projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: add READAHEAD extent buffer flag
[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>
7a36ddec 32#include <linux/ratelimit.h>
7e75bf3f 33#include <asm/unaligned.h>
4b4e25f2 34#include "compat.h"
eb60ceac
CM		 35#include "ctree.h"
36#include "disk-io.h"
e089f05c 37#include "transaction.h"
0f7d52f4 38#include "btrfs_inode.h"
0b86a832 39#include "volumes.h"
db94535d 40#include "print-tree.h"
8b712842 41#include "async-thread.h"
925baedd 42#include "locking.h"
e02119d5 43#include "tree-log.h"
fa9c0d79 44#include "free-space-cache.h"
581bb050 45#include "inode-map.h"
eb60ceac 46
d1310b2e 47static struct extent_io_ops btree_extent_io_ops;
8b712842 48static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 49static void free_fs_root(struct btrfs_root *root);
acce952b 50static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51 int read_only);
52static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63static int btrfs_cleanup_transaction(struct btrfs_root *root);
ce9adaa5 64
d352ac68
CM		 65/*
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
69 */
ce9adaa5
CM		 70struct end_io_wq {
71 struct bio *bio;
72 bio_end_io_t *end_io;
73 void *private;
74 struct btrfs_fs_info *info;
75 int error;
22c59948 76 int metadata;
ce9adaa5 77 struct list_head list;
8b712842 78 struct btrfs_work work;
ce9adaa5 79};
0da5468f 80
d352ac68
CM		 81/*
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
85 */
44b8bd7e
CM		 86struct async_submit_bio {
87 struct inode *inode;
88 struct bio *bio;
89 struct list_head list;
4a69a410
CM		 90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 92 int rw;
93 int mirror_num;
c8b97818 94 unsigned long bio_flags;
eaf25d93
CM		 95 /*
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
98 */
99 u64 bio_offset;
8b712842 100 struct btrfs_work work;
44b8bd7e
CM		 101};
102
85d4e461
CM		 103/*
104 * Lockdep class keys for extent_buffer->lock's in this root. For a given
105 * eb, the lockdep key is determined by the btrfs_root it belongs to and
106 * the level the eb occupies in the tree.
107 *
108 * Different roots are used for different purposes and may nest inside each
109 * other and they require separate keysets. As lockdep keys should be
110 * static, assign keysets according to the purpose of the root as indicated
111 * by btrfs_root->objectid. This ensures that all special purpose roots
112 * have separate keysets.
4008c04a 113 *
85d4e461
CM		 114 * Lock-nesting across peer nodes is always done with the immediate parent
115 * node locked thus preventing deadlock. As lockdep doesn't know this, use
116 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 117 *
85d4e461
CM		 118 * The key is set by the readpage_end_io_hook after the buffer has passed
119 * csum validation but before the pages are unlocked. It is also set by
120 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 121 *
85d4e461
CM		 122 * We also add a check to make sure the highest level of the tree is the
123 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
124 * needs update as well.
4008c04a
CM		 125 */
126#ifdef CONFIG_DEBUG_LOCK_ALLOC
127# if BTRFS_MAX_LEVEL != 8
128# error
129# endif
85d4e461
CM		 130
131static struct btrfs_lockdep_keyset {
132 u64 id; /* root objectid */
133 const char *name_stem; /* lock name stem */
134 char names[BTRFS_MAX_LEVEL + 1][20];
135 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
136} btrfs_lockdep_keysets[] = {
137 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
138 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
139 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
140 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
141 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
142 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
143 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
144 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
145 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
146 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
147 { .id = 0, .name_stem = "tree" },
4008c04a 148};
85d4e461
CM		 149
150void __init btrfs_init_lockdep(void)
151{
152 int i, j;
153
154 /* initialize lockdep class names */
155 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
157
158 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159 snprintf(ks->names[j], sizeof(ks->names[j]),
160 "btrfs-%s-%02d", ks->name_stem, j);
161 }
162}
163
164void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165 int level)
166{
167 struct btrfs_lockdep_keyset *ks;
168
169 BUG_ON(level >= ARRAY_SIZE(ks->keys));
170
171 /* find the matching keyset, id 0 is the default entry */
172 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173 if (ks->id == objectid)
174 break;
175
176 lockdep_set_class_and_name(&eb->lock,
177 &ks->keys[level], ks->names[level]);
178}
179
4008c04a
CM		 180#endif
181
d352ac68
CM		 182/*
183 * extents on the btree inode are pretty simple, there's one extent
184 * that covers the entire device
185 */
b2950863 186static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 187 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 188 int create)
7eccb903 189{
5f39d397
CM		 190 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191 struct extent_map *em;
192 int ret;
193
890871be 194 read_lock(&em_tree->lock);
d1310b2e 195 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 196 if (em) {
197 em->bdev =
198 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 199 read_unlock(&em_tree->lock);
5f39d397 200 goto out;
a061fc8d 201 }
890871be 202 read_unlock(&em_tree->lock);
7b13b7b1 203
172ddd60 204 em = alloc_extent_map();
5f39d397
CM		 205 if (!em) {
206 em = ERR_PTR(-ENOMEM);
207 goto out;
208 }
209 em->start = 0;
0afbaf8c 210 em->len = (u64)-1;
c8b97818 211 em->block_len = (u64)-1;
5f39d397 212 em->block_start = 0;
a061fc8d 213 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 214
890871be 215 write_lock(&em_tree->lock);
5f39d397
CM		 216 ret = add_extent_mapping(em_tree, em);
217 if (ret == -EEXIST) {
0afbaf8c
CM		 218 u64 failed_start = em->start;
219 u64 failed_len = em->len;
220
5f39d397 221 free_extent_map(em);
7b13b7b1 222 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 223 if (em) {
7b13b7b1 224 ret = 0;
0afbaf8c
CM		 225 } else {
226 em = lookup_extent_mapping(em_tree, failed_start,
227 failed_len);
7b13b7b1 228 ret = -EIO;
0afbaf8c 229 }
5f39d397 230 } else if (ret) {
7b13b7b1
CM		 231 free_extent_map(em);
232 em = NULL;
5f39d397 233 }
890871be 234 write_unlock(&em_tree->lock);
7b13b7b1
CM		 235
236 if (ret)
237 em = ERR_PTR(ret);
5f39d397
CM		 238out:
239 return em;
7eccb903
CM		 240}
241
19c00ddc
CM		 242u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
243{
163e783e 244 return crc32c(seed, data, len);
19c00ddc
CM		 245}
246
247void btrfs_csum_final(u32 crc, char *result)
248{
7e75bf3f 249 put_unaligned_le32(~crc, result);
19c00ddc
CM		 250}
251
d352ac68
CM		 252/*
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
255 */
19c00ddc
CM		 256static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 int verify)
258{
607d432d
JB		 259 u16 csum_size =
260 btrfs_super_csum_size(&root->fs_info->super_copy);
261 char *result = NULL;
19c00ddc
CM		 262 unsigned long len;
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 265 char *kaddr;
266 unsigned long map_start;
267 unsigned long map_len;
268 int err;
269 u32 crc = ~(u32)0;
607d432d 270 unsigned long inline_result;
19c00ddc
CM		 271
272 len = buf->len - offset;
d397712b 273 while (len > 0) {
19c00ddc 274 err = map_private_extent_buffer(buf, offset, 32,
a6591715 275 &kaddr, &map_start, &map_len);
d397712b 276 if (err)
19c00ddc 277 return 1;
19c00ddc
CM		 278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
280 crc, cur_len);
281 len -= cur_len;
282 offset += cur_len;
19c00ddc 283 }
607d432d
JB		 284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
286 if (!result)
287 return 1;
288 } else {
289 result = (char *)&inline_result;
290 }
291
19c00ddc
CM		 292 btrfs_csum_final(crc, result);
293
294 if (verify) {
607d432d 295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 296 u32 val;
297 u32 found = 0;
607d432d 298 memcpy(&found, result, csum_size);
e4204ded 299
607d432d 300 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 302 "failed on %llu wanted %X found %X "
303 "level %d\n",
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
607d432d
JB		 307 if (result != (char *)&inline_result)
308 kfree(result);
19c00ddc
CM		 309 return 1;
310 }
311 } else {
607d432d 312 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 313 }
607d432d
JB		 314 if (result != (char *)&inline_result)
315 kfree(result);
19c00ddc
CM		 316 return 0;
317}
318
d352ac68
CM		 319/*
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
324 */
1259ab75
CM		 325static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
327{
2ac55d41 328 struct extent_state *cached_state = NULL;
1259ab75
CM		 329 int ret;
330
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332 return 0;
333
2ac55d41
JB		 334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state, GFP_NOFS);
336 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 337 btrfs_header_generation(eb) == parent_transid) {
338 ret = 0;
339 goto out;
340 }
7a36ddec 341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM		 342 "found %llu\n",
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
1259ab75 346 ret = 1;
2ac55d41 347 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 348out:
2ac55d41
JB		 349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
1259ab75 351 return ret;
1259ab75
CM		 352}
353
d352ac68
CM		 354/*
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
357 */
f188591e
CM		 358static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
ca7a79ad 360 u64 start, u64 parent_transid)
f188591e
CM		 361{
362 struct extent_io_tree *io_tree;
363 int ret;
364 int num_copies = 0;
365 int mirror_num = 0;
366
a826d6dc 367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
369 while (1) {
bb82ab88
AJ		 370 ret = read_extent_buffer_pages(io_tree, eb, start,
371 WAIT_COMPLETE,
f188591e 372 btree_get_extent, mirror_num);
1259ab75
CM		 373 if (!ret &&
374 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 375 return ret;
d397712b 376
a826d6dc
JB		 377 /*
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
380 * any less wrong.
381 */
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
383 return ret;
384
f188591e
CM		 385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
386 eb->start, eb->len);
4235298e 387 if (num_copies == 1)
f188591e 388 return ret;
4235298e 389
f188591e 390 mirror_num++;
4235298e 391 if (mirror_num > num_copies)
f188591e 392 return ret;
f188591e 393 }
f188591e
CM		 394 return -EIO;
395}
19c00ddc 396
d352ac68 397/*
d397712b
CM		 398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 400 */
d397712b 401
b2950863 402static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 403{
d1310b2e 404 struct extent_io_tree *tree;
35ebb934 405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc 406 u64 found_start;
19c00ddc
CM		 407 unsigned long len;
408 struct extent_buffer *eb;
f188591e
CM		 409 int ret;
410
d1310b2e 411 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 412
eb14ab8e
CM		 413 if (page->private == EXTENT_PAGE_PRIVATE) {
414 WARN_ON(1);
19c00ddc 415 goto out;
eb14ab8e
CM		 416 }
417 if (!page->private) {
418 WARN_ON(1);
19c00ddc 419 goto out;
eb14ab8e 420 }
19c00ddc 421 len = page->private >> 2;
d397712b
CM		 422 WARN_ON(len == 0);
423
ba144192 424 eb = alloc_extent_buffer(tree, start, len, page);
91ca338d
TI		 425 if (eb == NULL) {
426 WARN_ON(1);
427 goto out;
428 }
ca7a79ad
CM		 429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
f188591e 431 BUG_ON(ret);
784b4e29
CM		 432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433
19c00ddc
CM		 434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
55c69072
CM		 436 WARN_ON(1);
437 goto err;
438 }
439 if (eb->first_page != page) {
55c69072
CM		 440 WARN_ON(1);
441 goto err;
442 }
443 if (!PageUptodate(page)) {
55c69072
CM		 444 WARN_ON(1);
445 goto err;
19c00ddc 446 }
19c00ddc 447 csum_tree_block(root, eb, 0);
55c69072 448err:
19c00ddc
CM		 449 free_extent_buffer(eb);
450out:
451 return 0;
452}
453
2b82032c
YZ		 454static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
456{
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
459 int ret = 1;
460
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462 BTRFS_FSID_SIZE);
463 while (fs_devices) {
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465 ret = 0;
466 break;
467 }
468 fs_devices = fs_devices->seed;
469 }
470 return ret;
471}
472
a826d6dc
JB		 473#define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
478
479static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
481{
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
485 int slot;
486
487 if (nritems == 0)
488 return 0;
489
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
494 return -EIO;
495 }
496
497 /*
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
502 * offset is correct.
503 */
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
511 return -EIO;
512 }
513
514 /*
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
517 * front.
