60caa68c3618d5a072f332be97e7db4ea7ba67b0
[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/scatterlist.h>
9 #include <linux/swap.h>
10 #include <linux/radix-tree.h>
11 #include <linux/writeback.h>
12 #include <linux/buffer_head.h>
13 #include <linux/workqueue.h>
14 #include <linux/kthread.h>
15 #include <linux/slab.h>
16 #include <linux/migrate.h>
17 #include <linux/ratelimit.h>
18 #include <linux/uuid.h>
19 #include <linux/semaphore.h>
20 #include <linux/error-injection.h>
21 #include <linux/crc32c.h>
22 #include <asm/unaligned.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 #ifdef CONFIG_X86
45 #include <asm/cpufeature.h>
46 #endif
47
48 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
49                                  BTRFS_HEADER_FLAG_RELOC |\
50                                  BTRFS_SUPER_FLAG_ERROR |\
51                                  BTRFS_SUPER_FLAG_SEEDING |\
52                                  BTRFS_SUPER_FLAG_METADUMP |\
53                                  BTRFS_SUPER_FLAG_METADUMP_V2)
54
55 static const struct extent_io_ops btree_extent_io_ops;
56 static void end_workqueue_fn(struct btrfs_work *work);
57 static void free_fs_root(struct btrfs_root *root);
58 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info);
59 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
60 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
61                                       struct btrfs_fs_info *fs_info);
62 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
63 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
64                                         struct extent_io_tree *dirty_pages,
65                                         int mark);
66 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
67                                        struct extent_io_tree *pinned_extents);
68 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
69 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
70
71 /*
72  * btrfs_end_io_wq structs are used to do processing in task context when an IO
73  * is complete.  This is used during reads to verify checksums, and it is used
74  * by writes to insert metadata for new file extents after IO is complete.
75  */
76 struct btrfs_end_io_wq {
77         struct bio *bio;
78         bio_end_io_t *end_io;
79         void *private;
80         struct btrfs_fs_info *info;
81         blk_status_t status;
82         enum btrfs_wq_endio_type metadata;
83         struct btrfs_work work;
84 };
85
86 static struct kmem_cache *btrfs_end_io_wq_cache;
87
88 int __init btrfs_end_io_wq_init(void)
89 {
90         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
91                                         sizeof(struct btrfs_end_io_wq),
92                                         0,
93                                         SLAB_MEM_SPREAD,
94                                         NULL);
95         if (!btrfs_end_io_wq_cache)
96                 return -ENOMEM;
97         return 0;
98 }
99
100 void __cold btrfs_end_io_wq_exit(void)
101 {
102         kmem_cache_destroy(btrfs_end_io_wq_cache);
103 }
104
105 /*
106  * async submit bios are used to offload expensive checksumming
107  * onto the worker threads.  They checksum file and metadata bios
108  * just before they are sent down the IO stack.
109  */
110 struct async_submit_bio {
111         void *private_data;
112         struct btrfs_fs_info *fs_info;
113         struct bio *bio;
114         extent_submit_bio_start_t *submit_bio_start;
115         extent_submit_bio_done_t *submit_bio_done;
116         int mirror_num;
117         unsigned long bio_flags;
118         /*
119          * bio_offset is optional, can be used if the pages in the bio
120          * can't tell us where in the file the bio should go
121          */
122         u64 bio_offset;
123         struct btrfs_work work;
124         blk_status_t status;
125 };
126
127 /*
128  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
129  * eb, the lockdep key is determined by the btrfs_root it belongs to and
130  * the level the eb occupies in the tree.
131  *
132  * Different roots are used for different purposes and may nest inside each
133  * other and they require separate keysets.  As lockdep keys should be
134  * static, assign keysets according to the purpose of the root as indicated
135  * by btrfs_root->objectid.  This ensures that all special purpose roots
136  * have separate keysets.
137  *
138  * Lock-nesting across peer nodes is always done with the immediate parent
139  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
140  * subclass to avoid triggering lockdep warning in such cases.
141  *
142  * The key is set by the readpage_end_io_hook after the buffer has passed
143  * csum validation but before the pages are unlocked.  It is also set by
144  * btrfs_init_new_buffer on freshly allocated blocks.
145  *
146  * We also add a check to make sure the highest level of the tree is the
147  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
148  * needs update as well.
149  */
150 #ifdef CONFIG_DEBUG_LOCK_ALLOC
151 # if BTRFS_MAX_LEVEL != 8
152 #  error
153 # endif
154
155 static struct btrfs_lockdep_keyset {
156         u64                     id;             /* root objectid */
157         const char              *name_stem;     /* lock name stem */
158         char                    names[BTRFS_MAX_LEVEL + 1][20];
159         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
160 } btrfs_lockdep_keysets[] = {
161         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
162         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
163         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
164         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
165         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
166         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
167         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
168         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
169         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
170         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
171         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
172         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
173         { .id = 0,                              .name_stem = "tree"     },
174 };
175
176 void __init btrfs_init_lockdep(void)
177 {
178         int i, j;
179
180         /* initialize lockdep class names */
181         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
182                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
183
184                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
185                         snprintf(ks->names[j], sizeof(ks->names[j]),
186                                  "btrfs-%s-%02d", ks->name_stem, j);
187         }
188 }
189
190 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
191                                     int level)
192 {
193         struct btrfs_lockdep_keyset *ks;
194
195         BUG_ON(level >= ARRAY_SIZE(ks->keys));
196
197         /* find the matching keyset, id 0 is the default entry */
198         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
199                 if (ks->id == objectid)
200                         break;
201
202         lockdep_set_class_and_name(&eb->lock,
203                                    &ks->keys[level], ks->names[level]);
204 }
205
206 #endif
207
208 /*
209  * extents on the btree inode are pretty simple, there's one extent
210  * that covers the entire device
211  */
212 struct extent_map *btree_get_extent(struct btrfs_inode *inode,
213                 struct page *page, size_t pg_offset, u64 start, u64 len,
214                 int create)
215 {
216         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
217         struct extent_map_tree *em_tree = &inode->extent_tree;
218         struct extent_map *em;
219         int ret;
220
221         read_lock(&em_tree->lock);
222         em = lookup_extent_mapping(em_tree, start, len);
223         if (em) {
224                 em->bdev = fs_info->fs_devices->latest_bdev;
225                 read_unlock(&em_tree->lock);
226                 goto out;
227         }
228         read_unlock(&em_tree->lock);
229
230         em = alloc_extent_map();
231         if (!em) {
232                 em = ERR_PTR(-ENOMEM);
233                 goto out;
234         }
235         em->start = 0;
236         em->len = (u64)-1;
237         em->block_len = (u64)-1;
238         em->block_start = 0;
239         em->bdev = fs_info->fs_devices->latest_bdev;
240
241         write_lock(&em_tree->lock);
242         ret = add_extent_mapping(em_tree, em, 0);
243         if (ret == -EEXIST) {
244                 free_extent_map(em);
245                 em = lookup_extent_mapping(em_tree, start, len);
246                 if (!em)
247                         em = ERR_PTR(-EIO);
248         } else if (ret) {
249                 free_extent_map(em);
250                 em = ERR_PTR(ret);
251         }
252         write_unlock(&em_tree->lock);
253
254 out:
255         return em;
256 }
257
258 u32 btrfs_csum_data(const char *data, u32 seed, size_t len)
259 {
260         return crc32c(seed, data, len);
261 }
262
263 void btrfs_csum_final(u32 crc, u8 *result)
264 {
265         put_unaligned_le32(~crc, result);
266 }
267
268 /*
269  * compute the csum for a btree block, and either verify it or write it
270  * into the csum field of the block.
271  */
272 static int csum_tree_block(struct btrfs_fs_info *fs_info,
273                            struct extent_buffer *buf,
274                            int verify)
275 {
276         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
277         char result[BTRFS_CSUM_SIZE];
278         unsigned long len;
279         unsigned long cur_len;
280         unsigned long offset = BTRFS_CSUM_SIZE;
281         char *kaddr;
282         unsigned long map_start;
283         unsigned long map_len;
284         int err;
285         u32 crc = ~(u32)0;
286
287         len = buf->len - offset;
288         while (len > 0) {
289                 err = map_private_extent_buffer(buf, offset, 32,
290                                         &kaddr, &map_start, &map_len);
291                 if (err)
292                         return err;
293                 cur_len = min(len, map_len - (offset - map_start));
294                 crc = btrfs_csum_data(kaddr + offset - map_start,
295                                       crc, cur_len);
296                 len -= cur_len;
297                 offset += cur_len;
298         }
299         memset(result, 0, BTRFS_CSUM_SIZE);
300
301         btrfs_csum_final(crc, result);
302
303         if (verify) {
304                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
305                         u32 val;
306                         u32 found = 0;
307                         memcpy(&found, result, csum_size);
308
309                         read_extent_buffer(buf, &val, 0, csum_size);
310                         btrfs_warn_rl(fs_info,
311                                 "%s checksum verify failed on %llu wanted %X found %X level %d",
312                                 fs_info->sb->s_id, buf->start,
313                                 val, found, btrfs_header_level(buf));
314                         return -EUCLEAN;
315                 }
316         } else {
317                 write_extent_buffer(buf, result, 0, csum_size);
318         }
319
320         return 0;
321 }
322
323 /*
324  * we can't consider a given block up to date unless the transid of the
325  * block matches the transid in the parent node's pointer.  This is how we
326  * detect blocks that either didn't get written at all or got written
327  * in the wrong place.
328  */
329 static int verify_parent_transid(struct extent_io_tree *io_tree,
330                                  struct extent_buffer *eb, u64 parent_transid,
331                                  int atomic)
332 {
333         struct extent_state *cached_state = NULL;
334         int ret;
335         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
336
337         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
338                 return 0;
339
340         if (atomic)
341                 return -EAGAIN;
342
343         if (need_lock) {
344                 btrfs_tree_read_lock(eb);
345                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
346         }
347
348         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
349                          &cached_state);
350         if (extent_buffer_uptodate(eb) &&
351             btrfs_header_generation(eb) == parent_transid) {
352                 ret = 0;
353                 goto out;
354         }
355         btrfs_err_rl(eb->fs_info,
356                 "parent transid verify failed on %llu wanted %llu found %llu",
357                         eb->start,
358                         parent_transid, btrfs_header_generation(eb));
359         ret = 1;
360
361         /*
362          * Things reading via commit roots that don't have normal protection,
363          * like send, can have a really old block in cache that may point at a
364          * block that has been freed and re-allocated.  So don't clear uptodate
365          * if we find an eb that is under IO (dirty/writeback) because we could
366          * end up reading in the stale data and then writing it back out and
367          * making everybody very sad.
368          */
369         if (!extent_buffer_under_io(eb))
370                 clear_extent_buffer_uptodate(eb);
371 out:
372         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
373                              &cached_state);
374         if (need_lock)
375                 btrfs_tree_read_unlock_blocking(eb);
376         return ret;
377 }
378
379 /*
380  * Return 0 if the superblock checksum type matches the checksum value of that
381  * algorithm. Pass the raw disk superblock data.
382  */
383 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
384                                   char *raw_disk_sb)
385 {
386         struct btrfs_super_block *disk_sb =
387                 (struct btrfs_super_block *)raw_disk_sb;
388         u16 csum_type = btrfs_super_csum_type(disk_sb);
389         int ret = 0;
390
391         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
392                 u32 crc = ~(u32)0;
393                 char result[sizeof(crc)];
394
395                 /*
396                  * The super_block structure does not span the whole
397                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
398                  * is filled with zeros and is included in the checksum.
399                  */
400                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
401                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
402                 btrfs_csum_final(crc, result);
403
404                 if (memcmp(raw_disk_sb, result, sizeof(result)))
405                         ret = 1;
406         }
407
408         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
409                 btrfs_err(fs_info, "unsupported checksum algorithm %u",
410                                 csum_type);
411                 ret = 1;
412         }
413
414         return ret;
415 }
416
417 static int verify_level_key(struct btrfs_fs_info *fs_info,
418                             struct extent_buffer *eb, int level,
419                             struct btrfs_key *first_key)
420 {
421         int found_level;
422         struct btrfs_key found_key;
423         int ret;
424
425         found_level = btrfs_header_level(eb);
426         if (found_level != level) {
427 #ifdef CONFIG_BTRFS_DEBUG
428                 WARN_ON(1);
429                 btrfs_err(fs_info,
430 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
431                           eb->start, level, found_level);
432 #endif
433                 return -EIO;
434         }
435
436         if (!first_key)
437                 return 0;
438
439         /*
440          * For live tree block (new tree blocks in current transaction),
441          * we need proper lock context to avoid race, which is impossible here.
442          * So we only checks tree blocks which is read from disk, whose
443          * generation <= fs_info->last_trans_committed.
444          */
445         if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
446                 return 0;
447         if (found_level)
448                 btrfs_node_key_to_cpu(eb, &found_key, 0);
449         else
450                 btrfs_item_key_to_cpu(eb, &found_key, 0);
451         ret = btrfs_comp_cpu_keys(first_key, &found_key);
452
453 #ifdef CONFIG_BTRFS_DEBUG
454         if (ret) {
455                 WARN_ON(1);
456                 btrfs_err(fs_info,
457 "tree first key mismatch detected, bytenr=%llu key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
458                           eb->start, first_key->objectid, first_key->type,
459                           first_key->offset, found_key.objectid,
460                           found_key.type, found_key.offset);
461         }
462 #endif
463         return ret;
464 }
465
466 /*
467  * helper to read a given tree block, doing retries as required when
468  * the checksums don't match and we have alternate mirrors to try.
469  *
470  * @parent_transid:     expected transid, skip check if 0
471  * @level:              expected level, mandatory check
472  * @first_key:          expected key of first slot, skip check if NULL
473  */
474 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
475                                           struct extent_buffer *eb,
476                                           u64 parent_transid, int level,
477                                           struct btrfs_key *first_key)
478 {
479         struct extent_io_tree *io_tree;
480         int failed = 0;
481         int ret;
482         int num_copies = 0;
483         int mirror_num = 0;
484         int failed_mirror = 0;
485
486         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
487         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
488         while (1) {
489                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
490                                                mirror_num);
491                 if (!ret) {
492                         if (verify_parent_transid(io_tree, eb,
493                                                    parent_transid, 0))
494                                 ret = -EIO;
495                         else if (verify_level_key(fs_info, eb, level,
496                                                   first_key))
497                                 ret = -EUCLEAN;
498                         else
499                                 break;
500                 }
501
502                 /*
503                  * This buffer's crc is fine, but its contents are corrupted, so
504                  * there is no reason to read the other copies, they won't be
505                  * any less wrong.
