Btrfs: self-tests: Support non-4k page size
[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_RECLAIM_ACCOUNT | 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 (!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 (!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_test_is_dummy_root(root))
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 (!buf)
1176                 return ERR_PTR(-ENOMEM);
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         root->node = NULL;
1237         root->commit_root = NULL;
1238         root->sectorsize = sectorsize;
1239         root->nodesize = nodesize;
1240         root->stripesize = stripesize;
1241         root->state = 0;
1242         root->orphan_cleanup_state = 0;
1243
1244         root->objectid = objectid;
1245         root->last_trans = 0;
1246         root->highest_objectid = 0;
1247         root->nr_delalloc_inodes = 0;
1248         root->nr_ordered_extents = 0;
1249         root->name = NULL;
1250         root->inode_tree = RB_ROOT;
1251         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1252         root->block_rsv = NULL;
1253         root->orphan_block_rsv = NULL;
1254
1255         INIT_LIST_HEAD(&root->dirty_list);
1256         INIT_LIST_HEAD(&root->root_list);
1257         INIT_LIST_HEAD(&root->delalloc_inodes);
1258         INIT_LIST_HEAD(&root->delalloc_root);
1259         INIT_LIST_HEAD(&root->ordered_extents);
1260         INIT_LIST_HEAD(&root->ordered_root);
1261         INIT_LIST_HEAD(&root->logged_list[0]);
1262         INIT_LIST_HEAD(&root->logged_list[1]);
1263         spin_lock_init(&root->orphan_lock);
1264         spin_lock_init(&root->inode_lock);
1265         spin_lock_init(&root->delalloc_lock);
1266         spin_lock_init(&root->ordered_extent_lock);
1267         spin_lock_init(&root->accounting_lock);
1268         spin_lock_init(&root->log_extents_lock[0]);
1269         spin_lock_init(&root->log_extents_lock[1]);
1270         mutex_init(&root->objectid_mutex);
1271         mutex_init(&root->log_mutex);
1272         mutex_init(&root->ordered_extent_mutex);
1273         mutex_init(&root->delalloc_mutex);
1274         init_waitqueue_head(&root->log_writer_wait);
1275         init_waitqueue_head(&root->log_commit_wait[0]);
1276         init_waitqueue_head(&root->log_commit_wait[1]);
1277         INIT_LIST_HEAD(&root->log_ctxs[0]);
1278         INIT_LIST_HEAD(&root->log_ctxs[1]);
1279         atomic_set(&root->log_commit[0], 0);
1280         atomic_set(&root->log_commit[1], 0);
1281         atomic_set(&root->log_writers, 0);
1282         atomic_set(&root->log_batch, 0);
1283         atomic_set(&root->orphan_inodes, 0);
1284         atomic_set(&root->refs, 1);
1285         atomic_set(&root->will_be_snapshoted, 0);
1286         atomic_set(&root->qgroup_meta_rsv, 0);
1287         root->log_transid = 0;
1288         root->log_transid_committed = -1;
1289         root->last_log_commit = 0;
1290         if (fs_info)
1291                 extent_io_tree_init(&root->dirty_log_pages,
1292                                      fs_info->btree_inode->i_mapping);
1293
1294         memset(&root->root_key, 0, sizeof(root->root_key));
1295         memset(&root->root_item, 0, sizeof(root->root_item));
1296         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1297         if (fs_info)
1298                 root->defrag_trans_start = fs_info->generation;
1299         else
1300                 root->defrag_trans_start = 0;
1301         root->root_key.objectid = objectid;
1302         root->anon_dev = 0;
1303
1304         spin_lock_init(&root->root_item_lock);
1305 }
1306
1307 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1308                 gfp_t flags)
1309 {
1310         struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1311         if (root)
1312                 root->fs_info = fs_info;
1313         return root;
1314 }
1315
1316 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1317 /* Should only be used by the testing infrastructure */
1318 struct btrfs_root *btrfs_alloc_dummy_root(u32 sectorsize, u32 nodesize)
1319 {
1320         struct btrfs_root *root;
1321
1322         root = btrfs_alloc_root(NULL, GFP_KERNEL);
1323         if (!root)
1324                 return ERR_PTR(-ENOMEM);
1325         /* We don't use the stripesize in selftest, set it as sectorsize */
1326         __setup_root(nodesize, sectorsize, sectorsize, root, NULL, 1);
1327         set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1328         root->alloc_bytenr = 0;
1329
1330         return root;
1331 }
1332 #endif
1333
1334 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1335                                      struct btrfs_fs_info *fs_info,
1336                                      u64 objectid)
1337 {
1338         struct extent_buffer *leaf;
1339         struct btrfs_root *tree_root = fs_info->tree_root;
1340         struct btrfs_root *root;
1341         struct btrfs_key key;
1342         int ret = 0;
1343         uuid_le uuid;
1344
1345         root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1346         if (!root)
1347                 return ERR_PTR(-ENOMEM);
1348
1349         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1350                 tree_root->stripesize, root, fs_info, objectid);
1351         root->root_key.objectid = objectid;
1352         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1353         root->root_key.offset = 0;
1354
1355         leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1356         if (IS_ERR(leaf)) {
1357                 ret = PTR_ERR(leaf);
1358                 leaf = NULL;
1359                 goto fail;
1360         }
1361
1362         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1363         btrfs_set_header_bytenr(leaf, leaf->start);
1364         btrfs_set_header_generation(leaf, trans->transid);
1365         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1366         btrfs_set_header_owner(leaf, objectid);
1367         root->node = leaf;
1368
1369         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1370                             BTRFS_FSID_SIZE);
1371         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1372                             btrfs_header_chunk_tree_uuid(leaf),
1373                             BTRFS_UUID_SIZE);
1374         btrfs_mark_buffer_dirty(leaf);
1375
1376         root->commit_root = btrfs_root_node(root);
1377         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1378
1379         root->root_item.flags = 0;
1380         root->root_item.byte_limit = 0;
1381         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1382         btrfs_set_root_generation(&root->root_item, trans->transid);
1383         btrfs_set_root_level(&root->root_item, 0);
1384         btrfs_set_root_refs(&root->root_item, 1);
1385         btrfs_set_root_used(&root->root_item, leaf->len);
1386         btrfs_set_root_last_snapshot(&root->root_item, 0);
1387         btrfs_set_root_dirid(&root->root_item, 0);
1388         uuid_le_gen(&uuid);
1389         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1390         root->root_item.drop_level = 0;
1391
1392         key.objectid = objectid;
1393         key.type = BTRFS_ROOT_ITEM_KEY;
1394         key.offset = 0;
1395         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1396         if (ret)
1397                 goto fail;
1398
1399         btrfs_tree_unlock(leaf);
1400
1401         return root;
1402
1403 fail:
1404         if (leaf) {
1405                 btrfs_tree_unlock(leaf);
1406                 free_extent_buffer(root->commit_root);
1407                 free_extent_buffer(leaf);
1408         }
1409         kfree(root);
1410
1411         return ERR_PTR(ret);
1412 }
1413
1414 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1415                                          struct btrfs_fs_info *fs_info)
1416 {
1417         struct btrfs_root *root;
1418         struct btrfs_root *tree_root = fs_info->tree_root;
1419         struct extent_buffer *leaf;
1420
1421         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1422         if (!root)
1423                 return ERR_PTR(-ENOMEM);
1424
1425         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1426                      tree_root->stripesize, root, fs_info,
1427                      BTRFS_TREE_LOG_OBJECTID);
1428
1429         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1430         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1431         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1432
1433         /*
1434          * DON'T set REF_COWS for log trees
1435          *
1436          * log trees do not get reference counted because they go away
1437          * before a real commit is actually done.  They do store pointers
1438          * to file data extents, and those reference counts still get
1439          * updated (along with back refs to the log tree).
1440          */
1441
1442         leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1443                         NULL, 0, 0, 0);
1444         if (IS_ERR(leaf)) {
1445                 kfree(root);
1446                 return ERR_CAST(leaf);
1447         }
1448
1449         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1450         btrfs_set_header_bytenr(leaf, leaf->start);
1451         btrfs_set_header_generation(leaf, trans->transid);
1452         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1453         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1454         root->node = leaf;
1455
1456         write_extent_buffer(root->node, root->fs_info->fsid,
1457                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1458         btrfs_mark_buffer_dirty(root->node);
1459         btrfs_tree_unlock(root->node);
1460         return root;
1461 }
1462
1463 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1464                              struct btrfs_fs_info *fs_info)
1465 {
1466         struct btrfs_root *log_root;
1467
1468         log_root = alloc_log_tree(trans, fs_info);
1469         if (IS_ERR(log_root))
1470                 return PTR_ERR(log_root);
1471         WARN_ON(fs_info->log_root_tree);
1472         fs_info->log_root_tree = log_root;
1473         return 0;
1474 }
1475
1476 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1477                        struct btrfs_root *root)
1478 {
1479         struct btrfs_root *log_root;
1480         struct btrfs_inode_item *inode_item;
1481
1482         log_root = alloc_log_tree(trans, root->fs_info);
1483         if (IS_ERR(log_root))
1484                 return PTR_ERR(log_root);
1485
1486         log_root->last_trans = trans->transid;
1487         log_root->root_key.offset = root->root_key.objectid;
1488
1489         inode_item = &log_root->root_item.inode;
1490         btrfs_set_stack_inode_generation(inode_item, 1);
1491         btrfs_set_stack_inode_size(inode_item, 3);
1492         btrfs_set_stack_inode_nlink(inode_item, 1);
1493         btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1494         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1495
1496         btrfs_set_root_node(&log_root->root_item, log_root->node);
1497
1498         WARN_ON(root->log_root);
1499         root->log_root = log_root;
1500         root->log_transid = 0;
1501         root->log_transid_committed = -1;
1502         root->last_log_commit = 0;
1503         return 0;
1504 }
1505
1506 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1507                                                struct btrfs_key *key)
1508 {
1509         struct btrfs_root *root;
1510         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1511         struct btrfs_path *path;
1512         u64 generation;
1513         int ret;
1514
1515         path = btrfs_alloc_path();
1516         if (!path)
1517                 return ERR_PTR(-ENOMEM);
1518
1519         root = btrfs_alloc_root(fs_info, GFP_NOFS);
1520         if (!root) {
1521                 ret = -ENOMEM;
1522                 goto alloc_fail;
1523         }
1524
1525         __setup_root(tree_root->nodesize, tree_root->sectorsize,
1526                 tree_root->stripesize, root, fs_info, key->objectid);
1527
1528         ret = btrfs_find_root(tree_root, key, path,
1529                               &root->root_item, &root->root_key);
1530         if (ret) {
1531                 if (ret > 0)
1532                         ret = -ENOENT;
1533                 goto find_fail;
1534         }
1535
1536         generation = btrfs_root_generation(&root->root_item);
1537         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1538                                      generation);
1539         if (IS_ERR(root->node)) {
1540                 ret = PTR_ERR(root->node);
1541                 goto find_fail;
1542         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1543                 ret = -EIO;
1544                 free_extent_buffer(root->node);
1545                 goto find_fail;
1546         }
1547         root->commit_root = btrfs_root_node(root);
1548 out:
1549         btrfs_free_path(path);
1550         return root;
1551
1552 find_fail:
1553         kfree(root);
1554 alloc_fail:
1555         root = ERR_PTR(ret);
1556         goto out;
1557 }
1558
1559 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1560                                       struct btrfs_key *location)
1561 {
1562         struct btrfs_root *root;
1563
1564         root = btrfs_read_tree_root(tree_root, location);
1565         if (IS_ERR(root))
1566                 return root;
1567
1568         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1569                 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1570                 btrfs_check_and_init_root_item(&root->root_item);
1571         }
1572
1573         return root;
1574 }
1575
1576 int btrfs_init_fs_root(struct btrfs_root *root)
1577 {
1578         int ret;
1579         struct btrfs_subvolume_writers *writers;
1580
1581         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1582         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1583                                         GFP_NOFS);
1584         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1585                 ret = -ENOMEM;
1586                 goto fail;
1587         }
1588
1589         writers = btrfs_alloc_subvolume_writers();
1590         if (IS_ERR(writers)) {
1591                 ret = PTR_ERR(writers);
1592                 goto fail;
1593         }
1594         root->subv_writers = writers;
1595
1596         btrfs_init_free_ino_ctl(root);
1597         spin_lock_init(&root->ino_cache_lock);
1598         init_waitqueue_head(&root->ino_cache_wait);
1599
1600         ret = get_anon_bdev(&root->anon_dev);
1601         if (ret)
1602                 goto free_writers;
1603
1604         mutex_lock(&root->objectid_mutex);
1605         ret = btrfs_find_highest_objectid(root,
1606                                         &root->highest_objectid);
1607         if (ret) {
1608                 mutex_unlock(&root->objectid_mutex);
1609                 goto free_root_dev;
1610         }
1611
1612         ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1613
1614         mutex_unlock(&root->objectid_mutex);
1615
1616         return 0;
1617
1618 free_root_dev:
1619         free_anon_bdev(root->anon_dev);
1620 free_writers:
1621         btrfs_free_subvolume_writers(root->subv_writers);
1622 fail:
1623         kfree(root->free_ino_ctl);
1624         kfree(root->free_ino_pinned);
1625         return ret;
1626 }
1627
1628 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1629                                                u64 root_id)
1630 {
1631         struct btrfs_root *root;
1632
1633         spin_lock(&fs_info->fs_roots_radix_lock);
1634         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1635                                  (unsigned long)root_id);
1636         spin_unlock(&fs_info->fs_roots_radix_lock);
1637         return root;
1638 }
1639
1640 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1641                          struct btrfs_root *root)
1642 {
1643         int ret;
1644
1645         ret = radix_tree_preload(GFP_NOFS);
1646         if (ret)
1647                 return ret;
1648
1649         spin_lock(&fs_info->fs_roots_radix_lock);
1650         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1651                                 (unsigned long)root->root_key.objectid,
