• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "hash.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
49 #include "raid56.h"
50 #include "sysfs.h"
51 #include "qgroup.h"
52 #include "compression.h"
53 
54 #ifdef CONFIG_X86
55 #include <asm/cpufeature.h>
56 #endif
57 
58 #define BTRFS_SUPER_FLAG_SUPP	(BTRFS_HEADER_FLAG_WRITTEN |\
59 				 BTRFS_HEADER_FLAG_RELOC |\
60 				 BTRFS_SUPER_FLAG_ERROR |\
61 				 BTRFS_SUPER_FLAG_SEEDING |\
62 				 BTRFS_SUPER_FLAG_METADUMP)
63 
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68 				    int read_only);
69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 				      struct btrfs_root *root);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 					struct extent_io_tree *dirty_pages,
75 					int mark);
76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 				       struct extent_io_tree *pinned_extents);
78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
79 static void btrfs_error_commit_super(struct btrfs_root *root);
80 
81 /*
82  * btrfs_end_io_wq structs are used to do processing in task context when an IO
83  * is complete.  This is used during reads to verify checksums, and it is used
84  * by writes to insert metadata for new file extents after IO is complete.
85  */
86 struct btrfs_end_io_wq {
87 	struct bio *bio;
88 	bio_end_io_t *end_io;
89 	void *private;
90 	struct btrfs_fs_info *info;
91 	int error;
92 	enum btrfs_wq_endio_type metadata;
93 	struct list_head list;
94 	struct btrfs_work work;
95 };
96 
97 static struct kmem_cache *btrfs_end_io_wq_cache;
98 
btrfs_end_io_wq_init(void)99 int __init btrfs_end_io_wq_init(void)
100 {
101 	btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 					sizeof(struct btrfs_end_io_wq),
103 					0,
104 					SLAB_MEM_SPREAD,
105 					NULL);
106 	if (!btrfs_end_io_wq_cache)
107 		return -ENOMEM;
108 	return 0;
109 }
110 
btrfs_end_io_wq_exit(void)111 void btrfs_end_io_wq_exit(void)
112 {
113 	kmem_cache_destroy(btrfs_end_io_wq_cache);
114 }
115 
116 /*
117  * async submit bios are used to offload expensive checksumming
118  * onto the worker threads.  They checksum file and metadata bios
119  * just before they are sent down the IO stack.
120  */
121 struct async_submit_bio {
122 	struct inode *inode;
123 	struct bio *bio;
124 	struct list_head list;
125 	extent_submit_bio_hook_t *submit_bio_start;
126 	extent_submit_bio_hook_t *submit_bio_done;
127 	int mirror_num;
128 	unsigned long bio_flags;
129 	/*
130 	 * bio_offset is optional, can be used if the pages in the bio
131 	 * can't tell us where in the file the bio should go
132 	 */
133 	u64 bio_offset;
134 	struct btrfs_work work;
135 	int error;
136 };
137 
138 /*
139  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
140  * eb, the lockdep key is determined by the btrfs_root it belongs to and
141  * the level the eb occupies in the tree.
142  *
143  * Different roots are used for different purposes and may nest inside each
144  * other and they require separate keysets.  As lockdep keys should be
145  * static, assign keysets according to the purpose of the root as indicated
146  * by btrfs_root->objectid.  This ensures that all special purpose roots
147  * have separate keysets.
148  *
149  * Lock-nesting across peer nodes is always done with the immediate parent
150  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
151  * subclass to avoid triggering lockdep warning in such cases.
152  *
153  * The key is set by the readpage_end_io_hook after the buffer has passed
154  * csum validation but before the pages are unlocked.  It is also set by
155  * btrfs_init_new_buffer on freshly allocated blocks.
156  *
157  * We also add a check to make sure the highest level of the tree is the
158  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
159  * needs update as well.
160  */
161 #ifdef CONFIG_DEBUG_LOCK_ALLOC
162 # if BTRFS_MAX_LEVEL != 8
163 #  error
164 # endif
165 
166 static struct btrfs_lockdep_keyset {
167 	u64			id;		/* root objectid */
168 	const char		*name_stem;	/* lock name stem */
169 	char			names[BTRFS_MAX_LEVEL + 1][20];
170 	struct lock_class_key	keys[BTRFS_MAX_LEVEL + 1];
171 } btrfs_lockdep_keysets[] = {
172 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	.name_stem = "root"	},
173 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	.name_stem = "extent"	},
174 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	.name_stem = "chunk"	},
175 	{ .id = BTRFS_DEV_TREE_OBJECTID,	.name_stem = "dev"	},
176 	{ .id = BTRFS_FS_TREE_OBJECTID,		.name_stem = "fs"	},
177 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	.name_stem = "csum"	},
178 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	.name_stem = "quota"	},
179 	{ .id = BTRFS_TREE_LOG_OBJECTID,	.name_stem = "log"	},
180 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	.name_stem = "treloc"	},
181 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	.name_stem = "dreloc"	},
182 	{ .id = BTRFS_UUID_TREE_OBJECTID,	.name_stem = "uuid"	},
183 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	.name_stem = "free-space" },
184 	{ .id = 0,				.name_stem = "tree"	},
185 };
186 
btrfs_init_lockdep(void)187 void __init btrfs_init_lockdep(void)
188 {
189 	int i, j;
190 
191 	/* initialize lockdep class names */
192 	for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
193 		struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
194 
195 		for (j = 0; j < ARRAY_SIZE(ks->names); j++)
196 			snprintf(ks->names[j], sizeof(ks->names[j]),
197 				 "btrfs-%s-%02d", ks->name_stem, j);
198 	}
199 }
200 
btrfs_set_buffer_lockdep_class(u64 objectid,struct extent_buffer * eb,int level)201 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
202 				    int level)
203 {
204 	struct btrfs_lockdep_keyset *ks;
205 
206 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
207 
208 	/* find the matching keyset, id 0 is the default entry */
209 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
210 		if (ks->id == objectid)
211 			break;
212 
213 	lockdep_set_class_and_name(&eb->lock,
214 				   &ks->keys[level], ks->names[level]);
215 }
216 
217 #endif
218 
219 /*
220  * extents on the btree inode are pretty simple, there's one extent
221  * that covers the entire device
222  */
btree_get_extent(struct inode * inode,struct page * page,size_t pg_offset,u64 start,u64 len,int create)223 static struct extent_map *btree_get_extent(struct inode *inode,
224 		struct page *page, size_t pg_offset, u64 start, u64 len,
225 		int create)
226 {
227 	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
228 	struct extent_map *em;
229 	int ret;
230 
231 	read_lock(&em_tree->lock);
232 	em = lookup_extent_mapping(em_tree, start, len);
233 	if (em) {
234 		em->bdev =
235 			BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
236 		read_unlock(&em_tree->lock);
237 		goto out;
238 	}
239 	read_unlock(&em_tree->lock);
240 
241 	em = alloc_extent_map();
242 	if (!em) {
243 		em = ERR_PTR(-ENOMEM);
244 		goto out;
245 	}
246 	em->start = 0;
247 	em->len = (u64)-1;
248 	em->block_len = (u64)-1;
249 	em->block_start = 0;
250 	em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
251 
252 	write_lock(&em_tree->lock);
253 	ret = add_extent_mapping(em_tree, em, 0);
254 	if (ret == -EEXIST) {
255 		free_extent_map(em);
256 		em = lookup_extent_mapping(em_tree, start, len);
257 		if (!em)
258 			em = ERR_PTR(-EIO);
259 	} else if (ret) {
260 		free_extent_map(em);
261 		em = ERR_PTR(ret);
262 	}
263 	write_unlock(&em_tree->lock);
264 
265 out:
266 	return em;
267 }
268 
btrfs_csum_data(char * data,u32 seed,size_t len)269 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
270 {
271 	return btrfs_crc32c(seed, data, len);
272 }
273 
btrfs_csum_final(u32 crc,char * result)274 void btrfs_csum_final(u32 crc, char *result)
275 {
276 	put_unaligned_le32(~crc, result);
277 }
278 
279 /*
280  * compute the csum for a btree block, and either verify it or write it
281  * into the csum field of the block.
282  */
csum_tree_block(struct btrfs_fs_info * fs_info,struct extent_buffer * buf,int verify)283 static int csum_tree_block(struct btrfs_fs_info *fs_info,
284 			   struct extent_buffer *buf,
285 			   int verify)
286 {
287 	u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
288 	char *result = NULL;
289 	unsigned long len;
290 	unsigned long cur_len;
291 	unsigned long offset = BTRFS_CSUM_SIZE;
292 	char *kaddr;
293 	unsigned long map_start;
294 	unsigned long map_len;
295 	int err;
296 	u32 crc = ~(u32)0;
297 	unsigned long inline_result;
298 
299 	len = buf->len - offset;
300 	while (len > 0) {
301 		err = map_private_extent_buffer(buf, offset, 32,
302 					&kaddr, &map_start, &map_len);
303 		if (err)
304 			return err;
305 		cur_len = min(len, map_len - (offset - map_start));
306 		crc = btrfs_csum_data(kaddr + offset - map_start,
307 				      crc, cur_len);
308 		len -= cur_len;
309 		offset += cur_len;
310 	}
311 	if (csum_size > sizeof(inline_result)) {
312 		result = kzalloc(csum_size, GFP_NOFS);
313 		if (!result)
314 			return -ENOMEM;
315 	} else {
316 		result = (char *)&inline_result;
317 	}
318 
319 	btrfs_csum_final(crc, result);
320 
321 	if (verify) {
322 		if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
323 			u32 val;
324 			u32 found = 0;
325 			memcpy(&found, result, csum_size);
326 
327 			read_extent_buffer(buf, &val, 0, csum_size);
328 			btrfs_warn_rl(fs_info,
329 				"%s checksum verify failed on %llu wanted %X found %X level %d",
330 				fs_info->sb->s_id, buf->start,
331 				val, found, btrfs_header_level(buf));
332 			if (result != (char *)&inline_result)
333 				kfree(result);
334 			return -EUCLEAN;
335 		}
336 	} else {
337 		write_extent_buffer(buf, result, 0, csum_size);
338 	}
339 	if (result != (char *)&inline_result)
340 		kfree(result);
341 	return 0;
342 }
343 
344 /*
345  * we can't consider a given block up to date unless the transid of the
346  * block matches the transid in the parent node's pointer.  This is how we
347  * detect blocks that either didn't get written at all or got written
348  * in the wrong place.
349  */
verify_parent_transid(struct extent_io_tree * io_tree,struct extent_buffer * eb,u64 parent_transid,int atomic)350 static int verify_parent_transid(struct extent_io_tree *io_tree,
351 				 struct extent_buffer *eb, u64 parent_transid,
352 				 int atomic)
353 {
354 	struct extent_state *cached_state = NULL;
355 	int ret;
356 	bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
357 
358 	if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
359 		return 0;
360 
361 	if (atomic)
362 		return -EAGAIN;
363 
364 	if (need_lock) {
365 		btrfs_tree_read_lock(eb);
366 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
367 	}
368 
369 	lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
370 			 &cached_state);
371 	if (extent_buffer_uptodate(eb) &&
372 	    btrfs_header_generation(eb) == parent_transid) {
373 		ret = 0;
374 		goto out;
375 	}
376 	btrfs_err_rl(eb->fs_info,
377 		"parent transid verify failed on %llu wanted %llu found %llu",
378 			eb->start,
379 			parent_transid, btrfs_header_generation(eb));
380 	ret = 1;
381 
382 	/*
383 	 * Things reading via commit roots that don't have normal protection,
384 	 * like send, can have a really old block in cache that may point at a
385 	 * block that has been freed and re-allocated.  So don't clear uptodate
386 	 * if we find an eb that is under IO (dirty/writeback) because we could
387 	 * end up reading in the stale data and then writing it back out and
388 	 * making everybody very sad.
389 	 */
390 	if (!extent_buffer_under_io(eb))
391 		clear_extent_buffer_uptodate(eb);
392 out:
393 	unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
394 			     &cached_state, GFP_NOFS);
395 	if (need_lock)
396 		btrfs_tree_read_unlock_blocking(eb);
397 	return ret;
398 }
399 
400 /*
401  * Return 0 if the superblock checksum type matches the checksum value of that
402  * algorithm. Pass the raw disk superblock data.
403  */
btrfs_check_super_csum(struct btrfs_fs_info * fs_info,char * raw_disk_sb)404 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
405 				  char *raw_disk_sb)
406 {
407 	struct btrfs_super_block *disk_sb =
408 		(struct btrfs_super_block *)raw_disk_sb;
409 	u16 csum_type = btrfs_super_csum_type(disk_sb);
410 	int ret = 0;
411 
412 	if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
413 		u32 crc = ~(u32)0;
414 		const int csum_size = sizeof(crc);
415 		char result[csum_size];
416 
417 		/*
418 		 * The super_block structure does not span the whole
419 		 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
420 		 * is filled with zeros and is included in the checksum.
421 		 */
422 		crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
423 				crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
424 		btrfs_csum_final(crc, result);
425 
426 		if (memcmp(raw_disk_sb, result, csum_size))
427 			ret = 1;
428 	}
429 
430 	if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431 		btrfs_err(fs_info, "unsupported checksum algorithm %u",
432 				csum_type);
433 		ret = 1;
434 	}
435 
436 	return ret;
437 }
438 
439 /*
440  * helper to read a given tree block, doing retries as required when
441  * the checksums don't match and we have alternate mirrors to try.
442  */
btree_read_extent_buffer_pages(struct btrfs_root * root,struct extent_buffer * eb,u64 parent_transid)443 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
444 					  struct extent_buffer *eb,
445 					  u64 parent_transid)
446 {
447 	struct extent_io_tree *io_tree;
448 	int failed = 0;
449 	int ret;
450 	int num_copies = 0;
451 	int mirror_num = 0;
452 	int failed_mirror = 0;
453 
454 	clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455 	io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
456 	while (1) {
457 		ret = read_extent_buffer_pages(io_tree, eb, WAIT_COMPLETE,
458 					       btree_get_extent, mirror_num);
459 		if (!ret) {
460 			if (!verify_parent_transid(io_tree, eb,
461 						   parent_transid, 0))
462 				break;
463 			else
464 				ret = -EIO;
465 		}
466 
467 		/*
468 		 * This buffer's crc is fine, but its contents are corrupted, so
469 		 * there is no reason to read the other copies, they won't be
470 		 * any less wrong.
471 		 */
472 		if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
473 			break;
474 
475 		num_copies = btrfs_num_copies(root->fs_info,
476 					      eb->start, eb->len);
477 		if (num_copies == 1)
478 			break;
479 
480 		if (!failed_mirror) {
481 			failed = 1;
482 			failed_mirror = eb->read_mirror;
483 		}
484 
485 		mirror_num++;
486 		if (mirror_num == failed_mirror)
487 			mirror_num++;
488 
489 		if (mirror_num > num_copies)
490 			break;
491 	}
492 
493 	if (failed && !ret && failed_mirror)
494 		repair_eb_io_failure(root, eb, failed_mirror);
495 
496 	return ret;
497 }
498 
499 /*
500  * checksum a dirty tree block before IO.  This has extra checks to make sure
501  * we only fill in the checksum field in the first page of a multi-page block
502  */
503 
csum_dirty_buffer(struct btrfs_fs_info * fs_info,struct page * page)504 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
505 {
506 	u64 start = page_offset(page);
507 	u64 found_start;
508 	struct extent_buffer *eb;
509 
510 	eb = (struct extent_buffer *)page->private;
511 	if (page != eb->pages[0])
512 		return 0;
513 
514 	found_start = btrfs_header_bytenr(eb);
515 	/*
516 	 * Please do not consolidate these warnings into a single if.
517 	 * It is useful to know what went wrong.
518 	 */
519 	if (WARN_ON(found_start != start))
520 		return -EUCLEAN;
521 	if (WARN_ON(!PageUptodate(page)))
522 		return -EUCLEAN;
523 
524 	ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
525 			btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
526 
527 	return csum_tree_block(fs_info, eb, 0);
528 }
529 
check_tree_block_fsid(struct btrfs_fs_info * fs_info,struct extent_buffer * eb)530 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
531 				 struct extent_buffer *eb)
532 {
533 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
534 	u8 fsid[BTRFS_UUID_SIZE];
535 	int ret = 1;
536 
537 	read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
538 	while (fs_devices) {
539 		if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
540 			ret = 0;
541 			break;
542 		}
543 		fs_devices = fs_devices->seed;
544 	}
545 	return ret;
546 }
547 
548 #define CORRUPT(reason, eb, root, slot)				\
549 	btrfs_crit(root->fs_info, "corrupt %s, %s: block=%llu,"	\
550 		   " root=%llu, slot=%d",			\
551 		   btrfs_header_level(eb) == 0 ? "leaf" : "node",\
552 		   reason, btrfs_header_bytenr(eb), root->objectid, slot)
553 
check_leaf(struct btrfs_root * root,struct extent_buffer * leaf)554 static noinline int check_leaf(struct btrfs_root *root,
555 			       struct extent_buffer *leaf)
556 {
557 	struct btrfs_key key;
558 	struct btrfs_key leaf_key;
559 	u32 nritems = btrfs_header_nritems(leaf);
560 	int slot;
561 
562 	/*
563 	 * Extent buffers from a relocation tree have a owner field that
564 	 * corresponds to the subvolume tree they are based on. So just from an
565 	 * extent buffer alone we can not find out what is the id of the
566 	 * corresponding subvolume tree, so we can not figure out if the extent
567 	 * buffer corresponds to the root of the relocation tree or not. So skip
568 	 * this check for relocation trees.
569 	 */
570 	if (nritems == 0 && !btrfs_header_flag(leaf, BTRFS_HEADER_FLAG_RELOC)) {
571 		struct btrfs_root *check_root;
572 
573 		key.objectid = btrfs_header_owner(leaf);
574 		key.type = BTRFS_ROOT_ITEM_KEY;
575 		key.offset = (u64)-1;
576 
577 		check_root = btrfs_get_fs_root(root->fs_info, &key, false);
578 		/*
579 		 * The only reason we also check NULL here is that during
580 		 * open_ctree() some roots has not yet been set up.
581 		 */
582 		if (!IS_ERR_OR_NULL(check_root)) {
583 			struct extent_buffer *eb;
584 
585 			eb = btrfs_root_node(check_root);
586 			/* if leaf is the root, then it's fine */
587 			if (leaf != eb) {
588 				CORRUPT("non-root leaf's nritems is 0",
589 					leaf, check_root, 0);
590 				free_extent_buffer(eb);
591 				return -EIO;
592 			}
593 			free_extent_buffer(eb);
594 		}
595 		return 0;
596 	}
597 
598 	if (nritems == 0)
599 		return 0;
600 
601 	/* Check the 0 item */
602 	if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
603 	    BTRFS_LEAF_DATA_SIZE(root)) {
604 		CORRUPT("invalid item offset size pair", leaf, root, 0);
605 		return -EIO;
606 	}
607 
608 	/*
609 	 * Check to make sure each items keys are in the correct order and their
610 	 * offsets make sense.  We only have to loop through nritems-1 because
611 	 * we check the current slot against the next slot, which verifies the
612 	 * next slot's offset+size makes sense and that the current's slot
613 	 * offset is correct.
614 	 */
615 	for (slot = 0; slot < nritems - 1; slot++) {
616 		btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
617 		btrfs_item_key_to_cpu(leaf, &key, slot + 1);
618 
619 		/* Make sure the keys are in the right order */
620 		if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
621 			CORRUPT("bad key order", leaf, root, slot);
622 			return -EIO;
623 		}
624 
625 		/*
626 		 * Make sure the offset and ends are right, remember that the
627 		 * item data starts at the end of the leaf and grows towards the
628 		 * front.
629 		 */
630 		if (btrfs_item_offset_nr(leaf, slot) !=
631 			btrfs_item_end_nr(leaf, slot + 1)) {
632 			CORRUPT("slot offset bad", leaf, root, slot);
633 			return -EIO;
634 		}
635 
636 		/*
637 		 * Check to make sure that we don't point outside of the leaf,
638 		 * just in case all the items are consistent to each other, but
639 		 * all point outside of the leaf.
