btrfs: relocation: Delay reloc tree deletion after merge_reloc_roots
[sfrench/cifs-2.6.git] / fs / btrfs / disk-io.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5
6 #include <linux/fs.h>
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/buffer_head.h>
11 #include <linux/workqueue.h>
12 #include <linux/kthread.h>
13 #include <linux/slab.h>
14 #include <linux/migrate.h>
15 #include <linux/ratelimit.h>
16 #include <linux/uuid.h>
17 #include <linux/semaphore.h>
18 #include <linux/error-injection.h>
19 #include <linux/crc32c.h>
20 #include <linux/sched/mm.h>
21 #include <asm/unaligned.h>
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "transaction.h"
25 #include "btrfs_inode.h"
26 #include "volumes.h"
27 #include "print-tree.h"
28 #include "locking.h"
29 #include "tree-log.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
36 #include "raid56.h"
37 #include "sysfs.h"
38 #include "qgroup.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42
43 #ifdef CONFIG_X86
44 #include <asm/cpufeature.h>
45 #endif
46
47 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
48                                  BTRFS_HEADER_FLAG_RELOC |\
49                                  BTRFS_SUPER_FLAG_ERROR |\
50                                  BTRFS_SUPER_FLAG_SEEDING |\
51                                  BTRFS_SUPER_FLAG_METADUMP |\
52                                  BTRFS_SUPER_FLAG_METADUMP_V2)
53
54 static const struct extent_io_ops btree_extent_io_ops;
55 static void end_workqueue_fn(struct btrfs_work *work);
56 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
57 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
58                                       struct btrfs_fs_info *fs_info);
59 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
61                                         struct extent_io_tree *dirty_pages,
62                                         int mark);
63 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
64                                        struct extent_io_tree *pinned_extents);
65 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
66 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67
68 /*
69  * btrfs_end_io_wq structs are used to do processing in task context when an IO
70  * is complete.  This is used during reads to verify checksums, and it is used
71  * by writes to insert metadata for new file extents after IO is complete.
72  */
73 struct btrfs_end_io_wq {
74         struct bio *bio;
75         bio_end_io_t *end_io;
76         void *private;
77         struct btrfs_fs_info *info;
78         blk_status_t status;
79         enum btrfs_wq_endio_type metadata;
80         struct btrfs_work work;
81 };
82
83 static struct kmem_cache *btrfs_end_io_wq_cache;
84
85 int __init btrfs_end_io_wq_init(void)
86 {
87         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
88                                         sizeof(struct btrfs_end_io_wq),
89                                         0,
90                                         SLAB_MEM_SPREAD,
91                                         NULL);
92         if (!btrfs_end_io_wq_cache)
93                 return -ENOMEM;
94         return 0;
95 }
96
97 void __cold btrfs_end_io_wq_exit(void)
98 {
99         kmem_cache_destroy(btrfs_end_io_wq_cache);
100 }
101
102 /*
103  * async submit bios are used to offload expensive checksumming
104  * onto the worker threads.  They checksum file and metadata bios
105  * just before they are sent down the IO stack.
106  */
107 struct async_submit_bio {
108         void *private_data;
109         struct bio *bio;
110         extent_submit_bio_start_t *submit_bio_start;
111         int mirror_num;
112         /*
113          * bio_offset is optional, can be used if the pages in the bio
114          * can't tell us where in the file the bio should go
115          */
116         u64 bio_offset;
117         struct btrfs_work work;
118         blk_status_t status;
119 };
120
121 /*
122  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
123  * eb, the lockdep key is determined by the btrfs_root it belongs to and
124  * the level the eb occupies in the tree.
125  *
126  * Different roots are used for different purposes and may nest inside each
127  * other and they require separate keysets.  As lockdep keys should be
128  * static, assign keysets according to the purpose of the root as indicated
129  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
130  * roots have separate keysets.
131  *
132  * Lock-nesting across peer nodes is always done with the immediate parent
133  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
134  * subclass to avoid triggering lockdep warning in such cases.
135  *
136  * The key is set by the readpage_end_io_hook after the buffer has passed
137  * csum validation but before the pages are unlocked.  It is also set by
138  * btrfs_init_new_buffer on freshly allocated blocks.
139  *
140  * We also add a check to make sure the highest level of the tree is the
141  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
142  * needs update as well.
143  */
144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
145 # if BTRFS_MAX_LEVEL != 8
146 #  error
147 # endif
148
149 static struct btrfs_lockdep_keyset {
150         u64                     id;             /* root objectid */
151         const char              *name_stem;     /* lock name stem */
152         char                    names[BTRFS_MAX_LEVEL + 1][20];
153         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
154 } btrfs_lockdep_keysets[] = {
155         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
156         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
157         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
158         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
159         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
160         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
161         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
162         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
163         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
164         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
165         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
166         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
167         { .id = 0,                              .name_stem = "tree"     },
168 };
169
170 void __init btrfs_init_lockdep(void)
171 {
172         int i, j;
173
174         /* initialize lockdep class names */
175         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
176                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
177
178                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
179                         snprintf(ks->names[j], sizeof(ks->names[j]),
180                                  "btrfs-%s-%02d", ks->name_stem, j);
181         }
182 }
183
184 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
185                                     int level)
186 {
187         struct btrfs_lockdep_keyset *ks;
188
189         BUG_ON(level >= ARRAY_SIZE(ks->keys));
190
191         /* find the matching keyset, id 0 is the default entry */
192         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
193                 if (ks->id == objectid)
194                         break;
195
196         lockdep_set_class_and_name(&eb->lock,
197                                    &ks->keys[level], ks->names[level]);
198 }
199
200 #endif
201
202 /*
203  * extents on the btree inode are pretty simple, there's one extent
204  * that covers the entire device
205  */
206 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
207                 struct page *page, size_t pg_offset, u64 start, u64 len,
208                 int create)
209 {
210         struct btrfs_fs_info *fs_info = inode->root->fs_info;
211         struct extent_map_tree *em_tree = &inode->extent_tree;
212         struct extent_map *em;
213         int ret;
214
215         read_lock(&em_tree->lock);
216         em = lookup_extent_mapping(em_tree, start, len);
217         if (em) {
218                 em->bdev = fs_info->fs_devices->latest_bdev;
219                 read_unlock(&em_tree->lock);
220                 goto out;
221         }
222         read_unlock(&em_tree->lock);
223
224         em = alloc_extent_map();
225         if (!em) {
226                 em = ERR_PTR(-ENOMEM);
227                 goto out;
228         }
229         em->start = 0;
230         em->len = (u64)-1;
231         em->block_len = (u64)-1;
232         em->block_start = 0;
233         em->bdev = fs_info->fs_devices->latest_bdev;
234
235         write_lock(&em_tree->lock);
236         ret = add_extent_mapping(em_tree, em, 0);
237         if (ret == -EEXIST) {
238                 free_extent_map(em);
239                 em = lookup_extent_mapping(em_tree, start, len);
240                 if (!em)
241                         em = ERR_PTR(-EIO);
242         } else if (ret) {
243                 free_extent_map(em);
244                 em = ERR_PTR(ret);
245         }
246         write_unlock(&em_tree->lock);
247
248 out:
249         return em;
250 }
251
252 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
253 {
254         return crc32c(seed, data, len);
255 }
256
257 void btrfs_csum_final(u32 crc, u8 *result)
258 {
259         put_unaligned_le32(~crc, result);
260 }
261
262 /*
263  * compute the csum for a btree block, and either verify it or write it
264  * into the csum field of the block.
265  */
266 static int csum_tree_block(struct btrfs_fs_info *fs_info,
267                            struct extent_buffer *buf,
268                            int verify)
269 {
270         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
271         char result[BTRFS_CSUM_SIZE];
272         unsigned long len;
273         unsigned long cur_len;
274         unsigned long offset = BTRFS_CSUM_SIZE;
275         char *kaddr;
276         unsigned long map_start;
277         unsigned long map_len;
278         int err;
279         u32 crc = ~(u32)0;
280
281         len = buf->len - offset;
282         while (len > 0) {
283                 /*
284                  * Note: we don't need to check for the err == 1 case here, as
285                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
286                  * and 'min_len = 32' and the currently implemented mapping
287                  * algorithm we cannot cross a page boundary.
288                  */
289                 err = map_private_extent_buffer(buf, offset, 32,
290                                         &kaddr, &map_start, &map_len);
291                 if (err)
292                         return err;
293                 cur_len = min(len, map_len - (offset - map_start));
294                 crc = btrfs_csum_data(kaddr + offset - map_start,
295                                       crc, cur_len);
296                 len -= cur_len;
297                 offset += cur_len;
298         }
299         memset(result, 0, BTRFS_CSUM_SIZE);
300
301         btrfs_csum_final(crc, result);
302
303         if (verify) {
304                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
305                         u32 val;
306                         u32 found = 0;
307                         memcpy(&found, result, csum_size);
308
309                         read_extent_buffer(buf, &val, 0, csum_size);
310                         btrfs_warn_rl(fs_info,
311                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
312                                 fs_info->sb->s_id, buf->start,
313                                 val, found, btrfs_header_level(buf));
314                         return -EUCLEAN;
315                 }
316         } else {
317                 write_extent_buffer(buf, result, 0, csum_size);
318         }
319
320         return 0;
321 }
322
323 /*
324  * we can't consider a given block up to date unless the transid of the
325  * block matches the transid in the parent node's pointer.  This is how we
326  * detect blocks that either didn't get written at all or got written
327  * in the wrong place.
328  */
329 static int verify_parent_transid(struct extent_io_tree *io_tree,
330                                  struct extent_buffer *eb, u64 parent_transid,
331                                  int atomic)
332 {
333         struct extent_state *cached_state = NULL;
334         int ret;
335         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336
337         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
338                 return 0;
339
340         if (atomic)
341                 return -EAGAIN;
342
343         if (need_lock) {
344                 btrfs_tree_read_lock(eb);
345                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
346         }
347
348         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349                          &cached_state);
350         if (extent_buffer_uptodate(eb) &&
351             btrfs_header_generation(eb) == parent_transid) {
352                 ret = 0;
353                 goto out;
354         }
355         btrfs_err_rl(eb->fs_info,
356                 "parent transid verify failed on %llu wanted %llu found %llu",
357                         eb->start,
358                         parent_transid, btrfs_header_generation(eb));
359         ret = 1;
360
361         /*
362          * Things reading via commit roots that don't have normal protection,
363          * like send, can have a really old block in cache that may point at a
364          * block that has been freed and re-allocated.  So don't clear uptodate
365          * if we find an eb that is under IO (dirty/writeback) because we could
366          * end up reading in the stale data and then writing it back out and
367          * making everybody very sad.
368          */
369         if (!extent_buffer_under_io(eb))
370                 clear_extent_buffer_uptodate(eb);
371 out:
372         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
373                              &cached_state);
374         if (need_lock)
375                 btrfs_tree_read_unlock_blocking(eb);
376         return ret;
377 }
378
379 /*
380  * Return 0 if the superblock checksum type matches the checksum value of that
381  * algorithm. Pass the raw disk superblock data.
382  */
383 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
384                                   char *raw_disk_sb)
385 {
386         struct btrfs_super_block *disk_sb =
387                 (struct btrfs_super_block *)raw_disk_sb;
388         u16 csum_type = btrfs_super_csum_type(disk_sb);
389         int ret = 0;
390
391         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392                 u32 crc = ~(u32)0;
393                 char result[sizeof(crc)];
394
395                 /*
396                  * The super_block structure does not span the whole
397                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
398                  * is filled with zeros and is included in the checksum.
399                  */
400                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
401                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
402                 btrfs_csum_final(crc, result);
403
404                 if (memcmp(raw_disk_sb, result, sizeof(result)))
405                         ret = 1;
406         }
407
408         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
409                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
410                                 csum_type);
411                 ret = 1;
412         }
413
414         return ret;
415 }
416
417 static int verify_level_key(struct btrfs_fs_info *fs_info,
418                             struct extent_buffer *eb, int level,
419                             struct btrfs_key *first_key, u64 parent_transid)
420 {
421         int found_level;
422         struct btrfs_key found_key;
423         int ret;
424
425         found_level = btrfs_header_level(eb);
426         if (found_level != level) {
427 #ifdef CONFIG_BTRFS_DEBUG
428                 WARN_ON(1);
429                 btrfs_err(fs_info,
430 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
431                           eb->start, level, found_level);
432 #endif
433                 return -EIO;
434         }
435
436         if (!first_key)
437                 return 0;
438
439         /*
440          * For live tree block (new tree blocks in current transaction),
441          * we need proper lock context to avoid race, which is impossible here.
