projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Introduce global metadata reservation
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
4b4e25f2 31#include "compat.h"
eb60ceac
CM		 32#include "ctree.h"
33#include "disk-io.h"
e089f05c 34#include "transaction.h"
0f7d52f4 35#include "btrfs_inode.h"
0b86a832 36#include "volumes.h"
db94535d 37#include "print-tree.h"
8b712842 38#include "async-thread.h"
925baedd 39#include "locking.h"
e02119d5 40#include "tree-log.h"
fa9c0d79 41#include "free-space-cache.h"
eb60ceac 42
d1310b2e 43static struct extent_io_ops btree_extent_io_ops;
8b712842 44static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 45static void free_fs_root(struct btrfs_root *root);
ce9adaa5 46
d352ac68
CM		 47/*
48 * end_io_wq structs are used to do processing in task context when an IO is
49 * complete. This is used during reads to verify checksums, and it is used
50 * by writes to insert metadata for new file extents after IO is complete.
51 */
ce9adaa5
CM		 52struct end_io_wq {
53 struct bio *bio;
54 bio_end_io_t *end_io;
55 void *private;
56 struct btrfs_fs_info *info;
57 int error;
22c59948 58 int metadata;
ce9adaa5 59 struct list_head list;
8b712842 60 struct btrfs_work work;
ce9adaa5 61};
0da5468f 62
d352ac68
CM		 63/*
64 * async submit bios are used to offload expensive checksumming
65 * onto the worker threads. They checksum file and metadata bios
66 * just before they are sent down the IO stack.
67 */
44b8bd7e
CM		 68struct async_submit_bio {
69 struct inode *inode;
70 struct bio *bio;
71 struct list_head list;
4a69a410
CM		 72 extent_submit_bio_hook_t *submit_bio_start;
73 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 74 int rw;
75 int mirror_num;
c8b97818 76 unsigned long bio_flags;
8b712842 77 struct btrfs_work work;
44b8bd7e
CM		 78};
79
4008c04a
CM		 80/* These are used to set the lockdep class on the extent buffer locks.
81 * The class is set by the readpage_end_io_hook after the buffer has
82 * passed csum validation but before the pages are unlocked.
83 *
84 * The lockdep class is also set by btrfs_init_new_buffer on freshly
85 * allocated blocks.
86 *
87 * The class is based on the level in the tree block, which allows lockdep
88 * to know that lower nodes nest inside the locks of higher nodes.
89 *
90 * We also add a check to make sure the highest level of the tree is
91 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
92 * code needs update as well.
93 */
94#ifdef CONFIG_DEBUG_LOCK_ALLOC
95# if BTRFS_MAX_LEVEL != 8
96# error
97# endif
98static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
99static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
100 /* leaf */
101 "btrfs-extent-00",
102 "btrfs-extent-01",
103 "btrfs-extent-02",
104 "btrfs-extent-03",
105 "btrfs-extent-04",
106 "btrfs-extent-05",
107 "btrfs-extent-06",
108 "btrfs-extent-07",
109 /* highest possible level */
110 "btrfs-extent-08",
111};
112#endif
113
d352ac68
CM		 114/*
115 * extents on the btree inode are pretty simple, there's one extent
116 * that covers the entire device
117 */
b2950863
CH		 118static struct extent_map *btree_get_extent(struct inode *inode,
119 struct page *page, size_t page_offset, u64 start, u64 len,
120 int create)
7eccb903 121{
5f39d397
CM		 122 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
123 struct extent_map *em;
124 int ret;
125
890871be 126 read_lock(&em_tree->lock);
d1310b2e 127 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 128 if (em) {
129 em->bdev =
130 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 131 read_unlock(&em_tree->lock);
5f39d397 132 goto out;
a061fc8d 133 }
890871be 134 read_unlock(&em_tree->lock);
7b13b7b1 135
5f39d397
CM		 136 em = alloc_extent_map(GFP_NOFS);
137 if (!em) {
138 em = ERR_PTR(-ENOMEM);
139 goto out;
140 }
141 em->start = 0;
0afbaf8c 142 em->len = (u64)-1;
c8b97818 143 em->block_len = (u64)-1;
5f39d397 144 em->block_start = 0;
a061fc8d 145 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 146
890871be 147 write_lock(&em_tree->lock);
5f39d397
CM		 148 ret = add_extent_mapping(em_tree, em);
149 if (ret == -EEXIST) {
0afbaf8c
CM		 150 u64 failed_start = em->start;
151 u64 failed_len = em->len;
152
5f39d397 153 free_extent_map(em);
7b13b7b1 154 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 155 if (em) {
7b13b7b1 156 ret = 0;
0afbaf8c
CM		 157 } else {
158 em = lookup_extent_mapping(em_tree, failed_start,
159 failed_len);
7b13b7b1 160 ret = -EIO;
0afbaf8c 161 }
5f39d397 162 } else if (ret) {
7b13b7b1
CM		 163 free_extent_map(em);
164 em = NULL;
5f39d397 165 }
890871be 166 write_unlock(&em_tree->lock);
7b13b7b1
CM		 167
168 if (ret)
169 em = ERR_PTR(ret);
5f39d397
CM		 170out:
171 return em;
7eccb903
CM		 172}
173
19c00ddc
CM		 174u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
175{
163e783e 176 return crc32c(seed, data, len);
19c00ddc
CM		 177}
178
179void btrfs_csum_final(u32 crc, char *result)
180{
181 *(__le32 *)result = ~cpu_to_le32(crc);
182}
183
d352ac68
CM		 184/*
185 * compute the csum for a btree block, and either verify it or write it
186 * into the csum field of the block.
187 */
19c00ddc
CM		 188static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
189 int verify)
190{
607d432d
JB		 191 u16 csum_size =
192 btrfs_super_csum_size(&root->fs_info->super_copy);
193 char *result = NULL;
19c00ddc
CM		 194 unsigned long len;
195 unsigned long cur_len;
196 unsigned long offset = BTRFS_CSUM_SIZE;
197 char *map_token = NULL;
198 char *kaddr;
199 unsigned long map_start;
200 unsigned long map_len;
201 int err;
202 u32 crc = ~(u32)0;
607d432d 203 unsigned long inline_result;
19c00ddc
CM		 204
205 len = buf->len - offset;
d397712b 206 while (len > 0) {
19c00ddc
CM		 207 err = map_private_extent_buffer(buf, offset, 32,
208 &map_token, &kaddr,
209 &map_start, &map_len, KM_USER0);
d397712b 210 if (err)
19c00ddc 211 return 1;
19c00ddc
CM		 212 cur_len = min(len, map_len - (offset - map_start));
213 crc = btrfs_csum_data(root, kaddr + offset - map_start,
214 crc, cur_len);
215 len -= cur_len;
216 offset += cur_len;
217 unmap_extent_buffer(buf, map_token, KM_USER0);
218 }
607d432d
JB		 219 if (csum_size > sizeof(inline_result)) {
220 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
221 if (!result)
222 return 1;
223 } else {
224 result = (char *)&inline_result;
225 }
226
19c00ddc
CM		 227 btrfs_csum_final(crc, result);
228
229 if (verify) {
607d432d 230 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 231 u32 val;
232 u32 found = 0;
607d432d 233 memcpy(&found, result, csum_size);
e4204ded 234
607d432d 235 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 236 if (printk_ratelimit()) {
237 printk(KERN_INFO "btrfs: %s checksum verify "
238 "failed on %llu wanted %X found %X "
239 "level %d\n",
240 root->fs_info->sb->s_id,
241 (unsigned long long)buf->start, val, found,
242 btrfs_header_level(buf));
243 }
607d432d
JB		 244 if (result != (char *)&inline_result)
245 kfree(result);
19c00ddc
CM		 246 return 1;
247 }
248 } else {
607d432d 249 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 250 }
607d432d
JB		 251 if (result != (char *)&inline_result)
252 kfree(result);
19c00ddc
CM		 253 return 0;
254}
255
d352ac68
CM		 256/*
257 * we can't consider a given block up to date unless the transid of the
258 * block matches the transid in the parent node's pointer. This is how we
259 * detect blocks that either didn't get written at all or got written
260 * in the wrong place.
261 */
1259ab75
CM		 262static int verify_parent_transid(struct extent_io_tree *io_tree,
263 struct extent_buffer *eb, u64 parent_transid)
264{
2ac55d41 265 struct extent_state *cached_state = NULL;
1259ab75
CM		 266 int ret;
267
268 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
269 return 0;
270
2ac55d41
JB		 271 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
272 0, &cached_state, GFP_NOFS);
273 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 274 btrfs_header_generation(eb) == parent_transid) {
275 ret = 0;
276 goto out;
277 }
193f284d
CM		 278 if (printk_ratelimit()) {
279 printk("parent transid verify failed on %llu wanted %llu "
280 "found %llu\n",
281 (unsigned long long)eb->start,
282 (unsigned long long)parent_transid,
283 (unsigned long long)btrfs_header_generation(eb));
284 }
1259ab75 285 ret = 1;
2ac55d41 286 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 287out:
2ac55d41
JB		 288 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
289 &cached_state, GFP_NOFS);
1259ab75 290 return ret;
1259ab75
CM		 291}
292
d352ac68
CM		 293/*
294 * helper to read a given tree block, doing retries as required when
295 * the checksums don't match and we have alternate mirrors to try.
