projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: cleanup some BUG_ON()
[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);
acce952b 47static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
48 int read_only);
49static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
50static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
51static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
52 struct btrfs_root *root);
53static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
54static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
55static int btrfs_destroy_marked_extents(struct btrfs_root *root,
56 struct extent_io_tree *dirty_pages,
57 int mark);
58static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
59 struct extent_io_tree *pinned_extents);
60static int btrfs_cleanup_transaction(struct btrfs_root *root);
ce9adaa5 61
d352ac68
CM		 62/*
63 * end_io_wq structs are used to do processing in task context when an IO is
64 * complete. This is used during reads to verify checksums, and it is used
65 * by writes to insert metadata for new file extents after IO is complete.
66 */
ce9adaa5
CM		 67struct end_io_wq {
68 struct bio *bio;
69 bio_end_io_t *end_io;
70 void *private;
71 struct btrfs_fs_info *info;
72 int error;
22c59948 73 int metadata;
ce9adaa5 74 struct list_head list;
8b712842 75 struct btrfs_work work;
ce9adaa5 76};
0da5468f 77
d352ac68
CM		 78/*
79 * async submit bios are used to offload expensive checksumming
80 * onto the worker threads. They checksum file and metadata bios
81 * just before they are sent down the IO stack.
82 */
44b8bd7e
CM		 83struct async_submit_bio {
84 struct inode *inode;
85 struct bio *bio;
86 struct list_head list;
4a69a410
CM		 87 extent_submit_bio_hook_t *submit_bio_start;
88 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 89 int rw;
90 int mirror_num;
c8b97818 91 unsigned long bio_flags;
eaf25d93
CM		 92 /*
93 * bio_offset is optional, can be used if the pages in the bio
94 * can't tell us where in the file the bio should go
95 */
96 u64 bio_offset;
8b712842 97 struct btrfs_work work;
44b8bd7e
CM		 98};
99
4008c04a
CM		 100/* These are used to set the lockdep class on the extent buffer locks.
101 * The class is set by the readpage_end_io_hook after the buffer has
102 * passed csum validation but before the pages are unlocked.
103 *
104 * The lockdep class is also set by btrfs_init_new_buffer on freshly
105 * allocated blocks.
106 *
107 * The class is based on the level in the tree block, which allows lockdep
108 * to know that lower nodes nest inside the locks of higher nodes.
109 *
110 * We also add a check to make sure the highest level of the tree is
111 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
112 * code needs update as well.
113 */
114#ifdef CONFIG_DEBUG_LOCK_ALLOC
115# if BTRFS_MAX_LEVEL != 8
116# error
117# endif
118static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
119static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
120 /* leaf */
121 "btrfs-extent-00",
122 "btrfs-extent-01",
123 "btrfs-extent-02",
124 "btrfs-extent-03",
125 "btrfs-extent-04",
126 "btrfs-extent-05",
127 "btrfs-extent-06",
128 "btrfs-extent-07",
129 /* highest possible level */
130 "btrfs-extent-08",
131};
132#endif
133
d352ac68
CM		 134/*
135 * extents on the btree inode are pretty simple, there's one extent
136 * that covers the entire device
137 */
b2950863
CH		 138static struct extent_map *btree_get_extent(struct inode *inode,
139 struct page *page, size_t page_offset, u64 start, u64 len,
140 int create)
7eccb903 141{
5f39d397
CM		 142 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
143 struct extent_map *em;
144 int ret;
145
890871be 146 read_lock(&em_tree->lock);
d1310b2e 147 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 148 if (em) {
149 em->bdev =
150 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 151 read_unlock(&em_tree->lock);
5f39d397 152 goto out;
a061fc8d 153 }
890871be 154 read_unlock(&em_tree->lock);
7b13b7b1 155
5f39d397
CM		 156 em = alloc_extent_map(GFP_NOFS);
157 if (!em) {
158 em = ERR_PTR(-ENOMEM);
159 goto out;
160 }
161 em->start = 0;
0afbaf8c 162 em->len = (u64)-1;
c8b97818 163 em->block_len = (u64)-1;
5f39d397 164 em->block_start = 0;
a061fc8d 165 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 166
890871be 167 write_lock(&em_tree->lock);
5f39d397
CM		 168 ret = add_extent_mapping(em_tree, em);
169 if (ret == -EEXIST) {
0afbaf8c
CM		 170 u64 failed_start = em->start;
171 u64 failed_len = em->len;
172
5f39d397 173 free_extent_map(em);
7b13b7b1 174 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 175 if (em) {
7b13b7b1 176 ret = 0;
0afbaf8c
CM		 177 } else {
178 em = lookup_extent_mapping(em_tree, failed_start,
179 failed_len);
7b13b7b1 180 ret = -EIO;
0afbaf8c 181 }
5f39d397 182 } else if (ret) {
7b13b7b1
CM		 183 free_extent_map(em);
184 em = NULL;
5f39d397 185 }
890871be 186 write_unlock(&em_tree->lock);
7b13b7b1
CM		 187
188 if (ret)
189 em = ERR_PTR(ret);
5f39d397
CM		 190out:
191 return em;
7eccb903
CM		 192}
193
19c00ddc
CM		 194u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
195{
163e783e 196 return crc32c(seed, data, len);
19c00ddc
CM		 197}
198
199void btrfs_csum_final(u32 crc, char *result)
200{
201 *(__le32 *)result = ~cpu_to_le32(crc);
202}
203
d352ac68
CM		 204/*
205 * compute the csum for a btree block, and either verify it or write it
206 * into the csum field of the block.
207 */
19c00ddc
CM		 208static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
209 int verify)
210{
607d432d
JB		 211 u16 csum_size =
212 btrfs_super_csum_size(&root->fs_info->super_copy);
213 char *result = NULL;
19c00ddc
CM		 214 unsigned long len;
215 unsigned long cur_len;
216 unsigned long offset = BTRFS_CSUM_SIZE;
217 char *map_token = NULL;
218 char *kaddr;
219 unsigned long map_start;
220 unsigned long map_len;
221 int err;
222 u32 crc = ~(u32)0;
607d432d 223 unsigned long inline_result;
19c00ddc
CM		 224
225 len = buf->len - offset;
d397712b 226 while (len > 0) {
19c00ddc
CM		 227 err = map_private_extent_buffer(buf, offset, 32,
228 &map_token, &kaddr,
229 &map_start, &map_len, KM_USER0);
d397712b 230 if (err)
19c00ddc 231 return 1;
19c00ddc
CM		 232 cur_len = min(len, map_len - (offset - map_start));
233 crc = btrfs_csum_data(root, kaddr + offset - map_start,
234 crc, cur_len);
235 len -= cur_len;
236 offset += cur_len;
237 unmap_extent_buffer(buf, map_token, KM_USER0);
238 }
607d432d
JB		 239 if (csum_size > sizeof(inline_result)) {
240 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
241 if (!result)
242 return 1;
243 } else {
244 result = (char *)&inline_result;
245 }
246
19c00ddc
CM		 247 btrfs_csum_final(crc, result);
248
249 if (verify) {
607d432d 250 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 251 u32 val;
252 u32 found = 0;
607d432d 253 memcpy(&found, result, csum_size);
e4204ded 254
607d432d 255 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 256 if (printk_ratelimit()) {
257 printk(KERN_INFO "btrfs: %s checksum verify "
258 "failed on %llu wanted %X found %X "
259 "level %d\n",
260 root->fs_info->sb->s_id,
261 (unsigned long long)buf->start, val, found,
262 btrfs_header_level(buf));
263 }
607d432d
JB		 264 if (result != (char *)&inline_result)
265 kfree(result);
19c00ddc
CM		 266 return 1;
267 }
268 } else {
607d432d 269 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 270 }
607d432d
JB		 271 if (result != (char *)&inline_result)
272 kfree(result);
19c00ddc
CM		 273 return 0;
274}
275
d352ac68
CM		 276/*
277 * we can't consider a given block up to date unless the transid of the
278 * block matches the transid in the parent node's pointer. This is how we
279 * detect blocks that either didn't get written at all or got written
280 * in the wrong place.
281 */
1259ab75
CM		 282static int verify_parent_transid(struct extent_io_tree *io_tree,
283 struct extent_buffer *eb, u64 parent_transid)
284{
2ac55d41 285 struct extent_state *cached_state = NULL;
1259ab75
CM		 286 int ret;
287
288 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
289 return 0;
290
2ac55d41
JB		 291 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
292 0, &cached_state, GFP_NOFS);
293 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 294 btrfs_header_generation(eb) == parent_transid) {
295 ret = 0;
296 goto out;
297 }
193f284d
CM		 298 if (printk_ratelimit()) {
299 printk("parent transid verify failed on %llu wanted %llu "
300 "found %llu\n",
301 (unsigned long long)eb->start,
302 (unsigned long long)parent_transid,
303 (unsigned long long)btrfs_header_generation(eb));
304 }
1259ab75 305 ret = 1;
2ac55d41 306 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 307out:
2ac55d41
JB		 308 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
309 &cached_state, GFP_NOFS);
1259ab75 310 return ret;
1259ab75
CM		 311}
312
d352ac68
CM		 313/*
314 * helper to read a given tree block, doing retries as required when
315 * the checksums don't match and we have alternate mirrors to try.
316 */
f188591e
CM		 317static int btree_read_extent_buffer_pages(struct btrfs_root *root,
318 struct extent_buffer *eb,
ca7a79ad 319 u64 start, u64 parent_transid)
f188591e
CM		 320{
321 struct extent_io_tree *io_tree;
322 int ret;
323 int num_copies = 0;
324 int mirror_num = 0;
325
a826d6dc 326 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 327 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
328 while (1) {
329 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
330 btree_get_extent, mirror_num);
1259ab75
CM		 331 if (!ret &&
332 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 333 return ret;
d397712b 334
a826d6dc
JB		 335 /*
336 * This buffer's crc is fine, but its contents are corrupted, so
337 * there is no reason to read the other copies, they won't be
338 * any less wrong.
339 */
340 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
341 return ret;
342
f188591e
CM		 343 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
344 eb->start, eb->len);
4235298e 345 if (num_copies == 1)
f188591e 346 return ret;
4235298e 347
f188591e 348 mirror_num++;
4235298e 349 if (mirror_num > num_copies)
f188591e 350 return ret;
f188591e 351 }
f188591e
CM		 352 return -EIO;
353}
19c00ddc 354
d352ac68 355/*
d397712b
CM		 356 * checksum a dirty tree block before IO. This has extra checks to make sure
357 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 358 */
d397712b 359
b2950863 360static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 361{
d1310b2e 362 struct extent_io_tree *tree;
35ebb934 363 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc 364 u64 found_start;
19c00ddc
CM		 365 unsigned long len;
366 struct extent_buffer *eb;
f188591e
CM		 367 int ret;
368
d1310b2e 369 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 370
eb14ab8e
CM		 371 if (page->private == EXTENT_PAGE_PRIVATE) {
372 WARN_ON(1);
19c00ddc 373 goto out;
eb14ab8e
CM		 374 }
375 if (!page->private) {
376 WARN_ON(1);
19c00ddc 377 goto out;
eb14ab8e 378 }
19c00ddc 379 len = page->private >> 2;
d397712b
CM		 380 WARN_ON(len == 0);
381
19c00ddc 382 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
91ca338d
TI		 383 if (eb == NULL) {
384 WARN_ON(1);
385 goto out;
386 }
ca7a79ad
CM		 387 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
388 btrfs_header_generation(eb));
f188591e 389 BUG_ON(ret);
784b4e29
CM		 390 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
391
19c00ddc
CM		 392 found_start = btrfs_header_bytenr(eb);
393 if (found_start != start) {
55c69072
CM		 394 WARN_ON(1);
395 goto err;
396 }
397 if (eb->first_page != page) {
55c69072
CM		 398 WARN_ON(1);
399 goto err;
400 }
401 if (!PageUptodate(page)) {
55c69072
CM		 402 WARN_ON(1);
403 goto err;
19c00ddc 404 }
19c00ddc 405 csum_tree_block(root, eb, 0);
55c69072 406err:
19c00ddc
CM		 407 free_extent_buffer(eb);
408out:
409 return 0;
410}
411
2b82032c
YZ		 412static int check_tree_block_fsid(struct btrfs_root *root,
413 struct extent_buffer *eb)
414{
415 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
416 u8 fsid[BTRFS_UUID_SIZE];
417 int ret = 1;
418
419 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
420 BTRFS_FSID_SIZE);
421 while (fs_devices) {
422 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
423 ret = 0;
424 break;
425 }
426 fs_devices = fs_devices->seed;
427 }
428 return ret;
429}
430
a826d6dc
JB		 431#define CORRUPT(reason, eb, root, slot) \
432 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
433 "root=%llu, slot=%d\n", reason, \
434 (unsigned long long)btrfs_header_bytenr(eb), \
435 (unsigned long long)root->objectid, slot)
436
437static noinline int check_leaf(struct btrfs_root *root,
438 struct extent_buffer *leaf)
439{
440 struct btrfs_key key;
441 struct btrfs_key leaf_key;
442 u32 nritems = btrfs_header_nritems(leaf);
443 int slot;
444
445 if (nritems == 0)
446 return 0;
447
448 /* Check the 0 item */
449 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
450 BTRFS_LEAF_DATA_SIZE(root)) {
451 CORRUPT("invalid item offset size pair", leaf, root, 0);
452 return -EIO;
453 }
454
455 /*
456 * Check to make sure each items keys are in the correct order and their
457 * offsets make sense. We only have to loop through nritems-1 because
458 * we check the current slot against the next slot, which verifies the
459 * next slot's offset+size makes sense and that the current's slot
460 * offset is correct.
