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, NULL);
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;
948 struct btrfs_inode_ref *ref;
950 unsigned long ptr_end;
951 unsigned long name_ptr;
957 path = btrfs_alloc_path();
961 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
965 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
967 if (key->type == BTRFS_INODE_EXTREF_KEY) {
968 if (btrfs_find_name_in_ext_backref(path->nodes[0],
977 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
978 ptr_end = ptr + item_size;
979 while (ptr < ptr_end) {
980 ref = (struct btrfs_inode_ref *)ptr;
981 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
982 if (found_name_len == namelen) {
983 name_ptr = (unsigned long)(ref + 1);
984 ret = memcmp_extent_buffer(path->nodes[0], name,
991 ptr = (unsigned long)(ref + 1) + found_name_len;
994 btrfs_free_path(path);
998 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
999 struct btrfs_root *root,
1000 struct btrfs_path *path,
1001 struct btrfs_root *log_root,
1002 struct btrfs_inode *dir,
1003 struct btrfs_inode *inode,
1004 u64 inode_objectid, u64 parent_objectid,
1005 u64 ref_index, char *name, int namelen,
1010 int victim_name_len;
1011 struct extent_buffer *leaf;
1012 struct btrfs_dir_item *di;
1013 struct btrfs_key search_key;
1014 struct btrfs_inode_extref *extref;
1017 /* Search old style refs */
1018 search_key.objectid = inode_objectid;
1019 search_key.type = BTRFS_INODE_REF_KEY;
1020 search_key.offset = parent_objectid;
1021 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1023 struct btrfs_inode_ref *victim_ref;
1025 unsigned long ptr_end;
1027 leaf = path->nodes[0];
1029 /* are we trying to overwrite a back ref for the root directory
1030 * if so, just jump out, we're done
1032 if (search_key.objectid == search_key.offset)
1035 /* check all the names in this back reference to see
1036 * if they are in the log. if so, we allow them to stay
1037 * otherwise they must be unlinked as a conflict
1039 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1040 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1041 while (ptr < ptr_end) {
1042 victim_ref = (struct btrfs_inode_ref *)ptr;
1043 victim_name_len = btrfs_inode_ref_name_len(leaf,
1045 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1049 read_extent_buffer(leaf, victim_name,
1050 (unsigned long)(victim_ref + 1),
1053 if (!backref_in_log(log_root, &search_key,
1057 inc_nlink(&inode->vfs_inode);
1058 btrfs_release_path(path);
1060 ret = btrfs_unlink_inode(trans, root, dir, inode,
1061 victim_name, victim_name_len);
1065 ret = btrfs_run_delayed_items(trans);
1073 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1077 * NOTE: we have searched root tree and checked the
1078 * corresponding ref, it does not need to check again.
1082 btrfs_release_path(path);
1084 /* Same search but for extended refs */
1085 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1086 inode_objectid, parent_objectid, 0,
1088 if (!IS_ERR_OR_NULL(extref)) {
1092 struct inode *victim_parent;
1094 leaf = path->nodes[0];
1096 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1097 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1099 while (cur_offset < item_size) {
1100 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1102 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1104 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1107 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1110 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1113 search_key.objectid = inode_objectid;
1114 search_key.type = BTRFS_INODE_EXTREF_KEY;
1115 search_key.offset = btrfs_extref_hash(parent_objectid,
1119 if (!backref_in_log(log_root, &search_key,
1120 parent_objectid, victim_name,
1123 victim_parent = read_one_inode(root,
1125 if (victim_parent) {
1126 inc_nlink(&inode->vfs_inode);
1127 btrfs_release_path(path);
1129 ret = btrfs_unlink_inode(trans, root,
1130 BTRFS_I(victim_parent),
1135 ret = btrfs_run_delayed_items(
1138 iput(victim_parent);
1147 cur_offset += victim_name_len + sizeof(*extref);
1151 btrfs_release_path(path);
1153 /* look for a conflicting sequence number */
1154 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1155 ref_index, name, namelen, 0);
1156 if (di && !IS_ERR(di)) {
1157 ret = drop_one_dir_item(trans, root, path, dir, di);
1161 btrfs_release_path(path);
1163 /* look for a conflicting name */
1164 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1166 if (di && !IS_ERR(di)) {
1167 ret = drop_one_dir_item(trans, root, path, dir, di);
1171 btrfs_release_path(path);
1176 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1177 u32 *namelen, char **name, u64 *index,
1178 u64 *parent_objectid)
1180 struct btrfs_inode_extref *extref;
1182 extref = (struct btrfs_inode_extref *)ref_ptr;
1184 *namelen = btrfs_inode_extref_name_len(eb, extref);
1185 *name = kmalloc(*namelen, GFP_NOFS);
1189 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1193 *index = btrfs_inode_extref_index(eb, extref);
1194 if (parent_objectid)
1195 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1200 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1201 u32 *namelen, char **name, u64 *index)
1203 struct btrfs_inode_ref *ref;
1205 ref = (struct btrfs_inode_ref *)ref_ptr;
1207 *namelen = btrfs_inode_ref_name_len(eb, ref);
1208 *name = kmalloc(*namelen, GFP_NOFS);
1212 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1215 *index = btrfs_inode_ref_index(eb, ref);
1221 * Take an inode reference item from the log tree and iterate all names from the
1222 * inode reference item in the subvolume tree with the same key (if it exists).
1223 * For any name that is not in the inode reference item from the log tree, do a
1224 * proper unlink of that name (that is, remove its entry from the inode
1225 * reference item and both dir index keys).
1227 static int unlink_old_inode_refs(struct btrfs_trans_handle *trans,
1228 struct btrfs_root *root,
1229 struct btrfs_path *path,
1230 struct btrfs_inode *inode,
1231 struct extent_buffer *log_eb,
1233 struct btrfs_key *key)
1236 unsigned long ref_ptr;
1237 unsigned long ref_end;
1238 struct extent_buffer *eb;
1241 btrfs_release_path(path);
1242 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
1250 eb = path->nodes[0];
1251 ref_ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
1252 ref_end = ref_ptr + btrfs_item_size_nr(eb, path->slots[0]);
1253 while (ref_ptr < ref_end) {
1258 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1259 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1262 parent_id = key->offset;
1263 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1269 if (key->type == BTRFS_INODE_EXTREF_KEY)
1270 ret = !!btrfs_find_name_in_ext_backref(log_eb, log_slot,
1274 ret = !!btrfs_find_name_in_backref(log_eb, log_slot,
1280 btrfs_release_path(path);
1281 dir = read_one_inode(root, parent_id);
1287 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
1288 inode, name, namelen);
1298 if (key->type == BTRFS_INODE_EXTREF_KEY)
1299 ref_ptr += sizeof(struct btrfs_inode_extref);
1301 ref_ptr += sizeof(struct btrfs_inode_ref);
1305 btrfs_release_path(path);
1309 static int btrfs_inode_ref_exists(struct inode *inode, struct inode *dir,
1310 const u8 ref_type, const char *name,
1313 struct btrfs_key key;
1314 struct btrfs_path *path;
1315 const u64 parent_id = btrfs_ino(BTRFS_I(dir));
1318 path = btrfs_alloc_path();
1322 key.objectid = btrfs_ino(BTRFS_I(inode));
1323 key.type = ref_type;
1324 if (key.type == BTRFS_INODE_REF_KEY)
1325 key.offset = parent_id;
1327 key.offset = btrfs_extref_hash(parent_id, name, namelen);
1329 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &key, path, 0, 0);
1336 if (key.type == BTRFS_INODE_EXTREF_KEY)
1337 ret = !!btrfs_find_name_in_ext_backref(path->nodes[0],
1338 path->slots[0], parent_id, name, namelen);
1340 ret = !!btrfs_find_name_in_backref(path->nodes[0], path->slots[0],
1344 btrfs_free_path(path);
1348 static int add_link(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1349 struct inode *dir, struct inode *inode, const char *name,
1350 int namelen, u64 ref_index)
1352 struct btrfs_dir_item *dir_item;
1353 struct btrfs_key key;
1354 struct btrfs_path *path;
1355 struct inode *other_inode = NULL;
1358 path = btrfs_alloc_path();
1362 dir_item = btrfs_lookup_dir_item(NULL, root, path,
1363 btrfs_ino(BTRFS_I(dir)),
1366 btrfs_release_path(path);
1368 } else if (IS_ERR(dir_item)) {
1369 ret = PTR_ERR(dir_item);
1374 * Our inode's dentry collides with the dentry of another inode which is
1375 * in the log but not yet processed since it has a higher inode number.
1376 * So delete that other dentry.
1378 btrfs_dir_item_key_to_cpu(path->nodes[0], dir_item, &key);
1379 btrfs_release_path(path);
1380 other_inode = read_one_inode(root, key.objectid);
1385 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(other_inode),
1390 * If we dropped the link count to 0, bump it so that later the iput()
1391 * on the inode will not free it. We will fixup the link count later.
1393 if (other_inode->i_nlink == 0)
1394 inc_nlink(other_inode);
1396 ret = btrfs_run_delayed_items(trans);
1400 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
1401 name, namelen, 0, ref_index);
1404 btrfs_free_path(path);
1410 * replay one inode back reference item found in the log tree.
1411 * eb, slot and key refer to the buffer and key found in the log tree.
1412 * root is the destination we are replaying into, and path is for temp
1413 * use by this function. (it should be released on return).
1415 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1416 struct btrfs_root *root,
1417 struct btrfs_root *log,
1418 struct btrfs_path *path,
1419 struct extent_buffer *eb, int slot,
1420 struct btrfs_key *key)
1422 struct inode *dir = NULL;
1423 struct inode *inode = NULL;
1424 unsigned long ref_ptr;
1425 unsigned long ref_end;
1429 int search_done = 0;
1430 int log_ref_ver = 0;
1431 u64 parent_objectid;
1434 int ref_struct_size;
1436 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1437 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1439 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1440 struct btrfs_inode_extref *r;
1442 ref_struct_size = sizeof(struct btrfs_inode_extref);
1444 r = (struct btrfs_inode_extref *)ref_ptr;
1445 parent_objectid = btrfs_inode_extref_parent(eb, r);
1447 ref_struct_size = sizeof(struct btrfs_inode_ref);
1448 parent_objectid = key->offset;
1450 inode_objectid = key->objectid;
1453 * it is possible that we didn't log all the parent directories
1454 * for a given inode. If we don't find the dir, just don't
1455 * copy the back ref in. The link count fixup code will take
1458 dir = read_one_inode(root, parent_objectid);
1464 inode = read_one_inode(root, inode_objectid);
1470 while (ref_ptr < ref_end) {
1472 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1473 &ref_index, &parent_objectid);
1475 * parent object can change from one array
1479 dir = read_one_inode(root, parent_objectid);
1485 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1491 /* if we already have a perfect match, we're done */
1492 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1493 btrfs_ino(BTRFS_I(inode)), ref_index,
1496 * look for a conflicting back reference in the
1497 * metadata. if we find one we have to unlink that name
1498 * of the file before we add our new link. Later on, we
1499 * overwrite any existing back reference, and we don't
1500 * want to create dangling pointers in the directory.
1504 ret = __add_inode_ref(trans, root, path, log,
1509 ref_index, name, namelen,
1519 * If a reference item already exists for this inode
1520 * with the same parent and name, but different index,
1521 * drop it and the corresponding directory index entries
1522 * from the parent before adding the new reference item
1523 * and dir index entries, otherwise we would fail with
1524 * -EEXIST returned from btrfs_add_link() below.
1526 ret = btrfs_inode_ref_exists(inode, dir, key->type,
1529 ret = btrfs_unlink_inode(trans, root,
1534 * If we dropped the link count to 0, bump it so
1535 * that later the iput() on the inode will not
1536 * free it. We will fixup the link count later.
1538 if (!ret && inode->i_nlink == 0)
1544 /* insert our name */
1545 ret = add_link(trans, root, dir, inode, name, namelen,
1550 btrfs_update_inode(trans, root, inode);
1553 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1563 * Before we overwrite the inode reference item in the subvolume tree
1564 * with the item from the log tree, we must unlink all names from the
1565 * parent directory that are in the subvolume's tree inode reference
1566 * item, otherwise we end up with an inconsistent subvolume tree where
1567 * dir index entries exist for a name but there is no inode reference
1568 * item with the same name.
1570 ret = unlink_old_inode_refs(trans, root, path, BTRFS_I(inode), eb, slot,
1575 /* finally write the back reference in the inode */
1576 ret = overwrite_item(trans, root, path, eb, slot, key);
1578 btrfs_release_path(path);
1585 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1586 struct btrfs_root *root, u64 ino)
1590 ret = btrfs_insert_orphan_item(trans, root, ino);
1597 static int count_inode_extrefs(struct btrfs_root *root,
1598 struct btrfs_inode *inode, struct btrfs_path *path)
1602 unsigned int nlink = 0;
1605 u64 inode_objectid = btrfs_ino(inode);
1608 struct btrfs_inode_extref *extref;
1609 struct extent_buffer *leaf;
1612 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1617 leaf = path->nodes[0];
1618 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1619 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1622 while (cur_offset < item_size) {
1623 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1624 name_len = btrfs_inode_extref_name_len(leaf, extref);
1628 cur_offset += name_len + sizeof(*extref);
1632 btrfs_release_path(path);
1634 btrfs_release_path(path);
1636 if (ret < 0 && ret != -ENOENT)
1641 static int count_inode_refs(struct btrfs_root *root,
1642 struct btrfs_inode *inode, struct btrfs_path *path)
1645 struct btrfs_key key;
1646 unsigned int nlink = 0;
1648 unsigned long ptr_end;
1650 u64 ino = btrfs_ino(inode);
1653 key.type = BTRFS_INODE_REF_KEY;
1654 key.offset = (u64)-1;
1657 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1661 if (path->slots[0] == 0)
1666 btrfs_item_key_to_cpu(path->nodes[0], &key,
1668 if (key.objectid != ino ||
1669 key.type != BTRFS_INODE_REF_KEY)
1671 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1672 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1674 while (ptr < ptr_end) {
1675 struct btrfs_inode_ref *ref;
1677 ref = (struct btrfs_inode_ref *)ptr;
1678 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1680 ptr = (unsigned long)(ref + 1) + name_len;
1684 if (key.offset == 0)
1686 if (path->slots[0] > 0) {
1691 btrfs_release_path(path);
1693 btrfs_release_path(path);
1699 * There are a few corners where the link count of the file can't
1700 * be properly maintained during replay. So, instead of adding
1701 * lots of complexity to the log code, we just scan the backrefs
1702 * for any file that has been through replay.
1704 * The scan will update the link count on the inode to reflect the
1705 * number of back refs found. If it goes down to zero, the iput
1706 * will free the inode.
