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>
23 #include <linux/iversion.h>
28 #include "print-tree.h"
30 #include "compression.h"
32 #include "inode-map.h"
34 /* magic values for the inode_only field in btrfs_log_inode:
36 * LOG_INODE_ALL means to log everything
37 * LOG_INODE_EXISTS means to log just enough to recreate the inode
40 #define LOG_INODE_ALL 0
41 #define LOG_INODE_EXISTS 1
42 #define LOG_OTHER_INODE 2
45 * directory trouble cases
47 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
48 * log, we must force a full commit before doing an fsync of the directory
49 * where the unlink was done.
50 * ---> record transid of last unlink/rename per directory
54 * rename foo/some_dir foo2/some_dir
56 * fsync foo/some_dir/some_file
58 * The fsync above will unlink the original some_dir without recording
59 * it in its new location (foo2). After a crash, some_dir will be gone
60 * unless the fsync of some_file forces a full commit
62 * 2) we must log any new names for any file or dir that is in the fsync
63 * log. ---> check inode while renaming/linking.
65 * 2a) we must log any new names for any file or dir during rename
66 * when the directory they are being removed from was logged.
67 * ---> check inode and old parent dir during rename
69 * 2a is actually the more important variant. With the extra logging
70 * a crash might unlink the old name without recreating the new one
72 * 3) after a crash, we must go through any directories with a link count
73 * of zero and redo the rm -rf
80 * The directory f1 was fully removed from the FS, but fsync was never
81 * called on f1, only its parent dir. After a crash the rm -rf must
82 * be replayed. This must be able to recurse down the entire
83 * directory tree. The inode link count fixup code takes care of the
88 * stages for the tree walking. The first
89 * stage (0) is to only pin down the blocks we find
90 * the second stage (1) is to make sure that all the inodes
91 * we find in the log are created in the subvolume.
93 * The last stage is to deal with directories and links and extents
94 * and all the other fun semantics
96 #define LOG_WALK_PIN_ONLY 0
97 #define LOG_WALK_REPLAY_INODES 1
98 #define LOG_WALK_REPLAY_DIR_INDEX 2
99 #define LOG_WALK_REPLAY_ALL 3
101 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
102 struct btrfs_root *root, struct btrfs_inode *inode,
106 struct btrfs_log_ctx *ctx);
107 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
108 struct btrfs_root *root,
109 struct btrfs_path *path, u64 objectid);
110 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
111 struct btrfs_root *root,
112 struct btrfs_root *log,
113 struct btrfs_path *path,
114 u64 dirid, int del_all);
117 * tree logging is a special write ahead log used to make sure that
118 * fsyncs and O_SYNCs can happen without doing full tree commits.
120 * Full tree commits are expensive because they require commonly
121 * modified blocks to be recowed, creating many dirty pages in the
122 * extent tree an 4x-6x higher write load than ext3.
124 * Instead of doing a tree commit on every fsync, we use the
125 * key ranges and transaction ids to find items for a given file or directory
126 * that have changed in this transaction. Those items are copied into
127 * a special tree (one per subvolume root), that tree is written to disk
128 * and then the fsync is considered complete.
130 * After a crash, items are copied out of the log-tree back into the
131 * subvolume tree. Any file data extents found are recorded in the extent
132 * allocation tree, and the log-tree freed.
134 * The log tree is read three times, once to pin down all the extents it is
135 * using in ram and once, once to create all the inodes logged in the tree
136 * and once to do all the other items.
140 * start a sub transaction and setup the log tree
141 * this increments the log tree writer count to make the people
142 * syncing the tree wait for us to finish
144 static int start_log_trans(struct btrfs_trans_handle *trans,
145 struct btrfs_root *root,
146 struct btrfs_log_ctx *ctx)
148 struct btrfs_fs_info *fs_info = root->fs_info;
151 mutex_lock(&root->log_mutex);
153 if (root->log_root) {
154 if (btrfs_need_log_full_commit(fs_info, trans)) {
159 if (!root->log_start_pid) {
160 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
161 root->log_start_pid = current->pid;
162 } else if (root->log_start_pid != current->pid) {
163 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
166 mutex_lock(&fs_info->tree_log_mutex);
167 if (!fs_info->log_root_tree)
168 ret = btrfs_init_log_root_tree(trans, fs_info);
169 mutex_unlock(&fs_info->tree_log_mutex);
173 ret = btrfs_add_log_tree(trans, root);
177 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
178 root->log_start_pid = current->pid;
181 atomic_inc(&root->log_batch);
182 atomic_inc(&root->log_writers);
184 int index = root->log_transid % 2;
185 list_add_tail(&ctx->list, &root->log_ctxs[index]);
186 ctx->log_transid = root->log_transid;
190 mutex_unlock(&root->log_mutex);
195 * returns 0 if there was a log transaction running and we were able
196 * to join, or returns -ENOENT if there were not transactions
199 static int join_running_log_trans(struct btrfs_root *root)
207 mutex_lock(&root->log_mutex);
208 if (root->log_root) {
210 atomic_inc(&root->log_writers);
212 mutex_unlock(&root->log_mutex);
217 * This either makes the current running log transaction wait
218 * until you call btrfs_end_log_trans() or it makes any future
219 * log transactions wait until you call btrfs_end_log_trans()
221 int btrfs_pin_log_trans(struct btrfs_root *root)
225 mutex_lock(&root->log_mutex);
226 atomic_inc(&root->log_writers);
227 mutex_unlock(&root->log_mutex);
232 * indicate we're done making changes to the log tree
233 * and wake up anyone waiting to do a sync
235 void btrfs_end_log_trans(struct btrfs_root *root)
237 if (atomic_dec_and_test(&root->log_writers)) {
239 * Implicit memory barrier after atomic_dec_and_test
241 if (waitqueue_active(&root->log_writer_wait))
242 wake_up(&root->log_writer_wait);
248 * the walk control struct is used to pass state down the chain when
249 * processing the log tree. The stage field tells us which part
250 * of the log tree processing we are currently doing. The others
251 * are state fields used for that specific part
253 struct walk_control {
254 /* should we free the extent on disk when done? This is used
255 * at transaction commit time while freeing a log tree
259 /* should we write out the extent buffer? This is used
260 * while flushing the log tree to disk during a sync
264 /* should we wait for the extent buffer io to finish? Also used
265 * while flushing the log tree to disk for a sync
269 /* pin only walk, we record which extents on disk belong to the
274 /* what stage of the replay code we're currently in */
277 /* the root we are currently replaying */
278 struct btrfs_root *replay_dest;
280 /* the trans handle for the current replay */
281 struct btrfs_trans_handle *trans;
283 /* the function that gets used to process blocks we find in the
284 * tree. Note the extent_buffer might not be up to date when it is
285 * passed in, and it must be checked or read if you need the data
288 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
289 struct walk_control *wc, u64 gen, int level);
293 * process_func used to pin down extents, write them or wait on them
295 static int process_one_buffer(struct btrfs_root *log,
296 struct extent_buffer *eb,
297 struct walk_control *wc, u64 gen, int level)
299 struct btrfs_fs_info *fs_info = log->fs_info;
303 * If this fs is mixed then we need to be able to process the leaves to
304 * pin down any logged extents, so we have to read the block.
306 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
307 ret = btrfs_read_buffer(eb, gen, level, NULL);
313 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
316 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
317 if (wc->pin && btrfs_header_level(eb) == 0)
318 ret = btrfs_exclude_logged_extents(fs_info, eb);
320 btrfs_write_tree_block(eb);
322 btrfs_wait_tree_block_writeback(eb);
328 * Item overwrite used by replay and tree logging. eb, slot and key all refer
329 * to the src data we are copying out.
331 * root is the tree we are copying into, and path is a scratch
332 * path for use in this function (it should be released on entry and
333 * will be released on exit).
335 * If the key is already in the destination tree the existing item is
336 * overwritten. If the existing item isn't big enough, it is extended.
337 * If it is too large, it is truncated.
339 * If the key isn't in the destination yet, a new item is inserted.
341 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
342 struct btrfs_root *root,
343 struct btrfs_path *path,
344 struct extent_buffer *eb, int slot,
345 struct btrfs_key *key)
347 struct btrfs_fs_info *fs_info = root->fs_info;
350 u64 saved_i_size = 0;
351 int save_old_i_size = 0;
352 unsigned long src_ptr;
353 unsigned long dst_ptr;
354 int overwrite_root = 0;
355 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
357 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
360 item_size = btrfs_item_size_nr(eb, slot);
361 src_ptr = btrfs_item_ptr_offset(eb, slot);
363 /* look for the key in the destination tree */
364 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
371 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
373 if (dst_size != item_size)
376 if (item_size == 0) {
377 btrfs_release_path(path);
380 dst_copy = kmalloc(item_size, GFP_NOFS);
381 src_copy = kmalloc(item_size, GFP_NOFS);
382 if (!dst_copy || !src_copy) {
383 btrfs_release_path(path);
389 read_extent_buffer(eb, src_copy, src_ptr, item_size);
391 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
392 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
394 ret = memcmp(dst_copy, src_copy, item_size);
399 * they have the same contents, just return, this saves
400 * us from cowing blocks in the destination tree and doing
401 * extra writes that may not have been done by a previous
405 btrfs_release_path(path);
410 * We need to load the old nbytes into the inode so when we
411 * replay the extents we've logged we get the right nbytes.
414 struct btrfs_inode_item *item;
418 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
419 struct btrfs_inode_item);
420 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
421 item = btrfs_item_ptr(eb, slot,
422 struct btrfs_inode_item);
423 btrfs_set_inode_nbytes(eb, item, nbytes);
426 * If this is a directory we need to reset the i_size to
427 * 0 so that we can set it up properly when replaying
428 * the rest of the items in this log.
430 mode = btrfs_inode_mode(eb, item);
432 btrfs_set_inode_size(eb, item, 0);
434 } else if (inode_item) {
435 struct btrfs_inode_item *item;
439 * New inode, set nbytes to 0 so that the nbytes comes out
440 * properly when we replay the extents.
442 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
443 btrfs_set_inode_nbytes(eb, item, 0);
446 * If this is a directory we need to reset the i_size to 0 so
447 * that we can set it up properly when replaying the rest of
448 * the items in this log.
450 mode = btrfs_inode_mode(eb, item);
452 btrfs_set_inode_size(eb, item, 0);
455 btrfs_release_path(path);
456 /* try to insert the key into the destination tree */
457 path->skip_release_on_error = 1;
458 ret = btrfs_insert_empty_item(trans, root, path,
460 path->skip_release_on_error = 0;
462 /* make sure any existing item is the correct size */
463 if (ret == -EEXIST || ret == -EOVERFLOW) {
465 found_size = btrfs_item_size_nr(path->nodes[0],
467 if (found_size > item_size)
468 btrfs_truncate_item(fs_info, path, item_size, 1);
469 else if (found_size < item_size)
470 btrfs_extend_item(fs_info, path,
471 item_size - found_size);
475 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
478 /* don't overwrite an existing inode if the generation number
479 * was logged as zero. This is done when the tree logging code
480 * is just logging an inode to make sure it exists after recovery.
482 * Also, don't overwrite i_size on directories during replay.
483 * log replay inserts and removes directory items based on the
484 * state of the tree found in the subvolume, and i_size is modified
487 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
488 struct btrfs_inode_item *src_item;
489 struct btrfs_inode_item *dst_item;
491 src_item = (struct btrfs_inode_item *)src_ptr;
492 dst_item = (struct btrfs_inode_item *)dst_ptr;
494 if (btrfs_inode_generation(eb, src_item) == 0) {
495 struct extent_buffer *dst_eb = path->nodes[0];
496 const u64 ino_size = btrfs_inode_size(eb, src_item);
499 * For regular files an ino_size == 0 is used only when
500 * logging that an inode exists, as part of a directory
501 * fsync, and the inode wasn't fsynced before. In this
502 * case don't set the size of the inode in the fs/subvol
503 * tree, otherwise we would be throwing valid data away.
505 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
506 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
508 struct btrfs_map_token token;
510 btrfs_init_map_token(&token);
511 btrfs_set_token_inode_size(dst_eb, dst_item,
517 if (overwrite_root &&
518 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
519 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
521 saved_i_size = btrfs_inode_size(path->nodes[0],
526 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
529 if (save_old_i_size) {
530 struct btrfs_inode_item *dst_item;
531 dst_item = (struct btrfs_inode_item *)dst_ptr;
532 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
535 /* make sure the generation is filled in */
536 if (key->type == BTRFS_INODE_ITEM_KEY) {
537 struct btrfs_inode_item *dst_item;
538 dst_item = (struct btrfs_inode_item *)dst_ptr;
539 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
540 btrfs_set_inode_generation(path->nodes[0], dst_item,
545 btrfs_mark_buffer_dirty(path->nodes[0]);
546 btrfs_release_path(path);
551 * simple helper to read an inode off the disk from a given root
552 * This can only be called for subvolume roots and not for the log
554 static noinline struct inode *read_one_inode(struct btrfs_root *root,
557 struct btrfs_key key;
560 key.objectid = objectid;
561 key.type = BTRFS_INODE_ITEM_KEY;
563 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
566 } else if (is_bad_inode(inode)) {
573 /* replays a single extent in 'eb' at 'slot' with 'key' into the
574 * subvolume 'root'. path is released on entry and should be released
577 * extents in the log tree have not been allocated out of the extent
578 * tree yet. So, this completes the allocation, taking a reference
579 * as required if the extent already exists or creating a new extent
580 * if it isn't in the extent allocation tree yet.
582 * The extent is inserted into the file, dropping any existing extents
583 * from the file that overlap the new one.
585 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path,
588 struct extent_buffer *eb, int slot,
589 struct btrfs_key *key)
591 struct btrfs_fs_info *fs_info = root->fs_info;
594 u64 start = key->offset;
596 struct btrfs_file_extent_item *item;
597 struct inode *inode = NULL;
601 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
602 found_type = btrfs_file_extent_type(eb, item);
604 if (found_type == BTRFS_FILE_EXTENT_REG ||
605 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
606 nbytes = btrfs_file_extent_num_bytes(eb, item);
607 extent_end = start + nbytes;
610 * We don't add to the inodes nbytes if we are prealloc or a
613 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
615 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
616 size = btrfs_file_extent_inline_len(eb, slot, item);
617 nbytes = btrfs_file_extent_ram_bytes(eb, item);
618 extent_end = ALIGN(start + size,
619 fs_info->sectorsize);
625 inode = read_one_inode(root, key->objectid);
632 * first check to see if we already have this extent in the
633 * file. This must be done before the btrfs_drop_extents run
634 * so we don't try to drop this extent.
636 ret = btrfs_lookup_file_extent(trans, root, path,
637 btrfs_ino(BTRFS_I(inode)), start, 0);
640 (found_type == BTRFS_FILE_EXTENT_REG ||
641 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
642 struct btrfs_file_extent_item cmp1;
643 struct btrfs_file_extent_item cmp2;
644 struct btrfs_file_extent_item *existing;
645 struct extent_buffer *leaf;
647 leaf = path->nodes[0];
648 existing = btrfs_item_ptr(leaf, path->slots[0],
649 struct btrfs_file_extent_item);
651 read_extent_buffer(eb, &cmp1, (unsigned long)item,
653 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
657 * we already have a pointer to this exact extent,
658 * we don't have to do anything
660 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
661 btrfs_release_path(path);
665 btrfs_release_path(path);
667 /* drop any overlapping extents */
668 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
672 if (found_type == BTRFS_FILE_EXTENT_REG ||
673 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
675 unsigned long dest_offset;
676 struct btrfs_key ins;
678 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
679 btrfs_fs_incompat(fs_info, NO_HOLES))
682 ret = btrfs_insert_empty_item(trans, root, path, key,
686 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
688 copy_extent_buffer(path->nodes[0], eb, dest_offset,
689 (unsigned long)item, sizeof(*item));
691 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
692 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
693 ins.type = BTRFS_EXTENT_ITEM_KEY;
694 offset = key->offset - btrfs_file_extent_offset(eb, item);
697 * Manually record dirty extent, as here we did a shallow
698 * file extent item copy and skip normal backref update,
699 * but modifying extent tree all by ourselves.
