]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/btrfs/disk-io.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: write out free space cache
[net-next-2.6.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 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
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
4b4e25f2 31#include "compat.h"
eb60ceac
CM		 32#include "ctree.h"
33#include "disk-io.h"
e089f05c 34#include "transaction.h"
0f7d52f4 35#include "btrfs_inode.h"
0b86a832 36#include "volumes.h"
db94535d 37#include "print-tree.h"
8b712842 38#include "async-thread.h"
925baedd 39#include "locking.h"
e02119d5 40#include "tree-log.h"
fa9c0d79 41#include "free-space-cache.h"
eb60ceac 42
d1310b2e 43static struct extent_io_ops btree_extent_io_ops;
8b712842 44static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 45static void free_fs_root(struct btrfs_root *root);
ce9adaa5 46
d352ac68
CM		 47/*
48 * end_io_wq structs are used to do processing in task context when an IO is
49 * complete. This is used during reads to verify checksums, and it is used
50 * by writes to insert metadata for new file extents after IO is complete.
51 */
ce9adaa5
CM		 52struct end_io_wq {
53 struct bio *bio;
54 bio_end_io_t *end_io;
55 void *private;
56 struct btrfs_fs_info *info;
57 int error;
22c59948 58 int metadata;
ce9adaa5 59 struct list_head list;
8b712842 60 struct btrfs_work work;
ce9adaa5 61};
0da5468f 62
d352ac68
CM		 63/*
64 * async submit bios are used to offload expensive checksumming
65 * onto the worker threads. They checksum file and metadata bios
66 * just before they are sent down the IO stack.
67 */
44b8bd7e
CM		 68struct async_submit_bio {
69 struct inode *inode;
70 struct bio *bio;
71 struct list_head list;
4a69a410
CM		 72 extent_submit_bio_hook_t *submit_bio_start;
73 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 74 int rw;
75 int mirror_num;
c8b97818 76 unsigned long bio_flags;
eaf25d93
CM		 77 /*
78 * bio_offset is optional, can be used if the pages in the bio
79 * can't tell us where in the file the bio should go
80 */
81 u64 bio_offset;
8b712842 82 struct btrfs_work work;
44b8bd7e
CM		 83};
84
4008c04a
CM		 85/* These are used to set the lockdep class on the extent buffer locks.
86 * The class is set by the readpage_end_io_hook after the buffer has
87 * passed csum validation but before the pages are unlocked.
88 *
89 * The lockdep class is also set by btrfs_init_new_buffer on freshly
90 * allocated blocks.
91 *
92 * The class is based on the level in the tree block, which allows lockdep
93 * to know that lower nodes nest inside the locks of higher nodes.
94 *
95 * We also add a check to make sure the highest level of the tree is
96 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
97 * code needs update as well.
98 */
99#ifdef CONFIG_DEBUG_LOCK_ALLOC
100# if BTRFS_MAX_LEVEL != 8
101# error
102# endif
103static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
104static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
105 /* leaf */
106 "btrfs-extent-00",
107 "btrfs-extent-01",
108 "btrfs-extent-02",
109 "btrfs-extent-03",
110 "btrfs-extent-04",
111 "btrfs-extent-05",
112 "btrfs-extent-06",
113 "btrfs-extent-07",
114 /* highest possible level */
115 "btrfs-extent-08",
116};
117#endif
118
d352ac68
CM		 119/*
120 * extents on the btree inode are pretty simple, there's one extent
121 * that covers the entire device
122 */
b2950863
CH		 123static struct extent_map *btree_get_extent(struct inode *inode,
124 struct page *page, size_t page_offset, u64 start, u64 len,
125 int create)
7eccb903 126{
5f39d397
CM		 127 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
128 struct extent_map *em;
129 int ret;
130
890871be 131 read_lock(&em_tree->lock);
d1310b2e 132 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 133 if (em) {
134 em->bdev =
135 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 136 read_unlock(&em_tree->lock);
5f39d397 137 goto out;
a061fc8d 138 }
890871be 139 read_unlock(&em_tree->lock);
7b13b7b1 140
5f39d397
CM		 141 em = alloc_extent_map(GFP_NOFS);
142 if (!em) {
143 em = ERR_PTR(-ENOMEM);
144 goto out;
145 }
146 em->start = 0;
0afbaf8c 147 em->len = (u64)-1;
c8b97818 148 em->block_len = (u64)-1;
5f39d397 149 em->block_start = 0;
a061fc8d 150 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 151
890871be 152 write_lock(&em_tree->lock);
5f39d397
CM		 153 ret = add_extent_mapping(em_tree, em);
154 if (ret == -EEXIST) {
0afbaf8c
CM		 155 u64 failed_start = em->start;
156 u64 failed_len = em->len;
157
5f39d397 158 free_extent_map(em);
7b13b7b1 159 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 160 if (em) {
7b13b7b1 161 ret = 0;
0afbaf8c
CM		 162 } else {
163 em = lookup_extent_mapping(em_tree, failed_start,
164 failed_len);
7b13b7b1 165 ret = -EIO;
0afbaf8c 166 }
5f39d397 167 } else if (ret) {
7b13b7b1
CM		 168 free_extent_map(em);
169 em = NULL;
5f39d397 170 }
890871be 171 write_unlock(&em_tree->lock);
7b13b7b1
CM		 172
173 if (ret)
174 em = ERR_PTR(ret);
5f39d397
CM		 175out:
176 return em;
7eccb903
CM		 177}
178
19c00ddc
CM		 179u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
180{
163e783e 181 return crc32c(seed, data, len);
19c00ddc
CM		 182}
183
184void btrfs_csum_final(u32 crc, char *result)
185{
186 *(__le32 *)result = ~cpu_to_le32(crc);
187}
188
d352ac68
CM		 189/*
190 * compute the csum for a btree block, and either verify it or write it
191 * into the csum field of the block.
192 */
19c00ddc
CM		 193static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
194 int verify)
195{
607d432d
JB		 196 u16 csum_size =
197 btrfs_super_csum_size(&root->fs_info->super_copy);
198 char *result = NULL;
19c00ddc
CM		 199 unsigned long len;
200 unsigned long cur_len;
201 unsigned long offset = BTRFS_CSUM_SIZE;
202 char *map_token = NULL;
203 char *kaddr;
204 unsigned long map_start;
205 unsigned long map_len;
206 int err;
207 u32 crc = ~(u32)0;
607d432d 208 unsigned long inline_result;
19c00ddc
CM		 209
210 len = buf->len - offset;
d397712b 211 while (len > 0) {
19c00ddc
CM		 212 err = map_private_extent_buffer(buf, offset, 32,
213 &map_token, &kaddr,
214 &map_start, &map_len, KM_USER0);
d397712b 215 if (err)
19c00ddc 216 return 1;
19c00ddc
CM		 217 cur_len = min(len, map_len - (offset - map_start));
218 crc = btrfs_csum_data(root, kaddr + offset - map_start,
219 crc, cur_len);
220 len -= cur_len;
221 offset += cur_len;
222 unmap_extent_buffer(buf, map_token, KM_USER0);
223 }
607d432d
JB		 224 if (csum_size > sizeof(inline_result)) {
225 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
226 if (!result)
227 return 1;
228 } else {
229 result = (char *)&inline_result;
230 }
231
19c00ddc
CM		 232 btrfs_csum_final(crc, result);
233
234 if (verify) {
607d432d 235 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 236 u32 val;
237 u32 found = 0;
607d432d 238 memcpy(&found, result, csum_size);
e4204ded 239
607d432d 240 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 241 if (printk_ratelimit()) {
242 printk(KERN_INFO "btrfs: %s checksum verify "
243 "failed on %llu wanted %X found %X "
244 "level %d\n",
245 root->fs_info->sb->s_id,
246 (unsigned long long)buf->start, val, found,
247 btrfs_header_level(buf));
248 }
607d432d
JB		 249 if (result != (char *)&inline_result)
250 kfree(result);
19c00ddc
CM		 251 return 1;
252 }
253 } else {
607d432d 254 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 255 }
607d432d
JB		 256 if (result != (char *)&inline_result)
257 kfree(result);
19c00ddc
CM		 258 return 0;
259}
260
d352ac68
CM		 261/*
262 * we can't consider a given block up to date unless the transid of the
263 * block matches the transid in the parent node's pointer. This is how we
264 * detect blocks that either didn't get written at all or got written
265 * in the wrong place.
266 */
1259ab75
CM		 267static int verify_parent_transid(struct extent_io_tree *io_tree,
268 struct extent_buffer *eb, u64 parent_transid)
269{
2ac55d41 270 struct extent_state *cached_state = NULL;
1259ab75
CM		 271 int ret;
272
273 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
274 return 0;
275
2ac55d41
JB		 276 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
277 0, &cached_state, GFP_NOFS);
278 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 279 btrfs_header_generation(eb) == parent_transid) {
280 ret = 0;
281 goto out;
282 }
193f284d
CM		 283 if (printk_ratelimit()) {
284 printk("parent transid verify failed on %llu wanted %llu "
285 "found %llu\n",
286 (unsigned long long)eb->start,
287 (unsigned long long)parent_transid,
288 (unsigned long long)btrfs_header_generation(eb));
289 }
1259ab75 290 ret = 1;
2ac55d41 291 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 292out:
2ac55d41
JB		 293 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
294 &cached_state, GFP_NOFS);
1259ab75 295 return ret;
1259ab75
CM		 296}
297
d352ac68
CM		 298/*
299 * helper to read a given tree block, doing retries as required when
300 * the checksums don't match and we have alternate mirrors to try.
