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, int slot,
1179 unsigned long ref_ptr, u32 *namelen, char **name,
1180 u64 *index, 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 if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)&extref->name,
1191 *name = kmalloc(*namelen, GFP_NOFS);
1195 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1198 *index = btrfs_inode_extref_index(eb, extref);
1199 if (parent_objectid)
1200 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1205 static int ref_get_fields(struct extent_buffer *eb, int slot,
1206 unsigned long ref_ptr, u32 *namelen, char **name,
1209 struct btrfs_inode_ref *ref;
1211 ref = (struct btrfs_inode_ref *)ref_ptr;
1213 *namelen = btrfs_inode_ref_name_len(eb, ref);
1214 if (!btrfs_is_name_len_valid(eb, slot, (unsigned long)(ref + 1),
1218 *name = kmalloc(*namelen, GFP_NOFS);
1222 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1224 *index = btrfs_inode_ref_index(eb, ref);
1230 * replay one inode back reference item found in the log tree.
1231 * eb, slot and key refer to the buffer and key found in the log tree.
1232 * root is the destination we are replaying into, and path is for temp
1233 * use by this function. (it should be released on return).
1235 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1236 struct btrfs_root *root,
1237 struct btrfs_root *log,
1238 struct btrfs_path *path,
1239 struct extent_buffer *eb, int slot,
1240 struct btrfs_key *key)
1242 struct inode *dir = NULL;
1243 struct inode *inode = NULL;
1244 unsigned long ref_ptr;
1245 unsigned long ref_end;
1249 int search_done = 0;
1250 int log_ref_ver = 0;
1251 u64 parent_objectid;
1254 int ref_struct_size;
1256 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1257 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1259 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1260 struct btrfs_inode_extref *r;
1262 ref_struct_size = sizeof(struct btrfs_inode_extref);
1264 r = (struct btrfs_inode_extref *)ref_ptr;
1265 parent_objectid = btrfs_inode_extref_parent(eb, r);
1267 ref_struct_size = sizeof(struct btrfs_inode_ref);
1268 parent_objectid = key->offset;
1270 inode_objectid = key->objectid;
1273 * it is possible that we didn't log all the parent directories
1274 * for a given inode. If we don't find the dir, just don't
1275 * copy the back ref in. The link count fixup code will take
1278 dir = read_one_inode(root, parent_objectid);
1284 inode = read_one_inode(root, inode_objectid);
1290 while (ref_ptr < ref_end) {
1292 ret = extref_get_fields(eb, slot, ref_ptr, &namelen,
1293 &name, &ref_index, &parent_objectid);
1295 * parent object can change from one array
1299 dir = read_one_inode(root, parent_objectid);
1305 ret = ref_get_fields(eb, slot, ref_ptr, &namelen,
1311 /* if we already have a perfect match, we're done */
1312 if (!inode_in_dir(root, path, btrfs_ino(BTRFS_I(dir)),
1313 btrfs_ino(BTRFS_I(inode)), ref_index,
1316 * look for a conflicting back reference in the
1317 * metadata. if we find one we have to unlink that name
1318 * of the file before we add our new link. Later on, we
1319 * overwrite any existing back reference, and we don't
1320 * want to create dangling pointers in the directory.
1324 ret = __add_inode_ref(trans, root, path, log,
1329 ref_index, name, namelen,
1338 /* insert our name */
1339 ret = btrfs_add_link(trans, BTRFS_I(dir),
1341 name, namelen, 0, ref_index);
1345 btrfs_update_inode(trans, root, inode);
1348 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1357 /* finally write the back reference in the inode */
1358 ret = overwrite_item(trans, root, path, eb, slot, key);
1360 btrfs_release_path(path);
1367 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1368 struct btrfs_root *root, u64 ino)
1372 ret = btrfs_insert_orphan_item(trans, root, ino);
1379 static int count_inode_extrefs(struct btrfs_root *root,
1380 struct btrfs_inode *inode, struct btrfs_path *path)
1384 unsigned int nlink = 0;
1387 u64 inode_objectid = btrfs_ino(inode);
1390 struct btrfs_inode_extref *extref;
1391 struct extent_buffer *leaf;
1394 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1399 leaf = path->nodes[0];
1400 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1401 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1404 while (cur_offset < item_size) {
1405 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1406 name_len = btrfs_inode_extref_name_len(leaf, extref);
1410 cur_offset += name_len + sizeof(*extref);
1414 btrfs_release_path(path);
1416 btrfs_release_path(path);
1418 if (ret < 0 && ret != -ENOENT)
1423 static int count_inode_refs(struct btrfs_root *root,
1424 struct btrfs_inode *inode, struct btrfs_path *path)
1427 struct btrfs_key key;
1428 unsigned int nlink = 0;
1430 unsigned long ptr_end;
1432 u64 ino = btrfs_ino(inode);
1435 key.type = BTRFS_INODE_REF_KEY;
1436 key.offset = (u64)-1;
1439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1443 if (path->slots[0] == 0)
1448 btrfs_item_key_to_cpu(path->nodes[0], &key,
1450 if (key.objectid != ino ||
1451 key.type != BTRFS_INODE_REF_KEY)
1453 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1454 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1456 while (ptr < ptr_end) {
1457 struct btrfs_inode_ref *ref;
1459 ref = (struct btrfs_inode_ref *)ptr;
1460 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1462 ptr = (unsigned long)(ref + 1) + name_len;
1466 if (key.offset == 0)
1468 if (path->slots[0] > 0) {
1473 btrfs_release_path(path);
1475 btrfs_release_path(path);
1481 * There are a few corners where the link count of the file can't
1482 * be properly maintained during replay. So, instead of adding
1483 * lots of complexity to the log code, we just scan the backrefs
1484 * for any file that has been through replay.
1486 * The scan will update the link count on the inode to reflect the
1487 * number of back refs found. If it goes down to zero, the iput
1488 * will free the inode.
1490 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1491 struct btrfs_root *root,
1492 struct inode *inode)
1494 struct btrfs_path *path;
1497 u64 ino = btrfs_ino(BTRFS_I(inode));
1499 path = btrfs_alloc_path();
1503 ret = count_inode_refs(root, BTRFS_I(inode), path);
1509 ret = count_inode_extrefs(root, BTRFS_I(inode), path);
1517 if (nlink != inode->i_nlink) {
1518 set_nlink(inode, nlink);
1519 btrfs_update_inode(trans, root, inode);
1521 BTRFS_I(inode)->index_cnt = (u64)-1;
1523 if (inode->i_nlink == 0) {
1524 if (S_ISDIR(inode->i_mode)) {
1525 ret = replay_dir_deletes(trans, root, NULL, path,
1530 ret = insert_orphan_item(trans, root, ino);
1534 btrfs_free_path(path);
1538 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root,
1540 struct btrfs_path *path)
1543 struct btrfs_key key;
1544 struct inode *inode;
1546 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1547 key.type = BTRFS_ORPHAN_ITEM_KEY;
1548 key.offset = (u64)-1;
1550 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1555 if (path->slots[0] == 0)
1560 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1561 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1562 key.type != BTRFS_ORPHAN_ITEM_KEY)
1565 ret = btrfs_del_item(trans, root, path);
1569 btrfs_release_path(path);
1570 inode = read_one_inode(root, key.offset);
1574 ret = fixup_inode_link_count(trans, root, inode);
1580 * fixup on a directory may create new entries,
1581 * make sure we always look for the highset possible
1584 key.offset = (u64)-1;
1588 btrfs_release_path(path);
1594 * record a given inode in the fixup dir so we can check its link
1595 * count when replay is done. The link count is incremented here
1596 * so the inode won't go away until we check it
1598 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1599 struct btrfs_root *root,
1600 struct btrfs_path *path,
1603 struct btrfs_key key;
1605 struct inode *inode;
1607 inode = read_one_inode(root, objectid);
1611 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1612 key.type = BTRFS_ORPHAN_ITEM_KEY;
1613 key.offset = objectid;
1615 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1617 btrfs_release_path(path);
1619 if (!inode->i_nlink)
1620 set_nlink(inode, 1);
1623 ret = btrfs_update_inode(trans, root, inode);
1624 } else if (ret == -EEXIST) {
1627 BUG(); /* Logic Error */
1635 * when replaying the log for a directory, we only insert names
1636 * for inodes that actually exist. This means an fsync on a directory
1637 * does not implicitly fsync all the new files in it
1639 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1640 struct btrfs_root *root,
1641 u64 dirid, u64 index,
1642 char *name, int name_len,
1643 struct btrfs_key *location)
1645 struct inode *inode;
1649 inode = read_one_inode(root, location->objectid);
1653 dir = read_one_inode(root, dirid);
1659 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
1660 name_len, 1, index);
1662 /* FIXME, put inode into FIXUP list */
1670 * Return true if an inode reference exists in the log for the given name,
1671 * inode and parent inode.
1673 static bool name_in_log_ref(struct btrfs_root *log_root,
1674 const char *name, const int name_len,
1675 const u64 dirid, const u64 ino)
1677 struct btrfs_key search_key;
1679 search_key.objectid = ino;
1680 search_key.type = BTRFS_INODE_REF_KEY;
1681 search_key.offset = dirid;
1682 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1685 search_key.type = BTRFS_INODE_EXTREF_KEY;
1686 search_key.offset = btrfs_extref_hash(dirid, name, name_len);
1687 if (backref_in_log(log_root, &search_key, dirid, name, name_len))
1694 * take a single entry in a log directory item and replay it into
1697 * if a conflicting item exists in the subdirectory already,
1698 * the inode it points to is unlinked and put into the link count
1701 * If a name from the log points to a file or directory that does
1702 * not exist in the FS, it is skipped. fsyncs on directories
1703 * do not force down inodes inside that directory, just changes to the
1704 * names or unlinks in a directory.
1706 * Returns < 0 on error, 0 if the name wasn't replayed (dentry points to a
1707 * non-existing inode) and 1 if the name was replayed.
1709 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1710 struct btrfs_root *root,
1711 struct btrfs_path *path,
1712 struct extent_buffer *eb,
1713 struct btrfs_dir_item *di,
1714 struct btrfs_key *key)
1718 struct btrfs_dir_item *dst_di;
1719 struct btrfs_key found_key;
1720 struct btrfs_key log_key;
1725 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1726 bool name_added = false;
1728 dir = read_one_inode(root, key->objectid);
1732 name_len = btrfs_dir_name_len(eb, di);
1733 name = kmalloc(name_len, GFP_NOFS);
1739 log_type = btrfs_dir_type(eb, di);
1740 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1743 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1744 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1749 btrfs_release_path(path);
1751 if (key->type == BTRFS_DIR_ITEM_KEY) {
1752 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1754 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1755 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1764 if (IS_ERR_OR_NULL(dst_di)) {
1765 /* we need a sequence number to insert, so we only
1766 * do inserts for the BTRFS_DIR_INDEX_KEY types
1768 if (key->type != BTRFS_DIR_INDEX_KEY)
1773 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1774 /* the existing item matches the logged item */
1775 if (found_key.objectid == log_key.objectid &&
1776 found_key.type == log_key.type &&
1777 found_key.offset == log_key.offset &&
1778 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1779 update_size = false;
1784 * don't drop the conflicting directory entry if the inode
1785 * for the new entry doesn't exist
1790 ret = drop_one_dir_item(trans, root, path, BTRFS_I(dir), dst_di);
1794 if (key->type == BTRFS_DIR_INDEX_KEY)
1797 btrfs_release_path(path);
1798 if (!ret && update_size) {
1799 btrfs_i_size_write(BTRFS_I(dir), dir->i_size + name_len * 2);
1800 ret = btrfs_update_inode(trans, root, dir);
1804 if (!ret && name_added)
1809 if (name_in_log_ref(root->log_root, name, name_len,
1810 key->objectid, log_key.objectid)) {
1811 /* The dentry will be added later. */
1813 update_size = false;
1816 btrfs_release_path(path);
1817 ret = insert_one_name(trans, root, key->objectid, key->offset,
1818 name, name_len, &log_key);
1819 if (ret && ret != -ENOENT && ret != -EEXIST)
1823 update_size = false;
1829 * find all the names in a directory item and reconcile them into
1830 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1831 * one name in a directory item, but the same code gets used for
1832 * both directory index types
1834 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1835 struct btrfs_root *root,
1836 struct btrfs_path *path,
1837 struct extent_buffer *eb, int slot,
1838 struct btrfs_key *key)
1840 struct btrfs_fs_info *fs_info = root->fs_info;
1842 u32 item_size = btrfs_item_size_nr(eb, slot);
1843 struct btrfs_dir_item *di;
1846 unsigned long ptr_end;
1847 struct btrfs_path *fixup_path = NULL;
1849 ptr = btrfs_item_ptr_offset(eb, slot);
1850 ptr_end = ptr + item_size;
1851 while (ptr < ptr_end) {
1852 di = (struct btrfs_dir_item *)ptr;
1853 if (verify_dir_item(fs_info, eb, slot, di))
1855 name_len = btrfs_dir_name_len(eb, di);
1856 ret = replay_one_name(trans, root, path, eb, di, key);
1859 ptr = (unsigned long)(di + 1);
1863 * If this entry refers to a non-directory (directories can not
1864 * have a link count > 1) and it was added in the transaction
1865 * that was not committed, make sure we fixup the link count of
1866 * the inode it the entry points to. Otherwise something like
1867 * the following would result in a directory pointing to an
1868 * inode with a wrong link that does not account for this dir
1876 * ln testdir/bar testdir/bar_link
1877 * ln testdir/foo testdir/foo_link
1878 * xfs_io -c "fsync" testdir/bar
1882 * mount fs, log replay happens
1884 * File foo would remain with a link count of 1 when it has two
1885 * entries pointing to it in the directory testdir. This would
1886 * make it impossible to ever delete the parent directory has
1887 * it would result in stale dentries that can never be deleted.
