projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: kill max_extent mount option
[deliverable/linux.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
4b4e25f2 30#include "compat.h"
eb60ceac
CM		 31#include "ctree.h"
32#include "disk-io.h"
e089f05c 33#include "transaction.h"
0f7d52f4 34#include "btrfs_inode.h"
0b86a832 35#include "volumes.h"
db94535d 36#include "print-tree.h"
8b712842 37#include "async-thread.h"
925baedd 38#include "locking.h"
e02119d5 39#include "tree-log.h"
fa9c0d79 40#include "free-space-cache.h"
eb60ceac 41
d1310b2e 42static struct extent_io_ops btree_extent_io_ops;
8b712842 43static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 44static void free_fs_root(struct btrfs_root *root);
ce9adaa5 45
ad081f14
JA		 46static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
47
d352ac68
CM		 48/*
49 * end_io_wq structs are used to do processing in task context when an IO is
50 * complete. This is used during reads to verify checksums, and it is used
51 * by writes to insert metadata for new file extents after IO is complete.
52 */
ce9adaa5
CM		 53struct end_io_wq {
54 struct bio *bio;
55 bio_end_io_t *end_io;
56 void *private;
57 struct btrfs_fs_info *info;
58 int error;
22c59948 59 int metadata;
ce9adaa5 60 struct list_head list;
8b712842 61 struct btrfs_work work;
ce9adaa5 62};
0da5468f 63
d352ac68
CM		 64/*
65 * async submit bios are used to offload expensive checksumming
66 * onto the worker threads. They checksum file and metadata bios
67 * just before they are sent down the IO stack.
68 */
44b8bd7e
CM		 69struct async_submit_bio {
70 struct inode *inode;
71 struct bio *bio;
72 struct list_head list;
4a69a410
CM		 73 extent_submit_bio_hook_t *submit_bio_start;
74 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 75 int rw;
76 int mirror_num;
c8b97818 77 unsigned long bio_flags;
8b712842 78 struct btrfs_work work;
44b8bd7e
CM		 79};
80
4008c04a
CM		 81/* These are used to set the lockdep class on the extent buffer locks.
82 * The class is set by the readpage_end_io_hook after the buffer has
83 * passed csum validation but before the pages are unlocked.
84 *
85 * The lockdep class is also set by btrfs_init_new_buffer on freshly
86 * allocated blocks.
87 *
88 * The class is based on the level in the tree block, which allows lockdep
89 * to know that lower nodes nest inside the locks of higher nodes.
90 *
91 * We also add a check to make sure the highest level of the tree is
92 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
93 * code needs update as well.
94 */
95#ifdef CONFIG_DEBUG_LOCK_ALLOC
96# if BTRFS_MAX_LEVEL != 8
97# error
98# endif
99static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
100static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
101 /* leaf */
102 "btrfs-extent-00",
103 "btrfs-extent-01",
104 "btrfs-extent-02",
105 "btrfs-extent-03",
106 "btrfs-extent-04",
107 "btrfs-extent-05",
108 "btrfs-extent-06",
109 "btrfs-extent-07",
110 /* highest possible level */
111 "btrfs-extent-08",
112};
113#endif
114
d352ac68
CM		 115/*
116 * extents on the btree inode are pretty simple, there's one extent
117 * that covers the entire device
118 */
b2950863
CH		 119static struct extent_map *btree_get_extent(struct inode *inode,
120 struct page *page, size_t page_offset, u64 start, u64 len,
121 int create)
7eccb903 122{
5f39d397
CM		 123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
124 struct extent_map *em;
125 int ret;
126
890871be 127 read_lock(&em_tree->lock);
d1310b2e 128 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 129 if (em) {
130 em->bdev =
131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 132 read_unlock(&em_tree->lock);
5f39d397 133 goto out;
a061fc8d 134 }
890871be 135 read_unlock(&em_tree->lock);
7b13b7b1 136
5f39d397
CM		 137 em = alloc_extent_map(GFP_NOFS);
138 if (!em) {
139 em = ERR_PTR(-ENOMEM);
140 goto out;
141 }
142 em->start = 0;
0afbaf8c 143 em->len = (u64)-1;
c8b97818 144 em->block_len = (u64)-1;
5f39d397 145 em->block_start = 0;
a061fc8d 146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 147
890871be 148 write_lock(&em_tree->lock);
5f39d397
CM		 149 ret = add_extent_mapping(em_tree, em);
150 if (ret == -EEXIST) {
0afbaf8c
CM		 151 u64 failed_start = em->start;
152 u64 failed_len = em->len;
153
5f39d397 154 free_extent_map(em);
7b13b7b1 155 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 156 if (em) {
7b13b7b1 157 ret = 0;
0afbaf8c
CM		 158 } else {
159 em = lookup_extent_mapping(em_tree, failed_start,
160 failed_len);
7b13b7b1 161 ret = -EIO;
0afbaf8c 162 }
5f39d397 163 } else if (ret) {
7b13b7b1
CM		 164 free_extent_map(em);
165 em = NULL;
5f39d397 166 }
890871be 167 write_unlock(&em_tree->lock);
7b13b7b1
CM		 168
169 if (ret)
170 em = ERR_PTR(ret);
5f39d397
CM		 171out:
172 return em;
7eccb903
CM		 173}
174
19c00ddc
CM		 175u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
176{
163e783e 177 return crc32c(seed, data, len);
19c00ddc
CM		 178}
179
180void btrfs_csum_final(u32 crc, char *result)
181{
182 *(__le32 *)result = ~cpu_to_le32(crc);
183}
184
d352ac68
CM		 185/*
186 * compute the csum for a btree block, and either verify it or write it
187 * into the csum field of the block.
188 */
19c00ddc
CM		 189static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
190 int verify)
191{
607d432d
JB		 192 u16 csum_size =
193 btrfs_super_csum_size(&root->fs_info->super_copy);
194 char *result = NULL;
19c00ddc
CM		 195 unsigned long len;
196 unsigned long cur_len;
197 unsigned long offset = BTRFS_CSUM_SIZE;
198 char *map_token = NULL;
199 char *kaddr;
200 unsigned long map_start;
201 unsigned long map_len;
202 int err;
203 u32 crc = ~(u32)0;
607d432d 204 unsigned long inline_result;
19c00ddc
CM		 205
206 len = buf->len - offset;
d397712b 207 while (len > 0) {
19c00ddc
CM		 208 err = map_private_extent_buffer(buf, offset, 32,
209 &map_token, &kaddr,
210 &map_start, &map_len, KM_USER0);
d397712b 211 if (err)
19c00ddc 212 return 1;
19c00ddc
CM		 213 cur_len = min(len, map_len - (offset - map_start));
214 crc = btrfs_csum_data(root, kaddr + offset - map_start,
215 crc, cur_len);
216 len -= cur_len;
217 offset += cur_len;
218 unmap_extent_buffer(buf, map_token, KM_USER0);
219 }
607d432d
JB		 220 if (csum_size > sizeof(inline_result)) {
221 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
222 if (!result)
223 return 1;
224 } else {
225 result = (char *)&inline_result;
226 }
227
19c00ddc
CM		 228 btrfs_csum_final(crc, result);
229
230 if (verify) {
607d432d 231 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 232 u32 val;
233 u32 found = 0;
607d432d 234 memcpy(&found, result, csum_size);
e4204ded 235
607d432d 236 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 237 if (printk_ratelimit()) {
238 printk(KERN_INFO "btrfs: %s checksum verify "
239 "failed on %llu wanted %X found %X "
240 "level %d\n",
241 root->fs_info->sb->s_id,
242 (unsigned long long)buf->start, val, found,
243 btrfs_header_level(buf));
244 }
607d432d
JB		 245 if (result != (char *)&inline_result)
246 kfree(result);
19c00ddc
CM		 247 return 1;
248 }
249 } else {
607d432d 250 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 251 }
607d432d
JB		 252 if (result != (char *)&inline_result)
253 kfree(result);
19c00ddc
CM		 254 return 0;
255}
256
d352ac68
CM		 257/*
258 * we can't consider a given block up to date unless the transid of the
259 * block matches the transid in the parent node's pointer. This is how we
260 * detect blocks that either didn't get written at all or got written
261 * in the wrong place.
262 */
1259ab75
CM		 263static int verify_parent_transid(struct extent_io_tree *io_tree,
264 struct extent_buffer *eb, u64 parent_transid)
265{
2ac55d41 266 struct extent_state *cached_state = NULL;
1259ab75
CM		 267 int ret;
268
269 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
270 return 0;
271
2ac55d41
JB		 272 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
273 0, &cached_state, GFP_NOFS);
274 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 275 btrfs_header_generation(eb) == parent_transid) {
276 ret = 0;
277 goto out;
278 }
193f284d
CM		 279 if (printk_ratelimit()) {
280 printk("parent transid verify failed on %llu wanted %llu "
281 "found %llu\n",
282 (unsigned long long)eb->start,
283 (unsigned long long)parent_transid,
284 (unsigned long long)btrfs_header_generation(eb));
285 }
1259ab75 286 ret = 1;
2ac55d41 287 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 288out:
2ac55d41
JB		 289 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
290 &cached_state, GFP_NOFS);
1259ab75 291 return ret;
1259ab75
CM		 292}
293
d352ac68
CM		 294/*
295 * helper to read a given tree block, doing retries as required when
296 * the checksums don't match and we have alternate mirrors to try.
