1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "inode-map.h"
21 #include "block-group.h"
22 #include "space-info.h"
24 /* magic values for the inode_only field in btrfs_log_inode:
26 * LOG_INODE_ALL means to log everything
27 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 * directory trouble cases
40 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
41 * log, we must force a full commit before doing an fsync of the directory
42 * where the unlink was done.
43 * ---> record transid of last unlink/rename per directory
47 * rename foo/some_dir foo2/some_dir
49 * fsync foo/some_dir/some_file
51 * The fsync above will unlink the original some_dir without recording
52 * it in its new location (foo2). After a crash, some_dir will be gone
53 * unless the fsync of some_file forces a full commit
55 * 2) we must log any new names for any file or dir that is in the fsync
56 * log. ---> check inode while renaming/linking.
58 * 2a) we must log any new names for any file or dir during rename
59 * when the directory they are being removed from was logged.
60 * ---> check inode and old parent dir during rename
62 * 2a is actually the more important variant. With the extra logging
63 * a crash might unlink the old name without recreating the new one
65 * 3) after a crash, we must go through any directories with a link count
66 * of zero and redo the rm -rf
73 * The directory f1 was fully removed from the FS, but fsync was never
74 * called on f1, only its parent dir. After a crash the rm -rf must
75 * be replayed. This must be able to recurse down the entire
76 * directory tree. The inode link count fixup code takes care of the
81 * stages for the tree walking. The first
82 * stage (0) is to only pin down the blocks we find
83 * the second stage (1) is to make sure that all the inodes
84 * we find in the log are created in the subvolume.
86 * The last stage is to deal with directories and links and extents
87 * and all the other fun semantics
91 LOG_WALK_REPLAY_INODES,
92 LOG_WALK_REPLAY_DIR_INDEX,
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct btrfs_inode *inode,
101 struct btrfs_log_ctx *ctx);
102 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
105 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
143 struct btrfs_fs_info *fs_info = root->fs_info;
146 mutex_lock(&root->log_mutex);
148 if (root->log_root) {
149 if (btrfs_need_log_full_commit(trans)) {
154 if (!root->log_start_pid) {
155 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
156 root->log_start_pid = current->pid;
157 } else if (root->log_start_pid != current->pid) {
158 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
161 mutex_lock(&fs_info->tree_log_mutex);
162 if (!fs_info->log_root_tree)
163 ret = btrfs_init_log_root_tree(trans, fs_info);
164 mutex_unlock(&fs_info->tree_log_mutex);
168 ret = btrfs_add_log_tree(trans, root);
172 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
173 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
174 root->log_start_pid = current->pid;
177 atomic_inc(&root->log_batch);
178 atomic_inc(&root->log_writers);
180 int index = root->log_transid % 2;
181 list_add_tail(&ctx->list, &root->log_ctxs[index]);
182 ctx->log_transid = root->log_transid;
186 mutex_unlock(&root->log_mutex);
191 * returns 0 if there was a log transaction running and we were able
192 * to join, or returns -ENOENT if there were not transactions
195 static int join_running_log_trans(struct btrfs_root *root)
199 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
202 mutex_lock(&root->log_mutex);
203 if (root->log_root) {
205 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
212 * This either makes the current running log transaction wait
213 * until you call btrfs_end_log_trans() or it makes any future
214 * log transactions wait until you call btrfs_end_log_trans()
216 void btrfs_pin_log_trans(struct btrfs_root *root)
218 mutex_lock(&root->log_mutex);
219 atomic_inc(&root->log_writers);
220 mutex_unlock(&root->log_mutex);
224 * indicate we're done making changes to the log tree
225 * and wake up anyone waiting to do a sync
227 void btrfs_end_log_trans(struct btrfs_root *root)
229 if (atomic_dec_and_test(&root->log_writers)) {
230 /* atomic_dec_and_test implies a barrier */
231 cond_wake_up_nomb(&root->log_writer_wait);
235 static int btrfs_write_tree_block(struct extent_buffer *buf)
237 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
238 buf->start + buf->len - 1);
241 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
243 filemap_fdatawait_range(buf->pages[0]->mapping,
244 buf->start, buf->start + buf->len - 1);
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 */
278 * Ignore any items from the inode currently being processed. Needs
279 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
280 * the LOG_WALK_REPLAY_INODES stage.
282 bool ignore_cur_inode;
284 /* the root we are currently replaying */
285 struct btrfs_root *replay_dest;
287 /* the trans handle for the current replay */
288 struct btrfs_trans_handle *trans;
290 /* the function that gets used to process blocks we find in the
291 * tree. Note the extent_buffer might not be up to date when it is
292 * passed in, and it must be checked or read if you need the data
295 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
296 struct walk_control *wc, u64 gen, int level);
300 * process_func used to pin down extents, write them or wait on them
302 static int process_one_buffer(struct btrfs_root *log,
303 struct extent_buffer *eb,
304 struct walk_control *wc, u64 gen, int level)
306 struct btrfs_fs_info *fs_info = log->fs_info;
310 * If this fs is mixed then we need to be able to process the leaves to
311 * pin down any logged extents, so we have to read the block.
313 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
314 ret = btrfs_read_buffer(eb, gen, level, NULL);
320 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
323 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
324 if (wc->pin && btrfs_header_level(eb) == 0)
325 ret = btrfs_exclude_logged_extents(eb);
327 btrfs_write_tree_block(eb);
329 btrfs_wait_tree_block_writeback(eb);
335 * Item overwrite used by replay and tree logging. eb, slot and key all refer
336 * to the src data we are copying out.
338 * root is the tree we are copying into, and path is a scratch
339 * path for use in this function (it should be released on entry and
340 * will be released on exit).
342 * If the key is already in the destination tree the existing item is
343 * overwritten. If the existing item isn't big enough, it is extended.
344 * If it is too large, it is truncated.
346 * If the key isn't in the destination yet, a new item is inserted.
348 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
349 struct btrfs_root *root,
350 struct btrfs_path *path,
351 struct extent_buffer *eb, int slot,
352 struct btrfs_key *key)
356 u64 saved_i_size = 0;
357 int save_old_i_size = 0;
358 unsigned long src_ptr;
359 unsigned long dst_ptr;
360 int overwrite_root = 0;
361 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
363 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
366 item_size = btrfs_item_size_nr(eb, slot);
367 src_ptr = btrfs_item_ptr_offset(eb, slot);
369 /* look for the key in the destination tree */
370 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
377 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
379 if (dst_size != item_size)
382 if (item_size == 0) {
383 btrfs_release_path(path);
386 dst_copy = kmalloc(item_size, GFP_NOFS);
387 src_copy = kmalloc(item_size, GFP_NOFS);
388 if (!dst_copy || !src_copy) {
389 btrfs_release_path(path);
395 read_extent_buffer(eb, src_copy, src_ptr, item_size);
397 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
398 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
400 ret = memcmp(dst_copy, src_copy, item_size);
405 * they have the same contents, just return, this saves
406 * us from cowing blocks in the destination tree and doing
407 * extra writes that may not have been done by a previous
411 btrfs_release_path(path);
416 * We need to load the old nbytes into the inode so when we
417 * replay the extents we've logged we get the right nbytes.
420 struct btrfs_inode_item *item;
424 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
425 struct btrfs_inode_item);
426 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
427 item = btrfs_item_ptr(eb, slot,
428 struct btrfs_inode_item);
429 btrfs_set_inode_nbytes(eb, item, nbytes);
432 * If this is a directory we need to reset the i_size to
433 * 0 so that we can set it up properly when replaying
434 * the rest of the items in this log.
436 mode = btrfs_inode_mode(eb, item);
438 btrfs_set_inode_size(eb, item, 0);
440 } else if (inode_item) {
441 struct btrfs_inode_item *item;
445 * New inode, set nbytes to 0 so that the nbytes comes out
446 * properly when we replay the extents.
448 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
449 btrfs_set_inode_nbytes(eb, item, 0);
452 * If this is a directory we need to reset the i_size to 0 so
453 * that we can set it up properly when replaying the rest of
454 * the items in this log.
456 mode = btrfs_inode_mode(eb, item);
458 btrfs_set_inode_size(eb, item, 0);
461 btrfs_release_path(path);
462 /* try to insert the key into the destination tree */
463 path->skip_release_on_error = 1;
464 ret = btrfs_insert_empty_item(trans, root, path,
466 path->skip_release_on_error = 0;
468 /* make sure any existing item is the correct size */
469 if (ret == -EEXIST || ret == -EOVERFLOW) {
471 found_size = btrfs_item_size_nr(path->nodes[0],
473 if (found_size > item_size)
474 btrfs_truncate_item(path, item_size, 1);
475 else if (found_size < item_size)
476 btrfs_extend_item(path, item_size - found_size);
480 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
483 /* don't overwrite an existing inode if the generation number
484 * was logged as zero. This is done when the tree logging code
485 * is just logging an inode to make sure it exists after recovery.
487 * Also, don't overwrite i_size on directories during replay.
488 * log replay inserts and removes directory items based on the
489 * state of the tree found in the subvolume, and i_size is modified
492 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
493 struct btrfs_inode_item *src_item;
494 struct btrfs_inode_item *dst_item;
496 src_item = (struct btrfs_inode_item *)src_ptr;
497 dst_item = (struct btrfs_inode_item *)dst_ptr;
499 if (btrfs_inode_generation(eb, src_item) == 0) {
500 struct extent_buffer *dst_eb = path->nodes[0];
501 const u64 ino_size = btrfs_inode_size(eb, src_item);
504 * For regular files an ino_size == 0 is used only when
505 * logging that an inode exists, as part of a directory
506 * fsync, and the inode wasn't fsynced before. In this
507 * case don't set the size of the inode in the fs/subvol
508 * tree, otherwise we would be throwing valid data away.
510 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
511 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
513 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
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,
559 inode = btrfs_iget(root->fs_info->sb, objectid, root);
565 /* replays a single extent in 'eb' at 'slot' with 'key' into the
566 * subvolume 'root'. path is released on entry and should be released
569 * extents in the log tree have not been allocated out of the extent
570 * tree yet. So, this completes the allocation, taking a reference
571 * as required if the extent already exists or creating a new extent
572 * if it isn't in the extent allocation tree yet.
574 * The extent is inserted into the file, dropping any existing extents
575 * from the file that overlap the new one.
577 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
578 struct btrfs_root *root,
579 struct btrfs_path *path,
580 struct extent_buffer *eb, int slot,
581 struct btrfs_key *key)
583 struct btrfs_fs_info *fs_info = root->fs_info;
586 u64 start = key->offset;
588 struct btrfs_file_extent_item *item;
589 struct inode *inode = NULL;
593 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
594 found_type = btrfs_file_extent_type(eb, item);
596 if (found_type == BTRFS_FILE_EXTENT_REG ||
597 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
598 nbytes = btrfs_file_extent_num_bytes(eb, item);
599 extent_end = start + nbytes;
602 * We don't add to the inodes nbytes if we are prealloc or a
605 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
607 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
608 size = btrfs_file_extent_ram_bytes(eb, item);
609 nbytes = btrfs_file_extent_ram_bytes(eb, item);
610 extent_end = ALIGN(start + size,
611 fs_info->sectorsize);
617 inode = read_one_inode(root, key->objectid);
624 * first check to see if we already have this extent in the
625 * file. This must be done before the btrfs_drop_extents run
626 * so we don't try to drop this extent.
628 ret = btrfs_lookup_file_extent(trans, root, path,
629 btrfs_ino(BTRFS_I(inode)), start, 0);
632 (found_type == BTRFS_FILE_EXTENT_REG ||
633 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
634 struct btrfs_file_extent_item cmp1;
635 struct btrfs_file_extent_item cmp2;
636 struct btrfs_file_extent_item *existing;
637 struct extent_buffer *leaf;
639 leaf = path->nodes[0];
640 existing = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_file_extent_item);
643 read_extent_buffer(eb, &cmp1, (unsigned long)item,
645 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
649 * we already have a pointer to this exact extent,
650 * we don't have to do anything
652 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
653 btrfs_release_path(path);
657 btrfs_release_path(path);
659 /* drop any overlapping extents */
660 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
664 if (found_type == BTRFS_FILE_EXTENT_REG ||
665 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
667 unsigned long dest_offset;
668 struct btrfs_key ins;
670 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
671 btrfs_fs_incompat(fs_info, NO_HOLES))
674 ret = btrfs_insert_empty_item(trans, root, path, key,
678 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
680 copy_extent_buffer(path->nodes[0], eb, dest_offset,
681 (unsigned long)item, sizeof(*item));
683 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
684 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
685 ins.type = BTRFS_EXTENT_ITEM_KEY;
686 offset = key->offset - btrfs_file_extent_offset(eb, item);
689 * Manually record dirty extent, as here we did a shallow
690 * file extent item copy and skip normal backref update,
691 * but modifying extent tree all by ourselves.
692 * So need to manually record dirty extent for qgroup,
693 * as the owner of the file extent changed from log tree
694 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
696 ret = btrfs_qgroup_trace_extent(trans,
697 btrfs_file_extent_disk_bytenr(eb, item),
698 btrfs_file_extent_disk_num_bytes(eb, item),
703 if (ins.objectid > 0) {
704 struct btrfs_ref ref = { 0 };
707 LIST_HEAD(ordered_sums);
710 * is this extent already allocated in the extent
711 * allocation tree? If so, just add a reference
713 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
716 btrfs_init_generic_ref(&ref,
717 BTRFS_ADD_DELAYED_REF,
718 ins.objectid, ins.offset, 0);
719 btrfs_init_data_ref(&ref,
720 root->root_key.objectid,
721 key->objectid, offset);
722 ret = btrfs_inc_extent_ref(trans, &ref);
727 * insert the extent pointer in the extent
730 ret = btrfs_alloc_logged_file_extent(trans,
731 root->root_key.objectid,
732 key->objectid, offset, &ins);
736 btrfs_release_path(path);
738 if (btrfs_file_extent_compression(eb, item)) {
739 csum_start = ins.objectid;
740 csum_end = csum_start + ins.offset;
742 csum_start = ins.objectid +
743 btrfs_file_extent_offset(eb, item);
744 csum_end = csum_start +
745 btrfs_file_extent_num_bytes(eb, item);
748 ret = btrfs_lookup_csums_range(root->log_root,
749 csum_start, csum_end - 1,
754 * Now delete all existing cums in the csum root that
755 * cover our range. We do this because we can have an
756 * extent that is completely referenced by one file
757 * extent item and partially referenced by another
758 * file extent item (like after using the clone or
759 * extent_same ioctls). In this case if we end up doing
760 * the replay of the one that partially references the
761 * extent first, and we do not do the csum deletion
762 * below, we can get 2 csum items in the csum tree that
763 * overlap each other. For example, imagine our log has
764 * the two following file extent items:
766 * key (257 EXTENT_DATA 409600)
767 * extent data disk byte 12845056 nr 102400
768 * extent data offset 20480 nr 20480 ram 102400
770 * key (257 EXTENT_DATA 819200)
771 * extent data disk byte 12845056 nr 102400
772 * extent data offset 0 nr 102400 ram 102400
774 * Where the second one fully references the 100K extent
775 * that starts at disk byte 12845056, and the log tree
776 * has a single csum item that covers the entire range
779 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
781 * After the first file extent item is replayed, the
782 * csum tree gets the following csum item:
784 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
786 * Which covers the 20K sub-range starting at offset 20K
787 * of our extent. Now when we replay the second file
788 * extent item, if we do not delete existing csum items
789 * that cover any of its blocks, we end up getting two
790 * csum items in our csum tree that overlap each other:
792 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
793 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
795 * Which is a problem, because after this anyone trying
796 * to lookup up for the checksum of any block of our
797 * extent starting at an offset of 40K or higher, will
798 * end up looking at the second csum item only, which
799 * does not contain the checksum for any block starting
800 * at offset 40K or higher of our extent.
802 while (!list_empty(&ordered_sums)) {
803 struct btrfs_ordered_sum *sums;
804 sums = list_entry(ordered_sums.next,
805 struct btrfs_ordered_sum,
808 ret = btrfs_del_csums(trans,
813 ret = btrfs_csum_file_blocks(trans,
814 fs_info->csum_root, sums);
815 list_del(&sums->list);
821 btrfs_release_path(path);
823 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
824 /* inline extents are easy, we just overwrite them */
825 ret = overwrite_item(trans, root, path, eb, slot, key);
830 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
835 inode_add_bytes(inode, nbytes);
837 ret = btrfs_update_inode(trans, root, inode);
845 * when cleaning up conflicts between the directory names in the
846 * subvolume, directory names in the log and directory names in the
847 * inode back references, we may have to unlink inodes from directories.