700 * So need to manually record dirty extent for qgroup,
701 * as the owner of the file extent changed from log tree
702 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
704 ret = btrfs_qgroup_trace_extent(trans, fs_info,
705 btrfs_file_extent_disk_bytenr(eb, item),
706 btrfs_file_extent_disk_num_bytes(eb, item),
711 if (ins.objectid > 0) {
714 LIST_HEAD(ordered_sums);
716 * is this extent already allocated in the extent
717 * allocation tree? If so, just add a reference
719 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
722 ret = btrfs_inc_extent_ref(trans, root,
723 ins.objectid, ins.offset,
724 0, root->root_key.objectid,
725 key->objectid, offset);
730 * insert the extent pointer in the extent
733 ret = btrfs_alloc_logged_file_extent(trans,
735 root->root_key.objectid,
736 key->objectid, offset, &ins);
740 btrfs_release_path(path);
742 if (btrfs_file_extent_compression(eb, item)) {
743 csum_start = ins.objectid;
744 csum_end = csum_start + ins.offset;
746 csum_start = ins.objectid +
747 btrfs_file_extent_offset(eb, item);
748 csum_end = csum_start +
749 btrfs_file_extent_num_bytes(eb, item);
752 ret = btrfs_lookup_csums_range(root->log_root,
753 csum_start, csum_end - 1,
758 * Now delete all existing cums in the csum root that
759 * cover our range. We do this because we can have an
760 * extent that is completely referenced by one file
761 * extent item and partially referenced by another
762 * file extent item (like after using the clone or
763 * extent_same ioctls). In this case if we end up doing
764 * the replay of the one that partially references the
765 * extent first, and we do not do the csum deletion
766 * below, we can get 2 csum items in the csum tree that
767 * overlap each other. For example, imagine our log has
768 * the two following file extent items:
770 * key (257 EXTENT_DATA 409600)
771 * extent data disk byte 12845056 nr 102400
772 * extent data offset 20480 nr 20480 ram 102400
774 * key (257 EXTENT_DATA 819200)
775 * extent data disk byte 12845056 nr 102400
776 * extent data offset 0 nr 102400 ram 102400
778 * Where the second one fully references the 100K extent
779 * that starts at disk byte 12845056, and the log tree
780 * has a single csum item that covers the entire range
783 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
785 * After the first file extent item is replayed, the
786 * csum tree gets the following csum item:
788 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
790 * Which covers the 20K sub-range starting at offset 20K
791 * of our extent. Now when we replay the second file
792 * extent item, if we do not delete existing csum items
793 * that cover any of its blocks, we end up getting two
794 * csum items in our csum tree that overlap each other:
796 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
797 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
799 * Which is a problem, because after this anyone trying
800 * to lookup up for the checksum of any block of our
801 * extent starting at an offset of 40K or higher, will
802 * end up looking at the second csum item only, which
803 * does not contain the checksum for any block starting
804 * at offset 40K or higher of our extent.
806 while (!list_empty(&ordered_sums)) {
807 struct btrfs_ordered_sum *sums;
808 sums = list_entry(ordered_sums.next,
809 struct btrfs_ordered_sum,
812 ret = btrfs_del_csums(trans, fs_info,
816 ret = btrfs_csum_file_blocks(trans,
817 fs_info->csum_root, sums);
818 list_del(&sums->list);
824 btrfs_release_path(path);
826 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
827 /* inline extents are easy, we just overwrite them */
828 ret = overwrite_item(trans, root, path, eb, slot, key);
833 inode_add_bytes(inode, nbytes);
835 ret = btrfs_update_inode(trans, root, inode);
843 * when cleaning up conflicts between the directory names in the
844 * subvolume, directory names in the log and directory names in the
845 * inode back references, we may have to unlink inodes from directories.
847 * This is a helper function to do the unlink of a specific directory
850 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
851 struct btrfs_root *root,
852 struct btrfs_path *path,
853 struct btrfs_inode *dir,
854 struct btrfs_dir_item *di)
859 struct extent_buffer *leaf;
860 struct btrfs_key location;
863 leaf = path->nodes[0];
865 btrfs_dir_item_key_to_cpu(leaf, di, &location);
866 name_len = btrfs_dir_name_len(leaf, di);
867 name = kmalloc(name_len, GFP_NOFS);
871 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
872 btrfs_release_path(path);
874 inode = read_one_inode(root, location.objectid);
880 ret = link_to_fixup_dir(trans, root, path, location.objectid);
884 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
889 ret = btrfs_run_delayed_items(trans);
897 * helper function to see if a given name and sequence number found
898 * in an inode back reference are already in a directory and correctly
899 * point to this inode
901 static noinline int inode_in_dir(struct btrfs_root *root,
902 struct btrfs_path *path,
903 u64 dirid, u64 objectid, u64 index,
904 const char *name, int name_len)
906 struct btrfs_dir_item *di;
907 struct btrfs_key location;
910 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
911 index, name, name_len, 0);
912 if (di && !IS_ERR(di)) {
913 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
914 if (location.objectid != objectid)
918 btrfs_release_path(path);
920 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
921 if (di && !IS_ERR(di)) {
922 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
923 if (location.objectid != objectid)
929 btrfs_release_path(path);
934 * helper function to check a log tree for a named back reference in
935 * an inode. This is used to decide if a back reference that is
936 * found in the subvolume conflicts with what we find in the log.
938 * inode backreferences may have multiple refs in a single item,
939 * during replay we process one reference at a time, and we don't
940 * want to delete valid links to a file from the subvolume if that
941 * link is also in the log.
943 static noinline int backref_in_log(struct btrfs_root *log,
944 struct btrfs_key *key,
946 const char *name, int namelen)
948 struct btrfs_path *path;
949 struct btrfs_inode_ref *ref;
951 unsigned long ptr_end;
952 unsigned long name_ptr;
958 path = btrfs_alloc_path();
962 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
966 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
968 if (key->type == BTRFS_INODE_EXTREF_KEY) {
969 if (btrfs_find_name_in_ext_backref(path->nodes[0],
972 name, namelen, NULL))
978 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
979 ptr_end = ptr + item_size;
980 while (ptr < ptr_end) {
981 ref = (struct btrfs_inode_ref *)ptr;
982 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
983 if (found_name_len == namelen) {
984 name_ptr = (unsigned long)(ref + 1);
985 ret = memcmp_extent_buffer(path->nodes[0], name,
992 ptr = (unsigned long)(ref + 1) + found_name_len;
995 btrfs_free_path(path);
999 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1000 struct btrfs_root *root,
1001 struct btrfs_path *path,
1002 struct btrfs_root *log_root,
1003 struct btrfs_inode *dir,
1004 struct btrfs_inode *inode,
1005 u64 inode_objectid, u64 parent_objectid,
1006 u64 ref_index, char *name, int namelen,
1011 int victim_name_len;
1012 struct extent_buffer *leaf;
1013 struct btrfs_dir_item *di;
1014 struct btrfs_key search_key;
1015 struct btrfs_inode_extref *extref;
1018 /* Search old style refs */
1019 search_key.objectid = inode_objectid;
1020 search_key.type = BTRFS_INODE_REF_KEY;
1021 search_key.offset = parent_objectid;
1022 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1024 struct btrfs_inode_ref *victim_ref;
1026 unsigned long ptr_end;
1028 leaf = path->nodes[0];
1030 /* are we trying to overwrite a back ref for the root directory
1031 * if so, just jump out, we're done
1033 if (search_key.objectid == search_key.offset)
1036 /* check all the names in this back reference to see
1037 * if they are in the log. if so, we allow them to stay
1038 * otherwise they must be unlinked as a conflict
1040 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1041 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1042 while (ptr < ptr_end) {
1043 victim_ref = (struct btrfs_inode_ref *)ptr;
1044 victim_name_len = btrfs_inode_ref_name_len(leaf,
1046 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1050 read_extent_buffer(leaf, victim_name,
1051 (unsigned long)(victim_ref + 1),
1054 if (!backref_in_log(log_root, &search_key,
1058 inc_nlink(&inode->vfs_inode);
1059 btrfs_release_path(path);
1061 ret = btrfs_unlink_inode(trans, root, dir, inode,
1062 victim_name, victim_name_len);
1066 ret = btrfs_run_delayed_items(trans);
1074 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1078 * NOTE: we have searched root tree and checked the
1079 * corresponding ref, it does not need to check again.
1083 btrfs_release_path(path);
1085 /* Same search but for extended refs */
1086 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1087 inode_objectid, parent_objectid, 0,
1089 if (!IS_ERR_OR_NULL(extref)) {
1093 struct inode *victim_parent;
1095 leaf = path->nodes[0];
1097 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1098 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1100 while (cur_offset < item_size) {
1101 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1103 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1105 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1108 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1111 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1114 search_key.objectid = inode_objectid;
1115 search_key.type = BTRFS_INODE_EXTREF_KEY;
1116 search_key.offset = btrfs_extref_hash(parent_objectid,
1120 if (!backref_in_log(log_root, &search_key,
1121 parent_objectid, victim_name,
1124 victim_parent = read_one_inode(root,
1126 if (victim_parent) {
1127 inc_nlink(&inode->vfs_inode);
1128 btrfs_release_path(path);
1130 ret = btrfs_unlink_inode(trans, root,
1131 BTRFS_I(victim_parent),
1136 ret = btrfs_run_delayed_items(
1139 iput(victim_parent);
1148 cur_offset += victim_name_len + sizeof(*extref);
1152 btrfs_release_path(path);
1154 /* look for a conflicting sequence number */
1155 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1156 ref_index, name, namelen, 0);
1157 if (di && !IS_ERR(di)) {
1158 ret = drop_one_dir_item(trans, root, path, dir, di);
1162 btrfs_release_path(path);
1164 /* look for a conflicing name */
1165 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1167 if (di && !IS_ERR(di)) {
1168 ret = drop_one_dir_item(trans, root, path, dir, di);
1172 btrfs_release_path(path);
1177 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1178 u32 *namelen, char **name, u64 *index,
1179 u64 *parent_objectid)
1181 struct btrfs_inode_extref *extref;
1183 extref = (struct btrfs_inode_extref *)ref_ptr;
1185 *namelen = btrfs_inode_extref_name_len(eb, extref);
1186 *name = kmalloc(*namelen, GFP_NOFS);
1190 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1194 *index = btrfs_inode_extref_index(eb, extref);
1195 if (parent_objectid)
1196 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1201 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1202 u32 *namelen, char **name, u64 *index)
1204 struct btrfs_inode_ref *ref;
1206 ref = (struct btrfs_inode_ref *)ref_ptr;
1208 *namelen = btrfs_inode_ref_name_len(eb, ref);
1209 *name = kmalloc(*namelen, GFP_NOFS);
1213 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1216 *index = btrfs_inode_ref_index(eb, ref);
1222 * Take an inode reference item from the log tree and iterate all names from the
1223 * inode reference item in the subvolume tree with the same key (if it exists).
1224 * For any name that is not in the inode reference item from the log tree, do a
1225 * proper unlink of that name (that is, remove its entry from the inode
1226 * reference item and both dir index keys).
1228 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1229 struct btrfs_root *root,
1230 struct btrfs_path *path,
1231 struct btrfs_inode *inode,
1232 struct extent_buffer *log_eb,
1234 struct btrfs_key *key)
1237 unsigned long ref_ptr;
1238 unsigned long ref_end;
1239 struct extent_buffer *eb;
1242 btrfs_release_path(path);
1243 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1251 eb = path->nodes[0];
1252 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1253 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1254 while (ref_ptr < ref_end) {
1259 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1260 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1263 parent_id = key->offset;
1264 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1270 if (key->type == BTRFS_INODE_EXTREF_KEY)
1271 ret = btrfs_find_name_in_ext_backref(log_eb, log_slot,
1275 ret = btrfs_find_name_in_backref(log_eb, log_slot, name,
1281 btrfs_release_path(path);
1282 dir = read_one_inode(root, parent_id);
1288 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1289 inode, name, namelen);
1299 if (key->type == BTRFS_INODE_EXTREF_KEY)
1300 ref_ptr += sizeof(struct btrfs_inode_extref);
1302 ref_ptr += sizeof(struct btrfs_inode_ref);
1306 btrfs_release_path(path);
1311 * replay one inode back reference item found in the log tree.
1312 * eb, slot and key refer to the buffer and key found in the log tree.
1313 * root is the destination we are replaying into, and path is for temp
1314 * use by this function. (it should be released on return).
1316 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root,
1318 struct btrfs_root *log,
1319 struct btrfs_path *path,
1320 struct extent_buffer *eb, int slot,
1321 struct btrfs_key *key)
1323 struct inode *dir = NULL;
1324 struct inode *inode = NULL;
1325 unsigned long ref_ptr;
1326 unsigned long ref_end;
1330 int search_done = 0;
1331 int log_ref_ver = 0;
1332 u64 parent_objectid;
1335 int ref_struct_size;
1337 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1338 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1340 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1341 struct btrfs_inode_extref *r;
1343 ref_struct_size = sizeof(struct btrfs_inode_extref);
1345 r = (struct btrfs_inode_extref *)ref_ptr;
1346 parent_objectid = btrfs_inode_extref_parent(eb, r);
1348 ref_struct_size = sizeof(struct btrfs_inode_ref);
1349 parent_objectid = key->offset;
1351 inode_objectid = key->objectid;
1354 * it is possible that we didn't log all the parent directories
1355 * for a given inode. If we don't find the dir, just don't
1356 * copy the back ref in. The link count fixup code will take
1359 dir = read_one_inode(root, parent_objectid);
1365 inode = read_one_inode(root, inode_objectid);
1371 while (ref_ptr < ref_end) {
1373 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1374 &ref_index, &parent_objectid);
1376 * parent object can change from one array
1380 dir = read_one_inode(root, parent_objectid);
1386 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1392 /* if we already have a perfect match, we're done */
1393 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1394 btrfs_ino(BTRFS_I(inode)), ref_index,
1397 * look for a conflicting back reference in the
1398 * metadata. if we find one we have to unlink that name
1399 * of the file before we add our new link. Later on, we
1400 * overwrite any existing back reference, and we don't
1401 * want to create dangling pointers in the directory.
1405 ret = __add_inode_ref(trans, root, path, log,
1410 ref_index, name, namelen,
1419 /* insert our name */
1420 ret = btrfs_add_link(trans, BTRFS_I(dir),
1422 name, namelen, 0, ref_index);
1426 btrfs_update_inode(trans, root, inode);
1429 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1439 * Before we overwrite the inode reference item in the subvolume tree
1440 * with the item from the log tree, we must unlink all names from the
1441 * parent directory that are in the subvolume's tree inode reference
1442 * item, otherwise we end up with an inconsistent subvolume tree where
1443 * dir index entries exist for a name but there is no inode reference
1444 * item with the same name.
1446 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1451 /* finally write the back reference in the inode */
1452 ret = overwrite_item(trans, root, path, eb, slot, key);
1454 btrfs_release_path(path);
1461 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1462 struct btrfs_root *root, u64 ino)
1466 ret = btrfs_insert_orphan_item(trans, root, ino);
1473 static int count_inode_extrefs(struct btrfs_root *root,
1474 struct btrfs_inode *inode, struct btrfs_path *path)
1478 unsigned int nlink = 0;
1481 u64 inode_objectid = btrfs_ino(inode);
1484 struct btrfs_inode_extref *extref;
1485 struct extent_buffer *leaf;
1488 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1493 leaf = path->nodes[0];
1494 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1495 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1498 while (cur_offset < item_size) {
1499 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1500 name_len = btrfs_inode_extref_name_len(leaf, extref);
1504 cur_offset += name_len + sizeof(*extref);
1508 btrfs_release_path(path);
1510 btrfs_release_path(path);
1512 if (ret < 0 && ret != -ENOENT)
1517 static int count_inode_refs(struct btrfs_root *root,
1518 struct btrfs_inode *inode, struct btrfs_path *path)
1521 struct btrfs_key key;
1522 unsigned int nlink = 0;
1524 unsigned long ptr_end;
1526 u64 ino = btrfs_ino(inode);
1529 key.type = BTRFS_INODE_REF_KEY;
1530 key.offset = (u64)-1;
1533 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1537 if (path->slots[0] == 0)
1542 btrfs_item_key_to_cpu(path->nodes[0], &key,
1544 if (key.objectid != ino ||
1545 key.type != BTRFS_INODE_REF_KEY)
1547 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1548 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1550 while (ptr < ptr_end) {
1551 struct btrfs_inode_ref *ref;
1553 ref = (struct btrfs_inode_ref *)ptr;
1554 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1556 ptr = (unsigned long)(ref + 1) + name_len;
1560 if (key.offset == 0)
1562 if (path->slots[0] > 0) {
1567 btrfs_release_path(path);
1569 btrfs_release_path(path);
1575 * There are a few corners where the link count of the file can't
1576 * be properly maintained during replay. So, instead of adding
1577 * lots of complexity to the log code, we just scan the backrefs
1578 * for any file that has been through replay.