297 */
f188591e
CM		 298static int btree_read_extent_buffer_pages(struct btrfs_root *root,
299 struct extent_buffer *eb,
ca7a79ad 300 u64 start, u64 parent_transid)
f188591e
CM		 301{
302 struct extent_io_tree *io_tree;
303 int ret;
304 int num_copies = 0;
305 int mirror_num = 0;
306
307 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
308 while (1) {
309 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
310 btree_get_extent, mirror_num);
1259ab75
CM		 311 if (!ret &&
312 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 313 return ret;
d397712b 314
f188591e
CM		 315 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
316 eb->start, eb->len);
4235298e 317 if (num_copies == 1)
f188591e 318 return ret;
4235298e 319
f188591e 320 mirror_num++;
4235298e 321 if (mirror_num > num_copies)
f188591e 322 return ret;
f188591e 323 }
f188591e
CM		 324 return -EIO;
325}
19c00ddc 326
d352ac68 327/*
d397712b
CM		 328 * checksum a dirty tree block before IO. This has extra checks to make sure
329 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 330 */
d397712b 331
b2950863 332static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 333{
d1310b2e 334 struct extent_io_tree *tree;
35ebb934 335 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 336 u64 found_start;
337 int found_level;
338 unsigned long len;
339 struct extent_buffer *eb;
f188591e
CM		 340 int ret;
341
d1310b2e 342 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 343
344 if (page->private == EXTENT_PAGE_PRIVATE)
345 goto out;
346 if (!page->private)
347 goto out;
348 len = page->private >> 2;
d397712b
CM		 349 WARN_ON(len == 0);
350
19c00ddc 351 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 352 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
353 btrfs_header_generation(eb));
f188591e 354 BUG_ON(ret);
19c00ddc
CM		 355 found_start = btrfs_header_bytenr(eb);
356 if (found_start != start) {
55c69072
CM		 357 WARN_ON(1);
358 goto err;
359 }
360 if (eb->first_page != page) {
55c69072
CM		 361 WARN_ON(1);
362 goto err;
363 }
364 if (!PageUptodate(page)) {
55c69072
CM		 365 WARN_ON(1);
366 goto err;
19c00ddc
CM		 367 }
368 found_level = btrfs_header_level(eb);
4bef0848 369
19c00ddc 370 csum_tree_block(root, eb, 0);
55c69072 371err:
19c00ddc
CM		 372 free_extent_buffer(eb);
373out:
374 return 0;
375}
376
2b82032c
YZ		 377static int check_tree_block_fsid(struct btrfs_root *root,
378 struct extent_buffer *eb)
379{
380 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
381 u8 fsid[BTRFS_UUID_SIZE];
382 int ret = 1;
383
384 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
385 BTRFS_FSID_SIZE);
386 while (fs_devices) {
387 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
388 ret = 0;
389 break;
390 }
391 fs_devices = fs_devices->seed;
392 }
393 return ret;
394}
395
4008c04a
CM		 396#ifdef CONFIG_DEBUG_LOCK_ALLOC
397void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
398{
399 lockdep_set_class_and_name(&eb->lock,
400 &btrfs_eb_class[level],
401 btrfs_eb_name[level]);
402}
403#endif
404
b2950863 405static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 406 struct extent_state *state)
407{
408 struct extent_io_tree *tree;
409 u64 found_start;
410 int found_level;
411 unsigned long len;
412 struct extent_buffer *eb;
413 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 414 int ret = 0;
ce9adaa5
CM		 415
416 tree = &BTRFS_I(page->mapping->host)->io_tree;
417 if (page->private == EXTENT_PAGE_PRIVATE)
418 goto out;
419 if (!page->private)
420 goto out;
d397712b 421
ce9adaa5 422 len = page->private >> 2;
d397712b
CM		 423 WARN_ON(len == 0);
424
ce9adaa5 425 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 426
ce9adaa5 427 found_start = btrfs_header_bytenr(eb);
23a07867 428 if (found_start != start) {
193f284d
CM		 429 if (printk_ratelimit()) {
430 printk(KERN_INFO "btrfs bad tree block start "
431 "%llu %llu\n",
432 (unsigned long long)found_start,
433 (unsigned long long)eb->start);
434 }
f188591e 435 ret = -EIO;
ce9adaa5
CM		 436 goto err;
437 }
438 if (eb->first_page != page) {
d397712b
CM		 439 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
440 eb->first_page->index, page->index);
ce9adaa5 441 WARN_ON(1);
f188591e 442 ret = -EIO;
ce9adaa5
CM		 443 goto err;
444 }
2b82032c 445 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 446 if (printk_ratelimit()) {
447 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
448 (unsigned long long)eb->start);
449 }
1259ab75
CM		 450 ret = -EIO;
451 goto err;
452 }
ce9adaa5
CM		 453 found_level = btrfs_header_level(eb);
454
4008c04a
CM		 455 btrfs_set_buffer_lockdep_class(eb, found_level);
456
ce9adaa5 457 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 458 if (ret)
459 ret = -EIO;
ce9adaa5
CM		 460
461 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
462 end = eb->start + end - 1;
ce9adaa5
CM		 463err:
464 free_extent_buffer(eb);
465out:
f188591e 466 return ret;
ce9adaa5
CM		 467}
468
ce9adaa5 469static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 470{
471 struct end_io_wq *end_io_wq = bio->bi_private;
472 struct btrfs_fs_info *fs_info;
ce9adaa5 473
ce9adaa5 474 fs_info = end_io_wq->info;
ce9adaa5 475 end_io_wq->error = err;
8b712842
CM		 476 end_io_wq->work.func = end_workqueue_fn;
477 end_io_wq->work.flags = 0;
d20f7043
CM		 478
479 if (bio->bi_rw & (1 << BIO_RW)) {
cad321ad
CM		 480 if (end_io_wq->metadata)
481 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
482 &end_io_wq->work);
483 else
484 btrfs_queue_worker(&fs_info->endio_write_workers,
485 &end_io_wq->work);
d20f7043
CM		 486 } else {
487 if (end_io_wq->metadata)
488 btrfs_queue_worker(&fs_info->endio_meta_workers,
489 &end_io_wq->work);
490 else
491 btrfs_queue_worker(&fs_info->endio_workers,
492 &end_io_wq->work);
493 }
ce9adaa5
CM		 494}
495
22c59948
CM		 496int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
497 int metadata)
0b86a832 498{
ce9adaa5 499 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 500 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
501 if (!end_io_wq)
502 return -ENOMEM;
503
504 end_io_wq->private = bio->bi_private;
505 end_io_wq->end_io = bio->bi_end_io;
22c59948 506 end_io_wq->info = info;
ce9adaa5
CM		 507 end_io_wq->error = 0;
508 end_io_wq->bio = bio;
22c59948 509 end_io_wq->metadata = metadata;
ce9adaa5
CM		 510
511 bio->bi_private = end_io_wq;
512 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 513 return 0;
514}
515
b64a2851 516unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 517{
4854ddd0
CM		 518 unsigned long limit = min_t(unsigned long,
519 info->workers.max_workers,
520 info->fs_devices->open_devices);
521 return 256 * limit;
522}
0986fe9e 523
4854ddd0
CM		 524int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
525{
b64a2851
CM		 526 return atomic_read(&info->nr_async_bios) >
527 btrfs_async_submit_limit(info);
0986fe9e
CM		 528}
529
4a69a410
CM		 530static void run_one_async_start(struct btrfs_work *work)
531{
532 struct btrfs_fs_info *fs_info;
533 struct async_submit_bio *async;
534
535 async = container_of(work, struct async_submit_bio, work);
536 fs_info = BTRFS_I(async->inode)->root->fs_info;
537 async->submit_bio_start(async->inode, async->rw, async->bio,
538 async->mirror_num, async->bio_flags);
539}
540
541static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 542{
543 struct btrfs_fs_info *fs_info;
544 struct async_submit_bio *async;
4854ddd0 545 int limit;
8b712842
CM		 546
547 async = container_of(work, struct async_submit_bio, work);
548 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 549
b64a2851 550 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 551 limit = limit * 2 / 3;
552
8b712842 553 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 554
b64a2851
CM		 555 if (atomic_read(&fs_info->nr_async_submits) < limit &&
556 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 557 wake_up(&fs_info->async_submit_wait);
558
4a69a410 559 async->submit_bio_done(async->inode, async->rw, async->bio,
c8b97818 560 async->mirror_num, async->bio_flags);
4a69a410
CM		 561}
562
563static void run_one_async_free(struct btrfs_work *work)
564{
565 struct async_submit_bio *async;
566
567 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 568 kfree(async);
569}
570
44b8bd7e
CM		 571int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
572 int rw, struct bio *bio, int mirror_num,
c8b97818 573 unsigned long bio_flags,
4a69a410
CM		 574 extent_submit_bio_hook_t *submit_bio_start,
575 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 576{
577 struct async_submit_bio *async;
578
579 async = kmalloc(sizeof(*async), GFP_NOFS);
580 if (!async)
581 return -ENOMEM;
582
583 async->inode = inode;
584 async->rw = rw;
585 async->bio = bio;
586 async->mirror_num = mirror_num;
4a69a410
CM		 587 async->submit_bio_start = submit_bio_start;
588 async->submit_bio_done = submit_bio_done;
589
590 async->work.func = run_one_async_start;
591 async->work.ordered_func = run_one_async_done;
592 async->work.ordered_free = run_one_async_free;
593
8b712842 594 async->work.flags = 0;
c8b97818 595 async->bio_flags = bio_flags;
8c8bee1d 596
cb03c743 597 atomic_inc(&fs_info->nr_async_submits);
d313d7a3
CM		 598
599 if (rw & (1 << BIO_RW_SYNCIO))
600 btrfs_set_work_high_prio(&async->work);
601
8b712842 602 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 603
d397712b 604 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 605 atomic_read(&fs_info->nr_async_submits)) {
606 wait_event(fs_info->async_submit_wait,
607 (atomic_read(&fs_info->nr_async_submits) == 0));
608 }
609
44b8bd7e
CM		 610 return 0;
611}
612
ce3ed71a
CM		 613static int btree_csum_one_bio(struct bio *bio)
614{
615 struct bio_vec *bvec = bio->bi_io_vec;
616 int bio_index = 0;
617 struct btrfs_root *root;
618
619 WARN_ON(bio->bi_vcnt <= 0);
d397712b 620 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 621 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
622 csum_dirty_buffer(root, bvec->bv_page);
623 bio_index++;
624 bvec++;
625 }
626 return 0;
627}
628
4a69a410
CM		 629static int __btree_submit_bio_start(struct inode *inode, int rw,
630 struct bio *bio, int mirror_num,
631 unsigned long bio_flags)
22c59948 632{
8b712842
CM		 633 /*
634 * when we're called for a write, we're already in the async
5443be45 635 * submission context. Just jump into btrfs_map_bio
8b712842 636 */
4a69a410
CM		 637 btree_csum_one_bio(bio);
638 return 0;
639}
22c59948 640
4a69a410
CM		 641static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
642 int mirror_num, unsigned long bio_flags)
643{
8b712842 644 /*
4a69a410
CM		 645 * when we're called for a write, we're already in the async
646 * submission context. Just jump into btrfs_map_bio
8b712842 647 */
8b712842 648 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 649}
650
44b8bd7e 651static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 652 int mirror_num, unsigned long bio_flags)
44b8bd7e 653{
cad321ad
CM		 654 int ret;
655
656 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
657 bio, 1);
658 BUG_ON(ret);
659
44b8bd7e 660 if (!(rw & (1 << BIO_RW))) {
4a69a410
CM		 661 /*
662 * called for a read, do the setup so that checksum validation
663 * can happen in the async kernel threads
664 */
4a69a410 665 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 666 mirror_num, 0);
44b8bd7e 667 }
d313d7a3 668
cad321ad
CM		 669 /*
670 * kthread helpers are used to submit writes so that checksumming
671 * can happen in parallel across all CPUs
672 */
44b8bd7e 673 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 674 inode, rw, bio, mirror_num, 0,
4a69a410
CM		 675 __btree_submit_bio_start,
676 __btree_submit_bio_done);
44b8bd7e
CM		 677}
678
0da5468f
CM		 679static int btree_writepage(struct page *page, struct writeback_control *wbc)
680{
d1310b2e 681 struct extent_io_tree *tree;
b9473439
CM		 682 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