640 		 */
641 		if (btrfs_item_end_nr(leaf, slot) >
642 		    BTRFS_LEAF_DATA_SIZE(root)) {
643 			CORRUPT("slot end outside of leaf", leaf, root, slot);
644 			return -EIO;
645 		}
646 	}
647 
648 	return 0;
649 }
650 
check_node(struct btrfs_root * root,struct extent_buffer * node)651 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
652 {
653 	unsigned long nr = btrfs_header_nritems(node);
654 	struct btrfs_key key, next_key;
655 	int slot;
656 	u64 bytenr;
657 	int ret = 0;
658 
659 	if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root)) {
660 		btrfs_crit(root->fs_info,
661 			   "corrupt node: block %llu root %llu nritems %lu",
662 			   node->start, root->objectid, nr);
663 		return -EIO;
664 	}
665 
666 	for (slot = 0; slot < nr - 1; slot++) {
667 		bytenr = btrfs_node_blockptr(node, slot);
668 		btrfs_node_key_to_cpu(node, &key, slot);
669 		btrfs_node_key_to_cpu(node, &next_key, slot + 1);
670 
671 		if (!bytenr) {
672 			CORRUPT("invalid item slot", node, root, slot);
673 			ret = -EIO;
674 			goto out;
675 		}
676 
677 		if (btrfs_comp_cpu_keys(&key, &next_key) >= 0) {
678 			CORRUPT("bad key order", node, root, slot);
679 			ret = -EIO;
680 			goto out;
681 		}
682 	}
683 out:
684 	return ret;
685 }
686 
btree_readpage_end_io_hook(struct btrfs_io_bio * io_bio,u64 phy_offset,struct page * page,u64 start,u64 end,int mirror)687 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
688 				      u64 phy_offset, struct page *page,
689 				      u64 start, u64 end, int mirror)
690 {
691 	u64 found_start;
692 	int found_level;
693 	struct extent_buffer *eb;
694 	struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
695 	struct btrfs_fs_info *fs_info = root->fs_info;
696 	int ret = 0;
697 	int reads_done;
698 
699 	if (!page->private)
700 		goto out;
701 
702 	eb = (struct extent_buffer *)page->private;
703 
704 	/* the pending IO might have been the only thing that kept this buffer
705 	 * in memory.  Make sure we have a ref for all this other checks
706 	 */
707 	extent_buffer_get(eb);
708 
709 	reads_done = atomic_dec_and_test(&eb->io_pages);
710 	if (!reads_done)
711 		goto err;
712 
713 	eb->read_mirror = mirror;
714 	if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
715 		ret = -EIO;
716 		goto err;
717 	}
718 
719 	found_start = btrfs_header_bytenr(eb);
720 	if (found_start != eb->start) {
721 		btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
722 			     found_start, eb->start);
723 		ret = -EIO;
724 		goto err;
725 	}
726 	if (check_tree_block_fsid(fs_info, eb)) {
727 		btrfs_err_rl(fs_info, "bad fsid on block %llu",
728 			     eb->start);
729 		ret = -EIO;
730 		goto err;
731 	}
732 	found_level = btrfs_header_level(eb);
733 	if (found_level >= BTRFS_MAX_LEVEL) {
734 		btrfs_err(fs_info, "bad tree block level %d",
735 			  (int)btrfs_header_level(eb));
736 		ret = -EIO;
737 		goto err;
738 	}
739 
740 	btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
741 				       eb, found_level);
742 
743 	ret = csum_tree_block(fs_info, eb, 1);
744 	if (ret)
745 		goto err;
746 
747 	/*
748 	 * If this is a leaf block and it is corrupt, set the corrupt bit so
749 	 * that we don't try and read the other copies of this block, just
750 	 * return -EIO.
751 	 */
752 	if (found_level == 0 && check_leaf(root, eb)) {
753 		set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
754 		ret = -EIO;
755 	}
756 
757 	if (found_level > 0 && check_node(root, eb))
758 		ret = -EIO;
759 
760 	if (!ret)
761 		set_extent_buffer_uptodate(eb);
762 err:
763 	if (reads_done &&
764 	    test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
765 		btree_readahead_hook(fs_info, eb, eb->start, ret);
766 
767 	if (ret) {
768 		/*
769 		 * our io error hook is going to dec the io pages
770 		 * again, we have to make sure it has something
771 		 * to decrement
772 		 */
773 		atomic_inc(&eb->io_pages);
774 		clear_extent_buffer_uptodate(eb);
775 	}
776 	free_extent_buffer(eb);
777 out:
778 	return ret;
779 }
780 
btree_io_failed_hook(struct page * page,int failed_mirror)781 static int btree_io_failed_hook(struct page *page, int failed_mirror)
782 {
783 	struct extent_buffer *eb;
784 
785 	eb = (struct extent_buffer *)page->private;
786 	set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
787 	eb->read_mirror = failed_mirror;
788 	atomic_dec(&eb->io_pages);
789 	if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
790 		btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
791 	return -EIO;	/* we fixed nothing */
792 }
793 
end_workqueue_bio(struct bio * bio)794 static void end_workqueue_bio(struct bio *bio)
795 {
796 	struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
797 	struct btrfs_fs_info *fs_info;
798 	struct btrfs_workqueue *wq;
799 	btrfs_work_func_t func;
800 
801 	fs_info = end_io_wq->info;
802 	end_io_wq->error = bio->bi_error;
803 
804 	if (bio_op(bio) == REQ_OP_WRITE) {
805 		if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
806 			wq = fs_info->endio_meta_write_workers;
807 			func = btrfs_endio_meta_write_helper;
808 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
809 			wq = fs_info->endio_freespace_worker;
810 			func = btrfs_freespace_write_helper;
811 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
812 			wq = fs_info->endio_raid56_workers;
813 			func = btrfs_endio_raid56_helper;
814 		} else {
815 			wq = fs_info->endio_write_workers;
816 			func = btrfs_endio_write_helper;
817 		}
818 	} else {
819 		if (unlikely(end_io_wq->metadata ==
820 			     BTRFS_WQ_ENDIO_DIO_REPAIR)) {
821 			wq = fs_info->endio_repair_workers;
822 			func = btrfs_endio_repair_helper;
823 		} else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
824 			wq = fs_info->endio_raid56_workers;
825 			func = btrfs_endio_raid56_helper;
826 		} else if (end_io_wq->metadata) {
827 			wq = fs_info->endio_meta_workers;
828 			func = btrfs_endio_meta_helper;
829 		} else {
830 			wq = fs_info->endio_workers;
831 			func = btrfs_endio_helper;
832 		}
833 	}
834 
835 	btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
836 	btrfs_queue_work(wq, &end_io_wq->work);
837 }
838 
btrfs_bio_wq_end_io(struct btrfs_fs_info * info,struct bio * bio,enum btrfs_wq_endio_type metadata)839 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
840 			enum btrfs_wq_endio_type metadata)
841 {
842 	struct btrfs_end_io_wq *end_io_wq;
843 
844 	end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
845 	if (!end_io_wq)
846 		return -ENOMEM;
847 
848 	end_io_wq->private = bio->bi_private;
849 	end_io_wq->end_io = bio->bi_end_io;
850 	end_io_wq->info = info;
851 	end_io_wq->error = 0;
852 	end_io_wq->bio = bio;
853 	end_io_wq->metadata = metadata;
854 
855 	bio->bi_private = end_io_wq;
856 	bio->bi_end_io = end_workqueue_bio;
857 	return 0;
858 }
859 
btrfs_async_submit_limit(struct btrfs_fs_info * info)860 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
861 {
862 	unsigned long limit = min_t(unsigned long,
863 				    info->thread_pool_size,
864 				    info->fs_devices->open_devices);
865 	return 256 * limit;
866 }
867 
run_one_async_start(struct btrfs_work * work)868 static void run_one_async_start(struct btrfs_work *work)
869 {
870 	struct async_submit_bio *async;
871 	int ret;
872 
873 	async = container_of(work, struct  async_submit_bio, work);
874 	ret = async->submit_bio_start(async->inode, async->bio,
875 				      async->mirror_num, async->bio_flags,
876 				      async->bio_offset);
877 	if (ret)
878 		async->error = ret;
879 }
880 
run_one_async_done(struct btrfs_work * work)881 static void run_one_async_done(struct btrfs_work *work)
882 {
883 	struct btrfs_fs_info *fs_info;
884 	struct async_submit_bio *async;
885 	int limit;
886 
887 	async = container_of(work, struct  async_submit_bio, work);
888 	fs_info = BTRFS_I(async->inode)->root->fs_info;
889 
890 	limit = btrfs_async_submit_limit(fs_info);
891 	limit = limit * 2 / 3;
892 
893 	/*
894 	 * atomic_dec_return implies a barrier for waitqueue_active
895 	 */
896 	if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
897 	    waitqueue_active(&fs_info->async_submit_wait))
898 		wake_up(&fs_info->async_submit_wait);
899 
900 	/* If an error occurred we just want to clean up the bio and move on */
901 	if (async->error) {
902 		async->bio->bi_error = async->error;
903 		bio_endio(async->bio);
904 		return;
905 	}
906 
907 	async->submit_bio_done(async->inode, async->bio, async->mirror_num,
908 			       async->bio_flags, async->bio_offset);
909 }
910 
run_one_async_free(struct btrfs_work * work)911 static void run_one_async_free(struct btrfs_work *work)
912 {
913 	struct async_submit_bio *async;
914 
915 	async = container_of(work, struct  async_submit_bio, work);
916 	kfree(async);
917 }
918 
btrfs_wq_submit_bio(struct btrfs_fs_info * fs_info,struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags,u64 bio_offset,extent_submit_bio_hook_t * submit_bio_start,extent_submit_bio_hook_t * submit_bio_done)919 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
920 			struct bio *bio, int mirror_num,
921 			unsigned long bio_flags,
922 			u64 bio_offset,
923 			extent_submit_bio_hook_t *submit_bio_start,
924 			extent_submit_bio_hook_t *submit_bio_done)
925 {
926 	struct async_submit_bio *async;
927 
928 	async = kmalloc(sizeof(*async), GFP_NOFS);
929 	if (!async)
930 		return -ENOMEM;
931 
932 	async->inode = inode;
933 	async->bio = bio;
934 	async->mirror_num = mirror_num;
935 	async->submit_bio_start = submit_bio_start;
936 	async->submit_bio_done = submit_bio_done;
937 
938 	btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
939 			run_one_async_done, run_one_async_free);
940 
941 	async->bio_flags = bio_flags;
942 	async->bio_offset = bio_offset;
943 
944 	async->error = 0;
945 
946 	atomic_inc(&fs_info->nr_async_submits);
947 
948 	if (bio->bi_opf & REQ_SYNC)
949 		btrfs_set_work_high_priority(&async->work);
950 
951 	btrfs_queue_work(fs_info->workers, &async->work);
952 
953 	while (atomic_read(&fs_info->async_submit_draining) &&
954 	      atomic_read(&fs_info->nr_async_submits)) {
955 		wait_event(fs_info->async_submit_wait,
956 			   (atomic_read(&fs_info->nr_async_submits) == 0));
957 	}
958 
959 	return 0;
960 }
961 
btree_csum_one_bio(struct bio * bio)962 static int btree_csum_one_bio(struct bio *bio)
963 {
964 	struct bio_vec *bvec;
965 	struct btrfs_root *root;
966 	int i, ret = 0;
967 
968 	bio_for_each_segment_all(bvec, bio, i) {
969 		root = BTRFS_I(bvec->bv_page->mapping->host)->root;
970 		ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
971 		if (ret)
972 			break;
973 	}
974 
975 	return ret;
976 }
977 
__btree_submit_bio_start(struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags,u64 bio_offset)978 static int __btree_submit_bio_start(struct inode *inode, struct bio *bio,
979 				    int mirror_num, unsigned long bio_flags,
980 				    u64 bio_offset)
981 {
982 	/*
983 	 * when we're called for a write, we're already in the async
984 	 * submission context.  Just jump into btrfs_map_bio
985 	 */
986 	return btree_csum_one_bio(bio);
987 }
988 
__btree_submit_bio_done(struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags,u64 bio_offset)989 static int __btree_submit_bio_done(struct inode *inode, struct bio *bio,
990 				 int mirror_num, unsigned long bio_flags,
991 				 u64 bio_offset)
992 {
993 	int ret;
994 
995 	/*
996 	 * when we're called for a write, we're already in the async
997 	 * submission context.  Just jump into btrfs_map_bio
998 	 */
999 	ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 1);
1000 	if (ret) {
1001 		bio->bi_error = ret;
1002 		bio_endio(bio);
1003 	}
1004 	return ret;
1005 }
1006 
check_async_write(struct inode * inode,unsigned long bio_flags)1007 static int check_async_write(struct inode *inode, unsigned long bio_flags)
1008 {
1009 	if (bio_flags & EXTENT_BIO_TREE_LOG)
1010 		return 0;
1011 #ifdef CONFIG_X86
1012 	if (static_cpu_has(X86_FEATURE_XMM4_2))
1013 		return 0;
1014 #endif
1015 	return 1;
1016 }
1017 
btree_submit_bio_hook(struct inode * inode,struct bio * bio,int mirror_num,unsigned long bio_flags,u64 bio_offset)1018 static int btree_submit_bio_hook(struct inode *inode, struct bio *bio,
1019 				 int mirror_num, unsigned long bio_flags,
1020 				 u64 bio_offset)
1021 {
1022 	int async = check_async_write(inode, bio_flags);
1023 	int ret;
1024 
1025 	if (bio_op(bio) != REQ_OP_WRITE) {
1026 		/*
1027 		 * called for a read, do the setup so that checksum validation
1028 		 * can happen in the async kernel threads
1029 		 */
1030 		ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
1031 					  bio, BTRFS_WQ_ENDIO_METADATA);
1032 		if (ret)
1033 			goto out_w_error;
1034 		ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
1035 	} else if (!async) {
1036 		ret = btree_csum_one_bio(bio);
1037 		if (ret)
1038 			goto out_w_error;
1039 		ret = btrfs_map_bio(BTRFS_I(inode)->root, bio, mirror_num, 0);
1040 	} else {
1041 		/*
1042 		 * kthread helpers are used to submit writes so that
1043 		 * checksumming can happen in parallel across all CPUs
1044 		 */
1045 		ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1046 					  inode, bio, mirror_num, 0,
1047 					  bio_offset,
1048 					  __btree_submit_bio_start,
1049 					  __btree_submit_bio_done);
1050 	}
1051 
1052 	if (ret)
1053 		goto out_w_error;
1054 	return 0;
1055 
1056 out_w_error:
1057 	bio->bi_error = ret;
1058 	bio_endio(bio);
1059 	return ret;
1060 }
1061 
1062 #ifdef CONFIG_MIGRATION
btree_migratepage(struct address_space * mapping,struct page * newpage,struct page * page,enum migrate_mode mode)1063 static int btree_migratepage(struct address_space *mapping,
1064 			struct page *newpage, struct page *page,
1065 			enum migrate_mode mode)
1066 {
1067 	/*
1068 	 * we can't safely write a btree page from here,
1069 	 * we haven't done the locking hook
1070 	 */
1071 	if (PageDirty(page))
1072 		return -EAGAIN;
1073 	/*
1074 	 * Buffers may be managed in a filesystem specific way.
1075 	 * We must have no buffers or drop them.
1076 	 */
1077 	if (page_has_private(page) &&
1078 	    !try_to_release_page(page, GFP_KERNEL))
1079 		return -EAGAIN;
1080 	return migrate_page(mapping, newpage, page, mode);
1081 }
1082 #endif
1083 
1084 
btree_writepages(struct address_space * mapping,struct writeback_control * wbc)1085 static int btree_writepages(struct address_space *mapping,
1086 			    struct writeback_control *wbc)
1087 {
1088 	struct btrfs_fs_info *fs_info;
1089 	int ret;
1090 
1091 	if (wbc->sync_mode == WB_SYNC_NONE) {
1092 
1093 		if (wbc->for_kupdate)
1094 			return 0;
1095 
1096 		fs_info = BTRFS_I(mapping->host)->root->fs_info;
1097 		/* this is a bit racy, but that's ok */
1098 		ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1099 					     BTRFS_DIRTY_METADATA_THRESH);
1100 		if (ret < 0)
1101 			return 0;
1102 	}
1103 	return btree_write_cache_pages(mapping, wbc);
1104 }
1105 
btree_readpage(struct file * file,struct page * page)1106 static int btree_readpage(struct file *file, struct page *page)
1107 {
1108 	struct extent_io_tree *tree;
1109 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1110 	return extent_read_full_page(tree, page, btree_get_extent, 0);
1111 }
1112 
btree_releasepage(struct page * page,gfp_t gfp_flags)1113 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1114 {
1115 	if (PageWriteback(page) || PageDirty(page))
1116 		return 0;
1117 
1118 	return try_release_extent_buffer(page);
1119 }
1120 
btree_invalidatepage(struct page * page,unsigned int offset,unsigned int length)1121 static void btree_invalidatepage(struct page *page, unsigned int offset,
1122 				 unsigned int length)
1123 {
1124 	struct extent_io_tree *tree;
1125 	tree = &BTRFS_I(page->mapping->host)->io_tree;
1126 	extent_invalidatepage(tree, page, offset);
1127 	btree_releasepage(page, GFP_NOFS);
1128 	if (PagePrivate(page)) {
1129 		btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1130 			   "page private not zero on page %llu",
1131 			   (unsigned long long)page_offset(page));
1132 		ClearPagePrivate(page);
1133 		set_page_private(page, 0);
1134 		put_page(page);
1135 	}
1136 }
1137 
btree_set_page_dirty(struct page * page)1138 static int btree_set_page_dirty(struct page *page)
1139 {
1140 #ifdef DEBUG
1141 	struct extent_buffer *eb;
1142 
1143 	BUG_ON(!PagePrivate(page));
1144 	eb = (struct extent_buffer *)page->private;
1145 	BUG_ON(!eb);
1146 	BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1147 	BUG_ON(!atomic_read(&eb->refs));
1148 	btrfs_assert_tree_locked(eb);
1149 #endif
1150 	return __set_page_dirty_nobuffers(page);
1151 }
1152 
1153 static const struct address_space_operations btree_aops = {
1154 	.readpage	= btree_readpage,
1155 	.writepages	= btree_writepages,
1156 	.releasepage	= btree_releasepage,
1157 	.invalidatepage = btree_invalidatepage,
1158 #ifdef CONFIG_MIGRATION
1159 	.migratepage	= btree_migratepage,
1160 #endif
1161 	.set_page_dirty = btree_set_page_dirty,
1162 };
1163 
readahead_tree_block(struct btrfs_root * root,u64 bytenr)1164 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1165 {
1166 	struct extent_buffer *buf = NULL;
1167 	struct inode *btree_inode = root->fs_info->btree_inode;
1168 
1169 	buf = btrfs_find_create_tree_block(root, bytenr);
1170 	if (IS_ERR(buf))
1171 		return;
1172 	read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1173 				 buf, WAIT_NONE, btree_get_extent, 0);
1174 	free_extent_buffer(buf);
1175 }
1176 
reada_tree_block_flagged(struct btrfs_root * root,u64 bytenr,int mirror_num,struct extent_buffer ** eb)1177 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1178 			 int mirror_num, struct extent_buffer **eb)
1179 {
1180 	struct extent_buffer *buf = NULL;
1181 	struct inode *btree_inode = root->fs_info->btree_inode;
1182 	struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1183 	int ret;
1184 
1185 	buf = btrfs_find_create_tree_block(root, bytenr);
1186 	if (IS_ERR(buf))
1187 		return 0;
1188 
1189 	set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1190 
1191 	ret = read_extent_buffer_pages(io_tree, buf, WAIT_PAGE_LOCK,
1192 				       btree_get_extent, mirror_num);
1193 	if (ret) {
1194 		free_extent_buffer(buf);
1195 		return ret;
1196 	}
1197 
1198 	if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1199 		free_extent_buffer(buf);
1200 		return -EIO;
1201 	} else if (extent_buffer_uptodate(buf)) {
1202 		*eb = buf;
1203 	} else {
1204 		free_extent_buffer(buf);
1205 	}
1206 	return 0;
1207 }
1208 
btrfs_find_tree_block(struct btrfs_fs_info * fs_info,u64 bytenr)1209 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1210 					    u64 bytenr)
1211 {
1212 	return find_extent_buffer(fs_info, bytenr);
1213 }
1214 
btrfs_find_create_tree_block(struct btrfs_root * root,u64 bytenr)1215 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1216 						 u64 bytenr)
1217 {
1218 	if (btrfs_is_testing(root->fs_info))
1219 		return alloc_test_extent_buffer(root->fs_info, bytenr,
1220 				root->nodesize);
1221 	return alloc_extent_buffer(root->fs_info, bytenr);
1222 }
1223 
1224 
btrfs_write_tree_block(struct extent_buffer * buf)1225 int btrfs_write_tree_block(struct extent_buffer *buf)
1226 {
1227 	return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1228 					buf->start + buf->len - 1);
1229 }
1230 
btrfs_wait_tree_block_writeback(struct extent_buffer * buf)1231 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1232 {
1233 	return filemap_fdatawait_range(buf->pages[0]->mapping,
1234 				       buf->start, buf->start + buf->len - 1);
1235 }
1236 
read_tree_block(struct btrfs_root * root,u64 bytenr,u64 parent_transid)1237 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1238 				      u64 parent_transid)
1239 {
1240 	struct extent_buffer *buf = NULL;
1241 	int ret;
1242 
1243 	buf = btrfs_find_create_tree_block(root, bytenr);
1244 	if (IS_ERR(buf))
1245 		return buf;
1246 
1247 	ret = btree_read_extent_buffer_pages(root, buf, parent_transid);
1248 	if (ret) {
1249 		free_extent_buffer(buf);
1250 		return ERR_PTR(ret);
1251 	}
1252 	return buf;
1253 
1254 }
1255 
clean_tree_block(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info,struct extent_buffer * buf)1256 void clean_tree_block(struct btrfs_trans_handle *trans,
1257 		      struct btrfs_fs_info *fs_info,
1258 		      struct extent_buffer *buf)
1259 {
1260 	if (btrfs_header_generation(buf) ==
1261 	    fs_info->running_transaction->transid) {
