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_for_log_replay(wc->trans,
280 log->fs_info->extent_root,
283 if (btrfs_buffer_uptodate(eb, gen)) {
285 btrfs_write_tree_block(eb);
287 btrfs_wait_tree_block_writeback(eb);
293 * Item overwrite used by replay and tree logging. eb, slot and key all refer
294 * to the src data we are copying out.
296 * root is the tree we are copying into, and path is a scratch
297 * path for use in this function (it should be released on entry and
298 * will be released on exit).
300 * If the key is already in the destination tree the existing item is
301 * overwritten. If the existing item isn't big enough, it is extended.
302 * If it is too large, it is truncated.
304 * If the key isn't in the destination yet, a new item is inserted.
306 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
307 struct btrfs_root *root,
308 struct btrfs_path *path,
309 struct extent_buffer *eb, int slot,
310 struct btrfs_key *key)
314 u64 saved_i_size = 0;
315 int save_old_i_size = 0;
316 unsigned long src_ptr;
317 unsigned long dst_ptr;
318 int overwrite_root = 0;
320 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
323 item_size = btrfs_item_size_nr(eb, slot);
324 src_ptr = btrfs_item_ptr_offset(eb, slot);
326 /* look for the key in the destination tree */
327 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
331 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
333 if (dst_size != item_size)
336 if (item_size == 0) {
337 btrfs_release_path(path);
340 dst_copy = kmalloc(item_size, GFP_NOFS);
341 src_copy = kmalloc(item_size, GFP_NOFS);
342 if (!dst_copy || !src_copy) {
343 btrfs_release_path(path);
349 read_extent_buffer(eb, src_copy, src_ptr, item_size);
351 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
352 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
354 ret = memcmp(dst_copy, src_copy, item_size);
359 * they have the same contents, just return, this saves
360 * us from cowing blocks in the destination tree and doing
361 * extra writes that may not have been done by a previous
365 btrfs_release_path(path);
371 btrfs_release_path(path);
372 /* try to insert the key into the destination tree */
373 ret = btrfs_insert_empty_item(trans, root, path,
376 /* make sure any existing item is the correct size */
377 if (ret == -EEXIST) {
379 found_size = btrfs_item_size_nr(path->nodes[0],
381 if (found_size > item_size) {
382 btrfs_truncate_item(trans, root, path, item_size, 1);
383 } else if (found_size < item_size) {
384 ret = btrfs_extend_item(trans, root, path,
385 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(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, btrfs_ino(inode),
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(path);
551 btrfs_release_path(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);
596 * insert the extent pointer in the extent
599 ret = btrfs_alloc_logged_file_extent(trans,
600 root, root->root_key.objectid,
601 key->objectid, offset, &ins);
604 btrfs_release_path(path);
606 if (btrfs_file_extent_compression(eb, item)) {
607 csum_start = ins.objectid;
608 csum_end = csum_start + ins.offset;
610 csum_start = ins.objectid +
611 btrfs_file_extent_offset(eb, item);
612 csum_end = csum_start +
613 btrfs_file_extent_num_bytes(eb, item);
616 ret = btrfs_lookup_csums_range(root->log_root,
617 csum_start, csum_end - 1,
620 while (!list_empty(&ordered_sums)) {
621 struct btrfs_ordered_sum *sums;
622 sums = list_entry(ordered_sums.next,
623 struct btrfs_ordered_sum,
625 ret = btrfs_csum_file_blocks(trans,
626 root->fs_info->csum_root,
629 list_del(&sums->list);
633 btrfs_release_path(path);
635 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
636 /* inline extents are easy, we just overwrite them */
637 ret = overwrite_item(trans, root, path, eb, slot, key);
641 inode_set_bytes(inode, saved_nbytes);
642 btrfs_update_inode(trans, root, inode);
650 * when cleaning up conflicts between the directory names in the
651 * subvolume, directory names in the log and directory names in the
652 * inode back references, we may have to unlink inodes from directories.
654 * This is a helper function to do the unlink of a specific directory
657 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
658 struct btrfs_root *root,
659 struct btrfs_path *path,
661 struct btrfs_dir_item *di)
666 struct extent_buffer *leaf;
667 struct btrfs_key location;
670 leaf = path->nodes[0];
672 btrfs_dir_item_key_to_cpu(leaf, di, &location);
673 name_len = btrfs_dir_name_len(leaf, di);
674 name = kmalloc(name_len, GFP_NOFS);
678 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
679 btrfs_release_path(path);
681 inode = read_one_inode(root, location.objectid);
687 ret = link_to_fixup_dir(trans, root, path, location.objectid);
690 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
699 * helper function to see if a given name and sequence number found
700 * in an inode back reference are already in a directory and correctly
701 * point to this inode
703 static noinline int inode_in_dir(struct btrfs_root *root,
704 struct btrfs_path *path,
705 u64 dirid, u64 objectid, u64 index,
706 const char *name, int name_len)
708 struct btrfs_dir_item *di;
709 struct btrfs_key location;
712 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
713 index, name, name_len, 0);
714 if (di && !IS_ERR(di)) {
715 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
716 if (location.objectid != objectid)
720 btrfs_release_path(path);
722 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
723 if (di && !IS_ERR(di)) {
724 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
725 if (location.objectid != objectid)
731 btrfs_release_path(path);
736 * helper function to check a log tree for a named back reference in
737 * an inode. This is used to decide if a back reference that is
738 * found in the subvolume conflicts with what we find in the log.
740 * inode backreferences may have multiple refs in a single item,
741 * during replay we process one reference at a time, and we don't
742 * want to delete valid links to a file from the subvolume if that
743 * link is also in the log.
745 static noinline int backref_in_log(struct btrfs_root *log,
746 struct btrfs_key *key,
747 char *name, int namelen)
749 struct btrfs_path *path;
750 struct btrfs_inode_ref *ref;
752 unsigned long ptr_end;
753 unsigned long name_ptr;
759 path = btrfs_alloc_path();
763 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
767 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
768 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
769 ptr_end = ptr + item_size;
770 while (ptr < ptr_end) {
771 ref = (struct btrfs_inode_ref *)ptr;
772 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
773 if (found_name_len == namelen) {
774 name_ptr = (unsigned long)(ref + 1);
775 ret = memcmp_extent_buffer(path->nodes[0], name,
782 ptr = (unsigned long)(ref + 1) + found_name_len;
785 btrfs_free_path(path);
791 * replay one inode back reference item found in the log tree.
792 * eb, slot and key refer to the buffer and key found in the log tree.
793 * root is the destination we are replaying into, and path is for temp
794 * use by this function. (it should be released on return).
796 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
797 struct btrfs_root *root,
798 struct btrfs_root *log,
799 struct btrfs_path *path,
800 struct extent_buffer *eb, int slot,
801 struct btrfs_key *key)
803 struct btrfs_inode_ref *ref;
804 struct btrfs_dir_item *di;
807 unsigned long ref_ptr;
808 unsigned long ref_end;
815 * it is possible that we didn't log all the parent directories
816 * for a given inode. If we don't find the dir, just don't
817 * copy the back ref in. The link count fixup code will take
820 dir = read_one_inode(root, key->offset);
824 inode = read_one_inode(root, key->objectid);
830 ref_ptr = btrfs_item_ptr_offset(eb, slot);
831 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
834 ref = (struct btrfs_inode_ref *)ref_ptr;
836 namelen = btrfs_inode_ref_name_len(eb, ref);
837 name = kmalloc(namelen, GFP_NOFS);
840 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
842 /* if we already have a perfect match, we're done */
843 if (inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
844 btrfs_inode_ref_index(eb, ref),
850 * look for a conflicting back reference in the metadata.
