btrfs: sink get_extent parameter to read_extent_buffer_pages
[sfrench/cifs-2.6.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/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "hash.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
49 #include "raid56.h"
50 #include "sysfs.h"
51 #include "qgroup.h"
52 #include "compression.h"
53 #include "tree-checker.h"
54 #include "ref-verify.h"
55
56 #ifdef CONFIG_X86
57 #include <asm/cpufeature.h>
58 #endif
59
60 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
61                                  BTRFS_HEADER_FLAG_RELOC |\
62                                  BTRFS_SUPER_FLAG_ERROR |\
63                                  BTRFS_SUPER_FLAG_SEEDING |\
64                                  BTRFS_SUPER_FLAG_METADUMP)
65
66 static const struct extent_io_ops btree_extent_io_ops;
67 static void end_workqueue_fn(struct btrfs_work *work);
68 static void free_fs_root(struct btrfs_root *root);
69 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
70 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
71 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
72                                       struct btrfs_fs_info *fs_info);
73 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
74 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
75                                         struct extent_io_tree *dirty_pages,
76                                         int mark);
77 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
78                                        struct extent_io_tree *pinned_extents);
79 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
80 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
81
82 /*
83  * btrfs_end_io_wq structs are used to do processing in task context when an IO
84  * is complete.  This is used during reads to verify checksums, and it is used
85  * by writes to insert metadata for new file extents after IO is complete.
86  */
87 struct btrfs_end_io_wq {
88         struct bio *bio;
89         bio_end_io_t *end_io;
90         void *private;
91         struct btrfs_fs_info *info;
92         blk_status_t status;
93         enum btrfs_wq_endio_type metadata;
94         struct btrfs_work work;
95 };
96
97 static struct kmem_cache *btrfs_end_io_wq_cache;
98
99 int __init btrfs_end_io_wq_init(void)
100 {
101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102                                         sizeof(struct btrfs_end_io_wq),
103                                         0,
104                                         SLAB_MEM_SPREAD,
105                                         NULL);
106         if (!btrfs_end_io_wq_cache)
107                 return -ENOMEM;
108         return 0;
109 }
110
111 void btrfs_end_io_wq_exit(void)
112 {
113         kmem_cache_destroy(btrfs_end_io_wq_cache);
114 }
115
116 /*
117  * async submit bios are used to offload expensive checksumming
118  * onto the worker threads.  They checksum file and metadata bios
119  * just before they are sent down the IO stack.
120  */
121 struct async_submit_bio {
122         void *private_data;
123         struct btrfs_fs_info *fs_info;
124         struct bio *bio;
125         extent_submit_bio_hook_t *submit_bio_start;
126         extent_submit_bio_hook_t *submit_bio_done;
127         int mirror_num;
128         unsigned long bio_flags;
129         /*
130          * bio_offset is optional, can be used if the pages in the bio
131          * can't tell us where in the file the bio should go
132          */
133         u64 bio_offset;
134         struct btrfs_work work;
135         blk_status_t status;
136 };
137
138 /*
139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
141  * the level the eb occupies in the tree.
142  *
143  * Different roots are used for different purposes and may nest inside each
144  * other and they require separate keysets.  As lockdep keys should be
145  * static, assign keysets according to the purpose of the root as indicated
146  * by btrfs_root->objectid.  This ensures that all special purpose roots
147  * have separate keysets.
148  *
149  * Lock-nesting across peer nodes is always done with the immediate parent
150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
151  * subclass to avoid triggering lockdep warning in such cases.
152  *
153  * The key is set by the readpage_end_io_hook after the buffer has passed
154  * csum validation but before the pages are unlocked.  It is also set by
155  * btrfs_init_new_buffer on freshly allocated blocks.
156  *
157  * We also add a check to make sure the highest level of the tree is the
158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
159  * needs update as well.
160  */
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
163 #  error
164 # endif
165
166 static struct btrfs_lockdep_keyset {
167         u64                     id;             /* root objectid */
168         const char              *name_stem;     /* lock name stem */
169         char                    names[BTRFS_MAX_LEVEL + 1][20];
170         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
171 } btrfs_lockdep_keysets[] = {
172         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
173         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
174         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
175         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
176         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
177         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
178         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
179         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
180         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
181         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
182         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
183         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
184         { .id = 0,                              .name_stem = "tree"     },
185 };
186
187 void __init btrfs_init_lockdep(void)
188 {
189         int i, j;
190
191         /* initialize lockdep class names */
192         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
194
195                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196                         snprintf(ks->names[j], sizeof(ks->names[j]),
197                                  "btrfs-%s-%02d", ks->name_stem, j);
198         }
199 }
200
201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
202                                     int level)
203 {
204         struct btrfs_lockdep_keyset *ks;
205
206         BUG_ON(level >= ARRAY_SIZE(ks->keys));
207
208         /* find the matching keyset, id 0 is the default entry */
209         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210                 if (ks->id == objectid)
211                         break;
212
213         lockdep_set_class_and_name(&eb->lock,
214                                    &ks->keys[level], ks->names[level]);
215 }
216
217 #endif
218
219 /*
220  * extents on the btree inode are pretty simple, there's one extent
221  * that covers the entire device
222  */
223 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
224                 struct page *page, size_t pg_offset, u64 start, u64 len,
225                 int create)
226 {
227         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
228         struct extent_map_tree *em_tree = &inode->extent_tree;
229         struct extent_map *em;
230         int ret;
231
232         read_lock(&em_tree->lock);
233         em = lookup_extent_mapping(em_tree, start, len);
234         if (em) {
235                 em->bdev = fs_info->fs_devices->latest_bdev;
236                 read_unlock(&em_tree->lock);
237                 goto out;
238         }
239         read_unlock(&em_tree->lock);
240
241         em = alloc_extent_map();
242         if (!em) {
243                 em = ERR_PTR(-ENOMEM);
244                 goto out;
245         }
246         em->start = 0;
247         em->len = (u64)-1;
248         em->block_len = (u64)-1;
249         em->block_start = 0;
250         em->bdev = fs_info->fs_devices->latest_bdev;
251
252         write_lock(&em_tree->lock);
253         ret = add_extent_mapping(em_tree, em, 0);
254         if (ret == -EEXIST) {
255                 free_extent_map(em);
256                 em = lookup_extent_mapping(em_tree, start, len);
257                 if (!em)
258                         em = ERR_PTR(-EIO);
259         } else if (ret) {
260                 free_extent_map(em);
261                 em = ERR_PTR(ret);
262         }
263         write_unlock(&em_tree->lock);
264
265 out:
266         return em;
267 }
268
269 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
270 {
271         return btrfs_crc32c(seed, data, len);
272 }
273
274 void btrfs_csum_final(u32 crc, u8 *result)
275 {
276         put_unaligned_le32(~crc, result);
277 }
278
279 /*
280  * compute the csum for a btree block, and either verify it or write it
281  * into the csum field of the block.
282  */
283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
284                            struct extent_buffer *buf,
285                            int verify)
286 {
287         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
288         char *result = NULL;
289         unsigned long len;
290         unsigned long cur_len;
291         unsigned long offset = BTRFS_CSUM_SIZE;
292         char *kaddr;
293         unsigned long map_start;
294         unsigned long map_len;
295         int err;
296         u32 crc = ~(u32)0;
297         unsigned long inline_result;
298
299         len = buf->len - offset;
300         while (len > 0) {
301                 err = map_private_extent_buffer(buf, offset, 32,
302                                         &kaddr, &map_start, &map_len);
303                 if (err)
304                         return err;
305                 cur_len = min(len, map_len - (offset - map_start));
306                 crc = btrfs_csum_data(kaddr + offset - map_start,
307                                       crc, cur_len);
308                 len -= cur_len;
309                 offset += cur_len;
310         }
311         if (csum_size > sizeof(inline_result)) {
312                 result = kzalloc(csum_size, GFP_NOFS);
313                 if (!result)
314                         return -ENOMEM;
315         } else {
316                 result = (char *)&inline_result;
317         }
318
319         btrfs_csum_final(crc, result);
320
321         if (verify) {
322                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
323                         u32 val;
324                         u32 found = 0;
325                         memcpy(&found, result, csum_size);
326
327                         read_extent_buffer(buf, &val, 0, csum_size);
328                         btrfs_warn_rl(fs_info,
329                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
330                                 fs_info->sb->s_id, buf->start,
331                                 val, found, btrfs_header_level(buf));
332                         if (result != (char *)&inline_result)
333                                 kfree(result);
334                         return -EUCLEAN;
335                 }
336         } else {
337                 write_extent_buffer(buf, result, 0, csum_size);
338         }
339         if (result != (char *)&inline_result)
340                 kfree(result);
341         return 0;
342 }
343
344 /*
345  * we can't consider a given block up to date unless the transid of the
346  * block matches the transid in the parent node's pointer.  This is how we
347  * detect blocks that either didn't get written at all or got written
348  * in the wrong place.
349  */
350 static int verify_parent_transid(struct extent_io_tree *io_tree,
351                                  struct extent_buffer *eb, u64 parent_transid,
352                                  int atomic)
353 {
354         struct extent_state *cached_state = NULL;
355         int ret;
356         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
357
358         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
359                 return 0;
360
361         if (atomic)
362                 return -EAGAIN;
363
364         if (need_lock) {
365                 btrfs_tree_read_lock(eb);
366                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
367         }
368
369         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
370                          &cached_state);
371         if (extent_buffer_uptodate(eb) &&
372             btrfs_header_generation(eb) == parent_transid) {
373                 ret = 0;
374                 goto out;
375         }
376         btrfs_err_rl(eb->fs_info,
377                 "parent transid verify failed on %llu wanted %llu found %llu",
378                         eb->start,
379                         parent_transid, btrfs_header_generation(eb));
380         ret = 1;
381
382         /*
383          * Things reading via commit roots that don't have normal protection,
384          * like send, can have a really old block in cache that may point at a
385          * block that has been freed and re-allocated.  So don't clear uptodate
386          * if we find an eb that is under IO (dirty/writeback) because we could
387          * end up reading in the stale data and then writing it back out and
388          * making everybody very sad.
389          */
390         if (!extent_buffer_under_io(eb))
391                 clear_extent_buffer_uptodate(eb);
392 out:
393         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
394                              &cached_state, GFP_NOFS);
395         if (need_lock)
396                 btrfs_tree_read_unlock_blocking(eb);
397         return ret;
398 }
399
400 /*
401  * Return 0 if the superblock checksum type matches the checksum value of that
402  * algorithm. Pass the raw disk superblock data.
403  */
404 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
405                                   char *raw_disk_sb)
406 {
407         struct btrfs_super_block *disk_sb =
408                 (struct btrfs_super_block *)raw_disk_sb;
409         u16 csum_type = btrfs_super_csum_type(disk_sb);
410         int ret = 0;
411
412         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
413                 u32 crc = ~(u32)0;
414                 const int csum_size = sizeof(crc);
415                 char result[csum_size];
416
417                 /*
418                  * The super_block structure does not span the whole
419                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420                  * is filled with zeros and is included in the checksum.
421                  */
422                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424                 btrfs_csum_final(crc, result);
425
426                 if (memcmp(raw_disk_sb, result, csum_size))
427                         ret = 1;
428         }
429
430         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
432                                 csum_type);
433                 ret = 1;
434         }
435
436         return ret;
437 }
438
439 /*
440  * helper to read a given tree block, doing retries as required when
441  * the checksums don't match and we have alternate mirrors to try.
442  */
443 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
444                                           struct extent_buffer *eb,
445                                           u64 parent_transid)
446 {
447         struct extent_io_tree *io_tree;
448         int failed = 0;
449         int ret;
450         int num_copies = 0;
451         int mirror_num = 0;
452         int failed_mirror = 0;
453
454         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
456         while (1) {
457                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
458                                                mirror_num);
459                 if (!ret) {
460                         if (!verify_parent_transid(io_tree, eb,
461                                                    parent_transid, 0))
462                                 break;
463                         else
464                                 ret = -EIO;
465                 }
466
467                 /*
468                  * This buffer's crc is fine, but its contents are corrupted, so
469                  * there is no reason to read the other copies, they won't be
470                  * any less wrong.
