?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
spi-topcliff-pch: Fix issue for transmitting over 4KByte
[zen-stable.git] / fs / btrfs / disk-io.c
blob534266fe505f25cf4a89e25036ff7994cb8c28d5
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
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 int read_only);
53 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_root *root);
65
66 /*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71 struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80 };
81
82 /*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87 struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
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
132 static 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
151 void __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
165 void 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 */
187 static 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);
239 out:
240 return em;
241 }
242
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 {
245 return crc32c(seed, data, len);
246 }
247
248 void 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 */
257 static 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 */
325 static 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, GFP_NOFS);
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);
348 out:
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 */
358 static 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
402 static 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;
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 goto out;
428 }
429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
431 BUG_ON(ret);
432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
436 WARN_ON(1);
437 goto err;
438 }
439 if (eb->first_page != page) {
440 WARN_ON(1);
441 goto err;
442 }
443 if (!PageUptodate(page)) {
444 WARN_ON(1);
445 goto err;
446 }
447 csum_tree_block(root, eb, 0);
448 err:
449 free_extent_buffer(eb);
450 out:
451 return 0;
452 }
453
454 static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
456 {
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
459 int ret = 1;
460
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462 BTRFS_FSID_SIZE);
463 while (fs_devices) {
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465 ret = 0;
466 break;
467 }
468 fs_devices = fs_devices->seed;
469 }
470 return ret;
471 }
472
473 #define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
478
479 static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
481 {
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
485 int slot;
486
487 if (nritems == 0)
488 return 0;
489
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
494 return -EIO;
495 }
496
497 /*
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
502 * offset is correct.
503 */
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
511 return -EIO;
512 }
513
514 /*
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
517 * front.
518 */
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
522 return -EIO;
523 }
524
525 /*
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
529 */
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
533 return -EIO;
534 }
535 }
536
537 return 0;
538 }
539
540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541 struct extent_state *state)
542 {
543 struct extent_io_tree *tree;
544 u64 found_start;
545 int found_level;
546 unsigned long len;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
549 int ret = 0;
550
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
553 goto out;
554 if (!page->private)
555 goto out;
556
557 len = page->private >> 2;
558 WARN_ON(len == 0);
559
560 eb = alloc_extent_buffer(tree, start, len, page);
561 if (eb == NULL) {
562 ret = -EIO;
563 goto out;
564 }
565
566 found_start = btrfs_header_bytenr(eb);
567 if (found_start != start) {
568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
569 "%llu %llu\n",
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
572 ret = -EIO;
573 goto err;
574 }
575 if (eb->first_page != page) {
576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
578 WARN_ON(1);
579 ret = -EIO;
580 goto err;
581 }
582 if (check_tree_block_fsid(root, eb)) {
583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584 (unsigned long long)eb->start);
585 ret = -EIO;
586 goto err;
587 }
588 found_level = btrfs_header_level(eb);
589
590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591 eb, found_level);
592
593 ret = csum_tree_block(root, eb, 1);
594 if (ret) {
595 ret = -EIO;
596 goto err;
597 }
598
599 /*
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
602 * return -EIO.
603 */
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606 ret = -EIO;
607 }
608
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
611 err:
612 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
613 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
614 btree_readahead_hook(root, eb, eb->start, ret);
615 }
616
617 free_extent_buffer(eb);
618 out:
619 return ret;
620 }
621
622 static int btree_io_failed_hook(struct bio *failed_bio,
623 struct page *page, u64 start, u64 end,
624 int mirror_num, struct extent_state *state)
625 {
626 struct extent_io_tree *tree;
627 unsigned long len;
628 struct extent_buffer *eb;
629 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
630
631 tree = &BTRFS_I(page->mapping->host)->io_tree;
632 if (page->private == EXTENT_PAGE_PRIVATE)
633 goto out;
634 if (!page->private)
635 goto out;
636
637 len = page->private >> 2;
638 WARN_ON(len == 0);
639
640 eb = alloc_extent_buffer(tree, start, len, page);
641 if (eb == NULL)
642 goto out;
643
644 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
645 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
646 btree_readahead_hook(root, eb, eb->start, -EIO);
647 }
648 free_extent_buffer(eb);
649
650 out:
651 return -EIO; /* we fixed nothing */
652 }
653
654 static void end_workqueue_bio(struct bio *bio, int err)
655 {
656 struct end_io_wq *end_io_wq = bio->bi_private;
657 struct btrfs_fs_info *fs_info;
658
659 fs_info = end_io_wq->info;
660 end_io_wq->error = err;
661 end_io_wq->work.func = end_workqueue_fn;
662 end_io_wq->work.flags = 0;
663
664 if (bio->bi_rw & REQ_WRITE) {
665 if (end_io_wq->metadata == 1)
666 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
667 &end_io_wq->work);
668 else if (end_io_wq->metadata == 2)
669 btrfs_queue_worker(&fs_info->endio_freespace_worker,
670 &end_io_wq->work);
671 else
672 btrfs_queue_worker(&fs_info->endio_write_workers,
673 &end_io_wq->work);
674 } else {
675 if (end_io_wq->metadata)
676 btrfs_queue_worker(&fs_info->endio_meta_workers,
677 &end_io_wq->work);
678 else
679 btrfs_queue_worker(&fs_info->endio_workers,
680 &end_io_wq->work);
681 }
682 }
683
684 /*
685 * For the metadata arg you want
686 *
687 * 0 - if data
688 * 1 - if normal metadta
689 * 2 - if writing to the free space cache area
690 */
691 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692 int metadata)
693 {
694 struct end_io_wq *end_io_wq;
695 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
696 if (!end_io_wq)
697 return -ENOMEM;
698
699 end_io_wq->private = bio->bi_private;
700 end_io_wq->end_io = bio->bi_end_io;
701 end_io_wq->info = info;
702 end_io_wq->error = 0;
703 end_io_wq->bio = bio;
704 end_io_wq->metadata = metadata;
705
706 bio->bi_private = end_io_wq;
707 bio->bi_end_io = end_workqueue_bio;
708 return 0;
709 }
710
711 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
712 {
713 unsigned long limit = min_t(unsigned long,
714 info->workers.max_workers,
715 info->fs_devices->open_devices);
716 return 256 * limit;
717 }
718
719 static void run_one_async_start(struct btrfs_work *work)
720 {
721 struct async_submit_bio *async;
722
723 async = container_of(work, struct async_submit_bio, work);
724 async->submit_bio_start(async->inode, async->rw, async->bio,
725 async->mirror_num, async->bio_flags,
726 async->bio_offset);
727 }
728
729 static void run_one_async_done(struct btrfs_work *work)
730 {
731 struct btrfs_fs_info *fs_info;
732 struct async_submit_bio *async;
733 int limit;
734
735 async = container_of(work, struct async_submit_bio, work);
736 fs_info = BTRFS_I(async->inode)->root->fs_info;
737
738 limit = btrfs_async_submit_limit(fs_info);
739 limit = limit * 2 / 3;
740
741 atomic_dec(&fs_info->nr_async_submits);
742
743 if (atomic_read(&fs_info->nr_async_submits) < limit &&
744 waitqueue_active(&fs_info->async_submit_wait))
745 wake_up(&fs_info->async_submit_wait);
746
747 async->submit_bio_done(async->inode, async->rw, async->bio,
748 async->mirror_num, async->bio_flags,
749 async->bio_offset);
750 }
751
752 static void run_one_async_free(struct btrfs_work *work)
753 {
754 struct async_submit_bio *async;
755
756 async = container_of(work, struct async_submit_bio, work);
757 kfree(async);
758 }
759
760 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
761 int rw, struct bio *bio, int mirror_num,
762 unsigned long bio_flags,
763 u64 bio_offset,
764 extent_submit_bio_hook_t *submit_bio_start,
765 extent_submit_bio_hook_t *submit_bio_done)
766 {
767 struct async_submit_bio *async;
768
769 async = kmalloc(sizeof(*async), GFP_NOFS);
770 if (!async)
771 return -ENOMEM;
772
773 async->inode = inode;
774 async->rw = rw;
775 async->bio = bio;
776 async->mirror_num = mirror_num;
777 async->submit_bio_start = submit_bio_start;
778 async->submit_bio_done = submit_bio_done;
779
780 async->work.func = run_one_async_start;
781 async->work.ordered_func = run_one_async_done;
782 async->work.ordered_free = run_one_async_free;
783
784 async->work.flags = 0;
785 async->bio_flags = bio_flags;
786 async->bio_offset = bio_offset;
787
788 atomic_inc(&fs_info->nr_async_submits);
789
790 if (rw & REQ_SYNC)
791 btrfs_set_work_high_prio(&async->work);
792
793 btrfs_queue_worker(&fs_info->workers, &async->work);
794
795 while (atomic_read(&fs_info->async_submit_draining) &&
796 atomic_read(&fs_info->nr_async_submits)) {
797 wait_event(fs_info->async_submit_wait,
798 (atomic_read(&fs_info->nr_async_submits) == 0));
799 }
800
801 return 0;
802 }
803
804 static int btree_csum_one_bio(struct bio *bio)
805 {
806 struct bio_vec *bvec = bio->bi_io_vec;
807 int bio_index = 0;
808 struct btrfs_root *root;
809
810 WARN_ON(bio->bi_vcnt <= 0);
811 while (bio_index < bio->bi_vcnt) {
812 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813 csum_dirty_buffer(root, bvec->bv_page);
814 bio_index++;
815 bvec++;
816 }
817 return 0;
818 }
819
820 static int __btree_submit_bio_start(struct inode *inode, int rw,
821 struct bio *bio, int mirror_num,
822 unsigned long bio_flags,
823 u64 bio_offset)
824 {
825 /*
826 * when we're called for a write, we're already in the async
827 * submission context. Just jump into btrfs_map_bio
828 */
829 btree_csum_one_bio(bio);
830 return 0;
831 }
832
833 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834 int mirror_num, unsigned long bio_flags,
835 u64 bio_offset)
836 {
837 /*
838 * when we're called for a write, we're already in the async
839 * submission context. Just jump into btrfs_map_bio
840 */
841 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
842 }
843
844 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
845 int mirror_num, unsigned long bio_flags,
846 u64 bio_offset)
847 {
848 int ret;
849
850 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
851 bio, 1);
852 BUG_ON(ret);
853
854 if (!(rw & REQ_WRITE)) {
855 /*
856 * called for a read, do the setup so that checksum validation
857 * can happen in the async kernel threads
858 */
859 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
860 mirror_num, 0);
861 }
862
863 /*
864 * kthread helpers are used to submit writes so that checksumming
865 * can happen in parallel across all CPUs
866 */
867 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
868 inode, rw, bio, mirror_num, 0,
869 bio_offset,
870 __btree_submit_bio_start,
871 __btree_submit_bio_done);
872 }
873
874 #ifdef CONFIG_MIGRATION
875 static int btree_migratepage(struct address_space *mapping,
876 struct page *newpage, struct page *page,
877 enum migrate_mode mode)
878 {
879 /*
880 * we can't safely write a btree page from here,
881 * we haven't done the locking hook
882 */
883 if (PageDirty(page))
884 return -EAGAIN;
885 /*
886 * Buffers may be managed in a filesystem specific way.
887 * We must have no buffers or drop them.
