btrfs: add boilerplate code for directly including the crypto framework
[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 <crypto/hash.h>
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "transaction.h"
26 #include "btrfs_inode.h"
27 #include "volumes.h"
28 #include "print-tree.h"
29 #include "locking.h"
30 #include "tree-log.h"
31 #include "free-space-cache.h"
32 #include "free-space-tree.h"
33 #include "inode-map.h"
34 #include "check-integrity.h"
35 #include "rcu-string.h"
36 #include "dev-replace.h"
37 #include "raid56.h"
38 #include "sysfs.h"
39 #include "qgroup.h"
40 #include "compression.h"
41 #include "tree-checker.h"
42 #include "ref-verify.h"
43
44 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
45                                  BTRFS_HEADER_FLAG_RELOC |\
46                                  BTRFS_SUPER_FLAG_ERROR |\
47                                  BTRFS_SUPER_FLAG_SEEDING |\
48                                  BTRFS_SUPER_FLAG_METADUMP |\
49                                  BTRFS_SUPER_FLAG_METADUMP_V2)
50
51 static const struct extent_io_ops btree_extent_io_ops;
52 static void end_workqueue_fn(struct btrfs_work *work);
53 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55                                       struct btrfs_fs_info *fs_info);
56 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
57 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
58                                         struct extent_io_tree *dirty_pages,
59                                         int mark);
60 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
61                                        struct extent_io_tree *pinned_extents);
62 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
63 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
64
65 /*
66  * btrfs_end_io_wq structs are used to do processing in task context when an IO
67  * is complete.  This is used during reads to verify checksums, and it is used
68  * by writes to insert metadata for new file extents after IO is complete.
69  */
70 struct btrfs_end_io_wq {
71         struct bio *bio;
72         bio_end_io_t *end_io;
73         void *private;
74         struct btrfs_fs_info *info;
75         blk_status_t status;
76         enum btrfs_wq_endio_type metadata;
77         struct btrfs_work work;
78 };
79
80 static struct kmem_cache *btrfs_end_io_wq_cache;
81
82 int __init btrfs_end_io_wq_init(void)
83 {
84         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
85                                         sizeof(struct btrfs_end_io_wq),
86                                         0,
87                                         SLAB_MEM_SPREAD,
88                                         NULL);
89         if (!btrfs_end_io_wq_cache)
90                 return -ENOMEM;
91         return 0;
92 }
93
94 void __cold btrfs_end_io_wq_exit(void)
95 {
96         kmem_cache_destroy(btrfs_end_io_wq_cache);
97 }
98
99 /*
100  * async submit bios are used to offload expensive checksumming
101  * onto the worker threads.  They checksum file and metadata bios
102  * just before they are sent down the IO stack.
103  */
104 struct async_submit_bio {
105         void *private_data;
106         struct bio *bio;
107         extent_submit_bio_start_t *submit_bio_start;
108         int mirror_num;
109         /*
110          * bio_offset is optional, can be used if the pages in the bio
111          * can't tell us where in the file the bio should go
112          */
113         u64 bio_offset;
114         struct btrfs_work work;
115         blk_status_t status;
116 };
117
118 /*
119  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
120  * eb, the lockdep key is determined by the btrfs_root it belongs to and
121  * the level the eb occupies in the tree.
122  *
123  * Different roots are used for different purposes and may nest inside each
124  * other and they require separate keysets.  As lockdep keys should be
125  * static, assign keysets according to the purpose of the root as indicated
126  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
127  * roots have separate keysets.
128  *
129  * Lock-nesting across peer nodes is always done with the immediate parent
130  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
131  * subclass to avoid triggering lockdep warning in such cases.
132  *
133  * The key is set by the readpage_end_io_hook after the buffer has passed
134  * csum validation but before the pages are unlocked.  It is also set by
135  * btrfs_init_new_buffer on freshly allocated blocks.
136  *
137  * We also add a check to make sure the highest level of the tree is the
138  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
139  * needs update as well.
140  */
141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
142 # if BTRFS_MAX_LEVEL != 8
143 #  error
144 # endif
145
146 static struct btrfs_lockdep_keyset {
147         u64                     id;             /* root objectid */
148         const char              *name_stem;     /* lock name stem */
149         char                    names[BTRFS_MAX_LEVEL + 1][20];
150         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
151 } btrfs_lockdep_keysets[] = {
152         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
153         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
154         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
155         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
156         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
157         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
158         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
159         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
160         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
161         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
162         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
163         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
164         { .id = 0,                              .name_stem = "tree"     },
165 };
166
167 void __init btrfs_init_lockdep(void)
168 {
169         int i, j;
170
171         /* initialize lockdep class names */
172         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
173                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
174
175                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
176                         snprintf(ks->names[j], sizeof(ks->names[j]),
177                                  "btrfs-%s-%02d", ks->name_stem, j);
178         }
179 }
180
181 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
182                                     int level)
183 {
184         struct btrfs_lockdep_keyset *ks;
185
186         BUG_ON(level >= ARRAY_SIZE(ks->keys));
187
188         /* find the matching keyset, id 0 is the default entry */
189         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
190                 if (ks->id == objectid)
191                         break;
192
193         lockdep_set_class_and_name(&eb->lock,
194                                    &ks->keys[level], ks->names[level]);
195 }
196
197 #endif
198
199 /*
200  * extents on the btree inode are pretty simple, there's one extent
201  * that covers the entire device
202  */
203 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
204                 struct page *page, size_t pg_offset, u64 start, u64 len,
205                 int create)
206 {
207         struct btrfs_fs_info *fs_info = inode->root->fs_info;
208         struct extent_map_tree *em_tree = &inode->extent_tree;
209         struct extent_map *em;
210         int ret;
211
212         read_lock(&em_tree->lock);
213         em = lookup_extent_mapping(em_tree, start, len);
214         if (em) {
215                 em->bdev = fs_info->fs_devices->latest_bdev;
216                 read_unlock(&em_tree->lock);
217                 goto out;
218         }
219         read_unlock(&em_tree->lock);
220
221         em = alloc_extent_map();
222         if (!em) {
223                 em = ERR_PTR(-ENOMEM);
224                 goto out;
225         }
226         em->start = 0;
227         em->len = (u64)-1;
228         em->block_len = (u64)-1;
229         em->block_start = 0;
230         em->bdev = fs_info->fs_devices->latest_bdev;
231
232         write_lock(&em_tree->lock);
233         ret = add_extent_mapping(em_tree, em, 0);
234         if (ret == -EEXIST) {
235                 free_extent_map(em);
236                 em = lookup_extent_mapping(em_tree, start, len);
237                 if (!em)
238                         em = ERR_PTR(-EIO);
239         } else if (ret) {
240                 free_extent_map(em);
241                 em = ERR_PTR(ret);
242         }
243         write_unlock(&em_tree->lock);
244
245 out:
246         return em;
247 }
248
249 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
250 {
251         return crc32c(seed, data, len);
252 }
253
254 void btrfs_csum_final(u32 crc, u8 *result)
255 {
256         put_unaligned_le32(~crc, result);
257 }
258
259 /*
260  * Compute the csum of a btree block and store the result to provided buffer.
261  *
262  * Returns error if the extent buffer cannot be mapped.
263  */
264 static int csum_tree_block(struct extent_buffer *buf, u8 *result)
265 {
266         unsigned long len;
267         unsigned long cur_len;
268         unsigned long offset = BTRFS_CSUM_SIZE;
269         char *kaddr;
270         unsigned long map_start;
271         unsigned long map_len;
272         int err;
273         u32 crc = ~(u32)0;
274
275         len = buf->len - offset;
276         while (len > 0) {
277                 /*
278                  * Note: we don't need to check for the err == 1 case here, as
279                  * with the given combination of 'start = BTRFS_CSUM_SIZE (32)'
280                  * and 'min_len = 32' and the currently implemented mapping
281                  * algorithm we cannot cross a page boundary.