1262 		btrfs_assert_tree_locked(buf);
1263 
1264 		if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1265 			__percpu_counter_add(&fs_info->dirty_metadata_bytes,
1266 					     -buf->len,
1267 					     fs_info->dirty_metadata_batch);
1268 			/* ugh, clear_extent_buffer_dirty needs to lock the page */
1269 			btrfs_set_lock_blocking(buf);
1270 			clear_extent_buffer_dirty(buf);
1271 		}
1272 	}
1273 }
1274 
btrfs_alloc_subvolume_writers(void)1275 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1276 {
1277 	struct btrfs_subvolume_writers *writers;
1278 	int ret;
1279 
1280 	writers = kmalloc(sizeof(*writers), GFP_NOFS);
1281 	if (!writers)
1282 		return ERR_PTR(-ENOMEM);
1283 
1284 	ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1285 	if (ret < 0) {
1286 		kfree(writers);
1287 		return ERR_PTR(ret);
1288 	}
1289 
1290 	init_waitqueue_head(&writers->wait);
1291 	return writers;
1292 }
1293 
1294 static void
btrfs_free_subvolume_writers(struct btrfs_subvolume_writers * writers)1295 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1296 {
1297 	percpu_counter_destroy(&writers->counter);
1298 	kfree(writers);
1299 }
1300 
__setup_root(u32 nodesize,u32 sectorsize,u32 stripesize,struct btrfs_root * root,struct btrfs_fs_info * fs_info,u64 objectid)1301 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1302 			 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1303 			 u64 objectid)
1304 {
1305 	bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1306 	root->node = NULL;
1307 	root->commit_root = NULL;
1308 	root->sectorsize = sectorsize;
1309 	root->nodesize = nodesize;
1310 	root->stripesize = stripesize;
1311 	root->state = 0;
1312 	root->orphan_cleanup_state = 0;
1313 
1314 	root->objectid = objectid;
1315 	root->last_trans = 0;
1316 	root->highest_objectid = 0;
1317 	root->nr_delalloc_inodes = 0;
1318 	root->nr_ordered_extents = 0;
1319 	root->name = NULL;
1320 	root->inode_tree = RB_ROOT;
1321 	INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1322 	root->block_rsv = NULL;
1323 	root->orphan_block_rsv = NULL;
1324 
1325 	INIT_LIST_HEAD(&root->dirty_list);
1326 	INIT_LIST_HEAD(&root->root_list);
1327 	INIT_LIST_HEAD(&root->delalloc_inodes);
1328 	INIT_LIST_HEAD(&root->delalloc_root);
1329 	INIT_LIST_HEAD(&root->ordered_extents);
1330 	INIT_LIST_HEAD(&root->ordered_root);
1331 	INIT_LIST_HEAD(&root->logged_list[0]);
1332 	INIT_LIST_HEAD(&root->logged_list[1]);
1333 	spin_lock_init(&root->orphan_lock);
1334 	spin_lock_init(&root->inode_lock);
1335 	spin_lock_init(&root->delalloc_lock);
1336 	spin_lock_init(&root->ordered_extent_lock);
1337 	spin_lock_init(&root->accounting_lock);
1338 	spin_lock_init(&root->log_extents_lock[0]);
1339 	spin_lock_init(&root->log_extents_lock[1]);
1340 	mutex_init(&root->objectid_mutex);
1341 	mutex_init(&root->log_mutex);
1342 	mutex_init(&root->ordered_extent_mutex);
1343 	mutex_init(&root->delalloc_mutex);
1344 	init_waitqueue_head(&root->log_writer_wait);
1345 	init_waitqueue_head(&root->log_commit_wait[0]);
1346 	init_waitqueue_head(&root->log_commit_wait[1]);
1347 	INIT_LIST_HEAD(&root->log_ctxs[0]);
1348 	INIT_LIST_HEAD(&root->log_ctxs[1]);
1349 	atomic_set(&root->log_commit[0], 0);
1350 	atomic_set(&root->log_commit[1], 0);
1351 	atomic_set(&root->log_writers, 0);
1352 	atomic_set(&root->log_batch, 0);
1353 	atomic_set(&root->orphan_inodes, 0);
1354 	atomic_set(&root->refs, 1);
1355 	atomic_set(&root->will_be_snapshoted, 0);
1356 	atomic_set(&root->qgroup_meta_rsv, 0);
1357 	root->log_transid = 0;
1358 	root->log_transid_committed = -1;
1359 	root->last_log_commit = 0;
1360 	if (!dummy)
1361 		extent_io_tree_init(&root->dirty_log_pages,
1362 				     fs_info->btree_inode->i_mapping);
1363 
1364 	memset(&root->root_key, 0, sizeof(root->root_key));
1365 	memset(&root->root_item, 0, sizeof(root->root_item));
1366 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1367 	if (!dummy)
1368 		root->defrag_trans_start = fs_info->generation;
1369 	else
1370 		root->defrag_trans_start = 0;
1371 	root->root_key.objectid = objectid;
1372 	root->anon_dev = 0;
1373 
1374 	spin_lock_init(&root->root_item_lock);
1375 }
1376 
btrfs_alloc_root(struct btrfs_fs_info * fs_info,gfp_t flags)1377 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1378 		gfp_t flags)
1379 {
1380 	struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1381 	if (root)
1382 		root->fs_info = fs_info;
1383 	return root;
1384 }
1385 
1386 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1387 /* Should only be used by the testing infrastructure */
btrfs_alloc_dummy_root(struct btrfs_fs_info * fs_info,u32 sectorsize,u32 nodesize)1388 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1389 					  u32 sectorsize, u32 nodesize)
1390 {
1391 	struct btrfs_root *root;
1392 
1393 	if (!fs_info)
1394 		return ERR_PTR(-EINVAL);
1395 
1396 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1397 	if (!root)
1398 		return ERR_PTR(-ENOMEM);
1399 	/* We don't use the stripesize in selftest, set it as sectorsize */
1400 	__setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1401 			BTRFS_ROOT_TREE_OBJECTID);
1402 	root->alloc_bytenr = 0;
1403 
1404 	return root;
1405 }
1406 #endif
1407 
btrfs_create_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info,u64 objectid)1408 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1409 				     struct btrfs_fs_info *fs_info,
1410 				     u64 objectid)
1411 {
1412 	struct extent_buffer *leaf;
1413 	struct btrfs_root *tree_root = fs_info->tree_root;
1414 	struct btrfs_root *root;
1415 	struct btrfs_key key;
1416 	int ret = 0;
1417 	uuid_le uuid;
1418 
1419 	root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1420 	if (!root)
1421 		return ERR_PTR(-ENOMEM);
1422 
1423 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1424 		tree_root->stripesize, root, fs_info, objectid);
1425 	root->root_key.objectid = objectid;
1426 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1427 	root->root_key.offset = 0;
1428 
1429 	leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1430 	if (IS_ERR(leaf)) {
1431 		ret = PTR_ERR(leaf);
1432 		leaf = NULL;
1433 		goto fail;
1434 	}
1435 
1436 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1437 	btrfs_set_header_bytenr(leaf, leaf->start);
1438 	btrfs_set_header_generation(leaf, trans->transid);
1439 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1440 	btrfs_set_header_owner(leaf, objectid);
1441 	root->node = leaf;
1442 
1443 	write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1444 			    BTRFS_FSID_SIZE);
1445 	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1446 			    btrfs_header_chunk_tree_uuid(leaf),
1447 			    BTRFS_UUID_SIZE);
1448 	btrfs_mark_buffer_dirty(leaf);
1449 
1450 	root->commit_root = btrfs_root_node(root);
1451 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1452 
1453 	root->root_item.flags = 0;
1454 	root->root_item.byte_limit = 0;
1455 	btrfs_set_root_bytenr(&root->root_item, leaf->start);
1456 	btrfs_set_root_generation(&root->root_item, trans->transid);
1457 	btrfs_set_root_level(&root->root_item, 0);
1458 	btrfs_set_root_refs(&root->root_item, 1);
1459 	btrfs_set_root_used(&root->root_item, leaf->len);
1460 	btrfs_set_root_last_snapshot(&root->root_item, 0);
1461 	btrfs_set_root_dirid(&root->root_item, 0);
1462 	uuid_le_gen(&uuid);
1463 	memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1464 	root->root_item.drop_level = 0;
1465 
1466 	key.objectid = objectid;
1467 	key.type = BTRFS_ROOT_ITEM_KEY;
1468 	key.offset = 0;
1469 	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1470 	if (ret)
1471 		goto fail;
1472 
1473 	btrfs_tree_unlock(leaf);
1474 
1475 	return root;
1476 
1477 fail:
1478 	if (leaf) {
1479 		btrfs_tree_unlock(leaf);
1480 		free_extent_buffer(root->commit_root);
1481 		free_extent_buffer(leaf);
1482 	}
1483 	kfree(root);
1484 
1485 	return ERR_PTR(ret);
1486 }
1487 
alloc_log_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)1488 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1489 					 struct btrfs_fs_info *fs_info)
1490 {
1491 	struct btrfs_root *root;
1492 	struct btrfs_root *tree_root = fs_info->tree_root;
1493 	struct extent_buffer *leaf;
1494 
1495 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1496 	if (!root)
1497 		return ERR_PTR(-ENOMEM);
1498 
1499 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1500 		     tree_root->stripesize, root, fs_info,
1501 		     BTRFS_TREE_LOG_OBJECTID);
1502 
1503 	root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1504 	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1505 	root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1506 
1507 	/*
1508 	 * DON'T set REF_COWS for log trees
1509 	 *
1510 	 * log trees do not get reference counted because they go away
1511 	 * before a real commit is actually done.  They do store pointers
1512 	 * to file data extents, and those reference counts still get
1513 	 * updated (along with back refs to the log tree).
1514 	 */
1515 
1516 	leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1517 			NULL, 0, 0, 0);
1518 	if (IS_ERR(leaf)) {
1519 		kfree(root);
1520 		return ERR_CAST(leaf);
1521 	}
1522 
1523 	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1524 	btrfs_set_header_bytenr(leaf, leaf->start);
1525 	btrfs_set_header_generation(leaf, trans->transid);
1526 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1527 	btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1528 	root->node = leaf;
1529 
1530 	write_extent_buffer(root->node, root->fs_info->fsid,
1531 			    btrfs_header_fsid(), BTRFS_FSID_SIZE);
1532 	btrfs_mark_buffer_dirty(root->node);
1533 	btrfs_tree_unlock(root->node);
1534 	return root;
1535 }
1536 
btrfs_init_log_root_tree(struct btrfs_trans_handle * trans,struct btrfs_fs_info * fs_info)1537 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1538 			     struct btrfs_fs_info *fs_info)
1539 {
1540 	struct btrfs_root *log_root;
1541 
1542 	log_root = alloc_log_tree(trans, fs_info);
1543 	if (IS_ERR(log_root))
1544 		return PTR_ERR(log_root);
1545 	WARN_ON(fs_info->log_root_tree);
1546 	fs_info->log_root_tree = log_root;
1547 	return 0;
1548 }
1549 
btrfs_add_log_tree(struct btrfs_trans_handle * trans,struct btrfs_root * root)1550 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1551 		       struct btrfs_root *root)
1552 {
1553 	struct btrfs_root *log_root;
1554 	struct btrfs_inode_item *inode_item;
1555 
1556 	log_root = alloc_log_tree(trans, root->fs_info);
1557 	if (IS_ERR(log_root))
1558 		return PTR_ERR(log_root);
1559 
1560 	log_root->last_trans = trans->transid;
1561 	log_root->root_key.offset = root->root_key.objectid;
1562 
1563 	inode_item = &log_root->root_item.inode;
1564 	btrfs_set_stack_inode_generation(inode_item, 1);
1565 	btrfs_set_stack_inode_size(inode_item, 3);
1566 	btrfs_set_stack_inode_nlink(inode_item, 1);
1567 	btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1568 	btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1569 
1570 	btrfs_set_root_node(&log_root->root_item, log_root->node);
1571 
1572 	WARN_ON(root->log_root);
1573 	root->log_root = log_root;
1574 	root->log_transid = 0;
1575 	root->log_transid_committed = -1;
1576 	root->last_log_commit = 0;
1577 	return 0;
1578 }
1579 
btrfs_read_tree_root(struct btrfs_root * tree_root,struct btrfs_key * key)1580 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1581 					       struct btrfs_key *key)
1582 {
1583 	struct btrfs_root *root;
1584 	struct btrfs_fs_info *fs_info = tree_root->fs_info;
1585 	struct btrfs_path *path;
1586 	u64 generation;
1587 	int ret;
1588 
1589 	path = btrfs_alloc_path();
1590 	if (!path)
1591 		return ERR_PTR(-ENOMEM);
1592 
1593 	root = btrfs_alloc_root(fs_info, GFP_NOFS);
1594 	if (!root) {
1595 		ret = -ENOMEM;
1596 		goto alloc_fail;
1597 	}
1598 
1599 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
1600 		tree_root->stripesize, root, fs_info, key->objectid);
1601 
1602 	ret = btrfs_find_root(tree_root, key, path,
1603 			      &root->root_item, &root->root_key);
1604 	if (ret) {
1605 		if (ret > 0)
1606 			ret = -ENOENT;
1607 		goto find_fail;
1608 	}
1609 
1610 	generation = btrfs_root_generation(&root->root_item);
1611 	root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1612 				     generation);
1613 	if (IS_ERR(root->node)) {
1614 		ret = PTR_ERR(root->node);
1615 		goto find_fail;
1616 	} else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1617 		ret = -EIO;
1618 		free_extent_buffer(root->node);
1619 		goto find_fail;
1620 	}
1621 	root->commit_root = btrfs_root_node(root);
1622 out:
1623 	btrfs_free_path(path);
1624 	return root;
1625 
1626 find_fail:
1627 	kfree(root);
1628 alloc_fail:
1629 	root = ERR_PTR(ret);
1630 	goto out;
1631 }
1632 
btrfs_read_fs_root(struct btrfs_root * tree_root,struct btrfs_key * location)1633 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1634 				      struct btrfs_key *location)
1635 {
1636 	struct btrfs_root *root;
1637 
1638 	root = btrfs_read_tree_root(tree_root, location);
1639 	if (IS_ERR(root))
1640 		return root;
1641 
1642 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1643 		set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1644 		btrfs_check_and_init_root_item(&root->root_item);
1645 	}
1646 
1647 	return root;
1648 }
1649 
btrfs_init_fs_root(struct btrfs_root * root)1650 int btrfs_init_fs_root(struct btrfs_root *root)
1651 {
1652 	int ret;
1653 	struct btrfs_subvolume_writers *writers;
1654 
1655 	root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1656 	root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1657 					GFP_NOFS);
1658 	if (!root->free_ino_pinned || !root->free_ino_ctl) {
1659 		ret = -ENOMEM;
1660 		goto fail;
1661 	}
1662 
1663 	writers = btrfs_alloc_subvolume_writers();
1664 	if (IS_ERR(writers)) {
1665 		ret = PTR_ERR(writers);
1666 		goto fail;
1667 	}
1668 	root->subv_writers = writers;
1669 
1670 	btrfs_init_free_ino_ctl(root);
1671 	spin_lock_init(&root->ino_cache_lock);
1672 	init_waitqueue_head(&root->ino_cache_wait);
1673 
1674 	ret = get_anon_bdev(&root->anon_dev);
1675 	if (ret)
1676 		goto fail;
1677 
1678 	mutex_lock(&root->objectid_mutex);
1679 	ret = btrfs_find_highest_objectid(root,
1680 					&root->highest_objectid);
1681 	if (ret) {
1682 		mutex_unlock(&root->objectid_mutex);
1683 		goto fail;
1684 	}
1685 
1686 	ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1687 
1688 	mutex_unlock(&root->objectid_mutex);
1689 
1690 	return 0;
1691 fail:
1692 	/* the caller is responsible to call free_fs_root */
1693 	return ret;
1694 }
1695 
btrfs_lookup_fs_root(struct btrfs_fs_info * fs_info,u64 root_id)1696 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1697 					u64 root_id)
1698 {
1699 	struct btrfs_root *root;
1700 
1701 	spin_lock(&fs_info->fs_roots_radix_lock);
1702 	root = radix_tree_lookup(&fs_info->fs_roots_radix,
1703 				 (unsigned long)root_id);
1704 	spin_unlock(&fs_info->fs_roots_radix_lock);
1705 	return root;
1706 }
1707 
btrfs_insert_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_root * root)1708 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1709 			 struct btrfs_root *root)
1710 {
1711 	int ret;
1712 
1713 	ret = radix_tree_preload(GFP_NOFS);
1714 	if (ret)
1715 		return ret;
1716 
1717 	spin_lock(&fs_info->fs_roots_radix_lock);
1718 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
1719 				(unsigned long)root->root_key.objectid,
1720 				root);
1721 	if (ret == 0)
1722 		set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1723 	spin_unlock(&fs_info->fs_roots_radix_lock);
1724 	radix_tree_preload_end();
1725 
1726 	return ret;
1727 }
1728 
btrfs_get_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_key * location,bool check_ref)1729 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1730 				     struct btrfs_key *location,
1731 				     bool check_ref)
1732 {
1733 	struct btrfs_root *root;
1734 	struct btrfs_path *path;
1735 	struct btrfs_key key;
1736 	int ret;
1737 
1738 	if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1739 		return fs_info->tree_root;
1740 	if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1741 		return fs_info->extent_root;
1742 	if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1743 		return fs_info->chunk_root;
1744 	if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1745 		return fs_info->dev_root;
1746 	if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1747 		return fs_info->csum_root;
1748 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1749 		return fs_info->quota_root ? fs_info->quota_root :
1750 					     ERR_PTR(-ENOENT);
1751 	if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1752 		return fs_info->uuid_root ? fs_info->uuid_root :
1753 					    ERR_PTR(-ENOENT);
1754 	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1755 		return fs_info->free_space_root ? fs_info->free_space_root :
1756 						  ERR_PTR(-ENOENT);
1757 again:
1758 	root = btrfs_lookup_fs_root(fs_info, location->objectid);
1759 	if (root) {
1760 		if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1761 			return ERR_PTR(-ENOENT);
1762 		return root;
1763 	}
1764 
1765 	root = btrfs_read_fs_root(fs_info->tree_root, location);
1766 	if (IS_ERR(root))
1767 		return root;
1768 
1769 	if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1770 		ret = -ENOENT;
1771 		goto fail;
1772 	}
1773 
1774 	ret = btrfs_init_fs_root(root);
1775 	if (ret)
1776 		goto fail;
1777 
1778 	path = btrfs_alloc_path();
1779 	if (!path) {
1780 		ret = -ENOMEM;
1781 		goto fail;
1782 	}
1783 	key.objectid = BTRFS_ORPHAN_OBJECTID;
1784 	key.type = BTRFS_ORPHAN_ITEM_KEY;
1785 	key.offset = location->objectid;
1786 
1787 	ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1788 	btrfs_free_path(path);
1789 	if (ret < 0)
1790 		goto fail;
1791 	if (ret == 0)
1792 		set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1793 
1794 	ret = btrfs_insert_fs_root(fs_info, root);
1795 	if (ret) {
1796 		if (ret == -EEXIST) {
1797 			free_fs_root(root);
1798 			goto again;
1799 		}
1800 		goto fail;
1801 	}
1802 	return root;
1803 fail:
1804 	free_fs_root(root);
1805 	return ERR_PTR(ret);
1806 }
1807 
btrfs_congested_fn(void * congested_data,int bdi_bits)1808 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1809 {
1810 	struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1811 	int ret = 0;
1812 	struct btrfs_device *device;
1813 	struct backing_dev_info *bdi;
1814 
1815 	rcu_read_lock();
1816 	list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1817 		if (!device->bdev)
1818 			continue;
1819 		bdi = blk_get_backing_dev_info(device->bdev);
1820 		if (bdi_congested(bdi, bdi_bits)) {
1821 			ret = 1;
1822 			break;
1823 		}
1824 	}
1825 	rcu_read_unlock();
1826 	return ret;
1827 }
1828 
setup_bdi(struct btrfs_fs_info * info,struct backing_dev_info * bdi)1829 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1830 {
1831 	int err;
1832 
1833 	err = bdi_setup_and_register(bdi, "btrfs");
1834 	if (err)
1835 		return err;
1836 
1837 	bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1838 	bdi->congested_fn	= btrfs_congested_fn;
1839 	bdi->congested_data	= info;
1840 	bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1841 	return 0;
1842 }
1843 
1844 /*
1845  * called by the kthread helper functions to finally call the bio end_io
1846  * functions.  This is where read checksum verification actually happens
1847  */
end_workqueue_fn(struct btrfs_work * work)1848 static void end_workqueue_fn(struct btrfs_work *work)
1849 {
1850 	struct bio *bio;
1851 	struct btrfs_end_io_wq *end_io_wq;
1852 
1853 	end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1854 	bio = end_io_wq->bio;
1855 
1856 	bio->bi_error = end_io_wq->error;
1857 	bio->bi_private = end_io_wq->private;
1858 	bio->bi_end_io = end_io_wq->end_io;
1859 	kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1860 	bio_endio(bio);
1861 }
1862 
cleaner_kthread(void * arg)1863 static int cleaner_kthread(void *arg)
1864 {
1865 	struct btrfs_root *root = arg;
1866 	int again;
1867 	struct btrfs_trans_handle *trans;
1868 
1869 	do {
1870 		again = 0;
1871 
1872 		/* Make the cleaner go to sleep early. */
1873 		if (btrfs_need_cleaner_sleep(root))
1874 			goto sleep;
1875 
1876 		/*
1877 		 * Do not do anything if we might cause open_ctree() to block
1878 		 * before we have finished mounting the filesystem.