296 */
f188591e
CM		 297static int btree_read_extent_buffer_pages(struct btrfs_root *root,
298 struct extent_buffer *eb,
ca7a79ad 299 u64 start, u64 parent_transid)
f188591e
CM		 300{
301 struct extent_io_tree *io_tree;
302 int ret;
303 int num_copies = 0;
304 int mirror_num = 0;
305
306 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
307 while (1) {
308 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
309 btree_get_extent, mirror_num);
1259ab75
CM		 310 if (!ret &&
311 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 312 return ret;
d397712b 313
f188591e
CM		 314 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
315 eb->start, eb->len);
4235298e 316 if (num_copies == 1)
f188591e 317 return ret;
4235298e 318
f188591e 319 mirror_num++;
4235298e 320 if (mirror_num > num_copies)
f188591e 321 return ret;
f188591e 322 }
f188591e
CM		 323 return -EIO;
324}
19c00ddc 325
d352ac68 326/*
d397712b
CM		 327 * checksum a dirty tree block before IO. This has extra checks to make sure
328 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 329 */
d397712b 330
b2950863 331static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 332{
d1310b2e 333 struct extent_io_tree *tree;
35ebb934 334 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 335 u64 found_start;
336 int found_level;
337 unsigned long len;
338 struct extent_buffer *eb;
f188591e
CM		 339 int ret;
340
d1310b2e 341 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 342
343 if (page->private == EXTENT_PAGE_PRIVATE)
344 goto out;
345 if (!page->private)
346 goto out;
347 len = page->private >> 2;
d397712b
CM		 348 WARN_ON(len == 0);
349
19c00ddc 350 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 351 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
352 btrfs_header_generation(eb));
f188591e 353 BUG_ON(ret);
19c00ddc
CM		 354 found_start = btrfs_header_bytenr(eb);
355 if (found_start != start) {
55c69072
CM		 356 WARN_ON(1);
357 goto err;
358 }
359 if (eb->first_page != page) {
55c69072
CM		 360 WARN_ON(1);
361 goto err;
362 }
363 if (!PageUptodate(page)) {
55c69072
CM		 364 WARN_ON(1);
365 goto err;
19c00ddc
CM		 366 }
367 found_level = btrfs_header_level(eb);
4bef0848 368
19c00ddc 369 csum_tree_block(root, eb, 0);
55c69072 370err:
19c00ddc
CM		 371 free_extent_buffer(eb);
372out:
373 return 0;
374}
375
2b82032c
YZ		 376static int check_tree_block_fsid(struct btrfs_root *root,
377 struct extent_buffer *eb)
378{
379 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
380 u8 fsid[BTRFS_UUID_SIZE];
381 int ret = 1;
382
383 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
384 BTRFS_FSID_SIZE);
385 while (fs_devices) {
386 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
387 ret = 0;
388 break;
389 }
390 fs_devices = fs_devices->seed;
391 }
392 return ret;
393}
394
4008c04a
CM		 395#ifdef CONFIG_DEBUG_LOCK_ALLOC
396void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
397{
398 lockdep_set_class_and_name(&eb->lock,
399 &btrfs_eb_class[level],
400 btrfs_eb_name[level]);
401}
402#endif
403
b2950863 404static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 405 struct extent_state *state)
406{
407 struct extent_io_tree *tree;
408 u64 found_start;
409 int found_level;
410 unsigned long len;
411 struct extent_buffer *eb;
412 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 413 int ret = 0;
ce9adaa5
CM		 414
415 tree = &BTRFS_I(page->mapping->host)->io_tree;
416 if (page->private == EXTENT_PAGE_PRIVATE)
417 goto out;
418 if (!page->private)
419 goto out;
d397712b 420
ce9adaa5 421 len = page->private >> 2;
d397712b
CM		 422 WARN_ON(len == 0);
423
ce9adaa5 424 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 425
ce9adaa5 426 found_start = btrfs_header_bytenr(eb);
23a07867 427 if (found_start != start) {
193f284d
CM		 428 if (printk_ratelimit()) {
429 printk(KERN_INFO "btrfs bad tree block start "
430 "%llu %llu\n",
431 (unsigned long long)found_start,
432 (unsigned long long)eb->start);
433 }
f188591e 434 ret = -EIO;
ce9adaa5
CM		 435 goto err;
436 }
437 if (eb->first_page != page) {
d397712b
CM		 438 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
439 eb->first_page->index, page->index);
ce9adaa5 440 WARN_ON(1);
f188591e 441 ret = -EIO;
ce9adaa5
CM		 442 goto err;
443 }
2b82032c 444 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 445 if (printk_ratelimit()) {
446 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
447 (unsigned long long)eb->start);
448 }
1259ab75
CM		 449 ret = -EIO;
450 goto err;
451 }
ce9adaa5
CM		 452 found_level = btrfs_header_level(eb);
453
4008c04a
CM		 454 btrfs_set_buffer_lockdep_class(eb, found_level);
455
ce9adaa5 456 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 457 if (ret)
458 ret = -EIO;
ce9adaa5
CM		 459
460 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
461 end = eb->start + end - 1;
ce9adaa5
CM		 462err:
463 free_extent_buffer(eb);
464out:
f188591e 465 return ret;
ce9adaa5
CM		 466}
467
ce9adaa5 468static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 469{
470 struct end_io_wq *end_io_wq = bio->bi_private;
471 struct btrfs_fs_info *fs_info;
ce9adaa5 472
ce9adaa5 473 fs_info = end_io_wq->info;
ce9adaa5 474 end_io_wq->error = err;
8b712842
CM		 475 end_io_wq->work.func = end_workqueue_fn;
476 end_io_wq->work.flags = 0;
d20f7043
CM		 477
478 if (bio->bi_rw & (1 << BIO_RW)) {
cad321ad
CM		 479 if (end_io_wq->metadata)
480 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
481 &end_io_wq->work);
482 else
483 btrfs_queue_worker(&fs_info->endio_write_workers,
484 &end_io_wq->work);
d20f7043
CM		 485 } else {
486 if (end_io_wq->metadata)
487 btrfs_queue_worker(&fs_info->endio_meta_workers,
488 &end_io_wq->work);
489 else
490 btrfs_queue_worker(&fs_info->endio_workers,
491 &end_io_wq->work);
492 }
ce9adaa5
CM		 493}
494
22c59948
CM		 495int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
496 int metadata)
0b86a832 497{
ce9adaa5 498 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 499 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
500 if (!end_io_wq)
501 return -ENOMEM;
502
503 end_io_wq->private = bio->bi_private;
504 end_io_wq->end_io = bio->bi_end_io;
22c59948 505 end_io_wq->info = info;
ce9adaa5
CM		 506 end_io_wq->error = 0;
507 end_io_wq->bio = bio;
22c59948 508 end_io_wq->metadata = metadata;
ce9adaa5
CM		 509
510 bio->bi_private = end_io_wq;
511 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 512 return 0;
513}
514
b64a2851 515unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 516{
4854ddd0
CM		 517 unsigned long limit = min_t(unsigned long,
518 info->workers.max_workers,
519 info->fs_devices->open_devices);
520 return 256 * limit;
521}
0986fe9e 522
4854ddd0
CM		 523int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
524{
b64a2851
CM		 525 return atomic_read(&info->nr_async_bios) >
526 btrfs_async_submit_limit(info);
0986fe9e
CM		 527}
528
4a69a410
CM		 529static void run_one_async_start(struct btrfs_work *work)
530{
531 struct btrfs_fs_info *fs_info;
532 struct async_submit_bio *async;
533
534 async = container_of(work, struct async_submit_bio, work);
535 fs_info = BTRFS_I(async->inode)->root->fs_info;
536 async->submit_bio_start(async->inode, async->rw, async->bio,
537 async->mirror_num, async->bio_flags);
538}
539
540static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 541{
542 struct btrfs_fs_info *fs_info;
543 struct async_submit_bio *async;
4854ddd0 544 int limit;
8b712842
CM		 545
546 async = container_of(work, struct async_submit_bio, work);
547 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 548
b64a2851 549 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 550 limit = limit * 2 / 3;
551
8b712842 552 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 553
b64a2851
CM		 554 if (atomic_read(&fs_info->nr_async_submits) < limit &&
555 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 556 wake_up(&fs_info->async_submit_wait);
557
4a69a410 558 async->submit_bio_done(async->inode, async->rw, async->bio,
c8b97818 559 async->mirror_num, async->bio_flags);
4a69a410
CM		 560}
561
562static void run_one_async_free(struct btrfs_work *work)
563{
564 struct async_submit_bio *async;
565
566 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 567 kfree(async);
568}
569
44b8bd7e
CM		 570int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
571 int rw, struct bio *bio, int mirror_num,
c8b97818 572 unsigned long bio_flags,
4a69a410
CM		 573 extent_submit_bio_hook_t *submit_bio_start,
574 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 575{
576 struct async_submit_bio *async;
577
578 async = kmalloc(sizeof(*async), GFP_NOFS);
579 if (!async)
580 return -ENOMEM;
581
582 async->inode = inode;
583 async->rw = rw;
584 async->bio = bio;
585 async->mirror_num = mirror_num;
4a69a410
CM		 586 async->submit_bio_start = submit_bio_start;
587 async->submit_bio_done = submit_bio_done;
588
589 async->work.func = run_one_async_start;
590 async->work.ordered_func = run_one_async_done;
591 async->work.ordered_free = run_one_async_free;
592
8b712842 593 async->work.flags = 0;
c8b97818 594 async->bio_flags = bio_flags;
8c8bee1d 595
cb03c743 596 atomic_inc(&fs_info->nr_async_submits);
d313d7a3
CM		 597
598 if (rw & (1 << BIO_RW_SYNCIO))
599 btrfs_set_work_high_prio(&async->work);
600
8b712842 601 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 602
d397712b 603 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 604 atomic_read(&fs_info->nr_async_submits)) {
605 wait_event(fs_info->async_submit_wait,
606 (atomic_read(&fs_info->nr_async_submits) == 0));
607 }
608
44b8bd7e
CM		 609 return 0;
610}
611
ce3ed71a
CM		 612static int btree_csum_one_bio(struct bio *bio)
613{
614 struct bio_vec *bvec = bio->bi_io_vec;
615 int bio_index = 0;
616 struct btrfs_root *root;
617
618 WARN_ON(bio->bi_vcnt <= 0);
d397712b 619 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 620 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
621 csum_dirty_buffer(root, bvec->bv_page);
622 bio_index++;
623 bvec++;
624 }
625 return 0;
626}
627
4a69a410
CM		 628static int __btree_submit_bio_start(struct inode *inode, int rw,
629 struct bio *bio, int mirror_num,
630 unsigned long bio_flags)
22c59948 631{
8b712842
CM		 632 /*
633 * when we're called for a write, we're already in the async
5443be45 634 * submission context. Just jump into btrfs_map_bio
8b712842 635 */
4a69a410
CM		 636 btree_csum_one_bio(bio);
637 return 0;
638}
22c59948 639
4a69a410
CM		 640static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
641 int mirror_num, unsigned long bio_flags)
642{
8b712842 643 /*
4a69a410
CM		 644 * when we're called for a write, we're already in the async
645 * submission context. Just jump into btrfs_map_bio
8b712842 646 */
8b712842 647 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 648}
649
44b8bd7e 650static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 651 int mirror_num, unsigned long bio_flags)
44b8bd7e 652{
cad321ad
CM		 653 int ret;
654
655 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
656 bio, 1);
657 BUG_ON(ret);
658
44b8bd7e 659 if (!(rw & (1 << BIO_RW))) {
4a69a410
CM		 660 /*
661 * called for a read, do the setup so that checksum validation
662 * can happen in the async kernel threads
663 */
4a69a410 664 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 665 mirror_num, 0);
44b8bd7e 666 }
d313d7a3 667
cad321ad
CM		 668 /*
669 * kthread helpers are used to submit writes so that checksumming
670 * can happen in parallel across all CPUs
671 */
44b8bd7e 672 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 673 inode, rw, bio, mirror_num, 0,
4a69a410
CM		 674 __btree_submit_bio_start,
675 __btree_submit_bio_done);
44b8bd7e
CM		 676}
677
0da5468f
CM		 678static int btree_writepage(struct page *page, struct writeback_control *wbc)
679{
d1310b2e 680 struct extent_io_tree *tree;
b9473439
CM		 681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682 struct extent_buffer *eb;
683 int was_dirty;
684
d1310b2e 685 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 686 if (!(current->flags & PF_MEMALLOC)) {
