Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c1d7c514 | 2 | |
d1310b2e CM |
3 | #include <linux/bitops.h> |
4 | #include <linux/slab.h> | |
5 | #include <linux/bio.h> | |
6 | #include <linux/mm.h> | |
d1310b2e CM |
7 | #include <linux/pagemap.h> |
8 | #include <linux/page-flags.h> | |
395cb57e | 9 | #include <linux/sched/mm.h> |
d1310b2e CM |
10 | #include <linux/spinlock.h> |
11 | #include <linux/blkdev.h> | |
12 | #include <linux/swap.h> | |
d1310b2e CM |
13 | #include <linux/writeback.h> |
14 | #include <linux/pagevec.h> | |
268bb0ce | 15 | #include <linux/prefetch.h> |
14605409 | 16 | #include <linux/fsverity.h> |
cea62800 | 17 | #include "misc.h" |
d1310b2e | 18 | #include "extent_io.h" |
9c7d3a54 | 19 | #include "extent-io-tree.h" |
d1310b2e | 20 | #include "extent_map.h" |
902b22f3 DW |
21 | #include "ctree.h" |
22 | #include "btrfs_inode.h" | |
4a54c8c1 | 23 | #include "volumes.h" |
21adbd5c | 24 | #include "check-integrity.h" |
0b32f4bb | 25 | #include "locking.h" |
606686ee | 26 | #include "rcu-string.h" |
fe09e16c | 27 | #include "backref.h" |
6af49dbd | 28 | #include "disk-io.h" |
760f991f | 29 | #include "subpage.h" |
d3575156 | 30 | #include "zoned.h" |
0bc09ca1 | 31 | #include "block-group.h" |
2a5232a8 | 32 | #include "compression.h" |
d1310b2e | 33 | |
d1310b2e CM |
34 | static struct kmem_cache *extent_state_cache; |
35 | static struct kmem_cache *extent_buffer_cache; | |
36 | ||
27a3507d FM |
37 | static inline bool extent_state_in_tree(const struct extent_state *state) |
38 | { | |
39 | return !RB_EMPTY_NODE(&state->rb_node); | |
40 | } | |
41 | ||
6d49ba1b | 42 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e | 43 | static LIST_HEAD(states); |
d397712b | 44 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b | 45 | |
3fd63727 JB |
46 | static inline void btrfs_leak_debug_add(spinlock_t *lock, |
47 | struct list_head *new, | |
48 | struct list_head *head) | |
6d49ba1b ES |
49 | { |
50 | unsigned long flags; | |
51 | ||
3fd63727 | 52 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 53 | list_add(new, head); |
3fd63727 | 54 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
55 | } |
56 | ||
3fd63727 JB |
57 | static inline void btrfs_leak_debug_del(spinlock_t *lock, |
58 | struct list_head *entry) | |
6d49ba1b ES |
59 | { |
60 | unsigned long flags; | |
61 | ||
3fd63727 | 62 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 63 | list_del(entry); |
3fd63727 | 64 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
65 | } |
66 | ||
3fd63727 | 67 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 68 | { |
6d49ba1b | 69 | struct extent_buffer *eb; |
3fd63727 | 70 | unsigned long flags; |
6d49ba1b | 71 | |
8c38938c JB |
72 | /* |
73 | * If we didn't get into open_ctree our allocated_ebs will not be | |
74 | * initialized, so just skip this. | |
75 | */ | |
76 | if (!fs_info->allocated_ebs.next) | |
77 | return; | |
78 | ||
b95b78e6 | 79 | WARN_ON(!list_empty(&fs_info->allocated_ebs)); |
3fd63727 JB |
80 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
81 | while (!list_empty(&fs_info->allocated_ebs)) { | |
82 | eb = list_first_entry(&fs_info->allocated_ebs, | |
83 | struct extent_buffer, leak_list); | |
8c38938c JB |
84 | pr_err( |
85 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
86 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
87 | btrfs_header_owner(eb)); | |
33ca832f JB |
88 | list_del(&eb->leak_list); |
89 | kmem_cache_free(extent_buffer_cache, eb); | |
90 | } | |
3fd63727 | 91 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f JB |
92 | } |
93 | ||
94 | static inline void btrfs_extent_state_leak_debug_check(void) | |
95 | { | |
96 | struct extent_state *state; | |
97 | ||
6d49ba1b ES |
98 | while (!list_empty(&states)) { |
99 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 100 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
101 | state->start, state->end, state->state, |
102 | extent_state_in_tree(state), | |
b7ac31b7 | 103 | refcount_read(&state->refs)); |
6d49ba1b ES |
104 | list_del(&state->leak_list); |
105 | kmem_cache_free(extent_state_cache, state); | |
106 | } | |
6d49ba1b | 107 | } |
8d599ae1 | 108 | |
a5dee37d JB |
109 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
110 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 111 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 112 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 113 | { |
65a680f6 NB |
114 | struct inode *inode = tree->private_data; |
115 | u64 isize; | |
116 | ||
117 | if (!inode || !is_data_inode(inode)) | |
118 | return; | |
119 | ||
120 | isize = i_size_read(inode); | |
121 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
122 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
123 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
124 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
125 | } | |
8d599ae1 | 126 | } |
6d49ba1b | 127 | #else |
3fd63727 JB |
128 | #define btrfs_leak_debug_add(lock, new, head) do {} while (0) |
129 | #define btrfs_leak_debug_del(lock, entry) do {} while (0) | |
33ca832f | 130 | #define btrfs_extent_state_leak_debug_check() do {} while (0) |
8d599ae1 | 131 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 132 | #endif |
d1310b2e | 133 | |
d1310b2e CM |
134 | struct tree_entry { |
135 | u64 start; | |
136 | u64 end; | |
d1310b2e CM |
137 | struct rb_node rb_node; |
138 | }; | |
139 | ||
7aab8b32 CH |
140 | /* |
141 | * Structure to record info about the bio being assembled, and other info like | |
142 | * how many bytes are there before stripe/ordered extent boundary. | |
143 | */ | |
144 | struct btrfs_bio_ctrl { | |
145 | struct bio *bio; | |
722c82ac | 146 | int mirror_num; |
0f07003b | 147 | enum btrfs_compression_type compress_type; |
7aab8b32 CH |
148 | u32 len_to_stripe_boundary; |
149 | u32 len_to_oe_boundary; | |
150 | }; | |
151 | ||
d1310b2e | 152 | struct extent_page_data { |
390ed29b | 153 | struct btrfs_bio_ctrl bio_ctrl; |
771ed689 CM |
154 | /* tells writepage not to lock the state bits for this range |
155 | * it still does the unlocking | |
156 | */ | |
ffbd517d CM |
157 | unsigned int extent_locked:1; |
158 | ||
70fd7614 | 159 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 160 | unsigned int sync_io:1; |
d1310b2e CM |
161 | }; |
162 | ||
f97e27e9 | 163 | static int add_extent_changeset(struct extent_state *state, u32 bits, |
d38ed27f QW |
164 | struct extent_changeset *changeset, |
165 | int set) | |
166 | { | |
167 | int ret; | |
168 | ||
169 | if (!changeset) | |
57599c7e | 170 | return 0; |
d38ed27f | 171 | if (set && (state->state & bits) == bits) |
57599c7e | 172 | return 0; |
fefdc557 | 173 | if (!set && (state->state & bits) == 0) |
57599c7e | 174 | return 0; |
d38ed27f | 175 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 176 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 177 | GFP_ATOMIC); |
57599c7e | 178 | return ret; |
d38ed27f QW |
179 | } |
180 | ||
722c82ac | 181 | static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl) |
bb58eb9e | 182 | { |
722c82ac | 183 | struct bio *bio; |
7aa51232 | 184 | struct bio_vec *bv; |
722c82ac CH |
185 | struct inode *inode; |
186 | int mirror_num; | |
187 | ||
188 | if (!bio_ctrl->bio) | |
189 | return; | |
bb58eb9e | 190 | |
722c82ac | 191 | bio = bio_ctrl->bio; |
7aa51232 CH |
192 | bv = bio_first_bvec_all(bio); |
193 | inode = bv->bv_page->mapping->host; | |
722c82ac | 194 | mirror_num = bio_ctrl->mirror_num; |
bb58eb9e | 195 | |
e0eefe07 QW |
196 | /* Caller should ensure the bio has at least some range added */ |
197 | ASSERT(bio->bi_iter.bi_size); | |
c9583ada | 198 | |
7aa51232 | 199 | btrfs_bio(bio)->file_offset = page_offset(bv->bv_page) + bv->bv_offset; |
bb58eb9e | 200 | |
c93104e7 CH |
201 | if (!is_data_inode(inode)) |
202 | btrfs_submit_metadata_bio(inode, bio, mirror_num); | |
203 | else if (btrfs_op(bio) == BTRFS_MAP_WRITE) | |
204 | btrfs_submit_data_write_bio(inode, bio, mirror_num); | |
908930f3 | 205 | else |
722c82ac CH |
206 | btrfs_submit_data_read_bio(inode, bio, mirror_num, |
207 | bio_ctrl->compress_type); | |
390ed29b | 208 | |
917f32a2 | 209 | /* The bio is owned by the end_io handler now */ |
722c82ac | 210 | bio_ctrl->bio = NULL; |
3065976b QW |
211 | } |
212 | ||
f4340622 | 213 | /* |
9845e5dd | 214 | * Submit or fail the current bio in an extent_page_data structure. |
f4340622 | 215 | */ |
9845e5dd | 216 | static void submit_write_bio(struct extent_page_data *epd, int ret) |
bb58eb9e | 217 | { |
390ed29b | 218 | struct bio *bio = epd->bio_ctrl.bio; |
bb58eb9e | 219 | |
9845e5dd CH |
220 | if (!bio) |
221 | return; | |
222 | ||
223 | if (ret) { | |
224 | ASSERT(ret < 0); | |
917f32a2 CH |
225 | btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); |
226 | /* The bio is owned by the end_io handler now */ | |
390ed29b | 227 | epd->bio_ctrl.bio = NULL; |
9845e5dd | 228 | } else { |
722c82ac | 229 | submit_one_bio(&epd->bio_ctrl); |
bb58eb9e QW |
230 | } |
231 | } | |
e2932ee0 | 232 | |
d45cfb88 | 233 | int __init extent_io_init(void) |
d1310b2e | 234 | { |
837e1972 | 235 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 236 | sizeof(struct extent_state), 0, |
fba4b697 | 237 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
238 | if (!extent_state_cache) |
239 | return -ENOMEM; | |
240 | ||
837e1972 | 241 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 242 | sizeof(struct extent_buffer), 0, |
fba4b697 | 243 | SLAB_MEM_SPREAD, NULL); |
d45cfb88 CH |
244 | if (!extent_buffer_cache) { |
245 | kmem_cache_destroy(extent_state_cache); | |
6f0d04f8 | 246 | return -ENOMEM; |
d45cfb88 | 247 | } |
b208c2f7 | 248 | |
d1310b2e | 249 | return 0; |
d1310b2e CM |
250 | } |
251 | ||
e67c718b | 252 | void __cold extent_io_exit(void) |
d1310b2e | 253 | { |
8c0a8537 KS |
254 | /* |
255 | * Make sure all delayed rcu free are flushed before we | |
256 | * destroy caches. | |
257 | */ | |
258 | rcu_barrier(); | |
5598e900 | 259 | kmem_cache_destroy(extent_buffer_cache); |
d45cfb88 CH |
260 | btrfs_extent_state_leak_debug_check(); |
261 | kmem_cache_destroy(extent_state_cache); | |
d1310b2e CM |
262 | } |
263 | ||
41a2ee75 JB |
264 | /* |
265 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
266 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
267 | * the tree lock and get the inode lock when setting delalloc. These two things | |
268 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
269 | * two locking patterns mixed up. | |
270 | */ | |
271 | static struct lock_class_key file_extent_tree_class; | |
272 | ||
c258d6e3 | 273 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
274 | struct extent_io_tree *tree, unsigned int owner, |
275 | void *private_data) | |
d1310b2e | 276 | { |
c258d6e3 | 277 | tree->fs_info = fs_info; |
6bef4d31 | 278 | tree->state = RB_ROOT; |
d1310b2e | 279 | tree->dirty_bytes = 0; |
70dec807 | 280 | spin_lock_init(&tree->lock); |
c6100a4b | 281 | tree->private_data = private_data; |
43eb5f29 | 282 | tree->owner = owner; |
41a2ee75 JB |
283 | if (owner == IO_TREE_INODE_FILE_EXTENT) |
284 | lockdep_set_class(&tree->lock, &file_extent_tree_class); | |
d1310b2e | 285 | } |
d1310b2e | 286 | |
41e7acd3 NB |
287 | void extent_io_tree_release(struct extent_io_tree *tree) |
288 | { | |
289 | spin_lock(&tree->lock); | |
290 | /* | |
291 | * Do a single barrier for the waitqueue_active check here, the state | |
292 | * of the waitqueue should not change once extent_io_tree_release is | |
293 | * called. | |
294 | */ | |
295 | smp_mb(); | |
296 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
297 | struct rb_node *node; | |
298 | struct extent_state *state; | |
299 | ||
300 | node = rb_first(&tree->state); | |
301 | state = rb_entry(node, struct extent_state, rb_node); | |
302 | rb_erase(&state->rb_node, &tree->state); | |
303 | RB_CLEAR_NODE(&state->rb_node); | |
304 | /* | |
305 | * btree io trees aren't supposed to have tasks waiting for | |
306 | * changes in the flags of extent states ever. | |
307 | */ | |
308 | ASSERT(!waitqueue_active(&state->wq)); | |
309 | free_extent_state(state); | |
310 | ||
311 | cond_resched_lock(&tree->lock); | |
312 | } | |
313 | spin_unlock(&tree->lock); | |
314 | } | |
315 | ||
b2950863 | 316 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
317 | { |
318 | struct extent_state *state; | |
d1310b2e | 319 | |
3ba7ab22 MH |
320 | /* |
321 | * The given mask might be not appropriate for the slab allocator, | |
322 | * drop the unsupported bits | |
323 | */ | |
324 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 325 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 326 | if (!state) |
d1310b2e CM |
327 | return state; |
328 | state->state = 0; | |
47dc196a | 329 | state->failrec = NULL; |
27a3507d | 330 | RB_CLEAR_NODE(&state->rb_node); |
3fd63727 | 331 | btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states); |
b7ac31b7 | 332 | refcount_set(&state->refs, 1); |
d1310b2e | 333 | init_waitqueue_head(&state->wq); |
143bede5 | 334 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
335 | return state; |
336 | } | |
d1310b2e | 337 | |
4845e44f | 338 | void free_extent_state(struct extent_state *state) |
d1310b2e | 339 | { |
d1310b2e CM |
340 | if (!state) |
341 | return; | |
b7ac31b7 | 342 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 343 | WARN_ON(extent_state_in_tree(state)); |
3fd63727 | 344 | btrfs_leak_debug_del(&leak_lock, &state->leak_list); |
143bede5 | 345 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
346 | kmem_cache_free(extent_state_cache, state); |
347 | } | |
348 | } | |
d1310b2e | 349 | |
8666e638 | 350 | /** |
3bed2da1 NB |
351 | * Search @tree for an entry that contains @offset. Such entry would have |
352 | * entry->start <= offset && entry->end >= offset. | |
8666e638 | 353 | * |
3bed2da1 NB |
354 | * @tree: the tree to search |
355 | * @offset: offset that should fall within an entry in @tree | |
9db33891 | 356 | * @node_ret: pointer where new node should be anchored (used when inserting an |
3bed2da1 NB |
357 | * entry in the tree) |
358 | * @parent_ret: points to entry which would have been the parent of the entry, | |
8666e638 NB |
359 | * containing @offset |
360 | * | |
9db33891 DS |
361 | * Return a pointer to the entry that contains @offset byte address and don't change |
362 | * @node_ret and @parent_ret. | |
363 | * | |
364 | * If no such entry exists, return pointer to entry that ends before @offset | |
365 | * and fill parameters @node_ret and @parent_ret, ie. does not return NULL. | |
8666e638 | 366 | */ |
9db33891 DS |
367 | static inline struct rb_node *tree_search_for_insert(struct extent_io_tree *tree, |
368 | u64 offset, | |
369 | struct rb_node ***node_ret, | |
370 | struct rb_node **parent_ret) | |
d1310b2e | 371 | { |
80ea96b1 | 372 | struct rb_root *root = &tree->state; |
9db33891 | 373 | struct rb_node **node = &root->rb_node; |
d1310b2e | 374 | struct rb_node *prev = NULL; |
d1310b2e | 375 | struct tree_entry *entry; |
d1310b2e | 376 | |
9db33891 DS |
377 | while (*node) { |
378 | prev = *node; | |
12cfbad9 | 379 | entry = rb_entry(prev, struct tree_entry, rb_node); |
d1310b2e CM |
380 | |
381 | if (offset < entry->start) | |
9db33891 | 382 | node = &(*node)->rb_left; |
d1310b2e | 383 | else if (offset > entry->end) |
9db33891 | 384 | node = &(*node)->rb_right; |
d397712b | 385 | else |
9db33891 | 386 | return *node; |
d1310b2e CM |
387 | } |
388 | ||
9db33891 DS |
389 | if (node_ret) |
390 | *node_ret = node; | |
12cfbad9 FDBM |
391 | if (parent_ret) |
392 | *parent_ret = prev; | |
393 | ||
9db33891 DS |
394 | /* Search neighbors until we find the first one past the end */ |
395 | while (prev && offset > entry->end) { | |
396 | prev = rb_next(prev); | |
397 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
398 | } |
399 | ||
9db33891 | 400 | return prev; |
d1310b2e CM |
401 | } |
402 | ||
9db33891 DS |
403 | /* |
404 | * Inexact rb-tree search, return the next entry if @offset is not found | |
405 | */ | |
406 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, u64 offset) | |
d1310b2e | 407 | { |
9db33891 | 408 | return tree_search_for_insert(tree, offset, NULL, NULL); |
d1310b2e CM |
409 | } |
410 | ||
9db33891 DS |
411 | /** |
412 | * Search offset in the tree or fill neighbor rbtree node pointers. | |
413 | * | |
414 | * @tree: the tree to search | |
415 | * @offset: offset that should fall within an entry in @tree | |
416 | * @next_ret: pointer to the first entry whose range ends after @offset | |
417 | * @prev_ret: pointer to the first entry whose range begins before @offset | |
418 | * | |
419 | * Return a pointer to the entry that contains @offset byte address. If no | |
420 | * such entry exists, then return NULL and fill @prev_ret and @next_ret. | |