506                  */
507                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
508                     ret == -EUCLEAN)
509                         break;
510
511                 num_copies = btrfs_num_copies(fs_info,
512                                               eb->start, eb->len);
513                 if (num_copies == 1)
514                         break;
515
516                 if (!failed_mirror) {
517                         failed = 1;
518                         failed_mirror = eb->read_mirror;
519                 }
520
521                 mirror_num++;
522                 if (mirror_num == failed_mirror)
523                         mirror_num++;
524
525                 if (mirror_num > num_copies)
526                         break;
527         }
528
529         if (failed && !ret && failed_mirror)
530                 repair_eb_io_failure(fs_info, eb, failed_mirror);
531
532         return ret;
533 }
534
535 /*
536  * checksum a dirty tree block before IO.  This has extra checks to make sure
537  * we only fill in the checksum field in the first page of a multi-page block
538  */
539
540 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
541 {
542         u64 start = page_offset(page);
543         u64 found_start;
544         struct extent_buffer *eb;
545
546         eb = (struct extent_buffer *)page->private;
547         if (page != eb->pages[0])
548                 return 0;
549
550         found_start = btrfs_header_bytenr(eb);
551         /*
552          * Please do not consolidate these warnings into a single if.
553          * It is useful to know what went wrong.
554          */
555         if (WARN_ON(found_start != start))
556                 return -EUCLEAN;
557         if (WARN_ON(!PageUptodate(page)))
558                 return -EUCLEAN;
559
560         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
561                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
562
563         return csum_tree_block(fs_info, eb, 0);
564 }
565
566 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
567                                  struct extent_buffer *eb)
568 {
569         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
570         u8 fsid[BTRFS_FSID_SIZE];
571         int ret = 1;
572
573         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
574         while (fs_devices) {
575                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
576                         ret = 0;
577                         break;
578                 }
579                 fs_devices = fs_devices->seed;
580         }
581         return ret;
582 }
583
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585                                       u64 phy_offset, struct page *page,
586                                       u64 start, u64 end, int mirror)
587 {
588         u64 found_start;
589         int found_level;
590         struct extent_buffer *eb;
591         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
592         struct btrfs_fs_info *fs_info = root->fs_info;
593         int ret = 0;
594         int reads_done;
595
596         if (!page->private)
597                 goto out;
598
599         eb = (struct extent_buffer *)page->private;
600
601         /* the pending IO might have been the only thing that kept this buffer
602          * in memory.  Make sure we have a ref for all this other checks
603          */
604         extent_buffer_get(eb);
605
606         reads_done = atomic_dec_and_test(&eb->io_pages);
607         if (!reads_done)
608                 goto err;
609
610         eb->read_mirror = mirror;
611         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
612                 ret = -EIO;
613                 goto err;
614         }
615
616         found_start = btrfs_header_bytenr(eb);
617         if (found_start != eb->start) {
618                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
619                              found_start, eb->start);
620                 ret = -EIO;
621                 goto err;
622         }
623         if (check_tree_block_fsid(fs_info, eb)) {
624                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
625                              eb->start);
626                 ret = -EIO;
627                 goto err;
628         }
629         found_level = btrfs_header_level(eb);
630         if (found_level >= BTRFS_MAX_LEVEL) {
631                 btrfs_err(fs_info, "bad tree block level %d",
632                           (int)btrfs_header_level(eb));
633                 ret = -EIO;
634                 goto err;
635         }
636
637         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
638                                        eb, found_level);
639
640         ret = csum_tree_block(fs_info, eb, 1);
641         if (ret)
642                 goto err;
643
644         /*
645          * If this is a leaf block and it is corrupt, set the corrupt bit so
646          * that we don't try and read the other copies of this block, just
647          * return -EIO.
648          */
649         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
650                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651                 ret = -EIO;
652         }
653
654         if (found_level > 0 && btrfs_check_node(fs_info, eb))
655                 ret = -EIO;
656
657         if (!ret)
658                 set_extent_buffer_uptodate(eb);
659 err:
660         if (reads_done &&
661             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662                 btree_readahead_hook(eb, ret);
663
664         if (ret) {
665                 /*
666                  * our io error hook is going to dec the io pages
667                  * again, we have to make sure it has something
668                  * to decrement
669                  */
670                 atomic_inc(&eb->io_pages);
671                 clear_extent_buffer_uptodate(eb);
672         }
673         free_extent_buffer(eb);
674 out:
675         return ret;
676 }
677
678 static int btree_io_failed_hook(struct page *page, int failed_mirror)
679 {
680         struct extent_buffer *eb;
681
682         eb = (struct extent_buffer *)page->private;
683         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
684         eb->read_mirror = failed_mirror;
685         atomic_dec(&eb->io_pages);
686         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687                 btree_readahead_hook(eb, -EIO);
688         return -EIO;    /* we fixed nothing */
689 }
690
691 static void end_workqueue_bio(struct bio *bio)
692 {
693         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
694         struct btrfs_fs_info *fs_info;
695         struct btrfs_workqueue *wq;
696         btrfs_work_func_t func;
697
698         fs_info = end_io_wq->info;
699         end_io_wq->status = bio->bi_status;
700
701         if (bio_op(bio) == REQ_OP_WRITE) {
702                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703                         wq = fs_info->endio_meta_write_workers;
704                         func = btrfs_endio_meta_write_helper;
705                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706                         wq = fs_info->endio_freespace_worker;
707                         func = btrfs_freespace_write_helper;
708                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709                         wq = fs_info->endio_raid56_workers;
710                         func = btrfs_endio_raid56_helper;
711                 } else {
712                         wq = fs_info->endio_write_workers;
713                         func = btrfs_endio_write_helper;
714                 }
715         } else {
716                 if (unlikely(end_io_wq->metadata ==
717                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718                         wq = fs_info->endio_repair_workers;
719                         func = btrfs_endio_repair_helper;
720                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721                         wq = fs_info->endio_raid56_workers;
722                         func = btrfs_endio_raid56_helper;
723                 } else if (end_io_wq->metadata) {
724                         wq = fs_info->endio_meta_workers;
725                         func = btrfs_endio_meta_helper;
726                 } else {
727                         wq = fs_info->endio_workers;
728                         func = btrfs_endio_helper;
729                 }
730         }
731
732         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733         btrfs_queue_work(wq, &end_io_wq->work);
734 }
735
736 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737                         enum btrfs_wq_endio_type metadata)
738 {
739         struct btrfs_end_io_wq *end_io_wq;
740
741         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
742         if (!end_io_wq)
743                 return BLK_STS_RESOURCE;
744
745         end_io_wq->private = bio->bi_private;
746         end_io_wq->end_io = bio->bi_end_io;
747         end_io_wq->info = info;
748         end_io_wq->status = 0;
749         end_io_wq->bio = bio;
750         end_io_wq->metadata = metadata;
751
752         bio->bi_private = end_io_wq;
753         bio->bi_end_io = end_workqueue_bio;
754         return 0;
755 }
756
757 static void run_one_async_start(struct btrfs_work *work)
758 {
759         struct async_submit_bio *async;
760         blk_status_t ret;
761
762         async = container_of(work, struct  async_submit_bio, work);
763         ret = async->submit_bio_start(async->private_data, async->bio,
764                                       async->bio_offset);
765         if (ret)
766                 async->status = ret;
767 }
768
769 static void run_one_async_done(struct btrfs_work *work)
770 {
771         struct async_submit_bio *async;
772
773         async = container_of(work, struct  async_submit_bio, work);
774
775         /* If an error occurred we just want to clean up the bio and move on */
776         if (async->status) {
777                 async->bio->bi_status = async->status;
778                 bio_endio(async->bio);
779                 return;
780         }
781
782         async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
783 }
784
785 static void run_one_async_free(struct btrfs_work *work)
786 {
787         struct async_submit_bio *async;
788
789         async = container_of(work, struct  async_submit_bio, work);
790         kfree(async);
791 }
792
793 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
794                                  int mirror_num, unsigned long bio_flags,
795                                  u64 bio_offset, void *private_data,
796                                  extent_submit_bio_start_t *submit_bio_start,
797                                  extent_submit_bio_done_t *submit_bio_done)
798 {
799         struct async_submit_bio *async;
800
801         async = kmalloc(sizeof(*async), GFP_NOFS);
802         if (!async)
803                 return BLK_STS_RESOURCE;
804
805         async->private_data = private_data;
806         async->fs_info = fs_info;
807         async->bio = bio;
808         async->mirror_num = mirror_num;
809         async->submit_bio_start = submit_bio_start;
810         async->submit_bio_done = submit_bio_done;
811
812         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
813                         run_one_async_done, run_one_async_free);
814
815         async->bio_flags = bio_flags;
816         async->bio_offset = bio_offset;
817
818         async->status = 0;
819
820         if (op_is_sync(bio->bi_opf))
821                 btrfs_set_work_high_priority(&async->work);
822
823         btrfs_queue_work(fs_info->workers, &async->work);
824         return 0;
825 }
826
827 static blk_status_t btree_csum_one_bio(struct bio *bio)
828 {
829         struct bio_vec *bvec;
830         struct btrfs_root *root;
831         int i, ret = 0;
832
833         ASSERT(!bio_flagged(bio, BIO_CLONED));
834         bio_for_each_segment_all(bvec, bio, i) {
835                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
836                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
837                 if (ret)
838                         break;
839         }
840
841         return errno_to_blk_status(ret);
842 }
843
844 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
845                                              u64 bio_offset)
846 {
847         /*
848          * when we're called for a write, we're already in the async
849          * submission context.  Just jump into btrfs_map_bio
850          */
851         return btree_csum_one_bio(bio);
852 }
853
854 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
855                                             int mirror_num)
856 {
857         struct inode *inode = private_data;
858         blk_status_t ret;
859
860         /*
861          * when we're called for a write, we're already in the async
862          * submission context.  Just jump into btrfs_map_bio
863          */
864         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
865         if (ret) {
866                 bio->bi_status = ret;
867                 bio_endio(bio);
868         }
869         return ret;
870 }
871
872 static int check_async_write(struct btrfs_inode *bi)
873 {
874         if (atomic_read(&bi->sync_writers))
875                 return 0;
876 #ifdef CONFIG_X86
877         if (static_cpu_has(X86_FEATURE_XMM4_2))
878                 return 0;
879 #endif
880         return 1;
881 }
882
883 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
884                                           int mirror_num, unsigned long bio_flags,
885                                           u64 bio_offset)
886 {
887         struct inode *inode = private_data;
888         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
889         int async = check_async_write(BTRFS_I(inode));
890         blk_status_t ret;
891
892         if (bio_op(bio) != REQ_OP_WRITE) {
893                 /*
894                  * called for a read, do the setup so that checksum validation
895                  * can happen in the async kernel threads
896                  */
897                 ret = btrfs_bio_wq_end_io(fs_info, bio,
898                                           BTRFS_WQ_ENDIO_METADATA);
899                 if (ret)
900                         goto out_w_error;
901                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
902         } else if (!async) {
903                 ret = btree_csum_one_bio(bio);
904                 if (ret)
905                         goto out_w_error;
906                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
907         } else {
908                 /*
909                  * kthread helpers are used to submit writes so that
910                  * checksumming can happen in parallel across all CPUs
911                  */
912                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
913                                           bio_offset, private_data,
914                                           btree_submit_bio_start,
915                                           btree_submit_bio_done);
916         }
917
918         if (ret)
919                 goto out_w_error;
920         return 0;
921
922 out_w_error:
923         bio->bi_status = ret;
924         bio_endio(bio);
925         return ret;
926 }
927
928 #ifdef CONFIG_MIGRATION
929 static int btree_migratepage(struct address_space *mapping,
930                         struct page *newpage, struct page *page,
931                         enum migrate_mode mode)
932 {
933         /*
934          * we can't safely write a btree page from here,
935          * we haven't done the locking hook
936          */
937         if (PageDirty(page))
938                 return -EAGAIN;
939         /*
940          * Buffers may be managed in a filesystem specific way.
941          * We must have no buffers or drop them.
