projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: deal with DIO bios that span more than one ordered extent
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
4b4e25f2 32#include "compat.h"
eb60ceac
CM		 33#include "ctree.h"
34#include "disk-io.h"
e089f05c 35#include "transaction.h"
0f7d52f4 36#include "btrfs_inode.h"
0b86a832 37#include "volumes.h"
db94535d 38#include "print-tree.h"
8b712842 39#include "async-thread.h"
925baedd 40#include "locking.h"
e02119d5 41#include "tree-log.h"
fa9c0d79 42#include "free-space-cache.h"
eb60ceac 43
d1310b2e 44static struct extent_io_ops btree_extent_io_ops;
8b712842 45static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 46static void free_fs_root(struct btrfs_root *root);
ce9adaa5 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;
eaf25d93
CM		 78 /*
79 * bio_offset is optional, can be used if the pages in the bio
80 * can't tell us where in the file the bio should go
81 */
82 u64 bio_offset;
8b712842 83 struct btrfs_work work;
44b8bd7e
CM		 84};
85
4008c04a
CM		 86/* These are used to set the lockdep class on the extent buffer locks.
87 * The class is set by the readpage_end_io_hook after the buffer has
88 * passed csum validation but before the pages are unlocked.
89 *
90 * The lockdep class is also set by btrfs_init_new_buffer on freshly
91 * allocated blocks.
92 *
93 * The class is based on the level in the tree block, which allows lockdep
94 * to know that lower nodes nest inside the locks of higher nodes.
95 *
96 * We also add a check to make sure the highest level of the tree is
97 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
98 * code needs update as well.
99 */
100#ifdef CONFIG_DEBUG_LOCK_ALLOC
101# if BTRFS_MAX_LEVEL != 8
102# error
103# endif
104static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
105static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
106 /* leaf */
107 "btrfs-extent-00",
108 "btrfs-extent-01",
109 "btrfs-extent-02",
110 "btrfs-extent-03",
111 "btrfs-extent-04",
112 "btrfs-extent-05",
113 "btrfs-extent-06",
114 "btrfs-extent-07",
115 /* highest possible level */
116 "btrfs-extent-08",
117};
118#endif
119
d352ac68
CM		 120/*
121 * extents on the btree inode are pretty simple, there's one extent
122 * that covers the entire device
123 */
b2950863
CH		 124static struct extent_map *btree_get_extent(struct inode *inode,
125 struct page *page, size_t page_offset, u64 start, u64 len,
126 int create)
7eccb903 127{
5f39d397
CM		 128 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
129 struct extent_map *em;
130 int ret;
131
890871be 132 read_lock(&em_tree->lock);
d1310b2e 133 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 134 if (em) {
135 em->bdev =
136 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 137 read_unlock(&em_tree->lock);
5f39d397 138 goto out;
a061fc8d 139 }
890871be 140 read_unlock(&em_tree->lock);
7b13b7b1 141
5f39d397
CM		 142 em = alloc_extent_map(GFP_NOFS);
143 if (!em) {
144 em = ERR_PTR(-ENOMEM);
145 goto out;
146 }
147 em->start = 0;
0afbaf8c 148 em->len = (u64)-1;
c8b97818 149 em->block_len = (u64)-1;
5f39d397 150 em->block_start = 0;
a061fc8d 151 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 152
890871be 153 write_lock(&em_tree->lock);
5f39d397
CM		 154 ret = add_extent_mapping(em_tree, em);
155 if (ret == -EEXIST) {
0afbaf8c
CM		 156 u64 failed_start = em->start;
157 u64 failed_len = em->len;
158
5f39d397 159 free_extent_map(em);
7b13b7b1 160 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 161 if (em) {
7b13b7b1 162 ret = 0;
0afbaf8c
CM		 163 } else {
164 em = lookup_extent_mapping(em_tree, failed_start,
165 failed_len);
7b13b7b1 166 ret = -EIO;
0afbaf8c 167 }
5f39d397 168 } else if (ret) {
7b13b7b1
CM		 169 free_extent_map(em);
170 em = NULL;
5f39d397 171 }
890871be 172 write_unlock(&em_tree->lock);
7b13b7b1
CM		 173
174 if (ret)
175 em = ERR_PTR(ret);
5f39d397
CM		 176out:
177 return em;
7eccb903
CM		 178}
179
19c00ddc
CM		 180u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
181{
163e783e 182 return crc32c(seed, data, len);
19c00ddc
CM		 183}
184
185void btrfs_csum_final(u32 crc, char *result)
186{
187 *(__le32 *)result = ~cpu_to_le32(crc);
188}
189
d352ac68
CM		 190/*
191 * compute the csum for a btree block, and either verify it or write it
192 * into the csum field of the block.
193 */
19c00ddc
CM		 194static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
195 int verify)
196{
607d432d
JB		 197 u16 csum_size =
198 btrfs_super_csum_size(&root->fs_info->super_copy);
199 char *result = NULL;
19c00ddc
CM		 200 unsigned long len;
201 unsigned long cur_len;
202 unsigned long offset = BTRFS_CSUM_SIZE;
203 char *map_token = NULL;
204 char *kaddr;
205 unsigned long map_start;
206 unsigned long map_len;
207 int err;
208 u32 crc = ~(u32)0;
607d432d 209 unsigned long inline_result;
19c00ddc
CM		 210
211 len = buf->len - offset;
d397712b 212 while (len > 0) {
19c00ddc
CM		 213 err = map_private_extent_buffer(buf, offset, 32,
214 &map_token, &kaddr,
215 &map_start, &map_len, KM_USER0);
d397712b 216 if (err)
19c00ddc 217 return 1;
19c00ddc
CM		 218 cur_len = min(len, map_len - (offset - map_start));
219 crc = btrfs_csum_data(root, kaddr + offset - map_start,
220 crc, cur_len);
221 len -= cur_len;
222 offset += cur_len;
223 unmap_extent_buffer(buf, map_token, KM_USER0);
224 }
607d432d
JB		 225 if (csum_size > sizeof(inline_result)) {
226 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
227 if (!result)
228 return 1;
229 } else {
230 result = (char *)&inline_result;
231 }
232
19c00ddc
CM		 233 btrfs_csum_final(crc, result);
234
235 if (verify) {
607d432d 236 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 237 u32 val;
238 u32 found = 0;
607d432d 239 memcpy(&found, result, csum_size);
e4204ded 240
607d432d 241 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 242 if (printk_ratelimit()) {
243 printk(KERN_INFO "btrfs: %s checksum verify "
244 "failed on %llu wanted %X found %X "
245 "level %d\n",
246 root->fs_info->sb->s_id,
247 (unsigned long long)buf->start, val, found,
248 btrfs_header_level(buf));
249 }
607d432d
JB		 250 if (result != (char *)&inline_result)
251 kfree(result);
19c00ddc
CM		 252 return 1;
253 }
254 } else {
607d432d 255 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 256 }
607d432d
JB		 257 if (result != (char *)&inline_result)
258 kfree(result);
19c00ddc
CM		 259 return 0;
260}
261
d352ac68
CM		 262/*
263 * we can't consider a given block up to date unless the transid of the
264 * block matches the transid in the parent node's pointer. This is how we
265 * detect blocks that either didn't get written at all or got written
266 * in the wrong place.
267 */
1259ab75
CM		 268static int verify_parent_transid(struct extent_io_tree *io_tree,
269 struct extent_buffer *eb, u64 parent_transid)
270{
2ac55d41 271 struct extent_state *cached_state = NULL;
1259ab75
CM		 272 int ret;
273
274 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
275 return 0;
276
2ac55d41
JB		 277 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
278 0, &cached_state, GFP_NOFS);
279 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 280 btrfs_header_generation(eb) == parent_transid) {
281 ret = 0;
282 goto out;
283 }
193f284d
CM		 284 if (printk_ratelimit()) {
285 printk("parent transid verify failed on %llu wanted %llu "
286 "found %llu\n",
287 (unsigned long long)eb->start,
288 (unsigned long long)parent_transid,
289 (unsigned long long)btrfs_header_generation(eb));
290 }
1259ab75 291 ret = 1;
2ac55d41 292 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 293out:
2ac55d41
JB		 294 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
295 &cached_state, GFP_NOFS);
1259ab75 296 return ret;
1259ab75
CM		 297}
298
d352ac68
CM		 299/*
300 * helper to read a given tree block, doing retries as required when
301 * the checksums don't match and we have alternate mirrors to try.