849 * This is a helper function to do the unlink of a specific directory
852 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
853 struct btrfs_root *root,
854 struct btrfs_path *path,
855 struct btrfs_inode *dir,
856 struct btrfs_dir_item *di)
861 struct extent_buffer *leaf;
862 struct btrfs_key location;
865 leaf = path->nodes[0];
867 btrfs_dir_item_key_to_cpu(leaf, di, &location);
868 name_len = btrfs_dir_name_len(leaf, di);
869 name = kmalloc(name_len, GFP_NOFS);
873 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
874 btrfs_release_path(path);
876 inode = read_one_inode(root, location.objectid);
882 ret = link_to_fixup_dir(trans, root, path, location.objectid);
886 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
891 ret = btrfs_run_delayed_items(trans);
899 * helper function to see if a given name and sequence number found
900 * in an inode back reference are already in a directory and correctly
901 * point to this inode
903 static noinline int inode_in_dir(struct btrfs_root *root,
904 struct btrfs_path *path,
905 u64 dirid, u64 objectid, u64 index,
906 const char *name, int name_len)
908 struct btrfs_dir_item *di;
909 struct btrfs_key location;
912 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
913 index, name, name_len, 0);
914 if (di && !IS_ERR(di)) {
915 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
916 if (location.objectid != objectid)
920 btrfs_release_path(path);
922 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
923 if (di && !IS_ERR(di)) {
924 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
925 if (location.objectid != objectid)
931 btrfs_release_path(path);
936 * helper function to check a log tree for a named back reference in
937 * an inode. This is used to decide if a back reference that is
938 * found in the subvolume conflicts with what we find in the log.
940 * inode backreferences may have multiple refs in a single item,
941 * during replay we process one reference at a time, and we don't
942 * want to delete valid links to a file from the subvolume if that
943 * link is also in the log.
945 static noinline int backref_in_log(struct btrfs_root *log,
946 struct btrfs_key *key,
948 const char *name, int namelen)
950 struct btrfs_path *path;
953 path = btrfs_alloc_path();
957 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
960 } else if (ret == 1) {
965 if (key->type == BTRFS_INODE_EXTREF_KEY)
966 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
971 ret = !!btrfs_find_name_in_backref(path->nodes[0],
975 btrfs_free_path(path);
979 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
981 struct btrfs_path *path,
982 struct btrfs_root *log_root,
983 struct btrfs_inode *dir,
984 struct btrfs_inode *inode,
985 u64 inode_objectid, u64 parent_objectid,
986 u64 ref_index, char *name, int namelen,
992 struct extent_buffer *leaf;
993 struct btrfs_dir_item *di;
994 struct btrfs_key search_key;
995 struct btrfs_inode_extref *extref;
998 /* Search old style refs */
999 search_key.objectid = inode_objectid;
1000 search_key.type = BTRFS_INODE_REF_KEY;
1001 search_key.offset = parent_objectid;
1002 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1004 struct btrfs_inode_ref *victim_ref;
1006 unsigned long ptr_end;
1008 leaf = path->nodes[0];
1010 /* are we trying to overwrite a back ref for the root directory
1011 * if so, just jump out, we're done
1013 if (search_key.objectid == search_key.offset)
1016 /* check all the names in this back reference to see
1017 * if they are in the log. if so, we allow them to stay
1018 * otherwise they must be unlinked as a conflict
1020 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1021 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1022 while (ptr < ptr_end) {
1023 victim_ref = (struct btrfs_inode_ref *)ptr;
1024 victim_name_len = btrfs_inode_ref_name_len(leaf,
1026 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1030 read_extent_buffer(leaf, victim_name,
1031 (unsigned long)(victim_ref + 1),
1034 ret = backref_in_log(log_root, &search_key,
1035 parent_objectid, victim_name,
1041 inc_nlink(&inode->vfs_inode);
1042 btrfs_release_path(path);
1044 ret = btrfs_unlink_inode(trans, root, dir, inode,
1045 victim_name, victim_name_len);
1049 ret = btrfs_run_delayed_items(trans);
1057 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1061 * NOTE: we have searched root tree and checked the
1062 * corresponding ref, it does not need to check again.
1066 btrfs_release_path(path);
1068 /* Same search but for extended refs */
1069 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1070 inode_objectid, parent_objectid, 0,
1072 if (!IS_ERR_OR_NULL(extref)) {
1076 struct inode *victim_parent;
1078 leaf = path->nodes[0];
1080 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1081 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1083 while (cur_offset < item_size) {
1084 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1086 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1088 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1091 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1094 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1097 search_key.objectid = inode_objectid;
1098 search_key.type = BTRFS_INODE_EXTREF_KEY;
1099 search_key.offset = btrfs_extref_hash(parent_objectid,
1102 ret = backref_in_log(log_root, &search_key,
1103 parent_objectid, victim_name,
1109 victim_parent = read_one_inode(root,
1111 if (victim_parent) {
1112 inc_nlink(&inode->vfs_inode);
1113 btrfs_release_path(path);
1115 ret = btrfs_unlink_inode(trans, root,
1116 BTRFS_I(victim_parent),
1121 ret = btrfs_run_delayed_items(
1124 iput(victim_parent);
1133 cur_offset += victim_name_len + sizeof(*extref);
1137 btrfs_release_path(path);
1139 /* look for a conflicting sequence number */
1140 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1141 ref_index, name, namelen, 0);
1142 if (di && !IS_ERR(di)) {
1143 ret = drop_one_dir_item(trans, root, path, dir, di);
1147 btrfs_release_path(path);
1149 /* look for a conflicting name */
1150 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1152 if (di && !IS_ERR(di)) {
1153 ret = drop_one_dir_item(trans, root, path, dir, di);
1157 btrfs_release_path(path);
1162 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1163 u32 *namelen, char **name, u64 *index,
1164 u64 *parent_objectid)
1166 struct btrfs_inode_extref *extref;
1168 extref = (struct btrfs_inode_extref *)ref_ptr;
1170 *namelen = btrfs_inode_extref_name_len(eb, extref);
1171 *name = kmalloc(*namelen, GFP_NOFS);
1175 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1179 *index = btrfs_inode_extref_index(eb, extref);
1180 if (parent_objectid)
1181 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1186 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1187 u32 *namelen, char **name, u64 *index)
1189 struct btrfs_inode_ref *ref;
1191 ref = (struct btrfs_inode_ref *)ref_ptr;
1193 *namelen = btrfs_inode_ref_name_len(eb, ref);
1194 *name = kmalloc(*namelen, GFP_NOFS);
1198 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1201 *index = btrfs_inode_ref_index(eb, ref);
1207 * Take an inode reference item from the log tree and iterate all names from the
1208 * inode reference item in the subvolume tree with the same key (if it exists).
1209 * For any name that is not in the inode reference item from the log tree, do a
1210 * proper unlink of that name (that is, remove its entry from the inode
1211 * reference item and both dir index keys).
1213 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1214 struct btrfs_root *root,
1215 struct btrfs_path *path,
1216 struct btrfs_inode *inode,
1217 struct extent_buffer *log_eb,
1219 struct btrfs_key *key)
1222 unsigned long ref_ptr;
1223 unsigned long ref_end;
1224 struct extent_buffer *eb;
1227 btrfs_release_path(path);
1228 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1236 eb = path->nodes[0];
1237 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1238 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1239 while (ref_ptr < ref_end) {
1244 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1245 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1248 parent_id = key->offset;
1249 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1255 if (key->type == BTRFS_INODE_EXTREF_KEY)
1256 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1260 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1266 btrfs_release_path(path);
1267 dir = read_one_inode(root, parent_id);
1273 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1274 inode, name, namelen);
1284 if (key->type == BTRFS_INODE_EXTREF_KEY)
1285 ref_ptr += sizeof(struct btrfs_inode_extref);
1287 ref_ptr += sizeof(struct btrfs_inode_ref);
1291 btrfs_release_path(path);
1295 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1296 const u8 ref_type, const char *name,
1299 struct btrfs_key key;
1300 struct btrfs_path *path;
1301 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1304 path = btrfs_alloc_path();
1308 key.objectid = btrfs_ino(BTRFS_I(inode));
1309 key.type = ref_type;
1310 if (key.type == BTRFS_INODE_REF_KEY)
1311 key.offset = parent_id;
1313 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1315 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1322 if (key.type == BTRFS_INODE_EXTREF_KEY)
1323 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1324 path->slots[0], parent_id, name, namelen);
1326 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1330 btrfs_free_path(path);
1334 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1335 struct inode *dir, struct inode *inode, const char *name,
1336 int namelen, u64 ref_index)
1338 struct btrfs_dir_item *dir_item;
1339 struct btrfs_key key;
1340 struct btrfs_path *path;
1341 struct inode *other_inode = NULL;
1344 path = btrfs_alloc_path();
1348 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1349 btrfs_ino(BTRFS_I(dir)),
1352 btrfs_release_path(path);
1354 } else if (IS_ERR(dir_item)) {
1355 ret = PTR_ERR(dir_item);
1360 * Our inode's dentry collides with the dentry of another inode which is
1361 * in the log but not yet processed since it has a higher inode number.
1362 * So delete that other dentry.
1364 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1365 btrfs_release_path(path);
1366 other_inode = read_one_inode(root, key.objectid);
1371 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1376 * If we dropped the link count to 0, bump it so that later the iput()
1377 * on the inode will not free it. We will fixup the link count later.
1379 if (other_inode->i_nlink == 0)
1380 inc_nlink(other_inode);
1382 ret = btrfs_run_delayed_items(trans);
1386 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1387 name, namelen, 0, ref_index);
1390 btrfs_free_path(path);
1396 * replay one inode back reference item found in the log tree.
1397 * eb, slot and key refer to the buffer and key found in the log tree.
1398 * root is the destination we are replaying into, and path is for temp
1399 * use by this function. (it should be released on return).
1401 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root,
1403 struct btrfs_root *log,
1404 struct btrfs_path *path,
1405 struct extent_buffer *eb, int slot,
1406 struct btrfs_key *key)
1408 struct inode *dir = NULL;
1409 struct inode *inode = NULL;
1410 unsigned long ref_ptr;
1411 unsigned long ref_end;
1415 int search_done = 0;
1416 int log_ref_ver = 0;
1417 u64 parent_objectid;
1420 int ref_struct_size;
1422 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1423 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1425 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1426 struct btrfs_inode_extref *r;
1428 ref_struct_size = sizeof(struct btrfs_inode_extref);
1430 r = (struct btrfs_inode_extref *)ref_ptr;
1431 parent_objectid = btrfs_inode_extref_parent(eb, r);
1433 ref_struct_size = sizeof(struct btrfs_inode_ref);
1434 parent_objectid = key->offset;
1436 inode_objectid = key->objectid;
1439 * it is possible that we didn't log all the parent directories
1440 * for a given inode. If we don't find the dir, just don't
1441 * copy the back ref in. The link count fixup code will take
1444 dir = read_one_inode(root, parent_objectid);
1450 inode = read_one_inode(root, inode_objectid);
1456 while (ref_ptr < ref_end) {
1458 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1459 &ref_index, &parent_objectid);
1461 * parent object can change from one array
1465 dir = read_one_inode(root, parent_objectid);
1471 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1477 /* if we already have a perfect match, we're done */
1478 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1479 btrfs_ino(BTRFS_I(inode)), ref_index,
1482 * look for a conflicting back reference in the
1483 * metadata. if we find one we have to unlink that name
1484 * of the file before we add our new link. Later on, we
1485 * overwrite any existing back reference, and we don't
1486 * want to create dangling pointers in the directory.
1490 ret = __add_inode_ref(trans, root, path, log,
1495 ref_index, name, namelen,
1505 * If a reference item already exists for this inode
1506 * with the same parent and name, but different index,
1507 * drop it and the corresponding directory index entries
1508 * from the parent before adding the new reference item
1509 * and dir index entries, otherwise we would fail with
1510 * -EEXIST returned from btrfs_add_link() below.
1512 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1515 ret = btrfs_unlink_inode(trans, root,
1520 * If we dropped the link count to 0, bump it so
1521 * that later the iput() on the inode will not
1522 * free it. We will fixup the link count later.
1524 if (!ret && inode->i_nlink == 0)
1530 /* insert our name */
1531 ret = add_link(trans, root, dir, inode, name, namelen,
1536 btrfs_update_inode(trans, root, inode);
1539 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1549 * Before we overwrite the inode reference item in the subvolume tree
1550 * with the item from the log tree, we must unlink all names from the
1551 * parent directory that are in the subvolume's tree inode reference
1552 * item, otherwise we end up with an inconsistent subvolume tree where
1553 * dir index entries exist for a name but there is no inode reference
1554 * item with the same name.
1556 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1561 /* finally write the back reference in the inode */
1562 ret = overwrite_item(trans, root, path, eb, slot, key);
1564 btrfs_release_path(path);
1571 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1572 struct btrfs_root *root, u64 ino)
1576 ret = btrfs_insert_orphan_item(trans, root, ino);
1583 static int count_inode_extrefs(struct btrfs_root *root,
1584 struct btrfs_inode *inode, struct btrfs_path *path)
1588 unsigned int nlink = 0;
1591 u64 inode_objectid = btrfs_ino(inode);
1594 struct btrfs_inode_extref *extref;
1595 struct extent_buffer *leaf;
1598 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1603 leaf = path->nodes[0];
1604 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1605 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1608 while (cur_offset < item_size) {
1609 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1610 name_len = btrfs_inode_extref_name_len(leaf, extref);
1614 cur_offset += name_len + sizeof(*extref);
1618 btrfs_release_path(path);
1620 btrfs_release_path(path);
1622 if (ret < 0 && ret != -ENOENT)
1627 static int count_inode_refs(struct btrfs_root *root,
1628 struct btrfs_inode *inode, struct btrfs_path *path)
1631 struct btrfs_key key;
1632 unsigned int nlink = 0;
1634 unsigned long ptr_end;
1636 u64 ino = btrfs_ino(inode);
1639 key.type = BTRFS_INODE_REF_KEY;
1640 key.offset = (u64)-1;
1643 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1647 if (path->slots[0] == 0)
1652 btrfs_item_key_to_cpu(path->nodes[0], &key,
1654 if (key.objectid != ino ||
1655 key.type != BTRFS_INODE_REF_KEY)
1657 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1658 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1660 while (ptr < ptr_end) {
1661 struct btrfs_inode_ref *ref;
1663 ref = (struct btrfs_inode_ref *)ptr;
1664 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1666 ptr = (unsigned long)(ref + 1) + name_len;
1670 if (key.offset == 0)
1672 if (path->slots[0] > 0) {
1677 btrfs_release_path(path);
1679 btrfs_release_path(path);
1685 * There are a few corners where the link count of the file can't
1686 * be properly maintained during replay. So, instead of adding
1687 * lots of complexity to the log code, we just scan the backrefs
1688 * for any file that has been through replay.
1690 * The scan will update the link count on the inode to reflect the
1691 * number of back refs found. If it goes down to zero, the iput
1692 * will free the inode.
1694 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root,
1696 struct inode *inode)
1698 struct btrfs_path *path;
1701 u64 ino = btrfs_ino(BTRFS_I(inode));
1703 path = btrfs_alloc_path();
1707 ret = count_inode_refs(root, BTRFS_I(inode), path);
1713 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1721 if (nlink != inode->i_nlink) {
1722 set_nlink(inode, nlink);
1723 btrfs_update_inode(trans, root, inode);
1725 BTRFS_I(inode)->index_cnt = (u64)-1;
1727 if (inode->i_nlink == 0) {
1728 if (S_ISDIR(inode->i_mode)) {
1729 ret = replay_dir_deletes(trans, root, NULL, path,
1734 ret = insert_orphan_item(trans, root, ino);
1738 btrfs_free_path(path);
1742 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1743 struct btrfs_root *root,
1744 struct btrfs_path *path)
1747 struct btrfs_key key;
1748 struct inode *inode;
1750 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1751 key.type = BTRFS_ORPHAN_ITEM_KEY;
1752 key.offset = (u64)-1;
1754 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1759 if (path->slots[0] == 0)
1764 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1765 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1766 key.type != BTRFS_ORPHAN_ITEM_KEY)
1769 ret = btrfs_del_item(trans, root, path);
1773 btrfs_release_path(path);
1774 inode = read_one_inode(root, key.offset);
1778 ret = fixup_inode_link_count(trans, root, inode);
1784 * fixup on a directory may create new entries,
1785 * make sure we always look for the highset possible
1788 key.offset = (u64)-1;
1792 btrfs_release_path(path);
1798 * record a given inode in the fixup dir so we can check its link
1799 * count when replay is done. The link count is incremented here
1800 * so the inode won't go away until we check it
1802 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_path *path,
1807 struct btrfs_key key;
1809 struct inode *inode;
1811 inode = read_one_inode(root, objectid);
1815 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1816 key.type = BTRFS_ORPHAN_ITEM_KEY;
1817 key.offset = objectid;
1819 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1821 btrfs_release_path(path);
1823 if (!inode->i_nlink)
1824 set_nlink(inode, 1);
1827 ret = btrfs_update_inode(trans, root, inode);
1828 } else if (ret == -EEXIST) {
1831 BUG(); /* Logic Error */
1839 * when replaying the log for a directory, we only insert names
1840 * for inodes that actually exist. This means an fsync on a directory
1841 * does not implicitly fsync all the new files in it
1843 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1844 struct btrfs_root *root,
1845 u64 dirid, u64 index,
1846 char *name, int name_len,
1847 struct btrfs_key *location)
1849 struct inode *inode;
1853 inode = read_one_inode(root, location->objectid);
1857 dir = read_one_inode(root, dirid);
1863 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1864 name_len, 1, index);
1866 /* FIXME, put inode into FIXUP list */
1874 * take a single entry in a log directory item and replay it into
1877 * if a conflicting item exists in the subdirectory already,
1878 * the inode it points to is unlinked and put into the link count
1881 * If a name from the log points to a file or directory that does
1882 * not exist in the FS, it is skipped. fsyncs on directories
1883 * do not force down inodes inside that directory, just changes to the
1884 * names or unlinks in a directory.
