1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/blkdev.h>
9 #include <linux/list_sort.h>
10 #include <linux/iversion.h>
16 #include "print-tree.h"
18 #include "compression.h"
20 #include "inode-map.h"
22 /* magic values for the inode_only field in btrfs_log_inode:
24 * LOG_INODE_ALL means to log everything
25 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 * directory trouble cases
38 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
39 * log, we must force a full commit before doing an fsync of the directory
40 * where the unlink was done.
41 * ---> record transid of last unlink/rename per directory
45 * rename foo/some_dir foo2/some_dir
47 * fsync foo/some_dir/some_file
49 * The fsync above will unlink the original some_dir without recording
50 * it in its new location (foo2). After a crash, some_dir will be gone
51 * unless the fsync of some_file forces a full commit
53 * 2) we must log any new names for any file or dir that is in the fsync
54 * log. ---> check inode while renaming/linking.
56 * 2a) we must log any new names for any file or dir during rename
57 * when the directory they are being removed from was logged.
58 * ---> check inode and old parent dir during rename
60 * 2a is actually the more important variant. With the extra logging
61 * a crash might unlink the old name without recreating the new one
63 * 3) after a crash, we must go through any directories with a link count
64 * of zero and redo the rm -rf
71 * The directory f1 was fully removed from the FS, but fsync was never
72 * called on f1, only its parent dir. After a crash the rm -rf must
73 * be replayed. This must be able to recurse down the entire
74 * directory tree. The inode link count fixup code takes care of the
79 * stages for the tree walking. The first
80 * stage (0) is to only pin down the blocks we find
81 * the second stage (1) is to make sure that all the inodes
82 * we find in the log are created in the subvolume.
84 * The last stage is to deal with directories and links and extents
85 * and all the other fun semantics
89 LOG_WALK_REPLAY_INODES,
90 LOG_WALK_REPLAY_DIR_INDEX,
94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root, struct btrfs_inode *inode,
99 struct btrfs_log_ctx *ctx);
100 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_path *path, u64 objectid);
103 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
104 struct btrfs_root *root,
105 struct btrfs_root *log,
106 struct btrfs_path *path,
107 u64 dirid, int del_all);
110 * tree logging is a special write ahead log used to make sure that
111 * fsyncs and O_SYNCs can happen without doing full tree commits.
113 * Full tree commits are expensive because they require commonly
114 * modified blocks to be recowed, creating many dirty pages in the
115 * extent tree an 4x-6x higher write load than ext3.
117 * Instead of doing a tree commit on every fsync, we use the
118 * key ranges and transaction ids to find items for a given file or directory
119 * that have changed in this transaction. Those items are copied into
120 * a special tree (one per subvolume root), that tree is written to disk
121 * and then the fsync is considered complete.
123 * After a crash, items are copied out of the log-tree back into the
124 * subvolume tree. Any file data extents found are recorded in the extent
125 * allocation tree, and the log-tree freed.
127 * The log tree is read three times, once to pin down all the extents it is
128 * using in ram and once, once to create all the inodes logged in the tree
129 * and once to do all the other items.
133 * start a sub transaction and setup the log tree
134 * this increments the log tree writer count to make the people
135 * syncing the tree wait for us to finish
137 static int start_log_trans(struct btrfs_trans_handle *trans,
138 struct btrfs_root *root,
139 struct btrfs_log_ctx *ctx)
141 struct btrfs_fs_info *fs_info = root->fs_info;
144 mutex_lock(&root->log_mutex);
146 if (root->log_root) {
147 if (btrfs_need_log_full_commit(trans)) {
152 if (!root->log_start_pid) {
153 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
154 root->log_start_pid = current->pid;
155 } else if (root->log_start_pid != current->pid) {
156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 mutex_lock(&fs_info->tree_log_mutex);
160 if (!fs_info->log_root_tree)
161 ret = btrfs_init_log_root_tree(trans, fs_info);
162 mutex_unlock(&fs_info->tree_log_mutex);
166 ret = btrfs_add_log_tree(trans, root);
170 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
171 root->log_start_pid = current->pid;
174 atomic_inc(&root->log_batch);
175 atomic_inc(&root->log_writers);
177 int index = root->log_transid % 2;
178 list_add_tail(&ctx->list, &root->log_ctxs[index]);
179 ctx->log_transid = root->log_transid;
183 mutex_unlock(&root->log_mutex);
188 * returns 0 if there was a log transaction running and we were able
189 * to join, or returns -ENOENT if there were not transactions
192 static int join_running_log_trans(struct btrfs_root *root)
196 mutex_lock(&root->log_mutex);
197 if (root->log_root) {
199 atomic_inc(&root->log_writers);
201 mutex_unlock(&root->log_mutex);
206 * This either makes the current running log transaction wait
207 * until you call btrfs_end_log_trans() or it makes any future
208 * log transactions wait until you call btrfs_end_log_trans()
210 void btrfs_pin_log_trans(struct btrfs_root *root)
212 mutex_lock(&root->log_mutex);
213 atomic_inc(&root->log_writers);
214 mutex_unlock(&root->log_mutex);
218 * indicate we're done making changes to the log tree
219 * and wake up anyone waiting to do a sync
221 void btrfs_end_log_trans(struct btrfs_root *root)
223 if (atomic_dec_and_test(&root->log_writers)) {
224 /* atomic_dec_and_test implies a barrier */
225 cond_wake_up_nomb(&root->log_writer_wait);
229 static int btrfs_write_tree_block(struct extent_buffer *buf)
231 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
232 buf->start + buf->len - 1);
235 static void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
237 filemap_fdatawait_range(buf->pages[0]->mapping,
238 buf->start, buf->start + buf->len - 1);
242 * the walk control struct is used to pass state down the chain when
243 * processing the log tree. The stage field tells us which part
244 * of the log tree processing we are currently doing. The others
245 * are state fields used for that specific part
247 struct walk_control {
248 /* should we free the extent on disk when done? This is used
249 * at transaction commit time while freeing a log tree
253 /* should we write out the extent buffer? This is used
254 * while flushing the log tree to disk during a sync
258 /* should we wait for the extent buffer io to finish? Also used
259 * while flushing the log tree to disk for a sync
263 /* pin only walk, we record which extents on disk belong to the
268 /* what stage of the replay code we're currently in */
272 * Ignore any items from the inode currently being processed. Needs
273 * to be set every time we find a BTRFS_INODE_ITEM_KEY and we are in
274 * the LOG_WALK_REPLAY_INODES stage.
276 bool ignore_cur_inode;
278 /* the root we are currently replaying */
279 struct btrfs_root *replay_dest;
281 /* the trans handle for the current replay */
282 struct btrfs_trans_handle *trans;
284 /* the function that gets used to process blocks we find in the
285 * tree. Note the extent_buffer might not be up to date when it is
286 * passed in, and it must be checked or read if you need the data
289 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
290 struct walk_control *wc, u64 gen, int level);
294 * process_func used to pin down extents, write them or wait on them
296 static int process_one_buffer(struct btrfs_root *log,
297 struct extent_buffer *eb,
298 struct walk_control *wc, u64 gen, int level)
300 struct btrfs_fs_info *fs_info = log->fs_info;
304 * If this fs is mixed then we need to be able to process the leaves to
305 * pin down any logged extents, so we have to read the block.
307 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
308 ret = btrfs_read_buffer(eb, gen, level, NULL);
314 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
317 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
318 if (wc->pin && btrfs_header_level(eb) == 0)
319 ret = btrfs_exclude_logged_extents(eb);
321 btrfs_write_tree_block(eb);
323 btrfs_wait_tree_block_writeback(eb);
329 * Item overwrite used by replay and tree logging. eb, slot and key all refer
330 * to the src data we are copying out.
332 * root is the tree we are copying into, and path is a scratch
333 * path for use in this function (it should be released on entry and
334 * will be released on exit).
336 * If the key is already in the destination tree the existing item is
337 * overwritten. If the existing item isn't big enough, it is extended.
338 * If it is too large, it is truncated.
340 * If the key isn't in the destination yet, a new item is inserted.
342 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
343 struct btrfs_root *root,
344 struct btrfs_path *path,
345 struct extent_buffer *eb, int slot,
346 struct btrfs_key *key)
350 u64 saved_i_size = 0;
351 int save_old_i_size = 0;
352 unsigned long src_ptr;
353 unsigned long dst_ptr;
354 int overwrite_root = 0;
355 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
357 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
360 item_size = btrfs_item_size_nr(eb, slot);
361 src_ptr = btrfs_item_ptr_offset(eb, slot);
363 /* look for the key in the destination tree */
364 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
371 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
373 if (dst_size != item_size)
376 if (item_size == 0) {
377 btrfs_release_path(path);
380 dst_copy = kmalloc(item_size, GFP_NOFS);
381 src_copy = kmalloc(item_size, GFP_NOFS);
382 if (!dst_copy || !src_copy) {
383 btrfs_release_path(path);
389 read_extent_buffer(eb, src_copy, src_ptr, item_size);
391 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
392 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
394 ret = memcmp(dst_copy, src_copy, item_size);
399 * they have the same contents, just return, this saves
400 * us from cowing blocks in the destination tree and doing
401 * extra writes that may not have been done by a previous
405 btrfs_release_path(path);
410 * We need to load the old nbytes into the inode so when we
411 * replay the extents we've logged we get the right nbytes.
414 struct btrfs_inode_item *item;
418 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
419 struct btrfs_inode_item);
420 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
421 item = btrfs_item_ptr(eb, slot,
422 struct btrfs_inode_item);
423 btrfs_set_inode_nbytes(eb, item, nbytes);
426 * If this is a directory we need to reset the i_size to
427 * 0 so that we can set it up properly when replaying
428 * the rest of the items in this log.
430 mode = btrfs_inode_mode(eb, item);
432 btrfs_set_inode_size(eb, item, 0);
434 } else if (inode_item) {
435 struct btrfs_inode_item *item;
439 * New inode, set nbytes to 0 so that the nbytes comes out
440 * properly when we replay the extents.
442 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
443 btrfs_set_inode_nbytes(eb, item, 0);
446 * If this is a directory we need to reset the i_size to 0 so
447 * that we can set it up properly when replaying the rest of
448 * the items in this log.
450 mode = btrfs_inode_mode(eb, item);
452 btrfs_set_inode_size(eb, item, 0);
455 btrfs_release_path(path);
456 /* try to insert the key into the destination tree */
457 path->skip_release_on_error = 1;
458 ret = btrfs_insert_empty_item(trans, root, path,
460 path->skip_release_on_error = 0;
462 /* make sure any existing item is the correct size */
463 if (ret == -EEXIST || ret == -EOVERFLOW) {
465 found_size = btrfs_item_size_nr(path->nodes[0],
467 if (found_size > item_size)
468 btrfs_truncate_item(path, item_size, 1);
469 else if (found_size < item_size)
470 btrfs_extend_item(path, item_size - found_size);
474 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
477 /* don't overwrite an existing inode if the generation number
478 * was logged as zero. This is done when the tree logging code
479 * is just logging an inode to make sure it exists after recovery.
481 * Also, don't overwrite i_size on directories during replay.
482 * log replay inserts and removes directory items based on the
483 * state of the tree found in the subvolume, and i_size is modified
486 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
487 struct btrfs_inode_item *src_item;
488 struct btrfs_inode_item *dst_item;
490 src_item = (struct btrfs_inode_item *)src_ptr;
491 dst_item = (struct btrfs_inode_item *)dst_ptr;
493 if (btrfs_inode_generation(eb, src_item) == 0) {
494 struct extent_buffer *dst_eb = path->nodes[0];
495 const u64 ino_size = btrfs_inode_size(eb, src_item);
498 * For regular files an ino_size == 0 is used only when
499 * logging that an inode exists, as part of a directory
500 * fsync, and the inode wasn't fsynced before. In this
501 * case don't set the size of the inode in the fs/subvol
502 * tree, otherwise we would be throwing valid data away.
504 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
505 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
507 struct btrfs_map_token token;
509 btrfs_init_map_token(&token, dst_eb);
510 btrfs_set_token_inode_size(dst_eb, dst_item,
516 if (overwrite_root &&
517 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
518 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
520 saved_i_size = btrfs_inode_size(path->nodes[0],
525 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
528 if (save_old_i_size) {
529 struct btrfs_inode_item *dst_item;
530 dst_item = (struct btrfs_inode_item *)dst_ptr;
531 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
534 /* make sure the generation is filled in */
535 if (key->type == BTRFS_INODE_ITEM_KEY) {
536 struct btrfs_inode_item *dst_item;
537 dst_item = (struct btrfs_inode_item *)dst_ptr;
538 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
539 btrfs_set_inode_generation(path->nodes[0], dst_item,
544 btrfs_mark_buffer_dirty(path->nodes[0]);
545 btrfs_release_path(path);
550 * simple helper to read an inode off the disk from a given root
551 * This can only be called for subvolume roots and not for the log
553 static noinline struct inode *read_one_inode(struct btrfs_root *root,
556 struct btrfs_key key;
559 key.objectid = objectid;
560 key.type = BTRFS_INODE_ITEM_KEY;
562 inode = btrfs_iget(root->fs_info->sb, &key, root);
568 /* replays a single extent in 'eb' at 'slot' with 'key' into the
569 * subvolume 'root'. path is released on entry and should be released
572 * extents in the log tree have not been allocated out of the extent
573 * tree yet. So, this completes the allocation, taking a reference
574 * as required if the extent already exists or creating a new extent
575 * if it isn't in the extent allocation tree yet.
577 * The extent is inserted into the file, dropping any existing extents
578 * from the file that overlap the new one.
580 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
581 struct btrfs_root *root,
582 struct btrfs_path *path,
583 struct extent_buffer *eb, int slot,
584 struct btrfs_key *key)
586 struct btrfs_fs_info *fs_info = root->fs_info;
589 u64 start = key->offset;
591 struct btrfs_file_extent_item *item;
592 struct inode *inode = NULL;
596 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
597 found_type = btrfs_file_extent_type(eb, item);
599 if (found_type == BTRFS_FILE_EXTENT_REG ||
600 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
601 nbytes = btrfs_file_extent_num_bytes(eb, item);
602 extent_end = start + nbytes;
605 * We don't add to the inodes nbytes if we are prealloc or a
608 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
610 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
611 size = btrfs_file_extent_ram_bytes(eb, item);
612 nbytes = btrfs_file_extent_ram_bytes(eb, item);
613 extent_end = ALIGN(start + size,
614 fs_info->sectorsize);
620 inode = read_one_inode(root, key->objectid);
627 * first check to see if we already have this extent in the
628 * file. This must be done before the btrfs_drop_extents run
629 * so we don't try to drop this extent.
631 ret = btrfs_lookup_file_extent(trans, root, path,
632 btrfs_ino(BTRFS_I(inode)), start, 0);
635 (found_type == BTRFS_FILE_EXTENT_REG ||
636 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
637 struct btrfs_file_extent_item cmp1;
638 struct btrfs_file_extent_item cmp2;
639 struct btrfs_file_extent_item *existing;
640 struct extent_buffer *leaf;
642 leaf = path->nodes[0];
643 existing = btrfs_item_ptr(leaf, path->slots[0],
644 struct btrfs_file_extent_item);
646 read_extent_buffer(eb, &cmp1, (unsigned long)item,
648 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
652 * we already have a pointer to this exact extent,
653 * we don't have to do anything
655 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
656 btrfs_release_path(path);
660 btrfs_release_path(path);
662 /* drop any overlapping extents */
663 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
667 if (found_type == BTRFS_FILE_EXTENT_REG ||
668 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
670 unsigned long dest_offset;
671 struct btrfs_key ins;
673 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
674 btrfs_fs_incompat(fs_info, NO_HOLES))
677 ret = btrfs_insert_empty_item(trans, root, path, key,
681 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
683 copy_extent_buffer(path->nodes[0], eb, dest_offset,
684 (unsigned long)item, sizeof(*item));
686 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
687 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
688 ins.type = BTRFS_EXTENT_ITEM_KEY;
689 offset = key->offset - btrfs_file_extent_offset(eb, item);
692 * Manually record dirty extent, as here we did a shallow
693 * file extent item copy and skip normal backref update,
694 * but modifying extent tree all by ourselves.
