• Home
  • History
  • Annotate
  • Line#
  • Navigate
  • Raw
  • Download
  • only in /asuswrt-rt-n18u-9.0.0.4.380.2695/release/src-rt-6.x.4708/linux/linux-2.6.36/fs/btrfs/
1/*
2 * Copyright (C) 2008 Oracle.  All rights reserved.
3 *
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.
7 *
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.
12 *
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.
17 */
18
19#include <linux/sched.h>
20#include <linux/slab.h>
21#include "ctree.h"
22#include "transaction.h"
23#include "disk-io.h"
24#include "locking.h"
25#include "print-tree.h"
26#include "compat.h"
27#include "tree-log.h"
28
29/* magic values for the inode_only field in btrfs_log_inode:
30 *
31 * LOG_INODE_ALL means to log everything
32 * LOG_INODE_EXISTS means to log just enough to recreate the inode
33 * during log replay
34 */
35#define LOG_INODE_ALL 0
36#define LOG_INODE_EXISTS 1
37
38/*
39 * directory trouble cases
40 *
41 * 1) on rename or unlink, if the inode being unlinked isn't in the fsync
42 * log, we must force a full commit before doing an fsync of the directory
43 * where the unlink was done.
44 * ---> record transid of last unlink/rename per directory
45 *
46 * mkdir foo/some_dir
47 * normal commit
48 * rename foo/some_dir foo2/some_dir
49 * mkdir foo/some_dir
50 * fsync foo/some_dir/some_file
51 *
52 * The fsync above will unlink the original some_dir without recording
53 * it in its new location (foo2).  After a crash, some_dir will be gone
54 * unless the fsync of some_file forces a full commit
55 *
56 * 2) we must log any new names for any file or dir that is in the fsync
57 * log. ---> check inode while renaming/linking.
58 *
59 * 2a) we must log any new names for any file or dir during rename
60 * when the directory they are being removed from was logged.
61 * ---> check inode and old parent dir during rename
62 *
63 *  2a is actually the more important variant.  With the extra logging
64 *  a crash might unlink the old name without recreating the new one
65 *
66 * 3) after a crash, we must go through any directories with a link count
67 * of zero and redo the rm -rf
68 *
69 * mkdir f1/foo
70 * normal commit
71 * rm -rf f1/foo
72 * fsync(f1)
73 *
74 * The directory f1 was fully removed from the FS, but fsync was never
75 * called on f1, only its parent dir.  After a crash the rm -rf must
76 * be replayed.  This must be able to recurse down the entire
77 * directory tree.  The inode link count fixup code takes care of the
78 * ugly details.
79 */
80
81/*
82 * stages for the tree walking.  The first
83 * stage (0) is to only pin down the blocks we find
84 * the second stage (1) is to make sure that all the inodes
85 * we find in the log are created in the subvolume.
86 *
87 * The last stage is to deal with directories and links and extents
88 * and all the other fun semantics
89 */
90#define LOG_WALK_PIN_ONLY 0
91#define LOG_WALK_REPLAY_INODES 1
92#define LOG_WALK_REPLAY_ALL 2
93
94static int btrfs_log_inode(struct btrfs_trans_handle *trans,
95			     struct btrfs_root *root, struct inode *inode,
96			     int inode_only);
97static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
98			     struct btrfs_root *root,
99			     struct btrfs_path *path, u64 objectid);
100static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
101				       struct btrfs_root *root,
102				       struct btrfs_root *log,
103				       struct btrfs_path *path,
104				       u64 dirid, int del_all);
105
106/*
107 * tree logging is a special write ahead log used to make sure that
108 * fsyncs and O_SYNCs can happen without doing full tree commits.
109 *
110 * Full tree commits are expensive because they require commonly
111 * modified blocks to be recowed, creating many dirty pages in the
112 * extent tree an 4x-6x higher write load than ext3.
113 *
114 * Instead of doing a tree commit on every fsync, we use the
115 * key ranges and transaction ids to find items for a given file or directory
116 * that have changed in this transaction.  Those items are copied into
117 * a special tree (one per subvolume root), that tree is written to disk
118 * and then the fsync is considered complete.
119 *
120 * After a crash, items are copied out of the log-tree back into the
121 * subvolume tree.  Any file data extents found are recorded in the extent
122 * allocation tree, and the log-tree freed.
123 *
124 * The log tree is read three times, once to pin down all the extents it is
125 * using in ram and once, once to create all the inodes logged in the tree
126 * and once to do all the other items.
127 */
128
129/*
130 * start a sub transaction and setup the log tree
131 * this increments the log tree writer count to make the people
132 * syncing the tree wait for us to finish
133 */
134static int start_log_trans(struct btrfs_trans_handle *trans,
135			   struct btrfs_root *root)
136{
137	int ret;
138	int err = 0;
139
140	mutex_lock(&root->log_mutex);
141	if (root->log_root) {
142		if (!root->log_start_pid) {
143			root->log_start_pid = current->pid;
144			root->log_multiple_pids = false;
145		} else if (root->log_start_pid != current->pid) {
146			root->log_multiple_pids = true;
147		}
148
149		root->log_batch++;
150		atomic_inc(&root->log_writers);
151		mutex_unlock(&root->log_mutex);
152		return 0;
153	}
154	root->log_multiple_pids = false;
155	root->log_start_pid = current->pid;
156	mutex_lock(&root->fs_info->tree_log_mutex);
157	if (!root->fs_info->log_root_tree) {
158		ret = btrfs_init_log_root_tree(trans, root->fs_info);
159		if (ret)
160			err = ret;
161	}
162	if (err == 0 && !root->log_root) {
163		ret = btrfs_add_log_tree(trans, root);
164		if (ret)
165			err = ret;
166	}
167	mutex_unlock(&root->fs_info->tree_log_mutex);
168	root->log_batch++;
169	atomic_inc(&root->log_writers);
170	mutex_unlock(&root->log_mutex);
171	return err;
172}
173
174/*
175 * returns 0 if there was a log transaction running and we were able
176 * to join, or returns -ENOENT if there were not transactions
177 * in progress
178 */
179static int join_running_log_trans(struct btrfs_root *root)
180{
181	int ret = -ENOENT;
182
183	smp_mb();
184	if (!root->log_root)
185		return -ENOENT;
186
187	mutex_lock(&root->log_mutex);
188	if (root->log_root) {
189		ret = 0;
190		atomic_inc(&root->log_writers);
191	}
192	mutex_unlock(&root->log_mutex);
193	return ret;
194}
195
196/*
197 * This either makes the current running log transaction wait
198 * until you call btrfs_end_log_trans() or it makes any future
199 * log transactions wait until you call btrfs_end_log_trans()
200 */
201int btrfs_pin_log_trans(struct btrfs_root *root)
202{
203	int ret = -ENOENT;
204
205	mutex_lock(&root->log_mutex);
206	atomic_inc(&root->log_writers);
207	mutex_unlock(&root->log_mutex);
208	return ret;
209}
210
211/*
212 * indicate we're done making changes to the log tree
213 * and wake up anyone waiting to do a sync
214 */
215int btrfs_end_log_trans(struct btrfs_root *root)
216{
217	if (atomic_dec_and_test(&root->log_writers)) {
218		smp_mb();
219		if (waitqueue_active(&root->log_writer_wait))
220			wake_up(&root->log_writer_wait);
221	}
222	return 0;
223}
224
225
226/*
227 * the walk control struct is used to pass state down the chain when
228 * processing the log tree.  The stage field tells us which part
229 * of the log tree processing we are currently doing.  The others
230 * are state fields used for that specific part
231 */
232struct walk_control {
233	/* should we free the extent on disk when done?  This is used
234	 * at transaction commit time while freeing a log tree
235	 */
236	int free;
237
238	/* should we write out the extent buffer?  This is used
239	 * while flushing the log tree to disk during a sync
240	 */
241	int write;
242
243	/* should we wait for the extent buffer io to finish?  Also used
244	 * while flushing the log tree to disk for a sync
245	 */
246	int wait;
247
248	/* pin only walk, we record which extents on disk belong to the
249	 * log trees
250	 */
251	int pin;
252
253	/* what stage of the replay code we're currently in */
254	int stage;
255
256	/* the root we are currently replaying */
257	struct btrfs_root *replay_dest;
258
259	/* the trans handle for the current replay */
260	struct btrfs_trans_handle *trans;
261
262	/* the function that gets used to process blocks we find in the
263	 * tree.  Note the extent_buffer might not be up to date when it is
264	 * passed in, and it must be checked or read if you need the data
265	 * inside it
266	 */
267	int (*process_func)(struct btrfs_root *log, struct extent_buffer *eb,
268			    struct walk_control *wc, u64 gen);
269};
270
271/*
272 * process_func used to pin down extents, write them or wait on them
273 */
274static int process_one_buffer(struct btrfs_root *log,
275			      struct extent_buffer *eb,
276			      struct walk_control *wc, u64 gen)
277{
278	if (wc->pin)
279		btrfs_pin_extent(log->fs_info->extent_root,
280				 eb->start, eb->len, 0);
281
282	if (btrfs_buffer_uptodate(eb, gen)) {
283		if (wc->write)
284			btrfs_write_tree_block(eb);
285		if (wc->wait)
286			btrfs_wait_tree_block_writeback(eb);
287	}
288	return 0;
289}
290
291/*
292 * Item overwrite used by replay and tree logging.  eb, slot and key all refer
293 * to the src data we are copying out.
294 *
295 * root is the tree we are copying into, and path is a scratch
296 * path for use in this function (it should be released on entry and
297 * will be released on exit).
298 *
299 * If the key is already in the destination tree the existing item is
300 * overwritten.  If the existing item isn't big enough, it is extended.
301 * If it is too large, it is truncated.
302 *
303 * If the key isn't in the destination yet, a new item is inserted.
304 */
305static noinline int overwrite_item(struct btrfs_trans_handle *trans,
306				   struct btrfs_root *root,
307				   struct btrfs_path *path,
308				   struct extent_buffer *eb, int slot,
309				   struct btrfs_key *key)
310{
311	int ret;
312	u32 item_size;
313	u64 saved_i_size = 0;
314	int save_old_i_size = 0;
315	unsigned long src_ptr;
316	unsigned long dst_ptr;
317	int overwrite_root = 0;
318
319	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
320		overwrite_root = 1;
321
322	item_size = btrfs_item_size_nr(eb, slot);
323	src_ptr = btrfs_item_ptr_offset(eb, slot);
324
325	/* look for the key in the destination tree */
326	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
327	if (ret == 0) {
328		char *src_copy;
329		char *dst_copy;
330		u32 dst_size = btrfs_item_size_nr(path->nodes[0],
331						  path->slots[0]);
332		if (dst_size != item_size)
333			goto insert;
334
335		if (item_size == 0) {
336			btrfs_release_path(root, path);
337			return 0;
338		}
339		dst_copy = kmalloc(item_size, GFP_NOFS);
340		src_copy = kmalloc(item_size, GFP_NOFS);
341
342		read_extent_buffer(eb, src_copy, src_ptr, item_size);
343
344		dst_ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
345		read_extent_buffer(path->nodes[0], dst_copy, dst_ptr,
346				   item_size);
347		ret = memcmp(dst_copy, src_copy, item_size);
348
349		kfree(dst_copy);
350		kfree(src_copy);
351		/*
352		 * they have the same contents, just return, this saves
353		 * us from cowing blocks in the destination tree and doing
354		 * extra writes that may not have been done by a previous
355		 * sync
356		 */
357		if (ret == 0) {
358			btrfs_release_path(root, path);
359			return 0;
360		}
361
362	}
363insert:
364	btrfs_release_path(root, path);
365	/* try to insert the key into the destination tree */
366	ret = btrfs_insert_empty_item(trans, root, path,
367				      key, item_size);
368
369	/* make sure any existing item is the correct size */
370	if (ret == -EEXIST) {
371		u32 found_size;
372		found_size = btrfs_item_size_nr(path->nodes[0],
373						path->slots[0]);
374		if (found_size > item_size) {
375			btrfs_truncate_item(trans, root, path, item_size, 1);
376		} else if (found_size < item_size) {
377			ret = btrfs_extend_item(trans, root, path,
378						item_size - found_size);
379			BUG_ON(ret);
380		}
381	} else if (ret) {
382		return ret;
383	}
384	dst_ptr = btrfs_item_ptr_offset(path->nodes[0],
385					path->slots[0]);
386
387	/* don't overwrite an existing inode if the generation number
388	 * was logged as zero.  This is done when the tree logging code
389	 * is just logging an inode to make sure it exists after recovery.
390	 *
391	 * Also, don't overwrite i_size on directories during replay.