888 */
889 if (page_has_private(page) &&
890 !try_to_release_page(page, GFP_KERNEL))
891 return -EAGAIN;
892 return migrate_page(mapping, newpage, page, mode);
893 }
894 #endif
895
896 static int btree_writepage(struct page *page, struct writeback_control *wbc)
897 {
898 struct extent_io_tree *tree;
899 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
900 struct extent_buffer *eb;
901 int was_dirty;
902
903 tree = &BTRFS_I(page->mapping->host)->io_tree;
904 if (!(current->flags & PF_MEMALLOC)) {
905 return extent_write_full_page(tree, page,
906 btree_get_extent, wbc);
907 }
908
909 redirty_page_for_writepage(wbc, page);
910 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
911 WARN_ON(!eb);
912
913 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
914 if (!was_dirty) {
915 spin_lock(&root->fs_info->delalloc_lock);
916 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
917 spin_unlock(&root->fs_info->delalloc_lock);
918 }
919 free_extent_buffer(eb);
920
921 unlock_page(page);
922 return 0;
923 }
924
925 static int btree_writepages(struct address_space *mapping,
926 struct writeback_control *wbc)
927 {
928 struct extent_io_tree *tree;
929 tree = &BTRFS_I(mapping->host)->io_tree;
930 if (wbc->sync_mode == WB_SYNC_NONE) {
931 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
932 u64 num_dirty;
933 unsigned long thresh = 32 * 1024 * 1024;
934
935 if (wbc->for_kupdate)
936 return 0;
937
938 /* this is a bit racy, but that's ok */
939 num_dirty = root->fs_info->dirty_metadata_bytes;
940 if (num_dirty < thresh)
941 return 0;
942 }
943 return extent_writepages(tree, mapping, btree_get_extent, wbc);
944 }
945
946 static int btree_readpage(struct file *file, struct page *page)
947 {
948 struct extent_io_tree *tree;
949 tree = &BTRFS_I(page->mapping->host)->io_tree;
950 return extent_read_full_page(tree, page, btree_get_extent, 0);
951 }
952
953 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
954 {
955 struct extent_io_tree *tree;
956 struct extent_map_tree *map;
957 int ret;
958
959 if (PageWriteback(page) || PageDirty(page))
960 return 0;
961
962 tree = &BTRFS_I(page->mapping->host)->io_tree;
963 map = &BTRFS_I(page->mapping->host)->extent_tree;
964
965 /*
966 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
967 * slab allocation from alloc_extent_state down the callchain where
968 * it'd hit a BUG_ON as those flags are not allowed.
969 */
970 gfp_flags &= ~GFP_SLAB_BUG_MASK;
971
972 ret = try_release_extent_state(map, tree, page, gfp_flags);
973 if (!ret)
974 return 0;
975
976 ret = try_release_extent_buffer(tree, page);
977 if (ret == 1) {
978 ClearPagePrivate(page);
979 set_page_private(page, 0);
980 page_cache_release(page);
981 }
982
983 return ret;
984 }
985
986 static void btree_invalidatepage(struct page *page, unsigned long offset)
987 {
988 struct extent_io_tree *tree;
989 tree = &BTRFS_I(page->mapping->host)->io_tree;
990 extent_invalidatepage(tree, page, offset);
991 btree_releasepage(page, GFP_NOFS);
992 if (PagePrivate(page)) {
993 printk(KERN_WARNING "btrfs warning page private not zero "
994 "on page %llu\n", (unsigned long long)page_offset(page));
995 ClearPagePrivate(page);
996 set_page_private(page, 0);
997 page_cache_release(page);
998 }
999 }
1000
1001 static const struct address_space_operations btree_aops = {
1002 .readpage = btree_readpage,
1003 .writepage = btree_writepage,
1004 .writepages = btree_writepages,
1005 .releasepage = btree_releasepage,
1006 .invalidatepage = btree_invalidatepage,
1007 #ifdef CONFIG_MIGRATION
1008 .migratepage = btree_migratepage,
1009 #endif
1010 };
1011
1012 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1013 u64 parent_transid)
1014 {
1015 struct extent_buffer *buf = NULL;
1016 struct inode *btree_inode = root->fs_info->btree_inode;
1017 int ret = 0;
1018
1019 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1020 if (!buf)
1021 return 0;
1022 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1023 buf, 0, WAIT_NONE, btree_get_extent, 0);
1024 free_extent_buffer(buf);
1025 return ret;
1026 }
1027
1028 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1029 int mirror_num, struct extent_buffer **eb)
1030 {
1031 struct extent_buffer *buf = NULL;
1032 struct inode *btree_inode = root->fs_info->btree_inode;
1033 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1034 int ret;
1035
1036 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1037 if (!buf)
1038 return 0;
1039
1040 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1041
1042 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1043 btree_get_extent, mirror_num);
1044 if (ret) {
1045 free_extent_buffer(buf);
1046 return ret;
1047 }
1048
1049 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1050 free_extent_buffer(buf);
1051 return -EIO;
1052 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1053 *eb = buf;
1054 } else {
1055 free_extent_buffer(buf);
1056 }
1057 return 0;
1058 }
1059
1060 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1061 u64 bytenr, u32 blocksize)
1062 {
1063 struct inode *btree_inode = root->fs_info->btree_inode;
1064 struct extent_buffer *eb;
1065 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1066 bytenr, blocksize);
1067 return eb;
1068 }
1069
1070 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1071 u64 bytenr, u32 blocksize)
1072 {
1073 struct inode *btree_inode = root->fs_info->btree_inode;
1074 struct extent_buffer *eb;
1075
1076 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1077 bytenr, blocksize, NULL);
1078 return eb;
1079 }
1080
1081
1082 int btrfs_write_tree_block(struct extent_buffer *buf)
1083 {
1084 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1085 buf->start + buf->len - 1);
1086 }
1087
1088 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1089 {
1090 return filemap_fdatawait_range(buf->first_page->mapping,
1091 buf->start, buf->start + buf->len - 1);
1092 }
1093
1094 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1095 u32 blocksize, u64 parent_transid)
1096 {
1097 struct extent_buffer *buf = NULL;
1098 int ret;
1099
1100 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1101 if (!buf)
1102 return NULL;
1103
1104 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1105
1106 if (ret == 0)
1107 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1108 return buf;
1109
1110 }
1111
1112 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1113 struct extent_buffer *buf)
1114 {
1115 struct inode *btree_inode = root->fs_info->btree_inode;
1116 if (btrfs_header_generation(buf) ==
1117 root->fs_info->running_transaction->transid) {
1118 btrfs_assert_tree_locked(buf);
1119
1120 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1121 spin_lock(&root->fs_info->delalloc_lock);
1122 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1123 root->fs_info->dirty_metadata_bytes -= buf->len;
1124 else
1125 WARN_ON(1);
1126 spin_unlock(&root->fs_info->delalloc_lock);
1127 }
1128
1129 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1130 btrfs_set_lock_blocking(buf);
1131 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1132 buf);
1133 }
1134 return 0;
1135 }
1136
1137 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1138 u32 stripesize, struct btrfs_root *root,
1139 struct btrfs_fs_info *fs_info,
1140 u64 objectid)
1141 {
1142 root->node = NULL;
1143 root->commit_root = NULL;
1144 root->sectorsize = sectorsize;
1145 root->nodesize = nodesize;
1146 root->leafsize = leafsize;
1147 root->stripesize = stripesize;
1148 root->ref_cows = 0;
1149 root->track_dirty = 0;
1150 root->in_radix = 0;
1151 root->orphan_item_inserted = 0;
1152 root->orphan_cleanup_state = 0;
1153
1154 root->objectid = objectid;
1155 root->last_trans = 0;
1156 root->highest_objectid = 0;
1157 root->name = NULL;
1158 root->inode_tree = RB_ROOT;
1159 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1160 root->block_rsv = NULL;
1161 root->orphan_block_rsv = NULL;
1162
1163 INIT_LIST_HEAD(&root->dirty_list);
1164 INIT_LIST_HEAD(&root->orphan_list);
1165 INIT_LIST_HEAD(&root->root_list);
1166 spin_lock_init(&root->orphan_lock);
1167 spin_lock_init(&root->inode_lock);
1168 spin_lock_init(&root->accounting_lock);
1169 mutex_init(&root->objectid_mutex);
1170 mutex_init(&root->log_mutex);
1171 init_waitqueue_head(&root->log_writer_wait);
1172 init_waitqueue_head(&root->log_commit_wait[0]);
1173 init_waitqueue_head(&root->log_commit_wait[1]);
1174 atomic_set(&root->log_commit[0], 0);
1175 atomic_set(&root->log_commit[1], 0);
1176 atomic_set(&root->log_writers, 0);
1177 root->log_batch = 0;
1178 root->log_transid = 0;
1179 root->last_log_commit = 0;
1180 extent_io_tree_init(&root->dirty_log_pages,
1181 fs_info->btree_inode->i_mapping);
1182
1183 memset(&root->root_key, 0, sizeof(root->root_key));
1184 memset(&root->root_item, 0, sizeof(root->root_item));
1185 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1186 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1187 root->defrag_trans_start = fs_info->generation;
1188 init_completion(&root->kobj_unregister);
1189 root->defrag_running = 0;
1190 root->root_key.objectid = objectid;
1191 root->anon_dev = 0;
1192 return 0;
1193 }
1194
1195 static int find_and_setup_root(struct btrfs_root *tree_root,
1196 struct btrfs_fs_info *fs_info,
1197 u64 objectid,
1198 struct btrfs_root *root)
1199 {
1200 int ret;
1201 u32 blocksize;
1202 u64 generation;
1203
1204 __setup_root(tree_root->nodesize, tree_root->leafsize,
1205 tree_root->sectorsize, tree_root->stripesize,
1206 root, fs_info, objectid);
1207 ret = btrfs_find_last_root(tree_root, objectid,
1208 &root->root_item, &root->root_key);
1209 if (ret > 0)
1210 return -ENOENT;
1211 BUG_ON(ret);
1212
1213 generation = btrfs_root_generation(&root->root_item);
1214 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1215 root->commit_root = NULL;
1216 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1217 blocksize, generation);
1218 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1219 free_extent_buffer(root->node);
1220 root->node = NULL;
1221 return -EIO;
1222 }
1223 root->commit_root = btrfs_root_node(root);
1224 return 0;
1225 }
1226
1227 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1228 {
1229 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1230 if (root)
1231 root->fs_info = fs_info;
1232 return root;
1233 }
1234
1235 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1236 struct btrfs_fs_info *fs_info)
1237 {
1238 struct btrfs_root *root;
1239 struct btrfs_root *tree_root = fs_info->tree_root;
1240 struct extent_buffer *leaf;
1241
1242 root = btrfs_alloc_root(fs_info);
1243 if (!root)
1244 return ERR_PTR(-ENOMEM);
1245
1246 __setup_root(tree_root->nodesize, tree_root->leafsize,
1247 tree_root->sectorsize, tree_root->stripesize,
1248 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1249
1250 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1251 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1252 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1253 /*
1254 * log trees do not get reference counted because they go away
1255 * before a real commit is actually done. They do store pointers
1256 * to file data extents, and those reference counts still get
1257 * updated (along with back refs to the log tree).