942          */
943         if (page_has_private(page) &&
944             !try_to_release_page(page, GFP_KERNEL))
945                 return -EAGAIN;
946         return migrate_page(mapping, newpage, page, mode);
947 }
948 #endif
949
950
951 static int btree_writepages(struct address_space *mapping,
952                             struct writeback_control *wbc)
953 {
954         struct btrfs_fs_info *fs_info;
955         int ret;
956
957         if (wbc->sync_mode == WB_SYNC_NONE) {
958
959                 if (wbc->for_kupdate)
960                         return 0;
961
962                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
963                 /* this is a bit racy, but that's ok */
964                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
965                                              BTRFS_DIRTY_METADATA_THRESH);
966                 if (ret < 0)
967                         return 0;
968         }
969         return btree_write_cache_pages(mapping, wbc);
970 }
971
972 static int btree_readpage(struct file *file, struct page *page)
973 {
974         struct extent_io_tree *tree;
975         tree = &BTRFS_I(page->mapping->host)->io_tree;
976         return extent_read_full_page(tree, page, btree_get_extent, 0);
977 }
978
979 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
980 {
981         if (PageWriteback(page) || PageDirty(page))
982                 return 0;
983
984         return try_release_extent_buffer(page);
985 }
986
987 static void btree_invalidatepage(struct page *page, unsigned int offset,
988                                  unsigned int length)
989 {
990         struct extent_io_tree *tree;
991         tree = &BTRFS_I(page->mapping->host)->io_tree;
992         extent_invalidatepage(tree, page, offset);
993         btree_releasepage(page, GFP_NOFS);
994         if (PagePrivate(page)) {
995                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
996                            "page private not zero on page %llu",
997                            (unsigned long long)page_offset(page));
998                 ClearPagePrivate(page);
999                 set_page_private(page, 0);
1000                 put_page(page);
1001         }
1002 }
1003
1004 static int btree_set_page_dirty(struct page *page)
1005 {
1006 #ifdef DEBUG
1007         struct extent_buffer *eb;
1008
1009         BUG_ON(!PagePrivate(page));
1010         eb = (struct extent_buffer *)page->private;
1011         BUG_ON(!eb);
1012         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1013         BUG_ON(!atomic_read(&eb->refs));
1014         btrfs_assert_tree_locked(eb);
1015 #endif
1016         return __set_page_dirty_nobuffers(page);
1017 }
1018
1019 static const struct address_space_operations btree_aops = {
1020         .readpage       = btree_readpage,
1021         .writepages     = btree_writepages,
1022         .releasepage    = btree_releasepage,
1023         .invalidatepage = btree_invalidatepage,
1024 #ifdef CONFIG_MIGRATION
1025         .migratepage    = btree_migratepage,
1026 #endif
1027         .set_page_dirty = btree_set_page_dirty,
1028 };
1029
1030 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1031 {
1032         struct extent_buffer *buf = NULL;
1033         struct inode *btree_inode = fs_info->btree_inode;
1034
1035         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1036         if (IS_ERR(buf))
1037                 return;
1038         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1039                                  buf, WAIT_NONE, 0);
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         struct inode *btree_inode = fs_info->btree_inode;
1048         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1049         int ret;
1050
1051         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1052         if (IS_ERR(buf))
1053                 return 0;
1054
1055         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1056
1057         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1058                                        mirror_num);
1059         if (ret) {
1060                 free_extent_buffer(buf);
1061                 return ret;
1062         }
1063
1064         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1065                 free_extent_buffer(buf);
1066                 return -EIO;
1067         } else if (extent_buffer_uptodate(buf)) {
1068                 *eb = buf;
1069         } else {
1070                 free_extent_buffer(buf);
1071         }
1072         return 0;
1073 }
1074
1075 struct extent_buffer *btrfs_find_create_tree_block(
1076                                                 struct btrfs_fs_info *fs_info,
1077                                                 u64 bytenr)
1078 {
1079         if (btrfs_is_testing(fs_info))
1080                 return alloc_test_extent_buffer(fs_info, bytenr);
1081         return alloc_extent_buffer(fs_info, bytenr);
1082 }
1083
1084
1085 int btrfs_write_tree_block(struct extent_buffer *buf)
1086 {
1087         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1088                                         buf->start + buf->len - 1);
1089 }
1090
1091 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1092 {
1093         filemap_fdatawait_range(buf->pages[0]->mapping,
1094                                 buf->start, buf->start + buf->len - 1);
1095 }
1096
1097 /*
1098  * Read tree block at logical address @bytenr and do variant basic but critical
1099  * verification.
1100  *
1101  * @parent_transid:     expected transid of this tree block, skip check if 0
1102  * @level:              expected level, mandatory check
1103  * @first_key:          expected key in slot 0, skip check if NULL
1104  */
1105 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1106                                       u64 parent_transid, int level,
1107                                       struct btrfs_key *first_key)
1108 {
1109         struct extent_buffer *buf = NULL;
1110         int ret;
1111
1112         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1113         if (IS_ERR(buf))
1114                 return buf;
1115
1116         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1117                                              level, first_key);
1118         if (ret) {
1119                 free_extent_buffer(buf);
1120                 return ERR_PTR(ret);
1121         }
1122         return buf;
1123
1124 }
1125
1126 void clean_tree_block(struct btrfs_fs_info *fs_info,
1127                       struct extent_buffer *buf)
1128 {
1129         if (btrfs_header_generation(buf) ==
1130             fs_info->running_transaction->transid) {
1131                 btrfs_assert_tree_locked(buf);
1132
1133                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1134                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1135                                                  -buf->len,
1136                                                  fs_info->dirty_metadata_batch);
1137                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1138                         btrfs_set_lock_blocking(buf);
1139                         clear_extent_buffer_dirty(buf);
1140                 }
1141         }
1142 }
1143
1144 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1145 {
1146         struct btrfs_subvolume_writers *writers;
1147         int ret;
1148
1149         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1150         if (!writers)
1151                 return ERR_PTR(-ENOMEM);
1152
1153         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1154         if (ret < 0) {
1155                 kfree(writers);
1156                 return ERR_PTR(ret);
1157         }
1158
1159         init_waitqueue_head(&writers->wait);
1160         return writers;
1161 }
1162
1163 static void
1164 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1165 {
1166         percpu_counter_destroy(&writers->counter);
1167         kfree(writers);
1168 }
1169
1170 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1171                          u64 objectid)
1172 {
1173         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1174         root->node = NULL;
1175         root->commit_root = NULL;
1176         root->state = 0;
1177         root->orphan_cleanup_state = 0;
1178
1179         root->objectid = objectid;
1180         root->last_trans = 0;
1181         root->highest_objectid = 0;
1182         root->nr_delalloc_inodes = 0;
1183         root->nr_ordered_extents = 0;
1184         root->name = NULL;
1185         root->inode_tree = RB_ROOT;
1186         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187         root->block_rsv = NULL;
1188         root->orphan_block_rsv = NULL;
1189
1190         INIT_LIST_HEAD(&root->dirty_list);
1191         INIT_LIST_HEAD(&root->root_list);
1192         INIT_LIST_HEAD(&root->delalloc_inodes);
1193         INIT_LIST_HEAD(&root->delalloc_root);
1194         INIT_LIST_HEAD(&root->ordered_extents);
1195         INIT_LIST_HEAD(&root->ordered_root);
1196         INIT_LIST_HEAD(&root->logged_list[0]);
1197         INIT_LIST_HEAD(&root->logged_list[1]);
1198         spin_lock_init(&root->orphan_lock);
1199         spin_lock_init(&root->inode_lock);
1200         spin_lock_init(&root->delalloc_lock);
1201         spin_lock_init(&root->ordered_extent_lock);
1202         spin_lock_init(&root->accounting_lock);
1203         spin_lock_init(&root->log_extents_lock[0]);
1204         spin_lock_init(&root->log_extents_lock[1]);
1205         spin_lock_init(&root->qgroup_meta_rsv_lock);
1206         mutex_init(&root->objectid_mutex);
1207         mutex_init(&root->log_mutex);
1208         mutex_init(&root->ordered_extent_mutex);
1209         mutex_init(&root->delalloc_mutex);
1210         init_waitqueue_head(&root->log_writer_wait);
1211         init_waitqueue_head(&root->log_commit_wait[0]);
1212         init_waitqueue_head(&root->log_commit_wait[1]);
1213         INIT_LIST_HEAD(&root->log_ctxs[0]);
1214         INIT_LIST_HEAD(&root->log_ctxs[1]);
1215         atomic_set(&root->log_commit[0], 0);
1216         atomic_set(&root->log_commit[1], 0);
1217         atomic_set(&root->log_writers, 0);
1218         atomic_set(&root->log_batch, 0);
1219         atomic_set(&root->orphan_inodes, 0);
1220         refcount_set(&root->refs, 1);
1221         atomic_set(&root->will_be_snapshotted, 0);
1222         root->log_transid = 0;
1223         root->log_transid_committed = -1;
1224         root->last_log_commit = 0;
1225         if (!dummy)
1226                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1227
1228         memset(&root->root_key, 0, sizeof(root->root_key));
1229         memset(&root->root_item, 0, sizeof(root->root_item));
1230         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1231         if (!dummy)
1232                 root->defrag_trans_start = fs_info->generation;
1233         else
1234                 root->defrag_trans_start = 0;
1235         root->root_key.objectid = objectid;
1236         root->anon_dev = 0;
1237
1238         spin_lock_init(&root->root_item_lock);
1239 }
1240
1241 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1242                 gfp_t flags)
1243 {
1244         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1245         if (root)
1246                 root->fs_info = fs_info;
1247         return root;
1248 }
1249
1250 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1251 /* Should only be used by the testing infrastructure */
1252 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1253 {
1254         struct btrfs_root *root;
1255
1256         if (!fs_info)
1257                 return ERR_PTR(-EINVAL);
1258
1259         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1260         if (!root)
1261                 return ERR_PTR(-ENOMEM);
1262
1263         /* We don't use the stripesize in selftest, set it as sectorsize */
1264         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1265         root->alloc_bytenr = 0;
1266
1267         return root;
1268 }
1269 #endif
1270
1271 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1272                                      struct btrfs_fs_info *fs_info,
1273                                      u64 objectid)
1274 {
1275         struct extent_buffer *leaf;
1276         struct btrfs_root *tree_root = fs_info->tree_root;
1277         struct btrfs_root *root;
1278         struct btrfs_key key;
1279         int ret = 0;
1280         uuid_le uuid = NULL_UUID_LE;
1281
1282         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1283         if (!root)
1284                 return ERR_PTR(-ENOMEM);
1285
1286         __setup_root(root, fs_info, objectid);
1287         root->root_key.objectid = objectid;
1288         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289         root->root_key.offset = 0;
1290
1291         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1292         if (IS_ERR(leaf)) {
1293                 ret = PTR_ERR(leaf);
1294                 leaf = NULL;
1295                 goto fail;
1296         }
1297
1298         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1299         btrfs_set_header_bytenr(leaf, leaf->start);
1300         btrfs_set_header_generation(leaf, trans->transid);
1301         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1302         btrfs_set_header_owner(leaf, objectid);
1303         root->node = leaf;
1304
1305         write_extent_buffer_fsid(leaf, fs_info->fsid);
1306         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1307         btrfs_mark_buffer_dirty(leaf);
1308
1309         root->commit_root = btrfs_root_node(root);
1310         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1311
1312         root->root_item.flags = 0;
1313         root->root_item.byte_limit = 0;
1314         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315         btrfs_set_root_generation(&root->root_item, trans->transid);
1316         btrfs_set_root_level(&root->root_item, 0);
1317         btrfs_set_root_refs(&root->root_item, 1);
1318         btrfs_set_root_used(&root->root_item, leaf->len);
1319         btrfs_set_root_last_snapshot(&root->root_item, 0);
1320         btrfs_set_root_dirid(&root->root_item, 0);
1321         if (is_fstree(objectid))
1322                 uuid_le_gen(&uuid);
1323         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1324         root->root_item.drop_level = 0;
1325
1326         key.objectid = objectid;
1327         key.type = BTRFS_ROOT_ITEM_KEY;
1328         key.offset = 0;
1329         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1330         if (ret)
1331                 goto fail;
1332
1333         btrfs_tree_unlock(leaf);
1334
1335         return root;
1336
1337 fail:
1338         if (leaf) {
1339                 btrfs_tree_unlock(leaf);
1340                 free_extent_buffer(root->commit_root);
1341                 free_extent_buffer(leaf);
1342         }
1343         kfree(root);
1344
1345         return ERR_PTR(ret);
1346 }
1347
1348 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1349                                          struct btrfs_fs_info *fs_info)
1350 {
1351         struct btrfs_root *root;
1352         struct extent_buffer *leaf;
1353
1354         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1355         if (!root)
1356                 return ERR_PTR(-ENOMEM);
1357
1358         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1359
1360         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1363
1364         /*
1365          * DON'T set REF_COWS for log trees
1366          *
1367          * log trees do not get reference counted because they go away
1368          * before a real commit is actually done.  They do store pointers
1369          * to file data extents, and those reference counts still get
1370          * updated (along with back refs to the log tree).