687 return extent_write_full_page(tree, page,
688 btree_get_extent, wbc);
689 }
5443be45 690
b9473439
CM		 691 redirty_page_for_writepage(wbc, page);
692 eb = btrfs_find_tree_block(root, page_offset(page),
693 PAGE_CACHE_SIZE);
694 WARN_ON(!eb);
695
696 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
697 if (!was_dirty) {
698 spin_lock(&root->fs_info->delalloc_lock);
699 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
700 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 701 }
b9473439
CM		 702 free_extent_buffer(eb);
703
704 unlock_page(page);
705 return 0;
5f39d397 706}
0da5468f
CM		 707
708static int btree_writepages(struct address_space *mapping,
709 struct writeback_control *wbc)
710{
d1310b2e
CM		 711 struct extent_io_tree *tree;
712 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 713 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 714 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 715 u64 num_dirty;
24ab9cd8 716 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 717
718 if (wbc->for_kupdate)
719 return 0;
720
b9473439
CM		 721 /* this is a bit racy, but that's ok */
722 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 723 if (num_dirty < thresh)
793955bc 724 return 0;
793955bc 725 }
0da5468f
CM		 726 return extent_writepages(tree, mapping, btree_get_extent, wbc);
727}
728
b2950863 729static int btree_readpage(struct file *file, struct page *page)
5f39d397 730{
d1310b2e
CM		 731 struct extent_io_tree *tree;
732 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 733 return extent_read_full_page(tree, page, btree_get_extent);
734}
22b0ebda 735
70dec807 736static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 737{
d1310b2e
CM		 738 struct extent_io_tree *tree;
739 struct extent_map_tree *map;
5f39d397 740 int ret;
d98237b3 741
98509cfc 742 if (PageWriteback(page) || PageDirty(page))
d397712b 743 return 0;
98509cfc 744
d1310b2e
CM		 745 tree = &BTRFS_I(page->mapping->host)->io_tree;
746 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 747
7b13b7b1 748 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 749 if (!ret)
6af118ce 750 return 0;
6af118ce
CM		 751
752 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 753 if (ret == 1) {
754 ClearPagePrivate(page);
755 set_page_private(page, 0);
756 page_cache_release(page);
757 }
6af118ce 758
d98237b3
CM		 759 return ret;
760}
761
5f39d397 762static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 763{
d1310b2e
CM		 764 struct extent_io_tree *tree;
765 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 766 extent_invalidatepage(tree, page, offset);
767 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 768 if (PagePrivate(page)) {
d397712b
CM		 769 printk(KERN_WARNING "btrfs warning page private not zero "
770 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 771 ClearPagePrivate(page);
772 set_page_private(page, 0);
773 page_cache_release(page);
774 }
d98237b3
CM		 775}
776
7f09410b 777static const struct address_space_operations btree_aops = {
d98237b3
CM		 778 .readpage = btree_readpage,
779 .writepage = btree_writepage,
0da5468f 780 .writepages = btree_writepages,
5f39d397
CM		 781 .releasepage = btree_releasepage,
782 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 783 .sync_page = block_sync_page,
784};
785
ca7a79ad
CM		 786int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
787 u64 parent_transid)
090d1875 788{
5f39d397
CM		 789 struct extent_buffer *buf = NULL;
790 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 791 int ret = 0;
090d1875 792
db94535d 793 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 794 if (!buf)
090d1875 795 return 0;
d1310b2e 796 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 797 buf, 0, 0, btree_get_extent, 0);
5f39d397 798 free_extent_buffer(buf);
de428b63 799 return ret;
090d1875
CM		 800}
801
0999df54
CM		 802struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
803 u64 bytenr, u32 blocksize)
804{
805 struct inode *btree_inode = root->fs_info->btree_inode;
806 struct extent_buffer *eb;
807 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
808 bytenr, blocksize, GFP_NOFS);
809 return eb;
810}
811
812struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
813 u64 bytenr, u32 blocksize)
814{
815 struct inode *btree_inode = root->fs_info->btree_inode;
816 struct extent_buffer *eb;
817
818 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
819 bytenr, blocksize, NULL, GFP_NOFS);
820 return eb;
821}
822
823
e02119d5
CM		 824int btrfs_write_tree_block(struct extent_buffer *buf)
825{
8aa38c31
CH		 826 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
827 buf->start + buf->len - 1);
e02119d5
CM		 828}
829
830int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
831{
8aa38c31
CH		 832 return filemap_fdatawait_range(buf->first_page->mapping,
833 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 834}
835
0999df54 836struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 837 u32 blocksize, u64 parent_transid)
0999df54
CM		 838{
839 struct extent_buffer *buf = NULL;
840 struct inode *btree_inode = root->fs_info->btree_inode;
841 struct extent_io_tree *io_tree;
842 int ret;
843
844 io_tree = &BTRFS_I(btree_inode)->io_tree;
845
846 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
847 if (!buf)
848 return NULL;
0999df54 849
ca7a79ad 850 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 851
d397712b 852 if (ret == 0)
b4ce94de 853 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 854 return buf;
ce9adaa5 855
eb60ceac
CM		 856}
857
e089f05c 858int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 859 struct extent_buffer *buf)
ed2ff2cb 860{
5f39d397 861 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 862 if (btrfs_header_generation(buf) ==
925baedd 863 root->fs_info->running_transaction->transid) {
b9447ef8 864 btrfs_assert_tree_locked(buf);
b4ce94de 865
b9473439
CM		 866 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
867 spin_lock(&root->fs_info->delalloc_lock);
868 if (root->fs_info->dirty_metadata_bytes >= buf->len)
869 root->fs_info->dirty_metadata_bytes -= buf->len;
870 else
871 WARN_ON(1);
872 spin_unlock(&root->fs_info->delalloc_lock);
873 }
b4ce94de 874
b9473439
CM		 875 /* ugh, clear_extent_buffer_dirty needs to lock the page */
876 btrfs_set_lock_blocking(buf);
d1310b2e 877 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 878 buf);
925baedd 879 }
5f39d397
CM		 880 return 0;
881}
882
db94535d 883static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 884 u32 stripesize, struct btrfs_root *root,
9f5fae2f 885 struct btrfs_fs_info *fs_info,
e20d96d6 886 u64 objectid)
d97e63b6 887{
cfaa7295 888 root->node = NULL;
a28ec197 889 root->commit_root = NULL;
db94535d
CM		 890 root->sectorsize = sectorsize;
891 root->nodesize = nodesize;
892 root->leafsize = leafsize;
87ee04eb 893 root->stripesize = stripesize;
123abc88 894 root->ref_cows = 0;
0b86a832 895 root->track_dirty = 0;
c71bf099
YZ		 896 root->in_radix = 0;
897 root->clean_orphans = 0;
0b86a832 898
9f5fae2f 899 root->fs_info = fs_info;
0f7d52f4
CM		 900 root->objectid = objectid;
901 root->last_trans = 0;
13a8a7c8 902 root->highest_objectid = 0;
58176a96 903 root->name = NULL;
4313b399 904 root->in_sysfs = 0;
6bef4d31 905 root->inode_tree = RB_ROOT;
f0486c68 906 root->block_rsv = NULL;
0b86a832
CM		 907
908 INIT_LIST_HEAD(&root->dirty_list);
7b128766 909 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 910 INIT_LIST_HEAD(&root->root_list);
925baedd 911 spin_lock_init(&root->node_lock);
bcc63abb 912 spin_lock_init(&root->list_lock);
5d4f98a2 913 spin_lock_init(&root->inode_lock);
f0486c68 914 spin_lock_init(&root->accounting_lock);
a2135011 915 mutex_init(&root->objectid_mutex);
e02119d5 916 mutex_init(&root->log_mutex);
7237f183
YZ		 917 init_waitqueue_head(&root->log_writer_wait);
918 init_waitqueue_head(&root->log_commit_wait[0]);
919 init_waitqueue_head(&root->log_commit_wait[1]);
920 atomic_set(&root->log_commit[0], 0);
921 atomic_set(&root->log_commit[1], 0);
922 atomic_set(&root->log_writers, 0);
923 root->log_batch = 0;
924 root->log_transid = 0;
257c62e1 925 root->last_log_commit = 0;
d0c803c4
CM		 926 extent_io_tree_init(&root->dirty_log_pages,
927 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 928
3768f368
CM		 929 memset(&root->root_key, 0, sizeof(root->root_key));
930 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 931 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 932 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 933 root->defrag_trans_start = fs_info->generation;
58176a96 934 init_completion(&root->kobj_unregister);
6702ed49 935 root->defrag_running = 0;
4d775673 936 root->root_key.objectid = objectid;
3394e160
CM		 937 root->anon_super.s_root = NULL;
938 root->anon_super.s_dev = 0;
939 INIT_LIST_HEAD(&root->anon_super.s_list);
940 INIT_LIST_HEAD(&root->anon_super.s_instances);
941 init_rwsem(&root->anon_super.s_umount);
942
3768f368
CM		 943 return 0;
944}
945
db94535d 946static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 947 struct btrfs_fs_info *fs_info,
948 u64 objectid,
e20d96d6 949 struct btrfs_root *root)
3768f368
CM		 950{
951 int ret;
db94535d 952 u32 blocksize;
84234f3a 953 u64 generation;
3768f368 954
db94535d 955 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 956 tree_root->sectorsize, tree_root->stripesize,
957 root, fs_info, objectid);
3768f368
CM		 958 ret = btrfs_find_last_root(tree_root, objectid,
959 &root->root_item, &root->root_key);
4df27c4d
YZ		 960 if (ret > 0)
961 return -ENOENT;
3768f368
CM		 962 BUG_ON(ret);
963
84234f3a 964 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 965 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
966 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 967 blocksize, generation);
3768f368 968 BUG_ON(!root->node);
4df27c4d 969 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 970 return 0;
971}
972
e02119d5
CM		 973int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
974 struct btrfs_fs_info *fs_info)
975{
976 struct extent_buffer *eb;
d0c803c4
CM		 977 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
978 u64 start = 0;
979 u64 end = 0;
e02119d5
CM		 980 int ret;
981
d0c803c4 982 if (!log_root_tree)
e02119d5
CM		 983 return 0;
984
d397712b 985 while (1) {
d0c803c4 986 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
8cef4e16 987 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
d0c803c4
CM		 988 if (ret)
989 break;
990
8cef4e16
YZ		 991 clear_extent_bits(&log_root_tree->dirty_log_pages, start, end,
992 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
d0c803c4 993 }
e02119d5
CM		 994 eb = fs_info->log_root_tree->node;
995
996 WARN_ON(btrfs_header_level(eb) != 0);
997 WARN_ON(btrfs_header_nritems(eb) != 0);
998
d00aff00
CM		 999 ret = btrfs_free_reserved_extent(fs_info->tree_root,
1000 eb->start, eb->len);
e02119d5
CM		 1001 BUG_ON(ret);
1002
1003 free_extent_buffer(eb);
1004 kfree(fs_info->log_root_tree);
1005 fs_info->log_root_tree = NULL;
1006 return 0;
1007}
1008
7237f183
YZ		 1009static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1010 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1011{
1012 struct btrfs_root *root;
1013 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1014 struct extent_buffer *leaf;
e02119d5
CM		 1015
1016 root = kzalloc(sizeof(*root), GFP_NOFS);
1017 if (!root)
7237f183 1018 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1019
1020 __setup_root(tree_root->nodesize, tree_root->leafsize,
1021 tree_root->sectorsize, tree_root->stripesize,
1022 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1023
1024 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1025 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1026 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1027 /*
1028 * log trees do not get reference counted because they go away
1029 * before a real commit is actually done. They do store pointers
1030 * to file data extents, and those reference counts still get
1031 * updated (along with back refs to the log tree).
