74cbc342b29ad35f8d6fb401c8bf6550c8cf27a3
[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         if (found_level)
440                 btrfs_node_key_to_cpu(eb, &found_key, 0);
441         else
442                 btrfs_item_key_to_cpu(eb, &found_key, 0);
443         ret = btrfs_comp_cpu_keys(first_key, &found_key);
444
445 #ifdef CONFIG_BTRFS_DEBUG
446         if (ret) {
447                 WARN_ON(1);
448                 btrfs_err(fs_info,
449 "tree first key mismatch detected, bytenr=%llu key expected=(%llu, %u, %llu) has=(%llu, %u, %llu)",
450                           eb->start, first_key->objectid, first_key->type,
451                           first_key->offset, found_key.objectid,
452                           found_key.type, found_key.offset);
453         }
454 #endif
455         return ret;
456 }
457
458 /*
459  * helper to read a given tree block, doing retries as required when
460  * the checksums don't match and we have alternate mirrors to try.
461  *
462  * @parent_transid:     expected transid, skip check if 0
463  * @level:              expected level, mandatory check
464  * @first_key:          expected key of first slot, skip check if NULL
465  */
466 static int btree_read_extent_buffer_pages(struct btrfs_fs_info *fs_info,
467                                           struct extent_buffer *eb,
468                                           u64 parent_transid, int level,
469                                           struct btrfs_key *first_key)
470 {
471         struct extent_io_tree *io_tree;
472         int failed = 0;
473         int ret;
474         int num_copies = 0;
475         int mirror_num = 0;
476         int failed_mirror = 0;
477
478         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
479         io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
480         while (1) {
481                 ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
482                                                mirror_num);
483                 if (!ret) {
484                         if (verify_parent_transid(io_tree, eb,
485                                                    parent_transid, 0))
486                                 ret = -EIO;
487                         else if (verify_level_key(fs_info, eb, level,
488                                                   first_key))
489                                 ret = -EUCLEAN;
490                         else
491                                 break;
492                 }
493
494                 /*
495                  * This buffer's crc is fine, but its contents are corrupted, so
496                  * there is no reason to read the other copies, they won't be
497                  * any less wrong.
498                  */
499                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags) ||
500                     ret == -EUCLEAN)
501                         break;
502
503                 num_copies = btrfs_num_copies(fs_info,
504                                               eb->start, eb->len);
505                 if (num_copies == 1)
506                         break;
507
508                 if (!failed_mirror) {
509                         failed = 1;
510                         failed_mirror = eb->read_mirror;
511                 }
512
513                 mirror_num++;
514                 if (mirror_num == failed_mirror)
515                         mirror_num++;
516
517                 if (mirror_num > num_copies)
518                         break;
519         }
520
521         if (failed && !ret && failed_mirror)
522                 repair_eb_io_failure(fs_info, eb, failed_mirror);
523
524         return ret;
525 }
526
527 /*
528  * checksum a dirty tree block before IO.  This has extra checks to make sure
529  * we only fill in the checksum field in the first page of a multi-page block
530  */
531
532 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
533 {
534         u64 start = page_offset(page);
535         u64 found_start;
536         struct extent_buffer *eb;
537
538         eb = (struct extent_buffer *)page->private;
539         if (page != eb->pages[0])
540                 return 0;
541
542         found_start = btrfs_header_bytenr(eb);
543         /*
544          * Please do not consolidate these warnings into a single if.
545          * It is useful to know what went wrong.
546          */
547         if (WARN_ON(found_start != start))
548                 return -EUCLEAN;
549         if (WARN_ON(!PageUptodate(page)))
550                 return -EUCLEAN;
551
552         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
553                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
554
555         return csum_tree_block(fs_info, eb, 0);
556 }
557
558 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
559                                  struct extent_buffer *eb)
560 {
561         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
562         u8 fsid[BTRFS_FSID_SIZE];
563         int ret = 1;
564
565         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
566         while (fs_devices) {
567                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
568                         ret = 0;
569                         break;
570                 }
571                 fs_devices = fs_devices->seed;
572         }
573         return ret;
574 }
575
576 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
577                                       u64 phy_offset, struct page *page,
578                                       u64 start, u64 end, int mirror)
579 {
580         u64 found_start;
581         int found_level;
582         struct extent_buffer *eb;
583         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
584         struct btrfs_fs_info *fs_info = root->fs_info;
585         int ret = 0;
586         int reads_done;
587
588         if (!page->private)
589                 goto out;
590
591         eb = (struct extent_buffer *)page->private;
592
593         /* the pending IO might have been the only thing that kept this buffer
594          * in memory.  Make sure we have a ref for all this other checks
595          */
596         extent_buffer_get(eb);
597
598         reads_done = atomic_dec_and_test(&eb->io_pages);
599         if (!reads_done)
600                 goto err;
601
602         eb->read_mirror = mirror;
603         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
604                 ret = -EIO;
605                 goto err;
606         }
607
608         found_start = btrfs_header_bytenr(eb);
609         if (found_start != eb->start) {
610                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
611                              found_start, eb->start);
612                 ret = -EIO;
613                 goto err;
614         }
615         if (check_tree_block_fsid(fs_info, eb)) {
616                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
617                              eb->start);
618                 ret = -EIO;
619                 goto err;
620         }
621         found_level = btrfs_header_level(eb);
622         if (found_level >= BTRFS_MAX_LEVEL) {
623                 btrfs_err(fs_info, "bad tree block level %d",
624                           (int)btrfs_header_level(eb));
625                 ret = -EIO;
626                 goto err;
627         }
628
629         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
630                                        eb, found_level);
631
632         ret = csum_tree_block(fs_info, eb, 1);
633         if (ret)
634                 goto err;
635
636         /*
637          * If this is a leaf block and it is corrupt, set the corrupt bit so
638          * that we don't try and read the other copies of this block, just
639          * return -EIO.
640          */
641         if (found_level == 0 && btrfs_check_leaf_full(fs_info, eb)) {
642                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
643                 ret = -EIO;
644         }
645
646         if (found_level > 0 && btrfs_check_node(fs_info, eb))
647                 ret = -EIO;
648
649         if (!ret)
650                 set_extent_buffer_uptodate(eb);
651 err:
652         if (reads_done &&
653             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
654                 btree_readahead_hook(eb, ret);
655
656         if (ret) {
657                 /*
658                  * our io error hook is going to dec the io pages
659                  * again, we have to make sure it has something
660                  * to decrement
661                  */
662                 atomic_inc(&eb->io_pages);
663                 clear_extent_buffer_uptodate(eb);
664         }
665         free_extent_buffer(eb);
666 out:
667         return ret;
668 }
669
670 static int btree_io_failed_hook(struct page *page, int failed_mirror)
671 {
672         struct extent_buffer *eb;
673
674         eb = (struct extent_buffer *)page->private;
675         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
676         eb->read_mirror = failed_mirror;
677         atomic_dec(&eb->io_pages);
678         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
679                 btree_readahead_hook(eb, -EIO);
680         return -EIO;    /* we fixed nothing */
681 }
682
683 static void end_workqueue_bio(struct bio *bio)
684 {
685         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
686         struct btrfs_fs_info *fs_info;
687         struct btrfs_workqueue *wq;
688         btrfs_work_func_t func;
689
690         fs_info = end_io_wq->info;
691         end_io_wq->status = bio->bi_status;
692
693         if (bio_op(bio) == REQ_OP_WRITE) {
694                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
695                         wq = fs_info->endio_meta_write_workers;
696                         func = btrfs_endio_meta_write_helper;
697                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
698                         wq = fs_info->endio_freespace_worker;
699                         func = btrfs_freespace_write_helper;
700                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
701                         wq = fs_info->endio_raid56_workers;
702                         func = btrfs_endio_raid56_helper;
703                 } else {
704                         wq = fs_info->endio_write_workers;
705                         func = btrfs_endio_write_helper;
706                 }
707         } else {
708                 if (unlikely(end_io_wq->metadata ==
709                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
710                         wq = fs_info->endio_repair_workers;
711                         func = btrfs_endio_repair_helper;
712                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
713                         wq = fs_info->endio_raid56_workers;
714                         func = btrfs_endio_raid56_helper;
715                 } else if (end_io_wq->metadata) {
716                         wq = fs_info->endio_meta_workers;
717                         func = btrfs_endio_meta_helper;
718                 } else {
719                         wq = fs_info->endio_workers;
720                         func = btrfs_endio_helper;
721                 }
722         }
723
724         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
725         btrfs_queue_work(wq, &end_io_wq->work);
726 }
727
728 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
729                         enum btrfs_wq_endio_type metadata)
730 {
731         struct btrfs_end_io_wq *end_io_wq;
732
733         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
734         if (!end_io_wq)
735                 return BLK_STS_RESOURCE;
736
737         end_io_wq->private = bio->bi_private;
738         end_io_wq->end_io = bio->bi_end_io;
739         end_io_wq->info = info;
740         end_io_wq->status = 0;
741         end_io_wq->bio = bio;
742         end_io_wq->metadata = metadata;
743
744         bio->bi_private = end_io_wq;
745         bio->bi_end_io = end_workqueue_bio;
746         return 0;
747 }
748
749 static void run_one_async_start(struct btrfs_work *work)
750 {
751         struct async_submit_bio *async;
752         blk_status_t ret;
753
754         async = container_of(work, struct  async_submit_bio, work);
755         ret = async->submit_bio_start(async->private_data, async->bio,
756                                       async->bio_offset);
757         if (ret)
758                 async->status = ret;
759 }
760
761 static void run_one_async_done(struct btrfs_work *work)
762 {
763         struct async_submit_bio *async;
764
765         async = container_of(work, struct  async_submit_bio, work);
766
767         /* If an error occurred we just want to clean up the bio and move on */
768         if (async->status) {
769                 async->bio->bi_status = async->status;
770                 bio_endio(async->bio);
771                 return;
772         }
773
774         async->submit_bio_done(async->private_data, async->bio, async->mirror_num);
775 }
776
777 static void run_one_async_free(struct btrfs_work *work)
778 {
779         struct async_submit_bio *async;
780
781         async = container_of(work, struct  async_submit_bio, work);
782         kfree(async);
783 }
784
785 blk_status_t btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
786                                  int mirror_num, unsigned long bio_flags,
787                                  u64 bio_offset, void *private_data,
788                                  extent_submit_bio_start_t *submit_bio_start,
789                                  extent_submit_bio_done_t *submit_bio_done)
790 {
791         struct async_submit_bio *async;
792
793         async = kmalloc(sizeof(*async), GFP_NOFS);
794         if (!async)
795                 return BLK_STS_RESOURCE;
796
797         async->private_data = private_data;
798         async->fs_info = fs_info;
799         async->bio = bio;
800         async->mirror_num = mirror_num;
801         async->submit_bio_start = submit_bio_start;
802         async->submit_bio_done = submit_bio_done;
803
804         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
805                         run_one_async_done, run_one_async_free);
806
807         async->bio_flags = bio_flags;
808         async->bio_offset = bio_offset;
809
810         async->status = 0;
811
812         if (op_is_sync(bio->bi_opf))
813                 btrfs_set_work_high_priority(&async->work);
814
815         btrfs_queue_work(fs_info->workers, &async->work);
816         return 0;
817 }
818
819 static blk_status_t btree_csum_one_bio(struct bio *bio)
820 {
821         struct bio_vec *bvec;
822         struct btrfs_root *root;
823         int i, ret = 0;
824
825         ASSERT(!bio_flagged(bio, BIO_CLONED));
826         bio_for_each_segment_all(bvec, bio, i) {
827                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
828                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
829                 if (ret)
830                         break;
831         }
832
833         return errno_to_blk_status(ret);
834 }
835
836 static blk_status_t btree_submit_bio_start(void *private_data, struct bio *bio,
837                                              u64 bio_offset)
838 {
839         /*
840          * when we're called for a write, we're already in the async
841          * submission context.  Just jump into btrfs_map_bio
842          */
843         return btree_csum_one_bio(bio);
844 }
845
846 static blk_status_t btree_submit_bio_done(void *private_data, struct bio *bio,
847                                             int mirror_num)
848 {
849         struct inode *inode = private_data;
850         blk_status_t ret;
851
852         /*
853          * when we're called for a write, we're already in the async
854          * submission context.  Just jump into btrfs_map_bio
855          */
856         ret = btrfs_map_bio(btrfs_sb(inode->i_sb), bio, mirror_num, 1);
857         if (ret) {
858                 bio->bi_status = ret;
859                 bio_endio(bio);
860         }
861         return ret;
862 }
863
864 static int check_async_write(struct btrfs_inode *bi)
865 {
866         if (atomic_read(&bi->sync_writers))
867                 return 0;
868 #ifdef CONFIG_X86
869         if (static_cpu_has(X86_FEATURE_XMM4_2))
870                 return 0;
871 #endif
872         return 1;
873 }
874
875 static blk_status_t btree_submit_bio_hook(void *private_data, struct bio *bio,
876                                           int mirror_num, unsigned long bio_flags,
877                                           u64 bio_offset)
878 {
879         struct inode *inode = private_data;
880         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
881         int async = check_async_write(BTRFS_I(inode));
882         blk_status_t ret;
883
884         if (bio_op(bio) != REQ_OP_WRITE) {
885                 /*
886                  * called for a read, do the setup so that checksum validation
887                  * can happen in the async kernel threads
888                  */
889                 ret = btrfs_bio_wq_end_io(fs_info, bio,
890                                           BTRFS_WQ_ENDIO_METADATA);
891                 if (ret)
892                         goto out_w_error;
893                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
894         } else if (!async) {
895                 ret = btree_csum_one_bio(bio);
896                 if (ret)
897                         goto out_w_error;
898                 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
899         } else {
900                 /*
901                  * kthread helpers are used to submit writes so that
902                  * checksumming can happen in parallel across all CPUs
903                  */
904                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, 0,
905                                           bio_offset, private_data,
906                                           btree_submit_bio_start,
907                                           btree_submit_bio_done);
908         }
909
910         if (ret)
911                 goto out_w_error;
912         return 0;
913
914 out_w_error:
915         bio->bi_status = ret;
916         bio_endio(bio);
917         return ret;
918 }
919
920 #ifdef CONFIG_MIGRATION
921 static int btree_migratepage(struct address_space *mapping,
922                         struct page *newpage, struct page *page,
923                         enum migrate_mode mode)
924 {
925         /*
926          * we can't safely write a btree page from here,
927          * we haven't done the locking hook
928          */
929         if (PageDirty(page))
930                 return -EAGAIN;
931         /*
932          * Buffers may be managed in a filesystem specific way.
