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);
933 btrfs_super_csum_size(&root->fs_info->super_copy);
934 unsigned long file_bytes;
935 struct btrfs_ordered_sum *sums;
936 struct btrfs_sector_sum *sector_sum;
939 file_bytes = (item_size / csum_size) * root->sectorsize;
940 sums = kzalloc(btrfs_ordered_sum_size(root, file_bytes), GFP_NOFS);
945 INIT_LIST_HEAD(&sums->list);
946 sums->len = file_bytes;
947 sums->bytenr = key->offset;
950 * copy all the sums into the ordered sum struct
952 sector_sum = sums->sums;
953 cur_offset = key->offset;
954 ptr = btrfs_item_ptr_offset(eb, slot);
955 while(item_size > 0) {
956 sector_sum->bytenr = cur_offset;
957 read_extent_buffer(eb, §or_sum->sum, ptr, csum_size);
959 item_size -= csum_size;
961 cur_offset += root->sectorsize;
964 /* let btrfs_csum_file_blocks add them into the file */
965 ret = btrfs_csum_file_blocks(trans, root->fs_info->csum_root, sums);
971 * There are a few corners where the link count of the file can't
972 * be properly maintained during replay. So, instead of adding
973 * lots of complexity to the log code, we just scan the backrefs
974 * for any file that has been through replay.
976 * The scan will update the link count on the inode to reflect the
977 * number of back refs found. If it goes down to zero, the iput
978 * will free the inode.
980 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
981 struct btrfs_root *root,
984 struct btrfs_path *path;
986 struct btrfs_key key;
989 unsigned long ptr_end;
992 key.objectid = inode->i_ino;
993 key.type = BTRFS_INODE_REF_KEY;
994 key.offset = (u64)-1;
996 path = btrfs_alloc_path();
999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1003 if (path->slots[0] == 0)
1007 btrfs_item_key_to_cpu(path->nodes[0], &key,
1009 if (key.objectid != inode->i_ino ||
1010 key.type != BTRFS_INODE_REF_KEY)
1012 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1013 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1015 while(ptr < ptr_end) {
1016 struct btrfs_inode_ref *ref;
1018 ref = (struct btrfs_inode_ref *)ptr;
1019 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1021 ptr = (unsigned long)(ref + 1) + name_len;
1025 if (key.offset == 0)
1028 btrfs_release_path(root, path);
1030 btrfs_free_path(path);
1031 if (nlink != inode->i_nlink) {
1032 inode->i_nlink = nlink;
1033 btrfs_update_inode(trans, root, inode);
1035 BTRFS_I(inode)->index_cnt = (u64)-1;
1040 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1041 struct btrfs_root *root,
1042 struct btrfs_path *path)
1045 struct btrfs_key key;
1046 struct inode *inode;
1048 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1049 key.type = BTRFS_ORPHAN_ITEM_KEY;
1050 key.offset = (u64)-1;
1052 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1057 if (path->slots[0] == 0)
1062 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1063 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1064 key.type != BTRFS_ORPHAN_ITEM_KEY)
1067 ret = btrfs_del_item(trans, root, path);
1070 btrfs_release_path(root, path);
1071 inode = read_one_inode(root, key.offset);
1074 ret = fixup_inode_link_count(trans, root, inode);
1079 if (key.offset == 0)
1083 btrfs_release_path(root, path);
1089 * record a given inode in the fixup dir so we can check its link
1090 * count when replay is done. The link count is incremented here
1091 * so the inode won't go away until we check it
1093 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1094 struct btrfs_root *root,
1095 struct btrfs_path *path,
1098 struct btrfs_key key;
1100 struct inode *inode;
1102 inode = read_one_inode(root, objectid);
1105 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1106 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1107 key.offset = objectid;
1109 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1111 btrfs_release_path(root, path);
1113 btrfs_inc_nlink(inode);
1114 btrfs_update_inode(trans, root, inode);
1115 } else if (ret == -EEXIST) {
1126 * when replaying the log for a directory, we only insert names
1127 * for inodes that actually exist. This means an fsync on a directory
1128 * does not implicitly fsync all the new files in it
1130 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1131 struct btrfs_root *root,
1132 struct btrfs_path *path,
1133 u64 dirid, u64 index,
1134 char *name, int name_len, u8 type,
1135 struct btrfs_key *location)
1137 struct inode *inode;
1141 inode = read_one_inode(root, location->objectid);
1145 dir = read_one_inode(root, dirid);
1150 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1152 /* FIXME, put inode into FIXUP list */
1160 * take a single entry in a log directory item and replay it into
1163 * if a conflicting item exists in the subdirectory already,
1164 * the inode it points to is unlinked and put into the link count
1167 * If a name from the log points to a file or directory that does
1168 * not exist in the FS, it is skipped. fsyncs on directories
1169 * do not force down inodes inside that directory, just changes to the
1170 * names or unlinks in a directory.
