(em_tree, em, 1);
8961                         write_unlock(&em_tree->lock);
8962                         if (ret != -EEXIST)
8963                                 break;
8964                         btrfs_drop_extent_cache(inode, cur_offset,
8965                                                 cur_offset + ins.offset - 1,
8966                                                 0);
8967                 }
8968                 free_extent_map(em);
8969 next:
8970                 num_bytes -= ins.offset;
8971                 cur_offset += ins.offset;
8972                 *alloc_hint = ins.objectid + ins.offset;
8973
8974                 inode_inc_iversion(inode);
8975                 inode->i_ctime = CURRENT_TIME;
8976                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8977                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8978                     (actual_len > inode->i_size) &&
8979                     (cur_offset > inode->i_size)) {
8980                         if (cur_offset > actual_len)
8981                                 i_size = actual_len;
8982                         else
8983                                 i_size = cur_offset;
8984                         i_size_write(inode, i_size);
8985                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8986                 }
8987
8988                 ret = btrfs_update_inode(trans, root, inode);
8989
8990                 if (ret) {
8991                         btrfs_abort_transaction(trans, root, ret);
8992                         if (own_trans)
8993                                 btrfs_end_transaction(trans, root);
8994                         break;
8995                 }
8996
8997                 if (own_trans)
8998                         btrfs_end_transaction(trans, root);
8999         }
9000         return ret;
9001 }
9002
9003 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9004                               u64 start, u64 num_bytes, u64 min_size,
9005                               loff_t actual_len, u64 *alloc_hint)
9006 {
9007         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9008                                            min_size, actual_len, alloc_hint,
9009                                            NULL);
9010 }
9011
9012 int btrfs_prealloc_file_range_trans(struct inode *inode,
9013                                     struct btrfs_trans_handle *trans, int mode,
9014                                     u64 start, u64 num_bytes, u64 min_size,
9015                                     loff_t actual_len, u64 *alloc_hint)
9016 {
9017         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9018                                            min_size, actual_len, alloc_hint, trans);
9019 }
9020
9021 static int btrfs_set_page_dirty(struct page *page)
9022 {
9023         return __set_page_dirty_nobuffers(page);
9024 }
9025
9026 static int btrfs_permission(struct inode *inode, int mask)
9027 {
9028         struct btrfs_root *root = BTRFS_I(inode)->root;
9029         umode_t mode = inode->i_mode;
9030
9031         if (mask & MAY_WRITE &&
9032             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9033                 if (btrfs_root_readonly(root))
9034                         return -EROFS;
9035                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9036                         return -EACCES;
9037         }
9038         return generic_permission(inode, mask);
9039 }
9040
9041 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9042 {
9043         struct btrfs_trans_handle *trans;
9044         struct btrfs_root *root = BTRFS_I(dir)->root;
9045         struct inode *inode = NULL;
9046         u64 objectid;
9047         u64 index;
9048         int ret = 0;
9049
9050         /*
9051          * 5 units required for adding orphan entry
9052          */
9053         trans = btrfs_start_transaction(root, 5);
9054         if (IS_ERR(trans))
9055                 return PTR_ERR(trans);
9056
9057         ret = btrfs_find_free_ino(root, &objectid);
9058         if (ret)
9059                 goto out;
9060
9061         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9062                                 btrfs_ino(dir), objectid, mode, &index);
9063         if (IS_ERR(inode)) {
9064                 ret = PTR_ERR(inode);
9065                 inode = NULL;
9066                 goto out;
9067         }
9068
9069         inode->i_fop = &btrfs_file_operations;
9070         inode->i_op = &btrfs_file_inode_operations;
9071
9072         inode->i_mapping->a_ops = &btrfs_aops;
9073         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
9074         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9075
9076         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9077         if (ret)
9078                 goto out_inode;
9079
9080         ret = btrfs_update_inode(trans, root, inode);
9081         if (ret)
9082                 goto out_inode;
9083         ret = btrfs_orphan_add(trans, inode);
9084         if (ret)
9085                 goto out_inode;
9086
9087         /*
9088          * We set number of links to 0 in btrfs_new_inode(), and here we set
9089          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9090          * through:
9091          *
9092          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9093          */
9094         set_nlink(inode, 1);
9095         unlock_new_inode(inode);
9096         d_tmpfile(dentry, inode);
9097         mark_inode_dirty(inode);
9098
9099 out:
9100         btrfs_end_transaction(trans, root);
9101         if (ret)
9102                 iput(inode);
9103         btrfs_balance_delayed_items(root);
9104         btrfs_btree_balance_dirty(root);
9105         return ret;
9106
9107 out_inode:
9108         unlock_new_inode(inode);
