root/fs/btrfs/tree-log.c

DEFINITIONS

1. start_log_trans
2. join_running_log_trans
3. btrfs_pin_log_trans
4. btrfs_end_log_trans
5. btrfs_write_tree_block
6. btrfs_wait_tree_block_writeback
7. process_one_buffer
8. overwrite_item
9. read_one_inode
10. replay_one_extent
11. drop_one_dir_item
12. inode_in_dir
13. backref_in_log
14. __add_inode_ref
15. extref_get_fields
16. ref_get_fields
17. unlink_old_inode_refs
18. btrfs_inode_ref_exists
19. add_link
20. add_inode_ref
21. insert_orphan_item
22. count_inode_extrefs
23. count_inode_refs
24. fixup_inode_link_count
25. fixup_inode_link_counts
26. link_to_fixup_dir
27. insert_one_name
28. name_in_log_ref
29. replay_one_name
30. replay_one_dir_item
31. find_dir_range
32. check_item_in_log
33. replay_xattr_deletes
34. replay_dir_deletes
35. replay_one_buffer
36. walk_down_log_tree
37. walk_up_log_tree
38. walk_log_tree
39. update_log_root
40. wait_log_commit
41. wait_for_writer
42. btrfs_remove_log_ctx
43. btrfs_remove_all_log_ctxs
44. btrfs_sync_log
45. free_log_tree
46. btrfs_free_log
47. btrfs_free_log_root_tree
48. inode_logged
49. btrfs_del_dir_entries_in_log
50. btrfs_del_inode_ref_in_log
51. insert_dir_log_key
52. log_dir_items
53. log_directory_changes
54. drop_objectid_items
55. fill_inode_item
56. log_inode_item
57. log_csums
58. copy_items
59. extent_cmp
60. log_extent_csums
61. log_one_extent
62. btrfs_log_prealloc_extents
63. btrfs_log_changed_extents
64. logged_inode_size
65. btrfs_log_all_xattrs
66. btrfs_log_holes
67. btrfs_check_ref_name_override
68. log_conflicting_inodes
69. btrfs_log_inode
70. btrfs_must_commit_transaction
71. check_parent_dirs_for_sync
72. log_new_dir_dentries
73. btrfs_log_all_parents
74. log_new_ancestors
75. log_new_ancestors_fast
76. log_all_new_ancestors
77. btrfs_log_inode_parent
78. btrfs_log_dentry_safe
79. btrfs_recover_log_trees
80. btrfs_record_unlink_dir
81. btrfs_record_snapshot_destroy
82. btrfs_log_new_name

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2008 Oracle.  All rights reserved.
   4  */
   5 
   6 #include <linux/sched.h>
   7 #include <linux/slab.h>
   8 #include <linux/blkdev.h>
   9 #include <linux/list_sort.h>
  10 #include <linux/iversion.h>
  11 #include "misc.h"
  12 #include "ctree.h"
  13 #include "tree-log.h"
  14 #include "disk-io.h"
  15 #include "locking.h"
  16 #include "print-tree.h"
  17 #include "backref.h"
  18 #include "compression.h"
  19 #include "qgroup.h"
  20 #include "inode-map.h"
  21 
  22 /* magic values for the inode_only field in btrfs_log_inode:
  23  *
  24  * LOG_INODE_ALL means to log everything
  25  * LOG_INODE_EXISTS means to log just enough to recreate the inode
  26  * during log replay
  27  */
  28 enum {
  29         LOG_INODE_ALL,
  30         LOG_INODE_EXISTS,
  31         LOG_OTHER_INODE,
  32         LOG_OTHER_INODE_ALL,
  33 };
  34 
  35 /*
  36  * directory trouble cases
  37  *
  38  * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
  39  * log, we must force a full commit before doing an fsync of the directory
  40  * where the unlink was done.
  41  * ---> record transid of last unlink/rename per directory
  42  *
  43  * mkdir foo/some_dir
  44  * normal commit
  45  * rename foo/some_dir foo2/some_dir
  46  * mkdir foo/some_dir
  47  * fsync foo/some_dir/some_file
  48  *
  49  * The fsync above will unlink the original some_dir without recording
  50  * it in its new location (foo2).  After a crash, some_dir will be gone
  51  * unless the fsync of some_file forces a full commit
  52  *
  53  * 2) we must log any new names for any file or dir that is in the fsync
  54  * log. ---> check inode while renaming/linking.
  55  *
  56  * 2a) we must log any new names for any file or dir during rename
  57  * when the directory they are being removed from was logged.
  58  * ---> check inode and old parent dir during rename
  59  *
  60  *  2a is actually the more important variant.  With the extra logging
  61  *  a crash might unlink the old name without recreating the new one
  62  *
  63  * 3) after a crash, we must go through any directories with a link count
  64  * of zero and redo the rm -rf
  65  *
  66  * mkdir f1/foo
  67  * normal commit
  68  * rm -rf f1/foo
  69  * fsync(f1)
  70  *
  71  * The directory f1 was fully removed from the FS, but fsync was never
  72  * called on f1, only its parent dir.  After a crash the rm -rf must
  73  * be replayed.  This must be able to recurse down the entire
  74  * directory tree.  The inode link count fixup code takes care of the
  75  * ugly details.
  76  */
  77 
  78 /*
  79  * stages for the tree walking.  The first
  80  * stage (0) is to only pin down the blocks we find
  81  * the second stage (1) is to make sure that all the inodes
  82  * we find in the log are created in the subvolume.
  83  *
  84  * The last stage is to deal with directories and links and extents
  85  * and all the other fun semantics
  86  */
  87 enum {
  88         LOG_WALK_PIN_ONLY,
  89         LOG_WALK_REPLAY_INODES,
  90         LOG_WALK_REPLAY_DIR_INDEX,
  91         LOG_WALK_REPLAY_ALL,
  92 };
  93 
  94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
  95                            struct btrfs_root *root, struct btrfs_inode *inode,
  96                            int inode_only,
  97                            const loff_t start,
  98                            const loff_t end,
  99                            struct btrfs_log_ctx *ctx);
 100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
 101                              struct btrfs_root *root,
 102                              struct btrfs_path *path, u64 objectid);
 103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
 104                                        struct btrfs_root *root,
 105                                        struct btrfs_root *log,
 106                                        struct btrfs_path *path,
 107                                        u64 dirid, int del_all);
 108 
 109 /*
 110  * tree logging is a special write ahead log used to make sure that
 111  * fsyncs and O_SYNCs can happen without doing full tree commits.
 112  *
 113  * Full tree commits are expensive because they require commonly
 114  * modified blocks to be recowed, creating many dirty pages in the
 115  * extent tree an 4x-6x higher write load than ext3.
 116  *
 117  * Instead of doing a tree commit on every fsync, we use the
 118  * key ranges and transaction ids to find items for a given file or directory
 119  * that have changed in this transaction.  Those items are copied into
 120  * a special tree (one per subvolume root), that tree is written to disk
 121  * and then the fsync is considered complete.
 122  *
 123  * After a crash, items are copied out of the log-tree back into the
 124  * subvolume tree.  Any file data extents found are recorded in the extent
 125  * allocation tree, and the log-tree freed.
 126  *
 127  * The log tree is read three times, once to pin down all the extents it is
 128  * using in ram and once, once to create all the inodes logged in the tree
 129  * and once to do all the other items.
 130  */
 131 
 132 /*
 133  * start a sub transaction and setup the log tree
 134  * this increments the log tree writer count to make the people
 135  * syncing the tree wait for us to finish
 136  */
 137 static int start_log_trans(struct btrfs_trans_handle *trans,
 138                            struct btrfs_root *root,
 139                            struct btrfs_log_ctx *ctx)
 140 {
 141         struct btrfs_fs_info *fs_info = root->fs_info;
 142         int ret = 0;
 143 
 144         mutex_lock(&root->log_mutex);
 145 
 146         if (root->log_root) {
 147                 if (btrfs_need_log_full_commit(trans)) {
 148                         ret = -EAGAIN;
 149                         goto out;
 150                 }
 151 
 152                 if (!root->log_start_pid) {
 153                         clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
 154                         root->log_start_pid = current->pid;
 155                 } else if (root->log_start_pid != current->pid) {
 156                         set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
 157                 }
 158         } else {
 159                 mutex_lock(&fs_info->tree_log_mutex);
 160                 if (!fs_info->log_root_tree)
 161                         ret = btrfs_init_log_root_tree(trans, fs_info);
 162                 mutex_unlock(&fs_info->tree_log_mutex);
 163                 if (ret)
 164                         goto out;
 165 
 166                 ret = btrfs_add_log_tree(trans, root);
 167                 if (ret)
 168                         goto out;
 169 
 170                 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
 171                 root->log_start_pid = current->pid;
 172         }
 173 
 174         atomic_inc(&root->log_batch);
 175         atomic_inc(&root->log_writers);
 176         if (ctx) {
 177                 int index = root->log_transid % 2;
 178                 list_add_tail(&ctx->list, &root->log_ctxs[index]);
 179                 ctx->log_transid = root->log_transid;
 180         }
 181 
 182 out:
 183         mutex_unlock(&root->log_mutex);
 184         return ret;
 185 }
 186 
 187 /*
 188  * returns 0 if there was a log transaction running and we were able
 189  * to join, or returns -ENOENT if there were not transactions
 190  * in progress
 191  */
 192 static int join_running_log_trans(struct btrfs_root *root)
 193 {
 194         int ret = -ENOENT;
 195 
 196         mutex_lock(&root->log_mutex);
 197         if (root->log_root) {
 198                 ret = 0;
 199                 atomic_inc(&root->log_writers);
 200         }
 201         mutex_unlock(&root->log_mutex);
 202         return ret;
 203 }
 204 
 205 /*
 206  * This either makes the current running log transaction wait
 207  * until you call btrfs_end_log_trans() or it makes any future
 208  * log transactions wait until you call btrfs_end_log_trans()
 209  */
 210 void btrfs_pin_log_trans(struct btrfs_root *root)
 211 {
 212         mutex_lock(&root->log_mutex);
 213         atomic_inc(&root->log_writers);
 214         mutex_unlock(&root->log_mutex);
 215 }
 216 
 217 /*
 218  * indicate we're done making changes to the log tree
 219  * and wake up anyone waiting to do a sync
 220  */
 221 void btrfs_end_log_trans(struct btrfs_root *root)
 222 {
 223         if (atomic_dec_and_test(&root->log_writers)) {
 224                 /* atomic_dec_and_test implies a barrier */
 225                 cond_wake_up_nomb(&root->log_writer_wait);
 226         }
 227 }
 228 
 229 static int btrfs_write_tree_block(struct extent_buffer *buf)
 230 {
 231         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
 232                                         buf->start + buf->len - 1);
 233 }
 234 
 235 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 236 {
 237         filemap_fdatawait_range(buf->pages[0]->mapping,
 238                                 buf->start, buf->start + buf->len - 1);
 239 }
 240 
 241 /*
 242  * the walk control struct is used to pass state down the chain when
 243  * processing the log tree.  The stage field tells us which part
 244  * of the log tree processing we are currently doing.  The others
 245  * are state fields used for that specific part
 246  */
 247 struct walk_control {
 248         /* should we free the extent on disk when done?  This is used
 249          * at transaction commit time while freeing a log tree
 250          */
 251         int free;
 252 
 253         /* should we write out the extent buffer?  This is used
 254          * while flushing the log tree to disk during a sync
 255          */
 256         int write;
 257 
 258         /* should we wait for the extent buffer io to finish?  Also used
 259          * while flushing the log tree to disk for a sync
 260          */
 261         int wait;
 262 
 263         /* pin only walk, we record which extents on disk belong to the
 264          * log trees
 265          */
 266         int pin;
 267 
 268         /* what stage of the replay code we're currently in */
 269         int stage;
 270 
 271         /*
 272          * Ignore any items from the inode currently being processed. Needs
 273          * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
 274          * the LOG_WALK_REPLAY_INODES stage.
 275          */
 276         bool ignore_cur_inode;
 277 
 278         /* the root we are currently replaying */
 279         struct btrfs_root *replay_dest;
 280 
 281         /* the trans handle for the current replay */
 282         struct btrfs_trans_handle *trans;
 283 
 284         /* the function that gets used to process blocks we find in the
 285          * tree.  Note the extent_buffer might not be up to date when it is
 286          * passed in, and it must be checked or read if you need the data
 287          * inside it
 288          */
 289         int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
 290                             struct walk_control *wc, u64 gen, int level);
 291 };
 292 
 293 /*
 294  * process_func used to pin down extents, write them or wait on them
 295  */
 296 static int process_one_buffer(struct btrfs_root *log,
 297                               struct extent_buffer *eb,
 298                               struct walk_control *wc, u64 gen, int level)
 299 {
 300         struct btrfs_fs_info *fs_info = log->fs_info;
 301         int ret = 0;
 302 
 303         /*
 304          * If this fs is mixed then we need to be able to process the leaves to
 305          * pin down any logged extents, so we have to read the block.
 306          */
 307         if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
 308                 ret = btrfs_read_buffer(eb, gen, level, NULL);
 309                 if (ret)
 310                         return ret;
 311         }
 312 
 313         if (wc->pin)
 314                 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
 315                                                       eb->len);
 316 
 317         if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
 318                 if (wc->pin && btrfs_header_level(eb) == 0)
 319                         ret = btrfs_exclude_logged_extents(eb);
 320                 if (wc->write)
 321                         btrfs_write_tree_block(eb);
 322                 if (wc->wait)
 323                         btrfs_wait_tree_block_writeback(eb);
 324         }
 325         return ret;
 326 }
 327 
 328 /*
 329  * Item overwrite used by replay and tree logging.  eb, slot and key all refer
 330  * to the src data we are copying out.
 331  *
 332  * root is the tree we are copying into, and path is a scratch
 333  * path for use in this function (it should be released on entry and
 334  * will be released on exit).
 335  *
 336  * If the key is already in the destination tree the existing item is
 337  * overwritten.  If the existing item isn't big enough, it is extended.
 338  * If it is too large, it is truncated.
 339  *
 340  * If the key isn't in the destination yet, a new item is inserted.
 341  */
 342 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
 343                                    struct btrfs_root *root,
 344                                    struct btrfs_path *path,
 345                                    struct extent_buffer *eb, int slot,
 346                                    struct btrfs_key *key)
 347 {
 348         int ret;
 349         u32 item_size;
 350         u64 saved_i_size = 0;
 351         int save_old_i_size = 0;
 352         unsigned long src_ptr;
 353         unsigned long dst_ptr;
 354         int overwrite_root = 0;
 355         bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
 356 
 357         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
 358                 overwrite_root = 1;
 359 
 360         item_size = btrfs_item_size_nr(eb, slot);
 361         src_ptr = btrfs_item_ptr_offset(eb, slot);
 362 
 363         /* look for the key in the destination tree */
 364         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
 365         if (ret < 0)
 366                 return ret;
 367 
 368         if (ret == 0) {
 369                 char *src_copy;
 370                 char *dst_copy;
 371                 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
 372                                                   path->slots[0]);
 373                 if (dst_size != item_size)
 374                         goto insert;
 375 
 376                 if (item_size == 0) {
 377                         btrfs_release_path(path);
 378                         return 0;
 379                 }
 380                 dst_copy = kmalloc(item_size, GFP_NOFS);
 381                 src_copy = kmalloc(item_size, GFP_NOFS);
 382                 if (!dst_copy || !src_copy) {
 383                         btrfs_release_path(path);
 384                         kfree(dst_copy);
 385                         kfree(src_copy);
 386                         return -ENOMEM;
 387                 }
 388 
 389                 read_extent_buffer(eb, src_copy, src_ptr, item_size);
 390 
 391                 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 392                 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
 393                                    item_size);
 394                 ret = memcmp(dst_copy, src_copy, item_size);
 395 
 396                 kfree(dst_copy);
 397                 kfree(src_copy);
 398                 /*
 399                  * they have the same contents, just return, this saves
 400                  * us from cowing blocks in the destination tree and doing
 401                  * extra writes that may not have been done by a previous
 402                  * sync
 403                  */
 404                 if (ret == 0) {
 405                         btrfs_release_path(path);
 406                         return 0;
 407                 }
 408 
 409                 /*
 410                  * We need to load the old nbytes into the inode so when we
 411                  * replay the extents we've logged we get the right nbytes.
 412                  */
 413                 if (inode_item) {
 414                         struct btrfs_inode_item *item;
 415                         u64 nbytes;
 416                         u32 mode;
 417 
 418                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 419                                               struct btrfs_inode_item);
 420                         nbytes = btrfs_inode_nbytes(path->nodes[0], item);
 421                         item = btrfs_item_ptr(eb, slot,
 422                                               struct btrfs_inode_item);
 423                         btrfs_set_inode_nbytes(eb, item, nbytes);
 424 
 425                         /*
 426                          * If this is a directory we need to reset the i_size to
 427                          * 0 so that we can set it up properly when replaying
 428                          * the rest of the items in this log.
 429                          */
 430                         mode = btrfs_inode_mode(eb, item);
 431                         if (S_ISDIR(mode))
 432                                 btrfs_set_inode_size(eb, item, 0);
 433                 }
 434         } else if (inode_item) {
 435                 struct btrfs_inode_item *item;
 436                 u32 mode;
 437 
 438                 /*
 439                  * New inode, set nbytes to 0 so that the nbytes comes out
 440                  * properly when we replay the extents.
 441                  */
 442                 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
 443                 btrfs_set_inode_nbytes(eb, item, 0);
 444 
 445                 /*
 446                  * If this is a directory we need to reset the i_size to 0 so
 447                  * that we can set it up properly when replaying the rest of
 448                  * the items in this log.
 449                  */
 450                 mode = btrfs_inode_mode(eb, item);
 451                 if (S_ISDIR(mode))
 452                         btrfs_set_inode_size(eb, item, 0);
 453         }
 454 insert:
 455         btrfs_release_path(path);
 456         /* try to insert the key into the destination tree */
 457         path->skip_release_on_error = 1;
 458         ret = btrfs_insert_empty_item(trans, root, path,
 459                                       key, item_size);
 460         path->skip_release_on_error = 0;
 461 
 462         /* make sure any existing item is the correct size */
 463         if (ret == -EEXIST || ret == -EOVERFLOW) {
 464                 u32 found_size;
 465                 found_size = btrfs_item_size_nr(path->nodes[0],
 466                                                 path->slots[0]);
 467                 if (found_size > item_size)
 468                         btrfs_truncate_item(path, item_size, 1);
 469                 else if (found_size < item_size)
 470                         btrfs_extend_item(path, item_size - found_size);
 471         } else if (ret) {
 472                 return ret;
 473         }
 474         dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
 475                                         path->slots[0]);
 476 
 477         /* don't overwrite an existing inode if the generation number
 478          * was logged as zero.  This is done when the tree logging code
 479          * is just logging an inode to make sure it exists after recovery.
 480          *
 481          * Also, don't overwrite i_size on directories during replay.
 482          * log replay inserts and removes directory items based on the
 483          * state of the tree found in the subvolume, and i_size is modified
 484          * as it goes
 485          */
 486         if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
 487                 struct btrfs_inode_item *src_item;
 488                 struct btrfs_inode_item *dst_item;
 489 
 490                 src_item = (struct btrfs_inode_item *)src_ptr;
 491                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 492 
 493                 if (btrfs_inode_generation(eb, src_item) == 0) {
 494                         struct extent_buffer *dst_eb = path->nodes[0];
 495                         const u64 ino_size = btrfs_inode_size(eb, src_item);
 496 
 497                         /*
 498                          * For regular files an ino_size == 0 is used only when
 499                          * logging that an inode exists, as part of a directory
 500                          * fsync, and the inode wasn't fsynced before. In this
 501                          * case don't set the size of the inode in the fs/subvol
 502                          * tree, otherwise we would be throwing valid data away.
 503                          */
 504                         if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
 505                             S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
 506                             ino_size != 0) {
 507                                 struct btrfs_map_token token;
 508 
 509                                 btrfs_init_map_token(&token, dst_eb);
 510                                 btrfs_set_token_inode_size(dst_eb, dst_item,
 511                                                            ino_size, &token);
 512                         }
 513                         goto no_copy;
 514                 }
 515 
 516                 if (overwrite_root &&
 517                     S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
 518                     S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
 519                         save_old_i_size = 1;
 520                         saved_i_size = btrfs_inode_size(path->nodes[0],
 521                                                         dst_item);
 522                 }
 523         }
 524 
 525         copy_extent_buffer(path->nodes[0], eb, dst_ptr,
 526                            src_ptr, item_size);
 527 
 528         if (save_old_i_size) {
 529                 struct btrfs_inode_item *dst_item;
 530                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 531                 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
 532         }
 533 
 534         /* make sure the generation is filled in */
 535         if (key->type == BTRFS_INODE_ITEM_KEY) {
 536                 struct btrfs_inode_item *dst_item;
 537                 dst_item = (struct btrfs_inode_item *)dst_ptr;
 538                 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
 539                         btrfs_set_inode_generation(path->nodes[0], dst_item,
 540                                                    trans->transid);
 541                 }
 542         }
 543 no_copy:
 544         btrfs_mark_buffer_dirty(path->nodes[0]);
 545         btrfs_release_path(path);
 546         return 0;
 547 }
 548 
 549 /*
 550  * simple helper to read an inode off the disk from a given root
 551  * This can only be called for subvolume roots and not for the log
 552  */
 553 static noinline struct inode *read_one_inode(struct btrfs_root *root,
 554                                              u64 objectid)
 555 {
 556         struct btrfs_key key;
 557         struct inode *inode;
 558 
 559         key.objectid = objectid;
 560         key.type = BTRFS_INODE_ITEM_KEY;
 561         key.offset = 0;
 562         inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
 563         if (IS_ERR(inode))
 564                 inode = NULL;
 565         return inode;
 566 }
 567 
 568 /* replays a single extent in 'eb' at 'slot' with 'key' into the
 569  * subvolume 'root'.  path is released on entry and should be released
 570  * on exit.
 571  *
 572  * extents in the log tree have not been allocated out of the extent
 573  * tree yet.  So, this completes the allocation, taking a reference
 574  * as required if the extent already exists or creating a new extent
 575  * if it isn't in the extent allocation tree yet.
 576  *
 577  * The extent is inserted into the file, dropping any existing extents
 578  * from the file that overlap the new one.
 579  */
 580 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 581                                       struct btrfs_root *root,
 582                                       struct btrfs_path *path,
 583                                       struct extent_buffer *eb, int slot,
 584                                       struct btrfs_key *key)
 585 {
 586         struct btrfs_fs_info *fs_info = root->fs_info;
 587         int found_type;
 588         u64 extent_end;
 589         u64 start = key->offset;
 590         u64 nbytes = 0;
 591         struct btrfs_file_extent_item *item;
 592         struct inode *inode = NULL;
 593         unsigned long size;
 594         int ret = 0;
 595 
 596         item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 597         found_type = btrfs_file_extent_type(eb, item);
 598 
 599         if (found_type == BTRFS_FILE_EXTENT_REG ||
 600             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 601                 nbytes = btrfs_file_extent_num_bytes(eb, item);
 602                 extent_end = start + nbytes;
 603 
 604                 /*
 605                  * We don't add to the inodes nbytes if we are prealloc or a
 606                  * hole.
 607                  */
 608                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
 609                         nbytes = 0;
 610         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 611                 size = btrfs_file_extent_ram_bytes(eb, item);
 612                 nbytes = btrfs_file_extent_ram_bytes(eb, item);
 613                 extent_end = ALIGN(start + size,
 614                                    fs_info->sectorsize);
 615         } else {
 616                 ret = 0;
 617                 goto out;
 618         }
 619 
 620         inode = read_one_inode(root, key->objectid);
 621         if (!inode) {
 622                 ret = -EIO;
 623                 goto out;
 624         }
 625 
 626         /*
 627          * first check to see if we already have this extent in the
 628          * file.  This must be done before the btrfs_drop_extents run
 629          * so we don't try to drop this extent.
 630          */
 631         ret = btrfs_lookup_file_extent(trans, root, path,
 632                         btrfs_ino(BTRFS_I(inode)), start, 0);
 633 
 634         if (ret == 0 &&
 635             (found_type == BTRFS_FILE_EXTENT_REG ||
 636              found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
 637                 struct btrfs_file_extent_item cmp1;
 638                 struct btrfs_file_extent_item cmp2;
 639                 struct btrfs_file_extent_item *existing;
 640                 struct extent_buffer *leaf;
 641 
 642                 leaf = path->nodes[0];
 643                 existing = btrfs_item_ptr(leaf, path->slots[0],
 644                                           struct btrfs_file_extent_item);
 645 
 646                 read_extent_buffer(eb, &cmp1, (unsigned long)item,
 647                                    sizeof(cmp1));
 648                 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
 649                                    sizeof(cmp2));
 650 
 651                 /*
 652                  * we already have a pointer to this exact extent,
 653                  * we don't have to do anything
 654                  */
 655                 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
 656                         btrfs_release_path(path);
 657                         goto out;
 658                 }
 659         }
 660         btrfs_release_path(path);
 661 
 662         /* drop any overlapping extents */
 663         ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
 664         if (ret)
 665                 goto out;
 666 
 667         if (found_type == BTRFS_FILE_EXTENT_REG ||
 668             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
 669                 u64 offset;
 670                 unsigned long dest_offset;
 671                 struct btrfs_key ins;
 672 
 673                 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
 674                     btrfs_fs_incompat(fs_info, NO_HOLES))
 675                         goto update_inode;
 676 
 677                 ret = btrfs_insert_empty_item(trans, root, path, key,
 678                                               sizeof(*item));
 679                 if (ret)
 680                         goto out;
 681                 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
 682                                                     path->slots[0]);
 683                 copy_extent_buffer(path->nodes[0], eb, dest_offset,
 684                                 (unsigned long)item,  sizeof(*item));
 685 
 686                 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
 687                 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
 688                 ins.type = BTRFS_EXTENT_ITEM_KEY;
 689                 offset = key->offset - btrfs_file_extent_offset(eb, item);
 690 
 691                 /*
 692                  * Manually record dirty extent, as here we did a shallow
 693                  * file extent item copy and skip normal backref update,
 694                  * but modifying extent tree all by ourselves.