421 | * Otherwise return the found entry and other pointers are left untouched. | |
422 | */ | |
423 | static struct rb_node *tree_search_prev_next(struct extent_io_tree *tree, | |
424 | u64 offset, | |
425 | struct rb_node **prev_ret, | |
426 | struct rb_node **next_ret) | |
12cfbad9 | 427 | { |
bebb22c1 DS |
428 | struct rb_root *root = &tree->state; |
429 | struct rb_node **node = &root->rb_node; | |
430 | struct rb_node *prev = NULL; | |
9db33891 | 431 | struct rb_node *orig_prev = NULL; |
bebb22c1 DS |
432 | struct tree_entry *entry; |
433 | ||
9db33891 DS |
434 | ASSERT(prev_ret); |
435 | ASSERT(next_ret); | |
436 | ||
bebb22c1 DS |
437 | while (*node) { |
438 | prev = *node; | |
439 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
440 | ||
441 | if (offset < entry->start) | |
442 | node = &(*node)->rb_left; | |
443 | else if (offset > entry->end) | |
444 | node = &(*node)->rb_right; | |
445 | else | |
446 | return *node; | |
447 | } | |
448 | ||
9db33891 | 449 | orig_prev = prev; |
bebb22c1 DS |
450 | while (prev && offset > entry->end) { |
451 | prev = rb_next(prev); | |
452 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
453 | } | |
9db33891 DS |
454 | *next_ret = prev; |
455 | prev = orig_prev; | |
bebb22c1 | 456 | |
9db33891 DS |
457 | entry = rb_entry(prev, struct tree_entry, rb_node); |
458 | while (prev && offset < entry->start) { | |
459 | prev = rb_prev(prev); | |
460 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
461 | } | |
462 | *prev_ret = prev; | |
12cfbad9 | 463 | |
9db33891 | 464 | return NULL; |
12cfbad9 FDBM |
465 | } |
466 | ||
d1310b2e CM |
467 | /* |
468 | * utility function to look for merge candidates inside a given range. | |
469 | * Any extents with matching state are merged together into a single | |
470 | * extent in the tree. Extents with EXTENT_IO in their state field | |
471 | * are not merged because the end_io handlers need to be able to do | |
472 | * operations on them without sleeping (or doing allocations/splits). | |
473 | * | |
474 | * This should be called with the tree lock held. | |
475 | */ | |
1bf85046 JM |
476 | static void merge_state(struct extent_io_tree *tree, |
477 | struct extent_state *state) | |
d1310b2e CM |
478 | { |
479 | struct extent_state *other; | |
480 | struct rb_node *other_node; | |
481 | ||
8882679e | 482 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 483 | return; |
d1310b2e CM |
484 | |
485 | other_node = rb_prev(&state->rb_node); | |
486 | if (other_node) { | |
487 | other = rb_entry(other_node, struct extent_state, rb_node); | |
488 | if (other->end == state->start - 1 && | |
489 | other->state == state->state) { | |
5c848198 NB |
490 | if (tree->private_data && |
491 | is_data_inode(tree->private_data)) | |
492 | btrfs_merge_delalloc_extent(tree->private_data, | |
493 | state, other); | |
d1310b2e | 494 | state->start = other->start; |
d1310b2e | 495 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 496 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
497 | free_extent_state(other); |
498 | } | |
499 | } | |
500 | other_node = rb_next(&state->rb_node); | |
501 | if (other_node) { | |
502 | other = rb_entry(other_node, struct extent_state, rb_node); | |
503 | if (other->start == state->end + 1 && | |
504 | other->state == state->state) { | |
5c848198 NB |
505 | if (tree->private_data && |
506 | is_data_inode(tree->private_data)) | |
507 | btrfs_merge_delalloc_extent(tree->private_data, | |
508 | state, other); | |
df98b6e2 | 509 | state->end = other->end; |
df98b6e2 | 510 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 511 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 512 | free_extent_state(other); |
d1310b2e CM |
513 | } |
514 | } | |
d1310b2e CM |
515 | } |
516 | ||
3150b699 | 517 | static void set_state_bits(struct extent_io_tree *tree, |
6d92b304 | 518 | struct extent_state *state, u32 bits, |
d38ed27f | 519 | struct extent_changeset *changeset); |
3150b699 | 520 | |
d1310b2e CM |
521 | /* |
522 | * insert an extent_state struct into the tree. 'bits' are set on the | |
523 | * struct before it is inserted. | |
524 | * | |
525 | * This may return -EEXIST if the extent is already there, in which case the | |
526 | * state struct is freed. | |
527 | * | |
528 | * The tree lock is not taken internally. This is a utility function and | |
529 | * probably isn't what you want to call (see set/clear_extent_bit). | |
530 | */ | |
531 | static int insert_state(struct extent_io_tree *tree, | |
cee51268 | 532 | struct extent_state *state, |
6d92b304 | 533 | u32 bits, struct extent_changeset *changeset) |
d1310b2e | 534 | { |
c7e118cf DS |
535 | struct rb_node **node; |
536 | struct rb_node *parent; | |
cee51268 | 537 | const u64 end = state->end; |
9ed74f2d | 538 | |
d38ed27f | 539 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 540 | |
c7e118cf DS |
541 | node = &tree->state.rb_node; |
542 | while (*node) { | |
543 | struct tree_entry *entry; | |
544 | ||
545 | parent = *node; | |
546 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
547 | ||
548 | if (end < entry->start) { | |
549 | node = &(*node)->rb_left; | |
550 | } else if (end > entry->end) { | |
551 | node = &(*node)->rb_right; | |
552 | } else { | |
553 | btrfs_err(tree->fs_info, | |
554 | "found node %llu %llu on insert of %llu %llu", | |
cee51268 | 555 | entry->start, entry->end, state->start, end); |
c7e118cf DS |
556 | return -EEXIST; |
557 | } | |
d1310b2e | 558 | } |
c7e118cf | 559 | |
c7e118cf DS |
560 | rb_link_node(&state->rb_node, parent, node); |
561 | rb_insert_color(&state->rb_node, &tree->state); | |
562 | ||
d1310b2e CM |
563 | merge_state(tree, state); |
564 | return 0; | |
565 | } | |
566 | ||
fb8f07d2 DS |
567 | /* |
568 | * Insert state to @tree to the location given by @node and @parent. | |
569 | */ | |
570 | static void insert_state_fast(struct extent_io_tree *tree, | |
571 | struct extent_state *state, struct rb_node **node, | |
572 | struct rb_node *parent, unsigned bits, | |
573 | struct extent_changeset *changeset) | |
574 | { | |
575 | set_state_bits(tree, state, bits, changeset); | |
576 | rb_link_node(&state->rb_node, parent, node); | |
577 | rb_insert_color(&state->rb_node, &tree->state); | |
578 | merge_state(tree, state); | |
579 | } | |
580 | ||
d1310b2e CM |
581 | /* |
582 | * split a given extent state struct in two, inserting the preallocated | |
583 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
584 | * offset inside 'orig' where it should be split. | |
585 | * | |
586 | * Before calling, | |
587 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
588 | * are two extent state structs in the tree: | |
589 | * prealloc: [orig->start, split - 1] | |
590 | * orig: [ split, orig->end ] | |
591 | * | |
592 | * The tree locks are not taken by this function. They need to be held | |
593 | * by the caller. | |
594 | */ | |
595 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
596 | struct extent_state *prealloc, u64 split) | |
597 | { | |
12c9cdda DS |
598 | struct rb_node *parent = NULL; |
599 | struct rb_node **node; | |
9ed74f2d | 600 | |
abbb55f4 NB |
601 | if (tree->private_data && is_data_inode(tree->private_data)) |
602 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 603 | |
d1310b2e CM |
604 | prealloc->start = orig->start; |
605 | prealloc->end = split - 1; | |
606 | prealloc->state = orig->state; | |
607 | orig->start = split; | |
608 | ||
12c9cdda DS |
609 | parent = &orig->rb_node; |
610 | node = &parent; | |
611 | while (*node) { | |
612 | struct tree_entry *entry; | |
613 | ||
614 | parent = *node; | |
615 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
616 | ||
617 | if (prealloc->end < entry->start) { | |
618 | node = &(*node)->rb_left; | |
619 | } else if (prealloc->end > entry->end) { | |
620 | node = &(*node)->rb_right; | |
621 | } else { | |
622 | free_extent_state(prealloc); | |
623 | return -EEXIST; | |
624 | } | |
d1310b2e | 625 | } |
12c9cdda DS |
626 | |
627 | rb_link_node(&prealloc->rb_node, parent, node); | |
628 | rb_insert_color(&prealloc->rb_node, &tree->state); | |
629 | ||
d1310b2e CM |
630 | return 0; |
631 | } | |
632 | ||
cdc6a395 LZ |
633 | static struct extent_state *next_state(struct extent_state *state) |
634 | { | |
635 | struct rb_node *next = rb_next(&state->rb_node); | |
636 | if (next) | |
637 | return rb_entry(next, struct extent_state, rb_node); | |
638 | else | |
639 | return NULL; | |
640 | } | |
641 | ||
d1310b2e CM |
642 | /* |
643 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 644 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
645 | * |
646 | * If no bits are set on the state struct after clearing things, the | |
647 | * struct is freed and removed from the tree | |
648 | */ | |
cdc6a395 LZ |
649 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
650 | struct extent_state *state, | |
6d92b304 | 651 | u32 bits, int wake, |
fefdc557 | 652 | struct extent_changeset *changeset) |
d1310b2e | 653 | { |
cdc6a395 | 654 | struct extent_state *next; |
6d92b304 | 655 | u32 bits_to_clear = bits & ~EXTENT_CTLBITS; |
57599c7e | 656 | int ret; |
d1310b2e | 657 | |
0ca1f7ce | 658 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
659 | u64 range = state->end - state->start + 1; |
660 | WARN_ON(range > tree->dirty_bytes); | |
661 | tree->dirty_bytes -= range; | |
662 | } | |
a36bb5f9 NB |
663 | |
664 | if (tree->private_data && is_data_inode(tree->private_data)) | |
665 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
666 | ||
57599c7e DS |
667 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
668 | BUG_ON(ret < 0); | |
32c00aff | 669 | state->state &= ~bits_to_clear; |
d1310b2e CM |
670 | if (wake) |
671 | wake_up(&state->wq); | |
0ca1f7ce | 672 | if (state->state == 0) { |
cdc6a395 | 673 | next = next_state(state); |
27a3507d | 674 | if (extent_state_in_tree(state)) { |
d1310b2e | 675 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 676 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
677 | free_extent_state(state); |
678 | } else { | |
679 | WARN_ON(1); | |
680 | } | |
681 | } else { | |
682 | merge_state(tree, state); | |
cdc6a395 | 683 | next = next_state(state); |
d1310b2e | 684 | } |
cdc6a395 | 685 | return next; |
d1310b2e CM |
686 | } |
687 | ||
8233767a XG |
688 | static struct extent_state * |
689 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
690 | { | |
691 | if (!prealloc) | |
692 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
693 | ||
694 | return prealloc; | |
695 | } | |
696 | ||
48a3b636 | 697 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 698 | { |
29b665cc | 699 | btrfs_panic(tree->fs_info, err, |
05912a3c | 700 | "locking error: extent tree was modified by another thread while locked"); |
c2d904e0 JM |
701 | } |
702 | ||
d1310b2e CM |
703 | /* |
704 | * clear some bits on a range in the tree. This may require splitting | |
705 | * or inserting elements in the tree, so the gfp mask is used to | |
706 | * indicate which allocations or sleeping are allowed. | |
707 | * | |
708 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
709 | * the given range from the tree regardless of state (ie for truncate). | |
710 | * | |
711 | * the range [start, end] is inclusive. | |
712 | * | |
6763af84 | 713 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 714 | */ |
66b0c887 | 715 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 QW |
716 | u32 bits, int wake, int delete, |
717 | struct extent_state **cached_state, | |
718 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
719 | { |
720 | struct extent_state *state; | |
2c64c53d | 721 | struct extent_state *cached; |
d1310b2e CM |
722 | struct extent_state *prealloc = NULL; |
723 | struct rb_node *node; | |
5c939df5 | 724 | u64 last_end; |
d1310b2e | 725 | int err; |
2ac55d41 | 726 | int clear = 0; |
d1310b2e | 727 | |
a5dee37d | 728 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 729 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 730 | |
7ee9e440 JB |
731 | if (bits & EXTENT_DELALLOC) |
732 | bits |= EXTENT_NORESERVE; | |
733 | ||
0ca1f7ce YZ |
734 | if (delete) |
735 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 736 | |
8882679e | 737 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 738 | clear = 1; |
d1310b2e | 739 | again: |
d0164adc | 740 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
741 | /* |
742 | * Don't care for allocation failure here because we might end | |
743 | * up not needing the pre-allocated extent state at all, which | |
744 | * is the case if we only have in the tree extent states that | |
745 | * cover our input range and don't cover too any other range. | |
746 | * If we end up needing a new extent state we allocate it later. | |
747 | */ | |
d1310b2e | 748 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
749 | } |
750 | ||
cad321ad | 751 | spin_lock(&tree->lock); |
2c64c53d CM |
752 | if (cached_state) { |
753 | cached = *cached_state; | |
2ac55d41 JB |
754 | |
755 | if (clear) { | |
756 | *cached_state = NULL; | |
757 | cached_state = NULL; | |
758 | } | |
759 | ||
27a3507d FM |
760 | if (cached && extent_state_in_tree(cached) && |
761 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 762 | if (clear) |
b7ac31b7 | 763 | refcount_dec(&cached->refs); |
2c64c53d | 764 | state = cached; |
42daec29 | 765 | goto hit_next; |
2c64c53d | 766 | } |
2ac55d41 JB |
767 | if (clear) |
768 | free_extent_state(cached); | |
2c64c53d | 769 | } |
d1310b2e CM |
770 | /* |
771 | * this search will find the extents that end after | |
772 | * our range starts | |
773 | */ | |
80ea96b1 | 774 | node = tree_search(tree, start); |
d1310b2e CM |
775 | if (!node) |
776 | goto out; | |
777 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 778 | hit_next: |
d1310b2e CM |
779 | if (state->start > end) |
780 | goto out; | |
781 | WARN_ON(state->end < start); | |
5c939df5 | 782 | last_end = state->end; |
d1310b2e | 783 | |
0449314a | 784 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
785 | if (!(state->state & bits)) { |
786 | state = next_state(state); | |
0449314a | 787 | goto next; |
cdc6a395 | 788 | } |
0449314a | 789 | |
d1310b2e CM |
790 | /* |
791 | * | ---- desired range ---- | | |
792 | * | state | or | |
793 | * | ------------- state -------------- | | |
794 | * | |
795 | * We need to split the extent we found, and may flip | |
796 | * bits on second half. | |
797 | * | |
798 | * If the extent we found extends past our range, we | |
799 | * just split and search again. It'll get split again | |
800 | * the next time though. | |
801 | * | |
802 | * If the extent we found is inside our range, we clear | |
803 | * the desired bit on it. | |
804 | */ | |
805 | ||
806 | if (state->start < start) { | |
8233767a XG |
807 | prealloc = alloc_extent_state_atomic(prealloc); |
808 | BUG_ON(!prealloc); | |
d1310b2e | 809 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
810 | if (err) |
811 | extent_io_tree_panic(tree, err); | |
812 | ||
d1310b2e CM |
813 | prealloc = NULL; |
814 | if (err) | |
815 | goto out; | |
816 | if (state->end <= end) { | |
6d92b304 | 817 | state = clear_state_bit(tree, state, bits, wake, changeset); |
d1ac6e41 | 818 | goto next; |
d1310b2e CM |
819 | } |
820 | goto search_again; | |
821 | } | |
822 | /* | |
823 | * | ---- desired range ---- | | |
824 | * | state | | |
825 | * We need to split the extent, and clear the bit | |
826 | * on the first half | |
827 | */ | |
828 | if (state->start <= end && state->end > end) { | |
8233767a XG |
829 | prealloc = alloc_extent_state_atomic(prealloc); |
830 | BUG_ON(!prealloc); | |
d1310b2e | 831 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
832 | if (err) |
833 | extent_io_tree_panic(tree, err); | |
834 | ||
d1310b2e CM |
835 | if (wake) |
836 | wake_up(&state->wq); | |
42daec29 | 837 | |
6d92b304 | 838 | clear_state_bit(tree, prealloc, bits, wake, changeset); |
9ed74f2d | 839 | |
d1310b2e CM |
840 | prealloc = NULL; |
841 | goto out; | |
842 | } | |
42daec29 | 843 | |
6d92b304 | 844 | state = clear_state_bit(tree, state, bits, wake, changeset); |
0449314a | 845 | next: |
5c939df5 YZ |
846 | if (last_end == (u64)-1) |
847 | goto out; | |
848 | start = last_end + 1; | |
cdc6a395 | 849 | if (start <= end && state && !need_resched()) |
692e5759 | 850 | goto hit_next; |
d1310b2e CM |
851 | |
852 | search_again: | |
853 | if (start > end) | |
854 | goto out; | |
cad321ad | 855 | spin_unlock(&tree->lock); |
d0164adc | 856 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
857 | cond_resched(); |
858 | goto again; | |
7ab5cb2a DS |
859 | |
860 | out: | |
861 | spin_unlock(&tree->lock); | |
862 | if (prealloc) | |
863 | free_extent_state(prealloc); | |
864 | ||
865 | return 0; | |
866 | ||
d1310b2e | 867 | } |
d1310b2e | 868 | |
143bede5 JM |
869 | static void wait_on_state(struct extent_io_tree *tree, |
