2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
29 #include "compression.h"
32 /* magic values for the inode_only field in btrfs_log_inode:
34 * LOG_INODE_ALL means to log everything
35 * LOG_INODE_EXISTS means to log just enough to recreate the inode
38 #define LOG_INODE_ALL 0
39 #define LOG_INODE_EXISTS 1
40 #define LOG_OTHER_INODE 2
43 * directory trouble cases
45 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
46 * log, we must force a full commit before doing an fsync of the directory
47 * where the unlink was done.
48 * ---> record transid of last unlink/rename per directory
52 * rename foo/some_dir foo2/some_dir
54 * fsync foo/some_dir/some_file
56 * The fsync above will unlink the original some_dir without recording
57 * it in its new location (foo2). After a crash, some_dir will be gone
58 * unless the fsync of some_file forces a full commit
60 * 2) we must log any new names for any file or dir that is in the fsync
61 * log. ---> check inode while renaming/linking.
63 * 2a) we must log any new names for any file or dir during rename
64 * when the directory they are being removed from was logged.
65 * ---> check inode and old parent dir during rename
67 * 2a is actually the more important variant. With the extra logging
68 * a crash might unlink the old name without recreating the new one
70 * 3) after a crash, we must go through any directories with a link count
71 * of zero and redo the rm -rf
78 * The directory f1 was fully removed from the FS, but fsync was never
79 * called on f1, only its parent dir. After a crash the rm -rf must
80 * be replayed. This must be able to recurse down the entire
81 * directory tree. The inode link count fixup code takes care of the
86 * stages for the tree walking. The first
87 * stage (0) is to only pin down the blocks we find
88 * the second stage (1) is to make sure that all the inodes
89 * we find in the log are created in the subvolume.
91 * The last stage is to deal with directories and links and extents
92 * and all the other fun semantics
94 #define LOG_WALK_PIN_ONLY 0
95 #define LOG_WALK_REPLAY_INODES 1
96 #define LOG_WALK_REPLAY_DIR_INDEX 2
97 #define LOG_WALK_REPLAY_ALL 3
99 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
100 struct btrfs_root *root, struct btrfs_inode *inode,
104 struct btrfs_log_ctx *ctx);
105 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_path *path, u64 objectid);
108 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root,
110 struct btrfs_root *log,
111 struct btrfs_path *path,
112 u64 dirid, int del_all);
115 * tree logging is a special write ahead log used to make sure that
116 * fsyncs and O_SYNCs can happen without doing full tree commits.
118 * Full tree commits are expensive because they require commonly
119 * modified blocks to be recowed, creating many dirty pages in the
120 * extent tree an 4x-6x higher write load than ext3.
122 * Instead of doing a tree commit on every fsync, we use the
123 * key ranges and transaction ids to find items for a given file or directory
124 * that have changed in this transaction. Those items are copied into
125 * a special tree (one per subvolume root), that tree is written to disk
126 * and then the fsync is considered complete.
128 * After a crash, items are copied out of the log-tree back into the
129 * subvolume tree. Any file data extents found are recorded in the extent
130 * allocation tree, and the log-tree freed.
132 * The log tree is read three times, once to pin down all the extents it is
133 * using in ram and once, once to create all the inodes logged in the tree
134 * and once to do all the other items.
138 * start a sub transaction and setup the log tree
139 * this increments the log tree writer count to make the people
140 * syncing the tree wait for us to finish
142 static int start_log_trans(struct btrfs_trans_handle *trans,
143 struct btrfs_root *root,
144 struct btrfs_log_ctx *ctx)
146 struct btrfs_fs_info *fs_info = root->fs_info;
149 mutex_lock(&root->log_mutex);
151 if (root->log_root) {
152 if (btrfs_need_log_full_commit(fs_info, trans)) {
157 if (!root->log_start_pid) {
158 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 root->log_start_pid = current->pid;
160 } else if (root->log_start_pid != current->pid) {
161 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
164 mutex_lock(&fs_info->tree_log_mutex);
165 if (!fs_info->log_root_tree)
166 ret = btrfs_init_log_root_tree(trans, fs_info);
167 mutex_unlock(&fs_info->tree_log_mutex);
171 ret = btrfs_add_log_tree(trans, root);
175 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
176 root->log_start_pid = current->pid;
179 atomic_inc(&root->log_batch);
180 atomic_inc(&root->log_writers);
182 int index = root->log_transid % 2;
183 list_add_tail(&ctx->list, &root->log_ctxs[index]);
184 ctx->log_transid = root->log_transid;
188 mutex_unlock(&root->log_mutex);
193 * returns 0 if there was a log transaction running and we were able
194 * to join, or returns -ENOENT if there were not transactions
197 static int join_running_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 if (root->log_root) {
208 atomic_inc(&root->log_writers);
210 mutex_unlock(&root->log_mutex);
215 * This either makes the current running log transaction wait
216 * until you call btrfs_end_log_trans() or it makes any future
217 * log transactions wait until you call btrfs_end_log_trans()
219 int btrfs_pin_log_trans(struct btrfs_root *root)
223 mutex_lock(&root->log_mutex);
224 atomic_inc(&root->log_writers);
225 mutex_unlock(&root->log_mutex);
230 * indicate we're done making changes to the log tree
231 * and wake up anyone waiting to do a sync
233 void btrfs_end_log_trans(struct btrfs_root *root)
235 if (atomic_dec_and_test(&root->log_writers)) {
237 * Implicit memory barrier after atomic_dec_and_test
239 if (waitqueue_active(&root->log_writer_wait))
240 wake_up(&root->log_writer_wait);
246 * the walk control struct is used to pass state down the chain when
247 * processing the log tree. The stage field tells us which part
248 * of the log tree processing we are currently doing. The others
249 * are state fields used for that specific part
251 struct walk_control {
252 /* should we free the extent on disk when done? This is used
253 * at transaction commit time while freeing a log tree
257 /* should we write out the extent buffer? This is used
258 * while flushing the log tree to disk during a sync
262 /* should we wait for the extent buffer io to finish? Also used
263 * while flushing the log tree to disk for a sync
267 /* pin only walk, we record which extents on disk belong to the
272 /* what stage of the replay code we're currently in */
275 /* the root we are currently replaying */
276 struct btrfs_root *replay_dest;
278 /* the trans handle for the current replay */
279 struct btrfs_trans_handle *trans;
281 /* the function that gets used to process blocks we find in the
282 * tree. Note the extent_buffer might not be up to date when it is
283 * passed in, and it must be checked or read if you need the data
286 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
287 struct walk_control *wc, u64 gen);
291 * process_func used to pin down extents, write them or wait on them
293 static int process_one_buffer(struct btrfs_root *log,
294 struct extent_buffer *eb,
295 struct walk_control *wc, u64 gen)
297 struct btrfs_fs_info *fs_info = log->fs_info;
301 * If this fs is mixed then we need to be able to process the leaves to
302 * pin down any logged extents, so we have to read the block.
304 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
305 ret = btrfs_read_buffer(eb, gen);
311 ret = btrfs_pin_extent_for_log_replay(fs_info, eb->start,
314 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
315 if (wc->pin && btrfs_header_level(eb) == 0)
316 ret = btrfs_exclude_logged_extents(fs_info, eb);
318 btrfs_write_tree_block(eb);
320 btrfs_wait_tree_block_writeback(eb);
326 * Item overwrite used by replay and tree logging. eb, slot and key all refer
327 * to the src data we are copying out.
329 * root is the tree we are copying into, and path is a scratch
330 * path for use in this function (it should be released on entry and
331 * will be released on exit).
333 * If the key is already in the destination tree the existing item is
334 * overwritten. If the existing item isn't big enough, it is extended.
335 * If it is too large, it is truncated.
337 * If the key isn't in the destination yet, a new item is inserted.
339 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
340 struct btrfs_root *root,
341 struct btrfs_path *path,
342 struct extent_buffer *eb, int slot,
343 struct btrfs_key *key)
345 struct btrfs_fs_info *fs_info = root->fs_info;
348 u64 saved_i_size = 0;
349 int save_old_i_size = 0;
350 unsigned long src_ptr;
351 unsigned long dst_ptr;
352 int overwrite_root = 0;
353 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
355 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
358 item_size = btrfs_item_size_nr(eb, slot);
359 src_ptr = btrfs_item_ptr_offset(eb, slot);
361 /* look for the key in the destination tree */
362 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
369 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
371 if (dst_size != item_size)
374 if (item_size == 0) {
375 btrfs_release_path(path);
378 dst_copy = kmalloc(item_size, GFP_NOFS);
379 src_copy = kmalloc(item_size, GFP_NOFS);
380 if (!dst_copy || !src_copy) {
381 btrfs_release_path(path);
387 read_extent_buffer(eb, src_copy, src_ptr, item_size);
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
390 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
392 ret = memcmp(dst_copy, src_copy, item_size);
397 * they have the same contents, just return, this saves
398 * us from cowing blocks in the destination tree and doing
399 * extra writes that may not have been done by a previous
403 btrfs_release_path(path);
408 * We need to load the old nbytes into the inode so when we
409 * replay the extents we've logged we get the right nbytes.
412 struct btrfs_inode_item *item;
416 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
417 struct btrfs_inode_item);
418 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
419 item = btrfs_item_ptr(eb, slot,
420 struct btrfs_inode_item);
421 btrfs_set_inode_nbytes(eb, item, nbytes);
424 * If this is a directory we need to reset the i_size to
425 * 0 so that we can set it up properly when replaying
426 * the rest of the items in this log.
428 mode = btrfs_inode_mode(eb, item);
430 btrfs_set_inode_size(eb, item, 0);
432 } else if (inode_item) {
433 struct btrfs_inode_item *item;
437 * New inode, set nbytes to 0 so that the nbytes comes out
438 * properly when we replay the extents.
440 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
441 btrfs_set_inode_nbytes(eb, item, 0);
444 * If this is a directory we need to reset the i_size to 0 so
445 * that we can set it up properly when replaying the rest of
446 * the items in this log.
448 mode = btrfs_inode_mode(eb, item);
450 btrfs_set_inode_size(eb, item, 0);
453 btrfs_release_path(path);
454 /* try to insert the key into the destination tree */
455 path->skip_release_on_error = 1;
456 ret = btrfs_insert_empty_item(trans, root, path,
458 path->skip_release_on_error = 0;
460 /* make sure any existing item is the correct size */
461 if (ret == -EEXIST || ret == -EOVERFLOW) {
463 found_size = btrfs_item_size_nr(path->nodes[0],
465 if (found_size > item_size)
466 btrfs_truncate_item(fs_info, path, item_size, 1);
467 else if (found_size < item_size)
468 btrfs_extend_item(fs_info, path,
469 item_size - found_size);
473 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
476 /* don't overwrite an existing inode if the generation number
477 * was logged as zero. This is done when the tree logging code
478 * is just logging an inode to make sure it exists after recovery.
480 * Also, don't overwrite i_size on directories during replay.
481 * log replay inserts and removes directory items based on the
482 * state of the tree found in the subvolume, and i_size is modified
485 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
486 struct btrfs_inode_item *src_item;
487 struct btrfs_inode_item *dst_item;
489 src_item = (struct btrfs_inode_item *)src_ptr;
490 dst_item = (struct btrfs_inode_item *)dst_ptr;
492 if (btrfs_inode_generation(eb, src_item) == 0) {
493 struct extent_buffer *dst_eb = path->nodes[0];
494 const u64 ino_size = btrfs_inode_size(eb, src_item);
497 * For regular files an ino_size == 0 is used only when
498 * logging that an inode exists, as part of a directory
499 * fsync, and the inode wasn't fsynced before. In this
500 * case don't set the size of the inode in the fs/subvol
501 * tree, otherwise we would be throwing valid data away.
503 if (S_ISREG(btrfs_inode_mode(eb, src_item)) &&
504 S_ISREG(btrfs_inode_mode(dst_eb, dst_item)) &&
506 struct btrfs_map_token token;
508 btrfs_init_map_token(&token);
509 btrfs_set_token_inode_size(dst_eb, dst_item,
515 if (overwrite_root &&
516 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
517 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
519 saved_i_size = btrfs_inode_size(path->nodes[0],
524 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
527 if (save_old_i_size) {
528 struct btrfs_inode_item *dst_item;
529 dst_item = (struct btrfs_inode_item *)dst_ptr;
530 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
533 /* make sure the generation is filled in */
534 if (key->type == BTRFS_INODE_ITEM_KEY) {
535 struct btrfs_inode_item *dst_item;
536 dst_item = (struct btrfs_inode_item *)dst_ptr;
537 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
538 btrfs_set_inode_generation(path->nodes[0], dst_item,
543 btrfs_mark_buffer_dirty(path->nodes[0]);
544 btrfs_release_path(path);
549 * simple helper to read an inode off the disk from a given root
550 * This can only be called for subvolume roots and not for the log
552 static noinline struct inode *read_one_inode(struct btrfs_root *root,
555 struct btrfs_key key;
558 key.objectid = objectid;
559 key.type = BTRFS_INODE_ITEM_KEY;
561 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
564 } else if (is_bad_inode(inode)) {
571 /* replays a single extent in 'eb' at 'slot' with 'key' into the
572 * subvolume 'root'. path is released on entry and should be released
575 * extents in the log tree have not been allocated out of the extent
576 * tree yet. So, this completes the allocation, taking a reference
577 * as required if the extent already exists or creating a new extent
578 * if it isn't in the extent allocation tree yet.
580 * The extent is inserted into the file, dropping any existing extents
581 * from the file that overlap the new one.
583 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
584 struct btrfs_root *root,
585 struct btrfs_path *path,
586 struct extent_buffer *eb, int slot,
587 struct btrfs_key *key)
589 struct btrfs_fs_info *fs_info = root->fs_info;
592 u64 start = key->offset;
594 struct btrfs_file_extent_item *item;
595 struct inode *inode = NULL;
599 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
600 found_type = btrfs_file_extent_type(eb, item);
602 if (found_type == BTRFS_FILE_EXTENT_REG ||
603 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
604 nbytes = btrfs_file_extent_num_bytes(eb, item);
605 extent_end = start + nbytes;
608 * We don't add to the inodes nbytes if we are prealloc or a
611 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
613 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
614 size = btrfs_file_extent_inline_len(eb, slot, item);
615 nbytes = btrfs_file_extent_ram_bytes(eb, item);
616 extent_end = ALIGN(start + size,
617 fs_info->sectorsize);
623 inode = read_one_inode(root, key->objectid);
630 * first check to see if we already have this extent in the
631 * file. This must be done before the btrfs_drop_extents run
632 * so we don't try to drop this extent.
634 ret = btrfs_lookup_file_extent(trans, root, path,
635 btrfs_ino(BTRFS_I(inode)), start, 0);
638 (found_type == BTRFS_FILE_EXTENT_REG ||
639 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
640 struct btrfs_file_extent_item cmp1;
641 struct btrfs_file_extent_item cmp2;
642 struct btrfs_file_extent_item *existing;
643 struct extent_buffer *leaf;
645 leaf = path->nodes[0];
646 existing = btrfs_item_ptr(leaf, path->slots[0],
647 struct btrfs_file_extent_item);
649 read_extent_buffer(eb, &cmp1, (unsigned long)item,
651 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
655 * we already have a pointer to this exact extent,
656 * we don't have to do anything
658 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
659 btrfs_release_path(path);
663 btrfs_release_path(path);
665 /* drop any overlapping extents */
666 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
670 if (found_type == BTRFS_FILE_EXTENT_REG ||
671 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
673 unsigned long dest_offset;
674 struct btrfs_key ins;
676 if (btrfs_file_extent_disk_bytenr(eb, item) == 0 &&
677 btrfs_fs_incompat(fs_info, NO_HOLES))
680 ret = btrfs_insert_empty_item(trans, root, path, key,
684 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
686 copy_extent_buffer(path->nodes[0], eb, dest_offset,
687 (unsigned long)item, sizeof(*item));
689 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
690 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
691 ins.type = BTRFS_EXTENT_ITEM_KEY;
692 offset = key->offset - btrfs_file_extent_offset(eb, item);
695 * Manually record dirty extent, as here we did a shallow
696 * file extent item copy and skip normal backref update,
697 * but modifying extent tree all by ourselves.