392	 * log replay inserts and removes directory items based on the
393	 * state of the tree found in the subvolume, and i_size is modified
394	 * as it goes
395	 */
396	if (key->type == BTRFS_INODE_ITEM_KEY && ret == -EEXIST) {
397		struct btrfs_inode_item *src_item;
398		struct btrfs_inode_item *dst_item;
399
400		src_item = (struct btrfs_inode_item *)src_ptr;
401		dst_item = (struct btrfs_inode_item *)dst_ptr;
402
403		if (btrfs_inode_generation(eb, src_item) == 0)
404			goto no_copy;
405
406		if (overwrite_root &&
407		    S_ISDIR(btrfs_inode_mode(eb, src_item)) &&
408		    S_ISDIR(btrfs_inode_mode(path->nodes[0], dst_item))) {
409			save_old_i_size = 1;
410			saved_i_size = btrfs_inode_size(path->nodes[0],
411							dst_item);
412		}
413	}
414
415	copy_extent_buffer(path->nodes[0], eb, dst_ptr,
416			   src_ptr, item_size);
417
418	if (save_old_i_size) {
419		struct btrfs_inode_item *dst_item;
420		dst_item = (struct btrfs_inode_item *)dst_ptr;
421		btrfs_set_inode_size(path->nodes[0], dst_item, saved_i_size);
422	}
423
424	/* make sure the generation is filled in */
425	if (key->type == BTRFS_INODE_ITEM_KEY) {
426		struct btrfs_inode_item *dst_item;
427		dst_item = (struct btrfs_inode_item *)dst_ptr;
428		if (btrfs_inode_generation(path->nodes[0], dst_item) == 0) {
429			btrfs_set_inode_generation(path->nodes[0], dst_item,
430						   trans->transid);
431		}
432	}
433no_copy:
434	btrfs_mark_buffer_dirty(path->nodes[0]);
435	btrfs_release_path(root, path);
436	return 0;
437}
438
439/*
440 * simple helper to read an inode off the disk from a given root
441 * This can only be called for subvolume roots and not for the log
442 */
443static noinline struct inode *read_one_inode(struct btrfs_root *root,
444					     u64 objectid)
445{
446	struct btrfs_key key;
447	struct inode *inode;
448
449	key.objectid = objectid;
450	key.type = BTRFS_INODE_ITEM_KEY;
451	key.offset = 0;
452	inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
453	if (IS_ERR(inode)) {
454		inode = NULL;
455	} else if (is_bad_inode(inode)) {
456		iput(inode);
457		inode = NULL;
458	}
459	return inode;
460}
461
462/* replays a single extent in 'eb' at 'slot' with 'key' into the
463 * subvolume 'root'.  path is released on entry and should be released
464 * on exit.
465 *
466 * extents in the log tree have not been allocated out of the extent
467 * tree yet.  So, this completes the allocation, taking a reference
468 * as required if the extent already exists or creating a new extent
469 * if it isn't in the extent allocation tree yet.
470 *
471 * The extent is inserted into the file, dropping any existing extents
472 * from the file that overlap the new one.
473 */
474static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
475				      struct btrfs_root *root,
476				      struct btrfs_path *path,
477				      struct extent_buffer *eb, int slot,
478				      struct btrfs_key *key)
479{
480	int found_type;
481	u64 mask = root->sectorsize - 1;
482	u64 extent_end;
483	u64 alloc_hint;
484	u64 start = key->offset;
485	u64 saved_nbytes;
486	struct btrfs_file_extent_item *item;
487	struct inode *inode = NULL;
488	unsigned long size;
489	int ret = 0;
490
491	item = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
492	found_type = btrfs_file_extent_type(eb, item);
493
494	if (found_type == BTRFS_FILE_EXTENT_REG ||
495	    found_type == BTRFS_FILE_EXTENT_PREALLOC)
496		extent_end = start + btrfs_file_extent_num_bytes(eb, item);
497	else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
498		size = btrfs_file_extent_inline_len(eb, item);
499		extent_end = (start + size + mask) & ~mask;
500	} else {
501		ret = 0;
502		goto out;
503	}
504
505	inode = read_one_inode(root, key->objectid);
506	if (!inode) {
507		ret = -EIO;
508		goto out;
509	}
510
511	/*
512	 * first check to see if we already have this extent in the
513	 * file.  This must be done before the btrfs_drop_extents run
514	 * so we don't try to drop this extent.
515	 */
516	ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
517				       start, 0);
518
519	if (ret == 0 &&
520	    (found_type == BTRFS_FILE_EXTENT_REG ||
521	     found_type == BTRFS_FILE_EXTENT_PREALLOC)) {
522		struct btrfs_file_extent_item cmp1;
523		struct btrfs_file_extent_item cmp2;
524		struct btrfs_file_extent_item *existing;
525		struct extent_buffer *leaf;
526
527		leaf = path->nodes[0];
528		existing = btrfs_item_ptr(leaf, path->slots[0],
529					  struct btrfs_file_extent_item);
530
531		read_extent_buffer(eb, &cmp1, (unsigned long)item,
532				   sizeof(cmp1));
533		read_extent_buffer(leaf, &cmp2, (unsigned long)existing,
534				   sizeof(cmp2));
535
536		/*
537		 * we already have a pointer to this exact extent,
538		 * we don't have to do anything
539		 */
540		if (memcmp(&cmp1, &cmp2, sizeof(cmp1)) == 0) {
541			btrfs_release_path(root, path);
542			goto out;
543		}
544	}
545	btrfs_release_path(root, path);
546
547	saved_nbytes = inode_get_bytes(inode);
548	/* drop any overlapping extents */
549	ret = btrfs_drop_extents(trans, inode, start, extent_end,
550				 &alloc_hint, 1);
551	BUG_ON(ret);
552
553	if (found_type == BTRFS_FILE_EXTENT_REG ||
554	    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
555		u64 offset;
556		unsigned long dest_offset;
557		struct btrfs_key ins;
558
559		ret = btrfs_insert_empty_item(trans, root, path, key,
560					      sizeof(*item));
561		BUG_ON(ret);
562		dest_offset = btrfs_item_ptr_offset(path->nodes[0],
563						    path->slots[0]);
564		copy_extent_buffer(path->nodes[0], eb, dest_offset,
565				(unsigned long)item,  sizeof(*item));
566
567		ins.objectid = btrfs_file_extent_disk_bytenr(eb, item);
568		ins.offset = btrfs_file_extent_disk_num_bytes(eb, item);
569		ins.type = BTRFS_EXTENT_ITEM_KEY;
570		offset = key->offset - btrfs_file_extent_offset(eb, item);
571
572		if (ins.objectid > 0) {
573			u64 csum_start;
574			u64 csum_end;
575			LIST_HEAD(ordered_sums);
576			/*
577			 * is this extent already allocated in the extent
578			 * allocation tree?  If so, just add a reference
579			 */
580			ret = btrfs_lookup_extent(root, ins.objectid,
581						ins.offset);
582			if (ret == 0) {
583				ret = btrfs_inc_extent_ref(trans, root,
584						ins.objectid, ins.offset,
585						0, root->root_key.objectid,
586						key->objectid, offset);
587			} else {
588				/*
589				 * insert the extent pointer in the extent
590				 * allocation tree
591				 */
592				ret = btrfs_alloc_logged_file_extent(trans,
593						root, root->root_key.objectid,
594						key->objectid, offset, &ins);
595				BUG_ON(ret);
596			}
597			btrfs_release_path(root, path);
598
599			if (btrfs_file_extent_compression(eb, item)) {
600				csum_start = ins.objectid;
601				csum_end = csum_start + ins.offset;
602			} else {
603				csum_start = ins.objectid +
604					btrfs_file_extent_offset(eb, item);
605				csum_end = csum_start +
606					btrfs_file_extent_num_bytes(eb, item);
607			}
608
609			ret = btrfs_lookup_csums_range(root->log_root,
610						csum_start, csum_end - 1,
611						&ordered_sums);
612			BUG_ON(ret);
613			while (!list_empty(&ordered_sums)) {
614				struct btrfs_ordered_sum *sums;
615				sums = list_entry(ordered_sums.next,
616						struct btrfs_ordered_sum,
617						list);
618				ret = btrfs_csum_file_blocks(trans,
619						root->fs_info->csum_root,
620						sums);
621				BUG_ON(ret);
622				list_del(&sums->list);
623				kfree(sums);
624			}
625		} else {
626			btrfs_release_path(root, path);
627		}
628	} else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
629		/* inline extents are easy, we just overwrite them */
630		ret = overwrite_item(trans, root, path, eb, slot, key);
631		BUG_ON(ret);
632	}
633
634	inode_set_bytes(inode, saved_nbytes);
635	btrfs_update_inode(trans, root, inode);
636out:
637	if (inode)
638		iput(inode);
639	return ret;
640}
641
642/*
643 * when cleaning up conflicts between the directory names in the
644 * subvolume, directory names in the log and directory names in the
645 * inode back references, we may have to unlink inodes from directories.
646 *
647 * This is a helper function to do the unlink of a specific directory
648 * item
649 */
650static noinline int drop_one_dir_item(struct btrfs_trans_handle *trans,
651				      struct btrfs_root *root,
652				      struct btrfs_path *path,
653				      struct inode *dir,
654				      struct btrfs_dir_item *di)
655{
656	struct inode *inode;
657	char *name;
658	int name_len;
659	struct extent_buffer *leaf;
660	struct btrfs_key location;
661	int ret;
662
663	leaf = path->nodes[0];
664
665	btrfs_dir_item_key_to_cpu(leaf, di, &location);
666	name_len = btrfs_dir_name_len(leaf, di);
667	name = kmalloc(name_len, GFP_NOFS);
668	read_extent_buffer(leaf, name, (unsigned long)(di + 1), name_len);
669	btrfs_release_path(root, path);
670
671	inode = read_one_inode(root, location.objectid);
672	BUG_ON(!inode);
673
674	ret = link_to_fixup_dir(trans, root, path, location.objectid);
675	BUG_ON(ret);
676
677	ret = btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
678	BUG_ON(ret);
679	kfree(name);
680
681	iput(inode);
682	return ret;
683}
684
685/*
686 * helper function to see if a given name and sequence number found
687 * in an inode back reference are already in a directory and correctly
688 * point to this inode
689 */
690static noinline int inode_in_dir(struct btrfs_root *root,
691				 struct btrfs_path *path,
692				 u64 dirid, u64 objectid, u64 index,
693				 const char *name, int name_len)
694{
695	struct btrfs_dir_item *di;
696	struct btrfs_key location;
697	int match = 0;
698
699	di = btrfs_lookup_dir_index_item(NULL, root, path, dirid,
700					 index, name, name_len, 0);
701	if (di && !IS_ERR(di)) {
702		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
703		if (location.objectid != objectid)
704			goto out;
705	} else
706		goto out;
707	btrfs_release_path(root, path);
708
709	di = btrfs_lookup_dir_item(NULL, root, path, dirid, name, name_len, 0);
710	if (di && !IS_ERR(di)) {
711		btrfs_dir_item_key_to_cpu(path->nodes[0], di, &location);
712		if (location.objectid != objectid)
713			goto out;
714	} else
715		goto out;
716	match = 1;
717out:
718	btrfs_release_path(root, path);
719	return match;
720}
721
722/*
723 * helper function to check a log tree for a named back reference in
724 * an inode.  This is used to decide if a back reference that is
725 * found in the subvolume conflicts with what we find in the log.
726 *
727 * inode backreferences may have multiple refs in a single item,
728 * during replay we process one reference at a time, and we don't
729 * want to delete valid links to a file from the subvolume if that
730 * link is also in the log.
731 */
732static noinline int backref_in_log(struct btrfs_root *log,
733				   struct btrfs_key *key,
734				   char *name, int namelen)
735{
736	struct btrfs_path *path;
737	struct btrfs_inode_ref *ref;
738	unsigned long ptr;
739	unsigned long ptr_end;
740	unsigned long name_ptr;
741	int found_name_len;
742	int item_size;
743	int ret;
744	int match = 0;
745
746	path = btrfs_alloc_path();
747	ret = btrfs_search_slot(NULL, log, key, path, 0, 0);
748	if (ret != 0)
749		goto out;
750
751	item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
752	ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
753	ptr_end = ptr + item_size;
754	while (ptr < ptr_end) {
755		ref = (struct btrfs_inode_ref *)ptr;
756		found_name_len = btrfs_inode_ref_name_len(path->nodes[0], ref);
757		if (found_name_len == namelen) {
758			name_ptr = (unsigned long)(ref + 1);
759			ret = memcmp_extent_buffer(path->nodes[0], name,
760						   name_ptr, namelen);
761			if (ret == 0) {
762				match = 1;
763				goto out;
764			}
765		}
766		ptr = (unsigned long)(ref + 1) + found_name_len;
767	}
768out:
769	btrfs_free_path(path);
770	return match;
771}
772
773
774/*
775 * replay one inode back reference item found in the log tree.
776 * eb, slot and key refer to the buffer and key found in the log tree.
777 * root is the destination we are replaying into, and path is for temp
778 * use by this function.  (it should be released on return).
779 */
780static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
781				  struct btrfs_root *root,
782				  struct btrfs_root *log,
783				  struct btrfs_path *path,
784				  struct extent_buffer *eb, int slot,
785				  struct btrfs_key *key)
786{
787	struct inode *dir;
788	int ret;
789	struct btrfs_key location;
790	struct btrfs_inode_ref *ref;
791	struct btrfs_dir_item *di;
792	struct inode *inode;
793	char *name;
794	int namelen;
795	unsigned long ref_ptr;
796	unsigned long ref_end;
797
798	location.objectid = key->objectid;
799	location.type = BTRFS_INODE_ITEM_KEY;
800	location.offset = 0;
801
802	/*
803	 * it is possible that we didn't log all the parent directories
804	 * for a given inode.  If we don't find the dir, just don't
805	 * copy the back ref in.  The link count fixup code will take
806	 * care of the rest
807	 */
808	dir = read_one_inode(root, key->offset);
809	if (!dir)
810		return -ENOENT;
811
812	inode = read_one_inode(root, key->objectid);
813	BUG_ON(!inode);
814
815	ref_ptr = btrfs_item_ptr_offset(eb, slot);
816	ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
817
818again:
819	ref = (struct btrfs_inode_ref *)ref_ptr;
820
821	namelen = btrfs_inode_ref_name_len(eb, ref);
822	name = kmalloc(namelen, GFP_NOFS);
823	BUG_ON(!name);
824
825	read_extent_buffer(eb, name, (unsigned long)(ref + 1), namelen);
826
827	/* if we already have a perfect match, we're done */
828	if (inode_in_dir(root, path, dir->i_ino, inode->i_ino,
829			 btrfs_inode_ref_index(eb, ref),
830			 name, namelen)) {
831		goto out;
832	}
833
834	/*
835	 * look for a conflicting back reference in the metadata.
836	 * if we find one we have to unlink that name of the file
837	 * before we add our new link.  Later on, we overwrite any
838	 * existing back reference, and we don't want to create
839	 * dangling pointers in the directory.