461 */
462 for (slot = 0; slot < nritems - 1; slot++) {
463 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
464 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
465
466 /* Make sure the keys are in the right order */
467 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
468 CORRUPT("bad key order", leaf, root, slot);
469 return -EIO;
470 }
471
472 /*
473 * Make sure the offset and ends are right, remember that the
474 * item data starts at the end of the leaf and grows towards the
475 * front.
476 */
477 if (btrfs_item_offset_nr(leaf, slot) !=
478 btrfs_item_end_nr(leaf, slot + 1)) {
479 CORRUPT("slot offset bad", leaf, root, slot);
480 return -EIO;
481 }
482
483 /*
484 * Check to make sure that we don't point outside of the leaf,
485 * just incase all the items are consistent to eachother, but
486 * all point outside of the leaf.
487 */
488 if (btrfs_item_end_nr(leaf, slot) >
489 BTRFS_LEAF_DATA_SIZE(root)) {
490 CORRUPT("slot end outside of leaf", leaf, root, slot);
491 return -EIO;
492 }
493 }
494
495 return 0;
496}
497
4008c04a
CM		 498#ifdef CONFIG_DEBUG_LOCK_ALLOC
499void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
500{
501 lockdep_set_class_and_name(&eb->lock,
502 &btrfs_eb_class[level],
503 btrfs_eb_name[level]);
504}
505#endif
506
b2950863 507static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 508 struct extent_state *state)
509{
510 struct extent_io_tree *tree;
511 u64 found_start;
512 int found_level;
513 unsigned long len;
514 struct extent_buffer *eb;
515 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 516 int ret = 0;
ce9adaa5
CM		 517
518 tree = &BTRFS_I(page->mapping->host)->io_tree;
519 if (page->private == EXTENT_PAGE_PRIVATE)
520 goto out;
521 if (!page->private)
522 goto out;
d397712b 523
ce9adaa5 524 len = page->private >> 2;
d397712b
CM		 525 WARN_ON(len == 0);
526
ce9adaa5 527 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
91ca338d
TI		 528 if (eb == NULL) {
529 ret = -EIO;
530 goto out;
531 }
f188591e 532
ce9adaa5 533 found_start = btrfs_header_bytenr(eb);
23a07867 534 if (found_start != start) {
193f284d
CM		 535 if (printk_ratelimit()) {
536 printk(KERN_INFO "btrfs bad tree block start "
537 "%llu %llu\n",
538 (unsigned long long)found_start,
539 (unsigned long long)eb->start);
540 }
f188591e 541 ret = -EIO;
ce9adaa5
CM		 542 goto err;
543 }
544 if (eb->first_page != page) {
d397712b
CM		 545 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
546 eb->first_page->index, page->index);
ce9adaa5 547 WARN_ON(1);
f188591e 548 ret = -EIO;
ce9adaa5
CM		 549 goto err;
550 }
2b82032c 551 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 552 if (printk_ratelimit()) {
553 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
554 (unsigned long long)eb->start);
555 }
1259ab75
CM		 556 ret = -EIO;
557 goto err;
558 }
ce9adaa5
CM		 559 found_level = btrfs_header_level(eb);
560
4008c04a
CM		 561 btrfs_set_buffer_lockdep_class(eb, found_level);
562
ce9adaa5 563 ret = csum_tree_block(root, eb, 1);
a826d6dc 564 if (ret) {
f188591e 565 ret = -EIO;
a826d6dc
JB		 566 goto err;
567 }
568
569 /*
570 * If this is a leaf block and it is corrupt, set the corrupt bit so
571 * that we don't try and read the other copies of this block, just
572 * return -EIO.
573 */
574 if (found_level == 0 && check_leaf(root, eb)) {
575 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
576 ret = -EIO;
577 }
ce9adaa5
CM		 578
579 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
580 end = eb->start + end - 1;
ce9adaa5
CM		 581err:
582 free_extent_buffer(eb);
583out:
f188591e 584 return ret;
ce9adaa5
CM		 585}
586
ce9adaa5 587static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 588{
589 struct end_io_wq *end_io_wq = bio->bi_private;
590 struct btrfs_fs_info *fs_info;
ce9adaa5 591
ce9adaa5 592 fs_info = end_io_wq->info;
ce9adaa5 593 end_io_wq->error = err;
8b712842
CM		 594 end_io_wq->work.func = end_workqueue_fn;
595 end_io_wq->work.flags = 0;
d20f7043 596
7b6d91da 597 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 598 if (end_io_wq->metadata == 1)
cad321ad
CM		 599 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
600 &end_io_wq->work);
0cb59c99
JB		 601 else if (end_io_wq->metadata == 2)
602 btrfs_queue_worker(&fs_info->endio_freespace_worker,
603 &end_io_wq->work);
cad321ad
CM		 604 else
605 btrfs_queue_worker(&fs_info->endio_write_workers,
606 &end_io_wq->work);
d20f7043
CM		 607 } else {
608 if (end_io_wq->metadata)
609 btrfs_queue_worker(&fs_info->endio_meta_workers,
610 &end_io_wq->work);
611 else
612 btrfs_queue_worker(&fs_info->endio_workers,
613 &end_io_wq->work);
614 }
ce9adaa5
CM		 615}
616
0cb59c99
JB		 617/*
618 * For the metadata arg you want
619 *
620 * 0 - if data
621 * 1 - if normal metadta
622 * 2 - if writing to the free space cache area
623 */
22c59948
CM		 624int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
625 int metadata)
0b86a832 626{
ce9adaa5 627 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 628 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
629 if (!end_io_wq)
630 return -ENOMEM;
631
632 end_io_wq->private = bio->bi_private;
633 end_io_wq->end_io = bio->bi_end_io;
22c59948 634 end_io_wq->info = info;
ce9adaa5
CM		 635 end_io_wq->error = 0;
636 end_io_wq->bio = bio;
22c59948 637 end_io_wq->metadata = metadata;
ce9adaa5
CM		 638
639 bio->bi_private = end_io_wq;
640 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 641 return 0;
642}
643
b64a2851 644unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 645{
4854ddd0
CM		 646 unsigned long limit = min_t(unsigned long,
647 info->workers.max_workers,
648 info->fs_devices->open_devices);
649 return 256 * limit;
650}
0986fe9e 651
4854ddd0
CM		 652int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
653{
b64a2851
CM		 654 return atomic_read(&info->nr_async_bios) >
655 btrfs_async_submit_limit(info);
0986fe9e
CM		 656}
657
4a69a410
CM		 658static void run_one_async_start(struct btrfs_work *work)
659{
4a69a410
CM		 660 struct async_submit_bio *async;
661
662 async = container_of(work, struct async_submit_bio, work);
4a69a410 663 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM		 664 async->mirror_num, async->bio_flags,
665 async->bio_offset);
4a69a410
CM		 666}
667
668static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 669{
670 struct btrfs_fs_info *fs_info;
671 struct async_submit_bio *async;
4854ddd0 672 int limit;
8b712842
CM		 673
674 async = container_of(work, struct async_submit_bio, work);
675 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 676
b64a2851 677 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 678 limit = limit * 2 / 3;
679
8b712842 680 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 681
b64a2851
CM		 682 if (atomic_read(&fs_info->nr_async_submits) < limit &&
683 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 684 wake_up(&fs_info->async_submit_wait);
685
4a69a410 686 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 687 async->mirror_num, async->bio_flags,
688 async->bio_offset);
4a69a410
CM		 689}
690
691static void run_one_async_free(struct btrfs_work *work)
692{
693 struct async_submit_bio *async;
694
695 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 696 kfree(async);
697}
698
44b8bd7e
CM		 699int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
700 int rw, struct bio *bio, int mirror_num,
c8b97818 701 unsigned long bio_flags,
eaf25d93 702 u64 bio_offset,
4a69a410
CM		 703 extent_submit_bio_hook_t *submit_bio_start,
704 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 705{
706 struct async_submit_bio *async;
707
708 async = kmalloc(sizeof(*async), GFP_NOFS);
709 if (!async)
710 return -ENOMEM;
711
712 async->inode = inode;
713 async->rw = rw;
714 async->bio = bio;
715 async->mirror_num = mirror_num;
4a69a410
CM		 716 async->submit_bio_start = submit_bio_start;
717 async->submit_bio_done = submit_bio_done;
718
719 async->work.func = run_one_async_start;
720 async->work.ordered_func = run_one_async_done;
721 async->work.ordered_free = run_one_async_free;
722
8b712842 723 async->work.flags = 0;
c8b97818 724 async->bio_flags = bio_flags;
eaf25d93 725 async->bio_offset = bio_offset;
8c8bee1d 726
cb03c743 727 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 728
7b6d91da 729 if (rw & REQ_SYNC)
d313d7a3
CM		 730 btrfs_set_work_high_prio(&async->work);
731
8b712842 732 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 733
d397712b 734 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 735 atomic_read(&fs_info->nr_async_submits)) {
736 wait_event(fs_info->async_submit_wait,
737 (atomic_read(&fs_info->nr_async_submits) == 0));
738 }
739
44b8bd7e
CM		 740 return 0;
741}
742
ce3ed71a
CM		 743static int btree_csum_one_bio(struct bio *bio)
744{
745 struct bio_vec *bvec = bio->bi_io_vec;
746 int bio_index = 0;
747 struct btrfs_root *root;
748
749 WARN_ON(bio->bi_vcnt <= 0);
d397712b 750 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 751 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
752 csum_dirty_buffer(root, bvec->bv_page);
753 bio_index++;
754 bvec++;
755 }
756 return 0;
757}
758
4a69a410
CM		 759static int __btree_submit_bio_start(struct inode *inode, int rw,
760 struct bio *bio, int mirror_num,
eaf25d93
CM		 761 unsigned long bio_flags,
762 u64 bio_offset)
22c59948 763{
8b712842
CM		 764 /*
765 * when we're called for a write, we're already in the async
5443be45 766 * submission context. Just jump into btrfs_map_bio
8b712842 767 */
4a69a410
CM		 768 btree_csum_one_bio(bio);
769 return 0;
770}
22c59948 771
4a69a410 772static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 773 int mirror_num, unsigned long bio_flags,
774 u64 bio_offset)
4a69a410 775{
8b712842 776 /*
4a69a410
CM		 777 * when we're called for a write, we're already in the async
778 * submission context. Just jump into btrfs_map_bio
8b712842 779 */
8b712842 780 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 781}
782
44b8bd7e 783static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 784 int mirror_num, unsigned long bio_flags,
785 u64 bio_offset)
44b8bd7e 786{
cad321ad
CM		 787 int ret;
788
789 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
790 bio, 1);
791 BUG_ON(ret);
792
7b6d91da 793 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 794 /*
795 * called for a read, do the setup so that checksum validation
796 * can happen in the async kernel threads
797 */
4a69a410 798 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 799 mirror_num, 0);
44b8bd7e 800 }
d313d7a3 801
cad321ad
CM		 802 /*
803 * kthread helpers are used to submit writes so that checksumming
804 * can happen in parallel across all CPUs
805 */
44b8bd7e 806 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 807 inode, rw, bio, mirror_num, 0,
eaf25d93 808 bio_offset,
4a69a410
CM		 809 __btree_submit_bio_start,
810 __btree_submit_bio_done);
44b8bd7e
CM		 811}
812
3dd1462e 813#ifdef CONFIG_MIGRATION
784b4e29
CM		 814static int btree_migratepage(struct address_space *mapping,
815 struct page *newpage, struct page *page)
816{
817 /*
818 * we can't safely write a btree page from here,
819 * we haven't done the locking hook
820 */
821 if (PageDirty(page))
822 return -EAGAIN;
823 /*
824 * Buffers may be managed in a filesystem specific way.