1172 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1173 struct btrfs_root *root,
1174 struct btrfs_path *path,
1175 struct extent_buffer *eb,
1176 struct btrfs_dir_item *di,
1177 struct btrfs_key *key)
1181 struct btrfs_dir_item *dst_di;
1182 struct btrfs_key found_key;
1183 struct btrfs_key log_key;
1189 dir = read_one_inode(root, key->objectid);
1192 name_len = btrfs_dir_name_len(eb, di);
1193 name = kmalloc(name_len, GFP_NOFS);
1194 log_type = btrfs_dir_type(eb, di);
1195 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1198 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1199 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1204 btrfs_release_path(root, path);
1206 if (key->type == BTRFS_DIR_ITEM_KEY) {
1207 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1210 else if (key->type == BTRFS_DIR_INDEX_KEY) {
1211 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1218 if (!dst_di || IS_ERR(dst_di)) {
1219 /* we need a sequence number to insert, so we only
1220 * do inserts for the BTRFS_DIR_INDEX_KEY types
1222 if (key->type != BTRFS_DIR_INDEX_KEY)
1227 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1228 /* the existing item matches the logged item */
1229 if (found_key.objectid == log_key.objectid &&
1230 found_key.type == log_key.type &&
1231 found_key.offset == log_key.offset &&
1232 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1237 * don't drop the conflicting directory entry if the inode
1238 * for the new entry doesn't exist
1243 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1246 if (key->type == BTRFS_DIR_INDEX_KEY)
1249 btrfs_release_path(root, path);
1255 btrfs_release_path(root, path);
1256 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1257 name, name_len, log_type, &log_key);
1259 if (ret && ret != -ENOENT)
1265 * find all the names in a directory item and reconcile them into
1266 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1267 * one name in a directory item, but the same code gets used for
1268 * both directory index types
1270 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root,
1272 struct btrfs_path *path,
1273 struct extent_buffer *eb, int slot,
1274 struct btrfs_key *key)
1277 u32 item_size = btrfs_item_size_nr(eb, slot);
1278 struct btrfs_dir_item *di;
1281 unsigned long ptr_end;
1283 ptr = btrfs_item_ptr_offset(eb, slot);
1284 ptr_end = ptr + item_size;
1285 while(ptr < ptr_end) {
1286 di = (struct btrfs_dir_item *)ptr;
1287 name_len = btrfs_dir_name_len(eb, di);
1288 ret = replay_one_name(trans, root, path, eb, di, key);
1290 ptr = (unsigned long)(di + 1);
1297 * directory replay has two parts. There are the standard directory
1298 * items in the log copied from the subvolume, and range items
1299 * created in the log while the subvolume was logged.
1301 * The range items tell us which parts of the key space the log
1302 * is authoritative for. During replay, if a key in the subvolume
1303 * directory is in a logged range item, but not actually in the log
1304 * that means it was deleted from the directory before the fsync
1305 * and should be removed.
1307 static noinline int find_dir_range(struct btrfs_root *root,
1308 struct btrfs_path *path,
1309 u64 dirid, int key_type,
1310 u64 *start_ret, u64 *end_ret)
1312 struct btrfs_key key;
1314 struct btrfs_dir_log_item *item;
1318 if (*start_ret == (u64)-1)
1321 key.objectid = dirid;
1322 key.type = key_type;
1323 key.offset = *start_ret;
1325 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1329 if (path->slots[0] == 0)
1334 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1336 if (key.type != key_type || key.objectid != dirid) {
1340 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1341 struct btrfs_dir_log_item);
1342 found_end = btrfs_dir_log_end(path->nodes[0], item);
1344 if (*start_ret >= key.offset && *start_ret <= found_end) {
1346 *start_ret = key.offset;
1347 *end_ret = found_end;
1352 /* check the next slot in the tree to see if it is a valid item */
1353 nritems = btrfs_header_nritems(path->nodes[0]);
1354 if (path->slots[0] >= nritems) {
1355 ret = btrfs_next_leaf(root, path);
1362 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1364 if (key.type != key_type || key.objectid != dirid) {