301 */
f188591e
CM		 302static int btree_read_extent_buffer_pages(struct btrfs_root *root,
303 struct extent_buffer *eb,
ca7a79ad 304 u64 start, u64 parent_transid)
f188591e
CM		 305{
306 struct extent_io_tree *io_tree;
307 int ret;
308 int num_copies = 0;
309 int mirror_num = 0;
310
311 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
312 while (1) {
313 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
314 btree_get_extent, mirror_num);
1259ab75
CM		 315 if (!ret &&
316 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 317 return ret;
d397712b 318
f188591e
CM		 319 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
320 eb->start, eb->len);
4235298e 321 if (num_copies == 1)
f188591e 322 return ret;
4235298e 323
f188591e 324 mirror_num++;
4235298e 325 if (mirror_num > num_copies)
f188591e 326 return ret;
f188591e 327 }
f188591e
CM		 328 return -EIO;
329}
19c00ddc 330
d352ac68 331/*
d397712b
CM		 332 * checksum a dirty tree block before IO. This has extra checks to make sure
333 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 334 */
d397712b 335
b2950863 336static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 337{
d1310b2e 338 struct extent_io_tree *tree;
35ebb934 339 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 340 u64 found_start;
341 int found_level;
342 unsigned long len;
343 struct extent_buffer *eb;
f188591e
CM		 344 int ret;
345
d1310b2e 346 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 347
348 if (page->private == EXTENT_PAGE_PRIVATE)
349 goto out;
350 if (!page->private)
351 goto out;
352 len = page->private >> 2;
d397712b
CM		 353 WARN_ON(len == 0);
354
19c00ddc 355 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 356 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
357 btrfs_header_generation(eb));
f188591e 358 BUG_ON(ret);
19c00ddc
CM		 359 found_start = btrfs_header_bytenr(eb);
360 if (found_start != start) {
55c69072
CM		 361 WARN_ON(1);
362 goto err;
363 }
364 if (eb->first_page != page) {
55c69072
CM		 365 WARN_ON(1);
366 goto err;
367 }
368 if (!PageUptodate(page)) {
55c69072
CM		 369 WARN_ON(1);
370 goto err;
19c00ddc
CM		 371 }
372 found_level = btrfs_header_level(eb);
4bef0848 373
19c00ddc 374 csum_tree_block(root, eb, 0);
55c69072 375err:
19c00ddc
CM		 376 free_extent_buffer(eb);
377out:
378 return 0;
379}
380
2b82032c
YZ		 381static int check_tree_block_fsid(struct btrfs_root *root,
382 struct extent_buffer *eb)
383{
384 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
385 u8 fsid[BTRFS_UUID_SIZE];
386 int ret = 1;
387
388 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
389 BTRFS_FSID_SIZE);
390 while (fs_devices) {
391 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
392 ret = 0;
393 break;
394 }
395 fs_devices = fs_devices->seed;
396 }
397 return ret;
398}
399
4008c04a
CM		 400#ifdef CONFIG_DEBUG_LOCK_ALLOC
401void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
402{
403 lockdep_set_class_and_name(&eb->lock,
404 &btrfs_eb_class[level],
405 btrfs_eb_name[level]);
406}
407#endif
408
b2950863 409static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 410 struct extent_state *state)
411{
412 struct extent_io_tree *tree;
413 u64 found_start;
414 int found_level;
415 unsigned long len;
416 struct extent_buffer *eb;
417 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 418 int ret = 0;
ce9adaa5
CM		 419
420 tree = &BTRFS_I(page->mapping->host)->io_tree;
421 if (page->private == EXTENT_PAGE_PRIVATE)
422 goto out;
423 if (!page->private)
424 goto out;
d397712b 425
ce9adaa5 426 len = page->private >> 2;
d397712b
CM		 427 WARN_ON(len == 0);
428
ce9adaa5 429 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 430
ce9adaa5 431 found_start = btrfs_header_bytenr(eb);
23a07867 432 if (found_start != start) {
193f284d
CM		 433 if (printk_ratelimit()) {
434 printk(KERN_INFO "btrfs bad tree block start "
435 "%llu %llu\n",
436 (unsigned long long)found_start,
437 (unsigned long long)eb->start);
438 }
f188591e 439 ret = -EIO;
ce9adaa5
CM		 440 goto err;
441 }
442 if (eb->first_page != page) {
d397712b
CM		 443 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
444 eb->first_page->index, page->index);
ce9adaa5 445 WARN_ON(1);
f188591e 446 ret = -EIO;
ce9adaa5
CM		 447 goto err;
448 }
2b82032c 449 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 450 if (printk_ratelimit()) {
451 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
452 (unsigned long long)eb->start);
453 }
1259ab75
CM		 454 ret = -EIO;
455 goto err;
456 }
ce9adaa5
CM		 457 found_level = btrfs_header_level(eb);
458
4008c04a
CM		 459 btrfs_set_buffer_lockdep_class(eb, found_level);
460
ce9adaa5 461 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 462 if (ret)
463 ret = -EIO;
ce9adaa5
CM		 464
465 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
466 end = eb->start + end - 1;
ce9adaa5
CM		 467err:
468 free_extent_buffer(eb);
469out:
f188591e 470 return ret;
ce9adaa5
CM		 471}
472
ce9adaa5 473static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 474{
475 struct end_io_wq *end_io_wq = bio->bi_private;
476 struct btrfs_fs_info *fs_info;
ce9adaa5 477
ce9adaa5 478 fs_info = end_io_wq->info;
ce9adaa5 479 end_io_wq->error = err;
8b712842
CM		 480 end_io_wq->work.func = end_workqueue_fn;
481 end_io_wq->work.flags = 0;
d20f7043 482
7b6d91da 483 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 484 if (end_io_wq->metadata == 1)
cad321ad
CM		 485 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
486 &end_io_wq->work);
0cb59c99
JB		 487 else if (end_io_wq->metadata == 2)
488 btrfs_queue_worker(&fs_info->endio_freespace_worker,
489 &end_io_wq->work);
cad321ad
CM		 490 else
491 btrfs_queue_worker(&fs_info->endio_write_workers,
492 &end_io_wq->work);
d20f7043
CM		 493 } else {
494 if (end_io_wq->metadata)
495 btrfs_queue_worker(&fs_info->endio_meta_workers,
496 &end_io_wq->work);
497 else
498 btrfs_queue_worker(&fs_info->endio_workers,
499 &end_io_wq->work);
500 }
ce9adaa5
CM		 501}
502
0cb59c99
JB		 503/*
504 * For the metadata arg you want
505 *
506 * 0 - if data
507 * 1 - if normal metadta
508 * 2 - if writing to the free space cache area
509 */
22c59948
CM		 510int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
511 int metadata)
0b86a832 512{
ce9adaa5 513 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 514 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
515 if (!end_io_wq)
516 return -ENOMEM;
517
518 end_io_wq->private = bio->bi_private;
519 end_io_wq->end_io = bio->bi_end_io;
22c59948 520 end_io_wq->info = info;
ce9adaa5
CM		 521 end_io_wq->error = 0;
522 end_io_wq->bio = bio;
22c59948 523 end_io_wq->metadata = metadata;
ce9adaa5
CM		 524
525 bio->bi_private = end_io_wq;
526 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 527 return 0;
528}
529
b64a2851 530unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 531{
4854ddd0
CM		 532 unsigned long limit = min_t(unsigned long,
533 info->workers.max_workers,
534 info->fs_devices->open_devices);
535 return 256 * limit;
536}
0986fe9e 537
4854ddd0
CM		 538int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
539{
b64a2851
CM		 540 return atomic_read(&info->nr_async_bios) >
541 btrfs_async_submit_limit(info);
0986fe9e
CM		 542}
543
4a69a410
CM		 544static void run_one_async_start(struct btrfs_work *work)
545{
546 struct btrfs_fs_info *fs_info;
547 struct async_submit_bio *async;
548
549 async = container_of(work, struct async_submit_bio, work);
550 fs_info = BTRFS_I(async->inode)->root->fs_info;
551 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM		 552 async->mirror_num, async->bio_flags,
553 async->bio_offset);
4a69a410
CM		 554}
555
556static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 557{
558 struct btrfs_fs_info *fs_info;
559 struct async_submit_bio *async;
4854ddd0 560 int limit;
8b712842
CM		 561
562 async = container_of(work, struct async_submit_bio, work);
563 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 564
b64a2851 565 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 566 limit = limit * 2 / 3;
567
8b712842 568 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 569
b64a2851
CM		 570 if (atomic_read(&fs_info->nr_async_submits) < limit &&
571 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 572 wake_up(&fs_info->async_submit_wait);
573
4a69a410 574 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 575 async->mirror_num, async->bio_flags,
576 async->bio_offset);
4a69a410
CM		 577}
578
579static void run_one_async_free(struct btrfs_work *work)
580{
581 struct async_submit_bio *async;
582
583 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 584 kfree(async);
585}
586
44b8bd7e
CM		 587int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
588 int rw, struct bio *bio, int mirror_num,
c8b97818 589 unsigned long bio_flags,
eaf25d93 590 u64 bio_offset,
4a69a410
CM		 591 extent_submit_bio_hook_t *submit_bio_start,
592 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 593{
594 struct async_submit_bio *async;
595
596 async = kmalloc(sizeof(*async), GFP_NOFS);
597 if (!async)
598 return -ENOMEM;
599
600 async->inode = inode;
601 async->rw = rw;
602 async->bio = bio;
603 async->mirror_num = mirror_num;
4a69a410
CM		 604 async->submit_bio_start = submit_bio_start;
605 async->submit_bio_done = submit_bio_done;
606
607 async->work.func = run_one_async_start;
608 async->work.ordered_func = run_one_async_done;
609 async->work.ordered_free = run_one_async_free;
610
8b712842 611 async->work.flags = 0;
c8b97818 612 async->bio_flags = bio_flags;
eaf25d93 613 async->bio_offset = bio_offset;
8c8bee1d 614
cb03c743 615 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 616
7b6d91da 617 if (rw & REQ_SYNC)
d313d7a3
CM		 618 btrfs_set_work_high_prio(&async->work);
619
8b712842 620 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 621
d397712b 622 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 623 atomic_read(&fs_info->nr_async_submits)) {
624 wait_event(fs_info->async_submit_wait,
625 (atomic_read(&fs_info->nr_async_submits) == 0));
626 }
627
44b8bd7e
CM		 628 return 0;
629}
630
ce3ed71a
CM		 631static int btree_csum_one_bio(struct bio *bio)
632{
633 struct bio_vec *bvec = bio->bi_io_vec;
634 int bio_index = 0;
635 struct btrfs_root *root;
636
637 WARN_ON(bio->bi_vcnt <= 0);
d397712b 638 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 639 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
640 csum_dirty_buffer(root, bvec->bv_page);
641 bio_index++;
642 bvec++;
643 }
644 return 0;
645}
646
4a69a410
CM		 647static int __btree_submit_bio_start(struct inode *inode, int rw,
648 struct bio *bio, int mirror_num,
eaf25d93
CM		 649 unsigned long bio_flags,
650 u64 bio_offset)
22c59948 651{
8b712842
CM		 652 /*
653 * when we're called for a write, we're already in the async
5443be45 654 * submission context. Just jump into btrfs_map_bio
8b712842 655 */
4a69a410
CM		 656 btree_csum_one_bio(bio);
657 return 0;
658}
22c59948 659
4a69a410 660static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 661 int mirror_num, unsigned long bio_flags,
662 u64 bio_offset)
4a69a410 663{
8b712842 664 /*
4a69a410
CM		 665 * when we're called for a write, we're already in the async
666 * submission context. Just jump into btrfs_map_bio
8b712842 667 */
8b712842 668 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 669}
670
44b8bd7e 671static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 672 int mirror_num, unsigned long bio_flags,
673 u64 bio_offset)
44b8bd7e 674{
cad321ad
CM		 675 int ret;
676
677 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
678 bio, 1);
679 BUG_ON(ret);
680
7b6d91da 681 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 682 /*
683 * called for a read, do the setup so that checksum validation
684 * can happen in the async kernel threads
685 */
4a69a410 686 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 687 mirror_num, 0);