840	 */
841conflict_again:
842	ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
843	if (ret == 0) {
844		char *victim_name;
845		int victim_name_len;
846		struct btrfs_inode_ref *victim_ref;
847		unsigned long ptr;
848		unsigned long ptr_end;
849		struct extent_buffer *leaf = path->nodes[0];
850
851		/* are we trying to overwrite a back ref for the root directory
852		 * if so, just jump out, we're done
853		 */
854		if (key->objectid == key->offset)
855			goto out_nowrite;
856
857		/* check all the names in this back reference to see
858		 * if they are in the log.  if so, we allow them to stay
859		 * otherwise they must be unlinked as a conflict
860		 */
861		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
862		ptr_end = ptr + btrfs_item_size_nr(leaf, path->slots[0]);
863		while (ptr < ptr_end) {
864			victim_ref = (struct btrfs_inode_ref *)ptr;
865			victim_name_len = btrfs_inode_ref_name_len(leaf,
866								   victim_ref);
867			victim_name = kmalloc(victim_name_len, GFP_NOFS);
868			BUG_ON(!victim_name);
869
870			read_extent_buffer(leaf, victim_name,
871					   (unsigned long)(victim_ref + 1),
872					   victim_name_len);
873
874			if (!backref_in_log(log, key, victim_name,
875					    victim_name_len)) {
876				btrfs_inc_nlink(inode);
877				btrfs_release_path(root, path);
878
879				ret = btrfs_unlink_inode(trans, root, dir,
880							 inode, victim_name,
881							 victim_name_len);
882				kfree(victim_name);
883				btrfs_release_path(root, path);
884				goto conflict_again;
885			}
886			kfree(victim_name);
887			ptr = (unsigned long)(victim_ref + 1) + victim_name_len;
888		}
889		BUG_ON(ret);
890	}
891	btrfs_release_path(root, path);
892
893	/* look for a conflicting sequence number */
894	di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
895					 btrfs_inode_ref_index(eb, ref),
896					 name, namelen, 0);
897	if (di && !IS_ERR(di)) {
898		ret = drop_one_dir_item(trans, root, path, dir, di);
899		BUG_ON(ret);
900	}
901	btrfs_release_path(root, path);
902
903
904	/* look for a conflicting name */
905	di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
906				   name, namelen, 0);
907	if (di && !IS_ERR(di)) {
908		ret = drop_one_dir_item(trans, root, path, dir, di);
909		BUG_ON(ret);
910	}
911	btrfs_release_path(root, path);
912
913	/* insert our name */
914	ret = btrfs_add_link(trans, dir, inode, name, namelen, 0,
915			     btrfs_inode_ref_index(eb, ref));
916	BUG_ON(ret);
917
918	btrfs_update_inode(trans, root, inode);
919
920out:
921	ref_ptr = (unsigned long)(ref + 1) + namelen;
922	kfree(name);
923	if (ref_ptr < ref_end)
924		goto again;
925
926	/* finally write the back reference in the inode */
927	ret = overwrite_item(trans, root, path, eb, slot, key);
928	BUG_ON(ret);
929
930out_nowrite:
931	btrfs_release_path(root, path);
932	iput(dir);
933	iput(inode);
934	return 0;
935}
936
937static int insert_orphan_item(struct btrfs_trans_handle *trans,
938			      struct btrfs_root *root, u64 offset)
939{
940	int ret;
941	ret = btrfs_find_orphan_item(root, offset);
942	if (ret > 0)
943		ret = btrfs_insert_orphan_item(trans, root, offset);
944	return ret;
945}
946
947
948/*
949 * There are a few corners where the link count of the file can't
950 * be properly maintained during replay.  So, instead of adding
951 * lots of complexity to the log code, we just scan the backrefs
952 * for any file that has been through replay.
953 *
954 * The scan will update the link count on the inode to reflect the
955 * number of back refs found.  If it goes down to zero, the iput
956 * will free the inode.
957 */
958static noinline int fixup_inode_link_count(struct btrfs_trans_handle *trans,
959					   struct btrfs_root *root,
960					   struct inode *inode)
961{
962	struct btrfs_path *path;
963	int ret;
964	struct btrfs_key key;
965	u64 nlink = 0;
966	unsigned long ptr;
967	unsigned long ptr_end;
968	int name_len;
969
970	key.objectid = inode->i_ino;
971	key.type = BTRFS_INODE_REF_KEY;
972	key.offset = (u64)-1;
973
974	path = btrfs_alloc_path();
975
976	while (1) {
977		ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
978		if (ret < 0)
979			break;
980		if (ret > 0) {
981			if (path->slots[0] == 0)
982				break;
983			path->slots[0]--;
984		}
985		btrfs_item_key_to_cpu(path->nodes[0], &key,
986				      path->slots[0]);
987		if (key.objectid != inode->i_ino ||
988		    key.type != BTRFS_INODE_REF_KEY)
989			break;
990		ptr = btrfs_item_ptr_offset(path->nodes[0], path->slots[0]);
991		ptr_end = ptr + btrfs_item_size_nr(path->nodes[0],
992						   path->slots[0]);
993		while (ptr < ptr_end) {
994			struct btrfs_inode_ref *ref;
995
996			ref = (struct btrfs_inode_ref *)ptr;
997			name_len = btrfs_inode_ref_name_len(path->nodes[0],
998							    ref);
999			ptr = (unsigned long)(ref + 1) + name_len;
1000			nlink++;
1001		}
1002
1003		if (key.offset == 0)
1004			break;
1005		key.offset--;
1006		btrfs_release_path(root, path);
1007	}
1008	btrfs_release_path(root, path);
1009	if (nlink != inode->i_nlink) {
1010		inode->i_nlink = nlink;
1011		btrfs_update_inode(trans, root, inode);
1012	}
1013	BTRFS_I(inode)->index_cnt = (u64)-1;
1014
1015	if (inode->i_nlink == 0) {
1016		if (S_ISDIR(inode->i_mode)) {
1017			ret = replay_dir_deletes(trans, root, NULL, path,
1018						 inode->i_ino, 1);
1019			BUG_ON(ret);
1020		}
1021		ret = insert_orphan_item(trans, root, inode->i_ino);
1022		BUG_ON(ret);
1023	}
1024	btrfs_free_path(path);
1025
1026	return 0;
1027}
1028
1029static noinline int fixup_inode_link_counts(struct btrfs_trans_handle *trans,
1030					    struct btrfs_root *root,
1031					    struct btrfs_path *path)
1032{
1033	int ret;
1034	struct btrfs_key key;
1035	struct inode *inode;
1036
1037	key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1038	key.type = BTRFS_ORPHAN_ITEM_KEY;
1039	key.offset = (u64)-1;
1040	while (1) {
1041		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1042		if (ret < 0)
1043			break;
1044
1045		if (ret == 1) {
1046			if (path->slots[0] == 0)
1047				break;
1048			path->slots[0]--;
1049		}
1050
1051		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1052		if (key.objectid != BTRFS_TREE_LOG_FIXUP_OBJECTID ||
1053		    key.type != BTRFS_ORPHAN_ITEM_KEY)
1054			break;
1055
1056		ret = btrfs_del_item(trans, root, path);
1057		BUG_ON(ret);
1058
1059		btrfs_release_path(root, path);
1060		inode = read_one_inode(root, key.offset);
1061		BUG_ON(!inode);
1062
1063		ret = fixup_inode_link_count(trans, root, inode);
1064		BUG_ON(ret);
1065
1066		iput(inode);
1067
1068		/*
1069		 * fixup on a directory may create new entries,
1070		 * make sure we always look for the highset possible
1071		 * offset
1072		 */
1073		key.offset = (u64)-1;
1074	}
1075	btrfs_release_path(root, path);
1076	return 0;
1077}
1078
1079
1080/*
1081 * record a given inode in the fixup dir so we can check its link
1082 * count when replay is done.  The link count is incremented here
1083 * so the inode won't go away until we check it
1084 */
1085static noinline int link_to_fixup_dir(struct btrfs_trans_handle *trans,
1086				      struct btrfs_root *root,
1087				      struct btrfs_path *path,
1088				      u64 objectid)
1089{
1090	struct btrfs_key key;
1091	int ret = 0;
1092	struct inode *inode;
1093
1094	inode = read_one_inode(root, objectid);
1095	BUG_ON(!inode);
1096
1097	key.objectid = BTRFS_TREE_LOG_FIXUP_OBJECTID;
1098	btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1099	key.offset = objectid;
1100
1101	ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1102
1103	btrfs_release_path(root, path);
1104	if (ret == 0) {
1105		btrfs_inc_nlink(inode);
1106		btrfs_update_inode(trans, root, inode);
1107	} else if (ret == -EEXIST) {
1108		ret = 0;
1109	} else {
1110		BUG();
1111	}
1112	iput(inode);
1113
1114	return ret;
1115}
1116
1117/*
1118 * when replaying the log for a directory, we only insert names
1119 * for inodes that actually exist.  This means an fsync on a directory
1120 * does not implicitly fsync all the new files in it
1121 */
1122static noinline int insert_one_name(struct btrfs_trans_handle *trans,
1123				    struct btrfs_root *root,
1124				    struct btrfs_path *path,
1125				    u64 dirid, u64 index,
1126				    char *name, int name_len, u8 type,
1127				    struct btrfs_key *location)
1128{
1129	struct inode *inode;
1130	struct inode *dir;
1131	int ret;
1132
1133	inode = read_one_inode(root, location->objectid);
1134	if (!inode)
1135		return -ENOENT;
1136
1137	dir = read_one_inode(root, dirid);
1138	if (!dir) {
1139		iput(inode);
1140		return -EIO;
1141	}
1142	ret = btrfs_add_link(trans, dir, inode, name, name_len, 1, index);
1143
1144
1145	iput(inode);
1146	iput(dir);
1147	return ret;
1148}
1149
1150/*
1151 * take a single entry in a log directory item and replay it into
1152 * the subvolume.
1153 *
1154 * if a conflicting item exists in the subdirectory already,
1155 * the inode it points to is unlinked and put into the link count
1156 * fix up tree.
1157 *
1158 * If a name from the log points to a file or directory that does
1159 * not exist in the FS, it is skipped.  fsyncs on directories
1160 * do not force down inodes inside that directory, just changes to the
1161 * names or unlinks in a directory.
1162 */
1163static noinline int replay_one_name(struct btrfs_trans_handle *trans,
1164				    struct btrfs_root *root,
1165				    struct btrfs_path *path,
1166				    struct extent_buffer *eb,
1167				    struct btrfs_dir_item *di,
1168				    struct btrfs_key *key)
1169{
1170	char *name;
1171	int name_len;
1172	struct btrfs_dir_item *dst_di;
1173	struct btrfs_key found_key;
1174	struct btrfs_key log_key;
1175	struct inode *dir;
1176	u8 log_type;
1177	int exists;
1178	int ret;
1179
1180	dir = read_one_inode(root, key->objectid);
1181	BUG_ON(!dir);
1182
1183	name_len = btrfs_dir_name_len(eb, di);
1184	name = kmalloc(name_len, GFP_NOFS);
1185	log_type = btrfs_dir_type(eb, di);
1186	read_extent_buffer(eb, name, (unsigned long)(di + 1),
1187		   name_len);
1188
1189	btrfs_dir_item_key_to_cpu(eb, di, &log_key);
1190	exists = btrfs_lookup_inode(trans, root, path, &log_key, 0);
1191	if (exists == 0)
1192		exists = 1;
1193	else
1194		exists = 0;
1195	btrfs_release_path(root, path);
1196
1197	if (key->type == BTRFS_DIR_ITEM_KEY) {
1198		dst_di = btrfs_lookup_dir_item(trans, root, path, key->objectid,
1199				       name, name_len, 1);
1200	} else if (key->type == BTRFS_DIR_INDEX_KEY) {
1201		dst_di = btrfs_lookup_dir_index_item(trans, root, path,
1202						     key->objectid,
1203						     key->offset, name,
1204						     name_len, 1);
1205	} else {
1206		BUG();
1207	}
1208	if (!dst_di || IS_ERR(dst_di)) {
1209		/* we need a sequence number to insert, so we only
1210		 * do inserts for the BTRFS_DIR_INDEX_KEY types
1211		 */
1212		if (key->type != BTRFS_DIR_INDEX_KEY)
1213			goto out;
1214		goto insert;
1215	}
1216
1217	btrfs_dir_item_key_to_cpu(path->nodes[0], dst_di, &found_key);
1218	/* the existing item matches the logged item */
1219	if (found_key.objectid == log_key.objectid &&
1220	    found_key.type == log_key.type &&
1221	    found_key.offset == log_key.offset &&
1222	    btrfs_dir_type(path->nodes[0], dst_di) == log_type) {
1223		goto out;
1224	}
1225
1226	/*
1227	 * don't drop the conflicting directory entry if the inode
1228	 * for the new entry doesn't exist
1229	 */
1230	if (!exists)
1231		goto out;
1232
1233	ret = drop_one_dir_item(trans, root, path, dir, dst_di);
1234	BUG_ON(ret);
1235
1236	if (key->type == BTRFS_DIR_INDEX_KEY)
1237		goto insert;
1238out:
1239	btrfs_release_path(root, path);
1240	kfree(name);
1241	iput(dir);
1242	return 0;
1243
1244insert:
1245	btrfs_release_path(root, path);
1246	ret = insert_one_name(trans, root, path, key->objectid, key->offset,
1247			      name, name_len, log_type, &log_key);
1248
1249	BUG_ON(ret && ret != -ENOENT);
1250	goto out;
1251}
1252
1253/*
1254 * find all the names in a directory item and reconcile them into
1255 * the subvolume.  Only BTRFS_DIR_ITEM_KEY types will have more than
1256 * one name in a directory item, but the same code gets used for
1257 * both directory index types
1258 */
1259static noinline int replay_one_dir_item(struct btrfs_trans_handle *trans,
1260					struct btrfs_root *root,
1261					struct btrfs_path *path,
1262					struct extent_buffer *eb, int slot,
1263					struct btrfs_key *key)
1264{
1265	int ret;
1266	u32 item_size = btrfs_item_size_nr(eb, slot);
1267	struct btrfs_dir_item *di;
1268	int name_len;
1269	unsigned long ptr;
1270	unsigned long ptr_end;
1271
1272	ptr = btrfs_item_ptr_offset(eb, slot);
1273	ptr_end = ptr + item_size;
1274	while (ptr < ptr_end) {
1275		di = (struct btrfs_dir_item *)ptr;
1276		name_len = btrfs_dir_name_len(eb, di);
1277		ret = replay_one_name(trans, root, path, eb, di, key);
1278		BUG_ON(ret);
1279		ptr = (unsigned long)(di + 1);
1280		ptr += name_len;
1281	}
1282	return 0;
1283}
1284
1285/*
1286 * directory replay has two parts.  There are the standard directory
1287 * items in the log copied from the subvolume, and range items
1288 * created in the log while the subvolume was logged.