1652                                 root);
1653         if (ret == 0)
1654                 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1655         spin_unlock(&fs_info->fs_roots_radix_lock);
1656         radix_tree_preload_end();
1657
1658         return ret;
1659 }
1660
1661 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1662                                      struct btrfs_key *location,
1663                                      bool check_ref)
1664 {
1665         struct btrfs_root *root;
1666         struct btrfs_path *path;
1667         struct btrfs_key key;
1668         int ret;
1669
1670         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1671                 return fs_info->tree_root;
1672         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1673                 return fs_info->extent_root;
1674         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1675                 return fs_info->chunk_root;
1676         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1677                 return fs_info->dev_root;
1678         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1679                 return fs_info->csum_root;
1680         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1681                 return fs_info->quota_root ? fs_info->quota_root :
1682                                              ERR_PTR(-ENOENT);
1683         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1684                 return fs_info->uuid_root ? fs_info->uuid_root :
1685                                             ERR_PTR(-ENOENT);
1686         if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1687                 return fs_info->free_space_root ? fs_info->free_space_root :
1688                                                   ERR_PTR(-ENOENT);
1689 again:
1690         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1691         if (root) {
1692                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1693                         return ERR_PTR(-ENOENT);
1694                 return root;
1695         }
1696
1697         root = btrfs_read_fs_root(fs_info->tree_root, location);
1698         if (IS_ERR(root))
1699                 return root;
1700
1701         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1702                 ret = -ENOENT;
1703                 goto fail;
1704         }
1705
1706         ret = btrfs_init_fs_root(root);
1707         if (ret)
1708                 goto fail;
1709
1710         path = btrfs_alloc_path();
1711         if (!path) {
1712                 ret = -ENOMEM;
1713                 goto fail;
1714         }
1715         key.objectid = BTRFS_ORPHAN_OBJECTID;
1716         key.type = BTRFS_ORPHAN_ITEM_KEY;
1717         key.offset = location->objectid;
1718
1719         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1720         btrfs_free_path(path);
1721         if (ret < 0)
1722                 goto fail;
1723         if (ret == 0)
1724                 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1725
1726         ret = btrfs_insert_fs_root(fs_info, root);
1727         if (ret) {
1728                 if (ret == -EEXIST) {
1729                         free_fs_root(root);
1730                         goto again;
1731                 }
1732                 goto fail;
1733         }
1734         return root;
1735 fail:
1736         free_fs_root(root);
1737         return ERR_PTR(ret);
1738 }
1739
1740 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1741 {
1742         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1743         int ret = 0;
1744         struct btrfs_device *device;
1745         struct backing_dev_info *bdi;
1746
1747         rcu_read_lock();
1748         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1749                 if (!device->bdev)
1750                         continue;
1751                 bdi = blk_get_backing_dev_info(device->bdev);
1752                 if (bdi_congested(bdi, bdi_bits)) {
1753                         ret = 1;
1754                         break;
1755                 }
1756         }
1757         rcu_read_unlock();
1758         return ret;
1759 }
1760
1761 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1762 {
1763         int err;
1764
1765         err = bdi_setup_and_register(bdi, "btrfs");
1766         if (err)
1767                 return err;
1768
1769         bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1770         bdi->congested_fn       = btrfs_congested_fn;
1771         bdi->congested_data     = info;
1772         bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1773         return 0;
1774 }
1775
1776 /*
1777  * called by the kthread helper functions to finally call the bio end_io
1778  * functions.  This is where read checksum verification actually happens
1779  */
1780 static void end_workqueue_fn(struct btrfs_work *work)
1781 {
1782         struct bio *bio;
1783         struct btrfs_end_io_wq *end_io_wq;
1784
1785         end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1786         bio = end_io_wq->bio;
1787
1788         bio->bi_error = end_io_wq->error;
1789         bio->bi_private = end_io_wq->private;
1790         bio->bi_end_io = end_io_wq->end_io;
1791         kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1792         bio_endio(bio);
1793 }
1794
1795 static int cleaner_kthread(void *arg)
1796 {
1797         struct btrfs_root *root = arg;
1798         int again;
1799         struct btrfs_trans_handle *trans;
1800
1801         do {
1802                 again = 0;
1803
1804                 /* Make the cleaner go to sleep early. */
1805                 if (btrfs_need_cleaner_sleep(root))
1806                         goto sleep;
1807
1808                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1809                         goto sleep;
1810
1811                 /*
1812                  * Avoid the problem that we change the status of the fs
1813                  * during the above check and trylock.
1814                  */
1815                 if (btrfs_need_cleaner_sleep(root)) {
1816                         mutex_unlock(&root->fs_info->cleaner_mutex);
1817                         goto sleep;
1818                 }
1819
1820                 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1821                 btrfs_run_delayed_iputs(root);
1822                 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1823
1824                 again = btrfs_clean_one_deleted_snapshot(root);
1825                 mutex_unlock(&root->fs_info->cleaner_mutex);
1826
1827                 /*
1828                  * The defragger has dealt with the R/O remount and umount,
1829                  * needn't do anything special here.
1830                  */
1831                 btrfs_run_defrag_inodes(root->fs_info);
1832
1833                 /*
1834                  * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1835                  * with relocation (btrfs_relocate_chunk) and relocation
1836                  * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1837                  * after acquiring fs_info->delete_unused_bgs_mutex. So we
1838                  * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1839                  * unused block groups.
1840                  */
1841                 btrfs_delete_unused_bgs(root->fs_info);
1842 sleep:
1843                 if (!again) {
1844                         set_current_state(TASK_INTERRUPTIBLE);
1845                         if (!kthread_should_stop())
1846                                 schedule();
1847                         __set_current_state(TASK_RUNNING);
1848                 }
1849         } while (!kthread_should_stop());
1850
1851         /*
1852          * Transaction kthread is stopped before us and wakes us up.
1853          * However we might have started a new transaction and COWed some
1854          * tree blocks when deleting unused block groups for example. So
1855          * make sure we commit the transaction we started to have a clean
1856          * shutdown when evicting the btree inode - if it has dirty pages
1857          * when we do the final iput() on it, eviction will trigger a
1858          * writeback for it which will fail with null pointer dereferences
1859          * since work queues and other resources were already released and
1860          * destroyed by the time the iput/eviction/writeback is made.
1861          */
1862         trans = btrfs_attach_transaction(root);
1863         if (IS_ERR(trans)) {
1864                 if (PTR_ERR(trans) != -ENOENT)
1865                         btrfs_err(root->fs_info,
1866                                   "cleaner transaction attach returned %ld",
1867                                   PTR_ERR(trans));
1868         } else {
1869                 int ret;
1870
1871                 ret = btrfs_commit_transaction(trans, root);
1872                 if (ret)
1873                         btrfs_err(root->fs_info,
1874                                   "cleaner open transaction commit returned %d",
1875                                   ret);
1876         }
1877
1878         return 0;
1879 }
1880
1881 static int transaction_kthread(void *arg)
1882 {
1883         struct btrfs_root *root = arg;
1884         struct btrfs_trans_handle *trans;
1885         struct btrfs_transaction *cur;
1886         u64 transid;
1887         unsigned long now;
1888         unsigned long delay;
1889         bool cannot_commit;
1890
1891         do {
1892                 cannot_commit = false;
1893                 delay = HZ * root->fs_info->commit_interval;
1894                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1895
1896                 spin_lock(&root->fs_info->trans_lock);
1897                 cur = root->fs_info->running_transaction;
1898                 if (!cur) {
1899                         spin_unlock(&root->fs_info->trans_lock);
1900                         goto sleep;
1901                 }
1902
1903                 now = get_seconds();
1904                 if (cur->state < TRANS_STATE_BLOCKED &&
1905                     (now < cur->start_time ||
1906                      now - cur->start_time < root->fs_info->commit_interval)) {
1907                         spin_unlock(&root->fs_info->trans_lock);
1908                         delay = HZ * 5;
1909                         goto sleep;
1910                 }
1911                 transid = cur->transid;
1912                 spin_unlock(&root->fs_info->trans_lock);
1913
1914                 /* If the file system is aborted, this will always fail. */
1915                 trans = btrfs_attach_transaction(root);
1916                 if (IS_ERR(trans)) {
1917                         if (PTR_ERR(trans) != -ENOENT)
1918                                 cannot_commit = true;
1919                         goto sleep;
1920                 }
1921                 if (transid == trans->transid) {
1922                         btrfs_commit_transaction(trans, root);
1923                 } else {
1924                         btrfs_end_transaction(trans, root);
1925                 }
1926 sleep:
1927                 wake_up_process(root->fs_info->cleaner_kthread);
1928                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1929
1930                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1931                                       &root->fs_info->fs_state)))
1932                         btrfs_cleanup_transaction(root);
1933                 set_current_state(TASK_INTERRUPTIBLE);
1934                 if (!kthread_should_stop() &&
1935                                 (!btrfs_transaction_blocked(root->fs_info) ||
1936                                  cannot_commit))
1937                         schedule_timeout(delay);
1938                 __set_current_state(TASK_RUNNING);
1939         } while (!kthread_should_stop());
1940         return 0;
1941 }
1942
1943 /*
1944  * this will find the highest generation in the array of
1945  * root backups.  The index of the highest array is returned,
1946  * or -1 if we can't find anything.
1947  *
1948  * We check to make sure the array is valid by comparing the
1949  * generation of the latest  root in the array with the generation
1950  * in the super block.  If they don't match we pitch it.
1951  */
1952 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1953 {
1954         u64 cur;
1955         int newest_index = -1;
1956         struct btrfs_root_backup *root_backup;
1957         int i;
1958
1959         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1960                 root_backup = info->super_copy->super_roots + i;
1961                 cur = btrfs_backup_tree_root_gen(root_backup);
1962                 if (cur == newest_gen)
1963                         newest_index = i;
1964         }
1965
1966         /* check to see if we actually wrapped around */
1967         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1968                 root_backup = info->super_copy->super_roots;
1969                 cur = btrfs_backup_tree_root_gen(root_backup);
1970                 if (cur == newest_gen)
1971                         newest_index = 0;
1972         }
1973         return newest_index;
1974 }
1975
1976
1977 /*
1978  * find the oldest backup so we know where to store new entries
1979  * in the backup array.  This will set the backup_root_index
1980  * field in the fs_info struct
1981  */
1982 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1983                                      u64 newest_gen)
1984 {
1985         int newest_index = -1;
1986
1987         newest_index = find_newest_super_backup(info, newest_gen);
1988         /* if there was garbage in there, just move along */
1989         if (newest_index == -1) {
1990                 info->backup_root_index = 0;
1991         } else {
1992                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1993         }
1994 }
1995
1996 /*
1997  * copy all the root pointers into the super backup array.
1998  * this will bump the backup pointer by one when it is
1999  * done
2000  */
2001 static void backup_super_roots(struct btrfs_fs_info *info)
2002 {
2003         int next_backup;
2004         struct btrfs_root_backup *root_backup;
2005         int last_backup;
2006
2007         next_backup = info->backup_root_index;
2008         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2009                 BTRFS_NUM_BACKUP_ROOTS;
2010
2011         /*
2012          * just overwrite the last backup if we're at the same generation
2013          * this happens only at umount
2014          */
2015         root_backup = info->super_for_commit->super_roots + last_backup;
2016         if (btrfs_backup_tree_root_gen(root_backup) ==
2017             btrfs_header_generation(info->tree_root->node))
2018                 next_backup = last_backup;
2019
2020         root_backup = info->super_for_commit->super_roots + next_backup;
2021
2022         /*
2023          * make sure all of our padding and empty slots get zero filled
2024          * regardless of which ones we use today
2025          */
2026         memset(root_backup, 0, sizeof(*root_backup));
2027
2028         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2029
2030         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2031         btrfs_set_backup_tree_root_gen(root_backup,
2032                                btrfs_header_generation(info->tree_root->node));
2033
2034         btrfs_set_backup_tree_root_level(root_backup,
2035                                btrfs_header_level(info->tree_root->node));
2036
2037         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2038         btrfs_set_backup_chunk_root_gen(root_backup,
2039                                btrfs_header_generation(info->chunk_root->node));
2040         btrfs_set_backup_chunk_root_level(root_backup,
2041                                btrfs_header_level(info->chunk_root->node));
2042
2043         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2044         btrfs_set_backup_extent_root_gen(root_backup,
2045                                btrfs_header_generation(info->extent_root->node));
2046         btrfs_set_backup_extent_root_level(root_backup,
2047                                btrfs_header_level(info->extent_root->node));
2048
2049         /*
2050          * we might commit during log recovery, which happens before we set
2051          * the fs_root.  Make sure it is valid before we fill it in.