518 */
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
522 return -EIO;
523 }
524
525 /*
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
529 */
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
533 return -EIO;
534 }
535 }
536
537 return 0;
538}
539
b2950863 540static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 541 struct extent_state *state)
542{
543 struct extent_io_tree *tree;
544 u64 found_start;
545 int found_level;
546 unsigned long len;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 549 int ret = 0;
ce9adaa5
CM		 550
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
553 goto out;
554 if (!page->private)
555 goto out;
d397712b 556
ce9adaa5 557 len = page->private >> 2;
d397712b
CM		 558 WARN_ON(len == 0);
559
ba144192 560 eb = alloc_extent_buffer(tree, start, len, page);
91ca338d
TI		 561 if (eb == NULL) {
562 ret = -EIO;
563 goto out;
564 }
f188591e 565
ce9adaa5 566 found_start = btrfs_header_bytenr(eb);
23a07867 567 if (found_start != start) {
7a36ddec 568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 569 "%llu %llu\n",
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
f188591e 572 ret = -EIO;
ce9adaa5
CM		 573 goto err;
574 }
575 if (eb->first_page != page) {
d397712b
CM		 576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
ce9adaa5 578 WARN_ON(1);
f188591e 579 ret = -EIO;
ce9adaa5
CM		 580 goto err;
581 }
2b82032c 582 if (check_tree_block_fsid(root, eb)) {
7a36ddec 583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 584 (unsigned long long)eb->start);
1259ab75
CM		 585 ret = -EIO;
586 goto err;
587 }
ce9adaa5
CM		 588 found_level = btrfs_header_level(eb);
589
85d4e461
CM		 590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591 eb, found_level);
4008c04a 592
ce9adaa5 593 ret = csum_tree_block(root, eb, 1);
a826d6dc 594 if (ret) {
f188591e 595 ret = -EIO;
a826d6dc
JB		 596 goto err;
597 }
598
599 /*
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
602 * return -EIO.
603 */
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606 ret = -EIO;
607 }
ce9adaa5
CM		 608
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
ce9adaa5
CM		 611err:
612 free_extent_buffer(eb);
613out:
f188591e 614 return ret;
ce9adaa5
CM		 615}
616
ce9adaa5 617static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 618{
619 struct end_io_wq *end_io_wq = bio->bi_private;
620 struct btrfs_fs_info *fs_info;
ce9adaa5 621
ce9adaa5 622 fs_info = end_io_wq->info;
ce9adaa5 623 end_io_wq->error = err;
8b712842
CM		 624 end_io_wq->work.func = end_workqueue_fn;
625 end_io_wq->work.flags = 0;
d20f7043 626
7b6d91da 627 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 628 if (end_io_wq->metadata == 1)
cad321ad
CM		 629 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
630 &end_io_wq->work);
0cb59c99
JB		 631 else if (end_io_wq->metadata == 2)
632 btrfs_queue_worker(&fs_info->endio_freespace_worker,
633 &end_io_wq->work);
cad321ad
CM		 634 else
635 btrfs_queue_worker(&fs_info->endio_write_workers,
636 &end_io_wq->work);
d20f7043
CM		 637 } else {
638 if (end_io_wq->metadata)
639 btrfs_queue_worker(&fs_info->endio_meta_workers,
640 &end_io_wq->work);
641 else
642 btrfs_queue_worker(&fs_info->endio_workers,
643 &end_io_wq->work);
644 }
ce9adaa5
CM		 645}
646
0cb59c99
JB		 647/*
648 * For the metadata arg you want
649 *
650 * 0 - if data
651 * 1 - if normal metadta
652 * 2 - if writing to the free space cache area
653 */
22c59948
CM		 654int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
655 int metadata)
0b86a832 656{
ce9adaa5 657 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 658 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
659 if (!end_io_wq)
660 return -ENOMEM;
661
662 end_io_wq->private = bio->bi_private;
663 end_io_wq->end_io = bio->bi_end_io;
22c59948 664 end_io_wq->info = info;
ce9adaa5
CM		 665 end_io_wq->error = 0;
666 end_io_wq->bio = bio;
22c59948 667 end_io_wq->metadata = metadata;
ce9adaa5
CM		 668
669 bio->bi_private = end_io_wq;
670 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 671 return 0;
672}
673
b64a2851 674unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 675{
4854ddd0
CM		 676 unsigned long limit = min_t(unsigned long,
677 info->workers.max_workers,
678 info->fs_devices->open_devices);
679 return 256 * limit;
680}
0986fe9e 681
4a69a410
CM		 682static void run_one_async_start(struct btrfs_work *work)
683{
4a69a410
CM		 684 struct async_submit_bio *async;
685
686 async = container_of(work, struct async_submit_bio, work);
4a69a410 687 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM		 688 async->mirror_num, async->bio_flags,
689 async->bio_offset);
4a69a410
CM		 690}
691
692static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 693{
694 struct btrfs_fs_info *fs_info;
695 struct async_submit_bio *async;
4854ddd0 696 int limit;
8b712842
CM		 697
698 async = container_of(work, struct async_submit_bio, work);
699 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 700
b64a2851 701 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 702 limit = limit * 2 / 3;
703
8b712842 704 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 705
b64a2851
CM		 706 if (atomic_read(&fs_info->nr_async_submits) < limit &&
707 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 708 wake_up(&fs_info->async_submit_wait);
709
4a69a410 710 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 711 async->mirror_num, async->bio_flags,
712 async->bio_offset);
4a69a410
CM		 713}
714
715static void run_one_async_free(struct btrfs_work *work)
716{
717 struct async_submit_bio *async;
718
719 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 720 kfree(async);
721}
722
44b8bd7e
CM		 723int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
724 int rw, struct bio *bio, int mirror_num,
c8b97818 725 unsigned long bio_flags,
eaf25d93 726 u64 bio_offset,
4a69a410
CM		 727 extent_submit_bio_hook_t *submit_bio_start,
728 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 729{
730 struct async_submit_bio *async;
731
732 async = kmalloc(sizeof(*async), GFP_NOFS);
733 if (!async)
734 return -ENOMEM;
735
736 async->inode = inode;
737 async->rw = rw;
738 async->bio = bio;
739 async->mirror_num = mirror_num;
4a69a410
CM		 740 async->submit_bio_start = submit_bio_start;
741 async->submit_bio_done = submit_bio_done;
742
743 async->work.func = run_one_async_start;
744 async->work.ordered_func = run_one_async_done;
745 async->work.ordered_free = run_one_async_free;
746
8b712842 747 async->work.flags = 0;
c8b97818 748 async->bio_flags = bio_flags;
eaf25d93 749 async->bio_offset = bio_offset;
8c8bee1d 750
cb03c743 751 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 752
7b6d91da 753 if (rw & REQ_SYNC)
d313d7a3
CM		 754 btrfs_set_work_high_prio(&async->work);
755
8b712842 756 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 757
d397712b 758 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 759 atomic_read(&fs_info->nr_async_submits)) {
760 wait_event(fs_info->async_submit_wait,
761 (atomic_read(&fs_info->nr_async_submits) == 0));
762 }
763
44b8bd7e
CM		 764 return 0;
765}
766
ce3ed71a
CM		 767static int btree_csum_one_bio(struct bio *bio)
768{
769 struct bio_vec *bvec = bio->bi_io_vec;
770 int bio_index = 0;
771 struct btrfs_root *root;
772
773 WARN_ON(bio->bi_vcnt <= 0);
d397712b 774 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 775 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
776 csum_dirty_buffer(root, bvec->bv_page);
777 bio_index++;
778 bvec++;
779 }
780 return 0;
781}
782
4a69a410
CM		 783static int __btree_submit_bio_start(struct inode *inode, int rw,
784 struct bio *bio, int mirror_num,
eaf25d93
CM		 785 unsigned long bio_flags,
786 u64 bio_offset)
22c59948 787{
8b712842
CM		 788 /*
789 * when we're called for a write, we're already in the async
5443be45 790 * submission context. Just jump into btrfs_map_bio
8b712842 791 */
4a69a410
CM		 792 btree_csum_one_bio(bio);
793 return 0;
794}
22c59948 795
4a69a410 796static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 797 int mirror_num, unsigned long bio_flags,
798 u64 bio_offset)
4a69a410 799{
8b712842 800 /*
4a69a410
CM		 801 * when we're called for a write, we're already in the async
802 * submission context. Just jump into btrfs_map_bio
8b712842 803 */
8b712842 804 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 805}
806
44b8bd7e 807static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 808 int mirror_num, unsigned long bio_flags,
809 u64 bio_offset)
44b8bd7e 810{
cad321ad
CM		 811 int ret;
812
813 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
814 bio, 1);
815 BUG_ON(ret);
816
7b6d91da 817 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 818 /*
819 * called for a read, do the setup so that checksum validation
820 * can happen in the async kernel threads
821 */
4a69a410 822 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 823 mirror_num, 0);
44b8bd7e 824 }
d313d7a3 825
cad321ad
CM		 826 /*
827 * kthread helpers are used to submit writes so that checksumming
828 * can happen in parallel across all CPUs
829 */
44b8bd7e 830 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 831 inode, rw, bio, mirror_num, 0,
eaf25d93 832 bio_offset,
4a69a410
CM		 833 __btree_submit_bio_start,
834 __btree_submit_bio_done);
44b8bd7e
CM		 835}
836
3dd1462e 837#ifdef CONFIG_MIGRATION
784b4e29
CM		 838static int btree_migratepage(struct address_space *mapping,
839 struct page *newpage, struct page *page)
840{
841 /*
842 * we can't safely write a btree page from here,
843 * we haven't done the locking hook
844 */
845 if (PageDirty(page))
846 return -EAGAIN;
847 /*
848 * Buffers may be managed in a filesystem specific way.
849 * We must have no buffers or drop them.
