2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/blkdev.h>
22 #include <linux/list_sort.h>
26 #include "print-tree.h"
30 /* magic values for the inode_only field in btrfs_log_inode:
32 * LOG_INODE_ALL means to log everything
33 * LOG_INODE_EXISTS means to log just enough to recreate the inode
36 #define LOG_INODE_ALL 0
37 #define LOG_INODE_EXISTS 1
40 * directory trouble cases
42 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
43 * log, we must force a full commit before doing an fsync of the directory
44 * where the unlink was done.
45 * ---> record transid of last unlink/rename per directory
49 * rename foo/some_dir foo2/some_dir
51 * fsync foo/some_dir/some_file
53 * The fsync above will unlink the original some_dir without recording
54 * it in its new location (foo2). After a crash, some_dir will be gone
55 * unless the fsync of some_file forces a full commit
57 * 2) we must log any new names for any file or dir that is in the fsync
58 * log. ---> check inode while renaming/linking.
60 * 2a) we must log any new names for any file or dir during rename
61 * when the directory they are being removed from was logged.
62 * ---> check inode and old parent dir during rename
64 * 2a is actually the more important variant. With the extra logging
65 * a crash might unlink the old name without recreating the new one
67 * 3) after a crash, we must go through any directories with a link count
68 * of zero and redo the rm -rf
75 * The directory f1 was fully removed from the FS, but fsync was never
76 * called on f1, only its parent dir. After a crash the rm -rf must
77 * be replayed. This must be able to recurse down the entire
78 * directory tree. The inode link count fixup code takes care of the
83 * stages for the tree walking. The first
84 * stage (0) is to only pin down the blocks we find
85 * the second stage (1) is to make sure that all the inodes
86 * we find in the log are created in the subvolume.
88 * The last stage is to deal with directories and links and extents
89 * and all the other fun semantics
91 #define LOG_WALK_PIN_ONLY 0
92 #define LOG_WALK_REPLAY_INODES 1
93 #define LOG_WALK_REPLAY_DIR_INDEX 2
94 #define LOG_WALK_REPLAY_ALL 3
96 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
97 struct btrfs_root *root, struct inode *inode,
101 struct btrfs_log_ctx *ctx);
102 static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
103 struct btrfs_root *root,
104 struct btrfs_path *path, u64 objectid);
105 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
106 struct btrfs_root *root,
107 struct btrfs_root *log,
108 struct btrfs_path *path,
109 u64 dirid, int del_all);
112 * tree logging is a special write ahead log used to make sure that
113 * fsyncs and O_SYNCs can happen without doing full tree commits.
115 * Full tree commits are expensive because they require commonly
116 * modified blocks to be recowed, creating many dirty pages in the
117 * extent tree an 4x-6x higher write load than ext3.
119 * Instead of doing a tree commit on every fsync, we use the
120 * key ranges and transaction ids to find items for a given file or directory
121 * that have changed in this transaction. Those items are copied into
122 * a special tree (one per subvolume root), that tree is written to disk
123 * and then the fsync is considered complete.
125 * After a crash, items are copied out of the log-tree back into the
126 * subvolume tree. Any file data extents found are recorded in the extent
127 * allocation tree, and the log-tree freed.
129 * The log tree is read three times, once to pin down all the extents it is
130 * using in ram and once, once to create all the inodes logged in the tree
131 * and once to do all the other items.
135 * start a sub transaction and setup the log tree
136 * this increments the log tree writer count to make the people
137 * syncing the tree wait for us to finish
139 static int start_log_trans(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct btrfs_log_ctx *ctx)
146 mutex_lock(&root->log_mutex);
147 if (root->log_root) {
148 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
152 if (!root->log_start_pid) {
153 root->log_start_pid = current->pid;
154 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
155 } else if (root->log_start_pid != current->pid) {
156 set_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
159 atomic_inc(&root->log_batch);
160 atomic_inc(&root->log_writers);
162 index = root->log_transid % 2;
163 list_add_tail(&ctx->list, &root->log_ctxs[index]);
164 ctx->log_transid = root->log_transid;
166 mutex_unlock(&root->log_mutex);
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);
174 mutex_unlock(&root->fs_info->tree_log_mutex);
178 if (!root->log_root) {
179 ret = btrfs_add_log_tree(trans, root);
183 clear_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state);
184 root->log_start_pid = current->pid;
185 atomic_inc(&root->log_batch);
186 atomic_inc(&root->log_writers);
188 index = root->log_transid % 2;
189 list_add_tail(&ctx->list, &root->log_ctxs[index]);
190 ctx->log_transid = root->log_transid;
193 mutex_unlock(&root->log_mutex);
198 * returns 0 if there was a log transaction running and we were able
199 * to join, or returns -ENOENT if there were not transactions
202 static int join_running_log_trans(struct btrfs_root *root)
210 mutex_lock(&root->log_mutex);
211 if (root->log_root) {
213 atomic_inc(&root->log_writers);
215 mutex_unlock(&root->log_mutex);
220 * This either makes the current running log transaction wait
221 * until you call btrfs_end_log_trans() or it makes any future
222 * log transactions wait until you call btrfs_end_log_trans()
224 int btrfs_pin_log_trans(struct btrfs_root *root)
228 mutex_lock(&root->log_mutex);
229 atomic_inc(&root->log_writers);
230 mutex_unlock(&root->log_mutex);
235 * indicate we're done making changes to the log tree
236 * and wake up anyone waiting to do a sync
238 void btrfs_end_log_trans(struct btrfs_root *root)
240 if (atomic_dec_and_test(&root->log_writers)) {
242 if (waitqueue_active(&root->log_writer_wait))
243 wake_up(&root->log_writer_wait);
249 * the walk control struct is used to pass state down the chain when
250 * processing the log tree. The stage field tells us which part
251 * of the log tree processing we are currently doing. The others
252 * are state fields used for that specific part
254 struct walk_control {
255 /* should we free the extent on disk when done? This is used
256 * at transaction commit time while freeing a log tree
260 /* should we write out the extent buffer? This is used
261 * while flushing the log tree to disk during a sync
265 /* should we wait for the extent buffer io to finish? Also used
266 * while flushing the log tree to disk for a sync
270 /* pin only walk, we record which extents on disk belong to the
275 /* what stage of the replay code we're currently in */
278 /* the root we are currently replaying */
279 struct btrfs_root *replay_dest;
281 /* the trans handle for the current replay */
282 struct btrfs_trans_handle *trans;
284 /* the function that gets used to process blocks we find in the
285 * tree. Note the extent_buffer might not be up to date when it is
286 * passed in, and it must be checked or read if you need the data
289 int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
290 struct walk_control *wc, u64 gen);
294 * process_func used to pin down extents, write them or wait on them
296 static int process_one_buffer(struct btrfs_root *log,
297 struct extent_buffer *eb,
298 struct walk_control *wc, u64 gen)
303 * If this fs is mixed then we need to be able to process the leaves to
304 * pin down any logged extents, so we have to read the block.
306 if (btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) {
307 ret = btrfs_read_buffer(eb, gen);
313 ret = btrfs_pin_extent_for_log_replay(log->fs_info->extent_root,
316 if (!ret && btrfs_buffer_uptodate(eb, gen, 0)) {
317 if (wc->pin && btrfs_header_level(eb) == 0)
318 ret = btrfs_exclude_logged_extents(log, eb);
320 btrfs_write_tree_block(eb);
322 btrfs_wait_tree_block_writeback(eb);
328 * Item overwrite used by replay and tree logging. eb, slot and key all refer
329 * to the src data we are copying out.
331 * root is the tree we are copying into, and path is a scratch
332 * path for use in this function (it should be released on entry and
333 * will be released on exit).
335 * If the key is already in the destination tree the existing item is
336 * overwritten. If the existing item isn't big enough, it is extended.
337 * If it is too large, it is truncated.
339 * If the key isn't in the destination yet, a new item is inserted.
341 static noinline int overwrite_item(struct btrfs_trans_handle *trans,
342 struct btrfs_root *root,
343 struct btrfs_path *path,
344 struct extent_buffer *eb, int slot,
345 struct btrfs_key *key)
349 u64 saved_i_size = 0;
350 int save_old_i_size = 0;
351 unsigned long src_ptr;
352 unsigned long dst_ptr;
353 int overwrite_root = 0;
354 bool inode_item = key->type == BTRFS_INODE_ITEM_KEY;
356 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
359 item_size = btrfs_item_size_nr(eb, slot);
360 src_ptr = btrfs_item_ptr_offset(eb, slot);
362 /* look for the key in the destination tree */
363 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
370 u32 dst_size = btrfs_item_size_nr(path->nodes[0],
372 if (dst_size != item_size)
375 if (item_size == 0) {
376 btrfs_release_path(path);
379 dst_copy = kmalloc(item_size, GFP_NOFS);
380 src_copy = kmalloc(item_size, GFP_NOFS);
381 if (!dst_copy || !src_copy) {
382 btrfs_release_path(path);
388 read_extent_buffer(eb, src_copy, src_ptr, item_size);
390 dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
391 read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
393 ret = memcmp(dst_copy, src_copy, item_size);
398 * they have the same contents, just return, this saves
399 * us from cowing blocks in the destination tree and doing
400 * extra writes that may not have been done by a previous
404 btrfs_release_path(path);
409 * We need to load the old nbytes into the inode so when we
410 * replay the extents we've logged we get the right nbytes.
413 struct btrfs_inode_item *item;
417 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
418 struct btrfs_inode_item);
419 nbytes = btrfs_inode_nbytes(path->nodes[0], item);
420 item = btrfs_item_ptr(eb, slot,
421 struct btrfs_inode_item);
422 btrfs_set_inode_nbytes(eb, item, nbytes);
425 * If this is a directory we need to reset the i_size to
426 * 0 so that we can set it up properly when replaying
427 * the rest of the items in this log.
429 mode = btrfs_inode_mode(eb, item);
431 btrfs_set_inode_size(eb, item, 0);
433 } else if (inode_item) {
434 struct btrfs_inode_item *item;
438 * New inode, set nbytes to 0 so that the nbytes comes out
439 * properly when we replay the extents.
441 item = btrfs_item_ptr(eb, slot, struct btrfs_inode_item);
442 btrfs_set_inode_nbytes(eb, item, 0);
445 * If this is a directory we need to reset the i_size to 0 so
446 * that we can set it up properly when replaying the rest of
447 * the items in this log.
449 mode = btrfs_inode_mode(eb, item);
451 btrfs_set_inode_size(eb, item, 0);
454 btrfs_release_path(path);
455 /* try to insert the key into the destination tree */
456 ret = btrfs_insert_empty_item(trans, root, path,
459 /* make sure any existing item is the correct size */
460 if (ret == -EEXIST) {
462 found_size = btrfs_item_size_nr(path->nodes[0],
464 if (found_size > item_size)
465 btrfs_truncate_item(root, path, item_size, 1);
466 else if (found_size < item_size)
467 btrfs_extend_item(root, path,
468 item_size - found_size);
472 dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
475 /* don't overwrite an existing inode if the generation number
476 * was logged as zero. This is done when the tree logging code
477 * is just logging an inode to make sure it exists after recovery.
479 * Also, don't overwrite i_size on directories during replay.
480 * log replay inserts and removes directory items based on the
481 * state of the tree found in the subvolume, and i_size is modified
484 if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
485 struct btrfs_inode_item *src_item;
486 struct btrfs_inode_item *dst_item;
488 src_item = (struct btrfs_inode_item *)src_ptr;
489 dst_item = (struct btrfs_inode_item *)dst_ptr;
491 if (btrfs_inode_generation(eb, src_item) == 0)
494 if (overwrite_root &&
495 S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
496 S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
498 saved_i_size = btrfs_inode_size(path->nodes[0],
503 copy_extent_buffer(path->nodes[0], eb, dst_ptr,
506 if (save_old_i_size) {
507 struct btrfs_inode_item *dst_item;
508 dst_item = (struct btrfs_inode_item *)dst_ptr;
509 btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
512 /* make sure the generation is filled in */
513 if (key->type == BTRFS_INODE_ITEM_KEY) {
514 struct btrfs_inode_item *dst_item;
515 dst_item = (struct btrfs_inode_item *)dst_ptr;
516 if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
517 btrfs_set_inode_generation(path->nodes[0], dst_item,
522 btrfs_mark_buffer_dirty(path->nodes[0]);
523 btrfs_release_path(path);
528 * simple helper to read an inode off the disk from a given root
529 * This can only be called for subvolume roots and not for the log
531 static noinline struct inode *read_one_inode(struct btrfs_root *root,
534 struct btrfs_key key;
537 key.objectid = objectid;
538 key.type = BTRFS_INODE_ITEM_KEY;
540 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
543 } else if (is_bad_inode(inode)) {
550 /* replays a single extent in 'eb' at 'slot' with 'key' into the
551 * subvolume 'root'. path is released on entry and should be released
554 * extents in the log tree have not been allocated out of the extent
555 * tree yet. So, this completes the allocation, taking a reference
556 * as required if the extent already exists or creating a new extent
557 * if it isn't in the extent allocation tree yet.
559 * The extent is inserted into the file, dropping any existing extents
560 * from the file that overlap the new one.