870 | struct extent_state *state) | |
641f5219 CH |
871 | __releases(tree->lock) |
872 | __acquires(tree->lock) | |
d1310b2e CM |
873 | { |
874 | DEFINE_WAIT(wait); | |
875 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 876 | spin_unlock(&tree->lock); |
d1310b2e | 877 | schedule(); |
cad321ad | 878 | spin_lock(&tree->lock); |
d1310b2e | 879 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
880 | } |
881 | ||
882 | /* | |
883 | * waits for one or more bits to clear on a range in the state tree. | |
884 | * The range [start, end] is inclusive. | |
885 | * The tree lock is taken by this function | |
886 | */ | |
41074888 | 887 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 888 | u32 bits) |
d1310b2e CM |
889 | { |
890 | struct extent_state *state; | |
891 | struct rb_node *node; | |
892 | ||
a5dee37d | 893 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 894 | |
cad321ad | 895 | spin_lock(&tree->lock); |
d1310b2e CM |
896 | again: |
897 | while (1) { | |
898 | /* | |
899 | * this search will find all the extents that end after | |
900 | * our range starts | |
901 | */ | |
80ea96b1 | 902 | node = tree_search(tree, start); |
c50d3e71 | 903 | process_node: |
d1310b2e CM |
904 | if (!node) |
905 | break; | |
906 | ||
907 | state = rb_entry(node, struct extent_state, rb_node); | |
908 | ||
909 | if (state->start > end) | |
910 | goto out; | |
911 | ||
912 | if (state->state & bits) { | |
913 | start = state->start; | |
b7ac31b7 | 914 | refcount_inc(&state->refs); |
d1310b2e CM |
915 | wait_on_state(tree, state); |
916 | free_extent_state(state); | |
917 | goto again; | |
918 | } | |
919 | start = state->end + 1; | |
920 | ||
921 | if (start > end) | |
922 | break; | |
923 | ||
c50d3e71 FM |
924 | if (!cond_resched_lock(&tree->lock)) { |
925 | node = rb_next(node); | |
926 | goto process_node; | |
927 | } | |
d1310b2e CM |
928 | } |
929 | out: | |
cad321ad | 930 | spin_unlock(&tree->lock); |
d1310b2e | 931 | } |
d1310b2e | 932 | |
1bf85046 | 933 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 934 | struct extent_state *state, |
6d92b304 | 935 | u32 bits, struct extent_changeset *changeset) |
d1310b2e | 936 | { |
6d92b304 | 937 | u32 bits_to_set = bits & ~EXTENT_CTLBITS; |
57599c7e | 938 | int ret; |
9ed74f2d | 939 | |
e06a1fc9 NB |
940 | if (tree->private_data && is_data_inode(tree->private_data)) |
941 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
942 | ||
0ca1f7ce | 943 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
944 | u64 range = state->end - state->start + 1; |
945 | tree->dirty_bytes += range; | |
946 | } | |
57599c7e DS |
947 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
948 | BUG_ON(ret < 0); | |
0ca1f7ce | 949 | state->state |= bits_to_set; |
d1310b2e CM |
950 | } |
951 | ||
e38e2ed7 FM |
952 | static void cache_state_if_flags(struct extent_state *state, |
953 | struct extent_state **cached_ptr, | |
9ee49a04 | 954 | unsigned flags) |
2c64c53d CM |
955 | { |
956 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 957 | if (!flags || (state->state & flags)) { |
2c64c53d | 958 | *cached_ptr = state; |
b7ac31b7 | 959 | refcount_inc(&state->refs); |
2c64c53d CM |
960 | } |
961 | } | |
962 | } | |
963 | ||
e38e2ed7 FM |
964 | static void cache_state(struct extent_state *state, |
965 | struct extent_state **cached_ptr) | |
966 | { | |
967 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 968 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
969 | } |
970 | ||
d1310b2e | 971 | /* |
1edbb734 CM |
972 | * set some bits on a range in the tree. This may require allocations or |
973 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 974 | * |
1edbb734 CM |
975 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
976 | * part of the range already has the desired bits set. The start of the | |
977 | * existing range is returned in failed_start in this case. | |
d1310b2e | 978 | * |
1edbb734 | 979 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 980 | */ |
f97e27e9 QW |
981 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits, |
982 | u32 exclusive_bits, u64 *failed_start, | |
1cab5e72 NB |
983 | struct extent_state **cached_state, gfp_t mask, |
984 | struct extent_changeset *changeset) | |
d1310b2e CM |
985 | { |
986 | struct extent_state *state; | |
987 | struct extent_state *prealloc = NULL; | |
988 | struct rb_node *node; | |
12cfbad9 FDBM |
989 | struct rb_node **p; |
990 | struct rb_node *parent; | |
d1310b2e | 991 | int err = 0; |
d1310b2e CM |
992 | u64 last_start; |
993 | u64 last_end; | |
42daec29 | 994 | |
a5dee37d | 995 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 996 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 997 | |
3f6bb4ae QW |
998 | if (exclusive_bits) |
999 | ASSERT(failed_start); | |
1000 | else | |
1001 | ASSERT(failed_start == NULL); | |
d1310b2e | 1002 | again: |
d0164adc | 1003 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
1004 | /* |
1005 | * Don't care for allocation failure here because we might end | |
1006 | * up not needing the pre-allocated extent state at all, which | |
1007 | * is the case if we only have in the tree extent states that | |
1008 | * cover our input range and don't cover too any other range. | |
1009 | * If we end up needing a new extent state we allocate it later. | |
1010 | */ | |
d1310b2e | 1011 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
1012 | } |
1013 | ||
cad321ad | 1014 | spin_lock(&tree->lock); |
9655d298 CM |
1015 | if (cached_state && *cached_state) { |
1016 | state = *cached_state; | |
df98b6e2 | 1017 | if (state->start <= start && state->end > start && |
27a3507d | 1018 | extent_state_in_tree(state)) { |
9655d298 CM |
1019 | node = &state->rb_node; |
1020 | goto hit_next; | |
1021 | } | |
1022 | } | |
d1310b2e CM |
1023 | /* |
1024 | * this search will find all the extents that end after | |
1025 | * our range starts. | |
1026 | */ | |
12cfbad9 | 1027 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1028 | if (!node) { |
8233767a XG |
1029 | prealloc = alloc_extent_state_atomic(prealloc); |
1030 | BUG_ON(!prealloc); | |
cee51268 DS |
1031 | prealloc->start = start; |
1032 | prealloc->end = end; | |
fb8f07d2 | 1033 | insert_state_fast(tree, prealloc, p, parent, bits, changeset); |
c42ac0bc | 1034 | cache_state(prealloc, cached_state); |
d1310b2e | 1035 | prealloc = NULL; |
d1310b2e CM |
1036 | goto out; |
1037 | } | |
d1310b2e | 1038 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1039 | hit_next: |
d1310b2e CM |
1040 | last_start = state->start; |
1041 | last_end = state->end; | |
1042 | ||
1043 | /* | |
1044 | * | ---- desired range ---- | | |
1045 | * | state | | |
1046 | * | |
1047 | * Just lock what we found and keep going | |
1048 | */ | |
1049 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1050 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1051 | *failed_start = state->start; |
1052 | err = -EEXIST; | |
1053 | goto out; | |
1054 | } | |
42daec29 | 1055 | |
6d92b304 | 1056 | set_state_bits(tree, state, bits, changeset); |
2c64c53d | 1057 | cache_state(state, cached_state); |
d1310b2e | 1058 | merge_state(tree, state); |
5c939df5 YZ |
1059 | if (last_end == (u64)-1) |
1060 | goto out; | |
1061 | start = last_end + 1; | |
d1ac6e41 LB |
1062 | state = next_state(state); |
1063 | if (start < end && state && state->start == start && | |
1064 | !need_resched()) | |
1065 | goto hit_next; | |
d1310b2e CM |
1066 | goto search_again; |
1067 | } | |
1068 | ||
1069 | /* | |
1070 | * | ---- desired range ---- | | |
1071 | * | state | | |
1072 | * or | |
1073 | * | ------------- state -------------- | | |
1074 | * | |
1075 | * We need to split the extent we found, and may flip bits on | |
1076 | * second half. | |
1077 | * | |
1078 | * If the extent we found extends past our | |
1079 | * range, we just split and search again. It'll get split | |
1080 | * again the next time though. | |
1081 | * | |
1082 | * If the extent we found is inside our range, we set the | |
1083 | * desired bit on it. | |
1084 | */ | |
1085 | if (state->start < start) { | |
1edbb734 | 1086 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1087 | *failed_start = start; |
1088 | err = -EEXIST; | |
1089 | goto out; | |
1090 | } | |
8233767a | 1091 | |
55ffaabe FM |
1092 | /* |
1093 | * If this extent already has all the bits we want set, then | |
1094 | * skip it, not necessary to split it or do anything with it. | |
1095 | */ | |
1096 | if ((state->state & bits) == bits) { | |
1097 | start = state->end + 1; | |
1098 | cache_state(state, cached_state); | |
1099 | goto search_again; | |
1100 | } | |
1101 | ||
8233767a XG |
1102 | prealloc = alloc_extent_state_atomic(prealloc); |
1103 | BUG_ON(!prealloc); | |
d1310b2e | 1104 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1105 | if (err) |
1106 | extent_io_tree_panic(tree, err); | |
1107 | ||
d1310b2e CM |
1108 | prealloc = NULL; |
1109 | if (err) | |
1110 | goto out; | |
1111 | if (state->end <= end) { | |
6d92b304 | 1112 | set_state_bits(tree, state, bits, changeset); |
2c64c53d | 1113 | cache_state(state, cached_state); |
d1310b2e | 1114 | merge_state(tree, state); |
5c939df5 YZ |
1115 | if (last_end == (u64)-1) |
1116 | goto out; | |
1117 | start = last_end + 1; | |
d1ac6e41 LB |
1118 | state = next_state(state); |
1119 | if (start < end && state && state->start == start && | |
1120 | !need_resched()) | |
1121 | goto hit_next; | |
d1310b2e CM |
1122 | } |
1123 | goto search_again; | |
1124 | } | |
1125 | /* | |
1126 | * | ---- desired range ---- | | |
1127 | * | state | or | state | | |
1128 | * | |
1129 | * There's a hole, we need to insert something in it and | |
1130 | * ignore the extent we found. | |
1131 | */ | |
1132 | if (state->start > start) { | |
1133 | u64 this_end; | |
1134 | if (end < last_start) | |
1135 | this_end = end; | |
1136 | else | |
d397712b | 1137 | this_end = last_start - 1; |
8233767a XG |
1138 | |
1139 | prealloc = alloc_extent_state_atomic(prealloc); | |
1140 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1141 | |
1142 | /* | |
1143 | * Avoid to free 'prealloc' if it can be merged with | |
1144 | * the later extent. | |
1145 | */ | |
cee51268 DS |
1146 | prealloc->start = start; |
1147 | prealloc->end = this_end; | |
c367602a | 1148 | err = insert_state(tree, prealloc, bits, changeset); |
c2d904e0 JM |
1149 | if (err) |
1150 | extent_io_tree_panic(tree, err); | |
1151 | ||
9ed74f2d JB |
1152 | cache_state(prealloc, cached_state); |
1153 | prealloc = NULL; | |
d1310b2e CM |
1154 | start = this_end + 1; |
1155 | goto search_again; | |
1156 | } | |
1157 | /* | |
1158 | * | ---- desired range ---- | | |
1159 | * | state | | |
1160 | * We need to split the extent, and set the bit | |
1161 | * on the first half | |
1162 | */ | |
1163 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1164 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1165 | *failed_start = start; |
1166 | err = -EEXIST; | |
1167 | goto out; | |
1168 | } | |
8233767a XG |
1169 | |
1170 | prealloc = alloc_extent_state_atomic(prealloc); | |
1171 | BUG_ON(!prealloc); | |
d1310b2e | 1172 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1173 | if (err) |
1174 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1175 | |
6d92b304 | 1176 | set_state_bits(tree, prealloc, bits, changeset); |
2c64c53d | 1177 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1178 | merge_state(tree, prealloc); |
1179 | prealloc = NULL; | |
1180 | goto out; | |
1181 | } | |
1182 | ||
b5a4ba14 DS |
1183 | search_again: |
1184 | if (start > end) | |
1185 | goto out; | |
1186 | spin_unlock(&tree->lock); | |
1187 | if (gfpflags_allow_blocking(mask)) | |
1188 | cond_resched(); | |
1189 | goto again; | |
d1310b2e CM |
1190 | |
1191 | out: | |
cad321ad | 1192 | spin_unlock(&tree->lock); |
d1310b2e CM |
1193 | if (prealloc) |
1194 | free_extent_state(prealloc); | |
1195 | ||
1196 | return err; | |
1197 | ||
d1310b2e | 1198 | } |
d1310b2e | 1199 | |
462d6fac | 1200 | /** |
10983f2e LB |
1201 | * convert_extent_bit - convert all bits in a given range from one bit to |
1202 | * another | |
462d6fac JB |
1203 | * @tree: the io tree to search |
1204 | * @start: the start offset in bytes | |
1205 | * @end: the end offset in bytes (inclusive) | |
1206 | * @bits: the bits to set in this range | |
1207 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1208 | * @cached_state: state that we're going to cache |
462d6fac JB |
1209 | * |
1210 | * This will go through and set bits for the given range. If any states exist | |
1211 | * already in this range they are set with the given bit and cleared of the | |
1212 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1213 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1214 | * boundary bits like LOCK. | |
210aa277 DS |
1215 | * |
1216 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1217 | */ |
1218 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 1219 | u32 bits, u32 clear_bits, |
210aa277 | 1220 | struct extent_state **cached_state) |
462d6fac JB |
1221 | { |
1222 | struct extent_state *state; | |
1223 | struct extent_state *prealloc = NULL; | |
1224 | struct rb_node *node; | |
12cfbad9 FDBM |
1225 | struct rb_node **p; |
1226 | struct rb_node *parent; | |
462d6fac JB |
1227 | int err = 0; |
1228 | u64 last_start; | |
1229 | u64 last_end; | |
c8fd3de7 | 1230 | bool first_iteration = true; |
462d6fac | 1231 | |
a5dee37d | 1232 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1233 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1234 | clear_bits); | |
8d599ae1 | 1235 | |
462d6fac | 1236 | again: |
210aa277 | 1237 | if (!prealloc) { |
c8fd3de7 FM |
1238 | /* |
1239 | * Best effort, don't worry if extent state allocation fails | |
1240 | * here for the first iteration. We might have a cached state | |
1241 | * that matches exactly the target range, in which case no | |
1242 | * extent state allocations are needed. We'll only know this | |
1243 | * after locking the tree. | |
1244 | */ | |
210aa277 | 1245 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1246 | if (!prealloc && !first_iteration) |
462d6fac JB |
1247 | return -ENOMEM; |
1248 | } | |
1249 | ||
1250 | spin_lock(&tree->lock); | |
e6138876 JB |
1251 | if (cached_state && *cached_state) { |
1252 | state = *cached_state; | |
1253 | if (state->start <= start && state->end > start && | |
27a3507d | 1254 | extent_state_in_tree(state)) { |
e6138876 JB |
1255 | node = &state->rb_node; |
1256 | goto hit_next; | |
1257 | } | |
1258 | } | |
1259 | ||
462d6fac JB |
1260 | /* |
1261 | * this search will find all the extents that end after | |
1262 | * our range starts. | |
1263 | */ | |
12cfbad9 | 1264 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1265 | if (!node) { |
1266 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1267 | if (!prealloc) { |
1268 | err = -ENOMEM; | |
1269 | goto out; | |
1270 | } | |
cee51268 DS |
1271 | prealloc->start = start; |
1272 | prealloc->end = end; | |
fb8f07d2 | 1273 | insert_state_fast(tree, prealloc, p, parent, bits, NULL); |
c42ac0bc FDBM |
1274 | cache_state(prealloc, cached_state); |
1275 | prealloc = NULL; | |
462d6fac JB |
1276 | goto out; |
1277 | } | |
1278 | state = rb_entry(node, struct extent_state, rb_node); | |
1279 | hit_next: | |
1280 | last_start = state->start; | |
1281 | last_end = state->end; | |
1282 | ||
1283 | /* | |
1284 | * | ---- desired range ---- | | |
1285 | * | state | | |
1286 | * | |
1287 | * Just lock what we found and keep going | |
1288 | */ | |
1289 | if (state->start == start && state->end <= end) { | |
6d92b304 | 1290 | set_state_bits(tree, state, bits, NULL); |
e6138876 | 1291 | cache_state(state, cached_state); |
6d92b304 | 1292 | state = clear_state_bit(tree, state, clear_bits, 0, NULL); |
462d6fac JB |
1293 | if (last_end == (u64)-1) |
1294 | goto out; | |
462d6fac | 1295 | start = last_end + 1; |
d1ac6e41 LB |
1296 | if (start < end && state && state->start == start && |
1297 | !need_resched()) | |
1298 | goto hit_next; | |
462d6fac JB |
1299 | goto search_again; |
1300 | } | |
1301 | ||
1302 | /* | |
1303 | * | ---- desired range ---- | | |
1304 | * | state | | |
1305 | * or | |
1306 | * | ------------- state -------------- | | |
1307 | * | |
1308 | * We need to split the extent we found, and may flip bits on | |
1309 | * second half. | |
1310 | * | |
1311 | * If the extent we found extends past our | |
1312 | * range, we just split and search again. It'll get split | |
1313 | * again the next time though. | |
1314 | * | |
1315 | * If the extent we found is inside our range, we set the | |
1316 | * desired bit on it. | |
1317 | */ | |
1318 | if (state->start < start) { | |
1319 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1320 | if (!prealloc) { |
1321 | err = -ENOMEM; | |