1889 if (ret == 1 && btrfs_dir_type(eb, di) != BTRFS_FT_DIR) {
1890 struct btrfs_key di_key;
1893 fixup_path = btrfs_alloc_path();
1900 btrfs_dir_item_key_to_cpu(eb, di, &di_key);
1901 ret = link_to_fixup_dir(trans, root, fixup_path,
1908 btrfs_free_path(fixup_path);
1913 * directory replay has two parts. There are the standard directory
1914 * items in the log copied from the subvolume, and range items
1915 * created in the log while the subvolume was logged.
1917 * The range items tell us which parts of the key space the log
1918 * is authoritative for. During replay, if a key in the subvolume
1919 * directory is in a logged range item, but not actually in the log
1920 * that means it was deleted from the directory before the fsync
1921 * and should be removed.
1923 static noinline int find_dir_range(struct btrfs_root *root,
1924 struct btrfs_path *path,
1925 u64 dirid, int key_type,
1926 u64 *start_ret, u64 *end_ret)
1928 struct btrfs_key key;
1930 struct btrfs_dir_log_item *item;
1934 if (*start_ret == (u64)-1)
1937 key.objectid = dirid;
1938 key.type = key_type;
1939 key.offset = *start_ret;
1941 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1945 if (path->slots[0] == 0)
1950 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1952 if (key.type != key_type || key.objectid != dirid) {
1956 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1957 struct btrfs_dir_log_item);
1958 found_end = btrfs_dir_log_end(path->nodes[0], item);
1960 if (*start_ret >= key.offset && *start_ret <= found_end) {
1962 *start_ret = key.offset;
1963 *end_ret = found_end;
1968 /* check the next slot in the tree to see if it is a valid item */
1969 nritems = btrfs_header_nritems(path->nodes[0]);
1971 if (path->slots[0] >= nritems) {
1972 ret = btrfs_next_leaf(root, path);
1977 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1979 if (key.type != key_type || key.objectid != dirid) {
1983 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1984 struct btrfs_dir_log_item);
1985 found_end = btrfs_dir_log_end(path->nodes[0], item);
1986 *start_ret = key.offset;
1987 *end_ret = found_end;
1990 btrfs_release_path(path);
1995 * this looks for a given directory item in the log. If the directory
1996 * item is not in the log, the item is removed and the inode it points
1999 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
2000 struct btrfs_root *root,
2001 struct btrfs_root *log,
2002 struct btrfs_path *path,
2003 struct btrfs_path *log_path,
2005 struct btrfs_key *dir_key)
2007 struct btrfs_fs_info *fs_info = root->fs_info;
2009 struct extent_buffer *eb;
2012 struct btrfs_dir_item *di;
2013 struct btrfs_dir_item *log_di;
2016 unsigned long ptr_end;
2018 struct inode *inode;
2019 struct btrfs_key location;
2022 eb = path->nodes[0];
2023 slot = path->slots[0];
2024 item_size = btrfs_item_size_nr(eb, slot);
2025 ptr = btrfs_item_ptr_offset(eb, slot);
2026 ptr_end = ptr + item_size;
2027 while (ptr < ptr_end) {
2028 di = (struct btrfs_dir_item *)ptr;
2029 if (verify_dir_item(fs_info, eb, slot, di)) {
2034 name_len = btrfs_dir_name_len(eb, di);
2035 name = kmalloc(name_len, GFP_NOFS);
2040 read_extent_buffer(eb, name, (unsigned long)(di + 1),
2043 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
2044 log_di = btrfs_lookup_dir_item(trans, log, log_path,
2047 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
2048 log_di = btrfs_lookup_dir_index_item(trans, log,
2054 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
2055 btrfs_dir_item_key_to_cpu(eb, di, &location);
2056 btrfs_release_path(path);
2057 btrfs_release_path(log_path);
2058 inode = read_one_inode(root, location.objectid);
2064 ret = link_to_fixup_dir(trans, root,
2065 path, location.objectid);
2073 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
2074 BTRFS_I(inode), name, name_len);
2076 ret = btrfs_run_delayed_items(trans, fs_info);
2082 /* there might still be more names under this key
2083 * check and repeat if required
2085 ret = btrfs_search_slot(NULL, root, dir_key, path,
2091 } else if (IS_ERR(log_di)) {
2093 return PTR_ERR(log_di);
2095 btrfs_release_path(log_path);
2098 ptr = (unsigned long)(di + 1);
2103 btrfs_release_path(path);
2104 btrfs_release_path(log_path);
2108 static int replay_xattr_deletes(struct btrfs_trans_handle *trans,
2109 struct btrfs_root *root,
2110 struct btrfs_root *log,
2111 struct btrfs_path *path,
2114 struct btrfs_fs_info *fs_info = root->fs_info;
2115 struct btrfs_key search_key;
2116 struct btrfs_path *log_path;
2121 log_path = btrfs_alloc_path();
2125 search_key.objectid = ino;
2126 search_key.type = BTRFS_XATTR_ITEM_KEY;
2127 search_key.offset = 0;
2129 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
2133 nritems = btrfs_header_nritems(path->nodes[0]);
2134 for (i = path->slots[0]; i < nritems; i++) {
2135 struct btrfs_key key;
2136 struct btrfs_dir_item *di;
2137 struct btrfs_dir_item *log_di;
2141 btrfs_item_key_to_cpu(path->nodes[0], &key, i);
2142 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY) {
2147 di = btrfs_item_ptr(path->nodes[0], i, struct btrfs_dir_item);
2148 total_size = btrfs_item_size_nr(path->nodes[0], i);
2150 while (cur < total_size) {
2151 u16 name_len = btrfs_dir_name_len(path->nodes[0], di);
2152 u16 data_len = btrfs_dir_data_len(path->nodes[0], di);
2153 u32 this_len = sizeof(*di) + name_len + data_len;
2156 ret = verify_dir_item(fs_info, path->nodes[0], i, di);
2161 name = kmalloc(name_len, GFP_NOFS);
2166 read_extent_buffer(path->nodes[0], name,
2167 (unsigned long)(di + 1), name_len);
2169 log_di = btrfs_lookup_xattr(NULL, log, log_path, ino,
2171 btrfs_release_path(log_path);
2173 /* Doesn't exist in log tree, so delete it. */
2174 btrfs_release_path(path);
2175 di = btrfs_lookup_xattr(trans, root, path, ino,
2176 name, name_len, -1);
2183 ret = btrfs_delete_one_dir_name(trans, root,
2187 btrfs_release_path(path);
2192 if (IS_ERR(log_di)) {
2193 ret = PTR_ERR(log_di);
2197 di = (struct btrfs_dir_item *)((char *)di + this_len);
2200 ret = btrfs_next_leaf(root, path);
2206 btrfs_free_path(log_path);
2207 btrfs_release_path(path);
2213 * deletion replay happens before we copy any new directory items
2214 * out of the log or out of backreferences from inodes. It
2215 * scans the log to find ranges of keys that log is authoritative for,
2216 * and then scans the directory to find items in those ranges that are
2217 * not present in the log.
2219 * Anything we don't find in the log is unlinked and removed from the
2222 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
2223 struct btrfs_root *root,
2224 struct btrfs_root *log,
2225 struct btrfs_path *path,
2226 u64 dirid, int del_all)
2230 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
2232 struct btrfs_key dir_key;
2233 struct btrfs_key found_key;
2234 struct btrfs_path *log_path;
2237 dir_key.objectid = dirid;
2238 dir_key.type = BTRFS_DIR_ITEM_KEY;
2239 log_path = btrfs_alloc_path();
2243 dir = read_one_inode(root, dirid);
2244 /* it isn't an error if the inode isn't there, that can happen
2245 * because we replay the deletes before we copy in the inode item
2249 btrfs_free_path(log_path);
2257 range_end = (u64)-1;
2259 ret = find_dir_range(log, path, dirid, key_type,
2260 &range_start, &range_end);
2265 dir_key.offset = range_start;
2268 ret = btrfs_search_slot(NULL, root, &dir_key, path,
2273 nritems = btrfs_header_nritems(path->nodes[0]);
2274 if (path->slots[0] >= nritems) {
2275 ret = btrfs_next_leaf(root, path);
2279 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2281 if (found_key.objectid != dirid ||
2282 found_key.type != dir_key.type)
2285 if (found_key.offset > range_end)
2288 ret = check_item_in_log(trans, root, log, path,
2293 if (found_key.offset == (u64)-1)
2295 dir_key.offset = found_key.offset + 1;
2297 btrfs_release_path(path);
2298 if (range_end == (u64)-1)
2300 range_start = range_end + 1;
2305 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2306 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2307 dir_key.type = BTRFS_DIR_INDEX_KEY;
2308 btrfs_release_path(path);
2312 btrfs_release_path(path);
2313 btrfs_free_path(log_path);
2319 * the process_func used to replay items from the log tree. This
2320 * gets called in two different stages. The first stage just looks
2321 * for inodes and makes sure they are all copied into the subvolume.
2323 * The second stage copies all the other item types from the log into
2324 * the subvolume. The two stage approach is slower, but gets rid of
2325 * lots of complexity around inodes referencing other inodes that exist
2326 * only in the log (references come from either directory items or inode
2329 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2330 struct walk_control *wc, u64 gen)
2333 struct btrfs_path *path;
2334 struct btrfs_root *root = wc->replay_dest;
2335 struct btrfs_key key;
2340 ret = btrfs_read_buffer(eb, gen);
2344 level = btrfs_header_level(eb);
2349 path = btrfs_alloc_path();
2353 nritems = btrfs_header_nritems(eb);
2354 for (i = 0; i < nritems; i++) {
2355 btrfs_item_key_to_cpu(eb, &key, i);
2357 /* inode keys are done during the first stage */
2358 if (key.type == BTRFS_INODE_ITEM_KEY &&
2359 wc->stage == LOG_WALK_REPLAY_INODES) {
2360 struct btrfs_inode_item *inode_item;
2363 inode_item = btrfs_item_ptr(eb, i,
2364 struct btrfs_inode_item);
2365 ret = replay_xattr_deletes(wc->trans, root, log,
2366 path, key.objectid);
2369 mode = btrfs_inode_mode(eb, inode_item);
2370 if (S_ISDIR(mode)) {
2371 ret = replay_dir_deletes(wc->trans,
2372 root, log, path, key.objectid, 0);
2376 ret = overwrite_item(wc->trans, root, path,
2381 /* for regular files, make sure corresponding
2382 * orphan item exist. extents past the new EOF
2383 * will be truncated later by orphan cleanup.