683 struct extent_buffer *eb;
684 int was_dirty;
685
d1310b2e 686 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 687 if (!(current->flags & PF_MEMALLOC)) {
688 return extent_write_full_page(tree, page,
689 btree_get_extent, wbc);
690 }
5443be45 691
b9473439
CM		 692 redirty_page_for_writepage(wbc, page);
693 eb = btrfs_find_tree_block(root, page_offset(page),
694 PAGE_CACHE_SIZE);
695 WARN_ON(!eb);
696
697 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
698 if (!was_dirty) {
699 spin_lock(&root->fs_info->delalloc_lock);
700 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
701 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 702 }
b9473439
CM		 703 free_extent_buffer(eb);
704
705 unlock_page(page);
706 return 0;
5f39d397 707}
0da5468f
CM		 708
709static int btree_writepages(struct address_space *mapping,
710 struct writeback_control *wbc)
711{
d1310b2e
CM		 712 struct extent_io_tree *tree;
713 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 714 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 715 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 716 u64 num_dirty;
24ab9cd8 717 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 718
719 if (wbc->for_kupdate)
720 return 0;
721
b9473439
CM		 722 /* this is a bit racy, but that's ok */
723 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 724 if (num_dirty < thresh)
793955bc 725 return 0;
793955bc 726 }
0da5468f
CM		 727 return extent_writepages(tree, mapping, btree_get_extent, wbc);
728}
729
b2950863 730static int btree_readpage(struct file *file, struct page *page)
5f39d397 731{
d1310b2e
CM		 732 struct extent_io_tree *tree;
733 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 734 return extent_read_full_page(tree, page, btree_get_extent);
735}
22b0ebda 736
70dec807 737static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 738{
d1310b2e
CM		 739 struct extent_io_tree *tree;
740 struct extent_map_tree *map;
5f39d397 741 int ret;
d98237b3 742
98509cfc 743 if (PageWriteback(page) || PageDirty(page))
d397712b 744 return 0;
98509cfc 745
d1310b2e
CM		 746 tree = &BTRFS_I(page->mapping->host)->io_tree;
747 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 748
7b13b7b1 749 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 750 if (!ret)
6af118ce 751 return 0;
6af118ce
CM		 752
753 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 754 if (ret == 1) {
755 ClearPagePrivate(page);
756 set_page_private(page, 0);
757 page_cache_release(page);
758 }
6af118ce 759
d98237b3
CM		 760 return ret;
761}
762
5f39d397 763static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 764{
d1310b2e
CM		 765 struct extent_io_tree *tree;
766 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 767 extent_invalidatepage(tree, page, offset);
768 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 769 if (PagePrivate(page)) {
d397712b
CM		 770 printk(KERN_WARNING "btrfs warning page private not zero "
771 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 772 ClearPagePrivate(page);
773 set_page_private(page, 0);
774 page_cache_release(page);
775 }
d98237b3
CM		 776}
777
7f09410b 778static const struct address_space_operations btree_aops = {
d98237b3
CM		 779 .readpage = btree_readpage,
780 .writepage = btree_writepage,
0da5468f 781 .writepages = btree_writepages,
5f39d397
CM		 782 .releasepage = btree_releasepage,
783 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 784 .sync_page = block_sync_page,
785};
786
ca7a79ad
CM		 787int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
788 u64 parent_transid)
090d1875 789{
5f39d397
CM		 790 struct extent_buffer *buf = NULL;
791 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 792 int ret = 0;
090d1875 793
db94535d 794 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 795 if (!buf)
090d1875 796 return 0;
d1310b2e 797 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 798 buf, 0, 0, btree_get_extent, 0);
5f39d397 799 free_extent_buffer(buf);
de428b63 800 return ret;
090d1875
CM		 801}
802
0999df54
CM		 803struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
804 u64 bytenr, u32 blocksize)
805{
806 struct inode *btree_inode = root->fs_info->btree_inode;
807 struct extent_buffer *eb;
808 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
809 bytenr, blocksize, GFP_NOFS);
810 return eb;
811}
812
813struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
814 u64 bytenr, u32 blocksize)
815{
816 struct inode *btree_inode = root->fs_info->btree_inode;
817 struct extent_buffer *eb;
818
819 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
820 bytenr, blocksize, NULL, GFP_NOFS);
821 return eb;
822}
823
824
e02119d5
CM		 825int btrfs_write_tree_block(struct extent_buffer *buf)
826{
8aa38c31
CH		 827 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
828 buf->start + buf->len - 1);
e02119d5
CM		 829}
830
831int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
832{
8aa38c31
CH		 833 return filemap_fdatawait_range(buf->first_page->mapping,
834 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 835}
836
0999df54 837struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 838 u32 blocksize, u64 parent_transid)
0999df54
CM		 839{
840 struct extent_buffer *buf = NULL;
841 struct inode *btree_inode = root->fs_info->btree_inode;
842 struct extent_io_tree *io_tree;
843 int ret;
844
845 io_tree = &BTRFS_I(btree_inode)->io_tree;
846
847 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
848 if (!buf)
849 return NULL;
0999df54 850
ca7a79ad 851 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 852
d397712b 853 if (ret == 0)
b4ce94de 854 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 855 return buf;
ce9adaa5 856
eb60ceac
CM		 857}
858
e089f05c 859int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 860 struct extent_buffer *buf)
ed2ff2cb 861{
5f39d397 862 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 863 if (btrfs_header_generation(buf) ==
925baedd 864 root->fs_info->running_transaction->transid) {
b9447ef8 865 btrfs_assert_tree_locked(buf);
b4ce94de 866
b9473439
CM		 867 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
868 spin_lock(&root->fs_info->delalloc_lock);
869 if (root->fs_info->dirty_metadata_bytes >= buf->len)
870 root->fs_info->dirty_metadata_bytes -= buf->len;
871 else
872 WARN_ON(1);
873 spin_unlock(&root->fs_info->delalloc_lock);
874 }
b4ce94de 875
b9473439
CM		 876 /* ugh, clear_extent_buffer_dirty needs to lock the page */
877 btrfs_set_lock_blocking(buf);
d1310b2e 878 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 879 buf);
925baedd 880 }
5f39d397
CM		 881 return 0;
882}
883
db94535d 884static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 885 u32 stripesize, struct btrfs_root *root,
9f5fae2f 886 struct btrfs_fs_info *fs_info,
e20d96d6 887 u64 objectid)
d97e63b6 888{
cfaa7295 889 root->node = NULL;
a28ec197 890 root->commit_root = NULL;
db94535d
CM		 891 root->sectorsize = sectorsize;
892 root->nodesize = nodesize;
893 root->leafsize = leafsize;
87ee04eb 894 root->stripesize = stripesize;
123abc88 895 root->ref_cows = 0;
0b86a832 896 root->track_dirty = 0;
c71bf099
YZ		 897 root->in_radix = 0;
898 root->clean_orphans = 0;
0b86a832 899
9f5fae2f 900 root->fs_info = fs_info;
0f7d52f4
CM		 901 root->objectid = objectid;
902 root->last_trans = 0;
13a8a7c8 903 root->highest_objectid = 0;
58176a96 904 root->name = NULL;
4313b399 905 root->in_sysfs = 0;
6bef4d31 906 root->inode_tree = RB_ROOT;
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);
a2135011 914 mutex_init(&root->objectid_mutex);
e02119d5 915 mutex_init(&root->log_mutex);
7237f183
YZ		 916 init_waitqueue_head(&root->log_writer_wait);
917 init_waitqueue_head(&root->log_commit_wait[0]);
918 init_waitqueue_head(&root->log_commit_wait[1]);
919 atomic_set(&root->log_commit[0], 0);
920 atomic_set(&root->log_commit[1], 0);
921 atomic_set(&root->log_writers, 0);
922 root->log_batch = 0;
923 root->log_transid = 0;
257c62e1 924 root->last_log_commit = 0;
d0c803c4
CM		 925 extent_io_tree_init(&root->dirty_log_pages,
926 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 927
3768f368
CM		 928 memset(&root->root_key, 0, sizeof(root->root_key));
929 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 930 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 931 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 932 root->defrag_trans_start = fs_info->generation;
58176a96 933 init_completion(&root->kobj_unregister);
6702ed49 934 root->defrag_running = 0;
4d775673 935 root->root_key.objectid = objectid;
3394e160
CM		 936 root->anon_super.s_root = NULL;
937 root->anon_super.s_dev = 0;
938 INIT_LIST_HEAD(&root->anon_super.s_list);
939 INIT_LIST_HEAD(&root->anon_super.s_instances);
940 init_rwsem(&root->anon_super.s_umount);
941
3768f368
CM		 942 return 0;
943}
944
db94535d 945static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 946 struct btrfs_fs_info *fs_info,
947 u64 objectid,
e20d96d6 948 struct btrfs_root *root)
3768f368
CM		 949{
950 int ret;
db94535d 951 u32 blocksize;
84234f3a 952 u64 generation;
3768f368 953
db94535d 954 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 955 tree_root->sectorsize, tree_root->stripesize,
956 root, fs_info, objectid);
3768f368
CM		 957 ret = btrfs_find_last_root(tree_root, objectid,
958 &root->root_item, &root->root_key);
4df27c4d
YZ		 959 if (ret > 0)
960 return -ENOENT;
3768f368
CM		 961 BUG_ON(ret);
962
84234f3a 963 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 964 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
965 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 966 blocksize, generation);
3768f368 967 BUG_ON(!root->node);
4df27c4d 968 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 969 return 0;
970}
971
e02119d5
CM		 972int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
973 struct btrfs_fs_info *fs_info)
974{
975 struct extent_buffer *eb;
d0c803c4
CM		 976 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
977 u64 start = 0;
978 u64 end = 0;
e02119d5
CM		 979 int ret;
980
d0c803c4 981 if (!log_root_tree)
e02119d5
CM		 982 return 0;
983
d397712b 984 while (1) {
d0c803c4 985 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
8cef4e16 986 0, &start, &end, EXTENT_DIRTY | EXTENT_NEW);
d0c803c4
CM		 987 if (ret)
988 break;
989
8cef4e16
YZ		 990 clear_extent_bits(&log_root_tree->dirty_log_pages, start, end,
991 EXTENT_DIRTY | EXTENT_NEW, GFP_NOFS);
d0c803c4 992 }
e02119d5
CM		 993 eb = fs_info->log_root_tree->node;
994
995 WARN_ON(btrfs_header_level(eb) != 0);
996 WARN_ON(btrfs_header_nritems(eb) != 0);
997
d00aff00
CM		 998 ret = btrfs_free_reserved_extent(fs_info->tree_root,
999 eb->start, eb->len);
e02119d5
CM		 1000 BUG_ON(ret);
1001
1002 free_extent_buffer(eb);
1003 kfree(fs_info->log_root_tree);
1004 fs_info->log_root_tree = NULL;
1005 return 0;
1006}
1007
7237f183
YZ		 1008static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1009 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1010{
1011 struct btrfs_root *root;
1012 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1013 struct extent_buffer *leaf;
e02119d5
CM		 1014
1015 root = kzalloc(sizeof(*root), GFP_NOFS);
1016 if (!root)
7237f183 1017 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1018
1019 __setup_root(tree_root->nodesize, tree_root->leafsize,
1020 tree_root->sectorsize, tree_root->stripesize,
1021 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1022
1023 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1024 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1025 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1026 /*
1027 * log trees do not get reference counted because they go away
1028 * before a real commit is actually done. They do store pointers
1029 * to file data extents, and those reference counts still get
1030 * updated (along with back refs to the log tree).
