projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix data space leak fix
[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>
4b4e25f2 30#include "compat.h"
eb60ceac
CM		 31#include "ctree.h"
32#include "disk-io.h"
e089f05c 33#include "transaction.h"
0f7d52f4 34#include "btrfs_inode.h"
0b86a832 35#include "volumes.h"
db94535d 36#include "print-tree.h"
8b712842 37#include "async-thread.h"
925baedd 38#include "locking.h"
e02119d5 39#include "tree-log.h"
fa9c0d79 40#include "free-space-cache.h"
eb60ceac 41
d1310b2e 42static struct extent_io_ops btree_extent_io_ops;
8b712842 43static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 44static void free_fs_root(struct btrfs_root *root);
ce9adaa5 45
ad081f14
JA		 46static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
47
d352ac68
CM		 48/*
49 * end_io_wq structs are used to do processing in task context when an IO is
50 * complete. This is used during reads to verify checksums, and it is used
51 * by writes to insert metadata for new file extents after IO is complete.
52 */
ce9adaa5
CM		 53struct end_io_wq {
54 struct bio *bio;
55 bio_end_io_t *end_io;
56 void *private;
57 struct btrfs_fs_info *info;
58 int error;
22c59948 59 int metadata;
ce9adaa5 60 struct list_head list;
8b712842 61 struct btrfs_work work;
ce9adaa5 62};
0da5468f 63
d352ac68
CM		 64/*
65 * async submit bios are used to offload expensive checksumming
66 * onto the worker threads. They checksum file and metadata bios
67 * just before they are sent down the IO stack.
68 */
44b8bd7e
CM		 69struct async_submit_bio {
70 struct inode *inode;
71 struct bio *bio;
72 struct list_head list;
4a69a410
CM		 73 extent_submit_bio_hook_t *submit_bio_start;
74 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 75 int rw;
76 int mirror_num;
c8b97818 77 unsigned long bio_flags;
8b712842 78 struct btrfs_work work;
44b8bd7e
CM		 79};
80
4008c04a
CM		 81/* These are used to set the lockdep class on the extent buffer locks.
82 * The class is set by the readpage_end_io_hook after the buffer has
83 * passed csum validation but before the pages are unlocked.
84 *
85 * The lockdep class is also set by btrfs_init_new_buffer on freshly
86 * allocated blocks.
87 *
88 * The class is based on the level in the tree block, which allows lockdep
89 * to know that lower nodes nest inside the locks of higher nodes.
90 *
91 * We also add a check to make sure the highest level of the tree is
92 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
93 * code needs update as well.
94 */
95#ifdef CONFIG_DEBUG_LOCK_ALLOC
96# if BTRFS_MAX_LEVEL != 8
97# error
98# endif
99static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
100static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
101 /* leaf */
102 "btrfs-extent-00",
103 "btrfs-extent-01",
104 "btrfs-extent-02",
105 "btrfs-extent-03",
106 "btrfs-extent-04",
107 "btrfs-extent-05",
108 "btrfs-extent-06",
109 "btrfs-extent-07",
110 /* highest possible level */
111 "btrfs-extent-08",
112};
113#endif
114
d352ac68
CM		 115/*
116 * extents on the btree inode are pretty simple, there's one extent
117 * that covers the entire device
118 */
b2950863
CH		 119static struct extent_map *btree_get_extent(struct inode *inode,
120 struct page *page, size_t page_offset, u64 start, u64 len,
121 int create)
7eccb903 122{
5f39d397
CM		 123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
124 struct extent_map *em;
125 int ret;
126
890871be 127 read_lock(&em_tree->lock);
d1310b2e 128 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 129 if (em) {
130 em->bdev =
131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 132 read_unlock(&em_tree->lock);
5f39d397 133 goto out;
a061fc8d 134 }
890871be 135 read_unlock(&em_tree->lock);
7b13b7b1 136
5f39d397
CM		 137 em = alloc_extent_map(GFP_NOFS);
138 if (!em) {
139 em = ERR_PTR(-ENOMEM);
140 goto out;
141 }
142 em->start = 0;
0afbaf8c 143 em->len = (u64)-1;
c8b97818 144 em->block_len = (u64)-1;
5f39d397 145 em->block_start = 0;
a061fc8d 146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 147
890871be 148 write_lock(&em_tree->lock);
5f39d397
CM		 149 ret = add_extent_mapping(em_tree, em);
150 if (ret == -EEXIST) {
0afbaf8c
CM		 151 u64 failed_start = em->start;
152 u64 failed_len = em->len;
153
5f39d397 154 free_extent_map(em);
7b13b7b1 155 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 156 if (em) {
7b13b7b1 157 ret = 0;
0afbaf8c
CM		 158 } else {
159 em = lookup_extent_mapping(em_tree, failed_start,
160 failed_len);
7b13b7b1 161 ret = -EIO;
0afbaf8c 162 }
5f39d397 163 } else if (ret) {
7b13b7b1
CM		 164 free_extent_map(em);
165 em = NULL;
5f39d397 166 }
890871be 167 write_unlock(&em_tree->lock);
7b13b7b1
CM		 168
169 if (ret)
170 em = ERR_PTR(ret);
5f39d397
CM		 171out:
172 return em;
7eccb903
CM		 173}
174
19c00ddc
CM		 175u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
176{
163e783e 177 return crc32c(seed, data, len);
19c00ddc
CM		 178}
179
180void btrfs_csum_final(u32 crc, char *result)
181{
182 *(__le32 *)result = ~cpu_to_le32(crc);
183}
184
d352ac68
CM		 185/*
186 * compute the csum for a btree block, and either verify it or write it
187 * into the csum field of the block.
188 */
19c00ddc
CM		 189static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
190 int verify)
191{
607d432d
JB		 192 u16 csum_size =
193 btrfs_super_csum_size(&root->fs_info->super_copy);
194 char *result = NULL;
19c00ddc
CM		 195 unsigned long len;
196 unsigned long cur_len;
197 unsigned long offset = BTRFS_CSUM_SIZE;
198 char *map_token = NULL;
199 char *kaddr;
200 unsigned long map_start;
201 unsigned long map_len;
202 int err;
203 u32 crc = ~(u32)0;
607d432d 204 unsigned long inline_result;
19c00ddc
CM		 205
206 len = buf->len - offset;
d397712b 207 while (len > 0) {
19c00ddc
CM		 208 err = map_private_extent_buffer(buf, offset, 32,
209 &map_token, &kaddr,
210 &map_start, &map_len, KM_USER0);
d397712b 211 if (err)
19c00ddc 212 return 1;
19c00ddc
CM		 213 cur_len = min(len, map_len - (offset - map_start));
214 crc = btrfs_csum_data(root, kaddr + offset - map_start,
215 crc, cur_len);
216 len -= cur_len;
217 offset += cur_len;
218 unmap_extent_buffer(buf, map_token, KM_USER0);
219 }
607d432d
JB		 220 if (csum_size > sizeof(inline_result)) {
221 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
222 if (!result)
223 return 1;
224 } else {
225 result = (char *)&inline_result;
226 }
227
19c00ddc
CM		 228 btrfs_csum_final(crc, result);
229
230 if (verify) {
607d432d 231 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 232 u32 val;
233 u32 found = 0;
607d432d 234 memcpy(&found, result, csum_size);
e4204ded 235
607d432d 236 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 237 if (printk_ratelimit()) {
238 printk(KERN_INFO "btrfs: %s checksum verify "
239 "failed on %llu wanted %X found %X "
240 "level %d\n",
241 root->fs_info->sb->s_id,
242 (unsigned long long)buf->start, val, found,
243 btrfs_header_level(buf));
244 }
607d432d
JB		 245 if (result != (char *)&inline_result)
246 kfree(result);
19c00ddc
CM		 247 return 1;
248 }
249 } else {
607d432d 250 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 251 }
607d432d
JB		 252 if (result != (char *)&inline_result)
253 kfree(result);
19c00ddc
CM		 254 return 0;
255}
256
d352ac68
CM		 257/*
258 * we can't consider a given block up to date unless the transid of the
259 * block matches the transid in the parent node's pointer. This is how we
260 * detect blocks that either didn't get written at all or got written
261 * in the wrong place.
262 */
1259ab75
CM		 263static int verify_parent_transid(struct extent_io_tree *io_tree,
264 struct extent_buffer *eb, u64 parent_transid)
265{
266 int ret;
267
268 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
269 return 0;
270
271 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
272 if (extent_buffer_uptodate(io_tree, eb) &&
273 btrfs_header_generation(eb) == parent_transid) {
274 ret = 0;
275 goto out;
276 }
193f284d
CM		 277 if (printk_ratelimit()) {
278 printk("parent transid verify failed on %llu wanted %llu "
279 "found %llu\n",
280 (unsigned long long)eb->start,
281 (unsigned long long)parent_transid,
282 (unsigned long long)btrfs_header_generation(eb));
283 }
1259ab75 284 ret = 1;
1259ab75 285 clear_extent_buffer_uptodate(io_tree, eb);
33958dc6 286out:
1259ab75
CM		 287 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
288 GFP_NOFS);
289 return ret;
1259ab75
CM		 290}
291
d352ac68
CM		 292/*
293 * helper to read a given tree block, doing retries as required when
294 * the checksums don't match and we have alternate mirrors to try.
