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 "block-group.h"
21 #include "space-info.h"
23 #include "inode-item.h"
25 /* magic values for the inode_only field in btrfs_log_inode:
27 * LOG_INODE_ALL means to log everything
28 * LOG_INODE_EXISTS means to log just enough to recreate the inode
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
92 LOG_WALK_REPLAY_INODES,
93 LOG_WALK_REPLAY_DIR_INDEX,
97 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
98 struct btrfs_inode *inode,
100 struct btrfs_log_ctx *ctx);
101 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
102 struct btrfs_root *root,
103 struct btrfs_path *path, u64 objectid);
104 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
106 struct btrfs_root *log,
107 struct btrfs_path *path,
108 u64 dirid, int del_all);
109 static void wait_log_commit(struct btrfs_root *root, int transid);
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;
144 struct btrfs_root *tree_root = fs_info->tree_root;
145 const bool zoned = btrfs_is_zoned(fs_info);
147 bool created = false;
150 * First check if the log root tree was already created. If not, create
151 * it before locking the root's log_mutex, just to keep lockdep happy.
153 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state)) {
154 mutex_lock(&tree_root->log_mutex);
155 if (!fs_info->log_root_tree) {
156 ret = btrfs_init_log_root_tree(trans, fs_info);
158 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &tree_root->state);
162 mutex_unlock(&tree_root->log_mutex);
167 mutex_lock(&root->log_mutex);
170 if (root->log_root) {
171 int index = (root->log_transid + 1) % 2;
173 if (btrfs_need_log_full_commit(trans)) {
178 if (zoned && atomic_read(&root->log_commit[index])) {
179 wait_log_commit(root, root->log_transid - 1);
183 if (!root->log_start_pid) {
184 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
185 root->log_start_pid = current->pid;
186 } else if (root->log_start_pid != current->pid) {
187 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
191 * This means fs_info->log_root_tree was already created
192 * for some other FS trees. Do the full commit not to mix
193 * nodes from multiple log transactions to do sequential
196 if (zoned && !created) {
201 ret = btrfs_add_log_tree(trans, root);
205 set_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
206 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
207 root->log_start_pid = current->pid;
210 atomic_inc(&root->log_writers);
211 if (!ctx->logging_new_name) {
212 int index = root->log_transid % 2;
213 list_add_tail(&ctx->list, &root->log_ctxs[index]);
214 ctx->log_transid = root->log_transid;
218 mutex_unlock(&root->log_mutex);
223 * returns 0 if there was a log transaction running and we were able
224 * to join, or returns -ENOENT if there were not transactions
227 static int join_running_log_trans(struct btrfs_root *root)
229 const bool zoned = btrfs_is_zoned(root->fs_info);
232 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state))
235 mutex_lock(&root->log_mutex);
237 if (root->log_root) {
238 int index = (root->log_transid + 1) % 2;
241 if (zoned && atomic_read(&root->log_commit[index])) {
242 wait_log_commit(root, root->log_transid - 1);
245 atomic_inc(&root->log_writers);
247 mutex_unlock(&root->log_mutex);
252 * This either makes the current running log transaction wait
253 * until you call btrfs_end_log_trans() or it makes any future
254 * log transactions wait until you call btrfs_end_log_trans()
256 void btrfs_pin_log_trans(struct btrfs_root *root)
258 atomic_inc(&root->log_writers);
262 * indicate we're done making changes to the log tree
263 * and wake up anyone waiting to do a sync
265 void btrfs_end_log_trans(struct btrfs_root *root)
267 if (atomic_dec_and_test(&root->log_writers)) {
268 /* atomic_dec_and_test implies a barrier */
269 cond_wake_up_nomb(&root->log_writer_wait);
273 static int btrfs_write_tree_block(struct extent_buffer *buf)
275 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
276 buf->start + buf->len - 1);
279 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
281 filemap_fdatawait_range(buf->pages[0]->mapping,
282 buf->start, buf->start + buf->len - 1);
286 * the walk control struct is used to pass state down the chain when
287 * processing the log tree. The stage field tells us which part
288 * of the log tree processing we are currently doing. The others
289 * are state fields used for that specific part
291 struct walk_control {
292 /* should we free the extent on disk when done? This is used
293 * at transaction commit time while freeing a log tree
297 /* should we write out the extent buffer? This is used
298 * while flushing the log tree to disk during a sync
302 /* should we wait for the extent buffer io to finish? Also used
303 * while flushing the log tree to disk for a sync
307 /* pin only walk, we record which extents on disk belong to the
312 /* what stage of the replay code we're currently in */
316 * Ignore any items from the inode currently being processed. Needs
317 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
318 * the LOG_WALK_REPLAY_INODES stage.
320 bool ignore_cur_inode;
322 /* the root we are currently replaying */
323 struct btrfs_root *replay_dest;
325 /* the trans handle for the current replay */
326 struct btrfs_trans_handle *trans;
328 /* the function that gets used to process blocks we find in the
329 * tree. Note the extent_buffer might not be up to date when it is
330 * passed in, and it must be checked or read if you need the data
333 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
334 struct walk_control *wc, u64 gen, int level);
338 * process_func used to pin down extents, write them or wait on them
340 static int process_one_buffer(struct btrfs_root *log,
341 struct extent_buffer *eb,
342 struct walk_control *wc, u64 gen, int level)
344 struct btrfs_fs_info *fs_info = log->fs_info;
348 * If this fs is mixed then we need to be able to process the leaves to
349 * pin down any logged extents, so we have to read the block.
351 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
352 ret = btrfs_read_buffer(eb, gen, level, NULL);
358 ret = btrfs_pin_extent_for_log_replay(wc->trans, eb->start,
361 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
362 if (wc->pin && btrfs_header_level(eb) == 0)
363 ret = btrfs_exclude_logged_extents(eb);
365 btrfs_write_tree_block(eb);
367 btrfs_wait_tree_block_writeback(eb);
372 static int do_overwrite_item(struct btrfs_trans_handle *trans,
373 struct btrfs_root *root,
374 struct btrfs_path *path,
375 struct extent_buffer *eb, int slot,
376 struct btrfs_key *key)
380 u64 saved_i_size = 0;
381 int save_old_i_size = 0;
382 unsigned long src_ptr;
383 unsigned long dst_ptr;
384 int overwrite_root = 0;
385 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
387 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
390 item_size = btrfs_item_size(eb, slot);
391 src_ptr = btrfs_item_ptr_offset(eb, slot);
393 /* Our caller must have done a search for the key for us. */
394 ASSERT(path->nodes[0] != NULL);
397 * And the slot must point to the exact key or the slot where the key
398 * should be at (the first item with a key greater than 'key')
400 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
401 struct btrfs_key found_key;
403 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
404 ret = btrfs_comp_cpu_keys(&found_key, key);
413 u32 dst_size = btrfs_item_size(path->nodes[0],
415 if (dst_size != item_size)
418 if (item_size == 0) {
419 btrfs_release_path(path);
422 dst_copy = kmalloc(item_size, GFP_NOFS);
423 src_copy = kmalloc(item_size, GFP_NOFS);
424 if (!dst_copy || !src_copy) {
425 btrfs_release_path(path);
431 read_extent_buffer(eb, src_copy, src_ptr, item_size);
433 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
434 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
436 ret = memcmp(dst_copy, src_copy, item_size);
441 * they have the same contents, just return, this saves
442 * us from cowing blocks in the destination tree and doing
443 * extra writes that may not have been done by a previous
447 btrfs_release_path(path);
452 * We need to load the old nbytes into the inode so when we
453 * replay the extents we've logged we get the right nbytes.
456 struct btrfs_inode_item *item;
460 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
461 struct btrfs_inode_item);
462 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
463 item = btrfs_item_ptr(eb, slot,
464 struct btrfs_inode_item);
465 btrfs_set_inode_nbytes(eb, item, nbytes);
468 * If this is a directory we need to reset the i_size to
469 * 0 so that we can set it up properly when replaying
470 * the rest of the items in this log.
472 mode = btrfs_inode_mode(eb, item);
474 btrfs_set_inode_size(eb, item, 0);
476 } else if (inode_item) {
477 struct btrfs_inode_item *item;
481 * New inode, set nbytes to 0 so that the nbytes comes out
482 * properly when we replay the extents.
484 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
485 btrfs_set_inode_nbytes(eb, item, 0);
488 * If this is a directory we need to reset the i_size to 0 so
489 * that we can set it up properly when replaying the rest of
490 * the items in this log.
492 mode = btrfs_inode_mode(eb, item);
494 btrfs_set_inode_size(eb, item, 0);
497 btrfs_release_path(path);
498 /* try to insert the key into the destination tree */
499 path->skip_release_on_error = 1;
500 ret = btrfs_insert_empty_item(trans, root, path,
502 path->skip_release_on_error = 0;
504 /* make sure any existing item is the correct size */
505 if (ret == -EEXIST || ret == -EOVERFLOW) {
507 found_size = btrfs_item_size(path->nodes[0],
509 if (found_size > item_size)
510 btrfs_truncate_item(path, item_size, 1);
511 else if (found_size < item_size)
512 btrfs_extend_item(path, item_size - found_size);
516 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
519 /* don't overwrite an existing inode if the generation number
520 * was logged as zero. This is done when the tree logging code
521 * is just logging an inode to make sure it exists after recovery.
523 * Also, don't overwrite i_size on directories during replay.
524 * log replay inserts and removes directory items based on the
525 * state of the tree found in the subvolume, and i_size is modified
528 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
529 struct btrfs_inode_item *src_item;
530 struct btrfs_inode_item *dst_item;
532 src_item = (struct btrfs_inode_item *)src_ptr;
533 dst_item = (struct btrfs_inode_item *)dst_ptr;
535 if (btrfs_inode_generation(eb, src_item) == 0) {
536 struct extent_buffer *dst_eb = path->nodes[0];
537 const u64 ino_size = btrfs_inode_size(eb, src_item);
540 * For regular files an ino_size == 0 is used only when
541 * logging that an inode exists, as part of a directory
542 * fsync, and the inode wasn't fsynced before. In this
543 * case don't set the size of the inode in the fs/subvol
544 * tree, otherwise we would be throwing valid data away.
546 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
547 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
549 btrfs_set_inode_size(dst_eb, dst_item, ino_size);
553 if (overwrite_root &&
554 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
555 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
557 saved_i_size = btrfs_inode_size(path->nodes[0],
562 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
565 if (save_old_i_size) {
566 struct btrfs_inode_item *dst_item;
567 dst_item = (struct btrfs_inode_item *)dst_ptr;
568 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
571 /* make sure the generation is filled in */
572 if (key->type == BTRFS_INODE_ITEM_KEY) {
573 struct btrfs_inode_item *dst_item;
574 dst_item = (struct btrfs_inode_item *)dst_ptr;
575 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
576 btrfs_set_inode_generation(path->nodes[0], dst_item,
581 btrfs_mark_buffer_dirty(path->nodes[0]);
582 btrfs_release_path(path);
587 * Item overwrite used by replay and tree logging. eb, slot and key all refer
588 * to the src data we are copying out.
590 * root is the tree we are copying into, and path is a scratch
591 * path for use in this function (it should be released on entry and
592 * will be released on exit).
594 * If the key is already in the destination tree the existing item is
595 * overwritten. If the existing item isn't big enough, it is extended.
596 * If it is too large, it is truncated.
598 * If the key isn't in the destination yet, a new item is inserted.
600 static int overwrite_item(struct btrfs_trans_handle *trans,
601 struct btrfs_root *root,
602 struct btrfs_path *path,
603 struct extent_buffer *eb, int slot,
604 struct btrfs_key *key)
608 /* Look for the key in the destination tree. */
609 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
613 return do_overwrite_item(trans, root, path, eb, slot, key);
617 * simple helper to read an inode off the disk from a given root
618 * This can only be called for subvolume roots and not for the log
620 static noinline struct inode *read_one_inode(struct btrfs_root *root,
625 inode = btrfs_iget(root->fs_info->sb, objectid, root);
631 /* replays a single extent in 'eb' at 'slot' with 'key' into the
632 * subvolume 'root'. path is released on entry and should be released
635 * extents in the log tree have not been allocated out of the extent
636 * tree yet. So, this completes the allocation, taking a reference
637 * as required if the extent already exists or creating a new extent
638 * if it isn't in the extent allocation tree yet.
640 * The extent is inserted into the file, dropping any existing extents
641 * from the file that overlap the new one.
643 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
644 struct btrfs_root *root,
645 struct btrfs_path *path,
646 struct extent_buffer *eb, int slot,
647 struct btrfs_key *key)
649 struct btrfs_drop_extents_args drop_args = { 0 };
650 struct btrfs_fs_info *fs_info = root->fs_info;
653 u64 start = key->offset;
655 struct btrfs_file_extent_item *item;
656 struct inode *inode = NULL;
660 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
661 found_type = btrfs_file_extent_type(eb, item);
663 if (found_type == BTRFS_FILE_EXTENT_REG ||
664 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
665 nbytes = btrfs_file_extent_num_bytes(eb, item);
666 extent_end = start + nbytes;
669 * We don't add to the inodes nbytes if we are prealloc or a
672 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
674 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
675 size = btrfs_file_extent_ram_bytes(eb, item);
676 nbytes = btrfs_file_extent_ram_bytes(eb, item);
677 extent_end = ALIGN(start + size,
678 fs_info->sectorsize);
684 inode = read_one_inode(root, key->objectid);
691 * first check to see if we already have this extent in the
692 * file. This must be done before the btrfs_drop_extents run
693 * so we don't try to drop this extent.
695 ret = btrfs_lookup_file_extent(trans, root, path,
696 btrfs_ino(BTRFS_I(inode)), start, 0);
699 (found_type == BTRFS_FILE_EXTENT_REG ||
700 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
701 struct btrfs_file_extent_item cmp1;
702 struct btrfs_file_extent_item cmp2;
703 struct btrfs_file_extent_item *existing;
704 struct extent_buffer *leaf;
706 leaf = path->nodes[0];
707 existing = btrfs_item_ptr(leaf, path->slots[0],
708 struct btrfs_file_extent_item);
710 read_extent_buffer(eb, &cmp1, (unsigned long)item,
712 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
716 * we already have a pointer to this exact extent,
717 * we don't have to do anything
719 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
720 btrfs_release_path(path);
724 btrfs_release_path(path);
726 /* drop any overlapping extents */
727 drop_args.start = start;
728 drop_args.end = extent_end;
729 drop_args.drop_cache = true;
730 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
734 if (found_type == BTRFS_FILE_EXTENT_REG ||
735 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
737 unsigned long dest_offset;
738 struct btrfs_key ins;
740 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
741 btrfs_fs_incompat(fs_info, NO_HOLES))
744 ret = btrfs_insert_empty_item(trans, root, path, key,
748 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
750 copy_extent_buffer(path->nodes[0], eb, dest_offset,
751 (unsigned long)item, sizeof(*item));
753 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
754 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
755 ins.type = BTRFS_EXTENT_ITEM_KEY;
756 offset = key->offset - btrfs_file_extent_offset(eb, item);
759 * Manually record dirty extent, as here we did a shallow
760 * file extent item copy and skip normal backref update,
761 * but modifying extent tree all by ourselves.
762 * So need to manually record dirty extent for qgroup,
763 * as the owner of the file extent changed from log tree
764 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
766 ret = btrfs_qgroup_trace_extent(trans,
767 btrfs_file_extent_disk_bytenr(eb, item),
768 btrfs_file_extent_disk_num_bytes(eb, item),
773 if (ins.objectid > 0) {
774 struct btrfs_ref ref = { 0 };
777 LIST_HEAD(ordered_sums);
780 * is this extent already allocated in the extent
781 * allocation tree? If so, just add a reference
783 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
787 } else if (ret == 0) {
788 btrfs_init_generic_ref(&ref,
789 BTRFS_ADD_DELAYED_REF,
790 ins.objectid, ins.offset, 0);
791 btrfs_init_data_ref(&ref,
792 root->root_key.objectid,
793 key->objectid, offset, 0, false);
794 ret = btrfs_inc_extent_ref(trans, &ref);
799 * insert the extent pointer in the extent
802 ret = btrfs_alloc_logged_file_extent(trans,
803 root->root_key.objectid,
804 key->objectid, offset, &ins);
808 btrfs_release_path(path);
810 if (btrfs_file_extent_compression(eb, item)) {
811 csum_start = ins.objectid;
812 csum_end = csum_start + ins.offset;
814 csum_start = ins.objectid +
815 btrfs_file_extent_offset(eb, item);
816 csum_end = csum_start +
817 btrfs_file_extent_num_bytes(eb, item);
820 ret = btrfs_lookup_csums_range(root->log_root,
821 csum_start, csum_end - 1,
826 * Now delete all existing cums in the csum root that
827 * cover our range. We do this because we can have an
828 * extent that is completely referenced by one file
829 * extent item and partially referenced by another
830 * file extent item (like after using the clone or
831 * extent_same ioctls). In this case if we end up doing
832 * the replay of the one that partially references the
833 * extent first, and we do not do the csum deletion
834 * below, we can get 2 csum items in the csum tree that
835 * overlap each other. For example, imagine our log has
836 * the two following file extent items:
838 * key (257 EXTENT_DATA 409600)
839 * extent data disk byte 12845056 nr 102400
840 * extent data offset 20480 nr 20480 ram 102400
842 * key (257 EXTENT_DATA 819200)
843 * extent data disk byte 12845056 nr 102400
844 * extent data offset 0 nr 102400 ram 102400
846 * Where the second one fully references the 100K extent
847 * that starts at disk byte 12845056, and the log tree
848 * has a single csum item that covers the entire range
851 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
853 * After the first file extent item is replayed, the
854 * csum tree gets the following csum item:
856 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
858 * Which covers the 20K sub-range starting at offset 20K
859 * of our extent. Now when we replay the second file
860 * extent item, if we do not delete existing csum items
861 * that cover any of its blocks, we end up getting two
862 * csum items in our csum tree that overlap each other:
864 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
865 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
867 * Which is a problem, because after this anyone trying
868 * to lookup up for the checksum of any block of our
869 * extent starting at an offset of 40K or higher, will
870 * end up looking at the second csum item only, which
871 * does not contain the checksum for any block starting
872 * at offset 40K or higher of our extent.
874 while (!list_empty(&ordered_sums)) {
875 struct btrfs_ordered_sum *sums;
876 struct btrfs_root *csum_root;
878 sums = list_entry(ordered_sums.next,
879 struct btrfs_ordered_sum,
881 csum_root = btrfs_csum_root(fs_info,
884 ret = btrfs_del_csums(trans, csum_root,
888 ret = btrfs_csum_file_blocks(trans,
891 list_del(&sums->list);
897 btrfs_release_path(path);
899 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
900 /* inline extents are easy, we just overwrite them */
901 ret = overwrite_item(trans, root, path, eb, slot, key);
906 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start,
912 btrfs_update_inode_bytes(BTRFS_I(inode), nbytes, drop_args.bytes_found);
913 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
921 * when cleaning up conflicts between the directory names in the
922 * subvolume, directory names in the log and directory names in the
923 * inode back references, we may have to unlink inodes from directories.
925 * This is a helper function to do the unlink of a specific directory
928 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
929 struct btrfs_path *path,
930 struct btrfs_inode *dir,
931 struct btrfs_dir_item *di)
933 struct btrfs_root *root = dir->root;
937 struct extent_buffer *leaf;
938 struct btrfs_key location;
941 leaf = path->nodes[0];
943 btrfs_dir_item_key_to_cpu(leaf, di, &location);
944 name_len = btrfs_dir_name_len(leaf, di);
945 name = kmalloc(name_len, GFP_NOFS);
949 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
950 btrfs_release_path(path);
952 inode = read_one_inode(root, location.objectid);
958 ret = link_to_fixup_dir(trans, root, path, location.objectid);
962 ret = btrfs_unlink_inode(trans, dir, BTRFS_I(inode), name,
967 ret = btrfs_run_delayed_items(trans);
975 * See if a given name and sequence number found in an inode back reference are
976 * already in a directory and correctly point to this inode.
