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_extend_item(trans, root, path,
378 item_size - found_size);
384 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
387 /* don't overwrite an existing inode if the generation number
388 * was logged as zero. This is done when the tree logging code
389 * is just logging an inode to make sure it exists after recovery.
391 * Also, don't overwrite i_size on directories during replay.
392 * log replay inserts and removes directory items based on the
393 * state of the tree found in the subvolume, and i_size is modified
396 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
397 struct btrfs_inode_item *src_item;
398 struct btrfs_inode_item *dst_item;
400 src_item = (struct btrfs_inode_item *)src_ptr;
401 dst_item = (struct btrfs_inode_item *)dst_ptr;
403 if (btrfs_inode_generation(eb, src_item) == 0)
406 if (overwrite_root &&
407 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
408 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
410 saved_i_size = btrfs_inode_size(path->nodes[0],
415 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
418 if (save_old_i_size) {
419 struct btrfs_inode_item *dst_item;
420 dst_item = (struct btrfs_inode_item *)dst_ptr;
421 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
424 /* make sure the generation is filled in */
425 if (key->type == BTRFS_INODE_ITEM_KEY) {
426 struct btrfs_inode_item *dst_item;
427 dst_item = (struct btrfs_inode_item *)dst_ptr;
428 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
429 btrfs_set_inode_generation(path->nodes[0], dst_item,
434 if (overwrite_root &&
435 key->type == BTRFS_EXTENT_DATA_KEY) {
437 struct btrfs_file_extent_item *fi;
439 fi = (struct btrfs_file_extent_item *)dst_ptr;
440 extent_type = btrfs_file_extent_type(path->nodes[0], fi);
441 if (extent_type == BTRFS_FILE_EXTENT_REG ||
442 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
443 struct btrfs_key ins;
444 ins.objectid = btrfs_file_extent_disk_bytenr(
446 ins.offset = btrfs_file_extent_disk_num_bytes(
448 ins.type = BTRFS_EXTENT_ITEM_KEY;
451 * is this extent already allocated in the extent
452 * allocation tree? If so, just add a reference
454 ret = btrfs_lookup_extent(root, ins.objectid,
457 ret = btrfs_inc_extent_ref(trans, root,
458 ins.objectid, ins.offset,
459 path->nodes[0]->start,
460 root->root_key.objectid,
461 trans->transid, key->objectid);
464 * insert the extent pointer in the extent
467 ret = btrfs_alloc_logged_extent(trans, root,
468 path->nodes[0]->start,
469 root->root_key.objectid,
470 trans->transid, key->objectid,
477 btrfs_mark_buffer_dirty(path->nodes[0]);
478 btrfs_release_path(root, path);
483 * simple helper to read an inode off the disk from a given root
484 * This can only be called for subvolume roots and not for the log
486 static noinline struct inode *read_one_inode(struct btrfs_root *root,
490 inode = btrfs_iget_locked(root->fs_info->sb, objectid, root);
491 if (inode->i_state & I_NEW) {
492 BTRFS_I(inode)->root = root;
493 BTRFS_I(inode)->location.objectid = objectid;
494 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
495 BTRFS_I(inode)->location.offset = 0;
496 btrfs_read_locked_inode(inode);
497 unlock_new_inode(inode);
500 if (is_bad_inode(inode)) {
507 /* replays a single extent in 'eb' at 'slot' with 'key' into the
508 * subvolume 'root'. path is released on entry and should be released
511 * extents in the log tree have not been allocated out of the extent
512 * tree yet. So, this completes the allocation, taking a reference
513 * as required if the extent already exists or creating a new extent
514 * if it isn't in the extent allocation tree yet.
516 * The extent is inserted into the file, dropping any existing extents
517 * from the file that overlap the new one.
519 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
520 struct btrfs_root *root,
521 struct btrfs_path *path,
522 struct extent_buffer *eb, int slot,
523 struct btrfs_key *key)
526 u64 mask = root->sectorsize - 1;
529 u64 start = key->offset;
530 struct btrfs_file_extent_item *item;
531 struct inode *inode = NULL;
535 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
536 found_type = btrfs_file_extent_type(eb, item);
538 if (found_type == BTRFS_FILE_EXTENT_REG ||
539 found_type == BTRFS_FILE_EXTENT_PREALLOC)
540 extent_end = start + btrfs_file_extent_num_bytes(eb, item);
541 else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
542 size = btrfs_file_extent_inline_len(eb, item);
543 extent_end = (start + size + mask) & ~mask;
549 inode = read_one_inode(root, key->objectid);
556 * first check to see if we already have this extent in the
557 * file. This must be done before the btrfs_drop_extents run
558 * so we don't try to drop this extent.
560 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
564 (found_type == BTRFS_FILE_EXTENT_REG ||
565 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
566 struct btrfs_file_extent_item cmp1;
567 struct btrfs_file_extent_item cmp2;
568 struct btrfs_file_extent_item *existing;
569 struct extent_buffer *leaf;
571 leaf = path->nodes[0];
572 existing = btrfs_item_ptr(leaf, path->slots[0],
573 struct btrfs_file_extent_item);
575 read_extent_buffer(eb, &cmp1, (unsigned long)item,
577 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
581 * we already have a pointer to this exact extent,
582 * we don't have to do anything
584 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
585 btrfs_release_path(root, path);
589 btrfs_release_path(root, path);
591 /* drop any overlapping extents */
592 ret = btrfs_drop_extents(trans, root, inode,
593 start, extent_end, start, &alloc_hint);
596 /* insert the extent */
597 ret = overwrite_item(trans, root, path, eb, slot, key);
600 /* btrfs_drop_extents changes i_bytes & i_blocks, update it here */
601 inode_add_bytes(inode, extent_end - start);
602 btrfs_update_inode(trans, root, inode);
610 * when cleaning up conflicts between the directory names in the
611 * subvolume, directory names in the log and directory names in the
612 * inode back references, we may have to unlink inodes from directories.
614 * This is a helper function to do the unlink of a specific directory
617 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
618 struct btrfs_root *root,
619 struct btrfs_path *path,
621 struct btrfs_dir_item *di)
626 struct extent_buffer *leaf;
627 struct btrfs_key location;
630 leaf = path->nodes[0];
632 btrfs_dir_item_key_to_cpu(leaf, di, &location);
633 name_len = btrfs_dir_name_len(leaf, di);
634 name = kmalloc(name_len, GFP_NOFS);
635 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
636 btrfs_release_path(root, path);
638 inode = read_one_inode(root, location.objectid);
641 btrfs_inc_nlink(inode);
642 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
650 * helper function to see if a given name and sequence number found
651 * in an inode back reference are already in a directory and correctly
652 * point to this inode
654 static noinline int inode_in_dir(struct btrfs_root *root,
655 struct btrfs_path *path,
656 u64 dirid, u64 objectid, u64 index,
657 const char *name, int name_len)
659 struct btrfs_dir_item *di;
660 struct btrfs_key location;
663 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
664 index, name, name_len, 0);
665 if (di && !IS_ERR(di)) {
666 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
667 if (location.objectid != objectid)
671 btrfs_release_path(root, path);
673 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
674 if (di && !IS_ERR(di)) {
675 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
676 if (location.objectid != objectid)
682 btrfs_release_path(root, path);
687 * helper function to check a log tree for a named back reference in
688 * an inode. This is used to decide if a back reference that is
689 * found in the subvolume conflicts with what we find in the log.