1322 | goto out; | |
1323 | } | |
462d6fac | 1324 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1325 | if (err) |
1326 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1327 | prealloc = NULL; |
1328 | if (err) | |
1329 | goto out; | |
1330 | if (state->end <= end) { | |
6d92b304 | 1331 | set_state_bits(tree, state, bits, NULL); |
e6138876 | 1332 | cache_state(state, cached_state); |
6d92b304 | 1333 | state = clear_state_bit(tree, state, clear_bits, 0, NULL); |
462d6fac JB |
1334 | if (last_end == (u64)-1) |
1335 | goto out; | |
1336 | start = last_end + 1; | |
d1ac6e41 LB |
1337 | if (start < end && state && state->start == start && |
1338 | !need_resched()) | |
1339 | goto hit_next; | |
462d6fac JB |
1340 | } |
1341 | goto search_again; | |
1342 | } | |
1343 | /* | |
1344 | * | ---- desired range ---- | | |
1345 | * | state | or | state | | |
1346 | * | |
1347 | * There's a hole, we need to insert something in it and | |
1348 | * ignore the extent we found. | |
1349 | */ | |
1350 | if (state->start > start) { | |
1351 | u64 this_end; | |
1352 | if (end < last_start) | |
1353 | this_end = end; | |
1354 | else | |
1355 | this_end = last_start - 1; | |
1356 | ||
1357 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1358 | if (!prealloc) { |
1359 | err = -ENOMEM; | |
1360 | goto out; | |
1361 | } | |
462d6fac JB |
1362 | |
1363 | /* | |
1364 | * Avoid to free 'prealloc' if it can be merged with | |
1365 | * the later extent. | |
1366 | */ | |
cee51268 DS |
1367 | prealloc->start = start; |
1368 | prealloc->end = this_end; | |
c367602a | 1369 | err = insert_state(tree, prealloc, bits, NULL); |
c2d904e0 JM |
1370 | if (err) |
1371 | extent_io_tree_panic(tree, err); | |
e6138876 | 1372 | cache_state(prealloc, cached_state); |
462d6fac JB |
1373 | prealloc = NULL; |
1374 | start = this_end + 1; | |
1375 | goto search_again; | |
1376 | } | |
1377 | /* | |
1378 | * | ---- desired range ---- | | |
1379 | * | state | | |
1380 | * We need to split the extent, and set the bit | |
1381 | * on the first half | |
1382 | */ | |
1383 | if (state->start <= end && state->end > end) { | |
1384 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1385 | if (!prealloc) { |
1386 | err = -ENOMEM; | |
1387 | goto out; | |
1388 | } | |
462d6fac JB |
1389 | |
1390 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1391 | if (err) |
1392 | extent_io_tree_panic(tree, err); | |
462d6fac | 1393 | |
6d92b304 | 1394 | set_state_bits(tree, prealloc, bits, NULL); |
e6138876 | 1395 | cache_state(prealloc, cached_state); |
6d92b304 | 1396 | clear_state_bit(tree, prealloc, clear_bits, 0, NULL); |
462d6fac JB |
1397 | prealloc = NULL; |
1398 | goto out; | |
1399 | } | |
1400 | ||
462d6fac JB |
1401 | search_again: |
1402 | if (start > end) | |
1403 | goto out; | |
1404 | spin_unlock(&tree->lock); | |
210aa277 | 1405 | cond_resched(); |
c8fd3de7 | 1406 | first_iteration = false; |
462d6fac | 1407 | goto again; |
462d6fac JB |
1408 | |
1409 | out: | |
1410 | spin_unlock(&tree->lock); | |
1411 | if (prealloc) | |
1412 | free_extent_state(prealloc); | |
1413 | ||
1414 | return err; | |
462d6fac JB |
1415 | } |
1416 | ||
d1310b2e | 1417 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1418 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1419 | u32 bits, struct extent_changeset *changeset) |
d38ed27f QW |
1420 | { |
1421 | /* | |
1422 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1423 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1424 | * either fail with -EEXIST or changeset will record the whole | |
1425 | * range. | |
1426 | */ | |
1427 | BUG_ON(bits & EXTENT_LOCKED); | |
1428 | ||
1cab5e72 NB |
1429 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
1430 | changeset); | |
d38ed27f QW |
1431 | } |
1432 | ||
4ca73656 | 1433 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1434 | u32 bits) |
4ca73656 | 1435 | { |
1cab5e72 NB |
1436 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, |
1437 | GFP_NOWAIT, NULL); | |
4ca73656 NB |
1438 | } |
1439 | ||
fefdc557 | 1440 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1441 | u32 bits, int wake, int delete, |
ae0f1625 | 1442 | struct extent_state **cached) |
fefdc557 QW |
1443 | { |
1444 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1445 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1446 | } |
1447 | ||
fefdc557 | 1448 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1449 | u32 bits, struct extent_changeset *changeset) |
fefdc557 QW |
1450 | { |
1451 | /* | |
1452 | * Don't support EXTENT_LOCKED case, same reason as | |
1453 | * set_record_extent_bits(). | |
1454 | */ | |
1455 | BUG_ON(bits & EXTENT_LOCKED); | |
1456 | ||
f734c44a | 1457 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1458 | changeset); |
1459 | } | |
1460 | ||
d352ac68 CM |
1461 | /* |
1462 | * either insert or lock state struct between start and end use mask to tell | |
1463 | * us if waiting is desired. | |
1464 | */ | |
1edbb734 | 1465 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1466 | struct extent_state **cached_state) |
d1310b2e CM |
1467 | { |
1468 | int err; | |
1469 | u64 failed_start; | |
9ee49a04 | 1470 | |
d1310b2e | 1471 | while (1) { |
1cab5e72 NB |
1472 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, |
1473 | EXTENT_LOCKED, &failed_start, | |
1474 | cached_state, GFP_NOFS, NULL); | |
d0082371 | 1475 | if (err == -EEXIST) { |
d1310b2e CM |
1476 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1477 | start = failed_start; | |
d0082371 | 1478 | } else |
d1310b2e | 1479 | break; |
d1310b2e CM |
1480 | WARN_ON(start > end); |
1481 | } | |
1482 | return err; | |
1483 | } | |
d1310b2e | 1484 | |
d0082371 | 1485 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1486 | { |
1487 | int err; | |
1488 | u64 failed_start; | |
1489 | ||
1cab5e72 NB |
1490 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1491 | &failed_start, NULL, GFP_NOFS, NULL); | |
6643558d YZ |
1492 | if (err == -EEXIST) { |
1493 | if (failed_start > start) | |
1494 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1495 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1496 | return 0; |
6643558d | 1497 | } |
25179201 JB |
1498 | return 1; |
1499 | } | |
25179201 | 1500 | |
bd1fa4f0 | 1501 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1502 | { |
09cbfeaf KS |
1503 | unsigned long index = start >> PAGE_SHIFT; |
1504 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1505 | struct page *page; |
1506 | ||
1507 | while (index <= end_index) { | |
1508 | page = find_get_page(inode->i_mapping, index); | |
1509 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1510 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1511 | put_page(page); |
4adaa611 CM |
1512 | index++; |
1513 | } | |
4adaa611 CM |
1514 | } |
1515 | ||
f6311572 | 1516 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1517 | { |
ebf55c88 | 1518 | struct address_space *mapping = inode->i_mapping; |
09cbfeaf KS |
1519 | unsigned long index = start >> PAGE_SHIFT; |
1520 | unsigned long end_index = end >> PAGE_SHIFT; | |
ebf55c88 | 1521 | struct folio *folio; |
4adaa611 CM |
1522 | |
1523 | while (index <= end_index) { | |
ebf55c88 MWO |
1524 | folio = filemap_get_folio(mapping, index); |
1525 | filemap_dirty_folio(mapping, folio); | |
1526 | folio_account_redirty(folio); | |
1527 | index += folio_nr_pages(folio); | |
1528 | folio_put(folio); | |
4adaa611 | 1529 | } |
4adaa611 CM |
1530 | } |
1531 | ||
d352ac68 CM |
1532 | /* find the first state struct with 'bits' set after 'start', and |
1533 | * return it. tree->lock must be held. NULL will returned if | |
1534 | * nothing was found after 'start' | |
1535 | */ | |
48a3b636 | 1536 | static struct extent_state * |
f97e27e9 | 1537 | find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits) |
d7fc640e CM |
1538 | { |
1539 | struct rb_node *node; | |
1540 | struct extent_state *state; | |
1541 | ||
1542 | /* | |
1543 | * this search will find all the extents that end after | |
1544 | * our range starts. | |
1545 | */ | |
1546 | node = tree_search(tree, start); | |
d397712b | 1547 | if (!node) |
d7fc640e | 1548 | goto out; |
d7fc640e | 1549 | |
d397712b | 1550 | while (1) { |
d7fc640e | 1551 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1552 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1553 | return state; |
d397712b | 1554 | |
d7fc640e CM |
1555 | node = rb_next(node); |
1556 | if (!node) | |
1557 | break; | |
1558 | } | |
1559 | out: | |
1560 | return NULL; | |
1561 | } | |
d7fc640e | 1562 | |
69261c4b | 1563 | /* |
03509b78 | 1564 | * Find the first offset in the io tree with one or more @bits set. |
69261c4b | 1565 | * |
03509b78 QW |
1566 | * Note: If there are multiple bits set in @bits, any of them will match. |
1567 | * | |
1568 | * Return 0 if we find something, and update @start_ret and @end_ret. | |
1569 | * Return 1 if we found nothing. | |
69261c4b XG |
1570 | */ |
1571 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1572 | u64 *start_ret, u64 *end_ret, u32 bits, |
e6138876 | 1573 | struct extent_state **cached_state) |
69261c4b XG |
1574 | { |
1575 | struct extent_state *state; | |
1576 | int ret = 1; | |
1577 | ||
1578 | spin_lock(&tree->lock); | |
e6138876 JB |
1579 | if (cached_state && *cached_state) { |
1580 | state = *cached_state; | |
27a3507d | 1581 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1582 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1583 | if (state->state & bits) |
1584 | goto got_it; | |
e6138876 JB |
1585 | } |
1586 | free_extent_state(*cached_state); | |
1587 | *cached_state = NULL; | |
1588 | goto out; | |
1589 | } | |
1590 | free_extent_state(*cached_state); | |
1591 | *cached_state = NULL; | |
1592 | } | |
1593 | ||
69261c4b | 1594 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1595 | got_it: |
69261c4b | 1596 | if (state) { |
e38e2ed7 | 1597 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1598 | *start_ret = state->start; |
1599 | *end_ret = state->end; | |
1600 | ret = 0; | |
1601 | } | |
e6138876 | 1602 | out: |
69261c4b XG |
1603 | spin_unlock(&tree->lock); |
1604 | return ret; | |
1605 | } | |
1606 | ||
41a2ee75 | 1607 | /** |
3bed2da1 NB |
1608 | * Find a contiguous area of bits |
1609 | * | |
1610 | * @tree: io tree to check | |
1611 | * @start: offset to start the search from | |
1612 | * @start_ret: the first offset we found with the bits set | |
1613 | * @end_ret: the final contiguous range of the bits that were set | |
1614 | * @bits: bits to look for | |
41a2ee75 JB |
1615 | * |
1616 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1617 | * to set bits appropriately, and then merge them again. During this time it | |
1618 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1619 | * contiguous area for given bits. We will search to the first bit we find, and | |
1620 | * then walk down the tree until we find a non-contiguous area. The area | |
1621 | * returned will be the full contiguous area with the bits set. | |
1622 | */ | |
1623 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1624 | u64 *start_ret, u64 *end_ret, u32 bits) |
41a2ee75 JB |
1625 | { |
1626 | struct extent_state *state; | |
1627 | int ret = 1; | |
1628 | ||
1629 | spin_lock(&tree->lock); | |
1630 | state = find_first_extent_bit_state(tree, start, bits); | |
1631 | if (state) { | |
1632 | *start_ret = state->start; | |
1633 | *end_ret = state->end; | |
1634 | while ((state = next_state(state)) != NULL) { | |
1635 | if (state->start > (*end_ret + 1)) | |
1636 | break; | |
1637 | *end_ret = state->end; | |
1638 | } | |
1639 | ret = 0; | |
1640 | } | |
1641 | spin_unlock(&tree->lock); | |
1642 | return ret; | |
1643 | } | |
1644 | ||
45bfcfc1 | 1645 | /** |
3bed2da1 NB |
1646 | * Find the first range that has @bits not set. This range could start before |
1647 | * @start. | |
45bfcfc1 | 1648 | * |
3bed2da1 NB |
1649 | * @tree: the tree to search |
1650 | * @start: offset at/after which the found extent should start | |
1651 | * @start_ret: records the beginning of the range | |
1652 | * @end_ret: records the end of the range (inclusive) | |
1653 | * @bits: the set of bits which must be unset | |
45bfcfc1 NB |
1654 | * |
1655 | * Since unallocated range is also considered one which doesn't have the bits | |
1656 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1657 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1658 | * trim @end_ret to the appropriate size. | |
1659 | */ | |
1660 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1661 | u64 *start_ret, u64 *end_ret, u32 bits) |
45bfcfc1 NB |
1662 | { |
1663 | struct extent_state *state; | |
1664 | struct rb_node *node, *prev = NULL, *next; | |
1665 | ||
1666 | spin_lock(&tree->lock); | |
1667 | ||
1668 | /* Find first extent with bits cleared */ | |
1669 | while (1) { | |
9db33891 | 1670 | node = tree_search_prev_next(tree, start, &prev, &next); |
5750c375 NB |
1671 | if (!node && !next && !prev) { |
1672 | /* | |
1673 | * Tree is completely empty, send full range and let | |
1674 | * caller deal with it | |
1675 | */ | |
1676 | *start_ret = 0; | |
1677 | *end_ret = -1; | |
1678 | goto out; | |
1679 | } else if (!node && !next) { | |
1680 | /* | |
1681 | * We are past the last allocated chunk, set start at | |
1682 | * the end of the last extent. | |
1683 | */ | |
1684 | state = rb_entry(prev, struct extent_state, rb_node); | |
1685 | *start_ret = state->end + 1; | |
1686 | *end_ret = -1; | |
1687 | goto out; | |
1688 | } else if (!node) { | |
45bfcfc1 | 1689 | node = next; |
45bfcfc1 | 1690 | } |
1eaebb34 NB |
1691 | /* |
1692 | * At this point 'node' either contains 'start' or start is | |
1693 | * before 'node' | |
1694 | */ | |
45bfcfc1 | 1695 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1696 | |
1697 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1698 | if (state->state & bits) { | |
1699 | /* | |
1700 | * |--range with bits sets--| | |
1701 | * | | |
1702 | * start | |
1703 | */ | |
1704 | start = state->end + 1; | |
1705 | } else { | |
1706 | /* | |
1707 | * 'start' falls within a range that doesn't | |
1708 | * have the bits set, so take its start as | |
1709 | * the beginning of the desired range | |
1710 | * | |
1711 | * |--range with bits cleared----| | |
1712 | * | | |
1713 | * start | |
1714 | */ | |
1715 | *start_ret = state->start; | |
1716 | break; | |
1717 | } | |
45bfcfc1 | 1718 | } else { |
1eaebb34 NB |
1719 | /* |
1720 | * |---prev range---|---hole/unset---|---node range---| | |
1721 | * | | |
1722 | * start | |
1723 | * | |
1724 | * or | |
1725 | * | |
1726 | * |---hole/unset--||--first node--| | |
1727 | * 0 | | |
1728 | * start | |
1729 | */ | |
1730 | if (prev) { | |
1731 | state = rb_entry(prev, struct extent_state, | |
1732 | rb_node); | |
1733 | *start_ret = state->end + 1; | |
1734 | } else { | |
1735 | *start_ret = 0; | |
1736 | } | |
45bfcfc1 NB |
1737 | break; |
1738 | } | |
1739 | } | |
1740 | ||
1741 | /* | |
1742 | * Find the longest stretch from start until an entry which has the | |
1743 | * bits set | |
1744 | */ | |
1745 | while (1) { | |
1746 | state = rb_entry(node, struct extent_state, rb_node); | |
1747 | if (state->end >= start && !(state->state & bits)) { | |
1748 | *end_ret = state->end; | |
1749 | } else { | |
1750 | *end_ret = state->start - 1; | |
1751 | break; | |
1752 | } | |
1753 | ||
1754 | node = rb_next(node); | |
1755 | if (!node) | |
1756 | break; | |
1757 | } | |
1758 | out: | |
1759 | spin_unlock(&tree->lock); | |
1760 | } | |
1761 | ||
d352ac68 CM |
1762 | /* |
1763 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1764 | * more than 'max_bytes'. start and end are used to return the range, | |
1765 | * | |
3522e903 | 1766 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1767 | */ |
083e75e7 JB |
1768 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1769 | u64 *end, u64 max_bytes, | |
1770 | struct extent_state **cached_state) | |
d1310b2e CM |
1771 | { |
1772 | struct rb_node *node; | |
1773 | struct extent_state *state; | |
1774 | u64 cur_start = *start; | |
3522e903 | 1775 | bool found = false; |
d1310b2e CM |
1776 | u64 total_bytes = 0; |
1777 | ||
cad321ad | 1778 | spin_lock(&tree->lock); |
c8b97818 | 1779 | |
d1310b2e CM |
1780 | /* |
1781 | * this search will find all the extents that end after | |
1782 | * our range starts. | |
1783 | */ | |
80ea96b1 | 1784 | node = tree_search(tree, cur_start); |
2b114d1d | 1785 | if (!node) { |
3522e903 | 1786 | *end = (u64)-1; |
d1310b2e CM |
1787 | goto out; |
1788 | } | |
1789 | ||
d397712b | 1790 | while (1) { |
d1310b2e | 1791 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1792 | if (found && (state->start != cur_start || |
1793 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1794 | goto out; |
1795 | } | |
