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>
22 #include "transaction.h"
25 #include "print-tree.h"
29 /* magic values for the inode_only field in btrfs_log_inode:
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
35 #define LOG_INODE_ALL 0
36 #define LOG_INODE_EXISTS 1
39 * directory trouble cases
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
48 * rename foo/some_dir foo2/some_dir
50 * fsync foo/some_dir/some_file
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2). After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
63 * 2a is actually the more important variant. With the extra logging
64 * a crash might unlink the old name without recreating the new one
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir. After a crash the rm -rf must
76 * be replayed. This must be able to recurse down the entire
77 * directory tree. The inode link count fixup code takes care of the
82 * stages for the tree walking. The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
90 #define LOG_WALK_PIN_ONLY 0
91 #define LOG_WALK_REPLAY_INODES 1
92 #define LOG_WALK_REPLAY_ALL 2
94 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root, struct inode *inode,
97 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root,
99 struct btrfs_path *path, u64 objectid);
100 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101 struct btrfs_root *root,
102 struct btrfs_root *log,
103 struct btrfs_path *path,
104 u64 dirid, int del_all);
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction. Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree. Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
134 static int start_log_trans(struct btrfs_trans_handle *trans,
135 struct btrfs_root *root)
140 mutex_lock(&root->log_mutex);
141 if (root->log_root) {
142 if (!root->log_start_pid) {
143 root->log_start_pid = current->pid;
144 root->log_multiple_pids = false;
145 } else if (root->log_start_pid != current->pid) {
146 root->log_multiple_pids = true;
150 atomic_inc(&root->log_writers);
151 mutex_unlock(&root->log_mutex);
154 root->log_multiple_pids = false;
155 root->log_start_pid = current->pid;
156 mutex_lock(&root->fs_info->tree_log_mutex);
157 if (!root->fs_info->log_root_tree) {
158 ret = btrfs_init_log_root_tree(trans, root->fs_info);
162 if (err == 0 && !root->log_root) {
163 ret = btrfs_add_log_tree(trans, root);
167 mutex_unlock(&root->fs_info->tree_log_mutex);
169 atomic_inc(&root->log_writers);
170 mutex_unlock(&root->log_mutex);
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
179 static int join_running_log_trans(struct btrfs_root *root)
187 mutex_lock(&root->log_mutex);
188 if (root->log_root) {
190 atomic_inc(&root->log_writers);
192 mutex_unlock(&root->log_mutex);
197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
201 int btrfs_pin_log_trans(struct btrfs_root *root)
205 mutex_lock(&root->log_mutex);
206 atomic_inc(&root->log_writers);
207 mutex_unlock(&root->log_mutex);
212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
215 int btrfs_end_log_trans(struct btrfs_root *root)
217 if (atomic_dec_and_test(&root->log_writers)) {
219 if (waitqueue_active(&root->log_writer_wait))
220 wake_up(&root->log_writer_wait);
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree. The stage field tells us which part
229 * of the log tree processing we are currently doing. The others
230 * are state fields used for that specific part
232 struct walk_control {
233 /* should we free the extent on disk when done? This is used
234 * at transaction commit time while freeing a log tree
238 /* should we write out the extent buffer? This is used
239 * while flushing the log tree to disk during a sync
243 /* should we wait for the extent buffer io to finish? Also used
244 * while flushing the log tree to disk for a sync
248 /* pin only walk, we record which extents on disk belong to the
253 /* what stage of the replay code we're currently in */
256 /* the root we are currently replaying */
257 struct btrfs_root *replay_dest;
259 /* the trans handle for the current replay */
260 struct btrfs_trans_handle *trans;
262 /* the function that gets used to process blocks we find in the
263 * tree. Note the extent_buffer might not be up to date when it is
264 * passed in, and it must be checked or read if you need the data
267 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
268 struct walk_control *wc, u64 gen);
272 * process_func used to pin down extents, write them or wait on them
274 static int process_one_buffer(struct btrfs_root *log,
275 struct extent_buffer *eb,
276 struct walk_control *wc, u64 gen)
279 btrfs_pin_extent(log->fs_info->extent_root,
280 eb->start, eb->len, 0);
282 if (btrfs_buffer_uptodate(eb, gen)) {
284 btrfs_write_tree_block(eb);
286 btrfs_wait_tree_block_writeback(eb);
292 * Item overwrite used by replay and tree logging. eb, slot and key all refer
293 * to the src data we are copying out.
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
299 * If the key is already in the destination tree the existing item is
300 * overwritten. If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
303 * If the key isn't in the destination yet, a new item is inserted.
305 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306 struct btrfs_root *root,
307 struct btrfs_path *path,
308 struct extent_buffer *eb, int slot,
309 struct btrfs_key *key)
313 u64 saved_i_size = 0;
314 int save_old_i_size = 0;
315 unsigned long src_ptr;
316 unsigned long dst_ptr;
317 int overwrite_root = 0;
319 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
322 item_size = btrfs_item_size_nr(eb, slot);
323 src_ptr = btrfs_item_ptr_offset(eb, slot);
325 /* look for the key in the destination tree */
326 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
330 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
332 if (dst_size != item_size)
335 if (item_size == 0) {
336 btrfs_release_path(root, path);
339 dst_copy = kmalloc(item_size, GFP_NOFS);
340 src_copy = kmalloc(item_size, GFP_NOFS);
341 if (!dst_copy || !src_copy) {
342 btrfs_release_path(root, path);
348 read_extent_buffer(eb, src_copy, src_ptr, item_size);
350 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
351 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
353 ret = memcmp(dst_copy, src_copy, item_size);
358 * they have the same contents, just return, this saves
359 * us from cowing blocks in the destination tree and doing
360 * extra writes that may not have been done by a previous
364 btrfs_release_path(root, path);
370 btrfs_release_path(root, path);
371 /* try to insert the key into the destination tree */
372 ret = btrfs_insert_empty_item(trans, root, path,
375 /* make sure any existing item is the correct size */
376 if (ret == -EEXIST) {
378 found_size = btrfs_item_size_nr(path->nodes[0],
380 if (found_size > item_size) {
381 btrfs_truncate_item(trans, root, path, item_size, 1);
382 } else if (found_size < item_size) {
383 ret = btrfs_extend_item(trans, root, path,
384 item_size - found_size);
390 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
393 /* don't overwrite an existing inode if the generation number
394 * was logged as zero. This is done when the tree logging code
395 * is just logging an inode to make sure it exists after recovery.
397 * Also, don't overwrite i_size on directories during replay.
398 * log replay inserts and removes directory items based on the
399 * state of the tree found in the subvolume, and i_size is modified
402 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
403 struct btrfs_inode_item *src_item;
404 struct btrfs_inode_item *dst_item;
406 src_item = (struct btrfs_inode_item *)src_ptr;
407 dst_item = (struct btrfs_inode_item *)dst_ptr;
409 if (btrfs_inode_generation(eb, src_item) == 0)
412 if (overwrite_root &&
413 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
414 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
416 saved_i_size = btrfs_inode_size(path->nodes[0],
421 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
424 if (save_old_i_size) {
425 struct btrfs_inode_item *dst_item;
426 dst_item = (struct btrfs_inode_item *)dst_ptr;
427 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
430 /* make sure the generation is filled in */
431 if (key->type == BTRFS_INODE_ITEM_KEY) {
432 struct btrfs_inode_item *dst_item;
433 dst_item = (struct btrfs_inode_item *)dst_ptr;
434 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
435 btrfs_set_inode_generation(path->nodes[0], dst_item,
440 btrfs_mark_buffer_dirty(path->nodes[0]);
441 btrfs_release_path(root, path);
446 * simple helper to read an inode off the disk from a given root
447 * This can only be called for subvolume roots and not for the log
449 static noinline struct inode *read_one_inode(struct btrfs_root *root,
452 struct btrfs_key key;
455 key.objectid = objectid;
456 key.type = BTRFS_INODE_ITEM_KEY;
458 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
461 } else if (is_bad_inode(inode)) {
468 /* replays a single extent in 'eb' at 'slot' with 'key' into the
469 * subvolume 'root'. path is released on entry and should be released
472 * extents in the log tree have not been allocated out of the extent
473 * tree yet. So, this completes the allocation, taking a reference
474 * as required if the extent already exists or creating a new extent
475 * if it isn't in the extent allocation tree yet.
477 * The extent is inserted into the file, dropping any existing extents
478 * from the file that overlap the new one.