698 * So need to manually record dirty extent for qgroup,
699 * as the owner of the file extent changed from log tree
700 * (doesn't affect qgroup) to fs/file tree(affects qgroup)
702 ret = btrfs_qgroup_trace_extent(trans, fs_info,
703 btrfs_file_extent_disk_bytenr(eb, item),
704 btrfs_file_extent_disk_num_bytes(eb, item),
709 if (ins.objectid > 0) {
712 LIST_HEAD(ordered_sums);
714 * is this extent already allocated in the extent
715 * allocation tree? If so, just add a reference
717 ret = btrfs_lookup_data_extent(fs_info, ins.objectid,
720 ret = btrfs_inc_extent_ref(trans, fs_info,
721 ins.objectid, ins.offset,
722 0, root->root_key.objectid,
723 key->objectid, offset);
728 * insert the extent pointer in the extent
731 ret = btrfs_alloc_logged_file_extent(trans,
733 root->root_key.objectid,
734 key->objectid, offset, &ins);
738 btrfs_release_path(path);
740 if (btrfs_file_extent_compression(eb, item)) {
741 csum_start = ins.objectid;
742 csum_end = csum_start + ins.offset;
744 csum_start = ins.objectid +
745 btrfs_file_extent_offset(eb, item);
746 csum_end = csum_start +
747 btrfs_file_extent_num_bytes(eb, item);
750 ret = btrfs_lookup_csums_range(root->log_root,
751 csum_start, csum_end - 1,
756 * Now delete all existing cums in the csum root that
757 * cover our range. We do this because we can have an
758 * extent that is completely referenced by one file
759 * extent item and partially referenced by another
760 * file extent item (like after using the clone or
761 * extent_same ioctls). In this case if we end up doing
762 * the replay of the one that partially references the
763 * extent first, and we do not do the csum deletion
764 * below, we can get 2 csum items in the csum tree that
765 * overlap each other. For example, imagine our log has
766 * the two following file extent items:
768 * key (257 EXTENT_DATA 409600)
769 * extent data disk byte 12845056 nr 102400
770 * extent data offset 20480 nr 20480 ram 102400
772 * key (257 EXTENT_DATA 819200)
773 * extent data disk byte 12845056 nr 102400
774 * extent data offset 0 nr 102400 ram 102400
776 * Where the second one fully references the 100K extent
777 * that starts at disk byte 12845056, and the log tree
778 * has a single csum item that covers the entire range
781 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
783 * After the first file extent item is replayed, the
784 * csum tree gets the following csum item:
786 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
788 * Which covers the 20K sub-range starting at offset 20K
789 * of our extent. Now when we replay the second file
790 * extent item, if we do not delete existing csum items
791 * that cover any of its blocks, we end up getting two
792 * csum items in our csum tree that overlap each other:
794 * key (EXTENT_CSUM EXTENT_CSUM 12845056) itemsize 100
795 * key (EXTENT_CSUM EXTENT_CSUM 12865536) itemsize 20
797 * Which is a problem, because after this anyone trying
798 * to lookup up for the checksum of any block of our
799 * extent starting at an offset of 40K or higher, will
800 * end up looking at the second csum item only, which
801 * does not contain the checksum for any block starting
802 * at offset 40K or higher of our extent.
804 while (!list_empty(&ordered_sums)) {
805 struct btrfs_ordered_sum *sums;
806 sums = list_entry(ordered_sums.next,
807 struct btrfs_ordered_sum,
810 ret = btrfs_del_csums(trans, fs_info,
814 ret = btrfs_csum_file_blocks(trans,
815 fs_info->csum_root, sums);
816 list_del(&sums->list);
822 btrfs_release_path(path);
824 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
825 /* inline extents are easy, we just overwrite them */
826 ret = overwrite_item(trans, root, path, eb, slot, key);
831 inode_add_bytes(inode, nbytes);
833 ret = btrfs_update_inode(trans, root, inode);
841 * when cleaning up conflicts between the directory names in the
842 * subvolume, directory names in the log and directory names in the
843 * inode back references, we may have to unlink inodes from directories.
845 * This is a helper function to do the unlink of a specific directory
848 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
849 struct btrfs_root *root,
850 struct btrfs_path *path,
851 struct btrfs_inode *dir,
852 struct btrfs_dir_item *di)
854 struct btrfs_fs_info *fs_info = root->fs_info;
858 struct extent_buffer *leaf;
859 struct btrfs_key location;
862 leaf = path->nodes[0];
864 btrfs_dir_item_key_to_cpu(leaf, di, &location);
865 name_len = btrfs_dir_name_len(leaf, di);
866 name = kmalloc(name_len, GFP_NOFS);
870 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
871 btrfs_release_path(path);
873 inode = read_one_inode(root, location.objectid);
879 ret = link_to_fixup_dir(trans, root, path, location.objectid);
883 ret = btrfs_unlink_inode(trans, root, dir, BTRFS_I(inode), name,
888 ret = btrfs_run_delayed_items(trans, fs_info);
896 * helper function to see if a given name and sequence number found
897 * in an inode back reference are already in a directory and correctly
898 * point to this inode
900 static noinline int inode_in_dir(struct btrfs_root *root,
901 struct btrfs_path *path,
902 u64 dirid, u64 objectid, u64 index,
903 const char *name, int name_len)
905 struct btrfs_dir_item *di;
906 struct btrfs_key location;
909 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
910 index, name, name_len, 0);
911 if (di && !IS_ERR(di)) {
912 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
913 if (location.objectid != objectid)
917 btrfs_release_path(path);
919 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
920 if (di && !IS_ERR(di)) {
921 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
922 if (location.objectid != objectid)
928 btrfs_release_path(path);
933 * helper function to check a log tree for a named back reference in
934 * an inode. This is used to decide if a back reference that is
935 * found in the subvolume conflicts with what we find in the log.
937 * inode backreferences may have multiple refs in a single item,
938 * during replay we process one reference at a time, and we don't
939 * want to delete valid links to a file from the subvolume if that
940 * link is also in the log.
942 static noinline int backref_in_log(struct btrfs_root *log,
943 struct btrfs_key *key,
945 const char *name, int namelen)
947 struct btrfs_path *path;
948 struct btrfs_inode_ref *ref;
950 unsigned long ptr_end;
951 unsigned long name_ptr;
957 path = btrfs_alloc_path();
961 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
965 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
967 if (key->type == BTRFS_INODE_EXTREF_KEY) {
968 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
969 name, namelen, NULL))
975 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
976 ptr_end = ptr + item_size;
977 while (ptr < ptr_end) {
978 ref = (struct btrfs_inode_ref *)ptr;
979 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
980 if (found_name_len == namelen) {
981 name_ptr = (unsigned long)(ref + 1);
982 ret = memcmp_extent_buffer(path->nodes[0], name,
989 ptr = (unsigned long)(ref + 1) + found_name_len;
992 btrfs_free_path(path);
996 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
997 struct btrfs_root *root,
998 struct btrfs_path *path,
999 struct btrfs_root *log_root,
1000 struct btrfs_inode *dir,
1001 struct btrfs_inode *inode,
1002 u64 inode_objectid, u64 parent_objectid,
1003 u64 ref_index, char *name, int namelen,
1006 struct btrfs_fs_info *fs_info = root->fs_info;
1009 int victim_name_len;
1010 struct extent_buffer *leaf;
1011 struct btrfs_dir_item *di;
1012 struct btrfs_key search_key;
1013 struct btrfs_inode_extref *extref;
1016 /* Search old style refs */
1017 search_key.objectid = inode_objectid;
1018 search_key.type = BTRFS_INODE_REF_KEY;
1019 search_key.offset = parent_objectid;
1020 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
1022 struct btrfs_inode_ref *victim_ref;
1024 unsigned long ptr_end;
1026 leaf = path->nodes[0];
1028 /* are we trying to overwrite a back ref for the root directory
1029 * if so, just jump out, we're done
1031 if (search_key.objectid == search_key.offset)
1034 /* check all the names in this back reference to see
1035 * if they are in the log. if so, we allow them to stay
1036 * otherwise they must be unlinked as a conflict
1038 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1039 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
1040 while (ptr < ptr_end) {
1041 victim_ref = (struct btrfs_inode_ref *)ptr;
1042 victim_name_len = btrfs_inode_ref_name_len(leaf,
1044 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1048 read_extent_buffer(leaf, victim_name,
1049 (unsigned long)(victim_ref + 1),
1052 if (!backref_in_log(log_root, &search_key,
1056 inc_nlink(&inode->vfs_inode);
1057 btrfs_release_path(path);
1059 ret = btrfs_unlink_inode(trans, root, dir, inode,
1060 victim_name, victim_name_len);
1064 ret = btrfs_run_delayed_items(trans, fs_info);
1072 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
1076 * NOTE: we have searched root tree and checked the
1077 * corresponding ref, it does not need to check again.
1081 btrfs_release_path(path);
1083 /* Same search but for extended refs */
1084 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
1085 inode_objectid, parent_objectid, 0,
1087 if (!IS_ERR_OR_NULL(extref)) {
1091 struct inode *victim_parent;
1093 leaf = path->nodes[0];
1095 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1096 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1098 while (cur_offset < item_size) {
1099 extref = (struct btrfs_inode_extref *)(base + cur_offset);
1101 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1103 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1106 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1109 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1112 search_key.objectid = inode_objectid;
1113 search_key.type = BTRFS_INODE_EXTREF_KEY;
1114 search_key.offset = btrfs_extref_hash(parent_objectid,
1118 if (!backref_in_log(log_root, &search_key,
1119 parent_objectid, victim_name,
1122 victim_parent = read_one_inode(root,
1124 if (victim_parent) {
1125 inc_nlink(&inode->vfs_inode);
1126 btrfs_release_path(path);
1128 ret = btrfs_unlink_inode(trans, root,
1129 BTRFS_I(victim_parent),
1134 ret = btrfs_run_delayed_items(
1138 iput(victim_parent);
1149 cur_offset += victim_name_len + sizeof(*extref);
1153 btrfs_release_path(path);
1155 /* look for a conflicting sequence number */
1156 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1157 ref_index, name, namelen, 0);
1158 if (di && !IS_ERR(di)) {
1159 ret = drop_one_dir_item(trans, root, path, dir, di);
1163 btrfs_release_path(path);
1165 /* look for a conflicing name */
1166 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1168 if (di && !IS_ERR(di)) {
1169 ret = drop_one_dir_item(trans, root, path, dir, di);
1173 btrfs_release_path(path);
1178 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1179 u32 *namelen, char **name, u64 *index,
1180 u64 *parent_objectid)
1182 struct btrfs_inode_extref *extref;
1184 extref = (struct btrfs_inode_extref *)ref_ptr;
1186 *namelen = btrfs_inode_extref_name_len(eb, extref);
1187 *name = kmalloc(*namelen, GFP_NOFS);
1191 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1194 *index = btrfs_inode_extref_index(eb, extref);
1195 if (parent_objectid)
1196 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1201 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1202 u32 *namelen, char **name, u64 *index)
1204 struct btrfs_inode_ref *ref;
1206 ref = (struct btrfs_inode_ref *)ref_ptr;
1208 *namelen = btrfs_inode_ref_name_len(eb, ref);
1209 *name = kmalloc(*namelen, GFP_NOFS);
1213 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1215 *index = btrfs_inode_ref_index(eb, ref);
1221 * replay one inode back reference item found in the log tree.
1222 * eb, slot and key refer to the buffer and key found in the log tree.
1223 * root is the destination we are replaying into, and path is for temp
1224 * use by this function. (it should be released on return).
1226 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1227 struct btrfs_root *root,
1228 struct btrfs_root *log,
1229 struct btrfs_path *path,
1230 struct extent_buffer *eb, int slot,
1231 struct btrfs_key *key)
1233 struct inode *dir = NULL;
1234 struct inode *inode = NULL;
1235 unsigned long ref_ptr;
1236 unsigned long ref_end;
1240 int search_done = 0;
1241 int log_ref_ver = 0;
1242 u64 parent_objectid;
1245 int ref_struct_size;
1247 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1248 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1250 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1251 struct btrfs_inode_extref *r;
1253 ref_struct_size = sizeof(struct btrfs_inode_extref);
1255 r = (struct btrfs_inode_extref *)ref_ptr;
1256 parent_objectid = btrfs_inode_extref_parent(eb, r);
1258 ref_struct_size = sizeof(struct btrfs_inode_ref);
1259 parent_objectid = key->offset;
1261 inode_objectid = key->objectid;
1264 * it is possible that we didn't log all the parent directories
1265 * for a given inode. If we don't find the dir, just don't
1266 * copy the back ref in. The link count fixup code will take
1269 dir = read_one_inode(root, parent_objectid);
1275 inode = read_one_inode(root, inode_objectid);
1281 while (ref_ptr < ref_end) {
1283 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1284 &ref_index, &parent_objectid);
1286 * parent object can change from one array
1290 dir = read_one_inode(root, parent_objectid);
1296 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1302 /* if we already have a perfect match, we're done */
1303 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1304 btrfs_ino(BTRFS_I(inode)), ref_index,
1307 * look for a conflicting back reference in the
1308 * metadata. if we find one we have to unlink that name
1309 * of the file before we add our new link. Later on, we
1310 * overwrite any existing back reference, and we don't
1311 * want to create dangling pointers in the directory.
1315 ret = __add_inode_ref(trans, root, path, log,
1320 ref_index, name, namelen,
1329 /* insert our name */
1330 ret = btrfs_add_link(trans, BTRFS_I(dir),
1332 name, namelen, 0, ref_index);
1336 btrfs_update_inode(trans, root, inode);
1339 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1348 /* finally write the back reference in the inode */
1349 ret = overwrite_item(trans, root, path, eb, slot, key);
1351 btrfs_release_path(path);
1358 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root, u64 ino)
1363 ret = btrfs_insert_orphan_item(trans, root, ino);
1370 static int count_inode_extrefs(struct btrfs_root *root,
1371 struct btrfs_inode *inode, struct btrfs_path *path)
1375 unsigned int nlink = 0;
1378 u64 inode_objectid = btrfs_ino(inode);
1381 struct btrfs_inode_extref *extref;
1382 struct extent_buffer *leaf;
1385 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1390 leaf = path->nodes[0];
1391 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1392 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1395 while (cur_offset < item_size) {
1396 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1397 name_len = btrfs_inode_extref_name_len(leaf, extref);
1401 cur_offset += name_len + sizeof(*extref);
1405 btrfs_release_path(path);
1407 btrfs_release_path(path);
1409 if (ret < 0 && ret != -ENOENT)
1414 static int count_inode_refs(struct btrfs_root *root,
1415 struct btrfs_inode *inode, struct btrfs_path *path)
1418 struct btrfs_key key;
1419 unsigned int nlink = 0;
1421 unsigned long ptr_end;
1423 u64 ino = btrfs_ino(inode);
1426 key.type = BTRFS_INODE_REF_KEY;
1427 key.offset = (u64)-1;
1430 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1434 if (path->slots[0] == 0)
1439 btrfs_item_key_to_cpu(path->nodes[0], &key,
1441 if (key.objectid != ino ||
1442 key.type != BTRFS_INODE_REF_KEY)
1444 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1445 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1447 while (ptr < ptr_end) {
1448 struct btrfs_inode_ref *ref;
1450 ref = (struct btrfs_inode_ref *)ptr;
1451 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1453 ptr = (unsigned long)(ref + 1) + name_len;
1457 if (key.offset == 0)
1459 if (path->slots[0] > 0) {
1464 btrfs_release_path(path);
1466 btrfs_release_path(path);
1472 * There are a few corners where the link count of the file can't
1473 * be properly maintained during replay. So, instead of adding
1474 * lots of complexity to the log code, we just scan the backrefs
1475 * for any file that has been through replay.
