projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: async block group caching
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
4b4e25f2 30#include "compat.h"
eb60ceac
CM		 31#include "ctree.h"
32#include "disk-io.h"
e089f05c 33#include "transaction.h"
0f7d52f4 34#include "btrfs_inode.h"
0b86a832 35#include "volumes.h"
db94535d 36#include "print-tree.h"
8b712842 37#include "async-thread.h"
925baedd 38#include "locking.h"
e02119d5 39#include "tree-log.h"
fa9c0d79 40#include "free-space-cache.h"
eb60ceac 41
d1310b2e 42static struct extent_io_ops btree_extent_io_ops;
8b712842 43static void end_workqueue_fn(struct btrfs_work *work);
ce9adaa5 44
d352ac68
CM		 45/*
46 * end_io_wq structs are used to do processing in task context when an IO is
47 * complete. This is used during reads to verify checksums, and it is used
48 * by writes to insert metadata for new file extents after IO is complete.
49 */
ce9adaa5
CM		 50struct end_io_wq {
51 struct bio *bio;
52 bio_end_io_t *end_io;
53 void *private;
54 struct btrfs_fs_info *info;
55 int error;
22c59948 56 int metadata;
ce9adaa5 57 struct list_head list;
8b712842 58 struct btrfs_work work;
ce9adaa5 59};
0da5468f 60
d352ac68
CM		 61/*
62 * async submit bios are used to offload expensive checksumming
63 * onto the worker threads. They checksum file and metadata bios
64 * just before they are sent down the IO stack.
65 */
44b8bd7e
CM		 66struct async_submit_bio {
67 struct inode *inode;
68 struct bio *bio;
69 struct list_head list;
4a69a410
CM		 70 extent_submit_bio_hook_t *submit_bio_start;
71 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 72 int rw;
73 int mirror_num;
c8b97818 74 unsigned long bio_flags;
8b712842 75 struct btrfs_work work;
44b8bd7e
CM		 76};
77
4008c04a
CM		 78/* These are used to set the lockdep class on the extent buffer locks.
79 * The class is set by the readpage_end_io_hook after the buffer has
80 * passed csum validation but before the pages are unlocked.
81 *
82 * The lockdep class is also set by btrfs_init_new_buffer on freshly
83 * allocated blocks.
84 *
85 * The class is based on the level in the tree block, which allows lockdep
86 * to know that lower nodes nest inside the locks of higher nodes.
87 *
88 * We also add a check to make sure the highest level of the tree is
89 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
90 * code needs update as well.
91 */
92#ifdef CONFIG_DEBUG_LOCK_ALLOC
93# if BTRFS_MAX_LEVEL != 8
94# error
95# endif
96static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
97static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
98 /* leaf */
99 "btrfs-extent-00",
100 "btrfs-extent-01",
101 "btrfs-extent-02",
102 "btrfs-extent-03",
103 "btrfs-extent-04",
104 "btrfs-extent-05",
105 "btrfs-extent-06",
106 "btrfs-extent-07",
107 /* highest possible level */
108 "btrfs-extent-08",
109};
110#endif
111
d352ac68
CM		 112/*
113 * extents on the btree inode are pretty simple, there's one extent
114 * that covers the entire device
115 */
b2950863
CH		 116static struct extent_map *btree_get_extent(struct inode *inode,
117 struct page *page, size_t page_offset, u64 start, u64 len,
118 int create)
7eccb903 119{
5f39d397
CM		 120 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
121 struct extent_map *em;
122 int ret;
123
d1310b2e
CM		 124 spin_lock(&em_tree->lock);
125 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 126 if (em) {
127 em->bdev =
128 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
129 spin_unlock(&em_tree->lock);
5f39d397 130 goto out;
a061fc8d
CM		 131 }
132 spin_unlock(&em_tree->lock);
7b13b7b1 133
5f39d397
CM		 134 em = alloc_extent_map(GFP_NOFS);
135 if (!em) {
136 em = ERR_PTR(-ENOMEM);
137 goto out;
138 }
139 em->start = 0;
0afbaf8c 140 em->len = (u64)-1;
c8b97818 141 em->block_len = (u64)-1;
5f39d397 142 em->block_start = 0;
a061fc8d 143 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e
CM		 144
145 spin_lock(&em_tree->lock);
5f39d397
CM		 146 ret = add_extent_mapping(em_tree, em);
147 if (ret == -EEXIST) {
0afbaf8c
CM		 148 u64 failed_start = em->start;
149 u64 failed_len = em->len;
150
5f39d397 151 free_extent_map(em);
7b13b7b1 152 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 153 if (em) {
7b13b7b1 154 ret = 0;
0afbaf8c
CM		 155 } else {
156 em = lookup_extent_mapping(em_tree, failed_start,
157 failed_len);
7b13b7b1 158 ret = -EIO;
0afbaf8c 159 }
5f39d397 160 } else if (ret) {
7b13b7b1
CM		 161 free_extent_map(em);
162 em = NULL;
5f39d397 163 }
7b13b7b1
CM		 164 spin_unlock(&em_tree->lock);
165
166 if (ret)
167 em = ERR_PTR(ret);
5f39d397
CM		 168out:
169 return em;
7eccb903
CM		 170}
171
19c00ddc
CM		 172u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
173{
163e783e 174 return crc32c(seed, data, len);
19c00ddc
CM		 175}
176
177void btrfs_csum_final(u32 crc, char *result)
178{
179 *(__le32 *)result = ~cpu_to_le32(crc);
180}
181
d352ac68
CM		 182/*
183 * compute the csum for a btree block, and either verify it or write it
184 * into the csum field of the block.
185 */
19c00ddc
CM		 186static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
187 int verify)
188{
607d432d
JB		 189 u16 csum_size =
190 btrfs_super_csum_size(&root->fs_info->super_copy);
191 char *result = NULL;
19c00ddc
CM		 192 unsigned long len;
193 unsigned long cur_len;
194 unsigned long offset = BTRFS_CSUM_SIZE;
195 char *map_token = NULL;
196 char *kaddr;
197 unsigned long map_start;
198 unsigned long map_len;
199 int err;
200 u32 crc = ~(u32)0;
607d432d 201 unsigned long inline_result;
19c00ddc
CM		 202
203 len = buf->len - offset;
d397712b 204 while (len > 0) {
19c00ddc
CM		 205 err = map_private_extent_buffer(buf, offset, 32,
206 &map_token, &kaddr,
207 &map_start, &map_len, KM_USER0);
d397712b 208 if (err)
19c00ddc 209 return 1;
19c00ddc
CM		 210 cur_len = min(len, map_len - (offset - map_start));
211 crc = btrfs_csum_data(root, kaddr + offset - map_start,
212 crc, cur_len);
213 len -= cur_len;
214 offset += cur_len;
215 unmap_extent_buffer(buf, map_token, KM_USER0);
216 }
607d432d
JB		 217 if (csum_size > sizeof(inline_result)) {
218 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
219 if (!result)
220 return 1;
221 } else {
222 result = (char *)&inline_result;
223 }
224
19c00ddc
CM		 225 btrfs_csum_final(crc, result);
226
227 if (verify) {
607d432d 228 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 229 u32 val;
230 u32 found = 0;
607d432d 231 memcpy(&found, result, csum_size);
e4204ded 232
607d432d 233 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 234 if (printk_ratelimit()) {
235 printk(KERN_INFO "btrfs: %s checksum verify "
236 "failed on %llu wanted %X found %X "
237 "level %d\n",
238 root->fs_info->sb->s_id,
239 (unsigned long long)buf->start, val, found,
240 btrfs_header_level(buf));
241 }
607d432d
JB		 242 if (result != (char *)&inline_result)
243 kfree(result);
19c00ddc
CM		 244 return 1;
245 }
246 } else {
607d432d 247 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 248 }
607d432d
JB		 249 if (result != (char *)&inline_result)
250 kfree(result);
19c00ddc
CM		 251 return 0;
252}
253
d352ac68
CM		 254/*
255 * we can't consider a given block up to date unless the transid of the
256 * block matches the transid in the parent node's pointer. This is how we
257 * detect blocks that either didn't get written at all or got written
258 * in the wrong place.
259 */
1259ab75
CM		 260static int verify_parent_transid(struct extent_io_tree *io_tree,
261 struct extent_buffer *eb, u64 parent_transid)
262{
263 int ret;
264
265 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
266 return 0;
267
268 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
269 if (extent_buffer_uptodate(io_tree, eb) &&
270 btrfs_header_generation(eb) == parent_transid) {
271 ret = 0;
272 goto out;
273 }
193f284d
CM		 274 if (printk_ratelimit()) {
275 printk("parent transid verify failed on %llu wanted %llu "
276 "found %llu\n",
277 (unsigned long long)eb->start,
278 (unsigned long long)parent_transid,
279 (unsigned long long)btrfs_header_generation(eb));
280 }
1259ab75 281 ret = 1;
1259ab75 282 clear_extent_buffer_uptodate(io_tree, eb);
33958dc6 283out:
1259ab75
CM		 284 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
285 GFP_NOFS);
286 return ret;
1259ab75
CM		 287}
288
d352ac68
CM		 289/*
290 * helper to read a given tree block, doing retries as required when
291 * the checksums don't match and we have alternate mirrors to try.
