2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
29 #include "compression.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
40 #define LOG_OTHER_INODE 2
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_DIR_INDEX 2
97 #define LOG_WALK_REPLAY_ALL 3
99 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root, struct inode *inode,
104 struct btrfs_log_ctx *ctx);
105 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_path *path, u64 objectid);
108 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root,
110 struct btrfs_root *log,
111 struct btrfs_path *path,
112 u64 dirid, int del_all);
115 * tree logging is a special write ahead log used to make sure that
116 * fsyncs and O_SYNCs can happen without doing full tree commits.
118 * Full tree commits are expensive because they require commonly
119 * modified blocks to be recowed, creating many dirty pages in the
120 * extent tree an 4x-6x higher write load than ext3.
122 * Instead of doing a tree commit on every fsync, we use the
123 * key ranges and transaction ids to find items for a given file or directory
124 * that have changed in this transaction. Those items are copied into
125 * a special tree (one per subvolume root), that tree is written to disk
126 * and then the fsync is considered complete.
128 * After a crash, items are copied out of the log-tree back into the
129 * subvolume tree. Any file data extents found are recorded in the extent
130 * allocation tree, and the log-tree freed.
132 * The log tree is read three times, once to pin down all the extents it is
133 * using in ram and once, once to create all the inodes logged in the tree
134 * and once to do all the other items.
138 * start a sub transaction and setup the log tree
139 * this increments the log tree writer count to make the people
140 * syncing the tree wait for us to finish
142 static int start_log_trans(struct btrfs_trans_handle *trans,
143 struct btrfs_root *root,
144 struct btrfs_log_ctx *ctx)
146 struct btrfs_fs_info *fs_info = root->fs_info;
149 mutex_lock(&root->log_mutex);
151 if (root->log_root) {
152 if (btrfs_need_log_full_commit(fs_info, trans)) {
157 if (!root->log_start_pid) {
158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 root->log_start_pid = current->pid;
160 } else if (root->log_start_pid != current->pid) {
161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
164 mutex_lock(&fs_info->tree_log_mutex);
165 if (!fs_info->log_root_tree)
166 ret = btrfs_init_log_root_tree(trans, fs_info);
167 mutex_unlock(&fs_info->tree_log_mutex);
171 ret = btrfs_add_log_tree(trans, root);
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
176 root->log_start_pid = current->pid;
179 atomic_inc(&root->log_batch);
180 atomic_inc(&root->log_writers);
182 int index = root->log_transid % 2;
183 list_add_tail(&ctx->list, &root->log_ctxs[index]);
184 ctx->log_transid = root->log_transid;
188 mutex_unlock(&root->log_mutex);
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
197 static int join_running_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 if (root->log_root) {
208 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
219 int btrfs_pin_log_trans(struct btrfs_root *root)
223 mutex_lock(&root->log_mutex);
224 atomic_inc(&root->log_writers);
225 mutex_unlock(&root->log_mutex);
230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
233 void btrfs_end_log_trans(struct btrfs_root *root)
235 if (atomic_dec_and_test(&root->log_writers)) {
237 * Implicit memory barrier after atomic_dec_and_test
239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
297 struct btrfs_fs_info *fs_info = log->fs_info;
301 * If this fs is mixed then we need to be able to process the leaves to
302 * pin down any logged extents, so we have to read the block.
304 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
305 ret = btrfs_read_buffer(eb, gen);
311 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
314 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
315 if (wc->pin && btrfs_header_level(eb) == 0)
316 ret = btrfs_exclude_logged_extents(fs_info, eb);
318 btrfs_write_tree_block(eb);
320 btrfs_wait_tree_block_writeback(eb);
326 * Item overwrite used by replay and tree logging. eb, slot and key all refer
327 * to the src data we are copying out.
329 * root is the tree we are copying into, and path is a scratch
330 * path for use in this function (it should be released on entry and
331 * will be released on exit).
333 * If the key is already in the destination tree the existing item is
334 * overwritten. If the existing item isn't big enough, it is extended.
335 * If it is too large, it is truncated.
337 * If the key isn't in the destination yet, a new item is inserted.
339 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
340 struct btrfs_root *root,
341 struct btrfs_path *path,
342 struct extent_buffer *eb, int slot,
343 struct btrfs_key *key)
345 struct btrfs_fs_info *fs_info = root->fs_info;
348 u64 saved_i_size = 0;
349 int save_old_i_size = 0;
350 unsigned long src_ptr;
351 unsigned long dst_ptr;
352 int overwrite_root = 0;
353 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
355 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
358 item_size = btrfs_item_size_nr(eb, slot);
359 src_ptr = btrfs_item_ptr_offset(eb, slot);
361 /* look for the key in the destination tree */
362 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
369 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
371 if (dst_size != item_size)
374 if (item_size == 0) {
375 btrfs_release_path(path);
378 dst_copy = kmalloc(item_size, GFP_NOFS);
379 src_copy = kmalloc(item_size, GFP_NOFS);
380 if (!dst_copy || !src_copy) {
381 btrfs_release_path(path);
387 read_extent_buffer(eb, src_copy, src_ptr, item_size);
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
390 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
392 ret = memcmp(dst_copy, src_copy, item_size);
397 * they have the same contents, just return, this saves
398 * us from cowing blocks in the destination tree and doing
399 * extra writes that may not have been done by a previous
403 btrfs_release_path(path);
408 * We need to load the old nbytes into the inode so when we
409 * replay the extents we've logged we get the right nbytes.
412 struct btrfs_inode_item *item;
416 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
417 struct btrfs_inode_item);
418 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
419 item = btrfs_item_ptr(eb, slot,
420 struct btrfs_inode_item);
421 btrfs_set_inode_nbytes(eb, item, nbytes);
424 * If this is a directory we need to reset the i_size to
425 * 0 so that we can set it up properly when replaying
426 * the rest of the items in this log.
428 mode = btrfs_inode_mode(eb, item);
430 btrfs_set_inode_size(eb, item, 0);
432 } else if (inode_item) {
433 struct btrfs_inode_item *item;
437 * New inode, set nbytes to 0 so that the nbytes comes out
438 * properly when we replay the extents.
440 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
441 btrfs_set_inode_nbytes(eb, item, 0);
444 * If this is a directory we need to reset the i_size to 0 so
445 * that we can set it up properly when replaying the rest of
446 * the items in this log.
448 mode = btrfs_inode_mode(eb, item);
450 btrfs_set_inode_size(eb, item, 0);
453 btrfs_release_path(path);
454 /* try to insert the key into the destination tree */
455 path->skip_release_on_error = 1;
456 ret = btrfs_insert_empty_item(trans, root, path,
458 path->skip_release_on_error = 0;
460 /* make sure any existing item is the correct size */
461 if (ret == -EEXIST || ret == -EOVERFLOW) {
463 found_size = btrfs_item_size_nr(path->nodes[0],
465 if (found_size > item_size)
466 btrfs_truncate_item(fs_info, path, item_size, 1);
467 else if (found_size < item_size)
468 btrfs_extend_item(fs_info, path,
469 item_size - found_size);
473 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
476 /* don't overwrite an existing inode if the generation number
477 * was logged as zero. This is done when the tree logging code
478 * is just logging an inode to make sure it exists after recovery.
480 * Also, don't overwrite i_size on directories during replay.
481 * log replay inserts and removes directory items based on the
482 * state of the tree found in the subvolume, and i_size is modified
485 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
486 struct btrfs_inode_item *src_item;
487 struct btrfs_inode_item *dst_item;
489 src_item = (struct btrfs_inode_item *)src_ptr;
490 dst_item = (struct btrfs_inode_item *)dst_ptr;
492 if (btrfs_inode_generation(eb, src_item) == 0) {
493 struct extent_buffer *dst_eb = path->nodes[0];
494 const u64 ino_size = btrfs_inode_size(eb, src_item);
497 * For regular files an ino_size == 0 is used only when
498 * logging that an inode exists, as part of a directory
499 * fsync, and the inode wasn't fsynced before. In this
500 * case don't set the size of the inode in the fs/subvol
501 * tree, otherwise we would be throwing valid data away.
503 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
504 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
506 struct btrfs_map_token token;
508 btrfs_init_map_token(&token);
509 btrfs_set_token_inode_size(dst_eb, dst_item,
515 if (overwrite_root &&
516 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
517 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
519 saved_i_size = btrfs_inode_size(path->nodes[0],
524 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
527 if (save_old_i_size) {
528 struct btrfs_inode_item *dst_item;
529 dst_item = (struct btrfs_inode_item *)dst_ptr;
530 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
533 /* make sure the generation is filled in */
534 if (key->type == BTRFS_INODE_ITEM_KEY) {
535 struct btrfs_inode_item *dst_item;
536 dst_item = (struct btrfs_inode_item *)dst_ptr;
537 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
538 btrfs_set_inode_generation(path->nodes[0], dst_item,
543 btrfs_mark_buffer_dirty(path->nodes[0]);
544 btrfs_release_path(path);
549 * simple helper to read an inode off the disk from a given root
550 * This can only be called for subvolume roots and not for the log
552 static noinline struct inode *read_one_inode(struct btrfs_root *root,
555 struct btrfs_key key;
558 key.objectid = objectid;
559 key.type = BTRFS_INODE_ITEM_KEY;
561 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
564 } else if (is_bad_inode(inode)) {
571 /* replays a single extent in 'eb' at 'slot' with 'key' into the
572 * subvolume 'root'. path is released on entry and should be released
575 * extents in the log tree have not been allocated out of the extent
576 * tree yet. So, this completes the allocation, taking a reference
577 * as required if the extent already exists or creating a new extent
578 * if it isn't in the extent allocation tree yet.
580 * The extent is inserted into the file, dropping any existing extents
581 * from the file that overlap the new one.
583 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root,
585 struct btrfs_path *path,
586 struct extent_buffer *eb, int slot,
587 struct btrfs_key *key)
589 struct btrfs_fs_info *fs_info = root->fs_info;
592 u64 start = key->offset;
594 struct btrfs_file_extent_item *item;
595 struct inode *inode = NULL;
599 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
600 found_type = btrfs_file_extent_type(eb, item);
602 if (found_type == BTRFS_FILE_EXTENT_REG ||
603 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
604 nbytes = btrfs_file_extent_num_bytes(eb, item);
605 extent_end = start + nbytes;
608 * We don't add to the inodes nbytes if we are prealloc or a
611 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
613 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
614 size = btrfs_file_extent_inline_len(eb, slot, item);
615 nbytes = btrfs_file_extent_ram_bytes(eb, item);
616 extent_end = ALIGN(start + size,
617 fs_info->sectorsize);
623 inode = read_one_inode(root, key->objectid);
630 * first check to see if we already have this extent in the
631 * file. This must be done before the btrfs_drop_extents run
632 * so we don't try to drop this extent.
634 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(BTRFS_I(inode)),
638 (found_type == BTRFS_FILE_EXTENT_REG ||
639 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
640 struct btrfs_file_extent_item cmp1;
641 struct btrfs_file_extent_item cmp2;
642 struct btrfs_file_extent_item *existing;
643 struct extent_buffer *leaf;
645 leaf = path->nodes[0];
646 existing = btrfs_item_ptr(leaf, path->slots[0],
647 struct btrfs_file_extent_item);
649 read_extent_buffer(eb, &cmp1, (unsigned long)item,
651 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
655 * we already have a pointer to this exact extent,
656 * we don't have to do anything
658 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
659 btrfs_release_path(path);
663 btrfs_release_path(path);
665 /* drop any overlapping extents */
666 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
670 if (found_type == BTRFS_FILE_EXTENT_REG ||
671 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
673 unsigned long dest_offset;
674 struct btrfs_key ins;
676 ret = btrfs_insert_empty_item(trans, root, path, key,
680 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
682 copy_extent_buffer(path->nodes[0], eb, dest_offset,
683 (unsigned long)item, sizeof(*item));
685 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
686 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
687 ins.type = BTRFS_EXTENT_ITEM_KEY;
688 offset = key->offset - btrfs_file_extent_offset(eb, item);
691 * Manually record dirty extent, as here we did a shallow
692 * file extent item copy and skip normal backref update,
693 * but modifying extent tree all by ourselves.
694 * So need to manually record dirty extent for qgroup,
695 * as the owner of the file extent changed from log tree
696 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
698 ret = btrfs_qgroup_trace_extent(trans, fs_info,
699 btrfs_file_extent_disk_bytenr(eb, item),
700 btrfs_file_extent_disk_num_bytes(eb, item),
705 if (ins.objectid > 0) {
708 LIST_HEAD(ordered_sums);
710 * is this extent already allocated in the extent
711 * allocation tree? If so, just add a reference
713 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
716 ret = btrfs_inc_extent_ref(trans, fs_info,
717 ins.objectid, ins.offset,
718 0, root->root_key.objectid,
719 key->objectid, offset);
724 * insert the extent pointer in the extent
727 ret = btrfs_alloc_logged_file_extent(trans,
729 root->root_key.objectid,
730 key->objectid, offset, &ins);
734 btrfs_release_path(path);
736 if (btrfs_file_extent_compression(eb, item)) {
737 csum_start = ins.objectid;
738 csum_end = csum_start + ins.offset;
740 csum_start = ins.objectid +
741 btrfs_file_extent_offset(eb, item);
742 csum_end = csum_start +
743 btrfs_file_extent_num_bytes(eb, item);
746 ret = btrfs_lookup_csums_range(root->log_root,
747 csum_start, csum_end - 1,
752 * Now delete all existing cums in the csum root that
753 * cover our range. We do this because we can have an
754 * extent that is completely referenced by one file
755 * extent item and partially referenced by another
756 * file extent item (like after using the clone or
757 * extent_same ioctls). In this case if we end up doing
758 * the replay of the one that partially references the
759 * extent first, and we do not do the csum deletion
760 * below, we can get 2 csum items in the csum tree that
761 * overlap each other. For example, imagine our log has
762 * the two following file extent items:
764 * key (257 EXTENT_DATA 409600)
765 * extent data disk byte 12845056 nr 102400
766 * extent data offset 20480 nr 20480 ram 102400
768 * key (257 EXTENT_DATA 819200)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 0 nr 102400 ram 102400
772 * Where the second one fully references the 100K extent
773 * that starts at disk byte 12845056, and the log tree
774 * has a single csum item that covers the entire range
777 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
779 * After the first file extent item is replayed, the
780 * csum tree gets the following csum item:
782 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
784 * Which covers the 20K sub-range starting at offset 20K
785 * of our extent. Now when we replay the second file
786 * extent item, if we do not delete existing csum items
787 * that cover any of its blocks, we end up getting two
788 * csum items in our csum tree that overlap each other:
790 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
791 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
793 * Which is a problem, because after this anyone trying
794 * to lookup up for the checksum of any block of our
795 * extent starting at an offset of 40K or higher, will
796 * end up looking at the second csum item only, which
797 * does not contain the checksum for any block starting
798 * at offset 40K or higher of our extent.
800 while (!list_empty(&ordered_sums)) {
801 struct btrfs_ordered_sum *sums;
802 sums = list_entry(ordered_sums.next,
803 struct btrfs_ordered_sum,
806 ret = btrfs_del_csums(trans, fs_info,
810 ret = btrfs_csum_file_blocks(trans,
811 fs_info->csum_root, sums);
812 list_del(&sums->list);
818 btrfs_release_path(path);
820 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
821 /* inline extents are easy, we just overwrite them */
822 ret = overwrite_item(trans, root, path, eb, slot, key);
827 inode_add_bytes(inode, nbytes);
828 ret = btrfs_update_inode(trans, root, inode);
836 * when cleaning up conflicts between the directory names in the
837 * subvolume, directory names in the log and directory names in the
838 * inode back references, we may have to unlink inodes from directories.