1258 */
1259 root->ref_cows = 0;
1260
1261 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1262 BTRFS_TREE_LOG_OBJECTID, NULL,
1263 0, 0, 0, 0);
1264 if (IS_ERR(leaf)) {
1265 kfree(root);
1266 return ERR_CAST(leaf);
1267 }
1268
1269 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1270 btrfs_set_header_bytenr(leaf, leaf->start);
1271 btrfs_set_header_generation(leaf, trans->transid);
1272 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1273 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1274 root->node = leaf;
1275
1276 write_extent_buffer(root->node, root->fs_info->fsid,
1277 (unsigned long)btrfs_header_fsid(root->node),
1278 BTRFS_FSID_SIZE);
1279 btrfs_mark_buffer_dirty(root->node);
1280 btrfs_tree_unlock(root->node);
1281 return root;
1282 }
1283
1284 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1285 struct btrfs_fs_info *fs_info)
1286 {
1287 struct btrfs_root *log_root;
1288
1289 log_root = alloc_log_tree(trans, fs_info);
1290 if (IS_ERR(log_root))
1291 return PTR_ERR(log_root);
1292 WARN_ON(fs_info->log_root_tree);
1293 fs_info->log_root_tree = log_root;
1294 return 0;
1295 }
1296
1297 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1298 struct btrfs_root *root)
1299 {
1300 struct btrfs_root *log_root;
1301 struct btrfs_inode_item *inode_item;
1302
1303 log_root = alloc_log_tree(trans, root->fs_info);
1304 if (IS_ERR(log_root))
1305 return PTR_ERR(log_root);
1306
1307 log_root->last_trans = trans->transid;
1308 log_root->root_key.offset = root->root_key.objectid;
1309
1310 inode_item = &log_root->root_item.inode;
1311 inode_item->generation = cpu_to_le64(1);
1312 inode_item->size = cpu_to_le64(3);
1313 inode_item->nlink = cpu_to_le32(1);
1314 inode_item->nbytes = cpu_to_le64(root->leafsize);
1315 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1316
1317 btrfs_set_root_node(&log_root->root_item, log_root->node);
1318
1319 WARN_ON(root->log_root);
1320 root->log_root = log_root;
1321 root->log_transid = 0;
1322 root->last_log_commit = 0;
1323 return 0;
1324 }
1325
1326 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1327 struct btrfs_key *location)
1328 {
1329 struct btrfs_root *root;
1330 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1331 struct btrfs_path *path;
1332 struct extent_buffer *l;
1333 u64 generation;
1334 u32 blocksize;
1335 int ret = 0;
1336
1337 root = btrfs_alloc_root(fs_info);
1338 if (!root)
1339 return ERR_PTR(-ENOMEM);
1340 if (location->offset == (u64)-1) {
1341 ret = find_and_setup_root(tree_root, fs_info,
1342 location->objectid, root);
1343 if (ret) {
1344 kfree(root);
1345 return ERR_PTR(ret);
1346 }
1347 goto out;
1348 }
1349
1350 __setup_root(tree_root->nodesize, tree_root->leafsize,
1351 tree_root->sectorsize, tree_root->stripesize,
1352 root, fs_info, location->objectid);
1353
1354 path = btrfs_alloc_path();
1355 if (!path) {
1356 kfree(root);
1357 return ERR_PTR(-ENOMEM);
1358 }
1359 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1360 if (ret == 0) {
1361 l = path->nodes[0];
1362 read_extent_buffer(l, &root->root_item,
1363 btrfs_item_ptr_offset(l, path->slots[0]),
1364 sizeof(root->root_item));
1365 memcpy(&root->root_key, location, sizeof(*location));
1366 }
1367 btrfs_free_path(path);
1368 if (ret) {
1369 kfree(root);
1370 if (ret > 0)
1371 ret = -ENOENT;
1372 return ERR_PTR(ret);
1373 }
1374
1375 generation = btrfs_root_generation(&root->root_item);
1376 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1377 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1378 blocksize, generation);
1379 root->commit_root = btrfs_root_node(root);
1380 BUG_ON(!root->node);
1381 out:
1382 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1383 root->ref_cows = 1;
1384 btrfs_check_and_init_root_item(&root->root_item);
1385 }
1386
1387 return root;
1388 }
1389
1390 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1391 struct btrfs_key *location)
1392 {
1393 struct btrfs_root *root;
1394 int ret;
1395
1396 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1397 return fs_info->tree_root;
1398 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1399 return fs_info->extent_root;
1400 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1401 return fs_info->chunk_root;
1402 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1403 return fs_info->dev_root;
1404 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1405 return fs_info->csum_root;
1406 again:
1407 spin_lock(&fs_info->fs_roots_radix_lock);
1408 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1409 (unsigned long)location->objectid);
1410 spin_unlock(&fs_info->fs_roots_radix_lock);
1411 if (root)
1412 return root;
1413
1414 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1415 if (IS_ERR(root))
1416 return root;
1417
1418 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1419 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1420 GFP_NOFS);
1421 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1422 ret = -ENOMEM;
1423 goto fail;
1424 }
1425
1426 btrfs_init_free_ino_ctl(root);
1427 mutex_init(&root->fs_commit_mutex);
1428 spin_lock_init(&root->cache_lock);
1429 init_waitqueue_head(&root->cache_wait);
1430
1431 ret = get_anon_bdev(&root->anon_dev);
1432 if (ret)
1433 goto fail;
1434
1435 if (btrfs_root_refs(&root->root_item) == 0) {
1436 ret = -ENOENT;
1437 goto fail;
1438 }
1439
1440 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1441 if (ret < 0)
1442 goto fail;
1443 if (ret == 0)
1444 root->orphan_item_inserted = 1;
1445
1446 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1447 if (ret)
1448 goto fail;
1449
1450 spin_lock(&fs_info->fs_roots_radix_lock);
1451 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1452 (unsigned long)root->root_key.objectid,
1453 root);
1454 if (ret == 0)
1455 root->in_radix = 1;
1456
1457 spin_unlock(&fs_info->fs_roots_radix_lock);
1458 radix_tree_preload_end();
1459 if (ret) {
1460 if (ret == -EEXIST) {
1461 free_fs_root(root);
1462 goto again;
1463 }
1464 goto fail;
1465 }
1466
1467 ret = btrfs_find_dead_roots(fs_info->tree_root,
1468 root->root_key.objectid);
1469 WARN_ON(ret);
1470 return root;
1471 fail:
1472 free_fs_root(root);
1473 return ERR_PTR(ret);
1474 }
1475
1476 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1477 {
1478 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1479 int ret = 0;
1480 struct btrfs_device *device;
1481 struct backing_dev_info *bdi;
1482
1483 rcu_read_lock();
1484 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1485 if (!device->bdev)
1486 continue;
1487 bdi = blk_get_backing_dev_info(device->bdev);
1488 if (bdi && bdi_congested(bdi, bdi_bits)) {
1489 ret = 1;
1490 break;
1491 }
1492 }
1493 rcu_read_unlock();
1494 return ret;
1495 }
1496
1497 /*
1498 * If this fails, caller must call bdi_destroy() to get rid of the
1499 * bdi again.
1500 */
1501 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1502 {
1503 int err;
1504
1505 bdi->capabilities = BDI_CAP_MAP_COPY;
1506 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1507 if (err)
1508 return err;
1509
1510 bdi->ra_pages = default_backing_dev_info.ra_pages;
1511 bdi->congested_fn = btrfs_congested_fn;
1512 bdi->congested_data = info;
1513 return 0;
1514 }
1515
1516 static int bio_ready_for_csum(struct bio *bio)
1517 {
1518 u64 length = 0;
1519 u64 buf_len = 0;
1520 u64 start = 0;
1521 struct page *page;
1522 struct extent_io_tree *io_tree = NULL;
1523 struct bio_vec *bvec;
1524 int i;
1525 int ret;
1526
1527 bio_for_each_segment(bvec, bio, i) {
1528 page = bvec->bv_page;
1529 if (page->private == EXTENT_PAGE_PRIVATE) {
1530 length += bvec->bv_len;
1531 continue;
1532 }
1533 if (!page->private) {
1534 length += bvec->bv_len;
1535 continue;
1536 }
1537 length = bvec->bv_len;
1538 buf_len = page->private >> 2;
1539 start = page_offset(page) + bvec->bv_offset;
1540 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1541 }
1542 /* are we fully contained in this bio? */
1543 if (buf_len <= length)
1544 return 1;
1545
1546 ret = extent_range_uptodate(io_tree, start + length,
1547 start + buf_len - 1);
1548 return ret;
1549 }
1550
1551 /*
1552 * called by the kthread helper functions to finally call the bio end_io
1553 * functions. This is where read checksum verification actually happens
1554 */
1555 static void end_workqueue_fn(struct btrfs_work *work)
1556 {
1557 struct bio *bio;
1558 struct end_io_wq *end_io_wq;
1559 struct btrfs_fs_info *fs_info;
1560 int error;
1561
1562 end_io_wq = container_of(work, struct end_io_wq, work);
1563 bio = end_io_wq->bio;
1564 fs_info = end_io_wq->info;
1565
1566 /* metadata bio reads are special because the whole tree block must
1567 * be checksummed at once. This makes sure the entire block is in
1568 * ram and up to date before trying to verify things. For
1569 * blocksize <= pagesize, it is basically a noop
1570 */
1571 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1572 !bio_ready_for_csum(bio)) {
1573 btrfs_queue_worker(&fs_info->endio_meta_workers,
1574 &end_io_wq->work);
1575 return;
1576 }
1577 error = end_io_wq->error;
1578 bio->bi_private = end_io_wq->private;
1579 bio->bi_end_io = end_io_wq->end_io;
1580 kfree(end_io_wq);
1581 bio_endio(bio, error);
1582 }
1583
1584 static int cleaner_kthread(void *arg)
1585 {
1586 struct btrfs_root *root = arg;
1587
1588 do {
1589 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1590
1591 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1592 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1593 btrfs_run_delayed_iputs(root);
1594 btrfs_clean_old_snapshots(root);
1595 mutex_unlock(&root->fs_info->cleaner_mutex);
1596 btrfs_run_defrag_inodes(root->fs_info);
1597 }
1598
1599 if (!try_to_freeze()) {
1600 set_current_state(TASK_INTERRUPTIBLE);
1601 if (!kthread_should_stop())
1602 schedule();
1603 __set_current_state(TASK_RUNNING);
1604 }
1605 } while (!kthread_should_stop());
1606 return 0;
1607 }
1608
1609 static int transaction_kthread(void *arg)
1610 {
1611 struct btrfs_root *root = arg;
1612 struct btrfs_trans_handle *trans;
1613 struct btrfs_transaction *cur;
1614 u64 transid;
1615 unsigned long now;
1616 unsigned long delay;
1617 int ret;
1618
1619 do {
1620 delay = HZ * 30;
1621 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1622 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1623
1624 spin_lock(&root->fs_info->trans_lock);
1625 cur = root->fs_info->running_transaction;
1626 if (!cur) {
1627 spin_unlock(&root->fs_info->trans_lock);
1628 goto sleep;
1629 }
1630
1631 now = get_seconds();
1632 if (!cur->blocked &&
1633 (now < cur->start_time || now - cur->start_time < 30)) {
1634 spin_unlock(&root->fs_info->trans_lock);
1635 delay = HZ * 5;
1636 goto sleep;
1637 }
1638 transid = cur->transid;
1639 spin_unlock(&root->fs_info->trans_lock);
1640
1641 trans = btrfs_join_transaction(root);
1642 BUG_ON(IS_ERR(trans));
1643 if (transid == trans->transid) {
1644 ret = btrfs_commit_transaction(trans, root);
1645 BUG_ON(ret);
1646 } else {
1647 btrfs_end_transaction(trans, root);
1648 }
1649 sleep:
1650 wake_up_process(root->fs_info->cleaner_kthread);
1651 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1652
1653 if (!try_to_freeze()) {
1654 set_current_state(TASK_INTERRUPTIBLE);
1655 if (!kthread_should_stop() &&
1656 !btrfs_transaction_blocked(root->fs_info))
1657 schedule_timeout(delay);
1658 __set_current_state(TASK_RUNNING);
1659 }
1660 } while (!kthread_should_stop());
1661 return 0;
1662 }
1663
1664 /*
1665 * this will find the highest generation in the array of
1666 * root backups. The index of the highest array is returned,
1667 * or -1 if we can't find anything.
1668 *
1669 * We check to make sure the array is valid by comparing the
1670 * generation of the latest root in the array with the generation
1671 * in the super block. If they don't match we pitch it.
1672 */
1673 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1674 {
1675 u64 cur;
1676 int newest_index = -1;
1677 struct btrfs_root_backup *root_backup;
1678 int i;
1679
1680 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1681 root_backup = info->super_copy->super_roots + i;
1682 cur = btrfs_backup_tree_root_gen(root_backup);
1683 if (cur == newest_gen)
1684 newest_index = i;
1685 }
1686
1687 /* check to see if we actually wrapped around */
1688 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1689 root_backup = info->super_copy->super_roots;
1690 cur = btrfs_backup_tree_root_gen(root_backup);
1691 if (cur == newest_gen)
1692 newest_index = 0;
1693 }
1694 return newest_index;
1695 }
1696
1697
1698 /*
1699 * find the oldest backup so we know where to store new entries
1700 * in the backup array. This will set the backup_root_index
1701 * field in the fs_info struct
1702 */
1703 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1704 u64 newest_gen)
1705 {
1706 int newest_index = -1;
1707
1708 newest_index = find_newest_super_backup(info, newest_gen);
1709 /* if there was garbage in there, just move along */
1710 if (newest_index == -1) {
1711 info->backup_root_index = 0;
1712 } else {
1713 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1714 }
1715 }
1716
1717 /*
1718 * copy all the root pointers into the super backup array.
