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>
21 #include "transaction.h"
24 #include "print-tree.h"
28 /* magic values for the inode_only field in btrfs_log_inode:
30 * LOG_INODE_ALL means to log everything
31 * LOG_INODE_EXISTS means to log just enough to recreate the inode
34 #define LOG_INODE_ALL 0
35 #define LOG_INODE_EXISTS 1
38 * stages for the tree walking. The first
39 * stage (0) is to only pin down the blocks we find
40 * the second stage (1) is to make sure that all the inodes
41 * we find in the log are created in the subvolume.
43 * The last stage is to deal with directories and links and extents
44 * and all the other fun semantics
46 #define LOG_WALK_PIN_ONLY 0
47 #define LOG_WALK_REPLAY_INODES 1
48 #define LOG_WALK_REPLAY_ALL 2
50 static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
51 struct btrfs_root *root, struct inode *inode,
55 * tree logging is a special write ahead log used to make sure that
56 * fsyncs and O_SYNCs can happen without doing full tree commits.
58 * Full tree commits are expensive because they require commonly
59 * modified blocks to be recowed, creating many dirty pages in the
60 * extent tree an 4x-6x higher write load than ext3.
62 * Instead of doing a tree commit on every fsync, we use the
63 * key ranges and transaction ids to find items for a given file or directory
64 * that have changed in this transaction. Those items are copied into
65 * a special tree (one per subvolume root), that tree is written to disk
66 * and then the fsync is considered complete.
68 * After a crash, items are copied out of the log-tree back into the
69 * subvolume tree. Any file data extents found are recorded in the extent
70 * allocation tree, and the log-tree freed.
72 * The log tree is read three times, once to pin down all the extents it is
73 * using in ram and once, once to create all the inodes logged in the tree
74 * and once to do all the other items.
78 * btrfs_add_log_tree adds a new per-subvolume log tree into the
79 * tree of log tree roots. This must be called with a tree log transaction
80 * running (see start_log_trans).
82 static int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
83 struct btrfs_root *root)
86 struct btrfs_root_item root_item;
87 struct btrfs_inode_item *inode_item;
88 struct extent_buffer *leaf;
89 struct btrfs_root *new_root = root;
91 u64 objectid = root->root_key.objectid;
93 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
94 BTRFS_TREE_LOG_OBJECTID,
95 trans->transid, 0, 0, 0);
101 btrfs_set_header_nritems(leaf, 0);
102 btrfs_set_header_level(leaf, 0);
103 btrfs_set_header_bytenr(leaf, leaf->start);
104 btrfs_set_header_generation(leaf, trans->transid);
105 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
107 write_extent_buffer(leaf, root->fs_info->fsid,
108 (unsigned long)btrfs_header_fsid(leaf),
110 btrfs_mark_buffer_dirty(leaf);
112 inode_item = &root_item.inode;
113 memset(inode_item, 0, sizeof(*inode_item));
114 inode_item->generation = cpu_to_le64(1);
115 inode_item->size = cpu_to_le64(3);
116 inode_item->nlink = cpu_to_le32(1);
117 inode_item->nbytes = cpu_to_le64(root->leafsize);
118 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
120 btrfs_set_root_bytenr(&root_item, leaf->start);
121 btrfs_set_root_generation(&root_item, trans->transid);
122 btrfs_set_root_level(&root_item, 0);
123 btrfs_set_root_refs(&root_item, 0);
124 btrfs_set_root_used(&root_item, 0);
126 memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
127 root_item.drop_level = 0;
129 btrfs_tree_unlock(leaf);
130 free_extent_buffer(leaf);
133 btrfs_set_root_dirid(&root_item, 0);
135 key.objectid = BTRFS_TREE_LOG_OBJECTID;
136 key.offset = objectid;
137 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
138 ret = btrfs_insert_root(trans, root->fs_info->log_root_tree, &key,
143 new_root = btrfs_read_fs_root_no_radix(root->fs_info->log_root_tree,
147 WARN_ON(root->log_root);
148 root->log_root = new_root;
151 * log trees do not get reference counted because they go away
152 * before a real commit is actually done. They do store pointers
153 * to file data extents, and those reference counts still get
154 * updated (along with back refs to the log tree).
156 new_root->ref_cows = 0;
157 new_root->last_trans = trans->transid;
163 * start a sub transaction and setup the log tree
164 * this increments the log tree writer count to make the people
165 * syncing the tree wait for us to finish
167 static int start_log_trans(struct btrfs_trans_handle *trans,
168 struct btrfs_root *root)
171 mutex_lock(&root->fs_info->tree_log_mutex);
172 if (!root->fs_info->log_root_tree) {
173 ret = btrfs_init_log_root_tree(trans, root->fs_info);
176 if (!root->log_root) {
177 ret = btrfs_add_log_tree(trans, root);
180 atomic_inc(&root->fs_info->tree_log_writers);
181 root->fs_info->tree_log_batch++;
182 mutex_unlock(&root->fs_info->tree_log_mutex);
187 * returns 0 if there was a log transaction running and we were able
188 * to join, or returns -ENOENT if there were not transactions
191 static int join_running_log_trans(struct btrfs_root *root)
199 mutex_lock(&root->fs_info->tree_log_mutex);
200 if (root->log_root) {
202 atomic_inc(&root->fs_info->tree_log_writers);
203 root->fs_info->tree_log_batch++;
205 mutex_unlock(&root->fs_info->tree_log_mutex);
210 * indicate we're done making changes to the log tree
211 * and wake up anyone waiting to do a sync
213 static int end_log_trans(struct btrfs_root *root)
215 atomic_dec(&root->fs_info->tree_log_writers);
217 if (waitqueue_active(&root->fs_info->tree_log_wait))
218 wake_up(&root->fs_info->tree_log_wait);
224 * the walk control struct is used to pass state down the chain when
225 * processing the log tree. The stage field tells us which part
226 * of the log tree processing we are currently doing. The others
227 * are state fields used for that specific part
229 struct walk_control {
230 /* should we free the extent on disk when done? This is used
231 * at transaction commit time while freeing a log tree
235 /* should we write out the extent buffer? This is used
236 * while flushing the log tree to disk during a sync
240 /* should we wait for the extent buffer io to finish? Also used
241 * while flushing the log tree to disk for a sync
245 /* pin only walk, we record which extents on disk belong to the
250 /* what stage of the replay code we're currently in */
253 /* the root we are currently replaying */
254 struct btrfs_root *replay_dest;
256 /* the trans handle for the current replay */
257 struct btrfs_trans_handle *trans;
259 /* the function that gets used to process blocks we find in the
260 * tree. Note the extent_buffer might not be up to date when it is
261 * passed in, and it must be checked or read if you need the data
264 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
265 struct walk_control *wc, u64 gen);
269 * process_func used to pin down extents, write them or wait on them
271 static int process_one_buffer(struct btrfs_root *log,
272 struct extent_buffer *eb,
273 struct walk_control *wc, u64 gen)
276 mutex_lock(&log->fs_info->pinned_mutex);
277 btrfs_update_pinned_extents(log->fs_info->extent_root,
278 eb->start, eb->len, 1);
279 mutex_unlock(&log->fs_info->pinned_mutex);
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);
342 read_extent_buffer(eb, src_copy, src_ptr, item_size);
344 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
345 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
347 ret = memcmp(dst_copy, src_copy, item_size);
352 * they have the same contents, just return, this saves
353 * us from cowing blocks in the destination tree and doing
354 * extra writes that may not have been done by a previous
358 btrfs_release_path(root, path);
364 btrfs_release_path(root, path);
365 /* try to insert the key into the destination tree */
366 ret = btrfs_insert_empty_item(trans, root, path,
369 /* make sure any existing item is the correct size */
370 if (ret == -EEXIST) {
372 found_size = btrfs_item_size_nr(path->nodes[0],
374 if (found_size > item_size) {
375 btrfs_truncate_item(trans, root, path, item_size, 1);
376 } else if (found_size < item_size) {
377 ret = btrfs_del_item(trans, root,
381 btrfs_release_path(root, path);
382 ret = btrfs_insert_empty_item(trans,
383 root, path, key, item_size);
389 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
392 /* don't overwrite an existing inode if the generation number
393 * was logged as zero. This is done when the tree logging code
394 * is just logging an inode to make sure it exists after recovery.