695 * So need to manually record dirty extent for qgroup,
696 * as the owner of the file extent changed from log tree
697 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
699 ret = btrfs_qgroup_trace_extent(trans,
700 btrfs_file_extent_disk_bytenr(eb, item),
701 btrfs_file_extent_disk_num_bytes(eb, item),
706 if (ins.objectid > 0) {
707 struct btrfs_ref ref = { 0 };
710 LIST_HEAD(ordered_sums);
713 * is this extent already allocated in the extent
714 * allocation tree? If so, just add a reference
716 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
719 btrfs_init_generic_ref(&ref,
720 BTRFS_ADD_DELAYED_REF,
721 ins.objectid, ins.offset, 0);
722 btrfs_init_data_ref(&ref,
723 root->root_key.objectid,
724 key->objectid, offset);
725 ret = btrfs_inc_extent_ref(trans, &ref);
730 * insert the extent pointer in the extent
733 ret = btrfs_alloc_logged_file_extent(trans,
734 root->root_key.objectid,
735 key->objectid, offset, &ins);
739 btrfs_release_path(path);
741 if (btrfs_file_extent_compression(eb, item)) {
742 csum_start = ins.objectid;
743 csum_end = csum_start + ins.offset;
745 csum_start = ins.objectid +
746 btrfs_file_extent_offset(eb, item);
747 csum_end = csum_start +
748 btrfs_file_extent_num_bytes(eb, item);
751 ret = btrfs_lookup_csums_range(root->log_root,
752 csum_start, csum_end - 1,
757 * Now delete all existing cums in the csum root that
758 * cover our range. We do this because we can have an
759 * extent that is completely referenced by one file
760 * extent item and partially referenced by another
761 * file extent item (like after using the clone or
762 * extent_same ioctls). In this case if we end up doing
763 * the replay of the one that partially references the
764 * extent first, and we do not do the csum deletion
765 * below, we can get 2 csum items in the csum tree that
766 * overlap each other. For example, imagine our log has
767 * the two following file extent items:
769 * key (257 EXTENT_DATA 409600)
770 * extent data disk byte 12845056 nr 102400
771 * extent data offset 20480 nr 20480 ram 102400
773 * key (257 EXTENT_DATA 819200)
774 * extent data disk byte 12845056 nr 102400
775 * extent data offset 0 nr 102400 ram 102400
777 * Where the second one fully references the 100K extent
778 * that starts at disk byte 12845056, and the log tree
779 * has a single csum item that covers the entire range
782 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
784 * After the first file extent item is replayed, the
785 * csum tree gets the following csum item:
787 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
789 * Which covers the 20K sub-range starting at offset 20K
790 * of our extent. Now when we replay the second file
791 * extent item, if we do not delete existing csum items
792 * that cover any of its blocks, we end up getting two
793 * csum items in our csum tree that overlap each other:
795 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
796 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
798 * Which is a problem, because after this anyone trying
799 * to lookup up for the checksum of any block of our
800 * extent starting at an offset of 40K or higher, will
801 * end up looking at the second csum item only, which
802 * does not contain the checksum for any block starting
803 * at offset 40K or higher of our extent.
805 while (!list_empty(&ordered_sums)) {
806 struct btrfs_ordered_sum *sums;
807 sums = list_entry(ordered_sums.next,
808 struct btrfs_ordered_sum,
811 ret = btrfs_del_csums(trans, fs_info,
815 ret = btrfs_csum_file_blocks(trans,
816 fs_info->csum_root, sums);
817 list_del(&sums->list);
823 btrfs_release_path(path);
825 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
826 /* inline extents are easy, we just overwrite them */
827 ret = overwrite_item(trans, root, path, eb, slot, key);
832 inode_add_bytes(inode, nbytes);
834 ret = btrfs_update_inode(trans, root, inode);
842 * when cleaning up conflicts between the directory names in the
843 * subvolume, directory names in the log and directory names in the
844 * inode back references, we may have to unlink inodes from directories.
846 * This is a helper function to do the unlink of a specific directory
849 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
850 struct btrfs_root *root,
851 struct btrfs_path *path,
852 struct btrfs_inode *dir,
853 struct btrfs_dir_item *di)
858 struct extent_buffer *leaf;
859 struct btrfs_key location;
862 leaf = path->nodes[0];
864 btrfs_dir_item_key_to_cpu(leaf, di, &location);
865 name_len = btrfs_dir_name_len(leaf, di);
866 name = kmalloc(name_len, GFP_NOFS);
870 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
871 btrfs_release_path(path);
873 inode = read_one_inode(root, location.objectid);
879 ret = link_to_fixup_dir(trans, root, path, location.objectid);
883 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
888 ret = btrfs_run_delayed_items(trans);
896 * helper function to see if a given name and sequence number found
897 * in an inode back reference are already in a directory and correctly
898 * point to this inode
900 static noinline int inode_in_dir(struct btrfs_root *root,
901 struct btrfs_path *path,
902 u64 dirid, u64 objectid, u64 index,
903 const char *name, int name_len)
905 struct btrfs_dir_item *di;
906 struct btrfs_key location;
909 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
910 index, name, name_len, 0);
911 if (di && !IS_ERR(di)) {
912 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
913 if (location.objectid != objectid)
917 btrfs_release_path(path);
919 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
920 if (di && !IS_ERR(di)) {
921 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
922 if (location.objectid != objectid)
928 btrfs_release_path(path);
933 * helper function to check a log tree for a named back reference in
934 * an inode. This is used to decide if a back reference that is
935 * found in the subvolume conflicts with what we find in the log.
937 * inode backreferences may have multiple refs in a single item,
938 * during replay we process one reference at a time, and we don't
939 * want to delete valid links to a file from the subvolume if that
940 * link is also in the log.
942 static noinline int backref_in_log(struct btrfs_root *log,
943 struct btrfs_key *key,
945 const char *name, int namelen)
947 struct btrfs_path *path;
950 path = btrfs_alloc_path();
954 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
957 } else if (ret == 1) {
962 if (key->type == BTRFS_INODE_EXTREF_KEY)
963 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
968 ret = !!btrfs_find_name_in_backref(path->nodes[0],
972 btrfs_free_path(path);
976 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
977 struct btrfs_root *root,
978 struct btrfs_path *path,
979 struct btrfs_root *log_root,
980 struct btrfs_inode *dir,
981 struct btrfs_inode *inode,
982 u64 inode_objectid, u64 parent_objectid,
983 u64 ref_index, char *name, int namelen,
989 struct extent_buffer *leaf;
990 struct btrfs_dir_item *di;
991 struct btrfs_key search_key;
992 struct btrfs_inode_extref *extref;
995 /* Search old style refs */
996 search_key.objectid = inode_objectid;
997 search_key.type = BTRFS_INODE_REF_KEY;
998 search_key.offset = parent_objectid;
999 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1001 struct btrfs_inode_ref *victim_ref;
1003 unsigned long ptr_end;
1005 leaf = path->nodes[0];
1007 /* are we trying to overwrite a back ref for the root directory
1008 * if so, just jump out, we're done
1010 if (search_key.objectid == search_key.offset)
1013 /* check all the names in this back reference to see
1014 * if they are in the log. if so, we allow them to stay
1015 * otherwise they must be unlinked as a conflict
1017 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1018 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1019 while (ptr < ptr_end) {
1020 victim_ref = (struct btrfs_inode_ref *)ptr;
1021 victim_name_len = btrfs_inode_ref_name_len(leaf,
1023 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1027 read_extent_buffer(leaf, victim_name,
1028 (unsigned long)(victim_ref + 1),
1031 ret = backref_in_log(log_root, &search_key,
1032 parent_objectid, victim_name,
1038 inc_nlink(&inode->vfs_inode);
1039 btrfs_release_path(path);
1041 ret = btrfs_unlink_inode(trans, root, dir, inode,
1042 victim_name, victim_name_len);
1046 ret = btrfs_run_delayed_items(trans);
1054 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1058 * NOTE: we have searched root tree and checked the
1059 * corresponding ref, it does not need to check again.
1063 btrfs_release_path(path);
1065 /* Same search but for extended refs */
1066 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1067 inode_objectid, parent_objectid, 0,
1069 if (!IS_ERR_OR_NULL(extref)) {
1073 struct inode *victim_parent;
1075 leaf = path->nodes[0];
1077 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1078 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1080 while (cur_offset < item_size) {
1081 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1083 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1085 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1088 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1091 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1094 search_key.objectid = inode_objectid;
1095 search_key.type = BTRFS_INODE_EXTREF_KEY;
1096 search_key.offset = btrfs_extref_hash(parent_objectid,
1099 ret = backref_in_log(log_root, &search_key,
1100 parent_objectid, victim_name,
1106 victim_parent = read_one_inode(root,
1108 if (victim_parent) {
1109 inc_nlink(&inode->vfs_inode);
1110 btrfs_release_path(path);
1112 ret = btrfs_unlink_inode(trans, root,
1113 BTRFS_I(victim_parent),
1118 ret = btrfs_run_delayed_items(
1121 iput(victim_parent);
1130 cur_offset += victim_name_len + sizeof(*extref);
1134 btrfs_release_path(path);
1136 /* look for a conflicting sequence number */
1137 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1138 ref_index, name, namelen, 0);
1139 if (di && !IS_ERR(di)) {
1140 ret = drop_one_dir_item(trans, root, path, dir, di);
1144 btrfs_release_path(path);
1146 /* look for a conflicting name */
1147 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1149 if (di && !IS_ERR(di)) {
1150 ret = drop_one_dir_item(trans, root, path, dir, di);
1154 btrfs_release_path(path);
1159 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1160 u32 *namelen, char **name, u64 *index,
1161 u64 *parent_objectid)
1163 struct btrfs_inode_extref *extref;
1165 extref = (struct btrfs_inode_extref *)ref_ptr;
1167 *namelen = btrfs_inode_extref_name_len(eb, extref);
1168 *name = kmalloc(*namelen, GFP_NOFS);
1172 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1176 *index = btrfs_inode_extref_index(eb, extref);
1177 if (parent_objectid)
1178 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1183 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1184 u32 *namelen, char **name, u64 *index)
1186 struct btrfs_inode_ref *ref;
1188 ref = (struct btrfs_inode_ref *)ref_ptr;
1190 *namelen = btrfs_inode_ref_name_len(eb, ref);
1191 *name = kmalloc(*namelen, GFP_NOFS);
1195 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1198 *index = btrfs_inode_ref_index(eb, ref);
1204 * Take an inode reference item from the log tree and iterate all names from the
1205 * inode reference item in the subvolume tree with the same key (if it exists).
1206 * For any name that is not in the inode reference item from the log tree, do a
1207 * proper unlink of that name (that is, remove its entry from the inode
1208 * reference item and both dir index keys).
1210 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1211 struct btrfs_root *root,
1212 struct btrfs_path *path,
1213 struct btrfs_inode *inode,
1214 struct extent_buffer *log_eb,
1216 struct btrfs_key *key)
1219 unsigned long ref_ptr;
1220 unsigned long ref_end;
1221 struct extent_buffer *eb;
1224 btrfs_release_path(path);
1225 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1233 eb = path->nodes[0];
1234 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1235 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1236 while (ref_ptr < ref_end) {
1241 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1242 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1245 parent_id = key->offset;
1246 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1252 if (key->type == BTRFS_INODE_EXTREF_KEY)
1253 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1257 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1263 btrfs_release_path(path);
1264 dir = read_one_inode(root, parent_id);
1270 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1271 inode, name, namelen);
1281 if (key->type == BTRFS_INODE_EXTREF_KEY)
1282 ref_ptr += sizeof(struct btrfs_inode_extref);
1284 ref_ptr += sizeof(struct btrfs_inode_ref);
1288 btrfs_release_path(path);
1292 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1293 const u8 ref_type, const char *name,
1296 struct btrfs_key key;
1297 struct btrfs_path *path;
1298 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1301 path = btrfs_alloc_path();
1305 key.objectid = btrfs_ino(BTRFS_I(inode));
1306 key.type = ref_type;
1307 if (key.type == BTRFS_INODE_REF_KEY)
1308 key.offset = parent_id;
1310 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1312 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1319 if (key.type == BTRFS_INODE_EXTREF_KEY)
1320 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1321 path->slots[0], parent_id, name, namelen);
1323 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1327 btrfs_free_path(path);
1331 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1332 struct inode *dir, struct inode *inode, const char *name,
1333 int namelen, u64 ref_index)
1335 struct btrfs_dir_item *dir_item;
1336 struct btrfs_key key;
1337 struct btrfs_path *path;
1338 struct inode *other_inode = NULL;
1341 path = btrfs_alloc_path();
1345 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1346 btrfs_ino(BTRFS_I(dir)),
1349 btrfs_release_path(path);
1351 } else if (IS_ERR(dir_item)) {
1352 ret = PTR_ERR(dir_item);
1357 * Our inode's dentry collides with the dentry of another inode which is
1358 * in the log but not yet processed since it has a higher inode number.
1359 * So delete that other dentry.
1361 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1362 btrfs_release_path(path);
1363 other_inode = read_one_inode(root, key.objectid);
1368 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1373 * If we dropped the link count to 0, bump it so that later the iput()
1374 * on the inode will not free it. We will fixup the link count later.
1376 if (other_inode->i_nlink == 0)
1377 inc_nlink(other_inode);
1379 ret = btrfs_run_delayed_items(trans);
1383 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1384 name, namelen, 0, ref_index);
1387 btrfs_free_path(path);
1393 * replay one inode back reference item found in the log tree.
1394 * eb, slot and key refer to the buffer and key found in the log tree.
1395 * root is the destination we are replaying into, and path is for temp
1396 * use by this function. (it should be released on return).
1398 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1399 struct btrfs_root *root,
1400 struct btrfs_root *log,
1401 struct btrfs_path *path,
1402 struct extent_buffer *eb, int slot,
1403 struct btrfs_key *key)
1405 struct inode *dir = NULL;
1406 struct inode *inode = NULL;
1407 unsigned long ref_ptr;
1408 unsigned long ref_end;
1412 int search_done = 0;
1413 int log_ref_ver = 0;
1414 u64 parent_objectid;
1417 int ref_struct_size;
1419 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1420 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1422 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1423 struct btrfs_inode_extref *r;
1425 ref_struct_size = sizeof(struct btrfs_inode_extref);
1427 r = (struct btrfs_inode_extref *)ref_ptr;
1428 parent_objectid = btrfs_inode_extref_parent(eb, r);
1430 ref_struct_size = sizeof(struct btrfs_inode_ref);
1431 parent_objectid = key->offset;
1433 inode_objectid = key->objectid;
1436 * it is possible that we didn't log all the parent directories
1437 * for a given inode. If we don't find the dir, just don't
1438 * copy the back ref in. The link count fixup code will take
1441 dir = read_one_inode(root, parent_objectid);
1447 inode = read_one_inode(root, inode_objectid);
1453 while (ref_ptr < ref_end) {
1455 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1456 &ref_index, &parent_objectid);
1458 * parent object can change from one array
1462 dir = read_one_inode(root, parent_objectid);
1468 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1474 /* if we already have a perfect match, we're done */
1475 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1476 btrfs_ino(BTRFS_I(inode)), ref_index,
1479 * look for a conflicting back reference in the
1480 * metadata. if we find one we have to unlink that name
1481 * of the file before we add our new link. Later on, we
1482 * overwrite any existing back reference, and we don't
1483 * want to create dangling pointers in the directory.
1487 ret = __add_inode_ref(trans, root, path, log,
1492 ref_index, name, namelen,
1502 * If a reference item already exists for this inode
1503 * with the same parent and name, but different index,
1504 * drop it and the corresponding directory index entries
1505 * from the parent before adding the new reference item
1506 * and dir index entries, otherwise we would fail with
1507 * -EEXIST returned from btrfs_add_link() below.
1509 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1512 ret = btrfs_unlink_inode(trans, root,
1517 * If we dropped the link count to 0, bump it so
1518 * that later the iput() on the inode will not
1519 * free it. We will fixup the link count later.
1521 if (!ret && inode->i_nlink == 0)
1527 /* insert our name */
1528 ret = add_link(trans, root, dir, inode, name, namelen,
1533 btrfs_update_inode(trans, root, inode);
1536 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1546 * Before we overwrite the inode reference item in the subvolume tree
1547 * with the item from the log tree, we must unlink all names from the
1548 * parent directory that are in the subvolume's tree inode reference
1549 * item, otherwise we end up with an inconsistent subvolume tree where
1550 * dir index entries exist for a name but there is no inode reference
1551 * item with the same name.
1553 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1558 /* finally write the back reference in the inode */
1559 ret = overwrite_item(trans, root, path, eb, slot, key);
1561 btrfs_release_path(path);
1568 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1569 struct btrfs_root *root, u64 ino)
1573 ret = btrfs_insert_orphan_item(trans, root, ino);
1580 static int count_inode_extrefs(struct btrfs_root *root,
1581 struct btrfs_inode *inode, struct btrfs_path *path)
1585 unsigned int nlink = 0;
1588 u64 inode_objectid = btrfs_ino(inode);
1591 struct btrfs_inode_extref *extref;
1592 struct extent_buffer *leaf;
1595 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1600 leaf = path->nodes[0];
1601 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1602 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1605 while (cur_offset < item_size) {
1606 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1607 name_len = btrfs_inode_extref_name_len(leaf, extref);
1611 cur_offset += name_len + sizeof(*extref);
1615 btrfs_release_path(path);
1617 btrfs_release_path(path);
1619 if (ret < 0 && ret != -ENOENT)
1624 static int count_inode_refs(struct btrfs_root *root,
1625 struct btrfs_inode *inode, struct btrfs_path *path)
1628 struct btrfs_key key;
1629 unsigned int nlink = 0;
1631 unsigned long ptr_end;
1633 u64 ino = btrfs_ino(inode);
1636 key.type = BTRFS_INODE_REF_KEY;
1637 key.offset = (u64)-1;
1640 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1644 if (path->slots[0] == 0)
1649 btrfs_item_key_to_cpu(path->nodes[0], &key,
1651 if (key.objectid != ino ||
1652 key.type != BTRFS_INODE_REF_KEY)
1654 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1655 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1657 while (ptr < ptr_end) {
1658 struct btrfs_inode_ref *ref;
1660 ref = (struct btrfs_inode_ref *)ptr;
1661 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1663 ptr = (unsigned long)(ref + 1) + name_len;
1667 if (key.offset == 0)
1669 if (path->slots[0] > 0) {
1674 btrfs_release_path(path);
1676 btrfs_release_path(path);
1682 * There are a few corners where the link count of the file can't
1683 * be properly maintained during replay. So, instead of adding
1684 * lots of complexity to the log code, we just scan the backrefs
1685 * for any file that has been through replay.
1687 * The scan will update the link count on the inode to reflect the
1688 * number of back refs found. If it goes down to zero, the iput
1689 * will free the inode.