825 * We must have no buffers or drop them.
826 */
827 if (page_has_private(page) &&
828 !try_to_release_page(page, GFP_KERNEL))
829 return -EAGAIN;
784b4e29
CM		 830 return migrate_page(mapping, newpage, page);
831}
3dd1462e 832#endif
784b4e29 833
0da5468f
CM		 834static int btree_writepage(struct page *page, struct writeback_control *wbc)
835{
d1310b2e 836 struct extent_io_tree *tree;
b9473439
CM		 837 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
838 struct extent_buffer *eb;
839 int was_dirty;
840
d1310b2e 841 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 842 if (!(current->flags & PF_MEMALLOC)) {
843 return extent_write_full_page(tree, page,
844 btree_get_extent, wbc);
845 }
5443be45 846
b9473439 847 redirty_page_for_writepage(wbc, page);
784b4e29 848 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
b9473439
CM		 849 WARN_ON(!eb);
850
851 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
852 if (!was_dirty) {
853 spin_lock(&root->fs_info->delalloc_lock);
854 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
855 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 856 }
b9473439
CM		 857 free_extent_buffer(eb);
858
859 unlock_page(page);
860 return 0;
5f39d397 861}
0da5468f
CM		 862
863static int btree_writepages(struct address_space *mapping,
864 struct writeback_control *wbc)
865{
d1310b2e
CM		 866 struct extent_io_tree *tree;
867 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 868 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 869 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 870 u64 num_dirty;
24ab9cd8 871 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 872
873 if (wbc->for_kupdate)
874 return 0;
875
b9473439
CM		 876 /* this is a bit racy, but that's ok */
877 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 878 if (num_dirty < thresh)
793955bc 879 return 0;
793955bc 880 }
0da5468f
CM		 881 return extent_writepages(tree, mapping, btree_get_extent, wbc);
882}
883
b2950863 884static int btree_readpage(struct file *file, struct page *page)
5f39d397 885{
d1310b2e
CM		 886 struct extent_io_tree *tree;
887 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 888 return extent_read_full_page(tree, page, btree_get_extent);
889}
22b0ebda 890
70dec807 891static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 892{
d1310b2e
CM		 893 struct extent_io_tree *tree;
894 struct extent_map_tree *map;
5f39d397 895 int ret;
d98237b3 896
98509cfc 897 if (PageWriteback(page) || PageDirty(page))
d397712b 898 return 0;
98509cfc 899
d1310b2e
CM		 900 tree = &BTRFS_I(page->mapping->host)->io_tree;
901 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 902
7b13b7b1 903 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 904 if (!ret)
6af118ce 905 return 0;
6af118ce
CM		 906
907 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 908 if (ret == 1) {
909 ClearPagePrivate(page);
910 set_page_private(page, 0);
911 page_cache_release(page);
912 }
6af118ce 913
d98237b3
CM		 914 return ret;
915}
916
5f39d397 917static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 918{
d1310b2e
CM		 919 struct extent_io_tree *tree;
920 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 921 extent_invalidatepage(tree, page, offset);
922 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 923 if (PagePrivate(page)) {
d397712b
CM		 924 printk(KERN_WARNING "btrfs warning page private not zero "
925 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 926 ClearPagePrivate(page);
927 set_page_private(page, 0);
928 page_cache_release(page);
929 }
d98237b3
CM		 930}
931
7f09410b 932static const struct address_space_operations btree_aops = {
d98237b3
CM		 933 .readpage = btree_readpage,
934 .writepage = btree_writepage,
0da5468f 935 .writepages = btree_writepages,
5f39d397
CM		 936 .releasepage = btree_releasepage,
937 .invalidatepage = btree_invalidatepage,
d98237b3 938 .sync_page = block_sync_page,
5a92bc88 939#ifdef CONFIG_MIGRATION
784b4e29 940 .migratepage = btree_migratepage,
5a92bc88 941#endif
d98237b3
CM		 942};
943
ca7a79ad
CM		 944int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
945 u64 parent_transid)
090d1875 946{
5f39d397
CM		 947 struct extent_buffer *buf = NULL;
948 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 949 int ret = 0;
090d1875 950
db94535d 951 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 952 if (!buf)
090d1875 953 return 0;
d1310b2e 954 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 955 buf, 0, 0, btree_get_extent, 0);
5f39d397 956 free_extent_buffer(buf);
de428b63 957 return ret;
090d1875
CM		 958}
959
0999df54
CM		 960struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
961 u64 bytenr, u32 blocksize)
962{
963 struct inode *btree_inode = root->fs_info->btree_inode;
964 struct extent_buffer *eb;
965 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
966 bytenr, blocksize, GFP_NOFS);
967 return eb;
968}
969
970struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
971 u64 bytenr, u32 blocksize)
972{
973 struct inode *btree_inode = root->fs_info->btree_inode;
974 struct extent_buffer *eb;
975
976 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
977 bytenr, blocksize, NULL, GFP_NOFS);
978 return eb;
979}
980
981
e02119d5
CM		 982int btrfs_write_tree_block(struct extent_buffer *buf)
983{
8aa38c31
CH		 984 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
985 buf->start + buf->len - 1);
e02119d5
CM		 986}
987
988int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
989{
8aa38c31
CH		 990 return filemap_fdatawait_range(buf->first_page->mapping,
991 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 992}
993
0999df54 994struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 995 u32 blocksize, u64 parent_transid)
0999df54
CM		 996{
997 struct extent_buffer *buf = NULL;
0999df54
CM		 998 int ret;
999
0999df54
CM		 1000 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1001 if (!buf)
1002 return NULL;
0999df54 1003
ca7a79ad 1004 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 1005
d397712b 1006 if (ret == 0)
b4ce94de 1007 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 1008 return buf;
ce9adaa5 1009
eb60ceac
CM		 1010}
1011
e089f05c 1012int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 1013 struct extent_buffer *buf)
ed2ff2cb 1014{
5f39d397 1015 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 1016 if (btrfs_header_generation(buf) ==
925baedd 1017 root->fs_info->running_transaction->transid) {
b9447ef8 1018 btrfs_assert_tree_locked(buf);
b4ce94de 1019
b9473439
CM		 1020 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1021 spin_lock(&root->fs_info->delalloc_lock);
1022 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1023 root->fs_info->dirty_metadata_bytes -= buf->len;
1024 else
1025 WARN_ON(1);
1026 spin_unlock(&root->fs_info->delalloc_lock);
1027 }
b4ce94de 1028
b9473439
CM		 1029 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1030 btrfs_set_lock_blocking(buf);
d1310b2e 1031 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 1032 buf);
925baedd 1033 }
5f39d397
CM		 1034 return 0;
1035}
1036
db94535d 1037static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 1038 u32 stripesize, struct btrfs_root *root,
9f5fae2f 1039 struct btrfs_fs_info *fs_info,
e20d96d6 1040 u64 objectid)
d97e63b6 1041{
cfaa7295 1042 root->node = NULL;
a28ec197 1043 root->commit_root = NULL;
db94535d
CM		 1044 root->sectorsize = sectorsize;
1045 root->nodesize = nodesize;
1046 root->leafsize = leafsize;
87ee04eb 1047 root->stripesize = stripesize;
123abc88 1048 root->ref_cows = 0;
0b86a832 1049 root->track_dirty = 0;
c71bf099 1050 root->in_radix = 0;
d68fc57b
YZ		 1051 root->orphan_item_inserted = 0;
1052 root->orphan_cleanup_state = 0;
0b86a832 1053
9f5fae2f 1054 root->fs_info = fs_info;
0f7d52f4
CM		 1055 root->objectid = objectid;
1056 root->last_trans = 0;
13a8a7c8 1057 root->highest_objectid = 0;
58176a96 1058 root->name = NULL;
4313b399 1059 root->in_sysfs = 0;
6bef4d31 1060 root->inode_tree = RB_ROOT;
f0486c68 1061 root->block_rsv = NULL;
d68fc57b 1062 root->orphan_block_rsv = NULL;
0b86a832
CM		 1063
1064 INIT_LIST_HEAD(&root->dirty_list);
7b128766 1065 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 1066 INIT_LIST_HEAD(&root->root_list);
925baedd 1067 spin_lock_init(&root->node_lock);
d68fc57b 1068 spin_lock_init(&root->orphan_lock);
5d4f98a2 1069 spin_lock_init(&root->inode_lock);
f0486c68 1070 spin_lock_init(&root->accounting_lock);
a2135011 1071 mutex_init(&root->objectid_mutex);
e02119d5 1072 mutex_init(&root->log_mutex);
7237f183
YZ		 1073 init_waitqueue_head(&root->log_writer_wait);
1074 init_waitqueue_head(&root->log_commit_wait[0]);
1075 init_waitqueue_head(&root->log_commit_wait[1]);
1076 atomic_set(&root->log_commit[0], 0);
1077 atomic_set(&root->log_commit[1], 0);
1078 atomic_set(&root->log_writers, 0);
1079 root->log_batch = 0;
1080 root->log_transid = 0;
257c62e1 1081 root->last_log_commit = 0;
d0c803c4
CM		 1082 extent_io_tree_init(&root->dirty_log_pages,
1083 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 1084
3768f368
CM		 1085 memset(&root->root_key, 0, sizeof(root->root_key));
1086 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1087 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1088 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1089 root->defrag_trans_start = fs_info->generation;
58176a96 1090 init_completion(&root->kobj_unregister);
6702ed49 1091 root->defrag_running = 0;
4d775673 1092 root->root_key.objectid = objectid;
3394e160
CM		 1093 root->anon_super.s_root = NULL;
1094 root->anon_super.s_dev = 0;
1095 INIT_LIST_HEAD(&root->anon_super.s_list);
1096 INIT_LIST_HEAD(&root->anon_super.s_instances);
1097 init_rwsem(&root->anon_super.s_umount);
1098
3768f368
CM		 1099 return 0;
1100}
1101
db94535d 1102static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 1103 struct btrfs_fs_info *fs_info,
1104 u64 objectid,
e20d96d6 1105 struct btrfs_root *root)
3768f368
CM		 1106{
1107 int ret;
db94535d 1108 u32 blocksize;
84234f3a 1109 u64 generation;
3768f368 1110
db94535d 1111 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1112 tree_root->sectorsize, tree_root->stripesize,
1113 root, fs_info, objectid);
3768f368
CM		 1114 ret = btrfs_find_last_root(tree_root, objectid,
1115 &root->root_item, &root->root_key);
4df27c4d
YZ		 1116 if (ret > 0)
1117 return -ENOENT;
3768f368
CM		 1118 BUG_ON(ret);
1119
84234f3a 1120 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1121 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1122 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1123 blocksize, generation);
68433b73
CM		 1124 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1125 free_extent_buffer(root->node);
1126 return -EIO;
1127 }
4df27c4d 1128 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1129 return 0;
1130}
1131
7237f183
YZ		 1132static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1133 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1134{
1135 struct btrfs_root *root;
1136 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1137 struct extent_buffer *leaf;
e02119d5
CM		 1138
1139 root = kzalloc(sizeof(*root), GFP_NOFS);
1140 if (!root)
7237f183 1141 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1142
1143 __setup_root(tree_root->nodesize, tree_root->leafsize,
1144 tree_root->sectorsize, tree_root->stripesize,
1145 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1146
1147 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1148 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1149 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1150 /*
1151 * log trees do not get reference counted because they go away
1152 * before a real commit is actually done. They do store pointers
1153 * to file data extents, and those reference counts still get
1154 * updated (along with back refs to the log tree).