396 * Also, don't overwrite i_size on directories during replay.
397 * log replay inserts and removes directory items based on the
398 * state of the tree found in the subvolume, and i_size is modified
401 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
402 struct btrfs_inode_item *src_item;
403 struct btrfs_inode_item *dst_item;
405 src_item = (struct btrfs_inode_item *)src_ptr;
406 dst_item = (struct btrfs_inode_item *)dst_ptr;
408 if (btrfs_inode_generation(eb, src_item) == 0)
411 if (overwrite_root &&
412 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
413 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
415 saved_i_size = btrfs_inode_size(path->nodes[0],
420 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
423 if (save_old_i_size) {
424 struct btrfs_inode_item *dst_item;
425 dst_item = (struct btrfs_inode_item *)dst_ptr;
426 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
429 /* make sure the generation is filled in */
430 if (key->type == BTRFS_INODE_ITEM_KEY) {
431 struct btrfs_inode_item *dst_item;
432 dst_item = (struct btrfs_inode_item *)dst_ptr;
433 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
434 btrfs_set_inode_generation(path->nodes[0], dst_item,
439 if (overwrite_root &&
440 key->type == BTRFS_EXTENT_DATA_KEY) {
442 struct btrfs_file_extent_item *fi;
444 fi = (struct btrfs_file_extent_item *)dst_ptr;
445 extent_type = btrfs_file_extent_type(path->nodes[0], fi);
446 if (extent_type == BTRFS_FILE_EXTENT_REG ||
447 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
448 struct btrfs_key ins;
449 ins.objectid = btrfs_file_extent_disk_bytenr(
451 ins.offset = btrfs_file_extent_disk_num_bytes(
453 ins.type = BTRFS_EXTENT_ITEM_KEY;
456 * is this extent already allocated in the extent
457 * allocation tree? If so, just add a reference
459 ret = btrfs_lookup_extent(root, ins.objectid,
462 ret = btrfs_inc_extent_ref(trans, root,
463 ins.objectid, ins.offset,
464 path->nodes[0]->start,
465 root->root_key.objectid,
466 trans->transid, key->objectid);
469 * insert the extent pointer in the extent
472 ret = btrfs_alloc_logged_extent(trans, root,
473 path->nodes[0]->start,
474 root->root_key.objectid,
475 trans->transid, key->objectid,
482 btrfs_mark_buffer_dirty(path->nodes[0]);
483 btrfs_release_path(root, path);
488 * simple helper to read an inode off the disk from a given root
489 * This can only be called for subvolume roots and not for the log
491 static noinline struct inode *read_one_inode(struct btrfs_root *root,
495 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
496 if (inode->i_state & I_NEW) {
497 BTRFS_I(inode)->root = root;
498 BTRFS_I(inode)->location.objectid = objectid;
499 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
500 BTRFS_I(inode)->location.offset = 0;
501 btrfs_read_locked_inode(inode);
502 unlock_new_inode(inode);
505 if (is_bad_inode(inode)) {
512 /* replays a single extent in 'eb' at 'slot' with 'key' into the
513 * subvolume 'root'. path is released on entry and should be released
516 * extents in the log tree have not been allocated out of the extent
517 * tree yet. So, this completes the allocation, taking a reference
518 * as required if the extent already exists or creating a new extent
519 * if it isn't in the extent allocation tree yet.
521 * The extent is inserted into the file, dropping any existing extents
522 * from the file that overlap the new one.
524 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
525 struct btrfs_root *root,
526 struct btrfs_path *path,
527 struct extent_buffer *eb, int slot,
528 struct btrfs_key *key)
531 u64 mask = root->sectorsize - 1;
534 u64 start = key->offset;
535 struct btrfs_file_extent_item *item;
536 struct inode *inode = NULL;
540 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
541 found_type = btrfs_file_extent_type(eb, item);
543 if (found_type == BTRFS_FILE_EXTENT_REG ||
544 found_type == BTRFS_FILE_EXTENT_PREALLOC)
545 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
546 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
547 size = btrfs_file_extent_inline_len(eb, item);
548 extent_end = (start + size + mask) & ~mask;
554 inode = read_one_inode(root, key->objectid);
561 * first check to see if we already have this extent in the
562 * file. This must be done before the btrfs_drop_extents run
563 * so we don't try to drop this extent.
565 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
569 (found_type == BTRFS_FILE_EXTENT_REG ||
570 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
571 struct btrfs_file_extent_item cmp1;
572 struct btrfs_file_extent_item cmp2;
573 struct btrfs_file_extent_item *existing;
574 struct extent_buffer *leaf;
576 leaf = path->nodes[0];
577 existing = btrfs_item_ptr(leaf, path->slots[0],
578 struct btrfs_file_extent_item);
580 read_extent_buffer(eb, &cmp1, (unsigned long)item,
582 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
586 * we already have a pointer to this exact extent,
587 * we don't have to do anything
589 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
590 btrfs_release_path(root, path);
594 btrfs_release_path(root, path);
596 /* drop any overlapping extents */
597 ret = btrfs_drop_extents(trans, root, inode,
598 start, extent_end, start, &alloc_hint);
601 /* insert the extent */
602 ret = overwrite_item(trans, root, path, eb, slot, key);
605 /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
606 inode_add_bytes(inode, extent_end - start);
607 btrfs_update_inode(trans, root, inode);
615 * when cleaning up conflicts between the directory names in the
616 * subvolume, directory names in the log and directory names in the
617 * inode back references, we may have to unlink inodes from directories.
619 * This is a helper function to do the unlink of a specific directory
622 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
623 struct btrfs_root *root,
624 struct btrfs_path *path,
626 struct btrfs_dir_item *di)
631 struct extent_buffer *leaf;
632 struct btrfs_key location;
635 leaf = path->nodes[0];
637 btrfs_dir_item_key_to_cpu(leaf, di, &location);
638 name_len = btrfs_dir_name_len(leaf, di);
639 name = kmalloc(name_len, GFP_NOFS);
640 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
641 btrfs_release_path(root, path);
643 inode = read_one_inode(root, location.objectid);
646 btrfs_inc_nlink(inode);
647 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
655 * helper function to see if a given name and sequence number found
656 * in an inode back reference are already in a directory and correctly
657 * point to this inode
659 static noinline int inode_in_dir(struct btrfs_root *root,
660 struct btrfs_path *path,
661 u64 dirid, u64 objectid, u64 index,
662 const char *name, int name_len)
664 struct btrfs_dir_item *di;
665 struct btrfs_key location;
668 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
669 index, name, name_len, 0);
670 if (di && !IS_ERR(di)) {
671 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
672 if (location.objectid != objectid)
676 btrfs_release_path(root, path);
678 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
679 if (di && !IS_ERR(di)) {
680 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
681 if (location.objectid != objectid)
687 btrfs_release_path(root, path);
692 * helper function to check a log tree for a named back reference in
693 * an inode. This is used to decide if a back reference that is
694 * found in the subvolume conflicts with what we find in the log.
696 * inode backreferences may have multiple refs in a single item,
697 * during replay we process one reference at a time, and we don't
698 * want to delete valid links to a file from the subvolume if that
699 * link is also in the log.
701 static noinline int backref_in_log(struct btrfs_root *log,
702 struct btrfs_key *key,
703 char *name, int namelen)
705 struct btrfs_path *path;
706 struct btrfs_inode_ref *ref;
708 unsigned long ptr_end;
709 unsigned long name_ptr;
715 path = btrfs_alloc_path();
716 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
720 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
721 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
722 ptr_end = ptr + item_size;
723 while (ptr < ptr_end) {
724 ref = (struct btrfs_inode_ref *)ptr;
725 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
726 if (found_name_len == namelen) {
727 name_ptr = (unsigned long)(ref + 1);
728 ret = memcmp_extent_buffer(path->nodes[0], name,
735 ptr = (unsigned long)(ref + 1) + found_name_len;
738 btrfs_free_path(path);
744 * replay one inode back reference item found in the log tree.
