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]);
1926 if (path->slots[0] >= nritems) {
1927 ret = btrfs_next_leaf(root, path);
1934 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1936 if (key.type != key_type || key.objectid != dirid) {
1940 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1941 struct btrfs_dir_log_item);
1942 found_end = btrfs_dir_log_end(path->nodes[0], item);
1943 *start_ret = key.offset;
1944 *end_ret = found_end;
1947 btrfs_release_path(path);
1952 * this looks for a given directory item in the log. If the directory
1953 * item is not in the log, the item is removed and the inode it points
1956 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1957 struct btrfs_root *root,
1958 struct btrfs_root *log,
1959 struct btrfs_path *path,
1960 struct btrfs_path *log_path,
1962 struct btrfs_key *dir_key)
1965 struct extent_buffer *eb;
1968 struct btrfs_dir_item *di;
1969 struct btrfs_dir_item *log_di;
1972 unsigned long ptr_end;
1974 struct inode *inode;
1975 struct btrfs_key location;
1978 eb = path->nodes[0];
1979 slot = path->slots[0];
1980 item_size = btrfs_item_size_nr(eb, slot);
1981 ptr = btrfs_item_ptr_offset(eb, slot);
1982 ptr_end = ptr + item_size;
1983 while (ptr < ptr_end) {
1984 di = (struct btrfs_dir_item *)ptr;
1985 if (verify_dir_item(root, eb, di)) {
1990 name_len = btrfs_dir_name_len(eb, di);
1991 name = kmalloc(name_len, GFP_NOFS);
1996 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1999 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2000 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2003 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2004 log_di = btrfs_lookup_dir_index_item(trans, log,
2010 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2011 btrfs_dir_item_key_to_cpu(eb, di, &location);
2012 btrfs_release_path(path);
2013 btrfs_release_path(log_path);
2014 inode = read_one_inode(root, location.objectid);
2020 ret = link_to_fixup_dir(trans, root,
2021 path, location.objectid);
2029 ret = btrfs_unlink_inode(trans, root, dir, inode,
2032 ret = btrfs_run_delayed_items(trans, root);
2038 /* there might still be more names under this key
2039 * check and repeat if required
2041 ret = btrfs_search_slot(NULL, root, dir_key, path,
2047 } else if (IS_ERR(log_di)) {
2049 return PTR_ERR(log_di);
2051 btrfs_release_path(log_path);
2054 ptr = (unsigned long)(di + 1);
2059 btrfs_release_path(path);
2060 btrfs_release_path(log_path);
2064 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2065 struct btrfs_root *root,
2066 struct btrfs_root *log,
2067 struct btrfs_path *path,
2070 struct btrfs_key search_key;
2071 struct btrfs_path *log_path;
2076 log_path = btrfs_alloc_path();
2080 search_key.objectid = ino;
2081 search_key.type = BTRFS_XATTR_ITEM_KEY;
2082 search_key.offset = 0;
2084 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2088 nritems = btrfs_header_nritems(path->nodes[0]);
2089 for (i = path->slots[0]; i < nritems; i++) {
2090 struct btrfs_key key;
2091 struct btrfs_dir_item *di;
2092 struct btrfs_dir_item *log_di;
2096 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2097 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2102 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2103 total_size = btrfs_item_size_nr(path->nodes[0], i);
2105 while (cur < total_size) {
2106 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2107 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2108 u32 this_len = sizeof(*di) + name_len + data_len;
2111 name = kmalloc(name_len, GFP_NOFS);
2116 read_extent_buffer(path->nodes[0], name,
2117 (unsigned long)(di + 1), name_len);
2119 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2121 btrfs_release_path(log_path);
2123 /* Doesn't exist in log tree, so delete it. */
2124 btrfs_release_path(path);
2125 di = btrfs_lookup_xattr(trans, root, path, ino,
2126 name, name_len, -1);
2133 ret = btrfs_delete_one_dir_name(trans, root,
2137 btrfs_release_path(path);
2142 if (IS_ERR(log_di)) {
2143 ret = PTR_ERR(log_di);
2147 di = (struct btrfs_dir_item *)((char *)di + this_len);
2150 ret = btrfs_next_leaf(root, path);
2156 btrfs_free_path(log_path);
2157 btrfs_release_path(path);
2163 * deletion replay happens before we copy any new directory items
2164 * out of the log or out of backreferences from inodes. It
2165 * scans the log to find ranges of keys that log is authoritative for,
2166 * and then scans the directory to find items in those ranges that are
2167 * not present in the log.
2169 * Anything we don't find in the log is unlinked and removed from the
2172 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2173 struct btrfs_root *root,
2174 struct btrfs_root *log,
2175 struct btrfs_path *path,
2176 u64 dirid, int del_all)
2180 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2182 struct btrfs_key dir_key;
2183 struct btrfs_key found_key;
2184 struct btrfs_path *log_path;
2187 dir_key.objectid = dirid;
2188 dir_key.type = BTRFS_DIR_ITEM_KEY;
2189 log_path = btrfs_alloc_path();
2193 dir = read_one_inode(root, dirid);
2194 /* it isn't an error if the inode isn't there, that can happen
2195 * because we replay the deletes before we copy in the inode item
2199 btrfs_free_path(log_path);
2207 range_end = (u64)-1;
2209 ret = find_dir_range(log, path, dirid, key_type,
2210 &range_start, &range_end);
2215 dir_key.offset = range_start;
2218 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2223 nritems = btrfs_header_nritems(path->nodes[0]);
2224 if (path->slots[0] >= nritems) {
2225 ret = btrfs_next_leaf(root, path);
2229 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2231 if (found_key.objectid != dirid ||
2232 found_key.type != dir_key.type)
2235 if (found_key.offset > range_end)
2238 ret = check_item_in_log(trans, root, log, path,
2243 if (found_key.offset == (u64)-1)
2245 dir_key.offset = found_key.offset + 1;
2247 btrfs_release_path(path);
2248 if (range_end == (u64)-1)
2250 range_start = range_end + 1;
2255 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2256 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2257 dir_key.type = BTRFS_DIR_INDEX_KEY;
2258 btrfs_release_path(path);
2262 btrfs_release_path(path);
2263 btrfs_free_path(log_path);
2269 * the process_func used to replay items from the log tree. This
2270 * gets called in two different stages. The first stage just looks
2271 * for inodes and makes sure they are all copied into the subvolume.
2273 * The second stage copies all the other item types from the log into
2274 * the subvolume. The two stage approach is slower, but gets rid of
2275 * lots of complexity around inodes referencing other inodes that exist
2276 * only in the log (references come from either directory items or inode
2279 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2280 struct walk_control *wc, u64 gen)
2283 struct btrfs_path *path;
2284 struct btrfs_root *root = wc->replay_dest;
2285 struct btrfs_key key;
2290 ret = btrfs_read_buffer(eb, gen);
2294 level = btrfs_header_level(eb);
2299 path = btrfs_alloc_path();
2303 nritems = btrfs_header_nritems(eb);
2304 for (i = 0; i < nritems; i++) {
2305 btrfs_item_key_to_cpu(eb, &key, i);
2307 /* inode keys are done during the first stage */
2308 if (key.type == BTRFS_INODE_ITEM_KEY &&
2309 wc->stage == LOG_WALK_REPLAY_INODES) {
2310 struct btrfs_inode_item *inode_item;
2313 inode_item = btrfs_item_ptr(eb, i,
2314 struct btrfs_inode_item);
2315 ret = replay_xattr_deletes(wc->trans, root, log,
2316 path, key.objectid);
2319 mode = btrfs_inode_mode(eb, inode_item);
2320 if (S_ISDIR(mode)) {
2321 ret = replay_dir_deletes(wc->trans,
2322 root, log, path, key.objectid, 0);
2326 ret = overwrite_item(wc->trans, root, path,
2331 /* for regular files, make sure corresponding
2332 * orhpan item exist. extents past the new EOF
2333 * will be truncated later by orphan cleanup.