1879 		 */
1880 		if (!test_bit(BTRFS_FS_OPEN, &root->fs_info->flags))
1881 			goto sleep;
1882 
1883 		if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1884 			goto sleep;
1885 
1886 		/*
1887 		 * Avoid the problem that we change the status of the fs
1888 		 * during the above check and trylock.
1889 		 */
1890 		if (btrfs_need_cleaner_sleep(root)) {
1891 			mutex_unlock(&root->fs_info->cleaner_mutex);
1892 			goto sleep;
1893 		}
1894 
1895 		mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1896 		btrfs_run_delayed_iputs(root);
1897 		mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1898 
1899 		again = btrfs_clean_one_deleted_snapshot(root);
1900 		mutex_unlock(&root->fs_info->cleaner_mutex);
1901 
1902 		/*
1903 		 * The defragger has dealt with the R/O remount and umount,
1904 		 * needn't do anything special here.
1905 		 */
1906 		btrfs_run_defrag_inodes(root->fs_info);
1907 
1908 		/*
1909 		 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1910 		 * with relocation (btrfs_relocate_chunk) and relocation
1911 		 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1912 		 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1913 		 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1914 		 * unused block groups.
1915 		 */
1916 		btrfs_delete_unused_bgs(root->fs_info);
1917 sleep:
1918 		if (!again) {
1919 			set_current_state(TASK_INTERRUPTIBLE);
1920 			if (!kthread_should_stop())
1921 				schedule();
1922 			__set_current_state(TASK_RUNNING);
1923 		}
1924 	} while (!kthread_should_stop());
1925 
1926 	/*
1927 	 * Transaction kthread is stopped before us and wakes us up.
1928 	 * However we might have started a new transaction and COWed some
1929 	 * tree blocks when deleting unused block groups for example. So
1930 	 * make sure we commit the transaction we started to have a clean
1931 	 * shutdown when evicting the btree inode - if it has dirty pages
1932 	 * when we do the final iput() on it, eviction will trigger a
1933 	 * writeback for it which will fail with null pointer dereferences
1934 	 * since work queues and other resources were already released and
1935 	 * destroyed by the time the iput/eviction/writeback is made.
1936 	 */
1937 	trans = btrfs_attach_transaction(root);
1938 	if (IS_ERR(trans)) {
1939 		if (PTR_ERR(trans) != -ENOENT)
1940 			btrfs_err(root->fs_info,
1941 				  "cleaner transaction attach returned %ld",
1942 				  PTR_ERR(trans));
1943 	} else {
1944 		int ret;
1945 
1946 		ret = btrfs_commit_transaction(trans, root);
1947 		if (ret)
1948 			btrfs_err(root->fs_info,
1949 				  "cleaner open transaction commit returned %d",
1950 				  ret);
1951 	}
1952 
1953 	return 0;
1954 }
1955 
transaction_kthread(void * arg)1956 static int transaction_kthread(void *arg)
1957 {
1958 	struct btrfs_root *root = arg;
1959 	struct btrfs_trans_handle *trans;
1960 	struct btrfs_transaction *cur;
1961 	u64 transid;
1962 	unsigned long now;
1963 	unsigned long delay;
1964 	bool cannot_commit;
1965 
1966 	do {
1967 		cannot_commit = false;
1968 		delay = HZ * root->fs_info->commit_interval;
1969 		mutex_lock(&root->fs_info->transaction_kthread_mutex);
1970 
1971 		spin_lock(&root->fs_info->trans_lock);
1972 		cur = root->fs_info->running_transaction;
1973 		if (!cur) {
1974 			spin_unlock(&root->fs_info->trans_lock);
1975 			goto sleep;
1976 		}
1977 
1978 		now = get_seconds();
1979 		if (cur->state < TRANS_STATE_BLOCKED &&
1980 		    (now < cur->start_time ||
1981 		     now - cur->start_time < root->fs_info->commit_interval)) {
1982 			spin_unlock(&root->fs_info->trans_lock);
1983 			delay = HZ * 5;
1984 			goto sleep;
1985 		}
1986 		transid = cur->transid;
1987 		spin_unlock(&root->fs_info->trans_lock);
1988 
1989 		/* If the file system is aborted, this will always fail. */
1990 		trans = btrfs_attach_transaction(root);
1991 		if (IS_ERR(trans)) {
1992 			if (PTR_ERR(trans) != -ENOENT)
1993 				cannot_commit = true;
1994 			goto sleep;
1995 		}
1996 		if (transid == trans->transid) {
1997 			btrfs_commit_transaction(trans, root);
1998 		} else {
1999 			btrfs_end_transaction(trans, root);
2000 		}
2001 sleep:
2002 		wake_up_process(root->fs_info->cleaner_kthread);
2003 		mutex_unlock(&root->fs_info->transaction_kthread_mutex);
2004 
2005 		if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
2006 				      &root->fs_info->fs_state)))
2007 			btrfs_cleanup_transaction(root);
2008 		set_current_state(TASK_INTERRUPTIBLE);
2009 		if (!kthread_should_stop() &&
2010 				(!btrfs_transaction_blocked(root->fs_info) ||
2011 				 cannot_commit))
2012 			schedule_timeout(delay);
2013 		__set_current_state(TASK_RUNNING);
2014 	} while (!kthread_should_stop());
2015 	return 0;
2016 }
2017 
2018 /*
2019  * this will find the highest generation in the array of
2020  * root backups.  The index of the highest array is returned,
2021  * or -1 if we can't find anything.
2022  *
2023  * We check to make sure the array is valid by comparing the
2024  * generation of the latest  root in the array with the generation
2025  * in the super block.  If they don't match we pitch it.
2026  */
find_newest_super_backup(struct btrfs_fs_info * info,u64 newest_gen)2027 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
2028 {
2029 	u64 cur;
2030 	int newest_index = -1;
2031 	struct btrfs_root_backup *root_backup;
2032 	int i;
2033 
2034 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2035 		root_backup = info->super_copy->super_roots + i;
2036 		cur = btrfs_backup_tree_root_gen(root_backup);
2037 		if (cur == newest_gen)
2038 			newest_index = i;
2039 	}
2040 
2041 	/* check to see if we actually wrapped around */
2042 	if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2043 		root_backup = info->super_copy->super_roots;
2044 		cur = btrfs_backup_tree_root_gen(root_backup);
2045 		if (cur == newest_gen)
2046 			newest_index = 0;
2047 	}
2048 	return newest_index;
2049 }
2050 
2051 
2052 /*
2053  * find the oldest backup so we know where to store new entries
2054  * in the backup array.  This will set the backup_root_index
2055  * field in the fs_info struct
2056  */
find_oldest_super_backup(struct btrfs_fs_info * info,u64 newest_gen)2057 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2058 				     u64 newest_gen)
2059 {
2060 	int newest_index = -1;
2061 
2062 	newest_index = find_newest_super_backup(info, newest_gen);
2063 	/* if there was garbage in there, just move along */
2064 	if (newest_index == -1) {
2065 		info->backup_root_index = 0;
2066 	} else {
2067 		info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2068 	}
2069 }
2070 
2071 /*
2072  * copy all the root pointers into the super backup array.
2073  * this will bump the backup pointer by one when it is
2074  * done
2075  */
backup_super_roots(struct btrfs_fs_info * info)2076 static void backup_super_roots(struct btrfs_fs_info *info)
2077 {
2078 	int next_backup;
2079 	struct btrfs_root_backup *root_backup;
2080 	int last_backup;
2081 
2082 	next_backup = info->backup_root_index;
2083 	last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2084 		BTRFS_NUM_BACKUP_ROOTS;
2085 
2086 	/*
2087 	 * just overwrite the last backup if we're at the same generation
2088 	 * this happens only at umount
2089 	 */
2090 	root_backup = info->super_for_commit->super_roots + last_backup;
2091 	if (btrfs_backup_tree_root_gen(root_backup) ==
2092 	    btrfs_header_generation(info->tree_root->node))
2093 		next_backup = last_backup;
2094 
2095 	root_backup = info->super_for_commit->super_roots + next_backup;
2096 
2097 	/*
2098 	 * make sure all of our padding and empty slots get zero filled
2099 	 * regardless of which ones we use today
2100 	 */
2101 	memset(root_backup, 0, sizeof(*root_backup));
2102 
2103 	info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2104 
2105 	btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2106 	btrfs_set_backup_tree_root_gen(root_backup,
2107 			       btrfs_header_generation(info->tree_root->node));
2108 
2109 	btrfs_set_backup_tree_root_level(root_backup,
2110 			       btrfs_header_level(info->tree_root->node));
2111 
2112 	btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2113 	btrfs_set_backup_chunk_root_gen(root_backup,
2114 			       btrfs_header_generation(info->chunk_root->node));
2115 	btrfs_set_backup_chunk_root_level(root_backup,
2116 			       btrfs_header_level(info->chunk_root->node));
2117 
2118 	btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2119 	btrfs_set_backup_extent_root_gen(root_backup,
2120 			       btrfs_header_generation(info->extent_root->node));
2121 	btrfs_set_backup_extent_root_level(root_backup,
2122 			       btrfs_header_level(info->extent_root->node));
2123 
2124 	/*
2125 	 * we might commit during log recovery, which happens before we set
2126 	 * the fs_root.  Make sure it is valid before we fill it in.
2127 	 */
2128 	if (info->fs_root && info->fs_root->node) {
2129 		btrfs_set_backup_fs_root(root_backup,
2130 					 info->fs_root->node->start);
2131 		btrfs_set_backup_fs_root_gen(root_backup,
2132 			       btrfs_header_generation(info->fs_root->node));
2133 		btrfs_set_backup_fs_root_level(root_backup,
2134 			       btrfs_header_level(info->fs_root->node));
2135 	}
2136 
2137 	btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2138 	btrfs_set_backup_dev_root_gen(root_backup,
2139 			       btrfs_header_generation(info->dev_root->node));
2140 	btrfs_set_backup_dev_root_level(root_backup,
2141 				       btrfs_header_level(info->dev_root->node));
2142 
2143 	btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2144 	btrfs_set_backup_csum_root_gen(root_backup,
2145 			       btrfs_header_generation(info->csum_root->node));
2146 	btrfs_set_backup_csum_root_level(root_backup,
2147 			       btrfs_header_level(info->csum_root->node));
2148 
2149 	btrfs_set_backup_total_bytes(root_backup,
2150 			     btrfs_super_total_bytes(info->super_copy));
2151 	btrfs_set_backup_bytes_used(root_backup,
2152 			     btrfs_super_bytes_used(info->super_copy));
2153 	btrfs_set_backup_num_devices(root_backup,
2154 			     btrfs_super_num_devices(info->super_copy));
2155 
2156 	/*
2157 	 * if we don't copy this out to the super_copy, it won't get remembered
2158 	 * for the next commit
2159 	 */
2160 	memcpy(&info->super_copy->super_roots,
2161 	       &info->super_for_commit->super_roots,
2162 	       sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2163 }
2164 
2165 /*
2166  * this copies info out of the root backup array and back into
2167  * the in-memory super block.  It is meant to help iterate through
2168  * the array, so you send it the number of backups you've already
2169  * tried and the last backup index you used.
2170  *
2171  * this returns -1 when it has tried all the backups
2172  */
next_root_backup(struct btrfs_fs_info * info,struct btrfs_super_block * super,int * num_backups_tried,int * backup_index)2173 static noinline int next_root_backup(struct btrfs_fs_info *info,
2174 				     struct btrfs_super_block *super,
2175 				     int *num_backups_tried, int *backup_index)
2176 {
2177 	struct btrfs_root_backup *root_backup;
2178 	int newest = *backup_index;
2179 
2180 	if (*num_backups_tried == 0) {
2181 		u64 gen = btrfs_super_generation(super);
2182 
2183 		newest = find_newest_super_backup(info, gen);
2184 		if (newest == -1)
2185 			return -1;
2186 
2187 		*backup_index = newest;
2188 		*num_backups_tried = 1;
2189 	} else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2190 		/* we've tried all the backups, all done */
2191 		return -1;
2192 	} else {
2193 		/* jump to the next oldest backup */
2194 		newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2195 			BTRFS_NUM_BACKUP_ROOTS;
2196 		*backup_index = newest;
2197 		*num_backups_tried += 1;
2198 	}
2199 	root_backup = super->super_roots + newest;
2200 
2201 	btrfs_set_super_generation(super,
2202 				   btrfs_backup_tree_root_gen(root_backup));
2203 	btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2204 	btrfs_set_super_root_level(super,
2205 				   btrfs_backup_tree_root_level(root_backup));
2206 	btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2207 
2208 	/*
2209 	 * fixme: the total bytes and num_devices need to match or we should
2210 	 * need a fsck
2211 	 */
2212 	btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2213 	btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2214 	return 0;
2215 }
2216 
2217 /* helper to cleanup workers */
btrfs_stop_all_workers(struct btrfs_fs_info * fs_info)2218 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2219 {
2220 	btrfs_destroy_workqueue(fs_info->fixup_workers);
2221 	btrfs_destroy_workqueue(fs_info->delalloc_workers);
2222 	btrfs_destroy_workqueue(fs_info->workers);
2223 	btrfs_destroy_workqueue(fs_info->endio_workers);
2224 	btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2225 	btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2226 	btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2227 	btrfs_destroy_workqueue(fs_info->rmw_workers);
2228 	btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2229 	btrfs_destroy_workqueue(fs_info->endio_write_workers);
2230 	btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2231 	btrfs_destroy_workqueue(fs_info->submit_workers);
2232 	btrfs_destroy_workqueue(fs_info->delayed_workers);
2233 	btrfs_destroy_workqueue(fs_info->caching_workers);
2234 	btrfs_destroy_workqueue(fs_info->readahead_workers);
2235 	btrfs_destroy_workqueue(fs_info->flush_workers);
2236 	btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2237 	btrfs_destroy_workqueue(fs_info->extent_workers);
2238 }
2239 
free_root_extent_buffers(struct btrfs_root * root)2240 static void free_root_extent_buffers(struct btrfs_root *root)
2241 {
2242 	if (root) {
2243 		free_extent_buffer(root->node);
2244 		free_extent_buffer(root->commit_root);
2245 		root->node = NULL;
2246 		root->commit_root = NULL;
2247 	}
2248 }
2249 
2250 /* helper to cleanup tree roots */
free_root_pointers(struct btrfs_fs_info * info,int chunk_root)2251 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2252 {
2253 	free_root_extent_buffers(info->tree_root);
2254 
2255 	free_root_extent_buffers(info->dev_root);
2256 	free_root_extent_buffers(info->extent_root);
2257 	free_root_extent_buffers(info->csum_root);
2258 	free_root_extent_buffers(info->quota_root);
2259 	free_root_extent_buffers(info->uuid_root);
2260 	if (chunk_root)
2261 		free_root_extent_buffers(info->chunk_root);
2262 	free_root_extent_buffers(info->free_space_root);
2263 }
2264 
btrfs_free_fs_roots(struct btrfs_fs_info * fs_info)2265 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2266 {
2267 	int ret;
2268 	struct btrfs_root *gang[8];
2269 	int i;
2270 
2271 	while (!list_empty(&fs_info->dead_roots)) {
2272 		gang[0] = list_entry(fs_info->dead_roots.next,
2273 				     struct btrfs_root, root_list);
2274 		list_del(&gang[0]->root_list);
2275 
2276 		if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2277 			btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2278 		} else {
2279 			free_extent_buffer(gang[0]->node);
2280 			free_extent_buffer(gang[0]->commit_root);
2281 			btrfs_put_fs_root(gang[0]);
2282 		}
2283 	}
2284 
2285 	while (1) {
2286 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2287 					     (void **)gang, 0,
2288 					     ARRAY_SIZE(gang));
2289 		if (!ret)
2290 			break;
2291 		for (i = 0; i < ret; i++)
2292 			btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2293 	}
2294 
2295 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2296 		btrfs_free_log_root_tree(NULL, fs_info);
2297 		btrfs_destroy_pinned_extent(fs_info->tree_root,
2298 					    fs_info->pinned_extents);
2299 	}
2300 }
2301 
btrfs_init_scrub(struct btrfs_fs_info * fs_info)2302 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2303 {
2304 	mutex_init(&fs_info->scrub_lock);
2305 	atomic_set(&fs_info->scrubs_running, 0);
2306 	atomic_set(&fs_info->scrub_pause_req, 0);
2307 	atomic_set(&fs_info->scrubs_paused, 0);
2308 	atomic_set(&fs_info->scrub_cancel_req, 0);
2309 	init_waitqueue_head(&fs_info->scrub_pause_wait);
2310 	fs_info->scrub_workers_refcnt = 0;
2311 }
2312 
btrfs_init_balance(struct btrfs_fs_info * fs_info)2313 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2314 {
2315 	spin_lock_init(&fs_info->balance_lock);
2316 	mutex_init(&fs_info->balance_mutex);
2317 	atomic_set(&fs_info->balance_running, 0);
2318 	atomic_set(&fs_info->balance_pause_req, 0);
2319 	atomic_set(&fs_info->balance_cancel_req, 0);
2320 	fs_info->balance_ctl = NULL;
2321 	init_waitqueue_head(&fs_info->balance_wait_q);
2322 }
2323 
btrfs_init_btree_inode(struct btrfs_fs_info * fs_info,struct btrfs_root * tree_root)2324 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2325 				   struct btrfs_root *tree_root)
2326 {
2327 	fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2328 	set_nlink(fs_info->btree_inode, 1);
2329 	/*
2330 	 * we set the i_size on the btree inode to the max possible int.
