btrfs: get fs_info from eb in btrfs_check_node
[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 of a btree block and store the result to provided buffer.
264  *
265  * Returns error if the extent buffer cannot be mapped.
266  */
267 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
268 {
269         unsigned long len;
270         unsigned long cur_len;
271         unsigned long offset = BTRFS_CSUM_SIZE;
272         char *kaddr;
273         unsigned long map_start;
274         unsigned long map_len;
275         int err;
276         u32 crc = ~(u32)0;
277
278         len = buf->len - offset;
279         while (len > 0) {
280                 /*
281                  * Note: we don't need to check for the err == 1 case here, as
282                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
283                  * and 'min_len = 32' and the currently implemented mapping
284                  * algorithm we cannot cross a page boundary.
285                  */
286                 err = map_private_extent_buffer(buf, offset, 32,
287                                         &kaddr, &map_start, &map_len);
288                 if (WARN_ON(err))
289                         return err;
290                 cur_len = min(len, map_len - (offset - map_start));
291                 crc = btrfs_csum_data(kaddr + offset - map_start,
292                                       crc, cur_len);
293                 len -= cur_len;
294                 offset += cur_len;
295         }
296         memset(result, 0, BTRFS_CSUM_SIZE);
297
298         btrfs_csum_final(crc, result);
299
300         return 0;
301 }
302
303 /*
304  * we can't consider a given block up to date unless the transid of the
305  * block matches the transid in the parent node's pointer.  This is how we
306  * detect blocks that either didn't get written at all or got written
307  * in the wrong place.
308  */
309 static int verify_parent_transid(struct extent_io_tree *io_tree,
310                                  struct extent_buffer *eb, u64 parent_transid,
311                                  int atomic)
312 {
313         struct extent_state *cached_state = NULL;
314         int ret;
315         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
316
317         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
318                 return 0;
319
320         if (atomic)
321                 return -EAGAIN;
322
323         if (need_lock) {
324                 btrfs_tree_read_lock(eb);
325                 btrfs_set_lock_blocking_read(eb);
326         }
327
328         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
329                          &cached_state);
330         if (extent_buffer_uptodate(eb) &&
331             btrfs_header_generation(eb) == parent_transid) {
332                 ret = 0;
333                 goto out;
334         }
335         btrfs_err_rl(eb->fs_info,
336                 "parent transid verify failed on %llu wanted %llu found %llu",
337                         eb->start,
338                         parent_transid, btrfs_header_generation(eb));
339         ret = 1;
340
341         /*
342          * Things reading via commit roots that don't have normal protection,
343          * like send, can have a really old block in cache that may point at a
344          * block that has been freed and re-allocated.  So don't clear uptodate
345          * if we find an eb that is under IO (dirty/writeback) because we could
346          * end up reading in the stale data and then writing it back out and
347          * making everybody very sad.
348          */
349         if (!extent_buffer_under_io(eb))
350                 clear_extent_buffer_uptodate(eb);
351 out:
352         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353                              &cached_state);
354         if (need_lock)
355                 btrfs_tree_read_unlock_blocking(eb);
356         return ret;
357 }
358
359 /*
360  * Return 0 if the superblock checksum type matches the checksum value of that
361  * algorithm. Pass the raw disk superblock data.
362  */
363 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
364                                   char *raw_disk_sb)
365 {
366         struct btrfs_super_block *disk_sb =
367                 (struct btrfs_super_block *)raw_disk_sb;
368         u16 csum_type = btrfs_super_csum_type(disk_sb);
369         int ret = 0;
370
371         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
372                 u32 crc = ~(u32)0;
373                 char result[sizeof(crc)];
374
375                 /*
376                  * The super_block structure does not span the whole
377                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378                  * is filled with zeros and is included in the checksum.
379                  */
380                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382                 btrfs_csum_final(crc, result);
383
384                 if (memcmp(raw_disk_sb, result, sizeof(result)))
385                         ret = 1;
386         }
387
388         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
389                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
390                                 csum_type);
391                 ret = 1;
392         }
393
394         return ret;
395 }
396
397 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
398                            struct btrfs_key *first_key, u64 parent_transid)
399 {
400         struct btrfs_fs_info *fs_info = eb->fs_info;
401         int found_level;
402         struct btrfs_key found_key;
403         int ret;
404
405         found_level = btrfs_header_level(eb);
406         if (found_level != level) {
407                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
408                      KERN_ERR "BTRFS: tree level check failed\n");
409                 btrfs_err(fs_info,
410 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
411                           eb->start, level, found_level);
412                 return -EIO;
413         }
414
415         if (!first_key)
416                 return 0;
417
418         /*
419          * For live tree block (new tree blocks in current transaction),
420          * we need proper lock context to avoid race, which is impossible here.
421          * So we only checks tree blocks which is read from disk, whose
422          * generation <= fs_info->last_trans_committed.
423          */
424         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
425                 return 0;
426         if (found_level)
427                 btrfs_node_key_to_cpu(eb, &found_key, 0);
428         else
429                 btrfs_item_key_to_cpu(eb, &found_key, 0);
430         ret = btrfs_comp_cpu_keys(first_key, &found_key);
431
432         if (ret) {
433                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
434                      KERN_ERR "BTRFS: tree first key check failed\n");
435                 btrfs_err(fs_info,
436 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
437                           eb->start, parent_transid, first_key->objectid,
438                           first_key->type, first_key->offset,
439                           found_key.objectid, found_key.type,
440                           found_key.offset);
441         }
442         return ret;
443 }
444
445 /*
446  * helper to read a given tree block, doing retries as required when
447  * the checksums don't match and we have alternate mirrors to try.
448  *
449  * @parent_transid:     expected transid, skip check if 0
450  * @level:              expected level, mandatory check
451  * @first_key:          expected key of first slot, skip check if NULL
452  */
453 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
454                                           u64 parent_transid, int level,
455                                           struct btrfs_key *first_key)
456 {
457         struct btrfs_fs_info *fs_info = eb->fs_info;
458         struct extent_io_tree *io_tree;
459         int failed = 0;
460         int ret;
461         int num_copies = 0;
462         int mirror_num = 0;
463         int failed_mirror = 0;
464
465         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
466         while (1) {
467                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
468                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
469                                                mirror_num);
470                 if (!ret) {
471                         if (verify_parent_transid(io_tree, eb,
472                                                    parent_transid, 0))
473                                 ret = -EIO;
474                         else if (btrfs_verify_level_key(eb, level,
475                                                 first_key, parent_transid))
476                                 ret = -EUCLEAN;
477                         else
478                                 break;
479                 }
480
481                 num_copies = btrfs_num_copies(fs_info,
482                                               eb->start, eb->len);
483                 if (num_copies == 1)
484                         break;
485
486                 if (!failed_mirror) {
487                         failed = 1;
488                         failed_mirror = eb->read_mirror;
489                 }
490
491                 mirror_num++;
492                 if (mirror_num == failed_mirror)
493                         mirror_num++;
494
495                 if (mirror_num > num_copies)
496                         break;
497         }
498
499         if (failed && !ret && failed_mirror)
500                 btrfs_repair_eb_io_failure(eb, failed_mirror);
501
502         return ret;
503 }
504
505 /*
506  * checksum a dirty tree block before IO.  This has extra checks to make sure
507  * we only fill in the checksum field in the first page of a multi-page block
508  */
509
510 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
511 {
512         u64 start = page_offset(page);
513         u64 found_start;
514         u8 result[BTRFS_CSUM_SIZE];
515         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
516         struct extent_buffer *eb;
517
518         eb = (struct extent_buffer *)page->private;
519         if (page != eb->pages[0])
520                 return 0;
521
522         found_start = btrfs_header_bytenr(eb);
523         /*
524          * Please do not consolidate these warnings into a single if.
525          * It is useful to know what went wrong.
526          */
527         if (WARN_ON(found_start != start))
528                 return -EUCLEAN;
529         if (WARN_ON(!PageUptodate(page)))
530                 return -EUCLEAN;
531
532         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
533                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
534
535         if (csum_tree_block(eb, result))
536                 return -EINVAL;
537
538         write_extent_buffer(eb, result, 0, csum_size);
539         return 0;
540 }
541
542 static int check_tree_block_fsid(struct extent_buffer *eb)
543 {
544         struct btrfs_fs_info *fs_info = eb->fs_info;
545         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
546         u8 fsid[BTRFS_FSID_SIZE];
547         int ret = 1;
548
549         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
550         while (fs_devices) {
551                 u8 *metadata_uuid;
552
553                 /*
554                  * Checking the incompat flag is only valid for the current
555                  * fs. For seed devices it's forbidden to have their uuid
556                  * changed so reading ->fsid in this case is fine
557                  */
558                 if (fs_devices == fs_info->fs_devices &&
559                     btrfs_fs_incompat(fs_info, METADATA_UUID))
560                         metadata_uuid = fs_devices->metadata_uuid;
561                 else
562                         metadata_uuid = fs_devices->fsid;
563
564                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
565                         ret = 0;
566                         break;
567                 }
568                 fs_devices = fs_devices->seed;
569         }
570         return ret;
571 }
572
573 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
574                                       u64 phy_offset, struct page *page,
575                                       u64 start, u64 end, int mirror)
576 {
577         u64 found_start;
578         int found_level;
579         struct extent_buffer *eb;
580         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
581         struct btrfs_fs_info *fs_info = root->fs_info;
582         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
583         int ret = 0;
584         u8 result[BTRFS_CSUM_SIZE];
585         int reads_done;
586
587         if (!page->private)
588                 goto out;
589
590         eb = (struct extent_buffer *)page->private;
591
592         /* the pending IO might have been the only thing that kept this buffer
593          * in memory.  Make sure we have a ref for all this other checks
594          */
595         extent_buffer_get(eb);
596
597         reads_done = atomic_dec_and_test(&eb->io_pages);
598         if (!reads_done)
599                 goto err;
600
601         eb->read_mirror = mirror;
602         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
603                 ret = -EIO;
604                 goto err;
605         }
606
607         found_start = btrfs_header_bytenr(eb);
608         if (found_start != eb->start) {
609                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
610                              eb->start, found_start);
611                 ret = -EIO;
612                 goto err;
613         }
614         if (check_tree_block_fsid(eb)) {
615                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
616                              eb->start);
617                 ret = -EIO;
618                 goto err;
619         }
620         found_level = btrfs_header_level(eb);
621         if (found_level >= BTRFS_MAX_LEVEL) {
622                 btrfs_err(fs_info, "bad tree block level %d on %llu",
623                           (int)btrfs_header_level(eb), eb->start);
624                 ret = -EIO;
625                 goto err;
626         }
627
628         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
629                                        eb, found_level);
630
631         ret = csum_tree_block(eb, result);
632         if (ret)
633                 goto err;
634
635         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
636                 u32 val;
637                 u32 found = 0;
638
639                 memcpy(&found, result, csum_size);
640
641                 read_extent_buffer(eb, &val, 0, csum_size);
642                 btrfs_warn_rl(fs_info,
643                 "%s checksum verify failed on %llu wanted %x found %x level %d",
644                               fs_info->sb->s_id, eb->start,
645                               val, found, btrfs_header_level(eb));
646                 ret = -EUCLEAN;
647                 goto err;
648         }
649
650         /*
651          * If this is a leaf block and it is corrupt, set the corrupt bit so
652          * that we don't try and read the other copies of this block, just
653          * return -EIO.