1477 * The scan will update the link count on the inode to reflect the
1478 * number of back refs found. If it goes down to zero, the iput
1479 * will free the inode.
1481 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root,
1483 struct inode *inode)
1485 struct btrfs_path *path;
1488 u64 ino = btrfs_ino(BTRFS_I(inode));
1490 path = btrfs_alloc_path();
1494 ret = count_inode_refs(root, BTRFS_I(inode), path);
1500 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1508 if (nlink != inode->i_nlink) {
1509 set_nlink(inode, nlink);
1510 btrfs_update_inode(trans, root, inode);
1512 BTRFS_I(inode)->index_cnt = (u64)-1;
1514 if (inode->i_nlink == 0) {
1515 if (S_ISDIR(inode->i_mode)) {
1516 ret = replay_dir_deletes(trans, root, NULL, path,
1521 ret = insert_orphan_item(trans, root, ino);
1525 btrfs_free_path(path);
1529 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root,
1531 struct btrfs_path *path)
1534 struct btrfs_key key;
1535 struct inode *inode;
1537 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1538 key.type = BTRFS_ORPHAN_ITEM_KEY;
1539 key.offset = (u64)-1;
1541 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1546 if (path->slots[0] == 0)
1551 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1552 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1553 key.type != BTRFS_ORPHAN_ITEM_KEY)
1556 ret = btrfs_del_item(trans, root, path);
1560 btrfs_release_path(path);
1561 inode = read_one_inode(root, key.offset);
1565 ret = fixup_inode_link_count(trans, root, inode);
1571 * fixup on a directory may create new entries,
1572 * make sure we always look for the highset possible
1575 key.offset = (u64)-1;
1579 btrfs_release_path(path);
1585 * record a given inode in the fixup dir so we can check its link
1586 * count when replay is done. The link count is incremented here
1587 * so the inode won't go away until we check it
1589 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1590 struct btrfs_root *root,
1591 struct btrfs_path *path,
1594 struct btrfs_key key;
1596 struct inode *inode;
1598 inode = read_one_inode(root, objectid);
1602 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1603 key.type = BTRFS_ORPHAN_ITEM_KEY;
1604 key.offset = objectid;
1606 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1608 btrfs_release_path(path);
1610 if (!inode->i_nlink)
1611 set_nlink(inode, 1);
1614 ret = btrfs_update_inode(trans, root, inode);
1615 } else if (ret == -EEXIST) {
1618 BUG(); /* Logic Error */
1626 * when replaying the log for a directory, we only insert names
1627 * for inodes that actually exist. This means an fsync on a directory
1628 * does not implicitly fsync all the new files in it
1630 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1631 struct btrfs_root *root,
1632 u64 dirid, u64 index,
1633 char *name, int name_len,
1634 struct btrfs_key *location)
1636 struct inode *inode;
1640 inode = read_one_inode(root, location->objectid);
1644 dir = read_one_inode(root, dirid);
1650 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1651 name_len, 1, index);
1653 /* FIXME, put inode into FIXUP list */
1661 * Return true if an inode reference exists in the log for the given name,
1662 * inode and parent inode.
1664 static bool name_in_log_ref(struct btrfs_root *log_root,
1665 const char *name, const int name_len,
1666 const u64 dirid, const u64 ino)
1668 struct btrfs_key search_key;
1670 search_key.objectid = ino;
1671 search_key.type = BTRFS_INODE_REF_KEY;
1672 search_key.offset = dirid;
1673 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1676 search_key.type = BTRFS_INODE_EXTREF_KEY;
1677 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1678 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1685 * take a single entry in a log directory item and replay it into
1688 * if a conflicting item exists in the subdirectory already,
1689 * the inode it points to is unlinked and put into the link count
1692 * If a name from the log points to a file or directory that does
1693 * not exist in the FS, it is skipped. fsyncs on directories
1694 * do not force down inodes inside that directory, just changes to the
1695 * names or unlinks in a directory.
1697 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1698 * non-existing inode) and 1 if the name was replayed.
1700 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1701 struct btrfs_root *root,
1702 struct btrfs_path *path,
1703 struct extent_buffer *eb,
1704 struct btrfs_dir_item *di,
1705 struct btrfs_key *key)
1709 struct btrfs_dir_item *dst_di;
1710 struct btrfs_key found_key;
1711 struct btrfs_key log_key;
1716 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1717 bool name_added = false;
1719 dir = read_one_inode(root, key->objectid);
1723 name_len = btrfs_dir_name_len(eb, di);
1724 name = kmalloc(name_len, GFP_NOFS);
1730 log_type = btrfs_dir_type(eb, di);
1731 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1734 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1735 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1740 btrfs_release_path(path);
1742 if (key->type == BTRFS_DIR_ITEM_KEY) {
1743 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1745 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1746 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1755 if (IS_ERR_OR_NULL(dst_di)) {
1756 /* we need a sequence number to insert, so we only
1757 * do inserts for the BTRFS_DIR_INDEX_KEY types
1759 if (key->type != BTRFS_DIR_INDEX_KEY)
1764 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1765 /* the existing item matches the logged item */
1766 if (found_key.objectid == log_key.objectid &&
1767 found_key.type == log_key.type &&
1768 found_key.offset == log_key.offset &&
1769 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1770 update_size = false;
1775 * don't drop the conflicting directory entry if the inode
1776 * for the new entry doesn't exist
1781 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1785 if (key->type == BTRFS_DIR_INDEX_KEY)
1788 btrfs_release_path(path);
1789 if (!ret && update_size) {
1790 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1791 ret = btrfs_update_inode(trans, root, dir);
1795 if (!ret && name_added)
1800 if (name_in_log_ref(root->log_root, name, name_len,
1801 key->objectid, log_key.objectid)) {
1802 /* The dentry will be added later. */
1804 update_size = false;
1807 btrfs_release_path(path);
1808 ret = insert_one_name(trans, root, key->objectid, key->offset,
1809 name, name_len, &log_key);
1810 if (ret && ret != -ENOENT && ret != -EEXIST)
1814 update_size = false;
1820 * find all the names in a directory item and reconcile them into
1821 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1822 * one name in a directory item, but the same code gets used for
1823 * both directory index types
1825 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1826 struct btrfs_root *root,
1827 struct btrfs_path *path,
1828 struct extent_buffer *eb, int slot,
1829 struct btrfs_key *key)
1831 struct btrfs_fs_info *fs_info = root->fs_info;
1833 u32 item_size = btrfs_item_size_nr(eb, slot);
1834 struct btrfs_dir_item *di;
1837 unsigned long ptr_end;
1838 struct btrfs_path *fixup_path = NULL;
1840 ptr = btrfs_item_ptr_offset(eb, slot);
1841 ptr_end = ptr + item_size;
1842 while (ptr < ptr_end) {
1843 di = (struct btrfs_dir_item *)ptr;
1844 if (verify_dir_item(fs_info, eb, di))
1846 name_len = btrfs_dir_name_len(eb, di);
1847 ret = replay_one_name(trans, root, path, eb, di, key);
1850 ptr = (unsigned long)(di + 1);
1854 * If this entry refers to a non-directory (directories can not
1855 * have a link count > 1) and it was added in the transaction
1856 * that was not committed, make sure we fixup the link count of
1857 * the inode it the entry points to. Otherwise something like
1858 * the following would result in a directory pointing to an
1859 * inode with a wrong link that does not account for this dir
1867 * ln testdir/bar testdir/bar_link
1868 * ln testdir/foo testdir/foo_link
1869 * xfs_io -c "fsync" testdir/bar
1873 * mount fs, log replay happens
1875 * File foo would remain with a link count of 1 when it has two
1876 * entries pointing to it in the directory testdir. This would
1877 * make it impossible to ever delete the parent directory has
1878 * it would result in stale dentries that can never be deleted.
1880 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1881 struct btrfs_key di_key;
1884 fixup_path = btrfs_alloc_path();
1891 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1892 ret = link_to_fixup_dir(trans, root, fixup_path,
1899 btrfs_free_path(fixup_path);
1904 * directory replay has two parts. There are the standard directory
1905 * items in the log copied from the subvolume, and range items
1906 * created in the log while the subvolume was logged.
1908 * The range items tell us which parts of the key space the log
1909 * is authoritative for. During replay, if a key in the subvolume
1910 * directory is in a logged range item, but not actually in the log
1911 * that means it was deleted from the directory before the fsync
1912 * and should be removed.
1914 static noinline int find_dir_range(struct btrfs_root *root,
1915 struct btrfs_path *path,
1916 u64 dirid, int key_type,
1917 u64 *start_ret, u64 *end_ret)
1919 struct btrfs_key key;
1921 struct btrfs_dir_log_item *item;
1925 if (*start_ret == (u64)-1)
1928 key.objectid = dirid;
1929 key.type = key_type;
1930 key.offset = *start_ret;
1932 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1936 if (path->slots[0] == 0)
1941 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1943 if (key.type != key_type || key.objectid != dirid) {
1947 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1948 struct btrfs_dir_log_item);
1949 found_end = btrfs_dir_log_end(path->nodes[0], item);
1951 if (*start_ret >= key.offset && *start_ret <= found_end) {
1953 *start_ret = key.offset;
1954 *end_ret = found_end;
1959 /* check the next slot in the tree to see if it is a valid item */
1960 nritems = btrfs_header_nritems(path->nodes[0]);
1962 if (path->slots[0] >= nritems) {
1963 ret = btrfs_next_leaf(root, path);
1968 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1970 if (key.type != key_type || key.objectid != dirid) {
1974 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1975 struct btrfs_dir_log_item);
1976 found_end = btrfs_dir_log_end(path->nodes[0], item);
1977 *start_ret = key.offset;
1978 *end_ret = found_end;
1981 btrfs_release_path(path);
1986 * this looks for a given directory item in the log. If the directory
1987 * item is not in the log, the item is removed and the inode it points
1990 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1991 struct btrfs_root *root,
1992 struct btrfs_root *log,
1993 struct btrfs_path *path,
1994 struct btrfs_path *log_path,
1996 struct btrfs_key *dir_key)
1998 struct btrfs_fs_info *fs_info = root->fs_info;
2000 struct extent_buffer *eb;
2003 struct btrfs_dir_item *di;
2004 struct btrfs_dir_item *log_di;
2007 unsigned long ptr_end;
2009 struct inode *inode;
2010 struct btrfs_key location;
2013 eb = path->nodes[0];
2014 slot = path->slots[0];
2015 item_size = btrfs_item_size_nr(eb, slot);
2016 ptr = btrfs_item_ptr_offset(eb, slot);
2017 ptr_end = ptr + item_size;
2018 while (ptr < ptr_end) {
2019 di = (struct btrfs_dir_item *)ptr;
2020 if (verify_dir_item(fs_info, eb, di)) {
2025 name_len = btrfs_dir_name_len(eb, di);
2026 name = kmalloc(name_len, GFP_NOFS);
2031 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2034 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2035 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2038 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2039 log_di = btrfs_lookup_dir_index_item(trans, log,
2045 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2046 btrfs_dir_item_key_to_cpu(eb, di, &location);
2047 btrfs_release_path(path);
2048 btrfs_release_path(log_path);
2049 inode = read_one_inode(root, location.objectid);
2055 ret = link_to_fixup_dir(trans, root,
2056 path, location.objectid);
2064 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2065 BTRFS_I(inode), name, name_len);
2067 ret = btrfs_run_delayed_items(trans, fs_info);
2073 /* there might still be more names under this key
2074 * check and repeat if required
2076 ret = btrfs_search_slot(NULL, root, dir_key, path,
2082 } else if (IS_ERR(log_di)) {
2084 return PTR_ERR(log_di);
2086 btrfs_release_path(log_path);
2089 ptr = (unsigned long)(di + 1);
2094 btrfs_release_path(path);
2095 btrfs_release_path(log_path);
2099 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_root *log,
2102 struct btrfs_path *path,
2105 struct btrfs_key search_key;
2106 struct btrfs_path *log_path;
2111 log_path = btrfs_alloc_path();
2115 search_key.objectid = ino;
2116 search_key.type = BTRFS_XATTR_ITEM_KEY;
2117 search_key.offset = 0;
2119 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2123 nritems = btrfs_header_nritems(path->nodes[0]);
2124 for (i = path->slots[0]; i < nritems; i++) {
2125 struct btrfs_key key;
2126 struct btrfs_dir_item *di;
2127 struct btrfs_dir_item *log_di;
2131 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2132 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2137 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2138 total_size = btrfs_item_size_nr(path->nodes[0], i);
2140 while (cur < total_size) {
2141 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2142 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2143 u32 this_len = sizeof(*di) + name_len + data_len;
2146 name = kmalloc(name_len, GFP_NOFS);
2151 read_extent_buffer(path->nodes[0], name,
2152 (unsigned long)(di + 1), name_len);
2154 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2156 btrfs_release_path(log_path);
2158 /* Doesn't exist in log tree, so delete it. */
2159 btrfs_release_path(path);
2160 di = btrfs_lookup_xattr(trans, root, path, ino,
2161 name, name_len, -1);
2168 ret = btrfs_delete_one_dir_name(trans, root,
2172 btrfs_release_path(path);
2177 if (IS_ERR(log_di)) {
2178 ret = PTR_ERR(log_di);
2182 di = (struct btrfs_dir_item *)((char *)di + this_len);
2185 ret = btrfs_next_leaf(root, path);
2191 btrfs_free_path(log_path);
2192 btrfs_release_path(path);
2198 * deletion replay happens before we copy any new directory items
2199 * out of the log or out of backreferences from inodes. It
2200 * scans the log to find ranges of keys that log is authoritative for,
2201 * and then scans the directory to find items in those ranges that are
2202 * not present in the log.
2204 * Anything we don't find in the log is unlinked and removed from the
2207 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2208 struct btrfs_root *root,
2209 struct btrfs_root *log,
2210 struct btrfs_path *path,
2211 u64 dirid, int del_all)
2215 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2217 struct btrfs_key dir_key;
2218 struct btrfs_key found_key;
2219 struct btrfs_path *log_path;
2222 dir_key.objectid = dirid;
2223 dir_key.type = BTRFS_DIR_ITEM_KEY;
2224 log_path = btrfs_alloc_path();
2228 dir = read_one_inode(root, dirid);
2229 /* it isn't an error if the inode isn't there, that can happen
2230 * because we replay the deletes before we copy in the inode item
2234 btrfs_free_path(log_path);
2242 range_end = (u64)-1;
2244 ret = find_dir_range(log, path, dirid, key_type,
2245 &range_start, &range_end);
2250 dir_key.offset = range_start;
2253 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2258 nritems = btrfs_header_nritems(path->nodes[0]);
2259 if (path->slots[0] >= nritems) {
2260 ret = btrfs_next_leaf(root, path);
2264 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2266 if (found_key.objectid != dirid ||
2267 found_key.type != dir_key.type)
2270 if (found_key.offset > range_end)
2273 ret = check_item_in_log(trans, root, log, path,
2278 if (found_key.offset == (u64)-1)
2280 dir_key.offset = found_key.offset + 1;
2282 btrfs_release_path(path);
2283 if (range_end == (u64)-1)
2285 range_start = range_end + 1;
2290 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2291 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2292 dir_key.type = BTRFS_DIR_INDEX_KEY;
2293 btrfs_release_path(path);
2297 btrfs_release_path(path);
2298 btrfs_free_path(log_path);
2304 * the process_func used to replay items from the log tree. This
2305 * gets called in two different stages. The first stage just looks
2306 * for inodes and makes sure they are all copied into the subvolume.