471                  */
472                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
473                         break;
474
475                 num_copies = btrfs_num_copies(fs_info,
476                                               eb->start, eb->len);
477                 if (num_copies == 1)
478                         break;
479
480                 if (!failed_mirror) {
481                         failed = 1;
482                         failed_mirror = eb->read_mirror;
483                 }
484
485                 mirror_num++;
486                 if (mirror_num == failed_mirror)
487                         mirror_num++;
488
489                 if (mirror_num > num_copies)
490                         break;
491         }
492
493         if (failed && !ret && failed_mirror)
494                 repair_eb_io_failure(fs_info, eb, failed_mirror);
495
496         return ret;
497 }
498
499 /*
500  * checksum a dirty tree block before IO.  This has extra checks to make sure
501  * we only fill in the checksum field in the first page of a multi-page block
502  */
503
504 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
505 {
506         u64 start = page_offset(page);
507         u64 found_start;
508         struct extent_buffer *eb;
509
510         eb = (struct extent_buffer *)page->private;
511         if (page != eb->pages[0])
512                 return 0;
513
514         found_start = btrfs_header_bytenr(eb);
515         /*
516          * Please do not consolidate these warnings into a single if.
517          * It is useful to know what went wrong.
518          */
519         if (WARN_ON(found_start != start))
520                 return -EUCLEAN;
521         if (WARN_ON(!PageUptodate(page)))
522                 return -EUCLEAN;
523
524         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
525                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
526
527         return csum_tree_block(fs_info, eb, 0);
528 }
529
530 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
531                                  struct extent_buffer *eb)
532 {
533         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
534         u8 fsid[BTRFS_FSID_SIZE];
535         int ret = 1;
536
537         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
538         while (fs_devices) {
539                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
540                         ret = 0;
541                         break;
542                 }
543                 fs_devices = fs_devices->seed;
544         }
545         return ret;
546 }
547
548 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
549                                       u64 phy_offset, struct page *page,
550                                       u64 start, u64 end, int mirror)
551 {
552         u64 found_start;
553         int found_level;
554         struct extent_buffer *eb;
555         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
556         struct btrfs_fs_info *fs_info = root->fs_info;
557         int ret = 0;
558         int reads_done;
559
560         if (!page->private)
561                 goto out;
562
563         eb = (struct extent_buffer *)page->private;
564
565         /* the pending IO might have been the only thing that kept this buffer
566          * in memory.  Make sure we have a ref for all this other checks
567          */
568         extent_buffer_get(eb);
569
570         reads_done = atomic_dec_and_test(&eb->io_pages);
571         if (!reads_done)
572                 goto err;
573
574         eb->read_mirror = mirror;
575         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
576                 ret = -EIO;
577                 goto err;
578         }
579
580         found_start = btrfs_header_bytenr(eb);
581         if (found_start != eb->start) {
582                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
583                              found_start, eb->start);
584                 ret = -EIO;
585                 goto err;
586         }
587         if (check_tree_block_fsid(fs_info, eb)) {
588                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
589                              eb->start);
590                 ret = -EIO;
591                 goto err;
592         }
593         found_level = btrfs_header_level(eb);
594         if (found_level >= BTRFS_MAX_LEVEL) {
595                 btrfs_err(fs_info, "bad tree block level %d",
596                           (int)btrfs_header_level(eb));
597                 ret = -EIO;
598                 goto err;
599         }
600
601         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
602                                        eb, found_level);
603
604         ret = csum_tree_block(fs_info, eb, 1);
605         if (ret)
606                 goto err;
607
608         /*
609          * If this is a leaf block and it is corrupt, set the corrupt bit so
610          * that we don't try and read the other copies of this block, just
611          * return -EIO.
612          */
613         if (found_level == 0 && btrfs_check_leaf_full(root, eb)) {
614                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
615                 ret = -EIO;
616         }
617
618         if (found_level > 0 && btrfs_check_node(root, eb))
619                 ret = -EIO;
620
621         if (!ret)
622                 set_extent_buffer_uptodate(eb);
623 err:
624         if (reads_done &&
625             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
626                 btree_readahead_hook(eb, ret);
627
628         if (ret) {
629                 /*
630                  * our io error hook is going to dec the io pages
631                  * again, we have to make sure it has something
632                  * to decrement
633                  */
634                 atomic_inc(&eb->io_pages);
635                 clear_extent_buffer_uptodate(eb);
636         }
637         free_extent_buffer(eb);
638 out:
639         return ret;
640 }
641
642 static int btree_io_failed_hook(struct page *page, int failed_mirror)
643 {
644         struct extent_buffer *eb;
645
646         eb = (struct extent_buffer *)page->private;
647         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
648         eb->read_mirror = failed_mirror;
649         atomic_dec(&eb->io_pages);
650         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
651                 btree_readahead_hook(eb, -EIO);
652         return -EIO;    /* we fixed nothing */
653 }
654
655 static void end_workqueue_bio(struct bio *bio)
656 {
657         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
658         struct btrfs_fs_info *fs_info;
659         struct btrfs_workqueue *wq;
660         btrfs_work_func_t func;
661
662         fs_info = end_io_wq->info;
663         end_io_wq->status = bio->bi_status;
664
665         if (bio_op(bio) == REQ_OP_WRITE) {
666                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
667                         wq = fs_info->endio_meta_write_workers;
668                         func = btrfs_endio_meta_write_helper;
669                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
670                         wq = fs_info->endio_freespace_worker;
671                         func = btrfs_freespace_write_helper;
672                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
673                         wq = fs_info->endio_raid56_workers;
674                         func = btrfs_endio_raid56_helper;
675                 } else {
676                         wq = fs_info->endio_write_workers;
677                         func = btrfs_endio_write_helper;
678                 }
679         } else {
680                 if (unlikely(end_io_wq->metadata ==
681                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
682                         wq = fs_info->endio_repair_workers;
683                         func = btrfs_endio_repair_helper;
684                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
685                         wq = fs_info->endio_raid56_workers;
686                         func = btrfs_endio_raid56_helper;
687                 } else if (end_io_wq->metadata) {
688                         wq = fs_info->endio_meta_workers;
689                         func = btrfs_endio_meta_helper;
690                 } else {
691                         wq = fs_info->endio_workers;
692                         func = btrfs_endio_helper;
693                 }
694         }
695
696         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
697         btrfs_queue_work(wq, &end_io_wq->work);
698 }
699
700 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
701                         enum btrfs_wq_endio_type metadata)
702 {
703         struct btrfs_end_io_wq *end_io_wq;
704
705         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
706         if (!end_io_wq)
707                 return BLK_STS_RESOURCE;
708
709         end_io_wq->private = bio->bi_private;
710         end_io_wq->end_io = bio->bi_end_io;
711         end_io_wq->info = info;
712         end_io_wq->status = 0;
713         end_io_wq->bio = bio;
714         end_io_wq->metadata = metadata;
715
716         bio->bi_private = end_io_wq;
717         bio->bi_end_io = end_workqueue_bio;
718         return 0;
719 }
720
721 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
722 {
723         unsigned long limit = min_t(unsigned long,
724                                     info->thread_pool_size,
725                                     info->fs_devices->open_devices);
726         return 256 * limit;
727 }
728
729 static void run_one_async_start(struct btrfs_work *work)
730 {
731         struct async_submit_bio *async;
732         blk_status_t ret;
733
734         async = container_of(work, struct  async_submit_bio, work);
735         ret = async->submit_bio_start(async->private_data, async->bio,
736                                       async->mirror_num, async->bio_flags,
737                                       async->bio_offset);
738         if (ret)
739                 async->status = ret;
740 }
741
742 static void run_one_async_done(struct btrfs_work *work)
743 {
744         struct async_submit_bio *async;
745
746         async = container_of(work, struct  async_submit_bio, work);
747
748         /* If an error occurred we just want to clean up the bio and move on */
749         if (async->status) {
750                 async->bio->bi_status = async->status;
751                 bio_endio(async->bio);
752                 return;
753         }
754
755         async->submit_bio_done(async->private_data, async->bio, async->mirror_num,
756                                async->bio_flags, async->bio_offset);
757 }
758
759 static void run_one_async_free(struct btrfs_work *work)
760 {
761         struct async_submit_bio *async;
762
763         async = container_of(work, struct  async_submit_bio, work);
764         kfree(async);
765 }
766
767 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
768                                  int mirror_num, unsigned long bio_flags,
769                                  u64 bio_offset, void *private_data,
770                                  extent_submit_bio_hook_t *submit_bio_start,
771                                  extent_submit_bio_hook_t *submit_bio_done)
772 {
773         struct async_submit_bio *async;
774
775         async = kmalloc(sizeof(*async), GFP_NOFS);
776         if (!async)
777                 return BLK_STS_RESOURCE;
778
779         async->private_data = private_data;
780         async->fs_info = fs_info;
781         async->bio = bio;
782         async->mirror_num = mirror_num;
783         async->submit_bio_start = submit_bio_start;
784         async->submit_bio_done = submit_bio_done;
785
786         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
787                         run_one_async_done, run_one_async_free);
788
789         async->bio_flags = bio_flags;
790         async->bio_offset = bio_offset;
791
792         async->status = 0;
793
794         if (op_is_sync(bio->bi_opf))
795                 btrfs_set_work_high_priority(&async->work);
796
797         btrfs_queue_work(fs_info->workers, &async->work);
798         return 0;
799 }
800
801 static blk_status_t btree_csum_one_bio(struct bio *bio)
802 {
803         struct bio_vec *bvec;
804         struct btrfs_root *root;
805         int i, ret = 0;
806
807         ASSERT(!bio_flagged(bio, BIO_CLONED));
808         bio_for_each_segment_all(bvec, bio, i) {
809                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
810                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
811                 if (ret)
812                         break;
813         }
814
815         return errno_to_blk_status(ret);
816 }
817
818 static blk_status_t __btree_submit_bio_start(void *private_data, struct bio *bio,
819                                              int mirror_num, unsigned long bio_flags,
820                                              u64 bio_offset)
821 {
822         /*
823          * when we're called for a write, we're already in the async
824          * submission context.  Just jump into btrfs_map_bio
825          */
826         return btree_csum_one_bio(bio);
827 }
828
829 static blk_status_t __btree_submit_bio_done(void *private_data, struct bio *bio,
830                                             int mirror_num, unsigned long bio_flags,
831                                             u64 bio_offset)
832 {
833         struct inode *inode = private_data;
834         blk_status_t ret;
835
836         /*
837          * when we're called for a write, we're already in the async
838          * submission context.  Just jump into btrfs_map_bio
839          */
840         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
841         if (ret) {
842                 bio->bi_status = ret;
843                 bio_endio(bio);
844         }
845         return ret;
846 }
847
848 static int check_async_write(struct btrfs_inode *bi)
849 {
850         if (atomic_read(&bi->sync_writers))
851                 return 0;
852 #ifdef CONFIG_X86
853         if (static_cpu_has(X86_FEATURE_XMM4_2))
854                 return 0;
855 #endif
856         return 1;
857 }
858
859 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
860                                           int mirror_num, unsigned long bio_flags,
861                                           u64 bio_offset)
862 {
863         struct inode *inode = private_data;
864         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
865         int async = check_async_write(BTRFS_I(inode));
866         blk_status_t ret;
867
868         if (bio_op(bio) != REQ_OP_WRITE) {
869                 /*
870                  * called for a read, do the setup so that checksum validation
871                  * can happen in the async kernel threads
872                  */
873                 ret = btrfs_bio_wq_end_io(fs_info, bio,
874                                           BTRFS_WQ_ENDIO_METADATA);
875                 if (ret)
876                         goto out_w_error;
877                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
878         } else if (!async) {
879                 ret = btree_csum_one_bio(bio);
880                 if (ret)
881                         goto out_w_error;
882                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
883         } else {
884                 /*
885                  * kthread helpers are used to submit writes so that
886                  * checksumming can happen in parallel across all CPUs
887                  */
888                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
889                                           bio_offset, private_data,
890                                           __btree_submit_bio_start,
891                                           __btree_submit_bio_done);
892         }
893
894         if (ret)
895                 goto out_w_error;
896         return 0;
897
898 out_w_error:
899         bio->bi_status = ret;
900         bio_endio(bio);
901         return ret;
902 }
903
904 #ifdef CONFIG_MIGRATION
905 static int btree_migratepage(struct address_space *mapping,
906                         struct page *newpage, struct page *page,
907                         enum migrate_mode mode)
908 {
909         /*
910          * we can't safely write a btree page from here,
911          * we haven't done the locking hook
912          */
913         if (PageDirty(page))
914                 return -EAGAIN;
915         /*
916          * Buffers may be managed in a filesystem specific way.