442          * So we only checks tree blocks which is read from disk, whose
443          * generation <= fs_info->last_trans_committed.
444          */
445         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
446                 return 0;
447         if (found_level)
448                 btrfs_node_key_to_cpu(eb, &found_key, 0);
449         else
450                 btrfs_item_key_to_cpu(eb, &found_key, 0);
451         ret = btrfs_comp_cpu_keys(first_key, &found_key);
452
453 #ifdef CONFIG_BTRFS_DEBUG
454         if (ret) {
455                 WARN_ON(1);
456                 btrfs_err(fs_info,
457 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
458                           eb->start, parent_transid, first_key->objectid,
459                           first_key->type, first_key->offset,
460                           found_key.objectid, found_key.type,
461                           found_key.offset);
462         }
463 #endif
464         return ret;
465 }
466
467 /*
468  * helper to read a given tree block, doing retries as required when
469  * the checksums don't match and we have alternate mirrors to try.
470  *
471  * @parent_transid:     expected transid, skip check if 0
472  * @level:              expected level, mandatory check
473  * @first_key:          expected key of first slot, skip check if NULL
474  */
475 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
476                                           struct extent_buffer *eb,
477                                           u64 parent_transid, int level,
478                                           struct btrfs_key *first_key)
479 {
480         struct extent_io_tree *io_tree;
481         int failed = 0;
482         int ret;
483         int num_copies = 0;
484         int mirror_num = 0;
485         int failed_mirror = 0;
486
487         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488         while (1) {
489                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
490                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
491                                                mirror_num);
492                 if (!ret) {
493                         if (verify_parent_transid(io_tree, eb,
494                                                    parent_transid, 0))
495                                 ret = -EIO;
496                         else if (verify_level_key(fs_info, eb, level,
497                                                   first_key, parent_transid))
498                                 ret = -EUCLEAN;
499                         else
500                                 break;
501                 }
502
503                 num_copies = btrfs_num_copies(fs_info,
504                                               eb->start, eb->len);
505                 if (num_copies == 1)
506                         break;
507
508                 if (!failed_mirror) {
509                         failed = 1;
510                         failed_mirror = eb->read_mirror;
511                 }
512
513                 mirror_num++;
514                 if (mirror_num == failed_mirror)
515                         mirror_num++;
516
517                 if (mirror_num > num_copies)
518                         break;
519         }
520
521         if (failed && !ret && failed_mirror)
522                 repair_eb_io_failure(fs_info, eb, failed_mirror);
523
524         return ret;
525 }
526
527 /*
528  * checksum a dirty tree block before IO.  This has extra checks to make sure
529  * we only fill in the checksum field in the first page of a multi-page block
530  */
531
532 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
533 {
534         u64 start = page_offset(page);
535         u64 found_start;
536         struct extent_buffer *eb;
537
538         eb = (struct extent_buffer *)page->private;
539         if (page != eb->pages[0])
540                 return 0;
541
542         found_start = btrfs_header_bytenr(eb);
543         /*
544          * Please do not consolidate these warnings into a single if.
545          * It is useful to know what went wrong.
546          */
547         if (WARN_ON(found_start != start))
548                 return -EUCLEAN;
549         if (WARN_ON(!PageUptodate(page)))
550                 return -EUCLEAN;
551
552         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
553                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
554
555         return csum_tree_block(fs_info, eb, 0);
556 }
557
558 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
559                                  struct extent_buffer *eb)
560 {
561         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562         u8 fsid[BTRFS_FSID_SIZE];
563         int ret = 1;
564
565         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
566         while (fs_devices) {
567                 u8 *metadata_uuid;
568
569                 /*
570                  * Checking the incompat flag is only valid for the current
571                  * fs. For seed devices it's forbidden to have their uuid
572                  * changed so reading ->fsid in this case is fine
573                  */
574                 if (fs_devices == fs_info->fs_devices &&
575                     btrfs_fs_incompat(fs_info, METADATA_UUID))
576                         metadata_uuid = fs_devices->metadata_uuid;
577                 else
578                         metadata_uuid = fs_devices->fsid;
579
580                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
581                         ret = 0;
582                         break;
583                 }
584                 fs_devices = fs_devices->seed;
585         }
586         return ret;
587 }
588
589 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
590                                       u64 phy_offset, struct page *page,
591                                       u64 start, u64 end, int mirror)
592 {
593         u64 found_start;
594         int found_level;
595         struct extent_buffer *eb;
596         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
597         struct btrfs_fs_info *fs_info = root->fs_info;
598         int ret = 0;
599         int reads_done;
600
601         if (!page->private)
602                 goto out;
603
604         eb = (struct extent_buffer *)page->private;
605
606         /* the pending IO might have been the only thing that kept this buffer
607          * in memory.  Make sure we have a ref for all this other checks
608          */
609         extent_buffer_get(eb);
610
611         reads_done = atomic_dec_and_test(&eb->io_pages);
612         if (!reads_done)
613                 goto err;
614
615         eb->read_mirror = mirror;
616         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
617                 ret = -EIO;
618                 goto err;
619         }
620
621         found_start = btrfs_header_bytenr(eb);
622         if (found_start != eb->start) {
623                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
624                              eb->start, found_start);
625                 ret = -EIO;
626                 goto err;
627         }
628         if (check_tree_block_fsid(fs_info, eb)) {
629                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
630                              eb->start);
631                 ret = -EIO;
632                 goto err;
633         }
634         found_level = btrfs_header_level(eb);
635         if (found_level >= BTRFS_MAX_LEVEL) {
636                 btrfs_err(fs_info, "bad tree block level %d on %llu",
637                           (int)btrfs_header_level(eb), eb->start);
638                 ret = -EIO;
639                 goto err;
640         }
641
642         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
643                                        eb, found_level);
644
645         ret = csum_tree_block(fs_info, eb, 1);
646         if (ret)
647                 goto err;
648
649         /*
650          * If this is a leaf block and it is corrupt, set the corrupt bit so
651          * that we don't try and read the other copies of this block, just
652          * return -EIO.
653          */
654         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
655                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
656                 ret = -EIO;
657         }
658
659         if (found_level > 0 && btrfs_check_node(fs_info, eb))
660                 ret = -EIO;
661
662         if (!ret)
663                 set_extent_buffer_uptodate(eb);
664 err:
665         if (reads_done &&
666             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
667                 btree_readahead_hook(eb, ret);
668
669         if (ret) {
670                 /*
671                  * our io error hook is going to dec the io pages
672                  * again, we have to make sure it has something
673                  * to decrement
674                  */
675                 atomic_inc(&eb->io_pages);
676                 clear_extent_buffer_uptodate(eb);
677         }
678         free_extent_buffer(eb);
679 out:
680         return ret;
681 }
682
683 static void end_workqueue_bio(struct bio *bio)
684 {
685         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
686         struct btrfs_fs_info *fs_info;
687         struct btrfs_workqueue *wq;
688         btrfs_work_func_t func;
689
690         fs_info = end_io_wq->info;
691         end_io_wq->status = bio->bi_status;
692
693         if (bio_op(bio) == REQ_OP_WRITE) {
694                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
695                         wq = fs_info->endio_meta_write_workers;
696                         func = btrfs_endio_meta_write_helper;
697                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
698                         wq = fs_info->endio_freespace_worker;
699                         func = btrfs_freespace_write_helper;
700                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
701                         wq = fs_info->endio_raid56_workers;
702                         func = btrfs_endio_raid56_helper;
703                 } else {
704                         wq = fs_info->endio_write_workers;
705                         func = btrfs_endio_write_helper;
706                 }
707         } else {
708                 if (unlikely(end_io_wq->metadata ==
709                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
710                         wq = fs_info->endio_repair_workers;
711                         func = btrfs_endio_repair_helper;
712                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
713                         wq = fs_info->endio_raid56_workers;
714                         func = btrfs_endio_raid56_helper;
715                 } else if (end_io_wq->metadata) {
716                         wq = fs_info->endio_meta_workers;
717                         func = btrfs_endio_meta_helper;
718                 } else {
719                         wq = fs_info->endio_workers;
720                         func = btrfs_endio_helper;
721                 }
722         }
723
724         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
725         btrfs_queue_work(wq, &end_io_wq->work);
726 }
727
728 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
729                         enum btrfs_wq_endio_type metadata)
730 {
731         struct btrfs_end_io_wq *end_io_wq;
732
733         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
734         if (!end_io_wq)
735                 return BLK_STS_RESOURCE;
736
737         end_io_wq->private = bio->bi_private;
738         end_io_wq->end_io = bio->bi_end_io;
739         end_io_wq->info = info;
740         end_io_wq->status = 0;
741         end_io_wq->bio = bio;
742         end_io_wq->metadata = metadata;
743
744         bio->bi_private = end_io_wq;
745         bio->bi_end_io = end_workqueue_bio;
746         return 0;
747 }
748
749 static void run_one_async_start(struct btrfs_work *work)
750 {
751         struct async_submit_bio *async;
752         blk_status_t ret;
753
754         async = container_of(work, struct  async_submit_bio, work);
755         ret = async->submit_bio_start(async->private_data, async->bio,
756                                       async->bio_offset);
757         if (ret)
758                 async->status = ret;
759 }
760
761 /*
762  * In order to insert checksums into the metadata in large chunks, we wait
763  * until bio submission time.   All the pages in the bio are checksummed and
764  * sums are attached onto the ordered extent record.
765  *
766  * At IO completion time the csums attached on the ordered extent record are
767  * inserted into the tree.
768  */
769 static void run_one_async_done(struct btrfs_work *work)
770 {
771         struct async_submit_bio *async;
772         struct inode *inode;
773         blk_status_t ret;
774
775         async = container_of(work, struct  async_submit_bio, work);
776         inode = async->private_data;
777
778         /* If an error occurred we just want to clean up the bio and move on */
779         if (async->status) {
780                 async->bio->bi_status = async->status;
781                 bio_endio(async->bio);
782                 return;
783         }
784
785         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
786                         async->mirror_num, 1);
787         if (ret) {
788                 async->bio->bi_status = ret;
789                 bio_endio(async->bio);
790         }
791 }
792
793 static void run_one_async_free(struct btrfs_work *work)
794 {
795         struct async_submit_bio *async;
796
797         async = container_of(work, struct  async_submit_bio, work);
798         kfree(async);
799 }
800
801 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
802                                  int mirror_num, unsigned long bio_flags,
803                                  u64 bio_offset, void *private_data,
804                                  extent_submit_bio_start_t *submit_bio_start)
805 {
806         struct async_submit_bio *async;
807
808         async = kmalloc(sizeof(*async), GFP_NOFS);
809         if (!async)
810                 return BLK_STS_RESOURCE;
811
812         async->private_data = private_data;
813         async->bio = bio;
814         async->mirror_num = mirror_num;
815         async->submit_bio_start = submit_bio_start;
816
817         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
818                         run_one_async_done, run_one_async_free);
819
820         async->bio_offset = bio_offset;
821
822         async->status = 0;
823
824         if (op_is_sync(bio->bi_opf))
825                 btrfs_set_work_high_priority(&async->work);
826
827         btrfs_queue_work(fs_info->workers, &async->work);
828         return 0;
829 }
830
831 static blk_status_t btree_csum_one_bio(struct bio *bio)
832 {
833         struct bio_vec *bvec;
834         struct btrfs_root *root;
835         int i, ret = 0;
836
837         ASSERT(!bio_flagged(bio, BIO_CLONED));
838         bio_for_each_segment_all(bvec, bio, i) {
839                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
840                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
841                 if (ret)
842                         break;
843         }
844
845         return errno_to_blk_status(ret);
846 }
847
848 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
849                                              u64 bio_offset)
850 {
851         /*
852          * when we're called for a write, we're already in the async
853          * submission context.  Just jump into btrfs_map_bio
854          */
855         return btree_csum_one_bio(bio);
856 }
857
858 static int check_async_write(struct btrfs_inode *bi)
859 {
860         if (atomic_read(&bi->sync_writers))
861                 return 0;
862 #ifdef CONFIG_X86
863         if (static_cpu_has(X86_FEATURE_XMM4_2))
864                 return 0;
865 #endif
866         return 1;
867 }
868
869 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
870                                           int mirror_num, unsigned long bio_flags,
871                                           u64 bio_offset)
872 {
873         struct inode *inode = private_data;
874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
875         int async = check_async_write(BTRFS_I(inode));
876         blk_status_t ret;
877
878         if (bio_op(bio) != REQ_OP_WRITE) {
879                 /*
880                  * called for a read, do the setup so that checksum validation
881                  * can happen in the async kernel threads
882                  */
883                 ret = btrfs_bio_wq_end_io(fs_info, bio,
884                                           BTRFS_WQ_ENDIO_METADATA);
885                 if (ret)
886                         goto out_w_error;
887                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
888         } else if (!async) {
889                 ret = btree_csum_one_bio(bio);
890                 if (ret)
891                         goto out_w_error;
892                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
893         } else {
894                 /*
895                  * kthread helpers are used to submit writes so that
896                  * checksumming can happen in parallel across all CPUs
897                  */
898                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
899                                           bio_offset, private_data,
900                                           btree_submit_bio_start);
901         }
902
903         if (ret)
904                 goto out_w_error;
905         return 0;
906
907 out_w_error:
908         bio->bi_status = ret;
909         bio_endio(bio);
910         return ret;
911 }
912
913 #ifdef CONFIG_MIGRATION
914 static int btree_migratepage(struct address_space *mapping,
915                         struct page *newpage, struct page *page,
916                         enum migrate_mode mode)
917 {
918         /*
919          * we can't safely write a btree page from here,
920          * we haven't done the locking hook
921          */
922         if (PageDirty(page))
923                 return -EAGAIN;
924         /*
925          * Buffers may be managed in a filesystem specific way.