2331 	 * the real end of the address space is determined by all of
2332 	 * the devices in the system
2333 	 */
2334 	fs_info->btree_inode->i_size = OFFSET_MAX;
2335 	fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2336 
2337 	RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2338 	extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2339 			     fs_info->btree_inode->i_mapping);
2340 	BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2341 	extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2342 
2343 	BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2344 
2345 	BTRFS_I(fs_info->btree_inode)->root = tree_root;
2346 	memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2347 	       sizeof(struct btrfs_key));
2348 	set_bit(BTRFS_INODE_DUMMY,
2349 		&BTRFS_I(fs_info->btree_inode)->runtime_flags);
2350 	btrfs_insert_inode_hash(fs_info->btree_inode);
2351 }
2352 
btrfs_init_dev_replace_locks(struct btrfs_fs_info * fs_info)2353 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2354 {
2355 	fs_info->dev_replace.lock_owner = 0;
2356 	atomic_set(&fs_info->dev_replace.nesting_level, 0);
2357 	mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2358 	rwlock_init(&fs_info->dev_replace.lock);
2359 	atomic_set(&fs_info->dev_replace.read_locks, 0);
2360 	atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2361 	init_waitqueue_head(&fs_info->replace_wait);
2362 	init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2363 }
2364 
btrfs_init_qgroup(struct btrfs_fs_info * fs_info)2365 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2366 {
2367 	spin_lock_init(&fs_info->qgroup_lock);
2368 	mutex_init(&fs_info->qgroup_ioctl_lock);
2369 	fs_info->qgroup_tree = RB_ROOT;
2370 	fs_info->qgroup_op_tree = RB_ROOT;
2371 	INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2372 	fs_info->qgroup_seq = 1;
2373 	fs_info->qgroup_ulist = NULL;
2374 	fs_info->qgroup_rescan_running = false;
2375 	mutex_init(&fs_info->qgroup_rescan_lock);
2376 }
2377 
btrfs_init_workqueues(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices)2378 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2379 		struct btrfs_fs_devices *fs_devices)
2380 {
2381 	int max_active = fs_info->thread_pool_size;
2382 	unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2383 
2384 	fs_info->workers =
2385 		btrfs_alloc_workqueue(fs_info, "worker",
2386 				      flags | WQ_HIGHPRI, max_active, 16);
2387 
2388 	fs_info->delalloc_workers =
2389 		btrfs_alloc_workqueue(fs_info, "delalloc",
2390 				      flags, max_active, 2);
2391 
2392 	fs_info->flush_workers =
2393 		btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2394 				      flags, max_active, 0);
2395 
2396 	fs_info->caching_workers =
2397 		btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2398 
2399 	/*
2400 	 * a higher idle thresh on the submit workers makes it much more
2401 	 * likely that bios will be send down in a sane order to the
2402 	 * devices
2403 	 */
2404 	fs_info->submit_workers =
2405 		btrfs_alloc_workqueue(fs_info, "submit", flags,
2406 				      min_t(u64, fs_devices->num_devices,
2407 					    max_active), 64);
2408 
2409 	fs_info->fixup_workers =
2410 		btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2411 
2412 	/*
2413 	 * endios are largely parallel and should have a very
2414 	 * low idle thresh
2415 	 */
2416 	fs_info->endio_workers =
2417 		btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2418 	fs_info->endio_meta_workers =
2419 		btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2420 				      max_active, 4);
2421 	fs_info->endio_meta_write_workers =
2422 		btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2423 				      max_active, 2);
2424 	fs_info->endio_raid56_workers =
2425 		btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2426 				      max_active, 4);
2427 	fs_info->endio_repair_workers =
2428 		btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2429 	fs_info->rmw_workers =
2430 		btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2431 	fs_info->endio_write_workers =
2432 		btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2433 				      max_active, 2);
2434 	fs_info->endio_freespace_worker =
2435 		btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2436 				      max_active, 0);
2437 	fs_info->delayed_workers =
2438 		btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2439 				      max_active, 0);
2440 	fs_info->readahead_workers =
2441 		btrfs_alloc_workqueue(fs_info, "readahead", flags,
2442 				      max_active, 2);
2443 	fs_info->qgroup_rescan_workers =
2444 		btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2445 	fs_info->extent_workers =
2446 		btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2447 				      min_t(u64, fs_devices->num_devices,
2448 					    max_active), 8);
2449 
2450 	if (!(fs_info->workers && fs_info->delalloc_workers &&
2451 	      fs_info->submit_workers && fs_info->flush_workers &&
2452 	      fs_info->endio_workers && fs_info->endio_meta_workers &&
2453 	      fs_info->endio_meta_write_workers &&
2454 	      fs_info->endio_repair_workers &&
2455 	      fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2456 	      fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2457 	      fs_info->caching_workers && fs_info->readahead_workers &&
2458 	      fs_info->fixup_workers && fs_info->delayed_workers &&
2459 	      fs_info->extent_workers &&
2460 	      fs_info->qgroup_rescan_workers)) {
2461 		return -ENOMEM;
2462 	}
2463 
2464 	return 0;
2465 }
2466 
btrfs_replay_log(struct btrfs_fs_info * fs_info,struct btrfs_fs_devices * fs_devices)2467 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2468 			    struct btrfs_fs_devices *fs_devices)
2469 {
2470 	int ret;
2471 	struct btrfs_root *tree_root = fs_info->tree_root;
2472 	struct btrfs_root *log_tree_root;
2473 	struct btrfs_super_block *disk_super = fs_info->super_copy;
2474 	u64 bytenr = btrfs_super_log_root(disk_super);
2475 
2476 	if (fs_devices->rw_devices == 0) {
2477 		btrfs_warn(fs_info, "log replay required on RO media");
2478 		return -EIO;
2479 	}
2480 
2481 	log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2482 	if (!log_tree_root)
2483 		return -ENOMEM;
2484 
2485 	__setup_root(tree_root->nodesize, tree_root->sectorsize,
2486 			tree_root->stripesize, log_tree_root, fs_info,
2487 			BTRFS_TREE_LOG_OBJECTID);
2488 
2489 	log_tree_root->node = read_tree_block(tree_root, bytenr,
2490 			fs_info->generation + 1);
2491 	if (IS_ERR(log_tree_root->node)) {
2492 		btrfs_warn(fs_info, "failed to read log tree");
2493 		ret = PTR_ERR(log_tree_root->node);
2494 		kfree(log_tree_root);
2495 		return ret;
2496 	} else if (!extent_buffer_uptodate(log_tree_root->node)) {
2497 		btrfs_err(fs_info, "failed to read log tree");
2498 		free_extent_buffer(log_tree_root->node);
2499 		kfree(log_tree_root);
2500 		return -EIO;
2501 	}
2502 	/* returns with log_tree_root freed on success */
2503 	ret = btrfs_recover_log_trees(log_tree_root);
2504 	if (ret) {
2505 		btrfs_handle_fs_error(tree_root->fs_info, ret,
2506 			    "Failed to recover log tree");
2507 		free_extent_buffer(log_tree_root->node);
2508 		kfree(log_tree_root);
2509 		return ret;
2510 	}
2511 
2512 	if (fs_info->sb->s_flags & MS_RDONLY) {
2513 		ret = btrfs_commit_super(tree_root);
2514 		if (ret)
2515 			return ret;
2516 	}
2517 
2518 	return 0;
2519 }
2520 
btrfs_read_roots(struct btrfs_fs_info * fs_info,struct btrfs_root * tree_root)2521 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2522 			    struct btrfs_root *tree_root)
2523 {
2524 	struct btrfs_root *root;
2525 	struct btrfs_key location;
2526 	int ret;
2527 
2528 	location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2529 	location.type = BTRFS_ROOT_ITEM_KEY;
2530 	location.offset = 0;
2531 
2532 	root = btrfs_read_tree_root(tree_root, &location);
2533 	if (IS_ERR(root))
2534 		return PTR_ERR(root);
2535 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2536 	fs_info->extent_root = root;
2537 
2538 	location.objectid = BTRFS_DEV_TREE_OBJECTID;
2539 	root = btrfs_read_tree_root(tree_root, &location);
2540 	if (IS_ERR(root))
2541 		return PTR_ERR(root);
2542 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2543 	fs_info->dev_root = root;
2544 	btrfs_init_devices_late(fs_info);
2545 
2546 	location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2547 	root = btrfs_read_tree_root(tree_root, &location);
2548 	if (IS_ERR(root))
2549 		return PTR_ERR(root);
2550 	set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2551 	fs_info->csum_root = root;
2552 
2553 	location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2554 	root = btrfs_read_tree_root(tree_root, &location);
2555 	if (!IS_ERR(root)) {
2556 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2557 		set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2558 		fs_info->quota_root = root;
2559 	}
2560 
2561 	location.objectid = BTRFS_UUID_TREE_OBJECTID;
2562 	root = btrfs_read_tree_root(tree_root, &location);
2563 	if (IS_ERR(root)) {
2564 		ret = PTR_ERR(root);
2565 		if (ret != -ENOENT)
2566 			return ret;
2567 	} else {
2568 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2569 		fs_info->uuid_root = root;
2570 	}
2571 
2572 	if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2573 		location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2574 		root = btrfs_read_tree_root(tree_root, &location);
2575 		if (IS_ERR(root))
2576 			return PTR_ERR(root);
2577 		set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2578 		fs_info->free_space_root = root;
2579 	}
2580 
2581 	return 0;
2582 }
2583 
open_ctree(struct super_block * sb,struct btrfs_fs_devices * fs_devices,char * options)2584 int open_ctree(struct super_block *sb,
2585 	       struct btrfs_fs_devices *fs_devices,
2586 	       char *options)
2587 {
2588 	u32 sectorsize;
2589 	u32 nodesize;
2590 	u32 stripesize;
2591 	u64 generation;
2592 	u64 features;
2593 	struct btrfs_key location;
2594 	struct buffer_head *bh;
2595 	struct btrfs_super_block *disk_super;
2596 	struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2597 	struct btrfs_root *tree_root;
2598 	struct btrfs_root *chunk_root;
2599 	int ret;
2600 	int err = -EINVAL;
2601 	int num_backups_tried = 0;
2602 	int backup_index = 0;
2603 	int max_active;
2604 	int clear_free_space_tree = 0;
2605 
2606 	tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2607 	chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2608 	if (!tree_root || !chunk_root) {
2609 		err = -ENOMEM;
2610 		goto fail;
2611 	}
2612 
2613 	ret = init_srcu_struct(&fs_info->subvol_srcu);
2614 	if (ret) {
2615 		err = ret;
2616 		goto fail;
2617 	}
2618 
2619 	ret = setup_bdi(fs_info, &fs_info->bdi);
2620 	if (ret) {
2621 		err = ret;
2622 		goto fail_srcu;
2623 	}
2624 
2625 	ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2626 	if (ret) {
2627 		err = ret;
2628 		goto fail_bdi;
2629 	}
2630 	fs_info->dirty_metadata_batch = PAGE_SIZE *
2631 					(1 + ilog2(nr_cpu_ids));
2632 
2633 	ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2634 	if (ret) {
2635 		err = ret;
2636 		goto fail_dirty_metadata_bytes;
2637 	}
2638 
2639 	ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2640 	if (ret) {
2641 		err = ret;
2642 		goto fail_delalloc_bytes;
2643 	}
2644 
2645 	fs_info->btree_inode = new_inode(sb);
2646 	if (!fs_info->btree_inode) {
2647 		err = -ENOMEM;
2648 		goto fail_bio_counter;
2649 	}
2650 
2651 	mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2652 
2653 	INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2654 	INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2655 	INIT_LIST_HEAD(&fs_info->trans_list);
2656 	INIT_LIST_HEAD(&fs_info->dead_roots);
2657 	INIT_LIST_HEAD(&fs_info->delayed_iputs);
2658 	INIT_LIST_HEAD(&fs_info->delalloc_roots);
2659 	INIT_LIST_HEAD(&fs_info->caching_block_groups);
2660 	spin_lock_init(&fs_info->delalloc_root_lock);
2661 	spin_lock_init(&fs_info->trans_lock);
2662 	spin_lock_init(&fs_info->fs_roots_radix_lock);
2663 	spin_lock_init(&fs_info->delayed_iput_lock);
2664 	spin_lock_init(&fs_info->defrag_inodes_lock);
2665 	spin_lock_init(&fs_info->free_chunk_lock);
2666 	spin_lock_init(&fs_info->tree_mod_seq_lock);
2667 	spin_lock_init(&fs_info->super_lock);
2668 	spin_lock_init(&fs_info->qgroup_op_lock);
2669 	spin_lock_init(&fs_info->buffer_lock);
2670 	spin_lock_init(&fs_info->unused_bgs_lock);
2671 	rwlock_init(&fs_info->tree_mod_log_lock);
2672 	mutex_init(&fs_info->unused_bg_unpin_mutex);
2673 	mutex_init(&fs_info->delete_unused_bgs_mutex);
2674 	mutex_init(&fs_info->reloc_mutex);
2675 	mutex_init(&fs_info->delalloc_root_mutex);
2676 	mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2677 	seqlock_init(&fs_info->profiles_lock);
2678 
2679 	INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2680 	INIT_LIST_HEAD(&fs_info->space_info);
2681 	INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2682 	INIT_LIST_HEAD(&fs_info->unused_bgs);
2683 	btrfs_mapping_init(&fs_info->mapping_tree);
2684 	btrfs_init_block_rsv(&fs_info->global_block_rsv,
2685 			     BTRFS_BLOCK_RSV_GLOBAL);
2686 	btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2687 			     BTRFS_BLOCK_RSV_DELALLOC);
2688 	btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2689 	btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2690 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2691 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2692 			     BTRFS_BLOCK_RSV_DELOPS);
2693 	atomic_set(&fs_info->nr_async_submits, 0);
2694 	atomic_set(&fs_info->async_delalloc_pages, 0);
2695 	atomic_set(&fs_info->async_submit_draining, 0);
2696 	atomic_set(&fs_info->nr_async_bios, 0);
2697 	atomic_set(&fs_info->defrag_running, 0);
2698 	atomic_set(&fs_info->qgroup_op_seq, 0);
2699 	atomic_set(&fs_info->reada_works_cnt, 0);
2700 	atomic64_set(&fs_info->tree_mod_seq, 0);
2701 	fs_info->fs_frozen = 0;
2702 	fs_info->sb = sb;
2703 	fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2704 	fs_info->metadata_ratio = 0;
2705 	fs_info->defrag_inodes = RB_ROOT;
2706 	fs_info->free_chunk_space = 0;
2707 	fs_info->tree_mod_log = RB_ROOT;
2708 	fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2709 	fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2710 	/* readahead state */
2711 	INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2712 	spin_lock_init(&fs_info->reada_lock);
2713 
2714 	fs_info->thread_pool_size = min_t(unsigned long,
2715 					  num_online_cpus() + 2, 8);
2716 
2717 	INIT_LIST_HEAD(&fs_info->ordered_roots);
2718 	spin_lock_init(&fs_info->ordered_root_lock);
2719 	fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2720 					GFP_KERNEL);
2721 	if (!fs_info->delayed_root) {
2722 		err = -ENOMEM;
2723 		goto fail_iput;
2724 	}
2725 	btrfs_init_delayed_root(fs_info->delayed_root);
2726 
2727 	btrfs_init_scrub(fs_info);
2728 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2729 	fs_info->check_integrity_print_mask = 0;
2730 #endif
2731 	btrfs_init_balance(fs_info);
2732 	btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2733 
2734 	sb->s_blocksize = 4096;
2735 	sb->s_blocksize_bits = blksize_bits(4096);
2736 	sb->s_bdi = &fs_info->bdi;
2737 
2738 	btrfs_init_btree_inode(fs_info, tree_root);
2739 
2740 	spin_lock_init(&fs_info->block_group_cache_lock);
2741 	fs_info->block_group_cache_tree = RB_ROOT;
2742 	fs_info->first_logical_byte = (u64)-1;
2743 
2744 	extent_io_tree_init(&fs_info->freed_extents[0],
2745 			     fs_info->btree_inode->i_mapping);
2746 	extent_io_tree_init(&fs_info->freed_extents[1],
2747 			     fs_info->btree_inode->i_mapping);
2748 	fs_info->pinned_extents = &fs_info->freed_extents[0];
2749 	set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2750 
2751 	mutex_init(&fs_info->ordered_operations_mutex);
2752 	mutex_init(&fs_info->tree_log_mutex);
2753 	mutex_init(&fs_info->chunk_mutex);
2754 	mutex_init(&fs_info->transaction_kthread_mutex);
2755 	mutex_init(&fs_info->cleaner_mutex);
2756 	mutex_init(&fs_info->volume_mutex);
2757 	mutex_init(&fs_info->ro_block_group_mutex);
2758 	init_rwsem(&fs_info->commit_root_sem);
2759 	init_rwsem(&fs_info->cleanup_work_sem);
2760 	init_rwsem(&fs_info->subvol_sem);
2761 	sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2762 
2763 	btrfs_init_dev_replace_locks(fs_info);
2764 	btrfs_init_qgroup(fs_info);
2765 
2766 	btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2767 	btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2768 
2769 	init_waitqueue_head(&fs_info->transaction_throttle);
2770 	init_waitqueue_head(&fs_info->transaction_wait);
2771 	init_waitqueue_head(&fs_info->transaction_blocked_wait);
2772 	init_waitqueue_head(&fs_info->async_submit_wait);
2773 
2774 	INIT_LIST_HEAD(&fs_info->pinned_chunks);
2775 
2776 	ret = btrfs_alloc_stripe_hash_table(fs_info);
2777 	if (ret) {
2778 		err = ret;
2779 		goto fail_alloc;
2780 	}
2781 
2782 	__setup_root(4096, 4096, 4096, tree_root,
2783 		     fs_info, BTRFS_ROOT_TREE_OBJECTID);
2784 
2785 	invalidate_bdev(fs_devices->latest_bdev);
2786 
2787 	/*
2788 	 * Read super block and check the signature bytes only
2789 	 */
2790 	bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2791 	if (IS_ERR(bh)) {
2792 		err = PTR_ERR(bh);
2793 		goto fail_alloc;
2794 	}
2795 
2796 	/*
2797 	 * We want to check superblock checksum, the type is stored inside.
2798 	 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2799 	 */
2800 	if (btrfs_check_super_csum(fs_info, bh->b_data)) {
2801 		btrfs_err(fs_info, "superblock checksum mismatch");
2802 		err = -EINVAL;
2803 		brelse(bh);
2804 		goto fail_alloc;
2805 	}
2806 
2807 	/*
2808 	 * super_copy is zeroed at allocation time and we never touch the
2809 	 * following bytes up to INFO_SIZE, the checksum is calculated from
2810 	 * the whole block of INFO_SIZE
2811 	 */
2812 	memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2813 	memcpy(fs_info->super_for_commit, fs_info->super_copy,
2814 	       sizeof(*fs_info->super_for_commit));
2815 	brelse(bh);
2816 
2817 	memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2818 
2819 	ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2820 	if (ret) {
2821 		btrfs_err(fs_info, "superblock contains fatal errors");
2822 		err = -EINVAL;
2823 		goto fail_alloc;
2824 	}
2825 
2826 	disk_super = fs_info->super_copy;
2827 	if (!btrfs_super_root(disk_super))
2828 		goto fail_alloc;
2829 
2830 	/* check FS state, whether FS is broken. */
2831 	if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2832 		set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2833 
2834 	/*
2835 	 * run through our array of backup supers and setup
2836 	 * our ring pointer to the oldest one
2837 	 */
2838 	generation = btrfs_super_generation(disk_super);
2839 	find_oldest_super_backup(fs_info, generation);
2840 
2841 	/*
2842 	 * In the long term, we'll store the compression type in the super
2843 	 * block, and it'll be used for per file compression control.
2844 	 */
2845 	fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2846 
2847 	ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2848 	if (ret) {
2849 		err = ret;
2850 		goto fail_alloc;
2851 	}
2852 
2853 	features = btrfs_super_incompat_flags(disk_super) &
2854 		~BTRFS_FEATURE_INCOMPAT_SUPP;
2855 	if (features) {
2856 		btrfs_err(fs_info,
2857 		    "cannot mount because of unsupported optional features (%llx)",
2858 		    features);
2859 		err = -EINVAL;
2860 		goto fail_alloc;
2861 	}
2862 
2863 	features = btrfs_super_incompat_flags(disk_super);
2864 	features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2865 	if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2866 		features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2867 
2868 	if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2869 		btrfs_info(fs_info, "has skinny extents");
2870 
2871 	/*
2872 	 * flag our filesystem as having big metadata blocks if
2873 	 * they are bigger than the page size
2874 	 */
2875 	if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2876 		if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2877 			btrfs_info(fs_info,
2878 				"flagging fs with big metadata feature");
2879 		features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2880 	}
2881 
2882 	nodesize = btrfs_super_nodesize(disk_super);
2883 	sectorsize = btrfs_super_sectorsize(disk_super);
2884 	stripesize = sectorsize;
2885 	fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2886 	fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2887 
2888 	/*
2889 	 * mixed block groups end up with duplicate but slightly offset
2890 	 * extent buffers for the same range.  It leads to corruptions
2891 	 */
2892 	if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2893 	    (sectorsize != nodesize)) {
2894 		btrfs_err(fs_info,
2895 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2896 			nodesize, sectorsize);
2897 		goto fail_alloc;
2898 	}
2899 
2900 	/*
2901 	 * Needn't use the lock because there is no other task which will
2902 	 * update the flag.