840 * This is a helper function to do the unlink of a specific directory
843 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
844 struct btrfs_root *root,
845 struct btrfs_path *path,
847 struct btrfs_dir_item *di)
849 struct btrfs_fs_info *fs_info = root->fs_info;
853 struct extent_buffer *leaf;
854 struct btrfs_key location;
857 leaf = path->nodes[0];
859 btrfs_dir_item_key_to_cpu(leaf, di, &location);
860 name_len = btrfs_dir_name_len(leaf, di);
861 name = kmalloc(name_len, GFP_NOFS);
865 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
866 btrfs_release_path(path);
868 inode = read_one_inode(root, location.objectid);
874 ret = link_to_fixup_dir(trans, root, path, location.objectid);
878 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir), BTRFS_I(inode),
883 ret = btrfs_run_delayed_items(trans, fs_info);
891 * helper function to see if a given name and sequence number found
892 * in an inode back reference are already in a directory and correctly
893 * point to this inode
895 static noinline int inode_in_dir(struct btrfs_root *root,
896 struct btrfs_path *path,
897 u64 dirid, u64 objectid, u64 index,
898 const char *name, int name_len)
900 struct btrfs_dir_item *di;
901 struct btrfs_key location;
904 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
905 index, name, name_len, 0);
906 if (di && !IS_ERR(di)) {
907 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
908 if (location.objectid != objectid)
912 btrfs_release_path(path);
914 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
915 if (di && !IS_ERR(di)) {
916 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
917 if (location.objectid != objectid)
923 btrfs_release_path(path);
928 * helper function to check a log tree for a named back reference in
929 * an inode. This is used to decide if a back reference that is
930 * found in the subvolume conflicts with what we find in the log.
932 * inode backreferences may have multiple refs in a single item,
933 * during replay we process one reference at a time, and we don't
934 * want to delete valid links to a file from the subvolume if that
935 * link is also in the log.
937 static noinline int backref_in_log(struct btrfs_root *log,
938 struct btrfs_key *key,
940 const char *name, int namelen)
942 struct btrfs_path *path;
943 struct btrfs_inode_ref *ref;
945 unsigned long ptr_end;
946 unsigned long name_ptr;
952 path = btrfs_alloc_path();
956 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
960 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
962 if (key->type == BTRFS_INODE_EXTREF_KEY) {
963 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
964 name, namelen, NULL))
970 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
971 ptr_end = ptr + item_size;
972 while (ptr < ptr_end) {
973 ref = (struct btrfs_inode_ref *)ptr;
974 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
975 if (found_name_len == namelen) {
976 name_ptr = (unsigned long)(ref + 1);
977 ret = memcmp_extent_buffer(path->nodes[0], name,
984 ptr = (unsigned long)(ref + 1) + found_name_len;
987 btrfs_free_path(path);
991 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
992 struct btrfs_root *root,
993 struct btrfs_path *path,
994 struct btrfs_root *log_root,
995 struct inode *dir, struct inode *inode,
996 struct extent_buffer *eb,
997 u64 inode_objectid, u64 parent_objectid,
998 u64 ref_index, char *name, int namelen,
1001 struct btrfs_fs_info *fs_info = root->fs_info;
1004 int victim_name_len;
1005 struct extent_buffer *leaf;
1006 struct btrfs_dir_item *di;
1007 struct btrfs_key search_key;
1008 struct btrfs_inode_extref *extref;
1011 /* Search old style refs */
1012 search_key.objectid = inode_objectid;
1013 search_key.type = BTRFS_INODE_REF_KEY;
1014 search_key.offset = parent_objectid;
1015 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1017 struct btrfs_inode_ref *victim_ref;
1019 unsigned long ptr_end;
1021 leaf = path->nodes[0];
1023 /* are we trying to overwrite a back ref for the root directory
1024 * if so, just jump out, we're done
1026 if (search_key.objectid == search_key.offset)
1029 /* check all the names in this back reference to see
1030 * if they are in the log. if so, we allow them to stay
1031 * otherwise they must be unlinked as a conflict
1033 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1034 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1035 while (ptr < ptr_end) {
1036 victim_ref = (struct btrfs_inode_ref *)ptr;
1037 victim_name_len = btrfs_inode_ref_name_len(leaf,
1039 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1043 read_extent_buffer(leaf, victim_name,
1044 (unsigned long)(victim_ref + 1),
1047 if (!backref_in_log(log_root, &search_key,
1052 btrfs_release_path(path);
1054 ret = btrfs_unlink_inode(trans, root,
1055 BTRFS_I(dir), BTRFS_I(inode),
1056 victim_name, victim_name_len);
1060 ret = btrfs_run_delayed_items(trans, fs_info);
1068 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1072 * NOTE: we have searched root tree and checked the
1073 * corresponding ref, it does not need to check again.
1077 btrfs_release_path(path);
1079 /* Same search but for extended refs */
1080 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1081 inode_objectid, parent_objectid, 0,
1083 if (!IS_ERR_OR_NULL(extref)) {
1087 struct inode *victim_parent;
1089 leaf = path->nodes[0];
1091 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1092 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1094 while (cur_offset < item_size) {
1095 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1097 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1099 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1102 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1105 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1108 search_key.objectid = inode_objectid;
1109 search_key.type = BTRFS_INODE_EXTREF_KEY;
1110 search_key.offset = btrfs_extref_hash(parent_objectid,
1114 if (!backref_in_log(log_root, &search_key,
1115 parent_objectid, victim_name,
1118 victim_parent = read_one_inode(root,
1120 if (victim_parent) {
1122 btrfs_release_path(path);
1124 ret = btrfs_unlink_inode(trans, root,
1125 BTRFS_I(victim_parent),
1130 ret = btrfs_run_delayed_items(
1134 iput(victim_parent);
1145 cur_offset += victim_name_len + sizeof(*extref);
1149 btrfs_release_path(path);
1151 /* look for a conflicting sequence number */
1152 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(BTRFS_I(dir)),
1153 ref_index, name, namelen, 0);
1154 if (di && !IS_ERR(di)) {
1155 ret = drop_one_dir_item(trans, root, path, dir, di);
1159 btrfs_release_path(path);
1161 /* look for a conflicing name */
1162 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(BTRFS_I(dir)),
1164 if (di && !IS_ERR(di)) {
1165 ret = drop_one_dir_item(trans, root, path, dir, di);
1169 btrfs_release_path(path);
1174 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1175 u32 *namelen, char **name, u64 *index,
1176 u64 *parent_objectid)
1178 struct btrfs_inode_extref *extref;
1180 extref = (struct btrfs_inode_extref *)ref_ptr;
1182 *namelen = btrfs_inode_extref_name_len(eb, extref);
1183 *name = kmalloc(*namelen, GFP_NOFS);
1187 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1190 *index = btrfs_inode_extref_index(eb, extref);
1191 if (parent_objectid)
1192 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1197 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1198 u32 *namelen, char **name, u64 *index)
1200 struct btrfs_inode_ref *ref;
1202 ref = (struct btrfs_inode_ref *)ref_ptr;
1204 *namelen = btrfs_inode_ref_name_len(eb, ref);
1205 *name = kmalloc(*namelen, GFP_NOFS);
1209 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1211 *index = btrfs_inode_ref_index(eb, ref);
1217 * replay one inode back reference item found in the log tree.
1218 * eb, slot and key refer to the buffer and key found in the log tree.
1219 * root is the destination we are replaying into, and path is for temp
1220 * use by this function. (it should be released on return).
1222 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1223 struct btrfs_root *root,
1224 struct btrfs_root *log,
1225 struct btrfs_path *path,
1226 struct extent_buffer *eb, int slot,
1227 struct btrfs_key *key)
1229 struct inode *dir = NULL;
1230 struct inode *inode = NULL;
1231 unsigned long ref_ptr;
1232 unsigned long ref_end;
1236 int search_done = 0;
1237 int log_ref_ver = 0;
1238 u64 parent_objectid;
1241 int ref_struct_size;
1243 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1244 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1246 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1247 struct btrfs_inode_extref *r;
1249 ref_struct_size = sizeof(struct btrfs_inode_extref);
1251 r = (struct btrfs_inode_extref *)ref_ptr;
1252 parent_objectid = btrfs_inode_extref_parent(eb, r);
1254 ref_struct_size = sizeof(struct btrfs_inode_ref);
1255 parent_objectid = key->offset;
1257 inode_objectid = key->objectid;
1260 * it is possible that we didn't log all the parent directories
1261 * for a given inode. If we don't find the dir, just don't
1262 * copy the back ref in. The link count fixup code will take
1265 dir = read_one_inode(root, parent_objectid);
1271 inode = read_one_inode(root, inode_objectid);
1277 while (ref_ptr < ref_end) {
1279 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1280 &ref_index, &parent_objectid);
1282 * parent object can change from one array
1286 dir = read_one_inode(root, parent_objectid);
1292 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1298 /* if we already have a perfect match, we're done */
1299 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)), btrfs_ino(BTRFS_I(inode)),
1300 ref_index, name, namelen)) {
1302 * look for a conflicting back reference in the
1303 * metadata. if we find one we have to unlink that name
1304 * of the file before we add our new link. Later on, we
1305 * overwrite any existing back reference, and we don't
1306 * want to create dangling pointers in the directory.
1310 ret = __add_inode_ref(trans, root, path, log,
1314 ref_index, name, namelen,
1323 /* insert our name */
1324 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1329 btrfs_update_inode(trans, root, inode);
1332 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1341 /* finally write the back reference in the inode */
1342 ret = overwrite_item(trans, root, path, eb, slot, key);
1344 btrfs_release_path(path);
1351 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1352 struct btrfs_root *root, u64 ino)
1356 ret = btrfs_insert_orphan_item(trans, root, ino);
1363 static int count_inode_extrefs(struct btrfs_root *root,
1364 struct inode *inode, struct btrfs_path *path)
1368 unsigned int nlink = 0;
1371 u64 inode_objectid = btrfs_ino(BTRFS_I(inode));
1374 struct btrfs_inode_extref *extref;
1375 struct extent_buffer *leaf;
1378 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1383 leaf = path->nodes[0];
1384 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1385 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1388 while (cur_offset < item_size) {
1389 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1390 name_len = btrfs_inode_extref_name_len(leaf, extref);
1394 cur_offset += name_len + sizeof(*extref);
1398 btrfs_release_path(path);
1400 btrfs_release_path(path);
1402 if (ret < 0 && ret != -ENOENT)
1407 static int count_inode_refs(struct btrfs_root *root,
1408 struct inode *inode, struct btrfs_path *path)
1411 struct btrfs_key key;
1412 unsigned int nlink = 0;
1414 unsigned long ptr_end;
1416 u64 ino = btrfs_ino(BTRFS_I(inode));
1419 key.type = BTRFS_INODE_REF_KEY;
1420 key.offset = (u64)-1;
1423 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1427 if (path->slots[0] == 0)
1432 btrfs_item_key_to_cpu(path->nodes[0], &key,
1434 if (key.objectid != ino ||
1435 key.type != BTRFS_INODE_REF_KEY)
1437 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1438 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1440 while (ptr < ptr_end) {
1441 struct btrfs_inode_ref *ref;
1443 ref = (struct btrfs_inode_ref *)ptr;
1444 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1446 ptr = (unsigned long)(ref + 1) + name_len;
1450 if (key.offset == 0)
1452 if (path->slots[0] > 0) {
1457 btrfs_release_path(path);
1459 btrfs_release_path(path);
1465 * There are a few corners where the link count of the file can't
1466 * be properly maintained during replay. So, instead of adding
1467 * lots of complexity to the log code, we just scan the backrefs
1468 * for any file that has been through replay.
1470 * The scan will update the link count on the inode to reflect the
1471 * number of back refs found. If it goes down to zero, the iput
1472 * will free the inode.
1474 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1475 struct btrfs_root *root,
1476 struct inode *inode)
1478 struct btrfs_path *path;
1481 u64 ino = btrfs_ino(BTRFS_I(inode));
1483 path = btrfs_alloc_path();
1487 ret = count_inode_refs(root, inode, path);
1493 ret = count_inode_extrefs(root, inode, path);
1501 if (nlink != inode->i_nlink) {
1502 set_nlink(inode, nlink);
1503 btrfs_update_inode(trans, root, inode);
1505 BTRFS_I(inode)->index_cnt = (u64)-1;
1507 if (inode->i_nlink == 0) {
1508 if (S_ISDIR(inode->i_mode)) {
1509 ret = replay_dir_deletes(trans, root, NULL, path,
1514 ret = insert_orphan_item(trans, root, ino);
1518 btrfs_free_path(path);
1522 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1523 struct btrfs_root *root,
1524 struct btrfs_path *path)
1527 struct btrfs_key key;
1528 struct inode *inode;
1530 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1531 key.type = BTRFS_ORPHAN_ITEM_KEY;
1532 key.offset = (u64)-1;
1534 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1539 if (path->slots[0] == 0)
1544 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1545 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1546 key.type != BTRFS_ORPHAN_ITEM_KEY)
1549 ret = btrfs_del_item(trans, root, path);
1553 btrfs_release_path(path);
1554 inode = read_one_inode(root, key.offset);
1558 ret = fixup_inode_link_count(trans, root, inode);
1564 * fixup on a directory may create new entries,
1565 * make sure we always look for the highset possible
1568 key.offset = (u64)-1;
1572 btrfs_release_path(path);
1578 * record a given inode in the fixup dir so we can check its link
1579 * count when replay is done. The link count is incremented here
1580 * so the inode won't go away until we check it
1582 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1583 struct btrfs_root *root,
1584 struct btrfs_path *path,
1587 struct btrfs_key key;
1589 struct inode *inode;
1591 inode = read_one_inode(root, objectid);
1595 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1596 key.type = BTRFS_ORPHAN_ITEM_KEY;
1597 key.offset = objectid;
1599 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1601 btrfs_release_path(path);
1603 if (!inode->i_nlink)
1604 set_nlink(inode, 1);
1607 ret = btrfs_update_inode(trans, root, inode);
1608 } else if (ret == -EEXIST) {
1611 BUG(); /* Logic Error */
1619 * when replaying the log for a directory, we only insert names
1620 * for inodes that actually exist. This means an fsync on a directory
1621 * does not implicitly fsync all the new files in it
1623 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1624 struct btrfs_root *root,
1625 u64 dirid, u64 index,
1626 char *name, int name_len,
1627 struct btrfs_key *location)
1629 struct inode *inode;
1633 inode = read_one_inode(root, location->objectid);
1637 dir = read_one_inode(root, dirid);
1643 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1645 /* FIXME, put inode into FIXUP list */
1653 * Return true if an inode reference exists in the log for the given name,
1654 * inode and parent inode.
1656 static bool name_in_log_ref(struct btrfs_root *log_root,
1657 const char *name, const int name_len,
1658 const u64 dirid, const u64 ino)
1660 struct btrfs_key search_key;
1662 search_key.objectid = ino;
1663 search_key.type = BTRFS_INODE_REF_KEY;
1664 search_key.offset = dirid;
1665 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1668 search_key.type = BTRFS_INODE_EXTREF_KEY;
1669 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1670 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1677 * take a single entry in a log directory item and replay it into
1680 * if a conflicting item exists in the subdirectory already,
1681 * the inode it points to is unlinked and put into the link count
1684 * If a name from the log points to a file or directory that does
1685 * not exist in the FS, it is skipped. fsyncs on directories
1686 * do not force down inodes inside that directory, just changes to the
1687 * names or unlinks in a directory.