 695                  * So need to manually record dirty extent for qgroup,
 696                  * as the owner of the file extent changed from log tree
 697                  * (doesn't affect qgroup) to fs/file tree(affects qgroup)
 698                  */
 699                 ret = btrfs_qgroup_trace_extent(trans,
 700                                 btrfs_file_extent_disk_bytenr(eb, item),
 701                                 btrfs_file_extent_disk_num_bytes(eb, item),
 702                                 GFP_NOFS);
 703                 if (ret < 0)
 704                         goto out;
 705 
 706                 if (ins.objectid > 0) {
 707                         struct btrfs_ref ref = { 0 };
 708                         u64 csum_start;
 709                         u64 csum_end;
 710                         LIST_HEAD(ordered_sums);
 711 
 712                         /*
 713                          * is this extent already allocated in the extent
 714                          * allocation tree?  If so, just add a reference
 715                          */
 716                         ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
 717                                                 ins.offset);
 718                         if (ret == 0) {
 719                                 btrfs_init_generic_ref(&ref,
 720                                                 BTRFS_ADD_DELAYED_REF,
 721                                                 ins.objectid, ins.offset, 0);
 722                                 btrfs_init_data_ref(&ref,
 723                                                 root->root_key.objectid,
 724                                                 key->objectid, offset);
 725                                 ret = btrfs_inc_extent_ref(trans, &ref);
 726                                 if (ret)
 727                                         goto out;
 728                         } else {
 729                                 /*
 730                                  * insert the extent pointer in the extent
 731                                  * allocation tree
 732                                  */
 733                                 ret = btrfs_alloc_logged_file_extent(trans,
 734                                                 root->root_key.objectid,
 735                                                 key->objectid, offset, &ins);
 736                                 if (ret)
 737                                         goto out;
 738                         }
 739                         btrfs_release_path(path);
 740 
 741                         if (btrfs_file_extent_compression(eb, item)) {
 742                                 csum_start = ins.objectid;
 743                                 csum_end = csum_start + ins.offset;
 744                         } else {
 745                                 csum_start = ins.objectid +
 746                                         btrfs_file_extent_offset(eb, item);
 747                                 csum_end = csum_start +
 748                                         btrfs_file_extent_num_bytes(eb, item);
 749                         }
 750 
 751                         ret = btrfs_lookup_csums_range(root->log_root,
 752                                                 csum_start, csum_end - 1,
 753                                                 &ordered_sums, 0);
 754                         if (ret)
 755                                 goto out;
 756                         /*
 757                          * Now delete all existing cums in the csum root that
 758                          * cover our range. We do this because we can have an
 759                          * extent that is completely referenced by one file
 760                          * extent item and partially referenced by another
 761                          * file extent item (like after using the clone or
 762                          * extent_same ioctls). In this case if we end up doing
 763                          * the replay of the one that partially references the
 764                          * extent first, and we do not do the csum deletion
 765                          * below, we can get 2 csum items in the csum tree that
 766                          * overlap each other. For example, imagine our log has
 767                          * the two following file extent items:
 768                          *
 769                          * key (257 EXTENT_DATA 409600)
 770                          *     extent data disk byte 12845056 nr 102400
 771                          *     extent data offset 20480 nr 20480 ram 102400
 772                          *
 773                          * key (257 EXTENT_DATA 819200)
 774                          *     extent data disk byte 12845056 nr 102400
 775                          *     extent data offset 0 nr 102400 ram 102400
 776                          *
 777                          * Where the second one fully references the 100K extent
 778                          * that starts at disk byte 12845056, and the log tree
 779                          * has a single csum item that covers the entire range
 780                          * of the extent:
 781                          *
 782                          * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
 783                          *
 784                          * After the first file extent item is replayed, the
 785                          * csum tree gets the following csum item:
 786                          *
 787                          * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
 788                          *
 789                          * Which covers the 20K sub-range starting at offset 20K
 790                          * of our extent. Now when we replay the second file
 791                          * extent item, if we do not delete existing csum items
 792                          * that cover any of its blocks, we end up getting two
 793                          * csum items in our csum tree that overlap each other:
 794                          *
 795                          * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
 796                          * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
 797                          *
 798                          * Which is a problem, because after this anyone trying
 799                          * to lookup up for the checksum of any block of our
 800                          * extent starting at an offset of 40K or higher, will
 801                          * end up looking at the second csum item only, which
 802                          * does not contain the checksum for any block starting
 803                          * at offset 40K or higher of our extent.
 804                          */
 805                         while (!list_empty(&ordered_sums)) {
 806                                 struct btrfs_ordered_sum *sums;
 807                                 sums = list_entry(ordered_sums.next,
 808                                                 struct btrfs_ordered_sum,
 809                                                 list);
 810                                 if (!ret)
 811                                         ret = btrfs_del_csums(trans,
 812                                                               fs_info->csum_root,
 813                                                               sums->bytenr,
 814                                                               sums->len);
 815                                 if (!ret)
 816                                         ret = btrfs_csum_file_blocks(trans,
 817                                                 fs_info->csum_root, sums);
 818                                 list_del(&sums->list);
 819                                 kfree(sums);
 820                         }
 821                         if (ret)
 822                                 goto out;
 823                 } else {
 824                         btrfs_release_path(path);
 825                 }
 826         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
 827                 /* inline extents are easy, we just overwrite them */
 828                 ret = overwrite_item(trans, root, path, eb, slot, key);
 829                 if (ret)
 830                         goto out;
 831         }
 832 
 833         inode_add_bytes(inode, nbytes);
 834 update_inode:
 835         ret = btrfs_update_inode(trans, root, inode);
 836 out:
 837         if (inode)
 838                 iput(inode);
 839         return ret;
 840 }
 841 
 842 /*
 843  * when cleaning up conflicts between the directory names in the
 844  * subvolume, directory names in the log and directory names in the
 845  * inode back references, we may have to unlink inodes from directories.
 846  *
 847  * This is a helper function to do the unlink of a specific directory
 848  * item
 849  */
 850 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
 851                                       struct btrfs_root *root,
 852                                       struct btrfs_path *path,
 853                                       struct btrfs_inode *dir,
 854                                       struct btrfs_dir_item *di)
 855 {
 856         struct inode *inode;
 857         char *name;
 858         int name_len;
 859         struct extent_buffer *leaf;
 860         struct btrfs_key location;
 861         int ret;
 862 
 863         leaf = path->nodes[0];
 864 
 865         btrfs_dir_item_key_to_cpu(leaf, di, &location);
 866         name_len = btrfs_dir_name_len(leaf, di);
 867         name = kmalloc(name_len, GFP_NOFS);
 868         if (!name)
 869                 return -ENOMEM;
 870 
 871         read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
 872         btrfs_release_path(path);
 873 
 874         inode = read_one_inode(root, location.objectid);
 875         if (!inode) {
 876                 ret = -EIO;
 877                 goto out;
 878         }
 879 
 880         ret = link_to_fixup_dir(trans, root, path, location.objectid);
 881         if (ret)
 882                 goto out;
 883 
 884         ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
 885                         name_len);
 886         if (ret)
 887                 goto out;
 888         else
 889                 ret = btrfs_run_delayed_items(trans);
 890 out:
 891         kfree(name);
 892         iput(inode);
 893         return ret;
 894 }
 895 
 896 /*
 897  * helper function to see if a given name and sequence number found
 898  * in an inode back reference are already in a directory and correctly
 899  * point to this inode
 900  */
 901 static noinline int inode_in_dir(struct btrfs_root *root,
 902                                  struct btrfs_path *path,
 903                                  u64 dirid, u64 objectid, u64 index,
 904                                  const char *name, int name_len)
 905 {
 906         struct btrfs_dir_item *di;
 907         struct btrfs_key location;
 908         int match = 0;
 909 
 910         di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
 911                                          index, name, name_len, 0);
 912         if (di && !IS_ERR(di)) {
 913                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 914                 if (location.objectid != objectid)
 915                         goto out;
 916         } else
 917                 goto out;
 918         btrfs_release_path(path);
 919 
 920         di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
 921         if (di && !IS_ERR(di)) {
 922                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
 923                 if (location.objectid != objectid)
 924                         goto out;
 925         } else
 926                 goto out;
 927         match = 1;
 928 out:
 929         btrfs_release_path(path);
 930         return match;
 931 }
 932 
 933 /*
 934  * helper function to check a log tree for a named back reference in
 935  * an inode.  This is used to decide if a back reference that is
 936  * found in the subvolume conflicts with what we find in the log.
 937  *
 938  * inode backreferences may have multiple refs in a single item,
 939  * during replay we process one reference at a time, and we don't
 940  * want to delete valid links to a file from the subvolume if that
 941  * link is also in the log.
 942  */
 943 static noinline int backref_in_log(struct btrfs_root *log,
 944                                    struct btrfs_key *key,
 945                                    u64 ref_objectid,
 946                                    const char *name, int namelen)
 947 {
 948         struct btrfs_path *path;
 949         struct btrfs_inode_ref *ref;
 950         unsigned long ptr;
 951         unsigned long ptr_end;
 952         unsigned long name_ptr;
 953         int found_name_len;
 954         int item_size;
 955         int ret;
 956         int match = 0;
 957 
 958         path = btrfs_alloc_path();
 959         if (!path)
 960                 return -ENOMEM;
 961 
 962         ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
 963         if (ret != 0)
 964                 goto out;
 965 
 966         ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
 967 
 968         if (key->type == BTRFS_INODE_EXTREF_KEY) {
 969                 if (btrfs_find_name_in_ext_backref(path->nodes[0],
 970                                                    path->slots[0],
 971                                                    ref_objectid,
 972                                                    name, namelen))
 973                         match = 1;
 974 
 975                 goto out;
 976         }
 977 
 978         item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
 979         ptr_end = ptr + item_size;
 980         while (ptr < ptr_end) {
 981                 ref = (struct btrfs_inode_ref *)ptr;
 982                 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
 983                 if (found_name_len == namelen) {
 984                         name_ptr = (unsigned long)(ref + 1);
 985                         ret = memcmp_extent_buffer(path->nodes[0], name,
 986                                                    name_ptr, namelen);
 987                         if (ret == 0) {
 988                                 match = 1;
 989                                 goto out;
 990                         }
 991                 }
 992                 ptr = (unsigned long)(ref + 1) + found_name_len;
 993         }
 994 out:
 995         btrfs_free_path(path);
 996         return match;
 997 }
 998 
 999 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1000                                   struct btrfs_root *root,
1001                                   struct btrfs_path *path,
1002                                   struct btrfs_root *log_root,
1003                                   struct btrfs_inode *dir,
1004                                   struct btrfs_inode *inode,
1005                                   u64 inode_objectid, u64 parent_objectid,
1006                                   u64 ref_index, char *name, int namelen,
1007                                   int *search_done)
1008 {
1009         int ret;
1010         char *victim_name;
1011         int victim_name_len;
1012         struct extent_buffer *leaf;
1013         struct btrfs_dir_item *di;
1014         struct btrfs_key search_key;
1015         struct btrfs_inode_extref *extref;
1016 
1017 again:
1018         /* Search old style refs */
1019         search_key.objectid = inode_objectid;
1020         search_key.type = BTRFS_INODE_REF_KEY;
1021         search_key.offset = parent_objectid;
1022         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1023         if (ret == 0) {
1024                 struct btrfs_inode_ref *victim_ref;
1025                 unsigned long ptr;
1026                 unsigned long ptr_end;
1027 
1028                 leaf = path->nodes[0];
1029 
1030                 /* are we trying to overwrite a back ref for the root directory
1031                  * if so, just jump out, we're done
1032                  */
1033                 if (search_key.objectid == search_key.offset)
1034                         return 1;
1035 
1036                 /* check all the names in this back reference to see
1037                  * if they are in the log.  if so, we allow them to stay
1038                  * otherwise they must be unlinked as a conflict
1039                  */
1040                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1041                 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1042                 while (ptr < ptr_end) {
1043                         victim_ref = (struct btrfs_inode_ref *)ptr;
1044                         victim_name_len = btrfs_inode_ref_name_len(leaf,
1045                                                                    victim_ref);
1046                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
1047                         if (!victim_name)
1048                                 return -ENOMEM;
1049 
1050                         read_extent_buffer(leaf, victim_name,
1051                                            (unsigned long)(victim_ref + 1),
1052                                            victim_name_len);
1053 
1054                         if (!backref_in_log(log_root, &search_key,
1055                                             parent_objectid,
1056                                             victim_name,
1057                                             victim_name_len)) {
1058                                 inc_nlink(&inode->vfs_inode);
1059                                 btrfs_release_path(path);
1060 
1061                                 ret = btrfs_unlink_inode(trans, root, dir, inode,
1062                                                 victim_name, victim_name_len);
1063                                 kfree(victim_name);
1064                                 if (ret)
1065                                         return ret;
1066                                 ret = btrfs_run_delayed_items(trans);
1067                                 if (ret)
1068                                         return ret;
1069                                 *search_done = 1;
1070                                 goto again;
1071                         }
1072                         kfree(victim_name);
1073 
1074                         ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1075                 }
1076 
1077                 /*
1078                  * NOTE: we have searched root tree and checked the
1079                  * corresponding ref, it does not need to check again.
1080                  */
1081                 *search_done = 1;
1082         }
1083         btrfs_release_path(path);
1084 
1085         /* Same search but for extended refs */
1086         extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1087                                            inode_objectid, parent_objectid, 0,
1088                                            0);
1089         if (!IS_ERR_OR_NULL(extref)) {
1090                 u32 item_size;
1091                 u32 cur_offset = 0;
1092                 unsigned long base;
1093                 struct inode *victim_parent;
1094 
1095                 leaf = path->nodes[0];
1096 
1097                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1098                 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1099 
1100                 while (cur_offset < item_size) {
1101                         extref = (struct btrfs_inode_extref *)(base + cur_offset);
1102 
1103                         victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1104 
1105                         if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1106                                 goto next;
1107 
1108                         victim_name = kmalloc(victim_name_len, GFP_NOFS);
1109                         if (!victim_name)
1110                                 return -ENOMEM;
1111                         read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1112                                            victim_name_len);
1113 
1114                         search_key.objectid = inode_objectid;
1115                         search_key.type = BTRFS_INODE_EXTREF_KEY;
1116                         search_key.offset = btrfs_extref_hash(parent_objectid,
1117                                                               victim_name,
1118                                                               victim_name_len);
1119                         ret = 0;
1120                         if (!backref_in_log(log_root, &search_key,
1121                                             parent_objectid, victim_name,
1122                                             victim_name_len)) {
1123                                 ret = -ENOENT;
1124                                 victim_parent = read_one_inode(root,
1125                                                 parent_objectid);
1126                                 if (victim_parent) {
1127                                         inc_nlink(&inode->vfs_inode);
1128                                         btrfs_release_path(path);
1129 
1130                                         ret = btrfs_unlink_inode(trans, root,
1131                                                         BTRFS_I(victim_parent),
1132                                                         inode,
1133                                                         victim_name,
1134                                                         victim_name_len);
1135                                         if (!ret)
1136                                                 ret = btrfs_run_delayed_items(
1137                                                                   trans);
1138                                 }
1139                                 iput(victim_parent);
1140                                 kfree(victim_name);
1141                                 if (ret)
1142                                         return ret;
1143                                 *search_done = 1;
1144                                 goto again;
1145                         }
1146                         kfree(victim_name);
1147 next:
1148                         cur_offset += victim_name_len + sizeof(*extref);
1149                 }
1150                 *search_done = 1;
1151         }
1152         btrfs_release_path(path);
1153 
1154         /* look for a conflicting sequence number */
1155         di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1156                                          ref_index, name, namelen, 0);
1157         if (di && !IS_ERR(di)) {
1158                 ret = drop_one_dir_item(trans, root, path, dir, di);
1159                 if (ret)
1160                         return ret;
1161         }
1162         btrfs_release_path(path);
1163 
1164         /* look for a conflicting name */
1165         di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1166                                    name, namelen, 0);
1167         if (di && !IS_ERR(di)) {
1168                 ret = drop_one_dir_item(trans, root, path, dir, di);
1169                 if (ret)
1170                         return ret;
1171         }
1172         btrfs_release_path(path);
1173 
1174         return 0;
1175 }
1176 
1177 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1178                              u32 *namelen, char **name, u64 *index,
1179                              u64 *parent_objectid)
1180 {
1181         struct btrfs_inode_extref *extref;
1182 
1183         extref = (struct btrfs_inode_extref *)ref_ptr;
1184 
1185         *namelen = btrfs_inode_extref_name_len(eb, extref);
1186         *name = kmalloc(*namelen, GFP_NOFS);
1187         if (*name == NULL)
1188                 return -ENOMEM;
1189 
1190         read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1191                            *namelen);
1192 
1193         if (index)
1194                 *index = btrfs_inode_extref_index(eb, extref);
1195         if (parent_objectid)
1196                 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1197 
1198         return 0;
1199 }
1200 
1201 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1202                           u32 *namelen, char **name, u64 *index)
1203 {
1204         struct btrfs_inode_ref *ref;
1205 
1206         ref = (struct btrfs_inode_ref *)ref_ptr;
1207 
1208         *namelen = btrfs_inode_ref_name_len(eb, ref);
1209         *name = kmalloc(*namelen, GFP_NOFS);
1210         if (*name == NULL)
1211                 return -ENOMEM;
1212 
1213         read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1214 
1215         if (index)
1216                 *index = btrfs_inode_ref_index(eb, ref);
1217 
1218         return 0;
1219 }
1220 
1221 /*
1222  * Take an inode reference item from the log tree and iterate all names from the
1223  * inode reference item in the subvolume tree with the same key (if it exists).
1224  * For any name that is not in the inode reference item from the log tree, do a
1225  * proper unlink of that name (that is, remove its entry from the inode
1226  * reference item and both dir index keys).
1227  */
1228 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1229                                  struct btrfs_root *root,
1230                                  struct btrfs_path *path,
1231                                  struct btrfs_inode *inode,
1232                                  struct extent_buffer *log_eb,
1233                                  int log_slot,
1234                                  struct btrfs_key *key)
1235 {
1236         int ret;
1237         unsigned long ref_ptr;
1238         unsigned long ref_end;
1239         struct extent_buffer *eb;
1240 
1241 again:
1242         btrfs_release_path(path);
1243         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1244         if (ret > 0) {
1245                 ret = 0;
1246                 goto out;
1247         }
1248         if (ret < 0)
1249                 goto out;
1250 
1251         eb = path->nodes[0];
1252         ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1253         ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1254         while (ref_ptr < ref_end) {
1255                 char *name = NULL;
1256                 int namelen;
1257                 u64 parent_id;
1258 
1259                 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1260                         ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1261                                                 NULL, &parent_id);
1262                 } else {
1263                         parent_id = key->offset;
1264                         ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1265                                              NULL);
1266                 }
1267                 if (ret)
1268                         goto out;
1269 
1270                 if (key->type == BTRFS_INODE_EXTREF_KEY)
1271                         ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1272                                                                parent_id, name,
1273                                                                namelen);
1274                 else
1275                         ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1276                                                            name, namelen);
1277 
1278                 if (!ret) {
1279                         struct inode *dir;
1280 
1281                         btrfs_release_path(path);
1282                         dir = read_one_inode(root, parent_id);
1283                         if (!dir) {
1284                                 ret = -ENOENT;
1285                                 kfree(name);
1286                                 goto out;
1287                         }
1288                         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1289                                                  inode, name, namelen);
1290                         kfree(name);
1291                         iput(dir);
1292                         if (ret)
1293                                 goto out;
1294                         goto again;
1295                 }
1296 
1297                 kfree(name);
1298                 ref_ptr += namelen;
1299                 if (key->type == BTRFS_INODE_EXTREF_KEY)
1300                         ref_ptr += sizeof(struct btrfs_inode_extref);
1301                 else
1302                         ref_ptr += sizeof(struct btrfs_inode_ref);
1303         }
1304         ret = 0;
1305  out:
1306         btrfs_release_path(path);
1307         return ret;
1308 }
1309 
1310 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1311                                   const u8 ref_type, const char *name,
1312                                   const int namelen)
1313 {
1314         struct btrfs_key key;
1315         struct btrfs_path *path;
1316         const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1317         int ret;
1318 
1319         path = btrfs_alloc_path();
1320         if (!path)
1321                 return -ENOMEM;
1322 
1323         key.objectid = btrfs_ino(BTRFS_I(inode));
1324         key.type = ref_type;
1325         if (key.type == BTRFS_INODE_REF_KEY)
1326                 key.offset = parent_id;
1327         else
1328                 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1329 
1330         ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1331         if (ret < 0)
1332                 goto out;
1333         if (ret > 0) {
1334                 ret = 0;
1335                 goto out;
1336         }
1337         if (key.type == BTRFS_INODE_EXTREF_KEY)
1338                 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1339                                 path->slots[0], parent_id, name, namelen);
1340         else
1341                 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1342                                                    name, namelen);
1343 
1344 out:
1345         btrfs_free_path(path);
1346         return ret;
1347 }
1348 
1349 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1350                     struct inode *dir, struct inode *inode, const char *name,
1351                     int namelen, u64 ref_index)
1352 {
1353         struct btrfs_dir_item *dir_item;
1354         struct btrfs_key key;
1355         struct btrfs_path *path;
1356         struct inode *other_inode = NULL;
1357         int ret;
1358 
1359         path = btrfs_alloc_path();
1360         if (!path)
1361                 return -ENOMEM;
1362 
1363         dir_item = btrfs_lookup_dir_item(NULL, root, path,
1364                                          btrfs_ino(BTRFS_I(dir)),
1365                                          name, namelen, 0);
1366         if (!dir_item) {
1367                 btrfs_release_path(path);
1368                 goto add_link;
1369         } else if (IS_ERR(dir_item)) {
1370                 ret = PTR_ERR(dir_item);
1371                 goto out;
1372         }
1373 
1374         /*
1375          * Our inode's dentry collides with the dentry of another inode which is
1376          * in the log but not yet processed since it has a higher inode number.
1377          * So delete that other dentry.
1378          */
1379         btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1380         btrfs_release_path(path);
1381         other_inode = read_one_inode(root, key.objectid);
1382         if (!other_inode) {
1383                 ret = -ENOENT;
1384                 goto out;
1385         }
1386         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1387                                  name, namelen);
1388         if (ret)
1389                 goto out;
1390         /*
1391          * If we dropped the link count to 0, bump it so that later the iput()
1392          * on the inode will not free it. We will fixup the link count later.
1393          */
1394         if (other_inode->i_nlink == 0)
1395                 inc_nlink(other_inode);
1396 
1397         ret = btrfs_run_delayed_items(trans);
1398         if (ret)
1399                 goto out;
1400 add_link:
1401         ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1402                              name, namelen, 0, ref_index);
1403 out:
1404         iput(other_inode);
1405         btrfs_free_path(path);
1406 
1407         return ret;
1408 }
1409 
1410 /*
1411  * replay one inode back reference item found in the log tree.
1412  * eb, slot and key refer to the buffer and key found in the log tree.
1413  * root is the destination we are replaying into, and path is for temp
1414  * use by this function.  (it should be released on return).
1415  */
1416 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1417                                   struct btrfs_root *root,
1418                                   struct btrfs_root *log,
1419                                   struct btrfs_path *path,
1420                                   struct extent_buffer *eb, int slot,
1421                                   struct btrfs_key *key)
1422 {
1423         struct inode *dir = NULL;
1424         struct inode *inode = NULL;
1425         unsigned long ref_ptr;
1426         unsigned long ref_end;
1427         char *name = NULL;
1428         int namelen;
1429         int ret;
1430         int search_done = 0;
1431         int log_ref_ver = 0;
1432         u64 parent_objectid;
1433         u64 inode_objectid;
1434         u64 ref_index = 0;
1435         int ref_struct_size;
1436 
1437         ref_ptr = btrfs_item_ptr_offset(eb, slot);
1438         ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1439 
1440         if (key->type == BTRFS_INODE_EXTREF_KEY) {
1441                 struct btrfs_inode_extref *r;
1442 
1443                 ref_struct_size = sizeof(struct btrfs_inode_extref);
1444                 log_ref_ver = 1;
1445                 r = (struct btrfs_inode_extref *)ref_ptr;
1446                 parent_objectid = btrfs_inode_extref_parent(eb, r);
1447         } else {
1448                 ref_struct_size = sizeof(struct btrfs_inode_ref);
1449                 parent_objectid = key->offset;
1450         }
1451         inode_objectid = key->objectid;
1452 
1453         /*
1454          * it is possible that we didn't log all the parent directories
1455          * for a given inode.  If we don't find the dir, just don't
1456          * copy the back ref in.  The link count fixup code will take
1457          * care of the rest
1458          */
1459         dir = read_one_inode(root, parent_objectid);
1460         if (!dir) {
1461                 ret = -ENOENT;
1462                 goto out;
1463         }
1464 
1465         inode = read_one_inode(root, inode_objectid);
1466         if (!inode) {
1467                 ret = -EIO;
1468                 goto out;
1469         }
1470 
1471         while (ref_ptr < ref_end) {
1472                 if (log_ref_ver) {
1473                         ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1474                                                 &ref_index, &parent_objectid);
1475                         /*
1476                          * parent object can change from one array
1477                          * item to another.
1478                          */
1479                         if (!dir)
1480                                 dir = read_one_inode(root, parent_objectid);
1481                         if (!dir) {
1482                                 ret = -ENOENT;
1483                                 goto out;
1484                         }
1485                 } else {
1486                         ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1487                                              &ref_index);
1488                 }
1489                 if (ret)
1490                         goto out;
1491 
1492                 /* if we already have a perfect match, we're done */
1493                 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1494                                         btrfs_ino(BTRFS_I(inode)), ref_index,
1495                                         name, namelen)) {
1496                         /*
1497                          * look for a conflicting back reference in the
1498                          * metadata. if we find one we have to unlink that name
1499                          * of the file before we add our new link.  Later on, we
1500                          * overwrite any existing back reference, and we don't
1501                          * want to create dangling pointers in the directory.
1502                          */
1503 
1504                         if (!search_done) {
1505                                 ret = __add_inode_ref(trans, root, path, log,
1506                                                       BTRFS_I(dir),
1507                                                       BTRFS_I(inode),
1508                                                       inode_objectid,
1509                                                       parent_objectid,
1510                                                       ref_index, name, namelen,
1511                                                       &search_done);
1512                                 if (ret) {
1513                                         if (ret == 1)
1514                                                 ret = 0;
1515                                         goto out;
1516                                 }
1517                         }
1518 
1519                         /*
1520                          * If a reference item already exists for this inode
1521                          * with the same parent and name, but different index,
1522                          * drop it and the corresponding directory index entries
1523                          * from the parent before adding the new reference item
1524                          * and dir index entries, otherwise we would fail with
1525                          * -EEXIST returned from btrfs_add_link() below.
1526                          */
1527                         ret = btrfs_inode_ref_exists(inode, dir, key->type,
1528                                                      name, namelen);
1529                         if (ret > 0) {
1530                                 ret = btrfs_unlink_inode(trans, root,
1531                                                          BTRFS_I(dir),
1532                                                          BTRFS_I(inode),
1533                                                          name, namelen);
1534                                 /*
1535                                  * If we dropped the link count to 0, bump it so
1536                                  * that later the iput() on the inode will not
1537                                  * free it. We will fixup the link count later.
1538                                  */
1539                                 if (!ret && inode->i_nlink == 0)
1540                                         inc_nlink(inode);
1541                         }
1542                         if (ret < 0)
1543                                 goto out;
1544 
1545                         /* insert our name */
1546                         ret = add_link(trans, root, dir, inode, name, namelen,
1547                                        ref_index);
1548                         if (ret)
1549                                 goto out;
1550 
1551                         btrfs_update_inode(trans, root, inode);
1552                 }
1553 
1554                 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1555                 kfree(name);
1556                 name = NULL;
1557                 if (log_ref_ver) {
1558                         iput(dir);
1559                         dir = NULL;
1560                 }
1561         }
1562 
1563         /*
1564          * Before we overwrite the inode reference item in the subvolume tree
1565          * with the item from the log tree, we must unlink all names from the
1566          * parent directory that are in the subvolume's tree inode reference
1567          * item, otherwise we end up with an inconsistent subvolume tree where
1568          * dir index entries exist for a name but there is no inode reference
1569          * item with the same name.