1796 | if (!(state->state & EXTENT_DELALLOC)) { | |
1797 | if (!found) | |
1798 | *end = state->end; | |
1799 | goto out; | |
1800 | } | |
c2a128d2 | 1801 | if (!found) { |
d1310b2e | 1802 | *start = state->start; |
c2a128d2 | 1803 | *cached_state = state; |
b7ac31b7 | 1804 | refcount_inc(&state->refs); |
c2a128d2 | 1805 | } |
3522e903 | 1806 | found = true; |
d1310b2e CM |
1807 | *end = state->end; |
1808 | cur_start = state->end + 1; | |
1809 | node = rb_next(node); | |
d1310b2e | 1810 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1811 | if (total_bytes >= max_bytes) |
573aecaf | 1812 | break; |
573aecaf | 1813 | if (!node) |
d1310b2e CM |
1814 | break; |
1815 | } | |
1816 | out: | |
cad321ad | 1817 | spin_unlock(&tree->lock); |
d1310b2e CM |
1818 | return found; |
1819 | } | |
1820 | ||
ed8f13bf QW |
1821 | /* |
1822 | * Process one page for __process_pages_contig(). | |
1823 | * | |
1824 | * Return >0 if we hit @page == @locked_page. | |
1825 | * Return 0 if we updated the page status. | |
1826 | * Return -EGAIN if the we need to try again. | |
1827 | * (For PAGE_LOCK case but got dirty page or page not belong to mapping) | |
1828 | */ | |
e38992be QW |
1829 | static int process_one_page(struct btrfs_fs_info *fs_info, |
1830 | struct address_space *mapping, | |
ed8f13bf | 1831 | struct page *page, struct page *locked_page, |
e38992be | 1832 | unsigned long page_ops, u64 start, u64 end) |
ed8f13bf | 1833 | { |
e38992be QW |
1834 | u32 len; |
1835 | ||
1836 | ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX); | |
1837 | len = end + 1 - start; | |
1838 | ||
ed8f13bf | 1839 | if (page_ops & PAGE_SET_ORDERED) |
b945a463 | 1840 | btrfs_page_clamp_set_ordered(fs_info, page, start, len); |
ed8f13bf | 1841 | if (page_ops & PAGE_SET_ERROR) |
e38992be | 1842 | btrfs_page_clamp_set_error(fs_info, page, start, len); |
ed8f13bf | 1843 | if (page_ops & PAGE_START_WRITEBACK) { |
e38992be QW |
1844 | btrfs_page_clamp_clear_dirty(fs_info, page, start, len); |
1845 | btrfs_page_clamp_set_writeback(fs_info, page, start, len); | |
ed8f13bf QW |
1846 | } |
1847 | if (page_ops & PAGE_END_WRITEBACK) | |
e38992be | 1848 | btrfs_page_clamp_clear_writeback(fs_info, page, start, len); |
a33a8e9a QW |
1849 | |
1850 | if (page == locked_page) | |
1851 | return 1; | |
1852 | ||
ed8f13bf | 1853 | if (page_ops & PAGE_LOCK) { |
1e1de387 QW |
1854 | int ret; |
1855 | ||
1856 | ret = btrfs_page_start_writer_lock(fs_info, page, start, len); | |
1857 | if (ret) | |
1858 | return ret; | |
ed8f13bf | 1859 | if (!PageDirty(page) || page->mapping != mapping) { |
1e1de387 | 1860 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1861 | return -EAGAIN; |
1862 | } | |
1863 | } | |
1864 | if (page_ops & PAGE_UNLOCK) | |
1e1de387 | 1865 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1866 | return 0; |
1867 | } | |
1868 | ||
da2c7009 LB |
1869 | static int __process_pages_contig(struct address_space *mapping, |
1870 | struct page *locked_page, | |
98af9ab1 | 1871 | u64 start, u64 end, unsigned long page_ops, |
ed8f13bf QW |
1872 | u64 *processed_end) |
1873 | { | |
e38992be | 1874 | struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); |
ed8f13bf QW |
1875 | pgoff_t start_index = start >> PAGE_SHIFT; |
1876 | pgoff_t end_index = end >> PAGE_SHIFT; | |
1877 | pgoff_t index = start_index; | |
1878 | unsigned long nr_pages = end_index - start_index + 1; | |
1879 | unsigned long pages_processed = 0; | |
1880 | struct page *pages[16]; | |
1881 | int err = 0; | |
1882 | int i; | |
1883 | ||
1884 | if (page_ops & PAGE_LOCK) { | |
1885 | ASSERT(page_ops == PAGE_LOCK); | |
1886 | ASSERT(processed_end && *processed_end == start); | |
1887 | } | |
1888 | ||
1889 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) | |
1890 | mapping_set_error(mapping, -EIO); | |
1891 | ||
1892 | while (nr_pages > 0) { | |
1893 | int found_pages; | |
1894 | ||
1895 | found_pages = find_get_pages_contig(mapping, index, | |
1896 | min_t(unsigned long, | |
1897 | nr_pages, ARRAY_SIZE(pages)), pages); | |
1898 | if (found_pages == 0) { | |
1899 | /* | |
1900 | * Only if we're going to lock these pages, we can find | |
1901 | * nothing at @index. | |
1902 | */ | |
1903 | ASSERT(page_ops & PAGE_LOCK); | |
1904 | err = -EAGAIN; | |
1905 | goto out; | |
1906 | } | |
1907 | ||
1908 | for (i = 0; i < found_pages; i++) { | |
1909 | int process_ret; | |
1910 | ||
e38992be QW |
1911 | process_ret = process_one_page(fs_info, mapping, |
1912 | pages[i], locked_page, page_ops, | |
1913 | start, end); | |
ed8f13bf QW |
1914 | if (process_ret < 0) { |
1915 | for (; i < found_pages; i++) | |
1916 | put_page(pages[i]); | |
1917 | err = -EAGAIN; | |
1918 | goto out; | |
1919 | } | |
1920 | put_page(pages[i]); | |
1921 | pages_processed++; | |
1922 | } | |
1923 | nr_pages -= found_pages; | |
1924 | index += found_pages; | |
1925 | cond_resched(); | |
1926 | } | |
1927 | out: | |
1928 | if (err && processed_end) { | |
1929 | /* | |
1930 | * Update @processed_end. I know this is awful since it has | |
1931 | * two different return value patterns (inclusive vs exclusive). | |
1932 | * | |
1933 | * But the exclusive pattern is necessary if @start is 0, or we | |
1934 | * underflow and check against processed_end won't work as | |
1935 | * expected. | |
1936 | */ | |
1937 | if (pages_processed) | |
1938 | *processed_end = min(end, | |
1939 | ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1); | |
1940 | else | |
1941 | *processed_end = start; | |
1942 | } | |
1943 | return err; | |
1944 | } | |
da2c7009 | 1945 | |
143bede5 JM |
1946 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1947 | struct page *locked_page, | |
1948 | u64 start, u64 end) | |
c8b97818 | 1949 | { |
09cbfeaf KS |
1950 | unsigned long index = start >> PAGE_SHIFT; |
1951 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1952 | |
76c0021d | 1953 | ASSERT(locked_page); |
c8b97818 | 1954 | if (index == locked_page->index && end_index == index) |
143bede5 | 1955 | return; |
c8b97818 | 1956 | |
98af9ab1 | 1957 | __process_pages_contig(inode->i_mapping, locked_page, start, end, |
76c0021d | 1958 | PAGE_UNLOCK, NULL); |
c8b97818 CM |
1959 | } |
1960 | ||
1961 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1962 | struct page *locked_page, | |
1963 | u64 delalloc_start, | |
1964 | u64 delalloc_end) | |
1965 | { | |
09cbfeaf | 1966 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
09cbfeaf | 1967 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
98af9ab1 | 1968 | u64 processed_end = delalloc_start; |
c8b97818 | 1969 | int ret; |
c8b97818 | 1970 | |
76c0021d | 1971 | ASSERT(locked_page); |
c8b97818 CM |
1972 | if (index == locked_page->index && index == end_index) |
1973 | return 0; | |
1974 | ||
98af9ab1 QW |
1975 | ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start, |
1976 | delalloc_end, PAGE_LOCK, &processed_end); | |
1977 | if (ret == -EAGAIN && processed_end > delalloc_start) | |
76c0021d | 1978 | __unlock_for_delalloc(inode, locked_page, delalloc_start, |
98af9ab1 | 1979 | processed_end); |
c8b97818 CM |
1980 | return ret; |
1981 | } | |
1982 | ||
1983 | /* | |
3522e903 | 1984 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
2749f7ef | 1985 | * more than @max_bytes. |
c8b97818 | 1986 | * |
2749f7ef QW |
1987 | * @start: The original start bytenr to search. |
1988 | * Will store the extent range start bytenr. | |
1989 | * @end: The original end bytenr of the search range | |
1990 | * Will store the extent range end bytenr. | |
1991 | * | |
1992 | * Return true if we find a delalloc range which starts inside the original | |
1993 | * range, and @start/@end will store the delalloc range start/end. | |
1994 | * | |
1995 | * Return false if we can't find any delalloc range which starts inside the | |
1996 | * original range, and @start/@end will be the non-delalloc range start/end. | |
c8b97818 | 1997 | */ |
ce9f967f | 1998 | EXPORT_FOR_TESTS |
3522e903 | 1999 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 2000 | struct page *locked_page, u64 *start, |
917aacec | 2001 | u64 *end) |
c8b97818 | 2002 | { |
f7b12a62 | 2003 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9978059b | 2004 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
2749f7ef QW |
2005 | const u64 orig_start = *start; |
2006 | const u64 orig_end = *end; | |
f7b12a62 NA |
2007 | /* The sanity tests may not set a valid fs_info. */ |
2008 | u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE; | |
c8b97818 CM |
2009 | u64 delalloc_start; |
2010 | u64 delalloc_end; | |
3522e903 | 2011 | bool found; |
9655d298 | 2012 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
2013 | int ret; |
2014 | int loops = 0; | |
2015 | ||
2749f7ef QW |
2016 | /* Caller should pass a valid @end to indicate the search range end */ |
2017 | ASSERT(orig_end > orig_start); | |
2018 | ||
2019 | /* The range should at least cover part of the page */ | |
2020 | ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE || | |
2021 | orig_end <= page_offset(locked_page))); | |
c8b97818 CM |
2022 | again: |
2023 | /* step one, find a bunch of delalloc bytes starting at start */ | |
2024 | delalloc_start = *start; | |
2025 | delalloc_end = 0; | |
083e75e7 JB |
2026 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
2027 | max_bytes, &cached_state); | |
2749f7ef | 2028 | if (!found || delalloc_end <= *start || delalloc_start > orig_end) { |
c8b97818 | 2029 | *start = delalloc_start; |
2749f7ef QW |
2030 | |
2031 | /* @delalloc_end can be -1, never go beyond @orig_end */ | |
2032 | *end = min(delalloc_end, orig_end); | |
c2a128d2 | 2033 | free_extent_state(cached_state); |
3522e903 | 2034 | return false; |
c8b97818 CM |
2035 | } |
2036 | ||
70b99e69 CM |
2037 | /* |
2038 | * start comes from the offset of locked_page. We have to lock | |
2039 | * pages in order, so we can't process delalloc bytes before | |
2040 | * locked_page | |
2041 | */ | |
d397712b | 2042 | if (delalloc_start < *start) |
70b99e69 | 2043 | delalloc_start = *start; |
70b99e69 | 2044 | |
c8b97818 CM |
2045 | /* |
2046 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 2047 | */ |
7bf811a5 JB |
2048 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
2049 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 2050 | |
c8b97818 CM |
2051 | /* step two, lock all the pages after the page that has start */ |
2052 | ret = lock_delalloc_pages(inode, locked_page, | |
2053 | delalloc_start, delalloc_end); | |
9bfd61d9 | 2054 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
2055 | if (ret == -EAGAIN) { |
2056 | /* some of the pages are gone, lets avoid looping by | |
2057 | * shortening the size of the delalloc range we're searching | |
2058 | */ | |
9655d298 | 2059 | free_extent_state(cached_state); |
7d788742 | 2060 | cached_state = NULL; |
c8b97818 | 2061 | if (!loops) { |
09cbfeaf | 2062 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
2063 | loops = 1; |
2064 | goto again; | |
2065 | } else { | |
3522e903 | 2066 | found = false; |
c8b97818 CM |
2067 | goto out_failed; |
2068 | } | |
2069 | } | |
c8b97818 CM |
2070 | |
2071 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 2072 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
2073 | |
2074 | /* then test to make sure it is all still delalloc */ | |
2075 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 2076 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 2077 | if (!ret) { |
9655d298 | 2078 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 2079 | &cached_state); |
c8b97818 CM |
2080 | __unlock_for_delalloc(inode, locked_page, |
2081 | delalloc_start, delalloc_end); | |
2082 | cond_resched(); | |
2083 | goto again; | |
2084 | } | |
9655d298 | 2085 | free_extent_state(cached_state); |
c8b97818 CM |
2086 | *start = delalloc_start; |
2087 | *end = delalloc_end; | |
2088 | out_failed: | |
2089 | return found; | |
2090 | } | |
2091 | ||
ad7ff17b | 2092 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 2093 | struct page *locked_page, |
f97e27e9 | 2094 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 2095 | { |
ad7ff17b | 2096 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL); |
873695b3 | 2097 | |
ad7ff17b | 2098 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
98af9ab1 | 2099 | start, end, page_ops, NULL); |
873695b3 LB |
2100 | } |
2101 | ||
d352ac68 CM |
2102 | /* |
2103 | * count the number of bytes in the tree that have a given bit(s) | |
2104 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2105 | * cached. The total number found is returned. | |
2106 | */ | |
d1310b2e CM |
2107 | u64 count_range_bits(struct extent_io_tree *tree, |
2108 | u64 *start, u64 search_end, u64 max_bytes, | |
f97e27e9 | 2109 | u32 bits, int contig) |
d1310b2e CM |
2110 | { |
2111 | struct rb_node *node; | |
2112 | struct extent_state *state; | |
2113 | u64 cur_start = *start; | |
2114 | u64 total_bytes = 0; | |
ec29ed5b | 2115 | u64 last = 0; |
d1310b2e CM |
2116 | int found = 0; |
2117 | ||
fae7f21c | 2118 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2119 | return 0; |
d1310b2e | 2120 | |
cad321ad | 2121 | spin_lock(&tree->lock); |
d1310b2e CM |
2122 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2123 | total_bytes = tree->dirty_bytes; | |
2124 | goto out; | |
2125 | } | |
2126 | /* | |
2127 | * this search will find all the extents that end after | |
2128 | * our range starts. | |
2129 | */ | |
80ea96b1 | 2130 | node = tree_search(tree, cur_start); |
d397712b | 2131 | if (!node) |
d1310b2e | 2132 | goto out; |
d1310b2e | 2133 | |
d397712b | 2134 | while (1) { |
d1310b2e CM |
2135 | state = rb_entry(node, struct extent_state, rb_node); |
2136 | if (state->start > search_end) | |
2137 | break; | |
ec29ed5b CM |
2138 | if (contig && found && state->start > last + 1) |
2139 | break; | |
2140 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2141 | total_bytes += min(search_end, state->end) + 1 - |
2142 | max(cur_start, state->start); | |
2143 | if (total_bytes >= max_bytes) | |
2144 | break; | |
2145 | if (!found) { | |
af60bed2 | 2146 | *start = max(cur_start, state->start); |
d1310b2e CM |
2147 | found = 1; |
2148 | } | |
ec29ed5b CM |
2149 | last = state->end; |
2150 | } else if (contig && found) { | |
2151 | break; | |
d1310b2e CM |
2152 | } |
2153 | node = rb_next(node); | |
2154 | if (!node) | |
2155 | break; | |
2156 | } | |
2157 | out: | |
cad321ad | 2158 | spin_unlock(&tree->lock); |
d1310b2e CM |
2159 | return total_bytes; |
2160 | } | |
b2950863 | 2161 | |
d352ac68 CM |
2162 | /* |
2163 | * set the private field for a given byte offset in the tree. If there isn't | |
2164 | * an extent_state there already, this does nothing. | |
2165 | */ | |
b3f167aa JB |
2166 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2167 | struct io_failure_record *failrec) | |
d1310b2e CM |
2168 | { |
2169 | struct rb_node *node; | |
2170 | struct extent_state *state; | |
2171 | int ret = 0; | |
2172 | ||
cad321ad | 2173 | spin_lock(&tree->lock); |
d1310b2e CM |
2174 | /* |
2175 | * this search will find all the extents that end after | |
2176 | * our range starts. | |
2177 | */ | |
80ea96b1 | 2178 | node = tree_search(tree, start); |
2b114d1d | 2179 | if (!node) { |
d1310b2e CM |
2180 | ret = -ENOENT; |
2181 | goto out; | |
2182 | } | |
2183 | state = rb_entry(node, struct extent_state, rb_node); | |
2184 | if (state->start != start) { | |
2185 | ret = -ENOENT; | |
2186 | goto out; | |
2187 | } | |
47dc196a | 2188 | state->failrec = failrec; |
d1310b2e | 2189 | out: |
cad321ad | 2190 | spin_unlock(&tree->lock); |
d1310b2e CM |
2191 | return ret; |
2192 | } | |
2193 | ||
2279a270 | 2194 | struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start) |
d1310b2e CM |
2195 | { |
2196 | struct rb_node *node; | |
2197 | struct extent_state *state; | |
2279a270 | 2198 | struct io_failure_record *failrec; |
d1310b2e | 2199 | |
cad321ad | 2200 | spin_lock(&tree->lock); |
d1310b2e CM |
2201 | /* |
2202 | * this search will find all the extents that end after | |
2203 | * our range starts. | |
2204 | */ | |
80ea96b1 | 2205 | node = tree_search(tree, start); |
2b114d1d | 2206 | if (!node) { |
2279a270 | 2207 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2208 | goto out; |
2209 | } | |
2210 | state = rb_entry(node, struct extent_state, rb_node); | |
2211 | if (state->start != start) { | |
2279a270 | 2212 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2213 | goto out; |
2214 | } | |
2279a270 NB |
2215 | |
2216 | failrec = state->failrec; | |
d1310b2e | 2217 | out: |
cad321ad | 2218 | spin_unlock(&tree->lock); |
2279a270 | 2219 | return failrec; |
d1310b2e CM |
2220 | } |
2221 | ||
2222 | /* | |
2223 | * searches a range in the state tree for a given mask. | |
70dec807 | 2224 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2225 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2226 | * range is found set. | |
