Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-drm-fsl-dcu.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 && !verify_parent_transid(io_tree, eb,
381                                                    parent_transid, 0))
382                         break;
383
384                 /*
385                  * This buffer's crc is fine, but its contents are corrupted, so
386                  * there is no reason to read the other copies, they won't be
387                  * any less wrong.
388                  */
389                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
390                         break;
391
392                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
393                                               eb->start, eb->len);
394                 if (num_copies == 1)
395                         break;
396
397                 if (!failed_mirror) {
398                         failed = 1;
399                         failed_mirror = eb->read_mirror;
400                 }
401
402                 mirror_num++;
403                 if (mirror_num == failed_mirror)
404                         mirror_num++;
405
406                 if (mirror_num > num_copies)
407                         break;
408         }
409
410         if (failed && !ret && failed_mirror)
411                 repair_eb_io_failure(root, eb, failed_mirror);
412
413         return ret;
414 }
415
416 /*
417  * checksum a dirty tree block before IO.  This has extra checks to make sure
418  * we only fill in the checksum field in the first page of a multi-page block
419  */
420
421 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
422 {
423         struct extent_io_tree *tree;
424         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
425         u64 found_start;
426         struct extent_buffer *eb;
427
428         tree = &BTRFS_I(page->mapping->host)->io_tree;
429
430         eb = (struct extent_buffer *)page->private;
431         if (page != eb->pages[0])
432                 return 0;
433         found_start = btrfs_header_bytenr(eb);
434         if (found_start != start) {
435                 WARN_ON(1);
436                 return 0;
437         }
438         if (eb->pages[0] != page) {
439                 WARN_ON(1);
440                 return 0;
441         }
442         if (!PageUptodate(page)) {
443                 WARN_ON(1);
444                 return 0;
445         }
446         csum_tree_block(root, eb, 0);
447         return 0;
448 }
449
450 static int check_tree_block_fsid(struct btrfs_root *root,
451                                  struct extent_buffer *eb)
452 {
453         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
454         u8 fsid[BTRFS_UUID_SIZE];
455         int ret = 1;
456
457         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
458                            BTRFS_FSID_SIZE);
459         while (fs_devices) {
460                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
461                         ret = 0;
462                         break;
463                 }
464                 fs_devices = fs_devices->seed;
465         }
466         return ret;
467 }
468
469 #define CORRUPT(reason, eb, root, slot)                         \
470         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471                "root=%llu, slot=%d\n", reason,                  \
472                (unsigned long long)btrfs_header_bytenr(eb),     \
473                (unsigned long long)root->objectid, slot)
474
475 static noinline int check_leaf(struct btrfs_root *root,
476                                struct extent_buffer *leaf)
477 {
478         struct btrfs_key key;
479         struct btrfs_key leaf_key;
480         u32 nritems = btrfs_header_nritems(leaf);
481         int slot;
482
483         if (nritems == 0)
484                 return 0;
485
486         /* Check the 0 item */
487         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
488             BTRFS_LEAF_DATA_SIZE(root)) {
489                 CORRUPT("invalid item offset size pair", leaf, root, 0);
490                 return -EIO;
491         }
492
493         /*
494          * Check to make sure each items keys are in the correct order and their
495          * offsets make sense.  We only have to loop through nritems-1 because
496          * we check the current slot against the next slot, which verifies the
497          * next slot's offset+size makes sense and that the current's slot
498          * offset is correct.
499          */
500         for (slot = 0; slot < nritems - 1; slot++) {
501                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
502                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
503
504                 /* Make sure the keys are in the right order */
505                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
506                         CORRUPT("bad key order", leaf, root, slot);
507                         return -EIO;
508                 }
509
510                 /*
511                  * Make sure the offset and ends are right, remember that the
512                  * item data starts at the end of the leaf and grows towards the
513                  * front.
514                  */
515                 if (btrfs_item_offset_nr(leaf, slot) !=
516                         btrfs_item_end_nr(leaf, slot + 1)) {
517                         CORRUPT("slot offset bad", leaf, root, slot);
518                         return -EIO;
519                 }
520
521                 /*
522                  * Check to make sure that we don't point outside of the leaf,
523                  * just incase all the items are consistent to eachother, but
524                  * all point outside of the leaf.
525                  */
526                 if (btrfs_item_end_nr(leaf, slot) >
527                     BTRFS_LEAF_DATA_SIZE(root)) {
528                         CORRUPT("slot end outside of leaf", leaf, root, slot);
529                         return -EIO;
530                 }
531         }
532
533         return 0;
534 }
535
536 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
537                                        struct page *page, int max_walk)
538 {
539         struct extent_buffer *eb;
540         u64 start = page_offset(page);
541         u64 target = start;
542         u64 min_start;
543
544         if (start < max_walk)
545                 min_start = 0;
546         else
547                 min_start = start - max_walk;
548
549         while (start >= min_start) {
550                 eb = find_extent_buffer(tree, start, 0);
551                 if (eb) {
552                         /*
553                          * we found an extent buffer and it contains our page
554                          * horray!
555                          */
556                         if (eb->start <= target &&
557                             eb->start + eb->len > target)
558                                 return eb;
559
560                         /* we found an extent buffer that wasn't for us */
561                         free_extent_buffer(eb);
562                         return NULL;
563                 }
564                 if (start == 0)
565                         break;
566                 start -= PAGE_CACHE_SIZE;
567         }
568         return NULL;
569 }
570
571 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
572                                struct extent_state *state, int mirror)
573 {
574         struct extent_io_tree *tree;
575         u64 found_start;
576         int found_level;
577         struct extent_buffer *eb;
578         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
579         int ret = 0;
580         int reads_done;
581
582         if (!page->private)
583                 goto out;
584
585         tree = &BTRFS_I(page->mapping->host)->io_tree;
586         eb = (struct extent_buffer *)page->private;
587
588         /* the pending IO might have been the only thing that kept this buffer
589          * in memory.  Make sure we have a ref for all this other checks
590          */
591         extent_buffer_get(eb);
592
593         reads_done = atomic_dec_and_test(&eb->io_pages);
594         if (!reads_done)
595                 goto err;
596
597         eb->read_mirror = mirror;
598         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
599                 ret = -EIO;
600                 goto err;
601         }
602
603         found_start = btrfs_header_bytenr(eb);
604         if (found_start != eb->start) {
605                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
606                                "%llu %llu\n",
607                                (unsigned long long)found_start,
608                                (unsigned long long)eb->start);
609                 ret = -EIO;
610                 goto err;
611         }
612         if (check_tree_block_fsid(root, eb)) {
613                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
614                                (unsigned long long)eb->start);
615                 ret = -EIO;
616                 goto err;
617         }
618         found_level = btrfs_header_level(eb);
619
620         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621                                        eb, found_level);
622
623         ret = csum_tree_block(root, eb, 1);
624         if (ret) {
625                 ret = -EIO;
626                 goto err;
627         }
628
629         /*
630          * If this is a leaf block and it is corrupt, set the corrupt bit so
631          * that we don't try and read the other copies of this block, just
632          * return -EIO.
633          */
634         if (found_level == 0 && check_leaf(root, eb)) {
635                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636                 ret = -EIO;
637         }
638
639         if (!ret)
640                 set_extent_buffer_uptodate(eb);
641 err:
642         if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
643                 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
644                 btree_readahead_hook(root, eb, eb->start, ret);
645         }
646
647         if (ret)
648                 clear_extent_buffer_uptodate(eb);
649         free_extent_buffer(eb);
650 out:
651         return ret;
652 }
653
654 static int btree_io_failed_hook(struct page *page, int failed_mirror)
655 {
656         struct extent_buffer *eb;
657         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
658
659         eb = (struct extent_buffer *)page->private;
660         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
661         eb->read_mirror = failed_mirror;
662         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663                 btree_readahead_hook(root, eb, eb->start, -EIO);
664         return -EIO;    /* we fixed nothing */
665 }
666
667 static void end_workqueue_bio(struct bio *bio, int err)
668 {
669         struct end_io_wq *end_io_wq = bio->bi_private;
670         struct btrfs_fs_info *fs_info;
671
672         fs_info = end_io_wq->info;
673         end_io_wq->error = err;
674         end_io_wq->work.func = end_workqueue_fn;
675         end_io_wq->work.flags = 0;
676
677         if (bio->bi_rw & REQ_WRITE) {
678                 if (end_io_wq->metadata == 1)
679                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
680                                            &end_io_wq->work);
681                 else if (end_io_wq->metadata == 2)
682                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
683                                            &end_io_wq->work);
684                 else
685                         btrfs_queue_worker(&fs_info->endio_write_workers,
686                                            &end_io_wq->work);
687         } else {
688                 if (end_io_wq->metadata)
689                         btrfs_queue_worker(&fs_info->endio_meta_workers,
690                                            &end_io_wq->work);
691                 else
692                         btrfs_queue_worker(&fs_info->endio_workers,
693                                            &end_io_wq->work);
694         }
695 }
696
697 /*
698  * For the metadata arg you want
699  *
700  * 0 - if data
701  * 1 - if normal metadta
702  * 2 - if writing to the free space cache area
703  */
704 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
705                         int metadata)
706 {
707         struct end_io_wq *end_io_wq;
708         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
709         if (!end_io_wq)
710                 return -ENOMEM;
711
712         end_io_wq->private = bio->bi_private;
713         end_io_wq->end_io = bio->bi_end_io;
714         end_io_wq->info = info;
715         end_io_wq->error = 0;
716         end_io_wq->bio = bio;
717         end_io_wq->metadata = metadata;
718
719         bio->bi_private = end_io_wq;
720         bio->bi_end_io = end_workqueue_bio;
721         return 0;
722 }
723
724 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
725 {
726         unsigned long limit = min_t(unsigned long,
727                                     info->workers.max_workers,
728                                     info->fs_devices->open_devices);
729         return 256 * limit;
730 }
731
732 static void run_one_async_start(struct btrfs_work *work)
733 {
734         struct async_submit_bio *async;
735         int ret;
736
737         async = container_of(work, struct  async_submit_bio, work);
738         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
739                                       async->mirror_num, async->bio_flags,
740                                       async->bio_offset);
741         if (ret)
742                 async->error = ret;
743 }
744
745 static void run_one_async_done(struct btrfs_work *work)
746 {
747         struct btrfs_fs_info *fs_info;
748         struct async_submit_bio *async;
749         int limit;
750
751         async = container_of(work, struct  async_submit_bio, work);
752         fs_info = BTRFS_I(async->inode)->root->fs_info;
753
754         limit = btrfs_async_submit_limit(fs_info);
755         limit = limit * 2 / 3;
756
757         atomic_dec(&fs_info->nr_async_submits);
758
759         if (atomic_read(&fs_info->nr_async_submits) < limit &&
760             waitqueue_active(&fs_info->async_submit_wait))
761                 wake_up(&fs_info->async_submit_wait);
762
763         /* If an error occured we just want to clean up the bio and move on */
764         if (async->error) {
765                 bio_endio(async->bio, async->error);
766                 return;
767         }
768
769         async->submit_bio_done(async->inode, async->rw, async->bio,
770                                async->mirror_num, async->bio_flags,
771                                async->bio_offset);
772 }
773
774 static void run_one_async_free(struct btrfs_work *work)
775 {
776         struct async_submit_bio *async;
777
778         async = container_of(work, struct  async_submit_bio, work);
779         kfree(async);
780 }
781
782 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
783                         int rw, struct bio *bio, int mirror_num,
784                         unsigned long bio_flags,
