ff2362dca91aa92ef77c719f911283456a04d1ad
[sfrench/cifs-2.6.git] / fs / btrfs / disk-io.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "hash.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
49 #include "raid56.h"
50 #include "sysfs.h"
51 #include "qgroup.h"
52 #include "compression.h"
53
54 #ifdef CONFIG_X86
55 #include <asm/cpufeature.h>
56 #endif
57
58 #define BTRFS_SUPER_FLAG_SUPP   (BTRFS_HEADER_FLAG_WRITTEN |\
59                                  BTRFS_HEADER_FLAG_RELOC |\
60                                  BTRFS_SUPER_FLAG_ERROR |\
61                                  BTRFS_SUPER_FLAG_SEEDING |\
62                                  BTRFS_SUPER_FLAG_METADUMP)
63
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68                                     int read_only);
69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71                                       struct btrfs_root *root);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74                                         struct extent_io_tree *dirty_pages,
75                                         int mark);
76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77                                        struct extent_io_tree *pinned_extents);
78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
79 static void btrfs_error_commit_super(struct btrfs_root *root);
80
81 /*
82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
83  * is complete.  This is used during reads to verify checksums, and it is used
84  * by writes to insert metadata for new file extents after IO is complete.
85  */
86 struct btrfs_end_io_wq {
87         struct bio *bio;
88         bio_end_io_t *end_io;
89         void *private;
90         struct btrfs_fs_info *info;
91         int error;
92         enum btrfs_wq_endio_type metadata;
93         struct list_head list;
94         struct btrfs_work work;
95 };
96
97 static struct kmem_cache *btrfs_end_io_wq_cache;
98
99 int __init btrfs_end_io_wq_init(void)
100 {
101         btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102                                         sizeof(struct btrfs_end_io_wq),
103                                         0,
104                                         SLAB_MEM_SPREAD,
105                                         NULL);
106         if (!btrfs_end_io_wq_cache)
107                 return -ENOMEM;
108         return 0;
109 }
110
111 void btrfs_end_io_wq_exit(void)
112 {
113         kmem_cache_destroy(btrfs_end_io_wq_cache);
114 }
115
116 /*
117  * async submit bios are used to offload expensive checksumming
118  * onto the worker threads.  They checksum file and metadata bios
119  * just before they are sent down the IO stack.
120  */
121 struct async_submit_bio {
122         struct inode *inode;
123         struct bio *bio;
124         struct list_head list;
125         extent_submit_bio_hook_t *submit_bio_start;
126         extent_submit_bio_hook_t *submit_bio_done;
127         int rw;
128         int mirror_num;
129         unsigned long bio_flags;
130         /*
131          * bio_offset is optional, can be used if the pages in the bio
132          * can't tell us where in the file the bio should go
133          */
134         u64 bio_offset;
135         struct btrfs_work work;
136         int error;
137 };
138
139 /*
140  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
141  * eb, the lockdep key is determined by the btrfs_root it belongs to and
142  * the level the eb occupies in the tree.
143  *
144  * Different roots are used for different purposes and may nest inside each
145  * other and they require separate keysets.  As lockdep keys should be
146  * static, assign keysets according to the purpose of the root as indicated
147  * by btrfs_root->objectid.  This ensures that all special purpose roots
148  * have separate keysets.
149  *
150  * Lock-nesting across peer nodes is always done with the immediate parent
151  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
152  * subclass to avoid triggering lockdep warning in such cases.
153  *
154  * The key is set by the readpage_end_io_hook after the buffer has passed
155  * csum validation but before the pages are unlocked.  It is also set by
156  * btrfs_init_new_buffer on freshly allocated blocks.
157  *
158  * We also add a check to make sure the highest level of the tree is the
159  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
160  * needs update as well.
161  */
162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
163 # if BTRFS_MAX_LEVEL != 8
164 #  error
165 # endif
166
167 static struct btrfs_lockdep_keyset {
168         u64                     id;             /* root objectid */
169         const char              *name_stem;     /* lock name stem */
170         char                    names[BTRFS_MAX_LEVEL + 1][20];
171         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
172 } btrfs_lockdep_keysets[] = {
173         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
174         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
175         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
176         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
177         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
178         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
179         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
180         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
181         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
182         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
183         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
184         { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185         { .id = 0,                              .name_stem = "tree"     },
186 };
187
188 void __init btrfs_init_lockdep(void)
189 {
190         int i, j;
191
192         /* initialize lockdep class names */
193         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195
196                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197                         snprintf(ks->names[j], sizeof(ks->names[j]),
198                                  "btrfs-%s-%02d", ks->name_stem, j);
199         }
200 }
201
202 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
203                                     int level)
204 {
205         struct btrfs_lockdep_keyset *ks;
206
207         BUG_ON(level >= ARRAY_SIZE(ks->keys));
208
209         /* find the matching keyset, id 0 is the default entry */
210         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211                 if (ks->id == objectid)
212                         break;
213
214         lockdep_set_class_and_name(&eb->lock,
215                                    &ks->keys[level], ks->names[level]);
216 }
217
218 #endif
219
220 /*
221  * extents on the btree inode are pretty simple, there's one extent
222  * that covers the entire device
223  */
224 static struct extent_map *btree_get_extent(struct inode *inode,
225                 struct page *page, size_t pg_offset, u64 start, u64 len,
226                 int create)
227 {
228         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229         struct extent_map *em;
230         int ret;
231
232         read_lock(&em_tree->lock);
233         em = lookup_extent_mapping(em_tree, start, len);
234         if (em) {
235                 em->bdev =
236                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
237                 read_unlock(&em_tree->lock);
238                 goto out;
239         }
240         read_unlock(&em_tree->lock);
241
242         em = alloc_extent_map();
243         if (!em) {
244                 em = ERR_PTR(-ENOMEM);
245                 goto out;
246         }
247         em->start = 0;
248         em->len = (u64)-1;
249         em->block_len = (u64)-1;
250         em->block_start = 0;
251         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
252
253         write_lock(&em_tree->lock);
254         ret = add_extent_mapping(em_tree, em, 0);
255         if (ret == -EEXIST) {
256                 free_extent_map(em);
257                 em = lookup_extent_mapping(em_tree, start, len);
258                 if (!em)
259                         em = ERR_PTR(-EIO);
260         } else if (ret) {
261                 free_extent_map(em);
262                 em = ERR_PTR(ret);
263         }
264         write_unlock(&em_tree->lock);
265
266 out:
267         return em;
268 }
269
270 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
271 {
272         return btrfs_crc32c(seed, data, len);
273 }
274
275 void btrfs_csum_final(u32 crc, char *result)
276 {
277         put_unaligned_le32(~crc, result);
278 }
279
280 /*
281  * compute the csum for a btree block, and either verify it or write it
282  * into the csum field of the block.
283  */
284 static int csum_tree_block(struct btrfs_fs_info *fs_info,
285                            struct extent_buffer *buf,
286                            int verify)
287 {
288         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
289         char *result = NULL;
290         unsigned long len;
291         unsigned long cur_len;
292         unsigned long offset = BTRFS_CSUM_SIZE;
293         char *kaddr;
294         unsigned long map_start;
295         unsigned long map_len;
296         int err;
297         u32 crc = ~(u32)0;
298         unsigned long inline_result;
299
300         len = buf->len - offset;
301         while (len > 0) {
302                 err = map_private_extent_buffer(buf, offset, 32,
303                                         &kaddr, &map_start, &map_len);
304                 if (err)
305                         return err;
306                 cur_len = min(len, map_len - (offset - map_start));
307                 crc = btrfs_csum_data(kaddr + offset - map_start,
308                                       crc, cur_len);
309                 len -= cur_len;
310                 offset += cur_len;
311         }
312         if (csum_size > sizeof(inline_result)) {
313                 result = kzalloc(csum_size, GFP_NOFS);
314                 if (!result)
315                         return -ENOMEM;
316         } else {
317                 result = (char *)&inline_result;
318         }
319
320         btrfs_csum_final(crc, result);
321
322         if (verify) {
323                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
324                         u32 val;
325                         u32 found = 0;
326                         memcpy(&found, result, csum_size);
327
328                         read_extent_buffer(buf, &val, 0, csum_size);
329                         btrfs_warn_rl(fs_info,
330                                 "%s checksum verify failed on %llu wanted %X found %X "
331                                 "level %d",
332                                 fs_info->sb->s_id, buf->start,
333                                 val, found, btrfs_header_level(buf));
334                         if (result != (char *)&inline_result)
335                                 kfree(result);
336                         return -EUCLEAN;
337                 }
338         } else {
339                 write_extent_buffer(buf, result, 0, csum_size);
340         }
341         if (result != (char *)&inline_result)
342                 kfree(result);
343         return 0;
344 }
345
346 /*
347  * we can't consider a given block up to date unless the transid of the
348  * block matches the transid in the parent node's pointer.  This is how we
349  * detect blocks that either didn't get written at all or got written
350  * in the wrong place.
351  */
352 static int verify_parent_transid(struct extent_io_tree *io_tree,
353                                  struct extent_buffer *eb, u64 parent_transid,
354                                  int atomic)
355 {
356         struct extent_state *cached_state = NULL;
357         int ret;
358         bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359
360         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
361                 return 0;
362
363         if (atomic)
364                 return -EAGAIN;
365
366         if (need_lock) {
367                 btrfs_tree_read_lock(eb);
368                 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
369         }
370
371         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372                          &cached_state);
373         if (extent_buffer_uptodate(eb) &&
374             btrfs_header_generation(eb) == parent_transid) {
375                 ret = 0;
376                 goto out;
377         }
378         btrfs_err_rl(eb->fs_info,
379                 "parent transid verify failed on %llu wanted %llu found %llu",
380                         eb->start,
381                         parent_transid, btrfs_header_generation(eb));
382         ret = 1;
383
384         /*
385          * Things reading via commit roots that don't have normal protection,
386          * like send, can have a really old block in cache that may point at a
387          * block that has been freed and re-allocated.  So don't clear uptodate
388          * if we find an eb that is under IO (dirty/writeback) because we could
389          * end up reading in the stale data and then writing it back out and
390          * making everybody very sad.
391          */
392         if (!extent_buffer_under_io(eb))
393                 clear_extent_buffer_uptodate(eb);
394 out:
395         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
396                              &cached_state, GFP_NOFS);
397         if (need_lock)
398                 btrfs_tree_read_unlock_blocking(eb);
399         return ret;
400 }
401
402 /*
403  * Return 0 if the superblock checksum type matches the checksum value of that
404  * algorithm. Pass the raw disk superblock data.
405  */
406 static int btrfs_check_super_csum(char *raw_disk_sb)
407 {
408         struct btrfs_super_block *disk_sb =
409                 (struct btrfs_super_block *)raw_disk_sb;
410         u16 csum_type = btrfs_super_csum_type(disk_sb);
411         int ret = 0;
412
413         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414                 u32 crc = ~(u32)0;
415                 const int csum_size = sizeof(crc);
416                 char result[csum_size];
417
418                 /*
419                  * The super_block structure does not span the whole
420                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421                  * is filled with zeros and is included in the checksum.
422                  */
423                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425                 btrfs_csum_final(crc, result);
426
427                 if (memcmp(raw_disk_sb, result, csum_size))
428                         ret = 1;
429         }
430
431         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
433                                 csum_type);
434                 ret = 1;
435         }
436
437         return ret;
438 }
439
440 /*
441  * helper to read a given tree block, doing retries as required when
442  * the checksums don't match and we have alternate mirrors to try.
443  */
444 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
445                                           struct extent_buffer *eb,
446                                           u64 start, u64 parent_transid)
447 {
448         struct extent_io_tree *io_tree;
449         int failed = 0;
450         int ret;
451         int num_copies = 0;
452         int mirror_num = 0;
453         int failed_mirror = 0;
454
455         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
457         while (1) {
458                 ret = read_extent_buffer_pages(io_tree, eb, start,
459                                                WAIT_COMPLETE,
460                                                btree_get_extent, mirror_num);
461                 if (!ret) {
462                         if (!verify_parent_transid(io_tree, eb,
463                                                    parent_transid, 0))
464                                 break;
465                         else
466                                 ret = -EIO;
467                 }
468
469                 /*
470                  * This buffer's crc is fine, but its contents are corrupted, so
471                  * there is no reason to read the other copies, they won't be
472                  * any less wrong.