295 */
f188591e
CM		 296static int btree_read_extent_buffer_pages(struct btrfs_root *root,
297 struct extent_buffer *eb,
ca7a79ad 298 u64 start, u64 parent_transid)
f188591e
CM		 299{
300 struct extent_io_tree *io_tree;
301 int ret;
302 int num_copies = 0;
303 int mirror_num = 0;
304
305 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
306 while (1) {
307 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
308 btree_get_extent, mirror_num);
1259ab75
CM		 309 if (!ret &&
310 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 311 return ret;
d397712b 312
f188591e
CM		 313 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
314 eb->start, eb->len);
4235298e 315 if (num_copies == 1)
f188591e 316 return ret;
4235298e 317
f188591e 318 mirror_num++;
4235298e 319 if (mirror_num > num_copies)
f188591e 320 return ret;
f188591e 321 }
f188591e
CM		 322 return -EIO;
323}
19c00ddc 324
d352ac68 325/*
d397712b
CM		 326 * checksum a dirty tree block before IO. This has extra checks to make sure
327 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 328 */
d397712b 329
b2950863 330static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 331{
d1310b2e 332 struct extent_io_tree *tree;
35ebb934 333 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 334 u64 found_start;
335 int found_level;
336 unsigned long len;
337 struct extent_buffer *eb;
f188591e
CM		 338 int ret;
339
d1310b2e 340 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 341
342 if (page->private == EXTENT_PAGE_PRIVATE)
343 goto out;
344 if (!page->private)
345 goto out;
346 len = page->private >> 2;
d397712b
CM		 347 WARN_ON(len == 0);
348
19c00ddc 349 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 350 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
351 btrfs_header_generation(eb));
f188591e 352 BUG_ON(ret);
19c00ddc
CM		 353 found_start = btrfs_header_bytenr(eb);
354 if (found_start != start) {
55c69072
CM		 355 WARN_ON(1);
356 goto err;
357 }
358 if (eb->first_page != page) {
55c69072
CM		 359 WARN_ON(1);
360 goto err;
361 }
362 if (!PageUptodate(page)) {
55c69072
CM		 363 WARN_ON(1);
364 goto err;
19c00ddc
CM		 365 }
366 found_level = btrfs_header_level(eb);
4bef0848 367
19c00ddc 368 csum_tree_block(root, eb, 0);
55c69072 369err:
19c00ddc
CM		 370 free_extent_buffer(eb);
371out:
372 return 0;
373}
374
2b82032c
YZ		 375static int check_tree_block_fsid(struct btrfs_root *root,
376 struct extent_buffer *eb)
377{
378 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
379 u8 fsid[BTRFS_UUID_SIZE];
380 int ret = 1;
381
382 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
383 BTRFS_FSID_SIZE);
384 while (fs_devices) {
385 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
386 ret = 0;
387 break;
388 }
389 fs_devices = fs_devices->seed;
390 }
391 return ret;
392}
393
4008c04a
CM		 394#ifdef CONFIG_DEBUG_LOCK_ALLOC
395void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
396{
397 lockdep_set_class_and_name(&eb->lock,
398 &btrfs_eb_class[level],
399 btrfs_eb_name[level]);
400}
401#endif
402
b2950863 403static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 404 struct extent_state *state)
405{
406 struct extent_io_tree *tree;
407 u64 found_start;
408 int found_level;
409 unsigned long len;
410 struct extent_buffer *eb;
411 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 412 int ret = 0;
ce9adaa5
CM		 413
414 tree = &BTRFS_I(page->mapping->host)->io_tree;
415 if (page->private == EXTENT_PAGE_PRIVATE)
416 goto out;
417 if (!page->private)
418 goto out;
d397712b 419
ce9adaa5 420 len = page->private >> 2;
d397712b
CM		 421 WARN_ON(len == 0);
422
ce9adaa5 423 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 424
ce9adaa5 425 found_start = btrfs_header_bytenr(eb);
23a07867 426 if (found_start != start) {
193f284d
CM		 427 if (printk_ratelimit()) {
428 printk(KERN_INFO "btrfs bad tree block start "
429 "%llu %llu\n",
430 (unsigned long long)found_start,
431 (unsigned long long)eb->start);
432 }
f188591e 433 ret = -EIO;
ce9adaa5
CM		 434 goto err;
435 }
436 if (eb->first_page != page) {
d397712b
CM		 437 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
438 eb->first_page->index, page->index);
ce9adaa5 439 WARN_ON(1);
f188591e 440 ret = -EIO;
ce9adaa5
CM		 441 goto err;
442 }
2b82032c 443 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 444 if (printk_ratelimit()) {
445 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
446 (unsigned long long)eb->start);
447 }
1259ab75
CM		 448 ret = -EIO;
449 goto err;
450 }
ce9adaa5
CM		 451 found_level = btrfs_header_level(eb);
452
4008c04a
CM		 453 btrfs_set_buffer_lockdep_class(eb, found_level);
454
ce9adaa5 455 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 456 if (ret)
457 ret = -EIO;
ce9adaa5
CM		 458
459 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
460 end = eb->start + end - 1;
ce9adaa5
CM		 461err:
462 free_extent_buffer(eb);
463out:
f188591e 464 return ret;
ce9adaa5
CM		 465}
466
ce9adaa5 467static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 468{
469 struct end_io_wq *end_io_wq = bio->bi_private;
470 struct btrfs_fs_info *fs_info;
ce9adaa5 471
ce9adaa5 472 fs_info = end_io_wq->info;
ce9adaa5 473 end_io_wq->error = err;
8b712842
CM		 474 end_io_wq->work.func = end_workqueue_fn;
475 end_io_wq->work.flags = 0;
d20f7043
CM		 476
477 if (bio->bi_rw & (1 << BIO_RW)) {
cad321ad
CM		 478 if (end_io_wq->metadata)
479 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
480 &end_io_wq->work);
481 else
482 btrfs_queue_worker(&fs_info->endio_write_workers,
483 &end_io_wq->work);
d20f7043
CM		 484 } else {
485 if (end_io_wq->metadata)
486 btrfs_queue_worker(&fs_info->endio_meta_workers,
487 &end_io_wq->work);
488 else
489 btrfs_queue_worker(&fs_info->endio_workers,
490 &end_io_wq->work);
491 }
ce9adaa5
CM		 492}
493
22c59948
CM		 494int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
495 int metadata)
0b86a832 496{
ce9adaa5 497 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 498 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
499 if (!end_io_wq)
500 return -ENOMEM;
501
502 end_io_wq->private = bio->bi_private;
503 end_io_wq->end_io = bio->bi_end_io;
22c59948 504 end_io_wq->info = info;
ce9adaa5
CM		 505 end_io_wq->error = 0;
506 end_io_wq->bio = bio;
22c59948 507 end_io_wq->metadata = metadata;
ce9adaa5
CM		 508
509 bio->bi_private = end_io_wq;
510 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 511 return 0;
512}
513
b64a2851 514unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 515{
4854ddd0
CM		 516 unsigned long limit = min_t(unsigned long,
517 info->workers.max_workers,
518 info->fs_devices->open_devices);
519 return 256 * limit;
520}
0986fe9e 521
4854ddd0
CM		 522int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
523{
b64a2851
CM		 524 return atomic_read(&info->nr_async_bios) >
525 btrfs_async_submit_limit(info);
0986fe9e
CM		 526}
527
4a69a410
CM		 528static void run_one_async_start(struct btrfs_work *work)
529{
530 struct btrfs_fs_info *fs_info;
531 struct async_submit_bio *async;
532
533 async = container_of(work, struct async_submit_bio, work);
534 fs_info = BTRFS_I(async->inode)->root->fs_info;
535 async->submit_bio_start(async->inode, async->rw, async->bio,
536 async->mirror_num, async->bio_flags);
537}
538
539static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 540{
541 struct btrfs_fs_info *fs_info;
542 struct async_submit_bio *async;
4854ddd0 543 int limit;
8b712842
CM		 544
545 async = container_of(work, struct async_submit_bio, work);
546 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 547
b64a2851 548 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 549 limit = limit * 2 / 3;
550
8b712842 551 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 552
b64a2851
CM		 553 if (atomic_read(&fs_info->nr_async_submits) < limit &&
554 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 555 wake_up(&fs_info->async_submit_wait);
556
4a69a410 557 async->submit_bio_done(async->inode, async->rw, async->bio,
c8b97818 558 async->mirror_num, async->bio_flags);
4a69a410
CM		 559}
560
561static void run_one_async_free(struct btrfs_work *work)
562{
563 struct async_submit_bio *async;
564
565 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 566 kfree(async);
567}
568
44b8bd7e
CM		 569int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
570 int rw, struct bio *bio, int mirror_num,
c8b97818 571 unsigned long bio_flags,
4a69a410
CM		 572 extent_submit_bio_hook_t *submit_bio_start,
573 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 574{
575 struct async_submit_bio *async;
576
577 async = kmalloc(sizeof(*async), GFP_NOFS);
578 if (!async)
579 return -ENOMEM;
580
581 async->inode = inode;
582 async->rw = rw;
583 async->bio = bio;
584 async->mirror_num = mirror_num;
4a69a410
CM		 585 async->submit_bio_start = submit_bio_start;
586 async->submit_bio_done = submit_bio_done;
587
588 async->work.func = run_one_async_start;
589 async->work.ordered_func = run_one_async_done;
590 async->work.ordered_free = run_one_async_free;
591
8b712842 592 async->work.flags = 0;
c8b97818 593 async->bio_flags = bio_flags;
8c8bee1d 594
cb03c743 595 atomic_inc(&fs_info->nr_async_submits);
d313d7a3
CM		 596
597 if (rw & (1 << BIO_RW_SYNCIO))
598 btrfs_set_work_high_prio(&async->work);
599
8b712842 600 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 601
d397712b 602 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 603 atomic_read(&fs_info->nr_async_submits)) {
604 wait_event(fs_info->async_submit_wait,
605 (atomic_read(&fs_info->nr_async_submits) == 0));
606 }
607
44b8bd7e
CM		 608 return 0;
609}
610
ce3ed71a
CM		 611static int btree_csum_one_bio(struct bio *bio)
612{
613 struct bio_vec *bvec = bio->bi_io_vec;
614 int bio_index = 0;
615 struct btrfs_root *root;
616
617 WARN_ON(bio->bi_vcnt <= 0);
d397712b 618 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 619 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
620 csum_dirty_buffer(root, bvec->bv_page);
621 bio_index++;
622 bvec++;
623 }
624 return 0;
625}
626
4a69a410
CM		 627static int __btree_submit_bio_start(struct inode *inode, int rw,
628 struct bio *bio, int mirror_num,
629 unsigned long bio_flags)
22c59948 630{
8b712842
CM		 631 /*
632 * when we're called for a write, we're already in the async
5443be45 633 * submission context. Just jump into btrfs_map_bio
8b712842 634 */
4a69a410
CM		 635 btree_csum_one_bio(bio);
636 return 0;
637}
22c59948 638
4a69a410
CM		 639static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
640 int mirror_num, unsigned long bio_flags)
641{
8b712842 642 /*
4a69a410
CM		 643 * when we're called for a write, we're already in the async
644 * submission context. Just jump into btrfs_map_bio
8b712842 645 */
8b712842 646 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 647}
648
44b8bd7e 649static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 650 int mirror_num, unsigned long bio_flags)
44b8bd7e 651{
cad321ad
CM		 652 int ret;
653
654 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
655 bio, 1);
656 BUG_ON(ret);
657
44b8bd7e 658 if (!(rw & (1 << BIO_RW))) {
4a69a410
CM		 659 /*
660 * called for a read, do the setup so that checksum validation
661 * can happen in the async kernel threads
662 */
4a69a410 663 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 664 mirror_num, 0);
44b8bd7e 665 }
d313d7a3 666
cad321ad
CM		 667 /*
668 * kthread helpers are used to submit writes so that checksumming
669 * can happen in parallel across all CPUs
670 */
44b8bd7e 671 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 672 inode, rw, bio, mirror_num, 0,
4a69a410
CM		 673 __btree_submit_bio_start,
674 __btree_submit_bio_done);
44b8bd7e
CM		 675}
676
0da5468f
CM		 677static int btree_writepage(struct page *page, struct writeback_control *wbc)
678{
d1310b2e 679 struct extent_io_tree *tree;
b9473439
CM		 680 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
681 struct extent_buffer *eb;