933          * We must have no buffers or drop them.
934          */
935         if (page_has_private(page) &&
936             !try_to_release_page(page, GFP_KERNEL))
937                 return -EAGAIN;
938         return migrate_page(mapping, newpage, page, mode);
939 }
940 #endif
941
942
943 static int btree_writepages(struct address_space *mapping,
944                             struct writeback_control *wbc)
945 {
946         struct btrfs_fs_info *fs_info;
947         int ret;
948
949         if (wbc->sync_mode == WB_SYNC_NONE) {
950
951                 if (wbc->for_kupdate)
952                         return 0;
953
954                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
955                 /* this is a bit racy, but that's ok */
956                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
957                                              BTRFS_DIRTY_METADATA_THRESH);
958                 if (ret < 0)
959                         return 0;
960         }
961         return btree_write_cache_pages(mapping, wbc);
962 }
963
964 static int btree_readpage(struct file *file, struct page *page)
965 {
966         struct extent_io_tree *tree;
967         tree = &BTRFS_I(page->mapping->host)->io_tree;
968         return extent_read_full_page(tree, page, btree_get_extent, 0);
969 }
970
971 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
972 {
973         if (PageWriteback(page) || PageDirty(page))
974                 return 0;
975
976         return try_release_extent_buffer(page);
977 }
978
979 static void btree_invalidatepage(struct page *page, unsigned int offset,
980                                  unsigned int length)
981 {
982         struct extent_io_tree *tree;
983         tree = &BTRFS_I(page->mapping->host)->io_tree;
984         extent_invalidatepage(tree, page, offset);
985         btree_releasepage(page, GFP_NOFS);
986         if (PagePrivate(page)) {
987                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
988                            "page private not zero on page %llu",
989                            (unsigned long long)page_offset(page));
990                 ClearPagePrivate(page);
991                 set_page_private(page, 0);
992                 put_page(page);
993         }
994 }
995
996 static int btree_set_page_dirty(struct page *page)
997 {
998 #ifdef DEBUG
999         struct extent_buffer *eb;
1000
1001         BUG_ON(!PagePrivate(page));
1002         eb = (struct extent_buffer *)page->private;
1003         BUG_ON(!eb);
1004         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1005         BUG_ON(!atomic_read(&eb->refs));
1006         btrfs_assert_tree_locked(eb);
1007 #endif
1008         return __set_page_dirty_nobuffers(page);
1009 }
1010
1011 static const struct address_space_operations btree_aops = {
1012         .readpage       = btree_readpage,
1013         .writepages     = btree_writepages,
1014         .releasepage    = btree_releasepage,
1015         .invalidatepage = btree_invalidatepage,
1016 #ifdef CONFIG_MIGRATION
1017         .migratepage    = btree_migratepage,
1018 #endif
1019         .set_page_dirty = btree_set_page_dirty,
1020 };
1021
1022 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
1023 {
1024         struct extent_buffer *buf = NULL;
1025         struct inode *btree_inode = fs_info->btree_inode;
1026
1027         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1028         if (IS_ERR(buf))
1029                 return;
1030         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1031                                  buf, WAIT_NONE, 0);
1032         free_extent_buffer(buf);
1033 }
1034
1035 int reada_tree_block_flagged(struct btrfs_fs_info *fs_info, u64 bytenr,
1036                          int mirror_num, struct extent_buffer **eb)
1037 {
1038         struct extent_buffer *buf = NULL;
1039         struct inode *btree_inode = fs_info->btree_inode;
1040         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1041         int ret;
1042
1043         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1044         if (IS_ERR(buf))
1045                 return 0;
1046
1047         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1048
1049         ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1050                                        mirror_num);
1051         if (ret) {
1052                 free_extent_buffer(buf);
1053                 return ret;
1054         }
1055
1056         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1057                 free_extent_buffer(buf);
1058                 return -EIO;
1059         } else if (extent_buffer_uptodate(buf)) {
1060                 *eb = buf;
1061         } else {
1062                 free_extent_buffer(buf);
1063         }
1064         return 0;
1065 }
1066
1067 struct extent_buffer *btrfs_find_create_tree_block(
1068                                                 struct btrfs_fs_info *fs_info,
1069                                                 u64 bytenr)
1070 {
1071         if (btrfs_is_testing(fs_info))
1072                 return alloc_test_extent_buffer(fs_info, bytenr);
1073         return alloc_extent_buffer(fs_info, bytenr);
1074 }
1075
1076
1077 int btrfs_write_tree_block(struct extent_buffer *buf)
1078 {
1079         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1080                                         buf->start + buf->len - 1);
1081 }
1082
1083 void btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1084 {
1085         filemap_fdatawait_range(buf->pages[0]->mapping,
1086                                 buf->start, buf->start + buf->len - 1);
1087 }
1088
1089 /*
1090  * Read tree block at logical address @bytenr and do variant basic but critical
1091  * verification.
1092  *
1093  * @parent_transid:     expected transid of this tree block, skip check if 0
1094  * @level:              expected level, mandatory check
1095  * @first_key:          expected key in slot 0, skip check if NULL
1096  */
1097 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
1098                                       u64 parent_transid, int level,
1099                                       struct btrfs_key *first_key)
1100 {
1101         struct extent_buffer *buf = NULL;
1102         int ret;
1103
1104         buf = btrfs_find_create_tree_block(fs_info, bytenr);
1105         if (IS_ERR(buf))
1106                 return buf;
1107
1108         ret = btree_read_extent_buffer_pages(fs_info, buf, parent_transid,
1109                                              level, first_key);
1110         if (ret) {
1111                 free_extent_buffer(buf);
1112                 return ERR_PTR(ret);
1113         }
1114         return buf;
1115
1116 }
1117
1118 void clean_tree_block(struct btrfs_fs_info *fs_info,
1119                       struct extent_buffer *buf)
1120 {
1121         if (btrfs_header_generation(buf) ==
1122             fs_info->running_transaction->transid) {
1123                 btrfs_assert_tree_locked(buf);
1124
1125                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1126                         percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1127                                                  -buf->len,
1128                                                  fs_info->dirty_metadata_batch);
1129                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1130                         btrfs_set_lock_blocking(buf);
1131                         clear_extent_buffer_dirty(buf);
1132                 }
1133         }
1134 }
1135
1136 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1137 {
1138         struct btrfs_subvolume_writers *writers;
1139         int ret;
1140
1141         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1142         if (!writers)
1143                 return ERR_PTR(-ENOMEM);
1144
1145         ret = percpu_counter_init(&writers->counter, 0, GFP_NOFS);
1146         if (ret < 0) {
1147                 kfree(writers);
1148                 return ERR_PTR(ret);
1149         }
1150
1151         init_waitqueue_head(&writers->wait);
1152         return writers;
1153 }
1154
1155 static void
1156 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1157 {
1158         percpu_counter_destroy(&writers->counter);
1159         kfree(writers);
1160 }
1161
1162 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1163                          u64 objectid)
1164 {
1165         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1166         root->node = NULL;
1167         root->commit_root = NULL;
1168         root->state = 0;
1169         root->orphan_cleanup_state = 0;
1170
1171         root->objectid = objectid;
1172         root->last_trans = 0;
1173         root->highest_objectid = 0;
1174         root->nr_delalloc_inodes = 0;
1175         root->nr_ordered_extents = 0;
1176         root->name = NULL;
1177         root->inode_tree = RB_ROOT;
1178         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1179         root->block_rsv = NULL;
1180         root->orphan_block_rsv = NULL;
1181
1182         INIT_LIST_HEAD(&root->dirty_list);
1183         INIT_LIST_HEAD(&root->root_list);
1184         INIT_LIST_HEAD(&root->delalloc_inodes);
1185         INIT_LIST_HEAD(&root->delalloc_root);
1186         INIT_LIST_HEAD(&root->ordered_extents);
1187         INIT_LIST_HEAD(&root->ordered_root);
1188         INIT_LIST_HEAD(&root->logged_list[0]);
1189         INIT_LIST_HEAD(&root->logged_list[1]);
1190         spin_lock_init(&root->orphan_lock);
1191         spin_lock_init(&root->inode_lock);
1192         spin_lock_init(&root->delalloc_lock);
1193         spin_lock_init(&root->ordered_extent_lock);
1194         spin_lock_init(&root->accounting_lock);
1195         spin_lock_init(&root->log_extents_lock[0]);
1196         spin_lock_init(&root->log_extents_lock[1]);
1197         spin_lock_init(&root->qgroup_meta_rsv_lock);
1198         mutex_init(&root->objectid_mutex);
1199         mutex_init(&root->log_mutex);
1200         mutex_init(&root->ordered_extent_mutex);
1201         mutex_init(&root->delalloc_mutex);
1202         init_waitqueue_head(&root->log_writer_wait);
1203         init_waitqueue_head(&root->log_commit_wait[0]);
1204         init_waitqueue_head(&root->log_commit_wait[1]);
1205         INIT_LIST_HEAD(&root->log_ctxs[0]);
1206         INIT_LIST_HEAD(&root->log_ctxs[1]);
1207         atomic_set(&root->log_commit[0], 0);
1208         atomic_set(&root->log_commit[1], 0);
1209         atomic_set(&root->log_writers, 0);
1210         atomic_set(&root->log_batch, 0);
1211         atomic_set(&root->orphan_inodes, 0);
1212         refcount_set(&root->refs, 1);
1213         atomic_set(&root->will_be_snapshotted, 0);
1214         root->log_transid = 0;
1215         root->log_transid_committed = -1;
1216         root->last_log_commit = 0;
1217         if (!dummy)
1218                 extent_io_tree_init(&root->dirty_log_pages, NULL);
1219
1220         memset(&root->root_key, 0, sizeof(root->root_key));
1221         memset(&root->root_item, 0, sizeof(root->root_item));
1222         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1223         if (!dummy)
1224                 root->defrag_trans_start = fs_info->generation;
1225         else
1226                 root->defrag_trans_start = 0;
1227         root->root_key.objectid = objectid;
1228         root->anon_dev = 0;
1229
1230         spin_lock_init(&root->root_item_lock);
1231 }
1232
1233 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1234                 gfp_t flags)
1235 {
1236         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1237         if (root)
1238                 root->fs_info = fs_info;
1239         return root;
1240 }
1241
1242 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1243 /* Should only be used by the testing infrastructure */
1244 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1245 {
1246         struct btrfs_root *root;
1247
1248         if (!fs_info)
1249                 return ERR_PTR(-EINVAL);
1250
1251         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1252         if (!root)
1253                 return ERR_PTR(-ENOMEM);
1254
1255         /* We don't use the stripesize in selftest, set it as sectorsize */
1256         __setup_root(root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
1257         root->alloc_bytenr = 0;
1258
1259         return root;
1260 }
1261 #endif
1262
1263 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1264                                      struct btrfs_fs_info *fs_info,
1265                                      u64 objectid)
1266 {
1267         struct extent_buffer *leaf;
1268         struct btrfs_root *tree_root = fs_info->tree_root;
1269         struct btrfs_root *root;
1270         struct btrfs_key key;
1271         int ret = 0;
1272         uuid_le uuid = NULL_UUID_LE;
1273
1274         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1275         if (!root)
1276                 return ERR_PTR(-ENOMEM);
1277
1278         __setup_root(root, fs_info, objectid);
1279         root->root_key.objectid = objectid;
1280         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1281         root->root_key.offset = 0;
1282
1283         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1284         if (IS_ERR(leaf)) {
1285                 ret = PTR_ERR(leaf);
1286                 leaf = NULL;
1287                 goto fail;
1288         }
1289
1290         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1291         btrfs_set_header_bytenr(leaf, leaf->start);
1292         btrfs_set_header_generation(leaf, trans->transid);
1293         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1294         btrfs_set_header_owner(leaf, objectid);
1295         root->node = leaf;
1296
1297         write_extent_buffer_fsid(leaf, fs_info->fsid);
1298         write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid);
1299         btrfs_mark_buffer_dirty(leaf);
1300
1301         root->commit_root = btrfs_root_node(root);
1302         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1303
1304         root->root_item.flags = 0;
1305         root->root_item.byte_limit = 0;
1306         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1307         btrfs_set_root_generation(&root->root_item, trans->transid);
1308         btrfs_set_root_level(&root->root_item, 0);
1309         btrfs_set_root_refs(&root->root_item, 1);
1310         btrfs_set_root_used(&root->root_item, leaf->len);
1311         btrfs_set_root_last_snapshot(&root->root_item, 0);
1312         btrfs_set_root_dirid(&root->root_item, 0);
1313         if (is_fstree(objectid))
1314                 uuid_le_gen(&uuid);
1315         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1316         root->root_item.drop_level = 0;
1317
1318         key.objectid = objectid;
1319         key.type = BTRFS_ROOT_ITEM_KEY;
1320         key.offset = 0;
1321         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1322         if (ret)
1323                 goto fail;
1324
1325         btrfs_tree_unlock(leaf);
1326
1327         return root;
1328
1329 fail:
1330         if (leaf) {
1331                 btrfs_tree_unlock(leaf);
1332                 free_extent_buffer(root->commit_root);
1333                 free_extent_buffer(leaf);
1334         }
1335         kfree(root);
1336
1337         return ERR_PTR(ret);
1338 }
1339
1340 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1341                                          struct btrfs_fs_info *fs_info)
1342 {
1343         struct btrfs_root *root;
1344         struct extent_buffer *leaf;
1345
1346         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1347         if (!root)
1348                 return ERR_PTR(-ENOMEM);
1349
1350         __setup_root(root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1351
1352         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1353         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1354         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1355
1356         /*
1357          * DON'T set REF_COWS for log trees
1358          *
1359          * log trees do not get reference counted because they go away
1360          * before a real commit is actually done.  They do store pointers
1361          * to file data extents, and those reference counts still get
1362          * updated (along with back refs to the log tree).