562 static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
563 struct btrfs_root *root,
564 struct btrfs_path *path,
565 struct extent_buffer *eb, int slot,
566 struct btrfs_key *key)
570 u64 start = key->offset;
572 struct btrfs_file_extent_item *item;
573 struct inode *inode = NULL;
577 item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
578 found_type = btrfs_file_extent_type(eb, item);
580 if (found_type == BTRFS_FILE_EXTENT_REG ||
581 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
582 nbytes = btrfs_file_extent_num_bytes(eb, item);
583 extent_end = start + nbytes;
586 * We don't add to the inodes nbytes if we are prealloc or a
589 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
591 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
592 size = btrfs_file_extent_inline_len(eb, slot, item);
593 nbytes = btrfs_file_extent_ram_bytes(eb, item);
594 extent_end = ALIGN(start + size, root->sectorsize);
600 inode = read_one_inode(root, key->objectid);
607 * first check to see if we already have this extent in the
608 * file. This must be done before the btrfs_drop_extents run
609 * so we don't try to drop this extent.
611 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
615 (found_type == BTRFS_FILE_EXTENT_REG ||
616 found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
617 struct btrfs_file_extent_item cmp1;
618 struct btrfs_file_extent_item cmp2;
619 struct btrfs_file_extent_item *existing;
620 struct extent_buffer *leaf;
622 leaf = path->nodes[0];
623 existing = btrfs_item_ptr(leaf, path->slots[0],
624 struct btrfs_file_extent_item);
626 read_extent_buffer(eb, &cmp1, (unsigned long)item,
628 read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
632 * we already have a pointer to this exact extent,
633 * we don't have to do anything
635 if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
636 btrfs_release_path(path);
640 btrfs_release_path(path);
642 /* drop any overlapping extents */
643 ret = btrfs_drop_extents(trans, root, inode, start, extent_end, 1);
647 if (found_type == BTRFS_FILE_EXTENT_REG ||
648 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
650 unsigned long dest_offset;
651 struct btrfs_key ins;
653 ret = btrfs_insert_empty_item(trans, root, path, key,
657 dest_offset = btrfs_item_ptr_offset(path->nodes[0],
659 copy_extent_buffer(path->nodes[0], eb, dest_offset,
660 (unsigned long)item, sizeof(*item));
662 ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
663 ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
664 ins.type = BTRFS_EXTENT_ITEM_KEY;
665 offset = key->offset - btrfs_file_extent_offset(eb, item);
667 if (ins.objectid > 0) {
670 LIST_HEAD(ordered_sums);
672 * is this extent already allocated in the extent
673 * allocation tree? If so, just add a reference
675 ret = btrfs_lookup_data_extent(root, ins.objectid,
678 ret = btrfs_inc_extent_ref(trans, root,
679 ins.objectid, ins.offset,
680 0, root->root_key.objectid,
681 key->objectid, offset, 0);
686 * insert the extent pointer in the extent
689 ret = btrfs_alloc_logged_file_extent(trans,
690 root, root->root_key.objectid,
691 key->objectid, offset, &ins);
695 btrfs_release_path(path);
697 if (btrfs_file_extent_compression(eb, item)) {
698 csum_start = ins.objectid;
699 csum_end = csum_start + ins.offset;
701 csum_start = ins.objectid +
702 btrfs_file_extent_offset(eb, item);
703 csum_end = csum_start +
704 btrfs_file_extent_num_bytes(eb, item);
707 ret = btrfs_lookup_csums_range(root->log_root,
708 csum_start, csum_end - 1,
712 while (!list_empty(&ordered_sums)) {
713 struct btrfs_ordered_sum *sums;
714 sums = list_entry(ordered_sums.next,
715 struct btrfs_ordered_sum,
718 ret = btrfs_csum_file_blocks(trans,
719 root->fs_info->csum_root,
721 list_del(&sums->list);
727 btrfs_release_path(path);
729 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
730 /* inline extents are easy, we just overwrite them */
731 ret = overwrite_item(trans, root, path, eb, slot, key);
736 inode_add_bytes(inode, nbytes);
737 ret = btrfs_update_inode(trans, root, inode);
745 * when cleaning up conflicts between the directory names in the
746 * subvolume, directory names in the log and directory names in the
747 * inode back references, we may have to unlink inodes from directories.
749 * This is a helper function to do the unlink of a specific directory
752 static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
753 struct btrfs_root *root,
754 struct btrfs_path *path,
756 struct btrfs_dir_item *di)
761 struct extent_buffer *leaf;
762 struct btrfs_key location;
765 leaf = path->nodes[0];
767 btrfs_dir_item_key_to_cpu(leaf, di, &location);
768 name_len = btrfs_dir_name_len(leaf, di);
769 name = kmalloc(name_len, GFP_NOFS);
773 read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
774 btrfs_release_path(path);
776 inode = read_one_inode(root, location.objectid);
782 ret = link_to_fixup_dir(trans, root, path, location.objectid);
786 ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
790 ret = btrfs_run_delayed_items(trans, root);
798 * helper function to see if a given name and sequence number found
799 * in an inode back reference are already in a directory and correctly
800 * point to this inode
802 static noinline int inode_in_dir(struct btrfs_root *root,
803 struct btrfs_path *path,
804 u64 dirid, u64 objectid, u64 index,
805 const char *name, int name_len)
807 struct btrfs_dir_item *di;
808 struct btrfs_key location;
811 di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
812 index, name, name_len, 0);
813 if (di && !IS_ERR(di)) {
814 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
815 if (location.objectid != objectid)
819 btrfs_release_path(path);
821 di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
822 if (di && !IS_ERR(di)) {
823 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
824 if (location.objectid != objectid)
830 btrfs_release_path(path);
835 * helper function to check a log tree for a named back reference in
836 * an inode. This is used to decide if a back reference that is
837 * found in the subvolume conflicts with what we find in the log.
839 * inode backreferences may have multiple refs in a single item,
840 * during replay we process one reference at a time, and we don't
841 * want to delete valid links to a file from the subvolume if that
842 * link is also in the log.
844 static noinline int backref_in_log(struct btrfs_root *log,
845 struct btrfs_key *key,
847 char *name, int namelen)
849 struct btrfs_path *path;
850 struct btrfs_inode_ref *ref;
852 unsigned long ptr_end;
853 unsigned long name_ptr;
859 path = btrfs_alloc_path();
863 ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
867 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
869 if (key->type == BTRFS_INODE_EXTREF_KEY) {
870 if (btrfs_find_name_in_ext_backref(path, ref_objectid,
871 name, namelen, NULL))
877 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
878 ptr_end = ptr + item_size;
879 while (ptr < ptr_end) {
880 ref = (struct btrfs_inode_ref *)ptr;
881 found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
882 if (found_name_len == namelen) {
883 name_ptr = (unsigned long)(ref + 1);
884 ret = memcmp_extent_buffer(path->nodes[0], name,
891 ptr = (unsigned long)(ref + 1) + found_name_len;
894 btrfs_free_path(path);
898 static inline int __add_inode_ref(struct btrfs_trans_handle *trans,
899 struct btrfs_root *root,
900 struct btrfs_path *path,
901 struct btrfs_root *log_root,
902 struct inode *dir, struct inode *inode,
903 struct extent_buffer *eb,
904 u64 inode_objectid, u64 parent_objectid,
905 u64 ref_index, char *name, int namelen,
911 struct extent_buffer *leaf;
912 struct btrfs_dir_item *di;
913 struct btrfs_key search_key;
914 struct btrfs_inode_extref *extref;
917 /* Search old style refs */
918 search_key.objectid = inode_objectid;
919 search_key.type = BTRFS_INODE_REF_KEY;
920 search_key.offset = parent_objectid;
921 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
923 struct btrfs_inode_ref *victim_ref;
925 unsigned long ptr_end;
927 leaf = path->nodes[0];
929 /* are we trying to overwrite a back ref for the root directory
930 * if so, just jump out, we're done
932 if (search_key.objectid == search_key.offset)
935 /* check all the names in this back reference to see
936 * if they are in the log. if so, we allow them to stay
937 * otherwise they must be unlinked as a conflict
939 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
940 ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
941 while (ptr < ptr_end) {
942 victim_ref = (struct btrfs_inode_ref *)ptr;
943 victim_name_len = btrfs_inode_ref_name_len(leaf,
945 victim_name = kmalloc(victim_name_len, GFP_NOFS);
949 read_extent_buffer(leaf, victim_name,
950 (unsigned long)(victim_ref + 1),
953 if (!backref_in_log(log_root, &search_key,
958 btrfs_release_path(path);
960 ret = btrfs_unlink_inode(trans, root, dir,
966 ret = btrfs_run_delayed_items(trans, root);
974 ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
978 * NOTE: we have searched root tree and checked the
979 * coresponding ref, it does not need to check again.
983 btrfs_release_path(path);
985 /* Same search but for extended refs */
986 extref = btrfs_lookup_inode_extref(NULL, root, path, name, namelen,
987 inode_objectid, parent_objectid, 0,
989 if (!IS_ERR_OR_NULL(extref)) {
993 struct inode *victim_parent;
995 leaf = path->nodes[0];
997 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
998 base = btrfs_item_ptr_offset(leaf, path->slots[0]);
1000 while (cur_offset < item_size) {
1001 extref = (struct btrfs_inode_extref *)base + cur_offset;
1003 victim_name_len = btrfs_inode_extref_name_len(leaf, extref);
1005 if (btrfs_inode_extref_parent(leaf, extref) != parent_objectid)
1008 victim_name = kmalloc(victim_name_len, GFP_NOFS);
1011 read_extent_buffer(leaf, victim_name, (unsigned long)&extref->name,
1014 search_key.objectid = inode_objectid;
1015 search_key.type = BTRFS_INODE_EXTREF_KEY;
1016 search_key.offset = btrfs_extref_hash(parent_objectid,
1020 if (!backref_in_log(log_root, &search_key,
1021 parent_objectid, victim_name,
1024 victim_parent = read_one_inode(root,
1026 if (victim_parent) {
1028 btrfs_release_path(path);
1030 ret = btrfs_unlink_inode(trans, root,
1036 ret = btrfs_run_delayed_items(
1039 iput(victim_parent);
1050 cur_offset += victim_name_len + sizeof(*extref);
1054 btrfs_release_path(path);
1056 /* look for a conflicting sequence number */
1057 di = btrfs_lookup_dir_index_item(trans, root, path, btrfs_ino(dir),
1058 ref_index, name, namelen, 0);
1059 if (di && !IS_ERR(di)) {
1060 ret = drop_one_dir_item(trans, root, path, dir, di);
1064 btrfs_release_path(path);
1066 /* look for a conflicing name */
1067 di = btrfs_lookup_dir_item(trans, root, path, btrfs_ino(dir),
1069 if (di && !IS_ERR(di)) {
1070 ret = drop_one_dir_item(trans, root, path, dir, di);
1074 btrfs_release_path(path);
1079 static int extref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1080 u32 *namelen, char **name, u64 *index,
1081 u64 *parent_objectid)
1083 struct btrfs_inode_extref *extref;
1085 extref = (struct btrfs_inode_extref *)ref_ptr;
1087 *namelen = btrfs_inode_extref_name_len(eb, extref);
1088 *name = kmalloc(*namelen, GFP_NOFS);
1092 read_extent_buffer(eb, *name, (unsigned long)&extref->name,
1095 *index = btrfs_inode_extref_index(eb, extref);
1096 if (parent_objectid)
1097 *parent_objectid = btrfs_inode_extref_parent(eb, extref);
1102 static int ref_get_fields(struct extent_buffer *eb, unsigned long ref_ptr,
1103 u32 *namelen, char **name, u64 *index)
1105 struct btrfs_inode_ref *ref;
1107 ref = (struct btrfs_inode_ref *)ref_ptr;
1109 *namelen = btrfs_inode_ref_name_len(eb, ref);
1110 *name = kmalloc(*namelen, GFP_NOFS);
1114 read_extent_buffer(eb, *name, (unsigned long)(ref + 1), *namelen);
1116 *index = btrfs_inode_ref_index(eb, ref);
1122 * replay one inode back reference item found in the log tree.
1123 * eb, slot and key refer to the buffer and key found in the log tree.
1124 * root is the destination we are replaying into, and path is for temp
1125 * use by this function. (it should be released on return).
1127 static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
1128 struct btrfs_root *root,
1129 struct btrfs_root *log,
1130 struct btrfs_path *path,
1131 struct extent_buffer *eb, int slot,
1132 struct btrfs_key *key)
1134 struct inode *dir = NULL;
1135 struct inode *inode = NULL;
1136 unsigned long ref_ptr;
1137 unsigned long ref_end;
1141 int search_done = 0;
1142 int log_ref_ver = 0;
1143 u64 parent_objectid;
1146 int ref_struct_size;
1148 ref_ptr = btrfs_item_ptr_offset(eb, slot);
1149 ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
1151 if (key->type == BTRFS_INODE_EXTREF_KEY) {
1152 struct btrfs_inode_extref *r;
1154 ref_struct_size = sizeof(struct btrfs_inode_extref);
1156 r = (struct btrfs_inode_extref *)ref_ptr;
1157 parent_objectid = btrfs_inode_extref_parent(eb, r);
1159 ref_struct_size = sizeof(struct btrfs_inode_ref);
1160 parent_objectid = key->offset;
1162 inode_objectid = key->objectid;
1165 * it is possible that we didn't log all the parent directories
1166 * for a given inode. If we don't find the dir, just don't
1167 * copy the back ref in. The link count fixup code will take
1170 dir = read_one_inode(root, parent_objectid);
1176 inode = read_one_inode(root, inode_objectid);
1182 while (ref_ptr < ref_end) {
1184 ret = extref_get_fields(eb, ref_ptr, &namelen, &name,
1185 &ref_index, &parent_objectid);
1187 * parent object can change from one array
1191 dir = read_one_inode(root, parent_objectid);
1197 ret = ref_get_fields(eb, ref_ptr, &namelen, &name,
1203 /* if we already have a perfect match, we're done */
1204 if (!inode_in_dir(root, path, btrfs_ino(dir), btrfs_ino(inode),
1205 ref_index, name, namelen)) {
1207 * look for a conflicting back reference in the
1208 * metadata. if we find one we have to unlink that name
1209 * of the file before we add our new link. Later on, we
1210 * overwrite any existing back reference, and we don't
1211 * want to create dangling pointers in the directory.