978 * Returns: < 0 on error, 0 if the directory entry does not exists and 1 if it
981 static noinline int inode_in_dir(struct btrfs_root *root,
982 struct btrfs_path *path,
983 u64 dirid, u64 objectid, u64 index,
984 const char *name, int name_len)
986 struct btrfs_dir_item *di;
987 struct btrfs_key location;
990 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
991 index, name, name_len, 0);
996 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
997 if (location.objectid != objectid)
1003 btrfs_release_path(path);
1004 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
1009 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
1010 if (location.objectid == objectid)
1014 btrfs_release_path(path);
1019 * helper function to check a log tree for a named back reference in
1020 * an inode. This is used to decide if a back reference that is
1021 * found in the subvolume conflicts with what we find in the log.
1023 * inode backreferences may have multiple refs in a single item,
1024 * during replay we process one reference at a time, and we don't
1025 * want to delete valid links to a file from the subvolume if that
1026 * link is also in the log.
1028 static noinline int backref_in_log(struct btrfs_root *log,
1029 struct btrfs_key *key,
1031 const char *name, int namelen)
1033 struct btrfs_path *path;
1036 path = btrfs_alloc_path();
1040 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
1043 } else if (ret == 1) {
1048 if (key->type == BTRFS_INODE_EXTREF_KEY)
1049 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1054 ret = !!btrfs_find_name_in_backref(path->nodes[0],
1058 btrfs_free_path(path);
1062 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
1063 struct btrfs_root *root,
1064 struct btrfs_path *path,
1065 struct btrfs_root *log_root,
1066 struct btrfs_inode *dir,
1067 struct btrfs_inode *inode,
1068 u64 inode_objectid, u64 parent_objectid,
1069 u64 ref_index, char *name, int namelen,
1074 int victim_name_len;
1075 struct extent_buffer *leaf;
1076 struct btrfs_dir_item *di;
1077 struct btrfs_key search_key;
1078 struct btrfs_inode_extref *extref;
1081 /* Search old style refs */
1082 search_key.objectid = inode_objectid;
1083 search_key.type = BTRFS_INODE_REF_KEY;
1084 search_key.offset = parent_objectid;
1085 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1087 struct btrfs_inode_ref *victim_ref;
1089 unsigned long ptr_end;
1091 leaf = path->nodes[0];
1093 /* are we trying to overwrite a back ref for the root directory
1094 * if so, just jump out, we're done
1096 if (search_key.objectid == search_key.offset)
1099 /* check all the names in this back reference to see
1100 * if they are in the log. if so, we allow them to stay
1101 * otherwise they must be unlinked as a conflict
1103 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1104 ptr_end = ptr + btrfs_item_size(leaf, path->slots[0]);
1105 while (ptr < ptr_end) {
1106 victim_ref = (struct btrfs_inode_ref *)ptr;
1107 victim_name_len = btrfs_inode_ref_name_len(leaf,
1109 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1113 read_extent_buffer(leaf, victim_name,
1114 (unsigned long)(victim_ref + 1),
1117 ret = backref_in_log(log_root, &search_key,
1118 parent_objectid, victim_name,
1124 inc_nlink(&inode->vfs_inode);
1125 btrfs_release_path(path);
1127 ret = btrfs_unlink_inode(trans, dir, inode,
1128 victim_name, victim_name_len);
1132 ret = btrfs_run_delayed_items(trans);
1140 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1144 * NOTE: we have searched root tree and checked the
1145 * corresponding ref, it does not need to check again.
1149 btrfs_release_path(path);
1151 /* Same search but for extended refs */
1152 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1153 inode_objectid, parent_objectid, 0,
1155 if (!IS_ERR_OR_NULL(extref)) {
1159 struct inode *victim_parent;
1161 leaf = path->nodes[0];
1163 item_size = btrfs_item_size(leaf, path->slots[0]);
1164 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1166 while (cur_offset < item_size) {
1167 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1169 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1171 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1174 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1177 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1180 search_key.objectid = inode_objectid;
1181 search_key.type = BTRFS_INODE_EXTREF_KEY;
1182 search_key.offset = btrfs_extref_hash(parent_objectid,
1185 ret = backref_in_log(log_root, &search_key,
1186 parent_objectid, victim_name,
1193 victim_parent = read_one_inode(root,
1195 if (victim_parent) {
1196 inc_nlink(&inode->vfs_inode);
1197 btrfs_release_path(path);
1199 ret = btrfs_unlink_inode(trans,
1200 BTRFS_I(victim_parent),
1205 ret = btrfs_run_delayed_items(
1208 iput(victim_parent);
1217 cur_offset += victim_name_len + sizeof(*extref);
1221 btrfs_release_path(path);
1223 /* look for a conflicting sequence number */
1224 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1225 ref_index, name, namelen, 0);
1229 ret = drop_one_dir_item(trans, path, dir, di);
1233 btrfs_release_path(path);
1235 /* look for a conflicting name */
1236 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1241 ret = drop_one_dir_item(trans, path, dir, di);
1245 btrfs_release_path(path);
1250 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1251 u32 *namelen, char **name, u64 *index,
1252 u64 *parent_objectid)
1254 struct btrfs_inode_extref *extref;
1256 extref = (struct btrfs_inode_extref *)ref_ptr;
1258 *namelen = btrfs_inode_extref_name_len(eb, extref);
1259 *name = kmalloc(*namelen, GFP_NOFS);
1263 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1267 *index = btrfs_inode_extref_index(eb, extref);
1268 if (parent_objectid)
1269 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1274 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1275 u32 *namelen, char **name, u64 *index)
1277 struct btrfs_inode_ref *ref;
1279 ref = (struct btrfs_inode_ref *)ref_ptr;
1281 *namelen = btrfs_inode_ref_name_len(eb, ref);
1282 *name = kmalloc(*namelen, GFP_NOFS);
1286 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1289 *index = btrfs_inode_ref_index(eb, ref);
1295 * Take an inode reference item from the log tree and iterate all names from the
1296 * inode reference item in the subvolume tree with the same key (if it exists).
1297 * For any name that is not in the inode reference item from the log tree, do a
1298 * proper unlink of that name (that is, remove its entry from the inode
1299 * reference item and both dir index keys).
1301 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1302 struct btrfs_root *root,
1303 struct btrfs_path *path,
1304 struct btrfs_inode *inode,
1305 struct extent_buffer *log_eb,
1307 struct btrfs_key *key)
1310 unsigned long ref_ptr;
1311 unsigned long ref_end;
1312 struct extent_buffer *eb;
1315 btrfs_release_path(path);
1316 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1324 eb = path->nodes[0];
1325 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1326 ref_end = ref_ptr + btrfs_item_size(eb, path->slots[0]);
1327 while (ref_ptr < ref_end) {
1332 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1333 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1336 parent_id = key->offset;
1337 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1343 if (key->type == BTRFS_INODE_EXTREF_KEY)
1344 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1348 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1354 btrfs_release_path(path);
1355 dir = read_one_inode(root, parent_id);
1361 ret = btrfs_unlink_inode(trans, BTRFS_I(dir),
1362 inode, name, namelen);
1372 if (key->type == BTRFS_INODE_EXTREF_KEY)
1373 ref_ptr += sizeof(struct btrfs_inode_extref);
1375 ref_ptr += sizeof(struct btrfs_inode_ref);
1379 btrfs_release_path(path);
1383 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1384 const u8 ref_type, const char *name,
1387 struct btrfs_key key;
1388 struct btrfs_path *path;
1389 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1392 path = btrfs_alloc_path();
1396 key.objectid = btrfs_ino(BTRFS_I(inode));
1397 key.type = ref_type;
1398 if (key.type == BTRFS_INODE_REF_KEY)
1399 key.offset = parent_id;
1401 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1403 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1410 if (key.type == BTRFS_INODE_EXTREF_KEY)
1411 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1412 path->slots[0], parent_id, name, namelen);
1414 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1418 btrfs_free_path(path);
1422 static int add_link(struct btrfs_trans_handle *trans,
1423 struct inode *dir, struct inode *inode, const char *name,
1424 int namelen, u64 ref_index)
1426 struct btrfs_root *root = BTRFS_I(dir)->root;
1427 struct btrfs_dir_item *dir_item;
1428 struct btrfs_key key;
1429 struct btrfs_path *path;
1430 struct inode *other_inode = NULL;
1433 path = btrfs_alloc_path();
1437 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1438 btrfs_ino(BTRFS_I(dir)),
1441 btrfs_release_path(path);
1443 } else if (IS_ERR(dir_item)) {
1444 ret = PTR_ERR(dir_item);
1449 * Our inode's dentry collides with the dentry of another inode which is
1450 * in the log but not yet processed since it has a higher inode number.
1451 * So delete that other dentry.
1453 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1454 btrfs_release_path(path);
1455 other_inode = read_one_inode(root, key.objectid);
1460 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(other_inode),
1465 * If we dropped the link count to 0, bump it so that later the iput()
1466 * on the inode will not free it. We will fixup the link count later.
1468 if (other_inode->i_nlink == 0)
1469 inc_nlink(other_inode);
1471 ret = btrfs_run_delayed_items(trans);
1475 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1476 name, namelen, 0, ref_index);
1479 btrfs_free_path(path);
1485 * replay one inode back reference item found in the log tree.
1486 * eb, slot and key refer to the buffer and key found in the log tree.
1487 * root is the destination we are replaying into, and path is for temp
1488 * use by this function. (it should be released on return).
1490 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1491 struct btrfs_root *root,
1492 struct btrfs_root *log,
1493 struct btrfs_path *path,
1494 struct extent_buffer *eb, int slot,
1495 struct btrfs_key *key)
1497 struct inode *dir = NULL;
1498 struct inode *inode = NULL;
1499 unsigned long ref_ptr;
1500 unsigned long ref_end;
1504 int search_done = 0;
1505 int log_ref_ver = 0;
1506 u64 parent_objectid;
1509 int ref_struct_size;
1511 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1512 ref_end = ref_ptr + btrfs_item_size(eb, slot);
1514 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1515 struct btrfs_inode_extref *r;
1517 ref_struct_size = sizeof(struct btrfs_inode_extref);
1519 r = (struct btrfs_inode_extref *)ref_ptr;
1520 parent_objectid = btrfs_inode_extref_parent(eb, r);
1522 ref_struct_size = sizeof(struct btrfs_inode_ref);
1523 parent_objectid = key->offset;
1525 inode_objectid = key->objectid;
1528 * it is possible that we didn't log all the parent directories
1529 * for a given inode. If we don't find the dir, just don't
1530 * copy the back ref in. The link count fixup code will take
1533 dir = read_one_inode(root, parent_objectid);
1539 inode = read_one_inode(root, inode_objectid);
1545 while (ref_ptr < ref_end) {
1547 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1548 &ref_index, &parent_objectid);
1550 * parent object can change from one array
1554 dir = read_one_inode(root, parent_objectid);
1560 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1566 ret = inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1567 btrfs_ino(BTRFS_I(inode)), ref_index,
1571 } else if (ret == 0) {
1573 * look for a conflicting back reference in the
1574 * metadata. if we find one we have to unlink that name
1575 * of the file before we add our new link. Later on, we
1576 * overwrite any existing back reference, and we don't
1577 * want to create dangling pointers in the directory.
1581 ret = __add_inode_ref(trans, root, path, log,
1586 ref_index, name, namelen,
1596 * If a reference item already exists for this inode
1597 * with the same parent and name, but different index,
1598 * drop it and the corresponding directory index entries
1599 * from the parent before adding the new reference item
1600 * and dir index entries, otherwise we would fail with
1601 * -EEXIST returned from btrfs_add_link() below.
1603 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1606 ret = btrfs_unlink_inode(trans,
1611 * If we dropped the link count to 0, bump it so
1612 * that later the iput() on the inode will not
1613 * free it. We will fixup the link count later.
1615 if (!ret && inode->i_nlink == 0)
1621 /* insert our name */
1622 ret = add_link(trans, dir, inode, name, namelen,
1627 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1631 /* Else, ret == 1, we already have a perfect match, we're done. */
1633 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1643 * Before we overwrite the inode reference item in the subvolume tree
1644 * with the item from the log tree, we must unlink all names from the
1645 * parent directory that are in the subvolume's tree inode reference
1646 * item, otherwise we end up with an inconsistent subvolume tree where
1647 * dir index entries exist for a name but there is no inode reference
1648 * item with the same name.
1650 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1655 /* finally write the back reference in the inode */
1656 ret = overwrite_item(trans, root, path, eb, slot, key);
1658 btrfs_release_path(path);
1665 static int count_inode_extrefs(struct btrfs_root *root,
1666 struct btrfs_inode *inode, struct btrfs_path *path)
1670 unsigned int nlink = 0;
1673 u64 inode_objectid = btrfs_ino(inode);
1676 struct btrfs_inode_extref *extref;
1677 struct extent_buffer *leaf;
1680 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1685 leaf = path->nodes[0];
1686 item_size = btrfs_item_size(leaf, path->slots[0]);
1687 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1690 while (cur_offset < item_size) {
1691 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1692 name_len = btrfs_inode_extref_name_len(leaf, extref);
1696 cur_offset += name_len + sizeof(*extref);
1700 btrfs_release_path(path);
1702 btrfs_release_path(path);
1704 if (ret < 0 && ret != -ENOENT)
1709 static int count_inode_refs(struct btrfs_root *root,
1710 struct btrfs_inode *inode, struct btrfs_path *path)
1713 struct btrfs_key key;
1714 unsigned int nlink = 0;
1716 unsigned long ptr_end;
1718 u64 ino = btrfs_ino(inode);
1721 key.type = BTRFS_INODE_REF_KEY;
1722 key.offset = (u64)-1;
1725 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1729 if (path->slots[0] == 0)
1734 btrfs_item_key_to_cpu(path->nodes[0], &key,
1736 if (key.objectid != ino ||
1737 key.type != BTRFS_INODE_REF_KEY)
1739 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1740 ptr_end = ptr + btrfs_item_size(path->nodes[0],
1742 while (ptr < ptr_end) {
1743 struct btrfs_inode_ref *ref;
1745 ref = (struct btrfs_inode_ref *)ptr;
1746 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1748 ptr = (unsigned long)(ref + 1) + name_len;
1752 if (key.offset == 0)
1754 if (path->slots[0] > 0) {
1759 btrfs_release_path(path);
1761 btrfs_release_path(path);
1767 * There are a few corners where the link count of the file can't
1768 * be properly maintained during replay. So, instead of adding
1769 * lots of complexity to the log code, we just scan the backrefs
1770 * for any file that has been through replay.
1772 * The scan will update the link count on the inode to reflect the
1773 * number of back refs found. If it goes down to zero, the iput
1774 * will free the inode.
1776 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1777 struct btrfs_root *root,
1778 struct inode *inode)
1780 struct btrfs_path *path;
1783 u64 ino = btrfs_ino(BTRFS_I(inode));
1785 path = btrfs_alloc_path();
1789 ret = count_inode_refs(root, BTRFS_I(inode), path);
1795 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1803 if (nlink != inode->i_nlink) {
1804 set_nlink(inode, nlink);
1805 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1809 BTRFS_I(inode)->index_cnt = (u64)-1;
1811 if (inode->i_nlink == 0) {
1812 if (S_ISDIR(inode->i_mode)) {
1813 ret = replay_dir_deletes(trans, root, NULL, path,
1818 ret = btrfs_insert_orphan_item(trans, root, ino);
1824 btrfs_free_path(path);
1828 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1829 struct btrfs_root *root,
1830 struct btrfs_path *path)
1833 struct btrfs_key key;
1834 struct inode *inode;
1836 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1837 key.type = BTRFS_ORPHAN_ITEM_KEY;
1838 key.offset = (u64)-1;
1840 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1846 if (path->slots[0] == 0)
1851 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1852 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1853 key.type != BTRFS_ORPHAN_ITEM_KEY)
1856 ret = btrfs_del_item(trans, root, path);
1860 btrfs_release_path(path);
1861 inode = read_one_inode(root, key.offset);
1867 ret = fixup_inode_link_count(trans, root, inode);
1873 * fixup on a directory may create new entries,
1874 * make sure we always look for the highset possible
1877 key.offset = (u64)-1;
1879 btrfs_release_path(path);
1885 * record a given inode in the fixup dir so we can check its link
1886 * count when replay is done. The link count is incremented here
1887 * so the inode won't go away until we check it
1889 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1890 struct btrfs_root *root,
1891 struct btrfs_path *path,
1894 struct btrfs_key key;
1896 struct inode *inode;
1898 inode = read_one_inode(root, objectid);
1902 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1903 key.type = BTRFS_ORPHAN_ITEM_KEY;
1904 key.offset = objectid;
1906 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1908 btrfs_release_path(path);
1910 if (!inode->i_nlink)
1911 set_nlink(inode, 1);
1914 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
1915 } else if (ret == -EEXIST) {
1924 * when replaying the log for a directory, we only insert names
1925 * for inodes that actually exist. This means an fsync on a directory
1926 * does not implicitly fsync all the new files in it
1928 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1929 struct btrfs_root *root,
1930 u64 dirid, u64 index,
1931 char *name, int name_len,
1932 struct btrfs_key *location)
1934 struct inode *inode;
1938 inode = read_one_inode(root, location->objectid);
1942 dir = read_one_inode(root, dirid);
1948 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1949 name_len, 1, index);
1951 /* FIXME, put inode into FIXUP list */
1958 static int delete_conflicting_dir_entry(struct btrfs_trans_handle *trans,
1959 struct btrfs_inode *dir,
1960 struct btrfs_path *path,
1961 struct btrfs_dir_item *dst_di,
1962 const struct btrfs_key *log_key,
1966 struct btrfs_key found_key;
1968 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1969 /* The existing dentry points to the same inode, don't delete it. */
1970 if (found_key.objectid == log_key->objectid &&
1971 found_key.type == log_key->type &&
1972 found_key.offset == log_key->offset &&
1973 btrfs_dir_type(path->nodes[0], dst_di) == log_type)
1977 * Don't drop the conflicting directory entry if the inode for the new
1978 * entry doesn't exist.
1983 return drop_one_dir_item(trans, path, dir, dst_di);
1987 * take a single entry in a log directory item and replay it into
1990 * if a conflicting item exists in the subdirectory already,
1991 * the inode it points to is unlinked and put into the link count
1994 * If a name from the log points to a file or directory that does
1995 * not exist in the FS, it is skipped. fsyncs on directories
1996 * do not force down inodes inside that directory, just changes to the
1997 * names or unlinks in a directory.
1999 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
2000 * non-existing inode) and 1 if the name was replayed.