1032 */
e02119d5
CM		 1033 root->ref_cows = 0;
1034
5d4f98a2
YZ		 1035 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1036 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1037 if (IS_ERR(leaf)) {
1038 kfree(root);
1039 return ERR_CAST(leaf);
1040 }
e02119d5 1041
5d4f98a2
YZ		 1042 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1043 btrfs_set_header_bytenr(leaf, leaf->start);
1044 btrfs_set_header_generation(leaf, trans->transid);
1045 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1046 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1047 root->node = leaf;
e02119d5
CM		 1048
1049 write_extent_buffer(root->node, root->fs_info->fsid,
1050 (unsigned long)btrfs_header_fsid(root->node),
1051 BTRFS_FSID_SIZE);
1052 btrfs_mark_buffer_dirty(root->node);
1053 btrfs_tree_unlock(root->node);
7237f183
YZ		 1054 return root;
1055}
1056
1057int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1058 struct btrfs_fs_info *fs_info)
1059{
1060 struct btrfs_root *log_root;
1061
1062 log_root = alloc_log_tree(trans, fs_info);
1063 if (IS_ERR(log_root))
1064 return PTR_ERR(log_root);
1065 WARN_ON(fs_info->log_root_tree);
1066 fs_info->log_root_tree = log_root;
1067 return 0;
1068}
1069
1070int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1071 struct btrfs_root *root)
1072{
1073 struct btrfs_root *log_root;
1074 struct btrfs_inode_item *inode_item;
1075
1076 log_root = alloc_log_tree(trans, root->fs_info);
1077 if (IS_ERR(log_root))
1078 return PTR_ERR(log_root);
1079
1080 log_root->last_trans = trans->transid;
1081 log_root->root_key.offset = root->root_key.objectid;
1082
1083 inode_item = &log_root->root_item.inode;
1084 inode_item->generation = cpu_to_le64(1);
1085 inode_item->size = cpu_to_le64(3);
1086 inode_item->nlink = cpu_to_le32(1);
1087 inode_item->nbytes = cpu_to_le64(root->leafsize);
1088 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1089
5d4f98a2 1090 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1091
1092 WARN_ON(root->log_root);
1093 root->log_root = log_root;
1094 root->log_transid = 0;
257c62e1 1095 root->last_log_commit = 0;
e02119d5
CM		 1096 return 0;
1097}
1098
1099struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1100 struct btrfs_key *location)
1101{
1102 struct btrfs_root *root;
1103 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1104 struct btrfs_path *path;
5f39d397 1105 struct extent_buffer *l;
84234f3a 1106 u64 generation;
db94535d 1107 u32 blocksize;
0f7d52f4
CM		 1108 int ret = 0;
1109
5eda7b5e 1110 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1111 if (!root)
0f7d52f4 1112 return ERR_PTR(-ENOMEM);
0f7d52f4 1113 if (location->offset == (u64)-1) {
db94535d 1114 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1115 location->objectid, root);
1116 if (ret) {
0f7d52f4
CM		 1117 kfree(root);
1118 return ERR_PTR(ret);
1119 }
13a8a7c8 1120 goto out;
0f7d52f4
CM		 1121 }
1122
db94535d 1123 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1124 tree_root->sectorsize, tree_root->stripesize,
1125 root, fs_info, location->objectid);
0f7d52f4
CM		 1126
1127 path = btrfs_alloc_path();
1128 BUG_ON(!path);
1129 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1130 if (ret == 0) {
1131 l = path->nodes[0];
1132 read_extent_buffer(l, &root->root_item,
1133 btrfs_item_ptr_offset(l, path->slots[0]),
1134 sizeof(root->root_item));
1135 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1136 }
0f7d52f4
CM		 1137 btrfs_free_path(path);
1138 if (ret) {
13a8a7c8
YZ		 1139 if (ret > 0)
1140 ret = -ENOENT;
0f7d52f4
CM		 1141 return ERR_PTR(ret);
1142 }
13a8a7c8 1143
84234f3a 1144 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1145 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1146 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1147 blocksize, generation);
5d4f98a2 1148 root->commit_root = btrfs_root_node(root);
0f7d52f4 1149 BUG_ON(!root->node);
13a8a7c8
YZ		 1150out:
1151 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1152 root->ref_cows = 1;
13a8a7c8 1153
5eda7b5e
CM		 1154 return root;
1155}
1156
dc17ff8f
CM		 1157struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1158 u64 root_objectid)
1159{
1160 struct btrfs_root *root;
1161
1162 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1163 return fs_info->tree_root;
1164 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1165 return fs_info->extent_root;
1166
1167 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1168 (unsigned long)root_objectid);
1169 return root;
1170}
1171
edbd8d4e
CM		 1172struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1173 struct btrfs_key *location)
5eda7b5e
CM		 1174{
1175 struct btrfs_root *root;
1176 int ret;
1177
edbd8d4e
CM		 1178 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1179 return fs_info->tree_root;
1180 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1181 return fs_info->extent_root;
8f18cf13
CM		 1182 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1183 return fs_info->chunk_root;
1184 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1185 return fs_info->dev_root;
0403e47e
YZ		 1186 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1187 return fs_info->csum_root;
4df27c4d
YZ		 1188again:
1189 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1190 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1191 (unsigned long)location->objectid);
4df27c4d 1192 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1193 if (root)
1194 return root;
1195
4df27c4d
YZ		 1196 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1197 if (ret == 0)
1198 ret = -ENOENT;
1199 if (ret < 0)
1200 return ERR_PTR(ret);
1201
e02119d5 1202 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1203 if (IS_ERR(root))
1204 return root;
3394e160 1205
4df27c4d 1206 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
3394e160
CM		 1207 set_anon_super(&root->anon_super, NULL);
1208
4df27c4d
YZ		 1209 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1210 if (ret)
1211 goto fail;
1212
1213 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1214 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1215 (unsigned long)root->root_key.objectid,
0f7d52f4 1216 root);
c71bf099 1217 if (ret == 0) {
4df27c4d 1218 root->in_radix = 1;
c71bf099
YZ		 1219 root->clean_orphans = 1;
1220 }
4df27c4d
YZ		 1221 spin_unlock(&fs_info->fs_roots_radix_lock);
1222 radix_tree_preload_end();
0f7d52f4 1223 if (ret) {
4df27c4d
YZ		 1224 if (ret == -EEXIST) {
1225 free_fs_root(root);
1226 goto again;
1227 }
1228 goto fail;
0f7d52f4 1229 }
4df27c4d
YZ		 1230
1231 ret = btrfs_find_dead_roots(fs_info->tree_root,
1232 root->root_key.objectid);
1233 WARN_ON(ret);
edbd8d4e 1234 return root;
4df27c4d
YZ		 1235fail:
1236 free_fs_root(root);
1237 return ERR_PTR(ret);
edbd8d4e
CM		 1238}
1239
1240struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1241 struct btrfs_key *location,
1242 const char *name, int namelen)
1243{
4df27c4d
YZ		 1244 return btrfs_read_fs_root_no_name(fs_info, location);
1245#if 0
edbd8d4e
CM		 1246 struct btrfs_root *root;
1247 int ret;
1248
1249 root = btrfs_read_fs_root_no_name(fs_info, location);
1250 if (!root)
1251 return NULL;
58176a96 1252
4313b399
CM		 1253 if (root->in_sysfs)
1254 return root;
1255
58176a96
JB		 1256 ret = btrfs_set_root_name(root, name, namelen);
1257 if (ret) {
5f39d397 1258 free_extent_buffer(root->node);
58176a96
JB		 1259 kfree(root);
1260 return ERR_PTR(ret);
1261 }
4df27c4d 1262
58176a96
JB		 1263 ret = btrfs_sysfs_add_root(root);
1264 if (ret) {
5f39d397 1265 free_extent_buffer(root->node);
58176a96
JB		 1266 kfree(root->name);
1267 kfree(root);
1268 return ERR_PTR(ret);
1269 }
4313b399 1270 root->in_sysfs = 1;
0f7d52f4 1271 return root;
4df27c4d 1272#endif
0f7d52f4 1273}
04160088
CM		 1274
1275static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1276{
1277 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1278 int ret = 0;
04160088
CM		 1279 struct btrfs_device *device;
1280 struct backing_dev_info *bdi;
b7967db7 1281
c6e30871 1282 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1283 if (!device->bdev)
1284 continue;
04160088
CM		 1285 bdi = blk_get_backing_dev_info(device->bdev);
1286 if (bdi && bdi_congested(bdi, bdi_bits)) {
1287 ret = 1;
1288 break;
1289 }
1290 }
1291 return ret;
1292}
1293
38b66988
CM		 1294/*
1295 * this unplugs every device on the box, and it is only used when page
1296 * is null
1297 */
1298static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1299{
38b66988
CM		 1300 struct btrfs_device *device;
1301 struct btrfs_fs_info *info;
1302
1303 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1304 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1305 if (!device->bdev)
1306 continue;
1307
38b66988 1308 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1309 if (bdi->unplug_io_fn)
38b66988 1310 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1311 }
1312}
1313
b2950863 1314static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1315{
38b66988 1316 struct inode *inode;
f2d8d74d
CM		 1317 struct extent_map_tree *em_tree;
1318 struct extent_map *em;
bcbfce8a 1319 struct address_space *mapping;
38b66988
CM		 1320 u64 offset;
1321
bcbfce8a 1322 /* the generic O_DIRECT read code does this */
9f0ba5bd 1323 if (1 || !page) {
38b66988
CM		 1324 __unplug_io_fn(bdi, page);
1325 return;
1326 }
1327
bcbfce8a
CM		 1328 /*
1329 * page->mapping may change at any time. Get a consistent copy
1330 * and use that for everything below
1331 */
1332 smp_mb();
1333 mapping = page->mapping;
1334 if (!mapping)
1335 return;
1336
1337 inode = mapping->host;
240d5d48
CM		 1338
1339 /*
1340 * don't do the expensive searching for a small number of
1341 * devices
1342 */
1343 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1344 __unplug_io_fn(bdi, page);
1345 return;
1346 }
1347
38b66988 1348 offset = page_offset(page);
04160088 1349
f2d8d74d 1350 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1351 read_lock(&em_tree->lock);
f2d8d74d 1352 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1353 read_unlock(&em_tree->lock);
89642229
CM		 1354 if (!em) {
1355 __unplug_io_fn(bdi, page);
f2d8d74d 1356 return;
89642229 1357 }
f2d8d74d 1358
89642229
CM		 1359 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1360 free_extent_map(em);
1361 __unplug_io_fn(bdi, page);
1362 return;
1363 }
f2d8d74d
CM		 1364 offset = offset - em->start;
1365 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1366 em->block_start + offset, page);
1367 free_extent_map(em);
04160088
CM		 1368}
1369
ad081f14
JA		 1370/*
1371 * If this fails, caller must call bdi_destroy() to get rid of the
1372 * bdi again.