1215 ret = __add_inode_ref(trans, root, path, log,
1219 ref_index, name, namelen,
1228 /* insert our name */
1229 ret = btrfs_add_link(trans, dir, inode, name, namelen,
1234 btrfs_update_inode(trans, root, inode);
1237 ref_ptr = (unsigned long)(ref_ptr + ref_struct_size) + namelen;
1246 /* finally write the back reference in the inode */
1247 ret = overwrite_item(trans, root, path, eb, slot, key);
1249 btrfs_release_path(path);
1256 static int insert_orphan_item(struct btrfs_trans_handle *trans,
1257 struct btrfs_root *root, u64 offset)
1260 ret = btrfs_find_item(root, NULL, BTRFS_ORPHAN_OBJECTID,
1261 offset, BTRFS_ORPHAN_ITEM_KEY, NULL);
1263 ret = btrfs_insert_orphan_item(trans, root, offset);
1267 static int count_inode_extrefs(struct btrfs_root *root,
1268 struct inode *inode, struct btrfs_path *path)
1272 unsigned int nlink = 0;
1275 u64 inode_objectid = btrfs_ino(inode);
1278 struct btrfs_inode_extref *extref;
1279 struct extent_buffer *leaf;
1282 ret = btrfs_find_one_extref(root, inode_objectid, offset, path,
1287 leaf = path->nodes[0];
1288 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1289 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1291 while (cur_offset < item_size) {
1292 extref = (struct btrfs_inode_extref *) (ptr + cur_offset);
1293 name_len = btrfs_inode_extref_name_len(leaf, extref);
1297 cur_offset += name_len + sizeof(*extref);
1301 btrfs_release_path(path);
1303 btrfs_release_path(path);
1310 static int count_inode_refs(struct btrfs_root *root,
1311 struct inode *inode, struct btrfs_path *path)
1314 struct btrfs_key key;
1315 unsigned int nlink = 0;
1317 unsigned long ptr_end;
1319 u64 ino = btrfs_ino(inode);
1322 key.type = BTRFS_INODE_REF_KEY;
1323 key.offset = (u64)-1;
1326 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1330 if (path->slots[0] == 0)
1335 btrfs_item_key_to_cpu(path->nodes[0], &key,
1337 if (key.objectid != ino ||
1338 key.type != BTRFS_INODE_REF_KEY)
1340 ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
1341 ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
1343 while (ptr < ptr_end) {
1344 struct btrfs_inode_ref *ref;
1346 ref = (struct btrfs_inode_ref *)ptr;
1347 name_len = btrfs_inode_ref_name_len(path->nodes[0],
1349 ptr = (unsigned long)(ref + 1) + name_len;
1353 if (key.offset == 0)
1355 if (path->slots[0] > 0) {
1360 btrfs_release_path(path);
1362 btrfs_release_path(path);
1368 * There are a few corners where the link count of the file can't
1369 * be properly maintained during replay. So, instead of adding
1370 * lots of complexity to the log code, we just scan the backrefs
1371 * for any file that has been through replay.
1373 * The scan will update the link count on the inode to reflect the
1374 * number of back refs found. If it goes down to zero, the iput
1375 * will free the inode.
1377 static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
1378 struct btrfs_root *root,
1379 struct inode *inode)
1381 struct btrfs_path *path;
1384 u64 ino = btrfs_ino(inode);
1386 path = btrfs_alloc_path();
1390 ret = count_inode_refs(root, inode, path);
1396 ret = count_inode_extrefs(root, inode, path);
1407 if (nlink != inode->i_nlink) {
1408 set_nlink(inode, nlink);
1409 btrfs_update_inode(trans, root, inode);
1411 BTRFS_I(inode)->index_cnt = (u64)-1;
1413 if (inode->i_nlink == 0) {
1414 if (S_ISDIR(inode->i_mode)) {
1415 ret = replay_dir_deletes(trans, root, NULL, path,
1420 ret = insert_orphan_item(trans, root, ino);
1424 btrfs_free_path(path);
1428 static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1429 struct btrfs_root *root,
1430 struct btrfs_path *path)
1433 struct btrfs_key key;
1434 struct inode *inode;
1436 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1437 key.type = BTRFS_ORPHAN_ITEM_KEY;
1438 key.offset = (u64)-1;
1440 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1445 if (path->slots[0] == 0)
1450 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1451 if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1452 key.type != BTRFS_ORPHAN_ITEM_KEY)
1455 ret = btrfs_del_item(trans, root, path);
1459 btrfs_release_path(path);
1460 inode = read_one_inode(root, key.offset);
1464 ret = fixup_inode_link_count(trans, root, inode);
1470 * fixup on a directory may create new entries,
1471 * make sure we always look for the highset possible
1474 key.offset = (u64)-1;
1478 btrfs_release_path(path);
1484 * record a given inode in the fixup dir so we can check its link
1485 * count when replay is done. The link count is incremented here
1486 * so the inode won't go away until we check it
1488 static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1489 struct btrfs_root *root,
1490 struct btrfs_path *path,
1493 struct btrfs_key key;
1495 struct inode *inode;
1497 inode = read_one_inode(root, objectid);
1501 key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1502 key.type = BTRFS_ORPHAN_ITEM_KEY;
1503 key.offset = objectid;
1505 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1507 btrfs_release_path(path);
1509 if (!inode->i_nlink)
1510 set_nlink(inode, 1);
1513 ret = btrfs_update_inode(trans, root, inode);
1514 } else if (ret == -EEXIST) {
1517 BUG(); /* Logic Error */
1525 * when replaying the log for a directory, we only insert names
1526 * for inodes that actually exist. This means an fsync on a directory
1527 * does not implicitly fsync all the new files in it
1529 static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root,
1531 struct btrfs_path *path,
1532 u64 dirid, u64 index,
1533 char *name, int name_len, u8 type,
1534 struct btrfs_key *location)
1536 struct inode *inode;
1540 inode = read_one_inode(root, location->objectid);
1544 dir = read_one_inode(root, dirid);
1550 ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1552 /* FIXME, put inode into FIXUP list */
1560 * take a single entry in a log directory item and replay it into
1563 * if a conflicting item exists in the subdirectory already,
1564 * the inode it points to is unlinked and put into the link count
1567 * If a name from the log points to a file or directory that does
1568 * not exist in the FS, it is skipped. fsyncs on directories
1569 * do not force down inodes inside that directory, just changes to the
1570 * names or unlinks in a directory.
1572 static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1573 struct btrfs_root *root,
1574 struct btrfs_path *path,
1575 struct extent_buffer *eb,
1576 struct btrfs_dir_item *di,
1577 struct btrfs_key *key)
1581 struct btrfs_dir_item *dst_di;
1582 struct btrfs_key found_key;
1583 struct btrfs_key log_key;
1588 bool update_size = (key->type == BTRFS_DIR_INDEX_KEY);
1590 dir = read_one_inode(root, key->objectid);
1594 name_len = btrfs_dir_name_len(eb, di);
1595 name = kmalloc(name_len, GFP_NOFS);
1601 log_type = btrfs_dir_type(eb, di);
1602 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1605 btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1606 exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1611 btrfs_release_path(path);
1613 if (key->type == BTRFS_DIR_ITEM_KEY) {
1614 dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1616 } else if (key->type == BTRFS_DIR_INDEX_KEY) {
1617 dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1626 if (IS_ERR_OR_NULL(dst_di)) {
1627 /* we need a sequence number to insert, so we only
1628 * do inserts for the BTRFS_DIR_INDEX_KEY types
1630 if (key->type != BTRFS_DIR_INDEX_KEY)
1635 btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1636 /* the existing item matches the logged item */
1637 if (found_key.objectid == log_key.objectid &&
1638 found_key.type == log_key.type &&
1639 found_key.offset == log_key.offset &&
1640 btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1641 update_size = false;
1646 * don't drop the conflicting directory entry if the inode
1647 * for the new entry doesn't exist
1652 ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1656 if (key->type == BTRFS_DIR_INDEX_KEY)
1659 btrfs_release_path(path);
1660 if (!ret && update_size) {
1661 btrfs_i_size_write(dir, dir->i_size + name_len * 2);
1662 ret = btrfs_update_inode(trans, root, dir);
1669 btrfs_release_path(path);
1670 ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1671 name, name_len, log_type, &log_key);
1672 if (ret && ret != -ENOENT)
1674 update_size = false;
1680 * find all the names in a directory item and reconcile them into
1681 * the subvolume. Only BTRFS_DIR_ITEM_KEY types will have more than
1682 * one name in a directory item, but the same code gets used for
1683 * both directory index types
1685 static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1686 struct btrfs_root *root,
1687 struct btrfs_path *path,
1688 struct extent_buffer *eb, int slot,
1689 struct btrfs_key *key)
1692 u32 item_size = btrfs_item_size_nr(eb, slot);
1693 struct btrfs_dir_item *di;
1696 unsigned long ptr_end;
1698 ptr = btrfs_item_ptr_offset(eb, slot);
1699 ptr_end = ptr + item_size;
1700 while (ptr < ptr_end) {
1701 di = (struct btrfs_dir_item *)ptr;
1702 if (verify_dir_item(root, eb, di))
1704 name_len = btrfs_dir_name_len(eb, di);
1705 ret = replay_one_name(trans, root, path, eb, di, key);
1708 ptr = (unsigned long)(di + 1);
1715 * directory replay has two parts. There are the standard directory
1716 * items in the log copied from the subvolume, and range items
1717 * created in the log while the subvolume was logged.
1719 * The range items tell us which parts of the key space the log
1720 * is authoritative for. During replay, if a key in the subvolume
1721 * directory is in a logged range item, but not actually in the log
1722 * that means it was deleted from the directory before the fsync
1723 * and should be removed.
1725 static noinline int find_dir_range(struct btrfs_root *root,
1726 struct btrfs_path *path,
1727 u64 dirid, int key_type,
1728 u64 *start_ret, u64 *end_ret)
1730 struct btrfs_key key;
1732 struct btrfs_dir_log_item *item;
1736 if (*start_ret == (u64)-1)
1739 key.objectid = dirid;
1740 key.type = key_type;
1741 key.offset = *start_ret;
1743 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1747 if (path->slots[0] == 0)
1752 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1754 if (key.type != key_type || key.objectid != dirid) {
1758 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1759 struct btrfs_dir_log_item);
1760 found_end = btrfs_dir_log_end(path->nodes[0], item);
1762 if (*start_ret >= key.offset && *start_ret <= found_end) {
1764 *start_ret = key.offset;
1765 *end_ret = found_end;
1770 /* check the next slot in the tree to see if it is a valid item */
1771 nritems = btrfs_header_nritems(path->nodes[0]);
1772 if (path->slots[0] >= nritems) {
1773 ret = btrfs_next_leaf(root, path);
1780 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1782 if (key.type != key_type || key.objectid != dirid) {
1786 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1787 struct btrfs_dir_log_item);
1788 found_end = btrfs_dir_log_end(path->nodes[0], item);
1789 *start_ret = key.offset;
1790 *end_ret = found_end;
1793 btrfs_release_path(path);
1798 * this looks for a given directory item in the log. If the directory
1799 * item is not in the log, the item is removed and the inode it points
1802 static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1803 struct btrfs_root *root,
1804 struct btrfs_root *log,
1805 struct btrfs_path *path,
1806 struct btrfs_path *log_path,
1808 struct btrfs_key *dir_key)
1811 struct extent_buffer *eb;
1814 struct btrfs_dir_item *di;
1815 struct btrfs_dir_item *log_di;
1818 unsigned long ptr_end;
1820 struct inode *inode;
1821 struct btrfs_key location;
1824 eb = path->nodes[0];
1825 slot = path->slots[0];
1826 item_size = btrfs_item_size_nr(eb, slot);
1827 ptr = btrfs_item_ptr_offset(eb, slot);
1828 ptr_end = ptr + item_size;
1829 while (ptr < ptr_end) {
1830 di = (struct btrfs_dir_item *)ptr;
1831 if (verify_dir_item(root, eb, di)) {
1836 name_len = btrfs_dir_name_len(eb, di);
1837 name = kmalloc(name_len, GFP_NOFS);
1842 read_extent_buffer(eb, name, (unsigned long)(di + 1),
1845 if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1846 log_di = btrfs_lookup_dir_item(trans, log, log_path,
1849 } else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1850 log_di = btrfs_lookup_dir_index_item(trans, log,
1856 if (!log_di || (IS_ERR(log_di) && PTR_ERR(log_di) == -ENOENT)) {
1857 btrfs_dir_item_key_to_cpu(eb, di, &location);
1858 btrfs_release_path(path);
1859 btrfs_release_path(log_path);
1860 inode = read_one_inode(root, location.objectid);
1866 ret = link_to_fixup_dir(trans, root,
1867 path, location.objectid);
1875 ret = btrfs_unlink_inode(trans, root, dir, inode,
1878 ret = btrfs_run_delayed_items(trans, root);
1884 /* there might still be more names under this key
1885 * check and repeat if required
1887 ret = btrfs_search_slot(NULL, root, dir_key, path,
1893 } else if (IS_ERR(log_di)) {
1895 return PTR_ERR(log_di);
1897 btrfs_release_path(log_path);
1900 ptr = (unsigned long)(di + 1);
1905 btrfs_release_path(path);
1906 btrfs_release_path(log_path);
1911 * deletion replay happens before we copy any new directory items
1912 * out of the log or out of backreferences from inodes. It
1913 * scans the log to find ranges of keys that log is authoritative for,
1914 * and then scans the directory to find items in those ranges that are
1915 * not present in the log.