302 */
f188591e
CM		 303static int btree_read_extent_buffer_pages(struct btrfs_root *root,
304 struct extent_buffer *eb,
ca7a79ad 305 u64 start, u64 parent_transid)
f188591e
CM		 306{
307 struct extent_io_tree *io_tree;
308 int ret;
309 int num_copies = 0;
310 int mirror_num = 0;
311
312 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
313 while (1) {
314 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
315 btree_get_extent, mirror_num);
1259ab75
CM		 316 if (!ret &&
317 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 318 return ret;
d397712b 319
f188591e
CM		 320 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
321 eb->start, eb->len);
4235298e 322 if (num_copies == 1)
f188591e 323 return ret;
4235298e 324
f188591e 325 mirror_num++;
4235298e 326 if (mirror_num > num_copies)
f188591e 327 return ret;
f188591e 328 }
f188591e
CM		 329 return -EIO;
330}
19c00ddc 331
d352ac68 332/*
d397712b
CM		 333 * checksum a dirty tree block before IO. This has extra checks to make sure
334 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 335 */
d397712b 336
b2950863 337static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 338{
d1310b2e 339 struct extent_io_tree *tree;
35ebb934 340 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc 341 u64 found_start;
19c00ddc
CM		 342 unsigned long len;
343 struct extent_buffer *eb;
f188591e
CM		 344 int ret;
345
d1310b2e 346 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc
CM		 347
348 if (page->private == EXTENT_PAGE_PRIVATE)
349 goto out;
350 if (!page->private)
351 goto out;
352 len = page->private >> 2;
d397712b
CM		 353 WARN_ON(len == 0);
354
19c00ddc 355 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
ca7a79ad
CM		 356 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
357 btrfs_header_generation(eb));
f188591e 358 BUG_ON(ret);
784b4e29
CM		 359 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
360
19c00ddc
CM		 361 found_start = btrfs_header_bytenr(eb);
362 if (found_start != start) {
55c69072
CM		 363 WARN_ON(1);
364 goto err;
365 }
366 if (eb->first_page != page) {
55c69072
CM		 367 WARN_ON(1);
368 goto err;
369 }
370 if (!PageUptodate(page)) {
55c69072
CM		 371 WARN_ON(1);
372 goto err;
19c00ddc 373 }
19c00ddc 374 csum_tree_block(root, eb, 0);
55c69072 375err:
19c00ddc
CM		 376 free_extent_buffer(eb);
377out:
378 return 0;
379}
380
2b82032c
YZ		 381static int check_tree_block_fsid(struct btrfs_root *root,
382 struct extent_buffer *eb)
383{
384 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
385 u8 fsid[BTRFS_UUID_SIZE];
386 int ret = 1;
387
388 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
389 BTRFS_FSID_SIZE);
390 while (fs_devices) {
391 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
392 ret = 0;
393 break;
394 }
395 fs_devices = fs_devices->seed;
396 }
397 return ret;
398}
399
4008c04a
CM		 400#ifdef CONFIG_DEBUG_LOCK_ALLOC
401void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
402{
403 lockdep_set_class_and_name(&eb->lock,
404 &btrfs_eb_class[level],
405 btrfs_eb_name[level]);
406}
407#endif
408
b2950863 409static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 410 struct extent_state *state)
411{
412 struct extent_io_tree *tree;
413 u64 found_start;
414 int found_level;
415 unsigned long len;
416 struct extent_buffer *eb;
417 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 418 int ret = 0;
ce9adaa5
CM		 419
420 tree = &BTRFS_I(page->mapping->host)->io_tree;
421 if (page->private == EXTENT_PAGE_PRIVATE)
422 goto out;
423 if (!page->private)
424 goto out;
d397712b 425
ce9adaa5 426 len = page->private >> 2;
d397712b
CM		 427 WARN_ON(len == 0);
428
ce9adaa5 429 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
f188591e 430
ce9adaa5 431 found_start = btrfs_header_bytenr(eb);
23a07867 432 if (found_start != start) {
193f284d
CM		 433 if (printk_ratelimit()) {
434 printk(KERN_INFO "btrfs bad tree block start "
435 "%llu %llu\n",
436 (unsigned long long)found_start,
437 (unsigned long long)eb->start);
438 }
f188591e 439 ret = -EIO;
ce9adaa5
CM		 440 goto err;
441 }
442 if (eb->first_page != page) {
d397712b
CM		 443 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
444 eb->first_page->index, page->index);
ce9adaa5 445 WARN_ON(1);
f188591e 446 ret = -EIO;
ce9adaa5
CM		 447 goto err;
448 }
2b82032c 449 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 450 if (printk_ratelimit()) {
451 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
452 (unsigned long long)eb->start);
453 }
1259ab75
CM		 454 ret = -EIO;
455 goto err;
456 }
ce9adaa5
CM		 457 found_level = btrfs_header_level(eb);
458
4008c04a
CM		 459 btrfs_set_buffer_lockdep_class(eb, found_level);
460
ce9adaa5 461 ret = csum_tree_block(root, eb, 1);
f188591e
CM		 462 if (ret)
463 ret = -EIO;
ce9adaa5
CM		 464
465 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
466 end = eb->start + end - 1;
ce9adaa5
CM		 467err:
468 free_extent_buffer(eb);
469out:
f188591e 470 return ret;
ce9adaa5
CM		 471}
472
ce9adaa5 473static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 474{
475 struct end_io_wq *end_io_wq = bio->bi_private;
476 struct btrfs_fs_info *fs_info;
ce9adaa5 477
ce9adaa5 478 fs_info = end_io_wq->info;
ce9adaa5 479 end_io_wq->error = err;
8b712842
CM		 480 end_io_wq->work.func = end_workqueue_fn;
481 end_io_wq->work.flags = 0;
d20f7043 482
7b6d91da 483 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 484 if (end_io_wq->metadata == 1)
cad321ad
CM		 485 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
486 &end_io_wq->work);
0cb59c99
JB		 487 else if (end_io_wq->metadata == 2)
488 btrfs_queue_worker(&fs_info->endio_freespace_worker,
489 &end_io_wq->work);
cad321ad
CM		 490 else
491 btrfs_queue_worker(&fs_info->endio_write_workers,
492 &end_io_wq->work);
d20f7043
CM		 493 } else {
494 if (end_io_wq->metadata)
495 btrfs_queue_worker(&fs_info->endio_meta_workers,
496 &end_io_wq->work);
497 else
498 btrfs_queue_worker(&fs_info->endio_workers,
499 &end_io_wq->work);
500 }
ce9adaa5
CM		 501}
502
0cb59c99
JB		 503/*
504 * For the metadata arg you want
505 *
506 * 0 - if data
507 * 1 - if normal metadta
508 * 2 - if writing to the free space cache area
509 */
22c59948
CM		 510int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
511 int metadata)
0b86a832 512{
ce9adaa5 513 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 514 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
515 if (!end_io_wq)
516 return -ENOMEM;
517
518 end_io_wq->private = bio->bi_private;
519 end_io_wq->end_io = bio->bi_end_io;
22c59948 520 end_io_wq->info = info;
ce9adaa5
CM		 521 end_io_wq->error = 0;
522 end_io_wq->bio = bio;
22c59948 523 end_io_wq->metadata = metadata;
ce9adaa5
CM		 524
525 bio->bi_private = end_io_wq;
526 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 527 return 0;
528}
529
b64a2851 530unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 531{
4854ddd0
CM		 532 unsigned long limit = min_t(unsigned long,
533 info->workers.max_workers,
534 info->fs_devices->open_devices);
535 return 256 * limit;
536}
0986fe9e 537
4854ddd0
CM		 538int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
539{
b64a2851
CM		 540 return atomic_read(&info->nr_async_bios) >
541 btrfs_async_submit_limit(info);
0986fe9e
CM		 542}
543
4a69a410
CM		 544static void run_one_async_start(struct btrfs_work *work)
545{
4a69a410
CM		 546 struct async_submit_bio *async;
547
548 async = container_of(work, struct async_submit_bio, work);
4a69a410 549 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM		 550 async->mirror_num, async->bio_flags,
551 async->bio_offset);
4a69a410
CM		 552}
553
554static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 555{
556 struct btrfs_fs_info *fs_info;
557 struct async_submit_bio *async;
4854ddd0 558 int limit;
8b712842
CM		 559
560 async = container_of(work, struct async_submit_bio, work);
561 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 562
b64a2851 563 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 564 limit = limit * 2 / 3;
565
8b712842 566 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 567
b64a2851
CM		 568 if (atomic_read(&fs_info->nr_async_submits) < limit &&
569 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 570 wake_up(&fs_info->async_submit_wait);
571
4a69a410 572 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 573 async->mirror_num, async->bio_flags,
574 async->bio_offset);
4a69a410
CM		 575}
576
577static void run_one_async_free(struct btrfs_work *work)
578{
579 struct async_submit_bio *async;
580
581 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 582 kfree(async);
583}
584
44b8bd7e
CM		 585int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
586 int rw, struct bio *bio, int mirror_num,
c8b97818 587 unsigned long bio_flags,
eaf25d93 588 u64 bio_offset,
4a69a410
CM		 589 extent_submit_bio_hook_t *submit_bio_start,
590 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 591{
592 struct async_submit_bio *async;
593
594 async = kmalloc(sizeof(*async), GFP_NOFS);
595 if (!async)
596 return -ENOMEM;
597
598 async->inode = inode;
599 async->rw = rw;
600 async->bio = bio;
601 async->mirror_num = mirror_num;
4a69a410
CM		 602 async->submit_bio_start = submit_bio_start;
603 async->submit_bio_done = submit_bio_done;
604
605 async->work.func = run_one_async_start;
606 async->work.ordered_func = run_one_async_done;
607 async->work.ordered_free = run_one_async_free;
608
8b712842 609 async->work.flags = 0;
c8b97818 610 async->bio_flags = bio_flags;
eaf25d93 611 async->bio_offset = bio_offset;
8c8bee1d 612
cb03c743 613 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 614
7b6d91da 615 if (rw & REQ_SYNC)
d313d7a3
CM		 616 btrfs_set_work_high_prio(&async->work);
617
8b712842 618 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 619
d397712b 620 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 621 atomic_read(&fs_info->nr_async_submits)) {
622 wait_event(fs_info->async_submit_wait,
623 (atomic_read(&fs_info->nr_async_submits) == 0));
624 }
625
44b8bd7e
CM		 626 return 0;
627}
628
ce3ed71a
CM		 629static int btree_csum_one_bio(struct bio *bio)
630{
631 struct bio_vec *bvec = bio->bi_io_vec;
632 int bio_index = 0;
633 struct btrfs_root *root;
634
635 WARN_ON(bio->bi_vcnt <= 0);
d397712b 636 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 637 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
638 csum_dirty_buffer(root, bvec->bv_page);
639 bio_index++;
640 bvec++;
641 }
642 return 0;
643}
644
4a69a410
CM		 645static int __btree_submit_bio_start(struct inode *inode, int rw,
646 struct bio *bio, int mirror_num,
eaf25d93
CM		 647 unsigned long bio_flags,
648 u64 bio_offset)
22c59948 649{
8b712842
CM		 650 /*
651 * when we're called for a write, we're already in the async
5443be45 652 * submission context. Just jump into btrfs_map_bio
8b712842 653 */
4a69a410
CM		 654 btree_csum_one_bio(bio);
655 return 0;
656}
22c59948 657
4a69a410 658static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 659 int mirror_num, unsigned long bio_flags,
660 u64 bio_offset)
4a69a410 661{
8b712842 662 /*
4a69a410
CM		 663 * when we're called for a write, we're already in the async
664 * submission context. Just jump into btrfs_map_bio
8b712842 665 */
8b712842 666 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 667}
668
44b8bd7e 669static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 670 int mirror_num, unsigned long bio_flags,
671 u64 bio_offset)
44b8bd7e 672{
cad321ad
CM		 673 int ret;
674
675 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
676 bio, 1);
677 BUG_ON(ret);
678
7b6d91da 679 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 680 /*
681 * called for a read, do the setup so that checksum validation
682 * can happen in the async kernel threads
683 */
4a69a410 684 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 685 mirror_num, 0);
44b8bd7e 686 }
d313d7a3 687
cad321ad
CM		 688 /*
689 * kthread helpers are used to submit writes so that checksumming
690 * can happen in parallel across all CPUs
691 */
44b8bd7e 692 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 693 inode, rw, bio, mirror_num, 0,
eaf25d93 694 bio_offset,
4a69a410
CM		 695 __btree_submit_bio_start,
696 __btree_submit_bio_done);
44b8bd7e
CM		 697}
698
784b4e29
CM		 699static int btree_migratepage(struct address_space *mapping,
700 struct page *newpage, struct page *page)
701{
702 /*
703 * we can't safely write a btree page from here,
704 * we haven't done the locking hook
705 */
706 if (PageDirty(page))
707 return -EAGAIN;
708 /*
709 * Buffers may be managed in a filesystem specific way.