682 int was_dirty;
683
d1310b2e 684 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 685 if (!(current->flags & PF_MEMALLOC)) {
686 return extent_write_full_page(tree, page,
687 btree_get_extent, wbc);
688 }
5443be45 689
b9473439
CM		 690 redirty_page_for_writepage(wbc, page);
691 eb = btrfs_find_tree_block(root, page_offset(page),
692 PAGE_CACHE_SIZE);
693 WARN_ON(!eb);
694
695 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
696 if (!was_dirty) {
697 spin_lock(&root->fs_info->delalloc_lock);
698 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
699 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 700 }
b9473439
CM		 701 free_extent_buffer(eb);
702
703 unlock_page(page);
704 return 0;
5f39d397 705}
0da5468f
CM		 706
707static int btree_writepages(struct address_space *mapping,
708 struct writeback_control *wbc)
709{
d1310b2e
CM		 710 struct extent_io_tree *tree;
711 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 712 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 713 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 714 u64 num_dirty;
24ab9cd8 715 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 716
717 if (wbc->for_kupdate)
718 return 0;
719
b9473439
CM		 720 /* this is a bit racy, but that's ok */
721 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 722 if (num_dirty < thresh)
793955bc 723 return 0;
793955bc 724 }
0da5468f
CM		 725 return extent_writepages(tree, mapping, btree_get_extent, wbc);
726}
727
b2950863 728static int btree_readpage(struct file *file, struct page *page)
5f39d397 729{
d1310b2e
CM		 730 struct extent_io_tree *tree;
731 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 732 return extent_read_full_page(tree, page, btree_get_extent);
733}
22b0ebda 734
70dec807 735static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 736{
d1310b2e
CM		 737 struct extent_io_tree *tree;
738 struct extent_map_tree *map;
5f39d397 739 int ret;
d98237b3 740
98509cfc 741 if (PageWriteback(page) || PageDirty(page))
d397712b 742 return 0;
98509cfc 743
d1310b2e
CM		 744 tree = &BTRFS_I(page->mapping->host)->io_tree;
745 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 746
7b13b7b1 747 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 748 if (!ret)
6af118ce 749 return 0;
6af118ce
CM		 750
751 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 752 if (ret == 1) {
753 ClearPagePrivate(page);
754 set_page_private(page, 0);
755 page_cache_release(page);
756 }
6af118ce 757
d98237b3
CM		 758 return ret;
759}
760
5f39d397 761static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 762{
d1310b2e
CM		 763 struct extent_io_tree *tree;
764 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 765 extent_invalidatepage(tree, page, offset);
766 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 767 if (PagePrivate(page)) {
d397712b
CM		 768 printk(KERN_WARNING "btrfs warning page private not zero "
769 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 770 ClearPagePrivate(page);
771 set_page_private(page, 0);
772 page_cache_release(page);
773 }
d98237b3
CM		 774}
775
d98237b3
CM		 776static struct address_space_operations btree_aops = {
777 .readpage = btree_readpage,
778 .writepage = btree_writepage,
0da5468f 779 .writepages = btree_writepages,
5f39d397
CM		 780 .releasepage = btree_releasepage,
781 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 782 .sync_page = block_sync_page,
783};
784
ca7a79ad
CM		 785int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
786 u64 parent_transid)
090d1875 787{
5f39d397
CM		 788 struct extent_buffer *buf = NULL;
789 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 790 int ret = 0;
090d1875 791
db94535d 792 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 793 if (!buf)
090d1875 794 return 0;
d1310b2e 795 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 796 buf, 0, 0, btree_get_extent, 0);
5f39d397 797 free_extent_buffer(buf);
de428b63 798 return ret;
090d1875
CM		 799}
800
0999df54
CM		 801struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
802 u64 bytenr, u32 blocksize)
803{
804 struct inode *btree_inode = root->fs_info->btree_inode;
805 struct extent_buffer *eb;
806 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
807 bytenr, blocksize, GFP_NOFS);
808 return eb;
809}
810
811struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
812 u64 bytenr, u32 blocksize)
813{
814 struct inode *btree_inode = root->fs_info->btree_inode;
815 struct extent_buffer *eb;
816
817 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
818 bytenr, blocksize, NULL, GFP_NOFS);
819 return eb;
820}
821
822
e02119d5
CM		 823int btrfs_write_tree_block(struct extent_buffer *buf)
824{
825 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
24ab9cd8 826 buf->start + buf->len - 1, WB_SYNC_ALL);
e02119d5
CM		 827}
828
829int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
830{
831 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
35d62a94
CH		 832 buf->start >> PAGE_CACHE_SHIFT,
833 (buf->start + buf->len - 1) >>
834 PAGE_CACHE_SHIFT);
e02119d5
CM		 835}
836
0999df54 837struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 838 u32 blocksize, u64 parent_transid)
0999df54
CM		 839{
840 struct extent_buffer *buf = NULL;
841 struct inode *btree_inode = root->fs_info->btree_inode;
842 struct extent_io_tree *io_tree;
843 int ret;
844
845 io_tree = &BTRFS_I(btree_inode)->io_tree;
846
847 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
848 if (!buf)
849 return NULL;
0999df54 850
ca7a79ad 851 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 852
d397712b 853 if (ret == 0)
b4ce94de 854 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 855 return buf;
ce9adaa5 856
eb60ceac
CM		 857}
858
e089f05c 859int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 860 struct extent_buffer *buf)
ed2ff2cb 861{
5f39d397 862 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 863 if (btrfs_header_generation(buf) ==
925baedd 864 root->fs_info->running_transaction->transid) {
b9447ef8 865 btrfs_assert_tree_locked(buf);
b4ce94de 866
b9473439
CM		 867 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
868 spin_lock(&root->fs_info->delalloc_lock);
869 if (root->fs_info->dirty_metadata_bytes >= buf->len)
870 root->fs_info->dirty_metadata_bytes -= buf->len;
871 else
872 WARN_ON(1);
873 spin_unlock(&root->fs_info->delalloc_lock);
874 }
b4ce94de 875
b9473439
CM		 876 /* ugh, clear_extent_buffer_dirty needs to lock the page */
877 btrfs_set_lock_blocking(buf);
d1310b2e 878 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 879 buf);
925baedd 880 }
5f39d397
CM		 881 return 0;
882}
883
db94535d 884static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 885 u32 stripesize, struct btrfs_root *root,
9f5fae2f 886 struct btrfs_fs_info *fs_info,
e20d96d6 887 u64 objectid)
d97e63b6 888{
cfaa7295 889 root->node = NULL;
a28ec197 890 root->commit_root = NULL;
db94535d
CM		 891 root->sectorsize = sectorsize;
892 root->nodesize = nodesize;
893 root->leafsize = leafsize;
87ee04eb 894 root->stripesize = stripesize;
123abc88 895 root->ref_cows = 0;
0b86a832
CM		 896 root->track_dirty = 0;
897
9f5fae2f 898 root->fs_info = fs_info;
0f7d52f4
CM		 899 root->objectid = objectid;
900 root->last_trans = 0;
13a8a7c8 901 root->highest_objectid = 0;
58176a96 902 root->name = NULL;
4313b399 903 root->in_sysfs = 0;
5d4f98a2 904 root->inode_tree.rb_node = NULL;
0b86a832
CM		 905
906 INIT_LIST_HEAD(&root->dirty_list);
7b128766 907 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 908 INIT_LIST_HEAD(&root->root_list);
925baedd 909 spin_lock_init(&root->node_lock);
bcc63abb 910 spin_lock_init(&root->list_lock);
5d4f98a2 911 spin_lock_init(&root->inode_lock);
a2135011 912 mutex_init(&root->objectid_mutex);
e02119d5 913 mutex_init(&root->log_mutex);
7237f183
YZ		 914 init_waitqueue_head(&root->log_writer_wait);
915 init_waitqueue_head(&root->log_commit_wait[0]);
916 init_waitqueue_head(&root->log_commit_wait[1]);
917 atomic_set(&root->log_commit[0], 0);
918 atomic_set(&root->log_commit[1], 0);
919 atomic_set(&root->log_writers, 0);
920 root->log_batch = 0;
921 root->log_transid = 0;
d0c803c4
CM		 922 extent_io_tree_init(&root->dirty_log_pages,
923 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 924
3768f368
CM		 925 memset(&root->root_key, 0, sizeof(root->root_key));
926 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 927 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 928 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 929 root->defrag_trans_start = fs_info->generation;
58176a96 930 init_completion(&root->kobj_unregister);
6702ed49
CM		 931 root->defrag_running = 0;
932 root->defrag_level = 0;
4d775673 933 root->root_key.objectid = objectid;
3394e160
CM		 934 root->anon_super.s_root = NULL;
935 root->anon_super.s_dev = 0;
936 INIT_LIST_HEAD(&root->anon_super.s_list);
937 INIT_LIST_HEAD(&root->anon_super.s_instances);
938 init_rwsem(&root->anon_super.s_umount);
939
3768f368
CM		 940 return 0;
941}
942
db94535d 943static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 944 struct btrfs_fs_info *fs_info,
945 u64 objectid,
e20d96d6 946 struct btrfs_root *root)
3768f368
CM		 947{
948 int ret;
db94535d 949 u32 blocksize;
84234f3a 950 u64 generation;
3768f368 951
db94535d 952 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 953 tree_root->sectorsize, tree_root->stripesize,
954 root, fs_info, objectid);
3768f368
CM		 955 ret = btrfs_find_last_root(tree_root, objectid,
956 &root->root_item, &root->root_key);
4df27c4d
YZ		 957 if (ret > 0)
958 return -ENOENT;
3768f368
CM		 959 BUG_ON(ret);
960
84234f3a 961 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 962 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
963 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 964 blocksize, generation);
3768f368 965 BUG_ON(!root->node);
4df27c4d 966 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 967 return 0;
968}
969
e02119d5
CM		 970int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
971 struct btrfs_fs_info *fs_info)
972{
973 struct extent_buffer *eb;
d0c803c4
CM		 974 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
975 u64 start = 0;
976 u64 end = 0;
e02119d5
CM		 977 int ret;
978
d0c803c4 979 if (!log_root_tree)
e02119d5
CM		 980 return 0;
981
d397712b 982 while (1) {
d0c803c4
CM		 983 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
984 0, &start, &end, EXTENT_DIRTY);
985 if (ret)
986 break;
987
988 clear_extent_dirty(&log_root_tree->dirty_log_pages,
989 start, end, GFP_NOFS);
990 }
e02119d5
CM		 991 eb = fs_info->log_root_tree->node;
992
993 WARN_ON(btrfs_header_level(eb) != 0);
994 WARN_ON(btrfs_header_nritems(eb) != 0);
995
d00aff00
CM		 996 ret = btrfs_free_reserved_extent(fs_info->tree_root,
997 eb->start, eb->len);
e02119d5
CM		 998 BUG_ON(ret);
999
1000 free_extent_buffer(eb);
1001 kfree(fs_info->log_root_tree);
1002 fs_info->log_root_tree = NULL;
1003 return 0;
1004}
1005
7237f183
YZ		 1006static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1007 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1008{
1009 struct btrfs_root *root;
1010 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1011 struct extent_buffer *leaf;
e02119d5
CM		 1012
1013 root = kzalloc(sizeof(*root), GFP_NOFS);
1014 if (!root)
7237f183 1015 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1016
1017 __setup_root(tree_root->nodesize, tree_root->leafsize,
1018 tree_root->sectorsize, tree_root->stripesize,
1019 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1020
1021 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1022 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1023 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1024 /*
1025 * log trees do not get reference counted because they go away
1026 * before a real commit is actually done. They do store pointers
1027 * to file data extents, and those reference counts still get
1028 * updated (along with back refs to the log tree).
