2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
30 /* magic values for the inode_only field in btrfs_log_inode:
32 * LOG_INODE_ALL means to log everything
33 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 #define LOG_INODE_ALL 0
37 #define LOG_INODE_EXISTS 1
40 * directory trouble cases
42 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
43 * log, we must force a full commit before doing an fsync of the directory
44 * where the unlink was done.
45 * ---> record transid of last unlink/rename per directory
49 * rename foo/some_dir foo2/some_dir
51 * fsync foo/some_dir/some_file
53 * The fsync above will unlink the original some_dir without recording
54 * it in its new location (foo2). After a crash, some_dir will be gone
55 * unless the fsync of some_file forces a full commit
57 * 2) we must log any new names for any file or dir that is in the fsync
58 * log. ---> check inode while renaming/linking.
60 * 2a) we must log any new names for any file or dir during rename
61 * when the directory they are being removed from was logged.
62 * ---> check inode and old parent dir during rename
64 * 2a is actually the more important variant. With the extra logging
65 * a crash might unlink the old name without recreating the new one
67 * 3) after a crash, we must go through any directories with a link count
68 * of zero and redo the rm -rf
75 * The directory f1 was fully removed from the FS, but fsync was never
76 * called on f1, only its parent dir. After a crash the rm -rf must
77 * be replayed. This must be able to recurse down the entire
78 * directory tree. The inode link count fixup code takes care of the
83 * stages for the tree walking. The first
84 * stage (0) is to only pin down the blocks we find
85 * the second stage (1) is to make sure that all the inodes
86 * we find in the log are created in the subvolume.
88 * The last stage is to deal with directories and links and extents
89 * and all the other fun semantics
91 #define LOG_WALK_PIN_ONLY 0
92 #define LOG_WALK_REPLAY_INODES 1
93 #define LOG_WALK_REPLAY_DIR_INDEX 2
94 #define LOG_WALK_REPLAY_ALL 3
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct inode *inode,
101 struct btrfs_log_ctx *ctx);
102 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
105 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
145 mutex_lock(&root->log_mutex);
147 if (root->log_root) {
148 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
153 if (!root->log_start_pid) {
154 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
155 root->log_start_pid = current->pid;
156 } else if (root->log_start_pid != current->pid) {
157 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
160 mutex_lock(&root->fs_info->tree_log_mutex);
161 if (!root->fs_info->log_root_tree)
162 ret = btrfs_init_log_root_tree(trans, root->fs_info);
163 mutex_unlock(&root->fs_info->tree_log_mutex);
167 ret = btrfs_add_log_tree(trans, root);
171 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
172 root->log_start_pid = current->pid;
175 atomic_inc(&root->log_batch);
176 atomic_inc(&root->log_writers);
178 int index = root->log_transid % 2;
179 list_add_tail(&ctx->list, &root->log_ctxs[index]);
180 ctx->log_transid = root->log_transid;
184 mutex_unlock(&root->log_mutex);
189 * returns 0 if there was a log transaction running and we were able
190 * to join, or returns -ENOENT if there were not transactions
193 static int join_running_log_trans(struct btrfs_root *root)
201 mutex_lock(&root->log_mutex);
202 if (root->log_root) {
204 atomic_inc(&root->log_writers);
206 mutex_unlock(&root->log_mutex);
211 * This either makes the current running log transaction wait
212 * until you call btrfs_end_log_trans() or it makes any future
213 * log transactions wait until you call btrfs_end_log_trans()
215 int btrfs_pin_log_trans(struct btrfs_root *root)
219 mutex_lock(&root->log_mutex);
220 atomic_inc(&root->log_writers);
221 mutex_unlock(&root->log_mutex);
226 * indicate we're done making changes to the log tree
227 * and wake up anyone waiting to do a sync
229 void btrfs_end_log_trans(struct btrfs_root *root)
231 if (atomic_dec_and_test(&root->log_writers)) {
233 * Implicit memory barrier after atomic_dec_and_test
235 if (waitqueue_active(&root->log_writer_wait))
236 wake_up(&root->log_writer_wait);
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
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
253 /* should we write out the extent buffer? This is used
254 * while flushing the log tree to disk during a sync
258 /* should we wait for the extent buffer io to finish? Also used
259 * while flushing the log tree to disk for a sync
263 /* pin only walk, we record which extents on disk belong to the
268 /* what stage of the replay code we're currently in */
271 /* the root we are currently replaying */
272 struct btrfs_root *replay_dest;
274 /* the trans handle for the current replay */
275 struct btrfs_trans_handle *trans;
277 /* the function that gets used to process blocks we find in the
278 * tree. Note the extent_buffer might not be up to date when it is
279 * passed in, and it must be checked or read if you need the data
282 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
283 struct walk_control *wc, u64 gen);
287 * process_func used to pin down extents, write them or wait on them
289 static int process_one_buffer(struct btrfs_root *log,
290 struct extent_buffer *eb,
291 struct walk_control *wc, u64 gen)
296 * If this fs is mixed then we need to be able to process the leaves to
297 * pin down any logged extents, so we have to read the block.
299 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
300 ret = btrfs_read_buffer(eb, gen);
306 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
309 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
310 if (wc->pin && btrfs_header_level(eb) == 0)
311 ret = btrfs_exclude_logged_extents(log, eb);
313 btrfs_write_tree_block(eb);
315 btrfs_wait_tree_block_writeback(eb);
321 * Item overwrite used by replay and tree logging. eb, slot and key all refer
322 * to the src data we are copying out.
324 * root is the tree we are copying into, and path is a scratch
325 * path for use in this function (it should be released on entry and
326 * will be released on exit).
328 * If the key is already in the destination tree the existing item is
329 * overwritten. If the existing item isn't big enough, it is extended.
330 * If it is too large, it is truncated.
332 * If the key isn't in the destination yet, a new item is inserted.
334 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
335 struct btrfs_root *root,
336 struct btrfs_path *path,
337 struct extent_buffer *eb, int slot,
338 struct btrfs_key *key)
342 u64 saved_i_size = 0;
343 int save_old_i_size = 0;
344 unsigned long src_ptr;
345 unsigned long dst_ptr;
346 int overwrite_root = 0;
347 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
349 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
352 item_size = btrfs_item_size_nr(eb, slot);
353 src_ptr = btrfs_item_ptr_offset(eb, slot);
355 /* look for the key in the destination tree */
356 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
363 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
365 if (dst_size != item_size)
368 if (item_size == 0) {
369 btrfs_release_path(path);
372 dst_copy = kmalloc(item_size, GFP_NOFS);
373 src_copy = kmalloc(item_size, GFP_NOFS);
374 if (!dst_copy || !src_copy) {
375 btrfs_release_path(path);
381 read_extent_buffer(eb, src_copy, src_ptr, item_size);
383 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
384 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
386 ret = memcmp(dst_copy, src_copy, item_size);
391 * they have the same contents, just return, this saves
392 * us from cowing blocks in the destination tree and doing
393 * extra writes that may not have been done by a previous
397 btrfs_release_path(path);
402 * We need to load the old nbytes into the inode so when we
403 * replay the extents we've logged we get the right nbytes.
406 struct btrfs_inode_item *item;
410 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
411 struct btrfs_inode_item);
412 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
413 item = btrfs_item_ptr(eb, slot,
414 struct btrfs_inode_item);
415 btrfs_set_inode_nbytes(eb, item, nbytes);
418 * If this is a directory we need to reset the i_size to
419 * 0 so that we can set it up properly when replaying
420 * the rest of the items in this log.
422 mode = btrfs_inode_mode(eb, item);
424 btrfs_set_inode_size(eb, item, 0);
426 } else if (inode_item) {
427 struct btrfs_inode_item *item;
431 * New inode, set nbytes to 0 so that the nbytes comes out
432 * properly when we replay the extents.
434 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
435 btrfs_set_inode_nbytes(eb, item, 0);
438 * If this is a directory we need to reset the i_size to 0 so
439 * that we can set it up properly when replaying the rest of
440 * the items in this log.
442 mode = btrfs_inode_mode(eb, item);
444 btrfs_set_inode_size(eb, item, 0);
447 btrfs_release_path(path);
448 /* try to insert the key into the destination tree */
449 path->skip_release_on_error = 1;
450 ret = btrfs_insert_empty_item(trans, root, path,
452 path->skip_release_on_error = 0;
454 /* make sure any existing item is the correct size */
455 if (ret == -EEXIST || ret == -EOVERFLOW) {
457 found_size = btrfs_item_size_nr(path->nodes[0],
459 if (found_size > item_size)
460 btrfs_truncate_item(root, path, item_size, 1);
461 else if (found_size < item_size)
462 btrfs_extend_item(root, path,
463 item_size - found_size);
467 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
470 /* don't overwrite an existing inode if the generation number
471 * was logged as zero. This is done when the tree logging code
472 * is just logging an inode to make sure it exists after recovery.
474 * Also, don't overwrite i_size on directories during replay.
475 * log replay inserts and removes directory items based on the
476 * state of the tree found in the subvolume, and i_size is modified
479 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
480 struct btrfs_inode_item *src_item;
481 struct btrfs_inode_item *dst_item;
483 src_item = (struct btrfs_inode_item *)src_ptr;
484 dst_item = (struct btrfs_inode_item *)dst_ptr;
486 if (btrfs_inode_generation(eb, src_item) == 0) {
487 struct extent_buffer *dst_eb = path->nodes[0];
488 const u64 ino_size = btrfs_inode_size(eb, src_item);
491 * For regular files an ino_size == 0 is used only when
492 * logging that an inode exists, as part of a directory
493 * fsync, and the inode wasn't fsynced before. In this
494 * case don't set the size of the inode in the fs/subvol
495 * tree, otherwise we would be throwing valid data away.
497 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
498 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
500 struct btrfs_map_token token;
502 btrfs_init_map_token(&token);
503 btrfs_set_token_inode_size(dst_eb, dst_item,
509 if (overwrite_root &&
510 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
511 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
513 saved_i_size = btrfs_inode_size(path->nodes[0],
518 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
521 if (save_old_i_size) {
522 struct btrfs_inode_item *dst_item;
523 dst_item = (struct btrfs_inode_item *)dst_ptr;
524 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
527 /* make sure the generation is filled in */
528 if (key->type == BTRFS_INODE_ITEM_KEY) {
529 struct btrfs_inode_item *dst_item;
530 dst_item = (struct btrfs_inode_item *)dst_ptr;
531 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
532 btrfs_set_inode_generation(path->nodes[0], dst_item,
537 btrfs_mark_buffer_dirty(path->nodes[0]);
538 btrfs_release_path(path);
543 * simple helper to read an inode off the disk from a given root
544 * This can only be called for subvolume roots and not for the log
546 static noinline struct inode *read_one_inode(struct btrfs_root *root,
549 struct btrfs_key key;
552 key.objectid = objectid;
553 key.type = BTRFS_INODE_ITEM_KEY;
555 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
558 } else if (is_bad_inode(inode)) {
565 /* replays a single extent in 'eb' at 'slot' with 'key' into the
566 * subvolume 'root'. path is released on entry and should be released
569 * extents in the log tree have not been allocated out of the extent
570 * tree yet. So, this completes the allocation, taking a reference
571 * as required if the extent already exists or creating a new extent
572 * if it isn't in the extent allocation tree yet.
574 * The extent is inserted into the file, dropping any existing extents
575 * from the file that overlap the new one.
577 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
578 struct btrfs_root *root,
579 struct btrfs_path *path,
580 struct extent_buffer *eb, int slot,
581 struct btrfs_key *key)
585 u64 start = key->offset;
587 struct btrfs_file_extent_item *item;
588 struct inode *inode = NULL;
592 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
593 found_type = btrfs_file_extent_type(eb, item);
595 if (found_type == BTRFS_FILE_EXTENT_REG ||
596 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
597 nbytes = btrfs_file_extent_num_bytes(eb, item);
598 extent_end = start + nbytes;
601 * We don't add to the inodes nbytes if we are prealloc or a
604 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
606 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
607 size = btrfs_file_extent_inline_len(eb, slot, item);
608 nbytes = btrfs_file_extent_ram_bytes(eb, item);
609 extent_end = ALIGN(start + size, root->sectorsize);
615 inode = read_one_inode(root, key->objectid);
622 * first check to see if we already have this extent in the
623 * file. This must be done before the btrfs_drop_extents run
624 * so we don't try to drop this extent.
626 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
630 (found_type == BTRFS_FILE_EXTENT_REG ||
631 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
632 struct btrfs_file_extent_item cmp1;
633 struct btrfs_file_extent_item cmp2;
634 struct btrfs_file_extent_item *existing;
635 struct extent_buffer *leaf;
637 leaf = path->nodes[0];
638 existing = btrfs_item_ptr(leaf, path->slots[0],
639 struct btrfs_file_extent_item);
641 read_extent_buffer(eb, &cmp1, (unsigned long)item,
643 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
647 * we already have a pointer to this exact extent,
648 * we don't have to do anything
650 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
651 btrfs_release_path(path);
655 btrfs_release_path(path);
657 /* drop any overlapping extents */
658 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
662 if (found_type == BTRFS_FILE_EXTENT_REG ||
663 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
665 unsigned long dest_offset;
666 struct btrfs_key ins;
668 ret = btrfs_insert_empty_item(trans, root, path, key,
672 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
674 copy_extent_buffer(path->nodes[0], eb, dest_offset,
675 (unsigned long)item, sizeof(*item));
677 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
678 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
679 ins.type = BTRFS_EXTENT_ITEM_KEY;
680 offset = key->offset - btrfs_file_extent_offset(eb, item);
682 if (ins.objectid > 0) {
685 LIST_HEAD(ordered_sums);
687 * is this extent already allocated in the extent
688 * allocation tree? If so, just add a reference
690 ret = btrfs_lookup_data_extent(root, ins.objectid,
693 ret = btrfs_inc_extent_ref(trans, root,
694 ins.objectid, ins.offset,
695 0, root->root_key.objectid,
696 key->objectid, offset);
701 * insert the extent pointer in the extent
704 ret = btrfs_alloc_logged_file_extent(trans,
705 root, root->root_key.objectid,
706 key->objectid, offset, &ins);
710 btrfs_release_path(path);
712 if (btrfs_file_extent_compression(eb, item)) {
713 csum_start = ins.objectid;
714 csum_end = csum_start + ins.offset;
716 csum_start = ins.objectid +
717 btrfs_file_extent_offset(eb, item);
718 csum_end = csum_start +
719 btrfs_file_extent_num_bytes(eb, item);
722 ret = btrfs_lookup_csums_range(root->log_root,
723 csum_start, csum_end - 1,
728 * Now delete all existing cums in the csum root that
729 * cover our range. We do this because we can have an
730 * extent that is completely referenced by one file
731 * extent item and partially referenced by another
732 * file extent item (like after using the clone or
733 * extent_same ioctls). In this case if we end up doing
734 * the replay of the one that partially references the
735 * extent first, and we do not do the csum deletion
736 * below, we can get 2 csum items in the csum tree that
737 * overlap each other. For example, imagine our log has
738 * the two following file extent items:
740 * key (257 EXTENT_DATA 409600)
741 * extent data disk byte 12845056 nr 102400
742 * extent data offset 20480 nr 20480 ram 102400
744 * key (257 EXTENT_DATA 819200)
745 * extent data disk byte 12845056 nr 102400
746 * extent data offset 0 nr 102400 ram 102400
748 * Where the second one fully references the 100K extent
749 * that starts at disk byte 12845056, and the log tree
750 * has a single csum item that covers the entire range
753 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
755 * After the first file extent item is replayed, the
756 * csum tree gets the following csum item:
758 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
760 * Which covers the 20K sub-range starting at offset 20K
761 * of our extent. Now when we replay the second file
762 * extent item, if we do not delete existing csum items
763 * that cover any of its blocks, we end up getting two
764 * csum items in our csum tree that overlap each other:
766 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
767 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
769 * Which is a problem, because after this anyone trying
770 * to lookup up for the checksum of any block of our
771 * extent starting at an offset of 40K or higher, will
772 * end up looking at the second csum item only, which
773 * does not contain the checksum for any block starting
774 * at offset 40K or higher of our extent.