2385 if (S_ISREG(mode)) {
2386 ret = insert_orphan_item(wc->trans, root,
2392 ret = link_to_fixup_dir(wc->trans, root,
2393 path, key.objectid);
2398 if (key.type == BTRFS_DIR_INDEX_KEY &&
2399 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2400 ret = replay_one_dir_item(wc->trans, root, path,
2406 if (wc->stage < LOG_WALK_REPLAY_ALL)
2409 /* these keys are simply copied */
2410 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2411 ret = overwrite_item(wc->trans, root, path,
2415 } else if (key.type == BTRFS_INODE_REF_KEY ||
2416 key.type == BTRFS_INODE_EXTREF_KEY) {
2417 ret = add_inode_ref(wc->trans, root, log, path,
2419 if (ret && ret != -ENOENT)
2422 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2423 ret = replay_one_extent(wc->trans, root, path,
2427 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2428 ret = replay_one_dir_item(wc->trans, root, path,
2434 btrfs_free_path(path);
2438 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2439 struct btrfs_root *root,
2440 struct btrfs_path *path, int *level,
2441 struct walk_control *wc)
2443 struct btrfs_fs_info *fs_info = root->fs_info;
2447 struct extent_buffer *next;
2448 struct extent_buffer *cur;
2449 struct extent_buffer *parent;
2453 WARN_ON(*level < 0);
2454 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2456 while (*level > 0) {
2457 WARN_ON(*level < 0);
2458 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2459 cur = path->nodes[*level];
2461 WARN_ON(btrfs_header_level(cur) != *level);
2463 if (path->slots[*level] >=
2464 btrfs_header_nritems(cur))
2467 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2468 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2469 blocksize = fs_info->nodesize;
2471 parent = path->nodes[*level];
2472 root_owner = btrfs_header_owner(parent);
2474 next = btrfs_find_create_tree_block(fs_info, bytenr);
2476 return PTR_ERR(next);
2479 ret = wc->process_func(root, next, wc, ptr_gen);
2481 free_extent_buffer(next);
2485 path->slots[*level]++;
2487 ret = btrfs_read_buffer(next, ptr_gen);
2489 free_extent_buffer(next);
2494 btrfs_tree_lock(next);
2495 btrfs_set_lock_blocking(next);
2496 clean_tree_block(fs_info, next);
2497 btrfs_wait_tree_block_writeback(next);
2498 btrfs_tree_unlock(next);
2501 WARN_ON(root_owner !=
2502 BTRFS_TREE_LOG_OBJECTID);
2503 ret = btrfs_free_and_pin_reserved_extent(
2507 free_extent_buffer(next);
2511 free_extent_buffer(next);
2514 ret = btrfs_read_buffer(next, ptr_gen);
2516 free_extent_buffer(next);
2520 WARN_ON(*level <= 0);
2521 if (path->nodes[*level-1])
2522 free_extent_buffer(path->nodes[*level-1]);
2523 path->nodes[*level-1] = next;
2524 *level = btrfs_header_level(next);
2525 path->slots[*level] = 0;
2528 WARN_ON(*level < 0);
2529 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2531 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2537 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2538 struct btrfs_root *root,
2539 struct btrfs_path *path, int *level,
2540 struct walk_control *wc)
2542 struct btrfs_fs_info *fs_info = root->fs_info;
2548 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2549 slot = path->slots[i];
2550 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2553 WARN_ON(*level == 0);
2556 struct extent_buffer *parent;
2557 if (path->nodes[*level] == root->node)
2558 parent = path->nodes[*level];
2560 parent = path->nodes[*level + 1];
2562 root_owner = btrfs_header_owner(parent);
2563 ret = wc->process_func(root, path->nodes[*level], wc,
2564 btrfs_header_generation(path->nodes[*level]));
2569 struct extent_buffer *next;
2571 next = path->nodes[*level];
2574 btrfs_tree_lock(next);
2575 btrfs_set_lock_blocking(next);
2576 clean_tree_block(fs_info, next);
2577 btrfs_wait_tree_block_writeback(next);
2578 btrfs_tree_unlock(next);
2581 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2582 ret = btrfs_free_and_pin_reserved_extent(
2584 path->nodes[*level]->start,
2585 path->nodes[*level]->len);
2589 free_extent_buffer(path->nodes[*level]);
2590 path->nodes[*level] = NULL;
2598 * drop the reference count on the tree rooted at 'snap'. This traverses
2599 * the tree freeing any blocks that have a ref count of zero after being
2602 static int walk_log_tree(struct btrfs_trans_handle *trans,
2603 struct btrfs_root *log, struct walk_control *wc)
2605 struct btrfs_fs_info *fs_info = log->fs_info;
2609 struct btrfs_path *path;
2612 path = btrfs_alloc_path();
2616 level = btrfs_header_level(log->node);
2618 path->nodes[level] = log->node;
2619 extent_buffer_get(log->node);
2620 path->slots[level] = 0;
2623 wret = walk_down_log_tree(trans, log, path, &level, wc);
2631 wret = walk_up_log_tree(trans, log, path, &level, wc);
2640 /* was the root node processed? if not, catch it here */
2641 if (path->nodes[orig_level]) {
2642 ret = wc->process_func(log, path->nodes[orig_level], wc,
2643 btrfs_header_generation(path->nodes[orig_level]));
2647 struct extent_buffer *next;
2649 next = path->nodes[orig_level];
2652 btrfs_tree_lock(next);
2653 btrfs_set_lock_blocking(next);
2654 clean_tree_block(fs_info, next);
2655 btrfs_wait_tree_block_writeback(next);
2656 btrfs_tree_unlock(next);
2659 WARN_ON(log->root_key.objectid !=
2660 BTRFS_TREE_LOG_OBJECTID);
2661 ret = btrfs_free_and_pin_reserved_extent(fs_info,
2662 next->start, next->len);
2669 btrfs_free_path(path);
2674 * helper function to update the item for a given subvolumes log root
2675 * in the tree of log roots
2677 static int update_log_root(struct btrfs_trans_handle *trans,
2678 struct btrfs_root *log)
2680 struct btrfs_fs_info *fs_info = log->fs_info;
2683 if (log->log_transid == 1) {
2684 /* insert root item on the first sync */
2685 ret = btrfs_insert_root(trans, fs_info->log_root_tree,
2686 &log->root_key, &log->root_item);
2688 ret = btrfs_update_root(trans, fs_info->log_root_tree,
2689 &log->root_key, &log->root_item);
2694 static void wait_log_commit(struct btrfs_root *root, int transid)
2697 int index = transid % 2;
2700 * we only allow two pending log transactions at a time,
2701 * so we know that if ours is more than 2 older than the
2702 * current transaction, we're done
2705 prepare_to_wait(&root->log_commit_wait[index],
2706 &wait, TASK_UNINTERRUPTIBLE);
2707 mutex_unlock(&root->log_mutex);
2709 if (root->log_transid_committed < transid &&
2710 atomic_read(&root->log_commit[index]))
2713 finish_wait(&root->log_commit_wait[index], &wait);
2714 mutex_lock(&root->log_mutex);
2715 } while (root->log_transid_committed < transid &&
2716 atomic_read(&root->log_commit[index]));
2719 static void wait_for_writer(struct btrfs_root *root)
2723 while (atomic_read(&root->log_writers)) {
2724 prepare_to_wait(&root->log_writer_wait,
2725 &wait, TASK_UNINTERRUPTIBLE);
2726 mutex_unlock(&root->log_mutex);
2727 if (atomic_read(&root->log_writers))
2729 finish_wait(&root->log_writer_wait, &wait);
2730 mutex_lock(&root->log_mutex);
2734 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2735 struct btrfs_log_ctx *ctx)
2740 mutex_lock(&root->log_mutex);
2741 list_del_init(&ctx->list);
2742 mutex_unlock(&root->log_mutex);
2746 * Invoked in log mutex context, or be sure there is no other task which
2747 * can access the list.
2749 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2750 int index, int error)
2752 struct btrfs_log_ctx *ctx;
2753 struct btrfs_log_ctx *safe;
2755 list_for_each_entry_safe(ctx, safe, &root->log_ctxs[index], list) {
2756 list_del_init(&ctx->list);
2757 ctx->log_ret = error;
2760 INIT_LIST_HEAD(&root->log_ctxs[index]);
2764 * btrfs_sync_log does sends a given tree log down to the disk and
2765 * updates the super blocks to record it. When this call is done,
2766 * you know that any inodes previously logged are safely on disk only
2769 * Any other return value means you need to call btrfs_commit_transaction.
2770 * Some of the edge cases for fsyncing directories that have had unlinks
2771 * or renames done in the past mean that sometimes the only safe
2772 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2773 * that has happened.
2775 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2776 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2782 struct btrfs_fs_info *fs_info = root->fs_info;
2783 struct btrfs_root *log = root->log_root;
2784 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
2785 int log_transid = 0;
2786 struct btrfs_log_ctx root_log_ctx;
2787 struct blk_plug plug;
2789 mutex_lock(&root->log_mutex);
2790 log_transid = ctx->log_transid;
2791 if (root->log_transid_committed >= log_transid) {
2792 mutex_unlock(&root->log_mutex);
2793 return ctx->log_ret;
2796 index1 = log_transid % 2;
2797 if (atomic_read(&root->log_commit[index1])) {
2798 wait_log_commit(root, log_transid);
2799 mutex_unlock(&root->log_mutex);
2800 return ctx->log_ret;
2802 ASSERT(log_transid == root->log_transid);
2803 atomic_set(&root->log_commit[index1], 1);
2805 /* wait for previous tree log sync to complete */
2806 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2807 wait_log_commit(root, log_transid - 1);
2810 int batch = atomic_read(&root->log_batch);
2811 /* when we're on an ssd, just kick the log commit out */
2812 if (!btrfs_test_opt(fs_info, SSD) &&
2813 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2814 mutex_unlock(&root->log_mutex);
2815 schedule_timeout_uninterruptible(1);
2816 mutex_lock(&root->log_mutex);
2818 wait_for_writer(root);
2819 if (batch == atomic_read(&root->log_batch))
2823 /* bail out if we need to do a full commit */
2824 if (btrfs_need_log_full_commit(fs_info, trans)) {
2826 btrfs_free_logged_extents(log, log_transid);
2827 mutex_unlock(&root->log_mutex);
2831 if (log_transid % 2 == 0)
2832 mark = EXTENT_DIRTY;
2836 /* we start IO on all the marked extents here, but we don't actually
2837 * wait for them until later.
2839 blk_start_plug(&plug);
2840 ret = btrfs_write_marked_extents(fs_info, &log->dirty_log_pages, mark);
2842 blk_finish_plug(&plug);
2843 btrfs_abort_transaction(trans, ret);
2844 btrfs_free_logged_extents(log, log_transid);
2845 btrfs_set_log_full_commit(fs_info, trans);
2846 mutex_unlock(&root->log_mutex);
2850 btrfs_set_root_node(&log->root_item, log->node);
2852 root->log_transid++;
2853 log->log_transid = root->log_transid;
2854 root->log_start_pid = 0;
2856 * IO has been started, blocks of the log tree have WRITTEN flag set
2857 * in their headers. new modifications of the log will be written to
2858 * new positions. so it's safe to allow log writers to go in.
2860 mutex_unlock(&root->log_mutex);
2862 btrfs_init_log_ctx(&root_log_ctx, NULL);
2864 mutex_lock(&log_root_tree->log_mutex);
2865 atomic_inc(&log_root_tree->log_batch);
2866 atomic_inc(&log_root_tree->log_writers);
2868 index2 = log_root_tree->log_transid % 2;
2869 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2870 root_log_ctx.log_transid = log_root_tree->log_transid;
2872 mutex_unlock(&log_root_tree->log_mutex);
2874 ret = update_log_root(trans, log);
2876 mutex_lock(&log_root_tree->log_mutex);
2877 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2879 * Implicit memory barrier after atomic_dec_and_test
2881 if (waitqueue_active(&log_root_tree->log_writer_wait))
2882 wake_up(&log_root_tree->log_writer_wait);
2886 if (!list_empty(&root_log_ctx.list))
2887 list_del_init(&root_log_ctx.list);
2889 blk_finish_plug(&plug);
2890 btrfs_set_log_full_commit(fs_info, trans);
2892 if (ret != -ENOSPC) {
2893 btrfs_abort_transaction(trans, ret);
2894 mutex_unlock(&log_root_tree->log_mutex);
2897 btrfs_wait_tree_log_extents(log, mark);
2898 btrfs_free_logged_extents(log, log_transid);
2899 mutex_unlock(&log_root_tree->log_mutex);
2904 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2905 blk_finish_plug(&plug);
2906 list_del_init(&root_log_ctx.list);
2907 mutex_unlock(&log_root_tree->log_mutex);
2908 ret = root_log_ctx.log_ret;
2912 index2 = root_log_ctx.log_transid % 2;
2913 if (atomic_read(&log_root_tree->log_commit[index2])) {
2914 blk_finish_plug(&plug);
2915 ret = btrfs_wait_tree_log_extents(log, mark);
2916 btrfs_wait_logged_extents(trans, log, log_transid);
2917 wait_log_commit(log_root_tree,
2918 root_log_ctx.log_transid);
2919 mutex_unlock(&log_root_tree->log_mutex);
2921 ret = root_log_ctx.log_ret;
2924 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2925 atomic_set(&log_root_tree->log_commit[index2], 1);
2927 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2928 wait_log_commit(log_root_tree,
2929 root_log_ctx.log_transid - 1);
2932 wait_for_writer(log_root_tree);
2935 * now that we've moved on to the tree of log tree roots,
2936 * check the full commit flag again
2938 if (btrfs_need_log_full_commit(fs_info, trans)) {
2939 blk_finish_plug(&plug);
2940 btrfs_wait_tree_log_extents(log, mark);
2941 btrfs_free_logged_extents(log, log_transid);
2942 mutex_unlock(&log_root_tree->log_mutex);
2944 goto out_wake_log_root;
2947 ret = btrfs_write_marked_extents(fs_info,
2948 &log_root_tree->dirty_log_pages,
2949 EXTENT_DIRTY | EXTENT_NEW);
2950 blk_finish_plug(&plug);
2952 btrfs_set_log_full_commit(fs_info, trans);
2953 btrfs_abort_transaction(trans, ret);
2954 btrfs_free_logged_extents(log, log_transid);
2955 mutex_unlock(&log_root_tree->log_mutex);
2956 goto out_wake_log_root;
2958 ret = btrfs_wait_tree_log_extents(log, mark);
2960 ret = btrfs_wait_tree_log_extents(log_root_tree,
2961 EXTENT_NEW | EXTENT_DIRTY);
2963 btrfs_set_log_full_commit(fs_info, trans);
2964 btrfs_free_logged_extents(log, log_transid);
2965 mutex_unlock(&log_root_tree->log_mutex);
2966 goto out_wake_log_root;
2968 btrfs_wait_logged_extents(trans, log, log_transid);
2970 btrfs_set_super_log_root(fs_info->super_for_commit,
2971 log_root_tree->node->start);
2972 btrfs_set_super_log_root_level(fs_info->super_for_commit,
2973 btrfs_header_level(log_root_tree->node));
2975 log_root_tree->log_transid++;
2976 mutex_unlock(&log_root_tree->log_mutex);
2979 * nobody else is going to jump in and write the the ctree
2980 * super here because the log_commit atomic below is protecting
2981 * us. We must be called with a transaction handle pinning
2982 * the running transaction open, so a full commit can't hop
2983 * in and cause problems either.
2985 ret = write_all_supers(fs_info, 1);
2987 btrfs_set_log_full_commit(fs_info, trans);
2988 btrfs_abort_transaction(trans, ret);
2989 goto out_wake_log_root;
2992 mutex_lock(&root->log_mutex);
2993 if (root->last_log_commit < log_transid)
2994 root->last_log_commit = log_transid;
2995 mutex_unlock(&root->log_mutex);
2998 mutex_lock(&log_root_tree->log_mutex);
2999 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
3001 log_root_tree->log_transid_committed++;
3002 atomic_set(&log_root_tree->log_commit[index2], 0);
3003 mutex_unlock(&log_root_tree->log_mutex);
3006 * The barrier before waitqueue_active is implied by mutex_unlock
3008 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
3009 wake_up(&log_root_tree->log_commit_wait[index2]);
3011 mutex_lock(&root->log_mutex);
3012 btrfs_remove_all_log_ctxs(root, index1, ret);
3013 root->log_transid_committed++;
3014 atomic_set(&root->log_commit[index1], 0);
3015 mutex_unlock(&root->log_mutex);
3018 * The barrier before waitqueue_active is implied by mutex_unlock
3020 if (waitqueue_active(&root->log_commit_wait[index1]))
3021 wake_up(&root->log_commit_wait[index1]);
3025 static void free_log_tree(struct btrfs_trans_handle *trans,
3026 struct btrfs_root *log)
3031 struct walk_control wc = {
3033 .process_func = process_one_buffer
3036 ret = walk_log_tree(trans, log, &wc);
3037 /* I don't think this can happen but just in case */
3039 btrfs_abort_transaction(trans, ret);
3042 ret = find_first_extent_bit(&log->dirty_log_pages,
3043 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
3048 clear_extent_bits(&log->dirty_log_pages, start, end,
3049 EXTENT_DIRTY | EXTENT_NEW);
3053 * We may have short-circuited the log tree with the full commit logic
3054 * and left ordered extents on our list, so clear these out to keep us
3055 * from leaking inodes and memory.