926          * We must have no buffers or drop them.
927          */
928         if (page_has_private(page) &&
929             !try_to_release_page(page, GFP_KERNEL))
930                 return -EAGAIN;
931         return migrate_page(mapping, newpage, page, mode);
932 }
933 #endif
934
935
936 static int btree_writepages(struct address_space *mapping,
937                             struct writeback_control *wbc)
938 {
939         struct btrfs_fs_info *fs_info;
940         int ret;
941
942         if (wbc->sync_mode == WB_SYNC_NONE) {
943
944                 if (wbc->for_kupdate)
945                         return 0;
946
947                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
948                 /* this is a bit racy, but that's ok */
949                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
950                                              BTRFS_DIRTY_METADATA_THRESH,
951                                              fs_info->dirty_metadata_batch);
952                 if (ret < 0)
953                         return 0;
954         }
955         return btree_write_cache_pages(mapping, wbc);
956 }
957
958 static int btree_readpage(struct file *file, struct page *page)
959 {
960         struct extent_io_tree *tree;
961         tree = &BTRFS_I(page->mapping->host)->io_tree;
962         return extent_read_full_page(tree, page, btree_get_extent, 0);
963 }
964
965 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
966 {
967         if (PageWriteback(page) || PageDirty(page))
968                 return 0;
969
970         return try_release_extent_buffer(page);
971 }
972
973 static void btree_invalidatepage(struct page *page, unsigned int offset,
974                                  unsigned int length)
975 {
976         struct extent_io_tree *tree;
977         tree = &BTRFS_I(page->mapping->host)->io_tree;
978         extent_invalidatepage(tree, page, offset);
979         btree_releasepage(page, GFP_NOFS);
980         if (PagePrivate(page)) {
981                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
982                            "page private not zero on page %llu",
983                            (unsigned long long)page_offset(page));
984                 ClearPagePrivate(page);
985                 set_page_private(page, 0);
986                 put_page(page);
987         }
988 }
989
990 static int btree_set_page_dirty(struct page *page)
991 {
992 #ifdef DEBUG
993         struct extent_buffer *eb;
994
995         BUG_ON(!PagePrivate(page));
996         eb = (struct extent_buffer *)page->private;
997         BUG_ON(!eb);
998         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
999         BUG_ON(!atomic_read(&eb->refs));
1000         btrfs_assert_tree_locked(eb);
1001 #endif
1002         return __set_page_dirty_nobuffers(page);
1003 }
1004
1005 static const struct address_space_operations btree_aops = {
1006         .readpage       = btree_readpage,
1007         .writepages     = btree_writepages,
1008         .releasepage    = btree_releasepage,
1009         .invalidatepage = btree_invalidatepage,
1010 #ifdef CONFIG_MIGRATION
1011         .migratepage    = btree_migratepage,
1012 #endif
1013         .set_page_dirty = btree_set_page_dirty,
1014 };
1015
1016 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1017 {
1018         struct extent_buffer *buf = NULL;
1019         struct inode *btree_inode = fs_info->btree_inode;
1020
1021         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1022         if (IS_ERR(buf))
1023                 return;
1024         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1025                                  buf, WAIT_NONE, 0);
1026         free_extent_buffer(buf);
1027 }
1028
1029 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1030                          int mirror_num, struct extent_buffer **eb)
1031 {
1032         struct extent_buffer *buf = NULL;
1033         struct inode *btree_inode = fs_info->btree_inode;
1034         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1035         int ret;
1036
1037         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1038         if (IS_ERR(buf))
1039                 return 0;
1040
1041         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1042
1043         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1044                                        mirror_num);
1045         if (ret) {
1046                 free_extent_buffer(buf);
1047                 return ret;
1048         }
1049
1050         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1051                 free_extent_buffer(buf);
1052                 return -EIO;
1053         } else if (extent_buffer_uptodate(buf)) {
1054                 *eb = buf;
1055         } else {
1056                 free_extent_buffer(buf);
1057         }
1058         return 0;
1059 }
1060
1061 struct extent_buffer *btrfs_find_create_tree_block(
1062                                                 struct btrfs_fs_info *fs_info,
1063                                                 u64 bytenr)
1064 {
1065         if (btrfs_is_testing(fs_info))
1066                 return alloc_test_extent_buffer(fs_info, bytenr);
1067         return alloc_extent_buffer(fs_info, bytenr);
1068 }
1069
1070
1071 int btrfs_write_tree_block(struct extent_buffer *buf)
1072 {
1073         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1074                                         buf->start + buf->len - 1);
1075 }
1076
1077 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1078 {
1079         filemap_fdatawait_range(buf->pages[0]->mapping,
1080                                 buf->start, buf->start + buf->len - 1);
1081 }
1082
1083 /*
1084  * Read tree block at logical address @bytenr and do variant basic but critical
1085  * verification.
1086  *
1087  * @parent_transid:     expected transid of this tree block, skip check if 0
1088  * @level:              expected level, mandatory check
1089  * @first_key:          expected key in slot 0, skip check if NULL
1090  */
1091 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1092                                       u64 parent_transid, int level,
1093                                       struct btrfs_key *first_key)
1094 {
1095         struct extent_buffer *buf = NULL;
1096         int ret;
1097
1098         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1099         if (IS_ERR(buf))
1100                 return buf;
1101
1102         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1103                                              level, first_key);
1104         if (ret) {
1105                 free_extent_buffer(buf);
1106                 return ERR_PTR(ret);
1107         }
1108         return buf;
1109
1110 }
1111
1112 void clean_tree_block(struct btrfs_fs_info *fs_info,
1113                       struct extent_buffer *buf)
1114 {
1115         if (btrfs_header_generation(buf) ==
1116             fs_info->running_transaction->transid) {
1117                 btrfs_assert_tree_locked(buf);
1118
1119                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1120                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1121                                                  -buf->len,
1122                                                  fs_info->dirty_metadata_batch);
1123                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1124                         btrfs_set_lock_blocking(buf);
1125                         clear_extent_buffer_dirty(buf);
1126                 }
1127         }
1128 }
1129
1130 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1131 {
1132         struct btrfs_subvolume_writers *writers;
1133         int ret;
1134
1135         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1136         if (!writers)
1137                 return ERR_PTR(-ENOMEM);
1138
1139         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1140         if (ret < 0) {
1141                 kfree(writers);
1142                 return ERR_PTR(ret);
1143         }
1144
1145         init_waitqueue_head(&writers->wait);
1146         return writers;
1147 }
1148
1149 static void
1150 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1151 {
1152         percpu_counter_destroy(&writers->counter);
1153         kfree(writers);
1154 }
1155
1156 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1157                          u64 objectid)
1158 {
1159         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1160         root->node = NULL;
1161         root->commit_root = NULL;
1162         root->state = 0;
1163         root->orphan_cleanup_state = 0;
1164
1165         root->last_trans = 0;
1166         root->highest_objectid = 0;
1167         root->nr_delalloc_inodes = 0;
1168         root->nr_ordered_extents = 0;
1169         root->inode_tree = RB_ROOT;
1170         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1171         root->block_rsv = NULL;
1172
1173         INIT_LIST_HEAD(&root->dirty_list);
1174         INIT_LIST_HEAD(&root->root_list);
1175         INIT_LIST_HEAD(&root->delalloc_inodes);
1176         INIT_LIST_HEAD(&root->delalloc_root);
1177         INIT_LIST_HEAD(&root->ordered_extents);
1178         INIT_LIST_HEAD(&root->ordered_root);
1179         INIT_LIST_HEAD(&root->reloc_dirty_list);
1180         INIT_LIST_HEAD(&root->logged_list[0]);
1181         INIT_LIST_HEAD(&root->logged_list[1]);
1182         spin_lock_init(&root->inode_lock);
1183         spin_lock_init(&root->delalloc_lock);
1184         spin_lock_init(&root->ordered_extent_lock);
1185         spin_lock_init(&root->accounting_lock);
1186         spin_lock_init(&root->log_extents_lock[0]);
1187         spin_lock_init(&root->log_extents_lock[1]);
1188         spin_lock_init(&root->qgroup_meta_rsv_lock);
1189         mutex_init(&root->objectid_mutex);
1190         mutex_init(&root->log_mutex);
1191         mutex_init(&root->ordered_extent_mutex);
1192         mutex_init(&root->delalloc_mutex);
1193         init_waitqueue_head(&root->log_writer_wait);
1194         init_waitqueue_head(&root->log_commit_wait[0]);
1195         init_waitqueue_head(&root->log_commit_wait[1]);
1196         INIT_LIST_HEAD(&root->log_ctxs[0]);
1197         INIT_LIST_HEAD(&root->log_ctxs[1]);
1198         atomic_set(&root->log_commit[0], 0);
1199         atomic_set(&root->log_commit[1], 0);
1200         atomic_set(&root->log_writers, 0);
1201         atomic_set(&root->log_batch, 0);
1202         refcount_set(&root->refs, 1);
1203         atomic_set(&root->will_be_snapshotted, 0);
1204         atomic_set(&root->snapshot_force_cow, 0);
1205         atomic_set(&root->nr_swapfiles, 0);
1206         root->log_transid = 0;
1207         root->log_transid_committed = -1;
1208         root->last_log_commit = 0;
1209         if (!dummy)
1210                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1211
1212         memset(&root->root_key, 0, sizeof(root->root_key));
1213         memset(&root->root_item, 0, sizeof(root->root_item));
1214         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1215         if (!dummy)
1216                 root->defrag_trans_start = fs_info->generation;
1217         else
1218                 root->defrag_trans_start = 0;
1219         root->root_key.objectid = objectid;
1220         root->anon_dev = 0;
1221
1222         spin_lock_init(&root->root_item_lock);
1223 }
1224
1225 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1226                 gfp_t flags)
1227 {
1228         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1229         if (root)
1230                 root->fs_info = fs_info;
1231         return root;
1232 }
1233
1234 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1235 /* Should only be used by the testing infrastructure */
1236 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1237 {
1238         struct btrfs_root *root;
1239
1240         if (!fs_info)
1241                 return ERR_PTR(-EINVAL);
1242
1243         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1244         if (!root)
1245                 return ERR_PTR(-ENOMEM);
1246
1247         /* We don't use the stripesize in selftest, set it as sectorsize */
1248         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1249         root->alloc_bytenr = 0;
1250
1251         return root;
1252 }
1253 #endif
1254
1255 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256                                      struct btrfs_fs_info *fs_info,
1257                                      u64 objectid)
1258 {
1259         struct extent_buffer *leaf;
1260         struct btrfs_root *tree_root = fs_info->tree_root;
1261         struct btrfs_root *root;
1262         struct btrfs_key key;
1263         unsigned int nofs_flag;
1264         int ret = 0;
1265         uuid_le uuid = NULL_UUID_LE;
1266
1267         /*
1268          * We're holding a transaction handle, so use a NOFS memory allocation
1269          * context to avoid deadlock if reclaim happens.