1689 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1690 * non-existing inode) and 1 if the name was replayed.
1692 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1693 struct btrfs_root *root,
1694 struct btrfs_path *path,
1695 struct extent_buffer *eb,
1696 struct btrfs_dir_item *di,
1697 struct btrfs_key *key)
1701 struct btrfs_dir_item *dst_di;
1702 struct btrfs_key found_key;
1703 struct btrfs_key log_key;
1708 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1709 bool name_added = false;
1711 dir = read_one_inode(root, key->objectid);
1715 name_len = btrfs_dir_name_len(eb, di);
1716 name = kmalloc(name_len, GFP_NOFS);
1722 log_type = btrfs_dir_type(eb, di);
1723 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1726 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1727 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1732 btrfs_release_path(path);
1734 if (key->type == BTRFS_DIR_ITEM_KEY) {
1735 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1737 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1738 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1747 if (IS_ERR_OR_NULL(dst_di)) {
1748 /* we need a sequence number to insert, so we only
1749 * do inserts for the BTRFS_DIR_INDEX_KEY types
1751 if (key->type != BTRFS_DIR_INDEX_KEY)
1756 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1757 /* the existing item matches the logged item */
1758 if (found_key.objectid == log_key.objectid &&
1759 found_key.type == log_key.type &&
1760 found_key.offset == log_key.offset &&
1761 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1762 update_size = false;
1767 * don't drop the conflicting directory entry if the inode
1768 * for the new entry doesn't exist
1773 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1777 if (key->type == BTRFS_DIR_INDEX_KEY)
1780 btrfs_release_path(path);
1781 if (!ret && update_size) {
1782 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1783 ret = btrfs_update_inode(trans, root, dir);
1787 if (!ret && name_added)
1792 if (name_in_log_ref(root->log_root, name, name_len,
1793 key->objectid, log_key.objectid)) {
1794 /* The dentry will be added later. */
1796 update_size = false;
1799 btrfs_release_path(path);
1800 ret = insert_one_name(trans, root, key->objectid, key->offset,
1801 name, name_len, &log_key);
1802 if (ret && ret != -ENOENT && ret != -EEXIST)
1806 update_size = false;
1812 * find all the names in a directory item and reconcile them into
1813 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1814 * one name in a directory item, but the same code gets used for
1815 * both directory index types
1817 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1818 struct btrfs_root *root,
1819 struct btrfs_path *path,
1820 struct extent_buffer *eb, int slot,
1821 struct btrfs_key *key)
1823 struct btrfs_fs_info *fs_info = root->fs_info;
1825 u32 item_size = btrfs_item_size_nr(eb, slot);
1826 struct btrfs_dir_item *di;
1829 unsigned long ptr_end;
1830 struct btrfs_path *fixup_path = NULL;
1832 ptr = btrfs_item_ptr_offset(eb, slot);
1833 ptr_end = ptr + item_size;
1834 while (ptr < ptr_end) {
1835 di = (struct btrfs_dir_item *)ptr;
1836 if (verify_dir_item(fs_info, eb, di))
1838 name_len = btrfs_dir_name_len(eb, di);
1839 ret = replay_one_name(trans, root, path, eb, di, key);
1842 ptr = (unsigned long)(di + 1);
1846 * If this entry refers to a non-directory (directories can not
1847 * have a link count > 1) and it was added in the transaction
1848 * that was not committed, make sure we fixup the link count of
1849 * the inode it the entry points to. Otherwise something like
1850 * the following would result in a directory pointing to an
1851 * inode with a wrong link that does not account for this dir
1859 * ln testdir/bar testdir/bar_link
1860 * ln testdir/foo testdir/foo_link
1861 * xfs_io -c "fsync" testdir/bar
1865 * mount fs, log replay happens
1867 * File foo would remain with a link count of 1 when it has two
1868 * entries pointing to it in the directory testdir. This would
1869 * make it impossible to ever delete the parent directory has
1870 * it would result in stale dentries that can never be deleted.
1872 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1873 struct btrfs_key di_key;
1876 fixup_path = btrfs_alloc_path();
1883 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1884 ret = link_to_fixup_dir(trans, root, fixup_path,
1891 btrfs_free_path(fixup_path);
1896 * directory replay has two parts. There are the standard directory
1897 * items in the log copied from the subvolume, and range items
1898 * created in the log while the subvolume was logged.
1900 * The range items tell us which parts of the key space the log
1901 * is authoritative for. During replay, if a key in the subvolume
1902 * directory is in a logged range item, but not actually in the log
1903 * that means it was deleted from the directory before the fsync
1904 * and should be removed.
1906 static noinline int find_dir_range(struct btrfs_root *root,
1907 struct btrfs_path *path,
1908 u64 dirid, int key_type,
1909 u64 *start_ret, u64 *end_ret)
1911 struct btrfs_key key;
1913 struct btrfs_dir_log_item *item;
1917 if (*start_ret == (u64)-1)
1920 key.objectid = dirid;
1921 key.type = key_type;
1922 key.offset = *start_ret;
1924 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1928 if (path->slots[0] == 0)
1933 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1935 if (key.type != key_type || key.objectid != dirid) {
1939 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1940 struct btrfs_dir_log_item);
1941 found_end = btrfs_dir_log_end(path->nodes[0], item);
1943 if (*start_ret >= key.offset && *start_ret <= found_end) {
1945 *start_ret = key.offset;
1946 *end_ret = found_end;
1951 /* check the next slot in the tree to see if it is a valid item */
1952 nritems = btrfs_header_nritems(path->nodes[0]);
1954 if (path->slots[0] >= nritems) {
1955 ret = btrfs_next_leaf(root, path);
1960 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1962 if (key.type != key_type || key.objectid != dirid) {
1966 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1967 struct btrfs_dir_log_item);
1968 found_end = btrfs_dir_log_end(path->nodes[0], item);
1969 *start_ret = key.offset;
1970 *end_ret = found_end;
1973 btrfs_release_path(path);
1978 * this looks for a given directory item in the log. If the directory
1979 * item is not in the log, the item is removed and the inode it points
1982 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1983 struct btrfs_root *root,
1984 struct btrfs_root *log,
1985 struct btrfs_path *path,
1986 struct btrfs_path *log_path,
1988 struct btrfs_key *dir_key)
1990 struct btrfs_fs_info *fs_info = root->fs_info;
1992 struct extent_buffer *eb;
1995 struct btrfs_dir_item *di;
1996 struct btrfs_dir_item *log_di;
1999 unsigned long ptr_end;
2001 struct inode *inode;
2002 struct btrfs_key location;
2005 eb = path->nodes[0];
2006 slot = path->slots[0];
2007 item_size = btrfs_item_size_nr(eb, slot);
2008 ptr = btrfs_item_ptr_offset(eb, slot);
2009 ptr_end = ptr + item_size;
2010 while (ptr < ptr_end) {
2011 di = (struct btrfs_dir_item *)ptr;
2012 if (verify_dir_item(fs_info, eb, di)) {
2017 name_len = btrfs_dir_name_len(eb, di);
2018 name = kmalloc(name_len, GFP_NOFS);
2023 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2026 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2027 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2030 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2031 log_di = btrfs_lookup_dir_index_item(trans, log,
2037 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2038 btrfs_dir_item_key_to_cpu(eb, di, &location);
2039 btrfs_release_path(path);
2040 btrfs_release_path(log_path);
2041 inode = read_one_inode(root, location.objectid);
2047 ret = link_to_fixup_dir(trans, root,
2048 path, location.objectid);
2056 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2057 BTRFS_I(inode), name, name_len);
2059 ret = btrfs_run_delayed_items(trans, fs_info);
2065 /* there might still be more names under this key
2066 * check and repeat if required
2068 ret = btrfs_search_slot(NULL, root, dir_key, path,
2074 } else if (IS_ERR(log_di)) {
2076 return PTR_ERR(log_di);
2078 btrfs_release_path(log_path);
2081 ptr = (unsigned long)(di + 1);
2086 btrfs_release_path(path);
2087 btrfs_release_path(log_path);
2091 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2092 struct btrfs_root *root,
2093 struct btrfs_root *log,
2094 struct btrfs_path *path,
2097 struct btrfs_key search_key;
2098 struct btrfs_path *log_path;
2103 log_path = btrfs_alloc_path();
2107 search_key.objectid = ino;
2108 search_key.type = BTRFS_XATTR_ITEM_KEY;
2109 search_key.offset = 0;
2111 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2115 nritems = btrfs_header_nritems(path->nodes[0]);
2116 for (i = path->slots[0]; i < nritems; i++) {
2117 struct btrfs_key key;
2118 struct btrfs_dir_item *di;
2119 struct btrfs_dir_item *log_di;
2123 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2124 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2129 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2130 total_size = btrfs_item_size_nr(path->nodes[0], i);
2132 while (cur < total_size) {
2133 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2134 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2135 u32 this_len = sizeof(*di) + name_len + data_len;
2138 name = kmalloc(name_len, GFP_NOFS);
2143 read_extent_buffer(path->nodes[0], name,
2144 (unsigned long)(di + 1), name_len);
2146 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2148 btrfs_release_path(log_path);
2150 /* Doesn't exist in log tree, so delete it. */
2151 btrfs_release_path(path);
2152 di = btrfs_lookup_xattr(trans, root, path, ino,
2153 name, name_len, -1);
2160 ret = btrfs_delete_one_dir_name(trans, root,
2164 btrfs_release_path(path);
2169 if (IS_ERR(log_di)) {
2170 ret = PTR_ERR(log_di);
2174 di = (struct btrfs_dir_item *)((char *)di + this_len);
2177 ret = btrfs_next_leaf(root, path);
2183 btrfs_free_path(log_path);
2184 btrfs_release_path(path);
2190 * deletion replay happens before we copy any new directory items
2191 * out of the log or out of backreferences from inodes. It
2192 * scans the log to find ranges of keys that log is authoritative for,
2193 * and then scans the directory to find items in those ranges that are
2194 * not present in the log.
2196 * Anything we don't find in the log is unlinked and removed from the
2199 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2200 struct btrfs_root *root,
2201 struct btrfs_root *log,
2202 struct btrfs_path *path,
2203 u64 dirid, int del_all)
2207 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2209 struct btrfs_key dir_key;
2210 struct btrfs_key found_key;
2211 struct btrfs_path *log_path;
2214 dir_key.objectid = dirid;
2215 dir_key.type = BTRFS_DIR_ITEM_KEY;
2216 log_path = btrfs_alloc_path();
2220 dir = read_one_inode(root, dirid);
2221 /* it isn't an error if the inode isn't there, that can happen
2222 * because we replay the deletes before we copy in the inode item
2226 btrfs_free_path(log_path);
2234 range_end = (u64)-1;
2236 ret = find_dir_range(log, path, dirid, key_type,
2237 &range_start, &range_end);
2242 dir_key.offset = range_start;
2245 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2250 nritems = btrfs_header_nritems(path->nodes[0]);
2251 if (path->slots[0] >= nritems) {
2252 ret = btrfs_next_leaf(root, path);
2256 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2258 if (found_key.objectid != dirid ||
2259 found_key.type != dir_key.type)
2262 if (found_key.offset > range_end)
2265 ret = check_item_in_log(trans, root, log, path,
2270 if (found_key.offset == (u64)-1)
2272 dir_key.offset = found_key.offset + 1;
2274 btrfs_release_path(path);
2275 if (range_end == (u64)-1)
2277 range_start = range_end + 1;
2282 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2283 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2284 dir_key.type = BTRFS_DIR_INDEX_KEY;
2285 btrfs_release_path(path);
2289 btrfs_release_path(path);
2290 btrfs_free_path(log_path);
2296 * the process_func used to replay items from the log tree. This
2297 * gets called in two different stages. The first stage just looks
2298 * for inodes and makes sure they are all copied into the subvolume.
2300 * The second stage copies all the other item types from the log into
2301 * the subvolume. The two stage approach is slower, but gets rid of
2302 * lots of complexity around inodes referencing other inodes that exist
2303 * only in the log (references come from either directory items or inode
2306 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2307 struct walk_control *wc, u64 gen)
2310 struct btrfs_path *path;
2311 struct btrfs_root *root = wc->replay_dest;
2312 struct btrfs_key key;
2317 ret = btrfs_read_buffer(eb, gen);
2321 level = btrfs_header_level(eb);
2326 path = btrfs_alloc_path();
2330 nritems = btrfs_header_nritems(eb);
2331 for (i = 0; i < nritems; i++) {
2332 btrfs_item_key_to_cpu(eb, &key, i);
2334 /* inode keys are done during the first stage */
2335 if (key.type == BTRFS_INODE_ITEM_KEY &&
2336 wc->stage == LOG_WALK_REPLAY_INODES) {
2337 struct btrfs_inode_item *inode_item;
2340 inode_item = btrfs_item_ptr(eb, i,
2341 struct btrfs_inode_item);
2342 ret = replay_xattr_deletes(wc->trans, root, log,
2343 path, key.objectid);
2346 mode = btrfs_inode_mode(eb, inode_item);
2347 if (S_ISDIR(mode)) {
2348 ret = replay_dir_deletes(wc->trans,
2349 root, log, path, key.objectid, 0);
2353 ret = overwrite_item(wc->trans, root, path,
2358 /* for regular files, make sure corresponding
2359 * orphan item exist. extents past the new EOF
2360 * will be truncated later by orphan cleanup.