850 */
851 if (page_has_private(page) &&
852 !try_to_release_page(page, GFP_KERNEL))
853 return -EAGAIN;
784b4e29
CM		 854 return migrate_page(mapping, newpage, page);
855}
3dd1462e 856#endif
784b4e29 857
0da5468f
CM		 858static int btree_writepage(struct page *page, struct writeback_control *wbc)
859{
d1310b2e 860 struct extent_io_tree *tree;
b9473439
CM		 861 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
862 struct extent_buffer *eb;
863 int was_dirty;
864
d1310b2e 865 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 866 if (!(current->flags & PF_MEMALLOC)) {
867 return extent_write_full_page(tree, page,
868 btree_get_extent, wbc);
869 }
5443be45 870
b9473439 871 redirty_page_for_writepage(wbc, page);
784b4e29 872 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
b9473439
CM		 873 WARN_ON(!eb);
874
875 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
876 if (!was_dirty) {
877 spin_lock(&root->fs_info->delalloc_lock);
878 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
879 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 880 }
b9473439
CM		 881 free_extent_buffer(eb);
882
883 unlock_page(page);
884 return 0;
5f39d397 885}
0da5468f
CM		 886
887static int btree_writepages(struct address_space *mapping,
888 struct writeback_control *wbc)
889{
d1310b2e
CM		 890 struct extent_io_tree *tree;
891 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 892 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 893 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 894 u64 num_dirty;
24ab9cd8 895 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 896
897 if (wbc->for_kupdate)
898 return 0;
899
b9473439
CM		 900 /* this is a bit racy, but that's ok */
901 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 902 if (num_dirty < thresh)
793955bc 903 return 0;
793955bc 904 }
0da5468f
CM		 905 return extent_writepages(tree, mapping, btree_get_extent, wbc);
906}
907
b2950863 908static int btree_readpage(struct file *file, struct page *page)
5f39d397 909{
d1310b2e
CM		 910 struct extent_io_tree *tree;
911 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 912 return extent_read_full_page(tree, page, btree_get_extent);
913}
22b0ebda 914
70dec807 915static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 916{
d1310b2e
CM		 917 struct extent_io_tree *tree;
918 struct extent_map_tree *map;
5f39d397 919 int ret;
d98237b3 920
98509cfc 921 if (PageWriteback(page) || PageDirty(page))
d397712b 922 return 0;
98509cfc 923
d1310b2e
CM		 924 tree = &BTRFS_I(page->mapping->host)->io_tree;
925 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 926
7b13b7b1 927 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 928 if (!ret)
6af118ce 929 return 0;
6af118ce
CM		 930
931 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 932 if (ret == 1) {
933 ClearPagePrivate(page);
934 set_page_private(page, 0);
935 page_cache_release(page);
936 }
6af118ce 937
d98237b3
CM		 938 return ret;
939}
940
5f39d397 941static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 942{
d1310b2e
CM		 943 struct extent_io_tree *tree;
944 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 945 extent_invalidatepage(tree, page, offset);
946 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 947 if (PagePrivate(page)) {
d397712b
CM		 948 printk(KERN_WARNING "btrfs warning page private not zero "
949 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 950 ClearPagePrivate(page);
951 set_page_private(page, 0);
952 page_cache_release(page);
953 }
d98237b3
CM		 954}
955
7f09410b 956static const struct address_space_operations btree_aops = {
d98237b3
CM		 957 .readpage = btree_readpage,
958 .writepage = btree_writepage,
0da5468f 959 .writepages = btree_writepages,
5f39d397
CM		 960 .releasepage = btree_releasepage,
961 .invalidatepage = btree_invalidatepage,
5a92bc88 962#ifdef CONFIG_MIGRATION
784b4e29 963 .migratepage = btree_migratepage,
5a92bc88 964#endif
d98237b3
CM		 965};
966
ca7a79ad
CM		 967int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
968 u64 parent_transid)
090d1875 969{
5f39d397
CM		 970 struct extent_buffer *buf = NULL;
971 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 972 int ret = 0;
090d1875 973
db94535d 974 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 975 if (!buf)
090d1875 976 return 0;
d1310b2e 977 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 978 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 979 free_extent_buffer(buf);
de428b63 980 return ret;
090d1875
CM		 981}
982
ab0fff03
AJ		 983int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
984 int mirror_num, struct extent_buffer **eb)
985{
986 struct extent_buffer *buf = NULL;
987 struct inode *btree_inode = root->fs_info->btree_inode;
988 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
989 int ret;
990
991 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
992 if (!buf)
993 return 0;
994
995 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
996
997 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
998 btree_get_extent, mirror_num);
999 if (ret) {
1000 free_extent_buffer(buf);
1001 return ret;
1002 }
1003
1004 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1005 free_extent_buffer(buf);
1006 return -EIO;
1007 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1008 *eb = buf;
1009 } else {
1010 free_extent_buffer(buf);
1011 }
1012 return 0;
1013}
1014
0999df54
CM		 1015struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1016 u64 bytenr, u32 blocksize)
1017{
1018 struct inode *btree_inode = root->fs_info->btree_inode;
1019 struct extent_buffer *eb;
1020 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1021 bytenr, blocksize);
0999df54
CM		 1022 return eb;
1023}
1024
1025struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1026 u64 bytenr, u32 blocksize)
1027{
1028 struct inode *btree_inode = root->fs_info->btree_inode;
1029 struct extent_buffer *eb;
1030
1031 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
ba144192 1032 bytenr, blocksize, NULL);
0999df54
CM		 1033 return eb;
1034}
1035
1036
e02119d5
CM		 1037int btrfs_write_tree_block(struct extent_buffer *buf)
1038{
8aa38c31
CH		 1039 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1040 buf->start + buf->len - 1);
e02119d5
CM		 1041}
1042
1043int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1044{
8aa38c31
CH		 1045 return filemap_fdatawait_range(buf->first_page->mapping,
1046 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1047}
1048
0999df54 1049struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1050 u32 blocksize, u64 parent_transid)
0999df54
CM		 1051{
1052 struct extent_buffer *buf = NULL;
0999df54
CM		 1053 int ret;
1054
0999df54
CM		 1055 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1056 if (!buf)
1057 return NULL;
0999df54 1058
ca7a79ad 1059 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 1060
d397712b 1061 if (ret == 0)
b4ce94de 1062 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 1063 return buf;
ce9adaa5 1064
eb60ceac
CM		 1065}
1066
e089f05c 1067int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 1068 struct extent_buffer *buf)
ed2ff2cb 1069{
5f39d397 1070 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 1071 if (btrfs_header_generation(buf) ==
925baedd 1072 root->fs_info->running_transaction->transid) {
b9447ef8 1073 btrfs_assert_tree_locked(buf);
b4ce94de 1074
b9473439
CM		 1075 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1076 spin_lock(&root->fs_info->delalloc_lock);
1077 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1078 root->fs_info->dirty_metadata_bytes -= buf->len;
1079 else
1080 WARN_ON(1);
1081 spin_unlock(&root->fs_info->delalloc_lock);
1082 }
b4ce94de 1083
b9473439
CM		 1084 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1085 btrfs_set_lock_blocking(buf);
d1310b2e 1086 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 1087 buf);
925baedd 1088 }
5f39d397
CM		 1089 return 0;
1090}
1091
db94535d 1092static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 1093 u32 stripesize, struct btrfs_root *root,
9f5fae2f 1094 struct btrfs_fs_info *fs_info,
e20d96d6 1095 u64 objectid)
d97e63b6 1096{
cfaa7295 1097 root->node = NULL;
a28ec197 1098 root->commit_root = NULL;
db94535d
CM		 1099 root->sectorsize = sectorsize;
1100 root->nodesize = nodesize;
1101 root->leafsize = leafsize;
87ee04eb 1102 root->stripesize = stripesize;
123abc88 1103 root->ref_cows = 0;
0b86a832 1104 root->track_dirty = 0;
c71bf099 1105 root->in_radix = 0;
d68fc57b
YZ		 1106 root->orphan_item_inserted = 0;
1107 root->orphan_cleanup_state = 0;
0b86a832 1108
9f5fae2f 1109 root->fs_info = fs_info;
0f7d52f4
CM		 1110 root->objectid = objectid;
1111 root->last_trans = 0;
13a8a7c8 1112 root->highest_objectid = 0;
58176a96 1113 root->name = NULL;
6bef4d31 1114 root->inode_tree = RB_ROOT;
16cdcec7 1115 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1116 root->block_rsv = NULL;
d68fc57b 1117 root->orphan_block_rsv = NULL;
0b86a832
CM		 1118
1119 INIT_LIST_HEAD(&root->dirty_list);
7b128766 1120 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 1121 INIT_LIST_HEAD(&root->root_list);
d68fc57b 1122 spin_lock_init(&root->orphan_lock);
5d4f98a2 1123 spin_lock_init(&root->inode_lock);
f0486c68 1124 spin_lock_init(&root->accounting_lock);
a2135011 1125 mutex_init(&root->objectid_mutex);
e02119d5 1126 mutex_init(&root->log_mutex);
7237f183
YZ		 1127 init_waitqueue_head(&root->log_writer_wait);
1128 init_waitqueue_head(&root->log_commit_wait[0]);
1129 init_waitqueue_head(&root->log_commit_wait[1]);
1130 atomic_set(&root->log_commit[0], 0);
1131 atomic_set(&root->log_commit[1], 0);
1132 atomic_set(&root->log_writers, 0);
1133 root->log_batch = 0;
1134 root->log_transid = 0;
257c62e1 1135 root->last_log_commit = 0;
d0c803c4 1136 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1137 fs_info->btree_inode->i_mapping);
017e5369 1138
3768f368
CM		 1139 memset(&root->root_key, 0, sizeof(root->root_key));
1140 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1141 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1142 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1143 root->defrag_trans_start = fs_info->generation;
58176a96 1144 init_completion(&root->kobj_unregister);
6702ed49 1145 root->defrag_running = 0;
4d775673 1146 root->root_key.objectid = objectid;
0ee5dc67 1147 root->anon_dev = 0;
3768f368
CM		 1148 return 0;
1149}
1150
db94535d 1151static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 1152 struct btrfs_fs_info *fs_info,
1153 u64 objectid,
e20d96d6 1154 struct btrfs_root *root)
3768f368
CM		 1155{
1156 int ret;
db94535d 1157 u32 blocksize;
84234f3a 1158 u64 generation;
3768f368 1159
db94535d 1160 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1161 tree_root->sectorsize, tree_root->stripesize,
1162 root, fs_info, objectid);
3768f368
CM		 1163 ret = btrfs_find_last_root(tree_root, objectid,
1164 &root->root_item, &root->root_key);
4df27c4d
YZ		 1165 if (ret > 0)
1166 return -ENOENT;
3768f368
CM		 1167 BUG_ON(ret);
1168
84234f3a 1169 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1170 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1171 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1172 blocksize, generation);
68433b73
CM		 1173 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1174 free_extent_buffer(root->node);
1175 return -EIO;
1176 }
4df27c4d 1177 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1178 return 0;
1179}
1180
7237f183
YZ		 1181static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1183{
1184 struct btrfs_root *root;
1185 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1186 struct extent_buffer *leaf;
e02119d5
CM		 1187
1188 root = kzalloc(sizeof(*root), GFP_NOFS);
1189 if (!root)
7237f183 1190 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1191
1192 __setup_root(tree_root->nodesize, tree_root->leafsize,
1193 tree_root->sectorsize, tree_root->stripesize,
1194 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1195
1196 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1197 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1198 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1199 /*
1200 * log trees do not get reference counted because they go away
1201 * before a real commit is actually done. They do store pointers
1202 * to file data extents, and those reference counts still get
1203 * updated (along with back refs to the log tree).