851 * if we find one we have to unlink that name of the file
852 * before we add our new link. Later on, we overwrite any
853 * existing back reference, and we don't want to create
854 * dangling pointers in the directory.
860 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
864 struct btrfs_inode_ref *victim_ref;
866 unsigned long ptr_end;
867 struct extent_buffer *leaf = path->nodes[0];
869 /* are we trying to overwrite a back ref for the root directory
870 * if so, just jump out, we're done
872 if (key->objectid == key->offset)
875 /* check all the names in this back reference to see
876 * if they are in the log. if so, we allow them to stay
877 * otherwise they must be unlinked as a conflict
879 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
880 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
881 while (ptr < ptr_end) {
882 victim_ref = (struct btrfs_inode_ref *)ptr;
883 victim_name_len = btrfs_inode_ref_name_len(leaf,
885 victim_name = kmalloc(victim_name_len, GFP_NOFS);
886 BUG_ON(!victim_name);
888 read_extent_buffer(leaf, victim_name,
889 (unsigned long)(victim_ref + 1),
892 if (!backref_in_log(log, key, victim_name,
894 btrfs_inc_nlink(inode);
895 btrfs_release_path(path);
897 ret = btrfs_unlink_inode(trans, root, dir,
902 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
907 * NOTE: we have searched root tree and checked the
908 * coresponding ref, it does not need to check again.
912 btrfs_release_path(path);
914 /* look for a conflicting sequence number */
915 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
916 btrfs_inode_ref_index(eb, ref),
918 if (di && !IS_ERR(di)) {
919 ret = drop_one_dir_item(trans, root, path, dir, di);
922 btrfs_release_path(path);
924 /* look for a conflicing name */
925 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
927 if (di && !IS_ERR(di)) {
928 ret = drop_one_dir_item(trans, root, path, dir, di);
931 btrfs_release_path(path);
934 /* insert our name */
935 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
936 btrfs_inode_ref_index(eb, ref));
939 btrfs_update_inode(trans, root, inode);
942 ref_ptr = (unsigned long)(ref + 1) + namelen;
944 if (ref_ptr < ref_end)
947 /* finally write the back reference in the inode */
948 ret = overwrite_item(trans, root, path, eb, slot, key);
952 btrfs_release_path(path);
958 static int insert_orphan_item(struct btrfs_trans_handle *trans,
959 struct btrfs_root *root, u64 offset)
962 ret = btrfs_find_orphan_item(root, offset);
964 ret = btrfs_insert_orphan_item(trans, root, offset);
970 * There are a few corners where the link count of the file can't
971 * be properly maintained during replay. So, instead of adding
972 * lots of complexity to the log code, we just scan the backrefs
973 * for any file that has been through replay.
975 * The scan will update the link count on the inode to reflect the
976 * number of back refs found. If it goes down to zero, the iput
977 * will free the inode.
979 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
980 struct btrfs_root *root,
983 struct btrfs_path *path;
985 struct btrfs_key key;
988 unsigned long ptr_end;
990 u64 ino = btrfs_ino(inode);
993 key.type = BTRFS_INODE_REF_KEY;
994 key.offset = (u64)-1;
996 path = btrfs_alloc_path();
1001 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1005 if (path->slots[0] == 0)
1009 btrfs_item_key_to_cpu(path->nodes[0], &key,
1011 if (key.objectid != ino ||
1012 key.type != BTRFS_INODE_REF_KEY)
1014 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1015 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1017 while (ptr < ptr_end) {
1018 struct btrfs_inode_ref *ref;
1020 ref = (struct btrfs_inode_ref *)ptr;
1021 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1023 ptr = (unsigned long)(ref + 1) + name_len;
1027 if (key.offset == 0)
1030 btrfs_release_path(path);
1032 btrfs_release_path(path);
1033 if (nlink != inode->i_nlink) {
1034 inode->i_nlink = nlink;
1035 btrfs_update_inode(trans, root, inode);
1037 BTRFS_I(inode)->index_cnt = (u64)-1;
1039 if (inode->i_nlink == 0) {
1040 if (S_ISDIR(inode->i_mode)) {
1041 ret = replay_dir_deletes(trans, root, NULL, path,
1045 ret = insert_orphan_item(trans, root, ino);
1048 btrfs_free_path(path);
1053 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1054 struct btrfs_root *root,
1055 struct btrfs_path *path)
1058 struct btrfs_key key;
1059 struct inode *inode;
1061 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1062 key.type = BTRFS_ORPHAN_ITEM_KEY;
1063 key.offset = (u64)-1;
1065 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1070 if (path->slots[0] == 0)
1075 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1076 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1077 key.type != BTRFS_ORPHAN_ITEM_KEY)
1080 ret = btrfs_del_item(trans, root, path);
1084 btrfs_release_path(path);
1085 inode = read_one_inode(root, key.offset);
1089 ret = fixup_inode_link_count(trans, root, inode);
1095 * fixup on a directory may create new entries,
1096 * make sure we always look for the highset possible
1099 key.offset = (u64)-1;
1103 btrfs_release_path(path);
1109 * record a given inode in the fixup dir so we can check its link
1110 * count when replay is done. The link count is incremented here
1111 * so the inode won't go away until we check it
1113 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1114 struct btrfs_root *root,
1115 struct btrfs_path *path,
1118 struct btrfs_key key;
1120 struct inode *inode;
1122 inode = read_one_inode(root, objectid);
1126 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1127 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1128 key.offset = objectid;
1130 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1132 btrfs_release_path(path);
1134 btrfs_inc_nlink(inode);
1135 btrfs_update_inode(trans, root, inode);
1136 } else if (ret == -EEXIST) {
1147 * when replaying the log for a directory, we only insert names
1148 * for inodes that actually exist. This means an fsync on a directory
1149 * does not implicitly fsync all the new files in it
1151 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1152 struct btrfs_root *root,
1153 struct btrfs_path *path,
1154 u64 dirid, u64 index,
1155 char *name, int name_len, u8 type,
1156 struct btrfs_key *location)
1158 struct inode *inode;
1162 inode = read_one_inode(root, location->objectid);
1166 dir = read_one_inode(root, dirid);
1171 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1173 /* FIXME, put inode into FIXUP list */
1181 * take a single entry in a log directory item and replay it into
1184 * if a conflicting item exists in the subdirectory already,
1185 * the inode it points to is unlinked and put into the link count
1188 * If a name from the log points to a file or directory that does
1189 * not exist in the FS, it is skipped. fsyncs on directories
1190 * do not force down inodes inside that directory, just changes to the
1191 * names or unlinks in a directory.
