projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: do not reuse objectid of deleted snapshot/subvol
[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
890871be 126 read_lock(&em_tree->lock);
d1310b2e 127 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 128 if (em) {
129 em->bdev =
130 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 131 read_unlock(&em_tree->lock);
5f39d397 132 goto out;
a061fc8d 133 }
890871be 134 read_unlock(&em_tree->lock);
7b13b7b1 135
5f39d397
CM		 136 em = alloc_extent_map(GFP_NOFS);
137 if (!em) {
138 em = ERR_PTR(-ENOMEM);
139 goto out;
140 }
141 em->start = 0;
0afbaf8c 142 em->len = (u64)-1;
c8b97818 143 em->block_len = (u64)-1;
5f39d397 144 em->block_start = 0;
a061fc8d 145 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 146
890871be 147 write_lock(&em_tree->lock);
5f39d397
CM		 148 ret = add_extent_mapping(em_tree, em);
149 if (ret == -EEXIST) {
0afbaf8c
CM		 150 u64 failed_start = em->start;
151 u64 failed_len = em->len;
152
5f39d397 153 free_extent_map(em);
7b13b7b1 154 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 155 if (em) {
7b13b7b1 156 ret = 0;
0afbaf8c
CM		 157 } else {
158 em = lookup_extent_mapping(em_tree, failed_start,
159 failed_len);
7b13b7b1 160 ret = -EIO;
0afbaf8c 161 }
5f39d397 162 } else if (ret) {
7b13b7b1
CM		 163 free_extent_map(em);
164 em = NULL;
5f39d397 165 }
890871be 166 write_unlock(&em_tree->lock);
7b13b7b1
CM		 167
168 if (ret)
169 em = ERR_PTR(ret);
5f39d397
CM		 170out:
171 return em;
7eccb903
CM		 172}
173
19c00ddc
CM		 174u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
175{
163e783e 176 return crc32c(seed, data, len);
19c00ddc
CM		 177}
178
179void btrfs_csum_final(u32 crc, char *result)
180{
181 *(__le32 *)result = ~cpu_to_le32(crc);
182}
183
d352ac68
CM		 184/*
185 * compute the csum for a btree block, and either verify it or write it
186 * into the csum field of the block.
187 */
19c00ddc
CM		 188static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
189 int verify)
190{
607d432d
JB		 191 u16 csum_size =
192 btrfs_super_csum_size(&root->fs_info->super_copy);
193 char *result = NULL;
19c00ddc
CM		 194 unsigned long len;
195 unsigned long cur_len;
196 unsigned long offset = BTRFS_CSUM_SIZE;
197 char *map_token = NULL;
198 char *kaddr;
199 unsigned long map_start;
200 unsigned long map_len;
201 int err;
202 u32 crc = ~(u32)0;
607d432d 203 unsigned long inline_result;
19c00ddc
CM		 204
205 len = buf->len - offset;
d397712b 206 while (len > 0) {
19c00ddc
CM		 207 err = map_private_extent_buffer(buf, offset, 32,
208 &map_token, &kaddr,
209 &map_start, &map_len, KM_USER0);
d397712b 210 if (err)
19c00ddc 211 return 1;
19c00ddc
CM		 212 cur_len = min(len, map_len - (offset - map_start));
213 crc = btrfs_csum_data(root, kaddr + offset - map_start,
214 crc, cur_len);
215 len -= cur_len;
216 offset += cur_len;
217 unmap_extent_buffer(buf, map_token, KM_USER0);
218 }
607d432d
JB		 219 if (csum_size > sizeof(inline_result)) {
220 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
221 if (!result)
222 return 1;
223 } else {
224 result = (char *)&inline_result;
225 }
226
19c00ddc
CM		 227 btrfs_csum_final(crc, result);
228
229 if (verify) {
607d432d 230 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 231 u32 val;
232 u32 found = 0;
607d432d 233 memcpy(&found, result, csum_size);
e4204ded 234
607d432d 235 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 236 if (printk_ratelimit()) {
237 printk(KERN_INFO "btrfs: %s checksum verify "
238 "failed on %llu wanted %X found %X "
239 "level %d\n",
240 root->fs_info->sb->s_id,
241 (unsigned long long)buf->start, val, found,
242 btrfs_header_level(buf));
243 }
607d432d
JB		 244 if (result != (char *)&inline_result)
245 kfree(result);
19c00ddc
CM		 246 return 1;
247 }
248 } else {
607d432d 249 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 250 }
607d432d
JB		 251 if (result != (char *)&inline_result)
252 kfree(result);
19c00ddc
CM		 253 return 0;
254}
255
d352ac68
CM		 256/*
257 * we can't consider a given block up to date unless the transid of the
258 * block matches the transid in the parent node's pointer. This is how we
259 * detect blocks that either didn't get written at all or got written
260 * in the wrong place.
261 */
1259ab75
CM		 262static int verify_parent_transid(struct extent_io_tree *io_tree,
263 struct extent_buffer *eb, u64 parent_transid)
264{
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;
13a8a7c8 898 root->highest_objectid = 0;
58176a96 899 root->name = NULL;
4313b399 900 root->in_sysfs = 0;
5d4f98a2 901 root->inode_tree.rb_node = NULL;
0b86a832
CM		 902
903 INIT_LIST_HEAD(&root->dirty_list);
7b128766 904 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 905 INIT_LIST_HEAD(&root->root_list);
925baedd 906 spin_lock_init(&root->node_lock);
bcc63abb 907 spin_lock_init(&root->list_lock);
5d4f98a2 908 spin_lock_init(&root->inode_lock);
a2135011 909 mutex_init(&root->objectid_mutex);
e02119d5 910 mutex_init(&root->log_mutex);
7237f183
YZ		 911 init_waitqueue_head(&root->log_writer_wait);
912 init_waitqueue_head(&root->log_commit_wait[0]);
913 init_waitqueue_head(&root->log_commit_wait[1]);
914 atomic_set(&root->log_commit[0], 0);
915 atomic_set(&root->log_commit[1], 0);
916 atomic_set(&root->log_writers, 0);
917 root->log_batch = 0;
918 root->log_transid = 0;
d0c803c4
CM		 919 extent_io_tree_init(&root->dirty_log_pages,
920 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 921
3768f368
CM		 922 memset(&root->root_key, 0, sizeof(root->root_key));
923 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 924 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 925 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 926 root->defrag_trans_start = fs_info->generation;
58176a96 927 init_completion(&root->kobj_unregister);
6702ed49
CM		 928 root->defrag_running = 0;
929 root->defrag_level = 0;
4d775673 930 root->root_key.objectid = objectid;
3394e160
CM		 931 root->anon_super.s_root = NULL;
932 root->anon_super.s_dev = 0;
933 INIT_LIST_HEAD(&root->anon_super.s_list);
934 INIT_LIST_HEAD(&root->anon_super.s_instances);
935 init_rwsem(&root->anon_super.s_umount);
936
3768f368
CM		 937 return 0;
938}
939
db94535d 940static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 941 struct btrfs_fs_info *fs_info,
942 u64 objectid,
e20d96d6 943 struct btrfs_root *root)
3768f368
CM		 944{
945 int ret;
db94535d 946 u32 blocksize;
84234f3a 947 u64 generation;
3768f368 948
db94535d 949 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 950 tree_root->sectorsize, tree_root->stripesize,
951 root, fs_info, objectid);
3768f368
CM		 952 ret = btrfs_find_last_root(tree_root, objectid,
953 &root->root_item, &root->root_key);
954 BUG_ON(ret);
955
84234f3a 956 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 957 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