1363          */
1364
1365         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1366                         NULL, 0, 0, 0);
1367         if (IS_ERR(leaf)) {
1368                 kfree(root);
1369                 return ERR_CAST(leaf);
1370         }
1371
1372         memzero_extent_buffer(leaf, 0, sizeof(struct btrfs_header));
1373         btrfs_set_header_bytenr(leaf, leaf->start);
1374         btrfs_set_header_generation(leaf, trans->transid);
1375         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1376         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1377         root->node = leaf;
1378
1379         write_extent_buffer_fsid(root->node, fs_info->fsid);
1380         btrfs_mark_buffer_dirty(root->node);
1381         btrfs_tree_unlock(root->node);
1382         return root;
1383 }
1384
1385 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1386                              struct btrfs_fs_info *fs_info)
1387 {
1388         struct btrfs_root *log_root;
1389
1390         log_root = alloc_log_tree(trans, fs_info);
1391         if (IS_ERR(log_root))
1392                 return PTR_ERR(log_root);
1393         WARN_ON(fs_info->log_root_tree);
1394         fs_info->log_root_tree = log_root;
1395         return 0;
1396 }
1397
1398 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1399                        struct btrfs_root *root)
1400 {
1401         struct btrfs_fs_info *fs_info = root->fs_info;
1402         struct btrfs_root *log_root;
1403         struct btrfs_inode_item *inode_item;
1404
1405         log_root = alloc_log_tree(trans, fs_info);
1406         if (IS_ERR(log_root))
1407                 return PTR_ERR(log_root);
1408
1409         log_root->last_trans = trans->transid;
1410         log_root->root_key.offset = root->root_key.objectid;
1411
1412         inode_item = &log_root->root_item.inode;
1413         btrfs_set_stack_inode_generation(inode_item, 1);
1414         btrfs_set_stack_inode_size(inode_item, 3);
1415         btrfs_set_stack_inode_nlink(inode_item, 1);
1416         btrfs_set_stack_inode_nbytes(inode_item,
1417                                      fs_info->nodesize);
1418         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1419
1420         btrfs_set_root_node(&log_root->root_item, log_root->node);
1421
1422         WARN_ON(root->log_root);
1423         root->log_root = log_root;
1424         root->log_transid = 0;
1425         root->log_transid_committed = -1;
1426         root->last_log_commit = 0;
1427         return 0;
1428 }
1429
1430 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1431                                                struct btrfs_key *key)
1432 {
1433         struct btrfs_root *root;
1434         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1435         struct btrfs_path *path;
1436         u64 generation;
1437         int ret;
1438         int level;
1439
1440         path = btrfs_alloc_path();
1441         if (!path)
1442                 return ERR_PTR(-ENOMEM);
1443
1444         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1445         if (!root) {
1446                 ret = -ENOMEM;
1447                 goto alloc_fail;
1448         }
1449
1450         __setup_root(root, fs_info, key->objectid);
1451
1452         ret = btrfs_find_root(tree_root, key, path,
1453                               &root->root_item, &root->root_key);
1454         if (ret) {
1455                 if (ret > 0)
1456                         ret = -ENOENT;
1457                 goto find_fail;
1458         }
1459
1460         generation = btrfs_root_generation(&root->root_item);
1461         level = btrfs_root_level(&root->root_item);
1462         root->node = read_tree_block(fs_info,
1463                                      btrfs_root_bytenr(&root->root_item),
1464                                      generation, level, NULL);
1465         if (IS_ERR(root->node)) {
1466                 ret = PTR_ERR(root->node);
1467                 goto find_fail;
1468         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1469                 ret = -EIO;
1470                 free_extent_buffer(root->node);
1471                 goto find_fail;
1472         }
1473         root->commit_root = btrfs_root_node(root);
1474 out:
1475         btrfs_free_path(path);
1476         return root;
1477
1478 find_fail:
1479         kfree(root);
1480 alloc_fail:
1481         root = ERR_PTR(ret);
1482         goto out;
1483 }
1484
1485 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1486                                       struct btrfs_key *location)
1487 {
1488         struct btrfs_root *root;
1489
1490         root = btrfs_read_tree_root(tree_root, location);
1491         if (IS_ERR(root))
1492                 return root;
1493
1494         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1495                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1496                 btrfs_check_and_init_root_item(&root->root_item);
1497         }
1498
1499         return root;
1500 }
1501
1502 int btrfs_init_fs_root(struct btrfs_root *root)
1503 {
1504         int ret;
1505         struct btrfs_subvolume_writers *writers;
1506
1507         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1508         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1509                                         GFP_NOFS);
1510         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1511                 ret = -ENOMEM;
1512                 goto fail;
1513         }
1514
1515         writers = btrfs_alloc_subvolume_writers();
1516         if (IS_ERR(writers)) {
1517                 ret = PTR_ERR(writers);
1518                 goto fail;
1519         }
1520         root->subv_writers = writers;
1521
1522         btrfs_init_free_ino_ctl(root);
1523         spin_lock_init(&root->ino_cache_lock);
1524         init_waitqueue_head(&root->ino_cache_wait);
1525
1526         ret = get_anon_bdev(&root->anon_dev);
1527         if (ret)
1528                 goto fail;
1529
1530         mutex_lock(&root->objectid_mutex);
1531         ret = btrfs_find_highest_objectid(root,
1532                                         &root->highest_objectid);
1533         if (ret) {
1534                 mutex_unlock(&root->objectid_mutex);
1535                 goto fail;
1536         }
1537
1538         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1539
1540         mutex_unlock(&root->objectid_mutex);
1541
1542         return 0;
1543 fail:
1544         /* the caller is responsible to call free_fs_root */
1545         return ret;
1546 }
1547
1548 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1549                                         u64 root_id)
1550 {
1551         struct btrfs_root *root;
1552
1553         spin_lock(&fs_info->fs_roots_radix_lock);
1554         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1555                                  (unsigned long)root_id);
1556         spin_unlock(&fs_info->fs_roots_radix_lock);
1557         return root;
1558 }
1559
1560 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1561                          struct btrfs_root *root)
1562 {
1563         int ret;
1564
1565         ret = radix_tree_preload(GFP_NOFS);
1566         if (ret)
1567                 return ret;
1568
1569         spin_lock(&fs_info->fs_roots_radix_lock);
1570         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1571                                 (unsigned long)root->root_key.objectid,
1572                                 root);
1573         if (ret == 0)
1574                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1575         spin_unlock(&fs_info->fs_roots_radix_lock);
1576         radix_tree_preload_end();
1577
1578         return ret;
1579 }
1580
1581 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1582                                      struct btrfs_key *location,
1583                                      bool check_ref)
1584 {
1585         struct btrfs_root *root;
1586         struct btrfs_path *path;
1587         struct btrfs_key key;
1588         int ret;
1589
1590         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1591                 return fs_info->tree_root;
1592         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1593                 return fs_info->extent_root;
1594         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1595                 return fs_info->chunk_root;
1596         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1597                 return fs_info->dev_root;
1598         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1599                 return fs_info->csum_root;
1600         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1601                 return fs_info->quota_root ? fs_info->quota_root :
1602                                              ERR_PTR(-ENOENT);
1603         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1604                 return fs_info->uuid_root ? fs_info->uuid_root :
1605                                             ERR_PTR(-ENOENT);
1606         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1607                 return fs_info->free_space_root ? fs_info->free_space_root :
1608                                                   ERR_PTR(-ENOENT);
1609 again:
1610         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1611         if (root) {
1612                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1613                         return ERR_PTR(-ENOENT);
1614                 return root;
1615         }
1616
1617         root = btrfs_read_fs_root(fs_info->tree_root, location);
1618         if (IS_ERR(root))
1619                 return root;
1620
1621         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1622                 ret = -ENOENT;
1623                 goto fail;
1624         }
1625
1626         ret = btrfs_init_fs_root(root);
1627         if (ret)
1628                 goto fail;
1629
1630         path = btrfs_alloc_path();
1631         if (!path) {
1632                 ret = -ENOMEM;
1633                 goto fail;
1634         }
1635         key.objectid = BTRFS_ORPHAN_OBJECTID;
1636         key.type = BTRFS_ORPHAN_ITEM_KEY;
1637         key.offset = location->objectid;
1638
1639         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1640         btrfs_free_path(path);
1641         if (ret < 0)
1642                 goto fail;
1643         if (ret == 0)
1644                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1645
1646         ret = btrfs_insert_fs_root(fs_info, root);
1647         if (ret) {
1648                 if (ret == -EEXIST) {
1649                         free_fs_root(root);
1650                         goto again;
1651                 }
1652                 goto fail;
1653         }
1654         return root;
1655 fail:
1656         free_fs_root(root);
1657         return ERR_PTR(ret);
1658 }
1659
1660 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1661 {
1662         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1663         int ret = 0;
1664         struct btrfs_device *device;
1665         struct backing_dev_info *bdi;
1666
1667         rcu_read_lock();
1668         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1669                 if (!device->bdev)
1670                         continue;
1671                 bdi = device->bdev->bd_bdi;
1672                 if (bdi_congested(bdi, bdi_bits)) {
1673                         ret = 1;
1674                         break;
1675                 }
1676         }
1677         rcu_read_unlock();
1678         return ret;
1679 }
1680
1681 /*
1682  * called by the kthread helper functions to finally call the bio end_io
1683  * functions.  This is where read checksum verification actually happens
1684  */
1685 static void end_workqueue_fn(struct btrfs_work *work)
1686 {
1687         struct bio *bio;
1688         struct btrfs_end_io_wq *end_io_wq;
1689
1690         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1691         bio = end_io_wq->bio;
1692
1693         bio->bi_status = end_io_wq->status;
1694         bio->bi_private = end_io_wq->private;
1695         bio->bi_end_io = end_io_wq->end_io;
1696         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1697         bio_endio(bio);
1698 }
1699
1700 static int cleaner_kthread(void *arg)
1701 {
1702         struct btrfs_root *root = arg;
1703         struct btrfs_fs_info *fs_info = root->fs_info;
1704         int again;
1705         struct btrfs_trans_handle *trans;
1706
1707         do {
1708                 again = 0;
1709
1710                 /* Make the cleaner go to sleep early. */
1711                 if (btrfs_need_cleaner_sleep(fs_info))
1712                         goto sleep;
1713
1714                 /*
1715                  * Do not do anything if we might cause open_ctree() to block
1716                  * before we have finished mounting the filesystem.
1717                  */
1718                 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1719                         goto sleep;
1720
1721                 if (!mutex_trylock(&fs_info->cleaner_mutex))
1722                         goto sleep;
1723
1724                 /*
1725                  * Avoid the problem that we change the status of the fs
1726                  * during the above check and trylock.
1727                  */
1728                 if (btrfs_need_cleaner_sleep(fs_info)) {
1729                         mutex_unlock(&fs_info->cleaner_mutex);
1730                         goto sleep;
1731                 }
1732
1733                 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
1734                 btrfs_run_delayed_iputs(fs_info);
1735                 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
1736
1737                 again = btrfs_clean_one_deleted_snapshot(root);
1738                 mutex_unlock(&fs_info->cleaner_mutex);
1739
1740                 /*
1741                  * The defragger has dealt with the R/O remount and umount,
1742                  * needn't do anything special here.
1743                  */
1744                 btrfs_run_defrag_inodes(fs_info);
1745
1746                 /*
1747                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1748                  * with relocation (btrfs_relocate_chunk) and relocation
1749                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1750                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1751                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1752                  * unused block groups.