1708 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1709 struct btrfs_root *root,
1710 struct inode *inode)
1712 struct btrfs_path *path;
1715 u64 ino = btrfs_ino(BTRFS_I(inode));
1717 path = btrfs_alloc_path();
1721 ret = count_inode_refs(root, BTRFS_I(inode), path);
1727 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1735 if (nlink != inode->i_nlink) {
1736 set_nlink(inode, nlink);
1737 btrfs_update_inode(trans, root, inode);
1739 BTRFS_I(inode)->index_cnt = (u64)-1;
1741 if (inode->i_nlink == 0) {
1742 if (S_ISDIR(inode->i_mode)) {
1743 ret = replay_dir_deletes(trans, root, NULL, path,
1748 ret = insert_orphan_item(trans, root, ino);
1752 btrfs_free_path(path);
1756 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1757 struct btrfs_root *root,
1758 struct btrfs_path *path)
1761 struct btrfs_key key;
1762 struct inode *inode;
1764 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1765 key.type = BTRFS_ORPHAN_ITEM_KEY;
1766 key.offset = (u64)-1;
1768 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1773 if (path->slots[0] == 0)
1778 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1779 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1780 key.type != BTRFS_ORPHAN_ITEM_KEY)
1783 ret = btrfs_del_item(trans, root, path);
1787 btrfs_release_path(path);
1788 inode = read_one_inode(root, key.offset);
1792 ret = fixup_inode_link_count(trans, root, inode);
1798 * fixup on a directory may create new entries,
1799 * make sure we always look for the highset possible
1802 key.offset = (u64)-1;
1806 btrfs_release_path(path);
1812 * record a given inode in the fixup dir so we can check its link
1813 * count when replay is done. The link count is incremented here
1814 * so the inode won't go away until we check it
1816 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1817 struct btrfs_root *root,
1818 struct btrfs_path *path,
1821 struct btrfs_key key;
1823 struct inode *inode;
1825 inode = read_one_inode(root, objectid);
1829 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1830 key.type = BTRFS_ORPHAN_ITEM_KEY;
1831 key.offset = objectid;
1833 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1835 btrfs_release_path(path);
1837 if (!inode->i_nlink)
1838 set_nlink(inode, 1);
1841 ret = btrfs_update_inode(trans, root, inode);
1842 } else if (ret == -EEXIST) {
1845 BUG(); /* Logic Error */
1853 * when replaying the log for a directory, we only insert names
1854 * for inodes that actually exist. This means an fsync on a directory
1855 * does not implicitly fsync all the new files in it
1857 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1858 struct btrfs_root *root,
1859 u64 dirid, u64 index,
1860 char *name, int name_len,
1861 struct btrfs_key *location)
1863 struct inode *inode;
1867 inode = read_one_inode(root, location->objectid);
1871 dir = read_one_inode(root, dirid);
1877 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1878 name_len, 1, index);
1880 /* FIXME, put inode into FIXUP list */
1888 * Return true if an inode reference exists in the log for the given name,
1889 * inode and parent inode.
1891 static bool name_in_log_ref(struct btrfs_root *log_root,
1892 const char *name, const int name_len,
1893 const u64 dirid, const u64 ino)
1895 struct btrfs_key search_key;
1897 search_key.objectid = ino;
1898 search_key.type = BTRFS_INODE_REF_KEY;
1899 search_key.offset = dirid;
1900 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1903 search_key.type = BTRFS_INODE_EXTREF_KEY;
1904 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1905 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1912 * take a single entry in a log directory item and replay it into
1915 * if a conflicting item exists in the subdirectory already,
1916 * the inode it points to is unlinked and put into the link count
1919 * If a name from the log points to a file or directory that does
1920 * not exist in the FS, it is skipped. fsyncs on directories
1921 * do not force down inodes inside that directory, just changes to the
1922 * names or unlinks in a directory.
1924 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1925 * non-existing inode) and 1 if the name was replayed.
1927 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *root,
1929 struct btrfs_path *path,
1930 struct extent_buffer *eb,
1931 struct btrfs_dir_item *di,
1932 struct btrfs_key *key)
1936 struct btrfs_dir_item *dst_di;
1937 struct btrfs_key found_key;
1938 struct btrfs_key log_key;
1943 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1944 bool name_added = false;
1946 dir = read_one_inode(root, key->objectid);
1950 name_len = btrfs_dir_name_len(eb, di);
1951 name = kmalloc(name_len, GFP_NOFS);
1957 log_type = btrfs_dir_type(eb, di);
1958 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1961 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1962 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1967 btrfs_release_path(path);
1969 if (key->type == BTRFS_DIR_ITEM_KEY) {
1970 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1972 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1973 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1982 if (IS_ERR_OR_NULL(dst_di)) {
1983 /* we need a sequence number to insert, so we only
1984 * do inserts for the BTRFS_DIR_INDEX_KEY types
1986 if (key->type != BTRFS_DIR_INDEX_KEY)
1991 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1992 /* the existing item matches the logged item */
1993 if (found_key.objectid == log_key.objectid &&
1994 found_key.type == log_key.type &&
1995 found_key.offset == log_key.offset &&
1996 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1997 update_size = false;
2002 * don't drop the conflicting directory entry if the inode
2003 * for the new entry doesn't exist
2008 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
2012 if (key->type == BTRFS_DIR_INDEX_KEY)
2015 btrfs_release_path(path);
2016 if (!ret && update_size) {
2017 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
2018 ret = btrfs_update_inode(trans, root, dir);
2022 if (!ret && name_added)
2027 if (name_in_log_ref(root->log_root, name, name_len,
2028 key->objectid, log_key.objectid)) {
2029 /* The dentry will be added later. */
2031 update_size = false;
2034 btrfs_release_path(path);
2035 ret = insert_one_name(trans, root, key->objectid, key->offset,
2036 name, name_len, &log_key);
2037 if (ret && ret != -ENOENT && ret != -EEXIST)
2041 update_size = false;
2047 * find all the names in a directory item and reconcile them into
2048 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
2049 * one name in a directory item, but the same code gets used for
2050 * both directory index types
2052 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
2053 struct btrfs_root *root,
2054 struct btrfs_path *path,
2055 struct extent_buffer *eb, int slot,
2056 struct btrfs_key *key)
2059 u32 item_size = btrfs_item_size_nr(eb, slot);
2060 struct btrfs_dir_item *di;
2063 unsigned long ptr_end;
2064 struct btrfs_path *fixup_path = NULL;
2066 ptr = btrfs_item_ptr_offset(eb, slot);
2067 ptr_end = ptr + item_size;
2068 while (ptr < ptr_end) {
2069 di = (struct btrfs_dir_item *)ptr;
2070 name_len = btrfs_dir_name_len(eb, di);
2071 ret = replay_one_name(trans, root, path, eb, di, key);
2074 ptr = (unsigned long)(di + 1);
2078 * If this entry refers to a non-directory (directories can not
2079 * have a link count > 1) and it was added in the transaction
2080 * that was not committed, make sure we fixup the link count of
2081 * the inode it the entry points to. Otherwise something like
2082 * the following would result in a directory pointing to an
2083 * inode with a wrong link that does not account for this dir
2091 * ln testdir/bar testdir/bar_link
2092 * ln testdir/foo testdir/foo_link
2093 * xfs_io -c "fsync" testdir/bar
2097 * mount fs, log replay happens
2099 * File foo would remain with a link count of 1 when it has two
2100 * entries pointing to it in the directory testdir. This would
2101 * make it impossible to ever delete the parent directory has
2102 * it would result in stale dentries that can never be deleted.
2104 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
2105 struct btrfs_key di_key;
2108 fixup_path = btrfs_alloc_path();
2115 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
2116 ret = link_to_fixup_dir(trans, root, fixup_path,
2123 btrfs_free_path(fixup_path);
2128 * directory replay has two parts. There are the standard directory
2129 * items in the log copied from the subvolume, and range items
2130 * created in the log while the subvolume was logged.
2132 * The range items tell us which parts of the key space the log
2133 * is authoritative for. During replay, if a key in the subvolume
2134 * directory is in a logged range item, but not actually in the log
2135 * that means it was deleted from the directory before the fsync
2136 * and should be removed.
2138 static noinline int find_dir_range(struct btrfs_root *root,
2139 struct btrfs_path *path,
2140 u64 dirid, int key_type,
2141 u64 *start_ret, u64 *end_ret)
2143 struct btrfs_key key;
2145 struct btrfs_dir_log_item *item;
2149 if (*start_ret == (u64)-1)
2152 key.objectid = dirid;
2153 key.type = key_type;
2154 key.offset = *start_ret;
2156 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2160 if (path->slots[0] == 0)
2165 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2167 if (key.type != key_type || key.objectid != dirid) {
2171 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2172 struct btrfs_dir_log_item);
2173 found_end = btrfs_dir_log_end(path->nodes[0], item);
2175 if (*start_ret >= key.offset && *start_ret <= found_end) {
2177 *start_ret = key.offset;
2178 *end_ret = found_end;
2183 /* check the next slot in the tree to see if it is a valid item */
2184 nritems = btrfs_header_nritems(path->nodes[0]);
2186 if (path->slots[0] >= nritems) {
2187 ret = btrfs_next_leaf(root, path);
2192 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2194 if (key.type != key_type || key.objectid != dirid) {
2198 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2199 struct btrfs_dir_log_item);
2200 found_end = btrfs_dir_log_end(path->nodes[0], item);
2201 *start_ret = key.offset;
2202 *end_ret = found_end;
2205 btrfs_release_path(path);
2210 * this looks for a given directory item in the log. If the directory
2211 * item is not in the log, the item is removed and the inode it points
2214 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2215 struct btrfs_root *root,
2216 struct btrfs_root *log,
2217 struct btrfs_path *path,
2218 struct btrfs_path *log_path,
2220 struct btrfs_key *dir_key)
2223 struct extent_buffer *eb;
2226 struct btrfs_dir_item *di;
2227 struct btrfs_dir_item *log_di;
2230 unsigned long ptr_end;
2232 struct inode *inode;
2233 struct btrfs_key location;
2236 eb = path->nodes[0];
2237 slot = path->slots[0];
2238 item_size = btrfs_item_size_nr(eb, slot);
2239 ptr = btrfs_item_ptr_offset(eb, slot);
2240 ptr_end = ptr + item_size;
2241 while (ptr < ptr_end) {
2242 di = (struct btrfs_dir_item *)ptr;
2243 name_len = btrfs_dir_name_len(eb, di);
2244 name = kmalloc(name_len, GFP_NOFS);
2249 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2252 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2253 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2256 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2257 log_di = btrfs_lookup_dir_index_item(trans, log,
2263 if (!log_di || log_di == ERR_PTR(-ENOENT)) {
2264 btrfs_dir_item_key_to_cpu(eb, di, &location);
2265 btrfs_release_path(path);
2266 btrfs_release_path(log_path);
2267 inode = read_one_inode(root, location.objectid);
2273 ret = link_to_fixup_dir(trans, root,
2274 path, location.objectid);
2282 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2283 BTRFS_I(inode), name, name_len);
2285 ret = btrfs_run_delayed_items(trans);
2291 /* there might still be more names under this key
2292 * check and repeat if required
2294 ret = btrfs_search_slot(NULL, root, dir_key, path,
2300 } else if (IS_ERR(log_di)) {
2302 return PTR_ERR(log_di);
2304 btrfs_release_path(log_path);
2307 ptr = (unsigned long)(di + 1);
2312 btrfs_release_path(path);
2313 btrfs_release_path(log_path);
2317 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2318 struct btrfs_root *root,
2319 struct btrfs_root *log,
2320 struct btrfs_path *path,
2323 struct btrfs_key search_key;
2324 struct btrfs_path *log_path;
2329 log_path = btrfs_alloc_path();
2333 search_key.objectid = ino;
2334 search_key.type = BTRFS_XATTR_ITEM_KEY;
2335 search_key.offset = 0;
2337 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2341 nritems = btrfs_header_nritems(path->nodes[0]);
2342 for (i = path->slots[0]; i < nritems; i++) {
2343 struct btrfs_key key;
2344 struct btrfs_dir_item *di;
2345 struct btrfs_dir_item *log_di;
2349 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2350 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2355 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2356 total_size = btrfs_item_size_nr(path->nodes[0], i);
2358 while (cur < total_size) {
2359 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2360 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2361 u32 this_len = sizeof(*di) + name_len + data_len;
2364 name = kmalloc(name_len, GFP_NOFS);
2369 read_extent_buffer(path->nodes[0], name,
2370 (unsigned long)(di + 1), name_len);
2372 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2374 btrfs_release_path(log_path);
2376 /* Doesn't exist in log tree, so delete it. */
2377 btrfs_release_path(path);
2378 di = btrfs_lookup_xattr(trans, root, path, ino,
2379 name, name_len, -1);
2386 ret = btrfs_delete_one_dir_name(trans, root,
2390 btrfs_release_path(path);
2395 if (IS_ERR(log_di)) {
2396 ret = PTR_ERR(log_di);
2400 di = (struct btrfs_dir_item *)((char *)di + this_len);
2403 ret = btrfs_next_leaf(root, path);
2409 btrfs_free_path(log_path);
2410 btrfs_release_path(path);
2416 * deletion replay happens before we copy any new directory items
2417 * out of the log or out of backreferences from inodes. It
2418 * scans the log to find ranges of keys that log is authoritative for,
2419 * and then scans the directory to find items in those ranges that are
2420 * not present in the log.
2422 * Anything we don't find in the log is unlinked and removed from the
2425 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2426 struct btrfs_root *root,
2427 struct btrfs_root *log,
2428 struct btrfs_path *path,
2429 u64 dirid, int del_all)
2433 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2435 struct btrfs_key dir_key;
2436 struct btrfs_key found_key;
2437 struct btrfs_path *log_path;
2440 dir_key.objectid = dirid;
2441 dir_key.type = BTRFS_DIR_ITEM_KEY;
2442 log_path = btrfs_alloc_path();
2446 dir = read_one_inode(root, dirid);
2447 /* it isn't an error if the inode isn't there, that can happen
2448 * because we replay the deletes before we copy in the inode item
2452 btrfs_free_path(log_path);
2460 range_end = (u64)-1;
2462 ret = find_dir_range(log, path, dirid, key_type,
2463 &range_start, &range_end);
2468 dir_key.offset = range_start;
2471 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2476 nritems = btrfs_header_nritems(path->nodes[0]);
2477 if (path->slots[0] >= nritems) {
2478 ret = btrfs_next_leaf(root, path);
2484 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2486 if (found_key.objectid != dirid ||
2487 found_key.type != dir_key.type)
2490 if (found_key.offset > range_end)
2493 ret = check_item_in_log(trans, root, log, path,
2498 if (found_key.offset == (u64)-1)
2500 dir_key.offset = found_key.offset + 1;
2502 btrfs_release_path(path);
2503 if (range_end == (u64)-1)
2505 range_start = range_end + 1;
2510 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2511 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2512 dir_key.type = BTRFS_DIR_INDEX_KEY;
2513 btrfs_release_path(path);
2517 btrfs_release_path(path);
2518 btrfs_free_path(log_path);
2524 * the process_func used to replay items from the log tree. This
2525 * gets called in two different stages. The first stage just looks
2526 * for inodes and makes sure they are all copied into the subvolume.
2528 * The second stage copies all the other item types from the log into
2529 * the subvolume. The two stage approach is slower, but gets rid of
2530 * lots of complexity around inodes referencing other inodes that exist
2531 * only in the log (references come from either directory items or inode
2534 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2535 struct walk_control *wc, u64 gen, int level)
2538 struct btrfs_path *path;
2539 struct btrfs_root *root = wc->replay_dest;
2540 struct btrfs_key key;
2544 ret = btrfs_read_buffer(eb, gen, level, NULL);
2548 level = btrfs_header_level(eb);
2553 path = btrfs_alloc_path();
2557 nritems = btrfs_header_nritems(eb);
2558 for (i = 0; i < nritems; i++) {
2559 btrfs_item_key_to_cpu(eb, &key, i);
2561 /* inode keys are done during the first stage */
2562 if (key.type == BTRFS_INODE_ITEM_KEY &&
2563 wc->stage == LOG_WALK_REPLAY_INODES) {
2564 struct btrfs_inode_item *inode_item;
2567 inode_item = btrfs_item_ptr(eb, i,
2568 struct btrfs_inode_item);
2570 * If we have a tmpfile (O_TMPFILE) that got fsync'ed
2571 * and never got linked before the fsync, skip it, as
2572 * replaying it is pointless since it would be deleted
2573 * later. We skip logging tmpfiles, but it's always
2574 * possible we are replaying a log created with a kernel
2575 * that used to log tmpfiles.