1719 * this will bump the backup pointer by one when it is
1720 * done
1721 */
1722 static void backup_super_roots(struct btrfs_fs_info *info)
1723 {
1724 int next_backup;
1725 struct btrfs_root_backup *root_backup;
1726 int last_backup;
1727
1728 next_backup = info->backup_root_index;
1729 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1730 BTRFS_NUM_BACKUP_ROOTS;
1731
1732 /*
1733 * just overwrite the last backup if we're at the same generation
1734 * this happens only at umount
1735 */
1736 root_backup = info->super_for_commit->super_roots + last_backup;
1737 if (btrfs_backup_tree_root_gen(root_backup) ==
1738 btrfs_header_generation(info->tree_root->node))
1739 next_backup = last_backup;
1740
1741 root_backup = info->super_for_commit->super_roots + next_backup;
1742
1743 /*
1744 * make sure all of our padding and empty slots get zero filled
1745 * regardless of which ones we use today
1746 */
1747 memset(root_backup, 0, sizeof(*root_backup));
1748
1749 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1750
1751 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1752 btrfs_set_backup_tree_root_gen(root_backup,
1753 btrfs_header_generation(info->tree_root->node));
1754
1755 btrfs_set_backup_tree_root_level(root_backup,
1756 btrfs_header_level(info->tree_root->node));
1757
1758 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1759 btrfs_set_backup_chunk_root_gen(root_backup,
1760 btrfs_header_generation(info->chunk_root->node));
1761 btrfs_set_backup_chunk_root_level(root_backup,
1762 btrfs_header_level(info->chunk_root->node));
1763
1764 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1765 btrfs_set_backup_extent_root_gen(root_backup,
1766 btrfs_header_generation(info->extent_root->node));
1767 btrfs_set_backup_extent_root_level(root_backup,
1768 btrfs_header_level(info->extent_root->node));
1769
1770 /*
1771 * we might commit during log recovery, which happens before we set
1772 * the fs_root. Make sure it is valid before we fill it in.
1773 */
1774 if (info->fs_root && info->fs_root->node) {
1775 btrfs_set_backup_fs_root(root_backup,
1776 info->fs_root->node->start);
1777 btrfs_set_backup_fs_root_gen(root_backup,
1778 btrfs_header_generation(info->fs_root->node));
1779 btrfs_set_backup_fs_root_level(root_backup,
1780 btrfs_header_level(info->fs_root->node));
1781 }
1782
1783 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1784 btrfs_set_backup_dev_root_gen(root_backup,
1785 btrfs_header_generation(info->dev_root->node));
1786 btrfs_set_backup_dev_root_level(root_backup,
1787 btrfs_header_level(info->dev_root->node));
1788
1789 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1790 btrfs_set_backup_csum_root_gen(root_backup,
1791 btrfs_header_generation(info->csum_root->node));
1792 btrfs_set_backup_csum_root_level(root_backup,
1793 btrfs_header_level(info->csum_root->node));
1794
1795 btrfs_set_backup_total_bytes(root_backup,
1796 btrfs_super_total_bytes(info->super_copy));
1797 btrfs_set_backup_bytes_used(root_backup,
1798 btrfs_super_bytes_used(info->super_copy));
1799 btrfs_set_backup_num_devices(root_backup,
1800 btrfs_super_num_devices(info->super_copy));
1801
1802 /*
1803 * if we don't copy this out to the super_copy, it won't get remembered
1804 * for the next commit
1805 */
1806 memcpy(&info->super_copy->super_roots,
1807 &info->super_for_commit->super_roots,
1808 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1809 }
1810
1811 /*
1812 * this copies info out of the root backup array and back into
1813 * the in-memory super block. It is meant to help iterate through
1814 * the array, so you send it the number of backups you've already
1815 * tried and the last backup index you used.
1816 *
1817 * this returns -1 when it has tried all the backups
1818 */
1819 static noinline int next_root_backup(struct btrfs_fs_info *info,
1820 struct btrfs_super_block *super,
1821 int *num_backups_tried, int *backup_index)
1822 {
1823 struct btrfs_root_backup *root_backup;
1824 int newest = *backup_index;
1825
1826 if (*num_backups_tried == 0) {
1827 u64 gen = btrfs_super_generation(super);
1828
1829 newest = find_newest_super_backup(info, gen);
1830 if (newest == -1)
1831 return -1;
1832
1833 *backup_index = newest;
1834 *num_backups_tried = 1;
1835 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1836 /* we've tried all the backups, all done */
1837 return -1;
1838 } else {
1839 /* jump to the next oldest backup */
1840 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1841 BTRFS_NUM_BACKUP_ROOTS;
1842 *backup_index = newest;
1843 *num_backups_tried += 1;
1844 }
1845 root_backup = super->super_roots + newest;
1846
1847 btrfs_set_super_generation(super,
1848 btrfs_backup_tree_root_gen(root_backup));
1849 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1850 btrfs_set_super_root_level(super,
1851 btrfs_backup_tree_root_level(root_backup));
1852 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1853
1854 /*
1855 * fixme: the total bytes and num_devices need to match or we should
1856 * need a fsck
1857 */
1858 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1859 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1860 return 0;
1861 }
1862
1863 /* helper to cleanup tree roots */
1864 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1865 {
1866 free_extent_buffer(info->tree_root->node);
1867 free_extent_buffer(info->tree_root->commit_root);
1868 free_extent_buffer(info->dev_root->node);
1869 free_extent_buffer(info->dev_root->commit_root);
1870 free_extent_buffer(info->extent_root->node);
1871 free_extent_buffer(info->extent_root->commit_root);
1872 free_extent_buffer(info->csum_root->node);
1873 free_extent_buffer(info->csum_root->commit_root);
1874
1875 info->tree_root->node = NULL;
1876 info->tree_root->commit_root = NULL;
1877 info->dev_root->node = NULL;
1878 info->dev_root->commit_root = NULL;
1879 info->extent_root->node = NULL;
1880 info->extent_root->commit_root = NULL;
1881 info->csum_root->node = NULL;
1882 info->csum_root->commit_root = NULL;
1883
1884 if (chunk_root) {
1885 free_extent_buffer(info->chunk_root->node);
1886 free_extent_buffer(info->chunk_root->commit_root);
1887 info->chunk_root->node = NULL;
1888 info->chunk_root->commit_root = NULL;
1889 }
1890 }
1891
1892
1893 int open_ctree(struct super_block *sb,
1894 struct btrfs_fs_devices *fs_devices,
1895 char *options)
1896 {
1897 u32 sectorsize;
1898 u32 nodesize;
1899 u32 leafsize;
1900 u32 blocksize;
1901 u32 stripesize;
1902 u64 generation;
1903 u64 features;
1904 struct btrfs_key location;
1905 struct buffer_head *bh;
1906 struct btrfs_super_block *disk_super;
1907 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1908 struct btrfs_root *tree_root;
1909 struct btrfs_root *extent_root;
1910 struct btrfs_root *csum_root;
1911 struct btrfs_root *chunk_root;
1912 struct btrfs_root *dev_root;
1913 struct btrfs_root *log_tree_root;
1914 int ret;
1915 int err = -EINVAL;
1916 int num_backups_tried = 0;
1917 int backup_index = 0;
1918
1919 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1920 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1921 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1922 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1923 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1924
1925 if (!tree_root || !extent_root || !csum_root ||
1926 !chunk_root || !dev_root) {
1927 err = -ENOMEM;
1928 goto fail;
1929 }
1930
1931 ret = init_srcu_struct(&fs_info->subvol_srcu);
1932 if (ret) {
1933 err = ret;
1934 goto fail;
1935 }
1936
1937 ret = setup_bdi(fs_info, &fs_info->bdi);
1938 if (ret) {
1939 err = ret;
1940 goto fail_srcu;
1941 }
1942
1943 fs_info->btree_inode = new_inode(sb);
1944 if (!fs_info->btree_inode) {
1945 err = -ENOMEM;
1946 goto fail_bdi;
1947 }
1948
1949 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1950
1951 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1952 INIT_LIST_HEAD(&fs_info->trans_list);
1953 INIT_LIST_HEAD(&fs_info->dead_roots);
1954 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1955 INIT_LIST_HEAD(&fs_info->hashers);
1956 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1957 INIT_LIST_HEAD(&fs_info->ordered_operations);
1958 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1959 spin_lock_init(&fs_info->delalloc_lock);
1960 spin_lock_init(&fs_info->trans_lock);
1961 spin_lock_init(&fs_info->ref_cache_lock);
1962 spin_lock_init(&fs_info->fs_roots_radix_lock);
1963 spin_lock_init(&fs_info->delayed_iput_lock);
1964 spin_lock_init(&fs_info->defrag_inodes_lock);
1965 spin_lock_init(&fs_info->free_chunk_lock);
1966 mutex_init(&fs_info->reloc_mutex);
1967
1968 init_completion(&fs_info->kobj_unregister);
1969 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1970 INIT_LIST_HEAD(&fs_info->space_info);
1971 btrfs_mapping_init(&fs_info->mapping_tree);
1972 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1973 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1974 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1975 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1976 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1977 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1978 atomic_set(&fs_info->nr_async_submits, 0);
1979 atomic_set(&fs_info->async_delalloc_pages, 0);
1980 atomic_set(&fs_info->async_submit_draining, 0);
1981 atomic_set(&fs_info->nr_async_bios, 0);
1982 atomic_set(&fs_info->defrag_running, 0);
1983 fs_info->sb = sb;
1984 fs_info->max_inline = 8192 * 1024;
1985 fs_info->metadata_ratio = 0;
1986 fs_info->defrag_inodes = RB_ROOT;
1987 fs_info->trans_no_join = 0;
1988 fs_info->free_chunk_space = 0;
1989
1990 /* readahead state */
1991 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1992 spin_lock_init(&fs_info->reada_lock);
1993
1994 fs_info->thread_pool_size = min_t(unsigned long,
1995 num_online_cpus() + 2, 8);
1996
1997 INIT_LIST_HEAD(&fs_info->ordered_extents);
1998 spin_lock_init(&fs_info->ordered_extent_lock);
1999 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2000 GFP_NOFS);
2001 if (!fs_info->delayed_root) {
2002 err = -ENOMEM;
2003 goto fail_iput;
2004 }
2005 btrfs_init_delayed_root(fs_info->delayed_root);
2006
2007 mutex_init(&fs_info->scrub_lock);
2008 atomic_set(&fs_info->scrubs_running, 0);
2009 atomic_set(&fs_info->scrub_pause_req, 0);
2010 atomic_set(&fs_info->scrubs_paused, 0);
2011 atomic_set(&fs_info->scrub_cancel_req, 0);
2012 init_waitqueue_head(&fs_info->scrub_pause_wait);
2013 init_rwsem(&fs_info->scrub_super_lock);
2014 fs_info->scrub_workers_refcnt = 0;
2015 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2016 fs_info->check_integrity_print_mask = 0;
2017 #endif
2018
2019 spin_lock_init(&fs_info->balance_lock);
2020 mutex_init(&fs_info->balance_mutex);
2021 atomic_set(&fs_info->balance_running, 0);
2022 atomic_set(&fs_info->balance_pause_req, 0);
2023 atomic_set(&fs_info->balance_cancel_req, 0);
2024 fs_info->balance_ctl = NULL;
2025 init_waitqueue_head(&fs_info->balance_wait_q);
2026
2027 sb->s_blocksize = 4096;
2028 sb->s_blocksize_bits = blksize_bits(4096);
2029 sb->s_bdi = &fs_info->bdi;
2030
2031 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2032 set_nlink(fs_info->btree_inode, 1);
2033 /*
2034 * we set the i_size on the btree inode to the max possible int.
