Btrfs: fix that repair code is spuriously executed for transid failures
[linux-3.10.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 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/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48
49 static struct extent_io_ops btree_extent_io_ops;
50 static void end_workqueue_fn(struct btrfs_work *work);
51 static void free_fs_root(struct btrfs_root *root);
52 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53                                     int read_only);
54 static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57                                       struct btrfs_root *root);
58 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61                                         struct extent_io_tree *dirty_pages,
62                                         int mark);
63 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64                                        struct extent_io_tree *pinned_extents);
65
66 /*
67  * end_io_wq structs are used to do processing in task context when an IO is
68  * complete.  This is used during reads to verify checksums, and it is used
69  * by writes to insert metadata for new file extents after IO is complete.
70  */
71 struct end_io_wq {
72         struct bio *bio;
73         bio_end_io_t *end_io;
74         void *private;
75         struct btrfs_fs_info *info;
76         int error;
77         int metadata;
78         struct list_head list;
79         struct btrfs_work work;
80 };
81
82 /*
83  * async submit bios are used to offload expensive checksumming
84  * onto the worker threads.  They checksum file and metadata bios
85  * just before they are sent down the IO stack.
86  */
87 struct async_submit_bio {
88         struct inode *inode;
89         struct bio *bio;
90         struct list_head list;
91         extent_submit_bio_hook_t *submit_bio_start;
92         extent_submit_bio_hook_t *submit_bio_done;
93         int rw;
94         int mirror_num;
95         unsigned long bio_flags;
96         /*
97          * bio_offset is optional, can be used if the pages in the bio
98          * can't tell us where in the file the bio should go
99          */
100         u64 bio_offset;
101         struct btrfs_work work;
102         int error;
103 };
104
105 /*
106  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
107  * eb, the lockdep key is determined by the btrfs_root it belongs to and
108  * the level the eb occupies in the tree.
109  *
110  * Different roots are used for different purposes and may nest inside each
111  * other and they require separate keysets.  As lockdep keys should be
112  * static, assign keysets according to the purpose of the root as indicated
113  * by btrfs_root->objectid.  This ensures that all special purpose roots
114  * have separate keysets.
115  *
116  * Lock-nesting across peer nodes is always done with the immediate parent
117  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
118  * subclass to avoid triggering lockdep warning in such cases.
119  *
120  * The key is set by the readpage_end_io_hook after the buffer has passed
121  * csum validation but before the pages are unlocked.  It is also set by
122  * btrfs_init_new_buffer on freshly allocated blocks.
123  *
124  * We also add a check to make sure the highest level of the tree is the
125  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
126  * needs update as well.
127  */
128 #ifdef CONFIG_DEBUG_LOCK_ALLOC
129 # if BTRFS_MAX_LEVEL != 8
130 #  error
131 # endif
132
133 static struct btrfs_lockdep_keyset {
134         u64                     id;             /* root objectid */
135         const char              *name_stem;     /* lock name stem */
136         char                    names[BTRFS_MAX_LEVEL + 1][20];
137         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
138 } btrfs_lockdep_keysets[] = {
139         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
140         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
141         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
142         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
143         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
144         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
145         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
146         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
147         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
148         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
149         { .id = 0,                              .name_stem = "tree"     },
150 };
151
152 void __init btrfs_init_lockdep(void)
153 {
154         int i, j;
155
156         /* initialize lockdep class names */
157         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159
160                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161                         snprintf(ks->names[j], sizeof(ks->names[j]),
162                                  "btrfs-%s-%02d", ks->name_stem, j);
163         }
164 }
165
166 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167                                     int level)
168 {
169         struct btrfs_lockdep_keyset *ks;
170
171         BUG_ON(level >= ARRAY_SIZE(ks->keys));
172
173         /* find the matching keyset, id 0 is the default entry */
174         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175                 if (ks->id == objectid)
176                         break;
177
178         lockdep_set_class_and_name(&eb->lock,
179                                    &ks->keys[level], ks->names[level]);
180 }
181
182 #endif
183
184 /*
185  * extents on the btree inode are pretty simple, there's one extent
186  * that covers the entire device
187  */
188 static struct extent_map *btree_get_extent(struct inode *inode,
189                 struct page *page, size_t pg_offset, u64 start, u64 len,
190                 int create)
191 {
192         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193         struct extent_map *em;
194         int ret;
195
196         read_lock(&em_tree->lock);
197         em = lookup_extent_mapping(em_tree, start, len);
198         if (em) {
199                 em->bdev =
200                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201                 read_unlock(&em_tree->lock);
202                 goto out;
203         }
204         read_unlock(&em_tree->lock);
205
206         em = alloc_extent_map();
207         if (!em) {
208                 em = ERR_PTR(-ENOMEM);
209                 goto out;
210         }
211         em->start = 0;
212         em->len = (u64)-1;
213         em->block_len = (u64)-1;
214         em->block_start = 0;
215         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216
217         write_lock(&em_tree->lock);
218         ret = add_extent_mapping(em_tree, em);
219         if (ret == -EEXIST) {
220                 u64 failed_start = em->start;
221                 u64 failed_len = em->len;
222
223                 free_extent_map(em);
224                 em = lookup_extent_mapping(em_tree, start, len);
225                 if (em) {
226                         ret = 0;
227                 } else {
228                         em = lookup_extent_mapping(em_tree, failed_start,
229                                                    failed_len);
230                         ret = -EIO;
231                 }
232         } else if (ret) {
233                 free_extent_map(em);
234                 em = NULL;
235         }
236         write_unlock(&em_tree->lock);
237
238         if (ret)
239                 em = ERR_PTR(ret);
240 out:
241         return em;
242 }
243
244 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245 {
246         return crc32c(seed, data, len);
247 }
248
249 void btrfs_csum_final(u32 crc, char *result)
250 {
251         put_unaligned_le32(~crc, result);
252 }
253
254 /*
255  * compute the csum for a btree block, and either verify it or write it
256  * into the csum field of the block.
257  */
258 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259                            int verify)
260 {
261         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262         char *result = NULL;
263         unsigned long len;
264         unsigned long cur_len;
265         unsigned long offset = BTRFS_CSUM_SIZE;
266         char *kaddr;
267         unsigned long map_start;
268         unsigned long map_len;
269         int err;
270         u32 crc = ~(u32)0;
271         unsigned long inline_result;
272
273         len = buf->len - offset;
274         while (len > 0) {
275                 err = map_private_extent_buffer(buf, offset, 32,
276                                         &kaddr, &map_start, &map_len);
277                 if (err)
278                         return 1;
279                 cur_len = min(len, map_len - (offset - map_start));
280                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
281                                       crc, cur_len);
282                 len -= cur_len;
283                 offset += cur_len;
284         }
285         if (csum_size > sizeof(inline_result)) {
286                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287                 if (!result)
288                         return 1;
289         } else {
290                 result = (char *)&inline_result;
291         }
292
293         btrfs_csum_final(crc, result);
294
295         if (verify) {
296                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297                         u32 val;
298                         u32 found = 0;
299                         memcpy(&found, result, csum_size);
300
301                         read_extent_buffer(buf, &val, 0, csum_size);
302                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303                                        "failed on %llu wanted %X found %X "
304                                        "level %d\n",
305                                        root->fs_info->sb->s_id,
306                                        (unsigned long long)buf->start, val, found,
307                                        btrfs_header_level(buf));
308                         if (result != (char *)&inline_result)
309                                 kfree(result);
310                         return 1;
311                 }
312         } else {
313                 write_extent_buffer(buf, result, 0, csum_size);
314         }
315         if (result != (char *)&inline_result)
316                 kfree(result);
317         return 0;
318 }
319
320 /*
321  * we can't consider a given block up to date unless the transid of the
322  * block matches the transid in the parent node's pointer.  This is how we
323  * detect blocks that either didn't get written at all or got written
324  * in the wrong place.
325  */
326 static int verify_parent_transid(struct extent_io_tree *io_tree,
327                                  struct extent_buffer *eb, u64 parent_transid,
328                                  int atomic)
329 {
330         struct extent_state *cached_state = NULL;
331         int ret;
332
333         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334                 return 0;
335
336         if (atomic)
337                 return -EAGAIN;
338
339         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340                          0, &cached_state);
341         if (extent_buffer_uptodate(eb) &&
342             btrfs_header_generation(eb) == parent_transid) {
343                 ret = 0;
344                 goto out;
345         }
346         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347                        "found %llu\n",
348                        (unsigned long long)eb->start,
349                        (unsigned long long)parent_transid,
350                        (unsigned long long)btrfs_header_generation(eb));
351         ret = 1;
352         clear_extent_buffer_uptodate(eb);
353 out:
354         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355                              &cached_state, GFP_NOFS);
356         return ret;
357 }
358
359 /*
360  * helper to read a given tree block, doing retries as required when
361  * the checksums don't match and we have alternate mirrors to try.
362  */
363 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
364                                           struct extent_buffer *eb,
365                                           u64 start, u64 parent_transid)
366 {
367         struct extent_io_tree *io_tree;
368         int failed = 0;
369         int ret;
370         int num_copies = 0;
371         int mirror_num = 0;
372         int failed_mirror = 0;
373
374         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
375         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
376         while (1) {
377                 ret = read_extent_buffer_pages(io_tree, eb, start,
378                                                WAIT_COMPLETE,
379                                                btree_get_extent, mirror_num);
380                 if (!ret) {
381                         if (!verify_parent_transid(io_tree, eb,
382                                                    parent_transid, 0))
383                                 break;
384                         else
385                                 ret = -EIO;
386                 }
387
388                 /*
389                  * This buffer's crc is fine, but its contents are corrupted, so
390                  * there is no reason to read the other copies, they won't be
391                  * any less wrong.
392                  */
393                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
394                         break;
395
396                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
397                                               eb->start, eb->len);
398                 if (num_copies == 1)
399                         break;
400
401                 if (!failed_mirror) {
402                         failed = 1;
403                         failed_mirror = eb->read_mirror;
404                 }
405
406                 mirror_num++;
407                 if (mirror_num == failed_mirror)
408                         mirror_num++;
409
410                 if (mirror_num > num_copies)
411                         break;
412         }
413
414         if (failed && !ret && failed_mirror)
415                 repair_eb_io_failure(root, eb, failed_mirror);
416
417         return ret;
418 }
419
420 /*
421  * checksum a dirty tree block before IO.  This has extra checks to make sure
422  * we only fill in the checksum field in the first page of a multi-page block
423  */
424
425 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
426 {
427         struct extent_io_tree *tree;
428         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
429         u64 found_start;
430         struct extent_buffer *eb;
431
432         tree = &BTRFS_I(page->mapping->host)->io_tree;
433
434         eb = (struct extent_buffer *)page->private;
435         if (page != eb->pages[0])
436                 return 0;
437         found_start = btrfs_header_bytenr(eb);
438         if (found_start != start) {
439                 WARN_ON(1);
440                 return 0;
441         }
442         if (eb->pages[0] != page) {
443                 WARN_ON(1);
444                 return 0;
445         }
446         if (!PageUptodate(page)) {
447                 WARN_ON(1);
448                 return 0;
449         }
450         csum_tree_block(root, eb, 0);
451         return 0;
452 }
453
454 static int check_tree_block_fsid(struct btrfs_root *root,
455                                  struct extent_buffer *eb)
456 {
457         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458         u8 fsid[BTRFS_UUID_SIZE];
459         int ret = 1;
460
461         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462                            BTRFS_FSID_SIZE);
463         while (fs_devices) {
464                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465                         ret = 0;
466                         break;
467                 }
468                 fs_devices = fs_devices->seed;
469         }
470         return ret;
471 }
472
473 #define CORRUPT(reason, eb, root, slot)                         \
474         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475                "root=%llu, slot=%d\n", reason,                  \
476                (unsigned long long)btrfs_header_bytenr(eb),     \
477                (unsigned long long)root->objectid, slot)
478
479 static noinline int check_leaf(struct btrfs_root *root,
480                                struct extent_buffer *leaf)
481 {
482         struct btrfs_key key;
483         struct btrfs_key leaf_key;
484         u32 nritems = btrfs_header_nritems(leaf);
485         int slot;
486
487         if (nritems == 0)
488                 return 0;
489
490         /* Check the 0 item */
491         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492             BTRFS_LEAF_DATA_SIZE(root)) {
493                 CORRUPT("invalid item offset size pair", leaf, root, 0);
494                 return -EIO;
495         }
496
497         /*
498          * Check to make sure each items keys are in the correct order and their
499          * offsets make sense.  We only have to loop through nritems-1 because
500          * we check the current slot against the next slot, which verifies the
501          * next slot's offset+size makes sense and that the current's slot
502          * offset is correct.
503          */
504         for (slot = 0; slot < nritems - 1; slot++) {
505                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508                 /* Make sure the keys are in the right order */
509                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510                         CORRUPT("bad key order", leaf, root, slot);
511                         return -EIO;
512                 }
513
514                 /*
515                  * Make sure the offset and ends are right, remember that the
516                  * item data starts at the end of the leaf and grows towards the
517                  * front.
518                  */
519                 if (btrfs_item_offset_nr(leaf, slot) !=
520                         btrfs_item_end_nr(leaf, slot + 1)) {
521                         CORRUPT("slot offset bad", leaf, root, slot);
522                         return -EIO;
523                 }
524
525                 /*
526                  * Check to make sure that we don't point outside of the leaf,
527                  * just incase all the items are consistent to eachother, but
528                  * all point outside of the leaf.
529                  */
530                 if (btrfs_item_end_nr(leaf, slot) >
531                     BTRFS_LEAF_DATA_SIZE(root)) {
532                         CORRUPT("slot end outside of leaf", leaf, root, slot);
533                         return -EIO;
534                 }
535         }
536
537         return 0;
538 }
539
540 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
541                                        struct page *page, int max_walk)
542 {
543         struct extent_buffer *eb;
544         u64 start = page_offset(page);
545         u64 target = start;
546         u64 min_start;
547
548         if (start < max_walk)
549                 min_start = 0;
550         else
551                 min_start = start - max_walk;
552
553         while (start >= min_start) {
554                 eb = find_extent_buffer(tree, start, 0);
555                 if (eb) {
556                         /*
557                          * we found an extent buffer and it contains our page
558                          * horray!