1193 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1194 struct btrfs_root *root,
1195 struct btrfs_path *path,
1196 struct extent_buffer *eb,
1197 struct btrfs_dir_item *di,
1198 struct btrfs_key *key)
1202 struct btrfs_dir_item *dst_di;
1203 struct btrfs_key found_key;
1204 struct btrfs_key log_key;
1210 dir = read_one_inode(root, key->objectid);
1214 name_len = btrfs_dir_name_len(eb, di);
1215 name = kmalloc(name_len, GFP_NOFS);
1219 log_type = btrfs_dir_type(eb, di);
1220 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1223 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1224 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1229 btrfs_release_path(path);
1231 if (key->type == BTRFS_DIR_ITEM_KEY) {
1232 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1234 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1235 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1242 if (IS_ERR_OR_NULL(dst_di)) {
1243 /* we need a sequence number to insert, so we only
1244 * do inserts for the BTRFS_DIR_INDEX_KEY types
1246 if (key->type != BTRFS_DIR_INDEX_KEY)
1251 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1252 /* the existing item matches the logged item */
1253 if (found_key.objectid == log_key.objectid &&
1254 found_key.type == log_key.type &&
1255 found_key.offset == log_key.offset &&
1256 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1261 * don't drop the conflicting directory entry if the inode
1262 * for the new entry doesn't exist
1267 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1270 if (key->type == BTRFS_DIR_INDEX_KEY)
1273 btrfs_release_path(path);
1279 btrfs_release_path(path);
1280 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1281 name, name_len, log_type, &log_key);
1283 BUG_ON(ret && ret != -ENOENT);
1288 * find all the names in a directory item and reconcile them into
1289 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1290 * one name in a directory item, but the same code gets used for
1291 * both directory index types
1293 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1294 struct btrfs_root *root,
1295 struct btrfs_path *path,
1296 struct extent_buffer *eb, int slot,
1297 struct btrfs_key *key)
1300 u32 item_size = btrfs_item_size_nr(eb, slot);
1301 struct btrfs_dir_item *di;
1304 unsigned long ptr_end;
1306 ptr = btrfs_item_ptr_offset(eb, slot);
1307 ptr_end = ptr + item_size;
1308 while (ptr < ptr_end) {
1309 di = (struct btrfs_dir_item *)ptr;
1310 if (verify_dir_item(root, eb, di))
1312 name_len = btrfs_dir_name_len(eb, di);
1313 ret = replay_one_name(trans, root, path, eb, di, key);
1315 ptr = (unsigned long)(di + 1);
1322 * directory replay has two parts. There are the standard directory
1323 * items in the log copied from the subvolume, and range items
1324 * created in the log while the subvolume was logged.
1326 * The range items tell us which parts of the key space the log
1327 * is authoritative for. During replay, if a key in the subvolume
1328 * directory is in a logged range item, but not actually in the log
1329 * that means it was deleted from the directory before the fsync
1330 * and should be removed.
1332 static noinline int find_dir_range(struct btrfs_root *root,
1333 struct btrfs_path *path,
1334 u64 dirid, int key_type,
1335 u64 *start_ret, u64 *end_ret)
1337 struct btrfs_key key;
1339 struct btrfs_dir_log_item *item;
1343 if (*start_ret == (u64)-1)
1346 key.objectid = dirid;
1347 key.type = key_type;
1348 key.offset = *start_ret;
1350 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1354 if (path->slots[0] == 0)
1359 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1361 if (key.type != key_type || key.objectid != dirid) {
1365 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1366 struct btrfs_dir_log_item);
1367 found_end = btrfs_dir_log_end(path->nodes[0], item);
1369 if (*start_ret >= key.offset && *start_ret <= found_end) {
1371 *start_ret = key.offset;
1372 *end_ret = found_end;
1377 /* check the next slot in the tree to see if it is a valid item */
1378 nritems = btrfs_header_nritems(path->nodes[0]);
1379 if (path->slots[0] >= nritems) {
1380 ret = btrfs_next_leaf(root, path);
1387 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1389 if (key.type != key_type || key.objectid != dirid) {
1393 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1394 struct btrfs_dir_log_item);
1395 found_end = btrfs_dir_log_end(path->nodes[0], item);
1396 *start_ret = key.offset;
1397 *end_ret = found_end;
1400 btrfs_release_path(path);
1405 * this looks for a given directory item in the log. If the directory
1406 * item is not in the log, the item is removed and the inode it points
1409 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1410 struct btrfs_root *root,
1411 struct btrfs_root *log,
1412 struct btrfs_path *path,
1413 struct btrfs_path *log_path,
1415 struct btrfs_key *dir_key)
1418 struct extent_buffer *eb;
1421 struct btrfs_dir_item *di;
1422 struct btrfs_dir_item *log_di;
1425 unsigned long ptr_end;
1427 struct inode *inode;
1428 struct btrfs_key location;
1431 eb = path->nodes[0];
1432 slot = path->slots[0];
1433 item_size = btrfs_item_size_nr(eb, slot);
1434 ptr = btrfs_item_ptr_offset(eb, slot);
1435 ptr_end = ptr + item_size;
1436 while (ptr < ptr_end) {
1437 di = (struct btrfs_dir_item *)ptr;
1438 if (verify_dir_item(root, eb, di)) {
1443 name_len = btrfs_dir_name_len(eb, di);
1444 name = kmalloc(name_len, GFP_NOFS);
1449 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1452 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1453 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1456 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1457 log_di = btrfs_lookup_dir_index_item(trans, log,
1463 if (IS_ERR_OR_NULL(log_di)) {
1464 btrfs_dir_item_key_to_cpu(eb, di, &location);
1465 btrfs_release_path(path);
1466 btrfs_release_path(log_path);
1467 inode = read_one_inode(root, location.objectid);
1473 ret = link_to_fixup_dir(trans, root,
1474 path, location.objectid);
1476 btrfs_inc_nlink(inode);
1477 ret = btrfs_unlink_inode(trans, root, dir, inode,
1483 /* there might still be more names under this key
1484 * check and repeat if required
1486 ret = btrfs_search_slot(NULL, root, dir_key, path,
1493 btrfs_release_path(log_path);
1496 ptr = (unsigned long)(di + 1);
1501 btrfs_release_path(path);
1502 btrfs_release_path(log_path);
1507 * deletion replay happens before we copy any new directory items
1508 * out of the log or out of backreferences from inodes. It
1509 * scans the log to find ranges of keys that log is authoritative for,
1510 * and then scans the directory to find items in those ranges that are
1511 * not present in the log.
1513 * Anything we don't find in the log is unlinked and removed from the
1516 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1517 struct btrfs_root *root,
1518 struct btrfs_root *log,
1519 struct btrfs_path *path,
1520 u64 dirid, int del_all)
1524 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1526 struct btrfs_key dir_key;
1527 struct btrfs_key found_key;
1528 struct btrfs_path *log_path;
1531 dir_key.objectid = dirid;
1532 dir_key.type = BTRFS_DIR_ITEM_KEY;
1533 log_path = btrfs_alloc_path();
1537 dir = read_one_inode(root, dirid);
1538 /* it isn't an error if the inode isn't there, that can happen
1539 * because we replay the deletes before we copy in the inode item
1543 btrfs_free_path(log_path);
1551 range_end = (u64)-1;
1553 ret = find_dir_range(log, path, dirid, key_type,
1554 &range_start, &range_end);
1559 dir_key.offset = range_start;
1562 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1567 nritems = btrfs_header_nritems(path->nodes[0]);
1568 if (path->slots[0] >= nritems) {
1569 ret = btrfs_next_leaf(root, path);
1573 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1575 if (found_key.objectid != dirid ||
1576 found_key.type != dir_key.type)
1579 if (found_key.offset > range_end)
1582 ret = check_item_in_log(trans, root, log, path,
1586 if (found_key.offset == (u64)-1)
1588 dir_key.offset = found_key.offset + 1;
1590 btrfs_release_path(path);
1591 if (range_end == (u64)-1)
1593 range_start = range_end + 1;
1598 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1599 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1600 dir_key.type = BTRFS_DIR_INDEX_KEY;
1601 btrfs_release_path(path);
1605 btrfs_release_path(path);
1606 btrfs_free_path(log_path);
1612 * the process_func used to replay items from the log tree. This
1613 * gets called in two different stages. The first stage just looks
1614 * for inodes and makes sure they are all copied into the subvolume.
