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/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/posix_acl_xattr.h>
  45 #include <linux/uio.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 #include "qgroup.h"
  63
  64 struct btrfs_iget_args {
  65         struct btrfs_key *location;
  66         struct btrfs_root *root;
  67 };
  68
  69 static const struct inode_operations btrfs_dir_inode_operations;
  70 static const struct inode_operations btrfs_symlink_inode_operations;
  71 static const struct inode_operations btrfs_dir_ro_inode_operations;
  72 static const struct inode_operations btrfs_special_inode_operations;
  73 static const struct inode_operations btrfs_file_inode_operations;
  74 static const struct address_space_operations btrfs_aops;
  75 static const struct address_space_operations btrfs_symlink_aops;
  76 static const struct file_operations btrfs_dir_file_operations;
  77 static struct extent_io_ops btrfs_extent_io_ops;
  78
  79 static struct kmem_cache *btrfs_inode_cachep;
  80 static struct kmem_cache *btrfs_delalloc_work_cachep;
  81 struct kmem_cache *btrfs_trans_handle_cachep;
  82 struct kmem_cache *btrfs_transaction_cachep;
  83 struct kmem_cache *btrfs_path_cachep;
  84 struct kmem_cache *btrfs_free_space_cachep;
  85
  86 #define S_SHIFT 12
  87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  88         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  89         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  90         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  91         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  92         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  93         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  94         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  95 };
  96
  97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  98 static int btrfs_truncate(struct inode *inode);
  99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 100 static noinline int cow_file_range(struct inode *inode,
 101                                    struct page *locked_page,
 102                                    u64 start, u64 end, int *page_started,
 103                                    unsigned long *nr_written, int unlock);
 104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 105                                            u64 len, u64 orig_start,
 106                                            u64 block_start, u64 block_len,
 107                                            u64 orig_block_len, u64 ram_bytes,
 108                                            int type);
 109
 110 static int btrfs_dirty_inode(struct inode *inode);
 111
 112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 113 void btrfs_test_inode_set_ops(struct inode *inode)
 114 {
 115         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 116 }
 117 #endif
 118
 119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 120                                      struct inode *inode,  struct inode *dir,
 121                                      const struct qstr *qstr)
 122 {
 123         int err;
 124
 125         err = btrfs_init_acl(trans, inode, dir);
 126         if (!err)
 127                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 128         return err;
 129 }
 130
 131 /*
 132  * this does all the hard work for inserting an inline extent into
 133  * the btree.  The caller should have done a btrfs_drop_extents so that
 134  * no overlapping inline items exist in the btree
 135  */
 136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 137                                 struct btrfs_path *path, int extent_inserted,
 138                                 struct btrfs_root *root, struct inode *inode,
 139                                 u64 start, size_t size, size_t compressed_size,
 140                                 int compress_type,
 141                                 struct page **compressed_pages)
 142 {
 143         struct extent_buffer *leaf;
 144         struct page *page = NULL;
 145         char *kaddr;
 146         unsigned long ptr;
 147         struct btrfs_file_extent_item *ei;
 148         int err = 0;
 149         int ret;
 150         size_t cur_size = size;
 151         unsigned long offset;
 152
 153         if (compressed_size && compressed_pages)
 154                 cur_size = compressed_size;
 155
 156         inode_add_bytes(inode, size);
 157
 158         if (!extent_inserted) {
 159                 struct btrfs_key key;
 160                 size_t datasize;
 161
 162                 key.objectid = btrfs_ino(inode);
 163                 key.offset = start;
 164                 key.type = BTRFS_EXTENT_DATA_KEY;
 165
 166                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 167                 path->leave_spinning = 1;
 168                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 169                                               datasize);
 170                 if (ret) {
 171                         err = ret;
 172                         goto fail;
 173                 }
 174         }
 175         leaf = path->nodes[0];
 176         ei = btrfs_item_ptr(leaf, path->slots[0],
 177                             struct btrfs_file_extent_item);
 178         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 179         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 180         btrfs_set_file_extent_encryption(leaf, ei, 0);
 181         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 182         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 183         ptr = btrfs_file_extent_inline_start(ei);
 184
 185         if (compress_type != BTRFS_COMPRESS_NONE) {
 186                 struct page *cpage;
 187                 int i = 0;
 188                 while (compressed_size > 0) {
 189                         cpage = compressed_pages[i];
 190                         cur_size = min_t(unsigned long, compressed_size,
 191                                        PAGE_CACHE_SIZE);
 192
 193                         kaddr = kmap_atomic(cpage);
 194                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 195                         kunmap_atomic(kaddr);
 196
 197                         i++;
 198                         ptr += cur_size;
 199                         compressed_size -= cur_size;
 200                 }
 201                 btrfs_set_file_extent_compression(leaf, ei,
 202                                                   compress_type);
 203         } else {
 204                 page = find_get_page(inode->i_mapping,
 205                                      start >> PAGE_CACHE_SHIFT);
 206                 btrfs_set_file_extent_compression(leaf, ei, 0);
 207                 kaddr = kmap_atomic(page);
 208                 offset = start & (PAGE_CACHE_SIZE - 1);
 209                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 210                 kunmap_atomic(kaddr);
 211                 page_cache_release(page);
 212         }
 213         btrfs_mark_buffer_dirty(leaf);
 214         btrfs_release_path(path);
 215
 216         /*
 217          * we're an inline extent, so nobody can
 218          * extend the file past i_size without locking
 219          * a page we already have locked.
 220          *
 221          * We must do any isize and inode updates
 222          * before we unlock the pages.  Otherwise we
 223          * could end up racing with unlink.
 224          */
 225         BTRFS_I(inode)->disk_i_size = inode->i_size;
 226         ret = btrfs_update_inode(trans, root, inode);
 227
 228         return ret;
 229 fail:
 230         return err;
 231 }
 232
 233
 234 /*
 235  * conditionally insert an inline extent into the file.  This
 236  * does the checks required to make sure the data is small enough
 237  * to fit as an inline extent.
 238  */
 239 static noinline int cow_file_range_inline(struct btrfs_root *root,
 240                                           struct inode *inode, u64 start,
 241                                           u64 end, size_t compressed_size,
 242                                           int compress_type,
 243                                           struct page **compressed_pages)
 244 {
 245         struct btrfs_trans_handle *trans;
 246         u64 isize = i_size_read(inode);
 247         u64 actual_end = min(end + 1, isize);
 248         u64 inline_len = actual_end - start;
 249         u64 aligned_end = ALIGN(end, root->sectorsize);
 250         u64 data_len = inline_len;
 251         int ret;
 252         struct btrfs_path *path;
 253         int extent_inserted = 0;
 254         u32 extent_item_size;
 255
 256         if (compressed_size)
 257                 data_len = compressed_size;
 258
 259         if (start > 0 ||
 260             actual_end > PAGE_CACHE_SIZE ||
 261             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 262             (!compressed_size &&
 263             (actual_end & (root->sectorsize - 1)) == 0) ||
 264             end + 1 < isize ||
 265             data_len > root->fs_info->max_inline) {
 266                 return 1;
 267         }
 268
 269         path = btrfs_alloc_path();
 270         if (!path)
 271                 return -ENOMEM;
 272
 273         trans = btrfs_join_transaction(root);
 274         if (IS_ERR(trans)) {
 275                 btrfs_free_path(path);
 276                 return PTR_ERR(trans);
 277         }
 278         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 279
 280         if (compressed_size && compressed_pages)
 281                 extent_item_size = btrfs_file_extent_calc_inline_size(
 282                    compressed_size);
 283         else
 284                 extent_item_size = btrfs_file_extent_calc_inline_size(
 285                     inline_len);
 286
 287         ret = __btrfs_drop_extents(trans, root, inode, path,
 288                                    start, aligned_end, NULL,
 289                                    1, 1, extent_item_size, &extent_inserted);
 290         if (ret) {
 291                 btrfs_abort_transaction(trans, root, ret);
 292                 goto out;
 293         }
 294
 295         if (isize > actual_end)
 296                 inline_len = min_t(u64, isize, actual_end);
 297         ret = insert_inline_extent(trans, path, extent_inserted,
 298                                    root, inode, start,
 299                                    inline_len, compressed_size,
 300                                    compress_type, compressed_pages);
 301         if (ret && ret != -ENOSPC) {
 302                 btrfs_abort_transaction(trans, root, ret);
 303                 goto out;
 304         } else if (ret == -ENOSPC) {
 305                 ret = 1;
 306                 goto out;
 307         }
 308
 309         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 310         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 311         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 312 out:
 313         btrfs_free_path(path);
 314         btrfs_end_transaction(trans, root);
 315         return ret;
 316 }
 317
 318 struct async_extent {
 319         u64 start;
 320         u64 ram_size;
 321         u64 compressed_size;
 322         struct page **pages;
 323         unsigned long nr_pages;
 324         int compress_type;
 325         struct list_head list;
 326 };
 327
 328 struct async_cow {
 329         struct inode *inode;
 330         struct btrfs_root *root;
 331         struct page *locked_page;
 332         u64 start;
 333         u64 end;
 334         struct list_head extents;
 335         struct btrfs_work work;
 336 };
 337
 338 static noinline int add_async_extent(struct async_cow *cow,
 339                                      u64 start, u64 ram_size,
 340                                      u64 compressed_size,
 341                                      struct page **pages,
 342                                      unsigned long nr_pages,
 343                                      int compress_type)
 344 {
 345         struct async_extent *async_extent;
 346
 347         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 348         BUG_ON(!async_extent); /* -ENOMEM */
 349         async_extent->start = start;
 350         async_extent->ram_size = ram_size;
 351         async_extent->compressed_size = compressed_size;
 352         async_extent->pages = pages;
 353         async_extent->nr_pages = nr_pages;
 354         async_extent->compress_type = compress_type;
 355         list_add_tail(&async_extent->list, &cow->extents);
 356         return 0;
 357 }
 358
 359 static inline int inode_need_compress(struct inode *inode)
 360 {
 361         struct btrfs_root *root = BTRFS_I(inode)->root;
 362
 363         /* force compress */
 364         if (btrfs_test_opt(root, FORCE_COMPRESS))
 365                 return 1;
 366         /* bad compression ratios */
 367         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 368                 return 0;
 369         if (btrfs_test_opt(root, COMPRESS) ||
 370             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 371             BTRFS_I(inode)->force_compress)
 372                 return 1;
 373         return 0;
 374 }
 375
 376 /*
 377  * we create compressed extents in two phases.  The first
 378  * phase compresses a range of pages that have already been
 379  * locked (both pages and state bits are locked).
 380  *
 381  * This is done inside an ordered work queue, and the compression
 382  * is spread across many cpus.  The actual IO submission is step
 383  * two, and the ordered work queue takes care of making sure that
 384  * happens in the same order things were put onto the queue by
 385  * writepages and friends.
 386  *
 387  * If this code finds it can't get good compression, it puts an
 388  * entry onto the work queue to write the uncompressed bytes.  This
 389  * makes sure that both compressed inodes and uncompressed inodes
 390  * are written in the same order that the flusher thread sent them
 391  * down.
 392  */
 393 static noinline void compress_file_range(struct inode *inode,
 394                                         struct page *locked_page,
 395                                         u64 start, u64 end,
 396                                         struct async_cow *async_cow,
 397                                         int *num_added)
 398 {
 399         struct btrfs_root *root = BTRFS_I(inode)->root;
 400         u64 num_bytes;
 401         u64 blocksize = root->sectorsize;
 402         u64 actual_end;
 403         u64 isize = i_size_read(inode);
 404         int ret = 0;
 405         struct page **pages = NULL;
 406         unsigned long nr_pages;
 407         unsigned long nr_pages_ret = 0;
 408         unsigned long total_compressed = 0;
 409         unsigned long total_in = 0;
 410         unsigned long max_compressed = 128 * 1024;
 411         unsigned long max_uncompressed = 128 * 1024;
 412         int i;
 413         int will_compress;
 414         int compress_type = root->fs_info->compress_type;
 415         int redirty = 0;
 416
 417         /* if this is a small write inside eof, kick off a defrag */
 418         if ((end - start + 1) < 16 * 1024 &&
 419             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 420                 btrfs_add_inode_defrag(NULL, inode);
 421
 422         actual_end = min_t(u64, isize, end + 1);
 423 again:
 424         will_compress = 0;
 425         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 426         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 427
 428         /*
 429          * we don't want to send crud past the end of i_size through
 430          * compression, that's just a waste of CPU time.  So, if the
 431          * end of the file is before the start of our current
 432          * requested range of bytes, we bail out to the uncompressed
 433          * cleanup code that can deal with all of this.
 434          *
 435          * It isn't really the fastest way to fix things, but this is a
 436          * very uncommon corner.
 437          */
 438         if (actual_end <= start)
 439                 goto cleanup_and_bail_uncompressed;
 440
 441         total_compressed = actual_end - start;
 442
 443         /*
 444          * skip compression for a small file range(<=blocksize) that
 445          * isn't an inline extent, since it dosen't save disk space at all.