1289 *
1290 * The range items tell us which parts of the key space the log
1291 * is authoritative for.  During replay, if a key in the subvolume
1292 * directory is in a logged range item, but not actually in the log
1293 * that means it was deleted from the directory before the fsync
1294 * and should be removed.
1295 */
1296static noinline int find_dir_range(struct btrfs_root *root,
1297				   struct btrfs_path *path,
1298				   u64 dirid, int key_type,
1299				   u64 *start_ret, u64 *end_ret)
1300{
1301	struct btrfs_key key;
1302	u64 found_end;
1303	struct btrfs_dir_log_item *item;
1304	int ret;
1305	int nritems;
1306
1307	if (*start_ret == (u64)-1)
1308		return 1;
1309
1310	key.objectid = dirid;
1311	key.type = key_type;
1312	key.offset = *start_ret;
1313
1314	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1315	if (ret < 0)
1316		goto out;
1317	if (ret > 0) {
1318		if (path->slots[0] == 0)
1319			goto out;
1320		path->slots[0]--;
1321	}
1322	if (ret != 0)
1323		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1324
1325	if (key.type != key_type || key.objectid != dirid) {
1326		ret = 1;
1327		goto next;
1328	}
1329	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1330			      struct btrfs_dir_log_item);
1331	found_end = btrfs_dir_log_end(path->nodes[0], item);
1332
1333	if (*start_ret >= key.offset && *start_ret <= found_end) {
1334		ret = 0;
1335		*start_ret = key.offset;
1336		*end_ret = found_end;
1337		goto out;
1338	}
1339	ret = 1;
1340next:
1341	/* check the next slot in the tree to see if it is a valid item */
1342	nritems = btrfs_header_nritems(path->nodes[0]);
1343	if (path->slots[0] >= nritems) {
1344		ret = btrfs_next_leaf(root, path);
1345		if (ret)
1346			goto out;
1347	} else {
1348		path->slots[0]++;
1349	}
1350
1351	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1352
1353	if (key.type != key_type || key.objectid != dirid) {
1354		ret = 1;
1355		goto out;
1356	}
1357	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1358			      struct btrfs_dir_log_item);
1359	found_end = btrfs_dir_log_end(path->nodes[0], item);
1360	*start_ret = key.offset;
1361	*end_ret = found_end;
1362	ret = 0;
1363out:
1364	btrfs_release_path(root, path);
1365	return ret;
1366}
1367
1368/*
1369 * this looks for a given directory item in the log.  If the directory
1370 * item is not in the log, the item is removed and the inode it points
1371 * to is unlinked
1372 */
1373static noinline int check_item_in_log(struct btrfs_trans_handle *trans,
1374				      struct btrfs_root *root,
1375				      struct btrfs_root *log,
1376				      struct btrfs_path *path,
1377				      struct btrfs_path *log_path,
1378				      struct inode *dir,
1379				      struct btrfs_key *dir_key)
1380{
1381	int ret;
1382	struct extent_buffer *eb;
1383	int slot;
1384	u32 item_size;
1385	struct btrfs_dir_item *di;
1386	struct btrfs_dir_item *log_di;
1387	int name_len;
1388	unsigned long ptr;
1389	unsigned long ptr_end;
1390	char *name;
1391	struct inode *inode;
1392	struct btrfs_key location;
1393
1394again:
1395	eb = path->nodes[0];
1396	slot = path->slots[0];
1397	item_size = btrfs_item_size_nr(eb, slot);
1398	ptr = btrfs_item_ptr_offset(eb, slot);
1399	ptr_end = ptr + item_size;
1400	while (ptr < ptr_end) {
1401		di = (struct btrfs_dir_item *)ptr;
1402		name_len = btrfs_dir_name_len(eb, di);
1403		name = kmalloc(name_len, GFP_NOFS);
1404		if (!name) {
1405			ret = -ENOMEM;
1406			goto out;
1407		}
1408		read_extent_buffer(eb, name, (unsigned long)(di + 1),
1409				  name_len);
1410		log_di = NULL;
1411		if (log && dir_key->type == BTRFS_DIR_ITEM_KEY) {
1412			log_di = btrfs_lookup_dir_item(trans, log, log_path,
1413						       dir_key->objectid,
1414						       name, name_len, 0);
1415		} else if (log && dir_key->type == BTRFS_DIR_INDEX_KEY) {
1416			log_di = btrfs_lookup_dir_index_item(trans, log,
1417						     log_path,
1418						     dir_key->objectid,
1419						     dir_key->offset,
1420						     name, name_len, 0);
1421		}
1422		if (!log_di || IS_ERR(log_di)) {
1423			btrfs_dir_item_key_to_cpu(eb, di, &location);
1424			btrfs_release_path(root, path);
1425			btrfs_release_path(log, log_path);
1426			inode = read_one_inode(root, location.objectid);
1427			BUG_ON(!inode);
1428
1429			ret = link_to_fixup_dir(trans, root,
1430						path, location.objectid);
1431			BUG_ON(ret);
1432			btrfs_inc_nlink(inode);
1433			ret = btrfs_unlink_inode(trans, root, dir, inode,
1434						 name, name_len);
1435			BUG_ON(ret);
1436			kfree(name);
1437			iput(inode);
1438
1439			/* there might still be more names under this key
1440			 * check and repeat if required
1441			 */
1442			ret = btrfs_search_slot(NULL, root, dir_key, path,
1443						0, 0);
1444			if (ret == 0)
1445				goto again;
1446			ret = 0;
1447			goto out;
1448		}
1449		btrfs_release_path(log, log_path);
1450		kfree(name);
1451
1452		ptr = (unsigned long)(di + 1);
1453		ptr += name_len;
1454	}
1455	ret = 0;
1456out:
1457	btrfs_release_path(root, path);
1458	btrfs_release_path(log, log_path);
1459	return ret;
1460}
1461
1462/*
1463 * deletion replay happens before we copy any new directory items
1464 * out of the log or out of backreferences from inodes.  It
1465 * scans the log to find ranges of keys that log is authoritative for,
1466 * and then scans the directory to find items in those ranges that are
1467 * not present in the log.
1468 *
1469 * Anything we don't find in the log is unlinked and removed from the
1470 * directory.
1471 */
1472static noinline int replay_dir_deletes(struct btrfs_trans_handle *trans,
1473				       struct btrfs_root *root,
1474				       struct btrfs_root *log,
1475				       struct btrfs_path *path,
1476				       u64 dirid, int del_all)
1477{
1478	u64 range_start;
1479	u64 range_end;
1480	int key_type = BTRFS_DIR_LOG_ITEM_KEY;
1481	int ret = 0;
1482	struct btrfs_key dir_key;
1483	struct btrfs_key found_key;
1484	struct btrfs_path *log_path;
1485	struct inode *dir;
1486
1487	dir_key.objectid = dirid;
1488	dir_key.type = BTRFS_DIR_ITEM_KEY;
1489	log_path = btrfs_alloc_path();
1490	if (!log_path)
1491		return -ENOMEM;
1492
1493	dir = read_one_inode(root, dirid);
1494	/* it isn't an error if the inode isn't there, that can happen
1495	 * because we replay the deletes before we copy in the inode item
1496	 * from the log
1497	 */
1498	if (!dir) {
1499		btrfs_free_path(log_path);
1500		return 0;
1501	}
1502again:
1503	range_start = 0;
1504	range_end = 0;
1505	while (1) {
1506		if (del_all)
1507			range_end = (u64)-1;
1508		else {
1509			ret = find_dir_range(log, path, dirid, key_type,
1510					     &range_start, &range_end);
1511			if (ret != 0)
1512				break;
1513		}
1514
1515		dir_key.offset = range_start;
1516		while (1) {
1517			int nritems;
1518			ret = btrfs_search_slot(NULL, root, &dir_key, path,
1519						0, 0);
1520			if (ret < 0)
1521				goto out;
1522
1523			nritems = btrfs_header_nritems(path->nodes[0]);
1524			if (path->slots[0] >= nritems) {
1525				ret = btrfs_next_leaf(root, path);
1526				if (ret)
1527					break;
1528			}
1529			btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1530					      path->slots[0]);
1531			if (found_key.objectid != dirid ||
1532			    found_key.type != dir_key.type)
1533				goto next_type;
1534
1535			if (found_key.offset > range_end)
1536				break;
1537
1538			ret = check_item_in_log(trans, root, log, path,
1539						log_path, dir,
1540						&found_key);
1541			BUG_ON(ret);
1542			if (found_key.offset == (u64)-1)
1543				break;
1544			dir_key.offset = found_key.offset + 1;
1545		}
1546		btrfs_release_path(root, path);
1547		if (range_end == (u64)-1)
1548			break;
1549		range_start = range_end + 1;
1550	}
1551
1552next_type:
1553	ret = 0;
1554	if (key_type == BTRFS_DIR_LOG_ITEM_KEY) {
1555		key_type = BTRFS_DIR_LOG_INDEX_KEY;
1556		dir_key.type = BTRFS_DIR_INDEX_KEY;
1557		btrfs_release_path(root, path);
1558		goto again;
1559	}
1560out:
1561	btrfs_release_path(root, path);
1562	btrfs_free_path(log_path);
1563	iput(dir);
1564	return ret;
1565}
1566
1567/*
1568 * the process_func used to replay items from the log tree.  This
1569 * gets called in two different stages.  The first stage just looks
1570 * for inodes and makes sure they are all copied into the subvolume.
1571 *
1572 * The second stage copies all the other item types from the log into
1573 * the subvolume.  The two stage approach is slower, but gets rid of
1574 * lots of complexity around inodes referencing other inodes that exist
1575 * only in the log (references come from either directory items or inode
1576 * back refs).
1577 */
1578static int replay_one_buffer(struct btrfs_root *log, struct extent_buffer *eb,
1579			     struct walk_control *wc, u64 gen)
1580{
1581	int nritems;
1582	struct btrfs_path *path;
1583	struct btrfs_root *root = wc->replay_dest;
1584	struct btrfs_key key;
1585	u32 item_size;
1586	int level;
1587	int i;
1588	int ret;
1589
1590	btrfs_read_buffer(eb, gen);
1591
1592	level = btrfs_header_level(eb);
1593
1594	if (level != 0)
1595		return 0;
1596
1597	path = btrfs_alloc_path();
1598	BUG_ON(!path);
1599
1600	nritems = btrfs_header_nritems(eb);
1601	for (i = 0; i < nritems; i++) {
1602		btrfs_item_key_to_cpu(eb, &key, i);
1603		item_size = btrfs_item_size_nr(eb, i);
1604
1605		/* inode keys are done during the first stage */
1606		if (key.type == BTRFS_INODE_ITEM_KEY &&
1607		    wc->stage == LOG_WALK_REPLAY_INODES) {
1608			struct btrfs_inode_item *inode_item;
1609			u32 mode;
1610
1611			inode_item = btrfs_item_ptr(eb, i,
1612					    struct btrfs_inode_item);
1613			mode = btrfs_inode_mode(eb, inode_item);
1614			if (S_ISDIR(mode)) {
1615				ret = replay_dir_deletes(wc->trans,
1616					 root, log, path, key.objectid, 0);
1617				BUG_ON(ret);
1618			}
1619			ret = overwrite_item(wc->trans, root, path,
1620					     eb, i, &key);
1621			BUG_ON(ret);
1622
1623			/* for regular files, make sure corresponding
1624			 * orhpan item exist. extents past the new EOF
1625			 * will be truncated later by orphan cleanup.
