projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: do not hold the file extent leaf locked when adding extent item
[linux-2.6-block.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
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/freezer.h>
29#include <linux/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <asm/unaligned.h>
34#include "compat.h"
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "async-thread.h"
42#include "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48
49#ifdef CONFIG_X86
50#include <asm/cpufeature.h>
51#endif
52
53static struct extent_io_ops btree_extent_io_ops;
54static void end_workqueue_fn(struct btrfs_work *work);
55static void free_fs_root(struct btrfs_root *root);
56static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
57 int read_only);
58static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
59static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
60static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
61 struct btrfs_root *root);
62static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
63static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
64static int btrfs_destroy_marked_extents(struct btrfs_root *root,
65 struct extent_io_tree *dirty_pages,
66 int mark);
67static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
68 struct extent_io_tree *pinned_extents);
69
70/*
71 * end_io_wq structs are used to do processing in task context when an IO is
72 * complete. This is used during reads to verify checksums, and it is used
73 * by writes to insert metadata for new file extents after IO is complete.
74 */
75struct end_io_wq {
76 struct bio *bio;
77 bio_end_io_t *end_io;
78 void *private;
79 struct btrfs_fs_info *info;
80 int error;
81 int metadata;
82 struct list_head list;
83 struct btrfs_work work;
84};
85
86/*
87 * async submit bios are used to offload expensive checksumming
88 * onto the worker threads. They checksum file and metadata bios
89 * just before they are sent down the IO stack.
90 */
91struct async_submit_bio {
92 struct inode *inode;
93 struct bio *bio;
94 struct list_head list;
95 extent_submit_bio_hook_t *submit_bio_start;
96 extent_submit_bio_hook_t *submit_bio_done;
97 int rw;
98 int mirror_num;
99 unsigned long bio_flags;
100 /*
101 * bio_offset is optional, can be used if the pages in the bio
102 * can't tell us where in the file the bio should go
103 */
104 u64 bio_offset;
105 struct btrfs_work work;
106 int error;
107};
108
109/*
110 * Lockdep class keys for extent_buffer->lock's in this root. For a given
111 * eb, the lockdep key is determined by the btrfs_root it belongs to and
112 * the level the eb occupies in the tree.
113 *
114 * Different roots are used for different purposes and may nest inside each
115 * other and they require separate keysets. As lockdep keys should be
116 * static, assign keysets according to the purpose of the root as indicated
117 * by btrfs_root->objectid. This ensures that all special purpose roots
118 * have separate keysets.
119 *
120 * Lock-nesting across peer nodes is always done with the immediate parent
121 * node locked thus preventing deadlock. As lockdep doesn't know this, use
122 * subclass to avoid triggering lockdep warning in such cases.
123 *
124 * The key is set by the readpage_end_io_hook after the buffer has passed
125 * csum validation but before the pages are unlocked. It is also set by
126 * btrfs_init_new_buffer on freshly allocated blocks.
127 *
128 * We also add a check to make sure the highest level of the tree is the
129 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
130 * needs update as well.
131 */
132#ifdef CONFIG_DEBUG_LOCK_ALLOC
133# if BTRFS_MAX_LEVEL != 8
134# error
135# endif
136
137static struct btrfs_lockdep_keyset {
138 u64 id; /* root objectid */
139 const char *name_stem; /* lock name stem */
140 char names[BTRFS_MAX_LEVEL + 1][20];
141 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
142} btrfs_lockdep_keysets[] = {
143 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
144 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
145 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
146 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
147 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
148 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
149 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
150 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
151 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
152 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
153 { .id = 0, .name_stem = "tree" },
154};
155
156void __init btrfs_init_lockdep(void)
157{
158 int i, j;
159
160 /* initialize lockdep class names */
161 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
162 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
163
164 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
165 snprintf(ks->names[j], sizeof(ks->names[j]),
166 "btrfs-%s-%02d", ks->name_stem, j);
167 }
168}
169
170void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
171 int level)
172{
173 struct btrfs_lockdep_keyset *ks;
174
175 BUG_ON(level >= ARRAY_SIZE(ks->keys));
176
177 /* find the matching keyset, id 0 is the default entry */
178 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
179 if (ks->id == objectid)
180 break;
181
182 lockdep_set_class_and_name(&eb->lock,
183 &ks->keys[level], ks->names[level]);
184}
185
186#endif
187
188/*
189 * extents on the btree inode are pretty simple, there's one extent
190 * that covers the entire device
191 */
192static struct extent_map *btree_get_extent(struct inode *inode,
193 struct page *page, size_t pg_offset, u64 start, u64 len,
194 int create)
195{
196 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
197 struct extent_map *em;
198 int ret;
199
200 read_lock(&em_tree->lock);
201 em = lookup_extent_mapping(em_tree, start, len);
202 if (em) {
203 em->bdev =
204 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
205 read_unlock(&em_tree->lock);
206 goto out;
207 }
208 read_unlock(&em_tree->lock);
209
210 em = alloc_extent_map();
211 if (!em) {
212 em = ERR_PTR(-ENOMEM);
213 goto out;
214 }
215 em->start = 0;
216 em->len = (u64)-1;
217 em->block_len = (u64)-1;
218 em->block_start = 0;
219 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
220
221 write_lock(&em_tree->lock);
222 ret = add_extent_mapping(em_tree, em);
223 if (ret == -EEXIST) {
224 free_extent_map(em);
225 em = lookup_extent_mapping(em_tree, start, len);
226 if (!em)
227 em = ERR_PTR(-EIO);
228 } else if (ret) {
229 free_extent_map(em);
230 em = ERR_PTR(ret);
231 }
232 write_unlock(&em_tree->lock);
233
234out:
235 return em;
236}
237
238u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
239{
240 return crc32c(seed, data, len);
241}
242
243void btrfs_csum_final(u32 crc, char *result)
244{
245 put_unaligned_le32(~crc, result);
246}
247
248/*
249 * compute the csum for a btree block, and either verify it or write it
250 * into the csum field of the block.
251 */
252static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
253 int verify)
254{
255 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
256 char *result = NULL;
257 unsigned long len;
258 unsigned long cur_len;
259 unsigned long offset = BTRFS_CSUM_SIZE;
260 char *kaddr;
261 unsigned long map_start;
262 unsigned long map_len;
263 int err;
264 u32 crc = ~(u32)0;
265 unsigned long inline_result;
266
267 len = buf->len - offset;
268 while (len > 0) {
269 err = map_private_extent_buffer(buf, offset, 32,
270 &kaddr, &map_start, &map_len);
271 if (err)
272 return 1;
273 cur_len = min(len, map_len - (offset - map_start));
274 crc = btrfs_csum_data(root, kaddr + offset - map_start,
275 crc, cur_len);
276 len -= cur_len;
277 offset += cur_len;
278 }
279 if (csum_size > sizeof(inline_result)) {
280 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
281 if (!result)
282 return 1;
283 } else {
284 result = (char *)&inline_result;
285 }
286
287 btrfs_csum_final(crc, result);
288
289 if (verify) {
290 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
291 u32 val;
292 u32 found = 0;
293 memcpy(&found, result, csum_size);
294
295 read_extent_buffer(buf, &val, 0, csum_size);
296 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
297 "failed on %llu wanted %X found %X "
298 "level %d\n",
299 root->fs_info->sb->s_id,
300 (unsigned long long)buf->start, val, found,
301 btrfs_header_level(buf));
302 if (result != (char *)&inline_result)
303 kfree(result);
304 return 1;
305 }
306 } else {
307 write_extent_buffer(buf, result, 0, csum_size);
308 }
309 if (result != (char *)&inline_result)
310 kfree(result);
311 return 0;
312}
313
314/*
315 * we can't consider a given block up to date unless the transid of the
316 * block matches the transid in the parent node's pointer. This is how we
317 * detect blocks that either didn't get written at all or got written
318 * in the wrong place.
319 */
320static int verify_parent_transid(struct extent_io_tree *io_tree,
321 struct extent_buffer *eb, u64 parent_transid,
322 int atomic)
323{
324 struct extent_state *cached_state = NULL;
325 int ret;
326
327 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
328 return 0;
329
330 if (atomic)
331 return -EAGAIN;
332
333 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334 0, &cached_state);
335 if (extent_buffer_uptodate(eb) &&
336 btrfs_header_generation(eb) == parent_transid) {
337 ret = 0;
338 goto out;
339 }
340 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
341 "found %llu\n",
342 (unsigned long long)eb->start,
343 (unsigned long long)parent_transid,
344 (unsigned long long)btrfs_header_generation(eb));
345 ret = 1;
346 clear_extent_buffer_uptodate(eb);
347out:
348 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349 &cached_state, GFP_NOFS);
350 return ret;
351}
352
353/*
354 * helper to read a given tree block, doing retries as required when
355 * the checksums don't match and we have alternate mirrors to try.
356 */
357static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358 struct extent_buffer *eb,
359 u64 start, u64 parent_transid)
360{
361 struct extent_io_tree *io_tree;
362 int failed = 0;
363 int ret;
364 int num_copies = 0;
365 int mirror_num = 0;
366 int failed_mirror = 0;
367
368 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
369 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370 while (1) {
371 ret = read_extent_buffer_pages(io_tree, eb, start,
372 WAIT_COMPLETE,
373 btree_get_extent, mirror_num);
374 if (!ret) {
375 if (!verify_parent_transid(io_tree, eb,
376 parent_transid, 0))
377 break;
378 else
379 ret = -EIO;
380 }
381
382 /*
383 * This buffer's crc is fine, but its contents are corrupted, so
384 * there is no reason to read the other copies, they won't be
385 * any less wrong.
386 */
387 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
388 break;
389
390 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
391 eb->start, eb->len);
392 if (num_copies == 1)
393 break;
394
395 if (!failed_mirror) {
396 failed = 1;
397 failed_mirror = eb->read_mirror;
398 }
399
400 mirror_num++;
401 if (mirror_num == failed_mirror)
402 mirror_num++;
403
404 if (mirror_num > num_copies)
405 break;
406 }
407
408 if (failed && !ret && failed_mirror)
409 repair_eb_io_failure(root, eb, failed_mirror);
410
411 return ret;
412}
413
414/*
415 * checksum a dirty tree block before IO. This has extra checks to make sure
416 * we only fill in the checksum field in the first page of a multi-page block
417 */
418
419static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
420{
421 struct extent_io_tree *tree;
422 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
423 u64 found_start;
424 struct extent_buffer *eb;
425
426 tree = &BTRFS_I(page->mapping->host)->io_tree;
427
428 eb = (struct extent_buffer *)page->private;
429 if (page != eb->pages[0])
430 return 0;
431 found_start = btrfs_header_bytenr(eb);
432 if (found_start != start) {
433 WARN_ON(1);
434 return 0;
435 }
436 if (eb->pages[0] != page) {
437 WARN_ON(1);
438 return 0;
439 }
440 if (!PageUptodate(page)) {
441 WARN_ON(1);
442 return 0;
443 }
444 csum_tree_block(root, eb, 0);
445 return 0;
446}
447
448static int check_tree_block_fsid(struct btrfs_root *root,
449 struct extent_buffer *eb)
450{
451 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
452 u8 fsid[BTRFS_UUID_SIZE];
453 int ret = 1;
454
455 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
456 BTRFS_FSID_SIZE);
457 while (fs_devices) {
458 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
459 ret = 0;
460 break;
461 }
462 fs_devices = fs_devices->seed;
463 }
464 return ret;
465}
466
467#define CORRUPT(reason, eb, root, slot) \
468 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
469 "root=%llu, slot=%d\n", reason, \
470 (unsigned long long)btrfs_header_bytenr(eb), \
471 (unsigned long long)root->objectid, slot)
472
473static noinline int check_leaf(struct btrfs_root *root,
474 struct extent_buffer *leaf)
475{
476 struct btrfs_key key;
477 struct btrfs_key leaf_key;
478 u32 nritems = btrfs_header_nritems(leaf);
479 int slot;
480
481 if (nritems == 0)
482 return 0;
483
484 /* Check the 0 item */
485 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
486 BTRFS_LEAF_DATA_SIZE(root)) {
487 CORRUPT("invalid item offset size pair", leaf, root, 0);
488 return -EIO;
489 }
490
491 /*
492 * Check to make sure each items keys are in the correct order and their
493 * offsets make sense. We only have to loop through nritems-1 because
494 * we check the current slot against the next slot, which verifies the
495 * next slot's offset+size makes sense and that the current's slot
496 * offset is correct.
497 */
498 for (slot = 0; slot < nritems - 1; slot++) {
499 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
500 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
501
502 /* Make sure the keys are in the right order */
503 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
504 CORRUPT("bad key order", leaf, root, slot);
505 return -EIO;
506 }
507
508 /*
509 * Make sure the offset and ends are right, remember that the
510 * item data starts at the end of the leaf and grows towards the
511 * front.
512 */
513 if (btrfs_item_offset_nr(leaf, slot) !=
514 btrfs_item_end_nr(leaf, slot + 1)) {
515 CORRUPT("slot offset bad", leaf, root, slot);
516 return -EIO;
517 }
518
519 /*
520 * Check to make sure that we don't point outside of the leaf,
521 * just incase all the items are consistent to eachother, but
522 * all point outside of the leaf.
523 */
524 if (btrfs_item_end_nr(leaf, slot) >
525 BTRFS_LEAF_DATA_SIZE(root)) {
526 CORRUPT("slot end outside of leaf", leaf, root, slot);
527 return -EIO;
528 }
529 }
530
531 return 0;
532}
533
534struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
535 struct page *page, int max_walk)
536{
537 struct extent_buffer *eb;
538 u64 start = page_offset(page);
539 u64 target = start;
540 u64 min_start;
541
542 if (start < max_walk)
543 min_start = 0;
544 else
545 min_start = start - max_walk;
546
547 while (start >= min_start) {
548 eb = find_extent_buffer(tree, start, 0);
549 if (eb) {
550 /*
551 * we found an extent buffer and it contains our page
552 * horray!