958 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 959 blocksize, generation);
5d4f98a2 960 root->commit_root = btrfs_root_node(root);
3768f368 961 BUG_ON(!root->node);
d97e63b6
CM		 962 return 0;
963}
964
e02119d5
CM		 965int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
966 struct btrfs_fs_info *fs_info)
967{
968 struct extent_buffer *eb;
d0c803c4
CM		 969 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
970 u64 start = 0;
971 u64 end = 0;
e02119d5
CM		 972 int ret;
973
d0c803c4 974 if (!log_root_tree)
e02119d5
CM		 975 return 0;
976
d397712b 977 while (1) {
d0c803c4
CM		 978 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
979 0, &start, &end, EXTENT_DIRTY);
980 if (ret)
981 break;
982
983 clear_extent_dirty(&log_root_tree->dirty_log_pages,
984 start, end, GFP_NOFS);
985 }
e02119d5
CM		 986 eb = fs_info->log_root_tree->node;
987
988 WARN_ON(btrfs_header_level(eb) != 0);
989 WARN_ON(btrfs_header_nritems(eb) != 0);
990
d00aff00
CM		 991 ret = btrfs_free_reserved_extent(fs_info->tree_root,
992 eb->start, eb->len);
e02119d5
CM		 993 BUG_ON(ret);
994
995 free_extent_buffer(eb);
996 kfree(fs_info->log_root_tree);
997 fs_info->log_root_tree = NULL;
998 return 0;
999}
1000
7237f183
YZ		 1001static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1002 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1003{
1004 struct btrfs_root *root;
1005 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1006 struct extent_buffer *leaf;
e02119d5
CM		 1007
1008 root = kzalloc(sizeof(*root), GFP_NOFS);
1009 if (!root)
7237f183 1010 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1011
1012 __setup_root(tree_root->nodesize, tree_root->leafsize,
1013 tree_root->sectorsize, tree_root->stripesize,
1014 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1015
1016 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1017 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1018 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1019 /*
1020 * log trees do not get reference counted because they go away
1021 * before a real commit is actually done. They do store pointers
1022 * to file data extents, and those reference counts still get
1023 * updated (along with back refs to the log tree).
1024 */
e02119d5
CM		 1025 root->ref_cows = 0;
1026
5d4f98a2
YZ		 1027 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1028 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1029 if (IS_ERR(leaf)) {
1030 kfree(root);
1031 return ERR_CAST(leaf);
1032 }
e02119d5 1033
5d4f98a2
YZ		 1034 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1035 btrfs_set_header_bytenr(leaf, leaf->start);
1036 btrfs_set_header_generation(leaf, trans->transid);
1037 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1038 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1039 root->node = leaf;
e02119d5
CM		 1040
1041 write_extent_buffer(root->node, root->fs_info->fsid,
1042 (unsigned long)btrfs_header_fsid(root->node),
1043 BTRFS_FSID_SIZE);
1044 btrfs_mark_buffer_dirty(root->node);
1045 btrfs_tree_unlock(root->node);
7237f183
YZ		 1046 return root;
1047}
1048
1049int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1050 struct btrfs_fs_info *fs_info)
1051{
1052 struct btrfs_root *log_root;
1053
1054 log_root = alloc_log_tree(trans, fs_info);
1055 if (IS_ERR(log_root))
1056 return PTR_ERR(log_root);
1057 WARN_ON(fs_info->log_root_tree);
1058 fs_info->log_root_tree = log_root;
1059 return 0;
1060}
1061
1062int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1063 struct btrfs_root *root)
1064{
1065 struct btrfs_root *log_root;
1066 struct btrfs_inode_item *inode_item;
1067
1068 log_root = alloc_log_tree(trans, root->fs_info);
1069 if (IS_ERR(log_root))
1070 return PTR_ERR(log_root);
1071
1072 log_root->last_trans = trans->transid;
1073 log_root->root_key.offset = root->root_key.objectid;
1074
1075 inode_item = &log_root->root_item.inode;
1076 inode_item->generation = cpu_to_le64(1);
1077 inode_item->size = cpu_to_le64(3);
1078 inode_item->nlink = cpu_to_le32(1);
1079 inode_item->nbytes = cpu_to_le64(root->leafsize);
1080 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1081
5d4f98a2 1082 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1083
1084 WARN_ON(root->log_root);
1085 root->log_root = log_root;
1086 root->log_transid = 0;
e02119d5
CM		 1087 return 0;
1088}
1089
1090struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1091 struct btrfs_key *location)
1092{
1093 struct btrfs_root *root;
1094 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1095 struct btrfs_path *path;
5f39d397 1096 struct extent_buffer *l;
84234f3a 1097 u64 generation;
db94535d 1098 u32 blocksize;
0f7d52f4
CM		 1099 int ret = 0;
1100
5eda7b5e 1101 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1102 if (!root)
0f7d52f4 1103 return ERR_PTR(-ENOMEM);
0f7d52f4 1104 if (location->offset == (u64)-1) {
db94535d 1105 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1106 location->objectid, root);
1107 if (ret) {
0f7d52f4
CM		 1108 kfree(root);
1109 return ERR_PTR(ret);
1110 }
13a8a7c8 1111 goto out;
0f7d52f4
CM		 1112 }
1113
db94535d 1114 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1115 tree_root->sectorsize, tree_root->stripesize,
1116 root, fs_info, location->objectid);
0f7d52f4
CM		 1117
1118 path = btrfs_alloc_path();
1119 BUG_ON(!path);
1120 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1121 if (ret == 0) {
1122 l = path->nodes[0];
1123 read_extent_buffer(l, &root->root_item,
1124 btrfs_item_ptr_offset(l, path->slots[0]),
1125 sizeof(root->root_item));
1126 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1127 }
0f7d52f4
CM		 1128 btrfs_free_path(path);
1129 if (ret) {
13a8a7c8
YZ		 1130 if (ret > 0)
1131 ret = -ENOENT;
0f7d52f4
CM		 1132 return ERR_PTR(ret);
1133 }
13a8a7c8 1134
84234f3a 1135 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1136 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1137 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1138 blocksize, generation);
5d4f98a2 1139 root->commit_root = btrfs_root_node(root);
0f7d52f4 1140 BUG_ON(!root->node);
13a8a7c8
YZ		 1141out:
1142 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1143 root->ref_cows = 1;
13a8a7c8 1144
5eda7b5e
CM		 1145 return root;
1146}
1147
dc17ff8f
CM		 1148struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1149 u64 root_objectid)
1150{
1151 struct btrfs_root *root;
1152
1153 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1154 return fs_info->tree_root;
1155 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1156 return fs_info->extent_root;
1157
1158 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1159 (unsigned long)root_objectid);
1160 return root;
1161}
1162
edbd8d4e
CM		 1163struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1164 struct btrfs_key *location)