1570          */
1571         ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1572                                     key);
1573         if (ret)
1574                 goto out;
1575 
1576         /* finally write the back reference in the inode */
1577         ret = overwrite_item(trans, root, path, eb, slot, key);
1578 out:
1579         btrfs_release_path(path);
1580         kfree(name);
1581         iput(dir);
1582         iput(inode);
1583         return ret;
1584 }
1585 
1586 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1587                               struct btrfs_root *root, u64 ino)
1588 {
1589         int ret;
1590 
1591         ret = btrfs_insert_orphan_item(trans, root, ino);
1592         if (ret == -EEXIST)
1593                 ret = 0;
1594 
1595         return ret;
1596 }
1597 
1598 static int count_inode_extrefs(struct btrfs_root *root,
1599                 struct btrfs_inode *inode, struct btrfs_path *path)
1600 {
1601         int ret = 0;
1602         int name_len;
1603         unsigned int nlink = 0;
1604         u32 item_size;
1605         u32 cur_offset = 0;
1606         u64 inode_objectid = btrfs_ino(inode);
1607         u64 offset = 0;
1608         unsigned long ptr;
1609         struct btrfs_inode_extref *extref;
1610         struct extent_buffer *leaf;
1611 
1612         while (1) {
1613                 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1614                                             &extref, &offset);
1615                 if (ret)
1616                         break;
1617 
1618                 leaf = path->nodes[0];
1619                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1620                 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1621                 cur_offset = 0;
1622 
1623                 while (cur_offset < item_size) {
1624                         extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1625                         name_len = btrfs_inode_extref_name_len(leaf, extref);
1626 
1627                         nlink++;
1628 
1629                         cur_offset += name_len + sizeof(*extref);
1630                 }
1631 
1632                 offset++;
1633                 btrfs_release_path(path);
1634         }
1635         btrfs_release_path(path);
1636 
1637         if (ret < 0 && ret != -ENOENT)
1638                 return ret;
1639         return nlink;
1640 }
1641 
1642 static int count_inode_refs(struct btrfs_root *root,
1643                         struct btrfs_inode *inode, struct btrfs_path *path)
1644 {
1645         int ret;
1646         struct btrfs_key key;
1647         unsigned int nlink = 0;
1648         unsigned long ptr;
1649         unsigned long ptr_end;
1650         int name_len;
1651         u64 ino = btrfs_ino(inode);
1652 
1653         key.objectid = ino;
1654         key.type = BTRFS_INODE_REF_KEY;
1655         key.offset = (u64)-1;
1656 
1657         while (1) {
1658                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1659                 if (ret < 0)
1660                         break;
1661                 if (ret > 0) {
1662                         if (path->slots[0] == 0)
1663                                 break;
1664                         path->slots[0]--;
1665                 }
1666 process_slot:
1667                 btrfs_item_key_to_cpu(path->nodes[0], &key,
1668                                       path->slots[0]);
1669                 if (key.objectid != ino ||
1670                     key.type != BTRFS_INODE_REF_KEY)
1671                         break;
1672                 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1673                 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1674                                                    path->slots[0]);
1675                 while (ptr < ptr_end) {
1676                         struct btrfs_inode_ref *ref;
1677 
1678                         ref = (struct btrfs_inode_ref *)ptr;
1679                         name_len = btrfs_inode_ref_name_len(path->nodes[0],
1680                                                             ref);
1681                         ptr = (unsigned long)(ref + 1) + name_len;
1682                         nlink++;
1683                 }
1684 
1685                 if (key.offset == 0)
1686                         break;
1687                 if (path->slots[0] > 0) {
1688                         path->slots[0]--;
1689                         goto process_slot;
1690                 }
1691                 key.offset--;
1692                 btrfs_release_path(path);
1693         }
1694         btrfs_release_path(path);
1695 
1696         return nlink;
1697 }
1698 
1699 /*
1700  * There are a few corners where the link count of the file can't
1701  * be properly maintained during replay.  So, instead of adding
1702  * lots of complexity to the log code, we just scan the backrefs
1703  * for any file that has been through replay.
1704  *
1705  * The scan will update the link count on the inode to reflect the
1706  * number of back refs found.  If it goes down to zero, the iput
1707  * will free the inode.
1708  */
1709 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1710                                            struct btrfs_root *root,
1711                                            struct inode *inode)
1712 {
1713         struct btrfs_path *path;
1714         int ret;
1715         u64 nlink = 0;
1716         u64 ino = btrfs_ino(BTRFS_I(inode));
1717 
1718         path = btrfs_alloc_path();
1719         if (!path)
1720                 return -ENOMEM;
1721 
1722         ret = count_inode_refs(root, BTRFS_I(inode), path);
1723         if (ret < 0)
1724                 goto out;
1725 
1726         nlink = ret;
1727 
1728         ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1729         if (ret < 0)
1730                 goto out;
1731 
1732         nlink += ret;
1733 
1734         ret = 0;
1735 
1736         if (nlink != inode->i_nlink) {
1737                 set_nlink(inode, nlink);
1738                 btrfs_update_inode(trans, root, inode);
1739         }
1740         BTRFS_I(inode)->index_cnt = (u64)-1;
1741 
1742         if (inode->i_nlink == 0) {
1743                 if (S_ISDIR(inode->i_mode)) {
1744                         ret = replay_dir_deletes(trans, root, NULL, path,
1745                                                  ino, 1);
1746                         if (ret)
1747                                 goto out;
1748                 }
1749                 ret = insert_orphan_item(trans, root, ino);
1750         }
1751 
1752 out:
1753         btrfs_free_path(path);
1754         return ret;
1755 }
1756 
1757 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1758                                             struct btrfs_root *root,
1759                                             struct btrfs_path *path)
1760 {
1761         int ret;
1762         struct btrfs_key key;
1763         struct inode *inode;
1764 
1765         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1766         key.type = BTRFS_ORPHAN_ITEM_KEY;
1767         key.offset = (u64)-1;
1768         while (1) {
1769                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1770                 if (ret < 0)
1771                         break;
1772 
1773                 if (ret == 1) {
1774                         if (path->slots[0] == 0)
1775                                 break;
1776                         path->slots[0]--;
1777                 }
1778 
1779                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1780                 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1781                     key.type != BTRFS_ORPHAN_ITEM_KEY)
1782                         break;
1783 
1784                 ret = btrfs_del_item(trans, root, path);
1785                 if (ret)
1786                         goto out;
1787 
1788                 btrfs_release_path(path);
1789                 inode = read_one_inode(root, key.offset);
1790                 if (!inode)
1791                         return -EIO;
1792 
1793                 ret = fixup_inode_link_count(trans, root, inode);
1794                 iput(inode);
1795                 if (ret)
1796                         goto out;
1797 
1798                 /*
1799                  * fixup on a directory may create new entries,
1800                  * make sure we always look for the highset possible
1801                  * offset
1802                  */
1803                 key.offset = (u64)-1;
1804         }
1805         ret = 0;
1806 out:
1807         btrfs_release_path(path);
1808         return ret;
1809 }
1810 
1811 
1812 /*
1813  * record a given inode in the fixup dir so we can check its link
1814  * count when replay is done.  The link count is incremented here
1815  * so the inode won't go away until we check it
1816  */
1817 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1818                                       struct btrfs_root *root,
1819                                       struct btrfs_path *path,
1820                                       u64 objectid)
1821 {
1822         struct btrfs_key key;
1823         int ret = 0;
1824         struct inode *inode;
1825 
1826         inode = read_one_inode(root, objectid);
1827         if (!inode)
1828                 return -EIO;
1829 
1830         key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1831         key.type = BTRFS_ORPHAN_ITEM_KEY;
1832         key.offset = objectid;
1833 
1834         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1835 
1836         btrfs_release_path(path);
1837         if (ret == 0) {
1838                 if (!inode->i_nlink)
1839                         set_nlink(inode, 1);
1840                 else
1841                         inc_nlink(inode);
1842                 ret = btrfs_update_inode(trans, root, inode);
1843         } else if (ret == -EEXIST) {
1844                 ret = 0;
1845         } else {
1846                 BUG(); /* Logic Error */
1847         }
1848         iput(inode);
1849 
1850         return ret;
1851 }
1852 
1853 /*
1854  * when replaying the log for a directory, we only insert names
1855  * for inodes that actually exist.  This means an fsync on a directory
1856  * does not implicitly fsync all the new files in it
1857  */
1858 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1859                                     struct btrfs_root *root,
1860                                     u64 dirid, u64 index,
1861                                     char *name, int name_len,
1862                                     struct btrfs_key *location)
1863 {
1864         struct inode *inode;
1865         struct inode *dir;
1866         int ret;
1867 
1868         inode = read_one_inode(root, location->objectid);
1869         if (!inode)
1870                 return -ENOENT;
1871 
1872         dir = read_one_inode(root, dirid);
1873         if (!dir) {
1874                 iput(inode);
1875                 return -EIO;
1876         }
1877 
1878         ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1879                         name_len, 1, index);
1880 
1881         /* FIXME, put inode into FIXUP list */
1882 
1883         iput(inode);
1884         iput(dir);
1885         return ret;
1886 }
1887 
1888 /*
1889  * Return true if an inode reference exists in the log for the given name,
1890  * inode and parent inode.
1891  */
1892 static bool name_in_log_ref(struct btrfs_root *log_root,
1893                             const char *name, const int name_len,
1894                             const u64 dirid, const u64 ino)
1895 {
1896         struct btrfs_key search_key;
1897 
1898         search_key.objectid = ino;
1899         search_key.type = BTRFS_INODE_REF_KEY;
1900         search_key.offset = dirid;
1901         if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1902                 return true;
1903 
1904         search_key.type = BTRFS_INODE_EXTREF_KEY;
1905         search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1906         if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1907                 return true;
1908 
1909         return false;
1910 }
1911 
1912 /*
1913  * take a single entry in a log directory item and replay it into
1914  * the subvolume.
1915  *
1916  * if a conflicting item exists in the subdirectory already,
1917  * the inode it points to is unlinked and put into the link count
1918  * fix up tree.
1919  *
1920  * If a name from the log points to a file or directory that does
1921  * not exist in the FS, it is skipped.  fsyncs on directories
1922  * do not force down inodes inside that directory, just changes to the
1923  * names or unlinks in a directory.
1924  *
1925  * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1926  * non-existing inode) and 1 if the name was replayed.
1927  */
1928 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1929                                     struct btrfs_root *root,
1930                                     struct btrfs_path *path,
1931                                     struct extent_buffer *eb,
1932                                     struct btrfs_dir_item *di,
1933                                     struct btrfs_key *key)
1934 {
1935         char *name;
1936         int name_len;
1937         struct btrfs_dir_item *dst_di;
1938         struct btrfs_key found_key;
1939         struct btrfs_key log_key;
1940         struct inode *dir;
1941         u8 log_type;
1942         int exists;
1943         int ret = 0;
1944         bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1945         bool name_added = false;
1946 
1947         dir = read_one_inode(root, key->objectid);
1948         if (!dir)
1949                 return -EIO;
1950 
1951         name_len = btrfs_dir_name_len(eb, di);
1952         name = kmalloc(name_len, GFP_NOFS);
1953         if (!name) {
1954                 ret = -ENOMEM;
1955                 goto out;
1956         }
1957 
1958         log_type = btrfs_dir_type(eb, di);
1959         read_extent_buffer(eb, name, (unsigned long)(di + 1),
1960                    name_len);
1961 
1962         btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1963         exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1964         if (exists == 0)
1965                 exists = 1;
1966         else
1967                 exists = 0;
1968         btrfs_release_path(path);
1969 
1970         if (key->type == BTRFS_DIR_ITEM_KEY) {
1971                 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1972                                        name, name_len, 1);
1973         } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1974                 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1975                                                      key->objectid,
1976                                                      key->offset, name,
1977                                                      name_len, 1);
1978         } else {
1979                 /* Corruption */
1980                 ret = -EINVAL;
1981                 goto out;
1982         }
1983         if (IS_ERR_OR_NULL(dst_di)) {
1984                 /* we need a sequence number to insert, so we only
1985                  * do inserts for the BTRFS_DIR_INDEX_KEY types
1986                  */
1987                 if (key->type != BTRFS_DIR_INDEX_KEY)
1988                         goto out;
1989                 goto insert;
1990         }
1991 
1992         btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1993         /* the existing item matches the logged item */
1994         if (found_key.objectid == log_key.objectid &&
1995             found_key.type == log_key.type &&
1996             found_key.offset == log_key.offset &&
1997             btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1998                 update_size = false;
1999                 goto out;
2000         }
2001 
2002         /*
2003          * don't drop the conflicting directory entry if the inode
2004          * for the new entry doesn't exist
2005          */
2006         if (!exists)
2007                 goto out;
2008 
2009         ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2010         if (ret)
2011                 goto out;
2012 
2013         if (key->type == BTRFS_DIR_INDEX_KEY)
2014                 goto insert;
2015 out:
2016         btrfs_release_path(path);
2017         if (!ret && update_size) {
2018                 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2019                 ret = btrfs_update_inode(trans, root, dir);
2020         }
2021         kfree(name);
2022         iput(dir);
2023         if (!ret && name_added)
2024                 ret = 1;
2025         return ret;
2026 
2027 insert:
2028         if (name_in_log_ref(root->log_root, name, name_len,
2029                             key->objectid, log_key.objectid)) {
2030                 /* The dentry will be added later. */
2031                 ret = 0;
2032                 update_size = false;
2033                 goto out;
2034         }
2035         btrfs_release_path(path);
2036         ret = insert_one_name(trans, root, key->objectid, key->offset,
2037                               name, name_len, &log_key);
2038         if (ret && ret != -ENOENT && ret != -EEXIST)
2039                 goto out;
2040         if (!ret)
2041                 name_added = true;
2042         update_size = false;
2043         ret = 0;
2044         goto out;
2045 }
2046 
2047 /*
2048  * find all the names in a directory item and reconcile them into
2049  * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
2050  * one name in a directory item, but the same code gets used for
2051  * both directory index types
2052  */
2053 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2054                                         struct btrfs_root *root,
2055                                         struct btrfs_path *path,
2056                                         struct extent_buffer *eb, int slot,
2057                                         struct btrfs_key *key)
2058 {
2059         int ret = 0;
2060         u32 item_size = btrfs_item_size_nr(eb, slot);
2061         struct btrfs_dir_item *di;
2062         int name_len;
2063         unsigned long ptr;
2064         unsigned long ptr_end;
2065         struct btrfs_path *fixup_path = NULL;
2066 
2067         ptr = btrfs_item_ptr_offset(eb, slot);
2068         ptr_end = ptr + item_size;
2069         while (ptr < ptr_end) {
2070                 di = (struct btrfs_dir_item *)ptr;
2071                 name_len = btrfs_dir_name_len(eb, di);
2072                 ret = replay_one_name(trans, root, path, eb, di, key);
2073                 if (ret < 0)
2074                         break;
2075                 ptr = (unsigned long)(di + 1);
2076                 ptr += name_len;
2077 
2078                 /*
2079                  * If this entry refers to a non-directory (directories can not
2080                  * have a link count > 1) and it was added in the transaction
2081                  * that was not committed, make sure we fixup the link count of
2082                  * the inode it the entry points to. Otherwise something like
2083                  * the following would result in a directory pointing to an
2084                  * inode with a wrong link that does not account for this dir
2085                  * entry:
2086                  *
2087                  * mkdir testdir
2088                  * touch testdir/foo
2089                  * touch testdir/bar
2090                  * sync
2091                  *
2092                  * ln testdir/bar testdir/bar_link
2093                  * ln testdir/foo testdir/foo_link
2094                  * xfs_io -c "fsync" testdir/bar
2095                  *
2096                  * <power failure>
2097                  *
2098                  * mount fs, log replay happens
2099                  *
2100                  * File foo would remain with a link count of 1 when it has two
2101                  * entries pointing to it in the directory testdir. This would
2102                  * make it impossible to ever delete the parent directory has
2103                  * it would result in stale dentries that can never be deleted.
2104                  */
2105                 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2106                         struct btrfs_key di_key;
2107 
2108                         if (!fixup_path) {
2109                                 fixup_path = btrfs_alloc_path();
2110                                 if (!fixup_path) {
2111                                         ret = -ENOMEM;
2112                                         break;
2113                                 }
2114                         }
2115 
2116                         btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2117                         ret = link_to_fixup_dir(trans, root, fixup_path,
2118                                                 di_key.objectid);
2119                         if (ret)
2120                                 break;
2121                 }
2122                 ret = 0;
2123         }
2124         btrfs_free_path(fixup_path);
2125         return ret;
2126 }
2127 
2128 /*
2129  * directory replay has two parts.  There are the standard directory
2130  * items in the log copied from the subvolume, and range items
2131  * created in the log while the subvolume was logged.
2132  *
2133  * The range items tell us which parts of the key space the log
2134  * is authoritative for.  During replay, if a key in the subvolume
2135  * directory is in a logged range item, but not actually in the log
2136  * that means it was deleted from the directory before the fsync
2137  * and should be removed.
2138  */
2139 static noinline int find_dir_range(struct btrfs_root *root,
2140                                    struct btrfs_path *path,
2141                                    u64 dirid, int key_type,
2142                                    u64 *start_ret, u64 *end_ret)
2143 {
2144         struct btrfs_key key;
2145         u64 found_end;
2146         struct btrfs_dir_log_item *item;
2147         int ret;
2148         int nritems;
2149 
2150         if (*start_ret == (u64)-1)
2151                 return 1;
2152 
2153         key.objectid = dirid;
2154         key.type = key_type;
2155         key.offset = *start_ret;
2156 
2157         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2158         if (ret < 0)
2159                 goto out;
2160         if (ret > 0) {
2161                 if (path->slots[0] == 0)
2162                         goto out;
2163                 path->slots[0]--;
2164         }
2165         if (ret != 0)
2166                 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2167 
2168         if (key.type != key_type || key.objectid != dirid) {
2169                 ret = 1;
2170                 goto next;
2171         }
2172         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2173                               struct btrfs_dir_log_item);
2174         found_end = btrfs_dir_log_end(path->nodes[0], item);
2175 
2176         if (*start_ret >= key.offset && *start_ret <= found_end) {
2177                 ret = 0;
2178                 *start_ret = key.offset;
2179                 *end_ret = found_end;
2180                 goto out;
2181         }
2182         ret = 1;
2183 next:
2184         /* check the next slot in the tree to see if it is a valid item */
2185         nritems = btrfs_header_nritems(path->nodes[0]);
2186         path->slots[0]++;
2187         if (path->slots[0] >= nritems) {
2188                 ret = btrfs_next_leaf(root, path);
2189                 if (ret)
2190                         goto out;
2191         }
2192 
2193         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2194 
2195         if (key.type != key_type || key.objectid != dirid) {
2196                 ret = 1;
2197                 goto out;
2198         }
2199         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2200                               struct btrfs_dir_log_item);
2201         found_end = btrfs_dir_log_end(path->nodes[0], item);
2202         *start_ret = key.offset;
2203         *end_ret = found_end;
2204         ret = 0;
2205 out:
2206         btrfs_release_path(path);
2207         return ret;
2208 }
2209 
2210 /*
2211  * this looks for a given directory item in the log.  If the directory
2212  * item is not in the log, the item is removed and the inode it points
2213  * to is unlinked
2214  */
2215 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2216                                       struct btrfs_root *root,
2217                                       struct btrfs_root *log,
2218                                       struct btrfs_path *path,
2219                                       struct btrfs_path *log_path,
2220                                       struct inode *dir,
2221                                       struct btrfs_key *dir_key)
2222 {
2223         int ret;
2224         struct extent_buffer *eb;
2225         int slot;
2226         u32 item_size;
2227         struct btrfs_dir_item *di;
2228         struct btrfs_dir_item *log_di;
2229         int name_len;
2230         unsigned long ptr;
2231         unsigned long ptr_end;
2232         char *name;
2233         struct inode *inode;
2234         struct btrfs_key location;
2235 
2236 again:
2237         eb = path->nodes[0];
2238         slot = path->slots[0];
2239         item_size = btrfs_item_size_nr(eb, slot);
2240         ptr = btrfs_item_ptr_offset(eb, slot);
2241         ptr_end = ptr + item_size;
2242         while (ptr < ptr_end) {
2243                 di = (struct btrfs_dir_item *)ptr;
2244                 name_len = btrfs_dir_name_len(eb, di);
2245                 name = kmalloc(name_len, GFP_NOFS);
2246                 if (!name) {
2247                         ret = -ENOMEM;
2248                         goto out;
2249                 }
2250                 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2251                                   name_len);
2252                 log_di = NULL;
2253                 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2254                         log_di = btrfs_lookup_dir_item(trans, log, log_path,
2255                                                        dir_key->objectid,
2256                                                        name, name_len, 0);
2257                 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2258                         log_di = btrfs_lookup_dir_index_item(trans, log,
2259                                                      log_path,
2260                                                      dir_key->objectid,
2261                                                      dir_key->offset,
2262                                                      name, name_len, 0);
2263                 }
2264                 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2265                         btrfs_dir_item_key_to_cpu(eb, di, &location);
2266                         btrfs_release_path(path);
2267                         btrfs_release_path(log_path);
2268                         inode = read_one_inode(root, location.objectid);
2269                         if (!inode) {
2270                                 kfree(name);
2271                                 return -EIO;
2272                         }
2273 
2274                         ret = link_to_fixup_dir(trans, root,
2275                                                 path, location.objectid);
2276                         if (ret) {
2277                                 kfree(name);
2278                                 iput(inode);
2279                                 goto out;
2280                         }
2281 
2282                         inc_nlink(inode);
2283                         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2284                                         BTRFS_I(inode), name, name_len);
2285                         if (!ret)
2286                                 ret = btrfs_run_delayed_items(trans);
2287                         kfree(name);
2288                         iput(inode);
2289                         if (ret)
2290                                 goto out;
2291 
2292                         /* there might still be more names under this key
2293                          * check and repeat if required
2294                          */
2295                         ret = btrfs_search_slot(NULL, root, dir_key, path,
2296                                                 0, 0);
2297                         if (ret == 0)
2298                                 goto again;
2299                         ret = 0;
2300                         goto out;
2301                 } else if (IS_ERR(log_di)) {
2302                         kfree(name);
2303                         return PTR_ERR(log_di);
2304                 }
2305                 btrfs_release_path(log_path);
2306                 kfree(name);
2307 
2308                 ptr = (unsigned long)(di + 1);
2309                 ptr += name_len;
2310         }
2311         ret = 0;
2312 out:
2313         btrfs_release_path(path);
2314         btrfs_release_path(log_path);
2315         return ret;
2316 }
2317 
2318 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2319                               struct btrfs_root *root,
2320                               struct btrfs_root *log,
2321                               struct btrfs_path *path,
2322                               const u64 ino)
2323 {
2324         struct btrfs_key search_key;
2325         struct btrfs_path *log_path;
2326         int i;
2327         int nritems;
2328         int ret;
2329 
2330         log_path = btrfs_alloc_path();
2331         if (!log_path)
2332                 return -ENOMEM;
2333 
2334         search_key.objectid = ino;
2335         search_key.type = BTRFS_XATTR_ITEM_KEY;
2336         search_key.offset = 0;
2337 again:
2338         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2339         if (ret < 0)
2340                 goto out;
2341 process_leaf:
2342         nritems = btrfs_header_nritems(path->nodes[0]);
2343         for (i = path->slots[0]; i < nritems; i++) {
2344                 struct btrfs_key key;
2345                 struct btrfs_dir_item *di;
2346                 struct btrfs_dir_item *log_di;
2347                 u32 total_size;
2348                 u32 cur;
2349 
2350                 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2351                 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2352                         ret = 0;
2353                         goto out;
2354                 }
2355 
2356                 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2357                 total_size = btrfs_item_size_nr(path->nodes[0], i);
2358                 cur = 0;
2359                 while (cur < total_size) {
2360                         u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2361                         u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2362                         u32 this_len = sizeof(*di) + name_len + data_len;
2363                         char *name;
2364 
2365                         name = kmalloc(name_len, GFP_NOFS);
2366                         if (!name) {
2367                                 ret = -ENOMEM;
2368                                 goto out;
2369                         }
2370                         read_extent_buffer(path->nodes[0], name,
2371                                            (unsigned long)(di + 1), name_len);
2372 
2373                         log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2374                                                     name, name_len, 0);
2375                         btrfs_release_path(log_path);
2376                         if (!log_di) {
2377                                 /* Doesn't exist in log tree, so delete it. */
2378                                 btrfs_release_path(path);
2379                                 di = btrfs_lookup_xattr(trans, root, path, ino,
2380                                                         name, name_len, -1);
2381                                 kfree(name);
2382                                 if (IS_ERR(di)) {
2383                                         ret = PTR_ERR(di);
2384                                         goto out;
2385                                 }
2386                                 ASSERT(di);
2387                                 ret = btrfs_delete_one_dir_name(trans, root,
2388                                                                 path, di);
2389                                 if (ret)
2390                                         goto out;
2391                                 btrfs_release_path(path);
2392                                 search_key = key;
2393                                 goto again;
2394                         }
2395                         kfree(name);
2396                         if (IS_ERR(log_di)) {
2397                                 ret = PTR_ERR(log_di);
2398                                 goto out;
2399                         }
2400                         cur += this_len;
2401                         di = (struct btrfs_dir_item *)((char *)di + this_len);
2402                 }
2403         }
2404         ret = btrfs_next_leaf(root, path);
2405         if (ret > 0)
2406                 ret = 0;
2407         else if (ret == 0)
2408                 goto process_leaf;
2409 out:
2410         btrfs_free_path(log_path);
2411         btrfs_release_path(path);
2412         return ret;
2413 }
2414 
2415 
2416 /*
2417  * deletion replay happens before we copy any new directory items
2418  * out of the log or out of backreferences from inodes.  It
2419  * scans the log to find ranges of keys that log is authoritative for,
2420  * and then scans the directory to find items in those ranges that are
2421  * not present in the log.
2422  *
2423  * Anything we don't find in the log is unlinked and removed from the
2424  * directory.