1204 */
e02119d5
CM		 1205 root->ref_cows = 0;
1206
5d4f98a2
YZ		 1207 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1208 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1209 if (IS_ERR(leaf)) {
1210 kfree(root);
1211 return ERR_CAST(leaf);
1212 }
e02119d5 1213
5d4f98a2
YZ		 1214 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1215 btrfs_set_header_bytenr(leaf, leaf->start);
1216 btrfs_set_header_generation(leaf, trans->transid);
1217 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1218 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1219 root->node = leaf;
e02119d5
CM		 1220
1221 write_extent_buffer(root->node, root->fs_info->fsid,
1222 (unsigned long)btrfs_header_fsid(root->node),
1223 BTRFS_FSID_SIZE);
1224 btrfs_mark_buffer_dirty(root->node);
1225 btrfs_tree_unlock(root->node);
7237f183
YZ		 1226 return root;
1227}
1228
1229int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1230 struct btrfs_fs_info *fs_info)
1231{
1232 struct btrfs_root *log_root;
1233
1234 log_root = alloc_log_tree(trans, fs_info);
1235 if (IS_ERR(log_root))
1236 return PTR_ERR(log_root);
1237 WARN_ON(fs_info->log_root_tree);
1238 fs_info->log_root_tree = log_root;
1239 return 0;
1240}
1241
1242int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1243 struct btrfs_root *root)
1244{
1245 struct btrfs_root *log_root;
1246 struct btrfs_inode_item *inode_item;
1247
1248 log_root = alloc_log_tree(trans, root->fs_info);
1249 if (IS_ERR(log_root))
1250 return PTR_ERR(log_root);
1251
1252 log_root->last_trans = trans->transid;
1253 log_root->root_key.offset = root->root_key.objectid;
1254
1255 inode_item = &log_root->root_item.inode;
1256 inode_item->generation = cpu_to_le64(1);
1257 inode_item->size = cpu_to_le64(3);
1258 inode_item->nlink = cpu_to_le32(1);
1259 inode_item->nbytes = cpu_to_le64(root->leafsize);
1260 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1261
5d4f98a2 1262 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1263
1264 WARN_ON(root->log_root);
1265 root->log_root = log_root;
1266 root->log_transid = 0;
257c62e1 1267 root->last_log_commit = 0;
e02119d5
CM		 1268 return 0;
1269}
1270
1271struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1272 struct btrfs_key *location)
1273{
1274 struct btrfs_root *root;
1275 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1276 struct btrfs_path *path;
5f39d397 1277 struct extent_buffer *l;
84234f3a 1278 u64 generation;
db94535d 1279 u32 blocksize;
0f7d52f4
CM		 1280 int ret = 0;
1281
5eda7b5e 1282 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1283 if (!root)
0f7d52f4 1284 return ERR_PTR(-ENOMEM);
0f7d52f4 1285 if (location->offset == (u64)-1) {
db94535d 1286 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1287 location->objectid, root);
1288 if (ret) {
0f7d52f4
CM		 1289 kfree(root);
1290 return ERR_PTR(ret);
1291 }
13a8a7c8 1292 goto out;
0f7d52f4
CM		 1293 }
1294
db94535d 1295 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1296 tree_root->sectorsize, tree_root->stripesize,
1297 root, fs_info, location->objectid);
0f7d52f4
CM		 1298
1299 path = btrfs_alloc_path();
db5b493a
TI		 1300 if (!path) {
1301 kfree(root);
1302 return ERR_PTR(-ENOMEM);
1303 }
0f7d52f4 1304 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1305 if (ret == 0) {
1306 l = path->nodes[0];
1307 read_extent_buffer(l, &root->root_item,
1308 btrfs_item_ptr_offset(l, path->slots[0]),
1309 sizeof(root->root_item));
1310 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1311 }
0f7d52f4
CM		 1312 btrfs_free_path(path);
1313 if (ret) {
5e540f77 1314 kfree(root);
13a8a7c8
YZ		 1315 if (ret > 0)
1316 ret = -ENOENT;
0f7d52f4
CM		 1317 return ERR_PTR(ret);
1318 }
13a8a7c8 1319
84234f3a 1320 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1321 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1322 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1323 blocksize, generation);
5d4f98a2 1324 root->commit_root = btrfs_root_node(root);
0f7d52f4 1325 BUG_ON(!root->node);
13a8a7c8 1326out:
08fe4db1 1327 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1328 root->ref_cows = 1;
08fe4db1
LZ		 1329 btrfs_check_and_init_root_item(&root->root_item);
1330 }
13a8a7c8 1331
5eda7b5e
CM		 1332 return root;
1333}
1334
edbd8d4e
CM		 1335struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1336 struct btrfs_key *location)
5eda7b5e
CM		 1337{
1338 struct btrfs_root *root;
1339 int ret;
1340
edbd8d4e
CM		 1341 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1342 return fs_info->tree_root;
1343 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1344 return fs_info->extent_root;
8f18cf13
CM		 1345 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1346 return fs_info->chunk_root;
1347 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1348 return fs_info->dev_root;
0403e47e
YZ		 1349 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1350 return fs_info->csum_root;
4df27c4d
YZ		 1351again:
1352 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1353 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1354 (unsigned long)location->objectid);
4df27c4d 1355 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1356 if (root)
1357 return root;
1358
e02119d5 1359 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1360 if (IS_ERR(root))
1361 return root;
3394e160 1362
581bb050 1363 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
581bb050
LZ		 1364 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1365 GFP_NOFS);
35a30d7c
DS		 1366 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1367 ret = -ENOMEM;
581bb050 1368 goto fail;
35a30d7c 1369 }
581bb050
LZ		 1370
1371 btrfs_init_free_ino_ctl(root);
1372 mutex_init(&root->fs_commit_mutex);
1373 spin_lock_init(&root->cache_lock);
1374 init_waitqueue_head(&root->cache_wait);
1375
0ee5dc67 1376 ret = get_anon_bdev(&root->anon_dev);
ac08aedf
CM		 1377 if (ret)
1378 goto fail;
3394e160 1379
d68fc57b
YZ		 1380 if (btrfs_root_refs(&root->root_item) == 0) {
1381 ret = -ENOENT;
1382 goto fail;
1383 }
1384
1385 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1386 if (ret < 0)
1387 goto fail;
1388 if (ret == 0)
1389 root->orphan_item_inserted = 1;
1390
4df27c4d
YZ		 1391 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1392 if (ret)
1393 goto fail;
1394
1395 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1396 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1397 (unsigned long)root->root_key.objectid,
0f7d52f4 1398 root);
d68fc57b 1399 if (ret == 0)
4df27c4d 1400 root->in_radix = 1;
d68fc57b 1401
4df27c4d
YZ		 1402 spin_unlock(&fs_info->fs_roots_radix_lock);
1403 radix_tree_preload_end();
0f7d52f4 1404 if (ret) {
4df27c4d
YZ		 1405 if (ret == -EEXIST) {
1406 free_fs_root(root);
1407 goto again;
1408 }
1409 goto fail;
0f7d52f4 1410 }
4df27c4d
YZ		 1411
1412 ret = btrfs_find_dead_roots(fs_info->tree_root,
1413 root->root_key.objectid);
1414 WARN_ON(ret);
edbd8d4e 1415 return root;
4df27c4d
YZ		 1416fail:
1417 free_fs_root(root);
1418 return ERR_PTR(ret);
edbd8d4e
CM		 1419}
1420
04160088
CM		 1421static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1422{
1423 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1424 int ret = 0;
04160088
CM		 1425 struct btrfs_device *device;
1426 struct backing_dev_info *bdi;
b7967db7 1427
1f78160c
XG		 1428 rcu_read_lock();
1429 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1430 if (!device->bdev)
1431 continue;
04160088
CM		 1432 bdi = blk_get_backing_dev_info(device->bdev);
1433 if (bdi && bdi_congested(bdi, bdi_bits)) {
1434 ret = 1;
1435 break;
1436 }
1437 }
1f78160c 1438 rcu_read_unlock();
04160088
CM		 1439 return ret;
1440}
1441
ad081f14
JA		 1442/*
1443 * If this fails, caller must call bdi_destroy() to get rid of the
1444 * bdi again.
1445 */
04160088
CM		 1446static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1447{
ad081f14
JA		 1448 int err;
1449
1450 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1451 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1452 if (err)
1453 return err;
1454
4575c9cc 1455 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1456 bdi->congested_fn = btrfs_congested_fn;
1457 bdi->congested_data = info;
1458 return 0;
1459}
1460
ce9adaa5
CM		 1461static int bio_ready_for_csum(struct bio *bio)
1462{
1463 u64 length = 0;
1464 u64 buf_len = 0;
1465 u64 start = 0;
1466 struct page *page;
1467 struct extent_io_tree *io_tree = NULL;
ce9adaa5
CM		 1468 struct bio_vec *bvec;
1469 int i;
1470 int ret;
1471
1472 bio_for_each_segment(bvec, bio, i) {
1473 page = bvec->bv_page;
1474 if (page->private == EXTENT_PAGE_PRIVATE) {
1475 length += bvec->bv_len;
1476 continue;
1477 }
1478 if (!page->private) {
1479 length += bvec->bv_len;
1480 continue;
1481 }
1482 length = bvec->bv_len;
1483 buf_len = page->private >> 2;
1484 start = page_offset(page) + bvec->bv_offset;
1485 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
ce9adaa5
CM		 1486 }
1487 /* are we fully contained in this bio? */
1488 if (buf_len <= length)
1489 return 1;
1490
1491 ret = extent_range_uptodate(io_tree, start + length,
1492 start + buf_len - 1);
ce9adaa5
CM		 1493 return ret;
1494}
1495
8b712842
CM		 1496/*
1497 * called by the kthread helper functions to finally call the bio end_io
1498 * functions. This is where read checksum verification actually happens
1499 */
1500static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1501{
ce9adaa5 1502 struct bio *bio;
8b712842
CM		 1503 struct end_io_wq *end_io_wq;
1504 struct btrfs_fs_info *fs_info;
ce9adaa5 1505 int error;
ce9adaa5 1506
8b712842
CM		 1507 end_io_wq = container_of(work, struct end_io_wq, work);
1508 bio = end_io_wq->bio;
1509 fs_info = end_io_wq->info;
ce9adaa5 1510
cad321ad 1511 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1512 * be checksummed at once. This makes sure the entire block is in
1513 * ram and up to date before trying to verify things. For
1514 * blocksize <= pagesize, it is basically a noop
1515 */
7b6d91da 1516 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1517 !bio_ready_for_csum(bio)) {
d20f7043 1518 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1519 &end_io_wq->work);
1520 return;
1521 }
1522 error = end_io_wq->error;
1523 bio->bi_private = end_io_wq->private;
1524 bio->bi_end_io = end_io_wq->end_io;
1525 kfree(end_io_wq);
8b712842 1526 bio_endio(bio, error);
44b8bd7e
CM		 1527}
1528
a74a4b97
CM		 1529static int cleaner_kthread(void *arg)
1530{
1531 struct btrfs_root *root = arg;
1532
1533 do {
a74a4b97 1534 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1535
1536 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1537 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1538 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1539 btrfs_clean_old_snapshots(root);
1540 mutex_unlock(&root->fs_info->cleaner_mutex);
4cb5300b 1541 btrfs_run_defrag_inodes(root->fs_info);
76dda93c 1542 }
a74a4b97
CM		 1543
1544 if (freezing(current)) {
1545 refrigerator();
1546 } else {
a74a4b97 1547 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1548 if (!kthread_should_stop())
1549 schedule();
a74a4b97
CM		 1550 __set_current_state(TASK_RUNNING);
1551 }
1552 } while (!kthread_should_stop());
1553 return 0;
1554}
1555
1556static int transaction_kthread(void *arg)
1557{
1558 struct btrfs_root *root = arg;
1559 struct btrfs_trans_handle *trans;
1560 struct btrfs_transaction *cur;
8929ecfa 1561 u64 transid;
a74a4b97
CM		 1562 unsigned long now;
1563 unsigned long delay;
1564 int ret;
1565
1566 do {
a74a4b97
CM		 1567 delay = HZ * 30;
1568 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1569 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1570
a4abeea4 1571 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1572 cur = root->fs_info->running_transaction;
1573 if (!cur) {
a4abeea4 1574 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1575 goto sleep;
1576 }
31153d81 1577
a74a4b97 1578 now = get_seconds();
8929ecfa
YZ		 1579 if (!cur->blocked &&
1580 (now < cur->start_time || now - cur->start_time < 30)) {
a4abeea4 1581 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1582 delay = HZ * 5;
1583 goto sleep;
1584 }
8929ecfa 1585 transid = cur->transid;
a4abeea4 1586 spin_unlock(&root->fs_info->trans_lock);
56bec294 1587
7a7eaa40 1588 trans = btrfs_join_transaction(root);
3612b495 1589 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 1590 if (transid == trans->transid) {
1591 ret = btrfs_commit_transaction(trans, root);
1592 BUG_ON(ret);
1593 } else {
1594 btrfs_end_transaction(trans, root);
1595 }
a74a4b97
CM		 1596sleep:
1597 wake_up_process(root->fs_info->cleaner_kthread);
1598 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1599
1600 if (freezing(current)) {
1601 refrigerator();
1602 } else {
a74a4b97 1603 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1604 if (!kthread_should_stop() &&
1605 !btrfs_transaction_blocked(root->fs_info))
1606 schedule_timeout(delay);
a74a4b97
CM		 1607 __set_current_state(TASK_RUNNING);
1608 }
1609 } while (!kthread_should_stop());
1610 return 0;
1611}
1612
8a4b83cc 1613struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1614 struct btrfs_fs_devices *fs_devices,
1615 char *options)
2e635a27 1616{
db94535d
CM		 1617 u32 sectorsize;
1618 u32 nodesize;
1619 u32 leafsize;
1620 u32 blocksize;
87ee04eb 1621 u32 stripesize;
84234f3a 1622 u64 generation;
f2b636e8 1623 u64 features;
3de4586c 1624 struct btrfs_key location;
a061fc8d 1625 struct buffer_head *bh;
e02119d5 1626 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1627 GFP_NOFS);
d20f7043
CM		 1628 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1629 GFP_NOFS);
450ba0ea 1630 struct btrfs_root *tree_root = btrfs_sb(sb);
4891aca2 1631 struct btrfs_fs_info *fs_info = NULL;
e02119d5 1632 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1633 GFP_NOFS);
e02119d5 1634 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1635 GFP_NOFS);
e02119d5
CM		 1636 struct btrfs_root *log_tree_root;
1637
eb60ceac 1638 int ret;
e58ca020 1639 int err = -EINVAL;
4543df7e 1640
2c90e5d6 1641 struct btrfs_super_block *disk_super;
8790d502 1642
4891aca2 1643 if (!extent_root || !tree_root || !tree_root->fs_info ||
d20f7043 1644 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1645 err = -ENOMEM;
1646 goto fail;
1647 }
4891aca2 1648 fs_info = tree_root->fs_info;
76dda93c
YZ		 1649
1650 ret = init_srcu_struct(&fs_info->subvol_srcu);
1651 if (ret) {
1652 err = ret;
1653 goto fail;
1654 }
1655
1656 ret = setup_bdi(fs_info, &fs_info->bdi);
1657 if (ret) {
1658 err = ret;
1659 goto fail_srcu;
1660 }
1661
1662 fs_info->btree_inode = new_inode(sb);
1663 if (!fs_info->btree_inode) {
1664 err = -ENOMEM;
1665 goto fail_bdi;
1666 }
1667
a6591715 1668 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 1669
76dda93c 1670 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1671 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1672 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1673 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1674 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1675 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1676 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1677 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1678 spin_lock_init(&fs_info->delalloc_lock);
a4abeea4 1679 spin_lock_init(&fs_info->trans_lock);
31153d81 1680 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1681 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1682 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 1683 spin_lock_init(&fs_info->defrag_inodes_lock);
7585717f 1684 mutex_init(&fs_info->reloc_mutex);
19c00ddc 1685
58176a96 1686 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1687 fs_info->tree_root = tree_root;
1688 fs_info->extent_root = extent_root;
d20f7043 1689 fs_info->csum_root = csum_root;
0b86a832
CM		 1690 fs_info->chunk_root = chunk_root;
1691 fs_info->dev_root = dev_root;
8a4b83cc 1692 fs_info->fs_devices = fs_devices;
0b86a832 1693 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1694 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1695 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1696 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1697 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1698 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1699 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1700 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1701 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1702 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1703 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1704 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1705 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1706 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 1707 atomic_set(&fs_info->defrag_running, 0);
e20d96d6 1708 fs_info->sb = sb;
6f568d35 1709 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1710 fs_info->metadata_ratio = 0;
4cb5300b 1711 fs_info->defrag_inodes = RB_ROOT;
a4abeea4 1712 fs_info->trans_no_join = 0;
c8b97818 1713
b34b086c
CM		 1714 fs_info->thread_pool_size = min_t(unsigned long,
1715 num_online_cpus() + 2, 8);
0afbaf8c 1716
3eaa2885
CM		 1717 INIT_LIST_HEAD(&fs_info->ordered_extents);
1718 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX		 1719 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1720 GFP_NOFS);
1721 if (!fs_info->delayed_root) {
1722 err = -ENOMEM;
1723 goto fail_iput;
1724 }
1725 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 1726
a2de733c
AJ		 1727 mutex_init(&fs_info->scrub_lock);
1728 atomic_set(&fs_info->scrubs_running, 0);
1729 atomic_set(&fs_info->scrub_pause_req, 0);
1730 atomic_set(&fs_info->scrubs_paused, 0);
1731 atomic_set(&fs_info->scrub_cancel_req, 0);
1732 init_waitqueue_head(&fs_info->scrub_pause_wait);
1733 init_rwsem(&fs_info->scrub_super_lock);
1734 fs_info->scrub_workers_refcnt = 0;
a2de733c 1735
a061fc8d
CM		 1736 sb->s_blocksize = 4096;
1737 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1738 sb->s_bdi = &fs_info->bdi;
a061fc8d 1739
76dda93c
YZ		 1740 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1741 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1742 /*
1743 * we set the i_size on the btree inode to the max possible int.