9109         goto out;
9110
9111 }
9112
9113 static const struct inode_operations btrfs_dir_inode_operations = {
9114         .getattr        = btrfs_getattr,
9115         .lookup         = btrfs_lookup,
9116         .create         = btrfs_create,
9117         .unlink         = btrfs_unlink,
9118         .link           = btrfs_link,
9119         .mkdir          = btrfs_mkdir,
9120         .rmdir          = btrfs_rmdir,
9121         .rename2        = btrfs_rename2,
9122         .symlink        = btrfs_symlink,
9123         .setattr        = btrfs_setattr,
9124         .mknod          = btrfs_mknod,
9125         .setxattr       = btrfs_setxattr,
9126         .getxattr       = btrfs_getxattr,
9127         .listxattr      = btrfs_listxattr,
9128         .removexattr    = btrfs_removexattr,
9129         .permission     = btrfs_permission,
9130         .get_acl        = btrfs_get_acl,
9131         .set_acl        = btrfs_set_acl,
9132         .update_time    = btrfs_update_time,
9133         .tmpfile        = btrfs_tmpfile,
9134 };
9135 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9136         .lookup         = btrfs_lookup,
9137         .permission     = btrfs_permission,
9138         .get_acl        = btrfs_get_acl,
9139         .set_acl        = btrfs_set_acl,
9140         .update_time    = btrfs_update_time,
9141 };
9142
9143 static const struct file_operations btrfs_dir_file_operations = {
9144         .llseek         = generic_file_llseek,
9145         .read           = generic_read_dir,
9146         .iterate        = btrfs_real_readdir,
9147         .unlocked_ioctl = btrfs_ioctl,
9148 #ifdef CONFIG_COMPAT
9149         .compat_ioctl   = btrfs_ioctl,
9150 #endif
9151         .release        = btrfs_release_file,
9152         .fsync          = btrfs_sync_file,
9153 };
9154
9155 static struct extent_io_ops btrfs_extent_io_ops = {
9156         .fill_delalloc = run_delalloc_range,
9157         .submit_bio_hook = btrfs_submit_bio_hook,
9158         .merge_bio_hook = btrfs_merge_bio_hook,
9159         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9160         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9161         .writepage_start_hook = btrfs_writepage_start_hook,
9162         .set_bit_hook = btrfs_set_bit_hook,
9163         .clear_bit_hook = btrfs_clear_bit_hook,
9164         .merge_extent_hook = btrfs_merge_extent_hook,
9165         .split_extent_hook = btrfs_split_extent_hook,
9166 };
9167
9168 /*
9169  * btrfs doesn't support the bmap operation because swapfiles
9170  * use bmap to make a mapping of extents in the file.  They assume
9171  * these extents won't change over the life of the file and they
9172  * use the bmap result to do IO directly to the drive.
9173  *
9174  * the btrfs bmap call would return logical addresses that aren't
9175  * suitable for IO and they also will change frequently as COW
9176  * operations happen.  So, swapfile + btrfs == corruption.
9177  *
9178  * For now we're avoiding this by dropping bmap.
9179  */
9180 static const struct address_space_operations btrfs_aops = {
9181         .readpage       = btrfs_readpage,
9182         .writepage      = btrfs_writepage,
9183         .writepages     = btrfs_writepages,
9184         .readpages      = btrfs_readpages,
9185         .direct_IO      = btrfs_direct_IO,
9186         .invalidatepage = btrfs_invalidatepage,
9187         .releasepage    = btrfs_releasepage,
9188         .set_page_dirty = btrfs_set_page_dirty,
9189         .error_remove_page = generic_error_remove_page,
9190 };
9191
9192 static const struct address_space_operations btrfs_symlink_aops = {
9193         .readpage       = btrfs_readpage,
9194         .writepage      = btrfs_writepage,
9195         .invalidatepage = btrfs_invalidatepage,
9196         .releasepage    = btrfs_releasepage,
9197 };
9198
9199 static const struct inode_operations btrfs_file_inode_operations = {
9200         .getattr        = btrfs_getattr,
9201         .setattr        = btrfs_setattr,
9202         .setxattr       = btrfs_setxattr,
9203         .getxattr       = btrfs_getxattr,
9204         .listxattr      = btrfs_listxattr,
9205         .removexattr    = btrfs_removexattr,
9206         .permission     = btrfs_permission,
9207         .fiemap         = btrfs_fiemap,
9208         .get_acl        = btrfs_get_acl,
9209         .set_acl        = btrfs_set_acl,
9210         .update_time    = btrfs_update_time,
9211 };
9212 static const struct inode_operations btrfs_special_inode_operations = {
9213         .getattr        = btrfs_getattr,
9214         .setattr        = btrfs_setattr,
9215         .permission     = btrfs_permission,
9216         .setxattr       = btrfs_setxattr,
9217         .getxattr       = btrfs_getxattr,
9218         .listxattr      = btrfs_listxattr,
9219         .removexattr    = btrfs_removexattr,
9220         .get_acl        = btrfs_get_acl,
9221         .set_acl        = btrfs_set_acl,
9222         .update_time    = btrfs_update_time,
9223 };
9224 static const struct inode_operations btrfs_symlink_inode_operations = {
9225         .readlink       = generic_readlink,
9226         .follow_link    = page_follow_link_light,
9227         .put_link       = page_put_link,
9228         .getattr        = btrfs_getattr,
9229         .setattr        = btrfs_setattr,
9230         .permission     = btrfs_permission,
9231         .setxattr       = btrfs_setxattr,
9232         .getxattr       = btrfs_getxattr,
9233         .listxattr      = btrfs_listxattr,
9234         .removexattr    = btrfs_removexattr,
9235         .update_time    = btrfs_update_time,
9236 };
9237
9238 const struct dentry_operations btrfs_dentry_operations = {
9239         .d_delete       = btrfs_dentry_delete,
9240         .d_release      = btrfs_dentry_release,
9241 };