1626			 */
1627			if (S_ISREG(mode)) {
1628				ret = insert_orphan_item(wc->trans, root,
1629							 key.objectid);
1630				BUG_ON(ret);
1631			}
1632
1633			ret = link_to_fixup_dir(wc->trans, root,
1634						path, key.objectid);
1635			BUG_ON(ret);
1636		}
1637		if (wc->stage < LOG_WALK_REPLAY_ALL)
1638			continue;
1639
1640		/* these keys are simply copied */
1641		if (key.type == BTRFS_XATTR_ITEM_KEY) {
1642			ret = overwrite_item(wc->trans, root, path,
1643					     eb, i, &key);
1644			BUG_ON(ret);
1645		} else if (key.type == BTRFS_INODE_REF_KEY) {
1646			ret = add_inode_ref(wc->trans, root, log, path,
1647					    eb, i, &key);
1648			BUG_ON(ret && ret != -ENOENT);
1649		} else if (key.type == BTRFS_EXTENT_DATA_KEY) {
1650			ret = replay_one_extent(wc->trans, root, path,
1651						eb, i, &key);
1652			BUG_ON(ret);
1653		} else if (key.type == BTRFS_DIR_ITEM_KEY ||
1654			   key.type == BTRFS_DIR_INDEX_KEY) {
1655			ret = replay_one_dir_item(wc->trans, root, path,
1656						  eb, i, &key);
1657			BUG_ON(ret);
1658		}
1659	}
1660	btrfs_free_path(path);
1661	return 0;
1662}
1663
1664static noinline int walk_down_log_tree(struct btrfs_trans_handle *trans,
1665				   struct btrfs_root *root,
1666				   struct btrfs_path *path, int *level,
1667				   struct walk_control *wc)
1668{
1669	u64 root_owner;
1670	u64 root_gen;
1671	u64 bytenr;
1672	u64 ptr_gen;
1673	struct extent_buffer *next;
1674	struct extent_buffer *cur;
1675	struct extent_buffer *parent;
1676	u32 blocksize;
1677	int ret = 0;
1678
1679	WARN_ON(*level < 0);
1680	WARN_ON(*level >= BTRFS_MAX_LEVEL);
1681
1682	while (*level > 0) {
1683		WARN_ON(*level < 0);
1684		WARN_ON(*level >= BTRFS_MAX_LEVEL);
1685		cur = path->nodes[*level];
1686
1687		if (btrfs_header_level(cur) != *level)
1688			WARN_ON(1);
1689
1690		if (path->slots[*level] >=
1691		    btrfs_header_nritems(cur))
1692			break;
1693
1694		bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1695		ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
1696		blocksize = btrfs_level_size(root, *level - 1);
1697
1698		parent = path->nodes[*level];
1699		root_owner = btrfs_header_owner(parent);
1700		root_gen = btrfs_header_generation(parent);
1701
1702		next = btrfs_find_create_tree_block(root, bytenr, blocksize);
1703
1704		if (*level == 1) {
1705			wc->process_func(root, next, wc, ptr_gen);
1706
1707			path->slots[*level]++;
1708			if (wc->free) {
1709				btrfs_read_buffer(next, ptr_gen);
1710
1711				btrfs_tree_lock(next);
1712				clean_tree_block(trans, root, next);
1713				btrfs_set_lock_blocking(next);
1714				btrfs_wait_tree_block_writeback(next);
1715				btrfs_tree_unlock(next);
1716
1717				WARN_ON(root_owner !=
1718					BTRFS_TREE_LOG_OBJECTID);
1719				ret = btrfs_free_reserved_extent(root,
1720							 bytenr, blocksize);
1721				BUG_ON(ret);
1722			}
1723			free_extent_buffer(next);
1724			continue;
1725		}
1726		btrfs_read_buffer(next, ptr_gen);
1727
1728		WARN_ON(*level <= 0);
1729		if (path->nodes[*level-1])
1730			free_extent_buffer(path->nodes[*level-1]);
1731		path->nodes[*level-1] = next;
1732		*level = btrfs_header_level(next);
1733		path->slots[*level] = 0;
1734		cond_resched();
1735	}
1736	WARN_ON(*level < 0);
1737	WARN_ON(*level >= BTRFS_MAX_LEVEL);
1738
1739	path->slots[*level] = btrfs_header_nritems(path->nodes[*level]);
1740
1741	cond_resched();
1742	return 0;
1743}
1744
1745static noinline int walk_up_log_tree(struct btrfs_trans_handle *trans,
1746				 struct btrfs_root *root,
1747				 struct btrfs_path *path, int *level,
1748				 struct walk_control *wc)
1749{
1750	u64 root_owner;
1751	u64 root_gen;
1752	int i;
1753	int slot;
1754	int ret;
1755
1756	for (i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1757		slot = path->slots[i];
1758		if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
1759			struct extent_buffer *node;
1760			node = path->nodes[i];
1761			path->slots[i]++;
1762			*level = i;
1763			WARN_ON(*level == 0);
1764			return 0;
1765		} else {
1766			struct extent_buffer *parent;
1767			if (path->nodes[*level] == root->node)
1768				parent = path->nodes[*level];
1769			else
1770				parent = path->nodes[*level + 1];
1771
1772			root_owner = btrfs_header_owner(parent);
1773			root_gen = btrfs_header_generation(parent);
1774			wc->process_func(root, path->nodes[*level], wc,
1775				 btrfs_header_generation(path->nodes[*level]));
1776			if (wc->free) {
1777				struct extent_buffer *next;
1778
1779				next = path->nodes[*level];
1780
1781				btrfs_tree_lock(next);
1782				clean_tree_block(trans, root, next);
1783				btrfs_set_lock_blocking(next);
1784				btrfs_wait_tree_block_writeback(next);
1785				btrfs_tree_unlock(next);
1786
1787				WARN_ON(root_owner != BTRFS_TREE_LOG_OBJECTID);
1788				ret = btrfs_free_reserved_extent(root,
1789						path->nodes[*level]->start,
1790						path->nodes[*level]->len);
1791				BUG_ON(ret);
1792			}
1793			free_extent_buffer(path->nodes[*level]);
1794			path->nodes[*level] = NULL;
1795			*level = i + 1;
1796		}
1797	}
1798	return 1;
1799}
1800
1801/*
1802 * drop the reference count on the tree rooted at 'snap'.  This traverses
1803 * the tree freeing any blocks that have a ref count of zero after being
1804 * decremented.
1805 */
1806static int walk_log_tree(struct btrfs_trans_handle *trans,
1807			 struct btrfs_root *log, struct walk_control *wc)
1808{
1809	int ret = 0;
1810	int wret;
1811	int level;
1812	struct btrfs_path *path;
1813	int i;
1814	int orig_level;
1815
1816	path = btrfs_alloc_path();
1817	BUG_ON(!path);
1818
1819	level = btrfs_header_level(log->node);
1820	orig_level = level;
1821	path->nodes[level] = log->node;
1822	extent_buffer_get(log->node);
1823	path->slots[level] = 0;
1824
1825	while (1) {
1826		wret = walk_down_log_tree(trans, log, path, &level, wc);
1827		if (wret > 0)
1828			break;
1829		if (wret < 0)
1830			ret = wret;
1831
1832		wret = walk_up_log_tree(trans, log, path, &level, wc);
1833		if (wret > 0)
1834			break;
1835		if (wret < 0)
1836			ret = wret;
1837	}
1838
1839	/* was the root node processed? if not, catch it here */
1840	if (path->nodes[orig_level]) {
1841		wc->process_func(log, path->nodes[orig_level], wc,
1842			 btrfs_header_generation(path->nodes[orig_level]));
1843		if (wc->free) {
1844			struct extent_buffer *next;
1845
1846			next = path->nodes[orig_level];
1847
1848			btrfs_tree_lock(next);
1849			clean_tree_block(trans, log, next);
1850			btrfs_set_lock_blocking(next);
1851			btrfs_wait_tree_block_writeback(next);
1852			btrfs_tree_unlock(next);
1853
1854			WARN_ON(log->root_key.objectid !=
1855				BTRFS_TREE_LOG_OBJECTID);
1856			ret = btrfs_free_reserved_extent(log, next->start,
1857							 next->len);
1858			BUG_ON(ret);
1859		}
1860	}
1861
1862	for (i = 0; i <= orig_level; i++) {
1863		if (path->nodes[i]) {
1864			free_extent_buffer(path->nodes[i]);
1865			path->nodes[i] = NULL;
1866		}
1867	}
1868	btrfs_free_path(path);
1869	return ret;
1870}
1871
1872/*
1873 * helper function to update the item for a given subvolumes log root
1874 * in the tree of log roots
1875 */
1876static int update_log_root(struct btrfs_trans_handle *trans,
1877			   struct btrfs_root *log)
1878{
1879	int ret;
1880
1881	if (log->log_transid == 1) {
1882		/* insert root item on the first sync */
1883		ret = btrfs_insert_root(trans, log->fs_info->log_root_tree,
1884				&log->root_key, &log->root_item);
1885	} else {
1886		ret = btrfs_update_root(trans, log->fs_info->log_root_tree,
1887				&log->root_key, &log->root_item);
1888	}
1889	return ret;
1890}
1891
1892static int wait_log_commit(struct btrfs_trans_handle *trans,
1893			   struct btrfs_root *root, unsigned long transid)
1894{
1895	DEFINE_WAIT(wait);
1896	int index = transid % 2;
1897
1898	/*
1899	 * we only allow two pending log transactions at a time,
1900	 * so we know that if ours is more than 2 older than the
1901	 * current transaction, we're done
1902	 */
1903	do {
1904		prepare_to_wait(&root->log_commit_wait[index],
1905				&wait, TASK_UNINTERRUPTIBLE);
1906		mutex_unlock(&root->log_mutex);
1907
1908		if (root->fs_info->last_trans_log_full_commit !=
1909		    trans->transid && root->log_transid < transid + 2 &&
1910		    atomic_read(&root->log_commit[index]))
1911			schedule();
1912
1913		finish_wait(&root->log_commit_wait[index], &wait);
1914		mutex_lock(&root->log_mutex);
1915	} while (root->log_transid < transid + 2 &&
1916		 atomic_read(&root->log_commit[index]));
1917	return 0;
1918}
1919
1920static int wait_for_writer(struct btrfs_trans_handle *trans,
1921			   struct btrfs_root *root)
1922{
1923	DEFINE_WAIT(wait);
1924	while (atomic_read(&root->log_writers)) {
1925		prepare_to_wait(&root->log_writer_wait,
1926				&wait, TASK_UNINTERRUPTIBLE);
1927		mutex_unlock(&root->log_mutex);
1928		if (root->fs_info->last_trans_log_full_commit !=
1929		    trans->transid && atomic_read(&root->log_writers))
1930			schedule();
1931		mutex_lock(&root->log_mutex);
1932		finish_wait(&root->log_writer_wait, &wait);
1933	}
1934	return 0;
1935}
1936
1937/*
1938 * btrfs_sync_log does sends a given tree log down to the disk and
1939 * updates the super blocks to record it.  When this call is done,
1940 * you know that any inodes previously logged are safely on disk only
1941 * if it returns 0.
1942 *
1943 * Any other return value means you need to call btrfs_commit_transaction.
1944 * Some of the edge cases for fsyncing directories that have had unlinks
1945 * or renames done in the past mean that sometimes the only safe
1946 * fsync is to commit the whole FS.  When btrfs_sync_log returns -EAGAIN,
1947 * that has happened.
1948 */
1949int btrfs_sync_log(struct btrfs_trans_handle *trans,
1950		   struct btrfs_root *root)
1951{
1952	int index1;
1953	int index2;
1954	int mark;
1955	int ret;
1956	struct btrfs_root *log = root->log_root;
1957	struct btrfs_root *log_root_tree = root->fs_info->log_root_tree;
1958	unsigned long log_transid = 0;
1959
1960	mutex_lock(&root->log_mutex);
1961	index1 = root->log_transid % 2;
1962	if (atomic_read(&root->log_commit[index1])) {
1963		wait_log_commit(trans, root, root->log_transid);
1964		mutex_unlock(&root->log_mutex);
1965		return 0;
1966	}
1967	atomic_set(&root->log_commit[index1], 1);
1968
1969	/* wait for previous tree log sync to complete */
1970	if (atomic_read(&root->log_commit[(index1 + 1) % 2]))
1971		wait_log_commit(trans, root, root->log_transid - 1);
1972
1973	while (1) {
1974		unsigned long batch = root->log_batch;
1975		if (root->log_multiple_pids) {
1976			mutex_unlock(&root->log_mutex);
1977			schedule_timeout_uninterruptible(1);
1978			mutex_lock(&root->log_mutex);
1979		}
1980		wait_for_writer(trans, root);
1981		if (batch == root->log_batch)
1982			break;
1983	}
1984
1985	/* bail out if we need to do a full commit */
1986	if (root->fs_info->last_trans_log_full_commit == trans->transid) {
1987		ret = -EAGAIN;
1988		mutex_unlock(&root->log_mutex);
1989		goto out;
1990	}
1991
1992	log_transid = root->log_transid;
1993	if (log_transid % 2 == 0)
1994		mark = EXTENT_DIRTY;
1995	else
1996		mark = EXTENT_NEW;
1997
1998	/* we start IO on  all the marked extents here, but we don't actually
1999	 * wait for them until later.
2000	 */
2001	ret = btrfs_write_marked_extents(log, &log->dirty_log_pages, mark);
2002	BUG_ON(ret);
2003
2004	btrfs_set_root_node(&log->root_item, log->node);
2005
2006	root->log_batch = 0;
2007	root->log_transid++;
2008	log->log_transid = root->log_transid;
2009	root->log_start_pid = 0;
2010	smp_mb();
2011	/*
2012	 * IO has been started, blocks of the log tree have WRITTEN flag set
2013	 * in their headers. new modifications of the log will be written to
2014	 * new positions. so it's safe to allow log writers to go in.
2015	 */
2016	mutex_unlock(&root->log_mutex);
2017
2018	mutex_lock(&log_root_tree->log_mutex);
2019	log_root_tree->log_batch++;
2020	atomic_inc(&log_root_tree->log_writers);
2021	mutex_unlock(&log_root_tree->log_mutex);
2022
2023	ret = update_log_root(trans, log);
2024
2025	mutex_lock(&log_root_tree->log_mutex);
2026	if (atomic_dec_and_test(&log_root_tree->log_writers)) {
2027		smp_mb();
2028		if (waitqueue_active(&log_root_tree->log_writer_wait))
2029			wake_up(&log_root_tree->log_writer_wait);
2030	}
2031
2032	if (ret) {
2033		BUG_ON(ret != -ENOSPC);
2034		root->fs_info->last_trans_log_full_commit = trans->transid;
2035		btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2036		mutex_unlock(&log_root_tree->log_mutex);
2037		ret = -EAGAIN;
2038		goto out;
2039	}
2040
2041	index2 = log_root_tree->log_transid % 2;
2042	if (atomic_read(&log_root_tree->log_commit[index2])) {
2043		btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2044		wait_log_commit(trans, log_root_tree,
2045				log_root_tree->log_transid);
2046		mutex_unlock(&log_root_tree->log_mutex);
2047		goto out;
2048	}
2049	atomic_set(&log_root_tree->log_commit[index2], 1);
2050
2051	if (atomic_read(&log_root_tree->log_commit[(index2 + 1) % 2])) {
2052		wait_log_commit(trans, log_root_tree,
2053				log_root_tree->log_transid - 1);
2054	}
2055
2056	wait_for_writer(trans, log_root_tree);
2057
2058	/*
2059	 * now that we've moved on to the tree of log tree roots,
2060	 * check the full commit flag again
2061	 */
2062	if (root->fs_info->last_trans_log_full_commit == trans->transid) {
2063		btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2064		mutex_unlock(&log_root_tree->log_mutex);
2065		ret = -EAGAIN;
2066		goto out_wake_log_root;
2067	}
2068
2069	ret = btrfs_write_and_wait_marked_extents(log_root_tree,
2070				&log_root_tree->dirty_log_pages,
2071				EXTENT_DIRTY | EXTENT_NEW);
2072	BUG_ON(ret);
2073	btrfs_wait_marked_extents(log, &log->dirty_log_pages, mark);
2074
2075	btrfs_set_super_log_root(&root->fs_info->super_for_commit,
2076				log_root_tree->node->start);
2077	btrfs_set_super_log_root_level(&root->fs_info->super_for_commit,
2078				btrfs_header_level(log_root_tree->node));
2079
2080	log_root_tree->log_batch = 0;
2081	log_root_tree->log_transid++;
2082	smp_mb();
2083
2084	mutex_unlock(&log_root_tree->log_mutex);
2085
2086	/*
2087	 * nobody else is going to jump in and write the the ctree
2088	 * super here because the log_commit atomic below is protecting
2089	 * us.  We must be called with a transaction handle pinning
2090	 * the running transaction open, so a full commit can't hop
2091	 * in and cause problems either.