1029 */
e02119d5
CM		 1030 root->ref_cows = 0;
1031
5d4f98a2
YZ		 1032 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1033 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1034 if (IS_ERR(leaf)) {
1035 kfree(root);
1036 return ERR_CAST(leaf);
1037 }
e02119d5 1038
5d4f98a2
YZ		 1039 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1040 btrfs_set_header_bytenr(leaf, leaf->start);
1041 btrfs_set_header_generation(leaf, trans->transid);
1042 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1043 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1044 root->node = leaf;
e02119d5
CM		 1045
1046 write_extent_buffer(root->node, root->fs_info->fsid,
1047 (unsigned long)btrfs_header_fsid(root->node),
1048 BTRFS_FSID_SIZE);
1049 btrfs_mark_buffer_dirty(root->node);
1050 btrfs_tree_unlock(root->node);
7237f183
YZ		 1051 return root;
1052}
1053
1054int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1055 struct btrfs_fs_info *fs_info)
1056{
1057 struct btrfs_root *log_root;
1058
1059 log_root = alloc_log_tree(trans, fs_info);
1060 if (IS_ERR(log_root))
1061 return PTR_ERR(log_root);
1062 WARN_ON(fs_info->log_root_tree);
1063 fs_info->log_root_tree = log_root;
1064 return 0;
1065}
1066
1067int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1068 struct btrfs_root *root)
1069{
1070 struct btrfs_root *log_root;
1071 struct btrfs_inode_item *inode_item;
1072
1073 log_root = alloc_log_tree(trans, root->fs_info);
1074 if (IS_ERR(log_root))
1075 return PTR_ERR(log_root);
1076
1077 log_root->last_trans = trans->transid;
1078 log_root->root_key.offset = root->root_key.objectid;
1079
1080 inode_item = &log_root->root_item.inode;
1081 inode_item->generation = cpu_to_le64(1);
1082 inode_item->size = cpu_to_le64(3);
1083 inode_item->nlink = cpu_to_le32(1);
1084 inode_item->nbytes = cpu_to_le64(root->leafsize);
1085 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1086
5d4f98a2 1087 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1088
1089 WARN_ON(root->log_root);
1090 root->log_root = log_root;
1091 root->log_transid = 0;
e02119d5
CM		 1092 return 0;
1093}
1094
1095struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1096 struct btrfs_key *location)
1097{
1098 struct btrfs_root *root;
1099 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1100 struct btrfs_path *path;
5f39d397 1101 struct extent_buffer *l;
84234f3a 1102 u64 generation;
db94535d 1103 u32 blocksize;
0f7d52f4
CM		 1104 int ret = 0;
1105
5eda7b5e 1106 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1107 if (!root)
0f7d52f4 1108 return ERR_PTR(-ENOMEM);
0f7d52f4 1109 if (location->offset == (u64)-1) {
db94535d 1110 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1111 location->objectid, root);
1112 if (ret) {
0f7d52f4
CM		 1113 kfree(root);
1114 return ERR_PTR(ret);
1115 }
13a8a7c8 1116 goto out;
0f7d52f4
CM		 1117 }
1118
db94535d 1119 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1120 tree_root->sectorsize, tree_root->stripesize,
1121 root, fs_info, location->objectid);
0f7d52f4
CM		 1122
1123 path = btrfs_alloc_path();
1124 BUG_ON(!path);
1125 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1126 if (ret == 0) {
1127 l = path->nodes[0];
1128 read_extent_buffer(l, &root->root_item,
1129 btrfs_item_ptr_offset(l, path->slots[0]),
1130 sizeof(root->root_item));
1131 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1132 }
0f7d52f4
CM		 1133 btrfs_free_path(path);
1134 if (ret) {
13a8a7c8
YZ		 1135 if (ret > 0)
1136 ret = -ENOENT;
0f7d52f4
CM		 1137 return ERR_PTR(ret);
1138 }
13a8a7c8 1139
84234f3a 1140 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1141 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1142 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1143 blocksize, generation);
5d4f98a2 1144 root->commit_root = btrfs_root_node(root);
0f7d52f4 1145 BUG_ON(!root->node);
13a8a7c8
YZ		 1146out:
1147 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1148 root->ref_cows = 1;
13a8a7c8 1149
5eda7b5e
CM		 1150 return root;
1151}
1152
dc17ff8f
CM		 1153struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1154 u64 root_objectid)
1155{
1156 struct btrfs_root *root;
1157
1158 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1159 return fs_info->tree_root;
1160 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1161 return fs_info->extent_root;
1162
1163 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1164 (unsigned long)root_objectid);
1165 return root;
1166}
1167
edbd8d4e
CM		 1168struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1169 struct btrfs_key *location)
5eda7b5e
CM		 1170{
1171 struct btrfs_root *root;
1172 int ret;
1173
edbd8d4e
CM		 1174 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1175 return fs_info->tree_root;
1176 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1177 return fs_info->extent_root;
8f18cf13
CM		 1178 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1179 return fs_info->chunk_root;
1180 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1181 return fs_info->dev_root;
0403e47e
YZ		 1182 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1183 return fs_info->csum_root;
4df27c4d
YZ		 1184again:
1185 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1186 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1187 (unsigned long)location->objectid);
4df27c4d 1188 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1189 if (root)
1190 return root;
1191
4df27c4d
YZ		 1192 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1193 if (ret == 0)
1194 ret = -ENOENT;
1195 if (ret < 0)
1196 return ERR_PTR(ret);
1197
e02119d5 1198 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1199 if (IS_ERR(root))
1200 return root;
3394e160 1201
4df27c4d 1202 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
3394e160
CM		 1203 set_anon_super(&root->anon_super, NULL);
1204
4df27c4d
YZ		 1205 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1206 if (ret)
1207 goto fail;
1208
1209 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1210 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1211 (unsigned long)root->root_key.objectid,
0f7d52f4 1212 root);
4df27c4d
YZ		 1213 if (ret == 0)
1214 root->in_radix = 1;
1215 spin_unlock(&fs_info->fs_roots_radix_lock);
1216 radix_tree_preload_end();
0f7d52f4 1217 if (ret) {
4df27c4d
YZ		 1218 if (ret == -EEXIST) {
1219 free_fs_root(root);
1220 goto again;
1221 }
1222 goto fail;
0f7d52f4 1223 }
4df27c4d
YZ		 1224
1225 ret = btrfs_find_dead_roots(fs_info->tree_root,
1226 root->root_key.objectid);
1227 WARN_ON(ret);
1228
1229 if (!(fs_info->sb->s_flags & MS_RDONLY))
c146afad 1230 btrfs_orphan_cleanup(root);
4df27c4d 1231
edbd8d4e 1232 return root;
4df27c4d
YZ		 1233fail:
1234 free_fs_root(root);
1235 return ERR_PTR(ret);
edbd8d4e
CM		 1236}
1237
1238struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1239 struct btrfs_key *location,
1240 const char *name, int namelen)
1241{
4df27c4d
YZ		 1242 return btrfs_read_fs_root_no_name(fs_info, location);
1243#if 0
edbd8d4e
CM		 1244 struct btrfs_root *root;
1245 int ret;
1246
1247 root = btrfs_read_fs_root_no_name(fs_info, location);
1248 if (!root)
1249 return NULL;
58176a96 1250
4313b399
CM		 1251 if (root->in_sysfs)
1252 return root;
1253
58176a96
JB		 1254 ret = btrfs_set_root_name(root, name, namelen);
1255 if (ret) {
5f39d397 1256 free_extent_buffer(root->node);
58176a96
JB		 1257 kfree(root);
1258 return ERR_PTR(ret);
1259 }
4df27c4d 1260
58176a96
JB		 1261 ret = btrfs_sysfs_add_root(root);
1262 if (ret) {
5f39d397 1263 free_extent_buffer(root->node);
58176a96
JB		 1264 kfree(root->name);
1265 kfree(root);
1266 return ERR_PTR(ret);
1267 }
4313b399 1268 root->in_sysfs = 1;
0f7d52f4 1269 return root;
4df27c4d 1270#endif
0f7d52f4 1271}
04160088
CM		 1272
1273static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1274{
1275 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1276 int ret = 0;
04160088
CM		 1277 struct btrfs_device *device;
1278 struct backing_dev_info *bdi;
b7967db7 1279
c6e30871 1280 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1281 if (!device->bdev)
1282 continue;
04160088
CM		 1283 bdi = blk_get_backing_dev_info(device->bdev);
1284 if (bdi && bdi_congested(bdi, bdi_bits)) {
1285 ret = 1;
1286 break;
1287 }
1288 }
1289 return ret;
1290}
1291
38b66988
CM		 1292/*
1293 * this unplugs every device on the box, and it is only used when page
1294 * is null
1295 */
1296static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1297{
38b66988
CM		 1298 struct btrfs_device *device;
1299 struct btrfs_fs_info *info;
1300
1301 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1302 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1303 if (!device->bdev)
1304 continue;
1305
38b66988 1306 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1307 if (bdi->unplug_io_fn)
38b66988 1308 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1309 }
1310}
1311
b2950863 1312static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1313{
38b66988 1314 struct inode *inode;
f2d8d74d
CM		 1315 struct extent_map_tree *em_tree;
1316 struct extent_map *em;
bcbfce8a 1317 struct address_space *mapping;
38b66988
CM		 1318 u64 offset;
1319
bcbfce8a 1320 /* the generic O_DIRECT read code does this */
9f0ba5bd 1321 if (1 || !page) {
38b66988
CM		 1322 __unplug_io_fn(bdi, page);
1323 return;
1324 }
1325
bcbfce8a
CM		 1326 /*
1327 * page->mapping may change at any time. Get a consistent copy
1328 * and use that for everything below
1329 */
1330 smp_mb();
1331 mapping = page->mapping;
1332 if (!mapping)
1333 return;
1334
1335 inode = mapping->host;
240d5d48
CM		 1336
1337 /*
1338 * don't do the expensive searching for a small number of
1339 * devices
1340 */
1341 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1342 __unplug_io_fn(bdi, page);