480 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
481 struct btrfs_root *root,
482 struct btrfs_path *path,
483 struct extent_buffer *eb, int slot,
484 struct btrfs_key *key)
487 u64 mask = root->sectorsize - 1;
490 u64 start = key->offset;
492 struct btrfs_file_extent_item *item;
493 struct inode *inode = NULL;
497 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
498 found_type = btrfs_file_extent_type(eb, item);
500 if (found_type == BTRFS_FILE_EXTENT_REG ||
501 found_type == BTRFS_FILE_EXTENT_PREALLOC)
502 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
503 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
504 size = btrfs_file_extent_inline_len(eb, item);
505 extent_end = (start + size + mask) & ~mask;
511 inode = read_one_inode(root, key->objectid);
518 * first check to see if we already have this extent in the
519 * file. This must be done before the btrfs_drop_extents run
520 * so we don't try to drop this extent.
522 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
526 (found_type == BTRFS_FILE_EXTENT_REG ||
527 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
528 struct btrfs_file_extent_item cmp1;
529 struct btrfs_file_extent_item cmp2;
530 struct btrfs_file_extent_item *existing;
531 struct extent_buffer *leaf;
533 leaf = path->nodes[0];
534 existing = btrfs_item_ptr(leaf, path->slots[0],
535 struct btrfs_file_extent_item);
537 read_extent_buffer(eb, &cmp1, (unsigned long)item,
539 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
543 * we already have a pointer to this exact extent,
544 * we don't have to do anything
546 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
547 btrfs_release_path(root, path);
551 btrfs_release_path(root, path);
553 saved_nbytes = inode_get_bytes(inode);
554 /* drop any overlapping extents */
555 ret = btrfs_drop_extents(trans, inode, start, extent_end,
559 if (found_type == BTRFS_FILE_EXTENT_REG ||
560 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
562 unsigned long dest_offset;
563 struct btrfs_key ins;
565 ret = btrfs_insert_empty_item(trans, root, path, key,
568 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
570 copy_extent_buffer(path->nodes[0], eb, dest_offset,
571 (unsigned long)item, sizeof(*item));
573 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
574 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
575 ins.type = BTRFS_EXTENT_ITEM_KEY;
576 offset = key->offset - btrfs_file_extent_offset(eb, item);
578 if (ins.objectid > 0) {
581 LIST_HEAD(ordered_sums);
583 * is this extent already allocated in the extent
584 * allocation tree? If so, just add a reference
586 ret = btrfs_lookup_extent(root, ins.objectid,
589 ret = btrfs_inc_extent_ref(trans, root,
590 ins.objectid, ins.offset,
591 0, root->root_key.objectid,
592 key->objectid, offset);
595 * insert the extent pointer in the extent
598 ret = btrfs_alloc_logged_file_extent(trans,
599 root, root->root_key.objectid,
600 key->objectid, offset, &ins);
603 btrfs_release_path(root, path);
605 if (btrfs_file_extent_compression(eb, item)) {
606 csum_start = ins.objectid;
607 csum_end = csum_start + ins.offset;
609 csum_start = ins.objectid +
610 btrfs_file_extent_offset(eb, item);
611 csum_end = csum_start +
612 btrfs_file_extent_num_bytes(eb, item);
615 ret = btrfs_lookup_csums_range(root->log_root,
616 csum_start, csum_end - 1,
619 while (!list_empty(&ordered_sums)) {
620 struct btrfs_ordered_sum *sums;
621 sums = list_entry(ordered_sums.next,
622 struct btrfs_ordered_sum,
624 ret = btrfs_csum_file_blocks(trans,
625 root->fs_info->csum_root,
628 list_del(&sums->list);
632 btrfs_release_path(root, path);
634 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
635 /* inline extents are easy, we just overwrite them */
636 ret = overwrite_item(trans, root, path, eb, slot, key);
640 inode_set_bytes(inode, saved_nbytes);
641 btrfs_update_inode(trans, root, inode);
649 * when cleaning up conflicts between the directory names in the
650 * subvolume, directory names in the log and directory names in the
651 * inode back references, we may have to unlink inodes from directories.
653 * This is a helper function to do the unlink of a specific directory
656 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
657 struct btrfs_root *root,
658 struct btrfs_path *path,
660 struct btrfs_dir_item *di)
665 struct extent_buffer *leaf;
666 struct btrfs_key location;
669 leaf = path->nodes[0];
671 btrfs_dir_item_key_to_cpu(leaf, di, &location);
672 name_len = btrfs_dir_name_len(leaf, di);
673 name = kmalloc(name_len, GFP_NOFS);
677 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
678 btrfs_release_path(root, path);
680 inode = read_one_inode(root, location.objectid);
683 ret = link_to_fixup_dir(trans, root, path, location.objectid);
686 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
695 * helper function to see if a given name and sequence number found
696 * in an inode back reference are already in a directory and correctly
697 * point to this inode
699 static noinline int inode_in_dir(struct btrfs_root *root,
700 struct btrfs_path *path,
701 u64 dirid, u64 objectid, u64 index,
702 const char *name, int name_len)
704 struct btrfs_dir_item *di;
705 struct btrfs_key location;
708 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
709 index, name, name_len, 0);
710 if (di && !IS_ERR(di)) {
711 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
712 if (location.objectid != objectid)
716 btrfs_release_path(root, path);
718 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
719 if (di && !IS_ERR(di)) {
720 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
721 if (location.objectid != objectid)
727 btrfs_release_path(root, path);
732 * helper function to check a log tree for a named back reference in
733 * an inode. This is used to decide if a back reference that is
734 * found in the subvolume conflicts with what we find in the log.
736 * inode backreferences may have multiple refs in a single item,
737 * during replay we process one reference at a time, and we don't
738 * want to delete valid links to a file from the subvolume if that
739 * link is also in the log.
741 static noinline int backref_in_log(struct btrfs_root *log,
742 struct btrfs_key *key,
743 char *name, int namelen)
745 struct btrfs_path *path;
746 struct btrfs_inode_ref *ref;
748 unsigned long ptr_end;
749 unsigned long name_ptr;
755 path = btrfs_alloc_path();
759 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
763 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
764 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
765 ptr_end = ptr + item_size;
766 while (ptr < ptr_end) {
767 ref = (struct btrfs_inode_ref *)ptr;
768 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
769 if (found_name_len == namelen) {
770 name_ptr = (unsigned long)(ref + 1);
771 ret = memcmp_extent_buffer(path->nodes[0], name,
778 ptr = (unsigned long)(ref + 1) + found_name_len;
781 btrfs_free_path(path);
787 * replay one inode back reference item found in the log tree.
788 * eb, slot and key refer to the buffer and key found in the log tree.
789 * root is the destination we are replaying into, and path is for temp
790 * use by this function. (it should be released on return).
792 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
793 struct btrfs_root *root,
794 struct btrfs_root *log,
795 struct btrfs_path *path,
796 struct extent_buffer *eb, int slot,
797 struct btrfs_key *key)
801 struct btrfs_inode_ref *ref;
802 struct btrfs_dir_item *di;
806 unsigned long ref_ptr;
807 unsigned long ref_end;
810 * it is possible that we didn't log all the parent directories
811 * for a given inode. If we don't find the dir, just don't
812 * copy the back ref in. The link count fixup code will take
815 dir = read_one_inode(root, key->offset);
819 inode = read_one_inode(root, key->objectid);
822 ref_ptr = btrfs_item_ptr_offset(eb, slot);
823 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
826 ref = (struct btrfs_inode_ref *)ref_ptr;
828 namelen = btrfs_inode_ref_name_len(eb, ref);
829 name = kmalloc(namelen, GFP_NOFS);
832 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
834 /* if we already have a perfect match, we're done */
835 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
836 btrfs_inode_ref_index(eb, ref),
842 * look for a conflicting back reference in the metadata.
843 * if we find one we have to unlink that name of the file
844 * before we add our new link. Later on, we overwrite any
845 * existing back reference, and we don't want to create
846 * dangling pointers in the directory.