282                  */
283                 err = map_private_extent_buffer(buf, offset, 32,
284                                         &kaddr, &map_start, &map_len);
285                 if (WARN_ON(err))
286                         return err;
287                 cur_len = min(len, map_len - (offset - map_start));
288                 crc = btrfs_csum_data(kaddr + offset - map_start,
289                                       crc, cur_len);
290                 len -= cur_len;
291                 offset += cur_len;
292         }
293         memset(result, 0, BTRFS_CSUM_SIZE);
294
295         btrfs_csum_final(crc, result);
296
297         return 0;
298 }
299
300 /*
301  * we can't consider a given block up to date unless the transid of the
302  * block matches the transid in the parent node's pointer.  This is how we
303  * detect blocks that either didn't get written at all or got written
304  * in the wrong place.
305  */
306 static int verify_parent_transid(struct extent_io_tree *io_tree,
307                                  struct extent_buffer *eb, u64 parent_transid,
308                                  int atomic)
309 {
310         struct extent_state *cached_state = NULL;
311         int ret;
312         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
313
314         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
315                 return 0;
316
317         if (atomic)
318                 return -EAGAIN;
319
320         if (need_lock) {
321                 btrfs_tree_read_lock(eb);
322                 btrfs_set_lock_blocking_read(eb);
323         }
324
325         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
326                          &cached_state);
327         if (extent_buffer_uptodate(eb) &&
328             btrfs_header_generation(eb) == parent_transid) {
329                 ret = 0;
330                 goto out;
331         }
332         btrfs_err_rl(eb->fs_info,
333                 "parent transid verify failed on %llu wanted %llu found %llu",
334                         eb->start,
335                         parent_transid, btrfs_header_generation(eb));
336         ret = 1;
337
338         /*
339          * Things reading via commit roots that don't have normal protection,
340          * like send, can have a really old block in cache that may point at a
341          * block that has been freed and re-allocated.  So don't clear uptodate
342          * if we find an eb that is under IO (dirty/writeback) because we could
343          * end up reading in the stale data and then writing it back out and
344          * making everybody very sad.
345          */
346         if (!extent_buffer_under_io(eb))
347                 clear_extent_buffer_uptodate(eb);
348 out:
349         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350                              &cached_state);
351         if (need_lock)
352                 btrfs_tree_read_unlock_blocking(eb);
353         return ret;
354 }
355
356 static bool btrfs_supported_super_csum(u16 csum_type)
357 {
358         switch (csum_type) {
359         case BTRFS_CSUM_TYPE_CRC32:
360                 return true;
361         default:
362                 return false;
363         }
364 }
365
366 /*
367  * Return 0 if the superblock checksum type matches the checksum value of that
368  * algorithm. Pass the raw disk superblock data.
369  */
370 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
371                                   char *raw_disk_sb)
372 {
373         struct btrfs_super_block *disk_sb =
374                 (struct btrfs_super_block *)raw_disk_sb;
375         u32 crc = ~(u32)0;
376         char result[BTRFS_CSUM_SIZE];
377
378         /*
379          * The super_block structure does not span the whole
380          * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
381          * filled with zeros and is included in the checksum.
382          */
383         crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
384                               crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
385         btrfs_csum_final(crc, result);
386
387         if (memcmp(disk_sb->csum, result, btrfs_super_csum_size(disk_sb)))
388                 return 1;
389
390         return 0;
391 }
392
393 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
394                            struct btrfs_key *first_key, u64 parent_transid)
395 {
396         struct btrfs_fs_info *fs_info = eb->fs_info;
397         int found_level;
398         struct btrfs_key found_key;
399         int ret;
400
401         found_level = btrfs_header_level(eb);
402         if (found_level != level) {
403                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
404                      KERN_ERR "BTRFS: tree level check failed\n");
405                 btrfs_err(fs_info,
406 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
407                           eb->start, level, found_level);
408                 return -EIO;
409         }
410
411         if (!first_key)
412                 return 0;
413
414         /*
415          * For live tree block (new tree blocks in current transaction),
416          * we need proper lock context to avoid race, which is impossible here.
417          * So we only checks tree blocks which is read from disk, whose
418          * generation <= fs_info->last_trans_committed.
419          */
420         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
421                 return 0;
422         if (found_level)
423                 btrfs_node_key_to_cpu(eb, &found_key, 0);
424         else
425                 btrfs_item_key_to_cpu(eb, &found_key, 0);
426         ret = btrfs_comp_cpu_keys(first_key, &found_key);
427
428         if (ret) {
429                 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
430                      KERN_ERR "BTRFS: tree first key check failed\n");
431                 btrfs_err(fs_info,
432 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
433                           eb->start, parent_transid, first_key->objectid,
434                           first_key->type, first_key->offset,
435                           found_key.objectid, found_key.type,
436                           found_key.offset);
437         }
438         return ret;
439 }
440
441 /*
442  * helper to read a given tree block, doing retries as required when
443  * the checksums don't match and we have alternate mirrors to try.
444  *
445  * @parent_transid:     expected transid, skip check if 0
446  * @level:              expected level, mandatory check
447  * @first_key:          expected key of first slot, skip check if NULL
448  */
449 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
450                                           u64 parent_transid, int level,
451                                           struct btrfs_key *first_key)
452 {
453         struct btrfs_fs_info *fs_info = eb->fs_info;
454         struct extent_io_tree *io_tree;
455         int failed = 0;
456         int ret;
457         int num_copies = 0;
458         int mirror_num = 0;
459         int failed_mirror = 0;
460
461         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
462         while (1) {
463                 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
464                 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
465                 if (!ret) {
466                         if (verify_parent_transid(io_tree, eb,
467                                                    parent_transid, 0))
468                                 ret = -EIO;
469                         else if (btrfs_verify_level_key(eb, level,
470                                                 first_key, parent_transid))
471                                 ret = -EUCLEAN;
472                         else
473                                 break;
474                 }
475
476                 num_copies = btrfs_num_copies(fs_info,
477                                               eb->start, eb->len);
478                 if (num_copies == 1)
479                         break;
480
481                 if (!failed_mirror) {
482                         failed = 1;
483                         failed_mirror = eb->read_mirror;
484                 }
485
486                 mirror_num++;
487                 if (mirror_num == failed_mirror)
488                         mirror_num++;
489
490                 if (mirror_num > num_copies)
491                         break;
492         }
493
494         if (failed && !ret && failed_mirror)
495                 btrfs_repair_eb_io_failure(eb, failed_mirror);
496
497         return ret;
498 }
499
500 /*
501  * checksum a dirty tree block before IO.  This has extra checks to make sure
502  * we only fill in the checksum field in the first page of a multi-page block
503  */
504
505 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
506 {
507         u64 start = page_offset(page);
508         u64 found_start;
509         u8 result[BTRFS_CSUM_SIZE];
510         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
511         struct extent_buffer *eb;
512         int ret;
513
514         eb = (struct extent_buffer *)page->private;
515         if (page != eb->pages[0])
516                 return 0;
517
518         found_start = btrfs_header_bytenr(eb);
519         /*
520          * Please do not consolidate these warnings into a single if.
521          * It is useful to know what went wrong.