1753                  */
1754                 btrfs_delete_unused_bgs(fs_info);
1755 sleep:
1756                 if (!again) {
1757                         set_current_state(TASK_INTERRUPTIBLE);
1758                         if (!kthread_should_stop())
1759                                 schedule();
1760                         __set_current_state(TASK_RUNNING);
1761                 }
1762         } while (!kthread_should_stop());
1763
1764         /*
1765          * Transaction kthread is stopped before us and wakes us up.
1766          * However we might have started a new transaction and COWed some
1767          * tree blocks when deleting unused block groups for example. So
1768          * make sure we commit the transaction we started to have a clean
1769          * shutdown when evicting the btree inode - if it has dirty pages
1770          * when we do the final iput() on it, eviction will trigger a
1771          * writeback for it which will fail with null pointer dereferences
1772          * since work queues and other resources were already released and
1773          * destroyed by the time the iput/eviction/writeback is made.
1774          */
1775         trans = btrfs_attach_transaction(root);
1776         if (IS_ERR(trans)) {
1777                 if (PTR_ERR(trans) != -ENOENT)
1778                         btrfs_err(fs_info,
1779                                   "cleaner transaction attach returned %ld",
1780                                   PTR_ERR(trans));
1781         } else {
1782                 int ret;
1783
1784                 ret = btrfs_commit_transaction(trans);
1785                 if (ret)
1786                         btrfs_err(fs_info,
1787                                   "cleaner open transaction commit returned %d",
1788                                   ret);
1789         }
1790
1791         return 0;
1792 }
1793
1794 static int transaction_kthread(void *arg)
1795 {
1796         struct btrfs_root *root = arg;
1797         struct btrfs_fs_info *fs_info = root->fs_info;
1798         struct btrfs_trans_handle *trans;
1799         struct btrfs_transaction *cur;
1800         u64 transid;
1801         unsigned long now;
1802         unsigned long delay;
1803         bool cannot_commit;
1804
1805         do {
1806                 cannot_commit = false;
1807                 delay = HZ * fs_info->commit_interval;
1808                 mutex_lock(&fs_info->transaction_kthread_mutex);
1809
1810                 spin_lock(&fs_info->trans_lock);
1811                 cur = fs_info->running_transaction;
1812                 if (!cur) {
1813                         spin_unlock(&fs_info->trans_lock);
1814                         goto sleep;
1815                 }
1816
1817                 now = get_seconds();
1818                 if (cur->state < TRANS_STATE_BLOCKED &&
1819                     (now < cur->start_time ||
1820                      now - cur->start_time < fs_info->commit_interval)) {
1821                         spin_unlock(&fs_info->trans_lock);
1822                         delay = HZ * 5;
1823                         goto sleep;
1824                 }
1825                 transid = cur->transid;
1826                 spin_unlock(&fs_info->trans_lock);
1827
1828                 /* If the file system is aborted, this will always fail. */
1829                 trans = btrfs_attach_transaction(root);
1830                 if (IS_ERR(trans)) {
1831                         if (PTR_ERR(trans) != -ENOENT)
1832                                 cannot_commit = true;
1833                         goto sleep;
1834                 }
1835                 if (transid == trans->transid) {
1836                         btrfs_commit_transaction(trans);
1837                 } else {
1838                         btrfs_end_transaction(trans);
1839                 }
1840 sleep:
1841                 wake_up_process(fs_info->cleaner_kthread);
1842                 mutex_unlock(&fs_info->transaction_kthread_mutex);
1843
1844                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1845                                       &fs_info->fs_state)))
1846                         btrfs_cleanup_transaction(fs_info);
1847                 if (!kthread_should_stop() &&
1848                                 (!btrfs_transaction_blocked(fs_info) ||
1849                                  cannot_commit))
1850                         schedule_timeout_interruptible(delay);
1851         } while (!kthread_should_stop());
1852         return 0;
1853 }
1854
1855 /*
1856  * this will find the highest generation in the array of
1857  * root backups.  The index of the highest array is returned,
1858  * or -1 if we can't find anything.
1859  *
1860  * We check to make sure the array is valid by comparing the
1861  * generation of the latest  root in the array with the generation
1862  * in the super block.  If they don't match we pitch it.
1863  */
1864 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1865 {
1866         u64 cur;
1867         int newest_index = -1;
1868         struct btrfs_root_backup *root_backup;
1869         int i;
1870
1871         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1872                 root_backup = info->super_copy->super_roots + i;
1873                 cur = btrfs_backup_tree_root_gen(root_backup);
1874                 if (cur == newest_gen)
1875                         newest_index = i;
1876         }
1877
1878         /* check to see if we actually wrapped around */
1879         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1880                 root_backup = info->super_copy->super_roots;
1881                 cur = btrfs_backup_tree_root_gen(root_backup);
1882                 if (cur == newest_gen)
1883                         newest_index = 0;
1884         }
1885         return newest_index;
1886 }
1887
1888
1889 /*
1890  * find the oldest backup so we know where to store new entries
1891  * in the backup array.  This will set the backup_root_index
1892  * field in the fs_info struct
1893  */
1894 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1895                                      u64 newest_gen)
1896 {
1897         int newest_index = -1;
1898
1899         newest_index = find_newest_super_backup(info, newest_gen);
1900         /* if there was garbage in there, just move along */
1901         if (newest_index == -1) {
1902                 info->backup_root_index = 0;
1903         } else {
1904                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1905         }
1906 }
1907
1908 /*
1909  * copy all the root pointers into the super backup array.
1910  * this will bump the backup pointer by one when it is
1911  * done
1912  */
1913 static void backup_super_roots(struct btrfs_fs_info *info)
1914 {
1915         int next_backup;
1916         struct btrfs_root_backup *root_backup;
1917         int last_backup;
1918
1919         next_backup = info->backup_root_index;
1920         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1921                 BTRFS_NUM_BACKUP_ROOTS;
1922
1923         /*
1924          * just overwrite the last backup if we're at the same generation
1925          * this happens only at umount
1926          */
1927         root_backup = info->super_for_commit->super_roots + last_backup;
1928         if (btrfs_backup_tree_root_gen(root_backup) ==
1929             btrfs_header_generation(info->tree_root->node))
1930                 next_backup = last_backup;
1931
1932         root_backup = info->super_for_commit->super_roots + next_backup;
1933
1934         /*
1935          * make sure all of our padding and empty slots get zero filled
1936          * regardless of which ones we use today
1937          */
1938         memset(root_backup, 0, sizeof(*root_backup));
1939
1940         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1941
1942         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1943         btrfs_set_backup_tree_root_gen(root_backup,
1944                                btrfs_header_generation(info->tree_root->node));
1945
1946         btrfs_set_backup_tree_root_level(root_backup,
1947                                btrfs_header_level(info->tree_root->node));
1948
1949         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1950         btrfs_set_backup_chunk_root_gen(root_backup,
1951                                btrfs_header_generation(info->chunk_root->node));
1952         btrfs_set_backup_chunk_root_level(root_backup,
1953                                btrfs_header_level(info->chunk_root->node));
1954
1955         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1956         btrfs_set_backup_extent_root_gen(root_backup,
1957                                btrfs_header_generation(info->extent_root->node));
1958         btrfs_set_backup_extent_root_level(root_backup,
1959                                btrfs_header_level(info->extent_root->node));
1960
1961         /*
1962          * we might commit during log recovery, which happens before we set
1963          * the fs_root.  Make sure it is valid before we fill it in.
1964          */
1965         if (info->fs_root && info->fs_root->node) {
1966                 btrfs_set_backup_fs_root(root_backup,
1967                                          info->fs_root->node->start);
1968                 btrfs_set_backup_fs_root_gen(root_backup,
1969                                btrfs_header_generation(info->fs_root->node));
1970                 btrfs_set_backup_fs_root_level(root_backup,
1971                                btrfs_header_level(info->fs_root->node));
1972         }
1973
1974         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1975         btrfs_set_backup_dev_root_gen(root_backup,
1976                                btrfs_header_generation(info->dev_root->node));
1977         btrfs_set_backup_dev_root_level(root_backup,
1978                                        btrfs_header_level(info->dev_root->node));
1979
1980         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1981         btrfs_set_backup_csum_root_gen(root_backup,
1982                                btrfs_header_generation(info->csum_root->node));
1983         btrfs_set_backup_csum_root_level(root_backup,
1984                                btrfs_header_level(info->csum_root->node));
1985
1986         btrfs_set_backup_total_bytes(root_backup,
1987                              btrfs_super_total_bytes(info->super_copy));
1988         btrfs_set_backup_bytes_used(root_backup,
1989                              btrfs_super_bytes_used(info->super_copy));
1990         btrfs_set_backup_num_devices(root_backup,
1991                              btrfs_super_num_devices(info->super_copy));
1992
1993         /*
1994          * if we don't copy this out to the super_copy, it won't get remembered
1995          * for the next commit
1996          */
1997         memcpy(&info->super_copy->super_roots,
1998                &info->super_for_commit->super_roots,
1999                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2000 }
2001
2002 /*
2003  * this copies info out of the root backup array and back into
2004  * the in-memory super block.  It is meant to help iterate through
2005  * the array, so you send it the number of backups you've already
2006  * tried and the last backup index you used.
2007  *
2008  * this returns -1 when it has tried all the backups
2009  */
2010 static noinline int next_root_backup(struct btrfs_fs_info *info,
2011                                      struct btrfs_super_block *super,
2012                                      int *num_backups_tried, int *backup_index)
2013 {
2014         struct btrfs_root_backup *root_backup;
2015         int newest = *backup_index;
2016
2017         if (*num_backups_tried == 0) {
2018                 u64 gen = btrfs_super_generation(super);
2019
2020                 newest = find_newest_super_backup(info, gen);
2021                 if (newest == -1)
2022                         return -1;
2023
2024                 *backup_index = newest;
2025                 *num_backups_tried = 1;
2026         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2027                 /* we've tried all the backups, all done */
2028                 return -1;
2029         } else {
2030                 /* jump to the next oldest backup */
2031                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2032                         BTRFS_NUM_BACKUP_ROOTS;
2033                 *backup_index = newest;
2034                 *num_backups_tried += 1;
2035         }
2036         root_backup = super->super_roots + newest;
2037
2038         btrfs_set_super_generation(super,
2039                                    btrfs_backup_tree_root_gen(root_backup));
2040         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2041         btrfs_set_super_root_level(super,
2042                                    btrfs_backup_tree_root_level(root_backup));
2043         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2044
2045         /*
2046          * fixme: the total bytes and num_devices need to match or we should
2047          * need a fsck
2048          */
2049         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2050         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2051         return 0;
2052 }
2053
2054 /* helper to cleanup workers */
2055 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2056 {
2057         btrfs_destroy_workqueue(fs_info->fixup_workers);
2058         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2059         btrfs_destroy_workqueue(fs_info->workers);
2060         btrfs_destroy_workqueue(fs_info->endio_workers);
2061         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2062         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2063         btrfs_destroy_workqueue(fs_info->rmw_workers);
2064         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2065         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2066         btrfs_destroy_workqueue(fs_info->submit_workers);
2067         btrfs_destroy_workqueue(fs_info->delayed_workers);
2068         btrfs_destroy_workqueue(fs_info->caching_workers);
2069         btrfs_destroy_workqueue(fs_info->readahead_workers);
2070         btrfs_destroy_workqueue(fs_info->flush_workers);
2071         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2072         btrfs_destroy_workqueue(fs_info->extent_workers);
2073         /*
2074          * Now that all other work queues are destroyed, we can safely destroy
2075          * the queues used for metadata I/O, since tasks from those other work
2076          * queues can do metadata I/O operations.
2077          */
2078         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2079         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2080 }
2081
2082 static void free_root_extent_buffers(struct btrfs_root *root)
2083 {
2084         if (root) {
2085                 free_extent_buffer(root->node);
2086                 free_extent_buffer(root->commit_root);
2087                 root->node = NULL;
2088                 root->commit_root = NULL;
2089         }
2090 }
2091
2092 /* helper to cleanup tree roots */
2093 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2094 {
2095         free_root_extent_buffers(info->tree_root);
2096
2097         free_root_extent_buffers(info->dev_root);
2098         free_root_extent_buffers(info->extent_root);
2099         free_root_extent_buffers(info->csum_root);
2100         free_root_extent_buffers(info->quota_root);
2101         free_root_extent_buffers(info->uuid_root);
2102         if (chunk_root)
2103                 free_root_extent_buffers(info->chunk_root);
2104         free_root_extent_buffers(info->free_space_root);
2105 }
2106
2107 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2108 {
2109         int ret;
2110         struct btrfs_root *gang[8];
2111         int i;
2112
2113         while (!list_empty(&fs_info->dead_roots)) {
2114                 gang[0] = list_entry(fs_info->dead_roots.next,
2115                                      struct btrfs_root, root_list);
2116                 list_del(&gang[0]->root_list);
2117
2118                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2119                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2120                 } else {
2121                         free_extent_buffer(gang[0]->node);
2122                         free_extent_buffer(gang[0]->commit_root);
2123                         btrfs_put_fs_root(gang[0]);
2124                 }
2125         }
2126
2127         while (1) {
2128                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2129                                              (void **)gang, 0,
2130                                              ARRAY_SIZE(gang));
2131                 if (!ret)
2132                         break;
2133                 for (i = 0; i < ret; i++)
2134                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2135         }
2136
2137         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2138                 btrfs_free_log_root_tree(NULL, fs_info);
2139                 btrfs_destroy_pinned_extent(fs_info, fs_info->pinned_extents);
2140         }
2141 }
2142
2143 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2144 {
2145         mutex_init(&fs_info->scrub_lock);
2146         atomic_set(&fs_info->scrubs_running, 0);
2147         atomic_set(&fs_info->scrub_pause_req, 0);
2148         atomic_set(&fs_info->scrubs_paused, 0);
2149         atomic_set(&fs_info->scrub_cancel_req, 0);
2150         init_waitqueue_head(&fs_info->scrub_pause_wait);
2151         fs_info->scrub_workers_refcnt = 0;
2152 }
2153
2154 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2155 {
2156         spin_lock_init(&fs_info->balance_lock);
2157         mutex_init(&fs_info->balance_mutex);
2158         atomic_set(&fs_info->balance_running, 0);
2159         atomic_set(&fs_info->balance_pause_req, 0);
2160         atomic_set(&fs_info->balance_cancel_req, 0);
2161         fs_info->balance_ctl = NULL;
2162         init_waitqueue_head(&fs_info->balance_wait_q);
2163 }
2164
2165 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2166 {
2167         struct inode *inode = fs_info->btree_inode;
2168
2169         inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2170         set_nlink(inode, 1);
2171         /*
2172          * we set the i_size on the btree inode to the max possible int.