849 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
853 struct btrfs_inode_ref *victim_ref;
855 unsigned long ptr_end;
856 struct extent_buffer *leaf = path->nodes[0];
858 /* are we trying to overwrite a back ref for the root directory
859 * if so, just jump out, we're done
861 if (key->objectid == key->offset)
864 /* check all the names in this back reference to see
865 * if they are in the log. if so, we allow them to stay
866 * otherwise they must be unlinked as a conflict
868 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
869 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
870 while (ptr < ptr_end) {
871 victim_ref = (struct btrfs_inode_ref *)ptr;
872 victim_name_len = btrfs_inode_ref_name_len(leaf,
874 victim_name = kmalloc(victim_name_len, GFP_NOFS);
875 BUG_ON(!victim_name);
877 read_extent_buffer(leaf, victim_name,
878 (unsigned long)(victim_ref + 1),
881 if (!backref_in_log(log, key, victim_name,
883 btrfs_inc_nlink(inode);
884 btrfs_release_path(root, path);
886 ret = btrfs_unlink_inode(trans, root, dir,
890 btrfs_release_path(root, path);
894 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
898 btrfs_release_path(root, path);
900 /* look for a conflicting sequence number */
901 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
902 btrfs_inode_ref_index(eb, ref),
904 if (di && !IS_ERR(di)) {
905 ret = drop_one_dir_item(trans, root, path, dir, di);
908 btrfs_release_path(root, path);
911 /* look for a conflicting name */
912 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
914 if (di && !IS_ERR(di)) {
915 ret = drop_one_dir_item(trans, root, path, dir, di);
918 btrfs_release_path(root, path);
920 /* insert our name */
921 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
922 btrfs_inode_ref_index(eb, ref));
925 btrfs_update_inode(trans, root, inode);
928 ref_ptr = (unsigned long)(ref + 1) + namelen;
930 if (ref_ptr < ref_end)
933 /* finally write the back reference in the inode */
934 ret = overwrite_item(trans, root, path, eb, slot, key);
938 btrfs_release_path(root, path);
944 static int insert_orphan_item(struct btrfs_trans_handle *trans,
945 struct btrfs_root *root, u64 offset)
948 ret = btrfs_find_orphan_item(root, offset);
950 ret = btrfs_insert_orphan_item(trans, root, offset);
956 * There are a few corners where the link count of the file can't
957 * be properly maintained during replay. So, instead of adding
958 * lots of complexity to the log code, we just scan the backrefs
959 * for any file that has been through replay.
961 * The scan will update the link count on the inode to reflect the
962 * number of back refs found. If it goes down to zero, the iput
963 * will free the inode.
965 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
966 struct btrfs_root *root,
969 struct btrfs_path *path;
971 struct btrfs_key key;
974 unsigned long ptr_end;
977 key.objectid = inode->i_ino;
978 key.type = BTRFS_INODE_REF_KEY;
979 key.offset = (u64)-1;
981 path = btrfs_alloc_path();
986 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
990 if (path->slots[0] == 0)
994 btrfs_item_key_to_cpu(path->nodes[0], &key,
996 if (key.objectid != inode->i_ino ||
997 key.type != BTRFS_INODE_REF_KEY)
999 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1000 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1002 while (ptr < ptr_end) {
1003 struct btrfs_inode_ref *ref;
1005 ref = (struct btrfs_inode_ref *)ptr;
1006 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1008 ptr = (unsigned long)(ref + 1) + name_len;
1012 if (key.offset == 0)
1015 btrfs_release_path(root, path);
1017 btrfs_release_path(root, path);
1018 if (nlink != inode->i_nlink) {
1019 inode->i_nlink = nlink;
1020 btrfs_update_inode(trans, root, inode);
1022 BTRFS_I(inode)->index_cnt = (u64)-1;
1024 if (inode->i_nlink == 0) {
1025 if (S_ISDIR(inode->i_mode)) {
1026 ret = replay_dir_deletes(trans, root, NULL, path,
1030 ret = insert_orphan_item(trans, root, inode->i_ino);
1033 btrfs_free_path(path);
1038 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1039 struct btrfs_root *root,
1040 struct btrfs_path *path)
1043 struct btrfs_key key;
1044 struct inode *inode;
1046 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1047 key.type = BTRFS_ORPHAN_ITEM_KEY;
1048 key.offset = (u64)-1;
1050 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1055 if (path->slots[0] == 0)
1060 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1061 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1062 key.type != BTRFS_ORPHAN_ITEM_KEY)
1065 ret = btrfs_del_item(trans, root, path);
1068 btrfs_release_path(root, path);
1069 inode = read_one_inode(root, key.offset);
1072 ret = fixup_inode_link_count(trans, root, inode);
1078 * fixup on a directory may create new entries,
1079 * make sure we always look for the highset possible
1082 key.offset = (u64)-1;
1084 btrfs_release_path(root, path);
1090 * record a given inode in the fixup dir so we can check its link
1091 * count when replay is done. The link count is incremented here
1092 * so the inode won't go away until we check it
1094 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1095 struct btrfs_root *root,
1096 struct btrfs_path *path,
1099 struct btrfs_key key;
1101 struct inode *inode;
1103 inode = read_one_inode(root, objectid);
1106 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1107 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1108 key.offset = objectid;
1110 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1112 btrfs_release_path(root, path);
1114 btrfs_inc_nlink(inode);
1115 btrfs_update_inode(trans, root, inode);
1116 } else if (ret == -EEXIST) {
1127 * when replaying the log for a directory, we only insert names
1128 * for inodes that actually exist. This means an fsync on a directory
1129 * does not implicitly fsync all the new files in it
1131 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1132 struct btrfs_root *root,
1133 struct btrfs_path *path,
1134 u64 dirid, u64 index,
1135 char *name, int name_len, u8 type,
1136 struct btrfs_key *location)
1138 struct inode *inode;
1142 inode = read_one_inode(root, location->objectid);
1146 dir = read_one_inode(root, dirid);
1151 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1153 /* FIXME, put inode into FIXUP list */
1161 * take a single entry in a log directory item and replay it into
1164 * if a conflicting item exists in the subdirectory already,
1165 * the inode it points to is unlinked and put into the link count
1168 * If a name from the log points to a file or directory that does
1169 * not exist in the FS, it is skipped. fsyncs on directories
1170 * do not force down inodes inside that directory, just changes to the
1171 * names or unlinks in a directory.
1173 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1174 struct btrfs_root *root,
1175 struct btrfs_path *path,
1176 struct extent_buffer *eb,
1177 struct btrfs_dir_item *di,
1178 struct btrfs_key *key)
1182 struct btrfs_dir_item *dst_di;
1183 struct btrfs_key found_key;
1184 struct btrfs_key log_key;
1190 dir = read_one_inode(root, key->objectid);
1193 name_len = btrfs_dir_name_len(eb, di);
1194 name = kmalloc(name_len, GFP_NOFS);
1198 log_type = btrfs_dir_type(eb, di);
1199 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1202 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1203 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1208 btrfs_release_path(root, path);
1210 if (key->type == BTRFS_DIR_ITEM_KEY) {
1211 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1213 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1214 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1221 if (!dst_di || IS_ERR(dst_di)) {
1222 /* we need a sequence number to insert, so we only
1223 * do inserts for the BTRFS_DIR_INDEX_KEY types
1225 if (key->type != BTRFS_DIR_INDEX_KEY)
1230 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1231 /* the existing item matches the logged item */
1232 if (found_key.objectid == log_key.objectid &&
1233 found_key.type == log_key.type &&
1234 found_key.offset == log_key.offset &&
1235 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1240 * don't drop the conflicting directory entry if the inode
1241 * for the new entry doesn't exist
1246 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1249 if (key->type == BTRFS_DIR_INDEX_KEY)
1252 btrfs_release_path(root, path);
1258 btrfs_release_path(root, path);
1259 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1260 name, name_len, log_type, &log_key);
1262 BUG_ON(ret && ret != -ENOENT);
1267 * find all the names in a directory item and reconcile them into
1268 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1269 * one name in a directory item, but the same code gets used for
1270 * both directory index types
1272 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1273 struct btrfs_root *root,
1274 struct btrfs_path *path,
1275 struct extent_buffer *eb, int slot,
1276 struct btrfs_key *key)
1279 u32 item_size = btrfs_item_size_nr(eb, slot);
1280 struct btrfs_dir_item *di;
1283 unsigned long ptr_end;
1285 ptr = btrfs_item_ptr_offset(eb, slot);
1286 ptr_end = ptr + item_size;
1287 while (ptr < ptr_end) {
1288 di = (struct btrfs_dir_item *)ptr;
1289 if (verify_dir_item(root, eb, di))
1291 name_len = btrfs_dir_name_len(eb, di);
1292 ret = replay_one_name(trans, root, path, eb, di, key);
1294 ptr = (unsigned long)(di + 1);
1301 * directory replay has two parts. There are the standard directory
1302 * items in the log copied from the subvolume, and range items
1303 * created in the log while the subvolume was logged.
1305 * The range items tell us which parts of the key space the log
1306 * is authoritative for. During replay, if a key in the subvolume
1307 * directory is in a logged range item, but not actually in the log
1308 * that means it was deleted from the directory before the fsync
1309 * and should be removed.