522          */
523         if (WARN_ON(found_start != start))
524                 return -EUCLEAN;
525         if (WARN_ON(!PageUptodate(page)))
526                 return -EUCLEAN;
527
528         ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
529                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
530
531         if (csum_tree_block(eb, result))
532                 return -EINVAL;
533
534         if (btrfs_header_level(eb))
535                 ret = btrfs_check_node(eb);
536         else
537                 ret = btrfs_check_leaf_full(eb);
538
539         if (ret < 0) {
540                 btrfs_err(fs_info,
541                 "block=%llu write time tree block corruption detected",
542                           eb->start);
543                 return ret;
544         }
545         write_extent_buffer(eb, result, 0, csum_size);
546
547         return 0;
548 }
549
550 static int check_tree_block_fsid(struct extent_buffer *eb)
551 {
552         struct btrfs_fs_info *fs_info = eb->fs_info;
553         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
554         u8 fsid[BTRFS_FSID_SIZE];
555         int ret = 1;
556
557         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
558         while (fs_devices) {
559                 u8 *metadata_uuid;
560
561                 /*
562                  * Checking the incompat flag is only valid for the current
563                  * fs. For seed devices it's forbidden to have their uuid
564                  * changed so reading ->fsid in this case is fine
565                  */
566                 if (fs_devices == fs_info->fs_devices &&
567                     btrfs_fs_incompat(fs_info, METADATA_UUID))
568                         metadata_uuid = fs_devices->metadata_uuid;
569                 else
570                         metadata_uuid = fs_devices->fsid;
571
572                 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE)) {
573                         ret = 0;
574                         break;
575                 }
576                 fs_devices = fs_devices->seed;
577         }
578         return ret;
579 }
580
581 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
582                                       u64 phy_offset, struct page *page,
583                                       u64 start, u64 end, int mirror)
584 {
585         u64 found_start;
586         int found_level;
587         struct extent_buffer *eb;
588         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
589         struct btrfs_fs_info *fs_info = root->fs_info;
590         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
591         int ret = 0;
592         u8 result[BTRFS_CSUM_SIZE];
593         int reads_done;
594
595         if (!page->private)
596                 goto out;
597
598         eb = (struct extent_buffer *)page->private;
599
600         /* the pending IO might have been the only thing that kept this buffer
601          * in memory.  Make sure we have a ref for all this other checks
602          */
603         extent_buffer_get(eb);
604
605         reads_done = atomic_dec_and_test(&eb->io_pages);
606         if (!reads_done)
607                 goto err;
608
609         eb->read_mirror = mirror;
610         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
611                 ret = -EIO;
612                 goto err;
613         }
614
615         found_start = btrfs_header_bytenr(eb);
616         if (found_start != eb->start) {
617                 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
618                              eb->start, found_start);
619                 ret = -EIO;
620                 goto err;
621         }
622         if (check_tree_block_fsid(eb)) {
623                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
624                              eb->start);
625                 ret = -EIO;
626                 goto err;
627         }
628         found_level = btrfs_header_level(eb);
629         if (found_level >= BTRFS_MAX_LEVEL) {
630                 btrfs_err(fs_info, "bad tree block level %d on %llu",
631                           (int)btrfs_header_level(eb), eb->start);
632                 ret = -EIO;
633                 goto err;
634         }
635
636         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
637                                        eb, found_level);
638
639         ret = csum_tree_block(eb, result);
640         if (ret)
641                 goto err;
642
643         if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
644                 u32 val;
645                 u32 found = 0;
646
647                 memcpy(&found, result, csum_size);
648
649                 read_extent_buffer(eb, &val, 0, csum_size);
650                 btrfs_warn_rl(fs_info,
651                 "%s checksum verify failed on %llu wanted %x found %x level %d",
652                               fs_info->sb->s_id, eb->start,
653                               val, found, btrfs_header_level(eb));
654                 ret = -EUCLEAN;
655                 goto err;
656         }
657
658         /*
659          * If this is a leaf block and it is corrupt, set the corrupt bit so
660          * that we don't try and read the other copies of this block, just
661          * return -EIO.
662          */
663         if (found_level == 0 && btrfs_check_leaf_full(eb)) {
664                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
665                 ret = -EIO;
666         }
667
668         if (found_level > 0 && btrfs_check_node(eb))
669                 ret = -EIO;
670
671         if (!ret)
672                 set_extent_buffer_uptodate(eb);
673         else
674                 btrfs_err(fs_info,
675                           "block=%llu read time tree block corruption detected",
676                           eb->start);
677 err:
678         if (reads_done &&
679             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
680                 btree_readahead_hook(eb, ret);
681
682         if (ret) {
683                 /*
684                  * our io error hook is going to dec the io pages
685                  * again, we have to make sure it has something
686                  * to decrement
687                  */
688                 atomic_inc(&eb->io_pages);
689                 clear_extent_buffer_uptodate(eb);
690         }
691         free_extent_buffer(eb);
692 out:
693         return ret;
694 }
695
696 static void end_workqueue_bio(struct bio *bio)
697 {
698         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
699         struct btrfs_fs_info *fs_info;
700         struct btrfs_workqueue *wq;
701         btrfs_work_func_t func;
702
703         fs_info = end_io_wq->info;
704         end_io_wq->status = bio->bi_status;
705
706         if (bio_op(bio) == REQ_OP_WRITE) {
707                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
708                         wq = fs_info->endio_meta_write_workers;
709                         func = btrfs_endio_meta_write_helper;
710                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
711                         wq = fs_info->endio_freespace_worker;
712                         func = btrfs_freespace_write_helper;
713                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
714                         wq = fs_info->endio_raid56_workers;
715                         func = btrfs_endio_raid56_helper;
716                 } else {
717                         wq = fs_info->endio_write_workers;
718                         func = btrfs_endio_write_helper;
719                 }
720         } else {
721                 if (unlikely(end_io_wq->metadata ==
722                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
723                         wq = fs_info->endio_repair_workers;
724                         func = btrfs_endio_repair_helper;
725                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
726                         wq = fs_info->endio_raid56_workers;
727                         func = btrfs_endio_raid56_helper;
728                 } else if (end_io_wq->metadata) {
729                         wq = fs_info->endio_meta_workers;
730                         func = btrfs_endio_meta_helper;
731                 } else {
732                         wq = fs_info->endio_workers;
733                         func = btrfs_endio_helper;
734                 }
735         }
736
737         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
738         btrfs_queue_work(wq, &end_io_wq->work);
739 }
740
741 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
742                         enum btrfs_wq_endio_type metadata)
743 {
744         struct btrfs_end_io_wq *end_io_wq;
745
746         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
747         if (!end_io_wq)
748                 return BLK_STS_RESOURCE;
749
750         end_io_wq->private = bio->bi_private;
751         end_io_wq->end_io = bio->bi_end_io;
752         end_io_wq->info = info;
753         end_io_wq->status = 0;
754         end_io_wq->bio = bio;
755         end_io_wq->metadata = metadata;
756
757         bio->bi_private = end_io_wq;
758         bio->bi_end_io = end_workqueue_bio;
759         return 0;
760 }
761
762 static void run_one_async_start(struct btrfs_work *work)
763 {
764         struct async_submit_bio *async;
765         blk_status_t ret;
766
767         async = container_of(work, struct  async_submit_bio, work);
768         ret = async->submit_bio_start(async->private_data, async->bio,
769                                       async->bio_offset);
770         if (ret)
771                 async->status = ret;
772 }
773
774 /*
775  * In order to insert checksums into the metadata in large chunks, we wait
776  * until bio submission time.   All the pages in the bio are checksummed and
777  * sums are attached onto the ordered extent record.
778  *
779  * At IO completion time the csums attached on the ordered extent record are
780  * inserted into the tree.