3057 btrfs_free_logged_extents(log, 0);
3058 btrfs_free_logged_extents(log, 1);
3060 free_extent_buffer(log->node);
3065 * free all the extents used by the tree log. This should be called
3066 * at commit time of the full transaction
3068 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
3070 if (root->log_root) {
3071 free_log_tree(trans, root->log_root);
3072 root->log_root = NULL;
3077 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
3078 struct btrfs_fs_info *fs_info)
3080 if (fs_info->log_root_tree) {
3081 free_log_tree(trans, fs_info->log_root_tree);
3082 fs_info->log_root_tree = NULL;
3088 * If both a file and directory are logged, and unlinks or renames are
3089 * mixed in, we have a few interesting corners:
3091 * create file X in dir Y
3092 * link file X to X.link in dir Y
3094 * unlink file X but leave X.link
3097 * After a crash we would expect only X.link to exist. But file X
3098 * didn't get fsync'd again so the log has back refs for X and X.link.
3100 * We solve this by removing directory entries and inode backrefs from the
3101 * log when a file that was logged in the current transaction is
3102 * unlinked. Any later fsync will include the updated log entries, and
3103 * we'll be able to reconstruct the proper directory items from backrefs.
3105 * This optimizations allows us to avoid relogging the entire inode
3106 * or the entire directory.
3108 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
3109 struct btrfs_root *root,
3110 const char *name, int name_len,
3111 struct btrfs_inode *dir, u64 index)
3113 struct btrfs_root *log;
3114 struct btrfs_dir_item *di;
3115 struct btrfs_path *path;
3119 u64 dir_ino = btrfs_ino(dir);
3121 if (dir->logged_trans < trans->transid)
3124 ret = join_running_log_trans(root);
3128 mutex_lock(&dir->log_mutex);
3130 log = root->log_root;
3131 path = btrfs_alloc_path();
3137 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
3138 name, name_len, -1);
3144 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3145 bytes_del += name_len;
3151 btrfs_release_path(path);
3152 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
3153 index, name, name_len, -1);
3159 ret = btrfs_delete_one_dir_name(trans, log, path, di);
3160 bytes_del += name_len;
3167 /* update the directory size in the log to reflect the names
3171 struct btrfs_key key;
3173 key.objectid = dir_ino;
3175 key.type = BTRFS_INODE_ITEM_KEY;
3176 btrfs_release_path(path);
3178 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
3184 struct btrfs_inode_item *item;
3187 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3188 struct btrfs_inode_item);
3189 i_size = btrfs_inode_size(path->nodes[0], item);
3190 if (i_size > bytes_del)
3191 i_size -= bytes_del;
3194 btrfs_set_inode_size(path->nodes[0], item, i_size);
3195 btrfs_mark_buffer_dirty(path->nodes[0]);
3198 btrfs_release_path(path);
3201 btrfs_free_path(path);
3203 mutex_unlock(&dir->log_mutex);
3204 if (ret == -ENOSPC) {
3205 btrfs_set_log_full_commit(root->fs_info, trans);
3208 btrfs_abort_transaction(trans, ret);
3210 btrfs_end_log_trans(root);
3215 /* see comments for btrfs_del_dir_entries_in_log */
3216 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root,
3218 const char *name, int name_len,
3219 struct btrfs_inode *inode, u64 dirid)
3221 struct btrfs_fs_info *fs_info = root->fs_info;
3222 struct btrfs_root *log;
3226 if (inode->logged_trans < trans->transid)
3229 ret = join_running_log_trans(root);
3232 log = root->log_root;
3233 mutex_lock(&inode->log_mutex);
3235 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
3237 mutex_unlock(&inode->log_mutex);
3238 if (ret == -ENOSPC) {
3239 btrfs_set_log_full_commit(fs_info, trans);
3241 } else if (ret < 0 && ret != -ENOENT)
3242 btrfs_abort_transaction(trans, ret);
3243 btrfs_end_log_trans(root);
3249 * creates a range item in the log for 'dirid'. first_offset and
3250 * last_offset tell us which parts of the key space the log should
3251 * be considered authoritative for.
3253 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
3254 struct btrfs_root *log,
3255 struct btrfs_path *path,
3256 int key_type, u64 dirid,
3257 u64 first_offset, u64 last_offset)
3260 struct btrfs_key key;
3261 struct btrfs_dir_log_item *item;
3263 key.objectid = dirid;
3264 key.offset = first_offset;
3265 if (key_type == BTRFS_DIR_ITEM_KEY)
3266 key.type = BTRFS_DIR_LOG_ITEM_KEY;
3268 key.type = BTRFS_DIR_LOG_INDEX_KEY;
3269 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
3273 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3274 struct btrfs_dir_log_item);
3275 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
3276 btrfs_mark_buffer_dirty(path->nodes[0]);
3277 btrfs_release_path(path);
3282 * log all the items included in the current transaction for a given
3283 * directory. This also creates the range items in the log tree required
3284 * to replay anything deleted before the fsync
3286 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
3287 struct btrfs_root *root, struct btrfs_inode *inode,
3288 struct btrfs_path *path,
3289 struct btrfs_path *dst_path, int key_type,
3290 struct btrfs_log_ctx *ctx,
3291 u64 min_offset, u64 *last_offset_ret)
3293 struct btrfs_key min_key;
3294 struct btrfs_root *log = root->log_root;
3295 struct extent_buffer *src;
3300 u64 first_offset = min_offset;
3301 u64 last_offset = (u64)-1;
3302 u64 ino = btrfs_ino(inode);
3304 log = root->log_root;
3306 min_key.objectid = ino;
3307 min_key.type = key_type;
3308 min_key.offset = min_offset;
3310 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3313 * we didn't find anything from this transaction, see if there
3314 * is anything at all
3316 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3317 min_key.objectid = ino;
3318 min_key.type = key_type;
3319 min_key.offset = (u64)-1;
3320 btrfs_release_path(path);
3321 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3323 btrfs_release_path(path);
3326 ret = btrfs_previous_item(root, path, ino, key_type);
3328 /* if ret == 0 there are items for this type,
3329 * create a range to tell us the last key of this type.
3330 * otherwise, there are no items in this directory after
3331 * *min_offset, and we create a range to indicate that.
3334 struct btrfs_key tmp;
3335 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3337 if (key_type == tmp.type)
3338 first_offset = max(min_offset, tmp.offset) + 1;
3343 /* go backward to find any previous key */
3344 ret = btrfs_previous_item(root, path, ino, key_type);
3346 struct btrfs_key tmp;
3347 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3348 if (key_type == tmp.type) {
3349 first_offset = tmp.offset;
3350 ret = overwrite_item(trans, log, dst_path,
3351 path->nodes[0], path->slots[0],
3359 btrfs_release_path(path);
3361 /* find the first key from this transaction again */
3362 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3363 if (WARN_ON(ret != 0))
3367 * we have a block from this transaction, log every item in it
3368 * from our directory
3371 struct btrfs_key tmp;
3372 src = path->nodes[0];
3373 nritems = btrfs_header_nritems(src);
3374 for (i = path->slots[0]; i < nritems; i++) {
3375 struct btrfs_dir_item *di;
3377 btrfs_item_key_to_cpu(src, &min_key, i);
3379 if (min_key.objectid != ino || min_key.type != key_type)
3381 ret = overwrite_item(trans, log, dst_path, src, i,
3389 * We must make sure that when we log a directory entry,
3390 * the corresponding inode, after log replay, has a
3391 * matching link count. For example:
3397 * xfs_io -c "fsync" mydir
3399 * <mount fs and log replay>
3401 * Would result in a fsync log that when replayed, our
3402 * file inode would have a link count of 1, but we get
3403 * two directory entries pointing to the same inode.
3404 * After removing one of the names, it would not be
3405 * possible to remove the other name, which resulted
3406 * always in stale file handle errors, and would not
3407 * be possible to rmdir the parent directory, since
3408 * its i_size could never decrement to the value
3409 * BTRFS_EMPTY_DIR_SIZE, resulting in -ENOTEMPTY errors.
3411 di = btrfs_item_ptr(src, i, struct btrfs_dir_item);
3412 btrfs_dir_item_key_to_cpu(src, di, &tmp);
3414 (btrfs_dir_transid(src, di) == trans->transid ||
3415 btrfs_dir_type(src, di) == BTRFS_FT_DIR) &&
3416 tmp.type != BTRFS_ROOT_ITEM_KEY)
3417 ctx->log_new_dentries = true;
3419 path->slots[0] = nritems;
3422 * look ahead to the next item and see if it is also
3423 * from this directory and from this transaction
3425 ret = btrfs_next_leaf(root, path);
3427 last_offset = (u64)-1;
3430 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3431 if (tmp.objectid != ino || tmp.type != key_type) {
3432 last_offset = (u64)-1;
3435 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3436 ret = overwrite_item(trans, log, dst_path,
3437 path->nodes[0], path->slots[0],
3442 last_offset = tmp.offset;
3447 btrfs_release_path(path);
3448 btrfs_release_path(dst_path);
3451 *last_offset_ret = last_offset;
3453 * insert the log range keys to indicate where the log
3456 ret = insert_dir_log_key(trans, log, path, key_type,
3457 ino, first_offset, last_offset);
3465 * logging directories is very similar to logging inodes, We find all the items
3466 * from the current transaction and write them to the log.
3468 * The recovery code scans the directory in the subvolume, and if it finds a
3469 * key in the range logged that is not present in the log tree, then it means
3470 * that dir entry was unlinked during the transaction.
3472 * In order for that scan to work, we must include one key smaller than
3473 * the smallest logged by this transaction and one key larger than the largest
3474 * key logged by this transaction.
3476 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3477 struct btrfs_root *root, struct btrfs_inode *inode,
3478 struct btrfs_path *path,
3479 struct btrfs_path *dst_path,
3480 struct btrfs_log_ctx *ctx)
3485 int key_type = BTRFS_DIR_ITEM_KEY;
3491 ret = log_dir_items(trans, root, inode, path, dst_path, key_type,
3492 ctx, min_key, &max_key);
3495 if (max_key == (u64)-1)
3497 min_key = max_key + 1;
3500 if (key_type == BTRFS_DIR_ITEM_KEY) {
3501 key_type = BTRFS_DIR_INDEX_KEY;
3508 * a helper function to drop items from the log before we relog an
3509 * inode. max_key_type indicates the highest item type to remove.
3510 * This cannot be run for file data extents because it does not
3511 * free the extents they point to.
3513 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3514 struct btrfs_root *log,
3515 struct btrfs_path *path,
3516 u64 objectid, int max_key_type)
3519 struct btrfs_key key;
3520 struct btrfs_key found_key;
3523 key.objectid = objectid;
3524 key.type = max_key_type;
3525 key.offset = (u64)-1;
3528 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3529 BUG_ON(ret == 0); /* Logic error */
3533 if (path->slots[0] == 0)
3537 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3540 if (found_key.objectid != objectid)
3543 found_key.offset = 0;
3545 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3548 ret = btrfs_del_items(trans, log, path, start_slot,
3549 path->slots[0] - start_slot + 1);
3551 * If start slot isn't 0 then we don't need to re-search, we've
3552 * found the last guy with the objectid in this tree.
3554 if (ret || start_slot != 0)
3556 btrfs_release_path(path);
3558 btrfs_release_path(path);
3564 static void fill_inode_item(struct btrfs_trans_handle *trans,
3565 struct extent_buffer *leaf,
3566 struct btrfs_inode_item *item,
3567 struct inode *inode, int log_inode_only,
3570 struct btrfs_map_token token;
3572 btrfs_init_map_token(&token);
3574 if (log_inode_only) {
3575 /* set the generation to zero so the recover code
3576 * can tell the difference between an logging
3577 * just to say 'this inode exists' and a logging
3578 * to say 'update this inode with these values'
3580 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3581 btrfs_set_token_inode_size(leaf, item, logged_isize, &token);
3583 btrfs_set_token_inode_generation(leaf, item,
3584 BTRFS_I(inode)->generation,
3586 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3589 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3590 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3591 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3592 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3594 btrfs_set_token_timespec_sec(leaf, &item->atime,
3595 inode->i_atime.tv_sec, &token);
3596 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3597 inode->i_atime.tv_nsec, &token);
3599 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3600 inode->i_mtime.tv_sec, &token);
3601 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3602 inode->i_mtime.tv_nsec, &token);
3604 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3605 inode->i_ctime.tv_sec, &token);
3606 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3607 inode->i_ctime.tv_nsec, &token);
3609 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3612 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3613 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3614 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3615 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3616 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3619 static int log_inode_item(struct btrfs_trans_handle *trans,
3620 struct btrfs_root *log, struct btrfs_path *path,
3621 struct btrfs_inode *inode)
3623 struct btrfs_inode_item *inode_item;
3626 ret = btrfs_insert_empty_item(trans, log, path,
3627 &inode->location, sizeof(*inode_item));
3628 if (ret && ret != -EEXIST)
3630 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3631 struct btrfs_inode_item);
3632 fill_inode_item(trans, path->nodes[0], inode_item, &inode->vfs_inode,
3634 btrfs_release_path(path);
3638 static noinline int copy_items(struct btrfs_trans_handle *trans,
3639 struct btrfs_inode *inode,
3640 struct btrfs_path *dst_path,
3641 struct btrfs_path *src_path, u64 *last_extent,
3642 int start_slot, int nr, int inode_only,
3645 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3646 unsigned long src_offset;
3647 unsigned long dst_offset;
3648 struct btrfs_root *log = inode->root->log_root;
3649 struct btrfs_file_extent_item *extent;
3650 struct btrfs_inode_item *inode_item;
3651 struct extent_buffer *src = src_path->nodes[0];
3652 struct btrfs_key first_key, last_key, key;
3654 struct btrfs_key *ins_keys;
3658 struct list_head ordered_sums;
3659 int skip_csum = inode->flags & BTRFS_INODE_NODATASUM;
3660 bool has_extents = false;
3661 bool need_find_last_extent = true;
3664 INIT_LIST_HEAD(&ordered_sums);
3666 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3667 nr * sizeof(u32), GFP_NOFS);
3671 first_key.objectid = (u64)-1;
3673 ins_sizes = (u32 *)ins_data;
3674 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3676 for (i = 0; i < nr; i++) {
3677 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3678 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3680 ret = btrfs_insert_empty_items(trans, log, dst_path,
3681 ins_keys, ins_sizes, nr);
3687 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3688 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3689 dst_path->slots[0]);
3691 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3693 if ((i == (nr - 1)))
3694 last_key = ins_keys[i];
3696 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3697 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3699 struct btrfs_inode_item);
3700 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3702 inode_only == LOG_INODE_EXISTS,
3705 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3706 src_offset, ins_sizes[i]);
3710 * We set need_find_last_extent here in case we know we were
3711 * processing other items and then walk into the first extent in
3712 * the inode. If we don't hit an extent then nothing changes,
3713 * we'll do the last search the next time around.