745 * eb, slot and key refer to the buffer and key found in the log tree.
746 * root is the destination we are replaying into, and path is for temp
747 * use by this function. (it should be released on return).
749 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
750 struct btrfs_root *root,
751 struct btrfs_root *log,
752 struct btrfs_path *path,
753 struct extent_buffer *eb, int slot,
754 struct btrfs_key *key)
758 struct btrfs_key location;
759 struct btrfs_inode_ref *ref;
760 struct btrfs_dir_item *di;
764 unsigned long ref_ptr;
765 unsigned long ref_end;
767 location.objectid = key->objectid;
768 location.type = BTRFS_INODE_ITEM_KEY;
772 * it is possible that we didn't log all the parent directories
773 * for a given inode. If we don't find the dir, just don't
774 * copy the back ref in. The link count fixup code will take
777 dir = read_one_inode(root, key->offset);
781 inode = read_one_inode(root, key->objectid);
784 ref_ptr = btrfs_item_ptr_offset(eb, slot);
785 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
788 ref = (struct btrfs_inode_ref *)ref_ptr;
790 namelen = btrfs_inode_ref_name_len(eb, ref);
791 name = kmalloc(namelen, GFP_NOFS);
794 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
796 /* if we already have a perfect match, we're done */
797 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
798 btrfs_inode_ref_index(eb, ref),
804 * look for a conflicting back reference in the metadata.
805 * if we find one we have to unlink that name of the file
806 * before we add our new link. Later on, we overwrite any
807 * existing back reference, and we don't want to create
808 * dangling pointers in the directory.
811 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
815 struct btrfs_inode_ref *victim_ref;
817 unsigned long ptr_end;
818 struct extent_buffer *leaf = path->nodes[0];
820 /* are we trying to overwrite a back ref for the root directory
821 * if so, just jump out, we're done
823 if (key->objectid == key->offset)
826 /* check all the names in this back reference to see
827 * if they are in the log. if so, we allow them to stay
828 * otherwise they must be unlinked as a conflict
830 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
831 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
832 while(ptr < ptr_end) {
833 victim_ref = (struct btrfs_inode_ref *)ptr;
834 victim_name_len = btrfs_inode_ref_name_len(leaf,
836 victim_name = kmalloc(victim_name_len, GFP_NOFS);
837 BUG_ON(!victim_name);
839 read_extent_buffer(leaf, victim_name,
840 (unsigned long)(victim_ref + 1),
843 if (!backref_in_log(log, key, victim_name,
845 btrfs_inc_nlink(inode);
846 btrfs_release_path(root, path);
847 ret = btrfs_unlink_inode(trans, root, dir,
851 btrfs_release_path(root, path);
855 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
859 btrfs_release_path(root, path);
861 /* look for a conflicting sequence number */
862 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
863 btrfs_inode_ref_index(eb, ref),
865 if (di && !IS_ERR(di)) {
866 ret = drop_one_dir_item(trans, root, path, dir, di);
869 btrfs_release_path(root, path);
872 /* look for a conflicting name */
873 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
875 if (di && !IS_ERR(di)) {
876 ret = drop_one_dir_item(trans, root, path, dir, di);
879 btrfs_release_path(root, path);
881 /* insert our name */
882 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
883 btrfs_inode_ref_index(eb, ref));
886 btrfs_update_inode(trans, root, inode);
889 ref_ptr = (unsigned long)(ref + 1) + namelen;
891 if (ref_ptr < ref_end)
894 /* finally write the back reference in the inode */
895 ret = overwrite_item(trans, root, path, eb, slot, key);
899 btrfs_release_path(root, path);
906 * replay one csum item from the log tree into the subvolume 'root'
907 * eb, slot and key all refer to the log tree
908 * path is for temp use by this function and should be released on return
910 * This copies the checksums out of the log tree and inserts them into
911 * the subvolume. Any existing checksums for this range in the file
912 * are overwritten, and new items are added where required.
914 * We keep this simple by reusing the btrfs_ordered_sum code from
915 * the data=ordered mode. This basically means making a copy
916 * of all the checksums in ram, which we have to do anyway for kmap
919 * The copy is then sent down to btrfs_csum_file_blocks, which
920 * does all the hard work of finding existing items in the file
921 * or adding new ones.
923 static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
924 struct btrfs_root *root,
925 struct btrfs_path *path,
926 struct extent_buffer *eb, int slot,
927 struct btrfs_key *key)
930 u32 item_size = btrfs_item_size_nr(eb, slot);
932 unsigned long file_bytes;
933 struct btrfs_ordered_sum *sums;
934 struct btrfs_sector_sum *sector_sum;
938 file_bytes = (item_size / BTRFS_CRC32_SIZE) * root->sectorsize;
939 inode = read_one_inode(root, key->objectid);
944 sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
950 INIT_LIST_HEAD(&sums->list);
951 sums->len = file_bytes;
952 sums->file_offset = key->offset;
955 * copy all the sums into the ordered sum struct
957 sector_sum = sums->sums;
958 cur_offset = key->offset;
959 ptr = btrfs_item_ptr_offset(eb, slot);
960 while(item_size > 0) {
961 sector_sum->offset = cur_offset;
962 read_extent_buffer(eb, §or_sum->sum, ptr, BTRFS_CRC32_SIZE);
964 item_size -= BTRFS_CRC32_SIZE;
965 ptr += BTRFS_CRC32_SIZE;
966 cur_offset += root->sectorsize;
969 /* let btrfs_csum_file_blocks add them into the file */
970 ret = btrfs_csum_file_blocks(trans, root, inode, sums);
978 * There are a few corners where the link count of the file can't
979 * be properly maintained during replay. So, instead of adding
980 * lots of complexity to the log code, we just scan the backrefs
981 * for any file that has been through replay.
983 * The scan will update the link count on the inode to reflect the
984 * number of back refs found. If it goes down to zero, the iput
985 * will free the inode.
987 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
988 struct btrfs_root *root,
991 struct btrfs_path *path;
993 struct btrfs_key key;
996 unsigned long ptr_end;
999 key.objectid = inode->i_ino;
1000 key.type = BTRFS_INODE_REF_KEY;
1001 key.offset = (u64)-1;
1003 path = btrfs_alloc_path();
1006 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1010 if (path->slots[0] == 0)
1014 btrfs_item_key_to_cpu(path->nodes[0], &key,
1016 if (key.objectid != inode->i_ino ||
1017 key.type != BTRFS_INODE_REF_KEY)
1019 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1020 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1022 while(ptr < ptr_end) {
1023 struct btrfs_inode_ref *ref;
1025 ref = (struct btrfs_inode_ref *)ptr;
1026 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1028 ptr = (unsigned long)(ref + 1) + name_len;
1032 if (key.offset == 0)
1035 btrfs_release_path(root, path);
1037 btrfs_free_path(path);
1038 if (nlink != inode->i_nlink) {
1039 inode->i_nlink = nlink;
1040 btrfs_update_inode(trans, root, inode);
1042 BTRFS_I(inode)->index_cnt = (u64)-1;
1047 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1048 struct btrfs_root *root,
1049 struct btrfs_path *path)
1052 struct btrfs_key key;
1053 struct inode *inode;
1055 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1056 key.type = BTRFS_ORPHAN_ITEM_KEY;
1057 key.offset = (u64)-1;
1059 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1064 if (path->slots[0] == 0)
1069 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1070 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1071 key.type != BTRFS_ORPHAN_ITEM_KEY)
1074 ret = btrfs_del_item(trans, root, path);
1077 btrfs_release_path(root, path);
1078 inode = read_one_inode(root, key.offset);
1081 ret = fixup_inode_link_count(trans, root, inode);
1086 if (key.offset == 0)
1090 btrfs_release_path(root, path);
1096 * record a given inode in the fixup dir so we can check its link
1097 * count when replay is done. The link count is incremented here
1098 * so the inode won't go away until we check it
1100 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1101 struct btrfs_root *root,
1102 struct btrfs_path *path,
1105 struct btrfs_key key;
1107 struct inode *inode;
1109 inode = read_one_inode(root, objectid);
1112 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1113 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1114 key.offset = objectid;
1116 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1118 btrfs_release_path(root, path);
1120 btrfs_inc_nlink(inode);
1121 btrfs_update_inode(trans, root, inode);
1122 } else if (ret == -EEXIST) {
1133 * when replaying the log for a directory, we only insert names
1134 * for inodes that actually exist. This means an fsync on a directory
1135 * does not implicitly fsync all the new files in it
1137 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1138 struct btrfs_root *root,
1139 struct btrfs_path *path,
1140 u64 dirid, u64 index,
1141 char *name, int name_len, u8 type,
1142 struct btrfs_key *location)
1144 struct inode *inode;
1148 inode = read_one_inode(root, location->objectid);
1152 dir = read_one_inode(root, dirid);
1157 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1159 /* FIXME, put inode into FIXUP list */
1167 * take a single entry in a log directory item and replay it into
1170 * if a conflicting item exists in the subdirectory already,
1171 * the inode it points to is unlinked and put into the link count
1174 * If a name from the log points to a file or directory that does
1175 * not exist in the FS, it is skipped. fsyncs on directories
1176 * do not force down inodes inside that directory, just changes to the
1177 * names or unlinks in a directory.