473                  */
474                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
475                         break;
476
477                 num_copies = btrfs_num_copies(root->fs_info,
478                                               eb->start, eb->len);
479                 if (num_copies == 1)
480                         break;
481
482                 if (!failed_mirror) {
483                         failed = 1;
484                         failed_mirror = eb->read_mirror;
485                 }
486
487                 mirror_num++;
488                 if (mirror_num == failed_mirror)
489                         mirror_num++;
490
491                 if (mirror_num > num_copies)
492                         break;
493         }
494
495         if (failed && !ret && failed_mirror)
496                 repair_eb_io_failure(root, eb, failed_mirror);
497
498         return ret;
499 }
500
501 /*
502  * checksum a dirty tree block before IO.  This has extra checks to make sure
503  * we only fill in the checksum field in the first page of a multi-page block
504  */
505
506 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507 {
508         u64 start = page_offset(page);
509         u64 found_start;
510         struct extent_buffer *eb;
511
512         eb = (struct extent_buffer *)page->private;
513         if (page != eb->pages[0])
514                 return 0;
515
516         found_start = btrfs_header_bytenr(eb);
517         /*
518          * Please do not consolidate these warnings into a single if.
519          * It is useful to know what went wrong.
520          */
521         if (WARN_ON(found_start != start))
522                 return -EUCLEAN;
523         if (WARN_ON(!PageUptodate(page)))
524                 return -EUCLEAN;
525
526         ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
527                         btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528
529         return csum_tree_block(fs_info, eb, 0);
530 }
531
532 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
533                                  struct extent_buffer *eb)
534 {
535         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
536         u8 fsid[BTRFS_UUID_SIZE];
537         int ret = 1;
538
539         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540         while (fs_devices) {
541                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
542                         ret = 0;
543                         break;
544                 }
545                 fs_devices = fs_devices->seed;
546         }
547         return ret;
548 }
549
550 #define CORRUPT(reason, eb, root, slot)                         \
551         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
552                    "root=%llu, slot=%d", reason,                        \
553                btrfs_header_bytenr(eb), root->objectid, slot)
554
555 static noinline int check_leaf(struct btrfs_root *root,
556                                struct extent_buffer *leaf)
557 {
558         struct btrfs_key key;
559         struct btrfs_key leaf_key;
560         u32 nritems = btrfs_header_nritems(leaf);
561         int slot;
562
563         if (nritems == 0)
564                 return 0;
565
566         /* Check the 0 item */
567         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
568             BTRFS_LEAF_DATA_SIZE(root)) {
569                 CORRUPT("invalid item offset size pair", leaf, root, 0);
570                 return -EIO;
571         }
572
573         /*
574          * Check to make sure each items keys are in the correct order and their
575          * offsets make sense.  We only have to loop through nritems-1 because
576          * we check the current slot against the next slot, which verifies the
577          * next slot's offset+size makes sense and that the current's slot
578          * offset is correct.
579          */
580         for (slot = 0; slot < nritems - 1; slot++) {
581                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
582                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
583
584                 /* Make sure the keys are in the right order */
585                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
586                         CORRUPT("bad key order", leaf, root, slot);
587                         return -EIO;
588                 }
589
590                 /*
591                  * Make sure the offset and ends are right, remember that the
592                  * item data starts at the end of the leaf and grows towards the
593                  * front.
594                  */
595                 if (btrfs_item_offset_nr(leaf, slot) !=
596                         btrfs_item_end_nr(leaf, slot + 1)) {
597                         CORRUPT("slot offset bad", leaf, root, slot);
598                         return -EIO;
599                 }
600
601                 /*
602                  * Check to make sure that we don't point outside of the leaf,
603                  * just in case all the items are consistent to each other, but
604                  * all point outside of the leaf.
605                  */
606                 if (btrfs_item_end_nr(leaf, slot) >
607                     BTRFS_LEAF_DATA_SIZE(root)) {
608                         CORRUPT("slot end outside of leaf", leaf, root, slot);
609                         return -EIO;
610                 }
611         }
612
613         return 0;
614 }
615
616 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
617                                       u64 phy_offset, struct page *page,
618                                       u64 start, u64 end, int mirror)
619 {
620         u64 found_start;
621         int found_level;
622         struct extent_buffer *eb;
623         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
624         struct btrfs_fs_info *fs_info = root->fs_info;
625         int ret = 0;
626         int reads_done;
627
628         if (!page->private)
629                 goto out;
630
631         eb = (struct extent_buffer *)page->private;
632
633         /* the pending IO might have been the only thing that kept this buffer
634          * in memory.  Make sure we have a ref for all this other checks
635          */
636         extent_buffer_get(eb);
637
638         reads_done = atomic_dec_and_test(&eb->io_pages);
639         if (!reads_done)
640                 goto err;
641
642         eb->read_mirror = mirror;
643         if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
644                 ret = -EIO;
645                 goto err;
646         }
647
648         found_start = btrfs_header_bytenr(eb);
649         if (found_start != eb->start) {
650                 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
651                              found_start, eb->start);
652                 ret = -EIO;
653                 goto err;
654         }
655         if (check_tree_block_fsid(fs_info, eb)) {
656                 btrfs_err_rl(fs_info, "bad fsid on block %llu",
657                              eb->start);
658                 ret = -EIO;
659                 goto err;
660         }
661         found_level = btrfs_header_level(eb);
662         if (found_level >= BTRFS_MAX_LEVEL) {
663                 btrfs_err(fs_info, "bad tree block level %d",
664                           (int)btrfs_header_level(eb));
665                 ret = -EIO;
666                 goto err;
667         }
668
669         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
670                                        eb, found_level);
671
672         ret = csum_tree_block(fs_info, eb, 1);
673         if (ret)
674                 goto err;
675
676         /*
677          * If this is a leaf block and it is corrupt, set the corrupt bit so
678          * that we don't try and read the other copies of this block, just
679          * return -EIO.
680          */
681         if (found_level == 0 && check_leaf(root, eb)) {
682                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
683                 ret = -EIO;
684         }
685
686         if (!ret)
687                 set_extent_buffer_uptodate(eb);
688 err:
689         if (reads_done &&
690             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691                 btree_readahead_hook(fs_info, eb, eb->start, ret);
692
693         if (ret) {
694                 /*
695                  * our io error hook is going to dec the io pages
696                  * again, we have to make sure it has something
697                  * to decrement
698                  */
699                 atomic_inc(&eb->io_pages);
700                 clear_extent_buffer_uptodate(eb);
701         }
702         free_extent_buffer(eb);
703 out:
704         return ret;
705 }
706
707 static int btree_io_failed_hook(struct page *page, int failed_mirror)
708 {
709         struct extent_buffer *eb;
710
711         eb = (struct extent_buffer *)page->private;
712         set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
713         eb->read_mirror = failed_mirror;
714         atomic_dec(&eb->io_pages);
715         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
716                 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
717         return -EIO;    /* we fixed nothing */
718 }
719
720 static void end_workqueue_bio(struct bio *bio)
721 {
722         struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
723         struct btrfs_fs_info *fs_info;
724         struct btrfs_workqueue *wq;
725         btrfs_work_func_t func;
726
727         fs_info = end_io_wq->info;
728         end_io_wq->error = bio->bi_error;
729
730         if (bio->bi_rw & REQ_WRITE) {
731                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
732                         wq = fs_info->endio_meta_write_workers;
733                         func = btrfs_endio_meta_write_helper;
734                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
735                         wq = fs_info->endio_freespace_worker;
736                         func = btrfs_freespace_write_helper;
737                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
738                         wq = fs_info->endio_raid56_workers;
739                         func = btrfs_endio_raid56_helper;
740                 } else {
741                         wq = fs_info->endio_write_workers;
742                         func = btrfs_endio_write_helper;
743                 }
744         } else {
745                 if (unlikely(end_io_wq->metadata ==
746                              BTRFS_WQ_ENDIO_DIO_REPAIR)) {
747                         wq = fs_info->endio_repair_workers;
748                         func = btrfs_endio_repair_helper;
749                 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
750                         wq = fs_info->endio_raid56_workers;
751                         func = btrfs_endio_raid56_helper;
752                 } else if (end_io_wq->metadata) {
753                         wq = fs_info->endio_meta_workers;
754                         func = btrfs_endio_meta_helper;
755                 } else {
756                         wq = fs_info->endio_workers;
757                         func = btrfs_endio_helper;
758                 }
759         }
760
761         btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
762         btrfs_queue_work(wq, &end_io_wq->work);
763 }
764
765 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
766                         enum btrfs_wq_endio_type metadata)
767 {
768         struct btrfs_end_io_wq *end_io_wq;
769
770         end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
771         if (!end_io_wq)
772                 return -ENOMEM;
773
774         end_io_wq->private = bio->bi_private;
775         end_io_wq->end_io = bio->bi_end_io;
776         end_io_wq->info = info;
777         end_io_wq->error = 0;
778         end_io_wq->bio = bio;
779         end_io_wq->metadata = metadata;
780
781         bio->bi_private = end_io_wq;
782         bio->bi_end_io = end_workqueue_bio;
783         return 0;
784 }
785
786 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
787 {
788         unsigned long limit = min_t(unsigned long,
789                                     info->thread_pool_size,
790                                     info->fs_devices->open_devices);
791         return 256 * limit;
792 }
793
794 static void run_one_async_start(struct btrfs_work *work)
795 {
796         struct async_submit_bio *async;
797         int ret;
798
799         async = container_of(work, struct  async_submit_bio, work);
800         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
801                                       async->mirror_num, async->bio_flags,
802                                       async->bio_offset);
803         if (ret)
804                 async->error = ret;
805 }
806
807 static void run_one_async_done(struct btrfs_work *work)
808 {
809         struct btrfs_fs_info *fs_info;
810         struct async_submit_bio *async;
811         int limit;
812
813         async = container_of(work, struct  async_submit_bio, work);
814         fs_info = BTRFS_I(async->inode)->root->fs_info;
815
816         limit = btrfs_async_submit_limit(fs_info);
817         limit = limit * 2 / 3;
818
819         /*
820          * atomic_dec_return implies a barrier for waitqueue_active
821          */
822         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
823             waitqueue_active(&fs_info->async_submit_wait))
824                 wake_up(&fs_info->async_submit_wait);
825
826         /* If an error occurred we just want to clean up the bio and move on */
827         if (async->error) {
828                 async->bio->bi_error = async->error;
829                 bio_endio(async->bio);
830                 return;
831         }
832
833         async->submit_bio_done(async->inode, async->rw, async->bio,
834                                async->mirror_num, async->bio_flags,
835                                async->bio_offset);
836 }
837
838 static void run_one_async_free(struct btrfs_work *work)
839 {
840         struct async_submit_bio *async;
841
842         async = container_of(work, struct  async_submit_bio, work);
843         kfree(async);
844 }
845
846 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
847                         int rw, struct bio *bio, int mirror_num,
848                         unsigned long bio_flags,
849                         u64 bio_offset,
850                         extent_submit_bio_hook_t *submit_bio_start,
851                         extent_submit_bio_hook_t *submit_bio_done)
852 {
853         struct async_submit_bio *async;
854
855         async = kmalloc(sizeof(*async), GFP_NOFS);
856         if (!async)
857                 return -ENOMEM;
858
859         async->inode = inode;
860         async->rw = rw;
861         async->bio = bio;
862         async->mirror_num = mirror_num;
863         async->submit_bio_start = submit_bio_start;
864         async->submit_bio_done = submit_bio_done;
865
866         btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
867                         run_one_async_done, run_one_async_free);
868
869         async->bio_flags = bio_flags;
870         async->bio_offset = bio_offset;
871
872         async->error = 0;
873
874         atomic_inc(&fs_info->nr_async_submits);
875
876         if (rw & REQ_SYNC)
877                 btrfs_set_work_high_priority(&async->work);
878
879         btrfs_queue_work(fs_info->workers, &async->work);
880
881         while (atomic_read(&fs_info->async_submit_draining) &&
882               atomic_read(&fs_info->nr_async_submits)) {
883                 wait_event(fs_info->async_submit_wait,
884                            (atomic_read(&fs_info->nr_async_submits) == 0));
885         }
886
887         return 0;
888 }
889
890 static int btree_csum_one_bio(struct bio *bio)
891 {
892         struct bio_vec *bvec;
893         struct btrfs_root *root;
894         int i, ret = 0;
895
896         bio_for_each_segment_all(bvec, bio, i) {
897                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
898                 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
899                 if (ret)
900                         break;
901         }
902
903         return ret;
904 }
905
906 static int __btree_submit_bio_start(struct inode *inode, int rw,
907                                     struct bio *bio, int mirror_num,
908                                     unsigned long bio_flags,
909                                     u64 bio_offset)
910 {
911         /*
912          * when we're called for a write, we're already in the async
913          * submission context.  Just jump into btrfs_map_bio
914          */
915         return btree_csum_one_bio(bio);
916 }
917
918 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
919                                  int mirror_num, unsigned long bio_flags,
920                                  u64 bio_offset)
921 {
922         int ret;
923
924         /*
925          * when we're called for a write, we're already in the async
926          * submission context.  Just jump into btrfs_map_bio
927          */
928         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
929         if (ret) {
930                 bio->bi_error = ret;
931                 bio_endio(bio);
932         }
933         return ret;
934 }
935
936 static int check_async_write(struct inode *inode, unsigned long bio_flags)
937 {
938         if (bio_flags & EXTENT_BIO_TREE_LOG)
939                 return 0;
940 #ifdef CONFIG_X86
941         if (static_cpu_has(X86_FEATURE_XMM4_2))
942                 return 0;
943 #endif
944         return 1;
945 }
946
947 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
948                                  int mirror_num, unsigned long bio_flags,
949                                  u64 bio_offset)
950 {
951         int async = check_async_write(inode, bio_flags);
952         int ret;
953
954         if (!(rw & REQ_WRITE)) {
955                 /*
956                  * called for a read, do the setup so that checksum validation
957                  * can happen in the async kernel threads
958                  */
959                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
960                                           bio, BTRFS_WQ_ENDIO_METADATA);
961                 if (ret)
962                         goto out_w_error;
963                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
964                                     mirror_num, 0);
965         } else if (!async) {
966                 ret = btree_csum_one_bio(bio);
967                 if (ret)
968                         goto out_w_error;
969                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
970                                     mirror_num, 0);
971         } else {
972                 /*
973                  * kthread helpers are used to submit writes so that
974                  * checksumming can happen in parallel across all CPUs
975                  */
976                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
977                                           inode, rw, bio, mirror_num, 0,
978                                           bio_offset,
979                                           __btree_submit_bio_start,
980                                           __btree_submit_bio_done);
981         }
982
983         if (ret)
984                 goto out_w_error;
985         return 0;
986
987 out_w_error:
988         bio->bi_error = ret;
989         bio_endio(bio);
990         return ret;
991 }
992
993 #ifdef CONFIG_MIGRATION
994 static int btree_migratepage(struct address_space *mapping,
995                         struct page *newpage, struct page *page,
996                         enum migrate_mode mode)
997 {
998         /*
999          * we can't safely write a btree page from here,
1000          * we haven't done the locking hook
1001          */
1002         if (PageDirty(page))
1003                 return -EAGAIN;
1004         /*
1005          * Buffers may be managed in a filesystem specific way.