5eda7b5e
CM		 1165{
1166 struct btrfs_root *root;
1167 int ret;
1168
edbd8d4e
CM		 1169 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1170 return fs_info->tree_root;
1171 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1172 return fs_info->extent_root;
8f18cf13
CM		 1173 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1174 return fs_info->chunk_root;
1175 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1176 return fs_info->dev_root;
0403e47e
YZ		 1177 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1178 return fs_info->csum_root;
edbd8d4e 1179
5eda7b5e
CM		 1180 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1181 (unsigned long)location->objectid);
1182 if (root)
1183 return root;
1184
e02119d5 1185 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1186 if (IS_ERR(root))
1187 return root;
3394e160
CM		 1188
1189 set_anon_super(&root->anon_super, NULL);
1190
2619ba1f
CM		 1191 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1192 (unsigned long)root->root_key.objectid,
0f7d52f4
CM		 1193 root);
1194 if (ret) {
5f39d397 1195 free_extent_buffer(root->node);
0f7d52f4
CM		 1196 kfree(root);
1197 return ERR_PTR(ret);
1198 }
c146afad
YZ		 1199 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1200 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 1201 root->root_key.objectid);
c146afad
YZ		 1202 BUG_ON(ret);
1203 btrfs_orphan_cleanup(root);
1204 }
edbd8d4e
CM		 1205 return root;
1206}
1207
1208struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1209 struct btrfs_key *location,
1210 const char *name, int namelen)
1211{
1212 struct btrfs_root *root;
1213 int ret;
1214
1215 root = btrfs_read_fs_root_no_name(fs_info, location);
1216 if (!root)
1217 return NULL;
58176a96 1218
4313b399
CM		 1219 if (root->in_sysfs)
1220 return root;
1221
58176a96
JB		 1222 ret = btrfs_set_root_name(root, name, namelen);
1223 if (ret) {
5f39d397 1224 free_extent_buffer(root->node);
58176a96
JB		 1225 kfree(root);
1226 return ERR_PTR(ret);
1227 }
ea9e8b11 1228#if 0
58176a96
JB		 1229 ret = btrfs_sysfs_add_root(root);
1230 if (ret) {
5f39d397 1231 free_extent_buffer(root->node);
58176a96
JB		 1232 kfree(root->name);
1233 kfree(root);
1234 return ERR_PTR(ret);
1235 }
ea9e8b11 1236#endif
4313b399 1237 root->in_sysfs = 1;
0f7d52f4
CM		 1238 return root;
1239}
04160088
CM		 1240
1241static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1242{
1243 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1244 int ret = 0;
04160088
CM		 1245 struct btrfs_device *device;
1246 struct backing_dev_info *bdi;
b7967db7 1247
c6e30871 1248 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1249 if (!device->bdev)
1250 continue;
04160088
CM		 1251 bdi = blk_get_backing_dev_info(device->bdev);
1252 if (bdi && bdi_congested(bdi, bdi_bits)) {
1253 ret = 1;
1254 break;
1255 }
1256 }
1257 return ret;
1258}
1259
38b66988
CM		 1260/*
1261 * this unplugs every device on the box, and it is only used when page
1262 * is null
1263 */
1264static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1265{
38b66988
CM		 1266 struct btrfs_device *device;
1267 struct btrfs_fs_info *info;
1268
1269 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1270 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1271 if (!device->bdev)
1272 continue;
1273
38b66988 1274 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1275 if (bdi->unplug_io_fn)
38b66988 1276 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1277 }
1278}
1279
b2950863 1280static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1281{
38b66988 1282 struct inode *inode;
f2d8d74d
CM		 1283 struct extent_map_tree *em_tree;
1284 struct extent_map *em;
bcbfce8a 1285 struct address_space *mapping;
38b66988
CM		 1286 u64 offset;
1287
bcbfce8a 1288 /* the generic O_DIRECT read code does this */
9f0ba5bd 1289 if (1 || !page) {
38b66988
CM		 1290 __unplug_io_fn(bdi, page);
1291 return;
1292 }
1293
bcbfce8a
CM		 1294 /*
1295 * page->mapping may change at any time. Get a consistent copy
1296 * and use that for everything below
1297 */
1298 smp_mb();
1299 mapping = page->mapping;
1300 if (!mapping)
1301 return;
1302
1303 inode = mapping->host;
240d5d48
CM		 1304
1305 /*
1306 * don't do the expensive searching for a small number of
1307 * devices
1308 */
1309 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1310 __unplug_io_fn(bdi, page);
1311 return;
1312 }
1313
38b66988 1314 offset = page_offset(page);
04160088 1315
f2d8d74d 1316 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1317 read_lock(&em_tree->lock);
f2d8d74d 1318 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1319 read_unlock(&em_tree->lock);
89642229
CM		 1320 if (!em) {
1321 __unplug_io_fn(bdi, page);
f2d8d74d 1322 return;
89642229 1323 }
f2d8d74d 1324
89642229
CM		 1325 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1326 free_extent_map(em);
1327 __unplug_io_fn(bdi, page);
1328 return;
1329 }
f2d8d74d
CM		 1330 offset = offset - em->start;
1331 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1332 em->block_start + offset, page);
1333 free_extent_map(em);
04160088
CM		 1334}
1335
ad081f14
JA		 1336/*
1337 * If this fails, caller must call bdi_destroy() to get rid of the
1338 * bdi again.
1339 */
04160088
CM		 1340static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1341{
ad081f14
JA		 1342 int err;
1343
1344 bdi->capabilities = BDI_CAP_MAP_COPY;
1345 err = bdi_init(bdi);
1346 if (err)
1347 return err;
1348
1349 err = bdi_register(bdi, NULL, "btrfs-%d",
1350 atomic_inc_return(&btrfs_bdi_num));
1351 if (err)
1352 return err;
1353
4575c9cc 1354 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1355 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1356 bdi->unplug_io_data = info;
1357 bdi->congested_fn = btrfs_congested_fn;
1358 bdi->congested_data = info;
1359 return 0;
1360}
1361
ce9adaa5
CM		 1362static int bio_ready_for_csum(struct bio *bio)
1363{
1364 u64 length = 0;
1365 u64 buf_len = 0;
1366 u64 start = 0;
1367 struct page *page;
1368 struct extent_io_tree *io_tree = NULL;
1369 struct btrfs_fs_info *info = NULL;
1370 struct bio_vec *bvec;
1371 int i;
1372 int ret;
1373
1374 bio_for_each_segment(bvec, bio, i) {
1375 page = bvec->bv_page;
1376 if (page->private == EXTENT_PAGE_PRIVATE) {
1377 length += bvec->bv_len;
1378 continue;
1379 }
1380 if (!page->private) {
1381 length += bvec->bv_len;
1382 continue;
1383 }
1384 length = bvec->bv_len;
1385 buf_len = page->private >> 2;
1386 start = page_offset(page) + bvec->bv_offset;
1387 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1388 info = BTRFS_I(page->mapping->host)->root->fs_info;
1389 }
1390 /* are we fully contained in this bio? */
1391 if (buf_len <= length)
1392 return 1;
1393
1394 ret = extent_range_uptodate(io_tree, start + length,
1395 start + buf_len - 1);
ce9adaa5
