projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi-misc-2.6
[linux-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);
ce9adaa5 44
ad081f14
JA		 45static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
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
d1310b2e
CM		 126 spin_lock(&em_tree->lock);
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;
131 spin_unlock(&em_tree->lock);
5f39d397 132 goto out;
a061fc8d
CM		 133 }
134 spin_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
CM		 146
147 spin_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 }
7b13b7b1
CM		 166 spin_unlock(&em_tree->lock);
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{
265 int ret;
266
267 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
268 return 0;
269
270 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
271 if (extent_buffer_uptodate(io_tree, eb) &&
272 btrfs_header_generation(eb) == parent_transid) {
273 ret = 0;
274 goto out;
275 }
193f284d
CM		 276 if (printk_ratelimit()) {
277 printk("parent transid verify failed on %llu wanted %llu "
278 "found %llu\n",
279 (unsigned long long)eb->start,
280 (unsigned long long)parent_transid,
281 (unsigned long long)btrfs_header_generation(eb));
282 }
1259ab75 283 ret = 1;
1259ab75 284 clear_extent_buffer_uptodate(io_tree, eb);
33958dc6 285out:
1259ab75
CM		 286 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
287 GFP_NOFS);
288 return ret;
1259ab75
CM		 289}
290
d352ac68
CM		 291/*
292 * helper to read a given tree block, doing retries as required when
293 * the checksums don't match and we have alternate mirrors to try.
294 */
f188591e
CM		 295static int btree_read_extent_buffer_pages(struct btrfs_root *root,
296 struct extent_buffer *eb,
ca7a79ad 297 u64 start, u64 parent_transid)
f188591e
CM		 298{
299 struct extent_io_tree *io_tree;
300 int ret;
301 int num_copies = 0;
302 int mirror_num = 0;
303
304 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
305 while (1) {
306 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
307 btree_get_extent, mirror_num);
1259ab75
CM		 308 if (!ret &&
309 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 310 return ret;
d397712b 311
f188591e
CM		 312 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
313 eb->start, eb->len);
4235298e 314 if (num_copies == 1)
f188591e 315 return ret;
4235298e 316
f188591e 317 mirror_num++;
4235298e 318 if (mirror_num > num_copies)
f188591e 319 return ret;
f188591e 320 }
f188591e
CM		 321 return -EIO;
322}
19c00ddc 323
d352ac68 324/*
d397712b
CM		 325 * checksum a dirty tree block before IO. This has extra checks to make sure
326 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 327 */
d397712b 328
b2950863 329static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 330{
d1310b2e 331 struct extent_io_tree *tree;
35ebb934 332 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 333 u64 found_start;
334 int found_level;
335 unsigned long len;
336 struct extent_buffer *eb;
f188591e
CM		 337 int ret;
338
d1310b2e 339 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 340
341 if (page->private == EXTENT_PAGE_PRIVATE)
342 goto out;
343 if (!page->private)
344 goto out;
345 len = page->private >> 2;
d397712b
CM		 346 WARN_ON(len == 0);
347
19c00ddc 348 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 349 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
350 btrfs_header_generation(eb));
f188591e 351 BUG_ON(ret);
19c00ddc
CM		 352 found_start = btrfs_header_bytenr(eb);
353 if (found_start != start) {
55c69072
CM		 354 WARN_ON(1);
355 goto err;
356 }
357 if (eb->first_page != page) {
55c69072
CM		 358 WARN_ON(1);
359 goto err;
360 }
361 if (!PageUptodate(page)) {
55c69072
CM		 362 WARN_ON(1);
363 goto err;
19c00ddc
CM		 364 }
365 found_level = btrfs_header_level(eb);
4bef0848 366
19c00ddc 367 csum_tree_block(root, eb, 0);
55c69072 368err:
19c00ddc
CM		 369 free_extent_buffer(eb);
370out:
371 return 0;
372}
373
2b82032c
YZ		 374static int check_tree_block_fsid(struct btrfs_root *root,
375 struct extent_buffer *eb)
376{
377 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
378 u8 fsid[BTRFS_UUID_SIZE];
379 int ret = 1;
380
381 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
382 BTRFS_FSID_SIZE);
383 while (fs_devices) {
384 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
385 ret = 0;
386 break;
387 }
388 fs_devices = fs_devices->seed;
389 }
390 return ret;
391}
392
4008c04a
CM		 393#ifdef CONFIG_DEBUG_LOCK_ALLOC
394void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
395{
396 lockdep_set_class_and_name(&eb->lock,
397 &btrfs_eb_class[level],
398 btrfs_eb_name[level]);
399}
400#endif
401
b2950863 402static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 403 struct extent_state *state)
404{
405 struct extent_io_tree *tree;
406 u64 found_start;
407 int found_level;
408 unsigned long len;
409 struct extent_buffer *eb;
410 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 411 int ret = 0;
ce9adaa5
CM		 412
413 tree = &BTRFS_I(page->mapping->host)->io_tree;
414 if (page->private == EXTENT_PAGE_PRIVATE)
415 goto out;
416 if (!page->private)
417 goto out;
d397712b 418
ce9adaa5 419 len = page->private >> 2;
d397712b
CM		 420 WARN_ON(len == 0);
421
ce9adaa5 422 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 423
ce9adaa5 424 found_start = btrfs_header_bytenr(eb);
23a07867 425 if (found_start != start) {
193f284d
CM		 426 if (printk_ratelimit()) {
427 printk(KERN_INFO "btrfs bad tree block start "
428 "%llu %llu\n",
429 (unsigned long long)found_start,
430 (unsigned long long)eb->start);
431 }
f188591e 432 ret = -EIO;
ce9adaa5
CM		 433 goto err;
434 }
435 if (eb->first_page != page) {
d397712b
CM		 436 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
437 eb->first_page->index, page->index);
ce9adaa5 438 WARN_ON(1);
f188591e 439 ret = -EIO;
ce9adaa5
CM		 440 goto err;
441 }
2b82032c 442 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 443 if (printk_ratelimit()) {
444 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
445 (unsigned long long)eb->start);
446 }
1259ab75
CM		 447 ret = -EIO;
448 goto err;
449 }
ce9adaa5
CM		 450 found_level = btrfs_header_level(eb);
451
4008c04a
CM		 452 btrfs_set_buffer_lockdep_class(eb, found_level);
453
ce9adaa5 454 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 455 if (ret)
456 ret = -EIO;
ce9adaa5
CM		 457
458 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
459 end = eb->start + end - 1;
ce9adaa5
CM		 460err:
461 free_extent_buffer(eb);
462out:
f188591e 463 return ret;
ce9adaa5
CM		 464}
465
ce9adaa5 466static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 467{
468 struct end_io_wq *end_io_wq = bio->bi_private;
469 struct btrfs_fs_info *fs_info;
ce9adaa5 470
ce9adaa5 471 fs_info = end_io_wq->info;
ce9adaa5 472 end_io_wq->error = err;
8b712842
CM		 473 end_io_wq->work.func = end_workqueue_fn;
474 end_io_wq->work.flags = 0;
d20f7043
CM		 475
476 if (bio->bi_rw & (1 << BIO_RW)) {
cad321ad
CM		 477 if (end_io_wq->metadata)
478 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
479 &end_io_wq->work);
480 else
481 btrfs_queue_worker(&fs_info->endio_write_workers,
482 &end_io_wq->work);
d20f7043
CM		 483 } else {
484 if (end_io_wq->metadata)
485 btrfs_queue_worker(&fs_info->endio_meta_workers,
486 &end_io_wq->work);
487 else
488 btrfs_queue_worker(&fs_info->endio_workers,
489 &end_io_wq->work);
490 }
ce9adaa5
CM		 491}
492
22c59948
CM		 493int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
494 int metadata)
0b86a832 495{
ce9adaa5 496 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 497 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
498 if (!end_io_wq)
499 return -ENOMEM;
500
501 end_io_wq->private = bio->bi_private;
502 end_io_wq->end_io = bio->bi_end_io;
22c59948 503 end_io_wq->info = info;
ce9adaa5
CM		 504 end_io_wq->error = 0;
505 end_io_wq->bio = bio;
22c59948 506 end_io_wq->metadata = metadata;
ce9adaa5
CM		 507
508 bio->bi_private = end_io_wq;
509 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 510 return 0;
511}
512
b64a2851 513unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 514{
4854ddd0
CM		 515 unsigned long limit = min_t(unsigned long,
516 info->workers.max_workers,
517 info->fs_devices->open_devices);
518 return 256 * limit;
519}
0986fe9e 520
4854ddd0
CM		 521int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
522{
b64a2851
CM		 523 return atomic_read(&info->nr_async_bios) >
524 btrfs_async_submit_limit(info);
0986fe9e
CM		 525}
526
4a69a410
CM		 527static void run_one_async_start(struct btrfs_work *work)
528{
529 struct btrfs_fs_info *fs_info;
530 struct async_submit_bio *async;
531
532 async = container_of(work, struct async_submit_bio, work);
533 fs_info = BTRFS_I(async->inode)->root->fs_info;
534 async->submit_bio_start(async->inode, async->rw, async->bio,
535 async->mirror_num, async->bio_flags);