2092	 */
2093	write_ctree_super(trans, root->fs_info->tree_root, 1);
2094	ret = 0;
2095
2096	mutex_lock(&root->log_mutex);
2097	if (root->last_log_commit < log_transid)
2098		root->last_log_commit = log_transid;
2099	mutex_unlock(&root->log_mutex);
2100
2101out_wake_log_root:
2102	atomic_set(&log_root_tree->log_commit[index2], 0);
2103	smp_mb();
2104	if (waitqueue_active(&log_root_tree->log_commit_wait[index2]))
2105		wake_up(&log_root_tree->log_commit_wait[index2]);
2106out:
2107	atomic_set(&root->log_commit[index1], 0);
2108	smp_mb();
2109	if (waitqueue_active(&root->log_commit_wait[index1]))
2110		wake_up(&root->log_commit_wait[index1]);
2111	return 0;
2112}
2113
2114static void free_log_tree(struct btrfs_trans_handle *trans,
2115			  struct btrfs_root *log)
2116{
2117	int ret;
2118	u64 start;
2119	u64 end;
2120	struct walk_control wc = {
2121		.free = 1,
2122		.process_func = process_one_buffer
2123	};
2124
2125	ret = walk_log_tree(trans, log, &wc);
2126	BUG_ON(ret);
2127
2128	while (1) {
2129		ret = find_first_extent_bit(&log->dirty_log_pages,
2130				0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
2131		if (ret)
2132			break;
2133
2134		clear_extent_bits(&log->dirty_log_pages, start, end,
2135				  EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
2136	}
2137
2138	free_extent_buffer(log->node);
2139	kfree(log);
2140}
2141
2142/*
2143 * free all the extents used by the tree log.  This should be called
2144 * at commit time of the full transaction
2145 */
2146int btrfs_free_log(struct btrfs_trans_handle *trans, struct btrfs_root *root)
2147{
2148	if (root->log_root) {
2149		free_log_tree(trans, root->log_root);
2150		root->log_root = NULL;
2151	}
2152	return 0;
2153}
2154
2155int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
2156			     struct btrfs_fs_info *fs_info)
2157{
2158	if (fs_info->log_root_tree) {
2159		free_log_tree(trans, fs_info->log_root_tree);
2160		fs_info->log_root_tree = NULL;
2161	}
2162	return 0;
2163}
2164
2165/*
2166 * If both a file and directory are logged, and unlinks or renames are
2167 * mixed in, we have a few interesting corners:
2168 *
2169 * create file X in dir Y
2170 * link file X to X.link in dir Y
2171 * fsync file X
2172 * unlink file X but leave X.link
2173 * fsync dir Y
2174 *
2175 * After a crash we would expect only X.link to exist.  But file X
2176 * didn't get fsync'd again so the log has back refs for X and X.link.
2177 *
2178 * We solve this by removing directory entries and inode backrefs from the
2179 * log when a file that was logged in the current transaction is
2180 * unlinked.  Any later fsync will include the updated log entries, and
2181 * we'll be able to reconstruct the proper directory items from backrefs.
2182 *
2183 * This optimizations allows us to avoid relogging the entire inode
2184 * or the entire directory.
2185 */
2186int btrfs_del_dir_entries_in_log(struct btrfs_trans_handle *trans,
2187				 struct btrfs_root *root,
2188				 const char *name, int name_len,
2189				 struct inode *dir, u64 index)
2190{
2191	struct btrfs_root *log;
2192	struct btrfs_dir_item *di;
2193	struct btrfs_path *path;
2194	int ret;
2195	int err = 0;
2196	int bytes_del = 0;
2197
2198	if (BTRFS_I(dir)->logged_trans < trans->transid)
2199		return 0;
2200
2201	ret = join_running_log_trans(root);
2202	if (ret)
2203		return 0;
2204
2205	mutex_lock(&BTRFS_I(dir)->log_mutex);
2206
2207	log = root->log_root;
2208	path = btrfs_alloc_path();
2209	di = btrfs_lookup_dir_item(trans, log, path, dir->i_ino,
2210				   name, name_len, -1);
2211	if (IS_ERR(di)) {
2212		err = PTR_ERR(di);
2213		goto fail;
2214	}
2215	if (di) {
2216		ret = btrfs_delete_one_dir_name(trans, log, path, di);
2217		bytes_del += name_len;
2218		BUG_ON(ret);
2219	}
2220	btrfs_release_path(log, path);
2221	di = btrfs_lookup_dir_index_item(trans, log, path, dir->i_ino,
2222					 index, name, name_len, -1);
2223	if (IS_ERR(di)) {
2224		err = PTR_ERR(di);
2225		goto fail;
2226	}
2227	if (di) {
2228		ret = btrfs_delete_one_dir_name(trans, log, path, di);
2229		bytes_del += name_len;
2230		BUG_ON(ret);
2231	}
2232
2233	/* update the directory size in the log to reflect the names
2234	 * we have removed
2235	 */
2236	if (bytes_del) {
2237		struct btrfs_key key;
2238
2239		key.objectid = dir->i_ino;
2240		key.offset = 0;
2241		key.type = BTRFS_INODE_ITEM_KEY;
2242		btrfs_release_path(log, path);
2243
2244		ret = btrfs_search_slot(trans, log, &key, path, 0, 1);
2245		if (ret < 0) {
2246			err = ret;
2247			goto fail;
2248		}
2249		if (ret == 0) {
2250			struct btrfs_inode_item *item;
2251			u64 i_size;
2252
2253			item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2254					      struct btrfs_inode_item);
2255			i_size = btrfs_inode_size(path->nodes[0], item);
2256			if (i_size > bytes_del)
2257				i_size -= bytes_del;
2258			else
2259				i_size = 0;
2260			btrfs_set_inode_size(path->nodes[0], item, i_size);
2261			btrfs_mark_buffer_dirty(path->nodes[0]);
2262		} else
2263			ret = 0;
2264		btrfs_release_path(log, path);
2265	}
2266fail:
2267	btrfs_free_path(path);
2268	mutex_unlock(&BTRFS_I(dir)->log_mutex);
2269	if (ret == -ENOSPC) {
2270		root->fs_info->last_trans_log_full_commit = trans->transid;
2271		ret = 0;
2272	}
2273	btrfs_end_log_trans(root);
2274
2275	return 0;
2276}
2277
2278/* see comments for btrfs_del_dir_entries_in_log */
2279int btrfs_del_inode_ref_in_log(struct btrfs_trans_handle *trans,
2280			       struct btrfs_root *root,
2281			       const char *name, int name_len,
2282			       struct inode *inode, u64 dirid)
2283{
2284	struct btrfs_root *log;
2285	u64 index;
2286	int ret;
2287
2288	if (BTRFS_I(inode)->logged_trans < trans->transid)
2289		return 0;
2290
2291	ret = join_running_log_trans(root);
2292	if (ret)
2293		return 0;
2294	log = root->log_root;
2295	mutex_lock(&BTRFS_I(inode)->log_mutex);
2296
2297	ret = btrfs_del_inode_ref(trans, log, name, name_len, inode->i_ino,
2298				  dirid, &index);
2299	mutex_unlock(&BTRFS_I(inode)->log_mutex);
2300	if (ret == -ENOSPC) {
2301		root->fs_info->last_trans_log_full_commit = trans->transid;
2302		ret = 0;
2303	}
2304	btrfs_end_log_trans(root);
2305
2306	return ret;
2307}
2308
2309/*
2310 * creates a range item in the log for 'dirid'.  first_offset and
2311 * last_offset tell us which parts of the key space the log should
2312 * be considered authoritative for.
2313 */
2314static noinline int insert_dir_log_key(struct btrfs_trans_handle *trans,
2315				       struct btrfs_root *log,
2316				       struct btrfs_path *path,
2317				       int key_type, u64 dirid,
2318				       u64 first_offset, u64 last_offset)
2319{
2320	int ret;
2321	struct btrfs_key key;
2322	struct btrfs_dir_log_item *item;
2323
2324	key.objectid = dirid;
2325	key.offset = first_offset;
2326	if (key_type == BTRFS_DIR_ITEM_KEY)
2327		key.type = BTRFS_DIR_LOG_ITEM_KEY;
2328	else
2329		key.type = BTRFS_DIR_LOG_INDEX_KEY;
2330	ret = btrfs_insert_empty_item(trans, log, path, &key, sizeof(*item));
2331	if (ret)
2332		return ret;
2333
2334	item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2335			      struct btrfs_dir_log_item);
2336	btrfs_set_dir_log_end(path->nodes[0], item, last_offset);
2337	btrfs_mark_buffer_dirty(path->nodes[0]);
2338	btrfs_release_path(log, path);
2339	return 0;
2340}
2341
2342/*
2343 * log all the items included in the current transaction for a given
2344 * directory.  This also creates the range items in the log tree required
2345 * to replay anything deleted before the fsync
2346 */
2347static noinline int log_dir_items(struct btrfs_trans_handle *trans,
2348			  struct btrfs_root *root, struct inode *inode,
2349			  struct btrfs_path *path,
2350			  struct btrfs_path *dst_path, int key_type,
2351			  u64 min_offset, u64 *last_offset_ret)
2352{
2353	struct btrfs_key min_key;
2354	struct btrfs_key max_key;
2355	struct btrfs_root *log = root->log_root;
2356	struct extent_buffer *src;
2357	int err = 0;
2358	int ret;
2359	int i;
2360	int nritems;
2361	u64 first_offset = min_offset;
2362	u64 last_offset = (u64)-1;
2363
2364	log = root->log_root;
2365	max_key.objectid = inode->i_ino;
2366	max_key.offset = (u64)-1;
2367	max_key.type = key_type;
2368
2369	min_key.objectid = inode->i_ino;
2370	min_key.type = key_type;
2371	min_key.offset = min_offset;
2372
2373	path->keep_locks = 1;
2374
2375	ret = btrfs_search_forward(root, &min_key, &max_key,
2376				   path, 0, trans->transid);
2377
2378	/*
2379	 * we didn't find anything from this transaction, see if there
2380	 * is anything at all
2381	 */
2382	if (ret != 0 || min_key.objectid != inode->i_ino ||
2383	    min_key.type != key_type) {
2384		min_key.objectid = inode->i_ino;
2385		min_key.type = key_type;
2386		min_key.offset = (u64)-1;
2387		btrfs_release_path(root, path);
2388		ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2389		if (ret < 0) {
2390			btrfs_release_path(root, path);
2391			return ret;
2392		}
2393		ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2394
2395		/* if ret == 0 there are items for this type,
2396		 * create a range to tell us the last key of this type.
2397		 * otherwise, there are no items in this directory after
2398		 * *min_offset, and we create a range to indicate that.
2399		 */
2400		if (ret == 0) {
2401			struct btrfs_key tmp;
2402			btrfs_item_key_to_cpu(path->nodes[0], &tmp,
2403					      path->slots[0]);
2404			if (key_type == tmp.type)
2405				first_offset = max(min_offset, tmp.offset) + 1;
2406		}
2407		goto done;
2408	}
2409
2410	/* go backward to find any previous key */
2411	ret = btrfs_previous_item(root, path, inode->i_ino, key_type);
2412	if (ret == 0) {
2413		struct btrfs_key tmp;
2414		btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2415		if (key_type == tmp.type) {
2416			first_offset = tmp.offset;
2417			ret = overwrite_item(trans, log, dst_path,
2418					     path->nodes[0], path->slots[0],
2419					     &tmp);
2420			if (ret) {
2421				err = ret;
2422				goto done;
2423			}
2424		}
2425	}
2426	btrfs_release_path(root, path);
2427
2428	/* find the first key from this transaction again */
2429	ret = btrfs_search_slot(NULL, root, &min_key, path, 0, 0);
2430	if (ret != 0) {
2431		WARN_ON(1);
2432		goto done;
2433	}
2434
2435	/*
2436	 * we have a block from this transaction, log every item in it
2437	 * from our directory
2438	 */
2439	while (1) {
2440		struct btrfs_key tmp;
2441		src = path->nodes[0];
2442		nritems = btrfs_header_nritems(src);
2443		for (i = path->slots[0]; i < nritems; i++) {
2444			btrfs_item_key_to_cpu(src, &min_key, i);
2445
2446			if (min_key.objectid != inode->i_ino ||
2447			    min_key.type != key_type)
2448				goto done;
2449			ret = overwrite_item(trans, log, dst_path, src, i,
2450					     &min_key);
2451			if (ret) {
2452				err = ret;
2453				goto done;
2454			}
2455		}
2456		path->slots[0] = nritems;
2457
2458		/*
2459		 * look ahead to the next item and see if it is also
2460		 * from this directory and from this transaction
2461		 */
2462		ret = btrfs_next_leaf(root, path);
2463		if (ret == 1) {
2464			last_offset = (u64)-1;
2465			goto done;
2466		}
2467		btrfs_item_key_to_cpu(path->nodes[0], &tmp, path->slots[0]);
2468		if (tmp.objectid != inode->i_ino || tmp.type != key_type) {
2469			last_offset = (u64)-1;
2470			goto done;
2471		}
2472		if (btrfs_header_generation(path->nodes[0]) != trans->transid) {
2473			ret = overwrite_item(trans, log, dst_path,
2474					     path->nodes[0], path->slots[0],
2475					     &tmp);
2476			if (ret)
2477				err = ret;
2478			else
2479				last_offset = tmp.offset;
2480			goto done;
2481		}
2482	}
2483done:
2484	btrfs_release_path(root, path);
2485	btrfs_release_path(log, dst_path);
2486
2487	if (err == 0) {
2488		*last_offset_ret = last_offset;
2489		/*
2490		 * insert the log range keys to indicate where the log
2491		 * is valid
2492		 */
2493		ret = insert_dir_log_key(trans, log, path, key_type,
2494					 inode->i_ino, first_offset,
2495					 last_offset);
2496		if (ret)
2497			err = ret;
2498	}
2499	return err;
2500}
2501
2502/*
2503 * logging directories is very similar to logging inodes, We find all the items
2504 * from the current transaction and write them to the log.