559                          */
560                         if (eb->start <= target &&
561                             eb->start + eb->len > target)
562                                 return eb;
563
564                         /* we found an extent buffer that wasn't for us */
565                         free_extent_buffer(eb);
566                         return NULL;
567                 }
568                 if (start == 0)
569                         break;
570                 start -= PAGE_CACHE_SIZE;
571         }
572         return NULL;
573 }
574
575 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
576                                struct extent_state *state, int mirror)
577 {
578         struct extent_io_tree *tree;
579         u64 found_start;
580         int found_level;
581         struct extent_buffer *eb;
582         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
583         int ret = 0;
584         int reads_done;
585
586         if (!page->private)
587                 goto out;
588
589         tree = &BTRFS_I(page->mapping->host)->io_tree;
590         eb = (struct extent_buffer *)page->private;
591
592         /* the pending IO might have been the only thing that kept this buffer
593          * in memory.  Make sure we have a ref for all this other checks
594          */
595         extent_buffer_get(eb);
596
597         reads_done = atomic_dec_and_test(&eb->io_pages);
598         if (!reads_done)
599                 goto err;
600
601         eb->read_mirror = mirror;
602         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
603                 ret = -EIO;
604                 goto err;
605         }
606
607         found_start = btrfs_header_bytenr(eb);
608         if (found_start != eb->start) {
609                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
610                                "%llu %llu\n",
611                                (unsigned long long)found_start,
612                                (unsigned long long)eb->start);
613                 ret = -EIO;
614                 goto err;
615         }
616         if (check_tree_block_fsid(root, eb)) {
617                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
618                                (unsigned long long)eb->start);
619                 ret = -EIO;
620                 goto err;
621         }
622         found_level = btrfs_header_level(eb);
623
624         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
625                                        eb, found_level);
626
627         ret = csum_tree_block(root, eb, 1);
628         if (ret) {
629                 ret = -EIO;
630                 goto err;
631         }
632
633         /*
634          * If this is a leaf block and it is corrupt, set the corrupt bit so
635          * that we don't try and read the other copies of this block, just
636          * return -EIO.
637          */
638         if (found_level == 0 && check_leaf(root, eb)) {
639                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
640                 ret = -EIO;
641         }
642
643         if (!ret)
644                 set_extent_buffer_uptodate(eb);
645 err:
646         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
647                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
648                 btree_readahead_hook(root, eb, eb->start, ret);
649         }
650
651         if (ret)
652                 clear_extent_buffer_uptodate(eb);
653         free_extent_buffer(eb);
654 out:
655         return ret;
656 }
657
658 static int btree_io_failed_hook(struct page *page, int failed_mirror)
659 {
660         struct extent_buffer *eb;
661         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
662
663         eb = (struct extent_buffer *)page->private;
664         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
665         eb->read_mirror = failed_mirror;
666         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
667                 btree_readahead_hook(root, eb, eb->start, -EIO);
668         return -EIO;    /* we fixed nothing */
669 }
670
671 static void end_workqueue_bio(struct bio *bio, int err)
672 {
673         struct end_io_wq *end_io_wq = bio->bi_private;
674         struct btrfs_fs_info *fs_info;
675
676         fs_info = end_io_wq->info;
677         end_io_wq->error = err;
678         end_io_wq->work.func = end_workqueue_fn;
679         end_io_wq->work.flags = 0;
680
681         if (bio->bi_rw & REQ_WRITE) {
682                 if (end_io_wq->metadata == 1)
683                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
684                                            &end_io_wq->work);
685                 else if (end_io_wq->metadata == 2)
686                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
687                                            &end_io_wq->work);
688                 else
689                         btrfs_queue_worker(&fs_info->endio_write_workers,
690                                            &end_io_wq->work);
691         } else {
692                 if (end_io_wq->metadata)
693                         btrfs_queue_worker(&fs_info->endio_meta_workers,
694                                            &end_io_wq->work);
695                 else
696                         btrfs_queue_worker(&fs_info->endio_workers,
697                                            &end_io_wq->work);
698         }
699 }
700
701 /*
702  * For the metadata arg you want
703  *
704  * 0 - if data
705  * 1 - if normal metadta
706  * 2 - if writing to the free space cache area
707  */
708 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
709                         int metadata)
710 {
711         struct end_io_wq *end_io_wq;
712         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
713         if (!end_io_wq)
714                 return -ENOMEM;
715
716         end_io_wq->private = bio->bi_private;
717         end_io_wq->end_io = bio->bi_end_io;
718         end_io_wq->info = info;
719         end_io_wq->error = 0;
720         end_io_wq->bio = bio;
721         end_io_wq->metadata = metadata;
722
723         bio->bi_private = end_io_wq;
724         bio->bi_end_io = end_workqueue_bio;
725         return 0;
726 }
727
728 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
729 {
730         unsigned long limit = min_t(unsigned long,
731                                     info->workers.max_workers,
732                                     info->fs_devices->open_devices);
733         return 256 * limit;
734 }
735
736 static void run_one_async_start(struct btrfs_work *work)
737 {
738         struct async_submit_bio *async;
739         int ret;
740
741         async = container_of(work, struct  async_submit_bio, work);
742         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
743                                       async->mirror_num, async->bio_flags,
744                                       async->bio_offset);
745         if (ret)
746                 async->error = ret;
747 }
748
749 static void run_one_async_done(struct btrfs_work *work)
750 {
751         struct btrfs_fs_info *fs_info;
752         struct async_submit_bio *async;
753         int limit;
754
755         async = container_of(work, struct  async_submit_bio, work);
756         fs_info = BTRFS_I(async->inode)->root->fs_info;
757
758         limit = btrfs_async_submit_limit(fs_info);
759         limit = limit * 2 / 3;
760
761         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
762             waitqueue_active(&fs_info->async_submit_wait))
763                 wake_up(&fs_info->async_submit_wait);
764
765         /* If an error occured we just want to clean up the bio and move on */
766         if (async->error) {
767                 bio_endio(async->bio, async->error);
768                 return;
769         }
770
771         async->submit_bio_done(async->inode, async->rw, async->bio,
772                                async->mirror_num, async->bio_flags,
773                                async->bio_offset);
774 }
775
776 static void run_one_async_free(struct btrfs_work *work)
777 {
778         struct async_submit_bio *async;
779
780         async = container_of(work, struct  async_submit_bio, work);
781         kfree(async);
782 }
783
784 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
785                         int rw, struct bio *bio, int mirror_num,
786                         unsigned long bio_flags,
787                         u64 bio_offset,
788                         extent_submit_bio_hook_t *submit_bio_start,
789                         extent_submit_bio_hook_t *submit_bio_done)
790 {
791         struct async_submit_bio *async;
792
793         async = kmalloc(sizeof(*async), GFP_NOFS);
794         if (!async)
795                 return -ENOMEM;
796
797         async->inode = inode;
798         async->rw = rw;
799         async->bio = bio;
800         async->mirror_num = mirror_num;
801         async->submit_bio_start = submit_bio_start;
802         async->submit_bio_done = submit_bio_done;
803
804         async->work.func = run_one_async_start;
805         async->work.ordered_func = run_one_async_done;
806         async->work.ordered_free = run_one_async_free;
807
808         async->work.flags = 0;
809         async->bio_flags = bio_flags;
810         async->bio_offset = bio_offset;
811
812         async->error = 0;
813
814         atomic_inc(&fs_info->nr_async_submits);
815
816         if (rw & REQ_SYNC)
817                 btrfs_set_work_high_prio(&async->work);
818
819         btrfs_queue_worker(&fs_info->workers, &async->work);
820
821         while (atomic_read(&fs_info->async_submit_draining) &&
822               atomic_read(&fs_info->nr_async_submits)) {
823                 wait_event(fs_info->async_submit_wait,
824                            (atomic_read(&fs_info->nr_async_submits) == 0));
825         }
826
827         return 0;
828 }
829
830 static int btree_csum_one_bio(struct bio *bio)
831 {
832         struct bio_vec *bvec = bio->bi_io_vec;
833         int bio_index = 0;
834         struct btrfs_root *root;
835         int ret = 0;
836
837         WARN_ON(bio->bi_vcnt <= 0);
838         while (bio_index < bio->bi_vcnt) {
839                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
840                 ret = csum_dirty_buffer(root, bvec->bv_page);
841                 if (ret)
842                         break;
843                 bio_index++;
844                 bvec++;
845         }
846         return ret;
847 }
848
849 static int __btree_submit_bio_start(struct inode *inode, int rw,
850                                     struct bio *bio, int mirror_num,
851                                     unsigned long bio_flags,
852                                     u64 bio_offset)
853 {
854         /*
855          * when we're called for a write, we're already in the async
856          * submission context.  Just jump into btrfs_map_bio
857          */
858         return btree_csum_one_bio(bio);
859 }
860
861 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
862                                  int mirror_num, unsigned long bio_flags,
863                                  u64 bio_offset)
864 {
865         /*
866          * when we're called for a write, we're already in the async
867          * submission context.  Just jump into btrfs_map_bio
868          */
869         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
870 }
871
872 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
873                                  int mirror_num, unsigned long bio_flags,
874                                  u64 bio_offset)
875 {
876         int ret;
877
878         if (!(rw & REQ_WRITE)) {
879
880                 /*
881                  * called for a read, do the setup so that checksum validation
882                  * can happen in the async kernel threads
883                  */
884                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
885                                           bio, 1);
886                 if (ret)
887                         return ret;
888                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
889                                      mirror_num, 0);
890         }
891
892         /*
893          * kthread helpers are used to submit writes so that checksumming
894          * can happen in parallel across all CPUs
895          */
896         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
897                                    inode, rw, bio, mirror_num, 0,
898                                    bio_offset,
899                                    __btree_submit_bio_start,
900                                    __btree_submit_bio_done);
901 }
902
903 #ifdef CONFIG_MIGRATION
904 static int btree_migratepage(struct address_space *mapping,
905                         struct page *newpage, struct page *page,
906                         enum migrate_mode mode)
907 {
908         /*
909          * we can't safely write a btree page from here,
910          * we haven't done the locking hook
911          */
912         if (PageDirty(page))
913                 return -EAGAIN;
914         /*
915          * Buffers may be managed in a filesystem specific way.
916          * We must have no buffers or drop them.
917          */
918         if (page_has_private(page) &&
919             !try_to_release_page(page, GFP_KERNEL))
920                 return -EAGAIN;
921         return migrate_page(mapping, newpage, page, mode);
922 }
923 #endif
924
925
926 static int btree_writepages(struct address_space *mapping,
927                             struct writeback_control *wbc)
928 {
929         struct extent_io_tree *tree;
930         tree = &BTRFS_I(mapping->host)->io_tree;
931         if (wbc->sync_mode == WB_SYNC_NONE) {
932                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
933                 u64 num_dirty;
934                 unsigned long thresh = 32 * 1024 * 1024;
935
936                 if (wbc->for_kupdate)
937                         return 0;
938
939                 /* this is a bit racy, but that's ok */
940                 num_dirty = root->fs_info->dirty_metadata_bytes;
941                 if (num_dirty < thresh)
942                         return 0;
943         }
944         return btree_write_cache_pages(mapping, wbc);
945 }
946
947 static int btree_readpage(struct file *file, struct page *page)
948 {
949         struct extent_io_tree *tree;
950         tree = &BTRFS_I(page->mapping->host)->io_tree;
951         return extent_read_full_page(tree, page, btree_get_extent, 0);
952 }
953
954 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
955 {
956         if (PageWriteback(page) || PageDirty(page))
957                 return 0;
958         /*
959          * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
960          * slab allocation from alloc_extent_state down the callchain where
961          * it'd hit a BUG_ON as those flags are not allowed.