1155 */
e02119d5
CM		 1156 root->ref_cows = 0;
1157
5d4f98a2
YZ		 1158 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1159 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1160 if (IS_ERR(leaf)) {
1161 kfree(root);
1162 return ERR_CAST(leaf);
1163 }
e02119d5 1164
5d4f98a2
YZ		 1165 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1166 btrfs_set_header_bytenr(leaf, leaf->start);
1167 btrfs_set_header_generation(leaf, trans->transid);
1168 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1169 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1170 root->node = leaf;
e02119d5
CM		 1171
1172 write_extent_buffer(root->node, root->fs_info->fsid,
1173 (unsigned long)btrfs_header_fsid(root->node),
1174 BTRFS_FSID_SIZE);
1175 btrfs_mark_buffer_dirty(root->node);
1176 btrfs_tree_unlock(root->node);
7237f183
YZ		 1177 return root;
1178}
1179
1180int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1181 struct btrfs_fs_info *fs_info)
1182{
1183 struct btrfs_root *log_root;
1184
1185 log_root = alloc_log_tree(trans, fs_info);
1186 if (IS_ERR(log_root))
1187 return PTR_ERR(log_root);
1188 WARN_ON(fs_info->log_root_tree);
1189 fs_info->log_root_tree = log_root;
1190 return 0;
1191}
1192
1193int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1194 struct btrfs_root *root)
1195{
1196 struct btrfs_root *log_root;
1197 struct btrfs_inode_item *inode_item;
1198
1199 log_root = alloc_log_tree(trans, root->fs_info);
1200 if (IS_ERR(log_root))
1201 return PTR_ERR(log_root);
1202
1203 log_root->last_trans = trans->transid;
1204 log_root->root_key.offset = root->root_key.objectid;
1205
1206 inode_item = &log_root->root_item.inode;
1207 inode_item->generation = cpu_to_le64(1);
1208 inode_item->size = cpu_to_le64(3);
1209 inode_item->nlink = cpu_to_le32(1);
1210 inode_item->nbytes = cpu_to_le64(root->leafsize);
1211 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1212
5d4f98a2 1213 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1214
1215 WARN_ON(root->log_root);
1216 root->log_root = log_root;
1217 root->log_transid = 0;
257c62e1 1218 root->last_log_commit = 0;
e02119d5
CM		 1219 return 0;
1220}
1221
1222struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1223 struct btrfs_key *location)
1224{
1225 struct btrfs_root *root;
1226 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1227 struct btrfs_path *path;
5f39d397 1228 struct extent_buffer *l;
84234f3a 1229 u64 generation;
db94535d 1230 u32 blocksize;
0f7d52f4
CM		 1231 int ret = 0;
1232
5eda7b5e 1233 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1234 if (!root)
0f7d52f4 1235 return ERR_PTR(-ENOMEM);
0f7d52f4 1236 if (location->offset == (u64)-1) {
db94535d 1237 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1238 location->objectid, root);
1239 if (ret) {
0f7d52f4
CM		 1240 kfree(root);
1241 return ERR_PTR(ret);
1242 }
13a8a7c8 1243 goto out;
0f7d52f4
CM		 1244 }
1245
db94535d 1246 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1247 tree_root->sectorsize, tree_root->stripesize,
1248 root, fs_info, location->objectid);
0f7d52f4
CM		 1249
1250 path = btrfs_alloc_path();
db5b493a
TI		 1251 if (!path) {
1252 kfree(root);
1253 return ERR_PTR(-ENOMEM);
1254 }
0f7d52f4 1255 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1256 if (ret == 0) {
1257 l = path->nodes[0];
1258 read_extent_buffer(l, &root->root_item,
1259 btrfs_item_ptr_offset(l, path->slots[0]),
1260 sizeof(root->root_item));
1261 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1262 }
0f7d52f4
CM		 1263 btrfs_free_path(path);
1264 if (ret) {
5e540f77 1265 kfree(root);
13a8a7c8
YZ		 1266 if (ret > 0)
1267 ret = -ENOENT;
0f7d52f4
CM		 1268 return ERR_PTR(ret);
1269 }
13a8a7c8 1270
84234f3a 1271 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1272 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1273 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1274 blocksize, generation);
5d4f98a2 1275 root->commit_root = btrfs_root_node(root);
0f7d52f4 1276 BUG_ON(!root->node);
13a8a7c8
YZ		 1277out:
1278 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
e02119d5 1279 root->ref_cows = 1;
13a8a7c8 1280
5eda7b5e
CM		 1281 return root;
1282}
1283
dc17ff8f
CM		 1284struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1285 u64 root_objectid)
1286{
1287 struct btrfs_root *root;
1288
1289 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1290 return fs_info->tree_root;
1291 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1292 return fs_info->extent_root;
1293
1294 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1295 (unsigned long)root_objectid);
1296 return root;
1297}
1298
edbd8d4e
CM		 1299struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1300 struct btrfs_key *location)
5eda7b5e
CM		 1301{
1302 struct btrfs_root *root;
1303 int ret;
1304
edbd8d4e
CM		 1305 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1306 return fs_info->tree_root;
1307 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1308 return fs_info->extent_root;
8f18cf13
CM		 1309 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1310 return fs_info->chunk_root;
1311 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1312 return fs_info->dev_root;
0403e47e
YZ		 1313 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1314 return fs_info->csum_root;
4df27c4d
YZ		 1315again:
1316 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1317 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1318 (unsigned long)location->objectid);
4df27c4d 1319 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1320 if (root)
1321 return root;
1322
e02119d5 1323 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1324 if (IS_ERR(root))
1325 return root;
3394e160
CM		 1326
1327 set_anon_super(&root->anon_super, NULL);
1328
d68fc57b
YZ		 1329 if (btrfs_root_refs(&root->root_item) == 0) {
1330 ret = -ENOENT;
1331 goto fail;
1332 }
1333
1334 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1335 if (ret < 0)
1336 goto fail;
1337 if (ret == 0)
1338 root->orphan_item_inserted = 1;
1339
4df27c4d
YZ		 1340 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1341 if (ret)
1342 goto fail;
1343
1344 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1345 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1346 (unsigned long)root->root_key.objectid,
0f7d52f4 1347 root);
d68fc57b 1348 if (ret == 0)
4df27c4d 1349 root->in_radix = 1;
d68fc57b 1350
4df27c4d
YZ		 1351 spin_unlock(&fs_info->fs_roots_radix_lock);
1352 radix_tree_preload_end();
0f7d52f4 1353 if (ret) {
4df27c4d
YZ		 1354 if (ret == -EEXIST) {
1355 free_fs_root(root);
1356 goto again;
1357 }
1358 goto fail;
0f7d52f4 1359 }
4df27c4d
YZ		 1360
1361 ret = btrfs_find_dead_roots(fs_info->tree_root,
1362 root->root_key.objectid);
1363 WARN_ON(ret);
edbd8d4e 1364 return root;
4df27c4d
YZ		 1365fail:
1366 free_fs_root(root);
1367 return ERR_PTR(ret);
edbd8d4e
CM		 1368}
1369
1370struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1371 struct btrfs_key *location,
1372 const char *name, int namelen)
1373{
4df27c4d
YZ		 1374 return btrfs_read_fs_root_no_name(fs_info, location);
1375#if 0
edbd8d4e
CM		 1376 struct btrfs_root *root;
1377 int ret;
1378
1379 root = btrfs_read_fs_root_no_name(fs_info, location);
1380 if (!root)
1381 return NULL;
58176a96 1382
4313b399
CM		 1383 if (root->in_sysfs)
1384 return root;
1385
58176a96
JB		 1386 ret = btrfs_set_root_name(root, name, namelen);
1387 if (ret) {
5f39d397 1388 free_extent_buffer(root->node);
58176a96
JB		 1389 kfree(root);
1390 return ERR_PTR(ret);
1391 }
4df27c4d 1392
58176a96
JB		 1393 ret = btrfs_sysfs_add_root(root);
1394 if (ret) {
5f39d397 1395 free_extent_buffer(root->node);
58176a96
JB		 1396 kfree(root->name);
1397 kfree(root);
1398 return ERR_PTR(ret);
1399 }
4313b399 1400 root->in_sysfs = 1;
0f7d52f4 1401 return root;
4df27c4d 1402#endif
0f7d52f4 1403}
04160088
CM		 1404
1405static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1406{
1407 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1408 int ret = 0;
04160088
CM		 1409 struct btrfs_device *device;
1410 struct backing_dev_info *bdi;
b7967db7 1411
c6e30871 1412 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1413 if (!device->bdev)
1414 continue;
04160088
CM		 1415 bdi = blk_get_backing_dev_info(device->bdev);
1416 if (bdi && bdi_congested(bdi, bdi_bits)) {
1417 ret = 1;
1418 break;
1419 }
1420 }
1421 return ret;
1422}
1423
38b66988
CM		 1424/*
1425 * this unplugs every device on the box, and it is only used when page
1426 * is null
1427 */
1428static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1429{
38b66988
CM		 1430 struct btrfs_device *device;
1431 struct btrfs_fs_info *info;
1432
1433 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1434 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1435 if (!device->bdev)
1436 continue;
1437
38b66988 1438 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1439 if (bdi->unplug_io_fn)
38b66988 1440 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1441 }
1442}
1443
b2950863 1444static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1445{
38b66988 1446 struct inode *inode;
f2d8d74d
CM		 1447 struct extent_map_tree *em_tree;
1448 struct extent_map *em;
bcbfce8a 1449 struct address_space *mapping;
38b66988
CM		 1450 u64 offset;
1451
bcbfce8a 1452 /* the generic O_DIRECT read code does this */
9f0ba5bd 1453 if (1 || !page) {
38b66988
CM		 1454 __unplug_io_fn(bdi, page);
1455 return;
1456 }
1457
bcbfce8a
CM		 1458 /*
1459 * page->mapping may change at any time. Get a consistent copy
1460 * and use that for everything below
1461 */
1462 smp_mb();
1463 mapping = page->mapping;
1464 if (!mapping)
1465 return;
1466
1467 inode = mapping->host;
240d5d48
CM		 1468
1469 /*
1470 * don't do the expensive searching for a small number of
1471 * devices
1472 */
1473 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1474 __unplug_io_fn(bdi, page);
1475 return;
1476 }
1477
38b66988 1478 offset = page_offset(page);
04160088 1479
f2d8d74d 1480 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1481 read_lock(&em_tree->lock);
f2d8d74d 1482 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1483 read_unlock(&em_tree->lock);
89642229
CM		 1484 if (!em) {
1485 __unplug_io_fn(bdi, page);
f2d8d74d 1486 return;
89642229 1487 }
f2d8d74d 1488
89642229
CM		 1489 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1490 free_extent_map(em);
1491 __unplug_io_fn(bdi, page);
1492 return;
1493 }
f2d8d74d
CM		 1494 offset = offset - em->start;
1495 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1496 em->block_start + offset, page);
1497 free_extent_map(em);
04160088
CM		 1498}
1499
ad081f14
JA		 1500/*
1501 * If this fails, caller must call bdi_destroy() to get rid of the
1502 * bdi again.