691 * inode backreferences may have multiple refs in a single item,
692 * during replay we process one reference at a time, and we don't
693 * want to delete valid links to a file from the subvolume if that
694 * link is also in the log.
696 static noinline int backref_in_log(struct btrfs_root *log,
697 struct btrfs_key *key,
698 char *name, int namelen)
700 struct btrfs_path *path;
701 struct btrfs_inode_ref *ref;
703 unsigned long ptr_end;
704 unsigned long name_ptr;
710 path = btrfs_alloc_path();
711 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
715 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
716 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
717 ptr_end = ptr + item_size;
718 while (ptr < ptr_end) {
719 ref = (struct btrfs_inode_ref *)ptr;
720 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
721 if (found_name_len == namelen) {
722 name_ptr = (unsigned long)(ref + 1);
723 ret = memcmp_extent_buffer(path->nodes[0], name,
730 ptr = (unsigned long)(ref + 1) + found_name_len;
733 btrfs_free_path(path);
739 * replay one inode back reference item found in the log tree.
740 * eb, slot and key refer to the buffer and key found in the log tree.
741 * root is the destination we are replaying into, and path is for temp
742 * use by this function. (it should be released on return).
744 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
745 struct btrfs_root *root,
746 struct btrfs_root *log,
747 struct btrfs_path *path,
748 struct extent_buffer *eb, int slot,
749 struct btrfs_key *key)
753 struct btrfs_key location;
754 struct btrfs_inode_ref *ref;
755 struct btrfs_dir_item *di;
759 unsigned long ref_ptr;
760 unsigned long ref_end;
762 location.objectid = key->objectid;
763 location.type = BTRFS_INODE_ITEM_KEY;
767 * it is possible that we didn't log all the parent directories
768 * for a given inode. If we don't find the dir, just don't
769 * copy the back ref in. The link count fixup code will take
772 dir = read_one_inode(root, key->offset);
776 inode = read_one_inode(root, key->objectid);
779 ref_ptr = btrfs_item_ptr_offset(eb, slot);
780 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
783 ref = (struct btrfs_inode_ref *)ref_ptr;
785 namelen = btrfs_inode_ref_name_len(eb, ref);
786 name = kmalloc(namelen, GFP_NOFS);
789 read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
791 /* if we already have a perfect match, we're done */
792 if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
793 btrfs_inode_ref_index(eb, ref),
799 * look for a conflicting back reference in the metadata.
800 * if we find one we have to unlink that name of the file
801 * before we add our new link. Later on, we overwrite any
802 * existing back reference, and we don't want to create
803 * dangling pointers in the directory.
806 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
810 struct btrfs_inode_ref *victim_ref;
812 unsigned long ptr_end;
813 struct extent_buffer *leaf = path->nodes[0];
815 /* are we trying to overwrite a back ref for the root directory
816 * if so, just jump out, we're done
818 if (key->objectid == key->offset)
821 /* check all the names in this back reference to see
822 * if they are in the log. if so, we allow them to stay
823 * otherwise they must be unlinked as a conflict
825 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
826 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
827 while(ptr < ptr_end) {
828 victim_ref = (struct btrfs_inode_ref *)ptr;
829 victim_name_len = btrfs_inode_ref_name_len(leaf,
831 victim_name = kmalloc(victim_name_len, GFP_NOFS);
832 BUG_ON(!victim_name);
834 read_extent_buffer(leaf, victim_name,
835 (unsigned long)(victim_ref + 1),
838 if (!backref_in_log(log, key, victim_name,
840 btrfs_inc_nlink(inode);
841 btrfs_release_path(root, path);
842 ret = btrfs_unlink_inode(trans, root, dir,
846 btrfs_release_path(root, path);
850 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
854 btrfs_release_path(root, path);
856 /* look for a conflicting sequence number */
857 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
858 btrfs_inode_ref_index(eb, ref),
860 if (di && !IS_ERR(di)) {
861 ret = drop_one_dir_item(trans, root, path, dir, di);
864 btrfs_release_path(root, path);
867 /* look for a conflicting name */
868 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
870 if (di && !IS_ERR(di)) {
871 ret = drop_one_dir_item(trans, root, path, dir, di);
874 btrfs_release_path(root, path);
876 /* insert our name */
877 ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
878 btrfs_inode_ref_index(eb, ref));
881 btrfs_update_inode(trans, root, inode);
884 ref_ptr = (unsigned long)(ref + 1) + namelen;
886 if (ref_ptr < ref_end)
889 /* finally write the back reference in the inode */
890 ret = overwrite_item(trans, root, path, eb, slot, key);
894 btrfs_release_path(root, path);
901 * replay one csum item from the log tree into the subvolume 'root'
902 * eb, slot and key all refer to the log tree
903 * path is for temp use by this function and should be released on return
905 * This copies the checksums out of the log tree and inserts them into
906 * the subvolume. Any existing checksums for this range in the file
907 * are overwritten, and new items are added where required.
909 * We keep this simple by reusing the btrfs_ordered_sum code from
910 * the data=ordered mode. This basically means making a copy
911 * of all the checksums in ram, which we have to do anyway for kmap
914 * The copy is then sent down to btrfs_csum_file_blocks, which
915 * does all the hard work of finding existing items in the file
916 * or adding new ones.
918 static noinline int replay_one_csum(struct btrfs_trans_handle *trans,
919 struct btrfs_root *root,
920 struct btrfs_path *path,
921 struct extent_buffer *eb, int slot,
922 struct btrfs_key *key)
925 u32 item_size = btrfs_item_size_nr(eb, slot);
928 btrfs_super_csum_size(&root->fs_info->super_copy);
929 unsigned long file_bytes;
930 struct btrfs_ordered_sum *sums;
931 struct btrfs_sector_sum *sector_sum;
934 file_bytes = (item_size / csum_size) * root->sectorsize;
935 sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
940 INIT_LIST_HEAD(&sums->list);
941 sums->len = file_bytes;
942 sums->bytenr = key->offset;
945 * copy all the sums into the ordered sum struct
947 sector_sum = sums->sums;
948 cur_offset = key->offset;
949 ptr = btrfs_item_ptr_offset(eb, slot);
950 while(item_size > 0) {
951 sector_sum->bytenr = cur_offset;
952 read_extent_buffer(eb, §or_sum->sum, ptr, csum_size);
954 item_size -= csum_size;
956 cur_offset += root->sectorsize;
959 /* let btrfs_csum_file_blocks add them into the file */
960 ret = btrfs_csum_file_blocks(trans, root->fs_info->csum_root, sums);
966 * There are a few corners where the link count of the file can't
967 * be properly maintained during replay. So, instead of adding
968 * lots of complexity to the log code, we just scan the backrefs
969 * for any file that has been through replay.
971 * The scan will update the link count on the inode to reflect the
972 * number of back refs found. If it goes down to zero, the iput
973 * will free the inode.