2362 if (S_ISREG(mode)) {
2363 ret = insert_orphan_item(wc->trans, root,
2369 ret = link_to_fixup_dir(wc->trans, root,
2370 path, key.objectid);
2375 if (key.type == BTRFS_DIR_INDEX_KEY &&
2376 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2377 ret = replay_one_dir_item(wc->trans, root, path,
2383 if (wc->stage < LOG_WALK_REPLAY_ALL)
2386 /* these keys are simply copied */
2387 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2388 ret = overwrite_item(wc->trans, root, path,
2392 } else if (key.type == BTRFS_INODE_REF_KEY ||
2393 key.type == BTRFS_INODE_EXTREF_KEY) {
2394 ret = add_inode_ref(wc->trans, root, log, path,
2396 if (ret && ret != -ENOENT)
2399 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2400 ret = replay_one_extent(wc->trans, root, path,
2404 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2405 ret = replay_one_dir_item(wc->trans, root, path,
2411 btrfs_free_path(path);
2415 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2416 struct btrfs_root *root,
2417 struct btrfs_path *path, int *level,
2418 struct walk_control *wc)
2420 struct btrfs_fs_info *fs_info = root->fs_info;
2424 struct extent_buffer *next;
2425 struct extent_buffer *cur;
2426 struct extent_buffer *parent;
2430 WARN_ON(*level < 0);
2431 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2433 while (*level > 0) {
2434 WARN_ON(*level < 0);
2435 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2436 cur = path->nodes[*level];
2438 WARN_ON(btrfs_header_level(cur) != *level);
2440 if (path->slots[*level] >=
2441 btrfs_header_nritems(cur))
2444 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2445 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2446 blocksize = fs_info->nodesize;
2448 parent = path->nodes[*level];
2449 root_owner = btrfs_header_owner(parent);
2451 next = btrfs_find_create_tree_block(fs_info, bytenr);
2453 return PTR_ERR(next);
2456 ret = wc->process_func(root, next, wc, ptr_gen);
2458 free_extent_buffer(next);
2462 path->slots[*level]++;
2464 ret = btrfs_read_buffer(next, ptr_gen);
2466 free_extent_buffer(next);
2471 btrfs_tree_lock(next);
2472 btrfs_set_lock_blocking(next);
2473 clean_tree_block(trans, fs_info, next);
2474 btrfs_wait_tree_block_writeback(next);
2475 btrfs_tree_unlock(next);
2478 WARN_ON(root_owner !=
2479 BTRFS_TREE_LOG_OBJECTID);
2480 ret = btrfs_free_and_pin_reserved_extent(
2484 free_extent_buffer(next);
2488 free_extent_buffer(next);
2491 ret = btrfs_read_buffer(next, ptr_gen);
2493 free_extent_buffer(next);
2497 WARN_ON(*level <= 0);
2498 if (path->nodes[*level-1])
2499 free_extent_buffer(path->nodes[*level-1]);
2500 path->nodes[*level-1] = next;
2501 *level = btrfs_header_level(next);
2502 path->slots[*level] = 0;
2505 WARN_ON(*level < 0);
2506 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2508 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2514 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2515 struct btrfs_root *root,
2516 struct btrfs_path *path, int *level,
2517 struct walk_control *wc)
2519 struct btrfs_fs_info *fs_info = root->fs_info;
2525 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2526 slot = path->slots[i];
2527 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2530 WARN_ON(*level == 0);
2533 struct extent_buffer *parent;
2534 if (path->nodes[*level] == root->node)
2535 parent = path->nodes[*level];
2537 parent = path->nodes[*level + 1];
2539 root_owner = btrfs_header_owner(parent);
2540 ret = wc->process_func(root, path->nodes[*level], wc,
2541 btrfs_header_generation(path->nodes[*level]));
2546 struct extent_buffer *next;
2548 next = path->nodes[*level];
2551 btrfs_tree_lock(next);
2552 btrfs_set_lock_blocking(next);
2553 clean_tree_block(trans, fs_info, next);
2554 btrfs_wait_tree_block_writeback(next);
2555 btrfs_tree_unlock(next);
2558 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2559 ret = btrfs_free_and_pin_reserved_extent(
2561 path->nodes[*level]->start,
2562 path->nodes[*level]->len);
2566 free_extent_buffer(path->nodes[*level]);
2567 path->nodes[*level] = NULL;
2575 * drop the reference count on the tree rooted at 'snap'. This traverses
2576 * the tree freeing any blocks that have a ref count of zero after being
2579 static int walk_log_tree(struct btrfs_trans_handle *trans,
2580 struct btrfs_root *log, struct walk_control *wc)
2582 struct btrfs_fs_info *fs_info = log->fs_info;
2586 struct btrfs_path *path;
2589 path = btrfs_alloc_path();
2593 level = btrfs_header_level(log->node);
2595 path->nodes[level] = log->node;
2596 extent_buffer_get(log->node);
2597 path->slots[level] = 0;
2600 wret = walk_down_log_tree(trans, log, path, &level, wc);
2608 wret = walk_up_log_tree(trans, log, path, &level, wc);
2617 /* was the root node processed? if not, catch it here */
2618 if (path->nodes[orig_level]) {
2619 ret = wc->process_func(log, path->nodes[orig_level], wc,
2620 btrfs_header_generation(path->nodes[orig_level]));
2624 struct extent_buffer *next;
2626 next = path->nodes[orig_level];
2629 btrfs_tree_lock(next);
2630 btrfs_set_lock_blocking(next);
2631 clean_tree_block(trans, fs_info, next);
2632 btrfs_wait_tree_block_writeback(next);
2633 btrfs_tree_unlock(next);
2636 WARN_ON(log->root_key.objectid !=
2637 BTRFS_TREE_LOG_OBJECTID);
2638 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2639 next->start, next->len);
2646 btrfs_free_path(path);
2651 * helper function to update the item for a given subvolumes log root
2652 * in the tree of log roots
2654 static int update_log_root(struct btrfs_trans_handle *trans,
2655 struct btrfs_root *log)
2657 struct btrfs_fs_info *fs_info = log->fs_info;
2660 if (log->log_transid == 1) {
2661 /* insert root item on the first sync */
2662 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2663 &log->root_key, &log->root_item);
2665 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2666 &log->root_key, &log->root_item);
2671 static void wait_log_commit(struct btrfs_root *root, int transid)
2674 int index = transid % 2;
2677 * we only allow two pending log transactions at a time,
2678 * so we know that if ours is more than 2 older than the
2679 * current transaction, we're done
2682 prepare_to_wait(&root->log_commit_wait[index],
2683 &wait, TASK_UNINTERRUPTIBLE);
2684 mutex_unlock(&root->log_mutex);
2686 if (root->log_transid_committed < transid &&
2687 atomic_read(&root->log_commit[index]))
2690 finish_wait(&root->log_commit_wait[index], &wait);
2691 mutex_lock(&root->log_mutex);
2692 } while (root->log_transid_committed < transid &&
2693 atomic_read(&root->log_commit[index]));
2696 static void wait_for_writer(struct btrfs_root *root)
2700 while (atomic_read(&root->log_writers)) {
2701 prepare_to_wait(&root->log_writer_wait,
2702 &wait, TASK_UNINTERRUPTIBLE);
2703 mutex_unlock(&root->log_mutex);
2704 if (atomic_read(&root->log_writers))
2706 finish_wait(&root->log_writer_wait, &wait);
2707 mutex_lock(&root->log_mutex);
2711 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2712 struct btrfs_log_ctx *ctx)
2717 mutex_lock(&root->log_mutex);
2718 list_del_init(&ctx->list);
2719 mutex_unlock(&root->log_mutex);
2723 * Invoked in log mutex context, or be sure there is no other task which
2724 * can access the list.
2726 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2727 int index, int error)
2729 struct btrfs_log_ctx *ctx;
2730 struct btrfs_log_ctx *safe;
2732 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2733 list_del_init(&ctx->list);
2734 ctx->log_ret = error;
2737 INIT_LIST_HEAD(&root->log_ctxs[index]);
2741 * btrfs_sync_log does sends a given tree log down to the disk and
2742 * updates the super blocks to record it. When this call is done,
2743 * you know that any inodes previously logged are safely on disk only
2746 * Any other return value means you need to call btrfs_commit_transaction.
2747 * Some of the edge cases for fsyncing directories that have had unlinks
2748 * or renames done in the past mean that sometimes the only safe
2749 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2750 * that has happened.
2752 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2753 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2759 struct btrfs_fs_info *fs_info = root->fs_info;
2760 struct btrfs_root *log = root->log_root;
2761 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2762 int log_transid = 0;
2763 struct btrfs_log_ctx root_log_ctx;
2764 struct blk_plug plug;
2766 mutex_lock(&root->log_mutex);
2767 log_transid = ctx->log_transid;
2768 if (root->log_transid_committed >= log_transid) {
2769 mutex_unlock(&root->log_mutex);
2770 return ctx->log_ret;
2773 index1 = log_transid % 2;
2774 if (atomic_read(&root->log_commit[index1])) {
2775 wait_log_commit(root, log_transid);
2776 mutex_unlock(&root->log_mutex);
2777 return ctx->log_ret;
2779 ASSERT(log_transid == root->log_transid);
2780 atomic_set(&root->log_commit[index1], 1);
2782 /* wait for previous tree log sync to complete */
2783 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2784 wait_log_commit(root, log_transid - 1);
2787 int batch = atomic_read(&root->log_batch);
2788 /* when we're on an ssd, just kick the log commit out */
2789 if (!btrfs_test_opt(fs_info, SSD) &&
2790 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2791 mutex_unlock(&root->log_mutex);
2792 schedule_timeout_uninterruptible(1);
2793 mutex_lock(&root->log_mutex);
2795 wait_for_writer(root);
2796 if (batch == atomic_read(&root->log_batch))
2800 /* bail out if we need to do a full commit */
2801 if (btrfs_need_log_full_commit(fs_info, trans)) {
2803 btrfs_free_logged_extents(log, log_transid);
2804 mutex_unlock(&root->log_mutex);
2808 if (log_transid % 2 == 0)
2809 mark = EXTENT_DIRTY;
2813 /* we start IO on all the marked extents here, but we don't actually
2814 * wait for them until later.
2816 blk_start_plug(&plug);
2817 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2819 blk_finish_plug(&plug);
2820 btrfs_abort_transaction(trans, ret);
2821 btrfs_free_logged_extents(log, log_transid);
2822 btrfs_set_log_full_commit(fs_info, trans);
2823 mutex_unlock(&root->log_mutex);
2827 btrfs_set_root_node(&log->root_item, log->node);
2829 root->log_transid++;
2830 log->log_transid = root->log_transid;
2831 root->log_start_pid = 0;
2833 * IO has been started, blocks of the log tree have WRITTEN flag set
2834 * in their headers. new modifications of the log will be written to
2835 * new positions. so it's safe to allow log writers to go in.
2837 mutex_unlock(&root->log_mutex);
2839 btrfs_init_log_ctx(&root_log_ctx, NULL);
2841 mutex_lock(&log_root_tree->log_mutex);
2842 atomic_inc(&log_root_tree->log_batch);
2843 atomic_inc(&log_root_tree->log_writers);
2845 index2 = log_root_tree->log_transid % 2;
2846 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2847 root_log_ctx.log_transid = log_root_tree->log_transid;
2849 mutex_unlock(&log_root_tree->log_mutex);
2851 ret = update_log_root(trans, log);
2853 mutex_lock(&log_root_tree->log_mutex);
2854 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2856 * Implicit memory barrier after atomic_dec_and_test
2858 if (waitqueue_active(&log_root_tree->log_writer_wait))
2859 wake_up(&log_root_tree->log_writer_wait);
2863 if (!list_empty(&root_log_ctx.list))
2864 list_del_init(&root_log_ctx.list);
2866 blk_finish_plug(&plug);
2867 btrfs_set_log_full_commit(fs_info, trans);
2869 if (ret != -ENOSPC) {
2870 btrfs_abort_transaction(trans, ret);
2871 mutex_unlock(&log_root_tree->log_mutex);
2874 btrfs_wait_tree_log_extents(log, mark);
2875 btrfs_free_logged_extents(log, log_transid);
2876 mutex_unlock(&log_root_tree->log_mutex);
2881 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2882 blk_finish_plug(&plug);
2883 list_del_init(&root_log_ctx.list);
2884 mutex_unlock(&log_root_tree->log_mutex);
2885 ret = root_log_ctx.log_ret;
2889 index2 = root_log_ctx.log_transid % 2;
2890 if (atomic_read(&log_root_tree->log_commit[index2])) {
2891 blk_finish_plug(&plug);
2892 ret = btrfs_wait_tree_log_extents(log, mark);
2893 btrfs_wait_logged_extents(trans, log, log_transid);
2894 wait_log_commit(log_root_tree,
2895 root_log_ctx.log_transid);
2896 mutex_unlock(&log_root_tree->log_mutex);
2898 ret = root_log_ctx.log_ret;
2901 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2902 atomic_set(&log_root_tree->log_commit[index2], 1);
2904 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2905 wait_log_commit(log_root_tree,
2906 root_log_ctx.log_transid - 1);
2909 wait_for_writer(log_root_tree);
2912 * now that we've moved on to the tree of log tree roots,
2913 * check the full commit flag again
2915 if (btrfs_need_log_full_commit(fs_info, trans)) {
2916 blk_finish_plug(&plug);
2917 btrfs_wait_tree_log_extents(log, mark);
2918 btrfs_free_logged_extents(log, log_transid);
2919 mutex_unlock(&log_root_tree->log_mutex);
2921 goto out_wake_log_root;
2924 ret = btrfs_write_marked_extents(fs_info,
2925 &log_root_tree->dirty_log_pages,
2926 EXTENT_DIRTY | EXTENT_NEW);
2927 blk_finish_plug(&plug);
2929 btrfs_set_log_full_commit(fs_info, trans);
2930 btrfs_abort_transaction(trans, ret);
2931 btrfs_free_logged_extents(log, log_transid);
2932 mutex_unlock(&log_root_tree->log_mutex);
2933 goto out_wake_log_root;
2935 ret = btrfs_wait_tree_log_extents(log, mark);
2937 ret = btrfs_wait_tree_log_extents(log_root_tree,
2938 EXTENT_NEW | EXTENT_DIRTY);
2940 btrfs_set_log_full_commit(fs_info, trans);
2941 btrfs_free_logged_extents(log, log_transid);
2942 mutex_unlock(&log_root_tree->log_mutex);
2943 goto out_wake_log_root;
2945 btrfs_wait_logged_extents(trans, log, log_transid);
2947 btrfs_set_super_log_root(fs_info->super_for_commit,
2948 log_root_tree->node->start);
2949 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2950 btrfs_header_level(log_root_tree->node));
2952 log_root_tree->log_transid++;
2953 mutex_unlock(&log_root_tree->log_mutex);
2956 * nobody else is going to jump in and write the the ctree
2957 * super here because the log_commit atomic below is protecting
2958 * us. We must be called with a transaction handle pinning
2959 * the running transaction open, so a full commit can't hop
2960 * in and cause problems either.
2962 ret = write_ctree_super(trans, fs_info, 1);
2964 btrfs_set_log_full_commit(fs_info, trans);
2965 btrfs_abort_transaction(trans, ret);
2966 goto out_wake_log_root;
2969 mutex_lock(&root->log_mutex);
2970 if (root->last_log_commit < log_transid)
2971 root->last_log_commit = log_transid;
2972 mutex_unlock(&root->log_mutex);
2975 mutex_lock(&log_root_tree->log_mutex);
2976 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2978 log_root_tree->log_transid_committed++;
2979 atomic_set(&log_root_tree->log_commit[index2], 0);
2980 mutex_unlock(&log_root_tree->log_mutex);
2983 * The barrier before waitqueue_active is implied by mutex_unlock
2985 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2986 wake_up(&log_root_tree->log_commit_wait[index2]);
2988 mutex_lock(&root->log_mutex);
2989 btrfs_remove_all_log_ctxs(root, index1, ret);
2990 root->log_transid_committed++;
2991 atomic_set(&root->log_commit[index1], 0);
2992 mutex_unlock(&root->log_mutex);
2995 * The barrier before waitqueue_active is implied by mutex_unlock
2997 if (waitqueue_active(&root->log_commit_wait[index1]))
2998 wake_up(&root->log_commit_wait[index1]);
3002 static void free_log_tree(struct btrfs_trans_handle *trans,
3003 struct btrfs_root *log)
3008 struct walk_control wc = {
3010 .process_func = process_one_buffer
3013 ret = walk_log_tree(trans, log, &wc);
3014 /* I don't think this can happen but just in case */
3016 btrfs_abort_transaction(trans, ret);
3019 ret = find_first_extent_bit(&log->dirty_log_pages,
3020 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3025 clear_extent_bits(&log->dirty_log_pages, start, end,
3026 EXTENT_DIRTY | EXTENT_NEW);
3030 * We may have short-circuited the log tree with the full commit logic
3031 * and left ordered extents on our list, so clear these out to keep us
3032 * from leaking inodes and memory.