1744 * the real end of the address space is determined by all of
1745 * the devices in the system
1746 */
1747 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1748 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1749 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1750
5d4f98a2 1751 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1752 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 1753 fs_info->btree_inode->i_mapping);
a8067e02 1754 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 1755
1756 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1757
76dda93c
YZ		 1758 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1759 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1760 sizeof(struct btrfs_key));
1761 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1762 insert_inode_hash(fs_info->btree_inode);
76dda93c 1763
0f9dd46c 1764 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1765 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1766
11833d66 1767 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 1768 fs_info->btree_inode->i_mapping);
11833d66 1769 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 1770 fs_info->btree_inode->i_mapping);
11833d66 1771 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1772 fs_info->do_barriers = 1;
e18e4809 1773
39279cc3 1774
5a3f23d5 1775 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1776 mutex_init(&fs_info->tree_log_mutex);
925baedd 1777 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1778 mutex_init(&fs_info->transaction_kthread_mutex);
1779 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1780 mutex_init(&fs_info->volume_mutex);
276e680d 1781 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1782 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1783 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1784
1785 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1786 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1787
e6dcd2dc 1788 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1789 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 1790 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 1791 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1792
0b86a832 1793 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1794 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1795
a512bbf8 1796 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 1797 if (!bh) {
1798 err = -EINVAL;
16cdcec7 1799 goto fail_alloc;
20b45077 1800 }
39279cc3 1801
a061fc8d 1802 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1803 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1804 sizeof(fs_info->super_for_commit));
a061fc8d 1805 brelse(bh);
5f39d397 1806
a061fc8d 1807 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1808
5f39d397 1809 disk_super = &fs_info->super_copy;
0f7d52f4 1810 if (!btrfs_super_root(disk_super))
16cdcec7 1811 goto fail_alloc;
0f7d52f4 1812
acce952b 1813 /* check FS state, whether FS is broken. */
1814 fs_info->fs_state |= btrfs_super_flags(disk_super);
1815
1816 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1817
75e7cb7f
LB		 1818 /*
1819 * In the long term, we'll store the compression type in the super
1820 * block, and it'll be used for per file compression control.
1821 */
1822 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1823
2b82032c
YZ		 1824 ret = btrfs_parse_options(tree_root, options);
1825 if (ret) {
1826 err = ret;
16cdcec7 1827 goto fail_alloc;
2b82032c 1828 }
dfe25020 1829
f2b636e8
JB		 1830 features = btrfs_super_incompat_flags(disk_super) &
1831 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1832 if (features) {
1833 printk(KERN_ERR "BTRFS: couldn't mount because of "
1834 "unsupported optional features (%Lx).\n",
21380931 1835 (unsigned long long)features);
f2b636e8 1836 err = -EINVAL;
16cdcec7 1837 goto fail_alloc;
f2b636e8
JB		 1838 }
1839
5d4f98a2 1840 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae
LZ		 1841 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1842 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1843 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1844 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 1845
f2b636e8
JB		 1846 features = btrfs_super_compat_ro_flags(disk_super) &
1847 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1848 if (!(sb->s_flags & MS_RDONLY) && features) {
1849 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1850 "unsupported option features (%Lx).\n",
21380931 1851 (unsigned long long)features);
f2b636e8 1852 err = -EINVAL;
16cdcec7 1853 goto fail_alloc;
f2b636e8 1854 }
61d92c32
CM		 1855
1856 btrfs_init_workers(&fs_info->generic_worker,
1857 "genwork", 1, NULL);
1858
5443be45 1859 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1860 fs_info->thread_pool_size,
1861 &fs_info->generic_worker);
c8b97818 1862
771ed689 1863 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1864 fs_info->thread_pool_size,
1865 &fs_info->generic_worker);
771ed689 1866
5443be45 1867 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1868 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1869 fs_info->thread_pool_size),
1870 &fs_info->generic_worker);
61b49440 1871
bab39bf9
JB		 1872 btrfs_init_workers(&fs_info->caching_workers, "cache",
1873 2, &fs_info->generic_worker);
1874
61b49440
CM		 1875 /* a higher idle thresh on the submit workers makes it much more
1876 * likely that bios will be send down in a sane order to the
1877 * devices
1878 */
1879 fs_info->submit_workers.idle_thresh = 64;
53863232 1880
771ed689 1881 fs_info->workers.idle_thresh = 16;
4a69a410 1882 fs_info->workers.ordered = 1;
61b49440 1883
771ed689
CM		 1884 fs_info->delalloc_workers.idle_thresh = 2;
1885 fs_info->delalloc_workers.ordered = 1;
1886
61d92c32
CM		 1887 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1888 &fs_info->generic_worker);
5443be45 1889 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1890 fs_info->thread_pool_size,
1891 &fs_info->generic_worker);
d20f7043 1892 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1893 fs_info->thread_pool_size,
1894 &fs_info->generic_worker);
cad321ad 1895 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1896 "endio-meta-write", fs_info->thread_pool_size,
1897 &fs_info->generic_worker);
5443be45 1898 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1899 fs_info->thread_pool_size,
1900 &fs_info->generic_worker);
0cb59c99
JB		 1901 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1902 1, &fs_info->generic_worker);
16cdcec7
MX		 1903 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1904 fs_info->thread_pool_size,
1905 &fs_info->generic_worker);
61b49440
CM		 1906
1907 /*
1908 * endios are largely parallel and should have a very
1909 * low idle thresh
1910 */
1911 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1912 fs_info->endio_meta_workers.idle_thresh = 4;
1913
9042846b
CM		 1914 fs_info->endio_write_workers.idle_thresh = 2;
1915 fs_info->endio_meta_write_workers.idle_thresh = 2;
1916
4543df7e 1917 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1918 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1919 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1920 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1921 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1922 btrfs_start_workers(&fs_info->endio_workers, 1);
1923 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1924 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1925 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1926 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
16cdcec7 1927 btrfs_start_workers(&fs_info->delayed_workers, 1);
bab39bf9 1928 btrfs_start_workers(&fs_info->caching_workers, 1);
4543df7e 1929
4575c9cc 1930 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1931 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1932 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1933
db94535d
CM		 1934 nodesize = btrfs_super_nodesize(disk_super);
1935 leafsize = btrfs_super_leafsize(disk_super);
1936 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1937 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1938 tree_root->nodesize = nodesize;
1939 tree_root->leafsize = leafsize;
1940 tree_root->sectorsize = sectorsize;
87ee04eb 1941 tree_root->stripesize = stripesize;
a061fc8d
CM		 1942
1943 sb->s_blocksize = sectorsize;
1944 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1945
39279cc3
CM		 1946 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1947 sizeof(disk_super->magic))) {
d397712b 1948 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1949 goto fail_sb_buffer;
1950 }
19c00ddc 1951
925baedd 1952 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1953 ret = btrfs_read_sys_array(tree_root);
925baedd 1954 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1955 if (ret) {
d397712b
CM		 1956 printk(KERN_WARNING "btrfs: failed to read the system "
1957 "array on %s\n", sb->s_id);
5d4f98a2 1958 goto fail_sb_buffer;
84eed90f 1959 }
0b86a832
CM		 1960
1961 blocksize = btrfs_level_size(tree_root,
1962 btrfs_super_chunk_root_level(disk_super));
84234f3a 1963 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1964
1965 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1966 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1967
1968 chunk_root->node = read_tree_block(chunk_root,
1969 btrfs_super_chunk_root(disk_super),
84234f3a 1970 blocksize, generation);
0b86a832 1971 BUG_ON(!chunk_root->node);
83121942
DW		 1972 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1973 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1974 sb->s_id);
1975 goto fail_chunk_root;
1976 }
5d4f98a2
YZ		 1977 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1978 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1979
e17cade2 1980 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 1981 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1982 BTRFS_UUID_SIZE);
e17cade2 1983
925baedd 1984 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1985 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1986 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1987 if (ret) {
d397712b
CM		 1988 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1989 sb->s_id);
2b82032c
YZ		 1990 goto fail_chunk_root;
1991 }
0b86a832 1992
dfe25020
CM		 1993 btrfs_close_extra_devices(fs_devices);
1994
db94535d
CM		 1995 blocksize = btrfs_level_size(tree_root,
1996 btrfs_super_root_level(disk_super));
84234f3a 1997 generation = btrfs_super_generation(disk_super);
0b86a832 1998
e20d96d6 1999 tree_root->node = read_tree_block(tree_root,
db94535d 2000 btrfs_super_root(disk_super),
84234f3a 2001 blocksize, generation);
39279cc3 2002 if (!tree_root->node)
2b82032c 2003 goto fail_chunk_root;
83121942
DW		 2004 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2005 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2006 sb->s_id);
2007 goto fail_tree_root;
2008 }
5d4f98a2
YZ		 2009 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2010 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 2011
2012 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 2013 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 2014 if (ret)
39279cc3 2015 goto fail_tree_root;
0b86a832
CM		 2016 extent_root->track_dirty = 1;
2017
2018 ret = find_and_setup_root(tree_root, fs_info,
2019 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 2020 if (ret)
2021 goto fail_extent_root;
5d4f98a2 2022 dev_root->track_dirty = 1;
3768f368 2023
d20f7043
CM		 2024 ret = find_and_setup_root(tree_root, fs_info,
2025 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2026 if (ret)
5d4f98a2 2027 goto fail_dev_root;
d20f7043
CM		 2028
2029 csum_root->track_dirty = 1;
2030
8929ecfa
YZ		 2031 fs_info->generation = generation;
2032 fs_info->last_trans_committed = generation;
2033 fs_info->data_alloc_profile = (u64)-1;