2052          */
2053         if (info->fs_root && info->fs_root->node) {
2054                 btrfs_set_backup_fs_root(root_backup,
2055                                          info->fs_root->node->start);
2056                 btrfs_set_backup_fs_root_gen(root_backup,
2057                                btrfs_header_generation(info->fs_root->node));
2058                 btrfs_set_backup_fs_root_level(root_backup,
2059                                btrfs_header_level(info->fs_root->node));
2060         }
2061
2062         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2063         btrfs_set_backup_dev_root_gen(root_backup,
2064                                btrfs_header_generation(info->dev_root->node));
2065         btrfs_set_backup_dev_root_level(root_backup,
2066                                        btrfs_header_level(info->dev_root->node));
2067
2068         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2069         btrfs_set_backup_csum_root_gen(root_backup,
2070                                btrfs_header_generation(info->csum_root->node));
2071         btrfs_set_backup_csum_root_level(root_backup,
2072                                btrfs_header_level(info->csum_root->node));
2073
2074         btrfs_set_backup_total_bytes(root_backup,
2075                              btrfs_super_total_bytes(info->super_copy));
2076         btrfs_set_backup_bytes_used(root_backup,
2077                              btrfs_super_bytes_used(info->super_copy));
2078         btrfs_set_backup_num_devices(root_backup,
2079                              btrfs_super_num_devices(info->super_copy));
2080
2081         /*
2082          * if we don't copy this out to the super_copy, it won't get remembered
2083          * for the next commit
2084          */
2085         memcpy(&info->super_copy->super_roots,
2086                &info->super_for_commit->super_roots,
2087                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2088 }
2089
2090 /*
2091  * this copies info out of the root backup array and back into
2092  * the in-memory super block.  It is meant to help iterate through
2093  * the array, so you send it the number of backups you've already
2094  * tried and the last backup index you used.
2095  *
2096  * this returns -1 when it has tried all the backups
2097  */
2098 static noinline int next_root_backup(struct btrfs_fs_info *info,
2099                                      struct btrfs_super_block *super,
2100                                      int *num_backups_tried, int *backup_index)
2101 {
2102         struct btrfs_root_backup *root_backup;
2103         int newest = *backup_index;
2104
2105         if (*num_backups_tried == 0) {
2106                 u64 gen = btrfs_super_generation(super);
2107
2108                 newest = find_newest_super_backup(info, gen);
2109                 if (newest == -1)
2110                         return -1;
2111
2112                 *backup_index = newest;
2113                 *num_backups_tried = 1;
2114         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2115                 /* we've tried all the backups, all done */
2116                 return -1;
2117         } else {
2118                 /* jump to the next oldest backup */
2119                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2120                         BTRFS_NUM_BACKUP_ROOTS;
2121                 *backup_index = newest;
2122                 *num_backups_tried += 1;
2123         }
2124         root_backup = super->super_roots + newest;
2125
2126         btrfs_set_super_generation(super,
2127                                    btrfs_backup_tree_root_gen(root_backup));
2128         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2129         btrfs_set_super_root_level(super,
2130                                    btrfs_backup_tree_root_level(root_backup));
2131         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2132
2133         /*
2134          * fixme: the total bytes and num_devices need to match or we should
2135          * need a fsck
2136          */
2137         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2138         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2139         return 0;
2140 }
2141
2142 /* helper to cleanup workers */
2143 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2144 {
2145         btrfs_destroy_workqueue(fs_info->fixup_workers);
2146         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2147         btrfs_destroy_workqueue(fs_info->workers);
2148         btrfs_destroy_workqueue(fs_info->endio_workers);
2149         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2150         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2151         btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2152         btrfs_destroy_workqueue(fs_info->rmw_workers);
2153         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2154         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2155         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2156         btrfs_destroy_workqueue(fs_info->submit_workers);
2157         btrfs_destroy_workqueue(fs_info->delayed_workers);
2158         btrfs_destroy_workqueue(fs_info->caching_workers);
2159         btrfs_destroy_workqueue(fs_info->readahead_workers);
2160         btrfs_destroy_workqueue(fs_info->flush_workers);
2161         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2162         btrfs_destroy_workqueue(fs_info->extent_workers);
2163 }
2164
2165 static void free_root_extent_buffers(struct btrfs_root *root)
2166 {
2167         if (root) {
2168                 free_extent_buffer(root->node);
2169                 free_extent_buffer(root->commit_root);
2170                 root->node = NULL;
2171                 root->commit_root = NULL;
2172         }
2173 }
2174
2175 /* helper to cleanup tree roots */
2176 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2177 {
2178         free_root_extent_buffers(info->tree_root);
2179
2180         free_root_extent_buffers(info->dev_root);
2181         free_root_extent_buffers(info->extent_root);
2182         free_root_extent_buffers(info->csum_root);
2183         free_root_extent_buffers(info->quota_root);
2184         free_root_extent_buffers(info->uuid_root);
2185         if (chunk_root)
2186                 free_root_extent_buffers(info->chunk_root);
2187         free_root_extent_buffers(info->free_space_root);
2188 }
2189
2190 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2191 {
2192         int ret;
2193         struct btrfs_root *gang[8];
2194         int i;
2195
2196         while (!list_empty(&fs_info->dead_roots)) {
2197                 gang[0] = list_entry(fs_info->dead_roots.next,
2198                                      struct btrfs_root, root_list);
2199                 list_del(&gang[0]->root_list);
2200
2201                 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2202                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2203                 } else {
2204                         free_extent_buffer(gang[0]->node);
2205                         free_extent_buffer(gang[0]->commit_root);
2206                         btrfs_put_fs_root(gang[0]);
2207                 }
2208         }
2209
2210         while (1) {
2211                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2212                                              (void **)gang, 0,
2213                                              ARRAY_SIZE(gang));
2214                 if (!ret)
2215                         break;
2216                 for (i = 0; i < ret; i++)
2217                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2218         }
2219
2220         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2221                 btrfs_free_log_root_tree(NULL, fs_info);
2222                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2223                                             fs_info->pinned_extents);
2224         }
2225 }
2226
2227 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2228 {
2229         mutex_init(&fs_info->scrub_lock);
2230         atomic_set(&fs_info->scrubs_running, 0);
2231         atomic_set(&fs_info->scrub_pause_req, 0);
2232         atomic_set(&fs_info->scrubs_paused, 0);
2233         atomic_set(&fs_info->scrub_cancel_req, 0);
2234         init_waitqueue_head(&fs_info->scrub_pause_wait);
2235         fs_info->scrub_workers_refcnt = 0;
2236 }
2237
2238 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2239 {
2240         spin_lock_init(&fs_info->balance_lock);
2241         mutex_init(&fs_info->balance_mutex);
2242         atomic_set(&fs_info->balance_running, 0);
2243         atomic_set(&fs_info->balance_pause_req, 0);
2244         atomic_set(&fs_info->balance_cancel_req, 0);
2245         fs_info->balance_ctl = NULL;
2246         init_waitqueue_head(&fs_info->balance_wait_q);
2247 }
2248
2249 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2250                                    struct btrfs_root *tree_root)
2251 {
2252         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2253         set_nlink(fs_info->btree_inode, 1);
2254         /*
2255          * we set the i_size on the btree inode to the max possible int.
2256          * the real end of the address space is determined by all of
2257          * the devices in the system
2258          */
2259         fs_info->btree_inode->i_size = OFFSET_MAX;
2260         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2261
2262         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2263         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2264                              fs_info->btree_inode->i_mapping);
2265         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2266         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2267
2268         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2269
2270         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2271         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2272                sizeof(struct btrfs_key));
2273         set_bit(BTRFS_INODE_DUMMY,
2274                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2275         btrfs_insert_inode_hash(fs_info->btree_inode);
2276 }
2277
2278 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2279 {
2280         fs_info->dev_replace.lock_owner = 0;
2281         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2282         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2283         rwlock_init(&fs_info->dev_replace.lock);
2284         atomic_set(&fs_info->dev_replace.read_locks, 0);
2285         atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2286         init_waitqueue_head(&fs_info->replace_wait);
2287         init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2288 }
2289
2290 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2291 {
2292         spin_lock_init(&fs_info->qgroup_lock);
2293         mutex_init(&fs_info->qgroup_ioctl_lock);
2294         fs_info->qgroup_tree = RB_ROOT;
2295         fs_info->qgroup_op_tree = RB_ROOT;
2296         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2297         fs_info->qgroup_seq = 1;
2298         fs_info->quota_enabled = 0;
2299         fs_info->pending_quota_state = 0;
2300         fs_info->qgroup_ulist = NULL;
2301         mutex_init(&fs_info->qgroup_rescan_lock);
2302 }
2303
2304 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2305                 struct btrfs_fs_devices *fs_devices)
2306 {
2307         int max_active = fs_info->thread_pool_size;
2308         unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2309
2310         fs_info->workers =
2311                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2312                                       max_active, 16);
2313
2314         fs_info->delalloc_workers =
2315                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2316
2317         fs_info->flush_workers =
2318                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2319
2320         fs_info->caching_workers =
2321                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2322
2323         /*
2324          * a higher idle thresh on the submit workers makes it much more
2325          * likely that bios will be send down in a sane order to the
2326          * devices
2327          */
2328         fs_info->submit_workers =
2329                 btrfs_alloc_workqueue("submit", flags,
2330                                       min_t(u64, fs_devices->num_devices,
2331                                             max_active), 64);
2332
2333         fs_info->fixup_workers =
2334                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2335
2336         /*
2337          * endios are largely parallel and should have a very
2338          * low idle thresh
2339          */
2340         fs_info->endio_workers =
2341                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2342         fs_info->endio_meta_workers =
2343                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2344         fs_info->endio_meta_write_workers =
2345                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2346         fs_info->endio_raid56_workers =
2347                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2348         fs_info->endio_repair_workers =
2349                 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2350         fs_info->rmw_workers =
2351                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2352         fs_info->endio_write_workers =
2353                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2354         fs_info->endio_freespace_worker =
2355                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2356         fs_info->delayed_workers =
2357                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2358         fs_info->readahead_workers =
2359                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2360         fs_info->qgroup_rescan_workers =
2361                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2362         fs_info->extent_workers =
2363                 btrfs_alloc_workqueue("extent-refs", flags,
2364                                       min_t(u64, fs_devices->num_devices,
2365                                             max_active), 8);
2366
2367         if (!(fs_info->workers && fs_info->delalloc_workers &&
2368               fs_info->submit_workers && fs_info->flush_workers &&
2369               fs_info->endio_workers && fs_info->endio_meta_workers &&
2370               fs_info->endio_meta_write_workers &&
2371               fs_info->endio_repair_workers &&
2372               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2373               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2374               fs_info->caching_workers && fs_info->readahead_workers &&
2375               fs_info->fixup_workers && fs_info->delayed_workers &&
2376               fs_info->extent_workers &&
2377               fs_info->qgroup_rescan_workers)) {
2378                 return -ENOMEM;
2379         }
2380
2381         return 0;
2382 }
2383
2384 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2385                             struct btrfs_fs_devices *fs_devices)
2386 {
2387         int ret;
2388         struct btrfs_root *tree_root = fs_info->tree_root;
2389         struct btrfs_root *log_tree_root;
2390         struct btrfs_super_block *disk_super = fs_info->super_copy;
2391         u64 bytenr = btrfs_super_log_root(disk_super);
2392
2393         if (fs_devices->rw_devices == 0) {
2394                 btrfs_warn(fs_info, "log replay required on RO media");
2395                 return -EIO;
2396         }
2397
2398         log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2399         if (!log_tree_root)
2400                 return -ENOMEM;
2401
2402         __setup_root(tree_root->nodesize, tree_root->sectorsize,
2403                         tree_root->stripesize, log_tree_root, fs_info,
2404                         BTRFS_TREE_LOG_OBJECTID);
2405
2406         log_tree_root->node = read_tree_block(tree_root, bytenr,
2407                         fs_info->generation + 1);
2408         if (IS_ERR(log_tree_root->node)) {
2409                 btrfs_warn(fs_info, "failed to read log tree");
2410                 ret = PTR_ERR(log_tree_root->node);
2411                 kfree(log_tree_root);
2412                 return ret;
2413         } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2414                 btrfs_err(fs_info, "failed to read log tree");
2415                 free_extent_buffer(log_tree_root->node);
2416                 kfree(log_tree_root);
2417                 return -EIO;