 446          */
 447         if (total_compressed <= blocksize &&
 448            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 449                 goto cleanup_and_bail_uncompressed;
 450
 451         /* we want to make sure that amount of ram required to uncompress
 452          * an extent is reasonable, so we limit the total size in ram
 453          * of a compressed extent to 128k.  This is a crucial number
 454          * because it also controls how easily we can spread reads across
 455          * cpus for decompression.
 456          *
 457          * We also want to make sure the amount of IO required to do
 458          * a random read is reasonably small, so we limit the size of
 459          * a compressed extent to 128k.
 460          */
 461         total_compressed = min(total_compressed, max_uncompressed);
 462         num_bytes = ALIGN(end - start + 1, blocksize);
 463         num_bytes = max(blocksize,  num_bytes);
 464         total_in = 0;
 465         ret = 0;
 466
 467         /*
 468          * we do compression for mount -o compress and when the
 469          * inode has not been flagged as nocompress.  This flag can
 470          * change at any time if we discover bad compression ratios.
 471          */
 472         if (inode_need_compress(inode)) {
 473                 WARN_ON(pages);
 474                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 475                 if (!pages) {
 476                         /* just bail out to the uncompressed code */
 477                         goto cont;
 478                 }
 479
 480                 if (BTRFS_I(inode)->force_compress)
 481                         compress_type = BTRFS_I(inode)->force_compress;
 482
 483                 /*
 484                  * we need to call clear_page_dirty_for_io on each
 485                  * page in the range.  Otherwise applications with the file
 486                  * mmap'd can wander in and change the page contents while
 487                  * we are compressing them.
 488                  *
 489                  * If the compression fails for any reason, we set the pages
 490                  * dirty again later on.
 491                  */
 492                 extent_range_clear_dirty_for_io(inode, start, end);
 493                 redirty = 1;
 494                 ret = btrfs_compress_pages(compress_type,
 495                                            inode->i_mapping, start,
 496                                            total_compressed, pages,
 497                                            nr_pages, &nr_pages_ret,
 498                                            &total_in,
 499                                            &total_compressed,
 500                                            max_compressed);
 501
 502                 if (!ret) {
 503                         unsigned long offset = total_compressed &
 504                                 (PAGE_CACHE_SIZE - 1);
 505                         struct page *page = pages[nr_pages_ret - 1];
 506                         char *kaddr;
 507
 508                         /* zero the tail end of the last page, we might be
 509                          * sending it down to disk
 510                          */
 511                         if (offset) {
 512                                 kaddr = kmap_atomic(page);
 513                                 memset(kaddr + offset, 0,
 514                                        PAGE_CACHE_SIZE - offset);
 515                                 kunmap_atomic(kaddr);
 516                         }
 517                         will_compress = 1;
 518                 }
 519         }
 520 cont:
 521         if (start == 0) {
 522                 /* lets try to make an inline extent */
 523                 if (ret || total_in < (actual_end - start)) {
 524                         /* we didn't compress the entire range, try
 525                          * to make an uncompressed inline extent.
 526                          */
 527                         ret = cow_file_range_inline(root, inode, start, end,
 528                                                     0, 0, NULL);
 529                 } else {
 530                         /* try making a compressed inline extent */
 531                         ret = cow_file_range_inline(root, inode, start, end,
 532                                                     total_compressed,
 533                                                     compress_type, pages);
 534                 }
 535                 if (ret <= 0) {
 536                         unsigned long clear_flags = EXTENT_DELALLOC |
 537                                 EXTENT_DEFRAG;
 538                         unsigned long page_error_op;
 539
 540                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 541                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 542
 543                         /*
 544                          * inline extent creation worked or returned error,
 545                          * we don't need to create any more async work items.
 546                          * Unlock and free up our temp pages.
 547                          */
 548                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 549                                                      clear_flags, PAGE_UNLOCK |
 550                                                      PAGE_CLEAR_DIRTY |
 551                                                      PAGE_SET_WRITEBACK |
 552                                                      page_error_op |
 553                                                      PAGE_END_WRITEBACK);
 554                         goto free_pages_out;
 555                 }
 556         }
 557
 558         if (will_compress) {
 559                 /*
 560                  * we aren't doing an inline extent round the compressed size
 561                  * up to a block size boundary so the allocator does sane
 562                  * things
 563                  */
 564                 total_compressed = ALIGN(total_compressed, blocksize);
 565
 566                 /*
 567                  * one last check to make sure the compression is really a
 568                  * win, compare the page count read with the blocks on disk
 569                  */
 570                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 571                 if (total_compressed >= total_in) {
 572                         will_compress = 0;
 573                 } else {
 574                         num_bytes = total_in;
 575                 }
 576         }
 577         if (!will_compress && pages) {
 578                 /*
 579                  * the compression code ran but failed to make things smaller,
 580                  * free any pages it allocated and our page pointer array
 581                  */
 582                 for (i = 0; i < nr_pages_ret; i++) {
 583                         WARN_ON(pages[i]->mapping);
 584                         page_cache_release(pages[i]);
 585                 }
 586                 kfree(pages);
 587                 pages = NULL;
 588                 total_compressed = 0;
 589                 nr_pages_ret = 0;
 590
 591                 /* flag the file so we don't compress in the future */
 592                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 593                     !(BTRFS_I(inode)->force_compress)) {
 594                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 595                 }
 596         }
 597         if (will_compress) {
 598                 *num_added += 1;
 599
 600                 /* the async work queues will take care of doing actual
 601                  * allocation on disk for these compressed pages,
 602                  * and will submit them to the elevator.
 603                  */
 604                 add_async_extent(async_cow, start, num_bytes,
 605                                  total_compressed, pages, nr_pages_ret,
 606                                  compress_type);
 607
 608                 if (start + num_bytes < end) {
 609                         start += num_bytes;
 610                         pages = NULL;
 611                         cond_resched();
 612                         goto again;
 613                 }
 614         } else {
 615 cleanup_and_bail_uncompressed:
 616                 /*
 617                  * No compression, but we still need to write the pages in
 618                  * the file we've been given so far.  redirty the locked
 619                  * page if it corresponds to our extent and set things up
 620                  * for the async work queue to run cow_file_range to do
 621                  * the normal delalloc dance
 622                  */
 623                 if (page_offset(locked_page) >= start &&
 624                     page_offset(locked_page) <= end) {
 625                         __set_page_dirty_nobuffers(locked_page);
 626                         /* unlocked later on in the async handlers */
 627                 }
 628                 if (redirty)
 629                         extent_range_redirty_for_io(inode, start, end);
 630                 add_async_extent(async_cow, start, end - start + 1,
 631                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 632                 *num_added += 1;
 633         }
 634
 635         return;
 636
 637 free_pages_out:
 638         for (i = 0; i < nr_pages_ret; i++) {
 639                 WARN_ON(pages[i]->mapping);
 640                 page_cache_release(pages[i]);
 641         }
 642         kfree(pages);
 643 }
 644
 645 static void free_async_extent_pages(struct async_extent *async_extent)
 646 {
 647         int i;
 648
 649         if (!async_extent->pages)
 650                 return;
 651
 652         for (i = 0; i < async_extent->nr_pages; i++) {
 653                 WARN_ON(async_extent->pages[i]->mapping);
 654                 page_cache_release(async_extent->pages[i]);
 655         }
 656         kfree(async_extent->pages);
 657         async_extent->nr_pages = 0;
 658         async_extent->pages = NULL;
 659 }
 660
 661 /*
 662  * phase two of compressed writeback.  This is the ordered portion
 663  * of the code, which only gets called in the order the work was
 664  * queued.  We walk all the async extents created by compress_file_range
 665  * and send them down to the disk.
 666  */
 667 static noinline void submit_compressed_extents(struct inode *inode,
 668                                               struct async_cow *async_cow)
 669 {
 670         struct async_extent *async_extent;
 671         u64 alloc_hint = 0;
 672         struct btrfs_key ins;
 673         struct extent_map *em;
 674         struct btrfs_root *root = BTRFS_I(inode)->root;
 675         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 676         struct extent_io_tree *io_tree;
 677         int ret = 0;
 678
 679 again:
 680         while (!list_empty(&async_cow->extents)) {
 681                 async_extent = list_entry(async_cow->extents.next,
 682                                           struct async_extent, list);
 683                 list_del(&async_extent->list);
 684
 685                 io_tree = &BTRFS_I(inode)->io_tree;
 686
 687 retry:
 688                 /* did the compression code fall back to uncompressed IO? */
 689                 if (!async_extent->pages) {
 690                         int page_started = 0;
 691                         unsigned long nr_written = 0;
 692
 693                         lock_extent(io_tree, async_extent->start,
 694                                          async_extent->start +
 695                                          async_extent->ram_size - 1);
 696
 697                         /* allocate blocks */
 698                         ret = cow_file_range(inode, async_cow->locked_page,
 699                                              async_extent->start,
 700                                              async_extent->start +
 701                                              async_extent->ram_size - 1,
 702                                              &page_started, &nr_written, 0);
 703
 704                         /* JDM XXX */
 705
 706                         /*
 707                          * if page_started, cow_file_range inserted an
 708                          * inline extent and took care of all the unlocking
 709                          * and IO for us.  Otherwise, we need to submit
 710                          * all those pages down to the drive.
 711                          */
 712                         if (!page_started && !ret)
 713                                 extent_write_locked_range(io_tree,
 714                                                   inode, async_extent->start,
 715                                                   async_extent->start +
 716                                                   async_extent->ram_size - 1,
 717                                                   btrfs_get_extent,
 718                                                   WB_SYNC_ALL);
 719                         else if (ret)
 720                                 unlock_page(async_cow->locked_page);
 721                         kfree(async_extent);
 722                         cond_resched();
 723                         continue;
 724                 }
 725
 726                 lock_extent(io_tree, async_extent->start,
 727                             async_extent->start + async_extent->ram_size - 1);
 728
 729                 ret = btrfs_reserve_extent(root,
 730                                            async_extent->compressed_size,
 731                                            async_extent->compressed_size,
 732                                            0, alloc_hint, &ins, 1, 1);
 733                 if (ret) {
 734                         free_async_extent_pages(async_extent);
 735
 736                         if (ret == -ENOSPC) {
 737                                 unlock_extent(io_tree, async_extent->start,
 738                                               async_extent->start +
 739                                               async_extent->ram_size - 1);
 740
 741                                 /*
 742                                  * we need to redirty the pages if we decide to
 743                                  * fallback to uncompressed IO, otherwise we
 744                                  * will not submit these pages down to lower
 745                                  * layers.
 746                                  */
 747                                 extent_range_redirty_for_io(inode,
 748                                                 async_extent->start,
 749                                                 async_extent->start +
 750                                                 async_extent->ram_size - 1);
 751
 752                                 goto retry;
 753                         }
 754                         goto out_free;
 755                 }
 756                 /*
 757                  * here we're doing allocation and writeback of the
 758                  * compressed pages
 759                  */
 760                 btrfs_drop_extent_cache(inode, async_extent->start,
 761                                         async_extent->start +
 762                                         async_extent->ram_size - 1, 0);
 763
 764                 em = alloc_extent_map();
 765                 if (!em) {
 766                         ret = -ENOMEM;
 767                         goto out_free_reserve;
 768                 }
 769                 em->start = async_extent->start;
 770                 em->len = async_extent->ram_size;
 771                 em->orig_start = em->start;
 772                 em->mod_start = em->start;
 773                 em->mod_len = em->len;
 774
 775                 em->block_start = ins.objectid;
 776                 em->block_len = ins.offset;
 777                 em->orig_block_len = ins.offset;
 778                 em->ram_bytes = async_extent->ram_size;
 779                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 780                 em->compress_type = async_extent->compress_type;
 781                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 782                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 783                 em->generation = -1;
 784
 785                 while (1) {
 786                         write_lock(&em_tree->lock);
 787                         ret = add_extent_mapping(em_tree, em, 1);
 788                         write_unlock(&em_tree->lock);
 789                         if (ret != -EEXIST) {
 790                                 free_extent_map(em);
 791                                 break;
 792                         }
 793                         btrfs_drop_extent_cache(inode, async_extent->start,
 794                                                 async_extent->start +
 795                                                 async_extent->ram_size - 1, 0);
 796                 }
 797
 798                 if (ret)
 799                         goto out_free_reserve;
 800
 801                 ret = btrfs_add_ordered_extent_compress(inode,
 802                                                 async_extent->start,
 803                                                 ins.objectid,
 804                                                 async_extent->ram_size,
 805                                                 ins.offset,
 806                                                 BTRFS_ORDERED_COMPRESSED,
 807                                                 async_extent->compress_type);
 808                 if (ret) {
 809                         btrfs_drop_extent_cache(inode, async_extent->start,
 810                                                 async_extent->start +
 811                                                 async_extent->ram_size - 1, 0);
 812                         goto out_free_reserve;
 813                 }
 814
 815                 /*
 816                  * clear dirty, set writeback and unlock the pages.