1886 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1887 * non-existing inode) and 1 if the name was replayed.
1889 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1890 struct btrfs_root *root,
1891 struct btrfs_path *path,
1892 struct extent_buffer *eb,
1893 struct btrfs_dir_item *di,
1894 struct btrfs_key *key)
1898 struct btrfs_dir_item *dst_di;
1899 struct btrfs_key found_key;
1900 struct btrfs_key log_key;
1905 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1906 bool name_added = false;
1908 dir = read_one_inode(root, key->objectid);
1912 name_len = btrfs_dir_name_len(eb, di);
1913 name = kmalloc(name_len, GFP_NOFS);
1919 log_type = btrfs_dir_type(eb, di);
1920 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1923 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1924 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1929 btrfs_release_path(path);
1931 if (key->type == BTRFS_DIR_ITEM_KEY) {
1932 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1934 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1935 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1944 if (IS_ERR_OR_NULL(dst_di)) {
1945 /* we need a sequence number to insert, so we only
1946 * do inserts for the BTRFS_DIR_INDEX_KEY types
1948 if (key->type != BTRFS_DIR_INDEX_KEY)
1953 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1954 /* the existing item matches the logged item */
1955 if (found_key.objectid == log_key.objectid &&
1956 found_key.type == log_key.type &&
1957 found_key.offset == log_key.offset &&
1958 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1959 update_size = false;
1964 * don't drop the conflicting directory entry if the inode
1965 * for the new entry doesn't exist
1970 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1974 if (key->type == BTRFS_DIR_INDEX_KEY)
1977 btrfs_release_path(path);
1978 if (!ret && update_size) {
1979 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1980 ret = btrfs_update_inode(trans, root, dir);
1984 if (!ret && name_added)
1990 * Check if the inode reference exists in the log for the given name,
1991 * inode and parent inode
1993 found_key.objectid = log_key.objectid;
1994 found_key.type = BTRFS_INODE_REF_KEY;
1995 found_key.offset = key->objectid;
1996 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
2000 /* The dentry will be added later. */
2002 update_size = false;
2006 found_key.objectid = log_key.objectid;
2007 found_key.type = BTRFS_INODE_EXTREF_KEY;
2008 found_key.offset = key->objectid;
2009 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2014 /* The dentry will be added later. */
2016 update_size = false;
2019 btrfs_release_path(path);
2020 ret = insert_one_name(trans, root, key->objectid, key->offset,
2021 name, name_len, &log_key);
2022 if (ret && ret != -ENOENT && ret != -EEXIST)
2026 update_size = false;
2032 * find all the names in a directory item and reconcile them into
2033 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2034 * one name in a directory item, but the same code gets used for
2035 * both directory index types
2037 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2038 struct btrfs_root *root,
2039 struct btrfs_path *path,
2040 struct extent_buffer *eb, int slot,
2041 struct btrfs_key *key)
2044 u32 item_size = btrfs_item_size_nr(eb, slot);
2045 struct btrfs_dir_item *di;
2048 unsigned long ptr_end;
2049 struct btrfs_path *fixup_path = NULL;
2051 ptr = btrfs_item_ptr_offset(eb, slot);
2052 ptr_end = ptr + item_size;
2053 while (ptr < ptr_end) {
2054 di = (struct btrfs_dir_item *)ptr;
2055 name_len = btrfs_dir_name_len(eb, di);
2056 ret = replay_one_name(trans, root, path, eb, di, key);
2059 ptr = (unsigned long)(di + 1);
2063 * If this entry refers to a non-directory (directories can not
2064 * have a link count > 1) and it was added in the transaction
2065 * that was not committed, make sure we fixup the link count of
2066 * the inode it the entry points to. Otherwise something like
2067 * the following would result in a directory pointing to an
2068 * inode with a wrong link that does not account for this dir
2076 * ln testdir/bar testdir/bar_link
2077 * ln testdir/foo testdir/foo_link
2078 * xfs_io -c "fsync" testdir/bar
2082 * mount fs, log replay happens
2084 * File foo would remain with a link count of 1 when it has two
2085 * entries pointing to it in the directory testdir. This would
2086 * make it impossible to ever delete the parent directory has
2087 * it would result in stale dentries that can never be deleted.
2089 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2090 struct btrfs_key di_key;
2093 fixup_path = btrfs_alloc_path();
2100 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2101 ret = link_to_fixup_dir(trans, root, fixup_path,
2108 btrfs_free_path(fixup_path);
2113 * directory replay has two parts. There are the standard directory
2114 * items in the log copied from the subvolume, and range items
2115 * created in the log while the subvolume was logged.
2117 * The range items tell us which parts of the key space the log
2118 * is authoritative for. During replay, if a key in the subvolume
2119 * directory is in a logged range item, but not actually in the log
2120 * that means it was deleted from the directory before the fsync
2121 * and should be removed.
2123 static noinline int find_dir_range(struct btrfs_root *root,
2124 struct btrfs_path *path,
2125 u64 dirid, int key_type,
2126 u64 *start_ret, u64 *end_ret)
2128 struct btrfs_key key;
2130 struct btrfs_dir_log_item *item;
2134 if (*start_ret == (u64)-1)
2137 key.objectid = dirid;
2138 key.type = key_type;
2139 key.offset = *start_ret;
2141 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2145 if (path->slots[0] == 0)
2150 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2152 if (key.type != key_type || key.objectid != dirid) {
2156 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2157 struct btrfs_dir_log_item);
2158 found_end = btrfs_dir_log_end(path->nodes[0], item);
2160 if (*start_ret >= key.offset && *start_ret <= found_end) {
2162 *start_ret = key.offset;
2163 *end_ret = found_end;
2168 /* check the next slot in the tree to see if it is a valid item */
2169 nritems = btrfs_header_nritems(path->nodes[0]);
2171 if (path->slots[0] >= nritems) {
2172 ret = btrfs_next_leaf(root, path);
2177 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2179 if (key.type != key_type || key.objectid != dirid) {
2183 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2184 struct btrfs_dir_log_item);
2185 found_end = btrfs_dir_log_end(path->nodes[0], item);
2186 *start_ret = key.offset;
2187 *end_ret = found_end;
2190 btrfs_release_path(path);
2195 * this looks for a given directory item in the log. If the directory
2196 * item is not in the log, the item is removed and the inode it points
2199 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2200 struct btrfs_root *root,
2201 struct btrfs_root *log,
2202 struct btrfs_path *path,
2203 struct btrfs_path *log_path,
2205 struct btrfs_key *dir_key)
2208 struct extent_buffer *eb;
2211 struct btrfs_dir_item *di;
2212 struct btrfs_dir_item *log_di;
2215 unsigned long ptr_end;
2217 struct inode *inode;
2218 struct btrfs_key location;
2221 eb = path->nodes[0];
2222 slot = path->slots[0];
2223 item_size = btrfs_item_size_nr(eb, slot);
2224 ptr = btrfs_item_ptr_offset(eb, slot);
2225 ptr_end = ptr + item_size;
2226 while (ptr < ptr_end) {
2227 di = (struct btrfs_dir_item *)ptr;
2228 name_len = btrfs_dir_name_len(eb, di);
2229 name = kmalloc(name_len, GFP_NOFS);
2234 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2237 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2238 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2241 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2242 log_di = btrfs_lookup_dir_index_item(trans, log,
2248 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2249 btrfs_dir_item_key_to_cpu(eb, di, &location);
2250 btrfs_release_path(path);
2251 btrfs_release_path(log_path);
2252 inode = read_one_inode(root, location.objectid);
2258 ret = link_to_fixup_dir(trans, root,
2259 path, location.objectid);
2267 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2268 BTRFS_I(inode), name, name_len);
2270 ret = btrfs_run_delayed_items(trans);
2276 /* there might still be more names under this key
2277 * check and repeat if required
2279 ret = btrfs_search_slot(NULL, root, dir_key, path,
2285 } else if (IS_ERR(log_di)) {
2287 return PTR_ERR(log_di);
2289 btrfs_release_path(log_path);
2292 ptr = (unsigned long)(di + 1);
2297 btrfs_release_path(path);
2298 btrfs_release_path(log_path);
2302 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2303 struct btrfs_root *root,
2304 struct btrfs_root *log,
2305 struct btrfs_path *path,
2308 struct btrfs_key search_key;
2309 struct btrfs_path *log_path;
2314 log_path = btrfs_alloc_path();
2318 search_key.objectid = ino;
2319 search_key.type = BTRFS_XATTR_ITEM_KEY;
2320 search_key.offset = 0;
2322 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2326 nritems = btrfs_header_nritems(path->nodes[0]);
2327 for (i = path->slots[0]; i < nritems; i++) {
2328 struct btrfs_key key;
2329 struct btrfs_dir_item *di;
2330 struct btrfs_dir_item *log_di;
2334 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2335 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2340 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2341 total_size = btrfs_item_size_nr(path->nodes[0], i);
2343 while (cur < total_size) {
2344 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2345 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2346 u32 this_len = sizeof(*di) + name_len + data_len;
2349 name = kmalloc(name_len, GFP_NOFS);
2354 read_extent_buffer(path->nodes[0], name,
2355 (unsigned long)(di + 1), name_len);
2357 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2359 btrfs_release_path(log_path);
2361 /* Doesn't exist in log tree, so delete it. */
2362 btrfs_release_path(path);
2363 di = btrfs_lookup_xattr(trans, root, path, ino,
2364 name, name_len, -1);
2371 ret = btrfs_delete_one_dir_name(trans, root,
2375 btrfs_release_path(path);
2380 if (IS_ERR(log_di)) {
2381 ret = PTR_ERR(log_di);
2385 di = (struct btrfs_dir_item *)((char *)di + this_len);
2388 ret = btrfs_next_leaf(root, path);
2394 btrfs_free_path(log_path);
2395 btrfs_release_path(path);
2401 * deletion replay happens before we copy any new directory items
2402 * out of the log or out of backreferences from inodes. It
2403 * scans the log to find ranges of keys that log is authoritative for,
2404 * and then scans the directory to find items in those ranges that are
2405 * not present in the log.
2407 * Anything we don't find in the log is unlinked and removed from the
2410 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2411 struct btrfs_root *root,
2412 struct btrfs_root *log,
2413 struct btrfs_path *path,
2414 u64 dirid, int del_all)
2418 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2420 struct btrfs_key dir_key;
2421 struct btrfs_key found_key;
2422 struct btrfs_path *log_path;
2425 dir_key.objectid = dirid;
2426 dir_key.type = BTRFS_DIR_ITEM_KEY;
2427 log_path = btrfs_alloc_path();
2431 dir = read_one_inode(root, dirid);
2432 /* it isn't an error if the inode isn't there, that can happen
2433 * because we replay the deletes before we copy in the inode item
2437 btrfs_free_path(log_path);
2445 range_end = (u64)-1;
2447 ret = find_dir_range(log, path, dirid, key_type,
2448 &range_start, &range_end);
2453 dir_key.offset = range_start;
2456 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2461 nritems = btrfs_header_nritems(path->nodes[0]);
2462 if (path->slots[0] >= nritems) {
2463 ret = btrfs_next_leaf(root, path);
2469 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2471 if (found_key.objectid != dirid ||
2472 found_key.type != dir_key.type)
2475 if (found_key.offset > range_end)
2478 ret = check_item_in_log(trans, root, log, path,
2483 if (found_key.offset == (u64)-1)
2485 dir_key.offset = found_key.offset + 1;
2487 btrfs_release_path(path);
2488 if (range_end == (u64)-1)
2490 range_start = range_end + 1;
2495 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2496 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2497 dir_key.type = BTRFS_DIR_INDEX_KEY;
2498 btrfs_release_path(path);
2502 btrfs_release_path(path);
2503 btrfs_free_path(log_path);
2509 * the process_func used to replay items from the log tree. This
2510 * gets called in two different stages. The first stage just looks
2511 * for inodes and makes sure they are all copied into the subvolume.
2513 * The second stage copies all the other item types from the log into
2514 * the subvolume. The two stage approach is slower, but gets rid of
2515 * lots of complexity around inodes referencing other inodes that exist
2516 * only in the log (references come from either directory items or inode
2519 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2520 struct walk_control *wc, u64 gen, int level)
2523 struct btrfs_path *path;
2524 struct btrfs_root *root = wc->replay_dest;
2525 struct btrfs_key key;
2529 ret = btrfs_read_buffer(eb, gen, level, NULL);
2533 level = btrfs_header_level(eb);
2538 path = btrfs_alloc_path();
2542 nritems = btrfs_header_nritems(eb);
2543 for (i = 0; i < nritems; i++) {
2544 btrfs_item_key_to_cpu(eb, &key, i);
2546 /* inode keys are done during the first stage */
2547 if (key.type == BTRFS_INODE_ITEM_KEY &&
2548 wc->stage == LOG_WALK_REPLAY_INODES) {
2549 struct btrfs_inode_item *inode_item;
2552 inode_item = btrfs_item_ptr(eb, i,
2553 struct btrfs_inode_item);
2555 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2556 * and never got linked before the fsync, skip it, as
2557 * replaying it is pointless since it would be deleted
2558 * later. We skip logging tmpfiles, but it's always
2559 * possible we are replaying a log created with a kernel
2560 * that used to log tmpfiles.
2562 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2563 wc->ignore_cur_inode = true;
2566 wc->ignore_cur_inode = false;
2568 ret = replay_xattr_deletes(wc->trans, root, log,
2569 path, key.objectid);
2572 mode = btrfs_inode_mode(eb, inode_item);
2573 if (S_ISDIR(mode)) {
2574 ret = replay_dir_deletes(wc->trans,
2575 root, log, path, key.objectid, 0);
2579 ret = overwrite_item(wc->trans, root, path,
2585 * Before replaying extents, truncate the inode to its
2586 * size. We need to do it now and not after log replay
2587 * because before an fsync we can have prealloc extents
2588 * added beyond the inode's i_size. If we did it after,
2589 * through orphan cleanup for example, we would drop
2590 * those prealloc extents just after replaying them.