1580 * The scan will update the link count on the inode to reflect the
1581 * number of back refs found. If it goes down to zero, the iput
1582 * will free the inode.
1584 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1585 struct btrfs_root *root,
1586 struct inode *inode)
1588 struct btrfs_path *path;
1591 u64 ino = btrfs_ino(BTRFS_I(inode));
1593 path = btrfs_alloc_path();
1597 ret = count_inode_refs(root, BTRFS_I(inode), path);
1603 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1611 if (nlink != inode->i_nlink) {
1612 set_nlink(inode, nlink);
1613 btrfs_update_inode(trans, root, inode);
1615 BTRFS_I(inode)->index_cnt = (u64)-1;
1617 if (inode->i_nlink == 0) {
1618 if (S_ISDIR(inode->i_mode)) {
1619 ret = replay_dir_deletes(trans, root, NULL, path,
1624 ret = insert_orphan_item(trans, root, ino);
1628 btrfs_free_path(path);
1632 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1633 struct btrfs_root *root,
1634 struct btrfs_path *path)
1637 struct btrfs_key key;
1638 struct inode *inode;
1640 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1641 key.type = BTRFS_ORPHAN_ITEM_KEY;
1642 key.offset = (u64)-1;
1644 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1649 if (path->slots[0] == 0)
1654 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1655 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1656 key.type != BTRFS_ORPHAN_ITEM_KEY)
1659 ret = btrfs_del_item(trans, root, path);
1663 btrfs_release_path(path);
1664 inode = read_one_inode(root, key.offset);
1668 ret = fixup_inode_link_count(trans, root, inode);
1674 * fixup on a directory may create new entries,
1675 * make sure we always look for the highset possible
1678 key.offset = (u64)-1;
1682 btrfs_release_path(path);
1688 * record a given inode in the fixup dir so we can check its link
1689 * count when replay is done. The link count is incremented here
1690 * so the inode won't go away until we check it
1692 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_path *path,
1697 struct btrfs_key key;
1699 struct inode *inode;
1701 inode = read_one_inode(root, objectid);
1705 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1706 key.type = BTRFS_ORPHAN_ITEM_KEY;
1707 key.offset = objectid;
1709 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1711 btrfs_release_path(path);
1713 if (!inode->i_nlink)
1714 set_nlink(inode, 1);
1717 ret = btrfs_update_inode(trans, root, inode);
1718 } else if (ret == -EEXIST) {
1721 BUG(); /* Logic Error */
1729 * when replaying the log for a directory, we only insert names
1730 * for inodes that actually exist. This means an fsync on a directory
1731 * does not implicitly fsync all the new files in it
1733 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1734 struct btrfs_root *root,
1735 u64 dirid, u64 index,
1736 char *name, int name_len,
1737 struct btrfs_key *location)
1739 struct inode *inode;
1743 inode = read_one_inode(root, location->objectid);
1747 dir = read_one_inode(root, dirid);
1753 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1754 name_len, 1, index);
1756 /* FIXME, put inode into FIXUP list */
1764 * Return true if an inode reference exists in the log for the given name,
1765 * inode and parent inode.
1767 static bool name_in_log_ref(struct btrfs_root *log_root,
1768 const char *name, const int name_len,
1769 const u64 dirid, const u64 ino)
1771 struct btrfs_key search_key;
1773 search_key.objectid = ino;
1774 search_key.type = BTRFS_INODE_REF_KEY;
1775 search_key.offset = dirid;
1776 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1779 search_key.type = BTRFS_INODE_EXTREF_KEY;
1780 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1781 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1788 * take a single entry in a log directory item and replay it into
1791 * if a conflicting item exists in the subdirectory already,
1792 * the inode it points to is unlinked and put into the link count
1795 * If a name from the log points to a file or directory that does
1796 * not exist in the FS, it is skipped. fsyncs on directories
1797 * do not force down inodes inside that directory, just changes to the
1798 * names or unlinks in a directory.
1800 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1801 * non-existing inode) and 1 if the name was replayed.
1803 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1804 struct btrfs_root *root,
1805 struct btrfs_path *path,
1806 struct extent_buffer *eb,
1807 struct btrfs_dir_item *di,
1808 struct btrfs_key *key)
1812 struct btrfs_dir_item *dst_di;
1813 struct btrfs_key found_key;
1814 struct btrfs_key log_key;
1819 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1820 bool name_added = false;
1822 dir = read_one_inode(root, key->objectid);
1826 name_len = btrfs_dir_name_len(eb, di);
1827 name = kmalloc(name_len, GFP_NOFS);
1833 log_type = btrfs_dir_type(eb, di);
1834 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1837 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1838 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1843 btrfs_release_path(path);
1845 if (key->type == BTRFS_DIR_ITEM_KEY) {
1846 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1848 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1849 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1858 if (IS_ERR_OR_NULL(dst_di)) {
1859 /* we need a sequence number to insert, so we only
1860 * do inserts for the BTRFS_DIR_INDEX_KEY types
1862 if (key->type != BTRFS_DIR_INDEX_KEY)
1867 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1868 /* the existing item matches the logged item */
1869 if (found_key.objectid == log_key.objectid &&
1870 found_key.type == log_key.type &&
1871 found_key.offset == log_key.offset &&
1872 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1873 update_size = false;
1878 * don't drop the conflicting directory entry if the inode
1879 * for the new entry doesn't exist
1884 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1888 if (key->type == BTRFS_DIR_INDEX_KEY)
1891 btrfs_release_path(path);
1892 if (!ret && update_size) {
1893 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1894 ret = btrfs_update_inode(trans, root, dir);
1898 if (!ret && name_added)
1903 if (name_in_log_ref(root->log_root, name, name_len,
1904 key->objectid, log_key.objectid)) {
1905 /* The dentry will be added later. */
1907 update_size = false;
1910 btrfs_release_path(path);
1911 ret = insert_one_name(trans, root, key->objectid, key->offset,
1912 name, name_len, &log_key);
1913 if (ret && ret != -ENOENT && ret != -EEXIST)
1917 update_size = false;
1923 * find all the names in a directory item and reconcile them into
1924 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1925 * one name in a directory item, but the same code gets used for
1926 * both directory index types
1928 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
1930 struct btrfs_path *path,
1931 struct extent_buffer *eb, int slot,
1932 struct btrfs_key *key)
1935 u32 item_size = btrfs_item_size_nr(eb, slot);
1936 struct btrfs_dir_item *di;
1939 unsigned long ptr_end;
1940 struct btrfs_path *fixup_path = NULL;
1942 ptr = btrfs_item_ptr_offset(eb, slot);
1943 ptr_end = ptr + item_size;
1944 while (ptr < ptr_end) {
1945 di = (struct btrfs_dir_item *)ptr;
1946 name_len = btrfs_dir_name_len(eb, di);
1947 ret = replay_one_name(trans, root, path, eb, di, key);
1950 ptr = (unsigned long)(di + 1);
1954 * If this entry refers to a non-directory (directories can not
1955 * have a link count > 1) and it was added in the transaction
1956 * that was not committed, make sure we fixup the link count of
1957 * the inode it the entry points to. Otherwise something like
1958 * the following would result in a directory pointing to an
1959 * inode with a wrong link that does not account for this dir
1967 * ln testdir/bar testdir/bar_link
1968 * ln testdir/foo testdir/foo_link
1969 * xfs_io -c "fsync" testdir/bar
1973 * mount fs, log replay happens
1975 * File foo would remain with a link count of 1 when it has two
1976 * entries pointing to it in the directory testdir. This would
1977 * make it impossible to ever delete the parent directory has
1978 * it would result in stale dentries that can never be deleted.
1980 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1981 struct btrfs_key di_key;
1984 fixup_path = btrfs_alloc_path();
1991 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1992 ret = link_to_fixup_dir(trans, root, fixup_path,
1999 btrfs_free_path(fixup_path);
2004 * directory replay has two parts. There are the standard directory
2005 * items in the log copied from the subvolume, and range items
2006 * created in the log while the subvolume was logged.
2008 * The range items tell us which parts of the key space the log
2009 * is authoritative for. During replay, if a key in the subvolume
2010 * directory is in a logged range item, but not actually in the log
2011 * that means it was deleted from the directory before the fsync
2012 * and should be removed.
2014 static noinline int find_dir_range(struct btrfs_root *root,
2015 struct btrfs_path *path,
2016 u64 dirid, int key_type,
2017 u64 *start_ret, u64 *end_ret)
2019 struct btrfs_key key;
2021 struct btrfs_dir_log_item *item;
2025 if (*start_ret == (u64)-1)
2028 key.objectid = dirid;
2029 key.type = key_type;
2030 key.offset = *start_ret;
2032 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2036 if (path->slots[0] == 0)
2041 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2043 if (key.type != key_type || key.objectid != dirid) {
2047 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2048 struct btrfs_dir_log_item);
2049 found_end = btrfs_dir_log_end(path->nodes[0], item);
2051 if (*start_ret >= key.offset && *start_ret <= found_end) {
2053 *start_ret = key.offset;
2054 *end_ret = found_end;
2059 /* check the next slot in the tree to see if it is a valid item */
2060 nritems = btrfs_header_nritems(path->nodes[0]);
2062 if (path->slots[0] >= nritems) {
2063 ret = btrfs_next_leaf(root, path);
2068 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2070 if (key.type != key_type || key.objectid != dirid) {
2074 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2075 struct btrfs_dir_log_item);
2076 found_end = btrfs_dir_log_end(path->nodes[0], item);
2077 *start_ret = key.offset;
2078 *end_ret = found_end;
2081 btrfs_release_path(path);
2086 * this looks for a given directory item in the log. If the directory
2087 * item is not in the log, the item is removed and the inode it points
2090 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2091 struct btrfs_root *root,
2092 struct btrfs_root *log,
2093 struct btrfs_path *path,
2094 struct btrfs_path *log_path,
2096 struct btrfs_key *dir_key)
2099 struct extent_buffer *eb;
2102 struct btrfs_dir_item *di;
2103 struct btrfs_dir_item *log_di;
2106 unsigned long ptr_end;
2108 struct inode *inode;
2109 struct btrfs_key location;
2112 eb = path->nodes[0];
2113 slot = path->slots[0];
2114 item_size = btrfs_item_size_nr(eb, slot);
2115 ptr = btrfs_item_ptr_offset(eb, slot);
2116 ptr_end = ptr + item_size;
2117 while (ptr < ptr_end) {
2118 di = (struct btrfs_dir_item *)ptr;
2119 name_len = btrfs_dir_name_len(eb, di);
2120 name = kmalloc(name_len, GFP_NOFS);
2125 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2128 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2129 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2132 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2133 log_di = btrfs_lookup_dir_index_item(trans, log,
2139 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2140 btrfs_dir_item_key_to_cpu(eb, di, &location);
2141 btrfs_release_path(path);
2142 btrfs_release_path(log_path);
2143 inode = read_one_inode(root, location.objectid);
2149 ret = link_to_fixup_dir(trans, root,
2150 path, location.objectid);
2158 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2159 BTRFS_I(inode), name, name_len);
2161 ret = btrfs_run_delayed_items(trans);
2167 /* there might still be more names under this key
2168 * check and repeat if required
2170 ret = btrfs_search_slot(NULL, root, dir_key, path,
2176 } else if (IS_ERR(log_di)) {
2178 return PTR_ERR(log_di);
2180 btrfs_release_path(log_path);
2183 ptr = (unsigned long)(di + 1);
2188 btrfs_release_path(path);
2189 btrfs_release_path(log_path);
2193 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2194 struct btrfs_root *root,
2195 struct btrfs_root *log,
2196 struct btrfs_path *path,
2199 struct btrfs_key search_key;
2200 struct btrfs_path *log_path;
2205 log_path = btrfs_alloc_path();
2209 search_key.objectid = ino;
2210 search_key.type = BTRFS_XATTR_ITEM_KEY;
2211 search_key.offset = 0;
2213 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2217 nritems = btrfs_header_nritems(path->nodes[0]);
2218 for (i = path->slots[0]; i < nritems; i++) {
2219 struct btrfs_key key;
2220 struct btrfs_dir_item *di;
2221 struct btrfs_dir_item *log_di;
2225 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2226 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2231 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2232 total_size = btrfs_item_size_nr(path->nodes[0], i);
2234 while (cur < total_size) {
2235 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2236 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2237 u32 this_len = sizeof(*di) + name_len + data_len;
2240 name = kmalloc(name_len, GFP_NOFS);
2245 read_extent_buffer(path->nodes[0], name,
2246 (unsigned long)(di + 1), name_len);
2248 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2250 btrfs_release_path(log_path);
2252 /* Doesn't exist in log tree, so delete it. */
2253 btrfs_release_path(path);
2254 di = btrfs_lookup_xattr(trans, root, path, ino,
2255 name, name_len, -1);
2262 ret = btrfs_delete_one_dir_name(trans, root,
2266 btrfs_release_path(path);
2271 if (IS_ERR(log_di)) {
2272 ret = PTR_ERR(log_di);
2276 di = (struct btrfs_dir_item *)((char *)di + this_len);
2279 ret = btrfs_next_leaf(root, path);
2285 btrfs_free_path(log_path);
2286 btrfs_release_path(path);
2292 * deletion replay happens before we copy any new directory items
2293 * out of the log or out of backreferences from inodes. It
2294 * scans the log to find ranges of keys that log is authoritative for,
2295 * and then scans the directory to find items in those ranges that are
2296 * not present in the log.
2298 * Anything we don't find in the log is unlinked and removed from the
2301 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2302 struct btrfs_root *root,
2303 struct btrfs_root *log,
2304 struct btrfs_path *path,
2305 u64 dirid, int del_all)
2309 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2311 struct btrfs_key dir_key;
2312 struct btrfs_key found_key;
2313 struct btrfs_path *log_path;
2316 dir_key.objectid = dirid;
2317 dir_key.type = BTRFS_DIR_ITEM_KEY;
2318 log_path = btrfs_alloc_path();
2322 dir = read_one_inode(root, dirid);
2323 /* it isn't an error if the inode isn't there, that can happen
2324 * because we replay the deletes before we copy in the inode item
2328 btrfs_free_path(log_path);
2336 range_end = (u64)-1;
2338 ret = find_dir_range(log, path, dirid, key_type,
2339 &range_start, &range_end);
2344 dir_key.offset = range_start;
2347 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2352 nritems = btrfs_header_nritems(path->nodes[0]);
2353 if (path->slots[0] >= nritems) {
2354 ret = btrfs_next_leaf(root, path);
2358 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2360 if (found_key.objectid != dirid ||
2361 found_key.type != dir_key.type)
2364 if (found_key.offset > range_end)
2367 ret = check_item_in_log(trans, root, log, path,
2372 if (found_key.offset == (u64)-1)
2374 dir_key.offset = found_key.offset + 1;
2376 btrfs_release_path(path);
2377 if (range_end == (u64)-1)
2379 range_start = range_end + 1;
2384 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2385 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2386 dir_key.type = BTRFS_DIR_INDEX_KEY;
2387 btrfs_release_path(path);
2391 btrfs_release_path(path);
2392 btrfs_free_path(log_path);
2398 * the process_func used to replay items from the log tree. This
2399 * gets called in two different stages. The first stage just looks
2400 * for inodes and makes sure they are all copied into the subvolume.