1343 return;
1344 }
1345
38b66988 1346 offset = page_offset(page);
04160088 1347
f2d8d74d 1348 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1349 read_lock(&em_tree->lock);
f2d8d74d 1350 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1351 read_unlock(&em_tree->lock);
89642229
CM		 1352 if (!em) {
1353 __unplug_io_fn(bdi, page);
f2d8d74d 1354 return;
89642229 1355 }
f2d8d74d 1356
89642229
CM		 1357 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1358 free_extent_map(em);
1359 __unplug_io_fn(bdi, page);
1360 return;
1361 }
f2d8d74d
CM		 1362 offset = offset - em->start;
1363 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1364 em->block_start + offset, page);
1365 free_extent_map(em);
04160088
CM		 1366}
1367
ad081f14
JA		 1368/*
1369 * If this fails, caller must call bdi_destroy() to get rid of the
1370 * bdi again.
1371 */
04160088
CM		 1372static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1373{
ad081f14
JA		 1374 int err;
1375
1376 bdi->capabilities = BDI_CAP_MAP_COPY;
1377 err = bdi_init(bdi);
1378 if (err)
1379 return err;
1380
1381 err = bdi_register(bdi, NULL, "btrfs-%d",
1382 atomic_inc_return(&btrfs_bdi_num));
76dda93c
YZ		 1383 if (err) {
1384 bdi_destroy(bdi);
ad081f14 1385 return err;
76dda93c 1386 }
ad081f14 1387
4575c9cc 1388 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1389 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1390 bdi->unplug_io_data = info;
1391 bdi->congested_fn = btrfs_congested_fn;
1392 bdi->congested_data = info;
1393 return 0;
1394}
1395
ce9adaa5
CM		 1396static int bio_ready_for_csum(struct bio *bio)
1397{
1398 u64 length = 0;
1399 u64 buf_len = 0;
1400 u64 start = 0;
1401 struct page *page;
1402 struct extent_io_tree *io_tree = NULL;
1403 struct btrfs_fs_info *info = NULL;
1404 struct bio_vec *bvec;
1405 int i;
1406 int ret;
1407
1408 bio_for_each_segment(bvec, bio, i) {
1409 page = bvec->bv_page;
1410 if (page->private == EXTENT_PAGE_PRIVATE) {
1411 length += bvec->bv_len;
1412 continue;
1413 }
1414 if (!page->private) {
1415 length += bvec->bv_len;
1416 continue;
1417 }
1418 length = bvec->bv_len;
1419 buf_len = page->private >> 2;
1420 start = page_offset(page) + bvec->bv_offset;
1421 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1422 info = BTRFS_I(page->mapping->host)->root->fs_info;
1423 }
1424 /* are we fully contained in this bio? */
1425 if (buf_len <= length)
1426 return 1;
1427
1428 ret = extent_range_uptodate(io_tree, start + length,
1429 start + buf_len - 1);
ce9adaa5
CM		 1430 return ret;
1431}
1432
8b712842
CM		 1433/*
1434 * called by the kthread helper functions to finally call the bio end_io
1435 * functions. This is where read checksum verification actually happens
1436 */
1437static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1438{
ce9adaa5 1439 struct bio *bio;
8b712842
CM		 1440 struct end_io_wq *end_io_wq;
1441 struct btrfs_fs_info *fs_info;
ce9adaa5 1442 int error;
ce9adaa5 1443
8b712842
CM		 1444 end_io_wq = container_of(work, struct end_io_wq, work);
1445 bio = end_io_wq->bio;
1446 fs_info = end_io_wq->info;
ce9adaa5 1447
cad321ad 1448 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1449 * be checksummed at once. This makes sure the entire block is in
1450 * ram and up to date before trying to verify things. For
1451 * blocksize <= pagesize, it is basically a noop
1452 */
cad321ad
CM		 1453 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1454 !bio_ready_for_csum(bio)) {
d20f7043 1455 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1456 &end_io_wq->work);
1457 return;
1458 }
1459 error = end_io_wq->error;
1460 bio->bi_private = end_io_wq->private;
1461 bio->bi_end_io = end_io_wq->end_io;
1462 kfree(end_io_wq);
8b712842 1463 bio_endio(bio, error);
44b8bd7e
CM		 1464}
1465
a74a4b97
CM		 1466static int cleaner_kthread(void *arg)
1467{
1468 struct btrfs_root *root = arg;
1469
1470 do {
1471 smp_mb();
1472 if (root->fs_info->closing)
1473 break;
1474
1475 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1476
1477 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1478 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1479 btrfs_clean_old_snapshots(root);
1480 mutex_unlock(&root->fs_info->cleaner_mutex);
1481 }
a74a4b97
CM		 1482
1483 if (freezing(current)) {
1484 refrigerator();
1485 } else {
1486 smp_mb();
1487 if (root->fs_info->closing)
1488 break;
1489 set_current_state(TASK_INTERRUPTIBLE);
1490 schedule();
1491 __set_current_state(TASK_RUNNING);
1492 }
1493 } while (!kthread_should_stop());
1494 return 0;
1495}
1496
1497static int transaction_kthread(void *arg)
1498{
1499 struct btrfs_root *root = arg;
1500 struct btrfs_trans_handle *trans;
1501 struct btrfs_transaction *cur;
1502 unsigned long now;
1503 unsigned long delay;
1504 int ret;
1505
1506 do {
1507 smp_mb();
1508 if (root->fs_info->closing)
1509 break;
1510
1511 delay = HZ * 30;
1512 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1513 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1514
1515 mutex_lock(&root->fs_info->trans_mutex);
1516 cur = root->fs_info->running_transaction;
1517 if (!cur) {
1518 mutex_unlock(&root->fs_info->trans_mutex);
1519 goto sleep;
1520 }
31153d81 1521
a74a4b97
CM		 1522 now = get_seconds();
1523 if (now < cur->start_time || now - cur->start_time < 30) {
1524 mutex_unlock(&root->fs_info->trans_mutex);
1525 delay = HZ * 5;
1526 goto sleep;
1527 }
1528 mutex_unlock(&root->fs_info->trans_mutex);
a74a4b97
CM		 1529 trans = btrfs_start_transaction(root, 1);
1530 ret = btrfs_commit_transaction(trans, root);
56bec294 1531
a74a4b97
CM		 1532sleep:
1533 wake_up_process(root->fs_info->cleaner_kthread);
1534 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1535
1536 if (freezing(current)) {
1537 refrigerator();
1538 } else {
1539 if (root->fs_info->closing)
1540 break;
1541 set_current_state(TASK_INTERRUPTIBLE);
1542 schedule_timeout(delay);
1543 __set_current_state(TASK_RUNNING);
1544 }
1545 } while (!kthread_should_stop());
1546 return 0;
1547}
1548
8a4b83cc 1549struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1550 struct btrfs_fs_devices *fs_devices,
1551 char *options)
2e635a27 1552{
db94535d
CM		 1553 u32 sectorsize;
1554 u32 nodesize;
1555 u32 leafsize;
1556 u32 blocksize;
87ee04eb 1557 u32 stripesize;
84234f3a 1558 u64 generation;
f2b636e8 1559 u64 features;
3de4586c 1560 struct btrfs_key location;
a061fc8d 1561 struct buffer_head *bh;
e02119d5 1562 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1563 GFP_NOFS);
d20f7043
CM		 1564 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1565 GFP_NOFS);
e02119d5 1566 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1567 GFP_NOFS);
8790d502 1568 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1569 GFP_NOFS);
e02119d5 1570 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1571 GFP_NOFS);
e02119d5 1572 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1573 GFP_NOFS);
e02119d5
CM		 1574 struct btrfs_root *log_tree_root;
1575
eb60ceac 1576 int ret;
e58ca020 1577 int err = -EINVAL;
4543df7e 1578
2c90e5d6 1579 struct btrfs_super_block *disk_super;
8790d502 1580
0463bb4e 1581 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1582 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1583 err = -ENOMEM;
1584 goto fail;
1585 }
76dda93c
YZ		 1586
1587 ret = init_srcu_struct(&fs_info->subvol_srcu);
1588 if (ret) {
1589 err = ret;
1590 goto fail;
1591 }
1592
1593 ret = setup_bdi(fs_info, &fs_info->bdi);
1594 if (ret) {
1595 err = ret;
1596 goto fail_srcu;
1597 }
1598
1599 fs_info->btree_inode = new_inode(sb);
1600 if (!fs_info->btree_inode) {
1601 err = -ENOMEM;
1602 goto fail_bdi;
1603 }
1604
1605 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1606 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1607 INIT_LIST_HEAD(&fs_info->dead_roots);
19c00ddc 1608 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1609 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1610 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1611 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1612 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1613 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1614 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1615 spin_lock_init(&fs_info->fs_roots_radix_lock);
19c00ddc 1616
58176a96 1617 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1618 fs_info->tree_root = tree_root;
1619 fs_info->extent_root = extent_root;
d20f7043 1620 fs_info->csum_root = csum_root;
0b86a832
CM		 1621 fs_info->chunk_root = chunk_root;
1622 fs_info->dev_root = dev_root;
8a4b83cc 1623 fs_info->fs_devices = fs_devices;
0b86a832 1624 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1625 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1626 btrfs_mapping_init(&fs_info->mapping_tree);
cb03c743 1627 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1628 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1629 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1630 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1631 fs_info->sb = sb;
c59f8951 1632 fs_info->max_extent = (u64)-1;
6f568d35 1633 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1634 fs_info->metadata_ratio = 0;
c8b97818 1635
b34b086c
CM		 1636 fs_info->thread_pool_size = min_t(unsigned long,
1637 num_online_cpus() + 2, 8);
0afbaf8c 1638
3eaa2885
CM		 1639 INIT_LIST_HEAD(&fs_info->ordered_extents);
1640 spin_lock_init(&fs_info->ordered_extent_lock);
1641
a061fc8d
CM		 1642 sb->s_blocksize = 4096;
1643 sb->s_blocksize_bits = blksize_bits(4096);
1644
76dda93c
YZ		 1645 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1646 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1647 /*
1648 * we set the i_size on the btree inode to the max possible int.