2418         }
2419         /* returns with log_tree_root freed on success */
2420         ret = btrfs_recover_log_trees(log_tree_root);
2421         if (ret) {
2422                 btrfs_handle_fs_error(tree_root->fs_info, ret,
2423                             "Failed to recover log tree");
2424                 free_extent_buffer(log_tree_root->node);
2425                 kfree(log_tree_root);
2426                 return ret;
2427         }
2428
2429         if (fs_info->sb->s_flags & MS_RDONLY) {
2430                 ret = btrfs_commit_super(tree_root);
2431                 if (ret)
2432                         return ret;
2433         }
2434
2435         return 0;
2436 }
2437
2438 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2439                             struct btrfs_root *tree_root)
2440 {
2441         struct btrfs_root *root;
2442         struct btrfs_key location;
2443         int ret;
2444
2445         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2446         location.type = BTRFS_ROOT_ITEM_KEY;
2447         location.offset = 0;
2448
2449         root = btrfs_read_tree_root(tree_root, &location);
2450         if (IS_ERR(root))
2451                 return PTR_ERR(root);
2452         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2453         fs_info->extent_root = root;
2454
2455         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2456         root = btrfs_read_tree_root(tree_root, &location);
2457         if (IS_ERR(root))
2458                 return PTR_ERR(root);
2459         set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2460         fs_info->dev_root = root;
2461         btrfs_init_devices_late(fs_info);
2462
2463         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
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->csum_root = root;
2469
2470         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2471         root = btrfs_read_tree_root(tree_root, &location);
2472         if (!IS_ERR(root)) {
2473                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2474                 fs_info->quota_enabled = 1;
2475                 fs_info->pending_quota_state = 1;
2476                 fs_info->quota_root = root;
2477         }
2478
2479         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2480         root = btrfs_read_tree_root(tree_root, &location);
2481         if (IS_ERR(root)) {
2482                 ret = PTR_ERR(root);
2483                 if (ret != -ENOENT)
2484                         return ret;
2485         } else {
2486                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2487                 fs_info->uuid_root = root;
2488         }
2489
2490         if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2491                 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2492                 root = btrfs_read_tree_root(tree_root, &location);
2493                 if (IS_ERR(root))
2494                         return PTR_ERR(root);
2495                 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2496                 fs_info->free_space_root = root;
2497         }
2498
2499         return 0;
2500 }
2501
2502 int open_ctree(struct super_block *sb,
2503                struct btrfs_fs_devices *fs_devices,
2504                char *options)
2505 {
2506         u32 sectorsize;
2507         u32 nodesize;
2508         u32 stripesize;
2509         u64 generation;
2510         u64 features;
2511         struct btrfs_key location;
2512         struct buffer_head *bh;
2513         struct btrfs_super_block *disk_super;
2514         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2515         struct btrfs_root *tree_root;
2516         struct btrfs_root *chunk_root;
2517         int ret;
2518         int err = -EINVAL;
2519         int num_backups_tried = 0;
2520         int backup_index = 0;
2521         int max_active;
2522         bool cleaner_mutex_locked = false;
2523
2524         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2525         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2526         if (!tree_root || !chunk_root) {
2527                 err = -ENOMEM;
2528                 goto fail;
2529         }
2530
2531         ret = init_srcu_struct(&fs_info->subvol_srcu);
2532         if (ret) {
2533                 err = ret;
2534                 goto fail;
2535         }
2536
2537         ret = setup_bdi(fs_info, &fs_info->bdi);
2538         if (ret) {
2539                 err = ret;
2540                 goto fail_srcu;
2541         }
2542
2543         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2544         if (ret) {
2545                 err = ret;
2546                 goto fail_bdi;
2547         }
2548         fs_info->dirty_metadata_batch = PAGE_SIZE *
2549                                         (1 + ilog2(nr_cpu_ids));
2550
2551         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2552         if (ret) {
2553                 err = ret;
2554                 goto fail_dirty_metadata_bytes;
2555         }
2556
2557         ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2558         if (ret) {
2559                 err = ret;
2560                 goto fail_delalloc_bytes;
2561         }
2562
2563         fs_info->btree_inode = new_inode(sb);
2564         if (!fs_info->btree_inode) {
2565                 err = -ENOMEM;
2566                 goto fail_bio_counter;
2567         }
2568
2569         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2570
2571         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2572         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2573         INIT_LIST_HEAD(&fs_info->trans_list);
2574         INIT_LIST_HEAD(&fs_info->dead_roots);
2575         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2576         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2577         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2578         spin_lock_init(&fs_info->delalloc_root_lock);
2579         spin_lock_init(&fs_info->trans_lock);
2580         spin_lock_init(&fs_info->fs_roots_radix_lock);
2581         spin_lock_init(&fs_info->delayed_iput_lock);
2582         spin_lock_init(&fs_info->defrag_inodes_lock);
2583         spin_lock_init(&fs_info->free_chunk_lock);
2584         spin_lock_init(&fs_info->tree_mod_seq_lock);
2585         spin_lock_init(&fs_info->super_lock);
2586         spin_lock_init(&fs_info->qgroup_op_lock);
2587         spin_lock_init(&fs_info->buffer_lock);
2588         spin_lock_init(&fs_info->unused_bgs_lock);
2589         rwlock_init(&fs_info->tree_mod_log_lock);
2590         mutex_init(&fs_info->unused_bg_unpin_mutex);
2591         mutex_init(&fs_info->delete_unused_bgs_mutex);
2592         mutex_init(&fs_info->reloc_mutex);
2593         mutex_init(&fs_info->delalloc_root_mutex);
2594         mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2595         seqlock_init(&fs_info->profiles_lock);
2596
2597         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2598         INIT_LIST_HEAD(&fs_info->space_info);
2599         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2600         INIT_LIST_HEAD(&fs_info->unused_bgs);
2601         btrfs_mapping_init(&fs_info->mapping_tree);
2602         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2603                              BTRFS_BLOCK_RSV_GLOBAL);
2604         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2605                              BTRFS_BLOCK_RSV_DELALLOC);
2606         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2607         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2608         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2609         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2610                              BTRFS_BLOCK_RSV_DELOPS);
2611         atomic_set(&fs_info->nr_async_submits, 0);
2612         atomic_set(&fs_info->async_delalloc_pages, 0);
2613         atomic_set(&fs_info->async_submit_draining, 0);
2614         atomic_set(&fs_info->nr_async_bios, 0);
2615         atomic_set(&fs_info->defrag_running, 0);
2616         atomic_set(&fs_info->qgroup_op_seq, 0);
2617         atomic_set(&fs_info->reada_works_cnt, 0);
2618         atomic64_set(&fs_info->tree_mod_seq, 0);
2619         fs_info->sb = sb;
2620         fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2621         fs_info->metadata_ratio = 0;
2622         fs_info->defrag_inodes = RB_ROOT;
2623         fs_info->free_chunk_space = 0;
2624         fs_info->tree_mod_log = RB_ROOT;
2625         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2626         fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2627         /* readahead state */
2628         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2629         spin_lock_init(&fs_info->reada_lock);
2630
2631         fs_info->thread_pool_size = min_t(unsigned long,
2632                                           num_online_cpus() + 2, 8);
2633
2634         INIT_LIST_HEAD(&fs_info->ordered_roots);
2635         spin_lock_init(&fs_info->ordered_root_lock);
2636         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2637                                         GFP_KERNEL);
2638         if (!fs_info->delayed_root) {
2639                 err = -ENOMEM;
2640                 goto fail_iput;
2641         }
2642         btrfs_init_delayed_root(fs_info->delayed_root);
2643
2644         btrfs_init_scrub(fs_info);
2645 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2646         fs_info->check_integrity_print_mask = 0;
2647 #endif
2648         btrfs_init_balance(fs_info);
2649         btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2650
2651         sb->s_blocksize = 4096;
2652         sb->s_blocksize_bits = blksize_bits(4096);
2653         sb->s_bdi = &fs_info->bdi;
2654
2655         btrfs_init_btree_inode(fs_info, tree_root);
2656
2657         spin_lock_init(&fs_info->block_group_cache_lock);
2658         fs_info->block_group_cache_tree = RB_ROOT;
2659         fs_info->first_logical_byte = (u64)-1;
2660
2661         extent_io_tree_init(&fs_info->freed_extents[0],
2662                              fs_info->btree_inode->i_mapping);
2663         extent_io_tree_init(&fs_info->freed_extents[1],
2664                              fs_info->btree_inode->i_mapping);
2665         fs_info->pinned_extents = &fs_info->freed_extents[0];
2666         fs_info->do_barriers = 1;
2667
2668
2669         mutex_init(&fs_info->ordered_operations_mutex);
2670         mutex_init(&fs_info->tree_log_mutex);
2671         mutex_init(&fs_info->chunk_mutex);
2672         mutex_init(&fs_info->transaction_kthread_mutex);
2673         mutex_init(&fs_info->cleaner_mutex);
2674         mutex_init(&fs_info->volume_mutex);
2675         mutex_init(&fs_info->ro_block_group_mutex);
2676         init_rwsem(&fs_info->commit_root_sem);
2677         init_rwsem(&fs_info->cleanup_work_sem);
2678         init_rwsem(&fs_info->subvol_sem);
2679         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2680
2681         btrfs_init_dev_replace_locks(fs_info);
2682         btrfs_init_qgroup(fs_info);
2683
2684         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2685         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2686
2687         init_waitqueue_head(&fs_info->transaction_throttle);
2688         init_waitqueue_head(&fs_info->transaction_wait);
2689         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2690         init_waitqueue_head(&fs_info->async_submit_wait);
2691
2692         INIT_LIST_HEAD(&fs_info->pinned_chunks);
2693
2694         ret = btrfs_alloc_stripe_hash_table(fs_info);
2695         if (ret) {
2696                 err = ret;
2697                 goto fail_alloc;
2698         }
2699
2700         __setup_root(4096, 4096, 4096, tree_root,
2701                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2702
2703         invalidate_bdev(fs_devices->latest_bdev);
2704
2705         /*
2706          * Read super block and check the signature bytes only
2707          */
2708         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2709         if (IS_ERR(bh)) {
2710                 err = PTR_ERR(bh);
2711                 goto fail_alloc;
2712         }
2713
2714         /*
2715          * We want to check superblock checksum, the type is stored inside.
2716          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2717          */
2718         if (btrfs_check_super_csum(bh->b_data)) {
2719                 btrfs_err(fs_info, "superblock checksum mismatch");
2720                 err = -EINVAL;
2721                 brelse(bh);
2722                 goto fail_alloc;
2723         }
2724
2725         /*
2726          * super_copy is zeroed at allocation time and we never touch the
2727          * following bytes up to INFO_SIZE, the checksum is calculated from
2728          * the whole block of INFO_SIZE
2729          */
2730         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2731         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2732                sizeof(*fs_info->super_for_commit));
2733         brelse(bh);
2734
2735         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2736
2737         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2738         if (ret) {
2739                 btrfs_err(fs_info, "superblock contains fatal errors");
2740                 err = -EINVAL;
2741                 goto fail_alloc;
2742         }
2743
2744         disk_super = fs_info->super_copy;
2745         if (!btrfs_super_root(disk_super))
2746                 goto fail_alloc;
2747
2748         /* check FS state, whether FS is broken. */
2749         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2750                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2751
2752         /*
2753          * run through our array of backup supers and setup
2754          * our ring pointer to the oldest one
2755          */
2756         generation = btrfs_super_generation(disk_super);
2757         find_oldest_super_backup(fs_info, generation);
2758
2759         /*
2760          * In the long term, we'll store the compression type in the super
2761          * block, and it'll be used for per file compression control.
2762          */
2763         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2764
2765         ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2766         if (ret) {
2767                 err = ret;
2768                 goto fail_alloc;
2769         }
2770
2771         features = btrfs_super_incompat_flags(disk_super) &
2772                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2773         if (features) {
2774                 btrfs_err(fs_info,
2775                     "cannot mount because of unsupported optional features (%llx)",
2776                     features);
2777                 err = -EINVAL;
2778                 goto fail_alloc;
2779         }
2780
2781         features = btrfs_super_incompat_flags(disk_super);
2782         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2783         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2784                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2785
2786         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2787                 btrfs_info(fs_info, "has skinny extents");
2788
2789         /*
2790          * flag our filesystem as having big metadata blocks if
2791          * they are bigger than the page size
2792          */
2793         if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2794                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2795                         btrfs_info(fs_info,
2796                                 "flagging fs with big metadata feature");
2797                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2798         }
2799
2800         nodesize = btrfs_super_nodesize(disk_super);
2801         sectorsize = btrfs_super_sectorsize(disk_super);
2802         stripesize = btrfs_super_stripesize(disk_super);
2803         fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2804         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2805
2806         /*
2807          * mixed block groups end up with duplicate but slightly offset
2808          * extent buffers for the same range.  It leads to corruptions
2809          */
2810         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2811             (sectorsize != nodesize)) {
2812                 btrfs_err(fs_info,
2813 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2814                         nodesize, sectorsize);
2815                 goto fail_alloc;
2816         }
2817
2818         /*
2819          * Needn't use the lock because there is no other task which will
2820          * update the flag.