2402 * The second stage copies all the other item types from the log into
2403 * the subvolume. The two stage approach is slower, but gets rid of
2404 * lots of complexity around inodes referencing other inodes that exist
2405 * only in the log (references come from either directory items or inode
2408 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2409 struct walk_control *wc, u64 gen, int level)
2412 struct btrfs_path *path;
2413 struct btrfs_root *root = wc->replay_dest;
2414 struct btrfs_key key;
2418 ret = btrfs_read_buffer(eb, gen, level, NULL);
2422 level = btrfs_header_level(eb);
2427 path = btrfs_alloc_path();
2431 nritems = btrfs_header_nritems(eb);
2432 for (i = 0; i < nritems; i++) {
2433 btrfs_item_key_to_cpu(eb, &key, i);
2435 /* inode keys are done during the first stage */
2436 if (key.type == BTRFS_INODE_ITEM_KEY &&
2437 wc->stage == LOG_WALK_REPLAY_INODES) {
2438 struct btrfs_inode_item *inode_item;
2441 inode_item = btrfs_item_ptr(eb, i,
2442 struct btrfs_inode_item);
2443 ret = replay_xattr_deletes(wc->trans, root, log,
2444 path, key.objectid);
2447 mode = btrfs_inode_mode(eb, inode_item);
2448 if (S_ISDIR(mode)) {
2449 ret = replay_dir_deletes(wc->trans,
2450 root, log, path, key.objectid, 0);
2454 ret = overwrite_item(wc->trans, root, path,
2459 /* for regular files, make sure corresponding
2460 * orphan item exist. extents past the new EOF
2461 * will be truncated later by orphan cleanup.
2463 if (S_ISREG(mode)) {
2464 ret = insert_orphan_item(wc->trans, root,
2470 ret = link_to_fixup_dir(wc->trans, root,
2471 path, key.objectid);
2476 if (key.type == BTRFS_DIR_INDEX_KEY &&
2477 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2478 ret = replay_one_dir_item(wc->trans, root, path,
2484 if (wc->stage < LOG_WALK_REPLAY_ALL)
2487 /* these keys are simply copied */
2488 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2489 ret = overwrite_item(wc->trans, root, path,
2493 } else if (key.type == BTRFS_INODE_REF_KEY ||
2494 key.type == BTRFS_INODE_EXTREF_KEY) {
2495 ret = add_inode_ref(wc->trans, root, log, path,
2497 if (ret && ret != -ENOENT)
2500 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2501 ret = replay_one_extent(wc->trans, root, path,
2505 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2506 ret = replay_one_dir_item(wc->trans, root, path,
2512 btrfs_free_path(path);
2516 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2517 struct btrfs_root *root,
2518 struct btrfs_path *path, int *level,
2519 struct walk_control *wc)
2521 struct btrfs_fs_info *fs_info = root->fs_info;
2525 struct extent_buffer *next;
2526 struct extent_buffer *cur;
2527 struct extent_buffer *parent;
2531 WARN_ON(*level < 0);
2532 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2534 while (*level > 0) {
2535 struct btrfs_key first_key;
2537 WARN_ON(*level < 0);
2538 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2539 cur = path->nodes[*level];
2541 WARN_ON(btrfs_header_level(cur) != *level);
2543 if (path->slots[*level] >=
2544 btrfs_header_nritems(cur))
2547 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2548 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2549 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2550 blocksize = fs_info->nodesize;
2552 parent = path->nodes[*level];
2553 root_owner = btrfs_header_owner(parent);
2555 next = btrfs_find_create_tree_block(fs_info, bytenr);
2557 return PTR_ERR(next);
2560 ret = wc->process_func(root, next, wc, ptr_gen,
2563 free_extent_buffer(next);
2567 path->slots[*level]++;
2569 ret = btrfs_read_buffer(next, ptr_gen,
2570 *level - 1, &first_key);
2572 free_extent_buffer(next);
2577 btrfs_tree_lock(next);
2578 btrfs_set_lock_blocking(next);
2579 clean_tree_block(fs_info, next);
2580 btrfs_wait_tree_block_writeback(next);
2581 btrfs_tree_unlock(next);
2583 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2584 clear_extent_buffer_dirty(next);
2587 WARN_ON(root_owner !=
2588 BTRFS_TREE_LOG_OBJECTID);
2589 ret = btrfs_free_and_pin_reserved_extent(
2593 free_extent_buffer(next);
2597 free_extent_buffer(next);
2600 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2602 free_extent_buffer(next);
2606 WARN_ON(*level <= 0);
2607 if (path->nodes[*level-1])
2608 free_extent_buffer(path->nodes[*level-1]);
2609 path->nodes[*level-1] = next;
2610 *level = btrfs_header_level(next);
2611 path->slots[*level] = 0;
2614 WARN_ON(*level < 0);
2615 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2617 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2623 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *root,
2625 struct btrfs_path *path, int *level,
2626 struct walk_control *wc)
2628 struct btrfs_fs_info *fs_info = root->fs_info;
2634 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2635 slot = path->slots[i];
2636 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2639 WARN_ON(*level == 0);
2642 struct extent_buffer *parent;
2643 if (path->nodes[*level] == root->node)
2644 parent = path->nodes[*level];
2646 parent = path->nodes[*level + 1];
2648 root_owner = btrfs_header_owner(parent);
2649 ret = wc->process_func(root, path->nodes[*level], wc,
2650 btrfs_header_generation(path->nodes[*level]),
2656 struct extent_buffer *next;
2658 next = path->nodes[*level];
2661 btrfs_tree_lock(next);
2662 btrfs_set_lock_blocking(next);
2663 clean_tree_block(fs_info, next);
2664 btrfs_wait_tree_block_writeback(next);
2665 btrfs_tree_unlock(next);
2667 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2668 clear_extent_buffer_dirty(next);
2671 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2672 ret = btrfs_free_and_pin_reserved_extent(
2674 path->nodes[*level]->start,
2675 path->nodes[*level]->len);
2679 free_extent_buffer(path->nodes[*level]);
2680 path->nodes[*level] = NULL;
2688 * drop the reference count on the tree rooted at 'snap'. This traverses
2689 * the tree freeing any blocks that have a ref count of zero after being
2692 static int walk_log_tree(struct btrfs_trans_handle *trans,
2693 struct btrfs_root *log, struct walk_control *wc)
2695 struct btrfs_fs_info *fs_info = log->fs_info;
2699 struct btrfs_path *path;
2702 path = btrfs_alloc_path();
2706 level = btrfs_header_level(log->node);
2708 path->nodes[level] = log->node;
2709 extent_buffer_get(log->node);
2710 path->slots[level] = 0;
2713 wret = walk_down_log_tree(trans, log, path, &level, wc);
2721 wret = walk_up_log_tree(trans, log, path, &level, wc);
2730 /* was the root node processed? if not, catch it here */
2731 if (path->nodes[orig_level]) {
2732 ret = wc->process_func(log, path->nodes[orig_level], wc,
2733 btrfs_header_generation(path->nodes[orig_level]),
2738 struct extent_buffer *next;
2740 next = path->nodes[orig_level];
2743 btrfs_tree_lock(next);
2744 btrfs_set_lock_blocking(next);
2745 clean_tree_block(fs_info, next);
2746 btrfs_wait_tree_block_writeback(next);
2747 btrfs_tree_unlock(next);
2749 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2750 clear_extent_buffer_dirty(next);
2753 WARN_ON(log->root_key.objectid !=
2754 BTRFS_TREE_LOG_OBJECTID);
2755 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2756 next->start, next->len);
2763 btrfs_free_path(path);
2768 * helper function to update the item for a given subvolumes log root
2769 * in the tree of log roots
2771 static int update_log_root(struct btrfs_trans_handle *trans,
2772 struct btrfs_root *log)
2774 struct btrfs_fs_info *fs_info = log->fs_info;
2777 if (log->log_transid == 1) {
2778 /* insert root item on the first sync */
2779 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2780 &log->root_key, &log->root_item);
2782 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2783 &log->root_key, &log->root_item);
2788 static void wait_log_commit(struct btrfs_root *root, int transid)
2791 int index = transid % 2;
2794 * we only allow two pending log transactions at a time,
2795 * so we know that if ours is more than 2 older than the
2796 * current transaction, we're done
2799 prepare_to_wait(&root->log_commit_wait[index],
2800 &wait, TASK_UNINTERRUPTIBLE);
2802 if (!(root->log_transid_committed < transid &&
2803 atomic_read(&root->log_commit[index])))
2806 mutex_unlock(&root->log_mutex);
2808 mutex_lock(&root->log_mutex);
2810 finish_wait(&root->log_commit_wait[index], &wait);
2813 static void wait_for_writer(struct btrfs_root *root)
2818 prepare_to_wait(&root->log_writer_wait, &wait,
2819 TASK_UNINTERRUPTIBLE);
2820 if (!atomic_read(&root->log_writers))
2823 mutex_unlock(&root->log_mutex);
2825 mutex_lock(&root->log_mutex);
2827 finish_wait(&root->log_writer_wait, &wait);
2830 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2831 struct btrfs_log_ctx *ctx)
2836 mutex_lock(&root->log_mutex);
2837 list_del_init(&ctx->list);
2838 mutex_unlock(&root->log_mutex);
2842 * Invoked in log mutex context, or be sure there is no other task which
2843 * can access the list.
2845 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2846 int index, int error)
2848 struct btrfs_log_ctx *ctx;
2849 struct btrfs_log_ctx *safe;
2851 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2852 list_del_init(&ctx->list);
2853 ctx->log_ret = error;
2856 INIT_LIST_HEAD(&root->log_ctxs[index]);
2860 * btrfs_sync_log does sends a given tree log down to the disk and
2861 * updates the super blocks to record it. When this call is done,
2862 * you know that any inodes previously logged are safely on disk only
2865 * Any other return value means you need to call btrfs_commit_transaction.
2866 * Some of the edge cases for fsyncing directories that have had unlinks
2867 * or renames done in the past mean that sometimes the only safe
2868 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2869 * that has happened.
2871 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2872 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2878 struct btrfs_fs_info *fs_info = root->fs_info;
2879 struct btrfs_root *log = root->log_root;
2880 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2881 int log_transid = 0;
2882 struct btrfs_log_ctx root_log_ctx;
2883 struct blk_plug plug;
2885 mutex_lock(&root->log_mutex);
2886 log_transid = ctx->log_transid;
2887 if (root->log_transid_committed >= log_transid) {
2888 mutex_unlock(&root->log_mutex);
2889 return ctx->log_ret;
2892 index1 = log_transid % 2;
2893 if (atomic_read(&root->log_commit[index1])) {
2894 wait_log_commit(root, log_transid);
2895 mutex_unlock(&root->log_mutex);
2896 return ctx->log_ret;
2898 ASSERT(log_transid == root->log_transid);
2899 atomic_set(&root->log_commit[index1], 1);
2901 /* wait for previous tree log sync to complete */
2902 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2903 wait_log_commit(root, log_transid - 1);
2906 int batch = atomic_read(&root->log_batch);
2907 /* when we're on an ssd, just kick the log commit out */
2908 if (!btrfs_test_opt(fs_info, SSD) &&
2909 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2910 mutex_unlock(&root->log_mutex);
2911 schedule_timeout_uninterruptible(1);
2912 mutex_lock(&root->log_mutex);
2914 wait_for_writer(root);
2915 if (batch == atomic_read(&root->log_batch))
2919 /* bail out if we need to do a full commit */
2920 if (btrfs_need_log_full_commit(fs_info, trans)) {
2922 btrfs_free_logged_extents(log, log_transid);
2923 mutex_unlock(&root->log_mutex);
2927 if (log_transid % 2 == 0)
2928 mark = EXTENT_DIRTY;
2932 /* we start IO on all the marked extents here, but we don't actually
2933 * wait for them until later.
2935 blk_start_plug(&plug);
2936 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2938 blk_finish_plug(&plug);
2939 btrfs_abort_transaction(trans, ret);
2940 btrfs_free_logged_extents(log, log_transid);
2941 btrfs_set_log_full_commit(fs_info, trans);
2942 mutex_unlock(&root->log_mutex);
2946 btrfs_set_root_node(&log->root_item, log->node);
2948 root->log_transid++;
2949 log->log_transid = root->log_transid;
2950 root->log_start_pid = 0;
2952 * IO has been started, blocks of the log tree have WRITTEN flag set
2953 * in their headers. new modifications of the log will be written to
2954 * new positions. so it's safe to allow log writers to go in.
2956 mutex_unlock(&root->log_mutex);
2958 btrfs_init_log_ctx(&root_log_ctx, NULL);
2960 mutex_lock(&log_root_tree->log_mutex);
2961 atomic_inc(&log_root_tree->log_batch);
2962 atomic_inc(&log_root_tree->log_writers);
2964 index2 = log_root_tree->log_transid % 2;
2965 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2966 root_log_ctx.log_transid = log_root_tree->log_transid;
2968 mutex_unlock(&log_root_tree->log_mutex);
2970 ret = update_log_root(trans, log);
2972 mutex_lock(&log_root_tree->log_mutex);
2973 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2975 * Implicit memory barrier after atomic_dec_and_test
2977 if (waitqueue_active(&log_root_tree->log_writer_wait))
2978 wake_up(&log_root_tree->log_writer_wait);
2982 if (!list_empty(&root_log_ctx.list))
2983 list_del_init(&root_log_ctx.list);
2985 blk_finish_plug(&plug);
2986 btrfs_set_log_full_commit(fs_info, trans);
2988 if (ret != -ENOSPC) {
2989 btrfs_abort_transaction(trans, ret);
2990 mutex_unlock(&log_root_tree->log_mutex);
2993 btrfs_wait_tree_log_extents(log, mark);
2994 btrfs_free_logged_extents(log, log_transid);
2995 mutex_unlock(&log_root_tree->log_mutex);
3000 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3001 blk_finish_plug(&plug);
3002 list_del_init(&root_log_ctx.list);
3003 mutex_unlock(&log_root_tree->log_mutex);
3004 ret = root_log_ctx.log_ret;
3008 index2 = root_log_ctx.log_transid % 2;
3009 if (atomic_read(&log_root_tree->log_commit[index2])) {
3010 blk_finish_plug(&plug);
3011 ret = btrfs_wait_tree_log_extents(log, mark);
3012 btrfs_wait_logged_extents(trans, log, log_transid);
3013 wait_log_commit(log_root_tree,
3014 root_log_ctx.log_transid);
3015 mutex_unlock(&log_root_tree->log_mutex);
3017 ret = root_log_ctx.log_ret;
3020 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3021 atomic_set(&log_root_tree->log_commit[index2], 1);
3023 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3024 wait_log_commit(log_root_tree,
3025 root_log_ctx.log_transid - 1);
3028 wait_for_writer(log_root_tree);
3031 * now that we've moved on to the tree of log tree roots,
3032 * check the full commit flag again
3034 if (btrfs_need_log_full_commit(fs_info, trans)) {
3035 blk_finish_plug(&plug);
3036 btrfs_wait_tree_log_extents(log, mark);
3037 btrfs_free_logged_extents(log, log_transid);
3038 mutex_unlock(&log_root_tree->log_mutex);
3040 goto out_wake_log_root;
3043 ret = btrfs_write_marked_extents(fs_info,
3044 &log_root_tree->dirty_log_pages,
3045 EXTENT_DIRTY | EXTENT_NEW);
3046 blk_finish_plug(&plug);
3048 btrfs_set_log_full_commit(fs_info, trans);
3049 btrfs_abort_transaction(trans, ret);
3050 btrfs_free_logged_extents(log, log_transid);
3051 mutex_unlock(&log_root_tree->log_mutex);
3052 goto out_wake_log_root;
3054 ret = btrfs_wait_tree_log_extents(log, mark);
3056 ret = btrfs_wait_tree_log_extents(log_root_tree,
3057 EXTENT_NEW | EXTENT_DIRTY);
3059 btrfs_set_log_full_commit(fs_info, trans);
3060 btrfs_free_logged_extents(log, log_transid);
3061 mutex_unlock(&log_root_tree->log_mutex);
3062 goto out_wake_log_root;
3064 btrfs_wait_logged_extents(trans, log, log_transid);
3066 btrfs_set_super_log_root(fs_info->super_for_commit,
3067 log_root_tree->node->start);
3068 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3069 btrfs_header_level(log_root_tree->node));
3071 log_root_tree->log_transid++;
3072 mutex_unlock(&log_root_tree->log_mutex);
3075 * nobody else is going to jump in and write the the ctree
3076 * super here because the log_commit atomic below is protecting
3077 * us. We must be called with a transaction handle pinning
3078 * the running transaction open, so a full commit can't hop
3079 * in and cause problems either.