292 */
f188591e
CM		 293static int btree_read_extent_buffer_pages(struct btrfs_root *root,
294 struct extent_buffer *eb,
ca7a79ad 295 u64 start, u64 parent_transid)
f188591e
CM		 296{
297 struct extent_io_tree *io_tree;
298 int ret;
299 int num_copies = 0;
300 int mirror_num = 0;
301
302 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
303 while (1) {
304 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
305 btree_get_extent, mirror_num);
1259ab75
CM		 306 if (!ret &&
307 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 308 return ret;
d397712b 309
f188591e
CM		 310 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
311 eb->start, eb->len);
4235298e 312 if (num_copies == 1)
f188591e 313 return ret;
4235298e 314
f188591e 315 mirror_num++;
4235298e 316 if (mirror_num > num_copies)
f188591e 317 return ret;
f188591e 318 }
f188591e
CM		 319 return -EIO;
320}
19c00ddc 321
d352ac68 322/*
d397712b
CM		 323 * checksum a dirty tree block before IO. This has extra checks to make sure
324 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 325 */
d397712b 326
b2950863 327static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 328{
d1310b2e 329 struct extent_io_tree *tree;
35ebb934 330 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc
CM		 331 u64 found_start;
332 int found_level;
333 unsigned long len;
334 struct extent_buffer *eb;
f188591e
CM		 335 int ret;
336
d1310b2e 337 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 338
339 if (page->private == EXTENT_PAGE_PRIVATE)
340 goto out;
341 if (!page->private)
342 goto out;
343 len = page->private >> 2;
d397712b
CM		 344 WARN_ON(len == 0);
345
19c00ddc 346 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 347 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
348 btrfs_header_generation(eb));
f188591e 349 BUG_ON(ret);
19c00ddc
CM		 350 found_start = btrfs_header_bytenr(eb);
351 if (found_start != start) {
55c69072
CM		 352 WARN_ON(1);
353 goto err;
354 }
355 if (eb->first_page != page) {
55c69072
CM		 356 WARN_ON(1);
357 goto err;
358 }
359 if (!PageUptodate(page)) {
55c69072
CM		 360 WARN_ON(1);
361 goto err;
19c00ddc
CM		 362 }
363 found_level = btrfs_header_level(eb);
4bef0848 364
19c00ddc 365 csum_tree_block(root, eb, 0);
55c69072 366err:
19c00ddc
CM		 367 free_extent_buffer(eb);
368out:
369 return 0;
370}
371
2b82032c
YZ		 372static int check_tree_block_fsid(struct btrfs_root *root,
373 struct extent_buffer *eb)
374{
375 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
376 u8 fsid[BTRFS_UUID_SIZE];
377 int ret = 1;
378
379 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
380 BTRFS_FSID_SIZE);
381 while (fs_devices) {
382 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
383 ret = 0;
384 break;
385 }
386 fs_devices = fs_devices->seed;
387 }
388 return ret;
389}
390
4008c04a
CM		 391#ifdef CONFIG_DEBUG_LOCK_ALLOC
392void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
393{
394 lockdep_set_class_and_name(&eb->lock,
395 &btrfs_eb_class[level],
396 btrfs_eb_name[level]);
397}
398#endif
399
b2950863 400static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 401 struct extent_state *state)
402{
403 struct extent_io_tree *tree;
404 u64 found_start;
405 int found_level;
406 unsigned long len;
407 struct extent_buffer *eb;
408 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 409 int ret = 0;
ce9adaa5
CM		 410
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
412 if (page->private == EXTENT_PAGE_PRIVATE)
413 goto out;
414 if (!page->private)
415 goto out;
d397712b 416
ce9adaa5 417 len = page->private >> 2;
d397712b
CM		 418 WARN_ON(len == 0);
419
ce9adaa5 420 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 421
ce9adaa5 422 found_start = btrfs_header_bytenr(eb);
23a07867 423 if (found_start != start) {
193f284d
CM		 424 if (printk_ratelimit()) {
425 printk(KERN_INFO "btrfs bad tree block start "
426 "%llu %llu\n",
427 (unsigned long long)found_start,
428 (unsigned long long)eb->start);
429 }
f188591e 430 ret = -EIO;
ce9adaa5
CM		 431 goto err;
432 }
433 if (eb->first_page != page) {
d397712b
CM		 434 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
435 eb->first_page->index, page->index);
ce9adaa5 436 WARN_ON(1);
f188591e 437 ret = -EIO;
ce9adaa5
CM		 438 goto err;
439 }
2b82032c 440 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 441 if (printk_ratelimit()) {
442 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
443 (unsigned long long)eb->start);
444 }
1259ab75
CM		 445 ret = -EIO;
446 goto err;
447 }
ce9adaa5
CM		 448 found_level = btrfs_header_level(eb);
449
4008c04a
CM		 450 btrfs_set_buffer_lockdep_class(eb, found_level);
451
ce9adaa5 452 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 453 if (ret)
454 ret = -EIO;
ce9adaa5
CM		 455
456 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
457 end = eb->start + end - 1;
ce9adaa5
CM		 458err:
459 free_extent_buffer(eb);
460out:
f188591e 461 return ret;
ce9adaa5
CM		 462}
463
ce9adaa5 464static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 465{
466 struct end_io_wq *end_io_wq = bio->bi_private;
467 struct btrfs_fs_info *fs_info;
ce9adaa5 468
ce9adaa5 469 fs_info = end_io_wq->info;
ce9adaa5 470 end_io_wq->error = err;
8b712842
CM		 471 end_io_wq->work.func = end_workqueue_fn;
472 end_io_wq->work.flags = 0;
d20f7043
CM		 473
474 if (bio->bi_rw & (1 << BIO_RW)) {
cad321ad
CM		 475 if (end_io_wq->metadata)
476 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
477 &end_io_wq->work);
478 else
479 btrfs_queue_worker(&fs_info->endio_write_workers,
480 &end_io_wq->work);
d20f7043
CM		 481 } else {
482 if (end_io_wq->metadata)
483 btrfs_queue_worker(&fs_info->endio_meta_workers,
484 &end_io_wq->work);
485 else
486 btrfs_queue_worker(&fs_info->endio_workers,
487 &end_io_wq->work);
488 }
ce9adaa5
CM		 489}
490
22c59948
CM		 491int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
492 int metadata)
0b86a832 493{
ce9adaa5 494 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 495 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
496 if (!end_io_wq)
497 return -ENOMEM;
498
499 end_io_wq->private = bio->bi_private;
500 end_io_wq->end_io = bio->bi_end_io;
22c59948 501 end_io_wq->info = info;
ce9adaa5
CM		 502 end_io_wq->error = 0;
503 end_io_wq->bio = bio;
22c59948 504 end_io_wq->metadata = metadata;
ce9adaa5
CM		 505
506 bio->bi_private = end_io_wq;
507 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 508 return 0;
509}
510
b64a2851 511unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 512{
4854ddd0
CM		 513 unsigned long limit = min_t(unsigned long,
514 info->workers.max_workers,
515 info->fs_devices->open_devices);
516 return 256 * limit;
517}
0986fe9e 518
4854ddd0
CM		 519int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
520{
b64a2851
CM		 521 return atomic_read(&info->nr_async_bios) >
522 btrfs_async_submit_limit(info);
0986fe9e
CM		 523}
524
4a69a410
CM		 525static void run_one_async_start(struct btrfs_work *work)
526{
527 struct btrfs_fs_info *fs_info;
528 struct async_submit_bio *async;
529
530 async = container_of(work, struct async_submit_bio, work);
531 fs_info = BTRFS_I(async->inode)->root->fs_info;
532 async->submit_bio_start(async->inode, async->rw, async->bio,
533 async->mirror_num, async->bio_flags);
534}
535
536static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 537{
538 struct btrfs_fs_info *fs_info;
539 struct async_submit_bio *async;
4854ddd0 540 int limit;
8b712842
CM		 541
542 async = container_of(work, struct async_submit_bio, work);
543 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 544
b64a2851 545 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 546 limit = limit * 2 / 3;
547
8b712842 548 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 549
b64a2851
CM		 550 if (atomic_read(&fs_info->nr_async_submits) < limit &&
551 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 552 wake_up(&fs_info->async_submit_wait);
553
4a69a410 554 async->submit_bio_done(async->inode, async->rw, async->bio,
c8b97818 555 async->mirror_num, async->bio_flags);
4a69a410
CM		 556}
557
558static void run_one_async_free(struct btrfs_work *work)
559{
560 struct async_submit_bio *async;
561
562 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 563 kfree(async);
564}
565
44b8bd7e
CM		 566int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
567 int rw, struct bio *bio, int mirror_num,
c8b97818 568 unsigned long bio_flags,
4a69a410
CM		 569 extent_submit_bio_hook_t *submit_bio_start,
570 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 571{
572 struct async_submit_bio *async;
573
574 async = kmalloc(sizeof(*async), GFP_NOFS);
575 if (!async)
576 return -ENOMEM;
577
578 async->inode = inode;
579 async->rw = rw;
580 async->bio = bio;
581 async->mirror_num = mirror_num;
4a69a410
CM		 582 async->submit_bio_start = submit_bio_start;
583 async->submit_bio_done = submit_bio_done;
584
585 async->work.func = run_one_async_start;
586 async->work.ordered_func = run_one_async_done;
587 async->work.ordered_free = run_one_async_free;
588
8b712842 589 async->work.flags = 0;
c8b97818 590 async->bio_flags = bio_flags;
8c8bee1d 591
cb03c743 592 atomic_inc(&fs_info->nr_async_submits);
d313d7a3
CM		 593
594 if (rw & (1 << BIO_RW_SYNCIO))
595 btrfs_set_work_high_prio(&async->work);
596
8b712842 597 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 598
d397712b 599 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 600 atomic_read(&fs_info->nr_async_submits)) {
601 wait_event(fs_info->async_submit_wait,
602 (atomic_read(&fs_info->nr_async_submits) == 0));
603 }
604
44b8bd7e
CM		 605 return 0;
606}
607
ce3ed71a
CM		 608static int btree_csum_one_bio(struct bio *bio)
609{
610 struct bio_vec *bvec = bio->bi_io_vec;
611 int bio_index = 0;
612 struct btrfs_root *root;
613
614 WARN_ON(bio->bi_vcnt <= 0);
d397712b 615 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 616 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
617 csum_dirty_buffer(root, bvec->bv_page);
618 bio_index++;
619 bvec++;
620 }
621 return 0;
622}
623
4a69a410
CM		 624static int __btree_submit_bio_start(struct inode *inode, int rw,
625 struct bio *bio, int mirror_num,
626 unsigned long bio_flags)
22c59948 627{
8b712842
CM		 628 /*
629 * when we're called for a write, we're already in the async
5443be45 630 * submission context. Just jump into btrfs_map_bio
8b712842 631 */
4a69a410
CM		 632 btree_csum_one_bio(bio);
633 return 0;
634}
22c59948 635
4a69a410
CM		 636static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
637 int mirror_num, unsigned long bio_flags)
638{
8b712842 639 /*
4a69a410
CM		 640 * when we're called for a write, we're already in the async
641 * submission context. Just jump into btrfs_map_bio
8b712842 642 */
8b712842 643 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 644}
645
44b8bd7e 646static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
c8b97818 647 int mirror_num, unsigned long bio_flags)
44b8bd7e 648{
cad321ad
CM		 649 int ret;
650
651 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
652 bio, 1);
653 BUG_ON(ret);
654
44b8bd7e 655 if (!(rw & (1 << BIO_RW))) {
4a69a410
CM		 656 /*
657 * called for a read, do the setup so that checksum validation
658 * can happen in the async kernel threads
659 */
4a69a410 660 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 661 mirror_num, 0);
44b8bd7e 662 }
d313d7a3 663
cad321ad
CM		 664 /*
665 * kthread helpers are used to submit writes so that checksumming
666 * can happen in parallel across all CPUs
667 */
44b8bd7e 668 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 669 inode, rw, bio, mirror_num, 0,
4a69a410
CM		 670 __btree_submit_bio_start,
671 __btree_submit_bio_done);
44b8bd7e
CM		 672}
673
0da5468f
CM		 674static int btree_writepage(struct page *page, struct writeback_control *wbc)
675{
d1310b2e 676 struct extent_io_tree *tree;
b9473439
CM		 677 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678 struct extent_buffer *eb;