1373 */
04160088
CM		 1374static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1375{
ad081f14
JA		 1376 int err;
1377
1378 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1379 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1380 if (err)
1381 return err;
1382
4575c9cc 1383 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1384 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1385 bdi->unplug_io_data = info;
1386 bdi->congested_fn = btrfs_congested_fn;
1387 bdi->congested_data = info;
1388 return 0;
1389}
1390
ce9adaa5
CM		 1391static int bio_ready_for_csum(struct bio *bio)
1392{
1393 u64 length = 0;
1394 u64 buf_len = 0;
1395 u64 start = 0;
1396 struct page *page;
1397 struct extent_io_tree *io_tree = NULL;
1398 struct btrfs_fs_info *info = NULL;
1399 struct bio_vec *bvec;
1400 int i;
1401 int ret;
1402
1403 bio_for_each_segment(bvec, bio, i) {
1404 page = bvec->bv_page;
1405 if (page->private == EXTENT_PAGE_PRIVATE) {
1406 length += bvec->bv_len;
1407 continue;
1408 }
1409 if (!page->private) {
1410 length += bvec->bv_len;
1411 continue;
1412 }
1413 length = bvec->bv_len;
1414 buf_len = page->private >> 2;
1415 start = page_offset(page) + bvec->bv_offset;
1416 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1417 info = BTRFS_I(page->mapping->host)->root->fs_info;
1418 }
1419 /* are we fully contained in this bio? */
1420 if (buf_len <= length)
1421 return 1;
1422
1423 ret = extent_range_uptodate(io_tree, start + length,
1424 start + buf_len - 1);
ce9adaa5
CM		 1425 return ret;
1426}
1427
8b712842
CM		 1428/*
1429 * called by the kthread helper functions to finally call the bio end_io
1430 * functions. This is where read checksum verification actually happens
1431 */
1432static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1433{
ce9adaa5 1434 struct bio *bio;
8b712842
CM		 1435 struct end_io_wq *end_io_wq;
1436 struct btrfs_fs_info *fs_info;
ce9adaa5 1437 int error;
ce9adaa5 1438
8b712842
CM		 1439 end_io_wq = container_of(work, struct end_io_wq, work);
1440 bio = end_io_wq->bio;
1441 fs_info = end_io_wq->info;
ce9adaa5 1442
cad321ad 1443 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1444 * be checksummed at once. This makes sure the entire block is in
1445 * ram and up to date before trying to verify things. For
1446 * blocksize <= pagesize, it is basically a noop
1447 */
cad321ad
CM		 1448 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1449 !bio_ready_for_csum(bio)) {
d20f7043 1450 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1451 &end_io_wq->work);
1452 return;
1453 }
1454 error = end_io_wq->error;
1455 bio->bi_private = end_io_wq->private;
1456 bio->bi_end_io = end_io_wq->end_io;
1457 kfree(end_io_wq);
8b712842 1458 bio_endio(bio, error);
44b8bd7e
CM		 1459}
1460
a74a4b97
CM		 1461static int cleaner_kthread(void *arg)
1462{
1463 struct btrfs_root *root = arg;
1464
1465 do {
a74a4b97 1466 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1467
1468 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1469 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1470 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1471 btrfs_clean_old_snapshots(root);
1472 mutex_unlock(&root->fs_info->cleaner_mutex);
1473 }
a74a4b97
CM		 1474
1475 if (freezing(current)) {
1476 refrigerator();
1477 } else {
a74a4b97 1478 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1479 if (!kthread_should_stop())
1480 schedule();
a74a4b97
CM		 1481 __set_current_state(TASK_RUNNING);
1482 }
1483 } while (!kthread_should_stop());
1484 return 0;
1485}
1486
1487static int transaction_kthread(void *arg)
1488{
1489 struct btrfs_root *root = arg;
1490 struct btrfs_trans_handle *trans;
1491 struct btrfs_transaction *cur;
8929ecfa 1492 u64 transid;
a74a4b97
CM		 1493 unsigned long now;
1494 unsigned long delay;
1495 int ret;
1496
1497 do {
a74a4b97
CM		 1498 delay = HZ * 30;
1499 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1500 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1501
8929ecfa 1502 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1503 cur = root->fs_info->running_transaction;
1504 if (!cur) {
8929ecfa 1505 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1506 goto sleep;
1507 }
31153d81 1508
a74a4b97 1509 now = get_seconds();
8929ecfa
YZ		 1510 if (!cur->blocked &&
1511 (now < cur->start_time || now - cur->start_time < 30)) {
1512 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1513 delay = HZ * 5;
1514 goto sleep;
1515 }
8929ecfa
YZ		 1516 transid = cur->transid;
1517 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1518
8929ecfa
YZ		 1519 trans = btrfs_join_transaction(root, 1);
1520 if (transid == trans->transid) {
1521 ret = btrfs_commit_transaction(trans, root);
1522 BUG_ON(ret);
1523 } else {
1524 btrfs_end_transaction(trans, root);
1525 }
a74a4b97
CM		 1526sleep:
1527 wake_up_process(root->fs_info->cleaner_kthread);
1528 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1529
1530 if (freezing(current)) {
1531 refrigerator();
1532 } else {
a74a4b97 1533 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1534 if (!kthread_should_stop() &&
1535 !btrfs_transaction_blocked(root->fs_info))
1536 schedule_timeout(delay);
a74a4b97
CM		 1537 __set_current_state(TASK_RUNNING);
1538 }
1539 } while (!kthread_should_stop());
1540 return 0;
1541}
1542
8a4b83cc 1543struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1544 struct btrfs_fs_devices *fs_devices,
1545 char *options)
2e635a27 1546{
db94535d
CM		 1547 u32 sectorsize;
1548 u32 nodesize;
1549 u32 leafsize;
1550 u32 blocksize;
87ee04eb 1551 u32 stripesize;
84234f3a 1552 u64 generation;
f2b636e8 1553 u64 features;
3de4586c 1554 struct btrfs_key location;
a061fc8d 1555 struct buffer_head *bh;
e02119d5 1556 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1557 GFP_NOFS);
d20f7043
CM		 1558 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1559 GFP_NOFS);
e02119d5 1560 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1561 GFP_NOFS);
8790d502 1562 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1563 GFP_NOFS);
e02119d5 1564 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1565 GFP_NOFS);
e02119d5 1566 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1567 GFP_NOFS);
e02119d5
CM		 1568 struct btrfs_root *log_tree_root;
1569
eb60ceac 1570 int ret;
e58ca020 1571 int err = -EINVAL;
4543df7e 1572
2c90e5d6 1573 struct btrfs_super_block *disk_super;
8790d502 1574
0463bb4e 1575 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1576 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1577 err = -ENOMEM;
1578 goto fail;
1579 }
76dda93c
YZ		 1580
1581 ret = init_srcu_struct(&fs_info->subvol_srcu);
1582 if (ret) {
1583 err = ret;
1584 goto fail;
1585 }
1586
1587 ret = setup_bdi(fs_info, &fs_info->bdi);
1588 if (ret) {
1589 err = ret;
1590 goto fail_srcu;
1591 }
1592
1593 fs_info->btree_inode = new_inode(sb);
1594 if (!fs_info->btree_inode) {
1595 err = -ENOMEM;
1596 goto fail_bdi;
1597 }
1598
1599 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1600 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1601 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1602 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1603 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1604 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1605 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1606 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1607 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1608 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1609 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1610 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1611 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1612
58176a96 1613 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1614 fs_info->tree_root = tree_root;
1615 fs_info->extent_root = extent_root;
d20f7043 1616 fs_info->csum_root = csum_root;
0b86a832
CM		 1617 fs_info->chunk_root = chunk_root;
1618 fs_info->dev_root = dev_root;
8a4b83cc 1619 fs_info->fs_devices = fs_devices;
0b86a832 1620 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1621 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1622 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1623 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1624 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1625 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1626 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1627 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1628 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1629 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1630 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1631 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1632 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1633 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1634 fs_info->sb = sb;
6f568d35 1635 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1636 fs_info->metadata_ratio = 0;
c8b97818 1637
b34b086c
CM		 1638 fs_info->thread_pool_size = min_t(unsigned long,
1639 num_online_cpus() + 2, 8);
0afbaf8c 1640
3eaa2885
CM		 1641 INIT_LIST_HEAD(&fs_info->ordered_extents);
1642 spin_lock_init(&fs_info->ordered_extent_lock);
1643
a061fc8d
CM		 1644 sb->s_blocksize = 4096;
1645 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1646 sb->s_bdi = &fs_info->bdi;
a061fc8d 1647
76dda93c
YZ		 1648 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1649 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1650 /*
1651 * we set the i_size on the btree inode to the max possible int.