3034 btrfs_free_logged_extents(log, 0);
3035 btrfs_free_logged_extents(log, 1);
3037 free_extent_buffer(log->node);
3042 * free all the extents used by the tree log. This should be called
3043 * at commit time of the full transaction
3045 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3047 if (root->log_root) {
3048 free_log_tree(trans, root->log_root);
3049 root->log_root = NULL;
3054 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3055 struct btrfs_fs_info *fs_info)
3057 if (fs_info->log_root_tree) {
3058 free_log_tree(trans, fs_info->log_root_tree);
3059 fs_info->log_root_tree = NULL;
3065 * If both a file and directory are logged, and unlinks or renames are
3066 * mixed in, we have a few interesting corners:
3068 * create file X in dir Y
3069 * link file X to X.link in dir Y
3071 * unlink file X but leave X.link
3074 * After a crash we would expect only X.link to exist. But file X
3075 * didn't get fsync'd again so the log has back refs for X and X.link.
3077 * We solve this by removing directory entries and inode backrefs from the
3078 * log when a file that was logged in the current transaction is
3079 * unlinked. Any later fsync will include the updated log entries, and
3080 * we'll be able to reconstruct the proper directory items from backrefs.
3082 * This optimizations allows us to avoid relogging the entire inode
3083 * or the entire directory.
3085 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3086 struct btrfs_root *root,
3087 const char *name, int name_len,
3088 struct btrfs_inode *dir, u64 index)
3090 struct btrfs_root *log;
3091 struct btrfs_dir_item *di;
3092 struct btrfs_path *path;
3096 u64 dir_ino = btrfs_ino(dir);
3098 if (dir->logged_trans < trans->transid)
3101 ret = join_running_log_trans(root);
3105 mutex_lock(&dir->log_mutex);
3107 log = root->log_root;
3108 path = btrfs_alloc_path();
3114 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3115 name, name_len, -1);
3121 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3122 bytes_del += name_len;
3128 btrfs_release_path(path);
3129 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3130 index, name, name_len, -1);
3136 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3137 bytes_del += name_len;
3144 /* update the directory size in the log to reflect the names
3148 struct btrfs_key key;
3150 key.objectid = dir_ino;
3152 key.type = BTRFS_INODE_ITEM_KEY;
3153 btrfs_release_path(path);
3155 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3161 struct btrfs_inode_item *item;
3164 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3165 struct btrfs_inode_item);
3166 i_size = btrfs_inode_size(path->nodes[0], item);
3167 if (i_size > bytes_del)
3168 i_size -= bytes_del;
3171 btrfs_set_inode_size(path->nodes[0], item, i_size);
3172 btrfs_mark_buffer_dirty(path->nodes[0]);
3175 btrfs_release_path(path);
3178 btrfs_free_path(path);
3180 mutex_unlock(&dir->log_mutex);
3181 if (ret == -ENOSPC) {
3182 btrfs_set_log_full_commit(root->fs_info, trans);
3185 btrfs_abort_transaction(trans, ret);
3187 btrfs_end_log_trans(root);
3192 /* see comments for btrfs_del_dir_entries_in_log */
3193 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3194 struct btrfs_root *root,
3195 const char *name, int name_len,
3196 struct btrfs_inode *inode, u64 dirid)
3198 struct btrfs_fs_info *fs_info = root->fs_info;
3199 struct btrfs_root *log;
3203 if (inode->logged_trans < trans->transid)
3206 ret = join_running_log_trans(root);
3209 log = root->log_root;
3210 mutex_lock(&inode->log_mutex);
3212 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3214 mutex_unlock(&inode->log_mutex);
3215 if (ret == -ENOSPC) {
3216 btrfs_set_log_full_commit(fs_info, trans);
3218 } else if (ret < 0 && ret != -ENOENT)
3219 btrfs_abort_transaction(trans, ret);
3220 btrfs_end_log_trans(root);
3226 * creates a range item in the log for 'dirid'. first_offset and
3227 * last_offset tell us which parts of the key space the log should
3228 * be considered authoritative for.
3230 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3231 struct btrfs_root *log,
3232 struct btrfs_path *path,
3233 int key_type, u64 dirid,
3234 u64 first_offset, u64 last_offset)
3237 struct btrfs_key key;
3238 struct btrfs_dir_log_item *item;
3240 key.objectid = dirid;
3241 key.offset = first_offset;
3242 if (key_type == BTRFS_DIR_ITEM_KEY)
3243 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3245 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3246 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3250 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3251 struct btrfs_dir_log_item);
3252 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3253 btrfs_mark_buffer_dirty(path->nodes[0]);
3254 btrfs_release_path(path);
3259 * log all the items included in the current transaction for a given
3260 * directory. This also creates the range items in the log tree required
3261 * to replay anything deleted before the fsync
3263 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3264 struct btrfs_root *root, struct btrfs_inode *inode,
3265 struct btrfs_path *path,
3266 struct btrfs_path *dst_path, int key_type,
3267 struct btrfs_log_ctx *ctx,
3268 u64 min_offset, u64 *last_offset_ret)
3270 struct btrfs_key min_key;
3271 struct btrfs_root *log = root->log_root;
3272 struct extent_buffer *src;
3277 u64 first_offset = min_offset;
3278 u64 last_offset = (u64)-1;
3279 u64 ino = btrfs_ino(inode);
3281 log = root->log_root;
3283 min_key.objectid = ino;
3284 min_key.type = key_type;
3285 min_key.offset = min_offset;
3287 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3290 * we didn't find anything from this transaction, see if there
3291 * is anything at all
3293 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3294 min_key.objectid = ino;
3295 min_key.type = key_type;
3296 min_key.offset = (u64)-1;
3297 btrfs_release_path(path);
3298 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3300 btrfs_release_path(path);
3303 ret = btrfs_previous_item(root, path, ino, key_type);
3305 /* if ret == 0 there are items for this type,
3306 * create a range to tell us the last key of this type.
3307 * otherwise, there are no items in this directory after
3308 * *min_offset, and we create a range to indicate that.
3311 struct btrfs_key tmp;
3312 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3314 if (key_type == tmp.type)
3315 first_offset = max(min_offset, tmp.offset) + 1;
3320 /* go backward to find any previous key */
3321 ret = btrfs_previous_item(root, path, ino, key_type);
3323 struct btrfs_key tmp;
3324 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3325 if (key_type == tmp.type) {
3326 first_offset = tmp.offset;
3327 ret = overwrite_item(trans, log, dst_path,
3328 path->nodes[0], path->slots[0],
3336 btrfs_release_path(path);
3338 /* find the first key from this transaction again */
3339 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3340 if (WARN_ON(ret != 0))
3344 * we have a block from this transaction, log every item in it
3345 * from our directory
3348 struct btrfs_key tmp;
3349 src = path->nodes[0];
3350 nritems = btrfs_header_nritems(src);
3351 for (i = path->slots[0]; i < nritems; i++) {
3352 struct btrfs_dir_item *di;
3354 btrfs_item_key_to_cpu(src, &min_key, i);
3356 if (min_key.objectid != ino || min_key.type != key_type)
3358 ret = overwrite_item(trans, log, dst_path, src, i,
3366 * We must make sure that when we log a directory entry,
3367 * the corresponding inode, after log replay, has a
3368 * matching link count. For example:
3374 * xfs_io -c "fsync" mydir
3376 * <mount fs and log replay>
3378 * Would result in a fsync log that when replayed, our
3379 * file inode would have a link count of 1, but we get
3380 * two directory entries pointing to the same inode.
3381 * After removing one of the names, it would not be
3382 * possible to remove the other name, which resulted
3383 * always in stale file handle errors, and would not
3384 * be possible to rmdir the parent directory, since
3385 * its i_size could never decrement to the value
3386 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3388 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3389 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3391 (btrfs_dir_transid(src, di) == trans->transid ||
3392 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3393 tmp.type != BTRFS_ROOT_ITEM_KEY)
3394 ctx->log_new_dentries = true;
3396 path->slots[0] = nritems;
3399 * look ahead to the next item and see if it is also
3400 * from this directory and from this transaction
3402 ret = btrfs_next_leaf(root, path);
3404 last_offset = (u64)-1;
3407 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3408 if (tmp.objectid != ino || tmp.type != key_type) {
3409 last_offset = (u64)-1;
3412 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3413 ret = overwrite_item(trans, log, dst_path,
3414 path->nodes[0], path->slots[0],
3419 last_offset = tmp.offset;
3424 btrfs_release_path(path);
3425 btrfs_release_path(dst_path);
3428 *last_offset_ret = last_offset;
3430 * insert the log range keys to indicate where the log
3433 ret = insert_dir_log_key(trans, log, path, key_type,
3434 ino, first_offset, last_offset);
3442 * logging directories is very similar to logging inodes, We find all the items
3443 * from the current transaction and write them to the log.
3445 * The recovery code scans the directory in the subvolume, and if it finds a
3446 * key in the range logged that is not present in the log tree, then it means
3447 * that dir entry was unlinked during the transaction.
3449 * In order for that scan to work, we must include one key smaller than
3450 * the smallest logged by this transaction and one key larger than the largest
3451 * key logged by this transaction.
3453 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3454 struct btrfs_root *root, struct btrfs_inode *inode,
3455 struct btrfs_path *path,
3456 struct btrfs_path *dst_path,
3457 struct btrfs_log_ctx *ctx)
3462 int key_type = BTRFS_DIR_ITEM_KEY;
3468 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3469 ctx, min_key, &max_key);
3472 if (max_key == (u64)-1)
3474 min_key = max_key + 1;
3477 if (key_type == BTRFS_DIR_ITEM_KEY) {
3478 key_type = BTRFS_DIR_INDEX_KEY;
3485 * a helper function to drop items from the log before we relog an
3486 * inode. max_key_type indicates the highest item type to remove.
3487 * This cannot be run for file data extents because it does not
3488 * free the extents they point to.
3490 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3491 struct btrfs_root *log,
3492 struct btrfs_path *path,
3493 u64 objectid, int max_key_type)
3496 struct btrfs_key key;
3497 struct btrfs_key found_key;
3500 key.objectid = objectid;
3501 key.type = max_key_type;
3502 key.offset = (u64)-1;
3505 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3506 BUG_ON(ret == 0); /* Logic error */
3510 if (path->slots[0] == 0)
3514 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3517 if (found_key.objectid != objectid)
3520 found_key.offset = 0;
3522 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3525 ret = btrfs_del_items(trans, log, path, start_slot,
3526 path->slots[0] - start_slot + 1);
3528 * If start slot isn't 0 then we don't need to re-search, we've
3529 * found the last guy with the objectid in this tree.
3531 if (ret || start_slot != 0)
3533 btrfs_release_path(path);
3535 btrfs_release_path(path);
3541 static void fill_inode_item(struct btrfs_trans_handle *trans,
3542 struct extent_buffer *leaf,
3543 struct btrfs_inode_item *item,
3544 struct inode *inode, int log_inode_only,
3547 struct btrfs_map_token token;
3549 btrfs_init_map_token(&token);
3551 if (log_inode_only) {
3552 /* set the generation to zero so the recover code
3553 * can tell the difference between an logging
3554 * just to say 'this inode exists' and a logging
3555 * to say 'update this inode with these values'
3557 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3558 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3560 btrfs_set_token_inode_generation(leaf, item,
3561 BTRFS_I(inode)->generation,
3563 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3566 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3567 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3568 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3569 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3571 btrfs_set_token_timespec_sec(leaf, &item->atime,
3572 inode->i_atime.tv_sec, &token);
3573 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3574 inode->i_atime.tv_nsec, &token);
3576 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3577 inode->i_mtime.tv_sec, &token);
3578 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3579 inode->i_mtime.tv_nsec, &token);
3581 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3582 inode->i_ctime.tv_sec, &token);
3583 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3584 inode->i_ctime.tv_nsec, &token);
3586 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3589 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3590 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3591 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3592 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3593 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3596 static int log_inode_item(struct btrfs_trans_handle *trans,
3597 struct btrfs_root *log, struct btrfs_path *path,
3598 struct inode *inode)
3600 struct btrfs_inode_item *inode_item;
3603 ret = btrfs_insert_empty_item(trans, log, path,
3604 &BTRFS_I(inode)->location,
3605 sizeof(*inode_item));
3606 if (ret && ret != -EEXIST)
3608 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3609 struct btrfs_inode_item);
3610 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3611 btrfs_release_path(path);
3615 static noinline int copy_items(struct btrfs_trans_handle *trans,
3616 struct btrfs_inode *inode,
3617 struct btrfs_path *dst_path,
3618 struct btrfs_path *src_path, u64 *last_extent,
3619 int start_slot, int nr, int inode_only,
3622 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3623 unsigned long src_offset;
3624 unsigned long dst_offset;
3625 struct btrfs_root *log = inode->root->log_root;
3626 struct btrfs_file_extent_item *extent;
3627 struct btrfs_inode_item *inode_item;
3628 struct extent_buffer *src = src_path->nodes[0];
3629 struct btrfs_key first_key, last_key, key;
3631 struct btrfs_key *ins_keys;
3635 struct list_head ordered_sums;
3636 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3637 bool has_extents = false;
3638 bool need_find_last_extent = true;
3641 INIT_LIST_HEAD(&ordered_sums);
3643 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3644 nr * sizeof(u32), GFP_NOFS);
3648 first_key.objectid = (u64)-1;
3650 ins_sizes = (u32 *)ins_data;
3651 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3653 for (i = 0; i < nr; i++) {
3654 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3655 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3657 ret = btrfs_insert_empty_items(trans, log, dst_path,
3658 ins_keys, ins_sizes, nr);
3664 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3665 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3666 dst_path->slots[0]);
3668 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3670 if ((i == (nr - 1)))
3671 last_key = ins_keys[i];
3673 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3674 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3676 struct btrfs_inode_item);
3677 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3678 &inode->vfs_inode, inode_only == LOG_INODE_EXISTS,
3681 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3682 src_offset, ins_sizes[i]);
3686 * We set need_find_last_extent here in case we know we were
3687 * processing other items and then walk into the first extent in
3688 * the inode. If we don't hit an extent then nothing changes,
3689 * we'll do the last search the next time around.
3691 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3693 if (first_key.objectid == (u64)-1)
3694 first_key = ins_keys[i];
3696 need_find_last_extent = false;
3699 /* take a reference on file data extents so that truncates
3700 * or deletes of this inode don't have to relog the inode
3703 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3706 extent = btrfs_item_ptr(src, start_slot + i,
3707 struct btrfs_file_extent_item);
3709 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3712 found_type = btrfs_file_extent_type(src, extent);
3713 if (found_type == BTRFS_FILE_EXTENT_REG) {
3715 ds = btrfs_file_extent_disk_bytenr(src,
3717 /* ds == 0 is a hole */
3721 dl = btrfs_file_extent_disk_num_bytes(src,
3723 cs = btrfs_file_extent_offset(src, extent);
3724 cl = btrfs_file_extent_num_bytes(src,
3726 if (btrfs_file_extent_compression(src,
3732 ret = btrfs_lookup_csums_range(
3734 ds + cs, ds + cs + cl - 1,
3737 btrfs_release_path(dst_path);
3745 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3746 btrfs_release_path(dst_path);
3750 * we have to do this after the loop above to avoid changing the
3751 * log tree while trying to change the log tree.
3754 while (!list_empty(&ordered_sums)) {
3755 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3756 struct btrfs_ordered_sum,
3759 ret = btrfs_csum_file_blocks(trans, log, sums);
3760 list_del(&sums->list);
3767 if (need_find_last_extent && *last_extent == first_key.offset) {
3769 * We don't have any leafs between our current one and the one
3770 * we processed before that can have file extent items for our
3771 * inode (and have a generation number smaller than our current
3774 need_find_last_extent = false;
3778 * Because we use btrfs_search_forward we could skip leaves that were
3779 * not modified and then assume *last_extent is valid when it really
3780 * isn't. So back up to the previous leaf and read the end of the last
3781 * extent before we go and fill in holes.