1649 * the real end of the address space is determined by all of
1650 * the devices in the system
1651 */
1652 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1653 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1654 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1655
5d4f98a2 1656 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1657 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1658 fs_info->btree_inode->i_mapping,
1659 GFP_NOFS);
d1310b2e
CM		 1660 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1661 GFP_NOFS);
1662
1663 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1664
76dda93c
YZ		 1665 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1666 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1667 sizeof(struct btrfs_key));
1668 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1669 insert_inode_hash(fs_info->btree_inode);
76dda93c 1670
0f9dd46c
JB		 1671 spin_lock_init(&fs_info->block_group_cache_lock);
1672 fs_info->block_group_cache_tree.rb_node = NULL;
1673
11833d66 1674 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1675 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1676 extent_io_tree_init(&fs_info->freed_extents[1],
1677 fs_info->btree_inode->i_mapping, GFP_NOFS);
1678 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1679 fs_info->do_barriers = 1;
e18e4809 1680
39279cc3 1681
79154b1b 1682 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1683 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1684 mutex_init(&fs_info->tree_log_mutex);
925baedd 1685 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1686 mutex_init(&fs_info->transaction_kthread_mutex);
1687 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1688 mutex_init(&fs_info->volume_mutex);
276e680d 1689 init_rwsem(&fs_info->extent_commit_sem);
76dda93c 1690 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1691
1692 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1693 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1694
e6dcd2dc 1695 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1696 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1697 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1698
0b86a832 1699 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1700 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1701
d98237b3 1702
a512bbf8 1703 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1704 if (!bh)
39279cc3 1705 goto fail_iput;
39279cc3 1706
a061fc8d 1707 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1708 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1709 sizeof(fs_info->super_for_commit));
a061fc8d 1710 brelse(bh);
5f39d397 1711
a061fc8d 1712 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1713
5f39d397 1714 disk_super = &fs_info->super_copy;
0f7d52f4 1715 if (!btrfs_super_root(disk_super))
c6e2bac1 1716 goto fail_iput;
0f7d52f4 1717
2b82032c
YZ		 1718 ret = btrfs_parse_options(tree_root, options);
1719 if (ret) {
1720 err = ret;
c6e2bac1 1721 goto fail_iput;
2b82032c 1722 }
dfe25020 1723
f2b636e8
JB		 1724 features = btrfs_super_incompat_flags(disk_super) &
1725 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1726 if (features) {
1727 printk(KERN_ERR "BTRFS: couldn't mount because of "
1728 "unsupported optional features (%Lx).\n",
21380931 1729 (unsigned long long)features);
f2b636e8 1730 err = -EINVAL;
c6e2bac1 1731 goto fail_iput;
f2b636e8
JB		 1732 }
1733
5d4f98a2
YZ		 1734 features = btrfs_super_incompat_flags(disk_super);
1735 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1736 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1737 btrfs_set_super_incompat_flags(disk_super, features);
1738 }
1739
f2b636e8
JB		 1740 features = btrfs_super_compat_ro_flags(disk_super) &
1741 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1742 if (!(sb->s_flags & MS_RDONLY) && features) {
1743 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1744 "unsupported option features (%Lx).\n",
21380931 1745 (unsigned long long)features);
f2b636e8 1746 err = -EINVAL;
c6e2bac1 1747 goto fail_iput;
f2b636e8 1748 }
9042846b 1749printk("thread pool is %d\n", fs_info->thread_pool_size);
4543df7e
CM		 1750 /*
1751 * we need to start all the end_io workers up front because the
1752 * queue work function gets called at interrupt time, and so it
1753 * cannot dynamically grow.
1754 */
5443be45
CM		 1755 btrfs_init_workers(&fs_info->workers, "worker",
1756 fs_info->thread_pool_size);
c8b97818 1757
771ed689
CM		 1758 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1759 fs_info->thread_pool_size);
1760
5443be45 1761 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209
CM		 1762 min_t(u64, fs_devices->num_devices,
1763 fs_info->thread_pool_size));
61b49440
CM		 1764
1765 /* a higher idle thresh on the submit workers makes it much more
1766 * likely that bios will be send down in a sane order to the
1767 * devices
1768 */
1769 fs_info->submit_workers.idle_thresh = 64;
53863232 1770
771ed689 1771 fs_info->workers.idle_thresh = 16;
4a69a410 1772 fs_info->workers.ordered = 1;
61b49440 1773
771ed689
CM		 1774 fs_info->delalloc_workers.idle_thresh = 2;
1775 fs_info->delalloc_workers.ordered = 1;
1776
5443be45
CM		 1777 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1778 btrfs_init_workers(&fs_info->endio_workers, "endio",
1779 fs_info->thread_pool_size);
d20f7043
CM		 1780 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1781 fs_info->thread_pool_size);
cad321ad
CM		 1782 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1783 "endio-meta-write", fs_info->thread_pool_size);
5443be45 1784 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
e6dcd2dc 1785 fs_info->thread_pool_size);
61b49440
CM		 1786
1787 /*
1788 * endios are largely parallel and should have a very
1789 * low idle thresh
1790 */
1791 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1792 fs_info->endio_meta_workers.idle_thresh = 4;
1793
9042846b
CM		 1794 fs_info->endio_write_workers.idle_thresh = 2;
1795 fs_info->endio_meta_write_workers.idle_thresh = 2;
1796
1797 fs_info->endio_workers.atomic_worker_start = 1;
1798 fs_info->endio_meta_workers.atomic_worker_start = 1;
1799 fs_info->endio_write_workers.atomic_worker_start = 1;
1800 fs_info->endio_meta_write_workers.atomic_worker_start = 1;
61b49440 1801
4543df7e 1802 btrfs_start_workers(&fs_info->workers, 1);
1cc127b5 1803 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1804 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1805 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1806 btrfs_start_workers(&fs_info->endio_workers, 1);
1807 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1808 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1809 btrfs_start_workers(&fs_info->endio_write_workers, 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
9078a3e1
CM		 1912 btrfs_read_block_groups(extent_root);
1913
2d69a0f8 1914 fs_info->generation = generation;
c146afad 1915 fs_info->last_trans_committed = generation;
d18a2c44
CM		 1916 fs_info->data_alloc_profile = (u64)-1;
1917 fs_info->metadata_alloc_profile = (u64)-1;
1918 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
a74a4b97
CM		 1919 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1920 "btrfs-cleaner");
57506d50 1921 if (IS_ERR(fs_info->cleaner_kthread))
d20f7043 1922 goto fail_csum_root;
a74a4b97
CM		 1923
1924 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1925 tree_root,
1926 "btrfs-transaction");
57506d50 1927 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1928 goto fail_cleaner;
a74a4b97 1929
c289811c
CM		 1930 if (!btrfs_test_opt(tree_root, SSD) &&
1931 !btrfs_test_opt(tree_root, NOSSD) &&
1932 !fs_info->fs_devices->rotating) {
1933 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1934 "mode\n");
1935 btrfs_set_opt(fs_info->mount_opt, SSD);
1936 }
1937
e02119d5 1938 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1939 u64 bytenr = btrfs_super_log_root(disk_super);
1940
7c2ca468 1941 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1942 printk(KERN_WARNING "Btrfs log replay required "
1943 "on RO media\n");
7c2ca468
CM		 1944 err = -EIO;
1945 goto fail_trans_kthread;
1946 }
e02119d5
CM		 1947 blocksize =
1948 btrfs_level_size(tree_root,
1949 btrfs_super_log_root_level(disk_super));
d18a2c44 1950
e02119d5
CM		 1951 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1952 GFP_NOFS);
1953
1954 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1955 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1956
1957 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1958 blocksize,
1959 generation + 1);
e02119d5
CM		 1960 ret = btrfs_recover_log_trees(log_tree_root);
1961 BUG_ON(ret);
e556ce2c
YZ		 1962
1963 if (sb->s_flags & MS_RDONLY) {
1964 ret = btrfs_commit_super(tree_root);
1965 BUG_ON(ret);
1966 }
e02119d5 1967 }
1a40e23b 1968
76dda93c
YZ		 1969 ret = btrfs_find_orphan_roots(tree_root);
1970 BUG_ON(ret);
1971
7c2ca468 1972 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1973 ret = btrfs_recover_relocation(tree_root);
7c2ca468
CM		 1974 BUG_ON(ret);
1975 }
1a40e23b 1976
3de4586c
CM		 1977 location.objectid = BTRFS_FS_TREE_OBJECTID;
1978 location.type = BTRFS_ROOT_ITEM_KEY;
1979 location.offset = (u64)-1;
1980
3de4586c
CM		 1981 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1982 if (!fs_info->fs_root)
7c2ca468 1983 goto fail_trans_kthread;
c289811c 1984
0f7d52f4 1985 return tree_root;
39279cc3 1986
7c2ca468
CM		 1987fail_trans_kthread:
1988 kthread_stop(fs_info->transaction_kthread);
3f157a2f 1989fail_cleaner:
a74a4b97 1990 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 1991
1992 /*
1993 * make sure we're done with the btree inode before we stop our
1994 * kthreads
1995 */