 817                  */
 818                 extent_clear_unlock_delalloc(inode, async_extent->start,
 819                                 async_extent->start +
 820                                 async_extent->ram_size - 1,
 821                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 822                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 823                                 PAGE_SET_WRITEBACK);
 824                 ret = btrfs_submit_compressed_write(inode,
 825                                     async_extent->start,
 826                                     async_extent->ram_size,
 827                                     ins.objectid,
 828                                     ins.offset, async_extent->pages,
 829                                     async_extent->nr_pages);
 830                 if (ret) {
 831                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 832                         struct page *p = async_extent->pages[0];
 833                         const u64 start = async_extent->start;
 834                         const u64 end = start + async_extent->ram_size - 1;
 835
 836                         p->mapping = inode->i_mapping;
 837                         tree->ops->writepage_end_io_hook(p, start, end,
 838                                                          NULL, 0);
 839                         p->mapping = NULL;
 840                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
 841                                                      PAGE_END_WRITEBACK |
 842                                                      PAGE_SET_ERROR);
 843                         free_async_extent_pages(async_extent);
 844                 }
 845                 alloc_hint = ins.objectid + ins.offset;
 846                 kfree(async_extent);
 847                 cond_resched();
 848         }
 849         return;
 850 out_free_reserve:
 851         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 852 out_free:
 853         extent_clear_unlock_delalloc(inode, async_extent->start,
 854                                      async_extent->start +
 855                                      async_extent->ram_size - 1,
 856                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 857                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 858                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 859                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 860                                      PAGE_SET_ERROR);
 861         free_async_extent_pages(async_extent);
 862         kfree(async_extent);
 863         goto again;
 864 }
 865
 866 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 867                                       u64 num_bytes)
 868 {
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         struct extent_map *em;
 871         u64 alloc_hint = 0;
 872
 873         read_lock(&em_tree->lock);
 874         em = search_extent_mapping(em_tree, start, num_bytes);
 875         if (em) {
 876                 /*
 877                  * if block start isn't an actual block number then find the
 878                  * first block in this inode and use that as a hint.  If that
 879                  * block is also bogus then just don't worry about it.
 880                  */
 881                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 882                         free_extent_map(em);
 883                         em = search_extent_mapping(em_tree, 0, 0);
 884                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 885                                 alloc_hint = em->block_start;
 886                         if (em)
 887                                 free_extent_map(em);
 888                 } else {
 889                         alloc_hint = em->block_start;
 890                         free_extent_map(em);
 891                 }
 892         }
 893         read_unlock(&em_tree->lock);
 894
 895         return alloc_hint;
 896 }
 897
 898 /*
 899  * when extent_io.c finds a delayed allocation range in the file,
 900  * the call backs end up in this code.  The basic idea is to
 901  * allocate extents on disk for the range, and create ordered data structs
 902  * in ram to track those extents.
 903  *
 904  * locked_page is the page that writepage had locked already.  We use
 905  * it to make sure we don't do extra locks or unlocks.
 906  *
 907  * *page_started is set to one if we unlock locked_page and do everything
 908  * required to start IO on it.  It may be clean and already done with
 909  * IO when we return.
 910  */
 911 static noinline int cow_file_range(struct inode *inode,
 912                                    struct page *locked_page,
 913                                    u64 start, u64 end, int *page_started,
 914                                    unsigned long *nr_written,
 915                                    int unlock)
 916 {
 917         struct btrfs_root *root = BTRFS_I(inode)->root;
 918         u64 alloc_hint = 0;
 919         u64 num_bytes;
 920         unsigned long ram_size;
 921         u64 disk_num_bytes;
 922         u64 cur_alloc_size;
 923         u64 blocksize = root->sectorsize;
 924         struct btrfs_key ins;
 925         struct extent_map *em;
 926         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 927         int ret = 0;
 928
 929         if (btrfs_is_free_space_inode(inode)) {
 930                 WARN_ON_ONCE(1);
 931                 ret = -EINVAL;
 932                 goto out_unlock;
 933         }
 934
 935         num_bytes = ALIGN(end - start + 1, blocksize);
 936         num_bytes = max(blocksize,  num_bytes);
 937         disk_num_bytes = num_bytes;
 938
 939         /* if this is a small write inside eof, kick off defrag */
 940         if (num_bytes < 64 * 1024 &&
 941             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 942                 btrfs_add_inode_defrag(NULL, inode);
 943
 944         if (start == 0) {
 945                 /* lets try to make an inline extent */
 946                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 947                                             NULL);
 948                 if (ret == 0) {
 949                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 950                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 951                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 952                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 953                                      PAGE_END_WRITEBACK);
 954
 955                         *nr_written = *nr_written +
 956                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 957                         *page_started = 1;
 958                         goto out;
 959                 } else if (ret < 0) {
 960                         goto out_unlock;
 961                 }
 962         }
 963
 964         BUG_ON(disk_num_bytes >
 965                btrfs_super_total_bytes(root->fs_info->super_copy));
 966
 967         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 968         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 969
 970         while (disk_num_bytes > 0) {
 971                 unsigned long op;
 972
 973                 cur_alloc_size = disk_num_bytes;
 974                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 975                                            root->sectorsize, 0, alloc_hint,
 976                                            &ins, 1, 1);
 977                 if (ret < 0)
 978                         goto out_unlock;
 979
 980                 em = alloc_extent_map();
 981                 if (!em) {
 982                         ret = -ENOMEM;
 983                         goto out_reserve;
 984                 }
 985                 em->start = start;
 986                 em->orig_start = em->start;
 987                 ram_size = ins.offset;
 988                 em->len = ins.offset;
 989                 em->mod_start = em->start;
 990                 em->mod_len = em->len;
 991
 992                 em->block_start = ins.objectid;
 993                 em->block_len = ins.offset;
 994                 em->orig_block_len = ins.offset;
 995                 em->ram_bytes = ram_size;
 996                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 997                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 998                 em->generation = -1;
 999
1000                 while (1) {
1001                         write_lock(&em_tree->lock);
1002                         ret = add_extent_mapping(em_tree, em, 1);
1003                         write_unlock(&em_tree->lock);
1004                         if (ret != -EEXIST) {
1005                                 free_extent_map(em);
1006                                 break;
1007                         }
1008                         btrfs_drop_extent_cache(inode, start,
1009                                                 start + ram_size - 1, 0);
1010                 }
1011                 if (ret)
1012                         goto out_reserve;
1013
1014                 cur_alloc_size = ins.offset;
1015                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1016                                                ram_size, cur_alloc_size, 0);
1017                 if (ret)
1018                         goto out_drop_extent_cache;
1019
1020                 if (root->root_key.objectid ==
1021                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1022                         ret = btrfs_reloc_clone_csums(inode, start,
1023                                                       cur_alloc_size);
1024                         if (ret)
1025                                 goto out_drop_extent_cache;
1026                 }
1027
1028                 if (disk_num_bytes < cur_alloc_size)
1029                         break;
1030
1031                 /* we're not doing compressed IO, don't unlock the first
1032                  * page (which the caller expects to stay locked), don't
1033                  * clear any dirty bits and don't set any writeback bits
1034                  *
1035                  * Do set the Private2 bit so we know this page was properly
1036                  * setup for writepage
1037                  */
1038                 op = unlock ? PAGE_UNLOCK : 0;
1039                 op |= PAGE_SET_PRIVATE2;
1040
1041                 extent_clear_unlock_delalloc(inode, start,
1042                                              start + ram_size - 1, locked_page,
1043                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1044                                              op);
1045                 disk_num_bytes -= cur_alloc_size;
1046                 num_bytes -= cur_alloc_size;
1047                 alloc_hint = ins.objectid + ins.offset;
1048                 start += cur_alloc_size;
1049         }
1050 out:
1051         return ret;
1052
1053 out_drop_extent_cache:
1054         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1055 out_reserve:
1056         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1057 out_unlock:
1058         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1059                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1060                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1061                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1062                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1063         goto out;
1064 }
1065
1066 /*
1067  * work queue call back to started compression on a file and pages
1068  */
1069 static noinline void async_cow_start(struct btrfs_work *work)
1070 {
1071         struct async_cow *async_cow;
1072         int num_added = 0;
1073         async_cow = container_of(work, struct async_cow, work);
1074
1075         compress_file_range(async_cow->inode, async_cow->locked_page,
1076                             async_cow->start, async_cow->end, async_cow,
1077                             &num_added);
1078         if (num_added == 0) {
1079                 btrfs_add_delayed_iput(async_cow->inode);
1080                 async_cow->inode = NULL;
1081         }
1082 }
1083
1084 /*
1085  * work queue call back to submit previously compressed pages
1086  */
1087 static noinline void async_cow_submit(struct btrfs_work *work)
1088 {
1089         struct async_cow *async_cow;
1090         struct btrfs_root *root;
1091         unsigned long nr_pages;
1092
1093         async_cow = container_of(work, struct async_cow, work);
1094
1095         root = async_cow->root;
1096         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1097                 PAGE_CACHE_SHIFT;
1098
1099         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1100             5 * 1024 * 1024 &&
1101             waitqueue_active(&root->fs_info->async_submit_wait))
1102                 wake_up(&root->fs_info->async_submit_wait);
1103
1104         if (async_cow->inode)
1105                 submit_compressed_extents(async_cow->inode, async_cow);
1106 }
1107
1108 static noinline void async_cow_free(struct btrfs_work *work)
1109 {
1110         struct async_cow *async_cow;
1111         async_cow = container_of(work, struct async_cow, work);
1112         if (async_cow->inode)
1113                 btrfs_add_delayed_iput(async_cow->inode);
1114         kfree(async_cow);
1115 }
1116
1117 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1118                                 u64 start, u64 end, int *page_started,
1119                                 unsigned long *nr_written)
1120 {
1121         struct async_cow *async_cow;
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         unsigned long nr_pages;
1124         u64 cur_end;
1125         int limit = 10 * 1024 * 1024;
1126
1127         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1128                          1, 0, NULL, GFP_NOFS);
1129         while (start < end) {
1130                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1131                 BUG_ON(!async_cow); /* -ENOMEM */
1132                 async_cow->inode = igrab(inode);
1133                 async_cow->root = root;
1134                 async_cow->locked_page = locked_page;
1135                 async_cow->start = start;
1136
1137                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1138                     !btrfs_test_opt(root, FORCE_COMPRESS))
1139                         cur_end = end;
1140                 else
1141                         cur_end = min(end, start + 512 * 1024 - 1);
1142
1143                 async_cow->end = cur_end;
1144                 INIT_LIST_HEAD(&async_cow->extents);
1145
1146                 btrfs_init_work(&async_cow->work,
1147                                 btrfs_delalloc_helper,
1148                                 async_cow_start, async_cow_submit,
1149                                 async_cow_free);
1150
1151                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1152                         PAGE_CACHE_SHIFT;
1153                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1154
1155                 btrfs_queue_work(root->fs_info->delalloc_workers,
1156                                  &async_cow->work);
1157
1158                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1161                             limit));
1162                 }
1163
1164                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1165                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1166                         wait_event(root->fs_info->async_submit_wait,
1167                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1168                            0));
1169                 }
1170
1171                 *nr_written += nr_pages;
1172                 start = cur_end + 1;
1173         }
1174         *page_started = 1;
1175         return 0;
1176 }
1177
1178 static noinline int csum_exist_in_range(struct btrfs_root *root,
1179                                         u64 bytenr, u64 num_bytes)
1180 {
1181         int ret;
1182         struct btrfs_ordered_sum *sums;
1183         LIST_HEAD(list);
1184
1185         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1186                                        bytenr + num_bytes - 1, &list, 0);
1187         if (ret == 0 && list_empty(&list))
1188                 return 0;
1189
1190         while (!list_empty(&list)) {
1191                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1192                 list_del(&sums->list);
1193                 kfree(sums);
1194         }
1195         return 1;
1196 }
1197
1198 /*
1199  * when nowcow writeback call back.  This checks for snapshots or COW copies
1200  * of the extents that exist in the file, and COWs the file as required.