1691 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1692 struct btrfs_root *root,
1693 struct inode *inode)
1695 struct btrfs_path *path;
1698 u64 ino = btrfs_ino(BTRFS_I(inode));
1700 path = btrfs_alloc_path();
1704 ret = count_inode_refs(root, BTRFS_I(inode), path);
1710 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1718 if (nlink != inode->i_nlink) {
1719 set_nlink(inode, nlink);
1720 btrfs_update_inode(trans, root, inode);
1722 BTRFS_I(inode)->index_cnt = (u64)-1;
1724 if (inode->i_nlink == 0) {
1725 if (S_ISDIR(inode->i_mode)) {
1726 ret = replay_dir_deletes(trans, root, NULL, path,
1731 ret = insert_orphan_item(trans, root, ino);
1735 btrfs_free_path(path);
1739 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1740 struct btrfs_root *root,
1741 struct btrfs_path *path)
1744 struct btrfs_key key;
1745 struct inode *inode;
1747 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1748 key.type = BTRFS_ORPHAN_ITEM_KEY;
1749 key.offset = (u64)-1;
1751 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1756 if (path->slots[0] == 0)
1761 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1762 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1763 key.type != BTRFS_ORPHAN_ITEM_KEY)
1766 ret = btrfs_del_item(trans, root, path);
1770 btrfs_release_path(path);
1771 inode = read_one_inode(root, key.offset);
1775 ret = fixup_inode_link_count(trans, root, inode);
1781 * fixup on a directory may create new entries,
1782 * make sure we always look for the highset possible
1785 key.offset = (u64)-1;
1789 btrfs_release_path(path);
1795 * record a given inode in the fixup dir so we can check its link
1796 * count when replay is done. The link count is incremented here
1797 * so the inode won't go away until we check it
1799 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1800 struct btrfs_root *root,
1801 struct btrfs_path *path,
1804 struct btrfs_key key;
1806 struct inode *inode;
1808 inode = read_one_inode(root, objectid);
1812 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1813 key.type = BTRFS_ORPHAN_ITEM_KEY;
1814 key.offset = objectid;
1816 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1818 btrfs_release_path(path);
1820 if (!inode->i_nlink)
1821 set_nlink(inode, 1);
1824 ret = btrfs_update_inode(trans, root, inode);
1825 } else if (ret == -EEXIST) {
1828 BUG(); /* Logic Error */
1836 * when replaying the log for a directory, we only insert names
1837 * for inodes that actually exist. This means an fsync on a directory
1838 * does not implicitly fsync all the new files in it
1840 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1841 struct btrfs_root *root,
1842 u64 dirid, u64 index,
1843 char *name, int name_len,
1844 struct btrfs_key *location)
1846 struct inode *inode;
1850 inode = read_one_inode(root, location->objectid);
1854 dir = read_one_inode(root, dirid);
1860 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1861 name_len, 1, index);
1863 /* FIXME, put inode into FIXUP list */
1871 * take a single entry in a log directory item and replay it into
1874 * if a conflicting item exists in the subdirectory already,
1875 * the inode it points to is unlinked and put into the link count
1878 * If a name from the log points to a file or directory that does
1879 * not exist in the FS, it is skipped. fsyncs on directories
1880 * do not force down inodes inside that directory, just changes to the
1881 * names or unlinks in a directory.
1883 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1884 * non-existing inode) and 1 if the name was replayed.
1886 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1887 struct btrfs_root *root,
1888 struct btrfs_path *path,
1889 struct extent_buffer *eb,
1890 struct btrfs_dir_item *di,
1891 struct btrfs_key *key)
1895 struct btrfs_dir_item *dst_di;
1896 struct btrfs_key found_key;
1897 struct btrfs_key log_key;
1902 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1903 bool name_added = false;
1905 dir = read_one_inode(root, key->objectid);
1909 name_len = btrfs_dir_name_len(eb, di);
1910 name = kmalloc(name_len, GFP_NOFS);
1916 log_type = btrfs_dir_type(eb, di);
1917 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1920 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1921 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1926 btrfs_release_path(path);
1928 if (key->type == BTRFS_DIR_ITEM_KEY) {
1929 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1931 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1932 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1941 if (IS_ERR_OR_NULL(dst_di)) {
1942 /* we need a sequence number to insert, so we only
1943 * do inserts for the BTRFS_DIR_INDEX_KEY types
1945 if (key->type != BTRFS_DIR_INDEX_KEY)
1950 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1951 /* the existing item matches the logged item */
1952 if (found_key.objectid == log_key.objectid &&
1953 found_key.type == log_key.type &&
1954 found_key.offset == log_key.offset &&
1955 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1956 update_size = false;
1961 * don't drop the conflicting directory entry if the inode
1962 * for the new entry doesn't exist
1967 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1971 if (key->type == BTRFS_DIR_INDEX_KEY)
1974 btrfs_release_path(path);
1975 if (!ret && update_size) {
1976 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1977 ret = btrfs_update_inode(trans, root, dir);
1981 if (!ret && name_added)
1987 * Check if the inode reference exists in the log for the given name,
1988 * inode and parent inode
1990 found_key.objectid = log_key.objectid;
1991 found_key.type = BTRFS_INODE_REF_KEY;
1992 found_key.offset = key->objectid;
1993 ret = backref_in_log(root->log_root, &found_key, 0, name, name_len);
1997 /* The dentry will be added later. */
1999 update_size = false;
2003 found_key.objectid = log_key.objectid;
2004 found_key.type = BTRFS_INODE_EXTREF_KEY;
2005 found_key.offset = key->objectid;
2006 ret = backref_in_log(root->log_root, &found_key, key->objectid, name,
2011 /* The dentry will be added later. */
2013 update_size = false;
2016 btrfs_release_path(path);
2017 ret = insert_one_name(trans, root, key->objectid, key->offset,
2018 name, name_len, &log_key);
2019 if (ret && ret != -ENOENT && ret != -EEXIST)
2023 update_size = false;
2029 * find all the names in a directory item and reconcile them into
2030 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2031 * one name in a directory item, but the same code gets used for
2032 * both directory index types
2034 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2035 struct btrfs_root *root,
2036 struct btrfs_path *path,
2037 struct extent_buffer *eb, int slot,
2038 struct btrfs_key *key)
2041 u32 item_size = btrfs_item_size_nr(eb, slot);
2042 struct btrfs_dir_item *di;
2045 unsigned long ptr_end;
2046 struct btrfs_path *fixup_path = NULL;
2048 ptr = btrfs_item_ptr_offset(eb, slot);
2049 ptr_end = ptr + item_size;
2050 while (ptr < ptr_end) {
2051 di = (struct btrfs_dir_item *)ptr;
2052 name_len = btrfs_dir_name_len(eb, di);
2053 ret = replay_one_name(trans, root, path, eb, di, key);
2056 ptr = (unsigned long)(di + 1);
2060 * If this entry refers to a non-directory (directories can not
2061 * have a link count > 1) and it was added in the transaction
2062 * that was not committed, make sure we fixup the link count of
2063 * the inode it the entry points to. Otherwise something like
2064 * the following would result in a directory pointing to an
2065 * inode with a wrong link that does not account for this dir
2073 * ln testdir/bar testdir/bar_link
2074 * ln testdir/foo testdir/foo_link
2075 * xfs_io -c "fsync" testdir/bar
2079 * mount fs, log replay happens
2081 * File foo would remain with a link count of 1 when it has two
2082 * entries pointing to it in the directory testdir. This would
2083 * make it impossible to ever delete the parent directory has
2084 * it would result in stale dentries that can never be deleted.
2086 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2087 struct btrfs_key di_key;
2090 fixup_path = btrfs_alloc_path();
2097 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2098 ret = link_to_fixup_dir(trans, root, fixup_path,
2105 btrfs_free_path(fixup_path);
2110 * directory replay has two parts. There are the standard directory
2111 * items in the log copied from the subvolume, and range items
2112 * created in the log while the subvolume was logged.
2114 * The range items tell us which parts of the key space the log
2115 * is authoritative for. During replay, if a key in the subvolume
2116 * directory is in a logged range item, but not actually in the log
2117 * that means it was deleted from the directory before the fsync
2118 * and should be removed.
2120 static noinline int find_dir_range(struct btrfs_root *root,
2121 struct btrfs_path *path,
2122 u64 dirid, int key_type,
2123 u64 *start_ret, u64 *end_ret)
2125 struct btrfs_key key;
2127 struct btrfs_dir_log_item *item;
2131 if (*start_ret == (u64)-1)
2134 key.objectid = dirid;
2135 key.type = key_type;
2136 key.offset = *start_ret;
2138 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2142 if (path->slots[0] == 0)
2147 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2149 if (key.type != key_type || key.objectid != dirid) {
2153 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2154 struct btrfs_dir_log_item);
2155 found_end = btrfs_dir_log_end(path->nodes[0], item);
2157 if (*start_ret >= key.offset && *start_ret <= found_end) {
2159 *start_ret = key.offset;
2160 *end_ret = found_end;
2165 /* check the next slot in the tree to see if it is a valid item */
2166 nritems = btrfs_header_nritems(path->nodes[0]);
2168 if (path->slots[0] >= nritems) {
2169 ret = btrfs_next_leaf(root, path);
2174 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2176 if (key.type != key_type || key.objectid != dirid) {
2180 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2181 struct btrfs_dir_log_item);
2182 found_end = btrfs_dir_log_end(path->nodes[0], item);
2183 *start_ret = key.offset;
2184 *end_ret = found_end;
2187 btrfs_release_path(path);
2192 * this looks for a given directory item in the log. If the directory
2193 * item is not in the log, the item is removed and the inode it points
2196 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 struct btrfs_root *log,
2199 struct btrfs_path *path,
2200 struct btrfs_path *log_path,
2202 struct btrfs_key *dir_key)
2205 struct extent_buffer *eb;
2208 struct btrfs_dir_item *di;
2209 struct btrfs_dir_item *log_di;
2212 unsigned long ptr_end;
2214 struct inode *inode;
2215 struct btrfs_key location;
2218 eb = path->nodes[0];
2219 slot = path->slots[0];
2220 item_size = btrfs_item_size_nr(eb, slot);
2221 ptr = btrfs_item_ptr_offset(eb, slot);
2222 ptr_end = ptr + item_size;
2223 while (ptr < ptr_end) {
2224 di = (struct btrfs_dir_item *)ptr;
2225 name_len = btrfs_dir_name_len(eb, di);
2226 name = kmalloc(name_len, GFP_NOFS);
2231 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2234 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2235 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2238 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2239 log_di = btrfs_lookup_dir_index_item(trans, log,
2245 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2246 btrfs_dir_item_key_to_cpu(eb, di, &location);
2247 btrfs_release_path(path);
2248 btrfs_release_path(log_path);
2249 inode = read_one_inode(root, location.objectid);
2255 ret = link_to_fixup_dir(trans, root,
2256 path, location.objectid);
2264 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2265 BTRFS_I(inode), name, name_len);
2267 ret = btrfs_run_delayed_items(trans);
2273 /* there might still be more names under this key
2274 * check and repeat if required
2276 ret = btrfs_search_slot(NULL, root, dir_key, path,
2282 } else if (IS_ERR(log_di)) {
2284 return PTR_ERR(log_di);
2286 btrfs_release_path(log_path);
2289 ptr = (unsigned long)(di + 1);
2294 btrfs_release_path(path);
2295 btrfs_release_path(log_path);
2299 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2300 struct btrfs_root *root,
2301 struct btrfs_root *log,
2302 struct btrfs_path *path,
2305 struct btrfs_key search_key;
2306 struct btrfs_path *log_path;
2311 log_path = btrfs_alloc_path();
2315 search_key.objectid = ino;
2316 search_key.type = BTRFS_XATTR_ITEM_KEY;
2317 search_key.offset = 0;
2319 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2323 nritems = btrfs_header_nritems(path->nodes[0]);
2324 for (i = path->slots[0]; i < nritems; i++) {
2325 struct btrfs_key key;
2326 struct btrfs_dir_item *di;
2327 struct btrfs_dir_item *log_di;
2331 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2332 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2337 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2338 total_size = btrfs_item_size_nr(path->nodes[0], i);
2340 while (cur < total_size) {
2341 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2342 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2343 u32 this_len = sizeof(*di) + name_len + data_len;
2346 name = kmalloc(name_len, GFP_NOFS);
2351 read_extent_buffer(path->nodes[0], name,
2352 (unsigned long)(di + 1), name_len);
2354 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2356 btrfs_release_path(log_path);
2358 /* Doesn't exist in log tree, so delete it. */
2359 btrfs_release_path(path);
2360 di = btrfs_lookup_xattr(trans, root, path, ino,
2361 name, name_len, -1);
2368 ret = btrfs_delete_one_dir_name(trans, root,
2372 btrfs_release_path(path);
2377 if (IS_ERR(log_di)) {
2378 ret = PTR_ERR(log_di);
2382 di = (struct btrfs_dir_item *)((char *)di + this_len);
2385 ret = btrfs_next_leaf(root, path);
2391 btrfs_free_path(log_path);
2392 btrfs_release_path(path);
2398 * deletion replay happens before we copy any new directory items
2399 * out of the log or out of backreferences from inodes. It
2400 * scans the log to find ranges of keys that log is authoritative for,
2401 * and then scans the directory to find items in those ranges that are
2402 * not present in the log.
2404 * Anything we don't find in the log is unlinked and removed from the
2407 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2408 struct btrfs_root *root,
2409 struct btrfs_root *log,
2410 struct btrfs_path *path,
2411 u64 dirid, int del_all)
2415 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2417 struct btrfs_key dir_key;
2418 struct btrfs_key found_key;
2419 struct btrfs_path *log_path;
2422 dir_key.objectid = dirid;
2423 dir_key.type = BTRFS_DIR_ITEM_KEY;
2424 log_path = btrfs_alloc_path();
2428 dir = read_one_inode(root, dirid);
2429 /* it isn't an error if the inode isn't there, that can happen
2430 * because we replay the deletes before we copy in the inode item
2434 btrfs_free_path(log_path);
2442 range_end = (u64)-1;
2444 ret = find_dir_range(log, path, dirid, key_type,
2445 &range_start, &range_end);
2450 dir_key.offset = range_start;
2453 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2458 nritems = btrfs_header_nritems(path->nodes[0]);
2459 if (path->slots[0] >= nritems) {
2460 ret = btrfs_next_leaf(root, path);
2466 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2468 if (found_key.objectid != dirid ||
2469 found_key.type != dir_key.type)
2472 if (found_key.offset > range_end)
2475 ret = check_item_in_log(trans, root, log, path,
2480 if (found_key.offset == (u64)-1)
2482 dir_key.offset = found_key.offset + 1;
2484 btrfs_release_path(path);
2485 if (range_end == (u64)-1)
2487 range_start = range_end + 1;
2492 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2493 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2494 dir_key.type = BTRFS_DIR_INDEX_KEY;
2495 btrfs_release_path(path);
2499 btrfs_release_path(path);
2500 btrfs_free_path(log_path);
2506 * the process_func used to replay items from the log tree. This
2507 * gets called in two different stages. The first stage just looks
2508 * for inodes and makes sure they are all copied into the subvolume.
2510 * The second stage copies all the other item types from the log into
2511 * the subvolume. The two stage approach is slower, but gets rid of
2512 * lots of complexity around inodes referencing other inodes that exist
2513 * only in the log (references come from either directory items or inode
2516 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2517 struct walk_control *wc, u64 gen, int level)
2520 struct btrfs_path *path;
2521 struct btrfs_root *root = wc->replay_dest;
2522 struct btrfs_key key;
2526 ret = btrfs_read_buffer(eb, gen, level, NULL);
2530 level = btrfs_header_level(eb);
2535 path = btrfs_alloc_path();
2539 nritems = btrfs_header_nritems(eb);
2540 for (i = 0; i < nritems; i++) {
2541 btrfs_item_key_to_cpu(eb, &key, i);
2543 /* inode keys are done during the first stage */
2544 if (key.type == BTRFS_INODE_ITEM_KEY &&
2545 wc->stage == LOG_WALK_REPLAY_INODES) {
2546 struct btrfs_inode_item *inode_item;
2549 inode_item = btrfs_item_ptr(eb, i,
2550 struct btrfs_inode_item);
2552 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2553 * and never got linked before the fsync, skip it, as
2554 * replaying it is pointless since it would be deleted
2555 * later. We skip logging tmpfiles, but it's always
2556 * possible we are replaying a log created with a kernel
2557 * that used to log tmpfiles.
2559 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2560 wc->ignore_cur_inode = true;
2563 wc->ignore_cur_inode = false;
2565 ret = replay_xattr_deletes(wc->trans, root, log,
2566 path, key.objectid);
2569 mode = btrfs_inode_mode(eb, inode_item);
2570 if (S_ISDIR(mode)) {
2571 ret = replay_dir_deletes(wc->trans,
2572 root, log, path, key.objectid, 0);
2576 ret = overwrite_item(wc->trans, root, path,
2582 * Before replaying extents, truncate the inode to its
2583 * size. We need to do it now and not after log replay
2584 * because before an fsync we can have prealloc extents
2585 * added beyond the inode's i_size. If we did it after,
2586 * through orphan cleanup for example, we would drop
2587 * those prealloc extents just after replaying them.