1652 * the real end of the address space is determined by all of
1653 * the devices in the system
1654 */
1655 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1656 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1657 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1658
5d4f98a2 1659 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1660 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1661 fs_info->btree_inode->i_mapping,
1662 GFP_NOFS);
d1310b2e
CM		 1663 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1664 GFP_NOFS);
1665
1666 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1667
76dda93c
YZ		 1668 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1669 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1670 sizeof(struct btrfs_key));
1671 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1672 insert_inode_hash(fs_info->btree_inode);
76dda93c 1673
0f9dd46c 1674 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1675 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1676
11833d66 1677 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1678 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1679 extent_io_tree_init(&fs_info->freed_extents[1],
1680 fs_info->btree_inode->i_mapping, GFP_NOFS);
1681 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1682 fs_info->do_barriers = 1;
e18e4809 1683
39279cc3 1684
79154b1b 1685 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1686 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1687 mutex_init(&fs_info->tree_log_mutex);
925baedd 1688 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1689 mutex_init(&fs_info->transaction_kthread_mutex);
1690 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1691 mutex_init(&fs_info->volume_mutex);
276e680d 1692 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1693 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1694 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1695
1696 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1697 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1698
e6dcd2dc 1699 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1700 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1701 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1702
0b86a832 1703 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1704 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1705
d98237b3 1706
a512bbf8 1707 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1708 if (!bh)
39279cc3 1709 goto fail_iput;
39279cc3 1710
a061fc8d 1711 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1712 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1713 sizeof(fs_info->super_for_commit));
a061fc8d 1714 brelse(bh);
5f39d397 1715
a061fc8d 1716 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1717
5f39d397 1718 disk_super = &fs_info->super_copy;
0f7d52f4 1719 if (!btrfs_super_root(disk_super))
c6e2bac1 1720 goto fail_iput;
0f7d52f4 1721
2b82032c
YZ		 1722 ret = btrfs_parse_options(tree_root, options);
1723 if (ret) {
1724 err = ret;
c6e2bac1 1725 goto fail_iput;
2b82032c 1726 }
dfe25020 1727
f2b636e8
JB		 1728 features = btrfs_super_incompat_flags(disk_super) &
1729 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1730 if (features) {
1731 printk(KERN_ERR "BTRFS: couldn't mount because of "
1732 "unsupported optional features (%Lx).\n",
21380931 1733 (unsigned long long)features);
f2b636e8 1734 err = -EINVAL;
c6e2bac1 1735 goto fail_iput;
f2b636e8
JB		 1736 }
1737
5d4f98a2
YZ		 1738 features = btrfs_super_incompat_flags(disk_super);
1739 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1740 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1741 btrfs_set_super_incompat_flags(disk_super, features);
1742 }
1743
f2b636e8
JB		 1744 features = btrfs_super_compat_ro_flags(disk_super) &
1745 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1746 if (!(sb->s_flags & MS_RDONLY) && features) {
1747 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1748 "unsupported option features (%Lx).\n",
21380931 1749 (unsigned long long)features);
f2b636e8 1750 err = -EINVAL;
c6e2bac1 1751 goto fail_iput;
f2b636e8 1752 }
61d92c32
CM		 1753
1754 btrfs_init_workers(&fs_info->generic_worker,
1755 "genwork", 1, NULL);
1756
5443be45 1757 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1758 fs_info->thread_pool_size,
1759 &fs_info->generic_worker);
c8b97818 1760
771ed689 1761 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1762 fs_info->thread_pool_size,
1763 &fs_info->generic_worker);
771ed689 1764
5443be45 1765 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1766 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1767 fs_info->thread_pool_size),
1768 &fs_info->generic_worker);
61b49440
CM		 1769
1770 /* a higher idle thresh on the submit workers makes it much more
1771 * likely that bios will be send down in a sane order to the
1772 * devices
1773 */
1774 fs_info->submit_workers.idle_thresh = 64;
53863232 1775
771ed689 1776 fs_info->workers.idle_thresh = 16;
4a69a410 1777 fs_info->workers.ordered = 1;
61b49440 1778
771ed689
CM		 1779 fs_info->delalloc_workers.idle_thresh = 2;
1780 fs_info->delalloc_workers.ordered = 1;
1781
61d92c32
CM		 1782 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1783 &fs_info->generic_worker);
5443be45 1784 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1785 fs_info->thread_pool_size,
1786 &fs_info->generic_worker);
d20f7043 1787 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1788 fs_info->thread_pool_size,
1789 &fs_info->generic_worker);
cad321ad 1790 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1791 "endio-meta-write", fs_info->thread_pool_size,
1792 &fs_info->generic_worker);
5443be45 1793 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1794 fs_info->thread_pool_size,
1795 &fs_info->generic_worker);
61b49440
CM		 1796
1797 /*
1798 * endios are largely parallel and should have a very
1799 * low idle thresh
1800 */
1801 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1802 fs_info->endio_meta_workers.idle_thresh = 4;
1803
9042846b
CM		 1804 fs_info->endio_write_workers.idle_thresh = 2;
1805 fs_info->endio_meta_write_workers.idle_thresh = 2;
1806
4543df7e 1807 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1808 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1809 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1810 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1811 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1812 btrfs_start_workers(&fs_info->endio_workers, 1);
1813 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1814 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1815 btrfs_start_workers(&fs_info->endio_write_workers, 1);
4543df7e 1816
4575c9cc 1817 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1818 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1819 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1820
db94535d
CM		 1821 nodesize = btrfs_super_nodesize(disk_super);
1822 leafsize = btrfs_super_leafsize(disk_super);
1823 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1824 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1825 tree_root->nodesize = nodesize;
1826 tree_root->leafsize = leafsize;
1827 tree_root->sectorsize = sectorsize;
87ee04eb 1828 tree_root->stripesize = stripesize;
a061fc8d
CM		 1829
1830 sb->s_blocksize = sectorsize;
1831 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1832
39279cc3
CM		 1833 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1834 sizeof(disk_super->magic))) {
d397712b 1835 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1836 goto fail_sb_buffer;
1837 }
19c00ddc 1838
925baedd 1839 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1840 ret = btrfs_read_sys_array(tree_root);
925baedd 1841 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1842 if (ret) {
d397712b
CM		 1843 printk(KERN_WARNING "btrfs: failed to read the system "
1844 "array on %s\n", sb->s_id);
5d4f98a2 1845 goto fail_sb_buffer;
84eed90f 1846 }
0b86a832
CM		 1847
1848 blocksize = btrfs_level_size(tree_root,
1849 btrfs_super_chunk_root_level(disk_super));
84234f3a 1850 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1851
1852 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1853 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1854
1855 chunk_root->node = read_tree_block(chunk_root,
1856 btrfs_super_chunk_root(disk_super),
84234f3a 1857 blocksize, generation);
0b86a832 1858 BUG_ON(!chunk_root->node);
83121942
DW		 1859 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1860 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1861 sb->s_id);
1862 goto fail_chunk_root;
1863 }
5d4f98a2
YZ		 1864 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1865 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1866
e17cade2 1867 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1868 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1869 BTRFS_UUID_SIZE);
e17cade2 1870
925baedd 1871 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1872 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1873 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1874 if (ret) {
d397712b
CM		 1875 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1876 sb->s_id);
2b82032c
YZ		 1877 goto fail_chunk_root;
1878 }
0b86a832 1879
dfe25020
CM		 1880 btrfs_close_extra_devices(fs_devices);
1881
db94535d
CM		 1882 blocksize = btrfs_level_size(tree_root,
1883 btrfs_super_root_level(disk_super));
84234f3a 1884 generation = btrfs_super_generation(disk_super);
0b86a832 1885
e20d96d6 1886 tree_root->node = read_tree_block(tree_root,
db94535d 1887 btrfs_super_root(disk_super),
84234f3a 1888 blocksize, generation);
39279cc3 1889 if (!tree_root->node)
2b82032c 1890 goto fail_chunk_root;
83121942
DW		 1891 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1892 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1893 sb->s_id);
1894 goto fail_tree_root;
1895 }
5d4f98a2
YZ		 1896 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1897 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1898
1899 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1900 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1901 if (ret)
39279cc3 1902 goto fail_tree_root;
0b86a832
CM		 1903 extent_root->track_dirty = 1;
1904
1905 ret = find_and_setup_root(tree_root, fs_info,
1906 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1907 if (ret)
1908 goto fail_extent_root;
5d4f98a2 1909 dev_root->track_dirty = 1;
3768f368 1910
d20f7043
CM		 1911 ret = find_and_setup_root(tree_root, fs_info,
1912 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1913 if (ret)
5d4f98a2 1914 goto fail_dev_root;
d20f7043
CM		 1915
1916 csum_root->track_dirty = 1;
1917
8929ecfa
YZ		 1918 fs_info->generation = generation;
1919 fs_info->last_trans_committed = generation;
1920 fs_info->data_alloc_profile = (u64)-1;
1921 fs_info->metadata_alloc_profile = (u64)-1;
1922 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1923
1b1d1f66
JB		 1924 ret = btrfs_read_block_groups(extent_root);
1925 if (ret) {
1926 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1927 goto fail_block_groups;
1928 }
9078a3e1 1929
a74a4b97
CM		 1930 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1931 "btrfs-cleaner");
57506d50 1932 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 1933 goto fail_block_groups;