2592 if (S_ISREG(mode)) {
2593 struct inode *inode;
2596 inode = read_one_inode(root, key.objectid);
2601 from = ALIGN(i_size_read(inode),
2602 root->fs_info->sectorsize);
2603 ret = btrfs_drop_extents(wc->trans, root, inode,
2606 /* Update the inode's nbytes. */
2607 ret = btrfs_update_inode(wc->trans,
2615 ret = link_to_fixup_dir(wc->trans, root,
2616 path, key.objectid);
2621 if (wc->ignore_cur_inode)
2624 if (key.type == BTRFS_DIR_INDEX_KEY &&
2625 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2626 ret = replay_one_dir_item(wc->trans, root, path,
2632 if (wc->stage < LOG_WALK_REPLAY_ALL)
2635 /* these keys are simply copied */
2636 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2637 ret = overwrite_item(wc->trans, root, path,
2641 } else if (key.type == BTRFS_INODE_REF_KEY ||
2642 key.type == BTRFS_INODE_EXTREF_KEY) {
2643 ret = add_inode_ref(wc->trans, root, log, path,
2645 if (ret && ret != -ENOENT)
2648 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2649 ret = replay_one_extent(wc->trans, root, path,
2653 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2654 ret = replay_one_dir_item(wc->trans, root, path,
2660 btrfs_free_path(path);
2665 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2667 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2669 struct btrfs_block_group *cache;
2671 cache = btrfs_lookup_block_group(fs_info, start);
2673 btrfs_err(fs_info, "unable to find block group for %llu", start);
2677 spin_lock(&cache->space_info->lock);
2678 spin_lock(&cache->lock);
2679 cache->reserved -= fs_info->nodesize;
2680 cache->space_info->bytes_reserved -= fs_info->nodesize;
2681 spin_unlock(&cache->lock);
2682 spin_unlock(&cache->space_info->lock);
2684 btrfs_put_block_group(cache);
2687 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2688 struct btrfs_root *root,
2689 struct btrfs_path *path, int *level,
2690 struct walk_control *wc)
2692 struct btrfs_fs_info *fs_info = root->fs_info;
2695 struct extent_buffer *next;
2696 struct extent_buffer *cur;
2700 while (*level > 0) {
2701 struct btrfs_key first_key;
2703 cur = path->nodes[*level];
2705 WARN_ON(btrfs_header_level(cur) != *level);
2707 if (path->slots[*level] >=
2708 btrfs_header_nritems(cur))
2711 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2712 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2713 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2714 blocksize = fs_info->nodesize;
2716 next = btrfs_find_create_tree_block(fs_info, bytenr);
2718 return PTR_ERR(next);
2721 ret = wc->process_func(root, next, wc, ptr_gen,
2724 free_extent_buffer(next);
2728 path->slots[*level]++;
2730 ret = btrfs_read_buffer(next, ptr_gen,
2731 *level - 1, &first_key);
2733 free_extent_buffer(next);
2738 btrfs_tree_lock(next);
2739 btrfs_set_lock_blocking_write(next);
2740 btrfs_clean_tree_block(next);
2741 btrfs_wait_tree_block_writeback(next);
2742 btrfs_tree_unlock(next);
2743 ret = btrfs_pin_reserved_extent(trans,
2746 free_extent_buffer(next);
2750 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2751 clear_extent_buffer_dirty(next);
2752 unaccount_log_buffer(fs_info, bytenr);
2755 free_extent_buffer(next);
2758 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2760 free_extent_buffer(next);
2764 if (path->nodes[*level-1])
2765 free_extent_buffer(path->nodes[*level-1]);
2766 path->nodes[*level-1] = next;
2767 *level = btrfs_header_level(next);
2768 path->slots[*level] = 0;
2771 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2777 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2778 struct btrfs_root *root,
2779 struct btrfs_path *path, int *level,
2780 struct walk_control *wc)
2782 struct btrfs_fs_info *fs_info = root->fs_info;
2787 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2788 slot = path->slots[i];
2789 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2792 WARN_ON(*level == 0);
2795 ret = wc->process_func(root, path->nodes[*level], wc,
2796 btrfs_header_generation(path->nodes[*level]),
2802 struct extent_buffer *next;
2804 next = path->nodes[*level];
2807 btrfs_tree_lock(next);
2808 btrfs_set_lock_blocking_write(next);
2809 btrfs_clean_tree_block(next);
2810 btrfs_wait_tree_block_writeback(next);
2811 btrfs_tree_unlock(next);
2812 ret = btrfs_pin_reserved_extent(trans,
2813 path->nodes[*level]->start,
2814 path->nodes[*level]->len);
2818 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2819 clear_extent_buffer_dirty(next);
2821 unaccount_log_buffer(fs_info,
2822 path->nodes[*level]->start);
2825 free_extent_buffer(path->nodes[*level]);
2826 path->nodes[*level] = NULL;
2834 * drop the reference count on the tree rooted at 'snap'. This traverses
2835 * the tree freeing any blocks that have a ref count of zero after being
2838 static int walk_log_tree(struct btrfs_trans_handle *trans,
2839 struct btrfs_root *log, struct walk_control *wc)
2841 struct btrfs_fs_info *fs_info = log->fs_info;
2845 struct btrfs_path *path;
2848 path = btrfs_alloc_path();
2852 level = btrfs_header_level(log->node);
2854 path->nodes[level] = log->node;
2855 atomic_inc(&log->node->refs);
2856 path->slots[level] = 0;
2859 wret = walk_down_log_tree(trans, log, path, &level, wc);
2867 wret = walk_up_log_tree(trans, log, path, &level, wc);
2876 /* was the root node processed? if not, catch it here */
2877 if (path->nodes[orig_level]) {
2878 ret = wc->process_func(log, path->nodes[orig_level], wc,
2879 btrfs_header_generation(path->nodes[orig_level]),
2884 struct extent_buffer *next;
2886 next = path->nodes[orig_level];
2889 btrfs_tree_lock(next);
2890 btrfs_set_lock_blocking_write(next);
2891 btrfs_clean_tree_block(next);
2892 btrfs_wait_tree_block_writeback(next);
2893 btrfs_tree_unlock(next);
2894 ret = btrfs_pin_reserved_extent(trans,
2895 next->start, next->len);
2899 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2900 clear_extent_buffer_dirty(next);
2901 unaccount_log_buffer(fs_info, next->start);
2907 btrfs_free_path(path);
2912 * helper function to update the item for a given subvolumes log root
2913 * in the tree of log roots
2915 static int update_log_root(struct btrfs_trans_handle *trans,
2916 struct btrfs_root *log,
2917 struct btrfs_root_item *root_item)
2919 struct btrfs_fs_info *fs_info = log->fs_info;
2922 if (log->log_transid == 1) {
2923 /* insert root item on the first sync */
2924 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2925 &log->root_key, root_item);
2927 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2928 &log->root_key, root_item);
2933 static void wait_log_commit(struct btrfs_root *root, int transid)
2936 int index = transid % 2;
2939 * we only allow two pending log transactions at a time,
2940 * so we know that if ours is more than 2 older than the
2941 * current transaction, we're done
2944 prepare_to_wait(&root->log_commit_wait[index],
2945 &wait, TASK_UNINTERRUPTIBLE);
2947 if (!(root->log_transid_committed < transid &&
2948 atomic_read(&root->log_commit[index])))
2951 mutex_unlock(&root->log_mutex);
2953 mutex_lock(&root->log_mutex);
2955 finish_wait(&root->log_commit_wait[index], &wait);
2958 static void wait_for_writer(struct btrfs_root *root)
2963 prepare_to_wait(&root->log_writer_wait, &wait,
2964 TASK_UNINTERRUPTIBLE);
2965 if (!atomic_read(&root->log_writers))
2968 mutex_unlock(&root->log_mutex);
2970 mutex_lock(&root->log_mutex);
2972 finish_wait(&root->log_writer_wait, &wait);
2975 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2976 struct btrfs_log_ctx *ctx)
2981 mutex_lock(&root->log_mutex);
2982 list_del_init(&ctx->list);
2983 mutex_unlock(&root->log_mutex);
2987 * Invoked in log mutex context, or be sure there is no other task which
2988 * can access the list.
2990 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2991 int index, int error)
2993 struct btrfs_log_ctx *ctx;
2994 struct btrfs_log_ctx *safe;
2996 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2997 list_del_init(&ctx->list);
2998 ctx->log_ret = error;
3001 INIT_LIST_HEAD(&root->log_ctxs[index]);
3005 * btrfs_sync_log does sends a given tree log down to the disk and
3006 * updates the super blocks to record it. When this call is done,
3007 * you know that any inodes previously logged are safely on disk only
3010 * Any other return value means you need to call btrfs_commit_transaction.
3011 * Some of the edge cases for fsyncing directories that have had unlinks
3012 * or renames done in the past mean that sometimes the only safe
3013 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3014 * that has happened.
3016 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3017 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3023 struct btrfs_fs_info *fs_info = root->fs_info;
3024 struct btrfs_root *log = root->log_root;
3025 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3026 struct btrfs_root_item new_root_item;
3027 int log_transid = 0;
3028 struct btrfs_log_ctx root_log_ctx;
3029 struct blk_plug plug;
3031 mutex_lock(&root->log_mutex);
3032 log_transid = ctx->log_transid;
3033 if (root->log_transid_committed >= log_transid) {
3034 mutex_unlock(&root->log_mutex);
3035 return ctx->log_ret;
3038 index1 = log_transid % 2;
3039 if (atomic_read(&root->log_commit[index1])) {
3040 wait_log_commit(root, log_transid);
3041 mutex_unlock(&root->log_mutex);
3042 return ctx->log_ret;
3044 ASSERT(log_transid == root->log_transid);
3045 atomic_set(&root->log_commit[index1], 1);
3047 /* wait for previous tree log sync to complete */
3048 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3049 wait_log_commit(root, log_transid - 1);
3052 int batch = atomic_read(&root->log_batch);
3053 /* when we're on an ssd, just kick the log commit out */
3054 if (!btrfs_test_opt(fs_info, SSD) &&
3055 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3056 mutex_unlock(&root->log_mutex);
3057 schedule_timeout_uninterruptible(1);
3058 mutex_lock(&root->log_mutex);
3060 wait_for_writer(root);
3061 if (batch == atomic_read(&root->log_batch))
3065 /* bail out if we need to do a full commit */
3066 if (btrfs_need_log_full_commit(trans)) {
3068 mutex_unlock(&root->log_mutex);
3072 if (log_transid % 2 == 0)
3073 mark = EXTENT_DIRTY;
3077 /* we start IO on all the marked extents here, but we don't actually
3078 * wait for them until later.
3080 blk_start_plug(&plug);
3081 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3083 blk_finish_plug(&plug);
3084 btrfs_abort_transaction(trans, ret);
3085 btrfs_set_log_full_commit(trans);
3086 mutex_unlock(&root->log_mutex);
3091 * We _must_ update under the root->log_mutex in order to make sure we
3092 * have a consistent view of the log root we are trying to commit at
3095 * We _must_ copy this into a local copy, because we are not holding the
3096 * log_root_tree->log_mutex yet. This is important because when we
3097 * commit the log_root_tree we must have a consistent view of the
3098 * log_root_tree when we update the super block to point at the
3099 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3100 * with the commit and possibly point at the new block which we may not
3103 btrfs_set_root_node(&log->root_item, log->node);
3104 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3106 root->log_transid++;
3107 log->log_transid = root->log_transid;
3108 root->log_start_pid = 0;
3110 * IO has been started, blocks of the log tree have WRITTEN flag set
3111 * in their headers. new modifications of the log will be written to
3112 * new positions. so it's safe to allow log writers to go in.
3114 mutex_unlock(&root->log_mutex);
3116 btrfs_init_log_ctx(&root_log_ctx, NULL);
3118 mutex_lock(&log_root_tree->log_mutex);
3119 atomic_inc(&log_root_tree->log_batch);
3120 atomic_inc(&log_root_tree->log_writers);
3122 index2 = log_root_tree->log_transid % 2;
3123 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3124 root_log_ctx.log_transid = log_root_tree->log_transid;
3126 mutex_unlock(&log_root_tree->log_mutex);
3128 mutex_lock(&log_root_tree->log_mutex);
3131 * Now we are safe to update the log_root_tree because we're under the
3132 * log_mutex, and we're a current writer so we're holding the commit
3133 * open until we drop the log_mutex.
3135 ret = update_log_root(trans, log, &new_root_item);
3137 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3138 /* atomic_dec_and_test implies a barrier */
3139 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3143 if (!list_empty(&root_log_ctx.list))
3144 list_del_init(&root_log_ctx.list);
3146 blk_finish_plug(&plug);
3147 btrfs_set_log_full_commit(trans);
3149 if (ret != -ENOSPC) {
3150 btrfs_abort_transaction(trans, ret);
3151 mutex_unlock(&log_root_tree->log_mutex);
3154 btrfs_wait_tree_log_extents(log, mark);
3155 mutex_unlock(&log_root_tree->log_mutex);
3160 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3161 blk_finish_plug(&plug);
3162 list_del_init(&root_log_ctx.list);
3163 mutex_unlock(&log_root_tree->log_mutex);
3164 ret = root_log_ctx.log_ret;
3168 index2 = root_log_ctx.log_transid % 2;
3169 if (atomic_read(&log_root_tree->log_commit[index2])) {
3170 blk_finish_plug(&plug);
3171 ret = btrfs_wait_tree_log_extents(log, mark);
3172 wait_log_commit(log_root_tree,
3173 root_log_ctx.log_transid);
3174 mutex_unlock(&log_root_tree->log_mutex);
3176 ret = root_log_ctx.log_ret;
3179 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3180 atomic_set(&log_root_tree->log_commit[index2], 1);
3182 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3183 wait_log_commit(log_root_tree,
3184 root_log_ctx.log_transid - 1);
3187 wait_for_writer(log_root_tree);
3190 * now that we've moved on to the tree of log tree roots,
3191 * check the full commit flag again
3193 if (btrfs_need_log_full_commit(trans)) {
3194 blk_finish_plug(&plug);
3195 btrfs_wait_tree_log_extents(log, mark);
3196 mutex_unlock(&log_root_tree->log_mutex);
3198 goto out_wake_log_root;
3201 ret = btrfs_write_marked_extents(fs_info,
3202 &log_root_tree->dirty_log_pages,
3203 EXTENT_DIRTY | EXTENT_NEW);
3204 blk_finish_plug(&plug);
3206 btrfs_set_log_full_commit(trans);
3207 btrfs_abort_transaction(trans, ret);
3208 mutex_unlock(&log_root_tree->log_mutex);
3209 goto out_wake_log_root;
3211 ret = btrfs_wait_tree_log_extents(log, mark);
3213 ret = btrfs_wait_tree_log_extents(log_root_tree,
3214 EXTENT_NEW | EXTENT_DIRTY);
3216 btrfs_set_log_full_commit(trans);
3217 mutex_unlock(&log_root_tree->log_mutex);
3218 goto out_wake_log_root;
3221 btrfs_set_super_log_root(fs_info->super_for_commit,
3222 log_root_tree->node->start);
3223 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3224 btrfs_header_level(log_root_tree->node));
3226 log_root_tree->log_transid++;
3227 mutex_unlock(&log_root_tree->log_mutex);
3230 * Nobody else is going to jump in and write the ctree
3231 * super here because the log_commit atomic below is protecting
3232 * us. We must be called with a transaction handle pinning
3233 * the running transaction open, so a full commit can't hop
3234 * in and cause problems either.
3236 ret = write_all_supers(fs_info, 1);
3238 btrfs_set_log_full_commit(trans);
3239 btrfs_abort_transaction(trans, ret);
3240 goto out_wake_log_root;
3243 mutex_lock(&root->log_mutex);
3244 if (root->last_log_commit < log_transid)
3245 root->last_log_commit = log_transid;
3246 mutex_unlock(&root->log_mutex);
3249 mutex_lock(&log_root_tree->log_mutex);
3250 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3252 log_root_tree->log_transid_committed++;
3253 atomic_set(&log_root_tree->log_commit[index2], 0);
3254 mutex_unlock(&log_root_tree->log_mutex);
3257 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3258 * all the updates above are seen by the woken threads. It might not be
3259 * necessary, but proving that seems to be hard.
3261 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3263 mutex_lock(&root->log_mutex);
3264 btrfs_remove_all_log_ctxs(root, index1, ret);
3265 root->log_transid_committed++;
3266 atomic_set(&root->log_commit[index1], 0);
3267 mutex_unlock(&root->log_mutex);
3270 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3271 * all the updates above are seen by the woken threads. It might not be
3272 * necessary, but proving that seems to be hard.
3274 cond_wake_up(&root->log_commit_wait[index1]);
3278 static void free_log_tree(struct btrfs_trans_handle *trans,
3279 struct btrfs_root *log)
3282 struct walk_control wc = {
3284 .process_func = process_one_buffer
3287 ret = walk_log_tree(trans, log, &wc);
3290 btrfs_abort_transaction(trans, ret);
3292 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3295 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3296 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3297 extent_io_tree_release(&log->log_csum_range);
3298 btrfs_put_root(log);
3302 * free all the extents used by the tree log. This should be called
3303 * at commit time of the full transaction
3305 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3307 if (root->log_root) {
3308 free_log_tree(trans, root->log_root);
3309 root->log_root = NULL;
3310 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3315 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3316 struct btrfs_fs_info *fs_info)
3318 if (fs_info->log_root_tree) {
3319 free_log_tree(trans, fs_info->log_root_tree);
3320 fs_info->log_root_tree = NULL;
3326 * Check if an inode was logged in the current transaction. We can't always rely
3327 * on an inode's logged_trans value, because it's an in-memory only field and
3328 * therefore not persisted. This means that its value is lost if the inode gets
3329 * evicted and loaded again from disk (in which case it has a value of 0, and
3330 * certainly it is smaller then any possible transaction ID), when that happens
3331 * the full_sync flag is set in the inode's runtime flags, so on that case we
3332 * assume eviction happened and ignore the logged_trans value, assuming the
3333 * worst case, that the inode was logged before in the current transaction.
3335 static bool inode_logged(struct btrfs_trans_handle *trans,
3336 struct btrfs_inode *inode)
3338 if (inode->logged_trans == trans->transid)
3341 if (inode->last_trans == trans->transid &&
3342 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3343 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3350 * If both a file and directory are logged, and unlinks or renames are
3351 * mixed in, we have a few interesting corners:
3353 * create file X in dir Y
3354 * link file X to X.link in dir Y
3356 * unlink file X but leave X.link
3359 * After a crash we would expect only X.link to exist. But file X
3360 * didn't get fsync'd again so the log has back refs for X and X.link.
3362 * We solve this by removing directory entries and inode backrefs from the
3363 * log when a file that was logged in the current transaction is
3364 * unlinked. Any later fsync will include the updated log entries, and
3365 * we'll be able to reconstruct the proper directory items from backrefs.
3367 * This optimizations allows us to avoid relogging the entire inode
3368 * or the entire directory.