975 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root,
979 struct btrfs_path *path;
981 struct btrfs_key key;
984 unsigned long ptr_end;
987 key.objectid = inode->i_ino;
988 key.type = BTRFS_INODE_REF_KEY;
989 key.offset = (u64)-1;
991 path = btrfs_alloc_path();
994 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
998 if (path->slots[0] == 0)
1002 btrfs_item_key_to_cpu(path->nodes[0], &key,
1004 if (key.objectid != inode->i_ino ||
1005 key.type != BTRFS_INODE_REF_KEY)
1007 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1008 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1010 while(ptr < ptr_end) {
1011 struct btrfs_inode_ref *ref;
1013 ref = (struct btrfs_inode_ref *)ptr;
1014 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1016 ptr = (unsigned long)(ref + 1) + name_len;
1020 if (key.offset == 0)
1023 btrfs_release_path(root, path);
1025 btrfs_free_path(path);
1026 if (nlink != inode->i_nlink) {
1027 inode->i_nlink = nlink;
1028 btrfs_update_inode(trans, root, inode);
1030 BTRFS_I(inode)->index_cnt = (u64)-1;
1035 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1036 struct btrfs_root *root,
1037 struct btrfs_path *path)
1040 struct btrfs_key key;
1041 struct inode *inode;
1043 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1044 key.type = BTRFS_ORPHAN_ITEM_KEY;
1045 key.offset = (u64)-1;
1047 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1052 if (path->slots[0] == 0)
1057 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1058 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1059 key.type != BTRFS_ORPHAN_ITEM_KEY)
1062 ret = btrfs_del_item(trans, root, path);
1065 btrfs_release_path(root, path);
1066 inode = read_one_inode(root, key.offset);
1069 ret = fixup_inode_link_count(trans, root, inode);
1074 if (key.offset == 0)
1078 btrfs_release_path(root, path);
1084 * record a given inode in the fixup dir so we can check its link
1085 * count when replay is done. The link count is incremented here
1086 * so the inode won't go away until we check it
1088 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1089 struct btrfs_root *root,
1090 struct btrfs_path *path,
1093 struct btrfs_key key;
1095 struct inode *inode;
1097 inode = read_one_inode(root, objectid);
1100 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1101 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1102 key.offset = objectid;
1104 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1106 btrfs_release_path(root, path);
1108 btrfs_inc_nlink(inode);
1109 btrfs_update_inode(trans, root, inode);
1110 } else if (ret == -EEXIST) {
1121 * when replaying the log for a directory, we only insert names
1122 * for inodes that actually exist. This means an fsync on a directory
1123 * does not implicitly fsync all the new files in it
1125 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1126 struct btrfs_root *root,
1127 struct btrfs_path *path,
1128 u64 dirid, u64 index,
1129 char *name, int name_len, u8 type,
1130 struct btrfs_key *location)
1132 struct inode *inode;
1136 inode = read_one_inode(root, location->objectid);
1140 dir = read_one_inode(root, dirid);
1145 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1147 /* FIXME, put inode into FIXUP list */
1155 * take a single entry in a log directory item and replay it into
1158 * if a conflicting item exists in the subdirectory already,
1159 * the inode it points to is unlinked and put into the link count
1162 * If a name from the log points to a file or directory that does
1163 * not exist in the FS, it is skipped. fsyncs on directories
1164 * do not force down inodes inside that directory, just changes to the
1165 * names or unlinks in a directory.
1167 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1168 struct btrfs_root *root,
1169 struct btrfs_path *path,
1170 struct extent_buffer *eb,
1171 struct btrfs_dir_item *di,
1172 struct btrfs_key *key)
1176 struct btrfs_dir_item *dst_di;
1177 struct btrfs_key found_key;
1178 struct btrfs_key log_key;
1184 dir = read_one_inode(root, key->objectid);
1187 name_len = btrfs_dir_name_len(eb, di);
1188 name = kmalloc(name_len, GFP_NOFS);
1189 log_type = btrfs_dir_type(eb, di);
1190 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1193 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1194 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1199 btrfs_release_path(root, path);
1201 if (key->type == BTRFS_DIR_ITEM_KEY) {
1202 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1205 else if (key->type == BTRFS_DIR_INDEX_KEY) {
1206 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1213 if (!dst_di || IS_ERR(dst_di)) {
1214 /* we need a sequence number to insert, so we only
1215 * do inserts for the BTRFS_DIR_INDEX_KEY types
1217 if (key->type != BTRFS_DIR_INDEX_KEY)
1222 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1223 /* the existing item matches the logged item */
1224 if (found_key.objectid == log_key.objectid &&
1225 found_key.type == log_key.type &&
1226 found_key.offset == log_key.offset &&
1227 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1232 * don't drop the conflicting directory entry if the inode
1233 * for the new entry doesn't exist
1238 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1241 if (key->type == BTRFS_DIR_INDEX_KEY)
1244 btrfs_release_path(root, path);
1250 btrfs_release_path(root, path);
1251 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1252 name, name_len, log_type, &log_key);
1254 if (ret && ret != -ENOENT)
1260 * find all the names in a directory item and reconcile them into
1261 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1262 * one name in a directory item, but the same code gets used for
1263 * both directory index types
1265 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1266 struct btrfs_root *root,
1267 struct btrfs_path *path,
1268 struct extent_buffer *eb, int slot,
1269 struct btrfs_key *key)
1272 u32 item_size = btrfs_item_size_nr(eb, slot);
1273 struct btrfs_dir_item *di;
1276 unsigned long ptr_end;
1278 ptr = btrfs_item_ptr_offset(eb, slot);
1279 ptr_end = ptr + item_size;
1280 while(ptr < ptr_end) {
1281 di = (struct btrfs_dir_item *)ptr;
1282 name_len = btrfs_dir_name_len(eb, di);
1283 ret = replay_one_name(trans, root, path, eb, di, key);
1285 ptr = (unsigned long)(di + 1);
1292 * directory replay has two parts. There are the standard directory
1293 * items in the log copied from the subvolume, and range items
1294 * created in the log while the subvolume was logged.
1296 * The range items tell us which parts of the key space the log
1297 * is authoritative for. During replay, if a key in the subvolume
1298 * directory is in a logged range item, but not actually in the log
1299 * that means it was deleted from the directory before the fsync
1300 * and should be removed.
1302 static noinline int find_dir_range(struct btrfs_root *root,
1303 struct btrfs_path *path,
1304 u64 dirid, int key_type,
1305 u64 *start_ret, u64 *end_ret)
1307 struct btrfs_key key;
1309 struct btrfs_dir_log_item *item;
1313 if (*start_ret == (u64)-1)
1316 key.objectid = dirid;
1317 key.type = key_type;
1318 key.offset = *start_ret;
1320 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1324 if (path->slots[0] == 0)
1329 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1331 if (key.type != key_type || key.objectid != dirid) {
1335 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1336 struct btrfs_dir_log_item);
1337 found_end = btrfs_dir_log_end(path->nodes[0], item);
1339 if (*start_ret >= key.offset && *start_ret <= found_end) {
1341 *start_ret = key.offset;
1342 *end_ret = found_end;
1347 /* check the next slot in the tree to see if it is a valid item */
1348 nritems = btrfs_header_nritems(path->nodes[0]);
1349 if (path->slots[0] >= nritems) {
1350 ret = btrfs_next_leaf(root, path);
1357 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1359 if (key.type != key_type || key.objectid != dirid) {
1363 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1364 struct btrfs_dir_log_item);
1365 found_end = btrfs_dir_log_end(path->nodes[0], item);
1366 *start_ret = key.offset;
1367 *end_ret = found_end;
1370 btrfs_release_path(root, path);
1375 * this looks for a given directory item in the log. If the directory
1376 * item is not in the log, the item is removed and the inode it points
1379 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1380 struct btrfs_root *root,
1381 struct btrfs_root *log,
1382 struct btrfs_path *path,
1383 struct btrfs_path *log_path,
1385 struct btrfs_key *dir_key)
1388 struct extent_buffer *eb;
1391 struct btrfs_dir_item *di;
1392 struct btrfs_dir_item *log_di;
1395 unsigned long ptr_end;
1397 struct inode *inode;
1398 struct btrfs_key location;
1401 eb = path->nodes[0];
1402 slot = path->slots[0];
1403 item_size = btrfs_item_size_nr(eb, slot);
1404 ptr = btrfs_item_ptr_offset(eb, slot);
1405 ptr_end = ptr + item_size;
1406 while(ptr < ptr_end) {
1407 di = (struct btrfs_dir_item *)ptr;
1408 name_len = btrfs_dir_name_len(eb, di);
1409 name = kmalloc(name_len, GFP_NOFS);
1414 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1417 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1418 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1421 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1422 log_di = btrfs_lookup_dir_index_item(trans, log,
1428 if (!log_di || IS_ERR(log_di)) {
1429 btrfs_dir_item_key_to_cpu(eb, di, &location);
1430 btrfs_release_path(root, path);
1431 btrfs_release_path(log, log_path);
1432 inode = read_one_inode(root, location.objectid);
1435 ret = link_to_fixup_dir(trans, root,
1436 path, location.objectid);
1438 btrfs_inc_nlink(inode);
1439 ret = btrfs_unlink_inode(trans, root, dir, inode,
1445 /* there might still be more names under this key
1446 * check and repeat if required
1448 ret = btrfs_search_slot(NULL, root, dir_key, path,
1455 btrfs_release_path(log, log_path);
1458 ptr = (unsigned long)(di + 1);
1463 btrfs_release_path(root, path);
1464 btrfs_release_path(log, log_path);
1469 * deletion replay happens before we copy any new directory items
1470 * out of the log or out of backreferences from inodes. It
1471 * scans the log to find ranges of keys that log is authoritative for,
1472 * and then scans the directory to find items in those ranges that are
1473 * not present in the log.