962          */
963         gfp_flags &= ~GFP_SLAB_BUG_MASK;
964
965         return try_release_extent_buffer(page, gfp_flags);
966 }
967
968 static void btree_invalidatepage(struct page *page, unsigned long offset)
969 {
970         struct extent_io_tree *tree;
971         tree = &BTRFS_I(page->mapping->host)->io_tree;
972         extent_invalidatepage(tree, page, offset);
973         btree_releasepage(page, GFP_NOFS);
974         if (PagePrivate(page)) {
975                 printk(KERN_WARNING "btrfs warning page private not zero "
976                        "on page %llu\n", (unsigned long long)page_offset(page));
977                 ClearPagePrivate(page);
978                 set_page_private(page, 0);
979                 page_cache_release(page);
980         }
981 }
982
983 static int btree_set_page_dirty(struct page *page)
984 {
985         struct extent_buffer *eb;
986
987         BUG_ON(!PagePrivate(page));
988         eb = (struct extent_buffer *)page->private;
989         BUG_ON(!eb);
990         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
991         BUG_ON(!atomic_read(&eb->refs));
992         btrfs_assert_tree_locked(eb);
993         return __set_page_dirty_nobuffers(page);
994 }
995
996 static const struct address_space_operations btree_aops = {
997         .readpage       = btree_readpage,
998         .writepages     = btree_writepages,
999         .releasepage    = btree_releasepage,
1000         .invalidatepage = btree_invalidatepage,
1001 #ifdef CONFIG_MIGRATION
1002         .migratepage    = btree_migratepage,
1003 #endif
1004         .set_page_dirty = btree_set_page_dirty,
1005 };
1006
1007 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1008                          u64 parent_transid)
1009 {
1010         struct extent_buffer *buf = NULL;
1011         struct inode *btree_inode = root->fs_info->btree_inode;
1012         int ret = 0;
1013
1014         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1015         if (!buf)
1016                 return 0;
1017         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1018                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1019         free_extent_buffer(buf);
1020         return ret;
1021 }
1022
1023 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1024                          int mirror_num, struct extent_buffer **eb)
1025 {
1026         struct extent_buffer *buf = NULL;
1027         struct inode *btree_inode = root->fs_info->btree_inode;
1028         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1029         int ret;
1030
1031         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1032         if (!buf)
1033                 return 0;
1034
1035         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1036
1037         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1038                                        btree_get_extent, mirror_num);
1039         if (ret) {
1040                 free_extent_buffer(buf);
1041                 return ret;
1042         }
1043
1044         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1045                 free_extent_buffer(buf);
1046                 return -EIO;
1047         } else if (extent_buffer_uptodate(buf)) {
1048                 *eb = buf;
1049         } else {
1050                 free_extent_buffer(buf);
1051         }
1052         return 0;
1053 }
1054
1055 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1056                                             u64 bytenr, u32 blocksize)
1057 {
1058         struct inode *btree_inode = root->fs_info->btree_inode;
1059         struct extent_buffer *eb;
1060         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1061                                 bytenr, blocksize);
1062         return eb;
1063 }
1064
1065 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1066                                                  u64 bytenr, u32 blocksize)
1067 {
1068         struct inode *btree_inode = root->fs_info->btree_inode;
1069         struct extent_buffer *eb;
1070
1071         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1072                                  bytenr, blocksize);
1073         return eb;
1074 }
1075
1076
1077 int btrfs_write_tree_block(struct extent_buffer *buf)
1078 {
1079         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1080                                         buf->start + buf->len - 1);
1081 }
1082
1083 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1084 {
1085         return filemap_fdatawait_range(buf->pages[0]->mapping,
1086                                        buf->start, buf->start + buf->len - 1);
1087 }
1088
1089 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1090                                       u32 blocksize, u64 parent_transid)
1091 {
1092         struct extent_buffer *buf = NULL;
1093         int ret;
1094
1095         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1096         if (!buf)
1097                 return NULL;
1098
1099         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1100         return buf;
1101
1102 }
1103
1104 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1105                       struct extent_buffer *buf)
1106 {
1107         if (btrfs_header_generation(buf) ==
1108             root->fs_info->running_transaction->transid) {
1109                 btrfs_assert_tree_locked(buf);
1110
1111                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1112                         spin_lock(&root->fs_info->delalloc_lock);
1113                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1114                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1115                         else {
1116                                 spin_unlock(&root->fs_info->delalloc_lock);
1117                                 btrfs_panic(root->fs_info, -EOVERFLOW,
1118                                           "Can't clear %lu bytes from "
1119                                           " dirty_mdatadata_bytes (%lu)",
1120                                           buf->len,
1121                                           root->fs_info->dirty_metadata_bytes);
1122                         }
1123                         spin_unlock(&root->fs_info->delalloc_lock);
1124                 }
1125
1126                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1127                 btrfs_set_lock_blocking(buf);
1128                 clear_extent_buffer_dirty(buf);
1129         }
1130 }
1131
1132 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1133                          u32 stripesize, struct btrfs_root *root,
1134                          struct btrfs_fs_info *fs_info,
1135                          u64 objectid)
1136 {
1137         root->node = NULL;
1138         root->commit_root = NULL;
1139         root->sectorsize = sectorsize;
1140         root->nodesize = nodesize;
1141         root->leafsize = leafsize;
1142         root->stripesize = stripesize;
1143         root->ref_cows = 0;
1144         root->track_dirty = 0;
1145         root->in_radix = 0;
1146         root->orphan_item_inserted = 0;
1147         root->orphan_cleanup_state = 0;
1148
1149         root->objectid = objectid;
1150         root->last_trans = 0;
1151         root->highest_objectid = 0;
1152         root->name = NULL;
1153         root->inode_tree = RB_ROOT;
1154         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1155         root->block_rsv = NULL;
1156         root->orphan_block_rsv = NULL;
1157
1158         INIT_LIST_HEAD(&root->dirty_list);
1159         INIT_LIST_HEAD(&root->root_list);
1160         spin_lock_init(&root->orphan_lock);
1161         spin_lock_init(&root->inode_lock);
1162         spin_lock_init(&root->accounting_lock);
1163         mutex_init(&root->objectid_mutex);
1164         mutex_init(&root->log_mutex);
1165         init_waitqueue_head(&root->log_writer_wait);
1166         init_waitqueue_head(&root->log_commit_wait[0]);
1167         init_waitqueue_head(&root->log_commit_wait[1]);
1168         atomic_set(&root->log_commit[0], 0);
1169         atomic_set(&root->log_commit[1], 0);
1170         atomic_set(&root->log_writers, 0);
1171         atomic_set(&root->orphan_inodes, 0);
1172         root->log_batch = 0;
1173         root->log_transid = 0;
1174         root->last_log_commit = 0;
1175         extent_io_tree_init(&root->dirty_log_pages,
1176                              fs_info->btree_inode->i_mapping);
1177
1178         memset(&root->root_key, 0, sizeof(root->root_key));
1179         memset(&root->root_item, 0, sizeof(root->root_item));
1180         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1181         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1182         root->defrag_trans_start = fs_info->generation;
1183         init_completion(&root->kobj_unregister);
1184         root->defrag_running = 0;
1185         root->root_key.objectid = objectid;
1186         root->anon_dev = 0;
1187
1188         spin_lock_init(&root->root_times_lock);
1189 }
1190
1191 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1192                                             struct btrfs_fs_info *fs_info,
1193                                             u64 objectid,
1194                                             struct btrfs_root *root)
1195 {
1196         int ret;
1197         u32 blocksize;
1198         u64 generation;
1199
1200         __setup_root(tree_root->nodesize, tree_root->leafsize,
1201                      tree_root->sectorsize, tree_root->stripesize,
1202                      root, fs_info, objectid);
1203         ret = btrfs_find_last_root(tree_root, objectid,
1204                                    &root->root_item, &root->root_key);
1205         if (ret > 0)
1206                 return -ENOENT;
1207         else if (ret < 0)
1208                 return ret;
1209
1210         generation = btrfs_root_generation(&root->root_item);
1211         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1212         root->commit_root = NULL;
1213         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1214                                      blocksize, generation);
1215         if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1216                 free_extent_buffer(root->node);
1217                 root->node = NULL;
1218                 return -EIO;
1219         }
1220         root->commit_root = btrfs_root_node(root);
1221         return 0;
1222 }
1223
1224 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1225 {
1226         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1227         if (root)
1228                 root->fs_info = fs_info;
1229         return root;
1230 }
1231
1232 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1233                                      struct btrfs_fs_info *fs_info,
1234                                      u64 objectid)
1235 {
1236         struct extent_buffer *leaf;
1237         struct btrfs_root *tree_root = fs_info->tree_root;
1238         struct btrfs_root *root;
1239         struct btrfs_key key;
1240         int ret = 0;
1241         u64 bytenr;
1242
1243         root = btrfs_alloc_root(fs_info);
1244         if (!root)
1245                 return ERR_PTR(-ENOMEM);
1246
1247         __setup_root(tree_root->nodesize, tree_root->leafsize,
1248                      tree_root->sectorsize, tree_root->stripesize,
1249                      root, fs_info, objectid);
1250         root->root_key.objectid = objectid;
1251         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1252         root->root_key.offset = 0;
1253
1254         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1255                                       0, objectid, NULL, 0, 0, 0);
1256         if (IS_ERR(leaf)) {
1257                 ret = PTR_ERR(leaf);
1258                 goto fail;
1259         }
1260
1261         bytenr = leaf->start;
1262         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1263         btrfs_set_header_bytenr(leaf, leaf->start);
1264         btrfs_set_header_generation(leaf, trans->transid);
1265         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1266         btrfs_set_header_owner(leaf, objectid);
1267         root->node = leaf;
1268
1269         write_extent_buffer(leaf, fs_info->fsid,
1270                             (unsigned long)btrfs_header_fsid(leaf),
1271                             BTRFS_FSID_SIZE);
1272         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1273                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1274                             BTRFS_UUID_SIZE);
1275         btrfs_mark_buffer_dirty(leaf);
1276
1277         root->commit_root = btrfs_root_node(root);
1278         root->track_dirty = 1;
1279
1280
1281         root->root_item.flags = 0;
1282         root->root_item.byte_limit = 0;
1283         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1284         btrfs_set_root_generation(&root->root_item, trans->transid);
1285         btrfs_set_root_level(&root->root_item, 0);
1286         btrfs_set_root_refs(&root->root_item, 1);
1287         btrfs_set_root_used(&root->root_item, leaf->len);
1288         btrfs_set_root_last_snapshot(&root->root_item, 0);
1289         btrfs_set_root_dirid(&root->root_item, 0);
1290         root->root_item.drop_level = 0;
1291
1292         key.objectid = objectid;
1293         key.type = BTRFS_ROOT_ITEM_KEY;
1294         key.offset = 0;
1295         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1296         if (ret)
1297                 goto fail;
1298
1299         btrfs_tree_unlock(leaf);
1300
1301 fail:
1302         if (ret)
1303                 return ERR_PTR(ret);
1304
1305         return root;
1306 }
1307
1308 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1309                                          struct btrfs_fs_info *fs_info)
1310 {
1311         struct btrfs_root *root;
1312         struct btrfs_root *tree_root = fs_info->tree_root;
1313         struct extent_buffer *leaf;
1314
1315         root = btrfs_alloc_root(fs_info);
1316         if (!root)
1317                 return ERR_PTR(-ENOMEM);
1318
1319         __setup_root(tree_root->nodesize, tree_root->leafsize,
1320                      tree_root->sectorsize, tree_root->stripesize,
1321                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1322
1323         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1324         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1325         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1326         /*
1327          * log trees do not get reference counted because they go away
1328          * before a real commit is actually done.  They do store pointers
1329          * to file data extents, and those reference counts still get
1330          * updated (along with back refs to the log tree).
1331          */
1332         root->ref_cows = 0;
1333
1334         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1335                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1336                                       0, 0, 0);
1337         if (IS_ERR(leaf)) {
1338                 kfree(root);
1339                 return ERR_CAST(leaf);
1340         }
1341
1342         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1343         btrfs_set_header_bytenr(leaf, leaf->start);
1344         btrfs_set_header_generation(leaf, trans->transid);
1345         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1346         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1347         root->node = leaf;
1348
1349         write_extent_buffer(root->node, root->fs_info->fsid,
1350                             (unsigned long)btrfs_header_fsid(root->node),
1351                             BTRFS_FSID_SIZE);
1352         btrfs_mark_buffer_dirty(root->node);
1353         btrfs_tree_unlock(root->node);
1354         return root;
1355 }
1356
1357 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1358                              struct btrfs_fs_info *fs_info)
1359 {
1360         struct btrfs_root *log_root;
1361
1362         log_root = alloc_log_tree(trans, fs_info);
1363         if (IS_ERR(log_root))
1364                 return PTR_ERR(log_root);
1365         WARN_ON(fs_info->log_root_tree);
1366         fs_info->log_root_tree = log_root;
1367         return 0;
1368 }
1369
1370 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1371                        struct btrfs_root *root)
1372 {
1373         struct btrfs_root *log_root;
1374         struct btrfs_inode_item *inode_item;
1375
1376         log_root = alloc_log_tree(trans, root->fs_info);
1377         if (IS_ERR(log_root))
1378                 return PTR_ERR(log_root);
1379
1380         log_root->last_trans = trans->transid;
1381         log_root->root_key.offset = root->root_key.objectid;
1382
1383         inode_item = &log_root->root_item.inode;
1384         inode_item->generation = cpu_to_le64(1);
1385         inode_item->size = cpu_to_le64(3);
1386         inode_item->nlink = cpu_to_le32(1);
1387         inode_item->nbytes = cpu_to_le64(root->leafsize);
1388         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1389
1390         btrfs_set_root_node(&log_root->root_item, log_root->node);
1391
1392         WARN_ON(root->log_root);
1393         root->log_root = log_root;
1394         root->log_transid = 0;
1395         root->last_log_commit = 0;
1396         return 0;
1397 }
1398
1399 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1400                                                struct btrfs_key *location)
1401 {
1402         struct btrfs_root *root;
1403         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1404         struct btrfs_path *path;
1405         struct extent_buffer *l;
1406         u64 generation;
1407         u32 blocksize;
1408         int ret = 0;
1409         int slot;
1410
1411         root = btrfs_alloc_root(fs_info);
1412         if (!root)
1413                 return ERR_PTR(-ENOMEM);
1414         if (location->offset == (u64)-1) {
1415                 ret = find_and_setup_root(tree_root, fs_info,
1416                                           location->objectid, root);
1417                 if (ret) {
1418                         kfree(root);
1419                         return ERR_PTR(ret);
1420                 }
1421                 goto out;
1422         }
1423
1424         __setup_root(tree_root->nodesize, tree_root->leafsize,
1425                      tree_root->sectorsize, tree_root->stripesize,
1426                      root, fs_info, location->objectid);
1427
1428         path = btrfs_alloc_path();
1429         if (!path) {
1430                 kfree(root);
1431                 return ERR_PTR(-ENOMEM);
1432         }
1433         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1434         if (ret == 0) {
1435                 l = path->nodes[0];
1436                 slot = path->slots[0];
1437                 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1438                 memcpy(&root->root_key, location, sizeof(*location));
1439         }
1440         btrfs_free_path(path);
1441         if (ret) {
1442                 kfree(root);
1443                 if (ret > 0)
1444                         ret = -ENOENT;
1445                 return ERR_PTR(ret);
1446         }
1447
1448         generation = btrfs_root_generation(&root->root_item);
1449         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1450         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1451                                      blocksize, generation);
1452         root->commit_root = btrfs_root_node(root);
1453         BUG_ON(!root->node); /* -ENOMEM */
1454 out:
1455         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1456                 root->ref_cows = 1;
1457                 btrfs_check_and_init_root_item(&root->root_item);
1458         }
1459
1460         return root;
1461 }
1462
1463 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1464                                               struct btrfs_key *location)
1465 {
1466         struct btrfs_root *root;
1467         int ret;
1468
1469         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1470                 return fs_info->tree_root;
1471         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1472                 return fs_info->extent_root;
1473         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1474                 return fs_info->chunk_root;
1475         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1476                 return fs_info->dev_root;
1477         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1478                 return fs_info->csum_root;
1479         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1480                 return fs_info->quota_root ? fs_info->quota_root :
1481                                              ERR_PTR(-ENOENT);
1482 again:
1483         spin_lock(&fs_info->fs_roots_radix_lock);
1484         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1485                                  (unsigned long)location->objectid);
1486         spin_unlock(&fs_info->fs_roots_radix_lock);
1487         if (root)
1488                 return root;
1489
1490         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1491         if (IS_ERR(root))
1492                 return root;
1493
1494         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1495         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1496                                         GFP_NOFS);
1497         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1498                 ret = -ENOMEM;
1499                 goto fail;
1500         }
1501
1502         btrfs_init_free_ino_ctl(root);
1503         mutex_init(&root->fs_commit_mutex);
1504         spin_lock_init(&root->cache_lock);
1505         init_waitqueue_head(&root->cache_wait);
1506
1507         ret = get_anon_bdev(&root->anon_dev);
1508         if (ret)
1509                 goto fail;
1510
1511         if (btrfs_root_refs(&root->root_item) == 0) {
1512                 ret = -ENOENT;
1513                 goto fail;
1514         }
1515
1516         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1517         if (ret < 0)
1518                 goto fail;
1519         if (ret == 0)
1520                 root->orphan_item_inserted = 1;
1521
1522         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1523         if (ret)
1524                 goto fail;
1525
1526         spin_lock(&fs_info->fs_roots_radix_lock);
1527         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1528                                 (unsigned long)root->root_key.objectid,
1529                                 root);
1530         if (ret == 0)
1531                 root->in_radix = 1;
1532
1533         spin_unlock(&fs_info->fs_roots_radix_lock);
1534         radix_tree_preload_end();
1535         if (ret) {
1536                 if (ret == -EEXIST) {
1537                         free_fs_root(root);
1538                         goto again;
1539                 }
1540                 goto fail;
1541         }
1542
1543         ret = btrfs_find_dead_roots(fs_info->tree_root,
1544                                     root->root_key.objectid);
1545         WARN_ON(ret);
1546         return root;
1547 fail:
1548         free_fs_root(root);
1549         return ERR_PTR(ret);
1550 }
1551
1552 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1553 {
1554         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1555         int ret = 0;
1556         struct btrfs_device *device;
1557         struct backing_dev_info *bdi;
1558
1559         rcu_read_lock();
1560         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1561                 if (!device->bdev)
1562                         continue;
1563                 bdi = blk_get_backing_dev_info(device->bdev);
1564                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1565                         ret = 1;
1566                         break;
1567                 }
1568         }
1569         rcu_read_unlock();
1570         return ret;
1571 }
1572
1573 /*
1574  * If this fails, caller must call bdi_destroy() to get rid of the
1575  * bdi again.