3715 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3717 if (first_key.objectid == (u64)-1)
3718 first_key = ins_keys[i];
3720 need_find_last_extent = false;
3723 /* take a reference on file data extents so that truncates
3724 * or deletes of this inode don't have to relog the inode
3727 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3730 extent = btrfs_item_ptr(src, start_slot + i,
3731 struct btrfs_file_extent_item);
3733 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3736 found_type = btrfs_file_extent_type(src, extent);
3737 if (found_type == BTRFS_FILE_EXTENT_REG) {
3739 ds = btrfs_file_extent_disk_bytenr(src,
3741 /* ds == 0 is a hole */
3745 dl = btrfs_file_extent_disk_num_bytes(src,
3747 cs = btrfs_file_extent_offset(src, extent);
3748 cl = btrfs_file_extent_num_bytes(src,
3750 if (btrfs_file_extent_compression(src,
3756 ret = btrfs_lookup_csums_range(
3758 ds + cs, ds + cs + cl - 1,
3761 btrfs_release_path(dst_path);
3769 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3770 btrfs_release_path(dst_path);
3774 * we have to do this after the loop above to avoid changing the
3775 * log tree while trying to change the log tree.
3778 while (!list_empty(&ordered_sums)) {
3779 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3780 struct btrfs_ordered_sum,
3783 ret = btrfs_csum_file_blocks(trans, log, sums);
3784 list_del(&sums->list);
3791 if (need_find_last_extent && *last_extent == first_key.offset) {
3793 * We don't have any leafs between our current one and the one
3794 * we processed before that can have file extent items for our
3795 * inode (and have a generation number smaller than our current
3798 need_find_last_extent = false;
3802 * Because we use btrfs_search_forward we could skip leaves that were
3803 * not modified and then assume *last_extent is valid when it really
3804 * isn't. So back up to the previous leaf and read the end of the last
3805 * extent before we go and fill in holes.
3807 if (need_find_last_extent) {
3810 ret = btrfs_prev_leaf(inode->root, src_path);
3815 if (src_path->slots[0])
3816 src_path->slots[0]--;
3817 src = src_path->nodes[0];
3818 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3819 if (key.objectid != btrfs_ino(inode) ||
3820 key.type != BTRFS_EXTENT_DATA_KEY)
3822 extent = btrfs_item_ptr(src, src_path->slots[0],
3823 struct btrfs_file_extent_item);
3824 if (btrfs_file_extent_type(src, extent) ==
3825 BTRFS_FILE_EXTENT_INLINE) {
3826 len = btrfs_file_extent_inline_len(src,
3829 *last_extent = ALIGN(key.offset + len,
3830 fs_info->sectorsize);
3832 len = btrfs_file_extent_num_bytes(src, extent);
3833 *last_extent = key.offset + len;
3837 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3838 * things could have happened
3840 * 1) A merge could have happened, so we could currently be on a leaf
3841 * that holds what we were copying in the first place.
3842 * 2) A split could have happened, and now not all of the items we want
3843 * are on the same leaf.
3845 * So we need to adjust how we search for holes, we need to drop the
3846 * path and re-search for the first extent key we found, and then walk
3847 * forward until we hit the last one we copied.
3849 if (need_find_last_extent) {
3850 /* btrfs_prev_leaf could return 1 without releasing the path */
3851 btrfs_release_path(src_path);
3852 ret = btrfs_search_slot(NULL, inode->root, &first_key,
3857 src = src_path->nodes[0];
3858 i = src_path->slots[0];
3864 * Ok so here we need to go through and fill in any holes we may have
3865 * to make sure that holes are punched for those areas in case they had
3866 * extents previously.
3872 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3873 ret = btrfs_next_leaf(inode->root, src_path);
3877 src = src_path->nodes[0];
3881 btrfs_item_key_to_cpu(src, &key, i);
3882 if (!btrfs_comp_cpu_keys(&key, &last_key))
3884 if (key.objectid != btrfs_ino(inode) ||
3885 key.type != BTRFS_EXTENT_DATA_KEY) {
3889 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3890 if (btrfs_file_extent_type(src, extent) ==
3891 BTRFS_FILE_EXTENT_INLINE) {
3892 len = btrfs_file_extent_inline_len(src, i, extent);
3893 extent_end = ALIGN(key.offset + len,
3894 fs_info->sectorsize);
3896 len = btrfs_file_extent_num_bytes(src, extent);
3897 extent_end = key.offset + len;
3901 if (*last_extent == key.offset) {
3902 *last_extent = extent_end;
3905 offset = *last_extent;
3906 len = key.offset - *last_extent;
3907 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3908 offset, 0, 0, len, 0, len, 0, 0, 0);
3911 *last_extent = extent_end;
3914 * Need to let the callers know we dropped the path so they should
3917 if (!ret && need_find_last_extent)
3922 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3924 struct extent_map *em1, *em2;
3926 em1 = list_entry(a, struct extent_map, list);
3927 em2 = list_entry(b, struct extent_map, list);
3929 if (em1->start < em2->start)
3931 else if (em1->start > em2->start)
3936 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3937 struct inode *inode,
3938 struct btrfs_root *root,
3939 const struct extent_map *em,
3940 const struct list_head *logged_list,
3941 bool *ordered_io_error)
3943 struct btrfs_fs_info *fs_info = root->fs_info;
3944 struct btrfs_ordered_extent *ordered;
3945 struct btrfs_root *log = root->log_root;
3946 u64 mod_start = em->mod_start;
3947 u64 mod_len = em->mod_len;
3948 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3951 LIST_HEAD(ordered_sums);
3954 *ordered_io_error = false;
3956 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3957 em->block_start == EXTENT_MAP_HOLE)
3961 * Wait far any ordered extent that covers our extent map. If it
3962 * finishes without an error, first check and see if our csums are on
3963 * our outstanding ordered extents.
3965 list_for_each_entry(ordered, logged_list, log_list) {
3966 struct btrfs_ordered_sum *sum;
3971 if (ordered->file_offset + ordered->len <= mod_start ||
3972 mod_start + mod_len <= ordered->file_offset)
3975 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3976 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3977 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3978 const u64 start = ordered->file_offset;
3979 const u64 end = ordered->file_offset + ordered->len - 1;
3981 WARN_ON(ordered->inode != inode);
3982 filemap_fdatawrite_range(inode->i_mapping, start, end);
3985 wait_event(ordered->wait,
3986 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3987 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3989 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3991 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3992 * i_mapping flags, so that the next fsync won't get
3993 * an outdated io error too.
3995 filemap_check_errors(inode->i_mapping);
3996 *ordered_io_error = true;
4000 * We are going to copy all the csums on this ordered extent, so
4001 * go ahead and adjust mod_start and mod_len in case this
4002 * ordered extent has already been logged.
4004 if (ordered->file_offset > mod_start) {
4005 if (ordered->file_offset + ordered->len >=
4006 mod_start + mod_len)
4007 mod_len = ordered->file_offset - mod_start;
4009 * If we have this case
4011 * |--------- logged extent ---------|
4012 * |----- ordered extent ----|
4014 * Just don't mess with mod_start and mod_len, we'll
4015 * just end up logging more csums than we need and it
4019 if (ordered->file_offset + ordered->len <
4020 mod_start + mod_len) {
4021 mod_len = (mod_start + mod_len) -
4022 (ordered->file_offset + ordered->len);
4023 mod_start = ordered->file_offset +
4034 * To keep us from looping for the above case of an ordered
4035 * extent that falls inside of the logged extent.
4037 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
4041 list_for_each_entry(sum, &ordered->list, list) {
4042 ret = btrfs_csum_file_blocks(trans, log, sum);
4048 if (*ordered_io_error || !mod_len || ret || skip_csum)
4051 if (em->compress_type) {
4053 csum_len = max(em->block_len, em->orig_block_len);
4055 csum_offset = mod_start - em->start;
4059 /* block start is already adjusted for the file extent offset. */
4060 ret = btrfs_lookup_csums_range(fs_info->csum_root,
4061 em->block_start + csum_offset,
4062 em->block_start + csum_offset +
4063 csum_len - 1, &ordered_sums, 0);
4067 while (!list_empty(&ordered_sums)) {
4068 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
4069 struct btrfs_ordered_sum,
4072 ret = btrfs_csum_file_blocks(trans, log, sums);
4073 list_del(&sums->list);
4080 static int log_one_extent(struct btrfs_trans_handle *trans,
4081 struct btrfs_inode *inode, struct btrfs_root *root,
4082 const struct extent_map *em,
4083 struct btrfs_path *path,
4084 const struct list_head *logged_list,
4085 struct btrfs_log_ctx *ctx)
4087 struct btrfs_root *log = root->log_root;
4088 struct btrfs_file_extent_item *fi;
4089 struct extent_buffer *leaf;
4090 struct btrfs_map_token token;
4091 struct btrfs_key key;
4092 u64 extent_offset = em->start - em->orig_start;
4095 int extent_inserted = 0;
4096 bool ordered_io_err = false;
4098 ret = wait_ordered_extents(trans, &inode->vfs_inode, root, em,
4099 logged_list, &ordered_io_err);
4103 if (ordered_io_err) {
4108 btrfs_init_map_token(&token);
4110 ret = __btrfs_drop_extents(trans, log, &inode->vfs_inode, path, em->start,
4111 em->start + em->len, NULL, 0, 1,
4112 sizeof(*fi), &extent_inserted);
4116 if (!extent_inserted) {
4117 key.objectid = btrfs_ino(inode);
4118 key.type = BTRFS_EXTENT_DATA_KEY;
4119 key.offset = em->start;
4121 ret = btrfs_insert_empty_item(trans, log, path, &key,
4126 leaf = path->nodes[0];
4127 fi = btrfs_item_ptr(leaf, path->slots[0],
4128 struct btrfs_file_extent_item);
4130 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
4132 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
4133 btrfs_set_token_file_extent_type(leaf, fi,
4134 BTRFS_FILE_EXTENT_PREALLOC,
4137 btrfs_set_token_file_extent_type(leaf, fi,
4138 BTRFS_FILE_EXTENT_REG,
4141 block_len = max(em->block_len, em->orig_block_len);
4142 if (em->compress_type != BTRFS_COMPRESS_NONE) {
4143 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4146 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4148 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
4149 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
4151 extent_offset, &token);
4152 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
4155 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
4156 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
4160 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
4161 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
4162 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
4163 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
4165 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
4166 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
4167 btrfs_mark_buffer_dirty(leaf);
4169 btrfs_release_path(path);
4174 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
4175 struct btrfs_root *root,
4176 struct btrfs_inode *inode,
4177 struct btrfs_path *path,
4178 struct list_head *logged_list,
4179 struct btrfs_log_ctx *ctx,
4183 struct extent_map *em, *n;
4184 struct list_head extents;
4185 struct extent_map_tree *tree = &inode->extent_tree;
4190 INIT_LIST_HEAD(&extents);
4192 down_write(&inode->dio_sem);
4193 write_lock(&tree->lock);
4194 test_gen = root->fs_info->last_trans_committed;
4196 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
4197 list_del_init(&em->list);
4200 * Just an arbitrary number, this can be really CPU intensive
4201 * once we start getting a lot of extents, and really once we
4202 * have a bunch of extents we just want to commit since it will
4205 if (++num > 32768) {
4206 list_del_init(&tree->modified_extents);
4211 if (em->generation <= test_gen)
4213 /* Need a ref to keep it from getting evicted from cache */
4214 refcount_inc(&em->refs);
4215 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
4216 list_add_tail(&em->list, &extents);
4220 list_sort(NULL, &extents, extent_cmp);
4221 btrfs_get_logged_extents(inode, logged_list, start, end);
4223 * Some ordered extents started by fsync might have completed
4224 * before we could collect them into the list logged_list, which
4225 * means they're gone, not in our logged_list nor in the inode's
4226 * ordered tree. We want the application/user space to know an
4227 * error happened while attempting to persist file data so that
4228 * it can take proper action. If such error happened, we leave
4229 * without writing to the log tree and the fsync must report the
4230 * file data write error and not commit the current transaction.