776 while (!list_empty(&ordered_sums)) {
777 struct btrfs_ordered_sum *sums;
778 sums = list_entry(ordered_sums.next,
779 struct btrfs_ordered_sum,
782 ret = btrfs_del_csums(trans,
783 root->fs_info->csum_root,
787 ret = btrfs_csum_file_blocks(trans,
788 root->fs_info->csum_root,
790 list_del(&sums->list);
796 btrfs_release_path(path);
798 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
799 /* inline extents are easy, we just overwrite them */
800 ret = overwrite_item(trans, root, path, eb, slot, key);
805 inode_add_bytes(inode, nbytes);
806 ret = btrfs_update_inode(trans, root, inode);
814 * when cleaning up conflicts between the directory names in the
815 * subvolume, directory names in the log and directory names in the
816 * inode back references, we may have to unlink inodes from directories.
818 * This is a helper function to do the unlink of a specific directory
821 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root,
823 struct btrfs_path *path,
825 struct btrfs_dir_item *di)
830 struct extent_buffer *leaf;
831 struct btrfs_key location;
834 leaf = path->nodes[0];
836 btrfs_dir_item_key_to_cpu(leaf, di, &location);
837 name_len = btrfs_dir_name_len(leaf, di);
838 name = kmalloc(name_len, GFP_NOFS);
842 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
843 btrfs_release_path(path);
845 inode = read_one_inode(root, location.objectid);
851 ret = link_to_fixup_dir(trans, root, path, location.objectid);
855 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
859 ret = btrfs_run_delayed_items(trans, root);
867 * helper function to see if a given name and sequence number found
868 * in an inode back reference are already in a directory and correctly
869 * point to this inode
871 static noinline int inode_in_dir(struct btrfs_root *root,
872 struct btrfs_path *path,
873 u64 dirid, u64 objectid, u64 index,
874 const char *name, int name_len)
876 struct btrfs_dir_item *di;
877 struct btrfs_key location;
880 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
881 index, name, name_len, 0);
882 if (di && !IS_ERR(di)) {
883 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
884 if (location.objectid != objectid)
888 btrfs_release_path(path);
890 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
891 if (di && !IS_ERR(di)) {
892 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
893 if (location.objectid != objectid)
899 btrfs_release_path(path);
904 * helper function to check a log tree for a named back reference in
905 * an inode. This is used to decide if a back reference that is
906 * found in the subvolume conflicts with what we find in the log.
908 * inode backreferences may have multiple refs in a single item,
909 * during replay we process one reference at a time, and we don't
910 * want to delete valid links to a file from the subvolume if that
911 * link is also in the log.
913 static noinline int backref_in_log(struct btrfs_root *log,
914 struct btrfs_key *key,
916 const char *name, int namelen)
918 struct btrfs_path *path;
919 struct btrfs_inode_ref *ref;
921 unsigned long ptr_end;
922 unsigned long name_ptr;
928 path = btrfs_alloc_path();
932 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
936 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
938 if (key->type == BTRFS_INODE_EXTREF_KEY) {
939 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
940 name, namelen, NULL))
946 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
947 ptr_end = ptr + item_size;
948 while (ptr < ptr_end) {
949 ref = (struct btrfs_inode_ref *)ptr;
950 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
951 if (found_name_len == namelen) {
952 name_ptr = (unsigned long)(ref + 1);
953 ret = memcmp_extent_buffer(path->nodes[0], name,
960 ptr = (unsigned long)(ref + 1) + found_name_len;
963 btrfs_free_path(path);
967 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
968 struct btrfs_root *root,
969 struct btrfs_path *path,
970 struct btrfs_root *log_root,
971 struct inode *dir, struct inode *inode,
972 struct extent_buffer *eb,
973 u64 inode_objectid, u64 parent_objectid,
974 u64 ref_index, char *name, int namelen,
980 struct extent_buffer *leaf;
981 struct btrfs_dir_item *di;
982 struct btrfs_key search_key;
983 struct btrfs_inode_extref *extref;
986 /* Search old style refs */
987 search_key.objectid = inode_objectid;
988 search_key.type = BTRFS_INODE_REF_KEY;
989 search_key.offset = parent_objectid;
990 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
992 struct btrfs_inode_ref *victim_ref;
994 unsigned long ptr_end;
996 leaf = path->nodes[0];
998 /* are we trying to overwrite a back ref for the root directory
999 * if so, just jump out, we're done
1001 if (search_key.objectid == search_key.offset)
1004 /* check all the names in this back reference to see
1005 * if they are in the log. if so, we allow them to stay
1006 * otherwise they must be unlinked as a conflict
1008 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1009 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1010 while (ptr < ptr_end) {
1011 victim_ref = (struct btrfs_inode_ref *)ptr;
1012 victim_name_len = btrfs_inode_ref_name_len(leaf,
1014 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1018 read_extent_buffer(leaf, victim_name,
1019 (unsigned long)(victim_ref + 1),
1022 if (!backref_in_log(log_root, &search_key,
1027 btrfs_release_path(path);
1029 ret = btrfs_unlink_inode(trans, root, dir,
1035 ret = btrfs_run_delayed_items(trans, root);
1043 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1047 * NOTE: we have searched root tree and checked the
1048 * coresponding ref, it does not need to check again.
1052 btrfs_release_path(path);
1054 /* Same search but for extended refs */
1055 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1056 inode_objectid, parent_objectid, 0,
1058 if (!IS_ERR_OR_NULL(extref)) {
1062 struct inode *victim_parent;
1064 leaf = path->nodes[0];
1066 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1067 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1069 while (cur_offset < item_size) {
1070 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1072 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1074 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1077 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1080 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1083 search_key.objectid = inode_objectid;
1084 search_key.type = BTRFS_INODE_EXTREF_KEY;
1085 search_key.offset = btrfs_extref_hash(parent_objectid,
1089 if (!backref_in_log(log_root, &search_key,
1090 parent_objectid, victim_name,
1093 victim_parent = read_one_inode(root,
1095 if (victim_parent) {
1097 btrfs_release_path(path);
1099 ret = btrfs_unlink_inode(trans, root,
1105 ret = btrfs_run_delayed_items(
1108 iput(victim_parent);
1119 cur_offset += victim_name_len + sizeof(*extref);
1123 btrfs_release_path(path);
1125 /* look for a conflicting sequence number */
1126 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1127 ref_index, name, namelen, 0);
1128 if (di && !IS_ERR(di)) {
1129 ret = drop_one_dir_item(trans, root, path, dir, di);
1133 btrfs_release_path(path);
1135 /* look for a conflicing name */
1136 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1138 if (di && !IS_ERR(di)) {
1139 ret = drop_one_dir_item(trans, root, path, dir, di);
1143 btrfs_release_path(path);
1148 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1149 u32 *namelen, char **name, u64 *index,
1150 u64 *parent_objectid)
1152 struct btrfs_inode_extref *extref;
1154 extref = (struct btrfs_inode_extref *)ref_ptr;
1156 *namelen = btrfs_inode_extref_name_len(eb, extref);
1157 *name = kmalloc(*namelen, GFP_NOFS);
1161 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1164 *index = btrfs_inode_extref_index(eb, extref);
1165 if (parent_objectid)
1166 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1171 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1172 u32 *namelen, char **name, u64 *index)
1174 struct btrfs_inode_ref *ref;
1176 ref = (struct btrfs_inode_ref *)ref_ptr;
1178 *namelen = btrfs_inode_ref_name_len(eb, ref);
1179 *name = kmalloc(*namelen, GFP_NOFS);
1183 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1185 *index = btrfs_inode_ref_index(eb, ref);
1191 * replay one inode back reference item found in the log tree.
1192 * eb, slot and key refer to the buffer and key found in the log tree.
1193 * root is the destination we are replaying into, and path is for temp
1194 * use by this function. (it should be released on return).
1196 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_root *log,
1199 struct btrfs_path *path,
1200 struct extent_buffer *eb, int slot,
1201 struct btrfs_key *key)
1203 struct inode *dir = NULL;
1204 struct inode *inode = NULL;
1205 unsigned long ref_ptr;
1206 unsigned long ref_end;
1210 int search_done = 0;
1211 int log_ref_ver = 0;
1212 u64 parent_objectid;
1215 int ref_struct_size;
1217 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1218 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1220 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1221 struct btrfs_inode_extref *r;
1223 ref_struct_size = sizeof(struct btrfs_inode_extref);
1225 r = (struct btrfs_inode_extref *)ref_ptr;
1226 parent_objectid = btrfs_inode_extref_parent(eb, r);
1228 ref_struct_size = sizeof(struct btrfs_inode_ref);
1229 parent_objectid = key->offset;
1231 inode_objectid = key->objectid;
1234 * it is possible that we didn't log all the parent directories
1235 * for a given inode. If we don't find the dir, just don't
1236 * copy the back ref in. The link count fixup code will take
1239 dir = read_one_inode(root, parent_objectid);
1245 inode = read_one_inode(root, inode_objectid);
1251 while (ref_ptr < ref_end) {
1253 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1254 &ref_index, &parent_objectid);
1256 * parent object can change from one array
1260 dir = read_one_inode(root, parent_objectid);
1266 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1272 /* if we already have a perfect match, we're done */
1273 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1274 ref_index, name, namelen)) {
1276 * look for a conflicting back reference in the
1277 * metadata. if we find one we have to unlink that name
1278 * of the file before we add our new link. Later on, we
1279 * overwrite any existing back reference, and we don't
1280 * want to create dangling pointers in the directory.
1284 ret = __add_inode_ref(trans, root, path, log,
1288 ref_index, name, namelen,
1297 /* insert our name */
1298 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1303 btrfs_update_inode(trans, root, inode);
1306 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1315 /* finally write the back reference in the inode */
1316 ret = overwrite_item(trans, root, path, eb, slot, key);
1318 btrfs_release_path(path);
1325 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root, u64 ino)
1330 ret = btrfs_insert_orphan_item(trans, root, ino);
1337 static int count_inode_extrefs(struct btrfs_root *root,
1338 struct inode *inode, struct btrfs_path *path)
1342 unsigned int nlink = 0;
1345 u64 inode_objectid = btrfs_ino(inode);
1348 struct btrfs_inode_extref *extref;
1349 struct extent_buffer *leaf;
1352 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1357 leaf = path->nodes[0];
1358 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1359 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1362 while (cur_offset < item_size) {
1363 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1364 name_len = btrfs_inode_extref_name_len(leaf, extref);
1368 cur_offset += name_len + sizeof(*extref);
1372 btrfs_release_path(path);
1374 btrfs_release_path(path);
1376 if (ret < 0 && ret != -ENOENT)
1381 static int count_inode_refs(struct btrfs_root *root,
1382 struct inode *inode, struct btrfs_path *path)
1385 struct btrfs_key key;
1386 unsigned int nlink = 0;
1388 unsigned long ptr_end;
1390 u64 ino = btrfs_ino(inode);
1393 key.type = BTRFS_INODE_REF_KEY;
1394 key.offset = (u64)-1;
1397 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1401 if (path->slots[0] == 0)
1406 btrfs_item_key_to_cpu(path->nodes[0], &key,
1408 if (key.objectid != ino ||
1409 key.type != BTRFS_INODE_REF_KEY)
1411 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1412 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1414 while (ptr < ptr_end) {
1415 struct btrfs_inode_ref *ref;
1417 ref = (struct btrfs_inode_ref *)ptr;
1418 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1420 ptr = (unsigned long)(ref + 1) + name_len;
1424 if (key.offset == 0)
1426 if (path->slots[0] > 0) {
1431 btrfs_release_path(path);
1433 btrfs_release_path(path);
1439 * There are a few corners where the link count of the file can't
1440 * be properly maintained during replay. So, instead of adding
1441 * lots of complexity to the log code, we just scan the backrefs
1442 * for any file that has been through replay.
1444 * The scan will update the link count on the inode to reflect the
1445 * number of back refs found. If it goes down to zero, the iput
1446 * will free the inode.
1448 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1449 struct btrfs_root *root,
1450 struct inode *inode)
1452 struct btrfs_path *path;
1455 u64 ino = btrfs_ino(inode);
1457 path = btrfs_alloc_path();
1461 ret = count_inode_refs(root, inode, path);
1467 ret = count_inode_extrefs(root, inode, path);
1475 if (nlink != inode->i_nlink) {
1476 set_nlink(inode, nlink);
1477 btrfs_update_inode(trans, root, inode);
1479 BTRFS_I(inode)->index_cnt = (u64)-1;
1481 if (inode->i_nlink == 0) {
1482 if (S_ISDIR(inode->i_mode)) {
1483 ret = replay_dir_deletes(trans, root, NULL, path,
1488 ret = insert_orphan_item(trans, root, ino);
1492 btrfs_free_path(path);
1496 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1497 struct btrfs_root *root,
1498 struct btrfs_path *path)
1501 struct btrfs_key key;
1502 struct inode *inode;
1504 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1505 key.type = BTRFS_ORPHAN_ITEM_KEY;
1506 key.offset = (u64)-1;
1508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1513 if (path->slots[0] == 0)
1518 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1519 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1520 key.type != BTRFS_ORPHAN_ITEM_KEY)
1523 ret = btrfs_del_item(trans, root, path);
1527 btrfs_release_path(path);
1528 inode = read_one_inode(root, key.offset);
1532 ret = fixup_inode_link_count(trans, root, inode);
1538 * fixup on a directory may create new entries,
1539 * make sure we always look for the highset possible
1542 key.offset = (u64)-1;
1546 btrfs_release_path(path);
1552 * record a given inode in the fixup dir so we can check its link
1553 * count when replay is done. The link count is incremented here
1554 * so the inode won't go away until we check it
1556 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1557 struct btrfs_root *root,
1558 struct btrfs_path *path,
1561 struct btrfs_key key;
1563 struct inode *inode;
1565 inode = read_one_inode(root, objectid);
1569 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1570 key.type = BTRFS_ORPHAN_ITEM_KEY;
1571 key.offset = objectid;
1573 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1575 btrfs_release_path(path);
1577 if (!inode->i_nlink)
1578 set_nlink(inode, 1);
1581 ret = btrfs_update_inode(trans, root, inode);
1582 } else if (ret == -EEXIST) {
1585 BUG(); /* Logic Error */
1593 * when replaying the log for a directory, we only insert names
1594 * for inodes that actually exist. This means an fsync on a directory
1595 * does not implicitly fsync all the new files in it
1597 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 u64 dirid, u64 index,
1600 char *name, int name_len,
1601 struct btrfs_key *location)
1603 struct inode *inode;
1607 inode = read_one_inode(root, location->objectid);
1611 dir = read_one_inode(root, dirid);
1617 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1619 /* FIXME, put inode into FIXUP list */
1627 * Return true if an inode reference exists in the log for the given name,
1628 * inode and parent inode.