1201  *
1202  * If no cow copies or snapshots exist, we write directly to the existing
1203  * blocks on disk
1204  */
1205 static noinline int run_delalloc_nocow(struct inode *inode,
1206                                        struct page *locked_page,
1207                               u64 start, u64 end, int *page_started, int force,
1208                               unsigned long *nr_written)
1209 {
1210         struct btrfs_root *root = BTRFS_I(inode)->root;
1211         struct btrfs_trans_handle *trans;
1212         struct extent_buffer *leaf;
1213         struct btrfs_path *path;
1214         struct btrfs_file_extent_item *fi;
1215         struct btrfs_key found_key;
1216         u64 cow_start;
1217         u64 cur_offset;
1218         u64 extent_end;
1219         u64 extent_offset;
1220         u64 disk_bytenr;
1221         u64 num_bytes;
1222         u64 disk_num_bytes;
1223         u64 ram_bytes;
1224         int extent_type;
1225         int ret, err;
1226         int type;
1227         int nocow;
1228         int check_prev = 1;
1229         bool nolock;
1230         u64 ino = btrfs_ino(inode);
1231
1232         path = btrfs_alloc_path();
1233         if (!path) {
1234                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1235                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1236                                              EXTENT_DO_ACCOUNTING |
1237                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1238                                              PAGE_CLEAR_DIRTY |
1239                                              PAGE_SET_WRITEBACK |
1240                                              PAGE_END_WRITEBACK);
1241                 return -ENOMEM;
1242         }
1243
1244         nolock = btrfs_is_free_space_inode(inode);
1245
1246         if (nolock)
1247                 trans = btrfs_join_transaction_nolock(root);
1248         else
1249                 trans = btrfs_join_transaction(root);
1250
1251         if (IS_ERR(trans)) {
1252                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1253                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1254                                              EXTENT_DO_ACCOUNTING |
1255                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1256                                              PAGE_CLEAR_DIRTY |
1257                                              PAGE_SET_WRITEBACK |
1258                                              PAGE_END_WRITEBACK);
1259                 btrfs_free_path(path);
1260                 return PTR_ERR(trans);
1261         }
1262
1263         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1264
1265         cow_start = (u64)-1;
1266         cur_offset = start;
1267         while (1) {
1268                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1269                                                cur_offset, 0);
1270                 if (ret < 0)
1271                         goto error;
1272                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1273                         leaf = path->nodes[0];
1274                         btrfs_item_key_to_cpu(leaf, &found_key,
1275                                               path->slots[0] - 1);
1276                         if (found_key.objectid == ino &&
1277                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1278                                 path->slots[0]--;
1279                 }
1280                 check_prev = 0;
1281 next_slot:
1282                 leaf = path->nodes[0];
1283                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1284                         ret = btrfs_next_leaf(root, path);
1285                         if (ret < 0)
1286                                 goto error;
1287                         if (ret > 0)
1288                                 break;
1289                         leaf = path->nodes[0];
1290                 }
1291
1292                 nocow = 0;
1293                 disk_bytenr = 0;
1294                 num_bytes = 0;
1295                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1296
1297                 if (found_key.objectid > ino ||
1298                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1299                     found_key.offset > end)
1300                         break;
1301
1302                 if (found_key.offset > cur_offset) {
1303                         extent_end = found_key.offset;
1304                         extent_type = 0;
1305                         goto out_check;
1306                 }
1307
1308                 fi = btrfs_item_ptr(leaf, path->slots[0],
1309                                     struct btrfs_file_extent_item);
1310                 extent_type = btrfs_file_extent_type(leaf, fi);
1311
1312                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1313                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1314                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1316                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1317                         extent_end = found_key.offset +
1318                                 btrfs_file_extent_num_bytes(leaf, fi);
1319                         disk_num_bytes =
1320                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1321                         if (extent_end <= start) {
1322                                 path->slots[0]++;
1323                                 goto next_slot;
1324                         }
1325                         if (disk_bytenr == 0)
1326                                 goto out_check;
1327                         if (btrfs_file_extent_compression(leaf, fi) ||
1328                             btrfs_file_extent_encryption(leaf, fi) ||
1329                             btrfs_file_extent_other_encoding(leaf, fi))
1330                                 goto out_check;
1331                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1332                                 goto out_check;
1333                         if (btrfs_extent_readonly(root, disk_bytenr))
1334                                 goto out_check;
1335                         if (btrfs_cross_ref_exist(trans, root, ino,
1336                                                   found_key.offset -
1337                                                   extent_offset, disk_bytenr))
1338                                 goto out_check;
1339                         disk_bytenr += extent_offset;
1340                         disk_bytenr += cur_offset - found_key.offset;
1341                         num_bytes = min(end + 1, extent_end) - cur_offset;
1342                         /*
1343                          * if there are pending snapshots for this root,
1344                          * we fall into common COW way.
1345                          */
1346                         if (!nolock) {
1347                                 err = btrfs_start_write_no_snapshoting(root);
1348                                 if (!err)
1349                                         goto out_check;
1350                         }
1351                         /*
1352                          * force cow if csum exists in the range.
1353                          * this ensure that csum for a given extent are
1354                          * either valid or do not exist.
1355                          */
1356                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1357                                 goto out_check;
1358                         nocow = 1;
1359                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1360                         extent_end = found_key.offset +
1361                                 btrfs_file_extent_inline_len(leaf,
1362                                                      path->slots[0], fi);
1363                         extent_end = ALIGN(extent_end, root->sectorsize);
1364                 } else {
1365                         BUG_ON(1);
1366                 }
1367 out_check:
1368                 if (extent_end <= start) {
1369                         path->slots[0]++;
1370                         if (!nolock && nocow)
1371                                 btrfs_end_write_no_snapshoting(root);
1372                         goto next_slot;
1373                 }
1374                 if (!nocow) {
1375                         if (cow_start == (u64)-1)
1376                                 cow_start = cur_offset;
1377                         cur_offset = extent_end;
1378                         if (cur_offset > end)
1379                                 break;
1380                         path->slots[0]++;
1381                         goto next_slot;
1382                 }
1383
1384                 btrfs_release_path(path);
1385                 if (cow_start != (u64)-1) {
1386                         ret = cow_file_range(inode, locked_page,
1387                                              cow_start, found_key.offset - 1,
1388                                              page_started, nr_written, 1);
1389                         if (ret) {
1390                                 if (!nolock && nocow)
1391                                         btrfs_end_write_no_snapshoting(root);
1392                                 goto error;
1393                         }
1394                         cow_start = (u64)-1;
1395                 }
1396
1397                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1398                         struct extent_map *em;
1399                         struct extent_map_tree *em_tree;
1400                         em_tree = &BTRFS_I(inode)->extent_tree;
1401                         em = alloc_extent_map();
1402                         BUG_ON(!em); /* -ENOMEM */
1403                         em->start = cur_offset;
1404                         em->orig_start = found_key.offset - extent_offset;
1405                         em->len = num_bytes;
1406                         em->block_len = num_bytes;
1407                         em->block_start = disk_bytenr;
1408                         em->orig_block_len = disk_num_bytes;
1409                         em->ram_bytes = ram_bytes;
1410                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1411                         em->mod_start = em->start;
1412                         em->mod_len = em->len;
1413                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1414                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1415                         em->generation = -1;
1416                         while (1) {
1417                                 write_lock(&em_tree->lock);
1418                                 ret = add_extent_mapping(em_tree, em, 1);
1419                                 write_unlock(&em_tree->lock);
1420                                 if (ret != -EEXIST) {
1421                                         free_extent_map(em);
1422                                         break;
1423                                 }
1424                                 btrfs_drop_extent_cache(inode, em->start,
1425                                                 em->start + em->len - 1, 0);
1426                         }
1427                         type = BTRFS_ORDERED_PREALLOC;
1428                 } else {
1429                         type = BTRFS_ORDERED_NOCOW;
1430                 }
1431
1432                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1433                                                num_bytes, num_bytes, type);
1434                 BUG_ON(ret); /* -ENOMEM */
1435
1436                 if (root->root_key.objectid ==
1437                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1438                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1439                                                       num_bytes);
1440                         if (ret) {
1441                                 if (!nolock && nocow)
1442                                         btrfs_end_write_no_snapshoting(root);
1443                                 goto error;
1444                         }
1445                 }
1446
1447                 extent_clear_unlock_delalloc(inode, cur_offset,
1448                                              cur_offset + num_bytes - 1,
1449                                              locked_page, EXTENT_LOCKED |
1450                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1451                                              PAGE_SET_PRIVATE2);
1452                 if (!nolock && nocow)
1453                         btrfs_end_write_no_snapshoting(root);
1454                 cur_offset = extent_end;
1455                 if (cur_offset > end)
1456                         break;
1457         }
1458         btrfs_release_path(path);
1459
1460         if (cur_offset <= end && cow_start == (u64)-1) {
1461                 cow_start = cur_offset;
1462                 cur_offset = end;
1463         }
1464
1465         if (cow_start != (u64)-1) {
1466                 ret = cow_file_range(inode, locked_page, cow_start, end,
1467                                      page_started, nr_written, 1);
1468                 if (ret)
1469                         goto error;
1470         }
1471
1472 error:
1473         err = btrfs_end_transaction(trans, root);
1474         if (!ret)
1475                 ret = err;
1476
1477         if (ret && cur_offset < end)
1478                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1479                                              locked_page, EXTENT_LOCKED |
1480                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1481                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1482                                              PAGE_CLEAR_DIRTY |
1483                                              PAGE_SET_WRITEBACK |
1484                                              PAGE_END_WRITEBACK);
1485         btrfs_free_path(path);
1486         return ret;
1487 }
1488
1489 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1490 {
1491
1492         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1493             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1494                 return 0;
1495
1496         /*
1497          * @defrag_bytes is a hint value, no spinlock held here,
1498          * if is not zero, it means the file is defragging.
1499          * Force cow if given extent needs to be defragged.
1500          */
1501         if (BTRFS_I(inode)->defrag_bytes &&
1502             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1503                            EXTENT_DEFRAG, 0, NULL))
1504                 return 1;
1505
1506         return 0;
1507 }
1508
1509 /*
1510  * extent_io.c call back to do delayed allocation processing
1511  */
1512 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1513                               u64 start, u64 end, int *page_started,
1514                               unsigned long *nr_written)
1515 {
1516         int ret;
1517         int force_cow = need_force_cow(inode, start, end);
1518
1519         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1520                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1521                                          page_started, 1, nr_written);
1522         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1523                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1524                                          page_started, 0, nr_written);
1525         } else if (!inode_need_compress(inode)) {
1526                 ret = cow_file_range(inode, locked_page, start, end,
1527                                       page_started, nr_written, 1);
1528         } else {
1529                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1530                         &BTRFS_I(inode)->runtime_flags);
1531                 ret = cow_file_range_async(inode, locked_page, start, end,
1532                                            page_started, nr_written);
1533         }
1534         return ret;
1535 }
1536
1537 static void btrfs_split_extent_hook(struct inode *inode,
1538                                     struct extent_state *orig, u64 split)
1539 {
1540         u64 size;
1541
1542         /* not delalloc, ignore it */
1543         if (!(orig->state & EXTENT_DELALLOC))
1544                 return;
1545
1546         size = orig->end - orig->start + 1;
1547         if (size > BTRFS_MAX_EXTENT_SIZE) {
1548                 u64 num_extents;
1549                 u64 new_size;
1550
1551                 /*
1552                  * See the explanation in btrfs_merge_extent_hook, the same
1553                  * applies here, just in reverse.
1554                  */
1555                 new_size = orig->end - split + 1;
1556                 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1557                                         BTRFS_MAX_EXTENT_SIZE);
1558                 new_size = split - orig->start;
1559                 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1560                                         BTRFS_MAX_EXTENT_SIZE);
1561                 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1562                               BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1563                         return;
1564         }
1565
1566         spin_lock(&BTRFS_I(inode)->lock);
1567         BTRFS_I(inode)->outstanding_extents++;
1568         spin_unlock(&BTRFS_I(inode)->lock);
1569 }
1570
1571 /*
1572  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1573  * extents so we can keep track of new extents that are just merged onto old
1574  * extents, such as when we are doing sequential writes, so we can properly
1575  * account for the metadata space we'll need.
1576  */
1577 static void btrfs_merge_extent_hook(struct inode *inode,
1578                                     struct extent_state *new,
1579                                     struct extent_state *other)
1580 {
1581         u64 new_size, old_size;
1582         u64 num_extents;
1583
1584         /* not delalloc, ignore it */
1585         if (!(other->state & EXTENT_DELALLOC))
1586                 return;
1587
1588         if (new->start > other->start)
1589                 new_size = new->end - other->start + 1;
1590         else
1591                 new_size = other->end - new->start + 1;
1592
1593         /* we're not bigger than the max, unreserve the space and go */
1594         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1595                 spin_lock(&BTRFS_I(inode)->lock);
1596                 BTRFS_I(inode)->outstanding_extents--;
1597                 spin_unlock(&BTRFS_I(inode)->lock);
1598                 return;
1599         }
1600
1601         /*
1602          * We have to add up either side to figure out how many extents were
1603          * accounted for before we merged into one big extent.  If the number of
1604          * extents we accounted for is <= the amount we need for the new range
1605          * then we can return, otherwise drop.  Think of it like this
1606          *
1607          * [ 4k][MAX_SIZE]
1608          *
1609          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1610          * need 2 outstanding extents, on one side we have 1 and the other side
1611          * we have 1 so they are == and we can return.  But in this case
1612          *
1613          * [MAX_SIZE+4k][MAX_SIZE+4k]
1614          *
1615          * Each range on their own accounts for 2 extents, but merged together
1616          * they are only 3 extents worth of accounting, so we need to drop in
1617          * this case.