3370 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root,
3372 const char *name, int name_len,
3373 struct btrfs_inode *dir, u64 index)
3375 struct btrfs_root *log;
3376 struct btrfs_dir_item *di;
3377 struct btrfs_path *path;
3381 u64 dir_ino = btrfs_ino(dir);
3383 if (!inode_logged(trans, dir))
3386 ret = join_running_log_trans(root);
3390 mutex_lock(&dir->log_mutex);
3392 log = root->log_root;
3393 path = btrfs_alloc_path();
3399 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3400 name, name_len, -1);
3406 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3407 bytes_del += name_len;
3413 btrfs_release_path(path);
3414 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3415 index, name, name_len, -1);
3421 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3422 bytes_del += name_len;
3429 /* update the directory size in the log to reflect the names
3433 struct btrfs_key key;
3435 key.objectid = dir_ino;
3437 key.type = BTRFS_INODE_ITEM_KEY;
3438 btrfs_release_path(path);
3440 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3446 struct btrfs_inode_item *item;
3449 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3450 struct btrfs_inode_item);
3451 i_size = btrfs_inode_size(path->nodes[0], item);
3452 if (i_size > bytes_del)
3453 i_size -= bytes_del;
3456 btrfs_set_inode_size(path->nodes[0], item, i_size);
3457 btrfs_mark_buffer_dirty(path->nodes[0]);
3460 btrfs_release_path(path);
3463 btrfs_free_path(path);
3465 mutex_unlock(&dir->log_mutex);
3466 if (ret == -ENOSPC) {
3467 btrfs_set_log_full_commit(trans);
3470 btrfs_abort_transaction(trans, ret);
3472 btrfs_end_log_trans(root);
3477 /* see comments for btrfs_del_dir_entries_in_log */
3478 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3479 struct btrfs_root *root,
3480 const char *name, int name_len,
3481 struct btrfs_inode *inode, u64 dirid)
3483 struct btrfs_root *log;
3487 if (!inode_logged(trans, inode))
3490 ret = join_running_log_trans(root);
3493 log = root->log_root;
3494 mutex_lock(&inode->log_mutex);
3496 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3498 mutex_unlock(&inode->log_mutex);
3499 if (ret == -ENOSPC) {
3500 btrfs_set_log_full_commit(trans);
3502 } else if (ret < 0 && ret != -ENOENT)
3503 btrfs_abort_transaction(trans, ret);
3504 btrfs_end_log_trans(root);
3510 * creates a range item in the log for 'dirid'. first_offset and
3511 * last_offset tell us which parts of the key space the log should
3512 * be considered authoritative for.
3514 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3515 struct btrfs_root *log,
3516 struct btrfs_path *path,
3517 int key_type, u64 dirid,
3518 u64 first_offset, u64 last_offset)
3521 struct btrfs_key key;
3522 struct btrfs_dir_log_item *item;
3524 key.objectid = dirid;
3525 key.offset = first_offset;
3526 if (key_type == BTRFS_DIR_ITEM_KEY)
3527 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3529 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3530 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3534 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3535 struct btrfs_dir_log_item);
3536 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3537 btrfs_mark_buffer_dirty(path->nodes[0]);
3538 btrfs_release_path(path);
3543 * log all the items included in the current transaction for a given
3544 * directory. This also creates the range items in the log tree required
3545 * to replay anything deleted before the fsync
3547 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3548 struct btrfs_root *root, struct btrfs_inode *inode,
3549 struct btrfs_path *path,
3550 struct btrfs_path *dst_path, int key_type,
3551 struct btrfs_log_ctx *ctx,
3552 u64 min_offset, u64 *last_offset_ret)
3554 struct btrfs_key min_key;
3555 struct btrfs_root *log = root->log_root;
3556 struct extent_buffer *src;
3561 u64 first_offset = min_offset;
3562 u64 last_offset = (u64)-1;
3563 u64 ino = btrfs_ino(inode);
3565 log = root->log_root;
3567 min_key.objectid = ino;
3568 min_key.type = key_type;
3569 min_key.offset = min_offset;
3571 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3574 * we didn't find anything from this transaction, see if there
3575 * is anything at all
3577 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3578 min_key.objectid = ino;
3579 min_key.type = key_type;
3580 min_key.offset = (u64)-1;
3581 btrfs_release_path(path);
3582 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3584 btrfs_release_path(path);
3587 ret = btrfs_previous_item(root, path, ino, key_type);
3589 /* if ret == 0 there are items for this type,
3590 * create a range to tell us the last key of this type.
3591 * otherwise, there are no items in this directory after
3592 * *min_offset, and we create a range to indicate that.
3595 struct btrfs_key tmp;
3596 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3598 if (key_type == tmp.type)
3599 first_offset = max(min_offset, tmp.offset) + 1;
3604 /* go backward to find any previous key */
3605 ret = btrfs_previous_item(root, path, ino, key_type);
3607 struct btrfs_key tmp;
3608 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3609 if (key_type == tmp.type) {
3610 first_offset = tmp.offset;
3611 ret = overwrite_item(trans, log, dst_path,
3612 path->nodes[0], path->slots[0],
3620 btrfs_release_path(path);
3623 * Find the first key from this transaction again. See the note for
3624 * log_new_dir_dentries, if we're logging a directory recursively we
3625 * won't be holding its i_mutex, which means we can modify the directory
3626 * while we're logging it. If we remove an entry between our first
3627 * search and this search we'll not find the key again and can just
3630 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3635 * we have a block from this transaction, log every item in it
3636 * from our directory
3639 struct btrfs_key tmp;
3640 src = path->nodes[0];
3641 nritems = btrfs_header_nritems(src);
3642 for (i = path->slots[0]; i < nritems; i++) {
3643 struct btrfs_dir_item *di;
3645 btrfs_item_key_to_cpu(src, &min_key, i);
3647 if (min_key.objectid != ino || min_key.type != key_type)
3649 ret = overwrite_item(trans, log, dst_path, src, i,
3657 * We must make sure that when we log a directory entry,
3658 * the corresponding inode, after log replay, has a
3659 * matching link count. For example:
3665 * xfs_io -c "fsync" mydir
3667 * <mount fs and log replay>
3669 * Would result in a fsync log that when replayed, our
3670 * file inode would have a link count of 1, but we get
3671 * two directory entries pointing to the same inode.
3672 * After removing one of the names, it would not be
3673 * possible to remove the other name, which resulted
3674 * always in stale file handle errors, and would not
3675 * be possible to rmdir the parent directory, since
3676 * its i_size could never decrement to the value
3677 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3679 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3680 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3682 (btrfs_dir_transid(src, di) == trans->transid ||
3683 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3684 tmp.type != BTRFS_ROOT_ITEM_KEY)
3685 ctx->log_new_dentries = true;
3687 path->slots[0] = nritems;
3690 * look ahead to the next item and see if it is also
3691 * from this directory and from this transaction
3693 ret = btrfs_next_leaf(root, path);
3696 last_offset = (u64)-1;
3701 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3702 if (tmp.objectid != ino || tmp.type != key_type) {
3703 last_offset = (u64)-1;
3706 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3707 ret = overwrite_item(trans, log, dst_path,
3708 path->nodes[0], path->slots[0],
3713 last_offset = tmp.offset;
3718 btrfs_release_path(path);
3719 btrfs_release_path(dst_path);
3722 *last_offset_ret = last_offset;
3724 * insert the log range keys to indicate where the log
3727 ret = insert_dir_log_key(trans, log, path, key_type,
3728 ino, first_offset, last_offset);
3736 * logging directories is very similar to logging inodes, We find all the items
3737 * from the current transaction and write them to the log.
3739 * The recovery code scans the directory in the subvolume, and if it finds a
3740 * key in the range logged that is not present in the log tree, then it means
3741 * that dir entry was unlinked during the transaction.
3743 * In order for that scan to work, we must include one key smaller than
3744 * the smallest logged by this transaction and one key larger than the largest
3745 * key logged by this transaction.
3747 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3748 struct btrfs_root *root, struct btrfs_inode *inode,
3749 struct btrfs_path *path,
3750 struct btrfs_path *dst_path,
3751 struct btrfs_log_ctx *ctx)
3756 int key_type = BTRFS_DIR_ITEM_KEY;
3762 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3763 ctx, min_key, &max_key);
3766 if (max_key == (u64)-1)
3768 min_key = max_key + 1;
3771 if (key_type == BTRFS_DIR_ITEM_KEY) {
3772 key_type = BTRFS_DIR_INDEX_KEY;
3779 * a helper function to drop items from the log before we relog an
3780 * inode. max_key_type indicates the highest item type to remove.
3781 * This cannot be run for file data extents because it does not
3782 * free the extents they point to.
3784 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3785 struct btrfs_root *log,
3786 struct btrfs_path *path,
3787 u64 objectid, int max_key_type)
3790 struct btrfs_key key;
3791 struct btrfs_key found_key;
3794 key.objectid = objectid;
3795 key.type = max_key_type;
3796 key.offset = (u64)-1;
3799 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3800 BUG_ON(ret == 0); /* Logic error */
3804 if (path->slots[0] == 0)
3808 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3811 if (found_key.objectid != objectid)
3814 found_key.offset = 0;
3816 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
3820 ret = btrfs_del_items(trans, log, path, start_slot,
3821 path->slots[0] - start_slot + 1);
3823 * If start slot isn't 0 then we don't need to re-search, we've
3824 * found the last guy with the objectid in this tree.
3826 if (ret || start_slot != 0)
3828 btrfs_release_path(path);
3830 btrfs_release_path(path);
3836 static void fill_inode_item(struct btrfs_trans_handle *trans,
3837 struct extent_buffer *leaf,
3838 struct btrfs_inode_item *item,
3839 struct inode *inode, int log_inode_only,
3842 struct btrfs_map_token token;
3844 btrfs_init_map_token(&token, leaf);
3846 if (log_inode_only) {
3847 /* set the generation to zero so the recover code
3848 * can tell the difference between an logging
3849 * just to say 'this inode exists' and a logging
3850 * to say 'update this inode with these values'
3852 btrfs_set_token_inode_generation(&token, item, 0);
3853 btrfs_set_token_inode_size(&token, item, logged_isize);
3855 btrfs_set_token_inode_generation(&token, item,
3856 BTRFS_I(inode)->generation);
3857 btrfs_set_token_inode_size(&token, item, inode->i_size);
3860 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3861 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3862 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3863 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3865 btrfs_set_token_timespec_sec(&token, &item->atime,
3866 inode->i_atime.tv_sec);
3867 btrfs_set_token_timespec_nsec(&token, &item->atime,
3868 inode->i_atime.tv_nsec);
3870 btrfs_set_token_timespec_sec(&token, &item->mtime,
3871 inode->i_mtime.tv_sec);
3872 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3873 inode->i_mtime.tv_nsec);
3875 btrfs_set_token_timespec_sec(&token, &item->ctime,
3876 inode->i_ctime.tv_sec);
3877 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3878 inode->i_ctime.tv_nsec);
3880 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3882 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3883 btrfs_set_token_inode_transid(&token, item, trans->transid);
3884 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3885 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3886 btrfs_set_token_inode_block_group(&token, item, 0);
3889 static int log_inode_item(struct btrfs_trans_handle *trans,
3890 struct btrfs_root *log, struct btrfs_path *path,
3891 struct btrfs_inode *inode)
3893 struct btrfs_inode_item *inode_item;
3896 ret = btrfs_insert_empty_item(trans, log, path,
3897 &inode->location, sizeof(*inode_item));
3898 if (ret && ret != -EEXIST)
3900 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3901 struct btrfs_inode_item);
3902 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3904 btrfs_release_path(path);
3908 static int log_csums(struct btrfs_trans_handle *trans,
3909 struct btrfs_root *log_root,
3910 struct btrfs_ordered_sum *sums)
3912 const u64 lock_end = sums->bytenr + sums->len - 1;
3913 struct extent_state *cached_state = NULL;
3917 * Serialize logging for checksums. This is to avoid racing with the
3918 * same checksum being logged by another task that is logging another
3919 * file which happens to refer to the same extent as well. Such races
3920 * can leave checksum items in the log with overlapping ranges.
3922 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
3923 lock_end, &cached_state);
3927 * Due to extent cloning, we might have logged a csum item that covers a
3928 * subrange of a cloned extent, and later we can end up logging a csum
3929 * item for a larger subrange of the same extent or the entire range.
3930 * This would leave csum items in the log tree that cover the same range
3931 * and break the searches for checksums in the log tree, resulting in
3932 * some checksums missing in the fs/subvolume tree. So just delete (or
3933 * trim and adjust) any existing csum items in the log for this range.
3935 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
3937 ret = btrfs_csum_file_blocks(trans, log_root, sums);
3939 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
3945 static noinline int copy_items(struct btrfs_trans_handle *trans,
3946 struct btrfs_inode *inode,
3947 struct btrfs_path *dst_path,
3948 struct btrfs_path *src_path,
3949 int start_slot, int nr, int inode_only,
3952 struct btrfs_fs_info *fs_info = trans->fs_info;
3953 unsigned long src_offset;
3954 unsigned long dst_offset;
3955 struct btrfs_root *log = inode->root->log_root;
3956 struct btrfs_file_extent_item *extent;
3957 struct btrfs_inode_item *inode_item;
3958 struct extent_buffer *src = src_path->nodes[0];
3960 struct btrfs_key *ins_keys;
3964 struct list_head ordered_sums;
3965 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3967 INIT_LIST_HEAD(&ordered_sums);
3969 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3970 nr * sizeof(u32), GFP_NOFS);
3974 ins_sizes = (u32 *)ins_data;
3975 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3977 for (i = 0; i < nr; i++) {
3978 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3979 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3981 ret = btrfs_insert_empty_items(trans, log, dst_path,
3982 ins_keys, ins_sizes, nr);
3988 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3989 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3990 dst_path->slots[0]);
3992 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3994 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3995 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3997 struct btrfs_inode_item);
3998 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4000 inode_only == LOG_INODE_EXISTS,
4003 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4004 src_offset, ins_sizes[i]);
4007 /* take a reference on file data extents so that truncates
4008 * or deletes of this inode don't have to relog the inode
4011 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4014 extent = btrfs_item_ptr(src, start_slot + i,
4015 struct btrfs_file_extent_item);
4017 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4020 found_type = btrfs_file_extent_type(src, extent);
4021 if (found_type == BTRFS_FILE_EXTENT_REG) {
4023 ds = btrfs_file_extent_disk_bytenr(src,
4025 /* ds == 0 is a hole */
4029 dl = btrfs_file_extent_disk_num_bytes(src,
4031 cs = btrfs_file_extent_offset(src, extent);
4032 cl = btrfs_file_extent_num_bytes(src,
4034 if (btrfs_file_extent_compression(src,
4040 ret = btrfs_lookup_csums_range(
4042 ds + cs, ds + cs + cl - 1,
4045 btrfs_release_path(dst_path);
4053 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4054 btrfs_release_path(dst_path);
4058 * we have to do this after the loop above to avoid changing the
4059 * log tree while trying to change the log tree.
4062 while (!list_empty(&ordered_sums)) {
4063 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4064 struct btrfs_ordered_sum,
4067 ret = log_csums(trans, log, sums);
4068 list_del(&sums->list);
4075 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4077 struct extent_map *em1, *em2;
4079 em1 = list_entry(a, struct extent_map, list);
4080 em2 = list_entry(b, struct extent_map, list);
4082 if (em1->start < em2->start)
4084 else if (em1->start > em2->start)
4089 static int log_extent_csums(struct btrfs_trans_handle *trans,
4090 struct btrfs_inode *inode,
4091 struct btrfs_root *log_root,
4092 const struct extent_map *em)
4096 LIST_HEAD(ordered_sums);
4099 if (inode->flags & BTRFS_INODE_NODATASUM ||
4100 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4101 em->block_start == EXTENT_MAP_HOLE)
4104 /* If we're compressed we have to save the entire range of csums. */
4105 if (em->compress_type) {
4107 csum_len = max(em->block_len, em->orig_block_len);
4109 csum_offset = em->mod_start - em->start;
4110 csum_len = em->mod_len;
4113 /* block start is already adjusted for the file extent offset. */
4114 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4115 em->block_start + csum_offset,
4116 em->block_start + csum_offset +
4117 csum_len - 1, &ordered_sums, 0);
4121 while (!list_empty(&ordered_sums)) {
4122 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4123 struct btrfs_ordered_sum,
4126 ret = log_csums(trans, log_root, sums);
4127 list_del(&sums->list);
4134 static int log_one_extent(struct btrfs_trans_handle *trans,
4135 struct btrfs_inode *inode, struct btrfs_root *root,
4136 const struct extent_map *em,
4137 struct btrfs_path *path,
4138 struct btrfs_log_ctx *ctx)
4140 struct btrfs_root *log = root->log_root;
4141 struct btrfs_file_extent_item *fi;
4142 struct extent_buffer *leaf;
4143 struct btrfs_map_token token;
4144 struct btrfs_key key;
4145 u64 extent_offset = em->start - em->orig_start;
4148 int extent_inserted = 0;
4150 ret = log_extent_csums(trans, inode, log, em);
4154 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4155 em->start + em->len, NULL, 0, 1,
4156 sizeof(*fi), &extent_inserted);
4160 if (!extent_inserted) {
4161 key.objectid = btrfs_ino(inode);
4162 key.type = BTRFS_EXTENT_DATA_KEY;
4163 key.offset = em->start;
4165 ret = btrfs_insert_empty_item(trans, log, path, &key,
4170 leaf = path->nodes[0];
4171 btrfs_init_map_token(&token, leaf);
4172 fi = btrfs_item_ptr(leaf, path->slots[0],
4173 struct btrfs_file_extent_item);
4175 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4176 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4177 btrfs_set_token_file_extent_type(&token, fi,
4178 BTRFS_FILE_EXTENT_PREALLOC);
4180 btrfs_set_token_file_extent_type(&token, fi,
4181 BTRFS_FILE_EXTENT_REG);
4183 block_len = max(em->block_len, em->orig_block_len);
4184 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4185 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4187 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4188 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4189 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4192 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4194 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4195 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4198 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4199 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4200 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4201 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4202 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4203 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4204 btrfs_mark_buffer_dirty(leaf);
4206 btrfs_release_path(path);
4212 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4213 * lose them after doing a fast fsync and replaying the log. We scan the
4214 * subvolume's root instead of iterating the inode's extent map tree because
4215 * otherwise we can log incorrect extent items based on extent map conversion.