1630 static bool name_in_log_ref(struct btrfs_root *log_root,
1631 const char *name, const int name_len,
1632 const u64 dirid, const u64 ino)
1634 struct btrfs_key search_key;
1636 search_key.objectid = ino;
1637 search_key.type = BTRFS_INODE_REF_KEY;
1638 search_key.offset = dirid;
1639 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1642 search_key.type = BTRFS_INODE_EXTREF_KEY;
1643 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1644 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1651 * take a single entry in a log directory item and replay it into
1654 * if a conflicting item exists in the subdirectory already,
1655 * the inode it points to is unlinked and put into the link count
1658 * If a name from the log points to a file or directory that does
1659 * not exist in the FS, it is skipped. fsyncs on directories
1660 * do not force down inodes inside that directory, just changes to the
1661 * names or unlinks in a directory.
1663 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1664 * non-existing inode) and 1 if the name was replayed.
1666 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1667 struct btrfs_root *root,
1668 struct btrfs_path *path,
1669 struct extent_buffer *eb,
1670 struct btrfs_dir_item *di,
1671 struct btrfs_key *key)
1675 struct btrfs_dir_item *dst_di;
1676 struct btrfs_key found_key;
1677 struct btrfs_key log_key;
1682 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1683 bool name_added = false;
1685 dir = read_one_inode(root, key->objectid);
1689 name_len = btrfs_dir_name_len(eb, di);
1690 name = kmalloc(name_len, GFP_NOFS);
1696 log_type = btrfs_dir_type(eb, di);
1697 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1700 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1701 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1706 btrfs_release_path(path);
1708 if (key->type == BTRFS_DIR_ITEM_KEY) {
1709 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1711 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1712 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1721 if (IS_ERR_OR_NULL(dst_di)) {
1722 /* we need a sequence number to insert, so we only
1723 * do inserts for the BTRFS_DIR_INDEX_KEY types
1725 if (key->type != BTRFS_DIR_INDEX_KEY)
1730 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1731 /* the existing item matches the logged item */
1732 if (found_key.objectid == log_key.objectid &&
1733 found_key.type == log_key.type &&
1734 found_key.offset == log_key.offset &&
1735 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1736 update_size = false;
1741 * don't drop the conflicting directory entry if the inode
1742 * for the new entry doesn't exist
1747 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1751 if (key->type == BTRFS_DIR_INDEX_KEY)
1754 btrfs_release_path(path);
1755 if (!ret && update_size) {
1756 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1757 ret = btrfs_update_inode(trans, root, dir);
1761 if (!ret && name_added)
1766 if (name_in_log_ref(root->log_root, name, name_len,
1767 key->objectid, log_key.objectid)) {
1768 /* The dentry will be added later. */
1770 update_size = false;
1773 btrfs_release_path(path);
1774 ret = insert_one_name(trans, root, key->objectid, key->offset,
1775 name, name_len, &log_key);
1776 if (ret && ret != -ENOENT && ret != -EEXIST)
1780 update_size = false;
1786 * find all the names in a directory item and reconcile them into
1787 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1788 * one name in a directory item, but the same code gets used for
1789 * both directory index types
1791 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct btrfs_path *path,
1794 struct extent_buffer *eb, int slot,
1795 struct btrfs_key *key)
1798 u32 item_size = btrfs_item_size_nr(eb, slot);
1799 struct btrfs_dir_item *di;
1802 unsigned long ptr_end;
1803 struct btrfs_path *fixup_path = NULL;
1805 ptr = btrfs_item_ptr_offset(eb, slot);
1806 ptr_end = ptr + item_size;
1807 while (ptr < ptr_end) {
1808 di = (struct btrfs_dir_item *)ptr;
1809 if (verify_dir_item(root, eb, di))
1811 name_len = btrfs_dir_name_len(eb, di);
1812 ret = replay_one_name(trans, root, path, eb, di, key);
1815 ptr = (unsigned long)(di + 1);
1819 * If this entry refers to a non-directory (directories can not
1820 * have a link count > 1) and it was added in the transaction
1821 * that was not committed, make sure we fixup the link count of
1822 * the inode it the entry points to. Otherwise something like
1823 * the following would result in a directory pointing to an
1824 * inode with a wrong link that does not account for this dir
1832 * ln testdir/bar testdir/bar_link
1833 * ln testdir/foo testdir/foo_link
1834 * xfs_io -c "fsync" testdir/bar
1838 * mount fs, log replay happens
1840 * File foo would remain with a link count of 1 when it has two
1841 * entries pointing to it in the directory testdir. This would
1842 * make it impossible to ever delete the parent directory has
1843 * it would result in stale dentries that can never be deleted.
1845 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1846 struct btrfs_key di_key;
1849 fixup_path = btrfs_alloc_path();
1856 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1857 ret = link_to_fixup_dir(trans, root, fixup_path,
1864 btrfs_free_path(fixup_path);
1869 * directory replay has two parts. There are the standard directory
1870 * items in the log copied from the subvolume, and range items
1871 * created in the log while the subvolume was logged.
1873 * The range items tell us which parts of the key space the log
1874 * is authoritative for. During replay, if a key in the subvolume
1875 * directory is in a logged range item, but not actually in the log
1876 * that means it was deleted from the directory before the fsync
1877 * and should be removed.
1879 static noinline int find_dir_range(struct btrfs_root *root,
1880 struct btrfs_path *path,
1881 u64 dirid, int key_type,
1882 u64 *start_ret, u64 *end_ret)
1884 struct btrfs_key key;
1886 struct btrfs_dir_log_item *item;
1890 if (*start_ret == (u64)-1)
1893 key.objectid = dirid;
1894 key.type = key_type;
1895 key.offset = *start_ret;
1897 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1901 if (path->slots[0] == 0)
1906 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1908 if (key.type != key_type || key.objectid != dirid) {
1912 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1913 struct btrfs_dir_log_item);
1914 found_end = btrfs_dir_log_end(path->nodes[0], item);
1916 if (*start_ret >= key.offset && *start_ret <= found_end) {
1918 *start_ret = key.offset;
1919 *end_ret = found_end;
1924 /* check the next slot in the tree to see if it is a valid item */
1925 nritems = btrfs_header_nritems(path->nodes[0]);
1927 if (path->slots[0] >= nritems) {
1928 ret = btrfs_next_leaf(root, path);
1933 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1935 if (key.type != key_type || key.objectid != dirid) {
1939 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1940 struct btrfs_dir_log_item);
1941 found_end = btrfs_dir_log_end(path->nodes[0], item);
1942 *start_ret = key.offset;
1943 *end_ret = found_end;
1946 btrfs_release_path(path);
1951 * this looks for a given directory item in the log. If the directory
1952 * item is not in the log, the item is removed and the inode it points
1955 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root,
1957 struct btrfs_root *log,
1958 struct btrfs_path *path,
1959 struct btrfs_path *log_path,
1961 struct btrfs_key *dir_key)
1964 struct extent_buffer *eb;
1967 struct btrfs_dir_item *di;
1968 struct btrfs_dir_item *log_di;
1971 unsigned long ptr_end;
1973 struct inode *inode;
1974 struct btrfs_key location;
1977 eb = path->nodes[0];
1978 slot = path->slots[0];
1979 item_size = btrfs_item_size_nr(eb, slot);
1980 ptr = btrfs_item_ptr_offset(eb, slot);
1981 ptr_end = ptr + item_size;
1982 while (ptr < ptr_end) {
1983 di = (struct btrfs_dir_item *)ptr;
1984 if (verify_dir_item(root, eb, di)) {
1989 name_len = btrfs_dir_name_len(eb, di);
1990 name = kmalloc(name_len, GFP_NOFS);
1995 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1998 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1999 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2002 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2003 log_di = btrfs_lookup_dir_index_item(trans, log,
2009 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2010 btrfs_dir_item_key_to_cpu(eb, di, &location);
2011 btrfs_release_path(path);
2012 btrfs_release_path(log_path);
2013 inode = read_one_inode(root, location.objectid);
2019 ret = link_to_fixup_dir(trans, root,
2020 path, location.objectid);
2028 ret = btrfs_unlink_inode(trans, root, dir, inode,
2031 ret = btrfs_run_delayed_items(trans, root);
2037 /* there might still be more names under this key
2038 * check and repeat if required
2040 ret = btrfs_search_slot(NULL, root, dir_key, path,
2046 } else if (IS_ERR(log_di)) {
2048 return PTR_ERR(log_di);
2050 btrfs_release_path(log_path);
2053 ptr = (unsigned long)(di + 1);
2058 btrfs_release_path(path);
2059 btrfs_release_path(log_path);
2063 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2064 struct btrfs_root *root,
2065 struct btrfs_root *log,
2066 struct btrfs_path *path,
2069 struct btrfs_key search_key;
2070 struct btrfs_path *log_path;
2075 log_path = btrfs_alloc_path();
2079 search_key.objectid = ino;
2080 search_key.type = BTRFS_XATTR_ITEM_KEY;
2081 search_key.offset = 0;
2083 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2087 nritems = btrfs_header_nritems(path->nodes[0]);
2088 for (i = path->slots[0]; i < nritems; i++) {
2089 struct btrfs_key key;
2090 struct btrfs_dir_item *di;
2091 struct btrfs_dir_item *log_di;
2095 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2096 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2101 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2102 total_size = btrfs_item_size_nr(path->nodes[0], i);
2104 while (cur < total_size) {
2105 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2106 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2107 u32 this_len = sizeof(*di) + name_len + data_len;
2110 name = kmalloc(name_len, GFP_NOFS);
2115 read_extent_buffer(path->nodes[0], name,
2116 (unsigned long)(di + 1), name_len);
2118 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2120 btrfs_release_path(log_path);
2122 /* Doesn't exist in log tree, so delete it. */
2123 btrfs_release_path(path);
2124 di = btrfs_lookup_xattr(trans, root, path, ino,
2125 name, name_len, -1);
2132 ret = btrfs_delete_one_dir_name(trans, root,
2136 btrfs_release_path(path);
2141 if (IS_ERR(log_di)) {
2142 ret = PTR_ERR(log_di);
2146 di = (struct btrfs_dir_item *)((char *)di + this_len);
2149 ret = btrfs_next_leaf(root, path);
2155 btrfs_free_path(log_path);
2156 btrfs_release_path(path);
2162 * deletion replay happens before we copy any new directory items
2163 * out of the log or out of backreferences from inodes. It
2164 * scans the log to find ranges of keys that log is authoritative for,
2165 * and then scans the directory to find items in those ranges that are
2166 * not present in the log.
2168 * Anything we don't find in the log is unlinked and removed from the
2171 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root,
2173 struct btrfs_root *log,
2174 struct btrfs_path *path,
2175 u64 dirid, int del_all)
2179 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2181 struct btrfs_key dir_key;
2182 struct btrfs_key found_key;
2183 struct btrfs_path *log_path;
2186 dir_key.objectid = dirid;
2187 dir_key.type = BTRFS_DIR_ITEM_KEY;
2188 log_path = btrfs_alloc_path();
2192 dir = read_one_inode(root, dirid);
2193 /* it isn't an error if the inode isn't there, that can happen
2194 * because we replay the deletes before we copy in the inode item
2198 btrfs_free_path(log_path);
2206 range_end = (u64)-1;
2208 ret = find_dir_range(log, path, dirid, key_type,
2209 &range_start, &range_end);
2214 dir_key.offset = range_start;
2217 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2222 nritems = btrfs_header_nritems(path->nodes[0]);
2223 if (path->slots[0] >= nritems) {
2224 ret = btrfs_next_leaf(root, path);
2228 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2230 if (found_key.objectid != dirid ||
2231 found_key.type != dir_key.type)
2234 if (found_key.offset > range_end)
2237 ret = check_item_in_log(trans, root, log, path,
2242 if (found_key.offset == (u64)-1)
2244 dir_key.offset = found_key.offset + 1;
2246 btrfs_release_path(path);
2247 if (range_end == (u64)-1)
2249 range_start = range_end + 1;
2254 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2255 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2256 dir_key.type = BTRFS_DIR_INDEX_KEY;
2257 btrfs_release_path(path);
2261 btrfs_release_path(path);
2262 btrfs_free_path(log_path);
2268 * the process_func used to replay items from the log tree. This
2269 * gets called in two different stages. The first stage just looks
2270 * for inodes and makes sure they are all copied into the subvolume.
2272 * The second stage copies all the other item types from the log into
2273 * the subvolume. The two stage approach is slower, but gets rid of
2274 * lots of complexity around inodes referencing other inodes that exist
2275 * only in the log (references come from either directory items or inode
2278 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2279 struct walk_control *wc, u64 gen)
2282 struct btrfs_path *path;
2283 struct btrfs_root *root = wc->replay_dest;
2284 struct btrfs_key key;
2289 ret = btrfs_read_buffer(eb, gen);
2293 level = btrfs_header_level(eb);
2298 path = btrfs_alloc_path();
2302 nritems = btrfs_header_nritems(eb);
2303 for (i = 0; i < nritems; i++) {
2304 btrfs_item_key_to_cpu(eb, &key, i);
2306 /* inode keys are done during the first stage */
2307 if (key.type == BTRFS_INODE_ITEM_KEY &&
2308 wc->stage == LOG_WALK_REPLAY_INODES) {
2309 struct btrfs_inode_item *inode_item;
2312 inode_item = btrfs_item_ptr(eb, i,
2313 struct btrfs_inode_item);
2314 ret = replay_xattr_deletes(wc->trans, root, log,
2315 path, key.objectid);
2318 mode = btrfs_inode_mode(eb, inode_item);
2319 if (S_ISDIR(mode)) {
2320 ret = replay_dir_deletes(wc->trans,
2321 root, log, path, key.objectid, 0);
2325 ret = overwrite_item(wc->trans, root, path,
2330 /* for regular files, make sure corresponding
2331 * orhpan item exist. extents past the new EOF
2332 * will be truncated later by orphan cleanup.