2903 	 */
2904 	btrfs_set_super_incompat_flags(disk_super, features);
2905 
2906 	features = btrfs_super_compat_ro_flags(disk_super) &
2907 		~BTRFS_FEATURE_COMPAT_RO_SUPP;
2908 	if (!(sb->s_flags & MS_RDONLY) && features) {
2909 		btrfs_err(fs_info,
2910 	"cannot mount read-write because of unsupported optional features (%llx)",
2911 		       features);
2912 		err = -EINVAL;
2913 		goto fail_alloc;
2914 	}
2915 
2916 	max_active = fs_info->thread_pool_size;
2917 
2918 	ret = btrfs_init_workqueues(fs_info, fs_devices);
2919 	if (ret) {
2920 		err = ret;
2921 		goto fail_sb_buffer;
2922 	}
2923 
2924 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2925 	fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2926 				    SZ_4M / PAGE_SIZE);
2927 
2928 	tree_root->nodesize = nodesize;
2929 	tree_root->sectorsize = sectorsize;
2930 	tree_root->stripesize = stripesize;
2931 
2932 	sb->s_blocksize = sectorsize;
2933 	sb->s_blocksize_bits = blksize_bits(sectorsize);
2934 
2935 	mutex_lock(&fs_info->chunk_mutex);
2936 	ret = btrfs_read_sys_array(tree_root);
2937 	mutex_unlock(&fs_info->chunk_mutex);
2938 	if (ret) {
2939 		btrfs_err(fs_info, "failed to read the system array: %d", ret);
2940 		goto fail_sb_buffer;
2941 	}
2942 
2943 	generation = btrfs_super_chunk_root_generation(disk_super);
2944 
2945 	__setup_root(nodesize, sectorsize, stripesize, chunk_root,
2946 		     fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2947 
2948 	chunk_root->node = read_tree_block(chunk_root,
2949 					   btrfs_super_chunk_root(disk_super),
2950 					   generation);
2951 	if (IS_ERR(chunk_root->node) ||
2952 	    !extent_buffer_uptodate(chunk_root->node)) {
2953 		btrfs_err(fs_info, "failed to read chunk root");
2954 		if (!IS_ERR(chunk_root->node))
2955 			free_extent_buffer(chunk_root->node);
2956 		chunk_root->node = NULL;
2957 		goto fail_tree_roots;
2958 	}
2959 	btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2960 	chunk_root->commit_root = btrfs_root_node(chunk_root);
2961 
2962 	read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2963 	   btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2964 
2965 	ret = btrfs_read_chunk_tree(chunk_root);
2966 	if (ret) {
2967 		btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2968 		goto fail_tree_roots;
2969 	}
2970 
2971 	/*
2972 	 * keep the device that is marked to be the target device for the
2973 	 * dev_replace procedure
2974 	 */
2975 	btrfs_close_extra_devices(fs_devices, 0);
2976 
2977 	if (!fs_devices->latest_bdev) {
2978 		btrfs_err(fs_info, "failed to read devices");
2979 		goto fail_tree_roots;
2980 	}
2981 
2982 retry_root_backup:
2983 	generation = btrfs_super_generation(disk_super);
2984 
2985 	tree_root->node = read_tree_block(tree_root,
2986 					  btrfs_super_root(disk_super),
2987 					  generation);
2988 	if (IS_ERR(tree_root->node) ||
2989 	    !extent_buffer_uptodate(tree_root->node)) {
2990 		btrfs_warn(fs_info, "failed to read tree root");
2991 		if (!IS_ERR(tree_root->node))
2992 			free_extent_buffer(tree_root->node);
2993 		tree_root->node = NULL;
2994 		goto recovery_tree_root;
2995 	}
2996 
2997 	btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2998 	tree_root->commit_root = btrfs_root_node(tree_root);
2999 	btrfs_set_root_refs(&tree_root->root_item, 1);
3000 
3001 	mutex_lock(&tree_root->objectid_mutex);
3002 	ret = btrfs_find_highest_objectid(tree_root,
3003 					&tree_root->highest_objectid);
3004 	if (ret) {
3005 		mutex_unlock(&tree_root->objectid_mutex);
3006 		goto recovery_tree_root;
3007 	}
3008 
3009 	ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
3010 
3011 	mutex_unlock(&tree_root->objectid_mutex);
3012 
3013 	ret = btrfs_read_roots(fs_info, tree_root);
3014 	if (ret)
3015 		goto recovery_tree_root;
3016 
3017 	fs_info->generation = generation;
3018 	fs_info->last_trans_committed = generation;
3019 
3020 	ret = btrfs_recover_balance(fs_info);
3021 	if (ret) {
3022 		btrfs_err(fs_info, "failed to recover balance: %d", ret);
3023 		goto fail_block_groups;
3024 	}
3025 
3026 	ret = btrfs_init_dev_stats(fs_info);
3027 	if (ret) {
3028 		btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3029 		goto fail_block_groups;
3030 	}
3031 
3032 	ret = btrfs_init_dev_replace(fs_info);
3033 	if (ret) {
3034 		btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3035 		goto fail_block_groups;
3036 	}
3037 
3038 	btrfs_close_extra_devices(fs_devices, 1);
3039 
3040 	ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3041 	if (ret) {
3042 		btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3043 				ret);
3044 		goto fail_block_groups;
3045 	}
3046 
3047 	ret = btrfs_sysfs_add_device(fs_devices);
3048 	if (ret) {
3049 		btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3050 				ret);
3051 		goto fail_fsdev_sysfs;
3052 	}
3053 
3054 	ret = btrfs_sysfs_add_mounted(fs_info);
3055 	if (ret) {
3056 		btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3057 		goto fail_fsdev_sysfs;
3058 	}
3059 
3060 	ret = btrfs_init_space_info(fs_info);
3061 	if (ret) {
3062 		btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3063 		goto fail_sysfs;
3064 	}
3065 
3066 	ret = btrfs_read_block_groups(fs_info->extent_root);
3067 	if (ret) {
3068 		btrfs_err(fs_info, "failed to read block groups: %d", ret);
3069 		goto fail_sysfs;
3070 	}
3071 	fs_info->num_tolerated_disk_barrier_failures =
3072 		btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3073 	if (fs_info->fs_devices->missing_devices >
3074 	     fs_info->num_tolerated_disk_barrier_failures &&
3075 	    !(sb->s_flags & MS_RDONLY)) {
3076 		btrfs_warn(fs_info,
3077 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3078 			fs_info->fs_devices->missing_devices,
3079 			fs_info->num_tolerated_disk_barrier_failures);
3080 		goto fail_sysfs;
3081 	}
3082 
3083 	fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3084 					       "btrfs-cleaner");
3085 	if (IS_ERR(fs_info->cleaner_kthread))
3086 		goto fail_sysfs;
3087 
3088 	fs_info->transaction_kthread = kthread_run(transaction_kthread,
3089 						   tree_root,
3090 						   "btrfs-transaction");
3091 	if (IS_ERR(fs_info->transaction_kthread))
3092 		goto fail_cleaner;
3093 
3094 	if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3095 	    !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3096 	    !fs_info->fs_devices->rotating) {
3097 		btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3098 		btrfs_set_opt(fs_info->mount_opt, SSD);
3099 	}
3100 
3101 	/*
3102 	 * Mount does not set all options immediately, we can do it now and do
3103 	 * not have to wait for transaction commit
3104 	 */
3105 	btrfs_apply_pending_changes(fs_info);
3106 
3107 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3108 	if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3109 		ret = btrfsic_mount(tree_root, fs_devices,
3110 				    btrfs_test_opt(tree_root->fs_info,
3111 					CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3112 				    1 : 0,
3113 				    fs_info->check_integrity_print_mask);
3114 		if (ret)
3115 			btrfs_warn(fs_info,
3116 				"failed to initialize integrity check module: %d",
3117 				ret);
3118 	}
3119 #endif
3120 	ret = btrfs_read_qgroup_config(fs_info);
3121 	if (ret)
3122 		goto fail_trans_kthread;
3123 
3124 	/* do not make disk changes in broken FS or nologreplay is given */
3125 	if (btrfs_super_log_root(disk_super) != 0 &&
3126 	    !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3127 		ret = btrfs_replay_log(fs_info, fs_devices);
3128 		if (ret) {
3129 			err = ret;
3130 			goto fail_qgroup;
3131 		}
3132 	}
3133 
3134 	ret = btrfs_find_orphan_roots(tree_root);
3135 	if (ret)
3136 		goto fail_qgroup;
3137 
3138 	if (!(sb->s_flags & MS_RDONLY)) {
3139 		ret = btrfs_cleanup_fs_roots(fs_info);
3140 		if (ret)
3141 			goto fail_qgroup;
3142 
3143 		mutex_lock(&fs_info->cleaner_mutex);
3144 		ret = btrfs_recover_relocation(tree_root);
3145 		mutex_unlock(&fs_info->cleaner_mutex);
3146 		if (ret < 0) {
3147 			btrfs_warn(fs_info, "failed to recover relocation: %d",
3148 					ret);
3149 			err = -EINVAL;
3150 			goto fail_qgroup;
3151 		}
3152 	}
3153 
3154 	location.objectid = BTRFS_FS_TREE_OBJECTID;
3155 	location.type = BTRFS_ROOT_ITEM_KEY;
3156 	location.offset = 0;
3157 
3158 	fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3159 	if (IS_ERR(fs_info->fs_root)) {
3160 		err = PTR_ERR(fs_info->fs_root);
3161 		goto fail_qgroup;
3162 	}
3163 
3164 	if (sb->s_flags & MS_RDONLY)
3165 		return 0;
3166 
3167 	if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3168 	    btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3169 		clear_free_space_tree = 1;
3170 	} else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3171 		   !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3172 		btrfs_warn(fs_info, "free space tree is invalid");
3173 		clear_free_space_tree = 1;
3174 	}
3175 
3176 	if (clear_free_space_tree) {
3177 		btrfs_info(fs_info, "clearing free space tree");
3178 		ret = btrfs_clear_free_space_tree(fs_info);
3179 		if (ret) {
3180 			btrfs_warn(fs_info,
3181 				   "failed to clear free space tree: %d", ret);
3182 			close_ctree(tree_root);
3183 			return ret;
3184 		}
3185 	}
3186 
3187 	if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3188 	    !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3189 		btrfs_info(fs_info, "creating free space tree");
3190 		ret = btrfs_create_free_space_tree(fs_info);
3191 		if (ret) {
3192 			btrfs_warn(fs_info,
3193 				"failed to create free space tree: %d", ret);
3194 			close_ctree(tree_root);
3195 			return ret;
3196 		}
3197 	}
3198 
3199 	down_read(&fs_info->cleanup_work_sem);
3200 	if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3201 	    (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3202 		up_read(&fs_info->cleanup_work_sem);
3203 		close_ctree(tree_root);
3204 		return ret;
3205 	}
3206 	up_read(&fs_info->cleanup_work_sem);
3207 
3208 	ret = btrfs_resume_balance_async(fs_info);
3209 	if (ret) {
3210 		btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3211 		close_ctree(tree_root);
3212 		return ret;
3213 	}
3214 
3215 	ret = btrfs_resume_dev_replace_async(fs_info);
3216 	if (ret) {
3217 		btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3218 		close_ctree(tree_root);
3219 		return ret;
3220 	}
3221 
3222 	btrfs_qgroup_rescan_resume(fs_info);
3223 
3224 	if (!fs_info->uuid_root) {
3225 		btrfs_info(fs_info, "creating UUID tree");
3226 		ret = btrfs_create_uuid_tree(fs_info);
3227 		if (ret) {
3228 			btrfs_warn(fs_info,
3229 				"failed to create the UUID tree: %d", ret);
3230 			close_ctree(tree_root);
3231 			return ret;
3232 		}
3233 	} else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3234 		   fs_info->generation !=
3235 				btrfs_super_uuid_tree_generation(disk_super)) {
3236 		btrfs_info(fs_info, "checking UUID tree");
3237 		ret = btrfs_check_uuid_tree(fs_info);
3238 		if (ret) {
3239 			btrfs_warn(fs_info,
3240 				"failed to check the UUID tree: %d", ret);
3241 			close_ctree(tree_root);
3242 			return ret;
3243 		}
3244 	} else {
3245 		set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3246 	}
3247 	set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3248 
3249 	/*
3250 	 * backuproot only affect mount behavior, and if open_ctree succeeded,
3251 	 * no need to keep the flag
3252 	 */
3253 	btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3254 
3255 	return 0;
3256 
3257 fail_qgroup:
3258 	btrfs_free_qgroup_config(fs_info);
3259 fail_trans_kthread:
3260 	kthread_stop(fs_info->transaction_kthread);
3261 	btrfs_cleanup_transaction(fs_info->tree_root);
3262 	btrfs_free_fs_roots(fs_info);
3263 fail_cleaner:
3264 	kthread_stop(fs_info->cleaner_kthread);
3265 
3266 	/*
3267 	 * make sure we're done with the btree inode before we stop our
3268 	 * kthreads
3269 	 */
3270 	filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3271 
3272 fail_sysfs:
3273 	btrfs_sysfs_remove_mounted(fs_info);
3274 
3275 fail_fsdev_sysfs:
3276 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3277 
3278 fail_block_groups:
3279 	btrfs_put_block_group_cache(fs_info);
3280 	btrfs_free_block_groups(fs_info);
3281 
3282 fail_tree_roots:
3283 	free_root_pointers(fs_info, 1);
3284 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3285 
3286 fail_sb_buffer:
3287 	btrfs_stop_all_workers(fs_info);
3288 fail_alloc:
3289 fail_iput:
3290 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
3291 
3292 	iput(fs_info->btree_inode);
3293 fail_bio_counter:
3294 	percpu_counter_destroy(&fs_info->bio_counter);
3295 fail_delalloc_bytes:
3296 	percpu_counter_destroy(&fs_info->delalloc_bytes);
3297 fail_dirty_metadata_bytes:
3298 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3299 fail_bdi:
3300 	bdi_destroy(&fs_info->bdi);
3301 fail_srcu:
3302 	cleanup_srcu_struct(&fs_info->subvol_srcu);
3303 fail:
3304 	btrfs_free_stripe_hash_table(fs_info);
3305 	btrfs_close_devices(fs_info->fs_devices);
3306 	return err;
3307 
3308 recovery_tree_root:
3309 	if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3310 		goto fail_tree_roots;
3311 
3312 	free_root_pointers(fs_info, 0);
3313 
3314 	/* don't use the log in recovery mode, it won't be valid */
3315 	btrfs_set_super_log_root(disk_super, 0);
3316 
3317 	/* we can't trust the free space cache either */
3318 	btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3319 
3320 	ret = next_root_backup(fs_info, fs_info->super_copy,
3321 			       &num_backups_tried, &backup_index);
3322 	if (ret == -1)
3323 		goto fail_block_groups;
3324 	goto retry_root_backup;
3325 }
3326 
btrfs_end_buffer_write_sync(struct buffer_head * bh,int uptodate)3327 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3328 {
3329 	if (uptodate) {
3330 		set_buffer_uptodate(bh);
3331 	} else {
3332 		struct btrfs_device *device = (struct btrfs_device *)
3333 			bh->b_private;
3334 
3335 		btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3336 				"lost page write due to IO error on %s",
3337 					  rcu_str_deref(device->name));
3338 		/* note, we don't set_buffer_write_io_error because we have
3339 		 * our own ways of dealing with the IO errors
3340 		 */
3341 		clear_buffer_uptodate(bh);
3342 		btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3343 	}
3344 	unlock_buffer(bh);
3345 	put_bh(bh);
3346 }
3347 
btrfs_read_dev_one_super(struct block_device * bdev,int copy_num,struct buffer_head ** bh_ret)3348 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3349 			struct buffer_head **bh_ret)
3350 {
3351 	struct buffer_head *bh;
3352 	struct btrfs_super_block *super;
3353 	u64 bytenr;
3354 
3355 	bytenr = btrfs_sb_offset(copy_num);
3356 	if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3357 		return -EINVAL;
3358 
3359 	bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3360 	/*
3361 	 * If we fail to read from the underlying devices, as of now
3362 	 * the best option we have is to mark it EIO.
3363 	 */
3364 	if (!bh)
3365 		return -EIO;
3366 
3367 	super = (struct btrfs_super_block *)bh->b_data;
3368 	if (btrfs_super_bytenr(super) != bytenr ||
3369 		    btrfs_super_magic(super) != BTRFS_MAGIC) {
3370 		brelse(bh);
3371 		return -EINVAL;
3372 	}
3373 
3374 	*bh_ret = bh;
3375 	return 0;
3376 }
3377 
3378 
btrfs_read_dev_super(struct block_device * bdev)3379 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3380 {
3381 	struct buffer_head *bh;
3382 	struct buffer_head *latest = NULL;
3383 	struct btrfs_super_block *super;
3384 	int i;
3385 	u64 transid = 0;
3386 	int ret = -EINVAL;
3387 
3388 	/* we would like to check all the supers, but that would make
3389 	 * a btrfs mount succeed after a mkfs from a different FS.
3390 	 * So, we need to add a special mount option to scan for
3391 	 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3392 	 */
3393 	for (i = 0; i < 1; i++) {
3394 		ret = btrfs_read_dev_one_super(bdev, i, &bh);
3395 		if (ret)
3396 			continue;
3397 
3398 		super = (struct btrfs_super_block *)bh->b_data;
3399 
3400 		if (!latest || btrfs_super_generation(super) > transid) {
3401 			brelse(latest);
3402 			latest = bh;
3403 			transid = btrfs_super_generation(super);
3404 		} else {
3405 			brelse(bh);
3406 		}
3407 	}
3408 
3409 	if (!latest)
3410 		return ERR_PTR(ret);
3411 
3412 	return latest;
3413 }
3414 
3415 /*
3416  * this should be called twice, once with wait == 0 and
3417  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3418  * we write are pinned.
3419  *
3420  * They are released when wait == 1 is done.
3421  * max_mirrors must be the same for both runs, and it indicates how
3422  * many supers on this one device should be written.
3423  *
3424  * max_mirrors == 0 means to write them all.
3425  */
write_dev_supers(struct btrfs_device * device,struct btrfs_super_block * sb,int do_barriers,int wait,int max_mirrors)3426 static int write_dev_supers(struct btrfs_device *device,
3427 			    struct btrfs_super_block *sb,
3428 			    int do_barriers, int wait, int max_mirrors)
3429 {
3430 	struct buffer_head *bh;
3431 	int i;
3432 	int ret;
3433 	int errors = 0;
3434 	u32 crc;
3435 	u64 bytenr;
3436 
3437 	if (max_mirrors == 0)
3438 		max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3439 
3440 	for (i = 0; i < max_mirrors; i++) {
3441 		bytenr = btrfs_sb_offset(i);
3442 		if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3443 		    device->commit_total_bytes)
3444 			break;
3445 
3446 		if (wait) {
3447 			bh = __find_get_block(device->bdev, bytenr / 4096,
3448 					      BTRFS_SUPER_INFO_SIZE);
3449 			if (!bh) {
3450 				errors++;
3451 				continue;
3452 			}
3453 			wait_on_buffer(bh);
3454 			if (!buffer_uptodate(bh))
3455 				errors++;
3456 
3457 			/* drop our reference */
3458 			brelse(bh);
3459 
3460 			/* drop the reference from the wait == 0 run */
3461 			brelse(bh);
3462 			continue;
3463 		} else {
3464 			btrfs_set_super_bytenr(sb, bytenr);
3465 
3466 			crc = ~(u32)0;
3467 			crc = btrfs_csum_data((char *)sb +
3468 					      BTRFS_CSUM_SIZE, crc,
3469 					      BTRFS_SUPER_INFO_SIZE -
3470 					      BTRFS_CSUM_SIZE);
3471 			btrfs_csum_final(crc, sb->csum);
3472 
3473 			/*
3474 			 * one reference for us, and we leave it for the
3475 			 * caller
3476 			 */
3477 			bh = __getblk(device->bdev, bytenr / 4096,
3478 				      BTRFS_SUPER_INFO_SIZE);
3479 			if (!bh) {
3480 				btrfs_err(device->dev_root->fs_info,
3481 				    "couldn't get super buffer head for bytenr %llu",
3482 				    bytenr);
3483 				errors++;
3484 				continue;
3485 			}
3486 
3487 			memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3488 
3489 			/* one reference for submit_bh */
3490 			get_bh(bh);
3491 
3492 			set_buffer_uptodate(bh);
3493 			lock_buffer(bh);
3494 			bh->b_end_io = btrfs_end_buffer_write_sync;
3495 			bh->b_private = device;
3496 		}
3497 
3498 		/*
3499 		 * we fua the first super.  The others we allow
3500 		 * to go down lazy.
3501 		 */
3502 		if (i == 0)
3503 			ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_FUA, bh);
3504 		else
3505 			ret = btrfsic_submit_bh(REQ_OP_WRITE, WRITE_SYNC, bh);
3506 		if (ret)
3507 			errors++;
3508 	}
3509 	return errors < i ? 0 : -1;
3510 }
3511 
3512 /*
3513  * endio for the write_dev_flush, this will wake anyone waiting
3514  * for the barrier when it is done
3515  */
btrfs_end_empty_barrier(struct bio * bio)3516 static void btrfs_end_empty_barrier(struct bio *bio)
3517 {
3518 	if (bio->bi_private)
3519 		complete(bio->bi_private);
3520 	bio_put(bio);
3521 }
3522 
3523 /*
3524  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3525  * sent down.  With wait == 1, it waits for the previous flush.