553 */
554 if (eb->start <= target &&
555 eb->start + eb->len > target)
556 return eb;
557
558 /* we found an extent buffer that wasn't for us */
559 free_extent_buffer(eb);
560 return NULL;
561 }
562 if (start == 0)
563 break;
564 start -= PAGE_CACHE_SIZE;
565 }
566 return NULL;
567}
568
569static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
570 struct extent_state *state, int mirror)
571{
572 struct extent_io_tree *tree;
573 u64 found_start;
574 int found_level;
575 struct extent_buffer *eb;
576 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
577 int ret = 0;
578 int reads_done;
579
580 if (!page->private)
581 goto out;
582
583 tree = &BTRFS_I(page->mapping->host)->io_tree;
584 eb = (struct extent_buffer *)page->private;
585
586 /* the pending IO might have been the only thing that kept this buffer
587 * in memory. Make sure we have a ref for all this other checks
588 */
589 extent_buffer_get(eb);
590
591 reads_done = atomic_dec_and_test(&eb->io_pages);
592 if (!reads_done)
593 goto err;
594
595 eb->read_mirror = mirror;
596 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
597 ret = -EIO;
598 goto err;
599 }
600
601 found_start = btrfs_header_bytenr(eb);
602 if (found_start != eb->start) {
603 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
604 "%llu %llu\n",
605 (unsigned long long)found_start,
606 (unsigned long long)eb->start);
607 ret = -EIO;
608 goto err;
609 }
610 if (check_tree_block_fsid(root, eb)) {
611 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
612 (unsigned long long)eb->start);
613 ret = -EIO;
614 goto err;
615 }
616 found_level = btrfs_header_level(eb);
617
618 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
619 eb, found_level);
620
621 ret = csum_tree_block(root, eb, 1);
622 if (ret) {
623 ret = -EIO;
624 goto err;
625 }
626
627 /*
628 * If this is a leaf block and it is corrupt, set the corrupt bit so
629 * that we don't try and read the other copies of this block, just
630 * return -EIO.
631 */
632 if (found_level == 0 && check_leaf(root, eb)) {
633 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
634 ret = -EIO;
635 }
636
637 if (!ret)
638 set_extent_buffer_uptodate(eb);
639err:
640 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
641 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
642 btree_readahead_hook(root, eb, eb->start, ret);
643 }
644
645 if (ret)
646 clear_extent_buffer_uptodate(eb);
647 free_extent_buffer(eb);
648out:
649 return ret;
650}
651
652static int btree_io_failed_hook(struct page *page, int failed_mirror)
653{
654 struct extent_buffer *eb;
655 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
656
657 eb = (struct extent_buffer *)page->private;
658 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
659 eb->read_mirror = failed_mirror;
660 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
661 btree_readahead_hook(root, eb, eb->start, -EIO);
662 return -EIO; /* we fixed nothing */
663}
664
665static void end_workqueue_bio(struct bio *bio, int err)
666{
667 struct end_io_wq *end_io_wq = bio->bi_private;
668 struct btrfs_fs_info *fs_info;
669
670 fs_info = end_io_wq->info;
671 end_io_wq->error = err;
672 end_io_wq->work.func = end_workqueue_fn;
673 end_io_wq->work.flags = 0;
674
675 if (bio->bi_rw & REQ_WRITE) {
676 if (end_io_wq->metadata == 1)
677 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
678 &end_io_wq->work);
679 else if (end_io_wq->metadata == 2)
680 btrfs_queue_worker(&fs_info->endio_freespace_worker,
681 &end_io_wq->work);
682 else
683 btrfs_queue_worker(&fs_info->endio_write_workers,
684 &end_io_wq->work);
685 } else {
686 if (end_io_wq->metadata)
687 btrfs_queue_worker(&fs_info->endio_meta_workers,
688 &end_io_wq->work);
689 else
690 btrfs_queue_worker(&fs_info->endio_workers,
691 &end_io_wq->work);
692 }
693}
694
695/*
696 * For the metadata arg you want
697 *
698 * 0 - if data
699 * 1 - if normal metadta
700 * 2 - if writing to the free space cache area
701 */
702int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
703 int metadata)
704{
705 struct end_io_wq *end_io_wq;
706 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
707 if (!end_io_wq)
708 return -ENOMEM;
709
710 end_io_wq->private = bio->bi_private;
711 end_io_wq->end_io = bio->bi_end_io;
712 end_io_wq->info = info;
713 end_io_wq->error = 0;
714 end_io_wq->bio = bio;
715 end_io_wq->metadata = metadata;
716
717 bio->bi_private = end_io_wq;
718 bio->bi_end_io = end_workqueue_bio;
719 return 0;
720}
721
722unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
723{
724 unsigned long limit = min_t(unsigned long,
725 info->workers.max_workers,
726 info->fs_devices->open_devices);
727 return 256 * limit;
728}
729
730static void run_one_async_start(struct btrfs_work *work)
731{
732 struct async_submit_bio *async;
733 int ret;
734
735 async = container_of(work, struct async_submit_bio, work);
736 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
737 async->mirror_num, async->bio_flags,
738 async->bio_offset);
739 if (ret)
740 async->error = ret;
741}
742
743static void run_one_async_done(struct btrfs_work *work)
744{
745 struct btrfs_fs_info *fs_info;
746 struct async_submit_bio *async;
747 int limit;
748
749 async = container_of(work, struct async_submit_bio, work);
750 fs_info = BTRFS_I(async->inode)->root->fs_info;
751
752 limit = btrfs_async_submit_limit(fs_info);
753 limit = limit * 2 / 3;
754
755 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
756 waitqueue_active(&fs_info->async_submit_wait))
757 wake_up(&fs_info->async_submit_wait);
758
759 /* If an error occured we just want to clean up the bio and move on */
760 if (async->error) {
761 bio_endio(async->bio, async->error);
762 return;
763 }
764
765 async->submit_bio_done(async->inode, async->rw, async->bio,
766 async->mirror_num, async->bio_flags,
767 async->bio_offset);
768}
769
770static void run_one_async_free(struct btrfs_work *work)
771{
772 struct async_submit_bio *async;
773
774 async = container_of(work, struct async_submit_bio, work);
775 kfree(async);
776}
777
778int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
779 int rw, struct bio *bio, int mirror_num,
780 unsigned long bio_flags,
781 u64 bio_offset,
782 extent_submit_bio_hook_t *submit_bio_start,
783 extent_submit_bio_hook_t *submit_bio_done)
784{
785 struct async_submit_bio *async;
786
787 async = kmalloc(sizeof(*async), GFP_NOFS);
788 if (!async)
789 return -ENOMEM;
790
791 async->inode = inode;
792 async->rw = rw;
793 async->bio = bio;
794 async->mirror_num = mirror_num;
795 async->submit_bio_start = submit_bio_start;
796 async->submit_bio_done = submit_bio_done;
797
798 async->work.func = run_one_async_start;
799 async->work.ordered_func = run_one_async_done;
800 async->work.ordered_free = run_one_async_free;
801
802 async->work.flags = 0;
803 async->bio_flags = bio_flags;
804 async->bio_offset = bio_offset;
805
806 async->error = 0;
807
808 atomic_inc(&fs_info->nr_async_submits);
809
810 if (rw & REQ_SYNC)
811 btrfs_set_work_high_prio(&async->work);
812
813 btrfs_queue_worker(&fs_info->workers, &async->work);
814
815 while (atomic_read(&fs_info->async_submit_draining) &&
816 atomic_read(&fs_info->nr_async_submits)) {
817 wait_event(fs_info->async_submit_wait,
818 (atomic_read(&fs_info->nr_async_submits) == 0));
819 }
820
821 return 0;
822}
823
824static int btree_csum_one_bio(struct bio *bio)
825{
826 struct bio_vec *bvec = bio->bi_io_vec;
827 int bio_index = 0;
828 struct btrfs_root *root;
829 int ret = 0;
830
831 WARN_ON(bio->bi_vcnt <= 0);
832 while (bio_index < bio->bi_vcnt) {
833 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
834 ret = csum_dirty_buffer(root, bvec->bv_page);
835 if (ret)
836 break;
837 bio_index++;
838 bvec++;
839 }
840 return ret;
841}
842
843static int __btree_submit_bio_start(struct inode *inode, int rw,
844 struct bio *bio, int mirror_num,
845 unsigned long bio_flags,
846 u64 bio_offset)
847{
848 /*
849 * when we're called for a write, we're already in the async
850 * submission context. Just jump into btrfs_map_bio
851 */
852 return btree_csum_one_bio(bio);
853}
854
855static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
856 int mirror_num, unsigned long bio_flags,
857 u64 bio_offset)
858{
859 /*
860 * when we're called for a write, we're already in the async
861 * submission context. Just jump into btrfs_map_bio
862 */
863 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
864}
865
866static int check_async_write(struct inode *inode, unsigned long bio_flags)
867{
868 if (bio_flags & EXTENT_BIO_TREE_LOG)
869 return 0;
870#ifdef CONFIG_X86
871 if (cpu_has_xmm4_2)
872 return 0;
873#endif
874 return 1;
875}
876
877static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
878 int mirror_num, unsigned long bio_flags,
879 u64 bio_offset)
880{
881 int async = check_async_write(inode, bio_flags);
882 int ret;
883
884 if (!(rw & REQ_WRITE)) {
885
886 /*
887 * called for a read, do the setup so that checksum validation
888 * can happen in the async kernel threads
889 */
890 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
891 bio, 1);
892 if (ret)
893 return ret;
894 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
895 mirror_num, 0);
896 } else if (!async) {
897 ret = btree_csum_one_bio(bio);
898 if (ret)
899 return ret;
900 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
901 mirror_num, 0);
902 }
903
904 /*
905 * kthread helpers are used to submit writes so that checksumming
906 * can happen in parallel across all CPUs
907 */
908 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
909 inode, rw, bio, mirror_num, 0,
910 bio_offset,
911 __btree_submit_bio_start,
912 __btree_submit_bio_done);
913}
914
915#ifdef CONFIG_MIGRATION
916static int btree_migratepage(struct address_space *mapping,
917 struct page *newpage, struct page *page,
918 enum migrate_mode mode)
919{
920 /*
921 * we can't safely write a btree page from here,
922 * we haven't done the locking hook
923 */
924 if (PageDirty(page))
925 return -EAGAIN;
926 /*
927 * Buffers may be managed in a filesystem specific way.
928 * We must have no buffers or drop them.
929 */
930 if (page_has_private(page) &&
931 !try_to_release_page(page, GFP_KERNEL))
932 return -EAGAIN;
933 return migrate_page(mapping, newpage, page, mode);
934}
935#endif
936
937
938static int btree_writepages(struct address_space *mapping,
939 struct writeback_control *wbc)
940{
941 struct extent_io_tree *tree;
942 tree = &BTRFS_I(mapping->host)->io_tree;
943 if (wbc->sync_mode == WB_SYNC_NONE) {
944 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
945 u64 num_dirty;
946 unsigned long thresh = 32 * 1024 * 1024;
947
948 if (wbc->for_kupdate)
949 return 0;
950
951 /* this is a bit racy, but that's ok */
952 num_dirty = root->fs_info->dirty_metadata_bytes;
953 if (num_dirty < thresh)
954 return 0;
955 }
956 return btree_write_cache_pages(mapping, wbc);
957}
958
959static int btree_readpage(struct file *file, struct page *page)
960{
961 struct extent_io_tree *tree;
962 tree = &BTRFS_I(page->mapping->host)->io_tree;
963 return extent_read_full_page(tree, page, btree_get_extent, 0);
964}
965
966static int btree_releasepage(struct page *page, gfp_t gfp_flags)
967{
968 if (PageWriteback(page) || PageDirty(page))
969 return 0;
970 /*
971 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
972 * slab allocation from alloc_extent_state down the callchain where
973 * it'd hit a BUG_ON as those flags are not allowed.