2334 if (S_ISREG(mode)) {
2335 ret = insert_orphan_item(wc->trans, root,
2341 ret = link_to_fixup_dir(wc->trans, root,
2342 path, key.objectid);
2347 if (key.type == BTRFS_DIR_INDEX_KEY &&
2348 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2349 ret = replay_one_dir_item(wc->trans, root, path,
2355 if (wc->stage < LOG_WALK_REPLAY_ALL)
2358 /* these keys are simply copied */
2359 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2360 ret = overwrite_item(wc->trans, root, path,
2364 } else if (key.type == BTRFS_INODE_REF_KEY ||
2365 key.type == BTRFS_INODE_EXTREF_KEY) {
2366 ret = add_inode_ref(wc->trans, root, log, path,
2368 if (ret && ret != -ENOENT)
2371 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2372 ret = replay_one_extent(wc->trans, root, path,
2376 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2377 ret = replay_one_dir_item(wc->trans, root, path,
2383 btrfs_free_path(path);
2387 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2388 struct btrfs_root *root,
2389 struct btrfs_path *path, int *level,
2390 struct walk_control *wc)
2395 struct extent_buffer *next;
2396 struct extent_buffer *cur;
2397 struct extent_buffer *parent;
2401 WARN_ON(*level < 0);
2402 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2404 while (*level > 0) {
2405 WARN_ON(*level < 0);
2406 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2407 cur = path->nodes[*level];
2409 WARN_ON(btrfs_header_level(cur) != *level);
2411 if (path->slots[*level] >=
2412 btrfs_header_nritems(cur))
2415 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2416 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2417 blocksize = root->nodesize;
2419 parent = path->nodes[*level];
2420 root_owner = btrfs_header_owner(parent);
2422 next = btrfs_find_create_tree_block(root, bytenr);
2427 ret = wc->process_func(root, next, wc, ptr_gen);
2429 free_extent_buffer(next);
2433 path->slots[*level]++;
2435 ret = btrfs_read_buffer(next, ptr_gen);
2437 free_extent_buffer(next);
2442 btrfs_tree_lock(next);
2443 btrfs_set_lock_blocking(next);
2444 clean_tree_block(trans, root->fs_info,
2446 btrfs_wait_tree_block_writeback(next);
2447 btrfs_tree_unlock(next);
2450 WARN_ON(root_owner !=
2451 BTRFS_TREE_LOG_OBJECTID);
2452 ret = btrfs_free_and_pin_reserved_extent(root,
2455 free_extent_buffer(next);
2459 free_extent_buffer(next);
2462 ret = btrfs_read_buffer(next, ptr_gen);
2464 free_extent_buffer(next);
2468 WARN_ON(*level <= 0);
2469 if (path->nodes[*level-1])
2470 free_extent_buffer(path->nodes[*level-1]);
2471 path->nodes[*level-1] = next;
2472 *level = btrfs_header_level(next);
2473 path->slots[*level] = 0;
2476 WARN_ON(*level < 0);
2477 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2479 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2485 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2486 struct btrfs_root *root,
2487 struct btrfs_path *path, int *level,
2488 struct walk_control *wc)
2495 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2496 slot = path->slots[i];
2497 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2500 WARN_ON(*level == 0);
2503 struct extent_buffer *parent;
2504 if (path->nodes[*level] == root->node)
2505 parent = path->nodes[*level];
2507 parent = path->nodes[*level + 1];
2509 root_owner = btrfs_header_owner(parent);
2510 ret = wc->process_func(root, path->nodes[*level], wc,
2511 btrfs_header_generation(path->nodes[*level]));
2516 struct extent_buffer *next;
2518 next = path->nodes[*level];
2521 btrfs_tree_lock(next);
2522 btrfs_set_lock_blocking(next);
2523 clean_tree_block(trans, root->fs_info,
2525 btrfs_wait_tree_block_writeback(next);
2526 btrfs_tree_unlock(next);
2529 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2530 ret = btrfs_free_and_pin_reserved_extent(root,
2531 path->nodes[*level]->start,
2532 path->nodes[*level]->len);
2536 free_extent_buffer(path->nodes[*level]);
2537 path->nodes[*level] = NULL;
2545 * drop the reference count on the tree rooted at 'snap'. This traverses
2546 * the tree freeing any blocks that have a ref count of zero after being
2549 static int walk_log_tree(struct btrfs_trans_handle *trans,
2550 struct btrfs_root *log, struct walk_control *wc)
2555 struct btrfs_path *path;
2558 path = btrfs_alloc_path();
2562 level = btrfs_header_level(log->node);
2564 path->nodes[level] = log->node;
2565 extent_buffer_get(log->node);
2566 path->slots[level] = 0;
2569 wret = walk_down_log_tree(trans, log, path, &level, wc);
2577 wret = walk_up_log_tree(trans, log, path, &level, wc);
2586 /* was the root node processed? if not, catch it here */
2587 if (path->nodes[orig_level]) {
2588 ret = wc->process_func(log, path->nodes[orig_level], wc,
2589 btrfs_header_generation(path->nodes[orig_level]));
2593 struct extent_buffer *next;
2595 next = path->nodes[orig_level];
2598 btrfs_tree_lock(next);
2599 btrfs_set_lock_blocking(next);
2600 clean_tree_block(trans, log->fs_info, next);
2601 btrfs_wait_tree_block_writeback(next);
2602 btrfs_tree_unlock(next);
2605 WARN_ON(log->root_key.objectid !=
2606 BTRFS_TREE_LOG_OBJECTID);
2607 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2615 btrfs_free_path(path);
2620 * helper function to update the item for a given subvolumes log root
2621 * in the tree of log roots
2623 static int update_log_root(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *log)
2628 if (log->log_transid == 1) {
2629 /* insert root item on the first sync */
2630 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2631 &log->root_key, &log->root_item);
2633 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2634 &log->root_key, &log->root_item);
2639 static void wait_log_commit(struct btrfs_root *root, int transid)
2642 int index = transid % 2;
2645 * we only allow two pending log transactions at a time,
2646 * so we know that if ours is more than 2 older than the
2647 * current transaction, we're done
2650 prepare_to_wait(&root->log_commit_wait[index],
2651 &wait, TASK_UNINTERRUPTIBLE);
2652 mutex_unlock(&root->log_mutex);
2654 if (root->log_transid_committed < transid &&
2655 atomic_read(&root->log_commit[index]))
2658 finish_wait(&root->log_commit_wait[index], &wait);
2659 mutex_lock(&root->log_mutex);
2660 } while (root->log_transid_committed < transid &&
2661 atomic_read(&root->log_commit[index]));
2664 static void wait_for_writer(struct btrfs_root *root)
2668 while (atomic_read(&root->log_writers)) {
2669 prepare_to_wait(&root->log_writer_wait,
2670 &wait, TASK_UNINTERRUPTIBLE);
2671 mutex_unlock(&root->log_mutex);
2672 if (atomic_read(&root->log_writers))
2674 finish_wait(&root->log_writer_wait, &wait);
2675 mutex_lock(&root->log_mutex);
2679 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2680 struct btrfs_log_ctx *ctx)
2685 mutex_lock(&root->log_mutex);
2686 list_del_init(&ctx->list);
2687 mutex_unlock(&root->log_mutex);
2691 * Invoked in log mutex context, or be sure there is no other task which
2692 * can access the list.
2694 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2695 int index, int error)
2697 struct btrfs_log_ctx *ctx;
2698 struct btrfs_log_ctx *safe;
2700 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2701 list_del_init(&ctx->list);
2702 ctx->log_ret = error;
2705 INIT_LIST_HEAD(&root->log_ctxs[index]);
2709 * btrfs_sync_log does sends a given tree log down to the disk and
2710 * updates the super blocks to record it. When this call is done,
2711 * you know that any inodes previously logged are safely on disk only
2714 * Any other return value means you need to call btrfs_commit_transaction.
2715 * Some of the edge cases for fsyncing directories that have had unlinks
2716 * or renames done in the past mean that sometimes the only safe
2717 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2718 * that has happened.
2720 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2721 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2727 struct btrfs_root *log = root->log_root;
2728 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2729 int log_transid = 0;
2730 struct btrfs_log_ctx root_log_ctx;
2731 struct blk_plug plug;
2733 mutex_lock(&root->log_mutex);
2734 log_transid = ctx->log_transid;
2735 if (root->log_transid_committed >= log_transid) {
2736 mutex_unlock(&root->log_mutex);
2737 return ctx->log_ret;
2740 index1 = log_transid % 2;
2741 if (atomic_read(&root->log_commit[index1])) {
2742 wait_log_commit(root, log_transid);
2743 mutex_unlock(&root->log_mutex);
2744 return ctx->log_ret;
2746 ASSERT(log_transid == root->log_transid);
2747 atomic_set(&root->log_commit[index1], 1);
2749 /* wait for previous tree log sync to complete */
2750 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2751 wait_log_commit(root, log_transid - 1);
2754 int batch = atomic_read(&root->log_batch);
2755 /* when we're on an ssd, just kick the log commit out */
2756 if (!btrfs_test_opt(root, SSD) &&
2757 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2758 mutex_unlock(&root->log_mutex);
2759 schedule_timeout_uninterruptible(1);
2760 mutex_lock(&root->log_mutex);
2762 wait_for_writer(root);
2763 if (batch == atomic_read(&root->log_batch))
2767 /* bail out if we need to do a full commit */
2768 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2770 btrfs_free_logged_extents(log, log_transid);
2771 mutex_unlock(&root->log_mutex);
2775 if (log_transid % 2 == 0)
2776 mark = EXTENT_DIRTY;
2780 /* we start IO on all the marked extents here, but we don't actually
2781 * wait for them until later.
2783 blk_start_plug(&plug);
2784 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2786 blk_finish_plug(&plug);
2787 btrfs_abort_transaction(trans, root, ret);
2788 btrfs_free_logged_extents(log, log_transid);
2789 btrfs_set_log_full_commit(root->fs_info, trans);
2790 mutex_unlock(&root->log_mutex);
2794 btrfs_set_root_node(&log->root_item, log->node);
2796 root->log_transid++;
2797 log->log_transid = root->log_transid;
2798 root->log_start_pid = 0;
2800 * IO has been started, blocks of the log tree have WRITTEN flag set
2801 * in their headers. new modifications of the log will be written to
2802 * new positions. so it's safe to allow log writers to go in.
2804 mutex_unlock(&root->log_mutex);
2806 btrfs_init_log_ctx(&root_log_ctx);
2808 mutex_lock(&log_root_tree->log_mutex);
2809 atomic_inc(&log_root_tree->log_batch);
2810 atomic_inc(&log_root_tree->log_writers);
2812 index2 = log_root_tree->log_transid % 2;
2813 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2814 root_log_ctx.log_transid = log_root_tree->log_transid;
2816 mutex_unlock(&log_root_tree->log_mutex);
2818 ret = update_log_root(trans, log);
2820 mutex_lock(&log_root_tree->log_mutex);
2821 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2823 * Implicit memory barrier after atomic_dec_and_test
2825 if (waitqueue_active(&log_root_tree->log_writer_wait))
2826 wake_up(&log_root_tree->log_writer_wait);
2830 if (!list_empty(&root_log_ctx.list))
2831 list_del_init(&root_log_ctx.list);
2833 blk_finish_plug(&plug);
2834 btrfs_set_log_full_commit(root->fs_info, trans);
2836 if (ret != -ENOSPC) {
2837 btrfs_abort_transaction(trans, root, ret);
2838 mutex_unlock(&log_root_tree->log_mutex);
2841 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2842 btrfs_free_logged_extents(log, log_transid);
2843 mutex_unlock(&log_root_tree->log_mutex);
2848 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2849 blk_finish_plug(&plug);
2850 list_del_init(&root_log_ctx.list);
2851 mutex_unlock(&log_root_tree->log_mutex);
2852 ret = root_log_ctx.log_ret;
2856 index2 = root_log_ctx.log_transid % 2;
2857 if (atomic_read(&log_root_tree->log_commit[index2])) {
2858 blk_finish_plug(&plug);
2859 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2861 btrfs_wait_logged_extents(trans, log, log_transid);
2862 wait_log_commit(log_root_tree,
2863 root_log_ctx.log_transid);
2864 mutex_unlock(&log_root_tree->log_mutex);
2866 ret = root_log_ctx.log_ret;
2869 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2870 atomic_set(&log_root_tree->log_commit[index2], 1);
2872 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2873 wait_log_commit(log_root_tree,
2874 root_log_ctx.log_transid - 1);
2877 wait_for_writer(log_root_tree);
2880 * now that we've moved on to the tree of log tree roots,
2881 * check the full commit flag again
2883 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2884 blk_finish_plug(&plug);
2885 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2886 btrfs_free_logged_extents(log, log_transid);
2887 mutex_unlock(&log_root_tree->log_mutex);
2889 goto out_wake_log_root;
2892 ret = btrfs_write_marked_extents(log_root_tree,
2893 &log_root_tree->dirty_log_pages,
2894 EXTENT_DIRTY | EXTENT_NEW);
2895 blk_finish_plug(&plug);
2897 btrfs_set_log_full_commit(root->fs_info, trans);
2898 btrfs_abort_transaction(trans, root, ret);
2899 btrfs_free_logged_extents(log, log_transid);
2900 mutex_unlock(&log_root_tree->log_mutex);
2901 goto out_wake_log_root;
2903 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2905 ret = btrfs_wait_marked_extents(log_root_tree,
2906 &log_root_tree->dirty_log_pages,
2907 EXTENT_NEW | EXTENT_DIRTY);
2909 btrfs_set_log_full_commit(root->fs_info, trans);
2910 btrfs_free_logged_extents(log, log_transid);
2911 mutex_unlock(&log_root_tree->log_mutex);
2912 goto out_wake_log_root;
2914 btrfs_wait_logged_extents(trans, log, log_transid);
2916 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2917 log_root_tree->node->start);
2918 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2919 btrfs_header_level(log_root_tree->node));
2921 log_root_tree->log_transid++;
2922 mutex_unlock(&log_root_tree->log_mutex);
2925 * nobody else is going to jump in and write the the ctree
2926 * super here because the log_commit atomic below is protecting
2927 * us. We must be called with a transaction handle pinning
2928 * the running transaction open, so a full commit can't hop
2929 * in and cause problems either.