917          * We must have no buffers or drop them.
918          */
919         if (page_has_private(page) &&
920             !try_to_release_page(page, GFP_KERNEL))
921                 return -EAGAIN;
922         return migrate_page(mapping, newpage, page, mode);
923 }
924 #endif
925
926
927 static int btree_writepages(struct address_space *mapping,
928                             struct writeback_control *wbc)
929 {
930         struct btrfs_fs_info *fs_info;
931         int ret;
932
933         if (wbc->sync_mode == WB_SYNC_NONE) {
934
935                 if (wbc->for_kupdate)
936                         return 0;
937
938                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
939                 /* this is a bit racy, but that's ok */
940                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
941                                              BTRFS_DIRTY_METADATA_THRESH);
942                 if (ret < 0)
943                         return 0;
944         }
945         return btree_write_cache_pages(mapping, wbc);
946 }
947
948 static int btree_readpage(struct file *file, struct page *page)
949 {
950         struct extent_io_tree *tree;
951         tree = &BTRFS_I(page->mapping->host)->io_tree;
952         return extent_read_full_page(tree, page, btree_get_extent, 0);
953 }
954
955 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
956 {
957         if (PageWriteback(page) || PageDirty(page))
958                 return 0;
959
960         return try_release_extent_buffer(page);
961 }
962
963 static void btree_invalidatepage(struct page *page, unsigned int offset,
964                                  unsigned int length)
965 {
966         struct extent_io_tree *tree;
967         tree = &BTRFS_I(page->mapping->host)->io_tree;
968         extent_invalidatepage(tree, page, offset);
969         btree_releasepage(page, GFP_NOFS);
970         if (PagePrivate(page)) {
971                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
972                            "page private not zero on page %llu",
973                            (unsigned long long)page_offset(page));
974                 ClearPagePrivate(page);
975                 set_page_private(page, 0);
976                 put_page(page);
977         }
978 }
979
980 static int btree_set_page_dirty(struct page *page)
981 {
982 #ifdef DEBUG
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 #endif
992         return __set_page_dirty_nobuffers(page);
993 }
994
995 static const struct address_space_operations btree_aops = {
996         .readpage       = btree_readpage,
997         .writepages     = btree_writepages,
998         .releasepage    = btree_releasepage,
999         .invalidatepage = btree_invalidatepage,
1000 #ifdef CONFIG_MIGRATION
1001         .migratepage    = btree_migratepage,
1002 #endif
1003         .set_page_dirty = btree_set_page_dirty,
1004 };
1005
1006 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1007 {
1008         struct extent_buffer *buf = NULL;
1009         struct inode *btree_inode = fs_info->btree_inode;
1010
1011         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1012         if (IS_ERR(buf))
1013                 return;
1014         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1015                                  buf, WAIT_NONE, 0);
1016         free_extent_buffer(buf);
1017 }
1018
1019 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1020                          int mirror_num, struct extent_buffer **eb)
1021 {
1022         struct extent_buffer *buf = NULL;
1023         struct inode *btree_inode = fs_info->btree_inode;
1024         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1025         int ret;
1026
1027         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1028         if (IS_ERR(buf))
1029                 return 0;
1030
1031         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1032
1033         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1034                                        mirror_num);
1035         if (ret) {
1036                 free_extent_buffer(buf);
1037                 return ret;
1038         }
1039
1040         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041                 free_extent_buffer(buf);
1042                 return -EIO;
1043         } else if (extent_buffer_uptodate(buf)) {
1044                 *eb = buf;
1045         } else {
1046                 free_extent_buffer(buf);
1047         }
1048         return 0;
1049 }
1050
1051 struct extent_buffer *btrfs_find_create_tree_block(
1052                                                 struct btrfs_fs_info *fs_info,
1053                                                 u64 bytenr)
1054 {
1055         if (btrfs_is_testing(fs_info))
1056                 return alloc_test_extent_buffer(fs_info, bytenr);
1057         return alloc_extent_buffer(fs_info, bytenr);
1058 }
1059
1060
1061 int btrfs_write_tree_block(struct extent_buffer *buf)
1062 {
1063         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1064                                         buf->start + buf->len - 1);
1065 }
1066
1067 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1068 {
1069         filemap_fdatawait_range(buf->pages[0]->mapping,
1070                                 buf->start, buf->start + buf->len - 1);
1071 }
1072
1073 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1074                                       u64 parent_transid)
1075 {
1076         struct extent_buffer *buf = NULL;
1077         int ret;
1078
1079         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1080         if (IS_ERR(buf))
1081                 return buf;
1082
1083         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid);
1084         if (ret) {
1085                 free_extent_buffer(buf);
1086                 return ERR_PTR(ret);
1087         }
1088         return buf;
1089
1090 }
1091
1092 void clean_tree_block(struct btrfs_fs_info *fs_info,
1093                       struct extent_buffer *buf)
1094 {
1095         if (btrfs_header_generation(buf) ==
1096             fs_info->running_transaction->transid) {
1097                 btrfs_assert_tree_locked(buf);
1098
1099                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1100                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1101                                                  -buf->len,
1102                                                  fs_info->dirty_metadata_batch);
1103                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1104                         btrfs_set_lock_blocking(buf);
1105                         clear_extent_buffer_dirty(buf);
1106                 }
1107         }
1108 }
1109
1110 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1111 {
1112         struct btrfs_subvolume_writers *writers;
1113         int ret;
1114
1115         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1116         if (!writers)
1117                 return ERR_PTR(-ENOMEM);
1118
1119         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1120         if (ret < 0) {
1121                 kfree(writers);
1122                 return ERR_PTR(ret);
1123         }
1124
1125         init_waitqueue_head(&writers->wait);
1126         return writers;
1127 }
1128
1129 static void
1130 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1131 {
1132         percpu_counter_destroy(&writers->counter);
1133         kfree(writers);
1134 }
1135
1136 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1137                          u64 objectid)
1138 {
1139         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1140         root->node = NULL;
1141         root->commit_root = NULL;
1142         root->state = 0;
1143         root->orphan_cleanup_state = 0;
1144
1145         root->objectid = objectid;
1146         root->last_trans = 0;
1147         root->highest_objectid = 0;
1148         root->nr_delalloc_inodes = 0;
1149         root->nr_ordered_extents = 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         INIT_LIST_HEAD(&root->delalloc_inodes);
1159         INIT_LIST_HEAD(&root->delalloc_root);
1160         INIT_LIST_HEAD(&root->ordered_extents);
1161         INIT_LIST_HEAD(&root->ordered_root);
1162         INIT_LIST_HEAD(&root->logged_list[0]);
1163         INIT_LIST_HEAD(&root->logged_list[1]);
1164         spin_lock_init(&root->orphan_lock);
1165         spin_lock_init(&root->inode_lock);
1166         spin_lock_init(&root->delalloc_lock);
1167         spin_lock_init(&root->ordered_extent_lock);
1168         spin_lock_init(&root->accounting_lock);
1169         spin_lock_init(&root->log_extents_lock[0]);
1170         spin_lock_init(&root->log_extents_lock[1]);
1171         mutex_init(&root->objectid_mutex);
1172         mutex_init(&root->log_mutex);
1173         mutex_init(&root->ordered_extent_mutex);
1174         mutex_init(&root->delalloc_mutex);
1175         init_waitqueue_head(&root->log_writer_wait);
1176         init_waitqueue_head(&root->log_commit_wait[0]);
1177         init_waitqueue_head(&root->log_commit_wait[1]);
1178         INIT_LIST_HEAD(&root->log_ctxs[0]);
1179         INIT_LIST_HEAD(&root->log_ctxs[1]);
1180         atomic_set(&root->log_commit[0], 0);
1181         atomic_set(&root->log_commit[1], 0);
1182         atomic_set(&root->log_writers, 0);
1183         atomic_set(&root->log_batch, 0);
1184         atomic_set(&root->orphan_inodes, 0);
1185         refcount_set(&root->refs, 1);
1186         atomic_set(&root->will_be_snapshotted, 0);
1187         atomic64_set(&root->qgroup_meta_rsv, 0);
1188         root->log_transid = 0;
1189         root->log_transid_committed = -1;
1190         root->last_log_commit = 0;
1191         if (!dummy)
1192                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1193
1194         memset(&root->root_key, 0, sizeof(root->root_key));
1195         memset(&root->root_item, 0, sizeof(root->root_item));
1196         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1197         if (!dummy)
1198                 root->defrag_trans_start = fs_info->generation;
1199         else
1200                 root->defrag_trans_start = 0;
1201         root->root_key.objectid = objectid;
1202         root->anon_dev = 0;
1203
1204         spin_lock_init(&root->root_item_lock);
1205 }
1206
1207 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1208                 gfp_t flags)
1209 {
1210         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1211         if (root)
1212                 root->fs_info = fs_info;
1213         return root;
1214 }
1215
1216 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1217 /* Should only be used by the testing infrastructure */
1218 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1219 {
1220         struct btrfs_root *root;
1221
1222         if (!fs_info)
1223                 return ERR_PTR(-EINVAL);
1224
1225         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1226         if (!root)
1227                 return ERR_PTR(-ENOMEM);
1228
1229         /* We don't use the stripesize in selftest, set it as sectorsize */
1230         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1231         root->alloc_bytenr = 0;
1232
1233         return root;
1234 }
1235 #endif
1236
1237 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1238                                      struct btrfs_fs_info *fs_info,
1239                                      u64 objectid)
1240 {
1241         struct extent_buffer *leaf;
1242         struct btrfs_root *tree_root = fs_info->tree_root;
1243         struct btrfs_root *root;
1244         struct btrfs_key key;
1245         int ret = 0;
1246         uuid_le uuid = NULL_UUID_LE;
1247
1248         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1249         if (!root)
1250                 return ERR_PTR(-ENOMEM);
1251
1252         __setup_root(root, fs_info, objectid);
1253         root->root_key.objectid = objectid;
1254         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1255         root->root_key.offset = 0;
1256
1257         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1258         if (IS_ERR(leaf)) {
1259                 ret = PTR_ERR(leaf);
1260                 leaf = NULL;
1261                 goto fail;
1262         }
1263
1264         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1265         btrfs_set_header_bytenr(leaf, leaf->start);
1266         btrfs_set_header_generation(leaf, trans->transid);
1267         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1268         btrfs_set_header_owner(leaf, objectid);
1269         root->node = leaf;
1270
1271         write_extent_buffer_fsid(leaf, fs_info->fsid);
1272         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1273         btrfs_mark_buffer_dirty(leaf);
1274
1275         root->commit_root = btrfs_root_node(root);
1276         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1277
1278         root->root_item.flags = 0;
1279         root->root_item.byte_limit = 0;
1280         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1281         btrfs_set_root_generation(&root->root_item, trans->transid);
1282         btrfs_set_root_level(&root->root_item, 0);
1283         btrfs_set_root_refs(&root->root_item, 1);
1284         btrfs_set_root_used(&root->root_item, leaf->len);
1285         btrfs_set_root_last_snapshot(&root->root_item, 0);
1286         btrfs_set_root_dirid(&root->root_item, 0);
1287         if (is_fstree(objectid))
1288                 uuid_le_gen(&uuid);
1289         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1290         root->root_item.drop_level = 0;
1291
1292         key.objectid = objectid;
1293         key.type = BTRFS_ROOT_ITEM_KEY;
1294         key.offset = 0;
1295         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1296         if (ret)
1297                 goto fail;
1298
1299         btrfs_tree_unlock(leaf);
1300
1301         return root;
1302
1303 fail:
1304         if (leaf) {
1305                 btrfs_tree_unlock(leaf);
1306                 free_extent_buffer(root->commit_root);
1307                 free_extent_buffer(leaf);
1308         }
1309         kfree(root);
1310
1311         return ERR_PTR(ret);
1312 }
1313
1314 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1315                                          struct btrfs_fs_info *fs_info)
1316 {
1317         struct btrfs_root *root;
1318         struct extent_buffer *leaf;
1319
1320         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1321         if (!root)
1322                 return ERR_PTR(-ENOMEM);
1323
1324         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1325
1326         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1327         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1328         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1329
1330         /*
1331          * DON'T set REF_COWS for log trees
1332          *
1333          * log trees do not get reference counted because they go away
1334          * before a real commit is actually done.  They do store pointers
1335          * to file data extents, and those reference counts still get
1336          * updated (along with back refs to the log tree).