679 int was_dirty;
680
d1310b2e 681 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 682 if (!(current->flags & PF_MEMALLOC)) {
683 return extent_write_full_page(tree, page,
684 btree_get_extent, wbc);
685 }
5443be45 686
b9473439
CM		 687 redirty_page_for_writepage(wbc, page);
688 eb = btrfs_find_tree_block(root, page_offset(page),
689 PAGE_CACHE_SIZE);
690 WARN_ON(!eb);
691
692 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
693 if (!was_dirty) {
694 spin_lock(&root->fs_info->delalloc_lock);
695 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
696 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 697 }
b9473439
CM		 698 free_extent_buffer(eb);
699
700 unlock_page(page);
701 return 0;
5f39d397 702}
0da5468f
CM		 703
704static int btree_writepages(struct address_space *mapping,
705 struct writeback_control *wbc)
706{
d1310b2e
CM		 707 struct extent_io_tree *tree;
708 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 709 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 710 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 711 u64 num_dirty;
24ab9cd8 712 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 713
714 if (wbc->for_kupdate)
715 return 0;
716
b9473439
CM		 717 /* this is a bit racy, but that's ok */
718 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 719 if (num_dirty < thresh)
793955bc 720 return 0;
793955bc 721 }
0da5468f
CM		 722 return extent_writepages(tree, mapping, btree_get_extent, wbc);
723}
724
b2950863 725static int btree_readpage(struct file *file, struct page *page)
5f39d397 726{
d1310b2e
CM		 727 struct extent_io_tree *tree;
728 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 729 return extent_read_full_page(tree, page, btree_get_extent);
730}
22b0ebda 731
70dec807 732static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 733{
d1310b2e
CM		 734 struct extent_io_tree *tree;
735 struct extent_map_tree *map;
5f39d397 736 int ret;
d98237b3 737
98509cfc 738 if (PageWriteback(page) || PageDirty(page))
d397712b 739 return 0;
98509cfc 740
d1310b2e
CM		 741 tree = &BTRFS_I(page->mapping->host)->io_tree;
742 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 743
7b13b7b1 744 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 745 if (!ret)
6af118ce 746 return 0;
6af118ce
CM		 747
748 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 749 if (ret == 1) {
750 ClearPagePrivate(page);
751 set_page_private(page, 0);
752 page_cache_release(page);
753 }
6af118ce 754
d98237b3
CM		 755 return ret;
756}
757
5f39d397 758static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 759{
d1310b2e
CM		 760 struct extent_io_tree *tree;
761 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 762 extent_invalidatepage(tree, page, offset);
763 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 764 if (PagePrivate(page)) {
d397712b
CM		 765 printk(KERN_WARNING "btrfs warning page private not zero "
766 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 767 ClearPagePrivate(page);
768 set_page_private(page, 0);
769 page_cache_release(page);
770 }
d98237b3
CM		 771}
772
d98237b3
CM		 773static struct address_space_operations btree_aops = {
774 .readpage = btree_readpage,
775 .writepage = btree_writepage,
0da5468f 776 .writepages = btree_writepages,
5f39d397
CM		 777 .releasepage = btree_releasepage,
778 .invalidatepage = btree_invalidatepage,
d98237b3
CM		 779 .sync_page = block_sync_page,
780};
781
ca7a79ad
CM		 782int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
783 u64 parent_transid)
090d1875 784{
5f39d397
CM		 785 struct extent_buffer *buf = NULL;
786 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 787 int ret = 0;
090d1875 788
db94535d 789 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 790 if (!buf)
090d1875 791 return 0;
d1310b2e 792 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 793 buf, 0, 0, btree_get_extent, 0);
5f39d397 794 free_extent_buffer(buf);
de428b63 795 return ret;
090d1875
CM		 796}
797
0999df54
CM		 798struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
799 u64 bytenr, u32 blocksize)
800{
801 struct inode *btree_inode = root->fs_info->btree_inode;
802 struct extent_buffer *eb;
803 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
804 bytenr, blocksize, GFP_NOFS);
805 return eb;
806}
807
808struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
809 u64 bytenr, u32 blocksize)
810{
811 struct inode *btree_inode = root->fs_info->btree_inode;
812 struct extent_buffer *eb;
813
814 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
815 bytenr, blocksize, NULL, GFP_NOFS);
816 return eb;
817}
818
819
e02119d5
CM		 820int btrfs_write_tree_block(struct extent_buffer *buf)
821{
822 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
24ab9cd8 823 buf->start + buf->len - 1, WB_SYNC_ALL);
e02119d5
CM		 824}
825
826int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
827{
828 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
d397712b 829 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 830}
831
0999df54 832struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 833 u32 blocksize, u64 parent_transid)
0999df54
CM		 834{
835 struct extent_buffer *buf = NULL;
836 struct inode *btree_inode = root->fs_info->btree_inode;
837 struct extent_io_tree *io_tree;
838 int ret;
839
840 io_tree = &BTRFS_I(btree_inode)->io_tree;
841
842 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
843 if (!buf)
844 return NULL;
0999df54 845
ca7a79ad 846 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 847
d397712b 848 if (ret == 0)
b4ce94de 849 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 850 return buf;
ce9adaa5 851
eb60ceac
CM		 852}
853
e089f05c 854int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 855 struct extent_buffer *buf)
ed2ff2cb 856{
5f39d397 857 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 858 if (btrfs_header_generation(buf) ==
925baedd 859 root->fs_info->running_transaction->transid) {
b9447ef8 860 btrfs_assert_tree_locked(buf);
b4ce94de 861
b9473439
CM		 862 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
863 spin_lock(&root->fs_info->delalloc_lock);
864 if (root->fs_info->dirty_metadata_bytes >= buf->len)
865 root->fs_info->dirty_metadata_bytes -= buf->len;
866 else
867 WARN_ON(1);
868 spin_unlock(&root->fs_info->delalloc_lock);
869 }
b4ce94de 870
b9473439
CM		 871 /* ugh, clear_extent_buffer_dirty needs to lock the page */
872 btrfs_set_lock_blocking(buf);
d1310b2e 873 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 874 buf);
925baedd 875 }
5f39d397
CM		 876 return 0;
877}
878
db94535d 879static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 880 u32 stripesize, struct btrfs_root *root,
9f5fae2f 881 struct btrfs_fs_info *fs_info,
e20d96d6 882 u64 objectid)
d97e63b6 883{
cfaa7295 884 root->node = NULL;
a28ec197 885 root->commit_root = NULL;
db94535d
CM		 886 root->sectorsize = sectorsize;
887 root->nodesize = nodesize;
888 root->leafsize = leafsize;
87ee04eb 889 root->stripesize = stripesize;
123abc88 890 root->ref_cows = 0;
0b86a832
CM		 891 root->track_dirty = 0;
892
9f5fae2f 893 root->fs_info = fs_info;
0f7d52f4
CM		 894 root->objectid = objectid;
895 root->last_trans = 0;
1b05da2e
CM		 896 root->highest_inode = 0;
897 root->last_inode_alloc = 0;
58176a96 898 root->name = NULL;
4313b399 899 root->in_sysfs = 0;
5d4f98a2 900 root->inode_tree.rb_node = NULL;
0b86a832
CM		 901
902 INIT_LIST_HEAD(&root->dirty_list);
7b128766 903 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 904 INIT_LIST_HEAD(&root->root_list);
925baedd 905 spin_lock_init(&root->node_lock);
bcc63abb 906 spin_lock_init(&root->list_lock);
5d4f98a2 907 spin_lock_init(&root->inode_lock);
a2135011 908 mutex_init(&root->objectid_mutex);
e02119d5 909 mutex_init(&root->log_mutex);
817d52f8 910 init_rwsem(&root->commit_root_sem);
7237f183
YZ		 911 init_waitqueue_head(&root->log_writer_wait);
912 init_waitqueue_head(&root->log_commit_wait[0]);
913 init_waitqueue_head(&root->log_commit_wait[1]);
914 atomic_set(&root->log_commit[0], 0);
915 atomic_set(&root->log_commit[1], 0);
916 atomic_set(&root->log_writers, 0);
917 root->log_batch = 0;
918 root->log_transid = 0;
d0c803c4
CM		 919 extent_io_tree_init(&root->dirty_log_pages,
920 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 921
3768f368
CM		 922 memset(&root->root_key, 0, sizeof(root->root_key));
923 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 924 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 925 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 926 root->defrag_trans_start = fs_info->generation;
58176a96 927 init_completion(&root->kobj_unregister);
6702ed49
CM		 928 root->defrag_running = 0;
929 root->defrag_level = 0;
4d775673 930 root->root_key.objectid = objectid;
3394e160
CM		 931 root->anon_super.s_root = NULL;
932 root->anon_super.s_dev = 0;
933 INIT_LIST_HEAD(&root->anon_super.s_list);
934 INIT_LIST_HEAD(&root->anon_super.s_instances);
935 init_rwsem(&root->anon_super.s_umount);
936
3768f368
CM		 937 return 0;
938}
939
db94535d 940static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 941 struct btrfs_fs_info *fs_info,
942 u64 objectid,
e20d96d6 943 struct btrfs_root *root)
3768f368
CM		 944{
945 int ret;
db94535d 946 u32 blocksize;
84234f3a 947 u64 generation;
3768f368 948
db94535d 949 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 950 tree_root->sectorsize, tree_root->stripesize,
951 root, fs_info, objectid);
3768f368
CM		 952 ret = btrfs_find_last_root(tree_root, objectid,
953 &root->root_item, &root->root_key);
954 BUG_ON(ret);
955
84234f3a 956 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 957 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
958 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 959 blocksize, generation);
5d4f98a2 960 root->commit_root = btrfs_root_node(root);
3768f368 961 BUG_ON(!root->node);
d97e63b6
CM		 962 return 0;
963}
964
e02119d5
CM		 965int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
966 struct btrfs_fs_info *fs_info)
967{
968 struct extent_buffer *eb;
d0c803c4
CM		 969 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
970 u64 start = 0;
971 u64 end = 0;
e02119d5
CM		 972 int ret;
973
d0c803c4 974 if (!log_root_tree)
e02119d5
CM		 975 return 0;
976
d397712b 977 while (1) {
d0c803c4
CM		 978 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
979 0, &start, &end, EXTENT_DIRTY);
980 if (ret)
981 break;
982
983 clear_extent_dirty(&log_root_tree->dirty_log_pages,
984 start, end, GFP_NOFS);
985 }
e02119d5
CM		 986 eb = fs_info->log_root_tree->node;
987
988 WARN_ON(btrfs_header_level(eb) != 0);
989 WARN_ON(btrfs_header_nritems(eb) != 0);
990
d00aff00
CM		 991 ret = btrfs_free_reserved_extent(fs_info->tree_root,
992 eb->start, eb->len);
e02119d5
CM		 993 BUG_ON(ret);
994
995 free_extent_buffer(eb);
996 kfree(fs_info->log_root_tree);
997 fs_info->log_root_tree = NULL;
998 return 0;
999}
1000
7237f183
YZ		 1001static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1002 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1003{
1004 struct btrfs_root *root;
1005 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1006 struct extent_buffer *leaf;
e02119d5
CM		 1007
1008 root = kzalloc(sizeof(*root), GFP_NOFS);
1009 if (!root)
7237f183 1010 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1011
1012 __setup_root(tree_root->nodesize, tree_root->leafsize,
1013 tree_root->sectorsize, tree_root->stripesize,
1014 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1015
1016 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1017 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1018 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1019 /*
1020 * log trees do not get reference counted because they go away
1021 * before a real commit is actually done. They do store pointers
1022 * to file data extents, and those reference counts still get
1023 * updated (along with back refs to the log tree).