974 */
975 gfp_flags &= ~GFP_SLAB_BUG_MASK;
976
977 return try_release_extent_buffer(page, gfp_flags);
978}
979
980static void btree_invalidatepage(struct page *page, unsigned long offset)
981{
982 struct extent_io_tree *tree;
983 tree = &BTRFS_I(page->mapping->host)->io_tree;
984 extent_invalidatepage(tree, page, offset);
985 btree_releasepage(page, GFP_NOFS);
986 if (PagePrivate(page)) {
987 printk(KERN_WARNING "btrfs warning page private not zero "
988 "on page %llu\n", (unsigned long long)page_offset(page));
989 ClearPagePrivate(page);
990 set_page_private(page, 0);
991 page_cache_release(page);
992 }
993}
994
995static int btree_set_page_dirty(struct page *page)
996{
997 struct extent_buffer *eb;
998
999 BUG_ON(!PagePrivate(page));
1000 eb = (struct extent_buffer *)page->private;
1001 BUG_ON(!eb);
1002 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1003 BUG_ON(!atomic_read(&eb->refs));
1004 btrfs_assert_tree_locked(eb);
1005 return __set_page_dirty_nobuffers(page);
1006}
1007
1008static const struct address_space_operations btree_aops = {
1009 .readpage = btree_readpage,
1010 .writepages = btree_writepages,
1011 .releasepage = btree_releasepage,
1012 .invalidatepage = btree_invalidatepage,
1013#ifdef CONFIG_MIGRATION
1014 .migratepage = btree_migratepage,
1015#endif
1016 .set_page_dirty = btree_set_page_dirty,
1017};
1018
1019int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1020 u64 parent_transid)
1021{
1022 struct extent_buffer *buf = NULL;
1023 struct inode *btree_inode = root->fs_info->btree_inode;
1024 int ret = 0;
1025
1026 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1027 if (!buf)
1028 return 0;
1029 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1030 buf, 0, WAIT_NONE, btree_get_extent, 0);
1031 free_extent_buffer(buf);
1032 return ret;
1033}
1034
1035int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1036 int mirror_num, struct extent_buffer **eb)
1037{
1038 struct extent_buffer *buf = NULL;
1039 struct inode *btree_inode = root->fs_info->btree_inode;
1040 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1041 int ret;
1042
1043 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1044 if (!buf)
1045 return 0;
1046
1047 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1048
1049 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1050 btree_get_extent, mirror_num);
1051 if (ret) {
1052 free_extent_buffer(buf);
1053 return ret;
1054 }
1055
1056 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1057 free_extent_buffer(buf);
1058 return -EIO;
1059 } else if (extent_buffer_uptodate(buf)) {
1060 *eb = buf;
1061 } else {
1062 free_extent_buffer(buf);
1063 }
1064 return 0;
1065}
1066
1067struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1068 u64 bytenr, u32 blocksize)
1069{
1070 struct inode *btree_inode = root->fs_info->btree_inode;
1071 struct extent_buffer *eb;
1072 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1073 bytenr, blocksize);
1074 return eb;
1075}
1076
1077struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1078 u64 bytenr, u32 blocksize)
1079{
1080 struct inode *btree_inode = root->fs_info->btree_inode;
1081 struct extent_buffer *eb;
1082
1083 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1084 bytenr, blocksize);
1085 return eb;
1086}
1087
1088
1089int btrfs_write_tree_block(struct extent_buffer *buf)
1090{
1091 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1092 buf->start + buf->len - 1);
1093}
1094
1095int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1096{
1097 return filemap_fdatawait_range(buf->pages[0]->mapping,
1098 buf->start, buf->start + buf->len - 1);
1099}
1100
1101struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1102 u32 blocksize, u64 parent_transid)
1103{
1104 struct extent_buffer *buf = NULL;
1105 int ret;
1106
1107 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1108 if (!buf)
1109 return NULL;
1110
1111 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1112 return buf;
1113
1114}
1115
1116void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1117 struct extent_buffer *buf)
1118{
1119 if (btrfs_header_generation(buf) ==
1120 root->fs_info->running_transaction->transid) {
1121 btrfs_assert_tree_locked(buf);
1122
1123 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1124 spin_lock(&root->fs_info->delalloc_lock);
1125 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1126 root->fs_info->dirty_metadata_bytes -= buf->len;
1127 else {
1128 spin_unlock(&root->fs_info->delalloc_lock);
1129 btrfs_panic(root->fs_info, -EOVERFLOW,
1130 "Can't clear %lu bytes from "
1131 " dirty_mdatadata_bytes (%llu)",
1132 buf->len,
1133 root->fs_info->dirty_metadata_bytes);
1134 }
1135 spin_unlock(&root->fs_info->delalloc_lock);
1136 }
1137
1138 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1139 btrfs_set_lock_blocking(buf);
1140 clear_extent_buffer_dirty(buf);
1141 }
1142}
1143
1144static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1145 u32 stripesize, struct btrfs_root *root,
1146 struct btrfs_fs_info *fs_info,
1147 u64 objectid)
1148{
1149 root->node = NULL;
1150 root->commit_root = NULL;
1151 root->sectorsize = sectorsize;
1152 root->nodesize = nodesize;
1153 root->leafsize = leafsize;
1154 root->stripesize = stripesize;
1155 root->ref_cows = 0;
1156 root->track_dirty = 0;
1157 root->in_radix = 0;
1158 root->orphan_item_inserted = 0;
1159 root->orphan_cleanup_state = 0;
1160
1161 root->objectid = objectid;
1162 root->last_trans = 0;
1163 root->highest_objectid = 0;
1164 root->name = NULL;
1165 root->inode_tree = RB_ROOT;
1166 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1167 root->block_rsv = NULL;
1168 root->orphan_block_rsv = NULL;
1169
1170 INIT_LIST_HEAD(&root->dirty_list);
1171 INIT_LIST_HEAD(&root->root_list);
1172 spin_lock_init(&root->orphan_lock);
1173 spin_lock_init(&root->inode_lock);
1174 spin_lock_init(&root->accounting_lock);
1175 mutex_init(&root->objectid_mutex);
1176 mutex_init(&root->log_mutex);
1177 init_waitqueue_head(&root->log_writer_wait);
1178 init_waitqueue_head(&root->log_commit_wait[0]);
1179 init_waitqueue_head(&root->log_commit_wait[1]);
1180 atomic_set(&root->log_commit[0], 0);
1181 atomic_set(&root->log_commit[1], 0);
1182 atomic_set(&root->log_writers, 0);
1183 atomic_set(&root->log_batch, 0);
1184 atomic_set(&root->orphan_inodes, 0);
1185 root->log_transid = 0;
1186 root->last_log_commit = 0;
1187 extent_io_tree_init(&root->dirty_log_pages,
1188 fs_info->btree_inode->i_mapping);
1189
1190 memset(&root->root_key, 0, sizeof(root->root_key));
1191 memset(&root->root_item, 0, sizeof(root->root_item));
1192 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1193 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1194 root->defrag_trans_start = fs_info->generation;
1195 init_completion(&root->kobj_unregister);
1196 root->defrag_running = 0;
1197 root->root_key.objectid = objectid;
1198 root->anon_dev = 0;
1199
1200 spin_lock_init(&root->root_times_lock);
1201}
1202
1203static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1204 struct btrfs_fs_info *fs_info,
1205 u64 objectid,
1206 struct btrfs_root *root)
1207{
1208 int ret;
1209 u32 blocksize;
1210 u64 generation;
1211
1212 __setup_root(tree_root->nodesize, tree_root->leafsize,
1213 tree_root->sectorsize, tree_root->stripesize,
1214 root, fs_info, objectid);
1215 ret = btrfs_find_last_root(tree_root, objectid,
1216 &root->root_item, &root->root_key);
1217 if (ret > 0)
1218 return -ENOENT;
1219 else if (ret < 0)
1220 return ret;
1221
1222 generation = btrfs_root_generation(&root->root_item);
1223 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1224 root->commit_root = NULL;
1225 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1226 blocksize, generation);
1227 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1228 free_extent_buffer(root->node);
1229 root->node = NULL;
1230 return -EIO;
1231 }
1232 root->commit_root = btrfs_root_node(root);
1233 return 0;
1234}
1235
1236static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1237{
1238 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1239 if (root)
1240 root->fs_info = fs_info;
1241 return root;
1242}
1243
1244struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1245 struct btrfs_fs_info *fs_info,
1246 u64 objectid)
1247{
1248 struct extent_buffer *leaf;
1249 struct btrfs_root *tree_root = fs_info->tree_root;
1250 struct btrfs_root *root;
1251 struct btrfs_key key;
1252 int ret = 0;
1253 u64 bytenr;
1254
1255 root = btrfs_alloc_root(fs_info);
1256 if (!root)
1257 return ERR_PTR(-ENOMEM);
1258
1259 __setup_root(tree_root->nodesize, tree_root->leafsize,
1260 tree_root->sectorsize, tree_root->stripesize,
1261 root, fs_info, objectid);
1262 root->root_key.objectid = objectid;
1263 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1264 root->root_key.offset = 0;
1265
1266 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1267 0, objectid, NULL, 0, 0, 0);
1268 if (IS_ERR(leaf)) {
1269 ret = PTR_ERR(leaf);
1270 goto fail;
1271 }
1272
1273 bytenr = leaf->start;
1274 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1275 btrfs_set_header_bytenr(leaf, leaf->start);
1276 btrfs_set_header_generation(leaf, trans->transid);
1277 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1278 btrfs_set_header_owner(leaf, objectid);
1279 root->node = leaf;
1280
1281 write_extent_buffer(leaf, fs_info->fsid,
1282 (unsigned long)btrfs_header_fsid(leaf),
1283 BTRFS_FSID_SIZE);
1284 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1285 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1286 BTRFS_UUID_SIZE);
1287 btrfs_mark_buffer_dirty(leaf);
1288
1289 root->commit_root = btrfs_root_node(root);
1290 root->track_dirty = 1;
1291
1292
1293 root->root_item.flags = 0;
1294 root->root_item.byte_limit = 0;
1295 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1296 btrfs_set_root_generation(&root->root_item, trans->transid);
1297 btrfs_set_root_level(&root->root_item, 0);
1298 btrfs_set_root_refs(&root->root_item, 1);
1299 btrfs_set_root_used(&root->root_item, leaf->len);
1300 btrfs_set_root_last_snapshot(&root->root_item, 0);
1301 btrfs_set_root_dirid(&root->root_item, 0);
1302 root->root_item.drop_level = 0;
1303
1304 key.objectid = objectid;
1305 key.type = BTRFS_ROOT_ITEM_KEY;
1306 key.offset = 0;
1307 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1308 if (ret)
1309 goto fail;
1310
1311 btrfs_tree_unlock(leaf);
1312
1313fail:
1314 if (ret)
1315 return ERR_PTR(ret);
1316
1317 return root;
1318}
1319
1320static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1321 struct btrfs_fs_info *fs_info)
1322{
1323 struct btrfs_root *root;
1324 struct btrfs_root *tree_root = fs_info->tree_root;
1325 struct extent_buffer *leaf;
1326
1327 root = btrfs_alloc_root(fs_info);
1328 if (!root)
1329 return ERR_PTR(-ENOMEM);
1330
1331 __setup_root(tree_root->nodesize, tree_root->leafsize,
1332 tree_root->sectorsize, tree_root->stripesize,
1333 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1334
1335 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1336 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1337 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1338 /*
1339 * log trees do not get reference counted because they go away
1340 * before a real commit is actually done. They do store pointers
1341 * to file data extents, and those reference counts still get
1342 * updated (along with back refs to the log tree).
1343 */
1344 root->ref_cows = 0;
1345
1346 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1347 BTRFS_TREE_LOG_OBJECTID, NULL,
1348 0, 0, 0);
1349 if (IS_ERR(leaf)) {
1350 kfree(root);
1351 return ERR_CAST(leaf);
1352 }
1353
1354 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1355 btrfs_set_header_bytenr(leaf, leaf->start);
1356 btrfs_set_header_generation(leaf, trans->transid);
1357 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1358 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1359 root->node = leaf;
1360
1361 write_extent_buffer(root->node, root->fs_info->fsid,
1362 (unsigned long)btrfs_header_fsid(root->node),
1363 BTRFS_FSID_SIZE);
1364 btrfs_mark_buffer_dirty(root->node);
1365 btrfs_tree_unlock(root->node);
1366 return root;
1367}
1368
1369int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1370 struct btrfs_fs_info *fs_info)
1371{
1372 struct btrfs_root *log_root;
1373
1374 log_root = alloc_log_tree(trans, fs_info);
1375 if (IS_ERR(log_root))
1376 return PTR_ERR(log_root);
1377 WARN_ON(fs_info->log_root_tree);
1378 fs_info->log_root_tree = log_root;
1379 return 0;
1380}
1381
1382int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1383 struct btrfs_root *root)
1384{
1385 struct btrfs_root *log_root;
1386 struct btrfs_inode_item *inode_item;
1387
1388 log_root = alloc_log_tree(trans, root->fs_info);
1389 if (IS_ERR(log_root))
1390 return PTR_ERR(log_root);
1391
1392 log_root->last_trans = trans->transid;
1393 log_root->root_key.offset = root->root_key.objectid;
1394
1395 inode_item = &log_root->root_item.inode;
1396 inode_item->generation = cpu_to_le64(1);
1397 inode_item->size = cpu_to_le64(3);
1398 inode_item->nlink = cpu_to_le32(1);
1399 inode_item->nbytes = cpu_to_le64(root->leafsize);
1400 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1401
1402 btrfs_set_root_node(&log_root->root_item, log_root->node);
1403
1404 WARN_ON(root->log_root);
1405 root->log_root = log_root;
1406 root->log_transid = 0;
1407 root->last_log_commit = 0;
1408 return 0;
1409}
1410
1411struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1412 struct btrfs_key *location)
1413{
1414 struct btrfs_root *root;
1415 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1416 struct btrfs_path *path;
1417 struct extent_buffer *l;
1418 u64 generation;
1419 u32 blocksize;
1420 int ret = 0;
1421 int slot;
1422
1423 root = btrfs_alloc_root(fs_info);
1424 if (!root)
1425 return ERR_PTR(-ENOMEM);
1426 if (location->offset == (u64)-1) {
1427 ret = find_and_setup_root(tree_root, fs_info,
1428 location->objectid, root);
1429 if (ret) {
1430 kfree(root);
1431 return ERR_PTR(ret);
1432 }
1433 goto out;
1434 }
1435
1436 __setup_root(tree_root->nodesize, tree_root->leafsize,
1437 tree_root->sectorsize, tree_root->stripesize,
1438 root, fs_info, location->objectid);
1439
1440 path = btrfs_alloc_path();
1441 if (!path) {
1442 kfree(root);
1443 return ERR_PTR(-ENOMEM);
1444 }
1445 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1446 if (ret == 0) {
1447 l = path->nodes[0];
1448 slot = path->slots[0];
1449 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1450 memcpy(&root->root_key, location, sizeof(*location));
1451 }
1452 btrfs_free_path(path);
1453 if (ret) {
1454 kfree(root);
1455 if (ret > 0)
1456 ret = -ENOENT;
1457 return ERR_PTR(ret);
1458 }
1459
1460 generation = btrfs_root_generation(&root->root_item);
1461 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1462 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1463 blocksize, generation);
1464 root->commit_root = btrfs_root_node(root);
1465 BUG_ON(!root->node); /* -ENOMEM */
1466out:
1467 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1468 root->ref_cows = 1;
1469 btrfs_check_and_init_root_item(&root->root_item);
1470 }
1471
1472 return root;
1473}
1474
1475struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1476 struct btrfs_key *location)
1477{
1478 struct btrfs_root *root;
1479 int ret;
1480
1481 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1482 return fs_info->tree_root;
1483 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1484 return fs_info->extent_root;
1485 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1486 return fs_info->chunk_root;
1487 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1488 return fs_info->dev_root;
1489 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1490 return fs_info->csum_root;
1491 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1492 return fs_info->quota_root ? fs_info->quota_root :
1493 ERR_PTR(-ENOENT);
1494again:
1495 spin_lock(&fs_info->fs_roots_radix_lock);
1496 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1497 (unsigned long)location->objectid);
1498 spin_unlock(&fs_info->fs_roots_radix_lock);
1499 if (root)
1500 return root;
1501
1502 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1503 if (IS_ERR(root))
1504 return root;
1505
1506 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1507 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1508 GFP_NOFS);
1509 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1510 ret = -ENOMEM;
1511 goto fail;
1512 }
1513
1514 btrfs_init_free_ino_ctl(root);
1515 mutex_init(&root->fs_commit_mutex);
1516 spin_lock_init(&root->cache_lock);
1517 init_waitqueue_head(&root->cache_wait);
1518
1519 ret = get_anon_bdev(&root->anon_dev);
1520 if (ret)
1521 goto fail;
1522
1523 if (btrfs_root_refs(&root->root_item) == 0) {
1524 ret = -ENOENT;
1525 goto fail;
1526 }
1527
1528 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1529 if (ret < 0)
1530 goto fail;
1531 if (ret == 0)
1532 root->orphan_item_inserted = 1;
1533
1534 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1535 if (ret)
1536 goto fail;
1537
1538 spin_lock(&fs_info->fs_roots_radix_lock);
1539 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1540 (unsigned long)root->root_key.objectid,
1541 root);
1542 if (ret == 0)
1543 root->in_radix = 1;
1544
1545 spin_unlock(&fs_info->fs_roots_radix_lock);
1546 radix_tree_preload_end();
1547 if (ret) {
1548 if (ret == -EEXIST) {
1549 free_fs_root(root);
1550 goto again;
1551 }
1552 goto fail;
1553 }
1554
1555 ret = btrfs_find_dead_roots(fs_info->tree_root,
1556 root->root_key.objectid);
1557 WARN_ON(ret);
1558 return root;
1559fail:
1560 free_fs_root(root);
1561 return ERR_PTR(ret);
1562}
1563
1564static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1565{
1566 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1567 int ret = 0;
1568 struct btrfs_device *device;
1569 struct backing_dev_info *bdi;
1570
1571 rcu_read_lock();
1572 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1573 if (!device->bdev)
1574 continue;
1575 bdi = blk_get_backing_dev_info(device->bdev);
1576 if (bdi && bdi_congested(bdi, bdi_bits)) {
1577 ret = 1;
1578 break;
1579 }
1580 }
1581 rcu_read_unlock();
1582 return ret;
1583}
1584
1585/*
1586 * If this fails, caller must call bdi_destroy() to get rid of the
1587 * bdi again.