2577 if (btrfs_inode_nlink(eb, inode_item) == 0) {
2578 wc->ignore_cur_inode = true;
2581 wc->ignore_cur_inode = false;
2583 ret = replay_xattr_deletes(wc->trans, root, log,
2584 path, key.objectid);
2587 mode = btrfs_inode_mode(eb, inode_item);
2588 if (S_ISDIR(mode)) {
2589 ret = replay_dir_deletes(wc->trans,
2590 root, log, path, key.objectid, 0);
2594 ret = overwrite_item(wc->trans, root, path,
2600 * Before replaying extents, truncate the inode to its
2601 * size. We need to do it now and not after log replay
2602 * because before an fsync we can have prealloc extents
2603 * added beyond the inode's i_size. If we did it after,
2604 * through orphan cleanup for example, we would drop
2605 * those prealloc extents just after replaying them.
2607 if (S_ISREG(mode)) {
2608 struct inode *inode;
2611 inode = read_one_inode(root, key.objectid);
2616 from = ALIGN(i_size_read(inode),
2617 root->fs_info->sectorsize);
2618 ret = btrfs_drop_extents(wc->trans, root, inode,
2621 /* Update the inode's nbytes. */
2622 ret = btrfs_update_inode(wc->trans,
2630 ret = link_to_fixup_dir(wc->trans, root,
2631 path, key.objectid);
2636 if (wc->ignore_cur_inode)
2639 if (key.type == BTRFS_DIR_INDEX_KEY &&
2640 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2641 ret = replay_one_dir_item(wc->trans, root, path,
2647 if (wc->stage < LOG_WALK_REPLAY_ALL)
2650 /* these keys are simply copied */
2651 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2652 ret = overwrite_item(wc->trans, root, path,
2656 } else if (key.type == BTRFS_INODE_REF_KEY ||
2657 key.type == BTRFS_INODE_EXTREF_KEY) {
2658 ret = add_inode_ref(wc->trans, root, log, path,
2660 if (ret && ret != -ENOENT)
2663 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2664 ret = replay_one_extent(wc->trans, root, path,
2668 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2669 ret = replay_one_dir_item(wc->trans, root, path,
2675 btrfs_free_path(path);
2679 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2680 struct btrfs_root *root,
2681 struct btrfs_path *path, int *level,
2682 struct walk_control *wc)
2684 struct btrfs_fs_info *fs_info = root->fs_info;
2688 struct extent_buffer *next;
2689 struct extent_buffer *cur;
2690 struct extent_buffer *parent;
2694 WARN_ON(*level < 0);
2695 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2697 while (*level > 0) {
2698 struct btrfs_key first_key;
2700 WARN_ON(*level < 0);
2701 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2702 cur = path->nodes[*level];
2704 WARN_ON(btrfs_header_level(cur) != *level);
2706 if (path->slots[*level] >=
2707 btrfs_header_nritems(cur))
2710 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2711 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2712 btrfs_node_key_to_cpu(cur, &first_key, path->slots[*level]);
2713 blocksize = fs_info->nodesize;
2715 parent = path->nodes[*level];
2716 root_owner = btrfs_header_owner(parent);
2718 next = btrfs_find_create_tree_block(fs_info, bytenr);
2720 return PTR_ERR(next);
2723 ret = wc->process_func(root, next, wc, ptr_gen,
2726 free_extent_buffer(next);
2730 path->slots[*level]++;
2732 ret = btrfs_read_buffer(next, ptr_gen,
2733 *level - 1, &first_key);
2735 free_extent_buffer(next);
2740 btrfs_tree_lock(next);
2741 btrfs_set_lock_blocking_write(next);
2742 btrfs_clean_tree_block(next);
2743 btrfs_wait_tree_block_writeback(next);
2744 btrfs_tree_unlock(next);
2746 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2747 clear_extent_buffer_dirty(next);
2750 WARN_ON(root_owner !=
2751 BTRFS_TREE_LOG_OBJECTID);
2752 ret = btrfs_free_and_pin_reserved_extent(
2756 free_extent_buffer(next);
2760 free_extent_buffer(next);
2763 ret = btrfs_read_buffer(next, ptr_gen, *level - 1, &first_key);
2765 free_extent_buffer(next);
2769 WARN_ON(*level <= 0);
2770 if (path->nodes[*level-1])
2771 free_extent_buffer(path->nodes[*level-1]);
2772 path->nodes[*level-1] = next;
2773 *level = btrfs_header_level(next);
2774 path->slots[*level] = 0;
2777 WARN_ON(*level < 0);
2778 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2780 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2786 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2787 struct btrfs_root *root,
2788 struct btrfs_path *path, int *level,
2789 struct walk_control *wc)
2791 struct btrfs_fs_info *fs_info = root->fs_info;
2797 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2798 slot = path->slots[i];
2799 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2802 WARN_ON(*level == 0);
2805 struct extent_buffer *parent;
2806 if (path->nodes[*level] == root->node)
2807 parent = path->nodes[*level];
2809 parent = path->nodes[*level + 1];
2811 root_owner = btrfs_header_owner(parent);
2812 ret = wc->process_func(root, path->nodes[*level], wc,
2813 btrfs_header_generation(path->nodes[*level]),
2819 struct extent_buffer *next;
2821 next = path->nodes[*level];
2824 btrfs_tree_lock(next);
2825 btrfs_set_lock_blocking_write(next);
2826 btrfs_clean_tree_block(next);
2827 btrfs_wait_tree_block_writeback(next);
2828 btrfs_tree_unlock(next);
2830 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2831 clear_extent_buffer_dirty(next);
2834 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2835 ret = btrfs_free_and_pin_reserved_extent(
2837 path->nodes[*level]->start,
2838 path->nodes[*level]->len);
2842 free_extent_buffer(path->nodes[*level]);
2843 path->nodes[*level] = NULL;
2851 * drop the reference count on the tree rooted at 'snap'. This traverses
2852 * the tree freeing any blocks that have a ref count of zero after being
2855 static int walk_log_tree(struct btrfs_trans_handle *trans,
2856 struct btrfs_root *log, struct walk_control *wc)
2858 struct btrfs_fs_info *fs_info = log->fs_info;
2862 struct btrfs_path *path;
2865 path = btrfs_alloc_path();
2869 level = btrfs_header_level(log->node);
2871 path->nodes[level] = log->node;
2872 extent_buffer_get(log->node);
2873 path->slots[level] = 0;
2876 wret = walk_down_log_tree(trans, log, path, &level, wc);
2884 wret = walk_up_log_tree(trans, log, path, &level, wc);
2893 /* was the root node processed? if not, catch it here */
2894 if (path->nodes[orig_level]) {
2895 ret = wc->process_func(log, path->nodes[orig_level], wc,
2896 btrfs_header_generation(path->nodes[orig_level]),
2901 struct extent_buffer *next;
2903 next = path->nodes[orig_level];
2906 btrfs_tree_lock(next);
2907 btrfs_set_lock_blocking_write(next);
2908 btrfs_clean_tree_block(next);
2909 btrfs_wait_tree_block_writeback(next);
2910 btrfs_tree_unlock(next);
2912 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &next->bflags))
2913 clear_extent_buffer_dirty(next);
2916 WARN_ON(log->root_key.objectid !=
2917 BTRFS_TREE_LOG_OBJECTID);
2918 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2919 next->start, next->len);
2926 btrfs_free_path(path);
2931 * helper function to update the item for a given subvolumes log root
2932 * in the tree of log roots
2934 static int update_log_root(struct btrfs_trans_handle *trans,
2935 struct btrfs_root *log,
2936 struct btrfs_root_item *root_item)
2938 struct btrfs_fs_info *fs_info = log->fs_info;
2941 if (log->log_transid == 1) {
2942 /* insert root item on the first sync */
2943 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2944 &log->root_key, root_item);
2946 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2947 &log->root_key, root_item);
2952 static void wait_log_commit(struct btrfs_root *root, int transid)
2955 int index = transid % 2;
2958 * we only allow two pending log transactions at a time,
2959 * so we know that if ours is more than 2 older than the
2960 * current transaction, we're done
2963 prepare_to_wait(&root->log_commit_wait[index],
2964 &wait, TASK_UNINTERRUPTIBLE);
2966 if (!(root->log_transid_committed < transid &&
2967 atomic_read(&root->log_commit[index])))
2970 mutex_unlock(&root->log_mutex);
2972 mutex_lock(&root->log_mutex);
2974 finish_wait(&root->log_commit_wait[index], &wait);
2977 static void wait_for_writer(struct btrfs_root *root)
2982 prepare_to_wait(&root->log_writer_wait, &wait,
2983 TASK_UNINTERRUPTIBLE);
2984 if (!atomic_read(&root->log_writers))
2987 mutex_unlock(&root->log_mutex);
2989 mutex_lock(&root->log_mutex);
2991 finish_wait(&root->log_writer_wait, &wait);
2994 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2995 struct btrfs_log_ctx *ctx)
3000 mutex_lock(&root->log_mutex);
3001 list_del_init(&ctx->list);
3002 mutex_unlock(&root->log_mutex);
3006 * Invoked in log mutex context, or be sure there is no other task which
3007 * can access the list.
3009 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
3010 int index, int error)
3012 struct btrfs_log_ctx *ctx;
3013 struct btrfs_log_ctx *safe;
3015 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
3016 list_del_init(&ctx->list);
3017 ctx->log_ret = error;
3020 INIT_LIST_HEAD(&root->log_ctxs[index]);
3024 * btrfs_sync_log does sends a given tree log down to the disk and
3025 * updates the super blocks to record it. When this call is done,
3026 * you know that any inodes previously logged are safely on disk only
3029 * Any other return value means you need to call btrfs_commit_transaction.
3030 * Some of the edge cases for fsyncing directories that have had unlinks
3031 * or renames done in the past mean that sometimes the only safe
3032 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
3033 * that has happened.
3035 int btrfs_sync_log(struct btrfs_trans_handle *trans,
3036 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
3042 struct btrfs_fs_info *fs_info = root->fs_info;
3043 struct btrfs_root *log = root->log_root;
3044 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
3045 struct btrfs_root_item new_root_item;
3046 int log_transid = 0;
3047 struct btrfs_log_ctx root_log_ctx;
3048 struct blk_plug plug;
3050 mutex_lock(&root->log_mutex);
3051 log_transid = ctx->log_transid;
3052 if (root->log_transid_committed >= log_transid) {
3053 mutex_unlock(&root->log_mutex);
3054 return ctx->log_ret;
3057 index1 = log_transid % 2;
3058 if (atomic_read(&root->log_commit[index1])) {
3059 wait_log_commit(root, log_transid);
3060 mutex_unlock(&root->log_mutex);
3061 return ctx->log_ret;
3063 ASSERT(log_transid == root->log_transid);
3064 atomic_set(&root->log_commit[index1], 1);
3066 /* wait for previous tree log sync to complete */
3067 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
3068 wait_log_commit(root, log_transid - 1);
3071 int batch = atomic_read(&root->log_batch);
3072 /* when we're on an ssd, just kick the log commit out */
3073 if (!btrfs_test_opt(fs_info, SSD) &&
3074 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
3075 mutex_unlock(&root->log_mutex);
3076 schedule_timeout_uninterruptible(1);
3077 mutex_lock(&root->log_mutex);
3079 wait_for_writer(root);
3080 if (batch == atomic_read(&root->log_batch))
3084 /* bail out if we need to do a full commit */
3085 if (btrfs_need_log_full_commit(trans)) {
3087 mutex_unlock(&root->log_mutex);
3091 if (log_transid % 2 == 0)
3092 mark = EXTENT_DIRTY;
3096 /* we start IO on all the marked extents here, but we don't actually
3097 * wait for them until later.
3099 blk_start_plug(&plug);
3100 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
3102 blk_finish_plug(&plug);
3103 btrfs_abort_transaction(trans, ret);
3104 btrfs_set_log_full_commit(trans);
3105 mutex_unlock(&root->log_mutex);
3110 * We _must_ update under the root->log_mutex in order to make sure we
3111 * have a consistent view of the log root we are trying to commit at
3114 * We _must_ copy this into a local copy, because we are not holding the
3115 * log_root_tree->log_mutex yet. This is important because when we
3116 * commit the log_root_tree we must have a consistent view of the
3117 * log_root_tree when we update the super block to point at the
3118 * log_root_tree bytenr. If we update the log_root_tree here we'll race
3119 * with the commit and possibly point at the new block which we may not
3122 btrfs_set_root_node(&log->root_item, log->node);
3123 memcpy(&new_root_item, &log->root_item, sizeof(new_root_item));
3125 root->log_transid++;
3126 log->log_transid = root->log_transid;
3127 root->log_start_pid = 0;
3129 * IO has been started, blocks of the log tree have WRITTEN flag set
3130 * in their headers. new modifications of the log will be written to
3131 * new positions. so it's safe to allow log writers to go in.
3133 mutex_unlock(&root->log_mutex);
3135 btrfs_init_log_ctx(&root_log_ctx, NULL);
3137 mutex_lock(&log_root_tree->log_mutex);
3138 atomic_inc(&log_root_tree->log_batch);
3139 atomic_inc(&log_root_tree->log_writers);
3141 index2 = log_root_tree->log_transid % 2;
3142 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
3143 root_log_ctx.log_transid = log_root_tree->log_transid;
3145 mutex_unlock(&log_root_tree->log_mutex);
3147 mutex_lock(&log_root_tree->log_mutex);
3150 * Now we are safe to update the log_root_tree because we're under the
3151 * log_mutex, and we're a current writer so we're holding the commit
3152 * open until we drop the log_mutex.
3154 ret = update_log_root(trans, log, &new_root_item);
3156 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
3157 /* atomic_dec_and_test implies a barrier */
3158 cond_wake_up_nomb(&log_root_tree->log_writer_wait);
3162 if (!list_empty(&root_log_ctx.list))
3163 list_del_init(&root_log_ctx.list);
3165 blk_finish_plug(&plug);
3166 btrfs_set_log_full_commit(trans);
3168 if (ret != -ENOSPC) {
3169 btrfs_abort_transaction(trans, ret);
3170 mutex_unlock(&log_root_tree->log_mutex);
3173 btrfs_wait_tree_log_extents(log, mark);
3174 mutex_unlock(&log_root_tree->log_mutex);
3179 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
3180 blk_finish_plug(&plug);
3181 list_del_init(&root_log_ctx.list);
3182 mutex_unlock(&log_root_tree->log_mutex);
3183 ret = root_log_ctx.log_ret;
3187 index2 = root_log_ctx.log_transid % 2;
3188 if (atomic_read(&log_root_tree->log_commit[index2])) {
3189 blk_finish_plug(&plug);
3190 ret = btrfs_wait_tree_log_extents(log, mark);
3191 wait_log_commit(log_root_tree,
3192 root_log_ctx.log_transid);
3193 mutex_unlock(&log_root_tree->log_mutex);
3195 ret = root_log_ctx.log_ret;
3198 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
3199 atomic_set(&log_root_tree->log_commit[index2], 1);
3201 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
3202 wait_log_commit(log_root_tree,
3203 root_log_ctx.log_transid - 1);
3206 wait_for_writer(log_root_tree);
3209 * now that we've moved on to the tree of log tree roots,
3210 * check the full commit flag again
3212 if (btrfs_need_log_full_commit(trans)) {
3213 blk_finish_plug(&plug);
3214 btrfs_wait_tree_log_extents(log, mark);
3215 mutex_unlock(&log_root_tree->log_mutex);
3217 goto out_wake_log_root;
3220 ret = btrfs_write_marked_extents(fs_info,
3221 &log_root_tree->dirty_log_pages,
3222 EXTENT_DIRTY | EXTENT_NEW);
3223 blk_finish_plug(&plug);
3225 btrfs_set_log_full_commit(trans);
3226 btrfs_abort_transaction(trans, ret);
3227 mutex_unlock(&log_root_tree->log_mutex);
3228 goto out_wake_log_root;
3230 ret = btrfs_wait_tree_log_extents(log, mark);
3232 ret = btrfs_wait_tree_log_extents(log_root_tree,
3233 EXTENT_NEW | EXTENT_DIRTY);
3235 btrfs_set_log_full_commit(trans);
3236 mutex_unlock(&log_root_tree->log_mutex);
3237 goto out_wake_log_root;
3240 btrfs_set_super_log_root(fs_info->super_for_commit,
3241 log_root_tree->node->start);
3242 btrfs_set_super_log_root_level(fs_info->super_for_commit,
3243 btrfs_header_level(log_root_tree->node));
3245 log_root_tree->log_transid++;
3246 mutex_unlock(&log_root_tree->log_mutex);
3249 * Nobody else is going to jump in and write the ctree
3250 * super here because the log_commit atomic below is protecting
3251 * us. We must be called with a transaction handle pinning
3252 * the running transaction open, so a full commit can't hop
3253 * in and cause problems either.