654          */
655         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
656                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
657                 ret = -EIO;
658         }
659
660         if (found_level > 0 && btrfs_check_node(eb))
661                 ret = -EIO;
662
663         if (!ret)
664                 set_extent_buffer_uptodate(eb);
665         else
666                 btrfs_err(fs_info,
667                           "block=%llu read time tree block corruption detected",
668                           eb->start);
669 err:
670         if (reads_done &&
671             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
672                 btree_readahead_hook(eb, ret);
673
674         if (ret) {
675                 /*
676                  * our io error hook is going to dec the io pages
677                  * again, we have to make sure it has something
678                  * to decrement
679                  */
680                 atomic_inc(&eb->io_pages);
681                 clear_extent_buffer_uptodate(eb);
682         }
683         free_extent_buffer(eb);
684 out:
685         return ret;
686 }
687
688 static void end_workqueue_bio(struct bio *bio)
689 {
690         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
691         struct btrfs_fs_info *fs_info;
692         struct btrfs_workqueue *wq;
693         btrfs_work_func_t func;
694
695         fs_info = end_io_wq->info;
696         end_io_wq->status = bio->bi_status;
697
698         if (bio_op(bio) == REQ_OP_WRITE) {
699                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
700                         wq = fs_info->endio_meta_write_workers;
701                         func = btrfs_endio_meta_write_helper;
702                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
703                         wq = fs_info->endio_freespace_worker;
704                         func = btrfs_freespace_write_helper;
705                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
706                         wq = fs_info->endio_raid56_workers;
707                         func = btrfs_endio_raid56_helper;
708                 } else {
709                         wq = fs_info->endio_write_workers;
710                         func = btrfs_endio_write_helper;
711                 }
712         } else {
713                 if (unlikely(end_io_wq->metadata ==
714                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
715                         wq = fs_info->endio_repair_workers;
716                         func = btrfs_endio_repair_helper;
717                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
718                         wq = fs_info->endio_raid56_workers;
719                         func = btrfs_endio_raid56_helper;
720                 } else if (end_io_wq->metadata) {
721                         wq = fs_info->endio_meta_workers;
722                         func = btrfs_endio_meta_helper;
723                 } else {
724                         wq = fs_info->endio_workers;
725                         func = btrfs_endio_helper;
726                 }
727         }
728
729         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
730         btrfs_queue_work(wq, &end_io_wq->work);
731 }
732
733 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
734                         enum btrfs_wq_endio_type metadata)
735 {
736         struct btrfs_end_io_wq *end_io_wq;
737
738         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
739         if (!end_io_wq)
740                 return BLK_STS_RESOURCE;
741
742         end_io_wq->private = bio->bi_private;
743         end_io_wq->end_io = bio->bi_end_io;
744         end_io_wq->info = info;
745         end_io_wq->status = 0;
746         end_io_wq->bio = bio;
747         end_io_wq->metadata = metadata;
748
749         bio->bi_private = end_io_wq;
750         bio->bi_end_io = end_workqueue_bio;
751         return 0;
752 }
753
754 static void run_one_async_start(struct btrfs_work *work)
755 {
756         struct async_submit_bio *async;
757         blk_status_t ret;
758
759         async = container_of(work, struct  async_submit_bio, work);
760         ret = async->submit_bio_start(async->private_data, async->bio,
761                                       async->bio_offset);
762         if (ret)
763                 async->status = ret;
764 }
765
766 /*
767  * In order to insert checksums into the metadata in large chunks, we wait
768  * until bio submission time.   All the pages in the bio are checksummed and
769  * sums are attached onto the ordered extent record.
770  *
771  * At IO completion time the csums attached on the ordered extent record are
772  * inserted into the tree.
773  */
774 static void run_one_async_done(struct btrfs_work *work)
775 {
776         struct async_submit_bio *async;
777         struct inode *inode;
778         blk_status_t ret;
779
780         async = container_of(work, struct  async_submit_bio, work);
781         inode = async->private_data;
782
783         /* If an error occurred we just want to clean up the bio and move on */
784         if (async->status) {
785                 async->bio->bi_status = async->status;
786                 bio_endio(async->bio);
787                 return;
788         }
789
790         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
791                         async->mirror_num, 1);
792         if (ret) {
793                 async->bio->bi_status = ret;
794                 bio_endio(async->bio);
795         }
796 }
797
798 static void run_one_async_free(struct btrfs_work *work)
799 {
800         struct async_submit_bio *async;
801
802         async = container_of(work, struct  async_submit_bio, work);
803         kfree(async);
804 }
805
806 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
807                                  int mirror_num, unsigned long bio_flags,
808                                  u64 bio_offset, void *private_data,
809                                  extent_submit_bio_start_t *submit_bio_start)
810 {
811         struct async_submit_bio *async;
812
813         async = kmalloc(sizeof(*async), GFP_NOFS);
814         if (!async)
815                 return BLK_STS_RESOURCE;
816
817         async->private_data = private_data;
818         async->bio = bio;
819         async->mirror_num = mirror_num;
820         async->submit_bio_start = submit_bio_start;
821
822         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
823                         run_one_async_done, run_one_async_free);
824
825         async->bio_offset = bio_offset;
826
827         async->status = 0;
828
829         if (op_is_sync(bio->bi_opf))
830                 btrfs_set_work_high_priority(&async->work);
831
832         btrfs_queue_work(fs_info->workers, &async->work);
833         return 0;
834 }
835
836 static blk_status_t btree_csum_one_bio(struct bio *bio)
837 {
838         struct bio_vec *bvec;
839         struct btrfs_root *root;
840         int i, ret = 0;
841         struct bvec_iter_all iter_all;
842
843         ASSERT(!bio_flagged(bio, BIO_CLONED));
844         bio_for_each_segment_all(bvec, bio, i, iter_all) {
845                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
846                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
847                 if (ret)
848                         break;
849         }
850
851         return errno_to_blk_status(ret);
852 }
853
854 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
855                                              u64 bio_offset)
856 {
857         /*
858          * when we're called for a write, we're already in the async
859          * submission context.  Just jump into btrfs_map_bio
860          */
861         return btree_csum_one_bio(bio);
862 }
863
864 static int check_async_write(struct btrfs_inode *bi)
865 {
866         if (atomic_read(&bi->sync_writers))
867                 return 0;
868 #ifdef CONFIG_X86
869         if (static_cpu_has(X86_FEATURE_XMM4_2))
870                 return 0;
871 #endif
872         return 1;
873 }
874
875 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
876                                           int mirror_num, unsigned long bio_flags,
877                                           u64 bio_offset)
878 {
879         struct inode *inode = private_data;
880         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
881         int async = check_async_write(BTRFS_I(inode));
882         blk_status_t ret;
883
884         if (bio_op(bio) != REQ_OP_WRITE) {
885                 /*
886                  * called for a read, do the setup so that checksum validation
887                  * can happen in the async kernel threads
888                  */
889                 ret = btrfs_bio_wq_end_io(fs_info, bio,
890                                           BTRFS_WQ_ENDIO_METADATA);
891                 if (ret)
892                         goto out_w_error;
893                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
894         } else if (!async) {
895                 ret = btree_csum_one_bio(bio);
896                 if (ret)
897                         goto out_w_error;
898                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
899         } else {
900                 /*
901                  * kthread helpers are used to submit writes so that
902                  * checksumming can happen in parallel across all CPUs
903                  */
904                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
905                                           bio_offset, private_data,
906                                           btree_submit_bio_start);
907         }
908
909         if (ret)
910                 goto out_w_error;
911         return 0;
912
913 out_w_error:
914         bio->bi_status = ret;
915         bio_endio(bio);
916         return ret;
917 }
918
919 #ifdef CONFIG_MIGRATION
920 static int btree_migratepage(struct address_space *mapping,
921                         struct page *newpage, struct page *page,
922                         enum migrate_mode mode)
923 {
924         /*
925          * we can't safely write a btree page from here,
926          * we haven't done the locking hook
927          */
928         if (PageDirty(page))
929                 return -EAGAIN;
930         /*
931          * Buffers may be managed in a filesystem specific way.
932          * We must have no buffers or drop them.