4232 ret = filemap_check_errors(inode->vfs_inode.i_mapping);
4236 while (!list_empty(&extents)) {
4237 em = list_entry(extents.next, struct extent_map, list);
4239 list_del_init(&em->list);
4242 * If we had an error we just need to delete everybody from our
4246 clear_em_logging(tree, em);
4247 free_extent_map(em);
4251 write_unlock(&tree->lock);
4253 ret = log_one_extent(trans, inode, root, em, path, logged_list,
4255 write_lock(&tree->lock);
4256 clear_em_logging(tree, em);
4257 free_extent_map(em);
4259 WARN_ON(!list_empty(&extents));
4260 write_unlock(&tree->lock);
4261 up_write(&inode->dio_sem);
4263 btrfs_release_path(path);
4267 static int logged_inode_size(struct btrfs_root *log, struct btrfs_inode *inode,
4268 struct btrfs_path *path, u64 *size_ret)
4270 struct btrfs_key key;
4273 key.objectid = btrfs_ino(inode);
4274 key.type = BTRFS_INODE_ITEM_KEY;
4277 ret = btrfs_search_slot(NULL, log, &key, path, 0, 0);
4280 } else if (ret > 0) {
4283 struct btrfs_inode_item *item;
4285 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4286 struct btrfs_inode_item);
4287 *size_ret = btrfs_inode_size(path->nodes[0], item);
4290 btrfs_release_path(path);
4295 * At the moment we always log all xattrs. This is to figure out at log replay
4296 * time which xattrs must have their deletion replayed. If a xattr is missing
4297 * in the log tree and exists in the fs/subvol tree, we delete it. This is
4298 * because if a xattr is deleted, the inode is fsynced and a power failure
4299 * happens, causing the log to be replayed the next time the fs is mounted,
4300 * we want the xattr to not exist anymore (same behaviour as other filesystems
4301 * with a journal, ext3/4, xfs, f2fs, etc).
4303 static int btrfs_log_all_xattrs(struct btrfs_trans_handle *trans,
4304 struct btrfs_root *root,
4305 struct btrfs_inode *inode,
4306 struct btrfs_path *path,
4307 struct btrfs_path *dst_path)
4310 struct btrfs_key key;
4311 const u64 ino = btrfs_ino(inode);
4316 key.type = BTRFS_XATTR_ITEM_KEY;
4319 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4324 int slot = path->slots[0];
4325 struct extent_buffer *leaf = path->nodes[0];
4326 int nritems = btrfs_header_nritems(leaf);
4328 if (slot >= nritems) {
4330 u64 last_extent = 0;
4332 ret = copy_items(trans, inode, dst_path, path,
4333 &last_extent, start_slot,
4335 /* can't be 1, extent items aren't processed */
4341 ret = btrfs_next_leaf(root, path);
4349 btrfs_item_key_to_cpu(leaf, &key, slot);
4350 if (key.objectid != ino || key.type != BTRFS_XATTR_ITEM_KEY)
4360 u64 last_extent = 0;
4362 ret = copy_items(trans, inode, dst_path, path,
4363 &last_extent, start_slot,
4365 /* can't be 1, extent items aren't processed */
4375 * If the no holes feature is enabled we need to make sure any hole between the
4376 * last extent and the i_size of our inode is explicitly marked in the log. This
4377 * is to make sure that doing something like:
4379 * 1) create file with 128Kb of data
4380 * 2) truncate file to 64Kb
4381 * 3) truncate file to 256Kb
4383 * 5) <crash/power failure>
4384 * 6) mount fs and trigger log replay
4386 * Will give us a file with a size of 256Kb, the first 64Kb of data match what
4387 * the file had in its first 64Kb of data at step 1 and the last 192Kb of the
4388 * file correspond to a hole. The presence of explicit holes in a log tree is
4389 * what guarantees that log replay will remove/adjust file extent items in the
4392 * Here we do not need to care about holes between extents, that is already done
4393 * by copy_items(). We also only need to do this in the full sync path, where we
4394 * lookup for extents from the fs/subvol tree only. In the fast path case, we
4395 * lookup the list of modified extent maps and if any represents a hole, we
4396 * insert a corresponding extent representing a hole in the log tree.
4398 static int btrfs_log_trailing_hole(struct btrfs_trans_handle *trans,
4399 struct btrfs_root *root,
4400 struct btrfs_inode *inode,
4401 struct btrfs_path *path)
4403 struct btrfs_fs_info *fs_info = root->fs_info;
4405 struct btrfs_key key;
4408 struct extent_buffer *leaf;
4409 struct btrfs_root *log = root->log_root;
4410 const u64 ino = btrfs_ino(inode);
4411 const u64 i_size = i_size_read(&inode->vfs_inode);
4413 if (!btrfs_fs_incompat(fs_info, NO_HOLES))
4417 key.type = BTRFS_EXTENT_DATA_KEY;
4418 key.offset = (u64)-1;
4420 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4425 ASSERT(path->slots[0] > 0);
4427 leaf = path->nodes[0];
4428 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4430 if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) {
4431 /* inode does not have any extents */
4435 struct btrfs_file_extent_item *extent;
4439 * If there's an extent beyond i_size, an explicit hole was
4440 * already inserted by copy_items().
4442 if (key.offset >= i_size)
4445 extent = btrfs_item_ptr(leaf, path->slots[0],
4446 struct btrfs_file_extent_item);
4448 if (btrfs_file_extent_type(leaf, extent) ==
4449 BTRFS_FILE_EXTENT_INLINE) {
4450 len = btrfs_file_extent_inline_len(leaf,
4453 ASSERT(len == i_size);
4457 len = btrfs_file_extent_num_bytes(leaf, extent);
4458 /* Last extent goes beyond i_size, no need to log a hole. */
4459 if (key.offset + len > i_size)
4461 hole_start = key.offset + len;
4462 hole_size = i_size - hole_start;
4464 btrfs_release_path(path);
4466 /* Last extent ends at i_size. */
4470 hole_size = ALIGN(hole_size, fs_info->sectorsize);
4471 ret = btrfs_insert_file_extent(trans, log, ino, hole_start, 0, 0,
4472 hole_size, 0, hole_size, 0, 0, 0);
4477 * When we are logging a new inode X, check if it doesn't have a reference that
4478 * matches the reference from some other inode Y created in a past transaction
4479 * and that was renamed in the current transaction. If we don't do this, then at
4480 * log replay time we can lose inode Y (and all its files if it's a directory):
4483 * echo "hello world" > /mnt/x/foobar
4486 * mkdir /mnt/x # or touch /mnt/x
4487 * xfs_io -c fsync /mnt/x
4489 * mount fs, trigger log replay
4491 * After the log replay procedure, we would lose the first directory and all its
4492 * files (file foobar).
4493 * For the case where inode Y is not a directory we simply end up losing it:
4495 * echo "123" > /mnt/foo
4497 * mv /mnt/foo /mnt/bar
4498 * echo "abc" > /mnt/foo
4499 * xfs_io -c fsync /mnt/foo
4502 * We also need this for cases where a snapshot entry is replaced by some other
4503 * entry (file or directory) otherwise we end up with an unreplayable log due to
4504 * attempts to delete the snapshot entry (entry of type BTRFS_ROOT_ITEM_KEY) as
4505 * if it were a regular entry:
4508 * btrfs subvolume snapshot /mnt /mnt/x/snap
4509 * btrfs subvolume delete /mnt/x/snap
4512 * fsync /mnt/x or fsync some new file inside it
4515 * The snapshot delete, rmdir of x, mkdir of a new x and the fsync all happen in
4516 * the same transaction.
4518 static int btrfs_check_ref_name_override(struct extent_buffer *eb,
4520 const struct btrfs_key *key,
4521 struct btrfs_inode *inode,
4525 struct btrfs_path *search_path;
4528 u32 item_size = btrfs_item_size_nr(eb, slot);
4530 unsigned long ptr = btrfs_item_ptr_offset(eb, slot);
4532 search_path = btrfs_alloc_path();
4535 search_path->search_commit_root = 1;
4536 search_path->skip_locking = 1;
4538 while (cur_offset < item_size) {
4542 unsigned long name_ptr;
4543 struct btrfs_dir_item *di;
4545 if (key->type == BTRFS_INODE_REF_KEY) {
4546 struct btrfs_inode_ref *iref;
4548 iref = (struct btrfs_inode_ref *)(ptr + cur_offset);
4549 parent = key->offset;
4550 this_name_len = btrfs_inode_ref_name_len(eb, iref);
4551 name_ptr = (unsigned long)(iref + 1);
4552 this_len = sizeof(*iref) + this_name_len;
4554 struct btrfs_inode_extref *extref;
4556 extref = (struct btrfs_inode_extref *)(ptr +
4558 parent = btrfs_inode_extref_parent(eb, extref);
4559 this_name_len = btrfs_inode_extref_name_len(eb, extref);
4560 name_ptr = (unsigned long)&extref->name;
4561 this_len = sizeof(*extref) + this_name_len;
4564 ret = btrfs_is_name_len_valid(eb, slot, name_ptr,
4570 if (this_name_len > name_len) {
4573 new_name = krealloc(name, this_name_len, GFP_NOFS);
4578 name_len = this_name_len;
4582 read_extent_buffer(eb, name, name_ptr, this_name_len);
4583 di = btrfs_lookup_dir_item(NULL, inode->root, search_path,
4584 parent, name, this_name_len, 0);
4585 if (di && !IS_ERR(di)) {
4586 struct btrfs_key di_key;
4588 btrfs_dir_item_key_to_cpu(search_path->nodes[0],
4590 if (di_key.type == BTRFS_INODE_ITEM_KEY) {
4592 *other_ino = di_key.objectid;
4597 } else if (IS_ERR(di)) {
4601 btrfs_release_path(search_path);
4603 cur_offset += this_len;
4607 btrfs_free_path(search_path);
4612 /* log a single inode in the tree log.
4613 * At least one parent directory for this inode must exist in the tree
4614 * or be logged already.
4616 * Any items from this inode changed by the current transaction are copied
4617 * to the log tree. An extra reference is taken on any extents in this
4618 * file, allowing us to avoid a whole pile of corner cases around logging
4619 * blocks that have been removed from the tree.
4621 * See LOG_INODE_ALL and related defines for a description of what inode_only
4624 * This handles both files and directories.
4626 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
4627 struct btrfs_root *root, struct btrfs_inode *inode,
4631 struct btrfs_log_ctx *ctx)
4633 struct btrfs_fs_info *fs_info = root->fs_info;
4634 struct btrfs_path *path;
4635 struct btrfs_path *dst_path;
4636 struct btrfs_key min_key;
4637 struct btrfs_key max_key;
4638 struct btrfs_root *log = root->log_root;
4639 struct extent_buffer *src = NULL;
4640 LIST_HEAD(logged_list);
4641 u64 last_extent = 0;
4645 int ins_start_slot = 0;
4647 bool fast_search = false;
4648 u64 ino = btrfs_ino(inode);
4649 struct extent_map_tree *em_tree = &inode->extent_tree;
4650 u64 logged_isize = 0;
4651 bool need_log_inode_item = true;
4653 path = btrfs_alloc_path();
4656 dst_path = btrfs_alloc_path();
4658 btrfs_free_path(path);
4662 min_key.objectid = ino;
4663 min_key.type = BTRFS_INODE_ITEM_KEY;
4666 max_key.objectid = ino;
4669 /* today the code can only do partial logging of directories */
4670 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4671 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4672 &inode->runtime_flags) &&
4673 inode_only >= LOG_INODE_EXISTS))
4674 max_key.type = BTRFS_XATTR_ITEM_KEY;
4676 max_key.type = (u8)-1;
4677 max_key.offset = (u64)-1;
4680 * Only run delayed items if we are a dir or a new file.
4681 * Otherwise commit the delayed inode only, which is needed in
4682 * order for the log replay code to mark inodes for link count
4683 * fixup (create temporary BTRFS_TREE_LOG_FIXUP_OBJECTID items).
4685 if (S_ISDIR(inode->vfs_inode.i_mode) ||
4686 inode->generation > fs_info->last_trans_committed)
4687 ret = btrfs_commit_inode_delayed_items(trans, inode);
4689 ret = btrfs_commit_inode_delayed_inode(inode);
4692 btrfs_free_path(path);
4693 btrfs_free_path(dst_path);
4697 if (inode_only == LOG_OTHER_INODE) {
4698 inode_only = LOG_INODE_EXISTS;
4699 mutex_lock_nested(&inode->log_mutex, SINGLE_DEPTH_NESTING);
4701 mutex_lock(&inode->log_mutex);
4705 * a brute force approach to making sure we get the most uptodate
4706 * copies of everything.
4708 if (S_ISDIR(inode->vfs_inode.i_mode)) {
4709 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
4711 if (inode_only == LOG_INODE_EXISTS)
4712 max_key_type = BTRFS_XATTR_ITEM_KEY;
4713 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
4715 if (inode_only == LOG_INODE_EXISTS) {
4717 * Make sure the new inode item we write to the log has
4718 * the same isize as the current one (if it exists).
4719 * This is necessary to prevent data loss after log
4720 * replay, and also to prevent doing a wrong expanding
4721 * truncate - for e.g. create file, write 4K into offset
4722 * 0, fsync, write 4K into offset 4096, add hard link,
4723 * fsync some other file (to sync log), power fail - if
4724 * we use the inode's current i_size, after log replay
4725 * we get a 8Kb file, with the last 4Kb extent as a hole
4726 * (zeroes), as if an expanding truncate happened,
4727 * instead of getting a file of 4Kb only.