2035 * the real end of the address space is determined by all of
2036 * the devices in the system
2037 */
2038 fs_info->btree_inode->i_size = OFFSET_MAX;
2039 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2040 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2041
2042 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2043 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2044 fs_info->btree_inode->i_mapping);
2045 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2046
2047 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2048
2049 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2050 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2051 sizeof(struct btrfs_key));
2052 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2053 insert_inode_hash(fs_info->btree_inode);
2054
2055 spin_lock_init(&fs_info->block_group_cache_lock);
2056 fs_info->block_group_cache_tree = RB_ROOT;
2057
2058 extent_io_tree_init(&fs_info->freed_extents[0],
2059 fs_info->btree_inode->i_mapping);
2060 extent_io_tree_init(&fs_info->freed_extents[1],
2061 fs_info->btree_inode->i_mapping);
2062 fs_info->pinned_extents = &fs_info->freed_extents[0];
2063 fs_info->do_barriers = 1;
2064
2065
2066 mutex_init(&fs_info->ordered_operations_mutex);
2067 mutex_init(&fs_info->tree_log_mutex);
2068 mutex_init(&fs_info->chunk_mutex);
2069 mutex_init(&fs_info->transaction_kthread_mutex);
2070 mutex_init(&fs_info->cleaner_mutex);
2071 mutex_init(&fs_info->volume_mutex);
2072 init_rwsem(&fs_info->extent_commit_sem);
2073 init_rwsem(&fs_info->cleanup_work_sem);
2074 init_rwsem(&fs_info->subvol_sem);
2075
2076 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2077 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2078
2079 init_waitqueue_head(&fs_info->transaction_throttle);
2080 init_waitqueue_head(&fs_info->transaction_wait);
2081 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2082 init_waitqueue_head(&fs_info->async_submit_wait);
2083
2084 __setup_root(4096, 4096, 4096, 4096, tree_root,
2085 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2086
2087 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2088 if (!bh) {
2089 err = -EINVAL;
2090 goto fail_alloc;
2091 }
2092
2093 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2094 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2095 sizeof(*fs_info->super_for_commit));
2096 brelse(bh);
2097
2098 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2099
2100 disk_super = fs_info->super_copy;
2101 if (!btrfs_super_root(disk_super))
2102 goto fail_alloc;
2103
2104 /* check FS state, whether FS is broken. */
2105 fs_info->fs_state |= btrfs_super_flags(disk_super);
2106
2107 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2108
2109 /*
2110 * run through our array of backup supers and setup
2111 * our ring pointer to the oldest one
2112 */
2113 generation = btrfs_super_generation(disk_super);
2114 find_oldest_super_backup(fs_info, generation);
2115
2116 /*
2117 * In the long term, we'll store the compression type in the super
2118 * block, and it'll be used for per file compression control.
2119 */
2120 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2121
2122 ret = btrfs_parse_options(tree_root, options);
2123 if (ret) {
2124 err = ret;
2125 goto fail_alloc;
2126 }
2127
2128 features = btrfs_super_incompat_flags(disk_super) &
2129 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2130 if (features) {
2131 printk(KERN_ERR "BTRFS: couldn't mount because of "
2132 "unsupported optional features (%Lx).\n",
2133 (unsigned long long)features);
2134 err = -EINVAL;
2135 goto fail_alloc;
2136 }
2137
2138 features = btrfs_super_incompat_flags(disk_super);
2139 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2140 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2141 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2142 btrfs_set_super_incompat_flags(disk_super, features);
2143
2144 features = btrfs_super_compat_ro_flags(disk_super) &
2145 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2146 if (!(sb->s_flags & MS_RDONLY) && features) {
2147 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2148 "unsupported option features (%Lx).\n",
2149 (unsigned long long)features);
2150 err = -EINVAL;
2151 goto fail_alloc;
2152 }
2153
2154 btrfs_init_workers(&fs_info->generic_worker,
2155 "genwork", 1, NULL);
2156
2157 btrfs_init_workers(&fs_info->workers, "worker",
2158 fs_info->thread_pool_size,
2159 &fs_info->generic_worker);
2160
2161 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2162 fs_info->thread_pool_size,
2163 &fs_info->generic_worker);
2164
2165 btrfs_init_workers(&fs_info->submit_workers, "submit",
2166 min_t(u64, fs_devices->num_devices,
2167 fs_info->thread_pool_size),
2168 &fs_info->generic_worker);
2169
2170 btrfs_init_workers(&fs_info->caching_workers, "cache",
2171 2, &fs_info->generic_worker);
2172
2173 /* a higher idle thresh on the submit workers makes it much more
2174 * likely that bios will be send down in a sane order to the
2175 * devices
2176 */
2177 fs_info->submit_workers.idle_thresh = 64;
2178
2179 fs_info->workers.idle_thresh = 16;
2180 fs_info->workers.ordered = 1;
2181
2182 fs_info->delalloc_workers.idle_thresh = 2;
2183 fs_info->delalloc_workers.ordered = 1;
2184
2185 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2186 &fs_info->generic_worker);
2187 btrfs_init_workers(&fs_info->endio_workers, "endio",
2188 fs_info->thread_pool_size,
2189 &fs_info->generic_worker);
2190 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2191 fs_info->thread_pool_size,
2192 &fs_info->generic_worker);
2193 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2194 "endio-meta-write", fs_info->thread_pool_size,
2195 &fs_info->generic_worker);
2196 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2197 fs_info->thread_pool_size,
2198 &fs_info->generic_worker);
2199 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2200 1, &fs_info->generic_worker);
2201 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2202 fs_info->thread_pool_size,
2203 &fs_info->generic_worker);
2204 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2205 fs_info->thread_pool_size,
2206 &fs_info->generic_worker);
2207
2208 /*
2209 * endios are largely parallel and should have a very
2210 * low idle thresh
2211 */
2212 fs_info->endio_workers.idle_thresh = 4;
2213 fs_info->endio_meta_workers.idle_thresh = 4;
2214
2215 fs_info->endio_write_workers.idle_thresh = 2;
2216 fs_info->endio_meta_write_workers.idle_thresh = 2;
2217 fs_info->readahead_workers.idle_thresh = 2;
2218
2219 /*
2220 * btrfs_start_workers can really only fail because of ENOMEM so just
2221 * return -ENOMEM if any of these fail.
2222 */
2223 ret = btrfs_start_workers(&fs_info->workers);
2224 ret |= btrfs_start_workers(&fs_info->generic_worker);
2225 ret |= btrfs_start_workers(&fs_info->submit_workers);
2226 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2227 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2228 ret |= btrfs_start_workers(&fs_info->endio_workers);
2229 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2230 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2231 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2232 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2233 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2234 ret |= btrfs_start_workers(&fs_info->caching_workers);
2235 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2236 if (ret) {
2237 ret = -ENOMEM;
2238 goto fail_sb_buffer;
2239 }
2240
2241 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2242 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2243 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2244
2245 nodesize = btrfs_super_nodesize(disk_super);
2246 leafsize = btrfs_super_leafsize(disk_super);
2247 sectorsize = btrfs_super_sectorsize(disk_super);
2248 stripesize = btrfs_super_stripesize(disk_super);
2249 tree_root->nodesize = nodesize;
2250 tree_root->leafsize = leafsize;
2251 tree_root->sectorsize = sectorsize;
2252 tree_root->stripesize = stripesize;
2253
2254 sb->s_blocksize = sectorsize;
2255 sb->s_blocksize_bits = blksize_bits(sectorsize);
2256
2257 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2258 sizeof(disk_super->magic))) {
2259 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2260 goto fail_sb_buffer;
2261 }
2262
2263 if (sectorsize < PAGE_SIZE) {
2264 printk(KERN_WARNING "btrfs: Incompatible sector size "
2265 "found on %s\n", sb->s_id);
2266 goto fail_sb_buffer;
2267 }
2268
2269 mutex_lock(&fs_info->chunk_mutex);
2270 ret = btrfs_read_sys_array(tree_root);
2271 mutex_unlock(&fs_info->chunk_mutex);
2272 if (ret) {
2273 printk(KERN_WARNING "btrfs: failed to read the system "
2274 "array on %s\n", sb->s_id);
2275 goto fail_sb_buffer;
2276 }
2277
2278 blocksize = btrfs_level_size(tree_root,
2279 btrfs_super_chunk_root_level(disk_super));
2280 generation = btrfs_super_chunk_root_generation(disk_super);
2281
2282 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2283 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2284
2285 chunk_root->node = read_tree_block(chunk_root,
2286 btrfs_super_chunk_root(disk_super),
2287 blocksize, generation);
2288 BUG_ON(!chunk_root->node);
2289 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2290 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2291 sb->s_id);
2292 goto fail_tree_roots;
2293 }
2294 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2295 chunk_root->commit_root = btrfs_root_node(chunk_root);
2296
2297 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2298 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2299 BTRFS_UUID_SIZE);
2300
2301 ret = btrfs_read_chunk_tree(chunk_root);
2302 if (ret) {
2303 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2304 sb->s_id);
2305 goto fail_tree_roots;
2306 }
2307
2308 btrfs_close_extra_devices(fs_devices);
2309
2310 if (!fs_devices->latest_bdev) {
2311 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2312 sb->s_id);
2313 goto fail_tree_roots;
2314 }
2315
2316 retry_root_backup:
2317 blocksize = btrfs_level_size(tree_root,
2318 btrfs_super_root_level(disk_super));
2319 generation = btrfs_super_generation(disk_super);
2320
2321 tree_root->node = read_tree_block(tree_root,
2322 btrfs_super_root(disk_super),
2323 blocksize, generation);
2324 if (!tree_root->node ||
2325 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2326 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2327 sb->s_id);
2328
2329 goto recovery_tree_root;
2330 }
2331
2332 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2333 tree_root->commit_root = btrfs_root_node(tree_root);
2334
2335 ret = find_and_setup_root(tree_root, fs_info,