4216 * That can happen due to the fact that extent maps are merged when they
4217 * are not in the extent map tree's list of modified extents.
4219 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4220 struct btrfs_inode *inode,
4221 struct btrfs_path *path)
4223 struct btrfs_root *root = inode->root;
4224 struct btrfs_key key;
4225 const u64 i_size = i_size_read(&inode->vfs_inode);
4226 const u64 ino = btrfs_ino(inode);
4227 struct btrfs_path *dst_path = NULL;
4228 bool dropped_extents = false;
4229 u64 truncate_offset = i_size;
4230 struct extent_buffer *leaf;
4236 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4240 key.type = BTRFS_EXTENT_DATA_KEY;
4241 key.offset = i_size;
4242 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4247 * We must check if there is a prealloc extent that starts before the
4248 * i_size and crosses the i_size boundary. This is to ensure later we
4249 * truncate down to the end of that extent and not to the i_size, as
4250 * otherwise we end up losing part of the prealloc extent after a log
4251 * replay and with an implicit hole if there is another prealloc extent
4252 * that starts at an offset beyond i_size.
4254 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4259 struct btrfs_file_extent_item *ei;
4261 leaf = path->nodes[0];
4262 slot = path->slots[0];
4263 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4265 if (btrfs_file_extent_type(leaf, ei) ==
4266 BTRFS_FILE_EXTENT_PREALLOC) {
4269 btrfs_item_key_to_cpu(leaf, &key, slot);
4270 extent_end = key.offset +
4271 btrfs_file_extent_num_bytes(leaf, ei);
4273 if (extent_end > i_size)
4274 truncate_offset = extent_end;
4281 leaf = path->nodes[0];
4282 slot = path->slots[0];
4284 if (slot >= btrfs_header_nritems(leaf)) {
4286 ret = copy_items(trans, inode, dst_path, path,
4287 start_slot, ins_nr, 1, 0);
4292 ret = btrfs_next_leaf(root, path);
4302 btrfs_item_key_to_cpu(leaf, &key, slot);
4303 if (key.objectid > ino)
4305 if (WARN_ON_ONCE(key.objectid < ino) ||
4306 key.type < BTRFS_EXTENT_DATA_KEY ||
4307 key.offset < i_size) {
4311 if (!dropped_extents) {
4313 * Avoid logging extent items logged in past fsync calls
4314 * and leading to duplicate keys in the log tree.
4317 ret = btrfs_truncate_inode_items(trans,
4321 BTRFS_EXTENT_DATA_KEY);
4322 } while (ret == -EAGAIN);
4325 dropped_extents = true;
4332 dst_path = btrfs_alloc_path();
4340 ret = copy_items(trans, inode, dst_path, path,
4341 start_slot, ins_nr, 1, 0);
4343 btrfs_release_path(path);
4344 btrfs_free_path(dst_path);
4348 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4349 struct btrfs_root *root,
4350 struct btrfs_inode *inode,
4351 struct btrfs_path *path,
4352 struct btrfs_log_ctx *ctx,
4356 struct extent_map *em, *n;
4357 struct list_head extents;
4358 struct extent_map_tree *tree = &inode->extent_tree;
4363 INIT_LIST_HEAD(&extents);
4365 write_lock(&tree->lock);
4366 test_gen = root->fs_info->last_trans_committed;
4368 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4370 * Skip extents outside our logging range. It's important to do
4371 * it for correctness because if we don't ignore them, we may
4372 * log them before their ordered extent completes, and therefore
4373 * we could log them without logging their respective checksums
4374 * (the checksum items are added to the csum tree at the very
4375 * end of btrfs_finish_ordered_io()). Also leave such extents
4376 * outside of our range in the list, since we may have another
4377 * ranged fsync in the near future that needs them. If an extent
4378 * outside our range corresponds to a hole, log it to avoid
4379 * leaving gaps between extents (fsck will complain when we are
4380 * not using the NO_HOLES feature).
4382 if ((em->start > end || em->start + em->len <= start) &&
4383 em->block_start != EXTENT_MAP_HOLE)
4386 list_del_init(&em->list);
4388 * Just an arbitrary number, this can be really CPU intensive
4389 * once we start getting a lot of extents, and really once we
4390 * have a bunch of extents we just want to commit since it will
4393 if (++num > 32768) {
4394 list_del_init(&tree->modified_extents);
4399 if (em->generation <= test_gen)
4402 /* We log prealloc extents beyond eof later. */
4403 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4404 em->start >= i_size_read(&inode->vfs_inode))
4407 /* Need a ref to keep it from getting evicted from cache */
4408 refcount_inc(&em->refs);
4409 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4410 list_add_tail(&em->list, &extents);
4414 list_sort(NULL, &extents, extent_cmp);
4416 while (!list_empty(&extents)) {
4417 em = list_entry(extents.next, struct extent_map, list);
4419 list_del_init(&em->list);
4422 * If we had an error we just need to delete everybody from our
4426 clear_em_logging(tree, em);
4427 free_extent_map(em);
4431 write_unlock(&tree->lock);
4433 ret = log_one_extent(trans, inode, root, em, path, ctx);
4434 write_lock(&tree->lock);
4435 clear_em_logging(tree, em);
4436 free_extent_map(em);
4438 WARN_ON(!list_empty(&extents));
4439 write_unlock(&tree->lock);
4441 btrfs_release_path(path);
4443 ret = btrfs_log_prealloc_extents(trans, inode, path);
4448 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4449 struct btrfs_path *path, u64 *size_ret)
4451 struct btrfs_key key;
4454 key.objectid = btrfs_ino(inode);
4455 key.type = BTRFS_INODE_ITEM_KEY;
4458 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4461 } else if (ret > 0) {
4464 struct btrfs_inode_item *item;
4466 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4467 struct btrfs_inode_item);
4468 *size_ret = btrfs_inode_size(path->nodes[0], item);
4470 * If the in-memory inode's i_size is smaller then the inode
4471 * size stored in the btree, return the inode's i_size, so
4472 * that we get a correct inode size after replaying the log
4473 * when before a power failure we had a shrinking truncate
4474 * followed by addition of a new name (rename / new hard link).
4475 * Otherwise return the inode size from the btree, to avoid
4476 * data loss when replaying a log due to previously doing a
4477 * write that expands the inode's size and logging a new name
4478 * immediately after.
4480 if (*size_ret > inode->vfs_inode.i_size)
4481 *size_ret = inode->vfs_inode.i_size;
4484 btrfs_release_path(path);
4489 * At the moment we always log all xattrs. This is to figure out at log replay
4490 * time which xattrs must have their deletion replayed. If a xattr is missing
4491 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4492 * because if a xattr is deleted, the inode is fsynced and a power failure
4493 * happens, causing the log to be replayed the next time the fs is mounted,
4494 * we want the xattr to not exist anymore (same behaviour as other filesystems
4495 * with a journal, ext3/4, xfs, f2fs, etc).
4497 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4498 struct btrfs_root *root,
4499 struct btrfs_inode *inode,
4500 struct btrfs_path *path,
4501 struct btrfs_path *dst_path)
4504 struct btrfs_key key;
4505 const u64 ino = btrfs_ino(inode);
4510 key.type = BTRFS_XATTR_ITEM_KEY;
4513 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4518 int slot = path->slots[0];
4519 struct extent_buffer *leaf = path->nodes[0];
4520 int nritems = btrfs_header_nritems(leaf);
4522 if (slot >= nritems) {
4524 ret = copy_items(trans, inode, dst_path, path,
4525 start_slot, ins_nr, 1, 0);
4530 ret = btrfs_next_leaf(root, path);
4538 btrfs_item_key_to_cpu(leaf, &key, slot);
4539 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4549 ret = copy_items(trans, inode, dst_path, path,
4550 start_slot, ins_nr, 1, 0);
4559 * When using the NO_HOLES feature if we punched a hole that causes the
4560 * deletion of entire leafs or all the extent items of the first leaf (the one
4561 * that contains the inode item and references) we may end up not processing
4562 * any extents, because there are no leafs with a generation matching the
4563 * current transaction that have extent items for our inode. So we need to find
4564 * if any holes exist and then log them. We also need to log holes after any
4565 * truncate operation that changes the inode's size.
4567 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
4568 struct btrfs_root *root,
4569 struct btrfs_inode *inode,
4570 struct btrfs_path *path)
4572 struct btrfs_fs_info *fs_info = root->fs_info;
4573 struct btrfs_key key;
4574 const u64 ino = btrfs_ino(inode);
4575 const u64 i_size = i_size_read(&inode->vfs_inode);
4576 u64 prev_extent_end = 0;
4579 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
4583 key.type = BTRFS_EXTENT_DATA_KEY;
4586 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4591 struct extent_buffer *leaf = path->nodes[0];
4593 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
4594 ret = btrfs_next_leaf(root, path);
4601 leaf = path->nodes[0];
4604 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4605 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
4608 /* We have a hole, log it. */
4609 if (prev_extent_end < key.offset) {
4610 const u64 hole_len = key.offset - prev_extent_end;
4613 * Release the path to avoid deadlocks with other code
4614 * paths that search the root while holding locks on
4615 * leafs from the log root.
4617 btrfs_release_path(path);
4618 ret = btrfs_insert_file_extent(trans, root->log_root,
4619 ino, prev_extent_end, 0,
4620 0, hole_len, 0, hole_len,
4626 * Search for the same key again in the root. Since it's
4627 * an extent item and we are holding the inode lock, the
4628 * key must still exist. If it doesn't just emit warning
4629 * and return an error to fall back to a transaction
4632 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4635 if (WARN_ON(ret > 0))
4637 leaf = path->nodes[0];
4640 prev_extent_end = btrfs_file_extent_end(path);
4645 if (prev_extent_end < i_size) {
4648 btrfs_release_path(path);
4649 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
4650 ret = btrfs_insert_file_extent(trans, root->log_root,
4651 ino, prev_extent_end, 0, 0,
4652 hole_len, 0, hole_len,
4662 * When we are logging a new inode X, check if it doesn't have a reference that
4663 * matches the reference from some other inode Y created in a past transaction
4664 * and that was renamed in the current transaction. If we don't do this, then at
4665 * log replay time we can lose inode Y (and all its files if it's a directory):
4668 * echo "hello world" > /mnt/x/foobar
4671 * mkdir /mnt/x # or touch /mnt/x
4672 * xfs_io -c fsync /mnt/x
4674 * mount fs, trigger log replay
4676 * After the log replay procedure, we would lose the first directory and all its
4677 * files (file foobar).
4678 * For the case where inode Y is not a directory we simply end up losing it:
4680 * echo "123" > /mnt/foo
4682 * mv /mnt/foo /mnt/bar
4683 * echo "abc" > /mnt/foo
4684 * xfs_io -c fsync /mnt/foo
4687 * We also need this for cases where a snapshot entry is replaced by some other
4688 * entry (file or directory) otherwise we end up with an unreplayable log due to
4689 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4690 * if it were a regular entry:
4693 * btrfs subvolume snapshot /mnt /mnt/x/snap
4694 * btrfs subvolume delete /mnt/x/snap
4697 * fsync /mnt/x or fsync some new file inside it
4700 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4701 * the same transaction.
4703 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4705 const struct btrfs_key *key,
4706 struct btrfs_inode *inode,
4707 u64 *other_ino, u64 *other_parent)
4710 struct btrfs_path *search_path;
4713 u32 item_size = btrfs_item_size_nr(eb, slot);
4715 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4717 search_path = btrfs_alloc_path();
4720 search_path->search_commit_root = 1;
4721 search_path->skip_locking = 1;
4723 while (cur_offset < item_size) {
4727 unsigned long name_ptr;
4728 struct btrfs_dir_item *di;
4730 if (key->type == BTRFS_INODE_REF_KEY) {
4731 struct btrfs_inode_ref *iref;
4733 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4734 parent = key->offset;
4735 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4736 name_ptr = (unsigned long)(iref + 1);
4737 this_len = sizeof(*iref) + this_name_len;
4739 struct btrfs_inode_extref *extref;
4741 extref = (struct btrfs_inode_extref *)(ptr +
4743 parent = btrfs_inode_extref_parent(eb, extref);
4744 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4745 name_ptr = (unsigned long)&extref->name;
4746 this_len = sizeof(*extref) + this_name_len;
4749 if (this_name_len > name_len) {
4752 new_name = krealloc(name, this_name_len, GFP_NOFS);
4757 name_len = this_name_len;
4761 read_extent_buffer(eb, name, name_ptr, this_name_len);
4762 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4763 parent, name, this_name_len, 0);
4764 if (di && !IS_ERR(di)) {
4765 struct btrfs_key di_key;
4767 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4769 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4770 if (di_key.objectid != key->objectid) {
4772 *other_ino = di_key.objectid;
4773 *other_parent = parent;
4781 } else if (IS_ERR(di)) {
4785 btrfs_release_path(search_path);
4787 cur_offset += this_len;
4791 btrfs_free_path(search_path);
4796 struct btrfs_ino_list {
4799 struct list_head list;
4802 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4803 struct btrfs_root *root,
4804 struct btrfs_path *path,
4805 struct btrfs_log_ctx *ctx,
4806 u64 ino, u64 parent)
4808 struct btrfs_ino_list *ino_elem;
4809 LIST_HEAD(inode_list);
4812 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4815 ino_elem->ino = ino;
4816 ino_elem->parent = parent;
4817 list_add_tail(&ino_elem->list, &inode_list);
4819 while (!list_empty(&inode_list)) {
4820 struct btrfs_fs_info *fs_info = root->fs_info;
4821 struct btrfs_key key;
4822 struct inode *inode;
4824 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4826 ino = ino_elem->ino;
4827 parent = ino_elem->parent;
4828 list_del(&ino_elem->list);
4833 btrfs_release_path(path);
4835 inode = btrfs_iget(fs_info->sb, ino, root);
4837 * If the other inode that had a conflicting dir entry was
4838 * deleted in the current transaction, we need to log its parent
4841 if (IS_ERR(inode)) {
4842 ret = PTR_ERR(inode);
4843 if (ret == -ENOENT) {
4844 inode = btrfs_iget(fs_info->sb, parent, root);
4845 if (IS_ERR(inode)) {
4846 ret = PTR_ERR(inode);
4848 ret = btrfs_log_inode(trans, root,
4850 LOG_OTHER_INODE_ALL,
4852 btrfs_add_delayed_iput(inode);
4858 * If the inode was already logged skip it - otherwise we can
4859 * hit an infinite loop. Example:
4861 * From the commit root (previous transaction) we have the
4864 * inode 257 a directory
4865 * inode 258 with references "zz" and "zz_link" on inode 257
4866 * inode 259 with reference "a" on inode 257
4868 * And in the current (uncommitted) transaction we have:
4870 * inode 257 a directory, unchanged
4871 * inode 258 with references "a" and "a2" on inode 257
4872 * inode 259 with reference "zz_link" on inode 257
4873 * inode 261 with reference "zz" on inode 257
4875 * When logging inode 261 the following infinite loop could
4876 * happen if we don't skip already logged inodes:
4878 * - we detect inode 258 as a conflicting inode, with inode 261
4879 * on reference "zz", and log it;
4881 * - we detect inode 259 as a conflicting inode, with inode 258
4882 * on reference "a", and log it;
4884 * - we detect inode 258 as a conflicting inode, with inode 259
4885 * on reference "zz_link", and log it - again! After this we
4886 * repeat the above steps forever.
4888 spin_lock(&BTRFS_I(inode)->lock);
4890 * Check the inode's logged_trans only instead of
4891 * btrfs_inode_in_log(). This is because the last_log_commit of
4892 * the inode is not updated when we only log that it exists and
4893 * and it has the full sync bit set (see btrfs_log_inode()).
4895 if (BTRFS_I(inode)->logged_trans == trans->transid) {
4896 spin_unlock(&BTRFS_I(inode)->lock);
4897 btrfs_add_delayed_iput(inode);
4900 spin_unlock(&BTRFS_I(inode)->lock);
4902 * We are safe logging the other inode without acquiring its
4903 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4904 * are safe against concurrent renames of the other inode as
4905 * well because during a rename we pin the log and update the
4906 * log with the new name before we unpin it.
