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
42 * directory trouble cases
44 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
45 * log, we must force a full commit before doing an fsync of the directory
46 * where the unlink was done.
47 * ---> record transid of last unlink/rename per directory
51 * rename foo/some_dir foo2/some_dir
53 * fsync foo/some_dir/some_file
55 * The fsync above will unlink the original some_dir without recording
56 * it in its new location (foo2). After a crash, some_dir will be gone
57 * unless the fsync of some_file forces a full commit
59 * 2) we must log any new names for any file or dir that is in the fsync
60 * log. ---> check inode while renaming/linking.
62 * 2a) we must log any new names for any file or dir during rename
63 * when the directory they are being removed from was logged.
64 * ---> check inode and old parent dir during rename
66 * 2a is actually the more important variant. With the extra logging
67 * a crash might unlink the old name without recreating the new one
69 * 3) after a crash, we must go through any directories with a link count
70 * of zero and redo the rm -rf
77 * The directory f1 was fully removed from the FS, but fsync was never
78 * called on f1, only its parent dir. After a crash the rm -rf must
79 * be replayed. This must be able to recurse down the entire
80 * directory tree. The inode link count fixup code takes care of the
85 * stages for the tree walking. The first
86 * stage (0) is to only pin down the blocks we find
87 * the second stage (1) is to make sure that all the inodes
88 * we find in the log are created in the subvolume.
90 * The last stage is to deal with directories and links and extents
91 * and all the other fun semantics
93 #define LOG_WALK_PIN_ONLY 0
94 #define LOG_WALK_REPLAY_INODES 1
95 #define LOG_WALK_REPLAY_DIR_INDEX 2
96 #define LOG_WALK_REPLAY_ALL 3
98 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
99 struct btrfs_root *root, struct inode *inode,
103 struct btrfs_log_ctx *ctx);
104 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
105 struct btrfs_root *root,
106 struct btrfs_path *path, u64 objectid);
107 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
108 struct btrfs_root *root,
109 struct btrfs_root *log,
110 struct btrfs_path *path,
111 u64 dirid, int del_all);
114 * tree logging is a special write ahead log used to make sure that
115 * fsyncs and O_SYNCs can happen without doing full tree commits.
117 * Full tree commits are expensive because they require commonly
118 * modified blocks to be recowed, creating many dirty pages in the
119 * extent tree an 4x-6x higher write load than ext3.
121 * Instead of doing a tree commit on every fsync, we use the
122 * key ranges and transaction ids to find items for a given file or directory
123 * that have changed in this transaction. Those items are copied into
124 * a special tree (one per subvolume root), that tree is written to disk
125 * and then the fsync is considered complete.
127 * After a crash, items are copied out of the log-tree back into the
128 * subvolume tree. Any file data extents found are recorded in the extent
129 * allocation tree, and the log-tree freed.
131 * The log tree is read three times, once to pin down all the extents it is
132 * using in ram and once, once to create all the inodes logged in the tree
133 * and once to do all the other items.
137 * start a sub transaction and setup the log tree
138 * this increments the log tree writer count to make the people
139 * syncing the tree wait for us to finish
141 static int start_log_trans(struct btrfs_trans_handle *trans,
142 struct btrfs_root *root,
143 struct btrfs_log_ctx *ctx)
147 mutex_lock(&root->log_mutex);
149 if (root->log_root) {
150 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
155 if (!root->log_start_pid) {
156 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
157 root->log_start_pid = current->pid;
158 } else if (root->log_start_pid != current->pid) {
159 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
162 mutex_lock(&root->fs_info->tree_log_mutex);
163 if (!root->fs_info->log_root_tree)
164 ret = btrfs_init_log_root_tree(trans, root->fs_info);
165 mutex_unlock(&root->fs_info->tree_log_mutex);
169 ret = btrfs_add_log_tree(trans, root);
173 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
174 root->log_start_pid = current->pid;
177 atomic_inc(&root->log_batch);
178 atomic_inc(&root->log_writers);
180 int index = root->log_transid % 2;
181 list_add_tail(&ctx->list, &root->log_ctxs[index]);
182 ctx->log_transid = root->log_transid;
186 mutex_unlock(&root->log_mutex);
191 * returns 0 if there was a log transaction running and we were able
192 * to join, or returns -ENOENT if there were not transactions
195 static int join_running_log_trans(struct btrfs_root *root)
203 mutex_lock(&root->log_mutex);
204 if (root->log_root) {
206 atomic_inc(&root->log_writers);
208 mutex_unlock(&root->log_mutex);
213 * This either makes the current running log transaction wait
214 * until you call btrfs_end_log_trans() or it makes any future
215 * log transactions wait until you call btrfs_end_log_trans()
217 int btrfs_pin_log_trans(struct btrfs_root *root)
221 mutex_lock(&root->log_mutex);
222 atomic_inc(&root->log_writers);
223 mutex_unlock(&root->log_mutex);
228 * indicate we're done making changes to the log tree
229 * and wake up anyone waiting to do a sync
231 void btrfs_end_log_trans(struct btrfs_root *root)
233 if (atomic_dec_and_test(&root->log_writers)) {
235 * Implicit memory barrier after atomic_dec_and_test
237 if (waitqueue_active(&root->log_writer_wait))
238 wake_up(&root->log_writer_wait);
244 * the walk control struct is used to pass state down the chain when
245 * processing the log tree. The stage field tells us which part
246 * of the log tree processing we are currently doing. The others
247 * are state fields used for that specific part
249 struct walk_control {
250 /* should we free the extent on disk when done? This is used
251 * at transaction commit time while freeing a log tree
255 /* should we write out the extent buffer? This is used
256 * while flushing the log tree to disk during a sync
260 /* should we wait for the extent buffer io to finish? Also used
261 * while flushing the log tree to disk for a sync
265 /* pin only walk, we record which extents on disk belong to the
270 /* what stage of the replay code we're currently in */
273 /* the root we are currently replaying */
274 struct btrfs_root *replay_dest;
276 /* the trans handle for the current replay */
277 struct btrfs_trans_handle *trans;
279 /* the function that gets used to process blocks we find in the
280 * tree. Note the extent_buffer might not be up to date when it is
281 * passed in, and it must be checked or read if you need the data
284 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
285 struct walk_control *wc, u64 gen);
289 * process_func used to pin down extents, write them or wait on them
291 static int process_one_buffer(struct btrfs_root *log,
292 struct extent_buffer *eb,
293 struct walk_control *wc, u64 gen)
298 * If this fs is mixed then we need to be able to process the leaves to
299 * pin down any logged extents, so we have to read the block.
301 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
302 ret = btrfs_read_buffer(eb, gen);
308 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
311 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
312 if (wc->pin && btrfs_header_level(eb) == 0)
313 ret = btrfs_exclude_logged_extents(log, eb);
315 btrfs_write_tree_block(eb);
317 btrfs_wait_tree_block_writeback(eb);
323 * Item overwrite used by replay and tree logging. eb, slot and key all refer
324 * to the src data we are copying out.
326 * root is the tree we are copying into, and path is a scratch
327 * path for use in this function (it should be released on entry and
328 * will be released on exit).
330 * If the key is already in the destination tree the existing item is
331 * overwritten. If the existing item isn't big enough, it is extended.
332 * If it is too large, it is truncated.
334 * If the key isn't in the destination yet, a new item is inserted.
336 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
337 struct btrfs_root *root,
338 struct btrfs_path *path,
339 struct extent_buffer *eb, int slot,
340 struct btrfs_key *key)
344 u64 saved_i_size = 0;
345 int save_old_i_size = 0;
346 unsigned long src_ptr;
347 unsigned long dst_ptr;
348 int overwrite_root = 0;
349 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
351 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
354 item_size = btrfs_item_size_nr(eb, slot);
355 src_ptr = btrfs_item_ptr_offset(eb, slot);
357 /* look for the key in the destination tree */
358 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
365 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
367 if (dst_size != item_size)
370 if (item_size == 0) {
371 btrfs_release_path(path);
374 dst_copy = kmalloc(item_size, GFP_NOFS);
375 src_copy = kmalloc(item_size, GFP_NOFS);
376 if (!dst_copy || !src_copy) {
377 btrfs_release_path(path);
383 read_extent_buffer(eb, src_copy, src_ptr, item_size);
385 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
386 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
388 ret = memcmp(dst_copy, src_copy, item_size);
393 * they have the same contents, just return, this saves
394 * us from cowing blocks in the destination tree and doing
395 * extra writes that may not have been done by a previous
399 btrfs_release_path(path);
404 * We need to load the old nbytes into the inode so when we
405 * replay the extents we've logged we get the right nbytes.
408 struct btrfs_inode_item *item;
412 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
413 struct btrfs_inode_item);
414 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
415 item = btrfs_item_ptr(eb, slot,
416 struct btrfs_inode_item);
417 btrfs_set_inode_nbytes(eb, item, nbytes);
420 * If this is a directory we need to reset the i_size to
421 * 0 so that we can set it up properly when replaying
422 * the rest of the items in this log.
424 mode = btrfs_inode_mode(eb, item);
426 btrfs_set_inode_size(eb, item, 0);
428 } else if (inode_item) {
429 struct btrfs_inode_item *item;
433 * New inode, set nbytes to 0 so that the nbytes comes out
434 * properly when we replay the extents.
436 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
437 btrfs_set_inode_nbytes(eb, item, 0);
440 * If this is a directory we need to reset the i_size to 0 so
441 * that we can set it up properly when replaying the rest of
442 * the items in this log.
444 mode = btrfs_inode_mode(eb, item);
446 btrfs_set_inode_size(eb, item, 0);
449 btrfs_release_path(path);
450 /* try to insert the key into the destination tree */
451 path->skip_release_on_error = 1;
452 ret = btrfs_insert_empty_item(trans, root, path,
454 path->skip_release_on_error = 0;
456 /* make sure any existing item is the correct size */
457 if (ret == -EEXIST || ret == -EOVERFLOW) {
459 found_size = btrfs_item_size_nr(path->nodes[0],
461 if (found_size > item_size)
462 btrfs_truncate_item(root, path, item_size, 1);
463 else if (found_size < item_size)
464 btrfs_extend_item(root, path,
465 item_size - found_size);
469 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
472 /* don't overwrite an existing inode if the generation number
473 * was logged as zero. This is done when the tree logging code
474 * is just logging an inode to make sure it exists after recovery.
476 * Also, don't overwrite i_size on directories during replay.
477 * log replay inserts and removes directory items based on the
478 * state of the tree found in the subvolume, and i_size is modified
481 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
482 struct btrfs_inode_item *src_item;
483 struct btrfs_inode_item *dst_item;
485 src_item = (struct btrfs_inode_item *)src_ptr;
486 dst_item = (struct btrfs_inode_item *)dst_ptr;
488 if (btrfs_inode_generation(eb, src_item) == 0) {
489 struct extent_buffer *dst_eb = path->nodes[0];
490 const u64 ino_size = btrfs_inode_size(eb, src_item);
493 * For regular files an ino_size == 0 is used only when
494 * logging that an inode exists, as part of a directory
495 * fsync, and the inode wasn't fsynced before. In this
496 * case don't set the size of the inode in the fs/subvol
497 * tree, otherwise we would be throwing valid data away.
499 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
500 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
502 struct btrfs_map_token token;
504 btrfs_init_map_token(&token);
505 btrfs_set_token_inode_size(dst_eb, dst_item,
511 if (overwrite_root &&
512 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
513 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
515 saved_i_size = btrfs_inode_size(path->nodes[0],
520 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
523 if (save_old_i_size) {
524 struct btrfs_inode_item *dst_item;
525 dst_item = (struct btrfs_inode_item *)dst_ptr;
526 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
529 /* make sure the generation is filled in */
530 if (key->type == BTRFS_INODE_ITEM_KEY) {
531 struct btrfs_inode_item *dst_item;
532 dst_item = (struct btrfs_inode_item *)dst_ptr;
533 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
534 btrfs_set_inode_generation(path->nodes[0], dst_item,
539 btrfs_mark_buffer_dirty(path->nodes[0]);
540 btrfs_release_path(path);
545 * simple helper to read an inode off the disk from a given root
546 * This can only be called for subvolume roots and not for the log
548 static noinline struct inode *read_one_inode(struct btrfs_root *root,
551 struct btrfs_key key;
554 key.objectid = objectid;
555 key.type = BTRFS_INODE_ITEM_KEY;
557 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
560 } else if (is_bad_inode(inode)) {
567 /* replays a single extent in 'eb' at 'slot' with 'key' into the
568 * subvolume 'root'. path is released on entry and should be released
571 * extents in the log tree have not been allocated out of the extent
572 * tree yet. So, this completes the allocation, taking a reference
573 * as required if the extent already exists or creating a new extent
574 * if it isn't in the extent allocation tree yet.
576 * The extent is inserted into the file, dropping any existing extents
577 * from the file that overlap the new one.
579 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
580 struct btrfs_root *root,
581 struct btrfs_path *path,
582 struct extent_buffer *eb, int slot,
583 struct btrfs_key *key)
587 u64 start = key->offset;
589 struct btrfs_file_extent_item *item;
590 struct inode *inode = NULL;
594 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
595 found_type = btrfs_file_extent_type(eb, item);
597 if (found_type == BTRFS_FILE_EXTENT_REG ||
598 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
599 nbytes = btrfs_file_extent_num_bytes(eb, item);
600 extent_end = start + nbytes;
603 * We don't add to the inodes nbytes if we are prealloc or a
606 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
608 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
609 size = btrfs_file_extent_inline_len(eb, slot, item);
610 nbytes = btrfs_file_extent_ram_bytes(eb, item);
611 extent_end = ALIGN(start + size, root->sectorsize);
617 inode = read_one_inode(root, key->objectid);
624 * first check to see if we already have this extent in the
625 * file. This must be done before the btrfs_drop_extents run
626 * so we don't try to drop this extent.
628 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
632 (found_type == BTRFS_FILE_EXTENT_REG ||
633 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
634 struct btrfs_file_extent_item cmp1;
635 struct btrfs_file_extent_item cmp2;
636 struct btrfs_file_extent_item *existing;
637 struct extent_buffer *leaf;
639 leaf = path->nodes[0];
640 existing = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_file_extent_item);
643 read_extent_buffer(eb, &cmp1, (unsigned long)item,
645 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
649 * we already have a pointer to this exact extent,
650 * we don't have to do anything
652 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
653 btrfs_release_path(path);
657 btrfs_release_path(path);
659 /* drop any overlapping extents */
660 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
664 if (found_type == BTRFS_FILE_EXTENT_REG ||
665 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
667 unsigned long dest_offset;
668 struct btrfs_key ins;
670 ret = btrfs_insert_empty_item(trans, root, path, key,
674 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
676 copy_extent_buffer(path->nodes[0], eb, dest_offset,
677 (unsigned long)item, sizeof(*item));
679 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
680 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
681 ins.type = BTRFS_EXTENT_ITEM_KEY;
682 offset = key->offset - btrfs_file_extent_offset(eb, item);
685 * Manually record dirty extent, as here we did a shallow
686 * file extent item copy and skip normal backref update,
687 * but modifying extent tree all by ourselves.