933          */
934         if (page_has_private(page) &&
935             !try_to_release_page(page, GFP_KERNEL))
936                 return -EAGAIN;
937         return migrate_page(mapping, newpage, page, mode);
938 }
939 #endif
940
941
942 static int btree_writepages(struct address_space *mapping,
943                             struct writeback_control *wbc)
944 {
945         struct btrfs_fs_info *fs_info;
946         int ret;
947
948         if (wbc->sync_mode == WB_SYNC_NONE) {
949
950                 if (wbc->for_kupdate)
951                         return 0;
952
953                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
954                 /* this is a bit racy, but that's ok */
955                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
956                                              BTRFS_DIRTY_METADATA_THRESH,
957                                              fs_info->dirty_metadata_batch);
958                 if (ret < 0)
959                         return 0;
960         }
961         return btree_write_cache_pages(mapping, wbc);
962 }
963
964 static int btree_readpage(struct file *file, struct page *page)
965 {
966         struct extent_io_tree *tree;
967         tree = &BTRFS_I(page->mapping->host)->io_tree;
968         return extent_read_full_page(tree, page, btree_get_extent, 0);
969 }
970
971 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
972 {
973         if (PageWriteback(page) || PageDirty(page))
974                 return 0;
975
976         return try_release_extent_buffer(page);
977 }
978
979 static void btree_invalidatepage(struct page *page, unsigned int offset,
980                                  unsigned int length)
981 {
982         struct extent_io_tree *tree;
983         tree = &BTRFS_I(page->mapping->host)->io_tree;
984         extent_invalidatepage(tree, page, offset);
985         btree_releasepage(page, GFP_NOFS);
986         if (PagePrivate(page)) {
987                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
988                            "page private not zero on page %llu",
989                            (unsigned long long)page_offset(page));
990                 ClearPagePrivate(page);
991                 set_page_private(page, 0);
992                 put_page(page);
993         }
994 }
995
996 static int btree_set_page_dirty(struct page *page)
997 {
998 #ifdef DEBUG
999         struct extent_buffer *eb;
1000
1001         BUG_ON(!PagePrivate(page));
1002         eb = (struct extent_buffer *)page->private;
1003         BUG_ON(!eb);
1004         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1005         BUG_ON(!atomic_read(&eb->refs));
1006         btrfs_assert_tree_locked(eb);
1007 #endif
1008         return __set_page_dirty_nobuffers(page);
1009 }
1010
1011 static const struct address_space_operations btree_aops = {
1012         .readpage       = btree_readpage,
1013         .writepages     = btree_writepages,
1014         .releasepage    = btree_releasepage,
1015         .invalidatepage = btree_invalidatepage,
1016 #ifdef CONFIG_MIGRATION
1017         .migratepage    = btree_migratepage,
1018 #endif
1019         .set_page_dirty = btree_set_page_dirty,
1020 };
1021
1022 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1023 {
1024         struct extent_buffer *buf = NULL;
1025         struct inode *btree_inode = fs_info->btree_inode;
1026         int ret;
1027
1028         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1029         if (IS_ERR(buf))
1030                 return;
1031
1032         ret = read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, buf,
1033                         WAIT_NONE, 0);
1034         if (ret < 0)
1035                 free_extent_buffer_stale(buf);
1036         else
1037                 free_extent_buffer(buf);
1038 }
1039
1040 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1041                          int mirror_num, struct extent_buffer **eb)
1042 {
1043         struct extent_buffer *buf = NULL;
1044         struct inode *btree_inode = fs_info->btree_inode;
1045         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1046         int ret;
1047
1048         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1049         if (IS_ERR(buf))
1050                 return 0;
1051
1052         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1053
1054         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1055                                        mirror_num);
1056         if (ret) {
1057                 free_extent_buffer_stale(buf);
1058                 return ret;
1059         }
1060
1061         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1062                 free_extent_buffer_stale(buf);
1063                 return -EIO;
1064         } else if (extent_buffer_uptodate(buf)) {
1065                 *eb = buf;
1066         } else {
1067                 free_extent_buffer(buf);
1068         }
1069         return 0;
1070 }
1071
1072 struct extent_buffer *btrfs_find_create_tree_block(
1073                                                 struct btrfs_fs_info *fs_info,
1074                                                 u64 bytenr)
1075 {
1076         if (btrfs_is_testing(fs_info))
1077                 return alloc_test_extent_buffer(fs_info, bytenr);
1078         return alloc_extent_buffer(fs_info, bytenr);
1079 }
1080
1081 /*
1082  * Read tree block at logical address @bytenr and do variant basic but critical
1083  * verification.
1084  *
1085  * @parent_transid:     expected transid of this tree block, skip check if 0
1086  * @level:              expected level, mandatory check
1087  * @first_key:          expected key in slot 0, skip check if NULL
1088  */
1089 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1090                                       u64 parent_transid, int level,
1091                                       struct btrfs_key *first_key)
1092 {
1093         struct extent_buffer *buf = NULL;
1094         int ret;
1095
1096         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1097         if (IS_ERR(buf))
1098                 return buf;
1099
1100         ret = btree_read_extent_buffer_pages(buf, parent_transid,
1101                                              level, first_key);
1102         if (ret) {
1103                 free_extent_buffer_stale(buf);
1104                 return ERR_PTR(ret);
1105         }
1106         return buf;
1107
1108 }
1109
1110 void btrfs_clean_tree_block(struct extent_buffer *buf)
1111 {
1112         struct btrfs_fs_info *fs_info = buf->fs_info;
1113         if (btrfs_header_generation(buf) ==
1114             fs_info->running_transaction->transid) {
1115                 btrfs_assert_tree_locked(buf);
1116
1117                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1118                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1119                                                  -buf->len,
1120                                                  fs_info->dirty_metadata_batch);
1121                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1122                         btrfs_set_lock_blocking_write(buf);
1123                         clear_extent_buffer_dirty(buf);
1124                 }
1125         }
1126 }
1127
1128 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1129 {
1130         struct btrfs_subvolume_writers *writers;
1131         int ret;
1132
1133         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1134         if (!writers)
1135                 return ERR_PTR(-ENOMEM);
1136
1137         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1138         if (ret < 0) {
1139                 kfree(writers);
1140                 return ERR_PTR(ret);
1141         }
1142
1143         init_waitqueue_head(&writers->wait);
1144         return writers;
1145 }
1146
1147 static void
1148 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1149 {
1150         percpu_counter_destroy(&writers->counter);
1151         kfree(writers);
1152 }
1153
1154 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1155                          u64 objectid)
1156 {
1157         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1158         root->node = NULL;
1159         root->commit_root = NULL;
1160         root->state = 0;
1161         root->orphan_cleanup_state = 0;
1162
1163         root->last_trans = 0;
1164         root->highest_objectid = 0;
1165         root->nr_delalloc_inodes = 0;
1166         root->nr_ordered_extents = 0;
1167         root->inode_tree = RB_ROOT;
1168         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1169         root->block_rsv = NULL;
1170
1171         INIT_LIST_HEAD(&root->dirty_list);
1172         INIT_LIST_HEAD(&root->root_list);
1173         INIT_LIST_HEAD(&root->delalloc_inodes);
1174         INIT_LIST_HEAD(&root->delalloc_root);
1175         INIT_LIST_HEAD(&root->ordered_extents);
1176         INIT_LIST_HEAD(&root->ordered_root);
1177         INIT_LIST_HEAD(&root->reloc_dirty_list);
1178         INIT_LIST_HEAD(&root->logged_list[0]);
1179         INIT_LIST_HEAD(&root->logged_list[1]);
1180         spin_lock_init(&root->inode_lock);
1181         spin_lock_init(&root->delalloc_lock);
1182         spin_lock_init(&root->ordered_extent_lock);
1183         spin_lock_init(&root->accounting_lock);
1184         spin_lock_init(&root->log_extents_lock[0]);
1185         spin_lock_init(&root->log_extents_lock[1]);
1186         spin_lock_init(&root->qgroup_meta_rsv_lock);
1187         mutex_init(&root->objectid_mutex);
1188         mutex_init(&root->log_mutex);
1189         mutex_init(&root->ordered_extent_mutex);
1190         mutex_init(&root->delalloc_mutex);
1191         init_waitqueue_head(&root->log_writer_wait);
1192         init_waitqueue_head(&root->log_commit_wait[0]);
1193         init_waitqueue_head(&root->log_commit_wait[1]);
1194         INIT_LIST_HEAD(&root->log_ctxs[0]);
1195         INIT_LIST_HEAD(&root->log_ctxs[1]);
1196         atomic_set(&root->log_commit[0], 0);
1197         atomic_set(&root->log_commit[1], 0);
1198         atomic_set(&root->log_writers, 0);
1199         atomic_set(&root->log_batch, 0);
1200         refcount_set(&root->refs, 1);
1201         atomic_set(&root->will_be_snapshotted, 0);
1202         atomic_set(&root->snapshot_force_cow, 0);
1203         atomic_set(&root->nr_swapfiles, 0);
1204         root->log_transid = 0;
1205         root->log_transid_committed = -1;
1206         root->last_log_commit = 0;
1207         if (!dummy)
1208                 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1209                                     IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1210
1211         memset(&root->root_key, 0, sizeof(root->root_key));
1212         memset(&root->root_item, 0, sizeof(root->root_item));
1213         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1214         if (!dummy)
1215                 root->defrag_trans_start = fs_info->generation;
1216         else
1217                 root->defrag_trans_start = 0;
1218         root->root_key.objectid = objectid;
1219         root->anon_dev = 0;
1220
1221         spin_lock_init(&root->root_item_lock);
1222         btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
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                 btrfs_run_delayed_iputs(fs_info);
1720
1721                 again = btrfs_clean_one_deleted_snapshot(root);
1722                 mutex_unlock(&fs_info->cleaner_mutex);
1723
1724                 /*
1725                  * The defragger has dealt with the R/O remount and umount,
1726                  * needn't do anything special here.