2034 fs_info->metadata_alloc_profile = (u64)-1;
2035 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2036
c59021f8 2037 ret = btrfs_init_space_info(fs_info);
2038 if (ret) {
2039 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2040 goto fail_block_groups;
2041 }
2042
1b1d1f66
JB		 2043 ret = btrfs_read_block_groups(extent_root);
2044 if (ret) {
2045 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2046 goto fail_block_groups;
2047 }
9078a3e1 2048
a74a4b97
CM		 2049 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2050 "btrfs-cleaner");
57506d50 2051 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2052 goto fail_block_groups;
a74a4b97
CM		 2053
2054 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2055 tree_root,
2056 "btrfs-transaction");
57506d50 2057 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2058 goto fail_cleaner;
a74a4b97 2059
c289811c
CM		 2060 if (!btrfs_test_opt(tree_root, SSD) &&
2061 !btrfs_test_opt(tree_root, NOSSD) &&
2062 !fs_info->fs_devices->rotating) {
2063 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2064 "mode\n");
2065 btrfs_set_opt(fs_info->mount_opt, SSD);
2066 }
2067
acce952b 2068 /* do not make disk changes in broken FS */
2069 if (btrfs_super_log_root(disk_super) != 0 &&
2070 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
e02119d5
CM		 2071 u64 bytenr = btrfs_super_log_root(disk_super);
2072
7c2ca468 2073 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2074 printk(KERN_WARNING "Btrfs log replay required "
2075 "on RO media\n");
7c2ca468
CM		 2076 err = -EIO;
2077 goto fail_trans_kthread;
2078 }
e02119d5
CM		 2079 blocksize =
2080 btrfs_level_size(tree_root,
2081 btrfs_super_log_root_level(disk_super));
d18a2c44 2082
676e4c86
DC		 2083 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2084 if (!log_tree_root) {
2085 err = -ENOMEM;
2086 goto fail_trans_kthread;
2087 }
e02119d5
CM		 2088
2089 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2090 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2091
2092 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2093 blocksize,
2094 generation + 1);
e02119d5
CM		 2095 ret = btrfs_recover_log_trees(log_tree_root);
2096 BUG_ON(ret);
e556ce2c
YZ		 2097
2098 if (sb->s_flags & MS_RDONLY) {
2099 ret = btrfs_commit_super(tree_root);
2100 BUG_ON(ret);
2101 }
e02119d5 2102 }
1a40e23b 2103
76dda93c
YZ		 2104 ret = btrfs_find_orphan_roots(tree_root);
2105 BUG_ON(ret);
2106
7c2ca468 2107 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ		 2108 ret = btrfs_cleanup_fs_roots(fs_info);
2109 BUG_ON(ret);
2110
5d4f98a2 2111 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2112 if (ret < 0) {
2113 printk(KERN_WARNING
2114 "btrfs: failed to recover relocation\n");
2115 err = -EINVAL;
2116 goto fail_trans_kthread;
2117 }
7c2ca468 2118 }
1a40e23b 2119
3de4586c
CM		 2120 location.objectid = BTRFS_FS_TREE_OBJECTID;
2121 location.type = BTRFS_ROOT_ITEM_KEY;
2122 location.offset = (u64)-1;
2123
3de4586c
CM		 2124 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2125 if (!fs_info->fs_root)
7c2ca468 2126 goto fail_trans_kthread;
3140c9a3
DC		 2127 if (IS_ERR(fs_info->fs_root)) {
2128 err = PTR_ERR(fs_info->fs_root);
2129 goto fail_trans_kthread;
2130 }
c289811c 2131
e3acc2a6
JB		 2132 if (!(sb->s_flags & MS_RDONLY)) {
2133 down_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2134 err = btrfs_orphan_cleanup(fs_info->fs_root);
2135 if (!err)
2136 err = btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6 2137 up_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2138 if (err) {
2139 close_ctree(tree_root);
2140 return ERR_PTR(err);
2141 }
e3acc2a6
JB		 2142 }
2143
0f7d52f4 2144 return tree_root;
39279cc3 2145
7c2ca468
CM		 2146fail_trans_kthread:
2147 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2148fail_cleaner:
a74a4b97 2149 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2150
2151 /*
2152 * make sure we're done with the btree inode before we stop our
2153 * kthreads
2154 */
2155 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2156 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2157
1b1d1f66
JB		 2158fail_block_groups:
2159 btrfs_free_block_groups(fs_info);
d20f7043 2160 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2161 free_extent_buffer(csum_root->commit_root);
2162fail_dev_root:
2163 free_extent_buffer(dev_root->node);
2164 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2165fail_extent_root:
2166 free_extent_buffer(extent_root->node);
5d4f98a2 2167 free_extent_buffer(extent_root->commit_root);
39279cc3 2168fail_tree_root:
5f39d397 2169 free_extent_buffer(tree_root->node);
5d4f98a2 2170 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2171fail_chunk_root:
2172 free_extent_buffer(chunk_root->node);
5d4f98a2 2173 free_extent_buffer(chunk_root->commit_root);
39279cc3 2174fail_sb_buffer:
61d92c32 2175 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2176 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2177 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2178 btrfs_stop_workers(&fs_info->workers);
2179 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2180 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2181 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2182 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2183 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2184 btrfs_stop_workers(&fs_info->submit_workers);
16cdcec7 2185 btrfs_stop_workers(&fs_info->delayed_workers);
bab39bf9 2186 btrfs_stop_workers(&fs_info->caching_workers);
16cdcec7
MX		 2187fail_alloc:
2188 kfree(fs_info->delayed_root);
4543df7e 2189fail_iput:
7c2ca468 2190 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2191 iput(fs_info->btree_inode);
7e662854 2192
dfe25020 2193 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2194 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2195fail_bdi:
7e662854 2196 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2197fail_srcu:
2198 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2199fail:
39279cc3
CM		 2200 kfree(extent_root);
2201 kfree(tree_root);
2202 kfree(fs_info);
83afeac4
JM		 2203 kfree(chunk_root);
2204 kfree(dev_root);
d20f7043 2205 kfree(csum_root);
39279cc3 2206 return ERR_PTR(err);
eb60ceac
CM		 2207}
2208
f2984462
CM		 2209static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2210{
2211 char b[BDEVNAME_SIZE];
2212
2213 if (uptodate) {
2214 set_buffer_uptodate(bh);
2215 } else {
7a36ddec 2216 printk_ratelimited(KERN_WARNING "lost page write due to "
f2984462
CM		 2217 "I/O error on %s\n",
2218 bdevname(bh->b_bdev, b));
1259ab75
CM		 2219 /* note, we dont' set_buffer_write_io_error because we have
2220 * our own ways of dealing with the IO errors
2221 */
f2984462
CM		 2222 clear_buffer_uptodate(bh);
2223 }
2224 unlock_buffer(bh);
2225 put_bh(bh);
2226}
2227
a512bbf8
YZ		 2228struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2229{
2230 struct buffer_head *bh;
2231 struct buffer_head *latest = NULL;
2232 struct btrfs_super_block *super;
2233 int i;
2234 u64 transid = 0;
2235 u64 bytenr;
2236
2237 /* we would like to check all the supers, but that would make
2238 * a btrfs mount succeed after a mkfs from a different FS.
2239 * So, we need to add a special mount option to scan for
2240 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2241 */
2242 for (i = 0; i < 1; i++) {
2243 bytenr = btrfs_sb_offset(i);
2244 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2245 break;
2246 bh = __bread(bdev, bytenr / 4096, 4096);
2247 if (!bh)
2248 continue;
2249
2250 super = (struct btrfs_super_block *)bh->b_data;
2251 if (btrfs_super_bytenr(super) != bytenr ||
2252 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2253 sizeof(super->magic))) {
2254 brelse(bh);
2255 continue;
2256 }
2257
2258 if (!latest || btrfs_super_generation(super) > transid) {
2259 brelse(latest);
2260 latest = bh;
2261 transid = btrfs_super_generation(super);
2262 } else {
2263 brelse(bh);
2264 }
2265 }
2266 return latest;
2267}
2268
4eedeb75
HH		 2269/*
2270 * this should be called twice, once with wait == 0 and
2271 * once with wait == 1. When wait == 0 is done, all the buffer heads
2272 * we write are pinned.
2273 *
2274 * They are released when wait == 1 is done.
2275 * max_mirrors must be the same for both runs, and it indicates how
2276 * many supers on this one device should be written.
2277 *
2278 * max_mirrors == 0 means to write them all.
2279 */
a512bbf8
YZ		 2280static int write_dev_supers(struct btrfs_device *device,
2281 struct btrfs_super_block *sb,
2282 int do_barriers, int wait, int max_mirrors)
2283{
2284 struct buffer_head *bh;
2285 int i;
2286 int ret;
2287 int errors = 0;
2288 u32 crc;
2289 u64 bytenr;
2290 int last_barrier = 0;
2291
2292 if (max_mirrors == 0)
2293 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2294
2295 /* make sure only the last submit_bh does a barrier */
2296 if (do_barriers) {
2297 for (i = 0; i < max_mirrors; i++) {
2298 bytenr = btrfs_sb_offset(i);
2299 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2300 device->total_bytes)
2301 break;
2302 last_barrier = i;
2303 }
2304 }
2305
2306 for (i = 0; i < max_mirrors; i++) {
2307 bytenr = btrfs_sb_offset(i);
2308 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2309 break;
2310
2311 if (wait) {
2312 bh = __find_get_block(device->bdev, bytenr / 4096,
2313 BTRFS_SUPER_INFO_SIZE);
2314 BUG_ON(!bh);
a512bbf8 2315 wait_on_buffer(bh);
4eedeb75
HH		 2316 if (!buffer_uptodate(bh))
2317 errors++;
2318
2319 /* drop our reference */
2320 brelse(bh);
2321
2322 /* drop the reference from the wait == 0 run */
2323 brelse(bh);
2324 continue;
a512bbf8
YZ		 2325 } else {
2326 btrfs_set_super_bytenr(sb, bytenr);
2327
2328 crc = ~(u32)0;
2329 crc = btrfs_csum_data(NULL, (char *)sb +
2330 BTRFS_CSUM_SIZE, crc,
2331 BTRFS_SUPER_INFO_SIZE -
2332 BTRFS_CSUM_SIZE);
2333 btrfs_csum_final(crc, sb->csum);
2334
4eedeb75
HH		 2335 /*
2336 * one reference for us, and we leave it for the
2337 * caller
2338 */
a512bbf8
YZ		 2339 bh = __getblk(device->bdev, bytenr / 4096,
2340 BTRFS_SUPER_INFO_SIZE);
2341 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2342
4eedeb75 2343 /* one reference for submit_bh */
a512bbf8 2344 get_bh(bh);
4eedeb75
HH		 2345
2346 set_buffer_uptodate(bh);
a512bbf8
YZ		 2347 lock_buffer(bh);
2348 bh->b_end_io = btrfs_end_buffer_write_sync;
2349 }
2350
c3b9a62c
CH		 2351 if (i == last_barrier && do_barriers)
2352 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2353 else
ffbd517d 2354 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8 2355
4eedeb75 2356 if (ret)
a512bbf8 2357 errors++;
a512bbf8
YZ		 2358 }
2359 return errors < i ? 0 : -1;
2360}
2361
2362int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2363{
e5e9a520 2364 struct list_head *head;
f2984462 2365 struct btrfs_device *dev;
a061fc8d 2366 struct btrfs_super_block *sb;
f2984462 2367 struct btrfs_dev_item *dev_item;
f2984462
CM		 2368 int ret;
2369 int do_barriers;
a236aed1
CM		 2370 int max_errors;
2371 int total_errors = 0;
a061fc8d 2372 u64 flags;
f2984462 2373
a236aed1 2374 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2375 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2376
a061fc8d
CM		 2377 sb = &root->fs_info->super_for_commit;
2378 dev_item = &sb->dev_item;
e5e9a520 2379
174ba509 2380 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 2381 head = &root->fs_info->fs_devices->devices;
1f78160c 2382 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 2383 if (!dev->bdev) {
2384 total_errors++;
2385 continue;
2386 }
2b82032c 2387 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2388 continue;
2389
2b82032c 2390 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2391 btrfs_set_stack_device_type(dev_item, dev->type);
2392 btrfs_set_stack_device_id(dev_item, dev->devid);