2173          * the real end of the address space is determined by all of
2174          * the devices in the system
2175          */
2176         inode->i_size = OFFSET_MAX;
2177         inode->i_mapping->a_ops = &btree_aops;
2178
2179         RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2180         extent_io_tree_init(&BTRFS_I(inode)->io_tree, inode);
2181         BTRFS_I(inode)->io_tree.track_uptodate = 0;
2182         extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2183
2184         BTRFS_I(inode)->io_tree.ops = &btree_extent_io_ops;
2185
2186         BTRFS_I(inode)->root = fs_info->tree_root;
2187         memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2188         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2189         btrfs_insert_inode_hash(inode);
2190 }
2191
2192 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2193 {
2194         fs_info->dev_replace.lock_owner = 0;
2195         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2196         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2197         rwlock_init(&fs_info->dev_replace.lock);
2198         atomic_set(&fs_info->dev_replace.read_locks, 0);
2199         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2200         init_waitqueue_head(&fs_info->replace_wait);
2201         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2202 }
2203
2204 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2205 {
2206         spin_lock_init(&fs_info->qgroup_lock);
2207         mutex_init(&fs_info->qgroup_ioctl_lock);
2208         fs_info->qgroup_tree = RB_ROOT;
2209         fs_info->qgroup_op_tree = RB_ROOT;
2210         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2211         fs_info->qgroup_seq = 1;
2212         fs_info->qgroup_ulist = NULL;
2213         fs_info->qgroup_rescan_running = false;
2214         mutex_init(&fs_info->qgroup_rescan_lock);
2215 }
2216
2217 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2218                 struct btrfs_fs_devices *fs_devices)
2219 {
2220         u32 max_active = fs_info->thread_pool_size;
2221         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2222
2223         fs_info->workers =
2224                 btrfs_alloc_workqueue(fs_info, "worker",
2225                                       flags | WQ_HIGHPRI, max_active, 16);
2226
2227         fs_info->delalloc_workers =
2228                 btrfs_alloc_workqueue(fs_info, "delalloc",
2229                                       flags, max_active, 2);
2230
2231         fs_info->flush_workers =
2232                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2233                                       flags, max_active, 0);
2234
2235         fs_info->caching_workers =
2236                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2237
2238         /*
2239          * a higher idle thresh on the submit workers makes it much more
2240          * likely that bios will be send down in a sane order to the
2241          * devices
2242          */
2243         fs_info->submit_workers =
2244                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2245                                       min_t(u64, fs_devices->num_devices,
2246                                             max_active), 64);
2247
2248         fs_info->fixup_workers =
2249                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2250
2251         /*
2252          * endios are largely parallel and should have a very
2253          * low idle thresh
2254          */
2255         fs_info->endio_workers =
2256                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2257         fs_info->endio_meta_workers =
2258                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2259                                       max_active, 4);
2260         fs_info->endio_meta_write_workers =
2261                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2262                                       max_active, 2);
2263         fs_info->endio_raid56_workers =
2264                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2265                                       max_active, 4);
2266         fs_info->endio_repair_workers =
2267                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2268         fs_info->rmw_workers =
2269                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2270         fs_info->endio_write_workers =
2271                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2272                                       max_active, 2);
2273         fs_info->endio_freespace_worker =
2274                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2275                                       max_active, 0);
2276         fs_info->delayed_workers =
2277                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2278                                       max_active, 0);
2279         fs_info->readahead_workers =
2280                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2281                                       max_active, 2);
2282         fs_info->qgroup_rescan_workers =
2283                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2284         fs_info->extent_workers =
2285                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2286                                       min_t(u64, fs_devices->num_devices,
2287                                             max_active), 8);
2288
2289         if (!(fs_info->workers && fs_info->delalloc_workers &&
2290               fs_info->submit_workers && fs_info->flush_workers &&
2291               fs_info->endio_workers && fs_info->endio_meta_workers &&
2292               fs_info->endio_meta_write_workers &&
2293               fs_info->endio_repair_workers &&
2294               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2295               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2296               fs_info->caching_workers && fs_info->readahead_workers &&
2297               fs_info->fixup_workers && fs_info->delayed_workers &&
2298               fs_info->extent_workers &&
2299               fs_info->qgroup_rescan_workers)) {
2300                 return -ENOMEM;
2301         }
2302
2303         return 0;
2304 }
2305
2306 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2307                             struct btrfs_fs_devices *fs_devices)
2308 {
2309         int ret;
2310         struct btrfs_root *log_tree_root;
2311         struct btrfs_super_block *disk_super = fs_info->super_copy;
2312         u64 bytenr = btrfs_super_log_root(disk_super);
2313         int level = btrfs_super_log_root_level(disk_super);
2314
2315         if (fs_devices->rw_devices == 0) {
2316                 btrfs_warn(fs_info, "log replay required on RO media");
2317                 return -EIO;
2318         }
2319
2320         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2321         if (!log_tree_root)
2322                 return -ENOMEM;
2323
2324         __setup_root(log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2325
2326         log_tree_root->node = read_tree_block(fs_info, bytenr,
2327                                               fs_info->generation + 1,
2328                                               level, NULL);
2329         if (IS_ERR(log_tree_root->node)) {
2330                 btrfs_warn(fs_info, "failed to read log tree");
2331                 ret = PTR_ERR(log_tree_root->node);
2332                 kfree(log_tree_root);
2333                 return ret;
2334         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2335                 btrfs_err(fs_info, "failed to read log tree");
2336                 free_extent_buffer(log_tree_root->node);
2337                 kfree(log_tree_root);
2338                 return -EIO;
2339         }
2340         /* returns with log_tree_root freed on success */
2341         ret = btrfs_recover_log_trees(log_tree_root);
2342         if (ret) {
2343                 btrfs_handle_fs_error(fs_info, ret,
2344                                       "Failed to recover log tree");
2345                 free_extent_buffer(log_tree_root->node);
2346                 kfree(log_tree_root);
2347                 return ret;
2348         }
2349
2350         if (sb_rdonly(fs_info->sb)) {
2351                 ret = btrfs_commit_super(fs_info);
2352                 if (ret)
2353                         return ret;
2354         }
2355
2356         return 0;
2357 }
2358
2359 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2360 {
2361         struct btrfs_root *tree_root = fs_info->tree_root;
2362         struct btrfs_root *root;
2363         struct btrfs_key location;
2364         int ret;
2365
2366         BUG_ON(!fs_info->tree_root);
2367
2368         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2369         location.type = BTRFS_ROOT_ITEM_KEY;
2370         location.offset = 0;
2371
2372         root = btrfs_read_tree_root(tree_root, &location);
2373         if (IS_ERR(root)) {
2374                 ret = PTR_ERR(root);
2375                 goto out;
2376         }
2377         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2378         fs_info->extent_root = root;
2379
2380         location.objectid = BTRFS_DEV_TREE_OBJECTID;
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->dev_root = root;
2388         btrfs_init_devices_late(fs_info);
2389
2390         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2391         root = btrfs_read_tree_root(tree_root, &location);
2392         if (IS_ERR(root)) {
2393                 ret = PTR_ERR(root);
2394                 goto out;
2395         }
2396         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2397         fs_info->csum_root = root;
2398
2399         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2400         root = btrfs_read_tree_root(tree_root, &location);
2401         if (!IS_ERR(root)) {
2402                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2403                 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2404                 fs_info->quota_root = root;
2405         }
2406
2407         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2408         root = btrfs_read_tree_root(tree_root, &location);
2409         if (IS_ERR(root)) {
2410                 ret = PTR_ERR(root);
2411                 if (ret != -ENOENT)
2412                         goto out;
2413         } else {
2414                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2415                 fs_info->uuid_root = root;
2416         }
2417
2418         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2419                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2420                 root = btrfs_read_tree_root(tree_root, &location);
2421                 if (IS_ERR(root)) {
2422                         ret = PTR_ERR(root);
2423                         goto out;
2424                 }
2425                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2426                 fs_info->free_space_root = root;
2427         }
2428
2429         return 0;
2430 out:
2431         btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2432                    location.objectid, ret);
2433         return ret;
2434 }
2435
2436 int open_ctree(struct super_block *sb,
2437                struct btrfs_fs_devices *fs_devices,
2438                char *options)
2439 {
2440         u32 sectorsize;
2441         u32 nodesize;
2442         u32 stripesize;
2443         u64 generation;
2444         u64 features;
2445         struct btrfs_key location;
2446         struct buffer_head *bh;
2447         struct btrfs_super_block *disk_super;
2448         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2449         struct btrfs_root *tree_root;
2450         struct btrfs_root *chunk_root;
2451         int ret;
2452         int err = -EINVAL;
2453         int num_backups_tried = 0;
2454         int backup_index = 0;
2455         int clear_free_space_tree = 0;
2456         int level;
2457
2458         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2459         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2460         if (!tree_root || !chunk_root) {
2461                 err = -ENOMEM;
2462                 goto fail;
2463         }
2464
2465         ret = init_srcu_struct(&fs_info->subvol_srcu);
2466         if (ret) {
2467                 err = ret;
2468                 goto fail;
2469         }
2470
2471         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2472         if (ret) {
2473                 err = ret;
2474                 goto fail_srcu;
2475         }
2476         fs_info->dirty_metadata_batch = PAGE_SIZE *
2477                                         (1 + ilog2(nr_cpu_ids));
2478
2479         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2480         if (ret) {
2481                 err = ret;
2482                 goto fail_dirty_metadata_bytes;
2483         }
2484
2485         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2486         if (ret) {
2487                 err = ret;
2488                 goto fail_delalloc_bytes;
2489         }
2490
2491         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2492         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2493         INIT_LIST_HEAD(&fs_info->trans_list);
2494         INIT_LIST_HEAD(&fs_info->dead_roots);
2495         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2496         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2497         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2498         INIT_LIST_HEAD(&fs_info->pending_raid_kobjs);
2499         spin_lock_init(&fs_info->pending_raid_kobjs_lock);
2500         spin_lock_init(&fs_info->delalloc_root_lock);
2501         spin_lock_init(&fs_info->trans_lock);
2502         spin_lock_init(&fs_info->fs_roots_radix_lock);
2503         spin_lock_init(&fs_info->delayed_iput_lock);
2504         spin_lock_init(&fs_info->defrag_inodes_lock);
2505         spin_lock_init(&fs_info->tree_mod_seq_lock);
2506         spin_lock_init(&fs_info->super_lock);
2507         spin_lock_init(&fs_info->qgroup_op_lock);
2508         spin_lock_init(&fs_info->buffer_lock);
2509         spin_lock_init(&fs_info->unused_bgs_lock);
2510         rwlock_init(&fs_info->tree_mod_log_lock);
2511         mutex_init(&fs_info->unused_bg_unpin_mutex);
2512         mutex_init(&fs_info->delete_unused_bgs_mutex);
2513         mutex_init(&fs_info->reloc_mutex);
2514         mutex_init(&fs_info->delalloc_root_mutex);
2515         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2516         seqlock_init(&fs_info->profiles_lock);
2517
2518         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2519         INIT_LIST_HEAD(&fs_info->space_info);
2520         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2521         INIT_LIST_HEAD(&fs_info->unused_bgs);
2522         btrfs_mapping_init(&fs_info->mapping_tree);
2523         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2524                              BTRFS_BLOCK_RSV_GLOBAL);
2525         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2526         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2527         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2528         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2529                              BTRFS_BLOCK_RSV_DELOPS);
2530         atomic_set(&fs_info->async_delalloc_pages, 0);
2531         atomic_set(&fs_info->defrag_running, 0);
2532         atomic_set(&fs_info->qgroup_op_seq, 0);
2533         atomic_set(&fs_info->reada_works_cnt, 0);
2534         atomic64_set(&fs_info->tree_mod_seq, 0);
2535         fs_info->sb = sb;
2536         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2537         fs_info->metadata_ratio = 0;
2538         fs_info->defrag_inodes = RB_ROOT;
2539         atomic64_set(&fs_info->free_chunk_space, 0);
2540         fs_info->tree_mod_log = RB_ROOT;
2541         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2542         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2543         /* readahead state */
2544         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2545         spin_lock_init(&fs_info->reada_lock);
2546         btrfs_init_ref_verify(fs_info);
2547
2548         fs_info->thread_pool_size = min_t(unsigned long,
2549                                           num_online_cpus() + 2, 8);
2550
2551         INIT_LIST_HEAD(&fs_info->ordered_roots);
2552         spin_lock_init(&fs_info->ordered_root_lock);
2553
2554         fs_info->btree_inode = new_inode(sb);
2555         if (!fs_info->btree_inode) {
2556                 err = -ENOMEM;
2557                 goto fail_bio_counter;
2558         }
2559         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2560
2561         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2562                                         GFP_KERNEL);
2563         if (!fs_info->delayed_root) {
2564                 err = -ENOMEM;
2565                 goto fail_iput;
2566         }
2567         btrfs_init_delayed_root(fs_info->delayed_root);
2568
2569         btrfs_init_scrub(fs_info);
2570 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2571         fs_info->check_integrity_print_mask = 0;
2572 #endif
2573         btrfs_init_balance(fs_info);
2574         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2575
2576         sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2577         sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2578
2579         btrfs_init_btree_inode(fs_info);
2580
2581         spin_lock_init(&fs_info->block_group_cache_lock);
2582         fs_info->block_group_cache_tree = RB_ROOT;
2583         fs_info->first_logical_byte = (u64)-1;
2584
2585         extent_io_tree_init(&fs_info->freed_extents[0], NULL);
2586         extent_io_tree_init(&fs_info->freed_extents[1], NULL);
2587         fs_info->pinned_extents = &fs_info->freed_extents[0];
2588         set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2589
2590         mutex_init(&fs_info->ordered_operations_mutex);
2591         mutex_init(&fs_info->tree_log_mutex);
2592         mutex_init(&fs_info->chunk_mutex);
2593         mutex_init(&fs_info->transaction_kthread_mutex);
2594         mutex_init(&fs_info->cleaner_mutex);
2595         mutex_init(&fs_info->volume_mutex);
2596         mutex_init(&fs_info->ro_block_group_mutex);
2597         init_rwsem(&fs_info->commit_root_sem);
2598         init_rwsem(&fs_info->cleanup_work_sem);
2599         init_rwsem(&fs_info->subvol_sem);
2600         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2601
2602         btrfs_init_dev_replace_locks(fs_info);
2603         btrfs_init_qgroup(fs_info);
2604
2605         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2606         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2607
2608         init_waitqueue_head(&fs_info->transaction_throttle);
2609         init_waitqueue_head(&fs_info->transaction_wait);
2610         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2611         init_waitqueue_head(&fs_info->async_submit_wait);
2612
2613         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2614
2615         /* Usable values until the real ones are cached from the superblock */
2616         fs_info->nodesize = 4096;
2617         fs_info->sectorsize = 4096;
2618         fs_info->stripesize = 4096;
2619
2620         ret = btrfs_alloc_stripe_hash_table(fs_info);
2621         if (ret) {
2622                 err = ret;
2623                 goto fail_alloc;
2624         }
2625
2626         __setup_root(tree_root, fs_info, BTRFS_ROOT_TREE_OBJECTID);
2627
2628         invalidate_bdev(fs_devices->latest_bdev);
2629
2630         /*
2631          * Read super block and check the signature bytes only
2632          */
2633         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2634         if (IS_ERR(bh)) {
2635                 err = PTR_ERR(bh);
2636                 goto fail_alloc;
2637         }
2638
2639         /*
2640          * We want to check superblock checksum, the type is stored inside.