1727                  */
1728                 btrfs_run_defrag_inodes(fs_info);
1729
1730                 /*
1731                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1732                  * with relocation (btrfs_relocate_chunk) and relocation
1733                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1734                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1735                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1736                  * unused block groups.
1737                  */
1738                 btrfs_delete_unused_bgs(fs_info);
1739 sleep:
1740                 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1741                 if (kthread_should_park())
1742                         kthread_parkme();
1743                 if (kthread_should_stop())
1744                         return 0;
1745                 if (!again) {
1746                         set_current_state(TASK_INTERRUPTIBLE);
1747                         schedule();
1748                         __set_current_state(TASK_RUNNING);
1749                 }
1750         }
1751 }
1752
1753 static int transaction_kthread(void *arg)
1754 {
1755         struct btrfs_root *root = arg;
1756         struct btrfs_fs_info *fs_info = root->fs_info;
1757         struct btrfs_trans_handle *trans;
1758         struct btrfs_transaction *cur;
1759         u64 transid;
1760         time64_t now;
1761         unsigned long delay;
1762         bool cannot_commit;
1763
1764         do {
1765                 cannot_commit = false;
1766                 delay = HZ * fs_info->commit_interval;
1767                 mutex_lock(&fs_info->transaction_kthread_mutex);
1768
1769                 spin_lock(&fs_info->trans_lock);
1770                 cur = fs_info->running_transaction;
1771                 if (!cur) {
1772                         spin_unlock(&fs_info->trans_lock);
1773                         goto sleep;
1774                 }
1775
1776                 now = ktime_get_seconds();
1777                 if (cur->state < TRANS_STATE_BLOCKED &&
1778                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1779                     (now < cur->start_time ||
1780                      now - cur->start_time < fs_info->commit_interval)) {
1781                         spin_unlock(&fs_info->trans_lock);
1782                         delay = HZ * 5;
1783                         goto sleep;
1784                 }
1785                 transid = cur->transid;
1786                 spin_unlock(&fs_info->trans_lock);
1787
1788                 /* If the file system is aborted, this will always fail. */
1789                 trans = btrfs_attach_transaction(root);
1790                 if (IS_ERR(trans)) {
1791                         if (PTR_ERR(trans) != -ENOENT)
1792                                 cannot_commit = true;
1793                         goto sleep;
1794                 }
1795                 if (transid == trans->transid) {
1796                         btrfs_commit_transaction(trans);
1797                 } else {
1798                         btrfs_end_transaction(trans);
1799                 }
1800 sleep:
1801                 wake_up_process(fs_info->cleaner_kthread);
1802                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1803
1804                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1805                                       &fs_info->fs_state)))
1806                         btrfs_cleanup_transaction(fs_info);
1807                 if (!kthread_should_stop() &&
1808                                 (!btrfs_transaction_blocked(fs_info) ||
1809                                  cannot_commit))
1810                         schedule_timeout_interruptible(delay);
1811         } while (!kthread_should_stop());
1812         return 0;
1813 }
1814
1815 /*
1816  * this will find the highest generation in the array of
1817  * root backups.  The index of the highest array is returned,
1818  * or -1 if we can't find anything.
1819  *
1820  * We check to make sure the array is valid by comparing the
1821  * generation of the latest  root in the array with the generation
1822  * in the super block.  If they don't match we pitch it.
1823  */
1824 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1825 {
1826         u64 cur;
1827         int newest_index = -1;
1828         struct btrfs_root_backup *root_backup;
1829         int i;
1830
1831         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1832                 root_backup = info->super_copy->super_roots + i;
1833                 cur = btrfs_backup_tree_root_gen(root_backup);
1834                 if (cur == newest_gen)
1835                         newest_index = i;
1836         }
1837
1838         /* check to see if we actually wrapped around */
1839         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1840                 root_backup = info->super_copy->super_roots;
1841                 cur = btrfs_backup_tree_root_gen(root_backup);
1842                 if (cur == newest_gen)
1843                         newest_index = 0;
1844         }
1845         return newest_index;
1846 }
1847
1848
1849 /*
1850  * find the oldest backup so we know where to store new entries
1851  * in the backup array.  This will set the backup_root_index
1852  * field in the fs_info struct
1853  */
1854 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1855                                      u64 newest_gen)
1856 {
1857         int newest_index = -1;
1858
1859         newest_index = find_newest_super_backup(info, newest_gen);
1860         /* if there was garbage in there, just move along */
1861         if (newest_index == -1) {
1862                 info->backup_root_index = 0;
1863         } else {
1864                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1865         }
1866 }
1867
1868 /*
1869  * copy all the root pointers into the super backup array.
1870  * this will bump the backup pointer by one when it is
1871  * done
1872  */
1873 static void backup_super_roots(struct btrfs_fs_info *info)
1874 {
1875         int next_backup;
1876         struct btrfs_root_backup *root_backup;
1877         int last_backup;
1878
1879         next_backup = info->backup_root_index;
1880         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1881                 BTRFS_NUM_BACKUP_ROOTS;
1882
1883         /*
1884          * just overwrite the last backup if we're at the same generation
1885          * this happens only at umount
1886          */
1887         root_backup = info->super_for_commit->super_roots + last_backup;
1888         if (btrfs_backup_tree_root_gen(root_backup) ==
1889             btrfs_header_generation(info->tree_root->node))
1890                 next_backup = last_backup;
1891
1892         root_backup = info->super_for_commit->super_roots + next_backup;
1893
1894         /*
1895          * make sure all of our padding and empty slots get zero filled
1896          * regardless of which ones we use today
1897          */
1898         memset(root_backup, 0, sizeof(*root_backup));
1899
1900         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1901
1902         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1903         btrfs_set_backup_tree_root_gen(root_backup,
1904                                btrfs_header_generation(info->tree_root->node));
1905
1906         btrfs_set_backup_tree_root_level(root_backup,
1907                                btrfs_header_level(info->tree_root->node));
1908
1909         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1910         btrfs_set_backup_chunk_root_gen(root_backup,
1911                                btrfs_header_generation(info->chunk_root->node));
1912         btrfs_set_backup_chunk_root_level(root_backup,
1913                                btrfs_header_level(info->chunk_root->node));
1914
1915         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1916         btrfs_set_backup_extent_root_gen(root_backup,
1917                                btrfs_header_generation(info->extent_root->node));
1918         btrfs_set_backup_extent_root_level(root_backup,
1919                                btrfs_header_level(info->extent_root->node));
1920
1921         /*
1922          * we might commit during log recovery, which happens before we set
1923          * the fs_root.  Make sure it is valid before we fill it in.
1924          */
1925         if (info->fs_root && info->fs_root->node) {
1926                 btrfs_set_backup_fs_root(root_backup,
1927                                          info->fs_root->node->start);
1928                 btrfs_set_backup_fs_root_gen(root_backup,
1929                                btrfs_header_generation(info->fs_root->node));
1930                 btrfs_set_backup_fs_root_level(root_backup,
1931                                btrfs_header_level(info->fs_root->node));
1932         }
1933
1934         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1935         btrfs_set_backup_dev_root_gen(root_backup,
1936                                btrfs_header_generation(info->dev_root->node));
1937         btrfs_set_backup_dev_root_level(root_backup,
1938                                        btrfs_header_level(info->dev_root->node));
1939
1940         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1941         btrfs_set_backup_csum_root_gen(root_backup,
1942                                btrfs_header_generation(info->csum_root->node));
1943         btrfs_set_backup_csum_root_level(root_backup,
1944                                btrfs_header_level(info->csum_root->node));
1945
1946         btrfs_set_backup_total_bytes(root_backup,
1947                              btrfs_super_total_bytes(info->super_copy));
1948         btrfs_set_backup_bytes_used(root_backup,
1949                              btrfs_super_bytes_used(info->super_copy));
1950         btrfs_set_backup_num_devices(root_backup,
1951                              btrfs_super_num_devices(info->super_copy));
1952
1953         /*
1954          * if we don't copy this out to the super_copy, it won't get remembered
1955          * for the next commit
1956          */
1957         memcpy(&info->super_copy->super_roots,
1958                &info->super_for_commit->super_roots,
1959                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1960 }
1961
1962 /*
1963  * this copies info out of the root backup array and back into
1964  * the in-memory super block.  It is meant to help iterate through
1965  * the array, so you send it the number of backups you've already
1966  * tried and the last backup index you used.