1618          */
1619         old_size = other->end - other->start + 1;
1620         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1621                                 BTRFS_MAX_EXTENT_SIZE);
1622         old_size = new->end - new->start + 1;
1623         num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1624                                  BTRFS_MAX_EXTENT_SIZE);
1625
1626         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1627                       BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1628                 return;
1629
1630         spin_lock(&BTRFS_I(inode)->lock);
1631         BTRFS_I(inode)->outstanding_extents--;
1632         spin_unlock(&BTRFS_I(inode)->lock);
1633 }
1634
1635 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1636                                       struct inode *inode)
1637 {
1638         spin_lock(&root->delalloc_lock);
1639         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1640                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1641                               &root->delalloc_inodes);
1642                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1643                         &BTRFS_I(inode)->runtime_flags);
1644                 root->nr_delalloc_inodes++;
1645                 if (root->nr_delalloc_inodes == 1) {
1646                         spin_lock(&root->fs_info->delalloc_root_lock);
1647                         BUG_ON(!list_empty(&root->delalloc_root));
1648                         list_add_tail(&root->delalloc_root,
1649                                       &root->fs_info->delalloc_roots);
1650                         spin_unlock(&root->fs_info->delalloc_root_lock);
1651                 }
1652         }
1653         spin_unlock(&root->delalloc_lock);
1654 }
1655
1656 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1657                                      struct inode *inode)
1658 {
1659         spin_lock(&root->delalloc_lock);
1660         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1661                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1662                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1663                           &BTRFS_I(inode)->runtime_flags);
1664                 root->nr_delalloc_inodes--;
1665                 if (!root->nr_delalloc_inodes) {
1666                         spin_lock(&root->fs_info->delalloc_root_lock);
1667                         BUG_ON(list_empty(&root->delalloc_root));
1668                         list_del_init(&root->delalloc_root);
1669                         spin_unlock(&root->fs_info->delalloc_root_lock);
1670                 }
1671         }
1672         spin_unlock(&root->delalloc_lock);
1673 }
1674
1675 /*
1676  * extent_io.c set_bit_hook, used to track delayed allocation
1677  * bytes in this file, and to maintain the list of inodes that
1678  * have pending delalloc work to be done.
1679  */
1680 static void btrfs_set_bit_hook(struct inode *inode,
1681                                struct extent_state *state, unsigned *bits)
1682 {
1683
1684         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1685                 WARN_ON(1);
1686         /*
1687          * set_bit and clear bit hooks normally require _irqsave/restore
1688          * but in this case, we are only testing for the DELALLOC
1689          * bit, which is only set or cleared with irqs on
1690          */
1691         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1692                 struct btrfs_root *root = BTRFS_I(inode)->root;
1693                 u64 len = state->end + 1 - state->start;
1694                 bool do_list = !btrfs_is_free_space_inode(inode);
1695
1696                 if (*bits & EXTENT_FIRST_DELALLOC) {
1697                         *bits &= ~EXTENT_FIRST_DELALLOC;
1698                 } else {
1699                         spin_lock(&BTRFS_I(inode)->lock);
1700                         BTRFS_I(inode)->outstanding_extents++;
1701                         spin_unlock(&BTRFS_I(inode)->lock);
1702                 }
1703
1704                 /* For sanity tests */
1705                 if (btrfs_test_is_dummy_root(root))
1706                         return;
1707
1708                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1709                                      root->fs_info->delalloc_batch);
1710                 spin_lock(&BTRFS_I(inode)->lock);
1711                 BTRFS_I(inode)->delalloc_bytes += len;
1712                 if (*bits & EXTENT_DEFRAG)
1713                         BTRFS_I(inode)->defrag_bytes += len;
1714                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715                                          &BTRFS_I(inode)->runtime_flags))
1716                         btrfs_add_delalloc_inodes(root, inode);
1717                 spin_unlock(&BTRFS_I(inode)->lock);
1718         }
1719 }
1720
1721 /*
1722  * extent_io.c clear_bit_hook, see set_bit_hook for why
1723  */
1724 static void btrfs_clear_bit_hook(struct inode *inode,
1725                                  struct extent_state *state,
1726                                  unsigned *bits)
1727 {
1728         u64 len = state->end + 1 - state->start;
1729         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1730                                     BTRFS_MAX_EXTENT_SIZE);
1731
1732         spin_lock(&BTRFS_I(inode)->lock);
1733         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1734                 BTRFS_I(inode)->defrag_bytes -= len;
1735         spin_unlock(&BTRFS_I(inode)->lock);
1736
1737         /*
1738          * set_bit and clear bit hooks normally require _irqsave/restore
1739          * but in this case, we are only testing for the DELALLOC
1740          * bit, which is only set or cleared with irqs on
1741          */
1742         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1743                 struct btrfs_root *root = BTRFS_I(inode)->root;
1744                 bool do_list = !btrfs_is_free_space_inode(inode);
1745
1746                 if (*bits & EXTENT_FIRST_DELALLOC) {
1747                         *bits &= ~EXTENT_FIRST_DELALLOC;
1748                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1749                         spin_lock(&BTRFS_I(inode)->lock);
1750                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1751                         spin_unlock(&BTRFS_I(inode)->lock);
1752                 }
1753
1754                 /*
1755                  * We don't reserve metadata space for space cache inodes so we
1756                  * don't need to call dellalloc_release_metadata if there is an
1757                  * error.
1758                  */
1759                 if (*bits & EXTENT_DO_ACCOUNTING &&
1760                     root != root->fs_info->tree_root)
1761                         btrfs_delalloc_release_metadata(inode, len);
1762
1763                 /* For sanity tests. */
1764                 if (btrfs_test_is_dummy_root(root))
1765                         return;
1766
1767                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1768                     && do_list && !(state->state & EXTENT_NORESERVE))
1769                         btrfs_free_reserved_data_space(inode, len);
1770
1771                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1772                                      root->fs_info->delalloc_batch);
1773                 spin_lock(&BTRFS_I(inode)->lock);
1774                 BTRFS_I(inode)->delalloc_bytes -= len;
1775                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1776                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1777                              &BTRFS_I(inode)->runtime_flags))
1778                         btrfs_del_delalloc_inode(root, inode);
1779                 spin_unlock(&BTRFS_I(inode)->lock);
1780         }
1781 }
1782
1783 /*
1784  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1785  * we don't create bios that span stripes or chunks
1786  */
1787 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1788                          size_t size, struct bio *bio,
1789                          unsigned long bio_flags)
1790 {
1791         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1792         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1793         u64 length = 0;
1794         u64 map_length;
1795         int ret;
1796
1797         if (bio_flags & EXTENT_BIO_COMPRESSED)
1798                 return 0;
1799
1800         length = bio->bi_iter.bi_size;
1801         map_length = length;
1802         ret = btrfs_map_block(root->fs_info, rw, logical,
1803                               &map_length, NULL, 0);
1804         /* Will always return 0 with map_multi == NULL */
1805         BUG_ON(ret < 0);
1806         if (map_length < length + size)
1807                 return 1;
1808         return 0;
1809 }
1810
1811 /*
1812  * in order to insert checksums into the metadata in large chunks,
1813  * we wait until bio submission time.   All the pages in the bio are
1814  * checksummed and sums are attached onto the ordered extent record.
1815  *
1816  * At IO completion time the cums attached on the ordered extent record
1817  * are inserted into the btree
1818  */
1819 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1820                                     struct bio *bio, int mirror_num,
1821                                     unsigned long bio_flags,
1822                                     u64 bio_offset)
1823 {
1824         struct btrfs_root *root = BTRFS_I(inode)->root;
1825         int ret = 0;
1826
1827         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1828         BUG_ON(ret); /* -ENOMEM */
1829         return 0;
1830 }
1831
1832 /*
1833  * in order to insert checksums into the metadata in large chunks,
1834  * we wait until bio submission time.   All the pages in the bio are
1835  * checksummed and sums are attached onto the ordered extent record.
1836  *
1837  * At IO completion time the cums attached on the ordered extent record
1838  * are inserted into the btree
1839  */
1840 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1841                           int mirror_num, unsigned long bio_flags,
1842                           u64 bio_offset)
1843 {
1844         struct btrfs_root *root = BTRFS_I(inode)->root;
1845         int ret;
1846
1847         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1848         if (ret) {
1849                 bio->bi_error = ret;
1850                 bio_endio(bio);
1851         }
1852         return ret;
1853 }
1854
1855 /*
1856  * extent_io.c submission hook. This does the right thing for csum calculation
1857  * on write, or reading the csums from the tree before a read
1858  */
1859 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1860                           int mirror_num, unsigned long bio_flags,
1861                           u64 bio_offset)
1862 {
1863         struct btrfs_root *root = BTRFS_I(inode)->root;
1864         int ret = 0;
1865         int skip_sum;
1866         int metadata = 0;
1867         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1868
1869         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1870
1871         if (btrfs_is_free_space_inode(inode))
1872                 metadata = 2;
1873
1874         if (!(rw & REQ_WRITE)) {
1875                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1876                 if (ret)
1877                         goto out;
1878
1879                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1880                         ret = btrfs_submit_compressed_read(inode, bio,
1881                                                            mirror_num,
1882                                                            bio_flags);
1883                         goto out;
1884                 } else if (!skip_sum) {
1885                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1886                         if (ret)
1887                                 goto out;
1888                 }
1889                 goto mapit;
1890         } else if (async && !skip_sum) {
1891                 /* csum items have already been cloned */
1892                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1893                         goto mapit;
1894                 /* we're doing a write, do the async checksumming */
1895                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1896                                    inode, rw, bio, mirror_num,
1897                                    bio_flags, bio_offset,
1898                                    __btrfs_submit_bio_start,
1899                                    __btrfs_submit_bio_done);
1900                 goto out;
1901         } else if (!skip_sum) {
1902                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1903                 if (ret)
1904                         goto out;
1905         }
1906
1907 mapit:
1908         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1909
1910 out:
1911         if (ret < 0) {
1912                 bio->bi_error = ret;
1913                 bio_endio(bio);
1914         }
1915         return ret;
1916 }
1917
1918 /*
1919  * given a list of ordered sums record them in the inode.  This happens
1920  * at IO completion time based on sums calculated at bio submission time.
1921  */
1922 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1923                              struct inode *inode, u64 file_offset,
1924                              struct list_head *list)
1925 {
1926         struct btrfs_ordered_sum *sum;
1927
1928         list_for_each_entry(sum, list, list) {
1929                 trans->adding_csums = 1;
1930                 btrfs_csum_file_blocks(trans,
1931                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1932                 trans->adding_csums = 0;
1933         }
1934         return 0;
1935 }
1936
1937 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1938                               struct extent_state **cached_state)
1939 {
1940         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1941         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1942                                    cached_state, GFP_NOFS);
1943 }
1944
1945 /* see btrfs_writepage_start_hook for details on why this is required */
1946 struct btrfs_writepage_fixup {
1947         struct page *page;
1948         struct btrfs_work work;
1949 };
1950
1951 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1952 {
1953         struct btrfs_writepage_fixup *fixup;
1954         struct btrfs_ordered_extent *ordered;
1955         struct extent_state *cached_state = NULL;
1956         struct page *page;
1957         struct inode *inode;
1958         u64 page_start;
1959         u64 page_end;
1960         int ret;
1961
1962         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1963         page = fixup->page;
1964 again:
1965         lock_page(page);
1966         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1967                 ClearPageChecked(page);
1968                 goto out_page;
1969         }
1970
1971         inode = page->mapping->host;
1972         page_start = page_offset(page);
1973         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1974
1975         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1976                          &cached_state);
1977
1978         /* already ordered? We're done */
1979         if (PagePrivate2(page))
1980                 goto out;
1981
1982         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1983         if (ordered) {
1984                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1985                                      page_end, &cached_state, GFP_NOFS);
1986                 unlock_page(page);
1987                 btrfs_start_ordered_extent(inode, ordered, 1);
1988                 btrfs_put_ordered_extent(ordered);
1989                 goto again;
1990         }
1991
1992         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1993         if (ret) {
1994                 mapping_set_error(page->mapping, ret);
1995                 end_extent_writepage(page, ret, page_start, page_end);
1996                 ClearPageChecked(page);
1997                 goto out;
1998          }
1999
2000         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2001         ClearPageChecked(page);
2002         set_page_dirty(page);
2003 out:
2004         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2005                              &cached_state, GFP_NOFS);
2006 out_page:
2007         unlock_page(page);
2008         page_cache_release(page);
2009         kfree(fixup);
2010 }
2011
2012 /*
2013  * There are a few paths in the higher layers of the kernel that directly
2014  * set the page dirty bit without asking the filesystem if it is a
2015  * good idea.  This causes problems because we want to make sure COW
2016  * properly happens and the data=ordered rules are followed.