1179 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1180 struct btrfs_root *root,
1181 struct btrfs_path *path,
1182 struct extent_buffer *eb,
1183 struct btrfs_dir_item *di,
1184 struct btrfs_key *key)
1188 struct btrfs_dir_item *dst_di;
1189 struct btrfs_key found_key;
1190 struct btrfs_key log_key;
1196 dir = read_one_inode(root, key->objectid);
1199 name_len = btrfs_dir_name_len(eb, di);
1200 name = kmalloc(name_len, GFP_NOFS);
1201 log_type = btrfs_dir_type(eb, di);
1202 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1205 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1206 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1211 btrfs_release_path(root, path);
1213 if (key->type == BTRFS_DIR_ITEM_KEY) {
1214 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1217 else if (key->type == BTRFS_DIR_INDEX_KEY) {
1218 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1225 if (!dst_di || IS_ERR(dst_di)) {
1226 /* we need a sequence number to insert, so we only
1227 * do inserts for the BTRFS_DIR_INDEX_KEY types
1229 if (key->type != BTRFS_DIR_INDEX_KEY)
1234 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1235 /* the existing item matches the logged item */
1236 if (found_key.objectid == log_key.objectid &&
1237 found_key.type == log_key.type &&
1238 found_key.offset == log_key.offset &&
1239 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1244 * don't drop the conflicting directory entry if the inode
1245 * for the new entry doesn't exist
1250 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1253 if (key->type == BTRFS_DIR_INDEX_KEY)
1256 btrfs_release_path(root, path);
1262 btrfs_release_path(root, path);
1263 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1264 name, name_len, log_type, &log_key);
1266 if (ret && ret != -ENOENT)
1272 * find all the names in a directory item and reconcile them into
1273 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1274 * one name in a directory item, but the same code gets used for
1275 * both directory index types
1277 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1278 struct btrfs_root *root,
1279 struct btrfs_path *path,
1280 struct extent_buffer *eb, int slot,
1281 struct btrfs_key *key)
1284 u32 item_size = btrfs_item_size_nr(eb, slot);
1285 struct btrfs_dir_item *di;
1288 unsigned long ptr_end;
1290 ptr = btrfs_item_ptr_offset(eb, slot);
1291 ptr_end = ptr + item_size;
1292 while(ptr < ptr_end) {
1293 di = (struct btrfs_dir_item *)ptr;
1294 name_len = btrfs_dir_name_len(eb, di);
1295 ret = replay_one_name(trans, root, path, eb, di, key);
1297 ptr = (unsigned long)(di + 1);
1304 * directory replay has two parts. There are the standard directory
1305 * items in the log copied from the subvolume, and range items
1306 * created in the log while the subvolume was logged.
1308 * The range items tell us which parts of the key space the log
1309 * is authoritative for. During replay, if a key in the subvolume
1310 * directory is in a logged range item, but not actually in the log
1311 * that means it was deleted from the directory before the fsync
1312 * and should be removed.
1314 static noinline int find_dir_range(struct btrfs_root *root,
1315 struct btrfs_path *path,
1316 u64 dirid, int key_type,
1317 u64 *start_ret, u64 *end_ret)
1319 struct btrfs_key key;
1321 struct btrfs_dir_log_item *item;
1325 if (*start_ret == (u64)-1)
1328 key.objectid = dirid;
1329 key.type = key_type;
1330 key.offset = *start_ret;
1332 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1336 if (path->slots[0] == 0)
1341 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1343 if (key.type != key_type || key.objectid != dirid) {
1347 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1348 struct btrfs_dir_log_item);
1349 found_end = btrfs_dir_log_end(path->nodes[0], item);
1351 if (*start_ret >= key.offset && *start_ret <= found_end) {
1353 *start_ret = key.offset;
1354 *end_ret = found_end;
1359 /* check the next slot in the tree to see if it is a valid item */
1360 nritems = btrfs_header_nritems(path->nodes[0]);
1361 if (path->slots[0] >= nritems) {
1362 ret = btrfs_next_leaf(root, path);
1369 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1371 if (key.type != key_type || key.objectid != dirid) {
1375 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1376 struct btrfs_dir_log_item);
1377 found_end = btrfs_dir_log_end(path->nodes[0], item);
1378 *start_ret = key.offset;
1379 *end_ret = found_end;
1382 btrfs_release_path(root, path);
1387 * this looks for a given directory item in the log. If the directory
1388 * item is not in the log, the item is removed and the inode it points
1391 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1392 struct btrfs_root *root,
1393 struct btrfs_root *log,
1394 struct btrfs_path *path,
1395 struct btrfs_path *log_path,
1397 struct btrfs_key *dir_key)
1400 struct extent_buffer *eb;
1403 struct btrfs_dir_item *di;
1404 struct btrfs_dir_item *log_di;
1407 unsigned long ptr_end;
1409 struct inode *inode;
1410 struct btrfs_key location;
1413 eb = path->nodes[0];
1414 slot = path->slots[0];
1415 item_size = btrfs_item_size_nr(eb, slot);
1416 ptr = btrfs_item_ptr_offset(eb, slot);
1417 ptr_end = ptr + item_size;
1418 while(ptr < ptr_end) {
1419 di = (struct btrfs_dir_item *)ptr;
1420 name_len = btrfs_dir_name_len(eb, di);
1421 name = kmalloc(name_len, GFP_NOFS);
1426 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1429 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1430 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1433 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1434 log_di = btrfs_lookup_dir_index_item(trans, log,
1440 if (!log_di || IS_ERR(log_di)) {
1441 btrfs_dir_item_key_to_cpu(eb, di, &location);
1442 btrfs_release_path(root, path);
1443 btrfs_release_path(log, log_path);
1444 inode = read_one_inode(root, location.objectid);
1447 ret = link_to_fixup_dir(trans, root,
1448 path, location.objectid);
1450 btrfs_inc_nlink(inode);
1451 ret = btrfs_unlink_inode(trans, root, dir, inode,
1457 /* there might still be more names under this key
1458 * check and repeat if required
1460 ret = btrfs_search_slot(NULL, root, dir_key, path,
1467 btrfs_release_path(log, log_path);
1470 ptr = (unsigned long)(di + 1);
1475 btrfs_release_path(root, path);
1476 btrfs_release_path(log, log_path);
1481 * deletion replay happens before we copy any new directory items
1482 * out of the log or out of backreferences from inodes. It
1483 * scans the log to find ranges of keys that log is authoritative for,
1484 * and then scans the directory to find items in those ranges that are
1485 * not present in the log.