3783 if (need_find_last_extent) {
3786 ret = btrfs_prev_leaf(inode->root, src_path);
3791 if (src_path->slots[0])
3792 src_path->slots[0]--;
3793 src = src_path->nodes[0];
3794 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3795 if (key.objectid != btrfs_ino(inode) ||
3796 key.type != BTRFS_EXTENT_DATA_KEY)
3798 extent = btrfs_item_ptr(src, src_path->slots[0],
3799 struct btrfs_file_extent_item);
3800 if (btrfs_file_extent_type(src, extent) ==
3801 BTRFS_FILE_EXTENT_INLINE) {
3802 len = btrfs_file_extent_inline_len(src,
3805 *last_extent = ALIGN(key.offset + len,
3806 fs_info->sectorsize);
3808 len = btrfs_file_extent_num_bytes(src, extent);
3809 *last_extent = key.offset + len;
3813 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3814 * things could have happened
3816 * 1) A merge could have happened, so we could currently be on a leaf
3817 * that holds what we were copying in the first place.
3818 * 2) A split could have happened, and now not all of the items we want
3819 * are on the same leaf.
3821 * So we need to adjust how we search for holes, we need to drop the
3822 * path and re-search for the first extent key we found, and then walk
3823 * forward until we hit the last one we copied.
3825 if (need_find_last_extent) {
3826 /* btrfs_prev_leaf could return 1 without releasing the path */
3827 btrfs_release_path(src_path);
3828 ret = btrfs_search_slot(NULL, inode->root, &first_key, src_path, 0, 0);
3832 src = src_path->nodes[0];
3833 i = src_path->slots[0];
3839 * Ok so here we need to go through and fill in any holes we may have
3840 * to make sure that holes are punched for those areas in case they had
3841 * extents previously.
3847 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3848 ret = btrfs_next_leaf(inode->root, src_path);
3852 src = src_path->nodes[0];
3856 btrfs_item_key_to_cpu(src, &key, i);
3857 if (!btrfs_comp_cpu_keys(&key, &last_key))
3859 if (key.objectid != btrfs_ino(inode) ||
3860 key.type != BTRFS_EXTENT_DATA_KEY) {
3864 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3865 if (btrfs_file_extent_type(src, extent) ==
3866 BTRFS_FILE_EXTENT_INLINE) {
3867 len = btrfs_file_extent_inline_len(src, i, extent);
3868 extent_end = ALIGN(key.offset + len,
3869 fs_info->sectorsize);
3871 len = btrfs_file_extent_num_bytes(src, extent);
3872 extent_end = key.offset + len;
3876 if (*last_extent == key.offset) {
3877 *last_extent = extent_end;
3880 offset = *last_extent;
3881 len = key.offset - *last_extent;
3882 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3883 offset, 0, 0, len, 0, len, 0, 0, 0);
3886 *last_extent = extent_end;
3889 * Need to let the callers know we dropped the path so they should
3892 if (!ret && need_find_last_extent)
3897 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3899 struct extent_map *em1, *em2;
3901 em1 = list_entry(a, struct extent_map, list);
3902 em2 = list_entry(b, struct extent_map, list);
3904 if (em1->start < em2->start)
3906 else if (em1->start > em2->start)
3911 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3912 struct inode *inode,
3913 struct btrfs_root *root,
3914 const struct extent_map *em,
3915 const struct list_head *logged_list,
3916 bool *ordered_io_error)
3918 struct btrfs_fs_info *fs_info = root->fs_info;
3919 struct btrfs_ordered_extent *ordered;
3920 struct btrfs_root *log = root->log_root;
3921 u64 mod_start = em->mod_start;
3922 u64 mod_len = em->mod_len;
3923 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3926 LIST_HEAD(ordered_sums);
3929 *ordered_io_error = false;
3931 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3932 em->block_start == EXTENT_MAP_HOLE)
3936 * Wait far any ordered extent that covers our extent map. If it
3937 * finishes without an error, first check and see if our csums are on
3938 * our outstanding ordered extents.
3940 list_for_each_entry(ordered, logged_list, log_list) {
3941 struct btrfs_ordered_sum *sum;
3946 if (ordered->file_offset + ordered->len <= mod_start ||
3947 mod_start + mod_len <= ordered->file_offset)
3950 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3951 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3952 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3953 const u64 start = ordered->file_offset;
3954 const u64 end = ordered->file_offset + ordered->len - 1;
3956 WARN_ON(ordered->inode != inode);
3957 filemap_fdatawrite_range(inode->i_mapping, start, end);
3960 wait_event(ordered->wait,
3961 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3962 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3964 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3966 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3967 * i_mapping flags, so that the next fsync won't get
3968 * an outdated io error too.
3970 filemap_check_errors(inode->i_mapping);
3971 *ordered_io_error = true;
3975 * We are going to copy all the csums on this ordered extent, so
3976 * go ahead and adjust mod_start and mod_len in case this
3977 * ordered extent has already been logged.
3979 if (ordered->file_offset > mod_start) {
3980 if (ordered->file_offset + ordered->len >=
3981 mod_start + mod_len)
3982 mod_len = ordered->file_offset - mod_start;
3984 * If we have this case
3986 * |--------- logged extent ---------|
3987 * |----- ordered extent ----|
3989 * Just don't mess with mod_start and mod_len, we'll
3990 * just end up logging more csums than we need and it
3994 if (ordered->file_offset + ordered->len <
3995 mod_start + mod_len) {
3996 mod_len = (mod_start + mod_len) -
3997 (ordered->file_offset + ordered->len);
3998 mod_start = ordered->file_offset +
4009 * To keep us from looping for the above case of an ordered
4010 * extent that falls inside of the logged extent.
4012 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4016 list_for_each_entry(sum, &ordered->list, list) {
4017 ret = btrfs_csum_file_blocks(trans, log, sum);
4023 if (*ordered_io_error || !mod_len || ret || skip_csum)
4026 if (em->compress_type) {
4028 csum_len = max(em->block_len, em->orig_block_len);
4030 csum_offset = mod_start - em->start;
4034 /* block start is already adjusted for the file extent offset. */
4035 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4036 em->block_start + csum_offset,
4037 em->block_start + csum_offset +
4038 csum_len - 1, &ordered_sums, 0);
4042 while (!list_empty(&ordered_sums)) {
4043 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4044 struct btrfs_ordered_sum,
4047 ret = btrfs_csum_file_blocks(trans, log, sums);
4048 list_del(&sums->list);
4055 static int log_one_extent(struct btrfs_trans_handle *trans,
4056 struct btrfs_inode *inode, struct btrfs_root *root,
4057 const struct extent_map *em,
4058 struct btrfs_path *path,
4059 const struct list_head *logged_list,
4060 struct btrfs_log_ctx *ctx)
4062 struct btrfs_root *log = root->log_root;
4063 struct btrfs_file_extent_item *fi;
4064 struct extent_buffer *leaf;
4065 struct btrfs_map_token token;
4066 struct btrfs_key key;
4067 u64 extent_offset = em->start - em->orig_start;
4070 int extent_inserted = 0;
4071 bool ordered_io_err = false;
4073 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em, logged_list,
4078 if (ordered_io_err) {
4083 btrfs_init_map_token(&token);
4085 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4086 em->start + em->len, NULL, 0, 1,
4087 sizeof(*fi), &extent_inserted);
4091 if (!extent_inserted) {
4092 key.objectid = btrfs_ino(inode);
4093 key.type = BTRFS_EXTENT_DATA_KEY;
4094 key.offset = em->start;
4096 ret = btrfs_insert_empty_item(trans, log, path, &key,
4101 leaf = path->nodes[0];
4102 fi = btrfs_item_ptr(leaf, path->slots[0],
4103 struct btrfs_file_extent_item);
4105 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4107 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4108 btrfs_set_token_file_extent_type(leaf, fi,
4109 BTRFS_FILE_EXTENT_PREALLOC,
4112 btrfs_set_token_file_extent_type(leaf, fi,
4113 BTRFS_FILE_EXTENT_REG,
4116 block_len = max(em->block_len, em->orig_block_len);
4117 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4118 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4121 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4123 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4124 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4126 extent_offset, &token);
4127 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4130 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4131 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4135 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4136 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4137 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4138 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4140 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4141 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4142 btrfs_mark_buffer_dirty(leaf);
4144 btrfs_release_path(path);
4149 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4150 struct btrfs_root *root,
4151 struct btrfs_inode *inode,
4152 struct btrfs_path *path,
4153 struct list_head *logged_list,
4154 struct btrfs_log_ctx *ctx,
4158 struct extent_map *em, *n;
4159 struct list_head extents;
4160 struct extent_map_tree *tree = &inode->extent_tree;
4165 INIT_LIST_HEAD(&extents);
4167 down_write(&inode->dio_sem);
4168 write_lock(&tree->lock);
4169 test_gen = root->fs_info->last_trans_committed;
4171 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4172 list_del_init(&em->list);
4175 * Just an arbitrary number, this can be really CPU intensive
4176 * once we start getting a lot of extents, and really once we
4177 * have a bunch of extents we just want to commit since it will
4180 if (++num > 32768) {
4181 list_del_init(&tree->modified_extents);
4186 if (em->generation <= test_gen)
4188 /* Need a ref to keep it from getting evicted from cache */
4189 atomic_inc(&em->refs);
4190 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4191 list_add_tail(&em->list, &extents);
4195 list_sort(NULL, &extents, extent_cmp);
4196 btrfs_get_logged_extents(inode, logged_list, start, end);
4198 * Some ordered extents started by fsync might have completed
4199 * before we could collect them into the list logged_list, which
4200 * means they're gone, not in our logged_list nor in the inode's
4201 * ordered tree. We want the application/user space to know an
4202 * error happened while attempting to persist file data so that
4203 * it can take proper action. If such error happened, we leave
4204 * without writing to the log tree and the fsync must report the
4205 * file data write error and not commit the current transaction.
4207 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4211 while (!list_empty(&extents)) {
4212 em = list_entry(extents.next, struct extent_map, list);
4214 list_del_init(&em->list);
4217 * If we had an error we just need to delete everybody from our
4221 clear_em_logging(tree, em);
4222 free_extent_map(em);
4226 write_unlock(&tree->lock);
4228 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4230 write_lock(&tree->lock);
4231 clear_em_logging(tree, em);
4232 free_extent_map(em);
4234 WARN_ON(!list_empty(&extents));
4235 write_unlock(&tree->lock);
4236 up_write(&inode->dio_sem);
4238 btrfs_release_path(path);
4242 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4243 struct btrfs_path *path, u64 *size_ret)
4245 struct btrfs_key key;
4248 key.objectid = btrfs_ino(inode);
4249 key.type = BTRFS_INODE_ITEM_KEY;
4252 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4255 } else if (ret > 0) {
4258 struct btrfs_inode_item *item;
4260 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4261 struct btrfs_inode_item);
4262 *size_ret = btrfs_inode_size(path->nodes[0], item);
4265 btrfs_release_path(path);
4270 * At the moment we always log all xattrs. This is to figure out at log replay
4271 * time which xattrs must have their deletion replayed. If a xattr is missing
4272 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4273 * because if a xattr is deleted, the inode is fsynced and a power failure
4274 * happens, causing the log to be replayed the next time the fs is mounted,
4275 * we want the xattr to not exist anymore (same behaviour as other filesystems
4276 * with a journal, ext3/4, xfs, f2fs, etc).
4278 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *root,
4280 struct btrfs_inode *inode,
4281 struct btrfs_path *path,
4282 struct btrfs_path *dst_path)
4285 struct btrfs_key key;
4286 const u64 ino = btrfs_ino(inode);
4291 key.type = BTRFS_XATTR_ITEM_KEY;
4294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4299 int slot = path->slots[0];
4300 struct extent_buffer *leaf = path->nodes[0];
4301 int nritems = btrfs_header_nritems(leaf);
4303 if (slot >= nritems) {
4305 u64 last_extent = 0;
4307 ret = copy_items(trans, inode, dst_path, path,
4308 &last_extent, start_slot,
4310 /* can't be 1, extent items aren't processed */
4316 ret = btrfs_next_leaf(root, path);
4324 btrfs_item_key_to_cpu(leaf, &key, slot);
4325 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4335 u64 last_extent = 0;
4337 ret = copy_items(trans, inode, dst_path, path,
4338 &last_extent, start_slot,
4340 /* can't be 1, extent items aren't processed */
4350 * If the no holes feature is enabled we need to make sure any hole between the
4351 * last extent and the i_size of our inode is explicitly marked in the log. This
4352 * is to make sure that doing something like:
4354 * 1) create file with 128Kb of data
4355 * 2) truncate file to 64Kb
4356 * 3) truncate file to 256Kb
4358 * 5) <crash/power failure>
4359 * 6) mount fs and trigger log replay
4361 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4362 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4363 * file correspond to a hole. The presence of explicit holes in a log tree is
4364 * what guarantees that log replay will remove/adjust file extent items in the
4367 * Here we do not need to care about holes between extents, that is already done
4368 * by copy_items(). We also only need to do this in the full sync path, where we
4369 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4370 * lookup the list of modified extent maps and if any represents a hole, we
4371 * insert a corresponding extent representing a hole in the log tree.
4373 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4374 struct btrfs_root *root,
4375 struct btrfs_inode *inode,
4376 struct btrfs_path *path)
4378 struct btrfs_fs_info *fs_info = root->fs_info;
4380 struct btrfs_key key;
4383 struct extent_buffer *leaf;
4384 struct btrfs_root *log = root->log_root;
4385 const u64 ino = btrfs_ino(inode);
4386 const u64 i_size = i_size_read(&inode->vfs_inode);
4388 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4392 key.type = BTRFS_EXTENT_DATA_KEY;
4393 key.offset = (u64)-1;
4395 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4400 ASSERT(path->slots[0] > 0);
4402 leaf = path->nodes[0];
4403 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4405 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4406 /* inode does not have any extents */
4410 struct btrfs_file_extent_item *extent;
4414 * If there's an extent beyond i_size, an explicit hole was
4415 * already inserted by copy_items().
4417 if (key.offset >= i_size)
4420 extent = btrfs_item_ptr(leaf, path->slots[0],
4421 struct btrfs_file_extent_item);
4423 if (btrfs_file_extent_type(leaf, extent) ==
4424 BTRFS_FILE_EXTENT_INLINE) {
4425 len = btrfs_file_extent_inline_len(leaf,
4428 ASSERT(len == i_size);
4432 len = btrfs_file_extent_num_bytes(leaf, extent);
4433 /* Last extent goes beyond i_size, no need to log a hole. */
4434 if (key.offset + len > i_size)
4436 hole_start = key.offset + len;
4437 hole_size = i_size - hole_start;
4439 btrfs_release_path(path);
4441 /* Last extent ends at i_size. */
4445 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4446 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4447 hole_size, 0, hole_size, 0, 0, 0);
4452 * When we are logging a new inode X, check if it doesn't have a reference that
4453 * matches the reference from some other inode Y created in a past transaction
4454 * and that was renamed in the current transaction. If we don't do this, then at
4455 * log replay time we can lose inode Y (and all its files if it's a directory):
4458 * echo "hello world" > /mnt/x/foobar
4461 * mkdir /mnt/x # or touch /mnt/x
4462 * xfs_io -c fsync /mnt/x
4464 * mount fs, trigger log replay
4466 * After the log replay procedure, we would lose the first directory and all its
4467 * files (file foobar).