1311 static noinline int find_dir_range(struct btrfs_root *root,
1312 struct btrfs_path *path,
1313 u64 dirid, int key_type,
1314 u64 *start_ret, u64 *end_ret)
1316 struct btrfs_key key;
1318 struct btrfs_dir_log_item *item;
1322 if (*start_ret == (u64)-1)
1325 key.objectid = dirid;
1326 key.type = key_type;
1327 key.offset = *start_ret;
1329 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1333 if (path->slots[0] == 0)
1338 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1340 if (key.type != key_type || key.objectid != dirid) {
1344 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1345 struct btrfs_dir_log_item);
1346 found_end = btrfs_dir_log_end(path->nodes[0], item);
1348 if (*start_ret >= key.offset && *start_ret <= found_end) {
1350 *start_ret = key.offset;
1351 *end_ret = found_end;
1356 /* check the next slot in the tree to see if it is a valid item */
1357 nritems = btrfs_header_nritems(path->nodes[0]);
1358 if (path->slots[0] >= nritems) {
1359 ret = btrfs_next_leaf(root, path);
1366 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1368 if (key.type != key_type || key.objectid != dirid) {
1372 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1373 struct btrfs_dir_log_item);
1374 found_end = btrfs_dir_log_end(path->nodes[0], item);
1375 *start_ret = key.offset;
1376 *end_ret = found_end;
1379 btrfs_release_path(root, path);
1384 * this looks for a given directory item in the log. If the directory
1385 * item is not in the log, the item is removed and the inode it points
1388 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1389 struct btrfs_root *root,
1390 struct btrfs_root *log,
1391 struct btrfs_path *path,
1392 struct btrfs_path *log_path,
1394 struct btrfs_key *dir_key)
1397 struct extent_buffer *eb;
1400 struct btrfs_dir_item *di;
1401 struct btrfs_dir_item *log_di;
1404 unsigned long ptr_end;
1406 struct inode *inode;
1407 struct btrfs_key location;
1410 eb = path->nodes[0];
1411 slot = path->slots[0];
1412 item_size = btrfs_item_size_nr(eb, slot);
1413 ptr = btrfs_item_ptr_offset(eb, slot);
1414 ptr_end = ptr + item_size;
1415 while (ptr < ptr_end) {
1416 di = (struct btrfs_dir_item *)ptr;
1417 if (verify_dir_item(root, eb, di)) {
1422 name_len = btrfs_dir_name_len(eb, di);
1423 name = kmalloc(name_len, GFP_NOFS);
1428 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1431 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1432 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1435 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1436 log_di = btrfs_lookup_dir_index_item(trans, log,
1442 if (!log_di || IS_ERR(log_di)) {
1443 btrfs_dir_item_key_to_cpu(eb, di, &location);
1444 btrfs_release_path(root, path);
1445 btrfs_release_path(log, log_path);
1446 inode = read_one_inode(root, location.objectid);
1449 ret = link_to_fixup_dir(trans, root,
1450 path, location.objectid);
1452 btrfs_inc_nlink(inode);
1453 ret = btrfs_unlink_inode(trans, root, dir, inode,
1459 /* there might still be more names under this key
1460 * check and repeat if required
1462 ret = btrfs_search_slot(NULL, root, dir_key, path,
1469 btrfs_release_path(log, log_path);
1472 ptr = (unsigned long)(di + 1);
1477 btrfs_release_path(root, path);
1478 btrfs_release_path(log, log_path);
1483 * deletion replay happens before we copy any new directory items
1484 * out of the log or out of backreferences from inodes. It
1485 * scans the log to find ranges of keys that log is authoritative for,
1486 * and then scans the directory to find items in those ranges that are
1487 * not present in the log.
1489 * Anything we don't find in the log is unlinked and removed from the
1492 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1493 struct btrfs_root *root,
1494 struct btrfs_root *log,
1495 struct btrfs_path *path,
1496 u64 dirid, int del_all)
1500 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1502 struct btrfs_key dir_key;
1503 struct btrfs_key found_key;
1504 struct btrfs_path *log_path;
1507 dir_key.objectid = dirid;
1508 dir_key.type = BTRFS_DIR_ITEM_KEY;
1509 log_path = btrfs_alloc_path();
1513 dir = read_one_inode(root, dirid);
1514 /* it isn't an error if the inode isn't there, that can happen
1515 * because we replay the deletes before we copy in the inode item
1519 btrfs_free_path(log_path);
1527 range_end = (u64)-1;
1529 ret = find_dir_range(log, path, dirid, key_type,
1530 &range_start, &range_end);
1535 dir_key.offset = range_start;
1538 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1543 nritems = btrfs_header_nritems(path->nodes[0]);
1544 if (path->slots[0] >= nritems) {
1545 ret = btrfs_next_leaf(root, path);
1549 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1551 if (found_key.objectid != dirid ||
1552 found_key.type != dir_key.type)
1555 if (found_key.offset > range_end)
1558 ret = check_item_in_log(trans, root, log, path,
1562 if (found_key.offset == (u64)-1)
1564 dir_key.offset = found_key.offset + 1;
1566 btrfs_release_path(root, path);
1567 if (range_end == (u64)-1)
1569 range_start = range_end + 1;
1574 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1575 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1576 dir_key.type = BTRFS_DIR_INDEX_KEY;
1577 btrfs_release_path(root, path);
1581 btrfs_release_path(root, path);
1582 btrfs_free_path(log_path);
1588 * the process_func used to replay items from the log tree. This
1589 * gets called in two different stages. The first stage just looks
1590 * for inodes and makes sure they are all copied into the subvolume.
1592 * The second stage copies all the other item types from the log into
1593 * the subvolume. The two stage approach is slower, but gets rid of
1594 * lots of complexity around inodes referencing other inodes that exist
1595 * only in the log (references come from either directory items or inode
1598 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1599 struct walk_control *wc, u64 gen)
1602 struct btrfs_path *path;
1603 struct btrfs_root *root = wc->replay_dest;
1604 struct btrfs_key key;
1609 btrfs_read_buffer(eb, gen);
1611 level = btrfs_header_level(eb);
1616 path = btrfs_alloc_path();
1619 nritems = btrfs_header_nritems(eb);
1620 for (i = 0; i < nritems; i++) {
1621 btrfs_item_key_to_cpu(eb, &key, i);
1623 /* inode keys are done during the first stage */
1624 if (key.type == BTRFS_INODE_ITEM_KEY &&
1625 wc->stage == LOG_WALK_REPLAY_INODES) {
1626 struct btrfs_inode_item *inode_item;
1629 inode_item = btrfs_item_ptr(eb, i,
1630 struct btrfs_inode_item);
1631 mode = btrfs_inode_mode(eb, inode_item);
1632 if (S_ISDIR(mode)) {
1633 ret = replay_dir_deletes(wc->trans,
1634 root, log, path, key.objectid, 0);
1637 ret = overwrite_item(wc->trans, root, path,
1641 /* for regular files, make sure corresponding
1642 * orhpan item exist. extents past the new EOF
1643 * will be truncated later by orphan cleanup.