1031 */
e02119d5
CM		 1032 root->ref_cows = 0;
1033
5d4f98a2
YZ		 1034 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1035 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1036 if (IS_ERR(leaf)) {
1037 kfree(root);
1038 return ERR_CAST(leaf);
1039 }
e02119d5 1040
5d4f98a2
YZ		 1041 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1042 btrfs_set_header_bytenr(leaf, leaf->start);
1043 btrfs_set_header_generation(leaf, trans->transid);
1044 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1045 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1046 root->node = leaf;
e02119d5
CM		 1047
1048 write_extent_buffer(root->node, root->fs_info->fsid,
1049 (unsigned long)btrfs_header_fsid(root->node),
1050 BTRFS_FSID_SIZE);
1051 btrfs_mark_buffer_dirty(root->node);
1052 btrfs_tree_unlock(root->node);
7237f183
YZ		 1053 return root;
1054}
1055
1056int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1057 struct btrfs_fs_info *fs_info)
1058{
1059 struct btrfs_root *log_root;
1060
1061 log_root = alloc_log_tree(trans, fs_info);
1062 if (IS_ERR(log_root))
1063 return PTR_ERR(log_root);
1064 WARN_ON(fs_info->log_root_tree);
1065 fs_info->log_root_tree = log_root;
1066 return 0;
1067}
1068
1069int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1070 struct btrfs_root *root)
1071{
1072 struct btrfs_root *log_root;
1073 struct btrfs_inode_item *inode_item;
1074
1075 log_root = alloc_log_tree(trans, root->fs_info);
1076 if (IS_ERR(log_root))
1077 return PTR_ERR(log_root);
1078
1079 log_root->last_trans = trans->transid;
1080 log_root->root_key.offset = root->root_key.objectid;
1081
1082 inode_item = &log_root->root_item.inode;
1083 inode_item->generation = cpu_to_le64(1);
1084 inode_item->size = cpu_to_le64(3);
1085 inode_item->nlink = cpu_to_le32(1);
1086 inode_item->nbytes = cpu_to_le64(root->leafsize);
1087 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1088
5d4f98a2 1089 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1090
1091 WARN_ON(root->log_root);
1092 root->log_root = log_root;
1093 root->log_transid = 0;
257c62e1 1094 root->last_log_commit = 0;
e02119d5
CM		 1095 return 0;
1096}
1097
1098struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1099 struct btrfs_key *location)
1100{
1101 struct btrfs_root *root;
1102 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1103 struct btrfs_path *path;
5f39d397 1104 struct extent_buffer *l;
84234f3a 1105 u64 generation;
db94535d 1106 u32 blocksize;
0f7d52f4
CM		 1107 int ret = 0;
1108
5eda7b5e 1109 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1110 if (!root)
0f7d52f4 1111 return ERR_PTR(-ENOMEM);
0f7d52f4 1112 if (location->offset == (u64)-1) {
db94535d 1113 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1114 location->objectid, root);
1115 if (ret) {
0f7d52f4
CM		 1116 kfree(root);
1117 return ERR_PTR(ret);
1118 }
13a8a7c8 1119 goto out;
0f7d52f4
CM		 1120 }
1121
db94535d 1122 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1123 tree_root->sectorsize, tree_root->stripesize,
1124 root, fs_info, location->objectid);
0f7d52f4
CM		 1125
1126 path = btrfs_alloc_path();
1127 BUG_ON(!path);
1128 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1129 if (ret == 0) {
1130 l = path->nodes[0];
1131 read_extent_buffer(l, &root->root_item,
1132 btrfs_item_ptr_offset(l, path->slots[0]),
1133 sizeof(root->root_item));
1134 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1135 }
0f7d52f4
CM		 1136 btrfs_free_path(path);
1137 if (ret) {
13a8a7c8
YZ		 1138 if (ret > 0)
1139 ret = -ENOENT;
0f7d52f4
CM		 1140 return ERR_PTR(ret);
1141 }
13a8a7c8 1142
84234f3a 1143 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1144 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1145 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1146 blocksize, generation);
5d4f98a2 1147 root->commit_root = btrfs_root_node(root);
0f7d52f4 1148 BUG_ON(!root->node);
13a8a7c8
YZ		 1149out:
1150 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1151 root->ref_cows = 1;
13a8a7c8 1152
5eda7b5e
CM		 1153 return root;
1154}
1155
dc17ff8f
CM		 1156struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1157 u64 root_objectid)
1158{
1159 struct btrfs_root *root;
1160
1161 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1162 return fs_info->tree_root;
1163 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1164 return fs_info->extent_root;
1165
1166 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1167 (unsigned long)root_objectid);
1168 return root;
1169}
1170
edbd8d4e
CM		 1171struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1172 struct btrfs_key *location)
5eda7b5e
CM		 1173{
1174 struct btrfs_root *root;
1175 int ret;
1176
edbd8d4e
CM		 1177 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1178 return fs_info->tree_root;
1179 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1180 return fs_info->extent_root;
8f18cf13
CM		 1181 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1182 return fs_info->chunk_root;
1183 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1184 return fs_info->dev_root;
0403e47e
YZ		 1185 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1186 return fs_info->csum_root;
4df27c4d
YZ		 1187again:
1188 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1189 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1190 (unsigned long)location->objectid);
4df27c4d 1191 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1192 if (root)
1193 return root;
1194
4df27c4d
YZ		 1195 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1196 if (ret == 0)
1197 ret = -ENOENT;
1198 if (ret < 0)
1199 return ERR_PTR(ret);
1200
e02119d5 1201 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1202 if (IS_ERR(root))
1203 return root;
3394e160 1204
4df27c4d 1205 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
3394e160
CM		 1206 set_anon_super(&root->anon_super, NULL);
1207
4df27c4d
YZ		 1208 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1209 if (ret)
1210 goto fail;
1211
1212 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1213 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1214 (unsigned long)root->root_key.objectid,
0f7d52f4 1215 root);
c71bf099 1216 if (ret == 0) {
4df27c4d 1217 root->in_radix = 1;
c71bf099
YZ		 1218 root->clean_orphans = 1;
1219 }
4df27c4d
YZ		 1220 spin_unlock(&fs_info->fs_roots_radix_lock);
1221 radix_tree_preload_end();
0f7d52f4 1222 if (ret) {
4df27c4d
YZ		 1223 if (ret == -EEXIST) {
1224 free_fs_root(root);
1225 goto again;
1226 }
1227 goto fail;
0f7d52f4 1228 }
4df27c4d
YZ		 1229
1230 ret = btrfs_find_dead_roots(fs_info->tree_root,
1231 root->root_key.objectid);
1232 WARN_ON(ret);
edbd8d4e 1233 return root;
4df27c4d
YZ		 1234fail:
1235 free_fs_root(root);
1236 return ERR_PTR(ret);
edbd8d4e
CM		 1237}
1238
1239struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1240 struct btrfs_key *location,
1241 const char *name, int namelen)
1242{
4df27c4d
YZ		 1243 return btrfs_read_fs_root_no_name(fs_info, location);
1244#if 0
edbd8d4e
CM		 1245 struct btrfs_root *root;
1246 int ret;
1247
1248 root = btrfs_read_fs_root_no_name(fs_info, location);
1249 if (!root)
1250 return NULL;
58176a96 1251
4313b399
CM		 1252 if (root->in_sysfs)
1253 return root;
1254
58176a96
JB		 1255 ret = btrfs_set_root_name(root, name, namelen);
1256 if (ret) {
5f39d397 1257 free_extent_buffer(root->node);
58176a96
JB		 1258 kfree(root);
1259 return ERR_PTR(ret);
1260 }
4df27c4d 1261
58176a96
JB		 1262 ret = btrfs_sysfs_add_root(root);
1263 if (ret) {
5f39d397 1264 free_extent_buffer(root->node);
58176a96
JB		 1265 kfree(root->name);
1266 kfree(root);
1267 return ERR_PTR(ret);
1268 }
4313b399 1269 root->in_sysfs = 1;
0f7d52f4 1270 return root;
4df27c4d 1271#endif
0f7d52f4 1272}
04160088
CM		 1273
1274static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1275{
1276 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1277 int ret = 0;
04160088
CM		 1278 struct btrfs_device *device;
1279 struct backing_dev_info *bdi;
b7967db7 1280
c6e30871 1281 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1282 if (!device->bdev)
1283 continue;
04160088
CM		 1284 bdi = blk_get_backing_dev_info(device->bdev);
1285 if (bdi && bdi_congested(bdi, bdi_bits)) {
1286 ret = 1;
1287 break;
1288 }
1289 }
1290 return ret;
1291}
1292
38b66988
CM		 1293/*
1294 * this unplugs every device on the box, and it is only used when page
1295 * is null
1296 */
1297static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1298{
38b66988
CM		 1299 struct btrfs_device *device;
1300 struct btrfs_fs_info *info;
1301
1302 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1303 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1304 if (!device->bdev)
1305 continue;
1306
38b66988 1307 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1308 if (bdi->unplug_io_fn)
38b66988 1309 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1310 }
1311}
1312
b2950863 1313static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1314{
38b66988 1315 struct inode *inode;
f2d8d74d
CM		 1316 struct extent_map_tree *em_tree;
1317 struct extent_map *em;
bcbfce8a 1318 struct address_space *mapping;
38b66988
CM		 1319 u64 offset;
1320
bcbfce8a 1321 /* the generic O_DIRECT read code does this */
9f0ba5bd 1322 if (1 || !page) {
38b66988
CM		 1323 __unplug_io_fn(bdi, page);
1324 return;
1325 }
1326
bcbfce8a
CM		 1327 /*
1328 * page->mapping may change at any time. Get a consistent copy
1329 * and use that for everything below
1330 */
1331 smp_mb();
1332 mapping = page->mapping;
1333 if (!mapping)
1334 return;
1335
1336 inode = mapping->host;
240d5d48
CM		 1337
1338 /*
1339 * don't do the expensive searching for a small number of
1340 * devices
1341 */
1342 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1343 __unplug_io_fn(bdi, page);
1344 return;
1345 }
1346
38b66988 1347 offset = page_offset(page);
04160088 1348
f2d8d74d 1349 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1350 read_lock(&em_tree->lock);
f2d8d74d 1351 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1352 read_unlock(&em_tree->lock);
89642229
CM		 1353 if (!em) {
1354 __unplug_io_fn(bdi, page);
f2d8d74d 1355 return;
89642229 1356 }
f2d8d74d 1357
89642229
CM		 1358 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1359 free_extent_map(em);
1360 __unplug_io_fn(bdi, page);
1361 return;
1362 }
f2d8d74d
CM		 1363 offset = offset - em->start;
1364 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1365 em->block_start + offset, page);
1366 free_extent_map(em);
04160088
CM		 1367}
1368
ad081f14
JA		 1369/*
1370 * If this fails, caller must call bdi_destroy() to get rid of the
1371 * bdi again.