2505 *
2506 * The recovery code scans the directory in the subvolume, and if it finds a
2507 * key in the range logged that is not present in the log tree, then it means
2508 * that dir entry was unlinked during the transaction.
2509 *
2510 * In order for that scan to work, we must include one key smaller than
2511 * the smallest logged by this transaction and one key larger than the largest
2512 * key logged by this transaction.
2513 */
2514static noinline int log_directory_changes(struct btrfs_trans_handle *trans,
2515			  struct btrfs_root *root, struct inode *inode,
2516			  struct btrfs_path *path,
2517			  struct btrfs_path *dst_path)
2518{
2519	u64 min_key;
2520	u64 max_key;
2521	int ret;
2522	int key_type = BTRFS_DIR_ITEM_KEY;
2523
2524again:
2525	min_key = 0;
2526	max_key = 0;
2527	while (1) {
2528		ret = log_dir_items(trans, root, inode, path,
2529				    dst_path, key_type, min_key,
2530				    &max_key);
2531		if (ret)
2532			return ret;
2533		if (max_key == (u64)-1)
2534			break;
2535		min_key = max_key + 1;
2536	}
2537
2538	if (key_type == BTRFS_DIR_ITEM_KEY) {
2539		key_type = BTRFS_DIR_INDEX_KEY;
2540		goto again;
2541	}
2542	return 0;
2543}
2544
2545/*
2546 * a helper function to drop items from the log before we relog an
2547 * inode.  max_key_type indicates the highest item type to remove.
2548 * This cannot be run for file data extents because it does not
2549 * free the extents they point to.
2550 */
2551static int drop_objectid_items(struct btrfs_trans_handle *trans,
2552				  struct btrfs_root *log,
2553				  struct btrfs_path *path,
2554				  u64 objectid, int max_key_type)
2555{
2556	int ret;
2557	struct btrfs_key key;
2558	struct btrfs_key found_key;
2559
2560	key.objectid = objectid;
2561	key.type = max_key_type;
2562	key.offset = (u64)-1;
2563
2564	while (1) {
2565		ret = btrfs_search_slot(trans, log, &key, path, -1, 1);
2566		BUG_ON(ret == 0);
2567		if (ret < 0)
2568			break;
2569
2570		if (path->slots[0] == 0)
2571			break;
2572
2573		path->slots[0]--;
2574		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2575				      path->slots[0]);
2576
2577		if (found_key.objectid != objectid)
2578			break;
2579
2580		ret = btrfs_del_item(trans, log, path);
2581		BUG_ON(ret);
2582		btrfs_release_path(log, path);
2583	}
2584	btrfs_release_path(log, path);
2585	return ret;
2586}
2587
2588static noinline int copy_items(struct btrfs_trans_handle *trans,
2589			       struct btrfs_root *log,
2590			       struct btrfs_path *dst_path,
2591			       struct extent_buffer *src,
2592			       int start_slot, int nr, int inode_only)
2593{
2594	unsigned long src_offset;
2595	unsigned long dst_offset;
2596	struct btrfs_file_extent_item *extent;
2597	struct btrfs_inode_item *inode_item;
2598	int ret;
2599	struct btrfs_key *ins_keys;
2600	u32 *ins_sizes;
2601	char *ins_data;
2602	int i;
2603	struct list_head ordered_sums;
2604
2605	INIT_LIST_HEAD(&ordered_sums);
2606
2607	ins_data = kmalloc(nr * sizeof(struct btrfs_key) +
2608			   nr * sizeof(u32), GFP_NOFS);
2609	ins_sizes = (u32 *)ins_data;
2610	ins_keys = (struct btrfs_key *)(ins_data + nr * sizeof(u32));
2611
2612	for (i = 0; i < nr; i++) {
2613		ins_sizes[i] = btrfs_item_size_nr(src, i + start_slot);
2614		btrfs_item_key_to_cpu(src, ins_keys + i, i + start_slot);
2615	}
2616	ret = btrfs_insert_empty_items(trans, log, dst_path,
2617				       ins_keys, ins_sizes, nr);
2618	if (ret) {
2619		kfree(ins_data);
2620		return ret;
2621	}
2622
2623	for (i = 0; i < nr; i++, dst_path->slots[0]++) {
2624		dst_offset = btrfs_item_ptr_offset(dst_path->nodes[0],
2625						   dst_path->slots[0]);
2626
2627		src_offset = btrfs_item_ptr_offset(src, start_slot + i);
2628
2629		copy_extent_buffer(dst_path->nodes[0], src, dst_offset,
2630				   src_offset, ins_sizes[i]);
2631
2632		if (inode_only == LOG_INODE_EXISTS &&
2633		    ins_keys[i].type == BTRFS_INODE_ITEM_KEY) {
2634			inode_item = btrfs_item_ptr(dst_path->nodes[0],
2635						    dst_path->slots[0],
2636						    struct btrfs_inode_item);
2637			btrfs_set_inode_size(dst_path->nodes[0], inode_item, 0);
2638
2639			/* set the generation to zero so the recover code
2640			 * can tell the difference between an logging
2641			 * just to say 'this inode exists' and a logging
2642			 * to say 'update this inode with these values'
2643			 */
2644			btrfs_set_inode_generation(dst_path->nodes[0],
2645						   inode_item, 0);
2646		}
2647		/* take a reference on file data extents so that truncates
2648		 * or deletes of this inode don't have to relog the inode
2649		 * again
2650		 */
2651		if (btrfs_key_type(ins_keys + i) == BTRFS_EXTENT_DATA_KEY) {
2652			int found_type;
2653			extent = btrfs_item_ptr(src, start_slot + i,
2654						struct btrfs_file_extent_item);
2655
2656			found_type = btrfs_file_extent_type(src, extent);
2657			if (found_type == BTRFS_FILE_EXTENT_REG ||
2658			    found_type == BTRFS_FILE_EXTENT_PREALLOC) {
2659				u64 ds, dl, cs, cl;
2660				ds = btrfs_file_extent_disk_bytenr(src,
2661								extent);
2662				/* ds == 0 is a hole */
2663				if (ds == 0)
2664					continue;
2665
2666				dl = btrfs_file_extent_disk_num_bytes(src,
2667								extent);
2668				cs = btrfs_file_extent_offset(src, extent);
2669				cl = btrfs_file_extent_num_bytes(src,
2670								extent);
2671				if (btrfs_file_extent_compression(src,
2672								  extent)) {
2673					cs = 0;
2674					cl = dl;
2675				}
2676
2677				ret = btrfs_lookup_csums_range(
2678						log->fs_info->csum_root,
2679						ds + cs, ds + cs + cl - 1,
2680						&ordered_sums);
2681				BUG_ON(ret);
2682			}
2683		}
2684	}
2685
2686	btrfs_mark_buffer_dirty(dst_path->nodes[0]);
2687	btrfs_release_path(log, dst_path);
2688	kfree(ins_data);
2689
2690	/*
2691	 * we have to do this after the loop above to avoid changing the
2692	 * log tree while trying to change the log tree.
2693	 */
2694	ret = 0;
2695	while (!list_empty(&ordered_sums)) {
2696		struct btrfs_ordered_sum *sums = list_entry(ordered_sums.next,
2697						   struct btrfs_ordered_sum,
2698						   list);
2699		if (!ret)
2700			ret = btrfs_csum_file_blocks(trans, log, sums);
2701		list_del(&sums->list);
2702		kfree(sums);
2703	}
2704	return ret;
2705}
2706
2707/* log a single inode in the tree log.
2708 * At least one parent directory for this inode must exist in the tree
2709 * or be logged already.
2710 *
2711 * Any items from this inode changed by the current transaction are copied
2712 * to the log tree.  An extra reference is taken on any extents in this
2713 * file, allowing us to avoid a whole pile of corner cases around logging
2714 * blocks that have been removed from the tree.
2715 *
2716 * See LOG_INODE_ALL and related defines for a description of what inode_only
2717 * does.
2718 *
2719 * This handles both files and directories.
2720 */
2721static int btrfs_log_inode(struct btrfs_trans_handle *trans,
2722			     struct btrfs_root *root, struct inode *inode,
2723			     int inode_only)
2724{
2725	struct btrfs_path *path;
2726	struct btrfs_path *dst_path;
2727	struct btrfs_key min_key;
2728	struct btrfs_key max_key;
2729	struct btrfs_root *log = root->log_root;
2730	struct extent_buffer *src = NULL;
2731	u32 size;
2732	int err = 0;
2733	int ret;
2734	int nritems;
2735	int ins_start_slot = 0;
2736	int ins_nr;
2737
2738	log = root->log_root;
2739
2740	path = btrfs_alloc_path();
2741	dst_path = btrfs_alloc_path();
2742
2743	min_key.objectid = inode->i_ino;
2744	min_key.type = BTRFS_INODE_ITEM_KEY;
2745	min_key.offset = 0;
2746
2747	max_key.objectid = inode->i_ino;
2748
2749	/* today the code can only do partial logging of directories */
2750	if (!S_ISDIR(inode->i_mode))
2751	    inode_only = LOG_INODE_ALL;
2752
2753	if (inode_only == LOG_INODE_EXISTS || S_ISDIR(inode->i_mode))
2754		max_key.type = BTRFS_XATTR_ITEM_KEY;
2755	else
2756		max_key.type = (u8)-1;
2757	max_key.offset = (u64)-1;
2758
2759	mutex_lock(&BTRFS_I(inode)->log_mutex);
2760
2761	/*
2762	 * a brute force approach to making sure we get the most uptodate
2763	 * copies of everything.
2764	 */
2765	if (S_ISDIR(inode->i_mode)) {
2766		int max_key_type = BTRFS_DIR_LOG_INDEX_KEY;
2767
2768		if (inode_only == LOG_INODE_EXISTS)
2769			max_key_type = BTRFS_XATTR_ITEM_KEY;
2770		ret = drop_objectid_items(trans, log, path,
2771					  inode->i_ino, max_key_type);
2772	} else {
2773		ret = btrfs_truncate_inode_items(trans, log, inode, 0, 0);
2774	}
2775	if (ret) {
2776		err = ret;
2777		goto out_unlock;
2778	}
2779	path->keep_locks = 1;
2780
2781	while (1) {
2782		ins_nr = 0;
2783		ret = btrfs_search_forward(root, &min_key, &max_key,
2784					   path, 0, trans->transid);
2785		if (ret != 0)
2786			break;
2787again:
2788		/* note, ins_nr might be > 0 here, cleanup outside the loop */
2789		if (min_key.objectid != inode->i_ino)
2790			break;
2791		if (min_key.type > max_key.type)
2792			break;
2793
2794		src = path->nodes[0];
2795		size = btrfs_item_size_nr(src, path->slots[0]);
2796		if (ins_nr && ins_start_slot + ins_nr == path->slots[0]) {
2797			ins_nr++;
2798			goto next_slot;
2799		} else if (!ins_nr) {
2800			ins_start_slot = path->slots[0];
2801			ins_nr = 1;
2802			goto next_slot;
2803		}
2804
2805		ret = copy_items(trans, log, dst_path, src, ins_start_slot,
2806				 ins_nr, inode_only);
2807		if (ret) {
2808			err = ret;
2809			goto out_unlock;
2810		}
2811		ins_nr = 1;
2812		ins_start_slot = path->slots[0];
2813next_slot:
2814
2815		nritems = btrfs_header_nritems(path->nodes[0]);
2816		path->slots[0]++;
2817		if (path->slots[0] < nritems) {
2818			btrfs_item_key_to_cpu(path->nodes[0], &min_key,
2819					      path->slots[0]);
2820			goto again;
2821		}
2822		if (ins_nr) {
2823			ret = copy_items(trans, log, dst_path, src,
2824					 ins_start_slot,
2825					 ins_nr, inode_only);
2826			if (ret) {
2827				err = ret;
2828				goto out_unlock;
2829			}
2830			ins_nr = 0;
2831		}
2832		btrfs_release_path(root, path);
2833
2834		if (min_key.offset < (u64)-1)
2835			min_key.offset++;
2836		else if (min_key.type < (u8)-1)
2837			min_key.type++;
2838		else if (min_key.objectid < (u64)-1)
2839			min_key.objectid++;
2840		else
2841			break;
2842	}
2843	if (ins_nr) {
2844		ret = copy_items(trans, log, dst_path, src,
2845				 ins_start_slot,
2846				 ins_nr, inode_only);
2847		if (ret) {
2848			err = ret;
2849			goto out_unlock;
2850		}
2851		ins_nr = 0;
2852	}
2853	WARN_ON(ins_nr);
2854	if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
2855		btrfs_release_path(root, path);
2856		btrfs_release_path(log, dst_path);
2857		ret = log_directory_changes(trans, root, inode, path, dst_path);
2858		if (ret) {
2859			err = ret;
2860			goto out_unlock;
2861		}
2862	}
2863	BTRFS_I(inode)->logged_trans = trans->transid;
2864out_unlock:
2865	mutex_unlock(&BTRFS_I(inode)->log_mutex);
2866
2867	btrfs_free_path(path);
2868	btrfs_free_path(dst_path);
2869	return err;
2870}
2871
2872/*
2873 * follow the dentry parent pointers up the chain and see if any
2874 * of the directories in it require a full commit before they can
2875 * be logged.  Returns zero if nothing special needs to be done or 1 if
2876 * a full commit is required.