2308 * The second stage copies all the other item types from the log into
2309 * the subvolume. The two stage approach is slower, but gets rid of
2310 * lots of complexity around inodes referencing other inodes that exist
2311 * only in the log (references come from either directory items or inode
2314 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2315 struct walk_control *wc, u64 gen)
2318 struct btrfs_path *path;
2319 struct btrfs_root *root = wc->replay_dest;
2320 struct btrfs_key key;
2325 ret = btrfs_read_buffer(eb, gen);
2329 level = btrfs_header_level(eb);
2334 path = btrfs_alloc_path();
2338 nritems = btrfs_header_nritems(eb);
2339 for (i = 0; i < nritems; i++) {
2340 btrfs_item_key_to_cpu(eb, &key, i);
2342 /* inode keys are done during the first stage */
2343 if (key.type == BTRFS_INODE_ITEM_KEY &&
2344 wc->stage == LOG_WALK_REPLAY_INODES) {
2345 struct btrfs_inode_item *inode_item;
2348 inode_item = btrfs_item_ptr(eb, i,
2349 struct btrfs_inode_item);
2350 ret = replay_xattr_deletes(wc->trans, root, log,
2351 path, key.objectid);
2354 mode = btrfs_inode_mode(eb, inode_item);
2355 if (S_ISDIR(mode)) {
2356 ret = replay_dir_deletes(wc->trans,
2357 root, log, path, key.objectid, 0);
2361 ret = overwrite_item(wc->trans, root, path,
2366 /* for regular files, make sure corresponding
2367 * orphan item exist. extents past the new EOF
2368 * will be truncated later by orphan cleanup.
2370 if (S_ISREG(mode)) {
2371 ret = insert_orphan_item(wc->trans, root,
2377 ret = link_to_fixup_dir(wc->trans, root,
2378 path, key.objectid);
2383 if (key.type == BTRFS_DIR_INDEX_KEY &&
2384 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2385 ret = replay_one_dir_item(wc->trans, root, path,
2391 if (wc->stage < LOG_WALK_REPLAY_ALL)
2394 /* these keys are simply copied */
2395 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2396 ret = overwrite_item(wc->trans, root, path,
2400 } else if (key.type == BTRFS_INODE_REF_KEY ||
2401 key.type == BTRFS_INODE_EXTREF_KEY) {
2402 ret = add_inode_ref(wc->trans, root, log, path,
2404 if (ret && ret != -ENOENT)
2407 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2408 ret = replay_one_extent(wc->trans, root, path,
2412 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2413 ret = replay_one_dir_item(wc->trans, root, path,
2419 btrfs_free_path(path);
2423 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2424 struct btrfs_root *root,
2425 struct btrfs_path *path, int *level,
2426 struct walk_control *wc)
2428 struct btrfs_fs_info *fs_info = root->fs_info;
2432 struct extent_buffer *next;
2433 struct extent_buffer *cur;
2434 struct extent_buffer *parent;
2438 WARN_ON(*level < 0);
2439 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2441 while (*level > 0) {
2442 WARN_ON(*level < 0);
2443 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2444 cur = path->nodes[*level];
2446 WARN_ON(btrfs_header_level(cur) != *level);
2448 if (path->slots[*level] >=
2449 btrfs_header_nritems(cur))
2452 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2453 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2454 blocksize = fs_info->nodesize;
2456 parent = path->nodes[*level];
2457 root_owner = btrfs_header_owner(parent);
2459 next = btrfs_find_create_tree_block(fs_info, bytenr);
2461 return PTR_ERR(next);
2464 ret = wc->process_func(root, next, wc, ptr_gen);
2466 free_extent_buffer(next);
2470 path->slots[*level]++;
2472 ret = btrfs_read_buffer(next, ptr_gen);
2474 free_extent_buffer(next);
2479 btrfs_tree_lock(next);
2480 btrfs_set_lock_blocking(next);
2481 clean_tree_block(fs_info, next);
2482 btrfs_wait_tree_block_writeback(next);
2483 btrfs_tree_unlock(next);
2486 WARN_ON(root_owner !=
2487 BTRFS_TREE_LOG_OBJECTID);
2488 ret = btrfs_free_and_pin_reserved_extent(
2492 free_extent_buffer(next);
2496 free_extent_buffer(next);
2499 ret = btrfs_read_buffer(next, ptr_gen);
2501 free_extent_buffer(next);
2505 WARN_ON(*level <= 0);
2506 if (path->nodes[*level-1])
2507 free_extent_buffer(path->nodes[*level-1]);
2508 path->nodes[*level-1] = next;
2509 *level = btrfs_header_level(next);
2510 path->slots[*level] = 0;
2513 WARN_ON(*level < 0);
2514 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2516 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2522 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2523 struct btrfs_root *root,
2524 struct btrfs_path *path, int *level,
2525 struct walk_control *wc)
2527 struct btrfs_fs_info *fs_info = root->fs_info;
2533 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2534 slot = path->slots[i];
2535 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2538 WARN_ON(*level == 0);
2541 struct extent_buffer *parent;
2542 if (path->nodes[*level] == root->node)
2543 parent = path->nodes[*level];
2545 parent = path->nodes[*level + 1];
2547 root_owner = btrfs_header_owner(parent);
2548 ret = wc->process_func(root, path->nodes[*level], wc,
2549 btrfs_header_generation(path->nodes[*level]));
2554 struct extent_buffer *next;
2556 next = path->nodes[*level];
2559 btrfs_tree_lock(next);
2560 btrfs_set_lock_blocking(next);
2561 clean_tree_block(fs_info, next);
2562 btrfs_wait_tree_block_writeback(next);
2563 btrfs_tree_unlock(next);
2566 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2567 ret = btrfs_free_and_pin_reserved_extent(
2569 path->nodes[*level]->start,
2570 path->nodes[*level]->len);
2574 free_extent_buffer(path->nodes[*level]);
2575 path->nodes[*level] = NULL;
2583 * drop the reference count on the tree rooted at 'snap'. This traverses
2584 * the tree freeing any blocks that have a ref count of zero after being
2587 static int walk_log_tree(struct btrfs_trans_handle *trans,
2588 struct btrfs_root *log, struct walk_control *wc)
2590 struct btrfs_fs_info *fs_info = log->fs_info;
2594 struct btrfs_path *path;
2597 path = btrfs_alloc_path();
2601 level = btrfs_header_level(log->node);
2603 path->nodes[level] = log->node;
2604 extent_buffer_get(log->node);
2605 path->slots[level] = 0;
2608 wret = walk_down_log_tree(trans, log, path, &level, wc);
2616 wret = walk_up_log_tree(trans, log, path, &level, wc);
2625 /* was the root node processed? if not, catch it here */
2626 if (path->nodes[orig_level]) {
2627 ret = wc->process_func(log, path->nodes[orig_level], wc,
2628 btrfs_header_generation(path->nodes[orig_level]));
2632 struct extent_buffer *next;
2634 next = path->nodes[orig_level];
2637 btrfs_tree_lock(next);
2638 btrfs_set_lock_blocking(next);
2639 clean_tree_block(fs_info, next);
2640 btrfs_wait_tree_block_writeback(next);
2641 btrfs_tree_unlock(next);
2644 WARN_ON(log->root_key.objectid !=
2645 BTRFS_TREE_LOG_OBJECTID);
2646 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2647 next->start, next->len);
2654 btrfs_free_path(path);
2659 * helper function to update the item for a given subvolumes log root
2660 * in the tree of log roots
2662 static int update_log_root(struct btrfs_trans_handle *trans,
2663 struct btrfs_root *log)
2665 struct btrfs_fs_info *fs_info = log->fs_info;
2668 if (log->log_transid == 1) {
2669 /* insert root item on the first sync */
2670 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2671 &log->root_key, &log->root_item);
2673 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2674 &log->root_key, &log->root_item);
2679 static void wait_log_commit(struct btrfs_root *root, int transid)
2682 int index = transid % 2;
2685 * we only allow two pending log transactions at a time,
2686 * so we know that if ours is more than 2 older than the
2687 * current transaction, we're done
2690 prepare_to_wait(&root->log_commit_wait[index],
2691 &wait, TASK_UNINTERRUPTIBLE);
2692 mutex_unlock(&root->log_mutex);
2694 if (root->log_transid_committed < transid &&
2695 atomic_read(&root->log_commit[index]))
2698 finish_wait(&root->log_commit_wait[index], &wait);
2699 mutex_lock(&root->log_mutex);
2700 } while (root->log_transid_committed < transid &&
2701 atomic_read(&root->log_commit[index]));
2704 static void wait_for_writer(struct btrfs_root *root)
2708 while (atomic_read(&root->log_writers)) {
2709 prepare_to_wait(&root->log_writer_wait,
2710 &wait, TASK_UNINTERRUPTIBLE);
2711 mutex_unlock(&root->log_mutex);
2712 if (atomic_read(&root->log_writers))
2714 finish_wait(&root->log_writer_wait, &wait);
2715 mutex_lock(&root->log_mutex);
2719 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2720 struct btrfs_log_ctx *ctx)
2725 mutex_lock(&root->log_mutex);
2726 list_del_init(&ctx->list);
2727 mutex_unlock(&root->log_mutex);
2731 * Invoked in log mutex context, or be sure there is no other task which
2732 * can access the list.
2734 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2735 int index, int error)
2737 struct btrfs_log_ctx *ctx;
2738 struct btrfs_log_ctx *safe;
2740 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2741 list_del_init(&ctx->list);
2742 ctx->log_ret = error;
2745 INIT_LIST_HEAD(&root->log_ctxs[index]);
2749 * btrfs_sync_log does sends a given tree log down to the disk and
2750 * updates the super blocks to record it. When this call is done,
2751 * you know that any inodes previously logged are safely on disk only
2754 * Any other return value means you need to call btrfs_commit_transaction.
2755 * Some of the edge cases for fsyncing directories that have had unlinks
2756 * or renames done in the past mean that sometimes the only safe
2757 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2758 * that has happened.
2760 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2761 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2767 struct btrfs_fs_info *fs_info = root->fs_info;
2768 struct btrfs_root *log = root->log_root;
2769 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2770 int log_transid = 0;
2771 struct btrfs_log_ctx root_log_ctx;
2772 struct blk_plug plug;
2774 mutex_lock(&root->log_mutex);
2775 log_transid = ctx->log_transid;
2776 if (root->log_transid_committed >= log_transid) {
2777 mutex_unlock(&root->log_mutex);
2778 return ctx->log_ret;
2781 index1 = log_transid % 2;
2782 if (atomic_read(&root->log_commit[index1])) {
2783 wait_log_commit(root, log_transid);
2784 mutex_unlock(&root->log_mutex);
2785 return ctx->log_ret;
2787 ASSERT(log_transid == root->log_transid);
2788 atomic_set(&root->log_commit[index1], 1);
2790 /* wait for previous tree log sync to complete */
2791 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2792 wait_log_commit(root, log_transid - 1);
2795 int batch = atomic_read(&root->log_batch);
2796 /* when we're on an ssd, just kick the log commit out */
2797 if (!btrfs_test_opt(fs_info, SSD) &&
2798 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2799 mutex_unlock(&root->log_mutex);
2800 schedule_timeout_uninterruptible(1);
2801 mutex_lock(&root->log_mutex);
2803 wait_for_writer(root);
2804 if (batch == atomic_read(&root->log_batch))
2808 /* bail out if we need to do a full commit */
2809 if (btrfs_need_log_full_commit(fs_info, trans)) {
2811 btrfs_free_logged_extents(log, log_transid);
2812 mutex_unlock(&root->log_mutex);
2816 if (log_transid % 2 == 0)
2817 mark = EXTENT_DIRTY;
2821 /* we start IO on all the marked extents here, but we don't actually
2822 * wait for them until later.
2824 blk_start_plug(&plug);
2825 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2827 blk_finish_plug(&plug);
2828 btrfs_abort_transaction(trans, ret);
2829 btrfs_free_logged_extents(log, log_transid);
2830 btrfs_set_log_full_commit(fs_info, trans);
2831 mutex_unlock(&root->log_mutex);
2835 btrfs_set_root_node(&log->root_item, log->node);
2837 root->log_transid++;
2838 log->log_transid = root->log_transid;
2839 root->log_start_pid = 0;
2841 * IO has been started, blocks of the log tree have WRITTEN flag set
2842 * in their headers. new modifications of the log will be written to
2843 * new positions. so it's safe to allow log writers to go in.
2845 mutex_unlock(&root->log_mutex);
2847 btrfs_init_log_ctx(&root_log_ctx, NULL);
2849 mutex_lock(&log_root_tree->log_mutex);
2850 atomic_inc(&log_root_tree->log_batch);
2851 atomic_inc(&log_root_tree->log_writers);
2853 index2 = log_root_tree->log_transid % 2;
2854 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2855 root_log_ctx.log_transid = log_root_tree->log_transid;
2857 mutex_unlock(&log_root_tree->log_mutex);
2859 ret = update_log_root(trans, log);
2861 mutex_lock(&log_root_tree->log_mutex);
2862 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2864 * Implicit memory barrier after atomic_dec_and_test
2866 if (waitqueue_active(&log_root_tree->log_writer_wait))
2867 wake_up(&log_root_tree->log_writer_wait);
2871 if (!list_empty(&root_log_ctx.list))
2872 list_del_init(&root_log_ctx.list);
2874 blk_finish_plug(&plug);
2875 btrfs_set_log_full_commit(fs_info, trans);
2877 if (ret != -ENOSPC) {
2878 btrfs_abort_transaction(trans, ret);
2879 mutex_unlock(&log_root_tree->log_mutex);
2882 btrfs_wait_tree_log_extents(log, mark);
2883 btrfs_free_logged_extents(log, log_transid);
2884 mutex_unlock(&log_root_tree->log_mutex);
2889 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2890 blk_finish_plug(&plug);
2891 list_del_init(&root_log_ctx.list);
2892 mutex_unlock(&log_root_tree->log_mutex);
2893 ret = root_log_ctx.log_ret;
2897 index2 = root_log_ctx.log_transid % 2;
2898 if (atomic_read(&log_root_tree->log_commit[index2])) {
2899 blk_finish_plug(&plug);
2900 ret = btrfs_wait_tree_log_extents(log, mark);
2901 btrfs_wait_logged_extents(trans, log, log_transid);
2902 wait_log_commit(log_root_tree,
2903 root_log_ctx.log_transid);
2904 mutex_unlock(&log_root_tree->log_mutex);
2906 ret = root_log_ctx.log_ret;
2909 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2910 atomic_set(&log_root_tree->log_commit[index2], 1);
2912 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2913 wait_log_commit(log_root_tree,
2914 root_log_ctx.log_transid - 1);
2917 wait_for_writer(log_root_tree);
2920 * now that we've moved on to the tree of log tree roots,
2921 * check the full commit flag again
2923 if (btrfs_need_log_full_commit(fs_info, trans)) {
2924 blk_finish_plug(&plug);
2925 btrfs_wait_tree_log_extents(log, mark);
2926 btrfs_free_logged_extents(log, log_transid);
2927 mutex_unlock(&log_root_tree->log_mutex);
2929 goto out_wake_log_root;
2932 ret = btrfs_write_marked_extents(fs_info,
2933 &log_root_tree->dirty_log_pages,
2934 EXTENT_DIRTY | EXTENT_NEW);
2935 blk_finish_plug(&plug);
2937 btrfs_set_log_full_commit(fs_info, trans);
2938 btrfs_abort_transaction(trans, ret);
2939 btrfs_free_logged_extents(log, log_transid);
2940 mutex_unlock(&log_root_tree->log_mutex);
2941 goto out_wake_log_root;
2943 ret = btrfs_wait_tree_log_extents(log, mark);
2945 ret = btrfs_wait_tree_log_extents(log_root_tree,
2946 EXTENT_NEW | EXTENT_DIRTY);
2948 btrfs_set_log_full_commit(fs_info, trans);
2949 btrfs_free_logged_extents(log, log_transid);
2950 mutex_unlock(&log_root_tree->log_mutex);
2951 goto out_wake_log_root;
2953 btrfs_wait_logged_extents(trans, log, log_transid);
2955 btrfs_set_super_log_root(fs_info->super_for_commit,
2956 log_root_tree->node->start);
2957 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2958 btrfs_header_level(log_root_tree->node));
2960 log_root_tree->log_transid++;
2961 mutex_unlock(&log_root_tree->log_mutex);
2964 * nobody else is going to jump in and write the the ctree
2965 * super here because the log_commit atomic below is protecting
2966 * us. We must be called with a transaction handle pinning
2967 * the running transaction open, so a full commit can't hop
2968 * in and cause problems either.