1588 */
1589static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1590{
1591 int err;
1592
1593 bdi->capabilities = BDI_CAP_MAP_COPY;
1594 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1595 if (err)
1596 return err;
1597
1598 bdi->ra_pages = default_backing_dev_info.ra_pages;
1599 bdi->congested_fn = btrfs_congested_fn;
1600 bdi->congested_data = info;
1601 return 0;
1602}
1603
1604/*
1605 * called by the kthread helper functions to finally call the bio end_io
1606 * functions. This is where read checksum verification actually happens
1607 */
1608static void end_workqueue_fn(struct btrfs_work *work)
1609{
1610 struct bio *bio;
1611 struct end_io_wq *end_io_wq;
1612 struct btrfs_fs_info *fs_info;
1613 int error;
1614
1615 end_io_wq = container_of(work, struct end_io_wq, work);
1616 bio = end_io_wq->bio;
1617 fs_info = end_io_wq->info;
1618
1619 error = end_io_wq->error;
1620 bio->bi_private = end_io_wq->private;
1621 bio->bi_end_io = end_io_wq->end_io;
1622 kfree(end_io_wq);
1623 bio_endio(bio, error);
1624}
1625
1626static int cleaner_kthread(void *arg)
1627{
1628 struct btrfs_root *root = arg;
1629
1630 do {
1631 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1632 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1633 btrfs_run_delayed_iputs(root);
1634 btrfs_clean_old_snapshots(root);
1635 mutex_unlock(&root->fs_info->cleaner_mutex);
1636 btrfs_run_defrag_inodes(root->fs_info);
1637 }
1638
1639 if (!try_to_freeze()) {
1640 set_current_state(TASK_INTERRUPTIBLE);
1641 if (!kthread_should_stop())
1642 schedule();
1643 __set_current_state(TASK_RUNNING);
1644 }
1645 } while (!kthread_should_stop());
1646 return 0;
1647}
1648
1649static int transaction_kthread(void *arg)
1650{
1651 struct btrfs_root *root = arg;
1652 struct btrfs_trans_handle *trans;
1653 struct btrfs_transaction *cur;
1654 u64 transid;
1655 unsigned long now;
1656 unsigned long delay;
1657 bool cannot_commit;
1658
1659 do {
1660 cannot_commit = false;
1661 delay = HZ * 30;
1662 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1663
1664 spin_lock(&root->fs_info->trans_lock);
1665 cur = root->fs_info->running_transaction;
1666 if (!cur) {
1667 spin_unlock(&root->fs_info->trans_lock);
1668 goto sleep;
1669 }
1670
1671 now = get_seconds();
1672 if (!cur->blocked &&
1673 (now < cur->start_time || now - cur->start_time < 30)) {
1674 spin_unlock(&root->fs_info->trans_lock);
1675 delay = HZ * 5;
1676 goto sleep;
1677 }
1678 transid = cur->transid;
1679 spin_unlock(&root->fs_info->trans_lock);
1680
1681 /* If the file system is aborted, this will always fail. */
1682 trans = btrfs_attach_transaction(root);
1683 if (IS_ERR(trans)) {
1684 if (PTR_ERR(trans) != -ENOENT)
1685 cannot_commit = true;
1686 goto sleep;
1687 }
1688 if (transid == trans->transid) {
1689 btrfs_commit_transaction(trans, root);
1690 } else {
1691 btrfs_end_transaction(trans, root);
1692 }
1693sleep:
1694 wake_up_process(root->fs_info->cleaner_kthread);
1695 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1696
1697 if (!try_to_freeze()) {
1698 set_current_state(TASK_INTERRUPTIBLE);
1699 if (!kthread_should_stop() &&
1700 (!btrfs_transaction_blocked(root->fs_info) ||
1701 cannot_commit))
1702 schedule_timeout(delay);
1703 __set_current_state(TASK_RUNNING);
1704 }
1705 } while (!kthread_should_stop());
1706 return 0;
1707}
1708
1709/*
1710 * this will find the highest generation in the array of
1711 * root backups. The index of the highest array is returned,
1712 * or -1 if we can't find anything.
1713 *
1714 * We check to make sure the array is valid by comparing the
1715 * generation of the latest root in the array with the generation
1716 * in the super block. If they don't match we pitch it.
1717 */
1718static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1719{
1720 u64 cur;
1721 int newest_index = -1;
1722 struct btrfs_root_backup *root_backup;
1723 int i;
1724
1725 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1726 root_backup = info->super_copy->super_roots + i;
1727 cur = btrfs_backup_tree_root_gen(root_backup);
1728 if (cur == newest_gen)
1729 newest_index = i;
1730 }
1731
1732 /* check to see if we actually wrapped around */
1733 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1734 root_backup = info->super_copy->super_roots;
1735 cur = btrfs_backup_tree_root_gen(root_backup);
1736 if (cur == newest_gen)
1737 newest_index = 0;
1738 }
1739 return newest_index;
1740}
1741
1742
1743/*
1744 * find the oldest backup so we know where to store new entries
1745 * in the backup array. This will set the backup_root_index
1746 * field in the fs_info struct
1747 */
1748static void find_oldest_super_backup(struct btrfs_fs_info *info,
1749 u64 newest_gen)
1750{
1751 int newest_index = -1;
1752
1753 newest_index = find_newest_super_backup(info, newest_gen);
1754 /* if there was garbage in there, just move along */
1755 if (newest_index == -1) {
1756 info->backup_root_index = 0;
1757 } else {
1758 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1759 }
1760}
1761
1762/*
1763 * copy all the root pointers into the super backup array.
1764 * this will bump the backup pointer by one when it is
1765 * done
1766 */
1767static void backup_super_roots(struct btrfs_fs_info *info)
1768{
1769 int next_backup;
1770 struct btrfs_root_backup *root_backup;
1771 int last_backup;
1772
1773 next_backup = info->backup_root_index;
1774 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1775 BTRFS_NUM_BACKUP_ROOTS;
1776
1777 /*
1778 * just overwrite the last backup if we're at the same generation
1779 * this happens only at umount
1780 */
1781 root_backup = info->super_for_commit->super_roots + last_backup;
1782 if (btrfs_backup_tree_root_gen(root_backup) ==
1783 btrfs_header_generation(info->tree_root->node))
1784 next_backup = last_backup;
1785
1786 root_backup = info->super_for_commit->super_roots + next_backup;
1787
1788 /*
1789 * make sure all of our padding and empty slots get zero filled
1790 * regardless of which ones we use today
1791 */
1792 memset(root_backup, 0, sizeof(*root_backup));
1793
1794 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1795
1796 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1797 btrfs_set_backup_tree_root_gen(root_backup,
1798 btrfs_header_generation(info->tree_root->node));
1799
1800 btrfs_set_backup_tree_root_level(root_backup,
1801 btrfs_header_level(info->tree_root->node));
1802
1803 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1804 btrfs_set_backup_chunk_root_gen(root_backup,
1805 btrfs_header_generation(info->chunk_root->node));
1806 btrfs_set_backup_chunk_root_level(root_backup,
1807 btrfs_header_level(info->chunk_root->node));
1808
1809 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1810 btrfs_set_backup_extent_root_gen(root_backup,
1811 btrfs_header_generation(info->extent_root->node));
1812 btrfs_set_backup_extent_root_level(root_backup,
1813 btrfs_header_level(info->extent_root->node));
1814
1815 /*
1816 * we might commit during log recovery, which happens before we set
1817 * the fs_root. Make sure it is valid before we fill it in.
1818 */
1819 if (info->fs_root && info->fs_root->node) {
1820 btrfs_set_backup_fs_root(root_backup,
1821 info->fs_root->node->start);
1822 btrfs_set_backup_fs_root_gen(root_backup,
1823 btrfs_header_generation(info->fs_root->node));
1824 btrfs_set_backup_fs_root_level(root_backup,
1825 btrfs_header_level(info->fs_root->node));
1826 }
1827
1828 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1829 btrfs_set_backup_dev_root_gen(root_backup,
1830 btrfs_header_generation(info->dev_root->node));
1831 btrfs_set_backup_dev_root_level(root_backup,
1832 btrfs_header_level(info->dev_root->node));
1833
1834 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1835 btrfs_set_backup_csum_root_gen(root_backup,
1836 btrfs_header_generation(info->csum_root->node));
1837 btrfs_set_backup_csum_root_level(root_backup,
1838 btrfs_header_level(info->csum_root->node));
1839
1840 btrfs_set_backup_total_bytes(root_backup,
1841 btrfs_super_total_bytes(info->super_copy));
1842 btrfs_set_backup_bytes_used(root_backup,
1843 btrfs_super_bytes_used(info->super_copy));
1844 btrfs_set_backup_num_devices(root_backup,
1845 btrfs_super_num_devices(info->super_copy));
1846
1847 /*
1848 * if we don't copy this out to the super_copy, it won't get remembered
1849 * for the next commit
1850 */
1851 memcpy(&info->super_copy->super_roots,
1852 &info->super_for_commit->super_roots,
1853 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1854}
1855
1856/*
1857 * this copies info out of the root backup array and back into
1858 * the in-memory super block. It is meant to help iterate through
1859 * the array, so you send it the number of backups you've already
1860 * tried and the last backup index you used.