2589 if (S_ISREG(mode)) {
2590 struct inode *inode;
2593 inode = read_one_inode(root, key.objectid);
2598 from = ALIGN(i_size_read(inode),
2599 root->fs_info->sectorsize);
2600 ret = btrfs_drop_extents(wc->trans, root, inode,
2603 /* Update the inode's nbytes. */
2604 ret = btrfs_update_inode(wc->trans,
2612 ret = link_to_fixup_dir(wc->trans, root,
2613 path, key.objectid);
2618 if (wc->ignore_cur_inode)
2621 if (key.type == BTRFS_DIR_INDEX_KEY &&
2622 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2623 ret = replay_one_dir_item(wc->trans, root, path,
2629 if (wc->stage < LOG_WALK_REPLAY_ALL)
2632 /* these keys are simply copied */
2633 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2634 ret = overwrite_item(wc->trans, root, path,
2638 } else if (key.type == BTRFS_INODE_REF_KEY ||
2639 key.type == BTRFS_INODE_EXTREF_KEY) {
2640 ret = add_inode_ref(wc->trans, root, log, path,
2642 if (ret && ret != -ENOENT)
2645 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2646 ret = replay_one_extent(wc->trans, root, path,
2650 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2651 ret = replay_one_dir_item(wc->trans, root, path,
2657 btrfs_free_path(path);
2661 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2662 struct btrfs_root *root,
2663 struct btrfs_path *path, int *level,
2664 struct walk_control *wc)
2666 struct btrfs_fs_info *fs_info = root->fs_info;
2670 struct extent_buffer *next;
2671 struct extent_buffer *cur;
2672 struct extent_buffer *parent;
2676 WARN_ON(*level < 0);
2677 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2679 while (*level > 0) {
2680 struct btrfs_key first_key;
2682 WARN_ON(*level < 0);
2683 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2684 cur = path->nodes[*level];
2686 WARN_ON(btrfs_header_level(cur) != *level);
2688 if (path->slots[*level] >=
2689 btrfs_header_nritems(cur))
2692 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2693 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2694 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2695 blocksize = fs_info->nodesize;
2697 parent = path->nodes[*level];
2698 root_owner = btrfs_header_owner(parent);
2700 next = btrfs_find_create_tree_block(fs_info, bytenr);
2702 return PTR_ERR(next);
2705 ret = wc->process_func(root, next, wc, ptr_gen,
2708 free_extent_buffer(next);
2712 path->slots[*level]++;
2714 ret = btrfs_read_buffer(next, ptr_gen,
2715 *level - 1, &first_key);
2717 free_extent_buffer(next);
2722 btrfs_tree_lock(next);
2723 btrfs_set_lock_blocking_write(next);
2724 btrfs_clean_tree_block(next);
2725 btrfs_wait_tree_block_writeback(next);
2726 btrfs_tree_unlock(next);
2728 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2729 clear_extent_buffer_dirty(next);
2732 WARN_ON(root_owner !=
2733 BTRFS_TREE_LOG_OBJECTID);
2734 ret = btrfs_free_and_pin_reserved_extent(
2738 free_extent_buffer(next);
2742 free_extent_buffer(next);
2745 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2747 free_extent_buffer(next);
2751 WARN_ON(*level <= 0);
2752 if (path->nodes[*level-1])
2753 free_extent_buffer(path->nodes[*level-1]);
2754 path->nodes[*level-1] = next;
2755 *level = btrfs_header_level(next);
2756 path->slots[*level] = 0;
2759 WARN_ON(*level < 0);
2760 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2762 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2768 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2769 struct btrfs_root *root,
2770 struct btrfs_path *path, int *level,
2771 struct walk_control *wc)
2773 struct btrfs_fs_info *fs_info = root->fs_info;
2779 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2780 slot = path->slots[i];
2781 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2784 WARN_ON(*level == 0);
2787 struct extent_buffer *parent;
2788 if (path->nodes[*level] == root->node)
2789 parent = path->nodes[*level];
2791 parent = path->nodes[*level + 1];
2793 root_owner = btrfs_header_owner(parent);
2794 ret = wc->process_func(root, path->nodes[*level], wc,
2795 btrfs_header_generation(path->nodes[*level]),
2801 struct extent_buffer *next;
2803 next = path->nodes[*level];
2806 btrfs_tree_lock(next);
2807 btrfs_set_lock_blocking_write(next);
2808 btrfs_clean_tree_block(next);
2809 btrfs_wait_tree_block_writeback(next);
2810 btrfs_tree_unlock(next);
2812 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2813 clear_extent_buffer_dirty(next);
2816 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2817 ret = btrfs_free_and_pin_reserved_extent(
2819 path->nodes[*level]->start,
2820 path->nodes[*level]->len);
2824 free_extent_buffer(path->nodes[*level]);
2825 path->nodes[*level] = NULL;
2833 * drop the reference count on the tree rooted at 'snap'. This traverses
2834 * the tree freeing any blocks that have a ref count of zero after being
2837 static int walk_log_tree(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *log, struct walk_control *wc)
2840 struct btrfs_fs_info *fs_info = log->fs_info;
2844 struct btrfs_path *path;
2847 path = btrfs_alloc_path();
2851 level = btrfs_header_level(log->node);
2853 path->nodes[level] = log->node;
2854 extent_buffer_get(log->node);
2855 path->slots[level] = 0;
2858 wret = walk_down_log_tree(trans, log, path, &level, wc);
2866 wret = walk_up_log_tree(trans, log, path, &level, wc);
2875 /* was the root node processed? if not, catch it here */
2876 if (path->nodes[orig_level]) {
2877 ret = wc->process_func(log, path->nodes[orig_level], wc,
2878 btrfs_header_generation(path->nodes[orig_level]),
2883 struct extent_buffer *next;
2885 next = path->nodes[orig_level];
2888 btrfs_tree_lock(next);
2889 btrfs_set_lock_blocking_write(next);
2890 btrfs_clean_tree_block(next);
2891 btrfs_wait_tree_block_writeback(next);
2892 btrfs_tree_unlock(next);
2894 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2895 clear_extent_buffer_dirty(next);
2898 WARN_ON(log->root_key.objectid !=
2899 BTRFS_TREE_LOG_OBJECTID);
2900 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2901 next->start, next->len);
2908 btrfs_free_path(path);
2913 * helper function to update the item for a given subvolumes log root
2914 * in the tree of log roots
2916 static int update_log_root(struct btrfs_trans_handle *trans,
2917 struct btrfs_root *log,
2918 struct btrfs_root_item *root_item)
2920 struct btrfs_fs_info *fs_info = log->fs_info;
2923 if (log->log_transid == 1) {
2924 /* insert root item on the first sync */
2925 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2926 &log->root_key, root_item);
2928 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2929 &log->root_key, root_item);
2934 static void wait_log_commit(struct btrfs_root *root, int transid)
2937 int index = transid % 2;
2940 * we only allow two pending log transactions at a time,
2941 * so we know that if ours is more than 2 older than the
2942 * current transaction, we're done
2945 prepare_to_wait(&root->log_commit_wait[index],
2946 &wait, TASK_UNINTERRUPTIBLE);
2948 if (!(root->log_transid_committed < transid &&
2949 atomic_read(&root->log_commit[index])))
2952 mutex_unlock(&root->log_mutex);
2954 mutex_lock(&root->log_mutex);
2956 finish_wait(&root->log_commit_wait[index], &wait);
2959 static void wait_for_writer(struct btrfs_root *root)
2964 prepare_to_wait(&root->log_writer_wait, &wait,
2965 TASK_UNINTERRUPTIBLE);
2966 if (!atomic_read(&root->log_writers))
2969 mutex_unlock(&root->log_mutex);
2971 mutex_lock(&root->log_mutex);
2973 finish_wait(&root->log_writer_wait, &wait);
2976 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2977 struct btrfs_log_ctx *ctx)
2982 mutex_lock(&root->log_mutex);
2983 list_del_init(&ctx->list);
2984 mutex_unlock(&root->log_mutex);
2988 * Invoked in log mutex context, or be sure there is no other task which
2989 * can access the list.
2991 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2992 int index, int error)
2994 struct btrfs_log_ctx *ctx;
2995 struct btrfs_log_ctx *safe;
2997 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2998 list_del_init(&ctx->list);
2999 ctx->log_ret = error;
3002 INIT_LIST_HEAD(&root->log_ctxs[index]);
3006 * btrfs_sync_log does sends a given tree log down to the disk and
3007 * updates the super blocks to record it. When this call is done,
3008 * you know that any inodes previously logged are safely on disk only
3011 * Any other return value means you need to call btrfs_commit_transaction.
3012 * Some of the edge cases for fsyncing directories that have had unlinks
3013 * or renames done in the past mean that sometimes the only safe
3014 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3015 * that has happened.
3017 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3018 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3024 struct btrfs_fs_info *fs_info = root->fs_info;
3025 struct btrfs_root *log = root->log_root;
3026 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3027 struct btrfs_root_item new_root_item;
3028 int log_transid = 0;
3029 struct btrfs_log_ctx root_log_ctx;
3030 struct blk_plug plug;
3032 mutex_lock(&root->log_mutex);
3033 log_transid = ctx->log_transid;
3034 if (root->log_transid_committed >= log_transid) {
3035 mutex_unlock(&root->log_mutex);
3036 return ctx->log_ret;
3039 index1 = log_transid % 2;
3040 if (atomic_read(&root->log_commit[index1])) {
3041 wait_log_commit(root, log_transid);
3042 mutex_unlock(&root->log_mutex);
3043 return ctx->log_ret;
3045 ASSERT(log_transid == root->log_transid);
3046 atomic_set(&root->log_commit[index1], 1);
3048 /* wait for previous tree log sync to complete */
3049 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3050 wait_log_commit(root, log_transid - 1);
3053 int batch = atomic_read(&root->log_batch);
3054 /* when we're on an ssd, just kick the log commit out */
3055 if (!btrfs_test_opt(fs_info, SSD) &&
3056 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3057 mutex_unlock(&root->log_mutex);
3058 schedule_timeout_uninterruptible(1);
3059 mutex_lock(&root->log_mutex);
3061 wait_for_writer(root);
3062 if (batch == atomic_read(&root->log_batch))
3066 /* bail out if we need to do a full commit */
3067 if (btrfs_need_log_full_commit(trans)) {
3069 mutex_unlock(&root->log_mutex);
3073 if (log_transid % 2 == 0)
3074 mark = EXTENT_DIRTY;
3078 /* we start IO on all the marked extents here, but we don't actually
3079 * wait for them until later.
3081 blk_start_plug(&plug);
3082 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3084 blk_finish_plug(&plug);
3085 btrfs_abort_transaction(trans, ret);
3086 btrfs_set_log_full_commit(trans);
3087 mutex_unlock(&root->log_mutex);
3092 * We _must_ update under the root->log_mutex in order to make sure we
3093 * have a consistent view of the log root we are trying to commit at
3096 * We _must_ copy this into a local copy, because we are not holding the
3097 * log_root_tree->log_mutex yet. This is important because when we
3098 * commit the log_root_tree we must have a consistent view of the
3099 * log_root_tree when we update the super block to point at the
3100 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3101 * with the commit and possibly point at the new block which we may not
3104 btrfs_set_root_node(&log->root_item, log->node);
3105 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3107 root->log_transid++;
3108 log->log_transid = root->log_transid;
3109 root->log_start_pid = 0;
3111 * IO has been started, blocks of the log tree have WRITTEN flag set
3112 * in their headers. new modifications of the log will be written to
3113 * new positions. so it's safe to allow log writers to go in.
3115 mutex_unlock(&root->log_mutex);
3117 btrfs_init_log_ctx(&root_log_ctx, NULL);
3119 mutex_lock(&log_root_tree->log_mutex);
3120 atomic_inc(&log_root_tree->log_batch);
3121 atomic_inc(&log_root_tree->log_writers);
3123 index2 = log_root_tree->log_transid % 2;
3124 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3125 root_log_ctx.log_transid = log_root_tree->log_transid;
3127 mutex_unlock(&log_root_tree->log_mutex);
3129 mutex_lock(&log_root_tree->log_mutex);
3132 * Now we are safe to update the log_root_tree because we're under the
3133 * log_mutex, and we're a current writer so we're holding the commit
3134 * open until we drop the log_mutex.
3136 ret = update_log_root(trans, log, &new_root_item);
3138 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3139 /* atomic_dec_and_test implies a barrier */
3140 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3144 if (!list_empty(&root_log_ctx.list))
3145 list_del_init(&root_log_ctx.list);
3147 blk_finish_plug(&plug);
3148 btrfs_set_log_full_commit(trans);
3150 if (ret != -ENOSPC) {
3151 btrfs_abort_transaction(trans, ret);
3152 mutex_unlock(&log_root_tree->log_mutex);
3155 btrfs_wait_tree_log_extents(log, mark);
3156 mutex_unlock(&log_root_tree->log_mutex);
3161 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3162 blk_finish_plug(&plug);
3163 list_del_init(&root_log_ctx.list);
3164 mutex_unlock(&log_root_tree->log_mutex);
3165 ret = root_log_ctx.log_ret;
3169 index2 = root_log_ctx.log_transid % 2;
3170 if (atomic_read(&log_root_tree->log_commit[index2])) {
3171 blk_finish_plug(&plug);
3172 ret = btrfs_wait_tree_log_extents(log, mark);
3173 wait_log_commit(log_root_tree,
3174 root_log_ctx.log_transid);
3175 mutex_unlock(&log_root_tree->log_mutex);
3177 ret = root_log_ctx.log_ret;
3180 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3181 atomic_set(&log_root_tree->log_commit[index2], 1);
3183 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3184 wait_log_commit(log_root_tree,
3185 root_log_ctx.log_transid - 1);
3188 wait_for_writer(log_root_tree);
3191 * now that we've moved on to the tree of log tree roots,
3192 * check the full commit flag again
3194 if (btrfs_need_log_full_commit(trans)) {
3195 blk_finish_plug(&plug);
3196 btrfs_wait_tree_log_extents(log, mark);
3197 mutex_unlock(&log_root_tree->log_mutex);
3199 goto out_wake_log_root;
3202 ret = btrfs_write_marked_extents(fs_info,
3203 &log_root_tree->dirty_log_pages,
3204 EXTENT_DIRTY | EXTENT_NEW);
3205 blk_finish_plug(&plug);
3207 btrfs_set_log_full_commit(trans);
3208 btrfs_abort_transaction(trans, ret);
3209 mutex_unlock(&log_root_tree->log_mutex);
3210 goto out_wake_log_root;
3212 ret = btrfs_wait_tree_log_extents(log, mark);
3214 ret = btrfs_wait_tree_log_extents(log_root_tree,
3215 EXTENT_NEW | EXTENT_DIRTY);
3217 btrfs_set_log_full_commit(trans);
3218 mutex_unlock(&log_root_tree->log_mutex);
3219 goto out_wake_log_root;
3222 btrfs_set_super_log_root(fs_info->super_for_commit,
3223 log_root_tree->node->start);
3224 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3225 btrfs_header_level(log_root_tree->node));
3227 log_root_tree->log_transid++;
3228 mutex_unlock(&log_root_tree->log_mutex);
3231 * Nobody else is going to jump in and write the ctree
3232 * super here because the log_commit atomic below is protecting
3233 * us. We must be called with a transaction handle pinning
3234 * the running transaction open, so a full commit can't hop
3235 * in and cause problems either.
3237 ret = write_all_supers(fs_info, 1);
3239 btrfs_set_log_full_commit(trans);
3240 btrfs_abort_transaction(trans, ret);
3241 goto out_wake_log_root;
3244 mutex_lock(&root->log_mutex);
3245 if (root->last_log_commit < log_transid)
3246 root->last_log_commit = log_transid;
3247 mutex_unlock(&root->log_mutex);
3250 mutex_lock(&log_root_tree->log_mutex);
3251 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3253 log_root_tree->log_transid_committed++;
3254 atomic_set(&log_root_tree->log_commit[index2], 0);
3255 mutex_unlock(&log_root_tree->log_mutex);
3258 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3259 * all the updates above are seen by the woken threads. It might not be
3260 * necessary, but proving that seems to be hard.
3262 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3264 mutex_lock(&root->log_mutex);
3265 btrfs_remove_all_log_ctxs(root, index1, ret);
3266 root->log_transid_committed++;
3267 atomic_set(&root->log_commit[index1], 0);
3268 mutex_unlock(&root->log_mutex);
3271 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3272 * all the updates above are seen by the woken threads. It might not be
3273 * necessary, but proving that seems to be hard.
3275 cond_wake_up(&root->log_commit_wait[index1]);
3279 static void free_log_tree(struct btrfs_trans_handle *trans,
3280 struct btrfs_root *log)
3283 struct walk_control wc = {
3285 .process_func = process_one_buffer
3288 ret = walk_log_tree(trans, log, &wc);
3291 btrfs_abort_transaction(trans, ret);
3293 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3296 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3297 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3298 free_extent_buffer(log->node);
3303 * free all the extents used by the tree log. This should be called
3304 * at commit time of the full transaction
3306 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3308 if (root->log_root) {
3309 free_log_tree(trans, root->log_root);
3310 root->log_root = NULL;
3315 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3316 struct btrfs_fs_info *fs_info)
3318 if (fs_info->log_root_tree) {
3319 free_log_tree(trans, fs_info->log_root_tree);
3320 fs_info->log_root_tree = NULL;
3326 * Check if an inode was logged in the current transaction. We can't always rely
3327 * on an inode's logged_trans value, because it's an in-memory only field and
3328 * therefore not persisted. This means that its value is lost if the inode gets
3329 * evicted and loaded again from disk (in which case it has a value of 0, and
3330 * certainly it is smaller then any possible transaction ID), when that happens
3331 * the full_sync flag is set in the inode's runtime flags, so on that case we
3332 * assume eviction happened and ignore the logged_trans value, assuming the
3333 * worst case, that the inode was logged before in the current transaction.
3335 static bool inode_logged(struct btrfs_trans_handle *trans,
3336 struct btrfs_inode *inode)
3338 if (inode->logged_trans == trans->transid)
3341 if (inode->last_trans == trans->transid &&
3342 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3343 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3350 * If both a file and directory are logged, and unlinks or renames are
3351 * mixed in, we have a few interesting corners:
3353 * create file X in dir Y
3354 * link file X to X.link in dir Y
3356 * unlink file X but leave X.link
3359 * After a crash we would expect only X.link to exist. But file X
3360 * didn't get fsync'd again so the log has back refs for X and X.link.