3255 ret = write_all_supers(fs_info, 1);
3257 btrfs_set_log_full_commit(trans);
3258 btrfs_abort_transaction(trans, ret);
3259 goto out_wake_log_root;
3262 mutex_lock(&root->log_mutex);
3263 if (root->last_log_commit < log_transid)
3264 root->last_log_commit = log_transid;
3265 mutex_unlock(&root->log_mutex);
3268 mutex_lock(&log_root_tree->log_mutex);
3269 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3271 log_root_tree->log_transid_committed++;
3272 atomic_set(&log_root_tree->log_commit[index2], 0);
3273 mutex_unlock(&log_root_tree->log_mutex);
3276 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3277 * all the updates above are seen by the woken threads. It might not be
3278 * necessary, but proving that seems to be hard.
3280 cond_wake_up(&log_root_tree->log_commit_wait[index2]);
3282 mutex_lock(&root->log_mutex);
3283 btrfs_remove_all_log_ctxs(root, index1, ret);
3284 root->log_transid_committed++;
3285 atomic_set(&root->log_commit[index1], 0);
3286 mutex_unlock(&root->log_mutex);
3289 * The barrier before waitqueue_active (in cond_wake_up) is needed so
3290 * all the updates above are seen by the woken threads. It might not be
3291 * necessary, but proving that seems to be hard.
3293 cond_wake_up(&root->log_commit_wait[index1]);
3297 static void free_log_tree(struct btrfs_trans_handle *trans,
3298 struct btrfs_root *log)
3301 struct walk_control wc = {
3303 .process_func = process_one_buffer
3306 ret = walk_log_tree(trans, log, &wc);
3309 btrfs_abort_transaction(trans, ret);
3311 btrfs_handle_fs_error(log->fs_info, ret, NULL);
3314 clear_extent_bits(&log->dirty_log_pages, 0, (u64)-1,
3315 EXTENT_DIRTY | EXTENT_NEW | EXTENT_NEED_WAIT);
3316 free_extent_buffer(log->node);
3321 * free all the extents used by the tree log. This should be called
3322 * at commit time of the full transaction
3324 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3326 if (root->log_root) {
3327 free_log_tree(trans, root->log_root);
3328 root->log_root = NULL;
3333 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3334 struct btrfs_fs_info *fs_info)
3336 if (fs_info->log_root_tree) {
3337 free_log_tree(trans, fs_info->log_root_tree);
3338 fs_info->log_root_tree = NULL;
3344 * Check if an inode was logged in the current transaction. We can't always rely
3345 * on an inode's logged_trans value, because it's an in-memory only field and
3346 * therefore not persisted. This means that its value is lost if the inode gets
3347 * evicted and loaded again from disk (in which case it has a value of 0, and
3348 * certainly it is smaller then any possible transaction ID), when that happens
3349 * the full_sync flag is set in the inode's runtime flags, so on that case we
3350 * assume eviction happened and ignore the logged_trans value, assuming the
3351 * worst case, that the inode was logged before in the current transaction.
3353 static bool inode_logged(struct btrfs_trans_handle *trans,
3354 struct btrfs_inode *inode)
3356 if (inode->logged_trans == trans->transid)
3359 if (inode->last_trans == trans->transid &&
3360 test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags) &&
3361 !test_bit(BTRFS_FS_LOG_RECOVERING, &trans->fs_info->flags))
3368 * If both a file and directory are logged, and unlinks or renames are
3369 * mixed in, we have a few interesting corners:
3371 * create file X in dir Y
3372 * link file X to X.link in dir Y
3374 * unlink file X but leave X.link
3377 * After a crash we would expect only X.link to exist. But file X
3378 * didn't get fsync'd again so the log has back refs for X and X.link.
3380 * We solve this by removing directory entries and inode backrefs from the
3381 * log when a file that was logged in the current transaction is
3382 * unlinked. Any later fsync will include the updated log entries, and
3383 * we'll be able to reconstruct the proper directory items from backrefs.
3385 * This optimizations allows us to avoid relogging the entire inode
3386 * or the entire directory.
3388 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3389 struct btrfs_root *root,
3390 const char *name, int name_len,
3391 struct btrfs_inode *dir, u64 index)
3393 struct btrfs_root *log;
3394 struct btrfs_dir_item *di;
3395 struct btrfs_path *path;
3399 u64 dir_ino = btrfs_ino(dir);
3401 if (!inode_logged(trans, dir))
3404 ret = join_running_log_trans(root);
3408 mutex_lock(&dir->log_mutex);
3410 log = root->log_root;
3411 path = btrfs_alloc_path();
3417 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3418 name, name_len, -1);
3424 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3425 bytes_del += name_len;
3431 btrfs_release_path(path);
3432 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3433 index, name, name_len, -1);
3439 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3440 bytes_del += name_len;
3447 /* update the directory size in the log to reflect the names
3451 struct btrfs_key key;
3453 key.objectid = dir_ino;
3455 key.type = BTRFS_INODE_ITEM_KEY;
3456 btrfs_release_path(path);
3458 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3464 struct btrfs_inode_item *item;
3467 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3468 struct btrfs_inode_item);
3469 i_size = btrfs_inode_size(path->nodes[0], item);
3470 if (i_size > bytes_del)
3471 i_size -= bytes_del;
3474 btrfs_set_inode_size(path->nodes[0], item, i_size);
3475 btrfs_mark_buffer_dirty(path->nodes[0]);
3478 btrfs_release_path(path);
3481 btrfs_free_path(path);
3483 mutex_unlock(&dir->log_mutex);
3484 if (ret == -ENOSPC) {
3485 btrfs_set_log_full_commit(trans);
3488 btrfs_abort_transaction(trans, ret);
3490 btrfs_end_log_trans(root);
3495 /* see comments for btrfs_del_dir_entries_in_log */
3496 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3497 struct btrfs_root *root,
3498 const char *name, int name_len,
3499 struct btrfs_inode *inode, u64 dirid)
3501 struct btrfs_root *log;
3505 if (!inode_logged(trans, inode))
3508 ret = join_running_log_trans(root);
3511 log = root->log_root;
3512 mutex_lock(&inode->log_mutex);
3514 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3516 mutex_unlock(&inode->log_mutex);
3517 if (ret == -ENOSPC) {
3518 btrfs_set_log_full_commit(trans);
3520 } else if (ret < 0 && ret != -ENOENT)
3521 btrfs_abort_transaction(trans, ret);
3522 btrfs_end_log_trans(root);
3528 * creates a range item in the log for 'dirid'. first_offset and
3529 * last_offset tell us which parts of the key space the log should
3530 * be considered authoritative for.
3532 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3533 struct btrfs_root *log,
3534 struct btrfs_path *path,
3535 int key_type, u64 dirid,
3536 u64 first_offset, u64 last_offset)
3539 struct btrfs_key key;
3540 struct btrfs_dir_log_item *item;
3542 key.objectid = dirid;
3543 key.offset = first_offset;
3544 if (key_type == BTRFS_DIR_ITEM_KEY)
3545 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3547 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3548 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3552 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3553 struct btrfs_dir_log_item);
3554 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3555 btrfs_mark_buffer_dirty(path->nodes[0]);
3556 btrfs_release_path(path);
3561 * log all the items included in the current transaction for a given
3562 * directory. This also creates the range items in the log tree required
3563 * to replay anything deleted before the fsync
3565 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3566 struct btrfs_root *root, struct btrfs_inode *inode,
3567 struct btrfs_path *path,
3568 struct btrfs_path *dst_path, int key_type,
3569 struct btrfs_log_ctx *ctx,
3570 u64 min_offset, u64 *last_offset_ret)
3572 struct btrfs_key min_key;
3573 struct btrfs_root *log = root->log_root;
3574 struct extent_buffer *src;
3579 u64 first_offset = min_offset;
3580 u64 last_offset = (u64)-1;
3581 u64 ino = btrfs_ino(inode);
3583 log = root->log_root;
3585 min_key.objectid = ino;
3586 min_key.type = key_type;
3587 min_key.offset = min_offset;
3589 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3592 * we didn't find anything from this transaction, see if there
3593 * is anything at all
3595 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3596 min_key.objectid = ino;
3597 min_key.type = key_type;
3598 min_key.offset = (u64)-1;
3599 btrfs_release_path(path);
3600 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3602 btrfs_release_path(path);
3605 ret = btrfs_previous_item(root, path, ino, key_type);
3607 /* if ret == 0 there are items for this type,
3608 * create a range to tell us the last key of this type.
3609 * otherwise, there are no items in this directory after
3610 * *min_offset, and we create a range to indicate that.
3613 struct btrfs_key tmp;
3614 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3616 if (key_type == tmp.type)
3617 first_offset = max(min_offset, tmp.offset) + 1;
3622 /* go backward to find any previous key */
3623 ret = btrfs_previous_item(root, path, ino, key_type);
3625 struct btrfs_key tmp;
3626 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3627 if (key_type == tmp.type) {
3628 first_offset = tmp.offset;
3629 ret = overwrite_item(trans, log, dst_path,
3630 path->nodes[0], path->slots[0],
3638 btrfs_release_path(path);
3641 * Find the first key from this transaction again. See the note for
3642 * log_new_dir_dentries, if we're logging a directory recursively we
3643 * won't be holding its i_mutex, which means we can modify the directory
3644 * while we're logging it. If we remove an entry between our first
3645 * search and this search we'll not find the key again and can just
3648 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3653 * we have a block from this transaction, log every item in it
3654 * from our directory
3657 struct btrfs_key tmp;
3658 src = path->nodes[0];
3659 nritems = btrfs_header_nritems(src);
3660 for (i = path->slots[0]; i < nritems; i++) {
3661 struct btrfs_dir_item *di;
3663 btrfs_item_key_to_cpu(src, &min_key, i);
3665 if (min_key.objectid != ino || min_key.type != key_type)
3667 ret = overwrite_item(trans, log, dst_path, src, i,
3675 * We must make sure that when we log a directory entry,
3676 * the corresponding inode, after log replay, has a
3677 * matching link count. For example:
3683 * xfs_io -c "fsync" mydir
3685 * <mount fs and log replay>
3687 * Would result in a fsync log that when replayed, our
3688 * file inode would have a link count of 1, but we get
3689 * two directory entries pointing to the same inode.
3690 * After removing one of the names, it would not be
3691 * possible to remove the other name, which resulted
3692 * always in stale file handle errors, and would not
3693 * be possible to rmdir the parent directory, since
3694 * its i_size could never decrement to the value
3695 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3697 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3698 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3700 (btrfs_dir_transid(src, di) == trans->transid ||
3701 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3702 tmp.type != BTRFS_ROOT_ITEM_KEY)
3703 ctx->log_new_dentries = true;
3705 path->slots[0] = nritems;
3708 * look ahead to the next item and see if it is also
3709 * from this directory and from this transaction
3711 ret = btrfs_next_leaf(root, path);
3714 last_offset = (u64)-1;
3719 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3720 if (tmp.objectid != ino || tmp.type != key_type) {
3721 last_offset = (u64)-1;
3724 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3725 ret = overwrite_item(trans, log, dst_path,
3726 path->nodes[0], path->slots[0],
3731 last_offset = tmp.offset;
3736 btrfs_release_path(path);
3737 btrfs_release_path(dst_path);
3740 *last_offset_ret = last_offset;
3742 * insert the log range keys to indicate where the log
3745 ret = insert_dir_log_key(trans, log, path, key_type,
3746 ino, first_offset, last_offset);
3754 * logging directories is very similar to logging inodes, We find all the items
3755 * from the current transaction and write them to the log.
3757 * The recovery code scans the directory in the subvolume, and if it finds a
3758 * key in the range logged that is not present in the log tree, then it means
3759 * that dir entry was unlinked during the transaction.
3761 * In order for that scan to work, we must include one key smaller than
3762 * the smallest logged by this transaction and one key larger than the largest
3763 * key logged by this transaction.
3765 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3766 struct btrfs_root *root, struct btrfs_inode *inode,
3767 struct btrfs_path *path,
3768 struct btrfs_path *dst_path,
3769 struct btrfs_log_ctx *ctx)
3774 int key_type = BTRFS_DIR_ITEM_KEY;
3780 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3781 ctx, min_key, &max_key);
3784 if (max_key == (u64)-1)
3786 min_key = max_key + 1;
3789 if (key_type == BTRFS_DIR_ITEM_KEY) {
3790 key_type = BTRFS_DIR_INDEX_KEY;
3797 * a helper function to drop items from the log before we relog an
3798 * inode. max_key_type indicates the highest item type to remove.
3799 * This cannot be run for file data extents because it does not
3800 * free the extents they point to.
3802 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3803 struct btrfs_root *log,
3804 struct btrfs_path *path,
3805 u64 objectid, int max_key_type)
3808 struct btrfs_key key;
3809 struct btrfs_key found_key;
3812 key.objectid = objectid;
3813 key.type = max_key_type;
3814 key.offset = (u64)-1;
3817 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3818 BUG_ON(ret == 0); /* Logic error */
3822 if (path->slots[0] == 0)
3826 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3829 if (found_key.objectid != objectid)
3832 found_key.offset = 0;
3834 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3839 ret = btrfs_del_items(trans, log, path, start_slot,
3840 path->slots[0] - start_slot + 1);
3842 * If start slot isn't 0 then we don't need to re-search, we've
3843 * found the last guy with the objectid in this tree.