1645 if (S_ISREG(mode)) {
1646 ret = insert_orphan_item(wc->trans, root,
1651 ret = link_to_fixup_dir(wc->trans, root,
1652 path, key.objectid);
1655 if (wc->stage < LOG_WALK_REPLAY_ALL)
1658 /* these keys are simply copied */
1659 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1660 ret = overwrite_item(wc->trans, root, path,
1663 } else if (key.type == BTRFS_INODE_REF_KEY) {
1664 ret = add_inode_ref(wc->trans, root, log, path,
1666 BUG_ON(ret && ret != -ENOENT);
1667 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1668 ret = replay_one_extent(wc->trans, root, path,
1671 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1672 key.type == BTRFS_DIR_INDEX_KEY) {
1673 ret = replay_one_dir_item(wc->trans, root, path,
1678 btrfs_free_path(path);
1682 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1683 struct btrfs_root *root,
1684 struct btrfs_path *path, int *level,
1685 struct walk_control *wc)
1690 struct extent_buffer *next;
1691 struct extent_buffer *cur;
1692 struct extent_buffer *parent;
1696 WARN_ON(*level < 0);
1697 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1699 while (*level > 0) {
1700 WARN_ON(*level < 0);
1701 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1702 cur = path->nodes[*level];
1704 if (btrfs_header_level(cur) != *level)
1707 if (path->slots[*level] >=
1708 btrfs_header_nritems(cur))
1711 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1712 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1713 blocksize = btrfs_level_size(root, *level - 1);
1715 parent = path->nodes[*level];
1716 root_owner = btrfs_header_owner(parent);
1718 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1723 wc->process_func(root, next, wc, ptr_gen);
1725 path->slots[*level]++;
1727 btrfs_read_buffer(next, ptr_gen);
1729 btrfs_tree_lock(next);
1730 clean_tree_block(trans, root, next);
1731 btrfs_set_lock_blocking(next);
1732 btrfs_wait_tree_block_writeback(next);
1733 btrfs_tree_unlock(next);
1735 WARN_ON(root_owner !=
1736 BTRFS_TREE_LOG_OBJECTID);
1737 ret = btrfs_free_reserved_extent(root,
1741 free_extent_buffer(next);
1744 btrfs_read_buffer(next, ptr_gen);
1746 WARN_ON(*level <= 0);
1747 if (path->nodes[*level-1])
1748 free_extent_buffer(path->nodes[*level-1]);
1749 path->nodes[*level-1] = next;
1750 *level = btrfs_header_level(next);
1751 path->slots[*level] = 0;
1754 WARN_ON(*level < 0);
1755 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1757 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1763 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1764 struct btrfs_root *root,
1765 struct btrfs_path *path, int *level,
1766 struct walk_control *wc)
1773 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1774 slot = path->slots[i];
1775 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1778 WARN_ON(*level == 0);
1781 struct extent_buffer *parent;
1782 if (path->nodes[*level] == root->node)
1783 parent = path->nodes[*level];
1785 parent = path->nodes[*level + 1];
1787 root_owner = btrfs_header_owner(parent);
1788 wc->process_func(root, path->nodes[*level], wc,
1789 btrfs_header_generation(path->nodes[*level]));
1791 struct extent_buffer *next;
1793 next = path->nodes[*level];
1795 btrfs_tree_lock(next);
1796 clean_tree_block(trans, root, next);
1797 btrfs_set_lock_blocking(next);
1798 btrfs_wait_tree_block_writeback(next);
1799 btrfs_tree_unlock(next);
1801 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1802 ret = btrfs_free_reserved_extent(root,
1803 path->nodes[*level]->start,
1804 path->nodes[*level]->len);
1807 free_extent_buffer(path->nodes[*level]);
1808 path->nodes[*level] = NULL;
1816 * drop the reference count on the tree rooted at 'snap'. This traverses
1817 * the tree freeing any blocks that have a ref count of zero after being
1820 static int walk_log_tree(struct btrfs_trans_handle *trans,
1821 struct btrfs_root *log, struct walk_control *wc)
1826 struct btrfs_path *path;
1830 path = btrfs_alloc_path();
1834 level = btrfs_header_level(log->node);
1836 path->nodes[level] = log->node;
1837 extent_buffer_get(log->node);
1838 path->slots[level] = 0;
1841 wret = walk_down_log_tree(trans, log, path, &level, wc);
1847 wret = walk_up_log_tree(trans, log, path, &level, wc);
1854 /* was the root node processed? if not, catch it here */
1855 if (path->nodes[orig_level]) {
1856 wc->process_func(log, path->nodes[orig_level], wc,
1857 btrfs_header_generation(path->nodes[orig_level]));
1859 struct extent_buffer *next;
1861 next = path->nodes[orig_level];
1863 btrfs_tree_lock(next);
1864 clean_tree_block(trans, log, next);
1865 btrfs_set_lock_blocking(next);
1866 btrfs_wait_tree_block_writeback(next);
1867 btrfs_tree_unlock(next);
1869 WARN_ON(log->root_key.objectid !=
1870 BTRFS_TREE_LOG_OBJECTID);
1871 ret = btrfs_free_reserved_extent(log, next->start,
1877 for (i = 0; i <= orig_level; i++) {
1878 if (path->nodes[i]) {
1879 free_extent_buffer(path->nodes[i]);
1880 path->nodes[i] = NULL;
1883 btrfs_free_path(path);
1888 * helper function to update the item for a given subvolumes log root
1889 * in the tree of log roots
1891 static int update_log_root(struct btrfs_trans_handle *trans,
1892 struct btrfs_root *log)
1896 if (log->log_transid == 1) {
1897 /* insert root item on the first sync */
1898 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1899 &log->root_key, &log->root_item);
1901 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1902 &log->root_key, &log->root_item);
1907 static int wait_log_commit(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root, unsigned long transid)
1911 int index = transid % 2;
1914 * we only allow two pending log transactions at a time,
1915 * so we know that if ours is more than 2 older than the
1916 * current transaction, we're done
1919 prepare_to_wait(&root->log_commit_wait[index],
1920 &wait, TASK_UNINTERRUPTIBLE);
1921 mutex_unlock(&root->log_mutex);
1923 if (root->fs_info->last_trans_log_full_commit !=
1924 trans->transid && root->log_transid < transid + 2 &&
1925 atomic_read(&root->log_commit[index]))
1928 finish_wait(&root->log_commit_wait[index], &wait);
1929 mutex_lock(&root->log_mutex);
1930 } while (root->log_transid < transid + 2 &&
1931 atomic_read(&root->log_commit[index]));
1935 static int wait_for_writer(struct btrfs_trans_handle *trans,
1936 struct btrfs_root *root)
1939 while (atomic_read(&root->log_writers)) {
1940 prepare_to_wait(&root->log_writer_wait,
1941 &wait, TASK_UNINTERRUPTIBLE);
1942 mutex_unlock(&root->log_mutex);
1943 if (root->fs_info->last_trans_log_full_commit !=
1944 trans->transid && atomic_read(&root->log_writers))
1946 mutex_lock(&root->log_mutex);
1947 finish_wait(&root->log_writer_wait, &wait);
1953 * btrfs_sync_log does sends a given tree log down to the disk and
1954 * updates the super blocks to record it. When this call is done,
1955 * you know that any inodes previously logged are safely on disk only
1958 * Any other return value means you need to call btrfs_commit_transaction.
1959 * Some of the edge cases for fsyncing directories that have had unlinks
1960 * or renames done in the past mean that sometimes the only safe
1961 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1962 * that has happened.
1964 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1965 struct btrfs_root *root)
1971 struct btrfs_root *log = root->log_root;
1972 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
1973 unsigned long log_transid = 0;
1975 mutex_lock(&root->log_mutex);
1976 index1 = root->log_transid % 2;
1977 if (atomic_read(&root->log_commit[index1])) {
1978 wait_log_commit(trans, root, root->log_transid);
1979 mutex_unlock(&root->log_mutex);
1982 atomic_set(&root->log_commit[index1], 1);
1984 /* wait for previous tree log sync to complete */
1985 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1986 wait_log_commit(trans, root, root->log_transid - 1);
1989 unsigned long batch = root->log_batch;
1990 if (root->log_multiple_pids) {
1991 mutex_unlock(&root->log_mutex);
1992 schedule_timeout_uninterruptible(1);
1993 mutex_lock(&root->log_mutex);
1995 wait_for_writer(trans, root);
1996 if (batch == root->log_batch)
2000 /* bail out if we need to do a full commit */
2001 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2003 mutex_unlock(&root->log_mutex);
2007 log_transid = root->log_transid;
2008 if (log_transid % 2 == 0)
2009 mark = EXTENT_DIRTY;
2013 /* we start IO on all the marked extents here, but we don't actually
2014 * wait for them until later.
2016 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2019 btrfs_set_root_node(&log->root_item, log->node);
2021 root->log_batch = 0;
2022 root->log_transid++;
2023 log->log_transid = root->log_transid;
2024 root->log_start_pid = 0;
2027 * IO has been started, blocks of the log tree have WRITTEN flag set
2028 * in their headers. new modifications of the log will be written to
2029 * new positions. so it's safe to allow log writers to go in.