4729 err = logged_inode_size(log, inode, path, &logged_isize);
4733 if (test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4734 &inode->runtime_flags)) {
4735 if (inode_only == LOG_INODE_EXISTS) {
4736 max_key.type = BTRFS_XATTR_ITEM_KEY;
4737 ret = drop_objectid_items(trans, log, path, ino,
4740 clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4741 &inode->runtime_flags);
4742 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4743 &inode->runtime_flags);
4745 ret = btrfs_truncate_inode_items(trans,
4746 log, &inode->vfs_inode, 0, 0);
4751 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4752 &inode->runtime_flags) ||
4753 inode_only == LOG_INODE_EXISTS) {
4754 if (inode_only == LOG_INODE_ALL)
4756 max_key.type = BTRFS_XATTR_ITEM_KEY;
4757 ret = drop_objectid_items(trans, log, path, ino,
4760 if (inode_only == LOG_INODE_ALL)
4773 ret = btrfs_search_forward(root, &min_key,
4774 path, trans->transid);
4782 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4783 if (min_key.objectid != ino)
4785 if (min_key.type > max_key.type)
4788 if (min_key.type == BTRFS_INODE_ITEM_KEY)
4789 need_log_inode_item = false;
4791 if ((min_key.type == BTRFS_INODE_REF_KEY ||
4792 min_key.type == BTRFS_INODE_EXTREF_KEY) &&
4793 inode->generation == trans->transid) {
4796 ret = btrfs_check_ref_name_override(path->nodes[0],
4797 path->slots[0], &min_key, inode,
4802 } else if (ret > 0 && ctx &&
4803 other_ino != btrfs_ino(BTRFS_I(ctx->inode))) {
4804 struct btrfs_key inode_key;
4805 struct inode *other_inode;
4811 ins_start_slot = path->slots[0];
4813 ret = copy_items(trans, inode, dst_path, path,
4814 &last_extent, ins_start_slot,
4822 btrfs_release_path(path);
4823 inode_key.objectid = other_ino;
4824 inode_key.type = BTRFS_INODE_ITEM_KEY;
4825 inode_key.offset = 0;
4826 other_inode = btrfs_iget(fs_info->sb,
4830 * If the other inode that had a conflicting dir
4831 * entry was deleted in the current transaction,
4832 * we don't need to do more work nor fallback to
4833 * a transaction commit.
4835 if (IS_ERR(other_inode) &&
4836 PTR_ERR(other_inode) == -ENOENT) {
4838 } else if (IS_ERR(other_inode)) {
4839 err = PTR_ERR(other_inode);
4843 * We are safe logging the other inode without
4844 * acquiring its i_mutex as long as we log with
4845 * the LOG_INODE_EXISTS mode. We're safe against
4846 * concurrent renames of the other inode as well
4847 * because during a rename we pin the log and
4848 * update the log with the new name before we
4851 err = btrfs_log_inode(trans, root,
4852 BTRFS_I(other_inode),
4853 LOG_OTHER_INODE, 0, LLONG_MAX,
4863 /* Skip xattrs, we log them later with btrfs_log_all_xattrs() */
4864 if (min_key.type == BTRFS_XATTR_ITEM_KEY) {
4867 ret = copy_items(trans, inode, dst_path, path,
4868 &last_extent, ins_start_slot,
4869 ins_nr, inode_only, logged_isize);
4876 btrfs_release_path(path);
4882 src = path->nodes[0];
4883 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4886 } else if (!ins_nr) {
4887 ins_start_slot = path->slots[0];
4892 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4893 ins_start_slot, ins_nr, inode_only,
4901 btrfs_release_path(path);
4905 ins_start_slot = path->slots[0];
4908 nritems = btrfs_header_nritems(path->nodes[0]);
4910 if (path->slots[0] < nritems) {
4911 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4916 ret = copy_items(trans, inode, dst_path, path,
4917 &last_extent, ins_start_slot,
4918 ins_nr, inode_only, logged_isize);
4926 btrfs_release_path(path);
4928 if (min_key.offset < (u64)-1) {
4930 } else if (min_key.type < max_key.type) {
4938 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4939 ins_start_slot, ins_nr, inode_only,
4949 btrfs_release_path(path);
4950 btrfs_release_path(dst_path);
4951 err = btrfs_log_all_xattrs(trans, root, inode, path, dst_path);
4954 if (max_key.type >= BTRFS_EXTENT_DATA_KEY && !fast_search) {
4955 btrfs_release_path(path);
4956 btrfs_release_path(dst_path);
4957 err = btrfs_log_trailing_hole(trans, root, inode, path);
4962 btrfs_release_path(path);
4963 btrfs_release_path(dst_path);
4964 if (need_log_inode_item) {
4965 err = log_inode_item(trans, log, dst_path, inode);
4970 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4971 &logged_list, ctx, start, end);
4976 } else if (inode_only == LOG_INODE_ALL) {
4977 struct extent_map *em, *n;
4979 write_lock(&em_tree->lock);
4981 * We can't just remove every em if we're called for a ranged
4982 * fsync - that is, one that doesn't cover the whole possible
4983 * file range (0 to LLONG_MAX). This is because we can have
4984 * em's that fall outside the range we're logging and therefore
4985 * their ordered operations haven't completed yet
4986 * (btrfs_finish_ordered_io() not invoked yet). This means we
4987 * didn't get their respective file extent item in the fs/subvol
4988 * tree yet, and need to let the next fast fsync (one which
4989 * consults the list of modified extent maps) find the em so
4990 * that it logs a matching file extent item and waits for the
4991 * respective ordered operation to complete (if it's still
4994 * Removing every em outside the range we're logging would make
4995 * the next fast fsync not log their matching file extent items,
4996 * therefore making us lose data after a log replay.
4998 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
5000 const u64 mod_end = em->mod_start + em->mod_len - 1;
5002 if (em->mod_start >= start && mod_end <= end)
5003 list_del_init(&em->list);
5005 write_unlock(&em_tree->lock);
5008 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->vfs_inode.i_mode)) {
5009 ret = log_directory_changes(trans, root, inode, path, dst_path,
5017 spin_lock(&inode->lock);
5018 inode->logged_trans = trans->transid;
5019 inode->last_log_commit = inode->last_sub_trans;
5020 spin_unlock(&inode->lock);
5023 btrfs_put_logged_extents(&logged_list);
5025 btrfs_submit_logged_extents(&logged_list, log);
5026 mutex_unlock(&inode->log_mutex);
5028 btrfs_free_path(path);
5029 btrfs_free_path(dst_path);
5034 * Check if we must fallback to a transaction commit when logging an inode.
5035 * This must be called after logging the inode and is used only in the context
5036 * when fsyncing an inode requires the need to log some other inode - in which
5037 * case we can't lock the i_mutex of each other inode we need to log as that
5038 * can lead to deadlocks with concurrent fsync against other inodes (as we can
5039 * log inodes up or down in the hierarchy) or rename operations for example. So
5040 * we take the log_mutex of the inode after we have logged it and then check for
5041 * its last_unlink_trans value - this is safe because any task setting
5042 * last_unlink_trans must take the log_mutex and it must do this before it does
5043 * the actual unlink operation, so if we do this check before a concurrent task
5044 * sets last_unlink_trans it means we've logged a consistent version/state of
5045 * all the inode items, otherwise we are not sure and must do a transaction
5046 * commit (the concurrent task might have only updated last_unlink_trans before
5047 * we logged the inode or it might have also done the unlink).
5049 static bool btrfs_must_commit_transaction(struct btrfs_trans_handle *trans,
5050 struct btrfs_inode *inode)
5052 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5055 mutex_lock(&inode->log_mutex);
5056 if (inode->last_unlink_trans > fs_info->last_trans_committed) {
5058 * Make sure any commits to the log are forced to be full
5061 btrfs_set_log_full_commit(fs_info, trans);
5064 mutex_unlock(&inode->log_mutex);
5070 * follow the dentry parent pointers up the chain and see if any
5071 * of the directories in it require a full commit before they can
5072 * be logged. Returns zero if nothing special needs to be done or 1 if
5073 * a full commit is required.
5075 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
5076 struct btrfs_inode *inode,
5077 struct dentry *parent,
5078 struct super_block *sb,
5082 struct dentry *old_parent = NULL;
5083 struct btrfs_inode *orig_inode = inode;
5086 * for regular files, if its inode is already on disk, we don't
5087 * have to worry about the parents at all. This is because
5088 * we can use the last_unlink_trans field to record renames
5089 * and other fun in this file.
5091 if (S_ISREG(inode->vfs_inode.i_mode) &&
5092 inode->generation <= last_committed &&
5093 inode->last_unlink_trans <= last_committed)
5096 if (!S_ISDIR(inode->vfs_inode.i_mode)) {
5097 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5099 inode = BTRFS_I(d_inode(parent));
5104 * If we are logging a directory then we start with our inode,
5105 * not our parent's inode, so we need to skip setting the
5106 * logged_trans so that further down in the log code we don't
5107 * think this inode has already been logged.
5109 if (inode != orig_inode)
5110 inode->logged_trans = trans->transid;
5113 if (btrfs_must_commit_transaction(trans, inode)) {
5118 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5121 if (IS_ROOT(parent)) {
5122 inode = BTRFS_I(d_inode(parent));
5123 if (btrfs_must_commit_transaction(trans, inode))
5128 parent = dget_parent(parent);
5130 old_parent = parent;
5131 inode = BTRFS_I(d_inode(parent));
5139 struct btrfs_dir_list {
5141 struct list_head list;
5145 * Log the inodes of the new dentries of a directory. See log_dir_items() for
5146 * details about the why it is needed.
5147 * This is a recursive operation - if an existing dentry corresponds to a
5148 * directory, that directory's new entries are logged too (same behaviour as
5149 * ext3/4, xfs, f2fs, reiserfs, nilfs2). Note that when logging the inodes
5150 * the dentries point to we do not lock their i_mutex, otherwise lockdep
5151 * complains about the following circular lock dependency / possible deadlock:
5155 * lock(&type->i_mutex_dir_key#3/2);
5156 * lock(sb_internal#2);
5157 * lock(&type->i_mutex_dir_key#3/2);
5158 * lock(&sb->s_type->i_mutex_key#14);
5160 * Where sb_internal is the lock (a counter that works as a lock) acquired by
5161 * sb_start_intwrite() in btrfs_start_transaction().
5162 * Not locking i_mutex of the inodes is still safe because:
5164 * 1) For regular files we log with a mode of LOG_INODE_EXISTS. It's possible
5165 * that while logging the inode new references (names) are added or removed
5166 * from the inode, leaving the logged inode item with a link count that does
5167 * not match the number of logged inode reference items. This is fine because
5168 * at log replay time we compute the real number of links and correct the
5169 * link count in the inode item (see replay_one_buffer() and
5170 * link_to_fixup_dir());
5172 * 2) For directories we log with a mode of LOG_INODE_ALL. It's possible that
5173 * while logging the inode's items new items with keys BTRFS_DIR_ITEM_KEY and
5174 * BTRFS_DIR_INDEX_KEY are added to fs/subvol tree and the logged inode item
5175 * has a size that doesn't match the sum of the lengths of all the logged
5176 * names. This does not result in a problem because if a dir_item key is
5177 * logged but its matching dir_index key is not logged, at log replay time we
5178 * don't use it to replay the respective name (see replay_one_name()). On the
5179 * other hand if only the dir_index key ends up being logged, the respective
5180 * name is added to the fs/subvol tree with both the dir_item and dir_index
5181 * keys created (see replay_one_name()).
5182 * The directory's inode item with a wrong i_size is not a problem as well,
5183 * since we don't use it at log replay time to set the i_size in the inode
5184 * item of the fs/subvol tree (see overwrite_item()).