785                         u64 bio_offset,
786                         extent_submit_bio_hook_t *submit_bio_start,
787                         extent_submit_bio_hook_t *submit_bio_done)
788 {
789         struct async_submit_bio *async;
790
791         async = kmalloc(sizeof(*async), GFP_NOFS);
792         if (!async)
793                 return -ENOMEM;
794
795         async->inode = inode;
796         async->rw = rw;
797         async->bio = bio;
798         async->mirror_num = mirror_num;
799         async->submit_bio_start = submit_bio_start;
800         async->submit_bio_done = submit_bio_done;
801
802         async->work.func = run_one_async_start;
803         async->work.ordered_func = run_one_async_done;
804         async->work.ordered_free = run_one_async_free;
805
806         async->work.flags = 0;
807         async->bio_flags = bio_flags;
808         async->bio_offset = bio_offset;
809
810         async->error = 0;
811
812         atomic_inc(&fs_info->nr_async_submits);
813
814         if (rw & REQ_SYNC)
815                 btrfs_set_work_high_prio(&async->work);
816
817         btrfs_queue_worker(&fs_info->workers, &async->work);
818
819         while (atomic_read(&fs_info->async_submit_draining) &&
820               atomic_read(&fs_info->nr_async_submits)) {
821                 wait_event(fs_info->async_submit_wait,
822                            (atomic_read(&fs_info->nr_async_submits) == 0));
823         }
824
825         return 0;
826 }
827
828 static int btree_csum_one_bio(struct bio *bio)
829 {
830         struct bio_vec *bvec = bio->bi_io_vec;
831         int bio_index = 0;
832         struct btrfs_root *root;
833         int ret = 0;
834
835         WARN_ON(bio->bi_vcnt <= 0);
836         while (bio_index < bio->bi_vcnt) {
837                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
838                 ret = csum_dirty_buffer(root, bvec->bv_page);
839                 if (ret)
840                         break;
841                 bio_index++;
842                 bvec++;
843         }
844         return ret;
845 }
846
847 static int __btree_submit_bio_start(struct inode *inode, int rw,
848                                     struct bio *bio, int mirror_num,
849                                     unsigned long bio_flags,
850                                     u64 bio_offset)
851 {
852         /*
853          * when we're called for a write, we're already in the async
854          * submission context.  Just jump into btrfs_map_bio
855          */
856         return btree_csum_one_bio(bio);
857 }
858
859 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
860                                  int mirror_num, unsigned long bio_flags,
861                                  u64 bio_offset)
862 {
863         /*
864          * when we're called for a write, we're already in the async
865          * submission context.  Just jump into btrfs_map_bio
866          */
867         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
868 }
869
870 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
871                                  int mirror_num, unsigned long bio_flags,
872                                  u64 bio_offset)
873 {
874         int ret;
875
876         if (!(rw & REQ_WRITE)) {
877
878                 /*
879                  * called for a read, do the setup so that checksum validation
880                  * can happen in the async kernel threads
881                  */
882                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
883                                           bio, 1);
884                 if (ret)
885                         return ret;
886                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
887                                      mirror_num, 0);
888         }
889
890         /*
891          * kthread helpers are used to submit writes so that checksumming
892          * can happen in parallel across all CPUs
893          */
894         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
895                                    inode, rw, bio, mirror_num, 0,
896                                    bio_offset,
897                                    __btree_submit_bio_start,
898                                    __btree_submit_bio_done);
899 }
900
901 #ifdef CONFIG_MIGRATION
902 static int btree_migratepage(struct address_space *mapping,
903                         struct page *newpage, struct page *page,
904                         enum migrate_mode mode)
905 {
906         /*
907          * we can't safely write a btree page from here,
908          * we haven't done the locking hook
909          */
910         if (PageDirty(page))
911                 return -EAGAIN;
912         /*
913          * Buffers may be managed in a filesystem specific way.
914          * We must have no buffers or drop them.
915          */
916         if (page_has_private(page) &&
917             !try_to_release_page(page, GFP_KERNEL))
918                 return -EAGAIN;
919         return migrate_page(mapping, newpage, page, mode);
920 }
921 #endif
922
923
924 static int btree_writepages(struct address_space *mapping,
925                             struct writeback_control *wbc)
926 {
927         struct extent_io_tree *tree;
928         tree = &BTRFS_I(mapping->host)->io_tree;
929         if (wbc->sync_mode == WB_SYNC_NONE) {
930                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931                 u64 num_dirty;
932                 unsigned long thresh = 32 * 1024 * 1024;
933
934                 if (wbc->for_kupdate)
935                         return 0;
936
937                 /* this is a bit racy, but that's ok */
938                 num_dirty = root->fs_info->dirty_metadata_bytes;
939                 if (num_dirty < thresh)
940                         return 0;
941         }
942         return btree_write_cache_pages(mapping, wbc);
943 }
944
945 static int btree_readpage(struct file *file, struct page *page)
946 {
947         struct extent_io_tree *tree;
948         tree = &BTRFS_I(page->mapping->host)->io_tree;
949         return extent_read_full_page(tree, page, btree_get_extent, 0);
950 }
951
952 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953 {
954         if (PageWriteback(page) || PageDirty(page))
955                 return 0;
956         /*
957          * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958          * slab allocation from alloc_extent_state down the callchain where
959          * it'd hit a BUG_ON as those flags are not allowed.
960          */
961         gfp_flags &= ~GFP_SLAB_BUG_MASK;
962
963         return try_release_extent_buffer(page, gfp_flags);
964 }
965
966 static void btree_invalidatepage(struct page *page, unsigned long offset)
967 {
968         struct extent_io_tree *tree;
969         tree = &BTRFS_I(page->mapping->host)->io_tree;
970         extent_invalidatepage(tree, page, offset);
971         btree_releasepage(page, GFP_NOFS);
972         if (PagePrivate(page)) {
973                 printk(KERN_WARNING "btrfs warning page private not zero "
974                        "on page %llu\n", (unsigned long long)page_offset(page));
975                 ClearPagePrivate(page);
976                 set_page_private(page, 0);
977                 page_cache_release(page);
978         }
979 }
980
981 static int btree_set_page_dirty(struct page *page)
982 {
983         struct extent_buffer *eb;
984
985         BUG_ON(!PagePrivate(page));
986         eb = (struct extent_buffer *)page->private;
987         BUG_ON(!eb);
988         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989         BUG_ON(!atomic_read(&eb->refs));
990         btrfs_assert_tree_locked(eb);
991         return __set_page_dirty_nobuffers(page);
992 }
993
994 static const struct address_space_operations btree_aops = {
995         .readpage       = btree_readpage,
996         .writepages     = btree_writepages,
997         .releasepage    = btree_releasepage,
998         .invalidatepage = btree_invalidatepage,
999 #ifdef CONFIG_MIGRATION
1000         .migratepage    = btree_migratepage,
1001 #endif
1002         .set_page_dirty = btree_set_page_dirty,
1003 };
1004
1005 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006                          u64 parent_transid)
1007 {
1008         struct extent_buffer *buf = NULL;
1009         struct inode *btree_inode = root->fs_info->btree_inode;
1010         int ret = 0;
1011
1012         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013         if (!buf)
1014                 return 0;
1015         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1017         free_extent_buffer(buf);
1018         return ret;
1019 }
1020
1021 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022                          int mirror_num, struct extent_buffer **eb)
1023 {
1024         struct extent_buffer *buf = NULL;
1025         struct inode *btree_inode = root->fs_info->btree_inode;
1026         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027         int ret;
1028
1029         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030         if (!buf)
1031                 return 0;
1032
1033         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036                                        btree_get_extent, mirror_num);
1037         if (ret) {
1038                 free_extent_buffer(buf);
1039                 return ret;
1040         }
1041
1042         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043                 free_extent_buffer(buf);
1044                 return -EIO;
1045         } else if (extent_buffer_uptodate(buf)) {
1046                 *eb = buf;
1047         } else {
1048                 free_extent_buffer(buf);
1049         }
1050         return 0;
1051 }
1052
1053 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054                                             u64 bytenr, u32 blocksize)
1055 {
1056         struct inode *btree_inode = root->fs_info->btree_inode;
1057         struct extent_buffer *eb;
1058         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059                                 bytenr, blocksize);
1060         return eb;
1061 }
1062
1063 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064                                                  u64 bytenr, u32 blocksize)
1065 {
1066         struct inode *btree_inode = root->fs_info->btree_inode;
1067         struct extent_buffer *eb;
1068
1069         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070                                  bytenr, blocksize);
1071         return eb;
1072 }
1073
1074
1075 int btrfs_write_tree_block(struct extent_buffer *buf)
1076 {
1077         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1078                                         buf->start + buf->len - 1);
1079 }
1080
1081 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082 {
1083         return filemap_fdatawait_range(buf->pages[0]->mapping,
1084                                        buf->start, buf->start + buf->len - 1);
1085 }
1086
1087 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088                                       u32 blocksize, u64 parent_transid)
1089 {
1090         struct extent_buffer *buf = NULL;
1091         int ret;
1092
1093         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094         if (!buf)
1095                 return NULL;
1096
1097         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098         return buf;
1099
1100 }
1101
1102 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1103                       struct extent_buffer *buf)
1104 {
1105         if (btrfs_header_generation(buf) ==
1106             root->fs_info->running_transaction->transid) {
1107                 btrfs_assert_tree_locked(buf);
1108
1109                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110                         spin_lock(&root->fs_info->delalloc_lock);
1111                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1112                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1113                         else {
1114                                 spin_unlock(&root->fs_info->delalloc_lock);
1115                                 btrfs_panic(root->fs_info, -EOVERFLOW,
1116                                           "Can't clear %lu bytes from "
1117                                           " dirty_mdatadata_bytes (%llu)",
1118                                           buf->len,
1119                                           root->fs_info->dirty_metadata_bytes);
1120                         }
1121                         spin_unlock(&root->fs_info->delalloc_lock);
1122                 }
1123
1124                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125                 btrfs_set_lock_blocking(buf);
1126                 clear_extent_buffer_dirty(buf);
1127         }
1128 }
1129
1130 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131                          u32 stripesize, struct btrfs_root *root,
1132                          struct btrfs_fs_info *fs_info,
1133                          u64 objectid)
1134 {
1135         root->node = NULL;
1136         root->commit_root = NULL;
1137         root->sectorsize = sectorsize;
1138         root->nodesize = nodesize;
1139         root->leafsize = leafsize;
1140         root->stripesize = stripesize;
1141         root->ref_cows = 0;
1142         root->track_dirty = 0;
1143         root->in_radix = 0;
1144         root->orphan_item_inserted = 0;
1145         root->orphan_cleanup_state = 0;
1146
1147         root->objectid = objectid;
1148         root->last_trans = 0;
1149         root->highest_objectid = 0;
1150         root->name = NULL;
1151         root->inode_tree = RB_ROOT;
1152         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153         root->block_rsv = NULL;
1154         root->orphan_block_rsv = NULL;
1155
1156         INIT_LIST_HEAD(&root->dirty_list);
1157         INIT_LIST_HEAD(&root->root_list);
1158         spin_lock_init(&root->orphan_lock);
1159         spin_lock_init(&root->inode_lock);
1160         spin_lock_init(&root->accounting_lock);
1161         mutex_init(&root->objectid_mutex);
1162         mutex_init(&root->log_mutex);
1163         init_waitqueue_head(&root->log_writer_wait);
1164         init_waitqueue_head(&root->log_commit_wait[0]);
1165         init_waitqueue_head(&root->log_commit_wait[1]);
1166         atomic_set(&root->log_commit[0], 0);
1167         atomic_set(&root->log_commit[1], 0);
1168         atomic_set(&root->log_writers, 0);
1169         atomic_set(&root->orphan_inodes, 0);
1170         root->log_batch = 0;
1171         root->log_transid = 0;
1172         root->last_log_commit = 0;
1173         extent_io_tree_init(&root->dirty_log_pages,
1174                              fs_info->btree_inode->i_mapping);
1175
1176         memset(&root->root_key, 0, sizeof(root->root_key));
1177         memset(&root->root_item, 0, sizeof(root->root_item));
1178         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180         root->defrag_trans_start = fs_info->generation;
1181         init_completion(&root->kobj_unregister);
1182         root->defrag_running = 0;
1183         root->root_key.objectid = objectid;
1184         root->anon_dev = 0;
1185