781  */
782 static void run_one_async_done(struct btrfs_work *work)
783 {
784         struct async_submit_bio *async;
785         struct inode *inode;
786         blk_status_t ret;
787
788         async = container_of(work, struct  async_submit_bio, work);
789         inode = async->private_data;
790
791         /* If an error occurred we just want to clean up the bio and move on */
792         if (async->status) {
793                 async->bio->bi_status = async->status;
794                 bio_endio(async->bio);
795                 return;
796         }
797
798         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio,
799                         async->mirror_num, 1);
800         if (ret) {
801                 async->bio->bi_status = ret;
802                 bio_endio(async->bio);
803         }
804 }
805
806 static void run_one_async_free(struct btrfs_work *work)
807 {
808         struct async_submit_bio *async;
809
810         async = container_of(work, struct  async_submit_bio, work);
811         kfree(async);
812 }
813
814 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
815                                  int mirror_num, unsigned long bio_flags,
816                                  u64 bio_offset, void *private_data,
817                                  extent_submit_bio_start_t *submit_bio_start)
818 {
819         struct async_submit_bio *async;
820
821         async = kmalloc(sizeof(*async), GFP_NOFS);
822         if (!async)
823                 return BLK_STS_RESOURCE;
824
825         async->private_data = private_data;
826         async->bio = bio;
827         async->mirror_num = mirror_num;
828         async->submit_bio_start = submit_bio_start;
829
830         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
831                         run_one_async_done, run_one_async_free);
832
833         async->bio_offset = bio_offset;
834
835         async->status = 0;
836
837         if (op_is_sync(bio->bi_opf))
838                 btrfs_set_work_high_priority(&async->work);
839
840         btrfs_queue_work(fs_info->workers, &async->work);
841         return 0;
842 }
843
844 static blk_status_t btree_csum_one_bio(struct bio *bio)
845 {
846         struct bio_vec *bvec;
847         struct btrfs_root *root;
848         int ret = 0;
849         struct bvec_iter_all iter_all;
850
851         ASSERT(!bio_flagged(bio, BIO_CLONED));
852         bio_for_each_segment_all(bvec, bio, iter_all) {
853                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
854                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
855                 if (ret)
856                         break;
857         }
858
859         return errno_to_blk_status(ret);
860 }
861
862 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
863                                              u64 bio_offset)
864 {
865         /*
866          * when we're called for a write, we're already in the async
867          * submission context.  Just jump into btrfs_map_bio
868          */
869         return btree_csum_one_bio(bio);
870 }
871
872 static int check_async_write(struct btrfs_fs_info *fs_info,
873                              struct btrfs_inode *bi)
874 {
875         if (atomic_read(&bi->sync_writers))
876                 return 0;
877         if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
878                 return 0;
879         return 1;
880 }
881
882 static blk_status_t btree_submit_bio_hook(struct inode *inode, struct bio *bio,
883                                           int mirror_num,
884                                           unsigned long bio_flags)
885 {
886         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
887         int async = check_async_write(fs_info, BTRFS_I(inode));
888         blk_status_t ret;
889
890         if (bio_op(bio) != REQ_OP_WRITE) {
891                 /*
892                  * called for a read, do the setup so that checksum validation
893                  * can happen in the async kernel threads
894                  */
895                 ret = btrfs_bio_wq_end_io(fs_info, bio,
896                                           BTRFS_WQ_ENDIO_METADATA);
897                 if (ret)
898                         goto out_w_error;
899                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
900         } else if (!async) {
901                 ret = btree_csum_one_bio(bio);
902                 if (ret)
903                         goto out_w_error;
904                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
905         } else {
906                 /*
907                  * kthread helpers are used to submit writes so that
908                  * checksumming can happen in parallel across all CPUs
909                  */
910                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
911                                           0, inode, btree_submit_bio_start);
912         }
913
914         if (ret)
915                 goto out_w_error;
916         return 0;
917
918 out_w_error:
919         bio->bi_status = ret;
920         bio_endio(bio);
921         return ret;
922 }
923
924 #ifdef CONFIG_MIGRATION
925 static int btree_migratepage(struct address_space *mapping,
926                         struct page *newpage, struct page *page,
927                         enum migrate_mode mode)
928 {
929         /*
930          * we can't safely write a btree page from here,
931          * we haven't done the locking hook
932          */
933         if (PageDirty(page))
934                 return -EAGAIN;
935         /*
936          * Buffers may be managed in a filesystem specific way.
937          * We must have no buffers or drop them.
938          */
939         if (page_has_private(page) &&
940             !try_to_release_page(page, GFP_KERNEL))
941                 return -EAGAIN;
942         return migrate_page(mapping, newpage, page, mode);
943 }
944 #endif
945
946
947 static int btree_writepages(struct address_space *mapping,
948                             struct writeback_control *wbc)
949 {
950         struct btrfs_fs_info *fs_info;
951         int ret;
952
953         if (wbc->sync_mode == WB_SYNC_NONE) {
954
955                 if (wbc->for_kupdate)
956                         return 0;
957
958                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
959                 /* this is a bit racy, but that's ok */
960                 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
961                                              BTRFS_DIRTY_METADATA_THRESH,
962                                              fs_info->dirty_metadata_batch);
963                 if (ret < 0)
964                         return 0;
965         }
966         return btree_write_cache_pages(mapping, wbc);
967 }
968
969 static int btree_readpage(struct file *file, struct page *page)
970 {
971         struct extent_io_tree *tree;
972         tree = &BTRFS_I(page->mapping->host)->io_tree;
973         return extent_read_full_page(tree, page, btree_get_extent, 0);
974 }
975
976 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
977 {
978         if (PageWriteback(page) || PageDirty(page))
979                 return 0;
980
981         return try_release_extent_buffer(page);
982 }
983
984 static void btree_invalidatepage(struct page *page, unsigned int offset,
985                                  unsigned int length)
986 {
987         struct extent_io_tree *tree;
988         tree = &BTRFS_I(page->mapping->host)->io_tree;
989         extent_invalidatepage(tree, page, offset);
990         btree_releasepage(page, GFP_NOFS);
991         if (PagePrivate(page)) {
992                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
993                            "page private not zero on page %llu",
994                            (unsigned long long)page_offset(page));
995                 ClearPagePrivate(page);
996                 set_page_private(page, 0);
997                 put_page(page);
998         }
999 }
1000
1001 static int btree_set_page_dirty(struct page *page)
1002 {
1003 #ifdef DEBUG
1004         struct extent_buffer *eb;
1005
1006         BUG_ON(!PagePrivate(page));
1007         eb = (struct extent_buffer *)page->private;
1008         BUG_ON(!eb);
1009         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1010         BUG_ON(!atomic_read(&eb->refs));
1011         btrfs_assert_tree_locked(eb);
1012 #endif
1013         return __set_page_dirty_nobuffers(page);
1014 }
1015
1016 static const struct address_space_operations btree_aops = {
1017         .readpage       = btree_readpage,
1018         .writepages     = btree_writepages,
1019         .releasepage    = btree_releasepage,
1020         .invalidatepage = btree_invalidatepage,
1021 #ifdef CONFIG_MIGRATION
1022         .migratepage    = btree_migratepage,
1023 #endif
1024         .set_page_dirty = btree_set_page_dirty,
1025 };
1026
1027 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1028 {
1029         struct extent_buffer *buf = NULL;
1030         int ret;
1031
1032         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1033         if (IS_ERR(buf))
1034                 return;
1035
1036         ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
1037         if (ret < 0)
1038                 free_extent_buffer_stale(buf);
1039         else
1040                 free_extent_buffer(buf);
1041 }
1042
1043 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1044                          int mirror_num, struct extent_buffer **eb)
1045 {
1046         struct extent_buffer *buf = NULL;
1047         int ret;
1048
1049         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1050         if (IS_ERR(buf))
1051                 return 0;
1052
1053         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1054
1055         ret = read_extent_buffer_pages(buf, WAIT_PAGE_LOCK, 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                                      u64 objectid)
1257 {
1258         struct btrfs_fs_info *fs_info = trans->fs_info;
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_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2261 {
2262         struct crypto_shash *csum_shash;
2263         const char *csum_name = btrfs_super_csum_name(csum_type);
2264
2265         csum_shash = crypto_alloc_shash(csum_name, 0, 0);
2266
2267         if (IS_ERR(csum_shash)) {
2268                 btrfs_err(fs_info, "error allocating %s hash for checksum",
2269                           csum_name);
2270                 return PTR_ERR(csum_shash);
2271         }
2272
2273         fs_info->csum_shash = csum_shash;
2274
2275         return 0;
2276 }
2277
2278 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
2279 {
2280         crypto_free_shash(fs_info->csum_shash);
2281 }
2282
2283 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2284                             struct btrfs_fs_devices *fs_devices)
2285 {
2286         int ret;
2287         struct btrfs_root *log_tree_root;
2288         struct btrfs_super_block *disk_super = fs_info->super_copy;
2289         u64 bytenr = btrfs_super_log_root(disk_super);
2290         int level = btrfs_super_log_root_level(disk_super);
2291
2292         if (fs_devices->rw_devices == 0) {
2293                 btrfs_warn(fs_info, "log replay required on RO media");
2294                 return -EIO;
2295         }
2296