2931 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2933 btrfs_set_log_full_commit(root->fs_info, trans);
2934 btrfs_abort_transaction(trans, root, ret);
2935 goto out_wake_log_root;
2938 mutex_lock(&root->log_mutex);
2939 if (root->last_log_commit < log_transid)
2940 root->last_log_commit = log_transid;
2941 mutex_unlock(&root->log_mutex);
2944 mutex_lock(&log_root_tree->log_mutex);
2945 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2947 log_root_tree->log_transid_committed++;
2948 atomic_set(&log_root_tree->log_commit[index2], 0);
2949 mutex_unlock(&log_root_tree->log_mutex);
2952 * The barrier before waitqueue_active is implied by mutex_unlock
2954 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2955 wake_up(&log_root_tree->log_commit_wait[index2]);
2957 mutex_lock(&root->log_mutex);
2958 btrfs_remove_all_log_ctxs(root, index1, ret);
2959 root->log_transid_committed++;
2960 atomic_set(&root->log_commit[index1], 0);
2961 mutex_unlock(&root->log_mutex);
2964 * The barrier before waitqueue_active is implied by mutex_unlock
2966 if (waitqueue_active(&root->log_commit_wait[index1]))
2967 wake_up(&root->log_commit_wait[index1]);
2971 static void free_log_tree(struct btrfs_trans_handle *trans,
2972 struct btrfs_root *log)
2977 struct walk_control wc = {
2979 .process_func = process_one_buffer
2982 ret = walk_log_tree(trans, log, &wc);
2983 /* I don't think this can happen but just in case */
2985 btrfs_abort_transaction(trans, log, ret);
2988 ret = find_first_extent_bit(&log->dirty_log_pages,
2989 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2994 clear_extent_bits(&log->dirty_log_pages, start, end,
2995 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2999 * We may have short-circuited the log tree with the full commit logic
3000 * and left ordered extents on our list, so clear these out to keep us
3001 * from leaking inodes and memory.
3003 btrfs_free_logged_extents(log, 0);
3004 btrfs_free_logged_extents(log, 1);
3006 free_extent_buffer(log->node);
3011 * free all the extents used by the tree log. This should be called
3012 * at commit time of the full transaction
3014 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3016 if (root->log_root) {
3017 free_log_tree(trans, root->log_root);
3018 root->log_root = NULL;
3023 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3024 struct btrfs_fs_info *fs_info)
3026 if (fs_info->log_root_tree) {
3027 free_log_tree(trans, fs_info->log_root_tree);
3028 fs_info->log_root_tree = NULL;
3034 * If both a file and directory are logged, and unlinks or renames are
3035 * mixed in, we have a few interesting corners:
3037 * create file X in dir Y
3038 * link file X to X.link in dir Y
3040 * unlink file X but leave X.link
3043 * After a crash we would expect only X.link to exist. But file X
3044 * didn't get fsync'd again so the log has back refs for X and X.link.
3046 * We solve this by removing directory entries and inode backrefs from the
3047 * log when a file that was logged in the current transaction is
3048 * unlinked. Any later fsync will include the updated log entries, and
3049 * we'll be able to reconstruct the proper directory items from backrefs.
3051 * This optimizations allows us to avoid relogging the entire inode
3052 * or the entire directory.
3054 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3055 struct btrfs_root *root,
3056 const char *name, int name_len,
3057 struct inode *dir, u64 index)
3059 struct btrfs_root *log;
3060 struct btrfs_dir_item *di;
3061 struct btrfs_path *path;
3065 u64 dir_ino = btrfs_ino(dir);
3067 if (BTRFS_I(dir)->logged_trans < trans->transid)
3070 ret = join_running_log_trans(root);
3074 mutex_lock(&BTRFS_I(dir)->log_mutex);
3076 log = root->log_root;
3077 path = btrfs_alloc_path();
3083 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3084 name, name_len, -1);
3090 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3091 bytes_del += name_len;
3097 btrfs_release_path(path);
3098 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3099 index, name, name_len, -1);
3105 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3106 bytes_del += name_len;
3113 /* update the directory size in the log to reflect the names
3117 struct btrfs_key key;
3119 key.objectid = dir_ino;
3121 key.type = BTRFS_INODE_ITEM_KEY;
3122 btrfs_release_path(path);
3124 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3130 struct btrfs_inode_item *item;
3133 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3134 struct btrfs_inode_item);
3135 i_size = btrfs_inode_size(path->nodes[0], item);
3136 if (i_size > bytes_del)
3137 i_size -= bytes_del;
3140 btrfs_set_inode_size(path->nodes[0], item, i_size);
3141 btrfs_mark_buffer_dirty(path->nodes[0]);
3144 btrfs_release_path(path);
3147 btrfs_free_path(path);
3149 mutex_unlock(&BTRFS_I(dir)->log_mutex);
3150 if (ret == -ENOSPC) {
3151 btrfs_set_log_full_commit(root->fs_info, trans);
3154 btrfs_abort_transaction(trans, root, ret);
3156 btrfs_end_log_trans(root);
3161 /* see comments for btrfs_del_dir_entries_in_log */
3162 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root,
3164 const char *name, int name_len,
3165 struct inode *inode, u64 dirid)
3167 struct btrfs_root *log;
3171 if (BTRFS_I(inode)->logged_trans < trans->transid)
3174 ret = join_running_log_trans(root);
3177 log = root->log_root;
3178 mutex_lock(&BTRFS_I(inode)->log_mutex);
3180 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3182 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3183 if (ret == -ENOSPC) {
3184 btrfs_set_log_full_commit(root->fs_info, trans);
3186 } else if (ret < 0 && ret != -ENOENT)
3187 btrfs_abort_transaction(trans, root, ret);
3188 btrfs_end_log_trans(root);
3194 * creates a range item in the log for 'dirid'. first_offset and
3195 * last_offset tell us which parts of the key space the log should
3196 * be considered authoritative for.
3198 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3199 struct btrfs_root *log,
3200 struct btrfs_path *path,
3201 int key_type, u64 dirid,
3202 u64 first_offset, u64 last_offset)
3205 struct btrfs_key key;
3206 struct btrfs_dir_log_item *item;
3208 key.objectid = dirid;
3209 key.offset = first_offset;
3210 if (key_type == BTRFS_DIR_ITEM_KEY)
3211 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3213 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3214 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3218 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3219 struct btrfs_dir_log_item);
3220 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3221 btrfs_mark_buffer_dirty(path->nodes[0]);
3222 btrfs_release_path(path);
3227 * log all the items included in the current transaction for a given
3228 * directory. This also creates the range items in the log tree required
3229 * to replay anything deleted before the fsync
3231 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3232 struct btrfs_root *root, struct inode *inode,
3233 struct btrfs_path *path,
3234 struct btrfs_path *dst_path, int key_type,
3235 struct btrfs_log_ctx *ctx,
3236 u64 min_offset, u64 *last_offset_ret)
3238 struct btrfs_key min_key;
3239 struct btrfs_root *log = root->log_root;
3240 struct extent_buffer *src;
3245 u64 first_offset = min_offset;
3246 u64 last_offset = (u64)-1;
3247 u64 ino = btrfs_ino(inode);
3249 log = root->log_root;
3251 min_key.objectid = ino;
3252 min_key.type = key_type;
3253 min_key.offset = min_offset;
3255 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3258 * we didn't find anything from this transaction, see if there
3259 * is anything at all
3261 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3262 min_key.objectid = ino;
3263 min_key.type = key_type;
3264 min_key.offset = (u64)-1;
3265 btrfs_release_path(path);
3266 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3268 btrfs_release_path(path);
3271 ret = btrfs_previous_item(root, path, ino, key_type);
3273 /* if ret == 0 there are items for this type,
3274 * create a range to tell us the last key of this type.
3275 * otherwise, there are no items in this directory after
3276 * *min_offset, and we create a range to indicate that.
3279 struct btrfs_key tmp;
3280 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3282 if (key_type == tmp.type)
3283 first_offset = max(min_offset, tmp.offset) + 1;
3288 /* go backward to find any previous key */
3289 ret = btrfs_previous_item(root, path, ino, key_type);
3291 struct btrfs_key tmp;
3292 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3293 if (key_type == tmp.type) {
3294 first_offset = tmp.offset;
3295 ret = overwrite_item(trans, log, dst_path,
3296 path->nodes[0], path->slots[0],
3304 btrfs_release_path(path);
3306 /* find the first key from this transaction again */
3307 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3308 if (WARN_ON(ret != 0))
3312 * we have a block from this transaction, log every item in it
3313 * from our directory
3316 struct btrfs_key tmp;
3317 src = path->nodes[0];
3318 nritems = btrfs_header_nritems(src);
3319 for (i = path->slots[0]; i < nritems; i++) {
3320 struct btrfs_dir_item *di;
3322 btrfs_item_key_to_cpu(src, &min_key, i);
3324 if (min_key.objectid != ino || min_key.type != key_type)
3326 ret = overwrite_item(trans, log, dst_path, src, i,
3334 * We must make sure that when we log a directory entry,
3335 * the corresponding inode, after log replay, has a
3336 * matching link count. For example:
3342 * xfs_io -c "fsync" mydir
3344 * <mount fs and log replay>
3346 * Would result in a fsync log that when replayed, our
3347 * file inode would have a link count of 1, but we get
3348 * two directory entries pointing to the same inode.
3349 * After removing one of the names, it would not be
3350 * possible to remove the other name, which resulted
3351 * always in stale file handle errors, and would not
3352 * be possible to rmdir the parent directory, since
3353 * its i_size could never decrement to the value
3354 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3356 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3357 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3359 (btrfs_dir_transid(src, di) == trans->transid ||
3360 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3361 tmp.type != BTRFS_ROOT_ITEM_KEY)
3362 ctx->log_new_dentries = true;
3364 path->slots[0] = nritems;
3367 * look ahead to the next item and see if it is also
3368 * from this directory and from this transaction
3370 ret = btrfs_next_leaf(root, path);
3372 last_offset = (u64)-1;
3375 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3376 if (tmp.objectid != ino || tmp.type != key_type) {
3377 last_offset = (u64)-1;
3380 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3381 ret = overwrite_item(trans, log, dst_path,
3382 path->nodes[0], path->slots[0],
3387 last_offset = tmp.offset;
3392 btrfs_release_path(path);
3393 btrfs_release_path(dst_path);
3396 *last_offset_ret = last_offset;
3398 * insert the log range keys to indicate where the log
3401 ret = insert_dir_log_key(trans, log, path, key_type,
3402 ino, first_offset, last_offset);
3410 * logging directories is very similar to logging inodes, We find all the items
3411 * from the current transaction and write them to the log.
3413 * The recovery code scans the directory in the subvolume, and if it finds a
3414 * key in the range logged that is not present in the log tree, then it means
3415 * that dir entry was unlinked during the transaction.
3417 * In order for that scan to work, we must include one key smaller than
3418 * the smallest logged by this transaction and one key larger than the largest
3419 * key logged by this transaction.
3421 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3422 struct btrfs_root *root, struct inode *inode,
3423 struct btrfs_path *path,
3424 struct btrfs_path *dst_path,
3425 struct btrfs_log_ctx *ctx)
3430 int key_type = BTRFS_DIR_ITEM_KEY;
3436 ret = log_dir_items(trans, root, inode, path,
3437 dst_path, key_type, ctx, min_key,
3441 if (max_key == (u64)-1)
3443 min_key = max_key + 1;
3446 if (key_type == BTRFS_DIR_ITEM_KEY) {
3447 key_type = BTRFS_DIR_INDEX_KEY;
3454 * a helper function to drop items from the log before we relog an
3455 * inode. max_key_type indicates the highest item type to remove.
3456 * This cannot be run for file data extents because it does not
3457 * free the extents they point to.
3459 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3460 struct btrfs_root *log,
3461 struct btrfs_path *path,
3462 u64 objectid, int max_key_type)
3465 struct btrfs_key key;
3466 struct btrfs_key found_key;
3469 key.objectid = objectid;
3470 key.type = max_key_type;
3471 key.offset = (u64)-1;
3474 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3475 BUG_ON(ret == 0); /* Logic error */
3479 if (path->slots[0] == 0)
3483 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3486 if (found_key.objectid != objectid)
3489 found_key.offset = 0;
3491 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3494 ret = btrfs_del_items(trans, log, path, start_slot,
3495 path->slots[0] - start_slot + 1);
3497 * If start slot isn't 0 then we don't need to re-search, we've
3498 * found the last guy with the objectid in this tree.
3500 if (ret || start_slot != 0)
3502 btrfs_release_path(path);
3504 btrfs_release_path(path);
3510 static void fill_inode_item(struct btrfs_trans_handle *trans,
3511 struct extent_buffer *leaf,
3512 struct btrfs_inode_item *item,
3513 struct inode *inode, int log_inode_only,
3516 struct btrfs_map_token token;
3518 btrfs_init_map_token(&token);
3520 if (log_inode_only) {
3521 /* set the generation to zero so the recover code
3522 * can tell the difference between an logging
3523 * just to say 'this inode exists' and a logging
3524 * to say 'update this inode with these values'
3526 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3527 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3529 btrfs_set_token_inode_generation(leaf, item,
3530 BTRFS_I(inode)->generation,
3532 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3535 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3536 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3537 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3538 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3540 btrfs_set_token_timespec_sec(leaf, &item->atime,
3541 inode->i_atime.tv_sec, &token);
3542 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3543 inode->i_atime.tv_nsec, &token);
3545 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3546 inode->i_mtime.tv_sec, &token);
3547 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3548 inode->i_mtime.tv_nsec, &token);
3550 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3551 inode->i_ctime.tv_sec, &token);
3552 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3553 inode->i_ctime.tv_nsec, &token);
3555 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3558 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3559 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3560 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3561 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3562 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3565 static int log_inode_item(struct btrfs_trans_handle *trans,
3566 struct btrfs_root *log, struct btrfs_path *path,
3567 struct inode *inode)
3569 struct btrfs_inode_item *inode_item;
3572 ret = btrfs_insert_empty_item(trans, log, path,
3573 &BTRFS_I(inode)->location,
3574 sizeof(*inode_item));
3575 if (ret && ret != -EEXIST)
3577 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3578 struct btrfs_inode_item);
3579 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3580 btrfs_release_path(path);
3584 static noinline int copy_items(struct btrfs_trans_handle *trans,
3585 struct inode *inode,
3586 struct btrfs_path *dst_path,
3587 struct btrfs_path *src_path, u64 *last_extent,
3588 int start_slot, int nr, int inode_only,
3591 unsigned long src_offset;
3592 unsigned long dst_offset;
3593 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3594 struct btrfs_file_extent_item *extent;
3595 struct btrfs_inode_item *inode_item;
3596 struct extent_buffer *src = src_path->nodes[0];
3597 struct btrfs_key first_key, last_key, key;
3599 struct btrfs_key *ins_keys;
3603 struct list_head ordered_sums;
3604 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3605 bool has_extents = false;
3606 bool need_find_last_extent = true;
3609 INIT_LIST_HEAD(&ordered_sums);
3611 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3612 nr * sizeof(u32), GFP_NOFS);
3616 first_key.objectid = (u64)-1;
3618 ins_sizes = (u32 *)ins_data;
3619 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3621 for (i = 0; i < nr; i++) {
3622 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3623 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3625 ret = btrfs_insert_empty_items(trans, log, dst_path,
3626 ins_keys, ins_sizes, nr);
3632 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3633 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3634 dst_path->slots[0]);
3636 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3638 if ((i == (nr - 1)))
3639 last_key = ins_keys[i];
3641 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3642 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3644 struct btrfs_inode_item);
3645 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3646 inode, inode_only == LOG_INODE_EXISTS,
3649 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3650 src_offset, ins_sizes[i]);
3654 * We set need_find_last_extent here in case we know we were
3655 * processing other items and then walk into the first extent in
3656 * the inode. If we don't hit an extent then nothing changes,
3657 * we'll do the last search the next time around.