1616 * The second stage copies all the other item types from the log into
1617 * the subvolume. The two stage approach is slower, but gets rid of
1618 * lots of complexity around inodes referencing other inodes that exist
1619 * only in the log (references come from either directory items or inode
1622 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1623 struct walk_control *wc, u64 gen)
1626 struct btrfs_path *path;
1627 struct btrfs_root *root = wc->replay_dest;
1628 struct btrfs_key key;
1633 btrfs_read_buffer(eb, gen);
1635 level = btrfs_header_level(eb);
1640 path = btrfs_alloc_path();
1644 nritems = btrfs_header_nritems(eb);
1645 for (i = 0; i < nritems; i++) {
1646 btrfs_item_key_to_cpu(eb, &key, i);
1648 /* inode keys are done during the first stage */
1649 if (key.type == BTRFS_INODE_ITEM_KEY &&
1650 wc->stage == LOG_WALK_REPLAY_INODES) {
1651 struct btrfs_inode_item *inode_item;
1654 inode_item = btrfs_item_ptr(eb, i,
1655 struct btrfs_inode_item);
1656 mode = btrfs_inode_mode(eb, inode_item);
1657 if (S_ISDIR(mode)) {
1658 ret = replay_dir_deletes(wc->trans,
1659 root, log, path, key.objectid, 0);
1662 ret = overwrite_item(wc->trans, root, path,
1666 /* for regular files, make sure corresponding
1667 * orhpan item exist. extents past the new EOF
1668 * will be truncated later by orphan cleanup.
1670 if (S_ISREG(mode)) {
1671 ret = insert_orphan_item(wc->trans, root,
1676 ret = link_to_fixup_dir(wc->trans, root,
1677 path, key.objectid);
1680 if (wc->stage < LOG_WALK_REPLAY_ALL)
1683 /* these keys are simply copied */
1684 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1685 ret = overwrite_item(wc->trans, root, path,
1688 } else if (key.type == BTRFS_INODE_REF_KEY) {
1689 ret = add_inode_ref(wc->trans, root, log, path,
1691 BUG_ON(ret && ret != -ENOENT);
1692 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1693 ret = replay_one_extent(wc->trans, root, path,
1696 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1697 key.type == BTRFS_DIR_INDEX_KEY) {
1698 ret = replay_one_dir_item(wc->trans, root, path,
1703 btrfs_free_path(path);
1707 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1708 struct btrfs_root *root,
1709 struct btrfs_path *path, int *level,
1710 struct walk_control *wc)
1715 struct extent_buffer *next;
1716 struct extent_buffer *cur;
1717 struct extent_buffer *parent;
1721 WARN_ON(*level < 0);
1722 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1724 while (*level > 0) {
1725 WARN_ON(*level < 0);
1726 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1727 cur = path->nodes[*level];
1729 if (btrfs_header_level(cur) != *level)
1732 if (path->slots[*level] >=
1733 btrfs_header_nritems(cur))
1736 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1737 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1738 blocksize = btrfs_level_size(root, *level - 1);
1740 parent = path->nodes[*level];
1741 root_owner = btrfs_header_owner(parent);
1743 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1748 ret = wc->process_func(root, next, wc, ptr_gen);
1752 path->slots[*level]++;
1754 btrfs_read_buffer(next, ptr_gen);
1756 btrfs_tree_lock(next);
1757 btrfs_set_lock_blocking(next);
1758 clean_tree_block(trans, root, next);
1759 btrfs_wait_tree_block_writeback(next);
1760 btrfs_tree_unlock(next);
1762 WARN_ON(root_owner !=
1763 BTRFS_TREE_LOG_OBJECTID);
1764 ret = btrfs_free_and_pin_reserved_extent(root,
1768 free_extent_buffer(next);
1771 btrfs_read_buffer(next, ptr_gen);
1773 WARN_ON(*level <= 0);
1774 if (path->nodes[*level-1])
1775 free_extent_buffer(path->nodes[*level-1]);
1776 path->nodes[*level-1] = next;
1777 *level = btrfs_header_level(next);
1778 path->slots[*level] = 0;
1781 WARN_ON(*level < 0);
1782 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1784 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1790 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1791 struct btrfs_root *root,
1792 struct btrfs_path *path, int *level,
1793 struct walk_control *wc)
1800 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1801 slot = path->slots[i];
1802 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1805 WARN_ON(*level == 0);
1808 struct extent_buffer *parent;
1809 if (path->nodes[*level] == root->node)
1810 parent = path->nodes[*level];
1812 parent = path->nodes[*level + 1];
1814 root_owner = btrfs_header_owner(parent);
1815 ret = wc->process_func(root, path->nodes[*level], wc,
1816 btrfs_header_generation(path->nodes[*level]));
1821 struct extent_buffer *next;
1823 next = path->nodes[*level];
1825 btrfs_tree_lock(next);
1826 btrfs_set_lock_blocking(next);
1827 clean_tree_block(trans, root, next);
1828 btrfs_wait_tree_block_writeback(next);
1829 btrfs_tree_unlock(next);
1831 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1832 ret = btrfs_free_and_pin_reserved_extent(root,
1833 path->nodes[*level]->start,
1834 path->nodes[*level]->len);
1837 free_extent_buffer(path->nodes[*level]);
1838 path->nodes[*level] = NULL;
1846 * drop the reference count on the tree rooted at 'snap'. This traverses
1847 * the tree freeing any blocks that have a ref count of zero after being
1850 static int walk_log_tree(struct btrfs_trans_handle *trans,
1851 struct btrfs_root *log, struct walk_control *wc)
1856 struct btrfs_path *path;
1860 path = btrfs_alloc_path();
1864 level = btrfs_header_level(log->node);
1866 path->nodes[level] = log->node;
1867 extent_buffer_get(log->node);
1868 path->slots[level] = 0;
1871 wret = walk_down_log_tree(trans, log, path, &level, wc);
1877 wret = walk_up_log_tree(trans, log, path, &level, wc);
1884 /* was the root node processed? if not, catch it here */
1885 if (path->nodes[orig_level]) {
1886 wc->process_func(log, path->nodes[orig_level], wc,
1887 btrfs_header_generation(path->nodes[orig_level]));
1889 struct extent_buffer *next;
1891 next = path->nodes[orig_level];
1893 btrfs_tree_lock(next);
1894 btrfs_set_lock_blocking(next);
1895 clean_tree_block(trans, log, next);
1896 btrfs_wait_tree_block_writeback(next);
1897 btrfs_tree_unlock(next);
1899 WARN_ON(log->root_key.objectid !=
1900 BTRFS_TREE_LOG_OBJECTID);
1901 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
1907 for (i = 0; i <= orig_level; i++) {
1908 if (path->nodes[i]) {
1909 free_extent_buffer(path->nodes[i]);
1910 path->nodes[i] = NULL;
1913 btrfs_free_path(path);
1918 * helper function to update the item for a given subvolumes log root
1919 * in the tree of log roots
1921 static int update_log_root(struct btrfs_trans_handle *trans,
1922 struct btrfs_root *log)
1926 if (log->log_transid == 1) {
1927 /* insert root item on the first sync */
1928 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1929 &log->root_key, &log->root_item);
1931 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1932 &log->root_key, &log->root_item);
1937 static int wait_log_commit(struct btrfs_trans_handle *trans,
1938 struct btrfs_root *root, unsigned long transid)
1941 int index = transid % 2;
1944 * we only allow two pending log transactions at a time,
1945 * so we know that if ours is more than 2 older than the
1946 * current transaction, we're done
1949 prepare_to_wait(&root->log_commit_wait[index],
1950 &wait, TASK_UNINTERRUPTIBLE);
1951 mutex_unlock(&root->log_mutex);
1953 if (root->fs_info->last_trans_log_full_commit !=
1954 trans->transid && root->log_transid < transid + 2 &&
1955 atomic_read(&root->log_commit[index]))
1958 finish_wait(&root->log_commit_wait[index], &wait);
1959 mutex_lock(&root->log_mutex);
1960 } while (root->log_transid < transid + 2 &&
1961 atomic_read(&root->log_commit[index]));
1965 static int wait_for_writer(struct btrfs_trans_handle *trans,
1966 struct btrfs_root *root)
1969 while (atomic_read(&root->log_writers)) {
1970 prepare_to_wait(&root->log_writer_wait,
1971 &wait, TASK_UNINTERRUPTIBLE);
1972 mutex_unlock(&root->log_mutex);
1973 if (root->fs_info->last_trans_log_full_commit !=
1974 trans->transid && atomic_read(&root->log_writers))
1976 mutex_lock(&root->log_mutex);
1977 finish_wait(&root->log_writer_wait, &wait);
1983 * btrfs_sync_log does sends a given tree log down to the disk and
1984 * updates the super blocks to record it. When this call is done,
1985 * you know that any inodes previously logged are safely on disk only
1988 * Any other return value means you need to call btrfs_commit_transaction.