2335 if (S_ISREG(mode)) {
2336 ret = insert_orphan_item(wc->trans, root,
2342 ret = link_to_fixup_dir(wc->trans, root,
2343 path, key.objectid);
2348 if (key.type == BTRFS_DIR_INDEX_KEY &&
2349 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2350 ret = replay_one_dir_item(wc->trans, root, path,
2356 if (wc->stage < LOG_WALK_REPLAY_ALL)
2359 /* these keys are simply copied */
2360 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2361 ret = overwrite_item(wc->trans, root, path,
2365 } else if (key.type == BTRFS_INODE_REF_KEY ||
2366 key.type == BTRFS_INODE_EXTREF_KEY) {
2367 ret = add_inode_ref(wc->trans, root, log, path,
2369 if (ret && ret != -ENOENT)
2372 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2373 ret = replay_one_extent(wc->trans, root, path,
2377 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2378 ret = replay_one_dir_item(wc->trans, root, path,
2384 btrfs_free_path(path);
2388 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2389 struct btrfs_root *root,
2390 struct btrfs_path *path, int *level,
2391 struct walk_control *wc)
2396 struct extent_buffer *next;
2397 struct extent_buffer *cur;
2398 struct extent_buffer *parent;
2402 WARN_ON(*level < 0);
2403 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2405 while (*level > 0) {
2406 WARN_ON(*level < 0);
2407 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2408 cur = path->nodes[*level];
2410 WARN_ON(btrfs_header_level(cur) != *level);
2412 if (path->slots[*level] >=
2413 btrfs_header_nritems(cur))
2416 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2417 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2418 blocksize = root->nodesize;
2420 parent = path->nodes[*level];
2421 root_owner = btrfs_header_owner(parent);
2423 next = btrfs_find_create_tree_block(root, bytenr);
2428 ret = wc->process_func(root, next, wc, ptr_gen);
2430 free_extent_buffer(next);
2434 path->slots[*level]++;
2436 ret = btrfs_read_buffer(next, ptr_gen);
2438 free_extent_buffer(next);
2443 btrfs_tree_lock(next);
2444 btrfs_set_lock_blocking(next);
2445 clean_tree_block(trans, root->fs_info,
2447 btrfs_wait_tree_block_writeback(next);
2448 btrfs_tree_unlock(next);
2451 WARN_ON(root_owner !=
2452 BTRFS_TREE_LOG_OBJECTID);
2453 ret = btrfs_free_and_pin_reserved_extent(root,
2456 free_extent_buffer(next);
2460 free_extent_buffer(next);
2463 ret = btrfs_read_buffer(next, ptr_gen);
2465 free_extent_buffer(next);
2469 WARN_ON(*level <= 0);
2470 if (path->nodes[*level-1])
2471 free_extent_buffer(path->nodes[*level-1]);
2472 path->nodes[*level-1] = next;
2473 *level = btrfs_header_level(next);
2474 path->slots[*level] = 0;
2477 WARN_ON(*level < 0);
2478 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2480 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2486 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2487 struct btrfs_root *root,
2488 struct btrfs_path *path, int *level,
2489 struct walk_control *wc)
2496 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2497 slot = path->slots[i];
2498 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2501 WARN_ON(*level == 0);
2504 struct extent_buffer *parent;
2505 if (path->nodes[*level] == root->node)
2506 parent = path->nodes[*level];
2508 parent = path->nodes[*level + 1];
2510 root_owner = btrfs_header_owner(parent);
2511 ret = wc->process_func(root, path->nodes[*level], wc,
2512 btrfs_header_generation(path->nodes[*level]));
2517 struct extent_buffer *next;
2519 next = path->nodes[*level];
2522 btrfs_tree_lock(next);
2523 btrfs_set_lock_blocking(next);
2524 clean_tree_block(trans, root->fs_info,
2526 btrfs_wait_tree_block_writeback(next);
2527 btrfs_tree_unlock(next);
2530 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2531 ret = btrfs_free_and_pin_reserved_extent(root,
2532 path->nodes[*level]->start,
2533 path->nodes[*level]->len);
2537 free_extent_buffer(path->nodes[*level]);
2538 path->nodes[*level] = NULL;
2546 * drop the reference count on the tree rooted at 'snap'. This traverses
2547 * the tree freeing any blocks that have a ref count of zero after being
2550 static int walk_log_tree(struct btrfs_trans_handle *trans,
2551 struct btrfs_root *log, struct walk_control *wc)
2556 struct btrfs_path *path;
2559 path = btrfs_alloc_path();
2563 level = btrfs_header_level(log->node);
2565 path->nodes[level] = log->node;
2566 extent_buffer_get(log->node);
2567 path->slots[level] = 0;
2570 wret = walk_down_log_tree(trans, log, path, &level, wc);
2578 wret = walk_up_log_tree(trans, log, path, &level, wc);
2587 /* was the root node processed? if not, catch it here */
2588 if (path->nodes[orig_level]) {
2589 ret = wc->process_func(log, path->nodes[orig_level], wc,
2590 btrfs_header_generation(path->nodes[orig_level]));
2594 struct extent_buffer *next;
2596 next = path->nodes[orig_level];
2599 btrfs_tree_lock(next);
2600 btrfs_set_lock_blocking(next);
2601 clean_tree_block(trans, log->fs_info, next);
2602 btrfs_wait_tree_block_writeback(next);
2603 btrfs_tree_unlock(next);
2606 WARN_ON(log->root_key.objectid !=
2607 BTRFS_TREE_LOG_OBJECTID);
2608 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2616 btrfs_free_path(path);
2621 * helper function to update the item for a given subvolumes log root
2622 * in the tree of log roots
2624 static int update_log_root(struct btrfs_trans_handle *trans,
2625 struct btrfs_root *log)
2629 if (log->log_transid == 1) {
2630 /* insert root item on the first sync */
2631 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2632 &log->root_key, &log->root_item);
2634 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2635 &log->root_key, &log->root_item);
2640 static void wait_log_commit(struct btrfs_root *root, int transid)
2643 int index = transid % 2;
2646 * we only allow two pending log transactions at a time,
2647 * so we know that if ours is more than 2 older than the
2648 * current transaction, we're done
2651 prepare_to_wait(&root->log_commit_wait[index],
2652 &wait, TASK_UNINTERRUPTIBLE);
2653 mutex_unlock(&root->log_mutex);
2655 if (root->log_transid_committed < transid &&
2656 atomic_read(&root->log_commit[index]))
2659 finish_wait(&root->log_commit_wait[index], &wait);
2660 mutex_lock(&root->log_mutex);
2661 } while (root->log_transid_committed < transid &&
2662 atomic_read(&root->log_commit[index]));
2665 static void wait_for_writer(struct btrfs_root *root)
2669 while (atomic_read(&root->log_writers)) {
2670 prepare_to_wait(&root->log_writer_wait,
2671 &wait, TASK_UNINTERRUPTIBLE);
2672 mutex_unlock(&root->log_mutex);
2673 if (atomic_read(&root->log_writers))
2675 finish_wait(&root->log_writer_wait, &wait);
2676 mutex_lock(&root->log_mutex);
2680 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2681 struct btrfs_log_ctx *ctx)
2686 mutex_lock(&root->log_mutex);
2687 list_del_init(&ctx->list);
2688 mutex_unlock(&root->log_mutex);
2692 * Invoked in log mutex context, or be sure there is no other task which
2693 * can access the list.
2695 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2696 int index, int error)
2698 struct btrfs_log_ctx *ctx;
2701 INIT_LIST_HEAD(&root->log_ctxs[index]);
2705 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2706 ctx->log_ret = error;
2708 INIT_LIST_HEAD(&root->log_ctxs[index]);
2712 * btrfs_sync_log does sends a given tree log down to the disk and
2713 * updates the super blocks to record it. When this call is done,
2714 * you know that any inodes previously logged are safely on disk only
2717 * Any other return value means you need to call btrfs_commit_transaction.
2718 * Some of the edge cases for fsyncing directories that have had unlinks
2719 * or renames done in the past mean that sometimes the only safe
2720 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2721 * that has happened.
2723 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2730 struct btrfs_root *log = root->log_root;
2731 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2732 int log_transid = 0;
2733 struct btrfs_log_ctx root_log_ctx;
2734 struct blk_plug plug;
2736 mutex_lock(&root->log_mutex);
2737 log_transid = ctx->log_transid;
2738 if (root->log_transid_committed >= log_transid) {
2739 mutex_unlock(&root->log_mutex);
2740 return ctx->log_ret;
2743 index1 = log_transid % 2;
2744 if (atomic_read(&root->log_commit[index1])) {
2745 wait_log_commit(root, log_transid);
2746 mutex_unlock(&root->log_mutex);
2747 return ctx->log_ret;
2749 ASSERT(log_transid == root->log_transid);
2750 atomic_set(&root->log_commit[index1], 1);
2752 /* wait for previous tree log sync to complete */
2753 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2754 wait_log_commit(root, log_transid - 1);
2757 int batch = atomic_read(&root->log_batch);
2758 /* when we're on an ssd, just kick the log commit out */
2759 if (!btrfs_test_opt(root, SSD) &&
2760 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2761 mutex_unlock(&root->log_mutex);
2762 schedule_timeout_uninterruptible(1);
2763 mutex_lock(&root->log_mutex);
2765 wait_for_writer(root);
2766 if (batch == atomic_read(&root->log_batch))
2770 /* bail out if we need to do a full commit */
2771 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2773 btrfs_free_logged_extents(log, log_transid);
2774 mutex_unlock(&root->log_mutex);
2778 if (log_transid % 2 == 0)
2779 mark = EXTENT_DIRTY;
2783 /* we start IO on all the marked extents here, but we don't actually
2784 * wait for them until later.
2786 blk_start_plug(&plug);
2787 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2789 blk_finish_plug(&plug);
2790 btrfs_abort_transaction(trans, root, ret);
2791 btrfs_free_logged_extents(log, log_transid);
2792 btrfs_set_log_full_commit(root->fs_info, trans);
2793 mutex_unlock(&root->log_mutex);
2797 btrfs_set_root_node(&log->root_item, log->node);
2799 root->log_transid++;
2800 log->log_transid = root->log_transid;
2801 root->log_start_pid = 0;
2803 * IO has been started, blocks of the log tree have WRITTEN flag set
2804 * in their headers. new modifications of the log will be written to
2805 * new positions. so it's safe to allow log writers to go in.
2807 mutex_unlock(&root->log_mutex);
2809 btrfs_init_log_ctx(&root_log_ctx);
2811 mutex_lock(&log_root_tree->log_mutex);
2812 atomic_inc(&log_root_tree->log_batch);
2813 atomic_inc(&log_root_tree->log_writers);
2815 index2 = log_root_tree->log_transid % 2;
2816 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2817 root_log_ctx.log_transid = log_root_tree->log_transid;
2819 mutex_unlock(&log_root_tree->log_mutex);
2821 ret = update_log_root(trans, log);
2823 mutex_lock(&log_root_tree->log_mutex);
2824 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2826 * Implicit memory barrier after atomic_dec_and_test
2828 if (waitqueue_active(&log_root_tree->log_writer_wait))
2829 wake_up(&log_root_tree->log_writer_wait);
2833 if (!list_empty(&root_log_ctx.list))
2834 list_del_init(&root_log_ctx.list);
2836 blk_finish_plug(&plug);
2837 btrfs_set_log_full_commit(root->fs_info, trans);
2839 if (ret != -ENOSPC) {
2840 btrfs_abort_transaction(trans, root, ret);
2841 mutex_unlock(&log_root_tree->log_mutex);
2844 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2845 btrfs_free_logged_extents(log, log_transid);
2846 mutex_unlock(&log_root_tree->log_mutex);
2851 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2852 blk_finish_plug(&plug);
2853 list_del_init(&root_log_ctx.list);
2854 mutex_unlock(&log_root_tree->log_mutex);
2855 ret = root_log_ctx.log_ret;
2859 index2 = root_log_ctx.log_transid % 2;
2860 if (atomic_read(&log_root_tree->log_commit[index2])) {
2861 blk_finish_plug(&plug);
2862 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2864 btrfs_wait_logged_extents(trans, log, log_transid);
2865 wait_log_commit(log_root_tree,
2866 root_log_ctx.log_transid);
2867 mutex_unlock(&log_root_tree->log_mutex);
2869 ret = root_log_ctx.log_ret;
2872 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2873 atomic_set(&log_root_tree->log_commit[index2], 1);
2875 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2876 wait_log_commit(log_root_tree,
2877 root_log_ctx.log_transid - 1);
2880 wait_for_writer(log_root_tree);
2883 * now that we've moved on to the tree of log tree roots,
2884 * check the full commit flag again
2886 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2887 blk_finish_plug(&plug);
2888 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2889 btrfs_free_logged_extents(log, log_transid);
2890 mutex_unlock(&log_root_tree->log_mutex);
2892 goto out_wake_log_root;
2895 ret = btrfs_write_marked_extents(log_root_tree,
2896 &log_root_tree->dirty_log_pages,
2897 EXTENT_DIRTY | EXTENT_NEW);
2898 blk_finish_plug(&plug);
2900 btrfs_set_log_full_commit(root->fs_info, trans);
2901 btrfs_abort_transaction(trans, root, ret);
2902 btrfs_free_logged_extents(log, log_transid);
2903 mutex_unlock(&log_root_tree->log_mutex);
2904 goto out_wake_log_root;
2906 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2908 ret = btrfs_wait_marked_extents(log_root_tree,
2909 &log_root_tree->dirty_log_pages,
2910 EXTENT_NEW | EXTENT_DIRTY);
2912 btrfs_set_log_full_commit(root->fs_info, trans);
2913 btrfs_free_logged_extents(log, log_transid);
2914 mutex_unlock(&log_root_tree->log_mutex);
2915 goto out_wake_log_root;
2917 btrfs_wait_logged_extents(trans, log, log_transid);
2919 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2920 log_root_tree->node->start);
2921 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2922 btrfs_header_level(log_root_tree->node));
2924 log_root_tree->log_transid++;
2925 mutex_unlock(&log_root_tree->log_mutex);
2928 * nobody else is going to jump in and write the the ctree
2929 * super here because the log_commit atomic below is protecting
2930 * us. We must be called with a transaction handle pinning
2931 * the running transaction open, so a full commit can't hop
2932 * in and cause problems either.