CM		 1396 return ret;
1397}
1398
8b712842
CM		 1399/*
1400 * called by the kthread helper functions to finally call the bio end_io
1401 * functions. This is where read checksum verification actually happens
1402 */
1403static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1404{
ce9adaa5 1405 struct bio *bio;
8b712842
CM		 1406 struct end_io_wq *end_io_wq;
1407 struct btrfs_fs_info *fs_info;
ce9adaa5 1408 int error;
ce9adaa5 1409
8b712842
CM		 1410 end_io_wq = container_of(work, struct end_io_wq, work);
1411 bio = end_io_wq->bio;
1412 fs_info = end_io_wq->info;
ce9adaa5 1413
cad321ad 1414 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1415 * be checksummed at once. This makes sure the entire block is in
1416 * ram and up to date before trying to verify things. For
1417 * blocksize <= pagesize, it is basically a noop
1418 */
cad321ad
CM		 1419 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1420 !bio_ready_for_csum(bio)) {
d20f7043 1421 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1422 &end_io_wq->work);
1423 return;
1424 }
1425 error = end_io_wq->error;
1426 bio->bi_private = end_io_wq->private;
1427 bio->bi_end_io = end_io_wq->end_io;
1428 kfree(end_io_wq);
8b712842 1429 bio_endio(bio, error);
44b8bd7e
CM		 1430}
1431
a74a4b97
CM		 1432static int cleaner_kthread(void *arg)
1433{
1434 struct btrfs_root *root = arg;
1435
1436 do {
1437 smp_mb();
1438 if (root->fs_info->closing)
1439 break;
1440
1441 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1442 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 1443 btrfs_clean_old_snapshots(root);
a74a4b97
CM		 1444 mutex_unlock(&root->fs_info->cleaner_mutex);
1445
1446 if (freezing(current)) {
1447 refrigerator();
1448 } else {
1449 smp_mb();
1450 if (root->fs_info->closing)
1451 break;
1452 set_current_state(TASK_INTERRUPTIBLE);
1453 schedule();
1454 __set_current_state(TASK_RUNNING);
1455 }
1456 } while (!kthread_should_stop());
1457 return 0;
1458}
1459
1460static int transaction_kthread(void *arg)
1461{
1462 struct btrfs_root *root = arg;
1463 struct btrfs_trans_handle *trans;
1464 struct btrfs_transaction *cur;
1465 unsigned long now;
1466 unsigned long delay;
1467 int ret;
1468
1469 do {
1470 smp_mb();
1471 if (root->fs_info->closing)
1472 break;
1473
1474 delay = HZ * 30;
1475 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1476 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1477
1478 mutex_lock(&root->fs_info->trans_mutex);
1479 cur = root->fs_info->running_transaction;
1480 if (!cur) {
1481 mutex_unlock(&root->fs_info->trans_mutex);
1482 goto sleep;
1483 }
31153d81 1484
a74a4b97
CM		 1485 now = get_seconds();
1486 if (now < cur->start_time || now - cur->start_time < 30) {
1487 mutex_unlock(&root->fs_info->trans_mutex);
1488 delay = HZ * 5;
1489 goto sleep;
1490 }
1491 mutex_unlock(&root->fs_info->trans_mutex);
a74a4b97
CM		 1492 trans = btrfs_start_transaction(root, 1);
1493 ret = btrfs_commit_transaction(trans, root);
56bec294 1494
a74a4b97
CM		 1495sleep:
1496 wake_up_process(root->fs_info->cleaner_kthread);
1497 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1498
1499 if (freezing(current)) {
1500 refrigerator();
1501 } else {
1502 if (root->fs_info->closing)
1503 break;
1504 set_current_state(TASK_INTERRUPTIBLE);
1505 schedule_timeout(delay);
1506 __set_current_state(TASK_RUNNING);
1507 }
1508 } while (!kthread_should_stop());
1509 return 0;
1510}
1511
8a4b83cc 1512struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1513 struct btrfs_fs_devices *fs_devices,
1514 char *options)
2e635a27 1515{
db94535d
CM		 1516 u32 sectorsize;
1517 u32 nodesize;
1518 u32 leafsize;
1519 u32 blocksize;
87ee04eb 1520 u32 stripesize;
84234f3a 1521 u64 generation;
f2b636e8 1522 u64 features;
3de4586c 1523 struct btrfs_key location;
a061fc8d 1524 struct buffer_head *bh;
e02119d5 1525 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1526 GFP_NOFS);
d20f7043
CM		 1527 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1528 GFP_NOFS);
e02119d5 1529 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1530 GFP_NOFS);
8790d502 1531 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1532 GFP_NOFS);
e02119d5 1533 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1534 GFP_NOFS);
e02119d5 1535 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1536 GFP_NOFS);
e02119d5
CM		 1537 struct btrfs_root *log_tree_root;
1538
eb60ceac 1539 int ret;
e58ca020 1540 int err = -EINVAL;
4543df7e 1541
2c90e5d6 1542 struct btrfs_super_block *disk_super;
8790d502 1543
0463bb4e 1544 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1545 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1546 err = -ENOMEM;
1547 goto fail;
1548 }
0f7d52f4 1549 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
8fd17795 1550 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1551 INIT_LIST_HEAD(&fs_info->dead_roots);
19c00ddc 1552 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1553 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1554 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1555 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1556 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1557 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1558 spin_lock_init(&fs_info->ref_cache_lock);
19c00ddc 1559
58176a96 1560 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1561 fs_info->tree_root = tree_root;
1562 fs_info->extent_root = extent_root;
d20f7043 1563 fs_info->csum_root = csum_root;
0b86a832
CM		 1564 fs_info->chunk_root = chunk_root;
1565 fs_info->dev_root = dev_root;
8a4b83cc 1566 fs_info->fs_devices = fs_devices;
0b86a832 1567 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1568 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1569 btrfs_mapping_init(&fs_info->mapping_tree);
cb03c743 1570 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1571 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1572 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1573 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1574 fs_info->sb = sb;
c59f8951 1575 fs_info->max_extent = (u64)-1;
6f568d35 1576 fs_info->max_inline = 8192 * 1024;
ad081f14
JA		 1577 if (setup_bdi(fs_info, &fs_info->bdi))
1578 goto fail_bdi;
d98237b3
CM		 1579 fs_info->btree_inode = new_inode(sb);
1580 fs_info->btree_inode->i_ino = 1;
2c90e5d6 1581 fs_info->btree_inode->i_nlink = 1;
97e728d4 1582 fs_info->metadata_ratio = 8;
c8b97818 1583
b34b086c
CM		 1584 fs_info->thread_pool_size = min_t(unsigned long,
1585 num_online_cpus() + 2, 8);
0afbaf8c 1586
3eaa2885
CM		 1587 INIT_LIST_HEAD(&fs_info->ordered_extents);
1588 spin_lock_init(&fs_info->ordered_extent_lock);
1589
a061fc8d
CM		 1590 sb->s_blocksize = 4096;
1591 sb->s_blocksize_bits = blksize_bits(4096);
1592
0afbaf8c
CM		 1593 /*
1594 * we set the i_size on the btree inode to the max possible int.