1989 * Some of the edge cases for fsyncing directories that have had unlinks
1990 * or renames done in the past mean that sometimes the only safe
1991 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
1992 * that has happened.
1994 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1995 struct btrfs_root *root)
2001 struct btrfs_root *log = root->log_root;
2002 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2003 unsigned long log_transid = 0;
2005 mutex_lock(&root->log_mutex);
2006 index1 = root->log_transid % 2;
2007 if (atomic_read(&root->log_commit[index1])) {
2008 wait_log_commit(trans, root, root->log_transid);
2009 mutex_unlock(&root->log_mutex);
2012 atomic_set(&root->log_commit[index1], 1);
2014 /* wait for previous tree log sync to complete */
2015 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2016 wait_log_commit(trans, root, root->log_transid - 1);
2018 unsigned long batch = root->log_batch;
2019 /* when we're on an ssd, just kick the log commit out */
2020 if (!btrfs_test_opt(root, SSD) && root->log_multiple_pids) {
2021 mutex_unlock(&root->log_mutex);
2022 schedule_timeout_uninterruptible(1);
2023 mutex_lock(&root->log_mutex);
2025 wait_for_writer(trans, root);
2026 if (batch == root->log_batch)
2030 /* bail out if we need to do a full commit */
2031 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2033 mutex_unlock(&root->log_mutex);
2037 log_transid = root->log_transid;
2038 if (log_transid % 2 == 0)
2039 mark = EXTENT_DIRTY;
2043 /* we start IO on all the marked extents here, but we don't actually
2044 * wait for them until later.
2046 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2049 btrfs_set_root_node(&log->root_item, log->node);
2051 root->log_batch = 0;
2052 root->log_transid++;
2053 log->log_transid = root->log_transid;
2054 root->log_start_pid = 0;
2057 * IO has been started, blocks of the log tree have WRITTEN flag set
2058 * in their headers. new modifications of the log will be written to
2059 * new positions. so it's safe to allow log writers to go in.
2061 mutex_unlock(&root->log_mutex);
2063 mutex_lock(&log_root_tree->log_mutex);
2064 log_root_tree->log_batch++;
2065 atomic_inc(&log_root_tree->log_writers);
2066 mutex_unlock(&log_root_tree->log_mutex);
2068 ret = update_log_root(trans, log);
2070 mutex_lock(&log_root_tree->log_mutex);
2071 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2073 if (waitqueue_active(&log_root_tree->log_writer_wait))
2074 wake_up(&log_root_tree->log_writer_wait);
2078 BUG_ON(ret != -ENOSPC);
2079 root->fs_info->last_trans_log_full_commit = trans->transid;
2080 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2081 mutex_unlock(&log_root_tree->log_mutex);
2086 index2 = log_root_tree->log_transid % 2;
2087 if (atomic_read(&log_root_tree->log_commit[index2])) {
2088 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2089 wait_log_commit(trans, log_root_tree,
2090 log_root_tree->log_transid);
2091 mutex_unlock(&log_root_tree->log_mutex);
2095 atomic_set(&log_root_tree->log_commit[index2], 1);
2097 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2098 wait_log_commit(trans, log_root_tree,
2099 log_root_tree->log_transid - 1);
2102 wait_for_writer(trans, log_root_tree);
2105 * now that we've moved on to the tree of log tree roots,
2106 * check the full commit flag again
2108 if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2109 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2110 mutex_unlock(&log_root_tree->log_mutex);
2112 goto out_wake_log_root;
2115 ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2116 &log_root_tree->dirty_log_pages,
2117 EXTENT_DIRTY | EXTENT_NEW);
2119 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2121 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2122 log_root_tree->node->start);
2123 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2124 btrfs_header_level(log_root_tree->node));
2126 log_root_tree->log_batch = 0;
2127 log_root_tree->log_transid++;
2130 mutex_unlock(&log_root_tree->log_mutex);
2133 * nobody else is going to jump in and write the the ctree
2134 * super here because the log_commit atomic below is protecting
2135 * us. We must be called with a transaction handle pinning
2136 * the running transaction open, so a full commit can't hop
2137 * in and cause problems either.
2139 btrfs_scrub_pause_super(root);
2140 write_ctree_super(trans, root->fs_info->tree_root, 1);
2141 btrfs_scrub_continue_super(root);
2144 mutex_lock(&root->log_mutex);
2145 if (root->last_log_commit < log_transid)
2146 root->last_log_commit = log_transid;
2147 mutex_unlock(&root->log_mutex);
2150 atomic_set(&log_root_tree->log_commit[index2], 0);
2152 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2153 wake_up(&log_root_tree->log_commit_wait[index2]);
2155 atomic_set(&root->log_commit[index1], 0);
2157 if (waitqueue_active(&root->log_commit_wait[index1]))
2158 wake_up(&root->log_commit_wait[index1]);
2162 static void free_log_tree(struct btrfs_trans_handle *trans,
2163 struct btrfs_root *log)
2168 struct walk_control wc = {
2170 .process_func = process_one_buffer
2173 ret = walk_log_tree(trans, log, &wc);
2177 ret = find_first_extent_bit(&log->dirty_log_pages,
2178 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2182 clear_extent_bits(&log->dirty_log_pages, start, end,
2183 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2186 free_extent_buffer(log->node);
2191 * free all the extents used by the tree log. This should be called
2192 * at commit time of the full transaction
2194 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2196 if (root->log_root) {
2197 free_log_tree(trans, root->log_root);
2198 root->log_root = NULL;
2203 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2204 struct btrfs_fs_info *fs_info)
2206 if (fs_info->log_root_tree) {
2207 free_log_tree(trans, fs_info->log_root_tree);
2208 fs_info->log_root_tree = NULL;
2214 * If both a file and directory are logged, and unlinks or renames are
2215 * mixed in, we have a few interesting corners:
2217 * create file X in dir Y
2218 * link file X to X.link in dir Y
2220 * unlink file X but leave X.link
2223 * After a crash we would expect only X.link to exist. But file X
2224 * didn't get fsync'd again so the log has back refs for X and X.link.
2226 * We solve this by removing directory entries and inode backrefs from the
2227 * log when a file that was logged in the current transaction is
2228 * unlinked. Any later fsync will include the updated log entries, and
2229 * we'll be able to reconstruct the proper directory items from backrefs.
2231 * This optimizations allows us to avoid relogging the entire inode
2232 * or the entire directory.