2227 | */ | |
2228 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 2229 | u32 bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2230 | { |
2231 | struct extent_state *state = NULL; | |
2232 | struct rb_node *node; | |
2233 | int bitset = 0; | |
d1310b2e | 2234 | |
cad321ad | 2235 | spin_lock(&tree->lock); |
27a3507d | 2236 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2237 | cached->end > start) |
9655d298 CM |
2238 | node = &cached->rb_node; |
2239 | else | |
2240 | node = tree_search(tree, start); | |
d1310b2e CM |
2241 | while (node && start <= end) { |
2242 | state = rb_entry(node, struct extent_state, rb_node); | |
2243 | ||
2244 | if (filled && state->start > start) { | |
2245 | bitset = 0; | |
2246 | break; | |
2247 | } | |
2248 | ||
2249 | if (state->start > end) | |
2250 | break; | |
2251 | ||
2252 | if (state->state & bits) { | |
2253 | bitset = 1; | |
2254 | if (!filled) | |
2255 | break; | |
2256 | } else if (filled) { | |
2257 | bitset = 0; | |
2258 | break; | |
2259 | } | |
46562cec CM |
2260 | |
2261 | if (state->end == (u64)-1) | |
2262 | break; | |
2263 | ||
d1310b2e CM |
2264 | start = state->end + 1; |
2265 | if (start > end) | |
2266 | break; | |
2267 | node = rb_next(node); | |
2268 | if (!node) { | |
2269 | if (filled) | |
2270 | bitset = 0; | |
2271 | break; | |
2272 | } | |
2273 | } | |
cad321ad | 2274 | spin_unlock(&tree->lock); |
d1310b2e CM |
2275 | return bitset; |
2276 | } | |
d1310b2e | 2277 | |
7870d082 JB |
2278 | int free_io_failure(struct extent_io_tree *failure_tree, |
2279 | struct extent_io_tree *io_tree, | |
2280 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2281 | { |
2282 | int ret; | |
2283 | int err = 0; | |
4a54c8c1 | 2284 | |
47dc196a | 2285 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2286 | ret = clear_extent_bits(failure_tree, rec->start, |
2287 | rec->start + rec->len - 1, | |
91166212 | 2288 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2289 | if (ret) |
2290 | err = ret; | |
2291 | ||
7870d082 | 2292 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2293 | rec->start + rec->len - 1, |
91166212 | 2294 | EXTENT_DAMAGED); |
53b381b3 DW |
2295 | if (ret && !err) |
2296 | err = ret; | |
4a54c8c1 JS |
2297 | |
2298 | kfree(rec); | |
2299 | return err; | |
2300 | } | |
2301 | ||
4a54c8c1 JS |
2302 | /* |
2303 | * this bypasses the standard btrfs submit functions deliberately, as | |
2304 | * the standard behavior is to write all copies in a raid setup. here we only | |
2305 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2306 | * submit_bio directly. | |
3ec706c8 | 2307 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2308 | * actually prevents the read that triggered the error from finishing. |
2309 | * currently, there can be no more than two copies of every data bit. thus, | |
2310 | * exactly one rewrite is required. | |
2311 | */ | |
38d5e541 QW |
2312 | static int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2313 | u64 length, u64 logical, struct page *page, | |
2314 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2315 | { |
4a54c8c1 | 2316 | struct btrfs_device *dev; |
e9458bfe CH |
2317 | struct bio_vec bvec; |
2318 | struct bio bio; | |
4a54c8c1 JS |
2319 | u64 map_length = 0; |
2320 | u64 sector; | |
4c664611 | 2321 | struct btrfs_io_context *bioc = NULL; |
e9458bfe | 2322 | int ret = 0; |
4a54c8c1 | 2323 | |
1751e8a6 | 2324 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2325 | BUG_ON(!mirror_num); |
2326 | ||
554aed7d JT |
2327 | if (btrfs_repair_one_zone(fs_info, logical)) |
2328 | return 0; | |
f7ef5287 | 2329 | |
4a54c8c1 JS |
2330 | map_length = length; |
2331 | ||
b5de8d0d | 2332 | /* |
4c664611 | 2333 | * Avoid races with device replace and make sure our bioc has devices |
b5de8d0d FM |
2334 | * associated to its stripes that don't go away while we are doing the |
2335 | * read repair operation. | |
2336 | */ | |
2337 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2338 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2339 | /* |
2340 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2341 | * to update all raid stripes, but here we just want to correct | |
2342 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2343 | * stripe's dev and sector. | |
2344 | */ | |
2345 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
4c664611 | 2346 | &map_length, &bioc, 0); |
e9458bfe CH |
2347 | if (ret) |
2348 | goto out_counter_dec; | |
4c664611 | 2349 | ASSERT(bioc->mirror_num == 1); |
c725328c LB |
2350 | } else { |
2351 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
4c664611 | 2352 | &map_length, &bioc, mirror_num); |
e9458bfe CH |
2353 | if (ret) |
2354 | goto out_counter_dec; | |
4c664611 | 2355 | BUG_ON(mirror_num != bioc->mirror_num); |
4a54c8c1 | 2356 | } |
c725328c | 2357 | |
4c664611 | 2358 | sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9; |
4c664611 QW |
2359 | dev = bioc->stripes[bioc->mirror_num - 1].dev; |
2360 | btrfs_put_bioc(bioc); | |
e9458bfe | 2361 | |
ebbede42 AJ |
2362 | if (!dev || !dev->bdev || |
2363 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
e9458bfe CH |
2364 | ret = -EIO; |
2365 | goto out_counter_dec; | |
4a54c8c1 | 2366 | } |
4a54c8c1 | 2367 | |
e9458bfe CH |
2368 | bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC); |
2369 | bio.bi_iter.bi_sector = sector; | |
2370 | __bio_add_page(&bio, page, length, pg_offset); | |
2371 | ||
2372 | btrfsic_check_bio(&bio); | |
2373 | ret = submit_bio_wait(&bio); | |
2374 | if (ret) { | |
4a54c8c1 | 2375 | /* try to remap that extent elsewhere? */ |
442a4f63 | 2376 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
e9458bfe | 2377 | goto out_bio_uninit; |
4a54c8c1 JS |
2378 | } |
2379 | ||
b14af3b4 DS |
2380 | btrfs_info_rl_in_rcu(fs_info, |
2381 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2382 | ino, start, |
1203b681 | 2383 | rcu_str_deref(dev->name), sector); |
e9458bfe CH |
2384 | ret = 0; |
2385 | ||
2386 | out_bio_uninit: | |
2387 | bio_uninit(&bio); | |
2388 | out_counter_dec: | |
b5de8d0d | 2389 | btrfs_bio_counter_dec(fs_info); |
e9458bfe | 2390 | return ret; |
4a54c8c1 JS |
2391 | } |
2392 | ||
2b48966a | 2393 | int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num) |
ea466794 | 2394 | { |
20a1fbf9 | 2395 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2396 | u64 start = eb->start; |
cc5e31a4 | 2397 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2398 | int ret = 0; |
ea466794 | 2399 | |
bc98a42c | 2400 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2401 | return -EROFS; |
2402 | ||
ea466794 | 2403 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2404 | struct page *p = eb->pages[i]; |
1203b681 | 2405 | |
6ec656bc | 2406 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2407 | start - page_offset(p), mirror_num); |
ea466794 JB |
2408 | if (ret) |
2409 | break; | |
09cbfeaf | 2410 | start += PAGE_SIZE; |
ea466794 JB |
2411 | } |
2412 | ||
2413 | return ret; | |
2414 | } | |
2415 | ||
c144c63f CH |
2416 | static int next_mirror(const struct io_failure_record *failrec, int cur_mirror) |
2417 | { | |
2418 | if (cur_mirror == failrec->num_copies) | |
2419 | return cur_mirror + 1 - failrec->num_copies; | |
2420 | return cur_mirror + 1; | |
2421 | } | |
2422 | ||
2423 | static int prev_mirror(const struct io_failure_record *failrec, int cur_mirror) | |
2424 | { | |
2425 | if (cur_mirror == 1) | |
2426 | return failrec->num_copies; | |
2427 | return cur_mirror - 1; | |
2428 | } | |
2429 | ||
4a54c8c1 JS |
2430 | /* |
2431 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2432 | * to see if we need to process or clean up an io_failure_record | |
2433 | */ | |
7870d082 JB |
2434 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2435 | struct extent_io_tree *failure_tree, | |
2436 | struct extent_io_tree *io_tree, u64 start, | |
2437 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2438 | { |
2439 | u64 private; | |
4a54c8c1 | 2440 | struct io_failure_record *failrec; |
4a54c8c1 | 2441 | struct extent_state *state; |
c144c63f | 2442 | int mirror; |
4a54c8c1 | 2443 | int ret; |
4a54c8c1 JS |
2444 | |
2445 | private = 0; | |
7870d082 JB |
2446 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2447 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2448 | if (!ret) |
2449 | return 0; | |
2450 | ||
2279a270 NB |
2451 | failrec = get_state_failrec(failure_tree, start); |
2452 | if (IS_ERR(failrec)) | |
4a54c8c1 JS |
2453 | return 0; |
2454 | ||
4a54c8c1 JS |
2455 | BUG_ON(!failrec->this_mirror); |
2456 | ||
bc98a42c | 2457 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2458 | goto out; |
4a54c8c1 | 2459 | |
7870d082 JB |
2460 | spin_lock(&io_tree->lock); |
2461 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2462 | failrec->start, |
2463 | EXTENT_LOCKED); | |
7870d082 | 2464 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2465 | |
c144c63f CH |
2466 | if (!state || state->start > failrec->start || |
2467 | state->end < failrec->start + failrec->len - 1) | |
2468 | goto out; | |
2469 | ||
2470 | mirror = failrec->this_mirror; | |
2471 | do { | |
2472 | mirror = prev_mirror(failrec, mirror); | |
2473 | repair_io_failure(fs_info, ino, start, failrec->len, | |
2474 | failrec->logical, page, pg_offset, mirror); | |
2475 | } while (mirror != failrec->failed_mirror); | |
4a54c8c1 JS |
2476 | |
2477 | out: | |
7870d082 | 2478 | free_io_failure(failure_tree, io_tree, failrec); |
454ff3de | 2479 | return 0; |
4a54c8c1 JS |
2480 | } |
2481 | ||
f612496b MX |
2482 | /* |
2483 | * Can be called when | |
2484 | * - hold extent lock | |
2485 | * - under ordered extent | |
2486 | * - the inode is freeing | |
2487 | */ | |
7ab7956e | 2488 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2489 | { |
7ab7956e | 2490 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2491 | struct io_failure_record *failrec; |
2492 | struct extent_state *state, *next; | |
2493 | ||
2494 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2495 | return; | |
2496 | ||
2497 | spin_lock(&failure_tree->lock); | |
2498 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2499 | while (state) { | |
2500 | if (state->start > end) | |
2501 | break; | |
2502 | ||
2503 | ASSERT(state->end <= end); | |
2504 | ||
2505 | next = next_state(state); | |
2506 | ||
47dc196a | 2507 | failrec = state->failrec; |
f612496b MX |
2508 | free_extent_state(state); |
2509 | kfree(failrec); | |
2510 | ||
2511 | state = next; | |
2512 | } | |
2513 | spin_unlock(&failure_tree->lock); | |
2514 | } | |
2515 | ||
3526302f | 2516 | static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode, |
7aa51232 CH |
2517 | struct btrfs_bio *bbio, |
2518 | unsigned int bio_offset) | |
4a54c8c1 | 2519 | { |
ab8d0fc4 | 2520 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7aa51232 | 2521 | u64 start = bbio->file_offset + bio_offset; |
2fe6303e | 2522 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2523 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2524 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
150e4b05 | 2525 | const u32 sectorsize = fs_info->sectorsize; |
4a54c8c1 | 2526 | int ret; |
4a54c8c1 | 2527 | |
2279a270 | 2528 | failrec = get_state_failrec(failure_tree, start); |
3526302f | 2529 | if (!IS_ERR(failrec)) { |
ab8d0fc4 | 2530 | btrfs_debug(fs_info, |
1245835d QW |
2531 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu", |
2532 | failrec->logical, failrec->start, failrec->len); | |
4a54c8c1 JS |
2533 | /* |
2534 | * when data can be on disk more than twice, add to failrec here | |
2535 | * (e.g. with a list for failed_mirror) to make | |
2536 | * clean_io_failure() clean all those errors at once. | |
2537 | */ | |
7aa51232 | 2538 | ASSERT(failrec->this_mirror == bbio->mirror_num); |
c144c63f | 2539 | ASSERT(failrec->len == fs_info->sectorsize); |
3526302f | 2540 | return failrec; |
4a54c8c1 | 2541 | } |
2fe6303e | 2542 | |
3526302f NB |
2543 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); |
2544 | if (!failrec) | |
2545 | return ERR_PTR(-ENOMEM); | |
2fe6303e | 2546 | |
3526302f | 2547 | failrec->start = start; |
150e4b05 | 2548 | failrec->len = sectorsize; |
7aa51232 CH |
2549 | failrec->failed_mirror = bbio->mirror_num; |
2550 | failrec->this_mirror = bbio->mirror_num; | |
81bd9328 | 2551 | failrec->logical = (bbio->iter.bi_sector << SECTOR_SHIFT) + bio_offset; |
3526302f | 2552 | |
3526302f | 2553 | btrfs_debug(fs_info, |
81bd9328 CH |
2554 | "new io failure record logical %llu start %llu", |
2555 | failrec->logical, start); | |
3526302f | 2556 | |
81bd9328 | 2557 | failrec->num_copies = btrfs_num_copies(fs_info, failrec->logical, sectorsize); |
c144c63f CH |
2558 | if (failrec->num_copies == 1) { |
2559 | /* | |
2560 | * We only have a single copy of the data, so don't bother with | |
2561 | * all the retry and error correction code that follows. No | |
2562 | * matter what the error is, it is very likely to persist. | |
2563 | */ | |
2564 | btrfs_debug(fs_info, | |
2565 | "cannot repair logical %llu num_copies %d", | |
2566 | failrec->logical, failrec->num_copies); | |
3526302f NB |
2567 | kfree(failrec); |
2568 | return ERR_PTR(-EIO); | |
2569 | } | |
2570 | ||
3526302f | 2571 | /* Set the bits in the private failure tree */ |
150e4b05 | 2572 | ret = set_extent_bits(failure_tree, start, start + sectorsize - 1, |
3526302f NB |
2573 | EXTENT_LOCKED | EXTENT_DIRTY); |
2574 | if (ret >= 0) { | |
2575 | ret = set_state_failrec(failure_tree, start, failrec); | |
2576 | /* Set the bits in the inode's tree */ | |
150e4b05 QW |
2577 | ret = set_extent_bits(tree, start, start + sectorsize - 1, |
2578 | EXTENT_DAMAGED); | |
3526302f NB |
2579 | } else if (ret < 0) { |
2580 | kfree(failrec); | |
2581 | return ERR_PTR(ret); | |
2582 | } | |
2583 | ||
2584 | return failrec; | |
2fe6303e MX |
2585 | } |
2586 | ||
7aa51232 CH |
2587 | int btrfs_repair_one_sector(struct inode *inode, struct btrfs_bio *failed_bbio, |
2588 | u32 bio_offset, struct page *page, unsigned int pgoff, | |
150e4b05 | 2589 | submit_bio_hook_t *submit_bio_hook) |
2fe6303e | 2590 | { |
7aa51232 | 2591 | u64 start = failed_bbio->file_offset + bio_offset; |
2fe6303e | 2592 | struct io_failure_record *failrec; |
77d5d689 | 2593 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2594 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2595 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
7aa51232 | 2596 | struct bio *failed_bio = &failed_bbio->bio; |
7ffd27e3 | 2597 | const int icsum = bio_offset >> fs_info->sectorsize_bits; |
77d5d689 | 2598 | struct bio *repair_bio; |
c3a3b19b | 2599 | struct btrfs_bio *repair_bbio; |
2fe6303e | 2600 | |
77d5d689 OS |
2601 | btrfs_debug(fs_info, |
2602 | "repair read error: read error at %llu", start); | |
2fe6303e | 2603 | |
1f7ad75b | 2604 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e | 2605 | |
7aa51232 | 2606 | failrec = btrfs_get_io_failure_record(inode, failed_bbio, bio_offset); |
3526302f | 2607 | if (IS_ERR(failrec)) |
150e4b05 | 2608 | return PTR_ERR(failrec); |
2fe6303e | 2609 | |
c144c63f CH |
2610 | /* |
2611 | * There are two premises: | |
2612 | * a) deliver good data to the caller | |
2613 | * b) correct the bad sectors on disk | |
2614 | * | |
2615 | * Since we're only doing repair for one sector, we only need to get | |
2616 | * a good copy of the failed sector and if we succeed, we have setup | |
2617 | * everything for repair_io_failure to do the rest for us. | |
2618 | */ | |
2619 | failrec->this_mirror = next_mirror(failrec, failrec->this_mirror); | |
2620 | if (failrec->this_mirror == failrec->failed_mirror) { | |
2621 | btrfs_debug(fs_info, | |
2622 | "failed to repair num_copies %d this_mirror %d failed_mirror %d", | |
2623 | failrec->num_copies, failrec->this_mirror, failrec->failed_mirror); | |
7870d082 | 2624 | free_io_failure(failure_tree, tree, failrec); |
150e4b05 | 2625 | return -EIO; |
2fe6303e MX |
2626 | } |
2627 | ||
917f32a2 CH |
2628 | repair_bio = btrfs_bio_alloc(1, REQ_OP_READ, failed_bbio->end_io, |
2629 | failed_bbio->private); | |
c3a3b19b | 2630 | repair_bbio = btrfs_bio(repair_bio); |
00d82525 | 2631 | repair_bbio->file_offset = start; |
77d5d689 | 2632 | repair_bio->bi_iter.bi_sector = failrec->logical >> 9; |
2fe6303e | 2633 | |
c3a3b19b | 2634 | if (failed_bbio->csum) { |
223486c2 | 2635 | const u32 csum_size = fs_info->csum_size; |
77d5d689 | 2636 | |
c3a3b19b QW |
2637 | repair_bbio->csum = repair_bbio->csum_inline; |
2638 | memcpy(repair_bbio->csum, | |
2639 | failed_bbio->csum + csum_size * icsum, csum_size); | |
77d5d689 | 2640 | } |
2fe6303e | 2641 | |
77d5d689 | 2642 | bio_add_page(repair_bio, page, failrec->len, pgoff); |
c3a3b19b | 2643 | repair_bbio->iter = repair_bio->bi_iter; |