1967  *
1968  * this returns -1 when it has tried all the backups
1969  */
1970 static noinline int next_root_backup(struct btrfs_fs_info *info,
1971                                      struct btrfs_super_block *super,
1972                                      int *num_backups_tried, int *backup_index)
1973 {
1974         struct btrfs_root_backup *root_backup;
1975         int newest = *backup_index;
1976
1977         if (*num_backups_tried == 0) {
1978                 u64 gen = btrfs_super_generation(super);
1979
1980                 newest = find_newest_super_backup(info, gen);
1981                 if (newest == -1)
1982                         return -1;
1983
1984                 *backup_index = newest;
1985                 *num_backups_tried = 1;
1986         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1987                 /* we've tried all the backups, all done */
1988                 return -1;
1989         } else {
1990                 /* jump to the next oldest backup */
1991                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1992                         BTRFS_NUM_BACKUP_ROOTS;
1993                 *backup_index = newest;
1994                 *num_backups_tried += 1;
1995         }
1996         root_backup = super->super_roots + newest;
1997
1998         btrfs_set_super_generation(super,
1999                                    btrfs_backup_tree_root_gen(root_backup));
2000         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2001         btrfs_set_super_root_level(super,
2002                                    btrfs_backup_tree_root_level(root_backup));
2003         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2004
2005         /*
2006          * fixme: the total bytes and num_devices need to match or we should
2007          * need a fsck
2008          */
2009         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2010         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2011         return 0;
2012 }
2013
2014 /* helper to cleanup workers */
2015 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2016 {
2017         btrfs_destroy_workqueue(fs_info->fixup_workers);
2018         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2019         btrfs_destroy_workqueue(fs_info->workers);
2020         btrfs_destroy_workqueue(fs_info->endio_workers);
2021         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2022         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2023         btrfs_destroy_workqueue(fs_info->rmw_workers);
2024         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2025         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2026         btrfs_destroy_workqueue(fs_info->submit_workers);
2027         btrfs_destroy_workqueue(fs_info->delayed_workers);
2028         btrfs_destroy_workqueue(fs_info->caching_workers);
2029         btrfs_destroy_workqueue(fs_info->readahead_workers);
2030         btrfs_destroy_workqueue(fs_info->flush_workers);
2031         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2032         btrfs_destroy_workqueue(fs_info->extent_workers);
2033         /*
2034          * Now that all other work queues are destroyed, we can safely destroy
2035          * the queues used for metadata I/O, since tasks from those other work
2036          * queues can do metadata I/O operations.
2037          */
2038         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2039         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2040 }
2041
2042 static void free_root_extent_buffers(struct btrfs_root *root)
2043 {
2044         if (root) {
2045                 free_extent_buffer(root->node);
2046                 free_extent_buffer(root->commit_root);
2047                 root->node = NULL;
2048                 root->commit_root = NULL;
2049         }
2050 }
2051
2052 /* helper to cleanup tree roots */
2053 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2054 {
2055         free_root_extent_buffers(info->tree_root);
2056
2057         free_root_extent_buffers(info->dev_root);
2058         free_root_extent_buffers(info->extent_root);
2059         free_root_extent_buffers(info->csum_root);
2060         free_root_extent_buffers(info->quota_root);
2061         free_root_extent_buffers(info->uuid_root);
2062         if (chunk_root)
2063                 free_root_extent_buffers(info->chunk_root);
2064         free_root_extent_buffers(info->free_space_root);
2065 }
2066
2067 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2068 {
2069         int ret;
2070         struct btrfs_root *gang[8];
2071         int i;
2072
2073         while (!list_empty(&fs_info->dead_roots)) {
2074                 gang[0] = list_entry(fs_info->dead_roots.next,
2075                                      struct btrfs_root, root_list);
2076                 list_del(&gang[0]->root_list);
2077
2078                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2079                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2080                 } else {
2081                         free_extent_buffer(gang[0]->node);
2082                         free_extent_buffer(gang[0]->commit_root);
2083                         btrfs_put_fs_root(gang[0]);
2084                 }
2085         }
2086
2087         while (1) {
2088                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2089                                              (void **)gang, 0,
2090                                              ARRAY_SIZE(gang));
2091                 if (!ret)
2092                         break;
2093                 for (i = 0; i < ret; i++)
2094                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2095         }
2096
2097         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2098                 btrfs_free_log_root_tree(NULL, fs_info);
2099                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2100         }
2101 }
2102
2103 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2104 {
2105         mutex_init(&fs_info->scrub_lock);
2106         atomic_set(&fs_info->scrubs_running, 0);
2107         atomic_set(&fs_info->scrub_pause_req, 0);
2108         atomic_set(&fs_info->scrubs_paused, 0);
2109         atomic_set(&fs_info->scrub_cancel_req, 0);
2110         init_waitqueue_head(&fs_info->scrub_pause_wait);
2111         refcount_set(&fs_info->scrub_workers_refcnt, 0);
2112 }
2113
2114 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2115 {
2116         spin_lock_init(&fs_info->balance_lock);
2117         mutex_init(&fs_info->balance_mutex);
2118         atomic_set(&fs_info->balance_pause_req, 0);
2119         atomic_set(&fs_info->balance_cancel_req, 0);
2120         fs_info->balance_ctl = NULL;
2121         init_waitqueue_head(&fs_info->balance_wait_q);
2122 }
2123
2124 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2125 {
2126         struct inode *inode = fs_info->btree_inode;
2127
2128         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2129         set_nlink(inode, 1);
2130         /*
2131          * we set the i_size on the btree inode to the max possible int.
2132          * the real end of the address space is determined by all of
2133          * the devices in the system
2134          */
2135         inode->i_size = OFFSET_MAX;
2136         inode->i_mapping->a_ops = &btree_aops;
2137
2138         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2139         extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2140                             IO_TREE_INODE_IO, inode);
2141         BTRFS_I(inode)->io_tree.track_uptodate = false;
2142         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2143
2144         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2145
2146         BTRFS_I(inode)->root = fs_info->tree_root;
2147         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2148         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2149         btrfs_insert_inode_hash(inode);
2150 }
2151
2152 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2153 {
2154         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2155         init_rwsem(&fs_info->dev_replace.rwsem);
2156         init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2157 }
2158
2159 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2160 {
2161         spin_lock_init(&fs_info->qgroup_lock);
2162         mutex_init(&fs_info->qgroup_ioctl_lock);
2163         fs_info->qgroup_tree = RB_ROOT;
2164         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2165         fs_info->qgroup_seq = 1;
2166         fs_info->qgroup_ulist = NULL;
2167         fs_info->qgroup_rescan_running = false;
2168         mutex_init(&fs_info->qgroup_rescan_lock);
2169 }
2170
2171 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2172                 struct btrfs_fs_devices *fs_devices)
2173 {
2174         u32 max_active = fs_info->thread_pool_size;
2175         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2176
2177         fs_info->workers =
2178                 btrfs_alloc_workqueue(fs_info, "worker",
2179                                       flags | WQ_HIGHPRI, max_active, 16);
2180
2181         fs_info->delalloc_workers =
2182                 btrfs_alloc_workqueue(fs_info, "delalloc",
2183                                       flags, max_active, 2);
2184
2185         fs_info->flush_workers =
2186                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2187                                       flags, max_active, 0);
2188
2189         fs_info->caching_workers =
2190                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2191
2192         /*
2193          * a higher idle thresh on the submit workers makes it much more
2194          * likely that bios will be send down in a sane order to the
2195          * devices
2196          */
2197         fs_info->submit_workers =
2198                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2199                                       min_t(u64, fs_devices->num_devices,
2200                                             max_active), 64);
2201
2202         fs_info->fixup_workers =
2203                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2204
2205         /*
2206          * endios are largely parallel and should have a very
2207          * low idle thresh
2208          */
2209         fs_info->endio_workers =
2210                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2211         fs_info->endio_meta_workers =
2212                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2213                                       max_active, 4);
2214         fs_info->endio_meta_write_workers =
2215                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2216                                       max_active, 2);
2217         fs_info->endio_raid56_workers =
2218                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2219                                       max_active, 4);
2220         fs_info->endio_repair_workers =
2221                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2222         fs_info->rmw_workers =
2223                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2224         fs_info->endio_write_workers =
2225                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2226                                       max_active, 2);
2227         fs_info->endio_freespace_worker =
2228                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2229                                       max_active, 0);
2230         fs_info->delayed_workers =
2231                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2232                                       max_active, 0);
2233         fs_info->readahead_workers =
2234                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2235                                       max_active, 2);
2236         fs_info->qgroup_rescan_workers =
2237                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2238         fs_info->extent_workers =
2239                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2240                                       min_t(u64, fs_devices->num_devices,
2241                                             max_active), 8);
2242
2243         if (!(fs_info->workers && fs_info->delalloc_workers &&
2244               fs_info->submit_workers && fs_info->flush_workers &&
2245               fs_info->endio_workers && fs_info->endio_meta_workers &&
2246               fs_info->endio_meta_write_workers &&
2247               fs_info->endio_repair_workers &&
2248               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2249               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2250               fs_info->caching_workers && fs_info->readahead_workers &&
2251               fs_info->fixup_workers && fs_info->delayed_workers &&
2252               fs_info->extent_workers &&
2253               fs_info->qgroup_rescan_workers)) {
2254                 return -ENOMEM;
2255         }
2256
2257         return 0;
2258 }
2259
2260 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2261                             struct btrfs_fs_devices *fs_devices)
2262 {
2263         int ret;
2264         struct btrfs_root *log_tree_root;
2265         struct btrfs_super_block *disk_super = fs_info->super_copy;
2266         u64 bytenr = btrfs_super_log_root(disk_super);
2267         int level = btrfs_super_log_root_level(disk_super);
2268
2269         if (fs_devices->rw_devices == 0) {
2270                 btrfs_warn(fs_info, "log replay required on RO media");
2271                 return -EIO;
2272         }
2273
2274         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2275         if (!log_tree_root)
2276                 return -ENOMEM;
2277
2278         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2279
2280         log_tree_root->node = read_tree_block(fs_info, bytenr,
2281                                               fs_info->generation + 1,
2282                                               level, NULL);
2283         if (IS_ERR(log_tree_root->node)) {
2284                 btrfs_warn(fs_info, "failed to read log tree");
2285                 ret = PTR_ERR(log_tree_root->node);
2286                 kfree(log_tree_root);
2287                 return ret;
2288         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2289                 btrfs_err(fs_info, "failed to read log tree");
2290                 free_extent_buffer(log_tree_root->node);
2291                 kfree(log_tree_root);
2292                 return -EIO;
2293         }
2294         /* returns with log_tree_root freed on success */
2295         ret = btrfs_recover_log_trees(log_tree_root);
2296         if (ret) {
2297                 btrfs_handle_fs_error(fs_info, ret,
2298                                       "Failed to recover log tree");
2299                 free_extent_buffer(log_tree_root->node);
2300                 kfree(log_tree_root);
2301                 return ret;
2302         }
2303
2304         if (sb_rdonly(fs_info->sb)) {
2305                 ret = btrfs_commit_super(fs_info);
2306                 if (ret)
2307                         return ret;
2308         }
2309
2310         return 0;
2311 }
2312
2313 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2314 {
2315         struct btrfs_root *tree_root = fs_info->tree_root;
2316         struct btrfs_root *root;
2317         struct btrfs_key location;
2318         int ret;
2319
2320         BUG_ON(!fs_info->tree_root);
2321
2322         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2323         location.type = BTRFS_ROOT_ITEM_KEY;
2324         location.offset = 0;
2325
2326         root = btrfs_read_tree_root(tree_root, &location);
2327         if (IS_ERR(root)) {
2328                 ret = PTR_ERR(root);
2329                 goto out;
2330         }
2331         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2332         fs_info->extent_root = root;
2333
2334         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2335         root = btrfs_read_tree_root(tree_root, &location);
2336         if (IS_ERR(root)) {
2337                 ret = PTR_ERR(root);
2338                 goto out;
2339         }
2340         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2341         fs_info->dev_root = root;
2342         btrfs_init_devices_late(fs_info);
2343
2344         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2345         root = btrfs_read_tree_root(tree_root, &location);
2346         if (IS_ERR(root)) {
2347                 ret = PTR_ERR(root);
2348                 goto out;
2349         }
2350         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2351         fs_info->csum_root = root;
2352
2353         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2354         root = btrfs_read_tree_root(tree_root, &location);
2355         if (!IS_ERR(root)) {
2356                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2357                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2358                 fs_info->quota_root = root;
2359         }
2360
2361         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2362         root = btrfs_read_tree_root(tree_root, &location);
2363         if (IS_ERR(root)) {
2364                 ret = PTR_ERR(root);
2365                 if (ret != -ENOENT)
2366                         goto out;
2367         } else {
2368                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369                 fs_info->uuid_root = root;
2370         }
2371
2372         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2373                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2374                 root = btrfs_read_tree_root(tree_root, &location);
2375                 if (IS_ERR(root)) {
2376                         ret = PTR_ERR(root);
2377                         goto out;
2378                 }
2379                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2380                 fs_info->free_space_root = root;
2381         }
2382
2383         return 0;
2384 out:
2385         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2386                    location.objectid, ret);
2387         return ret;
2388 }
2389
2390 /*
2391  * Real super block validation
2392  * NOTE: super csum type and incompat features will not be checked here.