1503 */
04160088
CM		 1504static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1505{
ad081f14
JA		 1506 int err;
1507
1508 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1509 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1510 if (err)
1511 return err;
1512
4575c9cc 1513 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1514 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1515 bdi->unplug_io_data = info;
1516 bdi->congested_fn = btrfs_congested_fn;
1517 bdi->congested_data = info;
1518 return 0;
1519}
1520
ce9adaa5
CM		 1521static int bio_ready_for_csum(struct bio *bio)
1522{
1523 u64 length = 0;
1524 u64 buf_len = 0;
1525 u64 start = 0;
1526 struct page *page;
1527 struct extent_io_tree *io_tree = NULL;
ce9adaa5
CM		 1528 struct bio_vec *bvec;
1529 int i;
1530 int ret;
1531
1532 bio_for_each_segment(bvec, bio, i) {
1533 page = bvec->bv_page;
1534 if (page->private == EXTENT_PAGE_PRIVATE) {
1535 length += bvec->bv_len;
1536 continue;
1537 }
1538 if (!page->private) {
1539 length += bvec->bv_len;
1540 continue;
1541 }
1542 length = bvec->bv_len;
1543 buf_len = page->private >> 2;
1544 start = page_offset(page) + bvec->bv_offset;
1545 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
ce9adaa5
CM		 1546 }
1547 /* are we fully contained in this bio? */
1548 if (buf_len <= length)
1549 return 1;
1550
1551 ret = extent_range_uptodate(io_tree, start + length,
1552 start + buf_len - 1);
ce9adaa5
CM		 1553 return ret;
1554}
1555
8b712842
CM		 1556/*
1557 * called by the kthread helper functions to finally call the bio end_io
1558 * functions. This is where read checksum verification actually happens
1559 */
1560static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1561{
ce9adaa5 1562 struct bio *bio;
8b712842
CM		 1563 struct end_io_wq *end_io_wq;
1564 struct btrfs_fs_info *fs_info;
ce9adaa5 1565 int error;
ce9adaa5 1566
8b712842
CM		 1567 end_io_wq = container_of(work, struct end_io_wq, work);
1568 bio = end_io_wq->bio;
1569 fs_info = end_io_wq->info;
ce9adaa5 1570
cad321ad 1571 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1572 * be checksummed at once. This makes sure the entire block is in
1573 * ram and up to date before trying to verify things. For
1574 * blocksize <= pagesize, it is basically a noop
1575 */
7b6d91da 1576 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1577 !bio_ready_for_csum(bio)) {
d20f7043 1578 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1579 &end_io_wq->work);
1580 return;
1581 }
1582 error = end_io_wq->error;
1583 bio->bi_private = end_io_wq->private;
1584 bio->bi_end_io = end_io_wq->end_io;
1585 kfree(end_io_wq);
8b712842 1586 bio_endio(bio, error);
44b8bd7e
CM		 1587}
1588
a74a4b97
CM		 1589static int cleaner_kthread(void *arg)
1590{
1591 struct btrfs_root *root = arg;
1592
1593 do {
a74a4b97 1594 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1595
1596 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1597 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1598 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1599 btrfs_clean_old_snapshots(root);
1600 mutex_unlock(&root->fs_info->cleaner_mutex);
1601 }
a74a4b97
CM		 1602
1603 if (freezing(current)) {
1604 refrigerator();
1605 } else {
a74a4b97 1606 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1607 if (!kthread_should_stop())
1608 schedule();
a74a4b97
CM		 1609 __set_current_state(TASK_RUNNING);
1610 }
1611 } while (!kthread_should_stop());
1612 return 0;
1613}
1614
1615static int transaction_kthread(void *arg)
1616{
1617 struct btrfs_root *root = arg;
1618 struct btrfs_trans_handle *trans;
1619 struct btrfs_transaction *cur;
8929ecfa 1620 u64 transid;
a74a4b97
CM		 1621 unsigned long now;
1622 unsigned long delay;
1623 int ret;
1624
1625 do {
a74a4b97
CM		 1626 delay = HZ * 30;
1627 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1628 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1629
8929ecfa 1630 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1631 cur = root->fs_info->running_transaction;
1632 if (!cur) {
8929ecfa 1633 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1634 goto sleep;
1635 }
31153d81 1636
a74a4b97 1637 now = get_seconds();
8929ecfa
YZ		 1638 if (!cur->blocked &&
1639 (now < cur->start_time || now - cur->start_time < 30)) {
1640 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1641 delay = HZ * 5;
1642 goto sleep;
1643 }
8929ecfa
YZ		 1644 transid = cur->transid;
1645 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1646
8929ecfa 1647 trans = btrfs_join_transaction(root, 1);
3612b495 1648 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 1649 if (transid == trans->transid) {
1650 ret = btrfs_commit_transaction(trans, root);
1651 BUG_ON(ret);
1652 } else {
1653 btrfs_end_transaction(trans, root);
1654 }
a74a4b97
CM		 1655sleep:
1656 wake_up_process(root->fs_info->cleaner_kthread);
1657 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1658
1659 if (freezing(current)) {
1660 refrigerator();
1661 } else {
a74a4b97 1662 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1663 if (!kthread_should_stop() &&
1664 !btrfs_transaction_blocked(root->fs_info))
1665 schedule_timeout(delay);
a74a4b97
CM		 1666 __set_current_state(TASK_RUNNING);
1667 }
1668 } while (!kthread_should_stop());
1669 return 0;
1670}
1671
8a4b83cc 1672struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1673 struct btrfs_fs_devices *fs_devices,
1674 char *options)
2e635a27 1675{
db94535d
CM		 1676 u32 sectorsize;
1677 u32 nodesize;
1678 u32 leafsize;
1679 u32 blocksize;
87ee04eb 1680 u32 stripesize;
84234f3a 1681 u64 generation;
f2b636e8 1682 u64 features;
3de4586c 1683 struct btrfs_key location;
a061fc8d 1684 struct buffer_head *bh;
e02119d5 1685 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1686 GFP_NOFS);
d20f7043
CM		 1687 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1688 GFP_NOFS);
450ba0ea
JB		 1689 struct btrfs_root *tree_root = btrfs_sb(sb);
1690 struct btrfs_fs_info *fs_info = tree_root->fs_info;
e02119d5 1691 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1692 GFP_NOFS);
e02119d5 1693 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1694 GFP_NOFS);
e02119d5
CM		 1695 struct btrfs_root *log_tree_root;
1696
eb60ceac 1697 int ret;
e58ca020 1698 int err = -EINVAL;
4543df7e 1699
2c90e5d6 1700 struct btrfs_super_block *disk_super;
8790d502 1701
0463bb4e 1702 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1703 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1704 err = -ENOMEM;
1705 goto fail;
1706 }
76dda93c
YZ		 1707
1708 ret = init_srcu_struct(&fs_info->subvol_srcu);
1709 if (ret) {
1710 err = ret;
1711 goto fail;
1712 }
1713
1714 ret = setup_bdi(fs_info, &fs_info->bdi);
1715 if (ret) {
1716 err = ret;
1717 goto fail_srcu;
1718 }
1719
1720 fs_info->btree_inode = new_inode(sb);
1721 if (!fs_info->btree_inode) {
1722 err = -ENOMEM;
1723 goto fail_bdi;
1724 }
1725
1726 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1727 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1728 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1729 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1730 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1731 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1732 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1733 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1734 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1735 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1736 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1737 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1738 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1739
58176a96 1740 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1741 fs_info->tree_root = tree_root;
1742 fs_info->extent_root = extent_root;
d20f7043 1743 fs_info->csum_root = csum_root;
0b86a832
CM		 1744 fs_info->chunk_root = chunk_root;
1745 fs_info->dev_root = dev_root;
8a4b83cc 1746 fs_info->fs_devices = fs_devices;
0b86a832 1747 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1748 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1749 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1750 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1751 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1752 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1753 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1754 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1755 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1756 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1757 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1758 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1759 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1760 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1761 fs_info->sb = sb;
6f568d35 1762 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1763 fs_info->metadata_ratio = 0;
c8b97818 1764
b34b086c
CM		 1765 fs_info->thread_pool_size = min_t(unsigned long,
1766 num_online_cpus() + 2, 8);
0afbaf8c 1767
3eaa2885
CM		 1768 INIT_LIST_HEAD(&fs_info->ordered_extents);
1769 spin_lock_init(&fs_info->ordered_extent_lock);
1770
a061fc8d
CM		 1771 sb->s_blocksize = 4096;
1772 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1773 sb->s_bdi = &fs_info->bdi;
a061fc8d 1774
76dda93c
YZ		 1775 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1776 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1777 /*
1778 * we set the i_size on the btree inode to the max possible int.
1779 * the real end of the address space is determined by all of
1780 * the devices in the system
1781 */
1782 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1783 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1784 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1785
5d4f98a2 1786 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1787 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1788 fs_info->btree_inode->i_mapping,
1789 GFP_NOFS);
d1310b2e
CM		 1790 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1791 GFP_NOFS);
1792
1793 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1794
76dda93c
YZ		 1795 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1796 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1797 sizeof(struct btrfs_key));
1798 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1799 insert_inode_hash(fs_info->btree_inode);
76dda93c 1800
0f9dd46c 1801 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1802 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1803
11833d66 1804 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1805 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1806 extent_io_tree_init(&fs_info->freed_extents[1],
1807 fs_info->btree_inode->i_mapping, GFP_NOFS);
1808 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1809 fs_info->do_barriers = 1;
e18e4809 1810
39279cc3 1811
79154b1b 1812 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1813 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1814 mutex_init(&fs_info->tree_log_mutex);
925baedd 1815 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1816 mutex_init(&fs_info->transaction_kthread_mutex);
1817 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1818 mutex_init(&fs_info->volume_mutex);
276e680d 1819 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1820 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1821 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1822
1823 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1824 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1825
e6dcd2dc 1826 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1827 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 1828 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 1829 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1830
0b86a832 1831 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1832 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1833
a512bbf8 1834 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 1835 if (!bh) {
1836 err = -EINVAL;
39279cc3 1837 goto fail_iput;
20b45077 1838 }
39279cc3 1839
a061fc8d 1840 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1841 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1842 sizeof(fs_info->super_for_commit));
a061fc8d 1843 brelse(bh);
5f39d397 1844
a061fc8d 1845 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1846
5f39d397 1847 disk_super = &fs_info->super_copy;
0f7d52f4 1848 if (!btrfs_super_root(disk_super))
c6e2bac1 1849 goto fail_iput;
0f7d52f4 1850
acce952b 1851 /* check FS state, whether FS is broken. */
1852 fs_info->fs_state |= btrfs_super_flags(disk_super);
1853
1854 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1855
2b82032c
YZ		 1856 ret = btrfs_parse_options(tree_root, options);
1857 if (ret) {
1858 err = ret;
c6e2bac1 1859 goto fail_iput;
2b82032c 1860 }
dfe25020 1861
f2b636e8
JB		 1862 features = btrfs_super_incompat_flags(disk_super) &
1863 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1864 if (features) {
1865 printk(KERN_ERR "BTRFS: couldn't mount because of "
1866 "unsupported optional features (%Lx).\n",
21380931 1867 (unsigned long long)features);
f2b636e8 1868 err = -EINVAL;
c6e2bac1 1869 goto fail_iput;
f2b636e8
JB		 1870 }
1871
5d4f98a2 1872 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae
LZ		 1873 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1874 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1875 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1876 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 1877
f2b636e8
JB		 1878 features = btrfs_super_compat_ro_flags(disk_super) &
1879 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1880 if (!(sb->s_flags & MS_RDONLY) && features) {
1881 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1882 "unsupported option features (%Lx).\n",
21380931 1883 (unsigned long long)features);
f2b636e8 1884 err = -EINVAL;
c6e2bac1 1885 goto fail_iput;
f2b636e8 1886 }
61d92c32
CM		 1887
1888 btrfs_init_workers(&fs_info->generic_worker,
1889 "genwork", 1, NULL);
1890
5443be45 1891 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1892 fs_info->thread_pool_size,
1893 &fs_info->generic_worker);
c8b97818 1894
771ed689 1895 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1896 fs_info->thread_pool_size,
1897 &fs_info->generic_worker);
771ed689 1898
5443be45 1899 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1900 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1901 fs_info->thread_pool_size),