4a54c8c1 | 2644 | |
ab8d0fc4 | 2645 | btrfs_debug(btrfs_sb(inode->i_sb), |
1245835d QW |
2646 | "repair read error: submitting new read to mirror %d", |
2647 | failrec->this_mirror); | |
4a54c8c1 | 2648 | |
8cbc3001 JB |
2649 | /* |
2650 | * At this point we have a bio, so any errors from submit_bio_hook() | |
2651 | * will be handled by the endio on the repair_bio, so we can't return an | |
2652 | * error here. | |
2653 | */ | |
81bd9328 | 2654 | submit_bio_hook(inode, repair_bio, failrec->this_mirror, 0); |
8cbc3001 | 2655 | return BLK_STS_OK; |
150e4b05 QW |
2656 | } |
2657 | ||
2658 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) | |
2659 | { | |
2660 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
2661 | ||
2662 | ASSERT(page_offset(page) <= start && | |
2663 | start + len <= page_offset(page) + PAGE_SIZE); | |
2664 | ||
150e4b05 | 2665 | if (uptodate) { |
14605409 BB |
2666 | if (fsverity_active(page->mapping->host) && |
2667 | !PageError(page) && | |
2668 | !PageUptodate(page) && | |
2669 | start < i_size_read(page->mapping->host) && | |
2670 | !fsverity_verify_page(page)) { | |
2671 | btrfs_page_set_error(fs_info, page, start, len); | |
2672 | } else { | |
2673 | btrfs_page_set_uptodate(fs_info, page, start, len); | |
2674 | } | |
150e4b05 QW |
2675 | } else { |
2676 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2677 | btrfs_page_set_error(fs_info, page, start, len); | |
2678 | } | |
2679 | ||
fbca46eb | 2680 | if (!btrfs_is_subpage(fs_info, page)) |
150e4b05 | 2681 | unlock_page(page); |
3d078efa | 2682 | else |
150e4b05 QW |
2683 | btrfs_subpage_end_reader(fs_info, page, start, len); |
2684 | } | |
2685 | ||
a5aa7ab6 CH |
2686 | static void end_sector_io(struct page *page, u64 offset, bool uptodate) |
2687 | { | |
2688 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); | |
2689 | const u32 sectorsize = inode->root->fs_info->sectorsize; | |
2690 | struct extent_state *cached = NULL; | |
2691 | ||
2692 | end_page_read(page, uptodate, offset, sectorsize); | |
2693 | if (uptodate) | |
2694 | set_extent_uptodate(&inode->io_tree, offset, | |
2695 | offset + sectorsize - 1, &cached, GFP_ATOMIC); | |
2696 | unlock_extent_cached_atomic(&inode->io_tree, offset, | |
2697 | offset + sectorsize - 1, &cached); | |
2698 | } | |
2699 | ||
7aa51232 CH |
2700 | static void submit_data_read_repair(struct inode *inode, |
2701 | struct btrfs_bio *failed_bbio, | |
fd5a6f63 | 2702 | u32 bio_offset, const struct bio_vec *bvec, |
7aa51232 | 2703 | unsigned int error_bitmap) |
150e4b05 | 2704 | { |
fd5a6f63 | 2705 | const unsigned int pgoff = bvec->bv_offset; |
150e4b05 | 2706 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
fd5a6f63 QW |
2707 | struct page *page = bvec->bv_page; |
2708 | const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset; | |
2709 | const u64 end = start + bvec->bv_len - 1; | |
150e4b05 QW |
2710 | const u32 sectorsize = fs_info->sectorsize; |
2711 | const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits; | |
150e4b05 QW |
2712 | int i; |
2713 | ||
7aa51232 | 2714 | BUG_ON(bio_op(&failed_bbio->bio) == REQ_OP_WRITE); |
150e4b05 | 2715 | |
c0111c44 QW |
2716 | /* This repair is only for data */ |
2717 | ASSERT(is_data_inode(inode)); | |
2718 | ||
150e4b05 QW |
2719 | /* We're here because we had some read errors or csum mismatch */ |
2720 | ASSERT(error_bitmap); | |
2721 | ||
2722 | /* | |
2723 | * We only get called on buffered IO, thus page must be mapped and bio | |
2724 | * must not be cloned. | |
2725 | */ | |
7aa51232 | 2726 | ASSERT(page->mapping && !bio_flagged(&failed_bbio->bio, BIO_CLONED)); |
150e4b05 QW |
2727 | |
2728 | /* Iterate through all the sectors in the range */ | |
2729 | for (i = 0; i < nr_bits; i++) { | |
2730 | const unsigned int offset = i * sectorsize; | |
150e4b05 QW |
2731 | bool uptodate = false; |
2732 | int ret; | |
2733 | ||
2734 | if (!(error_bitmap & (1U << i))) { | |
2735 | /* | |
2736 | * This sector has no error, just end the page read | |
2737 | * and unlock the range. | |
2738 | */ | |
2739 | uptodate = true; | |
2740 | goto next; | |
2741 | } | |
2742 | ||
7aa51232 CH |
2743 | ret = btrfs_repair_one_sector(inode, failed_bbio, |
2744 | bio_offset + offset, page, pgoff + offset, | |
2745 | btrfs_submit_data_read_bio); | |
150e4b05 QW |
2746 | if (!ret) { |
2747 | /* | |
2748 | * We have submitted the read repair, the page release | |
2749 | * will be handled by the endio function of the | |
2750 | * submitted repair bio. | |
2751 | * Thus we don't need to do any thing here. | |
2752 | */ | |
2753 | continue; | |
2754 | } | |
2755 | /* | |
fd5a6f63 QW |
2756 | * Continue on failed repair, otherwise the remaining sectors |
2757 | * will not be properly unlocked. | |
150e4b05 | 2758 | */ |
150e4b05 | 2759 | next: |
a5aa7ab6 | 2760 | end_sector_io(page, start + offset, uptodate); |
150e4b05 | 2761 | } |
4a54c8c1 JS |
2762 | } |
2763 | ||
d1310b2e CM |
2764 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2765 | ||
b5227c07 | 2766 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 2767 | { |
38a39ac7 | 2768 | struct btrfs_inode *inode; |
25c1252a | 2769 | const bool uptodate = (err == 0); |
3e2426bd | 2770 | int ret = 0; |
87826df0 | 2771 | |
38a39ac7 QW |
2772 | ASSERT(page && page->mapping); |
2773 | inode = BTRFS_I(page->mapping->host); | |
2774 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 2775 | |
87826df0 | 2776 | if (!uptodate) { |
963e4db8 QW |
2777 | const struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2778 | u32 len; | |
2779 | ||
2780 | ASSERT(end + 1 - start <= U32_MAX); | |
2781 | len = end + 1 - start; | |
2782 | ||
2783 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2784 | btrfs_page_set_error(fs_info, page, start, len); | |
bff5baf8 | 2785 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2786 | mapping_set_error(page->mapping, ret); |
87826df0 | 2787 | } |
87826df0 JM |
2788 | } |
2789 | ||
d1310b2e CM |
2790 | /* |
2791 | * after a writepage IO is done, we need to: | |
2792 | * clear the uptodate bits on error | |
2793 | * clear the writeback bits in the extent tree for this IO | |
2794 | * end_page_writeback if the page has no more pending IO | |
2795 | * | |
2796 | * Scheduling is not allowed, so the extent state tree is expected | |
2797 | * to have one and only one object corresponding to this IO. | |
2798 | */ | |
917f32a2 | 2799 | static void end_bio_extent_writepage(struct btrfs_bio *bbio) |
d1310b2e | 2800 | { |
917f32a2 | 2801 | struct bio *bio = &bbio->bio; |
4e4cbee9 | 2802 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2803 | struct bio_vec *bvec; |
d1310b2e CM |
2804 | u64 start; |
2805 | u64 end; | |
6dc4f100 | 2806 | struct bvec_iter_all iter_all; |
d8e3fb10 | 2807 | bool first_bvec = true; |
d1310b2e | 2808 | |
c09abff8 | 2809 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2810 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2811 | struct page *page = bvec->bv_page; |
0b246afa JM |
2812 | struct inode *inode = page->mapping->host; |
2813 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
2814 | const u32 sectorsize = fs_info->sectorsize; |
2815 | ||
2816 | /* Our read/write should always be sector aligned. */ | |
2817 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2818 | btrfs_err(fs_info, | |
2819 | "partial page write in btrfs with offset %u and length %u", | |
2820 | bvec->bv_offset, bvec->bv_len); | |
2821 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
2822 | btrfs_info(fs_info, | |
2823 | "incomplete page write with offset %u and length %u", | |
2824 | bvec->bv_offset, bvec->bv_len); | |
2825 | ||
2826 | start = page_offset(page) + bvec->bv_offset; | |
2827 | end = start + bvec->bv_len - 1; | |
d1310b2e | 2828 | |
d8e3fb10 NA |
2829 | if (first_bvec) { |
2830 | btrfs_record_physical_zoned(inode, start, bio); | |
2831 | first_bvec = false; | |
2832 | } | |
2833 | ||
4e4cbee9 | 2834 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
2835 | |
2836 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 2837 | } |
2b1f55b0 | 2838 | |
d1310b2e | 2839 | bio_put(bio); |
d1310b2e CM |
2840 | } |
2841 | ||
94e8c95c QW |
2842 | /* |
2843 | * Record previously processed extent range | |
2844 | * | |
2845 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
2846 | * the extent io operations. | |
2847 | */ | |
2848 | struct processed_extent { | |
2849 | struct btrfs_inode *inode; | |
2850 | /* Start of the range in @inode */ | |
2851 | u64 start; | |
2e626e56 | 2852 | /* End of the range in @inode */ |
94e8c95c QW |
2853 | u64 end; |
2854 | bool uptodate; | |
2855 | }; | |
2856 | ||
2857 | /* | |
2858 | * Try to release processed extent range | |
2859 | * | |
2860 | * May not release the extent range right now if the current range is | |
2861 | * contiguous to processed extent. | |
2862 | * | |
2863 | * Will release processed extent when any of @inode, @uptodate, the range is | |
2864 | * no longer contiguous to the processed range. | |
2865 | * | |
2866 | * Passing @inode == NULL will force processed extent to be released. | |
2867 | */ | |
2868 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
2869 | struct btrfs_inode *inode, u64 start, u64 end, | |
2870 | bool uptodate) | |
883d0de4 MX |
2871 | { |
2872 | struct extent_state *cached = NULL; | |
94e8c95c QW |
2873 | struct extent_io_tree *tree; |
2874 | ||
2875 | /* The first extent, initialize @processed */ | |
2876 | if (!processed->inode) | |
2877 | goto update; | |
883d0de4 | 2878 | |
94e8c95c QW |
2879 | /* |
2880 | * Contiguous to processed extent, just uptodate the end. | |
2881 | * | |
2882 | * Several things to notice: | |
2883 | * | |
2884 | * - bio can be merged as long as on-disk bytenr is contiguous | |
2885 | * This means we can have page belonging to other inodes, thus need to | |
2886 | * check if the inode still matches. | |
2887 | * - bvec can contain range beyond current page for multi-page bvec | |
2888 | * Thus we need to do processed->end + 1 >= start check | |
2889 | */ | |
2890 | if (processed->inode == inode && processed->uptodate == uptodate && | |
2891 | processed->end + 1 >= start && end >= processed->end) { | |
2892 | processed->end = end; | |
2893 | return; | |
2894 | } | |
2895 | ||
2896 | tree = &processed->inode->io_tree; | |
2897 | /* | |
2898 | * Now we don't have range contiguous to the processed range, release | |
2899 | * the processed range now. | |
2900 | */ | |
94e8c95c QW |
2901 | unlock_extent_cached_atomic(tree, processed->start, processed->end, |
2902 | &cached); | |
2903 | ||
2904 | update: | |
2905 | /* Update processed to current range */ | |
2906 | processed->inode = inode; | |
2907 | processed->start = start; | |
2908 | processed->end = end; | |
2909 | processed->uptodate = uptodate; | |
883d0de4 MX |
2910 | } |
2911 | ||
92082d40 QW |
2912 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
2913 | { | |
2914 | ASSERT(PageLocked(page)); | |
fbca46eb | 2915 | if (!btrfs_is_subpage(fs_info, page)) |
92082d40 QW |
2916 | return; |
2917 | ||
2918 | ASSERT(PagePrivate(page)); | |
2919 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
2920 | } | |
2921 | ||
d9bb77d5 | 2922 | /* |
01cd3909 | 2923 | * Find extent buffer for a givne bytenr. |
d9bb77d5 QW |
2924 | * |
2925 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
2926 | * in endio context. | |
2927 | */ | |
2928 | static struct extent_buffer *find_extent_buffer_readpage( | |
2929 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
2930 | { | |
2931 | struct extent_buffer *eb; | |
2932 | ||
2933 | /* | |
2934 | * For regular sectorsize, we can use page->private to grab extent | |
2935 | * buffer | |
2936 | */ | |
fbca46eb | 2937 | if (fs_info->nodesize >= PAGE_SIZE) { |
d9bb77d5 QW |
2938 | ASSERT(PagePrivate(page) && page->private); |
2939 | return (struct extent_buffer *)page->private; | |
2940 | } | |
2941 | ||
01cd3909 DS |
2942 | /* For subpage case, we need to lookup buffer radix tree */ |
2943 | rcu_read_lock(); | |
2944 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
2945 | bytenr >> fs_info->sectorsize_bits); | |
2946 | rcu_read_unlock(); | |
d9bb77d5 QW |
2947 | ASSERT(eb); |
2948 | return eb; | |
2949 | } | |
2950 | ||
d1310b2e CM |
2951 | /* |
2952 | * after a readpage IO is done, we need to: | |
2953 | * clear the uptodate bits on error | |
2954 | * set the uptodate bits if things worked | |
2955 | * set the page up to date if all extents in the tree are uptodate | |
2956 | * clear the lock bit in the extent tree | |
2957 | * unlock the page if there are no other extents locked for it | |
2958 | * | |
2959 | * Scheduling is not allowed, so the extent state tree is expected | |
2960 | * to have one and only one object corresponding to this IO. | |
2961 | */ | |
917f32a2 | 2962 | static void end_bio_extent_readpage(struct btrfs_bio *bbio) |
d1310b2e | 2963 | { |
917f32a2 | 2964 | struct bio *bio = &bbio->bio; |
2c30c71b | 2965 | struct bio_vec *bvec; |
7870d082 | 2966 | struct extent_io_tree *tree, *failure_tree; |
94e8c95c | 2967 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
2968 | /* |
2969 | * The offset to the beginning of a bio, since one bio can never be | |
2970 | * larger than UINT_MAX, u32 here is enough. | |
2971 | */ | |
2972 | u32 bio_offset = 0; | |
5cf1ab56 | 2973 | int mirror; |
6dc4f100 | 2974 | struct bvec_iter_all iter_all; |
d1310b2e | 2975 | |
c09abff8 | 2976 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2977 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 2978 | bool uptodate = !bio->bi_status; |
d1310b2e | 2979 | struct page *page = bvec->bv_page; |
a71754fc | 2980 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 2981 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 | 2982 | const u32 sectorsize = fs_info->sectorsize; |
150e4b05 | 2983 | unsigned int error_bitmap = (unsigned int)-1; |
97861cd1 | 2984 | bool repair = false; |
7ffd27e3 QW |
2985 | u64 start; |
2986 | u64 end; | |
2987 | u32 len; | |
507903b8 | 2988 | |
ab8d0fc4 JM |
2989 | btrfs_debug(fs_info, |
2990 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 2991 | bio->bi_iter.bi_sector, bio->bi_status, |
c3a3b19b | 2992 | bbio->mirror_num); |
a71754fc | 2993 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2994 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 2995 | |
8b8bbd46 QW |
2996 | /* |
2997 | * We always issue full-sector reads, but if some block in a | |
2998 | * page fails to read, blk_update_request() will advance | |
2999 | * bv_offset and adjust bv_len to compensate. Print a warning | |
3000 | * for unaligned offsets, and an error if they don't add up to | |
3001 | * a full sector. | |
3002 | */ | |
3003 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
3004 | btrfs_err(fs_info, | |
3005 | "partial page read in btrfs with offset %u and length %u", | |
3006 | bvec->bv_offset, bvec->bv_len); | |
3007 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
3008 | sectorsize)) | |
3009 | btrfs_info(fs_info, | |
3010 | "incomplete page read with offset %u and length %u", | |
3011 | bvec->bv_offset, bvec->bv_len); | |
3012 | ||
3013 | start = page_offset(page) + bvec->bv_offset; | |
3014 | end = start + bvec->bv_len - 1; | |
facc8a22 | 3015 | len = bvec->bv_len; |
d1310b2e | 3016 | |
c3a3b19b | 3017 | mirror = bbio->mirror_num; |
78e62c02 | 3018 | if (likely(uptodate)) { |
150e4b05 | 3019 | if (is_data_inode(inode)) { |
c3a3b19b | 3020 | error_bitmap = btrfs_verify_data_csum(bbio, |
5e295768 | 3021 | bio_offset, page, start, end); |
97861cd1 CH |
3022 | if (error_bitmap) |
3023 | uptodate = false; | |
150e4b05 | 3024 | } else { |
97861cd1 CH |
3025 | if (btrfs_validate_metadata_buffer(bbio, |
3026 | page, start, end, mirror)) | |
3027 | uptodate = false; | |
150e4b05 | 3028 | } |
d1310b2e | 3029 | } |
ea466794 | 3030 | |
883d0de4 | 3031 | if (likely(uptodate)) { |
a71754fc | 3032 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3033 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 3034 | |
97861cd1 CH |
3035 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
3036 | failure_tree, tree, start, page, | |
3037 | btrfs_ino(BTRFS_I(inode)), 0); | |
3038 | ||
c28ea613 QW |
3039 | /* |
3040 | * Zero out the remaining part if this range straddles | |
3041 | * i_size. | |
3042 | * | |
3043 | * Here we should only zero the range inside the bvec, | |
3044 | * not touch anything else. | |
3045 | * | |
3046 | * NOTE: i_size is exclusive while end is inclusive. | |
3047 | */ | |
3048 | if (page->index == end_index && i_size <= end) { | |
3049 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 3050 | offset_in_page(start)); |
c28ea613 QW |
3051 | |
3052 | zero_user_segment(page, zero_start, | |
3053 | offset_in_page(end) + 1); | |
3054 | } | |
97861cd1 CH |