2970 ret = write_all_supers(fs_info, 1);
2972 btrfs_set_log_full_commit(fs_info, trans);
2973 btrfs_abort_transaction(trans, ret);
2974 goto out_wake_log_root;
2977 mutex_lock(&root->log_mutex);
2978 if (root->last_log_commit < log_transid)
2979 root->last_log_commit = log_transid;
2980 mutex_unlock(&root->log_mutex);
2983 mutex_lock(&log_root_tree->log_mutex);
2984 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2986 log_root_tree->log_transid_committed++;
2987 atomic_set(&log_root_tree->log_commit[index2], 0);
2988 mutex_unlock(&log_root_tree->log_mutex);
2991 * The barrier before waitqueue_active is implied by mutex_unlock
2993 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2994 wake_up(&log_root_tree->log_commit_wait[index2]);
2996 mutex_lock(&root->log_mutex);
2997 btrfs_remove_all_log_ctxs(root, index1, ret);
2998 root->log_transid_committed++;
2999 atomic_set(&root->log_commit[index1], 0);
3000 mutex_unlock(&root->log_mutex);
3003 * The barrier before waitqueue_active is implied by mutex_unlock
3005 if (waitqueue_active(&root->log_commit_wait[index1]))
3006 wake_up(&root->log_commit_wait[index1]);
3010 static void free_log_tree(struct btrfs_trans_handle *trans,
3011 struct btrfs_root *log)
3016 struct walk_control wc = {
3018 .process_func = process_one_buffer
3021 ret = walk_log_tree(trans, log, &wc);
3022 /* I don't think this can happen but just in case */
3024 btrfs_abort_transaction(trans, ret);
3027 ret = find_first_extent_bit(&log->dirty_log_pages,
3028 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3033 clear_extent_bits(&log->dirty_log_pages, start, end,
3034 EXTENT_DIRTY | EXTENT_NEW);
3038 * We may have short-circuited the log tree with the full commit logic
3039 * and left ordered extents on our list, so clear these out to keep us
3040 * from leaking inodes and memory.
3042 btrfs_free_logged_extents(log, 0);
3043 btrfs_free_logged_extents(log, 1);
3045 free_extent_buffer(log->node);
3050 * free all the extents used by the tree log. This should be called
3051 * at commit time of the full transaction
3053 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3055 if (root->log_root) {
3056 free_log_tree(trans, root->log_root);
3057 root->log_root = NULL;
3062 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3063 struct btrfs_fs_info *fs_info)
3065 if (fs_info->log_root_tree) {
3066 free_log_tree(trans, fs_info->log_root_tree);
3067 fs_info->log_root_tree = NULL;
3073 * If both a file and directory are logged, and unlinks or renames are
3074 * mixed in, we have a few interesting corners:
3076 * create file X in dir Y
3077 * link file X to X.link in dir Y
3079 * unlink file X but leave X.link
3082 * After a crash we would expect only X.link to exist. But file X
3083 * didn't get fsync'd again so the log has back refs for X and X.link.
3085 * We solve this by removing directory entries and inode backrefs from the
3086 * log when a file that was logged in the current transaction is
3087 * unlinked. Any later fsync will include the updated log entries, and
3088 * we'll be able to reconstruct the proper directory items from backrefs.
3090 * This optimizations allows us to avoid relogging the entire inode
3091 * or the entire directory.
3093 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3094 struct btrfs_root *root,
3095 const char *name, int name_len,
3096 struct btrfs_inode *dir, u64 index)
3098 struct btrfs_root *log;
3099 struct btrfs_dir_item *di;
3100 struct btrfs_path *path;
3104 u64 dir_ino = btrfs_ino(dir);
3106 if (dir->logged_trans < trans->transid)
3109 ret = join_running_log_trans(root);
3113 mutex_lock(&dir->log_mutex);
3115 log = root->log_root;
3116 path = btrfs_alloc_path();
3122 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3123 name, name_len, -1);
3129 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3130 bytes_del += name_len;
3136 btrfs_release_path(path);
3137 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3138 index, name, name_len, -1);
3144 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3145 bytes_del += name_len;
3152 /* update the directory size in the log to reflect the names
3156 struct btrfs_key key;
3158 key.objectid = dir_ino;
3160 key.type = BTRFS_INODE_ITEM_KEY;
3161 btrfs_release_path(path);
3163 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3169 struct btrfs_inode_item *item;
3172 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3173 struct btrfs_inode_item);
3174 i_size = btrfs_inode_size(path->nodes[0], item);
3175 if (i_size > bytes_del)
3176 i_size -= bytes_del;
3179 btrfs_set_inode_size(path->nodes[0], item, i_size);
3180 btrfs_mark_buffer_dirty(path->nodes[0]);
3183 btrfs_release_path(path);
3186 btrfs_free_path(path);
3188 mutex_unlock(&dir->log_mutex);
3189 if (ret == -ENOSPC) {
3190 btrfs_set_log_full_commit(root->fs_info, trans);
3193 btrfs_abort_transaction(trans, ret);
3195 btrfs_end_log_trans(root);
3200 /* see comments for btrfs_del_dir_entries_in_log */
3201 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3202 struct btrfs_root *root,
3203 const char *name, int name_len,
3204 struct btrfs_inode *inode, u64 dirid)
3206 struct btrfs_fs_info *fs_info = root->fs_info;
3207 struct btrfs_root *log;
3211 if (inode->logged_trans < trans->transid)
3214 ret = join_running_log_trans(root);
3217 log = root->log_root;
3218 mutex_lock(&inode->log_mutex);
3220 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3222 mutex_unlock(&inode->log_mutex);
3223 if (ret == -ENOSPC) {
3224 btrfs_set_log_full_commit(fs_info, trans);
3226 } else if (ret < 0 && ret != -ENOENT)
3227 btrfs_abort_transaction(trans, ret);
3228 btrfs_end_log_trans(root);
3234 * creates a range item in the log for 'dirid'. first_offset and
3235 * last_offset tell us which parts of the key space the log should
3236 * be considered authoritative for.
3238 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3239 struct btrfs_root *log,
3240 struct btrfs_path *path,
3241 int key_type, u64 dirid,
3242 u64 first_offset, u64 last_offset)
3245 struct btrfs_key key;
3246 struct btrfs_dir_log_item *item;
3248 key.objectid = dirid;
3249 key.offset = first_offset;
3250 if (key_type == BTRFS_DIR_ITEM_KEY)
3251 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3253 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3254 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3258 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3259 struct btrfs_dir_log_item);
3260 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3261 btrfs_mark_buffer_dirty(path->nodes[0]);
3262 btrfs_release_path(path);
3267 * log all the items included in the current transaction for a given
3268 * directory. This also creates the range items in the log tree required
3269 * to replay anything deleted before the fsync
3271 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3272 struct btrfs_root *root, struct btrfs_inode *inode,
3273 struct btrfs_path *path,
3274 struct btrfs_path *dst_path, int key_type,
3275 struct btrfs_log_ctx *ctx,
3276 u64 min_offset, u64 *last_offset_ret)
3278 struct btrfs_key min_key;
3279 struct btrfs_root *log = root->log_root;
3280 struct extent_buffer *src;
3285 u64 first_offset = min_offset;
3286 u64 last_offset = (u64)-1;
3287 u64 ino = btrfs_ino(inode);
3289 log = root->log_root;
3291 min_key.objectid = ino;
3292 min_key.type = key_type;
3293 min_key.offset = min_offset;
3295 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3298 * we didn't find anything from this transaction, see if there
3299 * is anything at all
3301 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3302 min_key.objectid = ino;
3303 min_key.type = key_type;
3304 min_key.offset = (u64)-1;
3305 btrfs_release_path(path);
3306 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3308 btrfs_release_path(path);
3311 ret = btrfs_previous_item(root, path, ino, key_type);
3313 /* if ret == 0 there are items for this type,
3314 * create a range to tell us the last key of this type.
3315 * otherwise, there are no items in this directory after
3316 * *min_offset, and we create a range to indicate that.
3319 struct btrfs_key tmp;
3320 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3322 if (key_type == tmp.type)
3323 first_offset = max(min_offset, tmp.offset) + 1;
3328 /* go backward to find any previous key */
3329 ret = btrfs_previous_item(root, path, ino, key_type);
3331 struct btrfs_key tmp;
3332 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3333 if (key_type == tmp.type) {
3334 first_offset = tmp.offset;
3335 ret = overwrite_item(trans, log, dst_path,
3336 path->nodes[0], path->slots[0],
3344 btrfs_release_path(path);
3346 /* find the first key from this transaction again */
3347 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3348 if (WARN_ON(ret != 0))
3352 * we have a block from this transaction, log every item in it
3353 * from our directory
3356 struct btrfs_key tmp;
3357 src = path->nodes[0];
3358 nritems = btrfs_header_nritems(src);
3359 for (i = path->slots[0]; i < nritems; i++) {
3360 struct btrfs_dir_item *di;
3362 btrfs_item_key_to_cpu(src, &min_key, i);
3364 if (min_key.objectid != ino || min_key.type != key_type)
3366 ret = overwrite_item(trans, log, dst_path, src, i,
3374 * We must make sure that when we log a directory entry,
3375 * the corresponding inode, after log replay, has a
3376 * matching link count. For example:
3382 * xfs_io -c "fsync" mydir
3384 * <mount fs and log replay>
3386 * Would result in a fsync log that when replayed, our
3387 * file inode would have a link count of 1, but we get
3388 * two directory entries pointing to the same inode.
3389 * After removing one of the names, it would not be
3390 * possible to remove the other name, which resulted
3391 * always in stale file handle errors, and would not
3392 * be possible to rmdir the parent directory, since
3393 * its i_size could never decrement to the value
3394 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3396 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3397 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3399 (btrfs_dir_transid(src, di) == trans->transid ||
3400 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3401 tmp.type != BTRFS_ROOT_ITEM_KEY)
3402 ctx->log_new_dentries = true;
3404 path->slots[0] = nritems;
3407 * look ahead to the next item and see if it is also
3408 * from this directory and from this transaction
3410 ret = btrfs_next_leaf(root, path);
3412 last_offset = (u64)-1;
3415 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3416 if (tmp.objectid != ino || tmp.type != key_type) {
3417 last_offset = (u64)-1;
3420 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3421 ret = overwrite_item(trans, log, dst_path,
3422 path->nodes[0], path->slots[0],
3427 last_offset = tmp.offset;
3432 btrfs_release_path(path);
3433 btrfs_release_path(dst_path);
3436 *last_offset_ret = last_offset;
3438 * insert the log range keys to indicate where the log
3441 ret = insert_dir_log_key(trans, log, path, key_type,
3442 ino, first_offset, last_offset);
3450 * logging directories is very similar to logging inodes, We find all the items
3451 * from the current transaction and write them to the log.
3453 * The recovery code scans the directory in the subvolume, and if it finds a
3454 * key in the range logged that is not present in the log tree, then it means
3455 * that dir entry was unlinked during the transaction.
3457 * In order for that scan to work, we must include one key smaller than
3458 * the smallest logged by this transaction and one key larger than the largest
3459 * key logged by this transaction.
3461 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3462 struct btrfs_root *root, struct btrfs_inode *inode,
3463 struct btrfs_path *path,
3464 struct btrfs_path *dst_path,
3465 struct btrfs_log_ctx *ctx)
3470 int key_type = BTRFS_DIR_ITEM_KEY;
3476 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3477 ctx, min_key, &max_key);
3480 if (max_key == (u64)-1)
3482 min_key = max_key + 1;
3485 if (key_type == BTRFS_DIR_ITEM_KEY) {
3486 key_type = BTRFS_DIR_INDEX_KEY;
3493 * a helper function to drop items from the log before we relog an
3494 * inode. max_key_type indicates the highest item type to remove.
3495 * This cannot be run for file data extents because it does not
3496 * free the extents they point to.
3498 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3499 struct btrfs_root *log,
3500 struct btrfs_path *path,
3501 u64 objectid, int max_key_type)
3504 struct btrfs_key key;
3505 struct btrfs_key found_key;
3508 key.objectid = objectid;
3509 key.type = max_key_type;
3510 key.offset = (u64)-1;
3513 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3514 BUG_ON(ret == 0); /* Logic error */
3518 if (path->slots[0] == 0)
3522 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3525 if (found_key.objectid != objectid)
3528 found_key.offset = 0;
3530 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3533 ret = btrfs_del_items(trans, log, path, start_slot,
3534 path->slots[0] - start_slot + 1);
3536 * If start slot isn't 0 then we don't need to re-search, we've
3537 * found the last guy with the objectid in this tree.
3539 if (ret || start_slot != 0)
3541 btrfs_release_path(path);
3543 btrfs_release_path(path);
3549 static void fill_inode_item(struct btrfs_trans_handle *trans,
3550 struct extent_buffer *leaf,
3551 struct btrfs_inode_item *item,
3552 struct inode *inode, int log_inode_only,
3555 struct btrfs_map_token token;
3557 btrfs_init_map_token(&token);
3559 if (log_inode_only) {
3560 /* set the generation to zero so the recover code
3561 * can tell the difference between an logging
3562 * just to say 'this inode exists' and a logging
3563 * to say 'update this inode with these values'
3565 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3566 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3568 btrfs_set_token_inode_generation(leaf, item,
3569 BTRFS_I(inode)->generation,
3571 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3574 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3575 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3576 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3577 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3579 btrfs_set_token_timespec_sec(leaf, &item->atime,
3580 inode->i_atime.tv_sec, &token);
3581 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3582 inode->i_atime.tv_nsec, &token);
3584 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3585 inode->i_mtime.tv_sec, &token);
3586 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3587 inode->i_mtime.tv_nsec, &token);
3589 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3590 inode->i_ctime.tv_sec, &token);
3591 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3592 inode->i_ctime.tv_nsec, &token);
3594 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3597 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3598 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3599 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3600 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3601 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3604 static int log_inode_item(struct btrfs_trans_handle *trans,
3605 struct btrfs_root *log, struct btrfs_path *path,
3606 struct btrfs_inode *inode)
3608 struct btrfs_inode_item *inode_item;
3611 ret = btrfs_insert_empty_item(trans, log, path,
3612 &inode->location, sizeof(*inode_item));
3613 if (ret && ret != -EEXIST)
3615 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3616 struct btrfs_inode_item);
3617 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3619 btrfs_release_path(path);
3623 static noinline int copy_items(struct btrfs_trans_handle *trans,
3624 struct btrfs_inode *inode,
3625 struct btrfs_path *dst_path,
3626 struct btrfs_path *src_path, u64 *last_extent,
3627 int start_slot, int nr, int inode_only,
3630 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3631 unsigned long src_offset;
3632 unsigned long dst_offset;
3633 struct btrfs_root *log = inode->root->log_root;
3634 struct btrfs_file_extent_item *extent;
3635 struct btrfs_inode_item *inode_item;
3636 struct extent_buffer *src = src_path->nodes[0];
3637 struct btrfs_key first_key, last_key, key;
3639 struct btrfs_key *ins_keys;
3643 struct list_head ordered_sums;
3644 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3645 bool has_extents = false;
3646 bool need_find_last_extent = true;
3649 INIT_LIST_HEAD(&ordered_sums);
3651 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3652 nr * sizeof(u32), GFP_NOFS);
3656 first_key.objectid = (u64)-1;
3658 ins_sizes = (u32 *)ins_data;
3659 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3661 for (i = 0; i < nr; i++) {
3662 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3663 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3665 ret = btrfs_insert_empty_items(trans, log, dst_path,
3666 ins_keys, ins_sizes, nr);
3672 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3673 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3674 dst_path->slots[0]);
3676 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3678 if ((i == (nr - 1)))
3679 last_key = ins_keys[i];
3681 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3682 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3684 struct btrfs_inode_item);
3685 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3687 inode_only == LOG_INODE_EXISTS,
3690 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3691 src_offset, ins_sizes[i]);
3695 * We set need_find_last_extent here in case we know we were
3696 * processing other items and then walk into the first extent in
3697 * the inode. If we don't hit an extent then nothing changes,
3698 * we'll do the last search the next time around.