2234 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root,
2236 const char *name, int name_len,
2237 struct inode *dir, u64 index)
2239 struct btrfs_root *log;
2240 struct btrfs_dir_item *di;
2241 struct btrfs_path *path;
2245 u64 dir_ino = btrfs_ino(dir);
2247 if (BTRFS_I(dir)->logged_trans < trans->transid)
2250 ret = join_running_log_trans(root);
2254 mutex_lock(&BTRFS_I(dir)->log_mutex);
2256 log = root->log_root;
2257 path = btrfs_alloc_path();
2263 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2264 name, name_len, -1);
2270 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2271 bytes_del += name_len;
2274 btrfs_release_path(path);
2275 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2276 index, name, name_len, -1);
2282 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2283 bytes_del += name_len;
2287 /* update the directory size in the log to reflect the names
2291 struct btrfs_key key;
2293 key.objectid = dir_ino;
2295 key.type = BTRFS_INODE_ITEM_KEY;
2296 btrfs_release_path(path);
2298 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2304 struct btrfs_inode_item *item;
2307 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2308 struct btrfs_inode_item);
2309 i_size = btrfs_inode_size(path->nodes[0], item);
2310 if (i_size > bytes_del)
2311 i_size -= bytes_del;
2314 btrfs_set_inode_size(path->nodes[0], item, i_size);
2315 btrfs_mark_buffer_dirty(path->nodes[0]);
2318 btrfs_release_path(path);
2321 btrfs_free_path(path);
2323 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2324 if (ret == -ENOSPC) {
2325 root->fs_info->last_trans_log_full_commit = trans->transid;
2328 btrfs_end_log_trans(root);
2333 /* see comments for btrfs_del_dir_entries_in_log */
2334 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2335 struct btrfs_root *root,
2336 const char *name, int name_len,
2337 struct inode *inode, u64 dirid)
2339 struct btrfs_root *log;
2343 if (BTRFS_I(inode)->logged_trans < trans->transid)
2346 ret = join_running_log_trans(root);
2349 log = root->log_root;
2350 mutex_lock(&BTRFS_I(inode)->log_mutex);
2352 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2354 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2355 if (ret == -ENOSPC) {
2356 root->fs_info->last_trans_log_full_commit = trans->transid;
2359 btrfs_end_log_trans(root);
2365 * creates a range item in the log for 'dirid'. first_offset and
2366 * last_offset tell us which parts of the key space the log should
2367 * be considered authoritative for.
2369 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2370 struct btrfs_root *log,
2371 struct btrfs_path *path,
2372 int key_type, u64 dirid,
2373 u64 first_offset, u64 last_offset)
2376 struct btrfs_key key;
2377 struct btrfs_dir_log_item *item;
2379 key.objectid = dirid;
2380 key.offset = first_offset;
2381 if (key_type == BTRFS_DIR_ITEM_KEY)
2382 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2384 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2385 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2389 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2390 struct btrfs_dir_log_item);
2391 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2392 btrfs_mark_buffer_dirty(path->nodes[0]);
2393 btrfs_release_path(path);
2398 * log all the items included in the current transaction for a given
2399 * directory. This also creates the range items in the log tree required
2400 * to replay anything deleted before the fsync
2402 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2403 struct btrfs_root *root, struct inode *inode,
2404 struct btrfs_path *path,
2405 struct btrfs_path *dst_path, int key_type,
2406 u64 min_offset, u64 *last_offset_ret)
2408 struct btrfs_key min_key;
2409 struct btrfs_key max_key;
2410 struct btrfs_root *log = root->log_root;
2411 struct extent_buffer *src;
2416 u64 first_offset = min_offset;
2417 u64 last_offset = (u64)-1;
2418 u64 ino = btrfs_ino(inode);
2420 log = root->log_root;
2421 max_key.objectid = ino;
2422 max_key.offset = (u64)-1;
2423 max_key.type = key_type;
2425 min_key.objectid = ino;
2426 min_key.type = key_type;
2427 min_key.offset = min_offset;
2429 path->keep_locks = 1;
2431 ret = btrfs_search_forward(root, &min_key, &max_key,
2432 path, 0, trans->transid);
2435 * we didn't find anything from this transaction, see if there
2436 * is anything at all
2438 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
2439 min_key.objectid = ino;
2440 min_key.type = key_type;
2441 min_key.offset = (u64)-1;
2442 btrfs_release_path(path);
2443 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2445 btrfs_release_path(path);
2448 ret = btrfs_previous_item(root, path, ino, key_type);
2450 /* if ret == 0 there are items for this type,
2451 * create a range to tell us the last key of this type.
2452 * otherwise, there are no items in this directory after
2453 * *min_offset, and we create a range to indicate that.
2456 struct btrfs_key tmp;
2457 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2459 if (key_type == tmp.type)
2460 first_offset = max(min_offset, tmp.offset) + 1;
2465 /* go backward to find any previous key */
2466 ret = btrfs_previous_item(root, path, ino, key_type);
2468 struct btrfs_key tmp;
2469 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2470 if (key_type == tmp.type) {
2471 first_offset = tmp.offset;
2472 ret = overwrite_item(trans, log, dst_path,
2473 path->nodes[0], path->slots[0],
2481 btrfs_release_path(path);
2483 /* find the first key from this transaction again */
2484 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2491 * we have a block from this transaction, log every item in it
2492 * from our directory
2495 struct btrfs_key tmp;
2496 src = path->nodes[0];
2497 nritems = btrfs_header_nritems(src);
2498 for (i = path->slots[0]; i < nritems; i++) {
2499 btrfs_item_key_to_cpu(src, &min_key, i);
2501 if (min_key.objectid != ino || min_key.type != key_type)
2503 ret = overwrite_item(trans, log, dst_path, src, i,
2510 path->slots[0] = nritems;
2513 * look ahead to the next item and see if it is also
2514 * from this directory and from this transaction
2516 ret = btrfs_next_leaf(root, path);
2518 last_offset = (u64)-1;
2521 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2522 if (tmp.objectid != ino || tmp.type != key_type) {
2523 last_offset = (u64)-1;
2526 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2527 ret = overwrite_item(trans, log, dst_path,
2528 path->nodes[0], path->slots[0],
2533 last_offset = tmp.offset;
2538 btrfs_release_path(path);
2539 btrfs_release_path(dst_path);
2542 *last_offset_ret = last_offset;
2544 * insert the log range keys to indicate where the log
2547 ret = insert_dir_log_key(trans, log, path, key_type,
2548 ino, first_offset, last_offset);
2556 * logging directories is very similar to logging inodes, We find all the items
2557 * from the current transaction and write them to the log.
2559 * The recovery code scans the directory in the subvolume, and if it finds a
2560 * key in the range logged that is not present in the log tree, then it means
2561 * that dir entry was unlinked during the transaction.
2563 * In order for that scan to work, we must include one key smaller than
2564 * the smallest logged by this transaction and one key larger than the largest
2565 * key logged by this transaction.
2567 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2568 struct btrfs_root *root, struct inode *inode,
2569 struct btrfs_path *path,
2570 struct btrfs_path *dst_path)
2575 int key_type = BTRFS_DIR_ITEM_KEY;
2581 ret = log_dir_items(trans, root, inode, path,
2582 dst_path, key_type, min_key,
2586 if (max_key == (u64)-1)
2588 min_key = max_key + 1;
2591 if (key_type == BTRFS_DIR_ITEM_KEY) {
2592 key_type = BTRFS_DIR_INDEX_KEY;
2599 * a helper function to drop items from the log before we relog an
2600 * inode. max_key_type indicates the highest item type to remove.
2601 * This cannot be run for file data extents because it does not
2602 * free the extents they point to.