2934 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2936 btrfs_set_log_full_commit(root->fs_info, trans);
2937 btrfs_abort_transaction(trans, root, ret);
2938 goto out_wake_log_root;
2941 mutex_lock(&root->log_mutex);
2942 if (root->last_log_commit < log_transid)
2943 root->last_log_commit = log_transid;
2944 mutex_unlock(&root->log_mutex);
2948 * We needn't get log_mutex here because we are sure all
2949 * the other tasks are blocked.
2951 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2953 mutex_lock(&log_root_tree->log_mutex);
2954 log_root_tree->log_transid_committed++;
2955 atomic_set(&log_root_tree->log_commit[index2], 0);
2956 mutex_unlock(&log_root_tree->log_mutex);
2959 * The barrier before waitqueue_active is implied by mutex_unlock
2961 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2962 wake_up(&log_root_tree->log_commit_wait[index2]);
2965 btrfs_remove_all_log_ctxs(root, index1, ret);
2967 mutex_lock(&root->log_mutex);
2968 root->log_transid_committed++;
2969 atomic_set(&root->log_commit[index1], 0);
2970 mutex_unlock(&root->log_mutex);
2973 * The barrier before waitqueue_active is implied by mutex_unlock
2975 if (waitqueue_active(&root->log_commit_wait[index1]))
2976 wake_up(&root->log_commit_wait[index1]);
2980 static void free_log_tree(struct btrfs_trans_handle *trans,
2981 struct btrfs_root *log)
2986 struct walk_control wc = {
2988 .process_func = process_one_buffer
2991 ret = walk_log_tree(trans, log, &wc);
2992 /* I don't think this can happen but just in case */
2994 btrfs_abort_transaction(trans, log, ret);
2997 ret = find_first_extent_bit(&log->dirty_log_pages,
2998 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3003 clear_extent_bits(&log->dirty_log_pages, start, end,
3004 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
3008 * We may have short-circuited the log tree with the full commit logic
3009 * and left ordered extents on our list, so clear these out to keep us
3010 * from leaking inodes and memory.
3012 btrfs_free_logged_extents(log, 0);
3013 btrfs_free_logged_extents(log, 1);
3015 free_extent_buffer(log->node);
3020 * free all the extents used by the tree log. This should be called
3021 * at commit time of the full transaction
3023 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3025 if (root->log_root) {
3026 free_log_tree(trans, root->log_root);
3027 root->log_root = NULL;
3032 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3033 struct btrfs_fs_info *fs_info)
3035 if (fs_info->log_root_tree) {
3036 free_log_tree(trans, fs_info->log_root_tree);
3037 fs_info->log_root_tree = NULL;
3043 * If both a file and directory are logged, and unlinks or renames are
3044 * mixed in, we have a few interesting corners:
3046 * create file X in dir Y
3047 * link file X to X.link in dir Y
3049 * unlink file X but leave X.link
3052 * After a crash we would expect only X.link to exist. But file X
3053 * didn't get fsync'd again so the log has back refs for X and X.link.
3055 * We solve this by removing directory entries and inode backrefs from the
3056 * log when a file that was logged in the current transaction is
3057 * unlinked. Any later fsync will include the updated log entries, and
3058 * we'll be able to reconstruct the proper directory items from backrefs.
3060 * This optimizations allows us to avoid relogging the entire inode
3061 * or the entire directory.
3063 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3064 struct btrfs_root *root,
3065 const char *name, int name_len,
3066 struct inode *dir, u64 index)
3068 struct btrfs_root *log;
3069 struct btrfs_dir_item *di;
3070 struct btrfs_path *path;
3074 u64 dir_ino = btrfs_ino(dir);
3076 if (BTRFS_I(dir)->logged_trans < trans->transid)
3079 ret = join_running_log_trans(root);
3083 mutex_lock(&BTRFS_I(dir)->log_mutex);
3085 log = root->log_root;
3086 path = btrfs_alloc_path();
3092 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3093 name, name_len, -1);
3099 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3100 bytes_del += name_len;
3106 btrfs_release_path(path);
3107 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3108 index, name, name_len, -1);
3114 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3115 bytes_del += name_len;
3122 /* update the directory size in the log to reflect the names
3126 struct btrfs_key key;
3128 key.objectid = dir_ino;
3130 key.type = BTRFS_INODE_ITEM_KEY;
3131 btrfs_release_path(path);
3133 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3139 struct btrfs_inode_item *item;
3142 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3143 struct btrfs_inode_item);
3144 i_size = btrfs_inode_size(path->nodes[0], item);
3145 if (i_size > bytes_del)
3146 i_size -= bytes_del;
3149 btrfs_set_inode_size(path->nodes[0], item, i_size);
3150 btrfs_mark_buffer_dirty(path->nodes[0]);
3153 btrfs_release_path(path);
3156 btrfs_free_path(path);
3158 mutex_unlock(&BTRFS_I(dir)->log_mutex);
3159 if (ret == -ENOSPC) {
3160 btrfs_set_log_full_commit(root->fs_info, trans);
3163 btrfs_abort_transaction(trans, root, ret);
3165 btrfs_end_log_trans(root);
3170 /* see comments for btrfs_del_dir_entries_in_log */
3171 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3172 struct btrfs_root *root,
3173 const char *name, int name_len,
3174 struct inode *inode, u64 dirid)
3176 struct btrfs_root *log;
3180 if (BTRFS_I(inode)->logged_trans < trans->transid)
3183 ret = join_running_log_trans(root);
3186 log = root->log_root;
3187 mutex_lock(&BTRFS_I(inode)->log_mutex);
3189 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3191 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3192 if (ret == -ENOSPC) {
3193 btrfs_set_log_full_commit(root->fs_info, trans);
3195 } else if (ret < 0 && ret != -ENOENT)
3196 btrfs_abort_transaction(trans, root, ret);
3197 btrfs_end_log_trans(root);
3203 * creates a range item in the log for 'dirid'. first_offset and
3204 * last_offset tell us which parts of the key space the log should
3205 * be considered authoritative for.
3207 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3208 struct btrfs_root *log,
3209 struct btrfs_path *path,
3210 int key_type, u64 dirid,
3211 u64 first_offset, u64 last_offset)
3214 struct btrfs_key key;
3215 struct btrfs_dir_log_item *item;
3217 key.objectid = dirid;
3218 key.offset = first_offset;
3219 if (key_type == BTRFS_DIR_ITEM_KEY)
3220 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3222 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3223 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3227 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3228 struct btrfs_dir_log_item);
3229 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3230 btrfs_mark_buffer_dirty(path->nodes[0]);
3231 btrfs_release_path(path);
3236 * log all the items included in the current transaction for a given
3237 * directory. This also creates the range items in the log tree required
3238 * to replay anything deleted before the fsync
3240 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root, struct inode *inode,
3242 struct btrfs_path *path,
3243 struct btrfs_path *dst_path, int key_type,
3244 struct btrfs_log_ctx *ctx,
3245 u64 min_offset, u64 *last_offset_ret)
3247 struct btrfs_key min_key;
3248 struct btrfs_root *log = root->log_root;
3249 struct extent_buffer *src;
3254 u64 first_offset = min_offset;
3255 u64 last_offset = (u64)-1;
3256 u64 ino = btrfs_ino(inode);
3258 log = root->log_root;
3260 min_key.objectid = ino;
3261 min_key.type = key_type;
3262 min_key.offset = min_offset;
3264 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3267 * we didn't find anything from this transaction, see if there
3268 * is anything at all
3270 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3271 min_key.objectid = ino;
3272 min_key.type = key_type;
3273 min_key.offset = (u64)-1;
3274 btrfs_release_path(path);
3275 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3277 btrfs_release_path(path);
3280 ret = btrfs_previous_item(root, path, ino, key_type);
3282 /* if ret == 0 there are items for this type,
3283 * create a range to tell us the last key of this type.
3284 * otherwise, there are no items in this directory after
3285 * *min_offset, and we create a range to indicate that.