1371          */
1372
1373         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1374                         NULL, 0, 0, 0);
1375         if (IS_ERR(leaf)) {
1376                 kfree(root);
1377                 return ERR_CAST(leaf);
1378         }
1379
1380         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1381         btrfs_set_header_bytenr(leaf, leaf->start);
1382         btrfs_set_header_generation(leaf, trans->transid);
1383         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1385         root->node = leaf;
1386
1387         write_extent_buffer_fsid(root->node, fs_info->fsid);
1388         btrfs_mark_buffer_dirty(root->node);
1389         btrfs_tree_unlock(root->node);
1390         return root;
1391 }
1392
1393 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1394                              struct btrfs_fs_info *fs_info)
1395 {
1396         struct btrfs_root *log_root;
1397
1398         log_root = alloc_log_tree(trans, fs_info);
1399         if (IS_ERR(log_root))
1400                 return PTR_ERR(log_root);
1401         WARN_ON(fs_info->log_root_tree);
1402         fs_info->log_root_tree = log_root;
1403         return 0;
1404 }
1405
1406 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1407                        struct btrfs_root *root)
1408 {
1409         struct btrfs_fs_info *fs_info = root->fs_info;
1410         struct btrfs_root *log_root;
1411         struct btrfs_inode_item *inode_item;
1412
1413         log_root = alloc_log_tree(trans, fs_info);
1414         if (IS_ERR(log_root))
1415                 return PTR_ERR(log_root);
1416
1417         log_root->last_trans = trans->transid;
1418         log_root->root_key.offset = root->root_key.objectid;
1419
1420         inode_item = &log_root->root_item.inode;
1421         btrfs_set_stack_inode_generation(inode_item, 1);
1422         btrfs_set_stack_inode_size(inode_item, 3);
1423         btrfs_set_stack_inode_nlink(inode_item, 1);
1424         btrfs_set_stack_inode_nbytes(inode_item,
1425                                      fs_info->nodesize);
1426         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1427
1428         btrfs_set_root_node(&log_root->root_item, log_root->node);
1429
1430         WARN_ON(root->log_root);
1431         root->log_root = log_root;
1432         root->log_transid = 0;
1433         root->log_transid_committed = -1;
1434         root->last_log_commit = 0;
1435         return 0;
1436 }
1437
1438 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1439                                                struct btrfs_key *key)
1440 {
1441         struct btrfs_root *root;
1442         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1443         struct btrfs_path *path;
1444         u64 generation;
1445         int ret;
1446         int level;
1447
1448         path = btrfs_alloc_path();
1449         if (!path)
1450                 return ERR_PTR(-ENOMEM);
1451
1452         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1453         if (!root) {
1454                 ret = -ENOMEM;
1455                 goto alloc_fail;
1456         }
1457
1458         __setup_root(root, fs_info, key->objectid);
1459
1460         ret = btrfs_find_root(tree_root, key, path,
1461                               &root->root_item, &root->root_key);
1462         if (ret) {
1463                 if (ret > 0)
1464                         ret = -ENOENT;
1465                 goto find_fail;
1466         }
1467
1468         generation = btrfs_root_generation(&root->root_item);
1469         level = btrfs_root_level(&root->root_item);
1470         root->node = read_tree_block(fs_info,
1471                                      btrfs_root_bytenr(&root->root_item),
1472                                      generation, level, NULL);
1473         if (IS_ERR(root->node)) {
1474                 ret = PTR_ERR(root->node);
1475                 goto find_fail;
1476         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477                 ret = -EIO;
1478                 free_extent_buffer(root->node);
1479                 goto find_fail;
1480         }
1481         root->commit_root = btrfs_root_node(root);
1482 out:
1483         btrfs_free_path(path);
1484         return root;
1485
1486 find_fail:
1487         kfree(root);
1488 alloc_fail:
1489         root = ERR_PTR(ret);
1490         goto out;
1491 }
1492
1493 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494                                       struct btrfs_key *location)
1495 {
1496         struct btrfs_root *root;
1497
1498         root = btrfs_read_tree_root(tree_root, location);
1499         if (IS_ERR(root))
1500                 return root;
1501
1502         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1503                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1504                 btrfs_check_and_init_root_item(&root->root_item);
1505         }
1506
1507         return root;
1508 }
1509
1510 int btrfs_init_fs_root(struct btrfs_root *root)
1511 {
1512         int ret;
1513         struct btrfs_subvolume_writers *writers;
1514
1515         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517                                         GFP_NOFS);
1518         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1519                 ret = -ENOMEM;
1520                 goto fail;
1521         }
1522
1523         writers = btrfs_alloc_subvolume_writers();
1524         if (IS_ERR(writers)) {
1525                 ret = PTR_ERR(writers);
1526                 goto fail;
1527         }
1528         root->subv_writers = writers;
1529
1530         btrfs_init_free_ino_ctl(root);
1531         spin_lock_init(&root->ino_cache_lock);
1532         init_waitqueue_head(&root->ino_cache_wait);
1533
1534         ret = get_anon_bdev(&root->anon_dev);
1535         if (ret)
1536                 goto fail;
1537
1538         mutex_lock(&root->objectid_mutex);
1539         ret = btrfs_find_highest_objectid(root,
1540                                         &root->highest_objectid);
1541         if (ret) {
1542                 mutex_unlock(&root->objectid_mutex);
1543                 goto fail;
1544         }
1545
1546         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1547
1548         mutex_unlock(&root->objectid_mutex);
1549
1550         return 0;
1551 fail:
1552         /* the caller is responsible to call free_fs_root */
1553         return ret;
1554 }
1555
1556 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1557                                         u64 root_id)
1558 {
1559         struct btrfs_root *root;
1560
1561         spin_lock(&fs_info->fs_roots_radix_lock);
1562         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1563                                  (unsigned long)root_id);
1564         spin_unlock(&fs_info->fs_roots_radix_lock);
1565         return root;
1566 }
1567
1568 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1569                          struct btrfs_root *root)
1570 {
1571         int ret;
1572
1573         ret = radix_tree_preload(GFP_NOFS);
1574         if (ret)
1575                 return ret;
1576
1577         spin_lock(&fs_info->fs_roots_radix_lock);
1578         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1579                                 (unsigned long)root->root_key.objectid,
1580                                 root);
1581         if (ret == 0)
1582                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1583         spin_unlock(&fs_info->fs_roots_radix_lock);
1584         radix_tree_preload_end();
1585
1586         return ret;
1587 }
1588
1589 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1590                                      struct btrfs_key *location,
1591                                      bool check_ref)
1592 {
1593         struct btrfs_root *root;
1594         struct btrfs_path *path;
1595         struct btrfs_key key;
1596         int ret;
1597
1598         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1599                 return fs_info->tree_root;
1600         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1601                 return fs_info->extent_root;
1602         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1603                 return fs_info->chunk_root;
1604         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1605                 return fs_info->dev_root;
1606         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1607                 return fs_info->csum_root;
1608         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1609                 return fs_info->quota_root ? fs_info->quota_root :
1610                                              ERR_PTR(-ENOENT);
1611         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1612                 return fs_info->uuid_root ? fs_info->uuid_root :
1613                                             ERR_PTR(-ENOENT);
1614         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1615                 return fs_info->free_space_root ? fs_info->free_space_root :
1616                                                   ERR_PTR(-ENOENT);
1617 again:
1618         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1619         if (root) {
1620                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1621                         return ERR_PTR(-ENOENT);
1622                 return root;
1623         }
1624
1625         root = btrfs_read_fs_root(fs_info->tree_root, location);
1626         if (IS_ERR(root))
1627                 return root;
1628
1629         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1630                 ret = -ENOENT;
1631                 goto fail;
1632         }
1633
1634         ret = btrfs_init_fs_root(root);
1635         if (ret)
1636                 goto fail;
1637
1638         path = btrfs_alloc_path();
1639         if (!path) {
1640                 ret = -ENOMEM;
1641                 goto fail;
1642         }
1643         key.objectid = BTRFS_ORPHAN_OBJECTID;
1644         key.type = BTRFS_ORPHAN_ITEM_KEY;
1645         key.offset = location->objectid;
1646
1647         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1648         btrfs_free_path(path);
1649         if (ret < 0)
1650                 goto fail;
1651         if (ret == 0)
1652                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1653
1654         ret = btrfs_insert_fs_root(fs_info, root);
1655         if (ret) {
1656                 if (ret == -EEXIST) {
1657                         free_fs_root(root);
1658                         goto again;
1659                 }
1660                 goto fail;
1661         }
1662         return root;
1663 fail:
1664         free_fs_root(root);
1665         return ERR_PTR(ret);
1666 }
1667
1668 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1669 {
1670         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1671         int ret = 0;
1672         struct btrfs_device *device;
1673         struct backing_dev_info *bdi;
1674
1675         rcu_read_lock();
1676         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1677                 if (!device->bdev)
1678                         continue;
1679                 bdi = device->bdev->bd_bdi;
1680                 if (bdi_congested(bdi, bdi_bits)) {
1681                         ret = 1;
1682                         break;
1683                 }
1684         }
1685         rcu_read_unlock();
1686         return ret;
1687 }
1688
1689 /*
1690  * called by the kthread helper functions to finally call the bio end_io
1691  * functions.  This is where read checksum verification actually happens
1692  */
1693 static void end_workqueue_fn(struct btrfs_work *work)
1694 {
1695         struct bio *bio;
1696         struct btrfs_end_io_wq *end_io_wq;
1697
1698         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1699         bio = end_io_wq->bio;
1700
1701         bio->bi_status = end_io_wq->status;
1702         bio->bi_private = end_io_wq->private;
1703         bio->bi_end_io = end_io_wq->end_io;
1704         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1705         bio_endio(bio);
1706 }
1707
1708 static int cleaner_kthread(void *arg)
1709 {
1710         struct btrfs_root *root = arg;
1711         struct btrfs_fs_info *fs_info = root->fs_info;
1712         int again;
1713         struct btrfs_trans_handle *trans;
1714
1715         do {
1716                 again = 0;
1717
1718                 /* Make the cleaner go to sleep early. */
1719                 if (btrfs_need_cleaner_sleep(fs_info))
1720                         goto sleep;
1721
1722                 /*
1723                  * Do not do anything if we might cause open_ctree() to block
1724                  * before we have finished mounting the filesystem.
1725                  */
1726                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1727                         goto sleep;
1728
1729                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1730                         goto sleep;
1731
1732                 /*
1733                  * Avoid the problem that we change the status of the fs
1734                  * during the above check and trylock.
1735                  */
1736                 if (btrfs_need_cleaner_sleep(fs_info)) {
1737                         mutex_unlock(&fs_info->cleaner_mutex);
1738                         goto sleep;
1739                 }
1740
1741                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1742                 btrfs_run_delayed_iputs(fs_info);
1743                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1744
1745                 again = btrfs_clean_one_deleted_snapshot(root);
1746                 mutex_unlock(&fs_info->cleaner_mutex);
1747
1748                 /*
1749                  * The defragger has dealt with the R/O remount and umount,
1750                  * needn't do anything special here.
1751                  */
1752                 btrfs_run_defrag_inodes(fs_info);
1753
1754                 /*
1755                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1756                  * with relocation (btrfs_relocate_chunk) and relocation
1757                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1758                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1759                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1760                  * unused block groups.
1761                  */
1762                 btrfs_delete_unused_bgs(fs_info);
1763 sleep:
1764                 if (!again) {
1765                         set_current_state(TASK_INTERRUPTIBLE);
1766                         if (!kthread_should_stop())
1767                                 schedule();
1768                         __set_current_state(TASK_RUNNING);
1769                 }
1770         } while (!kthread_should_stop());
1771
1772         /*
1773          * Transaction kthread is stopped before us and wakes us up.
1774          * However we might have started a new transaction and COWed some
1775          * tree blocks when deleting unused block groups for example. So
1776          * make sure we commit the transaction we started to have a clean
1777          * shutdown when evicting the btree inode - if it has dirty pages
1778          * when we do the final iput() on it, eviction will trigger a
1779          * writeback for it which will fail with null pointer dereferences
1780          * since work queues and other resources were already released and
1781          * destroyed by the time the iput/eviction/writeback is made.
1782          */
1783         trans = btrfs_attach_transaction(root);
1784         if (IS_ERR(trans)) {
1785                 if (PTR_ERR(trans) != -ENOENT)
1786                         btrfs_err(fs_info,
1787                                   "cleaner transaction attach returned %ld",
1788                                   PTR_ERR(trans));
1789         } else {
1790                 int ret;
1791
1792                 ret = btrfs_commit_transaction(trans);
1793                 if (ret)
1794                         btrfs_err(fs_info,
1795                                   "cleaner open transaction commit returned %d",
1796                                   ret);
1797         }
1798
1799         return 0;
1800 }
1801
1802 static int transaction_kthread(void *arg)
1803 {
1804         struct btrfs_root *root = arg;
1805         struct btrfs_fs_info *fs_info = root->fs_info;
1806         struct btrfs_trans_handle *trans;
1807         struct btrfs_transaction *cur;
1808         u64 transid;
1809         unsigned long now;
1810         unsigned long delay;
1811         bool cannot_commit;
1812
1813         do {
1814                 cannot_commit = false;
1815                 delay = HZ * fs_info->commit_interval;
1816                 mutex_lock(&fs_info->transaction_kthread_mutex);
1817
1818                 spin_lock(&fs_info->trans_lock);
1819                 cur = fs_info->running_transaction;
1820                 if (!cur) {
1821                         spin_unlock(&fs_info->trans_lock);
1822                         goto sleep;
1823                 }
1824
1825                 now = get_seconds();
1826                 if (cur->state < TRANS_STATE_BLOCKED &&
1827                     !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
1828                     (now < cur->start_time ||
1829                      now - cur->start_time < fs_info->commit_interval)) {
1830                         spin_unlock(&fs_info->trans_lock);
1831                         delay = HZ * 5;
1832                         goto sleep;
1833                 }
1834                 transid = cur->transid;
1835                 spin_unlock(&fs_info->trans_lock);
1836
1837                 /* If the file system is aborted, this will always fail. */
1838                 trans = btrfs_attach_transaction(root);
1839                 if (IS_ERR(trans)) {
1840                         if (PTR_ERR(trans) != -ENOENT)
1841                                 cannot_commit = true;
1842                         goto sleep;
1843                 }
1844                 if (transid == trans->transid) {
1845                         btrfs_commit_transaction(trans);
1846                 } else {
1847                         btrfs_end_transaction(trans);
1848                 }
1849 sleep:
1850                 wake_up_process(fs_info->cleaner_kthread);
1851                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1852
1853                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1854                                       &fs_info->fs_state)))
1855                         btrfs_cleanup_transaction(fs_info);
1856                 if (!kthread_should_stop() &&
1857                                 (!btrfs_transaction_blocked(fs_info) ||
1858                                  cannot_commit))
1859                         schedule_timeout_interruptible(delay);
1860         } while (!kthread_should_stop());
1861         return 0;
1862 }
1863
1864 /*
1865  * this will find the highest generation in the array of
1866  * root backups.  The index of the highest array is returned,
1867  * or -1 if we can't find anything.
1868  *
1869  * We check to make sure the array is valid by comparing the
1870  * generation of the latest  root in the array with the generation
1871  * in the super block.  If they don't match we pitch it.
1872  */
1873 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1874 {
1875         u64 cur;
1876         int newest_index = -1;
1877         struct btrfs_root_backup *root_backup;
1878         int i;
1879
1880         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1881                 root_backup = info->super_copy->super_roots + i;
1882                 cur = btrfs_backup_tree_root_gen(root_backup);
1883                 if (cur == newest_gen)
1884                         newest_index = i;
1885         }
1886
1887         /* check to see if we actually wrapped around */
1888         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1889                 root_backup = info->super_copy->super_roots;
1890                 cur = btrfs_backup_tree_root_gen(root_backup);
1891                 if (cur == newest_gen)
1892                         newest_index = 0;
1893         }
1894         return newest_index;
1895 }
1896
1897
1898 /*
1899  * find the oldest backup so we know where to store new entries
1900  * in the backup array.  This will set the backup_root_index
1901  * field in the fs_info struct
1902  */
1903 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1904                                      u64 newest_gen)
1905 {
1906         int newest_index = -1;
1907
1908         newest_index = find_newest_super_backup(info, newest_gen);
1909         /* if there was garbage in there, just move along */
1910         if (newest_index == -1) {
1911                 info->backup_root_index = 0;
1912         } else {
1913                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1914         }
1915 }
1916
1917 /*
1918  * copy all the root pointers into the super backup array.
1919  * this will bump the backup pointer by one when it is
1920  * done
1921  */
1922 static void backup_super_roots(struct btrfs_fs_info *info)
1923 {
1924         int next_backup;
1925         struct btrfs_root_backup *root_backup;
1926         int last_backup;
1927
1928         next_backup = info->backup_root_index;
1929         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1930                 BTRFS_NUM_BACKUP_ROOTS;
1931
1932         /*
1933          * just overwrite the last backup if we're at the same generation
1934          * this happens only at umount
1935          */
1936         root_backup = info->super_for_commit->super_roots + last_backup;
1937         if (btrfs_backup_tree_root_gen(root_backup) ==
1938             btrfs_header_generation(info->tree_root->node))
1939                 next_backup = last_backup;
1940
1941         root_backup = info->super_for_commit->super_roots + next_backup;
1942
1943         /*
1944          * make sure all of our padding and empty slots get zero filled
1945          * regardless of which ones we use today
1946          */
1947         memset(root_backup, 0, sizeof(*root_backup));
1948
1949         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1950
1951         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1952         btrfs_set_backup_tree_root_gen(root_backup,
1953                                btrfs_header_generation(info->tree_root->node));
1954
1955         btrfs_set_backup_tree_root_level(root_backup,
1956                                btrfs_header_level(info->tree_root->node));
1957
1958         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1959         btrfs_set_backup_chunk_root_gen(root_backup,
1960                                btrfs_header_generation(info->chunk_root->node));
1961         btrfs_set_backup_chunk_root_level(root_backup,
1962                                btrfs_header_level(info->chunk_root->node));
1963
1964         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1965         btrfs_set_backup_extent_root_gen(root_backup,
1966                                btrfs_header_generation(info->extent_root->node));
1967         btrfs_set_backup_extent_root_level(root_backup,
1968                                btrfs_header_level(info->extent_root->node));
1969
1970         /*
1971          * we might commit during log recovery, which happens before we set
1972          * the fs_root.  Make sure it is valid before we fill it in.