2425  */
2426 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2427                                        struct btrfs_root *root,
2428                                        struct btrfs_root *log,
2429                                        struct btrfs_path *path,
2430                                        u64 dirid, int del_all)
2431 {
2432         u64 range_start;
2433         u64 range_end;
2434         int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2435         int ret = 0;
2436         struct btrfs_key dir_key;
2437         struct btrfs_key found_key;
2438         struct btrfs_path *log_path;
2439         struct inode *dir;
2440 
2441         dir_key.objectid = dirid;
2442         dir_key.type = BTRFS_DIR_ITEM_KEY;
2443         log_path = btrfs_alloc_path();
2444         if (!log_path)
2445                 return -ENOMEM;
2446 
2447         dir = read_one_inode(root, dirid);
2448         /* it isn't an error if the inode isn't there, that can happen
2449          * because we replay the deletes before we copy in the inode item
2450          * from the log
2451          */
2452         if (!dir) {
2453                 btrfs_free_path(log_path);
2454                 return 0;
2455         }
2456 again:
2457         range_start = 0;
2458         range_end = 0;
2459         while (1) {
2460                 if (del_all)
2461                         range_end = (u64)-1;
2462                 else {
2463                         ret = find_dir_range(log, path, dirid, key_type,
2464                                              &range_start, &range_end);
2465                         if (ret != 0)
2466                                 break;
2467                 }
2468 
2469                 dir_key.offset = range_start;
2470                 while (1) {
2471                         int nritems;
2472                         ret = btrfs_search_slot(NULL, root, &dir_key, path,
2473                                                 0, 0);
2474                         if (ret < 0)
2475                                 goto out;
2476 
2477                         nritems = btrfs_header_nritems(path->nodes[0]);
2478                         if (path->slots[0] >= nritems) {
2479                                 ret = btrfs_next_leaf(root, path);
2480                                 if (ret == 1)
2481                                         break;
2482                                 else if (ret < 0)
2483                                         goto out;
2484                         }
2485                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2486                                               path->slots[0]);
2487                         if (found_key.objectid != dirid ||
2488                             found_key.type != dir_key.type)
2489                                 goto next_type;
2490 
2491                         if (found_key.offset > range_end)
2492                                 break;
2493 
2494                         ret = check_item_in_log(trans, root, log, path,
2495                                                 log_path, dir,
2496                                                 &found_key);
2497                         if (ret)
2498                                 goto out;
2499                         if (found_key.offset == (u64)-1)
2500                                 break;
2501                         dir_key.offset = found_key.offset + 1;
2502                 }
2503                 btrfs_release_path(path);
2504                 if (range_end == (u64)-1)
2505                         break;
2506                 range_start = range_end + 1;
2507         }
2508 
2509 next_type:
2510         ret = 0;
2511         if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2512                 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2513                 dir_key.type = BTRFS_DIR_INDEX_KEY;
2514                 btrfs_release_path(path);
2515                 goto again;
2516         }
2517 out:
2518         btrfs_release_path(path);
2519         btrfs_free_path(log_path);
2520         iput(dir);
2521         return ret;
2522 }
2523 
2524 /*
2525  * the process_func used to replay items from the log tree.  This
2526  * gets called in two different stages.  The first stage just looks
2527  * for inodes and makes sure they are all copied into the subvolume.
2528  *
2529  * The second stage copies all the other item types from the log into
2530  * the subvolume.  The two stage approach is slower, but gets rid of
2531  * lots of complexity around inodes referencing other inodes that exist
2532  * only in the log (references come from either directory items or inode
2533  * back refs).
2534  */
2535 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2536                              struct walk_control *wc, u64 gen, int level)
2537 {
2538         int nritems;
2539         struct btrfs_path *path;
2540         struct btrfs_root *root = wc->replay_dest;
2541         struct btrfs_key key;
2542         int i;
2543         int ret;
2544 
2545         ret = btrfs_read_buffer(eb, gen, level, NULL);
2546         if (ret)
2547                 return ret;
2548 
2549         level = btrfs_header_level(eb);
2550 
2551         if (level != 0)
2552                 return 0;
2553 
2554         path = btrfs_alloc_path();
2555         if (!path)
2556                 return -ENOMEM;
2557 
2558         nritems = btrfs_header_nritems(eb);
2559         for (i = 0; i < nritems; i++) {
2560                 btrfs_item_key_to_cpu(eb, &key, i);
2561 
2562                 /* inode keys are done during the first stage */
2563                 if (key.type == BTRFS_INODE_ITEM_KEY &&
2564                     wc->stage == LOG_WALK_REPLAY_INODES) {
2565                         struct btrfs_inode_item *inode_item;
2566                         u32 mode;
2567 
2568                         inode_item = btrfs_item_ptr(eb, i,
2569                                             struct btrfs_inode_item);
2570                         /*
2571                          * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2572                          * and never got linked before the fsync, skip it, as
2573                          * replaying it is pointless since it would be deleted
2574                          * later. We skip logging tmpfiles, but it's always
2575                          * possible we are replaying a log created with a kernel
2576                          * that used to log tmpfiles.
2577                          */
2578                         if (btrfs_inode_nlink(eb, inode_item) == 0) {
2579                                 wc->ignore_cur_inode = true;
2580                                 continue;
2581                         } else {
2582                                 wc->ignore_cur_inode = false;
2583                         }
2584                         ret = replay_xattr_deletes(wc->trans, root, log,
2585                                                    path, key.objectid);
2586                         if (ret)
2587                                 break;
2588                         mode = btrfs_inode_mode(eb, inode_item);
2589                         if (S_ISDIR(mode)) {
2590                                 ret = replay_dir_deletes(wc->trans,
2591                                          root, log, path, key.objectid, 0);
2592                                 if (ret)
2593                                         break;
2594                         }
2595                         ret = overwrite_item(wc->trans, root, path,
2596                                              eb, i, &key);
2597                         if (ret)
2598                                 break;
2599 
2600                         /*
2601                          * Before replaying extents, truncate the inode to its
2602                          * size. We need to do it now and not after log replay
2603                          * because before an fsync we can have prealloc extents
2604                          * added beyond the inode's i_size. If we did it after,
2605                          * through orphan cleanup for example, we would drop
2606                          * those prealloc extents just after replaying them.
2607                          */
2608                         if (S_ISREG(mode)) {
2609                                 struct inode *inode;
2610                                 u64 from;
2611 
2612                                 inode = read_one_inode(root, key.objectid);
2613                                 if (!inode) {
2614                                         ret = -EIO;
2615                                         break;
2616                                 }
2617                                 from = ALIGN(i_size_read(inode),
2618                                              root->fs_info->sectorsize);
2619                                 ret = btrfs_drop_extents(wc->trans, root, inode,
2620                                                          from, (u64)-1, 1);
2621                                 if (!ret) {
2622                                         /* Update the inode's nbytes. */
2623                                         ret = btrfs_update_inode(wc->trans,
2624                                                                  root, inode);
2625                                 }
2626                                 iput(inode);
2627                                 if (ret)
2628                                         break;
2629                         }
2630 
2631                         ret = link_to_fixup_dir(wc->trans, root,
2632                                                 path, key.objectid);
2633                         if (ret)
2634                                 break;
2635                 }
2636 
2637                 if (wc->ignore_cur_inode)
2638                         continue;
2639 
2640                 if (key.type == BTRFS_DIR_INDEX_KEY &&
2641                     wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2642                         ret = replay_one_dir_item(wc->trans, root, path,
2643                                                   eb, i, &key);
2644                         if (ret)
2645                                 break;
2646                 }
2647 
2648                 if (wc->stage < LOG_WALK_REPLAY_ALL)
2649                         continue;
2650 
2651                 /* these keys are simply copied */
2652                 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2653                         ret = overwrite_item(wc->trans, root, path,
2654                                              eb, i, &key);
2655                         if (ret)
2656                                 break;
2657                 } else if (key.type == BTRFS_INODE_REF_KEY ||
2658                            key.type == BTRFS_INODE_EXTREF_KEY) {
2659                         ret = add_inode_ref(wc->trans, root, log, path,
2660                                             eb, i, &key);
2661                         if (ret && ret != -ENOENT)
2662                                 break;
2663                         ret = 0;
2664                 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2665                         ret = replay_one_extent(wc->trans, root, path,
2666                                                 eb, i, &key);
2667                         if (ret)
2668                                 break;
2669                 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2670                         ret = replay_one_dir_item(wc->trans, root, path,
2671                                                   eb, i, &key);
2672                         if (ret)
2673                                 break;
2674                 }
2675         }
2676         btrfs_free_path(path);
2677         return ret;
2678 }
2679 
2680 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2681                                    struct btrfs_root *root,
2682                                    struct btrfs_path *path, int *level,
2683                                    struct walk_control *wc)
2684 {
2685         struct btrfs_fs_info *fs_info = root->fs_info;
2686         u64 root_owner;
2687         u64 bytenr;
2688         u64 ptr_gen;
2689         struct extent_buffer *next;
2690         struct extent_buffer *cur;
2691         struct extent_buffer *parent;
2692         u32 blocksize;
2693         int ret = 0;
2694 
2695         WARN_ON(*level < 0);
2696         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2697 
2698         while (*level > 0) {
2699                 struct btrfs_key first_key;
2700 
2701                 WARN_ON(*level < 0);
2702                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2703                 cur = path->nodes[*level];
2704 
2705                 WARN_ON(btrfs_header_level(cur) != *level);
2706 
2707                 if (path->slots[*level] >=
2708                     btrfs_header_nritems(cur))
2709                         break;
2710 
2711                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2712                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2713                 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2714                 blocksize = fs_info->nodesize;
2715 
2716                 parent = path->nodes[*level];
2717                 root_owner = btrfs_header_owner(parent);
2718 
2719                 next = btrfs_find_create_tree_block(fs_info, bytenr);
2720                 if (IS_ERR(next))
2721                         return PTR_ERR(next);
2722 
2723                 if (*level == 1) {
2724                         ret = wc->process_func(root, next, wc, ptr_gen,
2725                                                *level - 1);
2726                         if (ret) {
2727                                 free_extent_buffer(next);
2728                                 return ret;
2729                         }
2730 
2731                         path->slots[*level]++;
2732                         if (wc->free) {
2733                                 ret = btrfs_read_buffer(next, ptr_gen,
2734                                                         *level - 1, &first_key);
2735                                 if (ret) {
2736                                         free_extent_buffer(next);
2737                                         return ret;
2738                                 }
2739 
2740                                 if (trans) {
2741                                         btrfs_tree_lock(next);
2742                                         btrfs_set_lock_blocking_write(next);
2743                                         btrfs_clean_tree_block(next);
2744                                         btrfs_wait_tree_block_writeback(next);
2745                                         btrfs_tree_unlock(next);
2746                                 } else {
2747                                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2748                                                 clear_extent_buffer_dirty(next);
2749                                 }
2750 
2751                                 WARN_ON(root_owner !=
2752                                         BTRFS_TREE_LOG_OBJECTID);
2753                                 ret = btrfs_free_and_pin_reserved_extent(
2754                                                         fs_info, bytenr,
2755                                                         blocksize);
2756                                 if (ret) {
2757                                         free_extent_buffer(next);
2758                                         return ret;
2759                                 }
2760                         }
2761                         free_extent_buffer(next);
2762                         continue;
2763                 }
2764                 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2765                 if (ret) {
2766                         free_extent_buffer(next);
2767                         return ret;
2768                 }
2769 
2770                 WARN_ON(*level <= 0);
2771                 if (path->nodes[*level-1])
2772                         free_extent_buffer(path->nodes[*level-1]);
2773                 path->nodes[*level-1] = next;
2774                 *level = btrfs_header_level(next);
2775                 path->slots[*level] = 0;
2776                 cond_resched();
2777         }
2778         WARN_ON(*level < 0);
2779         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2780 
2781         path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2782 
2783         cond_resched();
2784         return 0;
2785 }
2786 
2787 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2788                                  struct btrfs_root *root,
2789                                  struct btrfs_path *path, int *level,
2790                                  struct walk_control *wc)
2791 {
2792         struct btrfs_fs_info *fs_info = root->fs_info;
2793         u64 root_owner;
2794         int i;
2795         int slot;
2796         int ret;
2797 
2798         for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2799                 slot = path->slots[i];
2800                 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2801                         path->slots[i]++;
2802                         *level = i;
2803                         WARN_ON(*level == 0);
2804                         return 0;
2805                 } else {
2806                         struct extent_buffer *parent;
2807                         if (path->nodes[*level] == root->node)
2808                                 parent = path->nodes[*level];
2809                         else
2810                                 parent = path->nodes[*level + 1];
2811 
2812                         root_owner = btrfs_header_owner(parent);
2813                         ret = wc->process_func(root, path->nodes[*level], wc,
2814                                  btrfs_header_generation(path->nodes[*level]),
2815                                  *level);
2816                         if (ret)
2817                                 return ret;
2818 
2819                         if (wc->free) {
2820                                 struct extent_buffer *next;
2821 
2822                                 next = path->nodes[*level];
2823 
2824                                 if (trans) {
2825                                         btrfs_tree_lock(next);
2826                                         btrfs_set_lock_blocking_write(next);
2827                                         btrfs_clean_tree_block(next);
2828                                         btrfs_wait_tree_block_writeback(next);
2829                                         btrfs_tree_unlock(next);
2830                                 } else {
2831                                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2832                                                 clear_extent_buffer_dirty(next);
2833                                 }
2834 
2835                                 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2836                                 ret = btrfs_free_and_pin_reserved_extent(
2837                                                 fs_info,
2838                                                 path->nodes[*level]->start,
2839                                                 path->nodes[*level]->len);
2840                                 if (ret)
2841                                         return ret;
2842                         }
2843                         free_extent_buffer(path->nodes[*level]);
2844                         path->nodes[*level] = NULL;
2845                         *level = i + 1;
2846                 }
2847         }
2848         return 1;
2849 }
2850 
2851 /*
2852  * drop the reference count on the tree rooted at 'snap'.  This traverses
2853  * the tree freeing any blocks that have a ref count of zero after being
2854  * decremented.
2855  */
2856 static int walk_log_tree(struct btrfs_trans_handle *trans,
2857                          struct btrfs_root *log, struct walk_control *wc)
2858 {
2859         struct btrfs_fs_info *fs_info = log->fs_info;
2860         int ret = 0;
2861         int wret;
2862         int level;
2863         struct btrfs_path *path;
2864         int orig_level;
2865 
2866         path = btrfs_alloc_path();
2867         if (!path)
2868                 return -ENOMEM;
2869 
2870         level = btrfs_header_level(log->node);
2871         orig_level = level;
2872         path->nodes[level] = log->node;
2873         extent_buffer_get(log->node);
2874         path->slots[level] = 0;
2875 
2876         while (1) {
2877                 wret = walk_down_log_tree(trans, log, path, &level, wc);
2878                 if (wret > 0)
2879                         break;
2880                 if (wret < 0) {
2881                         ret = wret;
2882                         goto out;
2883                 }
2884 
2885                 wret = walk_up_log_tree(trans, log, path, &level, wc);
2886                 if (wret > 0)
2887                         break;
2888                 if (wret < 0) {
2889                         ret = wret;
2890                         goto out;
2891                 }
2892         }
2893 
2894         /* was the root node processed? if not, catch it here */
2895         if (path->nodes[orig_level]) {
2896                 ret = wc->process_func(log, path->nodes[orig_level], wc,
2897                          btrfs_header_generation(path->nodes[orig_level]),
2898                          orig_level);
2899                 if (ret)
2900                         goto out;
2901                 if (wc->free) {
2902                         struct extent_buffer *next;
2903 
2904                         next = path->nodes[orig_level];
2905 
2906                         if (trans) {
2907                                 btrfs_tree_lock(next);
2908                                 btrfs_set_lock_blocking_write(next);
2909                                 btrfs_clean_tree_block(next);
2910                                 btrfs_wait_tree_block_writeback(next);
2911                                 btrfs_tree_unlock(next);
2912                         } else {
2913                                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2914                                         clear_extent_buffer_dirty(next);
2915                         }
2916 
2917                         WARN_ON(log->root_key.objectid !=
2918                                 BTRFS_TREE_LOG_OBJECTID);
2919                         ret = btrfs_free_and_pin_reserved_extent(fs_info,
2920                                                         next->start, next->len);
2921                         if (ret)
2922                                 goto out;
2923                 }
2924         }
2925 
2926 out:
2927         btrfs_free_path(path);
2928         return ret;
2929 }
2930 
2931 /*
2932  * helper function to update the item for a given subvolumes log root
2933  * in the tree of log roots
2934  */
2935 static int update_log_root(struct btrfs_trans_handle *trans,
2936                            struct btrfs_root *log,
2937                            struct btrfs_root_item *root_item)
2938 {
2939         struct btrfs_fs_info *fs_info = log->fs_info;
2940         int ret;
2941 
2942         if (log->log_transid == 1) {
2943                 /* insert root item on the first sync */
2944                 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2945                                 &log->root_key, root_item);
2946         } else {
2947                 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2948                                 &log->root_key, root_item);
2949         }
2950         return ret;
2951 }
2952 
2953 static void wait_log_commit(struct btrfs_root *root, int transid)
2954 {
2955         DEFINE_WAIT(wait);
2956         int index = transid % 2;
2957 
2958         /*
2959          * we only allow two pending log transactions at a time,
2960          * so we know that if ours is more than 2 older than the
2961          * current transaction, we're done
2962          */
2963         for (;;) {
2964                 prepare_to_wait(&root->log_commit_wait[index],
2965                                 &wait, TASK_UNINTERRUPTIBLE);
2966 
2967                 if (!(root->log_transid_committed < transid &&
2968                       atomic_read(&root->log_commit[index])))
2969                         break;
2970 
2971                 mutex_unlock(&root->log_mutex);
2972                 schedule();
2973                 mutex_lock(&root->log_mutex);
2974         }
2975         finish_wait(&root->log_commit_wait[index], &wait);
2976 }
2977 
2978 static void wait_for_writer(struct btrfs_root *root)
2979 {
2980         DEFINE_WAIT(wait);
2981 
2982         for (;;) {
2983                 prepare_to_wait(&root->log_writer_wait, &wait,
2984                                 TASK_UNINTERRUPTIBLE);
2985                 if (!atomic_read(&root->log_writers))
2986                         break;
2987 
2988                 mutex_unlock(&root->log_mutex);
2989                 schedule();
2990                 mutex_lock(&root->log_mutex);
2991         }
2992         finish_wait(&root->log_writer_wait, &wait);
2993 }
2994 
2995 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2996                                         struct btrfs_log_ctx *ctx)
2997 {
2998         if (!ctx)
2999                 return;
3000 
3001         mutex_lock(&root->log_mutex);
3002         list_del_init(&ctx->list);
3003         mutex_unlock(&root->log_mutex);
3004 }
3005 
3006 /* 
3007  * Invoked in log mutex context, or be sure there is no other task which
3008  * can access the list.
3009  */
3010 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3011                                              int index, int error)
3012 {
3013         struct btrfs_log_ctx *ctx;
3014         struct btrfs_log_ctx *safe;
3015 
3016         list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3017                 list_del_init(&ctx->list);
3018                 ctx->log_ret = error;
3019         }
3020 
3021         INIT_LIST_HEAD(&root->log_ctxs[index]);
3022 }
3023 
3024 /*
3025  * btrfs_sync_log does sends a given tree log down to the disk and
3026  * updates the super blocks to record it.  When this call is done,
3027  * you know that any inodes previously logged are safely on disk only
3028  * if it returns 0.
3029  *
3030  * Any other return value means you need to call btrfs_commit_transaction.
3031  * Some of the edge cases for fsyncing directories that have had unlinks
3032  * or renames done in the past mean that sometimes the only safe
3033  * fsync is to commit the whole FS.  When btrfs_sync_log returns -EAGAIN,
3034  * that has happened.
3035  */
3036 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3037                    struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3038 {
3039         int index1;
3040         int index2;
3041         int mark;
3042         int ret;
3043         struct btrfs_fs_info *fs_info = root->fs_info;
3044         struct btrfs_root *log = root->log_root;
3045         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3046         struct btrfs_root_item new_root_item;
3047         int log_transid = 0;
3048         struct btrfs_log_ctx root_log_ctx;
3049         struct blk_plug plug;
3050 
3051         mutex_lock(&root->log_mutex);
3052         log_transid = ctx->log_transid;
3053         if (root->log_transid_committed >= log_transid) {
3054                 mutex_unlock(&root->log_mutex);
3055                 return ctx->log_ret;
3056         }
3057 
3058         index1 = log_transid % 2;
3059         if (atomic_read(&root->log_commit[index1])) {
3060                 wait_log_commit(root, log_transid);
3061                 mutex_unlock(&root->log_mutex);
3062                 return ctx->log_ret;
3063         }
3064         ASSERT(log_transid == root->log_transid);
3065         atomic_set(&root->log_commit[index1], 1);
3066 
3067         /* wait for previous tree log sync to complete */
3068         if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3069                 wait_log_commit(root, log_transid - 1);
3070 
3071         while (1) {
3072                 int batch = atomic_read(&root->log_batch);
3073                 /* when we're on an ssd, just kick the log commit out */
3074                 if (!btrfs_test_opt(fs_info, SSD) &&
3075                     test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3076                         mutex_unlock(&root->log_mutex);
3077                         schedule_timeout_uninterruptible(1);
3078                         mutex_lock(&root->log_mutex);
3079                 }
3080                 wait_for_writer(root);
3081                 if (batch == atomic_read(&root->log_batch))
3082                         break;
3083         }
3084 
3085         /* bail out if we need to do a full commit */
3086         if (btrfs_need_log_full_commit(trans)) {
3087                 ret = -EAGAIN;
3088                 mutex_unlock(&root->log_mutex);
3089                 goto out;
3090         }
3091 
3092         if (log_transid % 2 == 0)
3093                 mark = EXTENT_DIRTY;
3094         else
3095                 mark = EXTENT_NEW;
3096 
3097         /* we start IO on  all the marked extents here, but we don't actually
3098          * wait for them until later.
3099          */
3100         blk_start_plug(&plug);
3101         ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3102         if (ret) {
3103                 blk_finish_plug(&plug);
3104                 btrfs_abort_transaction(trans, ret);
3105                 btrfs_set_log_full_commit(trans);
3106                 mutex_unlock(&root->log_mutex);
3107                 goto out;
3108         }
3109 
3110         /*
3111          * We _must_ update under the root->log_mutex in order to make sure we
3112          * have a consistent view of the log root we are trying to commit at
3113          * this moment.
3114          *
3115          * We _must_ copy this into a local copy, because we are not holding the
3116          * log_root_tree->log_mutex yet.  This is important because when we
3117          * commit the log_root_tree we must have a consistent view of the
3118          * log_root_tree when we update the super block to point at the
3119          * log_root_tree bytenr.  If we update the log_root_tree here we'll race
3120          * with the commit and possibly point at the new block which we may not
3121          * have written out.
3122          */
3123         btrfs_set_root_node(&log->root_item, log->node);
3124         memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3125 
3126         root->log_transid++;
3127         log->log_transid = root->log_transid;
3128         root->log_start_pid = 0;
3129         /*
3130          * IO has been started, blocks of the log tree have WRITTEN flag set
3131          * in their headers. new modifications of the log will be written to
3132          * new positions. so it's safe to allow log writers to go in.
3133          */
3134         mutex_unlock(&root->log_mutex);
3135 
3136         btrfs_init_log_ctx(&root_log_ctx, NULL);
3137 
3138         mutex_lock(&log_root_tree->log_mutex);
3139         atomic_inc(&log_root_tree->log_batch);
3140         atomic_inc(&log_root_tree->log_writers);
3141 
3142         index2 = log_root_tree->log_transid % 2;
3143         list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3144         root_log_ctx.log_transid = log_root_tree->log_transid;
3145 
3146         mutex_unlock(&log_root_tree->log_mutex);
3147 
3148         mutex_lock(&log_root_tree->log_mutex);
3149 
3150         /*
3151          * Now we are safe to update the log_root_tree because we're under the
3152          * log_mutex, and we're a current writer so we're holding the commit
3153          * open until we drop the log_mutex.
3154          */
3155         ret = update_log_root(trans, log, &new_root_item);
3156 
3157         if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3158                 /* atomic_dec_and_test implies a barrier */
3159                 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3160         }
3161 
3162         if (ret) {
3163                 if (!list_empty(&root_log_ctx.list))
3164                         list_del_init(&root_log_ctx.list);
3165 
3166                 blk_finish_plug(&plug);
3167                 btrfs_set_log_full_commit(trans);
3168 
3169                 if (ret != -ENOSPC) {
3170                         btrfs_abort_transaction(trans, ret);
3171                         mutex_unlock(&log_root_tree->log_mutex);
3172                         goto out;
3173                 }
3174                 btrfs_wait_tree_log_extents(log, mark);
3175                 mutex_unlock(&log_root_tree->log_mutex);
3176                 ret = -EAGAIN;
3177                 goto out;
3178         }
3179 
3180         if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3181                 blk_finish_plug(&plug);
3182                 list_del_init(&root_log_ctx.list);
3183                 mutex_unlock(&log_root_tree->log_mutex);
3184                 ret = root_log_ctx.log_ret;
3185                 goto out;
3186         }
3187 
3188         index2 = root_log_ctx.log_transid % 2;
3189         if (atomic_read(&log_root_tree->log_commit[index2])) {
3190                 blk_finish_plug(&plug);
3191                 ret = btrfs_wait_tree_log_extents(log, mark);
3192                 wait_log_commit(log_root_tree,
3193                                 root_log_ctx.log_transid);
3194                 mutex_unlock(&log_root_tree->log_mutex);
3195                 if (!ret)
3196                         ret = root_log_ctx.log_ret;
3197                 goto out;
3198         }
3199         ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3200         atomic_set(&log_root_tree->log_commit[index2], 1);
3201 
3202         if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3203                 wait_log_commit(log_root_tree,
3204                                 root_log_ctx.log_transid - 1);
3205         }
3206 
3207         wait_for_writer(log_root_tree);
3208 
3209         /*
3210          * now that we've moved on to the tree of log tree roots,
3211          * check the full commit flag again
3212          */
3213         if (btrfs_need_log_full_commit(trans)) {
3214                 blk_finish_plug(&plug);
3215                 btrfs_wait_tree_log_extents(log, mark);
3216                 mutex_unlock(&log_root_tree->log_mutex);
3217                 ret = -EAGAIN;
3218                 goto out_wake_log_root;
3219         }
3220 
3221         ret = btrfs_write_marked_extents(fs_info,
3222                                          &log_root_tree->dirty_log_pages,
3223                                          EXTENT_DIRTY | EXTENT_NEW);
3224         blk_finish_plug(&plug);
3225         if (ret) {
3226                 btrfs_set_log_full_commit(trans);
3227                 btrfs_abort_transaction(trans, ret);
3228                 mutex_unlock(&log_root_tree->log_mutex);
3229                 goto out_wake_log_root;
3230         }
3231         ret = btrfs_wait_tree_log_extents(log, mark);
3232         if (!ret)
3233                 ret = btrfs_wait_tree_log_extents(log_root_tree,
3234                                                   EXTENT_NEW | EXTENT_DIRTY);
3235         if (ret) {
3236                 btrfs_set_log_full_commit(trans);
3237                 mutex_unlock(&log_root_tree->log_mutex);
3238                 goto out_wake_log_root;
3239         }
3240 
3241         btrfs_set_super_log_root(fs_info->super_for_commit,
3242                                  log_root_tree->node->start);
3243         btrfs_set_super_log_root_level(fs_info->super_for_commit,
3244                                        btrfs_header_level(log_root_tree->node));
3245 
3246         log_root_tree->log_transid++;
3247         mutex_unlock(&log_root_tree->log_mutex);
3248 
3249         /*
3250          * Nobody else is going to jump in and write the ctree
3251          * super here because the log_commit atomic below is protecting
3252          * us.  We must be called with a transaction handle pinning
3253          * the running transaction open, so a full commit can't hop
3254          * in and cause problems either.