4908 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4909 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
4911 btrfs_add_delayed_iput(inode);
4916 key.type = BTRFS_INODE_REF_KEY;
4918 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4920 btrfs_add_delayed_iput(inode);
4925 struct extent_buffer *leaf = path->nodes[0];
4926 int slot = path->slots[0];
4928 u64 other_parent = 0;
4930 if (slot >= btrfs_header_nritems(leaf)) {
4931 ret = btrfs_next_leaf(root, path);
4934 } else if (ret > 0) {
4941 btrfs_item_key_to_cpu(leaf, &key, slot);
4942 if (key.objectid != ino ||
4943 (key.type != BTRFS_INODE_REF_KEY &&
4944 key.type != BTRFS_INODE_EXTREF_KEY)) {
4949 ret = btrfs_check_ref_name_override(leaf, slot, &key,
4950 BTRFS_I(inode), &other_ino,
4955 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4960 ino_elem->ino = other_ino;
4961 ino_elem->parent = other_parent;
4962 list_add_tail(&ino_elem->list, &inode_list);
4967 btrfs_add_delayed_iput(inode);
4973 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
4974 struct btrfs_inode *inode,
4975 struct btrfs_key *min_key,
4976 const struct btrfs_key *max_key,
4977 struct btrfs_path *path,
4978 struct btrfs_path *dst_path,
4979 const u64 logged_isize,
4980 const bool recursive_logging,
4981 const int inode_only,
4982 struct btrfs_log_ctx *ctx,
4983 bool *need_log_inode_item)
4985 struct btrfs_root *root = inode->root;
4986 int ins_start_slot = 0;
4991 ret = btrfs_search_forward(root, min_key, path, trans->transid);
4999 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5000 if (min_key->objectid != max_key->objectid)
5002 if (min_key->type > max_key->type)
5005 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5006 *need_log_inode_item = false;
5008 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5009 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5010 inode->generation == trans->transid &&
5011 !recursive_logging) {
5013 u64 other_parent = 0;
5015 ret = btrfs_check_ref_name_override(path->nodes[0],
5016 path->slots[0], min_key, inode,
5017 &other_ino, &other_parent);
5020 } else if (ret > 0 && ctx &&
5021 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5026 ins_start_slot = path->slots[0];
5028 ret = copy_items(trans, inode, dst_path, path,
5029 ins_start_slot, ins_nr,
5030 inode_only, logged_isize);
5035 ret = log_conflicting_inodes(trans, root, path,
5036 ctx, other_ino, other_parent);
5039 btrfs_release_path(path);
5044 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5045 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5048 ret = copy_items(trans, inode, dst_path, path,
5050 ins_nr, inode_only, logged_isize);
5057 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5060 } else if (!ins_nr) {
5061 ins_start_slot = path->slots[0];
5066 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5067 ins_nr, inode_only, logged_isize);
5071 ins_start_slot = path->slots[0];
5074 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5075 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5080 ret = copy_items(trans, inode, dst_path, path,
5081 ins_start_slot, ins_nr, inode_only,
5087 btrfs_release_path(path);
5089 if (min_key->offset < (u64)-1) {
5091 } else if (min_key->type < max_key->type) {
5093 min_key->offset = 0;
5099 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5100 ins_nr, inode_only, logged_isize);
5105 /* log a single inode in the tree log.
5106 * At least one parent directory for this inode must exist in the tree
5107 * or be logged already.
5109 * Any items from this inode changed by the current transaction are copied
5110 * to the log tree. An extra reference is taken on any extents in this
5111 * file, allowing us to avoid a whole pile of corner cases around logging
5112 * blocks that have been removed from the tree.
5114 * See LOG_INODE_ALL and related defines for a description of what inode_only
5117 * This handles both files and directories.
5119 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5120 struct btrfs_root *root, struct btrfs_inode *inode,
5124 struct btrfs_log_ctx *ctx)
5126 struct btrfs_fs_info *fs_info = root->fs_info;
5127 struct btrfs_path *path;
5128 struct btrfs_path *dst_path;
5129 struct btrfs_key min_key;
5130 struct btrfs_key max_key;
5131 struct btrfs_root *log = root->log_root;
5134 bool fast_search = false;
5135 u64 ino = btrfs_ino(inode);
5136 struct extent_map_tree *em_tree = &inode->extent_tree;
5137 u64 logged_isize = 0;
5138 bool need_log_inode_item = true;
5139 bool xattrs_logged = false;
5140 bool recursive_logging = false;
5142 path = btrfs_alloc_path();
5145 dst_path = btrfs_alloc_path();
5147 btrfs_free_path(path);
5151 min_key.objectid = ino;
5152 min_key.type = BTRFS_INODE_ITEM_KEY;
5155 max_key.objectid = ino;
5158 /* today the code can only do partial logging of directories */
5159 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5160 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5161 &inode->runtime_flags) &&
5162 inode_only >= LOG_INODE_EXISTS))
5163 max_key.type = BTRFS_XATTR_ITEM_KEY;
5165 max_key.type = (u8)-1;
5166 max_key.offset = (u64)-1;
5169 * Only run delayed items if we are a dir or a new file.
5170 * Otherwise commit the delayed inode only, which is needed in
5171 * order for the log replay code to mark inodes for link count
5172 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5174 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5175 inode->generation > fs_info->last_trans_committed)
5176 ret = btrfs_commit_inode_delayed_items(trans, inode);
5178 ret = btrfs_commit_inode_delayed_inode(inode);
5181 btrfs_free_path(path);
5182 btrfs_free_path(dst_path);
5186 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5187 recursive_logging = true;
5188 if (inode_only == LOG_OTHER_INODE)
5189 inode_only = LOG_INODE_EXISTS;
5191 inode_only = LOG_INODE_ALL;
5192 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5194 mutex_lock(&inode->log_mutex);
5198 * a brute force approach to making sure we get the most uptodate
5199 * copies of everything.
5201 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5202 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5204 if (inode_only == LOG_INODE_EXISTS)
5205 max_key_type = BTRFS_XATTR_ITEM_KEY;
5206 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5208 if (inode_only == LOG_INODE_EXISTS) {
5210 * Make sure the new inode item we write to the log has
5211 * the same isize as the current one (if it exists).
5212 * This is necessary to prevent data loss after log
5213 * replay, and also to prevent doing a wrong expanding
5214 * truncate - for e.g. create file, write 4K into offset
5215 * 0, fsync, write 4K into offset 4096, add hard link,
5216 * fsync some other file (to sync log), power fail - if
5217 * we use the inode's current i_size, after log replay
5218 * we get a 8Kb file, with the last 4Kb extent as a hole
5219 * (zeroes), as if an expanding truncate happened,
5220 * instead of getting a file of 4Kb only.
5222 err = logged_inode_size(log, inode, path, &logged_isize);
5226 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5227 &inode->runtime_flags)) {
5228 if (inode_only == LOG_INODE_EXISTS) {
5229 max_key.type = BTRFS_XATTR_ITEM_KEY;
5230 ret = drop_objectid_items(trans, log, path, ino,
5233 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5234 &inode->runtime_flags);
5235 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5236 &inode->runtime_flags);
5238 ret = btrfs_truncate_inode_items(trans,
5239 log, &inode->vfs_inode, 0, 0);
5244 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5245 &inode->runtime_flags) ||
5246 inode_only == LOG_INODE_EXISTS) {
5247 if (inode_only == LOG_INODE_ALL)
5249 max_key.type = BTRFS_XATTR_ITEM_KEY;
5250 ret = drop_objectid_items(trans, log, path, ino,
5253 if (inode_only == LOG_INODE_ALL)
5264 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5265 path, dst_path, logged_isize,
5266 recursive_logging, inode_only, ctx,
5267 &need_log_inode_item);
5271 btrfs_release_path(path);
5272 btrfs_release_path(dst_path);
5273 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5276 xattrs_logged = true;
5277 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5278 btrfs_release_path(path);
5279 btrfs_release_path(dst_path);
5280 err = btrfs_log_holes(trans, root, inode, path);
5285 btrfs_release_path(path);
5286 btrfs_release_path(dst_path);
5287 if (need_log_inode_item) {
5288 err = log_inode_item(trans, log, dst_path, inode);
5289 if (!err && !xattrs_logged) {
5290 err = btrfs_log_all_xattrs(trans, root, inode, path,
5292 btrfs_release_path(path);
5298 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5304 } else if (inode_only == LOG_INODE_ALL) {
5305 struct extent_map *em, *n;
5307 write_lock(&em_tree->lock);
5309 * We can't just remove every em if we're called for a ranged
5310 * fsync - that is, one that doesn't cover the whole possible
5311 * file range (0 to LLONG_MAX). This is because we can have
5312 * em's that fall outside the range we're logging and therefore
5313 * their ordered operations haven't completed yet
5314 * (btrfs_finish_ordered_io() not invoked yet). This means we
5315 * didn't get their respective file extent item in the fs/subvol
5316 * tree yet, and need to let the next fast fsync (one which
5317 * consults the list of modified extent maps) find the em so
5318 * that it logs a matching file extent item and waits for the
5319 * respective ordered operation to complete (if it's still
5322 * Removing every em outside the range we're logging would make
5323 * the next fast fsync not log their matching file extent items,
5324 * therefore making us lose data after a log replay.
5326 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5328 const u64 mod_end = em->mod_start + em->mod_len - 1;
5330 if (em->mod_start >= start && mod_end <= end)
5331 list_del_init(&em->list);
5333 write_unlock(&em_tree->lock);
5336 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5337 ret = log_directory_changes(trans, root, inode, path, dst_path,
5346 * Don't update last_log_commit if we logged that an inode exists after
5347 * it was loaded to memory (full_sync bit set).
5348 * This is to prevent data loss when we do a write to the inode, then
5349 * the inode gets evicted after all delalloc was flushed, then we log
5350 * it exists (due to a rename for example) and then fsync it. This last
5351 * fsync would do nothing (not logging the extents previously written).
5353 spin_lock(&inode->lock);
5354 inode->logged_trans = trans->transid;
5355 if (inode_only != LOG_INODE_EXISTS ||
5356 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5357 inode->last_log_commit = inode->last_sub_trans;
5358 spin_unlock(&inode->lock);
5360 mutex_unlock(&inode->log_mutex);
5362 btrfs_free_path(path);
5363 btrfs_free_path(dst_path);
5368 * Check if we must fallback to a transaction commit when logging an inode.
5369 * This must be called after logging the inode and is used only in the context
5370 * when fsyncing an inode requires the need to log some other inode - in which
5371 * case we can't lock the i_mutex of each other inode we need to log as that
5372 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5373 * log inodes up or down in the hierarchy) or rename operations for example. So
5374 * we take the log_mutex of the inode after we have logged it and then check for
5375 * its last_unlink_trans value - this is safe because any task setting
5376 * last_unlink_trans must take the log_mutex and it must do this before it does
5377 * the actual unlink operation, so if we do this check before a concurrent task
5378 * sets last_unlink_trans it means we've logged a consistent version/state of
5379 * all the inode items, otherwise we are not sure and must do a transaction
5380 * commit (the concurrent task might have only updated last_unlink_trans before
5381 * we logged the inode or it might have also done the unlink).
5383 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5384 struct btrfs_inode *inode)
5386 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5389 mutex_lock(&inode->log_mutex);
5390 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5392 * Make sure any commits to the log are forced to be full
5395 btrfs_set_log_full_commit(trans);
5398 mutex_unlock(&inode->log_mutex);
5404 * follow the dentry parent pointers up the chain and see if any
5405 * of the directories in it require a full commit before they can
5406 * be logged. Returns zero if nothing special needs to be done or 1 if
5407 * a full commit is required.
5409 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5410 struct btrfs_inode *inode,
5411 struct dentry *parent,
5412 struct super_block *sb,
5416 struct dentry *old_parent = NULL;
5419 * for regular files, if its inode is already on disk, we don't
5420 * have to worry about the parents at all. This is because
5421 * we can use the last_unlink_trans field to record renames
5422 * and other fun in this file.
5424 if (S_ISREG(inode->vfs_inode.i_mode) &&
5425 inode->generation <= last_committed &&
5426 inode->last_unlink_trans <= last_committed)
5429 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5430 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5432 inode = BTRFS_I(d_inode(parent));
5436 if (btrfs_must_commit_transaction(trans, inode)) {
5441 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5444 if (IS_ROOT(parent)) {
5445 inode = BTRFS_I(d_inode(parent));
5446 if (btrfs_must_commit_transaction(trans, inode))
5451 parent = dget_parent(parent);
5453 old_parent = parent;
5454 inode = BTRFS_I(d_inode(parent));
5462 struct btrfs_dir_list {
5464 struct list_head list;
5468 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5469 * details about the why it is needed.
5470 * This is a recursive operation - if an existing dentry corresponds to a
5471 * directory, that directory's new entries are logged too (same behaviour as
5472 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5473 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5474 * complains about the following circular lock dependency / possible deadlock:
5478 * lock(&type->i_mutex_dir_key#3/2);
5479 * lock(sb_internal#2);
5480 * lock(&type->i_mutex_dir_key#3/2);
5481 * lock(&sb->s_type->i_mutex_key#14);
5483 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5484 * sb_start_intwrite() in btrfs_start_transaction().
5485 * Not locking i_mutex of the inodes is still safe because:
5487 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5488 * that while logging the inode new references (names) are added or removed
5489 * from the inode, leaving the logged inode item with a link count that does
5490 * not match the number of logged inode reference items. This is fine because
5491 * at log replay time we compute the real number of links and correct the
5492 * link count in the inode item (see replay_one_buffer() and
5493 * link_to_fixup_dir());
5495 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5496 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5497 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5498 * has a size that doesn't match the sum of the lengths of all the logged
5499 * names. This does not result in a problem because if a dir_item key is
5500 * logged but its matching dir_index key is not logged, at log replay time we
5501 * don't use it to replay the respective name (see replay_one_name()). On the
5502 * other hand if only the dir_index key ends up being logged, the respective
5503 * name is added to the fs/subvol tree with both the dir_item and dir_index
5504 * keys created (see replay_one_name()).
5505 * The directory's inode item with a wrong i_size is not a problem as well,
5506 * since we don't use it at log replay time to set the i_size in the inode
5507 * item of the fs/subvol tree (see overwrite_item()).
5509 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5510 struct btrfs_root *root,
5511 struct btrfs_inode *start_inode,
5512 struct btrfs_log_ctx *ctx)
5514 struct btrfs_fs_info *fs_info = root->fs_info;
5515 struct btrfs_root *log = root->log_root;
5516 struct btrfs_path *path;
5517 LIST_HEAD(dir_list);
5518 struct btrfs_dir_list *dir_elem;
5521 path = btrfs_alloc_path();
5525 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5527 btrfs_free_path(path);
5530 dir_elem->ino = btrfs_ino(start_inode);
5531 list_add_tail(&dir_elem->list, &dir_list);
5533 while (!list_empty(&dir_list)) {
5534 struct extent_buffer *leaf;
5535 struct btrfs_key min_key;
5539 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5542 goto next_dir_inode;
5544 min_key.objectid = dir_elem->ino;
5545 min_key.type = BTRFS_DIR_ITEM_KEY;
5548 btrfs_release_path(path);
5549 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5551 goto next_dir_inode;
5552 } else if (ret > 0) {
5554 goto next_dir_inode;
5558 leaf = path->nodes[0];
5559 nritems = btrfs_header_nritems(leaf);
5560 for (i = path->slots[0]; i < nritems; i++) {
5561 struct btrfs_dir_item *di;
5562 struct btrfs_key di_key;
5563 struct inode *di_inode;
5564 struct btrfs_dir_list *new_dir_elem;
5565 int log_mode = LOG_INODE_EXISTS;
5568 btrfs_item_key_to_cpu(leaf, &min_key, i);
5569 if (min_key.objectid != dir_elem->ino ||
5570 min_key.type != BTRFS_DIR_ITEM_KEY)
5571 goto next_dir_inode;
5573 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5574 type = btrfs_dir_type(leaf, di);
5575 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5576 type != BTRFS_FT_DIR)
5578 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5579 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5582 btrfs_release_path(path);
5583 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5584 if (IS_ERR(di_inode)) {
5585 ret = PTR_ERR(di_inode);
5586 goto next_dir_inode;
5589 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5590 btrfs_add_delayed_iput(di_inode);
5594 ctx->log_new_dentries = false;
5595 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5596 log_mode = LOG_INODE_ALL;
5597 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5598 log_mode, 0, LLONG_MAX, ctx);
5600 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5602 btrfs_add_delayed_iput(di_inode);
5604 goto next_dir_inode;
5605 if (ctx->log_new_dentries) {
5606 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5608 if (!new_dir_elem) {
5610 goto next_dir_inode;
5612 new_dir_elem->ino = di_key.objectid;
5613 list_add_tail(&new_dir_elem->list, &dir_list);
5618 ret = btrfs_next_leaf(log, path);
5620 goto next_dir_inode;
5621 } else if (ret > 0) {
5623 goto next_dir_inode;
5627 if (min_key.offset < (u64)-1) {
5632 list_del(&dir_elem->list);
5636 btrfs_free_path(path);
5640 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5641 struct btrfs_inode *inode,
5642 struct btrfs_log_ctx *ctx)
5644 struct btrfs_fs_info *fs_info = trans->fs_info;
5646 struct btrfs_path *path;
5647 struct btrfs_key key;
5648 struct btrfs_root *root = inode->root;
5649 const u64 ino = btrfs_ino(inode);
5651 path = btrfs_alloc_path();
5654 path->skip_locking = 1;
5655 path->search_commit_root = 1;
5658 key.type = BTRFS_INODE_REF_KEY;
5660 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5665 struct extent_buffer *leaf = path->nodes[0];
5666 int slot = path->slots[0];
5671 if (slot >= btrfs_header_nritems(leaf)) {
5672 ret = btrfs_next_leaf(root, path);
5680 btrfs_item_key_to_cpu(leaf, &key, slot);
5681 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5682 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5685 item_size = btrfs_item_size_nr(leaf, slot);
5686 ptr = btrfs_item_ptr_offset(leaf, slot);
5687 while (cur_offset < item_size) {
5688 struct btrfs_key inode_key;
5689 struct inode *dir_inode;
5691 inode_key.type = BTRFS_INODE_ITEM_KEY;
5692 inode_key.offset = 0;
5694 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5695 struct btrfs_inode_extref *extref;
5697 extref = (struct btrfs_inode_extref *)
5699 inode_key.objectid = btrfs_inode_extref_parent(
5701 cur_offset += sizeof(*extref);
5702 cur_offset += btrfs_inode_extref_name_len(leaf,
5705 inode_key.objectid = key.offset;
5706 cur_offset = item_size;
5709 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
5712 * If the parent inode was deleted, return an error to
5713 * fallback to a transaction commit. This is to prevent
5714 * getting an inode that was moved from one parent A to
5715 * a parent B, got its former parent A deleted and then
5716 * it got fsync'ed, from existing at both parents after
5717 * a log replay (and the old parent still existing).