1576  */
1577 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1578 {
1579         int err;
1580
1581         bdi->capabilities = BDI_CAP_MAP_COPY;
1582         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1583         if (err)
1584                 return err;
1585
1586         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1587         bdi->congested_fn       = btrfs_congested_fn;
1588         bdi->congested_data     = info;
1589         return 0;
1590 }
1591
1592 /*
1593  * called by the kthread helper functions to finally call the bio end_io
1594  * functions.  This is where read checksum verification actually happens
1595  */
1596 static void end_workqueue_fn(struct btrfs_work *work)
1597 {
1598         struct bio *bio;
1599         struct end_io_wq *end_io_wq;
1600         struct btrfs_fs_info *fs_info;
1601         int error;
1602
1603         end_io_wq = container_of(work, struct end_io_wq, work);
1604         bio = end_io_wq->bio;
1605         fs_info = end_io_wq->info;
1606
1607         error = end_io_wq->error;
1608         bio->bi_private = end_io_wq->private;
1609         bio->bi_end_io = end_io_wq->end_io;
1610         kfree(end_io_wq);
1611         bio_endio(bio, error);
1612 }
1613
1614 static int cleaner_kthread(void *arg)
1615 {
1616         struct btrfs_root *root = arg;
1617
1618         do {
1619                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1620
1621                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1622                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1623                         btrfs_run_delayed_iputs(root);
1624                         btrfs_clean_old_snapshots(root);
1625                         mutex_unlock(&root->fs_info->cleaner_mutex);
1626                         btrfs_run_defrag_inodes(root->fs_info);
1627                 }
1628
1629                 if (!try_to_freeze()) {
1630                         set_current_state(TASK_INTERRUPTIBLE);
1631                         if (!kthread_should_stop())
1632                                 schedule();
1633                         __set_current_state(TASK_RUNNING);
1634                 }
1635         } while (!kthread_should_stop());
1636         return 0;
1637 }
1638
1639 static int transaction_kthread(void *arg)
1640 {
1641         struct btrfs_root *root = arg;
1642         struct btrfs_trans_handle *trans;
1643         struct btrfs_transaction *cur;
1644         u64 transid;
1645         unsigned long now;
1646         unsigned long delay;
1647         bool cannot_commit;
1648
1649         do {
1650                 cannot_commit = false;
1651                 delay = HZ * 30;
1652                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1653                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1654
1655                 spin_lock(&root->fs_info->trans_lock);
1656                 cur = root->fs_info->running_transaction;
1657                 if (!cur) {
1658                         spin_unlock(&root->fs_info->trans_lock);
1659                         goto sleep;
1660                 }
1661
1662                 now = get_seconds();
1663                 if (!cur->blocked &&
1664                     (now < cur->start_time || now - cur->start_time < 30)) {
1665                         spin_unlock(&root->fs_info->trans_lock);
1666                         delay = HZ * 5;
1667                         goto sleep;
1668                 }
1669                 transid = cur->transid;
1670                 spin_unlock(&root->fs_info->trans_lock);
1671
1672                 /* If the file system is aborted, this will always fail. */
1673                 trans = btrfs_join_transaction(root);
1674                 if (IS_ERR(trans)) {
1675                         cannot_commit = true;
1676                         goto sleep;
1677                 }
1678                 if (transid == trans->transid) {
1679                         btrfs_commit_transaction(trans, root);
1680                 } else {
1681                         btrfs_end_transaction(trans, root);
1682                 }
1683 sleep:
1684                 wake_up_process(root->fs_info->cleaner_kthread);
1685                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1686
1687                 if (!try_to_freeze()) {
1688                         set_current_state(TASK_INTERRUPTIBLE);
1689                         if (!kthread_should_stop() &&
1690                             (!btrfs_transaction_blocked(root->fs_info) ||
1691                              cannot_commit))
1692                                 schedule_timeout(delay);
1693                         __set_current_state(TASK_RUNNING);
1694                 }
1695         } while (!kthread_should_stop());
1696         return 0;
1697 }
1698
1699 /*
1700  * this will find the highest generation in the array of
1701  * root backups.  The index of the highest array is returned,
1702  * or -1 if we can't find anything.
1703  *
1704  * We check to make sure the array is valid by comparing the
1705  * generation of the latest  root in the array with the generation
1706  * in the super block.  If they don't match we pitch it.
1707  */
1708 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1709 {
1710         u64 cur;
1711         int newest_index = -1;
1712         struct btrfs_root_backup *root_backup;
1713         int i;
1714
1715         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1716                 root_backup = info->super_copy->super_roots + i;
1717                 cur = btrfs_backup_tree_root_gen(root_backup);
1718                 if (cur == newest_gen)
1719                         newest_index = i;
1720         }
1721
1722         /* check to see if we actually wrapped around */
1723         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1724                 root_backup = info->super_copy->super_roots;
1725                 cur = btrfs_backup_tree_root_gen(root_backup);
1726                 if (cur == newest_gen)
1727                         newest_index = 0;
1728         }
1729         return newest_index;
1730 }
1731
1732
1733 /*
1734  * find the oldest backup so we know where to store new entries
1735  * in the backup array.  This will set the backup_root_index
1736  * field in the fs_info struct
1737  */
1738 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1739                                      u64 newest_gen)
1740 {
1741         int newest_index = -1;
1742
1743         newest_index = find_newest_super_backup(info, newest_gen);
1744         /* if there was garbage in there, just move along */
1745         if (newest_index == -1) {
1746                 info->backup_root_index = 0;
1747         } else {
1748                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1749         }
1750 }
1751
1752 /*
1753  * copy all the root pointers into the super backup array.
1754  * this will bump the backup pointer by one when it is
1755  * done
1756  */
1757 static void backup_super_roots(struct btrfs_fs_info *info)
1758 {
1759         int next_backup;
1760         struct btrfs_root_backup *root_backup;
1761         int last_backup;
1762
1763         next_backup = info->backup_root_index;
1764         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1765                 BTRFS_NUM_BACKUP_ROOTS;
1766
1767         /*
1768          * just overwrite the last backup if we're at the same generation
1769          * this happens only at umount
1770          */
1771         root_backup = info->super_for_commit->super_roots + last_backup;
1772         if (btrfs_backup_tree_root_gen(root_backup) ==
1773             btrfs_header_generation(info->tree_root->node))
1774                 next_backup = last_backup;
1775
1776         root_backup = info->super_for_commit->super_roots + next_backup;
1777
1778         /*
1779          * make sure all of our padding and empty slots get zero filled
1780          * regardless of which ones we use today
1781          */
1782         memset(root_backup, 0, sizeof(*root_backup));
1783
1784         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1785
1786         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1787         btrfs_set_backup_tree_root_gen(root_backup,
1788                                btrfs_header_generation(info->tree_root->node));
1789
1790         btrfs_set_backup_tree_root_level(root_backup,
1791                                btrfs_header_level(info->tree_root->node));
1792
1793         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1794         btrfs_set_backup_chunk_root_gen(root_backup,
1795                                btrfs_header_generation(info->chunk_root->node));
1796         btrfs_set_backup_chunk_root_level(root_backup,
1797                                btrfs_header_level(info->chunk_root->node));
1798
1799         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1800         btrfs_set_backup_extent_root_gen(root_backup,
1801                                btrfs_header_generation(info->extent_root->node));
1802         btrfs_set_backup_extent_root_level(root_backup,
1803                                btrfs_header_level(info->extent_root->node));
1804
1805         /*
1806          * we might commit during log recovery, which happens before we set
1807          * the fs_root.  Make sure it is valid before we fill it in.
1808          */
1809         if (info->fs_root && info->fs_root->node) {
1810                 btrfs_set_backup_fs_root(root_backup,
1811                                          info->fs_root->node->start);
1812                 btrfs_set_backup_fs_root_gen(root_backup,
1813                                btrfs_header_generation(info->fs_root->node));
1814                 btrfs_set_backup_fs_root_level(root_backup,
1815                                btrfs_header_level(info->fs_root->node));
1816         }
1817
1818         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1819         btrfs_set_backup_dev_root_gen(root_backup,
1820                                btrfs_header_generation(info->dev_root->node));
1821         btrfs_set_backup_dev_root_level(root_backup,
1822                                        btrfs_header_level(info->dev_root->node));
1823
1824         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1825         btrfs_set_backup_csum_root_gen(root_backup,
1826                                btrfs_header_generation(info->csum_root->node));
1827         btrfs_set_backup_csum_root_level(root_backup,
1828                                btrfs_header_level(info->csum_root->node));
1829
1830         btrfs_set_backup_total_bytes(root_backup,
1831                              btrfs_super_total_bytes(info->super_copy));
1832         btrfs_set_backup_bytes_used(root_backup,
1833                              btrfs_super_bytes_used(info->super_copy));
1834         btrfs_set_backup_num_devices(root_backup,
1835                              btrfs_super_num_devices(info->super_copy));
1836
1837         /*
1838          * if we don't copy this out to the super_copy, it won't get remembered
1839          * for the next commit
1840          */
1841         memcpy(&info->super_copy->super_roots,
1842                &info->super_for_commit->super_roots,
1843                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1844 }
1845
1846 /*
1847  * this copies info out of the root backup array and back into
1848  * the in-memory super block.  It is meant to help iterate through
1849  * the array, so you send it the number of backups you've already
1850  * tried and the last backup index you used.
1851  *
1852  * this returns -1 when it has tried all the backups
1853  */
1854 static noinline int next_root_backup(struct btrfs_fs_info *info,
1855                                      struct btrfs_super_block *super,
1856                                      int *num_backups_tried, int *backup_index)
1857 {
1858         struct btrfs_root_backup *root_backup;
1859         int newest = *backup_index;
1860
1861         if (*num_backups_tried == 0) {
1862                 u64 gen = btrfs_super_generation(super);
1863
1864                 newest = find_newest_super_backup(info, gen);
1865                 if (newest == -1)
1866                         return -1;
1867
1868                 *backup_index = newest;
1869                 *num_backups_tried = 1;
1870         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1871                 /* we've tried all the backups, all done */
1872                 return -1;
1873         } else {
1874                 /* jump to the next oldest backup */
1875                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1876                         BTRFS_NUM_BACKUP_ROOTS;
1877                 *backup_index = newest;
1878                 *num_backups_tried += 1;
1879         }
1880         root_backup = super->super_roots + newest;
1881
1882         btrfs_set_super_generation(super,
1883                                    btrfs_backup_tree_root_gen(root_backup));
1884         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1885         btrfs_set_super_root_level(super,
1886                                    btrfs_backup_tree_root_level(root_backup));
1887         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1888
1889         /*
1890          * fixme: the total bytes and num_devices need to match or we should
1891          * need a fsck
1892          */
1893         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1894         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1895         return 0;
1896 }
1897
1898 /* helper to cleanup tree roots */
1899 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1900 {
1901         free_extent_buffer(info->tree_root->node);
1902         free_extent_buffer(info->tree_root->commit_root);
1903         free_extent_buffer(info->dev_root->node);
1904         free_extent_buffer(info->dev_root->commit_root);
1905         free_extent_buffer(info->extent_root->node);
1906         free_extent_buffer(info->extent_root->commit_root);
1907         free_extent_buffer(info->csum_root->node);
1908         free_extent_buffer(info->csum_root->commit_root);
1909         if (info->quota_root) {
1910                 free_extent_buffer(info->quota_root->node);
1911                 free_extent_buffer(info->quota_root->commit_root);
1912         }
1913
1914         info->tree_root->node = NULL;
1915         info->tree_root->commit_root = NULL;
1916         info->dev_root->node = NULL;
1917         info->dev_root->commit_root = NULL;
1918         info->extent_root->node = NULL;
1919         info->extent_root->commit_root = NULL;
1920         info->csum_root->node = NULL;
1921         info->csum_root->commit_root = NULL;
1922         if (info->quota_root) {
1923                 info->quota_root->node = NULL;
1924                 info->quota_root->commit_root = NULL;
1925         }
1926
1927         if (chunk_root) {
1928                 free_extent_buffer(info->chunk_root->node);
1929                 free_extent_buffer(info->chunk_root->commit_root);
1930                 info->chunk_root->node = NULL;
1931                 info->chunk_root->commit_root = NULL;
1932         }
1933 }
1934
1935
1936 int open_ctree(struct super_block *sb,
1937                struct btrfs_fs_devices *fs_devices,
1938                char *options)
1939 {
1940         u32 sectorsize;
1941         u32 nodesize;
1942         u32 leafsize;
1943         u32 blocksize;
1944         u32 stripesize;
1945         u64 generation;
1946         u64 features;
1947         struct btrfs_key location;
1948         struct buffer_head *bh;
1949         struct btrfs_super_block *disk_super;
1950         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1951         struct btrfs_root *tree_root;
1952         struct btrfs_root *extent_root;
1953         struct btrfs_root *csum_root;
1954         struct btrfs_root *chunk_root;
1955         struct btrfs_root *dev_root;
1956         struct btrfs_root *quota_root;
1957         struct btrfs_root *log_tree_root;
1958         int ret;
1959         int err = -EINVAL;
1960         int num_backups_tried = 0;
1961         int backup_index = 0;
1962
1963         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1964         extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1965         csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1966         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1967         dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1968         quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1969
1970         if (!tree_root || !extent_root || !csum_root ||
1971             !chunk_root || !dev_root || !quota_root) {
1972                 err = -ENOMEM;
1973                 goto fail;
1974         }
1975
1976         ret = init_srcu_struct(&fs_info->subvol_srcu);
1977         if (ret) {
1978                 err = ret;
1979                 goto fail;
1980         }
1981
1982         ret = setup_bdi(fs_info, &fs_info->bdi);
1983         if (ret) {
1984                 err = ret;
1985                 goto fail_srcu;
1986         }
1987
1988         fs_info->btree_inode = new_inode(sb);
1989         if (!fs_info->btree_inode) {
1990                 err = -ENOMEM;
1991                 goto fail_bdi;
1992         }
1993
1994         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1995
1996         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1997         INIT_LIST_HEAD(&fs_info->trans_list);
1998         INIT_LIST_HEAD(&fs_info->dead_roots);
1999         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2000         INIT_LIST_HEAD(&fs_info->hashers);
2001         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2002         INIT_LIST_HEAD(&fs_info->ordered_operations);
2003         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2004         spin_lock_init(&fs_info->delalloc_lock);
2005         spin_lock_init(&fs_info->trans_lock);
2006         spin_lock_init(&fs_info->ref_cache_lock);
2007         spin_lock_init(&fs_info->fs_roots_radix_lock);
2008         spin_lock_init(&fs_info->delayed_iput_lock);
2009         spin_lock_init(&fs_info->defrag_inodes_lock);
2010         spin_lock_init(&fs_info->free_chunk_lock);
2011         spin_lock_init(&fs_info->tree_mod_seq_lock);
2012         rwlock_init(&fs_info->tree_mod_log_lock);
2013         mutex_init(&fs_info->reloc_mutex);
2014
2015         init_completion(&fs_info->kobj_unregister);
2016         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2017         INIT_LIST_HEAD(&fs_info->space_info);
2018         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2019         btrfs_mapping_init(&fs_info->mapping_tree);
2020         btrfs_init_block_rsv(&fs_info->global_block_rsv);
2021         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
2022         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
2023         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
2024         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
2025         btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
2026         atomic_set(&fs_info->nr_async_submits, 0);
2027         atomic_set(&fs_info->async_delalloc_pages, 0);
2028         atomic_set(&fs_info->async_submit_draining, 0);
2029         atomic_set(&fs_info->nr_async_bios, 0);
2030         atomic_set(&fs_info->defrag_running, 0);
2031         atomic_set(&fs_info->tree_mod_seq, 0);
2032         fs_info->sb = sb;
2033         fs_info->max_inline = 8192 * 1024;
2034         fs_info->metadata_ratio = 0;
2035         fs_info->defrag_inodes = RB_ROOT;
2036         fs_info->trans_no_join = 0;
2037         fs_info->free_chunk_space = 0;
2038         fs_info->tree_mod_log = RB_ROOT;
2039
2040         /* readahead state */
2041         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2042         spin_lock_init(&fs_info->reada_lock);
2043
2044         fs_info->thread_pool_size = min_t(unsigned long,
2045                                           num_online_cpus() + 2, 8);
2046
2047         INIT_LIST_HEAD(&fs_info->ordered_extents);
2048         spin_lock_init(&fs_info->ordered_extent_lock);
2049         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2050                                         GFP_NOFS);
2051         if (!fs_info->delayed_root) {
2052                 err = -ENOMEM;
2053                 goto fail_iput;
2054         }
2055         btrfs_init_delayed_root(fs_info->delayed_root);
2056
2057         mutex_init(&fs_info->scrub_lock);
2058         atomic_set(&fs_info->scrubs_running, 0);
2059         atomic_set(&fs_info->scrub_pause_req, 0);
2060         atomic_set(&fs_info->scrubs_paused, 0);
2061         atomic_set(&fs_info->scrub_cancel_req, 0);
2062         init_waitqueue_head(&fs_info->scrub_pause_wait);
2063         init_rwsem(&fs_info->scrub_super_lock);
2064         fs_info->scrub_workers_refcnt = 0;
2065 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2066         fs_info->check_integrity_print_mask = 0;
2067 #endif
2068
2069         spin_lock_init(&fs_info->balance_lock);
2070         mutex_init(&fs_info->balance_mutex);
2071         atomic_set(&fs_info->balance_running, 0);
2072         atomic_set(&fs_info->balance_pause_req, 0);
2073         atomic_set(&fs_info->balance_cancel_req, 0);
2074         fs_info->balance_ctl = NULL;
2075         init_waitqueue_head(&fs_info->balance_wait_q);
2076
2077         sb->s_blocksize = 4096;
2078         sb->s_blocksize_bits = blksize_bits(4096);
2079         sb->s_bdi = &fs_info->bdi;
2080
2081         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2082         set_nlink(fs_info->btree_inode, 1);
2083         /*
2084          * we set the i_size on the btree inode to the max possible int.