3255          */
3256         ret = write_all_supers(fs_info, 1);
3257         if (ret) {
3258                 btrfs_set_log_full_commit(trans);
3259                 btrfs_abort_transaction(trans, ret);
3260                 goto out_wake_log_root;
3261         }
3262 
3263         mutex_lock(&root->log_mutex);
3264         if (root->last_log_commit < log_transid)
3265                 root->last_log_commit = log_transid;
3266         mutex_unlock(&root->log_mutex);
3267 
3268 out_wake_log_root:
3269         mutex_lock(&log_root_tree->log_mutex);
3270         btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3271 
3272         log_root_tree->log_transid_committed++;
3273         atomic_set(&log_root_tree->log_commit[index2], 0);
3274         mutex_unlock(&log_root_tree->log_mutex);
3275 
3276         /*
3277          * The barrier before waitqueue_active (in cond_wake_up) is needed so
3278          * all the updates above are seen by the woken threads. It might not be
3279          * necessary, but proving that seems to be hard.
3280          */
3281         cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3282 out:
3283         mutex_lock(&root->log_mutex);
3284         btrfs_remove_all_log_ctxs(root, index1, ret);
3285         root->log_transid_committed++;
3286         atomic_set(&root->log_commit[index1], 0);
3287         mutex_unlock(&root->log_mutex);
3288 
3289         /*
3290          * The barrier before waitqueue_active (in cond_wake_up) is needed so
3291          * all the updates above are seen by the woken threads. It might not be
3292          * necessary, but proving that seems to be hard.
3293          */
3294         cond_wake_up(&root->log_commit_wait[index1]);
3295         return ret;
3296 }
3297 
3298 static void free_log_tree(struct btrfs_trans_handle *trans,
3299                           struct btrfs_root *log)
3300 {
3301         int ret;
3302         struct walk_control wc = {
3303                 .free = 1,
3304                 .process_func = process_one_buffer
3305         };
3306 
3307         ret = walk_log_tree(trans, log, &wc);
3308         if (ret) {
3309                 if (trans)
3310                         btrfs_abort_transaction(trans, ret);
3311                 else
3312                         btrfs_handle_fs_error(log->fs_info, ret, NULL);
3313         }
3314 
3315         clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3316                           EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3317         free_extent_buffer(log->node);
3318         kfree(log);
3319 }
3320 
3321 /*
3322  * free all the extents used by the tree log.  This should be called
3323  * at commit time of the full transaction
3324  */
3325 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3326 {
3327         if (root->log_root) {
3328                 free_log_tree(trans, root->log_root);
3329                 root->log_root = NULL;
3330         }
3331         return 0;
3332 }
3333 
3334 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3335                              struct btrfs_fs_info *fs_info)
3336 {
3337         if (fs_info->log_root_tree) {
3338                 free_log_tree(trans, fs_info->log_root_tree);
3339                 fs_info->log_root_tree = NULL;
3340         }
3341         return 0;
3342 }
3343 
3344 /*
3345  * Check if an inode was logged in the current transaction. We can't always rely
3346  * on an inode's logged_trans value, because it's an in-memory only field and
3347  * therefore not persisted. This means that its value is lost if the inode gets
3348  * evicted and loaded again from disk (in which case it has a value of 0, and
3349  * certainly it is smaller then any possible transaction ID), when that happens
3350  * the full_sync flag is set in the inode's runtime flags, so on that case we
3351  * assume eviction happened and ignore the logged_trans value, assuming the
3352  * worst case, that the inode was logged before in the current transaction.
3353  */
3354 static bool inode_logged(struct btrfs_trans_handle *trans,
3355                          struct btrfs_inode *inode)
3356 {
3357         if (inode->logged_trans == trans->transid)
3358                 return true;
3359 
3360         if (inode->last_trans == trans->transid &&
3361             test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3362             !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3363                 return true;
3364 
3365         return false;
3366 }
3367 
3368 /*
3369  * If both a file and directory are logged, and unlinks or renames are
3370  * mixed in, we have a few interesting corners:
3371  *
3372  * create file X in dir Y
3373  * link file X to X.link in dir Y
3374  * fsync file X
3375  * unlink file X but leave X.link
3376  * fsync dir Y
3377  *
3378  * After a crash we would expect only X.link to exist.  But file X
3379  * didn't get fsync'd again so the log has back refs for X and X.link.
3380  *
3381  * We solve this by removing directory entries and inode backrefs from the
3382  * log when a file that was logged in the current transaction is
3383  * unlinked.  Any later fsync will include the updated log entries, and
3384  * we'll be able to reconstruct the proper directory items from backrefs.
3385  *
3386  * This optimizations allows us to avoid relogging the entire inode
3387  * or the entire directory.
3388  */
3389 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3390                                  struct btrfs_root *root,
3391                                  const char *name, int name_len,
3392                                  struct btrfs_inode *dir, u64 index)
3393 {
3394         struct btrfs_root *log;
3395         struct btrfs_dir_item *di;
3396         struct btrfs_path *path;
3397         int ret;
3398         int err = 0;
3399         int bytes_del = 0;
3400         u64 dir_ino = btrfs_ino(dir);
3401 
3402         if (!inode_logged(trans, dir))
3403                 return 0;
3404 
3405         ret = join_running_log_trans(root);
3406         if (ret)
3407                 return 0;
3408 
3409         mutex_lock(&dir->log_mutex);
3410 
3411         log = root->log_root;
3412         path = btrfs_alloc_path();
3413         if (!path) {
3414                 err = -ENOMEM;
3415                 goto out_unlock;
3416         }
3417 
3418         di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3419                                    name, name_len, -1);
3420         if (IS_ERR(di)) {
3421                 err = PTR_ERR(di);
3422                 goto fail;
3423         }
3424         if (di) {
3425                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3426                 bytes_del += name_len;
3427                 if (ret) {
3428                         err = ret;
3429                         goto fail;
3430                 }
3431         }
3432         btrfs_release_path(path);
3433         di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3434                                          index, name, name_len, -1);
3435         if (IS_ERR(di)) {
3436                 err = PTR_ERR(di);
3437                 goto fail;
3438         }
3439         if (di) {
3440                 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3441                 bytes_del += name_len;
3442                 if (ret) {
3443                         err = ret;
3444                         goto fail;
3445                 }
3446         }
3447 
3448         /* update the directory size in the log to reflect the names
3449          * we have removed
3450          */
3451         if (bytes_del) {
3452                 struct btrfs_key key;
3453 
3454                 key.objectid = dir_ino;
3455                 key.offset = 0;
3456                 key.type = BTRFS_INODE_ITEM_KEY;
3457                 btrfs_release_path(path);
3458 
3459                 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3460                 if (ret < 0) {
3461                         err = ret;
3462                         goto fail;
3463                 }
3464                 if (ret == 0) {
3465                         struct btrfs_inode_item *item;
3466                         u64 i_size;
3467 
3468                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3469                                               struct btrfs_inode_item);
3470                         i_size = btrfs_inode_size(path->nodes[0], item);
3471                         if (i_size > bytes_del)
3472                                 i_size -= bytes_del;
3473                         else
3474                                 i_size = 0;
3475                         btrfs_set_inode_size(path->nodes[0], item, i_size);
3476                         btrfs_mark_buffer_dirty(path->nodes[0]);
3477                 } else
3478                         ret = 0;
3479                 btrfs_release_path(path);
3480         }
3481 fail:
3482         btrfs_free_path(path);
3483 out_unlock:
3484         mutex_unlock(&dir->log_mutex);
3485         if (ret == -ENOSPC) {
3486                 btrfs_set_log_full_commit(trans);
3487                 ret = 0;
3488         } else if (ret < 0)
3489                 btrfs_abort_transaction(trans, ret);
3490 
3491         btrfs_end_log_trans(root);
3492 
3493         return err;
3494 }
3495 
3496 /* see comments for btrfs_del_dir_entries_in_log */
3497 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3498                                struct btrfs_root *root,
3499                                const char *name, int name_len,
3500                                struct btrfs_inode *inode, u64 dirid)
3501 {
3502         struct btrfs_root *log;
3503         u64 index;
3504         int ret;
3505 
3506         if (!inode_logged(trans, inode))
3507                 return 0;
3508 
3509         ret = join_running_log_trans(root);
3510         if (ret)
3511                 return 0;
3512         log = root->log_root;
3513         mutex_lock(&inode->log_mutex);
3514 
3515         ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3516                                   dirid, &index);
3517         mutex_unlock(&inode->log_mutex);
3518         if (ret == -ENOSPC) {
3519                 btrfs_set_log_full_commit(trans);
3520                 ret = 0;
3521         } else if (ret < 0 && ret != -ENOENT)
3522                 btrfs_abort_transaction(trans, ret);
3523         btrfs_end_log_trans(root);
3524 
3525         return ret;
3526 }
3527 
3528 /*
3529  * creates a range item in the log for 'dirid'.  first_offset and
3530  * last_offset tell us which parts of the key space the log should
3531  * be considered authoritative for.
3532  */
3533 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3534                                        struct btrfs_root *log,
3535                                        struct btrfs_path *path,
3536                                        int key_type, u64 dirid,
3537                                        u64 first_offset, u64 last_offset)
3538 {
3539         int ret;
3540         struct btrfs_key key;
3541         struct btrfs_dir_log_item *item;
3542 
3543         key.objectid = dirid;
3544         key.offset = first_offset;
3545         if (key_type == BTRFS_DIR_ITEM_KEY)
3546                 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3547         else
3548                 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3549         ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3550         if (ret)
3551                 return ret;
3552 
3553         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3554                               struct btrfs_dir_log_item);
3555         btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3556         btrfs_mark_buffer_dirty(path->nodes[0]);
3557         btrfs_release_path(path);
3558         return 0;
3559 }
3560 
3561 /*
3562  * log all the items included in the current transaction for a given
3563  * directory.  This also creates the range items in the log tree required
3564  * to replay anything deleted before the fsync
3565  */
3566 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3567                           struct btrfs_root *root, struct btrfs_inode *inode,
3568                           struct btrfs_path *path,
3569                           struct btrfs_path *dst_path, int key_type,
3570                           struct btrfs_log_ctx *ctx,
3571                           u64 min_offset, u64 *last_offset_ret)
3572 {
3573         struct btrfs_key min_key;
3574         struct btrfs_root *log = root->log_root;
3575         struct extent_buffer *src;
3576         int err = 0;
3577         int ret;
3578         int i;
3579         int nritems;
3580         u64 first_offset = min_offset;
3581         u64 last_offset = (u64)-1;
3582         u64 ino = btrfs_ino(inode);
3583 
3584         log = root->log_root;
3585 
3586         min_key.objectid = ino;
3587         min_key.type = key_type;
3588         min_key.offset = min_offset;
3589 
3590         ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3591 
3592         /*
3593          * we didn't find anything from this transaction, see if there
3594          * is anything at all
3595          */
3596         if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3597                 min_key.objectid = ino;
3598                 min_key.type = key_type;
3599                 min_key.offset = (u64)-1;
3600                 btrfs_release_path(path);
3601                 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3602                 if (ret < 0) {
3603                         btrfs_release_path(path);
3604                         return ret;
3605                 }
3606                 ret = btrfs_previous_item(root, path, ino, key_type);
3607 
3608                 /* if ret == 0 there are items for this type,
3609                  * create a range to tell us the last key of this type.
3610                  * otherwise, there are no items in this directory after
3611                  * *min_offset, and we create a range to indicate that.
3612                  */
3613                 if (ret == 0) {
3614                         struct btrfs_key tmp;
3615                         btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3616                                               path->slots[0]);
3617                         if (key_type == tmp.type)
3618                                 first_offset = max(min_offset, tmp.offset) + 1;
3619                 }
3620                 goto done;
3621         }
3622 
3623         /* go backward to find any previous key */
3624         ret = btrfs_previous_item(root, path, ino, key_type);
3625         if (ret == 0) {
3626                 struct btrfs_key tmp;
3627                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3628                 if (key_type == tmp.type) {
3629                         first_offset = tmp.offset;
3630                         ret = overwrite_item(trans, log, dst_path,
3631                                              path->nodes[0], path->slots[0],
3632                                              &tmp);
3633                         if (ret) {
3634                                 err = ret;
3635                                 goto done;
3636                         }
3637                 }
3638         }
3639         btrfs_release_path(path);
3640 
3641         /*
3642          * Find the first key from this transaction again.  See the note for
3643          * log_new_dir_dentries, if we're logging a directory recursively we
3644          * won't be holding its i_mutex, which means we can modify the directory
3645          * while we're logging it.  If we remove an entry between our first
3646          * search and this search we'll not find the key again and can just
3647          * bail.
3648          */
3649         ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3650         if (ret != 0)
3651                 goto done;
3652 
3653         /*
3654          * we have a block from this transaction, log every item in it
3655          * from our directory
3656          */
3657         while (1) {
3658                 struct btrfs_key tmp;
3659                 src = path->nodes[0];
3660                 nritems = btrfs_header_nritems(src);
3661                 for (i = path->slots[0]; i < nritems; i++) {
3662                         struct btrfs_dir_item *di;
3663 
3664                         btrfs_item_key_to_cpu(src, &min_key, i);
3665 
3666                         if (min_key.objectid != ino || min_key.type != key_type)
3667                                 goto done;
3668                         ret = overwrite_item(trans, log, dst_path, src, i,
3669                                              &min_key);
3670                         if (ret) {
3671                                 err = ret;
3672                                 goto done;
3673                         }
3674 
3675                         /*
3676                          * We must make sure that when we log a directory entry,
3677                          * the corresponding inode, after log replay, has a
3678                          * matching link count. For example:
3679                          *
3680                          * touch foo
3681                          * mkdir mydir
3682                          * sync
3683                          * ln foo mydir/bar
3684                          * xfs_io -c "fsync" mydir
3685                          * <crash>
3686                          * <mount fs and log replay>
3687                          *
3688                          * Would result in a fsync log that when replayed, our
3689                          * file inode would have a link count of 1, but we get
3690                          * two directory entries pointing to the same inode.
3691                          * After removing one of the names, it would not be
3692                          * possible to remove the other name, which resulted
3693                          * always in stale file handle errors, and would not
3694                          * be possible to rmdir the parent directory, since
3695                          * its i_size could never decrement to the value
3696                          * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3697                          */
3698                         di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3699                         btrfs_dir_item_key_to_cpu(src, di, &tmp);
3700                         if (ctx &&
3701                             (btrfs_dir_transid(src, di) == trans->transid ||
3702                              btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3703                             tmp.type != BTRFS_ROOT_ITEM_KEY)
3704                                 ctx->log_new_dentries = true;
3705                 }
3706                 path->slots[0] = nritems;
3707 
3708                 /*
3709                  * look ahead to the next item and see if it is also
3710                  * from this directory and from this transaction
3711                  */
3712                 ret = btrfs_next_leaf(root, path);
3713                 if (ret) {
3714                         if (ret == 1)
3715                                 last_offset = (u64)-1;
3716                         else
3717                                 err = ret;
3718                         goto done;
3719                 }
3720                 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3721                 if (tmp.objectid != ino || tmp.type != key_type) {
3722                         last_offset = (u64)-1;
3723                         goto done;
3724                 }
3725                 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3726                         ret = overwrite_item(trans, log, dst_path,
3727                                              path->nodes[0], path->slots[0],
3728                                              &tmp);
3729                         if (ret)
3730                                 err = ret;
3731                         else
3732                                 last_offset = tmp.offset;
3733                         goto done;
3734                 }
3735         }
3736 done:
3737         btrfs_release_path(path);
3738         btrfs_release_path(dst_path);
3739 
3740         if (err == 0) {
3741                 *last_offset_ret = last_offset;
3742                 /*
3743                  * insert the log range keys to indicate where the log
3744                  * is valid
3745                  */
3746                 ret = insert_dir_log_key(trans, log, path, key_type,
3747                                          ino, first_offset, last_offset);
3748                 if (ret)
3749                         err = ret;
3750         }
3751         return err;
3752 }
3753 
3754 /*
3755  * logging directories is very similar to logging inodes, We find all the items
3756  * from the current transaction and write them to the log.
3757  *
3758  * The recovery code scans the directory in the subvolume, and if it finds a
3759  * key in the range logged that is not present in the log tree, then it means
3760  * that dir entry was unlinked during the transaction.
3761  *
3762  * In order for that scan to work, we must include one key smaller than
3763  * the smallest logged by this transaction and one key larger than the largest
3764  * key logged by this transaction.
3765  */
3766 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3767                           struct btrfs_root *root, struct btrfs_inode *inode,
3768                           struct btrfs_path *path,
3769                           struct btrfs_path *dst_path,
3770                           struct btrfs_log_ctx *ctx)
3771 {
3772         u64 min_key;
3773         u64 max_key;
3774         int ret;
3775         int key_type = BTRFS_DIR_ITEM_KEY;
3776 
3777 again:
3778         min_key = 0;
3779         max_key = 0;
3780         while (1) {
3781                 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3782                                 ctx, min_key, &max_key);
3783                 if (ret)
3784                         return ret;
3785                 if (max_key == (u64)-1)
3786                         break;
3787                 min_key = max_key + 1;
3788         }
3789 
3790         if (key_type == BTRFS_DIR_ITEM_KEY) {
3791                 key_type = BTRFS_DIR_INDEX_KEY;
3792                 goto again;
3793         }
3794         return 0;
3795 }
3796 
3797 /*
3798  * a helper function to drop items from the log before we relog an
3799  * inode.  max_key_type indicates the highest item type to remove.
3800  * This cannot be run for file data extents because it does not
3801  * free the extents they point to.
3802  */
3803 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3804                                   struct btrfs_root *log,
3805                                   struct btrfs_path *path,
3806                                   u64 objectid, int max_key_type)
3807 {
3808         int ret;
3809         struct btrfs_key key;
3810         struct btrfs_key found_key;
3811         int start_slot;
3812 
3813         key.objectid = objectid;
3814         key.type = max_key_type;
3815         key.offset = (u64)-1;
3816 
3817         while (1) {
3818                 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3819                 BUG_ON(ret == 0); /* Logic error */
3820                 if (ret < 0)
3821                         break;
3822 
3823                 if (path->slots[0] == 0)
3824                         break;
3825 
3826                 path->slots[0]--;
3827                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3828                                       path->slots[0]);
3829 
3830                 if (found_key.objectid != objectid)
3831                         break;
3832 
3833                 found_key.offset = 0;
3834                 found_key.type = 0;
3835                 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3836                                        &start_slot);
3837                 if (ret < 0)
3838                         break;
3839 
3840                 ret = btrfs_del_items(trans, log, path, start_slot,
3841                                       path->slots[0] - start_slot + 1);
3842                 /*
3843                  * If start slot isn't 0 then we don't need to re-search, we've
3844                  * found the last guy with the objectid in this tree.
3845                  */
3846                 if (ret || start_slot != 0)
3847                         break;
3848                 btrfs_release_path(path);
3849         }
3850         btrfs_release_path(path);
3851         if (ret > 0)
3852                 ret = 0;
3853         return ret;
3854 }
3855 
3856 static void fill_inode_item(struct btrfs_trans_handle *trans,
3857                             struct extent_buffer *leaf,
3858                             struct btrfs_inode_item *item,
3859                             struct inode *inode, int log_inode_only,
3860                             u64 logged_isize)
3861 {
3862         struct btrfs_map_token token;
3863 
3864         btrfs_init_map_token(&token, leaf);
3865 
3866         if (log_inode_only) {
3867                 /* set the generation to zero so the recover code
3868                  * can tell the difference between an logging
3869                  * just to say 'this inode exists' and a logging
3870                  * to say 'update this inode with these values'
3871                  */
3872                 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3873                 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3874         } else {
3875                 btrfs_set_token_inode_generation(leaf, item,
3876                                                  BTRFS_I(inode)->generation,
3877                                                  &token);
3878                 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3879         }
3880 
3881         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3882         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3883         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3884         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3885 
3886         btrfs_set_token_timespec_sec(leaf, &item->atime,
3887                                      inode->i_atime.tv_sec, &token);
3888         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3889                                       inode->i_atime.tv_nsec, &token);
3890 
3891         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3892                                      inode->i_mtime.tv_sec, &token);
3893         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3894                                       inode->i_mtime.tv_nsec, &token);
3895 
3896         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3897                                      inode->i_ctime.tv_sec, &token);
3898         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3899                                       inode->i_ctime.tv_nsec, &token);
3900 
3901         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3902                                      &token);
3903 
3904         btrfs_set_token_inode_sequence(leaf, item,
3905                                        inode_peek_iversion(inode), &token);
3906         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3907         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3908         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3909         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3910 }
3911 
3912 static int log_inode_item(struct btrfs_trans_handle *trans,
3913                           struct btrfs_root *log, struct btrfs_path *path,
3914                           struct btrfs_inode *inode)
3915 {
3916         struct btrfs_inode_item *inode_item;
3917         int ret;
3918 
3919         ret = btrfs_insert_empty_item(trans, log, path,
3920                                       &inode->location, sizeof(*inode_item));
3921         if (ret && ret != -EEXIST)
3922                 return ret;
3923         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3924                                     struct btrfs_inode_item);
3925         fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3926                         0, 0);
3927         btrfs_release_path(path);
3928         return 0;
3929 }
3930 
3931 static int log_csums(struct btrfs_trans_handle *trans,
3932                      struct btrfs_root *log_root,
3933                      struct btrfs_ordered_sum *sums)
3934 {
3935         int ret;
3936 
3937         /*
3938          * Due to extent cloning, we might have logged a csum item that covers a
3939          * subrange of a cloned extent, and later we can end up logging a csum
3940          * item for a larger subrange of the same extent or the entire range.
3941          * This would leave csum items in the log tree that cover the same range
3942          * and break the searches for checksums in the log tree, resulting in
3943          * some checksums missing in the fs/subvolume tree. So just delete (or
3944          * trim and adjust) any existing csum items in the log for this range.
3945          */
3946         ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3947         if (ret)
3948                 return ret;
3949 
3950         return btrfs_csum_file_blocks(trans, log_root, sums);
3951 }
3952 
3953 static noinline int copy_items(struct btrfs_trans_handle *trans,
3954                                struct btrfs_inode *inode,
3955                                struct btrfs_path *dst_path,
3956                                struct btrfs_path *src_path,
3957                                int start_slot, int nr, int inode_only,
3958                                u64 logged_isize)
3959 {
3960         struct btrfs_fs_info *fs_info = trans->fs_info;
3961         unsigned long src_offset;
3962         unsigned long dst_offset;
3963         struct btrfs_root *log = inode->root->log_root;
3964         struct btrfs_file_extent_item *extent;
3965         struct btrfs_inode_item *inode_item;
3966         struct extent_buffer *src = src_path->nodes[0];
3967         int ret;
3968         struct btrfs_key *ins_keys;
3969         u32 *ins_sizes;
3970         char *ins_data;
3971         int i;
3972         struct list_head ordered_sums;
3973         int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3974 
3975         INIT_LIST_HEAD(&ordered_sums);
3976 
3977         ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3978                            nr * sizeof(u32), GFP_NOFS);
3979         if (!ins_data)
3980                 return -ENOMEM;
3981 
3982         ins_sizes = (u32 *)ins_data;
3983         ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3984 
3985         for (i = 0; i < nr; i++) {
3986                 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3987                 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3988         }
3989         ret = btrfs_insert_empty_items(trans, log, dst_path,
3990                                        ins_keys, ins_sizes, nr);
3991         if (ret) {
3992                 kfree(ins_data);
3993                 return ret;
3994         }
3995 
3996         for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3997                 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3998                                                    dst_path->slots[0]);
3999 
4000                 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4001 
4002                 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4003                         inode_item = btrfs_item_ptr(dst_path->nodes[0],
4004                                                     dst_path->slots[0],
4005                                                     struct btrfs_inode_item);
4006                         fill_inode_item(trans, dst_path->nodes[0], inode_item,
4007                                         &inode->vfs_inode,
4008                                         inode_only == LOG_INODE_EXISTS,
4009                                         logged_isize);
4010                 } else {
4011                         copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4012                                            src_offset, ins_sizes[i]);
4013                 }
4014 
4015                 /* take a reference on file data extents so that truncates
4016                  * or deletes of this inode don't have to relog the inode
4017                  * again
4018                  */
4019                 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4020                     !skip_csum) {
4021                         int found_type;
4022                         extent = btrfs_item_ptr(src, start_slot + i,
4023                                                 struct btrfs_file_extent_item);
4024 
4025                         if (btrfs_file_extent_generation(src, extent) < trans->transid)
4026                                 continue;
4027 
4028                         found_type = btrfs_file_extent_type(src, extent);
4029                         if (found_type == BTRFS_FILE_EXTENT_REG) {
4030                                 u64 ds, dl, cs, cl;
4031                                 ds = btrfs_file_extent_disk_bytenr(src,
4032                                                                 extent);
4033                                 /* ds == 0 is a hole */
4034                                 if (ds == 0)
4035                                         continue;
4036 
4037                                 dl = btrfs_file_extent_disk_num_bytes(src,
4038                                                                 extent);
4039                                 cs = btrfs_file_extent_offset(src, extent);
4040                                 cl = btrfs_file_extent_num_bytes(src,
4041                                                                 extent);
4042                                 if (btrfs_file_extent_compression(src,
4043                                                                   extent)) {
4044                                         cs = 0;
4045                                         cl = dl;
4046                                 }
4047 
4048                                 ret = btrfs_lookup_csums_range(
4049                                                 fs_info->csum_root,
4050                                                 ds + cs, ds + cs + cl - 1,
4051                                                 &ordered_sums, 0);
4052                                 if (ret) {
4053                                         btrfs_release_path(dst_path);
4054                                         kfree(ins_data);
4055                                         return ret;
4056                                 }
4057                         }
4058                 }
4059         }
4060 
4061         btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4062         btrfs_release_path(dst_path);
4063         kfree(ins_data);
4064 
4065         /*
4066          * we have to do this after the loop above to avoid changing the
4067          * log tree while trying to change the log tree.