1372 */
04160088
CM		 1373static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1374{
ad081f14
JA		 1375 int err;
1376
d993831f 1377 bdi->name = "btrfs";
ad081f14
JA		 1378 bdi->capabilities = BDI_CAP_MAP_COPY;
1379 err = bdi_init(bdi);
1380 if (err)
1381 return err;
1382
1383 err = bdi_register(bdi, NULL, "btrfs-%d",
1384 atomic_inc_return(&btrfs_bdi_num));
76dda93c
YZ		 1385 if (err) {
1386 bdi_destroy(bdi);
ad081f14 1387 return err;
76dda93c 1388 }
ad081f14 1389
4575c9cc 1390 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1391 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1392 bdi->unplug_io_data = info;
1393 bdi->congested_fn = btrfs_congested_fn;
1394 bdi->congested_data = info;
1395 return 0;
1396}
1397
ce9adaa5
CM		 1398static int bio_ready_for_csum(struct bio *bio)
1399{
1400 u64 length = 0;
1401 u64 buf_len = 0;
1402 u64 start = 0;
1403 struct page *page;
1404 struct extent_io_tree *io_tree = NULL;
1405 struct btrfs_fs_info *info = NULL;
1406 struct bio_vec *bvec;
1407 int i;
1408 int ret;
1409
1410 bio_for_each_segment(bvec, bio, i) {
1411 page = bvec->bv_page;
1412 if (page->private == EXTENT_PAGE_PRIVATE) {
1413 length += bvec->bv_len;
1414 continue;
1415 }
1416 if (!page->private) {
1417 length += bvec->bv_len;
1418 continue;
1419 }
1420 length = bvec->bv_len;
1421 buf_len = page->private >> 2;
1422 start = page_offset(page) + bvec->bv_offset;
1423 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1424 info = BTRFS_I(page->mapping->host)->root->fs_info;
1425 }
1426 /* are we fully contained in this bio? */
1427 if (buf_len <= length)
1428 return 1;
1429
1430 ret = extent_range_uptodate(io_tree, start + length,
1431 start + buf_len - 1);
ce9adaa5
CM		 1432 return ret;
1433}
1434
8b712842
CM		 1435/*
1436 * called by the kthread helper functions to finally call the bio end_io
1437 * functions. This is where read checksum verification actually happens
1438 */
1439static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1440{
ce9adaa5 1441 struct bio *bio;
8b712842
CM		 1442 struct end_io_wq *end_io_wq;
1443 struct btrfs_fs_info *fs_info;
ce9adaa5 1444 int error;
ce9adaa5 1445
8b712842
CM		 1446 end_io_wq = container_of(work, struct end_io_wq, work);
1447 bio = end_io_wq->bio;
1448 fs_info = end_io_wq->info;
ce9adaa5 1449
cad321ad 1450 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1451 * be checksummed at once. This makes sure the entire block is in
1452 * ram and up to date before trying to verify things. For
1453 * blocksize <= pagesize, it is basically a noop
1454 */
cad321ad
CM		 1455 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1456 !bio_ready_for_csum(bio)) {
d20f7043 1457 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1458 &end_io_wq->work);
1459 return;
1460 }
1461 error = end_io_wq->error;
1462 bio->bi_private = end_io_wq->private;
1463 bio->bi_end_io = end_io_wq->end_io;
1464 kfree(end_io_wq);
8b712842 1465 bio_endio(bio, error);
44b8bd7e
CM		 1466}
1467
a74a4b97
CM		 1468static int cleaner_kthread(void *arg)
1469{
1470 struct btrfs_root *root = arg;
1471
1472 do {
1473 smp_mb();
1474 if (root->fs_info->closing)
1475 break;
1476
1477 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1478
1479 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1480 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1481 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1482 btrfs_clean_old_snapshots(root);
1483 mutex_unlock(&root->fs_info->cleaner_mutex);
1484 }
a74a4b97
CM		 1485
1486 if (freezing(current)) {
1487 refrigerator();
1488 } else {
1489 smp_mb();
1490 if (root->fs_info->closing)
1491 break;
1492 set_current_state(TASK_INTERRUPTIBLE);
1493 schedule();
1494 __set_current_state(TASK_RUNNING);
1495 }
1496 } while (!kthread_should_stop());
1497 return 0;
1498}
1499
1500static int transaction_kthread(void *arg)
1501{
1502 struct btrfs_root *root = arg;
1503 struct btrfs_trans_handle *trans;
1504 struct btrfs_transaction *cur;
1505 unsigned long now;
1506 unsigned long delay;
1507 int ret;
1508
1509 do {
1510 smp_mb();
1511 if (root->fs_info->closing)
1512 break;
1513
1514 delay = HZ * 30;
1515 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1516 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1517
1518 mutex_lock(&root->fs_info->trans_mutex);
1519 cur = root->fs_info->running_transaction;
1520 if (!cur) {
1521 mutex_unlock(&root->fs_info->trans_mutex);
1522 goto sleep;
1523 }
31153d81 1524
a74a4b97
CM		 1525 now = get_seconds();
1526 if (now < cur->start_time || now - cur->start_time < 30) {
1527 mutex_unlock(&root->fs_info->trans_mutex);
1528 delay = HZ * 5;
1529 goto sleep;
1530 }
1531 mutex_unlock(&root->fs_info->trans_mutex);
a74a4b97
CM		 1532 trans = btrfs_start_transaction(root, 1);
1533 ret = btrfs_commit_transaction(trans, root);
56bec294 1534
a74a4b97
CM		 1535sleep:
1536 wake_up_process(root->fs_info->cleaner_kthread);
1537 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1538
1539 if (freezing(current)) {
1540 refrigerator();
1541 } else {
1542 if (root->fs_info->closing)
1543 break;
1544 set_current_state(TASK_INTERRUPTIBLE);
1545 schedule_timeout(delay);
1546 __set_current_state(TASK_RUNNING);
1547 }
1548 } while (!kthread_should_stop());
1549 return 0;
1550}
1551
8a4b83cc 1552struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1553 struct btrfs_fs_devices *fs_devices,
1554 char *options)
2e635a27 1555{
db94535d
CM		 1556 u32 sectorsize;
1557 u32 nodesize;
1558 u32 leafsize;
1559 u32 blocksize;
87ee04eb 1560 u32 stripesize;
84234f3a 1561 u64 generation;
f2b636e8 1562 u64 features;
3de4586c 1563 struct btrfs_key location;
a061fc8d 1564 struct buffer_head *bh;
e02119d5 1565 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1566 GFP_NOFS);
d20f7043
CM		 1567 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1568 GFP_NOFS);
e02119d5 1569 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1570 GFP_NOFS);
8790d502 1571 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1572 GFP_NOFS);
e02119d5 1573 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1574 GFP_NOFS);
e02119d5 1575 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1576 GFP_NOFS);
e02119d5
CM		 1577 struct btrfs_root *log_tree_root;
1578
eb60ceac 1579 int ret;
e58ca020 1580 int err = -EINVAL;
4543df7e 1581
2c90e5d6 1582 struct btrfs_super_block *disk_super;
8790d502 1583
0463bb4e 1584 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1585 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1586 err = -ENOMEM;
1587 goto fail;
1588 }
76dda93c
YZ		 1589
1590 ret = init_srcu_struct(&fs_info->subvol_srcu);
1591 if (ret) {
1592 err = ret;
1593 goto fail;
1594 }
1595
1596 ret = setup_bdi(fs_info, &fs_info->bdi);
1597 if (ret) {
1598 err = ret;
1599 goto fail_srcu;
1600 }
1601
1602 fs_info->btree_inode = new_inode(sb);
1603 if (!fs_info->btree_inode) {
1604 err = -ENOMEM;
1605 goto fail_bdi;
1606 }
1607
1608 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1609 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1610 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1611 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1612 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1613 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1614 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1615 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1616 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1617 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1618 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1619 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1620 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1621
58176a96 1622 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1623 fs_info->tree_root = tree_root;
1624 fs_info->extent_root = extent_root;
d20f7043 1625 fs_info->csum_root = csum_root;
0b86a832
CM		 1626 fs_info->chunk_root = chunk_root;
1627 fs_info->dev_root = dev_root;
8a4b83cc 1628 fs_info->fs_devices = fs_devices;
0b86a832 1629 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1630 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1631 btrfs_mapping_init(&fs_info->mapping_tree);
cb03c743 1632 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1633 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1634 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1635 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1636 fs_info->sb = sb;
6f568d35 1637 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1638 fs_info->metadata_ratio = 0;
c8b97818 1639
b34b086c
CM		 1640 fs_info->thread_pool_size = min_t(unsigned long,
1641 num_online_cpus() + 2, 8);
0afbaf8c 1642
3eaa2885
CM		 1643 INIT_LIST_HEAD(&fs_info->ordered_extents);
1644 spin_lock_init(&fs_info->ordered_extent_lock);
1645
a061fc8d
CM		 1646 sb->s_blocksize = 4096;
1647 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1648 sb->s_bdi = &fs_info->bdi;
a061fc8d 1649
76dda93c
YZ		 1650 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1651 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1652 /*
1653 * we set the i_size on the btree inode to the max possible int.