710 * We must have no buffers or drop them.
711 */
712 if (page_has_private(page) &&
713 !try_to_release_page(page, GFP_KERNEL))
714 return -EAGAIN;
715
716 return migrate_page(mapping, newpage, page);
717}
718
0da5468f
CM		 719static int btree_writepage(struct page *page, struct writeback_control *wbc)
720{
d1310b2e 721 struct extent_io_tree *tree;
b9473439
CM		 722 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
723 struct extent_buffer *eb;
724 int was_dirty;
725
d1310b2e 726 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 727 if (!(current->flags & PF_MEMALLOC)) {
728 return extent_write_full_page(tree, page,
729 btree_get_extent, wbc);
730 }
5443be45 731
b9473439 732 redirty_page_for_writepage(wbc, page);
784b4e29 733 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
b9473439
CM		 734 WARN_ON(!eb);
735
736 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
737 if (!was_dirty) {
738 spin_lock(&root->fs_info->delalloc_lock);
739 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
740 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 741 }
b9473439
CM		 742 free_extent_buffer(eb);
743
744 unlock_page(page);
745 return 0;
5f39d397 746}
0da5468f
CM		 747
748static int btree_writepages(struct address_space *mapping,
749 struct writeback_control *wbc)
750{
d1310b2e
CM		 751 struct extent_io_tree *tree;
752 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 753 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 754 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 755 u64 num_dirty;
24ab9cd8 756 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 757
758 if (wbc->for_kupdate)
759 return 0;
760
b9473439
CM		 761 /* this is a bit racy, but that's ok */
762 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 763 if (num_dirty < thresh)
793955bc 764 return 0;
793955bc 765 }
0da5468f
CM		 766 return extent_writepages(tree, mapping, btree_get_extent, wbc);
767}
768
b2950863 769static int btree_readpage(struct file *file, struct page *page)
5f39d397 770{
d1310b2e
CM		 771 struct extent_io_tree *tree;
772 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 773 return extent_read_full_page(tree, page, btree_get_extent);
774}
22b0ebda 775
70dec807 776static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 777{
d1310b2e
CM		 778 struct extent_io_tree *tree;
779 struct extent_map_tree *map;
5f39d397 780 int ret;
d98237b3 781
98509cfc 782 if (PageWriteback(page) || PageDirty(page))
d397712b 783 return 0;
98509cfc 784
d1310b2e
CM		 785 tree = &BTRFS_I(page->mapping->host)->io_tree;
786 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 787
7b13b7b1 788 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 789 if (!ret)
6af118ce 790 return 0;
6af118ce
CM		 791
792 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 793 if (ret == 1) {
794 ClearPagePrivate(page);
795 set_page_private(page, 0);
796 page_cache_release(page);
797 }
6af118ce 798
d98237b3
CM		 799 return ret;
800}
801
5f39d397 802static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 803{
d1310b2e
CM		 804 struct extent_io_tree *tree;
805 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 806 extent_invalidatepage(tree, page, offset);
807 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 808 if (PagePrivate(page)) {
d397712b
CM		 809 printk(KERN_WARNING "btrfs warning page private not zero "
810 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 811 ClearPagePrivate(page);
812 set_page_private(page, 0);
813 page_cache_release(page);
814 }
d98237b3
CM		 815}
816
7f09410b 817static const struct address_space_operations btree_aops = {
d98237b3
CM		 818 .readpage = btree_readpage,
819 .writepage = btree_writepage,
0da5468f 820 .writepages = btree_writepages,
5f39d397
CM		 821 .releasepage = btree_releasepage,
822 .invalidatepage = btree_invalidatepage,
d98237b3 823 .sync_page = block_sync_page,
784b4e29 824 .migratepage = btree_migratepage,
d98237b3
CM		 825};
826
ca7a79ad
CM		 827int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
828 u64 parent_transid)
090d1875 829{
5f39d397
CM		 830 struct extent_buffer *buf = NULL;
831 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 832 int ret = 0;
090d1875 833
db94535d 834 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 835 if (!buf)
090d1875 836 return 0;
d1310b2e 837 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 838 buf, 0, 0, btree_get_extent, 0);
5f39d397 839 free_extent_buffer(buf);
de428b63 840 return ret;
090d1875
CM		 841}
842
0999df54
CM		 843struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
844 u64 bytenr, u32 blocksize)
845{
846 struct inode *btree_inode = root->fs_info->btree_inode;
847 struct extent_buffer *eb;
848 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
849 bytenr, blocksize, GFP_NOFS);
850 return eb;
851}
852
853struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
854 u64 bytenr, u32 blocksize)
855{
856 struct inode *btree_inode = root->fs_info->btree_inode;
857 struct extent_buffer *eb;
858
859 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
860 bytenr, blocksize, NULL, GFP_NOFS);
861 return eb;
862}
863
864
e02119d5
CM		 865int btrfs_write_tree_block(struct extent_buffer *buf)
866{
8aa38c31
CH		 867 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
868 buf->start + buf->len - 1);
e02119d5
CM		 869}
870
871int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
872{
8aa38c31
CH		 873 return filemap_fdatawait_range(buf->first_page->mapping,
874 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 875}
876
0999df54 877struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 878 u32 blocksize, u64 parent_transid)
0999df54
CM		 879{
880 struct extent_buffer *buf = NULL;
0999df54
CM		 881 int ret;
882
0999df54
CM		 883 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
884 if (!buf)
885 return NULL;
0999df54 886
ca7a79ad 887 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 888
d397712b 889 if (ret == 0)
b4ce94de 890 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 891 return buf;
ce9adaa5 892
eb60ceac
CM		 893}
894
e089f05c 895int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 896 struct extent_buffer *buf)
ed2ff2cb 897{
5f39d397 898 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 899 if (btrfs_header_generation(buf) ==
925baedd 900 root->fs_info->running_transaction->transid) {
b9447ef8 901 btrfs_assert_tree_locked(buf);
b4ce94de 902
b9473439
CM		 903 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
904 spin_lock(&root->fs_info->delalloc_lock);
905 if (root->fs_info->dirty_metadata_bytes >= buf->len)
906 root->fs_info->dirty_metadata_bytes -= buf->len;
907 else
908 WARN_ON(1);
909 spin_unlock(&root->fs_info->delalloc_lock);
910 }
b4ce94de 911
b9473439
CM		 912 /* ugh, clear_extent_buffer_dirty needs to lock the page */
913 btrfs_set_lock_blocking(buf);
d1310b2e 914 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 915 buf);
925baedd 916 }
5f39d397
CM		 917 return 0;
918}
919
db94535d 920static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 921 u32 stripesize, struct btrfs_root *root,
9f5fae2f 922 struct btrfs_fs_info *fs_info,
e20d96d6 923 u64 objectid)
d97e63b6 924{
cfaa7295 925 root->node = NULL;
a28ec197 926 root->commit_root = NULL;
db94535d
CM		 927 root->sectorsize = sectorsize;
928 root->nodesize = nodesize;
929 root->leafsize = leafsize;
87ee04eb 930 root->stripesize = stripesize;
123abc88 931 root->ref_cows = 0;
0b86a832 932 root->track_dirty = 0;
c71bf099 933 root->in_radix = 0;
d68fc57b
YZ		 934 root->orphan_item_inserted = 0;
935 root->orphan_cleanup_state = 0;
0b86a832 936
9f5fae2f 937 root->fs_info = fs_info;
0f7d52f4
CM		 938 root->objectid = objectid;
939 root->last_trans = 0;
13a8a7c8 940 root->highest_objectid = 0;
58176a96 941 root->name = NULL;
4313b399 942 root->in_sysfs = 0;
6bef4d31 943 root->inode_tree = RB_ROOT;
f0486c68 944 root->block_rsv = NULL;
d68fc57b 945 root->orphan_block_rsv = NULL;
0b86a832
CM		 946
947 INIT_LIST_HEAD(&root->dirty_list);
7b128766 948 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 949 INIT_LIST_HEAD(&root->root_list);
925baedd 950 spin_lock_init(&root->node_lock);
d68fc57b 951 spin_lock_init(&root->orphan_lock);
5d4f98a2 952 spin_lock_init(&root->inode_lock);
f0486c68 953 spin_lock_init(&root->accounting_lock);
a2135011 954 mutex_init(&root->objectid_mutex);
e02119d5 955 mutex_init(&root->log_mutex);
7237f183
YZ		 956 init_waitqueue_head(&root->log_writer_wait);
957 init_waitqueue_head(&root->log_commit_wait[0]);
958 init_waitqueue_head(&root->log_commit_wait[1]);
959 atomic_set(&root->log_commit[0], 0);
960 atomic_set(&root->log_commit[1], 0);
961 atomic_set(&root->log_writers, 0);
962 root->log_batch = 0;
963 root->log_transid = 0;
257c62e1 964 root->last_log_commit = 0;
d0c803c4
CM		 965 extent_io_tree_init(&root->dirty_log_pages,
966 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 967
3768f368
CM		 968 memset(&root->root_key, 0, sizeof(root->root_key));
969 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 970 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 971 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 972 root->defrag_trans_start = fs_info->generation;
58176a96 973 init_completion(&root->kobj_unregister);
6702ed49 974 root->defrag_running = 0;
4d775673 975 root->root_key.objectid = objectid;
3394e160
CM		 976 root->anon_super.s_root = NULL;
977 root->anon_super.s_dev = 0;
978 INIT_LIST_HEAD(&root->anon_super.s_list);
979 INIT_LIST_HEAD(&root->anon_super.s_instances);
980 init_rwsem(&root->anon_super.s_umount);
981
3768f368
CM		 982 return 0;
983}
984
db94535d 985static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 986 struct btrfs_fs_info *fs_info,
987 u64 objectid,
e20d96d6 988 struct btrfs_root *root)
3768f368
CM		 989{
990 int ret;
db94535d 991 u32 blocksize;
84234f3a 992 u64 generation;
3768f368 993
db94535d 994 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 995 tree_root->sectorsize, tree_root->stripesize,
996 root, fs_info, objectid);
3768f368