1917 * Anything we don't find in the log is unlinked and removed from the
1920 static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1921 struct btrfs_root *root,
1922 struct btrfs_root *log,
1923 struct btrfs_path *path,
1924 u64 dirid, int del_all)
1928 int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1930 struct btrfs_key dir_key;
1931 struct btrfs_key found_key;
1932 struct btrfs_path *log_path;
1935 dir_key.objectid = dirid;
1936 dir_key.type = BTRFS_DIR_ITEM_KEY;
1937 log_path = btrfs_alloc_path();
1941 dir = read_one_inode(root, dirid);
1942 /* it isn't an error if the inode isn't there, that can happen
1943 * because we replay the deletes before we copy in the inode item
1947 btrfs_free_path(log_path);
1955 range_end = (u64)-1;
1957 ret = find_dir_range(log, path, dirid, key_type,
1958 &range_start, &range_end);
1963 dir_key.offset = range_start;
1966 ret = btrfs_search_slot(NULL, root, &dir_key, path,
1971 nritems = btrfs_header_nritems(path->nodes[0]);
1972 if (path->slots[0] >= nritems) {
1973 ret = btrfs_next_leaf(root, path);
1977 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1979 if (found_key.objectid != dirid ||
1980 found_key.type != dir_key.type)
1983 if (found_key.offset > range_end)
1986 ret = check_item_in_log(trans, root, log, path,
1991 if (found_key.offset == (u64)-1)
1993 dir_key.offset = found_key.offset + 1;
1995 btrfs_release_path(path);
1996 if (range_end == (u64)-1)
1998 range_start = range_end + 1;
2003 if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
2004 key_type = BTRFS_DIR_LOG_INDEX_KEY;
2005 dir_key.type = BTRFS_DIR_INDEX_KEY;
2006 btrfs_release_path(path);
2010 btrfs_release_path(path);
2011 btrfs_free_path(log_path);
2017 * the process_func used to replay items from the log tree. This
2018 * gets called in two different stages. The first stage just looks
2019 * for inodes and makes sure they are all copied into the subvolume.
2021 * The second stage copies all the other item types from the log into
2022 * the subvolume. The two stage approach is slower, but gets rid of
2023 * lots of complexity around inodes referencing other inodes that exist
2024 * only in the log (references come from either directory items or inode
2027 static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
2028 struct walk_control *wc, u64 gen)
2031 struct btrfs_path *path;
2032 struct btrfs_root *root = wc->replay_dest;
2033 struct btrfs_key key;
2038 ret = btrfs_read_buffer(eb, gen);
2042 level = btrfs_header_level(eb);
2047 path = btrfs_alloc_path();
2051 nritems = btrfs_header_nritems(eb);
2052 for (i = 0; i < nritems; i++) {
2053 btrfs_item_key_to_cpu(eb, &key, i);
2055 /* inode keys are done during the first stage */
2056 if (key.type == BTRFS_INODE_ITEM_KEY &&
2057 wc->stage == LOG_WALK_REPLAY_INODES) {
2058 struct btrfs_inode_item *inode_item;
2061 inode_item = btrfs_item_ptr(eb, i,
2062 struct btrfs_inode_item);
2063 mode = btrfs_inode_mode(eb, inode_item);
2064 if (S_ISDIR(mode)) {
2065 ret = replay_dir_deletes(wc->trans,
2066 root, log, path, key.objectid, 0);
2070 ret = overwrite_item(wc->trans, root, path,
2075 /* for regular files, make sure corresponding
2076 * orhpan item exist. extents past the new EOF
2077 * will be truncated later by orphan cleanup.
2079 if (S_ISREG(mode)) {
2080 ret = insert_orphan_item(wc->trans, root,
2086 ret = link_to_fixup_dir(wc->trans, root,
2087 path, key.objectid);
2092 if (key.type == BTRFS_DIR_INDEX_KEY &&
2093 wc->stage == LOG_WALK_REPLAY_DIR_INDEX) {
2094 ret = replay_one_dir_item(wc->trans, root, path,
2100 if (wc->stage < LOG_WALK_REPLAY_ALL)
2103 /* these keys are simply copied */
2104 if (key.type == BTRFS_XATTR_ITEM_KEY) {
2105 ret = overwrite_item(wc->trans, root, path,
2109 } else if (key.type == BTRFS_INODE_REF_KEY ||
2110 key.type == BTRFS_INODE_EXTREF_KEY) {
2111 ret = add_inode_ref(wc->trans, root, log, path,
2113 if (ret && ret != -ENOENT)
2116 } else if (key.type == BTRFS_EXTENT_DATA_KEY) {
2117 ret = replay_one_extent(wc->trans, root, path,
2121 } else if (key.type == BTRFS_DIR_ITEM_KEY) {
2122 ret = replay_one_dir_item(wc->trans, root, path,
2128 btrfs_free_path(path);
2132 static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
2133 struct btrfs_root *root,
2134 struct btrfs_path *path, int *level,
2135 struct walk_control *wc)
2140 struct extent_buffer *next;
2141 struct extent_buffer *cur;
2142 struct extent_buffer *parent;
2146 WARN_ON(*level < 0);
2147 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2149 while (*level > 0) {
2150 WARN_ON(*level < 0);
2151 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2152 cur = path->nodes[*level];
2154 WARN_ON(btrfs_header_level(cur) != *level);
2156 if (path->slots[*level] >=
2157 btrfs_header_nritems(cur))
2160 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2161 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2162 blocksize = root->nodesize;
2164 parent = path->nodes[*level];
2165 root_owner = btrfs_header_owner(parent);
2167 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
2172 ret = wc->process_func(root, next, wc, ptr_gen);
2174 free_extent_buffer(next);
2178 path->slots[*level]++;
2180 ret = btrfs_read_buffer(next, ptr_gen);
2182 free_extent_buffer(next);
2187 btrfs_tree_lock(next);
2188 btrfs_set_lock_blocking(next);
2189 clean_tree_block(trans, root, next);
2190 btrfs_wait_tree_block_writeback(next);
2191 btrfs_tree_unlock(next);
2194 WARN_ON(root_owner !=
2195 BTRFS_TREE_LOG_OBJECTID);
2196 ret = btrfs_free_and_pin_reserved_extent(root,
2199 free_extent_buffer(next);
2203 free_extent_buffer(next);
2206 ret = btrfs_read_buffer(next, ptr_gen);
2208 free_extent_buffer(next);
2212 WARN_ON(*level <= 0);
2213 if (path->nodes[*level-1])
2214 free_extent_buffer(path->nodes[*level-1]);
2215 path->nodes[*level-1] = next;
2216 *level = btrfs_header_level(next);
2217 path->slots[*level] = 0;
2220 WARN_ON(*level < 0);
2221 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2223 path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
2229 static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
2230 struct btrfs_root *root,
2231 struct btrfs_path *path, int *level,
2232 struct walk_control *wc)
2239 for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2240 slot = path->slots[i];
2241 if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
2244 WARN_ON(*level == 0);
2247 struct extent_buffer *parent;
2248 if (path->nodes[*level] == root->node)
2249 parent = path->nodes[*level];
2251 parent = path->nodes[*level + 1];
2253 root_owner = btrfs_header_owner(parent);
2254 ret = wc->process_func(root, path->nodes[*level], wc,
2255 btrfs_header_generation(path->nodes[*level]));
2260 struct extent_buffer *next;
2262 next = path->nodes[*level];
2265 btrfs_tree_lock(next);
2266 btrfs_set_lock_blocking(next);
2267 clean_tree_block(trans, root, next);
2268 btrfs_wait_tree_block_writeback(next);
2269 btrfs_tree_unlock(next);
2272 WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
2273 ret = btrfs_free_and_pin_reserved_extent(root,
2274 path->nodes[*level]->start,
2275 path->nodes[*level]->len);
2279 free_extent_buffer(path->nodes[*level]);
2280 path->nodes[*level] = NULL;
2288 * drop the reference count on the tree rooted at 'snap'. This traverses
2289 * the tree freeing any blocks that have a ref count of zero after being
2292 static int walk_log_tree(struct btrfs_trans_handle *trans,
2293 struct btrfs_root *log, struct walk_control *wc)
2298 struct btrfs_path *path;
2301 path = btrfs_alloc_path();
2305 level = btrfs_header_level(log->node);
2307 path->nodes[level] = log->node;
2308 extent_buffer_get(log->node);
2309 path->slots[level] = 0;
2312 wret = walk_down_log_tree(trans, log, path, &level, wc);
2320 wret = walk_up_log_tree(trans, log, path, &level, wc);
2329 /* was the root node processed? if not, catch it here */
2330 if (path->nodes[orig_level]) {
2331 ret = wc->process_func(log, path->nodes[orig_level], wc,
2332 btrfs_header_generation(path->nodes[orig_level]));
2336 struct extent_buffer *next;
2338 next = path->nodes[orig_level];
2341 btrfs_tree_lock(next);
2342 btrfs_set_lock_blocking(next);
2343 clean_tree_block(trans, log, next);
2344 btrfs_wait_tree_block_writeback(next);
2345 btrfs_tree_unlock(next);
2348 WARN_ON(log->root_key.objectid !=
2349 BTRFS_TREE_LOG_OBJECTID);
2350 ret = btrfs_free_and_pin_reserved_extent(log, next->start,
2358 btrfs_free_path(path);
2363 * helper function to update the item for a given subvolumes log root
2364 * in the tree of log roots
2366 static int update_log_root(struct btrfs_trans_handle *trans,
2367 struct btrfs_root *log)
2371 if (log->log_transid == 1) {
2372 /* insert root item on the first sync */
2373 ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
2374 &log->root_key, &log->root_item);
2376 ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
2377 &log->root_key, &log->root_item);
2382 static void wait_log_commit(struct btrfs_trans_handle *trans,
2383 struct btrfs_root *root, int transid)
2386 int index = transid % 2;
2389 * we only allow two pending log transactions at a time,
2390 * so we know that if ours is more than 2 older than the
2391 * current transaction, we're done
2394 prepare_to_wait(&root->log_commit_wait[index],
2395 &wait, TASK_UNINTERRUPTIBLE);
2396 mutex_unlock(&root->log_mutex);
2398 if (root->log_transid_committed < transid &&
2399 atomic_read(&root->log_commit[index]))
2402 finish_wait(&root->log_commit_wait[index], &wait);
2403 mutex_lock(&root->log_mutex);
2404 } while (root->log_transid_committed < transid &&
2405 atomic_read(&root->log_commit[index]));
2408 static void wait_for_writer(struct btrfs_trans_handle *trans,
2409 struct btrfs_root *root)
2413 while (atomic_read(&root->log_writers)) {
2414 prepare_to_wait(&root->log_writer_wait,
2415 &wait, TASK_UNINTERRUPTIBLE);
2416 mutex_unlock(&root->log_mutex);
2417 if (atomic_read(&root->log_writers))
2419 mutex_lock(&root->log_mutex);
2420 finish_wait(&root->log_writer_wait, &wait);
2424 static inline void btrfs_remove_log_ctx(struct btrfs_root *root,
2425 struct btrfs_log_ctx *ctx)
2430 mutex_lock(&root->log_mutex);
2431 list_del_init(&ctx->list);
2432 mutex_unlock(&root->log_mutex);
2436 * Invoked in log mutex context, or be sure there is no other task which
2437 * can access the list.
2439 static inline void btrfs_remove_all_log_ctxs(struct btrfs_root *root,
2440 int index, int error)
2442 struct btrfs_log_ctx *ctx;
2445 INIT_LIST_HEAD(&root->log_ctxs[index]);
2449 list_for_each_entry(ctx, &root->log_ctxs[index], list)
2450 ctx->log_ret = error;
2452 INIT_LIST_HEAD(&root->log_ctxs[index]);
2456 * btrfs_sync_log does sends a given tree log down to the disk and
2457 * updates the super blocks to record it. When this call is done,
2458 * you know that any inodes previously logged are safely on disk only
2461 * Any other return value means you need to call btrfs_commit_transaction.
2462 * Some of the edge cases for fsyncing directories that have had unlinks
2463 * or renames done in the past mean that sometimes the only safe
2464 * fsync is to commit the whole FS. When btrfs_sync_log returns -EAGAIN,
2465 * that has happened.