1654 * the real end of the address space is determined by all of
1655 * the devices in the system
1656 */
1657 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1658 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1659 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1660
5d4f98a2 1661 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1662 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1663 fs_info->btree_inode->i_mapping,
1664 GFP_NOFS);
d1310b2e
CM		 1665 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1666 GFP_NOFS);
1667
1668 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1669
76dda93c
YZ		 1670 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1671 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1672 sizeof(struct btrfs_key));
1673 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1674 insert_inode_hash(fs_info->btree_inode);
76dda93c 1675
0f9dd46c 1676 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1677 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1678
11833d66 1679 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1680 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1681 extent_io_tree_init(&fs_info->freed_extents[1],
1682 fs_info->btree_inode->i_mapping, GFP_NOFS);
1683 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1684 fs_info->do_barriers = 1;
e18e4809 1685
39279cc3 1686
79154b1b 1687 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1688 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1689 mutex_init(&fs_info->tree_log_mutex);
925baedd 1690 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1691 mutex_init(&fs_info->transaction_kthread_mutex);
1692 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1693 mutex_init(&fs_info->volume_mutex);
276e680d 1694 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1695 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1696 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1697
1698 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1699 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1700
e6dcd2dc 1701 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1702 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1703 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1704
0b86a832 1705 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1706 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1707
d98237b3 1708
a512bbf8 1709 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1710 if (!bh)
39279cc3 1711 goto fail_iput;
39279cc3 1712
a061fc8d 1713 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1714 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1715 sizeof(fs_info->super_for_commit));
a061fc8d 1716 brelse(bh);
5f39d397 1717
a061fc8d 1718 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1719
5f39d397 1720 disk_super = &fs_info->super_copy;
0f7d52f4 1721 if (!btrfs_super_root(disk_super))
c6e2bac1 1722 goto fail_iput;
0f7d52f4 1723
2b82032c
YZ		 1724 ret = btrfs_parse_options(tree_root, options);
1725 if (ret) {
1726 err = ret;
c6e2bac1 1727 goto fail_iput;
2b82032c 1728 }
dfe25020 1729
f2b636e8
JB		 1730 features = btrfs_super_incompat_flags(disk_super) &
1731 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1732 if (features) {
1733 printk(KERN_ERR "BTRFS: couldn't mount because of "
1734 "unsupported optional features (%Lx).\n",
21380931 1735 (unsigned long long)features);
f2b636e8 1736 err = -EINVAL;
c6e2bac1 1737 goto fail_iput;
f2b636e8
JB		 1738 }
1739
5d4f98a2
YZ		 1740 features = btrfs_super_incompat_flags(disk_super);
1741 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1742 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1743 btrfs_set_super_incompat_flags(disk_super, features);
1744 }
1745
f2b636e8
JB		 1746 features = btrfs_super_compat_ro_flags(disk_super) &
1747 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1748 if (!(sb->s_flags & MS_RDONLY) && features) {
1749 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1750 "unsupported option features (%Lx).\n",
21380931 1751 (unsigned long long)features);
f2b636e8 1752 err = -EINVAL;
c6e2bac1 1753 goto fail_iput;
f2b636e8 1754 }
61d92c32
CM		 1755
1756 btrfs_init_workers(&fs_info->generic_worker,
1757 "genwork", 1, NULL);
1758
5443be45 1759 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1760 fs_info->thread_pool_size,
1761 &fs_info->generic_worker);
c8b97818 1762
771ed689 1763 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1764 fs_info->thread_pool_size,
1765 &fs_info->generic_worker);
771ed689 1766
5443be45 1767 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1768 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1769 fs_info->thread_pool_size),
1770 &fs_info->generic_worker);
e3ccfa98
JB		 1771 btrfs_init_workers(&fs_info->enospc_workers, "enospc",
1772 fs_info->thread_pool_size,
1773 &fs_info->generic_worker);
61b49440
CM		 1774
1775 /* a higher idle thresh on the submit workers makes it much more
1776 * likely that bios will be send down in a sane order to the
1777 * devices
1778 */
1779 fs_info->submit_workers.idle_thresh = 64;
53863232 1780
771ed689 1781 fs_info->workers.idle_thresh = 16;
4a69a410 1782 fs_info->workers.ordered = 1;
61b49440 1783
771ed689
CM		 1784 fs_info->delalloc_workers.idle_thresh = 2;
1785 fs_info->delalloc_workers.ordered = 1;
1786
61d92c32
CM		 1787 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1788 &fs_info->generic_worker);
5443be45 1789 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1790 fs_info->thread_pool_size,
1791 &fs_info->generic_worker);
d20f7043 1792 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1793 fs_info->thread_pool_size,
1794 &fs_info->generic_worker);
cad321ad 1795 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1796 "endio-meta-write", fs_info->thread_pool_size,
1797 &fs_info->generic_worker);
5443be45 1798 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1799 fs_info->thread_pool_size,
1800 &fs_info->generic_worker);
61b49440
CM		 1801
1802 /*
1803 * endios are largely parallel and should have a very
1804 * low idle thresh
1805 */
1806 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1807 fs_info->endio_meta_workers.idle_thresh = 4;
1808
9042846b
CM		 1809 fs_info->endio_write_workers.idle_thresh = 2;
1810 fs_info->endio_meta_write_workers.idle_thresh = 2;
1811
4543df7e 1812 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1813 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1814 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1815 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1816 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1817 btrfs_start_workers(&fs_info->endio_workers, 1);
1818 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1819 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1820 btrfs_start_workers(&fs_info->endio_write_workers, 1);
e3ccfa98 1821 btrfs_start_workers(&fs_info->enospc_workers, 1);
4543df7e 1822
4575c9cc 1823 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1824 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1825 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1826
db94535d
CM		 1827 nodesize = btrfs_super_nodesize(disk_super);
1828 leafsize = btrfs_super_leafsize(disk_super);
1829 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1830 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1831 tree_root->nodesize = nodesize;
1832 tree_root->leafsize = leafsize;
1833 tree_root->sectorsize = sectorsize;
87ee04eb 1834 tree_root->stripesize = stripesize;
a061fc8d
CM		 1835
1836 sb->s_blocksize = sectorsize;
1837 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1838
39279cc3
CM		 1839 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1840 sizeof(disk_super->magic))) {
d397712b 1841 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1842 goto fail_sb_buffer;
1843 }
19c00ddc 1844
925baedd 1845 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1846 ret = btrfs_read_sys_array(tree_root);
925baedd 1847 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1848 if (ret) {
d397712b
CM		 1849 printk(KERN_WARNING "btrfs: failed to read the system "
1850 "array on %s\n", sb->s_id);
5d4f98a2 1851 goto fail_sb_buffer;
84eed90f 1852 }
0b86a832
CM		 1853
1854 blocksize = btrfs_level_size(tree_root,
1855 btrfs_super_chunk_root_level(disk_super));
84234f3a 1856 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1857
1858 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1859 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1860
1861 chunk_root->node = read_tree_block(chunk_root,
1862 btrfs_super_chunk_root(disk_super),
84234f3a 1863 blocksize, generation);
0b86a832 1864 BUG_ON(!chunk_root->node);
83121942
DW		 1865 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1866 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1867 sb->s_id);
1868 goto fail_chunk_root;
1869 }
5d4f98a2
YZ		 1870 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1871 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1872
e17cade2 1873 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1874 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1875 BTRFS_UUID_SIZE);
e17cade2 1876
925baedd 1877 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1878 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1879 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1880 if (ret) {
d397712b
CM		 1881 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1882 sb->s_id);
2b82032c
YZ		 1883 goto fail_chunk_root;
1884 }
0b86a832 1885
dfe25020
CM		 1886 btrfs_close_extra_devices(fs_devices);
1887
db94535d
CM		 1888 blocksize = btrfs_level_size(tree_root,
1889 btrfs_super_root_level(disk_super));
84234f3a 1890 generation = btrfs_super_generation(disk_super);
0b86a832 1891
e20d96d6 1892 tree_root->node = read_tree_block(tree_root,
db94535d 1893 btrfs_super_root(disk_super),
84234f3a 1894 blocksize, generation);
39279cc3 1895 if (!tree_root->node)
2b82032c 1896 goto fail_chunk_root;
83121942
DW		 1897 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1898 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1899 sb->s_id);
1900 goto fail_tree_root;
1901 }
5d4f98a2
YZ		 1902 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1903 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1904
1905 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1906 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1907 if (ret)
39279cc3 1908 goto fail_tree_root;
0b86a832
CM		 1909 extent_root->track_dirty = 1;
1910
1911 ret = find_and_setup_root(tree_root, fs_info,
1912 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1913 if (ret)
1914 goto fail_extent_root;
5d4f98a2 1915 dev_root->track_dirty = 1;
3768f368 1916
d20f7043
CM		 1917 ret = find_and_setup_root(tree_root, fs_info,
1918 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1919 if (ret)
5d4f98a2 1920 goto fail_dev_root;
d20f7043
CM		 1921
1922 csum_root->track_dirty = 1;
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
2d69a0f8 1930 fs_info->generation = generation;
c146afad 1931 fs_info->last_trans_committed = generation;
d18a2c44
CM		 1932 fs_info->data_alloc_profile = (u64)-1;
1933 fs_info->metadata_alloc_profile = (u64)-1;
1934 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
a74a4b97
CM		 1935 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1936 "btrfs-cleaner");
57506d50 1937 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 1938 goto fail_block_groups;