2017  *
2018  * In our case any range that doesn't have the ORDERED bit set
2019  * hasn't been properly setup for IO.  We kick off an async process
2020  * to fix it up.  The async helper will wait for ordered extents, set
2021  * the delalloc bit and make it safe to write the page.
2022  */
2023 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2024 {
2025         struct inode *inode = page->mapping->host;
2026         struct btrfs_writepage_fixup *fixup;
2027         struct btrfs_root *root = BTRFS_I(inode)->root;
2028
2029         /* this page is properly in the ordered list */
2030         if (TestClearPagePrivate2(page))
2031                 return 0;
2032
2033         if (PageChecked(page))
2034                 return -EAGAIN;
2035
2036         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2037         if (!fixup)
2038                 return -EAGAIN;
2039
2040         SetPageChecked(page);
2041         page_cache_get(page);
2042         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2043                         btrfs_writepage_fixup_worker, NULL, NULL);
2044         fixup->page = page;
2045         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2046         return -EBUSY;
2047 }
2048
2049 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2050                                        struct inode *inode, u64 file_pos,
2051                                        u64 disk_bytenr, u64 disk_num_bytes,
2052                                        u64 num_bytes, u64 ram_bytes,
2053                                        u8 compression, u8 encryption,
2054                                        u16 other_encoding, int extent_type)
2055 {
2056         struct btrfs_root *root = BTRFS_I(inode)->root;
2057         struct btrfs_file_extent_item *fi;
2058         struct btrfs_path *path;
2059         struct extent_buffer *leaf;
2060         struct btrfs_key ins;
2061         int extent_inserted = 0;
2062         int ret;
2063
2064         path = btrfs_alloc_path();
2065         if (!path)
2066                 return -ENOMEM;
2067
2068         /*
2069          * we may be replacing one extent in the tree with another.
2070          * The new extent is pinned in the extent map, and we don't want
2071          * to drop it from the cache until it is completely in the btree.
2072          *
2073          * So, tell btrfs_drop_extents to leave this extent in the cache.
2074          * the caller is expected to unpin it and allow it to be merged
2075          * with the others.
2076          */
2077         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2078                                    file_pos + num_bytes, NULL, 0,
2079                                    1, sizeof(*fi), &extent_inserted);
2080         if (ret)
2081                 goto out;
2082
2083         if (!extent_inserted) {
2084                 ins.objectid = btrfs_ino(inode);
2085                 ins.offset = file_pos;
2086                 ins.type = BTRFS_EXTENT_DATA_KEY;
2087
2088                 path->leave_spinning = 1;
2089                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2090                                               sizeof(*fi));
2091                 if (ret)
2092                         goto out;
2093         }
2094         leaf = path->nodes[0];
2095         fi = btrfs_item_ptr(leaf, path->slots[0],
2096                             struct btrfs_file_extent_item);
2097         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2098         btrfs_set_file_extent_type(leaf, fi, extent_type);
2099         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2100         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2101         btrfs_set_file_extent_offset(leaf, fi, 0);
2102         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2103         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2104         btrfs_set_file_extent_compression(leaf, fi, compression);
2105         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2106         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2107
2108         btrfs_mark_buffer_dirty(leaf);
2109         btrfs_release_path(path);
2110
2111         inode_add_bytes(inode, num_bytes);
2112
2113         ins.objectid = disk_bytenr;
2114         ins.offset = disk_num_bytes;
2115         ins.type = BTRFS_EXTENT_ITEM_KEY;
2116         ret = btrfs_alloc_reserved_file_extent(trans, root,
2117                                         root->root_key.objectid,
2118                                         btrfs_ino(inode), file_pos, &ins);
2119 out:
2120         btrfs_free_path(path);
2121
2122         return ret;
2123 }
2124
2125 /* snapshot-aware defrag */
2126 struct sa_defrag_extent_backref {
2127         struct rb_node node;
2128         struct old_sa_defrag_extent *old;
2129         u64 root_id;
2130         u64 inum;
2131         u64 file_pos;
2132         u64 extent_offset;
2133         u64 num_bytes;
2134         u64 generation;
2135 };
2136
2137 struct old_sa_defrag_extent {
2138         struct list_head list;
2139         struct new_sa_defrag_extent *new;
2140
2141         u64 extent_offset;
2142         u64 bytenr;
2143         u64 offset;
2144         u64 len;
2145         int count;
2146 };
2147
2148 struct new_sa_defrag_extent {
2149         struct rb_root root;
2150         struct list_head head;
2151         struct btrfs_path *path;
2152         struct inode *inode;
2153         u64 file_pos;
2154         u64 len;
2155         u64 bytenr;
2156         u64 disk_len;
2157         u8 compress_type;
2158 };
2159
2160 static int backref_comp(struct sa_defrag_extent_backref *b1,
2161                         struct sa_defrag_extent_backref *b2)
2162 {
2163         if (b1->root_id < b2->root_id)
2164                 return -1;
2165         else if (b1->root_id > b2->root_id)
2166                 return 1;
2167
2168         if (b1->inum < b2->inum)
2169                 return -1;
2170         else if (b1->inum > b2->inum)
2171                 return 1;
2172
2173         if (b1->file_pos < b2->file_pos)
2174                 return -1;
2175         else if (b1->file_pos > b2->file_pos)
2176                 return 1;
2177
2178         /*
2179          * [------------------------------] ===> (a range of space)
2180          *     |<--->|   |<---->| =============> (fs/file tree A)
2181          * |<---------------------------->| ===> (fs/file tree B)
2182          *
2183          * A range of space can refer to two file extents in one tree while
2184          * refer to only one file extent in another tree.
2185          *
2186          * So we may process a disk offset more than one time(two extents in A)
2187          * and locate at the same extent(one extent in B), then insert two same
2188          * backrefs(both refer to the extent in B).
2189          */
2190         return 0;
2191 }
2192
2193 static void backref_insert(struct rb_root *root,
2194                            struct sa_defrag_extent_backref *backref)
2195 {
2196         struct rb_node **p = &root->rb_node;
2197         struct rb_node *parent = NULL;
2198         struct sa_defrag_extent_backref *entry;
2199         int ret;
2200
2201         while (*p) {
2202                 parent = *p;
2203                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2204
2205                 ret = backref_comp(backref, entry);
2206                 if (ret < 0)
2207                         p = &(*p)->rb_left;
2208                 else
2209                         p = &(*p)->rb_right;
2210         }
2211
2212         rb_link_node(&backref->node, parent, p);
2213         rb_insert_color(&backref->node, root);
2214 }
2215
2216 /*
2217  * Note the backref might has changed, and in this case we just return 0.
2218  */
2219 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2220                                        void *ctx)
2221 {
2222         struct btrfs_file_extent_item *extent;
2223         struct btrfs_fs_info *fs_info;
2224         struct old_sa_defrag_extent *old = ctx;
2225         struct new_sa_defrag_extent *new = old->new;
2226         struct btrfs_path *path = new->path;
2227         struct btrfs_key key;
2228         struct btrfs_root *root;
2229         struct sa_defrag_extent_backref *backref;
2230         struct extent_buffer *leaf;
2231         struct inode *inode = new->inode;
2232         int slot;
2233         int ret;
2234         u64 extent_offset;
2235         u64 num_bytes;
2236
2237         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2238             inum == btrfs_ino(inode))
2239                 return 0;
2240
2241         key.objectid = root_id;
2242         key.type = BTRFS_ROOT_ITEM_KEY;
2243         key.offset = (u64)-1;
2244
2245         fs_info = BTRFS_I(inode)->root->fs_info;
2246         root = btrfs_read_fs_root_no_name(fs_info, &key);
2247         if (IS_ERR(root)) {
2248                 if (PTR_ERR(root) == -ENOENT)
2249                         return 0;
2250                 WARN_ON(1);
2251                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2252                          inum, offset, root_id);
2253                 return PTR_ERR(root);
2254         }
2255
2256         key.objectid = inum;
2257         key.type = BTRFS_EXTENT_DATA_KEY;
2258         if (offset > (u64)-1 << 32)
2259                 key.offset = 0;
2260         else
2261                 key.offset = offset;
2262
2263         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2264         if (WARN_ON(ret < 0))
2265                 return ret;
2266         ret = 0;
2267
2268         while (1) {
2269                 cond_resched();
2270
2271                 leaf = path->nodes[0];
2272                 slot = path->slots[0];
2273
2274                 if (slot >= btrfs_header_nritems(leaf)) {
2275                         ret = btrfs_next_leaf(root, path);
2276                         if (ret < 0) {
2277                                 goto out;
2278                         } else if (ret > 0) {
2279                                 ret = 0;
2280                                 goto out;
2281                         }
2282                         continue;
2283                 }
2284
2285                 path->slots[0]++;
2286
2287                 btrfs_item_key_to_cpu(leaf, &key, slot);
2288
2289                 if (key.objectid > inum)
2290                         goto out;
2291
2292                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2293                         continue;
2294
2295                 extent = btrfs_item_ptr(leaf, slot,
2296                                         struct btrfs_file_extent_item);
2297
2298                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2299                         continue;
2300
2301                 /*
2302                  * 'offset' refers to the exact key.offset,
2303                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2304                  * (key.offset - extent_offset).
2305                  */
2306                 if (key.offset != offset)
2307                         continue;
2308
2309                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2310                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2311
2312                 if (extent_offset >= old->extent_offset + old->offset +
2313                     old->len || extent_offset + num_bytes <=
2314                     old->extent_offset + old->offset)
2315                         continue;
2316                 break;
2317         }
2318
2319         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2320         if (!backref) {
2321                 ret = -ENOENT;
2322                 goto out;
2323         }
2324
2325         backref->root_id = root_id;
2326         backref->inum = inum;
2327         backref->file_pos = offset;
2328         backref->num_bytes = num_bytes;
2329         backref->extent_offset = extent_offset;
2330         backref->generation = btrfs_file_extent_generation(leaf, extent);
2331         backref->old = old;
2332         backref_insert(&new->root, backref);
2333         old->count++;
2334 out:
2335         btrfs_release_path(path);
2336         WARN_ON(ret);
2337         return ret;
2338 }
2339
2340 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2341                                    struct new_sa_defrag_extent *new)
2342 {
2343         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2344         struct old_sa_defrag_extent *old, *tmp;
2345         int ret;
2346
2347         new->path = path;
2348
2349         list_for_each_entry_safe(old, tmp, &new->head, list) {
2350                 ret = iterate_inodes_from_logical(old->bytenr +
2351                                                   old->extent_offset, fs_info,
2352                                                   path, record_one_backref,
2353                                                   old);
2354                 if (ret < 0 && ret != -ENOENT)
2355                         return false;
2356
2357                 /* no backref to be processed for this extent */
2358                 if (!old->count) {
2359                         list_del(&old->list);
2360                         kfree(old);
2361                 }
2362         }
2363
2364         if (list_empty(&new->head))
2365                 return false;
2366
2367         return true;
2368 }
2369
2370 static int relink_is_mergable(struct extent_buffer *leaf,
2371                               struct btrfs_file_extent_item *fi,
2372                               struct new_sa_defrag_extent *new)
2373 {
2374         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2375                 return 0;
2376
2377         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2378                 return 0;
2379
2380         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2381                 return 0;
2382
2383         if (btrfs_file_extent_encryption(leaf, fi) ||
2384             btrfs_file_extent_other_encoding(leaf, fi))
2385                 return 0;
2386
2387         return 1;
2388 }
2389
2390 /*
2391  * Note the backref might has changed, and in this case we just return 0.