1024 */
e02119d5
CM		 1025 root->ref_cows = 0;
1026
5d4f98a2
YZ		 1027 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1028 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1029 if (IS_ERR(leaf)) {
1030 kfree(root);
1031 return ERR_CAST(leaf);
1032 }
e02119d5 1033
5d4f98a2
YZ		 1034 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1035 btrfs_set_header_bytenr(leaf, leaf->start);
1036 btrfs_set_header_generation(leaf, trans->transid);
1037 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1038 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1039 root->node = leaf;
e02119d5
CM		 1040
1041 write_extent_buffer(root->node, root->fs_info->fsid,
1042 (unsigned long)btrfs_header_fsid(root->node),
1043 BTRFS_FSID_SIZE);
1044 btrfs_mark_buffer_dirty(root->node);
1045 btrfs_tree_unlock(root->node);
7237f183
YZ		 1046 return root;
1047}
1048
1049int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1050 struct btrfs_fs_info *fs_info)
1051{
1052 struct btrfs_root *log_root;
1053
1054 log_root = alloc_log_tree(trans, fs_info);
1055 if (IS_ERR(log_root))
1056 return PTR_ERR(log_root);
1057 WARN_ON(fs_info->log_root_tree);
1058 fs_info->log_root_tree = log_root;
1059 return 0;
1060}
1061
1062int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1063 struct btrfs_root *root)
1064{
1065 struct btrfs_root *log_root;
1066 struct btrfs_inode_item *inode_item;
1067
1068 log_root = alloc_log_tree(trans, root->fs_info);
1069 if (IS_ERR(log_root))
1070 return PTR_ERR(log_root);
1071
1072 log_root->last_trans = trans->transid;
1073 log_root->root_key.offset = root->root_key.objectid;
1074
1075 inode_item = &log_root->root_item.inode;
1076 inode_item->generation = cpu_to_le64(1);
1077 inode_item->size = cpu_to_le64(3);
1078 inode_item->nlink = cpu_to_le32(1);
1079 inode_item->nbytes = cpu_to_le64(root->leafsize);
1080 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1081
5d4f98a2 1082 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1083
1084 WARN_ON(root->log_root);
1085 root->log_root = log_root;
1086 root->log_transid = 0;
e02119d5
CM		 1087 return 0;
1088}
1089
1090struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1091 struct btrfs_key *location)
1092{
1093 struct btrfs_root *root;
1094 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1095 struct btrfs_path *path;
5f39d397 1096 struct extent_buffer *l;
1b05da2e 1097 u64 highest_inode;
84234f3a 1098 u64 generation;
db94535d 1099 u32 blocksize;
0f7d52f4
CM		 1100 int ret = 0;
1101
5eda7b5e 1102 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1103 if (!root)
0f7d52f4 1104 return ERR_PTR(-ENOMEM);
0f7d52f4 1105 if (location->offset == (u64)-1) {
db94535d 1106 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1107 location->objectid, root);
1108 if (ret) {
0f7d52f4
CM		 1109 kfree(root);
1110 return ERR_PTR(ret);
1111 }
1112 goto insert;
1113 }
1114
db94535d 1115 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1116 tree_root->sectorsize, tree_root->stripesize,
1117 root, fs_info, location->objectid);
0f7d52f4
CM		 1118
1119 path = btrfs_alloc_path();
1120 BUG_ON(!path);
1121 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1122 if (ret != 0) {
0f7d52f4
CM		 1123 if (ret > 0)
1124 ret = -ENOENT;
1125 goto out;
1126 }
5f39d397
CM		 1127 l = path->nodes[0];
1128 read_extent_buffer(l, &root->root_item,
1129 btrfs_item_ptr_offset(l, path->slots[0]),
0f7d52f4 1130 sizeof(root->root_item));
44b36eb2 1131 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4
CM		 1132 ret = 0;
1133out:
1134 btrfs_release_path(root, path);
1135 btrfs_free_path(path);
1136 if (ret) {
1137 kfree(root);
1138 return ERR_PTR(ret);
1139 }
84234f3a 1140 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1141 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1142 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1143 blocksize, generation);
5d4f98a2 1144 root->commit_root = btrfs_root_node(root);
0f7d52f4
CM		 1145 BUG_ON(!root->node);
1146insert:
e02119d5
CM		 1147 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1148 root->ref_cows = 1;
1149 ret = btrfs_find_highest_inode(root, &highest_inode);
1150 if (ret == 0) {
1151 root->highest_inode = highest_inode;
1152 root->last_inode_alloc = highest_inode;
1153 }
5eda7b5e
CM		 1154 }
1155 return root;
1156}
1157
dc17ff8f
CM		 1158struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1159 u64 root_objectid)
1160{
1161 struct btrfs_root *root;
1162
1163 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1164 return fs_info->tree_root;
1165 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1166 return fs_info->extent_root;
1167
1168 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1169 (unsigned long)root_objectid);
1170 return root;
1171}
1172
edbd8d4e
CM		 1173struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1174 struct btrfs_key *location)
5eda7b5e
CM		 1175{
1176 struct btrfs_root *root;
1177 int ret;
1178
edbd8d4e
CM		 1179 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1180 return fs_info->tree_root;
1181 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1182 return fs_info->extent_root;
8f18cf13
CM		 1183 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1184 return fs_info->chunk_root;
1185 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1186 return fs_info->dev_root;
0403e47e
YZ		 1187 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1188 return fs_info->csum_root;
edbd8d4e 1189
5eda7b5e
CM		 1190 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1191 (unsigned long)location->objectid);
1192 if (root)
1193 return root;
1194
e02119d5 1195 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1196 if (IS_ERR(root))
1197 return root;
3394e160
CM		 1198
1199 set_anon_super(&root->anon_super, NULL);
1200
2619ba1f
CM		 1201 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1202 (unsigned long)root->root_key.objectid,
0f7d52f4
CM		 1203 root);
1204 if (ret) {
5f39d397 1205 free_extent_buffer(root->node);
0f7d52f4
CM		 1206 kfree(root);
1207 return ERR_PTR(ret);
1208 }
c146afad
YZ		 1209 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1210 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 1211 root->root_key.objectid);
c146afad
YZ		 1212 BUG_ON(ret);
1213 btrfs_orphan_cleanup(root);
1214 }
edbd8d4e
CM		 1215 return root;
1216}
1217
1218struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1219 struct btrfs_key *location,
1220 const char *name, int namelen)
1221{
1222 struct btrfs_root *root;
1223 int ret;
1224
1225 root = btrfs_read_fs_root_no_name(fs_info, location);
1226 if (!root)
1227 return NULL;
58176a96 1228
4313b399
CM		 1229 if (root->in_sysfs)
1230 return root;
1231
58176a96
JB		 1232 ret = btrfs_set_root_name(root, name, namelen);
1233 if (ret) {
5f39d397 1234 free_extent_buffer(root->node);
58176a96
JB		 1235 kfree(root);
1236 return ERR_PTR(ret);
1237 }
ea9e8b11 1238#if 0
58176a96
JB		 1239 ret = btrfs_sysfs_add_root(root);
1240 if (ret) {
5f39d397 1241 free_extent_buffer(root->node);
58176a96
JB		 1242 kfree(root->name);
1243 kfree(root);
1244 return ERR_PTR(ret);
1245 }
ea9e8b11 1246#endif
4313b399 1247 root->in_sysfs = 1;
0f7d52f4
CM		 1248 return root;
1249}
04160088
CM		 1250
1251static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1252{
1253 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1254 int ret = 0;
04160088
CM		 1255 struct btrfs_device *device;
1256 struct backing_dev_info *bdi;
b7967db7 1257
c6e30871 1258 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1259 if (!device->bdev)
1260 continue;
04160088
CM		 1261 bdi = blk_get_backing_dev_info(device->bdev);
1262 if (bdi && bdi_congested(bdi, bdi_bits)) {
1263 ret = 1;
1264 break;
1265 }
1266 }
1267 return ret;
1268}
1269
38b66988
CM		 1270/*
1271 * this unplugs every device on the box, and it is only used when page
1272 * is null
1273 */
1274static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1275{
38b66988
CM		 1276 struct btrfs_device *device;
1277 struct btrfs_fs_info *info;
1278
1279 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1280 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1281 if (!device->bdev)
1282 continue;
1283
38b66988 1284 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1285 if (bdi->unplug_io_fn)
38b66988 1286 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1287 }
1288}
1289
b2950863 1290static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1291{
38b66988 1292 struct inode *inode;
f2d8d74d
CM		 1293 struct extent_map_tree *em_tree;
1294 struct extent_map *em;
bcbfce8a 1295 struct address_space *mapping;
38b66988
CM		 1296 u64 offset;
1297
bcbfce8a 1298 /* the generic O_DIRECT read code does this */
9f0ba5bd 1299 if (1 || !page) {
38b66988
CM		 1300 __unplug_io_fn(bdi, page);
1301 return;
1302 }
1303
bcbfce8a
CM		 1304 /*
1305 * page->mapping may change at any time. Get a consistent copy
1306 * and use that for everything below
1307 */
1308 smp_mb();
1309 mapping = page->mapping;
1310 if (!mapping)
1311 return;
1312
1313 inode = mapping->host;
240d5d48
CM		 1314
1315 /*
1316 * don't do the expensive searching for a small number of