2641          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2642          */
2643         if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2644                 btrfs_err(fs_info, "superblock checksum mismatch");
2645                 err = -EINVAL;
2646                 brelse(bh);
2647                 goto fail_alloc;
2648         }
2649
2650         /*
2651          * super_copy is zeroed at allocation time and we never touch the
2652          * following bytes up to INFO_SIZE, the checksum is calculated from
2653          * the whole block of INFO_SIZE
2654          */
2655         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2656         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2657                sizeof(*fs_info->super_for_commit));
2658         brelse(bh);
2659
2660         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2661
2662         ret = btrfs_check_super_valid(fs_info);
2663         if (ret) {
2664                 btrfs_err(fs_info, "superblock contains fatal errors");
2665                 err = -EINVAL;
2666                 goto fail_alloc;
2667         }
2668
2669         disk_super = fs_info->super_copy;
2670         if (!btrfs_super_root(disk_super))
2671                 goto fail_alloc;
2672
2673         /* check FS state, whether FS is broken. */
2674         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2675                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2676
2677         /*
2678          * run through our array of backup supers and setup
2679          * our ring pointer to the oldest one
2680          */
2681         generation = btrfs_super_generation(disk_super);
2682         find_oldest_super_backup(fs_info, generation);
2683
2684         /*
2685          * In the long term, we'll store the compression type in the super
2686          * block, and it'll be used for per file compression control.
2687          */
2688         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2689
2690         ret = btrfs_parse_options(fs_info, options, sb->s_flags);
2691         if (ret) {
2692                 err = ret;
2693                 goto fail_alloc;
2694         }
2695
2696         features = btrfs_super_incompat_flags(disk_super) &
2697                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2698         if (features) {
2699                 btrfs_err(fs_info,
2700                     "cannot mount because of unsupported optional features (%llx)",
2701                     features);
2702                 err = -EINVAL;
2703                 goto fail_alloc;
2704         }
2705
2706         features = btrfs_super_incompat_flags(disk_super);
2707         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2708         if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
2709                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2710         else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
2711                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
2712
2713         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2714                 btrfs_info(fs_info, "has skinny extents");
2715
2716         /*
2717          * flag our filesystem as having big metadata blocks if
2718          * they are bigger than the page size
2719          */
2720         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2721                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2722                         btrfs_info(fs_info,
2723                                 "flagging fs with big metadata feature");
2724                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2725         }
2726
2727         nodesize = btrfs_super_nodesize(disk_super);
2728         sectorsize = btrfs_super_sectorsize(disk_super);
2729         stripesize = sectorsize;
2730         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2731         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2732
2733         /* Cache block sizes */
2734         fs_info->nodesize = nodesize;
2735         fs_info->sectorsize = sectorsize;
2736         fs_info->stripesize = stripesize;
2737
2738         /*
2739          * mixed block groups end up with duplicate but slightly offset
2740          * extent buffers for the same range.  It leads to corruptions
2741          */
2742         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2743             (sectorsize != nodesize)) {
2744                 btrfs_err(fs_info,
2745 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2746                         nodesize, sectorsize);
2747                 goto fail_alloc;
2748         }
2749
2750         /*
2751          * Needn't use the lock because there is no other task which will
2752          * update the flag.
2753          */
2754         btrfs_set_super_incompat_flags(disk_super, features);
2755
2756         features = btrfs_super_compat_ro_flags(disk_super) &
2757                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2758         if (!sb_rdonly(sb) && features) {
2759                 btrfs_err(fs_info,
2760         "cannot mount read-write because of unsupported optional features (%llx)",
2761                        features);
2762                 err = -EINVAL;
2763                 goto fail_alloc;
2764         }
2765
2766         ret = btrfs_init_workqueues(fs_info, fs_devices);
2767         if (ret) {
2768                 err = ret;
2769                 goto fail_sb_buffer;
2770         }
2771
2772         sb->s_bdi->congested_fn = btrfs_congested_fn;
2773         sb->s_bdi->congested_data = fs_info;
2774         sb->s_bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
2775         sb->s_bdi->ra_pages = VM_MAX_READAHEAD * SZ_1K / PAGE_SIZE;
2776         sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
2777         sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
2778
2779         sb->s_blocksize = sectorsize;
2780         sb->s_blocksize_bits = blksize_bits(sectorsize);
2781         memcpy(&sb->s_uuid, fs_info->fsid, BTRFS_FSID_SIZE);
2782
2783         mutex_lock(&fs_info->chunk_mutex);
2784         ret = btrfs_read_sys_array(fs_info);
2785         mutex_unlock(&fs_info->chunk_mutex);
2786         if (ret) {
2787                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2788                 goto fail_sb_buffer;
2789         }
2790
2791         generation = btrfs_super_chunk_root_generation(disk_super);
2792         level = btrfs_super_chunk_root_level(disk_super);
2793
2794         __setup_root(chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2795
2796         chunk_root->node = read_tree_block(fs_info,
2797                                            btrfs_super_chunk_root(disk_super),
2798                                            generation, level, NULL);
2799         if (IS_ERR(chunk_root->node) ||
2800             !extent_buffer_uptodate(chunk_root->node)) {
2801                 btrfs_err(fs_info, "failed to read chunk root");
2802                 if (!IS_ERR(chunk_root->node))
2803                         free_extent_buffer(chunk_root->node);
2804                 chunk_root->node = NULL;
2805                 goto fail_tree_roots;
2806         }
2807         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2808         chunk_root->commit_root = btrfs_root_node(chunk_root);
2809
2810         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2811            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2812
2813         ret = btrfs_read_chunk_tree(fs_info);
2814         if (ret) {
2815                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2816                 goto fail_tree_roots;
2817         }
2818
2819         /*
2820          * Keep the devid that is marked to be the target device for the
2821          * device replace procedure
2822          */
2823         btrfs_free_extra_devids(fs_devices, 0);
2824
2825         if (!fs_devices->latest_bdev) {
2826                 btrfs_err(fs_info, "failed to read devices");
2827                 goto fail_tree_roots;
2828         }
2829
2830 retry_root_backup:
2831         generation = btrfs_super_generation(disk_super);
2832         level = btrfs_super_root_level(disk_super);
2833
2834         tree_root->node = read_tree_block(fs_info,
2835                                           btrfs_super_root(disk_super),
2836                                           generation, level, NULL);
2837         if (IS_ERR(tree_root->node) ||
2838             !extent_buffer_uptodate(tree_root->node)) {
2839                 btrfs_warn(fs_info, "failed to read tree root");
2840                 if (!IS_ERR(tree_root->node))
2841                         free_extent_buffer(tree_root->node);
2842                 tree_root->node = NULL;
2843                 goto recovery_tree_root;
2844         }
2845
2846         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2847         tree_root->commit_root = btrfs_root_node(tree_root);
2848         btrfs_set_root_refs(&tree_root->root_item, 1);
2849
2850         mutex_lock(&tree_root->objectid_mutex);
2851         ret = btrfs_find_highest_objectid(tree_root,
2852                                         &tree_root->highest_objectid);
2853         if (ret) {
2854                 mutex_unlock(&tree_root->objectid_mutex);
2855                 goto recovery_tree_root;
2856         }
2857
2858         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2859
2860         mutex_unlock(&tree_root->objectid_mutex);
2861
2862         ret = btrfs_read_roots(fs_info);
2863         if (ret)
2864                 goto recovery_tree_root;
2865
2866         fs_info->generation = generation;
2867         fs_info->last_trans_committed = generation;
2868
2869         ret = btrfs_recover_balance(fs_info);
2870         if (ret) {
2871                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2872                 goto fail_block_groups;
2873         }
2874
2875         ret = btrfs_init_dev_stats(fs_info);
2876         if (ret) {
2877                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2878                 goto fail_block_groups;
2879         }
2880
2881         ret = btrfs_init_dev_replace(fs_info);
2882         if (ret) {
2883                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2884                 goto fail_block_groups;
2885         }
2886
2887         btrfs_free_extra_devids(fs_devices, 1);
2888
2889         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2890         if (ret) {
2891                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2892                                 ret);
2893                 goto fail_block_groups;
2894         }
2895
2896         ret = btrfs_sysfs_add_device(fs_devices);
2897         if (ret) {
2898                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2899                                 ret);
2900                 goto fail_fsdev_sysfs;
2901         }
2902
2903         ret = btrfs_sysfs_add_mounted(fs_info);
2904         if (ret) {
2905                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2906                 goto fail_fsdev_sysfs;
2907         }
2908
2909         ret = btrfs_init_space_info(fs_info);
2910         if (ret) {
2911                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2912                 goto fail_sysfs;
2913         }
2914
2915         ret = btrfs_read_block_groups(fs_info);
2916         if (ret) {
2917                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2918                 goto fail_sysfs;
2919         }
2920
2921         if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
2922                 btrfs_warn(fs_info,
2923                 "writeable mount is not allowed due to too many missing devices");
2924                 goto fail_sysfs;
2925         }
2926
2927         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2928                                                "btrfs-cleaner");
2929         if (IS_ERR(fs_info->cleaner_kthread))
2930                 goto fail_sysfs;
2931
2932         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2933                                                    tree_root,
2934                                                    "btrfs-transaction");
2935         if (IS_ERR(fs_info->transaction_kthread))
2936                 goto fail_cleaner;
2937
2938         if (!btrfs_test_opt(fs_info, NOSSD) &&
2939             !fs_info->fs_devices->rotating) {
2940                 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
2941         }
2942
2943         /*
2944          * Mount does not set all options immediately, we can do it now and do
2945          * not have to wait for transaction commit
2946          */
2947         btrfs_apply_pending_changes(fs_info);
2948
2949 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2950         if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
2951                 ret = btrfsic_mount(fs_info, fs_devices,
2952                                     btrfs_test_opt(fs_info,
2953                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2954                                     1 : 0,
2955                                     fs_info->check_integrity_print_mask);
2956                 if (ret)
2957                         btrfs_warn(fs_info,
2958                                 "failed to initialize integrity check module: %d",
2959                                 ret);
2960         }
2961 #endif
2962         ret = btrfs_read_qgroup_config(fs_info);
2963         if (ret)
2964                 goto fail_trans_kthread;
2965
2966         if (btrfs_build_ref_tree(fs_info))
2967                 btrfs_err(fs_info, "couldn't build ref tree");
2968
2969         /* do not make disk changes in broken FS or nologreplay is given */
2970         if (btrfs_super_log_root(disk_super) != 0 &&
2971             !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
2972                 ret = btrfs_replay_log(fs_info, fs_devices);
2973                 if (ret) {
2974                         err = ret;
2975                         goto fail_qgroup;