2393  *
2394  * @sb:         super block to check
2395  * @mirror_num: the super block number to check its bytenr:
2396  *              0       the primary (1st) sb
2397  *              1, 2    2nd and 3rd backup copy
2398  *             -1       skip bytenr check
2399  */
2400 static int validate_super(struct btrfs_fs_info *fs_info,
2401                             struct btrfs_super_block *sb, int mirror_num)
2402 {
2403         u64 nodesize = btrfs_super_nodesize(sb);
2404         u64 sectorsize = btrfs_super_sectorsize(sb);
2405         int ret = 0;
2406
2407         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2408                 btrfs_err(fs_info, "no valid FS found");
2409                 ret = -EINVAL;
2410         }
2411         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2412                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2413                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2414                 ret = -EINVAL;
2415         }
2416         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2417                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2418                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2419                 ret = -EINVAL;
2420         }
2421         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2422                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2423                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2424                 ret = -EINVAL;
2425         }
2426         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2427                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2428                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2429                 ret = -EINVAL;
2430         }
2431
2432         /*
2433          * Check sectorsize and nodesize first, other check will need it.
2434          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2435          */
2436         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2437             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2438                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2439                 ret = -EINVAL;
2440         }
2441         /* Only PAGE SIZE is supported yet */
2442         if (sectorsize != PAGE_SIZE) {
2443                 btrfs_err(fs_info,
2444                         "sectorsize %llu not supported yet, only support %lu",
2445                         sectorsize, PAGE_SIZE);
2446                 ret = -EINVAL;
2447         }
2448         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2449             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2450                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2451                 ret = -EINVAL;
2452         }
2453         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2454                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2455                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2456                 ret = -EINVAL;
2457         }
2458
2459         /* Root alignment check */
2460         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2461                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2462                            btrfs_super_root(sb));
2463                 ret = -EINVAL;
2464         }
2465         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2466                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2467                            btrfs_super_chunk_root(sb));
2468                 ret = -EINVAL;
2469         }
2470         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2471                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2472                            btrfs_super_log_root(sb));
2473                 ret = -EINVAL;
2474         }
2475
2476         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2477                    BTRFS_FSID_SIZE) != 0) {
2478                 btrfs_err(fs_info,
2479                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2480                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2481                 ret = -EINVAL;
2482         }
2483
2484         /*
2485          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2486          * done later
2487          */
2488         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2489                 btrfs_err(fs_info, "bytes_used is too small %llu",
2490                           btrfs_super_bytes_used(sb));
2491                 ret = -EINVAL;
2492         }
2493         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2494                 btrfs_err(fs_info, "invalid stripesize %u",
2495                           btrfs_super_stripesize(sb));
2496                 ret = -EINVAL;
2497         }
2498         if (btrfs_super_num_devices(sb) > (1UL << 31))
2499                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2500                            btrfs_super_num_devices(sb));
2501         if (btrfs_super_num_devices(sb) == 0) {
2502                 btrfs_err(fs_info, "number of devices is 0");
2503                 ret = -EINVAL;
2504         }
2505
2506         if (mirror_num >= 0 &&
2507             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2508                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2509                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2510                 ret = -EINVAL;
2511         }
2512
2513         /*
2514          * Obvious sys_chunk_array corruptions, it must hold at least one key
2515          * and one chunk
2516          */
2517         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2518                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2519                           btrfs_super_sys_array_size(sb),
2520                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2521                 ret = -EINVAL;
2522         }
2523         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2524                         + sizeof(struct btrfs_chunk)) {
2525                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2526                           btrfs_super_sys_array_size(sb),
2527                           sizeof(struct btrfs_disk_key)
2528                           + sizeof(struct btrfs_chunk));
2529                 ret = -EINVAL;
2530         }
2531
2532         /*
2533          * The generation is a global counter, we'll trust it more than the others
2534          * but it's still possible that it's the one that's wrong.
2535          */
2536         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2537                 btrfs_warn(fs_info,
2538                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2539                         btrfs_super_generation(sb),
2540                         btrfs_super_chunk_root_generation(sb));
2541         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2542             && btrfs_super_cache_generation(sb) != (u64)-1)
2543                 btrfs_warn(fs_info,
2544                         "suspicious: generation < cache_generation: %llu < %llu",
2545                         btrfs_super_generation(sb),
2546                         btrfs_super_cache_generation(sb));
2547
2548         return ret;
2549 }
2550
2551 /*
2552  * Validation of super block at mount time.
2553  * Some checks already done early at mount time, like csum type and incompat
2554  * flags will be skipped.
2555  */
2556 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2557 {
2558         return validate_super(fs_info, fs_info->super_copy, 0);
2559 }
2560
2561 /*
2562  * Validation of super block at write time.
2563  * Some checks like bytenr check will be skipped as their values will be
2564  * overwritten soon.
2565  * Extra checks like csum type and incompat flags will be done here.