1006          * We must have no buffers or drop them.
1007          */
1008         if (page_has_private(page) &&
1009             !try_to_release_page(page, GFP_KERNEL))
1010                 return -EAGAIN;
1011         return migrate_page(mapping, newpage, page, mode);
1012 }
1013 #endif
1014
1015
1016 static int btree_writepages(struct address_space *mapping,
1017                             struct writeback_control *wbc)
1018 {
1019         struct btrfs_fs_info *fs_info;
1020         int ret;
1021
1022         if (wbc->sync_mode == WB_SYNC_NONE) {
1023
1024                 if (wbc->for_kupdate)
1025                         return 0;
1026
1027                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1028                 /* this is a bit racy, but that's ok */
1029                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1030                                              BTRFS_DIRTY_METADATA_THRESH);
1031                 if (ret < 0)
1032                         return 0;
1033         }
1034         return btree_write_cache_pages(mapping, wbc);
1035 }
1036
1037 static int btree_readpage(struct file *file, struct page *page)
1038 {
1039         struct extent_io_tree *tree;
1040         tree = &BTRFS_I(page->mapping->host)->io_tree;
1041         return extent_read_full_page(tree, page, btree_get_extent, 0);
1042 }
1043
1044 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1045 {
1046         if (PageWriteback(page) || PageDirty(page))
1047                 return 0;
1048
1049         return try_release_extent_buffer(page);
1050 }
1051
1052 static void btree_invalidatepage(struct page *page, unsigned int offset,
1053                                  unsigned int length)
1054 {
1055         struct extent_io_tree *tree;
1056         tree = &BTRFS_I(page->mapping->host)->io_tree;
1057         extent_invalidatepage(tree, page, offset);
1058         btree_releasepage(page, GFP_NOFS);
1059         if (PagePrivate(page)) {
1060                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1061                            "page private not zero on page %llu",
1062                            (unsigned long long)page_offset(page));
1063                 ClearPagePrivate(page);
1064                 set_page_private(page, 0);
1065                 put_page(page);
1066         }
1067 }
1068
1069 static int btree_set_page_dirty(struct page *page)
1070 {
1071 #ifdef DEBUG
1072         struct extent_buffer *eb;
1073
1074         BUG_ON(!PagePrivate(page));
1075         eb = (struct extent_buffer *)page->private;
1076         BUG_ON(!eb);
1077         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1078         BUG_ON(!atomic_read(&eb->refs));
1079         btrfs_assert_tree_locked(eb);
1080 #endif
1081         return __set_page_dirty_nobuffers(page);
1082 }
1083
1084 static const struct address_space_operations btree_aops = {
1085         .readpage       = btree_readpage,
1086         .writepages     = btree_writepages,
1087         .releasepage    = btree_releasepage,
1088         .invalidatepage = btree_invalidatepage,
1089 #ifdef CONFIG_MIGRATION
1090         .migratepage    = btree_migratepage,
1091 #endif
1092         .set_page_dirty = btree_set_page_dirty,
1093 };
1094
1095 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1096 {
1097         struct extent_buffer *buf = NULL;
1098         struct inode *btree_inode = root->fs_info->btree_inode;
1099
1100         buf = btrfs_find_create_tree_block(root, bytenr);
1101         if (IS_ERR(buf))
1102                 return;
1103         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1104                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1105         free_extent_buffer(buf);
1106 }
1107
1108 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1109                          int mirror_num, struct extent_buffer **eb)
1110 {
1111         struct extent_buffer *buf = NULL;
1112         struct inode *btree_inode = root->fs_info->btree_inode;
1113         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1114         int ret;
1115
1116         buf = btrfs_find_create_tree_block(root, bytenr);
1117         if (IS_ERR(buf))
1118                 return 0;
1119
1120         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1121
1122         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1123                                        btree_get_extent, mirror_num);
1124         if (ret) {
1125                 free_extent_buffer(buf);
1126                 return ret;
1127         }
1128
1129         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1130                 free_extent_buffer(buf);
1131                 return -EIO;
1132         } else if (extent_buffer_uptodate(buf)) {
1133                 *eb = buf;
1134         } else {
1135                 free_extent_buffer(buf);
1136         }
1137         return 0;
1138 }
1139
1140 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1141                                             u64 bytenr)
1142 {
1143         return find_extent_buffer(fs_info, bytenr);
1144 }
1145
1146 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1147                                                  u64 bytenr)
1148 {
1149         if (btrfs_is_testing(root->fs_info))
1150                 return alloc_test_extent_buffer(root->fs_info, bytenr,
1151                                 root->nodesize);
1152         return alloc_extent_buffer(root->fs_info, bytenr);
1153 }
1154
1155
1156 int btrfs_write_tree_block(struct extent_buffer *buf)
1157 {
1158         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1159                                         buf->start + buf->len - 1);
1160 }
1161
1162 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1163 {
1164         return filemap_fdatawait_range(buf->pages[0]->mapping,
1165                                        buf->start, buf->start + buf->len - 1);
1166 }
1167
1168 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1169                                       u64 parent_transid)
1170 {
1171         struct extent_buffer *buf = NULL;
1172         int ret;
1173
1174         buf = btrfs_find_create_tree_block(root, bytenr);
1175         if (IS_ERR(buf))
1176                 return buf;
1177
1178         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1179         if (ret) {
1180                 free_extent_buffer(buf);
1181                 return ERR_PTR(ret);
1182         }
1183         return buf;
1184
1185 }
1186
1187 void clean_tree_block(struct btrfs_trans_handle *trans,
1188                       struct btrfs_fs_info *fs_info,
1189                       struct extent_buffer *buf)
1190 {
1191         if (btrfs_header_generation(buf) ==
1192             fs_info->running_transaction->transid) {
1193                 btrfs_assert_tree_locked(buf);
1194
1195                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1196                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1197                                              -buf->len,
1198                                              fs_info->dirty_metadata_batch);
1199                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1200                         btrfs_set_lock_blocking(buf);
1201                         clear_extent_buffer_dirty(buf);
1202                 }
1203         }
1204 }
1205
1206 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1207 {
1208         struct btrfs_subvolume_writers *writers;
1209         int ret;
1210
1211         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1212         if (!writers)
1213                 return ERR_PTR(-ENOMEM);
1214
1215         ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1216         if (ret < 0) {
1217                 kfree(writers);
1218                 return ERR_PTR(ret);
1219         }
1220
1221         init_waitqueue_head(&writers->wait);
1222         return writers;
1223 }
1224
1225 static void
1226 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1227 {
1228         percpu_counter_destroy(&writers->counter);
1229         kfree(writers);
1230 }
1231
1232 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1233                          struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1234                          u64 objectid)
1235 {
1236         bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1237         root->node = NULL;
1238         root->commit_root = NULL;
1239         root->sectorsize = sectorsize;
1240         root->nodesize = nodesize;
1241         root->stripesize = stripesize;
1242         root->state = 0;
1243         root->orphan_cleanup_state = 0;
1244
1245         root->objectid = objectid;
1246         root->last_trans = 0;
1247         root->highest_objectid = 0;
1248         root->nr_delalloc_inodes = 0;
1249         root->nr_ordered_extents = 0;
1250         root->name = NULL;
1251         root->inode_tree = RB_ROOT;
1252         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1253         root->block_rsv = NULL;
1254         root->orphan_block_rsv = NULL;
1255
1256         INIT_LIST_HEAD(&root->dirty_list);
1257         INIT_LIST_HEAD(&root->root_list);
1258         INIT_LIST_HEAD(&root->delalloc_inodes);
1259         INIT_LIST_HEAD(&root->delalloc_root);
1260         INIT_LIST_HEAD(&root->ordered_extents);
1261         INIT_LIST_HEAD(&root->ordered_root);
1262         INIT_LIST_HEAD(&root->logged_list[0]);
1263         INIT_LIST_HEAD(&root->logged_list[1]);
1264         spin_lock_init(&root->orphan_lock);
1265         spin_lock_init(&root->inode_lock);
1266         spin_lock_init(&root->delalloc_lock);
1267         spin_lock_init(&root->ordered_extent_lock);
1268         spin_lock_init(&root->accounting_lock);
1269         spin_lock_init(&root->log_extents_lock[0]);
1270         spin_lock_init(&root->log_extents_lock[1]);
1271         mutex_init(&root->objectid_mutex);
1272         mutex_init(&root->log_mutex);
1273         mutex_init(&root->ordered_extent_mutex);
1274         mutex_init(&root->delalloc_mutex);
1275         init_waitqueue_head(&root->log_writer_wait);
1276         init_waitqueue_head(&root->log_commit_wait[0]);
1277         init_waitqueue_head(&root->log_commit_wait[1]);
1278         INIT_LIST_HEAD(&root->log_ctxs[0]);
1279         INIT_LIST_HEAD(&root->log_ctxs[1]);
1280         atomic_set(&root->log_commit[0], 0);
1281         atomic_set(&root->log_commit[1], 0);
1282         atomic_set(&root->log_writers, 0);
1283         atomic_set(&root->log_batch, 0);
1284         atomic_set(&root->orphan_inodes, 0);
1285         atomic_set(&root->refs, 1);
1286         atomic_set(&root->will_be_snapshoted, 0);
1287         atomic_set(&root->qgroup_meta_rsv, 0);
1288         root->log_transid = 0;
1289         root->log_transid_committed = -1;
1290         root->last_log_commit = 0;
1291         if (!dummy)
1292                 extent_io_tree_init(&root->dirty_log_pages,
1293                                      fs_info->btree_inode->i_mapping);
1294
1295         memset(&root->root_key, 0, sizeof(root->root_key));
1296         memset(&root->root_item, 0, sizeof(root->root_item));
1297         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1298         if (!dummy)
1299                 root->defrag_trans_start = fs_info->generation;
1300         else
1301                 root->defrag_trans_start = 0;
1302         root->root_key.objectid = objectid;
1303         root->anon_dev = 0;
1304
1305         spin_lock_init(&root->root_item_lock);
1306 }
1307
1308 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1309                 gfp_t flags)
1310 {
1311         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1312         if (root)
1313                 root->fs_info = fs_info;
1314         return root;
1315 }
1316
1317 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1318 /* Should only be used by the testing infrastructure */
1319 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1320                                           u32 sectorsize, u32 nodesize)
1321 {
1322         struct btrfs_root *root;
1323
1324         if (!fs_info)
1325                 return ERR_PTR(-EINVAL);
1326
1327         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1328         if (!root)
1329                 return ERR_PTR(-ENOMEM);
1330         /* We don't use the stripesize in selftest, set it as sectorsize */
1331         __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1332                         BTRFS_ROOT_TREE_OBJECTID);
1333         root->alloc_bytenr = 0;
1334
1335         return root;
1336 }
1337 #endif
1338
1339 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1340                                      struct btrfs_fs_info *fs_info,
1341                                      u64 objectid)
1342 {
1343         struct extent_buffer *leaf;
1344         struct btrfs_root *tree_root = fs_info->tree_root;
1345         struct btrfs_root *root;
1346         struct btrfs_key key;
1347         int ret = 0;
1348         uuid_le uuid;
1349
1350         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1351         if (!root)
1352                 return ERR_PTR(-ENOMEM);
1353
1354         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1355                 tree_root->stripesize, root, fs_info, objectid);
1356         root->root_key.objectid = objectid;
1357         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1358         root->root_key.offset = 0;
1359
1360         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1361         if (IS_ERR(leaf)) {
1362                 ret = PTR_ERR(leaf);
1363                 leaf = NULL;
1364                 goto fail;
1365         }
1366
1367         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1368         btrfs_set_header_bytenr(leaf, leaf->start);
1369         btrfs_set_header_generation(leaf, trans->transid);
1370         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1371         btrfs_set_header_owner(leaf, objectid);
1372         root->node = leaf;
1373
1374         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1375                             BTRFS_FSID_SIZE);
1376         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1377                             btrfs_header_chunk_tree_uuid(leaf),
1378                             BTRFS_UUID_SIZE);
1379         btrfs_mark_buffer_dirty(leaf);
1380
1381         root->commit_root = btrfs_root_node(root);
1382         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1383
1384         root->root_item.flags = 0;
1385         root->root_item.byte_limit = 0;
1386         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1387         btrfs_set_root_generation(&root->root_item, trans->transid);
1388         btrfs_set_root_level(&root->root_item, 0);
1389         btrfs_set_root_refs(&root->root_item, 1);
1390         btrfs_set_root_used(&root->root_item, leaf->len);
1391         btrfs_set_root_last_snapshot(&root->root_item, 0);
1392         btrfs_set_root_dirid(&root->root_item, 0);
1393         uuid_le_gen(&uuid);