4468 * For the case where inode Y is not a directory we simply end up losing it:
4470 * echo "123" > /mnt/foo
4472 * mv /mnt/foo /mnt/bar
4473 * echo "abc" > /mnt/foo
4474 * xfs_io -c fsync /mnt/foo
4477 * We also need this for cases where a snapshot entry is replaced by some other
4478 * entry (file or directory) otherwise we end up with an unreplayable log due to
4479 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4480 * if it were a regular entry:
4483 * btrfs subvolume snapshot /mnt /mnt/x/snap
4484 * btrfs subvolume delete /mnt/x/snap
4487 * fsync /mnt/x or fsync some new file inside it
4490 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4491 * the same transaction.
4493 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4495 const struct btrfs_key *key,
4496 struct btrfs_inode *inode,
4500 struct btrfs_path *search_path;
4503 u32 item_size = btrfs_item_size_nr(eb, slot);
4505 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4507 search_path = btrfs_alloc_path();
4510 search_path->search_commit_root = 1;
4511 search_path->skip_locking = 1;
4513 while (cur_offset < item_size) {
4517 unsigned long name_ptr;
4518 struct btrfs_dir_item *di;
4520 if (key->type == BTRFS_INODE_REF_KEY) {
4521 struct btrfs_inode_ref *iref;
4523 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4524 parent = key->offset;
4525 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4526 name_ptr = (unsigned long)(iref + 1);
4527 this_len = sizeof(*iref) + this_name_len;
4529 struct btrfs_inode_extref *extref;
4531 extref = (struct btrfs_inode_extref *)(ptr +
4533 parent = btrfs_inode_extref_parent(eb, extref);
4534 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4535 name_ptr = (unsigned long)&extref->name;
4536 this_len = sizeof(*extref) + this_name_len;
4539 if (this_name_len > name_len) {
4542 new_name = krealloc(name, this_name_len, GFP_NOFS);
4547 name_len = this_name_len;
4551 read_extent_buffer(eb, name, name_ptr, this_name_len);
4552 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4553 parent, name, this_name_len, 0);
4554 if (di && !IS_ERR(di)) {
4555 struct btrfs_key di_key;
4557 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4559 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4561 *other_ino = di_key.objectid;
4566 } else if (IS_ERR(di)) {
4570 btrfs_release_path(search_path);
4572 cur_offset += this_len;
4576 btrfs_free_path(search_path);
4581 /* log a single inode in the tree log.
4582 * At least one parent directory for this inode must exist in the tree
4583 * or be logged already.
4585 * Any items from this inode changed by the current transaction are copied
4586 * to the log tree. An extra reference is taken on any extents in this
4587 * file, allowing us to avoid a whole pile of corner cases around logging
4588 * blocks that have been removed from the tree.
4590 * See LOG_INODE_ALL and related defines for a description of what inode_only
4593 * This handles both files and directories.
4595 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4596 struct btrfs_root *root, struct inode *inode,
4600 struct btrfs_log_ctx *ctx)
4602 struct btrfs_fs_info *fs_info = root->fs_info;
4603 struct btrfs_path *path;
4604 struct btrfs_path *dst_path;
4605 struct btrfs_key min_key;
4606 struct btrfs_key max_key;
4607 struct btrfs_root *log = root->log_root;
4608 struct extent_buffer *src = NULL;
4609 LIST_HEAD(logged_list);
4610 u64 last_extent = 0;
4614 int ins_start_slot = 0;
4616 bool fast_search = false;
4617 u64 ino = btrfs_ino(BTRFS_I(inode));
4618 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4619 u64 logged_isize = 0;
4620 bool need_log_inode_item = true;
4622 path = btrfs_alloc_path();
4625 dst_path = btrfs_alloc_path();
4627 btrfs_free_path(path);
4631 min_key.objectid = ino;
4632 min_key.type = BTRFS_INODE_ITEM_KEY;
4635 max_key.objectid = ino;
4638 /* today the code can only do partial logging of directories */
4639 if (S_ISDIR(inode->i_mode) ||
4640 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4641 &BTRFS_I(inode)->runtime_flags) &&
4642 inode_only >= LOG_INODE_EXISTS))
4643 max_key.type = BTRFS_XATTR_ITEM_KEY;
4645 max_key.type = (u8)-1;
4646 max_key.offset = (u64)-1;
4649 * Only run delayed items if we are a dir or a new file.
4650 * Otherwise commit the delayed inode only, which is needed in
4651 * order for the log replay code to mark inodes for link count
4652 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4654 if (S_ISDIR(inode->i_mode) ||
4655 BTRFS_I(inode)->generation > fs_info->last_trans_committed)
4656 ret = btrfs_commit_inode_delayed_items(trans, BTRFS_I(inode));
4658 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
4661 btrfs_free_path(path);
4662 btrfs_free_path(dst_path);
4666 if (inode_only == LOG_OTHER_INODE) {
4667 inode_only = LOG_INODE_EXISTS;
4668 mutex_lock_nested(&BTRFS_I(inode)->log_mutex,
4669 SINGLE_DEPTH_NESTING);
4671 mutex_lock(&BTRFS_I(inode)->log_mutex);
4675 * a brute force approach to making sure we get the most uptodate
4676 * copies of everything.
4678 if (S_ISDIR(inode->i_mode)) {
4679 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4681 if (inode_only == LOG_INODE_EXISTS)
4682 max_key_type = BTRFS_XATTR_ITEM_KEY;
4683 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4685 if (inode_only == LOG_INODE_EXISTS) {
4687 * Make sure the new inode item we write to the log has
4688 * the same isize as the current one (if it exists).
4689 * This is necessary to prevent data loss after log
4690 * replay, and also to prevent doing a wrong expanding
4691 * truncate - for e.g. create file, write 4K into offset
4692 * 0, fsync, write 4K into offset 4096, add hard link,
4693 * fsync some other file (to sync log), power fail - if
4694 * we use the inode's current i_size, after log replay
4695 * we get a 8Kb file, with the last 4Kb extent as a hole
4696 * (zeroes), as if an expanding truncate happened,
4697 * instead of getting a file of 4Kb only.
4699 err = logged_inode_size(log, BTRFS_I(inode), path,
4704 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4705 &BTRFS_I(inode)->runtime_flags)) {
4706 if (inode_only == LOG_INODE_EXISTS) {
4707 max_key.type = BTRFS_XATTR_ITEM_KEY;
4708 ret = drop_objectid_items(trans, log, path, ino,
4711 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4712 &BTRFS_I(inode)->runtime_flags);
4713 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4714 &BTRFS_I(inode)->runtime_flags);
4716 ret = btrfs_truncate_inode_items(trans,
4722 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4723 &BTRFS_I(inode)->runtime_flags) ||
4724 inode_only == LOG_INODE_EXISTS) {
4725 if (inode_only == LOG_INODE_ALL)
4727 max_key.type = BTRFS_XATTR_ITEM_KEY;
4728 ret = drop_objectid_items(trans, log, path, ino,
4731 if (inode_only == LOG_INODE_ALL)
4744 ret = btrfs_search_forward(root, &min_key,
4745 path, trans->transid);
4753 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4754 if (min_key.objectid != ino)
4756 if (min_key.type > max_key.type)
4759 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4760 need_log_inode_item = false;
4762 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4763 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4764 BTRFS_I(inode)->generation == trans->transid) {
4767 ret = btrfs_check_ref_name_override(path->nodes[0],
4769 &min_key, BTRFS_I(inode),
4774 } else if (ret > 0 && ctx &&
4775 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4776 struct btrfs_key inode_key;
4777 struct inode *other_inode;
4783 ins_start_slot = path->slots[0];
4785 ret = copy_items(trans, BTRFS_I(inode), dst_path, path,
4786 &last_extent, ins_start_slot,
4794 btrfs_release_path(path);
4795 inode_key.objectid = other_ino;
4796 inode_key.type = BTRFS_INODE_ITEM_KEY;
4797 inode_key.offset = 0;
4798 other_inode = btrfs_iget(fs_info->sb,
4802 * If the other inode that had a conflicting dir
4803 * entry was deleted in the current transaction,
4804 * we don't need to do more work nor fallback to
4805 * a transaction commit.
4807 if (IS_ERR(other_inode) &&
4808 PTR_ERR(other_inode) == -ENOENT) {
4810 } else if (IS_ERR(other_inode)) {
4811 err = PTR_ERR(other_inode);
4815 * We are safe logging the other inode without
4816 * acquiring its i_mutex as long as we log with
4817 * the LOG_INODE_EXISTS mode. We're safe against
4818 * concurrent renames of the other inode as well
4819 * because during a rename we pin the log and
4820 * update the log with the new name before we
4823 err = btrfs_log_inode(trans, root, other_inode,
4834 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4835 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4838 ret = copy_items(trans, BTRFS_I(inode), dst_path, path,
4839 &last_extent, ins_start_slot,
4840 ins_nr, inode_only, logged_isize);
4847 btrfs_release_path(path);
4853 src = path->nodes[0];
4854 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4857 } else if (!ins_nr) {
4858 ins_start_slot = path->slots[0];
4863 ret = copy_items(trans, BTRFS_I(inode), dst_path, path, &last_extent,
4864 ins_start_slot, ins_nr, inode_only,
4872 btrfs_release_path(path);
4876 ins_start_slot = path->slots[0];
4879 nritems = btrfs_header_nritems(path->nodes[0]);
4881 if (path->slots[0] < nritems) {
4882 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4887 ret = copy_items(trans, BTRFS_I(inode), dst_path, path,
4888 &last_extent, ins_start_slot,
4889 ins_nr, inode_only, logged_isize);
4897 btrfs_release_path(path);
4899 if (min_key.offset < (u64)-1) {
4901 } else if (min_key.type < max_key.type) {
4909 ret = copy_items(trans, BTRFS_I(inode), dst_path, path, &last_extent,
4910 ins_start_slot, ins_nr, inode_only,
4920 btrfs_release_path(path);
4921 btrfs_release_path(dst_path);
4922 err = btrfs_log_all_xattrs(trans, root, BTRFS_I(inode), path, dst_path);
4925 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4926 btrfs_release_path(path);
4927 btrfs_release_path(dst_path);
4928 err = btrfs_log_trailing_hole(trans, root, BTRFS_I(inode), path);
4933 btrfs_release_path(path);
4934 btrfs_release_path(dst_path);
4935 if (need_log_inode_item) {
4936 err = log_inode_item(trans, log, dst_path, inode);
4941 ret = btrfs_log_changed_extents(trans, root, BTRFS_I(inode), dst_path,
4942 &logged_list, ctx, start, end);
4947 } else if (inode_only == LOG_INODE_ALL) {
4948 struct extent_map *em, *n;
4950 write_lock(&em_tree->lock);
4952 * We can't just remove every em if we're called for a ranged
4953 * fsync - that is, one that doesn't cover the whole possible
4954 * file range (0 to LLONG_MAX). This is because we can have
4955 * em's that fall outside the range we're logging and therefore
4956 * their ordered operations haven't completed yet
4957 * (btrfs_finish_ordered_io() not invoked yet). This means we
4958 * didn't get their respective file extent item in the fs/subvol
4959 * tree yet, and need to let the next fast fsync (one which
4960 * consults the list of modified extent maps) find the em so
4961 * that it logs a matching file extent item and waits for the
4962 * respective ordered operation to complete (if it's still
4965 * Removing every em outside the range we're logging would make
4966 * the next fast fsync not log their matching file extent items,
4967 * therefore making us lose data after a log replay.
4969 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4971 const u64 mod_end = em->mod_start + em->mod_len - 1;
4973 if (em->mod_start >= start && mod_end <= end)
4974 list_del_init(&em->list);
4976 write_unlock(&em_tree->lock);
4979 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4980 ret = log_directory_changes(trans, root, BTRFS_I(inode), path,
4988 spin_lock(&BTRFS_I(inode)->lock);
4989 BTRFS_I(inode)->logged_trans = trans->transid;
4990 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4991 spin_unlock(&BTRFS_I(inode)->lock);
4994 btrfs_put_logged_extents(&logged_list);
4996 btrfs_submit_logged_extents(&logged_list, log);
4997 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4999 btrfs_free_path(path);
5000 btrfs_free_path(dst_path);
5005 * Check if we must fallback to a transaction commit when logging an inode.
5006 * This must be called after logging the inode and is used only in the context
5007 * when fsyncing an inode requires the need to log some other inode - in which
5008 * case we can't lock the i_mutex of each other inode we need to log as that
5009 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5010 * log inodes up or down in the hierarchy) or rename operations for example. So
5011 * we take the log_mutex of the inode after we have logged it and then check for
5012 * its last_unlink_trans value - this is safe because any task setting
5013 * last_unlink_trans must take the log_mutex and it must do this before it does
5014 * the actual unlink operation, so if we do this check before a concurrent task
5015 * sets last_unlink_trans it means we've logged a consistent version/state of
5016 * all the inode items, otherwise we are not sure and must do a transaction
5017 * commit (the concurrent task might have only updated last_unlink_trans before
5018 * we logged the inode or it might have also done the unlink).
5020 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5021 struct btrfs_inode *inode)
5023 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5026 mutex_lock(&inode->log_mutex);
5027 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5029 * Make sure any commits to the log are forced to be full
5032 btrfs_set_log_full_commit(fs_info, trans);
5035 mutex_unlock(&inode->log_mutex);
5041 * follow the dentry parent pointers up the chain and see if any
5042 * of the directories in it require a full commit before they can
5043 * be logged. Returns zero if nothing special needs to be done or 1 if
5044 * a full commit is required.
5046 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5047 struct inode *inode,
5048 struct dentry *parent,
5049 struct super_block *sb,
5053 struct dentry *old_parent = NULL;
5054 struct inode *orig_inode = inode;
5057 * for regular files, if its inode is already on disk, we don't
5058 * have to worry about the parents at all. This is because
5059 * we can use the last_unlink_trans field to record renames
5060 * and other fun in this file.
5062 if (S_ISREG(inode->i_mode) &&
5063 BTRFS_I(inode)->generation <= last_committed &&
5064 BTRFS_I(inode)->last_unlink_trans <= last_committed)
5067 if (!S_ISDIR(inode->i_mode)) {
5068 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5070 inode = d_inode(parent);
5075 * If we are logging a directory then we start with our inode,
5076 * not our parent's inode, so we need to skip setting the
5077 * logged_trans so that further down in the log code we don't
5078 * think this inode has already been logged.
5080 if (inode != orig_inode)
5081 BTRFS_I(inode)->logged_trans = trans->transid;
5084 if (btrfs_must_commit_transaction(trans, BTRFS_I(inode))) {
5089 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5092 if (IS_ROOT(parent)) {
5093 inode = d_inode(parent);
5094 if (btrfs_must_commit_transaction(trans, BTRFS_I(inode)))
5099 parent = dget_parent(parent);
5101 old_parent = parent;
5102 inode = d_inode(parent);
5110 struct btrfs_dir_list {
5112 struct list_head list;
5116 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5117 * details about the why it is needed.
5118 * This is a recursive operation - if an existing dentry corresponds to a
5119 * directory, that directory's new entries are logged too (same behaviour as
5120 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5121 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5122 * complains about the following circular lock dependency / possible deadlock:
5126 * lock(&type->i_mutex_dir_key#3/2);
5127 * lock(sb_internal#2);
5128 * lock(&type->i_mutex_dir_key#3/2);
5129 * lock(&sb->s_type->i_mutex_key#14);
5131 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5132 * sb_start_intwrite() in btrfs_start_transaction().