a74a4b97
CM		 1939
1940 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1941 tree_root,
1942 "btrfs-transaction");
57506d50 1943 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1944 goto fail_cleaner;
a74a4b97 1945
c289811c
CM		 1946 if (!btrfs_test_opt(tree_root, SSD) &&
1947 !btrfs_test_opt(tree_root, NOSSD) &&
1948 !fs_info->fs_devices->rotating) {
1949 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1950 "mode\n");
1951 btrfs_set_opt(fs_info->mount_opt, SSD);
1952 }
1953
e02119d5 1954 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1955 u64 bytenr = btrfs_super_log_root(disk_super);
1956
7c2ca468 1957 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1958 printk(KERN_WARNING "Btrfs log replay required "
1959 "on RO media\n");
7c2ca468
CM		 1960 err = -EIO;
1961 goto fail_trans_kthread;
1962 }
e02119d5
CM		 1963 blocksize =
1964 btrfs_level_size(tree_root,
1965 btrfs_super_log_root_level(disk_super));
d18a2c44 1966
e02119d5
CM		 1967 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1968 GFP_NOFS);
1969
1970 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1971 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1972
1973 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1974 blocksize,
1975 generation + 1);
e02119d5
CM		 1976 ret = btrfs_recover_log_trees(log_tree_root);
1977 BUG_ON(ret);
e556ce2c
YZ		 1978
1979 if (sb->s_flags & MS_RDONLY) {
1980 ret = btrfs_commit_super(tree_root);
1981 BUG_ON(ret);
1982 }
e02119d5 1983 }
1a40e23b 1984
76dda93c
YZ		 1985 ret = btrfs_find_orphan_roots(tree_root);
1986 BUG_ON(ret);
1987
7c2ca468 1988 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1989 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 1990 if (ret < 0) {
1991 printk(KERN_WARNING
1992 "btrfs: failed to recover relocation\n");
1993 err = -EINVAL;
1994 goto fail_trans_kthread;
1995 }
7c2ca468 1996 }
1a40e23b 1997
3de4586c
CM		 1998 location.objectid = BTRFS_FS_TREE_OBJECTID;
1999 location.type = BTRFS_ROOT_ITEM_KEY;
2000 location.offset = (u64)-1;
2001
3de4586c
CM		 2002 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2003 if (!fs_info->fs_root)
7c2ca468 2004 goto fail_trans_kthread;
c289811c 2005
e3acc2a6
JB		 2006 if (!(sb->s_flags & MS_RDONLY)) {
2007 down_read(&fs_info->cleanup_work_sem);
2008 btrfs_orphan_cleanup(fs_info->fs_root);
2009 up_read(&fs_info->cleanup_work_sem);
2010 }
2011
0f7d52f4 2012 return tree_root;
39279cc3 2013
7c2ca468
CM		 2014fail_trans_kthread:
2015 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2016fail_cleaner:
a74a4b97 2017 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2018
2019 /*
2020 * make sure we're done with the btree inode before we stop our
2021 * kthreads
2022 */
2023 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2024 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2025
1b1d1f66
JB		 2026fail_block_groups:
2027 btrfs_free_block_groups(fs_info);
d20f7043 2028 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2029 free_extent_buffer(csum_root->commit_root);
2030fail_dev_root:
2031 free_extent_buffer(dev_root->node);
2032 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2033fail_extent_root:
2034 free_extent_buffer(extent_root->node);
5d4f98a2 2035 free_extent_buffer(extent_root->commit_root);
39279cc3 2036fail_tree_root:
5f39d397 2037 free_extent_buffer(tree_root->node);
5d4f98a2 2038 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2039fail_chunk_root:
2040 free_extent_buffer(chunk_root->node);
5d4f98a2 2041 free_extent_buffer(chunk_root->commit_root);
39279cc3 2042fail_sb_buffer:
61d92c32 2043 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2044 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2045 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2046 btrfs_stop_workers(&fs_info->workers);
2047 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2048 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2049 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2050 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2051 btrfs_stop_workers(&fs_info->submit_workers);
e3ccfa98 2052 btrfs_stop_workers(&fs_info->enospc_workers);
4543df7e 2053fail_iput:
7c2ca468 2054 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2055 iput(fs_info->btree_inode);
7e662854 2056
dfe25020 2057 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2058 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2059fail_bdi:
7e662854 2060 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2061fail_srcu:
2062 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2063fail:
39279cc3
CM		 2064 kfree(extent_root);
2065 kfree(tree_root);
2066 kfree(fs_info);
83afeac4
JM		 2067 kfree(chunk_root);
2068 kfree(dev_root);
d20f7043 2069 kfree(csum_root);
39279cc3 2070 return ERR_PTR(err);
eb60ceac
CM		 2071}
2072
f2984462
CM		 2073static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2074{
2075 char b[BDEVNAME_SIZE];
2076
2077 if (uptodate) {
2078 set_buffer_uptodate(bh);
2079 } else {
2080 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2081 printk(KERN_WARNING "lost page write due to "
2082 "I/O error on %s\n",
2083 bdevname(bh->b_bdev, b));
2084 }
1259ab75
CM		 2085 /* note, we dont' set_buffer_write_io_error because we have
2086 * our own ways of dealing with the IO errors
2087 */
f2984462
CM		 2088 clear_buffer_uptodate(bh);
2089 }
2090 unlock_buffer(bh);
2091 put_bh(bh);
2092}
2093
a512bbf8
YZ		 2094struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2095{
2096 struct buffer_head *bh;
2097 struct buffer_head *latest = NULL;
2098 struct btrfs_super_block *super;
2099 int i;
2100 u64 transid = 0;
2101 u64 bytenr;
2102
2103 /* we would like to check all the supers, but that would make
2104 * a btrfs mount succeed after a mkfs from a different FS.
2105 * So, we need to add a special mount option to scan for
2106 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2107 */
2108 for (i = 0; i < 1; i++) {
2109 bytenr = btrfs_sb_offset(i);
2110 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2111 break;
2112 bh = __bread(bdev, bytenr / 4096, 4096);
2113 if (!bh)
2114 continue;
2115
2116 super = (struct btrfs_super_block *)bh->b_data;
2117 if (btrfs_super_bytenr(super) != bytenr ||
2118 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2119 sizeof(super->magic))) {
2120 brelse(bh);
2121 continue;
2122 }
2123
2124 if (!latest || btrfs_super_generation(super) > transid) {
2125 brelse(latest);
2126 latest = bh;
2127 transid = btrfs_super_generation(super);
2128 } else {
2129 brelse(bh);
2130 }
2131 }
2132 return latest;
2133}
2134
4eedeb75
HH		 2135/*
2136 * this should be called twice, once with wait == 0 and
2137 * once with wait == 1. When wait == 0 is done, all the buffer heads
2138 * we write are pinned.
2139 *
2140 * They are released when wait == 1 is done.
2141 * max_mirrors must be the same for both runs, and it indicates how
2142 * many supers on this one device should be written.
2143 *
2144 * max_mirrors == 0 means to write them all.
2145 */
a512bbf8
YZ		 2146static int write_dev_supers(struct btrfs_device *device,
2147 struct btrfs_super_block *sb,
2148 int do_barriers, int wait, int max_mirrors)
2149{
2150 struct buffer_head *bh;
2151 int i;
2152 int ret;
2153 int errors = 0;
2154 u32 crc;
2155 u64 bytenr;
2156 int last_barrier = 0;
2157
2158 if (max_mirrors == 0)
2159 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2160
2161 /* make sure only the last submit_bh does a barrier */
2162 if (do_barriers) {
2163 for (i = 0; i < max_mirrors; i++) {
2164 bytenr = btrfs_sb_offset(i);
2165 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2166 device->total_bytes)
2167 break;
2168 last_barrier = i;
2169 }
2170 }
2171
2172 for (i = 0; i < max_mirrors; i++) {
2173 bytenr = btrfs_sb_offset(i);
2174 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2175 break;
2176
2177 if (wait) {
2178 bh = __find_get_block(device->bdev, bytenr / 4096,
2179 BTRFS_SUPER_INFO_SIZE);
2180 BUG_ON(!bh);
a512bbf8 2181 wait_on_buffer(bh);
4eedeb75
HH		 2182 if (!buffer_uptodate(bh))
2183 errors++;
2184
2185 /* drop our reference */
2186 brelse(bh);
2187
2188 /* drop the reference from the wait == 0 run */
2189 brelse(bh);
2190 continue;
a512bbf8
YZ		 2191 } else {
2192 btrfs_set_super_bytenr(sb, bytenr);
2193
2194 crc = ~(u32)0;
2195 crc = btrfs_csum_data(NULL, (char *)sb +
2196 BTRFS_CSUM_SIZE, crc,
2197 BTRFS_SUPER_INFO_SIZE -
2198 BTRFS_CSUM_SIZE);
2199 btrfs_csum_final(crc, sb->csum);
2200
4eedeb75
HH		 2201 /*
2202 * one reference for us, and we leave it for the
2203 * caller
2204 */
a512bbf8
YZ		 2205 bh = __getblk(device->bdev, bytenr / 4096,
2206 BTRFS_SUPER_INFO_SIZE);
2207 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2208
4eedeb75 2209 /* one reference for submit_bh */
a512bbf8 2210 get_bh(bh);
4eedeb75
HH		 2211
2212 set_buffer_uptodate(bh);
a512bbf8
YZ		 2213 lock_buffer(bh);
2214 bh->b_end_io = btrfs_end_buffer_write_sync;
2215 }
2216
2217 if (i == last_barrier && do_barriers && device->barriers) {
2218 ret = submit_bh(WRITE_BARRIER, bh);
2219 if (ret == -EOPNOTSUPP) {
2220 printk("btrfs: disabling barriers on dev %s\n",
2221 device->name);
2222 set_buffer_uptodate(bh);
2223 device->barriers = 0;
4eedeb75 2224 /* one reference for submit_bh */
a512bbf8
YZ		 2225 get_bh(bh);
2226 lock_buffer(bh);
ffbd517d 2227 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2228 }
2229 } else {
ffbd517d 2230 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2231 }
2232
4eedeb75 2233 if (ret)
a512bbf8 2234 errors++;
a512bbf8
YZ		 2235 }
2236 return errors < i ? 0 : -1;
2237}
2238
2239int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2240{
e5e9a520 2241 struct list_head *head;
f2984462 2242 struct btrfs_device *dev;
a061fc8d 2243 struct btrfs_super_block *sb;
f2984462 2244 struct btrfs_dev_item *dev_item;
f2984462
CM		 2245 int ret;
2246 int do_barriers;
a236aed1
CM		 2247 int max_errors;
2248 int total_errors = 0;
a061fc8d 2249 u64 flags;
f2984462 2250
a236aed1 2251 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2252 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2253
a061fc8d
CM		 2254 sb = &root->fs_info->super_for_commit;
2255 dev_item = &sb->dev_item;
e5e9a520
CM		 2256
2257 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2258 head = &root->fs_info->fs_devices->devices;
c6e30871 2259 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2260 if (!dev->bdev) {
2261 total_errors++;
2262 continue;
2263 }
2b82032c 2264 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2265 continue;
2266
2b82032c 2267 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2268 btrfs_set_stack_device_type(dev_item, dev->type);
2269 btrfs_set_stack_device_id(dev_item, dev->devid);
2270 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2271 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2272 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2273 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2274 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2275 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2276 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2277
a061fc8d
CM		 2278 flags = btrfs_super_flags(sb);
2279 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2280
a512bbf8 2281 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2282 if (ret)
2283 total_errors++;
f2984462 2284 }
a236aed1 2285 if (total_errors > max_errors) {