1394         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1395         root->root_item.drop_level = 0;
1396
1397         key.objectid = objectid;
1398         key.type = BTRFS_ROOT_ITEM_KEY;
1399         key.offset = 0;
1400         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1401         if (ret)
1402                 goto fail;
1403
1404         btrfs_tree_unlock(leaf);
1405
1406         return root;
1407
1408 fail:
1409         if (leaf) {
1410                 btrfs_tree_unlock(leaf);
1411                 free_extent_buffer(root->commit_root);
1412                 free_extent_buffer(leaf);
1413         }
1414         kfree(root);
1415
1416         return ERR_PTR(ret);
1417 }
1418
1419 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1420                                          struct btrfs_fs_info *fs_info)
1421 {
1422         struct btrfs_root *root;
1423         struct btrfs_root *tree_root = fs_info->tree_root;
1424         struct extent_buffer *leaf;
1425
1426         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1427         if (!root)
1428                 return ERR_PTR(-ENOMEM);
1429
1430         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1431                      tree_root->stripesize, root, fs_info,
1432                      BTRFS_TREE_LOG_OBJECTID);
1433
1434         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1435         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1436         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1437
1438         /*
1439          * DON'T set REF_COWS for log trees
1440          *
1441          * log trees do not get reference counted because they go away
1442          * before a real commit is actually done.  They do store pointers
1443          * to file data extents, and those reference counts still get
1444          * updated (along with back refs to the log tree).
1445          */
1446
1447         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1448                         NULL, 0, 0, 0);
1449         if (IS_ERR(leaf)) {
1450                 kfree(root);
1451                 return ERR_CAST(leaf);
1452         }
1453
1454         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1455         btrfs_set_header_bytenr(leaf, leaf->start);
1456         btrfs_set_header_generation(leaf, trans->transid);
1457         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1458         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1459         root->node = leaf;
1460
1461         write_extent_buffer(root->node, root->fs_info->fsid,
1462                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1463         btrfs_mark_buffer_dirty(root->node);
1464         btrfs_tree_unlock(root->node);
1465         return root;
1466 }
1467
1468 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1469                              struct btrfs_fs_info *fs_info)
1470 {
1471         struct btrfs_root *log_root;
1472
1473         log_root = alloc_log_tree(trans, fs_info);
1474         if (IS_ERR(log_root))
1475                 return PTR_ERR(log_root);
1476         WARN_ON(fs_info->log_root_tree);
1477         fs_info->log_root_tree = log_root;
1478         return 0;
1479 }
1480
1481 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1482                        struct btrfs_root *root)
1483 {
1484         struct btrfs_root *log_root;
1485         struct btrfs_inode_item *inode_item;
1486
1487         log_root = alloc_log_tree(trans, root->fs_info);
1488         if (IS_ERR(log_root))
1489                 return PTR_ERR(log_root);
1490
1491         log_root->last_trans = trans->transid;
1492         log_root->root_key.offset = root->root_key.objectid;
1493
1494         inode_item = &log_root->root_item.inode;
1495         btrfs_set_stack_inode_generation(inode_item, 1);
1496         btrfs_set_stack_inode_size(inode_item, 3);
1497         btrfs_set_stack_inode_nlink(inode_item, 1);
1498         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1499         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1500
1501         btrfs_set_root_node(&log_root->root_item, log_root->node);
1502
1503         WARN_ON(root->log_root);
1504         root->log_root = log_root;
1505         root->log_transid = 0;
1506         root->log_transid_committed = -1;
1507         root->last_log_commit = 0;
1508         return 0;
1509 }
1510
1511 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1512                                                struct btrfs_key *key)
1513 {
1514         struct btrfs_root *root;
1515         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1516         struct btrfs_path *path;
1517         u64 generation;
1518         int ret;
1519
1520         path = btrfs_alloc_path();
1521         if (!path)
1522                 return ERR_PTR(-ENOMEM);
1523
1524         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1525         if (!root) {
1526                 ret = -ENOMEM;
1527                 goto alloc_fail;
1528         }
1529
1530         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1531                 tree_root->stripesize, root, fs_info, key->objectid);
1532
1533         ret = btrfs_find_root(tree_root, key, path,
1534                               &root->root_item, &root->root_key);
1535         if (ret) {
1536                 if (ret > 0)
1537                         ret = -ENOENT;
1538                 goto find_fail;
1539         }
1540
1541         generation = btrfs_root_generation(&root->root_item);
1542         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1543                                      generation);
1544         if (IS_ERR(root->node)) {
1545                 ret = PTR_ERR(root->node);
1546                 goto find_fail;
1547         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1548                 ret = -EIO;
1549                 free_extent_buffer(root->node);
1550                 goto find_fail;
1551         }
1552         root->commit_root = btrfs_root_node(root);
1553 out:
1554         btrfs_free_path(path);
1555         return root;
1556
1557 find_fail:
1558         kfree(root);
1559 alloc_fail:
1560         root = ERR_PTR(ret);
1561         goto out;
1562 }
1563
1564 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1565                                       struct btrfs_key *location)
1566 {
1567         struct btrfs_root *root;
1568
1569         root = btrfs_read_tree_root(tree_root, location);
1570         if (IS_ERR(root))
1571                 return root;
1572
1573         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1574                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1575                 btrfs_check_and_init_root_item(&root->root_item);
1576         }
1577
1578         return root;
1579 }
1580
1581 int btrfs_init_fs_root(struct btrfs_root *root)
1582 {
1583         int ret;
1584         struct btrfs_subvolume_writers *writers;
1585
1586         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1587         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1588                                         GFP_NOFS);
1589         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1590                 ret = -ENOMEM;
1591                 goto fail;
1592         }
1593
1594         writers = btrfs_alloc_subvolume_writers();
1595         if (IS_ERR(writers)) {
1596                 ret = PTR_ERR(writers);
1597                 goto fail;
1598         }
1599         root->subv_writers = writers;
1600
1601         btrfs_init_free_ino_ctl(root);
1602         spin_lock_init(&root->ino_cache_lock);
1603         init_waitqueue_head(&root->ino_cache_wait);
1604
1605         ret = get_anon_bdev(&root->anon_dev);
1606         if (ret)
1607                 goto fail;
1608
1609         mutex_lock(&root->objectid_mutex);
1610         ret = btrfs_find_highest_objectid(root,
1611                                         &root->highest_objectid);
1612         if (ret) {
1613                 mutex_unlock(&root->objectid_mutex);
1614                 goto fail;
1615         }
1616
1617         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1618
1619         mutex_unlock(&root->objectid_mutex);
1620
1621         return 0;
1622 fail:
1623         /* the caller is responsible to call free_fs_root */
1624         return ret;
1625 }
1626
1627 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1628                                                u64 root_id)
1629 {
1630         struct btrfs_root *root;
1631
1632         spin_lock(&fs_info->fs_roots_radix_lock);
1633         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1634                                  (unsigned long)root_id);
1635         spin_unlock(&fs_info->fs_roots_radix_lock);
1636         return root;
1637 }
1638
1639 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1640                          struct btrfs_root *root)
1641 {
1642         int ret;
1643
1644         ret = radix_tree_preload(GFP_NOFS);
1645         if (ret)
1646                 return ret;
1647
1648         spin_lock(&fs_info->fs_roots_radix_lock);
1649         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1650                                 (unsigned long)root->root_key.objectid,
1651                                 root);
1652         if (ret == 0)
1653                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1654         spin_unlock(&fs_info->fs_roots_radix_lock);
1655         radix_tree_preload_end();
1656
1657         return ret;
1658 }
1659
1660 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1661                                      struct btrfs_key *location,
1662                                      bool check_ref)
1663 {
1664         struct btrfs_root *root;
1665         struct btrfs_path *path;
1666         struct btrfs_key key;
1667         int ret;
1668
1669         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1670                 return fs_info->tree_root;
1671         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1672                 return fs_info->extent_root;
1673         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1674                 return fs_info->chunk_root;
1675         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1676                 return fs_info->dev_root;
1677         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1678                 return fs_info->csum_root;
1679         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1680                 return fs_info->quota_root ? fs_info->quota_root :
1681                                              ERR_PTR(-ENOENT);
1682         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1683                 return fs_info->uuid_root ? fs_info->uuid_root :
1684                                             ERR_PTR(-ENOENT);
1685         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1686                 return fs_info->free_space_root ? fs_info->free_space_root :
1687                                                   ERR_PTR(-ENOENT);
1688 again:
1689         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1690         if (root) {
1691                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1692                         return ERR_PTR(-ENOENT);
1693                 return root;
1694         }
1695
1696         root = btrfs_read_fs_root(fs_info->tree_root, location);
1697         if (IS_ERR(root))
1698                 return root;
1699
1700         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701                 ret = -ENOENT;
1702                 goto fail;
1703         }
1704
1705         ret = btrfs_init_fs_root(root);
1706         if (ret)
1707                 goto fail;
1708
1709         path = btrfs_alloc_path();
1710         if (!path) {
1711                 ret = -ENOMEM;
1712                 goto fail;
1713         }
1714         key.objectid = BTRFS_ORPHAN_OBJECTID;
1715         key.type = BTRFS_ORPHAN_ITEM_KEY;
1716         key.offset = location->objectid;
1717
1718         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719         btrfs_free_path(path);
1720         if (ret < 0)
1721                 goto fail;
1722         if (ret == 0)
1723                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725         ret = btrfs_insert_fs_root(fs_info, root);
1726         if (ret) {
1727                 if (ret == -EEXIST) {
1728                         free_fs_root(root);
1729                         goto again;
1730                 }
1731                 goto fail;
1732         }
1733         return root;
1734 fail:
1735         free_fs_root(root);
1736         return ERR_PTR(ret);
1737 }
1738
1739 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1740 {
1741         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1742         int ret = 0;
1743         struct btrfs_device *device;
1744         struct backing_dev_info *bdi;
1745
1746         rcu_read_lock();
1747         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1748                 if (!device->bdev)
1749                         continue;
1750                 bdi = blk_get_backing_dev_info(device->bdev);
1751                 if (bdi_congested(bdi, bdi_bits)) {
1752                         ret = 1;
1753                         break;
1754                 }
1755         }
1756         rcu_read_unlock();
1757         return ret;
1758 }
1759
1760 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1761 {
1762         int err;
1763
1764         err = bdi_setup_and_register(bdi, "btrfs");
1765         if (err)
1766                 return err;
1767
1768         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1769         bdi->congested_fn       = btrfs_congested_fn;
1770         bdi->congested_data     = info;
1771         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1772         return 0;
1773 }
1774
1775 /*
1776  * called by the kthread helper functions to finally call the bio end_io
1777  * functions.  This is where read checksum verification actually happens
1778  */
1779 static void end_workqueue_fn(struct btrfs_work *work)
1780 {
1781         struct bio *bio;
1782         struct btrfs_end_io_wq *end_io_wq;
1783
1784         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1785         bio = end_io_wq->bio;
1786
1787         bio->bi_error = end_io_wq->error;
1788         bio->bi_private = end_io_wq->private;
1789         bio->bi_end_io = end_io_wq->end_io;
1790         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1791         bio_endio(bio);
1792 }
1793
1794 static int cleaner_kthread(void *arg)
1795 {
1796         struct btrfs_root *root = arg;
1797         int again;
1798         struct btrfs_trans_handle *trans;
1799
1800         do {
1801                 again = 0;
1802
1803                 /* Make the cleaner go to sleep early. */
1804                 if (btrfs_need_cleaner_sleep(root))
1805                         goto sleep;
1806
1807                 /*
1808                  * Do not do anything if we might cause open_ctree() to block
1809                  * before we have finished mounting the filesystem.