4068          */
4069         ret = 0;
4070         while (!list_empty(&ordered_sums)) {
4071                 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4072                                                    struct btrfs_ordered_sum,
4073                                                    list);
4074                 if (!ret)
4075                         ret = log_csums(trans, log, sums);
4076                 list_del(&sums->list);
4077                 kfree(sums);
4078         }
4079 
4080         return ret;
4081 }
4082 
4083 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4084 {
4085         struct extent_map *em1, *em2;
4086 
4087         em1 = list_entry(a, struct extent_map, list);
4088         em2 = list_entry(b, struct extent_map, list);
4089 
4090         if (em1->start < em2->start)
4091                 return -1;
4092         else if (em1->start > em2->start)
4093                 return 1;
4094         return 0;
4095 }
4096 
4097 static int log_extent_csums(struct btrfs_trans_handle *trans,
4098                             struct btrfs_inode *inode,
4099                             struct btrfs_root *log_root,
4100                             const struct extent_map *em)
4101 {
4102         u64 csum_offset;
4103         u64 csum_len;
4104         LIST_HEAD(ordered_sums);
4105         int ret = 0;
4106 
4107         if (inode->flags & BTRFS_INODE_NODATASUM ||
4108             test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4109             em->block_start == EXTENT_MAP_HOLE)
4110                 return 0;
4111 
4112         /* If we're compressed we have to save the entire range of csums. */
4113         if (em->compress_type) {
4114                 csum_offset = 0;
4115                 csum_len = max(em->block_len, em->orig_block_len);
4116         } else {
4117                 csum_offset = em->mod_start - em->start;
4118                 csum_len = em->mod_len;
4119         }
4120 
4121         /* block start is already adjusted for the file extent offset. */
4122         ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4123                                        em->block_start + csum_offset,
4124                                        em->block_start + csum_offset +
4125                                        csum_len - 1, &ordered_sums, 0);
4126         if (ret)
4127                 return ret;
4128 
4129         while (!list_empty(&ordered_sums)) {
4130                 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4131                                                    struct btrfs_ordered_sum,
4132                                                    list);
4133                 if (!ret)
4134                         ret = log_csums(trans, log_root, sums);
4135                 list_del(&sums->list);
4136                 kfree(sums);
4137         }
4138 
4139         return ret;
4140 }
4141 
4142 static int log_one_extent(struct btrfs_trans_handle *trans,
4143                           struct btrfs_inode *inode, struct btrfs_root *root,
4144                           const struct extent_map *em,
4145                           struct btrfs_path *path,
4146                           struct btrfs_log_ctx *ctx)
4147 {
4148         struct btrfs_root *log = root->log_root;
4149         struct btrfs_file_extent_item *fi;
4150         struct extent_buffer *leaf;
4151         struct btrfs_map_token token;
4152         struct btrfs_key key;
4153         u64 extent_offset = em->start - em->orig_start;
4154         u64 block_len;
4155         int ret;
4156         int extent_inserted = 0;
4157 
4158         ret = log_extent_csums(trans, inode, log, em);
4159         if (ret)
4160                 return ret;
4161 
4162         ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4163                                    em->start + em->len, NULL, 0, 1,
4164                                    sizeof(*fi), &extent_inserted);
4165         if (ret)
4166                 return ret;
4167 
4168         if (!extent_inserted) {
4169                 key.objectid = btrfs_ino(inode);
4170                 key.type = BTRFS_EXTENT_DATA_KEY;
4171                 key.offset = em->start;
4172 
4173                 ret = btrfs_insert_empty_item(trans, log, path, &key,
4174                                               sizeof(*fi));
4175                 if (ret)
4176                         return ret;
4177         }
4178         leaf = path->nodes[0];
4179         btrfs_init_map_token(&token, leaf);
4180         fi = btrfs_item_ptr(leaf, path->slots[0],
4181                             struct btrfs_file_extent_item);
4182 
4183         btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4184                                                &token);
4185         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4186                 btrfs_set_token_file_extent_type(leaf, fi,
4187                                                  BTRFS_FILE_EXTENT_PREALLOC,
4188                                                  &token);
4189         else
4190                 btrfs_set_token_file_extent_type(leaf, fi,
4191                                                  BTRFS_FILE_EXTENT_REG,
4192                                                  &token);
4193 
4194         block_len = max(em->block_len, em->orig_block_len);
4195         if (em->compress_type != BTRFS_COMPRESS_NONE) {
4196                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4197                                                         em->block_start,
4198                                                         &token);
4199                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4200                                                            &token);
4201         } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4202                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4203                                                         em->block_start -
4204                                                         extent_offset, &token);
4205                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4206                                                            &token);
4207         } else {
4208                 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4209                 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4210                                                            &token);
4211         }
4212 
4213         btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4214         btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4215         btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4216         btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4217                                                 &token);
4218         btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4219         btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4220         btrfs_mark_buffer_dirty(leaf);
4221 
4222         btrfs_release_path(path);
4223 
4224         return ret;
4225 }
4226 
4227 /*
4228  * Log all prealloc extents beyond the inode's i_size to make sure we do not
4229  * lose them after doing a fast fsync and replaying the log. We scan the
4230  * subvolume's root instead of iterating the inode's extent map tree because
4231  * otherwise we can log incorrect extent items based on extent map conversion.
4232  * That can happen due to the fact that extent maps are merged when they
4233  * are not in the extent map tree's list of modified extents.
4234  */
4235 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4236                                       struct btrfs_inode *inode,
4237                                       struct btrfs_path *path)
4238 {
4239         struct btrfs_root *root = inode->root;
4240         struct btrfs_key key;
4241         const u64 i_size = i_size_read(&inode->vfs_inode);
4242         const u64 ino = btrfs_ino(inode);
4243         struct btrfs_path *dst_path = NULL;
4244         bool dropped_extents = false;
4245         u64 truncate_offset = i_size;
4246         struct extent_buffer *leaf;
4247         int slot;
4248         int ins_nr = 0;
4249         int start_slot;
4250         int ret;
4251 
4252         if (!(inode->flags & BTRFS_INODE_PREALLOC))
4253                 return 0;
4254 
4255         key.objectid = ino;
4256         key.type = BTRFS_EXTENT_DATA_KEY;
4257         key.offset = i_size;
4258         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4259         if (ret < 0)
4260                 goto out;
4261 
4262         /*
4263          * We must check if there is a prealloc extent that starts before the
4264          * i_size and crosses the i_size boundary. This is to ensure later we
4265          * truncate down to the end of that extent and not to the i_size, as
4266          * otherwise we end up losing part of the prealloc extent after a log
4267          * replay and with an implicit hole if there is another prealloc extent
4268          * that starts at an offset beyond i_size.
4269          */
4270         ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4271         if (ret < 0)
4272                 goto out;
4273 
4274         if (ret == 0) {
4275                 struct btrfs_file_extent_item *ei;
4276 
4277                 leaf = path->nodes[0];
4278                 slot = path->slots[0];
4279                 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4280 
4281                 if (btrfs_file_extent_type(leaf, ei) ==
4282                     BTRFS_FILE_EXTENT_PREALLOC) {
4283                         u64 extent_end;
4284 
4285                         btrfs_item_key_to_cpu(leaf, &key, slot);
4286                         extent_end = key.offset +
4287                                 btrfs_file_extent_num_bytes(leaf, ei);
4288 
4289                         if (extent_end > i_size)
4290                                 truncate_offset = extent_end;
4291                 }
4292         } else {
4293                 ret = 0;
4294         }
4295 
4296         while (true) {
4297                 leaf = path->nodes[0];
4298                 slot = path->slots[0];
4299 
4300                 if (slot >= btrfs_header_nritems(leaf)) {
4301                         if (ins_nr > 0) {
4302                                 ret = copy_items(trans, inode, dst_path, path,
4303                                                  start_slot, ins_nr, 1, 0);
4304                                 if (ret < 0)
4305                                         goto out;
4306                                 ins_nr = 0;
4307                         }
4308                         ret = btrfs_next_leaf(root, path);
4309                         if (ret < 0)
4310                                 goto out;
4311                         if (ret > 0) {
4312                                 ret = 0;
4313                                 break;
4314                         }
4315                         continue;
4316                 }
4317 
4318                 btrfs_item_key_to_cpu(leaf, &key, slot);
4319                 if (key.objectid > ino)
4320                         break;
4321                 if (WARN_ON_ONCE(key.objectid < ino) ||
4322                     key.type < BTRFS_EXTENT_DATA_KEY ||
4323                     key.offset < i_size) {
4324                         path->slots[0]++;
4325                         continue;
4326                 }
4327                 if (!dropped_extents) {
4328                         /*
4329                          * Avoid logging extent items logged in past fsync calls
4330                          * and leading to duplicate keys in the log tree.
4331                          */
4332                         do {
4333                                 ret = btrfs_truncate_inode_items(trans,
4334                                                          root->log_root,
4335                                                          &inode->vfs_inode,
4336                                                          truncate_offset,
4337                                                          BTRFS_EXTENT_DATA_KEY);
4338                         } while (ret == -EAGAIN);
4339                         if (ret)
4340                                 goto out;
4341                         dropped_extents = true;
4342                 }
4343                 if (ins_nr == 0)
4344                         start_slot = slot;
4345                 ins_nr++;
4346                 path->slots[0]++;
4347                 if (!dst_path) {
4348                         dst_path = btrfs_alloc_path();
4349                         if (!dst_path) {
4350                                 ret = -ENOMEM;
4351                                 goto out;
4352                         }
4353                 }
4354         }
4355         if (ins_nr > 0) {
4356                 ret = copy_items(trans, inode, dst_path, path,
4357                                  start_slot, ins_nr, 1, 0);
4358                 if (ret > 0)
4359                         ret = 0;
4360         }
4361 out:
4362         btrfs_release_path(path);
4363         btrfs_free_path(dst_path);
4364         return ret;
4365 }
4366 
4367 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4368                                      struct btrfs_root *root,
4369                                      struct btrfs_inode *inode,
4370                                      struct btrfs_path *path,
4371                                      struct btrfs_log_ctx *ctx,
4372                                      const u64 start,
4373                                      const u64 end)
4374 {
4375         struct extent_map *em, *n;
4376         struct list_head extents;
4377         struct extent_map_tree *tree = &inode->extent_tree;
4378         u64 test_gen;
4379         int ret = 0;
4380         int num = 0;
4381 
4382         INIT_LIST_HEAD(&extents);
4383 
4384         write_lock(&tree->lock);
4385         test_gen = root->fs_info->last_trans_committed;
4386 
4387         list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4388                 /*
4389                  * Skip extents outside our logging range. It's important to do
4390                  * it for correctness because if we don't ignore them, we may
4391                  * log them before their ordered extent completes, and therefore
4392                  * we could log them without logging their respective checksums
4393                  * (the checksum items are added to the csum tree at the very
4394                  * end of btrfs_finish_ordered_io()). Also leave such extents
4395                  * outside of our range in the list, since we may have another
4396                  * ranged fsync in the near future that needs them. If an extent
4397                  * outside our range corresponds to a hole, log it to avoid
4398                  * leaving gaps between extents (fsck will complain when we are
4399                  * not using the NO_HOLES feature).
4400                  */
4401                 if ((em->start > end || em->start + em->len <= start) &&
4402                     em->block_start != EXTENT_MAP_HOLE)
4403                         continue;
4404 
4405                 list_del_init(&em->list);
4406                 /*
4407                  * Just an arbitrary number, this can be really CPU intensive
4408                  * once we start getting a lot of extents, and really once we
4409                  * have a bunch of extents we just want to commit since it will
4410                  * be faster.
4411                  */
4412                 if (++num > 32768) {
4413                         list_del_init(&tree->modified_extents);
4414                         ret = -EFBIG;
4415                         goto process;
4416                 }
4417 
4418                 if (em->generation <= test_gen)
4419                         continue;
4420 
4421                 /* We log prealloc extents beyond eof later. */
4422                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4423                     em->start >= i_size_read(&inode->vfs_inode))
4424                         continue;
4425 
4426                 /* Need a ref to keep it from getting evicted from cache */
4427                 refcount_inc(&em->refs);
4428                 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4429                 list_add_tail(&em->list, &extents);
4430                 num++;
4431         }
4432 
4433         list_sort(NULL, &extents, extent_cmp);
4434 process:
4435         while (!list_empty(&extents)) {
4436                 em = list_entry(extents.next, struct extent_map, list);
4437 
4438                 list_del_init(&em->list);
4439 
4440                 /*
4441                  * If we had an error we just need to delete everybody from our
4442                  * private list.
4443                  */
4444                 if (ret) {
4445                         clear_em_logging(tree, em);
4446                         free_extent_map(em);
4447                         continue;
4448                 }
4449 
4450                 write_unlock(&tree->lock);
4451 
4452                 ret = log_one_extent(trans, inode, root, em, path, ctx);
4453                 write_lock(&tree->lock);
4454                 clear_em_logging(tree, em);
4455                 free_extent_map(em);
4456         }
4457         WARN_ON(!list_empty(&extents));
4458         write_unlock(&tree->lock);
4459 
4460         btrfs_release_path(path);
4461         if (!ret)
4462                 ret = btrfs_log_prealloc_extents(trans, inode, path);
4463 
4464         return ret;
4465 }
4466 
4467 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4468                              struct btrfs_path *path, u64 *size_ret)
4469 {
4470         struct btrfs_key key;
4471         int ret;
4472 
4473         key.objectid = btrfs_ino(inode);
4474         key.type = BTRFS_INODE_ITEM_KEY;
4475         key.offset = 0;
4476 
4477         ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4478         if (ret < 0) {
4479                 return ret;
4480         } else if (ret > 0) {
4481                 *size_ret = 0;
4482         } else {
4483                 struct btrfs_inode_item *item;
4484 
4485                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4486                                       struct btrfs_inode_item);
4487                 *size_ret = btrfs_inode_size(path->nodes[0], item);
4488                 /*
4489                  * If the in-memory inode's i_size is smaller then the inode
4490                  * size stored in the btree, return the inode's i_size, so
4491                  * that we get a correct inode size after replaying the log
4492                  * when before a power failure we had a shrinking truncate
4493                  * followed by addition of a new name (rename / new hard link).
4494                  * Otherwise return the inode size from the btree, to avoid
4495                  * data loss when replaying a log due to previously doing a
4496                  * write that expands the inode's size and logging a new name
4497                  * immediately after.
4498                  */
4499                 if (*size_ret > inode->vfs_inode.i_size)
4500                         *size_ret = inode->vfs_inode.i_size;
4501         }
4502 
4503         btrfs_release_path(path);
4504         return 0;
4505 }
4506 
4507 /*
4508  * At the moment we always log all xattrs. This is to figure out at log replay
4509  * time which xattrs must have their deletion replayed. If a xattr is missing
4510  * in the log tree and exists in the fs/subvol tree, we delete it. This is
4511  * because if a xattr is deleted, the inode is fsynced and a power failure
4512  * happens, causing the log to be replayed the next time the fs is mounted,
4513  * we want the xattr to not exist anymore (same behaviour as other filesystems
4514  * with a journal, ext3/4, xfs, f2fs, etc).
4515  */
4516 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4517                                 struct btrfs_root *root,
4518                                 struct btrfs_inode *inode,
4519                                 struct btrfs_path *path,
4520                                 struct btrfs_path *dst_path)
4521 {
4522         int ret;
4523         struct btrfs_key key;
4524         const u64 ino = btrfs_ino(inode);
4525         int ins_nr = 0;
4526         int start_slot = 0;
4527 
4528         key.objectid = ino;
4529         key.type = BTRFS_XATTR_ITEM_KEY;
4530         key.offset = 0;
4531 
4532         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4533         if (ret < 0)
4534                 return ret;
4535 
4536         while (true) {
4537                 int slot = path->slots[0];
4538                 struct extent_buffer *leaf = path->nodes[0];
4539                 int nritems = btrfs_header_nritems(leaf);
4540 
4541                 if (slot >= nritems) {
4542                         if (ins_nr > 0) {
4543                                 ret = copy_items(trans, inode, dst_path, path,
4544                                                  start_slot, ins_nr, 1, 0);
4545                                 if (ret < 0)
4546                                         return ret;
4547                                 ins_nr = 0;
4548                         }
4549                         ret = btrfs_next_leaf(root, path);
4550                         if (ret < 0)
4551                                 return ret;
4552                         else if (ret > 0)
4553                                 break;
4554                         continue;
4555                 }
4556 
4557                 btrfs_item_key_to_cpu(leaf, &key, slot);
4558                 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4559                         break;
4560 
4561                 if (ins_nr == 0)
4562                         start_slot = slot;
4563                 ins_nr++;
4564                 path->slots[0]++;
4565                 cond_resched();
4566         }
4567         if (ins_nr > 0) {
4568                 ret = copy_items(trans, inode, dst_path, path,
4569                                  start_slot, ins_nr, 1, 0);
4570                 if (ret < 0)
4571                         return ret;
4572         }
4573 
4574         return 0;
4575 }
4576 
4577 /*
4578  * When using the NO_HOLES feature if we punched a hole that causes the
4579  * deletion of entire leafs or all the extent items of the first leaf (the one
4580  * that contains the inode item and references) we may end up not processing
4581  * any extents, because there are no leafs with a generation matching the
4582  * current transaction that have extent items for our inode. So we need to find
4583  * if any holes exist and then log them. We also need to log holes after any
4584  * truncate operation that changes the inode's size.
4585  */
4586 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4587                            struct btrfs_root *root,
4588                            struct btrfs_inode *inode,
4589                            struct btrfs_path *path)
4590 {
4591         struct btrfs_fs_info *fs_info = root->fs_info;
4592         struct btrfs_key key;
4593         const u64 ino = btrfs_ino(inode);
4594         const u64 i_size = i_size_read(&inode->vfs_inode);
4595         u64 prev_extent_end = 0;
4596         int ret;
4597 
4598         if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4599                 return 0;
4600 
4601         key.objectid = ino;
4602         key.type = BTRFS_EXTENT_DATA_KEY;
4603         key.offset = 0;
4604 
4605         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4606         if (ret < 0)
4607                 return ret;
4608 
4609         while (true) {
4610                 struct btrfs_file_extent_item *extent;
4611                 struct extent_buffer *leaf = path->nodes[0];
4612                 u64 len;
4613 
4614                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4615                         ret = btrfs_next_leaf(root, path);
4616                         if (ret < 0)
4617                                 return ret;
4618                         if (ret > 0) {
4619                                 ret = 0;
4620                                 break;
4621                         }
4622                         leaf = path->nodes[0];
4623                 }
4624 
4625                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4626                 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4627                         break;
4628 
4629                 /* We have a hole, log it. */
4630                 if (prev_extent_end < key.offset) {
4631                         const u64 hole_len = key.offset - prev_extent_end;
4632 
4633                         /*
4634                          * Release the path to avoid deadlocks with other code
4635                          * paths that search the root while holding locks on
4636                          * leafs from the log root.
4637                          */
4638                         btrfs_release_path(path);
4639                         ret = btrfs_insert_file_extent(trans, root->log_root,
4640                                                        ino, prev_extent_end, 0,
4641                                                        0, hole_len, 0, hole_len,
4642                                                        0, 0, 0);
4643                         if (ret < 0)
4644                                 return ret;
4645 
4646                         /*
4647                          * Search for the same key again in the root. Since it's
4648                          * an extent item and we are holding the inode lock, the
4649                          * key must still exist. If it doesn't just emit warning
4650                          * and return an error to fall back to a transaction
4651                          * commit.
4652                          */
4653                         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4654                         if (ret < 0)
4655                                 return ret;
4656                         if (WARN_ON(ret > 0))
4657                                 return -ENOENT;
4658                         leaf = path->nodes[0];
4659                 }
4660 
4661                 extent = btrfs_item_ptr(leaf, path->slots[0],
4662                                         struct btrfs_file_extent_item);
4663                 if (btrfs_file_extent_type(leaf, extent) ==
4664                     BTRFS_FILE_EXTENT_INLINE) {
4665                         len = btrfs_file_extent_ram_bytes(leaf, extent);
4666                         prev_extent_end = ALIGN(key.offset + len,
4667                                                 fs_info->sectorsize);
4668                 } else {
4669                         len = btrfs_file_extent_num_bytes(leaf, extent);
4670                         prev_extent_end = key.offset + len;
4671                 }
4672 
4673                 path->slots[0]++;
4674                 cond_resched();
4675         }
4676 
4677         if (prev_extent_end < i_size) {
4678                 u64 hole_len;
4679 
4680                 btrfs_release_path(path);
4681                 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4682                 ret = btrfs_insert_file_extent(trans, root->log_root,
4683                                                ino, prev_extent_end, 0, 0,
4684                                                hole_len, 0, hole_len,
4685                                                0, 0, 0);
4686                 if (ret < 0)
4687                         return ret;
4688         }
4689 
4690         return 0;
4691 }
4692 
4693 /*
4694  * When we are logging a new inode X, check if it doesn't have a reference that
4695  * matches the reference from some other inode Y created in a past transaction
4696  * and that was renamed in the current transaction. If we don't do this, then at
4697  * log replay time we can lose inode Y (and all its files if it's a directory):
4698  *
4699  * mkdir /mnt/x
4700  * echo "hello world" > /mnt/x/foobar
4701  * sync
4702  * mv /mnt/x /mnt/y
4703  * mkdir /mnt/x                 # or touch /mnt/x
4704  * xfs_io -c fsync /mnt/x
4705  * <power fail>
4706  * mount fs, trigger log replay
4707  *
4708  * After the log replay procedure, we would lose the first directory and all its
4709  * files (file foobar).
4710  * For the case where inode Y is not a directory we simply end up losing it:
4711  *
4712  * echo "123" > /mnt/foo
4713  * sync
4714  * mv /mnt/foo /mnt/bar
4715  * echo "abc" > /mnt/foo
4716  * xfs_io -c fsync /mnt/foo
4717  * <power fail>
4718  *
4719  * We also need this for cases where a snapshot entry is replaced by some other
4720  * entry (file or directory) otherwise we end up with an unreplayable log due to
4721  * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4722  * if it were a regular entry:
4723  *
4724  * mkdir /mnt/x
4725  * btrfs subvolume snapshot /mnt /mnt/x/snap
4726  * btrfs subvolume delete /mnt/x/snap
4727  * rmdir /mnt/x
4728  * mkdir /mnt/x
4729  * fsync /mnt/x or fsync some new file inside it
4730  * <power fail>
4731  *
4732  * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4733  * the same transaction.
4734  */
4735 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4736                                          const int slot,
4737                                          const struct btrfs_key *key,
4738                                          struct btrfs_inode *inode,
4739                                          u64 *other_ino, u64 *other_parent)
4740 {
4741         int ret;
4742         struct btrfs_path *search_path;
4743         char *name = NULL;
4744         u32 name_len = 0;
4745         u32 item_size = btrfs_item_size_nr(eb, slot);
4746         u32 cur_offset = 0;
4747         unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4748 
4749         search_path = btrfs_alloc_path();
4750         if (!search_path)
4751                 return -ENOMEM;
4752         search_path->search_commit_root = 1;
4753         search_path->skip_locking = 1;
4754 
4755         while (cur_offset < item_size) {
4756                 u64 parent;
4757                 u32 this_name_len;
4758                 u32 this_len;
4759                 unsigned long name_ptr;
4760                 struct btrfs_dir_item *di;
4761 
4762                 if (key->type == BTRFS_INODE_REF_KEY) {
4763                         struct btrfs_inode_ref *iref;
4764 
4765                         iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4766                         parent = key->offset;
4767                         this_name_len = btrfs_inode_ref_name_len(eb, iref);
4768                         name_ptr = (unsigned long)(iref + 1);
4769                         this_len = sizeof(*iref) + this_name_len;
4770                 } else {
4771                         struct btrfs_inode_extref *extref;
4772 
4773                         extref = (struct btrfs_inode_extref *)(ptr +
4774                                                                cur_offset);
4775                         parent = btrfs_inode_extref_parent(eb, extref);
4776                         this_name_len = btrfs_inode_extref_name_len(eb, extref);
4777                         name_ptr = (unsigned long)&extref->name;
4778                         this_len = sizeof(*extref) + this_name_len;
4779                 }
4780 
4781                 if (this_name_len > name_len) {
4782                         char *new_name;
4783 
4784                         new_name = krealloc(name, this_name_len, GFP_NOFS);
4785                         if (!new_name) {
4786                                 ret = -ENOMEM;
4787                                 goto out;
4788                         }
4789                         name_len = this_name_len;
4790                         name = new_name;
4791                 }
4792 
4793                 read_extent_buffer(eb, name, name_ptr, this_name_len);
4794                 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4795                                 parent, name, this_name_len, 0);
4796                 if (di && !IS_ERR(di)) {
4797                         struct btrfs_key di_key;
4798 
4799                         btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4800                                                   di, &di_key);
4801                         if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4802                                 if (di_key.objectid != key->objectid) {
4803                                         ret = 1;
4804                                         *other_ino = di_key.objectid;
4805                                         *other_parent = parent;
4806                                 } else {
4807                                         ret = 0;
4808                                 }
4809                         } else {
4810                                 ret = -EAGAIN;
4811                         }
4812                         goto out;
4813                 } else if (IS_ERR(di)) {
4814                         ret = PTR_ERR(di);
4815                         goto out;
4816                 }
4817                 btrfs_release_path(search_path);
4818 
4819                 cur_offset += this_len;
4820         }
4821         ret = 0;
4822 out:
4823         btrfs_free_path(search_path);
4824         kfree(name);
4825         return ret;
4826 }
4827 
4828 struct btrfs_ino_list {
4829         u64 ino;
4830         u64 parent;
4831         struct list_head list;
4832 };
4833 
4834 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4835                                   struct btrfs_root *root,
4836                                   struct btrfs_path *path,
4837                                   struct btrfs_log_ctx *ctx,
4838                                   u64 ino, u64 parent)
4839 {
4840         struct btrfs_ino_list *ino_elem;
4841         LIST_HEAD(inode_list);
4842         int ret = 0;
4843 
4844         ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4845         if (!ino_elem)
4846                 return -ENOMEM;
4847         ino_elem->ino = ino;
4848         ino_elem->parent = parent;
4849         list_add_tail(&ino_elem->list, &inode_list);
4850 
4851         while (!list_empty(&inode_list)) {
4852                 struct btrfs_fs_info *fs_info = root->fs_info;
4853                 struct btrfs_key key;
4854                 struct inode *inode;
4855 
4856                 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4857                                             list);
4858                 ino = ino_elem->ino;
4859                 parent = ino_elem->parent;
4860                 list_del(&ino_elem->list);
4861                 kfree(ino_elem);
4862                 if (ret)
4863                         continue;
4864 
4865                 btrfs_release_path(path);
4866 
4867                 key.objectid = ino;
4868                 key.type = BTRFS_INODE_ITEM_KEY;
4869                 key.offset = 0;
4870                 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4871                 /*
4872                  * If the other inode that had a conflicting dir entry was
4873                  * deleted in the current transaction, we need to log its parent
4874                  * directory.