2976                 }
2977         }
2978
2979         ret = btrfs_find_orphan_roots(fs_info);
2980         if (ret)
2981                 goto fail_qgroup;
2982
2983         if (!sb_rdonly(sb)) {
2984                 ret = btrfs_cleanup_fs_roots(fs_info);
2985                 if (ret)
2986                         goto fail_qgroup;
2987
2988                 mutex_lock(&fs_info->cleaner_mutex);
2989                 ret = btrfs_recover_relocation(tree_root);
2990                 mutex_unlock(&fs_info->cleaner_mutex);
2991                 if (ret < 0) {
2992                         btrfs_warn(fs_info, "failed to recover relocation: %d",
2993                                         ret);
2994                         err = -EINVAL;
2995                         goto fail_qgroup;
2996                 }
2997         }
2998
2999         location.objectid = BTRFS_FS_TREE_OBJECTID;
3000         location.type = BTRFS_ROOT_ITEM_KEY;
3001         location.offset = 0;
3002
3003         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3004         if (IS_ERR(fs_info->fs_root)) {
3005                 err = PTR_ERR(fs_info->fs_root);
3006                 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3007                 goto fail_qgroup;
3008         }
3009
3010         if (sb_rdonly(sb))
3011                 return 0;
3012
3013         if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3014             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3015                 clear_free_space_tree = 1;
3016         } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3017                    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3018                 btrfs_warn(fs_info, "free space tree is invalid");
3019                 clear_free_space_tree = 1;
3020         }
3021
3022         if (clear_free_space_tree) {
3023                 btrfs_info(fs_info, "clearing free space tree");
3024                 ret = btrfs_clear_free_space_tree(fs_info);
3025                 if (ret) {
3026                         btrfs_warn(fs_info,
3027                                    "failed to clear free space tree: %d", ret);
3028                         close_ctree(fs_info);
3029                         return ret;
3030                 }
3031         }
3032
3033         if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3034             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3035                 btrfs_info(fs_info, "creating free space tree");
3036                 ret = btrfs_create_free_space_tree(fs_info);
3037                 if (ret) {
3038                         btrfs_warn(fs_info,
3039                                 "failed to create free space tree: %d", ret);
3040                         close_ctree(fs_info);
3041                         return ret;
3042                 }
3043         }
3044
3045         down_read(&fs_info->cleanup_work_sem);
3046         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3047             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3048                 up_read(&fs_info->cleanup_work_sem);
3049                 close_ctree(fs_info);
3050                 return ret;
3051         }
3052         up_read(&fs_info->cleanup_work_sem);
3053
3054         ret = btrfs_resume_balance_async(fs_info);
3055         if (ret) {
3056                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3057                 close_ctree(fs_info);
3058                 return ret;
3059         }
3060
3061         ret = btrfs_resume_dev_replace_async(fs_info);
3062         if (ret) {
3063                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3064                 close_ctree(fs_info);
3065                 return ret;
3066         }
3067
3068         btrfs_qgroup_rescan_resume(fs_info);
3069
3070         if (!fs_info->uuid_root) {
3071                 btrfs_info(fs_info, "creating UUID tree");
3072                 ret = btrfs_create_uuid_tree(fs_info);
3073                 if (ret) {
3074                         btrfs_warn(fs_info,
3075                                 "failed to create the UUID tree: %d", ret);
3076                         close_ctree(fs_info);
3077                         return ret;
3078                 }
3079         } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3080                    fs_info->generation !=
3081                                 btrfs_super_uuid_tree_generation(disk_super)) {
3082                 btrfs_info(fs_info, "checking UUID tree");
3083                 ret = btrfs_check_uuid_tree(fs_info);
3084                 if (ret) {
3085                         btrfs_warn(fs_info,
3086                                 "failed to check the UUID tree: %d", ret);
3087                         close_ctree(fs_info);
3088                         return ret;
3089                 }
3090         } else {
3091                 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3092         }
3093         set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3094
3095         /*
3096          * backuproot only affect mount behavior, and if open_ctree succeeded,
3097          * no need to keep the flag
3098          */
3099         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3100
3101         return 0;
3102
3103 fail_qgroup:
3104         btrfs_free_qgroup_config(fs_info);
3105 fail_trans_kthread:
3106         kthread_stop(fs_info->transaction_kthread);
3107         btrfs_cleanup_transaction(fs_info);
3108         btrfs_free_fs_roots(fs_info);
3109 fail_cleaner:
3110         kthread_stop(fs_info->cleaner_kthread);
3111
3112         /*
3113          * make sure we're done with the btree inode before we stop our
3114          * kthreads
3115          */
3116         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3117
3118 fail_sysfs:
3119         btrfs_sysfs_remove_mounted(fs_info);
3120
3121 fail_fsdev_sysfs:
3122         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3123
3124 fail_block_groups:
3125         btrfs_put_block_group_cache(fs_info);
3126
3127 fail_tree_roots:
3128         free_root_pointers(fs_info, 1);
3129         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3130
3131 fail_sb_buffer:
3132         btrfs_stop_all_workers(fs_info);
3133         btrfs_free_block_groups(fs_info);
3134 fail_alloc:
3135 fail_iput:
3136         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3137
3138         iput(fs_info->btree_inode);
3139 fail_bio_counter:
3140         percpu_counter_destroy(&fs_info->bio_counter);
3141 fail_delalloc_bytes:
3142         percpu_counter_destroy(&fs_info->delalloc_bytes);
3143 fail_dirty_metadata_bytes:
3144         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3145 fail_srcu:
3146         cleanup_srcu_struct(&fs_info->subvol_srcu);
3147 fail:
3148         btrfs_free_stripe_hash_table(fs_info);
3149         btrfs_close_devices(fs_info->fs_devices);
3150         return err;
3151
3152 recovery_tree_root:
3153         if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
3154                 goto fail_tree_roots;
3155
3156         free_root_pointers(fs_info, 0);
3157
3158         /* don't use the log in recovery mode, it won't be valid */
3159         btrfs_set_super_log_root(disk_super, 0);
3160
3161         /* we can't trust the free space cache either */
3162         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3163
3164         ret = next_root_backup(fs_info, fs_info->super_copy,
3165                                &num_backups_tried, &backup_index);
3166         if (ret == -1)
3167                 goto fail_block_groups;
3168         goto retry_root_backup;
3169 }
3170 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3171
3172 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3173 {
3174         if (uptodate) {
3175                 set_buffer_uptodate(bh);
3176         } else {
3177                 struct btrfs_device *device = (struct btrfs_device *)
3178                         bh->b_private;
3179
3180                 btrfs_warn_rl_in_rcu(device->fs_info,
3181                                 "lost page write due to IO error on %s",
3182                                           rcu_str_deref(device->name));
3183                 /* note, we don't set_buffer_write_io_error because we have
3184                  * our own ways of dealing with the IO errors
3185                  */
3186                 clear_buffer_uptodate(bh);
3187                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3188         }
3189         unlock_buffer(bh);
3190         put_bh(bh);
3191 }
3192
3193 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3194                         struct buffer_head **bh_ret)
3195 {
3196         struct buffer_head *bh;
3197         struct btrfs_super_block *super;
3198         u64 bytenr;
3199
3200         bytenr = btrfs_sb_offset(copy_num);
3201         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3202                 return -EINVAL;
3203
3204         bh = __bread(bdev, bytenr / BTRFS_BDEV_BLOCKSIZE, BTRFS_SUPER_INFO_SIZE);
3205         /*
3206          * If we fail to read from the underlying devices, as of now
3207          * the best option we have is to mark it EIO.
3208          */
3209         if (!bh)
3210                 return -EIO;
3211
3212         super = (struct btrfs_super_block *)bh->b_data;
3213         if (btrfs_super_bytenr(super) != bytenr ||
3214                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3215                 brelse(bh);
3216                 return -EINVAL;
3217         }
3218
3219         *bh_ret = bh;
3220         return 0;
3221 }
3222
3223
3224 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3225 {
3226         struct buffer_head *bh;
3227         struct buffer_head *latest = NULL;
3228         struct btrfs_super_block *super;
3229         int i;
3230         u64 transid = 0;
3231         int ret = -EINVAL;
3232
3233         /* we would like to check all the supers, but that would make
3234          * a btrfs mount succeed after a mkfs from a different FS.
3235          * So, we need to add a special mount option to scan for
3236          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3237          */
3238         for (i = 0; i < 1; i++) {
3239                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3240                 if (ret)
3241                         continue;
3242
3243                 super = (struct btrfs_super_block *)bh->b_data;
3244
3245                 if (!latest || btrfs_super_generation(super) > transid) {
3246                         brelse(latest);
3247                         latest = bh;
3248                         transid = btrfs_super_generation(super);
3249                 } else {
3250                         brelse(bh);
3251                 }
3252         }
3253
3254         if (!latest)
3255                 return ERR_PTR(ret);
3256
3257         return latest;
3258 }
3259
3260 /*
3261  * Write superblock @sb to the @device. Do not wait for completion, all the
3262  * buffer heads we write are pinned.
3263  *
3264  * Write @max_mirrors copies of the superblock, where 0 means default that fit
3265  * the expected device size at commit time. Note that max_mirrors must be
3266  * same for write and wait phases.
3267  *
3268  * Return number of errors when buffer head is not found or submission fails.
3269  */
3270 static int write_dev_supers(struct btrfs_device *device,
3271                             struct btrfs_super_block *sb, int max_mirrors)
3272 {
3273         struct buffer_head *bh;
3274         int i;
3275         int ret;
3276         int errors = 0;
3277         u32 crc;
3278         u64 bytenr;
3279         int op_flags;
3280
3281         if (max_mirrors == 0)
3282                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3283
3284         for (i = 0; i < max_mirrors; i++) {
3285                 bytenr = btrfs_sb_offset(i);
3286                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3287                     device->commit_total_bytes)
3288                         break;
3289
3290                 btrfs_set_super_bytenr(sb, bytenr);
3291
3292                 crc = ~(u32)0;
3293                 crc = btrfs_csum_data((const char *)sb + BTRFS_CSUM_SIZE, crc,
3294                                       BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
3295                 btrfs_csum_final(crc, sb->csum);
3296
3297                 /* One reference for us, and we leave it for the caller */
3298                 bh = __getblk(device->bdev, bytenr / BTRFS_BDEV_BLOCKSIZE,
3299                               BTRFS_SUPER_INFO_SIZE);
3300                 if (!bh) {
3301                         btrfs_err(device->fs_info,
3302                             "couldn't get super buffer head for bytenr %llu",
3303                             bytenr);
3304                         errors++;
3305                         continue;
3306                 }
3307
3308                 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3309
3310                 /* one reference for submit_bh */
3311                 get_bh(bh);
3312
3313                 set_buffer_uptodate(bh);
3314                 lock_buffer(bh);
3315                 bh->b_end_io = btrfs_end_buffer_write_sync;
3316                 bh->b_private = device;
3317
3318                 /*
3319                  * we fua the first super.  The others we allow
3320                  * to go down lazy.
3321                  */
3322                 op_flags = REQ_SYNC | REQ_META | REQ_PRIO;
3323                 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3324                         op_flags |= REQ_FUA;
3325                 ret = btrfsic_submit_bh(REQ_OP_WRITE, op_flags, bh);
3326                 if (ret)
3327                         errors++;
3328         }
3329         return errors < i ? 0 : -1;
3330 }
3331
3332 /*
3333  * Wait for write completion of superblocks done by write_dev_supers,
3334  * @max_mirrors same for write and wait phases.
3335  *
3336  * Return number of errors when buffer head is not found or not marked up to
3337  * date.