688 * So need to manually record dirty extent for qgroup,
689 * as the owner of the file extent changed from log tree
690 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
692 ret = btrfs_qgroup_insert_dirty_extent(trans, root->fs_info,
693 btrfs_file_extent_disk_bytenr(eb, item),
694 btrfs_file_extent_disk_num_bytes(eb, item),
699 if (ins.objectid > 0) {
702 LIST_HEAD(ordered_sums);
704 * is this extent already allocated in the extent
705 * allocation tree? If so, just add a reference
707 ret = btrfs_lookup_data_extent(root, ins.objectid,
710 ret = btrfs_inc_extent_ref(trans, root,
711 ins.objectid, ins.offset,
712 0, root->root_key.objectid,
713 key->objectid, offset);
718 * insert the extent pointer in the extent
721 ret = btrfs_alloc_logged_file_extent(trans,
722 root, root->root_key.objectid,
723 key->objectid, offset, &ins);
727 btrfs_release_path(path);
729 if (btrfs_file_extent_compression(eb, item)) {
730 csum_start = ins.objectid;
731 csum_end = csum_start + ins.offset;
733 csum_start = ins.objectid +
734 btrfs_file_extent_offset(eb, item);
735 csum_end = csum_start +
736 btrfs_file_extent_num_bytes(eb, item);
739 ret = btrfs_lookup_csums_range(root->log_root,
740 csum_start, csum_end - 1,
745 * Now delete all existing cums in the csum root that
746 * cover our range. We do this because we can have an
747 * extent that is completely referenced by one file
748 * extent item and partially referenced by another
749 * file extent item (like after using the clone or
750 * extent_same ioctls). In this case if we end up doing
751 * the replay of the one that partially references the
752 * extent first, and we do not do the csum deletion
753 * below, we can get 2 csum items in the csum tree that
754 * overlap each other. For example, imagine our log has
755 * the two following file extent items:
757 * key (257 EXTENT_DATA 409600)
758 * extent data disk byte 12845056 nr 102400
759 * extent data offset 20480 nr 20480 ram 102400
761 * key (257 EXTENT_DATA 819200)
762 * extent data disk byte 12845056 nr 102400
763 * extent data offset 0 nr 102400 ram 102400
765 * Where the second one fully references the 100K extent
766 * that starts at disk byte 12845056, and the log tree
767 * has a single csum item that covers the entire range
770 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
772 * After the first file extent item is replayed, the
773 * csum tree gets the following csum item:
775 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
777 * Which covers the 20K sub-range starting at offset 20K
778 * of our extent. Now when we replay the second file
779 * extent item, if we do not delete existing csum items
780 * that cover any of its blocks, we end up getting two
781 * csum items in our csum tree that overlap each other:
783 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
784 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
786 * Which is a problem, because after this anyone trying
787 * to lookup up for the checksum of any block of our
788 * extent starting at an offset of 40K or higher, will
789 * end up looking at the second csum item only, which
790 * does not contain the checksum for any block starting
791 * at offset 40K or higher of our extent.
793 while (!list_empty(&ordered_sums)) {
794 struct btrfs_ordered_sum *sums;
795 sums = list_entry(ordered_sums.next,
796 struct btrfs_ordered_sum,
799 ret = btrfs_del_csums(trans,
800 root->fs_info->csum_root,
804 ret = btrfs_csum_file_blocks(trans,
805 root->fs_info->csum_root,
807 list_del(&sums->list);
813 btrfs_release_path(path);
815 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
816 /* inline extents are easy, we just overwrite them */
817 ret = overwrite_item(trans, root, path, eb, slot, key);
822 inode_add_bytes(inode, nbytes);
823 ret = btrfs_update_inode(trans, root, inode);
831 * when cleaning up conflicts between the directory names in the
832 * subvolume, directory names in the log and directory names in the
833 * inode back references, we may have to unlink inodes from directories.
835 * This is a helper function to do the unlink of a specific directory
838 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
839 struct btrfs_root *root,
840 struct btrfs_path *path,
842 struct btrfs_dir_item *di)
847 struct extent_buffer *leaf;
848 struct btrfs_key location;
851 leaf = path->nodes[0];
853 btrfs_dir_item_key_to_cpu(leaf, di, &location);
854 name_len = btrfs_dir_name_len(leaf, di);
855 name = kmalloc(name_len, GFP_NOFS);
859 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
860 btrfs_release_path(path);
862 inode = read_one_inode(root, location.objectid);
868 ret = link_to_fixup_dir(trans, root, path, location.objectid);
872 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
876 ret = btrfs_run_delayed_items(trans, root);
884 * helper function to see if a given name and sequence number found
885 * in an inode back reference are already in a directory and correctly
886 * point to this inode
888 static noinline int inode_in_dir(struct btrfs_root *root,
889 struct btrfs_path *path,
890 u64 dirid, u64 objectid, u64 index,
891 const char *name, int name_len)
893 struct btrfs_dir_item *di;
894 struct btrfs_key location;
897 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
898 index, name, name_len, 0);
899 if (di && !IS_ERR(di)) {
900 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
901 if (location.objectid != objectid)
905 btrfs_release_path(path);
907 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
908 if (di && !IS_ERR(di)) {
909 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
910 if (location.objectid != objectid)
916 btrfs_release_path(path);
921 * helper function to check a log tree for a named back reference in
922 * an inode. This is used to decide if a back reference that is
923 * found in the subvolume conflicts with what we find in the log.
925 * inode backreferences may have multiple refs in a single item,
926 * during replay we process one reference at a time, and we don't
927 * want to delete valid links to a file from the subvolume if that
928 * link is also in the log.
930 static noinline int backref_in_log(struct btrfs_root *log,
931 struct btrfs_key *key,
933 const char *name, int namelen)
935 struct btrfs_path *path;
936 struct btrfs_inode_ref *ref;
938 unsigned long ptr_end;
939 unsigned long name_ptr;
945 path = btrfs_alloc_path();
949 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
953 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
955 if (key->type == BTRFS_INODE_EXTREF_KEY) {
956 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
957 name, namelen, NULL))
963 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
964 ptr_end = ptr + item_size;
965 while (ptr < ptr_end) {
966 ref = (struct btrfs_inode_ref *)ptr;
967 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
968 if (found_name_len == namelen) {
969 name_ptr = (unsigned long)(ref + 1);
970 ret = memcmp_extent_buffer(path->nodes[0], name,
977 ptr = (unsigned long)(ref + 1) + found_name_len;
980 btrfs_free_path(path);
984 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
985 struct btrfs_root *root,
986 struct btrfs_path *path,
987 struct btrfs_root *log_root,
988 struct inode *dir, struct inode *inode,
989 struct extent_buffer *eb,
990 u64 inode_objectid, u64 parent_objectid,
991 u64 ref_index, char *name, int namelen,
997 struct extent_buffer *leaf;
998 struct btrfs_dir_item *di;
999 struct btrfs_key search_key;
1000 struct btrfs_inode_extref *extref;
1003 /* Search old style refs */
1004 search_key.objectid = inode_objectid;
1005 search_key.type = BTRFS_INODE_REF_KEY;
1006 search_key.offset = parent_objectid;
1007 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1009 struct btrfs_inode_ref *victim_ref;
1011 unsigned long ptr_end;
1013 leaf = path->nodes[0];
1015 /* are we trying to overwrite a back ref for the root directory
1016 * if so, just jump out, we're done
1018 if (search_key.objectid == search_key.offset)
1021 /* check all the names in this back reference to see
1022 * if they are in the log. if so, we allow them to stay
1023 * otherwise they must be unlinked as a conflict
1025 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1026 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1027 while (ptr < ptr_end) {
1028 victim_ref = (struct btrfs_inode_ref *)ptr;
1029 victim_name_len = btrfs_inode_ref_name_len(leaf,
1031 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1035 read_extent_buffer(leaf, victim_name,
1036 (unsigned long)(victim_ref + 1),
1039 if (!backref_in_log(log_root, &search_key,
1044 btrfs_release_path(path);
1046 ret = btrfs_unlink_inode(trans, root, dir,
1052 ret = btrfs_run_delayed_items(trans, root);
1060 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1064 * NOTE: we have searched root tree and checked the
1065 * corresponding ref, it does not need to check again.
1069 btrfs_release_path(path);
1071 /* Same search but for extended refs */
1072 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1073 inode_objectid, parent_objectid, 0,
1075 if (!IS_ERR_OR_NULL(extref)) {
1079 struct inode *victim_parent;
1081 leaf = path->nodes[0];
1083 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1084 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1086 while (cur_offset < item_size) {
1087 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1089 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1091 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1094 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1097 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1100 search_key.objectid = inode_objectid;
1101 search_key.type = BTRFS_INODE_EXTREF_KEY;
1102 search_key.offset = btrfs_extref_hash(parent_objectid,
1106 if (!backref_in_log(log_root, &search_key,
1107 parent_objectid, victim_name,
1110 victim_parent = read_one_inode(root,
1112 if (victim_parent) {
1114 btrfs_release_path(path);
1116 ret = btrfs_unlink_inode(trans, root,
1122 ret = btrfs_run_delayed_items(
1125 iput(victim_parent);
1136 cur_offset += victim_name_len + sizeof(*extref);
1140 btrfs_release_path(path);
1142 /* look for a conflicting sequence number */
1143 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1144 ref_index, name, namelen, 0);
1145 if (di && !IS_ERR(di)) {
1146 ret = drop_one_dir_item(trans, root, path, dir, di);
1150 btrfs_release_path(path);
1152 /* look for a conflicing name */
1153 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1155 if (di && !IS_ERR(di)) {
1156 ret = drop_one_dir_item(trans, root, path, dir, di);
1160 btrfs_release_path(path);
1165 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1166 u32 *namelen, char **name, u64 *index,
1167 u64 *parent_objectid)
1169 struct btrfs_inode_extref *extref;
1171 extref = (struct btrfs_inode_extref *)ref_ptr;
1173 *namelen = btrfs_inode_extref_name_len(eb, extref);
1174 *name = kmalloc(*namelen, GFP_NOFS);
1178 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1181 *index = btrfs_inode_extref_index(eb, extref);
1182 if (parent_objectid)
1183 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1188 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1189 u32 *namelen, char **name, u64 *index)
1191 struct btrfs_inode_ref *ref;
1193 ref = (struct btrfs_inode_ref *)ref_ptr;
1195 *namelen = btrfs_inode_ref_name_len(eb, ref);
1196 *name = kmalloc(*namelen, GFP_NOFS);
1200 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1202 *index = btrfs_inode_ref_index(eb, ref);
1208 * replay one inode back reference item found in the log tree.
1209 * eb, slot and key refer to the buffer and key found in the log tree.
1210 * root is the destination we are replaying into, and path is for temp
1211 * use by this function. (it should be released on return).
1213 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1214 struct btrfs_root *root,
1215 struct btrfs_root *log,
1216 struct btrfs_path *path,
1217 struct extent_buffer *eb, int slot,
1218 struct btrfs_key *key)
1220 struct inode *dir = NULL;
1221 struct inode *inode = NULL;
1222 unsigned long ref_ptr;
1223 unsigned long ref_end;
1227 int search_done = 0;
1228 int log_ref_ver = 0;
1229 u64 parent_objectid;
1232 int ref_struct_size;
1234 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1235 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1237 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1238 struct btrfs_inode_extref *r;
1240 ref_struct_size = sizeof(struct btrfs_inode_extref);
1242 r = (struct btrfs_inode_extref *)ref_ptr;
1243 parent_objectid = btrfs_inode_extref_parent(eb, r);
1245 ref_struct_size = sizeof(struct btrfs_inode_ref);
1246 parent_objectid = key->offset;
1248 inode_objectid = key->objectid;
1251 * it is possible that we didn't log all the parent directories
1252 * for a given inode. If we don't find the dir, just don't
1253 * copy the back ref in. The link count fixup code will take
1256 dir = read_one_inode(root, parent_objectid);
1262 inode = read_one_inode(root, inode_objectid);
1268 while (ref_ptr < ref_end) {
1270 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1271 &ref_index, &parent_objectid);
1273 * parent object can change from one array
1277 dir = read_one_inode(root, parent_objectid);
1283 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1289 /* if we already have a perfect match, we're done */
1290 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1291 ref_index, name, namelen)) {
1293 * look for a conflicting back reference in the
1294 * metadata. if we find one we have to unlink that name
1295 * of the file before we add our new link. Later on, we
1296 * overwrite any existing back reference, and we don't
1297 * want to create dangling pointers in the directory.
1301 ret = __add_inode_ref(trans, root, path, log,
1305 ref_index, name, namelen,
1314 /* insert our name */
1315 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1320 btrfs_update_inode(trans, root, inode);
1323 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1332 /* finally write the back reference in the inode */
1333 ret = overwrite_item(trans, root, path, eb, slot, key);
1335 btrfs_release_path(path);
1342 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1343 struct btrfs_root *root, u64 ino)
1347 ret = btrfs_insert_orphan_item(trans, root, ino);
1354 static int count_inode_extrefs(struct btrfs_root *root,
1355 struct inode *inode, struct btrfs_path *path)
1359 unsigned int nlink = 0;
1362 u64 inode_objectid = btrfs_ino(inode);
1365 struct btrfs_inode_extref *extref;
1366 struct extent_buffer *leaf;
1369 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1374 leaf = path->nodes[0];
1375 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1376 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1379 while (cur_offset < item_size) {
1380 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1381 name_len = btrfs_inode_extref_name_len(leaf, extref);
1385 cur_offset += name_len + sizeof(*extref);
1389 btrfs_release_path(path);
1391 btrfs_release_path(path);
1393 if (ret < 0 && ret != -ENOENT)
1398 static int count_inode_refs(struct btrfs_root *root,
1399 struct inode *inode, struct btrfs_path *path)
1402 struct btrfs_key key;
1403 unsigned int nlink = 0;
1405 unsigned long ptr_end;
1407 u64 ino = btrfs_ino(inode);
1410 key.type = BTRFS_INODE_REF_KEY;
1411 key.offset = (u64)-1;
1414 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1418 if (path->slots[0] == 0)
1423 btrfs_item_key_to_cpu(path->nodes[0], &key,
1425 if (key.objectid != ino ||
1426 key.type != BTRFS_INODE_REF_KEY)
1428 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1429 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1431 while (ptr < ptr_end) {
1432 struct btrfs_inode_ref *ref;
1434 ref = (struct btrfs_inode_ref *)ptr;
1435 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1437 ptr = (unsigned long)(ref + 1) + name_len;
1441 if (key.offset == 0)
1443 if (path->slots[0] > 0) {
1448 btrfs_release_path(path);
1450 btrfs_release_path(path);
1456 * There are a few corners where the link count of the file can't
1457 * be properly maintained during replay. So, instead of adding
1458 * lots of complexity to the log code, we just scan the backrefs
1459 * for any file that has been through replay.
1461 * The scan will update the link count on the inode to reflect the
1462 * number of back refs found. If it goes down to zero, the iput
1463 * will free the inode.