2604 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2605 struct btrfs_root *log,
2606 struct btrfs_path *path,
2607 u64 objectid, int max_key_type)
2610 struct btrfs_key key;
2611 struct btrfs_key found_key;
2613 key.objectid = objectid;
2614 key.type = max_key_type;
2615 key.offset = (u64)-1;
2618 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2623 if (path->slots[0] == 0)
2627 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2630 if (found_key.objectid != objectid)
2633 ret = btrfs_del_item(trans, log, path);
2636 btrfs_release_path(path);
2638 btrfs_release_path(path);
2642 static noinline int copy_items(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *log,
2644 struct btrfs_path *dst_path,
2645 struct extent_buffer *src,
2646 int start_slot, int nr, int inode_only)
2648 unsigned long src_offset;
2649 unsigned long dst_offset;
2650 struct btrfs_file_extent_item *extent;
2651 struct btrfs_inode_item *inode_item;
2653 struct btrfs_key *ins_keys;
2657 struct list_head ordered_sums;
2659 INIT_LIST_HEAD(&ordered_sums);
2661 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2662 nr * sizeof(u32), GFP_NOFS);
2666 ins_sizes = (u32 *)ins_data;
2667 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2669 for (i = 0; i < nr; i++) {
2670 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2671 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2673 ret = btrfs_insert_empty_items(trans, log, dst_path,
2674 ins_keys, ins_sizes, nr);
2680 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2681 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2682 dst_path->slots[0]);
2684 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2686 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2687 src_offset, ins_sizes[i]);
2689 if (inode_only == LOG_INODE_EXISTS &&
2690 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2691 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2693 struct btrfs_inode_item);
2694 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2696 /* set the generation to zero so the recover code
2697 * can tell the difference between an logging
2698 * just to say 'this inode exists' and a logging
2699 * to say 'update this inode with these values'
2701 btrfs_set_inode_generation(dst_path->nodes[0],
2704 /* take a reference on file data extents so that truncates
2705 * or deletes of this inode don't have to relog the inode
2708 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2710 extent = btrfs_item_ptr(src, start_slot + i,
2711 struct btrfs_file_extent_item);
2713 if (btrfs_file_extent_generation(src, extent) < trans->transid)
2716 found_type = btrfs_file_extent_type(src, extent);
2717 if (found_type == BTRFS_FILE_EXTENT_REG ||
2718 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2720 ds = btrfs_file_extent_disk_bytenr(src,
2722 /* ds == 0 is a hole */
2726 dl = btrfs_file_extent_disk_num_bytes(src,
2728 cs = btrfs_file_extent_offset(src, extent);
2729 cl = btrfs_file_extent_num_bytes(src,
2731 if (btrfs_file_extent_compression(src,
2737 ret = btrfs_lookup_csums_range(
2738 log->fs_info->csum_root,
2739 ds + cs, ds + cs + cl - 1,
2746 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2747 btrfs_release_path(dst_path);
2751 * we have to do this after the loop above to avoid changing the
2752 * log tree while trying to change the log tree.
2755 while (!list_empty(&ordered_sums)) {
2756 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2757 struct btrfs_ordered_sum,
2760 ret = btrfs_csum_file_blocks(trans, log, sums);
2761 list_del(&sums->list);
2767 /* log a single inode in the tree log.
2768 * At least one parent directory for this inode must exist in the tree
2769 * or be logged already.
2771 * Any items from this inode changed by the current transaction are copied
2772 * to the log tree. An extra reference is taken on any extents in this
2773 * file, allowing us to avoid a whole pile of corner cases around logging
2774 * blocks that have been removed from the tree.
2776 * See LOG_INODE_ALL and related defines for a description of what inode_only
2779 * This handles both files and directories.
2781 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2782 struct btrfs_root *root, struct inode *inode,
2785 struct btrfs_path *path;
2786 struct btrfs_path *dst_path;
2787 struct btrfs_key min_key;
2788 struct btrfs_key max_key;
2789 struct btrfs_root *log = root->log_root;
2790 struct extent_buffer *src = NULL;
2794 int ins_start_slot = 0;
2796 u64 ino = btrfs_ino(inode);
2798 log = root->log_root;
2800 path = btrfs_alloc_path();
2803 dst_path = btrfs_alloc_path();
2805 btrfs_free_path(path);
2809 min_key.objectid = ino;
2810 min_key.type = BTRFS_INODE_ITEM_KEY;
2813 max_key.objectid = ino;
2815 /* today the code can only do partial logging of directories */
2816 if (!S_ISDIR(inode->i_mode))
2817 inode_only = LOG_INODE_ALL;
2819 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2820 max_key.type = BTRFS_XATTR_ITEM_KEY;
2822 max_key.type = (u8)-1;
2823 max_key.offset = (u64)-1;
2825 ret = btrfs_commit_inode_delayed_items(trans, inode);
2827 btrfs_free_path(path);
2828 btrfs_free_path(dst_path);
2832 mutex_lock(&BTRFS_I(inode)->log_mutex);
2835 * a brute force approach to making sure we get the most uptodate
2836 * copies of everything.
2838 if (S_ISDIR(inode->i_mode)) {
2839 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2841 if (inode_only == LOG_INODE_EXISTS)
2842 max_key_type = BTRFS_XATTR_ITEM_KEY;
2843 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
2845 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2851 path->keep_locks = 1;
2855 ret = btrfs_search_forward(root, &min_key, &max_key,
2856 path, 0, trans->transid);
2860 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2861 if (min_key.objectid != ino)
2863 if (min_key.type > max_key.type)
2866 src = path->nodes[0];
2867 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2870 } else if (!ins_nr) {
2871 ins_start_slot = path->slots[0];
2876 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2877 ins_nr, inode_only);
2883 ins_start_slot = path->slots[0];
2886 nritems = btrfs_header_nritems(path->nodes[0]);
2888 if (path->slots[0] < nritems) {
2889 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2894 ret = copy_items(trans, log, dst_path, src,
2896 ins_nr, inode_only);
2903 btrfs_release_path(path);
2905 if (min_key.offset < (u64)-1)
2907 else if (min_key.type < (u8)-1)
2909 else if (min_key.objectid < (u64)-1)
2915 ret = copy_items(trans, log, dst_path, src,
2917 ins_nr, inode_only);
2925 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2926 btrfs_release_path(path);
2927 btrfs_release_path(dst_path);
2928 ret = log_directory_changes(trans, root, inode, path, dst_path);
2934 BTRFS_I(inode)->logged_trans = trans->transid;
2936 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2938 btrfs_free_path(path);
2939 btrfs_free_path(dst_path);
2944 * follow the dentry parent pointers up the chain and see if any
2945 * of the directories in it require a full commit before they can
2946 * be logged. Returns zero if nothing special needs to be done or 1 if
2947 * a full commit is required.
2949 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2950 struct inode *inode,
2951 struct dentry *parent,
2952 struct super_block *sb,
2956 struct btrfs_root *root;
2957 struct dentry *old_parent = NULL;
2960 * for regular files, if its inode is already on disk, we don't
2961 * have to worry about the parents at all. This is because
2962 * we can use the last_unlink_trans field to record renames
2963 * and other fun in this file.