3526  *
3527  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3528  * capable
3529  */
write_dev_flush(struct btrfs_device * device,int wait)3530 static int write_dev_flush(struct btrfs_device *device, int wait)
3531 {
3532 	struct bio *bio;
3533 	int ret = 0;
3534 
3535 	if (device->nobarriers)
3536 		return 0;
3537 
3538 	if (wait) {
3539 		bio = device->flush_bio;
3540 		if (!bio)
3541 			return 0;
3542 
3543 		wait_for_completion(&device->flush_wait);
3544 
3545 		if (bio->bi_error) {
3546 			ret = bio->bi_error;
3547 			btrfs_dev_stat_inc_and_print(device,
3548 				BTRFS_DEV_STAT_FLUSH_ERRS);
3549 		}
3550 
3551 		/* drop the reference from the wait == 0 run */
3552 		bio_put(bio);
3553 		device->flush_bio = NULL;
3554 
3555 		return ret;
3556 	}
3557 
3558 	/*
3559 	 * one reference for us, and we leave it for the
3560 	 * caller
3561 	 */
3562 	device->flush_bio = NULL;
3563 	bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3564 	if (!bio)
3565 		return -ENOMEM;
3566 
3567 	bio->bi_end_io = btrfs_end_empty_barrier;
3568 	bio->bi_bdev = device->bdev;
3569 	bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
3570 	init_completion(&device->flush_wait);
3571 	bio->bi_private = &device->flush_wait;
3572 	device->flush_bio = bio;
3573 
3574 	bio_get(bio);
3575 	btrfsic_submit_bio(bio);
3576 
3577 	return 0;
3578 }
3579 
3580 /*
3581  * send an empty flush down to each device in parallel,
3582  * then wait for them
3583  */
barrier_all_devices(struct btrfs_fs_info * info)3584 static int barrier_all_devices(struct btrfs_fs_info *info)
3585 {
3586 	struct list_head *head;
3587 	struct btrfs_device *dev;
3588 	int errors_send = 0;
3589 	int errors_wait = 0;
3590 	int ret;
3591 
3592 	/* send down all the barriers */
3593 	head = &info->fs_devices->devices;
3594 	list_for_each_entry_rcu(dev, head, dev_list) {
3595 		if (dev->missing)
3596 			continue;
3597 		if (!dev->bdev) {
3598 			errors_send++;
3599 			continue;
3600 		}
3601 		if (!dev->in_fs_metadata || !dev->writeable)
3602 			continue;
3603 
3604 		ret = write_dev_flush(dev, 0);
3605 		if (ret)
3606 			errors_send++;
3607 	}
3608 
3609 	/* wait for all the barriers */
3610 	list_for_each_entry_rcu(dev, head, dev_list) {
3611 		if (dev->missing)
3612 			continue;
3613 		if (!dev->bdev) {
3614 			errors_wait++;
3615 			continue;
3616 		}
3617 		if (!dev->in_fs_metadata || !dev->writeable)
3618 			continue;
3619 
3620 		ret = write_dev_flush(dev, 1);
3621 		if (ret)
3622 			errors_wait++;
3623 	}
3624 	if (errors_send > info->num_tolerated_disk_barrier_failures ||
3625 	    errors_wait > info->num_tolerated_disk_barrier_failures)
3626 		return -EIO;
3627 	return 0;
3628 }
3629 
btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)3630 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3631 {
3632 	int raid_type;
3633 	int min_tolerated = INT_MAX;
3634 
3635 	if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3636 	    (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3637 		min_tolerated = min(min_tolerated,
3638 				    btrfs_raid_array[BTRFS_RAID_SINGLE].
3639 				    tolerated_failures);
3640 
3641 	for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3642 		if (raid_type == BTRFS_RAID_SINGLE)
3643 			continue;
3644 		if (!(flags & btrfs_raid_group[raid_type]))
3645 			continue;
3646 		min_tolerated = min(min_tolerated,
3647 				    btrfs_raid_array[raid_type].
3648 				    tolerated_failures);
3649 	}
3650 
3651 	if (min_tolerated == INT_MAX) {
3652 		pr_warn("BTRFS: unknown raid flag: %llu", flags);
3653 		min_tolerated = 0;
3654 	}
3655 
3656 	return min_tolerated;
3657 }
3658 
btrfs_calc_num_tolerated_disk_barrier_failures(struct btrfs_fs_info * fs_info)3659 int btrfs_calc_num_tolerated_disk_barrier_failures(
3660 	struct btrfs_fs_info *fs_info)
3661 {
3662 	struct btrfs_ioctl_space_info space;
3663 	struct btrfs_space_info *sinfo;
3664 	u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3665 		       BTRFS_BLOCK_GROUP_SYSTEM,
3666 		       BTRFS_BLOCK_GROUP_METADATA,
3667 		       BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3668 	int i;
3669 	int c;
3670 	int num_tolerated_disk_barrier_failures =
3671 		(int)fs_info->fs_devices->num_devices;
3672 
3673 	for (i = 0; i < ARRAY_SIZE(types); i++) {
3674 		struct btrfs_space_info *tmp;
3675 
3676 		sinfo = NULL;
3677 		rcu_read_lock();
3678 		list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3679 			if (tmp->flags == types[i]) {
3680 				sinfo = tmp;
3681 				break;
3682 			}
3683 		}
3684 		rcu_read_unlock();
3685 
3686 		if (!sinfo)
3687 			continue;
3688 
3689 		down_read(&sinfo->groups_sem);
3690 		for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3691 			u64 flags;
3692 
3693 			if (list_empty(&sinfo->block_groups[c]))
3694 				continue;
3695 
3696 			btrfs_get_block_group_info(&sinfo->block_groups[c],
3697 						   &space);
3698 			if (space.total_bytes == 0 || space.used_bytes == 0)
3699 				continue;
3700 			flags = space.flags;
3701 
3702 			num_tolerated_disk_barrier_failures = min(
3703 				num_tolerated_disk_barrier_failures,
3704 				btrfs_get_num_tolerated_disk_barrier_failures(
3705 					flags));
3706 		}
3707 		up_read(&sinfo->groups_sem);
3708 	}
3709 
3710 	return num_tolerated_disk_barrier_failures;
3711 }
3712 
write_all_supers(struct btrfs_root * root,int max_mirrors)3713 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3714 {
3715 	struct list_head *head;
3716 	struct btrfs_device *dev;
3717 	struct btrfs_super_block *sb;
3718 	struct btrfs_dev_item *dev_item;
3719 	int ret;
3720 	int do_barriers;
3721 	int max_errors;
3722 	int total_errors = 0;
3723 	u64 flags;
3724 
3725 	do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3726 	backup_super_roots(root->fs_info);
3727 
3728 	sb = root->fs_info->super_for_commit;
3729 	dev_item = &sb->dev_item;
3730 
3731 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3732 	head = &root->fs_info->fs_devices->devices;
3733 	max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3734 
3735 	if (do_barriers) {
3736 		ret = barrier_all_devices(root->fs_info);
3737 		if (ret) {
3738 			mutex_unlock(
3739 				&root->fs_info->fs_devices->device_list_mutex);
3740 			btrfs_handle_fs_error(root->fs_info, ret,
3741 				    "errors while submitting device barriers.");
3742 			return ret;
3743 		}
3744 	}
3745 
3746 	list_for_each_entry_rcu(dev, head, dev_list) {
3747 		if (!dev->bdev) {
3748 			total_errors++;
3749 			continue;
3750 		}
3751 		if (!dev->in_fs_metadata || !dev->writeable)
3752 			continue;
3753 
3754 		btrfs_set_stack_device_generation(dev_item, 0);
3755 		btrfs_set_stack_device_type(dev_item, dev->type);
3756 		btrfs_set_stack_device_id(dev_item, dev->devid);
3757 		btrfs_set_stack_device_total_bytes(dev_item,
3758 						   dev->commit_total_bytes);
3759 		btrfs_set_stack_device_bytes_used(dev_item,
3760 						  dev->commit_bytes_used);
3761 		btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3762 		btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3763 		btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3764 		memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3765 		memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3766 
3767 		flags = btrfs_super_flags(sb);
3768 		btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3769 
3770 		ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3771 		if (ret)
3772 			total_errors++;
3773 	}
3774 	if (total_errors > max_errors) {
3775 		btrfs_err(root->fs_info, "%d errors while writing supers",
3776 		       total_errors);
3777 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3778 
3779 		/* FUA is masked off if unsupported and can't be the reason */
3780 		btrfs_handle_fs_error(root->fs_info, -EIO,
3781 			    "%d errors while writing supers", total_errors);
3782 		return -EIO;
3783 	}
3784 
3785 	total_errors = 0;
3786 	list_for_each_entry_rcu(dev, head, dev_list) {
3787 		if (!dev->bdev)
3788 			continue;
3789 		if (!dev->in_fs_metadata || !dev->writeable)
3790 			continue;
3791 
3792 		ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3793 		if (ret)
3794 			total_errors++;
3795 	}
3796 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3797 	if (total_errors > max_errors) {
3798 		btrfs_handle_fs_error(root->fs_info, -EIO,
3799 			    "%d errors while writing supers", total_errors);
3800 		return -EIO;
3801 	}
3802 	return 0;
3803 }
3804 
write_ctree_super(struct btrfs_trans_handle * trans,struct btrfs_root * root,int max_mirrors)3805 int write_ctree_super(struct btrfs_trans_handle *trans,
3806 		      struct btrfs_root *root, int max_mirrors)
3807 {
3808 	return write_all_supers(root, max_mirrors);
3809 }
3810 
3811 /* Drop a fs root from the radix tree and free it. */
btrfs_drop_and_free_fs_root(struct btrfs_fs_info * fs_info,struct btrfs_root * root)3812 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3813 				  struct btrfs_root *root)
3814 {
3815 	spin_lock(&fs_info->fs_roots_radix_lock);
3816 	radix_tree_delete(&fs_info->fs_roots_radix,
3817 			  (unsigned long)root->root_key.objectid);
3818 	spin_unlock(&fs_info->fs_roots_radix_lock);
3819 
3820 	if (btrfs_root_refs(&root->root_item) == 0)
3821 		synchronize_srcu(&fs_info->subvol_srcu);
3822 
3823 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3824 		btrfs_free_log(NULL, root);
3825 		if (root->reloc_root) {
3826 			free_extent_buffer(root->reloc_root->node);
3827 			free_extent_buffer(root->reloc_root->commit_root);
3828 			btrfs_put_fs_root(root->reloc_root);
3829 			root->reloc_root = NULL;
3830 		}
3831 	}
3832 
3833 	if (root->free_ino_pinned)
3834 		__btrfs_remove_free_space_cache(root->free_ino_pinned);
3835 	if (root->free_ino_ctl)
3836 		__btrfs_remove_free_space_cache(root->free_ino_ctl);
3837 	free_fs_root(root);
3838 }
3839 
free_fs_root(struct btrfs_root * root)3840 static void free_fs_root(struct btrfs_root *root)
3841 {
3842 	iput(root->ino_cache_inode);
3843 	WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3844 	btrfs_free_block_rsv(root, root->orphan_block_rsv);
3845 	root->orphan_block_rsv = NULL;
3846 	if (root->anon_dev)
3847 		free_anon_bdev(root->anon_dev);
3848 	if (root->subv_writers)
3849 		btrfs_free_subvolume_writers(root->subv_writers);
3850 	free_extent_buffer(root->node);
3851 	free_extent_buffer(root->commit_root);
3852 	kfree(root->free_ino_ctl);
3853 	kfree(root->free_ino_pinned);
3854 	kfree(root->name);
3855 	btrfs_put_fs_root(root);
3856 }
3857 
btrfs_free_fs_root(struct btrfs_root * root)3858 void btrfs_free_fs_root(struct btrfs_root *root)
3859 {
3860 	free_fs_root(root);
3861 }
3862 
btrfs_cleanup_fs_roots(struct btrfs_fs_info * fs_info)3863 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3864 {
3865 	u64 root_objectid = 0;
3866 	struct btrfs_root *gang[8];
3867 	int i = 0;
3868 	int err = 0;
3869 	unsigned int ret = 0;
3870 	int index;
3871 
3872 	while (1) {
3873 		index = srcu_read_lock(&fs_info->subvol_srcu);
3874 		ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3875 					     (void **)gang, root_objectid,
3876 					     ARRAY_SIZE(gang));
3877 		if (!ret) {
3878 			srcu_read_unlock(&fs_info->subvol_srcu, index);
3879 			break;
3880 		}
3881 		root_objectid = gang[ret - 1]->root_key.objectid + 1;
3882 
3883 		for (i = 0; i < ret; i++) {
3884 			/* Avoid to grab roots in dead_roots */
3885 			if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3886 				gang[i] = NULL;
3887 				continue;
3888 			}
3889 			/* grab all the search result for later use */
3890 			gang[i] = btrfs_grab_fs_root(gang[i]);
3891 		}
3892 		srcu_read_unlock(&fs_info->subvol_srcu, index);
3893 
3894 		for (i = 0; i < ret; i++) {
3895 			if (!gang[i])
3896 				continue;
3897 			root_objectid = gang[i]->root_key.objectid;
3898 			err = btrfs_orphan_cleanup(gang[i]);
3899 			if (err)
3900 				break;
3901 			btrfs_put_fs_root(gang[i]);
3902 		}
3903 		root_objectid++;
3904 	}
3905 
3906 	/* release the uncleaned roots due to error */
3907 	for (; i < ret; i++) {
3908 		if (gang[i])
3909 			btrfs_put_fs_root(gang[i]);
3910 	}
3911 	return err;
3912 }
3913 
btrfs_commit_super(struct btrfs_root * root)3914 int btrfs_commit_super(struct btrfs_root *root)
3915 {
3916 	struct btrfs_trans_handle *trans;
3917 
3918 	mutex_lock(&root->fs_info->cleaner_mutex);
3919 	btrfs_run_delayed_iputs(root);
3920 	mutex_unlock(&root->fs_info->cleaner_mutex);
3921 	wake_up_process(root->fs_info->cleaner_kthread);
3922 
3923 	/* wait until ongoing cleanup work done */
3924 	down_write(&root->fs_info->cleanup_work_sem);
3925 	up_write(&root->fs_info->cleanup_work_sem);
3926 
3927 	trans = btrfs_join_transaction(root);
3928 	if (IS_ERR(trans))
3929 		return PTR_ERR(trans);
3930 	return btrfs_commit_transaction(trans, root);
3931 }
3932 
close_ctree(struct btrfs_root * root)3933 void close_ctree(struct btrfs_root *root)
3934 {
3935 	struct btrfs_fs_info *fs_info = root->fs_info;
3936 	int ret;
3937 
3938 	set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
3939 
3940 	/* wait for the qgroup rescan worker to stop */
3941 	btrfs_qgroup_wait_for_completion(fs_info, false);
3942 
3943 	/* wait for the uuid_scan task to finish */
3944 	down(&fs_info->uuid_tree_rescan_sem);
3945 	/* avoid complains from lockdep et al., set sem back to initial state */
3946 	up(&fs_info->uuid_tree_rescan_sem);
3947 
3948 	/* pause restriper - we want to resume on mount */
3949 	btrfs_pause_balance(fs_info);
3950 
3951 	btrfs_dev_replace_suspend_for_unmount(fs_info);
3952 
3953 	btrfs_scrub_cancel(fs_info);
3954 
3955 	/* wait for any defraggers to finish */
3956 	wait_event(fs_info->transaction_wait,
3957 		   (atomic_read(&fs_info->defrag_running) == 0));
3958 
3959 	/* clear out the rbtree of defraggable inodes */
3960 	btrfs_cleanup_defrag_inodes(fs_info);
3961 
3962 	cancel_work_sync(&fs_info->async_reclaim_work);
3963 
3964 	if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3965 		/*
3966 		 * If the cleaner thread is stopped and there are
3967 		 * block groups queued for removal, the deletion will be
3968 		 * skipped when we quit the cleaner thread.
3969 		 */
3970 		btrfs_delete_unused_bgs(root->fs_info);
3971 
3972 		ret = btrfs_commit_super(root);
3973 		if (ret)
3974 			btrfs_err(fs_info, "commit super ret %d", ret);
3975 	}
3976 
3977 	if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3978 		btrfs_error_commit_super(root);
3979 
3980 	kthread_stop(fs_info->transaction_kthread);
3981 	kthread_stop(fs_info->cleaner_kthread);
3982 
3983 	set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
3984 
3985 	btrfs_free_qgroup_config(fs_info);
3986 
3987 	if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3988 		btrfs_info(fs_info, "at unmount delalloc count %lld",
3989 		       percpu_counter_sum(&fs_info->delalloc_bytes));
3990 	}
3991 
3992 	btrfs_sysfs_remove_mounted(fs_info);
3993 	btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3994 
3995 	btrfs_free_fs_roots(fs_info);
3996 
3997 	btrfs_put_block_group_cache(fs_info);
3998 
3999 	btrfs_free_block_groups(fs_info);
4000 
4001 	/*
4002 	 * we must make sure there is not any read request to
4003 	 * submit after we stopping all workers.
4004 	 */
4005 	invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4006 	btrfs_stop_all_workers(fs_info);
4007 
4008 	clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4009 	free_root_pointers(fs_info, 1);
4010 
4011 	iput(fs_info->btree_inode);
4012 
4013 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4014 	if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
4015 		btrfsic_unmount(root, fs_info->fs_devices);
4016 #endif
4017 
4018 	btrfs_close_devices(fs_info->fs_devices);
4019 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
4020 
4021 	percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
4022 	percpu_counter_destroy(&fs_info->delalloc_bytes);
4023 	percpu_counter_destroy(&fs_info->bio_counter);
4024 	bdi_destroy(&fs_info->bdi);
4025 	cleanup_srcu_struct(&fs_info->subvol_srcu);
4026 
4027 	btrfs_free_stripe_hash_table(fs_info);
4028 
4029 	__btrfs_free_block_rsv(root->orphan_block_rsv);
4030 	root->orphan_block_rsv = NULL;
4031 
4032 	lock_chunks(root);
4033 	while (!list_empty(&fs_info->pinned_chunks)) {
4034 		struct extent_map *em;
4035 
4036 		em = list_first_entry(&fs_info->pinned_chunks,
4037 				      struct extent_map, list);
4038 		list_del_init(&em->list);
4039 		free_extent_map(em);
4040 	}
4041 	unlock_chunks(root);
4042 }
4043 
btrfs_buffer_uptodate(struct extent_buffer * buf,u64 parent_transid,int atomic)4044 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4045 			  int atomic)
4046 {
4047 	int ret;
4048 	struct inode *btree_inode = buf->pages[0]->mapping->host;
4049 
4050 	ret = extent_buffer_uptodate(buf);
4051 	if (!ret)
4052 		return ret;
4053 
4054 	ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4055 				    parent_transid, atomic);
4056 	if (ret == -EAGAIN)
4057 		return ret;
4058 	return !ret;
4059 }
4060 
btrfs_mark_buffer_dirty(struct extent_buffer * buf)4061 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4062 {
4063 	struct btrfs_root *root;
4064 	u64 transid = btrfs_header_generation(buf);
4065 	int was_dirty;
4066 
4067 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4068 	/*
4069 	 * This is a fast path so only do this check if we have sanity tests
4070 	 * enabled.  Normal people shouldn't be marking dummy buffers as dirty
4071 	 * outside of the sanity tests.
4072 	 */
4073 	if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4074 		return;
4075 #endif
4076 	root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4077 	btrfs_assert_tree_locked(buf);
4078 	if (transid != root->fs_info->generation)
4079 		WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4080 			buf->start, transid, root->fs_info->generation);
4081 	was_dirty = set_extent_buffer_dirty(buf);
4082 	if (!was_dirty)
4083 		__percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4084 				     buf->len,
4085 				     root->fs_info->dirty_metadata_batch);
4086 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4087 	if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4088 		btrfs_print_leaf(root, buf);
4089 		ASSERT(0);
4090 	}
4091 #endif
4092 }
4093 
__btrfs_btree_balance_dirty(struct btrfs_root * root,int flush_delayed)4094 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4095 					int flush_delayed)
4096 {
4097 	/*
4098 	 * looks as though older kernels can get into trouble with
4099 	 * this code, they end up stuck in balance_dirty_pages forever
4100 	 */
4101 	int ret;
4102 
4103 	if (current->flags & PF_MEMALLOC)
4104 		return;
4105 
4106 	if (flush_delayed)
4107 		btrfs_balance_delayed_items(root);
4108 
4109 	ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4110 				     BTRFS_DIRTY_METADATA_THRESH);
4111 	if (ret > 0) {
4112 		balance_dirty_pages_ratelimited(
4113 				   root->fs_info->btree_inode->i_mapping);
4114 	}
4115 }
4116 
btrfs_btree_balance_dirty(struct btrfs_root * root)4117 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4118 {
4119 	__btrfs_btree_balance_dirty(root, 1);
4120 }
4121 
btrfs_btree_balance_dirty_nodelay(struct btrfs_root * root)4122 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4123 {
4124 	__btrfs_btree_balance_dirty(root, 0);
4125 }
4126 
btrfs_read_buffer(struct extent_buffer * buf,u64 parent_transid)4127 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4128 {
4129 	struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4130 	return btree_read_extent_buffer_pages(root, buf, parent_transid);
4131 }
4132 
btrfs_check_super_valid(struct btrfs_fs_info * fs_info,int read_only)4133 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4134 			      int read_only)
4135 {
4136 	struct btrfs_super_block *sb = fs_info->super_copy;
4137 	u64 nodesize = btrfs_super_nodesize(sb);
4138 	u64 sectorsize = btrfs_super_sectorsize(sb);
4139 	int ret = 0;
4140 
4141 	if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4142 		btrfs_err(fs_info, "no valid FS found");
4143 		ret = -EINVAL;
4144 	}
4145 	if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4146 		btrfs_warn(fs_info, "unrecognized super flag: %llu",
4147 				btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4148 	if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4149 		btrfs_err(fs_info, "tree_root level too big: %d >= %d",
4150 				btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4151 		ret = -EINVAL;
4152 	}
4153 	if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4154 		btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
4155 				btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4156 		ret = -EINVAL;
4157 	}
4158 	if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4159 		btrfs_err(fs_info, "log_root level too big: %d >= %d",
4160 				btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4161 		ret = -EINVAL;
4162 	}
4163 
4164 	/*
4165 	 * Check sectorsize and nodesize first, other check will need it.