1368 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1369 struct btrfs_dir_log_item);
1370 found_end = btrfs_dir_log_end(path->nodes[0], item);
1371 *start_ret = key.offset;
1372 *end_ret = found_end;
1375 btrfs_release_path(root, path);
1380 * this looks for a given directory item in the log. If the directory
1381 * item is not in the log, the item is removed and the inode it points
1384 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1385 struct btrfs_root *root,
1386 struct btrfs_root *log,
1387 struct btrfs_path *path,
1388 struct btrfs_path *log_path,
1390 struct btrfs_key *dir_key)
1393 struct extent_buffer *eb;
1396 struct btrfs_dir_item *di;
1397 struct btrfs_dir_item *log_di;
1400 unsigned long ptr_end;
1402 struct inode *inode;
1403 struct btrfs_key location;
1406 eb = path->nodes[0];
1407 slot = path->slots[0];
1408 item_size = btrfs_item_size_nr(eb, slot);
1409 ptr = btrfs_item_ptr_offset(eb, slot);
1410 ptr_end = ptr + item_size;
1411 while(ptr < ptr_end) {
1412 di = (struct btrfs_dir_item *)ptr;
1413 name_len = btrfs_dir_name_len(eb, di);
1414 name = kmalloc(name_len, GFP_NOFS);
1419 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1422 if (dir_key->type == BTRFS_DIR_ITEM_KEY) {
1423 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1426 } else if (dir_key->type == BTRFS_DIR_INDEX_KEY) {
1427 log_di = btrfs_lookup_dir_index_item(trans, log,
1433 if (!log_di || IS_ERR(log_di)) {
1434 btrfs_dir_item_key_to_cpu(eb, di, &location);
1435 btrfs_release_path(root, path);
1436 btrfs_release_path(log, log_path);
1437 inode = read_one_inode(root, location.objectid);
1440 ret = link_to_fixup_dir(trans, root,
1441 path, location.objectid);
1443 btrfs_inc_nlink(inode);
1444 ret = btrfs_unlink_inode(trans, root, dir, inode,
1450 /* there might still be more names under this key
1451 * check and repeat if required
1453 ret = btrfs_search_slot(NULL, root, dir_key, path,
1460 btrfs_release_path(log, log_path);
1463 ptr = (unsigned long)(di + 1);
1468 btrfs_release_path(root, path);
1469 btrfs_release_path(log, log_path);
1474 * deletion replay happens before we copy any new directory items
1475 * out of the log or out of backreferences from inodes. It
1476 * scans the log to find ranges of keys that log is authoritative for,
1477 * and then scans the directory to find items in those ranges that are
1478 * not present in the log.
1480 * Anything we don't find in the log is unlinked and removed from the
1483 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1484 struct btrfs_root *root,
1485 struct btrfs_root *log,
1486 struct btrfs_path *path,
1491 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1493 struct btrfs_key dir_key;
1494 struct btrfs_key found_key;
1495 struct btrfs_path *log_path;
1498 dir_key.objectid = dirid;
1499 dir_key.type = BTRFS_DIR_ITEM_KEY;
1500 log_path = btrfs_alloc_path();
1504 dir = read_one_inode(root, dirid);
1505 /* it isn't an error if the inode isn't there, that can happen
1506 * because we replay the deletes before we copy in the inode item
1510 btrfs_free_path(log_path);
1517 ret = find_dir_range(log, path, dirid, key_type,
1518 &range_start, &range_end);
1522 dir_key.offset = range_start;
1525 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1530 nritems = btrfs_header_nritems(path->nodes[0]);
1531 if (path->slots[0] >= nritems) {
1532 ret = btrfs_next_leaf(root, path);
1536 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1538 if (found_key.objectid != dirid ||
1539 found_key.type != dir_key.type)
1542 if (found_key.offset > range_end)
1545 ret = check_item_in_log(trans, root, log, path,
1546 log_path, dir, &found_key);
1548 if (found_key.offset == (u64)-1)
1550 dir_key.offset = found_key.offset + 1;
1552 btrfs_release_path(root, path);
1553 if (range_end == (u64)-1)
1555 range_start = range_end + 1;
1560 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1561 key_type = BTRFS_DIR_LOG_INDEX_KEY;
1562 dir_key.type = BTRFS_DIR_INDEX_KEY;
1563 btrfs_release_path(root, path);
1567 btrfs_release_path(root, path);
1568 btrfs_free_path(log_path);
1574 * the process_func used to replay items from the log tree. This
1575 * gets called in two different stages. The first stage just looks
1576 * for inodes and makes sure they are all copied into the subvolume.