1595 * the real end of the address space is determined by all of
1596 * the devices in the system
1597 */
1598 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1599 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1600 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1601
5d4f98a2 1602 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1603 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1604 fs_info->btree_inode->i_mapping,
1605 GFP_NOFS);
d1310b2e
CM		 1606 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1607 GFP_NOFS);
1608
1609 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1610
0f9dd46c
JB		 1611 spin_lock_init(&fs_info->block_group_cache_lock);
1612 fs_info->block_group_cache_tree.rb_node = NULL;
1613
11833d66 1614 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1615 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1616 extent_io_tree_init(&fs_info->freed_extents[1],
1617 fs_info->btree_inode->i_mapping, GFP_NOFS);
1618 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1619 fs_info->do_barriers = 1;
e18e4809 1620
0f7d52f4
CM		 1621 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1622 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1623 sizeof(struct btrfs_key));
22b0ebda 1624 insert_inode_hash(fs_info->btree_inode);
39279cc3 1625
79154b1b 1626 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1627 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1628 mutex_init(&fs_info->tree_log_mutex);
a2135011 1629 mutex_init(&fs_info->drop_mutex);
925baedd 1630 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1631 mutex_init(&fs_info->transaction_kthread_mutex);
1632 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1633 mutex_init(&fs_info->volume_mutex);
1a40e23b 1634 mutex_init(&fs_info->tree_reloc_mutex);
276e680d 1635 init_rwsem(&fs_info->extent_commit_sem);
fa9c0d79
CM		 1636
1637 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1638 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1639
e6dcd2dc 1640 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1641 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1642 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1643
0b86a832 1644 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1645 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1646
d98237b3 1647
a512bbf8 1648 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1649 if (!bh)
39279cc3 1650 goto fail_iput;
39279cc3 1651
a061fc8d 1652 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1653 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1654 sizeof(fs_info->super_for_commit));
a061fc8d 1655 brelse(bh);
5f39d397 1656
a061fc8d 1657 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1658
5f39d397 1659 disk_super = &fs_info->super_copy;
0f7d52f4 1660 if (!btrfs_super_root(disk_super))
c6e2bac1 1661 goto fail_iput;
0f7d52f4 1662
2b82032c
YZ		 1663 ret = btrfs_parse_options(tree_root, options);
1664 if (ret) {
1665 err = ret;
c6e2bac1 1666 goto fail_iput;
2b82032c 1667 }
dfe25020 1668
f2b636e8
JB		 1669 features = btrfs_super_incompat_flags(disk_super) &
1670 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1671 if (features) {
1672 printk(KERN_ERR "BTRFS: couldn't mount because of "
1673 "unsupported optional features (%Lx).\n",
21380931 1674 (unsigned long long)features);
f2b636e8 1675 err = -EINVAL;
c6e2bac1 1676 goto fail_iput;
f2b636e8
JB		 1677 }
1678
5d4f98a2
YZ		 1679 features = btrfs_super_incompat_flags(disk_super);
1680 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1681 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1682 btrfs_set_super_incompat_flags(disk_super, features);
1683 }
1684
f2b636e8
JB		 1685 features = btrfs_super_compat_ro_flags(disk_super) &
1686 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1687 if (!(sb->s_flags & MS_RDONLY) && features) {
1688 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1689 "unsupported option features (%Lx).\n",
21380931 1690 (unsigned long long)features);
f2b636e8 1691 err = -EINVAL;
c6e2bac1 1692 goto fail_iput;
f2b636e8 1693 }
9042846b 1694printk("thread pool is %d\n", fs_info->thread_pool_size);
4543df7e
CM		 1695 /*
1696 * we need to start all the end_io workers up front because the
1697 * queue work function gets called at interrupt time, and so it
1698 * cannot dynamically grow.
1699 */
5443be45
CM		 1700 btrfs_init_workers(&fs_info->workers, "worker",
1701 fs_info->thread_pool_size);
c8b97818 1702
771ed689
CM		 1703 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1704 fs_info->thread_pool_size);
1705
5443be45 1706 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209
CM		 1707 min_t(u64, fs_devices->num_devices,
1708 fs_info->thread_pool_size));
61b49440
CM		 1709
1710 /* a higher idle thresh on the submit workers makes it much more
1711 * likely that bios will be send down in a sane order to the
1712 * devices
1713 */
1714 fs_info->submit_workers.idle_thresh = 64;
53863232 1715
771ed689 1716 fs_info->workers.idle_thresh = 16;
4a69a410 1717 fs_info->workers.ordered = 1;
61b49440 1718
771ed689
CM		 1719 fs_info->delalloc_workers.idle_thresh = 2;
1720 fs_info->delalloc_workers.ordered = 1;
1721
5443be45
CM		 1722 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1723 btrfs_init_workers(&fs_info->endio_workers, "endio",
1724 fs_info->thread_pool_size);
d20f7043
CM		 1725 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1726 fs_info->thread_pool_size);
cad321ad
CM		 1727 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1728 "endio-meta-write", fs_info->thread_pool_size);
5443be45 1729 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
e6dcd2dc 1730 fs_info->thread_pool_size);
61b49440
CM		 1731
1732 /*
1733 * endios are largely parallel and should have a very
1734 * low idle thresh
1735 */
1736 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1737 fs_info->endio_meta_workers.idle_thresh = 4;
1738
9042846b
CM		 1739 fs_info->endio_write_workers.idle_thresh = 2;
1740 fs_info->endio_meta_write_workers.idle_thresh = 2;
1741
1742 fs_info->endio_workers.atomic_worker_start = 1;
1743 fs_info->endio_meta_workers.atomic_worker_start = 1;
1744 fs_info->endio_write_workers.atomic_worker_start = 1;
1745 fs_info->endio_meta_write_workers.atomic_worker_start = 1;
61b49440 1746
4543df7e 1747 btrfs_start_workers(&fs_info->workers, 1);
1cc127b5 1748 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1749 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1750 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1751 btrfs_start_workers(&fs_info->endio_workers, 1);
1752 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1753 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1754 btrfs_start_workers(&fs_info->endio_write_workers, 1);
4543df7e 1755
4575c9cc 1756 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1757 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1758 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1759
db94535d
CM		 1760 nodesize = btrfs_super_nodesize(disk_super);
1761 leafsize = btrfs_super_leafsize(disk_super);
1762 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1763 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1764 tree_root->nodesize = nodesize;
1765 tree_root->leafsize = leafsize;
1766 tree_root->sectorsize = sectorsize;
87ee04eb 1767 tree_root->stripesize = stripesize;
a061fc8d
CM		 1768
1769 sb->s_blocksize = sectorsize;
1770 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1771
39279cc3
CM		 1772 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1773 sizeof(disk_super->magic))) {
d397712b 1774 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1775 goto fail_sb_buffer;
1776 }
19c00ddc 1777
925baedd 1778 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1779 ret = btrfs_read_sys_array(tree_root);
925baedd 1780 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1781 if (ret) {
d397712b
CM		 1782 printk(KERN_WARNING "btrfs: failed to read the system "
1783 "array on %s\n", sb->s_id);
5d4f98a2 1784 goto fail_sb_buffer;
84eed90f 1785 }
0b86a832
CM		 1786
1787 blocksize = btrfs_level_size(tree_root,
1788 btrfs_super_chunk_root_level(disk_super));
84234f3a 1789 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1790
1791 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1792 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1793
1794 chunk_root->node = read_tree_block(chunk_root,
1795 btrfs_super_chunk_root(disk_super),
84234f3a 1796 blocksize, generation);
0b86a832 1797 BUG_ON(!chunk_root->node);
83121942
DW		 1798 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1799 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1800 sb->s_id);
1801 goto fail_chunk_root;