2297         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2298         if (!log_tree_root)
2299                 return -ENOMEM;
2300
2301         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2302
2303         log_tree_root->node = read_tree_block(fs_info, bytenr,
2304                                               fs_info->generation + 1,
2305                                               level, NULL);
2306         if (IS_ERR(log_tree_root->node)) {
2307                 btrfs_warn(fs_info, "failed to read log tree");
2308                 ret = PTR_ERR(log_tree_root->node);
2309                 kfree(log_tree_root);
2310                 return ret;
2311         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2312                 btrfs_err(fs_info, "failed to read log tree");
2313                 free_extent_buffer(log_tree_root->node);
2314                 kfree(log_tree_root);
2315                 return -EIO;
2316         }
2317         /* returns with log_tree_root freed on success */
2318         ret = btrfs_recover_log_trees(log_tree_root);
2319         if (ret) {
2320                 btrfs_handle_fs_error(fs_info, ret,
2321                                       "Failed to recover log tree");
2322                 free_extent_buffer(log_tree_root->node);
2323                 kfree(log_tree_root);
2324                 return ret;
2325         }
2326
2327         if (sb_rdonly(fs_info->sb)) {
2328                 ret = btrfs_commit_super(fs_info);
2329                 if (ret)
2330                         return ret;
2331         }
2332
2333         return 0;
2334 }
2335
2336 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2337 {
2338         struct btrfs_root *tree_root = fs_info->tree_root;
2339         struct btrfs_root *root;
2340         struct btrfs_key location;
2341         int ret;
2342
2343         BUG_ON(!fs_info->tree_root);
2344
2345         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2346         location.type = BTRFS_ROOT_ITEM_KEY;
2347         location.offset = 0;
2348
2349         root = btrfs_read_tree_root(tree_root, &location);
2350         if (IS_ERR(root)) {
2351                 ret = PTR_ERR(root);
2352                 goto out;
2353         }
2354         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2355         fs_info->extent_root = root;
2356
2357         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2358         root = btrfs_read_tree_root(tree_root, &location);
2359         if (IS_ERR(root)) {
2360                 ret = PTR_ERR(root);
2361                 goto out;
2362         }
2363         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2364         fs_info->dev_root = root;
2365         btrfs_init_devices_late(fs_info);
2366
2367         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2368         root = btrfs_read_tree_root(tree_root, &location);
2369         if (IS_ERR(root)) {
2370                 ret = PTR_ERR(root);
2371                 goto out;
2372         }
2373         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2374         fs_info->csum_root = root;
2375
2376         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2377         root = btrfs_read_tree_root(tree_root, &location);
2378         if (!IS_ERR(root)) {
2379                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2380                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2381                 fs_info->quota_root = root;
2382         }
2383
2384         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2385         root = btrfs_read_tree_root(tree_root, &location);
2386         if (IS_ERR(root)) {
2387                 ret = PTR_ERR(root);
2388                 if (ret != -ENOENT)
2389                         goto out;
2390         } else {
2391                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2392                 fs_info->uuid_root = root;
2393         }
2394
2395         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2396                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2397                 root = btrfs_read_tree_root(tree_root, &location);
2398                 if (IS_ERR(root)) {
2399                         ret = PTR_ERR(root);
2400                         goto out;
2401                 }
2402                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2403                 fs_info->free_space_root = root;
2404         }
2405
2406         return 0;
2407 out:
2408         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2409                    location.objectid, ret);
2410         return ret;
2411 }
2412
2413 /*
2414  * Real super block validation
2415  * NOTE: super csum type and incompat features will not be checked here.
2416  *
2417  * @sb:         super block to check
2418  * @mirror_num: the super block number to check its bytenr:
2419  *              0       the primary (1st) sb
2420  *              1, 2    2nd and 3rd backup copy
2421  *             -1       skip bytenr check
2422  */
2423 static int validate_super(struct btrfs_fs_info *fs_info,
2424                             struct btrfs_super_block *sb, int mirror_num)
2425 {
2426         u64 nodesize = btrfs_super_nodesize(sb);
2427         u64 sectorsize = btrfs_super_sectorsize(sb);
2428         int ret = 0;
2429
2430         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2431                 btrfs_err(fs_info, "no valid FS found");
2432                 ret = -EINVAL;
2433         }
2434         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2435                 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2436                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2437                 ret = -EINVAL;
2438         }
2439         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2440                 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2441                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2442                 ret = -EINVAL;
2443         }
2444         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2445                 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2446                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2447                 ret = -EINVAL;
2448         }
2449         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2450                 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2451                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2452                 ret = -EINVAL;
2453         }
2454
2455         /*
2456          * Check sectorsize and nodesize first, other check will need it.
2457          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2458          */
2459         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2460             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2461                 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2462                 ret = -EINVAL;
2463         }
2464         /* Only PAGE SIZE is supported yet */
2465         if (sectorsize != PAGE_SIZE) {
2466                 btrfs_err(fs_info,
2467                         "sectorsize %llu not supported yet, only support %lu",
2468                         sectorsize, PAGE_SIZE);
2469                 ret = -EINVAL;
2470         }
2471         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2472             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2473                 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2474                 ret = -EINVAL;
2475         }
2476         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2477                 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2478                           le32_to_cpu(sb->__unused_leafsize), nodesize);
2479                 ret = -EINVAL;
2480         }
2481
2482         /* Root alignment check */
2483         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2484                 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2485                            btrfs_super_root(sb));
2486                 ret = -EINVAL;
2487         }
2488         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2489                 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2490                            btrfs_super_chunk_root(sb));
2491                 ret = -EINVAL;
2492         }
2493         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2494                 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2495                            btrfs_super_log_root(sb));
2496                 ret = -EINVAL;
2497         }
2498
2499         if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2500                    BTRFS_FSID_SIZE) != 0) {
2501                 btrfs_err(fs_info,
2502                         "dev_item UUID does not match metadata fsid: %pU != %pU",
2503                         fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2504                 ret = -EINVAL;
2505         }
2506
2507         /*
2508          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2509          * done later
2510          */
2511         if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2512                 btrfs_err(fs_info, "bytes_used is too small %llu",
2513                           btrfs_super_bytes_used(sb));
2514                 ret = -EINVAL;
2515         }
2516         if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2517                 btrfs_err(fs_info, "invalid stripesize %u",
2518                           btrfs_super_stripesize(sb));
2519                 ret = -EINVAL;
2520         }
2521         if (btrfs_super_num_devices(sb) > (1UL << 31))
2522                 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2523                            btrfs_super_num_devices(sb));
2524         if (btrfs_super_num_devices(sb) == 0) {
2525                 btrfs_err(fs_info, "number of devices is 0");
2526                 ret = -EINVAL;
2527         }
2528
2529         if (mirror_num >= 0 &&
2530             btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2531                 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2532                           btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2533                 ret = -EINVAL;
2534         }
2535
2536         /*
2537          * Obvious sys_chunk_array corruptions, it must hold at least one key
2538          * and one chunk
2539          */
2540         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2541                 btrfs_err(fs_info, "system chunk array too big %u > %u",
2542                           btrfs_super_sys_array_size(sb),
2543                           BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2544                 ret = -EINVAL;
2545         }
2546         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2547                         + sizeof(struct btrfs_chunk)) {
2548                 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2549                           btrfs_super_sys_array_size(sb),
2550                           sizeof(struct btrfs_disk_key)
2551                           + sizeof(struct btrfs_chunk));
2552                 ret = -EINVAL;
2553         }
2554
2555         /*
2556          * The generation is a global counter, we'll trust it more than the others
2557          * but it's still possible that it's the one that's wrong.