3700 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3702 if (first_key.objectid == (u64)-1)
3703 first_key = ins_keys[i];
3705 need_find_last_extent = false;
3708 /* take a reference on file data extents so that truncates
3709 * or deletes of this inode don't have to relog the inode
3712 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3715 extent = btrfs_item_ptr(src, start_slot + i,
3716 struct btrfs_file_extent_item);
3718 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3721 found_type = btrfs_file_extent_type(src, extent);
3722 if (found_type == BTRFS_FILE_EXTENT_REG) {
3724 ds = btrfs_file_extent_disk_bytenr(src,
3726 /* ds == 0 is a hole */
3730 dl = btrfs_file_extent_disk_num_bytes(src,
3732 cs = btrfs_file_extent_offset(src, extent);
3733 cl = btrfs_file_extent_num_bytes(src,
3735 if (btrfs_file_extent_compression(src,
3741 ret = btrfs_lookup_csums_range(
3743 ds + cs, ds + cs + cl - 1,
3746 btrfs_release_path(dst_path);
3754 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3755 btrfs_release_path(dst_path);
3759 * we have to do this after the loop above to avoid changing the
3760 * log tree while trying to change the log tree.
3763 while (!list_empty(&ordered_sums)) {
3764 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3765 struct btrfs_ordered_sum,
3768 ret = btrfs_csum_file_blocks(trans, log, sums);
3769 list_del(&sums->list);
3776 if (need_find_last_extent && *last_extent == first_key.offset) {
3778 * We don't have any leafs between our current one and the one
3779 * we processed before that can have file extent items for our
3780 * inode (and have a generation number smaller than our current
3783 need_find_last_extent = false;
3787 * Because we use btrfs_search_forward we could skip leaves that were
3788 * not modified and then assume *last_extent is valid when it really
3789 * isn't. So back up to the previous leaf and read the end of the last
3790 * extent before we go and fill in holes.
3792 if (need_find_last_extent) {
3795 ret = btrfs_prev_leaf(inode->root, src_path);
3800 if (src_path->slots[0])
3801 src_path->slots[0]--;
3802 src = src_path->nodes[0];
3803 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3804 if (key.objectid != btrfs_ino(inode) ||
3805 key.type != BTRFS_EXTENT_DATA_KEY)
3807 extent = btrfs_item_ptr(src, src_path->slots[0],
3808 struct btrfs_file_extent_item);
3809 if (btrfs_file_extent_type(src, extent) ==
3810 BTRFS_FILE_EXTENT_INLINE) {
3811 len = btrfs_file_extent_inline_len(src,
3814 *last_extent = ALIGN(key.offset + len,
3815 fs_info->sectorsize);
3817 len = btrfs_file_extent_num_bytes(src, extent);
3818 *last_extent = key.offset + len;
3822 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3823 * things could have happened
3825 * 1) A merge could have happened, so we could currently be on a leaf
3826 * that holds what we were copying in the first place.
3827 * 2) A split could have happened, and now not all of the items we want
3828 * are on the same leaf.
3830 * So we need to adjust how we search for holes, we need to drop the
3831 * path and re-search for the first extent key we found, and then walk
3832 * forward until we hit the last one we copied.
3834 if (need_find_last_extent) {
3835 /* btrfs_prev_leaf could return 1 without releasing the path */
3836 btrfs_release_path(src_path);
3837 ret = btrfs_search_slot(NULL, inode->root, &first_key,
3842 src = src_path->nodes[0];
3843 i = src_path->slots[0];
3849 * Ok so here we need to go through and fill in any holes we may have
3850 * to make sure that holes are punched for those areas in case they had
3851 * extents previously.
3857 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3858 ret = btrfs_next_leaf(inode->root, src_path);
3862 src = src_path->nodes[0];
3866 btrfs_item_key_to_cpu(src, &key, i);
3867 if (!btrfs_comp_cpu_keys(&key, &last_key))
3869 if (key.objectid != btrfs_ino(inode) ||
3870 key.type != BTRFS_EXTENT_DATA_KEY) {
3874 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3875 if (btrfs_file_extent_type(src, extent) ==
3876 BTRFS_FILE_EXTENT_INLINE) {
3877 len = btrfs_file_extent_inline_len(src, i, extent);
3878 extent_end = ALIGN(key.offset + len,
3879 fs_info->sectorsize);
3881 len = btrfs_file_extent_num_bytes(src, extent);
3882 extent_end = key.offset + len;
3886 if (*last_extent == key.offset) {
3887 *last_extent = extent_end;
3890 offset = *last_extent;
3891 len = key.offset - *last_extent;
3892 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3893 offset, 0, 0, len, 0, len, 0, 0, 0);
3896 *last_extent = extent_end;
3899 * Need to let the callers know we dropped the path so they should
3902 if (!ret && need_find_last_extent)
3907 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3909 struct extent_map *em1, *em2;
3911 em1 = list_entry(a, struct extent_map, list);
3912 em2 = list_entry(b, struct extent_map, list);
3914 if (em1->start < em2->start)
3916 else if (em1->start > em2->start)
3921 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3922 struct inode *inode,
3923 struct btrfs_root *root,
3924 const struct extent_map *em,
3925 const struct list_head *logged_list,
3926 bool *ordered_io_error)
3928 struct btrfs_fs_info *fs_info = root->fs_info;
3929 struct btrfs_ordered_extent *ordered;
3930 struct btrfs_root *log = root->log_root;
3931 u64 mod_start = em->mod_start;
3932 u64 mod_len = em->mod_len;
3933 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3936 LIST_HEAD(ordered_sums);
3939 *ordered_io_error = false;
3941 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3942 em->block_start == EXTENT_MAP_HOLE)
3946 * Wait far any ordered extent that covers our extent map. If it
3947 * finishes without an error, first check and see if our csums are on
3948 * our outstanding ordered extents.
3950 list_for_each_entry(ordered, logged_list, log_list) {
3951 struct btrfs_ordered_sum *sum;
3956 if (ordered->file_offset + ordered->len <= mod_start ||
3957 mod_start + mod_len <= ordered->file_offset)
3960 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3961 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3962 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3963 const u64 start = ordered->file_offset;
3964 const u64 end = ordered->file_offset + ordered->len - 1;
3966 WARN_ON(ordered->inode != inode);
3967 filemap_fdatawrite_range(inode->i_mapping, start, end);
3970 wait_event(ordered->wait,
3971 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3972 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3974 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3976 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3977 * i_mapping flags, so that the next fsync won't get
3978 * an outdated io error too.
3980 filemap_check_errors(inode->i_mapping);
3981 *ordered_io_error = true;
3985 * We are going to copy all the csums on this ordered extent, so
3986 * go ahead and adjust mod_start and mod_len in case this
3987 * ordered extent has already been logged.
3989 if (ordered->file_offset > mod_start) {
3990 if (ordered->file_offset + ordered->len >=
3991 mod_start + mod_len)
3992 mod_len = ordered->file_offset - mod_start;
3994 * If we have this case
3996 * |--------- logged extent ---------|
3997 * |----- ordered extent ----|
3999 * Just don't mess with mod_start and mod_len, we'll
4000 * just end up logging more csums than we need and it
4004 if (ordered->file_offset + ordered->len <
4005 mod_start + mod_len) {
4006 mod_len = (mod_start + mod_len) -
4007 (ordered->file_offset + ordered->len);
4008 mod_start = ordered->file_offset +
4019 * To keep us from looping for the above case of an ordered
4020 * extent that falls inside of the logged extent.
4022 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4026 list_for_each_entry(sum, &ordered->list, list) {
4027 ret = btrfs_csum_file_blocks(trans, log, sum);
4033 if (*ordered_io_error || !mod_len || ret || skip_csum)
4036 if (em->compress_type) {
4038 csum_len = max(em->block_len, em->orig_block_len);
4040 csum_offset = mod_start - em->start;
4044 /* block start is already adjusted for the file extent offset. */
4045 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4046 em->block_start + csum_offset,
4047 em->block_start + csum_offset +
4048 csum_len - 1, &ordered_sums, 0);
4052 while (!list_empty(&ordered_sums)) {
4053 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4054 struct btrfs_ordered_sum,
4057 ret = btrfs_csum_file_blocks(trans, log, sums);
4058 list_del(&sums->list);
4065 static int log_one_extent(struct btrfs_trans_handle *trans,
4066 struct btrfs_inode *inode, struct btrfs_root *root,
4067 const struct extent_map *em,
4068 struct btrfs_path *path,
4069 const struct list_head *logged_list,
4070 struct btrfs_log_ctx *ctx)
4072 struct btrfs_root *log = root->log_root;
4073 struct btrfs_file_extent_item *fi;
4074 struct extent_buffer *leaf;
4075 struct btrfs_map_token token;
4076 struct btrfs_key key;
4077 u64 extent_offset = em->start - em->orig_start;
4080 int extent_inserted = 0;
4081 bool ordered_io_err = false;
4083 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em,
4084 logged_list, &ordered_io_err);
4088 if (ordered_io_err) {
4093 btrfs_init_map_token(&token);
4095 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4096 em->start + em->len, NULL, 0, 1,
4097 sizeof(*fi), &extent_inserted);
4101 if (!extent_inserted) {
4102 key.objectid = btrfs_ino(inode);
4103 key.type = BTRFS_EXTENT_DATA_KEY;
4104 key.offset = em->start;
4106 ret = btrfs_insert_empty_item(trans, log, path, &key,
4111 leaf = path->nodes[0];
4112 fi = btrfs_item_ptr(leaf, path->slots[0],
4113 struct btrfs_file_extent_item);
4115 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4117 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4118 btrfs_set_token_file_extent_type(leaf, fi,
4119 BTRFS_FILE_EXTENT_PREALLOC,
4122 btrfs_set_token_file_extent_type(leaf, fi,
4123 BTRFS_FILE_EXTENT_REG,
4126 block_len = max(em->block_len, em->orig_block_len);
4127 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4128 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4131 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4133 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4134 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4136 extent_offset, &token);
4137 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4140 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4141 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4145 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4146 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4147 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4148 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4150 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4151 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4152 btrfs_mark_buffer_dirty(leaf);
4154 btrfs_release_path(path);
4159 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4160 struct btrfs_root *root,
4161 struct btrfs_inode *inode,
4162 struct btrfs_path *path,
4163 struct list_head *logged_list,
4164 struct btrfs_log_ctx *ctx,
4168 struct extent_map *em, *n;
4169 struct list_head extents;
4170 struct extent_map_tree *tree = &inode->extent_tree;
4175 INIT_LIST_HEAD(&extents);
4177 down_write(&inode->dio_sem);
4178 write_lock(&tree->lock);
4179 test_gen = root->fs_info->last_trans_committed;
4181 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4182 list_del_init(&em->list);
4185 * Just an arbitrary number, this can be really CPU intensive
4186 * once we start getting a lot of extents, and really once we
4187 * have a bunch of extents we just want to commit since it will
4190 if (++num > 32768) {
4191 list_del_init(&tree->modified_extents);
4196 if (em->generation <= test_gen)
4198 /* Need a ref to keep it from getting evicted from cache */
4199 refcount_inc(&em->refs);
4200 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4201 list_add_tail(&em->list, &extents);
4205 list_sort(NULL, &extents, extent_cmp);
4206 btrfs_get_logged_extents(inode, logged_list, start, end);
4208 * Some ordered extents started by fsync might have completed
4209 * before we could collect them into the list logged_list, which
4210 * means they're gone, not in our logged_list nor in the inode's
4211 * ordered tree. We want the application/user space to know an
4212 * error happened while attempting to persist file data so that
4213 * it can take proper action. If such error happened, we leave
4214 * without writing to the log tree and the fsync must report the
4215 * file data write error and not commit the current transaction.
4217 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4221 while (!list_empty(&extents)) {
4222 em = list_entry(extents.next, struct extent_map, list);
4224 list_del_init(&em->list);
4227 * If we had an error we just need to delete everybody from our
4231 clear_em_logging(tree, em);
4232 free_extent_map(em);
4236 write_unlock(&tree->lock);
4238 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4240 write_lock(&tree->lock);
4241 clear_em_logging(tree, em);
4242 free_extent_map(em);
4244 WARN_ON(!list_empty(&extents));
4245 write_unlock(&tree->lock);
4246 up_write(&inode->dio_sem);
4248 btrfs_release_path(path);
4252 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4253 struct btrfs_path *path, u64 *size_ret)
4255 struct btrfs_key key;
4258 key.objectid = btrfs_ino(inode);
4259 key.type = BTRFS_INODE_ITEM_KEY;
4262 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4265 } else if (ret > 0) {
4268 struct btrfs_inode_item *item;
4270 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4271 struct btrfs_inode_item);
4272 *size_ret = btrfs_inode_size(path->nodes[0], item);
4275 btrfs_release_path(path);
4280 * At the moment we always log all xattrs. This is to figure out at log replay
4281 * time which xattrs must have their deletion replayed. If a xattr is missing
4282 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4283 * because if a xattr is deleted, the inode is fsynced and a power failure
4284 * happens, causing the log to be replayed the next time the fs is mounted,
4285 * we want the xattr to not exist anymore (same behaviour as other filesystems
4286 * with a journal, ext3/4, xfs, f2fs, etc).
4288 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4289 struct btrfs_root *root,
4290 struct btrfs_inode *inode,
4291 struct btrfs_path *path,
4292 struct btrfs_path *dst_path)
4295 struct btrfs_key key;
4296 const u64 ino = btrfs_ino(inode);
4301 key.type = BTRFS_XATTR_ITEM_KEY;
4304 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4309 int slot = path->slots[0];
4310 struct extent_buffer *leaf = path->nodes[0];
4311 int nritems = btrfs_header_nritems(leaf);
4313 if (slot >= nritems) {
4315 u64 last_extent = 0;
4317 ret = copy_items(trans, inode, dst_path, path,
4318 &last_extent, start_slot,
4320 /* can't be 1, extent items aren't processed */
4326 ret = btrfs_next_leaf(root, path);
4334 btrfs_item_key_to_cpu(leaf, &key, slot);
4335 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4345 u64 last_extent = 0;
4347 ret = copy_items(trans, inode, dst_path, path,
4348 &last_extent, start_slot,
4350 /* can't be 1, extent items aren't processed */
4360 * If the no holes feature is enabled we need to make sure any hole between the
4361 * last extent and the i_size of our inode is explicitly marked in the log. This
4362 * is to make sure that doing something like:
4364 * 1) create file with 128Kb of data
4365 * 2) truncate file to 64Kb
4366 * 3) truncate file to 256Kb
4368 * 5) <crash/power failure>
4369 * 6) mount fs and trigger log replay
4371 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4372 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4373 * file correspond to a hole. The presence of explicit holes in a log tree is
4374 * what guarantees that log replay will remove/adjust file extent items in the
4377 * Here we do not need to care about holes between extents, that is already done
4378 * by copy_items(). We also only need to do this in the full sync path, where we
4379 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4380 * lookup the list of modified extent maps and if any represents a hole, we
4381 * insert a corresponding extent representing a hole in the log tree.
4383 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4384 struct btrfs_root *root,
4385 struct btrfs_inode *inode,
4386 struct btrfs_path *path)
4388 struct btrfs_fs_info *fs_info = root->fs_info;
4390 struct btrfs_key key;
4393 struct extent_buffer *leaf;
4394 struct btrfs_root *log = root->log_root;
4395 const u64 ino = btrfs_ino(inode);
4396 const u64 i_size = i_size_read(&inode->vfs_inode);
4398 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4402 key.type = BTRFS_EXTENT_DATA_KEY;
4403 key.offset = (u64)-1;
4405 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4410 ASSERT(path->slots[0] > 0);
4412 leaf = path->nodes[0];
4413 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4415 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4416 /* inode does not have any extents */
4420 struct btrfs_file_extent_item *extent;
4424 * If there's an extent beyond i_size, an explicit hole was
4425 * already inserted by copy_items().