1465 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1466 struct btrfs_root *root,
1467 struct inode *inode)
1469 struct btrfs_path *path;
1472 u64 ino = btrfs_ino(inode);
1474 path = btrfs_alloc_path();
1478 ret = count_inode_refs(root, inode, path);
1484 ret = count_inode_extrefs(root, inode, path);
1492 if (nlink != inode->i_nlink) {
1493 set_nlink(inode, nlink);
1494 btrfs_update_inode(trans, root, inode);
1496 BTRFS_I(inode)->index_cnt = (u64)-1;
1498 if (inode->i_nlink == 0) {
1499 if (S_ISDIR(inode->i_mode)) {
1500 ret = replay_dir_deletes(trans, root, NULL, path,
1505 ret = insert_orphan_item(trans, root, ino);
1509 btrfs_free_path(path);
1513 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1514 struct btrfs_root *root,
1515 struct btrfs_path *path)
1518 struct btrfs_key key;
1519 struct inode *inode;
1521 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1522 key.type = BTRFS_ORPHAN_ITEM_KEY;
1523 key.offset = (u64)-1;
1525 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1530 if (path->slots[0] == 0)
1535 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1536 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1537 key.type != BTRFS_ORPHAN_ITEM_KEY)
1540 ret = btrfs_del_item(trans, root, path);
1544 btrfs_release_path(path);
1545 inode = read_one_inode(root, key.offset);
1549 ret = fixup_inode_link_count(trans, root, inode);
1555 * fixup on a directory may create new entries,
1556 * make sure we always look for the highset possible
1559 key.offset = (u64)-1;
1563 btrfs_release_path(path);
1569 * record a given inode in the fixup dir so we can check its link
1570 * count when replay is done. The link count is incremented here
1571 * so the inode won't go away until we check it
1573 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1574 struct btrfs_root *root,
1575 struct btrfs_path *path,
1578 struct btrfs_key key;
1580 struct inode *inode;
1582 inode = read_one_inode(root, objectid);
1586 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1587 key.type = BTRFS_ORPHAN_ITEM_KEY;
1588 key.offset = objectid;
1590 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1592 btrfs_release_path(path);
1594 if (!inode->i_nlink)
1595 set_nlink(inode, 1);
1598 ret = btrfs_update_inode(trans, root, inode);
1599 } else if (ret == -EEXIST) {
1602 BUG(); /* Logic Error */
1610 * when replaying the log for a directory, we only insert names
1611 * for inodes that actually exist. This means an fsync on a directory
1612 * does not implicitly fsync all the new files in it
1614 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1615 struct btrfs_root *root,
1616 u64 dirid, u64 index,
1617 char *name, int name_len,
1618 struct btrfs_key *location)
1620 struct inode *inode;
1624 inode = read_one_inode(root, location->objectid);
1628 dir = read_one_inode(root, dirid);
1634 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1636 /* FIXME, put inode into FIXUP list */
1644 * Return true if an inode reference exists in the log for the given name,
1645 * inode and parent inode.
1647 static bool name_in_log_ref(struct btrfs_root *log_root,
1648 const char *name, const int name_len,
1649 const u64 dirid, const u64 ino)
1651 struct btrfs_key search_key;
1653 search_key.objectid = ino;
1654 search_key.type = BTRFS_INODE_REF_KEY;
1655 search_key.offset = dirid;
1656 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1659 search_key.type = BTRFS_INODE_EXTREF_KEY;
1660 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1661 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1668 * take a single entry in a log directory item and replay it into
1671 * if a conflicting item exists in the subdirectory already,
1672 * the inode it points to is unlinked and put into the link count
1675 * If a name from the log points to a file or directory that does
1676 * not exist in the FS, it is skipped. fsyncs on directories
1677 * do not force down inodes inside that directory, just changes to the
1678 * names or unlinks in a directory.
1680 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1681 * non-existing inode) and 1 if the name was replayed.
1683 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1684 struct btrfs_root *root,
1685 struct btrfs_path *path,
1686 struct extent_buffer *eb,
1687 struct btrfs_dir_item *di,
1688 struct btrfs_key *key)
1692 struct btrfs_dir_item *dst_di;
1693 struct btrfs_key found_key;
1694 struct btrfs_key log_key;
1699 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1700 bool name_added = false;
1702 dir = read_one_inode(root, key->objectid);
1706 name_len = btrfs_dir_name_len(eb, di);
1707 name = kmalloc(name_len, GFP_NOFS);
1713 log_type = btrfs_dir_type(eb, di);
1714 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1717 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1718 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1723 btrfs_release_path(path);
1725 if (key->type == BTRFS_DIR_ITEM_KEY) {
1726 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1728 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1729 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1738 if (IS_ERR_OR_NULL(dst_di)) {
1739 /* we need a sequence number to insert, so we only
1740 * do inserts for the BTRFS_DIR_INDEX_KEY types
1742 if (key->type != BTRFS_DIR_INDEX_KEY)
1747 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1748 /* the existing item matches the logged item */
1749 if (found_key.objectid == log_key.objectid &&
1750 found_key.type == log_key.type &&
1751 found_key.offset == log_key.offset &&
1752 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1753 update_size = false;
1758 * don't drop the conflicting directory entry if the inode
1759 * for the new entry doesn't exist
1764 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1768 if (key->type == BTRFS_DIR_INDEX_KEY)
1771 btrfs_release_path(path);
1772 if (!ret && update_size) {
1773 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1774 ret = btrfs_update_inode(trans, root, dir);
1778 if (!ret && name_added)
1783 if (name_in_log_ref(root->log_root, name, name_len,
1784 key->objectid, log_key.objectid)) {
1785 /* The dentry will be added later. */
1787 update_size = false;
1790 btrfs_release_path(path);
1791 ret = insert_one_name(trans, root, key->objectid, key->offset,
1792 name, name_len, &log_key);
1793 if (ret && ret != -ENOENT && ret != -EEXIST)
1797 update_size = false;
1803 * find all the names in a directory item and reconcile them into
1804 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1805 * one name in a directory item, but the same code gets used for
1806 * both directory index types
1808 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1809 struct btrfs_root *root,
1810 struct btrfs_path *path,
1811 struct extent_buffer *eb, int slot,
1812 struct btrfs_key *key)
1815 u32 item_size = btrfs_item_size_nr(eb, slot);
1816 struct btrfs_dir_item *di;
1819 unsigned long ptr_end;
1820 struct btrfs_path *fixup_path = NULL;
1822 ptr = btrfs_item_ptr_offset(eb, slot);
1823 ptr_end = ptr + item_size;
1824 while (ptr < ptr_end) {
1825 di = (struct btrfs_dir_item *)ptr;
1826 if (verify_dir_item(root, eb, di))
1828 name_len = btrfs_dir_name_len(eb, di);
1829 ret = replay_one_name(trans, root, path, eb, di, key);
1832 ptr = (unsigned long)(di + 1);
1836 * If this entry refers to a non-directory (directories can not
1837 * have a link count > 1) and it was added in the transaction
1838 * that was not committed, make sure we fixup the link count of
1839 * the inode it the entry points to. Otherwise something like
1840 * the following would result in a directory pointing to an
1841 * inode with a wrong link that does not account for this dir
1849 * ln testdir/bar testdir/bar_link
1850 * ln testdir/foo testdir/foo_link
1851 * xfs_io -c "fsync" testdir/bar
1855 * mount fs, log replay happens
1857 * File foo would remain with a link count of 1 when it has two
1858 * entries pointing to it in the directory testdir. This would
1859 * make it impossible to ever delete the parent directory has
1860 * it would result in stale dentries that can never be deleted.
1862 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1863 struct btrfs_key di_key;
1866 fixup_path = btrfs_alloc_path();
1873 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1874 ret = link_to_fixup_dir(trans, root, fixup_path,
1881 btrfs_free_path(fixup_path);
1886 * directory replay has two parts. There are the standard directory
1887 * items in the log copied from the subvolume, and range items
1888 * created in the log while the subvolume was logged.
1890 * The range items tell us which parts of the key space the log
1891 * is authoritative for. During replay, if a key in the subvolume
1892 * directory is in a logged range item, but not actually in the log
1893 * that means it was deleted from the directory before the fsync
1894 * and should be removed.
1896 static noinline int find_dir_range(struct btrfs_root *root,
1897 struct btrfs_path *path,
1898 u64 dirid, int key_type,
1899 u64 *start_ret, u64 *end_ret)
1901 struct btrfs_key key;
1903 struct btrfs_dir_log_item *item;
1907 if (*start_ret == (u64)-1)
1910 key.objectid = dirid;
1911 key.type = key_type;
1912 key.offset = *start_ret;
1914 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1918 if (path->slots[0] == 0)
1923 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1925 if (key.type != key_type || key.objectid != dirid) {
1929 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1930 struct btrfs_dir_log_item);
1931 found_end = btrfs_dir_log_end(path->nodes[0], item);
1933 if (*start_ret >= key.offset && *start_ret <= found_end) {
1935 *start_ret = key.offset;
1936 *end_ret = found_end;
1941 /* check the next slot in the tree to see if it is a valid item */
1942 nritems = btrfs_header_nritems(path->nodes[0]);
1943 if (path->slots[0] >= nritems) {
1944 ret = btrfs_next_leaf(root, path);
1951 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1953 if (key.type != key_type || key.objectid != dirid) {
1957 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1958 struct btrfs_dir_log_item);
1959 found_end = btrfs_dir_log_end(path->nodes[0], item);
1960 *start_ret = key.offset;
1961 *end_ret = found_end;
1964 btrfs_release_path(path);
1969 * this looks for a given directory item in the log. If the directory
1970 * item is not in the log, the item is removed and the inode it points
1973 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1974 struct btrfs_root *root,
1975 struct btrfs_root *log,
1976 struct btrfs_path *path,
1977 struct btrfs_path *log_path,
1979 struct btrfs_key *dir_key)
1982 struct extent_buffer *eb;
1985 struct btrfs_dir_item *di;
1986 struct btrfs_dir_item *log_di;
1989 unsigned long ptr_end;
1991 struct inode *inode;
1992 struct btrfs_key location;
1995 eb = path->nodes[0];
1996 slot = path->slots[0];
1997 item_size = btrfs_item_size_nr(eb, slot);
1998 ptr = btrfs_item_ptr_offset(eb, slot);
1999 ptr_end = ptr + item_size;
2000 while (ptr < ptr_end) {
2001 di = (struct btrfs_dir_item *)ptr;
2002 if (verify_dir_item(root, eb, di)) {
2007 name_len = btrfs_dir_name_len(eb, di);
2008 name = kmalloc(name_len, GFP_NOFS);
2013 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2016 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2017 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2020 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2021 log_di = btrfs_lookup_dir_index_item(trans, log,
2027 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2028 btrfs_dir_item_key_to_cpu(eb, di, &location);
2029 btrfs_release_path(path);
2030 btrfs_release_path(log_path);
2031 inode = read_one_inode(root, location.objectid);
2037 ret = link_to_fixup_dir(trans, root,
2038 path, location.objectid);
2046 ret = btrfs_unlink_inode(trans, root, dir, inode,
2049 ret = btrfs_run_delayed_items(trans, root);
2055 /* there might still be more names under this key
2056 * check and repeat if required
2058 ret = btrfs_search_slot(NULL, root, dir_key, path,
2064 } else if (IS_ERR(log_di)) {
2066 return PTR_ERR(log_di);
2068 btrfs_release_path(log_path);
2071 ptr = (unsigned long)(di + 1);
2076 btrfs_release_path(path);
2077 btrfs_release_path(log_path);
2081 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2082 struct btrfs_root *root,
2083 struct btrfs_root *log,
2084 struct btrfs_path *path,
2087 struct btrfs_key search_key;
2088 struct btrfs_path *log_path;
2093 log_path = btrfs_alloc_path();
2097 search_key.objectid = ino;
2098 search_key.type = BTRFS_XATTR_ITEM_KEY;
2099 search_key.offset = 0;
2101 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2105 nritems = btrfs_header_nritems(path->nodes[0]);
2106 for (i = path->slots[0]; i < nritems; i++) {
2107 struct btrfs_key key;
2108 struct btrfs_dir_item *di;
2109 struct btrfs_dir_item *log_di;
2113 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2114 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2119 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2120 total_size = btrfs_item_size_nr(path->nodes[0], i);
2122 while (cur < total_size) {
2123 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2124 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2125 u32 this_len = sizeof(*di) + name_len + data_len;
2128 name = kmalloc(name_len, GFP_NOFS);
2133 read_extent_buffer(path->nodes[0], name,
2134 (unsigned long)(di + 1), name_len);
2136 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2138 btrfs_release_path(log_path);
2140 /* Doesn't exist in log tree, so delete it. */
2141 btrfs_release_path(path);
2142 di = btrfs_lookup_xattr(trans, root, path, ino,
2143 name, name_len, -1);
2150 ret = btrfs_delete_one_dir_name(trans, root,
2154 btrfs_release_path(path);
2159 if (IS_ERR(log_di)) {
2160 ret = PTR_ERR(log_di);
2164 di = (struct btrfs_dir_item *)((char *)di + this_len);
2167 ret = btrfs_next_leaf(root, path);
2173 btrfs_free_path(log_path);
2174 btrfs_release_path(path);
2180 * deletion replay happens before we copy any new directory items
2181 * out of the log or out of backreferences from inodes. It
2182 * scans the log to find ranges of keys that log is authoritative for,
2183 * and then scans the directory to find items in those ranges that are
2184 * not present in the log.
2186 * Anything we don't find in the log is unlinked and removed from the
2189 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_root *log,
2192 struct btrfs_path *path,
2193 u64 dirid, int del_all)
2197 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2199 struct btrfs_key dir_key;
2200 struct btrfs_key found_key;
2201 struct btrfs_path *log_path;
2204 dir_key.objectid = dirid;
2205 dir_key.type = BTRFS_DIR_ITEM_KEY;
2206 log_path = btrfs_alloc_path();
2210 dir = read_one_inode(root, dirid);
2211 /* it isn't an error if the inode isn't there, that can happen
2212 * because we replay the deletes before we copy in the inode item
2216 btrfs_free_path(log_path);
2224 range_end = (u64)-1;
2226 ret = find_dir_range(log, path, dirid, key_type,
2227 &range_start, &range_end);
2232 dir_key.offset = range_start;
2235 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2240 nritems = btrfs_header_nritems(path->nodes[0]);
2241 if (path->slots[0] >= nritems) {
2242 ret = btrfs_next_leaf(root, path);
2246 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2248 if (found_key.objectid != dirid ||
2249 found_key.type != dir_key.type)
2252 if (found_key.offset > range_end)
2255 ret = check_item_in_log(trans, root, log, path,
2260 if (found_key.offset == (u64)-1)
2262 dir_key.offset = found_key.offset + 1;
2264 btrfs_release_path(path);
2265 if (range_end == (u64)-1)
2267 range_start = range_end + 1;
2272 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2273 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2274 dir_key.type = BTRFS_DIR_INDEX_KEY;
2275 btrfs_release_path(path);
2279 btrfs_release_path(path);
2280 btrfs_free_path(log_path);
2286 * the process_func used to replay items from the log tree. This
2287 * gets called in two different stages. The first stage just looks
2288 * for inodes and makes sure they are all copied into the subvolume.
2290 * The second stage copies all the other item types from the log into
2291 * the subvolume. The two stage approach is slower, but gets rid of
2292 * lots of complexity around inodes referencing other inodes that exist
2293 * only in the log (references come from either directory items or inode
2296 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2297 struct walk_control *wc, u64 gen)
2300 struct btrfs_path *path;
2301 struct btrfs_root *root = wc->replay_dest;
2302 struct btrfs_key key;
2307 ret = btrfs_read_buffer(eb, gen);
2311 level = btrfs_header_level(eb);
2316 path = btrfs_alloc_path();
2320 nritems = btrfs_header_nritems(eb);
2321 for (i = 0; i < nritems; i++) {
2322 btrfs_item_key_to_cpu(eb, &key, i);
2324 /* inode keys are done during the first stage */
2325 if (key.type == BTRFS_INODE_ITEM_KEY &&
2326 wc->stage == LOG_WALK_REPLAY_INODES) {
2327 struct btrfs_inode_item *inode_item;
2330 inode_item = btrfs_item_ptr(eb, i,
2331 struct btrfs_inode_item);
2332 ret = replay_xattr_deletes(wc->trans, root, log,
2333 path, key.objectid);
2336 mode = btrfs_inode_mode(eb, inode_item);
2337 if (S_ISDIR(mode)) {
2338 ret = replay_dir_deletes(wc->trans,
2339 root, log, path, key.objectid, 0);
2343 ret = overwrite_item(wc->trans, root, path,
2348 /* for regular files, make sure corresponding
2349 * orphan item exist. extents past the new EOF
2350 * will be truncated later by orphan cleanup.