1270          */
1271         nofs_flag = memalloc_nofs_save();
1272         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1273         memalloc_nofs_restore(nofs_flag);
1274         if (!root)
1275                 return ERR_PTR(-ENOMEM);
1276
1277         __setup_root(root, fs_info, objectid);
1278         root->root_key.objectid = objectid;
1279         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1280         root->root_key.offset = 0;
1281
1282         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1283         if (IS_ERR(leaf)) {
1284                 ret = PTR_ERR(leaf);
1285                 leaf = NULL;
1286                 goto fail;
1287         }
1288
1289         root->node = leaf;
1290         btrfs_mark_buffer_dirty(leaf);
1291
1292         root->commit_root = btrfs_root_node(root);
1293         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1294
1295         root->root_item.flags = 0;
1296         root->root_item.byte_limit = 0;
1297         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1298         btrfs_set_root_generation(&root->root_item, trans->transid);
1299         btrfs_set_root_level(&root->root_item, 0);
1300         btrfs_set_root_refs(&root->root_item, 1);
1301         btrfs_set_root_used(&root->root_item, leaf->len);
1302         btrfs_set_root_last_snapshot(&root->root_item, 0);
1303         btrfs_set_root_dirid(&root->root_item, 0);
1304         if (is_fstree(objectid))
1305                 uuid_le_gen(&uuid);
1306         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1307         root->root_item.drop_level = 0;
1308
1309         key.objectid = objectid;
1310         key.type = BTRFS_ROOT_ITEM_KEY;
1311         key.offset = 0;
1312         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1313         if (ret)
1314                 goto fail;
1315
1316         btrfs_tree_unlock(leaf);
1317
1318         return root;
1319
1320 fail:
1321         if (leaf) {
1322                 btrfs_tree_unlock(leaf);
1323                 free_extent_buffer(root->commit_root);
1324                 free_extent_buffer(leaf);
1325         }
1326         kfree(root);
1327
1328         return ERR_PTR(ret);
1329 }
1330
1331 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1332                                          struct btrfs_fs_info *fs_info)
1333 {
1334         struct btrfs_root *root;
1335         struct extent_buffer *leaf;
1336
1337         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1338         if (!root)
1339                 return ERR_PTR(-ENOMEM);
1340
1341         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1342
1343         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1344         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1345         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1346
1347         /*
1348          * DON'T set REF_COWS for log trees
1349          *
1350          * log trees do not get reference counted because they go away
1351          * before a real commit is actually done.  They do store pointers
1352          * to file data extents, and those reference counts still get
1353          * updated (along with back refs to the log tree).
1354          */
1355
1356         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1357                         NULL, 0, 0, 0);
1358         if (IS_ERR(leaf)) {
1359                 kfree(root);
1360                 return ERR_CAST(leaf);
1361         }
1362
1363         root->node = leaf;
1364
1365         btrfs_mark_buffer_dirty(root->node);
1366         btrfs_tree_unlock(root->node);
1367         return root;
1368 }
1369
1370 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1371                              struct btrfs_fs_info *fs_info)
1372 {
1373         struct btrfs_root *log_root;
1374
1375         log_root = alloc_log_tree(trans, fs_info);
1376         if (IS_ERR(log_root))
1377                 return PTR_ERR(log_root);
1378         WARN_ON(fs_info->log_root_tree);
1379         fs_info->log_root_tree = log_root;
1380         return 0;
1381 }
1382
1383 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1384                        struct btrfs_root *root)
1385 {
1386         struct btrfs_fs_info *fs_info = root->fs_info;
1387         struct btrfs_root *log_root;
1388         struct btrfs_inode_item *inode_item;
1389
1390         log_root = alloc_log_tree(trans, fs_info);
1391         if (IS_ERR(log_root))
1392                 return PTR_ERR(log_root);
1393
1394         log_root->last_trans = trans->transid;
1395         log_root->root_key.offset = root->root_key.objectid;
1396
1397         inode_item = &log_root->root_item.inode;
1398         btrfs_set_stack_inode_generation(inode_item, 1);
1399         btrfs_set_stack_inode_size(inode_item, 3);
1400         btrfs_set_stack_inode_nlink(inode_item, 1);
1401         btrfs_set_stack_inode_nbytes(inode_item,
1402                                      fs_info->nodesize);
1403         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1404
1405         btrfs_set_root_node(&log_root->root_item, log_root->node);
1406
1407         WARN_ON(root->log_root);
1408         root->log_root = log_root;
1409         root->log_transid = 0;
1410         root->log_transid_committed = -1;
1411         root->last_log_commit = 0;
1412         return 0;
1413 }
1414
1415 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1416                                                struct btrfs_key *key)
1417 {
1418         struct btrfs_root *root;
1419         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1420         struct btrfs_path *path;
1421         u64 generation;
1422         int ret;
1423         int level;
1424
1425         path = btrfs_alloc_path();
1426         if (!path)
1427                 return ERR_PTR(-ENOMEM);
1428
1429         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1430         if (!root) {
1431                 ret = -ENOMEM;
1432                 goto alloc_fail;
1433         }
1434
1435         __setup_root(root, fs_info, key->objectid);
1436
1437         ret = btrfs_find_root(tree_root, key, path,
1438                               &root->root_item, &root->root_key);
1439         if (ret) {
1440                 if (ret > 0)
1441                         ret = -ENOENT;
1442                 goto find_fail;
1443         }
1444
1445         generation = btrfs_root_generation(&root->root_item);
1446         level = btrfs_root_level(&root->root_item);
1447         root->node = read_tree_block(fs_info,
1448                                      btrfs_root_bytenr(&root->root_item),
1449                                      generation, level, NULL);
1450         if (IS_ERR(root->node)) {
1451                 ret = PTR_ERR(root->node);
1452                 goto find_fail;
1453         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1454                 ret = -EIO;
1455                 free_extent_buffer(root->node);
1456                 goto find_fail;
1457         }
1458         root->commit_root = btrfs_root_node(root);
1459 out:
1460         btrfs_free_path(path);
1461         return root;
1462
1463 find_fail:
1464         kfree(root);
1465 alloc_fail:
1466         root = ERR_PTR(ret);
1467         goto out;
1468 }
1469
1470 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1471                                       struct btrfs_key *location)
1472 {
1473         struct btrfs_root *root;
1474
1475         root = btrfs_read_tree_root(tree_root, location);
1476         if (IS_ERR(root))
1477                 return root;
1478
1479         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1480                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1481                 btrfs_check_and_init_root_item(&root->root_item);
1482         }
1483
1484         return root;
1485 }
1486
1487 int btrfs_init_fs_root(struct btrfs_root *root)
1488 {
1489         int ret;
1490         struct btrfs_subvolume_writers *writers;
1491
1492         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1493         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1494                                         GFP_NOFS);
1495         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1496                 ret = -ENOMEM;
1497                 goto fail;
1498         }
1499
1500         writers = btrfs_alloc_subvolume_writers();
1501         if (IS_ERR(writers)) {
1502                 ret = PTR_ERR(writers);
1503                 goto fail;
1504         }
1505         root->subv_writers = writers;
1506
1507         btrfs_init_free_ino_ctl(root);
1508         spin_lock_init(&root->ino_cache_lock);
1509         init_waitqueue_head(&root->ino_cache_wait);
1510
1511         ret = get_anon_bdev(&root->anon_dev);
1512         if (ret)
1513                 goto fail;
1514
1515         mutex_lock(&root->objectid_mutex);
1516         ret = btrfs_find_highest_objectid(root,
1517                                         &root->highest_objectid);
1518         if (ret) {
1519                 mutex_unlock(&root->objectid_mutex);
1520                 goto fail;
1521         }
1522
1523         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1524
1525         mutex_unlock(&root->objectid_mutex);
1526
1527         return 0;
1528 fail:
1529         /* The caller is responsible to call btrfs_free_fs_root */
1530         return ret;
1531 }
1532
1533 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1534                                         u64 root_id)
1535 {
1536         struct btrfs_root *root;
1537
1538         spin_lock(&fs_info->fs_roots_radix_lock);
1539         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1540                                  (unsigned long)root_id);
1541         spin_unlock(&fs_info->fs_roots_radix_lock);
1542         return root;
1543 }
1544
1545 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1546                          struct btrfs_root *root)
1547 {
1548         int ret;
1549
1550         ret = radix_tree_preload(GFP_NOFS);
1551         if (ret)
1552                 return ret;
1553
1554         spin_lock(&fs_info->fs_roots_radix_lock);
1555         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1556                                 (unsigned long)root->root_key.objectid,
1557                                 root);
1558         if (ret == 0)
1559                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1560         spin_unlock(&fs_info->fs_roots_radix_lock);
1561         radix_tree_preload_end();
1562
1563         return ret;
1564 }
1565
1566 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1567                                      struct btrfs_key *location,
1568                                      bool check_ref)
1569 {
1570         struct btrfs_root *root;
1571         struct btrfs_path *path;
1572         struct btrfs_key key;
1573         int ret;
1574
1575         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1576                 return fs_info->tree_root;
1577         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1578                 return fs_info->extent_root;
1579         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1580                 return fs_info->chunk_root;
1581         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1582                 return fs_info->dev_root;
1583         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1584                 return fs_info->csum_root;
1585         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1586                 return fs_info->quota_root ? fs_info->quota_root :
1587                                              ERR_PTR(-ENOENT);
1588         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1589                 return fs_info->uuid_root ? fs_info->uuid_root :
1590                                             ERR_PTR(-ENOENT);
1591         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1592                 return fs_info->free_space_root ? fs_info->free_space_root :
1593                                                   ERR_PTR(-ENOENT);
1594 again:
1595         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1596         if (root) {
1597                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1598                         return ERR_PTR(-ENOENT);
1599                 return root;
1600         }
1601
1602         root = btrfs_read_fs_root(fs_info->tree_root, location);
1603         if (IS_ERR(root))
1604                 return root;
1605
1606         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1607                 ret = -ENOENT;
1608                 goto fail;
1609         }
1610
1611         ret = btrfs_init_fs_root(root);
1612         if (ret)
1613                 goto fail;
1614
1615         path = btrfs_alloc_path();
1616         if (!path) {
1617                 ret = -ENOMEM;
1618                 goto fail;
1619         }
1620         key.objectid = BTRFS_ORPHAN_OBJECTID;
1621         key.type = BTRFS_ORPHAN_ITEM_KEY;
1622         key.offset = location->objectid;
1623
1624         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1625         btrfs_free_path(path);
1626         if (ret < 0)
1627                 goto fail;
1628         if (ret == 0)
1629                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1630
1631         ret = btrfs_insert_fs_root(fs_info, root);
1632         if (ret) {
1633                 if (ret == -EEXIST) {
1634                         btrfs_free_fs_root(root);
1635                         goto again;
1636                 }
1637                 goto fail;
1638         }
1639         return root;
1640 fail:
1641         btrfs_free_fs_root(root);
1642         return ERR_PTR(ret);
1643 }
1644
1645 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1646 {
1647         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1648         int ret = 0;
1649         struct btrfs_device *device;
1650         struct backing_dev_info *bdi;
1651
1652         rcu_read_lock();
1653         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1654                 if (!device->bdev)
1655                         continue;
1656                 bdi = device->bdev->bd_bdi;
1657                 if (bdi_congested(bdi, bdi_bits)) {
1658                         ret = 1;
1659                         break;
1660                 }
1661         }
1662         rcu_read_unlock();
1663         return ret;
1664 }
1665
1666 /*
1667  * called by the kthread helper functions to finally call the bio end_io
1668  * functions.  This is where read checksum verification actually happens
1669  */
1670 static void end_workqueue_fn(struct btrfs_work *work)
1671 {
1672         struct bio *bio;
1673         struct btrfs_end_io_wq *end_io_wq;
1674
1675         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1676         bio = end_io_wq->bio;
1677
1678         bio->bi_status = end_io_wq->status;
1679         bio->bi_private = end_io_wq->private;
1680         bio->bi_end_io = end_io_wq->end_io;
1681         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1682         bio_endio(bio);
1683 }
1684
1685 static int cleaner_kthread(void *arg)
1686 {
1687         struct btrfs_root *root = arg;
1688         struct btrfs_fs_info *fs_info = root->fs_info;
1689         int again;
1690
1691         while (1) {
1692                 again = 0;
1693
1694                 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1695
1696                 /* Make the cleaner go to sleep early. */
1697                 if (btrfs_need_cleaner_sleep(fs_info))
1698                         goto sleep;
1699
1700                 /*
1701                  * Do not do anything if we might cause open_ctree() to block
1702                  * before we have finished mounting the filesystem.