4875                  */
4876                 if (IS_ERR(inode)) {
4877                         ret = PTR_ERR(inode);
4878                         if (ret == -ENOENT) {
4879                                 key.objectid = parent;
4880                                 inode = btrfs_iget(fs_info->sb, &key, root,
4881                                                    NULL);
4882                                 if (IS_ERR(inode)) {
4883                                         ret = PTR_ERR(inode);
4884                                 } else {
4885                                         ret = btrfs_log_inode(trans, root,
4886                                                       BTRFS_I(inode),
4887                                                       LOG_OTHER_INODE_ALL,
4888                                                       0, LLONG_MAX, ctx);
4889                                         btrfs_add_delayed_iput(inode);
4890                                 }
4891                         }
4892                         continue;
4893                 }
4894                 /*
4895                  * If the inode was already logged skip it - otherwise we can
4896                  * hit an infinite loop. Example:
4897                  *
4898                  * From the commit root (previous transaction) we have the
4899                  * following inodes:
4900                  *
4901                  * inode 257 a directory
4902                  * inode 258 with references "zz" and "zz_link" on inode 257
4903                  * inode 259 with reference "a" on inode 257
4904                  *
4905                  * And in the current (uncommitted) transaction we have:
4906                  *
4907                  * inode 257 a directory, unchanged
4908                  * inode 258 with references "a" and "a2" on inode 257
4909                  * inode 259 with reference "zz_link" on inode 257
4910                  * inode 261 with reference "zz" on inode 257
4911                  *
4912                  * When logging inode 261 the following infinite loop could
4913                  * happen if we don't skip already logged inodes:
4914                  *
4915                  * - we detect inode 258 as a conflicting inode, with inode 261
4916                  *   on reference "zz", and log it;
4917                  *
4918                  * - we detect inode 259 as a conflicting inode, with inode 258
4919                  *   on reference "a", and log it;
4920                  *
4921                  * - we detect inode 258 as a conflicting inode, with inode 259
4922                  *   on reference "zz_link", and log it - again! After this we
4923                  *   repeat the above steps forever.
4924                  */
4925                 spin_lock(&BTRFS_I(inode)->lock);
4926                 /*
4927                  * Check the inode's logged_trans only instead of
4928                  * btrfs_inode_in_log(). This is because the last_log_commit of
4929                  * the inode is not updated when we only log that it exists and
4930                  * and it has the full sync bit set (see btrfs_log_inode()).
4931                  */
4932                 if (BTRFS_I(inode)->logged_trans == trans->transid) {
4933                         spin_unlock(&BTRFS_I(inode)->lock);
4934                         btrfs_add_delayed_iput(inode);
4935                         continue;
4936                 }
4937                 spin_unlock(&BTRFS_I(inode)->lock);
4938                 /*
4939                  * We are safe logging the other inode without acquiring its
4940                  * lock as long as we log with the LOG_INODE_EXISTS mode. We
4941                  * are safe against concurrent renames of the other inode as
4942                  * well because during a rename we pin the log and update the
4943                  * log with the new name before we unpin it.
4944                  */
4945                 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4946                                       LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4947                 if (ret) {
4948                         btrfs_add_delayed_iput(inode);
4949                         continue;
4950                 }
4951 
4952                 key.objectid = ino;
4953                 key.type = BTRFS_INODE_REF_KEY;
4954                 key.offset = 0;
4955                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4956                 if (ret < 0) {
4957                         btrfs_add_delayed_iput(inode);
4958                         continue;
4959                 }
4960 
4961                 while (true) {
4962                         struct extent_buffer *leaf = path->nodes[0];
4963                         int slot = path->slots[0];
4964                         u64 other_ino = 0;
4965                         u64 other_parent = 0;
4966 
4967                         if (slot >= btrfs_header_nritems(leaf)) {
4968                                 ret = btrfs_next_leaf(root, path);
4969                                 if (ret < 0) {
4970                                         break;
4971                                 } else if (ret > 0) {
4972                                         ret = 0;
4973                                         break;
4974                                 }
4975                                 continue;
4976                         }
4977 
4978                         btrfs_item_key_to_cpu(leaf, &key, slot);
4979                         if (key.objectid != ino ||
4980                             (key.type != BTRFS_INODE_REF_KEY &&
4981                              key.type != BTRFS_INODE_EXTREF_KEY)) {
4982                                 ret = 0;
4983                                 break;
4984                         }
4985 
4986                         ret = btrfs_check_ref_name_override(leaf, slot, &key,
4987                                         BTRFS_I(inode), &other_ino,
4988                                         &other_parent);
4989                         if (ret < 0)
4990                                 break;
4991                         if (ret > 0) {
4992                                 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4993                                 if (!ino_elem) {
4994                                         ret = -ENOMEM;
4995                                         break;
4996                                 }
4997                                 ino_elem->ino = other_ino;
4998                                 ino_elem->parent = other_parent;
4999                                 list_add_tail(&ino_elem->list, &inode_list);
5000                                 ret = 0;
5001                         }
5002                         path->slots[0]++;
5003                 }
5004                 btrfs_add_delayed_iput(inode);
5005         }
5006 
5007         return ret;
5008 }
5009 
5010 /* log a single inode in the tree log.
5011  * At least one parent directory for this inode must exist in the tree
5012  * or be logged already.
5013  *
5014  * Any items from this inode changed by the current transaction are copied
5015  * to the log tree.  An extra reference is taken on any extents in this
5016  * file, allowing us to avoid a whole pile of corner cases around logging
5017  * blocks that have been removed from the tree.
5018  *
5019  * See LOG_INODE_ALL and related defines for a description of what inode_only
5020  * does.
5021  *
5022  * This handles both files and directories.
5023  */
5024 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5025                            struct btrfs_root *root, struct btrfs_inode *inode,
5026                            int inode_only,
5027                            const loff_t start,
5028                            const loff_t end,
5029                            struct btrfs_log_ctx *ctx)
5030 {
5031         struct btrfs_fs_info *fs_info = root->fs_info;
5032         struct btrfs_path *path;
5033         struct btrfs_path *dst_path;
5034         struct btrfs_key min_key;
5035         struct btrfs_key max_key;
5036         struct btrfs_root *log = root->log_root;
5037         int err = 0;
5038         int ret;
5039         int nritems;
5040         int ins_start_slot = 0;
5041         int ins_nr;
5042         bool fast_search = false;
5043         u64 ino = btrfs_ino(inode);
5044         struct extent_map_tree *em_tree = &inode->extent_tree;
5045         u64 logged_isize = 0;
5046         bool need_log_inode_item = true;
5047         bool xattrs_logged = false;
5048         bool recursive_logging = false;
5049 
5050         path = btrfs_alloc_path();
5051         if (!path)
5052                 return -ENOMEM;
5053         dst_path = btrfs_alloc_path();
5054         if (!dst_path) {
5055                 btrfs_free_path(path);
5056                 return -ENOMEM;
5057         }
5058 
5059         min_key.objectid = ino;
5060         min_key.type = BTRFS_INODE_ITEM_KEY;
5061         min_key.offset = 0;
5062 
5063         max_key.objectid = ino;
5064 
5065 
5066         /* today the code can only do partial logging of directories */
5067         if (S_ISDIR(inode->vfs_inode.i_mode) ||
5068             (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5069                        &inode->runtime_flags) &&
5070              inode_only >= LOG_INODE_EXISTS))
5071                 max_key.type = BTRFS_XATTR_ITEM_KEY;
5072         else
5073                 max_key.type = (u8)-1;
5074         max_key.offset = (u64)-1;
5075 
5076         /*
5077          * Only run delayed items if we are a dir or a new file.
5078          * Otherwise commit the delayed inode only, which is needed in
5079          * order for the log replay code to mark inodes for link count
5080          * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5081          */
5082         if (S_ISDIR(inode->vfs_inode.i_mode) ||
5083             inode->generation > fs_info->last_trans_committed)
5084                 ret = btrfs_commit_inode_delayed_items(trans, inode);
5085         else
5086                 ret = btrfs_commit_inode_delayed_inode(inode);
5087 
5088         if (ret) {
5089                 btrfs_free_path(path);
5090                 btrfs_free_path(dst_path);
5091                 return ret;
5092         }
5093 
5094         if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5095                 recursive_logging = true;
5096                 if (inode_only == LOG_OTHER_INODE)
5097                         inode_only = LOG_INODE_EXISTS;
5098                 else
5099                         inode_only = LOG_INODE_ALL;
5100                 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5101         } else {
5102                 mutex_lock(&inode->log_mutex);
5103         }
5104 
5105         /*
5106          * a brute force approach to making sure we get the most uptodate
5107          * copies of everything.
5108          */
5109         if (S_ISDIR(inode->vfs_inode.i_mode)) {
5110                 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5111 
5112                 if (inode_only == LOG_INODE_EXISTS)
5113                         max_key_type = BTRFS_XATTR_ITEM_KEY;
5114                 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5115         } else {
5116                 if (inode_only == LOG_INODE_EXISTS) {
5117                         /*
5118                          * Make sure the new inode item we write to the log has
5119                          * the same isize as the current one (if it exists).
5120                          * This is necessary to prevent data loss after log
5121                          * replay, and also to prevent doing a wrong expanding
5122                          * truncate - for e.g. create file, write 4K into offset
5123                          * 0, fsync, write 4K into offset 4096, add hard link,
5124                          * fsync some other file (to sync log), power fail - if
5125                          * we use the inode's current i_size, after log replay
5126                          * we get a 8Kb file, with the last 4Kb extent as a hole
5127                          * (zeroes), as if an expanding truncate happened,
5128                          * instead of getting a file of 4Kb only.
5129                          */
5130                         err = logged_inode_size(log, inode, path, &logged_isize);
5131                         if (err)
5132                                 goto out_unlock;
5133                 }
5134                 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5135                              &inode->runtime_flags)) {
5136                         if (inode_only == LOG_INODE_EXISTS) {
5137                                 max_key.type = BTRFS_XATTR_ITEM_KEY;
5138                                 ret = drop_objectid_items(trans, log, path, ino,
5139                                                           max_key.type);
5140                         } else {
5141                                 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5142                                           &inode->runtime_flags);
5143                                 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5144                                           &inode->runtime_flags);
5145                                 while(1) {
5146                                         ret = btrfs_truncate_inode_items(trans,
5147                                                 log, &inode->vfs_inode, 0, 0);
5148                                         if (ret != -EAGAIN)
5149                                                 break;
5150                                 }
5151                         }
5152                 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5153                                               &inode->runtime_flags) ||
5154                            inode_only == LOG_INODE_EXISTS) {
5155                         if (inode_only == LOG_INODE_ALL)
5156                                 fast_search = true;
5157                         max_key.type = BTRFS_XATTR_ITEM_KEY;
5158                         ret = drop_objectid_items(trans, log, path, ino,
5159                                                   max_key.type);
5160                 } else {
5161                         if (inode_only == LOG_INODE_ALL)
5162                                 fast_search = true;
5163                         goto log_extents;
5164                 }
5165 
5166         }
5167         if (ret) {
5168                 err = ret;
5169                 goto out_unlock;
5170         }
5171 
5172         while (1) {
5173                 ins_nr = 0;
5174                 ret = btrfs_search_forward(root, &min_key,
5175                                            path, trans->transid);
5176                 if (ret < 0) {
5177                         err = ret;
5178                         goto out_unlock;
5179                 }
5180                 if (ret != 0)
5181                         break;
5182 again:
5183                 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5184                 if (min_key.objectid != ino)
5185                         break;
5186                 if (min_key.type > max_key.type)
5187                         break;
5188 
5189                 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5190                         need_log_inode_item = false;
5191 
5192                 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5193                      min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5194                     inode->generation == trans->transid &&
5195                     !recursive_logging) {
5196                         u64 other_ino = 0;
5197                         u64 other_parent = 0;
5198 
5199                         ret = btrfs_check_ref_name_override(path->nodes[0],
5200                                         path->slots[0], &min_key, inode,
5201                                         &other_ino, &other_parent);
5202                         if (ret < 0) {
5203                                 err = ret;
5204                                 goto out_unlock;
5205                         } else if (ret > 0 && ctx &&
5206                                    other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5207                                 if (ins_nr > 0) {
5208                                         ins_nr++;
5209                                 } else {
5210                                         ins_nr = 1;
5211                                         ins_start_slot = path->slots[0];
5212                                 }
5213                                 ret = copy_items(trans, inode, dst_path, path,
5214                                                  ins_start_slot,
5215                                                  ins_nr, inode_only,
5216                                                  logged_isize);
5217                                 if (ret < 0) {
5218                                         err = ret;
5219                                         goto out_unlock;
5220                                 }
5221                                 ins_nr = 0;
5222 
5223                                 err = log_conflicting_inodes(trans, root, path,
5224                                                 ctx, other_ino, other_parent);
5225                                 if (err)
5226                                         goto out_unlock;
5227                                 btrfs_release_path(path);
5228                                 goto next_key;
5229                         }
5230                 }
5231 
5232                 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5233                 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5234                         if (ins_nr == 0)
5235                                 goto next_slot;
5236                         ret = copy_items(trans, inode, dst_path, path,
5237                                          ins_start_slot,
5238                                          ins_nr, inode_only, logged_isize);
5239                         if (ret < 0) {
5240                                 err = ret;
5241                                 goto out_unlock;
5242                         }
5243                         ins_nr = 0;
5244                         goto next_slot;
5245                 }
5246 
5247                 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5248                         ins_nr++;
5249                         goto next_slot;
5250                 } else if (!ins_nr) {
5251                         ins_start_slot = path->slots[0];
5252                         ins_nr = 1;
5253                         goto next_slot;
5254                 }
5255 
5256                 ret = copy_items(trans, inode, dst_path, path,
5257                                  ins_start_slot, ins_nr, inode_only,
5258                                  logged_isize);
5259                 if (ret < 0) {
5260                         err = ret;
5261                         goto out_unlock;
5262                 }
5263                 ins_nr = 1;
5264                 ins_start_slot = path->slots[0];
5265 next_slot:
5266 
5267                 nritems = btrfs_header_nritems(path->nodes[0]);
5268                 path->slots[0]++;
5269                 if (path->slots[0] < nritems) {
5270                         btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5271                                               path->slots[0]);
5272                         goto again;
5273                 }
5274                 if (ins_nr) {
5275                         ret = copy_items(trans, inode, dst_path, path,
5276                                          ins_start_slot,
5277                                          ins_nr, inode_only, logged_isize);
5278                         if (ret < 0) {
5279                                 err = ret;
5280                                 goto out_unlock;
5281                         }
5282                         ins_nr = 0;
5283                 }
5284                 btrfs_release_path(path);
5285 next_key:
5286                 if (min_key.offset < (u64)-1) {
5287                         min_key.offset++;
5288                 } else if (min_key.type < max_key.type) {
5289                         min_key.type++;
5290                         min_key.offset = 0;
5291                 } else {
5292                         break;
5293                 }
5294         }
5295         if (ins_nr) {
5296                 ret = copy_items(trans, inode, dst_path, path,
5297                                  ins_start_slot, ins_nr, inode_only,
5298                                  logged_isize);
5299                 if (ret < 0) {
5300                         err = ret;
5301                         goto out_unlock;
5302                 }
5303                 ins_nr = 0;
5304         }
5305 
5306         btrfs_release_path(path);
5307         btrfs_release_path(dst_path);
5308         err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5309         if (err)
5310                 goto out_unlock;
5311         xattrs_logged = true;
5312         if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5313                 btrfs_release_path(path);
5314                 btrfs_release_path(dst_path);
5315                 err = btrfs_log_holes(trans, root, inode, path);
5316                 if (err)
5317                         goto out_unlock;
5318         }
5319 log_extents:
5320         btrfs_release_path(path);
5321         btrfs_release_path(dst_path);
5322         if (need_log_inode_item) {
5323                 err = log_inode_item(trans, log, dst_path, inode);
5324                 if (!err && !xattrs_logged) {
5325                         err = btrfs_log_all_xattrs(trans, root, inode, path,
5326                                                    dst_path);
5327                         btrfs_release_path(path);
5328                 }
5329                 if (err)
5330                         goto out_unlock;
5331         }
5332         if (fast_search) {
5333                 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5334                                                 ctx, start, end);
5335                 if (ret) {
5336                         err = ret;
5337                         goto out_unlock;
5338                 }
5339         } else if (inode_only == LOG_INODE_ALL) {
5340                 struct extent_map *em, *n;
5341 
5342                 write_lock(&em_tree->lock);
5343                 /*
5344                  * We can't just remove every em if we're called for a ranged
5345                  * fsync - that is, one that doesn't cover the whole possible
5346                  * file range (0 to LLONG_MAX). This is because we can have
5347                  * em's that fall outside the range we're logging and therefore
5348                  * their ordered operations haven't completed yet
5349                  * (btrfs_finish_ordered_io() not invoked yet). This means we
5350                  * didn't get their respective file extent item in the fs/subvol
5351                  * tree yet, and need to let the next fast fsync (one which
5352                  * consults the list of modified extent maps) find the em so
5353                  * that it logs a matching file extent item and waits for the
5354                  * respective ordered operation to complete (if it's still
5355                  * running).
5356                  *
5357                  * Removing every em outside the range we're logging would make
5358                  * the next fast fsync not log their matching file extent items,
5359                  * therefore making us lose data after a log replay.
5360                  */
5361                 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5362                                          list) {
5363                         const u64 mod_end = em->mod_start + em->mod_len - 1;
5364 
5365                         if (em->mod_start >= start && mod_end <= end)
5366                                 list_del_init(&em->list);
5367                 }
5368                 write_unlock(&em_tree->lock);
5369         }
5370 
5371         if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5372                 ret = log_directory_changes(trans, root, inode, path, dst_path,
5373                                         ctx);
5374                 if (ret) {
5375                         err = ret;
5376                         goto out_unlock;
5377                 }
5378         }
5379 
5380         /*
5381          * Don't update last_log_commit if we logged that an inode exists after
5382          * it was loaded to memory (full_sync bit set).
5383          * This is to prevent data loss when we do a write to the inode, then
5384          * the inode gets evicted after all delalloc was flushed, then we log
5385          * it exists (due to a rename for example) and then fsync it. This last
5386          * fsync would do nothing (not logging the extents previously written).
5387          */
5388         spin_lock(&inode->lock);
5389         inode->logged_trans = trans->transid;
5390         if (inode_only != LOG_INODE_EXISTS ||
5391             !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5392                 inode->last_log_commit = inode->last_sub_trans;
5393         spin_unlock(&inode->lock);
5394 out_unlock:
5395         mutex_unlock(&inode->log_mutex);
5396 
5397         btrfs_free_path(path);
5398         btrfs_free_path(dst_path);
5399         return err;
5400 }
5401 
5402 /*
5403  * Check if we must fallback to a transaction commit when logging an inode.
5404  * This must be called after logging the inode and is used only in the context
5405  * when fsyncing an inode requires the need to log some other inode - in which
5406  * case we can't lock the i_mutex of each other inode we need to log as that
5407  * can lead to deadlocks with concurrent fsync against other inodes (as we can
5408  * log inodes up or down in the hierarchy) or rename operations for example. So
5409  * we take the log_mutex of the inode after we have logged it and then check for
5410  * its last_unlink_trans value - this is safe because any task setting
5411  * last_unlink_trans must take the log_mutex and it must do this before it does
5412  * the actual unlink operation, so if we do this check before a concurrent task
5413  * sets last_unlink_trans it means we've logged a consistent version/state of
5414  * all the inode items, otherwise we are not sure and must do a transaction
5415  * commit (the concurrent task might have only updated last_unlink_trans before
5416  * we logged the inode or it might have also done the unlink).
5417  */
5418 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5419                                           struct btrfs_inode *inode)
5420 {
5421         struct btrfs_fs_info *fs_info = inode->root->fs_info;
5422         bool ret = false;
5423 
5424         mutex_lock(&inode->log_mutex);
5425         if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5426                 /*
5427                  * Make sure any commits to the log are forced to be full
5428                  * commits.
5429                  */
5430                 btrfs_set_log_full_commit(trans);
5431                 ret = true;
5432         }
5433         mutex_unlock(&inode->log_mutex);
5434 
5435         return ret;
5436 }
5437 
5438 /*
5439  * follow the dentry parent pointers up the chain and see if any
5440  * of the directories in it require a full commit before they can
5441  * be logged.  Returns zero if nothing special needs to be done or 1 if
5442  * a full commit is required.
5443  */
5444 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5445                                                struct btrfs_inode *inode,
5446                                                struct dentry *parent,
5447                                                struct super_block *sb,
5448                                                u64 last_committed)
5449 {
5450         int ret = 0;
5451         struct dentry *old_parent = NULL;
5452 
5453         /*
5454          * for regular files, if its inode is already on disk, we don't
5455          * have to worry about the parents at all.  This is because
5456          * we can use the last_unlink_trans field to record renames
5457          * and other fun in this file.
5458          */
5459         if (S_ISREG(inode->vfs_inode.i_mode) &&
5460             inode->generation <= last_committed &&
5461             inode->last_unlink_trans <= last_committed)
5462                 goto out;
5463 
5464         if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5465                 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5466                         goto out;
5467                 inode = BTRFS_I(d_inode(parent));
5468         }
5469 
5470         while (1) {
5471                 if (btrfs_must_commit_transaction(trans, inode)) {
5472                         ret = 1;
5473                         break;
5474                 }
5475 
5476                 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5477                         break;
5478 
5479                 if (IS_ROOT(parent)) {
5480                         inode = BTRFS_I(d_inode(parent));
5481                         if (btrfs_must_commit_transaction(trans, inode))
5482                                 ret = 1;
5483                         break;
5484                 }
5485 
5486                 parent = dget_parent(parent);
5487                 dput(old_parent);
5488                 old_parent = parent;
5489                 inode = BTRFS_I(d_inode(parent));
5490 
5491         }
5492         dput(old_parent);
5493 out:
5494         return ret;
5495 }
5496 
5497 struct btrfs_dir_list {
5498         u64 ino;
5499         struct list_head list;
5500 };
5501 
5502 /*
5503  * Log the inodes of the new dentries of a directory. See log_dir_items() for
5504  * details about the why it is needed.
5505  * This is a recursive operation - if an existing dentry corresponds to a
5506  * directory, that directory's new entries are logged too (same behaviour as
5507  * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5508  * the dentries point to we do not lock their i_mutex, otherwise lockdep
5509  * complains about the following circular lock dependency / possible deadlock:
5510  *
5511  *        CPU0                                        CPU1
5512  *        ----                                        ----
5513  * lock(&type->i_mutex_dir_key#3/2);
5514  *                                            lock(sb_internal#2);
5515  *                                            lock(&type->i_mutex_dir_key#3/2);
5516  * lock(&sb->s_type->i_mutex_key#14);
5517  *
5518  * Where sb_internal is the lock (a counter that works as a lock) acquired by
5519  * sb_start_intwrite() in btrfs_start_transaction().
5520  * Not locking i_mutex of the inodes is still safe because:
5521  *
5522  * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5523  *    that while logging the inode new references (names) are added or removed
5524  *    from the inode, leaving the logged inode item with a link count that does
5525  *    not match the number of logged inode reference items. This is fine because
5526  *    at log replay time we compute the real number of links and correct the
5527  *    link count in the inode item (see replay_one_buffer() and
5528  *    link_to_fixup_dir());
5529  *
5530  * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5531  *    while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5532  *    BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5533  *    has a size that doesn't match the sum of the lengths of all the logged
5534  *    names. This does not result in a problem because if a dir_item key is
5535  *    logged but its matching dir_index key is not logged, at log replay time we
5536  *    don't use it to replay the respective name (see replay_one_name()). On the
5537  *    other hand if only the dir_index key ends up being logged, the respective
5538  *    name is added to the fs/subvol tree with both the dir_item and dir_index
5539  *    keys created (see replay_one_name()).
5540  *    The directory's inode item with a wrong i_size is not a problem as well,
5541  *    since we don't use it at log replay time to set the i_size in the inode
5542  *    item of the fs/subvol tree (see overwrite_item()).
5543  */
5544 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5545                                 struct btrfs_root *root,
5546                                 struct btrfs_inode *start_inode,
5547                                 struct btrfs_log_ctx *ctx)
5548 {
5549         struct btrfs_fs_info *fs_info = root->fs_info;
5550         struct btrfs_root *log = root->log_root;
5551         struct btrfs_path *path;
5552         LIST_HEAD(dir_list);
5553         struct btrfs_dir_list *dir_elem;
5554         int ret = 0;
5555 
5556         path = btrfs_alloc_path();
5557         if (!path)
5558                 return -ENOMEM;
5559 
5560         dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5561         if (!dir_elem) {
5562                 btrfs_free_path(path);
5563                 return -ENOMEM;
5564         }
5565         dir_elem->ino = btrfs_ino(start_inode);
5566         list_add_tail(&dir_elem->list, &dir_list);
5567 
5568         while (!list_empty(&dir_list)) {
5569                 struct extent_buffer *leaf;
5570                 struct btrfs_key min_key;
5571                 int nritems;
5572                 int i;
5573 
5574                 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5575                                             list);
5576                 if (ret)
5577                         goto next_dir_inode;
5578 
5579                 min_key.objectid = dir_elem->ino;
5580                 min_key.type = BTRFS_DIR_ITEM_KEY;
5581                 min_key.offset = 0;
5582 again:
5583                 btrfs_release_path(path);
5584                 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5585                 if (ret < 0) {
5586                         goto next_dir_inode;
5587                 } else if (ret > 0) {
5588                         ret = 0;
5589                         goto next_dir_inode;
5590                 }
5591 
5592 process_leaf:
5593                 leaf = path->nodes[0];
5594                 nritems = btrfs_header_nritems(leaf);
5595                 for (i = path->slots[0]; i < nritems; i++) {
5596                         struct btrfs_dir_item *di;
5597                         struct btrfs_key di_key;
5598                         struct inode *di_inode;
5599                         struct btrfs_dir_list *new_dir_elem;
5600                         int log_mode = LOG_INODE_EXISTS;
5601                         int type;
5602 
5603                         btrfs_item_key_to_cpu(leaf, &min_key, i);
5604                         if (min_key.objectid != dir_elem->ino ||
5605                             min_key.type != BTRFS_DIR_ITEM_KEY)
5606                                 goto next_dir_inode;
5607 
5608                         di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5609                         type = btrfs_dir_type(leaf, di);
5610                         if (btrfs_dir_transid(leaf, di) < trans->transid &&
5611                             type != BTRFS_FT_DIR)
5612                                 continue;
5613                         btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5614                         if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5615                                 continue;
5616 
5617                         btrfs_release_path(path);
5618                         di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5619                         if (IS_ERR(di_inode)) {
5620                                 ret = PTR_ERR(di_inode);
5621                                 goto next_dir_inode;
5622                         }
5623 
5624                         if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5625                                 btrfs_add_delayed_iput(di_inode);
5626                                 break;
5627                         }
5628 
5629                         ctx->log_new_dentries = false;
5630                         if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5631                                 log_mode = LOG_INODE_ALL;
5632                         ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5633                                               log_mode, 0, LLONG_MAX, ctx);
5634                         if (!ret &&
5635                             btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5636                                 ret = 1;
5637                         btrfs_add_delayed_iput(di_inode);
5638                         if (ret)
5639                                 goto next_dir_inode;
5640                         if (ctx->log_new_dentries) {
5641                                 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5642                                                        GFP_NOFS);
5643                                 if (!new_dir_elem) {
5644                                         ret = -ENOMEM;
5645                                         goto next_dir_inode;
5646                                 }
5647                                 new_dir_elem->ino = di_key.objectid;
5648                                 list_add_tail(&new_dir_elem->list, &dir_list);
5649                         }
5650                         break;
5651                 }
5652                 if (i == nritems) {
5653                         ret = btrfs_next_leaf(log, path);
5654                         if (ret < 0) {
5655                                 goto next_dir_inode;
5656                         } else if (ret > 0) {
5657                                 ret = 0;
5658                                 goto next_dir_inode;
5659                         }
5660                         goto process_leaf;
5661                 }
5662                 if (min_key.offset < (u64)-1) {
5663                         min_key.offset++;
5664                         goto again;
5665                 }
5666 next_dir_inode:
5667                 list_del(&dir_elem->list);
5668                 kfree(dir_elem);
5669         }
5670 
5671         btrfs_free_path(path);
5672         return ret;
5673 }
5674 
5675 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5676                                  struct btrfs_inode *inode,
5677                                  struct btrfs_log_ctx *ctx)
5678 {
5679         struct btrfs_fs_info *fs_info = trans->fs_info;
5680         int ret;
5681         struct btrfs_path *path;
5682         struct btrfs_key key;
5683         struct btrfs_root *root = inode->root;
5684         const u64 ino = btrfs_ino(inode);
5685 
5686         path = btrfs_alloc_path();
5687         if (!path)
5688                 return -ENOMEM;
5689         path->skip_locking = 1;
5690         path->search_commit_root = 1;
5691 
5692         key.objectid = ino;
5693         key.type = BTRFS_INODE_REF_KEY;
5694         key.offset = 0;
5695         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5696         if (ret < 0)
5697                 goto out;
5698 
5699         while (true) {
5700                 struct extent_buffer *leaf = path->nodes[0];
5701                 int slot = path->slots[0];
5702                 u32 cur_offset = 0;
5703                 u32 item_size;
5704                 unsigned long ptr;
5705 
5706                 if (slot >= btrfs_header_nritems(leaf)) {
5707                         ret = btrfs_next_leaf(root, path);
5708                         if (ret < 0)
5709                                 goto out;
5710                         else if (ret > 0)
5711                                 break;
5712                         continue;
5713                 }
5714 
5715                 btrfs_item_key_to_cpu(leaf, &key, slot);
5716                 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5717                 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5718                         break;
5719 
5720                 item_size = btrfs_item_size_nr(leaf, slot);
5721                 ptr = btrfs_item_ptr_offset(leaf, slot);
5722                 while (cur_offset < item_size) {
5723                         struct btrfs_key inode_key;
5724                         struct inode *dir_inode;
5725 
5726                         inode_key.type = BTRFS_INODE_ITEM_KEY;
5727                         inode_key.offset = 0;
5728 
5729                         if (key.type == BTRFS_INODE_EXTREF_KEY) {
5730                                 struct btrfs_inode_extref *extref;
5731 
5732                                 extref = (struct btrfs_inode_extref *)
5733                                         (ptr + cur_offset);
5734                                 inode_key.objectid = btrfs_inode_extref_parent(
5735                                         leaf, extref);
5736                                 cur_offset += sizeof(*extref);
5737                                 cur_offset += btrfs_inode_extref_name_len(leaf,
5738                                         extref);
5739                         } else {
5740                                 inode_key.objectid = key.offset;
5741                                 cur_offset = item_size;
5742                         }
5743 
5744                         dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5745                                                root, NULL);
5746                         /*
5747                          * If the parent inode was deleted, return an error to
5748                          * fallback to a transaction commit. This is to prevent
5749                          * getting an inode that was moved from one parent A to
5750                          * a parent B, got its former parent A deleted and then
5751                          * it got fsync'ed, from existing at both parents after
5752                          * a log replay (and the old parent still existing).