1902 &fs_info->generic_worker);
61b49440
CM		 1903
1904 /* a higher idle thresh on the submit workers makes it much more
1905 * likely that bios will be send down in a sane order to the
1906 * devices
1907 */
1908 fs_info->submit_workers.idle_thresh = 64;
53863232 1909
771ed689 1910 fs_info->workers.idle_thresh = 16;
4a69a410 1911 fs_info->workers.ordered = 1;
61b49440 1912
771ed689
CM		 1913 fs_info->delalloc_workers.idle_thresh = 2;
1914 fs_info->delalloc_workers.ordered = 1;
1915
61d92c32
CM		 1916 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1917 &fs_info->generic_worker);
5443be45 1918 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1919 fs_info->thread_pool_size,
1920 &fs_info->generic_worker);
d20f7043 1921 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1922 fs_info->thread_pool_size,
1923 &fs_info->generic_worker);
cad321ad 1924 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1925 "endio-meta-write", fs_info->thread_pool_size,
1926 &fs_info->generic_worker);
5443be45 1927 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1928 fs_info->thread_pool_size,
1929 &fs_info->generic_worker);
0cb59c99
JB		 1930 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1931 1, &fs_info->generic_worker);
61b49440
CM		 1932
1933 /*
1934 * endios are largely parallel and should have a very
1935 * low idle thresh
1936 */
1937 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1938 fs_info->endio_meta_workers.idle_thresh = 4;
1939
9042846b
CM		 1940 fs_info->endio_write_workers.idle_thresh = 2;
1941 fs_info->endio_meta_write_workers.idle_thresh = 2;
1942
4543df7e 1943 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1944 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1945 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1946 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1947 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1948 btrfs_start_workers(&fs_info->endio_workers, 1);
1949 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1950 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1951 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1952 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
4543df7e 1953
4575c9cc 1954 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1955 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1956 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1957
db94535d
CM		 1958 nodesize = btrfs_super_nodesize(disk_super);
1959 leafsize = btrfs_super_leafsize(disk_super);
1960 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1961 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1962 tree_root->nodesize = nodesize;
1963 tree_root->leafsize = leafsize;
1964 tree_root->sectorsize = sectorsize;
87ee04eb 1965 tree_root->stripesize = stripesize;
a061fc8d
CM		 1966
1967 sb->s_blocksize = sectorsize;
1968 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1969
39279cc3
CM		 1970 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1971 sizeof(disk_super->magic))) {
d397712b 1972 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1973 goto fail_sb_buffer;
1974 }
19c00ddc 1975
925baedd 1976 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1977 ret = btrfs_read_sys_array(tree_root);
925baedd 1978 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1979 if (ret) {
d397712b
CM		 1980 printk(KERN_WARNING "btrfs: failed to read the system "
1981 "array on %s\n", sb->s_id);
5d4f98a2 1982 goto fail_sb_buffer;
84eed90f 1983 }
0b86a832
CM		 1984
1985 blocksize = btrfs_level_size(tree_root,
1986 btrfs_super_chunk_root_level(disk_super));
84234f3a 1987 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1988
1989 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1990 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1991
1992 chunk_root->node = read_tree_block(chunk_root,
1993 btrfs_super_chunk_root(disk_super),
84234f3a 1994 blocksize, generation);
0b86a832 1995 BUG_ON(!chunk_root->node);
83121942
DW		 1996 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1997 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1998 sb->s_id);
1999 goto fail_chunk_root;
2000 }
5d4f98a2
YZ		 2001 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2002 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2003
e17cade2 2004 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2005 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2006 BTRFS_UUID_SIZE);
e17cade2 2007
925baedd 2008 mutex_lock(&fs_info->chunk_mutex);
0b86a832 2009 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 2010 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 2011 if (ret) {
d397712b
CM		 2012 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2013 sb->s_id);
2b82032c
YZ		 2014 goto fail_chunk_root;
2015 }
0b86a832 2016
dfe25020
CM		 2017 btrfs_close_extra_devices(fs_devices);
2018
db94535d
CM		 2019 blocksize = btrfs_level_size(tree_root,
2020 btrfs_super_root_level(disk_super));
84234f3a 2021 generation = btrfs_super_generation(disk_super);
0b86a832 2022
e20d96d6 2023 tree_root->node = read_tree_block(tree_root,
db94535d 2024 btrfs_super_root(disk_super),
84234f3a 2025 blocksize, generation);
39279cc3 2026 if (!tree_root->node)
2b82032c 2027 goto fail_chunk_root;
83121942
DW		 2028 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2029 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2030 sb->s_id);
2031 goto fail_tree_root;
2032 }
5d4f98a2
YZ		 2033 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2034 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 2035
2036 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 2037 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 2038 if (ret)
39279cc3 2039 goto fail_tree_root;
0b86a832
CM		 2040 extent_root->track_dirty = 1;
2041
2042 ret = find_and_setup_root(tree_root, fs_info,
2043 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 2044 if (ret)
2045 goto fail_extent_root;
5d4f98a2 2046 dev_root->track_dirty = 1;
3768f368 2047
d20f7043
CM		 2048 ret = find_and_setup_root(tree_root, fs_info,
2049 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2050 if (ret)
5d4f98a2 2051 goto fail_dev_root;
d20f7043
CM		 2052
2053 csum_root->track_dirty = 1;
2054
8929ecfa
YZ		 2055 fs_info->generation = generation;
2056 fs_info->last_trans_committed = generation;
2057 fs_info->data_alloc_profile = (u64)-1;
2058 fs_info->metadata_alloc_profile = (u64)-1;
2059 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2060
1b1d1f66
JB		 2061 ret = btrfs_read_block_groups(extent_root);
2062 if (ret) {
2063 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2064 goto fail_block_groups;
2065 }
9078a3e1 2066
a74a4b97
CM		 2067 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2068 "btrfs-cleaner");
57506d50 2069 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2070 goto fail_block_groups;
a74a4b97
CM		 2071
2072 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2073 tree_root,
2074 "btrfs-transaction");
57506d50 2075 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2076 goto fail_cleaner;
a74a4b97 2077
c289811c
CM		 2078 if (!btrfs_test_opt(tree_root, SSD) &&
2079 !btrfs_test_opt(tree_root, NOSSD) &&
2080 !fs_info->fs_devices->rotating) {
2081 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2082 "mode\n");
2083 btrfs_set_opt(fs_info->mount_opt, SSD);
2084 }
2085
acce952b 2086 /* do not make disk changes in broken FS */
2087 if (btrfs_super_log_root(disk_super) != 0 &&
2088 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
e02119d5
CM		 2089 u64 bytenr = btrfs_super_log_root(disk_super);
2090
7c2ca468 2091 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2092 printk(KERN_WARNING "Btrfs log replay required "
2093 "on RO media\n");
7c2ca468
CM		 2094 err = -EIO;
2095 goto fail_trans_kthread;
2096 }
e02119d5
CM		 2097 blocksize =
2098 btrfs_level_size(tree_root,
2099 btrfs_super_log_root_level(disk_super));
d18a2c44 2100
676e4c86
DC		 2101 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2102 if (!log_tree_root) {
2103 err = -ENOMEM;
2104 goto fail_trans_kthread;
2105 }
e02119d5
CM		 2106
2107 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2108 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2109
2110 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2111 blocksize,
2112 generation + 1);
e02119d5
CM		 2113 ret = btrfs_recover_log_trees(log_tree_root);
2114 BUG_ON(ret);
e556ce2c
YZ		 2115
2116 if (sb->s_flags & MS_RDONLY) {
2117 ret = btrfs_commit_super(tree_root);
2118 BUG_ON(ret);
2119 }
e02119d5 2120 }
1a40e23b 2121
76dda93c
YZ		 2122 ret = btrfs_find_orphan_roots(tree_root);
2123 BUG_ON(ret);
2124
7c2ca468 2125 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ		 2126 ret = btrfs_cleanup_fs_roots(fs_info);
2127 BUG_ON(ret);
2128
5d4f98a2 2129 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2130 if (ret < 0) {
2131 printk(KERN_WARNING
2132 "btrfs: failed to recover relocation\n");
2133 err = -EINVAL;
2134 goto fail_trans_kthread;
2135 }
7c2ca468 2136 }
1a40e23b 2137
3de4586c
CM		 2138 location.objectid = BTRFS_FS_TREE_OBJECTID;
2139 location.type = BTRFS_ROOT_ITEM_KEY;
2140 location.offset = (u64)-1;
2141
3de4586c
CM		 2142 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2143 if (!fs_info->fs_root)
7c2ca468 2144 goto fail_trans_kthread;
3140c9a3
DC		 2145 if (IS_ERR(fs_info->fs_root)) {
2146 err = PTR_ERR(fs_info->fs_root);
2147 goto fail_trans_kthread;
2148 }
c289811c 2149
e3acc2a6
JB		 2150 if (!(sb->s_flags & MS_RDONLY)) {
2151 down_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2152 err = btrfs_orphan_cleanup(fs_info->fs_root);
2153 if (!err)
2154 err = btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6 2155 up_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2156 if (err) {
2157 close_ctree(tree_root);
2158 return ERR_PTR(err);
2159 }
e3acc2a6
JB		 2160 }
2161
0f7d52f4 2162 return tree_root;
39279cc3 2163
7c2ca468
CM		 2164fail_trans_kthread:
2165 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2166fail_cleaner:
a74a4b97 2167 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2168
2169 /*
2170 * make sure we're done with the btree inode before we stop our
2171 * kthreads
2172 */
2173 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2174 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2175
1b1d1f66
JB		 2176fail_block_groups:
2177 btrfs_free_block_groups(fs_info);
d20f7043 2178 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2179 free_extent_buffer(csum_root->commit_root);
2180fail_dev_root:
2181 free_extent_buffer(dev_root->node);
2182 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2183fail_extent_root:
2184 free_extent_buffer(extent_root->node);
5d4f98a2 2185 free_extent_buffer(extent_root->commit_root);
39279cc3 2186fail_tree_root:
5f39d397 2187 free_extent_buffer(tree_root->node);
5d4f98a2 2188 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2189fail_chunk_root:
2190 free_extent_buffer(chunk_root->node);
5d4f98a2 2191 free_extent_buffer(chunk_root->commit_root);
39279cc3 2192fail_sb_buffer:
61d92c32 2193 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2194 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2195 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2196 btrfs_stop_workers(&fs_info->workers);
2197 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2198 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2199 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2200 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2201 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2202 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2203fail_iput:
7c2ca468 2204 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2205 iput(fs_info->btree_inode);
7e662854 2206
dfe25020 2207 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2208 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2209fail_bdi:
7e662854 2210 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2211fail_srcu:
2212 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2213fail:
39279cc3
CM		 2214 kfree(extent_root);
2215 kfree(tree_root);
2216 kfree(fs_info);
83afeac4
JM		 2217 kfree(chunk_root);
2218 kfree(dev_root);
d20f7043 2219 kfree(csum_root);
39279cc3 2220 return ERR_PTR(err);
eb60ceac
CM		 2221}
2222
f2984462
CM		 2223static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2224{
2225 char b[BDEVNAME_SIZE];
2226
2227 if (uptodate) {
2228 set_buffer_uptodate(bh);
2229 } else {
c3b9a62c 2230 if (printk_ratelimit()) {
f2984462
CM		 2231 printk(KERN_WARNING "lost page write due to "
2232 "I/O error on %s\n",
2233 bdevname(bh->b_bdev, b));
2234 }
1259ab75
CM		 2235 /* note, we dont' set_buffer_write_io_error because we have
2236 * our own ways of dealing with the IO errors
2237 */
f2984462
CM		 2238 clear_buffer_uptodate(bh);
2239 }
2240 unlock_buffer(bh);
2241 put_bh(bh);
2242}
2243
a512bbf8
YZ		 2244struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2245{
2246 struct buffer_head *bh;
2247 struct buffer_head *latest = NULL;
2248 struct btrfs_super_block *super;
2249 int i;
2250 u64 transid = 0;
2251 u64 bytenr;
2252
2253 /* we would like to check all the supers, but that would make
2254 * a btrfs mount succeed after a mkfs from a different FS.
2255 * So, we need to add a special mount option to scan for
2256 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2257 */
2258 for (i = 0; i < 1; i++) {
2259 bytenr = btrfs_sb_offset(i);
2260 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2261 break;
2262 bh = __bread(bdev, bytenr / 4096, 4096);
2263 if (!bh)
2264 continue;
2265
2266 super = (struct btrfs_super_block *)bh->b_data;
2267 if (btrfs_super_bytenr(super) != bytenr ||
2268 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2269 sizeof(super->magic))) {
2270 brelse(bh);
2271 continue;
2272 }
2273
2274 if (!latest || btrfs_super_generation(super) > transid) {
2275 brelse(latest);
2276 latest = bh;
2277 transid = btrfs_super_generation(super);
2278 } else {
2279 brelse(bh);
2280 }
2281 }
2282 return latest;
2283}
2284
4eedeb75
HH		 2285/*
2286 * this should be called twice, once with wait == 0 and
2287 * once with wait == 1. When wait == 0 is done, all the buffer heads
2288 * we write are pinned.
2289 *
2290 * They are released when wait == 1 is done.
2291 * max_mirrors must be the same for both runs, and it indicates how
2292 * many supers on this one device should be written.
2293 *
2294 * max_mirrors == 0 means to write them all.