2336 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2337 if (ret)
2338 goto recovery_tree_root;
2339 extent_root->track_dirty = 1;
2340
2341 ret = find_and_setup_root(tree_root, fs_info,
2342 BTRFS_DEV_TREE_OBJECTID, dev_root);
2343 if (ret)
2344 goto recovery_tree_root;
2345 dev_root->track_dirty = 1;
2346
2347 ret = find_and_setup_root(tree_root, fs_info,
2348 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2349 if (ret)
2350 goto recovery_tree_root;
2351
2352 csum_root->track_dirty = 1;
2353
2354 fs_info->generation = generation;
2355 fs_info->last_trans_committed = generation;
2356
2357 ret = btrfs_init_space_info(fs_info);
2358 if (ret) {
2359 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2360 goto fail_block_groups;
2361 }
2362
2363 ret = btrfs_read_block_groups(extent_root);
2364 if (ret) {
2365 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2366 goto fail_block_groups;
2367 }
2368
2369 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2370 "btrfs-cleaner");
2371 if (IS_ERR(fs_info->cleaner_kthread))
2372 goto fail_block_groups;
2373
2374 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2375 tree_root,
2376 "btrfs-transaction");
2377 if (IS_ERR(fs_info->transaction_kthread))
2378 goto fail_cleaner;
2379
2380 if (!btrfs_test_opt(tree_root, SSD) &&
2381 !btrfs_test_opt(tree_root, NOSSD) &&
2382 !fs_info->fs_devices->rotating) {
2383 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2384 "mode\n");
2385 btrfs_set_opt(fs_info->mount_opt, SSD);
2386 }
2387
2388 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2389 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2390 ret = btrfsic_mount(tree_root, fs_devices,
2391 btrfs_test_opt(tree_root,
2392 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2393 1 : 0,
2394 fs_info->check_integrity_print_mask);
2395 if (ret)
2396 printk(KERN_WARNING "btrfs: failed to initialize"
2397 " integrity check module %s\n", sb->s_id);
2398 }
2399 #endif
2400
2401 /* do not make disk changes in broken FS */
2402 if (btrfs_super_log_root(disk_super) != 0 &&
2403 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2404 u64 bytenr = btrfs_super_log_root(disk_super);
2405
2406 if (fs_devices->rw_devices == 0) {
2407 printk(KERN_WARNING "Btrfs log replay required "
2408 "on RO media\n");
2409 err = -EIO;
2410 goto fail_trans_kthread;
2411 }
2412 blocksize =
2413 btrfs_level_size(tree_root,
2414 btrfs_super_log_root_level(disk_super));
2415
2416 log_tree_root = btrfs_alloc_root(fs_info);
2417 if (!log_tree_root) {
2418 err = -ENOMEM;
2419 goto fail_trans_kthread;
2420 }
2421
2422 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2423 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2424
2425 log_tree_root->node = read_tree_block(tree_root, bytenr,
2426 blocksize,
2427 generation + 1);
2428 ret = btrfs_recover_log_trees(log_tree_root);
2429 BUG_ON(ret);
2430
2431 if (sb->s_flags & MS_RDONLY) {
2432 ret = btrfs_commit_super(tree_root);
2433 BUG_ON(ret);
2434 }
2435 }
2436
2437 ret = btrfs_find_orphan_roots(tree_root);
2438 BUG_ON(ret);
2439
2440 if (!(sb->s_flags & MS_RDONLY)) {
2441 ret = btrfs_cleanup_fs_roots(fs_info);
2442 BUG_ON(ret);
2443
2444 ret = btrfs_recover_relocation(tree_root);
2445 if (ret < 0) {
2446 printk(KERN_WARNING
2447 "btrfs: failed to recover relocation\n");
2448 err = -EINVAL;
2449 goto fail_trans_kthread;
2450 }
2451 }
2452
2453 location.objectid = BTRFS_FS_TREE_OBJECTID;
2454 location.type = BTRFS_ROOT_ITEM_KEY;
2455 location.offset = (u64)-1;
2456
2457 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2458 if (!fs_info->fs_root)
2459 goto fail_trans_kthread;
2460 if (IS_ERR(fs_info->fs_root)) {
2461 err = PTR_ERR(fs_info->fs_root);
2462 goto fail_trans_kthread;
2463 }
2464
2465 if (!(sb->s_flags & MS_RDONLY)) {
2466 down_read(&fs_info->cleanup_work_sem);
2467 err = btrfs_orphan_cleanup(fs_info->fs_root);
2468 if (!err)
2469 err = btrfs_orphan_cleanup(fs_info->tree_root);
2470 up_read(&fs_info->cleanup_work_sem);
2471
2472 if (!err)
2473 err = btrfs_recover_balance(fs_info->tree_root);
2474
2475 if (err) {
2476 close_ctree(tree_root);
2477 return err;
2478 }
2479 }
2480
2481 return 0;
2482
2483 fail_trans_kthread:
2484 kthread_stop(fs_info->transaction_kthread);
2485 fail_cleaner:
2486 kthread_stop(fs_info->cleaner_kthread);
2487
2488 /*
2489 * make sure we're done with the btree inode before we stop our
2490 * kthreads
2491 */
2492 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2493 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2494
2495 fail_block_groups:
2496 btrfs_free_block_groups(fs_info);
2497
2498 fail_tree_roots:
2499 free_root_pointers(fs_info, 1);
2500
2501 fail_sb_buffer:
2502 btrfs_stop_workers(&fs_info->generic_worker);
2503 btrfs_stop_workers(&fs_info->readahead_workers);
2504 btrfs_stop_workers(&fs_info->fixup_workers);
2505 btrfs_stop_workers(&fs_info->delalloc_workers);
2506 btrfs_stop_workers(&fs_info->workers);
2507 btrfs_stop_workers(&fs_info->endio_workers);
2508 btrfs_stop_workers(&fs_info->endio_meta_workers);
2509 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2510 btrfs_stop_workers(&fs_info->endio_write_workers);
2511 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2512 btrfs_stop_workers(&fs_info->submit_workers);
2513 btrfs_stop_workers(&fs_info->delayed_workers);
2514 btrfs_stop_workers(&fs_info->caching_workers);
2515 fail_alloc:
2516 fail_iput:
2517 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2518
2519 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2520 iput(fs_info->btree_inode);
2521 fail_bdi:
2522 bdi_destroy(&fs_info->bdi);
2523 fail_srcu:
2524 cleanup_srcu_struct(&fs_info->subvol_srcu);
2525 fail:
2526 btrfs_close_devices(fs_info->fs_devices);
2527 return err;
2528
2529 recovery_tree_root:
2530 if (!btrfs_test_opt(tree_root, RECOVERY))
2531 goto fail_tree_roots;
2532
2533 free_root_pointers(fs_info, 0);
2534
2535 /* don't use the log in recovery mode, it won't be valid */
2536 btrfs_set_super_log_root(disk_super, 0);
2537
2538 /* we can't trust the free space cache either */
2539 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2540
2541 ret = next_root_backup(fs_info, fs_info->super_copy,
2542 &num_backups_tried, &backup_index);
2543 if (ret == -1)
2544 goto fail_block_groups;
2545 goto retry_root_backup;
2546 }
2547
2548 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2549 {
2550 char b[BDEVNAME_SIZE];
2551
2552 if (uptodate) {
2553 set_buffer_uptodate(bh);
2554 } else {
2555 printk_ratelimited(KERN_WARNING "lost page write due to "
2556 "I/O error on %s\n",
2557 bdevname(bh->b_bdev, b));
2558 /* note, we dont' set_buffer_write_io_error because we have
2559 * our own ways of dealing with the IO errors
2560 */
2561 clear_buffer_uptodate(bh);
2562 }
2563 unlock_buffer(bh);
2564 put_bh(bh);
2565 }
2566
2567 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2568 {
2569 struct buffer_head *bh;
2570 struct buffer_head *latest = NULL;
2571 struct btrfs_super_block *super;
2572 int i;
2573 u64 transid = 0;
2574 u64 bytenr;
2575
2576 /* we would like to check all the supers, but that would make
2577 * a btrfs mount succeed after a mkfs from a different FS.
2578 * So, we need to add a special mount option to scan for
2579 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2580 */
2581 for (i = 0; i < 1; i++) {
2582 bytenr = btrfs_sb_offset(i);
2583 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2584 break;
2585 bh = __bread(bdev, bytenr / 4096, 4096);
2586 if (!bh)
2587 continue;
2588
2589 super = (struct btrfs_super_block *)bh->b_data;
2590 if (btrfs_super_bytenr(super) != bytenr ||
2591 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2592 sizeof(super->magic))) {
2593 brelse(bh);
2594 continue;
2595 }
2596
2597 if (!latest || btrfs_super_generation(super) > transid) {
2598 brelse(latest);
2599 latest = bh;
2600 transid = btrfs_super_generation(super);
2601 } else {
2602 brelse(bh);
2603 }
2604 }
2605 return latest;
2606 }
2607
2608 /*
2609 * this should be called twice, once with wait == 0 and
2610 * once with wait == 1. When wait == 0 is done, all the buffer heads
2611 * we write are pinned.
2612 *
2613 * They are released when wait == 1 is done.
2614 * max_mirrors must be the same for both runs, and it indicates how
2615 * many supers on this one device should be written.
2616 *
2617 * max_mirrors == 0 means to write them all.
2618 */
2619 static int write_dev_supers(struct btrfs_device *device,
2620 struct btrfs_super_block *sb,
2621 int do_barriers, int wait, int max_mirrors)
2622 {
2623 struct buffer_head *bh;
2624 int i;
2625 int ret;
2626 int errors = 0;
2627 u32 crc;
2628 u64 bytenr;
2629
2630 if (max_mirrors == 0)
2631 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2632
2633 for (i = 0; i < max_mirrors; i++) {
2634 bytenr = btrfs_sb_offset(i);
2635 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2636 break;
2637
2638 if (wait) {
2639 bh = __find_get_block(device->bdev, bytenr / 4096,
2640 BTRFS_SUPER_INFO_SIZE);
2641 BUG_ON(!bh);
2642 wait_on_buffer(bh);
2643 if (!buffer_uptodate(bh))
2644 errors++;
2645
2646 /* drop our reference */
2647 brelse(bh);
2648
2649 /* drop the reference from the wait == 0 run */
2650 brelse(bh);
2651 continue;
2652 } else {
2653 btrfs_set_super_bytenr(sb, bytenr);
2654
2655 crc = ~(u32)0;
2656 crc = btrfs_csum_data(NULL, (char *)sb +
2657 BTRFS_CSUM_SIZE, crc,
2658 BTRFS_SUPER_INFO_SIZE -
2659 BTRFS_CSUM_SIZE);
2660 btrfs_csum_final(crc, sb->csum);
2661
2662 /*
2663 * one reference for us, and we leave it for the
2664 * caller
2665 */
2666 bh = __getblk(device->bdev, bytenr / 4096,
2667 BTRFS_SUPER_INFO_SIZE);
2668 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2669
2670 /* one reference for submit_bh */
2671 get_bh(bh);
2672
2673 set_buffer_uptodate(bh);
2674 lock_buffer(bh);
2675 bh->b_end_io = btrfs_end_buffer_write_sync;
2676 }
2677
2678 /*
2679 * we fua the first super. The others we allow
2680 * to go down lazy.
2681 */
2682 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2683 if (ret)
2684 errors++;
2685 }
2686 return errors < i ? 0 : -1;
2687 }
2688
2689 /*
2690 * endio for the write_dev_flush, this will wake anyone waiting
2691 * for the barrier when it is done
2692 */
2693 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2694 {
2695 if (err) {
2696 if (err == -EOPNOTSUPP)
2697 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2698 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2699 }
2700 if (bio->bi_private)
2701 complete(bio->bi_private);
2702 bio_put(bio);
2703 }
2704
2705 /*
2706 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2707 * sent down. With wait == 1, it waits for the previous flush.