1810                  */
1811                 if (!root->fs_info->open)
1812                         goto sleep;
1813
1814                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1815                         goto sleep;
1816
1817                 /*
1818                  * Avoid the problem that we change the status of the fs
1819                  * during the above check and trylock.
1820                  */
1821                 if (btrfs_need_cleaner_sleep(root)) {
1822                         mutex_unlock(&root->fs_info->cleaner_mutex);
1823                         goto sleep;
1824                 }
1825
1826                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1827                 btrfs_run_delayed_iputs(root);
1828                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1829
1830                 again = btrfs_clean_one_deleted_snapshot(root);
1831                 mutex_unlock(&root->fs_info->cleaner_mutex);
1832
1833                 /*
1834                  * The defragger has dealt with the R/O remount and umount,
1835                  * needn't do anything special here.
1836                  */
1837                 btrfs_run_defrag_inodes(root->fs_info);
1838
1839                 /*
1840                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1841                  * with relocation (btrfs_relocate_chunk) and relocation
1842                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1843                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1844                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1845                  * unused block groups.
1846                  */
1847                 btrfs_delete_unused_bgs(root->fs_info);
1848 sleep:
1849                 if (!again) {
1850                         set_current_state(TASK_INTERRUPTIBLE);
1851                         if (!kthread_should_stop())
1852                                 schedule();
1853                         __set_current_state(TASK_RUNNING);
1854                 }
1855         } while (!kthread_should_stop());
1856
1857         /*
1858          * Transaction kthread is stopped before us and wakes us up.
1859          * However we might have started a new transaction and COWed some
1860          * tree blocks when deleting unused block groups for example. So
1861          * make sure we commit the transaction we started to have a clean
1862          * shutdown when evicting the btree inode - if it has dirty pages
1863          * when we do the final iput() on it, eviction will trigger a
1864          * writeback for it which will fail with null pointer dereferences
1865          * since work queues and other resources were already released and
1866          * destroyed by the time the iput/eviction/writeback is made.
1867          */
1868         trans = btrfs_attach_transaction(root);
1869         if (IS_ERR(trans)) {
1870                 if (PTR_ERR(trans) != -ENOENT)
1871                         btrfs_err(root->fs_info,
1872                                   "cleaner transaction attach returned %ld",
1873                                   PTR_ERR(trans));
1874         } else {
1875                 int ret;
1876
1877                 ret = btrfs_commit_transaction(trans, root);
1878                 if (ret)
1879                         btrfs_err(root->fs_info,
1880                                   "cleaner open transaction commit returned %d",
1881                                   ret);
1882         }
1883
1884         return 0;
1885 }
1886
1887 static int transaction_kthread(void *arg)
1888 {
1889         struct btrfs_root *root = arg;
1890         struct btrfs_trans_handle *trans;
1891         struct btrfs_transaction *cur;
1892         u64 transid;
1893         unsigned long now;
1894         unsigned long delay;
1895         bool cannot_commit;
1896
1897         do {
1898                 cannot_commit = false;
1899                 delay = HZ * root->fs_info->commit_interval;
1900                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1901
1902                 spin_lock(&root->fs_info->trans_lock);
1903                 cur = root->fs_info->running_transaction;
1904                 if (!cur) {
1905                         spin_unlock(&root->fs_info->trans_lock);
1906                         goto sleep;
1907                 }
1908
1909                 now = get_seconds();
1910                 if (cur->state < TRANS_STATE_BLOCKED &&
1911                     (now < cur->start_time ||
1912                      now - cur->start_time < root->fs_info->commit_interval)) {
1913                         spin_unlock(&root->fs_info->trans_lock);
1914                         delay = HZ * 5;
1915                         goto sleep;
1916                 }
1917                 transid = cur->transid;
1918                 spin_unlock(&root->fs_info->trans_lock);
1919
1920                 /* If the file system is aborted, this will always fail. */
1921                 trans = btrfs_attach_transaction(root);
1922                 if (IS_ERR(trans)) {
1923                         if (PTR_ERR(trans) != -ENOENT)
1924                                 cannot_commit = true;
1925                         goto sleep;
1926                 }
1927                 if (transid == trans->transid) {
1928                         btrfs_commit_transaction(trans, root);
1929                 } else {
1930                         btrfs_end_transaction(trans, root);
1931                 }
1932 sleep:
1933                 wake_up_process(root->fs_info->cleaner_kthread);
1934                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1935
1936                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1937                                       &root->fs_info->fs_state)))
1938                         btrfs_cleanup_transaction(root);
1939                 set_current_state(TASK_INTERRUPTIBLE);
1940                 if (!kthread_should_stop() &&
1941                                 (!btrfs_transaction_blocked(root->fs_info) ||
1942                                  cannot_commit))
1943                         schedule_timeout(delay);
1944                 __set_current_state(TASK_RUNNING);
1945         } while (!kthread_should_stop());
1946         return 0;
1947 }
1948
1949 /*
1950  * this will find the highest generation in the array of
1951  * root backups.  The index of the highest array is returned,
1952  * or -1 if we can't find anything.
1953  *
1954  * We check to make sure the array is valid by comparing the
1955  * generation of the latest  root in the array with the generation
1956  * in the super block.  If they don't match we pitch it.
1957  */
1958 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1959 {
1960         u64 cur;
1961         int newest_index = -1;
1962         struct btrfs_root_backup *root_backup;
1963         int i;
1964
1965         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1966                 root_backup = info->super_copy->super_roots + i;
1967                 cur = btrfs_backup_tree_root_gen(root_backup);
1968                 if (cur == newest_gen)
1969                         newest_index = i;
1970         }
1971
1972         /* check to see if we actually wrapped around */
1973         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1974                 root_backup = info->super_copy->super_roots;
1975                 cur = btrfs_backup_tree_root_gen(root_backup);
1976                 if (cur == newest_gen)
1977                         newest_index = 0;
1978         }
1979         return newest_index;
1980 }
1981
1982
1983 /*
1984  * find the oldest backup so we know where to store new entries
1985  * in the backup array.  This will set the backup_root_index
1986  * field in the fs_info struct
1987  */
1988 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1989                                      u64 newest_gen)
1990 {
1991         int newest_index = -1;
1992
1993         newest_index = find_newest_super_backup(info, newest_gen);
1994         /* if there was garbage in there, just move along */
1995         if (newest_index == -1) {
1996                 info->backup_root_index = 0;
1997         } else {
1998                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1999         }
2000 }
2001
2002 /*
2003  * copy all the root pointers into the super backup array.
2004  * this will bump the backup pointer by one when it is
2005  * done
2006  */
2007 static void backup_super_roots(struct btrfs_fs_info *info)
2008 {
2009         int next_backup;
2010         struct btrfs_root_backup *root_backup;
2011         int last_backup;
2012
2013         next_backup = info->backup_root_index;
2014         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2015                 BTRFS_NUM_BACKUP_ROOTS;
2016
2017         /*
2018          * just overwrite the last backup if we're at the same generation
2019          * this happens only at umount
2020          */
2021         root_backup = info->super_for_commit->super_roots + last_backup;
2022         if (btrfs_backup_tree_root_gen(root_backup) ==
2023             btrfs_header_generation(info->tree_root->node))
2024                 next_backup = last_backup;
2025
2026         root_backup = info->super_for_commit->super_roots + next_backup;
2027
2028         /*
2029          * make sure all of our padding and empty slots get zero filled
2030          * regardless of which ones we use today
2031          */
2032         memset(root_backup, 0, sizeof(*root_backup));
2033
2034         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2035
2036         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2037         btrfs_set_backup_tree_root_gen(root_backup,
2038                                btrfs_header_generation(info->tree_root->node));
2039
2040         btrfs_set_backup_tree_root_level(root_backup,
2041                                btrfs_header_level(info->tree_root->node));
2042
2043         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2044         btrfs_set_backup_chunk_root_gen(root_backup,
2045                                btrfs_header_generation(info->chunk_root->node));
2046         btrfs_set_backup_chunk_root_level(root_backup,
2047                                btrfs_header_level(info->chunk_root->node));
2048
2049         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2050         btrfs_set_backup_extent_root_gen(root_backup,
2051                                btrfs_header_generation(info->extent_root->node));
2052         btrfs_set_backup_extent_root_level(root_backup,
2053                                btrfs_header_level(info->extent_root->node));
2054
2055         /*
2056          * we might commit during log recovery, which happens before we set
2057          * the fs_root.  Make sure it is valid before we fill it in.
2058          */
2059         if (info->fs_root && info->fs_root->node) {
2060                 btrfs_set_backup_fs_root(root_backup,
2061                                          info->fs_root->node->start);
2062                 btrfs_set_backup_fs_root_gen(root_backup,
2063                                btrfs_header_generation(info->fs_root->node));
2064                 btrfs_set_backup_fs_root_level(root_backup,
2065                                btrfs_header_level(info->fs_root->node));
2066         }
2067
2068         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2069         btrfs_set_backup_dev_root_gen(root_backup,
2070                                btrfs_header_generation(info->dev_root->node));
2071         btrfs_set_backup_dev_root_level(root_backup,
2072                                        btrfs_header_level(info->dev_root->node));
2073
2074         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2075         btrfs_set_backup_csum_root_gen(root_backup,
2076                                btrfs_header_generation(info->csum_root->node));
2077         btrfs_set_backup_csum_root_level(root_backup,
2078                                btrfs_header_level(info->csum_root->node));
2079
2080         btrfs_set_backup_total_bytes(root_backup,
2081                              btrfs_super_total_bytes(info->super_copy));
2082         btrfs_set_backup_bytes_used(root_backup,
2083                              btrfs_super_bytes_used(info->super_copy));
2084         btrfs_set_backup_num_devices(root_backup,
2085                              btrfs_super_num_devices(info->super_copy));
2086
2087         /*
2088          * if we don't copy this out to the super_copy, it won't get remembered
2089          * for the next commit
2090          */
2091         memcpy(&info->super_copy->super_roots,
2092                &info->super_for_commit->super_roots,
2093                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2094 }
2095
2096 /*
2097  * this copies info out of the root backup array and back into
2098  * the in-memory super block.  It is meant to help iterate through
2099  * the array, so you send it the number of backups you've already
2100  * tried and the last backup index you used.