536}
537
538static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 539{
540 struct btrfs_fs_info *fs_info;
541 struct async_submit_bio *async;
4854ddd0 542 int limit;
8b712842
CM		 543
544 async = container_of(work, struct async_submit_bio, work);
545 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 546
b64a2851 547 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 548 limit = limit * 2 / 3;
549
8b712842 550 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 551
b64a2851
CM		 552 if (atomic_read(&fs_info->nr_async_submits) < limit &&
553 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 554 wake_up(&fs_info->async_submit_wait);
555
4a69a410 556 async->submit_bio_done(async->inode, async->rw, async->bio,
c8b97818 557 async->mirror_num, async->bio_flags);
4a69a410
CM		 558}
559
560static void run_one_async_free(struct btrfs_work *work)
561{
562 struct async_submit_bio *async;
563
564 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 565 kfree(async);
566}
567
44b8bd7e
CM		 568int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
569 int rw, struct bio *bio, int mirror_num,
c8b97818 570 unsigned long bio_flags,
4a69a410
CM		 571 extent_submit_bio_hook_t *submit_bio_start,
572 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 573{
574 struct async_submit_bio *async;
575
576 async = kmalloc(sizeof(*async), GFP_NOFS);
577 if (!async)
578 return -ENOMEM;
579
580 async->inode = inode;
581 async->rw = rw;
582 async->bio = bio;
583 async->mirror_num = mirror_num;
4a69a410
CM		 584 async->submit_bio_start = submit_bio_start;
585 async->submit_bio_done = submit_bio_done;
586
587 async->work.func = run_one_async_start;
588 async->work.ordered_func = run_one_async_done;
589 async->work.ordered_free = run_one_async_free;
590
8b712842 591 async->work.flags = 0;
c8b97818 592 async->bio_flags = bio_flags;
8c8bee1d 593
cb03c743 594 atomic_inc(&fs_info->nr_async_submits);
d313d7a3
CM		 595
596 if (rw & (1 << BIO_RW_SYNCIO))
597 btrfs_set_work_high_prio(&async->work);
598
8b712842 599 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 600
d397712b 601 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 602 atomic_read(&fs_info->nr_async_submits)) {
603 wait_event(fs_info->async_submit_wait,
604 (atomic_read(&fs_info->nr_async_submits) == 0));
605 }
606
44b8bd7e
CM		 607 return 0;
608}
609
ce3ed71a
CM		 610static int btree_csum_one_bio(struct bio *bio)
611{
612 struct bio_vec *bvec = bio->bi_io_vec;
613 int bio_index = 0;
614 struct btrfs_root *root;
615
616 WARN_ON(bio->bi_vcnt <= 0);
d397712b 617 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 618 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
619 csum_dirty_buffer(root, bvec->bv_page);
620 bio_index++;
621 bvec++;
622 }
623 return 0;
624}
625
4a69a410
CM		 626static int __btree_submit_bio_start(struct inode *inode, int rw,
627 struct bio *bio, int mirror_num,
628 unsigned long bio_flags)
22c59948 629{
8b712842
CM		 630 /*
631 * when we're called for a write, we're already in the async
5443be45 632 * submission context. Just jump into btrfs_map_bio
8b712842 633 */
4a69a410
CM		 634 btree_csum_one_bio(bio);
635 return 0;
636}
22c59948 637
4a69a410
CM		 638static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
639 int mirror_num, unsigned long bio_flags)
640{
8b712842 641 /*
4a69a410
CM		 642 * when we're called for a write, we're already in the async
643 * submission context. Just jump into btrfs_map_bio
8b712842 644 */
8b712842 645 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 646}
647
44b8bd7e 648static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 649 int mirror_num, unsigned long bio_flags)
44b8bd7e 650{
cad321ad
CM		 651 int ret;
652
653 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
654 bio, 1);
655 BUG_ON(ret);
656
44b8bd7e 657 if (!(rw & (1 << BIO_RW))) {
4a69a410
CM		 658 /*
659 * called for a read, do the setup so that checksum validation
660 * can happen in the async kernel threads
661 */
4a69a410 662 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 663 mirror_num, 0);
44b8bd7e 664 }
d313d7a3 665
cad321ad
CM		 666 /*
667 * kthread helpers are used to submit writes so that checksumming
668 * can happen in parallel across all CPUs
669 */
44b8bd7e 670 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 671 inode, rw, bio, mirror_num, 0,
4a69a410
CM		 672 __btree_submit_bio_start,
673 __btree_submit_bio_done);
44b8bd7e
CM		 674}
675
0da5468f
CM		 676static int btree_writepage(struct page *page, struct writeback_control *wbc)
677{
d1310b2e 678 struct extent_io_tree *tree;
b9473439
CM		 679 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
680 struct extent_buffer *eb;
681 int was_dirty;
682
d1310b2e 683 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 684 if (!(current->flags & PF_MEMALLOC)) {
685 return extent_write_full_page(tree, page,
686 btree_get_extent, wbc);
687 }
5443be45 688
b9473439
CM		 689 redirty_page_for_writepage(wbc, page);
690 eb = btrfs_find_tree_block(root, page_offset(page),
691 PAGE_CACHE_SIZE);
692 WARN_ON(!eb);
693
694 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
695 if (!was_dirty) {
696 spin_lock(&root->fs_info->delalloc_lock);
697 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
698 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 699 }
b9473439
CM		 700 free_extent_buffer(eb);
701
702 unlock_page(page);
703 return 0;
5f39d397 704}
0da5468f
CM		 705
706static int btree_writepages(struct address_space *mapping,
707 struct writeback_control *wbc)
708{
d1310b2e
CM		 709 struct extent_io_tree *tree;
710 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 711 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 712 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 713 u64 num_dirty;
24ab9cd8 714 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 715
716 if (wbc->for_kupdate)
717 return 0;
718
b9473439
CM		 719 /* this is a bit racy, but that's ok */
720 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 721 if (num_dirty < thresh)
793955bc 722 return 0;
793955bc 723 }
0da5468f
CM		 724 return extent_writepages(tree, mapping, btree_get_extent, wbc);
725}
726
b2950863 727static int btree_readpage(struct file *file, struct page *page)
5f39d397 728{
d1310b2e
CM		 729 struct extent_io_tree *tree;
730 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 731 return extent_read_full_page(tree, page, btree_get_extent);
732}
22b0ebda 733
70dec807 734static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 735{
d1310b2e
CM		 736 struct extent_io_tree *tree;
737 struct extent_map_tree *map;
5f39d397 738 int ret;
d98237b3 739
98509cfc 740 if (PageWriteback(page) || PageDirty(page))
d397712b 741 return 0;
98509cfc 742
d1310b2e
CM		 743 tree = &BTRFS_I(page->mapping->host)->io_tree;
744 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 745
7b13b7b1 746 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 747 if (!ret)
6af118ce 748 return 0;
6af118ce
CM		 749
750 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 751 if (ret == 1) {
752 ClearPagePrivate(page);
753 set_page_private(page, 0);
754 page_cache_release(page);
755 }
6af118ce 756
d98237b3
CM		 757 return ret;
758}
759
5f39d397 760static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 761{
d1310b2e
CM		 762 struct extent_io_tree *tree;
763 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 764 extent_invalidatepage(tree, page, offset);
765 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 766 if (PagePrivate(page)) {
d397712b
CM		 767 printk(KERN_WARNING "btrfs warning page private not zero "
768 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 769 ClearPagePrivate(page);
770 set_page_private(page, 0);
771 page_cache_release(page);
772 }
d98237b3
CM		 773}
774
d98237b3
CM		 775static struct address_space_operations btree_aops = {
776 .readpage = btree_readpage,
777 .writepage = btree_writepage,
0da5468f 778 .writepages = btree_writepages,
5f39d397
CM		 779 .releasepage = btree_releasepage,
780 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 781 .sync_page = block_sync_page,
782};
783
ca7a79ad
CM		 784int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
785 u64 parent_transid)
090d1875 786{
5f39d397
CM		 787 struct extent_buffer *buf = NULL;
788 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 789 int ret = 0;
090d1875 790
db94535d 791 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 792 if (!buf)
090d1875 793 return 0;
d1310b2e 794 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 795 buf, 0, 0, btree_get_extent, 0);
5f39d397 796 free_extent_buffer(buf);
de428b63 797 return ret;
090d1875
CM		 798}
799
0999df54
CM		 800struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
801 u64 bytenr, u32 blocksize)
802{
803 struct inode *btree_inode = root->fs_info->btree_inode;
804 struct extent_buffer *eb;