2085          * the real end of the address space is determined by all of
2086          * the devices in the system
2087          */
2088         fs_info->btree_inode->i_size = OFFSET_MAX;
2089         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2090         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2091
2092         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2093         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2094                              fs_info->btree_inode->i_mapping);
2095         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2096         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2097
2098         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2099
2100         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2101         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2102                sizeof(struct btrfs_key));
2103         set_bit(BTRFS_INODE_DUMMY,
2104                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2105         insert_inode_hash(fs_info->btree_inode);
2106
2107         spin_lock_init(&fs_info->block_group_cache_lock);
2108         fs_info->block_group_cache_tree = RB_ROOT;
2109
2110         extent_io_tree_init(&fs_info->freed_extents[0],
2111                              fs_info->btree_inode->i_mapping);
2112         extent_io_tree_init(&fs_info->freed_extents[1],
2113                              fs_info->btree_inode->i_mapping);
2114         fs_info->pinned_extents = &fs_info->freed_extents[0];
2115         fs_info->do_barriers = 1;
2116
2117
2118         mutex_init(&fs_info->ordered_operations_mutex);
2119         mutex_init(&fs_info->tree_log_mutex);
2120         mutex_init(&fs_info->chunk_mutex);
2121         mutex_init(&fs_info->transaction_kthread_mutex);
2122         mutex_init(&fs_info->cleaner_mutex);
2123         mutex_init(&fs_info->volume_mutex);
2124         init_rwsem(&fs_info->extent_commit_sem);
2125         init_rwsem(&fs_info->cleanup_work_sem);
2126         init_rwsem(&fs_info->subvol_sem);
2127
2128         spin_lock_init(&fs_info->qgroup_lock);
2129         fs_info->qgroup_tree = RB_ROOT;
2130         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2131         fs_info->qgroup_seq = 1;
2132         fs_info->quota_enabled = 0;
2133         fs_info->pending_quota_state = 0;
2134
2135         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2136         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2137
2138         init_waitqueue_head(&fs_info->transaction_throttle);
2139         init_waitqueue_head(&fs_info->transaction_wait);
2140         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2141         init_waitqueue_head(&fs_info->async_submit_wait);
2142
2143         __setup_root(4096, 4096, 4096, 4096, tree_root,
2144                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2145
2146         invalidate_bdev(fs_devices->latest_bdev);
2147         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2148         if (!bh) {
2149                 err = -EINVAL;
2150                 goto fail_alloc;
2151         }
2152
2153         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2154         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2155                sizeof(*fs_info->super_for_commit));
2156         brelse(bh);
2157
2158         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2159
2160         disk_super = fs_info->super_copy;
2161         if (!btrfs_super_root(disk_super))
2162                 goto fail_alloc;
2163
2164         /* check FS state, whether FS is broken. */
2165         fs_info->fs_state |= btrfs_super_flags(disk_super);
2166
2167         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2168         if (ret) {
2169                 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2170                 err = ret;
2171                 goto fail_alloc;
2172         }
2173
2174         /*
2175          * run through our array of backup supers and setup
2176          * our ring pointer to the oldest one
2177          */
2178         generation = btrfs_super_generation(disk_super);
2179         find_oldest_super_backup(fs_info, generation);
2180
2181         /*
2182          * In the long term, we'll store the compression type in the super
2183          * block, and it'll be used for per file compression control.
2184          */
2185         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2186
2187         ret = btrfs_parse_options(tree_root, options);
2188         if (ret) {
2189                 err = ret;
2190                 goto fail_alloc;
2191         }
2192
2193         features = btrfs_super_incompat_flags(disk_super) &
2194                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2195         if (features) {
2196                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2197                        "unsupported optional features (%Lx).\n",
2198                        (unsigned long long)features);
2199                 err = -EINVAL;
2200                 goto fail_alloc;
2201         }
2202
2203         if (btrfs_super_leafsize(disk_super) !=
2204             btrfs_super_nodesize(disk_super)) {
2205                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2206                        "blocksizes don't match.  node %d leaf %d\n",
2207                        btrfs_super_nodesize(disk_super),
2208                        btrfs_super_leafsize(disk_super));
2209                 err = -EINVAL;
2210                 goto fail_alloc;
2211         }
2212         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2213                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2214                        "blocksize (%d) was too large\n",
2215                        btrfs_super_leafsize(disk_super));
2216                 err = -EINVAL;
2217                 goto fail_alloc;
2218         }
2219
2220         features = btrfs_super_incompat_flags(disk_super);
2221         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2222         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2223                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2224
2225         /*
2226          * flag our filesystem as having big metadata blocks if
2227          * they are bigger than the page size
2228          */
2229         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2230                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2231                         printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2232                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2233         }
2234
2235         nodesize = btrfs_super_nodesize(disk_super);
2236         leafsize = btrfs_super_leafsize(disk_super);
2237         sectorsize = btrfs_super_sectorsize(disk_super);
2238         stripesize = btrfs_super_stripesize(disk_super);
2239
2240         /*
2241          * mixed block groups end up with duplicate but slightly offset
2242          * extent buffers for the same range.  It leads to corruptions
2243          */
2244         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2245             (sectorsize != leafsize)) {
2246                 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2247                                 "are not allowed for mixed block groups on %s\n",
2248                                 sb->s_id);
2249                 goto fail_alloc;
2250         }
2251
2252         btrfs_set_super_incompat_flags(disk_super, features);
2253
2254         features = btrfs_super_compat_ro_flags(disk_super) &
2255                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2256         if (!(sb->s_flags & MS_RDONLY) && features) {
2257                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2258                        "unsupported option features (%Lx).\n",
2259                        (unsigned long long)features);
2260                 err = -EINVAL;
2261                 goto fail_alloc;
2262         }
2263
2264         btrfs_init_workers(&fs_info->generic_worker,
2265                            "genwork", 1, NULL);
2266
2267         btrfs_init_workers(&fs_info->workers, "worker",
2268                            fs_info->thread_pool_size,
2269                            &fs_info->generic_worker);
2270
2271         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2272                            fs_info->thread_pool_size,
2273                            &fs_info->generic_worker);
2274
2275         btrfs_init_workers(&fs_info->submit_workers, "submit",
2276                            min_t(u64, fs_devices->num_devices,
2277                            fs_info->thread_pool_size),
2278                            &fs_info->generic_worker);
2279
2280         btrfs_init_workers(&fs_info->caching_workers, "cache",
2281                            2, &fs_info->generic_worker);
2282
2283         /* a higher idle thresh on the submit workers makes it much more
2284          * likely that bios will be send down in a sane order to the
2285          * devices
2286          */
2287         fs_info->submit_workers.idle_thresh = 64;
2288
2289         fs_info->workers.idle_thresh = 16;
2290         fs_info->workers.ordered = 1;
2291
2292         fs_info->delalloc_workers.idle_thresh = 2;
2293         fs_info->delalloc_workers.ordered = 1;
2294
2295         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2296                            &fs_info->generic_worker);
2297         btrfs_init_workers(&fs_info->endio_workers, "endio",
2298                            fs_info->thread_pool_size,
2299                            &fs_info->generic_worker);
2300         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2301                            fs_info->thread_pool_size,
2302                            &fs_info->generic_worker);
2303         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2304                            "endio-meta-write", fs_info->thread_pool_size,
2305                            &fs_info->generic_worker);
2306         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2307                            fs_info->thread_pool_size,
2308                            &fs_info->generic_worker);
2309         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2310                            1, &fs_info->generic_worker);
2311         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2312                            fs_info->thread_pool_size,
2313                            &fs_info->generic_worker);
2314         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2315                            fs_info->thread_pool_size,
2316                            &fs_info->generic_worker);
2317
2318         /*
2319          * endios are largely parallel and should have a very
2320          * low idle thresh
2321          */
2322         fs_info->endio_workers.idle_thresh = 4;
2323         fs_info->endio_meta_workers.idle_thresh = 4;
2324
2325         fs_info->endio_write_workers.idle_thresh = 2;
2326         fs_info->endio_meta_write_workers.idle_thresh = 2;
2327         fs_info->readahead_workers.idle_thresh = 2;
2328
2329         /*
2330          * btrfs_start_workers can really only fail because of ENOMEM so just
2331          * return -ENOMEM if any of these fail.