CM		 997 ret = btrfs_find_last_root(tree_root, objectid,
998 &root->root_item, &root->root_key);
4df27c4d
YZ		 999 if (ret > 0)
1000 return -ENOENT;
3768f368
CM		 1001 BUG_ON(ret);
1002
84234f3a 1003 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1004 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1005 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1006 blocksize, generation);
3768f368 1007 BUG_ON(!root->node);
4df27c4d 1008 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1009 return 0;
1010}
1011
7237f183
YZ		 1012static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1013 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1014{
1015 struct btrfs_root *root;
1016 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1017 struct extent_buffer *leaf;
e02119d5
CM		 1018
1019 root = kzalloc(sizeof(*root), GFP_NOFS);
1020 if (!root)
7237f183 1021 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1022
1023 __setup_root(tree_root->nodesize, tree_root->leafsize,
1024 tree_root->sectorsize, tree_root->stripesize,
1025 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1026
1027 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1028 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1029 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1030 /*
1031 * log trees do not get reference counted because they go away
1032 * before a real commit is actually done. They do store pointers
1033 * to file data extents, and those reference counts still get
1034 * updated (along with back refs to the log tree).
1035 */
e02119d5
CM		 1036 root->ref_cows = 0;
1037
5d4f98a2
YZ		 1038 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1039 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1040 if (IS_ERR(leaf)) {
1041 kfree(root);
1042 return ERR_CAST(leaf);
1043 }
e02119d5 1044
5d4f98a2
YZ		 1045 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1046 btrfs_set_header_bytenr(leaf, leaf->start);
1047 btrfs_set_header_generation(leaf, trans->transid);
1048 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1049 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1050 root->node = leaf;
e02119d5
CM		 1051
1052 write_extent_buffer(root->node, root->fs_info->fsid,
1053 (unsigned long)btrfs_header_fsid(root->node),
1054 BTRFS_FSID_SIZE);
1055 btrfs_mark_buffer_dirty(root->node);
1056 btrfs_tree_unlock(root->node);
7237f183
YZ		 1057 return root;
1058}
1059
1060int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1061 struct btrfs_fs_info *fs_info)
1062{
1063 struct btrfs_root *log_root;
1064
1065 log_root = alloc_log_tree(trans, fs_info);
1066 if (IS_ERR(log_root))
1067 return PTR_ERR(log_root);
1068 WARN_ON(fs_info->log_root_tree);
1069 fs_info->log_root_tree = log_root;
1070 return 0;
1071}
1072
1073int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1074 struct btrfs_root *root)
1075{
1076 struct btrfs_root *log_root;
1077 struct btrfs_inode_item *inode_item;
1078
1079 log_root = alloc_log_tree(trans, root->fs_info);
1080 if (IS_ERR(log_root))
1081 return PTR_ERR(log_root);
1082
1083 log_root->last_trans = trans->transid;
1084 log_root->root_key.offset = root->root_key.objectid;
1085
1086 inode_item = &log_root->root_item.inode;
1087 inode_item->generation = cpu_to_le64(1);
1088 inode_item->size = cpu_to_le64(3);
1089 inode_item->nlink = cpu_to_le32(1);
1090 inode_item->nbytes = cpu_to_le64(root->leafsize);
1091 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1092
5d4f98a2 1093 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1094
1095 WARN_ON(root->log_root);
1096 root->log_root = log_root;
1097 root->log_transid = 0;
257c62e1 1098 root->last_log_commit = 0;
e02119d5
CM		 1099 return 0;
1100}
1101
1102struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1103 struct btrfs_key *location)
1104{
1105 struct btrfs_root *root;
1106 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1107 struct btrfs_path *path;
5f39d397 1108 struct extent_buffer *l;
84234f3a 1109 u64 generation;
db94535d 1110 u32 blocksize;
0f7d52f4
CM		 1111 int ret = 0;
1112
5eda7b5e 1113 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1114 if (!root)
0f7d52f4 1115 return ERR_PTR(-ENOMEM);
0f7d52f4 1116 if (location->offset == (u64)-1) {
db94535d 1117 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1118 location->objectid, root);
1119 if (ret) {
0f7d52f4
CM		 1120 kfree(root);
1121 return ERR_PTR(ret);
1122 }
13a8a7c8 1123 goto out;
0f7d52f4
CM		 1124 }
1125
db94535d 1126 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1127 tree_root->sectorsize, tree_root->stripesize,
1128 root, fs_info, location->objectid);
0f7d52f4
CM		 1129
1130 path = btrfs_alloc_path();
1131 BUG_ON(!path);
1132 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1133 if (ret == 0) {
1134 l = path->nodes[0];
1135 read_extent_buffer(l, &root->root_item,
1136 btrfs_item_ptr_offset(l, path->slots[0]),
1137 sizeof(root->root_item));
1138 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1139 }
0f7d52f4
CM		 1140 btrfs_free_path(path);
1141 if (ret) {
13a8a7c8
YZ		 1142 if (ret > 0)
1143 ret = -ENOENT;
0f7d52f4
CM		 1144 return ERR_PTR(ret);
1145 }
13a8a7c8 1146
84234f3a 1147 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1148 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1149 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1150 blocksize, generation);
5d4f98a2 1151 root->commit_root = btrfs_root_node(root);
0f7d52f4 1152 BUG_ON(!root->node);
13a8a7c8
YZ		 1153out:
1154 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1155 root->ref_cows = 1;
13a8a7c8 1156
5eda7b5e
CM		 1157 return root;
1158}
1159
dc17ff8f
CM		 1160struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1161 u64 root_objectid)
1162{
1163 struct btrfs_root *root;
1164
1165 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1166 return fs_info->tree_root;
1167 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1168 return fs_info->extent_root;
1169
1170 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1171 (unsigned long)root_objectid);
1172 return root;
1173}
1174
edbd8d4e
CM		 1175struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1176 struct btrfs_key *location)
5eda7b5e
CM		 1177{
1178 struct btrfs_root *root;
1179 int ret;
1180
edbd8d4e
CM		 1181 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1182 return fs_info->tree_root;
1183 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1184 return fs_info->extent_root;
8f18cf13
CM		 1185 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1186 return fs_info->chunk_root;
1187 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1188 return fs_info->dev_root;
0403e47e
YZ		 1189 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1190 return fs_info->csum_root;
4df27c4d
YZ		 1191again:
1192 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1193 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1194 (unsigned long)location->objectid);
4df27c4d 1195 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1196 if (root)
1197 return root;
1198
e02119d5 1199 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1200 if (IS_ERR(root))
1201 return root;
3394e160
CM		 1202
1203 set_anon_super(&root->anon_super, NULL);
1204
d68fc57b
YZ		 1205 if (btrfs_root_refs(&root->root_item) == 0) {
1206 ret = -ENOENT;
1207 goto fail;
1208 }
1209
1210 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1211 if (ret < 0)
1212 goto fail;
1213 if (ret == 0)
1214 root->orphan_item_inserted = 1;
1215
4df27c4d
YZ		 1216 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1217 if (ret)
1218 goto fail;
1219
1220 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1221 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1222 (unsigned long)root->root_key.objectid,
0f7d52f4 1223 root);
d68fc57b 1224 if (ret == 0)
4df27c4d 1225 root->in_radix = 1;
d68fc57b 1226
4df27c4d
YZ		 1227 spin_unlock(&fs_info->fs_roots_radix_lock);
1228 radix_tree_preload_end();
0f7d52f4 1229 if (ret) {
4df27c4d
YZ		 1230 if (ret == -EEXIST) {
1231 free_fs_root(root);
1232 goto again;
1233 }
1234 goto fail;
0f7d52f4 1235 }
4df27c4d
YZ		 1236
1237 ret = btrfs_find_dead_roots(fs_info->tree_root,
1238 root->root_key.objectid);
1239 WARN_ON(ret);
edbd8d4e 1240 return root;
4df27c4d
YZ		 1241fail:
1242 free_fs_root(root);
1243 return ERR_PTR(ret);
edbd8d4e
CM		 1244}
1245
1246struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1247 struct btrfs_key *location,
1248 const char *name, int namelen)
1249{
4df27c4d
YZ		 1250 return btrfs_read_fs_root_no_name(fs_info, location);
1251#if 0
edbd8d4e
CM		 1252 struct btrfs_root *root;
1253 int ret;
1254
1255 root = btrfs_read_fs_root_no_name(fs_info, location);
1256 if (!root)
1257 return NULL;
58176a96 1258
4313b399
CM		 1259 if (root->in_sysfs)
1260 return root;
1261
58176a96
JB		 1262 ret = btrfs_set_root_name(root, name, namelen);
1263 if (ret) {
5f39d397 1264 free_extent_buffer(root->node);
58176a96
JB		 1265 kfree(root);
1266 return ERR_PTR(ret);
1267 }
4df27c4d 1268
58176a96
JB		 1269 ret = btrfs_sysfs_add_root(root);
1270 if (ret) {
5f39d397 1271 free_extent_buffer(root->node);
58176a96
JB		 1272 kfree(root->name);
1273 kfree(root);
1274 return ERR_PTR(ret);
1275 }
4313b399 1276 root->in_sysfs = 1;
0f7d52f4 1277 return root;
4df27c4d 1278#endif
0f7d52f4 1279}
04160088
CM		 1280
1281static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1282{
1283 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1284 int ret = 0;
04160088
CM		 1285 struct btrfs_device *device;
1286 struct backing_dev_info *bdi;
b7967db7 1287
c6e30871 1288 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1289 if (!device->bdev)
1290 continue;
04160088
CM		 1291 bdi = blk_get_backing_dev_info(device->bdev);
1292 if (bdi && bdi_congested(bdi, bdi_bits)) {
1293 ret = 1;
1294 break;
1295 }
1296 }
1297 return ret;
1298}
1299
38b66988
CM		 1300/*
1301 * this unplugs every device on the box, and it is only used when page
1302 * is null
1303 */
1304static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1305{