3659 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3661 if (first_key.objectid == (u64)-1)
3662 first_key = ins_keys[i];
3664 need_find_last_extent = false;
3667 /* take a reference on file data extents so that truncates
3668 * or deletes of this inode don't have to relog the inode
3671 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3674 extent = btrfs_item_ptr(src, start_slot + i,
3675 struct btrfs_file_extent_item);
3677 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3680 found_type = btrfs_file_extent_type(src, extent);
3681 if (found_type == BTRFS_FILE_EXTENT_REG) {
3683 ds = btrfs_file_extent_disk_bytenr(src,
3685 /* ds == 0 is a hole */
3689 dl = btrfs_file_extent_disk_num_bytes(src,
3691 cs = btrfs_file_extent_offset(src, extent);
3692 cl = btrfs_file_extent_num_bytes(src,
3694 if (btrfs_file_extent_compression(src,
3700 ret = btrfs_lookup_csums_range(
3701 log->fs_info->csum_root,
3702 ds + cs, ds + cs + cl - 1,
3705 btrfs_release_path(dst_path);
3713 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3714 btrfs_release_path(dst_path);
3718 * we have to do this after the loop above to avoid changing the
3719 * log tree while trying to change the log tree.
3722 while (!list_empty(&ordered_sums)) {
3723 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3724 struct btrfs_ordered_sum,
3727 ret = btrfs_csum_file_blocks(trans, log, sums);
3728 list_del(&sums->list);
3735 if (need_find_last_extent && *last_extent == first_key.offset) {
3737 * We don't have any leafs between our current one and the one
3738 * we processed before that can have file extent items for our
3739 * inode (and have a generation number smaller than our current
3742 need_find_last_extent = false;
3746 * Because we use btrfs_search_forward we could skip leaves that were
3747 * not modified and then assume *last_extent is valid when it really
3748 * isn't. So back up to the previous leaf and read the end of the last
3749 * extent before we go and fill in holes.
3751 if (need_find_last_extent) {
3754 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3759 if (src_path->slots[0])
3760 src_path->slots[0]--;
3761 src = src_path->nodes[0];
3762 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3763 if (key.objectid != btrfs_ino(inode) ||
3764 key.type != BTRFS_EXTENT_DATA_KEY)
3766 extent = btrfs_item_ptr(src, src_path->slots[0],
3767 struct btrfs_file_extent_item);
3768 if (btrfs_file_extent_type(src, extent) ==
3769 BTRFS_FILE_EXTENT_INLINE) {
3770 len = btrfs_file_extent_inline_len(src,
3773 *last_extent = ALIGN(key.offset + len,
3776 len = btrfs_file_extent_num_bytes(src, extent);
3777 *last_extent = key.offset + len;
3781 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3782 * things could have happened
3784 * 1) A merge could have happened, so we could currently be on a leaf
3785 * that holds what we were copying in the first place.
3786 * 2) A split could have happened, and now not all of the items we want
3787 * are on the same leaf.
3789 * So we need to adjust how we search for holes, we need to drop the
3790 * path and re-search for the first extent key we found, and then walk
3791 * forward until we hit the last one we copied.
3793 if (need_find_last_extent) {
3794 /* btrfs_prev_leaf could return 1 without releasing the path */
3795 btrfs_release_path(src_path);
3796 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3801 src = src_path->nodes[0];
3802 i = src_path->slots[0];
3808 * Ok so here we need to go through and fill in any holes we may have
3809 * to make sure that holes are punched for those areas in case they had
3810 * extents previously.
3816 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3817 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3821 src = src_path->nodes[0];
3825 btrfs_item_key_to_cpu(src, &key, i);
3826 if (!btrfs_comp_cpu_keys(&key, &last_key))
3828 if (key.objectid != btrfs_ino(inode) ||
3829 key.type != BTRFS_EXTENT_DATA_KEY) {
3833 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3834 if (btrfs_file_extent_type(src, extent) ==
3835 BTRFS_FILE_EXTENT_INLINE) {
3836 len = btrfs_file_extent_inline_len(src, i, extent);
3837 extent_end = ALIGN(key.offset + len, log->sectorsize);
3839 len = btrfs_file_extent_num_bytes(src, extent);
3840 extent_end = key.offset + len;
3844 if (*last_extent == key.offset) {
3845 *last_extent = extent_end;
3848 offset = *last_extent;
3849 len = key.offset - *last_extent;
3850 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3851 offset, 0, 0, len, 0, len, 0,
3855 *last_extent = extent_end;
3858 * Need to let the callers know we dropped the path so they should
3861 if (!ret && need_find_last_extent)
3866 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3868 struct extent_map *em1, *em2;
3870 em1 = list_entry(a, struct extent_map, list);
3871 em2 = list_entry(b, struct extent_map, list);
3873 if (em1->start < em2->start)
3875 else if (em1->start > em2->start)
3880 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3881 struct inode *inode,
3882 struct btrfs_root *root,
3883 const struct extent_map *em,
3884 const struct list_head *logged_list,
3885 bool *ordered_io_error)
3887 struct btrfs_ordered_extent *ordered;
3888 struct btrfs_root *log = root->log_root;
3889 u64 mod_start = em->mod_start;
3890 u64 mod_len = em->mod_len;
3891 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3894 LIST_HEAD(ordered_sums);
3897 *ordered_io_error = false;
3899 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3900 em->block_start == EXTENT_MAP_HOLE)
3904 * Wait far any ordered extent that covers our extent map. If it
3905 * finishes without an error, first check and see if our csums are on
3906 * our outstanding ordered extents.
3908 list_for_each_entry(ordered, logged_list, log_list) {
3909 struct btrfs_ordered_sum *sum;
3914 if (ordered->file_offset + ordered->len <= mod_start ||
3915 mod_start + mod_len <= ordered->file_offset)
3918 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3919 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3920 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3921 const u64 start = ordered->file_offset;
3922 const u64 end = ordered->file_offset + ordered->len - 1;
3924 WARN_ON(ordered->inode != inode);
3925 filemap_fdatawrite_range(inode->i_mapping, start, end);
3928 wait_event(ordered->wait,
3929 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3930 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3932 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3934 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3935 * i_mapping flags, so that the next fsync won't get
3936 * an outdated io error too.
3938 btrfs_inode_check_errors(inode);
3939 *ordered_io_error = true;
3943 * We are going to copy all the csums on this ordered extent, so
3944 * go ahead and adjust mod_start and mod_len in case this
3945 * ordered extent has already been logged.
3947 if (ordered->file_offset > mod_start) {
3948 if (ordered->file_offset + ordered->len >=
3949 mod_start + mod_len)
3950 mod_len = ordered->file_offset - mod_start;
3952 * If we have this case
3954 * |--------- logged extent ---------|
3955 * |----- ordered extent ----|
3957 * Just don't mess with mod_start and mod_len, we'll
3958 * just end up logging more csums than we need and it
3962 if (ordered->file_offset + ordered->len <
3963 mod_start + mod_len) {
3964 mod_len = (mod_start + mod_len) -
3965 (ordered->file_offset + ordered->len);
3966 mod_start = ordered->file_offset +
3977 * To keep us from looping for the above case of an ordered
3978 * extent that falls inside of the logged extent.
3980 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3984 list_for_each_entry(sum, &ordered->list, list) {
3985 ret = btrfs_csum_file_blocks(trans, log, sum);
3991 if (*ordered_io_error || !mod_len || ret || skip_csum)
3994 if (em->compress_type) {
3996 csum_len = max(em->block_len, em->orig_block_len);
3998 csum_offset = mod_start - em->start;
4002 /* block start is already adjusted for the file extent offset. */
4003 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4004 em->block_start + csum_offset,
4005 em->block_start + csum_offset +
4006 csum_len - 1, &ordered_sums, 0);
4010 while (!list_empty(&ordered_sums)) {
4011 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4012 struct btrfs_ordered_sum,
4015 ret = btrfs_csum_file_blocks(trans, log, sums);
4016 list_del(&sums->list);
4023 static int log_one_extent(struct btrfs_trans_handle *trans,
4024 struct inode *inode, struct btrfs_root *root,
4025 const struct extent_map *em,
4026 struct btrfs_path *path,
4027 const struct list_head *logged_list,
4028 struct btrfs_log_ctx *ctx)
4030 struct btrfs_root *log = root->log_root;
4031 struct btrfs_file_extent_item *fi;
4032 struct extent_buffer *leaf;
4033 struct btrfs_map_token token;
4034 struct btrfs_key key;
4035 u64 extent_offset = em->start - em->orig_start;
4038 int extent_inserted = 0;
4039 bool ordered_io_err = false;
4041 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4046 if (ordered_io_err) {
4051 btrfs_init_map_token(&token);
4053 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4054 em->start + em->len, NULL, 0, 1,
4055 sizeof(*fi), &extent_inserted);
4059 if (!extent_inserted) {
4060 key.objectid = btrfs_ino(inode);
4061 key.type = BTRFS_EXTENT_DATA_KEY;
4062 key.offset = em->start;
4064 ret = btrfs_insert_empty_item(trans, log, path, &key,
4069 leaf = path->nodes[0];
4070 fi = btrfs_item_ptr(leaf, path->slots[0],
4071 struct btrfs_file_extent_item);
4073 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4075 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4076 btrfs_set_token_file_extent_type(leaf, fi,
4077 BTRFS_FILE_EXTENT_PREALLOC,
4080 btrfs_set_token_file_extent_type(leaf, fi,
4081 BTRFS_FILE_EXTENT_REG,
4084 block_len = max(em->block_len, em->orig_block_len);
4085 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4086 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4089 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4091 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4092 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4094 extent_offset, &token);
4095 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4098 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4099 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4103 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4104 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4105 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4106 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4108 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4109 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4110 btrfs_mark_buffer_dirty(leaf);
4112 btrfs_release_path(path);
4117 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4118 struct btrfs_root *root,
4119 struct inode *inode,
4120 struct btrfs_path *path,
4121 struct list_head *logged_list,
4122 struct btrfs_log_ctx *ctx)
4124 struct extent_map *em, *n;
4125 struct list_head extents;
4126 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4131 INIT_LIST_HEAD(&extents);
4133 write_lock(&tree->lock);
4134 test_gen = root->fs_info->last_trans_committed;
4136 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4137 list_del_init(&em->list);
4140 * Just an arbitrary number, this can be really CPU intensive
4141 * once we start getting a lot of extents, and really once we
4142 * have a bunch of extents we just want to commit since it will
4145 if (++num > 32768) {
4146 list_del_init(&tree->modified_extents);
4151 if (em->generation <= test_gen)
4153 /* Need a ref to keep it from getting evicted from cache */
4154 atomic_inc(&em->refs);
4155 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4156 list_add_tail(&em->list, &extents);
4160 list_sort(NULL, &extents, extent_cmp);
4163 while (!list_empty(&extents)) {
4164 em = list_entry(extents.next, struct extent_map, list);
4166 list_del_init(&em->list);
4169 * If we had an error we just need to delete everybody from our
4173 clear_em_logging(tree, em);
4174 free_extent_map(em);
4178 write_unlock(&tree->lock);
4180 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4182 write_lock(&tree->lock);
4183 clear_em_logging(tree, em);
4184 free_extent_map(em);
4186 WARN_ON(!list_empty(&extents));
4187 write_unlock(&tree->lock);
4189 btrfs_release_path(path);
4193 static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4194 struct btrfs_path *path, u64 *size_ret)
4196 struct btrfs_key key;
4199 key.objectid = btrfs_ino(inode);
4200 key.type = BTRFS_INODE_ITEM_KEY;
4203 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4206 } else if (ret > 0) {
4209 struct btrfs_inode_item *item;
4211 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4212 struct btrfs_inode_item);
4213 *size_ret = btrfs_inode_size(path->nodes[0], item);
4216 btrfs_release_path(path);
4221 * At the moment we always log all xattrs. This is to figure out at log replay
4222 * time which xattrs must have their deletion replayed. If a xattr is missing
4223 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4224 * because if a xattr is deleted, the inode is fsynced and a power failure
4225 * happens, causing the log to be replayed the next time the fs is mounted,
4226 * we want the xattr to not exist anymore (same behaviour as other filesystems
4227 * with a journal, ext3/4, xfs, f2fs, etc).
4229 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4230 struct btrfs_root *root,
4231 struct inode *inode,
4232 struct btrfs_path *path,
4233 struct btrfs_path *dst_path)
4236 struct btrfs_key key;
4237 const u64 ino = btrfs_ino(inode);
4242 key.type = BTRFS_XATTR_ITEM_KEY;
4245 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4250 int slot = path->slots[0];
4251 struct extent_buffer *leaf = path->nodes[0];
4252 int nritems = btrfs_header_nritems(leaf);
4254 if (slot >= nritems) {
4256 u64 last_extent = 0;
4258 ret = copy_items(trans, inode, dst_path, path,
4259 &last_extent, start_slot,
4261 /* can't be 1, extent items aren't processed */
4267 ret = btrfs_next_leaf(root, path);
4275 btrfs_item_key_to_cpu(leaf, &key, slot);
4276 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4286 u64 last_extent = 0;
4288 ret = copy_items(trans, inode, dst_path, path,
4289 &last_extent, start_slot,
4291 /* can't be 1, extent items aren't processed */
4301 * If the no holes feature is enabled we need to make sure any hole between the
4302 * last extent and the i_size of our inode is explicitly marked in the log. This
4303 * is to make sure that doing something like:
4305 * 1) create file with 128Kb of data
4306 * 2) truncate file to 64Kb
4307 * 3) truncate file to 256Kb
4309 * 5) <crash/power failure>
4310 * 6) mount fs and trigger log replay
4312 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4313 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4314 * file correspond to a hole. The presence of explicit holes in a log tree is
4315 * what guarantees that log replay will remove/adjust file extent items in the
4318 * Here we do not need to care about holes between extents, that is already done
4319 * by copy_items(). We also only need to do this in the full sync path, where we
4320 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4321 * lookup the list of modified extent maps and if any represents a hole, we
4322 * insert a corresponding extent representing a hole in the log tree.