3288 struct btrfs_key tmp;
3289 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3291 if (key_type == tmp.type)
3292 first_offset = max(min_offset, tmp.offset) + 1;
3297 /* go backward to find any previous key */
3298 ret = btrfs_previous_item(root, path, ino, key_type);
3300 struct btrfs_key tmp;
3301 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3302 if (key_type == tmp.type) {
3303 first_offset = tmp.offset;
3304 ret = overwrite_item(trans, log, dst_path,
3305 path->nodes[0], path->slots[0],
3313 btrfs_release_path(path);
3315 /* find the first key from this transaction again */
3316 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3317 if (WARN_ON(ret != 0))
3321 * we have a block from this transaction, log every item in it
3322 * from our directory
3325 struct btrfs_key tmp;
3326 src = path->nodes[0];
3327 nritems = btrfs_header_nritems(src);
3328 for (i = path->slots[0]; i < nritems; i++) {
3329 struct btrfs_dir_item *di;
3331 btrfs_item_key_to_cpu(src, &min_key, i);
3333 if (min_key.objectid != ino || min_key.type != key_type)
3335 ret = overwrite_item(trans, log, dst_path, src, i,
3343 * We must make sure that when we log a directory entry,
3344 * the corresponding inode, after log replay, has a
3345 * matching link count. For example:
3351 * xfs_io -c "fsync" mydir
3353 * <mount fs and log replay>
3355 * Would result in a fsync log that when replayed, our
3356 * file inode would have a link count of 1, but we get
3357 * two directory entries pointing to the same inode.
3358 * After removing one of the names, it would not be
3359 * possible to remove the other name, which resulted
3360 * always in stale file handle errors, and would not
3361 * be possible to rmdir the parent directory, since
3362 * its i_size could never decrement to the value
3363 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3365 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3366 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3368 (btrfs_dir_transid(src, di) == trans->transid ||
3369 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3370 tmp.type != BTRFS_ROOT_ITEM_KEY)
3371 ctx->log_new_dentries = true;
3373 path->slots[0] = nritems;
3376 * look ahead to the next item and see if it is also
3377 * from this directory and from this transaction
3379 ret = btrfs_next_leaf(root, path);
3381 last_offset = (u64)-1;
3384 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3385 if (tmp.objectid != ino || tmp.type != key_type) {
3386 last_offset = (u64)-1;
3389 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3390 ret = overwrite_item(trans, log, dst_path,
3391 path->nodes[0], path->slots[0],
3396 last_offset = tmp.offset;
3401 btrfs_release_path(path);
3402 btrfs_release_path(dst_path);
3405 *last_offset_ret = last_offset;
3407 * insert the log range keys to indicate where the log
3410 ret = insert_dir_log_key(trans, log, path, key_type,
3411 ino, first_offset, last_offset);
3419 * logging directories is very similar to logging inodes, We find all the items
3420 * from the current transaction and write them to the log.
3422 * The recovery code scans the directory in the subvolume, and if it finds a
3423 * key in the range logged that is not present in the log tree, then it means
3424 * that dir entry was unlinked during the transaction.
3426 * In order for that scan to work, we must include one key smaller than
3427 * the smallest logged by this transaction and one key larger than the largest
3428 * key logged by this transaction.
3430 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3431 struct btrfs_root *root, struct inode *inode,
3432 struct btrfs_path *path,
3433 struct btrfs_path *dst_path,
3434 struct btrfs_log_ctx *ctx)
3439 int key_type = BTRFS_DIR_ITEM_KEY;
3445 ret = log_dir_items(trans, root, inode, path,
3446 dst_path, key_type, ctx, min_key,
3450 if (max_key == (u64)-1)
3452 min_key = max_key + 1;
3455 if (key_type == BTRFS_DIR_ITEM_KEY) {
3456 key_type = BTRFS_DIR_INDEX_KEY;
3463 * a helper function to drop items from the log before we relog an
3464 * inode. max_key_type indicates the highest item type to remove.
3465 * This cannot be run for file data extents because it does not
3466 * free the extents they point to.
3468 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3469 struct btrfs_root *log,
3470 struct btrfs_path *path,
3471 u64 objectid, int max_key_type)
3474 struct btrfs_key key;
3475 struct btrfs_key found_key;
3478 key.objectid = objectid;
3479 key.type = max_key_type;
3480 key.offset = (u64)-1;
3483 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3484 BUG_ON(ret == 0); /* Logic error */
3488 if (path->slots[0] == 0)
3492 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3495 if (found_key.objectid != objectid)
3498 found_key.offset = 0;
3500 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3503 ret = btrfs_del_items(trans, log, path, start_slot,
3504 path->slots[0] - start_slot + 1);
3506 * If start slot isn't 0 then we don't need to re-search, we've
3507 * found the last guy with the objectid in this tree.
3509 if (ret || start_slot != 0)
3511 btrfs_release_path(path);
3513 btrfs_release_path(path);
3519 static void fill_inode_item(struct btrfs_trans_handle *trans,
3520 struct extent_buffer *leaf,
3521 struct btrfs_inode_item *item,
3522 struct inode *inode, int log_inode_only,
3525 struct btrfs_map_token token;
3527 btrfs_init_map_token(&token);
3529 if (log_inode_only) {
3530 /* set the generation to zero so the recover code
3531 * can tell the difference between an logging
3532 * just to say 'this inode exists' and a logging
3533 * to say 'update this inode with these values'
3535 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3536 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3538 btrfs_set_token_inode_generation(leaf, item,
3539 BTRFS_I(inode)->generation,
3541 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3544 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3545 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3546 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3547 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3549 btrfs_set_token_timespec_sec(leaf, &item->atime,
3550 inode->i_atime.tv_sec, &token);
3551 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3552 inode->i_atime.tv_nsec, &token);
3554 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3555 inode->i_mtime.tv_sec, &token);
3556 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3557 inode->i_mtime.tv_nsec, &token);
3559 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3560 inode->i_ctime.tv_sec, &token);
3561 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3562 inode->i_ctime.tv_nsec, &token);
3564 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3567 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3568 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3569 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3570 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3571 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3574 static int log_inode_item(struct btrfs_trans_handle *trans,
3575 struct btrfs_root *log, struct btrfs_path *path,
3576 struct inode *inode)
3578 struct btrfs_inode_item *inode_item;
3581 ret = btrfs_insert_empty_item(trans, log, path,
3582 &BTRFS_I(inode)->location,
3583 sizeof(*inode_item));
3584 if (ret && ret != -EEXIST)
3586 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3587 struct btrfs_inode_item);
3588 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3589 btrfs_release_path(path);
3593 static noinline int copy_items(struct btrfs_trans_handle *trans,
3594 struct inode *inode,
3595 struct btrfs_path *dst_path,
3596 struct btrfs_path *src_path, u64 *last_extent,
3597 int start_slot, int nr, int inode_only,
3600 unsigned long src_offset;
3601 unsigned long dst_offset;
3602 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3603 struct btrfs_file_extent_item *extent;
3604 struct btrfs_inode_item *inode_item;
3605 struct extent_buffer *src = src_path->nodes[0];
3606 struct btrfs_key first_key, last_key, key;
3608 struct btrfs_key *ins_keys;
3612 struct list_head ordered_sums;
3613 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3614 bool has_extents = false;
3615 bool need_find_last_extent = true;
3618 INIT_LIST_HEAD(&ordered_sums);
3620 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3621 nr * sizeof(u32), GFP_NOFS);
3625 first_key.objectid = (u64)-1;
3627 ins_sizes = (u32 *)ins_data;
3628 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3630 for (i = 0; i < nr; i++) {
3631 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3632 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3634 ret = btrfs_insert_empty_items(trans, log, dst_path,
3635 ins_keys, ins_sizes, nr);
3641 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3642 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3643 dst_path->slots[0]);
3645 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3647 if ((i == (nr - 1)))
3648 last_key = ins_keys[i];
3650 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3651 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3653 struct btrfs_inode_item);
3654 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3655 inode, inode_only == LOG_INODE_EXISTS,
3658 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3659 src_offset, ins_sizes[i]);
3663 * We set need_find_last_extent here in case we know we were
3664 * processing other items and then walk into the first extent in
3665 * the inode. If we don't hit an extent then nothing changes,
3666 * we'll do the last search the next time around.
3668 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3670 if (first_key.objectid == (u64)-1)
3671 first_key = ins_keys[i];
3673 need_find_last_extent = false;
3676 /* take a reference on file data extents so that truncates
3677 * or deletes of this inode don't have to relog the inode
3680 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3683 extent = btrfs_item_ptr(src, start_slot + i,
3684 struct btrfs_file_extent_item);
3686 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3689 found_type = btrfs_file_extent_type(src, extent);
3690 if (found_type == BTRFS_FILE_EXTENT_REG) {
3692 ds = btrfs_file_extent_disk_bytenr(src,
3694 /* ds == 0 is a hole */
3698 dl = btrfs_file_extent_disk_num_bytes(src,
3700 cs = btrfs_file_extent_offset(src, extent);
3701 cl = btrfs_file_extent_num_bytes(src,
3703 if (btrfs_file_extent_compression(src,
3709 ret = btrfs_lookup_csums_range(
3710 log->fs_info->csum_root,
3711 ds + cs, ds + cs + cl - 1,
3714 btrfs_release_path(dst_path);
3722 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3723 btrfs_release_path(dst_path);
3727 * we have to do this after the loop above to avoid changing the
3728 * log tree while trying to change the log tree.
3731 while (!list_empty(&ordered_sums)) {
3732 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3733 struct btrfs_ordered_sum,
3736 ret = btrfs_csum_file_blocks(trans, log, sums);
3737 list_del(&sums->list);
3744 if (need_find_last_extent && *last_extent == first_key.offset) {
3746 * We don't have any leafs between our current one and the one
3747 * we processed before that can have file extent items for our
3748 * inode (and have a generation number smaller than our current
3751 need_find_last_extent = false;
3755 * Because we use btrfs_search_forward we could skip leaves that were
3756 * not modified and then assume *last_extent is valid when it really
3757 * isn't. So back up to the previous leaf and read the end of the last
3758 * extent before we go and fill in holes.
3760 if (need_find_last_extent) {
3763 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3768 if (src_path->slots[0])
3769 src_path->slots[0]--;
3770 src = src_path->nodes[0];
3771 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3772 if (key.objectid != btrfs_ino(inode) ||
3773 key.type != BTRFS_EXTENT_DATA_KEY)
3775 extent = btrfs_item_ptr(src, src_path->slots[0],
3776 struct btrfs_file_extent_item);
3777 if (btrfs_file_extent_type(src, extent) ==
3778 BTRFS_FILE_EXTENT_INLINE) {
3779 len = btrfs_file_extent_inline_len(src,
3782 *last_extent = ALIGN(key.offset + len,
3785 len = btrfs_file_extent_num_bytes(src, extent);
3786 *last_extent = key.offset + len;
3790 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3791 * things could have happened
3793 * 1) A merge could have happened, so we could currently be on a leaf
3794 * that holds what we were copying in the first place.