2392  */
2393 static noinline int relink_extent_backref(struct btrfs_path *path,
2394                                  struct sa_defrag_extent_backref *prev,
2395                                  struct sa_defrag_extent_backref *backref)
2396 {
2397         struct btrfs_file_extent_item *extent;
2398         struct btrfs_file_extent_item *item;
2399         struct btrfs_ordered_extent *ordered;
2400         struct btrfs_trans_handle *trans;
2401         struct btrfs_fs_info *fs_info;
2402         struct btrfs_root *root;
2403         struct btrfs_key key;
2404         struct extent_buffer *leaf;
2405         struct old_sa_defrag_extent *old = backref->old;
2406         struct new_sa_defrag_extent *new = old->new;
2407         struct inode *src_inode = new->inode;
2408         struct inode *inode;
2409         struct extent_state *cached = NULL;
2410         int ret = 0;
2411         u64 start;
2412         u64 len;
2413         u64 lock_start;
2414         u64 lock_end;
2415         bool merge = false;
2416         int index;
2417
2418         if (prev && prev->root_id == backref->root_id &&
2419             prev->inum == backref->inum &&
2420             prev->file_pos + prev->num_bytes == backref->file_pos)
2421                 merge = true;
2422
2423         /* step 1: get root */
2424         key.objectid = backref->root_id;
2425         key.type = BTRFS_ROOT_ITEM_KEY;
2426         key.offset = (u64)-1;
2427
2428         fs_info = BTRFS_I(src_inode)->root->fs_info;
2429         index = srcu_read_lock(&fs_info->subvol_srcu);
2430
2431         root = btrfs_read_fs_root_no_name(fs_info, &key);
2432         if (IS_ERR(root)) {
2433                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2434                 if (PTR_ERR(root) == -ENOENT)
2435                         return 0;
2436                 return PTR_ERR(root);
2437         }
2438
2439         if (btrfs_root_readonly(root)) {
2440                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2441                 return 0;
2442         }
2443
2444         /* step 2: get inode */
2445         key.objectid = backref->inum;
2446         key.type = BTRFS_INODE_ITEM_KEY;
2447         key.offset = 0;
2448
2449         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2450         if (IS_ERR(inode)) {
2451                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2452                 return 0;
2453         }
2454
2455         srcu_read_unlock(&fs_info->subvol_srcu, index);
2456
2457         /* step 3: relink backref */
2458         lock_start = backref->file_pos;
2459         lock_end = backref->file_pos + backref->num_bytes - 1;
2460         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2461                          0, &cached);
2462
2463         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2464         if (ordered) {
2465                 btrfs_put_ordered_extent(ordered);
2466                 goto out_unlock;
2467         }
2468
2469         trans = btrfs_join_transaction(root);
2470         if (IS_ERR(trans)) {
2471                 ret = PTR_ERR(trans);
2472                 goto out_unlock;
2473         }
2474
2475         key.objectid = backref->inum;
2476         key.type = BTRFS_EXTENT_DATA_KEY;
2477         key.offset = backref->file_pos;
2478
2479         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2480         if (ret < 0) {
2481                 goto out_free_path;
2482         } else if (ret > 0) {
2483                 ret = 0;
2484                 goto out_free_path;
2485         }
2486
2487         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2488                                 struct btrfs_file_extent_item);
2489
2490         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2491             backref->generation)
2492                 goto out_free_path;
2493
2494         btrfs_release_path(path);
2495
2496         start = backref->file_pos;
2497         if (backref->extent_offset < old->extent_offset + old->offset)
2498                 start += old->extent_offset + old->offset -
2499                          backref->extent_offset;
2500
2501         len = min(backref->extent_offset + backref->num_bytes,
2502                   old->extent_offset + old->offset + old->len);
2503         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2504
2505         ret = btrfs_drop_extents(trans, root, inode, start,
2506                                  start + len, 1);
2507         if (ret)
2508                 goto out_free_path;
2509 again:
2510         key.objectid = btrfs_ino(inode);
2511         key.type = BTRFS_EXTENT_DATA_KEY;
2512         key.offset = start;
2513
2514         path->leave_spinning = 1;
2515         if (merge) {
2516                 struct btrfs_file_extent_item *fi;
2517                 u64 extent_len;
2518                 struct btrfs_key found_key;
2519
2520                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2521                 if (ret < 0)
2522                         goto out_free_path;
2523
2524                 path->slots[0]--;
2525                 leaf = path->nodes[0];
2526                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2527
2528                 fi = btrfs_item_ptr(leaf, path->slots[0],
2529                                     struct btrfs_file_extent_item);
2530                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2531
2532                 if (extent_len + found_key.offset == start &&
2533                     relink_is_mergable(leaf, fi, new)) {
2534                         btrfs_set_file_extent_num_bytes(leaf, fi,
2535                                                         extent_len + len);
2536                         btrfs_mark_buffer_dirty(leaf);
2537                         inode_add_bytes(inode, len);
2538
2539                         ret = 1;
2540                         goto out_free_path;
2541                 } else {
2542                         merge = false;
2543                         btrfs_release_path(path);
2544                         goto again;
2545                 }
2546         }
2547
2548         ret = btrfs_insert_empty_item(trans, root, path, &key,
2549                                         sizeof(*extent));
2550         if (ret) {
2551                 btrfs_abort_transaction(trans, root, ret);
2552                 goto out_free_path;
2553         }
2554
2555         leaf = path->nodes[0];
2556         item = btrfs_item_ptr(leaf, path->slots[0],
2557                                 struct btrfs_file_extent_item);
2558         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2559         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2560         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2561         btrfs_set_file_extent_num_bytes(leaf, item, len);
2562         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2563         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2564         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2565         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2566         btrfs_set_file_extent_encryption(leaf, item, 0);
2567         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2568
2569         btrfs_mark_buffer_dirty(leaf);
2570         inode_add_bytes(inode, len);
2571         btrfs_release_path(path);
2572
2573         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2574                         new->disk_len, 0,
2575                         backref->root_id, backref->inum,
2576                         new->file_pos, 0);      /* start - extent_offset */
2577         if (ret) {
2578                 btrfs_abort_transaction(trans, root, ret);
2579                 goto out_free_path;
2580         }
2581
2582         ret = 1;
2583 out_free_path:
2584         btrfs_release_path(path);
2585         path->leave_spinning = 0;
2586         btrfs_end_transaction(trans, root);
2587 out_unlock:
2588         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2589                              &cached, GFP_NOFS);
2590         iput(inode);
2591         return ret;
2592 }
2593
2594 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2595 {
2596         struct old_sa_defrag_extent *old, *tmp;
2597
2598         if (!new)
2599                 return;
2600
2601         list_for_each_entry_safe(old, tmp, &new->head, list) {
2602                 list_del(&old->list);
2603                 kfree(old);
2604         }
2605         kfree(new);
2606 }
2607
2608 static void relink_file_extents(struct new_sa_defrag_extent *new)
2609 {
2610         struct btrfs_path *path;
2611         struct sa_defrag_extent_backref *backref;
2612         struct sa_defrag_extent_backref *prev = NULL;
2613         struct inode *inode;
2614         struct btrfs_root *root;
2615         struct rb_node *node;
2616         int ret;
2617
2618         inode = new->inode;
2619         root = BTRFS_I(inode)->root;
2620
2621         path = btrfs_alloc_path();
2622         if (!path)
2623                 return;
2624
2625         if (!record_extent_backrefs(path, new)) {
2626                 btrfs_free_path(path);
2627                 goto out;
2628         }
2629         btrfs_release_path(path);
2630
2631         while (1) {
2632                 node = rb_first(&new->root);
2633                 if (!node)
2634                         break;
2635                 rb_erase(node, &new->root);
2636
2637                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2638
2639                 ret = relink_extent_backref(path, prev, backref);
2640                 WARN_ON(ret < 0);
2641
2642                 kfree(prev);
2643
2644                 if (ret == 1)
2645                         prev = backref;
2646                 else
2647                         prev = NULL;
2648                 cond_resched();
2649         }
2650         kfree(prev);
2651
2652         btrfs_free_path(path);
2653 out:
2654         free_sa_defrag_extent(new);
2655
2656         atomic_dec(&root->fs_info->defrag_running);
2657         wake_up(&root->fs_info->transaction_wait);
2658 }
2659
2660 static struct new_sa_defrag_extent *
2661 record_old_file_extents(struct inode *inode,
2662                         struct btrfs_ordered_extent *ordered)
2663 {
2664         struct btrfs_root *root = BTRFS_I(inode)->root;
2665         struct btrfs_path *path;
2666         struct btrfs_key key;
2667         struct old_sa_defrag_extent *old;
2668         struct new_sa_defrag_extent *new;
2669         int ret;
2670
2671         new = kmalloc(sizeof(*new), GFP_NOFS);
2672         if (!new)
2673                 return NULL;
2674
2675         new->inode = inode;
2676         new->file_pos = ordered->file_offset;
2677         new->len = ordered->len;
2678         new->bytenr = ordered->start;
2679         new->disk_len = ordered->disk_len;
2680         new->compress_type = ordered->compress_type;
2681         new->root = RB_ROOT;
2682         INIT_LIST_HEAD(&new->head);
2683
2684         path = btrfs_alloc_path();
2685         if (!path)
2686                 goto out_kfree;
2687
2688         key.objectid = btrfs_ino(inode);
2689         key.type = BTRFS_EXTENT_DATA_KEY;
2690         key.offset = new->file_pos;
2691
2692         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2693         if (ret < 0)
2694                 goto out_free_path;
2695         if (ret > 0 && path->slots[0] > 0)
2696                 path->slots[0]--;
2697
2698         /* find out all the old extents for the file range */
2699         while (1) {
2700                 struct btrfs_file_extent_item *extent;
2701                 struct extent_buffer *l;
2702                 int slot;
2703                 u64 num_bytes;
2704                 u64 offset;
2705                 u64 end;
2706                 u64 disk_bytenr;
2707                 u64 extent_offset;
2708
2709                 l = path->nodes[0];
2710                 slot = path->slots[0];
2711
2712                 if (slot >= btrfs_header_nritems(l)) {
2713                         ret = btrfs_next_leaf(root, path);
2714                         if (ret < 0)
2715                                 goto out_free_path;
2716                         else if (ret > 0)
2717                                 break;
2718                         continue;
2719                 }
2720
2721                 btrfs_item_key_to_cpu(l, &key, slot);
2722
2723                 if (key.objectid != btrfs_ino(inode))
2724                         break;
2725                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2726                         break;
2727                 if (key.offset >= new->file_pos + new->len)
2728                         break;
2729
2730                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2731
2732                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2733                 if (key.offset + num_bytes < new->file_pos)
2734                         goto next;
2735
2736                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2737                 if (!disk_bytenr)
2738                         goto next;
2739
2740                 extent_offset = btrfs_file_extent_offset(l, extent);
2741
2742                 old = kmalloc(sizeof(*old), GFP_NOFS);
2743                 if (!old)
2744                         goto out_free_path;
2745
2746                 offset = max(new->file_pos, key.offset);
2747                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2748
2749                 old->bytenr = disk_bytenr;
2750                 old->extent_offset = extent_offset;
2751                 old->offset = offset - key.offset;
2752                 old->len = end - offset;
2753                 old->new = new;
2754                 old->count = 0;
2755                 list_add_tail(&old->list, &new->head);
2756 next:
2757                 path->slots[0]++;
2758                 cond_resched();
2759         }
2760
2761         btrfs_free_path(path);
2762         atomic_inc(&root->fs_info->defrag_running);
2763
2764         return new;
2765
2766 out_free_path:
2767         btrfs_free_path(path);
2768 out_kfree:
2769         free_sa_defrag_extent(new);
2770         return NULL;
2771 }
2772
2773 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2774                                          u64 start, u64 len)
2775 {
2776         struct btrfs_block_group_cache *cache;
2777
2778         cache = btrfs_lookup_block_group(root->fs_info, start);
2779         ASSERT(cache);
2780
2781         spin_lock(&cache->lock);
2782         cache->delalloc_bytes -= len;
2783         spin_unlock(&cache->lock);
2784
2785         btrfs_put_block_group(cache);
2786 }
2787
2788 /* as ordered data IO finishes, this gets called so we can finish
2789  * an ordered extent if the range of bytes in the file it covers are
2790  * fully written.