1317 * devices
1318 */
1319 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1320 __unplug_io_fn(bdi, page);
1321 return;
1322 }
1323
38b66988 1324 offset = page_offset(page);
04160088 1325
f2d8d74d
CM		 1326 em_tree = &BTRFS_I(inode)->extent_tree;
1327 spin_lock(&em_tree->lock);
1328 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1329 spin_unlock(&em_tree->lock);
89642229
CM		 1330 if (!em) {
1331 __unplug_io_fn(bdi, page);
f2d8d74d 1332 return;
89642229 1333 }
f2d8d74d 1334
89642229
CM		 1335 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1336 free_extent_map(em);
1337 __unplug_io_fn(bdi, page);
1338 return;
1339 }
f2d8d74d
CM		 1340 offset = offset - em->start;
1341 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1342 em->block_start + offset, page);
1343 free_extent_map(em);
04160088
CM		 1344}
1345
1346static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1347{
1348 bdi_init(bdi);
4575c9cc 1349 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1350 bdi->state = 0;
1351 bdi->capabilities = default_backing_dev_info.capabilities;
1352 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1353 bdi->unplug_io_data = info;
1354 bdi->congested_fn = btrfs_congested_fn;
1355 bdi->congested_data = info;
1356 return 0;
1357}
1358
ce9adaa5
CM		 1359static int bio_ready_for_csum(struct bio *bio)
1360{
1361 u64 length = 0;
1362 u64 buf_len = 0;
1363 u64 start = 0;
1364 struct page *page;
1365 struct extent_io_tree *io_tree = NULL;
1366 struct btrfs_fs_info *info = NULL;
1367 struct bio_vec *bvec;
1368 int i;
1369 int ret;
1370
1371 bio_for_each_segment(bvec, bio, i) {
1372 page = bvec->bv_page;
1373 if (page->private == EXTENT_PAGE_PRIVATE) {
1374 length += bvec->bv_len;
1375 continue;
1376 }
1377 if (!page->private) {
1378 length += bvec->bv_len;
1379 continue;
1380 }
1381 length = bvec->bv_len;
1382 buf_len = page->private >> 2;
1383 start = page_offset(page) + bvec->bv_offset;
1384 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1385 info = BTRFS_I(page->mapping->host)->root->fs_info;
1386 }
1387 /* are we fully contained in this bio? */
1388 if (buf_len <= length)
1389 return 1;
1390
1391 ret = extent_range_uptodate(io_tree, start + length,
1392 start + buf_len - 1);
ce9adaa5
CM		 1393 return ret;
1394}
1395
8b712842
CM		 1396/*
1397 * called by the kthread helper functions to finally call the bio end_io
1398 * functions. This is where read checksum verification actually happens
1399 */
1400static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1401{
ce9adaa5 1402 struct bio *bio;
8b712842
CM		 1403 struct end_io_wq *end_io_wq;
1404 struct btrfs_fs_info *fs_info;
ce9adaa5 1405 int error;
ce9adaa5 1406
8b712842
CM		 1407 end_io_wq = container_of(work, struct end_io_wq, work);
1408 bio = end_io_wq->bio;
1409 fs_info = end_io_wq->info;
ce9adaa5 1410
cad321ad 1411 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1412 * be checksummed at once. This makes sure the entire block is in
1413 * ram and up to date before trying to verify things. For
1414 * blocksize <= pagesize, it is basically a noop
1415 */
cad321ad
CM		 1416 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1417 !bio_ready_for_csum(bio)) {
d20f7043 1418 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1419 &end_io_wq->work);
1420 return;
1421 }
1422 error = end_io_wq->error;
1423 bio->bi_private = end_io_wq->private;
1424 bio->bi_end_io = end_io_wq->end_io;
1425 kfree(end_io_wq);
8b712842 1426 bio_endio(bio, error);
44b8bd7e
CM		 1427}
1428
a74a4b97
CM		 1429static int cleaner_kthread(void *arg)
1430{
1431 struct btrfs_root *root = arg;
1432
1433 do {
1434 smp_mb();
1435 if (root->fs_info->closing)
1436 break;
1437
1438 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1439 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 1440 btrfs_clean_old_snapshots(root);
a74a4b97
CM		 1441 mutex_unlock(&root->fs_info->cleaner_mutex);
1442
1443 if (freezing(current)) {
1444 refrigerator();
1445 } else {
1446 smp_mb();
1447 if (root->fs_info->closing)
1448 break;
1449 set_current_state(TASK_INTERRUPTIBLE);
1450 schedule();
1451 __set_current_state(TASK_RUNNING);
1452 }
1453 } while (!kthread_should_stop());
1454 return 0;
1455}
1456
1457static int transaction_kthread(void *arg)
1458{
1459 struct btrfs_root *root = arg;
1460 struct btrfs_trans_handle *trans;
1461 struct btrfs_transaction *cur;
1462 unsigned long now;
1463 unsigned long delay;
1464 int ret;
1465
1466 do {
1467 smp_mb();
1468 if (root->fs_info->closing)
1469 break;
1470
1471 delay = HZ * 30;
1472 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1473 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1474
1475 mutex_lock(&root->fs_info->trans_mutex);
1476 cur = root->fs_info->running_transaction;
1477 if (!cur) {
1478 mutex_unlock(&root->fs_info->trans_mutex);
1479 goto sleep;
1480 }
31153d81 1481
a74a4b97
CM		 1482 now = get_seconds();
1483 if (now < cur->start_time || now - cur->start_time < 30) {
1484 mutex_unlock(&root->fs_info->trans_mutex);
1485 delay = HZ * 5;
1486 goto sleep;
1487 }
1488 mutex_unlock(&root->fs_info->trans_mutex);
a74a4b97
CM		 1489 trans = btrfs_start_transaction(root, 1);
1490 ret = btrfs_commit_transaction(trans, root);
56bec294 1491
a74a4b97
CM		 1492sleep:
1493 wake_up_process(root->fs_info->cleaner_kthread);
1494 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1495
1496 if (freezing(current)) {
1497 refrigerator();
1498 } else {
1499 if (root->fs_info->closing)
1500 break;
1501 set_current_state(TASK_INTERRUPTIBLE);
1502 schedule_timeout(delay);
1503 __set_current_state(TASK_RUNNING);
1504 }
1505 } while (!kthread_should_stop());
1506 return 0;
1507}
1508
8a4b83cc 1509struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1510 struct btrfs_fs_devices *fs_devices,
1511 char *options)
2e635a27 1512{
db94535d
CM		 1513 u32 sectorsize;
1514 u32 nodesize;
1515 u32 leafsize;
1516 u32 blocksize;
87ee04eb 1517 u32 stripesize;
84234f3a 1518 u64 generation;
f2b636e8 1519 u64 features;
3de4586c 1520 struct btrfs_key location;
a061fc8d 1521 struct buffer_head *bh;
e02119d5 1522 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1523 GFP_NOFS);
d20f7043
CM		 1524 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1525 GFP_NOFS);
e02119d5 1526 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1527 GFP_NOFS);
8790d502 1528 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
e20d96d6 1529 GFP_NOFS);
e02119d5 1530 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1531 GFP_NOFS);
e02119d5 1532 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1533 GFP_NOFS);
e02119d5
CM		 1534 struct btrfs_root *log_tree_root;
1535
eb60ceac 1536 int ret;
e58ca020 1537 int err = -EINVAL;
4543df7e 1538
2c90e5d6 1539 struct btrfs_super_block *disk_super;
8790d502 1540
0463bb4e 1541 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1542 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1543 err = -ENOMEM;
1544 goto fail;
1545 }
0f7d52f4 1546 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
8fd17795 1547 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1548 INIT_LIST_HEAD(&fs_info->dead_roots);
19c00ddc 1549 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1550 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1551 INIT_LIST_HEAD(&fs_info->ordered_operations);
1832a6d5 1552 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1553 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1554 spin_lock_init(&fs_info->ref_cache_lock);
19c00ddc 1555
58176a96 1556 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1557 fs_info->tree_root = tree_root;
1558 fs_info->extent_root = extent_root;
d20f7043 1559 fs_info->csum_root = csum_root;
0b86a832
CM		 1560 fs_info->chunk_root = chunk_root;
1561 fs_info->dev_root = dev_root;
8a4b83cc 1562 fs_info->fs_devices = fs_devices;
0b86a832 1563 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1564 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1565 btrfs_mapping_init(&fs_info->mapping_tree);
cb03c743 1566 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1567 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1568 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1569 atomic_set(&fs_info->nr_async_bios, 0);
817d52f8 1570 atomic_set(&fs_info->async_caching_threads, 0);
e20d96d6 1571 fs_info->sb = sb;
c59f8951 1572 fs_info->max_extent = (u64)-1;
6f568d35 1573 fs_info->max_inline = 8192 * 1024;
04160088 1574 setup_bdi(fs_info, &fs_info->bdi);
d98237b3
CM		 1575 fs_info->btree_inode = new_inode(sb);
1576 fs_info->btree_inode->i_ino = 1;
2c90e5d6 1577 fs_info->btree_inode->i_nlink = 1;
97e728d4 1578 fs_info->metadata_ratio = 8;
c8b97818 1579
b34b086c
CM		 1580 fs_info->thread_pool_size = min_t(unsigned long,
1581 num_online_cpus() + 2, 8);
0afbaf8c 1582
3eaa2885
CM		 1583 INIT_LIST_HEAD(&fs_info->ordered_extents);
1584 spin_lock_init(&fs_info->ordered_extent_lock);
1585
a061fc8d
CM		 1586 sb->s_blocksize = 4096;
1587 sb->s_blocksize_bits = blksize_bits(4096);
1588
0afbaf8c
CM		 1589 /*
1590 * we set the i_size on the btree inode to the max possible int.