1996 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1997 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1998
d20f7043
CM		 1999fail_csum_root:
2000 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2001 free_extent_buffer(csum_root->commit_root);
2002fail_dev_root:
2003 free_extent_buffer(dev_root->node);
2004 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2005fail_extent_root:
2006 free_extent_buffer(extent_root->node);
5d4f98a2 2007 free_extent_buffer(extent_root->commit_root);
39279cc3 2008fail_tree_root:
5f39d397 2009 free_extent_buffer(tree_root->node);
5d4f98a2 2010 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2011fail_chunk_root:
2012 free_extent_buffer(chunk_root->node);
5d4f98a2 2013 free_extent_buffer(chunk_root->commit_root);
39279cc3 2014fail_sb_buffer:
247e743c 2015 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2016 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2017 btrfs_stop_workers(&fs_info->workers);
2018 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2019 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2020 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2021 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2022 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2023fail_iput:
7c2ca468 2024 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2025 iput(fs_info->btree_inode);
7e662854 2026
dfe25020 2027 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2028 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2029fail_bdi:
7e662854 2030 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2031fail_srcu:
2032 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2033fail:
39279cc3
CM		 2034 kfree(extent_root);
2035 kfree(tree_root);
2036 kfree(fs_info);
83afeac4
JM		 2037 kfree(chunk_root);
2038 kfree(dev_root);
d20f7043 2039 kfree(csum_root);
39279cc3 2040 return ERR_PTR(err);
eb60ceac
CM		 2041}
2042
f2984462
CM		 2043static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2044{
2045 char b[BDEVNAME_SIZE];
2046
2047 if (uptodate) {
2048 set_buffer_uptodate(bh);
2049 } else {
2050 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2051 printk(KERN_WARNING "lost page write due to "
2052 "I/O error on %s\n",
2053 bdevname(bh->b_bdev, b));
2054 }
1259ab75
CM		 2055 /* note, we dont' set_buffer_write_io_error because we have
2056 * our own ways of dealing with the IO errors
2057 */
f2984462
CM		 2058 clear_buffer_uptodate(bh);
2059 }
2060 unlock_buffer(bh);
2061 put_bh(bh);
2062}
2063
a512bbf8
YZ		 2064struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2065{
2066 struct buffer_head *bh;
2067 struct buffer_head *latest = NULL;
2068 struct btrfs_super_block *super;
2069 int i;
2070 u64 transid = 0;
2071 u64 bytenr;
2072
2073 /* we would like to check all the supers, but that would make
2074 * a btrfs mount succeed after a mkfs from a different FS.
2075 * So, we need to add a special mount option to scan for
2076 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2077 */
2078 for (i = 0; i < 1; i++) {
2079 bytenr = btrfs_sb_offset(i);
2080 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2081 break;
2082 bh = __bread(bdev, bytenr / 4096, 4096);
2083 if (!bh)
2084 continue;
2085
2086 super = (struct btrfs_super_block *)bh->b_data;
2087 if (btrfs_super_bytenr(super) != bytenr ||
2088 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2089 sizeof(super->magic))) {
2090 brelse(bh);
2091 continue;
2092 }
2093
2094 if (!latest || btrfs_super_generation(super) > transid) {
2095 brelse(latest);
2096 latest = bh;
2097 transid = btrfs_super_generation(super);
2098 } else {
2099 brelse(bh);
2100 }
2101 }
2102 return latest;
2103}
2104
4eedeb75
HH		 2105/*
2106 * this should be called twice, once with wait == 0 and
2107 * once with wait == 1. When wait == 0 is done, all the buffer heads
2108 * we write are pinned.
2109 *
2110 * They are released when wait == 1 is done.
2111 * max_mirrors must be the same for both runs, and it indicates how
2112 * many supers on this one device should be written.
2113 *
2114 * max_mirrors == 0 means to write them all.
2115 */
a512bbf8
YZ		 2116static int write_dev_supers(struct btrfs_device *device,
2117 struct btrfs_super_block *sb,
2118 int do_barriers, int wait, int max_mirrors)
2119{
2120 struct buffer_head *bh;
2121 int i;
2122 int ret;
2123 int errors = 0;
2124 u32 crc;
2125 u64 bytenr;
2126 int last_barrier = 0;
2127
2128 if (max_mirrors == 0)
2129 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2130
2131 /* make sure only the last submit_bh does a barrier */
2132 if (do_barriers) {
2133 for (i = 0; i < max_mirrors; i++) {
2134 bytenr = btrfs_sb_offset(i);
2135 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2136 device->total_bytes)
2137 break;
2138 last_barrier = i;
2139 }
2140 }
2141
2142 for (i = 0; i < max_mirrors; i++) {
2143 bytenr = btrfs_sb_offset(i);
2144 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2145 break;
2146
2147 if (wait) {
2148 bh = __find_get_block(device->bdev, bytenr / 4096,
2149 BTRFS_SUPER_INFO_SIZE);
2150 BUG_ON(!bh);
a512bbf8 2151 wait_on_buffer(bh);
4eedeb75
HH		 2152 if (!buffer_uptodate(bh))
2153 errors++;
2154
2155 /* drop our reference */
2156 brelse(bh);
2157
2158 /* drop the reference from the wait == 0 run */
2159 brelse(bh);
2160 continue;
a512bbf8
YZ		 2161 } else {
2162 btrfs_set_super_bytenr(sb, bytenr);
2163
2164 crc = ~(u32)0;
2165 crc = btrfs_csum_data(NULL, (char *)sb +
2166 BTRFS_CSUM_SIZE, crc,
2167 BTRFS_SUPER_INFO_SIZE -
2168 BTRFS_CSUM_SIZE);
2169 btrfs_csum_final(crc, sb->csum);
2170
4eedeb75
HH		 2171 /*
2172 * one reference for us, and we leave it for the
2173 * caller
2174 */
a512bbf8
YZ		 2175 bh = __getblk(device->bdev, bytenr / 4096,
2176 BTRFS_SUPER_INFO_SIZE);
2177 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2178
4eedeb75 2179 /* one reference for submit_bh */
a512bbf8 2180 get_bh(bh);
4eedeb75
HH		 2181
2182 set_buffer_uptodate(bh);
a512bbf8
YZ		 2183 lock_buffer(bh);
2184 bh->b_end_io = btrfs_end_buffer_write_sync;
2185 }
2186
2187 if (i == last_barrier && do_barriers && device->barriers) {
2188 ret = submit_bh(WRITE_BARRIER, bh);
2189 if (ret == -EOPNOTSUPP) {
2190 printk("btrfs: disabling barriers on dev %s\n",
2191 device->name);
2192 set_buffer_uptodate(bh);
2193 device->barriers = 0;
4eedeb75 2194 /* one reference for submit_bh */
a512bbf8
YZ		 2195 get_bh(bh);
2196 lock_buffer(bh);
ffbd517d 2197 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2198 }
2199 } else {
ffbd517d 2200 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2201 }
2202
4eedeb75 2203 if (ret)
a512bbf8 2204 errors++;
a512bbf8
YZ		 2205 }
2206 return errors < i ? 0 : -1;
2207}
2208
2209int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2210{
e5e9a520 2211 struct list_head *head;
f2984462 2212 struct btrfs_device *dev;
a061fc8d 2213 struct btrfs_super_block *sb;
f2984462 2214 struct btrfs_dev_item *dev_item;
f2984462
CM		 2215 int ret;
2216 int do_barriers;
a236aed1
CM		 2217 int max_errors;
2218 int total_errors = 0;
a061fc8d 2219 u64 flags;
f2984462 2220
a236aed1 2221 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2222 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2223
a061fc8d
CM		 2224 sb = &root->fs_info->super_for_commit;
2225 dev_item = &sb->dev_item;
e5e9a520
CM		 2226
2227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2228 head = &root->fs_info->fs_devices->devices;
c6e30871 2229 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2230 if (!dev->bdev) {
2231 total_errors++;
2232 continue;
2233 }
2b82032c 2234 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2235 continue;
2236
2b82032c 2237 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2238 btrfs_set_stack_device_type(dev_item, dev->type);
2239 btrfs_set_stack_device_id(dev_item, dev->devid);
2240 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2241 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2242 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2243 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2244 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2245 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2246 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2247
a061fc8d
CM		 2248 flags = btrfs_super_flags(sb);
2249 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2250
a512bbf8 2251 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2252 if (ret)
2253 total_errors++;
f2984462 2254 }
a236aed1 2255 if (total_errors > max_errors) {
d397712b
CM		 2256 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2257 total_errors);
a236aed1
CM		 2258 BUG();
2259 }
f2984462 2260
a512bbf8 2261 total_errors = 0;
c6e30871 2262 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2263 if (!dev->bdev)
2264 continue;
2b82032c 2265 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2266 continue;
2267
a512bbf8
YZ		 2268 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2269 if (ret)
2270 total_errors++;
f2984462 2271 }
e5e9a520 2272 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2273 if (total_errors > max_errors) {
d397712b
CM		 2274 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2275 total_errors);
a236aed1
CM		 2276 BUG();
2277 }
f2984462
CM		 2278 return 0;
2279}
2280
a512bbf8
YZ		 2281int write_ctree_super(struct btrfs_trans_handle *trans,
2282 struct btrfs_root *root, int max_mirrors)
eb60ceac 2283{
e66f709b 2284 int ret;
5f39d397 2285
a512bbf8 2286 ret = write_all_supers(root, max_mirrors);
5f39d397 2287 return ret;
cfaa7295
CM		 2288}
2289
5eda7b5e 2290int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2291{
4df27c4d 2292 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2293 radix_tree_delete(&fs_info->fs_roots_radix,
2294 (unsigned long)root->root_key.objectid);