805 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
806 bytenr, blocksize, GFP_NOFS);
807 return eb;
808}
809
810struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
811 u64 bytenr, u32 blocksize)
812{
813 struct inode *btree_inode = root->fs_info->btree_inode;
814 struct extent_buffer *eb;
815
816 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
817 bytenr, blocksize, NULL, GFP_NOFS);
818 return eb;
819}
820
821
e02119d5
CM		 822int btrfs_write_tree_block(struct extent_buffer *buf)
823{
824 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
24ab9cd8 825 buf->start + buf->len - 1, WB_SYNC_ALL);
e02119d5
CM		 826}
827
828int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
829{
830 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
d397712b 831 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 832}
833
0999df54 834struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 835 u32 blocksize, u64 parent_transid)
0999df54
CM		 836{
837 struct extent_buffer *buf = NULL;
838 struct inode *btree_inode = root->fs_info->btree_inode;
839 struct extent_io_tree *io_tree;
840 int ret;
841
842 io_tree = &BTRFS_I(btree_inode)->io_tree;
843
844 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
845 if (!buf)
846 return NULL;
0999df54 847
ca7a79ad 848 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 849
d397712b 850 if (ret == 0)
b4ce94de 851 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 852 return buf;
ce9adaa5 853
eb60ceac
CM		 854}
855
e089f05c 856int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 857 struct extent_buffer *buf)
ed2ff2cb 858{
5f39d397 859 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 860 if (btrfs_header_generation(buf) ==
925baedd 861 root->fs_info->running_transaction->transid) {
b9447ef8 862 btrfs_assert_tree_locked(buf);
b4ce94de 863
b9473439
CM		 864 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
865 spin_lock(&root->fs_info->delalloc_lock);
866 if (root->fs_info->dirty_metadata_bytes >= buf->len)
867 root->fs_info->dirty_metadata_bytes -= buf->len;
868 else
869 WARN_ON(1);
870 spin_unlock(&root->fs_info->delalloc_lock);
871 }
b4ce94de 872
b9473439
CM		 873 /* ugh, clear_extent_buffer_dirty needs to lock the page */
874 btrfs_set_lock_blocking(buf);
d1310b2e 875 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 876 buf);
925baedd 877 }
5f39d397
CM		 878 return 0;
879}
880
db94535d 881static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 882 u32 stripesize, struct btrfs_root *root,
9f5fae2f 883 struct btrfs_fs_info *fs_info,
e20d96d6 884 u64 objectid)
d97e63b6 885{
cfaa7295 886 root->node = NULL;
a28ec197 887 root->commit_root = NULL;
db94535d
CM		 888 root->sectorsize = sectorsize;
889 root->nodesize = nodesize;
890 root->leafsize = leafsize;
87ee04eb 891 root->stripesize = stripesize;
123abc88 892 root->ref_cows = 0;
0b86a832
CM		 893 root->track_dirty = 0;
894
9f5fae2f 895 root->fs_info = fs_info;
0f7d52f4
CM		 896 root->objectid = objectid;
897 root->last_trans = 0;
1b05da2e
CM		 898 root->highest_inode = 0;
899 root->last_inode_alloc = 0;
58176a96 900 root->name = NULL;
4313b399 901 root->in_sysfs = 0;
5d4f98a2 902 root->inode_tree.rb_node = NULL;
0b86a832
CM		 903
904 INIT_LIST_HEAD(&root->dirty_list);
7b128766 905 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 906 INIT_LIST_HEAD(&root->root_list);
925baedd 907 spin_lock_init(&root->node_lock);
bcc63abb 908 spin_lock_init(&root->list_lock);
5d4f98a2 909 spin_lock_init(&root->inode_lock);
a2135011 910 mutex_init(&root->objectid_mutex);
e02119d5 911 mutex_init(&root->log_mutex);
7237f183
YZ		 912 init_waitqueue_head(&root->log_writer_wait);
913 init_waitqueue_head(&root->log_commit_wait[0]);
914 init_waitqueue_head(&root->log_commit_wait[1]);
915 atomic_set(&root->log_commit[0], 0);
916 atomic_set(&root->log_commit[1], 0);
917 atomic_set(&root->log_writers, 0);
918 root->log_batch = 0;
919 root->log_transid = 0;
d0c803c4
CM		 920 extent_io_tree_init(&root->dirty_log_pages,
921 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 922
3768f368
CM		 923 memset(&root->root_key, 0, sizeof(root->root_key));
924 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 925 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 926 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 927 root->defrag_trans_start = fs_info->generation;
58176a96 928 init_completion(&root->kobj_unregister);
6702ed49
CM		 929 root->defrag_running = 0;
930 root->defrag_level = 0;
4d775673 931 root->root_key.objectid = objectid;
3394e160
CM		 932 root->anon_super.s_root = NULL;
933 root->anon_super.s_dev = 0;
934 INIT_LIST_HEAD(&root->anon_super.s_list);
935 INIT_LIST_HEAD(&root->anon_super.s_instances);
936 init_rwsem(&root->anon_super.s_umount);
937
3768f368
CM		 938 return 0;
939}
940
db94535d 941static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 942 struct btrfs_fs_info *fs_info,
943 u64 objectid,
e20d96d6 944 struct btrfs_root *root)
3768f368
CM		 945{
946 int ret;
db94535d 947 u32 blocksize;
84234f3a 948 u64 generation;
3768f368 949
db94535d 950 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 951 tree_root->sectorsize, tree_root->stripesize,
952 root, fs_info, objectid);
3768f368
CM		 953 ret = btrfs_find_last_root(tree_root, objectid,
954 &root->root_item, &root->root_key);
955 BUG_ON(ret);
956
84234f3a 957 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 958 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
959 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 960 blocksize, generation);
5d4f98a2 961 root->commit_root = btrfs_root_node(root);
3768f368 962 BUG_ON(!root->node);
d97e63b6
CM		 963 return 0;
964}
965
e02119d5
CM		 966int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
967 struct btrfs_fs_info *fs_info)
968{
969 struct extent_buffer *eb;
d0c803c4
CM		 970 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
971 u64 start = 0;
972 u64 end = 0;
e02119d5
CM		 973 int ret;
974
d0c803c4 975 if (!log_root_tree)
e02119d5
CM		 976 return 0;
977
d397712b 978 while (1) {
d0c803c4
CM		 979 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
980 0, &start, &end, EXTENT_DIRTY);
981 if (ret)
982 break;
983
984 clear_extent_dirty(&log_root_tree->dirty_log_pages,
985 start, end, GFP_NOFS);
986 }
e02119d5
CM		 987 eb = fs_info->log_root_tree->node;
988
989 WARN_ON(btrfs_header_level(eb) != 0);
990 WARN_ON(btrfs_header_nritems(eb) != 0);
991
d00aff00
CM		 992 ret = btrfs_free_reserved_extent(fs_info->tree_root,
993 eb->start, eb->len);
e02119d5
CM		 994 BUG_ON(ret);
995
996 free_extent_buffer(eb);
997 kfree(fs_info->log_root_tree);
998 fs_info->log_root_tree = NULL;
999 return 0;
1000}
1001
7237f183
YZ		 1002static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1003 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1004{
1005 struct btrfs_root *root;
1006 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1007 struct extent_buffer *leaf;
e02119d5
CM		 1008
1009 root = kzalloc(sizeof(*root), GFP_NOFS);
1010 if (!root)
7237f183 1011 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1012
1013 __setup_root(tree_root->nodesize, tree_root->leafsize,
1014 tree_root->sectorsize, tree_root->stripesize,
1015 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1016
1017 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1018 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1019 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1020 /*
1021 * log trees do not get reference counted because they go away
1022 * before a real commit is actually done. They do store pointers
1023 * to file data extents, and those reference counts still get
1024 * updated (along with back refs to the log tree).
1025 */
e02119d5
CM		 1026 root->ref_cows = 0;
1027
5d4f98a2
YZ		 1028 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1029 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1030 if (IS_ERR(leaf)) {
1031 kfree(root);
1032 return ERR_CAST(leaf);
1033 }
e02119d5 1034
5d4f98a2
YZ		 1035 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1036 btrfs_set_header_bytenr(leaf, leaf->start);
1037 btrfs_set_header_generation(leaf, trans->transid);
1038 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1039 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1040 root->node = leaf;
e02119d5
CM		 1041
1042 write_extent_buffer(root->node, root->fs_info->fsid,
1043 (unsigned long)btrfs_header_fsid(root->node),
1044 BTRFS_FSID_SIZE);
1045 btrfs_mark_buffer_dirty(root->node);
1046 btrfs_tree_unlock(root->node);
7237f183
YZ		 1047 return root;
1048}
1049
1050int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1051 struct btrfs_fs_info *fs_info)
1052{
1053 struct btrfs_root *log_root;
1054
1055 log_root = alloc_log_tree(trans, fs_info);
1056 if (IS_ERR(log_root))
1057 return PTR_ERR(log_root);
1058 WARN_ON(fs_info->log_root_tree);
1059 fs_info->log_root_tree = log_root;