1591 * the real end of the address space is determined by all of
1592 * the devices in the system
1593 */
1594 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1595 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1596 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1597
5d4f98a2 1598 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1599 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1600 fs_info->btree_inode->i_mapping,
1601 GFP_NOFS);
d1310b2e
CM		 1602 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1603 GFP_NOFS);
1604
1605 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1606
0f9dd46c
JB		 1607 spin_lock_init(&fs_info->block_group_cache_lock);
1608 fs_info->block_group_cache_tree.rb_node = NULL;
1609
d1310b2e 1610 extent_io_tree_init(&fs_info->pinned_extents,
1a5bc167 1611 fs_info->btree_inode->i_mapping, GFP_NOFS);
e66f709b 1612 fs_info->do_barriers = 1;
e18e4809 1613
0f7d52f4
CM		 1614 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1615 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1616 sizeof(struct btrfs_key));
22b0ebda 1617 insert_inode_hash(fs_info->btree_inode);
39279cc3 1618
79154b1b 1619 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1620 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1621 mutex_init(&fs_info->tree_log_mutex);
a2135011 1622 mutex_init(&fs_info->drop_mutex);
925baedd 1623 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1624 mutex_init(&fs_info->transaction_kthread_mutex);
1625 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1626 mutex_init(&fs_info->volume_mutex);
1a40e23b 1627 mutex_init(&fs_info->tree_reloc_mutex);
fa9c0d79
CM		 1628
1629 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1630 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1631
e6dcd2dc 1632 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1633 init_waitqueue_head(&fs_info->transaction_wait);
4854ddd0 1634 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1635
0b86a832 1636 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1637 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1638
d98237b3 1639
a512bbf8 1640 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
a061fc8d 1641 if (!bh)
39279cc3 1642 goto fail_iput;
39279cc3 1643
a061fc8d 1644 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1645 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1646 sizeof(fs_info->super_for_commit));
a061fc8d 1647 brelse(bh);
5f39d397 1648
a061fc8d 1649 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1650
5f39d397 1651 disk_super = &fs_info->super_copy;
0f7d52f4 1652 if (!btrfs_super_root(disk_super))
c6e2bac1 1653 goto fail_iput;
0f7d52f4 1654
2b82032c
YZ		 1655 ret = btrfs_parse_options(tree_root, options);
1656 if (ret) {
1657 err = ret;
c6e2bac1 1658 goto fail_iput;
2b82032c 1659 }
dfe25020 1660
f2b636e8
JB		 1661 features = btrfs_super_incompat_flags(disk_super) &
1662 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1663 if (features) {
1664 printk(KERN_ERR "BTRFS: couldn't mount because of "
1665 "unsupported optional features (%Lx).\n",
21380931 1666 (unsigned long long)features);
f2b636e8 1667 err = -EINVAL;
c6e2bac1 1668 goto fail_iput;
f2b636e8
JB		 1669 }
1670
5d4f98a2
YZ		 1671 features = btrfs_super_incompat_flags(disk_super);
1672 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1673 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1674 btrfs_set_super_incompat_flags(disk_super, features);
1675 }
1676
f2b636e8
JB		 1677 features = btrfs_super_compat_ro_flags(disk_super) &
1678 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1679 if (!(sb->s_flags & MS_RDONLY) && features) {
1680 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1681 "unsupported option features (%Lx).\n",
21380931 1682 (unsigned long long)features);
f2b636e8 1683 err = -EINVAL;
c6e2bac1 1684 goto fail_iput;
f2b636e8
JB		 1685 }
1686
4543df7e
CM		 1687 /*
1688 * we need to start all the end_io workers up front because the
1689 * queue work function gets called at interrupt time, and so it
1690 * cannot dynamically grow.
1691 */
5443be45
CM		 1692 btrfs_init_workers(&fs_info->workers, "worker",
1693 fs_info->thread_pool_size);
c8b97818 1694
771ed689
CM		 1695 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1696 fs_info->thread_pool_size);
1697
5443be45 1698 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209
CM		 1699 min_t(u64, fs_devices->num_devices,
1700 fs_info->thread_pool_size));
61b49440
CM		 1701
1702 /* a higher idle thresh on the submit workers makes it much more
1703 * likely that bios will be send down in a sane order to the
1704 * devices
1705 */
1706 fs_info->submit_workers.idle_thresh = 64;
53863232 1707
771ed689 1708 fs_info->workers.idle_thresh = 16;
4a69a410 1709 fs_info->workers.ordered = 1;
61b49440 1710
771ed689
CM		 1711 fs_info->delalloc_workers.idle_thresh = 2;
1712 fs_info->delalloc_workers.ordered = 1;
1713
5443be45
CM		 1714 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1715 btrfs_init_workers(&fs_info->endio_workers, "endio",
1716 fs_info->thread_pool_size);
d20f7043
CM		 1717 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1718 fs_info->thread_pool_size);
cad321ad
CM		 1719 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1720 "endio-meta-write", fs_info->thread_pool_size);
5443be45 1721 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
e6dcd2dc 1722 fs_info->thread_pool_size);
61b49440
CM		 1723
1724 /*
1725 * endios are largely parallel and should have a very
1726 * low idle thresh
1727 */
1728 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1729 fs_info->endio_meta_workers.idle_thresh = 4;
1730
3a5f1d45 1731 fs_info->endio_write_workers.idle_thresh = 64;
cad321ad 1732 fs_info->endio_meta_write_workers.idle_thresh = 64;
61b49440 1733
4543df7e 1734 btrfs_start_workers(&fs_info->workers, 1);
1cc127b5 1735 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1736 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1737 btrfs_start_workers(&fs_info->fixup_workers, 1);
4543df7e 1738 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
d20f7043
CM		 1739 btrfs_start_workers(&fs_info->endio_meta_workers,
1740 fs_info->thread_pool_size);
cad321ad
CM		 1741 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1742 fs_info->thread_pool_size);
e6dcd2dc
CM		 1743 btrfs_start_workers(&fs_info->endio_write_workers,
1744 fs_info->thread_pool_size);
4543df7e 1745
4575c9cc 1746 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1747 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1748 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1749
db94535d
CM		 1750 nodesize = btrfs_super_nodesize(disk_super);
1751 leafsize = btrfs_super_leafsize(disk_super);
1752 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1753 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1754 tree_root->nodesize = nodesize;
1755 tree_root->leafsize = leafsize;
1756 tree_root->sectorsize = sectorsize;
87ee04eb 1757 tree_root->stripesize = stripesize;
a061fc8d
CM		 1758
1759 sb->s_blocksize = sectorsize;
1760 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1761
39279cc3
CM		 1762 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1763 sizeof(disk_super->magic))) {
d397712b 1764 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1765 goto fail_sb_buffer;
1766 }
19c00ddc 1767
925baedd 1768 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1769 ret = btrfs_read_sys_array(tree_root);
925baedd 1770 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1771 if (ret) {
d397712b
CM		 1772 printk(KERN_WARNING "btrfs: failed to read the system "
1773 "array on %s\n", sb->s_id);
5d4f98a2 1774 goto fail_sb_buffer;
84eed90f 1775 }
0b86a832
CM		 1776
1777 blocksize = btrfs_level_size(tree_root,
1778 btrfs_super_chunk_root_level(disk_super));
84234f3a 1779 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1780
1781 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1782 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1783
1784 chunk_root->node = read_tree_block(chunk_root,
1785 btrfs_super_chunk_root(disk_super),
84234f3a 1786 blocksize, generation);
0b86a832 1787 BUG_ON(!chunk_root->node);
83121942
DW		 1788 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1789 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1790 sb->s_id);
1791 goto fail_chunk_root;
1792 }
5d4f98a2
YZ		 1793 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1794 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1795
e17cade2 1796 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1797 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1798 BTRFS_UUID_SIZE);
e17cade2 1799
925baedd 1800 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1801 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1802 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1803 if (ret) {
d397712b
CM		 1804 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1805 sb->s_id);
2b82032c
YZ		 1806 goto fail_chunk_root;
1807 }
0b86a832 1808
dfe25020
CM		 1809 btrfs_close_extra_devices(fs_devices);
1810
db94535d
CM		 1811 blocksize = btrfs_level_size(tree_root,
1812 btrfs_super_root_level(disk_super));
84234f3a 1813 generation = btrfs_super_generation(disk_super);
0b86a832 1814
e20d96d6 1815 tree_root->node = read_tree_block(tree_root,
db94535d 1816 btrfs_super_root(disk_super),
84234f3a 1817 blocksize, generation);
39279cc3 1818 if (!tree_root->node)
2b82032c 1819 goto fail_chunk_root;
83121942
DW		 1820 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1821 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1822 sb->s_id);
1823 goto fail_tree_root;
1824 }
5d4f98a2
YZ		 1825 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1826 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1827
1828 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1829 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1830 if (ret)
39279cc3 1831 goto fail_tree_root;
0b86a832
CM		 1832 extent_root->track_dirty = 1;
1833
1834 ret = find_and_setup_root(tree_root, fs_info,
1835 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1836 if (ret)
1837 goto fail_extent_root;
5d4f98a2 1838 dev_root->track_dirty = 1;
3768f368 1839
d20f7043
CM		 1840 ret = find_and_setup_root(tree_root, fs_info,
1841 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1842 if (ret)
5d4f98a2 1843 goto fail_dev_root;
d20f7043
CM		 1844
1845 csum_root->track_dirty = 1;
1846
9078a3e1
CM		 1847 btrfs_read_block_groups(extent_root);
1848
2d69a0f8 1849 fs_info->generation = generation;
c146afad 1850 fs_info->last_trans_committed = generation;
d18a2c44
CM		 1851 fs_info->data_alloc_profile = (u64)-1;
1852 fs_info->metadata_alloc_profile = (u64)-1;