3362 * We solve this by removing directory entries and inode backrefs from the
3363 * log when a file that was logged in the current transaction is
3364 * unlinked. Any later fsync will include the updated log entries, and
3365 * we'll be able to reconstruct the proper directory items from backrefs.
3367 * This optimizations allows us to avoid relogging the entire inode
3368 * or the entire directory.
3370 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3371 struct btrfs_root *root,
3372 const char *name, int name_len,
3373 struct btrfs_inode *dir, u64 index)
3375 struct btrfs_root *log;
3376 struct btrfs_dir_item *di;
3377 struct btrfs_path *path;
3381 u64 dir_ino = btrfs_ino(dir);
3383 if (!inode_logged(trans, dir))
3386 ret = join_running_log_trans(root);
3390 mutex_lock(&dir->log_mutex);
3392 log = root->log_root;
3393 path = btrfs_alloc_path();
3399 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3400 name, name_len, -1);
3406 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3407 bytes_del += name_len;
3413 btrfs_release_path(path);
3414 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3415 index, name, name_len, -1);
3421 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3422 bytes_del += name_len;
3429 /* update the directory size in the log to reflect the names
3433 struct btrfs_key key;
3435 key.objectid = dir_ino;
3437 key.type = BTRFS_INODE_ITEM_KEY;
3438 btrfs_release_path(path);
3440 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3446 struct btrfs_inode_item *item;
3449 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3450 struct btrfs_inode_item);
3451 i_size = btrfs_inode_size(path->nodes[0], item);
3452 if (i_size > bytes_del)
3453 i_size -= bytes_del;
3456 btrfs_set_inode_size(path->nodes[0], item, i_size);
3457 btrfs_mark_buffer_dirty(path->nodes[0]);
3460 btrfs_release_path(path);
3463 btrfs_free_path(path);
3465 mutex_unlock(&dir->log_mutex);
3466 if (ret == -ENOSPC) {
3467 btrfs_set_log_full_commit(trans);
3470 btrfs_abort_transaction(trans, ret);
3472 btrfs_end_log_trans(root);
3477 /* see comments for btrfs_del_dir_entries_in_log */
3478 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3479 struct btrfs_root *root,
3480 const char *name, int name_len,
3481 struct btrfs_inode *inode, u64 dirid)
3483 struct btrfs_root *log;
3487 if (!inode_logged(trans, inode))
3490 ret = join_running_log_trans(root);
3493 log = root->log_root;
3494 mutex_lock(&inode->log_mutex);
3496 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3498 mutex_unlock(&inode->log_mutex);
3499 if (ret == -ENOSPC) {
3500 btrfs_set_log_full_commit(trans);
3502 } else if (ret < 0 && ret != -ENOENT)
3503 btrfs_abort_transaction(trans, ret);
3504 btrfs_end_log_trans(root);
3510 * creates a range item in the log for 'dirid'. first_offset and
3511 * last_offset tell us which parts of the key space the log should
3512 * be considered authoritative for.
3514 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3515 struct btrfs_root *log,
3516 struct btrfs_path *path,
3517 int key_type, u64 dirid,
3518 u64 first_offset, u64 last_offset)
3521 struct btrfs_key key;
3522 struct btrfs_dir_log_item *item;
3524 key.objectid = dirid;
3525 key.offset = first_offset;
3526 if (key_type == BTRFS_DIR_ITEM_KEY)
3527 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3529 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3530 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3534 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3535 struct btrfs_dir_log_item);
3536 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3537 btrfs_mark_buffer_dirty(path->nodes[0]);
3538 btrfs_release_path(path);
3543 * log all the items included in the current transaction for a given
3544 * directory. This also creates the range items in the log tree required
3545 * to replay anything deleted before the fsync
3547 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3548 struct btrfs_root *root, struct btrfs_inode *inode,
3549 struct btrfs_path *path,
3550 struct btrfs_path *dst_path, int key_type,
3551 struct btrfs_log_ctx *ctx,
3552 u64 min_offset, u64 *last_offset_ret)
3554 struct btrfs_key min_key;
3555 struct btrfs_root *log = root->log_root;
3556 struct extent_buffer *src;
3561 u64 first_offset = min_offset;
3562 u64 last_offset = (u64)-1;
3563 u64 ino = btrfs_ino(inode);
3565 log = root->log_root;
3567 min_key.objectid = ino;
3568 min_key.type = key_type;
3569 min_key.offset = min_offset;
3571 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3574 * we didn't find anything from this transaction, see if there
3575 * is anything at all
3577 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3578 min_key.objectid = ino;
3579 min_key.type = key_type;
3580 min_key.offset = (u64)-1;
3581 btrfs_release_path(path);
3582 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3584 btrfs_release_path(path);
3587 ret = btrfs_previous_item(root, path, ino, key_type);
3589 /* if ret == 0 there are items for this type,
3590 * create a range to tell us the last key of this type.
3591 * otherwise, there are no items in this directory after
3592 * *min_offset, and we create a range to indicate that.
3595 struct btrfs_key tmp;
3596 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3598 if (key_type == tmp.type)
3599 first_offset = max(min_offset, tmp.offset) + 1;
3604 /* go backward to find any previous key */
3605 ret = btrfs_previous_item(root, path, ino, key_type);
3607 struct btrfs_key tmp;
3608 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3609 if (key_type == tmp.type) {
3610 first_offset = tmp.offset;
3611 ret = overwrite_item(trans, log, dst_path,
3612 path->nodes[0], path->slots[0],
3620 btrfs_release_path(path);
3623 * Find the first key from this transaction again. See the note for
3624 * log_new_dir_dentries, if we're logging a directory recursively we
3625 * won't be holding its i_mutex, which means we can modify the directory
3626 * while we're logging it. If we remove an entry between our first
3627 * search and this search we'll not find the key again and can just
3630 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3635 * we have a block from this transaction, log every item in it
3636 * from our directory
3639 struct btrfs_key tmp;
3640 src = path->nodes[0];
3641 nritems = btrfs_header_nritems(src);
3642 for (i = path->slots[0]; i < nritems; i++) {
3643 struct btrfs_dir_item *di;
3645 btrfs_item_key_to_cpu(src, &min_key, i);
3647 if (min_key.objectid != ino || min_key.type != key_type)
3649 ret = overwrite_item(trans, log, dst_path, src, i,
3657 * We must make sure that when we log a directory entry,
3658 * the corresponding inode, after log replay, has a
3659 * matching link count. For example:
3665 * xfs_io -c "fsync" mydir
3667 * <mount fs and log replay>
3669 * Would result in a fsync log that when replayed, our
3670 * file inode would have a link count of 1, but we get
3671 * two directory entries pointing to the same inode.
3672 * After removing one of the names, it would not be
3673 * possible to remove the other name, which resulted
3674 * always in stale file handle errors, and would not
3675 * be possible to rmdir the parent directory, since
3676 * its i_size could never decrement to the value
3677 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3679 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3680 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3682 (btrfs_dir_transid(src, di) == trans->transid ||
3683 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3684 tmp.type != BTRFS_ROOT_ITEM_KEY)
3685 ctx->log_new_dentries = true;
3687 path->slots[0] = nritems;
3690 * look ahead to the next item and see if it is also
3691 * from this directory and from this transaction
3693 ret = btrfs_next_leaf(root, path);
3696 last_offset = (u64)-1;
3701 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3702 if (tmp.objectid != ino || tmp.type != key_type) {
3703 last_offset = (u64)-1;
3706 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3707 ret = overwrite_item(trans, log, dst_path,
3708 path->nodes[0], path->slots[0],
3713 last_offset = tmp.offset;
3718 btrfs_release_path(path);
3719 btrfs_release_path(dst_path);
3722 *last_offset_ret = last_offset;
3724 * insert the log range keys to indicate where the log
3727 ret = insert_dir_log_key(trans, log, path, key_type,
3728 ino, first_offset, last_offset);
3736 * logging directories is very similar to logging inodes, We find all the items
3737 * from the current transaction and write them to the log.
3739 * The recovery code scans the directory in the subvolume, and if it finds a
3740 * key in the range logged that is not present in the log tree, then it means
3741 * that dir entry was unlinked during the transaction.
3743 * In order for that scan to work, we must include one key smaller than
3744 * the smallest logged by this transaction and one key larger than the largest
3745 * key logged by this transaction.
3747 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3748 struct btrfs_root *root, struct btrfs_inode *inode,
3749 struct btrfs_path *path,
3750 struct btrfs_path *dst_path,
3751 struct btrfs_log_ctx *ctx)
3756 int key_type = BTRFS_DIR_ITEM_KEY;
3762 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3763 ctx, min_key, &max_key);
3766 if (max_key == (u64)-1)
3768 min_key = max_key + 1;
3771 if (key_type == BTRFS_DIR_ITEM_KEY) {
3772 key_type = BTRFS_DIR_INDEX_KEY;
3779 * a helper function to drop items from the log before we relog an
3780 * inode. max_key_type indicates the highest item type to remove.
3781 * This cannot be run for file data extents because it does not
3782 * free the extents they point to.
3784 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3785 struct btrfs_root *log,
3786 struct btrfs_path *path,
3787 u64 objectid, int max_key_type)
3790 struct btrfs_key key;
3791 struct btrfs_key found_key;
3794 key.objectid = objectid;
3795 key.type = max_key_type;
3796 key.offset = (u64)-1;
3799 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3800 BUG_ON(ret == 0); /* Logic error */
3804 if (path->slots[0] == 0)
3808 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3811 if (found_key.objectid != objectid)
3814 found_key.offset = 0;
3816 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3821 ret = btrfs_del_items(trans, log, path, start_slot,
3822 path->slots[0] - start_slot + 1);
3824 * If start slot isn't 0 then we don't need to re-search, we've
3825 * found the last guy with the objectid in this tree.
3827 if (ret || start_slot != 0)
3829 btrfs_release_path(path);
3831 btrfs_release_path(path);
3837 static void fill_inode_item(struct btrfs_trans_handle *trans,
3838 struct extent_buffer *leaf,
3839 struct btrfs_inode_item *item,
3840 struct inode *inode, int log_inode_only,
3843 struct btrfs_map_token token;
3845 btrfs_init_map_token(&token, leaf);
3847 if (log_inode_only) {
3848 /* set the generation to zero so the recover code
3849 * can tell the difference between an logging
3850 * just to say 'this inode exists' and a logging
3851 * to say 'update this inode with these values'
3853 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3854 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3856 btrfs_set_token_inode_generation(leaf, item,
3857 BTRFS_I(inode)->generation,
3859 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3862 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3863 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3864 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3865 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3867 btrfs_set_token_timespec_sec(leaf, &item->atime,
3868 inode->i_atime.tv_sec, &token);
3869 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3870 inode->i_atime.tv_nsec, &token);
3872 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3873 inode->i_mtime.tv_sec, &token);
3874 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3875 inode->i_mtime.tv_nsec, &token);
3877 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3878 inode->i_ctime.tv_sec, &token);
3879 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3880 inode->i_ctime.tv_nsec, &token);
3882 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3885 btrfs_set_token_inode_sequence(leaf, item,
3886 inode_peek_iversion(inode), &token);
3887 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3888 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3889 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3890 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3893 static int log_inode_item(struct btrfs_trans_handle *trans,
3894 struct btrfs_root *log, struct btrfs_path *path,
3895 struct btrfs_inode *inode)
3897 struct btrfs_inode_item *inode_item;
3900 ret = btrfs_insert_empty_item(trans, log, path,
3901 &inode->location, sizeof(*inode_item));
3902 if (ret && ret != -EEXIST)
3904 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3905 struct btrfs_inode_item);
3906 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3908 btrfs_release_path(path);
3912 static noinline int copy_items(struct btrfs_trans_handle *trans,
3913 struct btrfs_inode *inode,
3914 struct btrfs_path *dst_path,
3915 struct btrfs_path *src_path, u64 *last_extent,
3916 int start_slot, int nr, int inode_only,
3919 struct btrfs_fs_info *fs_info = trans->fs_info;
3920 unsigned long src_offset;
3921 unsigned long dst_offset;
3922 struct btrfs_root *log = inode->root->log_root;
3923 struct btrfs_file_extent_item *extent;
3924 struct btrfs_inode_item *inode_item;
3925 struct extent_buffer *src = src_path->nodes[0];
3926 struct btrfs_key first_key, last_key, key;
3928 struct btrfs_key *ins_keys;
3932 struct list_head ordered_sums;
3933 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3934 bool has_extents = false;
3935 bool need_find_last_extent = true;
3938 INIT_LIST_HEAD(&ordered_sums);
3940 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3941 nr * sizeof(u32), GFP_NOFS);
3945 first_key.objectid = (u64)-1;
3947 ins_sizes = (u32 *)ins_data;
3948 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3950 for (i = 0; i < nr; i++) {
3951 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3952 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3954 ret = btrfs_insert_empty_items(trans, log, dst_path,
3955 ins_keys, ins_sizes, nr);
3961 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3962 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3963 dst_path->slots[0]);
3965 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3968 last_key = ins_keys[i];
3970 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3971 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3973 struct btrfs_inode_item);
3974 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3976 inode_only == LOG_INODE_EXISTS,
3979 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3980 src_offset, ins_sizes[i]);
3984 * We set need_find_last_extent here in case we know we were
3985 * processing other items and then walk into the first extent in
3986 * the inode. If we don't hit an extent then nothing changes,
3987 * we'll do the last search the next time around.
3989 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3991 if (first_key.objectid == (u64)-1)
3992 first_key = ins_keys[i];
3994 need_find_last_extent = false;
3997 /* take a reference on file data extents so that truncates
3998 * or deletes of this inode don't have to relog the inode
4001 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4004 extent = btrfs_item_ptr(src, start_slot + i,
4005 struct btrfs_file_extent_item);
4007 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4010 found_type = btrfs_file_extent_type(src, extent);
4011 if (found_type == BTRFS_FILE_EXTENT_REG) {
4013 ds = btrfs_file_extent_disk_bytenr(src,
4015 /* ds == 0 is a hole */
4019 dl = btrfs_file_extent_disk_num_bytes(src,
4021 cs = btrfs_file_extent_offset(src, extent);
4022 cl = btrfs_file_extent_num_bytes(src,
4024 if (btrfs_file_extent_compression(src,
4030 ret = btrfs_lookup_csums_range(
4032 ds + cs, ds + cs + cl - 1,
4035 btrfs_release_path(dst_path);
4043 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4044 btrfs_release_path(dst_path);
4048 * we have to do this after the loop above to avoid changing the
4049 * log tree while trying to change the log tree.
4052 while (!list_empty(&ordered_sums)) {
4053 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4054 struct btrfs_ordered_sum,
4057 ret = btrfs_csum_file_blocks(trans, log, sums);
4058 list_del(&sums->list);
4065 if (need_find_last_extent && *last_extent == first_key.offset) {
4067 * We don't have any leafs between our current one and the one
4068 * we processed before that can have file extent items for our
4069 * inode (and have a generation number smaller than our current
4072 need_find_last_extent = false;
4076 * Because we use btrfs_search_forward we could skip leaves that were
4077 * not modified and then assume *last_extent is valid when it really
4078 * isn't. So back up to the previous leaf and read the end of the last
4079 * extent before we go and fill in holes.
4081 if (need_find_last_extent) {
4084 ret = btrfs_prev_leaf(inode->root, src_path);
4089 if (src_path->slots[0])
4090 src_path->slots[0]--;
4091 src = src_path->nodes[0];
4092 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
4093 if (key.objectid != btrfs_ino(inode) ||
4094 key.type != BTRFS_EXTENT_DATA_KEY)
4096 extent = btrfs_item_ptr(src, src_path->slots[0],
4097 struct btrfs_file_extent_item);
4098 if (btrfs_file_extent_type(src, extent) ==
4099 BTRFS_FILE_EXTENT_INLINE) {
4100 len = btrfs_file_extent_ram_bytes(src, extent);
4101 *last_extent = ALIGN(key.offset + len,
4102 fs_info->sectorsize);
4104 len = btrfs_file_extent_num_bytes(src, extent);
4105 *last_extent = key.offset + len;
4109 /* So we did prev_leaf, now we need to move to the next leaf, but a few
4110 * things could have happened
4112 * 1) A merge could have happened, so we could currently be on a leaf
4113 * that holds what we were copying in the first place.
4114 * 2) A split could have happened, and now not all of the items we want
4115 * are on the same leaf.
4117 * So we need to adjust how we search for holes, we need to drop the
4118 * path and re-search for the first extent key we found, and then walk
4119 * forward until we hit the last one we copied.
4121 if (need_find_last_extent) {
4122 /* btrfs_prev_leaf could return 1 without releasing the path */
4123 btrfs_release_path(src_path);
4124 ret = btrfs_search_slot(NULL, inode->root, &first_key,
4129 src = src_path->nodes[0];
4130 i = src_path->slots[0];
4136 * Ok so here we need to go through and fill in any holes we may have
4137 * to make sure that holes are punched for those areas in case they had
4138 * extents previously.