2002 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
2003 struct btrfs_root *root,
2004 struct btrfs_path *path,
2005 struct extent_buffer *eb,
2006 struct btrfs_dir_item *di,
2007 struct btrfs_key *key)
2011 struct btrfs_dir_item *dir_dst_di;
2012 struct btrfs_dir_item *index_dst_di;
2013 bool dir_dst_matches = false;
2014 bool index_dst_matches = false;
2015 struct btrfs_key log_key;
2016 struct btrfs_key search_key;
2021 bool update_size = true;
2022 bool name_added = false;
2024 dir = read_one_inode(root, key->objectid);
2028 name_len = btrfs_dir_name_len(eb, di);
2029 name = kmalloc(name_len, GFP_NOFS);
2035 log_type = btrfs_dir_type(eb, di);
2036 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2039 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
2040 ret = btrfs_lookup_inode(trans, root, path, &log_key, 0);
2041 btrfs_release_path(path);
2044 exists = (ret == 0);
2047 dir_dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
2049 if (IS_ERR(dir_dst_di)) {
2050 ret = PTR_ERR(dir_dst_di);
2052 } else if (dir_dst_di) {
2053 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
2054 dir_dst_di, &log_key, log_type,
2058 dir_dst_matches = (ret == 1);
2061 btrfs_release_path(path);
2063 index_dst_di = btrfs_lookup_dir_index_item(trans, root, path,
2064 key->objectid, key->offset,
2066 if (IS_ERR(index_dst_di)) {
2067 ret = PTR_ERR(index_dst_di);
2069 } else if (index_dst_di) {
2070 ret = delete_conflicting_dir_entry(trans, BTRFS_I(dir), path,
2071 index_dst_di, &log_key,
2075 index_dst_matches = (ret == 1);
2078 btrfs_release_path(path);
2080 if (dir_dst_matches && index_dst_matches) {
2082 update_size = false;
2087 * Check if the inode reference exists in the log for the given name,
2088 * inode and parent inode
2090 search_key.objectid = log_key.objectid;
2091 search_key.type = BTRFS_INODE_REF_KEY;
2092 search_key.offset = key->objectid;
2093 ret = backref_in_log(root->log_root, &search_key, 0, name, name_len);
2097 /* The dentry will be added later. */
2099 update_size = false;
2103 search_key.objectid = log_key.objectid;
2104 search_key.type = BTRFS_INODE_EXTREF_KEY;
2105 search_key.offset = key->objectid;
2106 ret = backref_in_log(root->log_root, &search_key, key->objectid, name,
2111 /* The dentry will be added later. */
2113 update_size = false;
2116 btrfs_release_path(path);
2117 ret = insert_one_name(trans, root, key->objectid, key->offset,
2118 name, name_len, &log_key);
2119 if (ret && ret != -ENOENT && ret != -EEXIST)
2123 update_size = false;
2127 if (!ret && update_size) {
2128 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2129 ret = btrfs_update_inode(trans, root, BTRFS_I(dir));
2133 if (!ret && name_added)
2138 /* Replay one dir item from a BTRFS_DIR_INDEX_KEY key. */
2139 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2140 struct btrfs_root *root,
2141 struct btrfs_path *path,
2142 struct extent_buffer *eb, int slot,
2143 struct btrfs_key *key)
2146 struct btrfs_dir_item *di;
2148 /* We only log dir index keys, which only contain a single dir item. */
2149 ASSERT(key->type == BTRFS_DIR_INDEX_KEY);
2151 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2152 ret = replay_one_name(trans, root, path, eb, di, key);
2157 * If this entry refers to a non-directory (directories can not have a
2158 * link count > 1) and it was added in the transaction that was not
2159 * committed, make sure we fixup the link count of the inode the entry
2160 * points to. Otherwise something like the following would result in a
2161 * directory pointing to an inode with a wrong link that does not account
2162 * for this dir entry:
2169 * ln testdir/bar testdir/bar_link
2170 * ln testdir/foo testdir/foo_link
2171 * xfs_io -c "fsync" testdir/bar
2175 * mount fs, log replay happens
2177 * File foo would remain with a link count of 1 when it has two entries
2178 * pointing to it in the directory testdir. This would make it impossible
2179 * to ever delete the parent directory has it would result in stale
2180 * dentries that can never be deleted.
2182 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2183 struct btrfs_path *fixup_path;
2184 struct btrfs_key di_key;
2186 fixup_path = btrfs_alloc_path();
2190 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2191 ret = link_to_fixup_dir(trans, root, fixup_path, di_key.objectid);
2192 btrfs_free_path(fixup_path);
2199 * directory replay has two parts. There are the standard directory
2200 * items in the log copied from the subvolume, and range items
2201 * created in the log while the subvolume was logged.
2203 * The range items tell us which parts of the key space the log
2204 * is authoritative for. During replay, if a key in the subvolume
2205 * directory is in a logged range item, but not actually in the log
2206 * that means it was deleted from the directory before the fsync
2207 * and should be removed.
2209 static noinline int find_dir_range(struct btrfs_root *root,
2210 struct btrfs_path *path,
2212 u64 *start_ret, u64 *end_ret)
2214 struct btrfs_key key;
2216 struct btrfs_dir_log_item *item;
2220 if (*start_ret == (u64)-1)
2223 key.objectid = dirid;
2224 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2225 key.offset = *start_ret;
2227 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2231 if (path->slots[0] == 0)
2236 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2238 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2242 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2243 struct btrfs_dir_log_item);
2244 found_end = btrfs_dir_log_end(path->nodes[0], item);
2246 if (*start_ret >= key.offset && *start_ret <= found_end) {
2248 *start_ret = key.offset;
2249 *end_ret = found_end;
2254 /* check the next slot in the tree to see if it is a valid item */
2255 nritems = btrfs_header_nritems(path->nodes[0]);
2257 if (path->slots[0] >= nritems) {
2258 ret = btrfs_next_leaf(root, path);
2263 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2265 if (key.type != BTRFS_DIR_LOG_INDEX_KEY || key.objectid != dirid) {
2269 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2270 struct btrfs_dir_log_item);
2271 found_end = btrfs_dir_log_end(path->nodes[0], item);
2272 *start_ret = key.offset;
2273 *end_ret = found_end;
2276 btrfs_release_path(path);
2281 * this looks for a given directory item in the log. If the directory
2282 * item is not in the log, the item is removed and the inode it points
2285 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2286 struct btrfs_root *log,
2287 struct btrfs_path *path,
2288 struct btrfs_path *log_path,
2290 struct btrfs_key *dir_key)
2292 struct btrfs_root *root = BTRFS_I(dir)->root;
2294 struct extent_buffer *eb;
2296 struct btrfs_dir_item *di;
2299 struct inode *inode = NULL;
2300 struct btrfs_key location;
2303 * Currenly we only log dir index keys. Even if we replay a log created
2304 * by an older kernel that logged both dir index and dir item keys, all
2305 * we need to do is process the dir index keys, we (and our caller) can
2306 * safely ignore dir item keys (key type BTRFS_DIR_ITEM_KEY).
2308 ASSERT(dir_key->type == BTRFS_DIR_INDEX_KEY);
2310 eb = path->nodes[0];
2311 slot = path->slots[0];
2312 di = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
2313 name_len = btrfs_dir_name_len(eb, di);
2314 name = kmalloc(name_len, GFP_NOFS);
2320 read_extent_buffer(eb, name, (unsigned long)(di + 1), name_len);
2323 struct btrfs_dir_item *log_di;
2325 log_di = btrfs_lookup_dir_index_item(trans, log, log_path,
2329 if (IS_ERR(log_di)) {
2330 ret = PTR_ERR(log_di);
2332 } else if (log_di) {
2333 /* The dentry exists in the log, we have nothing to do. */
2339 btrfs_dir_item_key_to_cpu(eb, di, &location);
2340 btrfs_release_path(path);
2341 btrfs_release_path(log_path);
2342 inode = read_one_inode(root, location.objectid);
2348 ret = link_to_fixup_dir(trans, root, path, location.objectid);
2353 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(inode), name,
2358 ret = btrfs_run_delayed_items(trans);
2363 * Unlike dir item keys, dir index keys can only have one name (entry) in
2364 * them, as there are no key collisions since each key has a unique offset
2365 * (an index number), so we're done.
2368 btrfs_release_path(path);
2369 btrfs_release_path(log_path);
2375 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *root,
2377 struct btrfs_root *log,
2378 struct btrfs_path *path,
2381 struct btrfs_key search_key;
2382 struct btrfs_path *log_path;
2387 log_path = btrfs_alloc_path();
2391 search_key.objectid = ino;
2392 search_key.type = BTRFS_XATTR_ITEM_KEY;
2393 search_key.offset = 0;
2395 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2399 nritems = btrfs_header_nritems(path->nodes[0]);
2400 for (i = path->slots[0]; i < nritems; i++) {
2401 struct btrfs_key key;
2402 struct btrfs_dir_item *di;
2403 struct btrfs_dir_item *log_di;
2407 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2408 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2413 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2414 total_size = btrfs_item_size(path->nodes[0], i);
2416 while (cur < total_size) {
2417 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2418 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2419 u32 this_len = sizeof(*di) + name_len + data_len;
2422 name = kmalloc(name_len, GFP_NOFS);
2427 read_extent_buffer(path->nodes[0], name,
2428 (unsigned long)(di + 1), name_len);
2430 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2432 btrfs_release_path(log_path);
2434 /* Doesn't exist in log tree, so delete it. */
2435 btrfs_release_path(path);
2436 di = btrfs_lookup_xattr(trans, root, path, ino,
2437 name, name_len, -1);
2444 ret = btrfs_delete_one_dir_name(trans, root,
2448 btrfs_release_path(path);
2453 if (IS_ERR(log_di)) {
2454 ret = PTR_ERR(log_di);
2458 di = (struct btrfs_dir_item *)((char *)di + this_len);
2461 ret = btrfs_next_leaf(root, path);
2467 btrfs_free_path(log_path);
2468 btrfs_release_path(path);
2474 * deletion replay happens before we copy any new directory items
2475 * out of the log or out of backreferences from inodes. It
2476 * scans the log to find ranges of keys that log is authoritative for,
2477 * and then scans the directory to find items in those ranges that are
2478 * not present in the log.
2480 * Anything we don't find in the log is unlinked and removed from the
2483 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2484 struct btrfs_root *root,
2485 struct btrfs_root *log,
2486 struct btrfs_path *path,
2487 u64 dirid, int del_all)
2492 struct btrfs_key dir_key;
2493 struct btrfs_key found_key;
2494 struct btrfs_path *log_path;
2497 dir_key.objectid = dirid;
2498 dir_key.type = BTRFS_DIR_INDEX_KEY;
2499 log_path = btrfs_alloc_path();
2503 dir = read_one_inode(root, dirid);
2504 /* it isn't an error if the inode isn't there, that can happen
2505 * because we replay the deletes before we copy in the inode item
2509 btrfs_free_path(log_path);
2517 range_end = (u64)-1;
2519 ret = find_dir_range(log, path, dirid,
2520 &range_start, &range_end);
2527 dir_key.offset = range_start;
2530 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2535 nritems = btrfs_header_nritems(path->nodes[0]);
2536 if (path->slots[0] >= nritems) {
2537 ret = btrfs_next_leaf(root, path);
2543 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2545 if (found_key.objectid != dirid ||
2546 found_key.type != dir_key.type) {
2551 if (found_key.offset > range_end)
2554 ret = check_item_in_log(trans, log, path,
2559 if (found_key.offset == (u64)-1)
2561 dir_key.offset = found_key.offset + 1;
2563 btrfs_release_path(path);
2564 if (range_end == (u64)-1)
2566 range_start = range_end + 1;
2570 btrfs_release_path(path);
2571 btrfs_free_path(log_path);
2577 * the process_func used to replay items from the log tree. This
2578 * gets called in two different stages. The first stage just looks
2579 * for inodes and makes sure they are all copied into the subvolume.
2581 * The second stage copies all the other item types from the log into
2582 * the subvolume. The two stage approach is slower, but gets rid of
2583 * lots of complexity around inodes referencing other inodes that exist
2584 * only in the log (references come from either directory items or inode
2587 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2588 struct walk_control *wc, u64 gen, int level)
2591 struct btrfs_path *path;
2592 struct btrfs_root *root = wc->replay_dest;
2593 struct btrfs_key key;
2597 ret = btrfs_read_buffer(eb, gen, level, NULL);
2601 level = btrfs_header_level(eb);
2606 path = btrfs_alloc_path();
2610 nritems = btrfs_header_nritems(eb);
2611 for (i = 0; i < nritems; i++) {
2612 btrfs_item_key_to_cpu(eb, &key, i);
2614 /* inode keys are done during the first stage */
2615 if (key.type == BTRFS_INODE_ITEM_KEY &&
2616 wc->stage == LOG_WALK_REPLAY_INODES) {
2617 struct btrfs_inode_item *inode_item;
2620 inode_item = btrfs_item_ptr(eb, i,
2621 struct btrfs_inode_item);
2623 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2624 * and never got linked before the fsync, skip it, as
2625 * replaying it is pointless since it would be deleted
2626 * later. We skip logging tmpfiles, but it's always
2627 * possible we are replaying a log created with a kernel
2628 * that used to log tmpfiles.
2630 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2631 wc->ignore_cur_inode = true;
2634 wc->ignore_cur_inode = false;
2636 ret = replay_xattr_deletes(wc->trans, root, log,
2637 path, key.objectid);
2640 mode = btrfs_inode_mode(eb, inode_item);
2641 if (S_ISDIR(mode)) {
2642 ret = replay_dir_deletes(wc->trans,
2643 root, log, path, key.objectid, 0);
2647 ret = overwrite_item(wc->trans, root, path,
2653 * Before replaying extents, truncate the inode to its
2654 * size. We need to do it now and not after log replay
2655 * because before an fsync we can have prealloc extents
2656 * added beyond the inode's i_size. If we did it after,
2657 * through orphan cleanup for example, we would drop
2658 * those prealloc extents just after replaying them.
2660 if (S_ISREG(mode)) {
2661 struct btrfs_drop_extents_args drop_args = { 0 };
2662 struct inode *inode;
2665 inode = read_one_inode(root, key.objectid);
2670 from = ALIGN(i_size_read(inode),
2671 root->fs_info->sectorsize);
2672 drop_args.start = from;
2673 drop_args.end = (u64)-1;
2674 drop_args.drop_cache = true;
2675 ret = btrfs_drop_extents(wc->trans, root,
2679 inode_sub_bytes(inode,
2680 drop_args.bytes_found);
2681 /* Update the inode's nbytes. */
2682 ret = btrfs_update_inode(wc->trans,
2683 root, BTRFS_I(inode));
2690 ret = link_to_fixup_dir(wc->trans, root,
2691 path, key.objectid);
2696 if (wc->ignore_cur_inode)
2699 if (key.type == BTRFS_DIR_INDEX_KEY &&
2700 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2701 ret = replay_one_dir_item(wc->trans, root, path,
2707 if (wc->stage < LOG_WALK_REPLAY_ALL)
2710 /* these keys are simply copied */
2711 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2712 ret = overwrite_item(wc->trans, root, path,
2716 } else if (key.type == BTRFS_INODE_REF_KEY ||
2717 key.type == BTRFS_INODE_EXTREF_KEY) {
2718 ret = add_inode_ref(wc->trans, root, log, path,
2720 if (ret && ret != -ENOENT)
2723 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2724 ret = replay_one_extent(wc->trans, root, path,
2730 * We don't log BTRFS_DIR_ITEM_KEY keys anymore, only the
2731 * BTRFS_DIR_INDEX_KEY items which we use to derive the
2732 * BTRFS_DIR_ITEM_KEY items. If we are replaying a log from an
2733 * older kernel with such keys, ignore them.
2736 btrfs_free_path(path);
2741 * Correctly adjust the reserved bytes occupied by a log tree extent buffer
2743 static void unaccount_log_buffer(struct btrfs_fs_info *fs_info, u64 start)
2745 struct btrfs_block_group *cache;
2747 cache = btrfs_lookup_block_group(fs_info, start);
2749 btrfs_err(fs_info, "unable to find block group for %llu", start);
2753 spin_lock(&cache->space_info->lock);
2754 spin_lock(&cache->lock);
2755 cache->reserved -= fs_info->nodesize;
2756 cache->space_info->bytes_reserved -= fs_info->nodesize;
2757 spin_unlock(&cache->lock);
2758 spin_unlock(&cache->space_info->lock);
2760 btrfs_put_block_group(cache);
2763 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 struct btrfs_path *path, int *level,
2766 struct walk_control *wc)
2768 struct btrfs_fs_info *fs_info = root->fs_info;
2771 struct extent_buffer *next;
2772 struct extent_buffer *cur;
2776 while (*level > 0) {
2777 struct btrfs_key first_key;
2779 cur = path->nodes[*level];
2781 WARN_ON(btrfs_header_level(cur) != *level);
2783 if (path->slots[*level] >=
2784 btrfs_header_nritems(cur))
2787 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2788 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2789 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2790 blocksize = fs_info->nodesize;
2792 next = btrfs_find_create_tree_block(fs_info, bytenr,
2793 btrfs_header_owner(cur),
2796 return PTR_ERR(next);
2799 ret = wc->process_func(root, next, wc, ptr_gen,
2802 free_extent_buffer(next);
2806 path->slots[*level]++;
2808 ret = btrfs_read_buffer(next, ptr_gen,
2809 *level - 1, &first_key);
2811 free_extent_buffer(next);
2816 btrfs_tree_lock(next);
2817 btrfs_clean_tree_block(next);
2818 btrfs_wait_tree_block_writeback(next);
2819 btrfs_tree_unlock(next);
2820 ret = btrfs_pin_reserved_extent(trans,
2823 free_extent_buffer(next);
2826 btrfs_redirty_list_add(
2827 trans->transaction, next);
2829 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2830 clear_extent_buffer_dirty(next);
2831 unaccount_log_buffer(fs_info, bytenr);
2834 free_extent_buffer(next);
2837 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2839 free_extent_buffer(next);
2843 if (path->nodes[*level-1])
2844 free_extent_buffer(path->nodes[*level-1]);
2845 path->nodes[*level-1] = next;
2846 *level = btrfs_header_level(next);
2847 path->slots[*level] = 0;
2850 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2856 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2857 struct btrfs_root *root,
2858 struct btrfs_path *path, int *level,
2859 struct walk_control *wc)
2861 struct btrfs_fs_info *fs_info = root->fs_info;
2866 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2867 slot = path->slots[i];
2868 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2871 WARN_ON(*level == 0);
2874 ret = wc->process_func(root, path->nodes[*level], wc,
2875 btrfs_header_generation(path->nodes[*level]),
2881 struct extent_buffer *next;
2883 next = path->nodes[*level];
2886 btrfs_tree_lock(next);
2887 btrfs_clean_tree_block(next);
2888 btrfs_wait_tree_block_writeback(next);
2889 btrfs_tree_unlock(next);
2890 ret = btrfs_pin_reserved_extent(trans,
2891 path->nodes[*level]->start,
2892 path->nodes[*level]->len);
2895 btrfs_redirty_list_add(trans->transaction,
2898 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2899 clear_extent_buffer_dirty(next);
2901 unaccount_log_buffer(fs_info,
2902 path->nodes[*level]->start);
2905 free_extent_buffer(path->nodes[*level]);
2906 path->nodes[*level] = NULL;
2914 * drop the reference count on the tree rooted at 'snap'. This traverses
2915 * the tree freeing any blocks that have a ref count of zero after being
2918 static int walk_log_tree(struct btrfs_trans_handle *trans,
2919 struct btrfs_root *log, struct walk_control *wc)
2921 struct btrfs_fs_info *fs_info = log->fs_info;
2925 struct btrfs_path *path;
2928 path = btrfs_alloc_path();
2932 level = btrfs_header_level(log->node);
2934 path->nodes[level] = log->node;
2935 atomic_inc(&log->node->refs);
2936 path->slots[level] = 0;
2939 wret = walk_down_log_tree(trans, log, path, &level, wc);
2947 wret = walk_up_log_tree(trans, log, path, &level, wc);
2956 /* was the root node processed? if not, catch it here */
2957 if (path->nodes[orig_level]) {
2958 ret = wc->process_func(log, path->nodes[orig_level], wc,
2959 btrfs_header_generation(path->nodes[orig_level]),
2964 struct extent_buffer *next;
2966 next = path->nodes[orig_level];
2969 btrfs_tree_lock(next);
2970 btrfs_clean_tree_block(next);
2971 btrfs_wait_tree_block_writeback(next);
2972 btrfs_tree_unlock(next);
2973 ret = btrfs_pin_reserved_extent(trans,
2974 next->start, next->len);
2977 btrfs_redirty_list_add(trans->transaction, next);
2979 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2980 clear_extent_buffer_dirty(next);
2981 unaccount_log_buffer(fs_info, next->start);
2987 btrfs_free_path(path);
2992 * helper function to update the item for a given subvolumes log root
2993 * in the tree of log roots
2995 static int update_log_root(struct btrfs_trans_handle *trans,
2996 struct btrfs_root *log,
2997 struct btrfs_root_item *root_item)
2999 struct btrfs_fs_info *fs_info = log->fs_info;
3002 if (log->log_transid == 1) {
3003 /* insert root item on the first sync */
3004 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
3005 &log->root_key, root_item);
3007 ret = btrfs_update_root(trans, fs_info->log_root_tree,
3008 &log->root_key, root_item);
3013 static void wait_log_commit(struct btrfs_root *root, int transid)
3016 int index = transid % 2;
3019 * we only allow two pending log transactions at a time,
3020 * so we know that if ours is more than 2 older than the
3021 * current transaction, we're done
3024 prepare_to_wait(&root->log_commit_wait[index],
3025 &wait, TASK_UNINTERRUPTIBLE);
3027 if (!(root->log_transid_committed < transid &&
3028 atomic_read(&root->log_commit[index])))
3031 mutex_unlock(&root->log_mutex);
3033 mutex_lock(&root->log_mutex);
3035 finish_wait(&root->log_commit_wait[index], &wait);
3038 static void wait_for_writer(struct btrfs_root *root)
3043 prepare_to_wait(&root->log_writer_wait, &wait,
3044 TASK_UNINTERRUPTIBLE);
3045 if (!atomic_read(&root->log_writers))
3048 mutex_unlock(&root->log_mutex);
3050 mutex_lock(&root->log_mutex);
3052 finish_wait(&root->log_writer_wait, &wait);
3055 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
3056 struct btrfs_log_ctx *ctx)
3058 mutex_lock(&root->log_mutex);
3059 list_del_init(&ctx->list);
3060 mutex_unlock(&root->log_mutex);
3064 * Invoked in log mutex context, or be sure there is no other task which
3065 * can access the list.