2965 if (S_ISREG(inode->i_mode) &&
2966 BTRFS_I(inode)->generation <= last_committed &&
2967 BTRFS_I(inode)->last_unlink_trans <= last_committed)
2970 if (!S_ISDIR(inode->i_mode)) {
2971 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2973 inode = parent->d_inode;
2977 BTRFS_I(inode)->logged_trans = trans->transid;
2980 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2981 root = BTRFS_I(inode)->root;
2984 * make sure any commits to the log are forced
2985 * to be full commits
2987 root->fs_info->last_trans_log_full_commit =
2993 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2996 if (IS_ROOT(parent))
2999 parent = dget_parent(parent);
3001 old_parent = parent;
3002 inode = parent->d_inode;
3010 static int inode_in_log(struct btrfs_trans_handle *trans,
3011 struct inode *inode)
3013 struct btrfs_root *root = BTRFS_I(inode)->root;
3016 mutex_lock(&root->log_mutex);
3017 if (BTRFS_I(inode)->logged_trans == trans->transid &&
3018 BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
3020 mutex_unlock(&root->log_mutex);
3026 * helper function around btrfs_log_inode to make sure newly created
3027 * parent directories also end up in the log. A minimal inode and backref
3028 * only logging is done of any parent directories that are older than
3029 * the last committed transaction
3031 int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
3032 struct btrfs_root *root, struct inode *inode,
3033 struct dentry *parent, int exists_only)
3035 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
3036 struct super_block *sb;
3037 struct dentry *old_parent = NULL;
3039 u64 last_committed = root->fs_info->last_trans_committed;
3043 if (btrfs_test_opt(root, NOTREELOG)) {
3048 if (root->fs_info->last_trans_log_full_commit >
3049 root->fs_info->last_trans_committed) {
3054 if (root != BTRFS_I(inode)->root ||
3055 btrfs_root_refs(&root->root_item) == 0) {
3060 ret = check_parent_dirs_for_sync(trans, inode, parent,
3061 sb, last_committed);
3065 if (inode_in_log(trans, inode)) {
3066 ret = BTRFS_NO_LOG_SYNC;
3070 ret = start_log_trans(trans, root);
3074 ret = btrfs_log_inode(trans, root, inode, inode_only);
3079 * for regular files, if its inode is already on disk, we don't
3080 * have to worry about the parents at all. This is because
3081 * we can use the last_unlink_trans field to record renames
3082 * and other fun in this file.
3084 if (S_ISREG(inode->i_mode) &&
3085 BTRFS_I(inode)->generation <= last_committed &&
3086 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3091 inode_only = LOG_INODE_EXISTS;
3093 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3096 inode = parent->d_inode;
3097 if (root != BTRFS_I(inode)->root)
3100 if (BTRFS_I(inode)->generation >
3101 root->fs_info->last_trans_committed) {
3102 ret = btrfs_log_inode(trans, root, inode, inode_only);
3106 if (IS_ROOT(parent))
3109 parent = dget_parent(parent);
3111 old_parent = parent;
3117 BUG_ON(ret != -ENOSPC);
3118 root->fs_info->last_trans_log_full_commit = trans->transid;
3121 btrfs_end_log_trans(root);
3127 * it is not safe to log dentry if the chunk root has added new
3128 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
3129 * If this returns 1, you must commit the transaction to safely get your
3132 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3133 struct btrfs_root *root, struct dentry *dentry)
3135 struct dentry *parent = dget_parent(dentry);
3138 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent, 0);
3145 * should be called during mount to recover any replay any log trees
3148 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3151 struct btrfs_path *path;
3152 struct btrfs_trans_handle *trans;
3153 struct btrfs_key key;
3154 struct btrfs_key found_key;
3155 struct btrfs_key tmp_key;
3156 struct btrfs_root *log;
3157 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3158 struct walk_control wc = {
3159 .process_func = process_one_buffer,
3163 path = btrfs_alloc_path();
3167 fs_info->log_root_recovering = 1;
3169 trans = btrfs_start_transaction(fs_info->tree_root, 0);
3170 BUG_ON(IS_ERR(trans));
3175 ret = walk_log_tree(trans, log_root_tree, &wc);
3179 key.objectid = BTRFS_TREE_LOG_OBJECTID;
3180 key.offset = (u64)-1;
3181 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3184 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3188 if (path->slots[0] == 0)
3192 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3194 btrfs_release_path(path);
3195 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3198 log = btrfs_read_fs_root_no_radix(log_root_tree,
3200 BUG_ON(IS_ERR(log));
3202 tmp_key.objectid = found_key.offset;
3203 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3204 tmp_key.offset = (u64)-1;
3206 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3207 BUG_ON(IS_ERR_OR_NULL(wc.replay_dest));
3209 wc.replay_dest->log_root = log;
3210 btrfs_record_root_in_trans(trans, wc.replay_dest);
3211 ret = walk_log_tree(trans, log, &wc);
3214 if (wc.stage == LOG_WALK_REPLAY_ALL) {
3215 ret = fixup_inode_link_counts(trans, wc.replay_dest,
3220 key.offset = found_key.offset - 1;
3221 wc.replay_dest->log_root = NULL;
3222 free_extent_buffer(log->node);
3223 free_extent_buffer(log->commit_root);
3226 if (found_key.offset == 0)
3229 btrfs_release_path(path);
3231 /* step one is to pin it all, step two is to replay just inodes */
3234 wc.process_func = replay_one_buffer;
3235 wc.stage = LOG_WALK_REPLAY_INODES;
3238 /* step three is to replay everything */
3239 if (wc.stage < LOG_WALK_REPLAY_ALL) {
3244 btrfs_free_path(path);
3246 free_extent_buffer(log_root_tree->node);
3247 log_root_tree->log_root = NULL;
3248 fs_info->log_root_recovering = 0;
3250 /* step 4: commit the transaction, which also unpins the blocks */
3251 btrfs_commit_transaction(trans, fs_info->tree_root);
3253 kfree(log_root_tree);
3258 * there are some corner cases where we want to force a full
3259 * commit instead of allowing a directory to be logged.
3261 * They revolve around files there were unlinked from the directory, and
3262 * this function updates the parent directory so that a full commit is
3263 * properly done if it is fsync'd later after the unlinks are done.
3265 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3266 struct inode *dir, struct inode *inode,
3270 * when we're logging a file, if it hasn't been renamed
3271 * or unlinked, and its inode is fully committed on disk,
3272 * we don't have to worry about walking up the directory chain
3273 * to log its parents.
3275 * So, we use the last_unlink_trans field to put this transid
3276 * into the file. When the file is logged we check it and
3277 * don't log the parents if the file is fully on disk.
3279 if (S_ISREG(inode->i_mode))
3280 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3283 * if this directory was already logged any new
3284 * names for this file/dir will get recorded
3287 if (BTRFS_I(dir)->logged_trans == trans->transid)
3291 * if the inode we're about to unlink was logged,
3292 * the log will be properly updated for any new names
3294 if (BTRFS_I(inode)->logged_trans == trans->transid)
3298 * when renaming files across directories, if the directory
3299 * there we're unlinking from gets fsync'd later on, there's
3300 * no way to find the destination directory later and fsync it
3301 * properly. So, we have to be conservative and force commits
3302 * so the new name gets discovered.
3307 /* we can safely do the unlink without any special recording */
3311 BTRFS_I(dir)->last_unlink_trans = trans->transid;
3315 * Call this after adding a new name for a file and it will properly
3316 * update the log to reflect the new name.
3318 * It will return zero if all goes well, and it will return 1 if a
3319 * full transaction commit is required.
3321 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3322 struct inode *inode, struct inode *old_dir,
3323 struct dentry *parent)
3325 struct btrfs_root * root = BTRFS_I(inode)->root;
3328 * this will force the logging code to walk the dentry chain
3331 if (S_ISREG(inode->i_mode))
3332 BTRFS_I(inode)->last_unlink_trans = trans->transid;
3335 * if this inode hasn't been logged and directory we're renaming it
3336 * from hasn't been logged, we don't need to log it
3338 if (BTRFS_I(inode)->logged_trans <=
3339 root->fs_info->last_trans_committed &&
3340 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3341 root->fs_info->last_trans_committed))
3344 return btrfs_log_inode_parent(trans, root, inode, parent, 1);