2352 if (S_ISREG(mode)) {
2353 ret = insert_orphan_item(wc->trans, root,
2359 ret = link_to_fixup_dir(wc->trans, root,
2360 path, key.objectid);
2365 if (key.type == BTRFS_DIR_INDEX_KEY &&
2366 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2367 ret = replay_one_dir_item(wc->trans, root, path,
2373 if (wc->stage < LOG_WALK_REPLAY_ALL)
2376 /* these keys are simply copied */
2377 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2378 ret = overwrite_item(wc->trans, root, path,
2382 } else if (key.type == BTRFS_INODE_REF_KEY ||
2383 key.type == BTRFS_INODE_EXTREF_KEY) {
2384 ret = add_inode_ref(wc->trans, root, log, path,
2386 if (ret && ret != -ENOENT)
2389 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2390 ret = replay_one_extent(wc->trans, root, path,
2394 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2395 ret = replay_one_dir_item(wc->trans, root, path,
2401 btrfs_free_path(path);
2405 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2406 struct btrfs_root *root,
2407 struct btrfs_path *path, int *level,
2408 struct walk_control *wc)
2413 struct extent_buffer *next;
2414 struct extent_buffer *cur;
2415 struct extent_buffer *parent;
2419 WARN_ON(*level < 0);
2420 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2422 while (*level > 0) {
2423 WARN_ON(*level < 0);
2424 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2425 cur = path->nodes[*level];
2427 WARN_ON(btrfs_header_level(cur) != *level);
2429 if (path->slots[*level] >=
2430 btrfs_header_nritems(cur))
2433 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2434 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2435 blocksize = root->nodesize;
2437 parent = path->nodes[*level];
2438 root_owner = btrfs_header_owner(parent);
2440 next = btrfs_find_create_tree_block(root, bytenr);
2442 return PTR_ERR(next);
2445 ret = wc->process_func(root, next, wc, ptr_gen);
2447 free_extent_buffer(next);
2451 path->slots[*level]++;
2453 ret = btrfs_read_buffer(next, ptr_gen);
2455 free_extent_buffer(next);
2460 btrfs_tree_lock(next);
2461 btrfs_set_lock_blocking(next);
2462 clean_tree_block(trans, root->fs_info,
2464 btrfs_wait_tree_block_writeback(next);
2465 btrfs_tree_unlock(next);
2468 WARN_ON(root_owner !=
2469 BTRFS_TREE_LOG_OBJECTID);
2470 ret = btrfs_free_and_pin_reserved_extent(root,
2473 free_extent_buffer(next);
2477 free_extent_buffer(next);
2480 ret = btrfs_read_buffer(next, ptr_gen);
2482 free_extent_buffer(next);
2486 WARN_ON(*level <= 0);
2487 if (path->nodes[*level-1])
2488 free_extent_buffer(path->nodes[*level-1]);
2489 path->nodes[*level-1] = next;
2490 *level = btrfs_header_level(next);
2491 path->slots[*level] = 0;
2494 WARN_ON(*level < 0);
2495 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2497 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2503 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2504 struct btrfs_root *root,
2505 struct btrfs_path *path, int *level,
2506 struct walk_control *wc)
2513 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2514 slot = path->slots[i];
2515 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2518 WARN_ON(*level == 0);
2521 struct extent_buffer *parent;
2522 if (path->nodes[*level] == root->node)
2523 parent = path->nodes[*level];
2525 parent = path->nodes[*level + 1];
2527 root_owner = btrfs_header_owner(parent);
2528 ret = wc->process_func(root, path->nodes[*level], wc,
2529 btrfs_header_generation(path->nodes[*level]));
2534 struct extent_buffer *next;
2536 next = path->nodes[*level];
2539 btrfs_tree_lock(next);
2540 btrfs_set_lock_blocking(next);
2541 clean_tree_block(trans, root->fs_info,
2543 btrfs_wait_tree_block_writeback(next);
2544 btrfs_tree_unlock(next);
2547 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2548 ret = btrfs_free_and_pin_reserved_extent(root,
2549 path->nodes[*level]->start,
2550 path->nodes[*level]->len);
2554 free_extent_buffer(path->nodes[*level]);
2555 path->nodes[*level] = NULL;
2563 * drop the reference count on the tree rooted at 'snap'. This traverses
2564 * the tree freeing any blocks that have a ref count of zero after being
2567 static int walk_log_tree(struct btrfs_trans_handle *trans,
2568 struct btrfs_root *log, struct walk_control *wc)
2573 struct btrfs_path *path;
2576 path = btrfs_alloc_path();
2580 level = btrfs_header_level(log->node);
2582 path->nodes[level] = log->node;
2583 extent_buffer_get(log->node);
2584 path->slots[level] = 0;
2587 wret = walk_down_log_tree(trans, log, path, &level, wc);
2595 wret = walk_up_log_tree(trans, log, path, &level, wc);
2604 /* was the root node processed? if not, catch it here */
2605 if (path->nodes[orig_level]) {
2606 ret = wc->process_func(log, path->nodes[orig_level], wc,
2607 btrfs_header_generation(path->nodes[orig_level]));
2611 struct extent_buffer *next;
2613 next = path->nodes[orig_level];
2616 btrfs_tree_lock(next);
2617 btrfs_set_lock_blocking(next);
2618 clean_tree_block(trans, log->fs_info, next);
2619 btrfs_wait_tree_block_writeback(next);
2620 btrfs_tree_unlock(next);
2623 WARN_ON(log->root_key.objectid !=
2624 BTRFS_TREE_LOG_OBJECTID);
2625 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2633 btrfs_free_path(path);
2638 * helper function to update the item for a given subvolumes log root
2639 * in the tree of log roots
2641 static int update_log_root(struct btrfs_trans_handle *trans,
2642 struct btrfs_root *log)
2646 if (log->log_transid == 1) {
2647 /* insert root item on the first sync */
2648 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2649 &log->root_key, &log->root_item);
2651 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2652 &log->root_key, &log->root_item);
2657 static void wait_log_commit(struct btrfs_root *root, int transid)
2660 int index = transid % 2;
2663 * we only allow two pending log transactions at a time,
2664 * so we know that if ours is more than 2 older than the
2665 * current transaction, we're done
2668 prepare_to_wait(&root->log_commit_wait[index],
2669 &wait, TASK_UNINTERRUPTIBLE);
2670 mutex_unlock(&root->log_mutex);
2672 if (root->log_transid_committed < transid &&
2673 atomic_read(&root->log_commit[index]))
2676 finish_wait(&root->log_commit_wait[index], &wait);
2677 mutex_lock(&root->log_mutex);
2678 } while (root->log_transid_committed < transid &&
2679 atomic_read(&root->log_commit[index]));
2682 static void wait_for_writer(struct btrfs_root *root)
2686 while (atomic_read(&root->log_writers)) {
2687 prepare_to_wait(&root->log_writer_wait,
2688 &wait, TASK_UNINTERRUPTIBLE);
2689 mutex_unlock(&root->log_mutex);
2690 if (atomic_read(&root->log_writers))
2692 finish_wait(&root->log_writer_wait, &wait);
2693 mutex_lock(&root->log_mutex);
2697 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2698 struct btrfs_log_ctx *ctx)
2703 mutex_lock(&root->log_mutex);
2704 list_del_init(&ctx->list);
2705 mutex_unlock(&root->log_mutex);
2709 * Invoked in log mutex context, or be sure there is no other task which
2710 * can access the list.
2712 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2713 int index, int error)
2715 struct btrfs_log_ctx *ctx;
2716 struct btrfs_log_ctx *safe;
2718 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2719 list_del_init(&ctx->list);
2720 ctx->log_ret = error;
2723 INIT_LIST_HEAD(&root->log_ctxs[index]);
2727 * btrfs_sync_log does sends a given tree log down to the disk and
2728 * updates the super blocks to record it. When this call is done,
2729 * you know that any inodes previously logged are safely on disk only
2732 * Any other return value means you need to call btrfs_commit_transaction.
2733 * Some of the edge cases for fsyncing directories that have had unlinks
2734 * or renames done in the past mean that sometimes the only safe
2735 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2736 * that has happened.
2738 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2739 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2745 struct btrfs_root *log = root->log_root;
2746 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2747 int log_transid = 0;
2748 struct btrfs_log_ctx root_log_ctx;
2749 struct blk_plug plug;
2751 mutex_lock(&root->log_mutex);
2752 log_transid = ctx->log_transid;
2753 if (root->log_transid_committed >= log_transid) {
2754 mutex_unlock(&root->log_mutex);
2755 return ctx->log_ret;
2758 index1 = log_transid % 2;
2759 if (atomic_read(&root->log_commit[index1])) {
2760 wait_log_commit(root, log_transid);
2761 mutex_unlock(&root->log_mutex);
2762 return ctx->log_ret;
2764 ASSERT(log_transid == root->log_transid);
2765 atomic_set(&root->log_commit[index1], 1);
2767 /* wait for previous tree log sync to complete */
2768 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2769 wait_log_commit(root, log_transid - 1);
2772 int batch = atomic_read(&root->log_batch);
2773 /* when we're on an ssd, just kick the log commit out */
2774 if (!btrfs_test_opt(root->fs_info, SSD) &&
2775 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2776 mutex_unlock(&root->log_mutex);
2777 schedule_timeout_uninterruptible(1);
2778 mutex_lock(&root->log_mutex);
2780 wait_for_writer(root);
2781 if (batch == atomic_read(&root->log_batch))
2785 /* bail out if we need to do a full commit */
2786 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2788 btrfs_free_logged_extents(log, log_transid);
2789 mutex_unlock(&root->log_mutex);
2793 if (log_transid % 2 == 0)
2794 mark = EXTENT_DIRTY;
2798 /* we start IO on all the marked extents here, but we don't actually
2799 * wait for them until later.
2801 blk_start_plug(&plug);
2802 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2804 blk_finish_plug(&plug);
2805 btrfs_abort_transaction(trans, ret);
2806 btrfs_free_logged_extents(log, log_transid);
2807 btrfs_set_log_full_commit(root->fs_info, trans);
2808 mutex_unlock(&root->log_mutex);
2812 btrfs_set_root_node(&log->root_item, log->node);
2814 root->log_transid++;
2815 log->log_transid = root->log_transid;
2816 root->log_start_pid = 0;
2818 * IO has been started, blocks of the log tree have WRITTEN flag set
2819 * in their headers. new modifications of the log will be written to
2820 * new positions. so it's safe to allow log writers to go in.
2822 mutex_unlock(&root->log_mutex);
2824 btrfs_init_log_ctx(&root_log_ctx, NULL);
2826 mutex_lock(&log_root_tree->log_mutex);
2827 atomic_inc(&log_root_tree->log_batch);
2828 atomic_inc(&log_root_tree->log_writers);
2830 index2 = log_root_tree->log_transid % 2;
2831 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2832 root_log_ctx.log_transid = log_root_tree->log_transid;
2834 mutex_unlock(&log_root_tree->log_mutex);
2836 ret = update_log_root(trans, log);
2838 mutex_lock(&log_root_tree->log_mutex);
2839 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2841 * Implicit memory barrier after atomic_dec_and_test
2843 if (waitqueue_active(&log_root_tree->log_writer_wait))
2844 wake_up(&log_root_tree->log_writer_wait);
2848 if (!list_empty(&root_log_ctx.list))
2849 list_del_init(&root_log_ctx.list);
2851 blk_finish_plug(&plug);
2852 btrfs_set_log_full_commit(root->fs_info, trans);
2854 if (ret != -ENOSPC) {
2855 btrfs_abort_transaction(trans, ret);
2856 mutex_unlock(&log_root_tree->log_mutex);
2859 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2860 btrfs_free_logged_extents(log, log_transid);
2861 mutex_unlock(&log_root_tree->log_mutex);
2866 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2867 blk_finish_plug(&plug);
2868 list_del_init(&root_log_ctx.list);
2869 mutex_unlock(&log_root_tree->log_mutex);
2870 ret = root_log_ctx.log_ret;
2874 index2 = root_log_ctx.log_transid % 2;
2875 if (atomic_read(&log_root_tree->log_commit[index2])) {
2876 blk_finish_plug(&plug);
2877 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2879 btrfs_wait_logged_extents(trans, log, log_transid);
2880 wait_log_commit(log_root_tree,
2881 root_log_ctx.log_transid);
2882 mutex_unlock(&log_root_tree->log_mutex);
2884 ret = root_log_ctx.log_ret;
2887 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2888 atomic_set(&log_root_tree->log_commit[index2], 1);
2890 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2891 wait_log_commit(log_root_tree,
2892 root_log_ctx.log_transid - 1);
2895 wait_for_writer(log_root_tree);
2898 * now that we've moved on to the tree of log tree roots,
2899 * check the full commit flag again
2901 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2902 blk_finish_plug(&plug);
2903 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2904 btrfs_free_logged_extents(log, log_transid);
2905 mutex_unlock(&log_root_tree->log_mutex);
2907 goto out_wake_log_root;
2910 ret = btrfs_write_marked_extents(log_root_tree,
2911 &log_root_tree->dirty_log_pages,
2912 EXTENT_DIRTY | EXTENT_NEW);
2913 blk_finish_plug(&plug);
2915 btrfs_set_log_full_commit(root->fs_info, trans);
2916 btrfs_abort_transaction(trans, ret);
2917 btrfs_free_logged_extents(log, log_transid);
2918 mutex_unlock(&log_root_tree->log_mutex);
2919 goto out_wake_log_root;
2921 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2923 ret = btrfs_wait_marked_extents(log_root_tree,
2924 &log_root_tree->dirty_log_pages,
2925 EXTENT_NEW | EXTENT_DIRTY);
2927 btrfs_set_log_full_commit(root->fs_info, trans);
2928 btrfs_free_logged_extents(log, log_transid);
2929 mutex_unlock(&log_root_tree->log_mutex);
2930 goto out_wake_log_root;
2932 btrfs_wait_logged_extents(trans, log, log_transid);
2934 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2935 log_root_tree->node->start);
2936 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2937 btrfs_header_level(log_root_tree->node));
2939 log_root_tree->log_transid++;
2940 mutex_unlock(&log_root_tree->log_mutex);
2943 * nobody else is going to jump in and write the the ctree
2944 * super here because the log_commit atomic below is protecting
2945 * us. We must be called with a transaction handle pinning
2946 * the running transaction open, so a full commit can't hop
2947 * in and cause problems either.
2949 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2951 btrfs_set_log_full_commit(root->fs_info, trans);
2952 btrfs_abort_transaction(trans, ret);
2953 goto out_wake_log_root;
2956 mutex_lock(&root->log_mutex);
2957 if (root->last_log_commit < log_transid)
2958 root->last_log_commit = log_transid;
2959 mutex_unlock(&root->log_mutex);
2962 mutex_lock(&log_root_tree->log_mutex);
2963 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2965 log_root_tree->log_transid_committed++;
2966 atomic_set(&log_root_tree->log_commit[index2], 0);
2967 mutex_unlock(&log_root_tree->log_mutex);
2970 * The barrier before waitqueue_active is implied by mutex_unlock
2972 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2973 wake_up(&log_root_tree->log_commit_wait[index2]);
2975 mutex_lock(&root->log_mutex);
2976 btrfs_remove_all_log_ctxs(root, index1, ret);
2977 root->log_transid_committed++;
2978 atomic_set(&root->log_commit[index1], 0);
2979 mutex_unlock(&root->log_mutex);
2982 * The barrier before waitqueue_active is implied by mutex_unlock
2984 if (waitqueue_active(&root->log_commit_wait[index1]))
2985 wake_up(&root->log_commit_wait[index1]);
2989 static void free_log_tree(struct btrfs_trans_handle *trans,
2990 struct btrfs_root *log)
2995 struct walk_control wc = {
2997 .process_func = process_one_buffer
3000 ret = walk_log_tree(trans, log, &wc);
3001 /* I don't think this can happen but just in case */
3003 btrfs_abort_transaction(trans, ret);
3006 ret = find_first_extent_bit(&log->dirty_log_pages,
3007 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3012 clear_extent_bits(&log->dirty_log_pages, start, end,
3013 EXTENT_DIRTY | EXTENT_NEW);
3017 * We may have short-circuited the log tree with the full commit logic
3018 * and left ordered extents on our list, so clear these out to keep us
3019 * from leaking inodes and memory.