1578 * The second stage copies all the other item types from the log into
1579 * the subvolume. The two stage approach is slower, but gets rid of
1580 * lots of complexity around inodes referencing other inodes that exist
1581 * only in the log (references come from either directory items or inode
1584 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1585 struct walk_control *wc, u64 gen)
1588 struct btrfs_path *path;
1589 struct btrfs_root *root = wc->replay_dest;
1590 struct btrfs_key key;
1596 btrfs_read_buffer(eb, gen);
1598 level = btrfs_header_level(eb);
1603 path = btrfs_alloc_path();
1606 nritems = btrfs_header_nritems(eb);
1607 for (i = 0; i < nritems; i++) {
1608 btrfs_item_key_to_cpu(eb, &key, i);
1609 item_size = btrfs_item_size_nr(eb, i);
1611 /* inode keys are done during the first stage */
1612 if (key.type == BTRFS_INODE_ITEM_KEY &&
1613 wc->stage == LOG_WALK_REPLAY_INODES) {
1614 struct inode *inode;
1615 struct btrfs_inode_item *inode_item;
1618 inode_item = btrfs_item_ptr(eb, i,
1619 struct btrfs_inode_item);
1620 mode = btrfs_inode_mode(eb, inode_item);
1621 if (S_ISDIR(mode)) {
1622 ret = replay_dir_deletes(wc->trans,
1623 root, log, path, key.objectid);
1626 ret = overwrite_item(wc->trans, root, path,
1630 /* for regular files, truncate away
1631 * extents past the new EOF
1633 if (S_ISREG(mode)) {
1634 inode = read_one_inode(root,
1638 ret = btrfs_truncate_inode_items(wc->trans,
1639 root, inode, inode->i_size,
1640 BTRFS_EXTENT_DATA_KEY);
1644 ret = link_to_fixup_dir(wc->trans, root,
1645 path, key.objectid);
1648 if (wc->stage < LOG_WALK_REPLAY_ALL)
1651 /* these keys are simply copied */
1652 if (key.type == BTRFS_XATTR_ITEM_KEY) {
1653 ret = overwrite_item(wc->trans, root, path,
1656 } else if (key.type == BTRFS_INODE_REF_KEY) {
1657 ret = add_inode_ref(wc->trans, root, log, path,
1659 BUG_ON(ret && ret != -ENOENT);
1660 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1661 ret = replay_one_extent(wc->trans, root, path,
1664 } else if (key.type == BTRFS_EXTENT_CSUM_KEY) {
1665 ret = replay_one_csum(wc->trans, root, path,
1668 } else if (key.type == BTRFS_DIR_ITEM_KEY ||
1669 key.type == BTRFS_DIR_INDEX_KEY) {
1670 ret = replay_one_dir_item(wc->trans, root, path,
1675 btrfs_free_path(path);
1679 static int noinline walk_down_log_tree(struct btrfs_trans_handle *trans,
1680 struct btrfs_root *root,
1681 struct btrfs_path *path, int *level,
1682 struct walk_control *wc)
1688 struct extent_buffer *next;
1689 struct extent_buffer *cur;
1690 struct extent_buffer *parent;
1694 WARN_ON(*level < 0);
1695 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1698 WARN_ON(*level < 0);
1699 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1700 cur = path->nodes[*level];
1702 if (btrfs_header_level(cur) != *level)
1705 if (path->slots[*level] >=
1706 btrfs_header_nritems(cur))
1709 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1710 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1711 blocksize = btrfs_level_size(root, *level - 1);
1713 parent = path->nodes[*level];
1714 root_owner = btrfs_header_owner(parent);
1715 root_gen = btrfs_header_generation(parent);
1717 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1719 wc->process_func(root, next, wc, ptr_gen);
1722 path->slots[*level]++;
1724 btrfs_read_buffer(next, ptr_gen);
1726 btrfs_tree_lock(next);
1727 clean_tree_block(trans, root, next);
1728 btrfs_wait_tree_block_writeback(next);
1729 btrfs_tree_unlock(next);
1731 ret = btrfs_drop_leaf_ref(trans, root, next);
1734 WARN_ON(root_owner !=
1735 BTRFS_TREE_LOG_OBJECTID);
1736 ret = btrfs_free_reserved_extent(root,
1740 free_extent_buffer(next);
1743 btrfs_read_buffer(next, ptr_gen);
1745 WARN_ON(*level <= 0);
1746 if (path->nodes[*level-1])
1747 free_extent_buffer(path->nodes[*level-1]);
1748 path->nodes[*level-1] = next;
1749 *level = btrfs_header_level(next);
1750 path->slots[*level] = 0;
1753 WARN_ON(*level < 0);
1754 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1756 if (path->nodes[*level] == root->node) {
1757 parent = path->nodes[*level];
1759 parent = path->nodes[*level + 1];
1761 bytenr = path->nodes[*level]->start;
1763 blocksize = btrfs_level_size(root, *level);
1764 root_owner = btrfs_header_owner(parent);
1765 root_gen = btrfs_header_generation(parent);
1767 wc->process_func(root, path->nodes[*level], wc,
1768 btrfs_header_generation(path->nodes[*level]));
1771 next = path->nodes[*level];
1772 btrfs_tree_lock(next);
1773 clean_tree_block(trans, root, next);
1774 btrfs_wait_tree_block_writeback(next);
1775 btrfs_tree_unlock(next);
1778 ret = btrfs_drop_leaf_ref(trans, root, next);
1781 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1782 ret = btrfs_free_reserved_extent(root, bytenr, blocksize);