1861 *
1862 * this returns -1 when it has tried all the backups
1863 */
1864static noinline int next_root_backup(struct btrfs_fs_info *info,
1865 struct btrfs_super_block *super,
1866 int *num_backups_tried, int *backup_index)
1867{
1868 struct btrfs_root_backup *root_backup;
1869 int newest = *backup_index;
1870
1871 if (*num_backups_tried == 0) {
1872 u64 gen = btrfs_super_generation(super);
1873
1874 newest = find_newest_super_backup(info, gen);
1875 if (newest == -1)
1876 return -1;
1877
1878 *backup_index = newest;
1879 *num_backups_tried = 1;
1880 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1881 /* we've tried all the backups, all done */
1882 return -1;
1883 } else {
1884 /* jump to the next oldest backup */
1885 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1886 BTRFS_NUM_BACKUP_ROOTS;
1887 *backup_index = newest;
1888 *num_backups_tried += 1;
1889 }
1890 root_backup = super->super_roots + newest;
1891
1892 btrfs_set_super_generation(super,
1893 btrfs_backup_tree_root_gen(root_backup));
1894 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1895 btrfs_set_super_root_level(super,
1896 btrfs_backup_tree_root_level(root_backup));
1897 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1898
1899 /*
1900 * fixme: the total bytes and num_devices need to match or we should
1901 * need a fsck
1902 */
1903 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1904 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1905 return 0;
1906}
1907
1908/* helper to cleanup tree roots */
1909static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1910{
1911 free_extent_buffer(info->tree_root->node);
1912 free_extent_buffer(info->tree_root->commit_root);
1913 free_extent_buffer(info->dev_root->node);
1914 free_extent_buffer(info->dev_root->commit_root);
1915 free_extent_buffer(info->extent_root->node);
1916 free_extent_buffer(info->extent_root->commit_root);
1917 free_extent_buffer(info->csum_root->node);
1918 free_extent_buffer(info->csum_root->commit_root);
1919 if (info->quota_root) {
1920 free_extent_buffer(info->quota_root->node);
1921 free_extent_buffer(info->quota_root->commit_root);
1922 }
1923
1924 info->tree_root->node = NULL;
1925 info->tree_root->commit_root = NULL;
1926 info->dev_root->node = NULL;
1927 info->dev_root->commit_root = NULL;
1928 info->extent_root->node = NULL;
1929 info->extent_root->commit_root = NULL;
1930 info->csum_root->node = NULL;
1931 info->csum_root->commit_root = NULL;
1932 if (info->quota_root) {
1933 info->quota_root->node = NULL;
1934 info->quota_root->commit_root = NULL;
1935 }
1936
1937 if (chunk_root) {
1938 free_extent_buffer(info->chunk_root->node);
1939 free_extent_buffer(info->chunk_root->commit_root);
1940 info->chunk_root->node = NULL;
1941 info->chunk_root->commit_root = NULL;
1942 }
1943}
1944
1945
1946int open_ctree(struct super_block *sb,
1947 struct btrfs_fs_devices *fs_devices,
1948 char *options)
1949{
1950 u32 sectorsize;
1951 u32 nodesize;
1952 u32 leafsize;
1953 u32 blocksize;
1954 u32 stripesize;
1955 u64 generation;
1956 u64 features;
1957 struct btrfs_key location;
1958 struct buffer_head *bh;
1959 struct btrfs_super_block *disk_super;
1960 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1961 struct btrfs_root *tree_root;
1962 struct btrfs_root *extent_root;
1963 struct btrfs_root *csum_root;
1964 struct btrfs_root *chunk_root;
1965 struct btrfs_root *dev_root;
1966 struct btrfs_root *quota_root;
1967 struct btrfs_root *log_tree_root;
1968 int ret;
1969 int err = -EINVAL;
1970 int num_backups_tried = 0;
1971 int backup_index = 0;
1972
1973 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1974 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1975 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1976 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1977 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1978 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1979
1980 if (!tree_root || !extent_root || !csum_root ||
1981 !chunk_root || !dev_root || !quota_root) {
1982 err = -ENOMEM;
1983 goto fail;
1984 }
1985
1986 ret = init_srcu_struct(&fs_info->subvol_srcu);
1987 if (ret) {
1988 err = ret;
1989 goto fail;
1990 }
1991
1992 ret = setup_bdi(fs_info, &fs_info->bdi);
1993 if (ret) {
1994 err = ret;
1995 goto fail_srcu;
1996 }
1997
1998 fs_info->btree_inode = new_inode(sb);
1999 if (!fs_info->btree_inode) {
2000 err = -ENOMEM;
2001 goto fail_bdi;
2002 }
2003
2004 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2005
2006 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2007 INIT_LIST_HEAD(&fs_info->trans_list);
2008 INIT_LIST_HEAD(&fs_info->dead_roots);
2009 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2010 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2011 INIT_LIST_HEAD(&fs_info->ordered_operations);
2012 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2013 spin_lock_init(&fs_info->delalloc_lock);
2014 spin_lock_init(&fs_info->trans_lock);
2015 spin_lock_init(&fs_info->fs_roots_radix_lock);
2016 spin_lock_init(&fs_info->delayed_iput_lock);
2017 spin_lock_init(&fs_info->defrag_inodes_lock);
2018 spin_lock_init(&fs_info->free_chunk_lock);
2019 spin_lock_init(&fs_info->tree_mod_seq_lock);
2020 rwlock_init(&fs_info->tree_mod_log_lock);
2021 mutex_init(&fs_info->reloc_mutex);
2022
2023 init_completion(&fs_info->kobj_unregister);
2024 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2025 INIT_LIST_HEAD(&fs_info->space_info);
2026 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2027 btrfs_mapping_init(&fs_info->mapping_tree);
2028 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2029 BTRFS_BLOCK_RSV_GLOBAL);
2030 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2031 BTRFS_BLOCK_RSV_DELALLOC);
2032 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2033 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2034 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2035 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2036 BTRFS_BLOCK_RSV_DELOPS);
2037 atomic_set(&fs_info->nr_async_submits, 0);
2038 atomic_set(&fs_info->async_delalloc_pages, 0);
2039 atomic_set(&fs_info->async_submit_draining, 0);
2040 atomic_set(&fs_info->nr_async_bios, 0);
2041 atomic_set(&fs_info->defrag_running, 0);
2042 atomic_set(&fs_info->tree_mod_seq, 0);
2043 fs_info->sb = sb;
2044 fs_info->max_inline = 8192 * 1024;
2045 fs_info->metadata_ratio = 0;
2046 fs_info->defrag_inodes = RB_ROOT;
2047 fs_info->trans_no_join = 0;
2048 fs_info->free_chunk_space = 0;
2049 fs_info->tree_mod_log = RB_ROOT;
2050
2051 /* readahead state */
2052 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2053 spin_lock_init(&fs_info->reada_lock);
2054
2055 fs_info->thread_pool_size = min_t(unsigned long,
2056 num_online_cpus() + 2, 8);
2057
2058 INIT_LIST_HEAD(&fs_info->ordered_extents);
2059 spin_lock_init(&fs_info->ordered_extent_lock);
2060 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2061 GFP_NOFS);
2062 if (!fs_info->delayed_root) {
2063 err = -ENOMEM;
2064 goto fail_iput;
2065 }
2066 btrfs_init_delayed_root(fs_info->delayed_root);
2067
2068 mutex_init(&fs_info->scrub_lock);
2069 atomic_set(&fs_info->scrubs_running, 0);
2070 atomic_set(&fs_info->scrub_pause_req, 0);
2071 atomic_set(&fs_info->scrubs_paused, 0);
2072 atomic_set(&fs_info->scrub_cancel_req, 0);
2073 init_waitqueue_head(&fs_info->scrub_pause_wait);
2074 init_rwsem(&fs_info->scrub_super_lock);
2075 fs_info->scrub_workers_refcnt = 0;
2076#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2077 fs_info->check_integrity_print_mask = 0;
2078#endif
2079
2080 spin_lock_init(&fs_info->balance_lock);
2081 mutex_init(&fs_info->balance_mutex);
2082 atomic_set(&fs_info->balance_running, 0);
2083 atomic_set(&fs_info->balance_pause_req, 0);
2084 atomic_set(&fs_info->balance_cancel_req, 0);
2085 fs_info->balance_ctl = NULL;
2086 init_waitqueue_head(&fs_info->balance_wait_q);
2087
2088 sb->s_blocksize = 4096;
2089 sb->s_blocksize_bits = blksize_bits(4096);
2090 sb->s_bdi = &fs_info->bdi;
2091
2092 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2093 set_nlink(fs_info->btree_inode, 1);
2094 /*
2095 * we set the i_size on the btree inode to the max possible int.
2096 * the real end of the address space is determined by all of
2097 * the devices in the system
2098 */
2099 fs_info->btree_inode->i_size = OFFSET_MAX;
2100 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2101 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2102
2103 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2104 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2105 fs_info->btree_inode->i_mapping);
2106 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2107 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2108
2109 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2110
2111 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2112 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2113 sizeof(struct btrfs_key));
2114 set_bit(BTRFS_INODE_DUMMY,
2115 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2116 insert_inode_hash(fs_info->btree_inode);
2117
2118 spin_lock_init(&fs_info->block_group_cache_lock);
2119 fs_info->block_group_cache_tree = RB_ROOT;
2120
2121 extent_io_tree_init(&fs_info->freed_extents[0],
2122 fs_info->btree_inode->i_mapping);
2123 extent_io_tree_init(&fs_info->freed_extents[1],
2124 fs_info->btree_inode->i_mapping);
2125 fs_info->pinned_extents = &fs_info->freed_extents[0];
2126 fs_info->do_barriers = 1;
2127
2128
2129 mutex_init(&fs_info->ordered_operations_mutex);
2130 mutex_init(&fs_info->tree_log_mutex);
2131 mutex_init(&fs_info->chunk_mutex);
2132 mutex_init(&fs_info->transaction_kthread_mutex);
2133 mutex_init(&fs_info->cleaner_mutex);
2134 mutex_init(&fs_info->volume_mutex);
2135 init_rwsem(&fs_info->extent_commit_sem);
2136 init_rwsem(&fs_info->cleanup_work_sem);
2137 init_rwsem(&fs_info->subvol_sem);
2138
2139 spin_lock_init(&fs_info->qgroup_lock);
2140 fs_info->qgroup_tree = RB_ROOT;
2141 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2142 fs_info->qgroup_seq = 1;
2143 fs_info->quota_enabled = 0;
2144 fs_info->pending_quota_state = 0;
2145
2146 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2147 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2148
2149 init_waitqueue_head(&fs_info->transaction_throttle);
2150 init_waitqueue_head(&fs_info->transaction_wait);
2151 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2152 init_waitqueue_head(&fs_info->async_submit_wait);
2153
2154 __setup_root(4096, 4096, 4096, 4096, tree_root,
2155 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2156
2157 invalidate_bdev(fs_devices->latest_bdev);
2158 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2159 if (!bh) {
2160 err = -EINVAL;
2161 goto fail_alloc;
2162 }
2163
2164 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2165 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2166 sizeof(*fs_info->super_for_commit));
2167 brelse(bh);
2168
2169 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2170
2171 disk_super = fs_info->super_copy;
2172 if (!btrfs_super_root(disk_super))
2173 goto fail_alloc;
2174
2175 /* check FS state, whether FS is broken. */
2176 fs_info->fs_state |= btrfs_super_flags(disk_super);
2177
2178 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2179 if (ret) {
2180 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2181 err = ret;
2182 goto fail_alloc;
2183 }
2184
2185 /*
2186 * run through our array of backup supers and setup
2187 * our ring pointer to the oldest one
2188 */
2189 generation = btrfs_super_generation(disk_super);
2190 find_oldest_super_backup(fs_info, generation);
2191
2192 /*
2193 * In the long term, we'll store the compression type in the super
2194 * block, and it'll be used for per file compression control.
2195 */
2196 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2197
2198 ret = btrfs_parse_options(tree_root, options);
2199 if (ret) {
2200 err = ret;
2201 goto fail_alloc;
2202 }
2203
2204 features = btrfs_super_incompat_flags(disk_super) &
2205 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2206 if (features) {
2207 printk(KERN_ERR "BTRFS: couldn't mount because of "
2208 "unsupported optional features (%Lx).\n",
2209 (unsigned long long)features);
2210 err = -EINVAL;
2211 goto fail_alloc;
2212 }
2213
2214 if (btrfs_super_leafsize(disk_super) !=
2215 btrfs_super_nodesize(disk_super)) {
2216 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2217 "blocksizes don't match. node %d leaf %d\n",
2218 btrfs_super_nodesize(disk_super),
2219 btrfs_super_leafsize(disk_super));
2220 err = -EINVAL;
2221 goto fail_alloc;
2222 }
2223 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2224 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2225 "blocksize (%d) was too large\n",
2226 btrfs_super_leafsize(disk_super));
2227 err = -EINVAL;
2228 goto fail_alloc;
2229 }
2230
2231 features = btrfs_super_incompat_flags(disk_super);
2232 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2233 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2234 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2235
2236 /*
2237 * flag our filesystem as having big metadata blocks if
2238 * they are bigger than the page size
2239 */
2240 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2241 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2242 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2243 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2244 }
2245
2246 nodesize = btrfs_super_nodesize(disk_super);
2247 leafsize = btrfs_super_leafsize(disk_super);
2248 sectorsize = btrfs_super_sectorsize(disk_super);
2249 stripesize = btrfs_super_stripesize(disk_super);
2250
2251 /*
2252 * mixed block groups end up with duplicate but slightly offset
2253 * extent buffers for the same range. It leads to corruptions
2254 */
2255 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2256 (sectorsize != leafsize)) {
2257 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2258 "are not allowed for mixed block groups on %s\n",
2259 sb->s_id);
2260 goto fail_alloc;
2261 }
2262
2263 btrfs_set_super_incompat_flags(disk_super, features);
2264
2265 features = btrfs_super_compat_ro_flags(disk_super) &
2266 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2267 if (!(sb->s_flags & MS_RDONLY) && features) {
2268 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2269 "unsupported option features (%Lx).\n",
2270 (unsigned long long)features);
2271 err = -EINVAL;
2272 goto fail_alloc;
2273 }
2274
2275 btrfs_init_workers(&fs_info->generic_worker,
2276 "genwork", 1, NULL);
2277
2278 btrfs_init_workers(&fs_info->workers, "worker",
2279 fs_info->thread_pool_size,
2280 &fs_info->generic_worker);
2281
2282 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2283 fs_info->thread_pool_size,
2284 &fs_info->generic_worker);
2285
2286 btrfs_init_workers(&fs_info->submit_workers, "submit",
2287 min_t(u64, fs_devices->num_devices,
2288 fs_info->thread_pool_size),
2289 &fs_info->generic_worker);
2290
2291 btrfs_init_workers(&fs_info->caching_workers, "cache",
2292 2, &fs_info->generic_worker);
2293
2294 /* a higher idle thresh on the submit workers makes it much more
2295 * likely that bios will be send down in a sane order to the
2296 * devices
2297 */
2298 fs_info->submit_workers.idle_thresh = 64;
2299
2300 fs_info->workers.idle_thresh = 16;
2301 fs_info->workers.ordered = 1;
2302
2303 fs_info->delalloc_workers.idle_thresh = 2;
2304 fs_info->delalloc_workers.ordered = 1;
2305
2306 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2307 &fs_info->generic_worker);
2308 btrfs_init_workers(&fs_info->endio_workers, "endio",
2309 fs_info->thread_pool_size,
2310 &fs_info->generic_worker);
2311 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2312 fs_info->thread_pool_size,
2313 &fs_info->generic_worker);
2314 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2315 "endio-meta-write", fs_info->thread_pool_size,
2316 &fs_info->generic_worker);
2317 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2318 fs_info->thread_pool_size,
2319 &fs_info->generic_worker);
2320 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2321 1, &fs_info->generic_worker);
2322 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2323 fs_info->thread_pool_size,
2324 &fs_info->generic_worker);
2325 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2326 fs_info->thread_pool_size,
2327 &fs_info->generic_worker);
2328
2329 /*
2330 * endios are largely parallel and should have a very
2331 * low idle thresh
2332 */
2333 fs_info->endio_workers.idle_thresh = 4;
2334 fs_info->endio_meta_workers.idle_thresh = 4;
2335
2336 fs_info->endio_write_workers.idle_thresh = 2;
2337 fs_info->endio_meta_write_workers.idle_thresh = 2;
2338 fs_info->readahead_workers.idle_thresh = 2;
2339
2340 /*
2341 * btrfs_start_workers can really only fail because of ENOMEM so just
2342 * return -ENOMEM if any of these fail.