38b66988
CM		 1306 struct btrfs_device *device;
1307 struct btrfs_fs_info *info;
1308
1309 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1310 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1311 if (!device->bdev)
1312 continue;
1313
38b66988 1314 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1315 if (bdi->unplug_io_fn)
38b66988 1316 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1317 }
1318}
1319
b2950863 1320static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1321{
38b66988 1322 struct inode *inode;
f2d8d74d
CM		 1323 struct extent_map_tree *em_tree;
1324 struct extent_map *em;
bcbfce8a 1325 struct address_space *mapping;
38b66988
CM		 1326 u64 offset;
1327
bcbfce8a 1328 /* the generic O_DIRECT read code does this */
9f0ba5bd 1329 if (1 || !page) {
38b66988
CM		 1330 __unplug_io_fn(bdi, page);
1331 return;
1332 }
1333
bcbfce8a
CM		 1334 /*
1335 * page->mapping may change at any time. Get a consistent copy
1336 * and use that for everything below
1337 */
1338 smp_mb();
1339 mapping = page->mapping;
1340 if (!mapping)
1341 return;
1342
1343 inode = mapping->host;
240d5d48
CM		 1344
1345 /*
1346 * don't do the expensive searching for a small number of
1347 * devices
1348 */
1349 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1350 __unplug_io_fn(bdi, page);
1351 return;
1352 }
1353
38b66988 1354 offset = page_offset(page);
04160088 1355
f2d8d74d 1356 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1357 read_lock(&em_tree->lock);
f2d8d74d 1358 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1359 read_unlock(&em_tree->lock);
89642229
CM		 1360 if (!em) {
1361 __unplug_io_fn(bdi, page);
f2d8d74d 1362 return;
89642229 1363 }
f2d8d74d 1364
89642229
CM		 1365 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1366 free_extent_map(em);
1367 __unplug_io_fn(bdi, page);
1368 return;
1369 }
f2d8d74d
CM		 1370 offset = offset - em->start;
1371 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1372 em->block_start + offset, page);
1373 free_extent_map(em);
04160088
CM		 1374}
1375
ad081f14
JA		 1376/*
1377 * If this fails, caller must call bdi_destroy() to get rid of the
1378 * bdi again.
1379 */
04160088
CM		 1380static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1381{
ad081f14
JA		 1382 int err;
1383
1384 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1385 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1386 if (err)
1387 return err;
1388
4575c9cc 1389 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1390 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1391 bdi->unplug_io_data = info;
1392 bdi->congested_fn = btrfs_congested_fn;
1393 bdi->congested_data = info;
1394 return 0;
1395}
1396
ce9adaa5
CM		 1397static int bio_ready_for_csum(struct bio *bio)
1398{
1399 u64 length = 0;
1400 u64 buf_len = 0;
1401 u64 start = 0;
1402 struct page *page;
1403 struct extent_io_tree *io_tree = NULL;
ce9adaa5
CM		 1404 struct bio_vec *bvec;
1405 int i;
1406 int ret;
1407
1408 bio_for_each_segment(bvec, bio, i) {
1409 page = bvec->bv_page;
1410 if (page->private == EXTENT_PAGE_PRIVATE) {
1411 length += bvec->bv_len;
1412 continue;
1413 }
1414 if (!page->private) {
1415 length += bvec->bv_len;
1416 continue;
1417 }
1418 length = bvec->bv_len;
1419 buf_len = page->private >> 2;
1420 start = page_offset(page) + bvec->bv_offset;
1421 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
ce9adaa5
CM		 1422 }
1423 /* are we fully contained in this bio? */
1424 if (buf_len <= length)
1425 return 1;
1426
1427 ret = extent_range_uptodate(io_tree, start + length,
1428 start + buf_len - 1);
ce9adaa5
CM		 1429 return ret;
1430}
1431
8b712842
CM		 1432/*
1433 * called by the kthread helper functions to finally call the bio end_io
1434 * functions. This is where read checksum verification actually happens
1435 */
1436static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1437{
ce9adaa5 1438 struct bio *bio;
8b712842
CM		 1439 struct end_io_wq *end_io_wq;
1440 struct btrfs_fs_info *fs_info;
ce9adaa5 1441 int error;
ce9adaa5 1442
8b712842
CM		 1443 end_io_wq = container_of(work, struct end_io_wq, work);
1444 bio = end_io_wq->bio;
1445 fs_info = end_io_wq->info;
ce9adaa5 1446
cad321ad 1447 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1448 * be checksummed at once. This makes sure the entire block is in
1449 * ram and up to date before trying to verify things. For
1450 * blocksize <= pagesize, it is basically a noop
1451 */
7b6d91da 1452 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1453 !bio_ready_for_csum(bio)) {
d20f7043 1454 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1455 &end_io_wq->work);
1456 return;
1457 }
1458 error = end_io_wq->error;
1459 bio->bi_private = end_io_wq->private;
1460 bio->bi_end_io = end_io_wq->end_io;
1461 kfree(end_io_wq);
8b712842 1462 bio_endio(bio, error);
44b8bd7e
CM		 1463}
1464
a74a4b97
CM		 1465static int cleaner_kthread(void *arg)
1466{
1467 struct btrfs_root *root = arg;
1468
1469 do {
a74a4b97 1470 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1471
1472 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1473 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1474 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1475 btrfs_clean_old_snapshots(root);
1476 mutex_unlock(&root->fs_info->cleaner_mutex);
1477 }
a74a4b97
CM		 1478
1479 if (freezing(current)) {
1480 refrigerator();
1481 } else {
a74a4b97 1482 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1483 if (!kthread_should_stop())
1484 schedule();
a74a4b97
CM		 1485 __set_current_state(TASK_RUNNING);
1486 }
1487 } while (!kthread_should_stop());
1488 return 0;
1489}
1490
1491static int transaction_kthread(void *arg)
1492{
1493 struct btrfs_root *root = arg;
1494 struct btrfs_trans_handle *trans;
1495 struct btrfs_transaction *cur;
8929ecfa 1496 u64 transid;
a74a4b97
CM		 1497 unsigned long now;
1498 unsigned long delay;
1499 int ret;
1500
1501 do {
a74a4b97
CM		 1502 delay = HZ * 30;
1503 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1504 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1505
8929ecfa 1506 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1507 cur = root->fs_info->running_transaction;
1508 if (!cur) {
8929ecfa 1509 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1510 goto sleep;
1511 }
31153d81 1512
a74a4b97 1513 now = get_seconds();
8929ecfa
YZ		 1514 if (!cur->blocked &&
1515 (now < cur->start_time || now - cur->start_time < 30)) {
1516 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1517 delay = HZ * 5;
1518 goto sleep;
1519 }
8929ecfa
YZ		 1520 transid = cur->transid;
1521 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1522
8929ecfa
YZ		 1523 trans = btrfs_join_transaction(root, 1);
1524 if (transid == trans->transid) {
1525 ret = btrfs_commit_transaction(trans, root);
1526 BUG_ON(ret);
1527 } else {
1528 btrfs_end_transaction(trans, root);
1529 }
a74a4b97
CM		 1530sleep:
1531 wake_up_process(root->fs_info->cleaner_kthread);
1532 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1533
1534 if (freezing(current)) {
1535 refrigerator();
1536 } else {
a74a4b97 1537 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1538 if (!kthread_should_stop() &&
1539 !btrfs_transaction_blocked(root->fs_info))
1540 schedule_timeout(delay);
a74a4b97
CM		 1541 __set_current_state(TASK_RUNNING);
1542 }
1543 } while (!kthread_should_stop());
1544 return 0;
1545}
1546
8a4b83cc 1547struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1548 struct btrfs_fs_devices *fs_devices,
1549 char *options)
2e635a27 1550{
db94535d
CM		 1551 u32 sectorsize;
1552 u32 nodesize;
1553 u32 leafsize;
1554 u32 blocksize;
87ee04eb 1555 u32 stripesize;
84234f3a 1556 u64 generation;
f2b636e8 1557 u64 features;
3de4586c 1558 struct btrfs_key location;
a061fc8d 1559 struct buffer_head *bh;
e02119d5 1560 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1561 GFP_NOFS);
d20f7043
CM		 1562 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1563 GFP_NOFS);
450ba0ea
JB		 1564 struct btrfs_root *tree_root = btrfs_sb(sb);
1565 struct btrfs_fs_info *fs_info = tree_root->fs_info;
e02119d5 1566 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1567 GFP_NOFS);
e02119d5 1568 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1569 GFP_NOFS);
e02119d5
CM		 1570 struct btrfs_root *log_tree_root;
1571
eb60ceac 1572 int ret;
e58ca020 1573 int err = -EINVAL;
4543df7e 1574
2c90e5d6 1575 struct btrfs_super_block *disk_super;
8790d502 1576
0463bb4e 1577 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1578 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1579 err = -ENOMEM;
1580 goto fail;
1581 }
76dda93c
YZ		 1582
1583 ret = init_srcu_struct(&fs_info->subvol_srcu);
1584 if (ret) {
1585 err = ret;
1586 goto fail;
1587 }
1588
1589 ret = setup_bdi(fs_info, &fs_info->bdi);
1590 if (ret) {
1591 err = ret;
1592 goto fail_srcu;
1593 }
1594
1595 fs_info->btree_inode = new_inode(sb);
1596 if (!fs_info->btree_inode) {
1597 err = -ENOMEM;
1598 goto fail_bdi;
1599 }
1600
1601 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1602 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1603 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1604 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1605 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1606 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1607 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1608 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1609 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1610 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1611 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1612 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1613 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1614
58176a96 1615 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1616 fs_info->tree_root = tree_root;
1617 fs_info->extent_root = extent_root;
d20f7043 1618 fs_info->csum_root = csum_root;
0b86a832