2467 int btrfs_sync_log(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *root, struct btrfs_log_ctx *ctx)
2474 struct btrfs_root *log = root->log_root;
2475 struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
2476 int log_transid = 0;
2477 struct btrfs_log_ctx root_log_ctx;
2478 struct blk_plug plug;
2480 mutex_lock(&root->log_mutex);
2481 log_transid = ctx->log_transid;
2482 if (root->log_transid_committed >= log_transid) {
2483 mutex_unlock(&root->log_mutex);
2484 return ctx->log_ret;
2487 index1 = log_transid % 2;
2488 if (atomic_read(&root->log_commit[index1])) {
2489 wait_log_commit(trans, root, log_transid);
2490 mutex_unlock(&root->log_mutex);
2491 return ctx->log_ret;
2493 ASSERT(log_transid == root->log_transid);
2494 atomic_set(&root->log_commit[index1], 1);
2496 /* wait for previous tree log sync to complete */
2497 if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
2498 wait_log_commit(trans, root, log_transid - 1);
2501 int batch = atomic_read(&root->log_batch);
2502 /* when we're on an ssd, just kick the log commit out */
2503 if (!btrfs_test_opt(root, SSD) &&
2504 test_bit(BTRFS_ROOT_MULTI_LOG_TASKS, &root->state)) {
2505 mutex_unlock(&root->log_mutex);
2506 schedule_timeout_uninterruptible(1);
2507 mutex_lock(&root->log_mutex);
2509 wait_for_writer(trans, root);
2510 if (batch == atomic_read(&root->log_batch))
2514 /* bail out if we need to do a full commit */
2515 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2517 btrfs_free_logged_extents(log, log_transid);
2518 mutex_unlock(&root->log_mutex);
2522 if (log_transid % 2 == 0)
2523 mark = EXTENT_DIRTY;
2527 /* we start IO on all the marked extents here, but we don't actually
2528 * wait for them until later.
2530 blk_start_plug(&plug);
2531 ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2533 blk_finish_plug(&plug);
2534 btrfs_abort_transaction(trans, root, ret);
2535 btrfs_free_logged_extents(log, log_transid);
2536 btrfs_set_log_full_commit(root->fs_info, trans);
2537 mutex_unlock(&root->log_mutex);
2541 btrfs_set_root_node(&log->root_item, log->node);
2543 root->log_transid++;
2544 log->log_transid = root->log_transid;
2545 root->log_start_pid = 0;
2547 * IO has been started, blocks of the log tree have WRITTEN flag set
2548 * in their headers. new modifications of the log will be written to
2549 * new positions. so it's safe to allow log writers to go in.
2551 mutex_unlock(&root->log_mutex);
2553 btrfs_init_log_ctx(&root_log_ctx);
2555 mutex_lock(&log_root_tree->log_mutex);
2556 atomic_inc(&log_root_tree->log_batch);
2557 atomic_inc(&log_root_tree->log_writers);
2559 index2 = log_root_tree->log_transid % 2;
2560 list_add_tail(&root_log_ctx.list, &log_root_tree->log_ctxs[index2]);
2561 root_log_ctx.log_transid = log_root_tree->log_transid;
2563 mutex_unlock(&log_root_tree->log_mutex);
2565 ret = update_log_root(trans, log);
2567 mutex_lock(&log_root_tree->log_mutex);
2568 if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2570 if (waitqueue_active(&log_root_tree->log_writer_wait))
2571 wake_up(&log_root_tree->log_writer_wait);
2575 if (!list_empty(&root_log_ctx.list))
2576 list_del_init(&root_log_ctx.list);
2578 blk_finish_plug(&plug);
2579 btrfs_set_log_full_commit(root->fs_info, trans);
2581 if (ret != -ENOSPC) {
2582 btrfs_abort_transaction(trans, root, ret);
2583 mutex_unlock(&log_root_tree->log_mutex);
2586 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2587 btrfs_free_logged_extents(log, log_transid);
2588 mutex_unlock(&log_root_tree->log_mutex);
2593 if (log_root_tree->log_transid_committed >= root_log_ctx.log_transid) {
2594 blk_finish_plug(&plug);
2595 mutex_unlock(&log_root_tree->log_mutex);
2596 ret = root_log_ctx.log_ret;
2600 index2 = root_log_ctx.log_transid % 2;
2601 if (atomic_read(&log_root_tree->log_commit[index2])) {
2602 blk_finish_plug(&plug);
2603 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages,
2605 btrfs_wait_logged_extents(trans, log, log_transid);
2606 wait_log_commit(trans, log_root_tree,
2607 root_log_ctx.log_transid);
2608 mutex_unlock(&log_root_tree->log_mutex);
2610 ret = root_log_ctx.log_ret;
2613 ASSERT(root_log_ctx.log_transid == log_root_tree->log_transid);
2614 atomic_set(&log_root_tree->log_commit[index2], 1);
2616 if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2617 wait_log_commit(trans, log_root_tree,
2618 root_log_ctx.log_transid - 1);
2621 wait_for_writer(trans, log_root_tree);
2624 * now that we've moved on to the tree of log tree roots,
2625 * check the full commit flag again
2627 if (btrfs_need_log_full_commit(root->fs_info, trans)) {
2628 blk_finish_plug(&plug);
2629 btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2630 btrfs_free_logged_extents(log, log_transid);
2631 mutex_unlock(&log_root_tree->log_mutex);
2633 goto out_wake_log_root;
2636 ret = btrfs_write_marked_extents(log_root_tree,
2637 &log_root_tree->dirty_log_pages,
2638 EXTENT_DIRTY | EXTENT_NEW);
2639 blk_finish_plug(&plug);
2641 btrfs_set_log_full_commit(root->fs_info, trans);
2642 btrfs_abort_transaction(trans, root, ret);
2643 btrfs_free_logged_extents(log, log_transid);
2644 mutex_unlock(&log_root_tree->log_mutex);
2645 goto out_wake_log_root;
2647 ret = btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2649 ret = btrfs_wait_marked_extents(log_root_tree,
2650 &log_root_tree->dirty_log_pages,
2651 EXTENT_NEW | EXTENT_DIRTY);
2653 btrfs_set_log_full_commit(root->fs_info, trans);
2654 btrfs_free_logged_extents(log, log_transid);
2655 mutex_unlock(&log_root_tree->log_mutex);
2656 goto out_wake_log_root;
2658 btrfs_wait_logged_extents(trans, log, log_transid);
2660 btrfs_set_super_log_root(root->fs_info->super_for_commit,
2661 log_root_tree->node->start);
2662 btrfs_set_super_log_root_level(root->fs_info->super_for_commit,
2663 btrfs_header_level(log_root_tree->node));
2665 log_root_tree->log_transid++;
2666 mutex_unlock(&log_root_tree->log_mutex);
2669 * nobody else is going to jump in and write the the ctree
2670 * super here because the log_commit atomic below is protecting
2671 * us. We must be called with a transaction handle pinning
2672 * the running transaction open, so a full commit can't hop
2673 * in and cause problems either.
2675 ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
2677 btrfs_set_log_full_commit(root->fs_info, trans);
2678 btrfs_abort_transaction(trans, root, ret);
2679 goto out_wake_log_root;
2682 mutex_lock(&root->log_mutex);
2683 if (root->last_log_commit < log_transid)
2684 root->last_log_commit = log_transid;
2685 mutex_unlock(&root->log_mutex);
2689 * We needn't get log_mutex here because we are sure all
2690 * the other tasks are blocked.
2692 btrfs_remove_all_log_ctxs(log_root_tree, index2, ret);
2694 mutex_lock(&log_root_tree->log_mutex);
2695 log_root_tree->log_transid_committed++;
2696 atomic_set(&log_root_tree->log_commit[index2], 0);
2697 mutex_unlock(&log_root_tree->log_mutex);
2699 if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2700 wake_up(&log_root_tree->log_commit_wait[index2]);
2703 btrfs_remove_all_log_ctxs(root, index1, ret);
2705 mutex_lock(&root->log_mutex);
2706 root->log_transid_committed++;
2707 atomic_set(&root->log_commit[index1], 0);
2708 mutex_unlock(&root->log_mutex);
2710 if (waitqueue_active(&root->log_commit_wait[index1]))
2711 wake_up(&root->log_commit_wait[index1]);
2715 static void free_log_tree(struct btrfs_trans_handle *trans,
2716 struct btrfs_root *log)
2721 struct walk_control wc = {
2723 .process_func = process_one_buffer
2726 ret = walk_log_tree(trans, log, &wc);
2727 /* I don't think this can happen but just in case */
2729 btrfs_abort_transaction(trans, log, ret);
2732 ret = find_first_extent_bit(&log->dirty_log_pages,
2733 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW,
2738 clear_extent_bits(&log->dirty_log_pages, start, end,
2739 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2743 * We may have short-circuited the log tree with the full commit logic
2744 * and left ordered extents on our list, so clear these out to keep us
2745 * from leaking inodes and memory.
2747 btrfs_free_logged_extents(log, 0);
2748 btrfs_free_logged_extents(log, 1);
2750 free_extent_buffer(log->node);
2755 * free all the extents used by the tree log. This should be called
2756 * at commit time of the full transaction
2758 int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2760 if (root->log_root) {
2761 free_log_tree(trans, root->log_root);
2762 root->log_root = NULL;
2767 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2768 struct btrfs_fs_info *fs_info)
2770 if (fs_info->log_root_tree) {
2771 free_log_tree(trans, fs_info->log_root_tree);
2772 fs_info->log_root_tree = NULL;
2778 * If both a file and directory are logged, and unlinks or renames are
2779 * mixed in, we have a few interesting corners:
2781 * create file X in dir Y
2782 * link file X to X.link in dir Y
2784 * unlink file X but leave X.link
2787 * After a crash we would expect only X.link to exist. But file X
2788 * didn't get fsync'd again so the log has back refs for X and X.link.
2790 * We solve this by removing directory entries and inode backrefs from the
2791 * log when a file that was logged in the current transaction is
2792 * unlinked. Any later fsync will include the updated log entries, and
2793 * we'll be able to reconstruct the proper directory items from backrefs.
2795 * This optimizations allows us to avoid relogging the entire inode
2796 * or the entire directory.
2798 int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2799 struct btrfs_root *root,
2800 const char *name, int name_len,
2801 struct inode *dir, u64 index)
2803 struct btrfs_root *log;
2804 struct btrfs_dir_item *di;
2805 struct btrfs_path *path;
2809 u64 dir_ino = btrfs_ino(dir);
2811 if (BTRFS_I(dir)->logged_trans < trans->transid)
2814 ret = join_running_log_trans(root);
2818 mutex_lock(&BTRFS_I(dir)->log_mutex);
2820 log = root->log_root;
2821 path = btrfs_alloc_path();
2827 di = btrfs_lookup_dir_item(trans, log, path, dir_ino,
2828 name, name_len, -1);
2834 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2835 bytes_del += name_len;
2841 btrfs_release_path(path);
2842 di = btrfs_lookup_dir_index_item(trans, log, path, dir_ino,
2843 index, name, name_len, -1);
2849 ret = btrfs_delete_one_dir_name(trans, log, path, di);
2850 bytes_del += name_len;
2857 /* update the directory size in the log to reflect the names
2861 struct btrfs_key key;
2863 key.objectid = dir_ino;
2865 key.type = BTRFS_INODE_ITEM_KEY;
2866 btrfs_release_path(path);
2868 ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2874 struct btrfs_inode_item *item;
2877 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2878 struct btrfs_inode_item);
2879 i_size = btrfs_inode_size(path->nodes[0], item);
2880 if (i_size > bytes_del)
2881 i_size -= bytes_del;
2884 btrfs_set_inode_size(path->nodes[0], item, i_size);
2885 btrfs_mark_buffer_dirty(path->nodes[0]);
2888 btrfs_release_path(path);
2891 btrfs_free_path(path);
2893 mutex_unlock(&BTRFS_I(dir)->log_mutex);
2894 if (ret == -ENOSPC) {
2895 btrfs_set_log_full_commit(root->fs_info, trans);
2898 btrfs_abort_transaction(trans, root, ret);
2900 btrfs_end_log_trans(root);
2905 /* see comments for btrfs_del_dir_entries_in_log */
2906 int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2907 struct btrfs_root *root,
2908 const char *name, int name_len,
2909 struct inode *inode, u64 dirid)
2911 struct btrfs_root *log;
2915 if (BTRFS_I(inode)->logged_trans < trans->transid)
2918 ret = join_running_log_trans(root);
2921 log = root->log_root;
2922 mutex_lock(&BTRFS_I(inode)->log_mutex);
2924 ret = btrfs_del_inode_ref(trans, log, name, name_len, btrfs_ino(inode),
2926 mutex_unlock(&BTRFS_I(inode)->log_mutex);
2927 if (ret == -ENOSPC) {
2928 btrfs_set_log_full_commit(root->fs_info, trans);
2930 } else if (ret < 0 && ret != -ENOENT)
2931 btrfs_abort_transaction(trans, root, ret);
2932 btrfs_end_log_trans(root);
2938 * creates a range item in the log for 'dirid'. first_offset and
2939 * last_offset tell us which parts of the key space the log should
2940 * be considered authoritative for.