2558          */
2559         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2560                 btrfs_warn(fs_info,
2561                         "suspicious: generation < chunk_root_generation: %llu < %llu",
2562                         btrfs_super_generation(sb),
2563                         btrfs_super_chunk_root_generation(sb));
2564         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2565             && btrfs_super_cache_generation(sb) != (u64)-1)
2566                 btrfs_warn(fs_info,
2567                         "suspicious: generation < cache_generation: %llu < %llu",
2568                         btrfs_super_generation(sb),
2569                         btrfs_super_cache_generation(sb));
2570
2571         return ret;
2572 }
2573
2574 /*
2575  * Validation of super block at mount time.
2576  * Some checks already done early at mount time, like csum type and incompat
2577  * flags will be skipped.
2578  */
2579 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2580 {
2581         return validate_super(fs_info, fs_info->super_copy, 0);
2582 }
2583
2584 /*
2585  * Validation of super block at write time.
2586  * Some checks like bytenr check will be skipped as their values will be
2587  * overwritten soon.
2588  * Extra checks like csum type and incompat flags will be done here.
2589  */
2590 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2591                                       struct btrfs_super_block *sb)
2592 {
2593         int ret;
2594
2595         ret = validate_super(fs_info, sb, -1);
2596         if (ret < 0)
2597                 goto out;
2598         if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2599                 ret = -EUCLEAN;
2600                 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2601                           btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2602                 goto out;
2603         }
2604         if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2605                 ret = -EUCLEAN;
2606                 btrfs_err(fs_info,
2607                 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2608                           btrfs_super_incompat_flags(sb),
2609                           (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2610                 goto out;
2611         }
2612 out:
2613         if (ret < 0)
2614                 btrfs_err(fs_info,
2615                 "super block corruption detected before writing it to disk");
2616         return ret;
2617 }
2618
2619 int open_ctree(struct super_block *sb,
2620                struct btrfs_fs_devices *fs_devices,
2621                char *options)
2622 {
2623         u32 sectorsize;
2624         u32 nodesize;
2625         u32 stripesize;
2626         u64 generation;
2627         u64 features;
2628         u16 csum_type;
2629         struct btrfs_key location;
2630         struct buffer_head *bh;
2631         struct btrfs_super_block *disk_super;
2632         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2633         struct btrfs_root *tree_root;
2634         struct btrfs_root *chunk_root;
2635         int ret;
2636         int err = -EINVAL;
2637         int num_backups_tried = 0;
2638         int backup_index = 0;
2639         int clear_free_space_tree = 0;
2640         int level;
2641
2642         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2643         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2644         if (!tree_root || !chunk_root) {
2645                 err = -ENOMEM;
2646                 goto fail;
2647         }
2648
2649         ret = init_srcu_struct(&fs_info->subvol_srcu);
2650         if (ret) {
2651                 err = ret;
2652                 goto fail;
2653         }
2654
2655         ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2656         if (ret) {
2657                 err = ret;
2658                 goto fail_srcu;
2659         }
2660
2661         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2662         if (ret) {
2663                 err = ret;
2664                 goto fail_dio_bytes;
2665         }
2666         fs_info->dirty_metadata_batch = PAGE_SIZE *
2667                                         (1 + ilog2(nr_cpu_ids));
2668
2669         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2670         if (ret) {
2671                 err = ret;
2672                 goto fail_dirty_metadata_bytes;
2673         }
2674
2675         ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2676                         GFP_KERNEL);
2677         if (ret) {
2678                 err = ret;
2679                 goto fail_delalloc_bytes;
2680         }
2681
2682         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2683         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2684         INIT_LIST_HEAD(&fs_info->trans_list);
2685         INIT_LIST_HEAD(&fs_info->dead_roots);
2686         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2687         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2688         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2689         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2690         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2691         spin_lock_init(&fs_info->delalloc_root_lock);
2692         spin_lock_init(&fs_info->trans_lock);
2693         spin_lock_init(&fs_info->fs_roots_radix_lock);
2694         spin_lock_init(&fs_info->delayed_iput_lock);
2695         spin_lock_init(&fs_info->defrag_inodes_lock);
2696         spin_lock_init(&fs_info->tree_mod_seq_lock);
2697         spin_lock_init(&fs_info->super_lock);
2698         spin_lock_init(&fs_info->buffer_lock);
2699         spin_lock_init(&fs_info->unused_bgs_lock);
2700         rwlock_init(&fs_info->tree_mod_log_lock);
2701         mutex_init(&fs_info->unused_bg_unpin_mutex);
2702         mutex_init(&fs_info->delete_unused_bgs_mutex);
2703         mutex_init(&fs_info->reloc_mutex);
2704         mutex_init(&fs_info->delalloc_root_mutex);
2705         seqlock_init(&fs_info->profiles_lock);
2706
2707         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2708         INIT_LIST_HEAD(&fs_info->space_info);
2709         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2710         INIT_LIST_HEAD(&fs_info->unused_bgs);
2711         extent_map_tree_init(&fs_info->mapping_tree);
2712         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2713                              BTRFS_BLOCK_RSV_GLOBAL);
2714         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2715         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2716         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2717         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2718                              BTRFS_BLOCK_RSV_DELOPS);
2719         btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2720                              BTRFS_BLOCK_RSV_DELREFS);
2721
2722         atomic_set(&fs_info->async_delalloc_pages, 0);
2723         atomic_set(&fs_info->defrag_running, 0);
2724         atomic_set(&fs_info->reada_works_cnt, 0);
2725         atomic_set(&fs_info->nr_delayed_iputs, 0);
2726         atomic64_set(&fs_info->tree_mod_seq, 0);
2727         fs_info->sb = sb;
2728         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2729         fs_info->metadata_ratio = 0;
2730         fs_info->defrag_inodes = RB_ROOT;
2731         atomic64_set(&fs_info->free_chunk_space, 0);
2732         fs_info->tree_mod_log = RB_ROOT;
2733         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2734         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2735         /* readahead state */
2736         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2737         spin_lock_init(&fs_info->reada_lock);
2738         btrfs_init_ref_verify(fs_info);
2739
2740         fs_info->thread_pool_size = min_t(unsigned long,
2741                                           num_online_cpus() + 2, 8);
2742
2743         INIT_LIST_HEAD(&fs_info->ordered_roots);
2744         spin_lock_init(&fs_info->ordered_root_lock);
2745
2746         fs_info->btree_inode = new_inode(sb);
2747         if (!fs_info->btree_inode) {
2748                 err = -ENOMEM;
2749                 goto fail_bio_counter;
2750         }
2751         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2752
2753         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2754                                         GFP_KERNEL);
2755         if (!fs_info->delayed_root) {
2756                 err = -ENOMEM;
2757                 goto fail_iput;
2758         }
2759         btrfs_init_delayed_root(fs_info->delayed_root);
2760
2761         btrfs_init_scrub(fs_info);
2762 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2763         fs_info->check_integrity_print_mask = 0;
2764 #endif
2765         btrfs_init_balance(fs_info);
2766         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2767
2768         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2769         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2770