3067 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3068 int index, int error)
3070 struct btrfs_log_ctx *ctx;
3071 struct btrfs_log_ctx *safe;
3073 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3074 list_del_init(&ctx->list);
3075 ctx->log_ret = error;
3080 * btrfs_sync_log does sends a given tree log down to the disk and
3081 * updates the super blocks to record it. When this call is done,
3082 * you know that any inodes previously logged are safely on disk only
3085 * Any other return value means you need to call btrfs_commit_transaction.
3086 * Some of the edge cases for fsyncing directories that have had unlinks
3087 * or renames done in the past mean that sometimes the only safe
3088 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3089 * that has happened.
3091 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3092 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3098 struct btrfs_fs_info *fs_info = root->fs_info;
3099 struct btrfs_root *log = root->log_root;
3100 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3101 struct btrfs_root_item new_root_item;
3102 int log_transid = 0;
3103 struct btrfs_log_ctx root_log_ctx;
3104 struct blk_plug plug;
3108 mutex_lock(&root->log_mutex);
3109 log_transid = ctx->log_transid;
3110 if (root->log_transid_committed >= log_transid) {
3111 mutex_unlock(&root->log_mutex);
3112 return ctx->log_ret;
3115 index1 = log_transid % 2;
3116 if (atomic_read(&root->log_commit[index1])) {
3117 wait_log_commit(root, log_transid);
3118 mutex_unlock(&root->log_mutex);
3119 return ctx->log_ret;
3121 ASSERT(log_transid == root->log_transid);
3122 atomic_set(&root->log_commit[index1], 1);
3124 /* wait for previous tree log sync to complete */
3125 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3126 wait_log_commit(root, log_transid - 1);
3129 int batch = atomic_read(&root->log_batch);
3130 /* when we're on an ssd, just kick the log commit out */
3131 if (!btrfs_test_opt(fs_info, SSD) &&
3132 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3133 mutex_unlock(&root->log_mutex);
3134 schedule_timeout_uninterruptible(1);
3135 mutex_lock(&root->log_mutex);
3137 wait_for_writer(root);
3138 if (batch == atomic_read(&root->log_batch))
3142 /* bail out if we need to do a full commit */
3143 if (btrfs_need_log_full_commit(trans)) {
3145 mutex_unlock(&root->log_mutex);
3149 if (log_transid % 2 == 0)
3150 mark = EXTENT_DIRTY;
3154 /* we start IO on all the marked extents here, but we don't actually
3155 * wait for them until later.
3157 blk_start_plug(&plug);
3158 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3160 * -EAGAIN happens when someone, e.g., a concurrent transaction
3161 * commit, writes a dirty extent in this tree-log commit. This
3162 * concurrent write will create a hole writing out the extents,
3163 * and we cannot proceed on a zoned filesystem, requiring
3164 * sequential writing. While we can bail out to a full commit
3165 * here, but we can continue hoping the concurrent writing fills
3168 if (ret == -EAGAIN && btrfs_is_zoned(fs_info))
3171 blk_finish_plug(&plug);
3172 btrfs_abort_transaction(trans, ret);
3173 btrfs_set_log_full_commit(trans);
3174 mutex_unlock(&root->log_mutex);
3179 * We _must_ update under the root->log_mutex in order to make sure we
3180 * have a consistent view of the log root we are trying to commit at
3183 * We _must_ copy this into a local copy, because we are not holding the
3184 * log_root_tree->log_mutex yet. This is important because when we
3185 * commit the log_root_tree we must have a consistent view of the
3186 * log_root_tree when we update the super block to point at the
3187 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3188 * with the commit and possibly point at the new block which we may not
3191 btrfs_set_root_node(&log->root_item, log->node);
3192 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3194 root->log_transid++;
3195 log->log_transid = root->log_transid;
3196 root->log_start_pid = 0;
3198 * IO has been started, blocks of the log tree have WRITTEN flag set
3199 * in their headers. new modifications of the log will be written to
3200 * new positions. so it's safe to allow log writers to go in.
3202 mutex_unlock(&root->log_mutex);
3204 if (btrfs_is_zoned(fs_info)) {
3205 mutex_lock(&fs_info->tree_root->log_mutex);
3206 if (!log_root_tree->node) {
3207 ret = btrfs_alloc_log_tree_node(trans, log_root_tree);
3209 mutex_unlock(&fs_info->tree_root->log_mutex);
3213 mutex_unlock(&fs_info->tree_root->log_mutex);
3216 btrfs_init_log_ctx(&root_log_ctx, NULL);
3218 mutex_lock(&log_root_tree->log_mutex);
3220 index2 = log_root_tree->log_transid % 2;
3221 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3222 root_log_ctx.log_transid = log_root_tree->log_transid;
3225 * Now we are safe to update the log_root_tree because we're under the
3226 * log_mutex, and we're a current writer so we're holding the commit
3227 * open until we drop the log_mutex.
3229 ret = update_log_root(trans, log, &new_root_item);
3231 if (!list_empty(&root_log_ctx.list))
3232 list_del_init(&root_log_ctx.list);
3234 blk_finish_plug(&plug);
3235 btrfs_set_log_full_commit(trans);
3237 if (ret != -ENOSPC) {
3238 btrfs_abort_transaction(trans, ret);
3239 mutex_unlock(&log_root_tree->log_mutex);
3242 btrfs_wait_tree_log_extents(log, mark);
3243 mutex_unlock(&log_root_tree->log_mutex);
3248 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3249 blk_finish_plug(&plug);
3250 list_del_init(&root_log_ctx.list);
3251 mutex_unlock(&log_root_tree->log_mutex);
3252 ret = root_log_ctx.log_ret;
3256 index2 = root_log_ctx.log_transid % 2;
3257 if (atomic_read(&log_root_tree->log_commit[index2])) {
3258 blk_finish_plug(&plug);
3259 ret = btrfs_wait_tree_log_extents(log, mark);
3260 wait_log_commit(log_root_tree,
3261 root_log_ctx.log_transid);
3262 mutex_unlock(&log_root_tree->log_mutex);
3264 ret = root_log_ctx.log_ret;
3267 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3268 atomic_set(&log_root_tree->log_commit[index2], 1);
3270 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3271 wait_log_commit(log_root_tree,
3272 root_log_ctx.log_transid - 1);
3276 * now that we've moved on to the tree of log tree roots,
3277 * check the full commit flag again
3279 if (btrfs_need_log_full_commit(trans)) {
3280 blk_finish_plug(&plug);
3281 btrfs_wait_tree_log_extents(log, mark);
3282 mutex_unlock(&log_root_tree->log_mutex);
3284 goto out_wake_log_root;
3287 ret = btrfs_write_marked_extents(fs_info,
3288 &log_root_tree->dirty_log_pages,
3289 EXTENT_DIRTY | EXTENT_NEW);
3290 blk_finish_plug(&plug);
3292 * As described above, -EAGAIN indicates a hole in the extents. We
3293 * cannot wait for these write outs since the waiting cause a
3294 * deadlock. Bail out to the full commit instead.
3296 if (ret == -EAGAIN && btrfs_is_zoned(fs_info)) {
3297 btrfs_set_log_full_commit(trans);
3298 btrfs_wait_tree_log_extents(log, mark);
3299 mutex_unlock(&log_root_tree->log_mutex);
3300 goto out_wake_log_root;
3302 btrfs_set_log_full_commit(trans);
3303 btrfs_abort_transaction(trans, ret);
3304 mutex_unlock(&log_root_tree->log_mutex);
3305 goto out_wake_log_root;
3307 ret = btrfs_wait_tree_log_extents(log, mark);
3309 ret = btrfs_wait_tree_log_extents(log_root_tree,
3310 EXTENT_NEW | EXTENT_DIRTY);
3312 btrfs_set_log_full_commit(trans);
3313 mutex_unlock(&log_root_tree->log_mutex);
3314 goto out_wake_log_root;
3317 log_root_start = log_root_tree->node->start;
3318 log_root_level = btrfs_header_level(log_root_tree->node);
3319 log_root_tree->log_transid++;
3320 mutex_unlock(&log_root_tree->log_mutex);
3323 * Here we are guaranteed that nobody is going to write the superblock
3324 * for the current transaction before us and that neither we do write
3325 * our superblock before the previous transaction finishes its commit
3326 * and writes its superblock, because:
3328 * 1) We are holding a handle on the current transaction, so no body
3329 * can commit it until we release the handle;
3331 * 2) Before writing our superblock we acquire the tree_log_mutex, so
3332 * if the previous transaction is still committing, and hasn't yet
3333 * written its superblock, we wait for it to do it, because a
3334 * transaction commit acquires the tree_log_mutex when the commit
3335 * begins and releases it only after writing its superblock.
3337 mutex_lock(&fs_info->tree_log_mutex);
3340 * The previous transaction writeout phase could have failed, and thus
3341 * marked the fs in an error state. We must not commit here, as we
3342 * could have updated our generation in the super_for_commit and
3343 * writing the super here would result in transid mismatches. If there
3344 * is an error here just bail.
3346 if (BTRFS_FS_ERROR(fs_info)) {
3348 btrfs_set_log_full_commit(trans);
3349 btrfs_abort_transaction(trans, ret);
3350 mutex_unlock(&fs_info->tree_log_mutex);
3351 goto out_wake_log_root;
3354 btrfs_set_super_log_root(fs_info->super_for_commit, log_root_start);
3355 btrfs_set_super_log_root_level(fs_info->super_for_commit, log_root_level);
3356 ret = write_all_supers(fs_info, 1);
3357 mutex_unlock(&fs_info->tree_log_mutex);
3359 btrfs_set_log_full_commit(trans);
3360 btrfs_abort_transaction(trans, ret);
3361 goto out_wake_log_root;
3365 * We know there can only be one task here, since we have not yet set
3366 * root->log_commit[index1] to 0 and any task attempting to sync the
3367 * log must wait for the previous log transaction to commit if it's
3368 * still in progress or wait for the current log transaction commit if
3369 * someone else already started it. We use <= and not < because the
3370 * first log transaction has an ID of 0.
3372 ASSERT(root->last_log_commit <= log_transid);
3373 root->last_log_commit = log_transid;
3376 mutex_lock(&log_root_tree->log_mutex);
3377 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3379 log_root_tree->log_transid_committed++;
3380 atomic_set(&log_root_tree->log_commit[index2], 0);
3381 mutex_unlock(&log_root_tree->log_mutex);
3384 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3385 * all the updates above are seen by the woken threads. It might not be
3386 * necessary, but proving that seems to be hard.
3388 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3390 mutex_lock(&root->log_mutex);
3391 btrfs_remove_all_log_ctxs(root, index1, ret);
3392 root->log_transid_committed++;
3393 atomic_set(&root->log_commit[index1], 0);
3394 mutex_unlock(&root->log_mutex);
3397 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3398 * all the updates above are seen by the woken threads. It might not be
3399 * necessary, but proving that seems to be hard.
3401 cond_wake_up(&root->log_commit_wait[index1]);
3405 static void free_log_tree(struct btrfs_trans_handle *trans,
3406 struct btrfs_root *log)
3409 struct walk_control wc = {
3411 .process_func = process_one_buffer
3415 ret = walk_log_tree(trans, log, &wc);
3418 * We weren't able to traverse the entire log tree, the
3419 * typical scenario is getting an -EIO when reading an
3420 * extent buffer of the tree, due to a previous writeback
3423 set_bit(BTRFS_FS_STATE_LOG_CLEANUP_ERROR,
3424 &log->fs_info->fs_state);
3427 * Some extent buffers of the log tree may still be dirty
3428 * and not yet written back to storage, because we may
3429 * have updates to a log tree without syncing a log tree,
3430 * such as during rename and link operations. So flush
3431 * them out and wait for their writeback to complete, so
3432 * that we properly cleanup their state and pages.
3434 btrfs_write_marked_extents(log->fs_info,
3435 &log->dirty_log_pages,
3436 EXTENT_DIRTY | EXTENT_NEW);
3437 btrfs_wait_tree_log_extents(log,
3438 EXTENT_DIRTY | EXTENT_NEW);
3441 btrfs_abort_transaction(trans, ret);
3443 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3447 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3448 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3449 extent_io_tree_release(&log->log_csum_range);
3451 btrfs_put_root(log);
3455 * free all the extents used by the tree log. This should be called
3456 * at commit time of the full transaction
3458 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3460 if (root->log_root) {
3461 free_log_tree(trans, root->log_root);
3462 root->log_root = NULL;
3463 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &root->state);
3468 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3469 struct btrfs_fs_info *fs_info)
3471 if (fs_info->log_root_tree) {
3472 free_log_tree(trans, fs_info->log_root_tree);
3473 fs_info->log_root_tree = NULL;
3474 clear_bit(BTRFS_ROOT_HAS_LOG_TREE, &fs_info->tree_root->state);
3480 * Check if an inode was logged in the current transaction. This may often
3481 * return some false positives, because logged_trans is an in memory only field,
3482 * not persisted anywhere. This is meant to be used in contexts where a false
3483 * positive has no functional consequences.
3485 static bool inode_logged(struct btrfs_trans_handle *trans,
3486 struct btrfs_inode *inode)
3488 if (inode->logged_trans == trans->transid)
3491 if (!test_bit(BTRFS_ROOT_HAS_LOG_TREE, &inode->root->state))
3495 * The inode's logged_trans is always 0 when we load it (because it is
3496 * not persisted in the inode item or elsewhere). So if it is 0, the
3497 * inode was last modified in the current transaction then the inode may
3498 * have been logged before in the current transaction, then evicted and
3499 * loaded again in the current transaction - or may have never been logged
3500 * in the current transaction, but since we can not be sure, we have to
3501 * assume it was, otherwise our callers can leave an inconsistent log.
3503 if (inode->logged_trans == 0 &&
3504 inode->last_trans == trans->transid &&
3505 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3512 * If both a file and directory are logged, and unlinks or renames are
3513 * mixed in, we have a few interesting corners:
3515 * create file X in dir Y
3516 * link file X to X.link in dir Y
3518 * unlink file X but leave X.link
3521 * After a crash we would expect only X.link to exist. But file X
3522 * didn't get fsync'd again so the log has back refs for X and X.link.
3524 * We solve this by removing directory entries and inode backrefs from the
3525 * log when a file that was logged in the current transaction is
3526 * unlinked. Any later fsync will include the updated log entries, and
3527 * we'll be able to reconstruct the proper directory items from backrefs.
3529 * This optimizations allows us to avoid relogging the entire inode
3530 * or the entire directory.
3532 void btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3533 struct btrfs_root *root,
3534 const char *name, int name_len,
3535 struct btrfs_inode *dir, u64 index)
3537 struct btrfs_root *log;
3538 struct btrfs_dir_item *di;
3539 struct btrfs_path *path;
3542 u64 dir_ino = btrfs_ino(dir);
3544 if (!inode_logged(trans, dir))
3547 ret = join_running_log_trans(root);
3551 mutex_lock(&dir->log_mutex);
3553 log = root->log_root;
3554 path = btrfs_alloc_path();
3561 * We only log dir index items of a directory, so we don't need to look
3562 * for dir item keys.
3564 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3565 index, name, name_len, -1);
3571 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3579 * We do not need to update the size field of the directory's inode item
3580 * because on log replay we update the field to reflect all existing
3581 * entries in the directory (see overwrite_item()).
3584 btrfs_free_path(path);
3586 mutex_unlock(&dir->log_mutex);
3588 btrfs_set_log_full_commit(trans);
3589 btrfs_end_log_trans(root);
3592 /* see comments for btrfs_del_dir_entries_in_log */
3593 void btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3594 struct btrfs_root *root,
3595 const char *name, int name_len,
3596 struct btrfs_inode *inode, u64 dirid)
3598 struct btrfs_root *log;
3602 if (!inode_logged(trans, inode))
3605 ret = join_running_log_trans(root);
3608 log = root->log_root;
3609 mutex_lock(&inode->log_mutex);
3611 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3613 mutex_unlock(&inode->log_mutex);
3614 if (ret < 0 && ret != -ENOENT)
3615 btrfs_set_log_full_commit(trans);
3616 btrfs_end_log_trans(root);
3620 * creates a range item in the log for 'dirid'. first_offset and
3621 * last_offset tell us which parts of the key space the log should
3622 * be considered authoritative for.
3624 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3625 struct btrfs_root *log,
3626 struct btrfs_path *path,
3628 u64 first_offset, u64 last_offset)
3631 struct btrfs_key key;
3632 struct btrfs_dir_log_item *item;
3634 key.objectid = dirid;
3635 key.offset = first_offset;
3636 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3637 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3641 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3642 struct btrfs_dir_log_item);
3643 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3644 btrfs_mark_buffer_dirty(path->nodes[0]);
3645 btrfs_release_path(path);
3649 static int flush_dir_items_batch(struct btrfs_trans_handle *trans,
3650 struct btrfs_root *log,
3651 struct extent_buffer *src,
3652 struct btrfs_path *dst_path,
3656 char *ins_data = NULL;
3657 struct btrfs_item_batch batch;
3658 struct extent_buffer *dst;
3659 unsigned long src_offset;
3660 unsigned long dst_offset;
3661 struct btrfs_key key;
3670 btrfs_item_key_to_cpu(src, &key, start_slot);
3671 item_size = btrfs_item_size(src, start_slot);
3673 batch.data_sizes = &item_size;
3674 batch.total_data_size = item_size;
3676 struct btrfs_key *ins_keys;
3679 ins_data = kmalloc(count * sizeof(u32) +
3680 count * sizeof(struct btrfs_key), GFP_NOFS);
3684 ins_sizes = (u32 *)ins_data;
3685 ins_keys = (struct btrfs_key *)(ins_data + count * sizeof(u32));
3686 batch.keys = ins_keys;
3687 batch.data_sizes = ins_sizes;
3688 batch.total_data_size = 0;
3690 for (i = 0; i < count; i++) {
3691 const int slot = start_slot + i;
3693 btrfs_item_key_to_cpu(src, &ins_keys[i], slot);
3694 ins_sizes[i] = btrfs_item_size(src, slot);
3695 batch.total_data_size += ins_sizes[i];
3699 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
3703 dst = dst_path->nodes[0];
3705 * Copy all the items in bulk, in a single copy operation. Item data is
3706 * organized such that it's placed at the end of a leaf and from right
3707 * to left. For example, the data for the second item ends at an offset
3708 * that matches the offset where the data for the first item starts, the
3709 * data for the third item ends at an offset that matches the offset
3710 * where the data of the second items starts, and so on.