2343 */
2344 ret = btrfs_start_workers(&fs_info->workers);
2345 ret |= btrfs_start_workers(&fs_info->generic_worker);
2346 ret |= btrfs_start_workers(&fs_info->submit_workers);
2347 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2348 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2349 ret |= btrfs_start_workers(&fs_info->endio_workers);
2350 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2351 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2352 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2353 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2354 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2355 ret |= btrfs_start_workers(&fs_info->caching_workers);
2356 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2357 if (ret) {
2358 err = -ENOMEM;
2359 goto fail_sb_buffer;
2360 }
2361
2362 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2363 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2364 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2365
2366 tree_root->nodesize = nodesize;
2367 tree_root->leafsize = leafsize;
2368 tree_root->sectorsize = sectorsize;
2369 tree_root->stripesize = stripesize;
2370
2371 sb->s_blocksize = sectorsize;
2372 sb->s_blocksize_bits = blksize_bits(sectorsize);
2373
2374 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2375 sizeof(disk_super->magic))) {
2376 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2377 goto fail_sb_buffer;
2378 }
2379
2380 if (sectorsize != PAGE_SIZE) {
2381 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2382 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2383 goto fail_sb_buffer;
2384 }
2385
2386 mutex_lock(&fs_info->chunk_mutex);
2387 ret = btrfs_read_sys_array(tree_root);
2388 mutex_unlock(&fs_info->chunk_mutex);
2389 if (ret) {
2390 printk(KERN_WARNING "btrfs: failed to read the system "
2391 "array on %s\n", sb->s_id);
2392 goto fail_sb_buffer;
2393 }
2394
2395 blocksize = btrfs_level_size(tree_root,
2396 btrfs_super_chunk_root_level(disk_super));
2397 generation = btrfs_super_chunk_root_generation(disk_super);
2398
2399 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2400 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2401
2402 chunk_root->node = read_tree_block(chunk_root,
2403 btrfs_super_chunk_root(disk_super),
2404 blocksize, generation);
2405 BUG_ON(!chunk_root->node); /* -ENOMEM */
2406 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2407 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2408 sb->s_id);
2409 goto fail_tree_roots;
2410 }
2411 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2412 chunk_root->commit_root = btrfs_root_node(chunk_root);
2413
2414 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2415 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2416 BTRFS_UUID_SIZE);
2417
2418 ret = btrfs_read_chunk_tree(chunk_root);
2419 if (ret) {
2420 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2421 sb->s_id);
2422 goto fail_tree_roots;
2423 }
2424
2425 btrfs_close_extra_devices(fs_devices);
2426
2427 if (!fs_devices->latest_bdev) {
2428 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2429 sb->s_id);
2430 goto fail_tree_roots;
2431 }
2432
2433retry_root_backup:
2434 blocksize = btrfs_level_size(tree_root,
2435 btrfs_super_root_level(disk_super));
2436 generation = btrfs_super_generation(disk_super);
2437
2438 tree_root->node = read_tree_block(tree_root,
2439 btrfs_super_root(disk_super),
2440 blocksize, generation);
2441 if (!tree_root->node ||
2442 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2443 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2444 sb->s_id);
2445
2446 goto recovery_tree_root;
2447 }
2448
2449 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2450 tree_root->commit_root = btrfs_root_node(tree_root);
2451
2452 ret = find_and_setup_root(tree_root, fs_info,
2453 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2454 if (ret)
2455 goto recovery_tree_root;
2456 extent_root->track_dirty = 1;
2457
2458 ret = find_and_setup_root(tree_root, fs_info,
2459 BTRFS_DEV_TREE_OBJECTID, dev_root);
2460 if (ret)
2461 goto recovery_tree_root;
2462 dev_root->track_dirty = 1;
2463
2464 ret = find_and_setup_root(tree_root, fs_info,
2465 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2466 if (ret)
2467 goto recovery_tree_root;
2468 csum_root->track_dirty = 1;
2469
2470 ret = find_and_setup_root(tree_root, fs_info,
2471 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2472 if (ret) {
2473 kfree(quota_root);
2474 quota_root = fs_info->quota_root = NULL;
2475 } else {
2476 quota_root->track_dirty = 1;
2477 fs_info->quota_enabled = 1;
2478 fs_info->pending_quota_state = 1;
2479 }
2480
2481 fs_info->generation = generation;
2482 fs_info->last_trans_committed = generation;
2483
2484 ret = btrfs_recover_balance(fs_info);
2485 if (ret) {
2486 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2487 goto fail_block_groups;
2488 }
2489
2490 ret = btrfs_init_dev_stats(fs_info);
2491 if (ret) {
2492 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2493 ret);
2494 goto fail_block_groups;
2495 }
2496
2497 ret = btrfs_init_space_info(fs_info);
2498 if (ret) {
2499 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2500 goto fail_block_groups;
2501 }
2502
2503 ret = btrfs_read_block_groups(extent_root);
2504 if (ret) {
2505 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2506 goto fail_block_groups;
2507 }
2508
2509 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2510 "btrfs-cleaner");
2511 if (IS_ERR(fs_info->cleaner_kthread))
2512 goto fail_block_groups;
2513
2514 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2515 tree_root,
2516 "btrfs-transaction");
2517 if (IS_ERR(fs_info->transaction_kthread))
2518 goto fail_cleaner;
2519
2520 if (!btrfs_test_opt(tree_root, SSD) &&
2521 !btrfs_test_opt(tree_root, NOSSD) &&
2522 !fs_info->fs_devices->rotating) {
2523 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2524 "mode\n");
2525 btrfs_set_opt(fs_info->mount_opt, SSD);
2526 }
2527
2528#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2529 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2530 ret = btrfsic_mount(tree_root, fs_devices,
2531 btrfs_test_opt(tree_root,
2532 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2533 1 : 0,
2534 fs_info->check_integrity_print_mask);
2535 if (ret)
2536 printk(KERN_WARNING "btrfs: failed to initialize"
2537 " integrity check module %s\n", sb->s_id);
2538 }
2539#endif
2540 ret = btrfs_read_qgroup_config(fs_info);
2541 if (ret)
2542 goto fail_trans_kthread;
2543
2544 /* do not make disk changes in broken FS */
2545 if (btrfs_super_log_root(disk_super) != 0) {
2546 u64 bytenr = btrfs_super_log_root(disk_super);
2547
2548 if (fs_devices->rw_devices == 0) {
2549 printk(KERN_WARNING "Btrfs log replay required "
2550 "on RO media\n");
2551 err = -EIO;
2552 goto fail_qgroup;
2553 }
2554 blocksize =
2555 btrfs_level_size(tree_root,
2556 btrfs_super_log_root_level(disk_super));
2557
2558 log_tree_root = btrfs_alloc_root(fs_info);
2559 if (!log_tree_root) {
2560 err = -ENOMEM;
2561 goto fail_qgroup;
2562 }
2563
2564 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2565 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2566
2567 log_tree_root->node = read_tree_block(tree_root, bytenr,
2568 blocksize,
2569 generation + 1);
2570 /* returns with log_tree_root freed on success */
2571 ret = btrfs_recover_log_trees(log_tree_root);
2572 if (ret) {
2573 btrfs_error(tree_root->fs_info, ret,
2574 "Failed to recover log tree");
2575 free_extent_buffer(log_tree_root->node);
2576 kfree(log_tree_root);
2577 goto fail_trans_kthread;
2578 }
2579
2580 if (sb->s_flags & MS_RDONLY) {
2581 ret = btrfs_commit_super(tree_root);
2582 if (ret)
2583 goto fail_trans_kthread;
2584 }
2585 }
2586
2587 ret = btrfs_find_orphan_roots(tree_root);
2588 if (ret)
2589 goto fail_trans_kthread;
2590
2591 if (!(sb->s_flags & MS_RDONLY)) {
2592 ret = btrfs_cleanup_fs_roots(fs_info);
2593 if (ret)
2594 goto fail_trans_kthread;
2595
2596 ret = btrfs_recover_relocation(tree_root);
2597 if (ret < 0) {
2598 printk(KERN_WARNING
2599 "btrfs: failed to recover relocation\n");
2600 err = -EINVAL;
2601 goto fail_qgroup;
2602 }
2603 }
2604
2605 location.objectid = BTRFS_FS_TREE_OBJECTID;
2606 location.type = BTRFS_ROOT_ITEM_KEY;
2607 location.offset = (u64)-1;
2608
2609 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2610 if (!fs_info->fs_root)
2611 goto fail_qgroup;
2612 if (IS_ERR(fs_info->fs_root)) {
2613 err = PTR_ERR(fs_info->fs_root);
2614 goto fail_qgroup;
2615 }
2616
2617 if (sb->s_flags & MS_RDONLY)
2618 return 0;
2619
2620 down_read(&fs_info->cleanup_work_sem);
2621 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2622 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2623 up_read(&fs_info->cleanup_work_sem);
2624 close_ctree(tree_root);
2625 return ret;
2626 }
2627 up_read(&fs_info->cleanup_work_sem);
2628
2629 ret = btrfs_resume_balance_async(fs_info);
2630 if (ret) {
2631 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2632 close_ctree(tree_root);
2633 return ret;
2634 }
2635
2636 return 0;
2637
2638fail_qgroup:
2639 btrfs_free_qgroup_config(fs_info);
2640fail_trans_kthread:
2641 kthread_stop(fs_info->transaction_kthread);
2642fail_cleaner:
2643 kthread_stop(fs_info->cleaner_kthread);
2644
2645 /*
2646 * make sure we're done with the btree inode before we stop our
2647 * kthreads
2648 */
2649 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2650 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2651
2652fail_block_groups:
2653 btrfs_free_block_groups(fs_info);
2654
2655fail_tree_roots:
2656 free_root_pointers(fs_info, 1);
2657
2658fail_sb_buffer:
2659 btrfs_stop_workers(&fs_info->generic_worker);
2660 btrfs_stop_workers(&fs_info->readahead_workers);
2661 btrfs_stop_workers(&fs_info->fixup_workers);
2662 btrfs_stop_workers(&fs_info->delalloc_workers);
2663 btrfs_stop_workers(&fs_info->workers);
2664 btrfs_stop_workers(&fs_info->endio_workers);
2665 btrfs_stop_workers(&fs_info->endio_meta_workers);
2666 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2667 btrfs_stop_workers(&fs_info->endio_write_workers);
2668 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2669 btrfs_stop_workers(&fs_info->submit_workers);
2670 btrfs_stop_workers(&fs_info->delayed_workers);
2671 btrfs_stop_workers(&fs_info->caching_workers);
2672fail_alloc:
2673fail_iput:
2674 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2675
2676 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2677 iput(fs_info->btree_inode);
2678fail_bdi:
2679 bdi_destroy(&fs_info->bdi);
2680fail_srcu:
2681 cleanup_srcu_struct(&fs_info->subvol_srcu);
2682fail:
2683 btrfs_close_devices(fs_info->fs_devices);
2684 return err;
2685
2686recovery_tree_root:
2687 if (!btrfs_test_opt(tree_root, RECOVERY))
2688 goto fail_tree_roots;
2689
2690 free_root_pointers(fs_info, 0);
2691
2692 /* don't use the log in recovery mode, it won't be valid */
2693 btrfs_set_super_log_root(disk_super, 0);
2694
2695 /* we can't trust the free space cache either */
2696 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2697
2698 ret = next_root_backup(fs_info, fs_info->super_copy,
2699 &num_backups_tried, &backup_index);
2700 if (ret == -1)
2701 goto fail_block_groups;
2702 goto retry_root_backup;
2703}
2704
2705static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2706{
2707 if (uptodate) {
2708 set_buffer_uptodate(bh);
2709 } else {
2710 struct btrfs_device *device = (struct btrfs_device *)
2711 bh->b_private;
2712
2713 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2714 "I/O error on %s\n",
2715 rcu_str_deref(device->name));
2716 /* note, we dont' set_buffer_write_io_error because we have
2717 * our own ways of dealing with the IO errors
2718 */
2719 clear_buffer_uptodate(bh);
2720 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2721 }
2722 unlock_buffer(bh);
2723 put_bh(bh);
2724}
2725
2726struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2727{
2728 struct buffer_head *bh;
2729 struct buffer_head *latest = NULL;
2730 struct btrfs_super_block *super;
2731 int i;
2732 u64 transid = 0;
2733 u64 bytenr;
2734
2735 /* we would like to check all the supers, but that would make
2736 * a btrfs mount succeed after a mkfs from a different FS.
2737 * So, we need to add a special mount option to scan for
2738 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2739 */
2740 for (i = 0; i < 1; i++) {
2741 bytenr = btrfs_sb_offset(i);
2742 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2743 break;
2744 bh = __bread(bdev, bytenr / 4096, 4096);
2745 if (!bh)
2746 continue;
2747
2748 super = (struct btrfs_super_block *)bh->b_data;
2749 if (btrfs_super_bytenr(super) != bytenr ||
2750 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2751 sizeof(super->magic))) {
2752 brelse(bh);
2753 continue;
2754 }
2755
2756 if (!latest || btrfs_super_generation(super) > transid) {
2757 brelse(latest);
2758 latest = bh;
2759 transid = btrfs_super_generation(super);
2760 } else {
2761 brelse(bh);
2762 }
2763 }
2764 return latest;
2765}
2766
2767/*
2768 * this should be called twice, once with wait == 0 and
2769 * once with wait == 1. When wait == 0 is done, all the buffer heads
2770 * we write are pinned.
2771 *
2772 * They are released when wait == 1 is done.
2773 * max_mirrors must be the same for both runs, and it indicates how
2774 * many supers on this one device should be written.
2775 *
2776 * max_mirrors == 0 means to write them all.
2777 */
2778static int write_dev_supers(struct btrfs_device *device,
2779 struct btrfs_super_block *sb,
2780 int do_barriers, int wait, int max_mirrors)
2781{
2782 struct buffer_head *bh;
2783 int i;
2784 int ret;
2785 int errors = 0;
2786 u32 crc;
2787 u64 bytenr;
2788
2789 if (max_mirrors == 0)
2790 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2791
2792 for (i = 0; i < max_mirrors; i++) {
2793 bytenr = btrfs_sb_offset(i);
2794 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2795 break;
2796
2797 if (wait) {
2798 bh = __find_get_block(device->bdev, bytenr / 4096,
2799 BTRFS_SUPER_INFO_SIZE);
2800 BUG_ON(!bh);
2801 wait_on_buffer(bh);
2802 if (!buffer_uptodate(bh))
2803 errors++;
2804
2805 /* drop our reference */
2806 brelse(bh);
2807
2808 /* drop the reference from the wait == 0 run */
2809 brelse(bh);
2810 continue;
2811 } else {
2812 btrfs_set_super_bytenr(sb, bytenr);
2813
2814 crc = ~(u32)0;
2815 crc = btrfs_csum_data(NULL, (char *)sb +
2816 BTRFS_CSUM_SIZE, crc,
2817 BTRFS_SUPER_INFO_SIZE -
2818 BTRFS_CSUM_SIZE);
2819 btrfs_csum_final(crc, sb->csum);
2820
2821 /*
2822 * one reference for us, and we leave it for the
2823 * caller
2824 */
2825 bh = __getblk(device->bdev, bytenr / 4096,
2826 BTRFS_SUPER_INFO_SIZE);
2827 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2828
2829 /* one reference for submit_bh */
2830 get_bh(bh);
2831
2832 set_buffer_uptodate(bh);
2833 lock_buffer(bh);
2834 bh->b_end_io = btrfs_end_buffer_write_sync;
2835 bh->b_private = device;
2836 }
2837
2838 /*
2839 * we fua the first super. The others we allow
2840 * to go down lazy.