1186         spin_lock_init(&root->root_times_lock);
1187 }
1188
1189 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1190                                             struct btrfs_fs_info *fs_info,
1191                                             u64 objectid,
1192                                             struct btrfs_root *root)
1193 {
1194         int ret;
1195         u32 blocksize;
1196         u64 generation;
1197
1198         __setup_root(tree_root->nodesize, tree_root->leafsize,
1199                      tree_root->sectorsize, tree_root->stripesize,
1200                      root, fs_info, objectid);
1201         ret = btrfs_find_last_root(tree_root, objectid,
1202                                    &root->root_item, &root->root_key);
1203         if (ret > 0)
1204                 return -ENOENT;
1205         else if (ret < 0)
1206                 return ret;
1207
1208         generation = btrfs_root_generation(&root->root_item);
1209         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1210         root->commit_root = NULL;
1211         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1212                                      blocksize, generation);
1213         if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1214                 free_extent_buffer(root->node);
1215                 root->node = NULL;
1216                 return -EIO;
1217         }
1218         root->commit_root = btrfs_root_node(root);
1219         return 0;
1220 }
1221
1222 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1223 {
1224         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1225         if (root)
1226                 root->fs_info = fs_info;
1227         return root;
1228 }
1229
1230 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1231                                      struct btrfs_fs_info *fs_info,
1232                                      u64 objectid)
1233 {
1234         struct extent_buffer *leaf;
1235         struct btrfs_root *tree_root = fs_info->tree_root;
1236         struct btrfs_root *root;
1237         struct btrfs_key key;
1238         int ret = 0;
1239         u64 bytenr;
1240
1241         root = btrfs_alloc_root(fs_info);
1242         if (!root)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         __setup_root(tree_root->nodesize, tree_root->leafsize,
1246                      tree_root->sectorsize, tree_root->stripesize,
1247                      root, fs_info, objectid);
1248         root->root_key.objectid = objectid;
1249         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1250         root->root_key.offset = 0;
1251
1252         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1253                                       0, objectid, NULL, 0, 0, 0);
1254         if (IS_ERR(leaf)) {
1255                 ret = PTR_ERR(leaf);
1256                 goto fail;
1257         }
1258
1259         bytenr = leaf->start;
1260         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1261         btrfs_set_header_bytenr(leaf, leaf->start);
1262         btrfs_set_header_generation(leaf, trans->transid);
1263         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1264         btrfs_set_header_owner(leaf, objectid);
1265         root->node = leaf;
1266
1267         write_extent_buffer(leaf, fs_info->fsid,
1268                             (unsigned long)btrfs_header_fsid(leaf),
1269                             BTRFS_FSID_SIZE);
1270         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1271                             (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1272                             BTRFS_UUID_SIZE);
1273         btrfs_mark_buffer_dirty(leaf);
1274
1275         root->commit_root = btrfs_root_node(root);
1276         root->track_dirty = 1;
1277
1278
1279         root->root_item.flags = 0;
1280         root->root_item.byte_limit = 0;
1281         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1282         btrfs_set_root_generation(&root->root_item, trans->transid);
1283         btrfs_set_root_level(&root->root_item, 0);
1284         btrfs_set_root_refs(&root->root_item, 1);
1285         btrfs_set_root_used(&root->root_item, leaf->len);
1286         btrfs_set_root_last_snapshot(&root->root_item, 0);
1287         btrfs_set_root_dirid(&root->root_item, 0);
1288         root->root_item.drop_level = 0;
1289
1290         key.objectid = objectid;
1291         key.type = BTRFS_ROOT_ITEM_KEY;
1292         key.offset = 0;
1293         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1294         if (ret)
1295                 goto fail;
1296
1297         btrfs_tree_unlock(leaf);
1298
1299 fail:
1300         if (ret)
1301                 return ERR_PTR(ret);
1302
1303         return root;
1304 }
1305
1306 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1307                                          struct btrfs_fs_info *fs_info)
1308 {
1309         struct btrfs_root *root;
1310         struct btrfs_root *tree_root = fs_info->tree_root;
1311         struct extent_buffer *leaf;
1312
1313         root = btrfs_alloc_root(fs_info);
1314         if (!root)
1315                 return ERR_PTR(-ENOMEM);
1316
1317         __setup_root(tree_root->nodesize, tree_root->leafsize,
1318                      tree_root->sectorsize, tree_root->stripesize,
1319                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1320
1321         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1322         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1323         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1324         /*
1325          * log trees do not get reference counted because they go away
1326          * before a real commit is actually done.  They do store pointers
1327          * to file data extents, and those reference counts still get
1328          * updated (along with back refs to the log tree).
1329          */
1330         root->ref_cows = 0;
1331
1332         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1333                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1334                                       0, 0, 0);
1335         if (IS_ERR(leaf)) {
1336                 kfree(root);
1337                 return ERR_CAST(leaf);
1338         }
1339
1340         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1341         btrfs_set_header_bytenr(leaf, leaf->start);
1342         btrfs_set_header_generation(leaf, trans->transid);
1343         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1344         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1345         root->node = leaf;
1346
1347         write_extent_buffer(root->node, root->fs_info->fsid,
1348                             (unsigned long)btrfs_header_fsid(root->node),
1349                             BTRFS_FSID_SIZE);
1350         btrfs_mark_buffer_dirty(root->node);
1351         btrfs_tree_unlock(root->node);
1352         return root;
1353 }
1354
1355 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1356                              struct btrfs_fs_info *fs_info)
1357 {
1358         struct btrfs_root *log_root;
1359
1360         log_root = alloc_log_tree(trans, fs_info);
1361         if (IS_ERR(log_root))
1362                 return PTR_ERR(log_root);
1363         WARN_ON(fs_info->log_root_tree);
1364         fs_info->log_root_tree = log_root;
1365         return 0;
1366 }
1367
1368 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1369                        struct btrfs_root *root)
1370 {
1371         struct btrfs_root *log_root;
1372         struct btrfs_inode_item *inode_item;
1373
1374         log_root = alloc_log_tree(trans, root->fs_info);
1375         if (IS_ERR(log_root))
1376                 return PTR_ERR(log_root);
1377
1378         log_root->last_trans = trans->transid;
1379         log_root->root_key.offset = root->root_key.objectid;
1380
1381         inode_item = &log_root->root_item.inode;
1382         inode_item->generation = cpu_to_le64(1);
1383         inode_item->size = cpu_to_le64(3);
1384         inode_item->nlink = cpu_to_le32(1);
1385         inode_item->nbytes = cpu_to_le64(root->leafsize);
1386         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1387
1388         btrfs_set_root_node(&log_root->root_item, log_root->node);
1389
1390         WARN_ON(root->log_root);
1391         root->log_root = log_root;
1392         root->log_transid = 0;
1393         root->last_log_commit = 0;
1394         return 0;
1395 }
1396
1397 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1398                                                struct btrfs_key *location)
1399 {
1400         struct btrfs_root *root;
1401         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1402         struct btrfs_path *path;
1403         struct extent_buffer *l;
1404         u64 generation;
1405         u32 blocksize;
1406         int ret = 0;
1407         int slot;
1408
1409         root = btrfs_alloc_root(fs_info);
1410         if (!root)
1411                 return ERR_PTR(-ENOMEM);
1412         if (location->offset == (u64)-1) {
1413                 ret = find_and_setup_root(tree_root, fs_info,
1414                                           location->objectid, root);
1415                 if (ret) {
1416                         kfree(root);
1417                         return ERR_PTR(ret);
1418                 }
1419                 goto out;
1420         }
1421
1422         __setup_root(tree_root->nodesize, tree_root->leafsize,
1423                      tree_root->sectorsize, tree_root->stripesize,
1424                      root, fs_info, location->objectid);
1425
1426         path = btrfs_alloc_path();
1427         if (!path) {
1428                 kfree(root);
1429                 return ERR_PTR(-ENOMEM);
1430         }
1431         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1432         if (ret == 0) {
1433                 l = path->nodes[0];
1434                 slot = path->slots[0];
1435                 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1436                 memcpy(&root->root_key, location, sizeof(*location));
1437         }
1438         btrfs_free_path(path);
1439         if (ret) {
1440                 kfree(root);
1441                 if (ret > 0)
1442                         ret = -ENOENT;
1443                 return ERR_PTR(ret);
1444         }
1445
1446         generation = btrfs_root_generation(&root->root_item);
1447         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1448         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1449                                      blocksize, generation);
1450         root->commit_root = btrfs_root_node(root);
1451         BUG_ON(!root->node); /* -ENOMEM */
1452 out:
1453         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1454                 root->ref_cows = 1;
1455                 btrfs_check_and_init_root_item(&root->root_item);
1456         }
1457
1458         return root;
1459 }
1460
1461 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1462                                               struct btrfs_key *location)
1463 {
1464         struct btrfs_root *root;
1465         int ret;
1466
1467         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1468                 return fs_info->tree_root;
1469         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1470                 return fs_info->extent_root;
1471         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1472                 return fs_info->chunk_root;
1473         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1474                 return fs_info->dev_root;
1475         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1476                 return fs_info->csum_root;
1477         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1478                 return fs_info->quota_root ? fs_info->quota_root :
1479                                              ERR_PTR(-ENOENT);
1480 again:
1481         spin_lock(&fs_info->fs_roots_radix_lock);
1482         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1483                                  (unsigned long)location->objectid);
1484         spin_unlock(&fs_info->fs_roots_radix_lock);
1485         if (root)
1486                 return root;
1487
1488         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1489         if (IS_ERR(root))
1490                 return root;
1491
1492         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1493         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1494                                         GFP_NOFS);
1495         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1496                 ret = -ENOMEM;
1497                 goto fail;
1498         }
1499
1500         btrfs_init_free_ino_ctl(root);
1501         mutex_init(&root->fs_commit_mutex);
1502         spin_lock_init(&root->cache_lock);
1503         init_waitqueue_head(&root->cache_wait);
1504
1505         ret = get_anon_bdev(&root->anon_dev);
1506         if (ret)
1507                 goto fail;
1508
1509         if (btrfs_root_refs(&root->root_item) == 0) {
1510                 ret = -ENOENT;
1511                 goto fail;
1512         }
1513
1514         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1515         if (ret < 0)
1516                 goto fail;
1517         if (ret == 0)
1518                 root->orphan_item_inserted = 1;
1519
1520         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1521         if (ret)
1522                 goto fail;
1523
1524         spin_lock(&fs_info->fs_roots_radix_lock);
1525         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1526                                 (unsigned long)root->root_key.objectid,
1527                                 root);
1528         if (ret == 0)
1529                 root->in_radix = 1;
1530
1531         spin_unlock(&fs_info->fs_roots_radix_lock);
1532         radix_tree_preload_end();
1533         if (ret) {
1534                 if (ret == -EEXIST) {
1535                         free_fs_root(root);
1536                         goto again;
1537                 }
1538                 goto fail;
1539         }
1540
1541         ret = btrfs_find_dead_roots(fs_info->tree_root,
1542                                     root->root_key.objectid);
1543         WARN_ON(ret);
1544         return root;
1545 fail:
1546         free_fs_root(root);
1547         return ERR_PTR(ret);
1548 }
1549
1550 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1551 {
1552         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1553         int ret = 0;
1554         struct btrfs_device *device;
1555         struct backing_dev_info *bdi;
1556
1557         rcu_read_lock();
1558         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1559                 if (!device->bdev)
1560                         continue;
1561                 bdi = blk_get_backing_dev_info(device->bdev);
1562                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1563                         ret = 1;
1564                         break;
1565                 }
1566         }
1567         rcu_read_unlock();
1568         return ret;
1569 }
1570
1571 /*
1572  * If this fails, caller must call bdi_destroy() to get rid of the
1573  * bdi again.