1475 * Anything we don't find in the log is unlinked and removed from the
1478 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root,
1480 struct btrfs_root *log,
1481 struct btrfs_path *path,
1486 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1488 struct btrfs_key dir_key;
1489 struct btrfs_key found_key;
1490 struct btrfs_path *log_path;
1493 dir_key.objectid = dirid;
1494 dir_key.type = BTRFS_DIR_ITEM_KEY;
1495 log_path = btrfs_alloc_path();
1499 dir = read_one_inode(root, dirid);
1500 /* it isn't an error if the inode isn't there, that can happen
1501 * because we replay the deletes before we copy in the inode item
1505 btrfs_free_path(log_path);
1512 ret = find_dir_range(log, path, dirid, key_type,
1513 &range_start, &range_end);
1517 dir_key.offset = range_start;
1520 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1525 nritems = btrfs_header_nritems(path->nodes[0]);
1526 if (path->slots[0] >= nritems) {
1527 ret = btrfs_next_leaf(root, path);
1531 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1533 if (found_key.objectid != dirid ||
1534 found_key.type != dir_key.type)
1537 if (found_key.offset > range_end)
1540 ret = check_item_in_log(trans, root, log, path,
1541 log_path, dir, &found_key);
1543 if (found_key.offset == (u64)-1)
1545 dir_key.offset = found_key.offset + 1;
1547 btrfs_release_path(root, path);
1548 if (range_end == (u64)-1)
1550 range_start = range_end + 1;
1555 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1556 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1557 dir_key.type = BTRFS_DIR_INDEX_KEY;
1558 btrfs_release_path(root, path);
1562 btrfs_release_path(root, path);
1563 btrfs_free_path(log_path);
1569 * the process_func used to replay items from the log tree. This
1570 * gets called in two different stages. The first stage just looks
1571 * for inodes and makes sure they are all copied into the subvolume.
1573 * The second stage copies all the other item types from the log into
1574 * the subvolume. The two stage approach is slower, but gets rid of
1575 * lots of complexity around inodes referencing other inodes that exist
1576 * only in the log (references come from either directory items or inode
1579 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1580 struct walk_control *wc, u64 gen)
1583 struct btrfs_path *path;
1584 struct btrfs_root *root = wc->replay_dest;
1585 struct btrfs_key key;
1591 btrfs_read_buffer(eb, gen);
1593 level = btrfs_header_level(eb);
1598 path = btrfs_alloc_path();
1601 nritems = btrfs_header_nritems(eb);
1602 for (i = 0; i < nritems; i++) {
1603 btrfs_item_key_to_cpu(eb, &key, i);
1604 item_size = btrfs_item_size_nr(eb, i);
1606 /* inode keys are done during the first stage */
1607 if (key.type == BTRFS_INODE_ITEM_KEY &&
1608 wc->stage == LOG_WALK_REPLAY_INODES) {
1609 struct inode *inode;
1610 struct btrfs_inode_item *inode_item;
1613 inode_item = btrfs_item_ptr(eb, i,
1614 struct btrfs_inode_item);
1615 mode = btrfs_inode_mode(eb, inode_item);
1616 if (S_ISDIR(mode)) {
1617 ret = replay_dir_deletes(wc->trans,
1618 root, log, path, key.objectid);
1621 ret = overwrite_item(wc->trans, root, path,
1625 /* for regular files, truncate away
1626 * extents past the new EOF
1628 if (S_ISREG(mode)) {
1629 inode = read_one_inode(root,
1633 ret = btrfs_truncate_inode_items(wc->trans,
1634 root, inode, inode->i_size,
1635 BTRFS_EXTENT_DATA_KEY);
1639 ret = link_to_fixup_dir(wc->trans, root,
1640 path, key.objectid);
1643 if (wc->stage < LOG_WALK_REPLAY_ALL)
1646 /* these keys are simply copied */
1647 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1648 ret = overwrite_item(wc->trans, root, path,
1651 } else if (key.type == BTRFS_INODE_REF_KEY) {
1652 ret = add_inode_ref(wc->trans, root, log, path,
1654 BUG_ON(ret && ret != -ENOENT);
1655 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1656 ret = replay_one_extent(wc->trans, root, path,
1659 } else if (key.type == BTRFS_EXTENT_CSUM_KEY) {
1660 ret = replay_one_csum(wc->trans, root, path,
1663 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1664 key.type == BTRFS_DIR_INDEX_KEY) {
1665 ret = replay_one_dir_item(wc->trans, root, path,
1670 btrfs_free_path(path);
1674 static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1675 struct btrfs_root *root,
1676 struct btrfs_path *path, int *level,
1677 struct walk_control *wc)
1683 struct extent_buffer *next;
1684 struct extent_buffer *cur;
1685 struct extent_buffer *parent;
1689 WARN_ON(*level < 0);
1690 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1693 WARN_ON(*level < 0);
1694 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1695 cur = path->nodes[*level];
1697 if (btrfs_header_level(cur) != *level)
1700 if (path->slots[*level] >=
1701 btrfs_header_nritems(cur))
1704 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1705 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1706 blocksize = btrfs_level_size(root, *level - 1);
1708 parent = path->nodes[*level];
1709 root_owner = btrfs_header_owner(parent);
1710 root_gen = btrfs_header_generation(parent);
1712 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1714 wc->process_func(root, next, wc, ptr_gen);
1717 path->slots[*level]++;
1719 btrfs_read_buffer(next, ptr_gen);
1721 btrfs_tree_lock(next);
1722 clean_tree_block(trans, root, next);
1723 btrfs_wait_tree_block_writeback(next);
1724 btrfs_tree_unlock(next);
1726 ret = btrfs_drop_leaf_ref(trans, root, next);
1729 WARN_ON(root_owner !=
1730 BTRFS_TREE_LOG_OBJECTID);
1731 ret = btrfs_free_reserved_extent(root,
1735 free_extent_buffer(next);
1738 btrfs_read_buffer(next, ptr_gen);
1740 WARN_ON(*level <= 0);
1741 if (path->nodes[*level-1])
1742 free_extent_buffer(path->nodes[*level-1]);
1743 path->nodes[*level-1] = next;
1744 *level = btrfs_header_level(next);
1745 path->slots[*level] = 0;
1748 WARN_ON(*level < 0);
1749 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1751 if (path->nodes[*level] == root->node) {
1752 parent = path->nodes[*level];
1754 parent = path->nodes[*level + 1];
1756 bytenr = path->nodes[*level]->start;
1758 blocksize = btrfs_level_size(root, *level);
1759 root_owner = btrfs_header_owner(parent);
1760 root_gen = btrfs_header_generation(parent);
1762 wc->process_func(root, path->nodes[*level], wc,
1763 btrfs_header_generation(path->nodes[*level]));
1766 next = path->nodes[*level];
1767 btrfs_tree_lock(next);
1768 clean_tree_block(trans, root, next);
1769 btrfs_wait_tree_block_writeback(next);
1770 btrfs_tree_unlock(next);
1773 ret = btrfs_drop_leaf_ref(trans, root, next);
1776 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1777 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1780 free_extent_buffer(path->nodes[*level]);
1781 path->nodes[*level] = NULL;