2393 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2394 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2395 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2396 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2397 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2398 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2399 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2400
a061fc8d
CM		 2401 flags = btrfs_super_flags(sb);
2402 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2403
a512bbf8 2404 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2405 if (ret)
2406 total_errors++;
f2984462 2407 }
a236aed1 2408 if (total_errors > max_errors) {
d397712b
CM		 2409 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2410 total_errors);
a236aed1
CM		 2411 BUG();
2412 }
f2984462 2413
a512bbf8 2414 total_errors = 0;
1f78160c 2415 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 2416 if (!dev->bdev)
2417 continue;
2b82032c 2418 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2419 continue;
2420
a512bbf8
YZ		 2421 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2422 if (ret)
2423 total_errors++;
f2984462 2424 }
174ba509 2425 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2426 if (total_errors > max_errors) {
d397712b
CM		 2427 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2428 total_errors);
a236aed1
CM		 2429 BUG();
2430 }
f2984462
CM		 2431 return 0;
2432}
2433
a512bbf8
YZ		 2434int write_ctree_super(struct btrfs_trans_handle *trans,
2435 struct btrfs_root *root, int max_mirrors)
eb60ceac 2436{
e66f709b 2437 int ret;
5f39d397 2438
a512bbf8 2439 ret = write_all_supers(root, max_mirrors);
5f39d397 2440 return ret;
cfaa7295
CM		 2441}
2442
5eda7b5e 2443int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2444{
4df27c4d 2445 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2446 radix_tree_delete(&fs_info->fs_roots_radix,
2447 (unsigned long)root->root_key.objectid);
4df27c4d 2448 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2449
2450 if (btrfs_root_refs(&root->root_item) == 0)
2451 synchronize_srcu(&fs_info->subvol_srcu);
2452
581bb050
LZ		 2453 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2454 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d
YZ		 2455 free_fs_root(root);
2456 return 0;
2457}
2458
2459static void free_fs_root(struct btrfs_root *root)
2460{
82d5902d 2461 iput(root->cache_inode);
4df27c4d 2462 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 2463 if (root->anon_dev)
2464 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 2465 free_extent_buffer(root->node);
2466 free_extent_buffer(root->commit_root);
581bb050
LZ		 2467 kfree(root->free_ino_ctl);
2468 kfree(root->free_ino_pinned);
d397712b 2469 kfree(root->name);
2619ba1f 2470 kfree(root);
2619ba1f
CM		 2471}
2472
35b7e476 2473static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2474{
2475 int ret;
2476 struct btrfs_root *gang[8];
2477 int i;
2478
76dda93c
YZ		 2479 while (!list_empty(&fs_info->dead_roots)) {
2480 gang[0] = list_entry(fs_info->dead_roots.next,
2481 struct btrfs_root, root_list);
2482 list_del(&gang[0]->root_list);
2483
2484 if (gang[0]->in_radix) {
2485 btrfs_free_fs_root(fs_info, gang[0]);
2486 } else {
2487 free_extent_buffer(gang[0]->node);
2488 free_extent_buffer(gang[0]->commit_root);
2489 kfree(gang[0]);
2490 }
2491 }
2492
d397712b 2493 while (1) {
0f7d52f4
CM		 2494 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2495 (void **)gang, 0,
2496 ARRAY_SIZE(gang));
2497 if (!ret)
2498 break;
2619ba1f 2499 for (i = 0; i < ret; i++)
5eda7b5e 2500 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2501 }
2502 return 0;
2503}
b4100d64 2504
c146afad 2505int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2506{
c146afad
YZ		 2507 u64 root_objectid = 0;
2508 struct btrfs_root *gang[8];
2509 int i;
3768f368 2510 int ret;
e089f05c 2511
c146afad
YZ		 2512 while (1) {
2513 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2514 (void **)gang, root_objectid,
2515 ARRAY_SIZE(gang));
2516 if (!ret)
2517 break;
5d4f98a2
YZ		 2518
2519 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 2520 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 2521 int err;
2522
c146afad 2523 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 2524 err = btrfs_orphan_cleanup(gang[i]);
2525 if (err)
2526 return err;
c146afad
YZ		 2527 }
2528 root_objectid++;
2529 }
2530 return 0;
2531}
a2135011 2532
c146afad
YZ		 2533int btrfs_commit_super(struct btrfs_root *root)
2534{
2535 struct btrfs_trans_handle *trans;
2536 int ret;
a74a4b97 2537
c146afad 2538 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2539 btrfs_run_delayed_iputs(root);
a74a4b97 2540 btrfs_clean_old_snapshots(root);
c146afad 2541 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2542
2543 /* wait until ongoing cleanup work done */
2544 down_write(&root->fs_info->cleanup_work_sem);
2545 up_write(&root->fs_info->cleanup_work_sem);
2546
7a7eaa40 2547 trans = btrfs_join_transaction(root);
3612b495
TI		 2548 if (IS_ERR(trans))
2549 return PTR_ERR(trans);
54aa1f4d 2550 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2551 BUG_ON(ret);
2552 /* run commit again to drop the original snapshot */
7a7eaa40 2553 trans = btrfs_join_transaction(root);
3612b495
TI		 2554 if (IS_ERR(trans))
2555 return PTR_ERR(trans);
79154b1b
CM		 2556 btrfs_commit_transaction(trans, root);
2557 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2558 BUG_ON(ret);
d6bfde87 2559
a512bbf8 2560 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2561 return ret;
2562}
2563
2564int close_ctree(struct btrfs_root *root)
2565{
2566 struct btrfs_fs_info *fs_info = root->fs_info;
2567 int ret;
2568
2569 fs_info->closing = 1;
2570 smp_mb();
2571
a2de733c 2572 btrfs_scrub_cancel(root);
4cb5300b
CM		 2573
2574 /* wait for any defraggers to finish */
2575 wait_event(fs_info->transaction_wait,
2576 (atomic_read(&fs_info->defrag_running) == 0));
2577
2578 /* clear out the rbtree of defraggable inodes */
2579 btrfs_run_defrag_inodes(root->fs_info);
2580
0af3d00b 2581 btrfs_put_block_group_cache(fs_info);
acce952b 2582
2583 /*
2584 * Here come 2 situations when btrfs is broken to flip readonly:
2585 *
2586 * 1. when btrfs flips readonly somewhere else before
2587 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2588 * and btrfs will skip to write sb directly to keep
2589 * ERROR state on disk.
2590 *
2591 * 2. when btrfs flips readonly just in btrfs_commit_super,
ae0e47f0 2592 * and in such case, btrfs cannot write sb via btrfs_commit_super,
acce952b 2593 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2594 * btrfs will cleanup all FS resources first and write sb then.
2595 */
c146afad 2596 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 2597 ret = btrfs_commit_super(root);
2598 if (ret)
2599 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2600 }
2601
2602 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2603 ret = btrfs_error_commit_super(root);
d397712b
CM		 2604 if (ret)
2605 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2606 }
0f7d52f4 2607
8929ecfa
YZ		 2608 kthread_stop(root->fs_info->transaction_kthread);
2609 kthread_stop(root->fs_info->cleaner_kthread);
2610
f25784b3
YZ		 2611 fs_info->closing = 2;
2612 smp_mb();
2613
b0c68f8b 2614 if (fs_info->delalloc_bytes) {
d397712b 2615 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2616 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2617 }
31153d81 2618 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2619 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2620 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2621 }
bcc63abb 2622
5d4f98a2
YZ		 2623 free_extent_buffer(fs_info->extent_root->node);
2624 free_extent_buffer(fs_info->extent_root->commit_root);
2625 free_extent_buffer(fs_info->tree_root->node);
2626 free_extent_buffer(fs_info->tree_root->commit_root);
2627 free_extent_buffer(root->fs_info->chunk_root->node);
2628 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2629 free_extent_buffer(root->fs_info->dev_root->node);
2630 free_extent_buffer(root->fs_info->dev_root->commit_root);
2631 free_extent_buffer(root->fs_info->csum_root->node);
2632 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2633
9078a3e1 2634 btrfs_free_block_groups(root->fs_info);
d10c5f31 2635
c146afad 2636 del_fs_roots(fs_info);
d10c5f31 2637
c146afad 2638 iput(fs_info->btree_inode);
16cdcec7 2639 kfree(fs_info->delayed_root);
9ad6b7bc 2640
61d92c32 2641 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2642 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2643 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2644 btrfs_stop_workers(&fs_info->workers);
2645 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2646 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2647 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2648 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2649 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2650 btrfs_stop_workers(&fs_info->submit_workers);
16cdcec7 2651 btrfs_stop_workers(&fs_info->delayed_workers);
bab39bf9 2652 btrfs_stop_workers(&fs_info->caching_workers);
d6bfde87 2653
dfe25020 2654 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2655 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2656
04160088 2657 bdi_destroy(&fs_info->bdi);
76dda93c 2658 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2659
0f7d52f4 2660 kfree(fs_info->extent_root);
0f7d52f4 2661 kfree(fs_info->tree_root);
0b86a832
CM		 2662 kfree(fs_info->chunk_root);
2663 kfree(fs_info->dev_root);
d20f7043 2664 kfree(fs_info->csum_root);
83a4d548
LZ		 2665 kfree(fs_info);
2666
eb60ceac
CM		 2667 return 0;
2668}
2669
1259ab75 2670int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2671{
1259ab75 2672 int ret;
810191ff 2673 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2674
2ac55d41
JB		 2675 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2676 NULL);
1259ab75
CM		 2677 if (!ret)
2678 return ret;
2679
2680 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2681 parent_transid);
2682 return !ret;
5f39d397
CM		 2683}
2684
2685int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2686{
810191ff 2687 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2688 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2689 buf);
2690}
6702ed49 2691
5f39d397
CM		 2692void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2693{
810191ff 2694 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2695 u64 transid = btrfs_header_generation(buf);
2696 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2697 int was_dirty;
b4ce94de 2698
b9447ef8 2699 btrfs_assert_tree_locked(buf);
ccd467d6 2700 if (transid != root->fs_info->generation) {
d397712b
CM		 2701 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2702 "found %llu running %llu\n",
db94535d 2703 (unsigned long long)buf->start,
d397712b
CM		 2704 (unsigned long long)transid,
2705 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2706 WARN_ON(1);
2707 }
b9473439
CM		 2708 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2709 buf);
2710 if (!was_dirty) {
2711 spin_lock(&root->fs_info->delalloc_lock);
2712 root->fs_info->dirty_metadata_bytes += buf->len;
2713 spin_unlock(&root->fs_info->delalloc_lock);
2714 }
eb60ceac
CM		 2715}
2716
d3c2fdcf 2717void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
16cdcec7
MX		 2718{
2719 /*
2720 * looks as though older kernels can get into trouble with
2721 * this code, they end up stuck in balance_dirty_pages forever
2722 */
2723 u64 num_dirty;
2724 unsigned long thresh = 32 * 1024 * 1024;
2725
2726 if (current->flags & PF_MEMALLOC)
2727 return;
2728
2729 btrfs_balance_delayed_items(root);
2730
2731 num_dirty = root->fs_info->dirty_metadata_bytes;
2732
2733 if (num_dirty > thresh) {
2734 balance_dirty_pages_ratelimited_nr(
2735 root->fs_info->btree_inode->i_mapping, 1);
2736 }
2737 return;
2738}
2739
2740void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2741{
188de649
CM		 2742 /*
2743 * looks as though older kernels can get into trouble with