1973          */
1974         if (info->fs_root && info->fs_root->node) {
1975                 btrfs_set_backup_fs_root(root_backup,
1976                                          info->fs_root->node->start);
1977                 btrfs_set_backup_fs_root_gen(root_backup,
1978                                btrfs_header_generation(info->fs_root->node));
1979                 btrfs_set_backup_fs_root_level(root_backup,
1980                                btrfs_header_level(info->fs_root->node));
1981         }
1982
1983         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1984         btrfs_set_backup_dev_root_gen(root_backup,
1985                                btrfs_header_generation(info->dev_root->node));
1986         btrfs_set_backup_dev_root_level(root_backup,
1987                                        btrfs_header_level(info->dev_root->node));
1988
1989         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1990         btrfs_set_backup_csum_root_gen(root_backup,
1991                                btrfs_header_generation(info->csum_root->node));
1992         btrfs_set_backup_csum_root_level(root_backup,
1993                                btrfs_header_level(info->csum_root->node));
1994
1995         btrfs_set_backup_total_bytes(root_backup,
1996                              btrfs_super_total_bytes(info->super_copy));
1997         btrfs_set_backup_bytes_used(root_backup,
1998                              btrfs_super_bytes_used(info->super_copy));
1999         btrfs_set_backup_num_devices(root_backup,
2000                              btrfs_super_num_devices(info->super_copy));
2001
2002         /*
2003          * if we don't copy this out to the super_copy, it won't get remembered
2004          * for the next commit
2005          */
2006         memcpy(&info->super_copy->super_roots,
2007                &info->super_for_commit->super_roots,
2008                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2009 }
2010
2011 /*
2012  * this copies info out of the root backup array and back into
2013  * the in-memory super block.  It is meant to help iterate through
2014  * the array, so you send it the number of backups you've already
2015  * tried and the last backup index you used.
2016  *
2017  * this returns -1 when it has tried all the backups
2018  */
2019 static noinline int next_root_backup(struct btrfs_fs_info *info,
2020                                      struct btrfs_super_block *super,
2021                                      int *num_backups_tried, int *backup_index)
2022 {
2023         struct btrfs_root_backup *root_backup;
2024         int newest = *backup_index;
2025
2026         if (*num_backups_tried == 0) {
2027                 u64 gen = btrfs_super_generation(super);
2028
2029                 newest = find_newest_super_backup(info, gen);
2030                 if (newest == -1)
2031                         return -1;
2032
2033                 *backup_index = newest;
2034                 *num_backups_tried = 1;
2035         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2036                 /* we've tried all the backups, all done */
2037                 return -1;
2038         } else {
2039                 /* jump to the next oldest backup */
2040                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2041                         BTRFS_NUM_BACKUP_ROOTS;
2042                 *backup_index = newest;
2043                 *num_backups_tried += 1;
2044         }
2045         root_backup = super->super_roots + newest;
2046
2047         btrfs_set_super_generation(super,
2048                                    btrfs_backup_tree_root_gen(root_backup));
2049         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2050         btrfs_set_super_root_level(super,
2051                                    btrfs_backup_tree_root_level(root_backup));
2052         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2053
2054         /*
2055          * fixme: the total bytes and num_devices need to match or we should
2056          * need a fsck
2057          */
2058         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2059         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2060         return 0;
2061 }
2062
2063 /* helper to cleanup workers */
2064 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2065 {
2066         btrfs_destroy_workqueue(fs_info->fixup_workers);
2067         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2068         btrfs_destroy_workqueue(fs_info->workers);
2069         btrfs_destroy_workqueue(fs_info->endio_workers);
2070         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2071         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2072         btrfs_destroy_workqueue(fs_info->rmw_workers);
2073         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2074         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2075         btrfs_destroy_workqueue(fs_info->submit_workers);
2076         btrfs_destroy_workqueue(fs_info->delayed_workers);
2077         btrfs_destroy_workqueue(fs_info->caching_workers);
2078         btrfs_destroy_workqueue(fs_info->readahead_workers);
2079         btrfs_destroy_workqueue(fs_info->flush_workers);
2080         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2081         btrfs_destroy_workqueue(fs_info->extent_workers);
2082         /*
2083          * Now that all other work queues are destroyed, we can safely destroy
2084          * the queues used for metadata I/O, since tasks from those other work
2085          * queues can do metadata I/O operations.
2086          */
2087         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2088         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2089 }
2090
2091 static void free_root_extent_buffers(struct btrfs_root *root)
2092 {
2093         if (root) {
2094                 free_extent_buffer(root->node);
2095                 free_extent_buffer(root->commit_root);
2096                 root->node = NULL;
2097                 root->commit_root = NULL;
2098         }
2099 }
2100
2101 /* helper to cleanup tree roots */
2102 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2103 {
2104         free_root_extent_buffers(info->tree_root);
2105
2106         free_root_extent_buffers(info->dev_root);
2107         free_root_extent_buffers(info->extent_root);
2108         free_root_extent_buffers(info->csum_root);
2109         free_root_extent_buffers(info->quota_root);
2110         free_root_extent_buffers(info->uuid_root);
2111         if (chunk_root)
2112                 free_root_extent_buffers(info->chunk_root);
2113         free_root_extent_buffers(info->free_space_root);
2114 }
2115
2116 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2117 {
2118         int ret;
2119         struct btrfs_root *gang[8];
2120         int i;
2121
2122         while (!list_empty(&fs_info->dead_roots)) {
2123                 gang[0] = list_entry(fs_info->dead_roots.next,
2124                                      struct btrfs_root, root_list);
2125                 list_del(&gang[0]->root_list);
2126
2127                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2128                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2129                 } else {
2130                         free_extent_buffer(gang[0]->node);
2131                         free_extent_buffer(gang[0]->commit_root);
2132                         btrfs_put_fs_root(gang[0]);
2133                 }
2134         }
2135
2136         while (1) {
2137                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2138                                              (void **)gang, 0,
2139                                              ARRAY_SIZE(gang));
2140                 if (!ret)
2141                         break;
2142                 for (i = 0; i < ret; i++)
2143                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2144         }
2145
2146         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2147                 btrfs_free_log_root_tree(NULL, fs_info);
2148                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2149         }
2150 }
2151
2152 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2153 {
2154         mutex_init(&fs_info->scrub_lock);
2155         atomic_set(&fs_info->scrubs_running, 0);
2156         atomic_set(&fs_info->scrub_pause_req, 0);
2157         atomic_set(&fs_info->scrubs_paused, 0);
2158         atomic_set(&fs_info->scrub_cancel_req, 0);
2159         init_waitqueue_head(&fs_info->scrub_pause_wait);
2160         fs_info->scrub_workers_refcnt = 0;
2161 }
2162
2163 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2164 {
2165         spin_lock_init(&fs_info->balance_lock);
2166         mutex_init(&fs_info->balance_mutex);
2167         atomic_set(&fs_info->balance_running, 0);
2168         atomic_set(&fs_info->balance_pause_req, 0);
2169         atomic_set(&fs_info->balance_cancel_req, 0);
2170         fs_info->balance_ctl = NULL;
2171         init_waitqueue_head(&fs_info->balance_wait_q);
2172 }
2173
2174 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2175 {
2176         struct inode *inode = fs_info->btree_inode;
2177
2178         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2179         set_nlink(inode, 1);
2180         /*
2181          * we set the i_size on the btree inode to the max possible int.
2182          * the real end of the address space is determined by all of
2183          * the devices in the system
2184          */
2185         inode->i_size = OFFSET_MAX;
2186         inode->i_mapping->a_ops = &btree_aops;
2187
2188         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2189         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2190         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2191         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2192
2193         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2194
2195         BTRFS_I(inode)->root = fs_info->tree_root;
2196         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2197         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2198         btrfs_insert_inode_hash(inode);
2199 }
2200
2201 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2202 {
2203         fs_info->dev_replace.lock_owner = 0;
2204         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2205         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2206         rwlock_init(&fs_info->dev_replace.lock);
2207         atomic_set(&fs_info->dev_replace.read_locks, 0);
2208         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2209         init_waitqueue_head(&fs_info->replace_wait);
2210         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2211 }
2212
2213 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2214 {
2215         spin_lock_init(&fs_info->qgroup_lock);
2216         mutex_init(&fs_info->qgroup_ioctl_lock);
2217         fs_info->qgroup_tree = RB_ROOT;
2218         fs_info->qgroup_op_tree = RB_ROOT;
2219         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2220         fs_info->qgroup_seq = 1;
2221         fs_info->qgroup_ulist = NULL;
2222         fs_info->qgroup_rescan_running = false;
2223         mutex_init(&fs_info->qgroup_rescan_lock);
2224 }
2225
2226 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2227                 struct btrfs_fs_devices *fs_devices)
2228 {
2229         u32 max_active = fs_info->thread_pool_size;
2230         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2231
2232         fs_info->workers =
2233                 btrfs_alloc_workqueue(fs_info, "worker",
2234                                       flags | WQ_HIGHPRI, max_active, 16);
2235
2236         fs_info->delalloc_workers =
2237                 btrfs_alloc_workqueue(fs_info, "delalloc",
2238                                       flags, max_active, 2);
2239
2240         fs_info->flush_workers =
2241                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2242                                       flags, max_active, 0);
2243
2244         fs_info->caching_workers =
2245                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2246
2247         /*
2248          * a higher idle thresh on the submit workers makes it much more
2249          * likely that bios will be send down in a sane order to the
2250          * devices
2251          */
2252         fs_info->submit_workers =
2253                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2254                                       min_t(u64, fs_devices->num_devices,
2255                                             max_active), 64);
2256
2257         fs_info->fixup_workers =
2258                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2259
2260         /*
2261          * endios are largely parallel and should have a very
2262          * low idle thresh
2263          */
2264         fs_info->endio_workers =
2265                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2266         fs_info->endio_meta_workers =
2267                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2268                                       max_active, 4);
2269         fs_info->endio_meta_write_workers =
2270                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2271                                       max_active, 2);
2272         fs_info->endio_raid56_workers =
2273                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2274                                       max_active, 4);
2275         fs_info->endio_repair_workers =
2276                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2277         fs_info->rmw_workers =
2278                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2279         fs_info->endio_write_workers =
2280                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2281                                       max_active, 2);
2282         fs_info->endio_freespace_worker =
2283                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2284                                       max_active, 0);
2285         fs_info->delayed_workers =
2286                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2287                                       max_active, 0);
2288         fs_info->readahead_workers =
2289                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2290                                       max_active, 2);
2291         fs_info->qgroup_rescan_workers =
2292                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2293         fs_info->extent_workers =
2294                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2295                                       min_t(u64, fs_devices->num_devices,
2296                                             max_active), 8);
2297
2298         if (!(fs_info->workers && fs_info->delalloc_workers &&
2299               fs_info->submit_workers && fs_info->flush_workers &&
2300               fs_info->endio_workers && fs_info->endio_meta_workers &&
2301               fs_info->endio_meta_write_workers &&
2302               fs_info->endio_repair_workers &&
2303               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2304               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2305               fs_info->caching_workers && fs_info->readahead_workers &&
2306               fs_info->fixup_workers && fs_info->delayed_workers &&
2307               fs_info->extent_workers &&
2308               fs_info->qgroup_rescan_workers)) {
2309                 return -ENOMEM;
2310         }
2311
2312         return 0;
2313 }
2314
2315 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2316                             struct btrfs_fs_devices *fs_devices)
2317 {
2318         int ret;
2319         struct btrfs_root *log_tree_root;
2320         struct btrfs_super_block *disk_super = fs_info->super_copy;
2321         u64 bytenr = btrfs_super_log_root(disk_super);
2322         int level = btrfs_super_log_root_level(disk_super);
2323
2324         if (fs_devices->rw_devices == 0) {
2325                 btrfs_warn(fs_info, "log replay required on RO media");
2326                 return -EIO;
2327         }
2328
2329         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2330         if (!log_tree_root)
2331                 return -ENOMEM;
2332
2333         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2334
2335         log_tree_root->node = read_tree_block(fs_info, bytenr,
2336                                               fs_info->generation + 1,
2337                                               level, NULL);
2338         if (IS_ERR(log_tree_root->node)) {
2339                 btrfs_warn(fs_info, "failed to read log tree");
2340                 ret = PTR_ERR(log_tree_root->node);
2341                 kfree(log_tree_root);
2342                 return ret;
2343         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2344                 btrfs_err(fs_info, "failed to read log tree");
2345                 free_extent_buffer(log_tree_root->node);
2346                 kfree(log_tree_root);
2347                 return -EIO;
2348         }
2349         /* returns with log_tree_root freed on success */