2332          */
2333         ret = btrfs_start_workers(&fs_info->workers);
2334         ret |= btrfs_start_workers(&fs_info->generic_worker);
2335         ret |= btrfs_start_workers(&fs_info->submit_workers);
2336         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2337         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2338         ret |= btrfs_start_workers(&fs_info->endio_workers);
2339         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2340         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2341         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2342         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2343         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2344         ret |= btrfs_start_workers(&fs_info->caching_workers);
2345         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2346         if (ret) {
2347                 err = -ENOMEM;
2348                 goto fail_sb_buffer;
2349         }
2350
2351         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2352         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2353                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2354
2355         tree_root->nodesize = nodesize;
2356         tree_root->leafsize = leafsize;
2357         tree_root->sectorsize = sectorsize;
2358         tree_root->stripesize = stripesize;
2359
2360         sb->s_blocksize = sectorsize;
2361         sb->s_blocksize_bits = blksize_bits(sectorsize);
2362
2363         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2364                     sizeof(disk_super->magic))) {
2365                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2366                 goto fail_sb_buffer;
2367         }
2368
2369         if (sectorsize != PAGE_SIZE) {
2370                 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2371                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2372                 goto fail_sb_buffer;
2373         }
2374
2375         mutex_lock(&fs_info->chunk_mutex);
2376         ret = btrfs_read_sys_array(tree_root);
2377         mutex_unlock(&fs_info->chunk_mutex);
2378         if (ret) {
2379                 printk(KERN_WARNING "btrfs: failed to read the system "
2380                        "array on %s\n", sb->s_id);
2381                 goto fail_sb_buffer;
2382         }
2383
2384         blocksize = btrfs_level_size(tree_root,
2385                                      btrfs_super_chunk_root_level(disk_super));
2386         generation = btrfs_super_chunk_root_generation(disk_super);
2387
2388         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2389                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2390
2391         chunk_root->node = read_tree_block(chunk_root,
2392                                            btrfs_super_chunk_root(disk_super),
2393                                            blocksize, generation);
2394         BUG_ON(!chunk_root->node); /* -ENOMEM */
2395         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2396                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2397                        sb->s_id);
2398                 goto fail_tree_roots;
2399         }
2400         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2401         chunk_root->commit_root = btrfs_root_node(chunk_root);
2402
2403         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2404            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2405            BTRFS_UUID_SIZE);
2406
2407         ret = btrfs_read_chunk_tree(chunk_root);
2408         if (ret) {
2409                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2410                        sb->s_id);
2411                 goto fail_tree_roots;
2412         }
2413
2414         btrfs_close_extra_devices(fs_devices);
2415
2416         if (!fs_devices->latest_bdev) {
2417                 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2418                        sb->s_id);
2419                 goto fail_tree_roots;
2420         }
2421
2422 retry_root_backup:
2423         blocksize = btrfs_level_size(tree_root,
2424                                      btrfs_super_root_level(disk_super));
2425         generation = btrfs_super_generation(disk_super);
2426
2427         tree_root->node = read_tree_block(tree_root,
2428                                           btrfs_super_root(disk_super),
2429                                           blocksize, generation);
2430         if (!tree_root->node ||
2431             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2432                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2433                        sb->s_id);
2434
2435                 goto recovery_tree_root;
2436         }
2437
2438         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2439         tree_root->commit_root = btrfs_root_node(tree_root);
2440
2441         ret = find_and_setup_root(tree_root, fs_info,
2442                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2443         if (ret)
2444                 goto recovery_tree_root;
2445         extent_root->track_dirty = 1;
2446
2447         ret = find_and_setup_root(tree_root, fs_info,
2448                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2449         if (ret)
2450                 goto recovery_tree_root;
2451         dev_root->track_dirty = 1;
2452
2453         ret = find_and_setup_root(tree_root, fs_info,
2454                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2455         if (ret)
2456                 goto recovery_tree_root;
2457         csum_root->track_dirty = 1;
2458
2459         ret = find_and_setup_root(tree_root, fs_info,
2460                                   BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2461         if (ret) {
2462                 kfree(quota_root);
2463                 quota_root = fs_info->quota_root = NULL;
2464         } else {
2465                 quota_root->track_dirty = 1;
2466                 fs_info->quota_enabled = 1;
2467                 fs_info->pending_quota_state = 1;
2468         }
2469
2470         fs_info->generation = generation;
2471         fs_info->last_trans_committed = generation;
2472
2473         ret = btrfs_recover_balance(fs_info);
2474         if (ret) {
2475                 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2476                 goto fail_block_groups;
2477         }
2478
2479         ret = btrfs_init_dev_stats(fs_info);
2480         if (ret) {
2481                 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2482                        ret);
2483                 goto fail_block_groups;
2484         }
2485
2486         ret = btrfs_init_space_info(fs_info);
2487         if (ret) {
2488                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2489                 goto fail_block_groups;
2490         }
2491
2492         ret = btrfs_read_block_groups(extent_root);
2493         if (ret) {
2494                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2495                 goto fail_block_groups;
2496         }
2497
2498         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2499                                                "btrfs-cleaner");
2500         if (IS_ERR(fs_info->cleaner_kthread))
2501                 goto fail_block_groups;
2502
2503         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2504                                                    tree_root,
2505                                                    "btrfs-transaction");
2506         if (IS_ERR(fs_info->transaction_kthread))
2507                 goto fail_cleaner;
2508
2509         if (!btrfs_test_opt(tree_root, SSD) &&
2510             !btrfs_test_opt(tree_root, NOSSD) &&
2511             !fs_info->fs_devices->rotating) {
2512                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2513                        "mode\n");
2514                 btrfs_set_opt(fs_info->mount_opt, SSD);
2515         }
2516
2517 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2518         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2519                 ret = btrfsic_mount(tree_root, fs_devices,
2520                                     btrfs_test_opt(tree_root,
2521                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2522                                     1 : 0,
2523                                     fs_info->check_integrity_print_mask);
2524                 if (ret)
2525                         printk(KERN_WARNING "btrfs: failed to initialize"
2526                                " integrity check module %s\n", sb->s_id);
2527         }
2528 #endif
2529         ret = btrfs_read_qgroup_config(fs_info);
2530         if (ret)
2531                 goto fail_trans_kthread;
2532
2533         /* do not make disk changes in broken FS */
2534         if (btrfs_super_log_root(disk_super) != 0) {
2535                 u64 bytenr = btrfs_super_log_root(disk_super);
2536
2537                 if (fs_devices->rw_devices == 0) {
2538                         printk(KERN_WARNING "Btrfs log replay required "
2539                                "on RO media\n");
2540                         err = -EIO;
2541                         goto fail_qgroup;
2542                 }
2543                 blocksize =
2544                      btrfs_level_size(tree_root,
2545                                       btrfs_super_log_root_level(disk_super));
2546
2547                 log_tree_root = btrfs_alloc_root(fs_info);
2548                 if (!log_tree_root) {
2549                         err = -ENOMEM;
2550                         goto fail_qgroup;
2551                 }
2552
2553                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2554                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2555
2556                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2557                                                       blocksize,
2558                                                       generation + 1);
2559                 /* returns with log_tree_root freed on success */
2560                 ret = btrfs_recover_log_trees(log_tree_root);
2561                 if (ret) {
2562                         btrfs_error(tree_root->fs_info, ret,
2563                                     "Failed to recover log tree");
2564                         free_extent_buffer(log_tree_root->node);
2565                         kfree(log_tree_root);
2566                         goto fail_trans_kthread;
2567                 }
2568
2569                 if (sb->s_flags & MS_RDONLY) {
2570                         ret = btrfs_commit_super(tree_root);
2571                         if (ret)
2572                                 goto fail_trans_kthread;
2573                 }
2574         }
2575
2576         ret = btrfs_find_orphan_roots(tree_root);
2577         if (ret)
2578                 goto fail_trans_kthread;
2579
2580         if (!(sb->s_flags & MS_RDONLY)) {
2581                 ret = btrfs_cleanup_fs_roots(fs_info);
2582                 if (ret)
2583                         goto fail_trans_kthread;
2584
2585                 ret = btrfs_recover_relocation(tree_root);
2586                 if (ret < 0) {
2587                         printk(KERN_WARNING
2588                                "btrfs: failed to recover relocation\n");
2589                         err = -EINVAL;
2590                         goto fail_qgroup;
2591                 }
2592         }
2593
2594         location.objectid = BTRFS_FS_TREE_OBJECTID;
2595         location.type = BTRFS_ROOT_ITEM_KEY;
2596         location.offset = (u64)-1;
2597
2598         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2599         if (!fs_info->fs_root)
2600                 goto fail_qgroup;
2601         if (IS_ERR(fs_info->fs_root)) {
2602                 err = PTR_ERR(fs_info->fs_root);
2603                 goto fail_qgroup;
2604         }
2605
2606         if (sb->s_flags & MS_RDONLY)
2607                 return 0;
2608
2609         down_read(&fs_info->cleanup_work_sem);
2610         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2611             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2612                 up_read(&fs_info->cleanup_work_sem);
2613                 close_ctree(tree_root);
2614                 return ret;
2615         }
2616         up_read(&fs_info->cleanup_work_sem);
2617
2618         ret = btrfs_resume_balance_async(fs_info);
2619         if (ret) {
2620                 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2621                 close_ctree(tree_root);
2622                 return ret;
2623         }
2624
2625         return 0;
2626
2627 fail_qgroup:
2628         btrfs_free_qgroup_config(fs_info);
2629 fail_trans_kthread:
2630         kthread_stop(fs_info->transaction_kthread);
2631 fail_cleaner:
2632         kthread_stop(fs_info->cleaner_kthread);
2633
2634         /*
2635          * make sure we're done with the btree inode before we stop our
2636          * kthreads
2637          */
2638         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2639         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2640
2641 fail_block_groups:
2642         btrfs_free_block_groups(fs_info);
2643
2644 fail_tree_roots:
2645         free_root_pointers(fs_info, 1);
2646
2647 fail_sb_buffer:
2648         btrfs_stop_workers(&fs_info->generic_worker);
2649         btrfs_stop_workers(&fs_info->readahead_workers);
2650         btrfs_stop_workers(&fs_info->fixup_workers);
2651         btrfs_stop_workers(&fs_info->delalloc_workers);
2652         btrfs_stop_workers(&fs_info->workers);
2653         btrfs_stop_workers(&fs_info->endio_workers);
2654         btrfs_stop_workers(&fs_info->endio_meta_workers);
2655         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2656         btrfs_stop_workers(&fs_info->endio_write_workers);
2657         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2658         btrfs_stop_workers(&fs_info->submit_workers);
2659         btrfs_stop_workers(&fs_info->delayed_workers);
2660         btrfs_stop_workers(&fs_info->caching_workers);
2661 fail_alloc:
2662 fail_iput:
2663         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2664
2665         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2666         iput(fs_info->btree_inode);
2667 fail_bdi:
2668         bdi_destroy(&fs_info->bdi);
2669 fail_srcu:
2670         cleanup_srcu_struct(&fs_info->subvol_srcu);
2671 fail:
2672         btrfs_close_devices(fs_info->fs_devices);
2673         return err;
2674
2675 recovery_tree_root:
2676         if (!btrfs_test_opt(tree_root, RECOVERY))
2677                 goto fail_tree_roots;
2678
2679         free_root_pointers(fs_info, 0);
2680
2681         /* don't use the log in recovery mode, it won't be valid */
2682         btrfs_set_super_log_root(disk_super, 0);
2683
2684         /* we can't trust the free space cache either */
2685         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2686
2687         ret = next_root_backup(fs_info, fs_info->super_copy,
2688                                &num_backups_tried, &backup_index);
2689         if (ret == -1)
2690                 goto fail_block_groups;
2691         goto retry_root_backup;
2692 }
2693
2694 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2695 {
2696         if (uptodate) {
2697                 set_buffer_uptodate(bh);
2698         } else {
2699                 struct btrfs_device *device = (struct btrfs_device *)
2700                         bh->b_private;
2701
2702                 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2703                                           "I/O error on %s\n",
2704                                           rcu_str_deref(device->name));
2705                 /* note, we dont' set_buffer_write_io_error because we have
2706                  * our own ways of dealing with the IO errors
2707                  */
2708                 clear_buffer_uptodate(bh);
2709                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2710         }
2711         unlock_buffer(bh);
2712         put_bh(bh);
2713 }
2714
2715 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2716 {
2717         struct buffer_head *bh;
2718         struct buffer_head *latest = NULL;
2719         struct btrfs_super_block *super;
2720         int i;
2721         u64 transid = 0;
2722         u64 bytenr;
2723
2724         /* we would like to check all the supers, but that would make
2725          * a btrfs mount succeed after a mkfs from a different FS.
2726          * So, we need to add a special mount option to scan for
2727          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2728          */
2729         for (i = 0; i < 1; i++) {
2730                 bytenr = btrfs_sb_offset(i);
2731                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2732                         break;
2733                 bh = __bread(bdev, bytenr / 4096, 4096);
2734                 if (!bh)
2735                         continue;
2736
2737                 super = (struct btrfs_super_block *)bh->b_data;
2738                 if (btrfs_super_bytenr(super) != bytenr ||
2739                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2740                             sizeof(super->magic))) {
2741                         brelse(bh);
2742                         continue;
2743                 }
2744
2745                 if (!latest || btrfs_super_generation(super) > transid) {
2746                         brelse(latest);
2747                         latest = bh;
2748                         transid = btrfs_super_generation(super);
2749                 } else {
2750                         brelse(bh);
2751                 }
2752         }
2753         return latest;
2754 }
2755
2756 /*
2757  * this should be called twice, once with wait == 0 and
2758  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2759  * we write are pinned.
2760  *
2761  * They are released when wait == 1 is done.
2762  * max_mirrors must be the same for both runs, and it indicates how
2763  * many supers on this one device should be written.
2764  *
2765  * max_mirrors == 0 means to write them all.
2766  */
2767 static int write_dev_supers(struct btrfs_device *device,
2768                             struct btrfs_super_block *sb,
2769                             int do_barriers, int wait, int max_mirrors)
2770 {
2771         struct buffer_head *bh;
2772         int i;
2773         int ret;
2774         int errors = 0;
2775         u32 crc;
2776         u64 bytenr;
2777
2778         if (max_mirrors == 0)
2779                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2780
2781         for (i = 0; i < max_mirrors; i++) {
2782                 bytenr = btrfs_sb_offset(i);
2783                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2784                         break;
2785
2786                 if (wait) {
2787                         bh = __find_get_block(device->bdev, bytenr / 4096,
2788                                               BTRFS_SUPER_INFO_SIZE);
2789                         BUG_ON(!bh);
2790                         wait_on_buffer(bh);
2791                         if (!buffer_uptodate(bh))
2792                                 errors++;
2793
2794                         /* drop our reference */
2795                         brelse(bh);
2796
2797                         /* drop the reference from the wait == 0 run */
2798                         brelse(bh);
2799                         continue;
2800                 } else {
2801                         btrfs_set_super_bytenr(sb, bytenr);
2802
2803                         crc = ~(u32)0;
2804                         crc = btrfs_csum_data(NULL, (char *)sb +
2805                                               BTRFS_CSUM_SIZE, crc,
2806                                               BTRFS_SUPER_INFO_SIZE -
2807                                               BTRFS_CSUM_SIZE);
2808                         btrfs_csum_final(crc, sb->csum);
2809
2810                         /*
2811                          * one reference for us, and we leave it for the
2812                          * caller
2813                          */
2814                         bh = __getblk(device->bdev, bytenr / 4096,
2815                                       BTRFS_SUPER_INFO_SIZE);
2816                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2817
2818                         /* one reference for submit_bh */
2819                         get_bh(bh);
2820
2821                         set_buffer_uptodate(bh);
2822                         lock_buffer(bh);
2823                         bh->b_end_io = btrfs_end_buffer_write_sync;
2824                         bh->b_private = device;
2825                 }
2826
2827                 /*
2828                  * we fua the first super.  The others we allow
2829                  * to go down lazy.
2830                  */
2831                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2832                 if (ret)
2833                         errors++;
2834         }
2835         return errors < i ? 0 : -1;
2836 }
2837
2838 /*
2839  * endio for the write_dev_flush, this will wake anyone waiting
2840  * for the barrier when it is done
2841  */
2842 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2843 {
2844         if (err) {
2845                 if (err == -EOPNOTSUPP)
2846                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2847                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2848         }
2849         if (bio->bi_private)
2850                 complete(bio->bi_private);
2851         bio_put(bio);
2852 }
2853
2854 /*
2855  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2856  * sent down.  With wait == 1, it waits for the previous flush.