1788 static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1789 struct btrfs_root *root,
1790 struct btrfs_path *path, int *level,
1791 struct walk_control *wc)
1799 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1800 slot = path->slots[i];
1801 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1802 struct extent_buffer *node;
1803 node = path->nodes[i];
1806 WARN_ON(*level == 0);
1809 struct extent_buffer *parent;
1810 if (path->nodes[*level] == root->node)
1811 parent = path->nodes[*level];
1813 parent = path->nodes[*level + 1];
1815 root_owner = btrfs_header_owner(parent);
1816 root_gen = btrfs_header_generation(parent);
1817 wc->process_func(root, path->nodes[*level], wc,
1818 btrfs_header_generation(path->nodes[*level]));
1820 struct extent_buffer *next;
1822 next = path->nodes[*level];
1824 btrfs_tree_lock(next);
1825 clean_tree_block(trans, root, next);
1826 btrfs_wait_tree_block_writeback(next);
1827 btrfs_tree_unlock(next);
1830 ret = btrfs_drop_leaf_ref(trans, root,
1835 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1836 ret = btrfs_free_reserved_extent(root,
1837 path->nodes[*level]->start,
1838 path->nodes[*level]->len);
1841 free_extent_buffer(path->nodes[*level]);
1842 path->nodes[*level] = NULL;
1850 * drop the reference count on the tree rooted at 'snap'. This traverses
1851 * the tree freeing any blocks that have a ref count of zero after being
1854 static int walk_log_tree(struct btrfs_trans_handle *trans,
1855 struct btrfs_root *log, struct walk_control *wc)
1860 struct btrfs_path *path;
1864 path = btrfs_alloc_path();
1867 level = btrfs_header_level(log->node);
1869 path->nodes[level] = log->node;
1870 extent_buffer_get(log->node);
1871 path->slots[level] = 0;
1874 wret = walk_down_log_tree(trans, log, path, &level, wc);
1880 wret = walk_up_log_tree(trans, log, path, &level, wc);
1887 /* was the root node processed? if not, catch it here */
1888 if (path->nodes[orig_level]) {
1889 wc->process_func(log, path->nodes[orig_level], wc,
1890 btrfs_header_generation(path->nodes[orig_level]));
1892 struct extent_buffer *next;
1894 next = path->nodes[orig_level];
1896 btrfs_tree_lock(next);
1897 clean_tree_block(trans, log, next);
1898 btrfs_wait_tree_block_writeback(next);
1899 btrfs_tree_unlock(next);
1901 if (orig_level == 0) {
1902 ret = btrfs_drop_leaf_ref(trans, log,
1906 WARN_ON(log->root_key.objectid !=
1907 BTRFS_TREE_LOG_OBJECTID);
1908 ret = btrfs_free_reserved_extent(log, next->start,
1914 for (i = 0; i <= orig_level; i++) {
1915 if (path->nodes[i]) {
1916 free_extent_buffer(path->nodes[i]);
1917 path->nodes[i] = NULL;
1920 btrfs_free_path(path);
1922 free_extent_buffer(log->node);
1926 static int wait_log_commit(struct btrfs_root *log)
1929 u64 transid = log->fs_info->tree_log_transid;
1932 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1933 TASK_UNINTERRUPTIBLE);
1934 mutex_unlock(&log->fs_info->tree_log_mutex);
1935 if (atomic_read(&log->fs_info->tree_log_commit))
1937 finish_wait(&log->fs_info->tree_log_wait, &wait);
1938 mutex_lock(&log->fs_info->tree_log_mutex);
1939 } while(transid == log->fs_info->tree_log_transid &&
1940 atomic_read(&log->fs_info->tree_log_commit));
1945 * btrfs_sync_log does sends a given tree log down to the disk and
1946 * updates the super blocks to record it. When this call is done,
1947 * you know that any inodes previously logged are safely on disk
1949 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1950 struct btrfs_root *root)
1953 unsigned long batch;
1954 struct btrfs_root *log = root->log_root;
1956 mutex_lock(&log->fs_info->tree_log_mutex);
1957 if (atomic_read(&log->fs_info->tree_log_commit)) {
1958 wait_log_commit(log);
1961 atomic_set(&log->fs_info->tree_log_commit, 1);
1964 batch = log->fs_info->tree_log_batch;
1965 mutex_unlock(&log->fs_info->tree_log_mutex);
1966 schedule_timeout_uninterruptible(1);
1967 mutex_lock(&log->fs_info->tree_log_mutex);
1969 while(atomic_read(&log->fs_info->tree_log_writers)) {
1971 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1972 TASK_UNINTERRUPTIBLE);
1973 mutex_unlock(&log->fs_info->tree_log_mutex);
1974 if (atomic_read(&log->fs_info->tree_log_writers))
1976 mutex_lock(&log->fs_info->tree_log_mutex);
1977 finish_wait(&log->fs_info->tree_log_wait, &wait);
1979 if (batch == log->fs_info->tree_log_batch)
1983 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1985 ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
1986 &root->fs_info->log_root_tree->dirty_log_pages);
1989 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
1990 log->fs_info->log_root_tree->node->start);
1991 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
1992 btrfs_header_level(log->fs_info->log_root_tree->node));
1994 write_ctree_super(trans, log->fs_info->tree_root, 2);
1995 log->fs_info->tree_log_transid++;
1996 log->fs_info->tree_log_batch = 0;
1997 atomic_set(&log->fs_info->tree_log_commit, 0);
1999 if (waitqueue_active(&log->fs_info->tree_log_wait))
2000 wake_up(&log->fs_info->tree_log_wait);
2002 mutex_unlock(&log->fs_info->tree_log_mutex);
2006 /* * free all the extents used by the tree log. This should be called
2007 * at commit time of the full transaction
2009 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2012 struct btrfs_root *log;
2016 struct walk_control wc = {
2018 .process_func = process_one_buffer
2021 if (!root->log_root)
2024 log = root->log_root;
2025 ret = walk_log_tree(trans, log, &wc);
2029 ret = find_first_extent_bit(&log->dirty_log_pages,
2030 0, &start, &end, EXTENT_DIRTY);
2034 clear_extent_dirty(&log->dirty_log_pages,
2035 start, end, GFP_NOFS);
2038 log = root->log_root;
2039 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2042 root->log_root = NULL;
2043 kfree(root->log_root);
2048 * helper function to update the item for a given subvolumes log root
2049 * in the tree of log roots
2051 static int update_log_root(struct btrfs_trans_handle *trans,
2052 struct btrfs_root *log)
2054 u64 bytenr = btrfs_root_bytenr(&log->root_item);
2057 if (log->node->start == bytenr)
2060 btrfs_set_root_bytenr(&log->root_item, log->node->start);
2061 btrfs_set_root_generation(&log->root_item, trans->transid);
2062 btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2063 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2064 &log->root_key, &log->root_item);
2070 * If both a file and directory are logged, and unlinks or renames are
2071 * mixed in, we have a few interesting corners:
2073 * create file X in dir Y
2074 * link file X to X.link in dir Y
2076 * unlink file X but leave X.link
2079 * After a crash we would expect only X.link to exist. But file X
2080 * didn't get fsync'd again so the log has back refs for X and X.link.