3081 ret = write_all_supers(fs_info, 1);
3083 btrfs_set_log_full_commit(fs_info, trans);
3084 btrfs_abort_transaction(trans, ret);
3085 goto out_wake_log_root;
3088 mutex_lock(&root->log_mutex);
3089 if (root->last_log_commit < log_transid)
3090 root->last_log_commit = log_transid;
3091 mutex_unlock(&root->log_mutex);
3094 mutex_lock(&log_root_tree->log_mutex);
3095 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3097 log_root_tree->log_transid_committed++;
3098 atomic_set(&log_root_tree->log_commit[index2], 0);
3099 mutex_unlock(&log_root_tree->log_mutex);
3102 * The barrier before waitqueue_active is implied by mutex_unlock
3104 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
3105 wake_up(&log_root_tree->log_commit_wait[index2]);
3107 mutex_lock(&root->log_mutex);
3108 btrfs_remove_all_log_ctxs(root, index1, ret);
3109 root->log_transid_committed++;
3110 atomic_set(&root->log_commit[index1], 0);
3111 mutex_unlock(&root->log_mutex);
3114 * The barrier before waitqueue_active is implied by mutex_unlock
3116 if (waitqueue_active(&root->log_commit_wait[index1]))
3117 wake_up(&root->log_commit_wait[index1]);
3121 static void free_log_tree(struct btrfs_trans_handle *trans,
3122 struct btrfs_root *log)
3127 struct walk_control wc = {
3129 .process_func = process_one_buffer
3132 ret = walk_log_tree(trans, log, &wc);
3133 /* I don't think this can happen but just in case */
3135 btrfs_abort_transaction(trans, ret);
3138 ret = find_first_extent_bit(&log->dirty_log_pages,
3140 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT,
3145 clear_extent_bits(&log->dirty_log_pages, start, end,
3146 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3150 * We may have short-circuited the log tree with the full commit logic
3151 * and left ordered extents on our list, so clear these out to keep us
3152 * from leaking inodes and memory.
3154 btrfs_free_logged_extents(log, 0);
3155 btrfs_free_logged_extents(log, 1);
3157 free_extent_buffer(log->node);
3162 * free all the extents used by the tree log. This should be called
3163 * at commit time of the full transaction
3165 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3167 if (root->log_root) {
3168 free_log_tree(trans, root->log_root);
3169 root->log_root = NULL;
3174 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3175 struct btrfs_fs_info *fs_info)
3177 if (fs_info->log_root_tree) {
3178 free_log_tree(trans, fs_info->log_root_tree);
3179 fs_info->log_root_tree = NULL;
3185 * If both a file and directory are logged, and unlinks or renames are
3186 * mixed in, we have a few interesting corners:
3188 * create file X in dir Y
3189 * link file X to X.link in dir Y
3191 * unlink file X but leave X.link
3194 * After a crash we would expect only X.link to exist. But file X
3195 * didn't get fsync'd again so the log has back refs for X and X.link.
3197 * We solve this by removing directory entries and inode backrefs from the
3198 * log when a file that was logged in the current transaction is
3199 * unlinked. Any later fsync will include the updated log entries, and
3200 * we'll be able to reconstruct the proper directory items from backrefs.
3202 * This optimizations allows us to avoid relogging the entire inode
3203 * or the entire directory.
3205 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3206 struct btrfs_root *root,
3207 const char *name, int name_len,
3208 struct btrfs_inode *dir, u64 index)
3210 struct btrfs_root *log;
3211 struct btrfs_dir_item *di;
3212 struct btrfs_path *path;
3216 u64 dir_ino = btrfs_ino(dir);
3218 if (dir->logged_trans < trans->transid)
3221 ret = join_running_log_trans(root);
3225 mutex_lock(&dir->log_mutex);
3227 log = root->log_root;
3228 path = btrfs_alloc_path();
3234 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3235 name, name_len, -1);
3241 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3242 bytes_del += name_len;
3248 btrfs_release_path(path);
3249 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3250 index, name, name_len, -1);
3256 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3257 bytes_del += name_len;
3264 /* update the directory size in the log to reflect the names
3268 struct btrfs_key key;
3270 key.objectid = dir_ino;
3272 key.type = BTRFS_INODE_ITEM_KEY;
3273 btrfs_release_path(path);
3275 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3281 struct btrfs_inode_item *item;
3284 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3285 struct btrfs_inode_item);
3286 i_size = btrfs_inode_size(path->nodes[0], item);
3287 if (i_size > bytes_del)
3288 i_size -= bytes_del;
3291 btrfs_set_inode_size(path->nodes[0], item, i_size);
3292 btrfs_mark_buffer_dirty(path->nodes[0]);
3295 btrfs_release_path(path);
3298 btrfs_free_path(path);
3300 mutex_unlock(&dir->log_mutex);
3301 if (ret == -ENOSPC) {
3302 btrfs_set_log_full_commit(root->fs_info, trans);
3305 btrfs_abort_transaction(trans, ret);
3307 btrfs_end_log_trans(root);
3312 /* see comments for btrfs_del_dir_entries_in_log */
3313 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 const char *name, int name_len,
3316 struct btrfs_inode *inode, u64 dirid)
3318 struct btrfs_fs_info *fs_info = root->fs_info;
3319 struct btrfs_root *log;
3323 if (inode->logged_trans < trans->transid)
3326 ret = join_running_log_trans(root);
3329 log = root->log_root;
3330 mutex_lock(&inode->log_mutex);
3332 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3334 mutex_unlock(&inode->log_mutex);
3335 if (ret == -ENOSPC) {
3336 btrfs_set_log_full_commit(fs_info, trans);
3338 } else if (ret < 0 && ret != -ENOENT)
3339 btrfs_abort_transaction(trans, ret);
3340 btrfs_end_log_trans(root);
3346 * creates a range item in the log for 'dirid'. first_offset and
3347 * last_offset tell us which parts of the key space the log should
3348 * be considered authoritative for.
3350 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3351 struct btrfs_root *log,
3352 struct btrfs_path *path,
3353 int key_type, u64 dirid,
3354 u64 first_offset, u64 last_offset)
3357 struct btrfs_key key;
3358 struct btrfs_dir_log_item *item;
3360 key.objectid = dirid;
3361 key.offset = first_offset;
3362 if (key_type == BTRFS_DIR_ITEM_KEY)
3363 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3365 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3366 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3370 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3371 struct btrfs_dir_log_item);
3372 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3373 btrfs_mark_buffer_dirty(path->nodes[0]);
3374 btrfs_release_path(path);
3379 * log all the items included in the current transaction for a given
3380 * directory. This also creates the range items in the log tree required
3381 * to replay anything deleted before the fsync
3383 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3384 struct btrfs_root *root, struct btrfs_inode *inode,
3385 struct btrfs_path *path,
3386 struct btrfs_path *dst_path, int key_type,
3387 struct btrfs_log_ctx *ctx,
3388 u64 min_offset, u64 *last_offset_ret)
3390 struct btrfs_key min_key;
3391 struct btrfs_root *log = root->log_root;
3392 struct extent_buffer *src;
3397 u64 first_offset = min_offset;
3398 u64 last_offset = (u64)-1;
3399 u64 ino = btrfs_ino(inode);
3401 log = root->log_root;
3403 min_key.objectid = ino;
3404 min_key.type = key_type;
3405 min_key.offset = min_offset;
3407 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3410 * we didn't find anything from this transaction, see if there
3411 * is anything at all
3413 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3414 min_key.objectid = ino;
3415 min_key.type = key_type;
3416 min_key.offset = (u64)-1;
3417 btrfs_release_path(path);
3418 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3420 btrfs_release_path(path);
3423 ret = btrfs_previous_item(root, path, ino, key_type);
3425 /* if ret == 0 there are items for this type,
3426 * create a range to tell us the last key of this type.
3427 * otherwise, there are no items in this directory after
3428 * *min_offset, and we create a range to indicate that.
3431 struct btrfs_key tmp;
3432 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3434 if (key_type == tmp.type)
3435 first_offset = max(min_offset, tmp.offset) + 1;
3440 /* go backward to find any previous key */
3441 ret = btrfs_previous_item(root, path, ino, key_type);
3443 struct btrfs_key tmp;
3444 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3445 if (key_type == tmp.type) {
3446 first_offset = tmp.offset;
3447 ret = overwrite_item(trans, log, dst_path,
3448 path->nodes[0], path->slots[0],
3456 btrfs_release_path(path);
3458 /* find the first key from this transaction again */
3459 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3460 if (WARN_ON(ret != 0))
3464 * we have a block from this transaction, log every item in it
3465 * from our directory
3468 struct btrfs_key tmp;
3469 src = path->nodes[0];
3470 nritems = btrfs_header_nritems(src);
3471 for (i = path->slots[0]; i < nritems; i++) {
3472 struct btrfs_dir_item *di;
3474 btrfs_item_key_to_cpu(src, &min_key, i);
3476 if (min_key.objectid != ino || min_key.type != key_type)
3478 ret = overwrite_item(trans, log, dst_path, src, i,
3486 * We must make sure that when we log a directory entry,
3487 * the corresponding inode, after log replay, has a
3488 * matching link count. For example:
3494 * xfs_io -c "fsync" mydir
3496 * <mount fs and log replay>
3498 * Would result in a fsync log that when replayed, our
3499 * file inode would have a link count of 1, but we get
3500 * two directory entries pointing to the same inode.
3501 * After removing one of the names, it would not be
3502 * possible to remove the other name, which resulted
3503 * always in stale file handle errors, and would not
3504 * be possible to rmdir the parent directory, since
3505 * its i_size could never decrement to the value
3506 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3508 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3509 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3511 (btrfs_dir_transid(src, di) == trans->transid ||
3512 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3513 tmp.type != BTRFS_ROOT_ITEM_KEY)
3514 ctx->log_new_dentries = true;
3516 path->slots[0] = nritems;
3519 * look ahead to the next item and see if it is also
3520 * from this directory and from this transaction
3522 ret = btrfs_next_leaf(root, path);
3524 last_offset = (u64)-1;
3527 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3528 if (tmp.objectid != ino || tmp.type != key_type) {
3529 last_offset = (u64)-1;
3532 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3533 ret = overwrite_item(trans, log, dst_path,
3534 path->nodes[0], path->slots[0],
3539 last_offset = tmp.offset;
3544 btrfs_release_path(path);
3545 btrfs_release_path(dst_path);
3548 *last_offset_ret = last_offset;
3550 * insert the log range keys to indicate where the log
3553 ret = insert_dir_log_key(trans, log, path, key_type,
3554 ino, first_offset, last_offset);
3562 * logging directories is very similar to logging inodes, We find all the items
3563 * from the current transaction and write them to the log.
3565 * The recovery code scans the directory in the subvolume, and if it finds a
3566 * key in the range logged that is not present in the log tree, then it means
3567 * that dir entry was unlinked during the transaction.
3569 * In order for that scan to work, we must include one key smaller than
3570 * the smallest logged by this transaction and one key larger than the largest
3571 * key logged by this transaction.
3573 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3574 struct btrfs_root *root, struct btrfs_inode *inode,
3575 struct btrfs_path *path,
3576 struct btrfs_path *dst_path,
3577 struct btrfs_log_ctx *ctx)
3582 int key_type = BTRFS_DIR_ITEM_KEY;
3588 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3589 ctx, min_key, &max_key);
3592 if (max_key == (u64)-1)
3594 min_key = max_key + 1;
3597 if (key_type == BTRFS_DIR_ITEM_KEY) {
3598 key_type = BTRFS_DIR_INDEX_KEY;
3605 * a helper function to drop items from the log before we relog an
3606 * inode. max_key_type indicates the highest item type to remove.
3607 * This cannot be run for file data extents because it does not
3608 * free the extents they point to.
3610 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3611 struct btrfs_root *log,
3612 struct btrfs_path *path,
3613 u64 objectid, int max_key_type)
3616 struct btrfs_key key;
3617 struct btrfs_key found_key;
3620 key.objectid = objectid;
3621 key.type = max_key_type;
3622 key.offset = (u64)-1;
3625 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3626 BUG_ON(ret == 0); /* Logic error */
3630 if (path->slots[0] == 0)
3634 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3637 if (found_key.objectid != objectid)
3640 found_key.offset = 0;
3642 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3645 ret = btrfs_del_items(trans, log, path, start_slot,
3646 path->slots[0] - start_slot + 1);
3648 * If start slot isn't 0 then we don't need to re-search, we've
3649 * found the last guy with the objectid in this tree.
3651 if (ret || start_slot != 0)
3653 btrfs_release_path(path);
3655 btrfs_release_path(path);
3661 static void fill_inode_item(struct btrfs_trans_handle *trans,
3662 struct extent_buffer *leaf,
3663 struct btrfs_inode_item *item,
3664 struct inode *inode, int log_inode_only,
3667 struct btrfs_map_token token;
3669 btrfs_init_map_token(&token);
3671 if (log_inode_only) {
3672 /* set the generation to zero so the recover code
3673 * can tell the difference between an logging
3674 * just to say 'this inode exists' and a logging
3675 * to say 'update this inode with these values'
3677 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3678 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3680 btrfs_set_token_inode_generation(leaf, item,
3681 BTRFS_I(inode)->generation,
3683 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3686 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3687 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3688 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3689 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3691 btrfs_set_token_timespec_sec(leaf, &item->atime,
3692 inode->i_atime.tv_sec, &token);
3693 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3694 inode->i_atime.tv_nsec, &token);
3696 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3697 inode->i_mtime.tv_sec, &token);
3698 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3699 inode->i_mtime.tv_nsec, &token);
3701 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3702 inode->i_ctime.tv_sec, &token);
3703 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3704 inode->i_ctime.tv_nsec, &token);
3706 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3709 btrfs_set_token_inode_sequence(leaf, item,
3710 inode_peek_iversion(inode), &token);
3711 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3712 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3713 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3714 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3717 static int log_inode_item(struct btrfs_trans_handle *trans,
3718 struct btrfs_root *log, struct btrfs_path *path,
3719 struct btrfs_inode *inode)
3721 struct btrfs_inode_item *inode_item;
3724 ret = btrfs_insert_empty_item(trans, log, path,
3725 &inode->location, sizeof(*inode_item));
3726 if (ret && ret != -EEXIST)
3728 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3729 struct btrfs_inode_item);
3730 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3732 btrfs_release_path(path);
3736 static noinline int copy_items(struct btrfs_trans_handle *trans,
3737 struct btrfs_inode *inode,
3738 struct btrfs_path *dst_path,
3739 struct btrfs_path *src_path, u64 *last_extent,
3740 int start_slot, int nr, int inode_only,
3743 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3744 unsigned long src_offset;
3745 unsigned long dst_offset;
3746 struct btrfs_root *log = inode->root->log_root;
3747 struct btrfs_file_extent_item *extent;
3748 struct btrfs_inode_item *inode_item;
3749 struct extent_buffer *src = src_path->nodes[0];
3750 struct btrfs_key first_key, last_key, key;
3752 struct btrfs_key *ins_keys;
3756 struct list_head ordered_sums;
3757 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3758 bool has_extents = false;
3759 bool need_find_last_extent = true;
3762 INIT_LIST_HEAD(&ordered_sums);
3764 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3765 nr * sizeof(u32), GFP_NOFS);
3769 first_key.objectid = (u64)-1;
3771 ins_sizes = (u32 *)ins_data;
3772 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3774 for (i = 0; i < nr; i++) {
3775 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3776 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3778 ret = btrfs_insert_empty_items(trans, log, dst_path,
3779 ins_keys, ins_sizes, nr);
3785 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3786 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3787 dst_path->slots[0]);
3789 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3792 last_key = ins_keys[i];
3794 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3795 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3797 struct btrfs_inode_item);
3798 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3800 inode_only == LOG_INODE_EXISTS,
3803 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3804 src_offset, ins_sizes[i]);
3808 * We set need_find_last_extent here in case we know we were
3809 * processing other items and then walk into the first extent in
3810 * the inode. If we don't hit an extent then nothing changes,
3811 * we'll do the last search the next time around.