1574  */
1575 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1576 {
1577         int err;
1578
1579         bdi->capabilities = BDI_CAP_MAP_COPY;
1580         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1581         if (err)
1582                 return err;
1583
1584         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1585         bdi->congested_fn       = btrfs_congested_fn;
1586         bdi->congested_data     = info;
1587         return 0;
1588 }
1589
1590 /*
1591  * called by the kthread helper functions to finally call the bio end_io
1592  * functions.  This is where read checksum verification actually happens
1593  */
1594 static void end_workqueue_fn(struct btrfs_work *work)
1595 {
1596         struct bio *bio;
1597         struct end_io_wq *end_io_wq;
1598         struct btrfs_fs_info *fs_info;
1599         int error;
1600
1601         end_io_wq = container_of(work, struct end_io_wq, work);
1602         bio = end_io_wq->bio;
1603         fs_info = end_io_wq->info;
1604
1605         error = end_io_wq->error;
1606         bio->bi_private = end_io_wq->private;
1607         bio->bi_end_io = end_io_wq->end_io;
1608         kfree(end_io_wq);
1609         bio_endio(bio, error);
1610 }
1611
1612 static int cleaner_kthread(void *arg)
1613 {
1614         struct btrfs_root *root = arg;
1615
1616         do {
1617                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1618                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1619                         btrfs_run_delayed_iputs(root);
1620                         btrfs_clean_old_snapshots(root);
1621                         mutex_unlock(&root->fs_info->cleaner_mutex);
1622                         btrfs_run_defrag_inodes(root->fs_info);
1623                 }
1624
1625                 if (!try_to_freeze()) {
1626                         set_current_state(TASK_INTERRUPTIBLE);
1627                         if (!kthread_should_stop())
1628                                 schedule();
1629                         __set_current_state(TASK_RUNNING);
1630                 }
1631         } while (!kthread_should_stop());
1632         return 0;
1633 }
1634
1635 static int transaction_kthread(void *arg)
1636 {
1637         struct btrfs_root *root = arg;
1638         struct btrfs_trans_handle *trans;
1639         struct btrfs_transaction *cur;
1640         u64 transid;
1641         unsigned long now;
1642         unsigned long delay;
1643         bool cannot_commit;
1644
1645         do {
1646                 cannot_commit = false;
1647                 delay = HZ * 30;
1648                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1649
1650                 spin_lock(&root->fs_info->trans_lock);
1651                 cur = root->fs_info->running_transaction;
1652                 if (!cur) {
1653                         spin_unlock(&root->fs_info->trans_lock);
1654                         goto sleep;
1655                 }
1656
1657                 now = get_seconds();
1658                 if (!cur->blocked &&
1659                     (now < cur->start_time || now - cur->start_time < 30)) {
1660                         spin_unlock(&root->fs_info->trans_lock);
1661                         delay = HZ * 5;
1662                         goto sleep;
1663                 }
1664                 transid = cur->transid;
1665                 spin_unlock(&root->fs_info->trans_lock);
1666
1667                 /* If the file system is aborted, this will always fail. */
1668                 trans = btrfs_join_transaction(root);
1669                 if (IS_ERR(trans)) {
1670                         cannot_commit = true;
1671                         goto sleep;
1672                 }
1673                 if (transid == trans->transid) {
1674                         btrfs_commit_transaction(trans, root);
1675                 } else {
1676                         btrfs_end_transaction(trans, root);
1677                 }
1678 sleep:
1679                 wake_up_process(root->fs_info->cleaner_kthread);
1680                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1681
1682                 if (!try_to_freeze()) {
1683                         set_current_state(TASK_INTERRUPTIBLE);
1684                         if (!kthread_should_stop() &&
1685                             (!btrfs_transaction_blocked(root->fs_info) ||
1686                              cannot_commit))
1687                                 schedule_timeout(delay);
1688                         __set_current_state(TASK_RUNNING);
1689                 }
1690         } while (!kthread_should_stop());
1691         return 0;
1692 }
1693
1694 /*
1695  * this will find the highest generation in the array of
1696  * root backups.  The index of the highest array is returned,
1697  * or -1 if we can't find anything.
1698  *
1699  * We check to make sure the array is valid by comparing the
1700  * generation of the latest  root in the array with the generation
1701  * in the super block.  If they don't match we pitch it.
1702  */
1703 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1704 {
1705         u64 cur;
1706         int newest_index = -1;
1707         struct btrfs_root_backup *root_backup;
1708         int i;
1709
1710         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1711                 root_backup = info->super_copy->super_roots + i;
1712                 cur = btrfs_backup_tree_root_gen(root_backup);
1713                 if (cur == newest_gen)
1714                         newest_index = i;
1715         }
1716
1717         /* check to see if we actually wrapped around */
1718         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1719                 root_backup = info->super_copy->super_roots;
1720                 cur = btrfs_backup_tree_root_gen(root_backup);
1721                 if (cur == newest_gen)
1722                         newest_index = 0;
1723         }
1724         return newest_index;
1725 }
1726
1727
1728 /*
1729  * find the oldest backup so we know where to store new entries
1730  * in the backup array.  This will set the backup_root_index
1731  * field in the fs_info struct
1732  */
1733 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1734                                      u64 newest_gen)
1735 {
1736         int newest_index = -1;
1737
1738         newest_index = find_newest_super_backup(info, newest_gen);
1739         /* if there was garbage in there, just move along */
1740         if (newest_index == -1) {
1741                 info->backup_root_index = 0;
1742         } else {
1743                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1744         }
1745 }
1746
1747 /*
1748  * copy all the root pointers into the super backup array.
1749  * this will bump the backup pointer by one when it is
1750  * done
1751  */
1752 static void backup_super_roots(struct btrfs_fs_info *info)
1753 {
1754         int next_backup;
1755         struct btrfs_root_backup *root_backup;
1756         int last_backup;
1757
1758         next_backup = info->backup_root_index;
1759         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1760                 BTRFS_NUM_BACKUP_ROOTS;
1761
1762         /*
1763          * just overwrite the last backup if we're at the same generation
1764          * this happens only at umount
1765          */
1766         root_backup = info->super_for_commit->super_roots + last_backup;
1767         if (btrfs_backup_tree_root_gen(root_backup) ==
1768             btrfs_header_generation(info->tree_root->node))
1769                 next_backup = last_backup;
1770
1771         root_backup = info->super_for_commit->super_roots + next_backup;
1772
1773         /*
1774          * make sure all of our padding and empty slots get zero filled
1775          * regardless of which ones we use today
1776          */
1777         memset(root_backup, 0, sizeof(*root_backup));
1778
1779         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1780
1781         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1782         btrfs_set_backup_tree_root_gen(root_backup,
1783                                btrfs_header_generation(info->tree_root->node));
1784
1785         btrfs_set_backup_tree_root_level(root_backup,
1786                                btrfs_header_level(info->tree_root->node));
1787
1788         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1789         btrfs_set_backup_chunk_root_gen(root_backup,
1790                                btrfs_header_generation(info->chunk_root->node));
1791         btrfs_set_backup_chunk_root_level(root_backup,
1792                                btrfs_header_level(info->chunk_root->node));
1793
1794         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1795         btrfs_set_backup_extent_root_gen(root_backup,
1796                                btrfs_header_generation(info->extent_root->node));
1797         btrfs_set_backup_extent_root_level(root_backup,
1798                                btrfs_header_level(info->extent_root->node));
1799
1800         /*
1801          * we might commit during log recovery, which happens before we set
1802          * the fs_root.  Make sure it is valid before we fill it in.
1803          */
1804         if (info->fs_root && info->fs_root->node) {
1805                 btrfs_set_backup_fs_root(root_backup,
1806                                          info->fs_root->node->start);
1807                 btrfs_set_backup_fs_root_gen(root_backup,
1808                                btrfs_header_generation(info->fs_root->node));
1809                 btrfs_set_backup_fs_root_level(root_backup,
1810                                btrfs_header_level(info->fs_root->node));
1811         }
1812
1813         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1814         btrfs_set_backup_dev_root_gen(root_backup,
1815                                btrfs_header_generation(info->dev_root->node));
1816         btrfs_set_backup_dev_root_level(root_backup,
1817                                        btrfs_header_level(info->dev_root->node));
1818
1819         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1820         btrfs_set_backup_csum_root_gen(root_backup,
1821                                btrfs_header_generation(info->csum_root->node));
1822         btrfs_set_backup_csum_root_level(root_backup,
1823                                btrfs_header_level(info->csum_root->node));
1824
1825         btrfs_set_backup_total_bytes(root_backup,
1826                              btrfs_super_total_bytes(info->super_copy));
1827         btrfs_set_backup_bytes_used(root_backup,
1828                              btrfs_super_bytes_used(info->super_copy));
1829         btrfs_set_backup_num_devices(root_backup,
1830                              btrfs_super_num_devices(info->super_copy));
1831
1832         /*
1833          * if we don't copy this out to the super_copy, it won't get remembered
1834          * for the next commit
1835          */
1836         memcpy(&info->super_copy->super_roots,
1837                &info->super_for_commit->super_roots,
1838                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1839 }
1840
1841 /*
1842  * this copies info out of the root backup array and back into
1843  * the in-memory super block.  It is meant to help iterate through
1844  * the array, so you send it the number of backups you've already
1845  * tried and the last backup index you used.