4427 if (key.offset >= i_size)
4430 extent = btrfs_item_ptr(leaf, path->slots[0],
4431 struct btrfs_file_extent_item);
4433 if (btrfs_file_extent_type(leaf, extent) ==
4434 BTRFS_FILE_EXTENT_INLINE) {
4435 len = btrfs_file_extent_inline_len(leaf,
4438 ASSERT(len == i_size);
4442 len = btrfs_file_extent_num_bytes(leaf, extent);
4443 /* Last extent goes beyond i_size, no need to log a hole. */
4444 if (key.offset + len > i_size)
4446 hole_start = key.offset + len;
4447 hole_size = i_size - hole_start;
4449 btrfs_release_path(path);
4451 /* Last extent ends at i_size. */
4455 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4456 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4457 hole_size, 0, hole_size, 0, 0, 0);
4462 * When we are logging a new inode X, check if it doesn't have a reference that
4463 * matches the reference from some other inode Y created in a past transaction
4464 * and that was renamed in the current transaction. If we don't do this, then at
4465 * log replay time we can lose inode Y (and all its files if it's a directory):
4468 * echo "hello world" > /mnt/x/foobar
4471 * mkdir /mnt/x # or touch /mnt/x
4472 * xfs_io -c fsync /mnt/x
4474 * mount fs, trigger log replay
4476 * After the log replay procedure, we would lose the first directory and all its
4477 * files (file foobar).
4478 * For the case where inode Y is not a directory we simply end up losing it:
4480 * echo "123" > /mnt/foo
4482 * mv /mnt/foo /mnt/bar
4483 * echo "abc" > /mnt/foo
4484 * xfs_io -c fsync /mnt/foo
4487 * We also need this for cases where a snapshot entry is replaced by some other
4488 * entry (file or directory) otherwise we end up with an unreplayable log due to
4489 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4490 * if it were a regular entry:
4493 * btrfs subvolume snapshot /mnt /mnt/x/snap
4494 * btrfs subvolume delete /mnt/x/snap
4497 * fsync /mnt/x or fsync some new file inside it
4500 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4501 * the same transaction.
4503 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4505 const struct btrfs_key *key,
4506 struct btrfs_inode *inode,
4510 struct btrfs_path *search_path;
4513 u32 item_size = btrfs_item_size_nr(eb, slot);
4515 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4517 search_path = btrfs_alloc_path();
4520 search_path->search_commit_root = 1;
4521 search_path->skip_locking = 1;
4523 while (cur_offset < item_size) {
4527 unsigned long name_ptr;
4528 struct btrfs_dir_item *di;
4530 if (key->type == BTRFS_INODE_REF_KEY) {
4531 struct btrfs_inode_ref *iref;
4533 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4534 parent = key->offset;
4535 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4536 name_ptr = (unsigned long)(iref + 1);
4537 this_len = sizeof(*iref) + this_name_len;
4539 struct btrfs_inode_extref *extref;
4541 extref = (struct btrfs_inode_extref *)(ptr +
4543 parent = btrfs_inode_extref_parent(eb, extref);
4544 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4545 name_ptr = (unsigned long)&extref->name;
4546 this_len = sizeof(*extref) + this_name_len;
4549 if (this_name_len > name_len) {
4552 new_name = krealloc(name, this_name_len, GFP_NOFS);
4557 name_len = this_name_len;
4561 read_extent_buffer(eb, name, name_ptr, this_name_len);
4562 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4563 parent, name, this_name_len, 0);
4564 if (di && !IS_ERR(di)) {
4565 struct btrfs_key di_key;
4567 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4569 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4571 *other_ino = di_key.objectid;
4576 } else if (IS_ERR(di)) {
4580 btrfs_release_path(search_path);
4582 cur_offset += this_len;
4586 btrfs_free_path(search_path);
4591 /* log a single inode in the tree log.
4592 * At least one parent directory for this inode must exist in the tree
4593 * or be logged already.
4595 * Any items from this inode changed by the current transaction are copied
4596 * to the log tree. An extra reference is taken on any extents in this
4597 * file, allowing us to avoid a whole pile of corner cases around logging
4598 * blocks that have been removed from the tree.
4600 * See LOG_INODE_ALL and related defines for a description of what inode_only
4603 * This handles both files and directories.
4605 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4606 struct btrfs_root *root, struct btrfs_inode *inode,
4610 struct btrfs_log_ctx *ctx)
4612 struct btrfs_fs_info *fs_info = root->fs_info;
4613 struct btrfs_path *path;
4614 struct btrfs_path *dst_path;
4615 struct btrfs_key min_key;
4616 struct btrfs_key max_key;
4617 struct btrfs_root *log = root->log_root;
4618 struct extent_buffer *src = NULL;
4619 LIST_HEAD(logged_list);
4620 u64 last_extent = 0;
4624 int ins_start_slot = 0;
4626 bool fast_search = false;
4627 u64 ino = btrfs_ino(inode);
4628 struct extent_map_tree *em_tree = &inode->extent_tree;
4629 u64 logged_isize = 0;
4630 bool need_log_inode_item = true;
4632 path = btrfs_alloc_path();
4635 dst_path = btrfs_alloc_path();
4637 btrfs_free_path(path);
4641 min_key.objectid = ino;
4642 min_key.type = BTRFS_INODE_ITEM_KEY;
4645 max_key.objectid = ino;
4648 /* today the code can only do partial logging of directories */
4649 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4650 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4651 &inode->runtime_flags) &&
4652 inode_only >= LOG_INODE_EXISTS))
4653 max_key.type = BTRFS_XATTR_ITEM_KEY;
4655 max_key.type = (u8)-1;
4656 max_key.offset = (u64)-1;
4659 * Only run delayed items if we are a dir or a new file.
4660 * Otherwise commit the delayed inode only, which is needed in
4661 * order for the log replay code to mark inodes for link count
4662 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4664 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4665 inode->generation > fs_info->last_trans_committed)
4666 ret = btrfs_commit_inode_delayed_items(trans, inode);
4668 ret = btrfs_commit_inode_delayed_inode(inode);
4671 btrfs_free_path(path);
4672 btrfs_free_path(dst_path);
4676 if (inode_only == LOG_OTHER_INODE) {
4677 inode_only = LOG_INODE_EXISTS;
4678 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
4680 mutex_lock(&inode->log_mutex);
4684 * a brute force approach to making sure we get the most uptodate
4685 * copies of everything.
4687 if (S_ISDIR(inode->vfs_inode.i_mode)) {
4688 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4690 if (inode_only == LOG_INODE_EXISTS)
4691 max_key_type = BTRFS_XATTR_ITEM_KEY;
4692 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4694 if (inode_only == LOG_INODE_EXISTS) {
4696 * Make sure the new inode item we write to the log has
4697 * the same isize as the current one (if it exists).
4698 * This is necessary to prevent data loss after log
4699 * replay, and also to prevent doing a wrong expanding
4700 * truncate - for e.g. create file, write 4K into offset
4701 * 0, fsync, write 4K into offset 4096, add hard link,
4702 * fsync some other file (to sync log), power fail - if
4703 * we use the inode's current i_size, after log replay
4704 * we get a 8Kb file, with the last 4Kb extent as a hole
4705 * (zeroes), as if an expanding truncate happened,
4706 * instead of getting a file of 4Kb only.
4708 err = logged_inode_size(log, inode, path, &logged_isize);
4712 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4713 &inode->runtime_flags)) {
4714 if (inode_only == LOG_INODE_EXISTS) {
4715 max_key.type = BTRFS_XATTR_ITEM_KEY;
4716 ret = drop_objectid_items(trans, log, path, ino,
4719 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4720 &inode->runtime_flags);
4721 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4722 &inode->runtime_flags);
4724 ret = btrfs_truncate_inode_items(trans,
4725 log, &inode->vfs_inode, 0, 0);
4730 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4731 &inode->runtime_flags) ||
4732 inode_only == LOG_INODE_EXISTS) {
4733 if (inode_only == LOG_INODE_ALL)
4735 max_key.type = BTRFS_XATTR_ITEM_KEY;
4736 ret = drop_objectid_items(trans, log, path, ino,
4739 if (inode_only == LOG_INODE_ALL)
4752 ret = btrfs_search_forward(root, &min_key,
4753 path, trans->transid);
4761 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4762 if (min_key.objectid != ino)
4764 if (min_key.type > max_key.type)
4767 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4768 need_log_inode_item = false;
4770 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4771 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4772 inode->generation == trans->transid) {
4775 ret = btrfs_check_ref_name_override(path->nodes[0],
4776 path->slots[0], &min_key, inode,
4781 } else if (ret > 0 && ctx &&
4782 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4783 struct btrfs_key inode_key;
4784 struct inode *other_inode;
4790 ins_start_slot = path->slots[0];
4792 ret = copy_items(trans, inode, dst_path, path,
4793 &last_extent, ins_start_slot,
4801 btrfs_release_path(path);
4802 inode_key.objectid = other_ino;
4803 inode_key.type = BTRFS_INODE_ITEM_KEY;
4804 inode_key.offset = 0;
4805 other_inode = btrfs_iget(fs_info->sb,
4809 * If the other inode that had a conflicting dir
4810 * entry was deleted in the current transaction,
4811 * we don't need to do more work nor fallback to
4812 * a transaction commit.
4814 if (IS_ERR(other_inode) &&
4815 PTR_ERR(other_inode) == -ENOENT) {
4817 } else if (IS_ERR(other_inode)) {
4818 err = PTR_ERR(other_inode);
4822 * We are safe logging the other inode without
4823 * acquiring its i_mutex as long as we log with
4824 * the LOG_INODE_EXISTS mode. We're safe against
4825 * concurrent renames of the other inode as well
4826 * because during a rename we pin the log and
4827 * update the log with the new name before we
4830 err = btrfs_log_inode(trans, root,
4831 BTRFS_I(other_inode),
4832 LOG_OTHER_INODE, 0, LLONG_MAX,
4842 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4843 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4846 ret = copy_items(trans, inode, dst_path, path,
4847 &last_extent, ins_start_slot,
4848 ins_nr, inode_only, logged_isize);
4855 btrfs_release_path(path);
4861 src = path->nodes[0];
4862 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4865 } else if (!ins_nr) {
4866 ins_start_slot = path->slots[0];
4871 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4872 ins_start_slot, ins_nr, inode_only,
4880 btrfs_release_path(path);
4884 ins_start_slot = path->slots[0];
4887 nritems = btrfs_header_nritems(path->nodes[0]);
4889 if (path->slots[0] < nritems) {
4890 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4895 ret = copy_items(trans, inode, dst_path, path,
4896 &last_extent, ins_start_slot,
4897 ins_nr, inode_only, logged_isize);
4905 btrfs_release_path(path);
4907 if (min_key.offset < (u64)-1) {
4909 } else if (min_key.type < max_key.type) {
4917 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4918 ins_start_slot, ins_nr, inode_only,
4928 btrfs_release_path(path);
4929 btrfs_release_path(dst_path);
4930 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4933 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4934 btrfs_release_path(path);
4935 btrfs_release_path(dst_path);
4936 err = btrfs_log_trailing_hole(trans, root, inode, path);
4941 btrfs_release_path(path);
4942 btrfs_release_path(dst_path);
4943 if (need_log_inode_item) {
4944 err = log_inode_item(trans, log, dst_path, inode);
4949 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4950 &logged_list, ctx, start, end);
4955 } else if (inode_only == LOG_INODE_ALL) {
4956 struct extent_map *em, *n;
4958 write_lock(&em_tree->lock);
4960 * We can't just remove every em if we're called for a ranged
4961 * fsync - that is, one that doesn't cover the whole possible
4962 * file range (0 to LLONG_MAX). This is because we can have
4963 * em's that fall outside the range we're logging and therefore
4964 * their ordered operations haven't completed yet
4965 * (btrfs_finish_ordered_io() not invoked yet). This means we
4966 * didn't get their respective file extent item in the fs/subvol
4967 * tree yet, and need to let the next fast fsync (one which
4968 * consults the list of modified extent maps) find the em so
4969 * that it logs a matching file extent item and waits for the
4970 * respective ordered operation to complete (if it's still
4973 * Removing every em outside the range we're logging would make
4974 * the next fast fsync not log their matching file extent items,
4975 * therefore making us lose data after a log replay.
4977 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4979 const u64 mod_end = em->mod_start + em->mod_len - 1;
4981 if (em->mod_start >= start && mod_end <= end)
4982 list_del_init(&em->list);
4984 write_unlock(&em_tree->lock);
4987 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
4988 ret = log_directory_changes(trans, root, inode, path, dst_path,
4996 spin_lock(&inode->lock);
4997 inode->logged_trans = trans->transid;
4998 inode->last_log_commit = inode->last_sub_trans;
4999 spin_unlock(&inode->lock);
5002 btrfs_put_logged_extents(&logged_list);
5004 btrfs_submit_logged_extents(&logged_list, log);
5005 mutex_unlock(&inode->log_mutex);
5007 btrfs_free_path(path);
5008 btrfs_free_path(dst_path);
5013 * Check if we must fallback to a transaction commit when logging an inode.
5014 * This must be called after logging the inode and is used only in the context
5015 * when fsyncing an inode requires the need to log some other inode - in which
5016 * case we can't lock the i_mutex of each other inode we need to log as that
5017 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5018 * log inodes up or down in the hierarchy) or rename operations for example. So
5019 * we take the log_mutex of the inode after we have logged it and then check for
5020 * its last_unlink_trans value - this is safe because any task setting
5021 * last_unlink_trans must take the log_mutex and it must do this before it does
5022 * the actual unlink operation, so if we do this check before a concurrent task
5023 * sets last_unlink_trans it means we've logged a consistent version/state of
5024 * all the inode items, otherwise we are not sure and must do a transaction
5025 * commit (the concurrent task might have only updated last_unlink_trans before
5026 * we logged the inode or it might have also done the unlink).
5028 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5029 struct btrfs_inode *inode)
5031 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5034 mutex_lock(&inode->log_mutex);
5035 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5037 * Make sure any commits to the log are forced to be full
5040 btrfs_set_log_full_commit(fs_info, trans);
5043 mutex_unlock(&inode->log_mutex);
5049 * follow the dentry parent pointers up the chain and see if any
5050 * of the directories in it require a full commit before they can
5051 * be logged. Returns zero if nothing special needs to be done or 1 if
5052 * a full commit is required.
5054 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5055 struct btrfs_inode *inode,
5056 struct dentry *parent,
5057 struct super_block *sb,
5061 struct dentry *old_parent = NULL;
5062 struct btrfs_inode *orig_inode = inode;
5065 * for regular files, if its inode is already on disk, we don't
5066 * have to worry about the parents at all. This is because
5067 * we can use the last_unlink_trans field to record renames
5068 * and other fun in this file.
5070 if (S_ISREG(inode->vfs_inode.i_mode) &&
5071 inode->generation <= last_committed &&
5072 inode->last_unlink_trans <= last_committed)
5075 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5076 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5078 inode = BTRFS_I(d_inode(parent));
5083 * If we are logging a directory then we start with our inode,
5084 * not our parent's inode, so we need to skip setting the
5085 * logged_trans so that further down in the log code we don't
5086 * think this inode has already been logged.
5088 if (inode != orig_inode)
5089 inode->logged_trans = trans->transid;
5092 if (btrfs_must_commit_transaction(trans, inode)) {
5097 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5100 if (IS_ROOT(parent)) {
5101 inode = BTRFS_I(d_inode(parent));
5102 if (btrfs_must_commit_transaction(trans, inode))
5107 parent = dget_parent(parent);
5109 old_parent = parent;
5110 inode = BTRFS_I(d_inode(parent));
5118 struct btrfs_dir_list {
5120 struct list_head list;
5124 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5125 * details about the why it is needed.