1337          */
1338
1339         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1340                         NULL, 0, 0, 0);
1341         if (IS_ERR(leaf)) {
1342                 kfree(root);
1343                 return ERR_CAST(leaf);
1344         }
1345
1346         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1347         btrfs_set_header_bytenr(leaf, leaf->start);
1348         btrfs_set_header_generation(leaf, trans->transid);
1349         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1350         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1351         root->node = leaf;
1352
1353         write_extent_buffer_fsid(root->node, fs_info->fsid);
1354         btrfs_mark_buffer_dirty(root->node);
1355         btrfs_tree_unlock(root->node);
1356         return root;
1357 }
1358
1359 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1360                              struct btrfs_fs_info *fs_info)
1361 {
1362         struct btrfs_root *log_root;
1363
1364         log_root = alloc_log_tree(trans, fs_info);
1365         if (IS_ERR(log_root))
1366                 return PTR_ERR(log_root);
1367         WARN_ON(fs_info->log_root_tree);
1368         fs_info->log_root_tree = log_root;
1369         return 0;
1370 }
1371
1372 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1373                        struct btrfs_root *root)
1374 {
1375         struct btrfs_fs_info *fs_info = root->fs_info;
1376         struct btrfs_root *log_root;
1377         struct btrfs_inode_item *inode_item;
1378
1379         log_root = alloc_log_tree(trans, fs_info);
1380         if (IS_ERR(log_root))
1381                 return PTR_ERR(log_root);
1382
1383         log_root->last_trans = trans->transid;
1384         log_root->root_key.offset = root->root_key.objectid;
1385
1386         inode_item = &log_root->root_item.inode;
1387         btrfs_set_stack_inode_generation(inode_item, 1);
1388         btrfs_set_stack_inode_size(inode_item, 3);
1389         btrfs_set_stack_inode_nlink(inode_item, 1);
1390         btrfs_set_stack_inode_nbytes(inode_item,
1391                                      fs_info->nodesize);
1392         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1393
1394         btrfs_set_root_node(&log_root->root_item, log_root->node);
1395
1396         WARN_ON(root->log_root);
1397         root->log_root = log_root;
1398         root->log_transid = 0;
1399         root->log_transid_committed = -1;
1400         root->last_log_commit = 0;
1401         return 0;
1402 }
1403
1404 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1405                                                struct btrfs_key *key)
1406 {
1407         struct btrfs_root *root;
1408         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1409         struct btrfs_path *path;
1410         u64 generation;
1411         int ret;
1412
1413         path = btrfs_alloc_path();
1414         if (!path)
1415                 return ERR_PTR(-ENOMEM);
1416
1417         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1418         if (!root) {
1419                 ret = -ENOMEM;
1420                 goto alloc_fail;
1421         }
1422
1423         __setup_root(root, fs_info, key->objectid);
1424
1425         ret = btrfs_find_root(tree_root, key, path,
1426                               &root->root_item, &root->root_key);
1427         if (ret) {
1428                 if (ret > 0)
1429                         ret = -ENOENT;
1430                 goto find_fail;
1431         }
1432
1433         generation = btrfs_root_generation(&root->root_item);
1434         root->node = read_tree_block(fs_info,
1435                                      btrfs_root_bytenr(&root->root_item),
1436                                      generation);
1437         if (IS_ERR(root->node)) {
1438                 ret = PTR_ERR(root->node);
1439                 goto find_fail;
1440         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1441                 ret = -EIO;
1442                 free_extent_buffer(root->node);
1443                 goto find_fail;
1444         }
1445         root->commit_root = btrfs_root_node(root);
1446 out:
1447         btrfs_free_path(path);
1448         return root;
1449
1450 find_fail:
1451         kfree(root);
1452 alloc_fail:
1453         root = ERR_PTR(ret);
1454         goto out;
1455 }
1456
1457 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1458                                       struct btrfs_key *location)
1459 {
1460         struct btrfs_root *root;
1461
1462         root = btrfs_read_tree_root(tree_root, location);
1463         if (IS_ERR(root))
1464                 return root;
1465
1466         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1467                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1468                 btrfs_check_and_init_root_item(&root->root_item);
1469         }
1470
1471         return root;
1472 }
1473
1474 int btrfs_init_fs_root(struct btrfs_root *root)
1475 {
1476         int ret;
1477         struct btrfs_subvolume_writers *writers;
1478
1479         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1480         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1481                                         GFP_NOFS);
1482         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1483                 ret = -ENOMEM;
1484                 goto fail;
1485         }
1486
1487         writers = btrfs_alloc_subvolume_writers();
1488         if (IS_ERR(writers)) {
1489                 ret = PTR_ERR(writers);
1490                 goto fail;
1491         }
1492         root->subv_writers = writers;
1493
1494         btrfs_init_free_ino_ctl(root);
1495         spin_lock_init(&root->ino_cache_lock);
1496         init_waitqueue_head(&root->ino_cache_wait);
1497
1498         ret = get_anon_bdev(&root->anon_dev);
1499         if (ret)
1500                 goto fail;
1501
1502         mutex_lock(&root->objectid_mutex);
1503         ret = btrfs_find_highest_objectid(root,
1504                                         &root->highest_objectid);
1505         if (ret) {
1506                 mutex_unlock(&root->objectid_mutex);
1507                 goto fail;
1508         }
1509
1510         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1511
1512         mutex_unlock(&root->objectid_mutex);
1513
1514         return 0;
1515 fail:
1516         /* the caller is responsible to call free_fs_root */
1517         return ret;
1518 }
1519
1520 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1521                                         u64 root_id)
1522 {
1523         struct btrfs_root *root;
1524
1525         spin_lock(&fs_info->fs_roots_radix_lock);
1526         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1527                                  (unsigned long)root_id);
1528         spin_unlock(&fs_info->fs_roots_radix_lock);
1529         return root;
1530 }
1531
1532 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1533                          struct btrfs_root *root)
1534 {
1535         int ret;
1536
1537         ret = radix_tree_preload(GFP_NOFS);
1538         if (ret)
1539                 return ret;
1540
1541         spin_lock(&fs_info->fs_roots_radix_lock);
1542         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1543                                 (unsigned long)root->root_key.objectid,
1544                                 root);
1545         if (ret == 0)
1546                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1547         spin_unlock(&fs_info->fs_roots_radix_lock);
1548         radix_tree_preload_end();
1549
1550         return ret;
1551 }
1552
1553 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1554                                      struct btrfs_key *location,
1555                                      bool check_ref)
1556 {
1557         struct btrfs_root *root;
1558         struct btrfs_path *path;
1559         struct btrfs_key key;
1560         int ret;
1561
1562         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1563                 return fs_info->tree_root;
1564         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1565                 return fs_info->extent_root;
1566         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1567                 return fs_info->chunk_root;
1568         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1569                 return fs_info->dev_root;
1570         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1571                 return fs_info->csum_root;
1572         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1573                 return fs_info->quota_root ? fs_info->quota_root :
1574                                              ERR_PTR(-ENOENT);
1575         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1576                 return fs_info->uuid_root ? fs_info->uuid_root :
1577                                             ERR_PTR(-ENOENT);
1578         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1579                 return fs_info->free_space_root ? fs_info->free_space_root :
1580                                                   ERR_PTR(-ENOENT);
1581 again:
1582         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1583         if (root) {
1584                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1585                         return ERR_PTR(-ENOENT);
1586                 return root;
1587         }
1588
1589         root = btrfs_read_fs_root(fs_info->tree_root, location);
1590         if (IS_ERR(root))
1591                 return root;
1592
1593         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1594                 ret = -ENOENT;
1595                 goto fail;
1596         }
1597
1598         ret = btrfs_init_fs_root(root);
1599         if (ret)
1600                 goto fail;
1601
1602         path = btrfs_alloc_path();
1603         if (!path) {
1604                 ret = -ENOMEM;
1605                 goto fail;
1606         }
1607         key.objectid = BTRFS_ORPHAN_OBJECTID;
1608         key.type = BTRFS_ORPHAN_ITEM_KEY;
1609         key.offset = location->objectid;
1610
1611         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1612         btrfs_free_path(path);
1613         if (ret < 0)
1614                 goto fail;
1615         if (ret == 0)
1616                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1617
1618         ret = btrfs_insert_fs_root(fs_info, root);
1619         if (ret) {
1620                 if (ret == -EEXIST) {
1621                         free_fs_root(root);
1622                         goto again;
1623                 }
1624                 goto fail;
1625         }
1626         return root;
1627 fail:
1628         free_fs_root(root);
1629         return ERR_PTR(ret);
1630 }
1631
1632 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1633 {
1634         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1635         int ret = 0;
1636         struct btrfs_device *device;
1637         struct backing_dev_info *bdi;
1638
1639         rcu_read_lock();
1640         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1641                 if (!device->bdev)
1642                         continue;
1643                 bdi = device->bdev->bd_bdi;
1644                 if (bdi_congested(bdi, bdi_bits)) {
1645                         ret = 1;
1646                         break;
1647                 }
1648         }
1649         rcu_read_unlock();
1650         return ret;
1651 }
1652
1653 /*
1654  * called by the kthread helper functions to finally call the bio end_io
1655  * functions.  This is where read checksum verification actually happens
1656  */
1657 static void end_workqueue_fn(struct btrfs_work *work)
1658 {
1659         struct bio *bio;
1660         struct btrfs_end_io_wq *end_io_wq;
1661
1662         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1663         bio = end_io_wq->bio;
1664
1665         bio->bi_status = end_io_wq->status;
1666         bio->bi_private = end_io_wq->private;
1667         bio->bi_end_io = end_io_wq->end_io;
1668         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1669         bio_endio(bio);
1670 }
1671
1672 static int cleaner_kthread(void *arg)
1673 {
1674         struct btrfs_root *root = arg;
1675         struct btrfs_fs_info *fs_info = root->fs_info;
1676         int again;
1677         struct btrfs_trans_handle *trans;
1678
1679         do {
1680                 again = 0;
1681
1682                 /* Make the cleaner go to sleep early. */
1683                 if (btrfs_need_cleaner_sleep(fs_info))
1684                         goto sleep;
1685
1686                 /*
1687                  * Do not do anything if we might cause open_ctree() to block
1688                  * before we have finished mounting the filesystem.
1689                  */
1690                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1691                         goto sleep;
1692
1693                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1694                         goto sleep;
1695
1696                 /*
1697                  * Avoid the problem that we change the status of the fs
1698                  * during the above check and trylock.
1699                  */
1700                 if (btrfs_need_cleaner_sleep(fs_info)) {
1701                         mutex_unlock(&fs_info->cleaner_mutex);
1702                         goto sleep;
1703                 }
1704
1705                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1706                 btrfs_run_delayed_iputs(fs_info);
1707                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1708
1709                 again = btrfs_clean_one_deleted_snapshot(root);
1710                 mutex_unlock(&fs_info->cleaner_mutex);
1711
1712                 /*
1713                  * The defragger has dealt with the R/O remount and umount,
1714                  * needn't do anything special here.
1715                  */
1716                 btrfs_run_defrag_inodes(fs_info);
1717
1718                 /*
1719                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1720                  * with relocation (btrfs_relocate_chunk) and relocation
1721                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1722                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1723                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1724                  * unused block groups.
1725                  */
1726                 btrfs_delete_unused_bgs(fs_info);
1727 sleep:
1728                 if (!again) {
1729                         set_current_state(TASK_INTERRUPTIBLE);
1730                         if (!kthread_should_stop())
1731                                 schedule();
1732                         __set_current_state(TASK_RUNNING);
1733                 }
1734         } while (!kthread_should_stop());
1735
1736         /*
1737          * Transaction kthread is stopped before us and wakes us up.
1738          * However we might have started a new transaction and COWed some
1739          * tree blocks when deleting unused block groups for example. So
1740          * make sure we commit the transaction we started to have a clean
1741          * shutdown when evicting the btree inode - if it has dirty pages
1742          * when we do the final iput() on it, eviction will trigger a
1743          * writeback for it which will fail with null pointer dereferences
1744          * since work queues and other resources were already released and
1745          * destroyed by the time the iput/eviction/writeback is made.