3711 * Therefore our source and destination start offsets for copy match the
3712 * offsets of the last items (highest slots).
3714 dst_offset = btrfs_item_ptr_offset(dst, dst_path->slots[0] + count - 1);
3715 src_offset = btrfs_item_ptr_offset(src, start_slot + count - 1);
3716 copy_extent_buffer(dst, src, dst_offset, src_offset, batch.total_data_size);
3717 btrfs_release_path(dst_path);
3724 static int process_dir_items_leaf(struct btrfs_trans_handle *trans,
3725 struct btrfs_inode *inode,
3726 struct btrfs_path *path,
3727 struct btrfs_path *dst_path,
3728 struct btrfs_log_ctx *ctx)
3730 struct btrfs_root *log = inode->root->log_root;
3731 struct extent_buffer *src = path->nodes[0];
3732 const int nritems = btrfs_header_nritems(src);
3733 const u64 ino = btrfs_ino(inode);
3734 const bool inode_logged_before = inode_logged(trans, inode);
3735 bool last_found = false;
3736 int batch_start = 0;
3740 for (i = path->slots[0]; i < nritems; i++) {
3741 struct btrfs_key key;
3744 btrfs_item_key_to_cpu(src, &key, i);
3746 if (key.objectid != ino || key.type != BTRFS_DIR_INDEX_KEY) {
3751 ctx->last_dir_item_offset = key.offset;
3753 * We must make sure that when we log a directory entry, the
3754 * corresponding inode, after log replay, has a matching link
3755 * count. For example:
3761 * xfs_io -c "fsync" mydir
3763 * <mount fs and log replay>
3765 * Would result in a fsync log that when replayed, our file inode
3766 * would have a link count of 1, but we get two directory entries
3767 * pointing to the same inode. After removing one of the names,
3768 * it would not be possible to remove the other name, which
3769 * resulted always in stale file handle errors, and would not be
3770 * possible to rmdir the parent directory, since its i_size could
3771 * never be decremented to the value BTRFS_EMPTY_DIR_SIZE,
3772 * resulting in -ENOTEMPTY errors.
3774 if (!ctx->log_new_dentries) {
3775 struct btrfs_dir_item *di;
3776 struct btrfs_key di_key;
3778 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3779 btrfs_dir_item_key_to_cpu(src, di, &di_key);
3780 if ((btrfs_dir_transid(src, di) == trans->transid ||
3781 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3782 di_key.type != BTRFS_ROOT_ITEM_KEY)
3783 ctx->log_new_dentries = true;
3786 if (!inode_logged_before)
3790 * If we were logged before and have logged dir items, we can skip
3791 * checking if any item with a key offset larger than the last one
3792 * we logged is in the log tree, saving time and avoiding adding
3793 * contention on the log tree.
3795 if (key.offset > inode->last_dir_index_offset)
3798 * Check if the key was already logged before. If not we can add
3799 * it to a batch for bulk insertion.
3801 ret = btrfs_search_slot(NULL, log, &key, dst_path, 0, 0);
3804 } else if (ret > 0) {
3805 btrfs_release_path(dst_path);
3810 * Item exists in the log. Overwrite the item in the log if it
3811 * has different content or do nothing if it has exactly the same
3812 * content. And then flush the current batch if any - do it after
3813 * overwriting the current item, or we would deadlock otherwise,
3814 * since we are holding a path for the existing item.
3816 ret = do_overwrite_item(trans, log, dst_path, src, i, &key);
3820 if (batch_size > 0) {
3821 ret = flush_dir_items_batch(trans, log, src, dst_path,
3822 batch_start, batch_size);
3829 if (batch_size == 0)
3834 if (batch_size > 0) {
3837 ret = flush_dir_items_batch(trans, log, src, dst_path,
3838 batch_start, batch_size);
3843 return last_found ? 1 : 0;
3847 * log all the items included in the current transaction for a given
3848 * directory. This also creates the range items in the log tree required
3849 * to replay anything deleted before the fsync
3851 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3852 struct btrfs_inode *inode,
3853 struct btrfs_path *path,
3854 struct btrfs_path *dst_path,
3855 struct btrfs_log_ctx *ctx,
3856 u64 min_offset, u64 *last_offset_ret)
3858 struct btrfs_key min_key;
3859 struct btrfs_root *root = inode->root;
3860 struct btrfs_root *log = root->log_root;
3863 u64 first_offset = min_offset;
3864 u64 last_offset = (u64)-1;
3865 u64 ino = btrfs_ino(inode);
3867 min_key.objectid = ino;
3868 min_key.type = BTRFS_DIR_INDEX_KEY;
3869 min_key.offset = min_offset;
3871 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3874 * we didn't find anything from this transaction, see if there
3875 * is anything at all
3877 if (ret != 0 || min_key.objectid != ino ||
3878 min_key.type != BTRFS_DIR_INDEX_KEY) {
3879 min_key.objectid = ino;
3880 min_key.type = BTRFS_DIR_INDEX_KEY;
3881 min_key.offset = (u64)-1;
3882 btrfs_release_path(path);
3883 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3885 btrfs_release_path(path);
3888 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3890 /* if ret == 0 there are items for this type,
3891 * create a range to tell us the last key of this type.
3892 * otherwise, there are no items in this directory after
3893 * *min_offset, and we create a range to indicate that.
3896 struct btrfs_key tmp;
3897 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3899 if (tmp.type == BTRFS_DIR_INDEX_KEY)
3900 first_offset = max(min_offset, tmp.offset) + 1;
3905 /* go backward to find any previous key */
3906 ret = btrfs_previous_item(root, path, ino, BTRFS_DIR_INDEX_KEY);
3908 struct btrfs_key tmp;
3909 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3910 if (tmp.type == BTRFS_DIR_INDEX_KEY) {
3911 first_offset = tmp.offset;
3912 ret = overwrite_item(trans, log, dst_path,
3913 path->nodes[0], path->slots[0],
3921 btrfs_release_path(path);
3924 * Find the first key from this transaction again. See the note for
3925 * log_new_dir_dentries, if we're logging a directory recursively we
3926 * won't be holding its i_mutex, which means we can modify the directory
3927 * while we're logging it. If we remove an entry between our first
3928 * search and this search we'll not find the key again and can just
3932 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3937 * we have a block from this transaction, log every item in it
3938 * from our directory
3941 ret = process_dir_items_leaf(trans, inode, path, dst_path, ctx);
3947 path->slots[0] = btrfs_header_nritems(path->nodes[0]);
3950 * look ahead to the next item and see if it is also
3951 * from this directory and from this transaction
3953 ret = btrfs_next_leaf(root, path);
3956 last_offset = (u64)-1;
3961 btrfs_item_key_to_cpu(path->nodes[0], &min_key, path->slots[0]);
3962 if (min_key.objectid != ino || min_key.type != BTRFS_DIR_INDEX_KEY) {
3963 last_offset = (u64)-1;
3966 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3967 ctx->last_dir_item_offset = min_key.offset;
3968 ret = overwrite_item(trans, log, dst_path,
3969 path->nodes[0], path->slots[0],
3974 last_offset = min_key.offset;
3977 if (need_resched()) {
3978 btrfs_release_path(path);
3984 btrfs_release_path(path);
3985 btrfs_release_path(dst_path);
3988 *last_offset_ret = last_offset;
3990 * insert the log range keys to indicate where the log
3993 ret = insert_dir_log_key(trans, log, path, ino, first_offset,
4002 * logging directories is very similar to logging inodes, We find all the items
4003 * from the current transaction and write them to the log.
4005 * The recovery code scans the directory in the subvolume, and if it finds a
4006 * key in the range logged that is not present in the log tree, then it means
4007 * that dir entry was unlinked during the transaction.
4009 * In order for that scan to work, we must include one key smaller than
4010 * the smallest logged by this transaction and one key larger than the largest
4011 * key logged by this transaction.
4013 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
4014 struct btrfs_inode *inode,
4015 struct btrfs_path *path,
4016 struct btrfs_path *dst_path,
4017 struct btrfs_log_ctx *ctx)
4024 * If this is the first time we are being logged in the current
4025 * transaction, or we were logged before but the inode was evicted and
4026 * reloaded later, in which case its logged_trans is 0, reset the value
4027 * of the last logged key offset. Note that we don't use the helper
4028 * function inode_logged() here - that is because the function returns
4029 * true after an inode eviction, assuming the worst case as it can not
4030 * know for sure if the inode was logged before. So we can not skip key
4031 * searches in the case the inode was evicted, because it may not have
4032 * been logged in this transaction and may have been logged in a past
4033 * transaction, so we need to reset the last dir index offset to (u64)-1.
4035 if (inode->logged_trans != trans->transid)
4036 inode->last_dir_index_offset = (u64)-1;
4040 ctx->last_dir_item_offset = inode->last_dir_index_offset;
4043 ret = log_dir_items(trans, inode, path, dst_path,
4044 ctx, min_key, &max_key);
4047 if (max_key == (u64)-1)
4049 min_key = max_key + 1;
4052 inode->last_dir_index_offset = ctx->last_dir_item_offset;
4058 * a helper function to drop items from the log before we relog an
4059 * inode. max_key_type indicates the highest item type to remove.
4060 * This cannot be run for file data extents because it does not
4061 * free the extents they point to.
4063 static int drop_inode_items(struct btrfs_trans_handle *trans,
4064 struct btrfs_root *log,
4065 struct btrfs_path *path,
4066 struct btrfs_inode *inode,
4070 struct btrfs_key key;
4071 struct btrfs_key found_key;
4074 if (!inode_logged(trans, inode))
4077 key.objectid = btrfs_ino(inode);
4078 key.type = max_key_type;
4079 key.offset = (u64)-1;
4082 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
4083 BUG_ON(ret == 0); /* Logic error */
4087 if (path->slots[0] == 0)
4091 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4094 if (found_key.objectid != key.objectid)
4097 found_key.offset = 0;
4099 ret = btrfs_bin_search(path->nodes[0], &found_key, &start_slot);
4103 ret = btrfs_del_items(trans, log, path, start_slot,
4104 path->slots[0] - start_slot + 1);
4106 * If start slot isn't 0 then we don't need to re-search, we've
4107 * found the last guy with the objectid in this tree.
4109 if (ret || start_slot != 0)
4111 btrfs_release_path(path);
4113 btrfs_release_path(path);
4119 static int truncate_inode_items(struct btrfs_trans_handle *trans,
4120 struct btrfs_root *log_root,
4121 struct btrfs_inode *inode,
4122 u64 new_size, u32 min_type)
4124 struct btrfs_truncate_control control = {
4125 .new_size = new_size,
4126 .ino = btrfs_ino(inode),
4127 .min_type = min_type,
4128 .skip_ref_updates = true,
4131 return btrfs_truncate_inode_items(trans, log_root, &control);
4134 static void fill_inode_item(struct btrfs_trans_handle *trans,
4135 struct extent_buffer *leaf,
4136 struct btrfs_inode_item *item,
4137 struct inode *inode, int log_inode_only,
4140 struct btrfs_map_token token;
4143 btrfs_init_map_token(&token, leaf);
4145 if (log_inode_only) {
4146 /* set the generation to zero so the recover code
4147 * can tell the difference between an logging
4148 * just to say 'this inode exists' and a logging
4149 * to say 'update this inode with these values'
4151 btrfs_set_token_inode_generation(&token, item, 0);
4152 btrfs_set_token_inode_size(&token, item, logged_isize);
4154 btrfs_set_token_inode_generation(&token, item,
4155 BTRFS_I(inode)->generation);
4156 btrfs_set_token_inode_size(&token, item, inode->i_size);
4159 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4160 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4161 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4162 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4164 btrfs_set_token_timespec_sec(&token, &item->atime,
4165 inode->i_atime.tv_sec);
4166 btrfs_set_token_timespec_nsec(&token, &item->atime,
4167 inode->i_atime.tv_nsec);
4169 btrfs_set_token_timespec_sec(&token, &item->mtime,
4170 inode->i_mtime.tv_sec);
4171 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4172 inode->i_mtime.tv_nsec);
4174 btrfs_set_token_timespec_sec(&token, &item->ctime,
4175 inode->i_ctime.tv_sec);
4176 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4177 inode->i_ctime.tv_nsec);
4180 * We do not need to set the nbytes field, in fact during a fast fsync
4181 * its value may not even be correct, since a fast fsync does not wait
4182 * for ordered extent completion, which is where we update nbytes, it
4183 * only waits for writeback to complete. During log replay as we find
4184 * file extent items and replay them, we adjust the nbytes field of the
4185 * inode item in subvolume tree as needed (see overwrite_item()).
4188 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4189 btrfs_set_token_inode_transid(&token, item, trans->transid);
4190 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4191 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4192 BTRFS_I(inode)->ro_flags);
4193 btrfs_set_token_inode_flags(&token, item, flags);
4194 btrfs_set_token_inode_block_group(&token, item, 0);
4197 static int log_inode_item(struct btrfs_trans_handle *trans,
4198 struct btrfs_root *log, struct btrfs_path *path,
4199 struct btrfs_inode *inode, bool inode_item_dropped)
4201 struct btrfs_inode_item *inode_item;
4205 * If we are doing a fast fsync and the inode was logged before in the
4206 * current transaction, then we know the inode was previously logged and
4207 * it exists in the log tree. For performance reasons, in this case use
4208 * btrfs_search_slot() directly with ins_len set to 0 so that we never
4209 * attempt a write lock on the leaf's parent, which adds unnecessary lock
4210 * contention in case there are concurrent fsyncs for other inodes of the
4211 * same subvolume. Using btrfs_insert_empty_item() when the inode item
4212 * already exists can also result in unnecessarily splitting a leaf.
4214 if (!inode_item_dropped && inode->logged_trans == trans->transid) {
4215 ret = btrfs_search_slot(trans, log, &inode->location, path, 0, 1);
4221 * This means it is the first fsync in the current transaction,
4222 * so the inode item is not in the log and we need to insert it.
4223 * We can never get -EEXIST because we are only called for a fast
4224 * fsync and in case an inode eviction happens after the inode was
4225 * logged before in the current transaction, when we load again
4226 * the inode, we set BTRFS_INODE_NEEDS_FULL_SYNC on its runtime
4227 * flags and set ->logged_trans to 0.
4229 ret = btrfs_insert_empty_item(trans, log, path, &inode->location,
4230 sizeof(*inode_item));
4231 ASSERT(ret != -EEXIST);
4235 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4236 struct btrfs_inode_item);
4237 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
4239 btrfs_release_path(path);
4243 static int log_csums(struct btrfs_trans_handle *trans,
4244 struct btrfs_inode *inode,
4245 struct btrfs_root *log_root,
4246 struct btrfs_ordered_sum *sums)
4248 const u64 lock_end = sums->bytenr + sums->len - 1;
4249 struct extent_state *cached_state = NULL;
4253 * If this inode was not used for reflink operations in the current
4254 * transaction with new extents, then do the fast path, no need to
4255 * worry about logging checksum items with overlapping ranges.
4257 if (inode->last_reflink_trans < trans->transid)
4258 return btrfs_csum_file_blocks(trans, log_root, sums);
4261 * Serialize logging for checksums. This is to avoid racing with the
4262 * same checksum being logged by another task that is logging another
4263 * file which happens to refer to the same extent as well. Such races
4264 * can leave checksum items in the log with overlapping ranges.
4266 ret = lock_extent_bits(&log_root->log_csum_range, sums->bytenr,
4267 lock_end, &cached_state);
4271 * Due to extent cloning, we might have logged a csum item that covers a
4272 * subrange of a cloned extent, and later we can end up logging a csum
4273 * item for a larger subrange of the same extent or the entire range.
4274 * This would leave csum items in the log tree that cover the same range
4275 * and break the searches for checksums in the log tree, resulting in
4276 * some checksums missing in the fs/subvolume tree. So just delete (or
4277 * trim and adjust) any existing csum items in the log for this range.
4279 ret = btrfs_del_csums(trans, log_root, sums->bytenr, sums->len);
4281 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4283 unlock_extent_cached(&log_root->log_csum_range, sums->bytenr, lock_end,
4289 static noinline int copy_items(struct btrfs_trans_handle *trans,
4290 struct btrfs_inode *inode,
4291 struct btrfs_path *dst_path,
4292 struct btrfs_path *src_path,
4293 int start_slot, int nr, int inode_only,
4296 struct btrfs_fs_info *fs_info = trans->fs_info;
4297 unsigned long src_offset;
4298 unsigned long dst_offset;
4299 struct btrfs_root *log = inode->root->log_root;
4300 struct btrfs_file_extent_item *extent;
4301 struct btrfs_inode_item *inode_item;
4302 struct extent_buffer *src = src_path->nodes[0];
4304 struct btrfs_key *ins_keys;
4306 struct btrfs_item_batch batch;
4309 struct list_head ordered_sums;
4310 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
4312 INIT_LIST_HEAD(&ordered_sums);
4314 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
4315 nr * sizeof(u32), GFP_NOFS);
4319 ins_sizes = (u32 *)ins_data;
4320 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
4321 batch.keys = ins_keys;
4322 batch.data_sizes = ins_sizes;
4323 batch.total_data_size = 0;
4326 for (i = 0; i < nr; i++) {
4327 ins_sizes[i] = btrfs_item_size(src, i + start_slot);
4328 batch.total_data_size += ins_sizes[i];
4329 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
4331 ret = btrfs_insert_empty_items(trans, log, dst_path, &batch);
4337 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
4338 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
4339 dst_path->slots[0]);
4341 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
4343 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
4344 inode_item = btrfs_item_ptr(dst_path->nodes[0],
4346 struct btrfs_inode_item);
4347 fill_inode_item(trans, dst_path->nodes[0], inode_item,
4349 inode_only == LOG_INODE_EXISTS,
4352 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
4353 src_offset, ins_sizes[i]);
4356 /* take a reference on file data extents so that truncates
4357 * or deletes of this inode don't have to relog the inode
4360 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4363 extent = btrfs_item_ptr(src, start_slot + i,
4364 struct btrfs_file_extent_item);
4366 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4369 found_type = btrfs_file_extent_type(src, extent);
4370 if (found_type == BTRFS_FILE_EXTENT_REG) {
4371 struct btrfs_root *csum_root;
4373 ds = btrfs_file_extent_disk_bytenr(src,
4375 /* ds == 0 is a hole */
4379 dl = btrfs_file_extent_disk_num_bytes(src,
4381 cs = btrfs_file_extent_offset(src, extent);
4382 cl = btrfs_file_extent_num_bytes(src,
4384 if (btrfs_file_extent_compression(src,
4390 csum_root = btrfs_csum_root(fs_info, ds);
4391 ret = btrfs_lookup_csums_range(csum_root,
4392 ds + cs, ds + cs + cl - 1,
4400 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4401 btrfs_release_path(dst_path);
4405 * we have to do this after the loop above to avoid changing the
4406 * log tree while trying to change the log tree.