2101  *
2102  * this returns -1 when it has tried all the backups
2103  */
2104 static noinline int next_root_backup(struct btrfs_fs_info *info,
2105                                      struct btrfs_super_block *super,
2106                                      int *num_backups_tried, int *backup_index)
2107 {
2108         struct btrfs_root_backup *root_backup;
2109         int newest = *backup_index;
2110
2111         if (*num_backups_tried == 0) {
2112                 u64 gen = btrfs_super_generation(super);
2113
2114                 newest = find_newest_super_backup(info, gen);
2115                 if (newest == -1)
2116                         return -1;
2117
2118                 *backup_index = newest;
2119                 *num_backups_tried = 1;
2120         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2121                 /* we've tried all the backups, all done */
2122                 return -1;
2123         } else {
2124                 /* jump to the next oldest backup */
2125                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2126                         BTRFS_NUM_BACKUP_ROOTS;
2127                 *backup_index = newest;
2128                 *num_backups_tried += 1;
2129         }
2130         root_backup = super->super_roots + newest;
2131
2132         btrfs_set_super_generation(super,
2133                                    btrfs_backup_tree_root_gen(root_backup));
2134         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2135         btrfs_set_super_root_level(super,
2136                                    btrfs_backup_tree_root_level(root_backup));
2137         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2138
2139         /*
2140          * fixme: the total bytes and num_devices need to match or we should
2141          * need a fsck
2142          */
2143         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2144         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2145         return 0;
2146 }
2147
2148 /* helper to cleanup workers */
2149 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2150 {
2151         btrfs_destroy_workqueue(fs_info->fixup_workers);
2152         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2153         btrfs_destroy_workqueue(fs_info->workers);
2154         btrfs_destroy_workqueue(fs_info->endio_workers);
2155         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2156         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2157         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2158         btrfs_destroy_workqueue(fs_info->rmw_workers);
2159         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2160         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2161         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2162         btrfs_destroy_workqueue(fs_info->submit_workers);
2163         btrfs_destroy_workqueue(fs_info->delayed_workers);
2164         btrfs_destroy_workqueue(fs_info->caching_workers);
2165         btrfs_destroy_workqueue(fs_info->readahead_workers);
2166         btrfs_destroy_workqueue(fs_info->flush_workers);
2167         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2168         btrfs_destroy_workqueue(fs_info->extent_workers);
2169 }
2170
2171 static void free_root_extent_buffers(struct btrfs_root *root)
2172 {
2173         if (root) {
2174                 free_extent_buffer(root->node);
2175                 free_extent_buffer(root->commit_root);
2176                 root->node = NULL;
2177                 root->commit_root = NULL;
2178         }
2179 }
2180
2181 /* helper to cleanup tree roots */
2182 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2183 {
2184         free_root_extent_buffers(info->tree_root);
2185
2186         free_root_extent_buffers(info->dev_root);
2187         free_root_extent_buffers(info->extent_root);
2188         free_root_extent_buffers(info->csum_root);
2189         free_root_extent_buffers(info->quota_root);
2190         free_root_extent_buffers(info->uuid_root);
2191         if (chunk_root)
2192                 free_root_extent_buffers(info->chunk_root);
2193         free_root_extent_buffers(info->free_space_root);
2194 }
2195
2196 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2197 {
2198         int ret;
2199         struct btrfs_root *gang[8];
2200         int i;
2201
2202         while (!list_empty(&fs_info->dead_roots)) {
2203                 gang[0] = list_entry(fs_info->dead_roots.next,
2204                                      struct btrfs_root, root_list);
2205                 list_del(&gang[0]->root_list);
2206
2207                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2208                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2209                 } else {
2210                         free_extent_buffer(gang[0]->node);
2211                         free_extent_buffer(gang[0]->commit_root);
2212                         btrfs_put_fs_root(gang[0]);
2213                 }
2214         }
2215
2216         while (1) {
2217                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2218                                              (void **)gang, 0,
2219                                              ARRAY_SIZE(gang));
2220                 if (!ret)
2221                         break;
2222                 for (i = 0; i < ret; i++)
2223                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2224         }
2225
2226         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2227                 btrfs_free_log_root_tree(NULL, fs_info);
2228                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2229                                             fs_info->pinned_extents);
2230         }
2231 }
2232
2233 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2234 {
2235         mutex_init(&fs_info->scrub_lock);
2236         atomic_set(&fs_info->scrubs_running, 0);
2237         atomic_set(&fs_info->scrub_pause_req, 0);
2238         atomic_set(&fs_info->scrubs_paused, 0);
2239         atomic_set(&fs_info->scrub_cancel_req, 0);
2240         init_waitqueue_head(&fs_info->scrub_pause_wait);
2241         fs_info->scrub_workers_refcnt = 0;
2242 }
2243
2244 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2245 {
2246         spin_lock_init(&fs_info->balance_lock);
2247         mutex_init(&fs_info->balance_mutex);
2248         atomic_set(&fs_info->balance_running, 0);
2249         atomic_set(&fs_info->balance_pause_req, 0);
2250         atomic_set(&fs_info->balance_cancel_req, 0);
2251         fs_info->balance_ctl = NULL;
2252         init_waitqueue_head(&fs_info->balance_wait_q);
2253 }
2254
2255 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2256                                    struct btrfs_root *tree_root)
2257 {
2258         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2259         set_nlink(fs_info->btree_inode, 1);
2260         /*
2261          * we set the i_size on the btree inode to the max possible int.
2262          * the real end of the address space is determined by all of
2263          * the devices in the system
2264          */
2265         fs_info->btree_inode->i_size = OFFSET_MAX;
2266         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2267
2268         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2269         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2270                              fs_info->btree_inode->i_mapping);
2271         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2272         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2273
2274         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2275
2276         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2277         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2278                sizeof(struct btrfs_key));
2279         set_bit(BTRFS_INODE_DUMMY,
2280                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2281         btrfs_insert_inode_hash(fs_info->btree_inode);
2282 }
2283
2284 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2285 {
2286         fs_info->dev_replace.lock_owner = 0;
2287         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2288         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2289         rwlock_init(&fs_info->dev_replace.lock);
2290         atomic_set(&fs_info->dev_replace.read_locks, 0);
2291         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2292         init_waitqueue_head(&fs_info->replace_wait);
2293         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2294 }
2295
2296 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2297 {
2298         spin_lock_init(&fs_info->qgroup_lock);
2299         mutex_init(&fs_info->qgroup_ioctl_lock);
2300         fs_info->qgroup_tree = RB_ROOT;
2301         fs_info->qgroup_op_tree = RB_ROOT;
2302         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2303         fs_info->qgroup_seq = 1;
2304         fs_info->quota_enabled = 0;
2305         fs_info->pending_quota_state = 0;
2306         fs_info->qgroup_ulist = NULL;
2307         mutex_init(&fs_info->qgroup_rescan_lock);
2308 }
2309
2310 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2311                 struct btrfs_fs_devices *fs_devices)
2312 {
2313         int max_active = fs_info->thread_pool_size;
2314         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2315
2316         fs_info->workers =
2317                 btrfs_alloc_workqueue(fs_info, "worker",
2318                                       flags | WQ_HIGHPRI, max_active, 16);
2319
2320         fs_info->delalloc_workers =
2321                 btrfs_alloc_workqueue(fs_info, "delalloc",
2322                                       flags, max_active, 2);
2323
2324         fs_info->flush_workers =
2325                 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2326                                       flags, max_active, 0);
2327
2328         fs_info->caching_workers =
2329                 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2330
2331         /*
2332          * a higher idle thresh on the submit workers makes it much more
2333          * likely that bios will be send down in a sane order to the
2334          * devices
2335          */
2336         fs_info->submit_workers =
2337                 btrfs_alloc_workqueue(fs_info, "submit", flags,
2338                                       min_t(u64, fs_devices->num_devices,
2339                                             max_active), 64);
2340
2341         fs_info->fixup_workers =
2342                 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2343
2344         /*
2345          * endios are largely parallel and should have a very
2346          * low idle thresh
2347          */
2348         fs_info->endio_workers =
2349                 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2350         fs_info->endio_meta_workers =
2351                 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2352                                       max_active, 4);
2353         fs_info->endio_meta_write_workers =
2354                 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2355                                       max_active, 2);
2356         fs_info->endio_raid56_workers =
2357                 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2358                                       max_active, 4);
2359         fs_info->endio_repair_workers =
2360                 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2361         fs_info->rmw_workers =
2362                 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2363         fs_info->endio_write_workers =
2364                 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2365                                       max_active, 2);
2366         fs_info->endio_freespace_worker =
2367                 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2368                                       max_active, 0);
2369         fs_info->delayed_workers =
2370                 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2371                                       max_active, 0);
2372         fs_info->readahead_workers =
2373                 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2374                                       max_active, 2);
2375         fs_info->qgroup_rescan_workers =
2376                 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2377         fs_info->extent_workers =
2378                 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2379                                       min_t(u64, fs_devices->num_devices,
2380                                             max_active), 8);
2381
2382         if (!(fs_info->workers && fs_info->delalloc_workers &&
2383               fs_info->submit_workers && fs_info->flush_workers &&
2384               fs_info->endio_workers && fs_info->endio_meta_workers &&
2385               fs_info->endio_meta_write_workers &&
2386               fs_info->endio_repair_workers &&
2387               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2388               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2389               fs_info->caching_workers && fs_info->readahead_workers &&
2390               fs_info->fixup_workers && fs_info->delayed_workers &&
2391               fs_info->extent_workers &&
2392               fs_info->qgroup_rescan_workers)) {
2393                 return -ENOMEM;
2394         }
2395
2396         return 0;
2397 }
2398
2399 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2400                             struct btrfs_fs_devices *fs_devices)
2401 {
2402         int ret;
2403         struct btrfs_root *tree_root = fs_info->tree_root;
2404         struct btrfs_root *log_tree_root;
2405         struct btrfs_super_block *disk_super = fs_info->super_copy;
2406         u64 bytenr = btrfs_super_log_root(disk_super);
2407
2408         if (fs_devices->rw_devices == 0) {
2409                 btrfs_warn(fs_info, "log replay required on RO media");
2410                 return -EIO;
2411         }
2412
2413         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2414         if (!log_tree_root)
2415                 return -ENOMEM;
2416
2417         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2418                         tree_root->stripesize, log_tree_root, fs_info,
2419                         BTRFS_TREE_LOG_OBJECTID);
2420
2421         log_tree_root->node = read_tree_block(tree_root, bytenr,
2422                         fs_info->generation + 1);
2423         if (IS_ERR(log_tree_root->node)) {
2424                 btrfs_warn(fs_info, "failed to read log tree");
2425                 ret = PTR_ERR(log_tree_root->node);
2426                 kfree(log_tree_root);
2427                 return ret;
2428         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2429                 btrfs_err(fs_info, "failed to read log tree");
2430                 free_extent_buffer(log_tree_root->node);
2431                 kfree(log_tree_root);
2432                 return -EIO;
2433         }
2434         /* returns with log_tree_root freed on success */
2435         ret = btrfs_recover_log_trees(log_tree_root);
2436         if (ret) {
2437                 btrfs_handle_fs_error(tree_root->fs_info, ret,
2438                             "Failed to recover log tree");
2439                 free_extent_buffer(log_tree_root->node);
2440                 kfree(log_tree_root);
2441                 return ret;
2442         }
2443
2444         if (fs_info->sb->s_flags & MS_RDONLY) {
2445                 ret = btrfs_commit_super(tree_root);
2446                 if (ret)
2447                         return ret;
2448         }
2449
2450         return 0;
2451 }
2452
2453 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2454                             struct btrfs_root *tree_root)
2455 {
2456         struct btrfs_root *root;
2457         struct btrfs_key location;