2841 */
2842 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2843 if (ret)
2844 errors++;
2845 }
2846 return errors < i ? 0 : -1;
2847}
2848
2849/*
2850 * endio for the write_dev_flush, this will wake anyone waiting
2851 * for the barrier when it is done
2852 */
2853static void btrfs_end_empty_barrier(struct bio *bio, int err)
2854{
2855 if (err) {
2856 if (err == -EOPNOTSUPP)
2857 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2858 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2859 }
2860 if (bio->bi_private)
2861 complete(bio->bi_private);
2862 bio_put(bio);
2863}
2864
2865/*
2866 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2867 * sent down. With wait == 1, it waits for the previous flush.
2868 *
2869 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2870 * capable
2871 */
2872static int write_dev_flush(struct btrfs_device *device, int wait)
2873{
2874 struct bio *bio;
2875 int ret = 0;
2876
2877 if (device->nobarriers)
2878 return 0;
2879
2880 if (wait) {
2881 bio = device->flush_bio;
2882 if (!bio)
2883 return 0;
2884
2885 wait_for_completion(&device->flush_wait);
2886
2887 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2888 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2889 rcu_str_deref(device->name));
2890 device->nobarriers = 1;
2891 }
2892 if (!bio_flagged(bio, BIO_UPTODATE)) {
2893 ret = -EIO;
2894 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2895 btrfs_dev_stat_inc_and_print(device,
2896 BTRFS_DEV_STAT_FLUSH_ERRS);
2897 }
2898
2899 /* drop the reference from the wait == 0 run */
2900 bio_put(bio);
2901 device->flush_bio = NULL;
2902
2903 return ret;
2904 }
2905
2906 /*
2907 * one reference for us, and we leave it for the
2908 * caller
2909 */
2910 device->flush_bio = NULL;
2911 bio = bio_alloc(GFP_NOFS, 0);
2912 if (!bio)
2913 return -ENOMEM;
2914
2915 bio->bi_end_io = btrfs_end_empty_barrier;
2916 bio->bi_bdev = device->bdev;
2917 init_completion(&device->flush_wait);
2918 bio->bi_private = &device->flush_wait;
2919 device->flush_bio = bio;
2920
2921 bio_get(bio);
2922 btrfsic_submit_bio(WRITE_FLUSH, bio);
2923
2924 return 0;
2925}
2926
2927/*
2928 * send an empty flush down to each device in parallel,
2929 * then wait for them
2930 */
2931static int barrier_all_devices(struct btrfs_fs_info *info)
2932{
2933 struct list_head *head;
2934 struct btrfs_device *dev;
2935 int errors = 0;
2936 int ret;
2937
2938 /* send down all the barriers */
2939 head = &info->fs_devices->devices;
2940 list_for_each_entry_rcu(dev, head, dev_list) {
2941 if (!dev->bdev) {
2942 errors++;
2943 continue;
2944 }
2945 if (!dev->in_fs_metadata || !dev->writeable)
2946 continue;
2947
2948 ret = write_dev_flush(dev, 0);
2949 if (ret)
2950 errors++;
2951 }
2952
2953 /* wait for all the barriers */
2954 list_for_each_entry_rcu(dev, head, dev_list) {
2955 if (!dev->bdev) {
2956 errors++;
2957 continue;
2958 }
2959 if (!dev->in_fs_metadata || !dev->writeable)
2960 continue;
2961
2962 ret = write_dev_flush(dev, 1);
2963 if (ret)
2964 errors++;
2965 }
2966 if (errors)
2967 return -EIO;
2968 return 0;
2969}
2970
2971int write_all_supers(struct btrfs_root *root, int max_mirrors)
2972{
2973 struct list_head *head;
2974 struct btrfs_device *dev;
2975 struct btrfs_super_block *sb;
2976 struct btrfs_dev_item *dev_item;
2977 int ret;
2978 int do_barriers;
2979 int max_errors;
2980 int total_errors = 0;
2981 u64 flags;
2982
2983 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2984 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2985 backup_super_roots(root->fs_info);
2986
2987 sb = root->fs_info->super_for_commit;
2988 dev_item = &sb->dev_item;
2989
2990 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2991 head = &root->fs_info->fs_devices->devices;
2992
2993 if (do_barriers)
2994 barrier_all_devices(root->fs_info);
2995
2996 list_for_each_entry_rcu(dev, head, dev_list) {
2997 if (!dev->bdev) {
2998 total_errors++;
2999 continue;
3000 }
3001 if (!dev->in_fs_metadata || !dev->writeable)
3002 continue;
3003
3004 btrfs_set_stack_device_generation(dev_item, 0);
3005 btrfs_set_stack_device_type(dev_item, dev->type);
3006 btrfs_set_stack_device_id(dev_item, dev->devid);
3007 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3008 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3009 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3010 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3011 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3012 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3013 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3014
3015 flags = btrfs_super_flags(sb);
3016 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3017
3018 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3019 if (ret)
3020 total_errors++;
3021 }
3022 if (total_errors > max_errors) {
3023 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3024 total_errors);
3025
3026 /* This shouldn't happen. FUA is masked off if unsupported */
3027 BUG();
3028 }
3029
3030 total_errors = 0;
3031 list_for_each_entry_rcu(dev, head, dev_list) {
3032 if (!dev->bdev)
3033 continue;
3034 if (!dev->in_fs_metadata || !dev->writeable)
3035 continue;
3036
3037 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3038 if (ret)
3039 total_errors++;
3040 }
3041 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3042 if (total_errors > max_errors) {
3043 btrfs_error(root->fs_info, -EIO,
3044 "%d errors while writing supers", total_errors);
3045 return -EIO;
3046 }
3047 return 0;
3048}
3049
3050int write_ctree_super(struct btrfs_trans_handle *trans,
3051 struct btrfs_root *root, int max_mirrors)
3052{
3053 int ret;
3054
3055 ret = write_all_supers(root, max_mirrors);
3056 return ret;
3057}
3058
3059void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3060{
3061 spin_lock(&fs_info->fs_roots_radix_lock);
3062 radix_tree_delete(&fs_info->fs_roots_radix,
3063 (unsigned long)root->root_key.objectid);
3064 spin_unlock(&fs_info->fs_roots_radix_lock);
3065
3066 if (btrfs_root_refs(&root->root_item) == 0)
3067 synchronize_srcu(&fs_info->subvol_srcu);
3068
3069 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3070 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3071 free_fs_root(root);
3072}
3073
3074static void free_fs_root(struct btrfs_root *root)
3075{
3076 iput(root->cache_inode);
3077 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3078 if (root->anon_dev)
3079 free_anon_bdev(root->anon_dev);
3080 free_extent_buffer(root->node);
3081 free_extent_buffer(root->commit_root);
3082 kfree(root->free_ino_ctl);
3083 kfree(root->free_ino_pinned);
3084 kfree(root->name);
3085 kfree(root);
3086}
3087
3088static void del_fs_roots(struct btrfs_fs_info *fs_info)
3089{
3090 int ret;
3091 struct btrfs_root *gang[8];
3092 int i;
3093
3094 while (!list_empty(&fs_info->dead_roots)) {
3095 gang[0] = list_entry(fs_info->dead_roots.next,
3096 struct btrfs_root, root_list);
3097 list_del(&gang[0]->root_list);
3098
3099 if (gang[0]->in_radix) {
3100 btrfs_free_fs_root(fs_info, gang[0]);
3101 } else {
3102 free_extent_buffer(gang[0]->node);
3103 free_extent_buffer(gang[0]->commit_root);
3104 kfree(gang[0]);
3105 }
3106 }
3107
3108 while (1) {
3109 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3110 (void **)gang, 0,
3111 ARRAY_SIZE(gang));
3112 if (!ret)
3113 break;
3114 for (i = 0; i < ret; i++)
3115 btrfs_free_fs_root(fs_info, gang[i]);
3116 }
3117}
3118
3119int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3120{
3121 u64 root_objectid = 0;
3122 struct btrfs_root *gang[8];
3123 int i;
3124 int ret;
3125
3126 while (1) {
3127 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3128 (void **)gang, root_objectid,
3129 ARRAY_SIZE(gang));
3130 if (!ret)
3131 break;
3132
3133 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3134 for (i = 0; i < ret; i++) {
3135 int err;
3136
3137 root_objectid = gang[i]->root_key.objectid;
3138 err = btrfs_orphan_cleanup(gang[i]);
3139 if (err)
3140 return err;
3141 }
3142 root_objectid++;
3143 }
3144 return 0;
3145}
3146
3147int btrfs_commit_super(struct btrfs_root *root)
3148{
3149 struct btrfs_trans_handle *trans;
3150 int ret;
3151
3152 mutex_lock(&root->fs_info->cleaner_mutex);
3153 btrfs_run_delayed_iputs(root);
3154 btrfs_clean_old_snapshots(root);
3155 mutex_unlock(&root->fs_info->cleaner_mutex);
3156
3157 /* wait until ongoing cleanup work done */
3158 down_write(&root->fs_info->cleanup_work_sem);
3159 up_write(&root->fs_info->cleanup_work_sem);
3160
3161 trans = btrfs_join_transaction(root);
3162 if (IS_ERR(trans))
3163 return PTR_ERR(trans);
3164 ret = btrfs_commit_transaction(trans, root);
3165 if (ret)
3166 return ret;
3167 /* run commit again to drop the original snapshot */
3168 trans = btrfs_join_transaction(root);
3169 if (IS_ERR(trans))
3170 return PTR_ERR(trans);
3171 ret = btrfs_commit_transaction(trans, root);
3172 if (ret)
3173 return ret;
3174 ret = btrfs_write_and_wait_transaction(NULL, root);
3175 if (ret) {
3176 btrfs_error(root->fs_info, ret,
3177 "Failed to sync btree inode to disk.");
3178 return ret;
3179 }
3180
3181 ret = write_ctree_super(NULL, root, 0);
3182 return ret;
3183}
3184
3185int close_ctree(struct btrfs_root *root)
3186{
3187 struct btrfs_fs_info *fs_info = root->fs_info;
3188 int ret;
3189
3190 fs_info->closing = 1;
3191 smp_mb();
3192
3193 /* pause restriper - we want to resume on mount */
3194 btrfs_pause_balance(root->fs_info);
3195
3196 btrfs_scrub_cancel(root);
3197
3198 /* wait for any defraggers to finish */
3199 wait_event(fs_info->transaction_wait,
3200 (atomic_read(&fs_info->defrag_running) == 0));
3201
3202 /* clear out the rbtree of defraggable inodes */
3203 btrfs_run_defrag_inodes(fs_info);
3204
3205 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3206 ret = btrfs_commit_super(root);
3207 if (ret)
3208 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3209 }
3210
3211 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3212 btrfs_error_commit_super(root);
3213
3214 btrfs_put_block_group_cache(fs_info);
3215
3216 kthread_stop(fs_info->transaction_kthread);
3217 kthread_stop(fs_info->cleaner_kthread);
3218
3219 fs_info->closing = 2;
3220 smp_mb();
3221
3222 btrfs_free_qgroup_config(root->fs_info);
3223
3224 if (fs_info->delalloc_bytes) {
3225 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3226 (unsigned long long)fs_info->delalloc_bytes);
3227 }
3228
3229 free_extent_buffer(fs_info->extent_root->node);
3230 free_extent_buffer(fs_info->extent_root->commit_root);
3231 free_extent_buffer(fs_info->tree_root->node);
3232 free_extent_buffer(fs_info->tree_root->commit_root);
3233 free_extent_buffer(fs_info->chunk_root->node);
3234 free_extent_buffer(fs_info->chunk_root->commit_root);
3235 free_extent_buffer(fs_info->dev_root->node);
3236 free_extent_buffer(fs_info->dev_root->commit_root);
3237 free_extent_buffer(fs_info->csum_root->node);
3238 free_extent_buffer(fs_info->csum_root->commit_root);
3239 if (fs_info->quota_root) {
3240 free_extent_buffer(fs_info->quota_root->node);
3241 free_extent_buffer(fs_info->quota_root->commit_root);
3242 }
3243
3244 btrfs_free_block_groups(fs_info);
3245
3246 del_fs_roots(fs_info);
3247
3248 iput(fs_info->btree_inode);
3249
3250 btrfs_stop_workers(&fs_info->generic_worker);
3251 btrfs_stop_workers(&fs_info->fixup_workers);
3252 btrfs_stop_workers(&fs_info->delalloc_workers);
3253 btrfs_stop_workers(&fs_info->workers);
3254 btrfs_stop_workers(&fs_info->endio_workers);
3255 btrfs_stop_workers(&fs_info->endio_meta_workers);
3256 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3257 btrfs_stop_workers(&fs_info->endio_write_workers);
3258 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3259 btrfs_stop_workers(&fs_info->submit_workers);
3260 btrfs_stop_workers(&fs_info->delayed_workers);
3261 btrfs_stop_workers(&fs_info->caching_workers);
3262 btrfs_stop_workers(&fs_info->readahead_workers);
3263
3264#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3265 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3266 btrfsic_unmount(root, fs_info->fs_devices);
3267#endif
3268
3269 btrfs_close_devices(fs_info->fs_devices);
3270 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3271
3272 bdi_destroy(&fs_info->bdi);
3273 cleanup_srcu_struct(&fs_info->subvol_srcu);
3274
3275 return 0;
3276}
3277
3278int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3279 int atomic)
3280{
3281 int ret;
3282 struct inode *btree_inode = buf->pages[0]->mapping->host;
3283
3284 ret = extent_buffer_uptodate(buf);
3285 if (!ret)
3286 return ret;
3287
3288 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3289 parent_transid, atomic);
3290 if (ret == -EAGAIN)
3291 return ret;
3292 return !ret;
3293}
3294
3295int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3296{
3297 return set_extent_buffer_uptodate(buf);
3298}
3299
3300void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3301{
3302 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3303 u64 transid = btrfs_header_generation(buf);