2771         btrfs_init_btree_inode(fs_info);
2772
2773         spin_lock_init(&fs_info->block_group_cache_lock);
2774         fs_info->block_group_cache_tree = RB_ROOT;
2775         fs_info->first_logical_byte = (u64)-1;
2776
2777         extent_io_tree_init(fs_info, &fs_info->freed_extents[0],
2778                             IO_TREE_FS_INFO_FREED_EXTENTS0, NULL);
2779         extent_io_tree_init(fs_info, &fs_info->freed_extents[1],
2780                             IO_TREE_FS_INFO_FREED_EXTENTS1, NULL);
2781         fs_info->pinned_extents = &fs_info->freed_extents[0];
2782         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2783
2784         mutex_init(&fs_info->ordered_operations_mutex);
2785         mutex_init(&fs_info->tree_log_mutex);
2786         mutex_init(&fs_info->chunk_mutex);
2787         mutex_init(&fs_info->transaction_kthread_mutex);
2788         mutex_init(&fs_info->cleaner_mutex);
2789         mutex_init(&fs_info->ro_block_group_mutex);
2790         init_rwsem(&fs_info->commit_root_sem);
2791         init_rwsem(&fs_info->cleanup_work_sem);
2792         init_rwsem(&fs_info->subvol_sem);
2793         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2794
2795         btrfs_init_dev_replace_locks(fs_info);
2796         btrfs_init_qgroup(fs_info);
2797
2798         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2799         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2800
2801         init_waitqueue_head(&fs_info->transaction_throttle);
2802         init_waitqueue_head(&fs_info->transaction_wait);
2803         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2804         init_waitqueue_head(&fs_info->async_submit_wait);
2805         init_waitqueue_head(&fs_info->delayed_iputs_wait);
2806
2807         /* Usable values until the real ones are cached from the superblock */
2808         fs_info->nodesize = 4096;
2809         fs_info->sectorsize = 4096;
2810         fs_info->stripesize = 4096;
2811
2812         spin_lock_init(&fs_info->swapfile_pins_lock);
2813         fs_info->swapfile_pins = RB_ROOT;
2814
2815         ret = btrfs_alloc_stripe_hash_table(fs_info);
2816         if (ret) {
2817                 err = ret;
2818                 goto fail_alloc;
2819         }
2820
2821         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2822
2823         invalidate_bdev(fs_devices->latest_bdev);
2824
2825         /*
2826          * Read super block and check the signature bytes only
2827          */
2828         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2829         if (IS_ERR(bh)) {
2830                 err = PTR_ERR(bh);
2831                 goto fail_alloc;
2832         }
2833
2834         /*
2835          * Verify the type first, if that or the the checksum value are
2836          * corrupted, we'll find out
2837          */
2838         csum_type = btrfs_super_csum_type((struct btrfs_super_block *)bh->b_data);
2839         if (!btrfs_supported_super_csum(csum_type)) {
2840                 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2841                           csum_type);
2842                 err = -EINVAL;
2843                 brelse(bh);
2844                 goto fail_alloc;
2845         }
2846
2847         ret = btrfs_init_csum_hash(fs_info, csum_type);
2848         if (ret) {
2849                 err = ret;
2850                 goto fail_alloc;
2851         }
2852
2853         /*
2854          * We want to check superblock checksum, the type is stored inside.
2855          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2856          */
2857         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2858                 btrfs_err(fs_info, "superblock checksum mismatch");
2859                 err = -EINVAL;
2860                 brelse(bh);
2861                 goto fail_csum;
2862         }
2863
2864         /*
2865          * super_copy is zeroed at allocation time and we never touch the
2866          * following bytes up to INFO_SIZE, the checksum is calculated from
2867          * the whole block of INFO_SIZE
2868          */
2869         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2870         brelse(bh);
2871
2872         disk_super = fs_info->super_copy;
2873
2874         ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2875                        BTRFS_FSID_SIZE));
2876
2877         if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
2878                 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
2879                                 fs_info->super_copy->metadata_uuid,
2880                                 BTRFS_FSID_SIZE));
2881         }
2882
2883         features = btrfs_super_flags(disk_super);
2884         if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
2885                 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
2886                 btrfs_set_super_flags(disk_super, features);
2887                 btrfs_info(fs_info,
2888                         "found metadata UUID change in progress flag, clearing");
2889         }
2890
2891         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2892                sizeof(*fs_info->super_for_commit));
2893
2894         ret = btrfs_validate_mount_super(fs_info);
2895         if (ret) {
2896                 btrfs_err(fs_info, "superblock contains fatal errors");
2897                 err = -EINVAL;
2898                 goto fail_csum;
2899         }
2900
2901         if (!btrfs_super_root(disk_super))
2902                 goto fail_csum;
2903
2904         /* check FS state, whether FS is broken. */
2905         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2906                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2907
2908         /*
2909          * run through our array of backup supers and setup
2910          * our ring pointer to the oldest one
2911          */
2912         generation = btrfs_super_generation(disk_super);
2913         find_oldest_super_backup(fs_info, generation);
2914
2915         /*
2916          * In the long term, we'll store the compression type in the super
2917          * block, and it'll be used for per file compression control.
2918          */
2919         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2920
2921         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2922         if (ret) {
2923                 err = ret;
2924                 goto fail_csum;
2925         }
2926
2927         features = btrfs_super_incompat_flags(disk_super) &
2928                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2929         if (features) {
2930                 btrfs_err(fs_info,
2931                     "cannot mount because of unsupported optional features (%llx)",
2932                     features);
2933                 err = -EINVAL;
2934                 goto fail_csum;
2935         }
2936
2937         features = btrfs_super_incompat_flags(disk_super);
2938         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2939         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2940                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2941         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2942                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2943
2944         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2945                 btrfs_info(fs_info, "has skinny extents");
2946
2947         /*
2948          * flag our filesystem as having big metadata blocks if
2949          * they are bigger than the page size
2950          */
2951         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2952                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2953                         btrfs_info(fs_info,
2954                                 "flagging fs with big metadata feature");
2955                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2956         }
2957
2958         nodesize = btrfs_super_nodesize(disk_super);
2959         sectorsize = btrfs_super_sectorsize(disk_super);
2960         stripesize = sectorsize;
2961         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2962         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2963
2964         /* Cache block sizes */
2965         fs_info->nodesize = nodesize;
2966         fs_info->sectorsize = sectorsize;
2967         fs_info->stripesize = stripesize;
2968
2969         /*
2970          * mixed block groups end up with duplicate but slightly offset
2971          * extent buffers for the same range.  It leads to corruptions
2972          */
2973         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2974             (sectorsize != nodesize)) {
2975                 btrfs_err(fs_info,
2976 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2977                         nodesize, sectorsize);
2978                 goto fail_csum;
2979         }
2980
2981         /*
2982          * Needn't use the lock because there is no other task which will
2983          * update the flag.