3813 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3815 if (first_key.objectid == (u64)-1)
3816 first_key = ins_keys[i];
3818 need_find_last_extent = false;
3821 /* take a reference on file data extents so that truncates
3822 * or deletes of this inode don't have to relog the inode
3825 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3828 extent = btrfs_item_ptr(src, start_slot + i,
3829 struct btrfs_file_extent_item);
3831 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3834 found_type = btrfs_file_extent_type(src, extent);
3835 if (found_type == BTRFS_FILE_EXTENT_REG) {
3837 ds = btrfs_file_extent_disk_bytenr(src,
3839 /* ds == 0 is a hole */
3843 dl = btrfs_file_extent_disk_num_bytes(src,
3845 cs = btrfs_file_extent_offset(src, extent);
3846 cl = btrfs_file_extent_num_bytes(src,
3848 if (btrfs_file_extent_compression(src,
3854 ret = btrfs_lookup_csums_range(
3856 ds + cs, ds + cs + cl - 1,
3859 btrfs_release_path(dst_path);
3867 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3868 btrfs_release_path(dst_path);
3872 * we have to do this after the loop above to avoid changing the
3873 * log tree while trying to change the log tree.
3876 while (!list_empty(&ordered_sums)) {
3877 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3878 struct btrfs_ordered_sum,
3881 ret = btrfs_csum_file_blocks(trans, log, sums);
3882 list_del(&sums->list);
3889 if (need_find_last_extent && *last_extent == first_key.offset) {
3891 * We don't have any leafs between our current one and the one
3892 * we processed before that can have file extent items for our
3893 * inode (and have a generation number smaller than our current
3896 need_find_last_extent = false;
3900 * Because we use btrfs_search_forward we could skip leaves that were
3901 * not modified and then assume *last_extent is valid when it really
3902 * isn't. So back up to the previous leaf and read the end of the last
3903 * extent before we go and fill in holes.
3905 if (need_find_last_extent) {
3908 ret = btrfs_prev_leaf(inode->root, src_path);
3913 if (src_path->slots[0])
3914 src_path->slots[0]--;
3915 src = src_path->nodes[0];
3916 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3917 if (key.objectid != btrfs_ino(inode) ||
3918 key.type != BTRFS_EXTENT_DATA_KEY)
3920 extent = btrfs_item_ptr(src, src_path->slots[0],
3921 struct btrfs_file_extent_item);
3922 if (btrfs_file_extent_type(src, extent) ==
3923 BTRFS_FILE_EXTENT_INLINE) {
3924 len = btrfs_file_extent_inline_len(src,
3927 *last_extent = ALIGN(key.offset + len,
3928 fs_info->sectorsize);
3930 len = btrfs_file_extent_num_bytes(src, extent);
3931 *last_extent = key.offset + len;
3935 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3936 * things could have happened
3938 * 1) A merge could have happened, so we could currently be on a leaf
3939 * that holds what we were copying in the first place.
3940 * 2) A split could have happened, and now not all of the items we want
3941 * are on the same leaf.
3943 * So we need to adjust how we search for holes, we need to drop the
3944 * path and re-search for the first extent key we found, and then walk
3945 * forward until we hit the last one we copied.
3947 if (need_find_last_extent) {
3948 /* btrfs_prev_leaf could return 1 without releasing the path */
3949 btrfs_release_path(src_path);
3950 ret = btrfs_search_slot(NULL, inode->root, &first_key,
3955 src = src_path->nodes[0];
3956 i = src_path->slots[0];
3962 * Ok so here we need to go through and fill in any holes we may have
3963 * to make sure that holes are punched for those areas in case they had
3964 * extents previously.
3970 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3971 ret = btrfs_next_leaf(inode->root, src_path);
3975 src = src_path->nodes[0];
3977 need_find_last_extent = true;
3980 btrfs_item_key_to_cpu(src, &key, i);
3981 if (!btrfs_comp_cpu_keys(&key, &last_key))
3983 if (key.objectid != btrfs_ino(inode) ||
3984 key.type != BTRFS_EXTENT_DATA_KEY) {
3988 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3989 if (btrfs_file_extent_type(src, extent) ==
3990 BTRFS_FILE_EXTENT_INLINE) {
3991 len = btrfs_file_extent_inline_len(src, i, extent);
3992 extent_end = ALIGN(key.offset + len,
3993 fs_info->sectorsize);
3995 len = btrfs_file_extent_num_bytes(src, extent);
3996 extent_end = key.offset + len;
4000 if (*last_extent == key.offset) {
4001 *last_extent = extent_end;
4004 offset = *last_extent;
4005 len = key.offset - *last_extent;
4006 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4007 offset, 0, 0, len, 0, len, 0, 0, 0);
4010 *last_extent = extent_end;
4014 * Check if there is a hole between the last extent found in our leaf
4015 * and the first extent in the next leaf. If there is one, we need to
4016 * log an explicit hole so that at replay time we can punch the hole.
4019 key.objectid == btrfs_ino(inode) &&
4020 key.type == BTRFS_EXTENT_DATA_KEY &&
4021 i == btrfs_header_nritems(src_path->nodes[0])) {
4022 ret = btrfs_next_leaf(inode->root, src_path);
4023 need_find_last_extent = true;
4026 } else if (ret == 0) {
4027 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4028 src_path->slots[0]);
4029 if (key.objectid == btrfs_ino(inode) &&
4030 key.type == BTRFS_EXTENT_DATA_KEY &&
4031 *last_extent < key.offset) {
4032 const u64 len = key.offset - *last_extent;
4034 ret = btrfs_insert_file_extent(trans, log,
4043 * Need to let the callers know we dropped the path so they should
4046 if (!ret && need_find_last_extent)
4051 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4053 struct extent_map *em1, *em2;
4055 em1 = list_entry(a, struct extent_map, list);
4056 em2 = list_entry(b, struct extent_map, list);
4058 if (em1->start < em2->start)
4060 else if (em1->start > em2->start)
4065 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
4066 struct inode *inode,
4067 struct btrfs_root *root,
4068 const struct extent_map *em,
4069 const struct list_head *logged_list,
4070 bool *ordered_io_error)
4072 struct btrfs_fs_info *fs_info = root->fs_info;
4073 struct btrfs_ordered_extent *ordered;
4074 struct btrfs_root *log = root->log_root;
4075 u64 mod_start = em->mod_start;
4076 u64 mod_len = em->mod_len;
4077 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
4080 LIST_HEAD(ordered_sums);
4083 *ordered_io_error = false;
4085 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4086 em->block_start == EXTENT_MAP_HOLE)
4090 * Wait far any ordered extent that covers our extent map. If it
4091 * finishes without an error, first check and see if our csums are on
4092 * our outstanding ordered extents.
4094 list_for_each_entry(ordered, logged_list, log_list) {
4095 struct btrfs_ordered_sum *sum;
4100 if (ordered->file_offset + ordered->len <= mod_start ||
4101 mod_start + mod_len <= ordered->file_offset)
4104 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
4105 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
4106 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
4107 const u64 start = ordered->file_offset;
4108 const u64 end = ordered->file_offset + ordered->len - 1;
4110 WARN_ON(ordered->inode != inode);
4111 filemap_fdatawrite_range(inode->i_mapping, start, end);
4114 wait_event(ordered->wait,
4115 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
4116 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
4118 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
4120 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
4121 * i_mapping flags, so that the next fsync won't get
4122 * an outdated io error too.
4124 filemap_check_errors(inode->i_mapping);
4125 *ordered_io_error = true;
4129 * We are going to copy all the csums on this ordered extent, so
4130 * go ahead and adjust mod_start and mod_len in case this
4131 * ordered extent has already been logged.
4133 if (ordered->file_offset > mod_start) {
4134 if (ordered->file_offset + ordered->len >=
4135 mod_start + mod_len)
4136 mod_len = ordered->file_offset - mod_start;
4138 * If we have this case
4140 * |--------- logged extent ---------|
4141 * |----- ordered extent ----|
4143 * Just don't mess with mod_start and mod_len, we'll
4144 * just end up logging more csums than we need and it
4148 if (ordered->file_offset + ordered->len <
4149 mod_start + mod_len) {
4150 mod_len = (mod_start + mod_len) -
4151 (ordered->file_offset + ordered->len);
4152 mod_start = ordered->file_offset +
4163 * To keep us from looping for the above case of an ordered
4164 * extent that falls inside of the logged extent.
4166 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4170 list_for_each_entry(sum, &ordered->list, list) {
4171 ret = btrfs_csum_file_blocks(trans, log, sum);
4177 if (*ordered_io_error || !mod_len || ret || skip_csum)
4180 if (em->compress_type) {
4182 csum_len = max(em->block_len, em->orig_block_len);
4184 csum_offset = mod_start - em->start;
4188 /* block start is already adjusted for the file extent offset. */
4189 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4190 em->block_start + csum_offset,
4191 em->block_start + csum_offset +
4192 csum_len - 1, &ordered_sums, 0);
4196 while (!list_empty(&ordered_sums)) {
4197 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4198 struct btrfs_ordered_sum,
4201 ret = btrfs_csum_file_blocks(trans, log, sums);
4202 list_del(&sums->list);
4209 static int log_one_extent(struct btrfs_trans_handle *trans,
4210 struct btrfs_inode *inode, struct btrfs_root *root,
4211 const struct extent_map *em,
4212 struct btrfs_path *path,
4213 const struct list_head *logged_list,
4214 struct btrfs_log_ctx *ctx)
4216 struct btrfs_root *log = root->log_root;
4217 struct btrfs_file_extent_item *fi;
4218 struct extent_buffer *leaf;
4219 struct btrfs_map_token token;
4220 struct btrfs_key key;
4221 u64 extent_offset = em->start - em->orig_start;
4224 int extent_inserted = 0;
4225 bool ordered_io_err = false;
4227 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em,
4228 logged_list, &ordered_io_err);
4232 if (ordered_io_err) {
4237 btrfs_init_map_token(&token);
4239 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4240 em->start + em->len, NULL, 0, 1,
4241 sizeof(*fi), &extent_inserted);
4245 if (!extent_inserted) {
4246 key.objectid = btrfs_ino(inode);
4247 key.type = BTRFS_EXTENT_DATA_KEY;
4248 key.offset = em->start;
4250 ret = btrfs_insert_empty_item(trans, log, path, &key,
4255 leaf = path->nodes[0];
4256 fi = btrfs_item_ptr(leaf, path->slots[0],
4257 struct btrfs_file_extent_item);
4259 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4261 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4262 btrfs_set_token_file_extent_type(leaf, fi,
4263 BTRFS_FILE_EXTENT_PREALLOC,
4266 btrfs_set_token_file_extent_type(leaf, fi,
4267 BTRFS_FILE_EXTENT_REG,
4270 block_len = max(em->block_len, em->orig_block_len);
4271 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4272 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4275 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4277 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4278 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4280 extent_offset, &token);
4281 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4284 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4285 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4289 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4290 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4291 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4292 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4294 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4295 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4296 btrfs_mark_buffer_dirty(leaf);
4298 btrfs_release_path(path);
4303 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4304 struct btrfs_root *root,
4305 struct btrfs_inode *inode,
4306 struct btrfs_path *path,
4307 struct list_head *logged_list,
4308 struct btrfs_log_ctx *ctx,
4312 struct extent_map *em, *n;
4313 struct list_head extents;
4314 struct extent_map_tree *tree = &inode->extent_tree;
4315 u64 logged_start, logged_end;
4320 INIT_LIST_HEAD(&extents);
4322 down_write(&inode->dio_sem);
4323 write_lock(&tree->lock);
4324 test_gen = root->fs_info->last_trans_committed;
4325 logged_start = start;
4328 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4329 list_del_init(&em->list);
4331 * Just an arbitrary number, this can be really CPU intensive
4332 * once we start getting a lot of extents, and really once we
4333 * have a bunch of extents we just want to commit since it will
4336 if (++num > 32768) {
4337 list_del_init(&tree->modified_extents);
4342 if (em->generation <= test_gen)
4345 if (em->start < logged_start)
4346 logged_start = em->start;
4347 if ((em->start + em->len - 1) > logged_end)
4348 logged_end = em->start + em->len - 1;
4350 /* Need a ref to keep it from getting evicted from cache */
4351 refcount_inc(&em->refs);
4352 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4353 list_add_tail(&em->list, &extents);
4357 list_sort(NULL, &extents, extent_cmp);
4358 btrfs_get_logged_extents(inode, logged_list, logged_start, logged_end);
4360 * Some ordered extents started by fsync might have completed
4361 * before we could collect them into the list logged_list, which
4362 * means they're gone, not in our logged_list nor in the inode's
4363 * ordered tree. We want the application/user space to know an
4364 * error happened while attempting to persist file data so that
4365 * it can take proper action. If such error happened, we leave
4366 * without writing to the log tree and the fsync must report the
4367 * file data write error and not commit the current transaction.
4369 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4373 while (!list_empty(&extents)) {
4374 em = list_entry(extents.next, struct extent_map, list);
4376 list_del_init(&em->list);
4379 * If we had an error we just need to delete everybody from our
4383 clear_em_logging(tree, em);
4384 free_extent_map(em);
4388 write_unlock(&tree->lock);
4390 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4392 write_lock(&tree->lock);
4393 clear_em_logging(tree, em);
4394 free_extent_map(em);
4396 WARN_ON(!list_empty(&extents));
4397 write_unlock(&tree->lock);
4398 up_write(&inode->dio_sem);
4400 btrfs_release_path(path);
4404 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4405 struct btrfs_path *path, u64 *size_ret)
4407 struct btrfs_key key;
4410 key.objectid = btrfs_ino(inode);
4411 key.type = BTRFS_INODE_ITEM_KEY;
4414 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4417 } else if (ret > 0) {
4420 struct btrfs_inode_item *item;
4422 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4423 struct btrfs_inode_item);
4424 *size_ret = btrfs_inode_size(path->nodes[0], item);
4427 btrfs_release_path(path);
4432 * At the moment we always log all xattrs. This is to figure out at log replay
4433 * time which xattrs must have their deletion replayed. If a xattr is missing
4434 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4435 * because if a xattr is deleted, the inode is fsynced and a power failure
4436 * happens, causing the log to be replayed the next time the fs is mounted,
4437 * we want the xattr to not exist anymore (same behaviour as other filesystems
4438 * with a journal, ext3/4, xfs, f2fs, etc).
4440 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4441 struct btrfs_root *root,
4442 struct btrfs_inode *inode,
4443 struct btrfs_path *path,
4444 struct btrfs_path *dst_path)
4447 struct btrfs_key key;
4448 const u64 ino = btrfs_ino(inode);
4453 key.type = BTRFS_XATTR_ITEM_KEY;
4456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4461 int slot = path->slots[0];
4462 struct extent_buffer *leaf = path->nodes[0];
4463 int nritems = btrfs_header_nritems(leaf);
4465 if (slot >= nritems) {
4467 u64 last_extent = 0;
4469 ret = copy_items(trans, inode, dst_path, path,
4470 &last_extent, start_slot,
4472 /* can't be 1, extent items aren't processed */
4478 ret = btrfs_next_leaf(root, path);
4486 btrfs_item_key_to_cpu(leaf, &key, slot);
4487 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4497 u64 last_extent = 0;
4499 ret = copy_items(trans, inode, dst_path, path,
4500 &last_extent, start_slot,
4502 /* can't be 1, extent items aren't processed */
4512 * If the no holes feature is enabled we need to make sure any hole between the
4513 * last extent and the i_size of our inode is explicitly marked in the log. This
4514 * is to make sure that doing something like:
4516 * 1) create file with 128Kb of data
4517 * 2) truncate file to 64Kb
4518 * 3) truncate file to 256Kb
4520 * 5) <crash/power failure>
4521 * 6) mount fs and trigger log replay
4523 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4524 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4525 * file correspond to a hole. The presence of explicit holes in a log tree is
4526 * what guarantees that log replay will remove/adjust file extent items in the
4529 * Here we do not need to care about holes between extents, that is already done
4530 * by copy_items(). We also only need to do this in the full sync path, where we
4531 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4532 * lookup the list of modified extent maps and if any represents a hole, we
4533 * insert a corresponding extent representing a hole in the log tree.