2821          */
2822         btrfs_set_super_incompat_flags(disk_super, features);
2823
2824         features = btrfs_super_compat_ro_flags(disk_super) &
2825                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2826         if (!(sb->s_flags & MS_RDONLY) && features) {
2827                 btrfs_err(fs_info,
2828         "cannot mount read-write because of unsupported optional features (%llx)",
2829                        features);
2830                 err = -EINVAL;
2831                 goto fail_alloc;
2832         }
2833
2834         max_active = fs_info->thread_pool_size;
2835
2836         ret = btrfs_init_workqueues(fs_info, fs_devices);
2837         if (ret) {
2838                 err = ret;
2839                 goto fail_sb_buffer;
2840         }
2841
2842         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2843         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2844                                     SZ_4M / PAGE_SIZE);
2845
2846         tree_root->nodesize = nodesize;
2847         tree_root->sectorsize = sectorsize;
2848         tree_root->stripesize = stripesize;
2849
2850         sb->s_blocksize = sectorsize;
2851         sb->s_blocksize_bits = blksize_bits(sectorsize);
2852
2853         mutex_lock(&fs_info->chunk_mutex);
2854         ret = btrfs_read_sys_array(tree_root);
2855         mutex_unlock(&fs_info->chunk_mutex);
2856         if (ret) {
2857                 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2858                 goto fail_sb_buffer;
2859         }
2860
2861         generation = btrfs_super_chunk_root_generation(disk_super);
2862
2863         __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2864                      fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2865
2866         chunk_root->node = read_tree_block(chunk_root,
2867                                            btrfs_super_chunk_root(disk_super),
2868                                            generation);
2869         if (IS_ERR(chunk_root->node) ||
2870             !extent_buffer_uptodate(chunk_root->node)) {
2871                 btrfs_err(fs_info, "failed to read chunk root");
2872                 if (!IS_ERR(chunk_root->node))
2873                         free_extent_buffer(chunk_root->node);
2874                 chunk_root->node = NULL;
2875                 goto fail_tree_roots;
2876         }
2877         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2878         chunk_root->commit_root = btrfs_root_node(chunk_root);
2879
2880         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2881            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2882
2883         ret = btrfs_read_chunk_tree(chunk_root);
2884         if (ret) {
2885                 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2886                 goto fail_tree_roots;
2887         }
2888
2889         /*
2890          * keep the device that is marked to be the target device for the
2891          * dev_replace procedure
2892          */
2893         btrfs_close_extra_devices(fs_devices, 0);
2894
2895         if (!fs_devices->latest_bdev) {
2896                 btrfs_err(fs_info, "failed to read devices");
2897                 goto fail_tree_roots;
2898         }
2899
2900 retry_root_backup:
2901         generation = btrfs_super_generation(disk_super);
2902
2903         tree_root->node = read_tree_block(tree_root,
2904                                           btrfs_super_root(disk_super),
2905                                           generation);
2906         if (IS_ERR(tree_root->node) ||
2907             !extent_buffer_uptodate(tree_root->node)) {
2908                 btrfs_warn(fs_info, "failed to read tree root");
2909                 if (!IS_ERR(tree_root->node))
2910                         free_extent_buffer(tree_root->node);
2911                 tree_root->node = NULL;
2912                 goto recovery_tree_root;
2913         }
2914
2915         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2916         tree_root->commit_root = btrfs_root_node(tree_root);
2917         btrfs_set_root_refs(&tree_root->root_item, 1);
2918
2919         mutex_lock(&tree_root->objectid_mutex);
2920         ret = btrfs_find_highest_objectid(tree_root,
2921                                         &tree_root->highest_objectid);
2922         if (ret) {
2923                 mutex_unlock(&tree_root->objectid_mutex);
2924                 goto recovery_tree_root;
2925         }
2926
2927         ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2928
2929         mutex_unlock(&tree_root->objectid_mutex);
2930
2931         ret = btrfs_read_roots(fs_info, tree_root);
2932         if (ret)
2933                 goto recovery_tree_root;
2934
2935         fs_info->generation = generation;
2936         fs_info->last_trans_committed = generation;
2937
2938         ret = btrfs_recover_balance(fs_info);
2939         if (ret) {
2940                 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2941                 goto fail_block_groups;
2942         }
2943
2944         ret = btrfs_init_dev_stats(fs_info);
2945         if (ret) {
2946                 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2947                 goto fail_block_groups;
2948         }
2949
2950         ret = btrfs_init_dev_replace(fs_info);
2951         if (ret) {
2952                 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2953                 goto fail_block_groups;
2954         }
2955
2956         btrfs_close_extra_devices(fs_devices, 1);
2957
2958         ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2959         if (ret) {
2960                 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2961                                 ret);
2962                 goto fail_block_groups;
2963         }
2964
2965         ret = btrfs_sysfs_add_device(fs_devices);
2966         if (ret) {
2967                 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2968                                 ret);
2969                 goto fail_fsdev_sysfs;
2970         }
2971
2972         ret = btrfs_sysfs_add_mounted(fs_info);
2973         if (ret) {
2974                 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2975                 goto fail_fsdev_sysfs;
2976         }
2977
2978         ret = btrfs_init_space_info(fs_info);
2979         if (ret) {
2980                 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2981                 goto fail_sysfs;
2982         }
2983
2984         ret = btrfs_read_block_groups(fs_info->extent_root);
2985         if (ret) {
2986                 btrfs_err(fs_info, "failed to read block groups: %d", ret);
2987                 goto fail_sysfs;
2988         }
2989         fs_info->num_tolerated_disk_barrier_failures =
2990                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2991         if (fs_info->fs_devices->missing_devices >
2992              fs_info->num_tolerated_disk_barrier_failures &&
2993             !(sb->s_flags & MS_RDONLY)) {
2994                 btrfs_warn(fs_info,
2995 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
2996                         fs_info->fs_devices->missing_devices,
2997                         fs_info->num_tolerated_disk_barrier_failures);
2998                 goto fail_sysfs;
2999         }
3000
3001         /*
3002          * Hold the cleaner_mutex thread here so that we don't block
3003          * for a long time on btrfs_recover_relocation.  cleaner_kthread
3004          * will wait for us to finish mounting the filesystem.
3005          */
3006         mutex_lock(&fs_info->cleaner_mutex);
3007         cleaner_mutex_locked = true;
3008         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3009                                                "btrfs-cleaner");
3010         if (IS_ERR(fs_info->cleaner_kthread))
3011                 goto fail_sysfs;
3012
3013         fs_info->transaction_kthread = kthread_run(transaction_kthread,
3014                                                    tree_root,
3015                                                    "btrfs-transaction");
3016         if (IS_ERR(fs_info->transaction_kthread))
3017                 goto fail_cleaner;
3018
3019         if (!btrfs_test_opt(tree_root, SSD) &&
3020             !btrfs_test_opt(tree_root, NOSSD) &&
3021             !fs_info->fs_devices->rotating) {
3022                 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3023                 btrfs_set_opt(fs_info->mount_opt, SSD);
3024         }
3025
3026         /*
3027          * Mount does not set all options immediately, we can do it now and do
3028          * not have to wait for transaction commit
3029          */
3030         btrfs_apply_pending_changes(fs_info);
3031
3032 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3033         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
3034                 ret = btrfsic_mount(tree_root, fs_devices,
3035                                     btrfs_test_opt(tree_root,
3036                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3037                                     1 : 0,
3038                                     fs_info->check_integrity_print_mask);
3039                 if (ret)
3040                         btrfs_warn(fs_info,
3041                                 "failed to initialize integrity check module: %d",
3042                                 ret);
3043         }
3044 #endif
3045         ret = btrfs_read_qgroup_config(fs_info);
3046         if (ret)
3047                 goto fail_trans_kthread;
3048
3049         /* do not make disk changes in broken FS or nologreplay is given */
3050         if (btrfs_super_log_root(disk_super) != 0 &&
3051             !btrfs_test_opt(tree_root, NOLOGREPLAY)) {
3052                 ret = btrfs_replay_log(fs_info, fs_devices);
3053                 if (ret) {
3054                         err = ret;
3055                         goto fail_qgroup;
3056                 }
3057         }
3058
3059         ret = btrfs_find_orphan_roots(tree_root);
3060         if (ret)
3061                 goto fail_qgroup;
3062
3063         if (!(sb->s_flags & MS_RDONLY)) {
3064                 ret = btrfs_cleanup_fs_roots(fs_info);
3065                 if (ret)
3066                         goto fail_qgroup;
3067                 /* We locked cleaner_mutex before creating cleaner_kthread. */
3068                 ret = btrfs_recover_relocation(tree_root);
3069                 if (ret < 0) {
3070                         btrfs_warn(fs_info, "failed to recover relocation: %d",
3071                                         ret);
3072                         err = -EINVAL;
3073                         goto fail_qgroup;
3074                 }
3075         }
3076         mutex_unlock(&fs_info->cleaner_mutex);
3077         cleaner_mutex_locked = false;
3078
3079         location.objectid = BTRFS_FS_TREE_OBJECTID;
3080         location.type = BTRFS_ROOT_ITEM_KEY;
3081         location.offset = 0;
3082
3083         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3084         if (IS_ERR(fs_info->fs_root)) {
3085                 err = PTR_ERR(fs_info->fs_root);
3086                 goto fail_qgroup;
3087         }
3088
3089         if (sb->s_flags & MS_RDONLY)
3090                 return 0;
3091
3092         if (btrfs_test_opt(tree_root, FREE_SPACE_TREE) &&
3093             !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3094                 btrfs_info(fs_info, "creating free space tree");
3095                 ret = btrfs_create_free_space_tree(fs_info);
3096                 if (ret) {
3097                         btrfs_warn(fs_info,
3098                                 "failed to create free space tree: %d", ret);
3099                         close_ctree(tree_root);
3100                         return ret;
3101                 }
3102         }
3103
3104         down_read(&fs_info->cleanup_work_sem);
3105         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3106             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3107                 up_read(&fs_info->cleanup_work_sem);
3108                 close_ctree(tree_root);
3109                 return ret;
3110         }
3111         up_read(&fs_info->cleanup_work_sem);
3112
3113         ret = btrfs_resume_balance_async(fs_info);
3114         if (ret) {
3115                 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3116                 close_ctree(tree_root);
3117                 return ret;
3118         }
3119
3120         ret = btrfs_resume_dev_replace_async(fs_info);
3121         if (ret) {
3122                 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3123                 close_ctree(tree_root);
3124                 return ret;
3125         }
3126
3127         btrfs_qgroup_rescan_resume(fs_info);
3128
3129         if (btrfs_test_opt(tree_root, CLEAR_CACHE) &&
3130             btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3131                 btrfs_info(fs_info, "clearing free space tree");
3132                 ret = btrfs_clear_free_space_tree(fs_info);
3133                 if (ret) {
3134                         btrfs_warn(fs_info,
3135                                 "failed to clear free space tree: %d", ret);
3136                         close_ctree(tree_root);
3137                         return ret;
3138                 }
3139         }
3140
3141         if (!fs_info->uuid_root) {
3142                 btrfs_info(fs_info, "creating UUID tree");
3143                 ret = btrfs_create_uuid_tree(fs_info);
3144                 if (ret) {
3145                         btrfs_warn(fs_info,
3146                                 "failed to create the UUID tree: %d", ret);
3147                         close_ctree(tree_root);
3148                         return ret;
3149                 }
3150         } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3151                    fs_info->generation !=
3152                                 btrfs_super_uuid_tree_generation(disk_super)) {
3153                 btrfs_info(fs_info, "checking UUID tree");
3154                 ret = btrfs_check_uuid_tree(fs_info);
3155                 if (ret) {
3156                         btrfs_warn(fs_info,
3157                                 "failed to check the UUID tree: %d", ret);
3158                         close_ctree(tree_root);
3159                         return ret;
3160                 }
3161         } else {
3162                 fs_info->update_uuid_tree_gen = 1;
3163         }
3164
3165         fs_info->open = 1;
3166
3167         /*
3168          * backuproot only affect mount behavior, and if open_ctree succeeded,
3169          * no need to keep the flag
3170          */
3171         btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3172
3173         return 0;
3174
3175 fail_qgroup:
3176         btrfs_free_qgroup_config(fs_info);
3177 fail_trans_kthread:
3178         kthread_stop(fs_info->transaction_kthread);
3179         btrfs_cleanup_transaction(fs_info->tree_root);
3180         btrfs_free_fs_roots(fs_info);
3181 fail_cleaner:
3182         kthread_stop(fs_info->cleaner_kthread);
3183
3184         /*
3185          * make sure we're done with the btree inode before we stop our
3186          * kthreads
3187          */
3188         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3189
3190 fail_sysfs:
3191         if (cleaner_mutex_locked) {
3192                 mutex_unlock(&fs_info->cleaner_mutex);
3193                 cleaner_mutex_locked = false;
3194         }
3195         btrfs_sysfs_remove_mounted(fs_info);
3196
3197 fail_fsdev_sysfs:
3198         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3199
3200 fail_block_groups:
3201         btrfs_put_block_group_cache(fs_info);
3202         btrfs_free_block_groups(fs_info);
3203
3204 fail_tree_roots:
3205         free_root_pointers(fs_info, 1);
3206         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3207
3208 fail_sb_buffer:
3209         btrfs_stop_all_workers(fs_info);
3210 fail_alloc:
3211 fail_iput:
3212         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3213
3214         iput(fs_info->btree_inode);
3215 fail_bio_counter:
3216         percpu_counter_destroy(&fs_info->bio_counter);
3217 fail_delalloc_bytes:
3218         percpu_counter_destroy(&fs_info->delalloc_bytes);
3219 fail_dirty_metadata_bytes:
3220         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3221 fail_bdi:
3222         bdi_destroy(&fs_info->bdi);
3223 fail_srcu:
3224         cleanup_srcu_struct(&fs_info->subvol_srcu);
3225 fail:
3226         btrfs_free_stripe_hash_table(fs_info);
3227         btrfs_close_devices(fs_info->fs_devices);
3228         return err;
3229
3230 recovery_tree_root:
3231         if (!btrfs_test_opt(tree_root, USEBACKUPROOT))
3232                 goto fail_tree_roots;
3233
3234         free_root_pointers(fs_info, 0);
3235
3236         /* don't use the log in recovery mode, it won't be valid */
3237         btrfs_set_super_log_root(disk_super, 0);
3238
3239         /* we can't trust the free space cache either */
3240         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3241
3242         ret = next_root_backup(fs_info, fs_info->super_copy,
3243                                &num_backups_tried, &backup_index);
3244         if (ret == -1)
3245                 goto fail_block_groups;
3246         goto retry_root_backup;
3247 }
3248
3249 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3250 {
3251         if (uptodate) {
3252                 set_buffer_uptodate(bh);
3253         } else {
3254                 struct btrfs_device *device = (struct btrfs_device *)
3255                         bh->b_private;
3256
3257                 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3258                                 "lost page write due to IO error on %s",
3259                                           rcu_str_deref(device->name));
3260                 /* note, we don't set_buffer_write_io_error because we have
3261                  * our own ways of dealing with the IO errors
3262                  */
3263                 clear_buffer_uptodate(bh);
3264                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3265         }
3266         unlock_buffer(bh);
3267         put_bh(bh);
3268 }
3269
3270 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3271                         struct buffer_head **bh_ret)
3272 {
3273         struct buffer_head *bh;
3274         struct btrfs_super_block *super;
3275         u64 bytenr;
3276
3277         bytenr = btrfs_sb_offset(copy_num);
3278         if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3279                 return -EINVAL;
3280
3281         bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3282         /*
3283          * If we fail to read from the underlying devices, as of now
3284          * the best option we have is to mark it EIO.