2744 * this code, they end up stuck in balance_dirty_pages forever
2745 */
d6bfde87 2746 u64 num_dirty;
771ed689 2747 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2748
6933c02e 2749 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2750 return;
2751
585ad2c3
CM		 2752 num_dirty = root->fs_info->dirty_metadata_bytes;
2753
d6bfde87
CM		 2754 if (num_dirty > thresh) {
2755 balance_dirty_pages_ratelimited_nr(
d7fc640e 2756 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2757 }
188de649 2758 return;
35b7e476 2759}
6b80053d 2760
ca7a79ad 2761int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2762{
810191ff 2763 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2764 int ret;
ca7a79ad 2765 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2766 if (ret == 0)
b4ce94de 2767 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2768 return ret;
6b80053d 2769}
0da5468f 2770
4bef0848
CM		 2771int btree_lock_page_hook(struct page *page)
2772{
2773 struct inode *inode = page->mapping->host;
b9473439 2774 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2775 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2776 struct extent_buffer *eb;
2777 unsigned long len;
2778 u64 bytenr = page_offset(page);
2779
2780 if (page->private == EXTENT_PAGE_PRIVATE)
2781 goto out;
2782
2783 len = page->private >> 2;
f09d1f60 2784 eb = find_extent_buffer(io_tree, bytenr, len);
4bef0848
CM		 2785 if (!eb)
2786 goto out;
2787
2788 btrfs_tree_lock(eb);
4bef0848 2789 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2790
2791 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2792 spin_lock(&root->fs_info->delalloc_lock);
2793 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2794 root->fs_info->dirty_metadata_bytes -= eb->len;
2795 else
2796 WARN_ON(1);
2797 spin_unlock(&root->fs_info->delalloc_lock);
2798 }
2799
4bef0848
CM		 2800 btrfs_tree_unlock(eb);
2801 free_extent_buffer(eb);
2802out:
2803 lock_page(page);
2804 return 0;
2805}
2806
acce952b 2807static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2808 int read_only)
2809{
2810 if (read_only)
2811 return;
2812
2813 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2814 printk(KERN_WARNING "warning: mount fs with errors, "
2815 "running btrfsck is recommended\n");
2816}
2817
2818int btrfs_error_commit_super(struct btrfs_root *root)
2819{
2820 int ret;
2821
2822 mutex_lock(&root->fs_info->cleaner_mutex);
2823 btrfs_run_delayed_iputs(root);
2824 mutex_unlock(&root->fs_info->cleaner_mutex);
2825
2826 down_write(&root->fs_info->cleanup_work_sem);
2827 up_write(&root->fs_info->cleanup_work_sem);
2828
2829 /* cleanup FS via transaction */
2830 btrfs_cleanup_transaction(root);
2831
2832 ret = write_ctree_super(NULL, root, 0);
2833
2834 return ret;
2835}
2836
2837static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2838{
2839 struct btrfs_inode *btrfs_inode;
2840 struct list_head splice;
2841
2842 INIT_LIST_HEAD(&splice);
2843
2844 mutex_lock(&root->fs_info->ordered_operations_mutex);
2845 spin_lock(&root->fs_info->ordered_extent_lock);
2846
2847 list_splice_init(&root->fs_info->ordered_operations, &splice);
2848 while (!list_empty(&splice)) {
2849 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2850 ordered_operations);
2851
2852 list_del_init(&btrfs_inode->ordered_operations);
2853
2854 btrfs_invalidate_inodes(btrfs_inode->root);
2855 }
2856
2857 spin_unlock(&root->fs_info->ordered_extent_lock);
2858 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2859
2860 return 0;
2861}
2862
2863static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2864{
2865 struct list_head splice;
2866 struct btrfs_ordered_extent *ordered;
2867 struct inode *inode;
2868
2869 INIT_LIST_HEAD(&splice);
2870
2871 spin_lock(&root->fs_info->ordered_extent_lock);
2872
2873 list_splice_init(&root->fs_info->ordered_extents, &splice);
2874 while (!list_empty(&splice)) {
2875 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2876 root_extent_list);
2877
2878 list_del_init(&ordered->root_extent_list);
2879 atomic_inc(&ordered->refs);
2880
2881 /* the inode may be getting freed (in sys_unlink path). */
2882 inode = igrab(ordered->inode);
2883
2884 spin_unlock(&root->fs_info->ordered_extent_lock);
2885 if (inode)
2886 iput(inode);
2887
2888 atomic_set(&ordered->refs, 1);
2889 btrfs_put_ordered_extent(ordered);
2890
2891 spin_lock(&root->fs_info->ordered_extent_lock);
2892 }
2893
2894 spin_unlock(&root->fs_info->ordered_extent_lock);
2895
2896 return 0;
2897}
2898
2899static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2900 struct btrfs_root *root)
2901{
2902 struct rb_node *node;
2903 struct btrfs_delayed_ref_root *delayed_refs;
2904 struct btrfs_delayed_ref_node *ref;
2905 int ret = 0;
2906
2907 delayed_refs = &trans->delayed_refs;
2908
2909 spin_lock(&delayed_refs->lock);
2910 if (delayed_refs->num_entries == 0) {
cfece4db 2911 spin_unlock(&delayed_refs->lock);
acce952b 2912 printk(KERN_INFO "delayed_refs has NO entry\n");
2913 return ret;
2914 }
2915
2916 node = rb_first(&delayed_refs->root);
2917 while (node) {
2918 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2919 node = rb_next(node);
2920
2921 ref->in_tree = 0;
2922 rb_erase(&ref->rb_node, &delayed_refs->root);
2923 delayed_refs->num_entries--;
2924
2925 atomic_set(&ref->refs, 1);
2926 if (btrfs_delayed_ref_is_head(ref)) {
2927 struct btrfs_delayed_ref_head *head;
2928
2929 head = btrfs_delayed_node_to_head(ref);
2930 mutex_lock(&head->mutex);
2931 kfree(head->extent_op);
2932 delayed_refs->num_heads--;
2933 if (list_empty(&head->cluster))
2934 delayed_refs->num_heads_ready--;
2935 list_del_init(&head->cluster);
2936 mutex_unlock(&head->mutex);
2937 }
2938
2939 spin_unlock(&delayed_refs->lock);
2940 btrfs_put_delayed_ref(ref);
2941
2942 cond_resched();
2943 spin_lock(&delayed_refs->lock);
2944 }
2945
2946 spin_unlock(&delayed_refs->lock);
2947
2948 return ret;
2949}
2950
2951static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2952{
2953 struct btrfs_pending_snapshot *snapshot;
2954 struct list_head splice;
2955
2956 INIT_LIST_HEAD(&splice);
2957
2958 list_splice_init(&t->pending_snapshots, &splice);
2959
2960 while (!list_empty(&splice)) {
2961 snapshot = list_entry(splice.next,
2962 struct btrfs_pending_snapshot,
2963 list);
2964
2965 list_del_init(&snapshot->list);
2966
2967 kfree(snapshot);
2968 }
2969
2970 return 0;
2971}
2972
2973static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2974{
2975 struct btrfs_inode *btrfs_inode;
2976 struct list_head splice;
2977
2978 INIT_LIST_HEAD(&splice);
2979
acce952b 2980 spin_lock(&root->fs_info->delalloc_lock);
5be76758 2981 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
acce952b 2982
2983 while (!list_empty(&splice)) {
2984 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2985 delalloc_inodes);
2986
2987 list_del_init(&btrfs_inode->delalloc_inodes);
2988
2989 btrfs_invalidate_inodes(btrfs_inode->root);
2990 }
2991
2992 spin_unlock(&root->fs_info->delalloc_lock);
2993
2994 return 0;
2995}
2996
2997static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2998 struct extent_io_tree *dirty_pages,
2999 int mark)
3000{
3001 int ret;
3002 struct page *page;
3003 struct inode *btree_inode = root->fs_info->btree_inode;
3004 struct extent_buffer *eb;
3005 u64 start = 0;
3006 u64 end;
3007 u64 offset;
3008 unsigned long index;
3009
3010 while (1) {
3011 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3012 mark);
3013 if (ret)
3014 break;
3015
3016 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3017 while (start <= end) {
3018 index = start >> PAGE_CACHE_SHIFT;
3019 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3020 page = find_get_page(btree_inode->i_mapping, index);
3021 if (!page)
3022 continue;
3023 offset = page_offset(page);
3024
3025 spin_lock(&dirty_pages->buffer_lock);
3026 eb = radix_tree_lookup(
3027 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3028 offset >> PAGE_CACHE_SHIFT);
3029 spin_unlock(&dirty_pages->buffer_lock);
3030 if (eb) {
3031 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3032 &eb->bflags);
3033 atomic_set(&eb->refs, 1);
3034 }
3035 if (PageWriteback(page))
3036 end_page_writeback(page);
3037
3038 lock_page(page);
3039 if (PageDirty(page)) {
3040 clear_page_dirty_for_io(page);
3041 spin_lock_irq(&page->mapping->tree_lock);
3042 radix_tree_tag_clear(&page->mapping->page_tree,
3043 page_index(page),
3044 PAGECACHE_TAG_DIRTY);
3045 spin_unlock_irq(&page->mapping->tree_lock);
3046 }
3047
3048 page->mapping->a_ops->invalidatepage(page, 0);
3049 unlock_page(page);
3050 }
3051 }
3052
3053 return ret;
3054}
3055
3056static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3057 struct extent_io_tree *pinned_extents)
3058{
3059 struct extent_io_tree *unpin;
3060 u64 start;
3061 u64 end;
3062 int ret;
3063
3064 unpin = pinned_extents;
3065 while (1) {
3066 ret = find_first_extent_bit(unpin, 0, &start, &end,
3067 EXTENT_DIRTY);
3068 if (ret)
3069 break;
3070
3071 /* opt_discard */
5378e607
LD		 3072 if (btrfs_test_opt(root, DISCARD))
3073 ret = btrfs_error_discard_extent(root, start,
3074 end + 1 - start,
3075 NULL);
acce952b 3076
3077 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3078 btrfs_error_unpin_extent_range(root, start, end);
3079 cond_resched();
3080 }
3081
3082 return 0;
3083}
3084
3085static int btrfs_cleanup_transaction(struct btrfs_root *root)
3086{
3087 struct btrfs_transaction *t;
3088 LIST_HEAD(list);
3089
3090 WARN_ON(1);
3091
acce952b 3092 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3093
a4abeea4 3094 spin_lock(&root->fs_info->trans_lock);
acce952b 3095 list_splice_init(&root->fs_info->trans_list, &list);
a4abeea4
JB		 3096 root->fs_info->trans_no_join = 1;
3097 spin_unlock(&root->fs_info->trans_lock);
3098
acce952b 3099 while (!list_empty(&list)) {
3100 t = list_entry(list.next, struct btrfs_transaction, list);
3101 if (!t)
3102 break;
3103
3104 btrfs_destroy_ordered_operations(root);
3105
3106 btrfs_destroy_ordered_extents(root);
3107
3108 btrfs_destroy_delayed_refs(t, root);
3109
3110 btrfs_block_rsv_release(root,
3111 &root->fs_info->trans_block_rsv,
3112 t->dirty_pages.dirty_bytes);
3113
3114 /* FIXME: cleanup wait for commit */
3115 t->in_commit = 1;
3116 t->blocked = 1;
3117 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3118 wake_up(&root->fs_info->transaction_blocked_wait);
3119
3120 t->blocked = 0;
3121 if (waitqueue_active(&root->fs_info->transaction_wait))
3122 wake_up(&root->fs_info->transaction_wait);
acce952b 3123
acce952b 3124 t->commit_done = 1;
3125 if (waitqueue_active(&t->commit_wait))
3126 wake_up(&t->commit_wait);
acce952b 3127
3128 btrfs_destroy_pending_snapshots(t);
3129
3130 btrfs_destroy_delalloc_inodes(root);
3131
a4abeea4 3132 spin_lock(&root->fs_info->trans_lock);
acce952b 3133 root->fs_info->running_transaction = NULL;
a4abeea4 3134 spin_unlock(&root->fs_info->trans_lock);
acce952b 3135
3136 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3137 EXTENT_DIRTY);
3138
3139 btrfs_destroy_pinned_extent(root,
3140 root->fs_info->pinned_extents);
3141
13c5a93e 3142 atomic_set(&t->use_count, 0);
acce952b 3143 list_del_init(&t->list);
3144 memset(t, 0, sizeof(*t));
3145 kmem_cache_free(btrfs_transaction_cachep, t);
3146 }
3147
a4abeea4
JB		 3148 spin_lock(&root->fs_info->trans_lock);
3149 root->fs_info->trans_no_join = 0;
3150 spin_unlock(&root->fs_info->trans_lock);
acce952b 3151 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 3152
3153 return 0;
3154}
3155
d1310b2e 3156static struct extent_io_ops btree_extent_io_ops = {
4bef0848 3157 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 3158 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 3159 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 3160 /* note we're sharing with inode.c for the merge bio hook */
3161 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3162};
Linux 6.x block layer and io_uring tree(s)