1703                  */
1704                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1705                         goto sleep;
1706
1707                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1708                         goto sleep;
1709
1710                 /*
1711                  * Avoid the problem that we change the status of the fs
1712                  * during the above check and trylock.
1713                  */
1714                 if (btrfs_need_cleaner_sleep(fs_info)) {
1715                         mutex_unlock(&fs_info->cleaner_mutex);
1716                         goto sleep;
1717                 }
1718
1719                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1720                 btrfs_run_delayed_iputs(fs_info);
1721                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1722
1723                 again = btrfs_clean_one_deleted_snapshot(root);
1724                 mutex_unlock(&fs_info->cleaner_mutex);
1725
1726                 /*
1727                  * The defragger has dealt with the R/O remount and umount,
1728                  * needn't do anything special here.
1729                  */
1730                 btrfs_run_defrag_inodes(fs_info);
1731
1732                 /*
1733                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1734                  * with relocation (btrfs_relocate_chunk) and relocation
1735                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1736                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1737                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1738                  * unused block groups.
1739                  */
1740                 btrfs_delete_unused_bgs(fs_info);
1741 sleep:
1742                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1743                 if (kthread_should_park())
1744                         kthread_parkme();
1745                 if (kthread_should_stop())
1746                         return 0;
1747                 if (!again) {
1748                         set_current_state(TASK_INTERRUPTIBLE);
1749                         schedule();
1750                         __set_current_state(TASK_RUNNING);
1751                 }
1752         }
1753 }
1754
1755 static int transaction_kthread(void *arg)
1756 {
1757         struct btrfs_root *root = arg;
1758         struct btrfs_fs_info *fs_info = root->fs_info;
1759         struct btrfs_trans_handle *trans;
1760         struct btrfs_transaction *cur;
1761         u64 transid;
1762         time64_t now;
1763         unsigned long delay;
1764         bool cannot_commit;
1765
1766         do {
1767                 cannot_commit = false;
1768                 delay = HZ * fs_info->commit_interval;
1769                 mutex_lock(&fs_info->transaction_kthread_mutex);
1770
1771                 spin_lock(&fs_info->trans_lock);
1772                 cur = fs_info->running_transaction;
1773                 if (!cur) {
1774                         spin_unlock(&fs_info->trans_lock);
1775                         goto sleep;
1776                 }
1777
1778                 now = ktime_get_seconds();
1779                 if (cur->state < TRANS_STATE_BLOCKED &&
1780                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1781                     (now < cur->start_time ||
1782                      now - cur->start_time < fs_info->commit_interval)) {
1783                         spin_unlock(&fs_info->trans_lock);
1784                         delay = HZ * 5;
1785                         goto sleep;
1786                 }
1787                 transid = cur->transid;
1788                 spin_unlock(&fs_info->trans_lock);
1789
1790                 /* If the file system is aborted, this will always fail. */
1791                 trans = btrfs_attach_transaction(root);
1792                 if (IS_ERR(trans)) {
1793                         if (PTR_ERR(trans) != -ENOENT)
1794                                 cannot_commit = true;
1795                         goto sleep;
1796                 }
1797                 if (transid == trans->transid) {
1798                         btrfs_commit_transaction(trans);
1799                 } else {
1800                         btrfs_end_transaction(trans);
1801                 }
1802 sleep:
1803                 wake_up_process(fs_info->cleaner_kthread);
1804                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1805
1806                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1807                                       &fs_info->fs_state)))
1808                         btrfs_cleanup_transaction(fs_info);
1809                 if (!kthread_should_stop() &&
1810                                 (!btrfs_transaction_blocked(fs_info) ||
1811                                  cannot_commit))
1812                         schedule_timeout_interruptible(delay);
1813         } while (!kthread_should_stop());
1814         return 0;
1815 }
1816
1817 /*
1818  * this will find the highest generation in the array of
1819  * root backups.  The index of the highest array is returned,
1820  * or -1 if we can't find anything.
1821  *
1822  * We check to make sure the array is valid by comparing the
1823  * generation of the latest  root in the array with the generation
1824  * in the super block.  If they don't match we pitch it.
1825  */
1826 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1827 {
1828         u64 cur;
1829         int newest_index = -1;
1830         struct btrfs_root_backup *root_backup;
1831         int i;
1832
1833         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1834                 root_backup = info->super_copy->super_roots + i;
1835                 cur = btrfs_backup_tree_root_gen(root_backup);
1836                 if (cur == newest_gen)
1837                         newest_index = i;
1838         }
1839
1840         /* check to see if we actually wrapped around */
1841         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1842                 root_backup = info->super_copy->super_roots;
1843                 cur = btrfs_backup_tree_root_gen(root_backup);
1844                 if (cur == newest_gen)
1845                         newest_index = 0;
1846         }
1847         return newest_index;
1848 }
1849
1850
1851 /*
1852  * find the oldest backup so we know where to store new entries
1853  * in the backup array.  This will set the backup_root_index
1854  * field in the fs_info struct
1855  */
1856 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1857                                      u64 newest_gen)
1858 {
1859         int newest_index = -1;
1860
1861         newest_index = find_newest_super_backup(info, newest_gen);
1862         /* if there was garbage in there, just move along */
1863         if (newest_index == -1) {
1864                 info->backup_root_index = 0;
1865         } else {
1866                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1867         }
1868 }
1869
1870 /*
1871  * copy all the root pointers into the super backup array.
1872  * this will bump the backup pointer by one when it is
1873  * done
1874  */
1875 static void backup_super_roots(struct btrfs_fs_info *info)
1876 {
1877         int next_backup;
1878         struct btrfs_root_backup *root_backup;
1879         int last_backup;
1880
1881         next_backup = info->backup_root_index;
1882         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1883                 BTRFS_NUM_BACKUP_ROOTS;
1884
1885         /*
1886          * just overwrite the last backup if we're at the same generation
1887          * this happens only at umount
1888          */
1889         root_backup = info->super_for_commit->super_roots + last_backup;
1890         if (btrfs_backup_tree_root_gen(root_backup) ==
1891             btrfs_header_generation(info->tree_root->node))
1892                 next_backup = last_backup;
1893
1894         root_backup = info->super_for_commit->super_roots + next_backup;
1895
1896         /*
1897          * make sure all of our padding and empty slots get zero filled
1898          * regardless of which ones we use today
1899          */
1900         memset(root_backup, 0, sizeof(*root_backup));
1901
1902         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1903
1904         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1905         btrfs_set_backup_tree_root_gen(root_backup,
1906                                btrfs_header_generation(info->tree_root->node));
1907
1908         btrfs_set_backup_tree_root_level(root_backup,
1909                                btrfs_header_level(info->tree_root->node));
1910
1911         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1912         btrfs_set_backup_chunk_root_gen(root_backup,
1913                                btrfs_header_generation(info->chunk_root->node));
1914         btrfs_set_backup_chunk_root_level(root_backup,
1915                                btrfs_header_level(info->chunk_root->node));
1916
1917         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1918         btrfs_set_backup_extent_root_gen(root_backup,
1919                                btrfs_header_generation(info->extent_root->node));
1920         btrfs_set_backup_extent_root_level(root_backup,
1921                                btrfs_header_level(info->extent_root->node));
1922
1923         /*
1924          * we might commit during log recovery, which happens before we set
1925          * the fs_root.  Make sure it is valid before we fill it in.
1926          */
1927         if (info->fs_root && info->fs_root->node) {
1928                 btrfs_set_backup_fs_root(root_backup,
1929                                          info->fs_root->node->start);
1930                 btrfs_set_backup_fs_root_gen(root_backup,
1931                                btrfs_header_generation(info->fs_root->node));
1932                 btrfs_set_backup_fs_root_level(root_backup,
1933                                btrfs_header_level(info->fs_root->node));
1934         }
1935
1936         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1937         btrfs_set_backup_dev_root_gen(root_backup,
1938                                btrfs_header_generation(info->dev_root->node));
1939         btrfs_set_backup_dev_root_level(root_backup,
1940                                        btrfs_header_level(info->dev_root->node));
1941
1942         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1943         btrfs_set_backup_csum_root_gen(root_backup,
1944                                btrfs_header_generation(info->csum_root->node));
1945         btrfs_set_backup_csum_root_level(root_backup,
1946                                btrfs_header_level(info->csum_root->node));
1947
1948         btrfs_set_backup_total_bytes(root_backup,
1949                              btrfs_super_total_bytes(info->super_copy));
1950         btrfs_set_backup_bytes_used(root_backup,
1951                              btrfs_super_bytes_used(info->super_copy));
1952         btrfs_set_backup_num_devices(root_backup,
1953                              btrfs_super_num_devices(info->super_copy));
1954
1955         /*
1956          * if we don't copy this out to the super_copy, it won't get remembered
1957          * for the next commit
1958          */
1959         memcpy(&info->super_copy->super_roots,
1960                &info->super_for_commit->super_roots,
1961                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1962 }
1963
1964 /*
1965  * this copies info out of the root backup array and back into
1966  * the in-memory super block.  It is meant to help iterate through
1967  * the array, so you send it the number of backups you've already
1968  * tried and the last backup index you used.
1969  *
1970  * this returns -1 when it has tried all the backups
1971  */
1972 static noinline int next_root_backup(struct btrfs_fs_info *info,
1973                                      struct btrfs_super_block *super,
1974                                      int *num_backups_tried, int *backup_index)
1975 {
1976         struct btrfs_root_backup *root_backup;
1977         int newest = *backup_index;
1978
1979         if (*num_backups_tried == 0) {
1980                 u64 gen = btrfs_super_generation(super);
1981
1982                 newest = find_newest_super_backup(info, gen);
1983                 if (newest == -1)
1984                         return -1;
1985
1986                 *backup_index = newest;
1987                 *num_backups_tried = 1;
1988         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1989                 /* we've tried all the backups, all done */
1990                 return -1;
1991         } else {
1992                 /* jump to the next oldest backup */
1993                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1994                         BTRFS_NUM_BACKUP_ROOTS;
1995                 *backup_index = newest;
1996                 *num_backups_tried += 1;
1997         }
1998         root_backup = super->super_roots + newest;
1999
2000         btrfs_set_super_generation(super,
2001                                    btrfs_backup_tree_root_gen(root_backup));
2002         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2003         btrfs_set_super_root_level(super,
2004                                    btrfs_backup_tree_root_level(root_backup));
2005         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2006
2007         /*
2008          * fixme: the total bytes and num_devices need to match or we should
2009          * need a fsck
2010          */
2011         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2012         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2013         return 0;
2014 }
2015
2016 /* helper to cleanup workers */
2017 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2018 {
2019         btrfs_destroy_workqueue(fs_info->fixup_workers);
2020         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2021         btrfs_destroy_workqueue(fs_info->workers);
2022         btrfs_destroy_workqueue(fs_info->endio_workers);
2023         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2024         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2025         btrfs_destroy_workqueue(fs_info->rmw_workers);
2026         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2027         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2028         btrfs_destroy_workqueue(fs_info->submit_workers);
2029         btrfs_destroy_workqueue(fs_info->delayed_workers);
2030         btrfs_destroy_workqueue(fs_info->caching_workers);
2031         btrfs_destroy_workqueue(fs_info->readahead_workers);
2032         btrfs_destroy_workqueue(fs_info->flush_workers);
2033         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2034         btrfs_destroy_workqueue(fs_info->extent_workers);
2035         /*
2036          * Now that all other work queues are destroyed, we can safely destroy
2037          * the queues used for metadata I/O, since tasks from those other work
2038          * queues can do metadata I/O operations.