a74a4b97
CM		 1934
1935 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1936 tree_root,
1937 "btrfs-transaction");
57506d50 1938 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1939 goto fail_cleaner;
a74a4b97 1940
c289811c
CM		 1941 if (!btrfs_test_opt(tree_root, SSD) &&
1942 !btrfs_test_opt(tree_root, NOSSD) &&
1943 !fs_info->fs_devices->rotating) {
1944 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1945 "mode\n");
1946 btrfs_set_opt(fs_info->mount_opt, SSD);
1947 }
1948
e02119d5 1949 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1950 u64 bytenr = btrfs_super_log_root(disk_super);
1951
7c2ca468 1952 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1953 printk(KERN_WARNING "Btrfs log replay required "
1954 "on RO media\n");
7c2ca468
CM		 1955 err = -EIO;
1956 goto fail_trans_kthread;
1957 }
e02119d5
CM		 1958 blocksize =
1959 btrfs_level_size(tree_root,
1960 btrfs_super_log_root_level(disk_super));
d18a2c44 1961
e02119d5
CM		 1962 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1963 GFP_NOFS);
1964
1965 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1966 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1967
1968 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1969 blocksize,
1970 generation + 1);
e02119d5
CM		 1971 ret = btrfs_recover_log_trees(log_tree_root);
1972 BUG_ON(ret);
e556ce2c
YZ		 1973
1974 if (sb->s_flags & MS_RDONLY) {
1975 ret = btrfs_commit_super(tree_root);
1976 BUG_ON(ret);
1977 }
e02119d5 1978 }
1a40e23b 1979
76dda93c
YZ		 1980 ret = btrfs_find_orphan_roots(tree_root);
1981 BUG_ON(ret);
1982
7c2ca468 1983 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1984 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 1985 if (ret < 0) {
1986 printk(KERN_WARNING
1987 "btrfs: failed to recover relocation\n");
1988 err = -EINVAL;
1989 goto fail_trans_kthread;
1990 }
7c2ca468 1991 }
1a40e23b 1992
3de4586c
CM		 1993 location.objectid = BTRFS_FS_TREE_OBJECTID;
1994 location.type = BTRFS_ROOT_ITEM_KEY;
1995 location.offset = (u64)-1;
1996
3de4586c
CM		 1997 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1998 if (!fs_info->fs_root)
7c2ca468 1999 goto fail_trans_kthread;
c289811c 2000
e3acc2a6
JB		 2001 if (!(sb->s_flags & MS_RDONLY)) {
2002 down_read(&fs_info->cleanup_work_sem);
2003 btrfs_orphan_cleanup(fs_info->fs_root);
2004 up_read(&fs_info->cleanup_work_sem);
2005 }
2006
0f7d52f4 2007 return tree_root;
39279cc3 2008
7c2ca468
CM		 2009fail_trans_kthread:
2010 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2011fail_cleaner:
a74a4b97 2012 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2013
2014 /*
2015 * make sure we're done with the btree inode before we stop our
2016 * kthreads
2017 */
2018 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2019 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2020
1b1d1f66
JB		 2021fail_block_groups:
2022 btrfs_free_block_groups(fs_info);
d20f7043 2023 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2024 free_extent_buffer(csum_root->commit_root);
2025fail_dev_root:
2026 free_extent_buffer(dev_root->node);
2027 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2028fail_extent_root:
2029 free_extent_buffer(extent_root->node);
5d4f98a2 2030 free_extent_buffer(extent_root->commit_root);
39279cc3 2031fail_tree_root:
5f39d397 2032 free_extent_buffer(tree_root->node);
5d4f98a2 2033 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2034fail_chunk_root:
2035 free_extent_buffer(chunk_root->node);
5d4f98a2 2036 free_extent_buffer(chunk_root->commit_root);
39279cc3 2037fail_sb_buffer:
61d92c32 2038 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2039 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2040 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2041 btrfs_stop_workers(&fs_info->workers);
2042 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2043 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2044 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2045 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2046 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2047fail_iput:
7c2ca468 2048 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2049 iput(fs_info->btree_inode);
7e662854 2050
dfe25020 2051 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2052 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2053fail_bdi:
7e662854 2054 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2055fail_srcu:
2056 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2057fail:
39279cc3
CM		 2058 kfree(extent_root);
2059 kfree(tree_root);
2060 kfree(fs_info);
83afeac4
JM		 2061 kfree(chunk_root);
2062 kfree(dev_root);
d20f7043 2063 kfree(csum_root);
39279cc3 2064 return ERR_PTR(err);
eb60ceac
CM		 2065}
2066
f2984462
CM		 2067static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2068{
2069 char b[BDEVNAME_SIZE];
2070
2071 if (uptodate) {
2072 set_buffer_uptodate(bh);
2073 } else {
2074 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2075 printk(KERN_WARNING "lost page write due to "
2076 "I/O error on %s\n",
2077 bdevname(bh->b_bdev, b));
2078 }
1259ab75
CM		 2079 /* note, we dont' set_buffer_write_io_error because we have
2080 * our own ways of dealing with the IO errors
2081 */
f2984462
CM		 2082 clear_buffer_uptodate(bh);
2083 }
2084 unlock_buffer(bh);
2085 put_bh(bh);
2086}
2087
a512bbf8
YZ		 2088struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2089{
2090 struct buffer_head *bh;
2091 struct buffer_head *latest = NULL;
2092 struct btrfs_super_block *super;
2093 int i;
2094 u64 transid = 0;
2095 u64 bytenr;
2096
2097 /* we would like to check all the supers, but that would make
2098 * a btrfs mount succeed after a mkfs from a different FS.
2099 * So, we need to add a special mount option to scan for
2100 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2101 */
2102 for (i = 0; i < 1; i++) {
2103 bytenr = btrfs_sb_offset(i);
2104 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2105 break;
2106 bh = __bread(bdev, bytenr / 4096, 4096);
2107 if (!bh)
2108 continue;
2109
2110 super = (struct btrfs_super_block *)bh->b_data;
2111 if (btrfs_super_bytenr(super) != bytenr ||
2112 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2113 sizeof(super->magic))) {
2114 brelse(bh);
2115 continue;
2116 }
2117
2118 if (!latest || btrfs_super_generation(super) > transid) {
2119 brelse(latest);
2120 latest = bh;
2121 transid = btrfs_super_generation(super);
2122 } else {
2123 brelse(bh);
2124 }
2125 }
2126 return latest;
2127}
2128
4eedeb75
HH		 2129/*
2130 * this should be called twice, once with wait == 0 and
2131 * once with wait == 1. When wait == 0 is done, all the buffer heads
2132 * we write are pinned.
2133 *
2134 * They are released when wait == 1 is done.
2135 * max_mirrors must be the same for both runs, and it indicates how
2136 * many supers on this one device should be written.
2137 *
2138 * max_mirrors == 0 means to write them all.
2139 */
a512bbf8
YZ		 2140static int write_dev_supers(struct btrfs_device *device,
2141 struct btrfs_super_block *sb,
2142 int do_barriers, int wait, int max_mirrors)
2143{
2144 struct buffer_head *bh;
2145 int i;
2146 int ret;
2147 int errors = 0;
2148 u32 crc;
2149 u64 bytenr;
2150 int last_barrier = 0;
2151
2152 if (max_mirrors == 0)
2153 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2154
2155 /* make sure only the last submit_bh does a barrier */
2156 if (do_barriers) {
2157 for (i = 0; i < max_mirrors; i++) {
2158 bytenr = btrfs_sb_offset(i);
2159 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2160 device->total_bytes)
2161 break;
2162 last_barrier = i;
2163 }
2164 }
2165
2166 for (i = 0; i < max_mirrors; i++) {
2167 bytenr = btrfs_sb_offset(i);
2168 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2169 break;
2170
2171 if (wait) {
2172 bh = __find_get_block(device->bdev, bytenr / 4096,
2173 BTRFS_SUPER_INFO_SIZE);
2174 BUG_ON(!bh);
a512bbf8 2175 wait_on_buffer(bh);
4eedeb75
HH		 2176 if (!buffer_uptodate(bh))
2177 errors++;
2178
2179 /* drop our reference */
2180 brelse(bh);
2181
2182 /* drop the reference from the wait == 0 run */
2183 brelse(bh);
2184 continue;
a512bbf8
YZ		 2185 } else {
2186 btrfs_set_super_bytenr(sb, bytenr);
2187
2188 crc = ~(u32)0;
2189 crc = btrfs_csum_data(NULL, (char *)sb +
2190 BTRFS_CSUM_SIZE, crc,
2191 BTRFS_SUPER_INFO_SIZE -
2192 BTRFS_CSUM_SIZE);
2193 btrfs_csum_final(crc, sb->csum);
2194
4eedeb75
HH		 2195 /*
2196 * one reference for us, and we leave it for the
2197 * caller
2198 */
a512bbf8
YZ		 2199 bh = __getblk(device->bdev, bytenr / 4096,
2200 BTRFS_SUPER_INFO_SIZE);
2201 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2202
4eedeb75 2203 /* one reference for submit_bh */
a512bbf8 2204 get_bh(bh);
4eedeb75
HH		 2205
2206 set_buffer_uptodate(bh);
a512bbf8
YZ		 2207 lock_buffer(bh);
2208 bh->b_end_io = btrfs_end_buffer_write_sync;
2209 }
2210
2211 if (i == last_barrier && do_barriers && device->barriers) {
2212 ret = submit_bh(WRITE_BARRIER, bh);
2213 if (ret == -EOPNOTSUPP) {
2214 printk("btrfs: disabling barriers on dev %s\n",
2215 device->name);
2216 set_buffer_uptodate(bh);
2217 device->barriers = 0;
4eedeb75 2218 /* one reference for submit_bh */
a512bbf8
YZ		 2219 get_bh(bh);
2220 lock_buffer(bh);
ffbd517d 2221 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2222 }
2223 } else {
ffbd517d 2224 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2225 }
2226
4eedeb75 2227 if (ret)
a512bbf8 2228 errors++;
a512bbf8
YZ		 2229 }
2230 return errors < i ? 0 : -1;
2231}
2232
2233int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2234{
e5e9a520 2235 struct list_head *head;
f2984462 2236 struct btrfs_device *dev;
a061fc8d 2237 struct btrfs_super_block *sb;
f2984462 2238 struct btrfs_dev_item *dev_item;
f2984462
CM		 2239 int ret;
2240 int do_barriers;
a236aed1
CM		 2241 int max_errors;
2242 int total_errors = 0;
a061fc8d 2243 u64 flags;
f2984462 2244
a236aed1 2245 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2246 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2247
a061fc8d
CM		 2248 sb = &root->fs_info->super_for_commit;
2249 dev_item = &sb->dev_item;
e5e9a520
CM		 2250
2251 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2252 head = &root->fs_info->fs_devices->devices;
c6e30871 2253 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2254 if (!dev->bdev) {
2255 total_errors++;
2256 continue;
2257 }
2b82032c 2258 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2259 continue;
2260
2b82032c 2261 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2262 btrfs_set_stack_device_type(dev_item, dev->type);
2263 btrfs_set_stack_device_id(dev_item, dev->devid);
2264 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2265 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2266 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2267 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2268 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2269 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2270 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2271
a061fc8d
CM		 2272 flags = btrfs_super_flags(sb);
2273 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2274
a512bbf8 2275 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2276 if (ret)
2277 total_errors++;
f2984462 2278 }
a236aed1 2279 if (total_errors > max_errors) {