2350         ret = btrfs_recover_log_trees(log_tree_root);
2351         if (ret) {
2352                 btrfs_handle_fs_error(fs_info, ret,
2353                                       "Failed to recover log tree");
2354                 free_extent_buffer(log_tree_root->node);
2355                 kfree(log_tree_root);
2356                 return ret;
2357         }
2358
2359         if (sb_rdonly(fs_info->sb)) {
2360                 ret = btrfs_commit_super(fs_info);
2361                 if (ret)
2362                         return ret;
2363         }
2364
2365         return 0;
2366 }
2367
2368 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2369 {
2370         struct btrfs_root *tree_root = fs_info->tree_root;
2371         struct btrfs_root *root;
2372         struct btrfs_key location;
2373         int ret;
2374
2375         BUG_ON(!fs_info->tree_root);
2376
2377         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2378         location.type = BTRFS_ROOT_ITEM_KEY;
2379         location.offset = 0;
2380
2381         root = btrfs_read_tree_root(tree_root, &location);
2382         if (IS_ERR(root)) {
2383                 ret = PTR_ERR(root);
2384                 goto out;
2385         }
2386         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2387         fs_info->extent_root = root;
2388
2389         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2390         root = btrfs_read_tree_root(tree_root, &location);
2391         if (IS_ERR(root)) {
2392                 ret = PTR_ERR(root);
2393                 goto out;
2394         }
2395         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2396         fs_info->dev_root = root;
2397         btrfs_init_devices_late(fs_info);
2398
2399         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2400         root = btrfs_read_tree_root(tree_root, &location);
2401         if (IS_ERR(root)) {
2402                 ret = PTR_ERR(root);
2403                 goto out;
2404         }
2405         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2406         fs_info->csum_root = root;
2407
2408         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2409         root = btrfs_read_tree_root(tree_root, &location);
2410         if (!IS_ERR(root)) {
2411                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2412                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2413                 fs_info->quota_root = root;
2414         }
2415
2416         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2417         root = btrfs_read_tree_root(tree_root, &location);
2418         if (IS_ERR(root)) {
2419                 ret = PTR_ERR(root);
2420                 if (ret != -ENOENT)
2421                         goto out;
2422         } else {
2423                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2424                 fs_info->uuid_root = root;
2425         }
2426
2427         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2428                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2429                 root = btrfs_read_tree_root(tree_root, &location);
2430                 if (IS_ERR(root)) {
2431                         ret = PTR_ERR(root);
2432                         goto out;
2433                 }
2434                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2435                 fs_info->free_space_root = root;
2436         }
2437
2438         return 0;
2439 out:
2440         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2441                    location.objectid, ret);
2442         return ret;
2443 }
2444
2445 int open_ctree(struct super_block *sb,
2446                struct btrfs_fs_devices *fs_devices,
2447                char *options)
2448 {
2449         u32 sectorsize;
2450         u32 nodesize;
2451         u32 stripesize;
2452         u64 generation;
2453         u64 features;
2454         struct btrfs_key location;
2455         struct buffer_head *bh;
2456         struct btrfs_super_block *disk_super;
2457         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2458         struct btrfs_root *tree_root;
2459         struct btrfs_root *chunk_root;
2460         int ret;
2461         int err = -EINVAL;
2462         int num_backups_tried = 0;
2463         int backup_index = 0;
2464         int clear_free_space_tree = 0;
2465         int level;
2466
2467         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2468         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2469         if (!tree_root || !chunk_root) {
2470                 err = -ENOMEM;
2471                 goto fail;
2472         }
2473
2474         ret = init_srcu_struct(&fs_info->subvol_srcu);
2475         if (ret) {
2476                 err = ret;
2477                 goto fail;
2478         }
2479
2480         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2481         if (ret) {
2482                 err = ret;
2483                 goto fail_srcu;
2484         }
2485         fs_info->dirty_metadata_batch = PAGE_SIZE *
2486                                         (1 + ilog2(nr_cpu_ids));
2487
2488         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2489         if (ret) {
2490                 err = ret;
2491                 goto fail_dirty_metadata_bytes;
2492         }
2493
2494         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2495         if (ret) {
2496                 err = ret;
2497                 goto fail_delalloc_bytes;
2498         }
2499
2500         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2501         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2502         INIT_LIST_HEAD(&fs_info->trans_list);
2503         INIT_LIST_HEAD(&fs_info->dead_roots);
2504         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2505         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2506         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2507         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2508         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2509         spin_lock_init(&fs_info->delalloc_root_lock);
2510         spin_lock_init(&fs_info->trans_lock);
2511         spin_lock_init(&fs_info->fs_roots_radix_lock);
2512         spin_lock_init(&fs_info->delayed_iput_lock);
2513         spin_lock_init(&fs_info->defrag_inodes_lock);
2514         spin_lock_init(&fs_info->tree_mod_seq_lock);
2515         spin_lock_init(&fs_info->super_lock);
2516         spin_lock_init(&fs_info->qgroup_op_lock);
2517         spin_lock_init(&fs_info->buffer_lock);
2518         spin_lock_init(&fs_info->unused_bgs_lock);
2519         rwlock_init(&fs_info->tree_mod_log_lock);
2520         mutex_init(&fs_info->unused_bg_unpin_mutex);
2521         mutex_init(&fs_info->delete_unused_bgs_mutex);
2522         mutex_init(&fs_info->reloc_mutex);
2523         mutex_init(&fs_info->delalloc_root_mutex);
2524         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2525         seqlock_init(&fs_info->profiles_lock);
2526
2527         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2528         INIT_LIST_HEAD(&fs_info->space_info);
2529         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2530         INIT_LIST_HEAD(&fs_info->unused_bgs);
2531         btrfs_mapping_init(&fs_info->mapping_tree);
2532         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2533                              BTRFS_BLOCK_RSV_GLOBAL);
2534         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2535         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2536         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2537         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2538                              BTRFS_BLOCK_RSV_DELOPS);
2539         atomic_set(&fs_info->async_delalloc_pages, 0);
2540         atomic_set(&fs_info->defrag_running, 0);
2541         atomic_set(&fs_info->qgroup_op_seq, 0);
2542         atomic_set(&fs_info->reada_works_cnt, 0);
2543         atomic64_set(&fs_info->tree_mod_seq, 0);
2544         fs_info->sb = sb;
2545         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2546         fs_info->metadata_ratio = 0;
2547         fs_info->defrag_inodes = RB_ROOT;
2548         atomic64_set(&fs_info->free_chunk_space, 0);
2549         fs_info->tree_mod_log = RB_ROOT;
2550         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2551         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2552         /* readahead state */
2553         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2554         spin_lock_init(&fs_info->reada_lock);
2555         btrfs_init_ref_verify(fs_info);
2556
2557         fs_info->thread_pool_size = min_t(unsigned long,
2558                                           num_online_cpus() + 2, 8);
2559
2560         INIT_LIST_HEAD(&fs_info->ordered_roots);
2561         spin_lock_init(&fs_info->ordered_root_lock);
2562
2563         fs_info->btree_inode = new_inode(sb);
2564         if (!fs_info->btree_inode) {
2565                 err = -ENOMEM;
2566                 goto fail_bio_counter;
2567         }
2568         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2569
2570         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2571                                         GFP_KERNEL);
2572         if (!fs_info->delayed_root) {
2573                 err = -ENOMEM;
2574                 goto fail_iput;
2575         }
2576         btrfs_init_delayed_root(fs_info->delayed_root);
2577
2578         btrfs_init_scrub(fs_info);
2579 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2580         fs_info->check_integrity_print_mask = 0;
2581 #endif
2582         btrfs_init_balance(fs_info);
2583         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2584
2585         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2586         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2587
2588         btrfs_init_btree_inode(fs_info);
2589
2590         spin_lock_init(&fs_info->block_group_cache_lock);
2591         fs_info->block_group_cache_tree = RB_ROOT;
2592         fs_info->first_logical_byte = (u64)-1;
2593
2594         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2595         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2596         fs_info->pinned_extents = &fs_info->freed_extents[0];
2597         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2598
2599         mutex_init(&fs_info->ordered_operations_mutex);
2600         mutex_init(&fs_info->tree_log_mutex);
2601         mutex_init(&fs_info->chunk_mutex);
2602         mutex_init(&fs_info->transaction_kthread_mutex);
2603         mutex_init(&fs_info->cleaner_mutex);
2604         mutex_init(&fs_info->volume_mutex);
2605         mutex_init(&fs_info->ro_block_group_mutex);
2606         init_rwsem(&fs_info->commit_root_sem);
2607         init_rwsem(&fs_info->cleanup_work_sem);
2608         init_rwsem(&fs_info->subvol_sem);
2609         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2610
2611         btrfs_init_dev_replace_locks(fs_info);
2612         btrfs_init_qgroup(fs_info);
2613
2614         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2615         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2616
2617         init_waitqueue_head(&fs_info->transaction_throttle);
2618         init_waitqueue_head(&fs_info->transaction_wait);
2619         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2620         init_waitqueue_head(&fs_info->async_submit_wait);
2621
2622         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2623
2624         /* Usable values until the real ones are cached from the superblock */
2625         fs_info->nodesize = 4096;
2626         fs_info->sectorsize = 4096;
2627         fs_info->stripesize = 4096;
2628
2629         ret = btrfs_alloc_stripe_hash_table(fs_info);
2630         if (ret) {
2631                 err = ret;
2632                 goto fail_alloc;
2633         }
2634
2635         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2636
2637         invalidate_bdev(fs_devices->latest_bdev);
2638
2639         /*
2640          * Read super block and check the signature bytes only
2641          */
2642         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2643         if (IS_ERR(bh)) {
2644                 err = PTR_ERR(bh);
2645                 goto fail_alloc;
2646         }
2647
2648         /*
2649          * We want to check superblock checksum, the type is stored inside.
2650          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2651          */
2652         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2653                 btrfs_err(fs_info, "superblock checksum mismatch");
2654                 err = -EINVAL;
2655                 brelse(bh);
2656                 goto fail_alloc;
2657         }
2658
2659         /*
2660          * super_copy is zeroed at allocation time and we never touch the
2661          * following bytes up to INFO_SIZE, the checksum is calculated from
2662          * the whole block of INFO_SIZE
2663          */
2664         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2665         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2666                sizeof(*fs_info->super_for_commit));
2667         brelse(bh);
2668
2669         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2670
2671         ret = btrfs_check_super_valid(fs_info);
2672         if (ret) {
2673                 btrfs_err(fs_info, "superblock contains fatal errors");
2674                 err = -EINVAL;
2675                 goto fail_alloc;
2676         }
2677
2678         disk_super = fs_info->super_copy;
2679         if (!btrfs_super_root(disk_super))
2680                 goto fail_alloc;
2681
2682         /* check FS state, whether FS is broken. */
2683         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2684                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2685
2686         /*
2687          * run through our array of backup supers and setup
2688          * our ring pointer to the oldest one
2689          */
2690         generation = btrfs_super_generation(disk_super);
2691         find_oldest_super_backup(fs_info, generation);
2692
2693         /*
2694          * In the long term, we'll store the compression type in the super
2695          * block, and it'll be used for per file compression control.
2696          */
2697         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2698
2699         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2700         if (ret) {
2701                 err = ret;
2702                 goto fail_alloc;
2703         }
2704
2705         features = btrfs_super_incompat_flags(disk_super) &
2706                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2707         if (features) {
2708                 btrfs_err(fs_info,
2709                     "cannot mount because of unsupported optional features (%llx)",
2710                     features);
2711                 err = -EINVAL;
2712                 goto fail_alloc;
2713         }
2714
2715         features = btrfs_super_incompat_flags(disk_super);
2716         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2717         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2718                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2719         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2720                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2721
2722         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2723                 btrfs_info(fs_info, "has skinny extents");
2724
2725         /*
2726          * flag our filesystem as having big metadata blocks if
2727          * they are bigger than the page size
2728          */
2729         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2730                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2731                         btrfs_info(fs_info,
2732                                 "flagging fs with big metadata feature");
2733                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2734         }
2735
2736         nodesize = btrfs_super_nodesize(disk_super);
2737         sectorsize = btrfs_super_sectorsize(disk_super);
2738         stripesize = sectorsize;
2739         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2740         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2741
2742         /* Cache block sizes */
2743         fs_info->nodesize = nodesize;
2744         fs_info->sectorsize = sectorsize;
2745         fs_info->stripesize = stripesize;
2746
2747         /*
2748          * mixed block groups end up with duplicate but slightly offset
2749          * extent buffers for the same range.  It leads to corruptions
2750          */
2751         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2752             (sectorsize != nodesize)) {
2753                 btrfs_err(fs_info,
2754 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2755                         nodesize, sectorsize);
2756                 goto fail_alloc;
2757         }
2758
2759         /*
2760          * Needn't use the lock because there is no other task which will
2761          * update the flag.