2039          */
2040         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2041         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2042 }
2043
2044 static void free_root_extent_buffers(struct btrfs_root *root)
2045 {
2046         if (root) {
2047                 free_extent_buffer(root->node);
2048                 free_extent_buffer(root->commit_root);
2049                 root->node = NULL;
2050                 root->commit_root = NULL;
2051         }
2052 }
2053
2054 /* helper to cleanup tree roots */
2055 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2056 {
2057         free_root_extent_buffers(info->tree_root);
2058
2059         free_root_extent_buffers(info->dev_root);
2060         free_root_extent_buffers(info->extent_root);
2061         free_root_extent_buffers(info->csum_root);
2062         free_root_extent_buffers(info->quota_root);
2063         free_root_extent_buffers(info->uuid_root);
2064         if (chunk_root)
2065                 free_root_extent_buffers(info->chunk_root);
2066         free_root_extent_buffers(info->free_space_root);
2067 }
2068
2069 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2070 {
2071         int ret;
2072         struct btrfs_root *gang[8];
2073         int i;
2074
2075         while (!list_empty(&fs_info->dead_roots)) {
2076                 gang[0] = list_entry(fs_info->dead_roots.next,
2077                                      struct btrfs_root, root_list);
2078                 list_del(&gang[0]->root_list);
2079
2080                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2081                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2082                 } else {
2083                         free_extent_buffer(gang[0]->node);
2084                         free_extent_buffer(gang[0]->commit_root);
2085                         btrfs_put_fs_root(gang[0]);
2086                 }
2087         }
2088
2089         while (1) {
2090                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2091                                              (void **)gang, 0,
2092                                              ARRAY_SIZE(gang));
2093                 if (!ret)
2094                         break;
2095                 for (i = 0; i < ret; i++)
2096                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2097         }
2098
2099         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2100                 btrfs_free_log_root_tree(NULL, fs_info);
2101                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2102         }
2103 }
2104
2105 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2106 {
2107         mutex_init(&fs_info->scrub_lock);
2108         atomic_set(&fs_info->scrubs_running, 0);
2109         atomic_set(&fs_info->scrub_pause_req, 0);
2110         atomic_set(&fs_info->scrubs_paused, 0);
2111         atomic_set(&fs_info->scrub_cancel_req, 0);
2112         init_waitqueue_head(&fs_info->scrub_pause_wait);
2113         fs_info->scrub_workers_refcnt = 0;
2114 }
2115
2116 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2117 {
2118         spin_lock_init(&fs_info->balance_lock);
2119         mutex_init(&fs_info->balance_mutex);
2120         atomic_set(&fs_info->balance_pause_req, 0);
2121         atomic_set(&fs_info->balance_cancel_req, 0);
2122         fs_info->balance_ctl = NULL;
2123         init_waitqueue_head(&fs_info->balance_wait_q);
2124 }
2125
2126 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2127 {
2128         struct inode *inode = fs_info->btree_inode;
2129
2130         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2131         set_nlink(inode, 1);
2132         /*
2133          * we set the i_size on the btree inode to the max possible int.
2134          * the real end of the address space is determined by all of
2135          * the devices in the system
2136          */
2137         inode->i_size = OFFSET_MAX;
2138         inode->i_mapping->a_ops = &btree_aops;
2139
2140         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2141         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2142         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2143         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2144
2145         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2146
2147         BTRFS_I(inode)->root = fs_info->tree_root;
2148         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2149         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2150         btrfs_insert_inode_hash(inode);
2151 }
2152
2153 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2154 {
2155         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2156         init_rwsem(&fs_info->dev_replace.rwsem);
2157         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2158 }
2159
2160 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2161 {
2162         spin_lock_init(&fs_info->qgroup_lock);
2163         mutex_init(&fs_info->qgroup_ioctl_lock);
2164         fs_info->qgroup_tree = RB_ROOT;
2165         fs_info->qgroup_op_tree = RB_ROOT;
2166         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2167         fs_info->qgroup_seq = 1;
2168         fs_info->qgroup_ulist = NULL;
2169         fs_info->qgroup_rescan_running = false;
2170         mutex_init(&fs_info->qgroup_rescan_lock);
2171 }
2172
2173 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2174                 struct btrfs_fs_devices *fs_devices)
2175 {
2176         u32 max_active = fs_info->thread_pool_size;
2177         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2178
2179         fs_info->workers =
2180                 btrfs_alloc_workqueue(fs_info, "worker",
2181                                       flags | WQ_HIGHPRI, max_active, 16);
2182
2183         fs_info->delalloc_workers =
2184                 btrfs_alloc_workqueue(fs_info, "delalloc",
2185                                       flags, max_active, 2);
2186
2187         fs_info->flush_workers =
2188                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2189                                       flags, max_active, 0);
2190
2191         fs_info->caching_workers =
2192                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2193
2194         /*
2195          * a higher idle thresh on the submit workers makes it much more
2196          * likely that bios will be send down in a sane order to the
2197          * devices
2198          */
2199         fs_info->submit_workers =
2200                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2201                                       min_t(u64, fs_devices->num_devices,
2202                                             max_active), 64);
2203
2204         fs_info->fixup_workers =
2205                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2206
2207         /*
2208          * endios are largely parallel and should have a very
2209          * low idle thresh
2210          */
2211         fs_info->endio_workers =
2212                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2213         fs_info->endio_meta_workers =
2214                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2215                                       max_active, 4);
2216         fs_info->endio_meta_write_workers =
2217                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2218                                       max_active, 2);
2219         fs_info->endio_raid56_workers =
2220                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2221                                       max_active, 4);
2222         fs_info->endio_repair_workers =
2223                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2224         fs_info->rmw_workers =
2225                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2226         fs_info->endio_write_workers =
2227                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2228                                       max_active, 2);
2229         fs_info->endio_freespace_worker =
2230                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2231                                       max_active, 0);
2232         fs_info->delayed_workers =
2233                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2234                                       max_active, 0);
2235         fs_info->readahead_workers =
2236                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2237                                       max_active, 2);
2238         fs_info->qgroup_rescan_workers =
2239                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2240         fs_info->extent_workers =
2241                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2242                                       min_t(u64, fs_devices->num_devices,
2243                                             max_active), 8);
2244
2245         if (!(fs_info->workers && fs_info->delalloc_workers &&
2246               fs_info->submit_workers && fs_info->flush_workers &&
2247               fs_info->endio_workers && fs_info->endio_meta_workers &&
2248               fs_info->endio_meta_write_workers &&
2249               fs_info->endio_repair_workers &&
2250               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2251               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2252               fs_info->caching_workers && fs_info->readahead_workers &&
2253               fs_info->fixup_workers && fs_info->delayed_workers &&
2254               fs_info->extent_workers &&
2255               fs_info->qgroup_rescan_workers)) {
2256                 return -ENOMEM;
2257         }
2258
2259         return 0;
2260 }
2261
2262 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2263                             struct btrfs_fs_devices *fs_devices)
2264 {
2265         int ret;
2266         struct btrfs_root *log_tree_root;
2267         struct btrfs_super_block *disk_super = fs_info->super_copy;
2268         u64 bytenr = btrfs_super_log_root(disk_super);
2269         int level = btrfs_super_log_root_level(disk_super);
2270
2271         if (fs_devices->rw_devices == 0) {
2272                 btrfs_warn(fs_info, "log replay required on RO media");
2273                 return -EIO;
2274         }
2275
2276         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2277         if (!log_tree_root)
2278                 return -ENOMEM;
2279
2280         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2281
2282         log_tree_root->node = read_tree_block(fs_info, bytenr,
2283                                               fs_info->generation + 1,
2284                                               level, NULL);
2285         if (IS_ERR(log_tree_root->node)) {
2286                 btrfs_warn(fs_info, "failed to read log tree");
2287                 ret = PTR_ERR(log_tree_root->node);
2288                 kfree(log_tree_root);
2289                 return ret;
2290         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2291                 btrfs_err(fs_info, "failed to read log tree");
2292                 free_extent_buffer(log_tree_root->node);
2293                 kfree(log_tree_root);
2294                 return -EIO;
2295         }
2296         /* returns with log_tree_root freed on success */
2297         ret = btrfs_recover_log_trees(log_tree_root);
2298         if (ret) {
2299                 btrfs_handle_fs_error(fs_info, ret,
2300                                       "Failed to recover log tree");
2301                 free_extent_buffer(log_tree_root->node);
2302                 kfree(log_tree_root);
2303                 return ret;
2304         }
2305
2306         if (sb_rdonly(fs_info->sb)) {
2307                 ret = btrfs_commit_super(fs_info);
2308                 if (ret)
2309                         return ret;
2310         }
2311
2312         return 0;
2313 }
2314
2315 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2316 {
2317         struct btrfs_root *tree_root = fs_info->tree_root;
2318         struct btrfs_root *root;
2319         struct btrfs_key location;
2320         int ret;
2321
2322         BUG_ON(!fs_info->tree_root);
2323
2324         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2325         location.type = BTRFS_ROOT_ITEM_KEY;
2326         location.offset = 0;
2327
2328         root = btrfs_read_tree_root(tree_root, &location);
2329         if (IS_ERR(root)) {
2330                 ret = PTR_ERR(root);
2331                 goto out;
2332         }
2333         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2334         fs_info->extent_root = root;
2335
2336         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2337         root = btrfs_read_tree_root(tree_root, &location);
2338         if (IS_ERR(root)) {
2339                 ret = PTR_ERR(root);
2340                 goto out;
2341         }
2342         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2343         fs_info->dev_root = root;
2344         btrfs_init_devices_late(fs_info);
2345
2346         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2347         root = btrfs_read_tree_root(tree_root, &location);
2348         if (IS_ERR(root)) {
2349                 ret = PTR_ERR(root);
2350                 goto out;
2351         }
2352         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2353         fs_info->csum_root = root;
2354
2355         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2356         root = btrfs_read_tree_root(tree_root, &location);
2357         if (!IS_ERR(root)) {
2358                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2359                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2360                 fs_info->quota_root = root;
2361         }
2362
2363         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2364         root = btrfs_read_tree_root(tree_root, &location);
2365         if (IS_ERR(root)) {
2366                 ret = PTR_ERR(root);
2367                 if (ret != -ENOENT)
2368                         goto out;
2369         } else {
2370                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2371                 fs_info->uuid_root = root;
2372         }
2373
2374         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2375                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2376                 root = btrfs_read_tree_root(tree_root, &location);
2377                 if (IS_ERR(root)) {
2378                         ret = PTR_ERR(root);
2379                         goto out;
2380                 }
2381                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382                 fs_info->free_space_root = root;
2383         }
2384
2385         return 0;
2386 out:
2387         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2388                    location.objectid, ret);
2389         return ret;
2390 }
2391
2392 /*
2393  * Real super block validation
2394  * NOTE: super csum type and incompat features will not be checked here.
2395  *
2396  * @sb:         super block to check
2397  * @mirror_num: the super block number to check its bytenr:
2398  *              0       the primary (1st) sb
2399  *              1, 2    2nd and 3rd backup copy
2400  *             -1       skip bytenr check
2401  */
2402 static int validate_super(struct btrfs_fs_info *fs_info,
2403                             struct btrfs_super_block *sb, int mirror_num)
2404 {
2405         u64 nodesize = btrfs_super_nodesize(sb);
2406         u64 sectorsize = btrfs_super_sectorsize(sb);
2407         int ret = 0;
2408
2409         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2410                 btrfs_err(fs_info, "no valid FS found");
2411                 ret = -EINVAL;
2412         }
2413         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2414                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2415                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2416                 ret = -EINVAL;
2417         }
2418         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2419                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2420                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2421                 ret = -EINVAL;
2422         }
2423         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2424                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2425                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2426                 ret = -EINVAL;
2427         }
2428         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2429                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2430                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2431                 ret = -EINVAL;
2432         }
2433
2434         /*
2435          * Check sectorsize and nodesize first, other check will need it.