5724 * mv /mnt/B/bar /mnt/A/bar
5725 * mv -T /mnt/A /mnt/B
5729 * If we ignore the old parent B which got deleted,
5730 * after a log replay we would have file bar linked
5731 * at both parents and the old parent B would still
5734 if (IS_ERR(dir_inode)) {
5735 ret = PTR_ERR(dir_inode);
5740 ctx->log_new_dentries = false;
5741 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5742 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5744 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5746 if (!ret && ctx && ctx->log_new_dentries)
5747 ret = log_new_dir_dentries(trans, root,
5748 BTRFS_I(dir_inode), ctx);
5749 btrfs_add_delayed_iput(dir_inode);
5757 btrfs_free_path(path);
5761 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5762 struct btrfs_root *root,
5763 struct btrfs_path *path,
5764 struct btrfs_log_ctx *ctx)
5766 struct btrfs_key found_key;
5768 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5771 struct btrfs_fs_info *fs_info = root->fs_info;
5772 const u64 last_committed = fs_info->last_trans_committed;
5773 struct extent_buffer *leaf = path->nodes[0];
5774 int slot = path->slots[0];
5775 struct btrfs_key search_key;
5776 struct inode *inode;
5780 btrfs_release_path(path);
5782 ino = found_key.offset;
5784 search_key.objectid = found_key.offset;
5785 search_key.type = BTRFS_INODE_ITEM_KEY;
5786 search_key.offset = 0;
5787 inode = btrfs_iget(fs_info->sb, ino, root);
5789 return PTR_ERR(inode);
5791 if (BTRFS_I(inode)->generation > last_committed)
5792 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5795 btrfs_add_delayed_iput(inode);
5799 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5802 search_key.type = BTRFS_INODE_REF_KEY;
5803 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5807 leaf = path->nodes[0];
5808 slot = path->slots[0];
5809 if (slot >= btrfs_header_nritems(leaf)) {
5810 ret = btrfs_next_leaf(root, path);
5815 leaf = path->nodes[0];
5816 slot = path->slots[0];
5819 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5820 if (found_key.objectid != search_key.objectid ||
5821 found_key.type != BTRFS_INODE_REF_KEY)
5827 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5828 struct btrfs_inode *inode,
5829 struct dentry *parent,
5830 struct btrfs_log_ctx *ctx)
5832 struct btrfs_root *root = inode->root;
5833 struct btrfs_fs_info *fs_info = root->fs_info;
5834 struct dentry *old_parent = NULL;
5835 struct super_block *sb = inode->vfs_inode.i_sb;
5839 if (!parent || d_really_is_negative(parent) ||
5843 inode = BTRFS_I(d_inode(parent));
5844 if (root != inode->root)
5847 if (inode->generation > fs_info->last_trans_committed) {
5848 ret = btrfs_log_inode(trans, root, inode,
5849 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5853 if (IS_ROOT(parent))
5856 parent = dget_parent(parent);
5858 old_parent = parent;
5865 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5866 struct btrfs_inode *inode,
5867 struct dentry *parent,
5868 struct btrfs_log_ctx *ctx)
5870 struct btrfs_root *root = inode->root;
5871 const u64 ino = btrfs_ino(inode);
5872 struct btrfs_path *path;
5873 struct btrfs_key search_key;
5877 * For a single hard link case, go through a fast path that does not
5878 * need to iterate the fs/subvolume tree.
5880 if (inode->vfs_inode.i_nlink < 2)
5881 return log_new_ancestors_fast(trans, inode, parent, ctx);
5883 path = btrfs_alloc_path();
5887 search_key.objectid = ino;
5888 search_key.type = BTRFS_INODE_REF_KEY;
5889 search_key.offset = 0;
5891 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5898 struct extent_buffer *leaf = path->nodes[0];
5899 int slot = path->slots[0];
5900 struct btrfs_key found_key;
5902 if (slot >= btrfs_header_nritems(leaf)) {
5903 ret = btrfs_next_leaf(root, path);
5911 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5912 if (found_key.objectid != ino ||
5913 found_key.type > BTRFS_INODE_EXTREF_KEY)
5917 * Don't deal with extended references because they are rare
5918 * cases and too complex to deal with (we would need to keep
5919 * track of which subitem we are processing for each item in
5920 * this loop, etc). So just return some error to fallback to
5921 * a transaction commit.
5923 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
5929 * Logging ancestors needs to do more searches on the fs/subvol
5930 * tree, so it releases the path as needed to avoid deadlocks.
5931 * Keep track of the last inode ref key and resume from that key
5932 * after logging all new ancestors for the current hard link.
5934 memcpy(&search_key, &found_key, sizeof(search_key));
5936 ret = log_new_ancestors(trans, root, path, ctx);
5939 btrfs_release_path(path);
5944 btrfs_free_path(path);
5949 * helper function around btrfs_log_inode to make sure newly created
5950 * parent directories also end up in the log. A minimal inode and backref
5951 * only logging is done of any parent directories that are older than
5952 * the last committed transaction
5954 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5955 struct btrfs_inode *inode,
5956 struct dentry *parent,
5960 struct btrfs_log_ctx *ctx)
5962 struct btrfs_root *root = inode->root;
5963 struct btrfs_fs_info *fs_info = root->fs_info;
5964 struct super_block *sb;
5966 u64 last_committed = fs_info->last_trans_committed;
5967 bool log_dentries = false;
5969 sb = inode->vfs_inode.i_sb;
5971 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5977 * The prev transaction commit doesn't complete, we need do
5978 * full commit by ourselves.
5980 if (fs_info->last_trans_log_full_commit >
5981 fs_info->last_trans_committed) {
5986 if (btrfs_root_refs(&root->root_item) == 0) {
5991 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5997 * Skip already logged inodes or inodes corresponding to tmpfiles
5998 * (since logging them is pointless, a link count of 0 means they
5999 * will never be accessible).
6001 if (btrfs_inode_in_log(inode, trans->transid) ||
6002 inode->vfs_inode.i_nlink == 0) {
6003 ret = BTRFS_NO_LOG_SYNC;
6007 ret = start_log_trans(trans, root, ctx);
6011 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
6016 * for regular files, if its inode is already on disk, we don't
6017 * have to worry about the parents at all. This is because
6018 * we can use the last_unlink_trans field to record renames
6019 * and other fun in this file.
6021 if (S_ISREG(inode->vfs_inode.i_mode) &&
6022 inode->generation <= last_committed &&
6023 inode->last_unlink_trans <= last_committed) {
6028 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6029 log_dentries = true;
6032 * On unlink we must make sure all our current and old parent directory
6033 * inodes are fully logged. This is to prevent leaving dangling
6034 * directory index entries in directories that were our parents but are
6035 * not anymore. Not doing this results in old parent directory being
6036 * impossible to delete after log replay (rmdir will always fail with
6037 * error -ENOTEMPTY).
6043 * ln testdir/foo testdir/bar
6045 * unlink testdir/bar
6046 * xfs_io -c fsync testdir/foo
6048 * mount fs, triggers log replay
6050 * If we don't log the parent directory (testdir), after log replay the
6051 * directory still has an entry pointing to the file inode using the bar
6052 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6053 * the file inode has a link count of 1.
6059 * ln foo testdir/foo2
6060 * ln foo testdir/foo3
6062 * unlink testdir/foo3
6063 * xfs_io -c fsync foo
6065 * mount fs, triggers log replay
6067 * Similar as the first example, after log replay the parent directory
6068 * testdir still has an entry pointing to the inode file with name foo3
6069 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6070 * and has a link count of 2.
6072 if (inode->last_unlink_trans > last_committed) {
6073 ret = btrfs_log_all_parents(trans, inode, ctx);
6078 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6083 ret = log_new_dir_dentries(trans, root, inode, ctx);
6088 btrfs_set_log_full_commit(trans);
6093 btrfs_remove_log_ctx(root, ctx);
6094 btrfs_end_log_trans(root);
6100 * it is not safe to log dentry if the chunk root has added new
6101 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6102 * If this returns 1, you must commit the transaction to safely get your
6105 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6106 struct dentry *dentry,
6109 struct btrfs_log_ctx *ctx)
6111 struct dentry *parent = dget_parent(dentry);
6114 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6115 start, end, LOG_INODE_ALL, ctx);
6122 * should be called during mount to recover any replay any log trees
6125 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6128 struct btrfs_path *path;
6129 struct btrfs_trans_handle *trans;
6130 struct btrfs_key key;
6131 struct btrfs_key found_key;
6132 struct btrfs_root *log;
6133 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6134 struct walk_control wc = {
6135 .process_func = process_one_buffer,
6136 .stage = LOG_WALK_PIN_ONLY,
6139 path = btrfs_alloc_path();
6143 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6145 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6146 if (IS_ERR(trans)) {
6147 ret = PTR_ERR(trans);
6154 ret = walk_log_tree(trans, log_root_tree, &wc);
6156 btrfs_handle_fs_error(fs_info, ret,
6157 "Failed to pin buffers while recovering log root tree.");
6162 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6163 key.offset = (u64)-1;
6164 key.type = BTRFS_ROOT_ITEM_KEY;
6167 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6170 btrfs_handle_fs_error(fs_info, ret,
6171 "Couldn't find tree log root.");
6175 if (path->slots[0] == 0)
6179 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6181 btrfs_release_path(path);
6182 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6185 log = btrfs_read_tree_root(log_root_tree, &found_key);
6188 btrfs_handle_fs_error(fs_info, ret,
6189 "Couldn't read tree log root.");
6193 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6195 if (IS_ERR(wc.replay_dest)) {
6196 ret = PTR_ERR(wc.replay_dest);
6199 * We didn't find the subvol, likely because it was
6200 * deleted. This is ok, simply skip this log and go to
6203 * We need to exclude the root because we can't have
6204 * other log replays overwriting this log as we'll read
6205 * it back in a few more times. This will keep our
6206 * block from being modified, and we'll just bail for
6207 * each subsequent pass.
6210 ret = btrfs_pin_extent_for_log_replay(trans,
6213 btrfs_put_root(log);
6217 btrfs_handle_fs_error(fs_info, ret,
6218 "Couldn't read target root for tree log recovery.");
6222 wc.replay_dest->log_root = log;
6223 btrfs_record_root_in_trans(trans, wc.replay_dest);
6224 ret = walk_log_tree(trans, log, &wc);
6226 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6227 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6231 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6232 struct btrfs_root *root = wc.replay_dest;
6234 btrfs_release_path(path);
6237 * We have just replayed everything, and the highest
6238 * objectid of fs roots probably has changed in case
6239 * some inode_item's got replayed.
6241 * root->objectid_mutex is not acquired as log replay
6242 * could only happen during mount.
6244 ret = btrfs_find_highest_objectid(root,
6245 &root->highest_objectid);
6248 wc.replay_dest->log_root = NULL;
6249 btrfs_put_root(wc.replay_dest);
6250 btrfs_put_root(log);
6255 if (found_key.offset == 0)
6257 key.offset = found_key.offset - 1;
6259 btrfs_release_path(path);
6261 /* step one is to pin it all, step two is to replay just inodes */
6264 wc.process_func = replay_one_buffer;
6265 wc.stage = LOG_WALK_REPLAY_INODES;
6268 /* step three is to replay everything */
6269 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6274 btrfs_free_path(path);
6276 /* step 4: commit the transaction, which also unpins the blocks */
6277 ret = btrfs_commit_transaction(trans);
6281 log_root_tree->log_root = NULL;
6282 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6283 btrfs_put_root(log_root_tree);
6288 btrfs_end_transaction(wc.trans);
6289 btrfs_free_path(path);
6294 * there are some corner cases where we want to force a full
6295 * commit instead of allowing a directory to be logged.
6297 * They revolve around files there were unlinked from the directory, and
6298 * this function updates the parent directory so that a full commit is
6299 * properly done if it is fsync'd later after the unlinks are done.
6301 * Must be called before the unlink operations (updates to the subvolume tree,
6302 * inodes, etc) are done.
6304 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6305 struct btrfs_inode *dir, struct btrfs_inode *inode,
6309 * when we're logging a file, if it hasn't been renamed
6310 * or unlinked, and its inode is fully committed on disk,
6311 * we don't have to worry about walking up the directory chain
6312 * to log its parents.
6314 * So, we use the last_unlink_trans field to put this transid
6315 * into the file. When the file is logged we check it and
6316 * don't log the parents if the file is fully on disk.
6318 mutex_lock(&inode->log_mutex);
6319 inode->last_unlink_trans = trans->transid;
6320 mutex_unlock(&inode->log_mutex);
6323 * if this directory was already logged any new
6324 * names for this file/dir will get recorded
6326 if (dir->logged_trans == trans->transid)
6330 * if the inode we're about to unlink was logged,
6331 * the log will be properly updated for any new names
6333 if (inode->logged_trans == trans->transid)
6337 * when renaming files across directories, if the directory
6338 * there we're unlinking from gets fsync'd later on, there's
6339 * no way to find the destination directory later and fsync it
6340 * properly. So, we have to be conservative and force commits
6341 * so the new name gets discovered.
6346 /* we can safely do the unlink without any special recording */
6350 mutex_lock(&dir->log_mutex);
6351 dir->last_unlink_trans = trans->transid;
6352 mutex_unlock(&dir->log_mutex);
6356 * Make sure that if someone attempts to fsync the parent directory of a deleted
6357 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6358 * that after replaying the log tree of the parent directory's root we will not
6359 * see the snapshot anymore and at log replay time we will not see any log tree
6360 * corresponding to the deleted snapshot's root, which could lead to replaying
6361 * it after replaying the log tree of the parent directory (which would replay
6362 * the snapshot delete operation).
6364 * Must be called before the actual snapshot destroy operation (updates to the
6365 * parent root and tree of tree roots trees, etc) are done.
6367 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6368 struct btrfs_inode *dir)
6370 mutex_lock(&dir->log_mutex);
6371 dir->last_unlink_trans = trans->transid;
6372 mutex_unlock(&dir->log_mutex);
6376 * Call this after adding a new name for a file and it will properly
6377 * update the log to reflect the new name.
6379 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6380 * true (because it's not used).
6382 * Return value depends on whether @sync_log is true or false.
6383 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6384 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6386 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6387 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6388 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6389 * committed (without attempting to sync the log).
6391 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6392 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6393 struct dentry *parent,
6394 bool sync_log, struct btrfs_log_ctx *ctx)
6396 struct btrfs_fs_info *fs_info = trans->fs_info;
6400 * this will force the logging code to walk the dentry chain
6403 if (!S_ISDIR(inode->vfs_inode.i_mode))
6404 inode->last_unlink_trans = trans->transid;
6407 * if this inode hasn't been logged and directory we're renaming it
6408 * from hasn't been logged, we don't need to log it
6410 if (inode->logged_trans <= fs_info->last_trans_committed &&
6411 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6412 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6413 BTRFS_DONT_NEED_LOG_SYNC;
6416 struct btrfs_log_ctx ctx2;
6418 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6419 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6420 LOG_INODE_EXISTS, &ctx2);
6421 if (ret == BTRFS_NO_LOG_SYNC)
6422 return BTRFS_DONT_NEED_TRANS_COMMIT;
6424 return BTRFS_NEED_TRANS_COMMIT;
6426 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6428 return BTRFS_NEED_TRANS_COMMIT;
6429 return BTRFS_DONT_NEED_TRANS_COMMIT;
6433 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6434 LOG_INODE_EXISTS, ctx);
6435 if (ret == BTRFS_NO_LOG_SYNC)
6436 return BTRFS_DONT_NEED_LOG_SYNC;
6438 return BTRFS_NEED_TRANS_COMMIT;
6440 return BTRFS_NEED_LOG_SYNC;