1746          */
1747         trans = btrfs_attach_transaction(root);
1748         if (IS_ERR(trans)) {
1749                 if (PTR_ERR(trans) != -ENOENT)
1750                         btrfs_err(fs_info,
1751                                   "cleaner transaction attach returned %ld",
1752                                   PTR_ERR(trans));
1753         } else {
1754                 int ret;
1755
1756                 ret = btrfs_commit_transaction(trans);
1757                 if (ret)
1758                         btrfs_err(fs_info,
1759                                   "cleaner open transaction commit returned %d",
1760                                   ret);
1761         }
1762
1763         return 0;
1764 }
1765
1766 static int transaction_kthread(void *arg)
1767 {
1768         struct btrfs_root *root = arg;
1769         struct btrfs_fs_info *fs_info = root->fs_info;
1770         struct btrfs_trans_handle *trans;
1771         struct btrfs_transaction *cur;
1772         u64 transid;
1773         unsigned long now;
1774         unsigned long delay;
1775         bool cannot_commit;
1776
1777         do {
1778                 cannot_commit = false;
1779                 delay = HZ * fs_info->commit_interval;
1780                 mutex_lock(&fs_info->transaction_kthread_mutex);
1781
1782                 spin_lock(&fs_info->trans_lock);
1783                 cur = fs_info->running_transaction;
1784                 if (!cur) {
1785                         spin_unlock(&fs_info->trans_lock);
1786                         goto sleep;
1787                 }
1788
1789                 now = get_seconds();
1790                 if (cur->state < TRANS_STATE_BLOCKED &&
1791                     (now < cur->start_time ||
1792                      now - cur->start_time < fs_info->commit_interval)) {
1793                         spin_unlock(&fs_info->trans_lock);
1794                         delay = HZ * 5;
1795                         goto sleep;
1796                 }
1797                 transid = cur->transid;
1798                 spin_unlock(&fs_info->trans_lock);
1799
1800                 /* If the file system is aborted, this will always fail. */
1801                 trans = btrfs_attach_transaction(root);
1802                 if (IS_ERR(trans)) {
1803                         if (PTR_ERR(trans) != -ENOENT)
1804                                 cannot_commit = true;
1805                         goto sleep;
1806                 }
1807                 if (transid == trans->transid) {
1808                         btrfs_commit_transaction(trans);
1809                 } else {
1810                         btrfs_end_transaction(trans);
1811                 }
1812 sleep:
1813                 wake_up_process(fs_info->cleaner_kthread);
1814                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1815
1816                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1817                                       &fs_info->fs_state)))
1818                         btrfs_cleanup_transaction(fs_info);
1819                 set_current_state(TASK_INTERRUPTIBLE);
1820                 if (!kthread_should_stop() &&
1821                                 (!btrfs_transaction_blocked(fs_info) ||
1822                                  cannot_commit))
1823                         schedule_timeout(delay);
1824                 __set_current_state(TASK_RUNNING);
1825         } while (!kthread_should_stop());
1826         return 0;
1827 }
1828
1829 /*
1830  * this will find the highest generation in the array of
1831  * root backups.  The index of the highest array is returned,
1832  * or -1 if we can't find anything.
1833  *
1834  * We check to make sure the array is valid by comparing the
1835  * generation of the latest  root in the array with the generation
1836  * in the super block.  If they don't match we pitch it.
1837  */
1838 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1839 {
1840         u64 cur;
1841         int newest_index = -1;
1842         struct btrfs_root_backup *root_backup;
1843         int i;
1844
1845         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1846                 root_backup = info->super_copy->super_roots + i;
1847                 cur = btrfs_backup_tree_root_gen(root_backup);
1848                 if (cur == newest_gen)
1849                         newest_index = i;
1850         }
1851
1852         /* check to see if we actually wrapped around */
1853         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1854                 root_backup = info->super_copy->super_roots;
1855                 cur = btrfs_backup_tree_root_gen(root_backup);
1856                 if (cur == newest_gen)
1857                         newest_index = 0;
1858         }
1859         return newest_index;
1860 }
1861
1862
1863 /*
1864  * find the oldest backup so we know where to store new entries
1865  * in the backup array.  This will set the backup_root_index
1866  * field in the fs_info struct
1867  */
1868 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1869                                      u64 newest_gen)
1870 {
1871         int newest_index = -1;
1872
1873         newest_index = find_newest_super_backup(info, newest_gen);
1874         /* if there was garbage in there, just move along */
1875         if (newest_index == -1) {
1876                 info->backup_root_index = 0;
1877         } else {
1878                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1879         }
1880 }
1881
1882 /*
1883  * copy all the root pointers into the super backup array.
1884  * this will bump the backup pointer by one when it is
1885  * done
1886  */
1887 static void backup_super_roots(struct btrfs_fs_info *info)
1888 {
1889         int next_backup;
1890         struct btrfs_root_backup *root_backup;
1891         int last_backup;
1892
1893         next_backup = info->backup_root_index;
1894         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1895                 BTRFS_NUM_BACKUP_ROOTS;
1896
1897         /*
1898          * just overwrite the last backup if we're at the same generation
1899          * this happens only at umount
1900          */
1901         root_backup = info->super_for_commit->super_roots + last_backup;
1902         if (btrfs_backup_tree_root_gen(root_backup) ==
1903             btrfs_header_generation(info->tree_root->node))
1904                 next_backup = last_backup;
1905
1906         root_backup = info->super_for_commit->super_roots + next_backup;
1907
1908         /*
1909          * make sure all of our padding and empty slots get zero filled
1910          * regardless of which ones we use today
1911          */
1912         memset(root_backup, 0, sizeof(*root_backup));
1913
1914         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1915
1916         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1917         btrfs_set_backup_tree_root_gen(root_backup,
1918                                btrfs_header_generation(info->tree_root->node));
1919
1920         btrfs_set_backup_tree_root_level(root_backup,
1921                                btrfs_header_level(info->tree_root->node));
1922
1923         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1924         btrfs_set_backup_chunk_root_gen(root_backup,
1925                                btrfs_header_generation(info->chunk_root->node));
1926         btrfs_set_backup_chunk_root_level(root_backup,
1927                                btrfs_header_level(info->chunk_root->node));
1928
1929         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1930         btrfs_set_backup_extent_root_gen(root_backup,
1931                                btrfs_header_generation(info->extent_root->node));
1932         btrfs_set_backup_extent_root_level(root_backup,
1933                                btrfs_header_level(info->extent_root->node));
1934
1935         /*
1936          * we might commit during log recovery, which happens before we set
1937          * the fs_root.  Make sure it is valid before we fill it in.
1938          */
1939         if (info->fs_root && info->fs_root->node) {
1940                 btrfs_set_backup_fs_root(root_backup,
1941                                          info->fs_root->node->start);
1942                 btrfs_set_backup_fs_root_gen(root_backup,
1943                                btrfs_header_generation(info->fs_root->node));
1944                 btrfs_set_backup_fs_root_level(root_backup,
1945                                btrfs_header_level(info->fs_root->node));
1946         }
1947
1948         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1949         btrfs_set_backup_dev_root_gen(root_backup,
1950                                btrfs_header_generation(info->dev_root->node));
1951         btrfs_set_backup_dev_root_level(root_backup,
1952                                        btrfs_header_level(info->dev_root->node));
1953
1954         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1955         btrfs_set_backup_csum_root_gen(root_backup,
1956                                btrfs_header_generation(info->csum_root->node));
1957         btrfs_set_backup_csum_root_level(root_backup,
1958                                btrfs_header_level(info->csum_root->node));
1959
1960         btrfs_set_backup_total_bytes(root_backup,
1961                              btrfs_super_total_bytes(info->super_copy));
1962         btrfs_set_backup_bytes_used(root_backup,
1963                              btrfs_super_bytes_used(info->super_copy));
1964         btrfs_set_backup_num_devices(root_backup,
1965                              btrfs_super_num_devices(info->super_copy));
1966
1967         /*
1968          * if we don't copy this out to the super_copy, it won't get remembered
1969          * for the next commit
1970          */
1971         memcpy(&info->super_copy->super_roots,
1972                &info->super_for_commit->super_roots,
1973                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1974 }
1975
1976 /*
1977  * this copies info out of the root backup array and back into
1978  * the in-memory super block.  It is meant to help iterate through
1979  * the array, so you send it the number of backups you've already
1980  * tried and the last backup index you used.
1981  *
1982  * this returns -1 when it has tried all the backups
1983  */
1984 static noinline int next_root_backup(struct btrfs_fs_info *info,
1985                                      struct btrfs_super_block *super,
1986                                      int *num_backups_tried, int *backup_index)
1987 {
1988         struct btrfs_root_backup *root_backup;
1989         int newest = *backup_index;
1990
1991         if (*num_backups_tried == 0) {
1992                 u64 gen = btrfs_super_generation(super);
1993
1994                 newest = find_newest_super_backup(info, gen);
1995                 if (newest == -1)
1996                         return -1;
1997
1998                 *backup_index = newest;
1999                 *num_backups_tried = 1;
2000         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2001                 /* we've tried all the backups, all done */
2002                 return -1;
2003         } else {
2004                 /* jump to the next oldest backup */
2005                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2006                         BTRFS_NUM_BACKUP_ROOTS;
2007                 *backup_index = newest;
2008                 *num_backups_tried += 1;
2009         }
2010         root_backup = super->super_roots + newest;
2011
2012         btrfs_set_super_generation(super,
2013                                    btrfs_backup_tree_root_gen(root_backup));
2014         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2015         btrfs_set_super_root_level(super,
2016                                    btrfs_backup_tree_root_level(root_backup));
2017         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2018
2019         /*
2020          * fixme: the total bytes and num_devices need to match or we should
2021          * need a fsck
2022          */
2023         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2024         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2025         return 0;
2026 }
2027
2028 /* helper to cleanup workers */
2029 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2030 {
2031         btrfs_destroy_workqueue(fs_info->fixup_workers);
2032         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2033         btrfs_destroy_workqueue(fs_info->workers);
2034         btrfs_destroy_workqueue(fs_info->endio_workers);
2035         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2036         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2037         btrfs_destroy_workqueue(fs_info->rmw_workers);
2038         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2039         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2040         btrfs_destroy_workqueue(fs_info->submit_workers);
2041         btrfs_destroy_workqueue(fs_info->delayed_workers);
2042         btrfs_destroy_workqueue(fs_info->caching_workers);
2043         btrfs_destroy_workqueue(fs_info->readahead_workers);
2044         btrfs_destroy_workqueue(fs_info->flush_workers);
2045         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2046         btrfs_destroy_workqueue(fs_info->extent_workers);
2047         /*
2048          * Now that all other work queues are destroyed, we can safely destroy
2049          * the queues used for metadata I/O, since tasks from those other work
2050          * queues can do metadata I/O operations.