3795 * 2) A split could have happened, and now not all of the items we want
3796 * are on the same leaf.
3798 * So we need to adjust how we search for holes, we need to drop the
3799 * path and re-search for the first extent key we found, and then walk
3800 * forward until we hit the last one we copied.
3802 if (need_find_last_extent) {
3803 /* btrfs_prev_leaf could return 1 without releasing the path */
3804 btrfs_release_path(src_path);
3805 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3810 src = src_path->nodes[0];
3811 i = src_path->slots[0];
3817 * Ok so here we need to go through and fill in any holes we may have
3818 * to make sure that holes are punched for those areas in case they had
3819 * extents previously.
3825 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3826 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3830 src = src_path->nodes[0];
3834 btrfs_item_key_to_cpu(src, &key, i);
3835 if (!btrfs_comp_cpu_keys(&key, &last_key))
3837 if (key.objectid != btrfs_ino(inode) ||
3838 key.type != BTRFS_EXTENT_DATA_KEY) {
3842 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3843 if (btrfs_file_extent_type(src, extent) ==
3844 BTRFS_FILE_EXTENT_INLINE) {
3845 len = btrfs_file_extent_inline_len(src, i, extent);
3846 extent_end = ALIGN(key.offset + len, log->sectorsize);
3848 len = btrfs_file_extent_num_bytes(src, extent);
3849 extent_end = key.offset + len;
3853 if (*last_extent == key.offset) {
3854 *last_extent = extent_end;
3857 offset = *last_extent;
3858 len = key.offset - *last_extent;
3859 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3860 offset, 0, 0, len, 0, len, 0,
3864 *last_extent = extent_end;
3867 * Need to let the callers know we dropped the path so they should
3870 if (!ret && need_find_last_extent)
3875 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3877 struct extent_map *em1, *em2;
3879 em1 = list_entry(a, struct extent_map, list);
3880 em2 = list_entry(b, struct extent_map, list);
3882 if (em1->start < em2->start)
3884 else if (em1->start > em2->start)
3889 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3890 struct inode *inode,
3891 struct btrfs_root *root,
3892 const struct extent_map *em,
3893 const struct list_head *logged_list,
3894 bool *ordered_io_error)
3896 struct btrfs_ordered_extent *ordered;
3897 struct btrfs_root *log = root->log_root;
3898 u64 mod_start = em->mod_start;
3899 u64 mod_len = em->mod_len;
3900 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3903 LIST_HEAD(ordered_sums);
3906 *ordered_io_error = false;
3908 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3909 em->block_start == EXTENT_MAP_HOLE)
3913 * Wait far any ordered extent that covers our extent map. If it
3914 * finishes without an error, first check and see if our csums are on
3915 * our outstanding ordered extents.
3917 list_for_each_entry(ordered, logged_list, log_list) {
3918 struct btrfs_ordered_sum *sum;
3923 if (ordered->file_offset + ordered->len <= mod_start ||
3924 mod_start + mod_len <= ordered->file_offset)
3927 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3928 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3929 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3930 const u64 start = ordered->file_offset;
3931 const u64 end = ordered->file_offset + ordered->len - 1;
3933 WARN_ON(ordered->inode != inode);
3934 filemap_fdatawrite_range(inode->i_mapping, start, end);
3937 wait_event(ordered->wait,
3938 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3939 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3941 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3943 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3944 * i_mapping flags, so that the next fsync won't get
3945 * an outdated io error too.
3947 btrfs_inode_check_errors(inode);
3948 *ordered_io_error = true;
3952 * We are going to copy all the csums on this ordered extent, so
3953 * go ahead and adjust mod_start and mod_len in case this
3954 * ordered extent has already been logged.
3956 if (ordered->file_offset > mod_start) {
3957 if (ordered->file_offset + ordered->len >=
3958 mod_start + mod_len)
3959 mod_len = ordered->file_offset - mod_start;
3961 * If we have this case
3963 * |--------- logged extent ---------|
3964 * |----- ordered extent ----|
3966 * Just don't mess with mod_start and mod_len, we'll
3967 * just end up logging more csums than we need and it
3971 if (ordered->file_offset + ordered->len <
3972 mod_start + mod_len) {
3973 mod_len = (mod_start + mod_len) -
3974 (ordered->file_offset + ordered->len);
3975 mod_start = ordered->file_offset +
3986 * To keep us from looping for the above case of an ordered
3987 * extent that falls inside of the logged extent.
3989 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3993 list_for_each_entry(sum, &ordered->list, list) {
3994 ret = btrfs_csum_file_blocks(trans, log, sum);
4000 if (*ordered_io_error || !mod_len || ret || skip_csum)
4003 if (em->compress_type) {
4005 csum_len = max(em->block_len, em->orig_block_len);
4007 csum_offset = mod_start - em->start;
4011 /* block start is already adjusted for the file extent offset. */
4012 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4013 em->block_start + csum_offset,
4014 em->block_start + csum_offset +
4015 csum_len - 1, &ordered_sums, 0);
4019 while (!list_empty(&ordered_sums)) {
4020 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4021 struct btrfs_ordered_sum,
4024 ret = btrfs_csum_file_blocks(trans, log, sums);
4025 list_del(&sums->list);
4032 static int log_one_extent(struct btrfs_trans_handle *trans,
4033 struct inode *inode, struct btrfs_root *root,
4034 const struct extent_map *em,
4035 struct btrfs_path *path,
4036 const struct list_head *logged_list,
4037 struct btrfs_log_ctx *ctx)
4039 struct btrfs_root *log = root->log_root;
4040 struct btrfs_file_extent_item *fi;
4041 struct extent_buffer *leaf;
4042 struct btrfs_map_token token;
4043 struct btrfs_key key;
4044 u64 extent_offset = em->start - em->orig_start;
4047 int extent_inserted = 0;
4048 bool ordered_io_err = false;
4050 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4055 if (ordered_io_err) {
4060 btrfs_init_map_token(&token);
4062 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4063 em->start + em->len, NULL, 0, 1,
4064 sizeof(*fi), &extent_inserted);
4068 if (!extent_inserted) {
4069 key.objectid = btrfs_ino(inode);
4070 key.type = BTRFS_EXTENT_DATA_KEY;
4071 key.offset = em->start;
4073 ret = btrfs_insert_empty_item(trans, log, path, &key,
4078 leaf = path->nodes[0];
4079 fi = btrfs_item_ptr(leaf, path->slots[0],
4080 struct btrfs_file_extent_item);
4082 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4084 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4085 btrfs_set_token_file_extent_type(leaf, fi,
4086 BTRFS_FILE_EXTENT_PREALLOC,
4089 btrfs_set_token_file_extent_type(leaf, fi,
4090 BTRFS_FILE_EXTENT_REG,
4093 block_len = max(em->block_len, em->orig_block_len);
4094 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4095 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4098 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4100 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4101 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4103 extent_offset, &token);
4104 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4107 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4108 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4112 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4113 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4114 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4115 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4117 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4118 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4119 btrfs_mark_buffer_dirty(leaf);
4121 btrfs_release_path(path);
4126 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4127 struct btrfs_root *root,
4128 struct inode *inode,
4129 struct btrfs_path *path,
4130 struct list_head *logged_list,
4131 struct btrfs_log_ctx *ctx)
4133 struct extent_map *em, *n;
4134 struct list_head extents;
4135 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4140 INIT_LIST_HEAD(&extents);
4142 write_lock(&tree->lock);
4143 test_gen = root->fs_info->last_trans_committed;
4145 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4146 list_del_init(&em->list);
4149 * Just an arbitrary number, this can be really CPU intensive
4150 * once we start getting a lot of extents, and really once we
4151 * have a bunch of extents we just want to commit since it will
4154 if (++num > 32768) {
4155 list_del_init(&tree->modified_extents);
4160 if (em->generation <= test_gen)
4162 /* Need a ref to keep it from getting evicted from cache */
4163 atomic_inc(&em->refs);
4164 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4165 list_add_tail(&em->list, &extents);
4169 list_sort(NULL, &extents, extent_cmp);
4172 while (!list_empty(&extents)) {
4173 em = list_entry(extents.next, struct extent_map, list);
4175 list_del_init(&em->list);
4178 * If we had an error we just need to delete everybody from our
4182 clear_em_logging(tree, em);
4183 free_extent_map(em);
4187 write_unlock(&tree->lock);
4189 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4191 write_lock(&tree->lock);
4192 clear_em_logging(tree, em);
4193 free_extent_map(em);
4195 WARN_ON(!list_empty(&extents));
4196 write_unlock(&tree->lock);
4198 btrfs_release_path(path);
4202 static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4203 struct btrfs_path *path, u64 *size_ret)
4205 struct btrfs_key key;
4208 key.objectid = btrfs_ino(inode);
4209 key.type = BTRFS_INODE_ITEM_KEY;
4212 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4215 } else if (ret > 0) {
4218 struct btrfs_inode_item *item;
4220 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4221 struct btrfs_inode_item);
4222 *size_ret = btrfs_inode_size(path->nodes[0], item);
4225 btrfs_release_path(path);
4230 * At the moment we always log all xattrs. This is to figure out at log replay
4231 * time which xattrs must have their deletion replayed. If a xattr is missing
4232 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4233 * because if a xattr is deleted, the inode is fsynced and a power failure
4234 * happens, causing the log to be replayed the next time the fs is mounted,
4235 * we want the xattr to not exist anymore (same behaviour as other filesystems
4236 * with a journal, ext3/4, xfs, f2fs, etc).