2942 static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2943 struct btrfs_root *log,
2944 struct btrfs_path *path,
2945 int key_type, u64 dirid,
2946 u64 first_offset, u64 last_offset)
2949 struct btrfs_key key;
2950 struct btrfs_dir_log_item *item;
2952 key.objectid = dirid;
2953 key.offset = first_offset;
2954 if (key_type == BTRFS_DIR_ITEM_KEY)
2955 key.type = BTRFS_DIR_LOG_ITEM_KEY;
2957 key.type = BTRFS_DIR_LOG_INDEX_KEY;
2958 ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2962 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2963 struct btrfs_dir_log_item);
2964 btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2965 btrfs_mark_buffer_dirty(path->nodes[0]);
2966 btrfs_release_path(path);
2971 * log all the items included in the current transaction for a given
2972 * directory. This also creates the range items in the log tree required
2973 * to replay anything deleted before the fsync
2975 static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2976 struct btrfs_root *root, struct inode *inode,
2977 struct btrfs_path *path,
2978 struct btrfs_path *dst_path, int key_type,
2979 u64 min_offset, u64 *last_offset_ret)
2981 struct btrfs_key min_key;
2982 struct btrfs_root *log = root->log_root;
2983 struct extent_buffer *src;
2988 u64 first_offset = min_offset;
2989 u64 last_offset = (u64)-1;
2990 u64 ino = btrfs_ino(inode);
2992 log = root->log_root;
2994 min_key.objectid = ino;
2995 min_key.type = key_type;
2996 min_key.offset = min_offset;
2998 ret = btrfs_search_forward(root, &min_key, path, trans->transid);
3001 * we didn't find anything from this transaction, see if there
3002 * is anything at all
3004 if (ret != 0 || min_key.objectid != ino || min_key.type != key_type) {
3005 min_key.objectid = ino;
3006 min_key.type = key_type;
3007 min_key.offset = (u64)-1;
3008 btrfs_release_path(path);
3009 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3011 btrfs_release_path(path);
3014 ret = btrfs_previous_item(root, path, ino, key_type);
3016 /* if ret == 0 there are items for this type,
3017 * create a range to tell us the last key of this type.
3018 * otherwise, there are no items in this directory after
3019 * *min_offset, and we create a range to indicate that.
3022 struct btrfs_key tmp;
3023 btrfs_item_key_to_cpu(path->nodes[0], &tmp,
3025 if (key_type == tmp.type)
3026 first_offset = max(min_offset, tmp.offset) + 1;
3031 /* go backward to find any previous key */
3032 ret = btrfs_previous_item(root, path, ino, key_type);
3034 struct btrfs_key tmp;
3035 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3036 if (key_type == tmp.type) {
3037 first_offset = tmp.offset;
3038 ret = overwrite_item(trans, log, dst_path,
3039 path->nodes[0], path->slots[0],
3047 btrfs_release_path(path);
3049 /* find the first key from this transaction again */
3050 ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
3051 if (WARN_ON(ret != 0))
3055 * we have a block from this transaction, log every item in it
3056 * from our directory
3059 struct btrfs_key tmp;
3060 src = path->nodes[0];
3061 nritems = btrfs_header_nritems(src);
3062 for (i = path->slots[0]; i < nritems; i++) {
3063 btrfs_item_key_to_cpu(src, &min_key, i);
3065 if (min_key.objectid != ino || min_key.type != key_type)
3067 ret = overwrite_item(trans, log, dst_path, src, i,
3074 path->slots[0] = nritems;
3077 * look ahead to the next item and see if it is also
3078 * from this directory and from this transaction
3080 ret = btrfs_next_leaf(root, path);
3082 last_offset = (u64)-1;
3085 btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
3086 if (tmp.objectid != ino || tmp.type != key_type) {
3087 last_offset = (u64)-1;
3090 if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
3091 ret = overwrite_item(trans, log, dst_path,
3092 path->nodes[0], path->slots[0],
3097 last_offset = tmp.offset;
3102 btrfs_release_path(path);
3103 btrfs_release_path(dst_path);
3106 *last_offset_ret = last_offset;
3108 * insert the log range keys to indicate where the log
3111 ret = insert_dir_log_key(trans, log, path, key_type,
3112 ino, first_offset, last_offset);
3120 * logging directories is very similar to logging inodes, We find all the items
3121 * from the current transaction and write them to the log.
3123 * The recovery code scans the directory in the subvolume, and if it finds a
3124 * key in the range logged that is not present in the log tree, then it means
3125 * that dir entry was unlinked during the transaction.
3127 * In order for that scan to work, we must include one key smaller than
3128 * the smallest logged by this transaction and one key larger than the largest
3129 * key logged by this transaction.
3131 static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
3132 struct btrfs_root *root, struct inode *inode,
3133 struct btrfs_path *path,
3134 struct btrfs_path *dst_path)
3139 int key_type = BTRFS_DIR_ITEM_KEY;
3145 ret = log_dir_items(trans, root, inode, path,
3146 dst_path, key_type, min_key,
3150 if (max_key == (u64)-1)
3152 min_key = max_key + 1;
3155 if (key_type == BTRFS_DIR_ITEM_KEY) {
3156 key_type = BTRFS_DIR_INDEX_KEY;
3163 * a helper function to drop items from the log before we relog an
3164 * inode. max_key_type indicates the highest item type to remove.
3165 * This cannot be run for file data extents because it does not
3166 * free the extents they point to.
3168 static int drop_objectid_items(struct btrfs_trans_handle *trans,
3169 struct btrfs_root *log,
3170 struct btrfs_path *path,
3171 u64 objectid, int max_key_type)
3174 struct btrfs_key key;
3175 struct btrfs_key found_key;
3178 key.objectid = objectid;
3179 key.type = max_key_type;
3180 key.offset = (u64)-1;
3183 ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
3184 BUG_ON(ret == 0); /* Logic error */
3188 if (path->slots[0] == 0)
3192 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3195 if (found_key.objectid != objectid)
3198 found_key.offset = 0;
3200 ret = btrfs_bin_search(path->nodes[0], &found_key, 0,
3203 ret = btrfs_del_items(trans, log, path, start_slot,
3204 path->slots[0] - start_slot + 1);
3206 * If start slot isn't 0 then we don't need to re-search, we've
3207 * found the last guy with the objectid in this tree.
3209 if (ret || start_slot != 0)
3211 btrfs_release_path(path);
3213 btrfs_release_path(path);
3219 static void fill_inode_item(struct btrfs_trans_handle *trans,
3220 struct extent_buffer *leaf,
3221 struct btrfs_inode_item *item,
3222 struct inode *inode, int log_inode_only)
3224 struct btrfs_map_token token;
3226 btrfs_init_map_token(&token);
3228 if (log_inode_only) {
3229 /* set the generation to zero so the recover code
3230 * can tell the difference between an logging
3231 * just to say 'this inode exists' and a logging
3232 * to say 'update this inode with these values'
3234 btrfs_set_token_inode_generation(leaf, item, 0, &token);
3235 btrfs_set_token_inode_size(leaf, item, 0, &token);
3237 btrfs_set_token_inode_generation(leaf, item,
3238 BTRFS_I(inode)->generation,
3240 btrfs_set_token_inode_size(leaf, item, inode->i_size, &token);
3243 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3244 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3245 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3246 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3248 btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3249 inode->i_atime.tv_sec, &token);
3250 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3251 inode->i_atime.tv_nsec, &token);
3253 btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3254 inode->i_mtime.tv_sec, &token);
3255 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3256 inode->i_mtime.tv_nsec, &token);
3258 btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3259 inode->i_ctime.tv_sec, &token);
3260 btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3261 inode->i_ctime.tv_nsec, &token);
3263 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3266 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3267 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3268 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3269 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3270 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3273 static int log_inode_item(struct btrfs_trans_handle *trans,
3274 struct btrfs_root *log, struct btrfs_path *path,
3275 struct inode *inode)
3277 struct btrfs_inode_item *inode_item;
3280 ret = btrfs_insert_empty_item(trans, log, path,
3281 &BTRFS_I(inode)->location,
3282 sizeof(*inode_item));
3283 if (ret && ret != -EEXIST)
3285 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3286 struct btrfs_inode_item);
3287 fill_inode_item(trans, path->nodes[0], inode_item, inode, 0);
3288 btrfs_release_path(path);
3292 static noinline int copy_items(struct btrfs_trans_handle *trans,
3293 struct inode *inode,
3294 struct btrfs_path *dst_path,
3295 struct btrfs_path *src_path, u64 *last_extent,
3296 int start_slot, int nr, int inode_only)
3298 unsigned long src_offset;
3299 unsigned long dst_offset;
3300 struct btrfs_root *log = BTRFS_I(inode)->root->log_root;
3301 struct btrfs_file_extent_item *extent;
3302 struct btrfs_inode_item *inode_item;
3303 struct extent_buffer *src = src_path->nodes[0];
3304 struct btrfs_key first_key, last_key, key;
3306 struct btrfs_key *ins_keys;
3310 struct list_head ordered_sums;
3311 int skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3312 bool has_extents = false;
3313 bool need_find_last_extent = true;
3316 INIT_LIST_HEAD(&ordered_sums);
3318 ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
3319 nr * sizeof(u32), GFP_NOFS);
3323 first_key.objectid = (u64)-1;
3325 ins_sizes = (u32 *)ins_data;
3326 ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
3328 for (i = 0; i < nr; i++) {
3329 ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
3330 btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
3332 ret = btrfs_insert_empty_items(trans, log, dst_path,
3333 ins_keys, ins_sizes, nr);
3339 for (i = 0; i < nr; i++, dst_path->slots[0]++) {
3340 dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
3341 dst_path->slots[0]);
3343 src_offset = btrfs_item_ptr_offset(src, start_slot + i);
3345 if ((i == (nr - 1)))
3346 last_key = ins_keys[i];
3348 if (ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
3349 inode_item = btrfs_item_ptr(dst_path->nodes[0],
3351 struct btrfs_inode_item);
3352 fill_inode_item(trans, dst_path->nodes[0], inode_item,
3353 inode, inode_only == LOG_INODE_EXISTS);
3355 copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
3356 src_offset, ins_sizes[i]);
3360 * We set need_find_last_extent here in case we know we were
3361 * processing other items and then walk into the first extent in
3362 * the inode. If we don't hit an extent then nothing changes,
3363 * we'll do the last search the next time around.
3365 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY) {
3367 if (first_key.objectid == (u64)-1)
3368 first_key = ins_keys[i];
3370 need_find_last_extent = false;
3373 /* take a reference on file data extents so that truncates
3374 * or deletes of this inode don't have to relog the inode
3377 if (ins_keys[i].type == BTRFS_EXTENT_DATA_KEY &&
3380 extent = btrfs_item_ptr(src, start_slot + i,
3381 struct btrfs_file_extent_item);
3383 if (btrfs_file_extent_generation(src, extent) < trans->transid)
3386 found_type = btrfs_file_extent_type(src, extent);
3387 if (found_type == BTRFS_FILE_EXTENT_REG) {
3389 ds = btrfs_file_extent_disk_bytenr(src,
3391 /* ds == 0 is a hole */
3395 dl = btrfs_file_extent_disk_num_bytes(src,
3397 cs = btrfs_file_extent_offset(src, extent);
3398 cl = btrfs_file_extent_num_bytes(src,
3400 if (btrfs_file_extent_compression(src,
3406 ret = btrfs_lookup_csums_range(
3407 log->fs_info->csum_root,
3408 ds + cs, ds + cs + cl - 1,
3411 btrfs_release_path(dst_path);
3419 btrfs_mark_buffer_dirty(dst_path->nodes[0]);
3420 btrfs_release_path(dst_path);
3424 * we have to do this after the loop above to avoid changing the
3425 * log tree while trying to change the log tree.
3428 while (!list_empty(&ordered_sums)) {
3429 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3430 struct btrfs_ordered_sum,
3433 ret = btrfs_csum_file_blocks(trans, log, sums);
3434 list_del(&sums->list);
3441 if (need_find_last_extent && *last_extent == first_key.offset) {
3443 * We don't have any leafs between our current one and the one
3444 * we processed before that can have file extent items for our
3445 * inode (and have a generation number smaller than our current
3448 need_find_last_extent = false;
3452 * Because we use btrfs_search_forward we could skip leaves that were
3453 * not modified and then assume *last_extent is valid when it really
3454 * isn't. So back up to the previous leaf and read the end of the last
3455 * extent before we go and fill in holes.
3457 if (need_find_last_extent) {
3460 ret = btrfs_prev_leaf(BTRFS_I(inode)->root, src_path);
3465 if (src_path->slots[0])
3466 src_path->slots[0]--;
3467 src = src_path->nodes[0];
3468 btrfs_item_key_to_cpu(src, &key, src_path->slots[0]);
3469 if (key.objectid != btrfs_ino(inode) ||
3470 key.type != BTRFS_EXTENT_DATA_KEY)
3472 extent = btrfs_item_ptr(src, src_path->slots[0],
3473 struct btrfs_file_extent_item);
3474 if (btrfs_file_extent_type(src, extent) ==
3475 BTRFS_FILE_EXTENT_INLINE) {
3476 len = btrfs_file_extent_inline_len(src,
3479 *last_extent = ALIGN(key.offset + len,
3482 len = btrfs_file_extent_num_bytes(src, extent);
3483 *last_extent = key.offset + len;
3487 /* So we did prev_leaf, now we need to move to the next leaf, but a few
3488 * things could have happened
3490 * 1) A merge could have happened, so we could currently be on a leaf
3491 * that holds what we were copying in the first place.
3492 * 2) A split could have happened, and now not all of the items we want
3493 * are on the same leaf.
3495 * So we need to adjust how we search for holes, we need to drop the
3496 * path and re-search for the first extent key we found, and then walk
3497 * forward until we hit the last one we copied.
3499 if (need_find_last_extent) {
3500 /* btrfs_prev_leaf could return 1 without releasing the path */
3501 btrfs_release_path(src_path);
3502 ret = btrfs_search_slot(NULL, BTRFS_I(inode)->root, &first_key,
3507 src = src_path->nodes[0];
3508 i = src_path->slots[0];
3514 * Ok so here we need to go through and fill in any holes we may have
3515 * to make sure that holes are punched for those areas in case they had
3516 * extents previously.