1487 * Anything we don't find in the log is unlinked and removed from the
1490 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1491 struct btrfs_root *root,
1492 struct btrfs_root *log,
1493 struct btrfs_path *path,
1498 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1500 struct btrfs_key dir_key;
1501 struct btrfs_key found_key;
1502 struct btrfs_path *log_path;
1505 dir_key.objectid = dirid;
1506 dir_key.type = BTRFS_DIR_ITEM_KEY;
1507 log_path = btrfs_alloc_path();
1511 dir = read_one_inode(root, dirid);
1512 /* it isn't an error if the inode isn't there, that can happen
1513 * because we replay the deletes before we copy in the inode item
1517 btrfs_free_path(log_path);
1524 ret = find_dir_range(log, path, dirid, key_type,
1525 &range_start, &range_end);
1529 dir_key.offset = range_start;
1532 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1537 nritems = btrfs_header_nritems(path->nodes[0]);
1538 if (path->slots[0] >= nritems) {
1539 ret = btrfs_next_leaf(root, path);
1543 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1545 if (found_key.objectid != dirid ||
1546 found_key.type != dir_key.type)
1549 if (found_key.offset > range_end)
1552 ret = check_item_in_log(trans, root, log, path,
1553 log_path, dir, &found_key);
1555 if (found_key.offset == (u64)-1)
1557 dir_key.offset = found_key.offset + 1;
1559 btrfs_release_path(root, path);
1560 if (range_end == (u64)-1)
1562 range_start = range_end + 1;
1567 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1568 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1569 dir_key.type = BTRFS_DIR_INDEX_KEY;
1570 btrfs_release_path(root, path);
1574 btrfs_release_path(root, path);
1575 btrfs_free_path(log_path);
1581 * the process_func used to replay items from the log tree. This
1582 * gets called in two different stages. The first stage just looks
1583 * for inodes and makes sure they are all copied into the subvolume.
1585 * The second stage copies all the other item types from the log into
1586 * the subvolume. The two stage approach is slower, but gets rid of
1587 * lots of complexity around inodes referencing other inodes that exist
1588 * only in the log (references come from either directory items or inode
1591 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1592 struct walk_control *wc, u64 gen)
1595 struct btrfs_path *path;
1596 struct btrfs_root *root = wc->replay_dest;
1597 struct btrfs_key key;
1603 btrfs_read_buffer(eb, gen);
1605 level = btrfs_header_level(eb);
1610 path = btrfs_alloc_path();
1613 nritems = btrfs_header_nritems(eb);
1614 for (i = 0; i < nritems; i++) {
1615 btrfs_item_key_to_cpu(eb, &key, i);
1616 item_size = btrfs_item_size_nr(eb, i);
1618 /* inode keys are done during the first stage */
1619 if (key.type == BTRFS_INODE_ITEM_KEY &&
1620 wc->stage == LOG_WALK_REPLAY_INODES) {
1621 struct inode *inode;
1622 struct btrfs_inode_item *inode_item;
1625 inode_item = btrfs_item_ptr(eb, i,
1626 struct btrfs_inode_item);
1627 mode = btrfs_inode_mode(eb, inode_item);
1628 if (S_ISDIR(mode)) {
1629 ret = replay_dir_deletes(wc->trans,
1630 root, log, path, key.objectid);
1633 ret = overwrite_item(wc->trans, root, path,
1637 /* for regular files, truncate away
1638 * extents past the new EOF
1640 if (S_ISREG(mode)) {
1641 inode = read_one_inode(root,
1645 ret = btrfs_truncate_inode_items(wc->trans,
1646 root, inode, inode->i_size,
1647 BTRFS_EXTENT_DATA_KEY);
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_CSUM_ITEM_KEY) {
1672 ret = replay_one_csum(wc->trans, root, path,
1675 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1676 key.type == BTRFS_DIR_INDEX_KEY) {
1677 ret = replay_one_dir_item(wc->trans, root, path,
1682 btrfs_free_path(path);
1686 static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1687 struct btrfs_root *root,
1688 struct btrfs_path *path, int *level,
1689 struct walk_control *wc)
1695 struct extent_buffer *next;
1696 struct extent_buffer *cur;
1697 struct extent_buffer *parent;
1701 WARN_ON(*level < 0);
1702 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1705 WARN_ON(*level < 0);
1706 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1707 cur = path->nodes[*level];
1709 if (btrfs_header_level(cur) != *level)
1712 if (path->slots[*level] >=
1713 btrfs_header_nritems(cur))
1716 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1717 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1718 blocksize = btrfs_level_size(root, *level - 1);
1720 parent = path->nodes[*level];
1721 root_owner = btrfs_header_owner(parent);
1722 root_gen = btrfs_header_generation(parent);
1724 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1726 wc->process_func(root, next, wc, ptr_gen);
1729 path->slots[*level]++;
1731 btrfs_read_buffer(next, ptr_gen);
1733 btrfs_tree_lock(next);
1734 clean_tree_block(trans, root, next);
1735 btrfs_wait_tree_block_writeback(next);
1736 btrfs_tree_unlock(next);
1738 ret = btrfs_drop_leaf_ref(trans, root, next);
1741 WARN_ON(root_owner !=
1742 BTRFS_TREE_LOG_OBJECTID);
1743 ret = btrfs_free_reserved_extent(root,
1747 free_extent_buffer(next);
1750 btrfs_read_buffer(next, ptr_gen);
1752 WARN_ON(*level <= 0);
1753 if (path->nodes[*level-1])
1754 free_extent_buffer(path->nodes[*level-1]);
1755 path->nodes[*level-1] = next;
1756 *level = btrfs_header_level(next);
1757 path->slots[*level] = 0;
1760 WARN_ON(*level < 0);
1761 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1763 if (path->nodes[*level] == root->node) {
1764 parent = path->nodes[*level];
1766 parent = path->nodes[*level + 1];
1768 bytenr = path->nodes[*level]->start;
1770 blocksize = btrfs_level_size(root, *level);
1771 root_owner = btrfs_header_owner(parent);
1772 root_gen = btrfs_header_generation(parent);
1774 wc->process_func(root, path->nodes[*level], wc,
1775 btrfs_header_generation(path->nodes[*level]));
1778 next = path->nodes[*level];
1779 btrfs_tree_lock(next);
1780 clean_tree_block(trans, root, next);
1781 btrfs_wait_tree_block_writeback(next);
1782 btrfs_tree_unlock(next);
1785 ret = btrfs_drop_leaf_ref(trans, root, next);
1788 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1789 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1792 free_extent_buffer(path->nodes[*level]);
1793 path->nodes[*level] = NULL;
1800 static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path, int *level,
1803 struct walk_control *wc)
1811 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1812 slot = path->slots[i];
1813 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1814 struct extent_buffer *node;
1815 node = path->nodes[i];
1818 WARN_ON(*level == 0);
1821 struct extent_buffer *parent;
1822 if (path->nodes[*level] == root->node)
1823 parent = path->nodes[*level];
1825 parent = path->nodes[*level + 1];
1827 root_owner = btrfs_header_owner(parent);
1828 root_gen = btrfs_header_generation(parent);
1829 wc->process_func(root, path->nodes[*level], wc,
1830 btrfs_header_generation(path->nodes[*level]));
1832 struct extent_buffer *next;
1834 next = path->nodes[*level];
1836 btrfs_tree_lock(next);
1837 clean_tree_block(trans, root, next);
1838 btrfs_wait_tree_block_writeback(next);
1839 btrfs_tree_unlock(next);
1842 ret = btrfs_drop_leaf_ref(trans, root,
1847 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1848 ret = btrfs_free_reserved_extent(root,
1849 path->nodes[*level]->start,
1850 path->nodes[*level]->len);
1853 free_extent_buffer(path->nodes[*level]);
1854 path->nodes[*level] = NULL;
1862 * drop the reference count on the tree rooted at 'snap'. This traverses
1863 * the tree freeing any blocks that have a ref count of zero after being
1866 static int walk_log_tree(struct btrfs_trans_handle *trans,
1867 struct btrfs_root *log, struct walk_control *wc)
1872 struct btrfs_path *path;
1876 path = btrfs_alloc_path();
1879 level = btrfs_header_level(log->node);