CM		 1619 fs_info->chunk_root = chunk_root;
1620 fs_info->dev_root = dev_root;
8a4b83cc 1621 fs_info->fs_devices = fs_devices;
0b86a832 1622 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1623 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1624 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1625 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1626 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1627 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1628 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1629 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1630 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1631 mutex_init(&fs_info->durable_block_rsv_mutex);
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);
bb9c12c9 1703 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 1704 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1705
0b86a832 1706 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1707 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 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);
61b49440
CM		 1771
1772 /* a higher idle thresh on the submit workers makes it much more
1773 * likely that bios will be send down in a sane order to the
1774 * devices
1775 */
1776 fs_info->submit_workers.idle_thresh = 64;
53863232 1777
771ed689 1778 fs_info->workers.idle_thresh = 16;
4a69a410 1779 fs_info->workers.ordered = 1;
61b49440 1780
771ed689
CM		 1781 fs_info->delalloc_workers.idle_thresh = 2;
1782 fs_info->delalloc_workers.ordered = 1;
1783
61d92c32
CM		 1784 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1785 &fs_info->generic_worker);
5443be45 1786 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1787 fs_info->thread_pool_size,
1788 &fs_info->generic_worker);
d20f7043 1789 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1790 fs_info->thread_pool_size,
1791 &fs_info->generic_worker);
cad321ad 1792 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1793 "endio-meta-write", fs_info->thread_pool_size,
1794 &fs_info->generic_worker);
5443be45 1795 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1796 fs_info->thread_pool_size,
1797 &fs_info->generic_worker);
0cb59c99
JB		 1798 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1799 1, &fs_info->generic_worker);
61b49440
CM		 1800
1801 /*
1802 * endios are largely parallel and should have a very
1803 * low idle thresh
1804 */
1805 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1806 fs_info->endio_meta_workers.idle_thresh = 4;
1807
9042846b
CM		 1808 fs_info->endio_write_workers.idle_thresh = 2;
1809 fs_info->endio_meta_write_workers.idle_thresh = 2;
1810
4543df7e 1811 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1812 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1813 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1814 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1815 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1816 btrfs_start_workers(&fs_info->endio_workers, 1);
1817 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1818 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1819 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1820 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
4543df7e 1821
4575c9cc 1822 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1823 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1824 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1825
db94535d
CM		 1826 nodesize = btrfs_super_nodesize(disk_super);
1827 leafsize = btrfs_super_leafsize(disk_super);
1828 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1829 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1830 tree_root->nodesize = nodesize;
1831 tree_root->leafsize = leafsize;
1832 tree_root->sectorsize = sectorsize;
87ee04eb 1833 tree_root->stripesize = stripesize;
a061fc8d
CM		 1834
1835 sb->s_blocksize = sectorsize;
1836 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1837
39279cc3
CM		 1838 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1839 sizeof(disk_super->magic))) {
d397712b 1840 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1841 goto fail_sb_buffer;
1842 }
19c00ddc 1843
925baedd 1844 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1845 ret = btrfs_read_sys_array(tree_root);
925baedd 1846 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1847 if (ret) {
d397712b
CM		 1848 printk(KERN_WARNING "btrfs: failed to read the system "
1849 "array on %s\n", sb->s_id);
5d4f98a2 1850 goto fail_sb_buffer;
84eed90f 1851 }
0b86a832
CM		 1852
1853 blocksize = btrfs_level_size(tree_root,
1854 btrfs_super_chunk_root_level(disk_super));
84234f3a 1855 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1856
1857 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1858 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1859
1860 chunk_root->node = read_tree_block(chunk_root,
1861 btrfs_super_chunk_root(disk_super),
84234f3a 1862 blocksize, generation);
0b86a832 1863 BUG_ON(!chunk_root->node);
83121942
DW		 1864 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1865 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1866 sb->s_id);
1867 goto fail_chunk_root;
1868 }
5d4f98a2
YZ		 1869 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1870 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1871
e17cade2 1872 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1873 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1874 BTRFS_UUID_SIZE);
e17cade2 1875
925baedd 1876 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1877 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1878 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1879 if (ret) {
d397712b
CM		 1880 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1881 sb->s_id);
2b82032c
YZ		 1882 goto fail_chunk_root;
1883 }
0b86a832 1884
dfe25020
CM		 1885 btrfs_close_extra_devices(fs_devices);
1886
db94535d
CM		 1887 blocksize = btrfs_level_size(tree_root,
1888 btrfs_super_root_level(disk_super));
84234f3a 1889 generation = btrfs_super_generation(disk_super);
0b86a832 1890
e20d96d6 1891 tree_root->node = read_tree_block(tree_root,
db94535d 1892 btrfs_super_root(disk_super),
84234f3a 1893 blocksize, generation);
39279cc3 1894 if (!tree_root->node)
2b82032c 1895 goto fail_chunk_root;
83121942
DW		 1896 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1897 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1898 sb->s_id);
1899 goto fail_tree_root;
1900 }
5d4f98a2
YZ		 1901 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1902 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 1903
1904 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1905 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1906 if (ret)
39279cc3 1907 goto fail_tree_root;
0b86a832
CM		 1908 extent_root->track_dirty = 1;
1909
1910 ret = find_and_setup_root(tree_root, fs_info,
1911 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 1912 if (ret)
1913 goto fail_extent_root;
5d4f98a2 1914 dev_root->track_dirty = 1;
3768f368 1915
d20f7043
CM		 1916 ret = find_and_setup_root(tree_root, fs_info,
1917 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1918 if (ret)
5d4f98a2 1919 goto fail_dev_root;
d20f7043
CM		 1920
1921 csum_root->track_dirty = 1;
1922
8929ecfa
YZ		 1923 fs_info->generation = generation;
1924 fs_info->last_trans_committed = generation;
1925 fs_info->data_alloc_profile = (u64)-1;
1926 fs_info->metadata_alloc_profile = (u64)-1;
1927 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1928
1b1d1f66
JB		 1929 ret = btrfs_read_block_groups(extent_root);
1930 if (ret) {
1931 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1932 goto fail_block_groups;
1933 }
9078a3e1 1934
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
676e4c86
DC		 1967 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1968 if (!log_tree_root) {
1969 err = -ENOMEM;
1970 goto fail_trans_kthread;
1971 }
e02119d5
CM		 1972
1973 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1974 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1975
1976 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 1977 blocksize,
1978 generation + 1);
e02119d5
CM		 1979 ret = btrfs_recover_log_trees(log_tree_root);
1980 BUG_ON(ret);
e556ce2c
YZ		 1981
1982 if (sb->s_flags & MS_RDONLY) {
1983 ret = btrfs_commit_super(tree_root);
1984 BUG_ON(ret);
1985 }
e02119d5 1986 }
1a40e23b 1987
76dda93c
YZ		 1988 ret = btrfs_find_orphan_roots(tree_root);
1989 BUG_ON(ret);
1990
7c2ca468 1991 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ		 1992 ret = btrfs_cleanup_fs_roots(fs_info);
1993 BUG_ON(ret);
1994
5d4f98a2 1995 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 1996 if (ret < 0) {
1997 printk(KERN_WARNING
1998 "btrfs: failed to recover relocation\n");
1999 err = -EINVAL;
2000 goto fail_trans_kthread;
2001 }
7c2ca468 2002 }
1a40e23b 2003
3de4586c
CM		 2004 location.objectid = BTRFS_FS_TREE_OBJECTID;
2005 location.type = BTRFS_ROOT_ITEM_KEY;
2006 location.offset = (u64)-1;
2007
3de4586c
CM		 2008 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2009 if (!fs_info->fs_root)
7c2ca468 2010 goto fail_trans_kthread;
3140c9a3
DC		 2011 if (IS_ERR(fs_info->fs_root)) {
2012 err = PTR_ERR(fs_info->fs_root);
2013 goto fail_trans_kthread;
2014 }
c289811c 2015
e3acc2a6
JB		 2016 if (!(sb->s_flags & MS_RDONLY)) {
2017 down_read(&fs_info->cleanup_work_sem);
2018 btrfs_orphan_cleanup(fs_info->fs_root);
0af3d00b 2019 btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6
JB		 2020 up_read(&fs_info->cleanup_work_sem);
2021 }
2022
0f7d52f4 2023 return tree_root;
39279cc3 2024
7c2ca468
CM		 2025fail_trans_kthread:
2026 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2027fail_cleaner:
a74a4b97 2028 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2029
2030 /*
2031 * make sure we're done with the btree inode before we stop our
2032 * kthreads
2033 */
2034 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2035 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2036
1b1d1f66
JB		 2037fail_block_groups:
2038 btrfs_free_block_groups(fs_info);
d20f7043 2039 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2040 free_extent_buffer(csum_root->commit_root);
2041fail_dev_root:
2042 free_extent_buffer(dev_root->node);
2043 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2044fail_extent_root:
2045 free_extent_buffer(extent_root->node);
5d4f98a2 2046 free_extent_buffer(extent_root->commit_root);
39279cc3 2047fail_tree_root:
5f39d397 2048 free_extent_buffer(tree_root->node);
5d4f98a2 2049 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2050fail_chunk_root:
2051 free_extent_buffer(chunk_root->node);
5d4f98a2 2052 free_extent_buffer(chunk_root->commit_root);
39279cc3 2053fail_sb_buffer:
61d92c32 2054 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2055 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2056 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2057 btrfs_stop_workers(&fs_info->workers);
2058 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2059 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2060 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2061 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2062 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2063 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2064fail_iput:
7c2ca468 2065 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2066 iput(fs_info->btree_inode);
7e662854 2067
dfe25020 2068 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2069 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2070fail_bdi:
7e662854 2071 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2072fail_srcu:
2073 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2074fail:
39279cc3
CM		 2075 kfree(extent_root);
2076 kfree(tree_root);
2077 kfree(fs_info);
83afeac4
JM		 2078 kfree(chunk_root);
2079 kfree(dev_root);
d20f7043 2080 kfree(csum_root);
39279cc3 2081 return ERR_PTR(err);
eb60ceac
CM		 2082}
2083
f2984462
CM		 2084static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2085{
2086 char b[BDEVNAME_SIZE];
2087
2088 if (uptodate) {
2089 set_buffer_uptodate(bh);
2090 } else {
2091 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2092 printk(KERN_WARNING "lost page write due to "
2093 "I/O error on %s\n",
2094 bdevname(bh->b_bdev, b));
2095 }
1259ab75
CM		 2096 /* note, we dont' set_buffer_write_io_error because we have
2097 * our own ways of dealing with the IO errors
2098 */
f2984462
CM		 2099 clear_buffer_uptodate(bh);
2100 }
2101 unlock_buffer(bh);
2102 put_bh(bh);
2103}
2104
a512bbf8
YZ		 2105struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2106{
2107 struct buffer_head *bh;
2108 struct buffer_head *latest = NULL;
2109 struct btrfs_super_block *super;
2110 int i;
2111 u64 transid = 0;
2112 u64 bytenr;
2113
2114 /* we would like to check all the supers, but that would make
2115 * a btrfs mount succeed after a mkfs from a different FS.
2116 * So, we need to add a special mount option to scan for
2117 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2118 */
2119 for (i = 0; i < 1; i++) {
2120 bytenr = btrfs_sb_offset(i);
2121 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2122 break;
2123 bh = __bread(bdev, bytenr / 4096, 4096);
2124 if (!bh)
2125 continue;
2126
2127 super = (struct btrfs_super_block *)bh->b_data;
2128 if (btrfs_super_bytenr(super) != bytenr ||
2129 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2130 sizeof(super->magic))) {
2131 brelse(bh);
2132 continue;
2133 }
2134
2135 if (!latest || btrfs_super_generation(super) > transid) {
2136 brelse(latest);
2137 latest = bh;
2138 transid = btrfs_super_generation(super);
2139 } else {
2140 brelse(bh);
2141 }
2142 }
2143 return latest;
2144}
2145
4eedeb75
HH		 2146/*
2147 * this should be called twice, once with wait == 0 and
2148 * once with wait == 1. When wait == 0 is done, all the buffer heads
2149 * we write are pinned.
2150 *
2151 * They are released when wait == 1 is done.
2152 * max_mirrors must be the same for both runs, and it indicates how
2153 * many supers on this one device should be written.
2154 *
2155 * max_mirrors == 0 means to write them all.
2156 */
a512bbf8
YZ		 2157static int write_dev_supers(struct btrfs_device *device,
2158 struct btrfs_super_block *sb,
2159 int do_barriers, int wait, int max_mirrors)
2160{
2161 struct buffer_head *bh;
2162 int i;
2163 int ret;
2164 int errors = 0;
2165 u32 crc;
2166 u64 bytenr;
2167 int last_barrier = 0;
2168
2169 if (max_mirrors == 0)
2170 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2171
2172 /* make sure only the last submit_bh does a barrier */
2173 if (do_barriers) {
2174 for (i = 0; i < max_mirrors; i++) {
2175 bytenr = btrfs_sb_offset(i);
2176 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2177 device->total_bytes)
2178 break;
2179 last_barrier = i;
2180 }
2181 }
2182
2183 for (i = 0; i < max_mirrors; i++) {
2184 bytenr = btrfs_sb_offset(i);
2185 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2186 break;
2187
2188 if (wait) {
2189 bh = __find_get_block(device->bdev, bytenr / 4096,
2190 BTRFS_SUPER_INFO_SIZE);
2191 BUG_ON(!bh);
a512bbf8 2192 wait_on_buffer(bh);
4eedeb75
HH		 2193 if (!buffer_uptodate(bh))
2194 errors++;
2195
2196 /* drop our reference */
2197 brelse(bh);
2198
2199 /* drop the reference from the wait == 0 run */
2200 brelse(bh);
2201 continue;
a512bbf8
YZ		 2202 } else {
2203 btrfs_set_super_bytenr(sb, bytenr);
2204
2205 crc = ~(u32)0;
2206 crc = btrfs_csum_data(NULL, (char *)sb +
2207 BTRFS_CSUM_SIZE, crc,
2208 BTRFS_SUPER_INFO_SIZE -
2209 BTRFS_CSUM_SIZE);
2210 btrfs_csum_final(crc, sb->csum);
2211
4eedeb75
HH		 2212 /*
2213 * one reference for us, and we leave it for the
2214 * caller
2215 */
a512bbf8
YZ		 2216 bh = __getblk(device->bdev, bytenr / 4096,
2217 BTRFS_SUPER_INFO_SIZE);
2218 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2219
4eedeb75 2220 /* one reference for submit_bh */
a512bbf8 2221 get_bh(bh);
4eedeb75
HH		 2222
2223 set_buffer_uptodate(bh);
a512bbf8
YZ		 2224 lock_buffer(bh);
2225 bh->b_end_io = btrfs_end_buffer_write_sync;
2226 }
2227
2228 if (i == last_barrier && do_barriers && device->barriers) {
2229 ret = submit_bh(WRITE_BARRIER, bh);
2230 if (ret == -EOPNOTSUPP) {
2231 printk("btrfs: disabling barriers on dev %s\n",
2232 device->name);
2233 set_buffer_uptodate(bh);
2234 device->barriers = 0;
4eedeb75 2235 /* one reference for submit_bh */
a512bbf8
YZ		 2236 get_bh(bh);
2237 lock_buffer(bh);
ffbd517d 2238 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2239 }
2240 } else {
ffbd517d 2241 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8
YZ		 2242 }
2243
4eedeb75 2244 if (ret)
a512bbf8 2245 errors++;
a512bbf8
YZ		 2246 }
2247 return errors < i ? 0 : -1;
2248}
2249
2250int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2251{
e5e9a520 2252 struct list_head *head;
f2984462 2253 struct btrfs_device *dev;
a061fc8d 2254 struct btrfs_super_block *sb;
f2984462 2255 struct btrfs_dev_item *dev_item;
f2984462
CM		 2256 int ret;
2257 int do_barriers;
a236aed1
CM		 2258 int max_errors;
2259 int total_errors = 0;
a061fc8d 2260 u64 flags;
f2984462 2261
a236aed1 2262 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2263 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2264
a061fc8d
CM		 2265 sb = &root->fs_info->super_for_commit;
2266 dev_item = &sb->dev_item;
e5e9a520
CM		 2267
2268 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2269 head = &root->fs_info->fs_devices->devices;
c6e30871 2270 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2271 if (!dev->bdev) {
2272 total_errors++;
2273 continue;
2274 }
2b82032c 2275 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2276 continue;
2277
2b82032c 2278 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2279 btrfs_set_stack_device_type(dev_item, dev->type);
2280 btrfs_set_stack_device_id(dev_item, dev->devid);
2281 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2282 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2283 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2284 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2285 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2286 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2287 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2288
a061fc8d
CM		 2289 flags = btrfs_super_flags(sb);
2290 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2291
a512bbf8 2292 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2293 if (ret)
2294 total_errors++;
f2984462 2295 }
a236aed1 2296 if (total_errors > max_errors) {
d397712b
CM		 2297 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2298 total_errors);
a236aed1
CM		 2299 BUG();
2300 }
f2984462 2301
a512bbf8 2302 total_errors = 0;
c6e30871 2303 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2304 if (!dev->bdev)
2305 continue;
2b82032c 2306 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2307 continue;
2308
a512bbf8
YZ		 2309 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2310 if (ret)
2311 total_errors++;
f2984462 2312 }
e5e9a520 2313 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2314 if (total_errors > max_errors) {
d397712b
CM		 2315 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2316 total_errors);
a236aed1
CM		 2317 BUG();
2318 }
f2984462
CM		 2319 return 0;
2320}
2321
a512bbf8
YZ		 2322int write_ctree_super(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root, int max_mirrors)
eb60ceac 2324{
e66f709b 2325 int ret;
5f39d397 2326
a512bbf8 2327 ret = write_all_supers(root, max_mirrors);
5f39d397 2328 return ret;
cfaa7295
CM		 2329}
2330
5eda7b5e 2331int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2332{
4df27c4d 2333 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2334 radix_tree_delete(&fs_info->fs_roots_radix,
2335 (unsigned long)root->root_key.objectid);
4df27c4d 2336 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2337
2338 if (btrfs_root_refs(&root->root_item) == 0)
2339 synchronize_srcu(&fs_info->subvol_srcu);
2340
4df27c4d
YZ		 2341 free_fs_root(root);