1846  *
1847  * this returns -1 when it has tried all the backups
1848  */
1849 static noinline int next_root_backup(struct btrfs_fs_info *info,
1850                                      struct btrfs_super_block *super,
1851                                      int *num_backups_tried, int *backup_index)
1852 {
1853         struct btrfs_root_backup *root_backup;
1854         int newest = *backup_index;
1855
1856         if (*num_backups_tried == 0) {
1857                 u64 gen = btrfs_super_generation(super);
1858
1859                 newest = find_newest_super_backup(info, gen);
1860                 if (newest == -1)
1861                         return -1;
1862
1863                 *backup_index = newest;
1864                 *num_backups_tried = 1;
1865         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1866                 /* we've tried all the backups, all done */
1867                 return -1;
1868         } else {
1869                 /* jump to the next oldest backup */
1870                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1871                         BTRFS_NUM_BACKUP_ROOTS;
1872                 *backup_index = newest;
1873                 *num_backups_tried += 1;
1874         }
1875         root_backup = super->super_roots + newest;
1876
1877         btrfs_set_super_generation(super,
1878                                    btrfs_backup_tree_root_gen(root_backup));
1879         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1880         btrfs_set_super_root_level(super,
1881                                    btrfs_backup_tree_root_level(root_backup));
1882         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1883
1884         /*
1885          * fixme: the total bytes and num_devices need to match or we should
1886          * need a fsck
1887          */
1888         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1889         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1890         return 0;
1891 }
1892
1893 /* helper to cleanup tree roots */
1894 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1895 {
1896         free_extent_buffer(info->tree_root->node);
1897         free_extent_buffer(info->tree_root->commit_root);
1898         free_extent_buffer(info->dev_root->node);
1899         free_extent_buffer(info->dev_root->commit_root);
1900         free_extent_buffer(info->extent_root->node);
1901         free_extent_buffer(info->extent_root->commit_root);
1902         free_extent_buffer(info->csum_root->node);
1903         free_extent_buffer(info->csum_root->commit_root);
1904         if (info->quota_root) {
1905                 free_extent_buffer(info->quota_root->node);
1906                 free_extent_buffer(info->quota_root->commit_root);
1907         }
1908
1909         info->tree_root->node = NULL;
1910         info->tree_root->commit_root = NULL;
1911         info->dev_root->node = NULL;
1912         info->dev_root->commit_root = NULL;
1913         info->extent_root->node = NULL;
1914         info->extent_root->commit_root = NULL;
1915         info->csum_root->node = NULL;
1916         info->csum_root->commit_root = NULL;
1917         if (info->quota_root) {
1918                 info->quota_root->node = NULL;
1919                 info->quota_root->commit_root = NULL;
1920         }
1921
1922         if (chunk_root) {
1923                 free_extent_buffer(info->chunk_root->node);
1924                 free_extent_buffer(info->chunk_root->commit_root);
1925                 info->chunk_root->node = NULL;
1926                 info->chunk_root->commit_root = NULL;
1927         }
1928 }
1929
1930
1931 int open_ctree(struct super_block *sb,
1932                struct btrfs_fs_devices *fs_devices,
1933                char *options)
1934 {
1935         u32 sectorsize;
1936         u32 nodesize;
1937         u32 leafsize;
1938         u32 blocksize;
1939         u32 stripesize;
1940         u64 generation;
1941         u64 features;
1942         struct btrfs_key location;
1943         struct buffer_head *bh;
1944         struct btrfs_super_block *disk_super;
1945         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1946         struct btrfs_root *tree_root;
1947         struct btrfs_root *extent_root;
1948         struct btrfs_root *csum_root;
1949         struct btrfs_root *chunk_root;
1950         struct btrfs_root *dev_root;
1951         struct btrfs_root *quota_root;
1952         struct btrfs_root *log_tree_root;
1953         int ret;
1954         int err = -EINVAL;
1955         int num_backups_tried = 0;
1956         int backup_index = 0;
1957
1958         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1959         extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1960         csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1961         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1962         dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1963         quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1964
1965         if (!tree_root || !extent_root || !csum_root ||
1966             !chunk_root || !dev_root || !quota_root) {
1967                 err = -ENOMEM;
1968                 goto fail;
1969         }
1970
1971         ret = init_srcu_struct(&fs_info->subvol_srcu);
1972         if (ret) {
1973                 err = ret;
1974                 goto fail;
1975         }
1976
1977         ret = setup_bdi(fs_info, &fs_info->bdi);
1978         if (ret) {
1979                 err = ret;
1980                 goto fail_srcu;
1981         }
1982
1983         fs_info->btree_inode = new_inode(sb);
1984         if (!fs_info->btree_inode) {
1985                 err = -ENOMEM;
1986                 goto fail_bdi;
1987         }
1988
1989         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1990
1991         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1992         INIT_LIST_HEAD(&fs_info->trans_list);
1993         INIT_LIST_HEAD(&fs_info->dead_roots);
1994         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1995         INIT_LIST_HEAD(&fs_info->hashers);
1996         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1997         INIT_LIST_HEAD(&fs_info->ordered_operations);
1998         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1999         spin_lock_init(&fs_info->delalloc_lock);
2000         spin_lock_init(&fs_info->trans_lock);
2001         spin_lock_init(&fs_info->ref_cache_lock);
2002         spin_lock_init(&fs_info->fs_roots_radix_lock);
2003         spin_lock_init(&fs_info->delayed_iput_lock);
2004         spin_lock_init(&fs_info->defrag_inodes_lock);
2005         spin_lock_init(&fs_info->free_chunk_lock);
2006         spin_lock_init(&fs_info->tree_mod_seq_lock);
2007         rwlock_init(&fs_info->tree_mod_log_lock);
2008         mutex_init(&fs_info->reloc_mutex);
2009
2010         init_completion(&fs_info->kobj_unregister);
2011         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2012         INIT_LIST_HEAD(&fs_info->space_info);
2013         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2014         btrfs_mapping_init(&fs_info->mapping_tree);
2015         btrfs_init_block_rsv(&fs_info->global_block_rsv);
2016         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
2017         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
2018         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
2019         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
2020         btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
2021         atomic_set(&fs_info->nr_async_submits, 0);
2022         atomic_set(&fs_info->async_delalloc_pages, 0);
2023         atomic_set(&fs_info->async_submit_draining, 0);
2024         atomic_set(&fs_info->nr_async_bios, 0);
2025         atomic_set(&fs_info->defrag_running, 0);
2026         atomic_set(&fs_info->tree_mod_seq, 0);
2027         fs_info->sb = sb;
2028         fs_info->max_inline = 8192 * 1024;
2029         fs_info->metadata_ratio = 0;
2030         fs_info->defrag_inodes = RB_ROOT;
2031         fs_info->trans_no_join = 0;
2032         fs_info->free_chunk_space = 0;
2033         fs_info->tree_mod_log = RB_ROOT;
2034
2035         init_waitqueue_head(&fs_info->tree_mod_seq_wait);
2036
2037         /* readahead state */
2038         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2039         spin_lock_init(&fs_info->reada_lock);
2040
2041         fs_info->thread_pool_size = min_t(unsigned long,
2042                                           num_online_cpus() + 2, 8);
2043
2044         INIT_LIST_HEAD(&fs_info->ordered_extents);
2045         spin_lock_init(&fs_info->ordered_extent_lock);
2046         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2047                                         GFP_NOFS);
2048         if (!fs_info->delayed_root) {
2049                 err = -ENOMEM;
2050                 goto fail_iput;
2051         }
2052         btrfs_init_delayed_root(fs_info->delayed_root);
2053
2054         mutex_init(&fs_info->scrub_lock);
2055         atomic_set(&fs_info->scrubs_running, 0);
2056         atomic_set(&fs_info->scrub_pause_req, 0);
2057         atomic_set(&fs_info->scrubs_paused, 0);
2058         atomic_set(&fs_info->scrub_cancel_req, 0);
2059         init_waitqueue_head(&fs_info->scrub_pause_wait);
2060         init_rwsem(&fs_info->scrub_super_lock);
2061         fs_info->scrub_workers_refcnt = 0;
2062 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2063         fs_info->check_integrity_print_mask = 0;
2064 #endif
2065
2066         spin_lock_init(&fs_info->balance_lock);
2067         mutex_init(&fs_info->balance_mutex);
2068         atomic_set(&fs_info->balance_running, 0);
2069         atomic_set(&fs_info->balance_pause_req, 0);
2070         atomic_set(&fs_info->balance_cancel_req, 0);
2071         fs_info->balance_ctl = NULL;
2072         init_waitqueue_head(&fs_info->balance_wait_q);
2073
2074         sb->s_blocksize = 4096;
2075         sb->s_blocksize_bits = blksize_bits(4096);
2076         sb->s_bdi = &fs_info->bdi;
2077
2078         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2079         set_nlink(fs_info->btree_inode, 1);
2080         /*
2081          * we set the i_size on the btree inode to the max possible int.
2082          * the real end of the address space is determined by all of
2083          * the devices in the system
2084          */
2085         fs_info->btree_inode->i_size = OFFSET_MAX;
2086         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2087         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2088
2089         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2090         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2091                              fs_info->btree_inode->i_mapping);
2092         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2093         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2094
2095         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2096
2097         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2098         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2099                sizeof(struct btrfs_key));
2100         set_bit(BTRFS_INODE_DUMMY,
2101                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2102         insert_inode_hash(fs_info->btree_inode);
2103
2104         spin_lock_init(&fs_info->block_group_cache_lock);
2105         fs_info->block_group_cache_tree = RB_ROOT;
2106
2107         extent_io_tree_init(&fs_info->freed_extents[0],
2108                              fs_info->btree_inode->i_mapping);
2109         extent_io_tree_init(&fs_info->freed_extents[1],
2110                              fs_info->btree_inode->i_mapping);
2111         fs_info->pinned_extents = &fs_info->freed_extents[0];
2112         fs_info->do_barriers = 1;
2113
2114
2115         mutex_init(&fs_info->ordered_operations_mutex);
2116         mutex_init(&fs_info->tree_log_mutex);
2117         mutex_init(&fs_info->chunk_mutex);
2118         mutex_init(&fs_info->transaction_kthread_mutex);
2119         mutex_init(&fs_info->cleaner_mutex);
2120         mutex_init(&fs_info->volume_mutex);
2121         init_rwsem(&fs_info->extent_commit_sem);
2122         init_rwsem(&fs_info->cleanup_work_sem);
2123         init_rwsem(&fs_info->subvol_sem);
2124
2125         spin_lock_init(&fs_info->qgroup_lock);
2126         fs_info->qgroup_tree = RB_ROOT;
2127         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2128         fs_info->qgroup_seq = 1;
2129         fs_info->quota_enabled = 0;
2130         fs_info->pending_quota_state = 0;
2131
2132         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2133         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2134
2135         init_waitqueue_head(&fs_info->transaction_throttle);
2136         init_waitqueue_head(&fs_info->transaction_wait);
2137         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2138         init_waitqueue_head(&fs_info->async_submit_wait);
2139
2140         __setup_root(4096, 4096, 4096, 4096, tree_root,
2141                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2142
2143         invalidate_bdev(fs_devices->latest_bdev);
2144         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2145         if (!bh) {
2146                 err = -EINVAL;
2147                 goto fail_alloc;
2148         }
2149
2150         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2151         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2152                sizeof(*fs_info->super_for_commit));
2153         brelse(bh);
2154
2155         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2156
2157         disk_super = fs_info->super_copy;
2158         if (!btrfs_super_root(disk_super))
2159                 goto fail_alloc;
2160
2161         /* check FS state, whether FS is broken. */
2162         fs_info->fs_state |= btrfs_super_flags(disk_super);
2163
2164         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2165         if (ret) {
2166                 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2167                 err = ret;
2168                 goto fail_alloc;
2169         }
2170
2171         /*
2172          * run through our array of backup supers and setup
2173          * our ring pointer to the oldest one
2174          */
2175         generation = btrfs_super_generation(disk_super);
2176         find_oldest_super_backup(fs_info, generation);
2177
2178         /*
2179          * In the long term, we'll store the compression type in the super
2180          * block, and it'll be used for per file compression control.