44b8bd7e 688 }
d313d7a3 689
cad321ad
CM		 690 /*
691 * kthread helpers are used to submit writes so that checksumming
692 * can happen in parallel across all CPUs
693 */
44b8bd7e 694 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 695 inode, rw, bio, mirror_num, 0,
eaf25d93 696 bio_offset,
4a69a410
CM		 697 __btree_submit_bio_start,
698 __btree_submit_bio_done);
44b8bd7e
CM		 699}
700
0da5468f
CM		 701static int btree_writepage(struct page *page, struct writeback_control *wbc)
702{
d1310b2e 703 struct extent_io_tree *tree;
b9473439
CM		 704 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
705 struct extent_buffer *eb;
706 int was_dirty;
707
d1310b2e 708 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 709 if (!(current->flags & PF_MEMALLOC)) {
710 return extent_write_full_page(tree, page,
711 btree_get_extent, wbc);
712 }
5443be45 713
b9473439
CM		 714 redirty_page_for_writepage(wbc, page);
715 eb = btrfs_find_tree_block(root, page_offset(page),
716 PAGE_CACHE_SIZE);
717 WARN_ON(!eb);
718
719 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
720 if (!was_dirty) {
721 spin_lock(&root->fs_info->delalloc_lock);
722 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
723 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 724 }
b9473439
CM		 725 free_extent_buffer(eb);
726
727 unlock_page(page);
728 return 0;
5f39d397 729}
0da5468f
CM		 730
731static int btree_writepages(struct address_space *mapping,
732 struct writeback_control *wbc)
733{
d1310b2e
CM		 734 struct extent_io_tree *tree;
735 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 736 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 737 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 738 u64 num_dirty;
24ab9cd8 739 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 740
741 if (wbc->for_kupdate)
742 return 0;
743
b9473439
CM		 744 /* this is a bit racy, but that's ok */
745 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 746 if (num_dirty < thresh)
793955bc 747 return 0;
793955bc 748 }
0da5468f
CM		 749 return extent_writepages(tree, mapping, btree_get_extent, wbc);
750}
751
b2950863 752static int btree_readpage(struct file *file, struct page *page)
5f39d397 753{
d1310b2e
CM		 754 struct extent_io_tree *tree;
755 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 756 return extent_read_full_page(tree, page, btree_get_extent);
757}
22b0ebda 758
70dec807 759static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 760{
d1310b2e
CM		 761 struct extent_io_tree *tree;
762 struct extent_map_tree *map;
5f39d397 763 int ret;
d98237b3 764
98509cfc 765 if (PageWriteback(page) || PageDirty(page))
d397712b 766 return 0;
98509cfc 767
d1310b2e
CM		 768 tree = &BTRFS_I(page->mapping->host)->io_tree;
769 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 770
7b13b7b1 771 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 772 if (!ret)
6af118ce 773 return 0;
6af118ce
CM		 774
775 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 776 if (ret == 1) {
777 ClearPagePrivate(page);
778 set_page_private(page, 0);
779 page_cache_release(page);
780 }
6af118ce 781
d98237b3
CM		 782 return ret;
783}
784
5f39d397 785static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 786{
d1310b2e
CM		 787 struct extent_io_tree *tree;
788 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 789 extent_invalidatepage(tree, page, offset);
790 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 791 if (PagePrivate(page)) {
d397712b
CM		 792 printk(KERN_WARNING "btrfs warning page private not zero "
793 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 794 ClearPagePrivate(page);
795 set_page_private(page, 0);
796 page_cache_release(page);
797 }
d98237b3
CM		 798}
799
7f09410b 800static const struct address_space_operations btree_aops = {
d98237b3
CM		 801 .readpage = btree_readpage,
802 .writepage = btree_writepage,
0da5468f 803 .writepages = btree_writepages,
5f39d397
CM		 804 .releasepage = btree_releasepage,
805 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 806 .sync_page = block_sync_page,
807};
808
ca7a79ad
CM		 809int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
810 u64 parent_transid)
090d1875 811{
5f39d397
CM		 812 struct extent_buffer *buf = NULL;
813 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 814 int ret = 0;
090d1875 815
db94535d 816 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 817 if (!buf)
090d1875 818 return 0;
d1310b2e 819 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 820 buf, 0, 0, btree_get_extent, 0);
5f39d397 821 free_extent_buffer(buf);
de428b63 822 return ret;
090d1875
CM		 823}
824
0999df54
CM		 825struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
826 u64 bytenr, u32 blocksize)
827{
828 struct inode *btree_inode = root->fs_info->btree_inode;
829 struct extent_buffer *eb;
830 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
831 bytenr, blocksize, GFP_NOFS);
832 return eb;
833}
834
835struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
836 u64 bytenr, u32 blocksize)
837{
838 struct inode *btree_inode = root->fs_info->btree_inode;
839 struct extent_buffer *eb;
840
841 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
842 bytenr, blocksize, NULL, GFP_NOFS);
843 return eb;
844}
845
846
e02119d5
CM		 847int btrfs_write_tree_block(struct extent_buffer *buf)
848{
8aa38c31
CH		 849 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
850 buf->start + buf->len - 1);
e02119d5
CM		 851}
852
853int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
854{
8aa38c31
CH		 855 return filemap_fdatawait_range(buf->first_page->mapping,
856 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 857}
858
0999df54 859struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 860 u32 blocksize, u64 parent_transid)
0999df54
CM		 861{
862 struct extent_buffer *buf = NULL;
863 struct inode *btree_inode = root->fs_info->btree_inode;
864 struct extent_io_tree *io_tree;
865 int ret;
866
867 io_tree = &BTRFS_I(btree_inode)->io_tree;
868
869 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
870 if (!buf)
871 return NULL;
0999df54 872
ca7a79ad 873 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 874
d397712b 875 if (ret == 0)
b4ce94de 876 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 877 return buf;
ce9adaa5 878
eb60ceac
CM		 879}
880
e089f05c 881int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 882 struct extent_buffer *buf)
ed2ff2cb 883{
5f39d397 884 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 885 if (btrfs_header_generation(buf) ==
925baedd 886 root->fs_info->running_transaction->transid) {
b9447ef8 887 btrfs_assert_tree_locked(buf);
b4ce94de 888
b9473439
CM		 889 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
890 spin_lock(&root->fs_info->delalloc_lock);
891 if (root->fs_info->dirty_metadata_bytes >= buf->len)
892 root->fs_info->dirty_metadata_bytes -= buf->len;
893 else
894 WARN_ON(1);
895 spin_unlock(&root->fs_info->delalloc_lock);
896 }
b4ce94de 897
b9473439
CM		 898 /* ugh, clear_extent_buffer_dirty needs to lock the page */
899 btrfs_set_lock_blocking(buf);
d1310b2e 900 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 901 buf);
925baedd 902 }
5f39d397
CM		 903 return 0;
904}
905
db94535d 906static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 907 u32 stripesize, struct btrfs_root *root,
9f5fae2f 908 struct btrfs_fs_info *fs_info,
e20d96d6 909 u64 objectid)
d97e63b6 910{
cfaa7295 911 root->node = NULL;
a28ec197 912 root->commit_root = NULL;
db94535d
CM		 913 root->sectorsize = sectorsize;
914 root->nodesize = nodesize;
915 root->leafsize = leafsize;
87ee04eb 916 root->stripesize = stripesize;
123abc88 917 root->ref_cows = 0;
0b86a832 918 root->track_dirty = 0;
c71bf099 919 root->in_radix = 0;
d68fc57b
YZ		 920 root->orphan_item_inserted = 0;
921 root->orphan_cleanup_state = 0;
0b86a832 922
9f5fae2f 923 root->fs_info = fs_info;
0f7d52f4
CM		 924 root->objectid = objectid;
925 root->last_trans = 0;
13a8a7c8 926 root->highest_objectid = 0;
58176a96 927 root->name = NULL;
4313b399 928 root->in_sysfs = 0;
6bef4d31 929 root->inode_tree = RB_ROOT;
f0486c68 930 root->block_rsv = NULL;
d68fc57b 931 root->orphan_block_rsv = NULL;
0b86a832
CM		 932
933 INIT_LIST_HEAD(&root->dirty_list);
7b128766 934 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 935 INIT_LIST_HEAD(&root->root_list);
925baedd 936 spin_lock_init(&root->node_lock);
d68fc57b 937 spin_lock_init(&root->orphan_lock);
5d4f98a2 938 spin_lock_init(&root->inode_lock);
f0486c68 939 spin_lock_init(&root->accounting_lock);
a2135011 940 mutex_init(&root->objectid_mutex);
e02119d5 941 mutex_init(&root->log_mutex);
7237f183
YZ		 942 init_waitqueue_head(&root->log_writer_wait);
943 init_waitqueue_head(&root->log_commit_wait[0]);
944 init_waitqueue_head(&root->log_commit_wait[1]);
945 atomic_set(&root->log_commit[0], 0);
946 atomic_set(&root->log_commit[1], 0);
947 atomic_set(&root->log_writers, 0);
948 root->log_batch = 0;
949 root->log_transid = 0;
257c62e1 950 root->last_log_commit = 0;
d0c803c4
CM		 951 extent_io_tree_init(&root->dirty_log_pages,
952 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 953
3768f368
CM		 954 memset(&root->root_key, 0, sizeof(root->root_key));
955 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 956 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 957 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 958 root->defrag_trans_start = fs_info->generation;
58176a96 959 init_completion(&root->kobj_unregister);
6702ed49 960 root->defrag_running = 0;
4d775673 961 root->root_key.objectid = objectid;
3394e160
CM		 962 root->anon_super.s_root = NULL;
963 root->anon_super.s_dev = 0;
964 INIT_LIST_HEAD(&root->anon_super.s_list);
965 INIT_LIST_HEAD(&root->anon_super.s_instances);
966 init_rwsem(&root->anon_super.s_umount);
967
3768f368
CM		 968 return 0;
969}
970
db94535d 971static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 972 struct btrfs_fs_info *fs_info,
973 u64 objectid,
e20d96d6 974 struct btrfs_root *root)
3768f368
CM		 975{
976 int ret;
db94535d 977 u32 blocksize;
84234f3a 978 u64 generation;
3768f368 979
db94535d 980 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 981 tree_root->sectorsize, tree_root->stripesize,
982 root, fs_info, objectid);
3768f368
CM		 983 ret = btrfs_find_last_root(tree_root, objectid,
984 &root->root_item, &root->root_key);
4df27c4d
YZ		 985 if (ret > 0)
986 return -ENOENT;
3768f368
CM		 987 BUG_ON(ret);
988
84234f3a 989 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 990 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
991 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 992 blocksize, generation);
3768f368 993 BUG_ON(!root->node);
4df27c4d 994 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 995 return 0;
996}
997
7237f183
YZ		 998static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
999 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1000{
1001 struct btrfs_root *root;
1002 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1003 struct extent_buffer *leaf;
e02119d5
CM		 1004
1005 root = kzalloc(sizeof(*root), GFP_NOFS);
1006 if (!root)
7237f183 1007 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1008
1009 __setup_root(tree_root->nodesize, tree_root->leafsize,
1010 tree_root->sectorsize, tree_root->stripesize,
1011 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1012
1013 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1014 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1015 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1016 /*
1017 * log trees do not get reference counted because they go away
1018 * before a real commit is actually done. They do store pointers
1019 * to file data extents, and those reference counts still get
1020 * updated (along with back refs to the log tree).
