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