2082 * We solve this by removing directory entries and inode backrefs from the
2083 * log when a file that was logged in the current transaction is
2084 * unlinked. Any later fsync will include the updated log entries, and
2085 * we'll be able to reconstruct the proper directory items from backrefs.
2087 * This optimizations allows us to avoid relogging the entire inode
2088 * or the entire directory.
2090 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2091 struct btrfs_root *root,
2092 const char *name, int name_len,
2093 struct inode *dir, u64 index)
2095 struct btrfs_root *log;
2096 struct btrfs_dir_item *di;
2097 struct btrfs_path *path;
2101 if (BTRFS_I(dir)->logged_trans < trans->transid)
2104 ret = join_running_log_trans(root);
2108 mutex_lock(&BTRFS_I(dir)->log_mutex);
2110 log = root->log_root;
2111 path = btrfs_alloc_path();
2112 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2113 name, name_len, -1);
2114 if (di && !IS_ERR(di)) {
2115 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2116 bytes_del += name_len;
2119 btrfs_release_path(log, path);
2120 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2121 index, name, name_len, -1);
2122 if (di && !IS_ERR(di)) {
2123 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2124 bytes_del += name_len;
2128 /* update the directory size in the log to reflect the names
2132 struct btrfs_key key;
2134 key.objectid = dir->i_ino;
2136 key.type = BTRFS_INODE_ITEM_KEY;
2137 btrfs_release_path(log, path);
2139 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2141 struct btrfs_inode_item *item;
2144 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2145 struct btrfs_inode_item);
2146 i_size = btrfs_inode_size(path->nodes[0], item);
2147 if (i_size > bytes_del)
2148 i_size -= bytes_del;
2151 btrfs_set_inode_size(path->nodes[0], item, i_size);
2152 btrfs_mark_buffer_dirty(path->nodes[0]);
2155 btrfs_release_path(log, path);
2158 btrfs_free_path(path);
2159 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2160 end_log_trans(root);
2165 /* see comments for btrfs_del_dir_entries_in_log */
2166 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2167 struct btrfs_root *root,
2168 const char *name, int name_len,
2169 struct inode *inode, u64 dirid)
2171 struct btrfs_root *log;
2175 if (BTRFS_I(inode)->logged_trans < trans->transid)
2178 ret = join_running_log_trans(root);
2181 log = root->log_root;
2182 mutex_lock(&BTRFS_I(inode)->log_mutex);
2184 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2186 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2187 end_log_trans(root);
2193 * creates a range item in the log for 'dirid'. first_offset and
2194 * last_offset tell us which parts of the key space the log should
2195 * be considered authoritative for.
2197 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2198 struct btrfs_root *log,
2199 struct btrfs_path *path,
2200 int key_type, u64 dirid,
2201 u64 first_offset, u64 last_offset)
2204 struct btrfs_key key;
2205 struct btrfs_dir_log_item *item;
2207 key.objectid = dirid;
2208 key.offset = first_offset;
2209 if (key_type == BTRFS_DIR_ITEM_KEY)
2210 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2212 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2213 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2216 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2217 struct btrfs_dir_log_item);
2218 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2219 btrfs_mark_buffer_dirty(path->nodes[0]);
2220 btrfs_release_path(log, path);
2225 * log all the items included in the current transaction for a given
2226 * directory. This also creates the range items in the log tree required
2227 * to replay anything deleted before the fsync
2229 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2230 struct btrfs_root *root, struct inode *inode,
2231 struct btrfs_path *path,
2232 struct btrfs_path *dst_path, int key_type,
2233 u64 min_offset, u64 *last_offset_ret)
2235 struct btrfs_key min_key;
2236 struct btrfs_key max_key;
2237 struct btrfs_root *log = root->log_root;
2238 struct extent_buffer *src;
2242 u64 first_offset = min_offset;
2243 u64 last_offset = (u64)-1;
2245 log = root->log_root;
2246 max_key.objectid = inode->i_ino;
2247 max_key.offset = (u64)-1;
2248 max_key.type = key_type;
2250 min_key.objectid = inode->i_ino;
2251 min_key.type = key_type;
2252 min_key.offset = min_offset;
2254 path->keep_locks = 1;
2256 ret = btrfs_search_forward(root, &min_key, &max_key,
2257 path, 0, trans->transid);
2260 * we didn't find anything from this transaction, see if there
2261 * is anything at all
2263 if (ret != 0 || min_key.objectid != inode->i_ino ||
2264 min_key.type != key_type) {
2265 min_key.objectid = inode->i_ino;
2266 min_key.type = key_type;
2267 min_key.offset = (u64)-1;
2268 btrfs_release_path(root, path);
2269 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2271 btrfs_release_path(root, path);
2274 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2276 /* if ret == 0 there are items for this type,
2277 * create a range to tell us the last key of this type.
2278 * otherwise, there are no items in this directory after
2279 * *min_offset, and we create a range to indicate that.
2282 struct btrfs_key tmp;
2283 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2285 if (key_type == tmp.type) {
2286 first_offset = max(min_offset, tmp.offset) + 1;
2292 /* go backward to find any previous key */
2293 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2295 struct btrfs_key tmp;
2296 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2297 if (key_type == tmp.type) {
2298 first_offset = tmp.offset;
2299 ret = overwrite_item(trans, log, dst_path,
2300 path->nodes[0], path->slots[0],
2304 btrfs_release_path(root, path);
2306 /* find the first key from this transaction again */
2307 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2314 * we have a block from this transaction, log every item in it
2315 * from our directory
2318 struct btrfs_key tmp;
2319 src = path->nodes[0];
2320 nritems = btrfs_header_nritems(src);
2321 for (i = path->slots[0]; i < nritems; i++) {
2322 btrfs_item_key_to_cpu(src, &min_key, i);
2324 if (min_key.objectid != inode->i_ino ||
2325 min_key.type != key_type)
2327 ret = overwrite_item(trans, log, dst_path, src, i,
2331 path->slots[0] = nritems;
2334 * look ahead to the next item and see if it is also
2335 * from this directory and from this transaction
2337 ret = btrfs_next_leaf(root, path);
2339 last_offset = (u64)-1;
2342 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2343 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2344 last_offset = (u64)-1;
2347 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2348 ret = overwrite_item(trans, log, dst_path,
2349 path->nodes[0], path->slots[0],
2353 last_offset = tmp.offset;
2358 *last_offset_ret = last_offset;
2359 btrfs_release_path(root, path);
2360 btrfs_release_path(log, dst_path);
2362 /* insert the log range keys to indicate where the log is valid */
2363 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2364 first_offset, last_offset);
2370 * logging directories is very similar to logging inodes, We find all the items
2371 * from the current transaction and write them to the log.