1021 */
e02119d5
CM		 1022 root->ref_cows = 0;
1023
5d4f98a2
YZ		 1024 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1025 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1026 if (IS_ERR(leaf)) {
1027 kfree(root);
1028 return ERR_CAST(leaf);
1029 }
e02119d5 1030
5d4f98a2
YZ		 1031 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1032 btrfs_set_header_bytenr(leaf, leaf->start);
1033 btrfs_set_header_generation(leaf, trans->transid);
1034 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1035 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1036 root->node = leaf;
e02119d5
CM		 1037
1038 write_extent_buffer(root->node, root->fs_info->fsid,
1039 (unsigned long)btrfs_header_fsid(root->node),
1040 BTRFS_FSID_SIZE);
1041 btrfs_mark_buffer_dirty(root->node);
1042 btrfs_tree_unlock(root->node);
7237f183
YZ		 1043 return root;
1044}
1045
1046int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1047 struct btrfs_fs_info *fs_info)
1048{
1049 struct btrfs_root *log_root;
1050
1051 log_root = alloc_log_tree(trans, fs_info);
1052 if (IS_ERR(log_root))
1053 return PTR_ERR(log_root);
1054 WARN_ON(fs_info->log_root_tree);
1055 fs_info->log_root_tree = log_root;
1056 return 0;
1057}
1058
1059int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1060 struct btrfs_root *root)
1061{
1062 struct btrfs_root *log_root;
1063 struct btrfs_inode_item *inode_item;
1064
1065 log_root = alloc_log_tree(trans, root->fs_info);
1066 if (IS_ERR(log_root))
1067 return PTR_ERR(log_root);
1068
1069 log_root->last_trans = trans->transid;
1070 log_root->root_key.offset = root->root_key.objectid;
1071
1072 inode_item = &log_root->root_item.inode;
1073 inode_item->generation = cpu_to_le64(1);
1074 inode_item->size = cpu_to_le64(3);
1075 inode_item->nlink = cpu_to_le32(1);
1076 inode_item->nbytes = cpu_to_le64(root->leafsize);
1077 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1078
5d4f98a2 1079 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1080
1081 WARN_ON(root->log_root);
1082 root->log_root = log_root;
1083 root->log_transid = 0;
257c62e1 1084 root->last_log_commit = 0;
e02119d5
CM		 1085 return 0;
1086}
1087
1088struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1089 struct btrfs_key *location)
1090{
1091 struct btrfs_root *root;
1092 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1093 struct btrfs_path *path;
5f39d397 1094 struct extent_buffer *l;
84234f3a 1095 u64 generation;
db94535d 1096 u32 blocksize;
0f7d52f4
CM		 1097 int ret = 0;
1098
5eda7b5e 1099 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1100 if (!root)
0f7d52f4 1101 return ERR_PTR(-ENOMEM);
0f7d52f4 1102 if (location->offset == (u64)-1) {
db94535d 1103 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1104 location->objectid, root);
1105 if (ret) {
0f7d52f4
CM		 1106 kfree(root);
1107 return ERR_PTR(ret);
1108 }
13a8a7c8 1109 goto out;
0f7d52f4
CM		 1110 }
1111
db94535d 1112 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1113 tree_root->sectorsize, tree_root->stripesize,
1114 root, fs_info, location->objectid);
0f7d52f4
CM		 1115
1116 path = btrfs_alloc_path();
1117 BUG_ON(!path);
1118 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1119 if (ret == 0) {
1120 l = path->nodes[0];
1121 read_extent_buffer(l, &root->root_item,
1122 btrfs_item_ptr_offset(l, path->slots[0]),
1123 sizeof(root->root_item));
1124 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1125 }
0f7d52f4
CM		 1126 btrfs_free_path(path);
1127 if (ret) {
13a8a7c8
YZ		 1128 if (ret > 0)
1129 ret = -ENOENT;
0f7d52f4
CM		 1130 return ERR_PTR(ret);
1131 }
13a8a7c8 1132
84234f3a 1133 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1134 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1135 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1136 blocksize, generation);
5d4f98a2 1137 root->commit_root = btrfs_root_node(root);
0f7d52f4 1138 BUG_ON(!root->node);
13a8a7c8
YZ		 1139out:
1140 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1141 root->ref_cows = 1;
13a8a7c8 1142
5eda7b5e
CM		 1143 return root;
1144}
1145
dc17ff8f
CM		 1146struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1147 u64 root_objectid)
1148{
1149 struct btrfs_root *root;
1150
1151 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1152 return fs_info->tree_root;
1153 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1154 return fs_info->extent_root;
1155
1156 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1157 (unsigned long)root_objectid);
1158 return root;
1159}
1160
edbd8d4e
CM		 1161struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1162 struct btrfs_key *location)
5eda7b5e
CM		 1163{
1164 struct btrfs_root *root;
1165 int ret;
1166
edbd8d4e
CM		 1167 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1168 return fs_info->tree_root;
1169 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1170 return fs_info->extent_root;
8f18cf13
CM		 1171 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1172 return fs_info->chunk_root;
1173 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1174 return fs_info->dev_root;
0403e47e
YZ		 1175 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1176 return fs_info->csum_root;
4df27c4d
YZ		 1177again:
1178 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1179 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1180 (unsigned long)location->objectid);
4df27c4d 1181 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1182 if (root)
1183 return root;
1184
e02119d5 1185 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1186 if (IS_ERR(root))
1187 return root;
3394e160
CM		 1188
1189 set_anon_super(&root->anon_super, NULL);
1190
d68fc57b
YZ		 1191 if (btrfs_root_refs(&root->root_item) == 0) {
1192 ret = -ENOENT;
1193 goto fail;
1194 }
1195
1196 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1197 if (ret < 0)
1198 goto fail;
1199 if (ret == 0)
1200 root->orphan_item_inserted = 1;
1201
4df27c4d
YZ		 1202 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1203 if (ret)
1204 goto fail;
1205
1206 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1207 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1208 (unsigned long)root->root_key.objectid,
0f7d52f4 1209 root);
d68fc57b 1210 if (ret == 0)
4df27c4d 1211 root->in_radix = 1;
d68fc57b 1212
4df27c4d
YZ		 1213 spin_unlock(&fs_info->fs_roots_radix_lock);
1214 radix_tree_preload_end();
0f7d52f4 1215 if (ret) {
4df27c4d
YZ		 1216 if (ret == -EEXIST) {
1217 free_fs_root(root);
1218 goto again;
1219 }
1220 goto fail;
0f7d52f4 1221 }
4df27c4d
YZ		 1222
1223 ret = btrfs_find_dead_roots(fs_info->tree_root,
1224 root->root_key.objectid);
1225 WARN_ON(ret);
edbd8d4e 1226 return root;
4df27c4d
YZ		 1227fail:
1228 free_fs_root(root);
1229 return ERR_PTR(ret);
edbd8d4e
CM		 1230}
1231
1232struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1233 struct btrfs_key *location,
1234 const char *name, int namelen)
1235{
4df27c4d
YZ		 1236 return btrfs_read_fs_root_no_name(fs_info, location);
1237#if 0
edbd8d4e
CM		 1238 struct btrfs_root *root;
1239 int ret;
1240
1241 root = btrfs_read_fs_root_no_name(fs_info, location);
1242 if (!root)
1243 return NULL;
58176a96 1244
4313b399
CM		 1245 if (root->in_sysfs)
1246 return root;
1247
58176a96
JB		 1248 ret = btrfs_set_root_name(root, name, namelen);
1249 if (ret) {
5f39d397 1250 free_extent_buffer(root->node);
58176a96
JB		 1251 kfree(root);
1252 return ERR_PTR(ret);
1253 }
4df27c4d 1254
58176a96
JB		 1255 ret = btrfs_sysfs_add_root(root);
1256 if (ret) {
5f39d397 1257 free_extent_buffer(root->node);
58176a96
JB		 1258 kfree(root->name);
1259 kfree(root);
1260 return ERR_PTR(ret);
1261 }
4313b399 1262 root->in_sysfs = 1;
0f7d52f4 1263 return root;
4df27c4d 1264#endif
0f7d52f4 1265}
04160088
CM		 1266
1267static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1268{
1269 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1270 int ret = 0;
04160088
CM		 1271 struct btrfs_device *device;
1272 struct backing_dev_info *bdi;
b7967db7 1273
c6e30871 1274 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1275 if (!device->bdev)
1276 continue;
04160088
CM		 1277 bdi = blk_get_backing_dev_info(device->bdev);
1278 if (bdi && bdi_congested(bdi, bdi_bits)) {
1279 ret = 1;
1280 break;
1281 }
1282 }
1283 return ret;
1284}
1285
38b66988
CM		 1286/*
1287 * this unplugs every device on the box, and it is only used when page
1288 * is null
1289 */
1290static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1291{
38b66988
CM		 1292 struct btrfs_device *device;
1293 struct btrfs_fs_info *info;
1294
1295 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1296 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1297 if (!device->bdev)
1298 continue;
1299
38b66988 1300 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1301 if (bdi->unplug_io_fn)
38b66988 1302 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1303 }
1304}
1305
b2950863 1306static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1307{
38b66988 1308 struct inode *inode;
f2d8d74d
CM		 1309 struct extent_map_tree *em_tree;
1310 struct extent_map *em;
bcbfce8a 1311 struct address_space *mapping;
38b66988
CM		 1312 u64 offset;
1313
bcbfce8a 1314 /* the generic O_DIRECT read code does this */
9f0ba5bd 1315 if (1 || !page) {
38b66988
CM		 1316 __unplug_io_fn(bdi, page);
1317 return;
1318 }
1319
bcbfce8a
CM		 1320 /*
1321 * page->mapping may change at any time. Get a consistent copy
1322 * and use that for everything below
1323 */
1324 smp_mb();
1325 mapping = page->mapping;
1326 if (!mapping)
1327 return;
1328
1329 inode = mapping->host;
240d5d48
CM		 1330
1331 /*
1332 * don't do the expensive searching for a small number of
1333 * devices
1334 */
1335 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1336 __unplug_io_fn(bdi, page);
1337 return;
1338 }
1339
38b66988 1340 offset = page_offset(page);
04160088 1341
f2d8d74d 1342 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1343 read_lock(&em_tree->lock);
f2d8d74d 1344 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1345 read_unlock(&em_tree->lock);
89642229
CM		 1346 if (!em) {
1347 __unplug_io_fn(bdi, page);
f2d8d74d 1348 return;
89642229 1349 }
f2d8d74d 1350
89642229
CM		 1351 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1352 free_extent_map(em);
1353 __unplug_io_fn(bdi, page);
1354 return;
1355 }
f2d8d74d
CM		 1356 offset = offset - em->start;
1357 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1358 em->block_start + offset, page);
1359 free_extent_map(em);
04160088
CM		 1360}
1361
ad081f14
JA		 1362/*
1363 * If this fails, caller must call bdi_destroy() to get rid of the
1364 * bdi again.