3845 if (ret || start_slot != 0)
3847 btrfs_release_path(path);
3849 btrfs_release_path(path);
3855 static void fill_inode_item(struct btrfs_trans_handle *trans,
3856 struct extent_buffer *leaf,
3857 struct btrfs_inode_item *item,
3858 struct inode *inode, int log_inode_only,
3861 struct btrfs_map_token token;
3863 btrfs_init_map_token(&token, leaf);
3865 if (log_inode_only) {
3866 /* set the generation to zero so the recover code
3867 * can tell the difference between an logging
3868 * just to say 'this inode exists' and a logging
3869 * to say 'update this inode with these values'
3871 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3872 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3874 btrfs_set_token_inode_generation(leaf, item,
3875 BTRFS_I(inode)->generation,
3877 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3880 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3881 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3882 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3883 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3885 btrfs_set_token_timespec_sec(leaf, &item->atime,
3886 inode->i_atime.tv_sec, &token);
3887 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3888 inode->i_atime.tv_nsec, &token);
3890 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3891 inode->i_mtime.tv_sec, &token);
3892 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3893 inode->i_mtime.tv_nsec, &token);
3895 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3896 inode->i_ctime.tv_sec, &token);
3897 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3898 inode->i_ctime.tv_nsec, &token);
3900 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3903 btrfs_set_token_inode_sequence(leaf, item,
3904 inode_peek_iversion(inode), &token);
3905 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3906 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3907 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3908 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3911 static int log_inode_item(struct btrfs_trans_handle *trans,
3912 struct btrfs_root *log, struct btrfs_path *path,
3913 struct btrfs_inode *inode)
3915 struct btrfs_inode_item *inode_item;
3918 ret = btrfs_insert_empty_item(trans, log, path,
3919 &inode->location, sizeof(*inode_item));
3920 if (ret && ret != -EEXIST)
3922 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3923 struct btrfs_inode_item);
3924 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3926 btrfs_release_path(path);
3930 static noinline int copy_items(struct btrfs_trans_handle *trans,
3931 struct btrfs_inode *inode,
3932 struct btrfs_path *dst_path,
3933 struct btrfs_path *src_path, u64 *last_extent,
3934 int start_slot, int nr, int inode_only,
3937 struct btrfs_fs_info *fs_info = trans->fs_info;
3938 unsigned long src_offset;
3939 unsigned long dst_offset;
3940 struct btrfs_root *log = inode->root->log_root;
3941 struct btrfs_file_extent_item *extent;
3942 struct btrfs_inode_item *inode_item;
3943 struct extent_buffer *src = src_path->nodes[0];
3944 struct btrfs_key first_key, last_key, key;
3946 struct btrfs_key *ins_keys;
3950 struct list_head ordered_sums;
3951 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3952 bool has_extents = false;
3953 bool need_find_last_extent = true;
3956 INIT_LIST_HEAD(&ordered_sums);
3958 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3959 nr * sizeof(u32), GFP_NOFS);
3963 first_key.objectid = (u64)-1;
3965 ins_sizes = (u32 *)ins_data;
3966 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3968 for (i = 0; i < nr; i++) {
3969 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3970 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3972 ret = btrfs_insert_empty_items(trans, log, dst_path,
3973 ins_keys, ins_sizes, nr);
3979 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3980 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3981 dst_path->slots[0]);
3983 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3986 last_key = ins_keys[i];
3988 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3989 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3991 struct btrfs_inode_item);
3992 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3994 inode_only == LOG_INODE_EXISTS,
3997 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3998 src_offset, ins_sizes[i]);
4002 * We set need_find_last_extent here in case we know we were
4003 * processing other items and then walk into the first extent in
4004 * the inode. If we don't hit an extent then nothing changes,
4005 * we'll do the last search the next time around.
4007 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
4009 if (first_key.objectid == (u64)-1)
4010 first_key = ins_keys[i];
4012 need_find_last_extent = false;
4015 /* take a reference on file data extents so that truncates
4016 * or deletes of this inode don't have to relog the inode
4019 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
4022 extent = btrfs_item_ptr(src, start_slot + i,
4023 struct btrfs_file_extent_item);
4025 if (btrfs_file_extent_generation(src, extent) < trans->transid)
4028 found_type = btrfs_file_extent_type(src, extent);
4029 if (found_type == BTRFS_FILE_EXTENT_REG) {
4031 ds = btrfs_file_extent_disk_bytenr(src,
4033 /* ds == 0 is a hole */
4037 dl = btrfs_file_extent_disk_num_bytes(src,
4039 cs = btrfs_file_extent_offset(src, extent);
4040 cl = btrfs_file_extent_num_bytes(src,
4042 if (btrfs_file_extent_compression(src,
4048 ret = btrfs_lookup_csums_range(
4050 ds + cs, ds + cs + cl - 1,
4053 btrfs_release_path(dst_path);
4061 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
4062 btrfs_release_path(dst_path);
4066 * we have to do this after the loop above to avoid changing the
4067 * log tree while trying to change the log tree.
4070 while (!list_empty(&ordered_sums)) {
4071 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4072 struct btrfs_ordered_sum,
4075 ret = btrfs_csum_file_blocks(trans, log, sums);
4076 list_del(&sums->list);
4083 if (need_find_last_extent && *last_extent == first_key.offset) {
4085 * We don't have any leafs between our current one and the one
4086 * we processed before that can have file extent items for our
4087 * inode (and have a generation number smaller than our current
4090 need_find_last_extent = false;
4094 * Because we use btrfs_search_forward we could skip leaves that were
4095 * not modified and then assume *last_extent is valid when it really
4096 * isn't. So back up to the previous leaf and read the end of the last
4097 * extent before we go and fill in holes.
4099 if (need_find_last_extent) {
4102 ret = btrfs_prev_leaf(inode->root, src_path);
4107 if (src_path->slots[0])
4108 src_path->slots[0]--;
4109 src = src_path->nodes[0];
4110 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
4111 if (key.objectid != btrfs_ino(inode) ||
4112 key.type != BTRFS_EXTENT_DATA_KEY)
4114 extent = btrfs_item_ptr(src, src_path->slots[0],
4115 struct btrfs_file_extent_item);
4116 if (btrfs_file_extent_type(src, extent) ==
4117 BTRFS_FILE_EXTENT_INLINE) {
4118 len = btrfs_file_extent_ram_bytes(src, extent);
4119 *last_extent = ALIGN(key.offset + len,
4120 fs_info->sectorsize);
4122 len = btrfs_file_extent_num_bytes(src, extent);
4123 *last_extent = key.offset + len;
4127 /* So we did prev_leaf, now we need to move to the next leaf, but a few
4128 * things could have happened
4130 * 1) A merge could have happened, so we could currently be on a leaf
4131 * that holds what we were copying in the first place.
4132 * 2) A split could have happened, and now not all of the items we want
4133 * are on the same leaf.
4135 * So we need to adjust how we search for holes, we need to drop the
4136 * path and re-search for the first extent key we found, and then walk
4137 * forward until we hit the last one we copied.
4139 if (need_find_last_extent) {
4140 /* btrfs_prev_leaf could return 1 without releasing the path */
4141 btrfs_release_path(src_path);
4142 ret = btrfs_search_slot(NULL, inode->root, &first_key,
4147 src = src_path->nodes[0];
4148 i = src_path->slots[0];
4154 * Ok so here we need to go through and fill in any holes we may have
4155 * to make sure that holes are punched for those areas in case they had
4156 * extents previously.
4162 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
4163 ret = btrfs_next_leaf(inode->root, src_path);
4167 src = src_path->nodes[0];
4169 need_find_last_extent = true;
4172 btrfs_item_key_to_cpu(src, &key, i);
4173 if (!btrfs_comp_cpu_keys(&key, &last_key))
4175 if (key.objectid != btrfs_ino(inode) ||
4176 key.type != BTRFS_EXTENT_DATA_KEY) {
4180 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
4181 if (btrfs_file_extent_type(src, extent) ==
4182 BTRFS_FILE_EXTENT_INLINE) {
4183 len = btrfs_file_extent_ram_bytes(src, extent);
4184 extent_end = ALIGN(key.offset + len,
4185 fs_info->sectorsize);
4187 len = btrfs_file_extent_num_bytes(src, extent);
4188 extent_end = key.offset + len;
4192 if (*last_extent == key.offset) {
4193 *last_extent = extent_end;
4196 offset = *last_extent;
4197 len = key.offset - *last_extent;
4198 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
4199 offset, 0, 0, len, 0, len, 0, 0, 0);
4202 *last_extent = extent_end;
4206 * Check if there is a hole between the last extent found in our leaf
4207 * and the first extent in the next leaf. If there is one, we need to
4208 * log an explicit hole so that at replay time we can punch the hole.
4211 key.objectid == btrfs_ino(inode) &&
4212 key.type == BTRFS_EXTENT_DATA_KEY &&
4213 i == btrfs_header_nritems(src_path->nodes[0])) {
4214 ret = btrfs_next_leaf(inode->root, src_path);
4215 need_find_last_extent = true;
4218 } else if (ret == 0) {
4219 btrfs_item_key_to_cpu(src_path->nodes[0], &key,
4220 src_path->slots[0]);
4221 if (key.objectid == btrfs_ino(inode) &&
4222 key.type == BTRFS_EXTENT_DATA_KEY &&
4223 *last_extent < key.offset) {
4224 const u64 len = key.offset - *last_extent;
4226 ret = btrfs_insert_file_extent(trans, log,
4231 *last_extent += len;
4236 * Need to let the callers know we dropped the path so they should
4239 if (!ret && need_find_last_extent)
4244 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
4246 struct extent_map *em1, *em2;
4248 em1 = list_entry(a, struct extent_map, list);
4249 em2 = list_entry(b, struct extent_map, list);
4251 if (em1->start < em2->start)
4253 else if (em1->start > em2->start)
4258 static int log_extent_csums(struct btrfs_trans_handle *trans,
4259 struct btrfs_inode *inode,
4260 struct btrfs_root *log_root,
4261 const struct extent_map *em)
4265 LIST_HEAD(ordered_sums);
4268 if (inode->flags & BTRFS_INODE_NODATASUM ||
4269 test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
4270 em->block_start == EXTENT_MAP_HOLE)
4273 /* If we're compressed we have to save the entire range of csums. */
4274 if (em->compress_type) {
4276 csum_len = max(em->block_len, em->orig_block_len);
4278 csum_offset = em->mod_start - em->start;
4279 csum_len = em->mod_len;
4282 /* block start is already adjusted for the file extent offset. */
4283 ret = btrfs_lookup_csums_range(trans->fs_info->csum_root,
4284 em->block_start + csum_offset,
4285 em->block_start + csum_offset +
4286 csum_len - 1, &ordered_sums, 0);
4290 while (!list_empty(&ordered_sums)) {
4291 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4292 struct btrfs_ordered_sum,
4295 ret = btrfs_csum_file_blocks(trans, log_root, sums);
4296 list_del(&sums->list);
4303 static int log_one_extent(struct btrfs_trans_handle *trans,
4304 struct btrfs_inode *inode, struct btrfs_root *root,
4305 const struct extent_map *em,
4306 struct btrfs_path *path,
4307 struct btrfs_log_ctx *ctx)
4309 struct btrfs_root *log = root->log_root;
4310 struct btrfs_file_extent_item *fi;
4311 struct extent_buffer *leaf;
4312 struct btrfs_map_token token;
4313 struct btrfs_key key;
4314 u64 extent_offset = em->start - em->orig_start;
4317 int extent_inserted = 0;
4319 ret = log_extent_csums(trans, inode, log, em);
4323 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4324 em->start + em->len, NULL, 0, 1,
4325 sizeof(*fi), &extent_inserted);
4329 if (!extent_inserted) {
4330 key.objectid = btrfs_ino(inode);
4331 key.type = BTRFS_EXTENT_DATA_KEY;
4332 key.offset = em->start;
4334 ret = btrfs_insert_empty_item(trans, log, path, &key,
4339 leaf = path->nodes[0];
4340 btrfs_init_map_token(&token, leaf);
4341 fi = btrfs_item_ptr(leaf, path->slots[0],
4342 struct btrfs_file_extent_item);
4344 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4346 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4347 btrfs_set_token_file_extent_type(leaf, fi,
4348 BTRFS_FILE_EXTENT_PREALLOC,
4351 btrfs_set_token_file_extent_type(leaf, fi,
4352 BTRFS_FILE_EXTENT_REG,
4355 block_len = max(em->block_len, em->orig_block_len);
4356 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4357 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4360 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4362 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4363 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4365 extent_offset, &token);
4366 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4369 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4370 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4374 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4375 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4376 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4377 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4379 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4380 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4381 btrfs_mark_buffer_dirty(leaf);
4383 btrfs_release_path(path);
4389 * Log all prealloc extents beyond the inode's i_size to make sure we do not
4390 * lose them after doing a fast fsync and replaying the log. We scan the
4391 * subvolume's root instead of iterating the inode's extent map tree because
4392 * otherwise we can log incorrect extent items based on extent map conversion.
4393 * That can happen due to the fact that extent maps are merged when they
4394 * are not in the extent map tree's list of modified extents.
4396 static int btrfs_log_prealloc_extents(struct btrfs_trans_handle *trans,
4397 struct btrfs_inode *inode,
4398 struct btrfs_path *path)
4400 struct btrfs_root *root = inode->root;
4401 struct btrfs_key key;
4402 const u64 i_size = i_size_read(&inode->vfs_inode);
4403 const u64 ino = btrfs_ino(inode);
4404 struct btrfs_path *dst_path = NULL;
4405 u64 last_extent = (u64)-1;
4410 if (!(inode->flags & BTRFS_INODE_PREALLOC))
4414 key.type = BTRFS_EXTENT_DATA_KEY;
4415 key.offset = i_size;
4416 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4421 struct extent_buffer *leaf = path->nodes[0];
4422 int slot = path->slots[0];
4424 if (slot >= btrfs_header_nritems(leaf)) {
4426 ret = copy_items(trans, inode, dst_path, path,
4427 &last_extent, start_slot,
4433 ret = btrfs_next_leaf(root, path);
4443 btrfs_item_key_to_cpu(leaf, &key, slot);
4444 if (key.objectid > ino)
4446 if (WARN_ON_ONCE(key.objectid < ino) ||
4447 key.type < BTRFS_EXTENT_DATA_KEY ||
4448 key.offset < i_size) {
4452 if (last_extent == (u64)-1) {
4453 last_extent = key.offset;
4455 * Avoid logging extent items logged in past fsync calls
4456 * and leading to duplicate keys in the log tree.
4459 ret = btrfs_truncate_inode_items(trans,
4463 BTRFS_EXTENT_DATA_KEY);
4464 } while (ret == -EAGAIN);
4473 dst_path = btrfs_alloc_path();
4481 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4482 start_slot, ins_nr, 1, 0);
4487 btrfs_release_path(path);
4488 btrfs_free_path(dst_path);
4492 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4493 struct btrfs_root *root,
4494 struct btrfs_inode *inode,
4495 struct btrfs_path *path,
4496 struct btrfs_log_ctx *ctx,
4500 struct extent_map *em, *n;
4501 struct list_head extents;
4502 struct extent_map_tree *tree = &inode->extent_tree;
4507 INIT_LIST_HEAD(&extents);
4509 write_lock(&tree->lock);
4510 test_gen = root->fs_info->last_trans_committed;
4512 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4514 * Skip extents outside our logging range. It's important to do
4515 * it for correctness because if we don't ignore them, we may
4516 * log them before their ordered extent completes, and therefore
4517 * we could log them without logging their respective checksums
4518 * (the checksum items are added to the csum tree at the very
4519 * end of btrfs_finish_ordered_io()). Also leave such extents
4520 * outside of our range in the list, since we may have another
4521 * ranged fsync in the near future that needs them. If an extent
4522 * outside our range corresponds to a hole, log it to avoid
4523 * leaving gaps between extents (fsck will complain when we are
4524 * not using the NO_HOLES feature).