3021 btrfs_free_logged_extents(log, 0);
3022 btrfs_free_logged_extents(log, 1);
3024 free_extent_buffer(log->node);
3029 * free all the extents used by the tree log. This should be called
3030 * at commit time of the full transaction
3032 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3034 if (root->log_root) {
3035 free_log_tree(trans, root->log_root);
3036 root->log_root = NULL;
3041 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3042 struct btrfs_fs_info *fs_info)
3044 if (fs_info->log_root_tree) {
3045 free_log_tree(trans, fs_info->log_root_tree);
3046 fs_info->log_root_tree = NULL;
3052 * If both a file and directory are logged, and unlinks or renames are
3053 * mixed in, we have a few interesting corners:
3055 * create file X in dir Y
3056 * link file X to X.link in dir Y
3058 * unlink file X but leave X.link
3061 * After a crash we would expect only X.link to exist. But file X
3062 * didn't get fsync'd again so the log has back refs for X and X.link.
3064 * We solve this by removing directory entries and inode backrefs from the
3065 * log when a file that was logged in the current transaction is
3066 * unlinked. Any later fsync will include the updated log entries, and
3067 * we'll be able to reconstruct the proper directory items from backrefs.
3069 * This optimizations allows us to avoid relogging the entire inode
3070 * or the entire directory.
3072 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3073 struct btrfs_root *root,
3074 const char *name, int name_len,
3075 struct inode *dir, u64 index)
3077 struct btrfs_root *log;
3078 struct btrfs_dir_item *di;
3079 struct btrfs_path *path;
3083 u64 dir_ino = btrfs_ino(dir);
3085 if (BTRFS_I(dir)->logged_trans < trans->transid)
3088 ret = join_running_log_trans(root);
3092 mutex_lock(&BTRFS_I(dir)->log_mutex);
3094 log = root->log_root;
3095 path = btrfs_alloc_path();
3101 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3102 name, name_len, -1);
3108 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3109 bytes_del += name_len;
3115 btrfs_release_path(path);
3116 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3117 index, name, name_len, -1);
3123 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3124 bytes_del += name_len;
3131 /* update the directory size in the log to reflect the names
3135 struct btrfs_key key;
3137 key.objectid = dir_ino;
3139 key.type = BTRFS_INODE_ITEM_KEY;
3140 btrfs_release_path(path);
3142 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3148 struct btrfs_inode_item *item;
3151 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3152 struct btrfs_inode_item);
3153 i_size = btrfs_inode_size(path->nodes[0], item);
3154 if (i_size > bytes_del)
3155 i_size -= bytes_del;
3158 btrfs_set_inode_size(path->nodes[0], item, i_size);
3159 btrfs_mark_buffer_dirty(path->nodes[0]);
3162 btrfs_release_path(path);
3165 btrfs_free_path(path);
3167 mutex_unlock(&BTRFS_I(dir)->log_mutex);
3168 if (ret == -ENOSPC) {
3169 btrfs_set_log_full_commit(root->fs_info, trans);
3172 btrfs_abort_transaction(trans, ret);
3174 btrfs_end_log_trans(root);
3179 /* see comments for btrfs_del_dir_entries_in_log */
3180 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3181 struct btrfs_root *root,
3182 const char *name, int name_len,
3183 struct inode *inode, u64 dirid)
3185 struct btrfs_root *log;
3189 if (BTRFS_I(inode)->logged_trans < trans->transid)
3192 ret = join_running_log_trans(root);
3195 log = root->log_root;
3196 mutex_lock(&BTRFS_I(inode)->log_mutex);
3198 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3200 mutex_unlock(&BTRFS_I(inode)->log_mutex);
3201 if (ret == -ENOSPC) {
3202 btrfs_set_log_full_commit(root->fs_info, trans);
3204 } else if (ret < 0 && ret != -ENOENT)
3205 btrfs_abort_transaction(trans, ret);
3206 btrfs_end_log_trans(root);
3212 * creates a range item in the log for 'dirid'. first_offset and
3213 * last_offset tell us which parts of the key space the log should
3214 * be considered authoritative for.
3216 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *log,
3218 struct btrfs_path *path,
3219 int key_type, u64 dirid,
3220 u64 first_offset, u64 last_offset)
3223 struct btrfs_key key;
3224 struct btrfs_dir_log_item *item;
3226 key.objectid = dirid;
3227 key.offset = first_offset;
3228 if (key_type == BTRFS_DIR_ITEM_KEY)
3229 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3231 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3232 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3236 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3237 struct btrfs_dir_log_item);
3238 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3239 btrfs_mark_buffer_dirty(path->nodes[0]);
3240 btrfs_release_path(path);
3245 * log all the items included in the current transaction for a given
3246 * directory. This also creates the range items in the log tree required
3247 * to replay anything deleted before the fsync
3249 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3250 struct btrfs_root *root, struct inode *inode,
3251 struct btrfs_path *path,
3252 struct btrfs_path *dst_path, int key_type,
3253 struct btrfs_log_ctx *ctx,
3254 u64 min_offset, u64 *last_offset_ret)
3256 struct btrfs_key min_key;
3257 struct btrfs_root *log = root->log_root;
3258 struct extent_buffer *src;
3263 u64 first_offset = min_offset;
3264 u64 last_offset = (u64)-1;
3265 u64 ino = btrfs_ino(inode);
3267 log = root->log_root;
3269 min_key.objectid = ino;
3270 min_key.type = key_type;
3271 min_key.offset = min_offset;
3273 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3276 * we didn't find anything from this transaction, see if there
3277 * is anything at all
3279 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3280 min_key.objectid = ino;
3281 min_key.type = key_type;
3282 min_key.offset = (u64)-1;
3283 btrfs_release_path(path);
3284 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3286 btrfs_release_path(path);
3289 ret = btrfs_previous_item(root, path, ino, key_type);
3291 /* if ret == 0 there are items for this type,
3292 * create a range to tell us the last key of this type.
3293 * otherwise, there are no items in this directory after
3294 * *min_offset, and we create a range to indicate that.
3297 struct btrfs_key tmp;
3298 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3300 if (key_type == tmp.type)
3301 first_offset = max(min_offset, tmp.offset) + 1;
3306 /* go backward to find any previous key */
3307 ret = btrfs_previous_item(root, path, ino, key_type);
3309 struct btrfs_key tmp;
3310 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3311 if (key_type == tmp.type) {
3312 first_offset = tmp.offset;
3313 ret = overwrite_item(trans, log, dst_path,
3314 path->nodes[0], path->slots[0],
3322 btrfs_release_path(path);
3324 /* find the first key from this transaction again */
3325 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3326 if (WARN_ON(ret != 0))
3330 * we have a block from this transaction, log every item in it
3331 * from our directory
3334 struct btrfs_key tmp;
3335 src = path->nodes[0];
3336 nritems = btrfs_header_nritems(src);
3337 for (i = path->slots[0]; i < nritems; i++) {
3338 struct btrfs_dir_item *di;
3340 btrfs_item_key_to_cpu(src, &min_key, i);
3342 if (min_key.objectid != ino || min_key.type != key_type)
3344 ret = overwrite_item(trans, log, dst_path, src, i,
3352 * We must make sure that when we log a directory entry,
3353 * the corresponding inode, after log replay, has a
3354 * matching link count. For example:
3360 * xfs_io -c "fsync" mydir
3362 * <mount fs and log replay>
3364 * Would result in a fsync log that when replayed, our
3365 * file inode would have a link count of 1, but we get
3366 * two directory entries pointing to the same inode.
3367 * After removing one of the names, it would not be
3368 * possible to remove the other name, which resulted
3369 * always in stale file handle errors, and would not
3370 * be possible to rmdir the parent directory, since
3371 * its i_size could never decrement to the value
3372 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3374 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3375 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3377 (btrfs_dir_transid(src, di) == trans->transid ||
3378 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3379 tmp.type != BTRFS_ROOT_ITEM_KEY)
3380 ctx->log_new_dentries = true;
3382 path->slots[0] = nritems;
3385 * look ahead to the next item and see if it is also
3386 * from this directory and from this transaction
3388 ret = btrfs_next_leaf(root, path);
3390 last_offset = (u64)-1;
3393 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3394 if (tmp.objectid != ino || tmp.type != key_type) {
3395 last_offset = (u64)-1;
3398 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3399 ret = overwrite_item(trans, log, dst_path,
3400 path->nodes[0], path->slots[0],
3405 last_offset = tmp.offset;
3410 btrfs_release_path(path);
3411 btrfs_release_path(dst_path);
3414 *last_offset_ret = last_offset;
3416 * insert the log range keys to indicate where the log
3419 ret = insert_dir_log_key(trans, log, path, key_type,
3420 ino, first_offset, last_offset);
3428 * logging directories is very similar to logging inodes, We find all the items
3429 * from the current transaction and write them to the log.
3431 * The recovery code scans the directory in the subvolume, and if it finds a
3432 * key in the range logged that is not present in the log tree, then it means
3433 * that dir entry was unlinked during the transaction.
3435 * In order for that scan to work, we must include one key smaller than
3436 * the smallest logged by this transaction and one key larger than the largest
3437 * key logged by this transaction.
3439 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3440 struct btrfs_root *root, struct inode *inode,
3441 struct btrfs_path *path,
3442 struct btrfs_path *dst_path,
3443 struct btrfs_log_ctx *ctx)
3448 int key_type = BTRFS_DIR_ITEM_KEY;
3454 ret = log_dir_items(trans, root, inode, path,
3455 dst_path, key_type, ctx, min_key,
3459 if (max_key == (u64)-1)
3461 min_key = max_key + 1;
3464 if (key_type == BTRFS_DIR_ITEM_KEY) {
3465 key_type = BTRFS_DIR_INDEX_KEY;
3472 * a helper function to drop items from the log before we relog an
3473 * inode. max_key_type indicates the highest item type to remove.
3474 * This cannot be run for file data extents because it does not
3475 * free the extents they point to.
3477 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3478 struct btrfs_root *log,
3479 struct btrfs_path *path,
3480 u64 objectid, int max_key_type)
3483 struct btrfs_key key;
3484 struct btrfs_key found_key;
3487 key.objectid = objectid;
3488 key.type = max_key_type;
3489 key.offset = (u64)-1;
3492 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3493 BUG_ON(ret == 0); /* Logic error */
3497 if (path->slots[0] == 0)
3501 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3504 if (found_key.objectid != objectid)
3507 found_key.offset = 0;
3509 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3512 ret = btrfs_del_items(trans, log, path, start_slot,
3513 path->slots[0] - start_slot + 1);
3515 * If start slot isn't 0 then we don't need to re-search, we've
3516 * found the last guy with the objectid in this tree.
3518 if (ret || start_slot != 0)
3520 btrfs_release_path(path);
3522 btrfs_release_path(path);
3528 static void fill_inode_item(struct btrfs_trans_handle *trans,
3529 struct extent_buffer *leaf,
3530 struct btrfs_inode_item *item,
3531 struct inode *inode, int log_inode_only,
3534 struct btrfs_map_token token;
3536 btrfs_init_map_token(&token);
3538 if (log_inode_only) {
3539 /* set the generation to zero so the recover code
3540 * can tell the difference between an logging
3541 * just to say 'this inode exists' and a logging
3542 * to say 'update this inode with these values'
3544 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3545 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3547 btrfs_set_token_inode_generation(leaf, item,
3548 BTRFS_I(inode)->generation,
3550 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3553 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3554 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3555 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3556 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3558 btrfs_set_token_timespec_sec(leaf, &item->atime,
3559 inode->i_atime.tv_sec, &token);
3560 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3561 inode->i_atime.tv_nsec, &token);
3563 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3564 inode->i_mtime.tv_sec, &token);
3565 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3566 inode->i_mtime.tv_nsec, &token);
3568 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3569 inode->i_ctime.tv_sec, &token);
3570 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3571 inode->i_ctime.tv_nsec, &token);
3573 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3576 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3577 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3578 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3579 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3580 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3583 static int log_inode_item(struct btrfs_trans_handle *trans,
3584 struct btrfs_root *log, struct btrfs_path *path,
3585 struct inode *inode)
3587 struct btrfs_inode_item *inode_item;
3590 ret = btrfs_insert_empty_item(trans, log, path,
3591 &BTRFS_I(inode)->location,
3592 sizeof(*inode_item));
3593 if (ret && ret != -EEXIST)
3595 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3596 struct btrfs_inode_item);
3597 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0, 0);
3598 btrfs_release_path(path);
3602 static noinline int copy_items(struct btrfs_trans_handle *trans,
3603 struct inode *inode,
3604 struct btrfs_path *dst_path,
3605 struct btrfs_path *src_path, u64 *last_extent,
3606 int start_slot, int nr, int inode_only,
3609 unsigned long src_offset;
3610 unsigned long dst_offset;
3611 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3612 struct btrfs_file_extent_item *extent;
3613 struct btrfs_inode_item *inode_item;
3614 struct extent_buffer *src = src_path->nodes[0];
3615 struct btrfs_key first_key, last_key, key;
3617 struct btrfs_key *ins_keys;
3621 struct list_head ordered_sums;
3622 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3623 bool has_extents = false;
3624 bool need_find_last_extent = true;
3627 INIT_LIST_HEAD(&ordered_sums);
3629 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3630 nr * sizeof(u32), GFP_NOFS);
3634 first_key.objectid = (u64)-1;
3636 ins_sizes = (u32 *)ins_data;
3637 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3639 for (i = 0; i < nr; i++) {
3640 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3641 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3643 ret = btrfs_insert_empty_items(trans, log, dst_path,
3644 ins_keys, ins_sizes, nr);
3650 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3651 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3652 dst_path->slots[0]);
3654 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3656 if ((i == (nr - 1)))
3657 last_key = ins_keys[i];
3659 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3660 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3662 struct btrfs_inode_item);
3663 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3664 inode, inode_only == LOG_INODE_EXISTS,
3667 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3668 src_offset, ins_sizes[i]);
3672 * We set need_find_last_extent here in case we know we were
3673 * processing other items and then walk into the first extent in
3674 * the inode. If we don't hit an extent then nothing changes,
3675 * we'll do the last search the next time around.
3677 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3679 if (first_key.objectid == (u64)-1)
3680 first_key = ins_keys[i];
3682 need_find_last_extent = false;
3685 /* take a reference on file data extents so that truncates
3686 * or deletes of this inode don't have to relog the inode
3689 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3692 extent = btrfs_item_ptr(src, start_slot + i,
3693 struct btrfs_file_extent_item);
3695 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3698 found_type = btrfs_file_extent_type(src, extent);
3699 if (found_type == BTRFS_FILE_EXTENT_REG) {
3701 ds = btrfs_file_extent_disk_bytenr(src,
3703 /* ds == 0 is a hole */
3707 dl = btrfs_file_extent_disk_num_bytes(src,
3709 cs = btrfs_file_extent_offset(src, extent);
3710 cl = btrfs_file_extent_num_bytes(src,
3712 if (btrfs_file_extent_compression(src,
3718 ret = btrfs_lookup_csums_range(
3719 log->fs_info->csum_root,
3720 ds + cs, ds + cs + cl - 1,
3723 btrfs_release_path(dst_path);
3731 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3732 btrfs_release_path(dst_path);
3736 * we have to do this after the loop above to avoid changing the
3737 * log tree while trying to change the log tree.
