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