5186 static int log_new_dir_dentries(struct btrfs_trans_handle *trans,
5187 struct btrfs_root *root,
5188 struct btrfs_inode *start_inode,
5189 struct btrfs_log_ctx *ctx)
5191 struct btrfs_fs_info *fs_info = root->fs_info;
5192 struct btrfs_root *log = root->log_root;
5193 struct btrfs_path *path;
5194 LIST_HEAD(dir_list);
5195 struct btrfs_dir_list *dir_elem;
5198 path = btrfs_alloc_path();
5202 dir_elem = kmalloc(sizeof(*dir_elem), GFP_NOFS);
5204 btrfs_free_path(path);
5207 dir_elem->ino = btrfs_ino(start_inode);
5208 list_add_tail(&dir_elem->list, &dir_list);
5210 while (!list_empty(&dir_list)) {
5211 struct extent_buffer *leaf;
5212 struct btrfs_key min_key;
5216 dir_elem = list_first_entry(&dir_list, struct btrfs_dir_list,
5219 goto next_dir_inode;
5221 min_key.objectid = dir_elem->ino;
5222 min_key.type = BTRFS_DIR_ITEM_KEY;
5225 btrfs_release_path(path);
5226 ret = btrfs_search_forward(log, &min_key, path, trans->transid);
5228 goto next_dir_inode;
5229 } else if (ret > 0) {
5231 goto next_dir_inode;
5235 leaf = path->nodes[0];
5236 nritems = btrfs_header_nritems(leaf);
5237 for (i = path->slots[0]; i < nritems; i++) {
5238 struct btrfs_dir_item *di;
5239 struct btrfs_key di_key;
5240 struct inode *di_inode;
5241 struct btrfs_dir_list *new_dir_elem;
5242 int log_mode = LOG_INODE_EXISTS;
5245 btrfs_item_key_to_cpu(leaf, &min_key, i);
5246 if (min_key.objectid != dir_elem->ino ||
5247 min_key.type != BTRFS_DIR_ITEM_KEY)
5248 goto next_dir_inode;
5250 di = btrfs_item_ptr(leaf, i, struct btrfs_dir_item);
5251 type = btrfs_dir_type(leaf, di);
5252 if (btrfs_dir_transid(leaf, di) < trans->transid &&
5253 type != BTRFS_FT_DIR)
5255 btrfs_dir_item_key_to_cpu(leaf, di, &di_key);
5256 if (di_key.type == BTRFS_ROOT_ITEM_KEY)
5259 btrfs_release_path(path);
5260 di_inode = btrfs_iget(fs_info->sb, &di_key, root, NULL);
5261 if (IS_ERR(di_inode)) {
5262 ret = PTR_ERR(di_inode);
5263 goto next_dir_inode;
5266 if (btrfs_inode_in_log(BTRFS_I(di_inode), trans->transid)) {
5271 ctx->log_new_dentries = false;
5272 if (type == BTRFS_FT_DIR || type == BTRFS_FT_SYMLINK)
5273 log_mode = LOG_INODE_ALL;
5274 ret = btrfs_log_inode(trans, root, BTRFS_I(di_inode),
5275 log_mode, 0, LLONG_MAX, ctx);
5277 btrfs_must_commit_transaction(trans, BTRFS_I(di_inode)))
5281 goto next_dir_inode;
5282 if (ctx->log_new_dentries) {
5283 new_dir_elem = kmalloc(sizeof(*new_dir_elem),
5285 if (!new_dir_elem) {
5287 goto next_dir_inode;
5289 new_dir_elem->ino = di_key.objectid;
5290 list_add_tail(&new_dir_elem->list, &dir_list);
5295 ret = btrfs_next_leaf(log, path);
5297 goto next_dir_inode;
5298 } else if (ret > 0) {
5300 goto next_dir_inode;
5304 if (min_key.offset < (u64)-1) {
5309 list_del(&dir_elem->list);
5313 btrfs_free_path(path);
5317 static int btrfs_log_all_parents(struct btrfs_trans_handle *trans,
5318 struct btrfs_inode *inode,
5319 struct btrfs_log_ctx *ctx)
5321 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5323 struct btrfs_path *path;
5324 struct btrfs_key key;
5325 struct btrfs_root *root = inode->root;
5326 const u64 ino = btrfs_ino(inode);
5328 path = btrfs_alloc_path();
5331 path->skip_locking = 1;
5332 path->search_commit_root = 1;
5335 key.type = BTRFS_INODE_REF_KEY;
5337 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5342 struct extent_buffer *leaf = path->nodes[0];
5343 int slot = path->slots[0];
5348 if (slot >= btrfs_header_nritems(leaf)) {
5349 ret = btrfs_next_leaf(root, path);
5357 btrfs_item_key_to_cpu(leaf, &key, slot);
5358 /* BTRFS_INODE_EXTREF_KEY is BTRFS_INODE_REF_KEY + 1 */
5359 if (key.objectid != ino || key.type > BTRFS_INODE_EXTREF_KEY)
5362 item_size = btrfs_item_size_nr(leaf, slot);
5363 ptr = btrfs_item_ptr_offset(leaf, slot);
5364 while (cur_offset < item_size) {
5365 struct btrfs_key inode_key;
5366 struct inode *dir_inode;
5368 inode_key.type = BTRFS_INODE_ITEM_KEY;
5369 inode_key.offset = 0;
5371 if (key.type == BTRFS_INODE_EXTREF_KEY) {
5372 struct btrfs_inode_extref *extref;
5374 extref = (struct btrfs_inode_extref *)
5376 inode_key.objectid = btrfs_inode_extref_parent(
5378 cur_offset += sizeof(*extref);
5379 cur_offset += btrfs_inode_extref_name_len(leaf,
5382 inode_key.objectid = key.offset;
5383 cur_offset = item_size;
5386 dir_inode = btrfs_iget(fs_info->sb, &inode_key,
5388 /* If parent inode was deleted, skip it. */
5389 if (IS_ERR(dir_inode))
5393 ctx->log_new_dentries = false;
5394 ret = btrfs_log_inode(trans, root, BTRFS_I(dir_inode),
5395 LOG_INODE_ALL, 0, LLONG_MAX, ctx);
5397 btrfs_must_commit_transaction(trans, BTRFS_I(dir_inode)))
5399 if (!ret && ctx && ctx->log_new_dentries)
5400 ret = log_new_dir_dentries(trans, root,
5401 BTRFS_I(dir_inode), ctx);
5410 btrfs_free_path(path);
5415 * helper function around btrfs_log_inode to make sure newly created
5416 * parent directories also end up in the log. A minimal inode and backref
5417 * only logging is done of any parent directories that are older than
5418 * the last committed transaction
5420 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
5421 struct btrfs_root *root,
5422 struct btrfs_inode *inode,
5423 struct dentry *parent,
5427 struct btrfs_log_ctx *ctx)
5429 struct btrfs_fs_info *fs_info = root->fs_info;
5430 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
5431 struct super_block *sb;
5432 struct dentry *old_parent = NULL;
5434 u64 last_committed = fs_info->last_trans_committed;
5435 bool log_dentries = false;
5436 struct btrfs_inode *orig_inode = inode;
5438 sb = inode->vfs_inode.i_sb;
5440 if (btrfs_test_opt(fs_info, NOTREELOG)) {
5446 * The prev transaction commit doesn't complete, we need do
5447 * full commit by ourselves.
5449 if (fs_info->last_trans_log_full_commit >
5450 fs_info->last_trans_committed) {
5455 if (root != inode->root || btrfs_root_refs(&root->root_item) == 0) {
5460 ret = check_parent_dirs_for_sync(trans, inode, parent, sb,
5465 if (btrfs_inode_in_log(inode, trans->transid)) {
5466 ret = BTRFS_NO_LOG_SYNC;
5470 ret = start_log_trans(trans, root, ctx);
5474 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
5479 * for regular files, if its inode is already on disk, we don't
5480 * have to worry about the parents at all. This is because
5481 * we can use the last_unlink_trans field to record renames
5482 * and other fun in this file.
5484 if (S_ISREG(inode->vfs_inode.i_mode) &&
5485 inode->generation <= last_committed &&
5486 inode->last_unlink_trans <= last_committed) {
5491 if (S_ISDIR(inode->vfs_inode.i_mode) && ctx && ctx->log_new_dentries)
5492 log_dentries = true;
5495 * On unlink we must make sure all our current and old parent directory
5496 * inodes are fully logged. This is to prevent leaving dangling
5497 * directory index entries in directories that were our parents but are
5498 * not anymore. Not doing this results in old parent directory being
5499 * impossible to delete after log replay (rmdir will always fail with
5500 * error -ENOTEMPTY).
5506 * ln testdir/foo testdir/bar
5508 * unlink testdir/bar
5509 * xfs_io -c fsync testdir/foo
5511 * mount fs, triggers log replay
5513 * If we don't log the parent directory (testdir), after log replay the
5514 * directory still has an entry pointing to the file inode using the bar
5515 * name, but a matching BTRFS_INODE_[REF|EXTREF]_KEY does not exist and
5516 * the file inode has a link count of 1.
5522 * ln foo testdir/foo2
5523 * ln foo testdir/foo3
5525 * unlink testdir/foo3
5526 * xfs_io -c fsync foo
5528 * mount fs, triggers log replay
5530 * Similar as the first example, after log replay the parent directory
5531 * testdir still has an entry pointing to the inode file with name foo3
5532 * but the file inode does not have a matching BTRFS_INODE_REF_KEY item
5533 * and has a link count of 2.
5535 if (inode->last_unlink_trans > last_committed) {
5536 ret = btrfs_log_all_parents(trans, orig_inode, ctx);
5542 if (!parent || d_really_is_negative(parent) || sb != parent->d_sb)
5545 inode = BTRFS_I(d_inode(parent));
5546 if (root != inode->root)
5549 if (inode->generation > last_committed) {
5550 ret = btrfs_log_inode(trans, root, inode,
5551 LOG_INODE_EXISTS, 0, LLONG_MAX, ctx);
5555 if (IS_ROOT(parent))
5558 parent = dget_parent(parent);
5560 old_parent = parent;
5563 ret = log_new_dir_dentries(trans, root, orig_inode, ctx);
5569 btrfs_set_log_full_commit(fs_info, trans);
5574 btrfs_remove_log_ctx(root, ctx);
5575 btrfs_end_log_trans(root);
5581 * it is not safe to log dentry if the chunk root has added new
5582 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
5583 * If this returns 1, you must commit the transaction to safely get your
5586 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
5587 struct btrfs_root *root, struct dentry *dentry,
5590 struct btrfs_log_ctx *ctx)
5592 struct dentry *parent = dget_parent(dentry);
5595 ret = btrfs_log_inode_parent(trans, root, BTRFS_I(d_inode(dentry)),
5596 parent, start, end, 0, ctx);
5603 * should be called during mount to recover any replay any log trees
5606 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
5609 struct btrfs_path *path;
5610 struct btrfs_trans_handle *trans;
5611 struct btrfs_key key;
5612 struct btrfs_key found_key;
5613 struct btrfs_key tmp_key;
5614 struct btrfs_root *log;
5615 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
5616 struct walk_control wc = {
5617 .process_func = process_one_buffer,
5621 path = btrfs_alloc_path();
5625 set_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5627 trans = btrfs_start_transaction(fs_info->tree_root, 0);
5628 if (IS_ERR(trans)) {
5629 ret = PTR_ERR(trans);
5636 ret = walk_log_tree(trans, log_root_tree, &wc);
5638 btrfs_handle_fs_error(fs_info, ret,
5639 "Failed to pin buffers while recovering log root tree.");
5644 key.objectid = BTRFS_TREE_LOG_OBJECTID;
5645 key.offset = (u64)-1;
5646 key.type = BTRFS_ROOT_ITEM_KEY;
5649 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
5652 btrfs_handle_fs_error(fs_info, ret,
5653 "Couldn't find tree log root.");
5657 if (path->slots[0] == 0)
5661 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
5663 btrfs_release_path(path);
5664 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
5667 log = btrfs_read_fs_root(log_root_tree, &found_key);
5670 btrfs_handle_fs_error(fs_info, ret,
5671 "Couldn't read tree log root.");
5675 tmp_key.objectid = found_key.offset;
5676 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
5677 tmp_key.offset = (u64)-1;
5679 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
5680 if (IS_ERR(wc.replay_dest)) {
5681 ret = PTR_ERR(wc.replay_dest);
5682 free_extent_buffer(log->node);
5683 free_extent_buffer(log->commit_root);
5685 btrfs_handle_fs_error(fs_info, ret,
5686 "Couldn't read target root for tree log recovery.");
5690 wc.replay_dest->log_root = log;
5691 btrfs_record_root_in_trans(trans, wc.replay_dest);
5692 ret = walk_log_tree(trans, log, &wc);
5694 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
5695 ret = fixup_inode_link_counts(trans, wc.replay_dest,
5699 key.offset = found_key.offset - 1;
5700 wc.replay_dest->log_root = NULL;
5701 free_extent_buffer(log->node);
5702 free_extent_buffer(log->commit_root);
5708 if (found_key.offset == 0)
5711 btrfs_release_path(path);
5713 /* step one is to pin it all, step two is to replay just inodes */
5716 wc.process_func = replay_one_buffer;
5717 wc.stage = LOG_WALK_REPLAY_INODES;
5720 /* step three is to replay everything */
5721 if (wc.stage < LOG_WALK_REPLAY_ALL) {
5726 btrfs_free_path(path);
5728 /* step 4: commit the transaction, which also unpins the blocks */
5729 ret = btrfs_commit_transaction(trans);
5733 free_extent_buffer(log_root_tree->node);
5734 log_root_tree->log_root = NULL;
5735 clear_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags);
5736 kfree(log_root_tree);
5741 btrfs_end_transaction(wc.trans);
5742 btrfs_free_path(path);
5747 * there are some corner cases where we want to force a full
5748 * commit instead of allowing a directory to be logged.
5750 * They revolve around files there were unlinked from the directory, and
5751 * this function updates the parent directory so that a full commit is
5752 * properly done if it is fsync'd later after the unlinks are done.
5754 * Must be called before the unlink operations (updates to the subvolume tree,
5755 * inodes, etc) are done.
5757 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
5758 struct btrfs_inode *dir, struct btrfs_inode *inode,
5762 * when we're logging a file, if it hasn't been renamed
5763 * or unlinked, and its inode is fully committed on disk,
5764 * we don't have to worry about walking up the directory chain
5765 * to log its parents.
5767 * So, we use the last_unlink_trans field to put this transid
5768 * into the file. When the file is logged we check it and
5769 * don't log the parents if the file is fully on disk.
5771 mutex_lock(&inode->log_mutex);
5772 inode->last_unlink_trans = trans->transid;
5773 mutex_unlock(&inode->log_mutex);
5776 * if this directory was already logged any new
5777 * names for this file/dir will get recorded
5780 if (dir->logged_trans == trans->transid)
5784 * if the inode we're about to unlink was logged,
5785 * the log will be properly updated for any new names
5787 if (inode->logged_trans == trans->transid)
5791 * when renaming files across directories, if the directory
5792 * there we're unlinking from gets fsync'd later on, there's
5793 * no way to find the destination directory later and fsync it
5794 * properly. So, we have to be conservative and force commits
5795 * so the new name gets discovered.
5800 /* we can safely do the unlink without any special recording */
5804 mutex_lock(&dir->log_mutex);
5805 dir->last_unlink_trans = trans->transid;
5806 mutex_unlock(&dir->log_mutex);
5810 * Make sure that if someone attempts to fsync the parent directory of a deleted
5811 * snapshot, it ends up triggering a transaction commit. This is to guarantee
5812 * that after replaying the log tree of the parent directory's root we will not
5813 * see the snapshot anymore and at log replay time we will not see any log tree
5814 * corresponding to the deleted snapshot's root, which could lead to replaying
5815 * it after replaying the log tree of the parent directory (which would replay
5816 * the snapshot delete operation).
5818 * Must be called before the actual snapshot destroy operation (updates to the
5819 * parent root and tree of tree roots trees, etc) are done.
5821 void btrfs_record_snapshot_destroy(struct btrfs_trans_handle *trans,
5822 struct btrfs_inode *dir)
5824 mutex_lock(&dir->log_mutex);
5825 dir->last_unlink_trans = trans->transid;
5826 mutex_unlock(&dir->log_mutex);
5830 * Call this after adding a new name for a file and it will properly
5831 * update the log to reflect the new name.
5833 * It will return zero if all goes well, and it will return 1 if a
5834 * full transaction commit is required.
5836 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
5837 struct btrfs_inode *inode, struct btrfs_inode *old_dir,
5838 struct dentry *parent)
5840 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5841 struct btrfs_root *root = inode->root;
5844 * this will force the logging code to walk the dentry chain
5847 if (S_ISREG(inode->vfs_inode.i_mode))
5848 inode->last_unlink_trans = trans->transid;
5851 * if this inode hasn't been logged and directory we're renaming it
5852 * from hasn't been logged, we don't need to log it
5854 if (inode->logged_trans <= fs_info->last_trans_committed &&
5855 (!old_dir || old_dir->logged_trans <= fs_info->last_trans_committed))
5858 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
5859 LLONG_MAX, 1, NULL);