4526 if ((em->start > end || em->start + em->len <= start) &&
4527 em->block_start != EXTENT_MAP_HOLE)
4530 list_del_init(&em->list);
4532 * Just an arbitrary number, this can be really CPU intensive
4533 * once we start getting a lot of extents, and really once we
4534 * have a bunch of extents we just want to commit since it will
4537 if (++num > 32768) {
4538 list_del_init(&tree->modified_extents);
4543 if (em->generation <= test_gen)
4546 /* We log prealloc extents beyond eof later. */
4547 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) &&
4548 em->start >= i_size_read(&inode->vfs_inode))
4551 /* Need a ref to keep it from getting evicted from cache */
4552 refcount_inc(&em->refs);
4553 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4554 list_add_tail(&em->list, &extents);
4558 list_sort(NULL, &extents, extent_cmp);
4560 while (!list_empty(&extents)) {
4561 em = list_entry(extents.next, struct extent_map, list);
4563 list_del_init(&em->list);
4566 * If we had an error we just need to delete everybody from our
4570 clear_em_logging(tree, em);
4571 free_extent_map(em);
4575 write_unlock(&tree->lock);
4577 ret = log_one_extent(trans, inode, root, em, path, ctx);
4578 write_lock(&tree->lock);
4579 clear_em_logging(tree, em);
4580 free_extent_map(em);
4582 WARN_ON(!list_empty(&extents));
4583 write_unlock(&tree->lock);
4585 btrfs_release_path(path);
4587 ret = btrfs_log_prealloc_extents(trans, inode, path);
4592 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4593 struct btrfs_path *path, u64 *size_ret)
4595 struct btrfs_key key;
4598 key.objectid = btrfs_ino(inode);
4599 key.type = BTRFS_INODE_ITEM_KEY;
4602 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4605 } else if (ret > 0) {
4608 struct btrfs_inode_item *item;
4610 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4611 struct btrfs_inode_item);
4612 *size_ret = btrfs_inode_size(path->nodes[0], item);
4614 * If the in-memory inode's i_size is smaller then the inode
4615 * size stored in the btree, return the inode's i_size, so
4616 * that we get a correct inode size after replaying the log
4617 * when before a power failure we had a shrinking truncate
4618 * followed by addition of a new name (rename / new hard link).
4619 * Otherwise return the inode size from the btree, to avoid
4620 * data loss when replaying a log due to previously doing a
4621 * write that expands the inode's size and logging a new name
4622 * immediately after.
4624 if (*size_ret > inode->vfs_inode.i_size)
4625 *size_ret = inode->vfs_inode.i_size;
4628 btrfs_release_path(path);
4633 * At the moment we always log all xattrs. This is to figure out at log replay
4634 * time which xattrs must have their deletion replayed. If a xattr is missing
4635 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4636 * because if a xattr is deleted, the inode is fsynced and a power failure
4637 * happens, causing the log to be replayed the next time the fs is mounted,
4638 * we want the xattr to not exist anymore (same behaviour as other filesystems
4639 * with a journal, ext3/4, xfs, f2fs, etc).
4641 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4642 struct btrfs_root *root,
4643 struct btrfs_inode *inode,
4644 struct btrfs_path *path,
4645 struct btrfs_path *dst_path)
4648 struct btrfs_key key;
4649 const u64 ino = btrfs_ino(inode);
4654 key.type = BTRFS_XATTR_ITEM_KEY;
4657 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4662 int slot = path->slots[0];
4663 struct extent_buffer *leaf = path->nodes[0];
4664 int nritems = btrfs_header_nritems(leaf);
4666 if (slot >= nritems) {
4668 u64 last_extent = 0;
4670 ret = copy_items(trans, inode, dst_path, path,
4671 &last_extent, start_slot,
4673 /* can't be 1, extent items aren't processed */
4679 ret = btrfs_next_leaf(root, path);
4687 btrfs_item_key_to_cpu(leaf, &key, slot);
4688 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4698 u64 last_extent = 0;
4700 ret = copy_items(trans, inode, dst_path, path,
4701 &last_extent, start_slot,
4703 /* can't be 1, extent items aren't processed */
4713 * If the no holes feature is enabled we need to make sure any hole between the
4714 * last extent and the i_size of our inode is explicitly marked in the log. This
4715 * is to make sure that doing something like:
4717 * 1) create file with 128Kb of data
4718 * 2) truncate file to 64Kb
4719 * 3) truncate file to 256Kb
4721 * 5) <crash/power failure>
4722 * 6) mount fs and trigger log replay
4724 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4725 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4726 * file correspond to a hole. The presence of explicit holes in a log tree is
4727 * what guarantees that log replay will remove/adjust file extent items in the
4730 * Here we do not need to care about holes between extents, that is already done
4731 * by copy_items(). We also only need to do this in the full sync path, where we
4732 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4733 * lookup the list of modified extent maps and if any represents a hole, we
4734 * insert a corresponding extent representing a hole in the log tree.
4736 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4737 struct btrfs_root *root,
4738 struct btrfs_inode *inode,
4739 struct btrfs_path *path)
4741 struct btrfs_fs_info *fs_info = root->fs_info;
4743 struct btrfs_key key;
4746 struct extent_buffer *leaf;
4747 struct btrfs_root *log = root->log_root;
4748 const u64 ino = btrfs_ino(inode);
4749 const u64 i_size = i_size_read(&inode->vfs_inode);
4751 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4755 key.type = BTRFS_EXTENT_DATA_KEY;
4756 key.offset = (u64)-1;
4758 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4763 ASSERT(path->slots[0] > 0);
4765 leaf = path->nodes[0];
4766 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4768 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4769 /* inode does not have any extents */
4773 struct btrfs_file_extent_item *extent;
4777 * If there's an extent beyond i_size, an explicit hole was
4778 * already inserted by copy_items().
4780 if (key.offset >= i_size)
4783 extent = btrfs_item_ptr(leaf, path->slots[0],
4784 struct btrfs_file_extent_item);
4786 if (btrfs_file_extent_type(leaf, extent) ==
4787 BTRFS_FILE_EXTENT_INLINE)
4790 len = btrfs_file_extent_num_bytes(leaf, extent);
4791 /* Last extent goes beyond i_size, no need to log a hole. */
4792 if (key.offset + len > i_size)
4794 hole_start = key.offset + len;
4795 hole_size = i_size - hole_start;
4797 btrfs_release_path(path);
4799 /* Last extent ends at i_size. */
4803 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4804 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4805 hole_size, 0, hole_size, 0, 0, 0);
4810 * When we are logging a new inode X, check if it doesn't have a reference that
4811 * matches the reference from some other inode Y created in a past transaction
4812 * and that was renamed in the current transaction. If we don't do this, then at
4813 * log replay time we can lose inode Y (and all its files if it's a directory):
4816 * echo "hello world" > /mnt/x/foobar
4819 * mkdir /mnt/x # or touch /mnt/x
4820 * xfs_io -c fsync /mnt/x
4822 * mount fs, trigger log replay
4824 * After the log replay procedure, we would lose the first directory and all its
4825 * files (file foobar).
4826 * For the case where inode Y is not a directory we simply end up losing it:
4828 * echo "123" > /mnt/foo
4830 * mv /mnt/foo /mnt/bar
4831 * echo "abc" > /mnt/foo
4832 * xfs_io -c fsync /mnt/foo
4835 * We also need this for cases where a snapshot entry is replaced by some other
4836 * entry (file or directory) otherwise we end up with an unreplayable log due to
4837 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4838 * if it were a regular entry:
4841 * btrfs subvolume snapshot /mnt /mnt/x/snap
4842 * btrfs subvolume delete /mnt/x/snap
4845 * fsync /mnt/x or fsync some new file inside it
4848 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4849 * the same transaction.
4851 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4853 const struct btrfs_key *key,
4854 struct btrfs_inode *inode,
4855 u64 *other_ino, u64 *other_parent)
4858 struct btrfs_path *search_path;
4861 u32 item_size = btrfs_item_size_nr(eb, slot);
4863 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4865 search_path = btrfs_alloc_path();
4868 search_path->search_commit_root = 1;
4869 search_path->skip_locking = 1;
4871 while (cur_offset < item_size) {
4875 unsigned long name_ptr;
4876 struct btrfs_dir_item *di;
4878 if (key->type == BTRFS_INODE_REF_KEY) {
4879 struct btrfs_inode_ref *iref;
4881 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4882 parent = key->offset;
4883 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4884 name_ptr = (unsigned long)(iref + 1);
4885 this_len = sizeof(*iref) + this_name_len;
4887 struct btrfs_inode_extref *extref;
4889 extref = (struct btrfs_inode_extref *)(ptr +
4891 parent = btrfs_inode_extref_parent(eb, extref);
4892 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4893 name_ptr = (unsigned long)&extref->name;
4894 this_len = sizeof(*extref) + this_name_len;
4897 if (this_name_len > name_len) {
4900 new_name = krealloc(name, this_name_len, GFP_NOFS);
4905 name_len = this_name_len;
4909 read_extent_buffer(eb, name, name_ptr, this_name_len);
4910 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4911 parent, name, this_name_len, 0);
4912 if (di && !IS_ERR(di)) {
4913 struct btrfs_key di_key;
4915 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4917 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4918 if (di_key.objectid != key->objectid) {
4920 *other_ino = di_key.objectid;
4921 *other_parent = parent;
4929 } else if (IS_ERR(di)) {
4933 btrfs_release_path(search_path);
4935 cur_offset += this_len;
4939 btrfs_free_path(search_path);
4944 struct btrfs_ino_list {
4947 struct list_head list;
4950 static int log_conflicting_inodes(struct btrfs_trans_handle *trans,
4951 struct btrfs_root *root,
4952 struct btrfs_path *path,
4953 struct btrfs_log_ctx *ctx,
4954 u64 ino, u64 parent)
4956 struct btrfs_ino_list *ino_elem;
4957 LIST_HEAD(inode_list);
4960 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
4963 ino_elem->ino = ino;
4964 ino_elem->parent = parent;
4965 list_add_tail(&ino_elem->list, &inode_list);
4967 while (!list_empty(&inode_list)) {
4968 struct btrfs_fs_info *fs_info = root->fs_info;
4969 struct btrfs_key key;
4970 struct inode *inode;
4972 ino_elem = list_first_entry(&inode_list, struct btrfs_ino_list,
4974 ino = ino_elem->ino;
4975 parent = ino_elem->parent;
4976 list_del(&ino_elem->list);
4981 btrfs_release_path(path);
4984 key.type = BTRFS_INODE_ITEM_KEY;
4986 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4988 * If the other inode that had a conflicting dir entry was
4989 * deleted in the current transaction, we need to log its parent
4992 if (IS_ERR(inode)) {
4993 ret = PTR_ERR(inode);
4994 if (ret == -ENOENT) {
4995 key.objectid = parent;
4996 inode = btrfs_iget(fs_info->sb, &key, root,
4998 if (IS_ERR(inode)) {
4999 ret = PTR_ERR(inode);
5001 ret = btrfs_log_inode(trans, root,
5003 LOG_OTHER_INODE_ALL,
5005 btrfs_add_delayed_iput(inode);
5011 * We are safe logging the other inode without acquiring its
5012 * lock as long as we log with the LOG_INODE_EXISTS mode. We
5013 * are safe against concurrent renames of the other inode as
5014 * well because during a rename we pin the log and update the
5015 * log with the new name before we unpin it.
5017 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5018 LOG_OTHER_INODE, 0, LLONG_MAX, ctx);
5020 btrfs_add_delayed_iput(inode);
5025 key.type = BTRFS_INODE_REF_KEY;
5027 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5029 btrfs_add_delayed_iput(inode);
5034 struct extent_buffer *leaf = path->nodes[0];
5035 int slot = path->slots[0];
5037 u64 other_parent = 0;
5039 if (slot >= btrfs_header_nritems(leaf)) {
5040 ret = btrfs_next_leaf(root, path);
5043 } else if (ret > 0) {
5050 btrfs_item_key_to_cpu(leaf, &key, slot);
5051 if (key.objectid != ino ||
5052 (key.type != BTRFS_INODE_REF_KEY &&
5053 key.type != BTRFS_INODE_EXTREF_KEY)) {
5058 ret = btrfs_check_ref_name_override(leaf, slot, &key,
5059 BTRFS_I(inode), &other_ino,
5064 ino_elem = kmalloc(sizeof(*ino_elem), GFP_NOFS);
5069 ino_elem->ino = other_ino;
5070 ino_elem->parent = other_parent;
5071 list_add_tail(&ino_elem->list, &inode_list);
5076 btrfs_add_delayed_iput(inode);
5082 /* log a single inode in the tree log.
5083 * At least one parent directory for this inode must exist in the tree
5084 * or be logged already.
5086 * Any items from this inode changed by the current transaction are copied
5087 * to the log tree. An extra reference is taken on any extents in this
5088 * file, allowing us to avoid a whole pile of corner cases around logging
5089 * blocks that have been removed from the tree.
5091 * See LOG_INODE_ALL and related defines for a description of what inode_only
5094 * This handles both files and directories.
5096 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
5097 struct btrfs_root *root, struct btrfs_inode *inode,
5101 struct btrfs_log_ctx *ctx)
5103 struct btrfs_fs_info *fs_info = root->fs_info;
5104 struct btrfs_path *path;
5105 struct btrfs_path *dst_path;
5106 struct btrfs_key min_key;
5107 struct btrfs_key max_key;
5108 struct btrfs_root *log = root->log_root;
5109 u64 last_extent = 0;
5113 int ins_start_slot = 0;
5115 bool fast_search = false;
5116 u64 ino = btrfs_ino(inode);
5117 struct extent_map_tree *em_tree = &inode->extent_tree;
5118 u64 logged_isize = 0;
5119 bool need_log_inode_item = true;
5120 bool xattrs_logged = false;
5121 bool recursive_logging = false;
5123 path = btrfs_alloc_path();
5126 dst_path = btrfs_alloc_path();
5128 btrfs_free_path(path);
5132 min_key.objectid = ino;
5133 min_key.type = BTRFS_INODE_ITEM_KEY;
5136 max_key.objectid = ino;
5139 /* today the code can only do partial logging of directories */
5140 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5141 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5142 &inode->runtime_flags) &&
5143 inode_only >= LOG_INODE_EXISTS))
5144 max_key.type = BTRFS_XATTR_ITEM_KEY;
5146 max_key.type = (u8)-1;
5147 max_key.offset = (u64)-1;
5150 * Only run delayed items if we are a dir or a new file.
5151 * Otherwise commit the delayed inode only, which is needed in
5152 * order for the log replay code to mark inodes for link count
5153 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
5155 if (S_ISDIR(inode->vfs_inode.i_mode) ||
5156 inode->generation > fs_info->last_trans_committed)
5157 ret = btrfs_commit_inode_delayed_items(trans, inode);
5159 ret = btrfs_commit_inode_delayed_inode(inode);
5162 btrfs_free_path(path);
5163 btrfs_free_path(dst_path);
5167 if (inode_only == LOG_OTHER_INODE || inode_only == LOG_OTHER_INODE_ALL) {
5168 recursive_logging = true;
5169 if (inode_only == LOG_OTHER_INODE)
5170 inode_only = LOG_INODE_EXISTS;
5172 inode_only = LOG_INODE_ALL;
5173 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
5175 mutex_lock(&inode->log_mutex);
5179 * a brute force approach to making sure we get the most uptodate
5180 * copies of everything.
5182 if (S_ISDIR(inode->vfs_inode.i_mode)) {
5183 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
5185 if (inode_only == LOG_INODE_EXISTS)
5186 max_key_type = BTRFS_XATTR_ITEM_KEY;
5187 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
5189 if (inode_only == LOG_INODE_EXISTS) {
5191 * Make sure the new inode item we write to the log has
5192 * the same isize as the current one (if it exists).
5193 * This is necessary to prevent data loss after log
5194 * replay, and also to prevent doing a wrong expanding
5195 * truncate - for e.g. create file, write 4K into offset
5196 * 0, fsync, write 4K into offset 4096, add hard link,
5197 * fsync some other file (to sync log), power fail - if
5198 * we use the inode's current i_size, after log replay
5199 * we get a 8Kb file, with the last 4Kb extent as a hole
5200 * (zeroes), as if an expanding truncate happened,
5201 * instead of getting a file of 4Kb only.