4238 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4239 struct btrfs_root *root,
4240 struct inode *inode,
4241 struct btrfs_path *path,
4242 struct btrfs_path *dst_path)
4245 struct btrfs_key key;
4246 const u64 ino = btrfs_ino(inode);
4251 key.type = BTRFS_XATTR_ITEM_KEY;
4254 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4259 int slot = path->slots[0];
4260 struct extent_buffer *leaf = path->nodes[0];
4261 int nritems = btrfs_header_nritems(leaf);
4263 if (slot >= nritems) {
4265 u64 last_extent = 0;
4267 ret = copy_items(trans, inode, dst_path, path,
4268 &last_extent, start_slot,
4270 /* can't be 1, extent items aren't processed */
4276 ret = btrfs_next_leaf(root, path);
4284 btrfs_item_key_to_cpu(leaf, &key, slot);
4285 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4295 u64 last_extent = 0;
4297 ret = copy_items(trans, inode, dst_path, path,
4298 &last_extent, start_slot,
4300 /* can't be 1, extent items aren't processed */
4310 * If the no holes feature is enabled we need to make sure any hole between the
4311 * last extent and the i_size of our inode is explicitly marked in the log. This
4312 * is to make sure that doing something like:
4314 * 1) create file with 128Kb of data
4315 * 2) truncate file to 64Kb
4316 * 3) truncate file to 256Kb
4318 * 5) <crash/power failure>
4319 * 6) mount fs and trigger log replay
4321 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4322 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4323 * file correspond to a hole. The presence of explicit holes in a log tree is
4324 * what guarantees that log replay will remove/adjust file extent items in the
4327 * Here we do not need to care about holes between extents, that is already done
4328 * by copy_items(). We also only need to do this in the full sync path, where we
4329 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4330 * lookup the list of modified extent maps and if any represents a hole, we
4331 * insert a corresponding extent representing a hole in the log tree.
4333 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4334 struct btrfs_root *root,
4335 struct inode *inode,
4336 struct btrfs_path *path)
4339 struct btrfs_key key;
4342 struct extent_buffer *leaf;
4343 struct btrfs_root *log = root->log_root;
4344 const u64 ino = btrfs_ino(inode);
4345 const u64 i_size = i_size_read(inode);
4347 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4351 key.type = BTRFS_EXTENT_DATA_KEY;
4352 key.offset = (u64)-1;
4354 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4359 ASSERT(path->slots[0] > 0);
4361 leaf = path->nodes[0];
4362 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4364 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4365 /* inode does not have any extents */
4369 struct btrfs_file_extent_item *extent;
4373 * If there's an extent beyond i_size, an explicit hole was
4374 * already inserted by copy_items().
4376 if (key.offset >= i_size)
4379 extent = btrfs_item_ptr(leaf, path->slots[0],
4380 struct btrfs_file_extent_item);
4382 if (btrfs_file_extent_type(leaf, extent) ==
4383 BTRFS_FILE_EXTENT_INLINE) {
4384 len = btrfs_file_extent_inline_len(leaf,
4387 ASSERT(len == i_size);
4391 len = btrfs_file_extent_num_bytes(leaf, extent);
4392 /* Last extent goes beyond i_size, no need to log a hole. */
4393 if (key.offset + len > i_size)
4395 hole_start = key.offset + len;
4396 hole_size = i_size - hole_start;
4398 btrfs_release_path(path);
4400 /* Last extent ends at i_size. */
4404 hole_size = ALIGN(hole_size, root->sectorsize);
4405 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4406 hole_size, 0, hole_size, 0, 0, 0);
4411 * When we are logging a new inode X, check if it doesn't have a reference that
4412 * matches the reference from some other inode Y created in a past transaction
4413 * and that was renamed in the current transaction. If we don't do this, then at
4414 * log replay time we can lose inode Y (and all its files if it's a directory):
4417 * echo "hello world" > /mnt/x/foobar
4420 * mkdir /mnt/x # or touch /mnt/x
4421 * xfs_io -c fsync /mnt/x
4423 * mount fs, trigger log replay
4425 * After the log replay procedure, we would lose the first directory and all its
4426 * files (file foobar).
4427 * For the case where inode Y is not a directory we simply end up losing it:
4429 * echo "123" > /mnt/foo
4431 * mv /mnt/foo /mnt/bar
4432 * echo "abc" > /mnt/foo
4433 * xfs_io -c fsync /mnt/foo
4436 * We also need this for cases where a snapshot entry is replaced by some other
4437 * entry (file or directory) otherwise we end up with an unreplayable log due to
4438 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4439 * if it were a regular entry:
4442 * btrfs subvolume snapshot /mnt /mnt/x/snap
4443 * btrfs subvolume delete /mnt/x/snap
4446 * fsync /mnt/x or fsync some new file inside it
4449 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4450 * the same transaction.
4452 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4454 const struct btrfs_key *key,
4455 struct inode *inode)
4458 struct btrfs_path *search_path;
4461 u32 item_size = btrfs_item_size_nr(eb, slot);
4463 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4465 search_path = btrfs_alloc_path();
4468 search_path->search_commit_root = 1;
4469 search_path->skip_locking = 1;
4471 while (cur_offset < item_size) {
4475 unsigned long name_ptr;
4476 struct btrfs_dir_item *di;
4478 if (key->type == BTRFS_INODE_REF_KEY) {
4479 struct btrfs_inode_ref *iref;
4481 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4482 parent = key->offset;
4483 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4484 name_ptr = (unsigned long)(iref + 1);
4485 this_len = sizeof(*iref) + this_name_len;
4487 struct btrfs_inode_extref *extref;
4489 extref = (struct btrfs_inode_extref *)(ptr +
4491 parent = btrfs_inode_extref_parent(eb, extref);
4492 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4493 name_ptr = (unsigned long)&extref->name;
4494 this_len = sizeof(*extref) + this_name_len;
4497 if (this_name_len > name_len) {
4500 new_name = krealloc(name, this_name_len, GFP_NOFS);
4505 name_len = this_name_len;
4509 read_extent_buffer(eb, name, name_ptr, this_name_len);
4510 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4511 search_path, parent,
4512 name, this_name_len, 0);
4513 if (di && !IS_ERR(di)) {
4516 } else if (IS_ERR(di)) {
4520 btrfs_release_path(search_path);
4522 cur_offset += this_len;
4526 btrfs_free_path(search_path);
4531 /* log a single inode in the tree log.
4532 * At least one parent directory for this inode must exist in the tree
4533 * or be logged already.
4535 * Any items from this inode changed by the current transaction are copied
4536 * to the log tree. An extra reference is taken on any extents in this
4537 * file, allowing us to avoid a whole pile of corner cases around logging
4538 * blocks that have been removed from the tree.
4540 * See LOG_INODE_ALL and related defines for a description of what inode_only
4543 * This handles both files and directories.
4545 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4546 struct btrfs_root *root, struct inode *inode,
4550 struct btrfs_log_ctx *ctx)
4552 struct btrfs_path *path;
4553 struct btrfs_path *dst_path;
4554 struct btrfs_key min_key;
4555 struct btrfs_key max_key;
4556 struct btrfs_root *log = root->log_root;
4557 struct extent_buffer *src = NULL;
4558 LIST_HEAD(logged_list);
4559 u64 last_extent = 0;
4563 int ins_start_slot = 0;
4565 bool fast_search = false;
4566 u64 ino = btrfs_ino(inode);
4567 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4568 u64 logged_isize = 0;
4569 bool need_log_inode_item = true;
4571 path = btrfs_alloc_path();
4574 dst_path = btrfs_alloc_path();
4576 btrfs_free_path(path);
4580 min_key.objectid = ino;
4581 min_key.type = BTRFS_INODE_ITEM_KEY;
4584 max_key.objectid = ino;
4587 /* today the code can only do partial logging of directories */
4588 if (S_ISDIR(inode->i_mode) ||
4589 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4590 &BTRFS_I(inode)->runtime_flags) &&
4591 inode_only == LOG_INODE_EXISTS))
4592 max_key.type = BTRFS_XATTR_ITEM_KEY;
4594 max_key.type = (u8)-1;
4595 max_key.offset = (u64)-1;
4598 * Only run delayed items if we are a dir or a new file.
4599 * Otherwise commit the delayed inode only, which is needed in
4600 * order for the log replay code to mark inodes for link count
4601 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4603 if (S_ISDIR(inode->i_mode) ||
4604 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
4605 ret = btrfs_commit_inode_delayed_items(trans, inode);
4607 ret = btrfs_commit_inode_delayed_inode(inode);
4610 btrfs_free_path(path);
4611 btrfs_free_path(dst_path);
4615 mutex_lock(&BTRFS_I(inode)->log_mutex);
4617 btrfs_get_logged_extents(inode, &logged_list, start, end);
4620 * a brute force approach to making sure we get the most uptodate
4621 * copies of everything.
4623 if (S_ISDIR(inode->i_mode)) {
4624 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4626 if (inode_only == LOG_INODE_EXISTS)
4627 max_key_type = BTRFS_XATTR_ITEM_KEY;
4628 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4630 if (inode_only == LOG_INODE_EXISTS) {
4632 * Make sure the new inode item we write to the log has
4633 * the same isize as the current one (if it exists).
4634 * This is necessary to prevent data loss after log
4635 * replay, and also to prevent doing a wrong expanding
4636 * truncate - for e.g. create file, write 4K into offset
4637 * 0, fsync, write 4K into offset 4096, add hard link,
4638 * fsync some other file (to sync log), power fail - if
4639 * we use the inode's current i_size, after log replay
4640 * we get a 8Kb file, with the last 4Kb extent as a hole
4641 * (zeroes), as if an expanding truncate happened,
4642 * instead of getting a file of 4Kb only.