4535 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4536 struct btrfs_root *root,
4537 struct btrfs_inode *inode,
4538 struct btrfs_path *path)
4540 struct btrfs_fs_info *fs_info = root->fs_info;
4542 struct btrfs_key key;
4545 struct extent_buffer *leaf;
4546 struct btrfs_root *log = root->log_root;
4547 const u64 ino = btrfs_ino(inode);
4548 const u64 i_size = i_size_read(&inode->vfs_inode);
4550 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4554 key.type = BTRFS_EXTENT_DATA_KEY;
4555 key.offset = (u64)-1;
4557 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4562 ASSERT(path->slots[0] > 0);
4564 leaf = path->nodes[0];
4565 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4567 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4568 /* inode does not have any extents */
4572 struct btrfs_file_extent_item *extent;
4576 * If there's an extent beyond i_size, an explicit hole was
4577 * already inserted by copy_items().
4579 if (key.offset >= i_size)
4582 extent = btrfs_item_ptr(leaf, path->slots[0],
4583 struct btrfs_file_extent_item);
4585 if (btrfs_file_extent_type(leaf, extent) ==
4586 BTRFS_FILE_EXTENT_INLINE) {
4587 len = btrfs_file_extent_inline_len(leaf,
4590 ASSERT(len == i_size ||
4591 (len == fs_info->sectorsize &&
4592 btrfs_file_extent_compression(leaf, extent) !=
4593 BTRFS_COMPRESS_NONE));
4597 len = btrfs_file_extent_num_bytes(leaf, extent);
4598 /* Last extent goes beyond i_size, no need to log a hole. */
4599 if (key.offset + len > i_size)
4601 hole_start = key.offset + len;
4602 hole_size = i_size - hole_start;
4604 btrfs_release_path(path);
4606 /* Last extent ends at i_size. */
4610 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4611 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4612 hole_size, 0, hole_size, 0, 0, 0);
4617 * When we are logging a new inode X, check if it doesn't have a reference that
4618 * matches the reference from some other inode Y created in a past transaction
4619 * and that was renamed in the current transaction. If we don't do this, then at
4620 * log replay time we can lose inode Y (and all its files if it's a directory):
4623 * echo "hello world" > /mnt/x/foobar
4626 * mkdir /mnt/x # or touch /mnt/x
4627 * xfs_io -c fsync /mnt/x
4629 * mount fs, trigger log replay
4631 * After the log replay procedure, we would lose the first directory and all its
4632 * files (file foobar).
4633 * For the case where inode Y is not a directory we simply end up losing it:
4635 * echo "123" > /mnt/foo
4637 * mv /mnt/foo /mnt/bar
4638 * echo "abc" > /mnt/foo
4639 * xfs_io -c fsync /mnt/foo
4642 * We also need this for cases where a snapshot entry is replaced by some other
4643 * entry (file or directory) otherwise we end up with an unreplayable log due to
4644 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4645 * if it were a regular entry:
4648 * btrfs subvolume snapshot /mnt /mnt/x/snap
4649 * btrfs subvolume delete /mnt/x/snap
4652 * fsync /mnt/x or fsync some new file inside it
4655 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4656 * the same transaction.
4658 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4660 const struct btrfs_key *key,
4661 struct btrfs_inode *inode,
4665 struct btrfs_path *search_path;
4668 u32 item_size = btrfs_item_size_nr(eb, slot);
4670 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4672 search_path = btrfs_alloc_path();
4675 search_path->search_commit_root = 1;
4676 search_path->skip_locking = 1;
4678 while (cur_offset < item_size) {
4682 unsigned long name_ptr;
4683 struct btrfs_dir_item *di;
4685 if (key->type == BTRFS_INODE_REF_KEY) {
4686 struct btrfs_inode_ref *iref;
4688 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4689 parent = key->offset;
4690 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4691 name_ptr = (unsigned long)(iref + 1);
4692 this_len = sizeof(*iref) + this_name_len;
4694 struct btrfs_inode_extref *extref;
4696 extref = (struct btrfs_inode_extref *)(ptr +
4698 parent = btrfs_inode_extref_parent(eb, extref);
4699 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4700 name_ptr = (unsigned long)&extref->name;
4701 this_len = sizeof(*extref) + this_name_len;
4704 if (this_name_len > name_len) {
4707 new_name = krealloc(name, this_name_len, GFP_NOFS);
4712 name_len = this_name_len;
4716 read_extent_buffer(eb, name, name_ptr, this_name_len);
4717 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4718 parent, name, this_name_len, 0);
4719 if (di && !IS_ERR(di)) {
4720 struct btrfs_key di_key;
4722 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4724 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4726 *other_ino = di_key.objectid;
4731 } else if (IS_ERR(di)) {
4735 btrfs_release_path(search_path);
4737 cur_offset += this_len;
4741 btrfs_free_path(search_path);
4746 /* log a single inode in the tree log.
4747 * At least one parent directory for this inode must exist in the tree
4748 * or be logged already.
4750 * Any items from this inode changed by the current transaction are copied
4751 * to the log tree. An extra reference is taken on any extents in this
4752 * file, allowing us to avoid a whole pile of corner cases around logging
4753 * blocks that have been removed from the tree.
4755 * See LOG_INODE_ALL and related defines for a description of what inode_only
4758 * This handles both files and directories.
4760 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4761 struct btrfs_root *root, struct btrfs_inode *inode,
4765 struct btrfs_log_ctx *ctx)
4767 struct btrfs_fs_info *fs_info = root->fs_info;
4768 struct btrfs_path *path;
4769 struct btrfs_path *dst_path;
4770 struct btrfs_key min_key;
4771 struct btrfs_key max_key;
4772 struct btrfs_root *log = root->log_root;
4773 LIST_HEAD(logged_list);
4774 u64 last_extent = 0;
4778 int ins_start_slot = 0;
4780 bool fast_search = false;
4781 u64 ino = btrfs_ino(inode);
4782 struct extent_map_tree *em_tree = &inode->extent_tree;
4783 u64 logged_isize = 0;
4784 bool need_log_inode_item = true;
4786 path = btrfs_alloc_path();
4789 dst_path = btrfs_alloc_path();
4791 btrfs_free_path(path);
4795 min_key.objectid = ino;
4796 min_key.type = BTRFS_INODE_ITEM_KEY;
4799 max_key.objectid = ino;
4802 /* today the code can only do partial logging of directories */
4803 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4804 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4805 &inode->runtime_flags) &&
4806 inode_only >= LOG_INODE_EXISTS))
4807 max_key.type = BTRFS_XATTR_ITEM_KEY;
4809 max_key.type = (u8)-1;
4810 max_key.offset = (u64)-1;
4813 * Only run delayed items if we are a dir or a new file.
4814 * Otherwise commit the delayed inode only, which is needed in
4815 * order for the log replay code to mark inodes for link count
4816 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4818 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4819 inode->generation > fs_info->last_trans_committed)
4820 ret = btrfs_commit_inode_delayed_items(trans, inode);
4822 ret = btrfs_commit_inode_delayed_inode(inode);
4825 btrfs_free_path(path);
4826 btrfs_free_path(dst_path);
4830 if (inode_only == LOG_OTHER_INODE) {
4831 inode_only = LOG_INODE_EXISTS;
4832 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
4834 mutex_lock(&inode->log_mutex);
4838 * a brute force approach to making sure we get the most uptodate
4839 * copies of everything.
4841 if (S_ISDIR(inode->vfs_inode.i_mode)) {
4842 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4844 if (inode_only == LOG_INODE_EXISTS)
4845 max_key_type = BTRFS_XATTR_ITEM_KEY;
4846 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4848 if (inode_only == LOG_INODE_EXISTS) {
4850 * Make sure the new inode item we write to the log has
4851 * the same isize as the current one (if it exists).
4852 * This is necessary to prevent data loss after log
4853 * replay, and also to prevent doing a wrong expanding
4854 * truncate - for e.g. create file, write 4K into offset
4855 * 0, fsync, write 4K into offset 4096, add hard link,
4856 * fsync some other file (to sync log), power fail - if
4857 * we use the inode's current i_size, after log replay
4858 * we get a 8Kb file, with the last 4Kb extent as a hole
4859 * (zeroes), as if an expanding truncate happened,
4860 * instead of getting a file of 4Kb only.
4862 err = logged_inode_size(log, inode, path, &logged_isize);
4866 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4867 &inode->runtime_flags)) {
4868 if (inode_only == LOG_INODE_EXISTS) {
4869 max_key.type = BTRFS_XATTR_ITEM_KEY;
4870 ret = drop_objectid_items(trans, log, path, ino,
4873 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4874 &inode->runtime_flags);
4875 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4876 &inode->runtime_flags);
4878 ret = btrfs_truncate_inode_items(trans,
4879 log, &inode->vfs_inode, 0, 0);
4884 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4885 &inode->runtime_flags) ||
4886 inode_only == LOG_INODE_EXISTS) {
4887 if (inode_only == LOG_INODE_ALL)
4889 max_key.type = BTRFS_XATTR_ITEM_KEY;
4890 ret = drop_objectid_items(trans, log, path, ino,
4893 if (inode_only == LOG_INODE_ALL)
4906 ret = btrfs_search_forward(root, &min_key,
4907 path, trans->transid);
4915 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4916 if (min_key.objectid != ino)
4918 if (min_key.type > max_key.type)
4921 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4922 need_log_inode_item = false;
4924 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4925 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4926 inode->generation == trans->transid) {
4929 ret = btrfs_check_ref_name_override(path->nodes[0],
4930 path->slots[0], &min_key, inode,
4935 } else if (ret > 0 && ctx &&
4936 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4937 struct btrfs_key inode_key;
4938 struct inode *other_inode;
4944 ins_start_slot = path->slots[0];
4946 ret = copy_items(trans, inode, dst_path, path,
4947 &last_extent, ins_start_slot,
4955 btrfs_release_path(path);
4956 inode_key.objectid = other_ino;
4957 inode_key.type = BTRFS_INODE_ITEM_KEY;
4958 inode_key.offset = 0;
4959 other_inode = btrfs_iget(fs_info->sb,
4963 * If the other inode that had a conflicting dir
4964 * entry was deleted in the current transaction,
4965 * we don't need to do more work nor fallback to
4966 * a transaction commit.
4968 if (IS_ERR(other_inode) &&
4969 PTR_ERR(other_inode) == -ENOENT) {
4971 } else if (IS_ERR(other_inode)) {
4972 err = PTR_ERR(other_inode);
4976 * We are safe logging the other inode without
4977 * acquiring its i_mutex as long as we log with
4978 * the LOG_INODE_EXISTS mode. We're safe against
4979 * concurrent renames of the other inode as well
4980 * because during a rename we pin the log and
4981 * update the log with the new name before we
4984 err = btrfs_log_inode(trans, root,
4985 BTRFS_I(other_inode),
4986 LOG_OTHER_INODE, 0, LLONG_MAX,
4996 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4997 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5000 ret = copy_items(trans, inode, dst_path, path,
5001 &last_extent, ins_start_slot,
5002 ins_nr, inode_only, logged_isize);
5009 btrfs_release_path(path);
5015 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5018 } else if (!ins_nr) {
5019 ins_start_slot = path->slots[0];
5024 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5025 ins_start_slot, ins_nr, inode_only,
5033 btrfs_release_path(path);
5037 ins_start_slot = path->slots[0];
5040 nritems = btrfs_header_nritems(path->nodes[0]);
5042 if (path->slots[0] < nritems) {
5043 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5048 ret = copy_items(trans, inode, dst_path, path,
5049 &last_extent, ins_start_slot,
5050 ins_nr, inode_only, logged_isize);
5058 btrfs_release_path(path);
5060 if (min_key.offset < (u64)-1) {
5062 } else if (min_key.type < max_key.type) {
5070 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5071 ins_start_slot, ins_nr, inode_only,
5081 btrfs_release_path(path);
5082 btrfs_release_path(dst_path);
5083 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5086 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5087 btrfs_release_path(path);
5088 btrfs_release_path(dst_path);
5089 err = btrfs_log_trailing_hole(trans, root, inode, path);
5094 btrfs_release_path(path);
5095 btrfs_release_path(dst_path);
5096 if (need_log_inode_item) {
5097 err = log_inode_item(trans, log, dst_path, inode);
5102 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5103 &logged_list, ctx, start, end);
5108 } else if (inode_only == LOG_INODE_ALL) {
5109 struct extent_map *em, *n;
5111 write_lock(&em_tree->lock);
5113 * We can't just remove every em if we're called for a ranged
5114 * fsync - that is, one that doesn't cover the whole possible
5115 * file range (0 to LLONG_MAX). This is because we can have
5116 * em's that fall outside the range we're logging and therefore
5117 * their ordered operations haven't completed yet
5118 * (btrfs_finish_ordered_io() not invoked yet). This means we
5119 * didn't get their respective file extent item in the fs/subvol
5120 * tree yet, and need to let the next fast fsync (one which
5121 * consults the list of modified extent maps) find the em so
5122 * that it logs a matching file extent item and waits for the
5123 * respective ordered operation to complete (if it's still
5126 * Removing every em outside the range we're logging would make
5127 * the next fast fsync not log their matching file extent items,
5128 * therefore making us lose data after a log replay.
5130 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5132 const u64 mod_end = em->mod_start + em->mod_len - 1;
5134 if (em->mod_start >= start && mod_end <= end)
5135 list_del_init(&em->list);
5137 write_unlock(&em_tree->lock);
5140 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5141 ret = log_directory_changes(trans, root, inode, path, dst_path,
5149 spin_lock(&inode->lock);
5150 inode->logged_trans = trans->transid;
5151 inode->last_log_commit = inode->last_sub_trans;
5152 spin_unlock(&inode->lock);
5155 btrfs_put_logged_extents(&logged_list);
5157 btrfs_submit_logged_extents(&logged_list, log);
5158 mutex_unlock(&inode->log_mutex);
5160 btrfs_free_path(path);
5161 btrfs_free_path(dst_path);
5166 * Check if we must fallback to a transaction commit when logging an inode.
5167 * This must be called after logging the inode and is used only in the context
5168 * when fsyncing an inode requires the need to log some other inode - in which
5169 * case we can't lock the i_mutex of each other inode we need to log as that
5170 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5171 * log inodes up or down in the hierarchy) or rename operations for example. So
5172 * we take the log_mutex of the inode after we have logged it and then check for
5173 * its last_unlink_trans value - this is safe because any task setting
5174 * last_unlink_trans must take the log_mutex and it must do this before it does
5175 * the actual unlink operation, so if we do this check before a concurrent task
5176 * sets last_unlink_trans it means we've logged a consistent version/state of
5177 * all the inode items, otherwise we are not sure and must do a transaction
5178 * commit (the concurrent task might have only updated last_unlink_trans before
5179 * we logged the inode or it might have also done the unlink).
5181 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5182 struct btrfs_inode *inode)
5184 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5187 mutex_lock(&inode->log_mutex);
5188 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5190 * Make sure any commits to the log are forced to be full
5193 btrfs_set_log_full_commit(fs_info, trans);
5196 mutex_unlock(&inode->log_mutex);
5202 * follow the dentry parent pointers up the chain and see if any
5203 * of the directories in it require a full commit before they can
5204 * be logged. Returns zero if nothing special needs to be done or 1 if
5205 * a full commit is required.
5207 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5208 struct btrfs_inode *inode,
5209 struct dentry *parent,
5210 struct super_block *sb,
5214 struct dentry *old_parent = NULL;
5215 struct btrfs_inode *orig_inode = inode;
5218 * for regular files, if its inode is already on disk, we don't
5219 * have to worry about the parents at all. This is because
5220 * we can use the last_unlink_trans field to record renames
5221 * and other fun in this file.
5223 if (S_ISREG(inode->vfs_inode.i_mode) &&
5224 inode->generation <= last_committed &&
5225 inode->last_unlink_trans <= last_committed)
5228 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5229 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5231 inode = BTRFS_I(d_inode(parent));
5236 * If we are logging a directory then we start with our inode,
5237 * not our parent's inode, so we need to skip setting the
5238 * logged_trans so that further down in the log code we don't
5239 * think this inode has already been logged.
5241 if (inode != orig_inode)
5242 inode->logged_trans = trans->transid;
5245 if (btrfs_must_commit_transaction(trans, inode)) {
5250 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5253 if (IS_ROOT(parent)) {
5254 inode = BTRFS_I(d_inode(parent));
5255 if (btrfs_must_commit_transaction(trans, inode))
5260 parent = dget_parent(parent);
5262 old_parent = parent;
5263 inode = BTRFS_I(d_inode(parent));
5271 struct btrfs_dir_list {
5273 struct list_head list;
5277 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5278 * details about the why it is needed.