1785 free_extent_buffer(path->nodes[*level]);
1786 path->nodes[*level] = NULL;
1793 static int noinline walk_up_log_tree(struct btrfs_trans_handle *trans,
1794 struct btrfs_root *root,
1795 struct btrfs_path *path, int *level,
1796 struct walk_control *wc)
1804 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1805 slot = path->slots[i];
1806 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1807 struct extent_buffer *node;
1808 node = path->nodes[i];
1811 WARN_ON(*level == 0);
1814 struct extent_buffer *parent;
1815 if (path->nodes[*level] == root->node)
1816 parent = path->nodes[*level];
1818 parent = path->nodes[*level + 1];
1820 root_owner = btrfs_header_owner(parent);
1821 root_gen = btrfs_header_generation(parent);
1822 wc->process_func(root, path->nodes[*level], wc,
1823 btrfs_header_generation(path->nodes[*level]));
1825 struct extent_buffer *next;
1827 next = path->nodes[*level];
1829 btrfs_tree_lock(next);
1830 clean_tree_block(trans, root, next);
1831 btrfs_wait_tree_block_writeback(next);
1832 btrfs_tree_unlock(next);
1835 ret = btrfs_drop_leaf_ref(trans, root,
1840 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1841 ret = btrfs_free_reserved_extent(root,
1842 path->nodes[*level]->start,
1843 path->nodes[*level]->len);
1846 free_extent_buffer(path->nodes[*level]);
1847 path->nodes[*level] = NULL;
1855 * drop the reference count on the tree rooted at 'snap'. This traverses
1856 * the tree freeing any blocks that have a ref count of zero after being
1859 static int walk_log_tree(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *log, struct walk_control *wc)
1865 struct btrfs_path *path;
1869 path = btrfs_alloc_path();
1872 level = btrfs_header_level(log->node);
1874 path->nodes[level] = log->node;
1875 extent_buffer_get(log->node);
1876 path->slots[level] = 0;
1879 wret = walk_down_log_tree(trans, log, path, &level, wc);
1885 wret = walk_up_log_tree(trans, log, path, &level, wc);
1892 /* was the root node processed? if not, catch it here */
1893 if (path->nodes[orig_level]) {
1894 wc->process_func(log, path->nodes[orig_level], wc,
1895 btrfs_header_generation(path->nodes[orig_level]));
1897 struct extent_buffer *next;
1899 next = path->nodes[orig_level];
1901 btrfs_tree_lock(next);
1902 clean_tree_block(trans, log, next);
1903 btrfs_wait_tree_block_writeback(next);
1904 btrfs_tree_unlock(next);
1906 if (orig_level == 0) {
1907 ret = btrfs_drop_leaf_ref(trans, log,
1911 WARN_ON(log->root_key.objectid !=
1912 BTRFS_TREE_LOG_OBJECTID);
1913 ret = btrfs_free_reserved_extent(log, next->start,
1919 for (i = 0; i <= orig_level; i++) {
1920 if (path->nodes[i]) {
1921 free_extent_buffer(path->nodes[i]);
1922 path->nodes[i] = NULL;
1925 btrfs_free_path(path);
1927 free_extent_buffer(log->node);
1931 static int wait_log_commit(struct btrfs_root *log)
1934 u64 transid = log->fs_info->tree_log_transid;
1937 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1938 TASK_UNINTERRUPTIBLE);
1939 mutex_unlock(&log->fs_info->tree_log_mutex);
1940 if (atomic_read(&log->fs_info->tree_log_commit))
1942 finish_wait(&log->fs_info->tree_log_wait, &wait);
1943 mutex_lock(&log->fs_info->tree_log_mutex);
1944 } while(transid == log->fs_info->tree_log_transid &&
1945 atomic_read(&log->fs_info->tree_log_commit));
1950 * btrfs_sync_log does sends a given tree log down to the disk and
1951 * updates the super blocks to record it. When this call is done,
1952 * you know that any inodes previously logged are safely on disk
1954 int btrfs_sync_log(struct btrfs_trans_handle *trans,
1955 struct btrfs_root *root)
1958 unsigned long batch;
1959 struct btrfs_root *log = root->log_root;
1961 mutex_lock(&log->fs_info->tree_log_mutex);
1962 if (atomic_read(&log->fs_info->tree_log_commit)) {
1963 wait_log_commit(log);
1966 atomic_set(&log->fs_info->tree_log_commit, 1);
1969 batch = log->fs_info->tree_log_batch;
1970 mutex_unlock(&log->fs_info->tree_log_mutex);
1971 schedule_timeout_uninterruptible(1);
1972 mutex_lock(&log->fs_info->tree_log_mutex);
1974 while(atomic_read(&log->fs_info->tree_log_writers)) {
1976 prepare_to_wait(&log->fs_info->tree_log_wait, &wait,
1977 TASK_UNINTERRUPTIBLE);
1978 mutex_unlock(&log->fs_info->tree_log_mutex);
1979 if (atomic_read(&log->fs_info->tree_log_writers))
1981 mutex_lock(&log->fs_info->tree_log_mutex);
1982 finish_wait(&log->fs_info->tree_log_wait, &wait);
1984 if (batch == log->fs_info->tree_log_batch)
1988 ret = btrfs_write_and_wait_marked_extents(log, &log->dirty_log_pages);
1990 ret = btrfs_write_and_wait_marked_extents(root->fs_info->log_root_tree,
1991 &root->fs_info->log_root_tree->dirty_log_pages);
1994 btrfs_set_super_log_root(&root->fs_info->super_for_commit,
1995 log->fs_info->log_root_tree->node->start);
1996 btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