4324 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4325 struct btrfs_root *root,
4326 struct inode *inode,
4327 struct btrfs_path *path)
4330 struct btrfs_key key;
4333 struct extent_buffer *leaf;
4334 struct btrfs_root *log = root->log_root;
4335 const u64 ino = btrfs_ino(inode);
4336 const u64 i_size = i_size_read(inode);
4338 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4342 key.type = BTRFS_EXTENT_DATA_KEY;
4343 key.offset = (u64)-1;
4345 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4350 ASSERT(path->slots[0] > 0);
4352 leaf = path->nodes[0];
4353 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4355 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4356 /* inode does not have any extents */
4360 struct btrfs_file_extent_item *extent;
4364 * If there's an extent beyond i_size, an explicit hole was
4365 * already inserted by copy_items().
4367 if (key.offset >= i_size)
4370 extent = btrfs_item_ptr(leaf, path->slots[0],
4371 struct btrfs_file_extent_item);
4373 if (btrfs_file_extent_type(leaf, extent) ==
4374 BTRFS_FILE_EXTENT_INLINE) {
4375 len = btrfs_file_extent_inline_len(leaf,
4378 ASSERT(len == i_size);
4382 len = btrfs_file_extent_num_bytes(leaf, extent);
4383 /* Last extent goes beyond i_size, no need to log a hole. */
4384 if (key.offset + len > i_size)
4386 hole_start = key.offset + len;
4387 hole_size = i_size - hole_start;
4389 btrfs_release_path(path);
4391 /* Last extent ends at i_size. */
4395 hole_size = ALIGN(hole_size, root->sectorsize);
4396 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4397 hole_size, 0, hole_size, 0, 0, 0);
4402 * When we are logging a new inode X, check if it doesn't have a reference that
4403 * matches the reference from some other inode Y created in a past transaction
4404 * and that was renamed in the current transaction. If we don't do this, then at
4405 * log replay time we can lose inode Y (and all its files if it's a directory):
4408 * echo "hello world" > /mnt/x/foobar
4411 * mkdir /mnt/x # or touch /mnt/x
4412 * xfs_io -c fsync /mnt/x
4414 * mount fs, trigger log replay
4416 * After the log replay procedure, we would lose the first directory and all its
4417 * files (file foobar).
4418 * For the case where inode Y is not a directory we simply end up losing it:
4420 * echo "123" > /mnt/foo
4422 * mv /mnt/foo /mnt/bar
4423 * echo "abc" > /mnt/foo
4424 * xfs_io -c fsync /mnt/foo
4427 * We also need this for cases where a snapshot entry is replaced by some other
4428 * entry (file or directory) otherwise we end up with an unreplayable log due to
4429 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4430 * if it were a regular entry:
4433 * btrfs subvolume snapshot /mnt /mnt/x/snap
4434 * btrfs subvolume delete /mnt/x/snap
4437 * fsync /mnt/x or fsync some new file inside it
4440 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4441 * the same transaction.
4443 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4445 const struct btrfs_key *key,
4446 struct inode *inode)
4449 struct btrfs_path *search_path;
4452 u32 item_size = btrfs_item_size_nr(eb, slot);
4454 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4456 search_path = btrfs_alloc_path();
4459 search_path->search_commit_root = 1;
4460 search_path->skip_locking = 1;
4462 while (cur_offset < item_size) {
4466 unsigned long name_ptr;
4467 struct btrfs_dir_item *di;
4469 if (key->type == BTRFS_INODE_REF_KEY) {
4470 struct btrfs_inode_ref *iref;
4472 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4473 parent = key->offset;
4474 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4475 name_ptr = (unsigned long)(iref + 1);
4476 this_len = sizeof(*iref) + this_name_len;
4478 struct btrfs_inode_extref *extref;
4480 extref = (struct btrfs_inode_extref *)(ptr +
4482 parent = btrfs_inode_extref_parent(eb, extref);
4483 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4484 name_ptr = (unsigned long)&extref->name;
4485 this_len = sizeof(*extref) + this_name_len;
4488 if (this_name_len > name_len) {
4491 new_name = krealloc(name, this_name_len, GFP_NOFS);
4496 name_len = this_name_len;
4500 read_extent_buffer(eb, name, name_ptr, this_name_len);
4501 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4502 search_path, parent,
4503 name, this_name_len, 0);
4504 if (di && !IS_ERR(di)) {
4507 } else if (IS_ERR(di)) {
4511 btrfs_release_path(search_path);
4513 cur_offset += this_len;
4517 btrfs_free_path(search_path);
4522 /* log a single inode in the tree log.
4523 * At least one parent directory for this inode must exist in the tree
4524 * or be logged already.
4526 * Any items from this inode changed by the current transaction are copied
4527 * to the log tree. An extra reference is taken on any extents in this
4528 * file, allowing us to avoid a whole pile of corner cases around logging
4529 * blocks that have been removed from the tree.
4531 * See LOG_INODE_ALL and related defines for a description of what inode_only
4534 * This handles both files and directories.
4536 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4537 struct btrfs_root *root, struct inode *inode,
4541 struct btrfs_log_ctx *ctx)
4543 struct btrfs_path *path;
4544 struct btrfs_path *dst_path;
4545 struct btrfs_key min_key;
4546 struct btrfs_key max_key;
4547 struct btrfs_root *log = root->log_root;
4548 struct extent_buffer *src = NULL;
4549 LIST_HEAD(logged_list);
4550 u64 last_extent = 0;
4554 int ins_start_slot = 0;
4556 bool fast_search = false;
4557 u64 ino = btrfs_ino(inode);
4558 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4559 u64 logged_isize = 0;
4560 bool need_log_inode_item = true;
4562 path = btrfs_alloc_path();
4565 dst_path = btrfs_alloc_path();
4567 btrfs_free_path(path);
4571 min_key.objectid = ino;
4572 min_key.type = BTRFS_INODE_ITEM_KEY;
4575 max_key.objectid = ino;
4578 /* today the code can only do partial logging of directories */
4579 if (S_ISDIR(inode->i_mode) ||
4580 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4581 &BTRFS_I(inode)->runtime_flags) &&
4582 inode_only == LOG_INODE_EXISTS))
4583 max_key.type = BTRFS_XATTR_ITEM_KEY;
4585 max_key.type = (u8)-1;
4586 max_key.offset = (u64)-1;
4589 * Only run delayed items if we are a dir or a new file.
4590 * Otherwise commit the delayed inode only, which is needed in
4591 * order for the log replay code to mark inodes for link count
4592 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4594 if (S_ISDIR(inode->i_mode) ||
4595 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
4596 ret = btrfs_commit_inode_delayed_items(trans, inode);
4598 ret = btrfs_commit_inode_delayed_inode(inode);
4601 btrfs_free_path(path);
4602 btrfs_free_path(dst_path);
4606 mutex_lock(&BTRFS_I(inode)->log_mutex);
4608 btrfs_get_logged_extents(inode, &logged_list, start, end);
4611 * a brute force approach to making sure we get the most uptodate
4612 * copies of everything.
4614 if (S_ISDIR(inode->i_mode)) {
4615 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4617 if (inode_only == LOG_INODE_EXISTS)
4618 max_key_type = BTRFS_XATTR_ITEM_KEY;
4619 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4621 if (inode_only == LOG_INODE_EXISTS) {
4623 * Make sure the new inode item we write to the log has
4624 * the same isize as the current one (if it exists).
4625 * This is necessary to prevent data loss after log
4626 * replay, and also to prevent doing a wrong expanding
4627 * truncate - for e.g. create file, write 4K into offset
4628 * 0, fsync, write 4K into offset 4096, add hard link,
4629 * fsync some other file (to sync log), power fail - if
4630 * we use the inode's current i_size, after log replay
4631 * we get a 8Kb file, with the last 4Kb extent as a hole
4632 * (zeroes), as if an expanding truncate happened,
4633 * instead of getting a file of 4Kb only.
4635 err = logged_inode_size(log, inode, path,
4640 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4641 &BTRFS_I(inode)->runtime_flags)) {
4642 if (inode_only == LOG_INODE_EXISTS) {
4643 max_key.type = BTRFS_XATTR_ITEM_KEY;
4644 ret = drop_objectid_items(trans, log, path, ino,
4647 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4648 &BTRFS_I(inode)->runtime_flags);
4649 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4650 &BTRFS_I(inode)->runtime_flags);
4652 ret = btrfs_truncate_inode_items(trans,
4658 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4659 &BTRFS_I(inode)->runtime_flags) ||
4660 inode_only == LOG_INODE_EXISTS) {
4661 if (inode_only == LOG_INODE_ALL)
4663 max_key.type = BTRFS_XATTR_ITEM_KEY;
4664 ret = drop_objectid_items(trans, log, path, ino,
4667 if (inode_only == LOG_INODE_ALL)
4680 ret = btrfs_search_forward(root, &min_key,
4681 path, trans->transid);
4685 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4686 if (min_key.objectid != ino)
4688 if (min_key.type > max_key.type)
4691 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4692 need_log_inode_item = false;
4694 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4695 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4696 BTRFS_I(inode)->generation == trans->transid) {
4697 ret = btrfs_check_ref_name_override(path->nodes[0],
4703 } else if (ret > 0) {
4705 btrfs_set_log_full_commit(root->fs_info, trans);
4710 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4711 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4714 ret = copy_items(trans, inode, dst_path, path,
4715 &last_extent, ins_start_slot,
4716 ins_nr, inode_only, logged_isize);
4723 btrfs_release_path(path);
4729 src = path->nodes[0];
4730 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4733 } else if (!ins_nr) {
4734 ins_start_slot = path->slots[0];
4739 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4740 ins_start_slot, ins_nr, inode_only,
4748 btrfs_release_path(path);
4752 ins_start_slot = path->slots[0];
4755 nritems = btrfs_header_nritems(path->nodes[0]);
4757 if (path->slots[0] < nritems) {
4758 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4763 ret = copy_items(trans, inode, dst_path, path,
4764 &last_extent, ins_start_slot,
4765 ins_nr, inode_only, logged_isize);
4773 btrfs_release_path(path);
4775 if (min_key.offset < (u64)-1) {
4777 } else if (min_key.type < max_key.type) {
4785 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4786 ins_start_slot, ins_nr, inode_only,
4796 btrfs_release_path(path);
4797 btrfs_release_path(dst_path);
4798 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4801 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4802 btrfs_release_path(path);
4803 btrfs_release_path(dst_path);
4804 err = btrfs_log_trailing_hole(trans, root, inode, path);
4809 btrfs_release_path(path);
4810 btrfs_release_path(dst_path);
4811 if (need_log_inode_item) {
4812 err = log_inode_item(trans, log, dst_path, inode);
4818 * Some ordered extents started by fsync might have completed
4819 * before we collected the ordered extents in logged_list, which
4820 * means they're gone, not in our logged_list nor in the inode's
4821 * ordered tree. We want the application/user space to know an
4822 * error happened while attempting to persist file data so that
4823 * it can take proper action. If such error happened, we leave
4824 * without writing to the log tree and the fsync must report the
4825 * file data write error and not commit the current transaction.
4827 err = btrfs_inode_check_errors(inode);
4832 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4838 } else if (inode_only == LOG_INODE_ALL) {
4839 struct extent_map *em, *n;
4841 write_lock(&em_tree->lock);
4843 * We can't just remove every em if we're called for a ranged
4844 * fsync - that is, one that doesn't cover the whole possible
4845 * file range (0 to LLONG_MAX). This is because we can have
4846 * em's that fall outside the range we're logging and therefore
4847 * their ordered operations haven't completed yet
4848 * (btrfs_finish_ordered_io() not invoked yet). This means we
4849 * didn't get their respective file extent item in the fs/subvol
4850 * tree yet, and need to let the next fast fsync (one which
4851 * consults the list of modified extent maps) find the em so
4852 * that it logs a matching file extent item and waits for the
4853 * respective ordered operation to complete (if it's still
4856 * Removing every em outside the range we're logging would make
4857 * the next fast fsync not log their matching file extent items,
4858 * therefore making us lose data after a log replay.
4860 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4862 const u64 mod_end = em->mod_start + em->mod_len - 1;
4864 if (em->mod_start >= start && mod_end <= end)
4865 list_del_init(&em->list);
4867 write_unlock(&em_tree->lock);
4870 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4871 ret = log_directory_changes(trans, root, inode, path, dst_path,
4879 spin_lock(&BTRFS_I(inode)->lock);
4880 BTRFS_I(inode)->logged_trans = trans->transid;
4881 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4882 spin_unlock(&BTRFS_I(inode)->lock);
4885 btrfs_put_logged_extents(&logged_list);
4887 btrfs_submit_logged_extents(&logged_list, log);
4888 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4890 btrfs_free_path(path);
4891 btrfs_free_path(dst_path);
4896 * follow the dentry parent pointers up the chain and see if any
4897 * of the directories in it require a full commit before they can
4898 * be logged. Returns zero if nothing special needs to be done or 1 if
4899 * a full commit is required.
4901 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4902 struct inode *inode,
4903 struct dentry *parent,
4904 struct super_block *sb,
4908 struct btrfs_root *root;
4909 struct dentry *old_parent = NULL;
4910 struct inode *orig_inode = inode;
4913 * for regular files, if its inode is already on disk, we don't
4914 * have to worry about the parents at all. This is because
4915 * we can use the last_unlink_trans field to record renames
4916 * and other fun in this file.
4918 if (S_ISREG(inode->i_mode) &&
4919 BTRFS_I(inode)->generation <= last_committed &&
4920 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4923 if (!S_ISDIR(inode->i_mode)) {
4924 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
4926 inode = d_inode(parent);
4931 * If we are logging a directory then we start with our inode,
4932 * not our parents inode, so we need to skipp setting the
4933 * logged_trans so that further down in the log code we don't
4934 * think this inode has already been logged.
4936 if (inode != orig_inode)
4937 BTRFS_I(inode)->logged_trans = trans->transid;
4940 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4941 root = BTRFS_I(inode)->root;
4944 * make sure any commits to the log are forced
4945 * to be full commits
4947 btrfs_set_log_full_commit(root->fs_info, trans);
4952 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
4955 if (IS_ROOT(parent))
4958 parent = dget_parent(parent);
4960 old_parent = parent;
4961 inode = d_inode(parent);
4969 struct btrfs_dir_list {
4971 struct list_head list;
4975 * Log the inodes of the new dentries of a directory. See log_dir_items() for
4976 * details about the why it is needed.