3740 while (!list_empty(&ordered_sums)) {
3741 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3742 struct btrfs_ordered_sum,
3745 ret = btrfs_csum_file_blocks(trans, log, sums);
3746 list_del(&sums->list);
3753 if (need_find_last_extent && *last_extent == first_key.offset) {
3755 * We don't have any leafs between our current one and the one
3756 * we processed before that can have file extent items for our
3757 * inode (and have a generation number smaller than our current
3760 need_find_last_extent = false;
3764 * Because we use btrfs_search_forward we could skip leaves that were
3765 * not modified and then assume *last_extent is valid when it really
3766 * isn't. So back up to the previous leaf and read the end of the last
3767 * extent before we go and fill in holes.
3769 if (need_find_last_extent) {
3772 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3777 if (src_path->slots[0])
3778 src_path->slots[0]--;
3779 src = src_path->nodes[0];
3780 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3781 if (key.objectid != btrfs_ino(inode) ||
3782 key.type != BTRFS_EXTENT_DATA_KEY)
3784 extent = btrfs_item_ptr(src, src_path->slots[0],
3785 struct btrfs_file_extent_item);
3786 if (btrfs_file_extent_type(src, extent) ==
3787 BTRFS_FILE_EXTENT_INLINE) {
3788 len = btrfs_file_extent_inline_len(src,
3791 *last_extent = ALIGN(key.offset + len,
3794 len = btrfs_file_extent_num_bytes(src, extent);
3795 *last_extent = key.offset + len;
3799 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3800 * things could have happened
3802 * 1) A merge could have happened, so we could currently be on a leaf
3803 * that holds what we were copying in the first place.
3804 * 2) A split could have happened, and now not all of the items we want
3805 * are on the same leaf.
3807 * So we need to adjust how we search for holes, we need to drop the
3808 * path and re-search for the first extent key we found, and then walk
3809 * forward until we hit the last one we copied.
3811 if (need_find_last_extent) {
3812 /* btrfs_prev_leaf could return 1 without releasing the path */
3813 btrfs_release_path(src_path);
3814 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3819 src = src_path->nodes[0];
3820 i = src_path->slots[0];
3826 * Ok so here we need to go through and fill in any holes we may have
3827 * to make sure that holes are punched for those areas in case they had
3828 * extents previously.
3834 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3835 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3839 src = src_path->nodes[0];
3843 btrfs_item_key_to_cpu(src, &key, i);
3844 if (!btrfs_comp_cpu_keys(&key, &last_key))
3846 if (key.objectid != btrfs_ino(inode) ||
3847 key.type != BTRFS_EXTENT_DATA_KEY) {
3851 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3852 if (btrfs_file_extent_type(src, extent) ==
3853 BTRFS_FILE_EXTENT_INLINE) {
3854 len = btrfs_file_extent_inline_len(src, i, extent);
3855 extent_end = ALIGN(key.offset + len, log->sectorsize);
3857 len = btrfs_file_extent_num_bytes(src, extent);
3858 extent_end = key.offset + len;
3862 if (*last_extent == key.offset) {
3863 *last_extent = extent_end;
3866 offset = *last_extent;
3867 len = key.offset - *last_extent;
3868 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3869 offset, 0, 0, len, 0, len, 0,
3873 *last_extent = extent_end;
3876 * Need to let the callers know we dropped the path so they should
3879 if (!ret && need_find_last_extent)
3884 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3886 struct extent_map *em1, *em2;
3888 em1 = list_entry(a, struct extent_map, list);
3889 em2 = list_entry(b, struct extent_map, list);
3891 if (em1->start < em2->start)
3893 else if (em1->start > em2->start)
3898 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3899 struct inode *inode,
3900 struct btrfs_root *root,
3901 const struct extent_map *em,
3902 const struct list_head *logged_list,
3903 bool *ordered_io_error)
3905 struct btrfs_ordered_extent *ordered;
3906 struct btrfs_root *log = root->log_root;
3907 u64 mod_start = em->mod_start;
3908 u64 mod_len = em->mod_len;
3909 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3912 LIST_HEAD(ordered_sums);
3915 *ordered_io_error = false;
3917 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3918 em->block_start == EXTENT_MAP_HOLE)
3922 * Wait far any ordered extent that covers our extent map. If it
3923 * finishes without an error, first check and see if our csums are on
3924 * our outstanding ordered extents.
3926 list_for_each_entry(ordered, logged_list, log_list) {
3927 struct btrfs_ordered_sum *sum;
3932 if (ordered->file_offset + ordered->len <= mod_start ||
3933 mod_start + mod_len <= ordered->file_offset)
3936 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3937 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3938 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3939 const u64 start = ordered->file_offset;
3940 const u64 end = ordered->file_offset + ordered->len - 1;
3942 WARN_ON(ordered->inode != inode);
3943 filemap_fdatawrite_range(inode->i_mapping, start, end);
3946 wait_event(ordered->wait,
3947 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3948 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3950 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3952 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3953 * i_mapping flags, so that the next fsync won't get
3954 * an outdated io error too.
3956 filemap_check_errors(inode->i_mapping);
3957 *ordered_io_error = true;
3961 * We are going to copy all the csums on this ordered extent, so
3962 * go ahead and adjust mod_start and mod_len in case this
3963 * ordered extent has already been logged.
3965 if (ordered->file_offset > mod_start) {
3966 if (ordered->file_offset + ordered->len >=
3967 mod_start + mod_len)
3968 mod_len = ordered->file_offset - mod_start;
3970 * If we have this case
3972 * |--------- logged extent ---------|
3973 * |----- ordered extent ----|
3975 * Just don't mess with mod_start and mod_len, we'll
3976 * just end up logging more csums than we need and it
3980 if (ordered->file_offset + ordered->len <
3981 mod_start + mod_len) {
3982 mod_len = (mod_start + mod_len) -
3983 (ordered->file_offset + ordered->len);
3984 mod_start = ordered->file_offset +
3995 * To keep us from looping for the above case of an ordered
3996 * extent that falls inside of the logged extent.
3998 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4002 list_for_each_entry(sum, &ordered->list, list) {
4003 ret = btrfs_csum_file_blocks(trans, log, sum);
4009 if (*ordered_io_error || !mod_len || ret || skip_csum)
4012 if (em->compress_type) {
4014 csum_len = max(em->block_len, em->orig_block_len);
4016 csum_offset = mod_start - em->start;
4020 /* block start is already adjusted for the file extent offset. */
4021 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
4022 em->block_start + csum_offset,
4023 em->block_start + csum_offset +
4024 csum_len - 1, &ordered_sums, 0);
4028 while (!list_empty(&ordered_sums)) {
4029 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4030 struct btrfs_ordered_sum,
4033 ret = btrfs_csum_file_blocks(trans, log, sums);
4034 list_del(&sums->list);
4041 static int log_one_extent(struct btrfs_trans_handle *trans,
4042 struct inode *inode, struct btrfs_root *root,
4043 const struct extent_map *em,
4044 struct btrfs_path *path,
4045 const struct list_head *logged_list,
4046 struct btrfs_log_ctx *ctx)
4048 struct btrfs_root *log = root->log_root;
4049 struct btrfs_file_extent_item *fi;
4050 struct extent_buffer *leaf;
4051 struct btrfs_map_token token;
4052 struct btrfs_key key;
4053 u64 extent_offset = em->start - em->orig_start;
4056 int extent_inserted = 0;
4057 bool ordered_io_err = false;
4059 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
4064 if (ordered_io_err) {
4069 btrfs_init_map_token(&token);
4071 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
4072 em->start + em->len, NULL, 0, 1,
4073 sizeof(*fi), &extent_inserted);
4077 if (!extent_inserted) {
4078 key.objectid = btrfs_ino(inode);
4079 key.type = BTRFS_EXTENT_DATA_KEY;
4080 key.offset = em->start;
4082 ret = btrfs_insert_empty_item(trans, log, path, &key,
4087 leaf = path->nodes[0];
4088 fi = btrfs_item_ptr(leaf, path->slots[0],
4089 struct btrfs_file_extent_item);
4091 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4093 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4094 btrfs_set_token_file_extent_type(leaf, fi,
4095 BTRFS_FILE_EXTENT_PREALLOC,
4098 btrfs_set_token_file_extent_type(leaf, fi,
4099 BTRFS_FILE_EXTENT_REG,
4102 block_len = max(em->block_len, em->orig_block_len);
4103 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4104 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4107 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4109 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4110 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4112 extent_offset, &token);
4113 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4116 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4117 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4121 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4122 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4123 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4124 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4126 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4127 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4128 btrfs_mark_buffer_dirty(leaf);
4130 btrfs_release_path(path);
4135 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4136 struct btrfs_root *root,
4137 struct inode *inode,
4138 struct btrfs_path *path,
4139 struct list_head *logged_list,
4140 struct btrfs_log_ctx *ctx,
4144 struct extent_map *em, *n;
4145 struct list_head extents;
4146 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
4151 INIT_LIST_HEAD(&extents);
4153 down_write(&BTRFS_I(inode)->dio_sem);
4154 write_lock(&tree->lock);
4155 test_gen = root->fs_info->last_trans_committed;
4157 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4158 list_del_init(&em->list);
4161 * Just an arbitrary number, this can be really CPU intensive
4162 * once we start getting a lot of extents, and really once we
4163 * have a bunch of extents we just want to commit since it will
4166 if (++num > 32768) {
4167 list_del_init(&tree->modified_extents);
4172 if (em->generation <= test_gen)
4174 /* Need a ref to keep it from getting evicted from cache */
4175 atomic_inc(&em->refs);
4176 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4177 list_add_tail(&em->list, &extents);
4181 list_sort(NULL, &extents, extent_cmp);
4182 btrfs_get_logged_extents(inode, logged_list, start, end);
4184 * Some ordered extents started by fsync might have completed
4185 * before we could collect them into the list logged_list, which
4186 * means they're gone, not in our logged_list nor in the inode's
4187 * ordered tree. We want the application/user space to know an
4188 * error happened while attempting to persist file data so that
4189 * it can take proper action. If such error happened, we leave
4190 * without writing to the log tree and the fsync must report the
4191 * file data write error and not commit the current transaction.
4193 ret = filemap_check_errors(inode->i_mapping);
4197 while (!list_empty(&extents)) {
4198 em = list_entry(extents.next, struct extent_map, list);
4200 list_del_init(&em->list);
4203 * If we had an error we just need to delete everybody from our
4207 clear_em_logging(tree, em);
4208 free_extent_map(em);
4212 write_unlock(&tree->lock);
4214 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4216 write_lock(&tree->lock);
4217 clear_em_logging(tree, em);
4218 free_extent_map(em);
4220 WARN_ON(!list_empty(&extents));
4221 write_unlock(&tree->lock);
4222 up_write(&BTRFS_I(inode)->dio_sem);
4224 btrfs_release_path(path);
4228 static int logged_inode_size(struct btrfs_root *log, struct inode *inode,
4229 struct btrfs_path *path, u64 *size_ret)
4231 struct btrfs_key key;
4234 key.objectid = btrfs_ino(inode);
4235 key.type = BTRFS_INODE_ITEM_KEY;
4238 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4241 } else if (ret > 0) {
4244 struct btrfs_inode_item *item;
4246 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4247 struct btrfs_inode_item);
4248 *size_ret = btrfs_inode_size(path->nodes[0], item);
4251 btrfs_release_path(path);
4256 * At the moment we always log all xattrs. This is to figure out at log replay
4257 * time which xattrs must have their deletion replayed. If a xattr is missing
4258 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4259 * because if a xattr is deleted, the inode is fsynced and a power failure
4260 * happens, causing the log to be replayed the next time the fs is mounted,
4261 * we want the xattr to not exist anymore (same behaviour as other filesystems
4262 * with a journal, ext3/4, xfs, f2fs, etc).
4264 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4265 struct btrfs_root *root,
4266 struct inode *inode,
4267 struct btrfs_path *path,
4268 struct btrfs_path *dst_path)
4271 struct btrfs_key key;
4272 const u64 ino = btrfs_ino(inode);
4277 key.type = BTRFS_XATTR_ITEM_KEY;
4280 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4285 int slot = path->slots[0];
4286 struct extent_buffer *leaf = path->nodes[0];
4287 int nritems = btrfs_header_nritems(leaf);
4289 if (slot >= nritems) {
4291 u64 last_extent = 0;
4293 ret = copy_items(trans, inode, dst_path, path,
4294 &last_extent, start_slot,
4296 /* can't be 1, extent items aren't processed */
4302 ret = btrfs_next_leaf(root, path);
4310 btrfs_item_key_to_cpu(leaf, &key, slot);
4311 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4321 u64 last_extent = 0;
4323 ret = copy_items(trans, inode, dst_path, path,
4324 &last_extent, start_slot,
4326 /* can't be 1, extent items aren't processed */
4336 * If the no holes feature is enabled we need to make sure any hole between the
4337 * last extent and the i_size of our inode is explicitly marked in the log. This
4338 * is to make sure that doing something like:
4340 * 1) create file with 128Kb of data
4341 * 2) truncate file to 64Kb
4342 * 3) truncate file to 256Kb
4344 * 5) <crash/power failure>
4345 * 6) mount fs and trigger log replay
4347 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4348 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4349 * file correspond to a hole. The presence of explicit holes in a log tree is
4350 * what guarantees that log replay will remove/adjust file extent items in the
4353 * Here we do not need to care about holes between extents, that is already done
4354 * by copy_items(). We also only need to do this in the full sync path, where we
4355 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4356 * lookup the list of modified extent maps and if any represents a hole, we
4357 * insert a corresponding extent representing a hole in the log tree.
4359 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4360 struct btrfs_root *root,
4361 struct inode *inode,
4362 struct btrfs_path *path)
4365 struct btrfs_key key;
4368 struct extent_buffer *leaf;
4369 struct btrfs_root *log = root->log_root;
4370 const u64 ino = btrfs_ino(inode);
4371 const u64 i_size = i_size_read(inode);
4373 if (!btrfs_fs_incompat(root->fs_info, NO_HOLES))
4377 key.type = BTRFS_EXTENT_DATA_KEY;
4378 key.offset = (u64)-1;
4380 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4385 ASSERT(path->slots[0] > 0);
4387 leaf = path->nodes[0];
4388 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4390 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4391 /* inode does not have any extents */
4395 struct btrfs_file_extent_item *extent;
4399 * If there's an extent beyond i_size, an explicit hole was
4400 * already inserted by copy_items().
4402 if (key.offset >= i_size)
4405 extent = btrfs_item_ptr(leaf, path->slots[0],
4406 struct btrfs_file_extent_item);
4408 if (btrfs_file_extent_type(leaf, extent) ==
4409 BTRFS_FILE_EXTENT_INLINE) {
4410 len = btrfs_file_extent_inline_len(leaf,
4413 ASSERT(len == i_size);
4417 len = btrfs_file_extent_num_bytes(leaf, extent);
4418 /* Last extent goes beyond i_size, no need to log a hole. */
4419 if (key.offset + len > i_size)
4421 hole_start = key.offset + len;
4422 hole_size = i_size - hole_start;
4424 btrfs_release_path(path);
4426 /* Last extent ends at i_size. */
4430 hole_size = ALIGN(hole_size, root->sectorsize);
4431 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4432 hole_size, 0, hole_size, 0, 0, 0);
4437 * When we are logging a new inode X, check if it doesn't have a reference that
4438 * matches the reference from some other inode Y created in a past transaction
4439 * and that was renamed in the current transaction. If we don't do this, then at
4440 * log replay time we can lose inode Y (and all its files if it's a directory):
4443 * echo "hello world" > /mnt/x/foobar
4446 * mkdir /mnt/x # or touch /mnt/x
4447 * xfs_io -c fsync /mnt/x
4449 * mount fs, trigger log replay
4451 * After the log replay procedure, we would lose the first directory and all its
4452 * files (file foobar).