2031 mutex_unlock(&root->log_mutex);
2033 mutex_lock(&log_root_tree->log_mutex);
2034 log_root_tree->log_batch++;
2035 atomic_inc(&log_root_tree->log_writers);
2036 mutex_unlock(&log_root_tree->log_mutex);
2038 ret = update_log_root(trans, log);
2040 mutex_lock(&log_root_tree->log_mutex);
2041 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2043 if (waitqueue_active(&log_root_tree->log_writer_wait))
2044 wake_up(&log_root_tree->log_writer_wait);
2048 BUG_ON(ret != -ENOSPC);
2049 root->fs_info->last_trans_log_full_commit = trans->transid;
2050 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2051 mutex_unlock(&log_root_tree->log_mutex);
2056 index2 = log_root_tree->log_transid % 2;
2057 if (atomic_read(&log_root_tree->log_commit[index2])) {
2058 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2059 wait_log_commit(trans, log_root_tree,
2060 log_root_tree->log_transid);
2061 mutex_unlock(&log_root_tree->log_mutex);
2065 atomic_set(&log_root_tree->log_commit[index2], 1);
2067 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2068 wait_log_commit(trans, log_root_tree,
2069 log_root_tree->log_transid - 1);
2072 wait_for_writer(trans, log_root_tree);
2075 * now that we've moved on to the tree of log tree roots,
2076 * check the full commit flag again
2078 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2079 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2080 mutex_unlock(&log_root_tree->log_mutex);
2082 goto out_wake_log_root;
2085 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2086 &log_root_tree->dirty_log_pages,
2087 EXTENT_DIRTY | EXTENT_NEW);
2089 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2091 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2092 log_root_tree->node->start);
2093 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2094 btrfs_header_level(log_root_tree->node));
2096 log_root_tree->log_batch = 0;
2097 log_root_tree->log_transid++;
2100 mutex_unlock(&log_root_tree->log_mutex);
2103 * nobody else is going to jump in and write the the ctree
2104 * super here because the log_commit atomic below is protecting
2105 * us. We must be called with a transaction handle pinning
2106 * the running transaction open, so a full commit can't hop
2107 * in and cause problems either.
2109 write_ctree_super(trans, root->fs_info->tree_root, 1);
2112 mutex_lock(&root->log_mutex);
2113 if (root->last_log_commit < log_transid)
2114 root->last_log_commit = log_transid;
2115 mutex_unlock(&root->log_mutex);
2118 atomic_set(&log_root_tree->log_commit[index2], 0);
2120 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2121 wake_up(&log_root_tree->log_commit_wait[index2]);
2123 atomic_set(&root->log_commit[index1], 0);
2125 if (waitqueue_active(&root->log_commit_wait[index1]))
2126 wake_up(&root->log_commit_wait[index1]);
2130 static void free_log_tree(struct btrfs_trans_handle *trans,
2131 struct btrfs_root *log)
2136 struct walk_control wc = {
2138 .process_func = process_one_buffer
2141 ret = walk_log_tree(trans, log, &wc);
2145 ret = find_first_extent_bit(&log->dirty_log_pages,
2146 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2150 clear_extent_bits(&log->dirty_log_pages, start, end,
2151 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2154 free_extent_buffer(log->node);
2159 * free all the extents used by the tree log. This should be called
2160 * at commit time of the full transaction
2162 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2164 if (root->log_root) {
2165 free_log_tree(trans, root->log_root);
2166 root->log_root = NULL;
2171 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2172 struct btrfs_fs_info *fs_info)
2174 if (fs_info->log_root_tree) {
2175 free_log_tree(trans, fs_info->log_root_tree);
2176 fs_info->log_root_tree = NULL;
2182 * If both a file and directory are logged, and unlinks or renames are
2183 * mixed in, we have a few interesting corners:
2185 * create file X in dir Y
2186 * link file X to X.link in dir Y
2188 * unlink file X but leave X.link
2191 * After a crash we would expect only X.link to exist. But file X
2192 * didn't get fsync'd again so the log has back refs for X and X.link.
2194 * We solve this by removing directory entries and inode backrefs from the
2195 * log when a file that was logged in the current transaction is
2196 * unlinked. Any later fsync will include the updated log entries, and
2197 * we'll be able to reconstruct the proper directory items from backrefs.
2199 * This optimizations allows us to avoid relogging the entire inode
2200 * or the entire directory.
2202 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2203 struct btrfs_root *root,
2204 const char *name, int name_len,
2205 struct inode *dir, u64 index)
2207 struct btrfs_root *log;
2208 struct btrfs_dir_item *di;
2209 struct btrfs_path *path;
2214 if (BTRFS_I(dir)->logged_trans < trans->transid)
2217 ret = join_running_log_trans(root);
2221 mutex_lock(&BTRFS_I(dir)->log_mutex);
2223 log = root->log_root;
2224 path = btrfs_alloc_path();
2228 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2229 name, name_len, -1);
2235 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2236 bytes_del += name_len;
2239 btrfs_release_path(log, path);
2240 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2241 index, name, name_len, -1);
2247 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2248 bytes_del += name_len;
2252 /* update the directory size in the log to reflect the names
2256 struct btrfs_key key;
2258 key.objectid = dir->i_ino;
2260 key.type = BTRFS_INODE_ITEM_KEY;
2261 btrfs_release_path(log, path);
2263 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2269 struct btrfs_inode_item *item;
2272 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2273 struct btrfs_inode_item);
2274 i_size = btrfs_inode_size(path->nodes[0], item);
2275 if (i_size > bytes_del)
2276 i_size -= bytes_del;
2279 btrfs_set_inode_size(path->nodes[0], item, i_size);
2280 btrfs_mark_buffer_dirty(path->nodes[0]);
2283 btrfs_release_path(log, path);
2286 btrfs_free_path(path);
2287 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2288 if (ret == -ENOSPC) {
2289 root->fs_info->last_trans_log_full_commit = trans->transid;
2292 btrfs_end_log_trans(root);
2297 /* see comments for btrfs_del_dir_entries_in_log */
2298 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 const char *name, int name_len,
2301 struct inode *inode, u64 dirid)
2303 struct btrfs_root *log;
2307 if (BTRFS_I(inode)->logged_trans < trans->transid)
2310 ret = join_running_log_trans(root);
2313 log = root->log_root;
2314 mutex_lock(&BTRFS_I(inode)->log_mutex);
2316 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2318 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2319 if (ret == -ENOSPC) {
2320 root->fs_info->last_trans_log_full_commit = trans->transid;
2323 btrfs_end_log_trans(root);
2329 * creates a range item in the log for 'dirid'. first_offset and
2330 * last_offset tell us which parts of the key space the log should
2331 * be considered authoritative for.
2333 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *log,
2335 struct btrfs_path *path,
2336 int key_type, u64 dirid,
2337 u64 first_offset, u64 last_offset)
2340 struct btrfs_key key;
2341 struct btrfs_dir_log_item *item;
2343 key.objectid = dirid;
2344 key.offset = first_offset;
2345 if (key_type == BTRFS_DIR_ITEM_KEY)
2346 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2348 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2349 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2353 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2354 struct btrfs_dir_log_item);
2355 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2356 btrfs_mark_buffer_dirty(path->nodes[0]);
2357 btrfs_release_path(log, path);
2362 * log all the items included in the current transaction for a given
2363 * directory. This also creates the range items in the log tree required
2364 * to replay anything deleted before the fsync
2366 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2367 struct btrfs_root *root, struct inode *inode,
2368 struct btrfs_path *path,
2369 struct btrfs_path *dst_path, int key_type,
2370 u64 min_offset, u64 *last_offset_ret)
2372 struct btrfs_key min_key;
2373 struct btrfs_key max_key;
2374 struct btrfs_root *log = root->log_root;
2375 struct extent_buffer *src;
2380 u64 first_offset = min_offset;
2381 u64 last_offset = (u64)-1;
2383 log = root->log_root;
2384 max_key.objectid = inode->i_ino;
2385 max_key.offset = (u64)-1;
2386 max_key.type = key_type;
2388 min_key.objectid = inode->i_ino;
2389 min_key.type = key_type;
2390 min_key.offset = min_offset;
2392 path->keep_locks = 1;
2394 ret = btrfs_search_forward(root, &min_key, &max_key,
2395 path, 0, trans->transid);
2398 * we didn't find anything from this transaction, see if there
2399 * is anything at all
2401 if (ret != 0 || min_key.objectid != inode->i_ino ||
2402 min_key.type != key_type) {
2403 min_key.objectid = inode->i_ino;
2404 min_key.type = key_type;
2405 min_key.offset = (u64)-1;
2406 btrfs_release_path(root, path);
2407 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2409 btrfs_release_path(root, path);
2412 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2414 /* if ret == 0 there are items for this type,
2415 * create a range to tell us the last key of this type.
2416 * otherwise, there are no items in this directory after
2417 * *min_offset, and we create a range to indicate that.