1881 path->nodes[level] = log->node;
1882 extent_buffer_get(log->node);
1883 path->slots[level] = 0;
1886 wret = walk_down_log_tree(trans, log, path, &level, wc);
1892 wret = walk_up_log_tree(trans, log, path, &level, wc);
1899 /* was the root node processed? if not, catch it here */
1900 if (path->nodes[orig_level]) {
1901 wc->process_func(log, path->nodes[orig_level], wc,
1902 btrfs_header_generation(path->nodes[orig_level]));
1904 struct extent_buffer *next;
1906 next = path->nodes[orig_level];
1908 btrfs_tree_lock(next);
1909 clean_tree_block(trans, log, next);
1910 btrfs_wait_tree_block_writeback(next);
1911 btrfs_tree_unlock(next);
1913 if (orig_level == 0) {
1914 ret = btrfs_drop_leaf_ref(trans, log,
1918 WARN_ON(log->root_key.objectid !=
1919 BTRFS_TREE_LOG_OBJECTID);
1920 ret = btrfs_free_reserved_extent(log, next->start,
1926 for (i = 0; i <= orig_level; i++) {
1927 if (path->nodes[i]) {
1928 free_extent_buffer(path->nodes[i]);
1929 path->nodes[i] = NULL;
1932 btrfs_free_path(path);
1934 free_extent_buffer(log->node);
1938 static int wait_log_commit(struct btrfs_root *log)
1941 u64 transid = log->fs_info->tree_log_transid;
1944 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1945 TASK_UNINTERRUPTIBLE);
1946 mutex_unlock(&log->fs_info->tree_log_mutex);
1947 if (atomic_read(&log->fs_info->tree_log_commit))
1949 finish_wait(&log->fs_info->tree_log_wait, &wait);
1950 mutex_lock(&log->fs_info->tree_log_mutex);
1951 } while(transid == log->fs_info->tree_log_transid &&
1952 atomic_read(&log->fs_info->tree_log_commit));
1957 * btrfs_sync_log does sends a given tree log down to the disk and
1958 * updates the super blocks to record it. When this call is done,
1959 * you know that any inodes previously logged are safely on disk
1961 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1962 struct btrfs_root *root)
1965 unsigned long batch;
1966 struct btrfs_root *log = root->log_root;
1968 mutex_lock(&log->fs_info->tree_log_mutex);
1969 if (atomic_read(&log->fs_info->tree_log_commit)) {
1970 wait_log_commit(log);
1973 atomic_set(&log->fs_info->tree_log_commit, 1);
1976 batch = log->fs_info->tree_log_batch;
1977 mutex_unlock(&log->fs_info->tree_log_mutex);
1978 schedule_timeout_uninterruptible(1);
1979 mutex_lock(&log->fs_info->tree_log_mutex);
1981 while(atomic_read(&log->fs_info->tree_log_writers)) {
1983 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1984 TASK_UNINTERRUPTIBLE);
1985 mutex_unlock(&log->fs_info->tree_log_mutex);
1986 if (atomic_read(&log->fs_info->tree_log_writers))
1988 mutex_lock(&log->fs_info->tree_log_mutex);
1989 finish_wait(&log->fs_info->tree_log_wait, &wait);
1991 if (batch == log->fs_info->tree_log_batch)
1995 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1997 ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
1998 &root->fs_info->log_root_tree->dirty_log_pages);
2001 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2002 log->fs_info->log_root_tree->node->start);
2003 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2004 btrfs_header_level(log->fs_info->log_root_tree->node));
2006 write_ctree_super(trans, log->fs_info->tree_root);
2007 log->fs_info->tree_log_transid++;
2008 log->fs_info->tree_log_batch = 0;
2009 atomic_set(&log->fs_info->tree_log_commit, 0);
2011 if (waitqueue_active(&log->fs_info->tree_log_wait))
2012 wake_up(&log->fs_info->tree_log_wait);
2014 mutex_unlock(&log->fs_info->tree_log_mutex);
2019 /* * free all the extents used by the tree log. This should be called
2020 * at commit time of the full transaction
2022 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2025 struct btrfs_root *log;
2029 struct walk_control wc = {
2031 .process_func = process_one_buffer
2034 if (!root->log_root)
2037 log = root->log_root;
2038 ret = walk_log_tree(trans, log, &wc);
2042 ret = find_first_extent_bit(&log->dirty_log_pages,
2043 0, &start, &end, EXTENT_DIRTY);
2047 clear_extent_dirty(&log->dirty_log_pages,
2048 start, end, GFP_NOFS);
2051 log = root->log_root;
2052 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2055 root->log_root = NULL;
2056 kfree(root->log_root);
2061 * helper function to update the item for a given subvolumes log root
2062 * in the tree of log roots
2064 static int update_log_root(struct btrfs_trans_handle *trans,
2065 struct btrfs_root *log)
2067 u64 bytenr = btrfs_root_bytenr(&log->root_item);
2070 if (log->node->start == bytenr)
2073 btrfs_set_root_bytenr(&log->root_item, log->node->start);
2074 btrfs_set_root_generation(&log->root_item, trans->transid);
2075 btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2076 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2077 &log->root_key, &log->root_item);
2083 * If both a file and directory are logged, and unlinks or renames are
2084 * mixed in, we have a few interesting corners:
2086 * create file X in dir Y
2087 * link file X to X.link in dir Y
2089 * unlink file X but leave X.link
2092 * After a crash we would expect only X.link to exist. But file X
2093 * didn't get fsync'd again so the log has back refs for X and X.link.
2095 * We solve this by removing directory entries and inode backrefs from the
2096 * log when a file that was logged in the current transaction is
2097 * unlinked. Any later fsync will include the updated log entries, and
2098 * we'll be able to reconstruct the proper directory items from backrefs.
2100 * This optimizations allows us to avoid relogging the entire inode
2101 * or the entire directory.
2103 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2104 struct btrfs_root *root,
2105 const char *name, int name_len,
2106 struct inode *dir, u64 index)
2108 struct btrfs_root *log;
2109 struct btrfs_dir_item *di;
2110 struct btrfs_path *path;
2114 if (BTRFS_I(dir)->logged_trans < trans->transid)
2117 ret = join_running_log_trans(root);
2121 mutex_lock(&BTRFS_I(dir)->log_mutex);
2123 log = root->log_root;
2124 path = btrfs_alloc_path();
2125 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2126 name, name_len, -1);
2127 if (di && !IS_ERR(di)) {
2128 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2129 bytes_del += name_len;
2132 btrfs_release_path(log, path);
2133 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2134 index, name, name_len, -1);
2135 if (di && !IS_ERR(di)) {
2136 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2137 bytes_del += name_len;
2141 /* update the directory size in the log to reflect the names
2145 struct btrfs_key key;
2147 key.objectid = dir->i_ino;
2149 key.type = BTRFS_INODE_ITEM_KEY;
2150 btrfs_release_path(log, path);
2152 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2154 struct btrfs_inode_item *item;
2157 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2158 struct btrfs_inode_item);
2159 i_size = btrfs_inode_size(path->nodes[0], item);
2160 if (i_size > bytes_del)
2161 i_size -= bytes_del;
2164 btrfs_set_inode_size(path->nodes[0], item, i_size);
2165 btrfs_mark_buffer_dirty(path->nodes[0]);
2168 btrfs_release_path(log, path);
2171 btrfs_free_path(path);
2172 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2173 end_log_trans(root);
2178 /* see comments for btrfs_del_dir_entries_in_log */
2179 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root,
2181 const char *name, int name_len,
2182 struct inode *inode, u64 dirid)
2184 struct btrfs_root *log;
2188 if (BTRFS_I(inode)->logged_trans < trans->transid)
2191 ret = join_running_log_trans(root);
2194 log = root->log_root;
2195 mutex_lock(&BTRFS_I(inode)->log_mutex);
2197 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2199 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2200 end_log_trans(root);
2206 * creates a range item in the log for 'dirid'. first_offset and
2207 * last_offset tell us which parts of the key space the log should
2208 * be considered authoritative for.