1365 */
04160088
CM		 1366static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1367{
ad081f14
JA		 1368 int err;
1369
1370 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1371 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1372 if (err)
1373 return err;
1374
4575c9cc 1375 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1376 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1377 bdi->unplug_io_data = info;
1378 bdi->congested_fn = btrfs_congested_fn;
1379 bdi->congested_data = info;
1380 return 0;
1381}
1382
ce9adaa5
CM		 1383static int bio_ready_for_csum(struct bio *bio)
1384{
1385 u64 length = 0;
1386 u64 buf_len = 0;
1387 u64 start = 0;
1388 struct page *page;
1389 struct extent_io_tree *io_tree = NULL;
1390 struct btrfs_fs_info *info = NULL;
1391 struct bio_vec *bvec;
1392 int i;
1393 int ret;
1394
1395 bio_for_each_segment(bvec, bio, i) {
1396 page = bvec->bv_page;
1397 if (page->private == EXTENT_PAGE_PRIVATE) {
1398 length += bvec->bv_len;
1399 continue;
1400 }
1401 if (!page->private) {
1402 length += bvec->bv_len;
1403 continue;
1404 }
1405 length = bvec->bv_len;
1406 buf_len = page->private >> 2;
1407 start = page_offset(page) + bvec->bv_offset;
1408 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1409 info = BTRFS_I(page->mapping->host)->root->fs_info;
1410 }
1411 /* are we fully contained in this bio? */
1412 if (buf_len <= length)
1413 return 1;
1414
1415 ret = extent_range_uptodate(io_tree, start + length,
1416 start + buf_len - 1);
ce9adaa5
CM		 1417 return ret;
1418}
1419
8b712842
CM		 1420/*
1421 * called by the kthread helper functions to finally call the bio end_io
1422 * functions. This is where read checksum verification actually happens
1423 */
1424static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1425{
ce9adaa5 1426 struct bio *bio;
8b712842
CM		 1427 struct end_io_wq *end_io_wq;
1428 struct btrfs_fs_info *fs_info;
ce9adaa5 1429 int error;
ce9adaa5 1430
8b712842
CM		 1431 end_io_wq = container_of(work, struct end_io_wq, work);
1432 bio = end_io_wq->bio;
1433 fs_info = end_io_wq->info;
ce9adaa5 1434
cad321ad 1435 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1436 * be checksummed at once. This makes sure the entire block is in
1437 * ram and up to date before trying to verify things. For
1438 * blocksize <= pagesize, it is basically a noop
1439 */
7b6d91da 1440 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1441 !bio_ready_for_csum(bio)) {
d20f7043 1442 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1443 &end_io_wq->work);
1444 return;
1445 }
1446 error = end_io_wq->error;
1447 bio->bi_private = end_io_wq->private;
1448 bio->bi_end_io = end_io_wq->end_io;
1449 kfree(end_io_wq);
8b712842 1450 bio_endio(bio, error);
44b8bd7e
CM		 1451}
1452
a74a4b97
CM		 1453static int cleaner_kthread(void *arg)
1454{
1455 struct btrfs_root *root = arg;
1456
1457 do {
a74a4b97 1458 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1459
1460 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1461 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1462 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1463 btrfs_clean_old_snapshots(root);
1464 mutex_unlock(&root->fs_info->cleaner_mutex);
1465 }
a74a4b97
CM		 1466
1467 if (freezing(current)) {
1468 refrigerator();
1469 } else {
a74a4b97 1470 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1471 if (!kthread_should_stop())
1472 schedule();
a74a4b97
CM		 1473 __set_current_state(TASK_RUNNING);
1474 }
1475 } while (!kthread_should_stop());
1476 return 0;
1477}
1478
1479static int transaction_kthread(void *arg)
1480{
1481 struct btrfs_root *root = arg;
1482 struct btrfs_trans_handle *trans;
1483 struct btrfs_transaction *cur;
8929ecfa 1484 u64 transid;
a74a4b97
CM		 1485 unsigned long now;
1486 unsigned long delay;
1487 int ret;
1488
1489 do {
a74a4b97
CM		 1490 delay = HZ * 30;
1491 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1492 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1493
8929ecfa 1494 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1495 cur = root->fs_info->running_transaction;
1496 if (!cur) {
8929ecfa 1497 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1498 goto sleep;
1499 }
31153d81 1500
a74a4b97 1501 now = get_seconds();
8929ecfa
YZ		 1502 if (!cur->blocked &&
1503 (now < cur->start_time || now - cur->start_time < 30)) {
1504 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1505 delay = HZ * 5;
1506 goto sleep;
1507 }
8929ecfa
YZ		 1508 transid = cur->transid;
1509 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1510
8929ecfa
YZ		 1511 trans = btrfs_join_transaction(root, 1);
1512 if (transid == trans->transid) {
1513 ret = btrfs_commit_transaction(trans, root);
1514 BUG_ON(ret);
1515 } else {
1516 btrfs_end_transaction(trans, root);
1517 }
a74a4b97
CM		 1518sleep:
1519 wake_up_process(root->fs_info->cleaner_kthread);
1520 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1521
1522 if (freezing(current)) {
1523 refrigerator();
1524 } else {
a74a4b97 1525 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1526 if (!kthread_should_stop() &&
1527 !btrfs_transaction_blocked(root->fs_info))
1528 schedule_timeout(delay);
a74a4b97
CM		 1529 __set_current_state(TASK_RUNNING);
1530 }
1531 } while (!kthread_should_stop());
1532 return 0;
1533}
1534
8a4b83cc 1535struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1536 struct btrfs_fs_devices *fs_devices,
1537 char *options)
2e635a27 1538{
db94535d
CM		 1539 u32 sectorsize;
1540 u32 nodesize;
1541 u32 leafsize;
1542 u32 blocksize;
87ee04eb 1543 u32 stripesize;
84234f3a 1544 u64 generation;
f2b636e8 1545 u64 features;
3de4586c 1546 struct btrfs_key location;
a061fc8d 1547 struct buffer_head *bh;
e02119d5 1548 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1549 GFP_NOFS);
d20f7043
CM		 1550 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1551 GFP_NOFS);
e02119d5 1552 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1553 GFP_NOFS);
8790d502 1554 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1555 GFP_NOFS);
e02119d5 1556 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1557 GFP_NOFS);
e02119d5 1558 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1559 GFP_NOFS);
e02119d5
CM		 1560 struct btrfs_root *log_tree_root;
1561
eb60ceac 1562 int ret;
e58ca020 1563 int err = -EINVAL;
4543df7e 1564
2c90e5d6 1565 struct btrfs_super_block *disk_super;
8790d502 1566
0463bb4e 1567 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1568 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1569 err = -ENOMEM;
1570 goto fail;
1571 }
76dda93c
YZ		 1572
1573 ret = init_srcu_struct(&fs_info->subvol_srcu);
1574 if (ret) {
1575 err = ret;
1576 goto fail;
1577 }
1578
1579 ret = setup_bdi(fs_info, &fs_info->bdi);
1580 if (ret) {
1581 err = ret;
1582 goto fail_srcu;
1583 }
1584
1585 fs_info->btree_inode = new_inode(sb);
1586 if (!fs_info->btree_inode) {
1587 err = -ENOMEM;
1588 goto fail_bdi;
1589 }
1590
1591 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1592 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1593 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1594 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1595 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1596 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1597 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1598 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1599 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1600 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1601 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1602 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1603 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1604
58176a96 1605 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1606 fs_info->tree_root = tree_root;
1607 fs_info->extent_root = extent_root;
d20f7043 1608 fs_info->csum_root = csum_root;
0b86a832
CM		 1609 fs_info->chunk_root = chunk_root;
1610 fs_info->dev_root = dev_root;
8a4b83cc 1611 fs_info->fs_devices = fs_devices;
0b86a832 1612 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1613 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1614 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1615 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1616 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1617 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1618 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1619 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1620 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1621 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1622 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1623 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1624 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1625 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1626 fs_info->sb = sb;
6f568d35 1627 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1628 fs_info->metadata_ratio = 0;
c8b97818 1629
b34b086c
CM		 1630 fs_info->thread_pool_size = min_t(unsigned long,
1631 num_online_cpus() + 2, 8);
0afbaf8c 1632
3eaa2885
CM		 1633 INIT_LIST_HEAD(&fs_info->ordered_extents);
1634 spin_lock_init(&fs_info->ordered_extent_lock);
1635
a061fc8d
CM		 1636 sb->s_blocksize = 4096;
1637 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1638 sb->s_bdi = &fs_info->bdi;
a061fc8d 1639
76dda93c
YZ		 1640 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1641 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1642 /*
1643 * we set the i_size on the btree inode to the max possible int.