3304 int was_dirty;
3305
3306 btrfs_assert_tree_locked(buf);
3307 if (transid != root->fs_info->generation) {
3308 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3309 "found %llu running %llu\n",
3310 (unsigned long long)buf->start,
3311 (unsigned long long)transid,
3312 (unsigned long long)root->fs_info->generation);
3313 WARN_ON(1);
3314 }
3315 was_dirty = set_extent_buffer_dirty(buf);
3316 if (!was_dirty) {
3317 spin_lock(&root->fs_info->delalloc_lock);
3318 root->fs_info->dirty_metadata_bytes += buf->len;
3319 spin_unlock(&root->fs_info->delalloc_lock);
3320 }
3321}
3322
3323void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3324{
3325 /*
3326 * looks as though older kernels can get into trouble with
3327 * this code, they end up stuck in balance_dirty_pages forever
3328 */
3329 u64 num_dirty;
3330 unsigned long thresh = 32 * 1024 * 1024;
3331
3332 if (current->flags & PF_MEMALLOC)
3333 return;
3334
3335 btrfs_balance_delayed_items(root);
3336
3337 num_dirty = root->fs_info->dirty_metadata_bytes;
3338
3339 if (num_dirty > thresh) {
3340 balance_dirty_pages_ratelimited_nr(
3341 root->fs_info->btree_inode->i_mapping, 1);
3342 }
3343 return;
3344}
3345
3346void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3347{
3348 /*
3349 * looks as though older kernels can get into trouble with
3350 * this code, they end up stuck in balance_dirty_pages forever
3351 */
3352 u64 num_dirty;
3353 unsigned long thresh = 32 * 1024 * 1024;
3354
3355 if (current->flags & PF_MEMALLOC)
3356 return;
3357
3358 num_dirty = root->fs_info->dirty_metadata_bytes;
3359
3360 if (num_dirty > thresh) {
3361 balance_dirty_pages_ratelimited_nr(
3362 root->fs_info->btree_inode->i_mapping, 1);
3363 }
3364 return;
3365}
3366
3367int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3368{
3369 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3370 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3371}
3372
3373static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3374 int read_only)
3375{
3376 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3377 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3378 return -EINVAL;
3379 }
3380
3381 if (read_only)
3382 return 0;
3383
3384 return 0;
3385}
3386
3387void btrfs_error_commit_super(struct btrfs_root *root)
3388{
3389 mutex_lock(&root->fs_info->cleaner_mutex);
3390 btrfs_run_delayed_iputs(root);
3391 mutex_unlock(&root->fs_info->cleaner_mutex);
3392
3393 down_write(&root->fs_info->cleanup_work_sem);
3394 up_write(&root->fs_info->cleanup_work_sem);
3395
3396 /* cleanup FS via transaction */
3397 btrfs_cleanup_transaction(root);
3398}
3399
3400static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3401{
3402 struct btrfs_inode *btrfs_inode;
3403 struct list_head splice;
3404
3405 INIT_LIST_HEAD(&splice);
3406
3407 mutex_lock(&root->fs_info->ordered_operations_mutex);
3408 spin_lock(&root->fs_info->ordered_extent_lock);
3409
3410 list_splice_init(&root->fs_info->ordered_operations, &splice);
3411 while (!list_empty(&splice)) {
3412 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3413 ordered_operations);
3414
3415 list_del_init(&btrfs_inode->ordered_operations);
3416
3417 btrfs_invalidate_inodes(btrfs_inode->root);
3418 }
3419
3420 spin_unlock(&root->fs_info->ordered_extent_lock);
3421 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3422}
3423
3424static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3425{
3426 struct list_head splice;
3427 struct btrfs_ordered_extent *ordered;
3428 struct inode *inode;
3429
3430 INIT_LIST_HEAD(&splice);
3431
3432 spin_lock(&root->fs_info->ordered_extent_lock);
3433
3434 list_splice_init(&root->fs_info->ordered_extents, &splice);
3435 while (!list_empty(&splice)) {
3436 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3437 root_extent_list);
3438
3439 list_del_init(&ordered->root_extent_list);
3440 atomic_inc(&ordered->refs);
3441
3442 /* the inode may be getting freed (in sys_unlink path). */
3443 inode = igrab(ordered->inode);
3444
3445 spin_unlock(&root->fs_info->ordered_extent_lock);
3446 if (inode)
3447 iput(inode);
3448
3449 atomic_set(&ordered->refs, 1);
3450 btrfs_put_ordered_extent(ordered);
3451
3452 spin_lock(&root->fs_info->ordered_extent_lock);
3453 }
3454
3455 spin_unlock(&root->fs_info->ordered_extent_lock);
3456}
3457
3458int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3459 struct btrfs_root *root)
3460{
3461 struct rb_node *node;
3462 struct btrfs_delayed_ref_root *delayed_refs;
3463 struct btrfs_delayed_ref_node *ref;
3464 int ret = 0;
3465
3466 delayed_refs = &trans->delayed_refs;
3467
3468 spin_lock(&delayed_refs->lock);
3469 if (delayed_refs->num_entries == 0) {
3470 spin_unlock(&delayed_refs->lock);
3471 printk(KERN_INFO "delayed_refs has NO entry\n");
3472 return ret;
3473 }
3474
3475 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3476 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3477
3478 atomic_set(&ref->refs, 1);
3479 if (btrfs_delayed_ref_is_head(ref)) {
3480 struct btrfs_delayed_ref_head *head;
3481
3482 head = btrfs_delayed_node_to_head(ref);
3483 if (!mutex_trylock(&head->mutex)) {
3484 atomic_inc(&ref->refs);
3485 spin_unlock(&delayed_refs->lock);
3486
3487 /* Need to wait for the delayed ref to run */
3488 mutex_lock(&head->mutex);
3489 mutex_unlock(&head->mutex);
3490 btrfs_put_delayed_ref(ref);
3491
3492 spin_lock(&delayed_refs->lock);
3493 continue;
3494 }
3495
3496 kfree(head->extent_op);
3497 delayed_refs->num_heads--;
3498 if (list_empty(&head->cluster))
3499 delayed_refs->num_heads_ready--;
3500 list_del_init(&head->cluster);
3501 }
3502 ref->in_tree = 0;
3503 rb_erase(&ref->rb_node, &delayed_refs->root);
3504 delayed_refs->num_entries--;
3505
3506 spin_unlock(&delayed_refs->lock);
3507 btrfs_put_delayed_ref(ref);
3508
3509 cond_resched();
3510 spin_lock(&delayed_refs->lock);
3511 }
3512
3513 spin_unlock(&delayed_refs->lock);
3514
3515 return ret;
3516}
3517
3518static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3519{
3520 struct btrfs_pending_snapshot *snapshot;
3521 struct list_head splice;
3522
3523 INIT_LIST_HEAD(&splice);
3524
3525 list_splice_init(&t->pending_snapshots, &splice);
3526
3527 while (!list_empty(&splice)) {
3528 snapshot = list_entry(splice.next,
3529 struct btrfs_pending_snapshot,
3530 list);
3531
3532 list_del_init(&snapshot->list);
3533
3534 kfree(snapshot);
3535 }
3536}
3537
3538static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3539{
3540 struct btrfs_inode *btrfs_inode;
3541 struct list_head splice;
3542
3543 INIT_LIST_HEAD(&splice);
3544
3545 spin_lock(&root->fs_info->delalloc_lock);
3546 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3547
3548 while (!list_empty(&splice)) {
3549 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3550 delalloc_inodes);
3551
3552 list_del_init(&btrfs_inode->delalloc_inodes);
3553
3554 btrfs_invalidate_inodes(btrfs_inode->root);
3555 }
3556
3557 spin_unlock(&root->fs_info->delalloc_lock);
3558}
3559
3560static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3561 struct extent_io_tree *dirty_pages,
3562 int mark)
3563{
3564 int ret;
3565 struct page *page;
3566 struct inode *btree_inode = root->fs_info->btree_inode;
3567 struct extent_buffer *eb;
3568 u64 start = 0;
3569 u64 end;
3570 u64 offset;
3571 unsigned long index;
3572
3573 while (1) {
3574 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3575 mark);
3576 if (ret)
3577 break;
3578
3579 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3580 while (start <= end) {
3581 index = start >> PAGE_CACHE_SHIFT;
3582 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3583 page = find_get_page(btree_inode->i_mapping, index);
3584 if (!page)
3585 continue;
3586 offset = page_offset(page);
3587
3588 spin_lock(&dirty_pages->buffer_lock);
3589 eb = radix_tree_lookup(
3590 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3591 offset >> PAGE_CACHE_SHIFT);
3592 spin_unlock(&dirty_pages->buffer_lock);
3593 if (eb)
3594 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3595 &eb->bflags);
3596 if (PageWriteback(page))
3597 end_page_writeback(page);
3598
3599 lock_page(page);
3600 if (PageDirty(page)) {
3601 clear_page_dirty_for_io(page);
3602 spin_lock_irq(&page->mapping->tree_lock);
3603 radix_tree_tag_clear(&page->mapping->page_tree,
3604 page_index(page),
3605 PAGECACHE_TAG_DIRTY);
3606 spin_unlock_irq(&page->mapping->tree_lock);
3607 }
3608
3609 unlock_page(page);
3610 page_cache_release(page);
3611 }
3612 }
3613
3614 return ret;
3615}
3616
3617static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3618 struct extent_io_tree *pinned_extents)
3619{
3620 struct extent_io_tree *unpin;
3621 u64 start;
3622 u64 end;
3623 int ret;
3624 bool loop = true;
3625
3626 unpin = pinned_extents;
3627again:
3628 while (1) {
3629 ret = find_first_extent_bit(unpin, 0, &start, &end,
3630 EXTENT_DIRTY);
3631 if (ret)
3632 break;
3633
3634 /* opt_discard */
3635 if (btrfs_test_opt(root, DISCARD))
3636 ret = btrfs_error_discard_extent(root, start,
3637 end + 1 - start,
3638 NULL);
3639
3640 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3641 btrfs_error_unpin_extent_range(root, start, end);
3642 cond_resched();
3643 }
3644
3645 if (loop) {
3646 if (unpin == &root->fs_info->freed_extents[0])
3647 unpin = &root->fs_info->freed_extents[1];
3648 else
3649 unpin = &root->fs_info->freed_extents[0];
3650 loop = false;
3651 goto again;
3652 }
3653
3654 return 0;
3655}
3656
3657void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3658 struct btrfs_root *root)
3659{
3660 btrfs_destroy_delayed_refs(cur_trans, root);
3661 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3662 cur_trans->dirty_pages.dirty_bytes);
3663
3664 /* FIXME: cleanup wait for commit */
3665 cur_trans->in_commit = 1;
3666 cur_trans->blocked = 1;
3667 wake_up(&root->fs_info->transaction_blocked_wait);
3668
3669 cur_trans->blocked = 0;
3670 wake_up(&root->fs_info->transaction_wait);
3671
3672 cur_trans->commit_done = 1;
3673 wake_up(&cur_trans->commit_wait);
3674
3675 btrfs_destroy_delayed_inodes(root);
3676 btrfs_assert_delayed_root_empty(root);
3677
3678 btrfs_destroy_pending_snapshots(cur_trans);
3679
3680 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3681 EXTENT_DIRTY);
3682 btrfs_destroy_pinned_extent(root,
3683 root->fs_info->pinned_extents);
3684
3685 /*
3686 memset(cur_trans, 0, sizeof(*cur_trans));
3687 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3688 */
3689}
3690
3691int btrfs_cleanup_transaction(struct btrfs_root *root)
3692{
3693 struct btrfs_transaction *t;
3694 LIST_HEAD(list);
3695
3696 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3697
3698 spin_lock(&root->fs_info->trans_lock);
3699 list_splice_init(&root->fs_info->trans_list, &list);
3700 root->fs_info->trans_no_join = 1;
3701 spin_unlock(&root->fs_info->trans_lock);
3702
3703 while (!list_empty(&list)) {
3704 t = list_entry(list.next, struct btrfs_transaction, list);
3705 if (!t)
3706 break;
3707
3708 btrfs_destroy_ordered_operations(root);
3709
3710 btrfs_destroy_ordered_extents(root);
3711
3712 btrfs_destroy_delayed_refs(t, root);
3713
3714 btrfs_block_rsv_release(root,
3715 &root->fs_info->trans_block_rsv,
3716 t->dirty_pages.dirty_bytes);
3717
3718 /* FIXME: cleanup wait for commit */
3719 t->in_commit = 1;
3720 t->blocked = 1;
3721 smp_mb();
3722 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3723 wake_up(&root->fs_info->transaction_blocked_wait);
3724
3725 t->blocked = 0;
3726 smp_mb();
3727 if (waitqueue_active(&root->fs_info->transaction_wait))
3728 wake_up(&root->fs_info->transaction_wait);
3729
3730 t->commit_done = 1;
3731 smp_mb();
3732 if (waitqueue_active(&t->commit_wait))
3733 wake_up(&t->commit_wait);
3734
3735 btrfs_destroy_delayed_inodes(root);
3736 btrfs_assert_delayed_root_empty(root);
3737
3738 btrfs_destroy_pending_snapshots(t);
3739
3740 btrfs_destroy_delalloc_inodes(root);
3741
3742 spin_lock(&root->fs_info->trans_lock);
3743 root->fs_info->running_transaction = NULL;
3744 spin_unlock(&root->fs_info->trans_lock);
3745
3746 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3747 EXTENT_DIRTY);
3748
3749 btrfs_destroy_pinned_extent(root,
3750 root->fs_info->pinned_extents);
3751
3752 atomic_set(&t->use_count, 0);
3753 list_del_init(&t->list);
3754 memset(t, 0, sizeof(*t));
3755 kmem_cache_free(btrfs_transaction_cachep, t);
3756 }
3757
3758 spin_lock(&root->fs_info->trans_lock);
3759 root->fs_info->trans_no_join = 0;
3760 spin_unlock(&root->fs_info->trans_lock);
3761 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3762
3763 return 0;
3764}
3765
3766static struct extent_io_ops btree_extent_io_ops = {
3767 .readpage_end_io_hook = btree_readpage_end_io_hook,
3768 .readpage_io_failed_hook = btree_io_failed_hook,
3769 .submit_bio_hook = btree_submit_bio_hook,
3770 /* note we're sharing with inode.c for the merge bio hook */
3771 .merge_bio_hook = btrfs_merge_bio_hook,
3772};
Linux 6.x block layer and io_uring tree(s)