2051          */
2052         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2053         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2054 }
2055
2056 static void free_root_extent_buffers(struct btrfs_root *root)
2057 {
2058         if (root) {
2059                 free_extent_buffer(root->node);
2060                 free_extent_buffer(root->commit_root);
2061                 root->node = NULL;
2062                 root->commit_root = NULL;
2063         }
2064 }
2065
2066 /* helper to cleanup tree roots */
2067 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2068 {
2069         free_root_extent_buffers(info->tree_root);
2070
2071         free_root_extent_buffers(info->dev_root);
2072         free_root_extent_buffers(info->extent_root);
2073         free_root_extent_buffers(info->csum_root);
2074         free_root_extent_buffers(info->quota_root);
2075         free_root_extent_buffers(info->uuid_root);
2076         if (chunk_root)
2077                 free_root_extent_buffers(info->chunk_root);
2078         free_root_extent_buffers(info->free_space_root);
2079 }
2080
2081 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2082 {
2083         int ret;
2084         struct btrfs_root *gang[8];
2085         int i;
2086
2087         while (!list_empty(&fs_info->dead_roots)) {
2088                 gang[0] = list_entry(fs_info->dead_roots.next,
2089                                      struct btrfs_root, root_list);
2090                 list_del(&gang[0]->root_list);
2091
2092                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2093                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2094                 } else {
2095                         free_extent_buffer(gang[0]->node);
2096                         free_extent_buffer(gang[0]->commit_root);
2097                         btrfs_put_fs_root(gang[0]);
2098                 }
2099         }
2100
2101         while (1) {
2102                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2103                                              (void **)gang, 0,
2104                                              ARRAY_SIZE(gang));
2105                 if (!ret)
2106                         break;
2107                 for (i = 0; i < ret; i++)
2108                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2109         }
2110
2111         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2112                 btrfs_free_log_root_tree(NULL, fs_info);
2113                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2114         }
2115 }
2116
2117 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2118 {
2119         mutex_init(&fs_info->scrub_lock);
2120         atomic_set(&fs_info->scrubs_running, 0);
2121         atomic_set(&fs_info->scrub_pause_req, 0);
2122         atomic_set(&fs_info->scrubs_paused, 0);
2123         atomic_set(&fs_info->scrub_cancel_req, 0);
2124         init_waitqueue_head(&fs_info->scrub_pause_wait);
2125         fs_info->scrub_workers_refcnt = 0;
2126 }
2127
2128 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2129 {
2130         spin_lock_init(&fs_info->balance_lock);
2131         mutex_init(&fs_info->balance_mutex);
2132         atomic_set(&fs_info->balance_running, 0);
2133         atomic_set(&fs_info->balance_pause_req, 0);
2134         atomic_set(&fs_info->balance_cancel_req, 0);
2135         fs_info->balance_ctl = NULL;
2136         init_waitqueue_head(&fs_info->balance_wait_q);
2137 }
2138
2139 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2140 {
2141         struct inode *inode = fs_info->btree_inode;
2142
2143         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2144         set_nlink(inode, 1);
2145         /*
2146          * we set the i_size on the btree inode to the max possible int.
2147          * the real end of the address space is determined by all of
2148          * the devices in the system
2149          */
2150         inode->i_size = OFFSET_MAX;
2151         inode->i_mapping->a_ops = &btree_aops;
2152
2153         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2154         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2155         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2156         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2157
2158         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2159
2160         BTRFS_I(inode)->root = fs_info->tree_root;
2161         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2162         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2163         btrfs_insert_inode_hash(inode);
2164 }
2165
2166 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2167 {
2168         fs_info->dev_replace.lock_owner = 0;
2169         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2170         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2171         rwlock_init(&fs_info->dev_replace.lock);
2172         atomic_set(&fs_info->dev_replace.read_locks, 0);
2173         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2174         init_waitqueue_head(&fs_info->replace_wait);
2175         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2176 }
2177
2178 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2179 {
2180         spin_lock_init(&fs_info->qgroup_lock);
2181         mutex_init(&fs_info->qgroup_ioctl_lock);
2182         fs_info->qgroup_tree = RB_ROOT;
2183         fs_info->qgroup_op_tree = RB_ROOT;
2184         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2185         fs_info->qgroup_seq = 1;
2186         fs_info->qgroup_ulist = NULL;
2187         fs_info->qgroup_rescan_running = false;
2188         mutex_init(&fs_info->qgroup_rescan_lock);
2189 }
2190
2191 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2192                 struct btrfs_fs_devices *fs_devices)
2193 {
2194         int max_active = fs_info->thread_pool_size;
2195         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2196
2197         fs_info->workers =
2198                 btrfs_alloc_workqueue(fs_info, "worker",
2199                                       flags | WQ_HIGHPRI, max_active, 16);
2200
2201         fs_info->delalloc_workers =
2202                 btrfs_alloc_workqueue(fs_info, "delalloc",
2203                                       flags, max_active, 2);
2204
2205         fs_info->flush_workers =
2206                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2207                                       flags, max_active, 0);
2208
2209         fs_info->caching_workers =
2210                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2211
2212         /*
2213          * a higher idle thresh on the submit workers makes it much more
2214          * likely that bios will be send down in a sane order to the
2215          * devices
2216          */
2217         fs_info->submit_workers =
2218                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2219                                       min_t(u64, fs_devices->num_devices,
2220                                             max_active), 64);
2221
2222         fs_info->fixup_workers =
2223                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2224
2225         /*
2226          * endios are largely parallel and should have a very
2227          * low idle thresh
2228          */
2229         fs_info->endio_workers =
2230                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2231         fs_info->endio_meta_workers =
2232                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2233                                       max_active, 4);
2234         fs_info->endio_meta_write_workers =
2235                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2236                                       max_active, 2);
2237         fs_info->endio_raid56_workers =
2238                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2239                                       max_active, 4);
2240         fs_info->endio_repair_workers =
2241                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2242         fs_info->rmw_workers =
2243                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2244         fs_info->endio_write_workers =
2245                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2246                                       max_active, 2);
2247         fs_info->endio_freespace_worker =
2248                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2249                                       max_active, 0);
2250         fs_info->delayed_workers =
2251                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2252                                       max_active, 0);
2253         fs_info->readahead_workers =
2254                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2255                                       max_active, 2);
2256         fs_info->qgroup_rescan_workers =
2257                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2258         fs_info->extent_workers =
2259                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2260                                       min_t(u64, fs_devices->num_devices,
2261                                             max_active), 8);
2262
2263         if (!(fs_info->workers && fs_info->delalloc_workers &&
2264               fs_info->submit_workers && fs_info->flush_workers &&
2265               fs_info->endio_workers && fs_info->endio_meta_workers &&
2266               fs_info->endio_meta_write_workers &&
2267               fs_info->endio_repair_workers &&
2268               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2269               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2270               fs_info->caching_workers && fs_info->readahead_workers &&
2271               fs_info->fixup_workers && fs_info->delayed_workers &&
2272               fs_info->extent_workers &&
2273               fs_info->qgroup_rescan_workers)) {
2274                 return -ENOMEM;
2275         }
2276
2277         return 0;
2278 }
2279
2280 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2281                             struct btrfs_fs_devices *fs_devices)
2282 {
2283         int ret;
2284         struct btrfs_root *log_tree_root;
2285         struct btrfs_super_block *disk_super = fs_info->super_copy;
2286         u64 bytenr = btrfs_super_log_root(disk_super);
2287
2288         if (fs_devices->rw_devices == 0) {
2289                 btrfs_warn(fs_info, "log replay required on RO media");
2290                 return -EIO;
2291         }
2292
2293         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2294         if (!log_tree_root)
2295                 return -ENOMEM;
2296
2297         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2298
2299         log_tree_root->node = read_tree_block(fs_info, bytenr,
2300                                               fs_info->generation + 1);
2301         if (IS_ERR(log_tree_root->node)) {
2302                 btrfs_warn(fs_info, "failed to read log tree");
2303                 ret = PTR_ERR(log_tree_root->node);
2304                 kfree(log_tree_root);
2305                 return ret;
2306         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2307                 btrfs_err(fs_info, "failed to read log tree");
2308                 free_extent_buffer(log_tree_root->node);
2309                 kfree(log_tree_root);
2310                 return -EIO;
2311         }
2312         /* returns with log_tree_root freed on success */
2313         ret = btrfs_recover_log_trees(log_tree_root);
2314         if (ret) {
2315                 btrfs_handle_fs_error(fs_info, ret,
2316                                       "Failed to recover log tree");
2317                 free_extent_buffer(log_tree_root->node);
2318                 kfree(log_tree_root);
2319                 return ret;
2320         }
2321
2322         if (sb_rdonly(fs_info->sb)) {
2323                 ret = btrfs_commit_super(fs_info);
2324                 if (ret)
2325                         return ret;
2326         }
2327
2328         return 0;
2329 }
2330
2331 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2332 {
2333         struct btrfs_root *tree_root = fs_info->tree_root;
2334         struct btrfs_root *root;
2335         struct btrfs_key location;
2336         int ret;
2337
2338         BUG_ON(!fs_info->tree_root);
2339
2340         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2341         location.type = BTRFS_ROOT_ITEM_KEY;
2342         location.offset = 0;
2343
2344         root = btrfs_read_tree_root(tree_root, &location);
2345         if (IS_ERR(root))
2346                 return PTR_ERR(root);
2347         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2348         fs_info->extent_root = root;
2349
2350         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2351         root = btrfs_read_tree_root(tree_root, &location);
2352         if (IS_ERR(root))
2353                 return PTR_ERR(root);
2354         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2355         fs_info->dev_root = root;
2356         btrfs_init_devices_late(fs_info);
2357
2358         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2359         root = btrfs_read_tree_root(tree_root, &location);
2360         if (IS_ERR(root))
2361                 return PTR_ERR(root);
2362         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2363         fs_info->csum_root = root;
2364
2365         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2366         root = btrfs_read_tree_root(tree_root, &location);
2367         if (!IS_ERR(root)) {
2368                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2370                 fs_info->quota_root = root;
2371         }
2372
2373         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2374         root = btrfs_read_tree_root(tree_root, &location);
2375         if (IS_ERR(root)) {
2376                 ret = PTR_ERR(root);
2377                 if (ret != -ENOENT)
2378                         return ret;
2379         } else {
2380                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2381                 fs_info->uuid_root = root;
2382         }
2383
2384         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2385                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2386                 root = btrfs_read_tree_root(tree_root, &location);
2387                 if (IS_ERR(root))
2388                         return PTR_ERR(root);
2389                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2390                 fs_info->free_space_root = root;
2391         }
2392
2393         return 0;
2394 }
2395
2396 int open_ctree(struct super_block *sb,
2397                struct btrfs_fs_devices *fs_devices,
2398                char *options)
2399 {
2400         u32 sectorsize;
2401         u32 nodesize;
2402         u32 stripesize;
2403         u64 generation;
2404         u64 features;
2405         struct btrfs_key location;
2406         struct buffer_head *bh;
2407         struct btrfs_super_block *disk_super;
2408         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2409         struct btrfs_root *tree_root;
2410         struct btrfs_root *chunk_root;
2411         int ret;
2412         int err = -EINVAL;
2413         int num_backups_tried = 0;
2414         int backup_index = 0;
2415         int max_active;
2416         int clear_free_space_tree = 0;
2417
2418         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2419         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2420         if (!tree_root || !chunk_root) {
2421                 err = -ENOMEM;
2422                 goto fail;
2423         }
2424
2425         ret = init_srcu_struct(&fs_info->subvol_srcu);
2426         if (ret) {
2427                 err = ret;
2428                 goto fail;
2429         }
2430
2431         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2432         if (ret) {
2433                 err = ret;
2434                 goto fail_srcu;
2435         }
2436         fs_info->dirty_metadata_batch = PAGE_SIZE *
2437                                         (1 + ilog2(nr_cpu_ids));
2438
2439         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2440         if (ret) {
2441                 err = ret;
2442                 goto fail_dirty_metadata_bytes;
2443         }
2444
2445         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2446         if (ret) {
2447                 err = ret;
2448                 goto fail_delalloc_bytes;
2449         }
2450
2451         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2452         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2453         INIT_LIST_HEAD(&fs_info->trans_list);
2454         INIT_LIST_HEAD(&fs_info->dead_roots);