3338  */
3339 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3340 {
3341         struct buffer_head *bh;
3342         int i;
3343         int errors = 0;
3344         bool primary_failed = false;
3345         u64 bytenr;
3346
3347         if (max_mirrors == 0)
3348                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3349
3350         for (i = 0; i < max_mirrors; i++) {
3351                 bytenr = btrfs_sb_offset(i);
3352                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3353                     device->commit_total_bytes)
3354                         break;
3355
3356                 bh = __find_get_block(device->bdev,
3357                                       bytenr / BTRFS_BDEV_BLOCKSIZE,
3358                                       BTRFS_SUPER_INFO_SIZE);
3359                 if (!bh) {
3360                         errors++;
3361                         if (i == 0)
3362                                 primary_failed = true;
3363                         continue;
3364                 }
3365                 wait_on_buffer(bh);
3366                 if (!buffer_uptodate(bh)) {
3367                         errors++;
3368                         if (i == 0)
3369                                 primary_failed = true;
3370                 }
3371
3372                 /* drop our reference */
3373                 brelse(bh);
3374
3375                 /* drop the reference from the writing run */
3376                 brelse(bh);
3377         }
3378
3379         /* log error, force error return */
3380         if (primary_failed) {
3381                 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3382                           device->devid);
3383                 return -1;
3384         }
3385
3386         return errors < i ? 0 : -1;
3387 }
3388
3389 /*
3390  * endio for the write_dev_flush, this will wake anyone waiting
3391  * for the barrier when it is done
3392  */
3393 static void btrfs_end_empty_barrier(struct bio *bio)
3394 {
3395         complete(bio->bi_private);
3396 }
3397
3398 /*
3399  * Submit a flush request to the device if it supports it. Error handling is
3400  * done in the waiting counterpart.
3401  */
3402 static void write_dev_flush(struct btrfs_device *device)
3403 {
3404         struct request_queue *q = bdev_get_queue(device->bdev);
3405         struct bio *bio = device->flush_bio;
3406
3407         if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3408                 return;
3409
3410         bio_reset(bio);
3411         bio->bi_end_io = btrfs_end_empty_barrier;
3412         bio_set_dev(bio, device->bdev);
3413         bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3414         init_completion(&device->flush_wait);
3415         bio->bi_private = &device->flush_wait;
3416
3417         btrfsic_submit_bio(bio);
3418         set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3419 }
3420
3421 /*
3422  * If the flush bio has been submitted by write_dev_flush, wait for it.
3423  */
3424 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3425 {
3426         struct bio *bio = device->flush_bio;
3427
3428         if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3429                 return BLK_STS_OK;
3430
3431         clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3432         wait_for_completion_io(&device->flush_wait);
3433
3434         return bio->bi_status;
3435 }
3436
3437 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3438 {
3439         if (!btrfs_check_rw_degradable(fs_info, NULL))
3440                 return -EIO;
3441         return 0;
3442 }
3443
3444 /*
3445  * send an empty flush down to each device in parallel,
3446  * then wait for them
3447  */
3448 static int barrier_all_devices(struct btrfs_fs_info *info)
3449 {
3450         struct list_head *head;
3451         struct btrfs_device *dev;
3452         int errors_wait = 0;
3453         blk_status_t ret;
3454
3455         lockdep_assert_held(&info->fs_devices->device_list_mutex);
3456         /* send down all the barriers */
3457         head = &info->fs_devices->devices;
3458         list_for_each_entry(dev, head, dev_list) {
3459                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3460                         continue;
3461                 if (!dev->bdev)
3462                         continue;
3463                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3464                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3465                         continue;
3466
3467                 write_dev_flush(dev);
3468                 dev->last_flush_error = BLK_STS_OK;
3469         }
3470
3471         /* wait for all the barriers */
3472         list_for_each_entry(dev, head, dev_list) {
3473                 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3474                         continue;
3475                 if (!dev->bdev) {
3476                         errors_wait++;
3477                         continue;
3478                 }
3479                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3480                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3481                         continue;
3482
3483                 ret = wait_dev_flush(dev);
3484                 if (ret) {
3485                         dev->last_flush_error = ret;
3486                         btrfs_dev_stat_inc_and_print(dev,
3487                                         BTRFS_DEV_STAT_FLUSH_ERRS);
3488                         errors_wait++;
3489                 }
3490         }
3491
3492         if (errors_wait) {
3493                 /*
3494                  * At some point we need the status of all disks
3495                  * to arrive at the volume status. So error checking
3496                  * is being pushed to a separate loop.
3497                  */
3498                 return check_barrier_error(info);
3499         }
3500         return 0;
3501 }
3502
3503 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3504 {
3505         int raid_type;
3506         int min_tolerated = INT_MAX;
3507
3508         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3509             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3510                 min_tolerated = min(min_tolerated,
3511                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3512                                     tolerated_failures);
3513
3514         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3515                 if (raid_type == BTRFS_RAID_SINGLE)
3516                         continue;
3517                 if (!(flags & btrfs_raid_group[raid_type]))
3518                         continue;
3519                 min_tolerated = min(min_tolerated,
3520                                     btrfs_raid_array[raid_type].
3521                                     tolerated_failures);
3522         }
3523
3524         if (min_tolerated == INT_MAX) {
3525                 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3526                 min_tolerated = 0;
3527         }
3528
3529         return min_tolerated;
3530 }
3531
3532 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3533 {
3534         struct list_head *head;
3535         struct btrfs_device *dev;
3536         struct btrfs_super_block *sb;
3537         struct btrfs_dev_item *dev_item;
3538         int ret;
3539         int do_barriers;
3540         int max_errors;
3541         int total_errors = 0;
3542         u64 flags;
3543
3544         do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3545
3546         /*
3547          * max_mirrors == 0 indicates we're from commit_transaction,
3548          * not from fsync where the tree roots in fs_info have not
3549          * been consistent on disk.
3550          */
3551         if (max_mirrors == 0)
3552                 backup_super_roots(fs_info);
3553
3554         sb = fs_info->super_for_commit;
3555         dev_item = &sb->dev_item;
3556
3557         mutex_lock(&fs_info->fs_devices->device_list_mutex);
3558         head = &fs_info->fs_devices->devices;
3559         max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3560
3561         if (do_barriers) {
3562                 ret = barrier_all_devices(fs_info);
3563                 if (ret) {
3564                         mutex_unlock(
3565                                 &fs_info->fs_devices->device_list_mutex);
3566                         btrfs_handle_fs_error(fs_info, ret,
3567                                               "errors while submitting device barriers.");
3568                         return ret;
3569                 }
3570         }
3571
3572         list_for_each_entry(dev, head, dev_list) {
3573                 if (!dev->bdev) {
3574                         total_errors++;
3575                         continue;
3576                 }
3577                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3578                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3579                         continue;
3580
3581                 btrfs_set_stack_device_generation(dev_item, 0);
3582                 btrfs_set_stack_device_type(dev_item, dev->type);
3583                 btrfs_set_stack_device_id(dev_item, dev->devid);
3584                 btrfs_set_stack_device_total_bytes(dev_item,
3585                                                    dev->commit_total_bytes);
3586                 btrfs_set_stack_device_bytes_used(dev_item,
3587                                                   dev->commit_bytes_used);
3588                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3589                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3590                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3591                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3592                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_FSID_SIZE);
3593
3594                 flags = btrfs_super_flags(sb);
3595                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3596
3597                 ret = write_dev_supers(dev, sb, max_mirrors);
3598                 if (ret)
3599                         total_errors++;
3600         }
3601         if (total_errors > max_errors) {
3602                 btrfs_err(fs_info, "%d errors while writing supers",
3603                           total_errors);
3604                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3605
3606                 /* FUA is masked off if unsupported and can't be the reason */
3607                 btrfs_handle_fs_error(fs_info, -EIO,
3608                                       "%d errors while writing supers",
3609                                       total_errors);
3610                 return -EIO;
3611         }
3612
3613         total_errors = 0;
3614         list_for_each_entry(dev, head, dev_list) {
3615                 if (!dev->bdev)
3616                         continue;
3617                 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3618                     !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3619                         continue;
3620
3621                 ret = wait_dev_supers(dev, max_mirrors);
3622                 if (ret)
3623                         total_errors++;
3624         }
3625         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3626         if (total_errors > max_errors) {
3627                 btrfs_handle_fs_error(fs_info, -EIO,
3628                                       "%d errors while writing supers",
3629                                       total_errors);
3630                 return -EIO;
3631         }
3632         return 0;
3633 }
3634
3635 /* Drop a fs root from the radix tree and free it. */
3636 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3637                                   struct btrfs_root *root)
3638 {
3639         spin_lock(&fs_info->fs_roots_radix_lock);
3640         radix_tree_delete(&fs_info->fs_roots_radix,
3641                           (unsigned long)root->root_key.objectid);
3642         spin_unlock(&fs_info->fs_roots_radix_lock);
3643
3644         if (btrfs_root_refs(&root->root_item) == 0)
3645                 synchronize_srcu(&fs_info->subvol_srcu);
3646
3647         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3648                 btrfs_free_log(NULL, root);
3649                 if (root->reloc_root) {
3650                         free_extent_buffer(root->reloc_root->node);
3651                         free_extent_buffer(root->reloc_root->commit_root);
3652                         btrfs_put_fs_root(root->reloc_root);
3653                         root->reloc_root = NULL;
3654                 }
3655         }
3656
3657         if (root->free_ino_pinned)
3658                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3659         if (root->free_ino_ctl)
3660                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3661         free_fs_root(root);
3662 }
3663
3664 static void free_fs_root(struct btrfs_root *root)
3665 {
3666         iput(root->ino_cache_inode);
3667         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3668         btrfs_free_block_rsv(root->fs_info, root->orphan_block_rsv);
3669         root->orphan_block_rsv = NULL;
3670         if (root->anon_dev)
3671                 free_anon_bdev(root->anon_dev);
3672         if (root->subv_writers)
3673                 btrfs_free_subvolume_writers(root->subv_writers);
3674         free_extent_buffer(root->node);
3675         free_extent_buffer(root->commit_root);
3676         kfree(root->free_ino_ctl);
3677         kfree(root->free_ino_pinned);
3678         kfree(root->name);
3679         btrfs_put_fs_root(root);
3680 }
3681
3682 void btrfs_free_fs_root(struct btrfs_root *root)
3683 {
3684         free_fs_root(root);
3685 }
3686
3687 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3688 {
3689         u64 root_objectid = 0;
3690         struct btrfs_root *gang[8];
3691         int i = 0;
3692         int err = 0;
3693         unsigned int ret = 0;
3694         int index;
3695
3696         while (1) {
3697                 index = srcu_read_lock(&fs_info->subvol_srcu);
3698                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3699                                              (void **)gang, root_objectid,
3700                                              ARRAY_SIZE(gang));
3701                 if (!ret) {
3702                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3703                         break;
3704                 }
3705                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3706
3707                 for (i = 0; i < ret; i++) {
3708                         /* Avoid to grab roots in dead_roots */
3709                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3710                                 gang[i] = NULL;
3711                                 continue;
3712                         }
3713                         /* grab all the search result for later use */
3714                         gang[i] = btrfs_grab_fs_root(gang[i]);
3715                 }
3716                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3717
3718                 for (i = 0; i < ret; i++) {
3719                         if (!gang[i])
3720                                 continue;
3721                         root_objectid = gang[i]->root_key.objectid;
3722                         err = btrfs_orphan_cleanup(gang[i]);
3723                         if (err)
3724                                 break;
3725                         btrfs_put_fs_root(gang[i]);
3726                 }
3727                 root_objectid++;
3728         }
3729
3730         /* release the uncleaned roots due to error */
3731         for (; i < ret; i++) {
3732                 if (gang[i])
3733                         btrfs_put_fs_root(gang[i]);
3734         }
3735         return err;
3736 }
3737
3738 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
3739 {
3740         struct btrfs_root *root = fs_info->tree_root;
3741         struct btrfs_trans_handle *trans;
3742
3743         mutex_lock(&fs_info->cleaner_mutex);
3744         btrfs_run_delayed_iputs(fs_info);
3745         mutex_unlock(&fs_info->cleaner_mutex);
3746         wake_up_process(fs_info->cleaner_kthread);
3747
3748         /* wait until ongoing cleanup work done */
3749         down_write(&fs_info->cleanup_work_sem);
3750         up_write(&fs_info->cleanup_work_sem);
3751
3752         trans = btrfs_join_transaction(root);
3753         if (IS_ERR(trans))
3754                 return PTR_ERR(trans);
3755         return btrfs_commit_transaction(trans);
3756 }
3757
3758 void close_ctree(struct btrfs_fs_info *fs_info)
3759 {
3760         struct btrfs_root *root = fs_info->tree_root;
3761         int ret;
3762
3763         set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3764
3765         /* wait for the qgroup rescan worker to stop */
3766         btrfs_qgroup_wait_for_completion(fs_info, false);
3767
3768         /* wait for the uuid_scan task to finish */
3769         down(&fs_info->uuid_tree_rescan_sem);
3770         /* avoid complains from lockdep et al., set sem back to initial state */
3771         up(&fs_info->uuid_tree_rescan_sem);
3772
3773         /* pause restriper - we want to resume on mount */
3774         btrfs_pause_balance(fs_info);
3775
3776         btrfs_dev_replace_suspend_for_unmount(fs_info);
3777
3778         btrfs_scrub_cancel(fs_info);
3779
3780         /* wait for any defraggers to finish */
3781         wait_event(fs_info->transaction_wait,
3782                    (atomic_read(&fs_info->defrag_running) == 0));
3783
3784         /* clear out the rbtree of defraggable inodes */
3785         btrfs_cleanup_defrag_inodes(fs_info);
3786
3787         cancel_work_sync(&fs_info->async_reclaim_work);
3788
3789         if (!sb_rdonly(fs_info->sb)) {
3790                 /*
3791                  * If the cleaner thread is stopped and there are
3792                  * block groups queued for removal, the deletion will be
3793                  * skipped when we quit the cleaner thread.
3794                  */
3795                 btrfs_delete_unused_bgs(fs_info);
3796
3797                 re