2762          */
2763         btrfs_set_super_incompat_flags(disk_super, features);
2764
2765         features = btrfs_super_compat_ro_flags(disk_super) &
2766                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2767         if (!sb_rdonly(sb) && features) {
2768                 btrfs_err(fs_info,
2769         "cannot mount read-write because of unsupported optional features (%llx)",
2770                        features);
2771                 err = -EINVAL;
2772                 goto fail_alloc;
2773         }
2774
2775         ret = btrfs_init_workqueues(fs_info, fs_devices);
2776         if (ret) {
2777                 err = ret;
2778                 goto fail_sb_buffer;
2779         }
2780
2781         sb->s_bdi->congested_fn = btrfs_congested_fn;
2782         sb->s_bdi->congested_data = fs_info;
2783         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2784         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2785         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2786         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2787
2788         sb->s_blocksize = sectorsize;
2789         sb->s_blocksize_bits = blksize_bits(sectorsize);
2790         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2791
2792         mutex_lock(&fs_info->chunk_mutex);
2793         ret = btrfs_read_sys_array(fs_info);
2794         mutex_unlock(&fs_info->chunk_mutex);
2795         if (ret) {
2796                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2797                 goto fail_sb_buffer;
2798         }
2799
2800         generation = btrfs_super_chunk_root_generation(disk_super);
2801         level = btrfs_super_chunk_root_level(disk_super);
2802
2803         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2804
2805         chunk_root->node = read_tree_block(fs_info,
2806                                            btrfs_super_chunk_root(disk_super),
2807                                            generation, level, NULL);
2808         if (IS_ERR(chunk_root->node) ||
2809             !extent_buffer_uptodate(chunk_root->node)) {
2810                 btrfs_err(fs_info, "failed to read chunk root");
2811                 if (!IS_ERR(chunk_root->node))
2812                         free_extent_buffer(chunk_root->node);
2813                 chunk_root->node = NULL;
2814                 goto fail_tree_roots;
2815         }
2816         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2817         chunk_root->commit_root = btrfs_root_node(chunk_root);
2818
2819         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2820            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2821
2822         ret = btrfs_read_chunk_tree(fs_info);
2823         if (ret) {
2824                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2825                 goto fail_tree_roots;
2826         }
2827
2828         /*
2829          * Keep the devid that is marked to be the target device for the
2830          * device replace procedure
2831          */
2832         btrfs_free_extra_devids(fs_devices, 0);
2833
2834         if (!fs_devices->latest_bdev) {
2835                 btrfs_err(fs_info, "failed to read devices");
2836                 goto fail_tree_roots;
2837         }
2838
2839 retry_root_backup:
2840         generation = btrfs_super_generation(disk_super);
2841         level = btrfs_super_root_level(disk_super);
2842
2843         tree_root->node = read_tree_block(fs_info,
2844                                           btrfs_super_root(disk_super),
2845                                           generation, level, NULL);
2846         if (IS_ERR(tree_root->node) ||
2847             !extent_buffer_uptodate(tree_root->node)) {
2848                 btrfs_warn(fs_info, "failed to read tree root");
2849                 if (!IS_ERR(tree_root->node))
2850                         free_extent_buffer(tree_root->node);
2851                 tree_root->node = NULL;
2852                 goto recovery_tree_root;
2853         }
2854
2855         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2856         tree_root->commit_root = btrfs_root_node(tree_root);
2857         btrfs_set_root_refs(&tree_root->root_item, 1);
2858
2859         mutex_lock(&tree_root->objectid_mutex);
2860         ret = btrfs_find_highest_objectid(tree_root,
2861                                         &tree_root->highest_objectid);
2862         if (ret) {
2863                 mutex_unlock(&tree_root->objectid_mutex);
2864                 goto recovery_tree_root;
2865         }
2866
2867         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2868
2869         mutex_unlock(&tree_root->objectid_mutex);
2870
2871         ret = btrfs_read_roots(fs_info);
2872         if (ret)
2873                 goto recovery_tree_root;
2874
2875         fs_info->generation = generation;
2876         fs_info->last_trans_committed = generation;
2877
2878         ret = btrfs_recover_balance(fs_info);
2879         if (ret) {
2880                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2881                 goto fail_block_groups;
2882         }
2883
2884         ret = btrfs_init_dev_stats(fs_info);
2885         if (ret) {
2886                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2887                 goto fail_block_groups;
2888         }
2889
2890         ret = btrfs_init_dev_replace(fs_info);
2891         if (ret) {
2892                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2893                 goto fail_block_groups;
2894         }
2895
2896         btrfs_free_extra_devids(fs_devices, 1);
2897
2898         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2899         if (ret) {
2900                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2901                                 ret);
2902                 goto fail_block_groups;
2903         }
2904
2905         ret = btrfs_sysfs_add_device(fs_devices);
2906         if (ret) {
2907                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2908                                 ret);
2909                 goto fail_fsdev_sysfs;
2910         }
2911
2912         ret = btrfs_sysfs_add_mounted(fs_info);
2913         if (ret) {
2914                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2915                 goto fail_fsdev_sysfs;
2916         }
2917
2918         ret = btrfs_init_space_info(fs_info);
2919         if (ret) {
2920                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2921                 goto fail_sysfs;
2922         }
2923
2924         ret = btrfs_read_block_groups(fs_info);
2925         if (ret) {
2926                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2927                 goto fail_sysfs;
2928         }
2929
2930         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2931                 btrfs_warn(fs_info,
2932                 "writeable mount is not allowed due to too many missing devices");
2933                 goto fail_sysfs;
2934         }
2935
2936         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2937                                                "btrfs-cleaner");
2938         if (IS_ERR(fs_info->cleaner_kthread))
2939                 goto fail_sysfs;
2940
2941         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2942                                                    tree_root,
2943                                                    "btrfs-transaction");
2944         if (IS_ERR(fs_info->transaction_kthread))
2945                 goto fail_cleaner;
2946
2947         if (!btrfs_test_opt(fs_info, NOSSD) &&
2948             !fs_info->fs_devices->rotating) {
2949                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2950         }
2951
2952         /*
2953          * Mount does not set all options immediately, we can do it now and do
2954          * not have to wait for transaction commit
2955          */
2956         btrfs_apply_pending_changes(fs_info);
2957
2958 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2959         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2960                 ret = btrfsic_mount(fs_info, fs_devices,
2961                                     btrfs_test_opt(fs_info,
2962                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2963                                     1 : 0,
2964                                     fs_info->check_integrity_print_mask);
2965                 if (ret)
2966                         btrfs_warn(fs_info,
2967                                 "failed to initialize integrity check module: %d",
2968                                 ret);
2969         }
2970 #endif
2971         ret = btrfs_read_qgroup_config(fs_info);
2972         if (ret)
2973                 goto fail_trans_kthread;
2974
2975         if (btrfs_build_ref_tree(fs_info))
2976                 btrfs_err(fs_info, "couldn't build ref tree");
2977
2978         /* do not make disk changes in broken FS or nologreplay is given */
2979         if (btrfs_super_log_root(disk_super) != 0 &&
2980             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2981                 ret = btrfs_replay_log(fs_info, fs_devices);
2982                 if (ret) {
2983                         err = ret;
2984                         goto fail_qgroup;
2985                 }
2986         }
2987
2988         ret = btrfs_find_orphan_roots(fs_info);
2989         if (ret)
2990                 goto fail_qgroup;
2991
2992         if (!sb_rdonly(sb)) {
2993                 ret = btrfs_cleanup_fs_roots(fs_info);
2994                 if (ret)
2995                         goto fail_qgroup;
2996
2997                 mutex_lock(&fs_info->cleaner_mutex);
2998                 ret = btrfs_recover_relocation(tree_root);
2999                 mutex_unlock(&fs_info->cleaner_mutex);
3000                 if (ret < 0) {
3001                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3002                                         ret);
3003                         err = -EINVAL;
3004                         goto fail_qgroup;
3005                 }
3006         }
3007
3008         location.objectid = BTRFS_FS_TREE_OBJECTID;
3009         location.type = BTRFS_ROOT_ITEM_KEY;
3010         location.offset = 0;
3011
3012         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3013         if (IS_ERR(fs_info->fs_root)) {
3014                 err = PTR_ERR(fs_info->fs_root);
3015                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3016                 goto fail_qgroup;
3017         }
3018
3019         if (sb_rdonly(sb))
3020                 return 0;
3021
3022         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3023             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3024                 clear_free_space_tree = 1;
3025         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3026                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3027                 btrfs_warn(fs_info, "free space tree is invalid");
3028                 clear_free_space_tree = 1;
3029         }
3030
3031         if (clear_free_space_tree) {
3032                 btrfs_info(fs_info, "clearing free space tree");
3033                 ret = btrfs_clear_free_space_tree(fs_info);
3034                 if (ret) {
3035                         btrfs_warn(fs_info,
3036                                    "failed to clear free space tree: %d", ret);
3037                         close_ctree(fs_info);
3038                         return ret;
3039                 }
3040         }
3041
3042         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3043             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3044                 btrfs_info(fs_info, "creating free space tree");
3045                 ret = btrfs_create_free_space_tree(fs_info);
3046                 if (ret) {
3047                         btrfs_warn(fs_info,
3048                                 "failed to create free space tree: %d", ret);
3049                         close_ctree(fs_info);
3050                         return ret;
3051                 }
3052         }
3053
3054         down_read(&fs_info->cleanup_work_sem);
3055         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3056             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3057                 up_read(&fs_info->cleanup_work_sem);
3058                 close_ctree(fs_info);
3059                 return ret;
3060         }
3061         up_read(&fs_info->cleanup_work_sem);
3062
3063         ret = btrfs_resume_balance_async(fs_info);
3064         if (ret) {
3065                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3066                 close_ctree(fs_info);
3067                 return ret;
3068         }
3069
3070         ret = btrfs_resume_dev_replace_async(fs_info);
3071         if (ret) {
3072                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3073                 close_ctree(fs_info);
3074                 return ret;
3075         }
3076
3077         btrfs_qgroup_rescan_resume(fs_info);
3078
3079         if (!fs_info->uuid_root) {
3080                 btrfs_info(fs_info, "creating UUID tree");
3081                 ret = btrfs_create_uuid_tree(fs_info);
3082                 if (ret) {
3083                         btrfs_warn(fs_info,
3084                                 "failed to create the UUID tree: %d", ret);
3085                         close_ctree(fs_info);
3086                         return ret;
3087                 }
3088         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3089                    fs_info->generation !=
3090                                 btrfs_super_uuid_tree_generation(disk_super)) {
3091                 btrfs_info(fs_info, "checking UUID tree");
3092                 ret = btrfs_check_uuid_tree(fs_info);
3093                 if (ret) {
3094                         btrfs_warn(fs_info,
3095                                 "failed to check the UUID tree: %d", ret);
3096                         close_ctree(fs_info);
3097                         return ret;
3098                 }
3099         } else {
3100                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3101         }
3102         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3103
3104         /*
3105          * backuproot only affect mount behavior, and if open_ctree succeeded,
3106          * no need to keep the flag
3107          */
3108         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3109
3110         return 0;
3111
3112 fail_qgroup:
3113         btrfs_free_qgroup_config(fs_info);
3114 fail_trans_kthread:
3115         kthread_stop(fs_info->transaction_kthread);
3116         btrfs_cleanup_transaction(fs_info);
3117         btrfs_free_fs_roots(fs_info);
3118 fail_cleaner:
3119         kthread_stop(fs_info->cleaner_kthread);
3120
3121         /*
3122          * make sure we're done with the btree inode before we stop our
3123          * kthreads
3124          */
3125         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3126
3127 fail_sysfs:
3128         btrfs_sysfs_remove_mounted(fs_info);
3129
3130 fail_fsdev_sysfs:
3131         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3132
3133 fail_block_groups:
3134         btrfs_put_block_group_cache(fs_info);
3135
3136 fail_tree_roots:
3137         free_root_pointers(fs_info, 1);
3138         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3139
3140 fail_sb_buffer:
3141         btrfs_stop_all_workers(fs_info);
3142         btrfs_free_block_groups(fs_info);
3143 fail_alloc:
3144 fail_iput:
3145         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3146
3147         iput(fs_info->btree_inode);
3148 fail_bio_counter:
3149         percpu_counter_destroy(&fs_info->bio_counter);
3150 fail_delalloc_bytes:
3151         percpu_counter_destroy(&fs_info->delalloc_bytes);
3152 fail_dirty_metadata_bytes:
3153         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3154 fail_srcu:
3155         cleanup_srcu_struct(&fs_info->subvol_srcu);
3156 fail:
3157         btrfs_free_stripe_hash_table(fs_info);
3158         btrfs_close_devices(fs_info->fs_devices);
3159         return err;
3160
3161 recovery_tree_root:
3162         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3163                 goto fail_tree_roots;
3164
3165         free_root_pointers(fs_info, 0);
3166
3167         /* don't use the log in recovery mode, it won't be valid */
3168         btrfs_set_super_log_root(disk_super, 0);
3169
3170         /* we can't trust the free space cache either */
3171         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3172
3173         ret = next_root_backup(fs_info, fs_info->super_copy,
3174                                &num_backups_tried, &backup_index);
3175         if (ret == -1)
3176                 goto fail_block_groups;
3177         goto retry_root_backup;
3178 }
3179 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3180
3181 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3182 {
3183         if (uptodate) {
3184                 set_buffer_uptodate(bh);
3185         } else {
3186                 struct btrfs_device *device = (struct btrfs_device *)
3187                         bh->b_private;
3188
3189                 btrfs_warn_rl_in_rcu(device->fs_info,
3190                                 "lost page write due to IO error on %s",
3191                                           rcu_str_deref(device->name));
3192                 /* note, we don't set_buffer_write_io_error because we have
3193                  * our own ways of dealing with the IO errors
3194                  */
3195                 clear_buffer_uptodate(bh);
3196                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3197         }
3198         unlock_buffer(bh);
3199         put_bh(bh);
3200 }
3201
3202 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3203                         struct buffer_head **bh_ret)
3204 {
3205         struct buffer_head *bh;
3206         struct btrfs_super_block *super;
3207         u64 bytenr;
3208
3209         bytenr = btrfs_sb_offset(copy_num);
3210         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3211                 return -EINVAL;
3212
3213         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3214         /*
3215          * If we fail to read from the underlying devices, as of now
3216          * the best option we have is to mark it EIO.
3217          */
3218         if (!bh)
3219                 return -EIO;
3220
3221         super = (struct btrfs_super_block *)bh->b_data;
3222         if (btrfs_super_bytenr(super) != bytenr ||
3223                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3224                 brelse(bh);
3225                 return -EINVAL;
3226         }
3227
3228         *bh_ret = bh;
3229         return 0;
3230 }
3231
3232
3233 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3234 {
3235         struct buffer_head *bh;
3236         struct buffer_head *latest = NULL;
3237         struct btrfs_super_block *super;
3238         int i;
3239         u64 transid = 0;
3240         int ret = -EINVAL;
3241
3242         /* we would like to check all the supers, but that would make
3243          * a btrfs mount succeed after a mkfs from a different FS.
3244          * So, we need to add a special mount option to scan for
3245          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3246          */
3247         for (i = 0; i < 1; i++) {
3248                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3249                 if (ret)
3250                         continue;
3251
3252                 super = (struct btrfs_super_block *)bh->b_data;
3253
3254                 if (!latest || btrfs_super_generation(super) > transid) {
3255                         brelse(latest);
3256                         latest = bh;
3257                         transid = btrfs_super_generation(super);
3258                 } else {
3259                         brelse(bh);
3260                 }
3261         }
3262
3263         if (!latest)
3264                 return ERR_PTR(ret);
3265
3266         return latest;
3267 }
3268
3269 /*
3270  * Write superblock @sb to the @device. Do not wait for completion, all the
3271  * buffer heads we write are pinned.
3272  *
3273  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3274  * the expected device size at commit time. Note that max_mirrors must be
3275  * same for write and wait phases.
3276  *
3277  * Return number of errors when buffer head is not found or submission fails.
3278  */
3279 static int write_dev_supers(struct btrfs_device *device,
3280                             struct btrfs_super_block *sb, int max_mirrors)
3281 {
3282         struct buffer_head *bh;
3283         int i;
3284         int ret;
3285         int errors = 0;
3286         u32 crc;
3287         u64 bytenr;
3288         int op_flags;
3289
3290         if (max_mirrors == 0)
3291                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3292
3293         for (i = 0; i < max_mirrors; i++) {
3294                 bytenr = btrfs_sb_offset(i);
3295                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3296                     device->commit_total_bytes)
3297                         break;
3298
3299                 btrfs_set_super_bytenr(sb, bytenr);
3300
3301                 crc = ~(u32)0;
3302                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3303                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3304                 btrfs_csum_final(crc, sb->csum);
3305
3306                 /* One reference for us, and we leave it for the caller */
3307                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3308                               BTRFS_SUPER_INFO_SIZE);
3309                 if (!bh) {
3310                         btrfs_err(device->fs_info,
3311                             "couldn't get super buffer head for bytenr %llu",
3312                             bytenr);
3313                         errors++;
3314                         continue;
3315                 }
3316
3317                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3318
3319                 /* one reference for submit_bh */
3320                 get_bh(bh);
3321
3322                 set_buffer_uptodate(bh);
3323                 lock_buffer(bh);
3324                 bh->b_end_io = btrfs_end_buffer_write_sync;
3325                 bh->b_private = device;
3326
3327                 /*
3328                  * we fua the first super.  The others we allow
3329                  * to go down lazy.