2455         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2456         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2457         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2458         spin_lock_init(&fs_info->delalloc_root_lock);
2459         spin_lock_init(&fs_info->trans_lock);
2460         spin_lock_init(&fs_info->fs_roots_radix_lock);
2461         spin_lock_init(&fs_info->delayed_iput_lock);
2462         spin_lock_init(&fs_info->defrag_inodes_lock);
2463         spin_lock_init(&fs_info->tree_mod_seq_lock);
2464         spin_lock_init(&fs_info->super_lock);
2465         spin_lock_init(&fs_info->qgroup_op_lock);
2466         spin_lock_init(&fs_info->buffer_lock);
2467         spin_lock_init(&fs_info->unused_bgs_lock);
2468         rwlock_init(&fs_info->tree_mod_log_lock);
2469         mutex_init(&fs_info->unused_bg_unpin_mutex);
2470         mutex_init(&fs_info->delete_unused_bgs_mutex);
2471         mutex_init(&fs_info->reloc_mutex);
2472         mutex_init(&fs_info->delalloc_root_mutex);
2473         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2474         seqlock_init(&fs_info->profiles_lock);
2475
2476         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2477         INIT_LIST_HEAD(&fs_info->space_info);
2478         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2479         INIT_LIST_HEAD(&fs_info->unused_bgs);
2480         btrfs_mapping_init(&fs_info->mapping_tree);
2481         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2482                              BTRFS_BLOCK_RSV_GLOBAL);
2483         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2484         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2485         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2486         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2487                              BTRFS_BLOCK_RSV_DELOPS);
2488         atomic_set(&fs_info->async_delalloc_pages, 0);
2489         atomic_set(&fs_info->defrag_running, 0);
2490         atomic_set(&fs_info->qgroup_op_seq, 0);
2491         atomic_set(&fs_info->reada_works_cnt, 0);
2492         atomic64_set(&fs_info->tree_mod_seq, 0);
2493         fs_info->sb = sb;
2494         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2495         fs_info->metadata_ratio = 0;
2496         fs_info->defrag_inodes = RB_ROOT;
2497         atomic64_set(&fs_info->free_chunk_space, 0);
2498         fs_info->tree_mod_log = RB_ROOT;
2499         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2500         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2501         /* readahead state */
2502         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2503         spin_lock_init(&fs_info->reada_lock);
2504         btrfs_init_ref_verify(fs_info);
2505
2506         fs_info->thread_pool_size = min_t(unsigned long,
2507                                           num_online_cpus() + 2, 8);
2508
2509         INIT_LIST_HEAD(&fs_info->ordered_roots);
2510         spin_lock_init(&fs_info->ordered_root_lock);
2511
2512         fs_info->btree_inode = new_inode(sb);
2513         if (!fs_info->btree_inode) {
2514                 err = -ENOMEM;
2515                 goto fail_bio_counter;
2516         }
2517         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2518
2519         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2520                                         GFP_KERNEL);
2521         if (!fs_info->delayed_root) {
2522                 err = -ENOMEM;
2523                 goto fail_iput;
2524         }
2525         btrfs_init_delayed_root(fs_info->delayed_root);
2526
2527         btrfs_init_scrub(fs_info);
2528 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2529         fs_info->check_integrity_print_mask = 0;
2530 #endif
2531         btrfs_init_balance(fs_info);
2532         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2533
2534         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2535         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2536
2537         btrfs_init_btree_inode(fs_info);
2538
2539         spin_lock_init(&fs_info->block_group_cache_lock);
2540         fs_info->block_group_cache_tree = RB_ROOT;
2541         fs_info->first_logical_byte = (u64)-1;
2542
2543         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2544         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2545         fs_info->pinned_extents = &fs_info->freed_extents[0];
2546         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2547
2548         mutex_init(&fs_info->ordered_operations_mutex);
2549         mutex_init(&fs_info->tree_log_mutex);
2550         mutex_init(&fs_info->chunk_mutex);
2551         mutex_init(&fs_info->transaction_kthread_mutex);
2552         mutex_init(&fs_info->cleaner_mutex);
2553         mutex_init(&fs_info->volume_mutex);
2554         mutex_init(&fs_info->ro_block_group_mutex);
2555         init_rwsem(&fs_info->commit_root_sem);
2556         init_rwsem(&fs_info->cleanup_work_sem);
2557         init_rwsem(&fs_info->subvol_sem);
2558         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2559
2560         btrfs_init_dev_replace_locks(fs_info);
2561         btrfs_init_qgroup(fs_info);
2562
2563         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2564         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2565
2566         init_waitqueue_head(&fs_info->transaction_throttle);
2567         init_waitqueue_head(&fs_info->transaction_wait);
2568         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2569         init_waitqueue_head(&fs_info->async_submit_wait);
2570
2571         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2572
2573         /* Usable values until the real ones are cached from the superblock */
2574         fs_info->nodesize = 4096;
2575         fs_info->sectorsize = 4096;
2576         fs_info->stripesize = 4096;
2577
2578         ret = btrfs_alloc_stripe_hash_table(fs_info);
2579         if (ret) {
2580                 err = ret;
2581                 goto fail_alloc;
2582         }
2583
2584         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2585
2586         invalidate_bdev(fs_devices->latest_bdev);
2587
2588         /*
2589          * Read super block and check the signature bytes only
2590          */
2591         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2592         if (IS_ERR(bh)) {
2593                 err = PTR_ERR(bh);
2594                 goto fail_alloc;
2595         }
2596
2597         /*
2598          * We want to check superblock checksum, the type is stored inside.
2599          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2600          */
2601         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2602                 btrfs_err(fs_info, "superblock checksum mismatch");
2603                 err = -EINVAL;
2604                 brelse(bh);
2605                 goto fail_alloc;
2606         }
2607
2608         /*
2609          * super_copy is zeroed at allocation time and we never touch the
2610          * following bytes up to INFO_SIZE, the checksum is calculated from
2611          * the whole block of INFO_SIZE
2612          */
2613         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2614         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2615                sizeof(*fs_info->super_for_commit));
2616         brelse(bh);
2617
2618         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2619
2620         ret = btrfs_check_super_valid(fs_info);
2621         if (ret) {
2622                 btrfs_err(fs_info, "superblock contains fatal errors");
2623                 err = -EINVAL;
2624                 goto fail_alloc;
2625         }
2626
2627         disk_super = fs_info->super_copy;
2628         if (!btrfs_super_root(disk_super))
2629                 goto fail_alloc;
2630
2631         /* check FS state, whether FS is broken. */
2632         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2633                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2634
2635         /*
2636          * run through our array of backup supers and setup
2637          * our ring pointer to the oldest one
2638          */
2639         generation = btrfs_super_generation(disk_super);
2640         find_oldest_super_backup(fs_info, generation);
2641
2642         /*
2643          * In the long term, we'll store the compression type in the super
2644          * block, and it'll be used for per file compression control.
2645          */
2646         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2647
2648         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2649         if (ret) {
2650                 err = ret;
2651                 goto fail_alloc;
2652         }
2653
2654         features = btrfs_super_incompat_flags(disk_super) &
2655                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2656         if (features) {
2657                 btrfs_err(fs_info,
2658                     "cannot mount because of unsupported optional features (%llx)",
2659                     features);
2660                 err = -EINVAL;
2661                 goto fail_alloc;
2662         }
2663
2664         features = btrfs_super_incompat_flags(disk_super);
2665         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2666         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2667                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2668         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2669                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2670
2671         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2672                 btrfs_info(fs_info, "has skinny extents");
2673
2674         /*
2675          * flag our filesystem as having big metadata blocks if
2676          * they are bigger than the page size
2677          */
2678         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2679                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2680                         btrfs_info(fs_info,
2681                                 "flagging fs with big metadata feature");
2682                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2683         }
2684
2685         nodesize = btrfs_super_nodesize(disk_super);
2686         sectorsize = btrfs_super_sectorsize(disk_super);
2687         stripesize = sectorsize;
2688         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2689         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2690
2691         /* Cache block sizes */
2692         fs_info->nodesize = nodesize;
2693         fs_info->sectorsize = sectorsize;
2694         fs_info->stripesize = stripesize;
2695
2696         /*
2697          * mixed block groups end up with duplicate but slightly offset
2698          * extent buffers for the same range.  It leads to corruptions
2699          */
2700         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2701             (sectorsize != nodesize)) {
2702                 btrfs_err(fs_info,
2703 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2704                         nodesize, sectorsize);
2705                 goto fail_alloc;
2706         }
2707
2708         /*
2709          * Needn't use the lock because there is no other task which will
2710          * update the flag.
2711          */
2712         btrfs_set_super_incompat_flags(disk_super, features);
2713
2714         features = btrfs_super_compat_ro_flags(disk_super) &
2715                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2716         if (!sb_rdonly(sb) && features) {
2717                 btrfs_err(fs_info,
2718         "cannot mount read-write because of unsupported optional features (%llx)",
2719                        features);
2720                 err = -EINVAL;
2721                 goto fail_alloc;
2722         }
2723
2724         max_active = fs_info->thread_pool_size;
2725
2726         ret = btrfs_init_workqueues(fs_info, fs_devices);
2727         if (ret) {
2728                 err = ret;
2729                 goto fail_sb_buffer;
2730         }
2731
2732         sb->s_bdi->congested_fn = btrfs_congested_fn;
2733         sb->s_bdi->congested_data = fs_info;
2734         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2735         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2736         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2737         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2738
2739         sb->s_blocksize = sectorsize;
2740         sb->s_blocksize_bits = blksize_bits(sectorsize);
2741         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2742
2743         mutex_lock(&fs_info->chunk_mutex);
2744         ret = btrfs_read_sys_array(fs_info);
2745         mutex_unlock(&fs_info->chunk_mutex);
2746         if (ret) {
2747                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2748                 goto fail_sb_buffer;
2749         }
2750
2751         generation = btrfs_super_chunk_root_generation(disk_super);
2752
2753         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2754
2755         chunk_root->node = read_tree_block(fs_info,
2756                                            btrfs_super_chunk_root(disk_super),
2757                                            generation);
2758         if (IS_ERR(chunk_root->node) ||
2759             !extent_buffer_uptodate(chunk_root->node)) {
2760                 btrfs_err(fs_info, "failed to read chunk root");
2761                 if (!IS_ERR(chunk_root->node))
2762                         free_extent_buffer(chunk_root->node);
2763                 chunk_root->node = NULL;
2764                 goto fail_tree_roots;
2765         }
2766         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2767         chunk_root->commit_root = btrfs_root_node(chunk_root);
2768
2769         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2770            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2771
2772         ret = btrfs_read_chunk_tree(fs_info);
2773         if (ret) {
2774                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2775                 goto fail_tree_roots;
2776         }
2777
2778         /*
2779          * keep the device that is marked to be the target device for the
2780          * dev_replace procedure
2781          */
2782         btrfs_close_extra_devices(fs_devices, 0);
2783
2784         if (!fs_devices->latest_bdev) {
2785                 btrfs_err(fs_info, "failed to read devices");
2786                 goto fail_tree_roots;
2787         }
2788
2789 retry_root_backup:
2790         generation = btrfs_super_generation(disk_super);
2791
2792         tree_root->node = read_tree_block(fs_info,
2793                                           btrfs_super_root(disk_super),
2794                                           generation);
2795         if (IS_ERR(tree_root->node) ||
2796             !extent_buffer_uptodate(tree_root->node)) {
2797                 btrfs_warn(fs_info, "failed to read tree root");
2798                 if (!IS_ERR(tree_root->node))
2799                         free_extent_buffer(tree_root->node);
2800                 tree_root->node = NULL;
2801                 goto recovery_tree_root;
2802         }
2803
2804         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2805         tree_root->commit_root = btrfs_root_node(tree_root);
2806         btrfs_set_root_refs(&tree_root->root_item, 1);
2807
2808         mutex_lock(&tree_root->objectid_mutex);
2809         ret = btrfs_find_highest_objectid(tree_root,
2810                                         &tree_root->highest_objectid);
2811         if (ret) {
2812                 mutex_unlock(&tree_root->objectid_mutex);
2813                 goto recovery_tree_root;
2814         }
2815
2816         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2817
2818         mutex_unlock(&tree_root->objectid_mutex);
2819
2820         ret = btrfs_read_roots(fs_info);
2821         if (ret)
2822                 goto recovery_tree_root;
2823
2824         fs_info->generation = generation;
2825         fs_info->last_trans_committed = generation;
2826
2827         ret = btrfs_recover_balance(fs_info);
2828         if (ret) {
2829                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2830                 goto fail_block_groups;
2831         }
2832
2833         ret = btrfs_init_dev_stats(fs_info);
2834         if (ret) {
2835                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2836                 goto fail_block_groups;
2837         }
2838
2839         ret = btrfs_init_dev_replace(fs_info);
2840         if (ret) {
2841                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2842                 goto fail_block_groups;
2843         }
2844
2845         btrfs_close_extra_devices(fs_devices, 1);
2846