1644 * the real end of the address space is determined by all of
1645 * the devices in the system
1646 */
1647 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1648 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1649 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1650
5d4f98a2 1651 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1652 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1653 fs_info->btree_inode->i_mapping,
1654 GFP_NOFS);
d1310b2e
CM		 1655 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1656 GFP_NOFS);
1657
1658 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1659
76dda93c
YZ		 1660 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1661 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1662 sizeof(struct btrfs_key));
1663 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1664 insert_inode_hash(fs_info->btree_inode);
76dda93c 1665
0f9dd46c 1666 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1667 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1668
11833d66 1669 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1670 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1671 extent_io_tree_init(&fs_info->freed_extents[1],
1672 fs_info->btree_inode->i_mapping, GFP_NOFS);
1673 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1674 fs_info->do_barriers = 1;
e18e4809 1675
39279cc3 1676
79154b1b 1677 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1678 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1679 mutex_init(&fs_info->tree_log_mutex);
925baedd 1680 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1681 mutex_init(&fs_info->transaction_kthread_mutex);
1682 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1683 mutex_init(&fs_info->volume_mutex);
276e680d 1684 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1685 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1686 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1687
1688 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1689 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1690
e6dcd2dc 1691 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1692 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1693 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1694
0b86a832 1695 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1696 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1697
a512bbf8 1698 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1699 if (!bh)
39279cc3 1700 goto fail_iput;
39279cc3 1701
a061fc8d 1702 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1703 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1704 sizeof(fs_info->super_for_commit));
a061fc8d 1705 brelse(bh);
5f39d397 1706
a061fc8d 1707 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1708
5f39d397 1709 disk_super = &fs_info->super_copy;
0f7d52f4 1710 if (!btrfs_super_root(disk_super))
c6e2bac1 1711 goto fail_iput;
0f7d52f4 1712
2b82032c
YZ		 1713 ret = btrfs_parse_options(tree_root, options);
1714 if (ret) {
1715 err = ret;
c6e2bac1 1716 goto fail_iput;
2b82032c 1717 }
dfe25020 1718
f2b636e8
JB		 1719 features = btrfs_super_incompat_flags(disk_super) &
1720 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1721 if (features) {
1722 printk(KERN_ERR "BTRFS: couldn't mount because of "
1723 "unsupported optional features (%Lx).\n",
21380931 1724 (unsigned long long)features);
f2b636e8 1725 err = -EINVAL;
c6e2bac1 1726 goto fail_iput;
f2b636e8
JB		 1727 }
1728
5d4f98a2
YZ		 1729 features = btrfs_super_incompat_flags(disk_super);
1730 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1731 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1732 btrfs_set_super_incompat_flags(disk_super, features);
1733 }
1734
f2b636e8
JB		 1735 features = btrfs_super_compat_ro_flags(disk_super) &
1736 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1737 if (!(sb->s_flags & MS_RDONLY) && features) {
1738 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1739 "unsupported option features (%Lx).\n",
21380931 1740 (unsigned long long)features);
f2b636e8 1741 err = -EINVAL;
c6e2bac1 1742 goto fail_iput;
f2b636e8 1743 }
61d92c32
CM		 1744
1745 btrfs_init_workers(&fs_info->generic_worker,
1746 "genwork", 1, NULL);
1747
5443be45 1748 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1749 fs_info->thread_pool_size,
1750 &fs_info->generic_worker);
c8b97818 1751
771ed689 1752 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1753 fs_info->thread_pool_size,
1754 &fs_info->generic_worker);
771ed689 1755
5443be45 1756 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1757 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1758 fs_info->thread_pool_size),
1759 &fs_info->generic_worker);
61b49440
CM		 1760
1761 /* a higher idle thresh on the submit workers makes it much more
1762 * likely that bios will be send down in a sane order to the
1763 * devices
1764 */
1765 fs_info->submit_workers.idle_thresh = 64;
53863232 1766
771ed689 1767 fs_info->workers.idle_thresh = 16;
4a69a410 1768 fs_info->workers.ordered = 1;
61b49440 1769
771ed689
CM		 1770 fs_info->delalloc_workers.idle_thresh = 2;
1771 fs_info->delalloc_workers.ordered = 1;
1772
61d92c32
CM		 1773 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1774 &fs_info->generic_worker);
5443be45 1775 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1776 fs_info->thread_pool_size,
1777 &fs_info->generic_worker);
d20f7043 1778 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1779 fs_info->thread_pool_size,
1780 &fs_info->generic_worker);
cad321ad 1781 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1782 "endio-meta-write", fs_info->thread_pool_size,
1783 &fs_info->generic_worker);
5443be45 1784 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1785 fs_info->thread_pool_size,
1786 &fs_info->generic_worker);
0cb59c99
JB		 1787 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1788 1, &fs_info->generic_worker);
61b49440
CM		 1789
1790 /*
1791 * endios are largely parallel and should have a very
1792 * low idle thresh
1793 */
1794 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1795 fs_info->endio_meta_workers.idle_thresh = 4;
1796
9042846b
CM		 1797 fs_info->endio_write_workers.idle_thresh = 2;
1798 fs_info->endio_meta_write_workers.idle_thresh = 2;
1799
4543df7e 1800 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1801 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1802 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1803 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1804 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1805 btrfs_start_workers(&fs_info->endio_workers, 1);
1806 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1807 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1808 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1809 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
4543df7e 1810
4575c9cc 1811 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1812 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1813 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1814
db94535d
CM		 1815 nodesize = btrfs_super_nodesize(disk_super);
1816 leafsize = btrfs_super_leafsize(disk_super);
1817 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1818 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1819 tree_root->nodesize = nodesize;
1820 tree_root->leafsize = leafsize;
1821 tree_root->sectorsize = sectorsize;
87ee04eb 1822 tree_root->stripesize = stripesize;
a061fc8d
CM		 1823
1824 sb->s_blocksize = sectorsize;
1825 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1826
39279cc3
CM		 1827 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1828 sizeof(disk_super->magic))) {
d397712b 1829 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1830 goto fail_sb_buffer;
1831 }
19c00ddc 1832
925baedd 1833 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1834 ret = btrfs_read_sys_array(tree_root);
925baedd 1835 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1836 if (ret) {
d397712b
CM		 1837 printk(KERN_WARNING "btrfs: failed to read the system "
1838 "array on %s\n", sb->s_id);
5d4f98a2 1839 goto fail_sb_buffer;
84eed90f 1840 }
0b86a832
CM		 1841
1842 blocksize = btrfs_level_size(tree_root,
1843 btrfs_super_chunk_root_level(disk_super));
84234f3a 1844 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1845
1846 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1847 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1848
1849 chunk_root->node = read_tree_block(chunk_root,
1850 btrfs_super_chunk_root(disk_super),
84234f3a 1851 blocksize, generation);
0b86a832 1852 BUG_ON(!chunk_root->node);
83121942
DW		 1853 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1854 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1855 sb->s_id);
1856 goto fail_chunk_root;
1857 }
5d4f98a2
YZ		 1858 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1859 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1860
e17cade2 1861 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1862 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1863 BTRFS_UUID_SIZE);
e17cade2 1864
925baedd 1865 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1866 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1867 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1868 if (ret) {
d397712b
CM		 1869 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1870 sb->s_id);
2b82032c
YZ		 1871 goto fail_chunk_root;
1872 }
0b86a832 1873
dfe25020
CM		 1874 btrfs_close_extra_devices(fs_devices);
1875
db94535d
CM		 1876 blocksize = btrfs_level_size(tree_root,
1877 btrfs_super_root_level(disk_super));
84234f3a 1878 generation = btrfs_super_generation(disk_super);
0b86a832 1879
e20d96d6 1880 tree_root->node = read_tree_block(tree_root,
db94535d 1881 btrfs_super_root(disk_super),
84234f3a 1882 blocksize, generation);
39279cc3 1883 if (!tree_root->node)
2b82032c 1884 goto fail_chunk_root;
83121942
DW		 1885 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1886 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1887 sb->s_id);
1888 goto fail_tree_root;
1889 }
5d4f98a2
YZ		 1890 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1891 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1892
1893 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1894 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1895 if (ret)
39279cc3 1896 goto fail_tree_root;
0b86a832
CM		 1897 extent_root->track_dirty = 1;
1898
1899 ret = find_and_setup_root(tree_root, fs_info,
1900 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1901 if (ret)
1902 goto fail_extent_root;
5d4f98a2 1903 dev_root->track_dirty = 1;
3768f368 1904
d20f7043
CM		 1905 ret = find_and_setup_root(tree_root, fs_info,
1906 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1907 if (ret)
5d4f98a2 1908 goto fail_dev_root;
d20f7043
CM		 1909
1910 csum_root->track_dirty = 1;
1911
8929ecfa
YZ		 1912 fs_info->generation = generation;
1913 fs_info->last_trans_committed = generation;
1914 fs_info->data_alloc_profile = (u64)-1;
1915 fs_info->metadata_alloc_profile = (u64)-1;
1916 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1917
1b1d1f66
JB		 1918 ret = btrfs_read_block_groups(extent_root);
1919 if (ret) {
1920 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1921 goto fail_block_groups;
1922 }
9078a3e1 1923
a74a4b97
CM		 1924 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1925 "btrfs-cleaner");
57506d50 1926 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 1927 goto fail_block_groups;
a74a4b97
CM		 1928