d397712b
CM		 2280 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2281 total_errors);
a236aed1
CM		 2282 BUG();
2283 }
f2984462 2284
a512bbf8 2285 total_errors = 0;
c6e30871 2286 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2287 if (!dev->bdev)
2288 continue;
2b82032c 2289 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2290 continue;
2291
a512bbf8
YZ		 2292 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2293 if (ret)
2294 total_errors++;
f2984462 2295 }
e5e9a520 2296 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2297 if (total_errors > max_errors) {
d397712b
CM		 2298 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2299 total_errors);
a236aed1
CM		 2300 BUG();
2301 }
f2984462
CM		 2302 return 0;
2303}
2304
a512bbf8
YZ		 2305int write_ctree_super(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root, int max_mirrors)
eb60ceac 2307{
e66f709b 2308 int ret;
5f39d397 2309
a512bbf8 2310 ret = write_all_supers(root, max_mirrors);
5f39d397 2311 return ret;
cfaa7295
CM		 2312}
2313
5eda7b5e 2314int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2315{
4df27c4d 2316 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2317 radix_tree_delete(&fs_info->fs_roots_radix,
2318 (unsigned long)root->root_key.objectid);
4df27c4d 2319 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2320
2321 if (btrfs_root_refs(&root->root_item) == 0)
2322 synchronize_srcu(&fs_info->subvol_srcu);
2323
4df27c4d
YZ		 2324 free_fs_root(root);
2325 return 0;
2326}
2327
2328static void free_fs_root(struct btrfs_root *root)
2329{
2330 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2331 if (root->anon_super.s_dev) {
2332 down_write(&root->anon_super.s_umount);
2333 kill_anon_super(&root->anon_super);
2334 }
4df27c4d
YZ		 2335 free_extent_buffer(root->node);
2336 free_extent_buffer(root->commit_root);
d397712b 2337 kfree(root->name);
2619ba1f 2338 kfree(root);
2619ba1f
CM		 2339}
2340
35b7e476 2341static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2342{
2343 int ret;
2344 struct btrfs_root *gang[8];
2345 int i;
2346
76dda93c
YZ		 2347 while (!list_empty(&fs_info->dead_roots)) {
2348 gang[0] = list_entry(fs_info->dead_roots.next,
2349 struct btrfs_root, root_list);
2350 list_del(&gang[0]->root_list);
2351
2352 if (gang[0]->in_radix) {
2353 btrfs_free_fs_root(fs_info, gang[0]);
2354 } else {
2355 free_extent_buffer(gang[0]->node);
2356 free_extent_buffer(gang[0]->commit_root);
2357 kfree(gang[0]);
2358 }
2359 }
2360
d397712b 2361 while (1) {
0f7d52f4
CM		 2362 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2363 (void **)gang, 0,
2364 ARRAY_SIZE(gang));
2365 if (!ret)
2366 break;
2619ba1f 2367 for (i = 0; i < ret; i++)
5eda7b5e 2368 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2369 }
2370 return 0;
2371}
b4100d64 2372
c146afad 2373int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2374{
c146afad
YZ		 2375 u64 root_objectid = 0;
2376 struct btrfs_root *gang[8];
2377 int i;
3768f368 2378 int ret;
e089f05c 2379
c146afad
YZ		 2380 while (1) {
2381 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2382 (void **)gang, root_objectid,
2383 ARRAY_SIZE(gang));
2384 if (!ret)
2385 break;
5d4f98a2
YZ		 2386
2387 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2388 for (i = 0; i < ret; i++) {
2389 root_objectid = gang[i]->root_key.objectid;
c146afad
YZ		 2390 btrfs_orphan_cleanup(gang[i]);
2391 }
2392 root_objectid++;
2393 }
2394 return 0;
2395}
a2135011 2396
c146afad
YZ		 2397int btrfs_commit_super(struct btrfs_root *root)
2398{
2399 struct btrfs_trans_handle *trans;
2400 int ret;
a74a4b97 2401
c146afad 2402 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2403 btrfs_run_delayed_iputs(root);
a74a4b97 2404 btrfs_clean_old_snapshots(root);
c146afad 2405 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2406
2407 /* wait until ongoing cleanup work done */
2408 down_write(&root->fs_info->cleanup_work_sem);
2409 up_write(&root->fs_info->cleanup_work_sem);
2410
a22285a6 2411 trans = btrfs_join_transaction(root, 1);
54aa1f4d 2412 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2413 BUG_ON(ret);
2414 /* run commit again to drop the original snapshot */
a22285a6 2415 trans = btrfs_join_transaction(root, 1);
79154b1b
CM		 2416 btrfs_commit_transaction(trans, root);
2417 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2418 BUG_ON(ret);
d6bfde87 2419
a512bbf8 2420 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2421 return ret;
2422}
2423
2424int close_ctree(struct btrfs_root *root)
2425{
2426 struct btrfs_fs_info *fs_info = root->fs_info;
2427 int ret;
2428
2429 fs_info->closing = 1;
2430 smp_mb();
2431
c146afad
YZ		 2432 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2433 ret = btrfs_commit_super(root);
d397712b
CM		 2434 if (ret)
2435 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2436 }
0f7d52f4 2437
8929ecfa
YZ		 2438 kthread_stop(root->fs_info->transaction_kthread);
2439 kthread_stop(root->fs_info->cleaner_kthread);
2440
f25784b3
YZ		 2441 fs_info->closing = 2;
2442 smp_mb();
2443
b0c68f8b 2444 if (fs_info->delalloc_bytes) {
d397712b 2445 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2446 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2447 }
31153d81 2448 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2449 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2450 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2451 }
bcc63abb 2452
5d4f98a2
YZ		 2453 free_extent_buffer(fs_info->extent_root->node);
2454 free_extent_buffer(fs_info->extent_root->commit_root);
2455 free_extent_buffer(fs_info->tree_root->node);
2456 free_extent_buffer(fs_info->tree_root->commit_root);
2457 free_extent_buffer(root->fs_info->chunk_root->node);
2458 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2459 free_extent_buffer(root->fs_info->dev_root->node);
2460 free_extent_buffer(root->fs_info->dev_root->commit_root);
2461 free_extent_buffer(root->fs_info->csum_root->node);
2462 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2463
9078a3e1 2464 btrfs_free_block_groups(root->fs_info);
d10c5f31 2465
c146afad 2466 del_fs_roots(fs_info);
d10c5f31 2467
c146afad 2468 iput(fs_info->btree_inode);
9ad6b7bc 2469
61d92c32 2470 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2471 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2472 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2473 btrfs_stop_workers(&fs_info->workers);
2474 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2475 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2476 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2477 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2478 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2479
dfe25020 2480 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2481 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2482
04160088 2483 bdi_destroy(&fs_info->bdi);
76dda93c 2484 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2485
0f7d52f4 2486 kfree(fs_info->extent_root);
0f7d52f4 2487 kfree(fs_info->tree_root);
0b86a832
CM		 2488 kfree(fs_info->chunk_root);
2489 kfree(fs_info->dev_root);
d20f7043 2490 kfree(fs_info->csum_root);
eb60ceac
CM		 2491 return 0;
2492}
2493
1259ab75 2494int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2495{
1259ab75 2496 int ret;
810191ff 2497 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2498
2ac55d41
JB		 2499 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2500 NULL);
1259ab75
CM		 2501 if (!ret)
2502 return ret;
2503
2504 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2505 parent_transid);
2506 return !ret;
5f39d397
CM		 2507}
2508
2509int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2510{
810191ff 2511 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2512 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2513 buf);
2514}
6702ed49 2515
5f39d397
CM		 2516void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2517{
810191ff 2518 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2519 u64 transid = btrfs_header_generation(buf);
2520 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2521 int was_dirty;
b4ce94de 2522
b9447ef8 2523 btrfs_assert_tree_locked(buf);
ccd467d6 2524 if (transid != root->fs_info->generation) {
d397712b
CM		 2525 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2526 "found %llu running %llu\n",
db94535d 2527 (unsigned long long)buf->start,
d397712b
CM		 2528 (unsigned long long)transid,
2529 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2530 WARN_ON(1);
2531 }
b9473439
CM		 2532 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2533 buf);
2534 if (!was_dirty) {
2535 spin_lock(&root->fs_info->delalloc_lock);
2536 root->fs_info->dirty_metadata_bytes += buf->len;
2537 spin_unlock(&root->fs_info->delalloc_lock);
2538 }
eb60ceac
CM		 2539}
2540
d3c2fdcf 2541void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2542{
188de649
CM		 2543 /*
2544 * looks as though older kernels can get into trouble with
2545 * this code, they end up stuck in balance_dirty_pages forever
2546 */
d6bfde87 2547 u64 num_dirty;
771ed689 2548 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2549
6933c02e 2550 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2551 return;
2552
585ad2c3
CM		 2553 num_dirty = root->fs_info->dirty_metadata_bytes;
2554
d6bfde87
CM		 2555 if (num_dirty > thresh) {
2556 balance_dirty_pages_ratelimited_nr(
d7fc640e 2557 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2558 }
188de649 2559 return;
35b7e476 2560}
6b80053d 2561
ca7a79ad 2562int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2563{
810191ff 2564 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2565 int ret;
ca7a79ad 2566 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2567 if (ret == 0)
b4ce94de 2568 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2569 return ret;
6b80053d 2570}
0da5468f 2571
4bef0848
CM		 2572int btree_lock_page_hook(struct page *page)
2573{
2574 struct inode *inode = page->mapping->host;
b9473439 2575 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2576 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2577 struct extent_buffer *eb;
2578 unsigned long len;
2579 u64 bytenr = page_offset(page);
2580
2581 if (page->private == EXTENT_PAGE_PRIVATE)
2582 goto out;
2583
2584 len = page->private >> 2;
2585 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2586 if (!eb)
2587 goto out;
2588
2589 btrfs_tree_lock(eb);
4bef0848 2590 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2591
2592 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2593 spin_lock(&root->fs_info->delalloc_lock);
2594 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2595 root->fs_info->dirty_metadata_bytes -= eb->len;
2596 else
2597 WARN_ON(1);
2598 spin_unlock(&root->fs_info->delalloc_lock);
2599 }
2600
4bef0848
CM		 2601 btrfs_tree_unlock(eb);
2602 free_extent_buffer(eb);
2603out:
2604 lock_page(page);
2605 return 0;
2606}
2607
d1310b2e 2608static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2609 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2610 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2611 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2612 /* note we're sharing with inode.c for the merge bio hook */
2613 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2614};
Linux 6.x block layer and io_uring tree(s)