1802 }
5d4f98a2
YZ		 1803 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1804 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1805
e17cade2 1806 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1807 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1808 BTRFS_UUID_SIZE);
e17cade2 1809
925baedd 1810 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1811 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1812 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1813 if (ret) {
d397712b
CM		 1814 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1815 sb->s_id);
2b82032c
YZ		 1816 goto fail_chunk_root;
1817 }
0b86a832 1818
dfe25020
CM		 1819 btrfs_close_extra_devices(fs_devices);
1820
db94535d
CM		 1821 blocksize = btrfs_level_size(tree_root,
1822 btrfs_super_root_level(disk_super));
84234f3a 1823 generation = btrfs_super_generation(disk_super);
0b86a832 1824
e20d96d6 1825 tree_root->node = read_tree_block(tree_root,
db94535d 1826 btrfs_super_root(disk_super),
84234f3a 1827 blocksize, generation);
39279cc3 1828 if (!tree_root->node)
2b82032c 1829 goto fail_chunk_root;
83121942
DW		 1830 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1831 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1832 sb->s_id);
1833 goto fail_tree_root;
1834 }
5d4f98a2
YZ		 1835 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1836 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1837
1838 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1839 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1840 if (ret)
39279cc3 1841 goto fail_tree_root;
0b86a832
CM		 1842 extent_root->track_dirty = 1;
1843
1844 ret = find_and_setup_root(tree_root, fs_info,
1845 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1846 if (ret)
1847 goto fail_extent_root;
5d4f98a2 1848 dev_root->track_dirty = 1;
3768f368 1849
d20f7043
CM		 1850 ret = find_and_setup_root(tree_root, fs_info,
1851 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1852 if (ret)
5d4f98a2 1853 goto fail_dev_root;
d20f7043
CM		 1854
1855 csum_root->track_dirty = 1;
1856
9078a3e1
CM		 1857 btrfs_read_block_groups(extent_root);
1858
2d69a0f8 1859 fs_info->generation = generation;
c146afad 1860 fs_info->last_trans_committed = generation;
d18a2c44
CM		 1861 fs_info->data_alloc_profile = (u64)-1;
1862 fs_info->metadata_alloc_profile = (u64)-1;
1863 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
a74a4b97
CM		 1864 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1865 "btrfs-cleaner");
57506d50 1866 if (IS_ERR(fs_info->cleaner_kthread))
d20f7043 1867 goto fail_csum_root;
a74a4b97
CM		 1868
1869 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1870 tree_root,
1871 "btrfs-transaction");
57506d50 1872 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1873 goto fail_cleaner;
a74a4b97 1874
c289811c
CM		 1875 if (!btrfs_test_opt(tree_root, SSD) &&
1876 !btrfs_test_opt(tree_root, NOSSD) &&
1877 !fs_info->fs_devices->rotating) {
1878 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1879 "mode\n");
1880 btrfs_set_opt(fs_info->mount_opt, SSD);
1881 }
1882
e02119d5 1883 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1884 u64 bytenr = btrfs_super_log_root(disk_super);
1885
7c2ca468 1886 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1887 printk(KERN_WARNING "Btrfs log replay required "
1888 "on RO media\n");
7c2ca468
CM		 1889 err = -EIO;
1890 goto fail_trans_kthread;
1891 }
e02119d5
CM		 1892 blocksize =
1893 btrfs_level_size(tree_root,
1894 btrfs_super_log_root_level(disk_super));
d18a2c44 1895
e02119d5
CM		 1896 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1897 GFP_NOFS);
1898
1899 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1900 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1901
1902 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1903 blocksize,
1904 generation + 1);
e02119d5
CM		 1905 ret = btrfs_recover_log_trees(log_tree_root);
1906 BUG_ON(ret);
e556ce2c
YZ		 1907
1908 if (sb->s_flags & MS_RDONLY) {
1909 ret = btrfs_commit_super(tree_root);
1910 BUG_ON(ret);
1911 }
e02119d5 1912 }
1a40e23b 1913
7c2ca468 1914 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1915 ret = btrfs_recover_relocation(tree_root);
7c2ca468
CM		 1916 BUG_ON(ret);
1917 }
1a40e23b 1918
3de4586c
CM		 1919 location.objectid = BTRFS_FS_TREE_OBJECTID;
1920 location.type = BTRFS_ROOT_ITEM_KEY;
1921 location.offset = (u64)-1;
1922
3de4586c
CM		 1923 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1924 if (!fs_info->fs_root)
7c2ca468 1925 goto fail_trans_kthread;
c289811c 1926
0f7d52f4 1927 return tree_root;
39279cc3 1928
7c2ca468
CM		 1929fail_trans_kthread:
1930 kthread_stop(fs_info->transaction_kthread);
3f157a2f 1931fail_cleaner:
a74a4b97 1932 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 1933
1934 /*
1935 * make sure we're done with the btree inode before we stop our
1936 * kthreads
1937 */
1938 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1939 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1940
d20f7043
CM		 1941fail_csum_root:
1942 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 1943 free_extent_buffer(csum_root->commit_root);
1944fail_dev_root:
1945 free_extent_buffer(dev_root->node);
1946 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 1947fail_extent_root:
1948 free_extent_buffer(extent_root->node);
5d4f98a2 1949 free_extent_buffer(extent_root->commit_root);
39279cc3 1950fail_tree_root:
5f39d397 1951 free_extent_buffer(tree_root->node);
5d4f98a2 1952 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 1953fail_chunk_root:
1954 free_extent_buffer(chunk_root->node);
5d4f98a2 1955 free_extent_buffer(chunk_root->commit_root);
39279cc3 1956fail_sb_buffer:
247e743c 1957 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 1958 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 1959 btrfs_stop_workers(&fs_info->workers);
1960 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 1961 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 1962 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 1963 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 1964 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 1965fail_iput:
7c2ca468 1966 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 1967 iput(fs_info->btree_inode);
7e662854 1968
dfe25020 1969 btrfs_close_devices(fs_info->fs_devices);
84eed90f 1970 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 1971fail_bdi:
7e662854 1972 bdi_destroy(&fs_info->bdi);
7e662854 1973fail:
39279cc3
CM		 1974 kfree(extent_root);
1975 kfree(tree_root);
1976 kfree(fs_info);
83afeac4
JM		 1977 kfree(chunk_root);
1978 kfree(dev_root);
d20f7043 1979 kfree(csum_root);
39279cc3 1980 return ERR_PTR(err);
eb60ceac
CM		 1981}
1982
f2984462
CM		 1983static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1984{
1985 char b[BDEVNAME_SIZE];
1986
1987 if (uptodate) {
1988 set_buffer_uptodate(bh);
1989 } else {
1990 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1991 printk(KERN_WARNING "lost page write due to "
1992 "I/O error on %s\n",
1993 bdevname(bh->b_bdev, b));
1994 }
1259ab75
CM		 1995 /* note, we dont' set_buffer_write_io_error because we have
1996 * our own ways of dealing with the IO errors
1997 */
f2984462
CM		 1998 clear_buffer_uptodate(bh);
1999 }
2000 unlock_buffer(bh);
2001 put_bh(bh);
2002}
2003
a512bbf8
YZ		 2004struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2005{
2006 struct buffer_head *bh;
2007 struct buffer_head *latest = NULL;
2008 struct btrfs_super_block *super;
2009 int i;
2010 u64 transid = 0;
2011 u64 bytenr;
2012
2013 /* we would like to check all the supers, but that would make
2014 * a btrfs mount succeed after a mkfs from a different FS.
2015 * So, we need to add a special mount option to scan for
2016 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2017 */
2018 for (i = 0; i < 1; i++) {
2019 bytenr = btrfs_sb_offset(i);
2020 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2021 break;
2022 bh = __bread(bdev, bytenr / 4096, 4096);
2023 if (!bh)
2024 continue;
2025
2026 super = (struct btrfs_super_block *)bh->b_data;
2027 if (btrfs_super_bytenr(super) != bytenr ||
2028 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2029 sizeof(super->magic))) {
2030 brelse(bh);
2031 continue;
2032 }
2033
2034 if (!latest || btrfs_super_generation(super) > transid) {
2035 brelse(latest);
2036 latest = bh;
2037 transid = btrfs_super_generation(super);
2038 } else {
2039 brelse(bh);
2040 }
2041 }
2042 return latest;
2043}
2044
4eedeb75
HH		 2045/*
2046 * this should be called twice, once with wait == 0 and
2047 * once with wait == 1. When wait == 0 is done, all the buffer heads
2048 * we write are pinned.