1929 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1930 tree_root,
1931 "btrfs-transaction");
57506d50 1932 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1933 goto fail_cleaner;
a74a4b97 1934
c289811c
CM		 1935 if (!btrfs_test_opt(tree_root, SSD) &&
1936 !btrfs_test_opt(tree_root, NOSSD) &&
1937 !fs_info->fs_devices->rotating) {
1938 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1939 "mode\n");
1940 btrfs_set_opt(fs_info->mount_opt, SSD);
1941 }
1942
e02119d5 1943 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1944 u64 bytenr = btrfs_super_log_root(disk_super);
1945
7c2ca468 1946 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1947 printk(KERN_WARNING "Btrfs log replay required "
1948 "on RO media\n");
7c2ca468
CM		 1949 err = -EIO;
1950 goto fail_trans_kthread;
1951 }
e02119d5
CM		 1952 blocksize =
1953 btrfs_level_size(tree_root,
1954 btrfs_super_log_root_level(disk_super));
d18a2c44 1955
676e4c86
DC		 1956 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1957 if (!log_tree_root) {
1958 err = -ENOMEM;
1959 goto fail_trans_kthread;
1960 }
e02119d5
CM		 1961
1962 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1963 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1964
1965 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1966 blocksize,
1967 generation + 1);
e02119d5
CM		 1968 ret = btrfs_recover_log_trees(log_tree_root);
1969 BUG_ON(ret);
e556ce2c
YZ		 1970
1971 if (sb->s_flags & MS_RDONLY) {
1972 ret = btrfs_commit_super(tree_root);
1973 BUG_ON(ret);
1974 }
e02119d5 1975 }
1a40e23b 1976
76dda93c
YZ		 1977 ret = btrfs_find_orphan_roots(tree_root);
1978 BUG_ON(ret);
1979
7c2ca468 1980 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ		 1981 ret = btrfs_cleanup_fs_roots(fs_info);
1982 BUG_ON(ret);
1983
5d4f98a2 1984 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 1985 if (ret < 0) {
1986 printk(KERN_WARNING
1987 "btrfs: failed to recover relocation\n");
1988 err = -EINVAL;
1989 goto fail_trans_kthread;
1990 }
7c2ca468 1991 }
1a40e23b 1992
3de4586c
CM		 1993 location.objectid = BTRFS_FS_TREE_OBJECTID;
1994 location.type = BTRFS_ROOT_ITEM_KEY;
1995 location.offset = (u64)-1;
1996
3de4586c
CM		 1997 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1998 if (!fs_info->fs_root)
7c2ca468 1999 goto fail_trans_kthread;
3140c9a3
DC		 2000 if (IS_ERR(fs_info->fs_root)) {
2001 err = PTR_ERR(fs_info->fs_root);
2002 goto fail_trans_kthread;
2003 }
c289811c 2004
e3acc2a6
JB		 2005 if (!(sb->s_flags & MS_RDONLY)) {
2006 down_read(&fs_info->cleanup_work_sem);
2007 btrfs_orphan_cleanup(fs_info->fs_root);
0af3d00b 2008 btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6
JB		 2009 up_read(&fs_info->cleanup_work_sem);
2010 }
2011
0f7d52f4 2012 return tree_root;
39279cc3 2013
7c2ca468
CM		 2014fail_trans_kthread:
2015 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2016fail_cleaner:
a74a4b97 2017 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2018
2019 /*
2020 * make sure we're done with the btree inode before we stop our
2021 * kthreads
2022 */
2023 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2024 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2025
1b1d1f66
JB		 2026fail_block_groups:
2027 btrfs_free_block_groups(fs_info);
d20f7043 2028 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2029 free_extent_buffer(csum_root->commit_root);
2030fail_dev_root:
2031 free_extent_buffer(dev_root->node);
2032 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2033fail_extent_root:
2034 free_extent_buffer(extent_root->node);
5d4f98a2 2035 free_extent_buffer(extent_root->commit_root);
39279cc3 2036fail_tree_root:
5f39d397 2037 free_extent_buffer(tree_root->node);
5d4f98a2 2038 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2039fail_chunk_root:
2040 free_extent_buffer(chunk_root->node);
5d4f98a2 2041 free_extent_buffer(chunk_root->commit_root);
39279cc3 2042fail_sb_buffer:
61d92c32 2043 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2044 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2045 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2046 btrfs_stop_workers(&fs_info->workers);
2047 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2048 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2049 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2050 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2051 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2052 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2053fail_iput:
7c2ca468 2054 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2055 iput(fs_info->btree_inode);
7e662854 2056
dfe25020 2057 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2058 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2059fail_bdi:
7e662854 2060 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2061fail_srcu:
2062 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2063fail:
39279cc3
CM		 2064 kfree(extent_root);
2065 kfree(tree_root);
2066 kfree(fs_info);
83afeac4
JM		 2067 kfree(chunk_root);
2068 kfree(dev_root);
d20f7043 2069 kfree(csum_root);
39279cc3 2070 return ERR_PTR(err);
eb60ceac
CM		 2071}
2072
f2984462
CM		 2073static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2074{
2075 char b[BDEVNAME_SIZE];
2076
2077 if (uptodate) {
2078 set_buffer_uptodate(bh);
2079 } else {
2080 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2081 printk(KERN_WARNING "lost page write due to "
2082 "I/O error on %s\n",
2083 bdevname(bh->b_bdev, b));
2084 }
1259ab75
CM		 2085 /* note, we dont' set_buffer_write_io_error because we have
2086 * our own ways of dealing with the IO errors
2087 */
f2984462
CM		 2088 clear_buffer_uptodate(bh);
2089 }
2090 unlock_buffer(bh);
2091 put_bh(bh);
2092}
2093
a512bbf8
YZ		 2094struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2095{
2096 struct buffer_head *bh;
2097 struct buffer_head *latest = NULL;
2098 struct btrfs_super_block *super;
2099 int i;
2100 u64 transid = 0;
2101 u64 bytenr;
2102
2103 /* we would like to check all the supers, but that would make
2104 * a btrfs mount succeed after a mkfs from a different FS.
2105 * So, we need to add a special mount option to scan for
2106 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2107 */
2108 for (i = 0; i < 1; i++) {
2109 bytenr = btrfs_sb_offset(i);
2110 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2111 break;
2112 bh = __bread(bdev, bytenr / 4096, 4096);
2113 if (!bh)
2114 continue;
2115
2116 super = (struct btrfs_super_block *)bh->b_data;
2117 if (btrfs_super_bytenr(super) != bytenr ||
2118 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2119 sizeof(super->magic))) {
2120 brelse(bh);
2121 continue;
2122 }
2123
2124 if (!latest || btrfs_super_generation(super) > transid) {
2125 brelse(latest);
2126 latest = bh;
2127 transid = btrfs_super_generation(super);
2128 } else {
2129 brelse(bh);
2130 }
2131 }
2132 return latest;
2133}
2134
4eedeb75
HH		 2135/*
2136 * this should be called twice, once with wait == 0 and
2137 * once with wait == 1. When wait == 0 is done, all the buffer heads
2138 * we write are pinned.
2139 *
2140 * They are released when wait == 1 is done.
2141 * max_mirrors must be the same for both runs, and it indicates how
2142 * many supers on this one device should be written.
2143 *
2144 * max_mirrors == 0 means to write them all.
2145 */
a512bbf8
YZ		 2146static int write_dev_supers(struct btrfs_device *device,
2147 struct btrfs_super_block *sb,
2148 int do_barriers, int wait, int max_mirrors)
2149{
2150 struct buffer_head *bh;
2151 int i;
2152 int ret;
2153 int errors = 0;
2154 u32 crc;
2155 u64 bytenr;
2156 int last_barrier = 0;
2157
2158 if (max_mirrors == 0)
2159 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2160
2161 /* make sure only the last submit_bh does a barrier */
2162 if (do_barriers) {
2163 for (i = 0; i < max_mirrors; i++) {
2164 bytenr = btrfs_sb_offset(i);
2165 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2166 device->total_bytes)
2167 break;
2168 last_barrier = i;
2169 }
2170 }
2171
2172 for (i = 0; i < max_mirrors; i++) {
2173 bytenr = btrfs_sb_offset(i);
2174 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2175 break;
2176
2177 if (wait) {
2178 bh = __find_get_block(device->bdev, bytenr / 4096,
2179 BTRFS_SUPER_INFO_SIZE);
2180 BUG_ON(!bh);
a512bbf8 2181 wait_on_buffer(bh);
4eedeb75
HH		 2182 if (!buffer_uptodate(bh))
2183 errors++;
2184
2185 /* drop our reference */
2186 brelse(bh);
2187
2188 /* drop the reference from the wait == 0 run */
2189 brelse(bh);
2190 continue;
a512bbf8
YZ		 2191 } else {
2192 btrfs_set_super_bytenr(sb, bytenr);
2193
2194 crc = ~(u32)0;
2195 crc = btrfs_csum_data(NULL, (char *)sb +
2196 BTRFS_CSUM_SIZE, crc,
2197 BTRFS_SUPER_INFO_SIZE -
2198 BTRFS_CSUM_SIZE);
2199 btrfs_csum_final(crc, sb->csum);
2200
4eedeb75
HH		 2201 /*
2202 * one reference for us, and we leave it for the
2203 * caller
2204 */
a512bbf8
YZ		 2205 bh = __getblk(device->bdev, bytenr / 4096,
2206 BTRFS_SUPER_INFO_SIZE);
2207 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2208
4eedeb75 2209 /* one reference for submit_bh */
a512bbf8 2210 get_bh(bh);
4eedeb75
HH		 2211
2212 set_buffer_uptodate(bh);
a512bbf8
YZ		 2213 lock_buffer(bh);
2214 bh->b_end_io = btrfs_end_buffer_write_sync;
2215 }
2216
2217 if (i == last_barrier && do_barriers && device->barriers) {
2218 ret = submit_bh(WRITE_BARRIER, bh);
2219 if (ret == -EOPNOTSUPP) {
2220 printk("btrfs: disabling barriers on dev %s\n",
2221 device->name);
2222 set_buffer_uptodate(bh);
2223 device->barriers = 0;
4eedeb75 2224 /* one reference for submit_bh */
a512bbf8
YZ		 2225 get_bh(bh);
2226 lock_buffer(bh);
ffbd517d 2227 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2228 }
2229 } else {
ffbd517d 2230 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2231 }
2232
4eedeb75 2233 if (ret)
a512bbf8 2234 errors++;
a512bbf8
YZ		 2235 }
2236 return errors < i ? 0 : -1;
2237}
2238
2239int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2240{
e5e9a520 2241 struct list_head *head;
f2984462 2242 struct btrfs_device *dev;
a061fc8d 2243 struct btrfs_super_block *sb;
f2984462 2244 struct btrfs_dev_item *dev_item;
f2984462
CM		 2245 int ret;
2246 int do_barriers;
a236aed1
CM		 2247 int max_errors;
2248 int total_errors = 0;
a061fc8d 2249 u64 flags;
f2984462 2250
a236aed1 2251 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2252 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2253
a061fc8d
CM		 2254 sb = &root->fs_info->super_for_commit;
2255 dev_item = &sb->dev_item;
e5e9a520
CM		 2256
2257 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2258 head = &root->fs_info->fs_devices->devices;
c6e30871 2259 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2260 if (!dev->bdev) {
2261 total_errors++;
2262 continue;
2263 }
2b82032c 2264 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2265 continue;
2266
2b82032c 2267 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2268 btrfs_set_stack_device_type(dev_item, dev->type);
2269 btrfs_set_stack_device_id(dev_item, dev->devid);
2270 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2271 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2272 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2273 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2274 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2275 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2276 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2277
a061fc8d
CM		 2278 flags = btrfs_super_flags(sb);
2279 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2280