1997 btrfs_header_level(log->fs_info->log_root_tree->node));
1999 write_ctree_super(trans, log->fs_info->tree_root, 2);
2000 log->fs_info->tree_log_transid++;
2001 log->fs_info->tree_log_batch = 0;
2002 atomic_set(&log->fs_info->tree_log_commit, 0);
2004 if (waitqueue_active(&log->fs_info->tree_log_wait))
2005 wake_up(&log->fs_info->tree_log_wait);
2007 mutex_unlock(&log->fs_info->tree_log_mutex);
2011 /* * free all the extents used by the tree log. This should be called
2012 * at commit time of the full transaction
2014 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2017 struct btrfs_root *log;
2021 struct walk_control wc = {
2023 .process_func = process_one_buffer
2026 if (!root->log_root)
2029 log = root->log_root;
2030 ret = walk_log_tree(trans, log, &wc);
2034 ret = find_first_extent_bit(&log->dirty_log_pages,
2035 0, &start, &end, EXTENT_DIRTY);
2039 clear_extent_dirty(&log->dirty_log_pages,
2040 start, end, GFP_NOFS);
2043 log = root->log_root;
2044 ret = btrfs_del_root(trans, root->fs_info->log_root_tree,
2047 root->log_root = NULL;
2048 kfree(root->log_root);
2053 * helper function to update the item for a given subvolumes log root
2054 * in the tree of log roots
2056 static int update_log_root(struct btrfs_trans_handle *trans,
2057 struct btrfs_root *log)
2059 u64 bytenr = btrfs_root_bytenr(&log->root_item);
2062 if (log->node->start == bytenr)
2065 btrfs_set_root_bytenr(&log->root_item, log->node->start);
2066 btrfs_set_root_generation(&log->root_item, trans->transid);
2067 btrfs_set_root_level(&log->root_item, btrfs_header_level(log->node));
2068 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2069 &log->root_key, &log->root_item);
2075 * If both a file and directory are logged, and unlinks or renames are
2076 * mixed in, we have a few interesting corners:
2078 * create file X in dir Y
2079 * link file X to X.link in dir Y
2081 * unlink file X but leave X.link
2084 * After a crash we would expect only X.link to exist. But file X
2085 * didn't get fsync'd again so the log has back refs for X and X.link.
2087 * We solve this by removing directory entries and inode backrefs from the
2088 * log when a file that was logged in the current transaction is
2089 * unlinked. Any later fsync will include the updated log entries, and
2090 * we'll be able to reconstruct the proper directory items from backrefs.
2092 * This optimizations allows us to avoid relogging the entire inode
2093 * or the entire directory.
2095 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2096 struct btrfs_root *root,
2097 const char *name, int name_len,
2098 struct inode *dir, u64 index)
2100 struct btrfs_root *log;
2101 struct btrfs_dir_item *di;
2102 struct btrfs_path *path;
2106 if (BTRFS_I(dir)->logged_trans < trans->transid)
2109 ret = join_running_log_trans(root);
2113 mutex_lock(&BTRFS_I(dir)->log_mutex);
2115 log = root->log_root;
2116 path = btrfs_alloc_path();
2117 di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2118 name, name_len, -1);
2119 if (di && !IS_ERR(di)) {
2120 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2121 bytes_del += name_len;
2124 btrfs_release_path(log, path);
2125 di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2126 index, 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;
2133 /* update the directory size in the log to reflect the names
2137 struct btrfs_key key;
2139 key.objectid = dir->i_ino;
2141 key.type = BTRFS_INODE_ITEM_KEY;
2142 btrfs_release_path(log, path);
2144 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2146 struct btrfs_inode_item *item;
2149 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2150 struct btrfs_inode_item);
2151 i_size = btrfs_inode_size(path->nodes[0], item);
2152 if (i_size > bytes_del)
2153 i_size -= bytes_del;
2156 btrfs_set_inode_size(path->nodes[0], item, i_size);
2157 btrfs_mark_buffer_dirty(path->nodes[0]);
2160 btrfs_release_path(log, path);
2163 btrfs_free_path(path);
2164 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2165 end_log_trans(root);
2170 /* see comments for btrfs_del_dir_entries_in_log */
2171 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root,
2173 const char *name, int name_len,
2174 struct inode *inode, u64 dirid)
2176 struct btrfs_root *log;
2180 if (BTRFS_I(inode)->logged_trans < trans->transid)
2183 ret = join_running_log_trans(root);
2186 log = root->log_root;
2187 mutex_lock(&BTRFS_I(inode)->log_mutex);
2189 ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2191 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2192 end_log_trans(root);
2198 * creates a range item in the log for 'dirid'. first_offset and
2199 * last_offset tell us which parts of the key space the log should
2200 * be considered authoritative for.