3330                  */
3331                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3332                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3333                         op_flags |= REQ_FUA;
3334                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3335                 if (ret)
3336                         errors++;
3337         }
3338         return errors < i ? 0 : -1;
3339 }
3340
3341 /*
3342  * Wait for write completion of superblocks done by write_dev_supers,
3343  * @max_mirrors same for write and wait phases.
3344  *
3345  * Return number of errors when buffer head is not found or not marked up to
3346  * date.
3347  */
3348 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3349 {
3350         struct buffer_head *bh;
3351         int i;
3352         int errors = 0;
3353         bool primary_failed = false;
3354         u64 bytenr;
3355
3356         if (max_mirrors == 0)
3357                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3358
3359         for (i = 0; i < max_mirrors; i++) {
3360                 bytenr = btrfs_sb_offset(i);
3361                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3362                     device->commit_total_bytes)
3363                         break;
3364
3365                 bh = __find_get_block(device->bdev,
3366                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3367                                       BTRFS_SUPER_INFO_SIZE);
3368                 if (!bh) {
3369                         errors++;
3370                         if (i == 0)
3371                                 primary_failed = true;
3372                         continue;
3373                 }
3374                 wait_on_buffer(bh);
3375                 if (!buffer_uptodate(bh)) {
3376                         errors++;
3377                         if (i == 0)
3378                                 primary_failed = true;
3379                 }
3380
3381                 /* drop our reference */
3382                 brelse(bh);
3383
3384                 /* drop the reference from the writing run */
3385                 brelse(bh);
3386         }
3387
3388         /* log error, force error return */
3389         if (primary_failed) {
3390                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3391                           device->devid);
3392                 return -1;
3393         }
3394
3395         return errors < i ? 0 : -1;
3396 }
3397
3398 /*
3399  * endio for the write_dev_flush, this will wake anyone waiting
3400  * for the barrier when it is done
3401  */
3402 static void btrfs_end_empty_barrier(struct bio *bio)
3403 {
3404         complete(bio->bi_private);
3405 }
3406
3407 /*
3408  * Submit a flush request to the device if it supports it. Error handling is
3409  * done in the waiting counterpart.
3410  */
3411 static void write_dev_flush(struct btrfs_device *device)
3412 {
3413         struct request_queue *q = bdev_get_queue(device->bdev);
3414         struct bio *bio = device->flush_bio;
3415
3416         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3417                 return;
3418
3419         bio_reset(bio);
3420         bio->bi_end_io = btrfs_end_empty_barrier;
3421         bio_set_dev(bio, device->bdev);
3422         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3423         init_completion(&device->flush_wait);
3424         bio->bi_private = &device->flush_wait;
3425
3426         btrfsic_submit_bio(bio);
3427         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3428 }
3429
3430 /*
3431  * If the flush bio has been submitted by write_dev_flush, wait for it.
3432  */
3433 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3434 {
3435         struct bio *bio = device->flush_bio;
3436
3437         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3438                 return BLK_STS_OK;
3439
3440         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3441         wait_for_completion_io(&device->flush_wait);
3442
3443         return bio->bi_status;
3444 }
3445
3446 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3447 {
3448         if (!btrfs_check_rw_degradable(fs_info, NULL))
3449                 return -EIO;
3450         return 0;
3451 }
3452
3453 /*
3454  * send an empty flush down to each device in parallel,
3455  * then wait for them
3456  */
3457 static int barrier_all_devices(struct btrfs_fs_info *info)
3458 {
3459         struct list_head *head;
3460         struct btrfs_device *dev;
3461         int errors_wait = 0;
3462         blk_status_t ret;
3463
3464         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3465         /* send down all the barriers */
3466         head = &info->fs_devices->devices;
3467         list_for_each_entry(dev, head, dev_list) {
3468                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3469                         continue;
3470                 if (!dev->bdev)
3471                         continue;
3472                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3473                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3474                         continue;
3475
3476                 write_dev_flush(dev);
3477                 dev->last_flush_error = BLK_STS_OK;
3478         }
3479
3480         /* wait for all the barriers */
3481         list_for_each_entry(dev, head, dev_list) {
3482                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3483                         continue;
3484                 if (!dev->bdev) {
3485                         errors_wait++;
3486                         continue;
3487                 }
3488                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3489                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3490                         continue;
3491
3492                 ret = wait_dev_flush(dev);
3493                 if (ret) {
3494                         dev->last_flush_error = ret;
3495                         btrfs_dev_stat_inc_and_print(dev,
3496                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3497                         errors_wait++;
3498                 }
3499         }
3500
3501         if (errors_wait) {
3502                 /*
3503                  * At some point we need the status of all disks
3504                  * to arrive at the volume status. So error checking
3505                  * is being pushed to a separate loop.
3506                  */
3507                 return check_barrier_error(info);
3508         }
3509         return 0;
3510 }
3511
3512 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3513 {
3514         int raid_type;
3515         int min_tolerated = INT_MAX;
3516
3517         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3518             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3519                 min_tolerated = min(min_tolerated,
3520                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3521                                     tolerated_failures);
3522
3523         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3524                 if (raid_type == BTRFS_RAID_SINGLE)
3525                         continue;
3526                 if (!(flags & btrfs_raid_group[raid_type]))
3527                         continue;
3528                 min_tolerated = min(min_tolerated,
3529                                     btrfs_raid_array[raid_type].
3530                                     tolerated_failures);
3531         }
3532
3533         if (min_tolerated == INT_MAX) {
3534                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3535                 min_tolerated = 0;
3536         }
3537
3538         return min_tolerated;
3539 }
3540
3541 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3542 {
3543         struct list_head *head;
3544         struct btrfs_device *dev;
3545         struct btrfs_super_block *sb;
3546         struct btrfs_dev_item *dev_item;
3547         int ret;
3548         int do_barriers;
3549         int max_errors;
3550         int total_errors = 0;
3551         u64 flags;
3552
3553         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3554
3555         /*
3556          * max_mirrors == 0 indicates we're from commit_transaction,
3557          * not from fsync where the tree roots in fs_info have not
3558          * been consistent on disk.
3559          */
3560         if (max_mirrors == 0)
3561                 backup_super_roots(fs_info);
3562
3563         sb = fs_info->super_for_commit;
3564         dev_item = &sb->dev_item;
3565
3566         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3567         head = &fs_info->fs_devices->devices;
3568         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3569
3570         if (do_barriers) {
3571                 ret = barrier_all_devices(fs_info);
3572                 if (ret) {
3573                         mutex_unlock(
3574                                 &fs_info->fs_devices->device_list_mutex);
3575                         btrfs_handle_fs_error(fs_info, ret,
3576                                               "errors while submitting device barriers.");
3577                         return ret;
3578                 }
3579         }
3580
3581         list_for_each_entry(dev, head, dev_list) {
3582                 if (!dev->bdev) {
3583                         total_errors++;
3584                         continue;
3585                 }
3586                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3587                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3588                         continue;
3589
3590                 btrfs_set_stack_device_generation(dev_item, 0);
3591                 btrfs_set_stack_device_type(dev_item, dev->type);
3592                 btrfs_set_stack_device_id(dev_item, dev->devid);
3593                 btrfs_set_stack_device_total_bytes(dev_item,
3594                                                    dev->commit_total_bytes);
3595                 btrfs_set_stack_device_bytes_used(dev_item,
3596                                                   dev->commit_bytes_used);
3597                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3598                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3599                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3600                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3601                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3602
3603                 flags = btrfs_super_flags(sb);
3604                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3605
3606                 ret = write_dev_supers(dev, sb, max_mirrors);
3607                 if (ret)
3608                         total_errors++;
3609         }
3610         if (total_errors > max_errors) {
3611                 btrfs_err(fs_info, "%d errors while writing supers",
3612                           total_errors);
3613                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3614
3615                 /* FUA is masked off if unsupported and can't be the reason */
3616                 btrfs_handle_fs_error(fs_info, -EIO,
3617                                       "%d errors while writing supers",
3618                                       total_errors);
3619                 return -EIO;
3620         }
3621
3622         total_errors = 0;
3623         list_for_each_entry(dev, head, dev_list) {
3624                 if (!dev->bdev)
3625                         continue;
3626                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3627                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3628                         continue;
3629
3630                 ret = wait_dev_supers(dev, max_mirrors);
3631                 if (ret)
3632                         total_errors++;
3633         }
3634         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3635         if (total_errors > max_errors) {
3636                 btrfs_handle_fs_error(fs_info, -EIO,
3637                                       "%d errors while writing supers",
3638                                       total_errors);
3639                 return -EIO;
3640         }
3641         return 0;
3642 }
3643
3644 /* Drop a fs root from the radix tree and free it. */
3645 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3646                                   struct btrfs_root *root)
3647 {
3648         spin_lock(&fs_info->fs_roots_radix_lock);
3649         radix_tree_delete(&fs_info->fs_roots_radix,
3650                           (unsigned long)root->root_key.objectid);
3651         spin_unlock(&fs_info->fs_roots_radix_lock);
3652
3653         if (btrfs_root_refs(&root->root_item) == 0)
3654                 synchronize_srcu(&fs_info->subvol_srcu);
3655
3656         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3657                 btrfs_free_log(NULL, root);
3658                 if (root->reloc_root) {
3659                         free_extent_buffer(root->reloc_root->node);
3660                         free_extent_buffer(root->reloc_root->commit_root);
3661                         btrfs_put_fs_root(root->reloc_root);
3662                         root->reloc_root = NULL;
3663                 }
3664         }
3665
3666         if (root->free_ino_pinned)
3667                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3668         if (root->free_ino_ctl)
3669                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3670         free_fs_root(root);
3671 }
3672
3673 static void free_fs_root(struct btrfs_root *root)
3674 {
3675         iput(root->ino_cache_inode);
3676         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3677         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3678         root->orphan_block_rsv = NULL;
3679         if (root->anon_dev)
3680                 free_anon_bdev(root->anon_dev);
3681         if (root->subv_writers)
3682                 btrfs_free_subvolume_writers(root->subv_writers);
3683         free_extent_buffer(root->node);
3684         free_extent_buffer(root->commit_root);
3685         kfree(root->free_ino_ctl);
3686         kfree(root->free_ino_pinned);
3687         kfree(root->name);
3688         btrfs_put_fs_root(root);
3689 }
3690
3691 void btrfs_free_fs_root(struct btrfs_root *root)
3692 {
3693         free_fs_root(root);
3694 }
3695
3696 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3697 {
3698         u64 root_objectid = 0;
3699         struct btrfs_root *gang[8];
3700         int i = 0;
3701         int err = 0;
3702         unsigned int ret = 0;
3703         int index;
3704
3705         while (1) {
3706                 index = srcu_read_lock(&fs_info->subvol_srcu);
3707                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3708                                              (void **)gang, root_objectid,
3709                                              ARRAY_SIZE(gang));
3710                 if (!ret) {
3711                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3712                         break;
3713                 }
3714                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3715
3716                 for (i = 0; i < ret; i++) {
3717                         /* Avoid to grab roots in dead_roots */
3718                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3719                                 gang[i] = NULL;
3720                                 continue;
3721                         }
3722                         /* grab all the search result for later use */
3723                         gang[i] = btrfs_grab_fs_root(gang[i]);
3724                 }
3725                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3726
3727                 for (i = 0; i < ret; i++) {
3728                         if (!gang[i])
3729                                 continue;
3730                         root_objectid = gang[i]->root_key.objectid;
3731                         err = btrfs_orphan_cleanup(gang[i]);
3732                         if (err)
3733                                 break;
3734                         btrfs_put_fs_root(gang[i]);
3735                 }
3736                 root_objectid++;
3737         }
3738
3739         /* release the uncleaned roots due to error */
3740         for (; i < ret; i++) {
3741                 if (gang[i])
3742                         btrfs_put_fs_root(gang[i]);
3743         }
3744         return err;
3745 }
3746
3747 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3748 {
3749         struct btrfs_root *root = fs_info->tree_root;
3750         struct btrfs_trans_handle *trans;
3751
3752         mutex_lock(&fs_info->cleaner_mutex);
3753         btrfs_run_delayed_iputs(fs_info);
3754         mutex_unlock(&fs_info->cleaner_mutex);
3755         wake_up_process(fs_info->cleaner_kthread);
3756
3757         /* wait until ongoing cleanup work done */
3758         down_write(&fs_info->cleanup_work_sem);
3759         up_write(&fs_info->cleanup_work_sem);
3760
3761         trans = btrfs_join_transaction(root);
3762         if (IS_ERR(trans))
3763                 return PTR_ERR(trans);
3764         return btrfs_commit_transaction(trans);
3765 }
3766
3767 void close_ctree(struct btrfs_fs_info *fs_info)
3768 {
3769         struct btrfs_root *root = fs_info->tree_root;
3770         int ret;
3771
3772         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3773
3774         /* wait for the qgroup rescan worker to stop */
3775         btrfs_qgroup_wait_for_completion(fs_info, false);
3776
3777         /* wait for the uuid_scan task to finish */
3778         down(&fs_info->uuid_tree_rescan_sem);
3779         /* avoid complains from lockdep et al., set sem back to initial state */
3780         up(&fs_info->uuid_tree_rescan_sem);
3781
3782         /* pause restriper - we want to resume on mount */
3783         btrfs_pause_balance(fs_info);
3784
3785         btrfs_dev_replace_suspend_for_unmount(fs_info);
3786
3787         btrfs_scrub_cancel(fs_info);
3788
3789         /* wait for any defraggers to finish */
3790         wait_event(fs_info->transaction_wait,
3791                    (atomic_read(&fs_info->defrag_running) == 0));
3792
3793         /* clear out the rbtree of defraggable inodes */
3794         btrfs_cleanup_defrag_inodes(fs_info);
3795
3796         cancel_work_sync(&fs_info->async_reclaim_work);