5279 * This is a recursive operation - if an existing dentry corresponds to a
5280 * directory, that directory's new entries are logged too (same behaviour as
5281 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5282 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5283 * complains about the following circular lock dependency / possible deadlock:
5287 * lock(&type->i_mutex_dir_key#3/2);
5288 * lock(sb_internal#2);
5289 * lock(&type->i_mutex_dir_key#3/2);
5290 * lock(&sb->s_type->i_mutex_key#14);
5292 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5293 * sb_start_intwrite() in btrfs_start_transaction().
5294 * Not locking i_mutex of the inodes is still safe because:
5296 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5297 * that while logging the inode new references (names) are added or removed
5298 * from the inode, leaving the logged inode item with a link count that does
5299 * not match the number of logged inode reference items. This is fine because
5300 * at log replay time we compute the real number of links and correct the
5301 * link count in the inode item (see replay_one_buffer() and
5302 * link_to_fixup_dir());
5304 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5305 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5306 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5307 * has a size that doesn't match the sum of the lengths of all the logged
5308 * names. This does not result in a problem because if a dir_item key is
5309 * logged but its matching dir_index key is not logged, at log replay time we
5310 * don't use it to replay the respective name (see replay_one_name()). On the
5311 * other hand if only the dir_index key ends up being logged, the respective
5312 * name is added to the fs/subvol tree with both the dir_item and dir_index
5313 * keys created (see replay_one_name()).
5314 * The directory's inode item with a wrong i_size is not a problem as well,
5315 * since we don't use it at log replay time to set the i_size in the inode
5316 * item of the fs/subvol tree (see overwrite_item()).
5318 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5319 struct btrfs_root *root,
5320 struct btrfs_inode *start_inode,
5321 struct btrfs_log_ctx *ctx)
5323 struct btrfs_fs_info *fs_info = root->fs_info;
5324 struct btrfs_root *log = root->log_root;
5325 struct btrfs_path *path;
5326 LIST_HEAD(dir_list);
5327 struct btrfs_dir_list *dir_elem;
5330 path = btrfs_alloc_path();
5334 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5336 btrfs_free_path(path);
5339 dir_elem->ino = btrfs_ino(start_inode);
5340 list_add_tail(&dir_elem->list, &dir_list);
5342 while (!list_empty(&dir_list)) {
5343 struct extent_buffer *leaf;
5344 struct btrfs_key min_key;
5348 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5351 goto next_dir_inode;
5353 min_key.objectid = dir_elem->ino;
5354 min_key.type = BTRFS_DIR_ITEM_KEY;
5357 btrfs_release_path(path);
5358 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5360 goto next_dir_inode;
5361 } else if (ret > 0) {
5363 goto next_dir_inode;
5367 leaf = path->nodes[0];
5368 nritems = btrfs_header_nritems(leaf);
5369 for (i = path->slots[0]; i < nritems; i++) {
5370 struct btrfs_dir_item *di;
5371 struct btrfs_key di_key;
5372 struct inode *di_inode;
5373 struct btrfs_dir_list *new_dir_elem;
5374 int log_mode = LOG_INODE_EXISTS;
5377 btrfs_item_key_to_cpu(leaf, &min_key, i);
5378 if (min_key.objectid != dir_elem->ino ||
5379 min_key.type != BTRFS_DIR_ITEM_KEY)
5380 goto next_dir_inode;
5382 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5383 type = btrfs_dir_type(leaf, di);
5384 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5385 type != BTRFS_FT_DIR)
5387 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5388 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5391 btrfs_release_path(path);
5392 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5393 if (IS_ERR(di_inode)) {
5394 ret = PTR_ERR(di_inode);
5395 goto next_dir_inode;
5398 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5403 ctx->log_new_dentries = false;
5404 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5405 log_mode = LOG_INODE_ALL;
5406 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5407 log_mode, 0, LLONG_MAX, ctx);
5409 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5413 goto next_dir_inode;
5414 if (ctx->log_new_dentries) {
5415 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5417 if (!new_dir_elem) {
5419 goto next_dir_inode;
5421 new_dir_elem->ino = di_key.objectid;
5422 list_add_tail(&new_dir_elem->list, &dir_list);
5427 ret = btrfs_next_leaf(log, path);
5429 goto next_dir_inode;
5430 } else if (ret > 0) {
5432 goto next_dir_inode;
5436 if (min_key.offset < (u64)-1) {
5441 list_del(&dir_elem->list);
5445 btrfs_free_path(path);
5449 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5450 struct btrfs_inode *inode,
5451 struct btrfs_log_ctx *ctx)
5453 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5455 struct btrfs_path *path;
5456 struct btrfs_key key;
5457 struct btrfs_root *root = inode->root;
5458 const u64 ino = btrfs_ino(inode);
5460 path = btrfs_alloc_path();
5463 path->skip_locking = 1;
5464 path->search_commit_root = 1;
5467 key.type = BTRFS_INODE_REF_KEY;
5469 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5474 struct extent_buffer *leaf = path->nodes[0];
5475 int slot = path->slots[0];
5480 if (slot >= btrfs_header_nritems(leaf)) {
5481 ret = btrfs_next_leaf(root, path);
5489 btrfs_item_key_to_cpu(leaf, &key, slot);
5490 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5491 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5494 item_size = btrfs_item_size_nr(leaf, slot);
5495 ptr = btrfs_item_ptr_offset(leaf, slot);
5496 while (cur_offset < item_size) {
5497 struct btrfs_key inode_key;
5498 struct inode *dir_inode;
5500 inode_key.type = BTRFS_INODE_ITEM_KEY;
5501 inode_key.offset = 0;
5503 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5504 struct btrfs_inode_extref *extref;
5506 extref = (struct btrfs_inode_extref *)
5508 inode_key.objectid = btrfs_inode_extref_parent(
5510 cur_offset += sizeof(*extref);
5511 cur_offset += btrfs_inode_extref_name_len(leaf,
5514 inode_key.objectid = key.offset;
5515 cur_offset = item_size;
5518 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5520 /* If parent inode was deleted, skip it. */
5521 if (IS_ERR(dir_inode))
5525 ctx->log_new_dentries = false;
5526 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5527 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5529 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5531 if (!ret && ctx && ctx->log_new_dentries)
5532 ret = log_new_dir_dentries(trans, root,
5533 BTRFS_I(dir_inode), ctx);
5542 btrfs_free_path(path);
5547 * helper function around btrfs_log_inode to make sure newly created
5548 * parent directories also end up in the log. A minimal inode and backref
5549 * only logging is done of any parent directories that are older than
5550 * the last committed transaction
5552 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5553 struct btrfs_inode *inode,
5554 struct dentry *parent,
5558 struct btrfs_log_ctx *ctx)
5560 struct btrfs_root *root = inode->root;
5561 struct btrfs_fs_info *fs_info = root->fs_info;
5562 struct super_block *sb;
5563 struct dentry *old_parent = NULL;
5565 u64 last_committed = fs_info->last_trans_committed;
5566 bool log_dentries = false;
5567 struct btrfs_inode *orig_inode = inode;
5569 sb = inode->vfs_inode.i_sb;
5571 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5577 * The prev transaction commit doesn't complete, we need do
5578 * full commit by ourselves.
5580 if (fs_info->last_trans_log_full_commit >
5581 fs_info->last_trans_committed) {
5586 if (btrfs_root_refs(&root->root_item) == 0) {
5591 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5596 if (btrfs_inode_in_log(inode, trans->transid)) {
5597 ret = BTRFS_NO_LOG_SYNC;
5601 ret = start_log_trans(trans, root, ctx);
5605 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5610 * for regular files, if its inode is already on disk, we don't
5611 * have to worry about the parents at all. This is because
5612 * we can use the last_unlink_trans field to record renames
5613 * and other fun in this file.
5615 if (S_ISREG(inode->vfs_inode.i_mode) &&
5616 inode->generation <= last_committed &&
5617 inode->last_unlink_trans <= last_committed) {
5622 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5623 log_dentries = true;
5626 * On unlink we must make sure all our current and old parent directory
5627 * inodes are fully logged. This is to prevent leaving dangling
5628 * directory index entries in directories that were our parents but are
5629 * not anymore. Not doing this results in old parent directory being
5630 * impossible to delete after log replay (rmdir will always fail with
5631 * error -ENOTEMPTY).
5637 * ln testdir/foo testdir/bar
5639 * unlink testdir/bar
5640 * xfs_io -c fsync testdir/foo
5642 * mount fs, triggers log replay
5644 * If we don't log the parent directory (testdir), after log replay the
5645 * directory still has an entry pointing to the file inode using the bar
5646 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5647 * the file inode has a link count of 1.
5653 * ln foo testdir/foo2
5654 * ln foo testdir/foo3
5656 * unlink testdir/foo3
5657 * xfs_io -c fsync foo
5659 * mount fs, triggers log replay
5661 * Similar as the first example, after log replay the parent directory
5662 * testdir still has an entry pointing to the inode file with name foo3
5663 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5664 * and has a link count of 2.
5666 if (inode->last_unlink_trans > last_committed) {
5667 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5673 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5676 inode = BTRFS_I(d_inode(parent));
5677 if (root != inode->root)
5680 if (inode->generation > last_committed) {
5681 ret = btrfs_log_inode(trans, root, inode,
5682 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5686 if (IS_ROOT(parent))
5689 parent = dget_parent(parent);
5691 old_parent = parent;
5694 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5700 btrfs_set_log_full_commit(fs_info, trans);
5705 btrfs_remove_log_ctx(root, ctx);
5706 btrfs_end_log_trans(root);
5712 * it is not safe to log dentry if the chunk root has added new
5713 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5714 * If this returns 1, you must commit the transaction to safely get your
5717 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5718 struct dentry *dentry,
5721 struct btrfs_log_ctx *ctx)
5723 struct dentry *parent = dget_parent(dentry);
5726 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
5727 start, end, LOG_INODE_ALL, ctx);
5734 * should be called during mount to recover any replay any log trees
5737 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5740 struct btrfs_path *path;
5741 struct btrfs_trans_handle *trans;
5742 struct btrfs_key key;
5743 struct btrfs_key found_key;
5744 struct btrfs_key tmp_key;
5745 struct btrfs_root *log;
5746 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5747 struct walk_control wc = {
5748 .process_func = process_one_buffer,
5752 path = btrfs_alloc_path();
5756 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5758 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5759 if (IS_ERR(trans)) {
5760 ret = PTR_ERR(trans);
5767 ret = walk_log_tree(trans, log_root_tree, &wc);
5769 btrfs_handle_fs_error(fs_info, ret,
5770 "Failed to pin buffers while recovering log root tree.");
5775 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5776 key.offset = (u64)-1;
5777 key.type = BTRFS_ROOT_ITEM_KEY;
5780 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5783 btrfs_handle_fs_error(fs_info, ret,
5784 "Couldn't find tree log root.");
5788 if (path->slots[0] == 0)
5792 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5794 btrfs_release_path(path);
5795 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5798 log = btrfs_read_fs_root(log_root_tree, &found_key);
5801 btrfs_handle_fs_error(fs_info, ret,
5802 "Couldn't read tree log root.");
5806 tmp_key.objectid = found_key.offset;
5807 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5808 tmp_key.offset = (u64)-1;
5810 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5811 if (IS_ERR(wc.replay_dest)) {
5812 ret = PTR_ERR(wc.replay_dest);
5813 free_extent_buffer(log->node);
5814 free_extent_buffer(log->commit_root);
5816 btrfs_handle_fs_error(fs_info, ret,
5817 "Couldn't read target root for tree log recovery.");
5821 wc.replay_dest->log_root = log;
5822 btrfs_record_root_in_trans(trans, wc.replay_dest);
5823 ret = walk_log_tree(trans, log, &wc);
5825 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5826 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5830 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5831 struct btrfs_root *root = wc.replay_dest;
5833 btrfs_release_path(path);
5836 * We have just replayed everything, and the highest
5837 * objectid of fs roots probably has changed in case
5838 * some inode_item's got replayed.
5840 * root->objectid_mutex is not acquired as log replay
5841 * could only happen during mount.
5843 ret = btrfs_find_highest_objectid(root,
5844 &root->highest_objectid);
5847 key.offset = found_key.offset - 1;
5848 wc.replay_dest->log_root = NULL;
5849 free_extent_buffer(log->node);
5850 free_extent_buffer(log->commit_root);
5856 if (found_key.offset == 0)
5859 btrfs_release_path(path);
5861 /* step one is to pin it all, step two is to replay just inodes */
5864 wc.process_func = replay_one_buffer;
5865 wc.stage = LOG_WALK_REPLAY_INODES;
5868 /* step three is to replay everything */
5869 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5874 btrfs_free_path(path);
5876 /* step 4: commit the transaction, which also unpins the blocks */
5877 ret = btrfs_commit_transaction(trans);
5881 free_extent_buffer(log_root_tree->node);
5882 log_root_tree->log_root = NULL;
5883 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5884 kfree(log_root_tree);
5889 btrfs_end_transaction(wc.trans);
5890 btrfs_free_path(path);
5895 * there are some corner cases where we want to force a full
5896 * commit instead of allowing a directory to be logged.
5898 * They revolve around files there were unlinked from the directory, and
5899 * this function updates the parent directory so that a full commit is
5900 * properly done if it is fsync'd later after the unlinks are done.
5902 * Must be called before the unlink operations (updates to the subvolume tree,
5903 * inodes, etc) are done.
5905 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5906 struct btrfs_inode *dir, struct btrfs_inode *inode,
5910 * when we're logging a file, if it hasn't been renamed
5911 * or unlinked, and its inode is fully committed on disk,
5912 * we don't have to worry about walking up the directory chain
5913 * to log its parents.
5915 * So, we use the last_unlink_trans field to put this transid
5916 * into the file. When the file is logged we check it and
5917 * don't log the parents if the file is fully on disk.
5919 mutex_lock(&inode->log_mutex);
5920 inode->last_unlink_trans = trans->transid;
5921 mutex_unlock(&inode->log_mutex);
5924 * if this directory was already logged any new
5925 * names for this file/dir will get recorded
5928 if (dir->logged_trans == trans->transid)
5932 * if the inode we're about to unlink was logged,
5933 * the log will be properly updated for any new names
5935 if (inode->logged_trans == trans->transid)
5939 * when renaming files across directories, if the directory
5940 * there we're unlinking from gets fsync'd later on, there's
5941 * no way to find the destination directory later and fsync it
5942 * properly. So, we have to be conservative and force commits
5943 * so the new name gets discovered.
5948 /* we can safely do the unlink without any special recording */
5952 mutex_lock(&dir->log_mutex);
5953 dir->last_unlink_trans = trans->transid;
5954 mutex_unlock(&dir->log_mutex);
5958 * Make sure that if someone attempts to fsync the parent directory of a deleted
5959 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5960 * that after replaying the log tree of the parent directory's root we will not
5961 * see the snapshot anymore and at log replay time we will not see any log tree
5962 * corresponding to the deleted snapshot's root, which could lead to replaying
5963 * it after replaying the log tree of the parent directory (which would replay
5964 * the snapshot delete operation).
5966 * Must be called before the actual snapshot destroy operation (updates to the
5967 * parent root and tree of tree roots trees, etc) are done.
5969 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5970 struct btrfs_inode *dir)
5972 mutex_lock(&dir->log_mutex);
5973 dir->last_unlink_trans = trans->transid;
5974 mutex_unlock(&dir->log_mutex);
5978 * Call this after adding a new name for a file and it will properly
5979 * update the log to reflect the new name.
5981 * It will return zero if all goes well, and it will return 1 if a
5982 * full transaction commit is required.
5984 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5985 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
5986 struct dentry *parent)
5988 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5991 * this will force the logging code to walk the dentry chain
5994 if (!S_ISDIR(inode->vfs_inode.i_mode))
5995 inode->last_unlink_trans = trans->transid;
5998 * if this inode hasn't been logged and directory we're renaming it
5999 * from hasn't been logged, we don't need to log it
6001 if (inode->logged_trans <= fs_info->last_trans_committed &&
6002 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6005 return btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6006 LOG_INODE_EXISTS, NULL);