d397712b
CM		 2286 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2287 total_errors);
a236aed1
CM		 2288 BUG();
2289 }
f2984462 2290
a512bbf8 2291 total_errors = 0;
c6e30871 2292 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2293 if (!dev->bdev)
2294 continue;
2b82032c 2295 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2296 continue;
2297
a512bbf8
YZ		 2298 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2299 if (ret)
2300 total_errors++;
f2984462 2301 }
e5e9a520 2302 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2303 if (total_errors > max_errors) {
d397712b
CM		 2304 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2305 total_errors);
a236aed1
CM		 2306 BUG();
2307 }
f2984462
CM		 2308 return 0;
2309}
2310
a512bbf8
YZ		 2311int write_ctree_super(struct btrfs_trans_handle *trans,
2312 struct btrfs_root *root, int max_mirrors)
eb60ceac 2313{
e66f709b 2314 int ret;
5f39d397 2315
a512bbf8 2316 ret = write_all_supers(root, max_mirrors);
5f39d397 2317 return ret;
cfaa7295
CM		 2318}
2319
5eda7b5e 2320int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2321{
4df27c4d 2322 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2323 radix_tree_delete(&fs_info->fs_roots_radix,
2324 (unsigned long)root->root_key.objectid);
4df27c4d 2325 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2326
2327 if (btrfs_root_refs(&root->root_item) == 0)
2328 synchronize_srcu(&fs_info->subvol_srcu);
2329
4df27c4d
YZ		 2330 free_fs_root(root);
2331 return 0;
2332}
2333
2334static void free_fs_root(struct btrfs_root *root)
2335{
2336 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2337 if (root->anon_super.s_dev) {
2338 down_write(&root->anon_super.s_umount);
2339 kill_anon_super(&root->anon_super);
2340 }
4df27c4d
YZ		 2341 free_extent_buffer(root->node);
2342 free_extent_buffer(root->commit_root);
d397712b 2343 kfree(root->name);
2619ba1f 2344 kfree(root);
2619ba1f
CM		 2345}
2346
35b7e476 2347static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2348{
2349 int ret;
2350 struct btrfs_root *gang[8];
2351 int i;
2352
76dda93c
YZ		 2353 while (!list_empty(&fs_info->dead_roots)) {
2354 gang[0] = list_entry(fs_info->dead_roots.next,
2355 struct btrfs_root, root_list);
2356 list_del(&gang[0]->root_list);
2357
2358 if (gang[0]->in_radix) {
2359 btrfs_free_fs_root(fs_info, gang[0]);
2360 } else {
2361 free_extent_buffer(gang[0]->node);
2362 free_extent_buffer(gang[0]->commit_root);
2363 kfree(gang[0]);
2364 }
2365 }
2366
d397712b 2367 while (1) {
0f7d52f4
CM		 2368 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2369 (void **)gang, 0,
2370 ARRAY_SIZE(gang));
2371 if (!ret)
2372 break;
2619ba1f 2373 for (i = 0; i < ret; i++)
5eda7b5e 2374 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2375 }
2376 return 0;
2377}
b4100d64 2378
c146afad 2379int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2380{
c146afad
YZ		 2381 u64 root_objectid = 0;
2382 struct btrfs_root *gang[8];
2383 int i;
3768f368 2384 int ret;
e089f05c 2385
c146afad
YZ		 2386 while (1) {
2387 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2388 (void **)gang, root_objectid,
2389 ARRAY_SIZE(gang));
2390 if (!ret)
2391 break;
5d4f98a2
YZ		 2392
2393 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2394 for (i = 0; i < ret; i++) {
2395 root_objectid = gang[i]->root_key.objectid;
c146afad
YZ		 2396 btrfs_orphan_cleanup(gang[i]);
2397 }
2398 root_objectid++;
2399 }
2400 return 0;
2401}
a2135011 2402
c146afad
YZ		 2403int btrfs_commit_super(struct btrfs_root *root)
2404{
2405 struct btrfs_trans_handle *trans;
2406 int ret;
a74a4b97 2407
c146afad 2408 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2409 btrfs_run_delayed_iputs(root);
a74a4b97 2410 btrfs_clean_old_snapshots(root);
c146afad 2411 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2412
2413 /* wait until ongoing cleanup work done */
2414 down_write(&root->fs_info->cleanup_work_sem);
2415 up_write(&root->fs_info->cleanup_work_sem);
2416
79154b1b 2417 trans = btrfs_start_transaction(root, 1);
54aa1f4d 2418 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2419 BUG_ON(ret);
2420 /* run commit again to drop the original snapshot */
79154b1b
CM		 2421 trans = btrfs_start_transaction(root, 1);
2422 btrfs_commit_transaction(trans, root);
2423 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2424 BUG_ON(ret);
d6bfde87 2425
a512bbf8 2426 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2427 return ret;
2428}
2429
2430int close_ctree(struct btrfs_root *root)
2431{
2432 struct btrfs_fs_info *fs_info = root->fs_info;
2433 int ret;
2434
2435 fs_info->closing = 1;
2436 smp_mb();
2437
2438 kthread_stop(root->fs_info->transaction_kthread);
2439 kthread_stop(root->fs_info->cleaner_kthread);
2440
2441 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2442 ret = btrfs_commit_super(root);
d397712b
CM		 2443 if (ret)
2444 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2445 }
0f7d52f4 2446
f25784b3
YZ		 2447 fs_info->closing = 2;
2448 smp_mb();
2449
b0c68f8b 2450 if (fs_info->delalloc_bytes) {
d397712b 2451 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2452 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2453 }
31153d81 2454 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2455 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2456 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2457 }
bcc63abb 2458
5d4f98a2
YZ		 2459 free_extent_buffer(fs_info->extent_root->node);
2460 free_extent_buffer(fs_info->extent_root->commit_root);
2461 free_extent_buffer(fs_info->tree_root->node);
2462 free_extent_buffer(fs_info->tree_root->commit_root);
2463 free_extent_buffer(root->fs_info->chunk_root->node);
2464 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2465 free_extent_buffer(root->fs_info->dev_root->node);
2466 free_extent_buffer(root->fs_info->dev_root->commit_root);
2467 free_extent_buffer(root->fs_info->csum_root->node);
2468 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2469
9078a3e1 2470 btrfs_free_block_groups(root->fs_info);
d10c5f31 2471
c146afad 2472 del_fs_roots(fs_info);
d10c5f31 2473
c146afad 2474 iput(fs_info->btree_inode);
9ad6b7bc 2475
61d92c32 2476 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2477 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2478 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2479 btrfs_stop_workers(&fs_info->workers);
2480 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2481 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2482 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2483 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2484 btrfs_stop_workers(&fs_info->submit_workers);
e3ccfa98 2485 btrfs_stop_workers(&fs_info->enospc_workers);
d6bfde87 2486
dfe25020 2487 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2488 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2489
04160088 2490 bdi_destroy(&fs_info->bdi);
76dda93c 2491 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2492
0f7d52f4 2493 kfree(fs_info->extent_root);
0f7d52f4 2494 kfree(fs_info->tree_root);
0b86a832
CM		 2495 kfree(fs_info->chunk_root);
2496 kfree(fs_info->dev_root);
d20f7043 2497 kfree(fs_info->csum_root);
eb60ceac
CM		 2498 return 0;
2499}
2500
1259ab75 2501int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2502{
1259ab75 2503 int ret;
810191ff 2504 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2505
2ac55d41
JB		 2506 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2507 NULL);
1259ab75
CM		 2508 if (!ret)
2509 return ret;
2510
2511 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2512 parent_transid);
2513 return !ret;
5f39d397
CM		 2514}
2515
2516int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2517{
810191ff 2518 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2519 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2520 buf);
2521}
6702ed49 2522
5f39d397
CM		 2523void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2524{
810191ff 2525 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2526 u64 transid = btrfs_header_generation(buf);
2527 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2528 int was_dirty;
b4ce94de 2529
b9447ef8 2530 btrfs_assert_tree_locked(buf);
ccd467d6 2531 if (transid != root->fs_info->generation) {
d397712b
CM		 2532 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2533 "found %llu running %llu\n",
db94535d 2534 (unsigned long long)buf->start,
d397712b
CM		 2535 (unsigned long long)transid,
2536 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2537 WARN_ON(1);
2538 }
b9473439
CM		 2539 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2540 buf);
2541 if (!was_dirty) {
2542 spin_lock(&root->fs_info->delalloc_lock);
2543 root->fs_info->dirty_metadata_bytes += buf->len;
2544 spin_unlock(&root->fs_info->delalloc_lock);
2545 }
eb60ceac
CM		 2546}
2547
d3c2fdcf 2548void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2549{
188de649
CM		 2550 /*
2551 * looks as though older kernels can get into trouble with
2552 * this code, they end up stuck in balance_dirty_pages forever
2553 */
d6bfde87 2554 u64 num_dirty;
771ed689 2555 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2556
6933c02e 2557 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2558 return;
2559
585ad2c3
CM		 2560 num_dirty = root->fs_info->dirty_metadata_bytes;
2561
d6bfde87
CM		 2562 if (num_dirty > thresh) {
2563 balance_dirty_pages_ratelimited_nr(
d7fc640e 2564 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2565 }
188de649 2566 return;
35b7e476 2567}
6b80053d 2568
ca7a79ad 2569int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2570{
810191ff 2571 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2572 int ret;
ca7a79ad 2573 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2574 if (ret == 0)
b4ce94de 2575 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2576 return ret;
6b80053d 2577}
0da5468f 2578
4bef0848
CM		 2579int btree_lock_page_hook(struct page *page)
2580{
2581 struct inode *inode = page->mapping->host;
b9473439 2582 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2583 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2584 struct extent_buffer *eb;
2585 unsigned long len;
2586 u64 bytenr = page_offset(page);
2587
2588 if (page->private == EXTENT_PAGE_PRIVATE)
2589 goto out;
2590
2591 len = page->private >> 2;
2592 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2593 if (!eb)
2594 goto out;
2595
2596 btrfs_tree_lock(eb);
4bef0848 2597 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2598
2599 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2600 spin_lock(&root->fs_info->delalloc_lock);
2601 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2602 root->fs_info->dirty_metadata_bytes -= eb->len;
2603 else
2604 WARN_ON(1);
2605 spin_unlock(&root->fs_info->delalloc_lock);
2606 }
2607
4bef0848
CM		 2608 btrfs_tree_unlock(eb);
2609 free_extent_buffer(eb);
2610out:
2611 lock_page(page);
2612 return 0;
2613}
2614
d1310b2e 2615static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2616 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2617 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2618 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2619 /* note we're sharing with inode.c for the merge bio hook */
2620 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2621};
Linux kernel - Deliverables
This page took 0.452037 seconds and 5 git commands to generate.