5126 * This is a recursive operation - if an existing dentry corresponds to a
5127 * directory, that directory's new entries are logged too (same behaviour as
5128 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5129 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5130 * complains about the following circular lock dependency / possible deadlock:
5134 * lock(&type->i_mutex_dir_key#3/2);
5135 * lock(sb_internal#2);
5136 * lock(&type->i_mutex_dir_key#3/2);
5137 * lock(&sb->s_type->i_mutex_key#14);
5139 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5140 * sb_start_intwrite() in btrfs_start_transaction().
5141 * Not locking i_mutex of the inodes is still safe because:
5143 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5144 * that while logging the inode new references (names) are added or removed
5145 * from the inode, leaving the logged inode item with a link count that does
5146 * not match the number of logged inode reference items. This is fine because
5147 * at log replay time we compute the real number of links and correct the
5148 * link count in the inode item (see replay_one_buffer() and
5149 * link_to_fixup_dir());
5151 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5152 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5153 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5154 * has a size that doesn't match the sum of the lengths of all the logged
5155 * names. This does not result in a problem because if a dir_item key is
5156 * logged but its matching dir_index key is not logged, at log replay time we
5157 * don't use it to replay the respective name (see replay_one_name()). On the
5158 * other hand if only the dir_index key ends up being logged, the respective
5159 * name is added to the fs/subvol tree with both the dir_item and dir_index
5160 * keys created (see replay_one_name()).
5161 * The directory's inode item with a wrong i_size is not a problem as well,
5162 * since we don't use it at log replay time to set the i_size in the inode
5163 * item of the fs/subvol tree (see overwrite_item()).
5165 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5166 struct btrfs_root *root,
5167 struct btrfs_inode *start_inode,
5168 struct btrfs_log_ctx *ctx)
5170 struct btrfs_fs_info *fs_info = root->fs_info;
5171 struct btrfs_root *log = root->log_root;
5172 struct btrfs_path *path;
5173 LIST_HEAD(dir_list);
5174 struct btrfs_dir_list *dir_elem;
5177 path = btrfs_alloc_path();
5181 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5183 btrfs_free_path(path);
5186 dir_elem->ino = btrfs_ino(start_inode);
5187 list_add_tail(&dir_elem->list, &dir_list);
5189 while (!list_empty(&dir_list)) {
5190 struct extent_buffer *leaf;
5191 struct btrfs_key min_key;
5195 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5198 goto next_dir_inode;
5200 min_key.objectid = dir_elem->ino;
5201 min_key.type = BTRFS_DIR_ITEM_KEY;
5204 btrfs_release_path(path);
5205 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5207 goto next_dir_inode;
5208 } else if (ret > 0) {
5210 goto next_dir_inode;
5214 leaf = path->nodes[0];
5215 nritems = btrfs_header_nritems(leaf);
5216 for (i = path->slots[0]; i < nritems; i++) {
5217 struct btrfs_dir_item *di;
5218 struct btrfs_key di_key;
5219 struct inode *di_inode;
5220 struct btrfs_dir_list *new_dir_elem;
5221 int log_mode = LOG_INODE_EXISTS;
5224 btrfs_item_key_to_cpu(leaf, &min_key, i);
5225 if (min_key.objectid != dir_elem->ino ||
5226 min_key.type != BTRFS_DIR_ITEM_KEY)
5227 goto next_dir_inode;
5229 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5230 type = btrfs_dir_type(leaf, di);
5231 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5232 type != BTRFS_FT_DIR)
5234 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5235 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5238 btrfs_release_path(path);
5239 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5240 if (IS_ERR(di_inode)) {
5241 ret = PTR_ERR(di_inode);
5242 goto next_dir_inode;
5245 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5250 ctx->log_new_dentries = false;
5251 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5252 log_mode = LOG_INODE_ALL;
5253 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5254 log_mode, 0, LLONG_MAX, ctx);
5256 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5260 goto next_dir_inode;
5261 if (ctx->log_new_dentries) {
5262 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5264 if (!new_dir_elem) {
5266 goto next_dir_inode;
5268 new_dir_elem->ino = di_key.objectid;
5269 list_add_tail(&new_dir_elem->list, &dir_list);
5274 ret = btrfs_next_leaf(log, path);
5276 goto next_dir_inode;
5277 } else if (ret > 0) {
5279 goto next_dir_inode;
5283 if (min_key.offset < (u64)-1) {
5288 list_del(&dir_elem->list);
5292 btrfs_free_path(path);
5296 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5297 struct btrfs_inode *inode,
5298 struct btrfs_log_ctx *ctx)
5300 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5302 struct btrfs_path *path;
5303 struct btrfs_key key;
5304 struct btrfs_root *root = inode->root;
5305 const u64 ino = btrfs_ino(inode);
5307 path = btrfs_alloc_path();
5310 path->skip_locking = 1;
5311 path->search_commit_root = 1;
5314 key.type = BTRFS_INODE_REF_KEY;
5316 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5321 struct extent_buffer *leaf = path->nodes[0];
5322 int slot = path->slots[0];
5327 if (slot >= btrfs_header_nritems(leaf)) {
5328 ret = btrfs_next_leaf(root, path);
5336 btrfs_item_key_to_cpu(leaf, &key, slot);
5337 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5338 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5341 item_size = btrfs_item_size_nr(leaf, slot);
5342 ptr = btrfs_item_ptr_offset(leaf, slot);
5343 while (cur_offset < item_size) {
5344 struct btrfs_key inode_key;
5345 struct inode *dir_inode;
5347 inode_key.type = BTRFS_INODE_ITEM_KEY;
5348 inode_key.offset = 0;
5350 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5351 struct btrfs_inode_extref *extref;
5353 extref = (struct btrfs_inode_extref *)
5355 inode_key.objectid = btrfs_inode_extref_parent(
5357 cur_offset += sizeof(*extref);
5358 cur_offset += btrfs_inode_extref_name_len(leaf,
5361 inode_key.objectid = key.offset;
5362 cur_offset = item_size;
5365 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5367 /* If parent inode was deleted, skip it. */
5368 if (IS_ERR(dir_inode))
5372 ctx->log_new_dentries = false;
5373 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5374 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5376 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5378 if (!ret && ctx && ctx->log_new_dentries)
5379 ret = log_new_dir_dentries(trans, root,
5380 BTRFS_I(dir_inode), ctx);
5389 btrfs_free_path(path);
5394 * helper function around btrfs_log_inode to make sure newly created
5395 * parent directories also end up in the log. A minimal inode and backref
5396 * only logging is done of any parent directories that are older than
5397 * the last committed transaction
5399 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5400 struct btrfs_root *root,
5401 struct btrfs_inode *inode,
5402 struct dentry *parent,
5406 struct btrfs_log_ctx *ctx)
5408 struct btrfs_fs_info *fs_info = root->fs_info;
5409 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5410 struct super_block *sb;
5411 struct dentry *old_parent = NULL;
5413 u64 last_committed = fs_info->last_trans_committed;
5414 bool log_dentries = false;
5415 struct btrfs_inode *orig_inode = inode;
5417 sb = inode->vfs_inode.i_sb;
5419 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5425 * The prev transaction commit doesn't complete, we need do
5426 * full commit by ourselves.
5428 if (fs_info->last_trans_log_full_commit >
5429 fs_info->last_trans_committed) {
5434 if (root != inode->root || btrfs_root_refs(&root->root_item) == 0) {
5439 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5444 if (btrfs_inode_in_log(inode, trans->transid)) {
5445 ret = BTRFS_NO_LOG_SYNC;
5449 ret = start_log_trans(trans, root, ctx);
5453 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5458 * for regular files, if its inode is already on disk, we don't
5459 * have to worry about the parents at all. This is because
5460 * we can use the last_unlink_trans field to record renames
5461 * and other fun in this file.
5463 if (S_ISREG(inode->vfs_inode.i_mode) &&
5464 inode->generation <= last_committed &&
5465 inode->last_unlink_trans <= last_committed) {
5470 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5471 log_dentries = true;
5474 * On unlink we must make sure all our current and old parent directory
5475 * inodes are fully logged. This is to prevent leaving dangling
5476 * directory index entries in directories that were our parents but are
5477 * not anymore. Not doing this results in old parent directory being
5478 * impossible to delete after log replay (rmdir will always fail with
5479 * error -ENOTEMPTY).
5485 * ln testdir/foo testdir/bar
5487 * unlink testdir/bar
5488 * xfs_io -c fsync testdir/foo
5490 * mount fs, triggers log replay
5492 * If we don't log the parent directory (testdir), after log replay the
5493 * directory still has an entry pointing to the file inode using the bar
5494 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5495 * the file inode has a link count of 1.
5501 * ln foo testdir/foo2
5502 * ln foo testdir/foo3
5504 * unlink testdir/foo3
5505 * xfs_io -c fsync foo
5507 * mount fs, triggers log replay
5509 * Similar as the first example, after log replay the parent directory
5510 * testdir still has an entry pointing to the inode file with name foo3
5511 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5512 * and has a link count of 2.
5514 if (inode->last_unlink_trans > last_committed) {
5515 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5521 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5524 inode = BTRFS_I(d_inode(parent));
5525 if (root != inode->root)
5528 if (inode->generation > last_committed) {
5529 ret = btrfs_log_inode(trans, root, inode,
5530 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5534 if (IS_ROOT(parent))
5537 parent = dget_parent(parent);
5539 old_parent = parent;
5542 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5548 btrfs_set_log_full_commit(fs_info, trans);
5553 btrfs_remove_log_ctx(root, ctx);
5554 btrfs_end_log_trans(root);
5560 * it is not safe to log dentry if the chunk root has added new
5561 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5562 * If this returns 1, you must commit the transaction to safely get your
5565 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5566 struct btrfs_root *root, struct dentry *dentry,
5569 struct btrfs_log_ctx *ctx)
5571 struct dentry *parent = dget_parent(dentry);
5574 ret = btrfs_log_inode_parent(trans, root, BTRFS_I(d_inode(dentry)),
5575 parent, start, end, 0, ctx);
5582 * should be called during mount to recover any replay any log trees
5585 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5588 struct btrfs_path *path;
5589 struct btrfs_trans_handle *trans;
5590 struct btrfs_key key;
5591 struct btrfs_key found_key;
5592 struct btrfs_key tmp_key;
5593 struct btrfs_root *log;
5594 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5595 struct walk_control wc = {
5596 .process_func = process_one_buffer,
5600 path = btrfs_alloc_path();
5604 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5606 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5607 if (IS_ERR(trans)) {
5608 ret = PTR_ERR(trans);
5615 ret = walk_log_tree(trans, log_root_tree, &wc);
5617 btrfs_handle_fs_error(fs_info, ret,
5618 "Failed to pin buffers while recovering log root tree.");
5623 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5624 key.offset = (u64)-1;
5625 key.type = BTRFS_ROOT_ITEM_KEY;
5628 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5631 btrfs_handle_fs_error(fs_info, ret,
5632 "Couldn't find tree log root.");
5636 if (path->slots[0] == 0)
5640 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5642 btrfs_release_path(path);
5643 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5646 log = btrfs_read_fs_root(log_root_tree, &found_key);
5649 btrfs_handle_fs_error(fs_info, ret,
5650 "Couldn't read tree log root.");
5654 tmp_key.objectid = found_key.offset;
5655 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5656 tmp_key.offset = (u64)-1;
5658 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5659 if (IS_ERR(wc.replay_dest)) {
5660 ret = PTR_ERR(wc.replay_dest);
5661 free_extent_buffer(log->node);
5662 free_extent_buffer(log->commit_root);
5664 btrfs_handle_fs_error(fs_info, ret,
5665 "Couldn't read target root for tree log recovery.");
5669 wc.replay_dest->log_root = log;
5670 btrfs_record_root_in_trans(trans, wc.replay_dest);
5671 ret = walk_log_tree(trans, log, &wc);
5673 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5674 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5678 key.offset = found_key.offset - 1;
5679 wc.replay_dest->log_root = NULL;
5680 free_extent_buffer(log->node);
5681 free_extent_buffer(log->commit_root);
5687 if (found_key.offset == 0)
5690 btrfs_release_path(path);
5692 /* step one is to pin it all, step two is to replay just inodes */
5695 wc.process_func = replay_one_buffer;
5696 wc.stage = LOG_WALK_REPLAY_INODES;
5699 /* step three is to replay everything */
5700 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5705 btrfs_free_path(path);
5707 /* step 4: commit the transaction, which also unpins the blocks */
5708 ret = btrfs_commit_transaction(trans);
5712 free_extent_buffer(log_root_tree->node);
5713 log_root_tree->log_root = NULL;
5714 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5715 kfree(log_root_tree);
5720 btrfs_end_transaction(wc.trans);
5721 btrfs_free_path(path);
5726 * there are some corner cases where we want to force a full
5727 * commit instead of allowing a directory to be logged.
5729 * They revolve around files there were unlinked from the directory, and
5730 * this function updates the parent directory so that a full commit is
5731 * properly done if it is fsync'd later after the unlinks are done.
5733 * Must be called before the unlink operations (updates to the subvolume tree,
5734 * inodes, etc) are done.
5736 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5737 struct btrfs_inode *dir, struct btrfs_inode *inode,
5741 * when we're logging a file, if it hasn't been renamed
5742 * or unlinked, and its inode is fully committed on disk,
5743 * we don't have to worry about walking up the directory chain
5744 * to log its parents.
5746 * So, we use the last_unlink_trans field to put this transid
5747 * into the file. When the file is logged we check it and
5748 * don't log the parents if the file is fully on disk.
5750 mutex_lock(&inode->log_mutex);
5751 inode->last_unlink_trans = trans->transid;
5752 mutex_unlock(&inode->log_mutex);
5755 * if this directory was already logged any new
5756 * names for this file/dir will get recorded
5759 if (dir->logged_trans == trans->transid)
5763 * if the inode we're about to unlink was logged,
5764 * the log will be properly updated for any new names
5766 if (inode->logged_trans == trans->transid)
5770 * when renaming files across directories, if the directory
5771 * there we're unlinking from gets fsync'd later on, there's
5772 * no way to find the destination directory later and fsync it
5773 * properly. So, we have to be conservative and force commits
5774 * so the new name gets discovered.
5779 /* we can safely do the unlink without any special recording */
5783 mutex_lock(&dir->log_mutex);
5784 dir->last_unlink_trans = trans->transid;
5785 mutex_unlock(&dir->log_mutex);
5789 * Make sure that if someone attempts to fsync the parent directory of a deleted
5790 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5791 * that after replaying the log tree of the parent directory's root we will not
5792 * see the snapshot anymore and at log replay time we will not see any log tree
5793 * corresponding to the deleted snapshot's root, which could lead to replaying
5794 * it after replaying the log tree of the parent directory (which would replay
5795 * the snapshot delete operation).
5797 * Must be called before the actual snapshot destroy operation (updates to the
5798 * parent root and tree of tree roots trees, etc) are done.
5800 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5801 struct btrfs_inode *dir)
5803 mutex_lock(&dir->log_mutex);
5804 dir->last_unlink_trans = trans->transid;
5805 mutex_unlock(&dir->log_mutex);
5809 * Call this after adding a new name for a file and it will properly
5810 * update the log to reflect the new name.
5812 * It will return zero if all goes well, and it will return 1 if a
5813 * full transaction commit is required.
5815 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5816 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
5817 struct dentry *parent)
5819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5820 struct btrfs_root *root = inode->root;
5823 * this will force the logging code to walk the dentry chain
5826 if (S_ISREG(inode->vfs_inode.i_mode))
5827 inode->last_unlink_trans = trans->transid;
5830 * if this inode hasn't been logged and directory we're renaming it
5831 * from hasn't been logged, we don't need to log it
5833 if (inode->logged_trans <= fs_info->last_trans_committed &&
5834 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
5837 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5838 LLONG_MAX, 1, NULL);