2295 */
a512bbf8
YZ		 2296static int write_dev_supers(struct btrfs_device *device,
2297 struct btrfs_super_block *sb,
2298 int do_barriers, int wait, int max_mirrors)
2299{
2300 struct buffer_head *bh;
2301 int i;
2302 int ret;
2303 int errors = 0;
2304 u32 crc;
2305 u64 bytenr;
2306 int last_barrier = 0;
2307
2308 if (max_mirrors == 0)
2309 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2310
2311 /* make sure only the last submit_bh does a barrier */
2312 if (do_barriers) {
2313 for (i = 0; i < max_mirrors; i++) {
2314 bytenr = btrfs_sb_offset(i);
2315 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2316 device->total_bytes)
2317 break;
2318 last_barrier = i;
2319 }
2320 }
2321
2322 for (i = 0; i < max_mirrors; i++) {
2323 bytenr = btrfs_sb_offset(i);
2324 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2325 break;
2326
2327 if (wait) {
2328 bh = __find_get_block(device->bdev, bytenr / 4096,
2329 BTRFS_SUPER_INFO_SIZE);
2330 BUG_ON(!bh);
a512bbf8 2331 wait_on_buffer(bh);
4eedeb75
HH		 2332 if (!buffer_uptodate(bh))
2333 errors++;
2334
2335 /* drop our reference */
2336 brelse(bh);
2337
2338 /* drop the reference from the wait == 0 run */
2339 brelse(bh);
2340 continue;
a512bbf8
YZ		 2341 } else {
2342 btrfs_set_super_bytenr(sb, bytenr);
2343
2344 crc = ~(u32)0;
2345 crc = btrfs_csum_data(NULL, (char *)sb +
2346 BTRFS_CSUM_SIZE, crc,
2347 BTRFS_SUPER_INFO_SIZE -
2348 BTRFS_CSUM_SIZE);
2349 btrfs_csum_final(crc, sb->csum);
2350
4eedeb75
HH		 2351 /*
2352 * one reference for us, and we leave it for the
2353 * caller
2354 */
a512bbf8
YZ		 2355 bh = __getblk(device->bdev, bytenr / 4096,
2356 BTRFS_SUPER_INFO_SIZE);
2357 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2358
4eedeb75 2359 /* one reference for submit_bh */
a512bbf8 2360 get_bh(bh);
4eedeb75
HH		 2361
2362 set_buffer_uptodate(bh);
a512bbf8
YZ		 2363 lock_buffer(bh);
2364 bh->b_end_io = btrfs_end_buffer_write_sync;
2365 }
2366
c3b9a62c
CH		 2367 if (i == last_barrier && do_barriers)
2368 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2369 else
ffbd517d 2370 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8 2371
4eedeb75 2372 if (ret)
a512bbf8 2373 errors++;
a512bbf8
YZ		 2374 }
2375 return errors < i ? 0 : -1;
2376}
2377
2378int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2379{
e5e9a520 2380 struct list_head *head;
f2984462 2381 struct btrfs_device *dev;
a061fc8d 2382 struct btrfs_super_block *sb;
f2984462 2383 struct btrfs_dev_item *dev_item;
f2984462
CM		 2384 int ret;
2385 int do_barriers;
a236aed1
CM		 2386 int max_errors;
2387 int total_errors = 0;
a061fc8d 2388 u64 flags;
f2984462 2389
a236aed1 2390 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2391 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2392
a061fc8d
CM		 2393 sb = &root->fs_info->super_for_commit;
2394 dev_item = &sb->dev_item;
e5e9a520
CM		 2395
2396 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2397 head = &root->fs_info->fs_devices->devices;
c6e30871 2398 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2399 if (!dev->bdev) {
2400 total_errors++;
2401 continue;
2402 }
2b82032c 2403 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2404 continue;
2405
2b82032c 2406 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2407 btrfs_set_stack_device_type(dev_item, dev->type);
2408 btrfs_set_stack_device_id(dev_item, dev->devid);
2409 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2410 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2411 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2412 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2413 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2414 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2415 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2416
a061fc8d
CM		 2417 flags = btrfs_super_flags(sb);
2418 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2419
a512bbf8 2420 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2421 if (ret)
2422 total_errors++;
f2984462 2423 }
a236aed1 2424 if (total_errors > max_errors) {
d397712b
CM		 2425 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2426 total_errors);
a236aed1
CM		 2427 BUG();
2428 }
f2984462 2429
a512bbf8 2430 total_errors = 0;
c6e30871 2431 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2432 if (!dev->bdev)
2433 continue;
2b82032c 2434 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2435 continue;
2436
a512bbf8
YZ		 2437 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2438 if (ret)
2439 total_errors++;
f2984462 2440 }
e5e9a520 2441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2442 if (total_errors > max_errors) {
d397712b
CM		 2443 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2444 total_errors);
a236aed1
CM		 2445 BUG();
2446 }
f2984462
CM		 2447 return 0;
2448}
2449
a512bbf8
YZ		 2450int write_ctree_super(struct btrfs_trans_handle *trans,
2451 struct btrfs_root *root, int max_mirrors)
eb60ceac 2452{
e66f709b 2453 int ret;
5f39d397 2454
a512bbf8 2455 ret = write_all_supers(root, max_mirrors);
5f39d397 2456 return ret;
cfaa7295
CM		 2457}
2458
5eda7b5e 2459int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2460{
4df27c4d 2461 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2462 radix_tree_delete(&fs_info->fs_roots_radix,
2463 (unsigned long)root->root_key.objectid);
4df27c4d 2464 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2465
2466 if (btrfs_root_refs(&root->root_item) == 0)
2467 synchronize_srcu(&fs_info->subvol_srcu);
2468
4df27c4d
YZ		 2469 free_fs_root(root);
2470 return 0;
2471}
2472
2473static void free_fs_root(struct btrfs_root *root)
2474{
2475 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2476 if (root->anon_super.s_dev) {
2477 down_write(&root->anon_super.s_umount);
2478 kill_anon_super(&root->anon_super);
2479 }
4df27c4d
YZ		 2480 free_extent_buffer(root->node);
2481 free_extent_buffer(root->commit_root);
d397712b 2482 kfree(root->name);
2619ba1f 2483 kfree(root);
2619ba1f
CM		 2484}
2485
35b7e476 2486static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2487{
2488 int ret;
2489 struct btrfs_root *gang[8];
2490 int i;
2491
76dda93c
YZ		 2492 while (!list_empty(&fs_info->dead_roots)) {
2493 gang[0] = list_entry(fs_info->dead_roots.next,
2494 struct btrfs_root, root_list);
2495 list_del(&gang[0]->root_list);
2496
2497 if (gang[0]->in_radix) {
2498 btrfs_free_fs_root(fs_info, gang[0]);
2499 } else {
2500 free_extent_buffer(gang[0]->node);
2501 free_extent_buffer(gang[0]->commit_root);
2502 kfree(gang[0]);
2503 }
2504 }
2505
d397712b 2506 while (1) {
0f7d52f4
CM		 2507 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2508 (void **)gang, 0,
2509 ARRAY_SIZE(gang));
2510 if (!ret)
2511 break;
2619ba1f 2512 for (i = 0; i < ret; i++)
5eda7b5e 2513 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2514 }
2515 return 0;
2516}
b4100d64 2517
c146afad 2518int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2519{
c146afad
YZ		 2520 u64 root_objectid = 0;
2521 struct btrfs_root *gang[8];
2522 int i;
3768f368 2523 int ret;
e089f05c 2524
c146afad
YZ		 2525 while (1) {
2526 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2527 (void **)gang, root_objectid,
2528 ARRAY_SIZE(gang));
2529 if (!ret)
2530 break;
5d4f98a2
YZ		 2531
2532 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 2533 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 2534 int err;
2535
c146afad 2536 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 2537 err = btrfs_orphan_cleanup(gang[i]);
2538 if (err)
2539 return err;
c146afad
YZ		 2540 }
2541 root_objectid++;
2542 }
2543 return 0;
2544}
a2135011 2545
c146afad
YZ		 2546int btrfs_commit_super(struct btrfs_root *root)
2547{
2548 struct btrfs_trans_handle *trans;
2549 int ret;
a74a4b97 2550
c146afad 2551 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2552 btrfs_run_delayed_iputs(root);
a74a4b97 2553 btrfs_clean_old_snapshots(root);
c146afad 2554 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2555
2556 /* wait until ongoing cleanup work done */
2557 down_write(&root->fs_info->cleanup_work_sem);
2558 up_write(&root->fs_info->cleanup_work_sem);
2559
a22285a6 2560 trans = btrfs_join_transaction(root, 1);
3612b495
TI		 2561 if (IS_ERR(trans))
2562 return PTR_ERR(trans);
54aa1f4d 2563 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2564 BUG_ON(ret);
2565 /* run commit again to drop the original snapshot */
a22285a6 2566 trans = btrfs_join_transaction(root, 1);
3612b495
TI		 2567 if (IS_ERR(trans))
2568 return PTR_ERR(trans);
79154b1b
CM		 2569 btrfs_commit_transaction(trans, root);
2570 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2571 BUG_ON(ret);
d6bfde87 2572
a512bbf8 2573 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2574 return ret;
2575}
2576
2577int close_ctree(struct btrfs_root *root)
2578{
2579 struct btrfs_fs_info *fs_info = root->fs_info;
2580 int ret;
2581
2582 fs_info->closing = 1;
2583 smp_mb();
2584
0af3d00b 2585 btrfs_put_block_group_cache(fs_info);
acce952b 2586
2587 /*
2588 * Here come 2 situations when btrfs is broken to flip readonly:
2589 *
2590 * 1. when btrfs flips readonly somewhere else before
2591 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2592 * and btrfs will skip to write sb directly to keep
2593 * ERROR state on disk.
2594 *
2595 * 2. when btrfs flips readonly just in btrfs_commit_super,
2596 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2597 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2598 * btrfs will cleanup all FS resources first and write sb then.
2599 */
c146afad 2600 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 2601 ret = btrfs_commit_super(root);
2602 if (ret)
2603 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2604 }
2605
2606 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2607 ret = btrfs_error_commit_super(root);
d397712b
CM		 2608 if (ret)
2609 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2610 }
0f7d52f4 2611
8929ecfa
YZ		 2612 kthread_stop(root->fs_info->transaction_kthread);
2613 kthread_stop(root->fs_info->cleaner_kthread);
2614
f25784b3
YZ		 2615 fs_info->closing = 2;
2616 smp_mb();
2617
b0c68f8b 2618 if (fs_info->delalloc_bytes) {
d397712b 2619 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2620 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2621 }
31153d81 2622 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2623 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2624 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2625 }
bcc63abb 2626
5d4f98a2
YZ		 2627 free_extent_buffer(fs_info->extent_root->node);
2628 free_extent_buffer(fs_info->extent_root->commit_root);
2629 free_extent_buffer(fs_info->tree_root->node);
2630 free_extent_buffer(fs_info->tree_root->commit_root);
2631 free_extent_buffer(root->fs_info->chunk_root->node);
2632 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2633 free_extent_buffer(root->fs_info->dev_root->node);
2634 free_extent_buffer(root->fs_info->dev_root->commit_root);
2635 free_extent_buffer(root->fs_info->csum_root->node);
2636 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2637
9078a3e1 2638 btrfs_free_block_groups(root->fs_info);
d10c5f31 2639
c146afad 2640 del_fs_roots(fs_info);
d10c5f31 2641
c146afad 2642 iput(fs_info->btree_inode);
9ad6b7bc 2643
61d92c32 2644 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2645 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2646 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2647 btrfs_stop_workers(&fs_info->workers);
2648 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2649 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2650 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2651 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2652 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2653 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2654
dfe25020 2655 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2656 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2657
04160088 2658 bdi_destroy(&fs_info->bdi);
76dda93c 2659 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2660
0f7d52f4 2661 kfree(fs_info->extent_root);
0f7d52f4 2662 kfree(fs_info->tree_root);
0b86a832
CM		 2663 kfree(fs_info->chunk_root);
2664 kfree(fs_info->dev_root);
d20f7043 2665 kfree(fs_info->csum_root);
83a4d548
LZ		 2666 kfree(fs_info);
2667
eb60ceac
CM		 2668 return 0;
2669}
2670
1259ab75 2671int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2672{
1259ab75 2673 int ret;
810191ff 2674 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2675
2ac55d41
JB		 2676 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2677 NULL);
1259ab75
CM		 2678 if (!ret)
2679 return ret;
2680
2681 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2682 parent_transid);
2683 return !ret;
5f39d397
CM		 2684}
2685
2686int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2687{