2984          */
2985         btrfs_set_super_incompat_flags(disk_super, features);
2986
2987         features = btrfs_super_compat_ro_flags(disk_super) &
2988                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2989         if (!sb_rdonly(sb) && features) {
2990                 btrfs_err(fs_info,
2991         "cannot mount read-write because of unsupported optional features (%llx)",
2992                        features);
2993                 err = -EINVAL;
2994                 goto fail_csum;
2995         }
2996
2997         ret = btrfs_init_workqueues(fs_info, fs_devices);
2998         if (ret) {
2999                 err = ret;
3000                 goto fail_sb_buffer;
3001         }
3002
3003         sb->s_bdi->congested_fn = btrfs_congested_fn;
3004         sb->s_bdi->congested_data = fs_info;
3005         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
3006         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
3007         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3008         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3009
3010         sb->s_blocksize = sectorsize;
3011         sb->s_blocksize_bits = blksize_bits(sectorsize);
3012         memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3013
3014         mutex_lock(&fs_info->chunk_mutex);
3015         ret = btrfs_read_sys_array(fs_info);
3016         mutex_unlock(&fs_info->chunk_mutex);
3017         if (ret) {
3018                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3019                 goto fail_sb_buffer;
3020         }
3021
3022         generation = btrfs_super_chunk_root_generation(disk_super);
3023         level = btrfs_super_chunk_root_level(disk_super);
3024
3025         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
3026
3027         chunk_root->node = read_tree_block(fs_info,
3028                                            btrfs_super_chunk_root(disk_super),
3029                                            generation, level, NULL);
3030         if (IS_ERR(chunk_root->node) ||
3031             !extent_buffer_uptodate(chunk_root->node)) {
3032                 btrfs_err(fs_info, "failed to read chunk root");
3033                 if (!IS_ERR(chunk_root->node))
3034                         free_extent_buffer(chunk_root->node);
3035                 chunk_root->node = NULL;
3036                 goto fail_tree_roots;
3037         }
3038         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3039         chunk_root->commit_root = btrfs_root_node(chunk_root);
3040
3041         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3042            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
3043
3044         ret = btrfs_read_chunk_tree(fs_info);
3045         if (ret) {
3046                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3047                 goto fail_tree_roots;
3048         }
3049
3050         /*
3051          * Keep the devid that is marked to be the target device for the
3052          * device replace procedure
3053          */
3054         btrfs_free_extra_devids(fs_devices, 0);
3055
3056         if (!fs_devices->latest_bdev) {
3057                 btrfs_err(fs_info, "failed to read devices");
3058                 goto fail_tree_roots;
3059         }
3060
3061 retry_root_backup:
3062         generation = btrfs_super_generation(disk_super);
3063         level = btrfs_super_root_level(disk_super);
3064
3065         tree_root->node = read_tree_block(fs_info,
3066                                           btrfs_super_root(disk_super),
3067                                           generation, level, NULL);
3068         if (IS_ERR(tree_root->node) ||
3069             !extent_buffer_uptodate(tree_root->node)) {
3070                 btrfs_warn(fs_info, "failed to read tree root");
3071                 if (!IS_ERR(tree_root->node))
3072                         free_extent_buffer(tree_root->node);
3073                 tree_root->node = NULL;
3074                 goto recovery_tree_root;
3075         }
3076
3077         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
3078         tree_root->commit_root = btrfs_root_node(tree_root);
3079         btrfs_set_root_refs(&tree_root->root_item, 1);
3080
3081         mutex_lock(&tree_root->objectid_mutex);
3082         ret = btrfs_find_highest_objectid(tree_root,
3083                                         &tree_root->highest_objectid);
3084         if (ret) {
3085                 mutex_unlock(&tree_root->objectid_mutex);
3086                 goto recovery_tree_root;
3087         }
3088
3089         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3090
3091         mutex_unlock(&tree_root->objectid_mutex);
3092
3093         ret = btrfs_read_roots(fs_info);
3094         if (ret)
3095                 goto recovery_tree_root;
3096
3097         fs_info->generation = generation;
3098         fs_info->last_trans_committed = generation;
3099
3100         ret = btrfs_verify_dev_extents(fs_info);
3101         if (ret) {
3102                 btrfs_err(fs_info,
3103                           "failed to verify dev extents against chunks: %d",
3104                           ret);
3105                 goto fail_block_groups;
3106         }
3107         ret = btrfs_recover_balance(fs_info);
3108         if (ret) {
3109                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3110                 goto fail_block_groups;
3111         }
3112
3113         ret = btrfs_init_dev_stats(fs_info);
3114         if (ret) {
3115                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3116                 goto fail_block_groups;
3117         }
3118
3119         ret = btrfs_init_dev_replace(fs_info);
3120         if (ret) {
3121                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3122                 goto fail_block_groups;
3123         }
3124
3125         btrfs_free_extra_devids(fs_devices, 1);
3126
3127         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3128         if (ret) {
3129                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3130                                 ret);
3131                 goto fail_block_groups;
3132         }
3133
3134         ret = btrfs_sysfs_add_device(fs_devices);
3135         if (ret) {
3136                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3137                                 ret);
3138                 goto fail_fsdev_sysfs;
3139         }
3140
3141         ret = btrfs_sysfs_add_mounted(fs_info);
3142         if (ret) {
3143                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3144                 goto fail_fsdev_sysfs;
3145         }
3146
3147         ret = btrfs_init_space_info(fs_info);
3148         if (ret) {
3149                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3150                 goto fail_sysfs;
3151         }
3152
3153         ret = btrfs_read_block_groups(fs_info);
3154         if (ret) {
3155                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3156                 goto fail_sysfs;
3157         }
3158
3159         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3160                 btrfs_warn(fs_info,
3161                 "writable mount is not allowed due to too many missing devices");
3162                 goto fail_sysfs;
3163         }
3164
3165         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3166                                                "btrfs-cleaner");
3167         if (IS_ERR(fs_info->cleaner_kthread))
3168                 goto fail_sysfs;
3169
3170         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3171                                                    tree_root,
3172                                                    "btrfs-transaction");
3173         if (IS_ERR(fs_info->transaction_kthread))
3174                 goto fail_cleaner;
3175
3176         if (!btrfs_test_opt(fs_info, NOSSD) &&
3177             !fs_info->fs_devices->rotating) {
3178                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3179         }
3180
3181         /*
3182          * Mount does not set all options immediately, we can do it now and do
3183          * not have to wait for transaction commit
3184          */
3185         btrfs_apply_pending_changes(fs_info);
3186
3187 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3188         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3189                 ret = btrfsic_mount(fs_info, fs_devices,
3190                                     btrfs_test_opt(fs_info,
3191                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3192                                     1 : 0,
3193                                     fs_info->check_integrity_print_mask);
3194                 if (ret)
3195                         btrfs_warn(fs_info,
3196                                 "failed to initialize integrity check module: %d",
3197                                 ret);
3198         }
3199 #endif
3200         ret = btrfs_read_qgroup_config(fs_info);
3201         if (ret)
3202                 goto fail_trans_kthread;
3203
3204         if (btrfs_build_ref_tree(fs_info))
3205                 btrfs_err(fs_info, "couldn't build ref tree");
3206
3207         /* do not make disk changes in broken FS or nologreplay is given */
3208         if (btrfs_super_log_root(disk_super) != 0 &&
3209             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3210                 ret = btrfs_replay_log(fs_info, fs_devices);
3211                 if (ret) {
3212                         err = ret;
3213                         goto fail_qgroup;
3214                 }
3215         }
3216
3217         ret = btrfs_find_orphan_roots(fs_info);
3218         if (ret)
3219                 goto fail_qgroup;
3220
3221         if (!sb_rdonly(sb)) {
3222                 ret = btrfs_cleanup_fs_roots(fs_info);
3223                 if (ret)
3224                         goto fail_qgroup;
3225
3226                 mutex_lock(&fs_info->cleaner_mutex);
3227                 ret = btrfs_recover_relocation(tree_root);
3228                 mutex_unlock(&fs_info->cleaner_mutex);
3229                 if (ret < 0) {
3230                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3231                                         ret);
3232                         err = -EINVAL;
3233                         goto fail_qgroup;
3234                 }
3235         }
3236
3237         location.objectid = BTRFS_FS_TREE_OBJECTID;
3238         location.type = BTRFS_ROOT_ITEM_KEY;
3239         location.offset = 0;
3240
3241         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3242         if (IS_ERR(fs_info->fs_root)) {
3243                 err = PTR_ERR(fs_info->fs_root);
3244                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3245                 goto fail_qgroup;
3246         }
3247
3248         if (sb_rdonly(sb))
3249                 return 0;
3250
3251         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3252             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3253                 clear_free_space_tree = 1;
3254         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3255                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3256                 btrfs_warn(fs_info, "free space tree is invalid");
3257                 clear_free_space_tree = 1;
3258         }
3259
3260         if (clear_free_space_tree) {
3261                 btrfs_info(fs_info, "clearing free space tree");
3262                 ret = btrfs_clear_free_space_tree(fs_info);
3263                 if (ret) {
3264                         btrfs_warn(fs_info,
3265                                    "failed to clear free space tree: %d"