4453 * For the case where inode Y is not a directory we simply end up losing it:
4455 * echo "123" > /mnt/foo
4457 * mv /mnt/foo /mnt/bar
4458 * echo "abc" > /mnt/foo
4459 * xfs_io -c fsync /mnt/foo
4462 * We also need this for cases where a snapshot entry is replaced by some other
4463 * entry (file or directory) otherwise we end up with an unreplayable log due to
4464 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4465 * if it were a regular entry:
4468 * btrfs subvolume snapshot /mnt /mnt/x/snap
4469 * btrfs subvolume delete /mnt/x/snap
4472 * fsync /mnt/x or fsync some new file inside it
4475 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4476 * the same transaction.
4478 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4480 const struct btrfs_key *key,
4481 struct inode *inode,
4485 struct btrfs_path *search_path;
4488 u32 item_size = btrfs_item_size_nr(eb, slot);
4490 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4492 search_path = btrfs_alloc_path();
4495 search_path->search_commit_root = 1;
4496 search_path->skip_locking = 1;
4498 while (cur_offset < item_size) {
4502 unsigned long name_ptr;
4503 struct btrfs_dir_item *di;
4505 if (key->type == BTRFS_INODE_REF_KEY) {
4506 struct btrfs_inode_ref *iref;
4508 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4509 parent = key->offset;
4510 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4511 name_ptr = (unsigned long)(iref + 1);
4512 this_len = sizeof(*iref) + this_name_len;
4514 struct btrfs_inode_extref *extref;
4516 extref = (struct btrfs_inode_extref *)(ptr +
4518 parent = btrfs_inode_extref_parent(eb, extref);
4519 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4520 name_ptr = (unsigned long)&extref->name;
4521 this_len = sizeof(*extref) + this_name_len;
4524 if (this_name_len > name_len) {
4527 new_name = krealloc(name, this_name_len, GFP_NOFS);
4532 name_len = this_name_len;
4536 read_extent_buffer(eb, name, name_ptr, this_name_len);
4537 di = btrfs_lookup_dir_item(NULL, BTRFS_I(inode)->root,
4538 search_path, parent,
4539 name, this_name_len, 0);
4540 if (di && !IS_ERR(di)) {
4541 struct btrfs_key di_key;
4543 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4545 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4547 *other_ino = di_key.objectid;
4552 } else if (IS_ERR(di)) {
4556 btrfs_release_path(search_path);
4558 cur_offset += this_len;
4562 btrfs_free_path(search_path);
4567 /* log a single inode in the tree log.
4568 * At least one parent directory for this inode must exist in the tree
4569 * or be logged already.
4571 * Any items from this inode changed by the current transaction are copied
4572 * to the log tree. An extra reference is taken on any extents in this
4573 * file, allowing us to avoid a whole pile of corner cases around logging
4574 * blocks that have been removed from the tree.
4576 * See LOG_INODE_ALL and related defines for a description of what inode_only
4579 * This handles both files and directories.
4581 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4582 struct btrfs_root *root, struct inode *inode,
4586 struct btrfs_log_ctx *ctx)
4588 struct btrfs_path *path;
4589 struct btrfs_path *dst_path;
4590 struct btrfs_key min_key;
4591 struct btrfs_key max_key;
4592 struct btrfs_root *log = root->log_root;
4593 struct extent_buffer *src = NULL;
4594 LIST_HEAD(logged_list);
4595 u64 last_extent = 0;
4599 int ins_start_slot = 0;
4601 bool fast_search = false;
4602 u64 ino = btrfs_ino(inode);
4603 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4604 u64 logged_isize = 0;
4605 bool need_log_inode_item = true;
4607 path = btrfs_alloc_path();
4610 dst_path = btrfs_alloc_path();
4612 btrfs_free_path(path);
4616 min_key.objectid = ino;
4617 min_key.type = BTRFS_INODE_ITEM_KEY;
4620 max_key.objectid = ino;
4623 /* today the code can only do partial logging of directories */
4624 if (S_ISDIR(inode->i_mode) ||
4625 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4626 &BTRFS_I(inode)->runtime_flags) &&
4627 inode_only == LOG_INODE_EXISTS))
4628 max_key.type = BTRFS_XATTR_ITEM_KEY;
4630 max_key.type = (u8)-1;
4631 max_key.offset = (u64)-1;
4634 * Only run delayed items if we are a dir or a new file.
4635 * Otherwise commit the delayed inode only, which is needed in
4636 * order for the log replay code to mark inodes for link count
4637 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4639 if (S_ISDIR(inode->i_mode) ||
4640 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed)
4641 ret = btrfs_commit_inode_delayed_items(trans, inode);
4643 ret = btrfs_commit_inode_delayed_inode(inode);
4646 btrfs_free_path(path);
4647 btrfs_free_path(dst_path);
4651 mutex_lock(&BTRFS_I(inode)->log_mutex);
4654 * a brute force approach to making sure we get the most uptodate
4655 * copies of everything.
4657 if (S_ISDIR(inode->i_mode)) {
4658 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4660 if (inode_only == LOG_INODE_EXISTS)
4661 max_key_type = BTRFS_XATTR_ITEM_KEY;
4662 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4664 if (inode_only == LOG_INODE_EXISTS) {
4666 * Make sure the new inode item we write to the log has
4667 * the same isize as the current one (if it exists).
4668 * This is necessary to prevent data loss after log
4669 * replay, and also to prevent doing a wrong expanding
4670 * truncate - for e.g. create file, write 4K into offset
4671 * 0, fsync, write 4K into offset 4096, add hard link,
4672 * fsync some other file (to sync log), power fail - if
4673 * we use the inode's current i_size, after log replay
4674 * we get a 8Kb file, with the last 4Kb extent as a hole
4675 * (zeroes), as if an expanding truncate happened,
4676 * instead of getting a file of 4Kb only.
4678 err = logged_inode_size(log, inode, path,
4683 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4684 &BTRFS_I(inode)->runtime_flags)) {
4685 if (inode_only == LOG_INODE_EXISTS) {
4686 max_key.type = BTRFS_XATTR_ITEM_KEY;
4687 ret = drop_objectid_items(trans, log, path, ino,
4690 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4691 &BTRFS_I(inode)->runtime_flags);
4692 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4693 &BTRFS_I(inode)->runtime_flags);
4695 ret = btrfs_truncate_inode_items(trans,
4701 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4702 &BTRFS_I(inode)->runtime_flags) ||
4703 inode_only == LOG_INODE_EXISTS) {
4704 if (inode_only == LOG_INODE_ALL)
4706 max_key.type = BTRFS_XATTR_ITEM_KEY;
4707 ret = drop_objectid_items(trans, log, path, ino,
4710 if (inode_only == LOG_INODE_ALL)
4723 ret = btrfs_search_forward(root, &min_key,
4724 path, trans->transid);
4732 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4733 if (min_key.objectid != ino)
4735 if (min_key.type > max_key.type)
4738 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4739 need_log_inode_item = false;
4741 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4742 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4743 BTRFS_I(inode)->generation == trans->transid) {
4746 ret = btrfs_check_ref_name_override(path->nodes[0],
4753 } else if (ret > 0 && ctx &&
4754 other_ino != btrfs_ino(ctx->inode)) {
4755 struct btrfs_key inode_key;
4756 struct inode *other_inode;
4762 ins_start_slot = path->slots[0];
4764 ret = copy_items(trans, inode, dst_path, path,
4765 &last_extent, ins_start_slot,
4773 btrfs_release_path(path);
4774 inode_key.objectid = other_ino;
4775 inode_key.type = BTRFS_INODE_ITEM_KEY;
4776 inode_key.offset = 0;
4777 other_inode = btrfs_iget(root->fs_info->sb,
4781 * If the other inode that had a conflicting dir
4782 * entry was deleted in the current transaction,
4783 * we don't need to do more work nor fallback to
4784 * a transaction commit.
4786 if (IS_ERR(other_inode) &&
4787 PTR_ERR(other_inode) == -ENOENT) {
4789 } else if (IS_ERR(other_inode)) {
4790 err = PTR_ERR(other_inode);
4794 * We are safe logging the other inode without
4795 * acquiring its i_mutex as long as we log with
4796 * the LOG_INODE_EXISTS mode. We're safe against
4797 * concurrent renames of the other inode as well
4798 * because during a rename we pin the log and
4799 * update the log with the new name before we
4802 err = btrfs_log_inode(trans, root, other_inode,
4813 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4814 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4817 ret = copy_items(trans, inode, dst_path, path,
4818 &last_extent, ins_start_slot,
4819 ins_nr, inode_only, logged_isize);
4826 btrfs_release_path(path);
4832 src = path->nodes[0];
4833 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4836 } else if (!ins_nr) {
4837 ins_start_slot = path->slots[0];
4842 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4843 ins_start_slot, ins_nr, inode_only,
4851 btrfs_release_path(path);
4855 ins_start_slot = path->slots[0];
4858 nritems = btrfs_header_nritems(path->nodes[0]);
4860 if (path->slots[0] < nritems) {
4861 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4866 ret = copy_items(trans, inode, dst_path, path,
4867 &last_extent, ins_start_slot,
4868 ins_nr, inode_only, logged_isize);
4876 btrfs_release_path(path);
4878 if (min_key.offset < (u64)-1) {
4880 } else if (min_key.type < max_key.type) {
4888 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4889 ins_start_slot, ins_nr, inode_only,
4899 btrfs_release_path(path);
4900 btrfs_release_path(dst_path);
4901 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4904 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4905 btrfs_release_path(path);
4906 btrfs_release_path(dst_path);
4907 err = btrfs_log_trailing_hole(trans, root, inode, path);
4912 btrfs_release_path(path);
4913 btrfs_release_path(dst_path);
4914 if (need_log_inode_item) {
4915 err = log_inode_item(trans, log, dst_path, inode);
4920 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4921 &logged_list, ctx, start, end);
4926 } else if (inode_only == LOG_INODE_ALL) {
4927 struct extent_map *em, *n;
4929 write_lock(&em_tree->lock);
4931 * We can't just remove every em if we're called for a ranged
4932 * fsync - that is, one that doesn't cover the whole possible
4933 * file range (0 to LLONG_MAX). This is because we can have
4934 * em's that fall outside the range we're logging and therefore
4935 * their ordered operations haven't completed yet
4936 * (btrfs_finish_ordered_io() not invoked yet). This means we
4937 * didn't get their respective file extent item in the fs/subvol
4938 * tree yet, and need to let the next fast fsync (one which
4939 * consults the list of modified extent maps) find the em so
4940 * that it logs a matching file extent item and waits for the
4941 * respective ordered operation to complete (if it's still
4944 * Removing every em outside the range we're logging would make
4945 * the next fast fsync not log their matching file extent items,
4946 * therefore making us lose data after a log replay.
4948 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4950 const u64 mod_end = em->mod_start + em->mod_len - 1;
4952 if (em->mod_start >= start && mod_end <= end)
4953 list_del_init(&em->list);
4955 write_unlock(&em_tree->lock);
4958 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4959 ret = log_directory_changes(trans, root, inode, path, dst_path,
4967 spin_lock(&BTRFS_I(inode)->lock);
4968 BTRFS_I(inode)->logged_trans = trans->transid;
4969 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4970 spin_unlock(&BTRFS_I(inode)->lock);
4973 btrfs_put_logged_extents(&logged_list);
4975 btrfs_submit_logged_extents(&logged_list, log);
4976 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4978 btrfs_free_path(path);
4979 btrfs_free_path(dst_path);
4984 * Check if we must fallback to a transaction commit when logging an inode.
4985 * This must be called after logging the inode and is used only in the context
4986 * when fsyncing an inode requires the need to log some other inode - in which
4987 * case we can't lock the i_mutex of each other inode we need to log as that
4988 * can lead to deadlocks with concurrent fsync against other inodes (as we can
4989 * log inodes up or down in the hierarchy) or rename operations for example. So
4990 * we take the log_mutex of the inode after we have logged it and then check for
4991 * its last_unlink_trans value - this is safe because any task setting
4992 * last_unlink_trans must take the log_mutex and it must do this before it does
4993 * the actual unlink operation, so if we do this check before a concurrent task
4994 * sets last_unlink_trans it means we've logged a consistent version/state of
4995 * all the inode items, otherwise we are not sure and must do a transaction
4996 * commit (the concurrent task might have only updated last_unlink_trans before
4997 * we logged the inode or it might have also done the unlink).
4999 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5000 struct inode *inode)
5002 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
5005 mutex_lock(&BTRFS_I(inode)->log_mutex);
5006 if (BTRFS_I(inode)->last_unlink_trans > fs_info->last_trans_committed) {
5008 * Make sure any commits to the log are forced to be full
5011 btrfs_set_log_full_commit(fs_info, trans);
5014 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5020 * follow the dentry parent pointers up the chain and see if any
5021 * of the directories in it require a full commit before they can
5022 * be logged. Returns zero if nothing special needs to be done or 1 if
5023 * a full commit is required.
5025 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5026 struct inode *inode,
5027 struct dentry *parent,
5028 struct super_block *sb,
5032 struct dentry *old_parent = NULL;
5033 struct inode *orig_inode = inode;
5036 * for regular files, if its inode is already on disk, we don't
5037 * have to worry about the parents at all. This is because
5038 * we can use the last_unlink_trans field to record renames
5039 * and other fun in this file.
5041 if (S_ISREG(inode->i_mode) &&
5042 BTRFS_I(inode)->generation <= last_committed &&
5043 BTRFS_I(inode)->last_unlink_trans <= last_committed)
5046 if (!S_ISDIR(inode->i_mode)) {
5047 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5049 inode = d_inode(parent);
5054 * If we are logging a directory then we start with our inode,
5055 * not our parent's inode, so we need to skip setting the
5056 * logged_trans so that further down in the log code we don't
5057 * think this inode has already been logged.
5059 if (inode != orig_inode)
5060 BTRFS_I(inode)->logged_trans = trans->transid;
5063 if (btrfs_must_commit_transaction(trans, inode)) {
5068 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5071 if (IS_ROOT(parent)) {
5072 inode = d_inode(parent);
5073 if (btrfs_must_commit_transaction(trans, inode))
5078 parent = dget_parent(parent);
5080 old_parent = parent;
5081 inode = d_inode(parent);
5089 struct btrfs_dir_list {
5091 struct list_head list;
5095 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5096 * details about the why it is needed.
5097 * This is a recursive operation - if an existing dentry corresponds to a
5098 * directory, that directory's new entries are logged too (same behaviour as
5099 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5100 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5101 * complains about the following circular lock dependency / possible deadlock:
5105 * lock(&type->i_mutex_dir_key#3/2);
5106 * lock(sb_internal#2);
5107 * lock(&type->i_mutex_dir_key#3/2);
5108 * lock(&sb->s_type->i_mutex_key#14);
5110 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5111 * sb_start_intwrite() in btrfs_start_transaction().