2181          */
2182         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2183
2184         ret = btrfs_parse_options(tree_root, options);
2185         if (ret) {
2186                 err = ret;
2187                 goto fail_alloc;
2188         }
2189
2190         features = btrfs_super_incompat_flags(disk_super) &
2191                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2192         if (features) {
2193                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2194                        "unsupported optional features (%Lx).\n",
2195                        (unsigned long long)features);
2196                 err = -EINVAL;
2197                 goto fail_alloc;
2198         }
2199
2200         if (btrfs_super_leafsize(disk_super) !=
2201             btrfs_super_nodesize(disk_super)) {
2202                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2203                        "blocksizes don't match.  node %d leaf %d\n",
2204                        btrfs_super_nodesize(disk_super),
2205                        btrfs_super_leafsize(disk_super));
2206                 err = -EINVAL;
2207                 goto fail_alloc;
2208         }
2209         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2210                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2211                        "blocksize (%d) was too large\n",
2212                        btrfs_super_leafsize(disk_super));
2213                 err = -EINVAL;
2214                 goto fail_alloc;
2215         }
2216
2217         features = btrfs_super_incompat_flags(disk_super);
2218         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2219         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2220                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2221
2222         /*
2223          * flag our filesystem as having big metadata blocks if
2224          * they are bigger than the page size
2225          */
2226         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2227                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2228                         printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2229                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2230         }
2231
2232         nodesize = btrfs_super_nodesize(disk_super);
2233         leafsize = btrfs_super_leafsize(disk_super);
2234         sectorsize = btrfs_super_sectorsize(disk_super);
2235         stripesize = btrfs_super_stripesize(disk_super);
2236
2237         /*
2238          * mixed block groups end up with duplicate but slightly offset
2239          * extent buffers for the same range.  It leads to corruptions
2240          */
2241         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2242             (sectorsize != leafsize)) {
2243                 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2244                                 "are not allowed for mixed block groups on %s\n",
2245                                 sb->s_id);
2246                 goto fail_alloc;
2247         }
2248
2249         btrfs_set_super_incompat_flags(disk_super, features);
2250
2251         features = btrfs_super_compat_ro_flags(disk_super) &
2252                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2253         if (!(sb->s_flags & MS_RDONLY) && features) {
2254                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2255                        "unsupported option features (%Lx).\n",
2256                        (unsigned long long)features);
2257                 err = -EINVAL;
2258                 goto fail_alloc;
2259         }
2260
2261         btrfs_init_workers(&fs_info->generic_worker,
2262                            "genwork", 1, NULL);
2263
2264         btrfs_init_workers(&fs_info->workers, "worker",
2265                            fs_info->thread_pool_size,
2266                            &fs_info->generic_worker);
2267
2268         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2269                            fs_info->thread_pool_size,
2270                            &fs_info->generic_worker);
2271
2272         btrfs_init_workers(&fs_info->submit_workers, "submit",
2273                            min_t(u64, fs_devices->num_devices,
2274                            fs_info->thread_pool_size),
2275                            &fs_info->generic_worker);
2276
2277         btrfs_init_workers(&fs_info->caching_workers, "cache",
2278                            2, &fs_info->generic_worker);
2279
2280         /* a higher idle thresh on the submit workers makes it much more
2281          * likely that bios will be send down in a sane order to the
2282          * devices
2283          */
2284         fs_info->submit_workers.idle_thresh = 64;
2285
2286         fs_info->workers.idle_thresh = 16;
2287         fs_info->workers.ordered = 1;
2288
2289         fs_info->delalloc_workers.idle_thresh = 2;
2290         fs_info->delalloc_workers.ordered = 1;
2291
2292         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2293                            &fs_info->generic_worker);
2294         btrfs_init_workers(&fs_info->endio_workers, "endio",
2295                            fs_info->thread_pool_size,
2296                            &fs_info->generic_worker);
2297         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2298                            fs_info->thread_pool_size,
2299                            &fs_info->generic_worker);
2300         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2301                            "endio-meta-write", fs_info->thread_pool_size,
2302                            &fs_info->generic_worker);
2303         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2304                            fs_info->thread_pool_size,
2305                            &fs_info->generic_worker);
2306         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2307                            1, &fs_info->generic_worker);
2308         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2309                            fs_info->thread_pool_size,
2310                            &fs_info->generic_worker);
2311         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2312                            fs_info->thread_pool_size,
2313                            &fs_info->generic_worker);
2314
2315         /*
2316          * endios are largely parallel and should have a very
2317          * low idle thresh
2318          */
2319         fs_info->endio_workers.idle_thresh = 4;
2320         fs_info->endio_meta_workers.idle_thresh = 4;
2321
2322         fs_info->endio_write_workers.idle_thresh = 2;
2323         fs_info->endio_meta_write_workers.idle_thresh = 2;
2324         fs_info->readahead_workers.idle_thresh = 2;
2325
2326         /*
2327          * btrfs_start_workers can really only fail because of ENOMEM so just
2328          * return -ENOMEM if any of these fail.
2329          */
2330         ret = btrfs_start_workers(&fs_info->workers);
2331         ret |= btrfs_start_workers(&fs_info->generic_worker);
2332         ret |= btrfs_start_workers(&fs_info->submit_workers);
2333         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2334         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2335         ret |= btrfs_start_workers(&fs_info->endio_workers);
2336         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2337         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2338         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2339         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2340         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2341         ret |= btrfs_start_workers(&fs_info->caching_workers);
2342         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2343         if (ret) {
2344                 err = -ENOMEM;
2345                 goto fail_sb_buffer;
2346         }
2347
2348         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2349         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2350                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2351
2352         tree_root->nodesize = nodesize;
2353         tree_root->leafsize = leafsize;
2354         tree_root->sectorsize = sectorsize;
2355         tree_root->stripesize = stripesize;
2356
2357         sb->s_blocksize = sectorsize;
2358         sb->s_blocksize_bits = blksize_bits(sectorsize);
2359
2360         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2361                     sizeof(disk_super->magic))) {
2362                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2363                 goto fail_sb_buffer;
2364         }
2365
2366         if (sectorsize != PAGE_SIZE) {
2367                 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2368                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2369                 goto fail_sb_buffer;
2370         }
2371
2372         mutex_lock(&fs_info->chunk_mutex);
2373         ret = btrfs_read_sys_array(tree_root);
2374         mutex_unlock(&fs_info->chunk_mutex);
2375         if (ret) {
2376                 printk(KERN_WARNING "btrfs: failed to read the system "
2377                        "array on %s\n", sb->s_id);
2378                 goto fail_sb_buffer;
2379         }
2380
2381         blocksize = btrfs_level_size(tree_root,
2382                                      btrfs_super_chunk_root_level(disk_super));
2383         generation = btrfs_super_chunk_root_generation(disk_super);
2384
2385         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2386                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2387
2388         chunk_root->node = read_tree_block(chunk_root,
2389                                            btrfs_super_chunk_root(disk_super),
2390                                            blocksize, generation);
2391         BUG_ON(!chunk_root->node); /* -ENOMEM */
2392         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2393                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2394                        sb->s_id);
2395                 goto fail_tree_roots;
2396         }
2397         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2398         chunk_root->commit_root = btrfs_root_node(chunk_root);
2399
2400         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2401            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2402            BTRFS_UUID_SIZE);
2403
2404         ret = btrfs_read_chunk_tree(chunk_root);
2405         if (ret) {
2406                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2407                        sb->s_id);
2408                 goto fail_tree_roots;
2409         }
2410
2411         btrfs_close_extra_devices(fs_devices);
2412
2413         if (!fs_devices->latest_bdev) {
2414                 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2415                        sb->s_id);
2416                 goto fail_tree_roots;
2417         }
2418
2419 retry_root_backup:
2420         blocksize = btrfs_level_size(tree_root,
2421                                      btrfs_super_root_level(disk_super));
2422         generation = btrfs_super_generation(disk_super);
2423
2424         tree_root->node = read_tree_block(tree_root,
2425                                           btrfs_super_root(disk_super),
2426                                           blocksize, generation);
2427         if (!tree_root->node ||
2428             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2429                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2430                        sb->s_id);
2431
2432                 goto recovery_tree_root;
2433         }
2434
2435         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2436         tree_root->commit_root = btrfs_root_node(tree_root);
2437
2438         ret = find_and_setup_root(tree_root, fs_info,
2439                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2440         if (ret)
2441                 goto recovery_tree_root;
2442         extent_root->track_dirty = 1;
2443
2444         ret = find_and_setup_root(tree_root, fs_info,
2445                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2446         if (ret)
2447                 goto recovery_tree_root;
2448         dev_root->track_dirty = 1;
2449
2450         ret = find_and_setup_root(tree_root, fs_info,
2451                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2452         if (ret)
2453                 goto recovery_tree_root;
2454         csum_root->track_dirty = 1;
2455
2456         ret = find_and_setup_root(tree_root, fs_info,
2457                                   BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2458         if (ret) {
2459                 kfree(quota_root);
2460                 quota_root = fs_info->quota_root = NULL;
2461         } else {
2462                 quota_root->track_dirty = 1;
2463                 fs_info->quota_enabled = 1;
2464                 fs_info->pending_quota_state = 1;
2465         }
2466
2467         fs_info->generation = generation;
2468         fs_info->last_trans_committed = generation;
2469
2470         ret = btrfs_recover_balance(fs_info);
2471         if (ret) {
2472                 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2473                 goto fail_block_groups;
2474         }
2475
2476         ret = btrfs_init_dev_stats(fs_info);
2477         if (ret) {
2478                 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2479                        ret);
2480                 goto fail_block_groups;
2481         }
2482
2483         ret = btrfs_init_space_info(fs_info);
2484         if (ret) {
2485                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2486                 goto fail_block_groups;
2487         }
2488
2489         ret = btrfs_read_block_groups(extent_root);
2490         if (ret) {
2491                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2492                 goto fail_block_groups;
2493         }
2494
2495         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2496                                                "btrfs-cleaner");
2497         if (IS_ERR(fs_info->cleaner_kthread))
2498                 goto fail_block_groups;
2499
2500         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2501                                                    tree_root,