4408 while (!list_empty(&ordered_sums)) {
4409 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4410 struct btrfs_ordered_sum,
4413 ret = log_csums(trans, inode, log, sums);
4414 list_del(&sums->list);
4421 static int extent_cmp(void *priv, const struct list_head *a,
4422 const struct list_head *b)
4424 const struct extent_map *em1, *em2;
4426 em1 = list_entry(a, struct extent_map, list);
4427 em2 = list_entry(b, struct extent_map, list);
4429 if (em1->start < em2->start)
4431 else if (em1->start > em2->start)
4436 static int log_extent_csums(struct btrfs_trans_handle *trans,
4437 struct btrfs_inode *inode,
4438 struct btrfs_root *log_root,
4439 const struct extent_map *em,
4440 struct btrfs_log_ctx *ctx)
4442 struct btrfs_ordered_extent *ordered;
4443 struct btrfs_root *csum_root;
4446 u64 mod_start = em->mod_start;
4447 u64 mod_len = em->mod_len;
4448 LIST_HEAD(ordered_sums);
4451 if (inode->flags & BTRFS_INODE_NODATASUM ||
4452 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4453 em->block_start == EXTENT_MAP_HOLE)
4456 list_for_each_entry(ordered, &ctx->ordered_extents, log_list) {
4457 const u64 ordered_end = ordered->file_offset + ordered->num_bytes;
4458 const u64 mod_end = mod_start + mod_len;
4459 struct btrfs_ordered_sum *sums;
4464 if (ordered_end <= mod_start)
4466 if (mod_end <= ordered->file_offset)
4470 * We are going to copy all the csums on this ordered extent, so
4471 * go ahead and adjust mod_start and mod_len in case this ordered
4472 * extent has already been logged.
4474 if (ordered->file_offset > mod_start) {
4475 if (ordered_end >= mod_end)
4476 mod_len = ordered->file_offset - mod_start;
4478 * If we have this case
4480 * |--------- logged extent ---------|
4481 * |----- ordered extent ----|
4483 * Just don't mess with mod_start and mod_len, we'll
4484 * just end up logging more csums than we need and it
4488 if (ordered_end < mod_end) {
4489 mod_len = mod_end - ordered_end;
4490 mod_start = ordered_end;
4497 * To keep us from looping for the above case of an ordered
4498 * extent that falls inside of the logged extent.
4500 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM, &ordered->flags))
4503 list_for_each_entry(sums, &ordered->list, list) {
4504 ret = log_csums(trans, inode, log_root, sums);
4510 /* We're done, found all csums in the ordered extents. */
4514 /* If we're compressed we have to save the entire range of csums. */
4515 if (em->compress_type) {
4517 csum_len = max(em->block_len, em->orig_block_len);
4519 csum_offset = mod_start - em->start;
4523 /* block start is already adjusted for the file extent offset. */
4524 csum_root = btrfs_csum_root(trans->fs_info, em->block_start);
4525 ret = btrfs_lookup_csums_range(csum_root,
4526 em->block_start + csum_offset,
4527 em->block_start + csum_offset +
4528 csum_len - 1, &ordered_sums, 0);
4532 while (!list_empty(&ordered_sums)) {
4533 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4534 struct btrfs_ordered_sum,
4537 ret = log_csums(trans, inode, log_root, sums);
4538 list_del(&sums->list);
4545 static int log_one_extent(struct btrfs_trans_handle *trans,
4546 struct btrfs_inode *inode,
4547 const struct extent_map *em,
4548 struct btrfs_path *path,
4549 struct btrfs_log_ctx *ctx)
4551 struct btrfs_drop_extents_args drop_args = { 0 };
4552 struct btrfs_root *log = inode->root->log_root;
4553 struct btrfs_file_extent_item *fi;
4554 struct extent_buffer *leaf;
4555 struct btrfs_map_token token;
4556 struct btrfs_key key;
4557 u64 extent_offset = em->start - em->orig_start;
4561 ret = log_extent_csums(trans, inode, log, em, ctx);
4566 * If this is the first time we are logging the inode in the current
4567 * transaction, we can avoid btrfs_drop_extents(), which is expensive
4568 * because it does a deletion search, which always acquires write locks
4569 * for extent buffers at levels 2, 1 and 0. This not only wastes time
4570 * but also adds significant contention in a log tree, since log trees
4571 * are small, with a root at level 2 or 3 at most, due to their short
4574 if (inode_logged(trans, inode)) {
4575 drop_args.path = path;
4576 drop_args.start = em->start;
4577 drop_args.end = em->start + em->len;
4578 drop_args.replace_extent = true;
4579 drop_args.extent_item_size = sizeof(*fi);
4580 ret = btrfs_drop_extents(trans, log, inode, &drop_args);
4585 if (!drop_args.extent_inserted) {
4586 key.objectid = btrfs_ino(inode);
4587 key.type = BTRFS_EXTENT_DATA_KEY;
4588 key.offset = em->start;
4590 ret = btrfs_insert_empty_item(trans, log, path, &key,
4595 leaf = path->nodes[0];
4596 btrfs_init_map_token(&token, leaf);
4597 fi = btrfs_item_ptr(leaf, path->slots[0],
4598 struct btrfs_file_extent_item);
4600 btrfs_set_token_file_extent_generation(&token, fi, trans->transid);
4601 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4602 btrfs_set_token_file_extent_type(&token, fi,
4603 BTRFS_FILE_EXTENT_PREALLOC);
4605 btrfs_set_token_file_extent_type(&token, fi,
4606 BTRFS_FILE_EXTENT_REG);
4608 block_len = max(em->block_len, em->orig_block_len);
4609 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4610 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4612 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4613 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4614 btrfs_set_token_file_extent_disk_bytenr(&token, fi,
4617 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, block_len);
4619 btrfs_set_token_file_extent_disk_bytenr(&token, fi, 0);
4620 btrfs_set_token_file_extent_disk_num_bytes(&token, fi, 0);
4623 btrfs_set_token_file_extent_offset(&token, fi, extent_offset);
4624 btrfs_set_token_file_extent_num_bytes(&token, fi, em->len);
4625 btrfs_set_token_file_extent_ram_bytes(&token, fi, em->ram_bytes);
4626 btrfs_set_token_file_extent_compression(&token, fi, em->compress_type);
4627 btrfs_set_token_file_extent_encryption(&token, fi, 0);
4628 btrfs_set_token_file_extent_other_encoding(&token, fi, 0);
4629 btrfs_mark_buffer_dirty(leaf);
4631 btrfs_release_path(path);
4637 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4638 * lose them after doing a fast fsync and replaying the log. We scan the
4639 * subvolume's root instead of iterating the inode's extent map tree because
4640 * otherwise we can log incorrect extent items based on extent map conversion.
4641 * That can happen due to the fact that extent maps are merged when they
4642 * are not in the extent map tree's list of modified extents.
4644 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4645 struct btrfs_inode *inode,
4646 struct btrfs_path *path)
4648 struct btrfs_root *root = inode->root;
4649 struct btrfs_key key;
4650 const u64 i_size = i_size_read(&inode->vfs_inode);
4651 const u64 ino = btrfs_ino(inode);
4652 struct btrfs_path *dst_path = NULL;
4653 bool dropped_extents = false;
4654 u64 truncate_offset = i_size;
4655 struct extent_buffer *leaf;
4661 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4665 key.type = BTRFS_EXTENT_DATA_KEY;
4666 key.offset = i_size;
4667 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4672 * We must check if there is a prealloc extent that starts before the
4673 * i_size and crosses the i_size boundary. This is to ensure later we
4674 * truncate down to the end of that extent and not to the i_size, as
4675 * otherwise we end up losing part of the prealloc extent after a log
4676 * replay and with an implicit hole if there is another prealloc extent
4677 * that starts at an offset beyond i_size.
4679 ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY);
4684 struct btrfs_file_extent_item *ei;
4686 leaf = path->nodes[0];
4687 slot = path->slots[0];
4688 ei = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4690 if (btrfs_file_extent_type(leaf, ei) ==
4691 BTRFS_FILE_EXTENT_PREALLOC) {
4694 btrfs_item_key_to_cpu(leaf, &key, slot);
4695 extent_end = key.offset +
4696 btrfs_file_extent_num_bytes(leaf, ei);
4698 if (extent_end > i_size)
4699 truncate_offset = extent_end;
4706 leaf = path->nodes[0];
4707 slot = path->slots[0];
4709 if (slot >= btrfs_header_nritems(leaf)) {
4711 ret = copy_items(trans, inode, dst_path, path,
4712 start_slot, ins_nr, 1, 0);
4717 ret = btrfs_next_leaf(root, path);
4727 btrfs_item_key_to_cpu(leaf, &key, slot);
4728 if (key.objectid > ino)
4730 if (WARN_ON_ONCE(key.objectid < ino) ||
4731 key.type < BTRFS_EXTENT_DATA_KEY ||
4732 key.offset < i_size) {
4736 if (!dropped_extents) {
4738 * Avoid logging extent items logged in past fsync calls
4739 * and leading to duplicate keys in the log tree.
4741 ret = truncate_inode_items(trans, root->log_root, inode,
4743 BTRFS_EXTENT_DATA_KEY);
4746 dropped_extents = true;
4753 dst_path = btrfs_alloc_path();
4761 ret = copy_items(trans, inode, dst_path, path,
4762 start_slot, ins_nr, 1, 0);
4764 btrfs_release_path(path);
4765 btrfs_free_path(dst_path);
4769 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4770 struct btrfs_inode *inode,
4771 struct btrfs_path *path,
4772 struct btrfs_log_ctx *ctx)
4774 struct btrfs_ordered_extent *ordered;
4775 struct btrfs_ordered_extent *tmp;
4776 struct extent_map *em, *n;
4777 struct list_head extents;
4778 struct extent_map_tree *tree = &inode->extent_tree;
4782 INIT_LIST_HEAD(&extents);
4784 write_lock(&tree->lock);
4786 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4787 list_del_init(&em->list);
4789 * Just an arbitrary number, this can be really CPU intensive
4790 * once we start getting a lot of extents, and really once we
4791 * have a bunch of extents we just want to commit since it will
4794 if (++num > 32768) {
4795 list_del_init(&tree->modified_extents);
4800 if (em->generation < trans->transid)
4803 /* We log prealloc extents beyond eof later. */
4804 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4805 em->start >= i_size_read(&inode->vfs_inode))
4808 /* Need a ref to keep it from getting evicted from cache */
4809 refcount_inc(&em->refs);
4810 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4811 list_add_tail(&em->list, &extents);
4815 list_sort(NULL, &extents, extent_cmp);
4817 while (!list_empty(&extents)) {
4818 em = list_entry(extents.next, struct extent_map, list);
4820 list_del_init(&em->list);
4823 * If we had an error we just need to delete everybody from our
4827 clear_em_logging(tree, em);
4828 free_extent_map(em);
4832 write_unlock(&tree->lock);
4834 ret = log_one_extent(trans, inode, em, path, ctx);
4835 write_lock(&tree->lock);
4836 clear_em_logging(tree, em);
4837 free_extent_map(em);
4839 WARN_ON(!list_empty(&extents));
4840 write_unlock(&tree->lock);
4842 btrfs_release_path(path);
4844 ret = btrfs_log_prealloc_extents(trans, inode, path);
4849 * We have logged all extents successfully, now make sure the commit of
4850 * the current transaction waits for the ordered extents to complete
4851 * before it commits and wipes out the log trees, otherwise we would
4852 * lose data if an ordered extents completes after the transaction
4853 * commits and a power failure happens after the transaction commit.
4855 list_for_each_entry_safe(ordered, tmp, &ctx->ordered_extents, log_list) {
4856 list_del_init(&ordered->log_list);
4857 set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags);
4859 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4860 spin_lock_irq(&inode->ordered_tree.lock);
4861 if (!test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) {
4862 set_bit(BTRFS_ORDERED_PENDING, &ordered->flags);
4863 atomic_inc(&trans->transaction->pending_ordered);
4865 spin_unlock_irq(&inode->ordered_tree.lock);
4867 btrfs_put_ordered_extent(ordered);
4873 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4874 struct btrfs_path *path, u64 *size_ret)
4876 struct btrfs_key key;
4879 key.objectid = btrfs_ino(inode);
4880 key.type = BTRFS_INODE_ITEM_KEY;
4883 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4886 } else if (ret > 0) {
4889 struct btrfs_inode_item *item;
4891 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4892 struct btrfs_inode_item);
4893 *size_ret = btrfs_inode_size(path->nodes[0], item);
4895 * If the in-memory inode's i_size is smaller then the inode
4896 * size stored in the btree, return the inode's i_size, so
4897 * that we get a correct inode size after replaying the log
4898 * when before a power failure we had a shrinking truncate
4899 * followed by addition of a new name (rename / new hard link).
4900 * Otherwise return the inode size from the btree, to avoid
4901 * data loss when replaying a log due to previously doing a
4902 * write that expands the inode's size and logging a new name
4903 * immediately after.
4905 if (*size_ret > inode->vfs_inode.i_size)
4906 *size_ret = inode->vfs_inode.i_size;
4909 btrfs_release_path(path);
4914 * At the moment we always log all xattrs. This is to figure out at log replay
4915 * time which xattrs must have their deletion replayed. If a xattr is missing
4916 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4917 * because if a xattr is deleted, the inode is fsynced and a power failure
4918 * happens, causing the log to be replayed the next time the fs is mounted,
4919 * we want the xattr to not exist anymore (same behaviour as other filesystems
4920 * with a journal, ext3/4, xfs, f2fs, etc).
4922 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4923 struct btrfs_inode *inode,
4924 struct btrfs_path *path,
4925 struct btrfs_path *dst_path)
4927 struct btrfs_root *root = inode->root;
4929 struct btrfs_key key;
4930 const u64 ino = btrfs_ino(inode);
4933 bool found_xattrs = false;
4935 if (test_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags))
4939 key.type = BTRFS_XATTR_ITEM_KEY;
4942 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4947 int slot = path->slots[0];
4948 struct extent_buffer *leaf = path->nodes[0];
4949 int nritems = btrfs_header_nritems(leaf);
4951 if (slot >= nritems) {
4953 ret = copy_items(trans, inode, dst_path, path,
4954 start_slot, ins_nr, 1, 0);
4959 ret = btrfs_next_leaf(root, path);
4967 btrfs_item_key_to_cpu(leaf, &key, slot);
4968 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4975 found_xattrs = true;
4979 ret = copy_items(trans, inode, dst_path, path,
4980 start_slot, ins_nr, 1, 0);
4986 set_bit(BTRFS_INODE_NO_XATTRS, &inode->runtime_flags);
4992 * When using the NO_HOLES feature if we punched a hole that causes the
4993 * deletion of entire leafs or all the extent items of the first leaf (the one
4994 * that contains the inode item and references) we may end up not processing
4995 * any extents, because there are no leafs with a generation matching the
4996 * current transaction that have extent items for our inode. So we need to find
4997 * if any holes exist and then log them. We also need to log holes after any
4998 * truncate operation that changes the inode's size.
5000 static int btrfs_log_holes(struct btrfs_trans_handle *trans,
5001 struct btrfs_inode *inode,
5002 struct btrfs_path *path)
5004 struct btrfs_root *root = inode->root;
5005 struct btrfs_fs_info *fs_info = root->fs_info;
5006 struct btrfs_key key;
5007 const u64 ino = btrfs_ino(inode);
5008 const u64 i_size = i_size_read(&inode->vfs_inode);
5009 u64 prev_extent_end = 0;
5012 if (!btrfs_fs_incompat(fs_info, NO_HOLES) || i_size == 0)
5016 key.type = BTRFS_EXTENT_DATA_KEY;
5019 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5024 struct extent_buffer *leaf = path->nodes[0];
5026 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
5027 ret = btrfs_next_leaf(root, path);
5034 leaf = path->nodes[0];
5037 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
5038 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY)
5041 /* We have a hole, log it. */
5042 if (prev_extent_end < key.offset) {
5043 const u64 hole_len = key.offset - prev_extent_end;
5046 * Release the path to avoid deadlocks with other code
5047 * paths that search the root while holding locks on
5048 * leafs from the log root.
5050 btrfs_release_path(path);
5051 ret = btrfs_insert_file_extent(trans, root->log_root,
5052 ino, prev_extent_end, 0,
5053 0, hole_len, 0, hole_len,
5059 * Search for the same key again in the root. Since it's
5060 * an extent item and we are holding the inode lock, the
5061 * key must still exist. If it doesn't just emit warning
5062 * and return an error to fall back to a transaction
5065 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5068 if (WARN_ON(ret > 0))
5070 leaf = path->nodes[0];
5073 prev_extent_end = btrfs_file_extent_end(path);
5078 if (prev_extent_end < i_size) {
5081 btrfs_release_path(path);
5082 hole_len = ALIGN(i_size - prev_extent_end, fs_info->sectorsize);
5083 ret = btrfs_insert_file_extent(trans, root->log_root,
5084 ino, prev_extent_end, 0, 0,
5085 hole_len, 0, hole_len,
5095 * When we are logging a new inode X, check if it doesn't have a reference that
5096 * matches the reference from some other inode Y created in a past transaction
5097 * and that was renamed in the current transaction. If we don't do this, then at
5098 * log replay time we can lose inode Y (and all its files if it's a directory):
5101 * echo "hello world" > /mnt/x/foobar
5104 * mkdir /mnt/x # or touch /mnt/x
5105 * xfs_io -c fsync /mnt/x
5107 * mount fs, trigger log replay
5109 * After the log replay procedure, we would lose the first directory and all its
5110 * files (file foobar).
5111 * For the case where inode Y is not a directory we simply end up losing it:
5113 * echo "123" > /mnt/foo
5115 * mv /mnt/foo /mnt/bar
5116 * echo "abc" > /mnt/foo
5117 * xfs_io -c fsync /mnt/foo
5120 * We also need this for cases where a snapshot entry is replaced by some other
5121 * entry (file or directory) otherwise we end up with an unreplayable log due to
5122 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
5123 * if it were a regular entry:
5126 * btrfs subvolume snapshot /mnt /mnt/x/snap
5127 * btrfs subvolume delete /mnt/x/snap
5130 * fsync /mnt/x or fsync some new file inside it
5133 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
5134 * the same transaction.