4644 err = logged_inode_size(log, inode, path,
4649 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4650 &BTRFS_I(inode)->runtime_flags)) {
4651 if (inode_only == LOG_INODE_EXISTS) {
4652 max_key.type = BTRFS_XATTR_ITEM_KEY;
4653 ret = drop_objectid_items(trans, log, path, ino,
4656 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4657 &BTRFS_I(inode)->runtime_flags);
4658 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4659 &BTRFS_I(inode)->runtime_flags);
4661 ret = btrfs_truncate_inode_items(trans,
4667 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4668 &BTRFS_I(inode)->runtime_flags) ||
4669 inode_only == LOG_INODE_EXISTS) {
4670 if (inode_only == LOG_INODE_ALL)
4672 max_key.type = BTRFS_XATTR_ITEM_KEY;
4673 ret = drop_objectid_items(trans, log, path, ino,
4676 if (inode_only == LOG_INODE_ALL)
4689 ret = btrfs_search_forward(root, &min_key,
4690 path, trans->transid);
4694 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4695 if (min_key.objectid != ino)
4697 if (min_key.type > max_key.type)
4700 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4701 need_log_inode_item = false;
4703 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4704 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4705 BTRFS_I(inode)->generation == trans->transid) {
4706 ret = btrfs_check_ref_name_override(path->nodes[0],
4712 } else if (ret > 0) {
4714 btrfs_set_log_full_commit(root->fs_info, trans);
4719 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4720 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4723 ret = copy_items(trans, inode, dst_path, path,
4724 &last_extent, ins_start_slot,
4725 ins_nr, inode_only, logged_isize);
4732 btrfs_release_path(path);
4738 src = path->nodes[0];
4739 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4742 } else if (!ins_nr) {
4743 ins_start_slot = path->slots[0];
4748 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4749 ins_start_slot, ins_nr, inode_only,
4757 btrfs_release_path(path);
4761 ins_start_slot = path->slots[0];
4764 nritems = btrfs_header_nritems(path->nodes[0]);
4766 if (path->slots[0] < nritems) {
4767 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4772 ret = copy_items(trans, inode, dst_path, path,
4773 &last_extent, ins_start_slot,
4774 ins_nr, inode_only, logged_isize);
4782 btrfs_release_path(path);
4784 if (min_key.offset < (u64)-1) {
4786 } else if (min_key.type < max_key.type) {
4794 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4795 ins_start_slot, ins_nr, inode_only,
4805 btrfs_release_path(path);
4806 btrfs_release_path(dst_path);
4807 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4810 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4811 btrfs_release_path(path);
4812 btrfs_release_path(dst_path);
4813 err = btrfs_log_trailing_hole(trans, root, inode, path);
4818 btrfs_release_path(path);
4819 btrfs_release_path(dst_path);
4820 if (need_log_inode_item) {
4821 err = log_inode_item(trans, log, dst_path, inode);
4827 * Some ordered extents started by fsync might have completed
4828 * before we collected the ordered extents in logged_list, which
4829 * means they're gone, not in our logged_list nor in the inode's
4830 * ordered tree. We want the application/user space to know an
4831 * error happened while attempting to persist file data so that
4832 * it can take proper action. If such error happened, we leave
4833 * without writing to the log tree and the fsync must report the
4834 * file data write error and not commit the current transaction.
4836 err = btrfs_inode_check_errors(inode);
4841 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4847 } else if (inode_only == LOG_INODE_ALL) {
4848 struct extent_map *em, *n;
4850 write_lock(&em_tree->lock);
4852 * We can't just remove every em if we're called for a ranged
4853 * fsync - that is, one that doesn't cover the whole possible
4854 * file range (0 to LLONG_MAX). This is because we can have
4855 * em's that fall outside the range we're logging and therefore
4856 * their ordered operations haven't completed yet
4857 * (btrfs_finish_ordered_io() not invoked yet). This means we
4858 * didn't get their respective file extent item in the fs/subvol
4859 * tree yet, and need to let the next fast fsync (one which
4860 * consults the list of modified extent maps) find the em so
4861 * that it logs a matching file extent item and waits for the
4862 * respective ordered operation to complete (if it's still
4865 * Removing every em outside the range we're logging would make
4866 * the next fast fsync not log their matching file extent items,
4867 * therefore making us lose data after a log replay.
4869 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4871 const u64 mod_end = em->mod_start + em->mod_len - 1;
4873 if (em->mod_start >= start && mod_end <= end)
4874 list_del_init(&em->list);
4876 write_unlock(&em_tree->lock);
4879 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4880 ret = log_directory_changes(trans, root, inode, path, dst_path,
4888 spin_lock(&BTRFS_I(inode)->lock);
4889 BTRFS_I(inode)->logged_trans = trans->transid;
4890 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4891 spin_unlock(&BTRFS_I(inode)->lock);
4894 btrfs_put_logged_extents(&logged_list);
4896 btrfs_submit_logged_extents(&logged_list, log);
4897 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4899 btrfs_free_path(path);
4900 btrfs_free_path(dst_path);
4905 * follow the dentry parent pointers up the chain and see if any
4906 * of the directories in it require a full commit before they can
4907 * be logged. Returns zero if nothing special needs to be done or 1 if
4908 * a full commit is required.
4910 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4911 struct inode *inode,
4912 struct dentry *parent,
4913 struct super_block *sb,
4917 struct btrfs_root *root;
4918 struct dentry *old_parent = NULL;
4919 struct inode *orig_inode = inode;
4922 * for regular files, if its inode is already on disk, we don't
4923 * have to worry about the parents at all. This is because
4924 * we can use the last_unlink_trans field to record renames
4925 * and other fun in this file.
4927 if (S_ISREG(inode->i_mode) &&
4928 BTRFS_I(inode)->generation <= last_committed &&
4929 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4932 if (!S_ISDIR(inode->i_mode)) {
4933 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
4935 inode = d_inode(parent);
4940 * If we are logging a directory then we start with our inode,
4941 * not our parents inode, so we need to skipp setting the
4942 * logged_trans so that further down in the log code we don't
4943 * think this inode has already been logged.
4945 if (inode != orig_inode)
4946 BTRFS_I(inode)->logged_trans = trans->transid;
4949 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4950 root = BTRFS_I(inode)->root;
4953 * make sure any commits to the log are forced
4954 * to be full commits
4956 btrfs_set_log_full_commit(root->fs_info, trans);
4961 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
4964 if (IS_ROOT(parent))
4967 parent = dget_parent(parent);
4969 old_parent = parent;
4970 inode = d_inode(parent);
4978 struct btrfs_dir_list {
4980 struct list_head list;
4984 * Log the inodes of the new dentries of a directory. See log_dir_items() for
4985 * details about the why it is needed.
4986 * This is a recursive operation - if an existing dentry corresponds to a
4987 * directory, that directory's new entries are logged too (same behaviour as
4988 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
4989 * the dentries point to we do not lock their i_mutex, otherwise lockdep
4990 * complains about the following circular lock dependency / possible deadlock:
4994 * lock(&type->i_mutex_dir_key#3/2);
4995 * lock(sb_internal#2);
4996 * lock(&type->i_mutex_dir_key#3/2);
4997 * lock(&sb->s_type->i_mutex_key#14);
4999 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5000 * sb_start_intwrite() in btrfs_start_transaction().
5001 * Not locking i_mutex of the inodes is still safe because:
5003 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5004 * that while logging the inode new references (names) are added or removed
5005 * from the inode, leaving the logged inode item with a link count that does
5006 * not match the number of logged inode reference items. This is fine because
5007 * at log replay time we compute the real number of links and correct the
5008 * link count in the inode item (see replay_one_buffer() and
5009 * link_to_fixup_dir());
5011 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5012 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5013 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5014 * has a size that doesn't match the sum of the lengths of all the logged
5015 * names. This does not result in a problem because if a dir_item key is
5016 * logged but its matching dir_index key is not logged, at log replay time we
5017 * don't use it to replay the respective name (see replay_one_name()). On the
5018 * other hand if only the dir_index key ends up being logged, the respective
5019 * name is added to the fs/subvol tree with both the dir_item and dir_index
5020 * keys created (see replay_one_name()).
5021 * The directory's inode item with a wrong i_size is not a problem as well,
5022 * since we don't use it at log replay time to set the i_size in the inode
5023 * item of the fs/subvol tree (see overwrite_item()).
5025 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5026 struct btrfs_root *root,
5027 struct inode *start_inode,
5028 struct btrfs_log_ctx *ctx)
5030 struct btrfs_root *log = root->log_root;
5031 struct btrfs_path *path;
5032 LIST_HEAD(dir_list);
5033 struct btrfs_dir_list *dir_elem;
5036 path = btrfs_alloc_path();
5040 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5042 btrfs_free_path(path);
5045 dir_elem->ino = btrfs_ino(start_inode);
5046 list_add_tail(&dir_elem->list, &dir_list);
5048 while (!list_empty(&dir_list)) {
5049 struct extent_buffer *leaf;
5050 struct btrfs_key min_key;
5054 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5057 goto next_dir_inode;
5059 min_key.objectid = dir_elem->ino;
5060 min_key.type = BTRFS_DIR_ITEM_KEY;
5063 btrfs_release_path(path);
5064 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5066 goto next_dir_inode;
5067 } else if (ret > 0) {
5069 goto next_dir_inode;
5073 leaf = path->nodes[0];
5074 nritems = btrfs_header_nritems(leaf);
5075 for (i = path->slots[0]; i < nritems; i++) {
5076 struct btrfs_dir_item *di;
5077 struct btrfs_key di_key;
5078 struct inode *di_inode;
5079 struct btrfs_dir_list *new_dir_elem;
5080 int log_mode = LOG_INODE_EXISTS;
5083 btrfs_item_key_to_cpu(leaf, &min_key, i);
5084 if (min_key.objectid != dir_elem->ino ||
5085 min_key.type != BTRFS_DIR_ITEM_KEY)
5086 goto next_dir_inode;
5088 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5089 type = btrfs_dir_type(leaf, di);
5090 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5091 type != BTRFS_FT_DIR)
5093 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5094 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5097 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5099 if (IS_ERR(di_inode)) {
5100 ret = PTR_ERR(di_inode);
5101 goto next_dir_inode;
5104 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5109 ctx->log_new_dentries = false;
5110 if (type == BTRFS_FT_DIR)
5111 log_mode = LOG_INODE_ALL;
5112 btrfs_release_path(path);
5113 ret = btrfs_log_inode(trans, root, di_inode,
5114 log_mode, 0, LLONG_MAX, ctx);
5117 goto next_dir_inode;
5118 if (ctx->log_new_dentries) {
5119 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5121 if (!new_dir_elem) {
5123 goto next_dir_inode;
5125 new_dir_elem->ino = di_key.objectid;
5126 list_add_tail(&new_dir_elem->list, &dir_list);
5131 ret = btrfs_next_leaf(log, path);
5133 goto next_dir_inode;
5134 } else if (ret > 0) {
5136 goto next_dir_inode;
5140 if (min_key.offset < (u64)-1) {
5145 list_del(&dir_elem->list);
5149 btrfs_free_path(path);
5153 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5154 struct inode *inode,
5155 struct btrfs_log_ctx *ctx)
5158 struct btrfs_path *path;
5159 struct btrfs_key key;
5160 struct btrfs_root *root = BTRFS_I(inode)->root;
5161 const u64 ino = btrfs_ino(inode);
5163 path = btrfs_alloc_path();
5166 path->skip_locking = 1;
5167 path->search_commit_root = 1;
5170 key.type = BTRFS_INODE_REF_KEY;
5172 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5177 struct extent_buffer *leaf = path->nodes[0];
5178 int slot = path->slots[0];
5183 if (slot >= btrfs_header_nritems(leaf)) {
5184 ret = btrfs_next_leaf(root, path);
5192 btrfs_item_key_to_cpu(leaf, &key, slot);
5193 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5194 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5197 item_size = btrfs_item_size_nr(leaf, slot);
5198 ptr = btrfs_item_ptr_offset(leaf, slot);
5199 while (cur_offset < item_size) {
5200 struct btrfs_key inode_key;
5201 struct inode *dir_inode;
5203 inode_key.type = BTRFS_INODE_ITEM_KEY;
5204 inode_key.offset = 0;
5206 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5207 struct btrfs_inode_extref *extref;
5209 extref = (struct btrfs_inode_extref *)
5211 inode_key.objectid = btrfs_inode_extref_parent(
5213 cur_offset += sizeof(*extref);
5214 cur_offset += btrfs_inode_extref_name_len(leaf,
5217 inode_key.objectid = key.offset;
5218 cur_offset = item_size;
5221 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5223 /* If parent inode was deleted, skip it. */
5224 if (IS_ERR(dir_inode))
5227 ret = btrfs_log_inode(trans, root, dir_inode,
5228 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5237 btrfs_free_path(path);
5242 * helper function around btrfs_log_inode to make sure newly created
5243 * parent directories also end up in the log. A minimal inode and backref
5244 * only logging is done of any parent directories that are older than
5245 * the last committed transaction
5247 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5248 struct btrfs_root *root, struct inode *inode,
5249 struct dentry *parent,
5253 struct btrfs_log_ctx *ctx)
5255 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5256 struct super_block *sb;
5257 struct dentry *old_parent = NULL;
5259 u64 last_committed = root->fs_info->last_trans_committed;
5260 bool log_dentries = false;
5261 struct inode *orig_inode = inode;
5265 if (btrfs_test_opt(root, NOTREELOG)) {
5271 * The prev transaction commit doesn't complete, we need do
5272 * full commit by ourselves.