a512bbf8 2281 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2282 if (ret)
2283 total_errors++;
f2984462 2284 }
a236aed1 2285 if (total_errors > max_errors) {
d397712b
CM		 2286 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2287 total_errors);
a236aed1
CM		 2288 BUG();
2289 }
f2984462 2290
a512bbf8 2291 total_errors = 0;
c6e30871 2292 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2293 if (!dev->bdev)
2294 continue;
2b82032c 2295 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2296 continue;
2297
a512bbf8
YZ		 2298 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2299 if (ret)
2300 total_errors++;
f2984462 2301 }
e5e9a520 2302 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2303 if (total_errors > max_errors) {
d397712b
CM		 2304 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2305 total_errors);
a236aed1
CM		 2306 BUG();
2307 }
f2984462
CM		 2308 return 0;
2309}
2310
a512bbf8
YZ		 2311int write_ctree_super(struct btrfs_trans_handle *trans,
2312 struct btrfs_root *root, int max_mirrors)
eb60ceac 2313{
e66f709b 2314 int ret;
5f39d397 2315
a512bbf8 2316 ret = write_all_supers(root, max_mirrors);
5f39d397 2317 return ret;
cfaa7295
CM		 2318}
2319
5eda7b5e 2320int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2321{
4df27c4d 2322 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2323 radix_tree_delete(&fs_info->fs_roots_radix,
2324 (unsigned long)root->root_key.objectid);
4df27c4d 2325 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2326
2327 if (btrfs_root_refs(&root->root_item) == 0)
2328 synchronize_srcu(&fs_info->subvol_srcu);
2329
4df27c4d
YZ		 2330 free_fs_root(root);
2331 return 0;
2332}
2333
2334static void free_fs_root(struct btrfs_root *root)
2335{
2336 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2337 if (root->anon_super.s_dev) {
2338 down_write(&root->anon_super.s_umount);
2339 kill_anon_super(&root->anon_super);
2340 }
4df27c4d
YZ		 2341 free_extent_buffer(root->node);
2342 free_extent_buffer(root->commit_root);
d397712b 2343 kfree(root->name);
2619ba1f 2344 kfree(root);
2619ba1f
CM		 2345}
2346
35b7e476 2347static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2348{
2349 int ret;
2350 struct btrfs_root *gang[8];
2351 int i;
2352
76dda93c
YZ		 2353 while (!list_empty(&fs_info->dead_roots)) {
2354 gang[0] = list_entry(fs_info->dead_roots.next,
2355 struct btrfs_root, root_list);
2356 list_del(&gang[0]->root_list);
2357
2358 if (gang[0]->in_radix) {
2359 btrfs_free_fs_root(fs_info, gang[0]);
2360 } else {
2361 free_extent_buffer(gang[0]->node);
2362 free_extent_buffer(gang[0]->commit_root);
2363 kfree(gang[0]);
2364 }
2365 }
2366
d397712b 2367 while (1) {
0f7d52f4
CM		 2368 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2369 (void **)gang, 0,
2370 ARRAY_SIZE(gang));
2371 if (!ret)
2372 break;
2619ba1f 2373 for (i = 0; i < ret; i++)
5eda7b5e 2374 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2375 }
2376 return 0;
2377}
b4100d64 2378
c146afad 2379int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2380{
c146afad
YZ		 2381 u64 root_objectid = 0;
2382 struct btrfs_root *gang[8];
2383 int i;
3768f368 2384 int ret;
e089f05c 2385
c146afad
YZ		 2386 while (1) {
2387 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2388 (void **)gang, root_objectid,
2389 ARRAY_SIZE(gang));
2390 if (!ret)
2391 break;
5d4f98a2
YZ		 2392
2393 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2394 for (i = 0; i < ret; i++) {
2395 root_objectid = gang[i]->root_key.objectid;
c146afad
YZ		 2396 btrfs_orphan_cleanup(gang[i]);
2397 }
2398 root_objectid++;
2399 }
2400 return 0;
2401}
a2135011 2402
c146afad
YZ		 2403int btrfs_commit_super(struct btrfs_root *root)
2404{
2405 struct btrfs_trans_handle *trans;
2406 int ret;
a74a4b97 2407
c146afad 2408 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2409 btrfs_run_delayed_iputs(root);
a74a4b97 2410 btrfs_clean_old_snapshots(root);
c146afad 2411 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2412
2413 /* wait until ongoing cleanup work done */
2414 down_write(&root->fs_info->cleanup_work_sem);
2415 up_write(&root->fs_info->cleanup_work_sem);
2416
a22285a6 2417 trans = btrfs_join_transaction(root, 1);
54aa1f4d 2418 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2419 BUG_ON(ret);
2420 /* run commit again to drop the original snapshot */
a22285a6 2421 trans = btrfs_join_transaction(root, 1);
79154b1b
CM		 2422 btrfs_commit_transaction(trans, root);
2423 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2424 BUG_ON(ret);
d6bfde87 2425
a512bbf8 2426 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2427 return ret;
2428}
2429
2430int close_ctree(struct btrfs_root *root)
2431{
2432 struct btrfs_fs_info *fs_info = root->fs_info;
2433 int ret;
2434
2435 fs_info->closing = 1;
2436 smp_mb();
2437
0af3d00b 2438 btrfs_put_block_group_cache(fs_info);
c146afad
YZ		 2439 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2440 ret = btrfs_commit_super(root);
d397712b
CM		 2441 if (ret)
2442 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2443 }
0f7d52f4 2444
8929ecfa
YZ		 2445 kthread_stop(root->fs_info->transaction_kthread);
2446 kthread_stop(root->fs_info->cleaner_kthread);
2447
f25784b3
YZ		 2448 fs_info->closing = 2;
2449 smp_mb();
2450
b0c68f8b 2451 if (fs_info->delalloc_bytes) {
d397712b 2452 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2453 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2454 }
31153d81 2455 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2456 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2457 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2458 }
bcc63abb 2459
5d4f98a2
YZ		 2460 free_extent_buffer(fs_info->extent_root->node);
2461 free_extent_buffer(fs_info->extent_root->commit_root);
2462 free_extent_buffer(fs_info->tree_root->node);
2463 free_extent_buffer(fs_info->tree_root->commit_root);
2464 free_extent_buffer(root->fs_info->chunk_root->node);
2465 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2466 free_extent_buffer(root->fs_info->dev_root->node);
2467 free_extent_buffer(root->fs_info->dev_root->commit_root);
2468 free_extent_buffer(root->fs_info->csum_root->node);
2469 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2470
9078a3e1 2471 btrfs_free_block_groups(root->fs_info);
d10c5f31 2472
c146afad 2473 del_fs_roots(fs_info);
d10c5f31 2474
c146afad 2475 iput(fs_info->btree_inode);
9ad6b7bc 2476
61d92c32 2477 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2478 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2479 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2480 btrfs_stop_workers(&fs_info->workers);
2481 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2482 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2483 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2484 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2485 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2486 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2487
dfe25020 2488 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2489 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2490
04160088 2491 bdi_destroy(&fs_info->bdi);
76dda93c 2492 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2493
0f7d52f4 2494 kfree(fs_info->extent_root);
0f7d52f4 2495 kfree(fs_info->tree_root);
0b86a832
CM		 2496 kfree(fs_info->chunk_root);
2497 kfree(fs_info->dev_root);
d20f7043 2498 kfree(fs_info->csum_root);
eb60ceac
CM		 2499 return 0;
2500}
2501
1259ab75 2502int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2503{
1259ab75 2504 int ret;
810191ff 2505 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2506
2ac55d41
JB		 2507 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2508 NULL);
1259ab75
CM		 2509 if (!ret)
2510 return ret;
2511
2512 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2513 parent_transid);
2514 return !ret;
5f39d397
CM		 2515}
2516
2517int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2518{
810191ff 2519 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2520 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2521 buf);
2522}
6702ed49 2523
5f39d397
CM		 2524void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2525{
810191ff 2526 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2527 u64 transid = btrfs_header_generation(buf);
2528 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2529 int was_dirty;
b4ce94de 2530
b9447ef8 2531 btrfs_assert_tree_locked(buf);
ccd467d6 2532 if (transid != root->fs_info->generation) {
d397712b
CM		 2533 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2534 "found %llu running %llu\n",
db94535d 2535 (unsigned long long)buf->start,
d397712b
CM		 2536 (unsigned long long)transid,
2537 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2538 WARN_ON(1);
2539 }
b9473439
CM		 2540 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2541 buf);
2542 if (!was_dirty) {
2543 spin_lock(&root->fs_info->delalloc_lock);
2544 root->fs_info->dirty_metadata_bytes += buf->len;
2545 spin_unlock(&root->fs_info->delalloc_lock);
2546 }
eb60ceac
CM		 2547}
2548
d3c2fdcf 2549void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2550{
188de649
CM		 2551 /*
2552 * looks as though older kernels can get into trouble with
2553 * this code, they end up stuck in balance_dirty_pages forever
2554 */
d6bfde87 2555 u64 num_dirty;
771ed689 2556 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2557
6933c02e 2558 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2559 return;
2560
585ad2c3
CM		 2561 num_dirty = root->fs_info->dirty_metadata_bytes;
2562
d6bfde87
CM		 2563 if (num_dirty > thresh) {
2564 balance_dirty_pages_ratelimited_nr(
d7fc640e 2565 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2566 }
188de649 2567 return;
35b7e476 2568}
6b80053d 2569
ca7a79ad 2570int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2571{
810191ff 2572 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2573 int ret;
ca7a79ad 2574 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2575 if (ret == 0)
b4ce94de 2576 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2577 return ret;
6b80053d 2578}
0da5468f 2579
4bef0848
CM		 2580int btree_lock_page_hook(struct page *page)
2581{
2582 struct inode *inode = page->mapping->host;
b9473439 2583 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2584 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2585 struct extent_buffer *eb;
2586 unsigned long len;
2587 u64 bytenr = page_offset(page);
2588
2589 if (page->private == EXTENT_PAGE_PRIVATE)
2590 goto out;
2591
2592 len = page->private >> 2;
2593 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2594 if (!eb)
2595 goto out;
2596
2597 btrfs_tree_lock(eb);
4bef0848 2598 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2599
2600 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2601 spin_lock(&root->fs_info->delalloc_lock);
2602 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2603 root->fs_info->dirty_metadata_bytes -= eb->len;
2604 else
2605 WARN_ON(1);
2606 spin_unlock(&root->fs_info->delalloc_lock);
2607 }
2608
4bef0848
CM		 2609 btrfs_tree_unlock(eb);
2610 free_extent_buffer(eb);
2611out:
2612 lock_page(page);
2613 return 0;
2614}
2615
d1310b2e 2616static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2617 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2618 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2619 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2620 /* note we're sharing with inode.c for the merge bio hook */
2621 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2622};
Clone of davem's net-next-2.6 tree