2342 return 0;
2343}
2344
2345static void free_fs_root(struct btrfs_root *root)
2346{
2347 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2348 if (root->anon_super.s_dev) {
2349 down_write(&root->anon_super.s_umount);
2350 kill_anon_super(&root->anon_super);
2351 }
4df27c4d
YZ		 2352 free_extent_buffer(root->node);
2353 free_extent_buffer(root->commit_root);
d397712b 2354 kfree(root->name);
2619ba1f 2355 kfree(root);
2619ba1f
CM		 2356}
2357
35b7e476 2358static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2359{
2360 int ret;
2361 struct btrfs_root *gang[8];
2362 int i;
2363
76dda93c
YZ		 2364 while (!list_empty(&fs_info->dead_roots)) {
2365 gang[0] = list_entry(fs_info->dead_roots.next,
2366 struct btrfs_root, root_list);
2367 list_del(&gang[0]->root_list);
2368
2369 if (gang[0]->in_radix) {
2370 btrfs_free_fs_root(fs_info, gang[0]);
2371 } else {
2372 free_extent_buffer(gang[0]->node);
2373 free_extent_buffer(gang[0]->commit_root);
2374 kfree(gang[0]);
2375 }
2376 }
2377
d397712b 2378 while (1) {
0f7d52f4
CM		 2379 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2380 (void **)gang, 0,
2381 ARRAY_SIZE(gang));
2382 if (!ret)
2383 break;
2619ba1f 2384 for (i = 0; i < ret; i++)
5eda7b5e 2385 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2386 }
2387 return 0;
2388}
b4100d64 2389
c146afad 2390int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2391{
c146afad
YZ		 2392 u64 root_objectid = 0;
2393 struct btrfs_root *gang[8];
2394 int i;
3768f368 2395 int ret;
e089f05c 2396
c146afad
YZ		 2397 while (1) {
2398 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2399 (void **)gang, root_objectid,
2400 ARRAY_SIZE(gang));
2401 if (!ret)
2402 break;
5d4f98a2
YZ		 2403
2404 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad
YZ		 2405 for (i = 0; i < ret; i++) {
2406 root_objectid = gang[i]->root_key.objectid;
c146afad
YZ		 2407 btrfs_orphan_cleanup(gang[i]);
2408 }
2409 root_objectid++;
2410 }
2411 return 0;
2412}
a2135011 2413
c146afad
YZ		 2414int btrfs_commit_super(struct btrfs_root *root)
2415{
2416 struct btrfs_trans_handle *trans;
2417 int ret;
a74a4b97 2418
c146afad 2419 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2420 btrfs_run_delayed_iputs(root);
a74a4b97 2421 btrfs_clean_old_snapshots(root);
c146afad 2422 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2423
2424 /* wait until ongoing cleanup work done */
2425 down_write(&root->fs_info->cleanup_work_sem);
2426 up_write(&root->fs_info->cleanup_work_sem);
2427
a22285a6 2428 trans = btrfs_join_transaction(root, 1);
54aa1f4d 2429 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2430 BUG_ON(ret);
2431 /* run commit again to drop the original snapshot */
a22285a6 2432 trans = btrfs_join_transaction(root, 1);
79154b1b
CM		 2433 btrfs_commit_transaction(trans, root);
2434 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2435 BUG_ON(ret);
d6bfde87 2436
a512bbf8 2437 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2438 return ret;
2439}
2440
2441int close_ctree(struct btrfs_root *root)
2442{
2443 struct btrfs_fs_info *fs_info = root->fs_info;
2444 int ret;
2445
2446 fs_info->closing = 1;
2447 smp_mb();
2448
0af3d00b 2449 btrfs_put_block_group_cache(fs_info);
c146afad
YZ		 2450 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2451 ret = btrfs_commit_super(root);
d397712b
CM		 2452 if (ret)
2453 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2454 }
0f7d52f4 2455
8929ecfa
YZ		 2456 kthread_stop(root->fs_info->transaction_kthread);
2457 kthread_stop(root->fs_info->cleaner_kthread);
2458
f25784b3
YZ		 2459 fs_info->closing = 2;
2460 smp_mb();
2461
b0c68f8b 2462 if (fs_info->delalloc_bytes) {
d397712b 2463 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2464 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2465 }
31153d81 2466 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2467 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2468 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2469 }
bcc63abb 2470
5d4f98a2
YZ		 2471 free_extent_buffer(fs_info->extent_root->node);
2472 free_extent_buffer(fs_info->extent_root->commit_root);
2473 free_extent_buffer(fs_info->tree_root->node);
2474 free_extent_buffer(fs_info->tree_root->commit_root);
2475 free_extent_buffer(root->fs_info->chunk_root->node);
2476 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2477 free_extent_buffer(root->fs_info->dev_root->node);
2478 free_extent_buffer(root->fs_info->dev_root->commit_root);
2479 free_extent_buffer(root->fs_info->csum_root->node);
2480 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2481
9078a3e1 2482 btrfs_free_block_groups(root->fs_info);
d10c5f31 2483
c146afad 2484 del_fs_roots(fs_info);
d10c5f31 2485
c146afad 2486 iput(fs_info->btree_inode);
9ad6b7bc 2487
61d92c32 2488 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2489 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2490 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2491 btrfs_stop_workers(&fs_info->workers);
2492 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2493 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2494 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2495 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2496 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2497 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2498
dfe25020 2499 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2500 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2501
04160088 2502 bdi_destroy(&fs_info->bdi);
76dda93c 2503 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2504
0f7d52f4 2505 kfree(fs_info->extent_root);
0f7d52f4 2506 kfree(fs_info->tree_root);
0b86a832
CM		 2507 kfree(fs_info->chunk_root);
2508 kfree(fs_info->dev_root);
d20f7043 2509 kfree(fs_info->csum_root);
eb60ceac
CM		 2510 return 0;
2511}
2512
1259ab75 2513int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2514{
1259ab75 2515 int ret;
810191ff 2516 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2517
2ac55d41
JB		 2518 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2519 NULL);
1259ab75
CM		 2520 if (!ret)
2521 return ret;
2522
2523 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2524 parent_transid);
2525 return !ret;
5f39d397
CM		 2526}
2527
2528int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2529{
810191ff 2530 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2531 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2532 buf);
2533}
6702ed49 2534
5f39d397
CM		 2535void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2536{
810191ff 2537 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2538 u64 transid = btrfs_header_generation(buf);
2539 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2540 int was_dirty;
b4ce94de 2541
b9447ef8 2542 btrfs_assert_tree_locked(buf);
ccd467d6 2543 if (transid != root->fs_info->generation) {
d397712b
CM		 2544 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2545 "found %llu running %llu\n",
db94535d 2546 (unsigned long long)buf->start,
d397712b
CM		 2547 (unsigned long long)transid,
2548 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2549 WARN_ON(1);
2550 }
b9473439
CM		 2551 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2552 buf);
2553 if (!was_dirty) {
2554 spin_lock(&root->fs_info->delalloc_lock);
2555 root->fs_info->dirty_metadata_bytes += buf->len;
2556 spin_unlock(&root->fs_info->delalloc_lock);
2557 }
eb60ceac
CM		 2558}
2559
d3c2fdcf 2560void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2561{
188de649
CM		 2562 /*
2563 * looks as though older kernels can get into trouble with
2564 * this code, they end up stuck in balance_dirty_pages forever
2565 */
d6bfde87 2566 u64 num_dirty;
771ed689 2567 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2568
6933c02e 2569 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2570 return;
2571
585ad2c3
CM		 2572 num_dirty = root->fs_info->dirty_metadata_bytes;
2573
d6bfde87
CM		 2574 if (num_dirty > thresh) {
2575 balance_dirty_pages_ratelimited_nr(
d7fc640e 2576 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2577 }
188de649 2578 return;
35b7e476 2579}
6b80053d 2580
ca7a79ad 2581int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2582{
810191ff 2583 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2584 int ret;
ca7a79ad 2585 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2586 if (ret == 0)
b4ce94de 2587 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2588 return ret;
6b80053d 2589}
0da5468f 2590
4bef0848
CM		 2591int btree_lock_page_hook(struct page *page)
2592{
2593 struct inode *inode = page->mapping->host;
b9473439 2594 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2595 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2596 struct extent_buffer *eb;
2597 unsigned long len;
2598 u64 bytenr = page_offset(page);
2599
2600 if (page->private == EXTENT_PAGE_PRIVATE)
2601 goto out;
2602
2603 len = page->private >> 2;
2604 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2605 if (!eb)
2606 goto out;
2607
2608 btrfs_tree_lock(eb);
4bef0848 2609 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2610
2611 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2612 spin_lock(&root->fs_info->delalloc_lock);
2613 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2614 root->fs_info->dirty_metadata_bytes -= eb->len;
2615 else
2616 WARN_ON(1);
2617 spin_unlock(&root->fs_info->delalloc_lock);
2618 }
2619
4bef0848
CM		 2620 btrfs_tree_unlock(eb);
2621 free_extent_buffer(eb);
2622out:
2623 lock_page(page);
2624 return 0;
2625}
2626
d1310b2e 2627static struct extent_io_ops btree_extent_io_ops = {
4bef0848 2628 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 2629 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 2630 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 2631 /* note we're sharing with inode.c for the merge bio hook */
2632 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 2633};
Linux 6.x block layer and io_uring tree(s)