2202 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2203 struct btrfs_root *log,
2204 struct btrfs_path *path,
2205 int key_type, u64 dirid,
2206 u64 first_offset, u64 last_offset)
2209 struct btrfs_key key;
2210 struct btrfs_dir_log_item *item;
2212 key.objectid = dirid;
2213 key.offset = first_offset;
2214 if (key_type == BTRFS_DIR_ITEM_KEY)
2215 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2217 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2218 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2221 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2222 struct btrfs_dir_log_item);
2223 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2224 btrfs_mark_buffer_dirty(path->nodes[0]);
2225 btrfs_release_path(log, path);
2230 * log all the items included in the current transaction for a given
2231 * directory. This also creates the range items in the log tree required
2232 * to replay anything deleted before the fsync
2234 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root, struct inode *inode,
2236 struct btrfs_path *path,
2237 struct btrfs_path *dst_path, int key_type,
2238 u64 min_offset, u64 *last_offset_ret)
2240 struct btrfs_key min_key;
2241 struct btrfs_key max_key;
2242 struct btrfs_root *log = root->log_root;
2243 struct extent_buffer *src;
2247 u64 first_offset = min_offset;
2248 u64 last_offset = (u64)-1;
2250 log = root->log_root;
2251 max_key.objectid = inode->i_ino;
2252 max_key.offset = (u64)-1;
2253 max_key.type = key_type;
2255 min_key.objectid = inode->i_ino;
2256 min_key.type = key_type;
2257 min_key.offset = min_offset;
2259 path->keep_locks = 1;
2261 ret = btrfs_search_forward(root, &min_key, &max_key,
2262 path, 0, trans->transid);
2265 * we didn't find anything from this transaction, see if there
2266 * is anything at all
2268 if (ret != 0 || min_key.objectid != inode->i_ino ||
2269 min_key.type != key_type) {
2270 min_key.objectid = inode->i_ino;
2271 min_key.type = key_type;
2272 min_key.offset = (u64)-1;
2273 btrfs_release_path(root, path);
2274 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2276 btrfs_release_path(root, path);
2279 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2281 /* if ret == 0 there are items for this type,
2282 * create a range to tell us the last key of this type.
2283 * otherwise, there are no items in this directory after
2284 * *min_offset, and we create a range to indicate that.
2287 struct btrfs_key tmp;
2288 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2290 if (key_type == tmp.type) {
2291 first_offset = max(min_offset, tmp.offset) + 1;
2297 /* go backward to find any previous key */
2298 ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2300 struct btrfs_key tmp;
2301 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2302 if (key_type == tmp.type) {
2303 first_offset = tmp.offset;
2304 ret = overwrite_item(trans, log, dst_path,
2305 path->nodes[0], path->slots[0],
2309 btrfs_release_path(root, path);
2311 /* find the first key from this transaction again */
2312 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2319 * we have a block from this transaction, log every item in it
2320 * from our directory
2323 struct btrfs_key tmp;
2324 src = path->nodes[0];
2325 nritems = btrfs_header_nritems(src);
2326 for (i = path->slots[0]; i < nritems; i++) {
2327 btrfs_item_key_to_cpu(src, &min_key, i);
2329 if (min_key.objectid != inode->i_ino ||
2330 min_key.type != key_type)
2332 ret = overwrite_item(trans, log, dst_path, src, i,
2336 path->slots[0] = nritems;
2339 * look ahead to the next item and see if it is also
2340 * from this directory and from this transaction
2342 ret = btrfs_next_leaf(root, path);
2344 last_offset = (u64)-1;
2347 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2348 if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2349 last_offset = (u64)-1;
2352 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2353 ret = overwrite_item(trans, log, dst_path,
2354 path->nodes[0], path->slots[0],
2358 last_offset = tmp.offset;
2363 *last_offset_ret = last_offset;
2364 btrfs_release_path(root, path);
2365 btrfs_release_path(log, dst_path);
2367 /* insert the log range keys to indicate where the log is valid */
2368 ret = insert_dir_log_key(trans, log, path, key_type, inode->i_ino,
2369 first_offset, last_offset);
2375 * logging directories is very similar to logging inodes, We find all the items
2376 * from the current transaction and write them to the log.
2378 * The recovery code scans the directory in the subvolume, and if it finds a
2379 * key in the range logged that is not present in the log tree, then it means
2380 * that dir entry was unlinked during the transaction.
2382 * In order for that scan to work, we must include one key smaller than
2383 * the smallest logged by this transaction and one key larger than the largest
2384 * key logged by this transaction.
2386 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2387 struct btrfs_root *root, struct inode *inode,
2388 struct btrfs_path *path,
2389 struct btrfs_path *dst_path)
2394 int key_type = BTRFS_DIR_ITEM_KEY;
2400 ret = log_dir_items(trans, root, inode, path,
2401 dst_path, key_type, min_key,
2404 if (max_key == (u64)-1)
2406 min_key = max_key + 1;
2409 if (key_type == BTRFS_DIR_ITEM_KEY) {
2410 key_type = BTRFS_DIR_INDEX_KEY;
2417 * a helper function to drop items from the log before we relog an
2418 * inode. max_key_type indicates the highest item type to remove.
2419 * This cannot be run for file data extents because it does not
2420 * free the extents they point to.