5274 if (root->fs_info->last_trans_log_full_commit >
5275 root->fs_info->last_trans_committed) {
5280 if (root != BTRFS_I(inode)->root ||
5281 btrfs_root_refs(&root->root_item) == 0) {
5286 ret = check_parent_dirs_for_sync(trans, inode, parent,
5287 sb, last_committed);
5291 if (btrfs_inode_in_log(inode, trans->transid)) {
5292 ret = BTRFS_NO_LOG_SYNC;
5296 ret = start_log_trans(trans, root, ctx);
5300 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5305 * for regular files, if its inode is already on disk, we don't
5306 * have to worry about the parents at all. This is because
5307 * we can use the last_unlink_trans field to record renames
5308 * and other fun in this file.
5310 if (S_ISREG(inode->i_mode) &&
5311 BTRFS_I(inode)->generation <= last_committed &&
5312 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5317 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5318 log_dentries = true;
5321 * On unlink we must make sure all our current and old parent directores
5322 * inodes are fully logged. This is to prevent leaving dangling
5323 * directory index entries in directories that were our parents but are
5324 * not anymore. Not doing this results in old parent directory being
5325 * impossible to delete after log replay (rmdir will always fail with
5326 * error -ENOTEMPTY).
5332 * ln testdir/foo testdir/bar
5334 * unlink testdir/bar
5335 * xfs_io -c fsync testdir/foo
5337 * mount fs, triggers log replay
5339 * If we don't log the parent directory (testdir), after log replay the
5340 * directory still has an entry pointing to the file inode using the bar
5341 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5342 * the file inode has a link count of 1.
5348 * ln foo testdir/foo2
5349 * ln foo testdir/foo3
5351 * unlink testdir/foo3
5352 * xfs_io -c fsync foo
5354 * mount fs, triggers log replay
5356 * Similar as the first example, after log replay the parent directory
5357 * testdir still has an entry pointing to the inode file with name foo3
5358 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5359 * and has a link count of 2.
5361 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5362 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5368 if (!parent || d_really_is_negative(parent) || sb != d_inode(parent)->i_sb)
5371 inode = d_inode(parent);
5372 if (root != BTRFS_I(inode)->root)
5375 if (BTRFS_I(inode)->generation > last_committed) {
5376 ret = btrfs_log_inode(trans, root, inode,
5382 if (IS_ROOT(parent))
5385 parent = dget_parent(parent);
5387 old_parent = parent;
5390 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5396 btrfs_set_log_full_commit(root->fs_info, trans);
5401 btrfs_remove_log_ctx(root, ctx);
5402 btrfs_end_log_trans(root);
5408 * it is not safe to log dentry if the chunk root has added new
5409 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5410 * If this returns 1, you must commit the transaction to safely get your
5413 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5414 struct btrfs_root *root, struct dentry *dentry,
5417 struct btrfs_log_ctx *ctx)
5419 struct dentry *parent = dget_parent(dentry);
5422 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5423 start, end, 0, ctx);
5430 * should be called during mount to recover any replay any log trees
5433 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5436 struct btrfs_path *path;
5437 struct btrfs_trans_handle *trans;
5438 struct btrfs_key key;
5439 struct btrfs_key found_key;
5440 struct btrfs_key tmp_key;
5441 struct btrfs_root *log;
5442 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5443 struct walk_control wc = {
5444 .process_func = process_one_buffer,
5448 path = btrfs_alloc_path();
5452 fs_info->log_root_recovering = 1;
5454 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5455 if (IS_ERR(trans)) {
5456 ret = PTR_ERR(trans);
5463 ret = walk_log_tree(trans, log_root_tree, &wc);
5465 btrfs_std_error(fs_info, ret, "Failed to pin buffers while "
5466 "recovering log root tree.");
5471 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5472 key.offset = (u64)-1;
5473 key.type = BTRFS_ROOT_ITEM_KEY;
5476 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5479 btrfs_std_error(fs_info, ret,
5480 "Couldn't find tree log root.");
5484 if (path->slots[0] == 0)
5488 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5490 btrfs_release_path(path);
5491 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5494 log = btrfs_read_fs_root(log_root_tree, &found_key);
5497 btrfs_std_error(fs_info, ret,
5498 "Couldn't read tree log root.");
5502 tmp_key.objectid = found_key.offset;
5503 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5504 tmp_key.offset = (u64)-1;
5506 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5507 if (IS_ERR(wc.replay_dest)) {
5508 ret = PTR_ERR(wc.replay_dest);
5509 free_extent_buffer(log->node);
5510 free_extent_buffer(log->commit_root);
5512 btrfs_std_error(fs_info, ret, "Couldn't read target root "
5513 "for tree log recovery.");
5517 wc.replay_dest->log_root = log;
5518 btrfs_record_root_in_trans(trans, wc.replay_dest);
5519 ret = walk_log_tree(trans, log, &wc);
5521 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5522 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5526 key.offset = found_key.offset - 1;
5527 wc.replay_dest->log_root = NULL;
5528 free_extent_buffer(log->node);
5529 free_extent_buffer(log->commit_root);
5535 if (found_key.offset == 0)
5538 btrfs_release_path(path);
5540 /* step one is to pin it all, step two is to replay just inodes */
5543 wc.process_func = replay_one_buffer;
5544 wc.stage = LOG_WALK_REPLAY_INODES;
5547 /* step three is to replay everything */
5548 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5553 btrfs_free_path(path);
5555 /* step 4: commit the transaction, which also unpins the blocks */
5556 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5560 free_extent_buffer(log_root_tree->node);
5561 log_root_tree->log_root = NULL;
5562 fs_info->log_root_recovering = 0;
5563 kfree(log_root_tree);
5568 btrfs_end_transaction(wc.trans, fs_info->tree_root);
5569 btrfs_free_path(path);
5574 * there are some corner cases where we want to force a full
5575 * commit instead of allowing a directory to be logged.
5577 * They revolve around files there were unlinked from the directory, and
5578 * this function updates the parent directory so that a full commit is
5579 * properly done if it is fsync'd later after the unlinks are done.
5581 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5582 struct inode *dir, struct inode *inode,
5586 * when we're logging a file, if it hasn't been renamed
5587 * or unlinked, and its inode is fully committed on disk,
5588 * we don't have to worry about walking up the directory chain
5589 * to log its parents.
5591 * So, we use the last_unlink_trans field to put this transid
5592 * into the file. When the file is logged we check it and
5593 * don't log the parents if the file is fully on disk.
5595 if (S_ISREG(inode->i_mode))
5596 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5599 * if this directory was already logged any new
5600 * names for this file/dir will get recorded
5603 if (BTRFS_I(dir)->logged_trans == trans->transid)
5607 * if the inode we're about to unlink was logged,
5608 * the log will be properly updated for any new names
5610 if (BTRFS_I(inode)->logged_trans == trans->transid)
5614 * when renaming files across directories, if the directory
5615 * there we're unlinking from gets fsync'd later on, there's
5616 * no way to find the destination directory later and fsync it
5617 * properly. So, we have to be conservative and force commits
5618 * so the new name gets discovered.
5623 /* we can safely do the unlink without any special recording */
5627 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5631 * Call this after adding a new name for a file and it will properly
5632 * update the log to reflect the new name.
5634 * It will return zero if all goes well, and it will return 1 if a
5635 * full transaction commit is required.
5637 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5638 struct inode *inode, struct inode *old_dir,
5639 struct dentry *parent)
5641 struct btrfs_root * root = BTRFS_I(inode)->root;
5644 * this will force the logging code to walk the dentry chain
5647 if (S_ISREG(inode->i_mode))
5648 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5651 * if this inode hasn't been logged and directory we're renaming it
5652 * from hasn't been logged, we don't need to log it
5654 if (BTRFS_I(inode)->logged_trans <=
5655 root->fs_info->last_trans_committed &&
5656 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5657 root->fs_info->last_trans_committed))
5660 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5661 LLONG_MAX, 1, NULL);