2373 * The recovery code scans the directory in the subvolume, and if it finds a
2374 * key in the range logged that is not present in the log tree, then it means
2375 * that dir entry was unlinked during the transaction.
2377 * In order for that scan to work, we must include one key smaller than
2378 * the smallest logged by this transaction and one key larger than the largest
2379 * key logged by this transaction.
2381 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2382 struct btrfs_root *root, struct inode *inode,
2383 struct btrfs_path *path,
2384 struct btrfs_path *dst_path)
2389 int key_type = BTRFS_DIR_ITEM_KEY;
2395 ret = log_dir_items(trans, root, inode, path,
2396 dst_path, key_type, min_key,
2399 if (max_key == (u64)-1)
2401 min_key = max_key + 1;
2404 if (key_type == BTRFS_DIR_ITEM_KEY) {
2405 key_type = BTRFS_DIR_INDEX_KEY;
2412 * a helper function to drop items from the log before we relog an
2413 * inode. max_key_type indicates the highest item type to remove.
2414 * This cannot be run for file data extents because it does not
2415 * free the extents they point to.
2417 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2418 struct btrfs_root *log,
2419 struct btrfs_path *path,
2420 u64 objectid, int max_key_type)
2423 struct btrfs_key key;
2424 struct btrfs_key found_key;
2426 key.objectid = objectid;
2427 key.type = max_key_type;
2428 key.offset = (u64)-1;
2431 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2436 if (path->slots[0] == 0)
2440 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2443 if (found_key.objectid != objectid)
2446 ret = btrfs_del_item(trans, log, path);
2448 btrfs_release_path(log, path);
2450 btrfs_release_path(log, path);
2454 static noinline int copy_extent_csums(struct btrfs_trans_handle *trans,
2455 struct list_head *list,
2456 struct btrfs_root *root,
2457 u64 disk_bytenr, u64 len)
2459 struct btrfs_ordered_sum *sums;
2460 struct btrfs_sector_sum *sector_sum;
2462 struct btrfs_path *path;
2463 struct btrfs_csum_item *item = NULL;
2464 u64 end = disk_bytenr + len;
2465 u64 item_start_offset = 0;
2466 u64 item_last_offset = 0;
2469 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
2471 sums = kzalloc(btrfs_ordered_sum_size(root, len), GFP_NOFS);
2473 sector_sum = sums->sums;
2474 sums->bytenr = disk_bytenr;
2476 list_add_tail(&sums->list, list);
2478 path = btrfs_alloc_path();
2479 while(disk_bytenr < end) {
2480 if (!item || disk_bytenr < item_start_offset ||
2481 disk_bytenr >= item_last_offset) {
2482 struct btrfs_key found_key;
2486 btrfs_release_path(root, path);
2487 item = btrfs_lookup_csum(NULL, root, path,
2490 ret = PTR_ERR(item);
2491 if (ret == -ENOENT || ret == -EFBIG)
2494 printk("log no csum found for byte %llu\n",
2495 (unsigned long long)disk_bytenr);
2497 btrfs_release_path(root, path);
2500 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2503 item_start_offset = found_key.offset;
2504 item_size = btrfs_item_size_nr(path->nodes[0],
2506 item_last_offset = item_start_offset +
2507 (item_size / csum_size) *
2509 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2510 struct btrfs_csum_item);
2513 * this byte range must be able to fit inside
2514 * a single leaf so it will also fit inside a u32
2516 diff = disk_bytenr - item_start_offset;
2517 diff = diff / root->sectorsize;
2518 diff = diff * csum_size;
2520 read_extent_buffer(path->nodes[0], &sum,
2521 ((unsigned long)item) + diff,
2524 sector_sum->bytenr = disk_bytenr;
2525 sector_sum->sum = sum;
2526 disk_bytenr += root->sectorsize;
2529 btrfs_free_path(path);
2533 static noinline int copy_items(struct btrfs_trans_handle *trans,
2534 struct btrfs_root *log,
2535 struct btrfs_path *dst_path,
2536 struct extent_buffer *src,
2537 int start_slot, int nr, int inode_only)
2539 unsigned long src_offset;
2540 unsigned long dst_offset;
2541 struct btrfs_file_extent_item *extent;
2542 struct btrfs_inode_item *inode_item;
2544 struct btrfs_key *ins_keys;
2548 struct list_head ordered_sums;
2550 INIT_LIST_HEAD(&ordered_sums);
2552 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2553 nr * sizeof(u32), GFP_NOFS);
2554 ins_sizes = (u32 *)ins_data;
2555 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2557 for (i = 0; i < nr; i++) {
2558 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2559 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2561 ret = btrfs_insert_empty_items(trans, log, dst_path,
2562 ins_keys, ins_sizes, nr);
2565 for (i = 0; i < nr; i++) {
2566 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2567 dst_path->slots[0]);
2569 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2571 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2572 src_offset, ins_sizes[i]);
2574 if (inode_only == LOG_INODE_EXISTS &&
2575 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2576 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2578 struct btrfs_inode_item);
2579 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2581 /* set the generation to zero so the recover code
2582 * can tell the difference between an logging
2583 * just to say 'this inode exists' and a logging
2584 * to say 'update this inode with these values'
2586 btrfs_set_inode_generation(dst_path->nodes[0],
2589 /* take a reference on file data extents so that truncates
2590 * or deletes of this inode don't have to relog the inode
2593 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2595 extent = btrfs_item_ptr(src, start_slot + i,
2596 struct btrfs_file_extent_item);
2598 found_type = btrfs_file_extent_type(src, extent);
2599 if (found_type == BTRFS_FILE_EXTENT_REG ||
2600 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2601 u64 ds = btrfs_file_extent_disk_bytenr(src,
2603 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2605 u64 cs = btrfs_file_extent_offset(src, extent);
2606 u64 cl = btrfs_file_extent_num_bytes(src,
2608 if (btrfs_file_extent_compression(src,
2613 /* ds == 0 is a hole */
2615 ret = btrfs_inc_extent_ref(trans, log,
2617 dst_path->nodes[0]->start,
2618 BTRFS_TREE_LOG_OBJECTID,
2620 ins_keys[i].objectid);
2622 ret = copy_extent_csums(trans,
2624 log->fs_info->csum_root,
2630 dst_path->slots[0]++;
2633 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2634 btrfs_release_path(log, dst_path);
2638 * we have to do this after the loop above to avoid changing the
2639 * log tree while trying to change the log tree.
2641 while(!list_empty(&ordered_sums)) {
2642 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2643 struct btrfs_ordered_sum,
2645 ret = btrfs_csum_file_blocks(trans, log, sums);
2647 list_del(&sums->list);
2653 /* log a single inode in the tree log.
2654 * At least one parent directory for this inode must exist in the tree
2655 * or be logged already.
2657 * Any items from this inode changed by the current transaction are copied
2658 * to the log tree. An extra reference is taken on any extents in this
2659 * file, allowing us to avoid a whole pile of corner cases around logging
2660 * blocks that have been removed from the tree.
2662 * See LOG_INODE_ALL and related defines for a description of what inode_only
2665 * This handles both files and directories.