2436          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2437          */
2438         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2439             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2440                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2441                 ret = -EINVAL;
2442         }
2443         /* Only PAGE SIZE is supported yet */
2444         if (sectorsize != PAGE_SIZE) {
2445                 btrfs_err(fs_info,
2446                         "sectorsize %llu not supported yet, only support %lu",
2447                         sectorsize, PAGE_SIZE);
2448                 ret = -EINVAL;
2449         }
2450         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2451             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2452                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2453                 ret = -EINVAL;
2454         }
2455         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2456                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2457                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2458                 ret = -EINVAL;
2459         }
2460
2461         /* Root alignment check */
2462         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2463                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2464                            btrfs_super_root(sb));
2465                 ret = -EINVAL;
2466         }
2467         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2468                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2469                            btrfs_super_chunk_root(sb));
2470                 ret = -EINVAL;
2471         }
2472         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2473                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2474                            btrfs_super_log_root(sb));
2475                 ret = -EINVAL;
2476         }
2477
2478         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2479                    BTRFS_FSID_SIZE) != 0) {
2480                 btrfs_err(fs_info,
2481                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2482                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2483                 ret = -EINVAL;
2484         }
2485
2486         /*
2487          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2488          * done later
2489          */
2490         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2491                 btrfs_err(fs_info, "bytes_used is too small %llu",
2492                           btrfs_super_bytes_used(sb));
2493                 ret = -EINVAL;
2494         }
2495         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2496                 btrfs_err(fs_info, "invalid stripesize %u",
2497                           btrfs_super_stripesize(sb));
2498                 ret = -EINVAL;
2499         }
2500         if (btrfs_super_num_devices(sb) > (1UL << 31))
2501                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2502                            btrfs_super_num_devices(sb));
2503         if (btrfs_super_num_devices(sb) == 0) {
2504                 btrfs_err(fs_info, "number of devices is 0");
2505                 ret = -EINVAL;
2506         }
2507
2508         if (mirror_num >= 0 &&
2509             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2510                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2511                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2512                 ret = -EINVAL;
2513         }
2514
2515         /*
2516          * Obvious sys_chunk_array corruptions, it must hold at least one key
2517          * and one chunk
2518          */
2519         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2520                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2521                           btrfs_super_sys_array_size(sb),
2522                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2523                 ret = -EINVAL;
2524         }
2525         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2526                         + sizeof(struct btrfs_chunk)) {
2527                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2528                           btrfs_super_sys_array_size(sb),
2529                           sizeof(struct btrfs_disk_key)
2530                           + sizeof(struct btrfs_chunk));
2531                 ret = -EINVAL;
2532         }
2533
2534         /*
2535          * The generation is a global counter, we'll trust it more than the others
2536          * but it's still possible that it's the one that's wrong.
2537          */
2538         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2539                 btrfs_warn(fs_info,
2540                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2541                         btrfs_super_generation(sb),
2542                         btrfs_super_chunk_root_generation(sb));
2543         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2544             && btrfs_super_cache_generation(sb) != (u64)-1)
2545                 btrfs_warn(fs_info,
2546                         "suspicious: generation < cache_generation: %llu < %llu",
2547                         btrfs_super_generation(sb),
2548                         btrfs_super_cache_generation(sb));
2549
2550         return ret;
2551 }
2552
2553 /*
2554  * Validation of super block at mount time.
2555  * Some checks already done early at mount time, like csum type and incompat
2556  * flags will be skipped.
2557  */
2558 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2559 {
2560         return validate_super(fs_info, fs_info->super_copy, 0);
2561 }
2562
2563 /*
2564  * Validation of super block at write time.
2565  * Some checks like bytenr check will be skipped as their values will be
2566  * overwritten soon.
2567  * Extra checks like csum type and incompat flags will be done here.
2568  */
2569 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2570                                       struct btrfs_super_block *sb)
2571 {
2572         int ret;
2573
2574         ret = validate_super(fs_info, sb, -1);
2575         if (ret < 0)
2576                 goto out;
2577         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2578                 ret = -EUCLEAN;
2579                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2580                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2581                 goto out;
2582         }
2583         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2584                 ret = -EUCLEAN;
2585                 btrfs_err(fs_info,
2586                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2587                           btrfs_super_incompat_flags(sb),
2588                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2589                 goto out;
2590         }
2591 out:
2592         if (ret < 0)
2593                 btrfs_err(fs_info,
2594                 "super block corruption detected before writing it to disk");
2595         return ret;
2596 }
2597
2598 int open_ctree(struct super_block *sb,
2599                struct btrfs_fs_devices *fs_devices,
2600                char *options)
2601 {
2602         u32 sectorsize;
2603         u32 nodesize;
2604         u32 stripesize;
2605         u64 generation;
2606         u64 features;
2607         struct btrfs_key location;
2608         struct buffer_head *bh;
2609         struct btrfs_super_block *disk_super;
2610         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2611         struct btrfs_root *tree_root;
2612         struct btrfs_root *chunk_root;
2613         int ret;
2614         int err = -EINVAL;
2615         int num_backups_tried = 0;
2616         int backup_index = 0;
2617         int clear_free_space_tree = 0;
2618         int level;
2619
2620         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2621         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2622         if (!tree_root || !chunk_root) {
2623                 err = -ENOMEM;
2624                 goto fail;
2625         }
2626
2627         ret = init_srcu_struct(&fs_info->subvol_srcu);
2628         if (ret) {
2629                 err = ret;
2630                 goto fail;
2631         }
2632
2633         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2634         if (ret) {
2635                 err = ret;
2636                 goto fail_srcu;
2637         }
2638         fs_info->dirty_metadata_batch = PAGE_SIZE *
2639                                         (1 + ilog2(nr_cpu_ids));
2640
2641         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2642         if (ret) {
2643                 err = ret;
2644                 goto fail_dirty_metadata_bytes;
2645         }
2646
2647         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2648                         GFP_KERNEL);
2649         if (ret) {
2650                 err = ret;
2651                 goto fail_delalloc_bytes;
2652         }
2653
2654         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2655         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2656         INIT_LIST_HEAD(&fs_info->trans_list);
2657         INIT_LIST_HEAD(&fs_info->dead_roots);
2658         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2659         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2660         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2661         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2662         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2663         spin_lock_init(&fs_info->delalloc_root_lock);
2664         spin_lock_init(&fs_info->trans_lock);
2665         spin_lock_init(&fs_info->fs_roots_radix_lock);
2666         spin_lock_init(&fs_info->delayed_iput_lock);
2667         spin_lock_init(&fs_info->defrag_inodes_lock);
2668         spin_lock_init(&fs_info->tree_mod_seq_lock);
2669         spin_lock_init(&fs_info->super_lock);
2670         spin_lock_init(&fs_info->qgroup_op_lock);
2671         spin_lock_init(&fs_info->buffer_lock);
2672         spin_lock_init(&fs_info->unused_bgs_lock);
2673         rwlock_init(&fs_info->tree_mod_log_lock);
2674         mutex_init(&fs_info->unused_bg_unpin_mutex);
2675         mutex_init(&fs_info->delete_unused_bgs_mutex);
2676         mutex_init(&fs_info->reloc_mutex);
2677         mutex_init(&fs_info->delalloc_root_mutex);
2678         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2679         seqlock_init(&fs_info->profiles_lock);
2680
2681         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2682         INIT_LIST_HEAD(&fs_info->space_info);
2683         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2684         INIT_LIST_HEAD(&fs_info->unused_bgs);
2685         btrfs_mapping_init(&fs_info->mapping_tree);
2686         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2687                              BTRFS_BLOCK_RSV_GLOBAL);
2688         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2689         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2690         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2691         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2692                              BTRFS_BLOCK_RSV_DELOPS);
2693         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2694                              BTRFS_BLOCK_RSV_DELREFS);
2695
2696         atomic_set(&fs_info->async_delalloc_pages, 0);
2697         atomic_set(&fs_info->defrag_running, 0);
2698         atomic_set(&fs_info->qgroup_op_seq, 0);
2699         atomic_set(&fs_info->reada_works_cnt, 0);
2700         atomic64_set(&fs_info->tree_mod_seq, 0);
2701         fs_info->sb = sb;
2702         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2703         fs_info->metadata_ratio = 0;
2704         fs_info->defrag_inodes = RB_ROOT;
2705         atomic64_set(&fs_info->free_chunk_space, 0);
2706         fs_info->tree_mod_log = RB_ROOT;
2707         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2708         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2709         /* readahead state */
2710         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2711         spin_lock_init(&fs_info->reada_lock);
2712         btrfs_init_ref_verify(fs_info);
2713
2714         fs_info->thread_pool_size = min_t(unsigned long,
2715                                           num_online_cpus() + 2, 8);
2716
2717         INIT_LIST_HEAD(&fs_info->ordered_roots);
2718         spin_lock_init(&fs_info->ordered_root_lock);
2719
2720         fs_info->btree_inode = new_inode(sb);
2721         if (!fs_info->btree_inode) {
2722                 err = -ENOMEM;
2723                 goto fail_bio_counter;
2724         }
2725         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2726
2727         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2728                                         GFP_KERNEL);
2729         if (!fs_info->delayed_root) {
2730                 err = -ENOMEM;
2731                 goto fail_iput;
2732         }
2733         btrfs_init_delayed_root(fs_info->delayed_root);
2734
2735         btrfs_init_scrub(fs_info);
2736 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2737         fs_info->check_integrity_print_mask = 0;
2738 #endif
2739         btrfs_init_balance(fs_info);
2740         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2741
2742         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2743         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2744
2745         btrfs_init_btree_inode(fs_info);
2746
2747         spin_lock_init(&fs_info->block_group_cache_lock);
2748         fs_info->block_group_cache_tree = RB_ROOT;
2749         fs_info->first_logical_byte = (u64)-1;
2750
2751         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2752         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2753         fs_info->pinned_extents = &fs_info->freed_extents[0];
2754         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2755
2756         mutex_init(&fs_info->ordered_operations_mutex);
2757         mutex_init(&fs_info->tree_log_mutex);
2758         mutex_init(&fs_info->chunk_mutex);
2759         mutex_init(&fs_info->transaction_kthread_mutex);
2760         mutex_init(&fs_info->cleaner_mutex);
2761         mutex_init(&fs_info->ro_block_group_mutex);
2762         init_rwsem(&fs_info->commit_root_sem);
2763         init_rwsem(&fs_info->cleanup_work_sem);
2764         init_rwsem(&fs_info->subvol_sem);
2765         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2766
2767         btrfs_init_dev_replace_locks(fs_info);
2768         btrfs_init_qgroup(fs_info);
2769
2770         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2771         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2772
2773         init_waitqueue_head(&fs_info->transaction_throttle);
2774         init_waitqueue_head(&fs_info->transaction_wait);
2775         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2776         init_waitqueue_head(&fs_info->async_submit_wait);
2777
2778         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2779
2780         /* Usable values until the real ones are cached from the superblock */
2781         fs_info->nodesize = 4096;
2782         fs_info->sectorsize = 4096;
2783         fs_info->stripesize = 4096;
2784
2785         spin_lock_init(&fs_info->swapfile_pins_lock);
2786         fs_info->swapfile_pins = RB_ROOT;
2787
2788         ret = btrfs_alloc_stripe_hash_table(fs_info);
2789         if (ret) {
2790                 err = ret;
2791                 goto fail_alloc;
2792         }
2793
2794         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2795
2796         invalidate_bdev(fs_devices->latest_bdev);
2797
2798         /*
2799          * Read super block and check the signature bytes only
2800          */
2801         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2802         if (IS_ERR(bh)) {
2803                 err = PTR_ERR(bh);
2804                 goto fail_alloc;
2805         }
2806
2807         /*
2808          * We want to check superblock checksum, the type is stored inside.
2809          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2810          */
2811         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2812                 btrfs_err(fs_info, "superblock checksum mismatch");
2813                 err = -EINVAL;
2814                 brelse(bh);
2815                 goto fail_alloc;
2816         }
2817
2818         /*
2819          * super_copy is zeroed at allocation time and we never touch the
2820          * following bytes up to INFO_SIZE, the checksum is calculated from
2821          * the whole block of INFO_SIZE
2822          */
2823         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2824         brelse(bh);
2825
2826         disk_super = fs_info->super_copy;
2827
2828         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2829                        BTRFS_FSID_SIZE));
2830
2831         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2832                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2833                                 fs_info->super_copy->metadata_uuid,
2834                                 BTRFS_FSID_SIZE));
2835         }
2836
2837         features = btrfs_super_flags(disk_super);
2838         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2839                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2840                 btrfs_set_super_flags(disk_super, features);
2841                 btrfs_info(fs_info,
2842                         "found metadata UUID change in progress flag, clearing");
2843         }
2844
2845         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2846                sizeof(*fs_info->super_for_commit));
2847
2848         ret = btrfs_validate_mount_super(fs_info);
2849         if (ret) {
2850                 btrfs_err(fs_info, "superblock contains fatal errors");
2851                 err = -EINVAL;
2852                 goto fail_alloc;
2853         }
2854
2855         if (!btrfs_super_root(disk_super))
2856                 goto fail_alloc;
2857
2858         /* check FS state, whether FS is broken. */
2859         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2860                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2861
2862         /*
2863          * run through our array of backup supers and setup
2864          * our ring pointer to the oldest one
2865          */
2866         generation = btrfs_super_generation(disk_super);
2867         find_oldest_super_backup(fs_info, generation);
2868
2869         /*
2870          * In the long term, we'll store the compression type in the super
2871          * block, and it'll be used for per file compression control.