1060 return 0;
1061}
1062
1063int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1064 struct btrfs_root *root)
1065{
1066 struct btrfs_root *log_root;
1067 struct btrfs_inode_item *inode_item;
1068
1069 log_root = alloc_log_tree(trans, root->fs_info);
1070 if (IS_ERR(log_root))
1071 return PTR_ERR(log_root);
1072
1073 log_root->last_trans = trans->transid;
1074 log_root->root_key.offset = root->root_key.objectid;
1075
1076 inode_item = &log_root->root_item.inode;
1077 inode_item->generation = cpu_to_le64(1);
1078 inode_item->size = cpu_to_le64(3);
1079 inode_item->nlink = cpu_to_le32(1);
1080 inode_item->nbytes = cpu_to_le64(root->leafsize);
1081 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1082
5d4f98a2 1083 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1084
1085 WARN_ON(root->log_root);
1086 root->log_root = log_root;
1087 root->log_transid = 0;
e02119d5
CM		 1088 return 0;
1089}
1090
1091struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1092 struct btrfs_key *location)
1093{
1094 struct btrfs_root *root;
1095 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1096 struct btrfs_path *path;
5f39d397 1097 struct extent_buffer *l;
1b05da2e 1098 u64 highest_inode;
84234f3a 1099 u64 generation;
db94535d 1100 u32 blocksize;
0f7d52f4
CM		 1101 int ret = 0;
1102
5eda7b5e 1103 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1104 if (!root)
0f7d52f4 1105 return ERR_PTR(-ENOMEM);
0f7d52f4 1106 if (location->offset == (u64)-1) {
db94535d 1107 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1108 location->objectid, root);
1109 if (ret) {
0f7d52f4
CM		 1110 kfree(root);
1111 return ERR_PTR(ret);
1112 }
1113 goto insert;
1114 }
1115
db94535d 1116 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1117 tree_root->sectorsize, tree_root->stripesize,
1118 root, fs_info, location->objectid);
0f7d52f4
CM		 1119
1120 path = btrfs_alloc_path();
1121 BUG_ON(!path);
1122 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1123 if (ret != 0) {
0f7d52f4
CM		 1124 if (ret > 0)
1125 ret = -ENOENT;
1126 goto out;
1127 }
5f39d397
CM		 1128 l = path->nodes[0];
1129 read_extent_buffer(l, &root->root_item,
1130 btrfs_item_ptr_offset(l, path->slots[0]),
0f7d52f4 1131 sizeof(root->root_item));
44b36eb2 1132 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4
CM		 1133 ret = 0;
1134out:
1135 btrfs_release_path(root, path);
1136 btrfs_free_path(path);
1137 if (ret) {
1138 kfree(root);
1139 return ERR_PTR(ret);
1140 }
84234f3a 1141 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1142 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1143 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1144 blocksize, generation);
5d4f98a2 1145 root->commit_root = btrfs_root_node(root);
0f7d52f4
CM		 1146 BUG_ON(!root->node);
1147insert:
e02119d5
CM		 1148 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1149 root->ref_cows = 1;
1150 ret = btrfs_find_highest_inode(root, &highest_inode);
1151 if (ret == 0) {
1152 root->highest_inode = highest_inode;
1153 root->last_inode_alloc = highest_inode;
1154 }
5eda7b5e
CM		 1155 }
1156 return root;
1157}
1158
dc17ff8f
CM		 1159struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1160 u64 root_objectid)
1161{
1162 struct btrfs_root *root;
1163
1164 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1165 return fs_info->tree_root;
1166 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1167 return fs_info->extent_root;
1168
1169 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170 (unsigned long)root_objectid);
1171 return root;
1172}
1173
edbd8d4e
CM		 1174struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1175 struct btrfs_key *location)
5eda7b5e
CM		 1176{
1177 struct btrfs_root *root;
1178 int ret;
1179
edbd8d4e
CM		 1180 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1181 return fs_info->tree_root;
1182 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1183 return fs_info->extent_root;
8f18cf13
CM		 1184 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1185 return fs_info->chunk_root;
1186 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1187 return fs_info->dev_root;
0403e47e
YZ		 1188 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1189 return fs_info->csum_root;
edbd8d4e 1190
5eda7b5e
CM		 1191 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1192 (unsigned long)location->objectid);
1193 if (root)
1194 return root;
1195
e02119d5 1196 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1197 if (IS_ERR(root))
1198 return root;
3394e160
CM		 1199
1200 set_anon_super(&root->anon_super, NULL);
1201
2619ba1f
CM		 1202 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1203 (unsigned long)root->root_key.objectid,
0f7d52f4
CM		 1204 root);
1205 if (ret) {
5f39d397 1206 free_extent_buffer(root->node);
0f7d52f4
CM		 1207 kfree(root);
1208 return ERR_PTR(ret);
1209 }
c146afad
YZ		 1210 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1211 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 1212 root->root_key.objectid);
c146afad
YZ		 1213 BUG_ON(ret);
1214 btrfs_orphan_cleanup(root);
1215 }
edbd8d4e
CM		 1216 return root;
1217}
1218
1219struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1220 struct btrfs_key *location,
1221 const char *name, int namelen)
1222{
1223 struct btrfs_root *root;
1224 int ret;
1225
1226 root = btrfs_read_fs_root_no_name(fs_info, location);
1227 if (!root)
1228 return NULL;
58176a96 1229
4313b399
CM		 1230 if (root->in_sysfs)
1231 return root;
1232
58176a96
JB		 1233 ret = btrfs_set_root_name(root, name, namelen);
1234 if (ret) {
5f39d397 1235 free_extent_buffer(root->node);
58176a96
JB		 1236 kfree(root);
1237 return ERR_PTR(ret);
1238 }
ea9e8b11 1239#if 0
58176a96
JB		 1240 ret = btrfs_sysfs_add_root(root);
1241 if (ret) {
5f39d397 1242 free_extent_buffer(root->node);
58176a96
JB		 1243 kfree(root->name);
1244 kfree(root);
1245 return ERR_PTR(ret);
1246 }
ea9e8b11 1247#endif
4313b399 1248 root->in_sysfs = 1;
0f7d52f4
CM		 1249 return root;
1250}
04160088
CM		 1251
1252static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1253{
1254 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1255 int ret = 0;
04160088
CM		 1256 struct btrfs_device *device;
1257 struct backing_dev_info *bdi;
b7967db7 1258
c6e30871 1259 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1260 if (!device->bdev)
1261 continue;
04160088
CM		 1262 bdi = blk_get_backing_dev_info(device->bdev);
1263 if (bdi && bdi_congested(bdi, bdi_bits)) {
1264 ret = 1;
1265 break;
1266 }
1267 }
1268 return ret;
1269}
1270
38b66988
CM		 1271/*
1272 * this unplugs every device on the box, and it is only used when page
1273 * is null
1274 */
1275static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1276{
38b66988
CM		 1277 struct btrfs_device *device;
1278 struct btrfs_fs_info *info;
1279
1280 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1281 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1282 if (!device->bdev)
1283 continue;
1284
38b66988 1285 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1286 if (bdi->unplug_io_fn)
38b66988 1287 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1288 }
1289}
1290
b2950863 1291static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1292{
38b66988 1293 struct inode *inode;
f2d8d74d
CM		 1294 struct extent_map_tree *em_tree;
1295 struct extent_map *em;
bcbfce8a 1296 struct address_space *mapping;
38b66988
CM		 1297 u64 offset;
1298
bcbfce8a 1299 /* the generic O_DIRECT read code does this */
9f0ba5bd 1300 if (1 || !page) {
38b66988
CM		 1301 __unplug_io_fn(bdi, page);
1302 return;
1303 }
1304
bcbfce8a
CM		 1305 /*
1306 * page->mapping may change at any time. Get a consistent copy
1307 * and use that for everything below
1308 */
1309 smp_mb();
1310 mapping = page->mapping;
1311 if (!mapping)
1312 return;
1313
1314 inode = mapping->host;
240d5d48
CM		 1315
1316 /*
1317 * don't do the expensive searching for a small number of
1318 * devices
1319 */
1320 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1321 __unplug_io_fn(bdi, page);
1322 return;
1323 }
1324
38b66988 1325 offset = page_offset(page);
04160088 1326
f2d8d74d
CM		 1327 em_tree = &BTRFS_I(inode)->extent_tree;
1328 spin_lock(&em_tree->lock);
1329 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1330 spin_unlock(&em_tree->lock);
89642229
CM		 1331 if (!em) {
1332 __unplug_io_fn(bdi, page);
f2d8d74d 1333 return;
89642229 1334 }
f2d8d74d 1335
89642229
CM		 1336 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1337 free_extent_map(em);
1338 __unplug_io_fn(bdi, page);
1339 return;
1340 }
f2d8d74d
CM		 1341 offset = offset - em->start;
1342 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1343 em->block_start + offset, page);
1344 free_extent_map(em);
04160088
CM		 1345}
1346
ad081f14
JA		 1347/*
1348 * If this fails, caller must call bdi_destroy() to get rid of the
1349 * bdi again.