4166 	 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4167 	 */
4168 	if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4169 	    sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4170 		btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
4171 		ret = -EINVAL;
4172 	}
4173 	/* Only PAGE SIZE is supported yet */
4174 	if (sectorsize != PAGE_SIZE) {
4175 		btrfs_err(fs_info,
4176 			"sectorsize %llu not supported yet, only support %lu",
4177 			sectorsize, PAGE_SIZE);
4178 		ret = -EINVAL;
4179 	}
4180 	if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4181 	    nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4182 		btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
4183 		ret = -EINVAL;
4184 	}
4185 	if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4186 		btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
4187 			  le32_to_cpu(sb->__unused_leafsize), nodesize);
4188 		ret = -EINVAL;
4189 	}
4190 
4191 	/* Root alignment check */
4192 	if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4193 		btrfs_warn(fs_info, "tree_root block unaligned: %llu",
4194 			   btrfs_super_root(sb));
4195 		ret = -EINVAL;
4196 	}
4197 	if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4198 		btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
4199 			   btrfs_super_chunk_root(sb));
4200 		ret = -EINVAL;
4201 	}
4202 	if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4203 		btrfs_warn(fs_info, "log_root block unaligned: %llu",
4204 			   btrfs_super_log_root(sb));
4205 		ret = -EINVAL;
4206 	}
4207 
4208 	if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4209 		btrfs_err(fs_info,
4210 			   "dev_item UUID does not match fsid: %pU != %pU",
4211 			   fs_info->fsid, sb->dev_item.fsid);
4212 		ret = -EINVAL;
4213 	}
4214 
4215 	/*
4216 	 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4217 	 * done later
4218 	 */
4219 	if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4220 		btrfs_err(fs_info, "bytes_used is too small %llu",
4221 			  btrfs_super_bytes_used(sb));
4222 		ret = -EINVAL;
4223 	}
4224 	if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4225 		btrfs_err(fs_info, "invalid stripesize %u",
4226 			  btrfs_super_stripesize(sb));
4227 		ret = -EINVAL;
4228 	}
4229 	if (btrfs_super_num_devices(sb) > (1UL << 31))
4230 		btrfs_warn(fs_info, "suspicious number of devices: %llu",
4231 			   btrfs_super_num_devices(sb));
4232 	if (btrfs_super_num_devices(sb) == 0) {
4233 		btrfs_err(fs_info, "number of devices is 0");
4234 		ret = -EINVAL;
4235 	}
4236 
4237 	if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4238 		btrfs_err(fs_info, "super offset mismatch %llu != %u",
4239 			  btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4240 		ret = -EINVAL;
4241 	}
4242 
4243 	/*
4244 	 * Obvious sys_chunk_array corruptions, it must hold at least one key
4245 	 * and one chunk
4246 	 */
4247 	if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4248 		btrfs_err(fs_info, "system chunk array too big %u > %u",
4249 			  btrfs_super_sys_array_size(sb),
4250 			  BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4251 		ret = -EINVAL;
4252 	}
4253 	if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4254 			+ sizeof(struct btrfs_chunk)) {
4255 		btrfs_err(fs_info, "system chunk array too small %u < %zu",
4256 			  btrfs_super_sys_array_size(sb),
4257 			  sizeof(struct btrfs_disk_key)
4258 			  + sizeof(struct btrfs_chunk));
4259 		ret = -EINVAL;
4260 	}
4261 
4262 	/*
4263 	 * The generation is a global counter, we'll trust it more than the others
4264 	 * but it's still possible that it's the one that's wrong.
4265 	 */
4266 	if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4267 		btrfs_warn(fs_info,
4268 			"suspicious: generation < chunk_root_generation: %llu < %llu",
4269 			btrfs_super_generation(sb),
4270 			btrfs_super_chunk_root_generation(sb));
4271 	if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4272 	    && btrfs_super_cache_generation(sb) != (u64)-1)
4273 		btrfs_warn(fs_info,
4274 			"suspicious: generation < cache_generation: %llu < %llu",
4275 			btrfs_super_generation(sb),
4276 			btrfs_super_cache_generation(sb));
4277 
4278 	return ret;
4279 }
4280 
btrfs_error_commit_super(struct btrfs_root * root)4281 static void btrfs_error_commit_super(struct btrfs_root *root)
4282 {
4283 	mutex_lock(&root->fs_info->cleaner_mutex);
4284 	btrfs_run_delayed_iputs(root);
4285 	mutex_unlock(&root->fs_info->cleaner_mutex);
4286 
4287 	down_write(&root->fs_info->cleanup_work_sem);
4288 	up_write(&root->fs_info->cleanup_work_sem);
4289 
4290 	/* cleanup FS via transaction */
4291 	btrfs_cleanup_transaction(root);
4292 }
4293 
btrfs_destroy_ordered_extents(struct btrfs_root * root)4294 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4295 {
4296 	struct btrfs_ordered_extent *ordered;
4297 
4298 	spin_lock(&root->ordered_extent_lock);
4299 	/*
4300 	 * This will just short circuit the ordered completion stuff which will
4301 	 * make sure the ordered extent gets properly cleaned up.
4302 	 */
4303 	list_for_each_entry(ordered, &root->ordered_extents,
4304 			    root_extent_list)
4305 		set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4306 	spin_unlock(&root->ordered_extent_lock);
4307 }
4308 
btrfs_destroy_all_ordered_extents(struct btrfs_fs_info * fs_info)4309 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4310 {
4311 	struct btrfs_root *root;
4312 	struct list_head splice;
4313 
4314 	INIT_LIST_HEAD(&splice);
4315 
4316 	spin_lock(&fs_info->ordered_root_lock);
4317 	list_splice_init(&fs_info->ordered_roots, &splice);
4318 	while (!list_empty(&splice)) {
4319 		root = list_first_entry(&splice, struct btrfs_root,
4320 					ordered_root);
4321 		list_move_tail(&root->ordered_root,
4322 			       &fs_info->ordered_roots);
4323 
4324 		spin_unlock(&fs_info->ordered_root_lock);
4325 		btrfs_destroy_ordered_extents(root);
4326 
4327 		cond_resched();
4328 		spin_lock(&fs_info->ordered_root_lock);
4329 	}
4330 	spin_unlock(&fs_info->ordered_root_lock);
4331 }
4332 
btrfs_destroy_delayed_refs(struct btrfs_transaction * trans,struct btrfs_root * root)4333 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4334 				      struct btrfs_root *root)
4335 {
4336 	struct rb_node *node;
4337 	struct btrfs_delayed_ref_root *delayed_refs;
4338 	struct btrfs_delayed_ref_node *ref;
4339 	int ret = 0;
4340 
4341 	delayed_refs = &trans->delayed_refs;
4342 
4343 	spin_lock(&delayed_refs->lock);
4344 	if (atomic_read(&delayed_refs->num_entries) == 0) {
4345 		spin_unlock(&delayed_refs->lock);
4346 		btrfs_info(root->fs_info, "delayed_refs has NO entry");
4347 		return ret;
4348 	}
4349 
4350 	while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4351 		struct btrfs_delayed_ref_head *head;
4352 		struct btrfs_delayed_ref_node *tmp;
4353 		bool pin_bytes = false;
4354 
4355 		head = rb_entry(node, struct btrfs_delayed_ref_head,
4356 				href_node);
4357 		if (!mutex_trylock(&head->mutex)) {
4358 			atomic_inc(&head->node.refs);
4359 			spin_unlock(&delayed_refs->lock);
4360 
4361 			mutex_lock(&head->mutex);
4362 			mutex_unlock(&head->mutex);
4363 			btrfs_put_delayed_ref(&head->node);
4364 			spin_lock(&delayed_refs->lock);
4365 			continue;
4366 		}
4367 		spin_lock(&head->lock);
4368 		list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4369 						 list) {
4370 			ref->in_tree = 0;
4371 			list_del(&ref->list);
4372 			atomic_dec(&delayed_refs->num_entries);
4373 			btrfs_put_delayed_ref(ref);
4374 		}
4375 		if (head->must_insert_reserved)
4376 			pin_bytes = true;
4377 		btrfs_free_delayed_extent_op(head->extent_op);
4378 		delayed_refs->num_heads--;
4379 		if (head->processing == 0)
4380 			delayed_refs->num_heads_ready--;
4381 		atomic_dec(&delayed_refs->num_entries);
4382 		head->node.in_tree = 0;
4383 		rb_erase(&head->href_node, &delayed_refs->href_root);
4384 		spin_unlock(&head->lock);
4385 		spin_unlock(&delayed_refs->lock);
4386 		mutex_unlock(&head->mutex);
4387 
4388 		if (pin_bytes)
4389 			btrfs_pin_extent(root, head->node.bytenr,
4390 					 head->node.num_bytes, 1);
4391 		btrfs_put_delayed_ref(&head->node);
4392 		cond_resched();
4393 		spin_lock(&delayed_refs->lock);
4394 	}
4395 
4396 	spin_unlock(&delayed_refs->lock);
4397 
4398 	return ret;
4399 }
4400 
btrfs_destroy_delalloc_inodes(struct btrfs_root * root)4401 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4402 {
4403 	struct btrfs_inode *btrfs_inode;
4404 	struct list_head splice;
4405 
4406 	INIT_LIST_HEAD(&splice);
4407 
4408 	spin_lock(&root->delalloc_lock);
4409 	list_splice_init(&root->delalloc_inodes, &splice);
4410 
4411 	while (!list_empty(&splice)) {
4412 		btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4413 					       delalloc_inodes);
4414 
4415 		list_del_init(&btrfs_inode->delalloc_inodes);
4416 		clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4417 			  &btrfs_inode->runtime_flags);
4418 		spin_unlock(&root->delalloc_lock);
4419 
4420 		btrfs_invalidate_inodes(btrfs_inode->root);
4421 
4422 		spin_lock(&root->delalloc_lock);
4423 	}
4424 
4425 	spin_unlock(&root->delalloc_lock);
4426 }
4427 
btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info * fs_info)4428 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4429 {
4430 	struct btrfs_root *root;
4431 	struct list_head splice;
4432 
4433 	INIT_LIST_HEAD(&splice);
4434 
4435 	spin_lock(&fs_info->delalloc_root_lock);
4436 	list_splice_init(&fs_info->delalloc_roots, &splice);
4437 	while (!list_empty(&splice)) {
4438 		root = list_first_entry(&splice, struct btrfs_root,
4439 					 delalloc_root);
4440 		list_del_init(&root->delalloc_root);
4441 		root = btrfs_grab_fs_root(root);
4442 		BUG_ON(!root);
4443 		spin_unlock(&fs_info->delalloc_root_lock);
4444 
4445 		btrfs_destroy_delalloc_inodes(root);
4446 		btrfs_put_fs_root(root);
4447 
4448 		spin_lock(&fs_info->delalloc_root_lock);
4449 	}
4450 	spin_unlock(&fs_info->delalloc_root_lock);
4451 }
4452 
btrfs_destroy_marked_extents(struct btrfs_root * root,struct extent_io_tree * dirty_pages,int mark)4453 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4454 					struct extent_io_tree *dirty_pages,
4455 					int mark)
4456 {
4457 	int ret;
4458 	struct extent_buffer *eb;
4459 	u64 start = 0;
4460 	u64 end;
4461 
4462 	while (1) {
4463 		ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4464 					    mark, NULL);
4465 		if (ret)
4466 			break;
4467 
4468 		clear_extent_bits(dirty_pages, start, end, mark);
4469 		while (start <= end) {
4470 			eb = btrfs_find_tree_block(root->fs_info, start);
4471 			start += root->nodesize;
4472 			if (!eb)
4473 				continue;
4474 			wait_on_extent_buffer_writeback(eb);
4475 
4476 			if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4477 					       &eb->bflags))
4478 				clear_extent_buffer_dirty(eb);
4479 			free_extent_buffer_stale(eb);
4480 		}
4481 	}
4482 
4483 	return ret;
4484 }
4485 
btrfs_destroy_pinned_extent(struct btrfs_root * root,struct extent_io_tree * pinned_extents)4486 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4487 				       struct extent_io_tree *pinned_extents)
4488 {
4489 	struct extent_io_tree *unpin;
4490 	u64 start;
4491 	u64 end;
4492 	int ret;
4493 	bool loop = true;
4494 
4495 	unpin = pinned_extents;
4496 again:
4497 	while (1) {
4498 		ret = find_first_extent_bit(unpin, 0, &start, &end,
4499 					    EXTENT_DIRTY, NULL);
4500 		if (ret)
4501 			break;
4502 
4503 		clear_extent_dirty(unpin, start, end);
4504 		btrfs_error_unpin_extent_range(root, start, end);
4505 		cond_resched();
4506 	}
4507 
4508 	if (loop) {
4509 		if (unpin == &root->fs_info->freed_extents[0])
4510 			unpin = &root->fs_info->freed_extents[1];
4511 		else
4512 			unpin = &root->fs_info->freed_extents[0];
4513 		loop = false;
4514 		goto again;
4515 	}
4516 
4517 	return 0;
4518 }
4519 
btrfs_cleanup_bg_io(struct btrfs_block_group_cache * cache)4520 static void btrfs_cleanup_bg_io(struct btrfs_block_group_cache *cache)
4521 {
4522 	struct inode *inode;
4523 
4524 	inode = cache->io_ctl.inode;
4525 	if (inode) {
4526 		invalidate_inode_pages2(inode->i_mapping);
4527 		BTRFS_I(inode)->generation = 0;
4528 		cache->io_ctl.inode = NULL;
4529 		iput(inode);
4530 	}
4531 	btrfs_put_block_group(cache);
4532 }
4533 
btrfs_cleanup_dirty_bgs(struct btrfs_transaction * cur_trans,struct btrfs_root * root)4534 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4535 			     struct btrfs_root *root)
4536 {
4537 	struct btrfs_block_group_cache *cache;
4538 
4539 	spin_lock(&cur_trans->dirty_bgs_lock);
4540 	while (!list_empty(&cur_trans->dirty_bgs)) {
4541 		cache = list_first_entry(&cur_trans->dirty_bgs,
4542 					 struct btrfs_block_group_cache,
4543 					 dirty_list);
4544 		if (!cache) {
4545 			btrfs_err(root->fs_info,
4546 				  "orphan block group dirty_bgs list");
4547 			spin_unlock(&cur_trans->dirty_bgs_lock);
4548 			return;
4549 		}
4550 
4551 		if (!list_empty(&cache->io_list)) {
4552 			spin_unlock(&cur_trans->dirty_bgs_lock);
4553 			list_del_init(&cache->io_list);
4554 			btrfs_cleanup_bg_io(cache);
4555 			spin_lock(&cur_trans->dirty_bgs_lock);
4556 		}
4557 
4558 		list_del_init(&cache->dirty_list);
4559 		spin_lock(&cache->lock);
4560 		cache->disk_cache_state = BTRFS_DC_ERROR;
4561 		spin_unlock(&cache->lock);
4562 
4563 		spin_unlock(&cur_trans->dirty_bgs_lock);
4564 		btrfs_put_block_group(cache);
4565 		spin_lock(&cur_trans->dirty_bgs_lock);
4566 	}
4567 	spin_unlock(&cur_trans->dirty_bgs_lock);
4568 
4569 	while (!list_empty(&cur_trans->io_bgs)) {
4570 		cache = list_first_entry(&cur_trans->io_bgs,
4571 					 struct btrfs_block_group_cache,
4572 					 io_list);
4573 		if (!cache) {
4574 			btrfs_err(root->fs_info,
4575 				  "orphan block group on io_bgs list");
4576 			return;
4577 		}
4578 
4579 		list_del_init(&cache->io_list);
4580 		spin_lock(&cache->lock);
4581 		cache->disk_cache_state = BTRFS_DC_ERROR;
4582 		spin_unlock(&cache->lock);
4583 		btrfs_cleanup_bg_io(cache);
4584 	}
4585 }
4586 
btrfs_cleanup_one_transaction(struct btrfs_transaction * cur_trans,struct btrfs_root * root)4587 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4588 				   struct btrfs_root *root)
4589 {
4590 	btrfs_cleanup_dirty_bgs(cur_trans, root);
4591 	ASSERT(list_empty(&cur_trans->dirty_bgs));
4592 	ASSERT(list_empty(&cur_trans->io_bgs));
4593 
4594 	btrfs_destroy_delayed_refs(cur_trans, root);
4595 
4596 	cur_trans->state = TRANS_STATE_COMMIT_START;
4597 	wake_up(&root->fs_info->transaction_blocked_wait);
4598 
4599 	cur_trans->state = TRANS_STATE_UNBLOCKED;
4600 	wake_up(&root->fs_info->transaction_wait);
4601 
4602 	btrfs_destroy_delayed_inodes(root);
4603 	btrfs_assert_delayed_root_empty(root);
4604 
4605 	btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4606 				     EXTENT_DIRTY);
4607 	btrfs_destroy_pinned_extent(root,
4608 				    root->fs_info->pinned_extents);
4609 
4610 	cur_trans->state =TRANS_STATE_COMPLETED;
4611 	wake_up(&cur_trans->commit_wait);
4612 
4613 	/*
4614 	memset(cur_trans, 0, sizeof(*cur_trans));
4615 	kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4616 	*/
4617 }
4618 
btrfs_cleanup_transaction(struct btrfs_root * root)4619 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4620 {
4621 	struct btrfs_transaction *t;
4622 
4623 	mutex_lock(&root->fs_info->transaction_kthread_mutex);
4624 
4625 	spin_lock(&root->fs_info->trans_lock);
4626 	while (!list_empty(&root->fs_info->trans_list)) {
4627 		t = list_first_entry(&root->fs_info->trans_list,
4628 				     struct btrfs_transaction, list);
4629 		if (t->state >= TRANS_STATE_COMMIT_START) {
4630 			atomic_inc(&t->use_count);
4631 			spin_unlock(&root->fs_info->trans_lock);
4632 			btrfs_wait_for_commit(root, t->transid);
4633 			btrfs_put_transaction(t);
4634 			spin_lock(&root->fs_info->trans_lock);
4635 			continue;
4636 		}
4637 		if (t == root->fs_info->running_transaction) {
4638 			t->state = TRANS_STATE_COMMIT_DOING;
4639 			spin_unlock(&root->fs_info->trans_lock);
4640 			/*
4641 			 * We wait for 0 num_writers since we don't hold a trans
4642 			 * handle open currently for this transaction.
4643 			 */
4644 			wait_event(t->writer_wait,
4645 				   atomic_read(&t->num_writers) == 0);
4646 		} else {
4647 			spin_unlock(&root->fs_info->trans_lock);
4648 		}
4649 		btrfs_cleanup_one_transaction(t, root);
4650 
4651 		spin_lock(&root->fs_info->trans_lock);
4652 		if (t == root->fs_info->running_transaction)
4653 			root->fs_info->running_transaction = NULL;
4654 		list_del_init(&t->list);
4655 		spin_unlock(&root->fs_info->trans_lock);
4656 
4657 		btrfs_put_transaction(t);
4658 		trace_btrfs_transaction_commit(root);
4659 		spin_lock(&root->fs_info->trans_lock);
4660 	}
4661 	spin_unlock(&root->fs_info->trans_lock);
4662 	btrfs_destroy_all_ordered_extents(root->fs_info);
4663 	btrfs_destroy_delayed_inodes(root);
4664 	btrfs_assert_delayed_root_empty(root);
4665 	btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4666 	btrfs_destroy_all_delalloc_inodes(root->fs_info);
4667 	mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4668 
4669 	return 0;
4670 }
4671 
4672 static const struct extent_io_ops btree_extent_io_ops = {
4673 	.readpage_end_io_hook = btree_readpage_end_io_hook,
4674 	.readpage_io_failed_hook = btree_io_failed_hook,
4675 	.submit_bio_hook = btree_submit_bio_hook,
4676 	/* note we're sharing with inode.c for the merge bio hook */
4677 	.merge_bio_hook = btrfs_merge_bio_hook,
4678 };
4679