3522 if (i >= btrfs_header_nritems(src_path->nodes[0])) {
3523 ret = btrfs_next_leaf(BTRFS_I(inode)->root, src_path);
3527 src = src_path->nodes[0];
3531 btrfs_item_key_to_cpu(src, &key, i);
3532 if (!btrfs_comp_cpu_keys(&key, &last_key))
3534 if (key.objectid != btrfs_ino(inode) ||
3535 key.type != BTRFS_EXTENT_DATA_KEY) {
3539 extent = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
3540 if (btrfs_file_extent_type(src, extent) ==
3541 BTRFS_FILE_EXTENT_INLINE) {
3542 len = btrfs_file_extent_inline_len(src, i, extent);
3543 extent_end = ALIGN(key.offset + len, log->sectorsize);
3545 len = btrfs_file_extent_num_bytes(src, extent);
3546 extent_end = key.offset + len;
3550 if (*last_extent == key.offset) {
3551 *last_extent = extent_end;
3554 offset = *last_extent;
3555 len = key.offset - *last_extent;
3556 ret = btrfs_insert_file_extent(trans, log, btrfs_ino(inode),
3557 offset, 0, 0, len, 0, len, 0,
3561 *last_extent = extent_end;
3564 * Need to let the callers know we dropped the path so they should
3567 if (!ret && need_find_last_extent)
3572 static int extent_cmp(void *priv, struct list_head *a, struct list_head *b)
3574 struct extent_map *em1, *em2;
3576 em1 = list_entry(a, struct extent_map, list);
3577 em2 = list_entry(b, struct extent_map, list);
3579 if (em1->start < em2->start)
3581 else if (em1->start > em2->start)
3586 static int wait_ordered_extents(struct btrfs_trans_handle *trans,
3587 struct inode *inode,
3588 struct btrfs_root *root,
3589 const struct extent_map *em,
3590 const struct list_head *logged_list,
3591 bool *ordered_io_error)
3593 struct btrfs_ordered_extent *ordered;
3594 struct btrfs_root *log = root->log_root;
3595 u64 mod_start = em->mod_start;
3596 u64 mod_len = em->mod_len;
3597 const bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
3600 LIST_HEAD(ordered_sums);
3603 *ordered_io_error = false;
3605 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
3606 em->block_start == EXTENT_MAP_HOLE)
3610 * Wait far any ordered extent that covers our extent map. If it
3611 * finishes without an error, first check and see if our csums are on
3612 * our outstanding ordered extents.
3614 list_for_each_entry(ordered, logged_list, log_list) {
3615 struct btrfs_ordered_sum *sum;
3620 if (ordered->file_offset + ordered->len <= mod_start ||
3621 mod_start + mod_len <= ordered->file_offset)
3624 if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) &&
3625 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3626 !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) {
3627 const u64 start = ordered->file_offset;
3628 const u64 end = ordered->file_offset + ordered->len - 1;
3630 WARN_ON(ordered->inode != inode);
3631 filemap_fdatawrite_range(inode->i_mapping, start, end);
3634 wait_event(ordered->wait,
3635 (test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) ||
3636 test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)));
3638 if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) {
3640 * Clear the AS_EIO/AS_ENOSPC flags from the inode's
3641 * i_mapping flags, so that the next fsync won't get
3642 * an outdated io error too.
3644 btrfs_inode_check_errors(inode);
3645 *ordered_io_error = true;
3649 * We are going to copy all the csums on this ordered extent, so
3650 * go ahead and adjust mod_start and mod_len in case this
3651 * ordered extent has already been logged.
3653 if (ordered->file_offset > mod_start) {
3654 if (ordered->file_offset + ordered->len >=
3655 mod_start + mod_len)
3656 mod_len = ordered->file_offset - mod_start;
3658 * If we have this case
3660 * |--------- logged extent ---------|
3661 * |----- ordered extent ----|
3663 * Just don't mess with mod_start and mod_len, we'll
3664 * just end up logging more csums than we need and it
3668 if (ordered->file_offset + ordered->len <
3669 mod_start + mod_len) {
3670 mod_len = (mod_start + mod_len) -
3671 (ordered->file_offset + ordered->len);
3672 mod_start = ordered->file_offset +
3683 * To keep us from looping for the above case of an ordered
3684 * extent that falls inside of the logged extent.
3686 if (test_and_set_bit(BTRFS_ORDERED_LOGGED_CSUM,
3690 if (ordered->csum_bytes_left) {
3691 btrfs_start_ordered_extent(inode, ordered, 0);
3692 wait_event(ordered->wait,
3693 ordered->csum_bytes_left == 0);
3696 list_for_each_entry(sum, &ordered->list, list) {
3697 ret = btrfs_csum_file_blocks(trans, log, sum);
3703 if (*ordered_io_error || !mod_len || ret || skip_csum)
3706 if (em->compress_type) {
3708 csum_len = max(em->block_len, em->orig_block_len);
3710 csum_offset = mod_start - em->start;
3714 /* block start is already adjusted for the file extent offset. */
3715 ret = btrfs_lookup_csums_range(log->fs_info->csum_root,
3716 em->block_start + csum_offset,
3717 em->block_start + csum_offset +
3718 csum_len - 1, &ordered_sums, 0);
3722 while (!list_empty(&ordered_sums)) {
3723 struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
3724 struct btrfs_ordered_sum,
3727 ret = btrfs_csum_file_blocks(trans, log, sums);
3728 list_del(&sums->list);
3735 static int log_one_extent(struct btrfs_trans_handle *trans,
3736 struct inode *inode, struct btrfs_root *root,
3737 const struct extent_map *em,
3738 struct btrfs_path *path,
3739 const struct list_head *logged_list,
3740 struct btrfs_log_ctx *ctx)
3742 struct btrfs_root *log = root->log_root;
3743 struct btrfs_file_extent_item *fi;
3744 struct extent_buffer *leaf;
3745 struct btrfs_map_token token;
3746 struct btrfs_key key;
3747 u64 extent_offset = em->start - em->orig_start;
3750 int extent_inserted = 0;
3751 bool ordered_io_err = false;
3753 ret = wait_ordered_extents(trans, inode, root, em, logged_list,
3758 if (ordered_io_err) {
3763 btrfs_init_map_token(&token);
3765 ret = __btrfs_drop_extents(trans, log, inode, path, em->start,
3766 em->start + em->len, NULL, 0, 1,
3767 sizeof(*fi), &extent_inserted);
3771 if (!extent_inserted) {
3772 key.objectid = btrfs_ino(inode);
3773 key.type = BTRFS_EXTENT_DATA_KEY;
3774 key.offset = em->start;
3776 ret = btrfs_insert_empty_item(trans, log, path, &key,
3781 leaf = path->nodes[0];
3782 fi = btrfs_item_ptr(leaf, path->slots[0],
3783 struct btrfs_file_extent_item);
3785 btrfs_set_token_file_extent_generation(leaf, fi, trans->transid,
3787 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
3788 btrfs_set_token_file_extent_type(leaf, fi,
3789 BTRFS_FILE_EXTENT_PREALLOC,
3792 btrfs_set_token_file_extent_type(leaf, fi,
3793 BTRFS_FILE_EXTENT_REG,
3796 block_len = max(em->block_len, em->orig_block_len);
3797 if (em->compress_type != BTRFS_COMPRESS_NONE) {
3798 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3801 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3803 } else if (em->block_start < EXTENT_MAP_LAST_BYTE) {
3804 btrfs_set_token_file_extent_disk_bytenr(leaf, fi,
3806 extent_offset, &token);
3807 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, block_len,
3810 btrfs_set_token_file_extent_disk_bytenr(leaf, fi, 0, &token);
3811 btrfs_set_token_file_extent_disk_num_bytes(leaf, fi, 0,
3815 btrfs_set_token_file_extent_offset(leaf, fi, extent_offset, &token);
3816 btrfs_set_token_file_extent_num_bytes(leaf, fi, em->len, &token);
3817 btrfs_set_token_file_extent_ram_bytes(leaf, fi, em->ram_bytes, &token);
3818 btrfs_set_token_file_extent_compression(leaf, fi, em->compress_type,
3820 btrfs_set_token_file_extent_encryption(leaf, fi, 0, &token);
3821 btrfs_set_token_file_extent_other_encoding(leaf, fi, 0, &token);
3822 btrfs_mark_buffer_dirty(leaf);
3824 btrfs_release_path(path);
3829 static int btrfs_log_changed_extents(struct btrfs_trans_handle *trans,
3830 struct btrfs_root *root,
3831 struct inode *inode,
3832 struct btrfs_path *path,
3833 struct list_head *logged_list,
3834 struct btrfs_log_ctx *ctx)
3836 struct extent_map *em, *n;
3837 struct list_head extents;
3838 struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
3843 INIT_LIST_HEAD(&extents);
3845 write_lock(&tree->lock);
3846 test_gen = root->fs_info->last_trans_committed;
3848 list_for_each_entry_safe(em, n, &tree->modified_extents, list) {
3849 list_del_init(&em->list);
3852 * Just an arbitrary number, this can be really CPU intensive
3853 * once we start getting a lot of extents, and really once we
3854 * have a bunch of extents we just want to commit since it will
3857 if (++num > 32768) {
3858 list_del_init(&tree->modified_extents);
3863 if (em->generation <= test_gen)
3865 /* Need a ref to keep it from getting evicted from cache */
3866 atomic_inc(&em->refs);
3867 set_bit(EXTENT_FLAG_LOGGING, &em->flags);
3868 list_add_tail(&em->list, &extents);
3872 list_sort(NULL, &extents, extent_cmp);
3875 while (!list_empty(&extents)) {
3876 em = list_entry(extents.next, struct extent_map, list);
3878 list_del_init(&em->list);
3881 * If we had an error we just need to delete everybody from our
3885 clear_em_logging(tree, em);
3886 free_extent_map(em);
3890 write_unlock(&tree->lock);
3892 ret = log_one_extent(trans, inode, root, em, path, logged_list,
3894 write_lock(&tree->lock);
3895 clear_em_logging(tree, em);
3896 free_extent_map(em);
3898 WARN_ON(!list_empty(&extents));
3899 write_unlock(&tree->lock);
3901 btrfs_release_path(path);
3905 /* log a single inode in the tree log.
3906 * At least one parent directory for this inode must exist in the tree
3907 * or be logged already.
3909 * Any items from this inode changed by the current transaction are copied
3910 * to the log tree. An extra reference is taken on any extents in this
3911 * file, allowing us to avoid a whole pile of corner cases around logging
3912 * blocks that have been removed from the tree.
3914 * See LOG_INODE_ALL and related defines for a description of what inode_only
3917 * This handles both files and directories.
3919 static int btrfs_log_inode(struct btrfs_trans_handle *trans,
3920 struct btrfs_root *root, struct inode *inode,
3924 struct btrfs_log_ctx *ctx)
3926 struct btrfs_path *path;
3927 struct btrfs_path *dst_path;
3928 struct btrfs_key min_key;
3929 struct btrfs_key max_key;
3930 struct btrfs_root *log = root->log_root;
3931 struct extent_buffer *src = NULL;
3932 LIST_HEAD(logged_list);
3933 u64 last_extent = 0;
3937 int ins_start_slot = 0;
3939 bool fast_search = false;
3940 u64 ino = btrfs_ino(inode);
3941 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3943 path = btrfs_alloc_path();
3946 dst_path = btrfs_alloc_path();
3948 btrfs_free_path(path);
3952 min_key.objectid = ino;
3953 min_key.type = BTRFS_INODE_ITEM_KEY;
3956 max_key.objectid = ino;
3959 /* today the code can only do partial logging of directories */
3960 if (S_ISDIR(inode->i_mode) ||
3961 (!test_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3962 &BTRFS_I(inode)->runtime_flags) &&
3963 inode_only == LOG_INODE_EXISTS))
3964 max_key.type = BTRFS_XATTR_ITEM_KEY;
3966 max_key.type = (u8)-1;
3967 max_key.offset = (u64)-1;
3969 /* Only run delayed items if we are a dir or a new file */
3970 if (S_ISDIR(inode->i_mode) ||
3971 BTRFS_I(inode)->generation > root->fs_info->last_trans_committed) {
3972 ret = btrfs_commit_inode_delayed_items(trans, inode);
3974 btrfs_free_path(path);
3975 btrfs_free_path(dst_path);
3980 mutex_lock(&BTRFS_I(inode)->log_mutex);
3982 btrfs_get_logged_extents(inode, &logged_list, start, end);
3985 * a brute force approach to making sure we get the most uptodate
3986 * copies of everything.