5753                          * Example:
5754                          *
5755                          * mkdir /mnt/A
5756                          * mkdir /mnt/B
5757                          * touch /mnt/B/bar
5758                          * sync
5759                          * mv /mnt/B/bar /mnt/A/bar
5760                          * mv -T /mnt/A /mnt/B
5761                          * fsync /mnt/B/bar
5762                          * <power fail>
5763                          *
5764                          * If we ignore the old parent B which got deleted,
5765                          * after a log replay we would have file bar linked
5766                          * at both parents and the old parent B would still
5767                          * exist.
5768                          */
5769                         if (IS_ERR(dir_inode)) {
5770                                 ret = PTR_ERR(dir_inode);
5771                                 goto out;
5772                         }
5773 
5774                         if (ctx)
5775                                 ctx->log_new_dentries = false;
5776                         ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5777                                               LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5778                         if (!ret &&
5779                             btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5780                                 ret = 1;
5781                         if (!ret && ctx && ctx->log_new_dentries)
5782                                 ret = log_new_dir_dentries(trans, root,
5783                                                    BTRFS_I(dir_inode), ctx);
5784                         btrfs_add_delayed_iput(dir_inode);
5785                         if (ret)
5786                                 goto out;
5787                 }
5788                 path->slots[0]++;
5789         }
5790         ret = 0;
5791 out:
5792         btrfs_free_path(path);
5793         return ret;
5794 }
5795 
5796 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5797                              struct btrfs_root *root,
5798                              struct btrfs_path *path,
5799                              struct btrfs_log_ctx *ctx)
5800 {
5801         struct btrfs_key found_key;
5802 
5803         btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5804 
5805         while (true) {
5806                 struct btrfs_fs_info *fs_info = root->fs_info;
5807                 const u64 last_committed = fs_info->last_trans_committed;
5808                 struct extent_buffer *leaf = path->nodes[0];
5809                 int slot = path->slots[0];
5810                 struct btrfs_key search_key;
5811                 struct inode *inode;
5812                 int ret = 0;
5813 
5814                 btrfs_release_path(path);
5815 
5816                 search_key.objectid = found_key.offset;
5817                 search_key.type = BTRFS_INODE_ITEM_KEY;
5818                 search_key.offset = 0;
5819                 inode = btrfs_iget(fs_info->sb, &search_key, root, NULL);
5820                 if (IS_ERR(inode))
5821                         return PTR_ERR(inode);
5822 
5823                 if (BTRFS_I(inode)->generation > last_committed)
5824                         ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5825                                               LOG_INODE_EXISTS,
5826                                               0, LLONG_MAX, ctx);
5827                 btrfs_add_delayed_iput(inode);
5828                 if (ret)
5829                         return ret;
5830 
5831                 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5832                         break;
5833 
5834                 search_key.type = BTRFS_INODE_REF_KEY;
5835                 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5836                 if (ret < 0)
5837                         return ret;
5838 
5839                 leaf = path->nodes[0];
5840                 slot = path->slots[0];
5841                 if (slot >= btrfs_header_nritems(leaf)) {
5842                         ret = btrfs_next_leaf(root, path);
5843                         if (ret < 0)
5844                                 return ret;
5845                         else if (ret > 0)
5846                                 return -ENOENT;
5847                         leaf = path->nodes[0];
5848                         slot = path->slots[0];
5849                 }
5850 
5851                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5852                 if (found_key.objectid != search_key.objectid ||
5853                     found_key.type != BTRFS_INODE_REF_KEY)
5854                         return -ENOENT;
5855         }
5856         return 0;
5857 }
5858 
5859 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5860                                   struct btrfs_inode *inode,
5861                                   struct dentry *parent,
5862                                   struct btrfs_log_ctx *ctx)
5863 {
5864         struct btrfs_root *root = inode->root;
5865         struct btrfs_fs_info *fs_info = root->fs_info;
5866         struct dentry *old_parent = NULL;
5867         struct super_block *sb = inode->vfs_inode.i_sb;
5868         int ret = 0;
5869 
5870         while (true) {
5871                 if (!parent || d_really_is_negative(parent) ||
5872                     sb != parent->d_sb)
5873                         break;
5874 
5875                 inode = BTRFS_I(d_inode(parent));
5876                 if (root != inode->root)
5877                         break;
5878 
5879                 if (inode->generation > fs_info->last_trans_committed) {
5880                         ret = btrfs_log_inode(trans, root, inode,
5881                                         LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5882                         if (ret)
5883                                 break;
5884                 }
5885                 if (IS_ROOT(parent))
5886                         break;
5887 
5888                 parent = dget_parent(parent);
5889                 dput(old_parent);
5890                 old_parent = parent;
5891         }
5892         dput(old_parent);
5893 
5894         return ret;
5895 }
5896 
5897 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5898                                  struct btrfs_inode *inode,
5899                                  struct dentry *parent,
5900                                  struct btrfs_log_ctx *ctx)
5901 {
5902         struct btrfs_root *root = inode->root;
5903         const u64 ino = btrfs_ino(inode);
5904         struct btrfs_path *path;
5905         struct btrfs_key search_key;
5906         int ret;
5907 
5908         /*
5909          * For a single hard link case, go through a fast path that does not
5910          * need to iterate the fs/subvolume tree.
5911          */
5912         if (inode->vfs_inode.i_nlink < 2)
5913                 return log_new_ancestors_fast(trans, inode, parent, ctx);
5914 
5915         path = btrfs_alloc_path();
5916         if (!path)
5917                 return -ENOMEM;
5918 
5919         search_key.objectid = ino;
5920         search_key.type = BTRFS_INODE_REF_KEY;
5921         search_key.offset = 0;
5922 again:
5923         ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5924         if (ret < 0)
5925                 goto out;
5926         if (ret == 0)
5927                 path->slots[0]++;
5928 
5929         while (true) {
5930                 struct extent_buffer *leaf = path->nodes[0];
5931                 int slot = path->slots[0];
5932                 struct btrfs_key found_key;
5933 
5934                 if (slot >= btrfs_header_nritems(leaf)) {
5935                         ret = btrfs_next_leaf(root, path);
5936                         if (ret < 0)
5937                                 goto out;
5938                         else if (ret > 0)
5939                                 break;
5940                         continue;
5941                 }
5942 
5943                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5944                 if (found_key.objectid != ino ||
5945                     found_key.type > BTRFS_INODE_EXTREF_KEY)
5946                         break;
5947 
5948                 /*
5949                  * Don't deal with extended references because they are rare
5950                  * cases and too complex to deal with (we would need to keep
5951                  * track of which subitem we are processing for each item in
5952                  * this loop, etc). So just return some error to fallback to
5953                  * a transaction commit.
5954                  */
5955                 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
5956                         ret = -EMLINK;
5957                         goto out;
5958                 }
5959 
5960                 /*
5961                  * Logging ancestors needs to do more searches on the fs/subvol
5962                  * tree, so it releases the path as needed to avoid deadlocks.
5963                  * Keep track of the last inode ref key and resume from that key
5964                  * after logging all new ancestors for the current hard link.
5965                  */
5966                 memcpy(&search_key, &found_key, sizeof(search_key));
5967 
5968                 ret = log_new_ancestors(trans, root, path, ctx);
5969                 if (ret)
5970                         goto out;
5971                 btrfs_release_path(path);
5972                 goto again;
5973         }
5974         ret = 0;
5975 out:
5976         btrfs_free_path(path);
5977         return ret;
5978 }
5979 
5980 /*
5981  * helper function around btrfs_log_inode to make sure newly created
5982  * parent directories also end up in the log.  A minimal inode and backref
5983  * only logging is done of any parent directories that are older than
5984  * the last committed transaction
5985  */
5986 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5987                                   struct btrfs_inode *inode,
5988                                   struct dentry *parent,
5989                                   const loff_t start,
5990                                   const loff_t end,
5991                                   int inode_only,
5992                                   struct btrfs_log_ctx *ctx)
5993 {
5994         struct btrfs_root *root = inode->root;
5995         struct btrfs_fs_info *fs_info = root->fs_info;
5996         struct super_block *sb;
5997         int ret = 0;
5998         u64 last_committed = fs_info->last_trans_committed;
5999         bool log_dentries = false;
6000 
6001         sb = inode->vfs_inode.i_sb;
6002 
6003         if (btrfs_test_opt(fs_info, NOTREELOG)) {
6004                 ret = 1;
6005                 goto end_no_trans;
6006         }
6007 
6008         /*
6009          * The prev transaction commit doesn't complete, we need do
6010          * full commit by ourselves.
6011          */
6012         if (fs_info->last_trans_log_full_commit >
6013             fs_info->last_trans_committed) {
6014                 ret = 1;
6015                 goto end_no_trans;
6016         }
6017 
6018         if (btrfs_root_refs(&root->root_item) == 0) {
6019                 ret = 1;
6020                 goto end_no_trans;
6021         }
6022 
6023         ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
6024                         last_committed);
6025         if (ret)
6026                 goto end_no_trans;
6027 
6028         /*
6029          * Skip already logged inodes or inodes corresponding to tmpfiles
6030          * (since logging them is pointless, a link count of 0 means they
6031          * will never be accessible).
6032          */
6033         if (btrfs_inode_in_log(inode, trans->transid) ||
6034             inode->vfs_inode.i_nlink == 0) {
6035                 ret = BTRFS_NO_LOG_SYNC;
6036                 goto end_no_trans;
6037         }
6038 
6039         ret = start_log_trans(trans, root, ctx);
6040         if (ret)
6041                 goto end_no_trans;
6042 
6043         ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
6044         if (ret)
6045                 goto end_trans;
6046 
6047         /*
6048          * for regular files, if its inode is already on disk, we don't
6049          * have to worry about the parents at all.  This is because
6050          * we can use the last_unlink_trans field to record renames
6051          * and other fun in this file.
6052          */
6053         if (S_ISREG(inode->vfs_inode.i_mode) &&
6054             inode->generation <= last_committed &&
6055             inode->last_unlink_trans <= last_committed) {
6056                 ret = 0;
6057                 goto end_trans;
6058         }
6059 
6060         if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6061                 log_dentries = true;
6062 
6063         /*
6064          * On unlink we must make sure all our current and old parent directory
6065          * inodes are fully logged. This is to prevent leaving dangling
6066          * directory index entries in directories that were our parents but are
6067          * not anymore. Not doing this results in old parent directory being
6068          * impossible to delete after log replay (rmdir will always fail with
6069          * error -ENOTEMPTY).
6070          *
6071          * Example 1:
6072          *
6073          * mkdir testdir
6074          * touch testdir/foo
6075          * ln testdir/foo testdir/bar
6076          * sync
6077          * unlink testdir/bar
6078          * xfs_io -c fsync testdir/foo
6079          * <power failure>
6080          * mount fs, triggers log replay
6081          *
6082          * If we don't log the parent directory (testdir), after log replay the
6083          * directory still has an entry pointing to the file inode using the bar
6084          * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6085          * the file inode has a link count of 1.
6086          *
6087          * Example 2:
6088          *
6089          * mkdir testdir
6090          * touch foo
6091          * ln foo testdir/foo2
6092          * ln foo testdir/foo3
6093          * sync
6094          * unlink testdir/foo3
6095          * xfs_io -c fsync foo
6096          * <power failure>
6097          * mount fs, triggers log replay
6098          *
6099          * Similar as the first example, after log replay the parent directory
6100          * testdir still has an entry pointing to the inode file with name foo3
6101          * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6102          * and has a link count of 2.
6103          */
6104         if (inode->last_unlink_trans > last_committed) {
6105                 ret = btrfs_log_all_parents(trans, inode, ctx);
6106                 if (ret)
6107                         goto end_trans;
6108         }
6109 
6110         ret = log_all_new_ancestors(trans, inode, parent, ctx);
6111         if (ret)
6112                 goto end_trans;
6113 
6114         if (log_dentries)
6115                 ret = log_new_dir_dentries(trans, root, inode, ctx);
6116         else
6117                 ret = 0;
6118 end_trans:
6119         if (ret < 0) {
6120                 btrfs_set_log_full_commit(trans);
6121                 ret = 1;
6122         }
6123 
6124         if (ret)
6125                 btrfs_remove_log_ctx(root, ctx);
6126         btrfs_end_log_trans(root);
6127 end_no_trans:
6128         return ret;
6129 }
6130 
6131 /*
6132  * it is not safe to log dentry if the chunk root has added new
6133  * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
6134  * If this returns 1, you must commit the transaction to safely get your
6135  * data on disk.
6136  */
6137 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6138                           struct dentry *dentry,
6139                           const loff_t start,
6140                           const loff_t end,
6141                           struct btrfs_log_ctx *ctx)
6142 {
6143         struct dentry *parent = dget_parent(dentry);
6144         int ret;
6145 
6146         ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6147                                      start, end, LOG_INODE_ALL, ctx);
6148         dput(parent);
6149 
6150         return ret;
6151 }
6152 
6153 /*
6154  * should be called during mount to recover any replay any log trees
6155  * from the FS
6156  */
6157 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6158 {
6159         int ret;
6160         struct btrfs_path *path;
6161         struct btrfs_trans_handle *trans;
6162         struct btrfs_key key;
6163         struct btrfs_key found_key;
6164         struct btrfs_key tmp_key;
6165         struct btrfs_root *log;
6166         struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6167         struct walk_control wc = {
6168                 .process_func = process_one_buffer,
6169                 .stage = LOG_WALK_PIN_ONLY,
6170         };
6171 
6172         path = btrfs_alloc_path();
6173         if (!path)
6174                 return -ENOMEM;
6175 
6176         set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6177 
6178         trans = btrfs_start_transaction(fs_info->tree_root, 0);
6179         if (IS_ERR(trans)) {
6180                 ret = PTR_ERR(trans);
6181                 goto error;
6182         }
6183 
6184         wc.trans = trans;
6185         wc.pin = 1;
6186 
6187         ret = walk_log_tree(trans, log_root_tree, &wc);
6188         if (ret) {
6189                 btrfs_handle_fs_error(fs_info, ret,
6190                         "Failed to pin buffers while recovering log root tree.");
6191                 goto error;
6192         }
6193 
6194 again:
6195         key.objectid = BTRFS_TREE_LOG_OBJECTID;
6196         key.offset = (u64)-1;
6197         key.type = BTRFS_ROOT_ITEM_KEY;
6198 
6199         while (1) {
6200                 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6201 
6202                 if (ret < 0) {
6203                         btrfs_handle_fs_error(fs_info, ret,
6204                                     "Couldn't find tree log root.");
6205                         goto error;
6206                 }
6207                 if (ret > 0) {
6208                         if (path->slots[0] == 0)
6209                                 break;
6210                         path->slots[0]--;
6211                 }
6212                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6213                                       path->slots[0]);
6214                 btrfs_release_path(path);
6215                 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6216                         break;
6217 
6218                 log = btrfs_read_fs_root(log_root_tree, &found_key);
6219                 if (IS_ERR(log)) {
6220                         ret = PTR_ERR(log);
6221                         btrfs_handle_fs_error(fs_info, ret,
6222                                     "Couldn't read tree log root.");
6223                         goto error;
6224                 }
6225 
6226                 tmp_key.objectid = found_key.offset;
6227                 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6228                 tmp_key.offset = (u64)-1;
6229 
6230                 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6231                 if (IS_ERR(wc.replay_dest)) {
6232                         ret = PTR_ERR(wc.replay_dest);
6233 
6234                         /*
6235                          * We didn't find the subvol, likely because it was
6236                          * deleted.  This is ok, simply skip this log and go to
6237                          * the next one.
6238                          *
6239                          * We need to exclude the root because we can't have
6240                          * other log replays overwriting this log as we'll read
6241                          * it back in a few more times.  This will keep our
6242                          * block from being modified, and we'll just bail for
6243                          * each subsequent pass.
6244                          */
6245                         if (ret == -ENOENT)
6246                                 ret = btrfs_pin_extent_for_log_replay(fs_info,
6247                                                         log->node->start,
6248                                                         log->node->len);
6249                         free_extent_buffer(log->node);
6250                         free_extent_buffer(log->commit_root);
6251                         kfree(log);
6252 
6253                         if (!ret)
6254                                 goto next;
6255                         btrfs_handle_fs_error(fs_info, ret,
6256                                 "Couldn't read target root for tree log recovery.");
6257                         goto error;
6258                 }
6259 
6260                 wc.replay_dest->log_root = log;
6261                 btrfs_record_root_in_trans(trans, wc.replay_dest);
6262                 ret = walk_log_tree(trans, log, &wc);
6263 
6264                 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6265                         ret = fixup_inode_link_counts(trans, wc.replay_dest,
6266                                                       path);
6267                 }
6268 
6269                 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6270                         struct btrfs_root *root = wc.replay_dest;
6271 
6272                         btrfs_release_path(path);
6273 
6274                         /*
6275                          * We have just replayed everything, and the highest
6276                          * objectid of fs roots probably has changed in case
6277                          * some inode_item's got replayed.
6278                          *
6279                          * root->objectid_mutex is not acquired as log replay
6280                          * could only happen during mount.
6281                          */
6282                         ret = btrfs_find_highest_objectid(root,
6283                                                   &root->highest_objectid);
6284                 }
6285 
6286                 wc.replay_dest->log_root = NULL;
6287                 free_extent_buffer(log->node);
6288                 free_extent_buffer(log->commit_root);
6289                 kfree(log);
6290 
6291                 if (ret)
6292                         goto error;
6293 next:
6294                 if (found_key.offset == 0)
6295                         break;
6296                 key.offset = found_key.offset - 1;
6297         }
6298         btrfs_release_path(path);
6299 
6300         /* step one is to pin it all, step two is to replay just inodes */
6301         if (wc.pin) {
6302                 wc.pin = 0;
6303                 wc.process_func = replay_one_buffer;
6304                 wc.stage = LOG_WALK_REPLAY_INODES;
6305                 goto again;
6306         }
6307         /* step three is to replay everything */
6308         if (wc.stage < LOG_WALK_REPLAY_ALL) {
6309                 wc.stage++;
6310                 goto again;
6311         }
6312 
6313         btrfs_free_path(path);
6314 
6315         /* step 4: commit the transaction, which also unpins the blocks */
6316         ret = btrfs_commit_transaction(trans);
6317         if (ret)
6318                 return ret;
6319 
6320         free_extent_buffer(log_root_tree->node);
6321         log_root_tree->log_root = NULL;
6322         clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6323         kfree(log_root_tree);
6324 
6325         return 0;
6326 error:
6327         if (wc.trans)
6328                 btrfs_end_transaction(wc.trans);
6329         btrfs_free_path(path);
6330         return ret;
6331 }
6332 
6333 /*
6334  * there are some corner cases where we want to force a full
6335  * commit instead of allowing a directory to be logged.
6336  *
6337  * They revolve around files there were unlinked from the directory, and
6338  * this function updates the parent directory so that a full commit is
6339  * properly done if it is fsync'd later after the unlinks are done.
6340  *
6341  * Must be called before the unlink operations (updates to the subvolume tree,
6342  * inodes, etc) are done.
6343  */
6344 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6345                              struct btrfs_inode *dir, struct btrfs_inode *inode,
6346                              int for_rename)
6347 {
6348         /*
6349          * when we're logging a file, if it hasn't been renamed
6350          * or unlinked, and its inode is fully committed on disk,
6351          * we don't have to worry about walking up the directory chain
6352          * to log its parents.
6353          *
6354          * So, we use the last_unlink_trans field to put this transid
6355          * into the file.  When the file is logged we check it and
6356          * don't log the parents if the file is fully on disk.
6357          */
6358         mutex_lock(&inode->log_mutex);
6359         inode->last_unlink_trans = trans->transid;
6360         mutex_unlock(&inode->log_mutex);
6361 
6362         /*
6363          * if this directory was already logged any new
6364          * names for this file/dir will get recorded
6365          */
6366         if (dir->logged_trans == trans->transid)
6367                 return;
6368 
6369         /*
6370          * if the inode we're about to unlink was logged,
6371          * the log will be properly updated for any new names
6372          */
6373         if (inode->logged_trans == trans->transid)
6374                 return;
6375 
6376         /*
6377          * when renaming files across directories, if the directory
6378          * there we're unlinking from gets fsync'd later on, there's
6379          * no way to find the destination directory later and fsync it
6380          * properly.  So, we have to be conservative and force commits
6381          * so the new name gets discovered.
6382          */
6383         if (for_rename)
6384                 goto record;
6385 
6386         /* we can safely do the unlink without any special recording */
6387         return;
6388 
6389 record:
6390         mutex_lock(&dir->log_mutex);
6391         dir->last_unlink_trans = trans->transid;
6392         mutex_unlock(&dir->log_mutex);
6393 }
6394 
6395 /*
6396  * Make sure that if someone attempts to fsync the parent directory of a deleted
6397  * snapshot, it ends up triggering a transaction commit. This is to guarantee
6398  * that after replaying the log tree of the parent directory's root we will not
6399  * see the snapshot anymore and at log replay time we will not see any log tree
6400  * corresponding to the deleted snapshot's root, which could lead to replaying
6401  * it after replaying the log tree of the parent directory (which would replay
6402  * the snapshot delete operation).
6403  *
6404  * Must be called before the actual snapshot destroy operation (updates to the
6405  * parent root and tree of tree roots trees, etc) are done.
6406  */
6407 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6408                                    struct btrfs_inode *dir)
6409 {
6410         mutex_lock(&dir->log_mutex);
6411         dir->last_unlink_trans = trans->transid;
6412         mutex_unlock(&dir->log_mutex);
6413 }
6414 
6415 /*
6416  * Call this after adding a new name for a file and it will properly
6417  * update the log to reflect the new name.
6418  *
6419  * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6420  * true (because it's not used).
6421  *
6422  * Return value depends on whether @sync_log is true or false.
6423  * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6424  *            committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6425  *            otherwise.
6426  * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6427  *             to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6428  *             or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6429  *             committed (without attempting to sync the log).
6430  */
6431 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6432                         struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6433                         struct dentry *parent,
6434                         bool sync_log, struct btrfs_log_ctx *ctx)
6435 {
6436         struct btrfs_fs_info *fs_info = trans->fs_info;
6437         int ret;
6438 
6439         /*
6440          * this will force the logging code to walk the dentry chain
6441          * up for the file
6442          */
6443         if (!S_ISDIR(inode->vfs_inode.i_mode))
6444                 inode->last_unlink_trans = trans->transid;
6445 
6446         /*
6447          * if this inode hasn't been logged and directory we're renaming it
6448          * from hasn't been logged, we don't need to log it
6449          */
6450         if (inode->logged_trans <= fs_info->last_trans_committed &&
6451             (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6452                 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6453                         BTRFS_DONT_NEED_LOG_SYNC;
6454 
6455         if (sync_log) {
6456                 struct btrfs_log_ctx ctx2;
6457 
6458                 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6459                 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6460                                              LOG_INODE_EXISTS, &ctx2);
6461                 if (ret == BTRFS_NO_LOG_SYNC)
6462                         return BTRFS_DONT_NEED_TRANS_COMMIT;
6463                 else if (ret)
6464                         return BTRFS_NEED_TRANS_COMMIT;
6465 
6466                 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6467                 if (ret)
6468                         return BTRFS_NEED_TRANS_COMMIT;
6469                 return BTRFS_DONT_NEED_TRANS_COMMIT;
6470         }
6471 
6472         ASSERT(ctx);
6473         ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6474                                      LOG_INODE_EXISTS, ctx);
6475         if (ret == BTRFS_NO_LOG_SYNC)
6476                 return BTRFS_DONT_NEED_LOG_SYNC;
6477         else if (ret)
6478                 return BTRFS_NEED_TRANS_COMMIT;
6479 
6480         return BTRFS_NEED_LOG_SYNC;
6481 }
6482