2502                                                    "btrfs-transaction");
2503         if (IS_ERR(fs_info->transaction_kthread))
2504                 goto fail_cleaner;
2505
2506         if (!btrfs_test_opt(tree_root, SSD) &&
2507             !btrfs_test_opt(tree_root, NOSSD) &&
2508             !fs_info->fs_devices->rotating) {
2509                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2510                        "mode\n");
2511                 btrfs_set_opt(fs_info->mount_opt, SSD);
2512         }
2513
2514 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2515         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2516                 ret = btrfsic_mount(tree_root, fs_devices,
2517                                     btrfs_test_opt(tree_root,
2518                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2519                                     1 : 0,
2520                                     fs_info->check_integrity_print_mask);
2521                 if (ret)
2522                         printk(KERN_WARNING "btrfs: failed to initialize"
2523                                " integrity check module %s\n", sb->s_id);
2524         }
2525 #endif
2526         ret = btrfs_read_qgroup_config(fs_info);
2527         if (ret)
2528                 goto fail_trans_kthread;
2529
2530         /* do not make disk changes in broken FS */
2531         if (btrfs_super_log_root(disk_super) != 0 &&
2532             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2533                 u64 bytenr = btrfs_super_log_root(disk_super);
2534
2535                 if (fs_devices->rw_devices == 0) {
2536                         printk(KERN_WARNING "Btrfs log replay required "
2537                                "on RO media\n");
2538                         err = -EIO;
2539                         goto fail_qgroup;
2540                 }
2541                 blocksize =
2542                      btrfs_level_size(tree_root,
2543                                       btrfs_super_log_root_level(disk_super));
2544
2545                 log_tree_root = btrfs_alloc_root(fs_info);
2546                 if (!log_tree_root) {
2547                         err = -ENOMEM;
2548                         goto fail_qgroup;
2549                 }
2550
2551                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2552                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2553
2554                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2555                                                       blocksize,
2556                                                       generation + 1);
2557                 /* returns with log_tree_root freed on success */
2558                 ret = btrfs_recover_log_trees(log_tree_root);
2559                 if (ret) {
2560                         btrfs_error(tree_root->fs_info, ret,
2561                                     "Failed to recover log tree");
2562                         free_extent_buffer(log_tree_root->node);
2563                         kfree(log_tree_root);
2564                         goto fail_trans_kthread;
2565                 }
2566
2567                 if (sb->s_flags & MS_RDONLY) {
2568                         ret = btrfs_commit_super(tree_root);
2569                         if (ret)
2570                                 goto fail_trans_kthread;
2571                 }
2572         }
2573
2574         ret = btrfs_find_orphan_roots(tree_root);
2575         if (ret)
2576                 goto fail_trans_kthread;
2577
2578         if (!(sb->s_flags & MS_RDONLY)) {
2579                 ret = btrfs_cleanup_fs_roots(fs_info);
2580                 if (ret)
2581                         goto fail_trans_kthread;
2582
2583                 ret = btrfs_recover_relocation(tree_root);
2584                 if (ret < 0) {
2585                         printk(KERN_WARNING
2586                                "btrfs: failed to recover relocation\n");
2587                         err = -EINVAL;
2588                         goto fail_qgroup;
2589                 }
2590         }
2591
2592         location.objectid = BTRFS_FS_TREE_OBJECTID;
2593         location.type = BTRFS_ROOT_ITEM_KEY;
2594         location.offset = (u64)-1;
2595
2596         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2597         if (!fs_info->fs_root)
2598                 goto fail_qgroup;
2599         if (IS_ERR(fs_info->fs_root)) {
2600                 err = PTR_ERR(fs_info->fs_root);
2601                 goto fail_qgroup;
2602         }
2603
2604         if (sb->s_flags & MS_RDONLY)
2605                 return 0;
2606
2607         down_read(&fs_info->cleanup_work_sem);
2608         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2609             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2610                 up_read(&fs_info->cleanup_work_sem);
2611                 close_ctree(tree_root);
2612                 return ret;
2613         }
2614         up_read(&fs_info->cleanup_work_sem);
2615
2616         ret = btrfs_resume_balance_async(fs_info);
2617         if (ret) {
2618                 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2619                 close_ctree(tree_root);
2620                 return ret;
2621         }
2622
2623         return 0;
2624
2625 fail_qgroup:
2626         btrfs_free_qgroup_config(fs_info);
2627 fail_trans_kthread:
2628         kthread_stop(fs_info->transaction_kthread);
2629 fail_cleaner:
2630         kthread_stop(fs_info->cleaner_kthread);
2631
2632         /*
2633          * make sure we're done with the btree inode before we stop our
2634          * kthreads
2635          */
2636         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2637         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2638
2639 fail_block_groups:
2640         btrfs_free_block_groups(fs_info);
2641
2642 fail_tree_roots:
2643         free_root_pointers(fs_info, 1);
2644
2645 fail_sb_buffer:
2646         btrfs_stop_workers(&fs_info->generic_worker);
2647         btrfs_stop_workers(&fs_info->readahead_workers);
2648         btrfs_stop_workers(&fs_info->fixup_workers);
2649         btrfs_stop_workers(&fs_info->delalloc_workers);
2650         btrfs_stop_workers(&fs_info->workers);
2651         btrfs_stop_workers(&fs_info->endio_workers);
2652         btrfs_stop_workers(&fs_info->endio_meta_workers);
2653         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2654         btrfs_stop_workers(&fs_info->endio_write_workers);
2655         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2656         btrfs_stop_workers(&fs_info->submit_workers);
2657         btrfs_stop_workers(&fs_info->delayed_workers);
2658         btrfs_stop_workers(&fs_info->caching_workers);
2659 fail_alloc:
2660 fail_iput:
2661         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2662
2663         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2664         iput(fs_info->btree_inode);
2665 fail_bdi:
2666         bdi_destroy(&fs_info->bdi);
2667 fail_srcu:
2668         cleanup_srcu_struct(&fs_info->subvol_srcu);
2669 fail:
2670         btrfs_close_devices(fs_info->fs_devices);
2671         return err;
2672
2673 recovery_tree_root:
2674         if (!btrfs_test_opt(tree_root, RECOVERY))
2675                 goto fail_tree_roots;
2676
2677         free_root_pointers(fs_info, 0);
2678
2679         /* don't use the log in recovery mode, it won't be valid */
2680         btrfs_set_super_log_root(disk_super, 0);
2681
2682         /* we can't trust the free space cache either */
2683         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2684
2685         ret = next_root_backup(fs_info, fs_info->super_copy,
2686                                &num_backups_tried, &backup_index);
2687         if (ret == -1)
2688                 goto fail_block_groups;
2689         goto retry_root_backup;
2690 }
2691
2692 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2693 {
2694         if (uptodate) {
2695                 set_buffer_uptodate(bh);
2696         } else {
2697                 struct btrfs_device *device = (struct btrfs_device *)
2698                         bh->b_private;
2699
2700                 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2701                                           "I/O error on %s\n",
2702                                           rcu_str_deref(device->name));
2703                 /* note, we dont' set_buffer_write_io_error because we have
2704                  * our own ways of dealing with the IO errors
2705                  */
2706                 clear_buffer_uptodate(bh);
2707                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2708         }
2709         unlock_buffer(bh);
2710         put_bh(bh);
2711 }
2712
2713 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2714 {
2715         struct buffer_head *bh;
2716         struct buffer_head *latest = NULL;
2717         struct btrfs_super_block *super;
2718         int i;
2719         u64 transid = 0;
2720         u64 bytenr;
2721
2722         /* we would like to check all the supers, but that would make
2723          * a btrfs mount succeed after a mkfs from a different FS.
2724          * So, we need to add a special mount option to scan for
2725          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2726          */
2727         for (i = 0; i < 1; i++) {
2728                 bytenr = btrfs_sb_offset(i);
2729                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2730                         break;
2731                 bh = __bread(bdev, bytenr / 4096, 4096);
2732                 if (!bh)
2733                         continue;
2734
2735                 super = (struct btrfs_super_block *)bh->b_data;
2736                 if (btrfs_super_bytenr(super) != bytenr ||
2737                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2738                             sizeof(super->magic))) {
2739                         brelse(bh);
2740                         continue;
2741                 }
2742
2743                 if (!latest || btrfs_super_generation(super) > transid) {
2744                         brelse(latest);
2745                         latest = bh;
2746                         transid = btrfs_super_generation(super);
2747                 } else {
2748                         brelse(bh);
2749                 }
2750         }
2751         return latest;
2752 }
2753
2754 /*
2755  * this should be called twice, once with wait == 0 and
2756  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2757  * we write are pinned.
2758  *
2759  * They are released when wait == 1 is done.
2760  * max_mirrors must be the same for both runs, and it indicates how
2761  * many supers on this one device should be written.
2762  *
2763  * max_mirrors == 0 means to write them all.
2764  */
2765 static int write_dev_supers(struct btrfs_device *device,
2766                             struct btrfs_super_block *sb,
2767                             int do_barriers, int wait, int max_mirrors)
2768 {
2769         struct buffer_head *bh;
2770         int i;
2771         int ret;
2772         int errors = 0;
2773         u32 crc;
2774         u64 bytenr;
2775
2776         if (max_mirrors == 0)
2777                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2778
2779         for (i = 0; i < max_mirrors; i++) {
2780                 bytenr = btrfs_sb_offset(i);
2781                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2782                         break;
2783
2784                 if (wait) {
2785                         bh = __find_get_block(device->bdev, bytenr / 4096,
2786                                               BTRFS_SUPER_INFO_SIZE);
2787                         BUG_ON(!bh);
2788                         wait_on_buffer(bh);
2789                         if (!buffer_uptodate(bh))
2790                                 errors++;
2791
2792                         /* drop our reference */
2793                         brelse(bh);
2794
2795                         /* drop the reference from the wait == 0 run */
2796                         brelse(bh);
2797                         continue;
2798                 } else {
2799                         btrfs_set_super_bytenr(sb, bytenr);
2800
2801                         crc = ~(u32)0;
2802                         crc = btrfs_csum_data(NULL, (char *)sb +
2803                                               BTRFS_CSUM_SIZE, crc,
2804                                               BTRFS_SUPER_INFO_SIZE -
2805                                               BTRFS_CSUM_SIZE);
2806                         btrfs_csum_final(crc, sb->csum);
2807
2808                         /*
2809                          * one reference for us, and we leave it for the
2810                          * caller
2811                          */
2812                         bh = __getblk(device->bdev, bytenr / 4096,
2813                                       BTRFS_SUPER_INFO_SIZE);
2814                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2815
2816                         /* one reference for submit_bh */
2817                         get_bh(bh);
2818
2819                         set_buffer_uptodate(bh);
2820                         lock_buffer(bh);
2821                         bh->b_end_io = btrfs_end_buffer_write_sync;
2822                         bh->b_private = device;
2823                 }
2824
2825                 /*
2826                  * we fua the first super.  The others we allow
2827                  * to go down lazy.
2828                  */
2829                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2830                 if (ret)
2831                         errors++;
2832         }
2833         return errors < i ? 0 : -1;
2834 }
2835
2836 /*
2837  * endio for the write_dev_flush, this will wake anyone waiting
2838  * for the barrier when it is done
2839  */
2840 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2841 {
2842         if (err) {
2843                 if (err == -EOPNOTSUPP)
2844                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2845                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2846         }
2847         if (bio->bi_private)
2848                 complete(bio->bi_private);
2849         bio_put(bio);
2850 }
2851
2852 /*
2853  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
2854  * sent down.  With wait == 1, it waits for the previous flush.