5112 * Not locking i_mutex of the inodes is still safe because:
5114 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5115 * that while logging the inode new references (names) are added or removed
5116 * from the inode, leaving the logged inode item with a link count that does
5117 * not match the number of logged inode reference items. This is fine because
5118 * at log replay time we compute the real number of links and correct the
5119 * link count in the inode item (see replay_one_buffer() and
5120 * link_to_fixup_dir());
5122 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5123 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5124 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5125 * has a size that doesn't match the sum of the lengths of all the logged
5126 * names. This does not result in a problem because if a dir_item key is
5127 * logged but its matching dir_index key is not logged, at log replay time we
5128 * don't use it to replay the respective name (see replay_one_name()). On the
5129 * other hand if only the dir_index key ends up being logged, the respective
5130 * name is added to the fs/subvol tree with both the dir_item and dir_index
5131 * keys created (see replay_one_name()).
5132 * The directory's inode item with a wrong i_size is not a problem as well,
5133 * since we don't use it at log replay time to set the i_size in the inode
5134 * item of the fs/subvol tree (see overwrite_item()).
5136 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5137 struct btrfs_root *root,
5138 struct inode *start_inode,
5139 struct btrfs_log_ctx *ctx)
5141 struct btrfs_root *log = root->log_root;
5142 struct btrfs_path *path;
5143 LIST_HEAD(dir_list);
5144 struct btrfs_dir_list *dir_elem;
5147 path = btrfs_alloc_path();
5151 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5153 btrfs_free_path(path);
5156 dir_elem->ino = btrfs_ino(start_inode);
5157 list_add_tail(&dir_elem->list, &dir_list);
5159 while (!list_empty(&dir_list)) {
5160 struct extent_buffer *leaf;
5161 struct btrfs_key min_key;
5165 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5168 goto next_dir_inode;
5170 min_key.objectid = dir_elem->ino;
5171 min_key.type = BTRFS_DIR_ITEM_KEY;
5174 btrfs_release_path(path);
5175 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5177 goto next_dir_inode;
5178 } else if (ret > 0) {
5180 goto next_dir_inode;
5184 leaf = path->nodes[0];
5185 nritems = btrfs_header_nritems(leaf);
5186 for (i = path->slots[0]; i < nritems; i++) {
5187 struct btrfs_dir_item *di;
5188 struct btrfs_key di_key;
5189 struct inode *di_inode;
5190 struct btrfs_dir_list *new_dir_elem;
5191 int log_mode = LOG_INODE_EXISTS;
5194 btrfs_item_key_to_cpu(leaf, &min_key, i);
5195 if (min_key.objectid != dir_elem->ino ||
5196 min_key.type != BTRFS_DIR_ITEM_KEY)
5197 goto next_dir_inode;
5199 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5200 type = btrfs_dir_type(leaf, di);
5201 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5202 type != BTRFS_FT_DIR)
5204 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5205 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5208 di_inode = btrfs_iget(root->fs_info->sb, &di_key,
5210 if (IS_ERR(di_inode)) {
5211 ret = PTR_ERR(di_inode);
5212 goto next_dir_inode;
5215 if (btrfs_inode_in_log(di_inode, trans->transid)) {
5220 ctx->log_new_dentries = false;
5221 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5222 log_mode = LOG_INODE_ALL;
5223 btrfs_release_path(path);
5224 ret = btrfs_log_inode(trans, root, di_inode,
5225 log_mode, 0, LLONG_MAX, ctx);
5227 btrfs_must_commit_transaction(trans, di_inode))
5231 goto next_dir_inode;
5232 if (ctx->log_new_dentries) {
5233 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5235 if (!new_dir_elem) {
5237 goto next_dir_inode;
5239 new_dir_elem->ino = di_key.objectid;
5240 list_add_tail(&new_dir_elem->list, &dir_list);
5245 ret = btrfs_next_leaf(log, path);
5247 goto next_dir_inode;
5248 } else if (ret > 0) {
5250 goto next_dir_inode;
5254 if (min_key.offset < (u64)-1) {
5259 list_del(&dir_elem->list);
5263 btrfs_free_path(path);
5267 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5268 struct inode *inode,
5269 struct btrfs_log_ctx *ctx)
5272 struct btrfs_path *path;
5273 struct btrfs_key key;
5274 struct btrfs_root *root = BTRFS_I(inode)->root;
5275 const u64 ino = btrfs_ino(inode);
5277 path = btrfs_alloc_path();
5280 path->skip_locking = 1;
5281 path->search_commit_root = 1;
5284 key.type = BTRFS_INODE_REF_KEY;
5286 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5291 struct extent_buffer *leaf = path->nodes[0];
5292 int slot = path->slots[0];
5297 if (slot >= btrfs_header_nritems(leaf)) {
5298 ret = btrfs_next_leaf(root, path);
5306 btrfs_item_key_to_cpu(leaf, &key, slot);
5307 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5308 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5311 item_size = btrfs_item_size_nr(leaf, slot);
5312 ptr = btrfs_item_ptr_offset(leaf, slot);
5313 while (cur_offset < item_size) {
5314 struct btrfs_key inode_key;
5315 struct inode *dir_inode;
5317 inode_key.type = BTRFS_INODE_ITEM_KEY;
5318 inode_key.offset = 0;
5320 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5321 struct btrfs_inode_extref *extref;
5323 extref = (struct btrfs_inode_extref *)
5325 inode_key.objectid = btrfs_inode_extref_parent(
5327 cur_offset += sizeof(*extref);
5328 cur_offset += btrfs_inode_extref_name_len(leaf,
5331 inode_key.objectid = key.offset;
5332 cur_offset = item_size;
5335 dir_inode = btrfs_iget(root->fs_info->sb, &inode_key,
5337 /* If parent inode was deleted, skip it. */
5338 if (IS_ERR(dir_inode))
5342 ctx->log_new_dentries = false;
5343 ret = btrfs_log_inode(trans, root, dir_inode,
5344 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5346 btrfs_must_commit_transaction(trans, dir_inode))
5348 if (!ret && ctx && ctx->log_new_dentries)
5349 ret = log_new_dir_dentries(trans, root,
5359 btrfs_free_path(path);
5364 * helper function around btrfs_log_inode to make sure newly created
5365 * parent directories also end up in the log. A minimal inode and backref
5366 * only logging is done of any parent directories that are older than
5367 * the last committed transaction
5369 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5370 struct btrfs_root *root, struct inode *inode,
5371 struct dentry *parent,
5375 struct btrfs_log_ctx *ctx)
5377 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5378 struct super_block *sb;
5379 struct dentry *old_parent = NULL;
5381 u64 last_committed = root->fs_info->last_trans_committed;
5382 bool log_dentries = false;
5383 struct inode *orig_inode = inode;
5387 if (btrfs_test_opt(root->fs_info, NOTREELOG)) {
5393 * The prev transaction commit doesn't complete, we need do
5394 * full commit by ourselves.
5396 if (root->fs_info->last_trans_log_full_commit >
5397 root->fs_info->last_trans_committed) {
5402 if (root != BTRFS_I(inode)->root ||
5403 btrfs_root_refs(&root->root_item) == 0) {
5408 ret = check_parent_dirs_for_sync(trans, inode, parent,
5409 sb, last_committed);
5413 if (btrfs_inode_in_log(inode, trans->transid)) {
5414 ret = BTRFS_NO_LOG_SYNC;
5418 ret = start_log_trans(trans, root, ctx);
5422 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5427 * for regular files, if its inode is already on disk, we don't
5428 * have to worry about the parents at all. This is because
5429 * we can use the last_unlink_trans field to record renames
5430 * and other fun in this file.
5432 if (S_ISREG(inode->i_mode) &&
5433 BTRFS_I(inode)->generation <= last_committed &&
5434 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
5439 if (S_ISDIR(inode->i_mode) && ctx && ctx->log_new_dentries)
5440 log_dentries = true;
5443 * On unlink we must make sure all our current and old parent directory
5444 * inodes are fully logged. This is to prevent leaving dangling
5445 * directory index entries in directories that were our parents but are
5446 * not anymore. Not doing this results in old parent directory being
5447 * impossible to delete after log replay (rmdir will always fail with
5448 * error -ENOTEMPTY).
5454 * ln testdir/foo testdir/bar
5456 * unlink testdir/bar
5457 * xfs_io -c fsync testdir/foo
5459 * mount fs, triggers log replay
5461 * If we don't log the parent directory (testdir), after log replay the
5462 * directory still has an entry pointing to the file inode using the bar
5463 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5464 * the file inode has a link count of 1.
5470 * ln foo testdir/foo2
5471 * ln foo testdir/foo3
5473 * unlink testdir/foo3
5474 * xfs_io -c fsync foo
5476 * mount fs, triggers log replay
5478 * Similar as the first example, after log replay the parent directory
5479 * testdir still has an entry pointing to the inode file with name foo3
5480 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5481 * and has a link count of 2.
5483 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
5484 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5490 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5493 inode = d_inode(parent);
5494 if (root != BTRFS_I(inode)->root)
5497 if (BTRFS_I(inode)->generation > last_committed) {
5498 ret = btrfs_log_inode(trans, root, inode,
5504 if (IS_ROOT(parent))
5507 parent = dget_parent(parent);
5509 old_parent = parent;
5512 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5518 btrfs_set_log_full_commit(root->fs_info, trans);
5523 btrfs_remove_log_ctx(root, ctx);
5524 btrfs_end_log_trans(root);
5530 * it is not safe to log dentry if the chunk root has added new
5531 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5532 * If this returns 1, you must commit the transaction to safely get your
5535 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5536 struct btrfs_root *root, struct dentry *dentry,
5539 struct btrfs_log_ctx *ctx)
5541 struct dentry *parent = dget_parent(dentry);
5544 ret = btrfs_log_inode_parent(trans, root, d_inode(dentry), parent,
5545 start, end, 0, ctx);
5552 * should be called during mount to recover any replay any log trees
5555 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5558 struct btrfs_path *path;
5559 struct btrfs_trans_handle *trans;
5560 struct btrfs_key key;
5561 struct btrfs_key found_key;
5562 struct btrfs_key tmp_key;
5563 struct btrfs_root *log;
5564 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5565 struct walk_control wc = {
5566 .process_func = process_one_buffer,
5570 path = btrfs_alloc_path();
5574 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5576 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5577 if (IS_ERR(trans)) {
5578 ret = PTR_ERR(trans);
5585 ret = walk_log_tree(trans, log_root_tree, &wc);
5587 btrfs_handle_fs_error(fs_info, ret,
5588 "Failed to pin buffers while recovering log root tree.");
5593 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5594 key.offset = (u64)-1;
5595 key.type = BTRFS_ROOT_ITEM_KEY;
5598 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5601 btrfs_handle_fs_error(fs_info, ret,
5602 "Couldn't find tree log root.");
5606 if (path->slots[0] == 0)
5610 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5612 btrfs_release_path(path);
5613 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5616 log = btrfs_read_fs_root(log_root_tree, &found_key);
5619 btrfs_handle_fs_error(fs_info, ret,
5620 "Couldn't read tree log root.");
5624 tmp_key.objectid = found_key.offset;
5625 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5626 tmp_key.offset = (u64)-1;
5628 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5629 if (IS_ERR(wc.replay_dest)) {
5630 ret = PTR_ERR(wc.replay_dest);
5631 free_extent_buffer(log->node);
5632 free_extent_buffer(log->commit_root);
5634 btrfs_handle_fs_error(fs_info, ret,
5635 "Couldn't read target root for tree log recovery.");
5639 wc.replay_dest->log_root = log;
5640 btrfs_record_root_in_trans(trans, wc.replay_dest);
5641 ret = walk_log_tree(trans, log, &wc);
5643 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5644 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5648 key.offset = found_key.offset - 1;
5649 wc.replay_dest->log_root = NULL;
5650 free_extent_buffer(log->node);
5651 free_extent_buffer(log->commit_root);
5657 if (found_key.offset == 0)
5660 btrfs_release_path(path);
5662 /* step one is to pin it all, step two is to replay just inodes */
5665 wc.process_func = replay_one_buffer;
5666 wc.stage = LOG_WALK_REPLAY_INODES;
5669 /* step three is to replay everything */
5670 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5675 btrfs_free_path(path);
5677 /* step 4: commit the transaction, which also unpins the blocks */
5678 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
5682 free_extent_buffer(log_root_tree->node);
5683 log_root_tree->log_root = NULL;
5684 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5685 kfree(log_root_tree);
5690 btrfs_end_transaction(wc.trans, fs_info->tree_root);
5691 btrfs_free_path(path);
5696 * there are some corner cases where we want to force a full
5697 * commit instead of allowing a directory to be logged.
5699 * They revolve around files there were unlinked from the directory, and
5700 * this function updates the parent directory so that a full commit is
5701 * properly done if it is fsync'd later after the unlinks are done.
5703 * Must be called before the unlink operations (updates to the subvolume tree,
5704 * inodes, etc) are done.
5706 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5707 struct inode *dir, struct inode *inode,
5711 * when we're logging a file, if it hasn't been renamed
5712 * or unlinked, and its inode is fully committed on disk,
5713 * we don't have to worry about walking up the directory chain
5714 * to log its parents.
5716 * So, we use the last_unlink_trans field to put this transid
5717 * into the file. When the file is logged we check it and
5718 * don't log the parents if the file is fully on disk.
5720 mutex_lock(&BTRFS_I(inode)->log_mutex);
5721 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5722 mutex_unlock(&BTRFS_I(inode)->log_mutex);
5725 * if this directory was already logged any new
5726 * names for this file/dir will get recorded
5729 if (BTRFS_I(dir)->logged_trans == trans->transid)
5733 * if the inode we're about to unlink was logged,
5734 * the log will be properly updated for any new names
5736 if (BTRFS_I(inode)->logged_trans == trans->transid)
5740 * when renaming files across directories, if the directory
5741 * there we're unlinking from gets fsync'd later on, there's
5742 * no way to find the destination directory later and fsync it
5743 * properly. So, we have to be conservative and force commits
5744 * so the new name gets discovered.
5749 /* we can safely do the unlink without any special recording */
5753 mutex_lock(&BTRFS_I(dir)->log_mutex);
5754 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5755 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5759 * Make sure that if someone attempts to fsync the parent directory of a deleted
5760 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5761 * that after replaying the log tree of the parent directory's root we will not
5762 * see the snapshot anymore and at log replay time we will not see any log tree
5763 * corresponding to the deleted snapshot's root, which could lead to replaying
5764 * it after replaying the log tree of the parent directory (which would replay
5765 * the snapshot delete operation).
5767 * Must be called before the actual snapshot destroy operation (updates to the
5768 * parent root and tree of tree roots trees, etc) are done.
5770 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5773 mutex_lock(&BTRFS_I(dir)->log_mutex);
5774 BTRFS_I(dir)->last_unlink_trans = trans->transid;
5775 mutex_unlock(&BTRFS_I(dir)->log_mutex);
5779 * Call this after adding a new name for a file and it will properly
5780 * update the log to reflect the new name.
5782 * It will return zero if all goes well, and it will return 1 if a
5783 * full transaction commit is required.
5785 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5786 struct inode *inode, struct inode *old_dir,
5787 struct dentry *parent)
5789 struct btrfs_root * root = BTRFS_I(inode)->root;
5792 * this will force the logging code to walk the dentry chain
5795 if (S_ISREG(inode->i_mode))
5796 BTRFS_I(inode)->last_unlink_trans = trans->transid;
5799 * if this inode hasn't been logged and directory we're renaming it
5800 * from hasn't been logged, we don't need to log it
5802 if (BTRFS_I(inode)->logged_trans <=
5803 root->fs_info->last_trans_committed &&
5804 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
5805 root->fs_info->last_trans_committed))
5808 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5809 LLONG_MAX, 1, NULL);