2857  *
2858  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2859  * capable
2860  */
2861 static int write_dev_flush(struct btrfs_device *device, int wait)
2862 {
2863         struct bio *bio;
2864         int ret = 0;
2865
2866         if (device->nobarriers)
2867                 return 0;
2868
2869         if (wait) {
2870                 bio = device->flush_bio;
2871                 if (!bio)
2872                         return 0;
2873
2874                 wait_for_completion(&device->flush_wait);
2875
2876                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2877                         printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2878                                       rcu_str_deref(device->name));
2879                         device->nobarriers = 1;
2880                 }
2881                 if (!bio_flagged(bio, BIO_UPTODATE)) {
2882                         ret = -EIO;
2883                         if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2884                                 btrfs_dev_stat_inc_and_print(device,
2885                                         BTRFS_DEV_STAT_FLUSH_ERRS);
2886                 }
2887
2888                 /* drop the reference from the wait == 0 run */
2889                 bio_put(bio);
2890                 device->flush_bio = NULL;
2891
2892                 return ret;
2893         }
2894
2895         /*
2896          * one reference for us, and we leave it for the
2897          * caller
2898          */
2899         device->flush_bio = NULL;
2900         bio = bio_alloc(GFP_NOFS, 0);
2901         if (!bio)
2902                 return -ENOMEM;
2903
2904         bio->bi_end_io = btrfs_end_empty_barrier;
2905         bio->bi_bdev = device->bdev;
2906         init_completion(&device->flush_wait);
2907         bio->bi_private = &device->flush_wait;
2908         device->flush_bio = bio;
2909
2910         bio_get(bio);
2911         btrfsic_submit_bio(WRITE_FLUSH, bio);
2912
2913         return 0;
2914 }
2915
2916 /*
2917  * send an empty flush down to each device in parallel,
2918  * then wait for them
2919  */
2920 static int barrier_all_devices(struct btrfs_fs_info *info)
2921 {
2922         struct list_head *head;
2923         struct btrfs_device *dev;
2924         int errors = 0;
2925         int ret;
2926
2927         /* send down all the barriers */
2928         head = &info->fs_devices->devices;
2929         list_for_each_entry_rcu(dev, head, dev_list) {
2930                 if (!dev->bdev) {
2931                         errors++;
2932                         continue;
2933                 }
2934                 if (!dev->in_fs_metadata || !dev->writeable)
2935                         continue;
2936
2937                 ret = write_dev_flush(dev, 0);
2938                 if (ret)
2939                         errors++;
2940         }
2941
2942         /* wait for all the barriers */
2943         list_for_each_entry_rcu(dev, head, dev_list) {
2944                 if (!dev->bdev) {
2945                         errors++;
2946                         continue;
2947                 }
2948                 if (!dev->in_fs_metadata || !dev->writeable)
2949                         continue;
2950
2951                 ret = write_dev_flush(dev, 1);
2952                 if (ret)
2953                         errors++;
2954         }
2955         if (errors)
2956                 return -EIO;
2957         return 0;
2958 }
2959
2960 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2961 {
2962         struct list_head *head;
2963         struct btrfs_device *dev;
2964         struct btrfs_super_block *sb;
2965         struct btrfs_dev_item *dev_item;
2966         int ret;
2967         int do_barriers;
2968         int max_errors;
2969         int total_errors = 0;
2970         u64 flags;
2971
2972         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2973         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2974         backup_super_roots(root->fs_info);
2975
2976         sb = root->fs_info->super_for_commit;
2977         dev_item = &sb->dev_item;
2978
2979         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2980         head = &root->fs_info->fs_devices->devices;
2981
2982         if (do_barriers)
2983                 barrier_all_devices(root->fs_info);
2984
2985         list_for_each_entry_rcu(dev, head, dev_list) {
2986                 if (!dev->bdev) {
2987                         total_errors++;
2988                         continue;
2989                 }
2990                 if (!dev->in_fs_metadata || !dev->writeable)
2991                         continue;
2992
2993                 btrfs_set_stack_device_generation(dev_item, 0);
2994                 btrfs_set_stack_device_type(dev_item, dev->type);
2995                 btrfs_set_stack_device_id(dev_item, dev->devid);
2996                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2997                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2998                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2999                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3000                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3001                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3002                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3003
3004                 flags = btrfs_super_flags(sb);
3005                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3006
3007                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3008                 if (ret)
3009                         total_errors++;
3010         }
3011         if (total_errors > max_errors) {
3012                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3013                        total_errors);
3014
3015                 /* This shouldn't happen. FUA is masked off if unsupported */
3016                 BUG();
3017         }
3018
3019         total_errors = 0;
3020         list_for_each_entry_rcu(dev, head, dev_list) {
3021                 if (!dev->bdev)
3022                         continue;
3023                 if (!dev->in_fs_metadata || !dev->writeable)
3024                         continue;
3025
3026                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3027                 if (ret)
3028                         total_errors++;
3029         }
3030         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3031         if (total_errors > max_errors) {
3032                 btrfs_error(root->fs_info, -EIO,
3033                             "%d errors while writing supers", total_errors);
3034                 return -EIO;
3035         }
3036         return 0;
3037 }
3038
3039 int write_ctree_super(struct btrfs_trans_handle *trans,
3040                       struct btrfs_root *root, int max_mirrors)
3041 {
3042         int ret;
3043
3044         ret = write_all_supers(root, max_mirrors);
3045         return ret;
3046 }
3047
3048 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3049 {
3050         spin_lock(&fs_info->fs_roots_radix_lock);
3051         radix_tree_delete(&fs_info->fs_roots_radix,
3052                           (unsigned long)root->root_key.objectid);
3053         spin_unlock(&fs_info->fs_roots_radix_lock);
3054
3055         if (btrfs_root_refs(&root->root_item) == 0)
3056                 synchronize_srcu(&fs_info->subvol_srcu);
3057
3058         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3059         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3060         free_fs_root(root);
3061 }
3062
3063 static void free_fs_root(struct btrfs_root *root)
3064 {
3065         iput(root->cache_inode);
3066         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3067         if (root->anon_dev)
3068                 free_anon_bdev(root->anon_dev);
3069         free_extent_buffer(root->node);
3070         free_extent_buffer(root->commit_root);
3071         kfree(root->free_ino_ctl);
3072         kfree(root->free_ino_pinned);
3073         kfree(root->name);
3074         kfree(root);
3075 }
3076
3077 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3078 {
3079         int ret;
3080         struct btrfs_root *gang[8];
3081         int i;
3082
3083         while (!list_empty(&fs_info->dead_roots)) {
3084                 gang[0] = list_entry(fs_info->dead_roots.next,
3085                                      struct btrfs_root, root_list);
3086                 list_del(&gang[0]->root_list);
3087
3088                 if (gang[0]->in_radix) {
3089                         btrfs_free_fs_root(fs_info, gang[0]);
3090                 } else {
3091                         free_extent_buffer(gang[0]->node);
3092                         free_extent_buffer(gang[0]->commit_root);
3093                         kfree(gang[0]);
3094                 }
3095         }
3096
3097         while (1) {
3098                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3099                                              (void **)gang, 0,
3100                                              ARRAY_SIZE(gang));
3101                 if (!ret)
3102                         break;
3103                 for (i = 0; i < ret; i++)
3104                         btrfs_free_fs_root(fs_info, gang[i]);
3105         }
3106 }
3107
3108 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3109 {
3110         u64 root_objectid = 0;
3111         struct btrfs_root *gang[8];
3112         int i;
3113         int ret;
3114
3115         while (1) {
3116                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3117                                              (void **)gang, root_objectid,
3118                                              ARRAY_SIZE(gang));
3119                 if (!ret)
3120                         break;
3121
3122                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3123                 for (i = 0; i < ret; i++) {
3124                         int err;
3125
3126                         root_objectid = gang[i]->root_key.objectid;
3127                         err = btrfs_orphan_cleanup(gang[i]);
3128                         if (err)
3129                                 return err;
3130                 }
3131                 root_objectid++;
3132         }
3133         return 0;
3134 }
3135
3136 int btrfs_commit_super(struct btrfs_root *root)
3137 {
3138         struct btrfs_trans_handle *trans;
3139         int ret;
3140
3141         mutex_lock(&root->fs_info->cleaner_mutex);
3142         btrfs_run_delayed_iputs(root);
3143         btrfs_clean_old_snapshots(root);
3144         mutex_unlock(&root->fs_info->cleaner_mutex);
3145
3146         /* wait until ongoing cleanup work done */
3147         down_write(&root->fs_info->cleanup_work_sem);
3148         up_write(&root->fs_info->cleanup_work_sem);
3149
3150         trans = btrfs_join_transaction(root);
3151         if (IS_ERR(trans))
3152                 return PTR_ERR(trans);
3153         ret = btrfs_commit_transaction(trans, root);
3154         if (ret)
3155                 return ret;
3156         /* run commit again to drop the original snapshot */
3157         trans = btrfs_join_transaction(root);
3158         if (IS_ERR(trans))
3159                 return PTR_ERR(trans);
3160         ret = btrfs_commit_transaction(trans, root);
3161         if (ret)
3162                 return ret;
3163         ret = btrfs_write_and_wait_transaction(NULL, root);
3164         if (ret) {
3165                 btrfs_error(root->fs_info, ret,
3166                             "Failed to sync btree inode to disk.");
3167                 return ret;
3168         }
3169
3170         ret = write_ctree_super(NULL, root, 0);
3171         return ret;
3172 }
3173
3174 int close_ctree(struct btrfs_root *root)
3175 {
3176         struct btrfs_fs_info *fs_info = root->fs_info;
3177         int ret;
3178
3179         fs_info->closing = 1;
3180         smp_mb();
3181
3182         /* pause restriper - we want to resume on mount */
3183         btrfs_pause_balance(root->fs_info);
3184
3185         btrfs_scrub_cancel(root);
3186
3187         /* wait for any defraggers to finish */
3188         wait_event(fs_info->transaction_wait,
3189                    (atomic_read(&fs_info->defrag_running) == 0));
3190
3191         /* clear out the rbtree of defraggable inodes */
3192         btrfs_run_defrag_inodes(fs_info);
3193
3194         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3195                 ret = btrfs_commit_super(root);
3196                 if (ret)
3197                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3198         }
3199
3200         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3201                 btrfs_error_commit_super(root);
3202
3203         btrfs_put_block_group_cache(fs_info);
3204
3205         kthread_stop(fs_info->transaction_kthread);
3206         kthread_stop(fs_info->cleaner_kthread);
3207
3208         fs_info->closing = 2;
3209         smp_mb();
3210
3211         btrfs_free_qgroup_config(root->fs_info);
3212
3213         if (fs_info->delalloc_bytes) {
3214                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3215                        (unsigned long long)fs_info->delalloc_bytes);
3216         }
3217         if (fs_info->total_ref_cache_size) {
3218                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3219                        (unsigned long long)fs_info->total_ref_cache_size);
3220         }
3221
3222         free_extent_buffer(fs_info->extent_root->node);
3223         free_extent_buffer(fs_info->extent_root->commit_root);
3224         free_extent_buffer(fs_info->tree_root->node);
3225         free_extent_buffer(fs_info->tree_root->commit_root);
3226         free_extent_buffer(fs_info->chunk_root->node);
3227         free_extent_buffer(fs_info->chunk_root->commit_root);
3228         free_extent_buffer(fs_info->dev_root->node);
3229         free_extent_buffer(fs_info->dev_root->commit_root);
3230         free_extent_buffer(fs_info->csum_root->node);
3231         free_extent_buffer(fs_info->csum_root->commit_root);
3232         if (fs_info->quota_root) {
3233                 free_extent_buffer(fs_info->quota_root->node);
3234                 free_extent_buffer(fs_info->quota_root->commit_root);
3235         }
3236
3237         btrfs_free_block_groups(fs_info);
3238
3239         del_fs_roots(fs_info);
3240
3241         iput(fs_info->btree_inode);
3242
3243         btrfs_stop_workers(&fs_info->generic_worker);
3244         btrfs_stop_workers(&fs_info->fixup_workers);
3245         btrfs_stop_workers(&fs_info->delalloc_workers);
3246         btrfs_stop_workers(&fs_info->workers);
3247         btrfs_stop_workers(&fs_info->endio_workers);
3248         btrfs_stop_workers(&fs_info->endio_meta_workers);
3249         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3250         btrfs_stop_workers(&fs_info->endio_write_workers);
3251         btrfs_stop_workers(&fs_info->endio_freespace_worker);
3252         btrfs_stop_workers(&fs_info->submit_workers);
3253         btrfs_stop_workers(&fs_info->delayed_workers);
3254         btrfs_stop_workers(&fs_info->caching_workers);
3255         btrfs_stop_workers(&fs_info->readahead_workers);
3256
3257 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3258         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3259                 btrfsic_unmount(root, fs_info->fs_devices);
3260 #endif
3261
3262         btrfs_close_devices(fs_info->fs_devices);
3263         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3264
3265         bdi_destroy(&fs_info->bdi);
3266         cleanup_srcu_struct(&fs_info->subvol_srcu);
3267
3268         return 0;
3269 }
3270
3271 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3272                           int atomic)
3273 {
3274         int ret;
3275         struct inode *btree_inode = buf->pages[0]->mapping->host;
3276
3277         ret = extent_buffer_uptodate(buf);
3278         if (!ret)
3279                 return ret;
3280
3281         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3282                                     parent_transid, atomic);
3283         if (ret == -EAGAIN)
3284                 return ret;
3285         return !ret;
3286 }
3287
3288 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3289 {
3290         return set_extent_buffer_uptodate(buf);
3291 }
3292
3293 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3294 {
3295         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3296         u64 transid = btrfs_header_generation(buf);
3297         int was_dirty;
3298
3299         btrfs_assert_tree_locked(buf);
3300         if (transid != root->fs_info->generation) {
3301                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3302                        "found %llu running %llu\n",
3303                         (unsigned long long)buf->start,
3304                         (unsigned long long)transid,
3305                         (unsigned long long)root->fs_info->generation);
3306                 WARN_ON(1);
3307         }
3308         was_dirty = set_extent_buffer_dirty(buf);
3309         if (!was_dirty) {
3310                 spin_lock(&root->fs_info->delalloc_lock);
3311                 root->fs_info->dirty_metadata_bytes += buf->len;
3312                 spin_unlock(&root->fs_info->delalloc_lock);
3313         }
3314 }
3315
3316 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3317 {
3318         /*
3319          * looks as though older kernels can get into trouble with
3320          * this code, they end up stuck in balance_dirty_pages forever
3321          */
3322         u64 num_dirty;
3323         unsigned long thresh = 32 * 1024 * 1024;
3324
3325         if (current->flags & PF_MEMALLOC)
3326                 return;
3327
3328         btrfs_balance_delayed_items(root);
3329
3330         num_dirty = root->fs_info->dirty_metadata_bytes;
3331
3332         if (num_dirty > thresh) {
3333                 balance_dirty_pages_ratelimited_nr(
3334                                    root->fs_info->btree_inode->i_mapping, 1);
3335         }
3336         return;
3337 }
3338
3339 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3340 {
3341         /*
3342          * looks as though older kernels can get into trouble with
3343          * this code, they end up stuck in balance_dirty_pages forever
3344          */
3345         u64 num_dirty;
3346         unsigned long thresh = 32 * 1024 * 1024;
3347
3348         if (current->flags & PF_MEMALLOC)
3349                 return;
3350
3351         num_dirty = root->fs_info->dirty_metadata_bytes;
3352
3353         if (num_dirty > thresh) {
3354                 balance_dirty_pages_ratelimited_nr(
3355                                    root->fs_info->btree_inode->i_mapping, 1);
3356         }
3357         return;
3358 }
3359
3360 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3361 {
3362         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3363         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3364 }
3365
3366 int btree_lock_page_hook(struct page *page, void *data,
3367                                 void (*flush_fn)(void *))
3368 {
3369         struct inode *inode = page->mapping->host;
3370         struct btrfs_root *root = BTRFS_I(inode)->root;
3371         struct extent_buffer *eb;
3372
3373         /*
3374          * We culled this eb but the page is still hanging out on the mapping,
3375          * carry on.