2791  */
2792 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2793 {
2794         struct inode *inode = ordered_extent->inode;
2795         struct btrfs_root *root = BTRFS_I(inode)->root;
2796         struct btrfs_trans_handle *trans = NULL;
2797         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2798         struct extent_state *cached_state = NULL;
2799         struct new_sa_defrag_extent *new = NULL;
2800         int compress_type = 0;
2801         int ret = 0;
2802         u64 logical_len = ordered_extent->len;
2803         bool nolock;
2804         bool truncated = false;
2805
2806         nolock = btrfs_is_free_space_inode(inode);
2807
2808         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2809                 ret = -EIO;
2810                 goto out;
2811         }
2812
2813         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2814                                      ordered_extent->file_offset +
2815                                      ordered_extent->len - 1);
2816
2817         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2818                 truncated = true;
2819                 logical_len = ordered_extent->truncated_len;
2820                 /* Truncated the entire extent, don't bother adding */
2821                 if (!logical_len)
2822                         goto out;
2823         }
2824
2825         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2826                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2827                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2828                 if (nolock)
2829                         trans = btrfs_join_transaction_nolock(root);
2830                 else
2831                         trans = btrfs_join_transaction(root);
2832                 if (IS_ERR(trans)) {
2833                         ret = PTR_ERR(trans);
2834                         trans = NULL;
2835                         goto out;
2836                 }
2837                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2838                 ret = btrfs_update_inode_fallback(trans, root, inode);
2839                 if (ret) /* -ENOMEM or corruption */
2840                         btrfs_abort_transaction(trans, root, ret);
2841                 goto out;
2842         }
2843
2844         lock_extent_bits(io_tree, ordered_extent->file_offset,
2845                          ordered_extent->file_offset + ordered_extent->len - 1,
2846                          0, &cached_state);
2847
2848         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2849                         ordered_extent->file_offset + ordered_extent->len - 1,
2850                         EXTENT_DEFRAG, 1, cached_state);
2851         if (ret) {
2852                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2853                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2854                         /* the inode is shared */
2855                         new = record_old_file_extents(inode, ordered_extent);
2856
2857                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2858                         ordered_extent->file_offset + ordered_extent->len - 1,
2859                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2860         }
2861
2862         if (nolock)
2863                 trans = btrfs_join_transaction_nolock(root);
2864         else
2865                 trans = btrfs_join_transaction(root);
2866         if (IS_ERR(trans)) {
2867                 ret = PTR_ERR(trans);
2868                 trans = NULL;
2869                 goto out_unlock;
2870         }
2871
2872         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2873
2874         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2875                 compress_type = ordered_extent->compress_type;
2876         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2877                 BUG_ON(compress_type);
2878                 ret = btrfs_mark_extent_written(trans, inode,
2879                                                 ordered_extent->file_offset,
2880                                                 ordered_extent->file_offset +
2881                                                 logical_len);
2882         } else {
2883                 BUG_ON(root == root->fs_info->tree_root);
2884                 ret = insert_reserved_file_extent(trans, inode,
2885                                                 ordered_extent->file_offset,
2886                                                 ordered_extent->start,
2887                                                 ordered_extent->disk_len,
2888                                                 logical_len, logical_len,
2889                                                 compress_type, 0, 0,
2890                                                 BTRFS_FILE_EXTENT_REG);
2891                 if (!ret)
2892                         btrfs_release_delalloc_bytes(root,
2893                                                      ordered_extent->start,
2894                                                      ordered_extent->disk_len);
2895         }
2896         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2897                            ordered_extent->file_offset, ordered_extent->len,
2898                            trans->transid);
2899         if (ret < 0) {
2900                 btrfs_abort_transaction(trans, root, ret);
2901                 goto out_unlock;
2902         }
2903
2904         add_pending_csums(trans, inode, ordered_extent->file_offset,
2905                           &ordered_extent->list);
2906
2907         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2908         ret = btrfs_update_inode_fallback(trans, root, inode);
2909         if (ret) { /* -ENOMEM or corruption */
2910                 btrfs_abort_transaction(trans, root, ret);
2911                 goto out_unlock;
2912         }
2913         ret = 0;
2914 out_unlock:
2915         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2916                              ordered_extent->file_offset +
2917                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2918 out:
2919         if (root != root->fs_info->tree_root)
2920                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2921         if (trans)
2922                 btrfs_end_transaction(trans, root);
2923
2924         if (ret || truncated) {
2925                 u64 start, end;
2926
2927                 if (truncated)
2928                         start = ordered_extent->file_offset + logical_len;
2929                 else
2930                         start = ordered_extent->file_offset;
2931                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2932                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2933
2934                 /* Drop the cache for the part of the extent we didn't write. */
2935                 btrfs_drop_extent_cache(inode, start, end, 0);
2936
2937                 /*
2938                  * If the ordered extent had an IOERR or something else went
2939                  * wrong we need to return the space for this ordered extent
2940                  * back to the allocator.  We only free the extent in the
2941                  * truncated case if we didn't write out the extent at all.
2942                  */
2943                 if ((ret || !logical_len) &&
2944                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2945                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2946                         btrfs_free_reserved_extent(root, ordered_extent->start,
2947                                                    ordered_extent->disk_len, 1);
2948         }
2949
2950
2951         /*
2952          * This needs to be done to make sure anybody waiting knows we are done
2953          * updating everything for this ordered extent.
2954          */
2955         btrfs_remove_ordered_extent(inode, ordered_extent);
2956
2957         /* for snapshot-aware defrag */
2958         if (new) {
2959                 if (ret) {
2960                         free_sa_defrag_extent(new);
2961                         atomic_dec(&root->fs_info->defrag_running);
2962                 } else {
2963                         relink_file_extents(new);
2964                 }
2965         }
2966
2967         /* once for us */
2968         btrfs_put_ordered_extent(ordered_extent);
2969         /* once for the tree */
2970         btrfs_put_ordered_extent(ordered_extent);
2971
2972         return ret;
2973 }
2974
2975 static void finish_ordered_fn(struct btrfs_work *work)
2976 {
2977         struct btrfs_ordered_extent *ordered_extent;
2978         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2979         btrfs_finish_ordered_io(ordered_extent);
2980 }
2981
2982 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2983                                 struct extent_state *state, int uptodate)
2984 {
2985         struct inode *inode = page->mapping->host;
2986         struct btrfs_root *root = BTRFS_I(inode)->root;
2987         struct btrfs_ordered_extent *ordered_extent = NULL;
2988         struct btrfs_workqueue *wq;
2989         btrfs_work_func_t func;
2990
2991         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2992
2993         ClearPagePrivate2(page);
2994         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2995                                             end - start + 1, uptodate))
2996                 return 0;
2997
2998         if (btrfs_is_free_space_inode(inode)) {
2999                 wq = root->fs_info->endio_freespace_worker;
3000                 func = btrfs_freespace_write_helper;
3001         } else {
3002                 wq = root->fs_info->endio_write_workers;
3003                 func = btrfs_endio_write_helper;
3004         }
3005
3006         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3007                         NULL);
3008         btrfs_queue_work(wq, &ordered_extent->work);
3009
3010         return 0;
3011 }
3012
3013 static int __readpage_endio_check(struct inode *inode,
3014                                   struct btrfs_io_bio *io_bio,
3015                                   int icsum, struct page *page,
3016                                   int pgoff, u64 start, size_t len)
3017 {
3018         char *kaddr;
3019         u32 csum_expected;
3020         u32 csum = ~(u32)0;
3021         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
3022                                       DEFAULT_RATELIMIT_BURST);
3023
3024         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3025
3026         kaddr = kmap_atomic(page);
3027         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3028         btrfs_csum_final(csum, (char *)&csum);
3029         if (csum != csum_expected)
3030                 goto zeroit;
3031
3032         kunmap_atomic(kaddr);
3033         return 0;
3034 zeroit:
3035         if (__ratelimit(&_rs))
3036                 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3037                            "csum failed ino %llu off %llu csum %u expected csum %u",
3038                            btrfs_ino(inode), start, csum, csum_expected);
3039         memset(kaddr + pgoff, 1, len);
3040         flush_dcache_page(page);
3041         kunmap_atomic(kaddr);
3042         if (csum_expected == 0)
3043                 return 0;
3044         return -EIO;
3045 }
3046
3047 /*
3048  * when reads are done, we need to check csums to verify the data is correct
3049  * if there's a match, we allow the bio to finish.  If not, the code in
3050  * extent_io.c will try to find good copies for us.
3051  */
3052 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3053                                       u64 phy_offset, struct page *page,
3054                                       u64 start, u64 end, int mirror)
3055 {
3056         size_t offset = start - page_offset(page);
3057         struct inode *inode = page->mapping->host;
3058         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3059         struct btrfs_root *root = BTRFS_I(inode)->root;
3060
3061         if (PageChecked(page)) {
3062                 ClearPageChecked(page);
3063                 return 0;
3064         }
3065
3066         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3067                 return 0;
3068
3069         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3070             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3071                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3072                                   GFP_NOFS);
3073                 return 0;
3074         }
3075
3076         phy_offset >>= inode->i_sb->s_blocksize_bits;
3077         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3078                                       start, (size_t)(end - start + 1));
3079 }
3080
3081 struct delayed_iput {
3082         struct list_head list;
3083         struct inode *inode;
3084 };
3085
3086 /* JDM: If this is fs-wide, why can't we add a pointer to
3087  * btrfs_inode instead and avoid the allocation? */
3088 void btrfs_add_delayed_iput(struct inode *inode)
3089 {
3090         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3091         struct delayed_iput *delayed;
3092
3093         if (atomic_add_unless(&inode->i_count, -1, 1))
3094                 return;
3095
3096         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3097         delayed->inode = inode;
3098
3099         spin_lock(&fs_info->delayed_iput_lock);
3100         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3101         spin_unlock(&fs_info->delayed_iput_lock);
3102 }
3103
3104 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3105 {
3106         LIST_HEAD(list);
3107         struct btrfs_fs_info *fs_info = root->fs_info;
3108         struct delayed_iput *delayed;
3109         int empty;
3110
3111         spin_lock(&fs_info->delayed_iput_lock);
3112         empty = list_empty(&fs_info->delayed_iputs);
3113         spin_unlock(&fs_info->delayed_iput_lock);
3114         if (empty)
3115                 return;
3116
3117         down_read(&fs_info->delayed_iput_sem);
3118
3119         spin_lock(&fs_info->delayed_iput_lock);
3120         list_splice_init(&fs_info->delayed_iputs, &list);
3121         spin_unlock(&fs_info->delayed_iput_lock);
3122
3123         while (!list_empty(&list)) {
3124                 delayed = list_entry(list.next, struct delayed_iput, list);
3125                 list_del(&delayed->list);
3126                 iput(delayed->inode);
3127                 kfree(delayed);
3128         }
3129
3130         up_read(&root->fs_info->delayed_iput_sem);
3131 }
3132
3133 /*
3134  * This is called in transaction commit time. If there are no orphan
3135  * files in the subvolume, it removes orphan item and frees block_rsv
3136  * structure.
3137  */
3138 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3139                               struct btrfs_root *root)
3140 {
3141         struct btrfs_block_rsv *block_rsv;
3142         int ret;
3143
3144         if (atomic_read(&root->orphan_inodes) ||
3145             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3146                 return;
3147
3148         spin_lock(&root->orphan_lock);
3149         if (atomic_read(&root->orphan_inodes)) {
3150                 spin_unlock(&root->orphan_lock);
3151                 return;
3152         }
3153
3154         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3155                 spin_unlock(&root->orphan_lock);
3156                 return;
3157         }
3158
3159         block_rsv = root->orphan_block_rsv;
3160         root->orphan_block_rsv = NULL;
3161         spin_unlock(&root->orphan_lock);
3162
3163         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3164             btrfs_root_refs(&root->root_item) > 0) {
3165                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3166                                             root->root_key.objectid);
3167                 if (ret)
3168                         btrfs_abort_transaction(trans, root, ret);
3169                 else
3170                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3171                                   &root->state);
3172         }
3173
3174         if (block_rsv) {
3175                 WARN_ON(block_rsv->size > 0);
3176                 btrfs_free_block_rsv(root, block_rsv);
3177         }
3178 }
3179
3180 /*
3181  * This creates an orphan entry for the given inode in case something goes
3182  * wrong in the middle of an unlink/truncate.
3183  *
3184  * NOTE: caller of this function should reserve 5 units of metadata for
3185  *       this function.
3186  */
3187 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3188 {
3189         struct btrfs_root *root = BTRFS_I(inode)->root;
3190         struct btrfs_block_rsv *block_rsv = NULL;
3191         int reserve = 0;
3192         int insert = 0;
3193         int ret;
3194
3195         if (!root->orphan_block_rsv) {
3196                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3197                 if (!block_rsv)
3198                         return -ENOMEM;
3199         }
3200
3201         spin_lock(&root->orphan_lock);
3202         if (!root->orphan_block_rsv) {
3203                 root->orphan_block_rsv = block_rsv;
3204         } else if (block_rsv) {
3205                 btrfs_free_block_rsv(root, block_rsv);
3206                 block_rsv = NULL;
3207         }
3208
3209         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3210                               &BTRFS_I(inode)->runtime_flags)) {
3211 #if 0
3212                 /*
3213                  * For proper ENOSPC handling, we should do orphan
3214                  * cleanup when mounting. But this introduces backward
3215                  * compatibility issue.
3216                  */
3217                 if (!xchg(&root->orphan_item_inserted, 1))
3218                         insert = 2;
3219                 else
3220                         insert = 1;
3221 #endif
3222                 insert = 1;
3223                 atomic_inc(&root->orphan_inodes);
3224         }
3225
3226         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3227                               &BTRFS_I(inode)->runtime_flags))
3228                 reserve = 1;
3229         spin_unlock(&root->orphan_lock);
3230
3231         /* grab metadata reservation from transaction handle */
3232         if (reserve) {
3233                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3234                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3235         }
3236
3237         /* insert an orphan item to track this unlinked/truncated file */
3238         if (insert >= 1) {
3239                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3240                 if (ret) {
3241                         atomic_dec(&root->orphan_inodes);
3242                         if (reserve) {
3243                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3244                                           &BTRFS_I(inode)->runtime_flags);
3245                                 btrfs_orphan_release_metadata(inode);
3246                         }
3247                         if (ret != -EEXIST) {
3248                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3249                                           &BTRFS_I(inode)->runtime_flags);
3250                                 btrfs_abort_transaction(trans, root, ret);
3251                                 return ret;
3252                         }
3253                 }
3254                 ret = 0;
3255         }
3256
3257         /* insert an orphan item to track subvolume contains orphan files */
3258         if (insert >= 2) {
3259                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3260                                                root->root_key.objectid);