3988 if (S_ISDIR(inode->i_mode)) {
3989 int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
3991 if (inode_only == LOG_INODE_EXISTS)
3992 max_key_type = BTRFS_XATTR_ITEM_KEY;
3993 ret = drop_objectid_items(trans, log, path, ino, max_key_type);
3995 if (test_and_clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3996 &BTRFS_I(inode)->runtime_flags)) {
3997 clear_bit(BTRFS_INODE_COPY_EVERYTHING,
3998 &BTRFS_I(inode)->runtime_flags);
3999 ret = btrfs_truncate_inode_items(trans, log,
4001 } else if (test_and_clear_bit(BTRFS_INODE_COPY_EVERYTHING,
4002 &BTRFS_I(inode)->runtime_flags) ||
4003 inode_only == LOG_INODE_EXISTS) {
4004 if (inode_only == LOG_INODE_ALL)
4006 max_key.type = BTRFS_XATTR_ITEM_KEY;
4007 ret = drop_objectid_items(trans, log, path, ino,
4010 if (inode_only == LOG_INODE_ALL)
4012 ret = log_inode_item(trans, log, dst_path, inode);
4028 ret = btrfs_search_forward(root, &min_key,
4029 path, trans->transid);
4033 /* note, ins_nr might be > 0 here, cleanup outside the loop */
4034 if (min_key.objectid != ino)
4036 if (min_key.type > max_key.type)
4039 src = path->nodes[0];
4040 if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
4043 } else if (!ins_nr) {
4044 ins_start_slot = path->slots[0];
4049 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4050 ins_start_slot, ins_nr, inode_only);
4057 btrfs_release_path(path);
4061 ins_start_slot = path->slots[0];
4064 nritems = btrfs_header_nritems(path->nodes[0]);
4066 if (path->slots[0] < nritems) {
4067 btrfs_item_key_to_cpu(path->nodes[0], &min_key,
4072 ret = copy_items(trans, inode, dst_path, path,
4073 &last_extent, ins_start_slot,
4074 ins_nr, inode_only);
4082 btrfs_release_path(path);
4084 if (min_key.offset < (u64)-1) {
4086 } else if (min_key.type < max_key.type) {
4094 ret = copy_items(trans, inode, dst_path, path, &last_extent,
4095 ins_start_slot, ins_nr, inode_only);
4105 btrfs_release_path(path);
4106 btrfs_release_path(dst_path);
4109 * Some ordered extents started by fsync might have completed
4110 * before we collected the ordered extents in logged_list, which
4111 * means they're gone, not in our logged_list nor in the inode's
4112 * ordered tree. We want the application/user space to know an
4113 * error happened while attempting to persist file data so that
4114 * it can take proper action. If such error happened, we leave
4115 * without writing to the log tree and the fsync must report the
4116 * file data write error and not commit the current transaction.
4118 err = btrfs_inode_check_errors(inode);
4123 ret = btrfs_log_changed_extents(trans, root, inode, dst_path,
4129 } else if (inode_only == LOG_INODE_ALL) {
4130 struct extent_map *em, *n;
4132 write_lock(&em_tree->lock);
4134 * We can't just remove every em if we're called for a ranged
4135 * fsync - that is, one that doesn't cover the whole possible
4136 * file range (0 to LLONG_MAX). This is because we can have
4137 * em's that fall outside the range we're logging and therefore
4138 * their ordered operations haven't completed yet
4139 * (btrfs_finish_ordered_io() not invoked yet). This means we
4140 * didn't get their respective file extent item in the fs/subvol
4141 * tree yet, and need to let the next fast fsync (one which
4142 * consults the list of modified extent maps) find the em so
4143 * that it logs a matching file extent item and waits for the
4144 * respective ordered operation to complete (if it's still
4147 * Removing every em outside the range we're logging would make
4148 * the next fast fsync not log their matching file extent items,
4149 * therefore making us lose data after a log replay.
4151 list_for_each_entry_safe(em, n, &em_tree->modified_extents,
4153 const u64 mod_end = em->mod_start + em->mod_len - 1;
4155 if (em->mod_start >= start && mod_end <= end)
4156 list_del_init(&em->list);
4158 write_unlock(&em_tree->lock);
4161 if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
4162 ret = log_directory_changes(trans, root, inode, path, dst_path);
4169 BTRFS_I(inode)->logged_trans = trans->transid;
4170 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
4173 btrfs_put_logged_extents(&logged_list);
4175 btrfs_submit_logged_extents(&logged_list, log);
4176 mutex_unlock(&BTRFS_I(inode)->log_mutex);
4178 btrfs_free_path(path);
4179 btrfs_free_path(dst_path);
4184 * follow the dentry parent pointers up the chain and see if any
4185 * of the directories in it require a full commit before they can
4186 * be logged. Returns zero if nothing special needs to be done or 1 if
4187 * a full commit is required.
4189 static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
4190 struct inode *inode,
4191 struct dentry *parent,
4192 struct super_block *sb,
4196 struct btrfs_root *root;
4197 struct dentry *old_parent = NULL;
4198 struct inode *orig_inode = inode;
4201 * for regular files, if its inode is already on disk, we don't
4202 * have to worry about the parents at all. This is because
4203 * we can use the last_unlink_trans field to record renames
4204 * and other fun in this file.
4206 if (S_ISREG(inode->i_mode) &&
4207 BTRFS_I(inode)->generation <= last_committed &&
4208 BTRFS_I(inode)->last_unlink_trans <= last_committed)
4211 if (!S_ISDIR(inode->i_mode)) {
4212 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4214 inode = parent->d_inode;
4219 * If we are logging a directory then we start with our inode,
4220 * not our parents inode, so we need to skipp setting the
4221 * logged_trans so that further down in the log code we don't
4222 * think this inode has already been logged.
4224 if (inode != orig_inode)
4225 BTRFS_I(inode)->logged_trans = trans->transid;
4228 if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
4229 root = BTRFS_I(inode)->root;
4232 * make sure any commits to the log are forced
4233 * to be full commits
4235 btrfs_set_log_full_commit(root->fs_info, trans);
4240 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4243 if (IS_ROOT(parent))
4246 parent = dget_parent(parent);
4248 old_parent = parent;
4249 inode = parent->d_inode;
4258 * helper function around btrfs_log_inode to make sure newly created
4259 * parent directories also end up in the log. A minimal inode and backref
4260 * only logging is done of any parent directories that are older than
4261 * the last committed transaction
4263 static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
4264 struct btrfs_root *root, struct inode *inode,
4265 struct dentry *parent,
4269 struct btrfs_log_ctx *ctx)
4271 int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
4272 struct super_block *sb;
4273 struct dentry *old_parent = NULL;
4275 u64 last_committed = root->fs_info->last_trans_committed;
4279 if (btrfs_test_opt(root, NOTREELOG)) {
4285 * The prev transaction commit doesn't complete, we need do
4286 * full commit by ourselves.
4288 if (root->fs_info->last_trans_log_full_commit >
4289 root->fs_info->last_trans_committed) {
4294 if (root != BTRFS_I(inode)->root ||
4295 btrfs_root_refs(&root->root_item) == 0) {
4300 ret = check_parent_dirs_for_sync(trans, inode, parent,
4301 sb, last_committed);
4305 if (btrfs_inode_in_log(inode, trans->transid)) {
4306 ret = BTRFS_NO_LOG_SYNC;
4310 ret = start_log_trans(trans, root, ctx);
4314 ret = btrfs_log_inode(trans, root, inode, inode_only, start, end, ctx);
4319 * for regular files, if its inode is already on disk, we don't
4320 * have to worry about the parents at all. This is because
4321 * we can use the last_unlink_trans field to record renames
4322 * and other fun in this file.
4324 if (S_ISREG(inode->i_mode) &&
4325 BTRFS_I(inode)->generation <= last_committed &&
4326 BTRFS_I(inode)->last_unlink_trans <= last_committed) {
4331 inode_only = LOG_INODE_EXISTS;
4333 if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
4336 inode = parent->d_inode;
4337 if (root != BTRFS_I(inode)->root)
4340 if (BTRFS_I(inode)->generation >
4341 root->fs_info->last_trans_committed) {
4342 ret = btrfs_log_inode(trans, root, inode, inode_only,
4347 if (IS_ROOT(parent))
4350 parent = dget_parent(parent);
4352 old_parent = parent;
4358 btrfs_set_log_full_commit(root->fs_info, trans);
4363 btrfs_remove_log_ctx(root, ctx);
4364 btrfs_end_log_trans(root);
4370 * it is not safe to log dentry if the chunk root has added new
4371 * chunks. This returns 0 if the dentry was logged, and 1 otherwise.
4372 * If this returns 1, you must commit the transaction to safely get your
4375 int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
4376 struct btrfs_root *root, struct dentry *dentry,
4379 struct btrfs_log_ctx *ctx)
4381 struct dentry *parent = dget_parent(dentry);
4384 ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent,
4385 start, end, 0, ctx);
4392 * should be called during mount to recover any replay any log trees
4395 int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
4398 struct btrfs_path *path;
4399 struct btrfs_trans_handle *trans;
4400 struct btrfs_key key;
4401 struct btrfs_key found_key;
4402 struct btrfs_key tmp_key;
4403 struct btrfs_root *log;
4404 struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
4405 struct walk_control wc = {
4406 .process_func = process_one_buffer,
4410 path = btrfs_alloc_path();
4414 fs_info->log_root_recovering = 1;
4416 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4417 if (IS_ERR(trans)) {
4418 ret = PTR_ERR(trans);
4425 ret = walk_log_tree(trans, log_root_tree, &wc);
4427 btrfs_error(fs_info, ret, "Failed to pin buffers while "
4428 "recovering log root tree.");
4433 key.objectid = BTRFS_TREE_LOG_OBJECTID;
4434 key.offset = (u64)-1;
4435 key.type = BTRFS_ROOT_ITEM_KEY;
4438 ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
4441 btrfs_error(fs_info, ret,
4442 "Couldn't find tree log root.");
4446 if (path->slots[0] == 0)
4450 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
4452 btrfs_release_path(path);
4453 if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4456 log = btrfs_read_fs_root(log_root_tree, &found_key);
4459 btrfs_error(fs_info, ret,
4460 "Couldn't read tree log root.");
4464 tmp_key.objectid = found_key.offset;
4465 tmp_key.type = BTRFS_ROOT_ITEM_KEY;
4466 tmp_key.offset = (u64)-1;
4468 wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
4469 if (IS_ERR(wc.replay_dest)) {
4470 ret = PTR_ERR(wc.replay_dest);
4471 free_extent_buffer(log->node);
4472 free_extent_buffer(log->commit_root);
4474 btrfs_error(fs_info, ret, "Couldn't read target root "
4475 "for tree log recovery.");
4479 wc.replay_dest->log_root = log;
4480 btrfs_record_root_in_trans(trans, wc.replay_dest);
4481 ret = walk_log_tree(trans, log, &wc);
4483 if (!ret && wc.stage == LOG_WALK_REPLAY_ALL) {
4484 ret = fixup_inode_link_counts(trans, wc.replay_dest,
4488 key.offset = found_key.offset - 1;
4489 wc.replay_dest->log_root = NULL;
4490 free_extent_buffer(log->node);
4491 free_extent_buffer(log->commit_root);
4497 if (found_key.offset == 0)
4500 btrfs_release_path(path);
4502 /* step one is to pin it all, step two is to replay just inodes */
4505 wc.process_func = replay_one_buffer;
4506 wc.stage = LOG_WALK_REPLAY_INODES;
4509 /* step three is to replay everything */
4510 if (wc.stage < LOG_WALK_REPLAY_ALL) {
4515 btrfs_free_path(path);
4517 /* step 4: commit the transaction, which also unpins the blocks */
4518 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
4522 free_extent_buffer(log_root_tree->node);
4523 log_root_tree->log_root = NULL;
4524 fs_info->log_root_recovering = 0;
4525 kfree(log_root_tree);
4530 btrfs_end_transaction(wc.trans, fs_info->tree_root);
4531 btrfs_free_path(path);
4536 * there are some corner cases where we want to force a full
4537 * commit instead of allowing a directory to be logged.
4539 * They revolve around files there were unlinked from the directory, and
4540 * this function updates the parent directory so that a full commit is
4541 * properly done if it is fsync'd later after the unlinks are done.
4543 void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
4544 struct inode *dir, struct inode *inode,
4548 * when we're logging a file, if it hasn't been renamed
4549 * or unlinked, and its inode is fully committed on disk,
4550 * we don't have to worry about walking up the directory chain
4551 * to log its parents.
4553 * So, we use the last_unlink_trans field to put this transid
4554 * into the file. When the file is logged we check it and
4555 * don't log the parents if the file is fully on disk.
4557 if (S_ISREG(inode->i_mode))
4558 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4561 * if this directory was already logged any new
4562 * names for this file/dir will get recorded
4565 if (BTRFS_I(dir)->logged_trans == trans->transid)
4569 * if the inode we're about to unlink was logged,
4570 * the log will be properly updated for any new names
4572 if (BTRFS_I(inode)->logged_trans == trans->transid)
4576 * when renaming files across directories, if the directory
4577 * there we're unlinking from gets fsync'd later on, there's
4578 * no way to find the destination directory later and fsync it
4579 * properly. So, we have to be conservative and force commits
4580 * so the new name gets discovered.
4585 /* we can safely do the unlink without any special recording */
4589 BTRFS_I(dir)->last_unlink_trans = trans->transid;
4593 * Call this after adding a new name for a file and it will properly
4594 * update the log to reflect the new name.
4596 * It will return zero if all goes well, and it will return 1 if a
4597 * full transaction commit is required.
4599 int btrfs_log_new_name(struct btrfs_trans_handle *trans,
4600 struct inode *inode, struct inode *old_dir,
4601 struct dentry *parent)
4603 struct btrfs_root * root = BTRFS_I(inode)->root;
4606 * this will force the logging code to walk the dentry chain
4609 if (S_ISREG(inode->i_mode))
4610 BTRFS_I(inode)->last_unlink_trans = trans->transid;
4613 * if this inode hasn't been logged and directory we're renaming it
4614 * from hasn't been logged, we don't need to log it
4616 if (BTRFS_I(inode)->logged_trans <=
4617 root->fs_info->last_trans_committed &&
4618 (!old_dir || BTRFS_I(old_dir)->logged_trans <=
4619 root->fs_info->last_trans_committed))
4622 return btrfs_log_inode_parent(trans, root, inode, parent, 0,
4623 LLONG_MAX, 1, NULL);