5203 err = logged_inode_size(log, inode, path, &logged_isize);
5207 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5208 &inode->runtime_flags)) {
5209 if (inode_only == LOG_INODE_EXISTS) {
5210 max_key.type = BTRFS_XATTR_ITEM_KEY;
5211 ret = drop_objectid_items(trans, log, path, ino,
5214 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5215 &inode->runtime_flags);
5216 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5217 &inode->runtime_flags);
5219 ret = btrfs_truncate_inode_items(trans,
5220 log, &inode->vfs_inode, 0, 0);
5225 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
5226 &inode->runtime_flags) ||
5227 inode_only == LOG_INODE_EXISTS) {
5228 if (inode_only == LOG_INODE_ALL)
5230 max_key.type = BTRFS_XATTR_ITEM_KEY;
5231 ret = drop_objectid_items(trans, log, path, ino,
5234 if (inode_only == LOG_INODE_ALL)
5247 ret = btrfs_search_forward(root, &min_key,
5248 path, trans->transid);
5256 /* note, ins_nr might be > 0 here, cleanup outside the loop */
5257 if (min_key.objectid != ino)
5259 if (min_key.type > max_key.type)
5262 if (min_key.type == BTRFS_INODE_ITEM_KEY)
5263 need_log_inode_item = false;
5265 if ((min_key.type == BTRFS_INODE_REF_KEY ||
5266 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
5267 inode->generation == trans->transid &&
5268 !recursive_logging) {
5270 u64 other_parent = 0;
5272 ret = btrfs_check_ref_name_override(path->nodes[0],
5273 path->slots[0], &min_key, inode,
5274 &other_ino, &other_parent);
5278 } else if (ret > 0 && ctx &&
5279 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
5284 ins_start_slot = path->slots[0];
5286 ret = copy_items(trans, inode, dst_path, path,
5287 &last_extent, ins_start_slot,
5296 err = log_conflicting_inodes(trans, root, path,
5297 ctx, other_ino, other_parent);
5300 btrfs_release_path(path);
5305 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
5306 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
5309 ret = copy_items(trans, inode, dst_path, path,
5310 &last_extent, ins_start_slot,
5311 ins_nr, inode_only, logged_isize);
5318 btrfs_release_path(path);
5324 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
5327 } else if (!ins_nr) {
5328 ins_start_slot = path->slots[0];
5333 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5334 ins_start_slot, ins_nr, inode_only,
5342 btrfs_release_path(path);
5346 ins_start_slot = path->slots[0];
5349 nritems = btrfs_header_nritems(path->nodes[0]);
5351 if (path->slots[0] < nritems) {
5352 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
5357 ret = copy_items(trans, inode, dst_path, path,
5358 &last_extent, ins_start_slot,
5359 ins_nr, inode_only, logged_isize);
5367 btrfs_release_path(path);
5369 if (min_key.offset < (u64)-1) {
5371 } else if (min_key.type < max_key.type) {
5379 ret = copy_items(trans, inode, dst_path, path, &last_extent,
5380 ins_start_slot, ins_nr, inode_only,
5390 btrfs_release_path(path);
5391 btrfs_release_path(dst_path);
5392 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
5395 xattrs_logged = true;
5396 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
5397 btrfs_release_path(path);
5398 btrfs_release_path(dst_path);
5399 err = btrfs_log_trailing_hole(trans, root, inode, path);
5404 btrfs_release_path(path);
5405 btrfs_release_path(dst_path);
5406 if (need_log_inode_item) {
5407 err = log_inode_item(trans, log, dst_path, inode);
5408 if (!err && !xattrs_logged) {
5409 err = btrfs_log_all_xattrs(trans, root, inode, path,
5411 btrfs_release_path(path);
5417 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
5423 } else if (inode_only == LOG_INODE_ALL) {
5424 struct extent_map *em, *n;
5426 write_lock(&em_tree->lock);
5428 * We can't just remove every em if we're called for a ranged
5429 * fsync - that is, one that doesn't cover the whole possible
5430 * file range (0 to LLONG_MAX). This is because we can have
5431 * em's that fall outside the range we're logging and therefore
5432 * their ordered operations haven't completed yet
5433 * (btrfs_finish_ordered_io() not invoked yet). This means we
5434 * didn't get their respective file extent item in the fs/subvol
5435 * tree yet, and need to let the next fast fsync (one which
5436 * consults the list of modified extent maps) find the em so
5437 * that it logs a matching file extent item and waits for the
5438 * respective ordered operation to complete (if it's still
5441 * Removing every em outside the range we're logging would make
5442 * the next fast fsync not log their matching file extent items,
5443 * therefore making us lose data after a log replay.
5445 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5447 const u64 mod_end = em->mod_start + em->mod_len - 1;
5449 if (em->mod_start >= start && mod_end <= end)
5450 list_del_init(&em->list);
5452 write_unlock(&em_tree->lock);
5455 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5456 ret = log_directory_changes(trans, root, inode, path, dst_path,
5465 * Don't update last_log_commit if we logged that an inode exists after
5466 * it was loaded to memory (full_sync bit set).
5467 * This is to prevent data loss when we do a write to the inode, then
5468 * the inode gets evicted after all delalloc was flushed, then we log
5469 * it exists (due to a rename for example) and then fsync it. This last
5470 * fsync would do nothing (not logging the extents previously written).
5472 spin_lock(&inode->lock);
5473 inode->logged_trans = trans->transid;
5474 if (inode_only != LOG_INODE_EXISTS ||
5475 !test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags))
5476 inode->last_log_commit = inode->last_sub_trans;
5477 spin_unlock(&inode->lock);
5479 mutex_unlock(&inode->log_mutex);
5481 btrfs_free_path(path);
5482 btrfs_free_path(dst_path);
5487 * Check if we must fallback to a transaction commit when logging an inode.
5488 * This must be called after logging the inode and is used only in the context
5489 * when fsyncing an inode requires the need to log some other inode - in which
5490 * case we can't lock the i_mutex of each other inode we need to log as that
5491 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5492 * log inodes up or down in the hierarchy) or rename operations for example. So
5493 * we take the log_mutex of the inode after we have logged it and then check for
5494 * its last_unlink_trans value - this is safe because any task setting
5495 * last_unlink_trans must take the log_mutex and it must do this before it does
5496 * the actual unlink operation, so if we do this check before a concurrent task
5497 * sets last_unlink_trans it means we've logged a consistent version/state of
5498 * all the inode items, otherwise we are not sure and must do a transaction
5499 * commit (the concurrent task might have only updated last_unlink_trans before
5500 * we logged the inode or it might have also done the unlink).
5502 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5503 struct btrfs_inode *inode)
5505 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5508 mutex_lock(&inode->log_mutex);
5509 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5511 * Make sure any commits to the log are forced to be full
5514 btrfs_set_log_full_commit(trans);
5517 mutex_unlock(&inode->log_mutex);
5523 * follow the dentry parent pointers up the chain and see if any
5524 * of the directories in it require a full commit before they can
5525 * be logged. Returns zero if nothing special needs to be done or 1 if
5526 * a full commit is required.
5528 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5529 struct btrfs_inode *inode,
5530 struct dentry *parent,
5531 struct super_block *sb,
5535 struct dentry *old_parent = NULL;
5538 * for regular files, if its inode is already on disk, we don't
5539 * have to worry about the parents at all. This is because
5540 * we can use the last_unlink_trans field to record renames
5541 * and other fun in this file.
5543 if (S_ISREG(inode->vfs_inode.i_mode) &&
5544 inode->generation <= last_committed &&
5545 inode->last_unlink_trans <= last_committed)
5548 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5549 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5551 inode = BTRFS_I(d_inode(parent));
5555 if (btrfs_must_commit_transaction(trans, inode)) {
5560 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5563 if (IS_ROOT(parent)) {
5564 inode = BTRFS_I(d_inode(parent));
5565 if (btrfs_must_commit_transaction(trans, inode))
5570 parent = dget_parent(parent);
5572 old_parent = parent;
5573 inode = BTRFS_I(d_inode(parent));
5581 struct btrfs_dir_list {
5583 struct list_head list;
5587 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5588 * details about the why it is needed.
5589 * This is a recursive operation - if an existing dentry corresponds to a
5590 * directory, that directory's new entries are logged too (same behaviour as
5591 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5592 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5593 * complains about the following circular lock dependency / possible deadlock:
5597 * lock(&type->i_mutex_dir_key#3/2);
5598 * lock(sb_internal#2);
5599 * lock(&type->i_mutex_dir_key#3/2);
5600 * lock(&sb->s_type->i_mutex_key#14);
5602 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5603 * sb_start_intwrite() in btrfs_start_transaction().
5604 * Not locking i_mutex of the inodes is still safe because:
5606 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5607 * that while logging the inode new references (names) are added or removed
5608 * from the inode, leaving the logged inode item with a link count that does
5609 * not match the number of logged inode reference items. This is fine because
5610 * at log replay time we compute the real number of links and correct the
5611 * link count in the inode item (see replay_one_buffer() and
5612 * link_to_fixup_dir());
5614 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5615 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5616 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5617 * has a size that doesn't match the sum of the lengths of all the logged
5618 * names. This does not result in a problem because if a dir_item key is
5619 * logged but its matching dir_index key is not logged, at log replay time we
5620 * don't use it to replay the respective name (see replay_one_name()). On the
5621 * other hand if only the dir_index key ends up being logged, the respective
5622 * name is added to the fs/subvol tree with both the dir_item and dir_index
5623 * keys created (see replay_one_name()).
5624 * The directory's inode item with a wrong i_size is not a problem as well,
5625 * since we don't use it at log replay time to set the i_size in the inode
5626 * item of the fs/subvol tree (see overwrite_item()).
5628 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5629 struct btrfs_root *root,
5630 struct btrfs_inode *start_inode,
5631 struct btrfs_log_ctx *ctx)
5633 struct btrfs_fs_info *fs_info = root->fs_info;
5634 struct btrfs_root *log = root->log_root;
5635 struct btrfs_path *path;
5636 LIST_HEAD(dir_list);
5637 struct btrfs_dir_list *dir_elem;
5640 path = btrfs_alloc_path();
5644 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5646 btrfs_free_path(path);
5649 dir_elem->ino = btrfs_ino(start_inode);
5650 list_add_tail(&dir_elem->list, &dir_list);
5652 while (!list_empty(&dir_list)) {
5653 struct extent_buffer *leaf;
5654 struct btrfs_key min_key;
5658 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5661 goto next_dir_inode;
5663 min_key.objectid = dir_elem->ino;
5664 min_key.type = BTRFS_DIR_ITEM_KEY;
5667 btrfs_release_path(path);
5668 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5670 goto next_dir_inode;
5671 } else if (ret > 0) {
5673 goto next_dir_inode;
5677 leaf = path->nodes[0];
5678 nritems = btrfs_header_nritems(leaf);
5679 for (i = path->slots[0]; i < nritems; i++) {
5680 struct btrfs_dir_item *di;
5681 struct btrfs_key di_key;
5682 struct inode *di_inode;
5683 struct btrfs_dir_list *new_dir_elem;
5684 int log_mode = LOG_INODE_EXISTS;
5687 btrfs_item_key_to_cpu(leaf, &min_key, i);
5688 if (min_key.objectid != dir_elem->ino ||
5689 min_key.type != BTRFS_DIR_ITEM_KEY)
5690 goto next_dir_inode;
5692 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5693 type = btrfs_dir_type(leaf, di);
5694 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5695 type != BTRFS_FT_DIR)
5697 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5698 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5701 btrfs_release_path(path);
5702 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5703 if (IS_ERR(di_inode)) {
5704 ret = PTR_ERR(di_inode);
5705 goto next_dir_inode;
5708 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5709 btrfs_add_delayed_iput(di_inode);
5713 ctx->log_new_dentries = false;
5714 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5715 log_mode = LOG_INODE_ALL;
5716 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5717 log_mode, 0, LLONG_MAX, ctx);
5719 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5721 btrfs_add_delayed_iput(di_inode);
5723 goto next_dir_inode;
5724 if (ctx->log_new_dentries) {
5725 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5727 if (!new_dir_elem) {
5729 goto next_dir_inode;
5731 new_dir_elem->ino = di_key.objectid;
5732 list_add_tail(&new_dir_elem->list, &dir_list);
5737 ret = btrfs_next_leaf(log, path);
5739 goto next_dir_inode;
5740 } else if (ret > 0) {
5742 goto next_dir_inode;
5746 if (min_key.offset < (u64)-1) {
5751 list_del(&dir_elem->list);
5755 btrfs_free_path(path);
5759 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5760 struct btrfs_inode *inode,
5761 struct btrfs_log_ctx *ctx)
5763 struct btrfs_fs_info *fs_info = trans->fs_info;
5765 struct btrfs_path *path;
5766 struct btrfs_key key;
5767 struct btrfs_root *root = inode->root;
5768 const u64 ino = btrfs_ino(inode);
5770 path = btrfs_alloc_path();
5773 path->skip_locking = 1;
5774 path->search_commit_root = 1;
5777 key.type = BTRFS_INODE_REF_KEY;
5779 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5784 struct extent_buffer *leaf = path->nodes[0];
5785 int slot = path->slots[0];
5790 if (slot >= btrfs_header_nritems(leaf)) {
5791 ret = btrfs_next_leaf(root, path);
5799 btrfs_item_key_to_cpu(leaf, &key, slot);
5800 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5801 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5804 item_size = btrfs_item_size_nr(leaf, slot);
5805 ptr = btrfs_item_ptr_offset(leaf, slot);
5806 while (cur_offset < item_size) {
5807 struct btrfs_key inode_key;
5808 struct inode *dir_inode;
5810 inode_key.type = BTRFS_INODE_ITEM_KEY;
5811 inode_key.offset = 0;
5813 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5814 struct btrfs_inode_extref *extref;
5816 extref = (struct btrfs_inode_extref *)
5818 inode_key.objectid = btrfs_inode_extref_parent(
5820 cur_offset += sizeof(*extref);
5821 cur_offset += btrfs_inode_extref_name_len(leaf,
5824 inode_key.objectid = key.offset;
5825 cur_offset = item_size;
5828 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5831 * If the parent inode was deleted, return an error to
5832 * fallback to a transaction commit. This is to prevent
5833 * getting an inode that was moved from one parent A to
5834 * a parent B, got its former parent A deleted and then
5835 * it got fsync'ed, from existing at both parents after
5836 * a log replay (and the old parent still existing).