3285          */
3286         if (!bh)
3287                 return -EIO;
3288
3289         super = (struct btrfs_super_block *)bh->b_data;
3290         if (btrfs_super_bytenr(super) != bytenr ||
3291                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3292                 brelse(bh);
3293                 return -EINVAL;
3294         }
3295
3296         *bh_ret = bh;
3297         return 0;
3298 }
3299
3300
3301 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3302 {
3303         struct buffer_head *bh;
3304         struct buffer_head *latest = NULL;
3305         struct btrfs_super_block *super;
3306         int i;
3307         u64 transid = 0;
3308         int ret = -EINVAL;
3309
3310         /* we would like to check all the supers, but that would make
3311          * a btrfs mount succeed after a mkfs from a different FS.
3312          * So, we need to add a special mount option to scan for
3313          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3314          */
3315         for (i = 0; i < 1; i++) {
3316                 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3317                 if (ret)
3318                         continue;
3319
3320                 super = (struct btrfs_super_block *)bh->b_data;
3321
3322                 if (!latest || btrfs_super_generation(super) > transid) {
3323                         brelse(latest);
3324                         latest = bh;
3325                         transid = btrfs_super_generation(super);
3326                 } else {
3327                         brelse(bh);
3328                 }
3329         }
3330
3331         if (!latest)
3332                 return ERR_PTR(ret);
3333
3334         return latest;
3335 }
3336
3337 /*
3338  * this should be called twice, once with wait == 0 and
3339  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3340  * we write are pinned.
3341  *
3342  * They are released when wait == 1 is done.
3343  * max_mirrors must be the same for both runs, and it indicates how
3344  * many supers on this one device should be written.
3345  *
3346  * max_mirrors == 0 means to write them all.
3347  */
3348 static int write_dev_supers(struct btrfs_device *device,
3349                             struct btrfs_super_block *sb,
3350                             int do_barriers, int wait, int max_mirrors)
3351 {
3352         struct buffer_head *bh;
3353         int i;
3354         int ret;
3355         int errors = 0;
3356         u32 crc;
3357         u64 bytenr;
3358
3359         if (max_mirrors == 0)
3360                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3361
3362         for (i = 0; i < max_mirrors; i++) {
3363                 bytenr = btrfs_sb_offset(i);
3364                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3365                     device->commit_total_bytes)
3366                         break;
3367
3368                 if (wait) {
3369                         bh = __find_get_block(device->bdev, bytenr / 4096,
3370                                               BTRFS_SUPER_INFO_SIZE);
3371                         if (!bh) {
3372                                 errors++;
3373                                 continue;
3374                         }
3375                         wait_on_buffer(bh);
3376                         if (!buffer_uptodate(bh))
3377                                 errors++;
3378
3379                         /* drop our reference */
3380                         brelse(bh);
3381
3382                         /* drop the reference from the wait == 0 run */
3383                         brelse(bh);
3384                         continue;
3385                 } else {
3386                         btrfs_set_super_bytenr(sb, bytenr);
3387
3388                         crc = ~(u32)0;
3389                         crc = btrfs_csum_data((char *)sb +
3390                                               BTRFS_CSUM_SIZE, crc,
3391                                               BTRFS_SUPER_INFO_SIZE -
3392                                               BTRFS_CSUM_SIZE);
3393                         btrfs_csum_final(crc, sb->csum);
3394
3395                         /*
3396                          * one reference for us, and we leave it for the
3397                          * caller
3398                          */
3399                         bh = __getblk(device->bdev, bytenr / 4096,
3400                                       BTRFS_SUPER_INFO_SIZE);
3401                         if (!bh) {
3402                                 btrfs_err(device->dev_root->fs_info,
3403                                     "couldn't get super buffer head for bytenr %llu",
3404                                     bytenr);
3405                                 errors++;
3406                                 continue;
3407                         }
3408
3409                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3410
3411                         /* one reference for submit_bh */
3412                         get_bh(bh);
3413
3414                         set_buffer_uptodate(bh);
3415                         lock_buffer(bh);
3416                         bh->b_end_io = btrfs_end_buffer_write_sync;
3417                         bh->b_private = device;
3418                 }
3419
3420                 /*
3421                  * we fua the first super.  The others we allow
3422                  * to go down lazy.
3423                  */
3424                 if (i == 0)
3425                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3426                 else
3427                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3428                 if (ret)
3429                         errors++;
3430         }
3431         return errors < i ? 0 : -1;
3432 }
3433
3434 /*
3435  * endio for the write_dev_flush, this will wake anyone waiting
3436  * for the barrier when it is done
3437  */
3438 static void btrfs_end_empty_barrier(struct bio *bio)
3439 {
3440         if (bio->bi_private)
3441                 complete(bio->bi_private);
3442         bio_put(bio);
3443 }
3444
3445 /*
3446  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3447  * sent down.  With wait == 1, it waits for the previous flush.
3448  *
3449  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3450  * capable
3451  */
3452 static int write_dev_flush(struct btrfs_device *device, int wait)
3453 {
3454         struct bio *bio;
3455         int ret = 0;
3456
3457         if (device->nobarriers)
3458                 return 0;
3459
3460         if (wait) {
3461                 bio = device->flush_bio;
3462                 if (!bio)
3463                         return 0;
3464
3465                 wait_for_completion(&device->flush_wait);
3466
3467                 if (bio->bi_error) {
3468                         ret = bio->bi_error;
3469                         btrfs_dev_stat_inc_and_print(device,
3470                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3471                 }
3472
3473                 /* drop the reference from the wait == 0 run */
3474                 bio_put(bio);
3475                 device->flush_bio = NULL;
3476
3477                 return ret;
3478         }
3479
3480         /*
3481          * one reference for us, and we leave it for the
3482          * caller
3483          */
3484         device->flush_bio = NULL;
3485         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3486         if (!bio)
3487                 return -ENOMEM;
3488
3489         bio->bi_end_io = btrfs_end_empty_barrier;
3490         bio->bi_bdev = device->bdev;
3491         init_completion(&device->flush_wait);
3492         bio->bi_private = &device->flush_wait;
3493         device->flush_bio = bio;
3494
3495         bio_get(bio);
3496         btrfsic_submit_bio(WRITE_FLUSH, bio);
3497
3498         return 0;
3499 }
3500
3501 /*
3502  * send an empty flush down to each device in parallel,
3503  * then wait for them
3504  */
3505 static int barrier_all_devices(struct btrfs_fs_info *info)
3506 {
3507         struct list_head *head;
3508         struct btrfs_device *dev;
3509         int errors_send = 0;
3510         int errors_wait = 0;
3511         int ret;
3512
3513         /* send down all the barriers */
3514         head = &info->fs_devices->devices;
3515         list_for_each_entry_rcu(dev, head, dev_list) {
3516                 if (dev->missing)
3517                         continue;
3518                 if (!dev->bdev) {
3519                         errors_send++;
3520                         continue;
3521                 }
3522                 if (!dev->in_fs_metadata || !dev->writeable)
3523                         continue;
3524
3525                 ret = write_dev_flush(dev, 0);
3526                 if (ret)
3527                         errors_send++;
3528         }
3529
3530         /* wait for all the barriers */
3531         list_for_each_entry_rcu(dev, head, dev_list) {
3532                 if (dev->missing)
3533                         continue;
3534                 if (!dev->bdev) {
3535                         errors_wait++;
3536                         continue;
3537                 }
3538                 if (!dev->in_fs_metadata || !dev->writeable)
3539                         continue;
3540
3541                 ret = write_dev_flush(dev, 1);
3542                 if (ret)
3543                         errors_wait++;
3544         }
3545         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3546             errors_wait > info->num_tolerated_disk_barrier_failures)
3547                 return -EIO;
3548         return 0;
3549 }
3550
3551 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3552 {
3553         int raid_type;
3554         int min_tolerated = INT_MAX;
3555
3556         if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3557             (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3558                 min_tolerated = min(min_tolerated,
3559                                     btrfs_raid_array[BTRFS_RAID_SINGLE].
3560                                     tolerated_failures);
3561
3562         for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3563                 if (raid_type == BTRFS_RAID_SINGLE)
3564                         continue;
3565                 if (!(flags & btrfs_raid_group[raid_type]))
3566                         continue;
3567                 min_tolerated = min(min_tolerated,
3568                                     btrfs_raid_array[raid_type].