2422 static int drop_objectid_items(struct btrfs_trans_handle *trans,
2423 struct btrfs_root *log,
2424 struct btrfs_path *path,
2425 u64 objectid, int max_key_type)
2428 struct btrfs_key key;
2429 struct btrfs_key found_key;
2431 key.objectid = objectid;
2432 key.type = max_key_type;
2433 key.offset = (u64)-1;
2436 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2441 if (path->slots[0] == 0)
2445 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2448 if (found_key.objectid != objectid)
2451 ret = btrfs_del_item(trans, log, path);
2453 btrfs_release_path(log, path);
2455 btrfs_release_path(log, path);
2459 static noinline int copy_extent_csums(struct btrfs_trans_handle *trans,
2460 struct list_head *list,
2461 struct btrfs_root *root,
2462 u64 disk_bytenr, u64 len)
2464 struct btrfs_ordered_sum *sums;
2465 struct btrfs_sector_sum *sector_sum;
2467 struct btrfs_path *path;
2468 struct btrfs_csum_item *item = NULL;
2469 u64 end = disk_bytenr + len;
2470 u64 item_start_offset = 0;
2471 u64 item_last_offset = 0;
2474 u16 csum_size = btrfs_super_csum_size(&root->fs_info->super_copy);
2476 sums = kzalloc(btrfs_ordered_sum_size(root, len), GFP_NOFS);
2478 sector_sum = sums->sums;
2479 sums->bytenr = disk_bytenr;
2481 list_add_tail(&sums->list, list);
2483 path = btrfs_alloc_path();
2484 while(disk_bytenr < end) {
2485 if (!item || disk_bytenr < item_start_offset ||
2486 disk_bytenr >= item_last_offset) {
2487 struct btrfs_key found_key;
2491 btrfs_release_path(root, path);
2492 item = btrfs_lookup_csum(NULL, root, path,
2495 ret = PTR_ERR(item);
2496 if (ret == -ENOENT || ret == -EFBIG)
2499 printk("log no csum found for byte %llu\n",
2500 (unsigned long long)disk_bytenr);
2502 btrfs_release_path(root, path);
2505 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2508 item_start_offset = found_key.offset;
2509 item_size = btrfs_item_size_nr(path->nodes[0],
2511 item_last_offset = item_start_offset +
2512 (item_size / csum_size) *
2514 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2515 struct btrfs_csum_item);
2518 * this byte range must be able to fit inside
2519 * a single leaf so it will also fit inside a u32
2521 diff = disk_bytenr - item_start_offset;
2522 diff = diff / root->sectorsize;
2523 diff = diff * csum_size;
2525 read_extent_buffer(path->nodes[0], &sum,
2526 ((unsigned long)item) + diff,
2529 sector_sum->bytenr = disk_bytenr;
2530 sector_sum->sum = sum;
2531 disk_bytenr += root->sectorsize;
2534 btrfs_free_path(path);
2538 static noinline int copy_items(struct btrfs_trans_handle *trans,
2539 struct btrfs_root *log,
2540 struct btrfs_path *dst_path,
2541 struct extent_buffer *src,
2542 int start_slot, int nr, int inode_only)
2544 unsigned long src_offset;
2545 unsigned long dst_offset;
2546 struct btrfs_file_extent_item *extent;
2547 struct btrfs_inode_item *inode_item;
2549 struct btrfs_key *ins_keys;
2553 struct list_head ordered_sums;
2555 INIT_LIST_HEAD(&ordered_sums);
2557 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2558 nr * sizeof(u32), GFP_NOFS);
2559 ins_sizes = (u32 *)ins_data;
2560 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2562 for (i = 0; i < nr; i++) {
2563 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2564 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2566 ret = btrfs_insert_empty_items(trans, log, dst_path,
2567 ins_keys, ins_sizes, nr);
2570 for (i = 0; i < nr; i++) {
2571 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2572 dst_path->slots[0]);
2574 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2576 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2577 src_offset, ins_sizes[i]);
2579 if (inode_only == LOG_INODE_EXISTS &&
2580 ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2581 inode_item = btrfs_item_ptr(dst_path->nodes[0],
2583 struct btrfs_inode_item);
2584 btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2586 /* set the generation to zero so the recover code
2587 * can tell the difference between an logging
2588 * just to say 'this inode exists' and a logging
2589 * to say 'update this inode with these values'
2591 btrfs_set_inode_generation(dst_path->nodes[0],
2594 /* take a reference on file data extents so that truncates
2595 * or deletes of this inode don't have to relog the inode
2598 if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2600 extent = btrfs_item_ptr(src, start_slot + i,
2601 struct btrfs_file_extent_item);
2603 found_type = btrfs_file_extent_type(src, extent);
2604 if (found_type == BTRFS_FILE_EXTENT_REG ||
2605 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2606 u64 ds = btrfs_file_extent_disk_bytenr(src,
2608 u64 dl = btrfs_file_extent_disk_num_bytes(src,
2610 u64 cs = btrfs_file_extent_offset(src, extent);
2611 u64 cl = btrfs_file_extent_num_bytes(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);