4144 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
4145 ret = btrfs_next_leaf(inode->root, src_path);
4149 src = src_path->nodes[0];
4151 need_find_last_extent = true;
4154 btrfs_item_key_to_cpu(src, &key, i);
4155 if (!btrfs_comp_cpu_keys(&key, &last_key))
4157 if (key.objectid != btrfs_ino(inode) ||
4158 key.type != BTRFS_EXTENT_DATA_KEY) {
4162 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4163 if (btrfs_file_extent_type(src, extent) ==
4164 BTRFS_FILE_EXTENT_INLINE) {
4165 len = btrfs_file_extent_ram_bytes(src, extent);
4166 extent_end = ALIGN(key.offset + len,
4167 fs_info->sectorsize);
4169 len = btrfs_file_extent_num_bytes(src, extent);
4170 extent_end = key.offset + len;
4174 if (*last_extent == key.offset) {
4175 *last_extent = extent_end;
4178 offset = *last_extent;
4179 len = key.offset - *last_extent;
4180 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4181 offset, 0, 0, len, 0, len, 0, 0, 0);
4184 *last_extent = extent_end;
4188 * Check if there is a hole between the last extent found in our leaf
4189 * and the first extent in the next leaf. If there is one, we need to
4190 * log an explicit hole so that at replay time we can punch the hole.
4193 key.objectid == btrfs_ino(inode) &&
4194 key.type == BTRFS_EXTENT_DATA_KEY &&
4195 i == btrfs_header_nritems(src_path->nodes[0])) {
4196 ret = btrfs_next_leaf(inode->root, src_path);
4197 need_find_last_extent = true;
4200 } else if (ret == 0) {
4201 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4202 src_path->slots[0]);
4203 if (key.objectid == btrfs_ino(inode) &&
4204 key.type == BTRFS_EXTENT_DATA_KEY &&
4205 *last_extent < key.offset) {
4206 const u64 len = key.offset - *last_extent;
4208 ret = btrfs_insert_file_extent(trans, log,
4213 *last_extent += len;
4218 * Need to let the callers know we dropped the path so they should
4221 if (!ret && need_find_last_extent)
4226 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4228 struct extent_map *em1, *em2;
4230 em1 = list_entry(a, struct extent_map, list);
4231 em2 = list_entry(b, struct extent_map, list);
4233 if (em1->start < em2->start)
4235 else if (em1->start > em2->start)
4240 static int log_extent_csums(struct btrfs_trans_handle *trans,
4241 struct btrfs_inode *inode,
4242 struct btrfs_root *log_root,
4243 const struct extent_map *em)
4247 LIST_HEAD(ordered_sums);
4250 if (inode->flags & BTRFS_INODE_NODATASUM ||
4251 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4252 em->block_start == EXTENT_MAP_HOLE)
4255 /* If we're compressed we have to save the entire range of csums. */
4256 if (em->compress_type) {
4258 csum_len = max(em->block_len, em->orig_block_len);
4260 csum_offset = em->mod_start - em->start;
4261 csum_len = em->mod_len;
4264 /* block start is already adjusted for the file extent offset. */
4265 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4266 em->block_start + csum_offset,
4267 em->block_start + csum_offset +
4268 csum_len - 1, &ordered_sums, 0);
4272 while (!list_empty(&ordered_sums)) {
4273 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4274 struct btrfs_ordered_sum,
4277 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4278 list_del(&sums->list);
4285 static int log_one_extent(struct btrfs_trans_handle *trans,
4286 struct btrfs_inode *inode, struct btrfs_root *root,
4287 const struct extent_map *em,
4288 struct btrfs_path *path,
4289 struct btrfs_log_ctx *ctx)
4291 struct btrfs_root *log = root->log_root;
4292 struct btrfs_file_extent_item *fi;
4293 struct extent_buffer *leaf;
4294 struct btrfs_map_token token;
4295 struct btrfs_key key;
4296 u64 extent_offset = em->start - em->orig_start;
4299 int extent_inserted = 0;
4301 ret = log_extent_csums(trans, inode, log, em);
4305 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4306 em->start + em->len, NULL, 0, 1,
4307 sizeof(*fi), &extent_inserted);
4311 if (!extent_inserted) {
4312 key.objectid = btrfs_ino(inode);
4313 key.type = BTRFS_EXTENT_DATA_KEY;
4314 key.offset = em->start;
4316 ret = btrfs_insert_empty_item(trans, log, path, &key,
4321 leaf = path->nodes[0];
4322 btrfs_init_map_token(&token, leaf);
4323 fi = btrfs_item_ptr(leaf, path->slots[0],
4324 struct btrfs_file_extent_item);
4326 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4328 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4329 btrfs_set_token_file_extent_type(leaf, fi,
4330 BTRFS_FILE_EXTENT_PREALLOC,
4333 btrfs_set_token_file_extent_type(leaf, fi,
4334 BTRFS_FILE_EXTENT_REG,
4337 block_len = max(em->block_len, em->orig_block_len);
4338 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4339 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4342 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4344 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4345 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4347 extent_offset, &token);
4348 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4351 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4352 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4356 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4357 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4358 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4359 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4361 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4362 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4363 btrfs_mark_buffer_dirty(leaf);
4365 btrfs_release_path(path);
4371 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4372 * lose them after doing a fast fsync and replaying the log. We scan the
4373 * subvolume's root instead of iterating the inode's extent map tree because
4374 * otherwise we can log incorrect extent items based on extent map conversion.
4375 * That can happen due to the fact that extent maps are merged when they
4376 * are not in the extent map tree's list of modified extents.
4378 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4379 struct btrfs_inode *inode,
4380 struct btrfs_path *path)
4382 struct btrfs_root *root = inode->root;
4383 struct btrfs_key key;
4384 const u64 i_size = i_size_read(&inode->vfs_inode);
4385 const u64 ino = btrfs_ino(inode);
4386 struct btrfs_path *dst_path = NULL;
4387 u64 last_extent = (u64)-1;
4392 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4396 key.type = BTRFS_EXTENT_DATA_KEY;
4397 key.offset = i_size;
4398 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4403 struct extent_buffer *leaf = path->nodes[0];
4404 int slot = path->slots[0];
4406 if (slot >= btrfs_header_nritems(leaf)) {
4408 ret = copy_items(trans, inode, dst_path, path,
4409 &last_extent, start_slot,
4415 ret = btrfs_next_leaf(root, path);
4425 btrfs_item_key_to_cpu(leaf, &key, slot);
4426 if (key.objectid > ino)
4428 if (WARN_ON_ONCE(key.objectid < ino) ||
4429 key.type < BTRFS_EXTENT_DATA_KEY ||
4430 key.offset < i_size) {
4434 if (last_extent == (u64)-1) {
4435 last_extent = key.offset;
4437 * Avoid logging extent items logged in past fsync calls
4438 * and leading to duplicate keys in the log tree.
4441 ret = btrfs_truncate_inode_items(trans,
4445 BTRFS_EXTENT_DATA_KEY);
4446 } while (ret == -EAGAIN);
4455 dst_path = btrfs_alloc_path();
4463 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4464 start_slot, ins_nr, 1, 0);
4469 btrfs_release_path(path);
4470 btrfs_free_path(dst_path);
4474 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4475 struct btrfs_root *root,
4476 struct btrfs_inode *inode,
4477 struct btrfs_path *path,
4478 struct btrfs_log_ctx *ctx,
4482 struct extent_map *em, *n;
4483 struct list_head extents;
4484 struct extent_map_tree *tree = &inode->extent_tree;
4489 INIT_LIST_HEAD(&extents);
4491 write_lock(&tree->lock);
4492 test_gen = root->fs_info->last_trans_committed;
4494 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4496 * Skip extents outside our logging range. It's important to do
4497 * it for correctness because if we don't ignore them, we may
4498 * log them before their ordered extent completes, and therefore
4499 * we could log them without logging their respective checksums
4500 * (the checksum items are added to the csum tree at the very
4501 * end of btrfs_finish_ordered_io()). Also leave such extents
4502 * outside of our range in the list, since we may have another
4503 * ranged fsync in the near future that needs them. If an extent
4504 * outside our range corresponds to a hole, log it to avoid
4505 * leaving gaps between extents (fsck will complain when we are
4506 * not using the NO_HOLES feature).
4508 if ((em->start > end || em->start + em->len <= start) &&
4509 em->block_start != EXTENT_MAP_HOLE)
4512 list_del_init(&em->list);
4514 * Just an arbitrary number, this can be really CPU intensive
4515 * once we start getting a lot of extents, and really once we
4516 * have a bunch of extents we just want to commit since it will
4519 if (++num > 32768) {
4520 list_del_init(&tree->modified_extents);
4525 if (em->generation <= test_gen)
4528 /* We log prealloc extents beyond eof later. */
4529 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4530 em->start >= i_size_read(&inode->vfs_inode))
4533 /* Need a ref to keep it from getting evicted from cache */
4534 refcount_inc(&em->refs);
4535 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4536 list_add_tail(&em->list, &extents);
4540 list_sort(NULL, &extents, extent_cmp);
4542 while (!list_empty(&extents)) {
4543 em = list_entry(extents.next, struct extent_map, list);
4545 list_del_init(&em->list);
4548 * If we had an error we just need to delete everybody from our
4552 clear_em_logging(tree, em);
4553 free_extent_map(em);
4557 write_unlock(&tree->lock);
4559 ret = log_one_extent(trans, inode, root, em, path, ctx);
4560 write_lock(&tree->lock);
4561 clear_em_logging(tree, em);
4562 free_extent_map(em);
4564 WARN_ON(!list_empty(&extents));
4565 write_unlock(&tree->lock);
4567 btrfs_release_path(path);
4569 ret = btrfs_log_prealloc_extents(trans, inode, path);
4574 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4575 struct btrfs_path *path, u64 *size_ret)
4577 struct btrfs_key key;
4580 key.objectid = btrfs_ino(inode);
4581 key.type = BTRFS_INODE_ITEM_KEY;
4584 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4587 } else if (ret > 0) {
4590 struct btrfs_inode_item *item;
4592 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4593 struct btrfs_inode_item);
4594 *size_ret = btrfs_inode_size(path->nodes[0], item);
4596 * If the in-memory inode's i_size is smaller then the inode
4597 * size stored in the btree, return the inode's i_size, so
4598 * that we get a correct inode size after replaying the log
4599 * when before a power failure we had a shrinking truncate
4600 * followed by addition of a new name (rename / new hard link).
4601 * Otherwise return the inode size from the btree, to avoid
4602 * data loss when replaying a log due to previously doing a
4603 * write that expands the inode's size and logging a new name
4604 * immediately after.
4606 if (*size_ret > inode->vfs_inode.i_size)
4607 *size_ret = inode->vfs_inode.i_size;
4610 btrfs_release_path(path);
4615 * At the moment we always log all xattrs. This is to figure out at log replay
4616 * time which xattrs must have their deletion replayed. If a xattr is missing
4617 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4618 * because if a xattr is deleted, the inode is fsynced and a power failure
4619 * happens, causing the log to be replayed the next time the fs is mounted,
4620 * we want the xattr to not exist anymore (same behaviour as other filesystems
4621 * with a journal, ext3/4, xfs, f2fs, etc).
4623 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4624 struct btrfs_root *root,
4625 struct btrfs_inode *inode,
4626 struct btrfs_path *path,
4627 struct btrfs_path *dst_path)
4630 struct btrfs_key key;
4631 const u64 ino = btrfs_ino(inode);
4636 key.type = BTRFS_XATTR_ITEM_KEY;
4639 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4644 int slot = path->slots[0];
4645 struct extent_buffer *leaf = path->nodes[0];
4646 int nritems = btrfs_header_nritems(leaf);
4648 if (slot >= nritems) {
4650 u64 last_extent = 0;
4652 ret = copy_items(trans, inode, dst_path, path,
4653 &last_extent, start_slot,
4655 /* can't be 1, extent items aren't processed */
4661 ret = btrfs_next_leaf(root, path);
4669 btrfs_item_key_to_cpu(leaf, &key, slot);
4670 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4680 u64 last_extent = 0;
4682 ret = copy_items(trans, inode, dst_path, path,
4683 &last_extent, start_slot,
4685 /* can't be 1, extent items aren't processed */
4695 * If the no holes feature is enabled we need to make sure any hole between the
4696 * last extent and the i_size of our inode is explicitly marked in the log. This
4697 * is to make sure that doing something like:
4699 * 1) create file with 128Kb of data
4700 * 2) truncate file to 64Kb
4701 * 3) truncate file to 256Kb
4703 * 5) <crash/power failure>
4704 * 6) mount fs and trigger log replay
4706 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4707 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4708 * file correspond to a hole. The presence of explicit holes in a log tree is
4709 * what guarantees that log replay will remove/adjust file extent items in the
4712 * Here we do not need to care about holes between extents, that is already done
4713 * by copy_items(). We also only need to do this in the full sync path, where we
4714 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4715 * lookup the list of modified extent maps and if any represents a hole, we
4716 * insert a corresponding extent representing a hole in the log tree.
4718 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4719 struct btrfs_root *root,
4720 struct btrfs_inode *inode,
4721 struct btrfs_path *path)
4723 struct btrfs_fs_info *fs_info = root->fs_info;
4725 struct btrfs_key key;
4728 struct extent_buffer *leaf;
4729 struct btrfs_root *log = root->log_root;
4730 const u64 ino = btrfs_ino(inode);
4731 const u64 i_size = i_size_read(&inode->vfs_inode);
4733 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4737 key.type = BTRFS_EXTENT_DATA_KEY;
4738 key.offset = (u64)-1;
4740 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4745 ASSERT(path->slots[0] > 0);
4747 leaf = path->nodes[0];
4748 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4750 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4751 /* inode does not have any extents */
4755 struct btrfs_file_extent_item *extent;
4759 * If there's an extent beyond i_size, an explicit hole was
4760 * already inserted by copy_items().
4762 if (key.offset >= i_size)
4765 extent = btrfs_item_ptr(leaf, path->slots[0],
4766 struct btrfs_file_extent_item);
4768 if (btrfs_file_extent_type(leaf, extent) ==
4769 BTRFS_FILE_EXTENT_INLINE)
4772 len = btrfs_file_extent_num_bytes(leaf, extent);
4773 /* Last extent goes beyond i_size, no need to log a hole. */
4774 if (key.offset + len > i_size)
4776 hole_start = key.offset + len;
4777 hole_size = i_size - hole_start;
4779 btrfs_release_path(path);
4781 /* Last extent ends at i_size. */
4785 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4786 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4787 hole_size, 0, hole_size, 0, 0, 0);
4792 * When we are logging a new inode X, check if it doesn't have a reference that
4793 * matches the reference from some other inode Y created in a past transaction
4794 * and that was renamed in the current transaction. If we don't do this, then at
4795 * log replay time we can lose inode Y (and all its files if it's a directory):
4798 * echo "hello world" > /mnt/x/foobar
4801 * mkdir /mnt/x # or touch /mnt/x
4802 * xfs_io -c fsync /mnt/x
4804 * mount fs, trigger log replay
4806 * After the log replay procedure, we would lose the first directory and all its
4807 * files (file foobar).
4808 * For the case where inode Y is not a directory we simply end up losing it:
4810 * echo "123" > /mnt/foo
4812 * mv /mnt/foo /mnt/bar
4813 * echo "abc" > /mnt/foo
4814 * xfs_io -c fsync /mnt/foo
4817 * We also need this for cases where a snapshot entry is replaced by some other
4818 * entry (file or directory) otherwise we end up with an unreplayable log due to
4819 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4820 * if it were a regular entry:
4823 * btrfs subvolume snapshot /mnt /mnt/x/snap
4824 * btrfs subvolume delete /mnt/x/snap
4827 * fsync /mnt/x or fsync some new file inside it
4830 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4831 * the same transaction.
4833 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4835 const struct btrfs_key *key,
4836 struct btrfs_inode *inode,
4837 u64 *other_ino, u64 *other_parent)
4840 struct btrfs_path *search_path;
4843 u32 item_size = btrfs_item_size_nr(eb, slot);
4845 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4847 search_path = btrfs_alloc_path();
4850 search_path->search_commit_root = 1;
4851 search_path->skip_locking = 1;
4853 while (cur_offset < item_size) {
4857 unsigned long name_ptr;
4858 struct btrfs_dir_item *di;
4860 if (key->type == BTRFS_INODE_REF_KEY) {
4861 struct btrfs_inode_ref *iref;
4863 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4864 parent = key->offset;
4865 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4866 name_ptr = (unsigned long)(iref + 1);
4867 this_len = sizeof(*iref) + this_name_len;
4869 struct btrfs_inode_extref *extref;
4871 extref = (struct btrfs_inode_extref *)(ptr +
4873 parent = btrfs_inode_extref_parent(eb, extref);
4874 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4875 name_ptr = (unsigned long)&extref->name;
4876 this_len = sizeof(*extref) + this_name_len;
4879 if (this_name_len > name_len) {
4882 new_name = krealloc(name, this_name_len, GFP_NOFS);
4887 name_len = this_name_len;
4891 read_extent_buffer(eb, name, name_ptr, this_name_len);
4892 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4893 parent, name, this_name_len, 0);
4894 if (di && !IS_ERR(di)) {
4895 struct btrfs_key di_key;
4897 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4899 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4900 if (di_key.objectid != key->objectid) {
4902 *other_ino = di_key.objectid;
4903 *other_parent = parent;
4911 } else if (IS_ERR(di)) {
4915 btrfs_release_path(search_path);
4917 cur_offset += this_len;
4921 btrfs_free_path(search_path);
4926 struct btrfs_ino_list {
4929 struct list_head list;
4932 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4933 struct btrfs_root *root,
4934 struct btrfs_path *path,
4935 struct btrfs_log_ctx *ctx,
4936 u64 ino, u64 parent)
4938 struct btrfs_ino_list *ino_elem;
4939 LIST_HEAD(inode_list);
4942 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4945 ino_elem->ino = ino;
4946 ino_elem->parent = parent;
4947 list_add_tail(&ino_elem->list, &inode_list);
4949 while (!list_empty(&inode_list)) {
4950 struct btrfs_fs_info *fs_info = root->fs_info;
4951 struct btrfs_key key;
4952 struct inode *inode;
4954 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4956 ino = ino_elem->ino;
4957 parent = ino_elem->parent;
4958 list_del(&ino_elem->list);
4963 btrfs_release_path(path);
4966 key.type = BTRFS_INODE_ITEM_KEY;
4968 inode = btrfs_iget(fs_info->sb, &key, root);
4970 * If the other inode that had a conflicting dir entry was
4971 * deleted in the current transaction, we need to log its parent
4974 if (IS_ERR(inode)) {
4975 ret = PTR_ERR(inode);
4976 if (ret == -ENOENT) {
4977 key.objectid = parent;
4978 inode = btrfs_iget(fs_info->sb, &key, root);
4979 if (IS_ERR(inode)) {
4980 ret = PTR_ERR(inode);
4982 ret = btrfs_log_inode(trans, root,
4984 LOG_OTHER_INODE_ALL,
4986 btrfs_add_delayed_iput(inode);
4992 * We are safe logging the other inode without acquiring its
4993 * lock as long as we log with the LOG_INODE_EXISTS mode. We
4994 * are safe against concurrent renames of the other inode as
4995 * well because during a rename we pin the log and update the
4996 * log with the new name before we unpin it.