5136 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
5138 const struct btrfs_key *key,
5139 struct btrfs_inode *inode,
5140 u64 *other_ino, u64 *other_parent)
5143 struct btrfs_path *search_path;
5146 u32 item_size = btrfs_item_size(eb, slot);
5148 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
5150 search_path = btrfs_alloc_path();
5153 search_path->search_commit_root = 1;
5154 search_path->skip_locking = 1;
5156 while (cur_offset < item_size) {
5160 unsigned long name_ptr;
5161 struct btrfs_dir_item *di;
5163 if (key->type == BTRFS_INODE_REF_KEY) {
5164 struct btrfs_inode_ref *iref;
5166 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
5167 parent = key->offset;
5168 this_name_len = btrfs_inode_ref_name_len(eb, iref);
5169 name_ptr = (unsigned long)(iref + 1);
5170 this_len = sizeof(*iref) + this_name_len;
5172 struct btrfs_inode_extref *extref;
5174 extref = (struct btrfs_inode_extref *)(ptr +
5176 parent = btrfs_inode_extref_parent(eb, extref);
5177 this_name_len = btrfs_inode_extref_name_len(eb, extref);
5178 name_ptr = (unsigned long)&extref->name;
5179 this_len = sizeof(*extref) + this_name_len;
5182 if (this_name_len > name_len) {
5185 new_name = krealloc(name, this_name_len, GFP_NOFS);
5190 name_len = this_name_len;
5194 read_extent_buffer(eb, name, name_ptr, this_name_len);
5195 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
5196 parent, name, this_name_len, 0);
5197 if (di && !IS_ERR(di)) {
5198 struct btrfs_key di_key;
5200 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
5202 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
5203 if (di_key.objectid != key->objectid) {
5205 *other_ino = di_key.objectid;
5206 *other_parent = parent;
5214 } else if (IS_ERR(di)) {
5218 btrfs_release_path(search_path);
5220 cur_offset += this_len;
5224 btrfs_free_path(search_path);
5229 struct btrfs_ino_list {
5232 struct list_head list;
5235 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
5236 struct btrfs_root *root,
5237 struct btrfs_path *path,
5238 struct btrfs_log_ctx *ctx,
5239 u64 ino, u64 parent)
5241 struct btrfs_ino_list *ino_elem;
5242 LIST_HEAD(inode_list);
5245 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5248 ino_elem->ino = ino;
5249 ino_elem->parent = parent;
5250 list_add_tail(&ino_elem->list, &inode_list);
5252 while (!list_empty(&inode_list)) {
5253 struct btrfs_fs_info *fs_info = root->fs_info;
5254 struct btrfs_key key;
5255 struct inode *inode;
5257 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
5259 ino = ino_elem->ino;
5260 parent = ino_elem->parent;
5261 list_del(&ino_elem->list);
5266 btrfs_release_path(path);
5268 inode = btrfs_iget(fs_info->sb, ino, root);
5270 * If the other inode that had a conflicting dir entry was
5271 * deleted in the current transaction, we need to log its parent
5274 if (IS_ERR(inode)) {
5275 ret = PTR_ERR(inode);
5276 if (ret == -ENOENT) {
5277 inode = btrfs_iget(fs_info->sb, parent, root);
5278 if (IS_ERR(inode)) {
5279 ret = PTR_ERR(inode);
5281 ret = btrfs_log_inode(trans,
5283 LOG_OTHER_INODE_ALL,
5285 btrfs_add_delayed_iput(inode);
5291 * If the inode was already logged skip it - otherwise we can
5292 * hit an infinite loop. Example:
5294 * From the commit root (previous transaction) we have the
5297 * inode 257 a directory
5298 * inode 258 with references "zz" and "zz_link" on inode 257
5299 * inode 259 with reference "a" on inode 257
5301 * And in the current (uncommitted) transaction we have:
5303 * inode 257 a directory, unchanged
5304 * inode 258 with references "a" and "a2" on inode 257
5305 * inode 259 with reference "zz_link" on inode 257
5306 * inode 261 with reference "zz" on inode 257
5308 * When logging inode 261 the following infinite loop could
5309 * happen if we don't skip already logged inodes:
5311 * - we detect inode 258 as a conflicting inode, with inode 261
5312 * on reference "zz", and log it;
5314 * - we detect inode 259 as a conflicting inode, with inode 258
5315 * on reference "a", and log it;
5317 * - we detect inode 258 as a conflicting inode, with inode 259
5318 * on reference "zz_link", and log it - again! After this we
5319 * repeat the above steps forever.
5321 spin_lock(&BTRFS_I(inode)->lock);
5323 * Check the inode's logged_trans only instead of
5324 * btrfs_inode_in_log(). This is because the last_log_commit of
5325 * the inode is not updated when we only log that it exists (see
5326 * btrfs_log_inode()).
5328 if (BTRFS_I(inode)->logged_trans == trans->transid) {
5329 spin_unlock(&BTRFS_I(inode)->lock);
5330 btrfs_add_delayed_iput(inode);
5333 spin_unlock(&BTRFS_I(inode)->lock);
5335 * We are safe logging the other inode without acquiring its
5336 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5337 * are safe against concurrent renames of the other inode as
5338 * well because during a rename we pin the log and update the
5339 * log with the new name before we unpin it.
5341 ret = btrfs_log_inode(trans, BTRFS_I(inode), LOG_OTHER_INODE, ctx);
5343 btrfs_add_delayed_iput(inode);
5348 key.type = BTRFS_INODE_REF_KEY;
5350 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5352 btrfs_add_delayed_iput(inode);
5357 struct extent_buffer *leaf = path->nodes[0];
5358 int slot = path->slots[0];
5360 u64 other_parent = 0;
5362 if (slot >= btrfs_header_nritems(leaf)) {
5363 ret = btrfs_next_leaf(root, path);
5366 } else if (ret > 0) {
5373 btrfs_item_key_to_cpu(leaf, &key, slot);
5374 if (key.objectid != ino ||
5375 (key.type != BTRFS_INODE_REF_KEY &&
5376 key.type != BTRFS_INODE_EXTREF_KEY)) {
5381 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5382 BTRFS_I(inode), &other_ino,
5387 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5392 ino_elem->ino = other_ino;
5393 ino_elem->parent = other_parent;
5394 list_add_tail(&ino_elem->list, &inode_list);
5399 btrfs_add_delayed_iput(inode);
5405 static int copy_inode_items_to_log(struct btrfs_trans_handle *trans,
5406 struct btrfs_inode *inode,
5407 struct btrfs_key *min_key,
5408 const struct btrfs_key *max_key,
5409 struct btrfs_path *path,
5410 struct btrfs_path *dst_path,
5411 const u64 logged_isize,
5412 const bool recursive_logging,
5413 const int inode_only,
5414 struct btrfs_log_ctx *ctx,
5415 bool *need_log_inode_item)
5417 struct btrfs_root *root = inode->root;
5418 int ins_start_slot = 0;
5423 ret = btrfs_search_forward(root, min_key, path, trans->transid);
5431 /* Note, ins_nr might be > 0 here, cleanup outside the loop */
5432 if (min_key->objectid != max_key->objectid)
5434 if (min_key->type > max_key->type)
5437 if (min_key->type == BTRFS_INODE_ITEM_KEY)
5438 *need_log_inode_item = false;
5440 if ((min_key->type == BTRFS_INODE_REF_KEY ||
5441 min_key->type == BTRFS_INODE_EXTREF_KEY) &&
5442 inode->generation == trans->transid &&
5443 !recursive_logging) {
5445 u64 other_parent = 0;
5447 ret = btrfs_check_ref_name_override(path->nodes[0],
5448 path->slots[0], min_key, inode,
5449 &other_ino, &other_parent);
5452 } else if (ret > 0 &&
5453 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5458 ins_start_slot = path->slots[0];
5460 ret = copy_items(trans, inode, dst_path, path,
5461 ins_start_slot, ins_nr,
5462 inode_only, logged_isize);
5467 ret = log_conflicting_inodes(trans, root, path,
5468 ctx, other_ino, other_parent);
5471 btrfs_release_path(path);
5476 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5477 if (min_key->type == BTRFS_XATTR_ITEM_KEY) {
5480 ret = copy_items(trans, inode, dst_path, path,
5482 ins_nr, inode_only, logged_isize);
5489 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5492 } else if (!ins_nr) {
5493 ins_start_slot = path->slots[0];
5498 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5499 ins_nr, inode_only, logged_isize);
5503 ins_start_slot = path->slots[0];
5506 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) {
5507 btrfs_item_key_to_cpu(path->nodes[0], min_key,
5512 ret = copy_items(trans, inode, dst_path, path,
5513 ins_start_slot, ins_nr, inode_only,
5519 btrfs_release_path(path);
5521 if (min_key->offset < (u64)-1) {
5523 } else if (min_key->type < max_key->type) {
5525 min_key->offset = 0;
5531 ret = copy_items(trans, inode, dst_path, path, ins_start_slot,
5532 ins_nr, inode_only, logged_isize);
5537 /* log a single inode in the tree log.
5538 * At least one parent directory for this inode must exist in the tree
5539 * or be logged already.
5541 * Any items from this inode changed by the current transaction are copied
5542 * to the log tree. An extra reference is taken on any extents in this
5543 * file, allowing us to avoid a whole pile of corner cases around logging
5544 * blocks that have been removed from the tree.
5546 * See LOG_INODE_ALL and related defines for a description of what inode_only
5549 * This handles both files and directories.
5551 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5552 struct btrfs_inode *inode,
5554 struct btrfs_log_ctx *ctx)
5556 struct btrfs_path *path;
5557 struct btrfs_path *dst_path;
5558 struct btrfs_key min_key;
5559 struct btrfs_key max_key;
5560 struct btrfs_root *log = inode->root->log_root;
5563 bool fast_search = false;
5564 u64 ino = btrfs_ino(inode);
5565 struct extent_map_tree *em_tree = &inode->extent_tree;
5566 u64 logged_isize = 0;
5567 bool need_log_inode_item = true;
5568 bool xattrs_logged = false;
5569 bool recursive_logging = false;
5570 bool inode_item_dropped = true;
5572 path = btrfs_alloc_path();
5575 dst_path = btrfs_alloc_path();
5577 btrfs_free_path(path);
5581 min_key.objectid = ino;
5582 min_key.type = BTRFS_INODE_ITEM_KEY;
5585 max_key.objectid = ino;
5588 /* today the code can only do partial logging of directories */
5589 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5590 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5591 &inode->runtime_flags) &&
5592 inode_only >= LOG_INODE_EXISTS))
5593 max_key.type = BTRFS_XATTR_ITEM_KEY;
5595 max_key.type = (u8)-1;
5596 max_key.offset = (u64)-1;
5599 * Only run delayed items if we are a directory. We want to make sure
5600 * all directory indexes hit the fs/subvolume tree so we can find them
5601 * and figure out which index ranges have to be logged.
5603 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5604 err = btrfs_commit_inode_delayed_items(trans, inode);
5609 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5610 recursive_logging = true;
5611 if (inode_only == LOG_OTHER_INODE)
5612 inode_only = LOG_INODE_EXISTS;
5614 inode_only = LOG_INODE_ALL;
5615 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5617 mutex_lock(&inode->log_mutex);
5621 * This is for cases where logging a directory could result in losing a
5622 * a file after replaying the log. For example, if we move a file from a
5623 * directory A to a directory B, then fsync directory A, we have no way
5624 * to known the file was moved from A to B, so logging just A would
5625 * result in losing the file after a log replay.
5627 if (S_ISDIR(inode->vfs_inode.i_mode) &&
5628 inode_only == LOG_INODE_ALL &&
5629 inode->last_unlink_trans >= trans->transid) {
5630 btrfs_set_log_full_commit(trans);
5636 * a brute force approach to making sure we get the most uptodate
5637 * copies of everything.
5639 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5640 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5642 clear_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags);
5643 if (inode_only == LOG_INODE_EXISTS)
5644 max_key_type = BTRFS_XATTR_ITEM_KEY;
5645 ret = drop_inode_items(trans, log, path, inode, max_key_type);
5647 if (inode_only == LOG_INODE_EXISTS && inode_logged(trans, inode)) {
5649 * Make sure the new inode item we write to the log has
5650 * the same isize as the current one (if it exists).
5651 * This is necessary to prevent data loss after log
5652 * replay, and also to prevent doing a wrong expanding
5653 * truncate - for e.g. create file, write 4K into offset
5654 * 0, fsync, write 4K into offset 4096, add hard link,
5655 * fsync some other file (to sync log), power fail - if
5656 * we use the inode's current i_size, after log replay
5657 * we get a 8Kb file, with the last 4Kb extent as a hole
5658 * (zeroes), as if an expanding truncate happened,
5659 * instead of getting a file of 4Kb only.
5661 err = logged_inode_size(log, inode, path, &logged_isize);
5665 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5666 &inode->runtime_flags)) {
5667 if (inode_only == LOG_INODE_EXISTS) {
5668 max_key.type = BTRFS_XATTR_ITEM_KEY;
5669 ret = drop_inode_items(trans, log, path, inode,
5672 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5673 &inode->runtime_flags);
5674 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5675 &inode->runtime_flags);
5676 if (inode_logged(trans, inode))
5677 ret = truncate_inode_items(trans, log,
5680 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5681 &inode->runtime_flags) ||
5682 inode_only == LOG_INODE_EXISTS) {
5683 if (inode_only == LOG_INODE_ALL)
5685 max_key.type = BTRFS_XATTR_ITEM_KEY;
5686 ret = drop_inode_items(trans, log, path, inode,
5689 if (inode_only == LOG_INODE_ALL)
5691 inode_item_dropped = false;
5701 err = copy_inode_items_to_log(trans, inode, &min_key, &max_key,
5702 path, dst_path, logged_isize,
5703 recursive_logging, inode_only, ctx,
5704 &need_log_inode_item);
5708 btrfs_release_path(path);
5709 btrfs_release_path(dst_path);
5710 err = btrfs_log_all_xattrs(trans, inode, path, dst_path);
5713 xattrs_logged = true;
5714 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5715 btrfs_release_path(path);
5716 btrfs_release_path(dst_path);
5717 err = btrfs_log_holes(trans, inode, path);
5722 btrfs_release_path(path);
5723 btrfs_release_path(dst_path);
5724 if (need_log_inode_item) {
5725 err = log_inode_item(trans, log, dst_path, inode, inode_item_dropped);
5729 * If we are doing a fast fsync and the inode was logged before
5730 * in this transaction, we don't need to log the xattrs because
5731 * they were logged before. If xattrs were added, changed or
5732 * deleted since the last time we logged the inode, then we have
5733 * already logged them because the inode had the runtime flag
5734 * BTRFS_INODE_COPY_EVERYTHING set.
5736 if (!xattrs_logged && inode->logged_trans < trans->transid) {
5737 err = btrfs_log_all_xattrs(trans, inode, path, dst_path);
5740 btrfs_release_path(path);
5744 ret = btrfs_log_changed_extents(trans, inode, dst_path, ctx);
5749 } else if (inode_only == LOG_INODE_ALL) {
5750 struct extent_map *em, *n;
5752 write_lock(&em_tree->lock);
5753 list_for_each_entry_safe(em, n, &em_tree->modified_extents, list)
5754 list_del_init(&em->list);
5755 write_unlock(&em_tree->lock);
5758 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5759 ret = log_directory_changes(trans, inode, path, dst_path, ctx);
5766 spin_lock(&inode->lock);
5767 inode->logged_trans = trans->transid;
5769 * Don't update last_log_commit if we logged that an inode exists.
5770 * We do this for three reasons:
5772 * 1) We might have had buffered writes to this inode that were
5773 * flushed and had their ordered extents completed in this
5774 * transaction, but we did not previously log the inode with
5775 * LOG_INODE_ALL. Later the inode was evicted and after that
5776 * it was loaded again and this LOG_INODE_EXISTS log operation
5777 * happened. We must make sure that if an explicit fsync against
5778 * the inode is performed later, it logs the new extents, an
5779 * updated inode item, etc, and syncs the log. The same logic
5780 * applies to direct IO writes instead of buffered writes.
5782 * 2) When we log the inode with LOG_INODE_EXISTS, its inode item
5783 * is logged with an i_size of 0 or whatever value was logged
5784 * before. If later the i_size of the inode is increased by a
5785 * truncate operation, the log is synced through an fsync of
5786 * some other inode and then finally an explicit fsync against
5787 * this inode is made, we must make sure this fsync logs the
5788 * inode with the new i_size, the hole between old i_size and
5789 * the new i_size, and syncs the log.
5791 * 3) If we are logging that an ancestor inode exists as part of
5792 * logging a new name from a link or rename operation, don't update
5793 * its last_log_commit - otherwise if an explicit fsync is made
5794 * against an ancestor, the fsync considers the inode in the log
5795 * and doesn't sync the log, resulting in the ancestor missing after
5796 * a power failure unless the log was synced as part of an fsync
5797 * against any other unrelated inode.
5799 if (inode_only != LOG_INODE_EXISTS)
5800 inode->last_log_commit = inode->last_sub_trans;
5801 spin_unlock(&inode->lock);
5803 mutex_unlock(&inode->log_mutex);
5805 btrfs_free_path(path);
5806 btrfs_free_path(dst_path);
5811 * Check if we need to log an inode. This is used in contexts where while
5812 * logging an inode we need to log another inode (either that it exists or in
5813 * full mode). This is used instead of btrfs_inode_in_log() because the later
5814 * requires the inode to be in the log and have the log transaction committed,
5815 * while here we do not care if the log transaction was already committed - our
5816 * caller will commit the log later - and we want to avoid logging an inode
5817 * multiple times when multiple tasks have joined the same log transaction.
5819 static bool need_log_inode(struct btrfs_trans_handle *trans,
5820 struct btrfs_inode *inode)
5823 * If a directory was not modified, no dentries added or removed, we can
5824 * and should avoid logging it.
5826 if (S_ISDIR(inode->vfs_inode.i_mode) && inode->last_trans < trans->transid)
5830 * If this inode does not have new/updated/deleted xattrs since the last
5831 * time it was logged and is flagged as logged in the current transaction,
5832 * we can skip logging it. As for new/deleted names, those are updated in
5833 * the log by link/unlink/rename operations.
5834 * In case the inode was logged and then evicted and reloaded, its
5835 * logged_trans will be 0, in which case we have to fully log it since
5836 * logged_trans is a transient field, not persisted.
5838 if (inode->logged_trans == trans->transid &&
5839 !test_bit(BTRFS_INODE_COPY_EVERYTHING, &inode->runtime_flags))
5845 struct btrfs_dir_list {
5847 struct list_head list;
5851 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5852 * details about the why it is needed.
5853 * This is a recursive operation - if an existing dentry corresponds to a
5854 * directory, that directory's new entries are logged too (same behaviour as
5855 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5856 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5857 * complains about the following circular lock dependency / possible deadlock:
5861 * lock(&type->i_mutex_dir_key#3/2);
5862 * lock(sb_internal#2);
5863 * lock(&type->i_mutex_dir_key#3/2);
5864 * lock(&sb->s_type->i_mutex_key#14);
5866 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5867 * sb_start_intwrite() in btrfs_start_transaction().
5868 * Not locking i_mutex of the inodes is still safe because:
5870 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5871 * that while logging the inode new references (names) are added or removed
5872 * from the inode, leaving the logged inode item with a link count that does
5873 * not match the number of logged inode reference items. This is fine because
5874 * at log replay time we compute the real number of links and correct the
5875 * link count in the inode item (see replay_one_buffer() and
5876 * link_to_fixup_dir());
5878 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5879 * while logging the inode's items new index items (key type
5880 * BTRFS_DIR_INDEX_KEY) are added to fs/subvol tree and the logged inode item
5881 * has a size that doesn't match the sum of the lengths of all the logged
5882 * names - this is ok, not a problem, because at log replay time we set the
5883 * directory's i_size to the correct value (see replay_one_name() and
5884 * do_overwrite_item()).
5886 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5887 struct btrfs_root *root,
5888 struct btrfs_inode *start_inode,
5889 struct btrfs_log_ctx *ctx)
5891 struct btrfs_fs_info *fs_info = root->fs_info;
5892 struct btrfs_root *log = root->log_root;
5893 struct btrfs_path *path;
5894 LIST_HEAD(dir_list);
5895 struct btrfs_dir_list *dir_elem;
5899 * If we are logging a new name, as part of a link or rename operation,
5900 * don't bother logging new dentries, as we just want to log the names
5901 * of an inode and that any new parents exist.