2708 *
2709 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2710 * capable
2711 */
2712 static int write_dev_flush(struct btrfs_device *device, int wait)
2713 {
2714 struct bio *bio;
2715 int ret = 0;
2716
2717 if (device->nobarriers)
2718 return 0;
2719
2720 if (wait) {
2721 bio = device->flush_bio;
2722 if (!bio)
2723 return 0;
2724
2725 wait_for_completion(&device->flush_wait);
2726
2727 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2728 printk("btrfs: disabling barriers on dev %s\n",
2729 device->name);
2730 device->nobarriers = 1;
2731 }
2732 if (!bio_flagged(bio, BIO_UPTODATE)) {
2733 ret = -EIO;
2734 }
2735
2736 /* drop the reference from the wait == 0 run */
2737 bio_put(bio);
2738 device->flush_bio = NULL;
2739
2740 return ret;
2741 }
2742
2743 /*
2744 * one reference for us, and we leave it for the
2745 * caller
2746 */
2747 device->flush_bio = NULL;;
2748 bio = bio_alloc(GFP_NOFS, 0);
2749 if (!bio)
2750 return -ENOMEM;
2751
2752 bio->bi_end_io = btrfs_end_empty_barrier;
2753 bio->bi_bdev = device->bdev;
2754 init_completion(&device->flush_wait);
2755 bio->bi_private = &device->flush_wait;
2756 device->flush_bio = bio;
2757
2758 bio_get(bio);
2759 btrfsic_submit_bio(WRITE_FLUSH, bio);
2760
2761 return 0;
2762 }
2763
2764 /*
2765 * send an empty flush down to each device in parallel,
2766 * then wait for them
2767 */
2768 static int barrier_all_devices(struct btrfs_fs_info *info)
2769 {
2770 struct list_head *head;
2771 struct btrfs_device *dev;
2772 int errors = 0;
2773 int ret;
2774
2775 /* send down all the barriers */
2776 head = &info->fs_devices->devices;
2777 list_for_each_entry_rcu(dev, head, dev_list) {
2778 if (!dev->bdev) {
2779 errors++;
2780 continue;
2781 }
2782 if (!dev->in_fs_metadata || !dev->writeable)
2783 continue;
2784
2785 ret = write_dev_flush(dev, 0);
2786 if (ret)
2787 errors++;
2788 }
2789
2790 /* wait for all the barriers */
2791 list_for_each_entry_rcu(dev, head, dev_list) {
2792 if (!dev->bdev) {
2793 errors++;
2794 continue;
2795 }
2796 if (!dev->in_fs_metadata || !dev->writeable)
2797 continue;
2798
2799 ret = write_dev_flush(dev, 1);
2800 if (ret)
2801 errors++;
2802 }
2803 if (errors)
2804 return -EIO;
2805 return 0;
2806 }
2807
2808 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2809 {
2810 struct list_head *head;
2811 struct btrfs_device *dev;
2812 struct btrfs_super_block *sb;
2813 struct btrfs_dev_item *dev_item;
2814 int ret;
2815 int do_barriers;
2816 int max_errors;
2817 int total_errors = 0;
2818 u64 flags;
2819
2820 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2821 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2822 backup_super_roots(root->fs_info);
2823
2824 sb = root->fs_info->super_for_commit;
2825 dev_item = &sb->dev_item;
2826
2827 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2828 head = &root->fs_info->fs_devices->devices;
2829
2830 if (do_barriers)
2831 barrier_all_devices(root->fs_info);
2832
2833 list_for_each_entry_rcu(dev, head, dev_list) {
2834 if (!dev->bdev) {
2835 total_errors++;
2836 continue;
2837 }
2838 if (!dev->in_fs_metadata || !dev->writeable)
2839 continue;
2840
2841 btrfs_set_stack_device_generation(dev_item, 0);
2842 btrfs_set_stack_device_type(dev_item, dev->type);
2843 btrfs_set_stack_device_id(dev_item, dev->devid);
2844 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2845 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2846 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2847 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2848 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2849 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2850 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2851
2852 flags = btrfs_super_flags(sb);
2853 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2854
2855 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2856 if (ret)
2857 total_errors++;
2858 }
2859 if (total_errors > max_errors) {
2860 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2861 total_errors);
2862 BUG();
2863 }
2864
2865 total_errors = 0;
2866 list_for_each_entry_rcu(dev, head, dev_list) {
2867 if (!dev->bdev)
2868 continue;
2869 if (!dev->in_fs_metadata || !dev->writeable)
2870 continue;
2871
2872 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2873 if (ret)
2874 total_errors++;
2875 }
2876 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2877 if (total_errors > max_errors) {
2878 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2879 total_errors);
2880 BUG();
2881 }
2882 return 0;
2883 }
2884
2885 int write_ctree_super(struct btrfs_trans_handle *trans,
2886 struct btrfs_root *root, int max_mirrors)
2887 {
2888 int ret;
2889
2890 ret = write_all_supers(root, max_mirrors);
2891 return ret;
2892 }
2893
2894 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2895 {
2896 spin_lock(&fs_info->fs_roots_radix_lock);
2897 radix_tree_delete(&fs_info->fs_roots_radix,
2898 (unsigned long)root->root_key.objectid);
2899 spin_unlock(&fs_info->fs_roots_radix_lock);
2900
2901 if (btrfs_root_refs(&root->root_item) == 0)
2902 synchronize_srcu(&fs_info->subvol_srcu);
2903
2904 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2905 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2906 free_fs_root(root);
2907 return 0;
2908 }
2909
2910 static void free_fs_root(struct btrfs_root *root)
2911 {
2912 iput(root->cache_inode);
2913 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2914 if (root->anon_dev)
2915 free_anon_bdev(root->anon_dev);
2916 free_extent_buffer(root->node);
2917 free_extent_buffer(root->commit_root);
2918 kfree(root->free_ino_ctl);
2919 kfree(root->free_ino_pinned);
2920 kfree(root->name);
2921 kfree(root);
2922 }
2923
2924 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2925 {
2926 int ret;
2927 struct btrfs_root *gang[8];
2928 int i;
2929
2930 while (!list_empty(&fs_info->dead_roots)) {
2931 gang[0] = list_entry(fs_info->dead_roots.next,
2932 struct btrfs_root, root_list);
2933 list_del(&gang[0]->root_list);
2934
2935 if (gang[0]->in_radix) {
2936 btrfs_free_fs_root(fs_info, gang[0]);
2937 } else {
2938 free_extent_buffer(gang[0]->node);
2939 free_extent_buffer(gang[0]->commit_root);
2940 kfree(gang[0]);
2941 }
2942 }
2943
2944 while (1) {
2945 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2946 (void **)gang, 0,
2947 ARRAY_SIZE(gang));
2948 if (!ret)
2949 break;
2950 for (i = 0; i < ret; i++)
2951 btrfs_free_fs_root(fs_info, gang[i]);
2952 }
2953 return 0;
2954 }
2955
2956 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2957 {
2958 u64 root_objectid = 0;
2959 struct btrfs_root *gang[8];
2960 int i;
2961 int ret;
2962
2963 while (1) {
2964 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2965 (void **)gang, root_objectid,
2966 ARRAY_SIZE(gang));
2967 if (!ret)
2968 break;
2969
2970 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2971 for (i = 0; i < ret; i++) {
2972 int err;
2973
2974 root_objectid = gang[i]->root_key.objectid;
2975 err = btrfs_orphan_cleanup(gang[i]);
2976 if (err)
2977 return err;
2978 }
2979 root_objectid++;
2980 }
2981 return 0;
2982 }
2983
2984 int btrfs_commit_super(struct btrfs_root *root)
2985 {
2986 struct btrfs_trans_handle *trans;
2987 int ret;
2988
2989 mutex_lock(&root->fs_info->cleaner_mutex);
2990 btrfs_run_delayed_iputs(root);
2991 btrfs_clean_old_snapshots(root);
2992 mutex_unlock(&root->fs_info->cleaner_mutex);
2993
2994 /* wait until ongoing cleanup work done */
2995 down_write(&root->fs_info->cleanup_work_sem);
2996 up_write(&root->fs_info->cleanup_work_sem);
2997
2998 trans = btrfs_join_transaction(root);
2999 if (IS_ERR(trans))
3000 return PTR_ERR(trans);
3001 ret = btrfs_commit_transaction(trans, root);
3002 BUG_ON(ret);
3003 /* run commit again to drop the original snapshot */
3004 trans = btrfs_join_transaction(root);
3005 if (IS_ERR(trans))
3006 return PTR_ERR(trans);
3007 btrfs_commit_transaction(trans, root);
3008 ret = btrfs_write_and_wait_transaction(NULL, root);
3009 BUG_ON(ret);
3010
3011 ret = write_ctree_super(NULL, root, 0);
3012 return ret;
3013 }
3014
3015 int close_ctree(struct btrfs_root *root)
3016 {
3017 struct btrfs_fs_info *fs_info = root->fs_info;
3018 int ret;
3019
3020 fs_info->closing = 1;
3021 smp_mb();
3022
3023 /* pause restriper - we want to resume on mount */
3024 btrfs_pause_balance(root->fs_info);
3025
3026 btrfs_scrub_cancel(root);
3027
3028 /* wait for any defraggers to finish */
3029 wait_event(fs_info->transaction_wait,
3030 (atomic_read(&fs_info->defrag_running) == 0));
3031
3032 /* clear out the rbtree of defraggable inodes */
3033 btrfs_run_defrag_inodes(fs_info);
3034
3035 /*
3036 * Here come 2 situations when btrfs is broken to flip readonly:
3037 *
3038 * 1. when btrfs flips readonly somewhere else before
3039 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3040 * and btrfs will skip to write sb directly to keep
3041 * ERROR state on disk.
3042 *
3043 * 2. when btrfs flips readonly just in btrfs_commit_super,
3044 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3045 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3046 * btrfs will cleanup all FS resources first and write sb then.
3047 */
3048 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3049 ret = btrfs_commit_super(root);
3050 if (ret)
3051 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3052 }
3053
3054 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3055 ret = btrfs_error_commit_super(root);
3056 if (ret)
3057 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3058 }
3059
3060 btrfs_put_block_group_cache(fs_info);
3061
3062 kthread_stop(fs_info->transaction_kthread);
3063 kthread_stop(fs_info->cleaner_kthread);
3064
3065 fs_info->closing = 2;
3066 smp_mb();
3067
3068 if (fs_info->delalloc_bytes) {
3069 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3070 (unsigned long long)fs_info->delalloc_bytes);
3071 }
3072 if (fs_info->total_ref_cache_size) {
3073 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3074 (unsigned long long)fs_info->total_ref_cache_size);
3075 }
3076
3077 free_extent_buffer(fs_info->extent_root->node);
3078 free_extent_buffer(fs_info->extent_root->commit_root);
3079 free_extent_buffer(fs_info->tree_root->node);
3080 free_extent_buffer(fs_info->tree_root->commit_root);
3081 free_extent_buffer(fs_info->chunk_root->node);
3082 free_extent_buffer(fs_info->chunk_root->commit_root);
3083 free_extent_buffer(fs_info->dev_root->node);
3084 free_extent_buffer(fs_info->dev_root->commit_root);
3085 free_extent_buffer(fs_info->csum_root->node);
3086 free_extent_buffer(fs_info->csum_root->commit_root);
3087
3088 btrfs_free_block_groups(fs_info);
3089
3090 del_fs_roots(fs_info);
3091
3092 iput(fs_info->btree_inode);
3093
3094 btrfs_stop_workers(&fs_info->generic_worker);
3095 btrfs_stop_workers(&fs_info->fixup_workers);
3096 btrfs_stop_workers(&fs_info->delalloc_workers);
3097 btrfs_stop_workers(&fs_info->workers);
3098 btrfs_stop_workers(&fs_info->endio_workers);
3099 btrfs_stop_workers(&fs_info->endio_meta_workers);
3100 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3101 btrfs_stop_workers(&fs_info->endio_write_workers);
3102 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3103 btrfs_stop_workers(&fs_info->submit_workers);
3104 btrfs_stop_workers(&fs_info->delayed_workers);
3105 btrfs_stop_workers(&fs_info->caching_workers);
3106 btrfs_stop_workers(&fs_info->readahead_workers);
3107
3108 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3109 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3110 btrfsic_unmount(root, fs_info->fs_devices);
3111 #endif
3112
3113 btrfs_close_devices(fs_info->fs_devices);
3114 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3115
3116 bdi_destroy(&fs_info->bdi);
3117 cleanup_srcu_struct(&fs_info->subvol_srcu);
3118
3119 return 0;
3120 }
3121
3122 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3123 {
3124 int ret;
3125 struct inode *btree_inode = buf->first_page->mapping->host;
3126
3127 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3128 NULL);
3129 if (!ret)
3130 return ret;
3131
3132 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3133 parent_transid);
3134 return !ret;
3135 }
3136
3137 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3138 {
3139 struct inode *btree_inode = buf->first_page->mapping->host;
3140 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3141 buf);
3142 }
3143
3144 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3145 {
3146 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3147 u64 transid = btrfs_header_generation(buf);
3148 struct inode *btree_inode = root->fs_info->btree_inode;
3149 int was_dirty;
3150
3151 btrfs_assert_tree_locked(buf);
3152 if (transid != root->fs_info->generation) {