4998 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
4999 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
5001 btrfs_add_delayed_iput(inode);
5006 key.type = BTRFS_INODE_REF_KEY;
5008 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5010 btrfs_add_delayed_iput(inode);
5015 struct extent_buffer *leaf = path->nodes[0];
5016 int slot = path->slots[0];
5018 u64 other_parent = 0;
5020 if (slot >= btrfs_header_nritems(leaf)) {
5021 ret = btrfs_next_leaf(root, path);
5024 } else if (ret > 0) {
5031 btrfs_item_key_to_cpu(leaf, &key, slot);
5032 if (key.objectid != ino ||
5033 (key.type != BTRFS_INODE_REF_KEY &&
5034 key.type != BTRFS_INODE_EXTREF_KEY)) {
5039 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5040 BTRFS_I(inode), &other_ino,
5045 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5050 ino_elem->ino = other_ino;
5051 ino_elem->parent = other_parent;
5052 list_add_tail(&ino_elem->list, &inode_list);
5057 btrfs_add_delayed_iput(inode);
5063 /* log a single inode in the tree log.
5064 * At least one parent directory for this inode must exist in the tree
5065 * or be logged already.
5067 * Any items from this inode changed by the current transaction are copied
5068 * to the log tree. An extra reference is taken on any extents in this
5069 * file, allowing us to avoid a whole pile of corner cases around logging
5070 * blocks that have been removed from the tree.
5072 * See LOG_INODE_ALL and related defines for a description of what inode_only
5075 * This handles both files and directories.
5077 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5078 struct btrfs_root *root, struct btrfs_inode *inode,
5082 struct btrfs_log_ctx *ctx)
5084 struct btrfs_fs_info *fs_info = root->fs_info;
5085 struct btrfs_path *path;
5086 struct btrfs_path *dst_path;
5087 struct btrfs_key min_key;
5088 struct btrfs_key max_key;
5089 struct btrfs_root *log = root->log_root;
5090 u64 last_extent = 0;
5094 int ins_start_slot = 0;
5096 bool fast_search = false;
5097 u64 ino = btrfs_ino(inode);
5098 struct extent_map_tree *em_tree = &inode->extent_tree;
5099 u64 logged_isize = 0;
5100 bool need_log_inode_item = true;
5101 bool xattrs_logged = false;
5102 bool recursive_logging = false;
5104 path = btrfs_alloc_path();
5107 dst_path = btrfs_alloc_path();
5109 btrfs_free_path(path);
5113 min_key.objectid = ino;
5114 min_key.type = BTRFS_INODE_ITEM_KEY;
5117 max_key.objectid = ino;
5120 /* today the code can only do partial logging of directories */
5121 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5122 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5123 &inode->runtime_flags) &&
5124 inode_only >= LOG_INODE_EXISTS))
5125 max_key.type = BTRFS_XATTR_ITEM_KEY;
5127 max_key.type = (u8)-1;
5128 max_key.offset = (u64)-1;
5131 * Only run delayed items if we are a dir or a new file.
5132 * Otherwise commit the delayed inode only, which is needed in
5133 * order for the log replay code to mark inodes for link count
5134 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5136 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5137 inode->generation > fs_info->last_trans_committed)
5138 ret = btrfs_commit_inode_delayed_items(trans, inode);
5140 ret = btrfs_commit_inode_delayed_inode(inode);
5143 btrfs_free_path(path);
5144 btrfs_free_path(dst_path);
5148 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5149 recursive_logging = true;
5150 if (inode_only == LOG_OTHER_INODE)
5151 inode_only = LOG_INODE_EXISTS;
5153 inode_only = LOG_INODE_ALL;
5154 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5156 mutex_lock(&inode->log_mutex);
5160 * a brute force approach to making sure we get the most uptodate
5161 * copies of everything.
5163 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5164 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5166 if (inode_only == LOG_INODE_EXISTS)
5167 max_key_type = BTRFS_XATTR_ITEM_KEY;
5168 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5170 if (inode_only == LOG_INODE_EXISTS) {
5172 * Make sure the new inode item we write to the log has
5173 * the same isize as the current one (if it exists).
5174 * This is necessary to prevent data loss after log
5175 * replay, and also to prevent doing a wrong expanding
5176 * truncate - for e.g. create file, write 4K into offset
5177 * 0, fsync, write 4K into offset 4096, add hard link,
5178 * fsync some other file (to sync log), power fail - if
5179 * we use the inode's current i_size, after log replay
5180 * we get a 8Kb file, with the last 4Kb extent as a hole
5181 * (zeroes), as if an expanding truncate happened,
5182 * instead of getting a file of 4Kb only.
5184 err = logged_inode_size(log, inode, path, &logged_isize);
5188 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5189 &inode->runtime_flags)) {
5190 if (inode_only == LOG_INODE_EXISTS) {
5191 max_key.type = BTRFS_XATTR_ITEM_KEY;
5192 ret = drop_objectid_items(trans, log, path, ino,
5195 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5196 &inode->runtime_flags);
5197 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5198 &inode->runtime_flags);
5200 ret = btrfs_truncate_inode_items(trans,
5201 log, &inode->vfs_inode, 0, 0);
5206 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5207 &inode->runtime_flags) ||
5208 inode_only == LOG_INODE_EXISTS) {
5209 if (inode_only == LOG_INODE_ALL)
5211 max_key.type = BTRFS_XATTR_ITEM_KEY;
5212 ret = drop_objectid_items(trans, log, path, ino,
5215 if (inode_only == LOG_INODE_ALL)
5228 ret = btrfs_search_forward(root, &min_key,
5229 path, trans->transid);
5237 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5238 if (min_key.objectid != ino)
5240 if (min_key.type > max_key.type)
5243 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5244 need_log_inode_item = false;
5246 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5247 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5248 inode->generation == trans->transid &&
5249 !recursive_logging) {
5251 u64 other_parent = 0;
5253 ret = btrfs_check_ref_name_override(path->nodes[0],
5254 path->slots[0], &min_key, inode,
5255 &other_ino, &other_parent);
5259 } else if (ret > 0 && ctx &&
5260 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5265 ins_start_slot = path->slots[0];
5267 ret = copy_items(trans, inode, dst_path, path,
5268 &last_extent, ins_start_slot,
5277 err = log_conflicting_inodes(trans, root, path,
5278 ctx, other_ino, other_parent);
5281 btrfs_release_path(path);
5286 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5287 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5290 ret = copy_items(trans, inode, dst_path, path,
5291 &last_extent, ins_start_slot,
5292 ins_nr, inode_only, logged_isize);
5299 btrfs_release_path(path);
5305 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5308 } else if (!ins_nr) {
5309 ins_start_slot = path->slots[0];
5314 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5315 ins_start_slot, ins_nr, inode_only,
5323 btrfs_release_path(path);
5327 ins_start_slot = path->slots[0];
5330 nritems = btrfs_header_nritems(path->nodes[0]);
5332 if (path->slots[0] < nritems) {
5333 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5338 ret = copy_items(trans, inode, dst_path, path,
5339 &last_extent, ins_start_slot,
5340 ins_nr, inode_only, logged_isize);
5348 btrfs_release_path(path);
5350 if (min_key.offset < (u64)-1) {
5352 } else if (min_key.type < max_key.type) {
5360 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5361 ins_start_slot, ins_nr, inode_only,
5371 btrfs_release_path(path);
5372 btrfs_release_path(dst_path);
5373 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5376 xattrs_logged = true;
5377 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5378 btrfs_release_path(path);
5379 btrfs_release_path(dst_path);
5380 err = btrfs_log_trailing_hole(trans, root, inode, path);
5385 btrfs_release_path(path);
5386 btrfs_release_path(dst_path);
5387 if (need_log_inode_item) {
5388 err = log_inode_item(trans, log, dst_path, inode);
5389 if (!err && !xattrs_logged) {
5390 err = btrfs_log_all_xattrs(trans, root, inode, path,
5392 btrfs_release_path(path);
5398 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5404 } else if (inode_only == LOG_INODE_ALL) {
5405 struct extent_map *em, *n;
5407 write_lock(&em_tree->lock);
5409 * We can't just remove every em if we're called for a ranged
5410 * fsync - that is, one that doesn't cover the whole possible
5411 * file range (0 to LLONG_MAX). This is because we can have
5412 * em's that fall outside the range we're logging and therefore
5413 * their ordered operations haven't completed yet
5414 * (btrfs_finish_ordered_io() not invoked yet). This means we
5415 * didn't get their respective file extent item in the fs/subvol
5416 * tree yet, and need to let the next fast fsync (one which
5417 * consults the list of modified extent maps) find the em so
5418 * that it logs a matching file extent item and waits for the
5419 * respective ordered operation to complete (if it's still
5422 * Removing every em outside the range we're logging would make
5423 * the next fast fsync not log their matching file extent items,
5424 * therefore making us lose data after a log replay.
5426 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5428 const u64 mod_end = em->mod_start + em->mod_len - 1;
5430 if (em->mod_start >= start && mod_end <= end)
5431 list_del_init(&em->list);
5433 write_unlock(&em_tree->lock);
5436 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5437 ret = log_directory_changes(trans, root, inode, path, dst_path,
5446 * Don't update last_log_commit if we logged that an inode exists after
5447 * it was loaded to memory (full_sync bit set).
5448 * This is to prevent data loss when we do a write to the inode, then
5449 * the inode gets evicted after all delalloc was flushed, then we log
5450 * it exists (due to a rename for example) and then fsync it. This last
5451 * fsync would do nothing (not logging the extents previously written).
5453 spin_lock(&inode->lock);
5454 inode->logged_trans = trans->transid;
5455 if (inode_only != LOG_INODE_EXISTS ||
5456 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5457 inode->last_log_commit = inode->last_sub_trans;
5458 spin_unlock(&inode->lock);
5460 mutex_unlock(&inode->log_mutex);
5462 btrfs_free_path(path);
5463 btrfs_free_path(dst_path);
5468 * Check if we must fallback to a transaction commit when logging an inode.
5469 * This must be called after logging the inode and is used only in the context
5470 * when fsyncing an inode requires the need to log some other inode - in which
5471 * case we can't lock the i_mutex of each other inode we need to log as that
5472 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5473 * log inodes up or down in the hierarchy) or rename operations for example. So
5474 * we take the log_mutex of the inode after we have logged it and then check for
5475 * its last_unlink_trans value - this is safe because any task setting
5476 * last_unlink_trans must take the log_mutex and it must do this before it does
5477 * the actual unlink operation, so if we do this check before a concurrent task
5478 * sets last_unlink_trans it means we've logged a consistent version/state of
5479 * all the inode items, otherwise we are not sure and must do a transaction
5480 * commit (the concurrent task might have only updated last_unlink_trans before
5481 * we logged the inode or it might have also done the unlink).
5483 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5484 struct btrfs_inode *inode)
5486 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5489 mutex_lock(&inode->log_mutex);
5490 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5492 * Make sure any commits to the log are forced to be full
5495 btrfs_set_log_full_commit(trans);
5498 mutex_unlock(&inode->log_mutex);
5504 * follow the dentry parent pointers up the chain and see if any
5505 * of the directories in it require a full commit before they can
5506 * be logged. Returns zero if nothing special needs to be done or 1 if
5507 * a full commit is required.
5509 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5510 struct btrfs_inode *inode,
5511 struct dentry *parent,
5512 struct super_block *sb,
5516 struct dentry *old_parent = NULL;
5519 * for regular files, if its inode is already on disk, we don't
5520 * have to worry about the parents at all. This is because
5521 * we can use the last_unlink_trans field to record renames
5522 * and other fun in this file.
5524 if (S_ISREG(inode->vfs_inode.i_mode) &&
5525 inode->generation <= last_committed &&
5526 inode->last_unlink_trans <= last_committed)
5529 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5530 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5532 inode = BTRFS_I(d_inode(parent));
5536 if (btrfs_must_commit_transaction(trans, inode)) {
5541 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5544 if (IS_ROOT(parent)) {
5545 inode = BTRFS_I(d_inode(parent));
5546 if (btrfs_must_commit_transaction(trans, inode))
5551 parent = dget_parent(parent);
5553 old_parent = parent;
5554 inode = BTRFS_I(d_inode(parent));
5562 struct btrfs_dir_list {
5564 struct list_head list;
5568 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5569 * details about the why it is needed.
5570 * This is a recursive operation - if an existing dentry corresponds to a
5571 * directory, that directory's new entries are logged too (same behaviour as
5572 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5573 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5574 * complains about the following circular lock dependency / possible deadlock:
5578 * lock(&type->i_mutex_dir_key#3/2);
5579 * lock(sb_internal#2);
5580 * lock(&type->i_mutex_dir_key#3/2);
5581 * lock(&sb->s_type->i_mutex_key#14);
5583 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5584 * sb_start_intwrite() in btrfs_start_transaction().
5585 * Not locking i_mutex of the inodes is still safe because:
5587 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5588 * that while logging the inode new references (names) are added or removed
5589 * from the inode, leaving the logged inode item with a link count that does
5590 * not match the number of logged inode reference items. This is fine because
5591 * at log replay time we compute the real number of links and correct the
5592 * link count in the inode item (see replay_one_buffer() and
5593 * link_to_fixup_dir());
5595 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5596 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5597 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5598 * has a size that doesn't match the sum of the lengths of all the logged
5599 * names. This does not result in a problem because if a dir_item key is
5600 * logged but its matching dir_index key is not logged, at log replay time we
5601 * don't use it to replay the respective name (see replay_one_name()). On the
5602 * other hand if only the dir_index key ends up being logged, the respective
5603 * name is added to the fs/subvol tree with both the dir_item and dir_index
5604 * keys created (see replay_one_name()).
5605 * The directory's inode item with a wrong i_size is not a problem as well,
5606 * since we don't use it at log replay time to set the i_size in the inode
5607 * item of the fs/subvol tree (see overwrite_item()).
5609 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5610 struct btrfs_root *root,
5611 struct btrfs_inode *start_inode,
5612 struct btrfs_log_ctx *ctx)
5614 struct btrfs_fs_info *fs_info = root->fs_info;
5615 struct btrfs_root *log = root->log_root;
5616 struct btrfs_path *path;
5617 LIST_HEAD(dir_list);
5618 struct btrfs_dir_list *dir_elem;
5621 path = btrfs_alloc_path();
5625 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5627 btrfs_free_path(path);
5630 dir_elem->ino = btrfs_ino(start_inode);
5631 list_add_tail(&dir_elem->list, &dir_list);
5633 while (!list_empty(&dir_list)) {
5634 struct extent_buffer *leaf;
5635 struct btrfs_key min_key;
5639 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5642 goto next_dir_inode;
5644 min_key.objectid = dir_elem->ino;
5645 min_key.type = BTRFS_DIR_ITEM_KEY;
5648 btrfs_release_path(path);
5649 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5651 goto next_dir_inode;
5652 } else if (ret > 0) {
5654 goto next_dir_inode;
5658 leaf = path->nodes[0];
5659 nritems = btrfs_header_nritems(leaf);
5660 for (i = path->slots[0]; i < nritems; i++) {
5661 struct btrfs_dir_item *di;
5662 struct btrfs_key di_key;
5663 struct inode *di_inode;
5664 struct btrfs_dir_list *new_dir_elem;
5665 int log_mode = LOG_INODE_EXISTS;
5668 btrfs_item_key_to_cpu(leaf, &min_key, i);
5669 if (min_key.objectid != dir_elem->ino ||
5670 min_key.type != BTRFS_DIR_ITEM_KEY)
5671 goto next_dir_inode;
5673 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5674 type = btrfs_dir_type(leaf, di);
5675 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5676 type != BTRFS_FT_DIR)
5678 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5679 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5682 btrfs_release_path(path);
5683 di_inode = btrfs_iget(fs_info->sb, &di_key, root);
5684 if (IS_ERR(di_inode)) {
5685 ret = PTR_ERR(di_inode);
5686 goto next_dir_inode;
5689 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5690 btrfs_add_delayed_iput(di_inode);
5694 ctx->log_new_dentries = false;
5695 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5696 log_mode = LOG_INODE_ALL;
5697 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5698 log_mode, 0, LLONG_MAX, ctx);
5700 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5702 btrfs_add_delayed_iput(di_inode);
5704 goto next_dir_inode;
5705 if (ctx->log_new_dentries) {
5706 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5708 if (!new_dir_elem) {
5710 goto next_dir_inode;
5712 new_dir_elem->ino = di_key.objectid;
5713 list_add_tail(&new_dir_elem->list, &dir_list);
5718 ret = btrfs_next_leaf(log, path);
5720 goto next_dir_inode;
5721 } else if (ret > 0) {
5723 goto next_dir_inode;
5727 if (min_key.offset < (u64)-1) {
5732 list_del(&dir_elem->list);
5736 btrfs_free_path(path);
5740 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5741 struct btrfs_inode *inode,
5742 struct btrfs_log_ctx *ctx)
5744 struct btrfs_fs_info *fs_info = trans->fs_info;
5746 struct btrfs_path *path;
5747 struct btrfs_key key;
5748 struct btrfs_root *root = inode->root;
5749 const u64 ino = btrfs_ino(inode);
5751 path = btrfs_alloc_path();
5754 path->skip_locking = 1;
5755 path->search_commit_root = 1;
5758 key.type = BTRFS_INODE_REF_KEY;
5760 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5765 struct extent_buffer *leaf = path->nodes[0];
5766 int slot = path->slots[0];
5771 if (slot >= btrfs_header_nritems(leaf)) {
5772 ret = btrfs_next_leaf(root, path);
5780 btrfs_item_key_to_cpu(leaf, &key, slot);
5781 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5782 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5785 item_size = btrfs_item_size_nr(leaf, slot);
5786 ptr = btrfs_item_ptr_offset(leaf, slot);
5787 while (cur_offset < item_size) {
5788 struct btrfs_key inode_key;
5789 struct inode *dir_inode;
5791 inode_key.type = BTRFS_INODE_ITEM_KEY;
5792 inode_key.offset = 0;
5794 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5795 struct btrfs_inode_extref *extref;
5797 extref = (struct btrfs_inode_extref *)
5799 inode_key.objectid = btrfs_inode_extref_parent(
5801 cur_offset += sizeof(*extref);
5802 cur_offset += btrfs_inode_extref_name_len(leaf,
5805 inode_key.objectid = key.offset;
5806 cur_offset = item_size;
5809 dir_inode = btrfs_iget(fs_info->sb, &inode_key, root);
5811 * If the parent inode was deleted, return an error to
5812 * fallback to a transaction commit. This is to prevent
5813 * getting an inode that was moved from one parent A to
5814 * a parent B, got its former parent A deleted and then
5815 * it got fsync'ed, from existing at both parents after
5816 * a log replay (and the old parent still existing).