3569                                     tolerated_failures);
3570         }
3571
3572         if (min_tolerated == INT_MAX) {
3573                 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3574                 min_tolerated = 0;
3575         }
3576
3577         return min_tolerated;
3578 }
3579
3580 int btrfs_calc_num_tolerated_disk_barrier_failures(
3581         struct btrfs_fs_info *fs_info)
3582 {
3583         struct btrfs_ioctl_space_info space;
3584         struct btrfs_space_info *sinfo;
3585         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3586                        BTRFS_BLOCK_GROUP_SYSTEM,
3587                        BTRFS_BLOCK_GROUP_METADATA,
3588                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3589         int i;
3590         int c;
3591         int num_tolerated_disk_barrier_failures =
3592                 (int)fs_info->fs_devices->num_devices;
3593
3594         for (i = 0; i < ARRAY_SIZE(types); i++) {
3595                 struct btrfs_space_info *tmp;
3596
3597                 sinfo = NULL;
3598                 rcu_read_lock();
3599                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3600                         if (tmp->flags == types[i]) {
3601                                 sinfo = tmp;
3602                                 break;
3603                         }
3604                 }
3605                 rcu_read_unlock();
3606
3607                 if (!sinfo)
3608                         continue;
3609
3610                 down_read(&sinfo->groups_sem);
3611                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3612                         u64 flags;
3613
3614                         if (list_empty(&sinfo->block_groups[c]))
3615                                 continue;
3616
3617                         btrfs_get_block_group_info(&sinfo->block_groups[c],
3618                                                    &space);
3619                         if (space.total_bytes == 0 || space.used_bytes == 0)
3620                                 continue;
3621                         flags = space.flags;
3622
3623                         num_tolerated_disk_barrier_failures = min(
3624                                 num_tolerated_disk_barrier_failures,
3625                                 btrfs_get_num_tolerated_disk_barrier_failures(
3626                                         flags));
3627                 }
3628                 up_read(&sinfo->groups_sem);
3629         }
3630
3631         return num_tolerated_disk_barrier_failures;
3632 }
3633
3634 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3635 {
3636         struct list_head *head;
3637         struct btrfs_device *dev;
3638         struct btrfs_super_block *sb;
3639         struct btrfs_dev_item *dev_item;
3640         int ret;
3641         int do_barriers;
3642         int max_errors;
3643         int total_errors = 0;
3644         u64 flags;
3645
3646         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3647         backup_super_roots(root->fs_info);
3648
3649         sb = root->fs_info->super_for_commit;
3650         dev_item = &sb->dev_item;
3651
3652         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3653         head = &root->fs_info->fs_devices->devices;
3654         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3655
3656         if (do_barriers) {
3657                 ret = barrier_all_devices(root->fs_info);
3658                 if (ret) {
3659                         mutex_unlock(
3660                                 &root->fs_info->fs_devices->device_list_mutex);
3661                         btrfs_handle_fs_error(root->fs_info, ret,
3662                                     "errors while submitting device barriers.");
3663                         return ret;
3664                 }
3665         }
3666
3667         list_for_each_entry_rcu(dev, head, dev_list) {
3668                 if (!dev->bdev) {
3669                         total_errors++;
3670                         continue;
3671                 }
3672                 if (!dev->in_fs_metadata || !dev->writeable)
3673                         continue;
3674
3675                 btrfs_set_stack_device_generation(dev_item, 0);
3676                 btrfs_set_stack_device_type(dev_item, dev->type);
3677                 btrfs_set_stack_device_id(dev_item, dev->devid);
3678                 btrfs_set_stack_device_total_bytes(dev_item,
3679                                                    dev->commit_total_bytes);
3680                 btrfs_set_stack_device_bytes_used(dev_item,
3681                                                   dev->commit_bytes_used);
3682                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3683                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3684                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3685                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3686                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3687
3688                 flags = btrfs_super_flags(sb);
3689                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3690
3691                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3692                 if (ret)
3693                         total_errors++;
3694         }
3695         if (total_errors > max_errors) {
3696                 btrfs_err(root->fs_info, "%d errors while writing supers",
3697                        total_errors);
3698                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3699
3700                 /* FUA is masked off if unsupported and can't be the reason */
3701                 btrfs_handle_fs_error(root->fs_info, -EIO,
3702                             "%d errors while writing supers", total_errors);
3703                 return -EIO;
3704         }
3705
3706         total_errors = 0;
3707         list_for_each_entry_rcu(dev, head, dev_list) {
3708                 if (!dev->bdev)
3709                         continue;
3710                 if (!dev->in_fs_metadata || !dev->writeable)
3711                         continue;
3712
3713                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3714                 if (ret)
3715                         total_errors++;
3716         }
3717         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3718         if (total_errors > max_errors) {
3719                 btrfs_handle_fs_error(root->fs_info, -EIO,
3720                             "%d errors while writing supers", total_errors);
3721                 return -EIO;
3722         }
3723         return 0;
3724 }
3725
3726 int write_ctree_super(struct btrfs_trans_handle *trans,
3727                       struct btrfs_root *root, int max_mirrors)
3728 {
3729         return write_all_supers(root, max_mirrors);
3730 }
3731
3732 /* Drop a fs root from the radix tree and free it. */
3733 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3734                                   struct btrfs_root *root)
3735 {
3736         spin_lock(&fs_info->fs_roots_radix_lock);
3737         radix_tree_delete(&fs_info->fs_roots_radix,
3738                           (unsigned long)root->root_key.objectid);
3739         spin_unlock(&fs_info->fs_roots_radix_lock);
3740
3741         if (btrfs_root_refs(&root->root_item) == 0)
3742                 synchronize_srcu(&fs_info->subvol_srcu);
3743
3744         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3745                 btrfs_free_log(NULL, root);
3746
3747         if (root->free_ino_pinned)
3748                 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3749         if (root->free_ino_ctl)
3750                 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3751         free_fs_root(root);
3752 }
3753
3754 static void free_fs_root(struct btrfs_root *root)
3755 {
3756         iput(root->ino_cache_inode);
3757         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3758         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3759         root->orphan_block_rsv = NULL;
3760         if (root->anon_dev)
3761                 free_anon_bdev(root->anon_dev);
3762         if (root->subv_writers)
3763                 btrfs_free_subvolume_writers(root->subv_writers);
3764         free_extent_buffer(root->node);
3765         free_extent_buffer(root->commit_root);
3766         kfree(root->free_ino_ctl);
3767         kfree(root->free_ino_pinned);
3768         kfree(root->name);
3769         btrfs_put_fs_root(root);
3770 }
3771
3772 void btrfs_free_fs_root(struct btrfs_root *root)
3773 {
3774         free_fs_root(root);
3775 }
3776
3777 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3778 {
3779         u64 root_objectid = 0;
3780         struct btrfs_root *gang[8];
3781         int i = 0;
3782         int err = 0;
3783         unsigned int ret = 0;
3784         int index;
3785
3786         while (1) {
3787                 index = srcu_read_lock(&fs_info->subvol_srcu);
3788                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3789                                              (void **)gang, root_objectid,
3790                                              ARRAY_SIZE(gang));
3791                 if (!ret) {
3792                         srcu_read_unlock(&fs_info->subvol_srcu, index);
3793                         break;
3794                 }
3795                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3796
3797                 for (i = 0; i < ret; i++) {
3798                         /* Avoid to grab roots in dead_roots */
3799                         if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3800                                 gang[i] = NULL;
3801                                 continue;
3802                         }
3803                         /* grab all the search result for later use */
3804                         gang[i] = btrfs_grab_fs_root(gang[i]);
3805                 }
3806                 srcu_read_unlock(&fs_info->subvol_srcu, index);
3807
3808                 for (i = 0; i < ret; i++) {
3809                         if (!gang[i])
3810                                 continue;
3811                         root_objectid = gang[i]->root_key.objectid;
3812                         err = btrfs_orphan_cleanup(gang[i]);
3813                         if (err)
3814                                 break;
3815                         btrfs_put_fs_root(gang[i]);
3816                 }
3817                 root_objectid++;
3818         }
3819
3820         /* release the uncleaned roots due to error */
3821         for (; i < ret; i++) {
3822                 if (gang[i])
3823                         btrfs_put_fs_root(gang[i]);
3824         }
3825         return err;
3826 }
3827
3828 int btrfs_commit_super(struct btrfs_root *root)
3829 {
3830         struct btrfs_trans_handle *trans;
3831
3832         mutex_lock(&root->fs_info->cleaner_mutex);
3833         btrfs_run_delayed_iputs(root);
3834         mutex_unlock(&root->fs_info->cleaner_mutex);
3835         wake_up_process(root->fs_info->cleaner_kthread);
3836
3837         /* wait until ongoing cleanup work done */
3838         down_write(&root->fs_info->cleanup_work_sem);
3839         up_write(&root->fs_info->cleanup_work_sem);
3840
3841         trans = btrfs_join_transaction(root);
3842         if (IS_ERR(trans))
3843                 return PTR_ERR(trans);
3844         return btrfs_commit_transaction(trans, root);
3845 }
3846
3847 void close_ctree(struct btrfs_root *root)
3848 {
3849         struct btrfs_fs_info *fs_info = root->fs_info;
3850         int ret;
3851
3852         fs_info->closing = 1;
3853         smp_mb();
3854
3855         /* wait for the qgroup rescan worker to stop */
3856         btrfs_qgroup_wait_for_completion(fs_info);
3857
3858         /* wait for the uuid_scan task to finish */
3859         down(&fs_info->uuid_tree_rescan_sem);
3860         /* avoid complains from lockdep et al., set sem back to initial state */
3861         up(&fs_info->uuid_tree_rescan_sem);
3862
3863         /* pause restriper - we want to resume on mount */
3864         btrfs_pause_balance(fs_info);
3865
3866         btrfs_dev_replace_suspend_for_unmount(fs_info);
3867
3868         btrfs_scrub_cancel(fs_info);
3869
3870         /* wait for any defraggers to finish */
3871         wait_event(fs_info->transaction_wait,
3872                    (atomic_read(&fs_info->defrag_running) == 0));
3873
3874         /* clear out the rbtree of defraggable inodes */
3875         btrfs_cleanup_defrag_inodes(fs_info);
3876
3877         cancel_work_sync(&fs_info->async_reclaim_work);
3878
3879         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3880                 /*
3881                  * If the cleaner thread is stopped and there are
3882                  * block groups queued for removal, the deletion will be
3883                  * skipped when we quit the cleaner thread.
3884                  */
3885                 btrfs_delete_unused_bgs(root->fs_info);
3886
3887                 ret = btrfs_commit_super(root);
3888                 if (ret)
3889                         btrfs_err(fs_info, "commit super ret %d", ret);
3890         }
3891
3892         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3893                 btrfs_error_commit_super(root);
3894
3895         kthread_stop(fs_info->transaction_kthread);
3896         kthread_stop(fs_info->cleaner_kthread);
3897
3898         fs_info->closing = 2;
3899         smp_mb();
3900
3901         btrfs_free_qgroup_config(fs_info);
3902
3903         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3904                 btrfs_info(fs_info, "at unmount delalloc count %lld",
3905                        percpu_counter_sum(&fs_info->delalloc_bytes));
3906         }
3907
3908         btrfs_sysfs_remove_mounted(fs_info);
3909         btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3910
3911         btrfs_free_fs_roots(fs_info);
3912
3913         btrfs_put_block_group_cache(fs_info);
3914
3915         btrfs_free_block_groups(fs_info);
3916
3917         /*
3918          * we must make sure there is not any read request to
3919          * submit after we stopping all workers.
3920          */
3921         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3922         btrfs_stop_all_workers(fs_info);
3923
3924         fs_info->open = 0;
3925         free_root_pointers(fs_info, 1);
3926
3927         iput(fs_info->btree_inode);
3928
3929 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3930         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3931                 btrfsic_unmount(root, fs_info->fs_devices);
3932 #endif
3933
3934         btrfs_close_devices(fs_info->fs_devices);
3935         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3936
3937         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3938         percpu_counter_destroy(&fs_info->delalloc_bytes);
3939         percpu_counter_destroy(&fs_info->bio_counter);
3940         bdi_destroy(&fs_info->bdi);
3941         cleanup_srcu_struct(&fs_info->subvol_srcu);
3942
3943         btrfs_free_stripe_hash_table(fs_info);
3944
3945         __btrfs_free_block_rsv(root->orphan_block_rsv);
3946         root->orphan_block_rsv = NULL;
3947
3948         lock_chunks(root);
3949         while (!list_empty(&fs_info->pinned_chunks)) {
3950                 struct extent_map *em;
3951
3952                 em = list_first_entry(&fs_info->pinned_chunks,
3953                                       struct extent_map, list);
3954                 list_del_init(&em->list);
3955                 free_extent_map(em);
3956         }
3957         unlock_chunks(root);
3958 }
3959
3960 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3961                           int atomic)
3962 {
3963         int ret;
3964         struct inode *btree_inode = buf->pages[0]->mapping->host;
3965
3966         ret = extent_buffer_uptodate(buf);
3967         if (!ret)
3968                 return ret;
3969
3970         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3971                                     parent_transid, atomic);
3972         if (ret == -EAGAIN)
3973                 return ret;
3974         return !ret;
3975 }
3976
3977 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3978 {
3979         struct btrfs_root *root;
3980         u64 transid = btrfs_header_generation(buf);
3981         int was_dirty;
3982
3983 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3984         /*
3985          * This is a fast path so only do this check if we have sanity tests
3986          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3987          * outside of the sanity tests.
3988          */
3989         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3990                 return;
3991 #endif
3992         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3993         btrfs_assert_tree_locked(buf);
3994         if (transid != root->fs_info->generation)
3995                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3996                        "found %llu running %llu\n",
3997                         buf->start, transid, root->fs_info->generation);
3998         was_dirty = set_extent_buffer_dirty(buf);
3999         if (!was_dirty)
4000                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4001                                      buf->len,
4002                                      root->fs_info->dirty_metadata_batch);
4003 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4004         if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4005                 btrfs_print_leaf(root, buf);
4006                 ASSERT(0);
4007         }
4008 #endif
4009 }
4010
4011 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4012                                         int flush_delayed)
4013 {
4014         /*
4015          * looks as though older kernels can get into trouble with
4016          * this code, they end up stuck in balance_dirty_pages forever
4017          */
4018         int ret;
4019
4020         if (current->flags & PF_MEMALLOC)
4021                 return;
4022
4023         if (flush_delayed)
4024                 btrfs_balance_delayed_items(root);
4025
4026         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4027                                      BTRFS_DIRTY_METADATA_THRESH);
4028         if (ret > 0) {
4029                 balance_dirty_pages_ratelimited(
4030                                    root->fs_info->btree_inode->i_mapping);
4031         }
4032 }
4033
4034 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4035 {
4036         __btrfs_btree_balance_dirty(root, 1);
4037 }
4038
4039 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4040 {
4041         __btrfs_btree_balance_dirty(root, 0);
4042 }
4043
4044 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4045 {
4046         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4047         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4048 }
4049
4050 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4051                               int read_only)
4052 {
4053         struct btrfs_super_block *sb = fs_info->super_copy;
4054         u64 nodesize = btrfs_super_nodesize(sb);
4055         u64 sectorsize = btrfs_super_sectorsize(sb);
4056         int ret = 0;
4057
4058         if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4059                 printk(KERN_ERR "BTRFS: no valid FS found\n");
4060                 ret = -EINVAL;
4061         }
4062         if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4063                 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4064                                 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4065         if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4066                 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4067                                 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4068                 ret = -EINVAL;
4069         }
4070         if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4071                 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4072                                 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4073                 ret = -EINVAL;
4074         }
4075         if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4076                 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4077                                 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4078                 ret = -EINVAL;
4079         }
4080
4081         /*
4082          * Check sectorsize and nodesize first, other check will need it.