2049 *
2050 * They are released when wait == 1 is done.
2051 * max_mirrors must be the same for both runs, and it indicates how
2052 * many supers on this one device should be written.
2053 *
2054 * max_mirrors == 0 means to write them all.
2055 */
a512bbf8
YZ		 2056static int write_dev_supers(struct btrfs_device *device,
2057 struct btrfs_super_block *sb,
2058 int do_barriers, int wait, int max_mirrors)
2059{
2060 struct buffer_head *bh;
2061 int i;
2062 int ret;
2063 int errors = 0;
2064 u32 crc;
2065 u64 bytenr;
2066 int last_barrier = 0;
2067
2068 if (max_mirrors == 0)
2069 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2070
2071 /* make sure only the last submit_bh does a barrier */
2072 if (do_barriers) {
2073 for (i = 0; i < max_mirrors; i++) {
2074 bytenr = btrfs_sb_offset(i);
2075 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2076 device->total_bytes)
2077 break;
2078 last_barrier = i;
2079 }
2080 }
2081
2082 for (i = 0; i < max_mirrors; i++) {
2083 bytenr = btrfs_sb_offset(i);
2084 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2085 break;
2086
2087 if (wait) {
2088 bh = __find_get_block(device->bdev, bytenr / 4096,
2089 BTRFS_SUPER_INFO_SIZE);
2090 BUG_ON(!bh);
a512bbf8 2091 wait_on_buffer(bh);
4eedeb75
HH		 2092 if (!buffer_uptodate(bh))
2093 errors++;
2094
2095 /* drop our reference */
2096 brelse(bh);
2097
2098 /* drop the reference from the wait == 0 run */
2099 brelse(bh);
2100 continue;
a512bbf8
YZ		 2101 } else {
2102 btrfs_set_super_bytenr(sb, bytenr);
2103
2104 crc = ~(u32)0;
2105 crc = btrfs_csum_data(NULL, (char *)sb +
2106 BTRFS_CSUM_SIZE, crc,
2107 BTRFS_SUPER_INFO_SIZE -
2108 BTRFS_CSUM_SIZE);
2109 btrfs_csum_final(crc, sb->csum);
2110
4eedeb75
HH		 2111 /*
2112 * one reference for us, and we leave it for the
2113 * caller
2114 */
a512bbf8
YZ		 2115 bh = __getblk(device->bdev, bytenr / 4096,
2116 BTRFS_SUPER_INFO_SIZE);
2117 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2118
4eedeb75 2119 /* one reference for submit_bh */
a512bbf8 2120 get_bh(bh);
4eedeb75
HH		 2121
2122 set_buffer_uptodate(bh);
a512bbf8
YZ		 2123 lock_buffer(bh);
2124 bh->b_end_io = btrfs_end_buffer_write_sync;
2125 }
2126
2127 if (i == last_barrier && do_barriers && device->barriers) {
2128 ret = submit_bh(WRITE_BARRIER, bh);
2129 if (ret == -EOPNOTSUPP) {
2130 printk("btrfs: disabling barriers on dev %s\n",
2131 device->name);
2132 set_buffer_uptodate(bh);
2133 device->barriers = 0;
4eedeb75 2134 /* one reference for submit_bh */
a512bbf8
YZ		 2135 get_bh(bh);
2136 lock_buffer(bh);
ffbd517d 2137 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2138 }
2139 } else {
ffbd517d 2140 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2141 }
2142
4eedeb75 2143 if (ret)
a512bbf8 2144 errors++;
a512bbf8
YZ		 2145 }
2146 return errors < i ? 0 : -1;
2147}
2148
2149int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2150{
e5e9a520 2151 struct list_head *head;
f2984462 2152 struct btrfs_device *dev;
a061fc8d 2153 struct btrfs_super_block *sb;
f2984462 2154 struct btrfs_dev_item *dev_item;
f2984462
CM		 2155 int ret;
2156 int do_barriers;
a236aed1
CM		 2157 int max_errors;
2158 int total_errors = 0;
a061fc8d 2159 u64 flags;
f2984462 2160
a236aed1 2161 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2162 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2163
a061fc8d
CM		 2164 sb = &root->fs_info->super_for_commit;
2165 dev_item = &sb->dev_item;
e5e9a520
CM		 2166
2167 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2168 head = &root->fs_info->fs_devices->devices;
c6e30871 2169 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2170 if (!dev->bdev) {
2171 total_errors++;
2172 continue;
2173 }
2b82032c 2174 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2175 continue;
2176
2b82032c 2177 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2178 btrfs_set_stack_device_type(dev_item, dev->type);
2179 btrfs_set_stack_device_id(dev_item, dev->devid);
2180 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2181 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2182 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2183 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2184 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2185 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2186 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2187
a061fc8d
CM		 2188 flags = btrfs_super_flags(sb);
2189 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2190
a512bbf8 2191 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2192 if (ret)
2193 total_errors++;
f2984462 2194 }
a236aed1 2195 if (total_errors > max_errors) {
d397712b
CM		 2196 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2197 total_errors);
a236aed1
CM		 2198 BUG();
2199 }
f2984462 2200
a512bbf8 2201 total_errors = 0;
c6e30871 2202 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2203 if (!dev->bdev)
2204 continue;
2b82032c 2205 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2206 continue;
2207
a512bbf8
YZ		 2208 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2209 if (ret)
2210 total_errors++;
f2984462 2211 }
e5e9a520 2212 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2213 if (total_errors > max_errors) {
d397712b
CM		 2214 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2215 total_errors);
a236aed1
CM		 2216 BUG();
2217 }
f2984462
CM		 2218 return 0;
2219}
2220
a512bbf8
YZ		 2221int write_ctree_super(struct btrfs_trans_handle *trans,
2222 struct btrfs_root *root, int max_mirrors)
eb60ceac 2223{
e66f709b 2224 int ret;
5f39d397 2225
a512bbf8 2226 ret = write_all_supers(root, max_mirrors);
5f39d397 2227 return ret;
cfaa7295
CM		 2228}
2229
5eda7b5e 2230int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2231{
5d4f98a2 2232 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2619ba1f
CM		 2233 radix_tree_delete(&fs_info->fs_roots_radix,
2234 (unsigned long)root->root_key.objectid);
3394e160
CM		 2235 if (root->anon_super.s_dev) {
2236 down_write(&root->anon_super.s_umount);
2237 kill_anon_super(&root->anon_super);
2238 }
2619ba1f 2239 if (root->node)
5f39d397 2240 free_extent_buffer(root->node);
2619ba1f 2241 if (root->commit_root)
5f39d397 2242 free_extent_buffer(root->commit_root);
d397712b 2243 kfree(root->name);
2619ba1f
CM		 2244 kfree(root);
2245 return 0;
2246}
2247
35b7e476 2248static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2249{
2250 int ret;
2251 struct btrfs_root *gang[8];
2252 int i;
2253
d397712b 2254 while (1) {
0f7d52f4
CM		 2255 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2256 (void **)gang, 0,
2257 ARRAY_SIZE(gang));
2258 if (!ret)
2259 break;
2619ba1f 2260 for (i = 0; i < ret; i++)
5eda7b5e 2261 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2262 }
2263 return 0;
2264}
b4100d64 2265
c146afad 2266int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2267{
c146afad
YZ		 2268 u64 root_objectid = 0;
2269 struct btrfs_root *gang[8];
2270 int i;
3768f368 2271 int ret;
e089f05c 2272
c146afad
YZ		 2273 while (1) {
2274 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2275 (void **)gang, root_objectid,
2276 ARRAY_SIZE(gang));
2277 if (!ret)
2278 break;
5d4f98a2
YZ		 2279
2280 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2281 for (i = 0; i < ret; i++) {
2282 root_objectid = gang[i]->root_key.objectid;
2283 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 2284 root_objectid);
c146afad
YZ		 2285 BUG_ON(ret);
2286 btrfs_orphan_cleanup(gang[i]);
2287 }
2288 root_objectid++;
2289 }
2290 return 0;