2847         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2848         if (ret) {
2849                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2850                                 ret);
2851                 goto fail_block_groups;
2852         }
2853
2854         ret = btrfs_sysfs_add_device(fs_devices);
2855         if (ret) {
2856                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2857                                 ret);
2858                 goto fail_fsdev_sysfs;
2859         }
2860
2861         ret = btrfs_sysfs_add_mounted(fs_info);
2862         if (ret) {
2863                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2864                 goto fail_fsdev_sysfs;
2865         }
2866
2867         ret = btrfs_init_space_info(fs_info);
2868         if (ret) {
2869                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2870                 goto fail_sysfs;
2871         }
2872
2873         ret = btrfs_read_block_groups(fs_info);
2874         if (ret) {
2875                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2876                 goto fail_sysfs;
2877         }
2878
2879         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info)) {
2880                 btrfs_warn(fs_info,
2881                 "writeable mount is not allowed due to too many missing devices");
2882                 goto fail_sysfs;
2883         }
2884
2885         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2886                                                "btrfs-cleaner");
2887         if (IS_ERR(fs_info->cleaner_kthread))
2888                 goto fail_sysfs;
2889
2890         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2891                                                    tree_root,
2892                                                    "btrfs-transaction");
2893         if (IS_ERR(fs_info->transaction_kthread))
2894                 goto fail_cleaner;
2895
2896         if (!btrfs_test_opt(fs_info, NOSSD) &&
2897             !fs_info->fs_devices->rotating) {
2898                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2899         }
2900
2901         /*
2902          * Mount does not set all options immediately, we can do it now and do
2903          * not have to wait for transaction commit
2904          */
2905         btrfs_apply_pending_changes(fs_info);
2906
2907 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2908         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2909                 ret = btrfsic_mount(fs_info, fs_devices,
2910                                     btrfs_test_opt(fs_info,
2911                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2912                                     1 : 0,
2913                                     fs_info->check_integrity_print_mask);
2914                 if (ret)
2915                         btrfs_warn(fs_info,
2916                                 "failed to initialize integrity check module: %d",
2917                                 ret);
2918         }
2919 #endif
2920         ret = btrfs_read_qgroup_config(fs_info);
2921         if (ret)
2922                 goto fail_trans_kthread;
2923
2924         if (btrfs_build_ref_tree(fs_info))
2925                 btrfs_err(fs_info, "couldn't build ref tree");
2926
2927         /* do not make disk changes in broken FS or nologreplay is given */
2928         if (btrfs_super_log_root(disk_super) != 0 &&
2929             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2930                 ret = btrfs_replay_log(fs_info, fs_devices);
2931                 if (ret) {
2932                         err = ret;
2933                         goto fail_qgroup;
2934                 }
2935         }
2936
2937         ret = btrfs_find_orphan_roots(fs_info);
2938         if (ret)
2939                 goto fail_qgroup;
2940
2941         if (!sb_rdonly(sb)) {
2942                 ret = btrfs_cleanup_fs_roots(fs_info);
2943                 if (ret)
2944                         goto fail_qgroup;
2945
2946                 mutex_lock(&fs_info->cleaner_mutex);
2947                 ret = btrfs_recover_relocation(tree_root);
2948                 mutex_unlock(&fs_info->cleaner_mutex);
2949                 if (ret < 0) {
2950                         btrfs_warn(fs_info, "failed to recover relocation: %d",
2951                                         ret);
2952                         err = -EINVAL;
2953                         goto fail_qgroup;
2954                 }
2955         }
2956
2957         location.objectid = BTRFS_FS_TREE_OBJECTID;
2958         location.type = BTRFS_ROOT_ITEM_KEY;
2959         location.offset = 0;
2960
2961         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2962         if (IS_ERR(fs_info->fs_root)) {
2963                 err = PTR_ERR(fs_info->fs_root);
2964                 goto fail_qgroup;
2965         }
2966
2967         if (sb_rdonly(sb))
2968                 return 0;
2969
2970         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
2971             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2972                 clear_free_space_tree = 1;
2973         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
2974                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
2975                 btrfs_warn(fs_info, "free space tree is invalid");
2976                 clear_free_space_tree = 1;
2977         }
2978
2979         if (clear_free_space_tree) {
2980                 btrfs_info(fs_info, "clearing free space tree");
2981                 ret = btrfs_clear_free_space_tree(fs_info);
2982                 if (ret) {
2983                         btrfs_warn(fs_info,
2984                                    "failed to clear free space tree: %d", ret);
2985                         close_ctree(fs_info);
2986                         return ret;
2987                 }
2988         }
2989
2990         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
2991             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2992                 btrfs_info(fs_info, "creating free space tree");
2993                 ret = btrfs_create_free_space_tree(fs_info);
2994                 if (ret) {
2995                         btrfs_warn(fs_info,
2996                                 "failed to create free space tree: %d", ret);
2997                         close_ctree(fs_info);
2998                         return ret;
2999                 }
3000         }
3001
3002         down_read(&fs_info->cleanup_work_sem);
3003         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3004             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3005                 up_read(&fs_info->cleanup_work_sem);
3006                 close_ctree(fs_info);
3007                 return ret;
3008         }
3009         up_read(&fs_info->cleanup_work_sem);
3010
3011         ret = btrfs_resume_balance_async(fs_info);
3012         if (ret) {
3013                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3014                 close_ctree(fs_info);
3015                 return ret;
3016         }
3017
3018         ret = btrfs_resume_dev_replace_async(fs_info);
3019         if (ret) {
3020                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3021                 close_ctree(fs_info);
3022                 return ret;
3023         }
3024
3025         btrfs_qgroup_rescan_resume(fs_info);
3026
3027         if (!fs_info->uuid_root) {
3028                 btrfs_info(fs_info, "creating UUID tree");
3029                 ret = btrfs_create_uuid_tree(fs_info);
3030                 if (ret) {
3031                         btrfs_warn(fs_info,
3032                                 "failed to create the UUID tree: %d", ret);
3033                         close_ctree(fs_info);
3034                         return ret;
3035                 }
3036         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3037                    fs_info->generation !=
3038                                 btrfs_super_uuid_tree_generation(disk_super)) {
3039                 btrfs_info(fs_info, "checking UUID tree");
3040                 ret = btrfs_check_uuid_tree(fs_info);
3041                 if (ret) {
3042                         btrfs_warn(fs_info,
3043                                 "failed to check the UUID tree: %d", ret);
3044                         close_ctree(fs_info);
3045                         return ret;
3046                 }
3047         } else {
3048                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3049         }
3050         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3051
3052         /*
3053          * backuproot only affect mount behavior, and if open_ctree succeeded,
3054          * no need to keep the flag
3055          */
3056         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3057
3058         return 0;
3059
3060 fail_qgroup:
3061         btrfs_free_qgroup_config(fs_info);
3062 fail_trans_kthread:
3063         kthread_stop(fs_info->transaction_kthread);
3064         btrfs_cleanup_transaction(fs_info);
3065         btrfs_free_fs_roots(fs_info);
3066 fail_cleaner:
3067         kthread_stop(fs_info->cleaner_kthread);
3068
3069         /*
3070          * make sure we're done with the btree inode before we stop our
3071          * kthreads
3072          */
3073         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3074
3075 fail_sysfs:
3076         btrfs_sysfs_remove_mounted(fs_info);
3077
3078 fail_fsdev_sysfs:
3079         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3080
3081 fail_block_groups:
3082         btrfs_put_block_group_cache(fs_info);
3083
3084 fail_tree_roots:
3085         free_root_pointers(fs_info, 1);
3086         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3087
3088 fail_sb_buffer:
3089         btrfs_stop_all_workers(fs_info);
3090         btrfs_free_block_groups(fs_info);
3091 fail_alloc:
3092 fail_iput:
3093         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3094
3095         iput(fs_info->btree_inode);
3096 fail_bio_counter:
3097         percpu_counter_destroy(&fs_info->bio_counter);
3098 fail_delalloc_bytes:
3099         percpu_counter_destroy(&fs_info->delalloc_bytes);
3100 fail_dirty_metadata_bytes:
3101         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3102 fail_srcu:
3103         cleanup_srcu_struct(&fs_info->subvol_srcu);
3104 fail:
3105         btrfs_free_stripe_hash_table(fs_info);
3106         btrfs_close_devices(fs_info->fs_devices);
3107         return err;
3108
3109 recovery_tree_root:
3110         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3111                 goto fail_tree_roots;
3112
3113         free_root_pointers(fs_info, 0);
3114
3115         /* don't use the log in recovery mode, it won't be valid */
3116         btrfs_set_super_log_root(disk_super, 0);
3117
3118         /* we can't trust the free space cache either */
3119         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3120
3121         ret = next_root_backup(fs_info, fs_info->super_copy,
3122                                &num_backups_tried, &backup_index);
3123         if (ret == -1)
3124                 goto fail_block_groups;
3125         goto retry_root_backup;
3126 }
3127
3128 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3129 {
3130         if (uptodate) {
3131                 set_buffer_uptodate(bh);
3132         } else {
3133                 struct btrfs_device *device = (struct btrfs_device *)
3134                         bh->b_private;
3135
3136                 btrfs_warn_rl_in_rcu(device->fs_info,
3137                                 "lost page write due to IO error on %s",
3138                                           rcu_str_deref(device->name));
3139                 /* note, we don't set_buffer_write_io_error because we have
3140                  * our own ways of dealing with the IO errors
3141                  */
3142                 clear_buffer_uptodate(bh);
3143                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3144         }
3145         unlock_buffer(bh);
3146         put_bh(bh);
3147 }
3148
3149 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3150                         struct buffer_head **bh_ret)
3151 {
3152         struct buffer_head *bh;
3153         struct btrfs_super_block *super;
3154         u64 bytenr;
3155
3156         bytenr = btrfs_sb_offset(copy_num);
3157         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3158                 return -EINVAL;
3159
3160         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3161         /*
3162          * If we fail to read from the underlying devices, as of now
3163          * the best option we have is to mark it EIO.
3164          */
3165         if (!bh)
3166                 return -EIO;
3167
3168         super = (struct btrfs_super_block *)bh->b_data;
3169         if (btrfs_super_bytenr(super) != bytenr ||
3170                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3171                 brelse(bh);
3172                 return -EINVAL;
3173         }
3174
3175         *bh_ret = bh;
3176         return 0;
3177 }
3178
3179
3180 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3181 {
3182         struct buffer_head *bh;
3183         struct buffer_head *latest = NULL;
3184         struct btrfs_super_block *super;
3185         int i;
3186         u64 transid = 0;
3187         int ret = -EINVAL;
3188
3189         /* we would like to check all the supers, but that would make
3190          * a btrfs mount succeed after a mkfs from a different FS.
3191          * So, we need to add a special mount option to scan for
3192          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3193          */
3194         for (i = 0; i < 1; i++) {
3195                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3196                 if (ret)
3197                         continue;
3198
3199                 super = (struct btrfs_super_block *)bh->b_data;
3200
3201                 if (!latest || btrfs_super_generation(super) > transid) {
3202                         brelse(latest);
3203                         latest = bh;
3204                         transid = btrfs_super_generation(super);
3205                 } else {
3206                         brelse(bh);
3207                 }
3208         }
3209
3210         if (!latest)
3211                 return ERR_PTR(ret);
3212
3213         return latest;
3214 }
3215
3216 /*
3217  * Write superblock @sb to the @device. Do not wait for completion, all the
3218  * buffer heads we write are pinned.
3219  *
3220  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3221  * the expected device size at commit time. Note that max_mirrors must be
3222  * same for write and wait phases.
3223  *
3224  * Return number of errors when buffer head is not found or submission fails.
3225  */
3226 static int write_dev_supers(struct btrfs_device *device,
3227                             struct btrfs_super_block *sb, int max_mirrors)
3228 {
3229         struct buffer_head *bh;
3230         int i;
3231         int ret;
3232         int errors = 0;
3233         u32 crc;
3234         u64 bytenr;
3235         int op_flags;
3236
3237         if (max_mirrors == 0)
3238                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3239
3240         for (i = 0; i < max_mirrors; i++) {
3241                 bytenr = btrfs_sb_offset(i);
3242                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3243                     device->commit_total_bytes)
3244                         break;
3245
3246                 btrfs_set_super_bytenr(sb, bytenr);
3247
3248                 crc = ~(u32)0;
3249                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3250                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3251                 btrfs_csum_final(crc, sb->csum);
3252
3253                 /* One reference for us, and we leave it for the caller */
3254                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3255                               BTRFS_SUPER_INFO_SIZE);
3256                 if (!bh) {
3257                         btrfs_err(device->fs_info,
3258                             "couldn't get super buffer head for bytenr %llu",
3259                             bytenr);
3260                         errors++;
3261                         continue;
3262                 }
3263
3264                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3265
3266                 /* one reference for submit_bh */
3267                 get_bh(bh);
3268
3269                 set_buffer_uptodate(bh);
3270                 lock_buffer(bh);
3271                 bh->b_end_io = btrfs_end_buffer_write_sync;
3272                 bh->b_private = device;
3273
3274                 /*
3275                  * we fua the first super.  The others we allow
3276                  * to go down lazy.