810191ff 2688 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2689 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2690 buf);
2691}
6702ed49 2692
5f39d397
CM		 2693void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2694{
810191ff 2695 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2696 u64 transid = btrfs_header_generation(buf);
2697 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2698 int was_dirty;
b4ce94de 2699
b9447ef8 2700 btrfs_assert_tree_locked(buf);
ccd467d6 2701 if (transid != root->fs_info->generation) {
d397712b
CM		 2702 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2703 "found %llu running %llu\n",
db94535d 2704 (unsigned long long)buf->start,
d397712b
CM		 2705 (unsigned long long)transid,
2706 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2707 WARN_ON(1);
2708 }
b9473439
CM		 2709 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2710 buf);
2711 if (!was_dirty) {
2712 spin_lock(&root->fs_info->delalloc_lock);
2713 root->fs_info->dirty_metadata_bytes += buf->len;
2714 spin_unlock(&root->fs_info->delalloc_lock);
2715 }
eb60ceac
CM		 2716}
2717
d3c2fdcf 2718void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2719{
188de649
CM		 2720 /*
2721 * looks as though older kernels can get into trouble with
2722 * this code, they end up stuck in balance_dirty_pages forever
2723 */
d6bfde87 2724 u64 num_dirty;
771ed689 2725 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2726
6933c02e 2727 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2728 return;
2729
585ad2c3
CM		 2730 num_dirty = root->fs_info->dirty_metadata_bytes;
2731
d6bfde87
CM		 2732 if (num_dirty > thresh) {
2733 balance_dirty_pages_ratelimited_nr(
d7fc640e 2734 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2735 }
188de649 2736 return;
35b7e476 2737}
6b80053d 2738
ca7a79ad 2739int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2740{
810191ff 2741 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2742 int ret;
ca7a79ad 2743 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2744 if (ret == 0)
b4ce94de 2745 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2746 return ret;
6b80053d 2747}
0da5468f 2748
4bef0848
CM		 2749int btree_lock_page_hook(struct page *page)
2750{
2751 struct inode *inode = page->mapping->host;
b9473439 2752 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2753 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2754 struct extent_buffer *eb;
2755 unsigned long len;
2756 u64 bytenr = page_offset(page);
2757
2758 if (page->private == EXTENT_PAGE_PRIVATE)
2759 goto out;
2760
2761 len = page->private >> 2;
2762 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2763 if (!eb)
2764 goto out;
2765
2766 btrfs_tree_lock(eb);
4bef0848 2767 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2768
2769 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2770 spin_lock(&root->fs_info->delalloc_lock);
2771 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2772 root->fs_info->dirty_metadata_bytes -= eb->len;
2773 else
2774 WARN_ON(1);
2775 spin_unlock(&root->fs_info->delalloc_lock);
2776 }
2777
4bef0848
CM		 2778 btrfs_tree_unlock(eb);
2779 free_extent_buffer(eb);
2780out:
2781 lock_page(page);
2782 return 0;
2783}
2784
acce952b 2785static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2786 int read_only)
2787{
2788 if (read_only)
2789 return;
2790
2791 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2792 printk(KERN_WARNING "warning: mount fs with errors, "
2793 "running btrfsck is recommended\n");
2794}
2795
2796int btrfs_error_commit_super(struct btrfs_root *root)
2797{
2798 int ret;
2799
2800 mutex_lock(&root->fs_info->cleaner_mutex);
2801 btrfs_run_delayed_iputs(root);
2802 mutex_unlock(&root->fs_info->cleaner_mutex);
2803
2804 down_write(&root->fs_info->cleanup_work_sem);
2805 up_write(&root->fs_info->cleanup_work_sem);
2806
2807 /* cleanup FS via transaction */
2808 btrfs_cleanup_transaction(root);
2809
2810 ret = write_ctree_super(NULL, root, 0);
2811
2812 return ret;
2813}
2814
2815static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2816{
2817 struct btrfs_inode *btrfs_inode;
2818 struct list_head splice;
2819
2820 INIT_LIST_HEAD(&splice);
2821
2822 mutex_lock(&root->fs_info->ordered_operations_mutex);
2823 spin_lock(&root->fs_info->ordered_extent_lock);
2824
2825 list_splice_init(&root->fs_info->ordered_operations, &splice);
2826 while (!list_empty(&splice)) {
2827 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2828 ordered_operations);
2829
2830 list_del_init(&btrfs_inode->ordered_operations);
2831
2832 btrfs_invalidate_inodes(btrfs_inode->root);
2833 }
2834
2835 spin_unlock(&root->fs_info->ordered_extent_lock);
2836 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2837
2838 return 0;
2839}
2840
2841static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2842{
2843 struct list_head splice;
2844 struct btrfs_ordered_extent *ordered;
2845 struct inode *inode;
2846
2847 INIT_LIST_HEAD(&splice);
2848
2849 spin_lock(&root->fs_info->ordered_extent_lock);
2850
2851 list_splice_init(&root->fs_info->ordered_extents, &splice);
2852 while (!list_empty(&splice)) {
2853 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2854 root_extent_list);
2855
2856 list_del_init(&ordered->root_extent_list);
2857 atomic_inc(&ordered->refs);
2858
2859 /* the inode may be getting freed (in sys_unlink path). */
2860 inode = igrab(ordered->inode);
2861
2862 spin_unlock(&root->fs_info->ordered_extent_lock);
2863 if (inode)
2864 iput(inode);
2865
2866 atomic_set(&ordered->refs, 1);
2867 btrfs_put_ordered_extent(ordered);
2868
2869 spin_lock(&root->fs_info->ordered_extent_lock);
2870 }
2871
2872 spin_unlock(&root->fs_info->ordered_extent_lock);
2873
2874 return 0;
2875}
2876
2877static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2878 struct btrfs_root *root)
2879{
2880 struct rb_node *node;
2881 struct btrfs_delayed_ref_root *delayed_refs;
2882 struct btrfs_delayed_ref_node *ref;
2883 int ret = 0;
2884
2885 delayed_refs = &trans->delayed_refs;
2886
2887 spin_lock(&delayed_refs->lock);
2888 if (delayed_refs->num_entries == 0) {
2889 printk(KERN_INFO "delayed_refs has NO entry\n");
2890 return ret;
2891 }
2892
2893 node = rb_first(&delayed_refs->root);
2894 while (node) {
2895 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2896 node = rb_next(node);
2897
2898 ref->in_tree = 0;
2899 rb_erase(&ref->rb_node, &delayed_refs->root);
2900 delayed_refs->num_entries--;
2901
2902 atomic_set(&ref->refs, 1);
2903 if (btrfs_delayed_ref_is_head(ref)) {
2904 struct btrfs_delayed_ref_head *head;
2905
2906 head = btrfs_delayed_node_to_head(ref);
2907 mutex_lock(&head->mutex);
2908 kfree(head->extent_op);
2909 delayed_refs->num_heads--;
2910 if (list_empty(&head->cluster))
2911 delayed_refs->num_heads_ready--;
2912 list_del_init(&head->cluster);
2913 mutex_unlock(&head->mutex);
2914 }
2915
2916 spin_unlock(&delayed_refs->lock);
2917 btrfs_put_delayed_ref(ref);
2918
2919 cond_resched();
2920 spin_lock(&delayed_refs->lock);
2921 }
2922
2923 spin_unlock(&delayed_refs->lock);
2924
2925 return ret;
2926}
2927
2928static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2929{
2930 struct btrfs_pending_snapshot *snapshot;
2931 struct list_head splice;
2932
2933 INIT_LIST_HEAD(&splice);
2934
2935 list_splice_init(&t->pending_snapshots, &splice);
2936
2937 while (!list_empty(&splice)) {
2938 snapshot = list_entry(splice.next,
2939 struct btrfs_pending_snapshot,
2940 list);
2941
2942 list_del_init(&snapshot->list);
2943
2944 kfree(snapshot);
2945 }
2946
2947 return 0;
2948}
2949
2950static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2951{
2952 struct btrfs_inode *btrfs_inode;
2953 struct list_head splice;
2954
2955 INIT_LIST_HEAD(&splice);
2956
2957 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2958
2959 spin_lock(&root->fs_info->delalloc_lock);
2960
2961 while (!list_empty(&splice)) {
2962 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2963 delalloc_inodes);
2964
2965 list_del_init(&btrfs_inode->delalloc_inodes);
2966
2967 btrfs_invalidate_inodes(btrfs_inode->root);
2968 }
2969
2970 spin_unlock(&root->fs_info->delalloc_lock);
2971
2972 return 0;
2973}
2974
2975static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2976 struct extent_io_tree *dirty_pages,
2977 int mark)
2978{
2979 int ret;
2980 struct page *page;
2981 struct inode *btree_inode = root->fs_info->btree_inode;
2982 struct extent_buffer *eb;
2983 u64 start = 0;
2984 u64 end;
2985 u64 offset;
2986 unsigned long index;
2987
2988 while (1) {
2989 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2990 mark);
2991 if (ret)
2992 break;
2993
2994 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2995 while (start <= end) {
2996 index = start >> PAGE_CACHE_SHIFT;
2997 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2998 page = find_get_page(btree_inode->i_mapping, index);
2999 if (!page)
3000 continue;
3001 offset = page_offset(page);
3002
3003 spin_lock(&dirty_pages->buffer_lock);
3004 eb = radix_tree_lookup(
3005 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3006 offset >> PAGE_CACHE_SHIFT);
3007 spin_unlock(&dirty_pages->buffer_lock);
3008 if (eb) {
3009 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3010 &eb->bflags);
3011 atomic_set(&eb->refs, 1);
3012 }
3013 if (PageWriteback(page))
3014 end_page_writeback(page);
3015
3016 lock_page(page);
3017 if (PageDirty(page)) {
3018 clear_page_dirty_for_io(page);
3019 spin_lock_irq(&page->mapping->tree_lock);
3020 radix_tree_tag_clear(&page->mapping->page_tree,
3021 page_index(page),
3022 PAGECACHE_TAG_DIRTY);
3023 spin_unlock_irq(&page->mapping->tree_lock);
3024 }
3025
3026 page->mapping->a_ops->invalidatepage(page, 0);
3027 unlock_page(page);
3028 }
3029 }
3030
3031 return ret;
3032}
3033
3034static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3035 struct extent_io_tree *pinned_extents)
3036{
3037 struct extent_io_tree *unpin;
3038 u64 start;
3039 u64 end;
3040 int ret;
3041
3042 unpin = pinned_extents;
3043 while (1) {
3044 ret = find_first_extent_bit(unpin, 0, &start, &end,
3045 EXTENT_DIRTY);
3046 if (ret)
3047 break;
3048
3049 /* opt_discard */
3050 ret = btrfs_error_discard_extent(root, start, end + 1 - start);
3051
3052 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3053 btrfs_error_unpin_extent_range(root, start, end);
3054 cond_resched();
3055 }
3056
3057 return 0;
3058}
3059
3060static int btrfs_cleanup_transaction(struct btrfs_root *root)
3061{
3062 struct btrfs_transaction *t;
3063 LIST_HEAD(list);
3064
3065 WARN_ON(1);
3066
3067 mutex_lock(&root->fs_info->trans_mutex);
3068 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3069
3070 list_splice_init(&root->fs_info->trans_list, &list);
3071 while (!list_empty(&list)) {
3072 t = list_entry(list.next, struct btrfs_transaction, list);
3073 if (!t)
3074 break;
3075
3076 btrfs_destroy_ordered_operations(root);
3077
3078 btrfs_destroy_ordered_extents(root);
3079
3080 btrfs_destroy_delayed_refs(t, root);
3081
3082 btrfs_block_rsv_release(root,
3083 &root->fs_info->trans_block_rsv,
3084 t->dirty_pages.dirty_bytes);
3085
3086 /* FIXME: cleanup wait for commit */
3087 t->in_commit = 1;
3088 t->blocked = 1;
3089 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3090 wake_up(&root->fs_info->transaction_blocked_wait);
3091
3092 t->blocked = 0;
3093 if (waitqueue_active(&root->fs_info->transaction_wait))
3094 wake_up(&root->fs_info->transaction_wait);
3095 mutex_unlock(&root->fs_info->trans_mutex);
3096
3097 mutex_lock(&root->fs_info->trans_mutex);
3098 t->commit_done = 1;
3099 if (waitqueue_active(&t->commit_wait))
3100 wake_up(&t->commit_wait);
3101 mutex_unlock(&root->fs_info->trans_mutex);
3102
3103 mutex_lock(&root->fs_info->trans_mutex);
3104
3105 btrfs_destroy_pending_snapshots(t);
3106
3107 btrfs_destroy_delalloc_inodes(root);
3108
3109 spin_lock(&root->fs_info->new_trans_lock);
3110 root->fs_info->running_transaction = NULL;
3111 spin_unlock(&root->fs_info->new_trans_lock);
3112
3113 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3114 EXTENT_DIRTY);
3115
3116 btrfs_destroy_pinned_extent(root,
3117 root->fs_info->pinned_extents);
3118
3119 t->use_count = 0;
3120 list_del_init(&t->list);
3121 memset(t, 0, sizeof(*t));
3122 kmem_cache_free(btrfs_transaction_cachep, t);
3123 }
3124
3125 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3126 mutex_unlock(&root->fs_info->trans_mutex);
3127
3128 return 0;
3129}
3130
d1310b2e 3131static struct extent_io_ops btree_extent_io_ops = {
4bef0848 3132 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 3133 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 3134 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 3135 /* note we're sharing with inode.c for the merge bio hook */
3136 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3137};
Linux 6.x block layer and io_uring tree(s)