4083          * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4084          */
4085         if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4086             sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4087                 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4088                 ret = -EINVAL;
4089         }
4090         /* Only PAGE SIZE is supported yet */
4091         if (sectorsize != PAGE_SIZE) {
4092                 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4093                                 sectorsize, PAGE_SIZE);
4094                 ret = -EINVAL;
4095         }
4096         if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4097             nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4098                 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4099                 ret = -EINVAL;
4100         }
4101         if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4102                 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4103                                 le32_to_cpu(sb->__unused_leafsize),
4104                                 nodesize);
4105                 ret = -EINVAL;
4106         }
4107
4108         /* Root alignment check */
4109         if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4110                 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4111                                 btrfs_super_root(sb));
4112                 ret = -EINVAL;
4113         }
4114         if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4115                 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4116                                 btrfs_super_chunk_root(sb));
4117                 ret = -EINVAL;
4118         }
4119         if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4120                 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4121                                 btrfs_super_log_root(sb));
4122                 ret = -EINVAL;
4123         }
4124
4125         if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4126                 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4127                                 fs_info->fsid, sb->dev_item.fsid);
4128                 ret = -EINVAL;
4129         }
4130
4131         /*
4132          * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4133          * done later
4134          */
4135         if (btrfs_super_num_devices(sb) > (1UL << 31))
4136                 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4137                                 btrfs_super_num_devices(sb));
4138         if (btrfs_super_num_devices(sb) == 0) {
4139                 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4140                 ret = -EINVAL;
4141         }
4142
4143         if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4144                 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4145                                 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4146                 ret = -EINVAL;
4147         }
4148
4149         /*
4150          * Obvious sys_chunk_array corruptions, it must hold at least one key
4151          * and one chunk
4152          */
4153         if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4154                 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4155                                 btrfs_super_sys_array_size(sb),
4156                                 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4157                 ret = -EINVAL;
4158         }
4159         if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4160                         + sizeof(struct btrfs_chunk)) {
4161                 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4162                                 btrfs_super_sys_array_size(sb),
4163                                 sizeof(struct btrfs_disk_key)
4164                                 + sizeof(struct btrfs_chunk));
4165                 ret = -EINVAL;
4166         }
4167
4168         /*
4169          * The generation is a global counter, we'll trust it more than the others
4170          * but it's still possible that it's the one that's wrong.
4171          */
4172         if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4173                 printk(KERN_WARNING
4174                         "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4175                         btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4176         if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4177             && btrfs_super_cache_generation(sb) != (u64)-1)
4178                 printk(KERN_WARNING
4179                         "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4180                         btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4181
4182         return ret;
4183 }
4184
4185 static void btrfs_error_commit_super(struct btrfs_root *root)
4186 {
4187         mutex_lock(&root->fs_info->cleaner_mutex);
4188         btrfs_run_delayed_iputs(root);
4189         mutex_unlock(&root->fs_info->cleaner_mutex);
4190
4191         down_write(&root->fs_info->cleanup_work_sem);
4192         up_write(&root->fs_info->cleanup_work_sem);
4193
4194         /* cleanup FS via transaction */
4195         btrfs_cleanup_transaction(root);
4196 }
4197
4198 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4199 {
4200         struct btrfs_ordered_extent *ordered;
4201
4202         spin_lock(&root->ordered_extent_lock);
4203         /*
4204          * This will just short circuit the ordered completion stuff which will
4205          * make sure the ordered extent gets properly cleaned up.
4206          */
4207         list_for_each_entry(ordered, &root->ordered_extents,
4208                             root_extent_list)
4209                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4210         spin_unlock(&root->ordered_extent_lock);
4211 }
4212
4213 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4214 {
4215         struct btrfs_root *root;
4216         struct list_head splice;
4217
4218         INIT_LIST_HEAD(&splice);
4219
4220         spin_lock(&fs_info->ordered_root_lock);
4221         list_splice_init(&fs_info->ordered_roots, &splice);
4222         while (!list_empty(&splice)) {
4223                 root = list_first_entry(&splice, struct btrfs_root,
4224                                         ordered_root);
4225                 list_move_tail(&root->ordered_root,
4226                                &fs_info->ordered_roots);
4227
4228                 spin_unlock(&fs_info->ordered_root_lock);
4229                 btrfs_destroy_ordered_extents(root);
4230
4231                 cond_resched();
4232                 spin_lock(&fs_info->ordered_root_lock);
4233         }
4234         spin_unlock(&fs_info->ordered_root_lock);
4235 }
4236
4237 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4238                                       struct btrfs_root *root)
4239 {
4240         struct rb_node *node;
4241         struct btrfs_delayed_ref_root *delayed_refs;
4242         struct btrfs_delayed_ref_node *ref;
4243         int ret = 0;
4244
4245         delayed_refs = &trans->delayed_refs;
4246
4247         spin_lock(&delayed_refs->lock);
4248         if (atomic_read(&delayed_refs->num_entries) == 0) {
4249                 spin_unlock(&delayed_refs->lock);
4250                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4251                 return ret;
4252         }
4253
4254         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4255                 struct btrfs_delayed_ref_head *head;
4256                 struct btrfs_delayed_ref_node *tmp;
4257                 bool pin_bytes = false;
4258
4259                 head = rb_entry(node, struct btrfs_delayed_ref_head,
4260                                 href_node);
4261                 if (!mutex_trylock(&head->mutex)) {
4262                         atomic_inc(&head->node.refs);
4263                         spin_unlock(&delayed_refs->lock);
4264
4265                         mutex_lock(&head->mutex);
4266                         mutex_unlock(&head->mutex);
4267                         btrfs_put_delayed_ref(&head->node);
4268                         spin_lock(&delayed_refs->lock);
4269                         continue;
4270                 }
4271                 spin_lock(&head->lock);
4272                 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4273                                                  list) {
4274                         ref->in_tree = 0;
4275                         list_del(&ref->list);
4276                         atomic_dec(&delayed_refs->num_entries);
4277                         btrfs_put_delayed_ref(ref);
4278                 }
4279                 if (head->must_insert_reserved)
4280                         pin_bytes = true;
4281                 btrfs_free_delayed_extent_op(head->extent_op);
4282                 delayed_refs->num_heads--;
4283                 if (head->processing == 0)
4284                         delayed_refs->num_heads_ready--;
4285                 atomic_dec(&delayed_refs->num_entries);
4286                 head->node.in_tree = 0;
4287                 rb_erase(&head->href_node, &delayed_refs->href_root);
4288                 spin_unlock(&head->lock);
4289                 spin_unlock(&delayed_refs->lock);
4290                 mutex_unlock(&head->mutex);
4291
4292                 if (pin_bytes)
4293                         btrfs_pin_extent(root, head->node.bytenr,
4294                                          head->node.num_bytes, 1);
4295                 btrfs_put_delayed_ref(&head->node);
4296                 cond_resched();
4297                 spin_lock(&delayed_refs->lock);
4298         }
4299
4300         spin_unlock(&delayed_refs->lock);
4301
4302         return ret;
4303 }
4304
4305 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4306 {
4307         struct btrfs_inode *btrfs_inode;
4308         struct list_head splice;
4309
4310         INIT_LIST_HEAD(&splice);
4311
4312         spin_lock(&root->delalloc_lock);
4313         list_splice_init(&root->delalloc_inodes, &splice);
4314
4315         while (!list_empty(&splice)) {
4316                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4317                                                delalloc_inodes);
4318
4319                 list_del_init(&btrfs_inode->delalloc_inodes);
4320                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4321                           &btrfs_inode->runtime_flags);
4322                 spin_unlock(&root->delalloc_lock);
4323
4324                 btrfs_invalidate_inodes(btrfs_inode->root);
4325
4326                 spin_lock(&root->delalloc_lock);
4327         }
4328
4329         spin_unlock(&root->delalloc_lock);
4330 }
4331
4332 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4333 {
4334         struct btrfs_root *root;
4335         struct list_head splice;
4336
4337         INIT_LIST_HEAD(&splice);
4338
4339         spin_lock(&fs_info->delalloc_root_lock);
4340         list_splice_init(&fs_info->delalloc_roots, &splice);
4341         while (!list_empty(&splice)) {
4342                 root = list_first_entry(&splice, struct btrfs_root,
4343                                          delalloc_root);
4344                 list_del_init(&root->delalloc_root);
4345                 root = btrfs_grab_fs_root(root);
4346                 BUG_ON(!root);
4347                 spin_unlock(&fs_info->delalloc_root_lock);
4348
4349                 btrfs_destroy_delalloc_inodes(root);
4350                 btrfs_put_fs_root(root);
4351
4352                 spin_lock(&fs_info->delalloc_root_lock);
4353         }
4354         spin_unlock(&fs_info->delalloc_root_lock);
4355 }
4356
4357 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4358                                         struct extent_io_tree *dirty_pages,
4359                                         int mark)
4360 {
4361         int ret;
4362         struct extent_buffer *eb;
4363         u64 start = 0;
4364         u64 end;
4365
4366         while (1) {
4367                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4368                                             mark, NULL);
4369                 if (ret)
4370                         break;
4371
4372                 clear_extent_bits(dirty_pages, start, end, mark);
4373                 while (start <= end) {
4374                         eb = btrfs_find_tree_block(root->fs_info, start);
4375                         start += root->nodesize;
4376                         if (!eb)
4377                                 continue;
4378                         wait_on_extent_buffer_writeback(eb);
4379
4380                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4381                                                &eb->bflags))
4382                                 clear_extent_buffer_dirty(eb);
4383                         free_extent_buffer_stale(eb);
4384                 }
4385         }
4386
4387         return ret;
4388 }
4389
4390 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4391                                        struct extent_io_tree *pinned_extents)
4392 {
4393         struct extent_io_tree *unpin;
4394         u64 start;
4395         u64 end;
4396         int ret;
4397         bool loop = true;
4398
4399         unpin = pinned_extents;
4400 again:
4401         while (1) {
4402                 ret = find_first_extent_bit(unpin, 0, &start, &end,
4403                                             EXTENT_DIRTY, NULL);
4404                 if (ret)
4405                         break;
4406
4407                 clear_extent_dirty(unpin, start, end);
4408                 btrfs_error_unpin_extent_range(root, start, end);
4409                 cond_resched();
4410         }
4411
4412         if (loop) {
4413                 if (unpin == &root->fs_info->freed_extents[0])
4414                         unpin = &root->fs_info->freed_extents[1];
4415                 else
4416                         unpin = &root->fs_info->freed_extents[0];
4417                 loop = false;
4418                 goto again;
4419         }
4420
4421         return 0;
4422 }
4423
4424 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4425                                    struct btrfs_root *root)
4426 {
4427         btrfs_destroy_delayed_refs(cur_trans, root);
4428
4429         cur_trans->state = TRANS_STATE_COMMIT_START;
4430         wake_up(&root->fs_info->transaction_blocked_wait);
4431
4432         cur_trans->state = TRANS_STATE_UNBLOCKED;
4433         wake_up(&root->fs_info->transaction_wait);
4434
4435         btrfs_destroy_delayed_inodes(root);
4436         btrfs_assert_delayed_root_empty(root);
4437
4438         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4439                                      EXTENT_DIRTY);
4440         btrfs_destroy_pinned_extent(root,
4441                                     root->fs_info->pinned_extents);
4442
4443         cur_trans->state =TRANS_STATE_COMPLETED;
4444         wake_up(&cur_trans->commit_wait);
4445
4446         /*
4447         memset(cur_trans, 0, sizeof(*cur_trans));
4448         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4449         */
4450 }
4451
4452 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4453 {
4454         struct btrfs_transaction *t;
4455
4456         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4457
4458         spin_lock(&root->fs_info->trans_lock);
4459         while (!list_empty(&root->fs_info->trans_list)) {
4460                 t = list_first_entry(&root->fs_info->trans_list,
4461                                      struct btrfs_transaction, list);
4462                 if (t->state >= TRANS_STATE_COMMIT_START) {
4463                         atomic_inc(&t->use_count);
4464                         spin_unlock(&root->fs_info->trans_lock);
4465                         btrfs_wait_for_commit(root, t->transid);
4466                         btrfs_put_transaction(t);
4467                         spin_lock(&root->fs_info->trans_lock);
4468                         continue;
4469                 }
4470                 if (t == root->fs_info->running_transaction) {
4471                         t->state = TRANS_STATE_COMMIT_DOING;
4472                         spin_unlock(&root->fs_info->trans_lock);
4473                         /*
4474                          * We wait for 0 num_writers since we don't hold a trans
4475                          * handle open currently for this transaction.
4476                          */
4477                         wait_event(t->writer_wait,
4478                                    atomic_read(&t->num_writers) == 0);
4479                 } else {
4480                         spin_unlock(&root->fs_info->trans_lock);
4481                 }
4482                 btrfs_cleanup_one_transaction(t, root);
4483
4484                 spin_lock(&root->fs_info->trans_lock);
4485                 if (t == root->fs_info->running_transaction)
4486                         root->fs_info->running_transaction = NULL;
4487                 list_del_init(&t->list);
4488                 spin_unlock(&root->fs_info->trans_lock);
4489
4490                 btrfs_put_transaction(t);
4491                 trace_btrfs_transaction_commit(root);
4492                 spin_lock(&root->fs_info->trans_lock);
4493         }
4494         spin_unlock(&root->fs_info->trans_lock);
4495         btrfs_destroy_all_ordered_extents(root->fs_info);
4496         btrfs_destroy_delayed_inodes(root);
4497         btrfs_assert_delayed_root_empty(root);
4498         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4499         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4500         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4501
4502         return 0;
4503 }
4504
4505 static const struct extent_io_ops btree_extent_io_ops = {
4506         .readpage_end_io_hook = btree_readpage_end_io_hook,
4507         .readpage_io_failed_hook = btree_io_failed_hook,
4508         .submit_bio_hook = btree_submit_bio_hook,
4509         /* note we're sharing with inode.c for the merge bio hook */
4510         .merge_bio_hook = btrfs_merge_bio_hook,
4511 };