2210 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2211 struct btrfs_root *log,
2212 struct btrfs_path *path,
2213 int key_type, u64 dirid,
2214 u64 first_offset, u64 last_offset)
2217 struct btrfs_key key;
2218 struct btrfs_dir_log_item *item;
2220 key.objectid = dirid;
2221 key.offset = first_offset;
2222 if (key_type == BTRFS_DIR_ITEM_KEY)
2223 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2225 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2226 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2229 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2230 struct btrfs_dir_log_item);
2231 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2232 btrfs_mark_buffer_dirty(path->nodes[0]);
2233 btrfs_release_path(log, path);
2238 * log all the items included in the current transaction for a given
2239 * directory. This also creates the range items in the log tree required
2240 * to replay anything deleted before the fsync
2242 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2243 struct btrfs_root *root, struct inode *inode,
2244 struct btrfs_path *path,
2245 struct btrfs_path *dst_path, int key_type,
2246 u64 min_offset, u64 *last_offset_ret)
2248 struct btrfs_key min_key;
2249 struct btrfs_key max_key;
2250 struct btrfs_root *log = root->log_root;
2251 struct extent_buffer *src;
2255 u64 first_offset = min_offset;
2256 u64 last_offset = (u64)-1;
2258 log = root->log_root;
2259 max_key.objectid = inode->i_ino;
2260 max_key.offset = (u64)-1;
2261 max_key.type = key_type;
2263 min_key.objectid = inode->i_ino;
2264 min_key.type = key_type;
2265 min_key.offset = min_offset;
2267 path->keep_locks = 1;
2269 ret = btrfs_search_forward(root, &min_key, &max_key,
2270 path, 0, trans->transid);
2273 * we didn't find anything from this transaction, see if there
2274 * is anything at all
2276 if (ret != 0 || min_key.objectid != inode->i_ino ||
2277 min_key.type != key_type) {
2278 min_key.objectid = inode->i_ino;
2279 min_key.type = key_type;
2280 min_key.offset = (u64)-1;
2281 btrfs_release_path(root, path);
2282 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2284 btrfs_release_path(root, path);
2287 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2289 /* if ret == 0 there are items for this type,
2290 * create a range to tell us the last key of this type.
2291 * otherwise, there are no items in this directory after
2292 * *min_offset, and we create a range to indicate that.
2295 struct btrfs_key tmp;
2296 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2298 if (key_type == tmp.type) {
2299 first_offset = max(min_offset, tmp.offset) + 1;
2305 /* go backward to find any previous key */
2306 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2308 struct btrfs_key tmp;
2309 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2310 if (key_type == tmp.type) {
2311 first_offset = tmp.offset;
2312 ret = overwrite_item(trans, log, dst_path,
2313 path->nodes[0], path->slots[0],
2317 btrfs_release_path(root, path);
2319 /* find the first key from this transaction again */
2320 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2327 * we have a block from this transaction, log every item in it
2328 * from our directory
2331 struct btrfs_key tmp;
2332 src = path->nodes[0];
2333 nritems = btrfs_header_nritems(src);
2334 for (i = path->slots[0]; i < nritems; i++) {
2335 btrfs_item_key_to_cpu(src, &min_key, i);
2337 if (min_key.objectid != inode->i_ino ||
2338 min_key.type != key_type)
2340 ret = overwrite_item(trans, log, dst_path, src, i,
2344 path->slots[0] = nritems;
2347 * look ahead to the next item and see if it is also
2348 * from this directory and from this transaction
2350 ret = btrfs_next_leaf(root, path);
2352 last_offset = (u64)-1;
2355 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2356 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2357 last_offset = (u64)-1;
2360 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2361 ret = overwrite_item(trans, log, dst_path,
2362 path->nodes[0], path->slots[0],
2366 last_offset = tmp.offset;
2371 *last_offset_ret = last_offset;
2372 btrfs_release_path(root, path);
2373 btrfs_release_path(log, dst_path);
2375 /* insert the log range keys to indicate where the log is valid */
2376 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2377 first_offset, last_offset);
2383 * logging directories is very similar to logging inodes, We find all the items
2384 * from the current transaction and write them to the log.
2386 * The recovery code scans the directory in the subvolume, and if it finds a
2387 * key in the range logged that is not present in the log tree, then it means
2388 * that dir entry was unlinked during the transaction.
2390 * In order for that scan to work, we must include one key smaller than
2391 * the smallest logged by this transaction and one key larger than the largest
2392 * key logged by this transaction.
2394 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2395 struct btrfs_root *root, struct inode *inode,
2396 struct btrfs_path *path,
2397 struct btrfs_path *dst_path)
2402 int key_type = BTRFS_DIR_ITEM_KEY;
2408 ret = log_dir_items(trans, root, inode, path,
2409 dst_path, key_type, min_key,
2412 if (max_key == (u64)-1)
2414 min_key = max_key + 1;
2417 if (key_type == BTRFS_DIR_ITEM_KEY) {
2418 key_type = BTRFS_DIR_INDEX_KEY;
2425 * a helper function to drop items from the log before we relog an
2426 * inode. max_key_type indicates the highest item type to remove.
2427 * This cannot be run for file data extents because it does not
2428 * free the extents they point to.
2430 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2431 struct btrfs_root *log,
2432 struct btrfs_path *path,
2433 u64 objectid, int max_key_type)
2436 struct btrfs_key key;
2437 struct btrfs_key found_key;
2439 key.objectid = objectid;
2440 key.type = max_key_type;
2441 key.offset = (u64)-1;
2444 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2449 if (path->slots[0] == 0)
2453 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2456 if (found_key.objectid != objectid)
2459 ret = btrfs_del_item(trans, log, path);
2461 btrfs_release_path(log, path);
2463 btrfs_release_path(log, path);
2467 static noinline int copy_items(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *log,
2469 struct btrfs_path *dst_path,
2470 struct extent_buffer *src,
2471 int start_slot, int nr, int inode_only)
2473 unsigned long src_offset;
2474 unsigned long dst_offset;
2475 struct btrfs_file_extent_item *extent;
2476 struct btrfs_inode_item *inode_item;
2478 struct btrfs_key *ins_keys;
2483 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2484 nr * sizeof(u32), GFP_NOFS);
2485 ins_sizes = (u32 *)ins_data;
2486 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2488 for (i = 0; i < nr; i++) {
2489 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2490 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2492 ret = btrfs_insert_empty_items(trans, log, dst_path,
2493 ins_keys, ins_sizes, nr);
2496 for (i = 0; i < nr; i++) {
2497 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2498 dst_path->slots[0]);
2500 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2502 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2503 src_offset, ins_sizes[i]);
2505 if (inode_only == LOG_INODE_EXISTS &&
2506 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2507 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2509 struct btrfs_inode_item);
2510 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2512 /* set the generation to zero so the recover code
2513 * can tell the difference between an logging
2514 * just to say 'this inode exists' and a logging
2515 * to say 'update this inode with these values'
2517 btrfs_set_inode_generation(dst_path->nodes[0],
2520 /* take a reference on file data extents so that truncates
2521 * or deletes of this inode don't have to relog the inode
2524 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2526 extent = btrfs_item_ptr(src, start_slot + i,
2527 struct btrfs_file_extent_item);
2529 found_type = btrfs_file_extent_type(src, extent);
2530 if (found_type == BTRFS_FILE_EXTENT_REG ||
2531 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2532 u64 ds = btrfs_file_extent_disk_bytenr(src,
2534 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2536 /* ds == 0 is a hole */
2538 ret = btrfs_inc_extent_ref(trans, log,
2540 dst_path->nodes[0]->start,
2541 BTRFS_TREE_LOG_OBJECTID,
2543 ins_keys[i].objectid);
2548 dst_path->slots[0]++;
2551 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2552 btrfs_release_path(log, dst_path);
2557 /* log a single inode in the tree log.