1853 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
a74a4b97
CM		 1854 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1855 "btrfs-cleaner");
57506d50 1856 if (IS_ERR(fs_info->cleaner_kthread))
d20f7043 1857 goto fail_csum_root;
a74a4b97
CM		 1858
1859 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1860 tree_root,
1861 "btrfs-transaction");
57506d50 1862 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1863 goto fail_cleaner;
a74a4b97 1864
c289811c
CM		 1865 if (!btrfs_test_opt(tree_root, SSD) &&
1866 !btrfs_test_opt(tree_root, NOSSD) &&
1867 !fs_info->fs_devices->rotating) {
1868 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1869 "mode\n");
1870 btrfs_set_opt(fs_info->mount_opt, SSD);
1871 }
1872
e02119d5 1873 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 1874 u64 bytenr = btrfs_super_log_root(disk_super);
1875
7c2ca468 1876 if (fs_devices->rw_devices == 0) {
d397712b
CM		 1877 printk(KERN_WARNING "Btrfs log replay required "
1878 "on RO media\n");
7c2ca468
CM		 1879 err = -EIO;
1880 goto fail_trans_kthread;
1881 }
e02119d5
CM		 1882 blocksize =
1883 btrfs_level_size(tree_root,
1884 btrfs_super_log_root_level(disk_super));
d18a2c44 1885
e02119d5
CM		 1886 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1887 GFP_NOFS);
1888
1889 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1890 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1891
1892 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1893 blocksize,
1894 generation + 1);
e02119d5
CM		 1895 ret = btrfs_recover_log_trees(log_tree_root);
1896 BUG_ON(ret);
e556ce2c
YZ		 1897
1898 if (sb->s_flags & MS_RDONLY) {
1899 ret = btrfs_commit_super(tree_root);
1900 BUG_ON(ret);
1901 }
e02119d5 1902 }
1a40e23b 1903
7c2ca468 1904 if (!(sb->s_flags & MS_RDONLY)) {
5d4f98a2 1905 ret = btrfs_recover_relocation(tree_root);
7c2ca468
CM		 1906 BUG_ON(ret);
1907 }
1a40e23b 1908
3de4586c
CM		 1909 location.objectid = BTRFS_FS_TREE_OBJECTID;
1910 location.type = BTRFS_ROOT_ITEM_KEY;
1911 location.offset = (u64)-1;
1912
3de4586c
CM		 1913 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1914 if (!fs_info->fs_root)
7c2ca468 1915 goto fail_trans_kthread;
c289811c 1916
0f7d52f4 1917 return tree_root;
39279cc3 1918
7c2ca468
CM		 1919fail_trans_kthread:
1920 kthread_stop(fs_info->transaction_kthread);
3f157a2f 1921fail_cleaner:
a74a4b97 1922 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 1923
1924 /*
1925 * make sure we're done with the btree inode before we stop our
1926 * kthreads
1927 */
1928 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1929 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1930
d20f7043
CM		 1931fail_csum_root:
1932 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 1933 free_extent_buffer(csum_root->commit_root);
1934fail_dev_root:
1935 free_extent_buffer(dev_root->node);
1936 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 1937fail_extent_root:
1938 free_extent_buffer(extent_root->node);
5d4f98a2 1939 free_extent_buffer(extent_root->commit_root);
39279cc3 1940fail_tree_root:
5f39d397 1941 free_extent_buffer(tree_root->node);
5d4f98a2 1942 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 1943fail_chunk_root:
1944 free_extent_buffer(chunk_root->node);
5d4f98a2 1945 free_extent_buffer(chunk_root->commit_root);
39279cc3 1946fail_sb_buffer:
247e743c 1947 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 1948 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 1949 btrfs_stop_workers(&fs_info->workers);
1950 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 1951 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 1952 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 1953 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 1954 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 1955fail_iput:
7c2ca468 1956 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 1957 iput(fs_info->btree_inode);
7e662854 1958
dfe25020 1959 btrfs_close_devices(fs_info->fs_devices);
84eed90f 1960 btrfs_mapping_tree_free(&fs_info->mapping_tree);
7e662854 1961 bdi_destroy(&fs_info->bdi);
84eed90f 1962
7e662854 1963fail:
39279cc3
CM		 1964 kfree(extent_root);
1965 kfree(tree_root);
1966 kfree(fs_info);
83afeac4
JM		 1967 kfree(chunk_root);
1968 kfree(dev_root);
d20f7043 1969 kfree(csum_root);
39279cc3 1970 return ERR_PTR(err);
eb60ceac
CM		 1971}
1972
f2984462
CM		 1973static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1974{
1975 char b[BDEVNAME_SIZE];
1976
1977 if (uptodate) {
1978 set_buffer_uptodate(bh);
1979 } else {
1980 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1981 printk(KERN_WARNING "lost page write due to "
1982 "I/O error on %s\n",
1983 bdevname(bh->b_bdev, b));
1984 }
1259ab75
CM		 1985 /* note, we dont' set_buffer_write_io_error because we have
1986 * our own ways of dealing with the IO errors
1987 */
f2984462
CM		 1988 clear_buffer_uptodate(bh);
1989 }
1990 unlock_buffer(bh);
1991 put_bh(bh);
1992}
1993
a512bbf8
YZ		 1994struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1995{
1996 struct buffer_head *bh;
1997 struct buffer_head *latest = NULL;
1998 struct btrfs_super_block *super;
1999 int i;
2000 u64 transid = 0;
2001 u64 bytenr;
2002
2003 /* we would like to check all the supers, but that would make
2004 * a btrfs mount succeed after a mkfs from a different FS.
2005 * So, we need to add a special mount option to scan for
2006 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2007 */
2008 for (i = 0; i < 1; i++) {
2009 bytenr = btrfs_sb_offset(i);
2010 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2011 break;
2012 bh = __bread(bdev, bytenr / 4096, 4096);
2013 if (!bh)
2014 continue;
2015
2016 super = (struct btrfs_super_block *)bh->b_data;
2017 if (btrfs_super_bytenr(super) != bytenr ||
2018 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2019 sizeof(super->magic))) {
2020 brelse(bh);
2021 continue;
2022 }
2023
2024 if (!latest || btrfs_super_generation(super) > transid) {
2025 brelse(latest);
2026 latest = bh;
2027 transid = btrfs_super_generation(super);
2028 } else {
2029 brelse(bh);
2030 }
2031 }
2032 return latest;
2033}
2034
4eedeb75
HH		 2035/*
2036 * this should be called twice, once with wait == 0 and
2037 * once with wait == 1. When wait == 0 is done, all the buffer heads
2038 * we write are pinned.
2039 *
2040 * They are released when wait == 1 is done.
2041 * max_mirrors must be the same for both runs, and it indicates how
2042 * many supers on this one device should be written.
2043 *
2044 * max_mirrors == 0 means to write them all.
2045 */
a512bbf8
YZ		 2046static int write_dev_supers(struct btrfs_device *device,
2047 struct btrfs_super_block *sb,
2048 int do_barriers, int wait, int max_mirrors)
2049{
2050 struct buffer_head *bh;
2051 int i;
2052 int ret;
2053 int errors = 0;
2054 u32 crc;
2055 u64 bytenr;
2056 int last_barrier = 0;
2057
2058 if (max_mirrors == 0)
2059 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2060
2061 /* make sure only the last submit_bh does a barrier */
2062 if (do_barriers) {
2063 for (i = 0; i < max_mirrors; i++) {
2064 bytenr = btrfs_sb_offset(i);
2065 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2066 device->total_bytes)
2067 break;
2068 last_barrier = i;
2069 }
2070 }
2071
2072 for (i = 0; i < max_mirrors; i++) {
2073 bytenr = btrfs_sb_offset(i);
2074 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2075 break;
2076
2077 if (wait) {
2078 bh = __find_get_block(device->bdev, bytenr / 4096,
2079 BTRFS_SUPER_INFO_SIZE);
2080 BUG_ON(!bh);
a512bbf8 2081 wait_on_buffer(bh);
4eedeb75
HH		 2082 if (!buffer_uptodate(bh))
2083 errors++;
2084
2085 /* drop our reference */
2086 brelse(bh);
2087
2088 /* drop the reference from the wait == 0 run */
2089 brelse(bh);
2090 continue;
a512bbf8
YZ		 2091 } else {
2092 btrfs_set_super_bytenr(sb, bytenr);
2093
2094 crc = ~(u32)0;
2095 crc = btrfs_csum_data(NULL, (char *)sb +
2096 BTRFS_CSUM_SIZE, crc,
2097 BTRFS_SUPER_INFO_SIZE -
2098 BTRFS_CSUM_SIZE);
2099 btrfs_csum_final(crc, sb->csum);
2100
4eedeb75
HH		 2101 /*
2102 * one reference for us, and we leave it for the
2103 * caller
2104 */
a512bbf8
YZ		 2105 bh = __getblk(device->bdev, bytenr / 4096,
2106 BTRFS_SUPER_INFO_SIZE);
2107 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2108
4eedeb75 2109 /* one reference for submit_bh */
a512bbf8 2110 get_bh(bh);
4eedeb75
HH		 2111
2112 set_buffer_uptodate(bh);
a512bbf8
YZ		 2113 lock_buffer(bh);
2114 bh->b_end_io = btrfs_end_buffer_write_sync;
2115 }
2116
2117 if (i == last_barrier && do_barriers && device->barriers) {
2118 ret = submit_bh(WRITE_BARRIER, bh);
2119 if (ret == -EOPNOTSUPP) {
2120 printk("btrfs: disabling barriers on dev %s\n",
2121 device->name);
2122 set_buffer_uptodate(bh);
2123 device->barriers = 0;
4eedeb75 2124 /* one reference for submit_bh */
a512bbf8
YZ		 2125 get_bh(bh);
2126 lock_buffer(bh);
ffbd517d 2127 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2128 }
2129 } else {
ffbd517d 2130 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2131 }
2132
4eedeb75 2133 if (ret)
a512bbf8 2134 errors++;
a512bbf8
YZ		 2135 }
2136 return errors < i ? 0 : -1;
2137}
2138
2139int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2140{
e5e9a520 2141 struct list_head *head;
f2984462 2142 struct btrfs_device *dev;
a061fc8d 2143 struct btrfs_super_block *sb;
f2984462 2144 struct btrfs_dev_item *dev_item;
f2984462
CM		 2145 int ret;
2146 int do_barriers;
a236aed1
CM		 2147 int max_errors;
2148 int total_errors = 0;
a061fc8d 2149 u64 flags;
f2984462 2150
a236aed1 2151 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2152 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2153
a061fc8d
CM		 2154 sb = &root->fs_info->super_for_commit;
2155 dev_item = &sb->dev_item;
e5e9a520
CM		 2156
2157 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2158 head = &root->fs_info->fs_devices->devices;
c6e30871 2159 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2160 if (!dev->bdev) {
2161 total_errors++;
2162 continue;
2163 }
2b82032c 2164 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2165 continue;
2166
2b82032c 2167 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2168 btrfs_set_stack_device_type(dev_item, dev->type);
2169 btrfs_set_stack_device_id(dev_item, dev->devid);
2170 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2171 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2172 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2173 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2174 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2175 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2176 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2177