2855  *
2856  * any device where the flush fails with eopnotsupp are flagged as not-barrier
2857  * capable
2858  */
2859 static int write_dev_flush(struct btrfs_device *device, int wait)
2860 {
2861         struct bio *bio;
2862         int ret = 0;
2863
2864         if (device->nobarriers)
2865                 return 0;
2866
2867         if (wait) {
2868                 bio = device->flush_bio;
2869                 if (!bio)
2870                         return 0;
2871
2872                 wait_for_completion(&device->flush_wait);
2873
2874                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2875                         printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2876                                       rcu_str_deref(device->name));
2877                         device->nobarriers = 1;
2878                 }
2879                 if (!bio_flagged(bio, BIO_UPTODATE)) {
2880                         ret = -EIO;
2881                         if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2882                                 btrfs_dev_stat_inc_and_print(device,
2883                                         BTRFS_DEV_STAT_FLUSH_ERRS);
2884                 }
2885
2886                 /* drop the reference from the wait == 0 run */
2887                 bio_put(bio);
2888                 device->flush_bio = NULL;
2889
2890                 return ret;
2891         }
2892
2893         /*
2894          * one reference for us, and we leave it for the
2895          * caller
2896          */
2897         device->flush_bio = NULL;
2898         bio = bio_alloc(GFP_NOFS, 0);
2899         if (!bio)
2900                 return -ENOMEM;
2901
2902         bio->bi_end_io = btrfs_end_empty_barrier;
2903         bio->bi_bdev = device->bdev;
2904         init_completion(&device->flush_wait);
2905         bio->bi_private = &device->flush_wait;
2906         device->flush_bio = bio;
2907
2908         bio_get(bio);
2909         btrfsic_submit_bio(WRITE_FLUSH, bio);
2910
2911         return 0;
2912 }
2913
2914 /*
2915  * send an empty flush down to each device in parallel,
2916  * then wait for them
2917  */
2918 static int barrier_all_devices(struct btrfs_fs_info *info)
2919 {
2920         struct list_head *head;
2921         struct btrfs_device *dev;
2922         int errors = 0;
2923         int ret;
2924
2925         /* send down all the barriers */
2926         head = &info->fs_devices->devices;
2927         list_for_each_entry_rcu(dev, head, dev_list) {
2928                 if (!dev->bdev) {
2929                         errors++;
2930                         continue;
2931                 }
2932                 if (!dev->in_fs_metadata || !dev->writeable)
2933                         continue;
2934
2935                 ret = write_dev_flush(dev, 0);
2936                 if (ret)
2937                         errors++;
2938         }
2939
2940         /* wait for all the barriers */
2941         list_for_each_entry_rcu(dev, head, dev_list) {
2942                 if (!dev->bdev) {
2943                         errors++;
2944                         continue;
2945                 }
2946                 if (!dev->in_fs_metadata || !dev->writeable)
2947                         continue;
2948
2949                 ret = write_dev_flush(dev, 1);
2950                 if (ret)
2951                         errors++;
2952         }
2953         if (errors)
2954                 return -EIO;
2955         return 0;
2956 }
2957
2958 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2959 {
2960         struct list_head *head;
2961         struct btrfs_device *dev;
2962         struct btrfs_super_block *sb;
2963         struct btrfs_dev_item *dev_item;
2964         int ret;
2965         int do_barriers;
2966         int max_errors;
2967         int total_errors = 0;
2968         u64 flags;
2969
2970         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2971         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2972         backup_super_roots(root->fs_info);
2973
2974         sb = root->fs_info->super_for_commit;
2975         dev_item = &sb->dev_item;
2976
2977         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2978         head = &root->fs_info->fs_devices->devices;
2979
2980         if (do_barriers)
2981                 barrier_all_devices(root->fs_info);
2982
2983         list_for_each_entry_rcu(dev, head, dev_list) {
2984                 if (!dev->bdev) {
2985                         total_errors++;
2986                         continue;
2987                 }
2988                 if (!dev->in_fs_metadata || !dev->writeable)
2989                         continue;
2990
2991                 btrfs_set_stack_device_generation(dev_item, 0);
2992                 btrfs_set_stack_device_type(dev_item, dev->type);
2993                 btrfs_set_stack_device_id(dev_item, dev->devid);
2994                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2995                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2996                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2997                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2998                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2999                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3000                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3001
3002                 flags = btrfs_super_flags(sb);
3003                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3004
3005                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3006                 if (ret)
3007                         total_errors++;
3008         }
3009         if (total_errors > max_errors) {
3010                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3011                        total_errors);
3012
3013                 /* This shouldn't happen. FUA is masked off if unsupported */
3014                 BUG();
3015         }
3016
3017         total_errors = 0;
3018         list_for_each_entry_rcu(dev, head, dev_list) {
3019                 if (!dev->bdev)
3020                         continue;
3021                 if (!dev->in_fs_metadata || !dev->writeable)
3022                         continue;
3023
3024                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3025                 if (ret)
3026                         total_errors++;
3027         }
3028         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3029         if (total_errors > max_errors) {
3030                 btrfs_error(root->fs_info, -EIO,
3031                             "%d errors while writing supers", total_errors);
3032                 return -EIO;
3033         }
3034         return 0;
3035 }
3036
3037 int write_ctree_super(struct btrfs_trans_handle *trans,
3038                       struct btrfs_root *root, int max_mirrors)
3039 {
3040         int ret;
3041
3042         ret = write_all_supers(root, max_mirrors);
3043         return ret;
3044 }
3045
3046 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3047 {
3048         spin_lock(&fs_info->fs_roots_radix_lock);
3049         radix_tree_delete(&fs_info->fs_roots_radix,
3050                           (unsigned long)root->root_key.objectid);
3051         spin_unlock(&fs_info->fs_roots_radix_lock);
3052
3053         if (btrfs_root_refs(&root->root_item) == 0)
3054                 synchronize_srcu(&fs_info->subvol_srcu);
3055
3056         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3057         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3058         free_fs_root(root);
3059 }
3060
3061 static void free_fs_root(struct btrfs_root *root)
3062 {
3063         iput(root->cache_inode);
3064         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3065         if (root->anon_dev)
3066                 free_anon_bdev(root->anon_dev);
3067         free_extent_buffer(root->node);
3068         free_extent_buffer(root->commit_root);
3069         kfree(root->free_ino_ctl);
3070         kfree(root->free_ino_pinned);
3071         kfree(root->name);
3072         kfree(root);
3073 }
3074
3075 static void del_fs_roots(struct btrfs_fs_info *fs_info)
3076 {
3077         int ret;
3078         struct btrfs_root *gang[8];
3079         int i;
3080
3081         while (!list_empty(&fs_info->dead_roots)) {
3082                 gang[0] = list_entry(fs_info->dead_roots.next,
3083                                      struct btrfs_root, root_list);
3084                 list_del(&gang[0]->root_list);
3085
3086                 if (gang[0]->in_radix) {
3087                         btrfs_free_fs_root(fs_info, gang[0]);
3088                 } else {
3089                         free_extent_buffer(gang[0]->node);
3090                         free_extent_buffer(gang[0]->commit_root);
3091                         kfree(gang[0]);
3092                 }
3093         }
3094
3095         while (1) {
3096                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3097                                              (void **)gang, 0,
3098                                              ARRAY_SIZE(gang));
3099                 if (!ret)
3100                         break;
3101                 for (i = 0; i < ret; i++)
3102                         btrfs_free_fs_root(fs_info, gang[i]);
3103         }
3104 }
3105
3106 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3107 {
3108         u64 root_objectid = 0;
3109         struct btrfs_root *gang[8];
3110         int i;
3111         int ret;
3112
3113         while (1) {
3114                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3115                                              (void **)gang, root_objectid,
3116                                              ARRAY_SIZE(gang));
3117                 if (!ret)
3118                         break;
3119
3120                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3121                 for (i = 0; i < ret; i++) {
3122                         int err;
3123
3124                         root_objectid = gang[i]->root_key.objectid;
3125                         err = btrfs_orphan_cleanup(gang[i]);
3126                         if (err)
3127                                 return err;
3128                 }
3129                 root_objectid++;
3130         }
3131         return 0;
3132 }
3133
3134 int btrfs_commit_super(struct btrfs_root *root)
3135 {
3136         struct btrfs_trans_handle *trans;
3137         int ret;
3138
3139         mutex_lock(&root->fs_info->cleaner_mutex);
3140         btrfs_run_delayed_iputs(root);
3141         btrfs_clean_old_snapshots(root);
3142         mutex_unlock(&root->fs_info->cleaner_mutex);
3143
3144         /* wait until ongoing cleanup work done */
3145         down_write(&root->fs_info->cleanup_work_sem);
3146         up_write(&root->fs_info->cleanup_work_sem);
3147
3148         trans = btrfs_join_transaction(root);
3149         if (IS_ERR(trans))
3150                 return PTR_ERR(trans);
3151         ret = btrfs_commit_transaction(trans, root);
3152         if (ret)
3153                 return ret;
3154         /* run commit again to drop the original snapshot */
3155         trans = btrfs_join_transaction(root);
3156         if (IS_ERR(trans))
3157                 return PTR_ERR(trans);
3158         ret = btrfs_commit_transaction(trans, root);
3159         if (ret)
3160                 return ret;
3161         ret = btrfs_write_and_wait_transaction(NULL, root);
3162         if (ret) {
3163                 btrfs_error(root->fs_info, ret,
3164                             "Failed to sync btree inode to disk.");
3165                 return ret;
3166         }
3167
3168         ret = write_ctree_super(NULL, root, 0);
3169         return ret;
3170 }
3171
3172 int close_ctree(struct btrfs_root *root)
3173 {
3174         struct btrfs_fs_info *fs_info = root->fs_info;
3175         int ret;
3176
3177         fs_info->closing = 1;
3178         smp_mb();
3179
3180         /* pause restriper - we want to resume on mount */
3181         btrfs_pause_balance(root->fs_info);
3182
3183         btrfs_scrub_cancel(root);
3184
3185         /* wait for any defraggers to finish */
3186         wait_event(fs_info->transaction_wait,
3187                    (atomic_read(&fs_info->defrag_running) == 0));
3188
3189         /* clear out the rbtree of defraggable inodes */
3190         btrfs_run_defrag_inodes(fs_info);
3191
3192         /*
3193          * Here come 2 situations when btrfs is broken to flip readonly:
3194          *
3195          * 1. when btrfs flips readonly somewhere else before
3196          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3197          * and btrfs will skip to write sb directly to keep
3198          * ERROR state on disk.
3199          *
3200          * 2. when btrfs flips readonly just in btrfs_commit_super,
3201          * and in such case, btrfs cannot write sb via btrfs_commit_super,
3202          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3203          * btrfs will cleanup all FS resources first and write sb then.
3204          */
3205         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3206                 ret = btrfs_commit_super(root);
3207                 if (ret)
3208                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3209         }
3210
3211         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3212                 ret = btrfs_error_commit_super(root);
3213                 if (ret)
3214                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3215         }
3216
3217         btrfs_put_block_group_cache(fs_info);
3218
3219         kthread_stop(fs_info->transaction_kthread);
3220         kthread_stop(fs_info->cleaner_kthread);
3221
3222         fs_info->closing = 2;
3223         smp_mb();
3224
3225         btrfs_free_qgroup_config(root->fs_info);
3226
3227         if (fs_info->delalloc_bytes) {
3228                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3229                        (unsigned long long)fs_info->delalloc_bytes);
3230         }
3231         if (fs_info->total_ref_cache_size) {
3232                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3233                        (unsigned long long)fs_info->total_ref_cache_size);
3234         }
3235
3236         free_extent_buffer(fs_info->extent_root->node);
3237         free_extent_buffer(fs_info->extent_root->commit_root);
3238         free_extent_buffer(fs_info->tree_root->node);
3239         free_extent_buffer(fs_info->tree_root->commit_root);
3240         free_extent_buffer(fs_info->chunk_root->node);
3241         free_extent_buffer(fs_info->chunk_root->commit_root);
3242         free_extent_buffer(fs_info->dev_root->node);
3243         free_extent_buffer(fs_info->dev_root->commit_root);
3244         free_extent_buffer(fs_info->csum_root->node);
3245         free_extent_buffer(fs_info->csum_root->commit_root);
3246         if (fs_info->quota_root) {
3247                 free_extent_buffer(fs_info->quota_root->node);
3248                 free_extent_buffer(fs_info->quota_root->commit_root);
3249         }
3250
3251         btrfs_free_block_groups(fs_info);
3252
3253         del_fs_roots(fs_info);
3254
3255         iput(fs_info->btree_inode);
3256
3257         btrfs_stop_workers(&fs_info->generic_worker);
3258         btrfs_stop_workers(&fs_info->fixup_workers);
3259         btrfs_stop_workers(&fs_info->delalloc_workers);
3260         btrfs_stop_workers(&fs_info->workers);
3261         btrfs_stop_workers(&fs_info->endio_workers);
3262         btrfs_stop_workers(&fs_info->endio_meta_workers);