2877 */
2878static noinline int check_parent_dirs_for_sync(struct btrfs_trans_handle *trans,
2879					       struct inode *inode,
2880					       struct dentry *parent,
2881					       struct super_block *sb,
2882					       u64 last_committed)
2883{
2884	int ret = 0;
2885	struct btrfs_root *root;
2886
2887	/*
2888	 * for regular files, if its inode is already on disk, we don't
2889	 * have to worry about the parents at all.  This is because
2890	 * we can use the last_unlink_trans field to record renames
2891	 * and other fun in this file.
2892	 */
2893	if (S_ISREG(inode->i_mode) &&
2894	    BTRFS_I(inode)->generation <= last_committed &&
2895	    BTRFS_I(inode)->last_unlink_trans <= last_committed)
2896			goto out;
2897
2898	if (!S_ISDIR(inode->i_mode)) {
2899		if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2900			goto out;
2901		inode = parent->d_inode;
2902	}
2903
2904	while (1) {
2905		BTRFS_I(inode)->logged_trans = trans->transid;
2906		smp_mb();
2907
2908		if (BTRFS_I(inode)->last_unlink_trans > last_committed) {
2909			root = BTRFS_I(inode)->root;
2910
2911			/*
2912			 * make sure any commits to the log are forced
2913			 * to be full commits
2914			 */
2915			root->fs_info->last_trans_log_full_commit =
2916				trans->transid;
2917			ret = 1;
2918			break;
2919		}
2920
2921		if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
2922			break;
2923
2924		if (IS_ROOT(parent))
2925			break;
2926
2927		parent = parent->d_parent;
2928		inode = parent->d_inode;
2929
2930	}
2931out:
2932	return ret;
2933}
2934
2935static int inode_in_log(struct btrfs_trans_handle *trans,
2936		 struct inode *inode)
2937{
2938	struct btrfs_root *root = BTRFS_I(inode)->root;
2939	int ret = 0;
2940
2941	mutex_lock(&root->log_mutex);
2942	if (BTRFS_I(inode)->logged_trans == trans->transid &&
2943	    BTRFS_I(inode)->last_sub_trans <= root->last_log_commit)
2944		ret = 1;
2945	mutex_unlock(&root->log_mutex);
2946	return ret;
2947}
2948
2949
2950/*
2951 * helper function around btrfs_log_inode to make sure newly created
2952 * parent directories also end up in the log.  A minimal inode and backref
2953 * only logging is done of any parent directories that are older than
2954 * the last committed transaction
2955 */
2956int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
2957		    struct btrfs_root *root, struct inode *inode,
2958		    struct dentry *parent, int exists_only)
2959{
2960	int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
2961	struct super_block *sb;
2962	int ret = 0;
2963	u64 last_committed = root->fs_info->last_trans_committed;
2964
2965	sb = inode->i_sb;
2966
2967	if (btrfs_test_opt(root, NOTREELOG)) {
2968		ret = 1;
2969		goto end_no_trans;
2970	}
2971
2972	if (root->fs_info->last_trans_log_full_commit >
2973	    root->fs_info->last_trans_committed) {
2974		ret = 1;
2975		goto end_no_trans;
2976	}
2977
2978	if (root != BTRFS_I(inode)->root ||
2979	    btrfs_root_refs(&root->root_item) == 0) {
2980		ret = 1;
2981		goto end_no_trans;
2982	}
2983
2984	ret = check_parent_dirs_for_sync(trans, inode, parent,
2985					 sb, last_committed);
2986	if (ret)
2987		goto end_no_trans;
2988
2989	if (inode_in_log(trans, inode)) {
2990		ret = BTRFS_NO_LOG_SYNC;
2991		goto end_no_trans;
2992	}
2993
2994	ret = start_log_trans(trans, root);
2995	if (ret)
2996		goto end_trans;
2997
2998	ret = btrfs_log_inode(trans, root, inode, inode_only);
2999	if (ret)
3000		goto end_trans;
3001
3002	/*
3003	 * for regular files, if its inode is already on disk, we don't
3004	 * have to worry about the parents at all.  This is because
3005	 * we can use the last_unlink_trans field to record renames
3006	 * and other fun in this file.
3007	 */
3008	if (S_ISREG(inode->i_mode) &&
3009	    BTRFS_I(inode)->generation <= last_committed &&
3010	    BTRFS_I(inode)->last_unlink_trans <= last_committed) {
3011		ret = 0;
3012		goto end_trans;
3013	}
3014
3015	inode_only = LOG_INODE_EXISTS;
3016	while (1) {
3017		if (!parent || !parent->d_inode || sb != parent->d_inode->i_sb)
3018			break;
3019
3020		inode = parent->d_inode;
3021		if (root != BTRFS_I(inode)->root)
3022			break;
3023
3024		if (BTRFS_I(inode)->generation >
3025		    root->fs_info->last_trans_committed) {
3026			ret = btrfs_log_inode(trans, root, inode, inode_only);
3027			if (ret)
3028				goto end_trans;
3029		}
3030		if (IS_ROOT(parent))
3031			break;
3032
3033		parent = parent->d_parent;
3034	}
3035	ret = 0;
3036end_trans:
3037	if (ret < 0) {
3038		BUG_ON(ret != -ENOSPC);
3039		root->fs_info->last_trans_log_full_commit = trans->transid;
3040		ret = 1;
3041	}
3042	btrfs_end_log_trans(root);
3043end_no_trans:
3044	return ret;
3045}
3046
3047/*
3048 * it is not safe to log dentry if the chunk root has added new
3049 * chunks.  This returns 0 if the dentry was logged, and 1 otherwise.
3050 * If this returns 1, you must commit the transaction to safely get your
3051 * data on disk.
3052 */
3053int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
3054			  struct btrfs_root *root, struct dentry *dentry)
3055{
3056	return btrfs_log_inode_parent(trans, root, dentry->d_inode,
3057				      dentry->d_parent, 0);
3058}
3059
3060/*
3061 * should be called during mount to recover any replay any log trees
3062 * from the FS
3063 */
3064int btrfs_recover_log_trees(struct btrfs_root *log_root_tree)
3065{
3066	int ret;
3067	struct btrfs_path *path;
3068	struct btrfs_trans_handle *trans;
3069	struct btrfs_key key;
3070	struct btrfs_key found_key;
3071	struct btrfs_key tmp_key;
3072	struct btrfs_root *log;
3073	struct btrfs_fs_info *fs_info = log_root_tree->fs_info;
3074	struct walk_control wc = {
3075		.process_func = process_one_buffer,
3076		.stage = 0,
3077	};
3078
3079	fs_info->log_root_recovering = 1;
3080	path = btrfs_alloc_path();
3081	BUG_ON(!path);
3082
3083	trans = btrfs_start_transaction(fs_info->tree_root, 0);
3084
3085	wc.trans = trans;
3086	wc.pin = 1;
3087
3088	walk_log_tree(trans, log_root_tree, &wc);
3089
3090again:
3091	key.objectid = BTRFS_TREE_LOG_OBJECTID;
3092	key.offset = (u64)-1;
3093	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
3094
3095	while (1) {
3096		ret = btrfs_search_slot(NULL, log_root_tree, &key, path, 0, 0);
3097		if (ret < 0)
3098			break;
3099		if (ret > 0) {
3100			if (path->slots[0] == 0)
3101				break;
3102			path->slots[0]--;
3103		}
3104		btrfs_item_key_to_cpu(path->nodes[0], &found_key,
3105				      path->slots[0]);
3106		btrfs_release_path(log_root_tree, path);
3107		if (found_key.objectid != BTRFS_TREE_LOG_OBJECTID)
3108			break;
3109
3110		log = btrfs_read_fs_root_no_radix(log_root_tree,
3111						  &found_key);
3112		BUG_ON(!log);
3113
3114
3115		tmp_key.objectid = found_key.offset;
3116		tmp_key.type = BTRFS_ROOT_ITEM_KEY;
3117		tmp_key.offset = (u64)-1;
3118
3119		wc.replay_dest = btrfs_read_fs_root_no_name(fs_info, &tmp_key);
3120		BUG_ON(!wc.replay_dest);
3121
3122		wc.replay_dest->log_root = log;
3123		btrfs_record_root_in_trans(trans, wc.replay_dest);
3124		ret = walk_log_tree(trans, log, &wc);
3125		BUG_ON(ret);
3126
3127		if (wc.stage == LOG_WALK_REPLAY_ALL) {
3128			ret = fixup_inode_link_counts(trans, wc.replay_dest,
3129						      path);
3130			BUG_ON(ret);
3131		}
3132
3133		key.offset = found_key.offset - 1;
3134		wc.replay_dest->log_root = NULL;
3135		free_extent_buffer(log->node);
3136		free_extent_buffer(log->commit_root);
3137		kfree(log);
3138
3139		if (found_key.offset == 0)
3140			break;
3141	}
3142	btrfs_release_path(log_root_tree, path);
3143
3144	/* step one is to pin it all, step two is to replay just inodes */
3145	if (wc.pin) {
3146		wc.pin = 0;
3147		wc.process_func = replay_one_buffer;
3148		wc.stage = LOG_WALK_REPLAY_INODES;
3149		goto again;
3150	}
3151	/* step three is to replay everything */
3152	if (wc.stage < LOG_WALK_REPLAY_ALL) {
3153		wc.stage++;
3154		goto again;
3155	}
3156
3157	btrfs_free_path(path);
3158
3159	free_extent_buffer(log_root_tree->node);
3160	log_root_tree->log_root = NULL;
3161	fs_info->log_root_recovering = 0;
3162
3163	/* step 4: commit the transaction, which also unpins the blocks */
3164	btrfs_commit_transaction(trans, fs_info->tree_root);
3165
3166	kfree(log_root_tree);
3167	return 0;
3168}
3169
3170/*
3171 * there are some corner cases where we want to force a full
3172 * commit instead of allowing a directory to be logged.
3173 *
3174 * They revolve around files there were unlinked from the directory, and
3175 * this function updates the parent directory so that a full commit is
3176 * properly done if it is fsync'd later after the unlinks are done.
3177 */
3178void btrfs_record_unlink_dir(struct btrfs_trans_handle *trans,
3179			     struct inode *dir, struct inode *inode,
3180			     int for_rename)
3181{
3182	/*
3183	 * when we're logging a file, if it hasn't been renamed
3184	 * or unlinked, and its inode is fully committed on disk,
3185	 * we don't have to worry about walking up the directory chain
3186	 * to log its parents.
3187	 *
3188	 * So, we use the last_unlink_trans field to put this transid
3189	 * into the file.  When the file is logged we check it and
3190	 * don't log the parents if the file is fully on disk.
3191	 */
3192	if (S_ISREG(inode->i_mode))
3193		BTRFS_I(inode)->last_unlink_trans = trans->transid;
3194
3195	/*
3196	 * if this directory was already logged any new
3197	 * names for this file/dir will get recorded
3198	 */
3199	smp_mb();
3200	if (BTRFS_I(dir)->logged_trans == trans->transid)
3201		return;
3202
3203	/*
3204	 * if the inode we're about to unlink was logged,
3205	 * the log will be properly updated for any new names
3206	 */
3207	if (BTRFS_I(inode)->logged_trans == trans->transid)
3208		return;
3209
3210	/*
3211	 * when renaming files across directories, if the directory
3212	 * there we're unlinking from gets fsync'd later on, there's
3213	 * no way to find the destination directory later and fsync it
3214	 * properly.  So, we have to be conservative and force commits
3215	 * so the new name gets discovered.
3216	 */
3217	if (for_rename)
3218		goto record;
3219
3220	/* we can safely do the unlink without any special recording */
3221	return;
3222
3223record:
3224	BTRFS_I(dir)->last_unlink_trans = trans->transid;
3225}
3226
3227/*
3228 * Call this after adding a new name for a file and it will properly
3229 * update the log to reflect the new name.
3230 *
3231 * It will return zero if all goes well, and it will return 1 if a
3232 * full transaction commit is required.
3233 */
3234int btrfs_log_new_name(struct btrfs_trans_handle *trans,
3235			struct inode *inode, struct inode *old_dir,
3236			struct dentry *parent)
3237{
3238	struct btrfs_root * root = BTRFS_I(inode)->root;
3239
3240	/*
3241	 * this will force the logging code to walk the dentry chain
3242	 * up for the file
3243	 */
3244	if (S_ISREG(inode->i_mode))
3245		BTRFS_I(inode)->last_unlink_trans = trans->transid;
3246
3247	/*
3248	 * if this inode hasn't been logged and directory we're renaming it
3249	 * from hasn't been logged, we don't need to log it
3250	 */
3251	if (BTRFS_I(inode)->logged_trans <=
3252	    root->fs_info->last_trans_committed &&
3253	    (!old_dir || BTRFS_I(old_dir)->logged_trans <=
3254		    root->fs_info->last_trans_committed))
3255		return 0;
3256
3257	return btrfs_log_inode_parent(trans, root, inode, parent, 1);
3258}
3259