1350 */
04160088
CM		 1351static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1352{
ad081f14
JA		 1353 int err;
1354
d993831f 1355 bdi->name = "btrfs";
ad081f14
JA		 1356 bdi->capabilities = BDI_CAP_MAP_COPY;
1357 err = bdi_init(bdi);
1358 if (err)
1359 return err;
1360
1361 err = bdi_register(bdi, NULL, "btrfs-%d",
1362 atomic_inc_return(&btrfs_bdi_num));
1363 if (err)
1364 return err;
1365
4575c9cc 1366 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1367 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1368 bdi->unplug_io_data = info;
1369 bdi->congested_fn = btrfs_congested_fn;
1370 bdi->congested_data = info;
1371 return 0;
1372}
1373
ce9adaa5
CM		 1374static int bio_ready_for_csum(struct bio *bio)
1375{
1376 u64 length = 0;
1377 u64 buf_len = 0;
1378 u64 start = 0;
1379 struct page *page;
1380 struct extent_io_tree *io_tree = NULL;
1381 struct btrfs_fs_info *info = NULL;
1382 struct bio_vec *bvec;
1383 int i;
1384 int ret;
1385
1386 bio_for_each_segment(bvec, bio, i) {
1387 page = bvec->bv_page;
1388 if (page->private == EXTENT_PAGE_PRIVATE) {
1389 length += bvec->bv_len;
1390 continue;
1391 }
1392 if (!page->private) {
1393 length += bvec->bv_len;
1394 continue;
1395 }
1396 length = bvec->bv_len;
1397 buf_len = page->private >> 2;
1398 start = page_offset(page) + bvec->bv_offset;
1399 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1400 info = BTRFS_I(page->mapping->host)->root->fs_info;
1401 }
1402 /* are we fully contained in this bio? */
1403 if (buf_len <= length)
1404 return 1;
1405
1406 ret = extent_range_uptodate(io_tree, start + length,
1407 start + buf_len - 1);
ce9adaa5
CM		 1408 return ret;
1409}
1410
8b712842
CM		 1411/*
1412 * called by the kthread helper functions to finally call the bio end_io
1413 * functions. This is where read checksum verification actually happens
1414 */
1415static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1416{
ce9adaa5 1417 struct bio *bio;
8b712842
CM		 1418 struct end_io_wq *end_io_wq;
1419 struct btrfs_fs_info *fs_info;
ce9adaa5 1420 int error;
ce9adaa5 1421
8b712842
CM		 1422 end_io_wq = container_of(work, struct end_io_wq, work);
1423 bio = end_io_wq->bio;
1424 fs_info = end_io_wq->info;
ce9adaa5 1425
cad321ad 1426 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1427 * be checksummed at once. This makes sure the entire block is in
1428 * ram and up to date before trying to verify things. For
1429 * blocksize <= pagesize, it is basically a noop
1430 */
cad321ad
CM		 1431 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1432 !bio_ready_for_csum(bio)) {
d20f7043 1433 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1434 &end_io_wq->work);
1435 return;
1436 }
1437 error = end_io_wq->error;
1438 bio->bi_private = end_io_wq->private;
1439 bio->bi_end_io = end_io_wq->end_io;
1440 kfree(end_io_wq);
8b712842 1441 bio_endio(bio, error);
44b8bd7e
CM		 1442}
1443
a74a4b97
CM		 1444static int cleaner_kthread(void *arg)
1445{
1446 struct btrfs_root *root = arg;
1447
1448 do {
1449 smp_mb();
1450 if (root->fs_info->closing)
1451 break;
1452
1453 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1454 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 1455 btrfs_clean_old_snapshots(root);
a74a4b97
CM		 1456 mutex_unlock(&root->fs_info->cleaner_mutex);
1457
1458 if (freezing(current)) {
1459 refrigerator();
1460 } else {
1461 smp_mb();
1462 if (root->fs_info->closing)
1463 break;
1464 set_current_state(TASK_INTERRUPTIBLE);
1465 schedule();
1466 __set_current_state(TASK_RUNNING);
1467 }
1468 } while (!kthread_should_stop());
1469 return 0;
1470}
1471
1472static int transaction_kthread(void *arg)
1473{
1474 struct btrfs_root *root = arg;
1475 struct btrfs_trans_handle *trans;
1476 struct btrfs_transaction *cur;
1477 unsigned long now;
1478 unsigned long delay;
1479 int ret;
1480
1481 do {
1482 smp_mb();
1483 if (root->fs_info->closing)
1484 break;
1485
1486 delay = HZ * 30;
1487 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1488 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1489
1490 mutex_lock(&root->fs_info->trans_mutex);
1491 cur = root->fs_info->running_transaction;
1492 if (!cur) {
1493 mutex_unlock(&root->fs_info->trans_mutex);
1494 goto sleep;
1495 }
31153d81 1496
a74a4b97
CM		 1497 now = get_seconds();
1498 if (now < cur->start_time || now - cur->start_time < 30) {
1499 mutex_unlock(&root->fs_info->trans_mutex);
1500 delay = HZ * 5;
1501 goto sleep;
1502 }
1503 mutex_unlock(&root->fs_info->trans_mutex);
a74a4b97
CM		 1504 trans = btrfs_start_transaction(root, 1);
1505 ret = btrfs_commit_transaction(trans, root);
56bec294 1506
a74a4b97
CM		 1507sleep:
1508 wake_up_process(root->fs_info->cleaner_kthread);
1509 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1510
1511 if (freezing(current)) {
1512 refrigerator();
1513 } else {
1514 if (root->fs_info->closing)
1515 break;
1516 set_current_state(TASK_INTERRUPTIBLE);
1517 schedule_timeout(delay);
1518 __set_current_state(TASK_RUNNING);
1519 }
1520 } while (!kthread_should_stop());
1521 return 0;
1522}
1523
8a4b83cc 1524struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1525 struct btrfs_fs_devices *fs_devices,
1526 char *options)
2e635a27 1527{
db94535d
CM		 1528 u32 sectorsize;
1529 u32 nodesize;
1530 u32 leafsize;
1531 u32 blocksize;
87ee04eb 1532 u32 stripesize;
84234f3a 1533 u64 generation;
f2b636e8 1534 u64 features;
3de4586c 1535 struct btrfs_key location;
a061fc8d 1536 struct buffer_head *bh;
e02119d5 1537 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1538 GFP_NOFS);
d20f7043
CM		 1539 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1540 GFP_NOFS);
e02119d5 1541 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1542 GFP_NOFS);
8790d502 1543 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1544 GFP_NOFS);
e02119d5 1545 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1546 GFP_NOFS);
e02119d5 1547 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1548 GFP_NOFS);
e02119d5
CM		 1549 struct btrfs_root *log_tree_root;
1550
eb60ceac 1551 int ret;
e58ca020 1552 int err = -EINVAL;
4543df7e 1553
2c90e5d6 1554 struct btrfs_super_block *disk_super;
8790d502 1555
0463bb4e 1556 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1557 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1558 err = -ENOMEM;
1559 goto fail;
1560 }
0f7d52f4 1561 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
8fd17795 1562 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1563 INIT_LIST_HEAD(&fs_info->dead_roots);
19c00ddc 1564 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1565 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1566 INIT_LIST_HEAD(&fs_info->ordered_operations);
1832a6d5 1567 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1568 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1569 spin_lock_init(&fs_info->ref_cache_lock);
19c00ddc 1570
58176a96 1571 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1572 fs_info->tree_root = tree_root;
1573 fs_info->extent_root = extent_root;
d20f7043 1574 fs_info->csum_root = csum_root;
0b86a832
CM		 1575 fs_info->chunk_root = chunk_root;
1576 fs_info->dev_root = dev_root;
8a4b83cc 1577 fs_info->fs_devices = fs_devices;
0b86a832 1578 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1579 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1580 btrfs_mapping_init(&fs_info->mapping_tree);
cb03c743 1581 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1582 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1583 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1584 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1585 fs_info->sb = sb;
c59f8951 1586 fs_info->max_extent = (u64)-1;
6f568d35 1587 fs_info->max_inline = 8192 * 1024;
ad081f14
JA		 1588 if (setup_bdi(fs_info, &fs_info->bdi))
1589 goto fail_bdi;
d98237b3
CM		 1590 fs_info->btree_inode = new_inode(sb);
1591 fs_info->btree_inode->i_ino = 1;
2c90e5d6 1592 fs_info->btree_inode->i_nlink = 1;
97e728d4 1593 fs_info->metadata_ratio = 8;
c8b97818 1594
b34b086c
CM		 1595 fs_info->thread_pool_size = min_t(unsigned long,
1596 num_online_cpus() + 2, 8);
0afbaf8c 1597
3eaa2885
CM		 1598 INIT_LIST_HEAD(&fs_info->ordered_extents);
1599 spin_lock_init(&fs_info->ordered_extent_lock);
1600
a061fc8d
CM		 1601 sb->s_blocksize = 4096;
1602 sb->s_blocksize_bits = blksize_bits(4096);
1603
0afbaf8c
CM		 1604 /*
1605 * we set the i_size on the btree inode to the max possible int.
1606 * the real end of the address space is determined by all of
1607 * the devices in the system
1608 */
1609 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1610 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1611 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1612
5d4f98a2 1613 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1614 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1615 fs_info->btree_inode->i_mapping,
1616 GFP_NOFS);
d1310b2e
CM		 1617 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1618 GFP_NOFS);
1619
1620 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1621
0f9dd46c
JB		 1622 spin_lock_init(&fs_info->block_group_cache_lock);
1623 fs_info->block_group_cache_tree.rb_node = NULL;
1624
d1310b2e 1625 extent_io_tree_init(&fs_info->pinned_extents,
1a5bc167 1626 fs_info->btree_inode->i_mapping, GFP_NOFS);
e66f709b 1627 fs_info->do_barriers = 1;
e18e4809 1628
0f7d52f4
CM		 1629 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1630 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1631 sizeof(struct btrfs_key));
22b0ebda 1632 insert_inode_hash(fs_info->btree_inode);
39279cc3 1633
79154b1b 1634 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1635 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1636 mutex_init(&fs_info->tree_log_mutex);
a2135011 1637 mutex_init(&fs_info->drop_mutex);
925baedd 1638 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1639 mutex_init(&fs_info->transaction_kthread_mutex);
1640 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1641 mutex_init(&fs_info->volume_mutex);
1a40e23b 1642 mutex_init(&fs_info->tree_reloc_mutex);
276e680d 1643 init_rwsem(&fs_info->extent_commit_sem);
fa9c0d79
CM		 1644
1645 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1646 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1647
e6dcd2dc 1648 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1649 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1650 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1651
0b86a832 1652 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1653 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1654
d98237b3 1655
a512bbf8 1656 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1657 if (!bh)
39279cc3 1658 goto fail_iput;
39279cc3 1659
a061fc8d 1660 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1661 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1662 sizeof(fs_info->super_for_commit));
a061fc8d 1663 brelse(bh);
5f39d397 1664
a061fc8d 1665 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1666
5f39d397 1667 disk_super = &fs_info->super_copy;
0f7d52f4 1668 if (!btrfs_super_root(disk_super))
c6e2bac1 1669 goto fail_iput;
0f7d52f4 1670
2b82032c
YZ		 1671 ret = btrfs_parse_options(tree_root, options);
1672 if (ret) {
1673 err = ret;
c6e2bac1 1674 goto fail_iput;
2b82032c 1675 }
dfe25020 1676
f2b636e8
JB		 1677 features = btrfs_super_incompat_flags(disk_super) &
1678 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1679 if (features) {
1680 printk(KERN_ERR "BTRFS: couldn't mount because of "
1681 "unsupported optional features (%Lx).\n",
21380931 1682 (unsigned long long)features);
f2b636e8 1683 err = -EINVAL;
c6e2bac1 1684 goto fail_iput;
f2b636e8
JB		 1685 }
1686
5d4f98a2
YZ		 1687 features = btrfs_super_incompat_flags(disk_super);
1688 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1689 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1690 btrfs_set_super_incompat_flags(disk_super, features);
1691 }
1692
f2b636e8
JB		 1693 features = btrfs_super_compat_ro_flags(disk_super) &
1694 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1695 if (!(sb->s_flags & MS_RDONLY) && features) {
1696 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1697 "unsupported option features (%Lx).\n",
21380931 1698 (unsigned long long)features);
f2b636e8 1699 err = -EINVAL;
c6e2bac1 1700 goto fail_iput;
f2b636e8
JB		 1701 }
1702
4543df7e
CM		 1703 /*
1704 * we need to start all the end_io workers up front because the
1705 * queue work function gets called at interrupt time, and so it
1706 * cannot dynamically grow.