2566  */
2567 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2568                                       struct btrfs_super_block *sb)
2569 {
2570         int ret;
2571
2572         ret = validate_super(fs_info, sb, -1);
2573         if (ret < 0)
2574                 goto out;
2575         if (btrfs_super_csum_type(sb) != BTRFS_CSUM_TYPE_CRC32) {
2576                 ret = -EUCLEAN;
2577                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2578                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2579                 goto out;
2580         }
2581         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2582                 ret = -EUCLEAN;
2583                 btrfs_err(fs_info,
2584                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2585                           btrfs_super_incompat_flags(sb),
2586                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2587                 goto out;
2588         }
2589 out:
2590         if (ret < 0)
2591                 btrfs_err(fs_info,
2592                 "super block corruption detected before writing it to disk");
2593         return ret;
2594 }
2595
2596 int open_ctree(struct super_block *sb,
2597                struct btrfs_fs_devices *fs_devices,
2598                char *options)
2599 {
2600         u32 sectorsize;
2601         u32 nodesize;
2602         u32 stripesize;
2603         u64 generation;
2604         u64 features;
2605         struct btrfs_key location;
2606         struct buffer_head *bh;
2607         struct btrfs_super_block *disk_super;
2608         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2609         struct btrfs_root *tree_root;
2610         struct btrfs_root *chunk_root;
2611         int ret;
2612         int err = -EINVAL;
2613         int num_backups_tried = 0;
2614         int backup_index = 0;
2615         int clear_free_space_tree = 0;
2616         int level;
2617
2618         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2619         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2620         if (!tree_root || !chunk_root) {
2621                 err = -ENOMEM;
2622                 goto fail;
2623         }
2624
2625         ret = init_srcu_struct(&fs_info->subvol_srcu);
2626         if (ret) {
2627                 err = ret;
2628                 goto fail;
2629         }
2630
2631         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2632         if (ret) {
2633                 err = ret;
2634                 goto fail_srcu;
2635         }
2636         fs_info->dirty_metadata_batch = PAGE_SIZE *
2637                                         (1 + ilog2(nr_cpu_ids));
2638
2639         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2640         if (ret) {
2641                 err = ret;
2642                 goto fail_dirty_metadata_bytes;
2643         }
2644
2645         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2646                         GFP_KERNEL);
2647         if (ret) {
2648                 err = ret;
2649                 goto fail_delalloc_bytes;
2650         }
2651
2652         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2653         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2654         INIT_LIST_HEAD(&fs_info->trans_list);
2655         INIT_LIST_HEAD(&fs_info->dead_roots);
2656         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2657         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2658         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2659         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2660         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2661         spin_lock_init(&fs_info->delalloc_root_lock);
2662         spin_lock_init(&fs_info->trans_lock);
2663         spin_lock_init(&fs_info->fs_roots_radix_lock);
2664         spin_lock_init(&fs_info->delayed_iput_lock);
2665         spin_lock_init(&fs_info->defrag_inodes_lock);
2666         spin_lock_init(&fs_info->tree_mod_seq_lock);
2667         spin_lock_init(&fs_info->super_lock);
2668         spin_lock_init(&fs_info->buffer_lock);
2669         spin_lock_init(&fs_info->unused_bgs_lock);
2670         rwlock_init(&fs_info->tree_mod_log_lock);
2671         mutex_init(&fs_info->unused_bg_unpin_mutex);
2672         mutex_init(&fs_info->delete_unused_bgs_mutex);
2673         mutex_init(&fs_info->reloc_mutex);
2674         mutex_init(&fs_info->delalloc_root_mutex);
2675         seqlock_init(&fs_info->profiles_lock);
2676
2677         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2678         INIT_LIST_HEAD(&fs_info->space_info);
2679         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2680         INIT_LIST_HEAD(&fs_info->unused_bgs);
2681         btrfs_mapping_init(&fs_info->mapping_tree);
2682         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2683                              BTRFS_BLOCK_RSV_GLOBAL);
2684         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2685         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2686         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2687         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2688                              BTRFS_BLOCK_RSV_DELOPS);
2689         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2690                              BTRFS_BLOCK_RSV_DELREFS);
2691
2692         atomic_set(&fs_info->async_delalloc_pages, 0);
2693         atomic_set(&fs_info->defrag_running, 0);
2694         atomic_set(&fs_info->reada_works_cnt, 0);
2695         atomic_set(&fs_info->nr_delayed_iputs, 0);
2696         atomic64_set(&fs_info->tree_mod_seq, 0);
2697         fs_info->sb = sb;
2698         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2699         fs_info->metadata_ratio = 0;
2700         fs_info->defrag_inodes = RB_ROOT;
2701         atomic64_set(&fs_info->free_chunk_space, 0);
2702         fs_info->tree_mod_log = RB_ROOT;
2703         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2704         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2705         /* readahead state */
2706         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2707         spin_lock_init(&fs_info->reada_lock);
2708         btrfs_init_ref_verify(fs_info);
2709
2710         fs_info->thread_pool_size = min_t(unsigned long,
2711                                           num_online_cpus() + 2, 8);
2712
2713         INIT_LIST_HEAD(&fs_info->ordered_roots);
2714         spin_lock_init(&fs_info->ordered_root_lock);
2715
2716         fs_info->btree_inode = new_inode(sb);
2717         if (!fs_info->btree_inode) {
2718                 err = -ENOMEM;
2719                 goto fail_bio_counter;
2720         }
2721         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2722
2723         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2724                                         GFP_KERNEL);
2725         if (!fs_info->delayed_root) {
2726                 err = -ENOMEM;
2727                 goto fail_iput;
2728         }
2729         btrfs_init_delayed_root(fs_info->delayed_root);
2730
2731         btrfs_init_scrub(fs_info);
2732 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2733         fs_info->check_integrity_print_mask = 0;
2734 #endif
2735         btrfs_init_balance(fs_info);
2736         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2737
2738         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2739         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2740
2741         btrfs_init_btree_inode(fs_info);
2742
2743         spin_lock_init(&fs_info->block_group_cache_lock);
2744         fs_info->block_group_cache_tree = RB_ROOT;
2745         fs_info->first_logical_byte = (u64)-1;
2746
2747         extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2748                             IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2749         extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2750                             IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2751         fs_info->pinned_extents = &fs_info->freed_extents[0];
2752         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2753
2754         mutex_init(&fs_info->ordered_operations_mutex);
2755         mutex_init(&fs_info->tree_log_mutex);
2756         mutex_init(&fs_info->chunk_mutex);
2757         mutex_init(&fs_info->transaction_kthread_mutex);
2758         mutex_init(&fs_info->cleaner_mutex);
2759         mutex_init(&fs_info->ro_block_group_mutex);
2760         init_rwsem(&fs_info->commit_root_sem);
2761         init_rwsem(&fs_info->cleanup_work_sem);
2762         init_rwsem(&fs_info->subvol_sem);
2763         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2764
2765         btrfs_init_dev_replace_locks(fs_info);
2766         btrfs_init_qgroup(fs_info);
2767
2768         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2769         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2770
2771         init_waitqueue_head(&fs_info->transaction_throttle);
2772         init_waitqueue_head(&fs_info->transaction_wait);
2773         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2774         init_waitqueue_head(&fs_info->async_submit_wait);
2775         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2776
2777         /* Usable values until the real ones are cached from the superblock */
2778         fs_info->nodesize = 4096;
2779         fs_info->sectorsize = 4096;
2780         fs_info->stripesize = 4096;
2781
2782         spin_lock_init(&fs_info->swapfile_pins_lock);
2783         fs_info->swapfile_pins = RB_ROOT;
2784
2785         ret = btrfs_alloc_stripe_hash_table(fs_info);
2786         if (ret) {
2787                 err = ret;
2788                 goto fail_alloc;
2789         }
2790
2791         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2792
2793         invalidate_bdev(fs_devices->latest_bdev);
2794
2795         /*
2796          * Read super block and check the signature bytes only
2797          */
2798         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2799         if (IS_ERR(bh)) {
2800                 err = PTR_ERR(bh);
2801                 goto fail_alloc;
2802         }
2803
2804         /*
2805          * We want to check superblock checksum, the type is stored inside.
2806          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2807          */
2808         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2809                 btrfs_err(fs_info, "superblock checksum mismatch");
2810                 err = -EINVAL;
2811                 brelse(bh);
2812                 goto fail_alloc;
2813         }
2814
2815         /*
2816          * super_copy is zeroed at allocation time and we never touch the
2817          * following bytes up to INFO_SIZE, the checksum is calculated from
2818          * the whole block of INFO_SIZE
2819          */
2820         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2821         brelse(bh);
2822
2823         disk_super = fs_info->super_copy;
2824
2825         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2826                        BTRFS_FSID_SIZE));
2827
2828         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2829                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2830                                 fs_info->super_copy->metadata_uuid,
2831                                 BTRFS_FSID_SIZE));
2832         }
2833
2834         features = btrfs_super_flags(disk_super);
2835         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2836                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2837                 btrfs_set_super_flags(disk_super, features);
2838                 btrfs_info(fs_info,
2839                         "found metadata UUID change in progress flag, clearing");
2840         }
2841
2842         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2843                sizeof(*fs_info->super_for_commit));
2844
2845         ret = btrfs_validate_mount_super(fs_info);
2846         if (ret) {
2847                 btrfs_err(fs_info, "superblock contains fatal errors");
2848                 err = -EINVAL;
2849                 goto fail_alloc;
2850         }
2851
2852         if (!btrfs_super_root(disk_super))
2853                 goto fail_alloc;
2854
2855         /* check FS state, whether FS is broken. */
2856         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2857                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2858
2859         /*
2860          * run through our array of backup supers and setup
2861          * our ring pointer to the oldest one
2862          */
2863         generation = btrfs_super_generation(disk_super);
2864         find_oldest_super_backup(fs_info, generation);
2865
2866         /*
2867          * In the long term, we'll store the compression type in the super
2868          * block, and it'll be used for per file compression control.
2869          */
2870         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2871
2872         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2873         if (ret) {
2874                 err = ret;
2875                 goto fail_alloc;
2876         }
2877
2878         features = btrfs_super_incompat_flags(disk_super) &
2879                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2880         if (features) {
2881                 btrfs_err(fs_info,
2882                     "cannot mount because of unsupported optional features (%llx)",
2883                     features);
2884                 err = -EINVAL;
2885                 goto fail_alloc;
2886         }
2887
2888         features = btrfs_super_incompat_flags(disk_super);
2889         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2890         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2891                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2892         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2893                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2894
2895         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2896                 btrfs_info(fs_info, "has skinny extents");
2897
2898         /*
2899          * flag our filesystem as having big metadata blocks if
2900          * they are bigger than the page size
2901          */
2902         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2903                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2904                         btrfs_info(fs_info,
2905                                 "flagging fs with big metadata feature");
2906                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2907         }
2908
2909         nodesize = btrfs_super_nodesize(disk_super);
2910         sectorsize = btrfs_super_sectorsize(disk_super);
2911         stripesize = sectorsize;
2912         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2913         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2914
2915         /* Cache block sizes */
2916         fs_info->nodesize = nodesize;
2917         fs_info->sectorsize = sectorsize;
2918         fs_info->stripesize = stripesize;
2919
2920         /*
2921          * mixed block groups end up with duplicate but slightly offset
2922          * extent buffers for the same range.  It leads to corruptions
2923          */
2924         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2925             (sectorsize != nodesize)) {
2926                 btrfs_err(fs_info,
2927 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2928                         nodesize, sectorsize);
2929                 goto fail_alloc;
2930         }
2931
2932         /*
2933          * Needn't use the lock because there is no other task which will
2934          * update the flag.