5823 * mv /mnt/B/bar /mnt/A/bar
5824 * mv -T /mnt/A /mnt/B
5828 * If we ignore the old parent B which got deleted,
5829 * after a log replay we would have file bar linked
5830 * at both parents and the old parent B would still
5833 if (IS_ERR(dir_inode)) {
5834 ret = PTR_ERR(dir_inode);
5839 ctx->log_new_dentries = false;
5840 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5841 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5843 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5845 if (!ret && ctx && ctx->log_new_dentries)
5846 ret = log_new_dir_dentries(trans, root,
5847 BTRFS_I(dir_inode), ctx);
5848 btrfs_add_delayed_iput(dir_inode);
5856 btrfs_free_path(path);
5860 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5861 struct btrfs_root *root,
5862 struct btrfs_path *path,
5863 struct btrfs_log_ctx *ctx)
5865 struct btrfs_key found_key;
5867 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5870 struct btrfs_fs_info *fs_info = root->fs_info;
5871 const u64 last_committed = fs_info->last_trans_committed;
5872 struct extent_buffer *leaf = path->nodes[0];
5873 int slot = path->slots[0];
5874 struct btrfs_key search_key;
5875 struct inode *inode;
5878 btrfs_release_path(path);
5880 search_key.objectid = found_key.offset;
5881 search_key.type = BTRFS_INODE_ITEM_KEY;
5882 search_key.offset = 0;
5883 inode = btrfs_iget(fs_info->sb, &search_key, root);
5885 return PTR_ERR(inode);
5887 if (BTRFS_I(inode)->generation > last_committed)
5888 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5891 btrfs_add_delayed_iput(inode);
5895 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5898 search_key.type = BTRFS_INODE_REF_KEY;
5899 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5903 leaf = path->nodes[0];
5904 slot = path->slots[0];
5905 if (slot >= btrfs_header_nritems(leaf)) {
5906 ret = btrfs_next_leaf(root, path);
5911 leaf = path->nodes[0];
5912 slot = path->slots[0];
5915 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5916 if (found_key.objectid != search_key.objectid ||
5917 found_key.type != BTRFS_INODE_REF_KEY)
5923 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5924 struct btrfs_inode *inode,
5925 struct dentry *parent,
5926 struct btrfs_log_ctx *ctx)
5928 struct btrfs_root *root = inode->root;
5929 struct btrfs_fs_info *fs_info = root->fs_info;
5930 struct dentry *old_parent = NULL;
5931 struct super_block *sb = inode->vfs_inode.i_sb;
5935 if (!parent || d_really_is_negative(parent) ||
5939 inode = BTRFS_I(d_inode(parent));
5940 if (root != inode->root)
5943 if (inode->generation > fs_info->last_trans_committed) {
5944 ret = btrfs_log_inode(trans, root, inode,
5945 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5949 if (IS_ROOT(parent))
5952 parent = dget_parent(parent);
5954 old_parent = parent;
5961 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5962 struct btrfs_inode *inode,
5963 struct dentry *parent,
5964 struct btrfs_log_ctx *ctx)
5966 struct btrfs_root *root = inode->root;
5967 const u64 ino = btrfs_ino(inode);
5968 struct btrfs_path *path;
5969 struct btrfs_key search_key;
5973 * For a single hard link case, go through a fast path that does not
5974 * need to iterate the fs/subvolume tree.
5976 if (inode->vfs_inode.i_nlink < 2)
5977 return log_new_ancestors_fast(trans, inode, parent, ctx);
5979 path = btrfs_alloc_path();
5983 search_key.objectid = ino;
5984 search_key.type = BTRFS_INODE_REF_KEY;
5985 search_key.offset = 0;
5987 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5994 struct extent_buffer *leaf = path->nodes[0];
5995 int slot = path->slots[0];
5996 struct btrfs_key found_key;
5998 if (slot >= btrfs_header_nritems(leaf)) {
5999 ret = btrfs_next_leaf(root, path);
6007 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6008 if (found_key.objectid != ino ||
6009 found_key.type > BTRFS_INODE_EXTREF_KEY)
6013 * Don't deal with extended references because they are rare
6014 * cases and too complex to deal with (we would need to keep
6015 * track of which subitem we are processing for each item in
6016 * this loop, etc). So just return some error to fallback to
6017 * a transaction commit.
6019 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6025 * Logging ancestors needs to do more searches on the fs/subvol
6026 * tree, so it releases the path as needed to avoid deadlocks.
6027 * Keep track of the last inode ref key and resume from that key
6028 * after logging all new ancestors for the current hard link.
6030 memcpy(&search_key, &found_key, sizeof(search_key));
6032 ret = log_new_ancestors(trans, root, path, ctx);
6035 btrfs_release_path(path);
6040 btrfs_free_path(path);
6045 * helper function around btrfs_log_inode to make sure newly created
6046 * parent directories also end up in the log. A minimal inode and backref
6047 * only logging is done of any parent directories that are older than
6048 * the last committed transaction
6050 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6051 struct btrfs_inode *inode,
6052 struct dentry *parent,
6056 struct btrfs_log_ctx *ctx)
6058 struct btrfs_root *root = inode->root;
6059 struct btrfs_fs_info *fs_info = root->fs_info;
6060 struct super_block *sb;
6062 u64 last_committed = fs_info->last_trans_committed;
6063 bool log_dentries = false;
6065 sb = inode->vfs_inode.i_sb;
6067 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6073 * The prev transaction commit doesn't complete, we need do
6074 * full commit by ourselves.
6076 if (fs_info->last_trans_log_full_commit >
6077 fs_info->last_trans_committed) {
6082 if (btrfs_root_refs(&root->root_item) == 0) {
6087 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
6093 * Skip already logged inodes or inodes corresponding to tmpfiles
6094 * (since logging them is pointless, a link count of 0 means they
6095 * will never be accessible).
6097 if (btrfs_inode_in_log(inode, trans->transid) ||
6098 inode->vfs_inode.i_nlink == 0) {
6099 ret = BTRFS_NO_LOG_SYNC;
6103 ret = start_log_trans(trans, root, ctx);
6107 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
6112 * for regular files, if its inode is already on disk, we don't
6113 * have to worry about the parents at all. This is because
6114 * we can use the last_unlink_trans field to record renames
6115 * and other fun in this file.
6117 if (S_ISREG(inode->vfs_inode.i_mode) &&
6118 inode->generation <= last_committed &&
6119 inode->last_unlink_trans <= last_committed) {
6124 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6125 log_dentries = true;
6128 * On unlink we must make sure all our current and old parent directory
6129 * inodes are fully logged. This is to prevent leaving dangling
6130 * directory index entries in directories that were our parents but are
6131 * not anymore. Not doing this results in old parent directory being
6132 * impossible to delete after log replay (rmdir will always fail with
6133 * error -ENOTEMPTY).
6139 * ln testdir/foo testdir/bar
6141 * unlink testdir/bar
6142 * xfs_io -c fsync testdir/foo
6144 * mount fs, triggers log replay
6146 * If we don't log the parent directory (testdir), after log replay the
6147 * directory still has an entry pointing to the file inode using the bar
6148 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6149 * the file inode has a link count of 1.
6155 * ln foo testdir/foo2
6156 * ln foo testdir/foo3
6158 * unlink testdir/foo3
6159 * xfs_io -c fsync foo
6161 * mount fs, triggers log replay
6163 * Similar as the first example, after log replay the parent directory
6164 * testdir still has an entry pointing to the inode file with name foo3
6165 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6166 * and has a link count of 2.
6168 if (inode->last_unlink_trans > last_committed) {
6169 ret = btrfs_log_all_parents(trans, inode, ctx);
6174 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6179 ret = log_new_dir_dentries(trans, root, inode, ctx);
6184 btrfs_set_log_full_commit(trans);
6189 btrfs_remove_log_ctx(root, ctx);
6190 btrfs_end_log_trans(root);
6196 * it is not safe to log dentry if the chunk root has added new
6197 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6198 * If this returns 1, you must commit the transaction to safely get your
6201 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6202 struct dentry *dentry,
6205 struct btrfs_log_ctx *ctx)
6207 struct dentry *parent = dget_parent(dentry);
6210 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6211 start, end, LOG_INODE_ALL, ctx);
6218 * should be called during mount to recover any replay any log trees
6221 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6224 struct btrfs_path *path;
6225 struct btrfs_trans_handle *trans;
6226 struct btrfs_key key;
6227 struct btrfs_key found_key;
6228 struct btrfs_key tmp_key;
6229 struct btrfs_root *log;
6230 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6231 struct walk_control wc = {
6232 .process_func = process_one_buffer,
6233 .stage = LOG_WALK_PIN_ONLY,
6236 path = btrfs_alloc_path();
6240 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6242 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6243 if (IS_ERR(trans)) {
6244 ret = PTR_ERR(trans);
6251 ret = walk_log_tree(trans, log_root_tree, &wc);
6253 btrfs_handle_fs_error(fs_info, ret,
6254 "Failed to pin buffers while recovering log root tree.");
6259 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6260 key.offset = (u64)-1;
6261 key.type = BTRFS_ROOT_ITEM_KEY;
6264 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6267 btrfs_handle_fs_error(fs_info, ret,
6268 "Couldn't find tree log root.");
6272 if (path->slots[0] == 0)
6276 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6278 btrfs_release_path(path);
6279 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6282 log = btrfs_read_fs_root(log_root_tree, &found_key);
6285 btrfs_handle_fs_error(fs_info, ret,
6286 "Couldn't read tree log root.");
6290 tmp_key.objectid = found_key.offset;
6291 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6292 tmp_key.offset = (u64)-1;
6294 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6295 if (IS_ERR(wc.replay_dest)) {
6296 ret = PTR_ERR(wc.replay_dest);
6297 free_extent_buffer(log->node);
6298 free_extent_buffer(log->commit_root);
6300 btrfs_handle_fs_error(fs_info, ret,
6301 "Couldn't read target root for tree log recovery.");
6305 wc.replay_dest->log_root = log;
6306 btrfs_record_root_in_trans(trans, wc.replay_dest);
6307 ret = walk_log_tree(trans, log, &wc);
6309 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6310 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6314 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6315 struct btrfs_root *root = wc.replay_dest;
6317 btrfs_release_path(path);
6320 * We have just replayed everything, and the highest
6321 * objectid of fs roots probably has changed in case
6322 * some inode_item's got replayed.
6324 * root->objectid_mutex is not acquired as log replay
6325 * could only happen during mount.
6327 ret = btrfs_find_highest_objectid(root,
6328 &root->highest_objectid);
6331 key.offset = found_key.offset - 1;
6332 wc.replay_dest->log_root = NULL;
6333 free_extent_buffer(log->node);
6334 free_extent_buffer(log->commit_root);
6340 if (found_key.offset == 0)
6343 btrfs_release_path(path);
6345 /* step one is to pin it all, step two is to replay just inodes */
6348 wc.process_func = replay_one_buffer;
6349 wc.stage = LOG_WALK_REPLAY_INODES;
6352 /* step three is to replay everything */
6353 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6358 btrfs_free_path(path);
6360 /* step 4: commit the transaction, which also unpins the blocks */
6361 ret = btrfs_commit_transaction(trans);
6365 free_extent_buffer(log_root_tree->node);
6366 log_root_tree->log_root = NULL;
6367 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6368 kfree(log_root_tree);
6373 btrfs_end_transaction(wc.trans);
6374 btrfs_free_path(path);
6379 * there are some corner cases where we want to force a full
6380 * commit instead of allowing a directory to be logged.
6382 * They revolve around files there were unlinked from the directory, and
6383 * this function updates the parent directory so that a full commit is
6384 * properly done if it is fsync'd later after the unlinks are done.
6386 * Must be called before the unlink operations (updates to the subvolume tree,
6387 * inodes, etc) are done.
6389 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6390 struct btrfs_inode *dir, struct btrfs_inode *inode,
6394 * when we're logging a file, if it hasn't been renamed
6395 * or unlinked, and its inode is fully committed on disk,
6396 * we don't have to worry about walking up the directory chain
6397 * to log its parents.
6399 * So, we use the last_unlink_trans field to put this transid
6400 * into the file. When the file is logged we check it and
6401 * don't log the parents if the file is fully on disk.
6403 mutex_lock(&inode->log_mutex);
6404 inode->last_unlink_trans = trans->transid;
6405 mutex_unlock(&inode->log_mutex);
6408 * if this directory was already logged any new
6409 * names for this file/dir will get recorded
6411 if (dir->logged_trans == trans->transid)
6415 * if the inode we're about to unlink was logged,
6416 * the log will be properly updated for any new names
6418 if (inode->logged_trans == trans->transid)
6422 * when renaming files across directories, if the directory
6423 * there we're unlinking from gets fsync'd later on, there's
6424 * no way to find the destination directory later and fsync it
6425 * properly. So, we have to be conservative and force commits
6426 * so the new name gets discovered.
6431 /* we can safely do the unlink without any special recording */
6435 mutex_lock(&dir->log_mutex);
6436 dir->last_unlink_trans = trans->transid;
6437 mutex_unlock(&dir->log_mutex);
6441 * Make sure that if someone attempts to fsync the parent directory of a deleted
6442 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6443 * that after replaying the log tree of the parent directory's root we will not
6444 * see the snapshot anymore and at log replay time we will not see any log tree
6445 * corresponding to the deleted snapshot's root, which could lead to replaying
6446 * it after replaying the log tree of the parent directory (which would replay
6447 * the snapshot delete operation).
6449 * Must be called before the actual snapshot destroy operation (updates to the
6450 * parent root and tree of tree roots trees, etc) are done.
6452 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6453 struct btrfs_inode *dir)
6455 mutex_lock(&dir->log_mutex);
6456 dir->last_unlink_trans = trans->transid;
6457 mutex_unlock(&dir->log_mutex);
6461 * Call this after adding a new name for a file and it will properly
6462 * update the log to reflect the new name.
6464 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6465 * true (because it's not used).
6467 * Return value depends on whether @sync_log is true or false.
6468 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6469 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6471 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6472 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6473 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6474 * committed (without attempting to sync the log).
6476 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6477 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6478 struct dentry *parent,
6479 bool sync_log, struct btrfs_log_ctx *ctx)
6481 struct btrfs_fs_info *fs_info = trans->fs_info;
6485 * this will force the logging code to walk the dentry chain
6488 if (!S_ISDIR(inode->vfs_inode.i_mode))
6489 inode->last_unlink_trans = trans->transid;
6492 * if this inode hasn't been logged and directory we're renaming it
6493 * from hasn't been logged, we don't need to log it
6495 if (inode->logged_trans <= fs_info->last_trans_committed &&
6496 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6497 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6498 BTRFS_DONT_NEED_LOG_SYNC;
6501 struct btrfs_log_ctx ctx2;
6503 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6504 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6505 LOG_INODE_EXISTS, &ctx2);
6506 if (ret == BTRFS_NO_LOG_SYNC)
6507 return BTRFS_DONT_NEED_TRANS_COMMIT;
6509 return BTRFS_NEED_TRANS_COMMIT;
6511 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6513 return BTRFS_NEED_TRANS_COMMIT;
6514 return BTRFS_DONT_NEED_TRANS_COMMIT;
6518 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6519 LOG_INODE_EXISTS, ctx);
6520 if (ret == BTRFS_NO_LOG_SYNC)
6521 return BTRFS_DONT_NEED_LOG_SYNC;
6523 return BTRFS_NEED_TRANS_COMMIT;
6525 return BTRFS_NEED_LOG_SYNC;