2458         int ret;
2459
2460         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2461         location.type = BTRFS_ROOT_ITEM_KEY;
2462         location.offset = 0;
2463
2464         root = btrfs_read_tree_root(tree_root, &location);
2465         if (IS_ERR(root))
2466                 return PTR_ERR(root);
2467         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2468         fs_info->extent_root = root;
2469
2470         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2471         root = btrfs_read_tree_root(tree_root, &location);
2472         if (IS_ERR(root))
2473                 return PTR_ERR(root);
2474         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2475         fs_info->dev_root = root;
2476         btrfs_init_devices_late(fs_info);
2477
2478         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2479         root = btrfs_read_tree_root(tree_root, &location);
2480         if (IS_ERR(root))
2481                 return PTR_ERR(root);
2482         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2483         fs_info->csum_root = root;
2484
2485         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2486         root = btrfs_read_tree_root(tree_root, &location);
2487         if (!IS_ERR(root)) {
2488                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2489                 fs_info->quota_enabled = 1;
2490                 fs_info->pending_quota_state = 1;
2491                 fs_info->quota_root = root;
2492         }
2493
2494         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2495         root = btrfs_read_tree_root(tree_root, &location);
2496         if (IS_ERR(root)) {
2497                 ret = PTR_ERR(root);
2498                 if (ret != -ENOENT)
2499                         return ret;
2500         } else {
2501                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2502                 fs_info->uuid_root = root;
2503         }
2504
2505         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2506                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2507                 root = btrfs_read_tree_root(tree_root, &location);
2508                 if (IS_ERR(root))
2509                         return PTR_ERR(root);
2510                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2511                 fs_info->free_space_root = root;
2512         }
2513
2514         return 0;
2515 }
2516
2517 int open_ctree(struct super_block *sb,
2518                struct btrfs_fs_devices *fs_devices,
2519                char *options)
2520 {
2521         u32 sectorsize;
2522         u32 nodesize;
2523         u32 stripesize;
2524         u64 generation;
2525         u64 features;
2526         struct btrfs_key location;
2527         struct buffer_head *bh;
2528         struct btrfs_super_block *disk_super;
2529         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2530         struct btrfs_root *tree_root;
2531         struct btrfs_root *chunk_root;
2532         int ret;
2533         int err = -EINVAL;
2534         int num_backups_tried = 0;
2535         int backup_index = 0;
2536         int max_active;
2537
2538         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2539         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2540         if (!tree_root || !chunk_root) {
2541                 err = -ENOMEM;
2542                 goto fail;
2543         }
2544
2545         ret = init_srcu_struct(&fs_info->subvol_srcu);
2546         if (ret) {
2547                 err = ret;
2548                 goto fail;
2549         }
2550
2551         ret = setup_bdi(fs_info, &fs_info->bdi);
2552         if (ret) {
2553                 err = ret;
2554                 goto fail_srcu;
2555         }
2556
2557         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2558         if (ret) {
2559                 err = ret;
2560                 goto fail_bdi;
2561         }
2562         fs_info->dirty_metadata_batch = PAGE_SIZE *
2563                                         (1 + ilog2(nr_cpu_ids));
2564
2565         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2566         if (ret) {
2567                 err = ret;
2568                 goto fail_dirty_metadata_bytes;
2569         }
2570
2571         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2572         if (ret) {
2573                 err = ret;
2574                 goto fail_delalloc_bytes;
2575         }
2576
2577         fs_info->btree_inode = new_inode(sb);
2578         if (!fs_info->btree_inode) {
2579                 err = -ENOMEM;
2580                 goto fail_bio_counter;
2581         }
2582
2583         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2584
2585         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2586         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2587         INIT_LIST_HEAD(&fs_info->trans_list);
2588         INIT_LIST_HEAD(&fs_info->dead_roots);
2589         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2590         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2591         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2592         spin_lock_init(&fs_info->delalloc_root_lock);
2593         spin_lock_init(&fs_info->trans_lock);
2594         spin_lock_init(&fs_info->fs_roots_radix_lock);
2595         spin_lock_init(&fs_info->delayed_iput_lock);
2596         spin_lock_init(&fs_info->defrag_inodes_lock);
2597         spin_lock_init(&fs_info->free_chunk_lock);
2598         spin_lock_init(&fs_info->tree_mod_seq_lock);
2599         spin_lock_init(&fs_info->super_lock);
2600         spin_lock_init(&fs_info->qgroup_op_lock);
2601         spin_lock_init(&fs_info->buffer_lock);
2602         spin_lock_init(&fs_info->unused_bgs_lock);
2603         rwlock_init(&fs_info->tree_mod_log_lock);
2604         mutex_init(&fs_info->unused_bg_unpin_mutex);
2605         mutex_init(&fs_info->delete_unused_bgs_mutex);
2606         mutex_init(&fs_info->reloc_mutex);
2607         mutex_init(&fs_info->delalloc_root_mutex);
2608         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2609         seqlock_init(&fs_info->profiles_lock);
2610
2611         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2612         INIT_LIST_HEAD(&fs_info->space_info);
2613         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2614         INIT_LIST_HEAD(&fs_info->unused_bgs);
2615         btrfs_mapping_init(&fs_info->mapping_tree);
2616         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2617                              BTRFS_BLOCK_RSV_GLOBAL);
2618         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2619                              BTRFS_BLOCK_RSV_DELALLOC);
2620         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2621         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2622         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2623         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2624                              BTRFS_BLOCK_RSV_DELOPS);
2625         atomic_set(&fs_info->nr_async_submits, 0);
2626         atomic_set(&fs_info->async_delalloc_pages, 0);
2627         atomic_set(&fs_info->async_submit_draining, 0);
2628         atomic_set(&fs_info->nr_async_bios, 0);
2629         atomic_set(&fs_info->defrag_running, 0);
2630         atomic_set(&fs_info->qgroup_op_seq, 0);
2631         atomic_set(&fs_info->reada_works_cnt, 0);
2632         atomic64_set(&fs_info->tree_mod_seq, 0);
2633         fs_info->sb = sb;
2634         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2635         fs_info->metadata_ratio = 0;
2636         fs_info->defrag_inodes = RB_ROOT;
2637         fs_info->free_chunk_space = 0;
2638         fs_info->tree_mod_log = RB_ROOT;
2639         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2640         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2641         /* readahead state */
2642         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2643         spin_lock_init(&fs_info->reada_lock);
2644
2645         fs_info->thread_pool_size = min_t(unsigned long,
2646                                           num_online_cpus() + 2, 8);
2647
2648         INIT_LIST_HEAD(&fs_info->ordered_roots);
2649         spin_lock_init(&fs_info->ordered_root_lock);
2650         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2651                                         GFP_KERNEL);
2652         if (!fs_info->delayed_root) {
2653                 err = -ENOMEM;
2654                 goto fail_iput;
2655         }
2656         btrfs_init_delayed_root(fs_info->delayed_root);
2657
2658         btrfs_init_scrub(fs_info);
2659 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2660         fs_info->check_integrity_print_mask = 0;
2661 #endif
2662         btrfs_init_balance(fs_info);
2663         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2664
2665         sb->s_blocksize = 4096;
2666         sb->s_blocksize_bits = blksize_bits(4096);
2667         sb->s_bdi = &fs_info->bdi;
2668
2669         btrfs_init_btree_inode(fs_info, tree_root);
2670
2671         spin_lock_init(&fs_info->block_group_cache_lock);
2672         fs_info->block_group_cache_tree = RB_ROOT;
2673         fs_info->first_logical_byte = (u64)-1;
2674
2675         extent_io_tree_init(&fs_info->freed_extents[0],
2676                              fs_info->btree_inode->i_mapping);
2677         extent_io_tree_init(&fs_info->freed_extents[1],
2678                              fs_info->btree_inode->i_mapping);
2679         fs_info->pinned_extents = &fs_info->freed_extents[0];
2680         fs_info->do_barriers = 1;
2681
2682
2683         mutex_init(&fs_info->ordered_operations_mutex);
2684         mutex_init(&fs_info->tree_log_mutex);
2685         mutex_init(&fs_info->chunk_mutex);
2686         mutex_init(&fs_info->transaction_kthread_mutex);
2687         mutex_init(&fs_info->cleaner_mutex);
2688         mutex_init(&fs_info->volume_mutex);
2689         mutex_init(&fs_info->ro_block_group_mutex);
2690         init_rwsem(&fs_info->commit_root_sem);
2691         init_rwsem(&fs_info->cleanup_work_sem);
2692         init_rwsem(&fs_info->subvol_sem);
2693         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2694
2695         btrfs_init_dev_replace_locks(fs_info);
2696         btrfs_init_qgroup(fs_info);
2697
2698         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2699         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2700
2701         init_waitqueue_head(&fs_info->transaction_throttle);
2702         init_waitqueue_head(&fs_info->transaction_wait);
2703         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2704         init_waitqueue_head(&fs_info->async_submit_wait);
2705
2706         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2707
2708         ret = btrfs_alloc_stripe_hash_table(fs_info);
2709         if (ret) {
2710                 err = ret;
2711                 goto fail_alloc;
2712         }
2713
2714         __setup_root(4096, 4096, 4096, tree_root,
2715                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2716
2717         invalidate_bdev(fs_devices->latest_bdev);
2718
2719         /*
2720          * Read super block and check the signature bytes only
2721          */
2722         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2723         if (IS_ERR(bh)) {
2724                 err = PTR_ERR(bh);
2725                 goto fail_alloc;
2726         }
2727
2728         /*
2729          * We want to check superblock checksum, the type is stored inside.
2730          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2731          */
2732         if (btrfs_check_super_csum(bh->b_data)) {
2733                 btrfs_err(fs_info, "superblock checksum mismatch");
2734                 err = -EINVAL;
2735                 brelse(bh);
2736                 goto fail_alloc;
2737         }
2738
2739         /*
2740          * super_copy is zeroed at allocation time and we never touch the
2741          * following bytes up to INFO_SIZE, the checksum is calculated from
2742          * the whole block of INFO_SIZE
2743          */
2744         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2745         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2746                sizeof(*fs_info->super_for_commit));
2747         brelse(bh);
2748
2749         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2750
2751         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2752         if (ret) {
2753                 btrfs_err(fs_info, "superblock contains fatal errors");
2754                 err = -EINVAL;
2755                 goto fail_alloc;
2756         }
2757
2758         disk_super = fs_info->super_copy;
2759         if (!btrfs_super_root(disk_super))
2760                 goto fail_alloc;
2761
2762         /* check FS state, whether FS is broken. */
2763         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2764                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2765
2766         /*
2767          * run through our array of backup supers and setup
2768          * our ring pointer to the oldest one
2769          */
2770         generation = btrfs_super_generation(disk_super);
2771         find_oldest_super_backup(fs_info, generation);
2772
2773         /*
2774          * In the long term, we'll store the compression type in the super
2775          * block, and it'll be used for per file compression control.
2776          */
2777         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2778
2779         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2780         if (ret) {
2781                 err = ret;
2782                 goto fail_alloc;
2783         }
2784
2785         features = btrfs_super_incompat_flags(disk_super) &
2786                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2787         if (features) {
2788                 btrfs_err(fs_info,
2789                     "cannot mount because of unsupported optional features (%llx)",
2790                     features);
2791                 err = -EINVAL;
2792                 goto fail_alloc;
2793         }
2794
2795         features = btrfs_super_incompat_flags(disk_super);
2796         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2797         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2798                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2799
2800         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2801                 btrfs_info(fs_info, "has skinny extents");
2802
2803         /*
2804          * flag our filesystem as having big metadata blocks if
2805          * they are bigger than the page size
2806          */
2807         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2808                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2809                         btrfs_info(fs_info,
2810                                 "flagging fs with big metadata feature");
2811                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2812         }
2813
2814         nodesize = btrfs_super_nodesize(disk_super);
2815         sectorsize = btrfs_super_sectorsize(disk_super);
2816         stripesize = sectorsize;
2817         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2818         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2819
2820         /*
2821          * mixed block groups end up with duplicate but slightly offset
2822          * extent buffers for the same range.  It leads to corruptions
2823          */
2824         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2825             (sectorsize != nodesize)) {
2826                 btrfs_err(fs_info,
2827 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2828                         nodesize, sectorsize);
2829                 goto fail_alloc;
2830         }
2831
2832         /*
2833          * Needn't use the lock because there is no other task which will
2834          * update the flag.