4977 * This is a recursive operation - if an existing dentry corresponds to a
4978 * directory, that directory's new entries are logged too (same behaviour as
4979 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
4980 * the dentries point to we do not lock their i_mutex, otherwise lockdep
4981 * complains about the following circular lock dependency / possible deadlock:
4985 * lock(&type->i_mutex_dir_key#3/2);
4986 * lock(sb_internal#2);
4987 * lock(&type->i_mutex_dir_key#3/2);
4988 * lock(&sb->s_type->i_mutex_key#14);
4990 * Where sb_internal is the lock (a counter that works as a lock) acquired by
4991 * sb_start_intwrite() in btrfs_start_transaction().
4992 * Not locking i_mutex of the inodes is still safe because:
4994 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
4995 * that while logging the inode new references (names) are added or removed
4996 * from the inode, leaving the logged inode item with a link count that does
4997 * not match the number of logged inode reference items. This is fine because
4998 * at log replay time we compute the real number of links and correct the
4999 * link count in the inode item (see replay_one_buffer() and
5000 * link_to_fixup_dir());
5002 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5003 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5004 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5005 * has a size that doesn't match the sum of the lengths of all the logged
5006 * names. This does not result in a problem because if a dir_item key is
5007 * logged but its matching dir_index key is not logged, at log replay time we
5008 * don't use it to replay the respective name (see replay_one_name()). On the
5009 * other hand if only the dir_index key ends up being logged, the respective
5010 * name is added to the fs/subvol tree with both the dir_item and dir_index
5011 * keys created (see replay_one_name()).
5012 * The directory's inode item with a wrong i_size is not a problem as well,
5013 * since we don't use it at log replay time to set the i_size in the inode
5014 * item of the fs/subvol tree (see overwrite_item()).
5016 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *root,
5018 struct inode *start_inode,
5019 struct btrfs_log_ctx *ctx)
5021 struct btrfs_root *log = root->log_root;
5022 struct btrfs_path *path;
5023 LIST_HEAD(dir_list);
5024 struct btrfs_dir_list *dir_elem;
5027 path = btrfs_alloc_path();
5031 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5033 btrfs_free_path(path);
5036 dir_elem->ino = btrfs_ino(start_inode);
5037 list_add_tail(&dir_elem->list, &dir_list);
5039 while (!list_empty(&dir_list)) {
5040 struct extent_buffer *leaf;
5041 struct btrfs_key min_key;
5045 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5048 goto next_dir_inode;
5050 min_key.objectid = dir_elem->ino;
5051 min_key.type = BTRFS_DIR_ITEM_KEY;
5054 btrfs_release_path(path);
5055 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5057 goto next_dir_inode;
5058 } else if (ret > 0) {
5060 goto next_dir_inode;
5064 leaf = path->nodes[0];
5065 nritems = btrfs_header_nritems(leaf);
5066 for (i = path->slots[0]; i < nritems; i++) {
5067 struct btrfs_dir_item *di;
5068 struct btrfs_key di_key;
5069 struct inode *di_inode;
5070 struct btrfs_dir_list *new_dir_elem;
5071 int log_mode = LOG_INODE_EXISTS;
5074 btrfs_item_key_to_cpu(leaf, &min_key, i);
5075 if (min_key.objectid != dir_elem->ino ||
5076 min_key.type != BTRFS_DIR_ITEM_KEY)
5077 goto next_dir_inode;
5079 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5080 type = btrfs_dir_type(leaf, di);
5081 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5082 type != BTRFS_FT_DIR)
5084 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5085 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5088 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5090 if (IS_ERR(di_inode)) {
5091 ret = PTR_ERR(di_inode);
5092 goto next_dir_inode;
5095 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5100 ctx->log_new_dentries = false;
5101 if (type == BTRFS_FT_DIR)
5102 log_mode = LOG_INODE_ALL;
5103 btrfs_release_path(path);
5104 ret = btrfs_log_inode(trans, root, di_inode,
5105 log_mode, 0, LLONG_MAX, ctx);
5108 goto next_dir_inode;
5109 if (ctx->log_new_dentries) {
5110 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5112 if (!new_dir_elem) {
5114 goto next_dir_inode;
5116 new_dir_elem->ino = di_key.objectid;
5117 list_add_tail(&new_dir_elem->list, &dir_list);
5122 ret = btrfs_next_leaf(log, path);
5124 goto next_dir_inode;
5125 } else if (ret > 0) {
5127 goto next_dir_inode;
5131 if (min_key.offset < (u64)-1) {
5136 list_del(&dir_elem->list);
5140 btrfs_free_path(path);
5144 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5145 struct inode *inode,
5146 struct btrfs_log_ctx *ctx)
5149 struct btrfs_path *path;
5150 struct btrfs_key key;
5151 struct btrfs_root *root = BTRFS_I(inode)->root;
5152 const u64 ino = btrfs_ino(inode);
5154 path = btrfs_alloc_path();
5157 path->skip_locking = 1;
5158 path->search_commit_root = 1;
5161 key.type = BTRFS_INODE_REF_KEY;
5163 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5168 struct extent_buffer *leaf = path->nodes[0];
5169 int slot = path->slots[0];
5174 if (slot >= btrfs_header_nritems(leaf)) {
5175 ret = btrfs_next_leaf(root, path);
5183 btrfs_item_key_to_cpu(leaf, &key, slot);
5184 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5185 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5188 item_size = btrfs_item_size_nr(leaf, slot);
5189 ptr = btrfs_item_ptr_offset(leaf, slot);
5190 while (cur_offset < item_size) {
5191 struct btrfs_key inode_key;
5192 struct inode *dir_inode;
5194 inode_key.type = BTRFS_INODE_ITEM_KEY;
5195 inode_key.offset = 0;
5197 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5198 struct btrfs_inode_extref *extref;
5200 extref = (struct btrfs_inode_extref *)
5202 inode_key.objectid = btrfs_inode_extref_parent(
5204 cur_offset += sizeof(*extref);
5205 cur_offset += btrfs_inode_extref_name_len(leaf,
5208 inode_key.objectid = key.offset;
5209 cur_offset = item_size;
5212 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5214 /* If parent inode was deleted, skip it. */
5215 if (IS_ERR(dir_inode))
5218 ret = btrfs_log_inode(trans, root, dir_inode,
5219 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5228 btrfs_free_path(path);
5233 * helper function around btrfs_log_inode to make sure newly created
5234 * parent directories also end up in the log. A minimal inode and backref
5235 * only logging is done of any parent directories that are older than
5236 * the last committed transaction
5238 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5239 struct btrfs_root *root, struct inode *inode,
5240 struct dentry *parent,
5244 struct btrfs_log_ctx *ctx)
5246 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5247 struct super_block *sb;
5248 struct dentry *old_parent = NULL;
5250 u64 last_committed = root->fs_info->last_trans_committed;
5251 bool log_dentries = false;
5252 struct inode *orig_inode = inode;
5256 if (btrfs_test_opt(root, NOTREELOG)) {
5262 * The prev transaction commit doesn't complete, we need do
5263 * full commit by ourselves.
5265 if (root->fs_info->last_trans_log_full_commit >
5266 root->fs_info->last_trans_committed) {
5271 if (root != BTRFS_I(inode)->root ||
5272 btrfs_root_refs(&root->root_item) == 0) {
5277 ret = check_parent_dirs_for_sync(trans, inode, parent,
5278 sb, last_committed);
5282 if (btrfs_inode_in_log(inode, trans->transid)) {
5283 ret = BTRFS_NO_LOG_SYNC;
5287 ret = start_log_trans(trans, root, ctx);
5291 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5296 * for regular files, if its inode is already on disk, we don't
5297 * have to worry about the parents at all. This is because
5298 * we can use the last_unlink_trans field to record renames
5299 * and other fun in this file.
5301 if (S_ISREG(inode->i_mode) &&
5302 BTRFS_I(inode)->generation <= last_committed &&
5303 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5308 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5309 log_dentries = true;
5312 * On unlink we must make sure all our current and old parent directores
5313 * inodes are fully logged. This is to prevent leaving dangling
5314 * directory index entries in directories that were our parents but are
5315 * not anymore. Not doing this results in old parent directory being
5316 * impossible to delete after log replay (rmdir will always fail with
5317 * error -ENOTEMPTY).
5323 * ln testdir/foo testdir/bar
5325 * unlink testdir/bar
5326 * xfs_io -c fsync testdir/foo
5328 * mount fs, triggers log replay
5330 * If we don't log the parent directory (testdir), after log replay the
5331 * directory still has an entry pointing to the file inode using the bar
5332 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5333 * the file inode has a link count of 1.
5339 * ln foo testdir/foo2
5340 * ln foo testdir/foo3
5342 * unlink testdir/foo3
5343 * xfs_io -c fsync foo
5345 * mount fs, triggers log replay
5347 * Similar as the first example, after log replay the parent directory
5348 * testdir still has an entry pointing to the inode file with name foo3
5349 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5350 * and has a link count of 2.
5352 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5353 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5359 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
5362 inode = d_inode(parent);
5363 if (root != BTRFS_I(inode)->root)
5366 if (BTRFS_I(inode)->generation > last_committed) {
5367 ret = btrfs_log_inode(trans, root, inode,
5373 if (IS_ROOT(parent))
5376 parent = dget_parent(parent);
5378 old_parent = parent;
5381 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5387 btrfs_set_log_full_commit(root->fs_info, trans);
5392 btrfs_remove_log_ctx(root, ctx);
5393 btrfs_end_log_trans(root);
5399 * it is not safe to log dentry if the chunk root has added new
5400 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5401 * If this returns 1, you must commit the transaction to safely get your
5404 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5405 struct btrfs_root *root, struct dentry *dentry,
5408 struct btrfs_log_ctx *ctx)
5410 struct dentry *parent = dget_parent(dentry);
5413 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5414 start, end, 0, ctx);
5421 * should be called during mount to recover any replay any log trees
5424 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5427 struct btrfs_path *path;
5428 struct btrfs_trans_handle *trans;
5429 struct btrfs_key key;
5430 struct btrfs_key found_key;
5431 struct btrfs_key tmp_key;
5432 struct btrfs_root *log;
5433 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5434 struct walk_control wc = {
5435 .process_func = process_one_buffer,
5439 path = btrfs_alloc_path();
5443 fs_info->log_root_recovering = 1;
5445 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5446 if (IS_ERR(trans)) {
5447 ret = PTR_ERR(trans);
5454 ret = walk_log_tree(trans, log_root_tree, &wc);
5456 btrfs_std_error(fs_info, ret, "Failed to pin buffers while "
5457 "recovering log root tree.");
5462 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5463 key.offset = (u64)-1;
5464 key.type = BTRFS_ROOT_ITEM_KEY;
5467 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5470 btrfs_std_error(fs_info, ret,
5471 "Couldn't find tree log root.");
5475 if (path->slots[0] == 0)
5479 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5481 btrfs_release_path(path);
5482 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5485 log = btrfs_read_fs_root(log_root_tree, &found_key);
5488 btrfs_std_error(fs_info, ret,
5489 "Couldn't read tree log root.");
5493 tmp_key.objectid = found_key.offset;
5494 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5495 tmp_key.offset = (u64)-1;
5497 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5498 if (IS_ERR(wc.replay_dest)) {
5499 ret = PTR_ERR(wc.replay_dest);
5500 free_extent_buffer(log->node);
5501 free_extent_buffer(log->commit_root);
5503 btrfs_std_error(fs_info, ret, "Couldn't read target root "
5504 "for tree log recovery.");
5508 wc.replay_dest->log_root = log;
5509 btrfs_record_root_in_trans(trans, wc.replay_dest);
5510 ret = walk_log_tree(trans, log, &wc);
5512 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5513 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5517 key.offset = found_key.offset - 1;
5518 wc.replay_dest->log_root = NULL;
5519 free_extent_buffer(log->node);
5520 free_extent_buffer(log->commit_root);
5526 if (found_key.offset == 0)
5529 btrfs_release_path(path);
5531 /* step one is to pin it all, step two is to replay just inodes */
5534 wc.process_func = replay_one_buffer;
5535 wc.stage = LOG_WALK_REPLAY_INODES;
5538 /* step three is to replay everything */
5539 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5544 btrfs_free_path(path);
5546 /* step 4: commit the transaction, which also unpins the blocks */
5547 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5551 free_extent_buffer(log_root_tree->node);
5552 log_root_tree->log_root = NULL;
5553 fs_info->log_root_recovering = 0;
5554 kfree(log_root_tree);
5559 btrfs_end_transaction(wc.trans, fs_info->tree_root);
5560 btrfs_free_path(path);
5565 * there are some corner cases where we want to force a full
5566 * commit instead of allowing a directory to be logged.
5568 * They revolve around files there were unlinked from the directory, and
5569 * this function updates the parent directory so that a full commit is
5570 * properly done if it is fsync'd later after the unlinks are done.
5572 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5573 struct inode *dir, struct inode *inode,
5577 * when we're logging a file, if it hasn't been renamed
5578 * or unlinked, and its inode is fully committed on disk,
5579 * we don't have to worry about walking up the directory chain
5580 * to log its parents.
5582 * So, we use the last_unlink_trans field to put this transid
5583 * into the file. When the file is logged we check it and
5584 * don't log the parents if the file is fully on disk.
5586 if (S_ISREG(inode->i_mode))
5587 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5590 * if this directory was already logged any new
5591 * names for this file/dir will get recorded
5594 if (BTRFS_I(dir)->logged_trans == trans->transid)
5598 * if the inode we're about to unlink was logged,
5599 * the log will be properly updated for any new names
5601 if (BTRFS_I(inode)->logged_trans == trans->transid)
5605 * when renaming files across directories, if the directory
5606 * there we're unlinking from gets fsync'd later on, there's
5607 * no way to find the destination directory later and fsync it
5608 * properly. So, we have to be conservative and force commits
5609 * so the new name gets discovered.
5614 /* we can safely do the unlink without any special recording */
5618 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5622 * Call this after adding a new name for a file and it will properly
5623 * update the log to reflect the new name.
5625 * It will return zero if all goes well, and it will return 1 if a
5626 * full transaction commit is required.
5628 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5629 struct inode *inode, struct inode *old_dir,
5630 struct dentry *parent)
5632 struct btrfs_root * root = BTRFS_I(inode)->root;
5635 * this will force the logging code to walk the dentry chain
5638 if (S_ISREG(inode->i_mode))
5639 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5642 * if this inode hasn't been logged and directory we're renaming it
5643 * from hasn't been logged, we don't need to log it
5645 if (BTRFS_I(inode)->logged_trans <=
5646 root->fs_info->last_trans_committed &&
5647 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5648 root->fs_info->last_trans_committed))
5651 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5652 LLONG_MAX, 1, NULL);