a061fc8d
CM		 2178 flags = btrfs_super_flags(sb);
2179 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2180
a512bbf8 2181 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2182 if (ret)
2183 total_errors++;
f2984462 2184 }
a236aed1 2185 if (total_errors > max_errors) {
d397712b
CM		 2186 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2187 total_errors);
a236aed1
CM		 2188 BUG();
2189 }
f2984462 2190
a512bbf8 2191 total_errors = 0;
c6e30871 2192 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2193 if (!dev->bdev)
2194 continue;
2b82032c 2195 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2196 continue;
2197
a512bbf8
YZ		 2198 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2199 if (ret)
2200 total_errors++;
f2984462 2201 }
e5e9a520 2202 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2203 if (total_errors > max_errors) {
d397712b
CM		 2204 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2205 total_errors);
a236aed1
CM		 2206 BUG();
2207 }
f2984462
CM		 2208 return 0;
2209}
2210
a512bbf8
YZ		 2211int write_ctree_super(struct btrfs_trans_handle *trans,
2212 struct btrfs_root *root, int max_mirrors)
eb60ceac 2213{
e66f709b 2214 int ret;
5f39d397 2215
a512bbf8 2216 ret = write_all_supers(root, max_mirrors);
5f39d397 2217 return ret;
cfaa7295
CM		 2218}
2219
5eda7b5e 2220int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2221{
5d4f98a2 2222 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2619ba1f
CM		 2223 radix_tree_delete(&fs_info->fs_roots_radix,
2224 (unsigned long)root->root_key.objectid);
3394e160
CM		 2225 if (root->anon_super.s_dev) {
2226 down_write(&root->anon_super.s_umount);
2227 kill_anon_super(&root->anon_super);
2228 }
2619ba1f 2229 if (root->node)
5f39d397 2230 free_extent_buffer(root->node);
2619ba1f 2231 if (root->commit_root)
5f39d397 2232 free_extent_buffer(root->commit_root);
d397712b 2233 kfree(root->name);
2619ba1f
CM		 2234 kfree(root);
2235 return 0;
2236}
2237
35b7e476 2238static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2239{
2240 int ret;
2241 struct btrfs_root *gang[8];
2242 int i;
2243
d397712b 2244 while (1) {
0f7d52f4
CM		 2245 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2246 (void **)gang, 0,
2247 ARRAY_SIZE(gang));
2248 if (!ret)
2249 break;
2619ba1f 2250 for (i = 0; i < ret; i++)
5eda7b5e 2251 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2252 }
2253 return 0;
2254}
b4100d64 2255
c146afad 2256int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2257{
c146afad
YZ		 2258 u64 root_objectid = 0;
2259 struct btrfs_root *gang[8];
2260 int i;
3768f368 2261 int ret;
e089f05c 2262
c146afad
YZ		 2263 while (1) {
2264 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2265 (void **)gang, root_objectid,
2266 ARRAY_SIZE(gang));
2267 if (!ret)
2268 break;
5d4f98a2
YZ		 2269
2270 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2271 for (i = 0; i < ret; i++) {
2272 root_objectid = gang[i]->root_key.objectid;
2273 ret = btrfs_find_dead_roots(fs_info->tree_root,
5d4f98a2 2274 root_objectid);
c146afad
YZ		 2275 BUG_ON(ret);
2276 btrfs_orphan_cleanup(gang[i]);
2277 }
2278 root_objectid++;
2279 }
2280 return 0;
2281}
a2135011 2282
c146afad
YZ		 2283int btrfs_commit_super(struct btrfs_root *root)
2284{
2285 struct btrfs_trans_handle *trans;
2286 int ret;
a74a4b97 2287
c146afad 2288 mutex_lock(&root->fs_info->cleaner_mutex);
a74a4b97 2289 btrfs_clean_old_snapshots(root);
c146afad 2290 mutex_unlock(&root->fs_info->cleaner_mutex);
79154b1b 2291 trans = btrfs_start_transaction(root, 1);
54aa1f4d 2292 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2293 BUG_ON(ret);
2294 /* run commit again to drop the original snapshot */
79154b1b
CM		 2295 trans = btrfs_start_transaction(root, 1);
2296 btrfs_commit_transaction(trans, root);
2297 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2298 BUG_ON(ret);
d6bfde87 2299
a512bbf8 2300 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2301 return ret;
2302}
2303
2304int close_ctree(struct btrfs_root *root)
2305{
2306 struct btrfs_fs_info *fs_info = root->fs_info;
2307 int ret;
2308
2309 fs_info->closing = 1;
2310 smp_mb();
2311
2312 kthread_stop(root->fs_info->transaction_kthread);
2313 kthread_stop(root->fs_info->cleaner_kthread);
2314
2315 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2316 ret = btrfs_commit_super(root);
d397712b
CM		 2317 if (ret)
2318 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2319 }
0f7d52f4 2320
b0c68f8b 2321 if (fs_info->delalloc_bytes) {
d397712b 2322 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2323 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2324 }
31153d81 2325 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2326 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2327 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2328 }
bcc63abb 2329
5d4f98a2
YZ		 2330 free_extent_buffer(fs_info->extent_root->node);
2331 free_extent_buffer(fs_info->extent_root->commit_root);
2332 free_extent_buffer(fs_info->tree_root->node);
2333 free_extent_buffer(fs_info->tree_root->commit_root);
2334 free_extent_buffer(root->fs_info->chunk_root->node);
2335 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2336 free_extent_buffer(root->fs_info->dev_root->node);
2337 free_extent_buffer(root->fs_info->dev_root->commit_root);
2338 free_extent_buffer(root->fs_info->csum_root->node);
2339 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2340
9078a3e1 2341 btrfs_free_block_groups(root->fs_info);
817d52f8 2342 btrfs_free_super_mirror_extents(root->fs_info);
d10c5f31 2343
c146afad 2344 del_fs_roots(fs_info);
d10c5f31 2345
c146afad 2346 iput(fs_info->btree_inode);
9ad6b7bc 2347
247e743c 2348 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2349 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2350 btrfs_stop_workers(&fs_info->workers);
2351 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2352 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2353 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2354 btrfs_stop_workers(&fs_info->endio_write_workers);
1cc127b5 2355 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2356
dfe25020 2357 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2358 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2359
04160088 2360 bdi_destroy(&fs_info->bdi);
0b86a832 2361
0f7d52f4 2362 kfree(fs_info->extent_root);
0f7d52f4 2363 kfree(fs_info->tree_root);
0b86a832
CM		 2364 kfree(fs_info->chunk_root);
2365 kfree(fs_info->dev_root);
d20f7043 2366 kfree(fs_info->csum_root);
eb60ceac
CM		 2367 return 0;
2368}
2369
1259ab75 2370int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2371{
1259ab75 2372 int ret;
810191ff 2373 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75
CM		 2374
2375 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2376 if (!ret)
2377 return ret;
2378
2379 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2380 parent_transid);
2381 return !ret;
5f39d397
CM		 2382}
2383
2384int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2385{
810191ff 2386 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2387 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2388 buf);
2389}
6702ed49 2390
5f39d397
CM		 2391void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2392{
810191ff 2393 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2394 u64 transid = btrfs_header_generation(buf);
2395 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2396 int was_dirty;
b4ce94de 2397
b9447ef8 2398 btrfs_assert_tree_locked(buf);
ccd467d6 2399 if (transid != root->fs_info->generation) {
d397712b
CM		 2400 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2401 "found %llu running %llu\n",
db94535d 2402 (unsigned long long)buf->start,
d397712b
CM		 2403 (unsigned long long)transid,
2404 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2405 WARN_ON(1);
2406 }
b9473439
CM		 2407 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2408 buf);
2409 if (!was_dirty) {
2410 spin_lock(&root->fs_info->delalloc_lock);
2411 root->fs_info->dirty_metadata_bytes += buf->len;
2412 spin_unlock(&root->fs_info->delalloc_lock);
2413 }
eb60ceac
CM		 2414}
2415
d3c2fdcf 2416void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2417{
188de649
CM		 2418 /*
2419 * looks as though older kernels can get into trouble with
2420 * this code, they end up stuck in balance_dirty_pages forever
2421 */
d6bfde87 2422 u64 num_dirty;
771ed689 2423 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2424
6933c02e 2425 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2426 return;
2427
585ad2c3
CM		 2428 num_dirty = root->fs_info->dirty_metadata_bytes;
2429
d6bfde87
CM		 2430 if (num_dirty > thresh) {
2431 balance_dirty_pages_ratelimited_nr(
d7fc640e 2432 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2433 }
188de649 2434 return;
35b7e476 2435}
6b80053d 2436
ca7a79ad 2437int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2438{
810191ff 2439 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2440 int ret;
ca7a79ad 2441 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2442 if (ret == 0)
b4ce94de 2443 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2444 return ret;
6b80053d 2445}
0da5468f 2446
4bef0848
CM		 2447int btree_lock_page_hook(struct page *page)
2448{
2449 struct inode *inode = page->mapping->host;
b9473439 2450 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2451 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2452 struct extent_buffer *eb;
2453 unsigned long len;
2454 u64 bytenr = page_offset(page);
2455
2456 if (page->private == EXTENT_PAGE_PRIVATE)
2457 goto out;
2458
2459 len = page->private >> 2;
2460 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2461 if (!eb)
2462 goto out;
2463
2464 btrfs_tree_lock(eb);
4bef0848 2465 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2466
2467 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2468 spin_lock(&root->fs_info->delalloc_lock);
2469 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2470 root->fs_info->dirty_metadata_bytes -= eb->len;
2471 else
2472 WARN_ON(1);
2473 spin_unlock(&root->fs_info->delalloc_lock);
2474 }
2475
4bef0848
CM		 2476 btrfs_tree_unlock(eb);
2477 free_extent_buffer(eb);
2478out:
2479 lock_page(page);
2480 return 0;
2481}
2482
d1310b2e 2483static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2484 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2485 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2486 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2487 /* note we're sharing with inode.c for the merge bio hook */
2488 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2489};
Linux 6.x block layer and io_uring tree(s)