2872          */
2873         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2874
2875         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2876         if (ret) {
2877                 err = ret;
2878                 goto fail_alloc;
2879         }
2880
2881         features = btrfs_super_incompat_flags(disk_super) &
2882                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2883         if (features) {
2884                 btrfs_err(fs_info,
2885                     "cannot mount because of unsupported optional features (%llx)",
2886                     features);
2887                 err = -EINVAL;
2888                 goto fail_alloc;
2889         }
2890
2891         features = btrfs_super_incompat_flags(disk_super);
2892         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2893         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2894                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2895         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2896                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2897
2898         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2899                 btrfs_info(fs_info, "has skinny extents");
2900
2901         /*
2902          * flag our filesystem as having big metadata blocks if
2903          * they are bigger than the page size
2904          */
2905         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2906                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2907                         btrfs_info(fs_info,
2908                                 "flagging fs with big metadata feature");
2909                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2910         }
2911
2912         nodesize = btrfs_super_nodesize(disk_super);
2913         sectorsize = btrfs_super_sectorsize(disk_super);
2914         stripesize = sectorsize;
2915         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2916         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2917
2918         /* Cache block sizes */
2919         fs_info->nodesize = nodesize;
2920         fs_info->sectorsize = sectorsize;
2921         fs_info->stripesize = stripesize;
2922
2923         /*
2924          * mixed block groups end up with duplicate but slightly offset
2925          * extent buffers for the same range.  It leads to corruptions
2926          */
2927         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2928             (sectorsize != nodesize)) {
2929                 btrfs_err(fs_info,
2930 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2931                         nodesize, sectorsize);
2932                 goto fail_alloc;
2933         }
2934
2935         /*
2936          * Needn't use the lock because there is no other task which will
2937          * update the flag.
2938          */
2939         btrfs_set_super_incompat_flags(disk_super, features);
2940
2941         features = btrfs_super_compat_ro_flags(disk_super) &
2942                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2943         if (!sb_rdonly(sb) && features) {
2944                 btrfs_err(fs_info,
2945         "cannot mount read-write because of unsupported optional features (%llx)",
2946                        features);
2947                 err = -EINVAL;
2948                 goto fail_alloc;
2949         }
2950
2951         ret = btrfs_init_workqueues(fs_info, fs_devices);
2952         if (ret) {
2953                 err = ret;
2954                 goto fail_sb_buffer;
2955         }
2956
2957         sb->s_bdi->congested_fn = btrfs_congested_fn;
2958         sb->s_bdi->congested_data = fs_info;
2959         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2960         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2961         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2962         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2963
2964         sb->s_blocksize = sectorsize;
2965         sb->s_blocksize_bits = blksize_bits(sectorsize);
2966         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2967
2968         mutex_lock(&fs_info->chunk_mutex);
2969         ret = btrfs_read_sys_array(fs_info);
2970         mutex_unlock(&fs_info->chunk_mutex);
2971         if (ret) {
2972                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2973                 goto fail_sb_buffer;
2974         }
2975
2976         generation = btrfs_super_chunk_root_generation(disk_super);
2977         level = btrfs_super_chunk_root_level(disk_super);
2978
2979         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2980
2981         chunk_root->node = read_tree_block(fs_info,
2982                                            btrfs_super_chunk_root(disk_super),
2983                                            generation, level, NULL);
2984         if (IS_ERR(chunk_root->node) ||
2985             !extent_buffer_uptodate(chunk_root->node)) {
2986                 btrfs_err(fs_info, "failed to read chunk root");
2987                 if (!IS_ERR(chunk_root->node))
2988                         free_extent_buffer(chunk_root->node);
2989                 chunk_root->node = NULL;
2990                 goto fail_tree_roots;
2991         }
2992         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2993         chunk_root->commit_root = btrfs_root_node(chunk_root);
2994
2995         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2996            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2997
2998         ret = btrfs_read_chunk_tree(fs_info);
2999         if (ret) {
3000                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3001                 goto fail_tree_roots;
3002         }
3003
3004         /*
3005          * Keep the devid that is marked to be the target device for the
3006          * device replace procedure
3007          */
3008         btrfs_free_extra_devids(fs_devices, 0);
3009
3010         if (!fs_devices->latest_bdev) {
3011                 btrfs_err(fs_info, "failed to read devices");
3012                 goto fail_tree_roots;
3013         }
3014
3015 retry_root_backup:
3016         generation = btrfs_super_generation(disk_super);
3017         level = btrfs_super_root_level(disk_super);
3018
3019         tree_root->node = read_tree_block(fs_info,
3020                                           btrfs_super_root(disk_super),
3021                                           generation, level, NULL);
3022         if (IS_ERR(tree_root->node) ||
3023             !extent_buffer_uptodate(tree_root->node)) {
3024                 btrfs_warn(fs_info, "failed to read tree root");
3025                 if (!IS_ERR(tree_root->node))
3026                         free_extent_buffer(tree_root->node);
3027                 tree_root->node = NULL;
3028                 goto recovery_tree_root;
3029         }
3030
3031         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3032         tree_root->commit_root = btrfs_root_node(tree_root);
3033         btrfs_set_root_refs(&tree_root->root_item, 1);
3034
3035         mutex_lock(&tree_root->objectid_mutex);
3036         ret = btrfs_find_highest_objectid(tree_root,
3037                                         &tree_root->highest_objectid);
3038         if (ret) {
3039                 mutex_unlock(&tree_root->objectid_mutex);
3040                 goto recovery_tree_root;
3041         }
3042
3043         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3044
3045         mutex_unlock(&tree_root->objectid_mutex);
3046
3047         ret = btrfs_read_roots(fs_info);
3048         if (ret)
3049                 goto recovery_tree_root;
3050
3051         fs_info->generation = generation;
3052         fs_info->last_trans_committed = generation;
3053
3054         ret = btrfs_verify_dev_extents(fs_info);
3055         if (ret) {
3056                 btrfs_err(fs_info,
3057                           "failed to verify dev extents against chunks: %d",
3058                           ret);
3059                 goto fail_block_groups;
3060         }
3061         ret = btrfs_recover_balance(fs_info);
3062         if (ret) {
3063                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3064                 goto fail_block_groups;
3065         }
3066
3067         ret = btrfs_init_dev_stats(fs_info);
3068         if (ret) {
3069                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3070                 goto fail_block_groups;
3071         }
3072
3073         ret = btrfs_init_dev_replace(fs_info);
3074         if (ret) {
3075                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3076                 goto fail_block_groups;
3077         }
3078
3079         btrfs_free_extra_devids(fs_devices, 1);
3080
3081         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3082         if (ret) {
3083                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3084                                 ret);
3085                 goto fail_block_groups;
3086         }
3087
3088         ret = btrfs_sysfs_add_device(fs_devices);
3089         if (ret) {
3090                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3091                                 ret);
3092                 goto fail_fsdev_sysfs;
3093         }
3094
3095         ret = btrfs_sysfs_add_mounted(fs_info);
3096         if (ret) {
3097                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3098                 goto fail_fsdev_sysfs;
3099         }
3100
3101         ret = btrfs_init_space_info(fs_info);
3102         if (ret) {
3103                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3104                 goto fail_sysfs;
3105         }
3106
3107         ret = btrfs_read_block_groups(fs_info);
3108         if (ret) {
3109                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3110                 goto fail_sysfs;
3111         }
3112
3113         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3114                 btrfs_warn(fs_info,
3115                 "writable mount is not allowed due to too many missing devices");
3116                 goto fail_sysfs;
3117         }
3118
3119         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3120                                                "btrfs-cleaner");
3121         if (IS_ERR(fs_info->cleaner_kthread))
3122                 goto fail_sysfs;
3123
3124         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3125                                                    tree_root,
3126                                                    "btrfs-transaction");
3127         if (IS_ERR(fs_info->transaction_kthread))
3128                 goto fail_cleaner;
3129
3130         if (!btrfs_test_opt(fs_info, NOSSD) &&
3131             !fs_info->fs_devices->rotating) {
3132                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3133         }
3134
3135         /*
3136          * Mount does not set all options immediately, we can do it now and do
3137          * not have to wait for transaction commit
3138          */
3139         btrfs_apply_pending_changes(fs_info);
3140
3141 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3142         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3143                 ret = btrfsic_mount(fs_info, fs_devices,
3144                                     btrfs_test_opt(fs_info,
3145                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3146                                     1 : 0,
3147                                     fs_info->check_integrity_print_mask);
3148                 if (ret)
3149                         btrfs_warn(fs_info,
3150                                 "failed to initialize integrity check module: %d",
3151                                 ret);
3152         }
3153 #endif
3154         ret = btrfs_read_qgroup_config(fs_info);
3155         if (ret)
3156                 goto fail_trans_kthread;
3157
3158         if (btrfs_build_ref_tree(fs_info))
3159                 btrfs_err(fs_info, "couldn't build ref tree");
3160
3161         /* do not make disk changes in broken FS or nologreplay is given */
3162         if (btrfs_super_log_root(disk_super) != 0 &&
3163             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3164                 ret = btrfs_replay_log(fs_info, fs_devices);
3165                 if (ret) {
3166                         err = ret;
3167                         goto fail_qgroup;
3168                 }
3169         }
3170
3171         ret = btrfs_find_orphan_roots(fs_info);
3172         if (ret)
3173                 goto fail_qgroup;
3174
3175         if (!sb_rdonly(sb)) {
3176                 ret = btrfs_cleanup_fs_roots(fs_info);
3177                 if (ret)
3178                         goto fail_qgroup;
3179
3180                 mutex_lock(&fs_info->cleaner_mutex);
3181                 ret = btrfs_recover_relocation(tree_root);
3182                 mutex_unlock(&fs_info->cleaner_mutex);
3183                 if (ret < 0) {
3184                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3185                                         ret);
3186                         err = -EINVAL;
3187                         goto fail_qgroup;
3188                 }
3189         }
3190
3191         location.objectid = BTRFS_FS_TREE_OBJECTID;
3192         location.type = BTRFS_ROOT_ITEM_KEY;
3193         location.offset = 0;
3194
3195         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3196         if (IS_ERR(fs_info->fs_root)) {
3197                 err = PTR_ERR(fs_info->fs_root);
3198                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3199                 goto fail_qgroup;
3200         }
3201
3202         if (sb_rdonly(sb))
3203                 return 0;
3204
3205         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3206             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3207                 clear_free_space_tree = 1;
3208         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3209                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3210                 btrfs_warn(fs_info, "free space tree is invalid");
3211                 clear_free_space_tree = 1;
3212         }
3213
3214         if (clear_free_space_tree) {
3215                 btrfs_info(fs_info, "clearing free space tree");
3216                 ret = btrfs_clear_free_space_tree(fs_info);
3217                 if (ret) {
3218                         btrfs_warn(fs_info,
3219                                    "failed to clear free space tree: %d", ret);
3220                         close_ctree(fs_info);
3221                         return ret;
3222                 }
3223         }
3224
3225         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3226             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3227                 btrfs_info(fs_info, "creating free space tree");
3228                 ret = btrfs_create_free_space_tree(fs_info);
3229                 if (ret) {
3230                         btrfs_warn(fs_info,
3231                                 "failed to create free space tree: %d", ret);
3232                         close_ctree(fs_info);
3233                         return ret;
3234                 }
3235         }
3236
3237         down_read(&fs_info->cleanup_work_sem);
3238         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3239             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3240                 up_read(&fs_info->cleanup_work_sem);
3241                 close_ctree(fs_info);
3242                 return ret;
3243         }
3244         up_read(&fs_info->cleanup_work_sem);
3245
3246         ret = btrfs_resume_balance_async(fs_info);
3247         if (ret) {
3248                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3249                 close_ctree(fs_info);
3250                 return ret;
3251         }
3252
3253         ret = btrfs_resume_dev_replace_async(fs_info);
3254         if (ret) {
3255                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3256                 close_ctree(fs_info);
3257   &nb