2935          */
2936         btrfs_set_super_incompat_flags(disk_super, features);
2937
2938         features = btrfs_super_compat_ro_flags(disk_super) &
2939                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2940         if (!sb_rdonly(sb) && features) {
2941                 btrfs_err(fs_info,
2942         "cannot mount read-write because of unsupported optional features (%llx)",
2943                        features);
2944                 err = -EINVAL;
2945                 goto fail_alloc;
2946         }
2947
2948         ret = btrfs_init_workqueues(fs_info, fs_devices);
2949         if (ret) {
2950                 err = ret;
2951                 goto fail_sb_buffer;
2952         }
2953
2954         sb->s_bdi->congested_fn = btrfs_congested_fn;
2955         sb->s_bdi->congested_data = fs_info;
2956         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2957         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
2958         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2959         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2960
2961         sb->s_blocksize = sectorsize;
2962         sb->s_blocksize_bits = blksize_bits(sectorsize);
2963         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
2964
2965         mutex_lock(&fs_info->chunk_mutex);
2966         ret = btrfs_read_sys_array(fs_info);
2967         mutex_unlock(&fs_info->chunk_mutex);
2968         if (ret) {
2969                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2970                 goto fail_sb_buffer;
2971         }
2972
2973         generation = btrfs_super_chunk_root_generation(disk_super);
2974         level = btrfs_super_chunk_root_level(disk_super);
2975
2976         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2977
2978         chunk_root->node = read_tree_block(fs_info,
2979                                            btrfs_super_chunk_root(disk_super),
2980                                            generation, level, NULL);
2981         if (IS_ERR(chunk_root->node) ||
2982             !extent_buffer_uptodate(chunk_root->node)) {
2983                 btrfs_err(fs_info, "failed to read chunk root");
2984                 if (!IS_ERR(chunk_root->node))
2985                         free_extent_buffer(chunk_root->node);
2986                 chunk_root->node = NULL;
2987                 goto fail_tree_roots;
2988         }
2989         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2990         chunk_root->commit_root = btrfs_root_node(chunk_root);
2991
2992         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2993            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2994
2995         ret = btrfs_read_chunk_tree(fs_info);
2996         if (ret) {
2997                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2998                 goto fail_tree_roots;
2999         }
3000
3001         /*
3002          * Keep the devid that is marked to be the target device for the
3003          * device replace procedure
3004          */
3005         btrfs_free_extra_devids(fs_devices, 0);
3006
3007         if (!fs_devices->latest_bdev) {
3008                 btrfs_err(fs_info, "failed to read devices");
3009                 goto fail_tree_roots;
3010         }
3011
3012 retry_root_backup:
3013         generation = btrfs_super_generation(disk_super);
3014         level = btrfs_super_root_level(disk_super);
3015
3016         tree_root->node = read_tree_block(fs_info,
3017                                           btrfs_super_root(disk_super),
3018                                           generation, level, NULL);
3019         if (IS_ERR(tree_root->node) ||
3020             !extent_buffer_uptodate(tree_root->node)) {
3021                 btrfs_warn(fs_info, "failed to read tree root");
3022                 if (!IS_ERR(tree_root->node))
3023                         free_extent_buffer(tree_root->node);
3024                 tree_root->node = NULL;
3025                 goto recovery_tree_root;
3026         }
3027
3028         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3029         tree_root->commit_root = btrfs_root_node(tree_root);
3030         btrfs_set_root_refs(&tree_root->root_item, 1);
3031
3032         mutex_lock(&tree_root->objectid_mutex);
3033         ret = btrfs_find_highest_objectid(tree_root,
3034                                         &tree_root->highest_objectid);
3035         if (ret) {
3036                 mutex_unlock(&tree_root->objectid_mutex);
3037                 goto recovery_tree_root;
3038         }
3039
3040         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3041
3042         mutex_unlock(&tree_root->objectid_mutex);
3043
3044         ret = btrfs_read_roots(fs_info);
3045         if (ret)
3046                 goto recovery_tree_root;
3047
3048         fs_info->generation = generation;
3049         fs_info->last_trans_committed = generation;
3050
3051         ret = btrfs_verify_dev_extents(fs_info);
3052         if (ret) {
3053                 btrfs_err(fs_info,
3054                           "failed to verify dev extents against chunks: %d",
3055                           ret);
3056                 goto fail_block_groups;
3057         }
3058         ret = btrfs_recover_balance(fs_info);
3059         if (ret) {
3060                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3061                 goto fail_block_groups;
3062         }
3063
3064         ret = btrfs_init_dev_stats(fs_info);
3065         if (ret) {
3066                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3067                 goto fail_block_groups;
3068         }
3069
3070         ret = btrfs_init_dev_replace(fs_info);
3071         if (ret) {
3072                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3073                 goto fail_block_groups;
3074         }
3075
3076         btrfs_free_extra_devids(fs_devices, 1);
3077
3078         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3079         if (ret) {
3080                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3081                                 ret);
3082                 goto fail_block_groups;
3083         }
3084
3085         ret = btrfs_sysfs_add_device(fs_devices);
3086         if (ret) {
3087                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3088                                 ret);
3089                 goto fail_fsdev_sysfs;
3090         }
3091
3092         ret = btrfs_sysfs_add_mounted(fs_info);
3093         if (ret) {
3094                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3095                 goto fail_fsdev_sysfs;
3096         }
3097
3098         ret = btrfs_init_space_info(fs_info);
3099         if (ret) {
3100                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3101                 goto fail_sysfs;
3102         }
3103
3104         ret = btrfs_read_block_groups(fs_info);
3105         if (ret) {
3106                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3107                 goto fail_sysfs;
3108         }
3109
3110         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3111                 btrfs_warn(fs_info,
3112                 "writable mount is not allowed due to too many missing devices");
3113                 goto fail_sysfs;
3114         }
3115
3116         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3117                                                "btrfs-cleaner");
3118         if (IS_ERR(fs_info->cleaner_kthread))
3119                 goto fail_sysfs;
3120
3121         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3122                                                    tree_root,
3123                                                    "btrfs-transaction");
3124         if (IS_ERR(fs_info->transaction_kthread))
3125                 goto fail_cleaner;
3126
3127         if (!btrfs_test_opt(fs_info, NOSSD) &&
3128             !fs_info->fs_devices->rotating) {
3129                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3130         }
3131
3132         /*
3133          * Mount does not set all options immediately, we can do it now and do
3134          * not have to wait for transaction commit
3135          */
3136         btrfs_apply_pending_changes(fs_info);
3137
3138 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3139         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3140                 ret = btrfsic_mount(fs_info, fs_devices,
3141                                     btrfs_test_opt(fs_info,
3142                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3143                                     1 : 0,
3144                                     fs_info->check_integrity_print_mask);
3145                 if (ret)
3146                         btrfs_warn(fs_info,
3147                                 "failed to initialize integrity check module: %d",
3148                                 ret);
3149         }
3150 #endif
3151         ret = btrfs_read_qgroup_config(fs_info);
3152         if (ret)
3153                 goto fail_trans_kthread;
3154
3155         if (btrfs_build_ref_tree(fs_info))
3156                 btrfs_err(fs_info, "couldn't build ref tree");
3157
3158         /* do not make disk changes in broken FS or nologreplay is given */
3159         if (btrfs_super_log_root(disk_super) != 0 &&
3160             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3161                 ret = btrfs_replay_log(fs_info, fs_devices);
3162                 if (ret) {
3163                         err = ret;
3164                         goto fail_qgroup;
3165                 }
3166         }
3167
3168         ret = btrfs_find_orphan_roots(fs_info);
3169         if (ret)
3170                 goto fail_qgroup;
3171
3172         if (!sb_rdonly(sb)) {
3173                 ret = btrfs_cleanup_fs_roots(fs_info);
3174                 if (ret)
3175                         goto fail_qgroup;
3176
3177                 mutex_lock(&fs_info->cleaner_mutex);
3178                 ret = btrfs_recover_relocation(tree_root);
3179                 mutex_unlock(&fs_info->cleaner_mutex);
3180                 if (ret < 0) {
3181                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3182                                         ret);
3183                         err = -EINVAL;
3184                         goto fail_qgroup;
3185                 }
3186         }
3187
3188         location.objectid = BTRFS_FS_TREE_OBJECTID;
3189         location.type = BTRFS_ROOT_ITEM_KEY;
3190         location.offset = 0;
3191
3192         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3193         if (IS_ERR(fs_info->fs_root)) {
3194                 err = PTR_ERR(fs_info->fs_root);
3195                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3196                 goto fail_qgroup;
3197         }
3198
3199         if (sb_rdonly(sb))
3200                 return 0;
3201
3202         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3203             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3204                 clear_free_space_tree = 1;
3205         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3206                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3207                 btrfs_warn(fs_info, "free space tree is invalid");
3208                 clear_free_space_tree = 1;
3209         }
3210
3211         if (clear_free_space_tree) {
3212                 btrfs_info(fs_info, "clearing free space tree");
3213                 ret = btrfs_clear_free_space_tree(fs_info);
3214                 if (ret) {
3215                         btrfs_warn(fs_info,
3216                                    "failed to clear free space tree: %d", ret);
3217                         close_ctree(fs_info);
3218                         return ret;
3219                 }
3220         }
3221
3222         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3223             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3224                 btrfs_info(fs_info, "creating free space tree");
3225                 ret = btrfs_create_free_space_tree(fs_info);
3226                 if (ret) {
3227                         btrfs_warn(fs_info,
3228                                 "failed to create free space tree: %d", ret);
3229                         close_ctree(fs_info);
3230                         return ret;
3231                 }
3232         }
3233
3234         down_read(&fs_info->cleanup_work_sem);
3235         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3236             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3237                 up_read(&fs_info->cleanup_work_sem);
3238                 close_ctree(fs_info);
3239                 return ret;
3240         }
3241         up_read(&fs_info->cleanup_work_sem);
3242
3243         ret = btrfs_resume_balance_async(fs_info);
3244         if (ret) {
3245                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3246                 close_ctree(fs_info);
3247                 return ret;
3248         }
3249
3250         ret = btrfs_resume_dev_replace_async(fs_info);
3251         if (ret) {
3252                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3253                 close_ctree(fs_info);
3254                 return ret;
3255         }
3256
3257         btrfs_qgroup_rescan_resume(fs_info);