5903 if (ctx->logging_new_name)
5906 path = btrfs_alloc_path();
5910 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5912 btrfs_free_path(path);
5915 dir_elem->ino = btrfs_ino(start_inode);
5916 list_add_tail(&dir_elem->list, &dir_list);
5918 while (!list_empty(&dir_list)) {
5919 struct extent_buffer *leaf;
5920 struct btrfs_key min_key;
5924 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5927 goto next_dir_inode;
5929 min_key.objectid = dir_elem->ino;
5930 min_key.type = BTRFS_DIR_INDEX_KEY;
5933 btrfs_release_path(path);
5934 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5936 goto next_dir_inode;
5937 } else if (ret > 0) {
5939 goto next_dir_inode;
5943 leaf = path->nodes[0];
5944 nritems = btrfs_header_nritems(leaf);
5945 for (i = path->slots[0]; i < nritems; i++) {
5946 struct btrfs_dir_item *di;
5947 struct btrfs_key di_key;
5948 struct inode *di_inode;
5949 struct btrfs_dir_list *new_dir_elem;
5950 int log_mode = LOG_INODE_EXISTS;
5953 btrfs_item_key_to_cpu(leaf, &min_key, i);
5954 if (min_key.objectid != dir_elem->ino ||
5955 min_key.type != BTRFS_DIR_INDEX_KEY)
5956 goto next_dir_inode;
5958 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5959 type = btrfs_dir_type(leaf, di);
5960 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5961 type != BTRFS_FT_DIR)
5963 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5964 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5967 btrfs_release_path(path);
5968 di_inode = btrfs_iget(fs_info->sb, di_key.objectid, root);
5969 if (IS_ERR(di_inode)) {
5970 ret = PTR_ERR(di_inode);
5971 goto next_dir_inode;
5974 if (!need_log_inode(trans, BTRFS_I(di_inode))) {
5975 btrfs_add_delayed_iput(di_inode);
5979 ctx->log_new_dentries = false;
5980 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5981 log_mode = LOG_INODE_ALL;
5982 ret = btrfs_log_inode(trans, BTRFS_I(di_inode),
5984 btrfs_add_delayed_iput(di_inode);
5986 goto next_dir_inode;
5987 if (ctx->log_new_dentries) {
5988 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5990 if (!new_dir_elem) {
5992 goto next_dir_inode;
5994 new_dir_elem->ino = di_key.objectid;
5995 list_add_tail(&new_dir_elem->list, &dir_list);
6000 ret = btrfs_next_leaf(log, path);
6002 goto next_dir_inode;
6003 } else if (ret > 0) {
6005 goto next_dir_inode;
6009 if (min_key.offset < (u64)-1) {
6014 list_del(&dir_elem->list);
6018 btrfs_free_path(path);
6022 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
6023 struct btrfs_inode *inode,
6024 struct btrfs_log_ctx *ctx)
6026 struct btrfs_fs_info *fs_info = trans->fs_info;
6028 struct btrfs_path *path;
6029 struct btrfs_key key;
6030 struct btrfs_root *root = inode->root;
6031 const u64 ino = btrfs_ino(inode);
6033 path = btrfs_alloc_path();
6036 path->skip_locking = 1;
6037 path->search_commit_root = 1;
6040 key.type = BTRFS_INODE_REF_KEY;
6042 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6047 struct extent_buffer *leaf = path->nodes[0];
6048 int slot = path->slots[0];
6053 if (slot >= btrfs_header_nritems(leaf)) {
6054 ret = btrfs_next_leaf(root, path);
6062 btrfs_item_key_to_cpu(leaf, &key, slot);
6063 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
6064 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
6067 item_size = btrfs_item_size(leaf, slot);
6068 ptr = btrfs_item_ptr_offset(leaf, slot);
6069 while (cur_offset < item_size) {
6070 struct btrfs_key inode_key;
6071 struct inode *dir_inode;
6073 inode_key.type = BTRFS_INODE_ITEM_KEY;
6074 inode_key.offset = 0;
6076 if (key.type == BTRFS_INODE_EXTREF_KEY) {
6077 struct btrfs_inode_extref *extref;
6079 extref = (struct btrfs_inode_extref *)
6081 inode_key.objectid = btrfs_inode_extref_parent(
6083 cur_offset += sizeof(*extref);
6084 cur_offset += btrfs_inode_extref_name_len(leaf,
6087 inode_key.objectid = key.offset;
6088 cur_offset = item_size;
6091 dir_inode = btrfs_iget(fs_info->sb, inode_key.objectid,
6094 * If the parent inode was deleted, return an error to
6095 * fallback to a transaction commit. This is to prevent
6096 * getting an inode that was moved from one parent A to
6097 * a parent B, got its former parent A deleted and then
6098 * it got fsync'ed, from existing at both parents after
6099 * a log replay (and the old parent still existing).
6106 * mv /mnt/B/bar /mnt/A/bar
6107 * mv -T /mnt/A /mnt/B
6111 * If we ignore the old parent B which got deleted,
6112 * after a log replay we would have file bar linked
6113 * at both parents and the old parent B would still
6116 if (IS_ERR(dir_inode)) {
6117 ret = PTR_ERR(dir_inode);
6121 if (!need_log_inode(trans, BTRFS_I(dir_inode))) {
6122 btrfs_add_delayed_iput(dir_inode);
6126 ctx->log_new_dentries = false;
6127 ret = btrfs_log_inode(trans, BTRFS_I(dir_inode),
6128 LOG_INODE_ALL, ctx);
6129 if (!ret && ctx->log_new_dentries)
6130 ret = log_new_dir_dentries(trans, root,
6131 BTRFS_I(dir_inode), ctx);
6132 btrfs_add_delayed_iput(dir_inode);
6140 btrfs_free_path(path);
6144 static int log_new_ancestors(struct btrfs_trans_handle *trans,
6145 struct btrfs_root *root,
6146 struct btrfs_path *path,
6147 struct btrfs_log_ctx *ctx)
6149 struct btrfs_key found_key;
6151 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
6154 struct btrfs_fs_info *fs_info = root->fs_info;
6155 struct extent_buffer *leaf = path->nodes[0];
6156 int slot = path->slots[0];
6157 struct btrfs_key search_key;
6158 struct inode *inode;
6162 btrfs_release_path(path);
6164 ino = found_key.offset;
6166 search_key.objectid = found_key.offset;
6167 search_key.type = BTRFS_INODE_ITEM_KEY;
6168 search_key.offset = 0;
6169 inode = btrfs_iget(fs_info->sb, ino, root);
6171 return PTR_ERR(inode);
6173 if (BTRFS_I(inode)->generation >= trans->transid &&
6174 need_log_inode(trans, BTRFS_I(inode)))
6175 ret = btrfs_log_inode(trans, BTRFS_I(inode),
6176 LOG_INODE_EXISTS, ctx);
6177 btrfs_add_delayed_iput(inode);
6181 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
6184 search_key.type = BTRFS_INODE_REF_KEY;
6185 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6189 leaf = path->nodes[0];
6190 slot = path->slots[0];
6191 if (slot >= btrfs_header_nritems(leaf)) {
6192 ret = btrfs_next_leaf(root, path);
6197 leaf = path->nodes[0];
6198 slot = path->slots[0];
6201 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6202 if (found_key.objectid != search_key.objectid ||
6203 found_key.type != BTRFS_INODE_REF_KEY)
6209 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
6210 struct btrfs_inode *inode,
6211 struct dentry *parent,
6212 struct btrfs_log_ctx *ctx)
6214 struct btrfs_root *root = inode->root;
6215 struct dentry *old_parent = NULL;
6216 struct super_block *sb = inode->vfs_inode.i_sb;
6220 if (!parent || d_really_is_negative(parent) ||
6224 inode = BTRFS_I(d_inode(parent));
6225 if (root != inode->root)
6228 if (inode->generation >= trans->transid &&
6229 need_log_inode(trans, inode)) {
6230 ret = btrfs_log_inode(trans, inode,
6231 LOG_INODE_EXISTS, ctx);
6235 if (IS_ROOT(parent))
6238 parent = dget_parent(parent);
6240 old_parent = parent;
6247 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
6248 struct btrfs_inode *inode,
6249 struct dentry *parent,
6250 struct btrfs_log_ctx *ctx)
6252 struct btrfs_root *root = inode->root;
6253 const u64 ino = btrfs_ino(inode);
6254 struct btrfs_path *path;
6255 struct btrfs_key search_key;
6259 * For a single hard link case, go through a fast path that does not
6260 * need to iterate the fs/subvolume tree.
6262 if (inode->vfs_inode.i_nlink < 2)
6263 return log_new_ancestors_fast(trans, inode, parent, ctx);
6265 path = btrfs_alloc_path();
6269 search_key.objectid = ino;
6270 search_key.type = BTRFS_INODE_REF_KEY;
6271 search_key.offset = 0;
6273 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6280 struct extent_buffer *leaf = path->nodes[0];
6281 int slot = path->slots[0];
6282 struct btrfs_key found_key;
6284 if (slot >= btrfs_header_nritems(leaf)) {
6285 ret = btrfs_next_leaf(root, path);
6293 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6294 if (found_key.objectid != ino ||
6295 found_key.type > BTRFS_INODE_EXTREF_KEY)
6299 * Don't deal with extended references because they are rare
6300 * cases and too complex to deal with (we would need to keep
6301 * track of which subitem we are processing for each item in
6302 * this loop, etc). So just return some error to fallback to
6303 * a transaction commit.
6305 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6311 * Logging ancestors needs to do more searches on the fs/subvol
6312 * tree, so it releases the path as needed to avoid deadlocks.
6313 * Keep track of the last inode ref key and resume from that key
6314 * after logging all new ancestors for the current hard link.
6316 memcpy(&search_key, &found_key, sizeof(search_key));
6318 ret = log_new_ancestors(trans, root, path, ctx);
6321 btrfs_release_path(path);
6326 btrfs_free_path(path);
6331 * helper function around btrfs_log_inode to make sure newly created
6332 * parent directories also end up in the log. A minimal inode and backref
6333 * only logging is done of any parent directories that are older than
6334 * the last committed transaction
6336 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6337 struct btrfs_inode *inode,
6338 struct dentry *parent,
6340 struct btrfs_log_ctx *ctx)
6342 struct btrfs_root *root = inode->root;
6343 struct btrfs_fs_info *fs_info = root->fs_info;
6345 bool log_dentries = false;
6347 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6352 if (btrfs_root_refs(&root->root_item) == 0) {
6358 * Skip already logged inodes or inodes corresponding to tmpfiles
6359 * (since logging them is pointless, a link count of 0 means they
6360 * will never be accessible).
6362 if ((btrfs_inode_in_log(inode, trans->transid) &&
6363 list_empty(&ctx->ordered_extents)) ||
6364 inode->vfs_inode.i_nlink == 0) {
6365 ret = BTRFS_NO_LOG_SYNC;
6369 ret = start_log_trans(trans, root, ctx);
6373 ret = btrfs_log_inode(trans, inode, inode_only, ctx);
6378 * for regular files, if its inode is already on disk, we don't
6379 * have to worry about the parents at all. This is because
6380 * we can use the last_unlink_trans field to record renames
6381 * and other fun in this file.
6383 if (S_ISREG(inode->vfs_inode.i_mode) &&
6384 inode->generation < trans->transid &&
6385 inode->last_unlink_trans < trans->transid) {
6390 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx->log_new_dentries)
6391 log_dentries = true;
6394 * On unlink we must make sure all our current and old parent directory
6395 * inodes are fully logged. This is to prevent leaving dangling
6396 * directory index entries in directories that were our parents but are
6397 * not anymore. Not doing this results in old parent directory being
6398 * impossible to delete after log replay (rmdir will always fail with
6399 * error -ENOTEMPTY).
6405 * ln testdir/foo testdir/bar
6407 * unlink testdir/bar
6408 * xfs_io -c fsync testdir/foo
6410 * mount fs, triggers log replay
6412 * If we don't log the parent directory (testdir), after log replay the
6413 * directory still has an entry pointing to the file inode using the bar
6414 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6415 * the file inode has a link count of 1.
6421 * ln foo testdir/foo2
6422 * ln foo testdir/foo3
6424 * unlink testdir/foo3
6425 * xfs_io -c fsync foo
6427 * mount fs, triggers log replay
6429 * Similar as the first example, after log replay the parent directory
6430 * testdir still has an entry pointing to the inode file with name foo3
6431 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6432 * and has a link count of 2.
6434 if (inode->last_unlink_trans >= trans->transid) {
6435 ret = btrfs_log_all_parents(trans, inode, ctx);
6440 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6445 ret = log_new_dir_dentries(trans, root, inode, ctx);
6450 btrfs_set_log_full_commit(trans);
6455 btrfs_remove_log_ctx(root, ctx);
6456 btrfs_end_log_trans(root);
6462 * it is not safe to log dentry if the chunk root has added new
6463 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6464 * If this returns 1, you must commit the transaction to safely get your
6467 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6468 struct dentry *dentry,
6469 struct btrfs_log_ctx *ctx)
6471 struct dentry *parent = dget_parent(dentry);
6474 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6475 LOG_INODE_ALL, ctx);
6482 * should be called during mount to recover any replay any log trees
6485 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6488 struct btrfs_path *path;
6489 struct btrfs_trans_handle *trans;
6490 struct btrfs_key key;
6491 struct btrfs_key found_key;
6492 struct btrfs_root *log;
6493 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6494 struct walk_control wc = {
6495 .process_func = process_one_buffer,
6496 .stage = LOG_WALK_PIN_ONLY,
6499 path = btrfs_alloc_path();
6503 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6505 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6506 if (IS_ERR(trans)) {
6507 ret = PTR_ERR(trans);
6514 ret = walk_log_tree(trans, log_root_tree, &wc);
6516 btrfs_abort_transaction(trans, ret);
6521 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6522 key.offset = (u64)-1;
6523 key.type = BTRFS_ROOT_ITEM_KEY;
6526 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6529 btrfs_abort_transaction(trans, ret);
6533 if (path->slots[0] == 0)
6537 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6539 btrfs_release_path(path);
6540 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6543 log = btrfs_read_tree_root(log_root_tree, &found_key);
6546 btrfs_abort_transaction(trans, ret);
6550 wc.replay_dest = btrfs_get_fs_root(fs_info, found_key.offset,
6552 if (IS_ERR(wc.replay_dest)) {
6553 ret = PTR_ERR(wc.replay_dest);
6556 * We didn't find the subvol, likely because it was
6557 * deleted. This is ok, simply skip this log and go to
6560 * We need to exclude the root because we can't have
6561 * other log replays overwriting this log as we'll read
6562 * it back in a few more times. This will keep our
6563 * block from being modified, and we'll just bail for
6564 * each subsequent pass.
6567 ret = btrfs_pin_extent_for_log_replay(trans,
6570 btrfs_put_root(log);
6574 btrfs_abort_transaction(trans, ret);
6578 wc.replay_dest->log_root = log;
6579 ret = btrfs_record_root_in_trans(trans, wc.replay_dest);
6581 /* The loop needs to continue due to the root refs */
6582 btrfs_abort_transaction(trans, ret);
6584 ret = walk_log_tree(trans, log, &wc);
6586 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6587 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6590 btrfs_abort_transaction(trans, ret);
6593 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6594 struct btrfs_root *root = wc.replay_dest;
6596 btrfs_release_path(path);
6599 * We have just replayed everything, and the highest
6600 * objectid of fs roots probably has changed in case
6601 * some inode_item's got replayed.
6603 * root->objectid_mutex is not acquired as log replay
6604 * could only happen during mount.
6606 ret = btrfs_init_root_free_objectid(root);
6608 btrfs_abort_transaction(trans, ret);
6611 wc.replay_dest->log_root = NULL;
6612 btrfs_put_root(wc.replay_dest);
6613 btrfs_put_root(log);
6618 if (found_key.offset == 0)
6620 key.offset = found_key.offset - 1;
6622 btrfs_release_path(path);
6624 /* step one is to pin it all, step two is to replay just inodes */
6627 wc.process_func = replay_one_buffer;
6628 wc.stage = LOG_WALK_REPLAY_INODES;
6631 /* step three is to replay everything */
6632 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6637 btrfs_free_path(path);
6639 /* step 4: commit the transaction, which also unpins the blocks */
6640 ret = btrfs_commit_transaction(trans);
6644 log_root_tree->log_root = NULL;
6645 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6646 btrfs_put_root(log_root_tree);
6651 btrfs_end_transaction(wc.trans);
6652 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6653 btrfs_free_path(path);
6658 * there are some corner cases where we want to force a full
6659 * commit instead of allowing a directory to be logged.
6661 * They revolve around files there were unlinked from the directory, and
6662 * this function updates the parent directory so that a full commit is
6663 * properly done if it is fsync'd later after the unlinks are done.
6665 * Must be called before the unlink operations (updates to the subvolume tree,
6666 * inodes, etc) are done.
6668 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6669 struct btrfs_inode *dir, struct btrfs_inode *inode,
6673 * when we're logging a file, if it hasn't been renamed
6674 * or unlinked, and its inode is fully committed on disk,
6675 * we don't have to worry about walking up the directory chain
6676 * to log its parents.
6678 * So, we use the last_unlink_trans field to put this transid
6679 * into the file. When the file is logged we check it and
6680 * don't log the parents if the file is fully on disk.
6682 mutex_lock(&inode->log_mutex);
6683 inode->last_unlink_trans = trans->transid;
6684 mutex_unlock(&inode->log_mutex);
6687 * if this directory was already logged any new
6688 * names for this file/dir will get recorded
6690 if (dir->logged_trans == trans->transid)
6694 * if the inode we're about to unlink was logged,
6695 * the log will be properly updated for any new names
6697 if (inode->logged_trans == trans->transid)
6701 * when renaming files across directories, if the directory
6702 * there we're unlinking from gets fsync'd later on, there's
6703 * no way to find the destination directory later and fsync it
6704 * properly. So, we have to be conservative and force commits
6705 * so the new name gets discovered.
6710 /* we can safely do the unlink without any special recording */
6714 mutex_lock(&dir->log_mutex);
6715 dir->last_unlink_trans = trans->transid;
6716 mutex_unlock(&dir->log_mutex);
6720 * Make sure that if someone attempts to fsync the parent directory of a deleted
6721 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6722 * that after replaying the log tree of the parent directory's root we will not
6723 * see the snapshot anymore and at log replay time we will not see any log tree
6724 * corresponding to the deleted snapshot's root, which could lead to replaying
6725 * it after replaying the log tree of the parent directory (which would replay
6726 * the snapshot delete operation).
6728 * Must be called before the actual snapshot destroy operation (updates to the
6729 * parent root and tree of tree roots trees, etc) are done.
6731 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6732 struct btrfs_inode *dir)
6734 mutex_lock(&dir->log_mutex);
6735 dir->last_unlink_trans = trans->transid;
6736 mutex_unlock(&dir->log_mutex);
6740 * Call this after adding a new name for a file and it will properly
6741 * update the log to reflect the new name.
6743 void btrfs_log_new_name(struct btrfs_trans_handle *trans,
6744 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6745 struct dentry *parent)
6747 struct btrfs_log_ctx ctx;
6750 * this will force the logging code to walk the dentry chain
6753 if (!S_ISDIR(inode->vfs_inode.i_mode))
6754 inode->last_unlink_trans = trans->transid;
6757 * if this inode hasn't been logged and directory we're renaming it
6758 * from hasn't been logged, we don't need to log it
6760 if (!inode_logged(trans, inode) &&
6761 (!old_dir || !inode_logged(trans, old_dir)))
6765 * If we are doing a rename (old_dir is not NULL) from a directory that
6766 * was previously logged, make sure the next log attempt on the directory
6767 * is not skipped and logs the inode again. This is because the log may
6768 * not currently be authoritative for a range including the old
6769 * BTRFS_DIR_INDEX_KEY key, so we want to make sure after a log replay we
6770 * do not end up with both the new and old dentries around (in case the
6771 * inode is a directory we would have a directory with two hard links and
6772 * 2 inode references for different parents). The next log attempt of
6773 * old_dir will happen at btrfs_log_all_parents(), called through
6774 * btrfs_log_inode_parent() below, because we have previously set
6775 * inode->last_unlink_trans to the current transaction ID, either here or
6776 * at btrfs_record_unlink_dir() in case the inode is a directory.
6779 old_dir->logged_trans = 0;
6781 btrfs_init_log_ctx(&ctx, &inode->vfs_inode);
6782 ctx.logging_new_name = true;
6784 * We don't care about the return value. If we fail to log the new name
6785 * then we know the next attempt to sync the log will fallback to a full
6786 * transaction commit (due to a call to btrfs_set_log_full_commit()), so
6787 * we don't need to worry about getting a log committed that has an
6788 * inconsistent state after a rename operation.
6790 btrfs_log_inode_parent(trans, inode, parent, LOG_INODE_EXISTS, &ctx);