2835          */
2836         btrfs_set_super_incompat_flags(disk_super, features);
2837
2838         features = btrfs_super_compat_ro_flags(disk_super) &
2839                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2840         if (!(sb->s_flags & MS_RDONLY) && features) {
2841                 btrfs_err(fs_info,
2842         "cannot mount read-write because of unsupported optional features (%llx)",
2843                        features);
2844                 err = -EINVAL;
2845                 goto fail_alloc;
2846         }
2847
2848         max_active = fs_info->thread_pool_size;
2849
2850         ret = btrfs_init_workqueues(fs_info, fs_devices);
2851         if (ret) {
2852                 err = ret;
2853                 goto fail_sb_buffer;
2854         }
2855
2856         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2857         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2858                                     SZ_4M / PAGE_SIZE);
2859
2860         tree_root->nodesize = nodesize;
2861         tree_root->sectorsize = sectorsize;
2862         tree_root->stripesize = stripesize;
2863
2864         sb->s_blocksize = sectorsize;
2865         sb->s_blocksize_bits = blksize_bits(sectorsize);
2866
2867         mutex_lock(&fs_info->chunk_mutex);
2868         ret = btrfs_read_sys_array(tree_root);
2869         mutex_unlock(&fs_info->chunk_mutex);
2870         if (ret) {
2871                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2872                 goto fail_sb_buffer;
2873         }
2874
2875         generation = btrfs_super_chunk_root_generation(disk_super);
2876
2877         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2878                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2879
2880         chunk_root->node = read_tree_block(chunk_root,
2881                                            btrfs_super_chunk_root(disk_super),
2882                                            generation);
2883         if (IS_ERR(chunk_root->node) ||
2884             !extent_buffer_uptodate(chunk_root->node)) {
2885                 btrfs_err(fs_info, "failed to read chunk root");
2886                 if (!IS_ERR(chunk_root->node))
2887                         free_extent_buffer(chunk_root->node);
2888                 chunk_root->node = NULL;
2889                 goto fail_tree_roots;
2890         }
2891         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2892         chunk_root->commit_root = btrfs_root_node(chunk_root);
2893
2894         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2895            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2896
2897         ret = btrfs_read_chunk_tree(chunk_root);
2898         if (ret) {
2899                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2900                 goto fail_tree_roots;
2901         }
2902
2903         /*
2904          * keep the device that is marked to be the target device for the
2905          * dev_replace procedure
2906          */
2907         btrfs_close_extra_devices(fs_devices, 0);
2908
2909         if (!fs_devices->latest_bdev) {
2910                 btrfs_err(fs_info, "failed to read devices");
2911                 goto fail_tree_roots;
2912         }
2913
2914 retry_root_backup:
2915         generation = btrfs_super_generation(disk_super);
2916
2917         tree_root->node = read_tree_block(tree_root,
2918                                           btrfs_super_root(disk_super),
2919                                           generation);
2920         if (IS_ERR(tree_root->node) ||
2921             !extent_buffer_uptodate(tree_root->node)) {
2922                 btrfs_warn(fs_info, "failed to read tree root");
2923                 if (!IS_ERR(tree_root->node))
2924                         free_extent_buffer(tree_root->node);
2925                 tree_root->node = NULL;
2926                 goto recovery_tree_root;
2927         }
2928
2929         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2930         tree_root->commit_root = btrfs_root_node(tree_root);
2931         btrfs_set_root_refs(&tree_root->root_item, 1);
2932
2933         mutex_lock(&tree_root->objectid_mutex);
2934         ret = btrfs_find_highest_objectid(tree_root,
2935                                         &tree_root->highest_objectid);
2936         if (ret) {
2937                 mutex_unlock(&tree_root->objectid_mutex);
2938                 goto recovery_tree_root;
2939         }
2940
2941         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2942
2943         mutex_unlock(&tree_root->objectid_mutex);
2944
2945         ret = btrfs_read_roots(fs_info, tree_root);
2946         if (ret)
2947                 goto recovery_tree_root;
2948
2949         fs_info->generation = generation;
2950         fs_info->last_trans_committed = generation;
2951
2952         ret = btrfs_recover_balance(fs_info);
2953         if (ret) {
2954                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2955                 goto fail_block_groups;
2956         }
2957
2958         ret = btrfs_init_dev_stats(fs_info);
2959         if (ret) {
2960                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2961                 goto fail_block_groups;
2962         }
2963
2964         ret = btrfs_init_dev_replace(fs_info);
2965         if (ret) {
2966                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2967                 goto fail_block_groups;
2968         }
2969
2970         btrfs_close_extra_devices(fs_devices, 1);
2971
2972         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2973         if (ret) {
2974                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2975                                 ret);
2976                 goto fail_block_groups;
2977         }
2978
2979         ret = btrfs_sysfs_add_device(fs_devices);
2980         if (ret) {
2981                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2982                                 ret);
2983                 goto fail_fsdev_sysfs;
2984         }
2985
2986         ret = btrfs_sysfs_add_mounted(fs_info);
2987         if (ret) {
2988                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2989                 goto fail_fsdev_sysfs;
2990         }
2991
2992         ret = btrfs_init_space_info(fs_info);
2993         if (ret) {
2994                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2995                 goto fail_sysfs;
2996         }
2997
2998         ret = btrfs_read_block_groups(fs_info->extent_root);
2999         if (ret) {
3000                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3001                 goto fail_sysfs;
3002         }
3003         fs_info->num_tolerated_disk_barrier_failures =
3004                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3005         if (fs_info->fs_devices->missing_devices >
3006              fs_info->num_tolerated_disk_barrier_failures &&
3007             !(sb->s_flags & MS_RDONLY)) {
3008                 btrfs_warn(fs_info,
3009 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3010                         fs_info->fs_devices->missing_devices,
3011                         fs_info->num_tolerated_disk_barrier_failures);
3012                 goto fail_sysfs;
3013         }
3014
3015         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3016                                                "btrfs-cleaner");
3017         if (IS_ERR(fs_info->cleaner_kthread))
3018                 goto fail_sysfs;
3019
3020         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3021                                                    tree_root,
3022                                                    "btrfs-transaction");
3023         if (IS_ERR(fs_info->transaction_kthread))
3024                 goto fail_cleaner;
3025
3026         if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3027             !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3028             !fs_info->fs_devices->rotating) {
3029                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3030                 btrfs_set_opt(fs_info->mount_opt, SSD);
3031         }
3032
3033         /*
3034          * Mount does not set all options immediately, we can do it now and do
3035          * not have to wait for transaction commit
3036          */
3037         btrfs_apply_pending_changes(fs_info);
3038
3039 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3040         if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3041                 ret = btrfsic_mount(tree_root, fs_devices,
3042                                     btrfs_test_opt(tree_root->fs_info,
3043                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3044                                     1 : 0,
3045                                     fs_info->check_integrity_print_mask);
3046                 if (ret)
3047                         btrfs_warn(fs_info,
3048                                 "failed to initialize integrity check module: %d",
3049                                 ret);
3050         }
3051 #endif
3052         ret = btrfs_read_qgroup_config(fs_info);
3053         if (ret)
3054                 goto fail_trans_kthread;
3055
3056         /* do not make disk changes in broken FS or nologreplay is given */
3057         if (btrfs_super_log_root(disk_super) != 0 &&
3058             !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3059                 ret = btrfs_replay_log(fs_info, fs_devices);
3060                 if (ret) {
3061                         err = ret;
3062                         goto fail_qgroup;
3063                 }
3064         }
3065
3066         ret = btrfs_find_orphan_roots(tree_root);
3067         if (ret)
3068                 goto fail_qgroup;
3069
3070         if (!(sb->s_flags & MS_RDONLY)) {
3071                 ret = btrfs_cleanup_fs_roots(fs_info);
3072                 if (ret)
3073                         goto fail_qgroup;
3074
3075                 mutex_lock(&fs_info->cleaner_mutex);
3076                 ret = btrfs_recover_relocation(tree_root);
3077                 mutex_unlock(&fs_info->cleaner_mutex);
3078                 if (ret < 0) {
3079                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3080                                         ret);
3081                         err = -EINVAL;
3082                         goto fail_qgroup;
3083                 }
3084         }
3085
3086         location.objectid = BTRFS_FS_TREE_OBJECTID;
3087         location.type = BTRFS_ROOT_ITEM_KEY;
3088         location.offset = 0;
3089
3090         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3091         if (IS_ERR(fs_info->fs_root)) {
3092                 err = PTR_ERR(fs_info->fs_root);
3093                 goto fail_qgroup;
3094         }
3095
3096         if (sb->s_flags & MS_RDONLY)
3097                 return 0;
3098
3099         if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3100             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3101                 btrfs_info(fs_info, "creating free space tree");
3102                 ret = btrfs_create_free_space_tree(fs_info);
3103                 if (ret) {
3104                         btrfs_warn(fs_info,
3105                                 "failed to create free space tree: %d", ret);
3106                         close_ctree(tree_root);
3107                         return ret;
3108                 }
3109         }
3110
3111         down_read(&fs_info->cleanup_work_sem);
3112         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3113             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3114                 up_read(&fs_info->cleanup_work_sem);
3115                 close_ctree(tree_root);
3116                 return ret;
3117         }
3118         up_read(&fs_info->cleanup_work_sem);
3119
3120         ret = btrfs_resume_balance_async(fs_info);
3121         if (ret) {
3122                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3123                 close_ctree(tree_root);
3124                 return ret;
3125         }
3126
3127         ret = btrfs_resume_dev_replace_async(fs_info);
3128         if (ret) {
3129                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3130                 close_ctree(tree_root);
3131                 return ret;
3132         }
3133
3134         btrfs_qgroup_rescan_resume(fs_info);
3135
3136         if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3137             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3138                 btrfs_info(fs_info, "clearing free space tree");
3139                 ret = btrfs_clear_free_space_tree(fs_info);
3140                 if (ret) {
3141                         btrfs_warn(fs_info,
3142                                 "failed to clear free space tree: %d", ret);
3143                         close_ctree(tree_root);
3144                         return ret;
3145                 }
3146         }
3147
3148         if (!fs_info->uuid_root) {
3149                 btrfs_info(fs_info, "creating UUID tree");
3150                 ret = btrfs_create_uuid_tree(fs_info);
3151                 if (ret) {
3152                         btrfs_warn(fs_info,
3153                                 "failed to create the UUID tree: %d", ret);
3154                         close_ctree(tree_root);
3155                         return ret;
3156                 }
3157         } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3158                    fs_info->generation !=
3159                                 btrfs_super_uuid_tree_generation(disk_super)) {
3160                 btrfs_info(fs_info, "checking UUID tree");
3161                 ret = btrfs_check_uuid_tree(fs_info);
3162                 if (ret) {
3163                         btrfs_warn(fs_info,
3164                                 "failed to check the UUID tree: %d", ret);
3165                         close_ctree(tree_root);
3166                         return ret;
3167                 }
3168         } else {
3169                 fs_info->update_uuid_tree_gen = 1;
3170         }
3171
3172         fs_info->open = 1;
3173
3174         /*
3175          * backuproot only affect mount behavior, and if open_ctree succeeded,
3176          * no need to keep the flag
3177          */
3178         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3179
3180         return 0;
3181
3182 fail_qgroup:
3183         btrfs_free_qgroup_config(fs_info);
3184 fail_trans_kthread:
3185         kthread_stop(fs_info->transaction_kthread);
3186         btrfs_cleanup_transaction(fs_info->tree_root);
3187         btrfs_free_fs_roots(fs_info);
3188 fail_cleaner:
3189         kthread_stop(fs_info->cleaner_kthread);
3190
3191         /*
3192          * make sure we're done with the btree inode before we stop our
3193          * kthreads
3194          */
3195         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3196
3197 fail_sysfs:
3198         btrfs_sysfs_remove_mounted(fs_info);
3199
3200 fail_fsdev_sysfs:
3201         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3202
3203 fail_block_groups:
3204         btrfs_put_block_group_cache(fs_info);
3205         btrfs_free_block_groups(fs_info);
3206
3207 fail_tree_roots:
3208         free_root_pointers(fs_info, 1);
3209         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3210
3211 fail_sb_buffer:
3212         btrfs_stop_all_workers(fs_info);
3213 fail_alloc:
3214 fail_iput:
3215         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3216
3217         iput(fs_info->btree_inode);
3218 fail_bio_counter:
3219         percpu_counter_destroy(&fs_info->bio_counter);
3220 fail_delalloc_bytes:
3221         percpu_counter_destroy(&fs_info->delalloc_bytes);
3222 fail_dirty_metadata_bytes:
3223         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3224 fail_bdi:
3225         bdi_destroy(&fs_info->bdi);
3226 fail_srcu:
3227         cleanup_srcu_struct(&fs_info->subvol_srcu);
3228 fail:
3229         btrfs_free_stripe_hash_table(fs_info);
3230         btrfs_close_devices(fs_info->fs_devices);
3231         return err;
3232
3233 recovery_tree_root:
3234         if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3235                 goto fail_tree_roots;
3236
3237         free_root_pointers(fs_info, 0);
3238
3239         /* don't use the log in recovery mode, it won't be valid */
3240         btrfs_set_super_log_root(disk_super, 0);
3241
3242         /* we can't trust the free space cache either */
3243         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3244
3245         ret = next_root_backup(fs_info, fs_info->super_copy,
3246                                &num_backups_tried, &backup_index);
3247         if (ret == -1)
3248                 goto fail_block_groups;
3249         goto retry_root_backup;
3250 }
3251
3252 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3253 {
3254         if (uptodate) {
3255                 set_buffer_uptodate(bh);
3256         } else {
3257                 struct btrfs_device *device = (struct btrfs_device *)
3258                         bh->b_private;
3259
3260                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3261                                 "lost page write due to IO error on %s",
3262                                           rcu_str_deref(device->name));
3263                 /* note, we don't set_buffer_write_io_error because we have
3264                  * our own ways of dealing with the IO errors
3265                  */