2420 struct btrfs_key tmp;
2421 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2423 if (key_type == tmp.type)
2424 first_offset = max(min_offset, tmp.offset) + 1;
2429 /* go backward to find any previous key */
2430 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2432 struct btrfs_key tmp;
2433 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2434 if (key_type == tmp.type) {
2435 first_offset = tmp.offset;
2436 ret = overwrite_item(trans, log, dst_path,
2437 path->nodes[0], path->slots[0],
2445 btrfs_release_path(root, path);
2447 /* find the first key from this transaction again */
2448 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2455 * we have a block from this transaction, log every item in it
2456 * from our directory
2459 struct btrfs_key tmp;
2460 src = path->nodes[0];
2461 nritems = btrfs_header_nritems(src);
2462 for (i = path->slots[0]; i < nritems; i++) {
2463 btrfs_item_key_to_cpu(src, &min_key, i);
2465 if (min_key.objectid != inode->i_ino ||
2466 min_key.type != key_type)
2468 ret = overwrite_item(trans, log, dst_path, src, i,
2475 path->slots[0] = nritems;
2478 * look ahead to the next item and see if it is also
2479 * from this directory and from this transaction
2481 ret = btrfs_next_leaf(root, path);
2483 last_offset = (u64)-1;
2486 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2487 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2488 last_offset = (u64)-1;
2491 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2492 ret = overwrite_item(trans, log, dst_path,
2493 path->nodes[0], path->slots[0],
2498 last_offset = tmp.offset;
2503 btrfs_release_path(root, path);
2504 btrfs_release_path(log, dst_path);
2507 *last_offset_ret = last_offset;
2509 * insert the log range keys to indicate where the log
2512 ret = insert_dir_log_key(trans, log, path, key_type,
2513 inode->i_ino, first_offset,
2522 * logging directories is very similar to logging inodes, We find all the items
2523 * from the current transaction and write them to the log.
2525 * The recovery code scans the directory in the subvolume, and if it finds a
2526 * key in the range logged that is not present in the log tree, then it means
2527 * that dir entry was unlinked during the transaction.
2529 * In order for that scan to work, we must include one key smaller than
2530 * the smallest logged by this transaction and one key larger than the largest
2531 * key logged by this transaction.
2533 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2534 struct btrfs_root *root, struct inode *inode,
2535 struct btrfs_path *path,
2536 struct btrfs_path *dst_path)
2541 int key_type = BTRFS_DIR_ITEM_KEY;
2547 ret = log_dir_items(trans, root, inode, path,
2548 dst_path, key_type, min_key,
2552 if (max_key == (u64)-1)
2554 min_key = max_key + 1;
2557 if (key_type == BTRFS_DIR_ITEM_KEY) {
2558 key_type = BTRFS_DIR_INDEX_KEY;
2565 * a helper function to drop items from the log before we relog an
2566 * inode. max_key_type indicates the highest item type to remove.
2567 * This cannot be run for file data extents because it does not
2568 * free the extents they point to.
2570 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2571 struct btrfs_root *log,
2572 struct btrfs_path *path,
2573 u64 objectid, int max_key_type)
2576 struct btrfs_key key;
2577 struct btrfs_key found_key;
2579 key.objectid = objectid;
2580 key.type = max_key_type;
2581 key.offset = (u64)-1;
2584 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2589 if (path->slots[0] == 0)
2593 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2596 if (found_key.objectid != objectid)
2599 ret = btrfs_del_item(trans, log, path);
2601 btrfs_release_path(log, path);
2603 btrfs_release_path(log, path);
2607 static noinline int copy_items(struct btrfs_trans_handle *trans,
2608 struct btrfs_root *log,
2609 struct btrfs_path *dst_path,
2610 struct extent_buffer *src,
2611 int start_slot, int nr, int inode_only)
2613 unsigned long src_offset;
2614 unsigned long dst_offset;
2615 struct btrfs_file_extent_item *extent;
2616 struct btrfs_inode_item *inode_item;
2618 struct btrfs_key *ins_keys;
2622 struct list_head ordered_sums;
2624 INIT_LIST_HEAD(&ordered_sums);
2626 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2627 nr * sizeof(u32), GFP_NOFS);
2631 ins_sizes = (u32 *)ins_data;
2632 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2634 for (i = 0; i < nr; i++) {
2635 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2636 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2638 ret = btrfs_insert_empty_items(trans, log, dst_path,
2639 ins_keys, ins_sizes, nr);
2645 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2646 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2647 dst_path->slots[0]);
2649 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2651 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2652 src_offset, ins_sizes[i]);
2654 if (inode_only == LOG_INODE_EXISTS &&
2655 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2656 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2658 struct btrfs_inode_item);
2659 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2661 /* set the generation to zero so the recover code
2662 * can tell the difference between an logging
2663 * just to say 'this inode exists' and a logging
2664 * to say 'update this inode with these values'
2666 btrfs_set_inode_generation(dst_path->nodes[0],
2669 /* take a reference on file data extents so that truncates
2670 * or deletes of this inode don't have to relog the inode
2673 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2675 extent = btrfs_item_ptr(src, start_slot + i,
2676 struct btrfs_file_extent_item);
2678 found_type = btrfs_file_extent_type(src, extent);
2679 if (found_type == BTRFS_FILE_EXTENT_REG ||
2680 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2682 ds = btrfs_file_extent_disk_bytenr(src,
2684 /* ds == 0 is a hole */
2688 dl = btrfs_file_extent_disk_num_bytes(src,
2690 cs = btrfs_file_extent_offset(src, extent);
2691 cl = btrfs_file_extent_num_bytes(src,
2693 if (btrfs_file_extent_compression(src,
2699 ret = btrfs_lookup_csums_range(
2700 log->fs_info->csum_root,
2701 ds + cs, ds + cs + cl - 1,
2708 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2709 btrfs_release_path(log, dst_path);
2713 * we have to do this after the loop above to avoid changing the
2714 * log tree while trying to change the log tree.
2717 while (!list_empty(&ordered_sums)) {
2718 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2719 struct btrfs_ordered_sum,
2722 ret = btrfs_csum_file_blocks(trans, log, sums);
2723 list_del(&sums->list);
2729 /* log a single inode in the tree log.
2730 * At least one parent directory for this inode must exist in the tree
2731 * or be logged already.
2733 * Any items from this inode changed by the current transaction are copied
2734 * to the log tree. An extra reference is taken on any extents in this
2735 * file, allowing us to avoid a whole pile of corner cases around logging
2736 * blocks that have been removed from the tree.
2738 * See LOG_INODE_ALL and related defines for a description of what inode_only
2741 * This handles both files and directories.
2743 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2744 struct btrfs_root *root, struct inode *inode,
2747 struct btrfs_path *path;
2748 struct btrfs_path *dst_path;
2749 struct btrfs_key min_key;
2750 struct btrfs_key max_key;
2751 struct btrfs_root *log = root->log_root;
2752 struct extent_buffer *src = NULL;
2756 int ins_start_slot = 0;
2759 log = root->log_root;
2761 path = btrfs_alloc_path();
2764 dst_path = btrfs_alloc_path();
2766 btrfs_free_path(path);
2770 min_key.objectid = inode->i_ino;
2771 min_key.type = BTRFS_INODE_ITEM_KEY;
2774 max_key.objectid = inode->i_ino;
2776 /* today the code can only do partial logging of directories */
2777 if (!S_ISDIR(inode->i_mode))
2778 inode_only = LOG_INODE_ALL;
2780 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2781 max_key.type = BTRFS_XATTR_ITEM_KEY;
2783 max_key.type = (u8)-1;
2784 max_key.offset = (u64)-1;
2786 mutex_lock(&BTRFS_I(inode)->log_mutex);
2789 * a brute force approach to making sure we get the most uptodate
2790 * copies of everything.
2792 if (S_ISDIR(inode->i_mode)) {
2793 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2795 if (inode_only == LOG_INODE_EXISTS)
2796 max_key_type = BTRFS_XATTR_ITEM_KEY;
2797 ret = drop_objectid_items(trans, log, path,
2798 inode->i_ino, max_key_type);
2800 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2806 path->keep_locks = 1;
2810 ret = btrfs_search_forward(root, &min_key, &max_key,
2811 path, 0, trans->transid);
2815 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2816 if (min_key.objectid != inode->i_ino)
2818 if (min_key.type > max_key.type)
2821 src = path->nodes[0];
2822 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2825 } else if (!ins_nr) {
2826 ins_start_slot = path->slots[0];
2831 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2832 ins_nr, inode_only);
2838 ins_start_slot = path->slots[0];
2841 nritems = btrfs_header_nritems(path->nodes[0]);
2843 if (path->slots[0] < nritems) {
2844 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2849 ret = copy_items(trans, log, dst_path, src,
2851 ins_nr, inode_only);
2858 btrfs_release_path(root, path);
2860 if (min_key.offset < (u64)-1)
2862 else if (min_key.type < (u8)-1)
2864 else if (min_key.objectid < (u64)-1)
2870 ret = copy_items(trans, log, dst_path, src,
2872 ins_nr, inode_only);
2880 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2881 btrfs_release_path(root, path);
2882 btrfs_release_path(log, dst_path);
2883 ret = log_directory_changes(trans, root, inode, path, dst_path);
2889 BTRFS_I(inode)->logged_trans = trans->transid;
2891 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2893 btrfs_free_path(path);
2894 btrfs_free_path(dst_path);
2899 * follow the dentry parent pointers up the chain and see if any
2900 * of the directories in it require a full commit before they can
2901 * be logged. Returns zero if nothing special needs to be done or 1 if
2902 * a full commit is required.