5843 * mv /mnt/B/bar /mnt/A/bar
5844 * mv -T /mnt/A /mnt/B
5848 * If we ignore the old parent B which got deleted,
5849 * after a log replay we would have file bar linked
5850 * at both parents and the old parent B would still
5853 if (IS_ERR(dir_inode)) {
5854 ret = PTR_ERR(dir_inode);
5859 ctx->log_new_dentries = false;
5860 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5861 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5863 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5865 if (!ret && ctx && ctx->log_new_dentries)
5866 ret = log_new_dir_dentries(trans, root,
5867 BTRFS_I(dir_inode), ctx);
5868 btrfs_add_delayed_iput(dir_inode);
5876 btrfs_free_path(path);
5880 static int log_new_ancestors(struct btrfs_trans_handle *trans,
5881 struct btrfs_root *root,
5882 struct btrfs_path *path,
5883 struct btrfs_log_ctx *ctx)
5885 struct btrfs_key found_key;
5887 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
5890 struct btrfs_fs_info *fs_info = root->fs_info;
5891 const u64 last_committed = fs_info->last_trans_committed;
5892 struct extent_buffer *leaf = path->nodes[0];
5893 int slot = path->slots[0];
5894 struct btrfs_key search_key;
5895 struct inode *inode;
5898 btrfs_release_path(path);
5900 search_key.objectid = found_key.offset;
5901 search_key.type = BTRFS_INODE_ITEM_KEY;
5902 search_key.offset = 0;
5903 inode = btrfs_iget(fs_info->sb, &search_key, root, NULL);
5905 return PTR_ERR(inode);
5907 if (BTRFS_I(inode)->generation > last_committed)
5908 ret = btrfs_log_inode(trans, root, BTRFS_I(inode),
5911 btrfs_add_delayed_iput(inode);
5915 if (search_key.objectid == BTRFS_FIRST_FREE_OBJECTID)
5918 search_key.type = BTRFS_INODE_REF_KEY;
5919 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
5923 leaf = path->nodes[0];
5924 slot = path->slots[0];
5925 if (slot >= btrfs_header_nritems(leaf)) {
5926 ret = btrfs_next_leaf(root, path);
5931 leaf = path->nodes[0];
5932 slot = path->slots[0];
5935 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5936 if (found_key.objectid != search_key.objectid ||
5937 found_key.type != BTRFS_INODE_REF_KEY)
5943 static int log_new_ancestors_fast(struct btrfs_trans_handle *trans,
5944 struct btrfs_inode *inode,
5945 struct dentry *parent,
5946 struct btrfs_log_ctx *ctx)
5948 struct btrfs_root *root = inode->root;
5949 struct btrfs_fs_info *fs_info = root->fs_info;
5950 struct dentry *old_parent = NULL;
5951 struct super_block *sb = inode->vfs_inode.i_sb;
5955 if (!parent || d_really_is_negative(parent) ||
5959 inode = BTRFS_I(d_inode(parent));
5960 if (root != inode->root)
5963 if (inode->generation > fs_info->last_trans_committed) {
5964 ret = btrfs_log_inode(trans, root, inode,
5965 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5969 if (IS_ROOT(parent))
5972 parent = dget_parent(parent);
5974 old_parent = parent;
5981 static int log_all_new_ancestors(struct btrfs_trans_handle *trans,
5982 struct btrfs_inode *inode,
5983 struct dentry *parent,
5984 struct btrfs_log_ctx *ctx)
5986 struct btrfs_root *root = inode->root;
5987 const u64 ino = btrfs_ino(inode);
5988 struct btrfs_path *path;
5989 struct btrfs_key search_key;
5993 * For a single hard link case, go through a fast path that does not
5994 * need to iterate the fs/subvolume tree.
5996 if (inode->vfs_inode.i_nlink < 2)
5997 return log_new_ancestors_fast(trans, inode, parent, ctx);
5999 path = btrfs_alloc_path();
6003 search_key.objectid = ino;
6004 search_key.type = BTRFS_INODE_REF_KEY;
6005 search_key.offset = 0;
6007 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
6014 struct extent_buffer *leaf = path->nodes[0];
6015 int slot = path->slots[0];
6016 struct btrfs_key found_key;
6018 if (slot >= btrfs_header_nritems(leaf)) {
6019 ret = btrfs_next_leaf(root, path);
6027 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6028 if (found_key.objectid != ino ||
6029 found_key.type > BTRFS_INODE_EXTREF_KEY)
6033 * Don't deal with extended references because they are rare
6034 * cases and too complex to deal with (we would need to keep
6035 * track of which subitem we are processing for each item in
6036 * this loop, etc). So just return some error to fallback to
6037 * a transaction commit.
6039 if (found_key.type == BTRFS_INODE_EXTREF_KEY) {
6045 * Logging ancestors needs to do more searches on the fs/subvol
6046 * tree, so it releases the path as needed to avoid deadlocks.
6047 * Keep track of the last inode ref key and resume from that key
6048 * after logging all new ancestors for the current hard link.
6050 memcpy(&search_key, &found_key, sizeof(search_key));
6052 ret = log_new_ancestors(trans, root, path, ctx);
6055 btrfs_release_path(path);
6060 btrfs_free_path(path);
6065 * helper function around btrfs_log_inode to make sure newly created
6066 * parent directories also end up in the log. A minimal inode and backref
6067 * only logging is done of any parent directories that are older than
6068 * the last committed transaction
6070 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
6071 struct btrfs_inode *inode,
6072 struct dentry *parent,
6076 struct btrfs_log_ctx *ctx)
6078 struct btrfs_root *root = inode->root;
6079 struct btrfs_fs_info *fs_info = root->fs_info;
6080 struct super_block *sb;
6082 u64 last_committed = fs_info->last_trans_committed;
6083 bool log_dentries = false;
6085 sb = inode->vfs_inode.i_sb;
6087 if (btrfs_test_opt(fs_info, NOTREELOG)) {
6093 * The prev transaction commit doesn't complete, we need do
6094 * full commit by ourselves.
6096 if (fs_info->last_trans_log_full_commit >
6097 fs_info->last_trans_committed) {
6102 if (btrfs_root_refs(&root->root_item) == 0) {
6107 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
6113 * Skip already logged inodes or inodes corresponding to tmpfiles
6114 * (since logging them is pointless, a link count of 0 means they
6115 * will never be accessible).
6117 if (btrfs_inode_in_log(inode, trans->transid) ||
6118 inode->vfs_inode.i_nlink == 0) {
6119 ret = BTRFS_NO_LOG_SYNC;
6123 ret = start_log_trans(trans, root, ctx);
6127 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
6132 * for regular files, if its inode is already on disk, we don't
6133 * have to worry about the parents at all. This is because
6134 * we can use the last_unlink_trans field to record renames
6135 * and other fun in this file.
6137 if (S_ISREG(inode->vfs_inode.i_mode) &&
6138 inode->generation <= last_committed &&
6139 inode->last_unlink_trans <= last_committed) {
6144 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
6145 log_dentries = true;
6148 * On unlink we must make sure all our current and old parent directory
6149 * inodes are fully logged. This is to prevent leaving dangling
6150 * directory index entries in directories that were our parents but are
6151 * not anymore. Not doing this results in old parent directory being
6152 * impossible to delete after log replay (rmdir will always fail with
6153 * error -ENOTEMPTY).
6159 * ln testdir/foo testdir/bar
6161 * unlink testdir/bar
6162 * xfs_io -c fsync testdir/foo
6164 * mount fs, triggers log replay
6166 * If we don't log the parent directory (testdir), after log replay the
6167 * directory still has an entry pointing to the file inode using the bar
6168 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
6169 * the file inode has a link count of 1.
6175 * ln foo testdir/foo2
6176 * ln foo testdir/foo3
6178 * unlink testdir/foo3
6179 * xfs_io -c fsync foo
6181 * mount fs, triggers log replay
6183 * Similar as the first example, after log replay the parent directory
6184 * testdir still has an entry pointing to the inode file with name foo3
6185 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
6186 * and has a link count of 2.
6188 if (inode->last_unlink_trans > last_committed) {
6189 ret = btrfs_log_all_parents(trans, inode, ctx);
6194 ret = log_all_new_ancestors(trans, inode, parent, ctx);
6199 ret = log_new_dir_dentries(trans, root, inode, ctx);
6204 btrfs_set_log_full_commit(trans);
6209 btrfs_remove_log_ctx(root, ctx);
6210 btrfs_end_log_trans(root);
6216 * it is not safe to log dentry if the chunk root has added new
6217 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
6218 * If this returns 1, you must commit the transaction to safely get your
6221 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
6222 struct dentry *dentry,
6225 struct btrfs_log_ctx *ctx)
6227 struct dentry *parent = dget_parent(dentry);
6230 ret = btrfs_log_inode_parent(trans, BTRFS_I(d_inode(dentry)), parent,
6231 start, end, LOG_INODE_ALL, ctx);
6238 * should be called during mount to recover any replay any log trees
6241 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
6244 struct btrfs_path *path;
6245 struct btrfs_trans_handle *trans;
6246 struct btrfs_key key;
6247 struct btrfs_key found_key;
6248 struct btrfs_key tmp_key;
6249 struct btrfs_root *log;
6250 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
6251 struct walk_control wc = {
6252 .process_func = process_one_buffer,
6253 .stage = LOG_WALK_PIN_ONLY,
6256 path = btrfs_alloc_path();
6260 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6262 trans = btrfs_start_transaction(fs_info->tree_root, 0);
6263 if (IS_ERR(trans)) {
6264 ret = PTR_ERR(trans);
6271 ret = walk_log_tree(trans, log_root_tree, &wc);
6273 btrfs_handle_fs_error(fs_info, ret,
6274 "Failed to pin buffers while recovering log root tree.");
6279 key.objectid = BTRFS_TREE_LOG_OBJECTID;
6280 key.offset = (u64)-1;
6281 key.type = BTRFS_ROOT_ITEM_KEY;
6284 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
6287 btrfs_handle_fs_error(fs_info, ret,
6288 "Couldn't find tree log root.");
6292 if (path->slots[0] == 0)
6296 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
6298 btrfs_release_path(path);
6299 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
6302 log = btrfs_read_fs_root(log_root_tree, &found_key);
6305 btrfs_handle_fs_error(fs_info, ret,
6306 "Couldn't read tree log root.");
6310 tmp_key.objectid = found_key.offset;
6311 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
6312 tmp_key.offset = (u64)-1;
6314 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
6315 if (IS_ERR(wc.replay_dest)) {
6316 ret = PTR_ERR(wc.replay_dest);
6317 free_extent_buffer(log->node);
6318 free_extent_buffer(log->commit_root);
6320 btrfs_handle_fs_error(fs_info, ret,
6321 "Couldn't read target root for tree log recovery.");
6325 wc.replay_dest->log_root = log;
6326 btrfs_record_root_in_trans(trans, wc.replay_dest);
6327 ret = walk_log_tree(trans, log, &wc);
6329 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6330 ret = fixup_inode_link_counts(trans, wc.replay_dest,
6334 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
6335 struct btrfs_root *root = wc.replay_dest;
6337 btrfs_release_path(path);
6340 * We have just replayed everything, and the highest
6341 * objectid of fs roots probably has changed in case
6342 * some inode_item's got replayed.
6344 * root->objectid_mutex is not acquired as log replay
6345 * could only happen during mount.
6347 ret = btrfs_find_highest_objectid(root,
6348 &root->highest_objectid);
6351 key.offset = found_key.offset - 1;
6352 wc.replay_dest->log_root = NULL;
6353 free_extent_buffer(log->node);
6354 free_extent_buffer(log->commit_root);
6360 if (found_key.offset == 0)
6363 btrfs_release_path(path);
6365 /* step one is to pin it all, step two is to replay just inodes */
6368 wc.process_func = replay_one_buffer;
6369 wc.stage = LOG_WALK_REPLAY_INODES;
6372 /* step three is to replay everything */
6373 if (wc.stage < LOG_WALK_REPLAY_ALL) {
6378 btrfs_free_path(path);
6380 /* step 4: commit the transaction, which also unpins the blocks */
6381 ret = btrfs_commit_transaction(trans);
6385 free_extent_buffer(log_root_tree->node);
6386 log_root_tree->log_root = NULL;
6387 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
6388 kfree(log_root_tree);
6393 btrfs_end_transaction(wc.trans);
6394 btrfs_free_path(path);
6399 * there are some corner cases where we want to force a full
6400 * commit instead of allowing a directory to be logged.
6402 * They revolve around files there were unlinked from the directory, and
6403 * this function updates the parent directory so that a full commit is
6404 * properly done if it is fsync'd later after the unlinks are done.
6406 * Must be called before the unlink operations (updates to the subvolume tree,
6407 * inodes, etc) are done.
6409 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
6410 struct btrfs_inode *dir, struct btrfs_inode *inode,
6414 * when we're logging a file, if it hasn't been renamed
6415 * or unlinked, and its inode is fully committed on disk,
6416 * we don't have to worry about walking up the directory chain
6417 * to log its parents.
6419 * So, we use the last_unlink_trans field to put this transid
6420 * into the file. When the file is logged we check it and
6421 * don't log the parents if the file is fully on disk.
6423 mutex_lock(&inode->log_mutex);
6424 inode->last_unlink_trans = trans->transid;
6425 mutex_unlock(&inode->log_mutex);
6428 * if this directory was already logged any new
6429 * names for this file/dir will get recorded
6431 if (dir->logged_trans == trans->transid)
6435 * if the inode we're about to unlink was logged,
6436 * the log will be properly updated for any new names
6438 if (inode->logged_trans == trans->transid)
6442 * when renaming files across directories, if the directory
6443 * there we're unlinking from gets fsync'd later on, there's
6444 * no way to find the destination directory later and fsync it
6445 * properly. So, we have to be conservative and force commits
6446 * so the new name gets discovered.
6451 /* we can safely do the unlink without any special recording */
6455 mutex_lock(&dir->log_mutex);
6456 dir->last_unlink_trans = trans->transid;
6457 mutex_unlock(&dir->log_mutex);
6461 * Make sure that if someone attempts to fsync the parent directory of a deleted
6462 * snapshot, it ends up triggering a transaction commit. This is to guarantee
6463 * that after replaying the log tree of the parent directory's root we will not
6464 * see the snapshot anymore and at log replay time we will not see any log tree
6465 * corresponding to the deleted snapshot's root, which could lead to replaying
6466 * it after replaying the log tree of the parent directory (which would replay
6467 * the snapshot delete operation).
6469 * Must be called before the actual snapshot destroy operation (updates to the
6470 * parent root and tree of tree roots trees, etc) are done.
6472 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
6473 struct btrfs_inode *dir)
6475 mutex_lock(&dir->log_mutex);
6476 dir->last_unlink_trans = trans->transid;
6477 mutex_unlock(&dir->log_mutex);
6481 * Call this after adding a new name for a file and it will properly
6482 * update the log to reflect the new name.
6484 * @ctx can not be NULL when @sync_log is false, and should be NULL when it's
6485 * true (because it's not used).
6487 * Return value depends on whether @sync_log is true or false.
6488 * When true: returns BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6489 * committed by the caller, and BTRFS_DONT_NEED_TRANS_COMMIT
6491 * When false: returns BTRFS_DONT_NEED_LOG_SYNC if the caller does not need to
6492 * to sync the log, BTRFS_NEED_LOG_SYNC if it needs to sync the log,
6493 * or BTRFS_NEED_TRANS_COMMIT if the transaction needs to be
6494 * committed (without attempting to sync the log).
6496 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
6497 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
6498 struct dentry *parent,
6499 bool sync_log, struct btrfs_log_ctx *ctx)
6501 struct btrfs_fs_info *fs_info = trans->fs_info;
6505 * this will force the logging code to walk the dentry chain
6508 if (!S_ISDIR(inode->vfs_inode.i_mode))
6509 inode->last_unlink_trans = trans->transid;
6512 * if this inode hasn't been logged and directory we're renaming it
6513 * from hasn't been logged, we don't need to log it
6515 if (inode->logged_trans <= fs_info->last_trans_committed &&
6516 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
6517 return sync_log ? BTRFS_DONT_NEED_TRANS_COMMIT :
6518 BTRFS_DONT_NEED_LOG_SYNC;
6521 struct btrfs_log_ctx ctx2;
6523 btrfs_init_log_ctx(&ctx2, &inode->vfs_inode);
6524 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6525 LOG_INODE_EXISTS, &ctx2);
6526 if (ret == BTRFS_NO_LOG_SYNC)
6527 return BTRFS_DONT_NEED_TRANS_COMMIT;
6529 return BTRFS_NEED_TRANS_COMMIT;
6531 ret = btrfs_sync_log(trans, inode->root, &ctx2);
6533 return BTRFS_NEED_TRANS_COMMIT;
6534 return BTRFS_DONT_NEED_TRANS_COMMIT;
6538 ret = btrfs_log_inode_parent(trans, inode, parent, 0, LLONG_MAX,
6539 LOG_INODE_EXISTS, ctx);
6540 if (ret == BTRFS_NO_LOG_SYNC)
6541 return BTRFS_DONT_NEED_LOG_SYNC;
6543 return BTRFS_NEED_TRANS_COMMIT;
6545 return BTRFS_NEED_LOG_SYNC;