1707 */
5443be45
CM		 1708 btrfs_init_workers(&fs_info->workers, "worker",
1709 fs_info->thread_pool_size);
c8b97818 1710
771ed689
CM		 1711 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1712 fs_info->thread_pool_size);
1713
5443be45 1714 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209
CM		 1715 min_t(u64, fs_devices->num_devices,
1716 fs_info->thread_pool_size));
61b49440
CM		 1717
1718 /* a higher idle thresh on the submit workers makes it much more
1719 * likely that bios will be send down in a sane order to the
1720 * devices
1721 */
1722 fs_info->submit_workers.idle_thresh = 64;
53863232 1723
771ed689 1724 fs_info->workers.idle_thresh = 16;
4a69a410 1725 fs_info->workers.ordered = 1;
61b49440 1726
771ed689
CM		 1727 fs_info->delalloc_workers.idle_thresh = 2;
1728 fs_info->delalloc_workers.ordered = 1;
1729
5443be45
CM		 1730 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1731 btrfs_init_workers(&fs_info->endio_workers, "endio",
1732 fs_info->thread_pool_size);
d20f7043
CM		 1733 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1734 fs_info->thread_pool_size);
cad321ad
CM		 1735 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1736 "endio-meta-write", fs_info->thread_pool_size);
5443be45 1737 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
e6dcd2dc 1738 fs_info->thread_pool_size);
61b49440
CM		 1739
1740 /*
1741 * endios are largely parallel and should have a very
1742 * low idle thresh
1743 */
1744 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1745 fs_info->endio_meta_workers.idle_thresh = 4;
1746
3a5f1d45 1747 fs_info->endio_write_workers.idle_thresh = 64;
cad321ad 1748 fs_info->endio_meta_write_workers.idle_thresh = 64;
61b49440 1749
4543df7e 1750 btrfs_start_workers(&fs_info->workers, 1);
1cc127b5 1751 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1752 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1753 btrfs_start_workers(&fs_info->fixup_workers, 1);
4543df7e 1754 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
d20f7043
CM		 1755 btrfs_start_workers(&fs_info->endio_meta_workers,
1756 fs_info->thread_pool_size);
cad321ad
CM		 1757 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1758 fs_info->thread_pool_size);
e6dcd2dc
CM		 1759 btrfs_start_workers(&fs_info->endio_write_workers,
1760 fs_info->thread_pool_size);
4543df7e 1761
4575c9cc 1762 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1763 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1764 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1765
db94535d
CM		 1766 nodesize = btrfs_super_nodesize(disk_super);
1767 leafsize = btrfs_super_leafsize(disk_super);
1768 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1769 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1770 tree_root->nodesize = nodesize;
1771 tree_root->leafsize = leafsize;
1772 tree_root->sectorsize = sectorsize;
87ee04eb 1773 tree_root->stripesize = stripesize;
a061fc8d
CM		 1774
1775 sb->s_blocksize = sectorsize;
1776 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1777
39279cc3
CM		 1778 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1779 sizeof(disk_super->magic))) {
d397712b 1780 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1781 goto fail_sb_buffer;
1782 }
19c00ddc 1783
925baedd 1784 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1785 ret = btrfs_read_sys_array(tree_root);
925baedd 1786 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1787 if (ret) {
d397712b
CM		 1788 printk(KERN_WARNING "btrfs: failed to read the system "
1789 "array on %s\n", sb->s_id);
5d4f98a2 1790 goto fail_sb_buffer;
84eed90f 1791 }
0b86a832
CM		 1792
1793 blocksize = btrfs_level_size(tree_root,
1794 btrfs_super_chunk_root_level(disk_super));
84234f3a 1795 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1796
1797 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1798 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1799
1800 chunk_root->node = read_tree_block(chunk_root,
1801 btrfs_super_chunk_root(disk_super),
84234f3a 1802 blocksize, generation);
0b86a832 1803 BUG_ON(!chunk_root->node);
83121942
DW		 1804 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1805 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1806 sb->s_id);
1807 goto fail_chunk_root;
1808 }
5d4f98a2
YZ		 1809 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1810 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1811
e17cade2 1812 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1813 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1814 BTRFS_UUID_SIZE);
e17cade2 1815
925baedd 1816 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1817 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1818 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1819 if (ret) {
d397712b
CM		 1820 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1821 sb->s_id);
2b82032c
YZ		 1822 goto fail_chunk_root;
1823 }
0b86a832 1824
dfe25020
CM		 1825 btrfs_close_extra_devices(fs_devices);
1826
db94535d
CM		 1827 blocksize = btrfs_level_size(tree_root,
1828 btrfs_super_root_level(disk_super));
84234f3a 1829 generation = btrfs_super_generation(disk_super);
0b86a832 1830
e20d96d6 1831 tree_root->node = read_tree_block(tree_root,
db94535d 1832 btrfs_super_root(disk_super),
84234f3a 1833 blocksize, generation);
39279cc3 1834 if (!tree_root->node)
2b82032c 1835 goto fail_chunk_root;
83121942
DW		 1836 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1837 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1838 sb->s_id);
1839 goto fail_tree_root;
1840 }
5d4f98a2
YZ		 1841 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1842 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1843
1844 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1845 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1846 if (ret)
39279cc3 1847 goto fail_tree_root;
0b86a832
CM		 1848 extent_root->track_dirty = 1;
1849
1850 ret = find_and_setup_root(tree_root, fs_info,
1851 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1852 if (ret)
1853 goto fail_extent_root;
5d4f98a2 1854 dev_root->track_dirty = 1;
3768f368 1855
d20f7043
CM		 1856 ret = find_and_setup_root(tree_root, fs_info,
1857 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1858 if (ret)
5d4f98a2 1859 goto fail_dev_root;
d20f7043
CM		 1860
1861 csum_root->track_dirty = 1;
1862
9078a3e1
CM		 1863 btrfs_read_block_groups(extent_root);
1864
2d69a0f8 1865 fs_info->generation = generation;
c146afad 1866 fs_info->last_trans_committed = generation;
d18a2c44
CM		 1867 fs_info->data_alloc_profile = (u64)-1;
1868 fs_info->metadata_alloc_profile = (u64)-1;
1869 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
a74a4b97
CM		 1870 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1871 "btrfs-cleaner");
57506d50 1872 if (IS_ERR(fs_info->cleaner_kthread))
d20f7043 1873 goto fail_csum_root;
a74a4b97
CM		 1874
1875 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1876 tree_root,
1877 "btrfs-transaction");
57506d50 1878 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1879 goto fail_cleaner;
a74a4b97 1880
c289811c
CM		 1881 if (!btrfs_test_opt(tree_root, SSD) &&
1882 !btrfs_test_opt(tree_root, NOSSD) &&
1883 !fs_info->fs_devices->rotating) {
1884 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1885 "mode\n");
1886 btrfs_set_opt(fs_info->mount_opt, SSD);
1887 }
1888
e02119d5 1889 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1890 u64 bytenr = btrfs_super_log_root(disk_super);
1891
7c2ca468 1892 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1893 printk(KERN_WARNING "Btrfs log replay required "
1894 "on RO media\n");
7c2ca468
CM		 1895 err = -EIO;
1896 goto fail_trans_kthread;
1897 }
e02119d5
CM		 1898 blocksize =
1899 btrfs_level_size(tree_root,
1900 btrfs_super_log_root_level(disk_super));
d18a2c44 1901
e02119d5
CM		 1902 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1903 GFP_NOFS);
1904
1905 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1906 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1907
1908 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1909 blocksize,
1910 generation + 1);
e02119d5
CM		 1911 ret = btrfs_recover_log_trees(log_tree_root);
1912 BUG_ON(ret);
e556ce2c
YZ		 1913
1914 if (sb->s_flags & MS_RDONLY) {
1915 ret = btrfs_commit_super(tree_root);
1916 BUG_ON(ret);
1917 }
e02119d5 1918 }
1a40e23b 1919
7c2ca468 1920 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1921 ret = btrfs_recover_relocation(tree_root);
7c2ca468
CM		 1922 BUG_ON(ret);
1923 }
1a40e23b 1924
3de4586c
CM		 1925 location.objectid = BTRFS_FS_TREE_OBJECTID;
1926 location.type = BTRFS_ROOT_ITEM_KEY;
1927 location.offset = (u64)-1;
1928
3de4586c
CM		 1929 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1930 if (!fs_info->fs_root)
7c2ca468 1931 goto fail_trans_kthread;
c289811c 1932
0f7d52f4 1933 return tree_root;
39279cc3 1934
7c2ca468
CM		 1935fail_trans_kthread:
1936 kthread_stop(fs_info->transaction_kthread);
3f157a2f 1937fail_cleaner:
a74a4b97 1938 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 1939
1940 /*
1941 * make sure we're done with the btree inode before we stop our
1942 * kthreads
1943 */
1944 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1945 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1946
d20f7043
CM		 1947fail_csum_root:
1948 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 1949 free_extent_buffer(csum_root->commit_root);
1950fail_dev_root:
1951 free_extent_buffer(dev_root->node);
1952 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 1953fail_extent_root:
1954 free_extent_buffer(extent_root->node);
5d4f98a2 1955 free_extent_buffer(extent_root->commit_root);
39279cc3 1956fail_tree_root:
5f39d397 1957 free_extent_buffer(tree_root->node);
5d4f98a2 1958 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 1959fail_chunk_root:
1960 free_extent_buffer(chunk_root->node);
5d4f98a2 1961 free_extent_buffer(chunk_root->commit_root);
39279cc3 1962fail_sb_buffer:
247e743c 1963 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 1964 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 1965 btrfs_stop_workers(&fs_info->workers);
1966 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 1967 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 1968 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 1969 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 1970 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 1971fail_iput:
7c2ca468 1972 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 1973 iput(fs_info->btree_inode);
7e662854 1974
dfe25020 1975 btrfs_close_devices(fs_info->fs_devices);
84eed90f 1976 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 1977fail_bdi:
7e662854 1978 bdi_destroy(&fs_info->bdi);
7e662854 1979fail:
39279cc3
CM		 1980 kfree(extent_root);
1981 kfree(tree_root);
1982 kfree(fs_info);
83afeac4
JM		 1983 kfree(chunk_root);
1984 kfree(dev_root);
d20f7043 1985 kfree(csum_root);
39279cc3 1986 return ERR_PTR(err);
eb60ceac
CM		 1987}
1988
f2984462
CM		 1989static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1990{
1991 char b[BDEVNAME_SIZE];
1992
1993 if (uptodate) {
1994 set_buffer_uptodate(bh);
1995 } else {
1996 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1997 printk(KERN_WARNING "lost page write due to "
1998 "I/O error on %s\n",
1999 bdevname(bh->b_bdev, b));
2000 }
1259ab75
CM		 2001 /* note, we dont' set_buffer_write_io_error because we have
2002 * our own ways of dealing with the IO errors
2003 */
f2984462
CM		 2004 clear_buffer_uptodate(bh);
2005 }
2006 unlock_buffer(bh);
2007 put_bh(bh);
2008}
2009
a512bbf8
YZ		 2010struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2011{
2012 struct buffer_head *bh;
2013 struct buffer_head *latest = NULL;
2014 struct btrfs_super_block *super;
2015 int i;
2016 u64 transid = 0;
2017 u64 bytenr;
2018
2019 /* we would like to check all the supers, but that would make
2020 * a btrfs mount succeed after a mkfs from a different FS.
2021 * So, we need to add a special mount option to scan for
2022 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2023 */
2024 for (i = 0; i < 1; i++) {
2025 bytenr = btrfs_sb_offset(i);
2026 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2027 break;
2028 bh = __bread(bdev, bytenr / 4096, 4096);
2029 if (!bh)
2030 continue;
2031
2032 super = (struct btrfs_super_block *)bh->b_data;
2033 if (btrfs_super_bytenr(super) != bytenr ||
2034 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2035 sizeof(super->magic))) {
2036 brelse(bh);
2037 continue;
2038 }
2039
2040 if (!latest || btrfs_super_generation(super) > transid) {
2041 brelse(latest);
2042 latest = bh;
2043 transid = btrfs_super_generation(super);
2044 } else {
2045 brelse(bh);
2046 }
2047 }
2048 return latest;
2049}
2050
4eedeb75
HH		 2051/*
2052 * this should be called twice, once with wait == 0 and
2053 * once with wait == 1. When wait == 0 is done, all the buffer heads
2054 * we write are pinned.
2055 *
2056 * They are released when wait == 1 is done.
2057 * max_mirrors must be the same for both runs, and it indicates how
2058 * many supers on this one device should be written.
2059 *
2060 * max_mirrors == 0 means to write them all.
2061 */
a512bbf8
YZ		 2062static int write_dev_supers(struct btrfs_device *device,
2063 struct btrfs_super_block *sb,
2064 int do_barriers, int wait, int max_mirrors)
2065{
2066 struct buffer_head *bh;
2067 int i;
2068 int ret;
2069 int errors = 0;
2070 u32 crc;
2071 u64 bytenr;
2072 int last_barrier = 0;
2073
2074 if (max_mirrors == 0)
2075 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2076
2077 /* make sure only the last submit_bh does a barrier */
2078 if (do_barriers) {
2079 for (i = 0; i < max_mirrors; i++) {
2080 bytenr = btrfs_sb_offset(i);
2081 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2082 device->total_bytes)
2083 break;
2084 last_barrier = i;
2085 }
2086 }
2087
2088 for (i = 0; i < max_mirrors; i++) {
2089 bytenr = btrfs_sb_offset(i);
2090 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2091 break;
2092
2093 if (wait) {
2094 bh = __find_get_block(device->bdev, bytenr / 4096,
2095 BTRFS_SUPER_INFO_SIZE);
2096 BUG_ON(!bh);
a512bbf8 2097 wait_on_buffer(bh);
4eedeb75
HH		 2098 if (!buffer_uptodate(bh))
2099 errors++;
2100
2101 /* drop our reference */
2102 brelse(bh);
2103
2104 /* drop the reference from the wait == 0 run */
2105 brelse(bh);
2106 continue;
a512bbf8
YZ		 2107 } else {
2108 btrfs_set_super_bytenr(sb, bytenr);
2109
2110 crc = ~(u32)0;
2111 crc = btrfs_csum_data(NULL, (char *)sb +
2112 BTRFS_CSUM_SIZE, crc,
2113 BTRFS_SUPER_INFO_SIZE -
2114 BTRFS_CSUM_SIZE);
2115 btrfs_csum_final(crc, sb->csum);
2116
4eedeb75
HH		 2117 /*
2118 * one reference for us, and we leave it for the
2119 * caller
2120 */
a512bbf8
YZ		 2121 bh = __getblk(device->bdev, bytenr / 4096,
2122 BTRFS_SUPER_INFO_SIZE);
2123 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2124
4eedeb75 2125 /* one reference for submit_bh */
a512bbf8 2126 get_bh(bh);
4eedeb75
HH		 2127
2128 set_buffer_uptodate(bh);
a512bbf8
YZ		 2129 lock_buffer(bh);
2130 bh->b_end_io = btrfs_end_buffer_write_sync;
2131 }
2132
2133 if (i == last_barrier && do_barriers && device->barriers) {
2134 ret = submit_bh(WRITE_BARRIER, bh);
2135 if (ret == -EOPNOTSUPP) {
2136 printk("btrfs: disabling barriers on dev %s\n",
2137 device->name);
2138 set_buffer_uptodate(bh);
2139 device->barriers = 0;
4eedeb75 2140 /* one reference for submit_bh */
a512bbf8
YZ		 2141 get_bh(bh);
2142 lock_buffer(bh);
ffbd517d 2143 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2144 }
2145 } else {
ffbd517d 2146 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2147 }
2148
4eedeb75 2149 if (ret)
a512bbf8 2150 errors++;
a512bbf8
YZ		 2151 }
2152 return errors < i ? 0 : -1;
2153}
2154
2155int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2156{
e5e9a520 2157 struct list_head *head;
f2984462 2158 struct btrfs_device *dev;
a061fc8d 2159 struct btrfs_super_block *sb;
f2984462 2160 struct btrfs_dev_item *dev_item;
f2984462
CM		 2161 int ret;
2162 int do_barriers;
a236aed1
CM		 2163 int max_errors;
2164 int total_errors = 0;
a061fc8d 2165 u64 flags;
f2984462 2166
a236aed1 2167 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2168 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2169
a061fc8d
CM		 2170 sb = &root->fs_info->super_for_commit;
2171 dev_item = &sb->dev_item;
e5e9a520
CM		 2172
2173 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2174 head = &root->fs_info->fs_devices->devices;
c6e30871 2175 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2176 if (!dev->bdev) {
2177 total_errors++;
2178 continue;
2179 }
2b82032c 2180 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2181 continue;
2182
2b82032c 2183 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2184 btrfs_set_stack_device_type(dev_item, dev->type);
2185 btrfs_set_stack_device_id(dev_item, dev->devid);
2186 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2187 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2188 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2189 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2190 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2191 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2192 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2193
a061fc8d
CM		 2194 flags = btrfs_super_flags(sb);
2195 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2196
a512bbf8 2197 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2198 if (ret)
2199 total_errors++;
f2984462 2200 }
a236aed1 2201 if (total_errors > max_errors) {
d397712b
CM		 2202 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2203 total_errors);
a236aed1
CM		 2204 BUG();
2205 }
f2984462 2206
a512bbf8 2207 total_errors = 0;
c6e30871 2208 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2209 if (!dev->bdev)
2210 continue;
2b82032c 2211 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2212 continue;
2213
a512bbf8
YZ		 2214 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2215 if (ret)
2216 total_errors++;
f2984462 2217 }
e5e9a520 2218 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2219 if (total_errors > max_errors) {
d397712b
CM		 2220 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2221 total_errors);
a236aed1
CM		 2222 BUG();
2223 }
f2984462
CM		 2224 return 0;
2225}
2226
a512bbf8
YZ		 2227int write_ctree_super(struct btrfs_trans_handle *trans,
2228 struct btrfs_root *root, int max_mirrors)
eb60ceac 2229{
e66f709b 2230 int ret;
5f39d397 2231
a512bbf8 2232 ret = write_all_supers(root, max_mirrors);
5f39d397 2233 return ret;
cfaa7295
CM		 2234}
2235
5eda7b5e 2236int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2237{