3153 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3154 "found %llu running %llu\n",
3155 (unsigned long long)buf->start,
3156 (unsigned long long)transid,
3157 (unsigned long long)root->fs_info->generation);
3158 WARN_ON(1);
3159 }
3160 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3161 buf);
3162 if (!was_dirty) {
3163 spin_lock(&root->fs_info->delalloc_lock);
3164 root->fs_info->dirty_metadata_bytes += buf->len;
3165 spin_unlock(&root->fs_info->delalloc_lock);
3166 }
3167 }
3168
3169 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3170 {
3171 /*
3172 * looks as though older kernels can get into trouble with
3173 * this code, they end up stuck in balance_dirty_pages forever
3174 */
3175 u64 num_dirty;
3176 unsigned long thresh = 32 * 1024 * 1024;
3177
3178 if (current->flags & PF_MEMALLOC)
3179 return;
3180
3181 btrfs_balance_delayed_items(root);
3182
3183 num_dirty = root->fs_info->dirty_metadata_bytes;
3184
3185 if (num_dirty > thresh) {
3186 balance_dirty_pages_ratelimited_nr(
3187 root->fs_info->btree_inode->i_mapping, 1);
3188 }
3189 return;
3190 }
3191
3192 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3193 {
3194 /*
3195 * looks as though older kernels can get into trouble with
3196 * this code, they end up stuck in balance_dirty_pages forever
3197 */
3198 u64 num_dirty;
3199 unsigned long thresh = 32 * 1024 * 1024;
3200
3201 if (current->flags & PF_MEMALLOC)
3202 return;
3203
3204 num_dirty = root->fs_info->dirty_metadata_bytes;
3205
3206 if (num_dirty > thresh) {
3207 balance_dirty_pages_ratelimited_nr(
3208 root->fs_info->btree_inode->i_mapping, 1);
3209 }
3210 return;
3211 }
3212
3213 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3214 {
3215 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3216 int ret;
3217 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3218 if (ret == 0)
3219 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3220 return ret;
3221 }
3222
3223 static int btree_lock_page_hook(struct page *page, void *data,
3224 void (*flush_fn)(void *))
3225 {
3226 struct inode *inode = page->mapping->host;
3227 struct btrfs_root *root = BTRFS_I(inode)->root;
3228 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3229 struct extent_buffer *eb;
3230 unsigned long len;
3231 u64 bytenr = page_offset(page);
3232
3233 if (page->private == EXTENT_PAGE_PRIVATE)
3234 goto out;
3235
3236 len = page->private >> 2;
3237 eb = find_extent_buffer(io_tree, bytenr, len);
3238 if (!eb)
3239 goto out;
3240
3241 if (!btrfs_try_tree_write_lock(eb)) {
3242 flush_fn(data);
3243 btrfs_tree_lock(eb);
3244 }
3245 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3246
3247 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3248 spin_lock(&root->fs_info->delalloc_lock);
3249 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3250 root->fs_info->dirty_metadata_bytes -= eb->len;
3251 else
3252 WARN_ON(1);
3253 spin_unlock(&root->fs_info->delalloc_lock);
3254 }
3255
3256 btrfs_tree_unlock(eb);
3257 free_extent_buffer(eb);
3258 out:
3259 if (!trylock_page(page)) {
3260 flush_fn(data);
3261 lock_page(page);
3262 }
3263 return 0;
3264 }
3265
3266 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3267 int read_only)
3268 {
3269 if (read_only)
3270 return;
3271
3272 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3273 printk(KERN_WARNING "warning: mount fs with errors, "
3274 "running btrfsck is recommended\n");
3275 }
3276
3277 int btrfs_error_commit_super(struct btrfs_root *root)
3278 {
3279 int ret;
3280
3281 mutex_lock(&root->fs_info->cleaner_mutex);
3282 btrfs_run_delayed_iputs(root);
3283 mutex_unlock(&root->fs_info->cleaner_mutex);
3284
3285 down_write(&root->fs_info->cleanup_work_sem);
3286 up_write(&root->fs_info->cleanup_work_sem);
3287
3288 /* cleanup FS via transaction */
3289 btrfs_cleanup_transaction(root);
3290
3291 ret = write_ctree_super(NULL, root, 0);
3292
3293 return ret;
3294 }
3295
3296 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3297 {
3298 struct btrfs_inode *btrfs_inode;
3299 struct list_head splice;
3300
3301 INIT_LIST_HEAD(&splice);
3302
3303 mutex_lock(&root->fs_info->ordered_operations_mutex);
3304 spin_lock(&root->fs_info->ordered_extent_lock);
3305
3306 list_splice_init(&root->fs_info->ordered_operations, &splice);
3307 while (!list_empty(&splice)) {
3308 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3309 ordered_operations);
3310
3311 list_del_init(&btrfs_inode->ordered_operations);
3312
3313 btrfs_invalidate_inodes(btrfs_inode->root);
3314 }
3315
3316 spin_unlock(&root->fs_info->ordered_extent_lock);
3317 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3318
3319 return 0;
3320 }
3321
3322 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3323 {
3324 struct list_head splice;
3325 struct btrfs_ordered_extent *ordered;
3326 struct inode *inode;
3327
3328 INIT_LIST_HEAD(&splice);
3329
3330 spin_lock(&root->fs_info->ordered_extent_lock);
3331
3332 list_splice_init(&root->fs_info->ordered_extents, &splice);
3333 while (!list_empty(&splice)) {
3334 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3335 root_extent_list);
3336
3337 list_del_init(&ordered->root_extent_list);
3338 atomic_inc(&ordered->refs);
3339
3340 /* the inode may be getting freed (in sys_unlink path). */
3341 inode = igrab(ordered->inode);
3342
3343 spin_unlock(&root->fs_info->ordered_extent_lock);
3344 if (inode)
3345 iput(inode);
3346
3347 atomic_set(&ordered->refs, 1);
3348 btrfs_put_ordered_extent(ordered);
3349
3350 spin_lock(&root->fs_info->ordered_extent_lock);
3351 }
3352
3353 spin_unlock(&root->fs_info->ordered_extent_lock);
3354
3355 return 0;
3356 }
3357
3358 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3359 struct btrfs_root *root)
3360 {
3361 struct rb_node *node;
3362 struct btrfs_delayed_ref_root *delayed_refs;
3363 struct btrfs_delayed_ref_node *ref;
3364 int ret = 0;
3365
3366 delayed_refs = &trans->delayed_refs;
3367
3368 spin_lock(&delayed_refs->lock);
3369 if (delayed_refs->num_entries == 0) {
3370 spin_unlock(&delayed_refs->lock);
3371 printk(KERN_INFO "delayed_refs has NO entry\n");
3372 return ret;
3373 }
3374
3375 node = rb_first(&delayed_refs->root);
3376 while (node) {
3377 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3378 node = rb_next(node);
3379
3380 ref->in_tree = 0;
3381 rb_erase(&ref->rb_node, &delayed_refs->root);
3382 delayed_refs->num_entries--;
3383
3384 atomic_set(&ref->refs, 1);
3385 if (btrfs_delayed_ref_is_head(ref)) {
3386 struct btrfs_delayed_ref_head *head;
3387
3388 head = btrfs_delayed_node_to_head(ref);
3389 mutex_lock(&head->mutex);
3390 kfree(head->extent_op);
3391 delayed_refs->num_heads--;
3392 if (list_empty(&head->cluster))
3393 delayed_refs->num_heads_ready--;
3394 list_del_init(&head->cluster);
3395 mutex_unlock(&head->mutex);
3396 }
3397
3398 spin_unlock(&delayed_refs->lock);
3399 btrfs_put_delayed_ref(ref);
3400
3401 cond_resched();
3402 spin_lock(&delayed_refs->lock);
3403 }
3404
3405 spin_unlock(&delayed_refs->lock);
3406
3407 return ret;
3408 }
3409
3410 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3411 {
3412 struct btrfs_pending_snapshot *snapshot;
3413 struct list_head splice;
3414
3415 INIT_LIST_HEAD(&splice);
3416
3417 list_splice_init(&t->pending_snapshots, &splice);
3418
3419 while (!list_empty(&splice)) {
3420 snapshot = list_entry(splice.next,
3421 struct btrfs_pending_snapshot,
3422 list);
3423
3424 list_del_init(&snapshot->list);
3425
3426 kfree(snapshot);
3427 }
3428
3429 return 0;
3430 }
3431
3432 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3433 {
3434 struct btrfs_inode *btrfs_inode;
3435 struct list_head splice;
3436
3437 INIT_LIST_HEAD(&splice);
3438
3439 spin_lock(&root->fs_info->delalloc_lock);
3440 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3441
3442 while (!list_empty(&splice)) {
3443 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3444 delalloc_inodes);
3445
3446 list_del_init(&btrfs_inode->delalloc_inodes);
3447
3448 btrfs_invalidate_inodes(btrfs_inode->root);
3449 }
3450
3451 spin_unlock(&root->fs_info->delalloc_lock);
3452
3453 return 0;
3454 }
3455
3456 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3457 struct extent_io_tree *dirty_pages,
3458 int mark)
3459 {
3460 int ret;
3461 struct page *page;
3462 struct inode *btree_inode = root->fs_info->btree_inode;
3463 struct extent_buffer *eb;
3464 u64 start = 0;
3465 u64 end;
3466 u64 offset;
3467 unsigned long index;
3468
3469 while (1) {
3470 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3471 mark);
3472 if (ret)
3473 break;
3474
3475 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3476 while (start <= end) {
3477 index = start >> PAGE_CACHE_SHIFT;
3478 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3479 page = find_get_page(btree_inode->i_mapping, index);
3480 if (!page)
3481 continue;
3482 offset = page_offset(page);
3483
3484 spin_lock(&dirty_pages->buffer_lock);
3485 eb = radix_tree_lookup(
3486 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3487 offset >> PAGE_CACHE_SHIFT);
3488 spin_unlock(&dirty_pages->buffer_lock);
3489 if (eb) {
3490 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3491 &eb->bflags);
3492 atomic_set(&eb->refs, 1);
3493 }
3494 if (PageWriteback(page))
3495 end_page_writeback(page);
3496
3497 lock_page(page);
3498 if (PageDirty(page)) {
3499 clear_page_dirty_for_io(page);
3500 spin_lock_irq(&page->mapping->tree_lock);
3501 radix_tree_tag_clear(&page->mapping->page_tree,
3502 page_index(page),
3503 PAGECACHE_TAG_DIRTY);
3504 spin_unlock_irq(&page->mapping->tree_lock);
3505 }
3506
3507 page->mapping->a_ops->invalidatepage(page, 0);
3508 unlock_page(page);
3509 }
3510 }
3511
3512 return ret;
3513 }
3514
3515 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3516 struct extent_io_tree *pinned_extents)
3517 {
3518 struct extent_io_tree *unpin;
3519 u64 start;
3520 u64 end;
3521 int ret;
3522
3523 unpin = pinned_extents;
3524 while (1) {
3525 ret = find_first_extent_bit(unpin, 0, &start, &end,
3526 EXTENT_DIRTY);
3527 if (ret)
3528 break;
3529
3530 /* opt_discard */
3531 if (btrfs_test_opt(root, DISCARD))
3532 ret = btrfs_error_discard_extent(root, start,
3533 end + 1 - start,
3534 NULL);
3535
3536 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3537 btrfs_error_unpin_extent_range(root, start, end);
3538 cond_resched();
3539 }
3540
3541 return 0;
3542 }
3543
3544 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3545 {
3546 struct btrfs_transaction *t;
3547 LIST_HEAD(list);
3548
3549 WARN_ON(1);
3550
3551 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3552
3553 spin_lock(&root->fs_info->trans_lock);
3554 list_splice_init(&root->fs_info->trans_list, &list);
3555 root->fs_info->trans_no_join = 1;
3556 spin_unlock(&root->fs_info->trans_lock);
3557
3558 while (!list_empty(&list)) {
3559 t = list_entry(list.next, struct btrfs_transaction, list);
3560 if (!t)
3561 break;
3562
3563 btrfs_destroy_ordered_operations(root);
3564
3565 btrfs_destroy_ordered_extents(root);
3566
3567 btrfs_destroy_delayed_refs(t, root);
3568
3569 btrfs_block_rsv_release(root,
3570 &root->fs_info->trans_block_rsv,
3571 t->dirty_pages.dirty_bytes);
3572
3573 /* FIXME: cleanup wait for commit */
3574 t->in_commit = 1;
3575 t->blocked = 1;
3576 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3577 wake_up(&root->fs_info->transaction_blocked_wait);
3578
3579 t->blocked = 0;
3580 if (waitqueue_active(&root->fs_info->transaction_wait))
3581 wake_up(&root->fs_info->transaction_wait);
3582
3583 t->commit_done = 1;
3584 if (waitqueue_active(&t->commit_wait))
3585 wake_up(&t->commit_wait);
3586
3587 btrfs_destroy_pending_snapshots(t);
3588
3589 btrfs_destroy_delalloc_inodes(root);
3590
3591 spin_lock(&root->fs_info->trans_lock);
3592 root->fs_info->running_transaction = NULL;
3593 spin_unlock(&root->fs_info->trans_lock);
3594
3595 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3596 EXTENT_DIRTY);
3597
3598 btrfs_destroy_pinned_extent(root,
3599 root->fs_info->pinned_extents);
3600
3601 atomic_set(&t->use_count, 0);
3602 list_del_init(&t->list);
3603 memset(t, 0, sizeof(*t));
3604 kmem_cache_free(btrfs_transaction_cachep, t);
3605 }
3606
3607 spin_lock(&root->fs_info->trans_lock);
3608 root->fs_info->trans_no_join = 0;
3609 spin_unlock(&root->fs_info->trans_lock);
3610 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3611
3612 return 0;
3613 }
3614
3615 static struct extent_io_ops btree_extent_io_ops = {
3616 .write_cache_pages_lock_hook = btree_lock_page_hook,
3617 .readpage_end_io_hook = btree_readpage_end_io_hook,
3618 .readpage_io_failed_hook = btree_io_failed_hook,
3619 .submit_bio_hook = btree_submit_bio_hook,
3620 /* note we're sharing with inode.c for the merge bio hook */
3621 .merge_bio_hook = btrfs_merge_bio_hook,
3622 };