4df27c4d 2295 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2296
2297 if (btrfs_root_refs(&root->root_item) == 0)
2298 synchronize_srcu(&fs_info->subvol_srcu);
2299
4df27c4d
YZ		 2300 free_fs_root(root);
2301 return 0;
2302}
2303
2304static void free_fs_root(struct btrfs_root *root)
2305{
2306 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2307 if (root->anon_super.s_dev) {
2308 down_write(&root->anon_super.s_umount);
2309 kill_anon_super(&root->anon_super);
2310 }
4df27c4d
YZ		 2311 free_extent_buffer(root->node);
2312 free_extent_buffer(root->commit_root);
d397712b 2313 kfree(root->name);
2619ba1f 2314 kfree(root);
2619ba1f
CM		 2315}
2316
35b7e476 2317static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2318{
2319 int ret;
2320 struct btrfs_root *gang[8];
2321 int i;
2322
76dda93c
YZ		 2323 while (!list_empty(&fs_info->dead_roots)) {
2324 gang[0] = list_entry(fs_info->dead_roots.next,
2325 struct btrfs_root, root_list);
2326 list_del(&gang[0]->root_list);
2327
2328 if (gang[0]->in_radix) {
2329 btrfs_free_fs_root(fs_info, gang[0]);
2330 } else {
2331 free_extent_buffer(gang[0]->node);
2332 free_extent_buffer(gang[0]->commit_root);
2333 kfree(gang[0]);
2334 }
2335 }
2336
d397712b 2337 while (1) {
0f7d52f4
CM		 2338 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2339 (void **)gang, 0,
2340 ARRAY_SIZE(gang));
2341 if (!ret)
2342 break;
2619ba1f 2343 for (i = 0; i < ret; i++)
5eda7b5e 2344 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2345 }
2346 return 0;
2347}
b4100d64 2348
c146afad 2349int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2350{
c146afad
YZ		 2351 u64 root_objectid = 0;
2352 struct btrfs_root *gang[8];
2353 int i;
3768f368 2354 int ret;
e089f05c 2355
c146afad
YZ		 2356 while (1) {
2357 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2358 (void **)gang, root_objectid,
2359 ARRAY_SIZE(gang));
2360 if (!ret)
2361 break;
5d4f98a2
YZ		 2362
2363 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2364 for (i = 0; i < ret; i++) {
2365 root_objectid = gang[i]->root_key.objectid;
c146afad
YZ		 2366 btrfs_orphan_cleanup(gang[i]);
2367 }
2368 root_objectid++;
2369 }
2370 return 0;
2371}
a2135011 2372
c146afad
YZ		 2373int btrfs_commit_super(struct btrfs_root *root)
2374{
2375 struct btrfs_trans_handle *trans;
2376 int ret;
a74a4b97 2377
c146afad 2378 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 2379 btrfs_clean_old_snapshots(root);
c146afad 2380 mutex_unlock(&root->fs_info->cleaner_mutex);
79154b1b 2381 trans = btrfs_start_transaction(root, 1);
54aa1f4d 2382 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2383 BUG_ON(ret);
2384 /* run commit again to drop the original snapshot */
79154b1b
CM		 2385 trans = btrfs_start_transaction(root, 1);
2386 btrfs_commit_transaction(trans, root);
2387 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2388 BUG_ON(ret);
d6bfde87 2389
a512bbf8 2390 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2391 return ret;
2392}
2393
2394int close_ctree(struct btrfs_root *root)
2395{
2396 struct btrfs_fs_info *fs_info = root->fs_info;
2397 int ret;
2398
2399 fs_info->closing = 1;
2400 smp_mb();
2401
2402 kthread_stop(root->fs_info->transaction_kthread);
2403 kthread_stop(root->fs_info->cleaner_kthread);
2404
2405 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2406 ret = btrfs_commit_super(root);
d397712b
CM		 2407 if (ret)
2408 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2409 }
0f7d52f4 2410
f25784b3
YZ		 2411 fs_info->closing = 2;
2412 smp_mb();
2413
b0c68f8b 2414 if (fs_info->delalloc_bytes) {
d397712b 2415 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2416 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2417 }
31153d81 2418 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2419 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2420 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2421 }
bcc63abb 2422
5d4f98a2
YZ		 2423 free_extent_buffer(fs_info->extent_root->node);
2424 free_extent_buffer(fs_info->extent_root->commit_root);
2425 free_extent_buffer(fs_info->tree_root->node);
2426 free_extent_buffer(fs_info->tree_root->commit_root);
2427 free_extent_buffer(root->fs_info->chunk_root->node);
2428 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2429 free_extent_buffer(root->fs_info->dev_root->node);
2430 free_extent_buffer(root->fs_info->dev_root->commit_root);
2431 free_extent_buffer(root->fs_info->csum_root->node);
2432 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2433
9078a3e1 2434 btrfs_free_block_groups(root->fs_info);
d10c5f31 2435
c146afad 2436 del_fs_roots(fs_info);
d10c5f31 2437
c146afad 2438 iput(fs_info->btree_inode);
9ad6b7bc 2439
247e743c 2440 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2441 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2442 btrfs_stop_workers(&fs_info->workers);
2443 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2444 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2445 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2446 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2447 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2448
dfe25020 2449 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2450 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2451
04160088 2452 bdi_destroy(&fs_info->bdi);
76dda93c 2453 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2454
0f7d52f4 2455 kfree(fs_info->extent_root);
0f7d52f4 2456 kfree(fs_info->tree_root);
0b86a832
CM		 2457 kfree(fs_info->chunk_root);
2458 kfree(fs_info->dev_root);
d20f7043 2459 kfree(fs_info->csum_root);
eb60ceac
CM		 2460 return 0;
2461}
2462
1259ab75 2463int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2464{
1259ab75 2465 int ret;
810191ff 2466 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75
CM		 2467
2468 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2469 if (!ret)
2470 return ret;
2471
2472 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2473 parent_transid);
2474 return !ret;
5f39d397
CM		 2475}
2476
2477int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2478{
810191ff 2479 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2480 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2481 buf);
2482}
6702ed49 2483
5f39d397
CM		 2484void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2485{
810191ff 2486 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2487 u64 transid = btrfs_header_generation(buf);
2488 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2489 int was_dirty;
b4ce94de 2490
b9447ef8 2491 btrfs_assert_tree_locked(buf);
ccd467d6 2492 if (transid != root->fs_info->generation) {
d397712b
CM		 2493 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2494 "found %llu running %llu\n",
db94535d 2495 (unsigned long long)buf->start,
d397712b
CM		 2496 (unsigned long long)transid,
2497 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2498 WARN_ON(1);
2499 }
b9473439
CM		 2500 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2501 buf);
2502 if (!was_dirty) {
2503 spin_lock(&root->fs_info->delalloc_lock);
2504 root->fs_info->dirty_metadata_bytes += buf->len;
2505 spin_unlock(&root->fs_info->delalloc_lock);
2506 }
eb60ceac
CM		 2507}
2508
d3c2fdcf 2509void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2510{
188de649
CM		 2511 /*
2512 * looks as though older kernels can get into trouble with
2513 * this code, they end up stuck in balance_dirty_pages forever
2514 */
d6bfde87 2515 u64 num_dirty;
771ed689 2516 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2517
6933c02e 2518 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2519 return;
2520
585ad2c3
CM		 2521 num_dirty = root->fs_info->dirty_metadata_bytes;
2522
d6bfde87
CM		 2523 if (num_dirty > thresh) {
2524 balance_dirty_pages_ratelimited_nr(
d7fc640e 2525 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2526 }
188de649 2527 return;
35b7e476 2528}
6b80053d 2529
ca7a79ad 2530int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2531{
810191ff 2532 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2533 int ret;
ca7a79ad 2534 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2535 if (ret == 0)
b4ce94de 2536 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2537 return ret;
6b80053d 2538}
0da5468f 2539
4bef0848
CM		 2540int btree_lock_page_hook(struct page *page)
2541{
2542 struct inode *inode = page->mapping->host;
b9473439 2543 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2544 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2545 struct extent_buffer *eb;
2546 unsigned long len;
2547 u64 bytenr = page_offset(page);
2548
2549 if (page->private == EXTENT_PAGE_PRIVATE)
2550 goto out;
2551
2552 len = page->private >> 2;
2553 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2554 if (!eb)
2555 goto out;
2556
2557 btrfs_tree_lock(eb);
4bef0848 2558 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2559
2560 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2561 spin_lock(&root->fs_info->delalloc_lock);
2562 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2563 root->fs_info->dirty_metadata_bytes -= eb->len;
2564 else
2565 WARN_ON(1);
2566 spin_unlock(&root->fs_info->delalloc_lock);
2567 }
2568
4bef0848
CM		 2569 btrfs_tree_unlock(eb);
2570 free_extent_buffer(eb);
2571out:
2572 lock_page(page);
2573 return 0;
2574}
2575
d1310b2e 2576static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2577 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2578 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2579 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2580 /* note we're sharing with inode.c for the merge bio hook */
2581 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2582};
Linux 6.x block layer and io_uring tree(s)