5133 * Not locking i_mutex of the inodes is still safe because:
5135 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5136 * that while logging the inode new references (names) are added or removed
5137 * from the inode, leaving the logged inode item with a link count that does
5138 * not match the number of logged inode reference items. This is fine because
5139 * at log replay time we compute the real number of links and correct the
5140 * link count in the inode item (see replay_one_buffer() and
5141 * link_to_fixup_dir());
5143 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5144 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5145 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5146 * has a size that doesn't match the sum of the lengths of all the logged
5147 * names. This does not result in a problem because if a dir_item key is
5148 * logged but its matching dir_index key is not logged, at log replay time we
5149 * don't use it to replay the respective name (see replay_one_name()). On the
5150 * other hand if only the dir_index key ends up being logged, the respective
5151 * name is added to the fs/subvol tree with both the dir_item and dir_index
5152 * keys created (see replay_one_name()).
5153 * The directory's inode item with a wrong i_size is not a problem as well,
5154 * since we don't use it at log replay time to set the i_size in the inode
5155 * item of the fs/subvol tree (see overwrite_item()).
5157 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5158 struct btrfs_root *root,
5159 struct btrfs_inode *start_inode,
5160 struct btrfs_log_ctx *ctx)
5162 struct btrfs_fs_info *fs_info = root->fs_info;
5163 struct btrfs_root *log = root->log_root;
5164 struct btrfs_path *path;
5165 LIST_HEAD(dir_list);
5166 struct btrfs_dir_list *dir_elem;
5169 path = btrfs_alloc_path();
5173 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5175 btrfs_free_path(path);
5178 dir_elem->ino = btrfs_ino(start_inode);
5179 list_add_tail(&dir_elem->list, &dir_list);
5181 while (!list_empty(&dir_list)) {
5182 struct extent_buffer *leaf;
5183 struct btrfs_key min_key;
5187 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5190 goto next_dir_inode;
5192 min_key.objectid = dir_elem->ino;
5193 min_key.type = BTRFS_DIR_ITEM_KEY;
5196 btrfs_release_path(path);
5197 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5199 goto next_dir_inode;
5200 } else if (ret > 0) {
5202 goto next_dir_inode;
5206 leaf = path->nodes[0];
5207 nritems = btrfs_header_nritems(leaf);
5208 for (i = path->slots[0]; i < nritems; i++) {
5209 struct btrfs_dir_item *di;
5210 struct btrfs_key di_key;
5211 struct inode *di_inode;
5212 struct btrfs_dir_list *new_dir_elem;
5213 int log_mode = LOG_INODE_EXISTS;
5216 btrfs_item_key_to_cpu(leaf, &min_key, i);
5217 if (min_key.objectid != dir_elem->ino ||
5218 min_key.type != BTRFS_DIR_ITEM_KEY)
5219 goto next_dir_inode;
5221 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5222 type = btrfs_dir_type(leaf, di);
5223 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5224 type != BTRFS_FT_DIR)
5226 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5227 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5230 btrfs_release_path(path);
5231 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5232 if (IS_ERR(di_inode)) {
5233 ret = PTR_ERR(di_inode);
5234 goto next_dir_inode;
5237 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5242 ctx->log_new_dentries = false;
5243 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5244 log_mode = LOG_INODE_ALL;
5245 ret = btrfs_log_inode(trans, root, di_inode,
5246 log_mode, 0, LLONG_MAX, ctx);
5248 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5252 goto next_dir_inode;
5253 if (ctx->log_new_dentries) {
5254 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5256 if (!new_dir_elem) {
5258 goto next_dir_inode;
5260 new_dir_elem->ino = di_key.objectid;
5261 list_add_tail(&new_dir_elem->list, &dir_list);
5266 ret = btrfs_next_leaf(log, path);
5268 goto next_dir_inode;
5269 } else if (ret > 0) {
5271 goto next_dir_inode;
5275 if (min_key.offset < (u64)-1) {
5280 list_del(&dir_elem->list);
5284 btrfs_free_path(path);
5288 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5289 struct inode *inode,
5290 struct btrfs_log_ctx *ctx)
5292 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5294 struct btrfs_path *path;
5295 struct btrfs_key key;
5296 struct btrfs_root *root = BTRFS_I(inode)->root;
5297 const u64 ino = btrfs_ino(BTRFS_I(inode));
5299 path = btrfs_alloc_path();
5302 path->skip_locking = 1;
5303 path->search_commit_root = 1;
5306 key.type = BTRFS_INODE_REF_KEY;
5308 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5313 struct extent_buffer *leaf = path->nodes[0];
5314 int slot = path->slots[0];
5319 if (slot >= btrfs_header_nritems(leaf)) {
5320 ret = btrfs_next_leaf(root, path);
5328 btrfs_item_key_to_cpu(leaf, &key, slot);
5329 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5330 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5333 item_size = btrfs_item_size_nr(leaf, slot);
5334 ptr = btrfs_item_ptr_offset(leaf, slot);
5335 while (cur_offset < item_size) {
5336 struct btrfs_key inode_key;
5337 struct inode *dir_inode;
5339 inode_key.type = BTRFS_INODE_ITEM_KEY;
5340 inode_key.offset = 0;
5342 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5343 struct btrfs_inode_extref *extref;
5345 extref = (struct btrfs_inode_extref *)
5347 inode_key.objectid = btrfs_inode_extref_parent(
5349 cur_offset += sizeof(*extref);
5350 cur_offset += btrfs_inode_extref_name_len(leaf,
5353 inode_key.objectid = key.offset;
5354 cur_offset = item_size;
5357 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5359 /* If parent inode was deleted, skip it. */
5360 if (IS_ERR(dir_inode))
5364 ctx->log_new_dentries = false;
5365 ret = btrfs_log_inode(trans, root, dir_inode,
5366 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5368 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5370 if (!ret && ctx && ctx->log_new_dentries)
5371 ret = log_new_dir_dentries(trans, root,
5372 BTRFS_I(dir_inode), ctx);
5381 btrfs_free_path(path);
5386 * helper function around btrfs_log_inode to make sure newly created
5387 * parent directories also end up in the log. A minimal inode and backref
5388 * only logging is done of any parent directories that are older than
5389 * the last committed transaction
5391 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5392 struct btrfs_root *root, struct inode *inode,
5393 struct dentry *parent,
5397 struct btrfs_log_ctx *ctx)
5399 struct btrfs_fs_info *fs_info = root->fs_info;
5400 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5401 struct super_block *sb;
5402 struct dentry *old_parent = NULL;
5404 u64 last_committed = fs_info->last_trans_committed;
5405 bool log_dentries = false;
5406 struct inode *orig_inode = inode;
5410 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5416 * The prev transaction commit doesn't complete, we need do
5417 * full commit by ourselves.
5419 if (fs_info->last_trans_log_full_commit >
5420 fs_info->last_trans_committed) {
5425 if (root != BTRFS_I(inode)->root ||
5426 btrfs_root_refs(&root->root_item) == 0) {
5431 ret = check_parent_dirs_for_sync(trans, inode, parent,
5432 sb, last_committed);
5436 if (btrfs_inode_in_log(BTRFS_I(inode), trans->transid)) {
5437 ret = BTRFS_NO_LOG_SYNC;
5441 ret = start_log_trans(trans, root, ctx);
5445 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5450 * for regular files, if its inode is already on disk, we don't
5451 * have to worry about the parents at all. This is because
5452 * we can use the last_unlink_trans field to record renames
5453 * and other fun in this file.
5455 if (S_ISREG(inode->i_mode) &&
5456 BTRFS_I(inode)->generation <= last_committed &&
5457 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5462 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5463 log_dentries = true;
5466 * On unlink we must make sure all our current and old parent directory
5467 * inodes are fully logged. This is to prevent leaving dangling
5468 * directory index entries in directories that were our parents but are
5469 * not anymore. Not doing this results in old parent directory being
5470 * impossible to delete after log replay (rmdir will always fail with
5471 * error -ENOTEMPTY).
5477 * ln testdir/foo testdir/bar
5479 * unlink testdir/bar
5480 * xfs_io -c fsync testdir/foo
5482 * mount fs, triggers log replay
5484 * If we don't log the parent directory (testdir), after log replay the
5485 * directory still has an entry pointing to the file inode using the bar
5486 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5487 * the file inode has a link count of 1.
5493 * ln foo testdir/foo2
5494 * ln foo testdir/foo3
5496 * unlink testdir/foo3
5497 * xfs_io -c fsync foo
5499 * mount fs, triggers log replay
5501 * Similar as the first example, after log replay the parent directory
5502 * testdir still has an entry pointing to the inode file with name foo3
5503 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5504 * and has a link count of 2.
5506 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5507 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5513 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5516 inode = d_inode(parent);
5517 if (root != BTRFS_I(inode)->root)
5520 if (BTRFS_I(inode)->generation > last_committed) {
5521 ret = btrfs_log_inode(trans, root, inode,
5527 if (IS_ROOT(parent))
5530 parent = dget_parent(parent);
5532 old_parent = parent;
5535 ret = log_new_dir_dentries(trans, root, BTRFS_I(orig_inode), ctx);
5541 btrfs_set_log_full_commit(fs_info, trans);
5546 btrfs_remove_log_ctx(root, ctx);
5547 btrfs_end_log_trans(root);
5553 * it is not safe to log dentry if the chunk root has added new
5554 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5555 * If this returns 1, you must commit the transaction to safely get your
5558 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5559 struct btrfs_root *root, struct dentry *dentry,
5562 struct btrfs_log_ctx *ctx)
5564 struct dentry *parent = dget_parent(dentry);
5567 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5568 start, end, 0, ctx);
5575 * should be called during mount to recover any replay any log trees
5578 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5581 struct btrfs_path *path;
5582 struct btrfs_trans_handle *trans;
5583 struct btrfs_key key;
5584 struct btrfs_key found_key;
5585 struct btrfs_key tmp_key;
5586 struct btrfs_root *log;
5587 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5588 struct walk_control wc = {
5589 .process_func = process_one_buffer,
5593 path = btrfs_alloc_path();
5597 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5599 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5600 if (IS_ERR(trans)) {
5601 ret = PTR_ERR(trans);
5608 ret = walk_log_tree(trans, log_root_tree, &wc);
5610 btrfs_handle_fs_error(fs_info, ret,
5611 "Failed to pin buffers while recovering log root tree.");
5616 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5617 key.offset = (u64)-1;
5618 key.type = BTRFS_ROOT_ITEM_KEY;
5621 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5624 btrfs_handle_fs_error(fs_info, ret,
5625 "Couldn't find tree log root.");
5629 if (path->slots[0] == 0)
5633 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5635 btrfs_release_path(path);
5636 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5639 log = btrfs_read_fs_root(log_root_tree, &found_key);
5642 btrfs_handle_fs_error(fs_info, ret,
5643 "Couldn't read tree log root.");
5647 tmp_key.objectid = found_key.offset;
5648 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5649 tmp_key.offset = (u64)-1;
5651 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5652 if (IS_ERR(wc.replay_dest)) {
5653 ret = PTR_ERR(wc.replay_dest);
5654 free_extent_buffer(log->node);
5655 free_extent_buffer(log->commit_root);
5657 btrfs_handle_fs_error(fs_info, ret,
5658 "Couldn't read target root for tree log recovery.");
5662 wc.replay_dest->log_root = log;
5663 btrfs_record_root_in_trans(trans, wc.replay_dest);
5664 ret = walk_log_tree(trans, log, &wc);
5666 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5667 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5671 key.offset = found_key.offset - 1;
5672 wc.replay_dest->log_root = NULL;
5673 free_extent_buffer(log->node);
5674 free_extent_buffer(log->commit_root);
5680 if (found_key.offset == 0)
5683 btrfs_release_path(path);
5685 /* step one is to pin it all, step two is to replay just inodes */
5688 wc.process_func = replay_one_buffer;
5689 wc.stage = LOG_WALK_REPLAY_INODES;
5692 /* step three is to replay everything */
5693 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5698 btrfs_free_path(path);
5700 /* step 4: commit the transaction, which also unpins the blocks */
5701 ret = btrfs_commit_transaction(trans);
5705 free_extent_buffer(log_root_tree->node);
5706 log_root_tree->log_root = NULL;
5707 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5708 kfree(log_root_tree);
5713 btrfs_end_transaction(wc.trans);
5714 btrfs_free_path(path);
5719 * there are some corner cases where we want to force a full
5720 * commit instead of allowing a directory to be logged.
5722 * They revolve around files there were unlinked from the directory, and
5723 * this function updates the parent directory so that a full commit is
5724 * properly done if it is fsync'd later after the unlinks are done.
5726 * Must be called before the unlink operations (updates to the subvolume tree,
5727 * inodes, etc) are done.
5729 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5730 struct btrfs_inode *dir, struct btrfs_inode *inode,
5734 * when we're logging a file, if it hasn't been renamed
5735 * or unlinked, and its inode is fully committed on disk,
5736 * we don't have to worry about walking up the directory chain
5737 * to log its parents.
5739 * So, we use the last_unlink_trans field to put this transid
5740 * into the file. When the file is logged we check it and
5741 * don't log the parents if the file is fully on disk.
5743 mutex_lock(&inode->log_mutex);
5744 inode->last_unlink_trans = trans->transid;
5745 mutex_unlock(&inode->log_mutex);
5748 * if this directory was already logged any new
5749 * names for this file/dir will get recorded
5752 if (dir->logged_trans == trans->transid)
5756 * if the inode we're about to unlink was logged,
5757 * the log will be properly updated for any new names
5759 if (inode->logged_trans == trans->transid)
5763 * when renaming files across directories, if the directory
5764 * there we're unlinking from gets fsync'd later on, there's
5765 * no way to find the destination directory later and fsync it
5766 * properly. So, we have to be conservative and force commits
5767 * so the new name gets discovered.
5772 /* we can safely do the unlink without any special recording */
5776 mutex_lock(&dir->log_mutex);
5777 dir->last_unlink_trans = trans->transid;
5778 mutex_unlock(&dir->log_mutex);
5782 * Make sure that if someone attempts to fsync the parent directory of a deleted
5783 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5784 * that after replaying the log tree of the parent directory's root we will not
5785 * see the snapshot anymore and at log replay time we will not see any log tree
5786 * corresponding to the deleted snapshot's root, which could lead to replaying
5787 * it after replaying the log tree of the parent directory (which would replay
5788 * the snapshot delete operation).
5790 * Must be called before the actual snapshot destroy operation (updates to the
5791 * parent root and tree of tree roots trees, etc) are done.
5793 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5794 struct btrfs_inode *dir)
5796 mutex_lock(&dir->log_mutex);
5797 dir->last_unlink_trans = trans->transid;
5798 mutex_unlock(&dir->log_mutex);
5802 * Call this after adding a new name for a file and it will properly
5803 * update the log to reflect the new name.
5805 * It will return zero if all goes well, and it will return 1 if a
5806 * full transaction commit is required.
5808 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5809 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
5810 struct dentry *parent)
5812 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5813 struct btrfs_root * root = inode->root;
5816 * this will force the logging code to walk the dentry chain
5819 if (S_ISREG(inode->vfs_inode.i_mode))
5820 inode->last_unlink_trans = trans->transid;
5823 * if this inode hasn't been logged and directory we're renaming it
5824 * from hasn't been logged, we don't need to log it
5826 if (inode->logged_trans <= fs_info->last_trans_committed &&
5827 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
5830 return btrfs_log_inode_parent(trans, root, &inode->vfs_inode, parent, 0,
5831 LLONG_MAX, 1, NULL);