2291}
a2135011 2292
c146afad
YZ		 2293int btrfs_commit_super(struct btrfs_root *root)
2294{
2295 struct btrfs_trans_handle *trans;
2296 int ret;
a74a4b97 2297
c146afad 2298 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 2299 btrfs_clean_old_snapshots(root);
c146afad 2300 mutex_unlock(&root->fs_info->cleaner_mutex);
79154b1b 2301 trans = btrfs_start_transaction(root, 1);
54aa1f4d 2302 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2303 BUG_ON(ret);
2304 /* run commit again to drop the original snapshot */
79154b1b
CM		 2305 trans = btrfs_start_transaction(root, 1);
2306 btrfs_commit_transaction(trans, root);
2307 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2308 BUG_ON(ret);
d6bfde87 2309
a512bbf8 2310 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2311 return ret;
2312}
2313
2314int close_ctree(struct btrfs_root *root)
2315{
2316 struct btrfs_fs_info *fs_info = root->fs_info;
2317 int ret;
2318
2319 fs_info->closing = 1;
2320 smp_mb();
2321
2322 kthread_stop(root->fs_info->transaction_kthread);
2323 kthread_stop(root->fs_info->cleaner_kthread);
2324
2325 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2326 ret = btrfs_commit_super(root);
d397712b
CM		 2327 if (ret)
2328 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2329 }
0f7d52f4 2330
f25784b3
YZ		 2331 fs_info->closing = 2;
2332 smp_mb();
2333
b0c68f8b 2334 if (fs_info->delalloc_bytes) {
d397712b 2335 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2336 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2337 }
31153d81 2338 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2339 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2340 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2341 }
bcc63abb 2342
5d4f98a2
YZ		 2343 free_extent_buffer(fs_info->extent_root->node);
2344 free_extent_buffer(fs_info->extent_root->commit_root);
2345 free_extent_buffer(fs_info->tree_root->node);
2346 free_extent_buffer(fs_info->tree_root->commit_root);
2347 free_extent_buffer(root->fs_info->chunk_root->node);
2348 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2349 free_extent_buffer(root->fs_info->dev_root->node);
2350 free_extent_buffer(root->fs_info->dev_root->commit_root);
2351 free_extent_buffer(root->fs_info->csum_root->node);
2352 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2353
9078a3e1 2354 btrfs_free_block_groups(root->fs_info);
d10c5f31 2355
c146afad 2356 del_fs_roots(fs_info);
d10c5f31 2357
c146afad 2358 iput(fs_info->btree_inode);
9ad6b7bc 2359
247e743c 2360 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2361 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2362 btrfs_stop_workers(&fs_info->workers);
2363 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2364 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2365 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2366 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2367 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2368
dfe25020 2369 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2370 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2371
04160088 2372 bdi_destroy(&fs_info->bdi);
0b86a832 2373
0f7d52f4 2374 kfree(fs_info->extent_root);
0f7d52f4 2375 kfree(fs_info->tree_root);
0b86a832
CM		 2376 kfree(fs_info->chunk_root);
2377 kfree(fs_info->dev_root);
d20f7043 2378 kfree(fs_info->csum_root);
eb60ceac
CM		 2379 return 0;
2380}
2381
1259ab75 2382int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2383{
1259ab75 2384 int ret;
810191ff 2385 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75
CM		 2386
2387 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2388 if (!ret)
2389 return ret;
2390
2391 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2392 parent_transid);
2393 return !ret;
5f39d397
CM		 2394}
2395
2396int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2397{
810191ff 2398 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2399 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2400 buf);
2401}
6702ed49 2402
5f39d397
CM		 2403void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2404{
810191ff 2405 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2406 u64 transid = btrfs_header_generation(buf);
2407 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2408 int was_dirty;
b4ce94de 2409
b9447ef8 2410 btrfs_assert_tree_locked(buf);
ccd467d6 2411 if (transid != root->fs_info->generation) {
d397712b
CM		 2412 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2413 "found %llu running %llu\n",
db94535d 2414 (unsigned long long)buf->start,
d397712b
CM		 2415 (unsigned long long)transid,
2416 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2417 WARN_ON(1);
2418 }
b9473439
CM		 2419 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2420 buf);
2421 if (!was_dirty) {
2422 spin_lock(&root->fs_info->delalloc_lock);
2423 root->fs_info->dirty_metadata_bytes += buf->len;
2424 spin_unlock(&root->fs_info->delalloc_lock);
2425 }
eb60ceac
CM		 2426}
2427
d3c2fdcf 2428void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2429{
188de649
CM		 2430 /*
2431 * looks as though older kernels can get into trouble with
2432 * this code, they end up stuck in balance_dirty_pages forever
2433 */
d6bfde87 2434 u64 num_dirty;
771ed689 2435 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2436
6933c02e 2437 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2438 return;
2439
585ad2c3
CM		 2440 num_dirty = root->fs_info->dirty_metadata_bytes;
2441
d6bfde87
CM		 2442 if (num_dirty > thresh) {
2443 balance_dirty_pages_ratelimited_nr(
d7fc640e 2444 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2445 }
188de649 2446 return;
35b7e476 2447}
6b80053d 2448
ca7a79ad 2449int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2450{
810191ff 2451 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2452 int ret;
ca7a79ad 2453 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2454 if (ret == 0)
b4ce94de 2455 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2456 return ret;
6b80053d 2457}
0da5468f 2458
4bef0848
CM		 2459int btree_lock_page_hook(struct page *page)
2460{
2461 struct inode *inode = page->mapping->host;
b9473439 2462 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2463 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2464 struct extent_buffer *eb;
2465 unsigned long len;
2466 u64 bytenr = page_offset(page);
2467
2468 if (page->private == EXTENT_PAGE_PRIVATE)
2469 goto out;
2470
2471 len = page->private >> 2;
2472 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2473 if (!eb)
2474 goto out;
2475
2476 btrfs_tree_lock(eb);
4bef0848 2477 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2478
2479 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2480 spin_lock(&root->fs_info->delalloc_lock);
2481 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2482 root->fs_info->dirty_metadata_bytes -= eb->len;
2483 else
2484 WARN_ON(1);
2485 spin_unlock(&root->fs_info->delalloc_lock);
2486 }
2487
4bef0848
CM		 2488 btrfs_tree_unlock(eb);
2489 free_extent_buffer(eb);
2490out:
2491 lock_page(page);
2492 return 0;
2493}
2494
d1310b2e 2495static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2496 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2497 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2498 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2499 /* note we're sharing with inode.c for the merge bio hook */
2500 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2501};
Linux 6.x block layer and io_uring tree(s)