2667 static int __btrfs_log_inode(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root, struct inode *inode,
2671 struct btrfs_path *path;
2672 struct btrfs_path *dst_path;
2673 struct btrfs_key min_key;
2674 struct btrfs_key max_key;
2675 struct btrfs_root *log = root->log_root;
2676 struct extent_buffer *src = NULL;
2680 int ins_start_slot = 0;
2683 log = root->log_root;
2685 path = btrfs_alloc_path();
2686 dst_path = btrfs_alloc_path();
2688 min_key.objectid = inode->i_ino;
2689 min_key.type = BTRFS_INODE_ITEM_KEY;
2692 max_key.objectid = inode->i_ino;
2693 if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2694 max_key.type = BTRFS_XATTR_ITEM_KEY;
2696 max_key.type = (u8)-1;
2697 max_key.offset = (u64)-1;
2700 * if this inode has already been logged and we're in inode_only
2701 * mode, we don't want to delete the things that have already
2702 * been written to the log.
2704 * But, if the inode has been through an inode_only log,
2705 * the logged_trans field is not set. This allows us to catch
2706 * any new names for this inode in the backrefs by logging it
2709 if (inode_only == LOG_INODE_EXISTS &&
2710 BTRFS_I(inode)->logged_trans == trans->transid) {
2711 btrfs_free_path(path);
2712 btrfs_free_path(dst_path);
2715 mutex_lock(&BTRFS_I(inode)->log_mutex);
2718 * a brute force approach to making sure we get the most uptodate
2719 * copies of everything.
2721 if (S_ISDIR(inode->i_mode)) {
2722 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2724 if (inode_only == LOG_INODE_EXISTS)
2725 max_key_type = BTRFS_XATTR_ITEM_KEY;
2726 ret = drop_objectid_items(trans, log, path,
2727 inode->i_ino, max_key_type);
2729 ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2732 path->keep_locks = 1;
2736 ret = btrfs_search_forward(root, &min_key, &max_key,
2737 path, 0, trans->transid);
2741 /* note, ins_nr might be > 0 here, cleanup outside the loop */
2742 if (min_key.objectid != inode->i_ino)
2744 if (min_key.type > max_key.type)
2747 src = path->nodes[0];
2748 size = btrfs_item_size_nr(src, path->slots[0]);
2749 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2752 } else if (!ins_nr) {
2753 ins_start_slot = path->slots[0];
2758 ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2759 ins_nr, inode_only);
2762 ins_start_slot = path->slots[0];
2765 nritems = btrfs_header_nritems(path->nodes[0]);
2767 if (path->slots[0] < nritems) {
2768 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2773 ret = copy_items(trans, log, dst_path, src,
2775 ins_nr, inode_only);
2779 btrfs_release_path(root, path);
2781 if (min_key.offset < (u64)-1)
2783 else if (min_key.type < (u8)-1)
2785 else if (min_key.objectid < (u64)-1)
2791 ret = copy_items(trans, log, dst_path, src,
2793 ins_nr, inode_only);
2798 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2799 btrfs_release_path(root, path);
2800 btrfs_release_path(log, dst_path);
2801 BTRFS_I(inode)->log_dirty_trans = 0;
2802 ret = log_directory_changes(trans, root, inode, path, dst_path);
2805 BTRFS_I(inode)->logged_trans = trans->transid;
2806 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2808 btrfs_free_path(path);
2809 btrfs_free_path(dst_path);
2811 mutex_lock(&root->fs_info->tree_log_mutex);
2812 ret = update_log_root(trans, log);
2814 mutex_unlock(&root->fs_info->tree_log_mutex);
2819 int btrfs_log_inode(struct btrfs_trans_handle *trans,
2820 struct btrfs_root *root, struct inode *inode,
2825 start_log_trans(trans, root);
2826 ret = __btrfs_log_inode(trans, root, inode, inode_only);
2827 end_log_trans(root);
2832 * helper function around btrfs_log_inode to make sure newly created
2833 * parent directories also end up in the log. A minimal inode and backref
2834 * only logging is done of any parent directories that are older than
2835 * the last committed transaction
2837 int btrfs_log_dentry(struct btrfs_trans_handle *trans,
2838 struct btrfs_root *root, struct dentry *dentry)
2840 int inode_only = LOG_INODE_ALL;
2841 struct super_block *sb;
2844 start_log_trans(trans, root);
2845 sb = dentry->d_inode->i_sb;
2847 ret = __btrfs_log_inode(trans, root, dentry->d_inode,
2850 inode_only = LOG_INODE_EXISTS;
2852 dentry = dentry->d_parent;
2853 if (!dentry || !dentry->d_inode || sb != dentry->d_inode->i_sb)
2856 if (BTRFS_I(dentry->d_inode)->generation <=
2857 root->fs_info->last_trans_committed)
2860 end_log_trans(root);
2865 * it is not safe to log dentry if the chunk root has added new
2866 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
2867 * If this returns 1, you must commit the transaction to safely get your
2870 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
2871 struct btrfs_root *root, struct dentry *dentry)
2874 gen = root->fs_info->last_trans_new_blockgroup;
2875 if (gen > root->fs_info->last_trans_committed)
2878 return btrfs_log_dentry(trans, root, dentry);
2882 * should be called during mount to recover any replay any log trees
2885 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
2888 struct btrfs_path *path;
2889 struct btrfs_trans_handle *trans;
2890 struct btrfs_key key;
2891 struct btrfs_key found_key;
2892 struct btrfs_key tmp_key;
2893 struct btrfs_root *log;
2894 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
2896 struct walk_control wc = {
2897 .process_func = process_one_buffer,
2901 fs_info->log_root_recovering = 1;
2902 path = btrfs_alloc_path();
2905 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2910 walk_log_tree(trans, log_root_tree, &wc);
2913 key.objectid = BTRFS_TREE_LOG_OBJECTID;
2914 key.offset = (u64)-1;
2915 btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2918 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
2922 if (path->slots[0] == 0)
2926 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2928 btrfs_release_path(log_root_tree, path);
2929 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
2932 log = btrfs_read_fs_root_no_radix(log_root_tree,
2937 tmp_key.objectid = found_key.offset;
2938 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
2939 tmp_key.offset = (u64)-1;
2941 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
2943 BUG_ON(!wc.replay_dest);
2945 btrfs_record_root_in_trans(wc.replay_dest);
2946 ret = walk_log_tree(trans, log, &wc);
2949 if (wc.stage == LOG_WALK_REPLAY_ALL) {
2950 ret = fixup_inode_link_counts(trans, wc.replay_dest,
2954 ret = btrfs_find_highest_inode(wc.replay_dest, &highest_inode);
2956 wc.replay_dest->highest_inode = highest_inode;
2957 wc.replay_dest->last_inode_alloc = highest_inode;
2960 key.offset = found_key.offset - 1;
2961 free_extent_buffer(log->node);
2964 if (found_key.offset == 0)
2967 btrfs_release_path(log_root_tree, path);
2969 /* step one is to pin it all, step two is to replay just inodes */
2972 wc.process_func = replay_one_buffer;
2973 wc.stage = LOG_WALK_REPLAY_INODES;
2976 /* step three is to replay everything */
2977 if (wc.stage < LOG_WALK_REPLAY_ALL) {
2982 btrfs_free_path(path);
2984 free_extent_buffer(log_root_tree->node);
2985 log_root_tree->log_root = NULL;
2986 fs_info->log_root_recovering = 0;
2988 /* step 4: commit the transaction, which also unpins the blocks */
2989 btrfs_commit_transaction(trans, fs_info->tree_root);
2991 kfree(log_root_tree);