2904 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2905 struct inode *inode,
2906 struct dentry *parent,
2907 struct super_block *sb,
2911 struct btrfs_root *root;
2912 struct dentry *old_parent = NULL;
2915 * for regular files, if its inode is already on disk, we don't
2916 * have to worry about the parents at all. This is because
2917 * we can use the last_unlink_trans field to record renames
2918 * and other fun in this file.
2920 if (S_ISREG(inode->i_mode) &&
2921 BTRFS_I(inode)->generation <= last_committed &&
2922 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2925 if (!S_ISDIR(inode->i_mode)) {
2926 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2928 inode = parent->d_inode;
2932 BTRFS_I(inode)->logged_trans = trans->transid;
2935 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2936 root = BTRFS_I(inode)->root;
2939 * make sure any commits to the log are forced
2940 * to be full commits
2942 root->fs_info->last_trans_log_full_commit =
2948 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2951 if (IS_ROOT(parent))
2954 parent = dget_parent(parent);
2956 old_parent = parent;
2957 inode = parent->d_inode;
2965 static int inode_in_log(struct btrfs_trans_handle *trans,
2966 struct inode *inode)
2968 struct btrfs_root *root = BTRFS_I(inode)->root;
2971 mutex_lock(&root->log_mutex);
2972 if (BTRFS_I(inode)->logged_trans == trans->transid &&
2973 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2975 mutex_unlock(&root->log_mutex);
2981 * helper function around btrfs_log_inode to make sure newly created
2982 * parent directories also end up in the log. A minimal inode and backref
2983 * only logging is done of any parent directories that are older than
2984 * the last committed transaction
2986 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2987 struct btrfs_root *root, struct inode *inode,
2988 struct dentry *parent, int exists_only)
2990 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
2991 struct super_block *sb;
2992 struct dentry *old_parent = NULL;
2994 u64 last_committed = root->fs_info->last_trans_committed;
2998 if (btrfs_test_opt(root, NOTREELOG)) {
3003 if (root->fs_info->last_trans_log_full_commit >
3004 root->fs_info->last_trans_committed) {
3009 if (root != BTRFS_I(inode)->root ||
3010 btrfs_root_refs(&root->root_item) == 0) {
3015 ret = check_parent_dirs_for_sync(trans, inode, parent,
3016 sb, last_committed);
3020 if (inode_in_log(trans, inode)) {
3021 ret = BTRFS_NO_LOG_SYNC;
3025 ret = start_log_trans(trans, root);
3029 ret = btrfs_log_inode(trans, root, inode, inode_only);
3034 * for regular files, if its inode is already on disk, we don't
3035 * have to worry about the parents at all. This is because
3036 * we can use the last_unlink_trans field to record renames
3037 * and other fun in this file.
3039 if (S_ISREG(inode->i_mode) &&
3040 BTRFS_I(inode)->generation <= last_committed &&
3041 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3046 inode_only = LOG_INODE_EXISTS;
3048 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3051 inode = parent->d_inode;
3052 if (root != BTRFS_I(inode)->root)
3055 if (BTRFS_I(inode)->generation >
3056 root->fs_info->last_trans_committed) {
3057 ret = btrfs_log_inode(trans, root, inode, inode_only);
3061 if (IS_ROOT(parent))
3064 parent = dget_parent(parent);
3066 old_parent = parent;
3072 BUG_ON(ret != -ENOSPC);
3073 root->fs_info->last_trans_log_full_commit = trans->transid;
3076 btrfs_end_log_trans(root);
3082 * it is not safe to log dentry if the chunk root has added new
3083 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3084 * If this returns 1, you must commit the transaction to safely get your
3087 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3088 struct btrfs_root *root, struct dentry *dentry)
3090 struct dentry *parent = dget_parent(dentry);
3093 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3100 * should be called during mount to recover any replay any log trees
3103 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3106 struct btrfs_path *path;
3107 struct btrfs_trans_handle *trans;
3108 struct btrfs_key key;
3109 struct btrfs_key found_key;
3110 struct btrfs_key tmp_key;
3111 struct btrfs_root *log;
3112 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3113 struct walk_control wc = {
3114 .process_func = process_one_buffer,
3118 path = btrfs_alloc_path();
3122 fs_info->log_root_recovering = 1;
3124 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3125 BUG_ON(IS_ERR(trans));
3130 ret = walk_log_tree(trans, log_root_tree, &wc);
3134 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3135 key.offset = (u64)-1;
3136 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3139 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3143 if (path->slots[0] == 0)
3147 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3149 btrfs_release_path(log_root_tree, path);
3150 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3153 log = btrfs_read_fs_root_no_radix(log_root_tree,
3155 BUG_ON(IS_ERR(log));
3157 tmp_key.objectid = found_key.offset;
3158 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3159 tmp_key.offset = (u64)-1;
3161 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3162 BUG_ON(!wc.replay_dest);
3164 wc.replay_dest->log_root = log;
3165 btrfs_record_root_in_trans(trans, wc.replay_dest);
3166 ret = walk_log_tree(trans, log, &wc);
3169 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3170 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3175 key.offset = found_key.offset - 1;
3176 wc.replay_dest->log_root = NULL;
3177 free_extent_buffer(log->node);
3178 free_extent_buffer(log->commit_root);
3181 if (found_key.offset == 0)
3184 btrfs_release_path(log_root_tree, path);
3186 /* step one is to pin it all, step two is to replay just inodes */
3189 wc.process_func = replay_one_buffer;
3190 wc.stage = LOG_WALK_REPLAY_INODES;
3193 /* step three is to replay everything */
3194 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3199 btrfs_free_path(path);
3201 free_extent_buffer(log_root_tree->node);
3202 log_root_tree->log_root = NULL;
3203 fs_info->log_root_recovering = 0;
3205 /* step 4: commit the transaction, which also unpins the blocks */
3206 btrfs_commit_transaction(trans, fs_info->tree_root);
3208 kfree(log_root_tree);
3213 * there are some corner cases where we want to force a full
3214 * commit instead of allowing a directory to be logged.
3216 * They revolve around files there were unlinked from the directory, and
3217 * this function updates the parent directory so that a full commit is
3218 * properly done if it is fsync'd later after the unlinks are done.
3220 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3221 struct inode *dir, struct inode *inode,
3225 * when we're logging a file, if it hasn't been renamed
3226 * or unlinked, and its inode is fully committed on disk,
3227 * we don't have to worry about walking up the directory chain
3228 * to log its parents.
3230 * So, we use the last_unlink_trans field to put this transid
3231 * into the file. When the file is logged we check it and
3232 * don't log the parents if the file is fully on disk.
3234 if (S_ISREG(inode->i_mode))
3235 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3238 * if this directory was already logged any new
3239 * names for this file/dir will get recorded
3242 if (BTRFS_I(dir)->logged_trans == trans->transid)
3246 * if the inode we're about to unlink was logged,
3247 * the log will be properly updated for any new names
3249 if (BTRFS_I(inode)->logged_trans == trans->transid)
3253 * when renaming files across directories, if the directory
3254 * there we're unlinking from gets fsync'd later on, there's
3255 * no way to find the destination directory later and fsync it
3256 * properly. So, we have to be conservative and force commits
3257 * so the new name gets discovered.
3262 /* we can safely do the unlink without any special recording */
3266 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3270 * Call this after adding a new name for a file and it will properly
3271 * update the log to reflect the new name.
3273 * It will return zero if all goes well, and it will return 1 if a
3274 * full transaction commit is required.
3276 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3277 struct inode *inode, struct inode *old_dir,
3278 struct dentry *parent)
3280 struct btrfs_root * root = BTRFS_I(inode)->root;
3283 * this will force the logging code to walk the dentry chain
3286 if (S_ISREG(inode->i_mode))
3287 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3290 * if this inode hasn't been logged and directory we're renaming it
3291 * from hasn't been logged, we don't need to log it
3293 if (BTRFS_I(inode)->logged_trans <=
3294 root->fs_info->last_trans_committed &&
3295 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3296 root->fs_info->last_trans_committed))
3299 return btrfs_log_inode_parent(trans, root, inode, parent, 1);