3376          */
3377         if (!PagePrivate(page))
3378                 goto out;
3379
3380         eb = (struct extent_buffer *)page->private;
3381         if (!eb) {
3382                 WARN_ON(1);
3383                 goto out;
3384         }
3385         if (page != eb->pages[0])
3386                 goto out;
3387
3388         if (!btrfs_try_tree_write_lock(eb)) {
3389                 flush_fn(data);
3390                 btrfs_tree_lock(eb);
3391         }
3392         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3393
3394         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3395                 spin_lock(&root->fs_info->delalloc_lock);
3396                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3397                         root->fs_info->dirty_metadata_bytes -= eb->len;
3398                 else
3399                         WARN_ON(1);
3400                 spin_unlock(&root->fs_info->delalloc_lock);
3401         }
3402
3403         btrfs_tree_unlock(eb);
3404 out:
3405         if (!trylock_page(page)) {
3406                 flush_fn(data);
3407                 lock_page(page);
3408         }
3409         return 0;
3410 }
3411
3412 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3413                               int read_only)
3414 {
3415         if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3416                 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3417                 return -EINVAL;
3418         }
3419
3420         if (read_only)
3421                 return 0;
3422
3423         return 0;
3424 }
3425
3426 void btrfs_error_commit_super(struct btrfs_root *root)
3427 {
3428         mutex_lock(&root->fs_info->cleaner_mutex);
3429         btrfs_run_delayed_iputs(root);
3430         mutex_unlock(&root->fs_info->cleaner_mutex);
3431
3432         down_write(&root->fs_info->cleanup_work_sem);
3433         up_write(&root->fs_info->cleanup_work_sem);
3434
3435         /* cleanup FS via transaction */
3436         btrfs_cleanup_transaction(root);
3437 }
3438
3439 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3440 {
3441         struct btrfs_inode *btrfs_inode;
3442         struct list_head splice;
3443
3444         INIT_LIST_HEAD(&splice);
3445
3446         mutex_lock(&root->fs_info->ordered_operations_mutex);
3447         spin_lock(&root->fs_info->ordered_extent_lock);
3448
3449         list_splice_init(&root->fs_info->ordered_operations, &splice);
3450         while (!list_empty(&splice)) {
3451                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3452                                          ordered_operations);
3453
3454                 list_del_init(&btrfs_inode->ordered_operations);
3455
3456                 btrfs_invalidate_inodes(btrfs_inode->root);
3457         }
3458
3459         spin_unlock(&root->fs_info->ordered_extent_lock);
3460         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3461 }
3462
3463 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3464 {
3465         struct list_head splice;
3466         struct btrfs_ordered_extent *ordered;
3467         struct inode *inode;
3468
3469         INIT_LIST_HEAD(&splice);
3470
3471         spin_lock(&root->fs_info->ordered_extent_lock);
3472
3473         list_splice_init(&root->fs_info->ordered_extents, &splice);
3474         while (!list_empty(&splice)) {
3475                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3476                                      root_extent_list);
3477
3478                 list_del_init(&ordered->root_extent_list);
3479                 atomic_inc(&ordered->refs);
3480
3481                 /* the inode may be getting freed (in sys_unlink path). */
3482                 inode = igrab(ordered->inode);
3483
3484                 spin_unlock(&root->fs_info->ordered_extent_lock);
3485                 if (inode)
3486                         iput(inode);
3487
3488                 atomic_set(&ordered->refs, 1);
3489                 btrfs_put_ordered_extent(ordered);
3490
3491                 spin_lock(&root->fs_info->ordered_extent_lock);
3492         }
3493
3494         spin_unlock(&root->fs_info->ordered_extent_lock);
3495 }
3496
3497 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3498                                struct btrfs_root *root)
3499 {
3500         struct rb_node *node;
3501         struct btrfs_delayed_ref_root *delayed_refs;
3502         struct btrfs_delayed_ref_node *ref;
3503         int ret = 0;
3504
3505         delayed_refs = &trans->delayed_refs;
3506
3507         spin_lock(&delayed_refs->lock);
3508         if (delayed_refs->num_entries == 0) {
3509                 spin_unlock(&delayed_refs->lock);
3510                 printk(KERN_INFO "delayed_refs has NO entry\n");
3511                 return ret;
3512         }
3513
3514         while ((node = rb_first(&delayed_refs->root)) != NULL) {
3515                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3516
3517                 atomic_set(&ref->refs, 1);
3518                 if (btrfs_delayed_ref_is_head(ref)) {
3519                         struct btrfs_delayed_ref_head *head;
3520
3521                         head = btrfs_delayed_node_to_head(ref);
3522                         if (!mutex_trylock(&head->mutex)) {
3523                                 atomic_inc(&ref->refs);
3524                                 spin_unlock(&delayed_refs->lock);
3525
3526                                 /* Need to wait for the delayed ref to run */
3527                                 mutex_lock(&head->mutex);
3528                                 mutex_unlock(&head->mutex);
3529                                 btrfs_put_delayed_ref(ref);
3530
3531                                 spin_lock(&delayed_refs->lock);
3532                                 continue;
3533                         }
3534
3535                         kfree(head->extent_op);
3536                         delayed_refs->num_heads--;
3537                         if (list_empty(&head->cluster))
3538                                 delayed_refs->num_heads_ready--;
3539                         list_del_init(&head->cluster);
3540                 }
3541                 ref->in_tree = 0;
3542                 rb_erase(&ref->rb_node, &delayed_refs->root);
3543                 delayed_refs->num_entries--;
3544
3545                 spin_unlock(&delayed_refs->lock);
3546                 btrfs_put_delayed_ref(ref);
3547
3548                 cond_resched();
3549                 spin_lock(&delayed_refs->lock);
3550         }
3551
3552         spin_unlock(&delayed_refs->lock);
3553
3554         return ret;
3555 }
3556
3557 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3558 {
3559         struct btrfs_pending_snapshot *snapshot;
3560         struct list_head splice;
3561
3562         INIT_LIST_HEAD(&splice);
3563
3564         list_splice_init(&t->pending_snapshots, &splice);
3565
3566         while (!list_empty(&splice)) {
3567                 snapshot = list_entry(splice.next,
3568                                       struct btrfs_pending_snapshot,
3569                                       list);
3570
3571                 list_del_init(&snapshot->list);
3572
3573                 kfree(snapshot);
3574         }
3575 }
3576
3577 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3578 {
3579         struct btrfs_inode *btrfs_inode;
3580         struct list_head splice;
3581
3582         INIT_LIST_HEAD(&splice);
3583
3584         spin_lock(&root->fs_info->delalloc_lock);
3585         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3586
3587         while (!list_empty(&splice)) {
3588                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3589                                     delalloc_inodes);
3590
3591                 list_del_init(&btrfs_inode->delalloc_inodes);
3592
3593                 btrfs_invalidate_inodes(btrfs_inode->root);
3594         }
3595
3596         spin_unlock(&root->fs_info->delalloc_lock);
3597 }
3598
3599 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3600                                         struct extent_io_tree *dirty_pages,
3601                                         int mark)
3602 {
3603         int ret;
3604         struct page *page;
3605         struct inode *btree_inode = root->fs_info->btree_inode;
3606         struct extent_buffer *eb;
3607         u64 start = 0;
3608         u64 end;
3609         u64 offset;
3610         unsigned long index;
3611
3612         while (1) {
3613                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3614                                             mark);
3615                 if (ret)
3616                         break;
3617
3618                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3619                 while (start <= end) {
3620                         index = start >> PAGE_CACHE_SHIFT;
3621                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3622                         page = find_get_page(btree_inode->i_mapping, index);
3623                         if (!page)
3624                                 continue;
3625                         offset = page_offset(page);
3626
3627                         spin_lock(&dirty_pages->buffer_lock);
3628                         eb = radix_tree_lookup(
3629                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3630                                                offset >> PAGE_CACHE_SHIFT);
3631                         spin_unlock(&dirty_pages->buffer_lock);
3632                         if (eb)
3633                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3634                                                          &eb->bflags);
3635                         if (PageWriteback(page))
3636                                 end_page_writeback(page);
3637
3638                         lock_page(page);
3639                         if (PageDirty(page)) {
3640                                 clear_page_dirty_for_io(page);
3641                                 spin_lock_irq(&page->mapping->tree_lock);
3642                                 radix_tree_tag_clear(&page->mapping->page_tree,
3643                                                         page_index(page),
3644                                                         PAGECACHE_TAG_DIRTY);
3645                                 spin_unlock_irq(&page->mapping->tree_lock);
3646                         }
3647
3648                         unlock_page(page);
3649                         page_cache_release(page);
3650                 }
3651         }
3652
3653         return ret;
3654 }
3655
3656 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3657                                        struct extent_io_tree *pinned_extents)
3658 {
3659         struct extent_io_tree *unpin;
3660         u64 start;
3661         u64 end;
3662         int ret;
3663         bool loop = true;
3664
3665         unpin = pinned_extents;
3666 again:
3667         while (1) {
3668                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3669                                             EXTENT_DIRTY);
3670                 if (ret)
3671                         break;
3672
3673                 /* opt_discard */
3674                 if (btrfs_test_opt(root, DISCARD))
3675                         ret = btrfs_error_discard_extent(root, start,
3676                                                          end + 1 - start,
3677                                                          NULL);
3678
3679                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3680                 btrfs_error_unpin_extent_range(root, start, end);
3681                 cond_resched();
3682         }
3683
3684         if (loop) {
3685                 if (unpin == &root->fs_info->freed_extents[0])
3686                         unpin = &root->fs_info->freed_extents[1];
3687                 else
3688                         unpin = &root->fs_info->freed_extents[0];
3689                 loop = false;
3690                 goto again;
3691         }
3692
3693         return 0;
3694 }
3695
3696 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3697                                    struct btrfs_root *root)
3698 {
3699         btrfs_destroy_delayed_refs(cur_trans, root);
3700         btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3701                                 cur_trans->dirty_pages.dirty_bytes);
3702
3703         /* FIXME: cleanup wait for commit */
3704         cur_trans->in_commit = 1;
3705         cur_trans->blocked = 1;
3706         wake_up(&root->fs_info->transaction_blocked_wait);
3707
3708         cur_trans->blocked = 0;
3709         wake_up(&root->fs_info->transaction_wait);
3710
3711         cur_trans->commit_done = 1;
3712         wake_up(&cur_trans->commit_wait);
3713
3714         btrfs_destroy_delayed_inodes(root);
3715         btrfs_assert_delayed_root_empty(root);
3716
3717         btrfs_destroy_pending_snapshots(cur_trans);
3718
3719         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3720                                      EXTENT_DIRTY);
3721         btrfs_destroy_pinned_extent(root,
3722                                     root->fs_info->pinned_extents);
3723
3724         /*
3725         memset(cur_trans, 0, sizeof(*cur_trans));
3726         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3727         */
3728 }
3729
3730 int btrfs_cleanup_transaction(struct btrfs_root *root)
3731 {
3732         struct btrfs_transaction *t;
3733         LIST_HEAD(list);
3734
3735         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3736
3737         spin_lock(&root->fs_info->trans_lock);
3738         list_splice_init(&root->fs_info->trans_list, &list);
3739         root->fs_info->trans_no_join = 1;
3740         spin_unlock(&root->fs_info->trans_lock);
3741
3742         while (!list_empty(&list)) {
3743                 t = list_entry(list.next, struct btrfs_transaction, list);
3744                 if (!t)
3745                         break;
3746
3747                 btrfs_destroy_ordered_operations(root);
3748
3749                 btrfs_destroy_ordered_extents(root);
3750
3751                 btrfs_destroy_delayed_refs(t, root);
3752
3753                 btrfs_block_rsv_release(root,
3754                                         &root->fs_info->trans_block_rsv,
3755                                         t->dirty_pages.dirty_bytes);
3756
3757                 /* FIXME: cleanup wait for commit */
3758                 t->in_commit = 1;
3759                 t->blocked = 1;
3760                 smp_mb();
3761                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3762                         wake_up(&root->fs_info->transaction_blocked_wait);
3763
3764                 t->blocked = 0;
3765                 smp_mb();
3766                 if (waitqueue_active(&root->fs_info->transaction_wait))
3767                         wake_up(&root->fs_info->transaction_wait);
3768
3769                 t->commit_done = 1;
3770                 smp_mb();
3771                 if (waitqueue_active(&t->commit_wait))
3772                         wake_up(&t->commit_wait);
3773
3774                 btrfs_destroy_delayed_inodes(root);
3775                 btrfs_assert_delayed_root_empty(root);
3776
3777                 btrfs_destroy_pending_snapshots(t);
3778
3779                 btrfs_destroy_delalloc_inodes(root);
3780
3781                 spin_lock(&root->fs_info->trans_lock);
3782                 root->fs_info->running_transaction = NULL;
3783                 spin_unlock(&root->fs_info->trans_lock);
3784
3785                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3786                                              EXTENT_DIRTY);
3787
3788                 btrfs_destroy_pinned_extent(root,
3789                                             root->fs_info->pinned_extents);
3790
3791                 atomic_set(&t->use_count, 0);
3792                 list_del_init(&t->list);
3793                 memset(t, 0, sizeof(*t));
3794                 kmem_cache_free(btrfs_transaction_cachep, t);
3795         }
3796
3797         spin_lock(&root->fs_info->trans_lock);
3798         root->fs_info->trans_no_join = 0;
3799         spin_unlock(&root->fs_info->trans_lock);
3800         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3801
3802         return 0;
3803 }
3804
3805 static struct extent_io_ops btree_extent_io_ops = {
3806         .write_cache_pages_lock_hook = btree_lock_page_hook,
3807         .readpage_end_io_hook = btree_readpage_end_io_hook,
3808         .readpage_io_failed_hook = btree_io_failed_hook,
3809         .submit_bio_hook = btree_submit_bio_hook,
3810         /* note we're sharing with inode.c for the merge bio hook */
3811         .merge_bio_hook = btrfs_merge_bio_hook,
3812 };