5d4f98a2 2238 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2619ba1f
CM		 2239 radix_tree_delete(&fs_info->fs_roots_radix,
2240 (unsigned long)root->root_key.objectid);
3394e160
CM		 2241 if (root->anon_super.s_dev) {
2242 down_write(&root->anon_super.s_umount);
2243 kill_anon_super(&root->anon_super);
2244 }
2619ba1f 2245 if (root->node)
5f39d397 2246 free_extent_buffer(root->node);
2619ba1f 2247 if (root->commit_root)
5f39d397 2248 free_extent_buffer(root->commit_root);
d397712b 2249 kfree(root->name);
2619ba1f
CM		 2250 kfree(root);
2251 return 0;
2252}
2253
35b7e476 2254static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2255{
2256 int ret;
2257 struct btrfs_root *gang[8];
2258 int i;
2259
d397712b 2260 while (1) {
0f7d52f4
CM		 2261 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2262 (void **)gang, 0,
2263 ARRAY_SIZE(gang));
2264 if (!ret)
2265 break;
2619ba1f 2266 for (i = 0; i < ret; i++)
5eda7b5e 2267 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2268 }
2269 return 0;
2270}
b4100d64 2271
c146afad 2272int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2273{
c146afad
YZ		 2274 u64 root_objectid = 0;
2275 struct btrfs_root *gang[8];
2276 int i;
3768f368 2277 int ret;
e089f05c 2278
c146afad
YZ		 2279 while (1) {
2280 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2281 (void **)gang, root_objectid,
2282 ARRAY_SIZE(gang));
2283 if (!ret)
2284 break;
5d4f98a2
YZ		 2285
2286 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2287 for (i = 0; i < ret; i++) {
2288 root_objectid = gang[i]->root_key.objectid;
2289 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 2290 root_objectid);
c146afad
YZ		 2291 BUG_ON(ret);
2292 btrfs_orphan_cleanup(gang[i]);
2293 }
2294 root_objectid++;
2295 }
2296 return 0;
2297}
a2135011 2298
c146afad
YZ		 2299int btrfs_commit_super(struct btrfs_root *root)
2300{
2301 struct btrfs_trans_handle *trans;
2302 int ret;
a74a4b97 2303
c146afad 2304 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 2305 btrfs_clean_old_snapshots(root);
c146afad 2306 mutex_unlock(&root->fs_info->cleaner_mutex);
79154b1b 2307 trans = btrfs_start_transaction(root, 1);
54aa1f4d 2308 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2309 BUG_ON(ret);
2310 /* run commit again to drop the original snapshot */
79154b1b
CM		 2311 trans = btrfs_start_transaction(root, 1);
2312 btrfs_commit_transaction(trans, root);
2313 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2314 BUG_ON(ret);
d6bfde87 2315
a512bbf8 2316 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2317 return ret;
2318}
2319
2320int close_ctree(struct btrfs_root *root)
2321{
2322 struct btrfs_fs_info *fs_info = root->fs_info;
2323 int ret;
2324
2325 fs_info->closing = 1;
2326 smp_mb();
2327
2328 kthread_stop(root->fs_info->transaction_kthread);
2329 kthread_stop(root->fs_info->cleaner_kthread);
2330
2331 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2332 ret = btrfs_commit_super(root);
d397712b
CM		 2333 if (ret)
2334 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2335 }
0f7d52f4 2336
f25784b3
YZ		 2337 fs_info->closing = 2;
2338 smp_mb();
2339
b0c68f8b 2340 if (fs_info->delalloc_bytes) {
d397712b 2341 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2342 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2343 }
31153d81 2344 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2345 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2346 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2347 }
bcc63abb 2348
5d4f98a2
YZ		 2349 free_extent_buffer(fs_info->extent_root->node);
2350 free_extent_buffer(fs_info->extent_root->commit_root);
2351 free_extent_buffer(fs_info->tree_root->node);
2352 free_extent_buffer(fs_info->tree_root->commit_root);
2353 free_extent_buffer(root->fs_info->chunk_root->node);
2354 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2355 free_extent_buffer(root->fs_info->dev_root->node);
2356 free_extent_buffer(root->fs_info->dev_root->commit_root);
2357 free_extent_buffer(root->fs_info->csum_root->node);
2358 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2359
9078a3e1 2360 btrfs_free_block_groups(root->fs_info);
68b38550 2361 btrfs_free_pinned_extents(root->fs_info);
d10c5f31 2362
c146afad 2363 del_fs_roots(fs_info);
d10c5f31 2364
c146afad 2365 iput(fs_info->btree_inode);
9ad6b7bc 2366
247e743c 2367 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2368 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2369 btrfs_stop_workers(&fs_info->workers);
2370 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2371 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2372 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2373 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2374 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2375
dfe25020 2376 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2377 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2378
04160088 2379 bdi_destroy(&fs_info->bdi);
0b86a832 2380
0f7d52f4 2381 kfree(fs_info->extent_root);
0f7d52f4 2382 kfree(fs_info->tree_root);
0b86a832
CM		 2383 kfree(fs_info->chunk_root);
2384 kfree(fs_info->dev_root);
d20f7043 2385 kfree(fs_info->csum_root);
eb60ceac
CM		 2386 return 0;
2387}
2388
1259ab75 2389int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2390{
1259ab75 2391 int ret;
810191ff 2392 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75
CM		 2393
2394 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2395 if (!ret)
2396 return ret;
2397
2398 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2399 parent_transid);
2400 return !ret;
5f39d397
CM		 2401}
2402
2403int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2404{
810191ff 2405 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2406 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2407 buf);
2408}
6702ed49 2409
5f39d397
CM		 2410void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2411{
810191ff 2412 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2413 u64 transid = btrfs_header_generation(buf);
2414 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2415 int was_dirty;
b4ce94de 2416
b9447ef8 2417 btrfs_assert_tree_locked(buf);
ccd467d6 2418 if (transid != root->fs_info->generation) {
d397712b
CM		 2419 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2420 "found %llu running %llu\n",
db94535d 2421 (unsigned long long)buf->start,
d397712b
CM		 2422 (unsigned long long)transid,
2423 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2424 WARN_ON(1);
2425 }
b9473439
CM		 2426 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2427 buf);
2428 if (!was_dirty) {
2429 spin_lock(&root->fs_info->delalloc_lock);
2430 root->fs_info->dirty_metadata_bytes += buf->len;
2431 spin_unlock(&root->fs_info->delalloc_lock);
2432 }
eb60ceac
CM		 2433}
2434
d3c2fdcf 2435void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2436{
188de649
CM		 2437 /*
2438 * looks as though older kernels can get into trouble with
2439 * this code, they end up stuck in balance_dirty_pages forever
2440 */
d6bfde87 2441 u64 num_dirty;
771ed689 2442 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2443
6933c02e 2444 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2445 return;
2446
585ad2c3
CM		 2447 num_dirty = root->fs_info->dirty_metadata_bytes;
2448
d6bfde87
CM		 2449 if (num_dirty > thresh) {
2450 balance_dirty_pages_ratelimited_nr(
d7fc640e 2451 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2452 }
188de649 2453 return;
35b7e476 2454}
6b80053d 2455
ca7a79ad 2456int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2457{
810191ff 2458 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2459 int ret;
ca7a79ad 2460 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2461 if (ret == 0)
b4ce94de 2462 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2463 return ret;
6b80053d 2464}
0da5468f 2465
4bef0848
CM		 2466int btree_lock_page_hook(struct page *page)
2467{
2468 struct inode *inode = page->mapping->host;
b9473439 2469 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2470 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2471 struct extent_buffer *eb;
2472 unsigned long len;
2473 u64 bytenr = page_offset(page);
2474
2475 if (page->private == EXTENT_PAGE_PRIVATE)
2476 goto out;
2477
2478 len = page->private >> 2;
2479 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2480 if (!eb)
2481 goto out;
2482
2483 btrfs_tree_lock(eb);
4bef0848 2484 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2485
2486 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2487 spin_lock(&root->fs_info->delalloc_lock);
2488 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2489 root->fs_info->dirty_metadata_bytes -= eb->len;
2490 else
2491 WARN_ON(1);
2492 spin_unlock(&root->fs_info->delalloc_lock);
2493 }
2494
4bef0848
CM		 2495 btrfs_tree_unlock(eb);
2496 free_extent_buffer(eb);
2497out:
2498 lock_page(page);
2499 return 0;
2500}
2501
d1310b2e 2502static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2503 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2504 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2505 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2506 /* note we're sharing with inode.c for the merge bio hook */
2507 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2508};
Linux 6.x block layer and io_uring tree(s)