2558 * At least one parent directory for this inode must exist in the tree
2559 * or be logged already.
2561 * Any items from this inode changed by the current transaction are copied
2562 * to the log tree. An extra reference is taken on any extents in this
2563 * file, allowing us to avoid a whole pile of corner cases around logging
2564 * blocks that have been removed from the tree.
2566 * See LOG_INODE_ALL and related defines for a description of what inode_only
2569 * This handles both files and directories.
2571 static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2572 struct btrfs_root *root, struct inode *inode,
2575 struct btrfs_path *path;
2576 struct btrfs_path *dst_path;
2577 struct btrfs_key min_key;
2578 struct btrfs_key max_key;
2579 struct btrfs_root *log = root->log_root;
2580 struct extent_buffer *src = NULL;
2584 int ins_start_slot = 0;
2587 log = root->log_root;
2589 path = btrfs_alloc_path();
2590 dst_path = btrfs_alloc_path();
2592 min_key.objectid = inode->i_ino;
2593 min_key.type = BTRFS_INODE_ITEM_KEY;
2596 max_key.objectid = inode->i_ino;
2597 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2598 max_key.type = BTRFS_XATTR_ITEM_KEY;
2600 max_key.type = (u8)-1;
2601 max_key.offset = (u64)-1;
2604 * if this inode has already been logged and we're in inode_only
2605 * mode, we don't want to delete the things that have already
2606 * been written to the log.
2608 * But, if the inode has been through an inode_only log,
2609 * the logged_trans field is not set. This allows us to catch
2610 * any new names for this inode in the backrefs by logging it
2613 if (inode_only == LOG_INODE_EXISTS &&
2614 BTRFS_I(inode)->logged_trans == trans->transid) {
2615 btrfs_free_path(path);
2616 btrfs_free_path(dst_path);
2619 mutex_lock(&BTRFS_I(inode)->log_mutex);
2622 * a brute force approach to making sure we get the most uptodate
2623 * copies of everything.
2625 if (S_ISDIR(inode->i_mode)) {
2626 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2628 if (inode_only == LOG_INODE_EXISTS)
2629 max_key_type = BTRFS_XATTR_ITEM_KEY;
2630 ret = drop_objectid_items(trans, log, path,
2631 inode->i_ino, max_key_type);
2633 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2636 path->keep_locks = 1;
2640 ret = btrfs_search_forward(root, &min_key, &max_key,
2641 path, 0, trans->transid);
2645 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2646 if (min_key.objectid != inode->i_ino)
2648 if (min_key.type > max_key.type)
2651 src = path->nodes[0];
2652 size = btrfs_item_size_nr(src, path->slots[0]);
2653 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2656 } else if (!ins_nr) {
2657 ins_start_slot = path->slots[0];
2662 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2663 ins_nr, inode_only);
2666 ins_start_slot = path->slots[0];
2669 nritems = btrfs_header_nritems(path->nodes[0]);
2671 if (path->slots[0] < nritems) {
2672 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2677 ret = copy_items(trans, log, dst_path, src,
2679 ins_nr, inode_only);
2683 btrfs_release_path(root, path);
2685 if (min_key.offset < (u64)-1)
2687 else if (min_key.type < (u8)-1)
2689 else if (min_key.objectid < (u64)-1)
2695 ret = copy_items(trans, log, dst_path, src,
2697 ins_nr, inode_only);
2702 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2703 btrfs_release_path(root, path);
2704 btrfs_release_path(log, dst_path);
2705 BTRFS_I(inode)->log_dirty_trans = 0;
2706 ret = log_directory_changes(trans, root, inode, path, dst_path);
2709 BTRFS_I(inode)->logged_trans = trans->transid;
2710 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2712 btrfs_free_path(path);
2713 btrfs_free_path(dst_path);
2715 mutex_lock(&root->fs_info->tree_log_mutex);
2716 ret = update_log_root(trans, log);
2718 mutex_unlock(&root->fs_info->tree_log_mutex);
2723 int btrfs_log_inode(struct btrfs_trans_handle *trans,
2724 struct btrfs_root *root, struct inode *inode,
2729 start_log_trans(trans, root);
2730 ret = __btrfs_log_inode(trans, root, inode, inode_only);
2731 end_log_trans(root);
2736 * helper function around btrfs_log_inode to make sure newly created
2737 * parent directories also end up in the log. A minimal inode and backref
2738 * only logging is done of any parent directories that are older than
2739 * the last committed transaction
2741 int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2742 struct btrfs_root *root, struct dentry *dentry)
2744 int inode_only = LOG_INODE_ALL;
2745 struct super_block *sb;
2748 start_log_trans(trans, root);
2749 sb = dentry->d_inode->i_sb;
2751 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2754 inode_only = LOG_INODE_EXISTS;
2756 dentry = dentry->d_parent;
2757 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
2760 if (BTRFS_I(dentry->d_inode)->generation <=
2761 root->fs_info->last_trans_committed)
2764 end_log_trans(root);
2769 * it is not safe to log dentry if the chunk root has added new
2770 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2771 * If this returns 1, you must commit the transaction to safely get your
2774 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2775 struct btrfs_root *root, struct dentry *dentry)
2778 gen = root->fs_info->last_trans_new_blockgroup;
2779 if (gen > root->fs_info->last_trans_committed)
2782 return btrfs_log_dentry(trans, root, dentry);
2786 * should be called during mount to recover any replay any log trees
2789 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2792 struct btrfs_path *path;
2793 struct btrfs_trans_handle *trans;
2794 struct btrfs_key key;
2795 struct btrfs_key found_key;
2796 struct btrfs_key tmp_key;
2797 struct btrfs_root *log;
2798 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2800 struct walk_control wc = {
2801 .process_func = process_one_buffer,
2805 fs_info->log_root_recovering = 1;
2806 path = btrfs_alloc_path();
2809 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2814 walk_log_tree(trans, log_root_tree, &wc);
2817 key.objectid = BTRFS_TREE_LOG_OBJECTID;
2818 key.offset = (u64)-1;
2819 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2822 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2826 if (path->slots[0] == 0)
2830 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2832 btrfs_release_path(log_root_tree, path);
2833 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2836 log = btrfs_read_fs_root_no_radix(log_root_tree,
2841 tmp_key.objectid = found_key.offset;
2842 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2843 tmp_key.offset = (u64)-1;
2845 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
2847 BUG_ON(!wc.replay_dest);
2849 btrfs_record_root_in_trans(wc.replay_dest);
2850 ret = walk_log_tree(trans, log, &wc);
2853 if (wc.stage == LOG_WALK_REPLAY_ALL) {
2854 ret = fixup_inode_link_counts(trans, wc.replay_dest,
2858 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
2860 wc.replay_dest->highest_inode = highest_inode;
2861 wc.replay_dest->last_inode_alloc = highest_inode;
2864 key.offset = found_key.offset - 1;
2865 free_extent_buffer(log->node);
2868 if (found_key.offset == 0)
2871 btrfs_release_path(log_root_tree, path);
2873 /* step one is to pin it all, step two is to replay just inodes */
2876 wc.process_func = replay_one_buffer;
2877 wc.stage = LOG_WALK_REPLAY_INODES;
2880 /* step three is to replay everything */
2881 if (wc.stage < LOG_WALK_REPLAY_ALL) {
2886 btrfs_free_path(path);
2888 free_extent_buffer(log_root_tree->node);
2889 log_root_tree->log_root = NULL;
2890 fs_info->log_root_recovering = 0;
2892 /* step 4: commit the transaction, which also unpins the blocks */
2893 btrfs_commit_transaction(trans, fs_info->tree_root);
2895 kfree(log_root_tree);