3055 | } else if (is_data_inode(inode)) { |
3056 | /* | |
3057 | * Only try to repair bios that actually made it to a | |
3058 | * device. If the bio failed to be submitted mirror | |
3059 | * is 0 and we need to fail it without retrying. | |
81bd9328 CH |
3060 | * |
3061 | * This also includes the high level bios for compressed | |
3062 | * extents - these never make it to a device and repair | |
3063 | * is already handled on the lower compressed bio. | |
97861cd1 CH |
3064 | */ |
3065 | if (mirror > 0) | |
3066 | repair = true; | |
3067 | } else { | |
3068 | struct extent_buffer *eb; | |
3069 | ||
3070 | eb = find_extent_buffer_readpage(fs_info, page, start); | |
3071 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
3072 | eb->read_mirror = mirror; | |
3073 | atomic_dec(&eb->io_pages); | |
70dec807 | 3074 | } |
97861cd1 CH |
3075 | |
3076 | if (repair) { | |
3077 | /* | |
3078 | * submit_data_read_repair() will handle all the good | |
3079 | * and bad sectors, we just continue to the next bvec. | |
3080 | */ | |
7aa51232 CH |
3081 | submit_data_read_repair(inode, bbio, bio_offset, bvec, |
3082 | error_bitmap); | |
97861cd1 CH |
3083 | } else { |
3084 | /* Update page status and unlock */ | |
3085 | end_page_read(page, uptodate, start, len); | |
3086 | endio_readpage_release_extent(&processed, BTRFS_I(inode), | |
3087 | start, end, PageUptodate(page)); | |
70dec807 | 3088 | } |
97861cd1 | 3089 | |
7ffd27e3 QW |
3090 | ASSERT(bio_offset + len > bio_offset); |
3091 | bio_offset += len; | |
883d0de4 | 3092 | |
2c30c71b | 3093 | } |
94e8c95c QW |
3094 | /* Release the last extent */ |
3095 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
c3a3b19b | 3096 | btrfs_bio_free_csum(bbio); |
d1310b2e | 3097 | bio_put(bio); |
d1310b2e CM |
3098 | } |
3099 | ||
dd137dd1 STD |
3100 | /** |
3101 | * Populate every free slot in a provided array with pages. | |
3102 | * | |
3103 | * @nr_pages: number of pages to allocate | |
3104 | * @page_array: the array to fill with pages; any existing non-null entries in | |
3105 | * the array will be skipped | |
3106 | * | |
3107 | * Return: 0 if all pages were able to be allocated; | |
3108 | * -ENOMEM otherwise, and the caller is responsible for freeing all | |
3109 | * non-null page pointers in the array. | |
3110 | */ | |
3111 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array) | |
3112 | { | |
91d6ac1d | 3113 | unsigned int allocated; |
dd137dd1 | 3114 | |
91d6ac1d STD |
3115 | for (allocated = 0; allocated < nr_pages;) { |
3116 | unsigned int last = allocated; | |
dd137dd1 | 3117 | |
91d6ac1d STD |
3118 | allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array); |
3119 | ||
395cb57e STD |
3120 | if (allocated == nr_pages) |
3121 | return 0; | |
3122 | ||
91d6ac1d STD |
3123 | /* |
3124 | * During this iteration, no page could be allocated, even | |
3125 | * though alloc_pages_bulk_array() falls back to alloc_page() | |
3126 | * if it could not bulk-allocate. So we must be out of memory. | |
3127 | */ | |
3128 | if (allocated == last) | |
dd137dd1 | 3129 | return -ENOMEM; |
395cb57e STD |
3130 | |
3131 | memalloc_retry_wait(GFP_NOFS); | |
dd137dd1 STD |
3132 | } |
3133 | return 0; | |
3134 | } | |
3135 | ||
953651eb NA |
3136 | /** |
3137 | * Attempt to add a page to bio | |
3138 | * | |
be8d1a2a | 3139 | * @bio_ctrl: record both the bio, and its bio_flags |
953651eb NA |
3140 | * @page: page to add to the bio |
3141 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
3142 | * a contiguous page to the previous one | |
953651eb | 3143 | * @size: portion of page that we want to write |
be8d1a2a | 3144 | * @pg_offset: starting offset in the page |
cb3a12d9 | 3145 | * @compress_type: compression type of the current bio to see if we can merge them |
953651eb NA |
3146 | * |
3147 | * Attempt to add a page to bio considering stripe alignment etc. | |
3148 | * | |
e0eefe07 QW |
3149 | * Return >= 0 for the number of bytes added to the bio. |
3150 | * Can return 0 if the current bio is already at stripe/zone boundary. | |
3151 | * Return <0 for error. | |
953651eb | 3152 | */ |
e0eefe07 QW |
3153 | static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
3154 | struct page *page, | |
3155 | u64 disk_bytenr, unsigned int size, | |
3156 | unsigned int pg_offset, | |
cb3a12d9 | 3157 | enum btrfs_compression_type compress_type) |
953651eb | 3158 | { |
390ed29b QW |
3159 | struct bio *bio = bio_ctrl->bio; |
3160 | u32 bio_size = bio->bi_iter.bi_size; | |
e0eefe07 | 3161 | u32 real_size; |
953651eb | 3162 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
4a445b7b | 3163 | bool contig = false; |
e1326f03 | 3164 | int ret; |
953651eb | 3165 | |
390ed29b QW |
3166 | ASSERT(bio); |
3167 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
3168 | ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary); | |
0f07003b | 3169 | if (bio_ctrl->compress_type != compress_type) |
e0eefe07 | 3170 | return 0; |
953651eb | 3171 | |
4a445b7b QW |
3172 | |
3173 | if (bio->bi_iter.bi_size == 0) { | |
3174 | /* We can always add a page into an empty bio. */ | |
3175 | contig = true; | |
3176 | } else if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE) { | |
3177 | struct bio_vec *bvec = bio_last_bvec_all(bio); | |
3178 | ||
3179 | /* | |
3180 | * The contig check requires the following conditions to be met: | |
3181 | * 1) The pages are belonging to the same inode | |
3182 | * This is implied by the call chain. | |
3183 | * | |
3184 | * 2) The range has adjacent logical bytenr | |
3185 | * | |
3186 | * 3) The range has adjacent file offset | |
3187 | * This is required for the usage of btrfs_bio->file_offset. | |
3188 | */ | |
3189 | if (bio_end_sector(bio) == sector && | |
3190 | page_offset(bvec->bv_page) + bvec->bv_offset + | |
3191 | bvec->bv_len == page_offset(page) + pg_offset) | |
3192 | contig = true; | |
3193 | } else { | |
3194 | /* | |
3195 | * For compression, all IO should have its logical bytenr | |
3196 | * set to the starting bytenr of the compressed extent. | |
3197 | */ | |
953651eb | 3198 | contig = bio->bi_iter.bi_sector == sector; |
4a445b7b QW |
3199 | } |
3200 | ||
953651eb | 3201 | if (!contig) |
e0eefe07 | 3202 | return 0; |
953651eb | 3203 | |
e0eefe07 QW |
3204 | real_size = min(bio_ctrl->len_to_oe_boundary, |
3205 | bio_ctrl->len_to_stripe_boundary) - bio_size; | |
3206 | real_size = min(real_size, size); | |
3207 | ||
3208 | /* | |
3209 | * If real_size is 0, never call bio_add_*_page(), as even size is 0, | |
3210 | * bio will still execute its endio function on the page! | |
3211 | */ | |
3212 | if (real_size == 0) | |
3213 | return 0; | |
953651eb | 3214 | |
390ed29b | 3215 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
e0eefe07 | 3216 | ret = bio_add_zone_append_page(bio, page, real_size, pg_offset); |
390ed29b | 3217 | else |
e0eefe07 | 3218 | ret = bio_add_page(bio, page, real_size, pg_offset); |
e1326f03 | 3219 | |
e0eefe07 | 3220 | return ret; |
953651eb NA |
3221 | } |
3222 | ||
390ed29b | 3223 | static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
939c7feb | 3224 | struct btrfs_inode *inode, u64 file_offset) |
390ed29b QW |
3225 | { |
3226 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3227 | struct btrfs_io_geometry geom; | |
3228 | struct btrfs_ordered_extent *ordered; | |
3229 | struct extent_map *em; | |
3230 | u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT); | |
3231 | int ret; | |
3232 | ||
3233 | /* | |
3234 | * Pages for compressed extent are never submitted to disk directly, | |
3235 | * thus it has no real boundary, just set them to U32_MAX. | |
3236 | * | |
3237 | * The split happens for real compressed bio, which happens in | |
3238 | * btrfs_submit_compressed_read/write(). | |
3239 | */ | |
0f07003b | 3240 | if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) { |
390ed29b QW |
3241 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3242 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3243 | return 0; | |
3244 | } | |
3245 | em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize); | |
3246 | if (IS_ERR(em)) | |
3247 | return PTR_ERR(em); | |
3248 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio), | |
3249 | logical, &geom); | |
3250 | free_extent_map(em); | |
3251 | if (ret < 0) { | |
3252 | return ret; | |
3253 | } | |
3254 | if (geom.len > U32_MAX) | |
3255 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3256 | else | |
3257 | bio_ctrl->len_to_stripe_boundary = (u32)geom.len; | |
3258 | ||
73672710 | 3259 | if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) { |
390ed29b QW |
3260 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3261 | return 0; | |
3262 | } | |
3263 | ||
390ed29b | 3264 | /* Ordered extent not yet created, so we're good */ |
939c7feb | 3265 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
390ed29b QW |
3266 | if (!ordered) { |
3267 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3268 | return 0; | |
3269 | } | |
3270 | ||
3271 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
3272 | ordered->disk_bytenr + ordered->disk_num_bytes - logical); | |
3273 | btrfs_put_ordered_extent(ordered); | |
3274 | return 0; | |
3275 | } | |
3276 | ||
e0eefe07 QW |
3277 | static int alloc_new_bio(struct btrfs_inode *inode, |
3278 | struct btrfs_bio_ctrl *bio_ctrl, | |
3279 | struct writeback_control *wbc, | |
bf9486d6 | 3280 | blk_opf_t opf, |
917f32a2 | 3281 | btrfs_bio_end_io_t end_io_func, |
939c7feb | 3282 | u64 disk_bytenr, u32 offset, u64 file_offset, |
cb3a12d9 | 3283 | enum btrfs_compression_type compress_type) |
e0eefe07 QW |
3284 | { |
3285 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3286 | struct bio *bio; | |
3287 | int ret; | |
3288 | ||
917f32a2 | 3289 | bio = btrfs_bio_alloc(BIO_MAX_VECS, opf, end_io_func, NULL); |
e0eefe07 QW |
3290 | /* |
3291 | * For compressed page range, its disk_bytenr is always @disk_bytenr | |
3292 | * passed in, no matter if we have added any range into previous bio. | |
3293 | */ | |
cb3a12d9 | 3294 | if (compress_type != BTRFS_COMPRESS_NONE) |
cd8e0cca | 3295 | bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
e0eefe07 | 3296 | else |
cd8e0cca | 3297 | bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT; |
e0eefe07 | 3298 | bio_ctrl->bio = bio; |
0f07003b | 3299 | bio_ctrl->compress_type = compress_type; |
939c7feb NA |
3300 | ret = calc_bio_boundaries(bio_ctrl, inode, file_offset); |
3301 | if (ret < 0) | |
3302 | goto error; | |
e0eefe07 | 3303 | |
50f1cff3 CH |
3304 | if (wbc) { |
3305 | /* | |
3306 | * For Zone append we need the correct block_device that we are | |
3307 | * going to write to set in the bio to be able to respect the | |
3308 | * hardware limitation. Look it up here: | |
3309 | */ | |
3310 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { | |
3311 | struct btrfs_device *dev; | |
3312 | ||
3313 | dev = btrfs_zoned_get_device(fs_info, disk_bytenr, | |
3314 | fs_info->sectorsize); | |
3315 | if (IS_ERR(dev)) { | |
3316 | ret = PTR_ERR(dev); | |
3317 | goto error; | |
3318 | } | |
e0eefe07 | 3319 | |
50f1cff3 CH |
3320 | bio_set_dev(bio, dev->bdev); |
3321 | } else { | |
3322 | /* | |
3323 | * Otherwise pick the last added device to support | |
3324 | * cgroup writeback. For multi-device file systems this | |
3325 | * means blk-cgroup policies have to always be set on the | |
3326 | * last added/replaced device. This is a bit odd but has | |
3327 | * been like that for a long time. | |
3328 | */ | |
3329 | bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev); | |
e0eefe07 | 3330 | } |
50f1cff3 CH |
3331 | wbc_init_bio(wbc, bio); |
3332 | } else { | |
3333 | ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND); | |
e0eefe07 QW |
3334 | } |
3335 | return 0; | |
3336 | error: | |
3337 | bio_ctrl->bio = NULL; | |
917f32a2 | 3338 | btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); |
e0eefe07 QW |
3339 | return ret; |
3340 | } | |
3341 | ||
4b81ba48 DS |
3342 | /* |
3343 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3344 | * @wbc: optional writeback control for io accounting |
3345 | * @page: page to add to the bio | |
0c64c33c QW |
3346 | * @disk_bytenr: logical bytenr where the write will be |
3347 | * @size: portion of page that we want to write to | |
b8b3d625 DS |
3348 | * @pg_offset: offset of the new bio or to check whether we are adding |
3349 | * a contiguous page to the previous one | |
5c2b1fd7 | 3350 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3351 | * @end_io_func: end_io callback for new bio |
3352 | * @mirror_num: desired mirror to read/write | |
3353 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
cb3a12d9 | 3354 | * @compress_type: compress type for current bio |
4b81ba48 | 3355 | */ |
bf9486d6 | 3356 | static int submit_extent_page(blk_opf_t opf, |
da2f0f74 | 3357 | struct writeback_control *wbc, |
390ed29b | 3358 | struct btrfs_bio_ctrl *bio_ctrl, |
0c64c33c | 3359 | struct page *page, u64 disk_bytenr, |
6c5a4e2c | 3360 | size_t size, unsigned long pg_offset, |
917f32a2 | 3361 | btrfs_bio_end_io_t end_io_func, |
cb3a12d9 | 3362 | enum btrfs_compression_type compress_type, |
005efedf | 3363 | bool force_bio_submit) |
d1310b2e CM |
3364 | { |
3365 | int ret = 0; | |
e1326f03 | 3366 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
e0eefe07 | 3367 | unsigned int cur = pg_offset; |
d1310b2e | 3368 | |
390ed29b | 3369 | ASSERT(bio_ctrl); |
5c2b1fd7 | 3370 | |
390ed29b QW |
3371 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
3372 | pg_offset + size <= PAGE_SIZE); | |
722c82ac CH |
3373 | if (force_bio_submit) |
3374 | submit_one_bio(bio_ctrl); | |
e0eefe07 QW |
3375 | |
3376 | while (cur < pg_offset + size) { | |
3377 | u32 offset = cur - pg_offset; | |
3378 | int added; | |
3379 | ||
3380 | /* Allocate new bio if needed */ | |
3381 | if (!bio_ctrl->bio) { | |
3382 | ret = alloc_new_bio(inode, bio_ctrl, wbc, opf, | |
3383 | end_io_func, disk_bytenr, offset, | |
939c7feb | 3384 | page_offset(page) + cur, |
cb3a12d9 | 3385 | compress_type); |
e0eefe07 QW |
3386 | if (ret < 0) |
3387 | return ret; | |
3388 | } | |
3389 | /* | |
3390 | * We must go through btrfs_bio_add_page() to ensure each | |
3391 | * page range won't cross various boundaries. | |
3392 | */ | |
cb3a12d9 | 3393 | if (compress_type != BTRFS_COMPRESS_NONE) |
e0eefe07 QW |
3394 | added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, |
3395 | size - offset, pg_offset + offset, | |
cb3a12d9 | 3396 | compress_type); |
e0eefe07 QW |
3397 | else |
3398 | added = btrfs_bio_add_page(bio_ctrl, page, | |
3399 | disk_bytenr + offset, size - offset, | |
cb3a12d9 | 3400 | pg_offset + offset, compress_type); |
e0eefe07 QW |
3401 | |
3402 | /* Metadata page range should never be split */ | |
3403 | if (!is_data_inode(&inode->vfs_inode)) | |
3404 | ASSERT(added == 0 || added == size - offset); | |
3405 | ||
3406 | /* At least we added some page, update the account */ | |
3407 | if (wbc && added) | |
3408 | wbc_account_cgroup_owner(wbc, page, added); | |
3409 | ||
3410 | /* We have reached boundary, submit right now */ | |
3411 | if (added < size - offset) { | |
3412 | /* The bio should contain some page(s) */ | |
3413 | ASSERT(bio_ctrl->bio->bi_iter.bi_size); | |
722c82ac | 3414 | submit_one_bio(bio_ctrl); |
d1310b2e | 3415 | } |
e0eefe07 | 3416 | cur += added; |
d1310b2e | 3417 | } |
e0eefe07 | 3418 | return 0; |
d1310b2e CM |
3419 | } |
3420 | ||
760f991f QW |
3421 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
3422 | struct page *page, | |
3423 | struct btrfs_subpage *prealloc) | |
d1310b2e | 3424 | { |
760f991f QW |
3425 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3426 | int ret = 0; | |
3427 | ||
0d01e247 QW |
3428 | /* |
3429 | * If the page is mapped to btree inode, we should hold the private | |
3430 | * lock to prevent race. | |
3431 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3432 | * will not race with any other ebs. | |
3433 | */ | |
3434 | if (page->mapping) | |
3435 | lockdep_assert_held(&page->mapping->private_lock); | |
3436 | ||
fbca46eb | 3437 | if (fs_info->nodesize >= PAGE_SIZE) { |
760f991f QW |
3438 | if (!PagePrivate(page)) |
3439 | attach_page_private(page, eb); | |
3440 | else | |
3441 | WARN_ON(page->private != (unsigned long)eb); | |
3442 | return 0; | |
3443 | } | |
3444 | ||
3445 | /* Already mapped, just free prealloc */ | |
3446 | if (PagePrivate(page)) { | |
3447 | btrfs_free_subpage(prealloc); | |
3448 | return 0; | |
3449 | } | |
3450 | ||
3451 | if (prealloc) | |
3452 | /* Has preallocated memory for subpage */ | |
3453 | attach_page_private(page, prealloc); | |
d1b89bc0 | 3454 | else |
760f991f QW |
3455 | /* Do new allocation to attach subpage */ |
3456 | ret = btrfs_attach_subpage(fs_info, page, | |
3457 | BTRFS_SUBPAGE_METADATA); | |
3458 | return ret; | |
d1310b2e CM |
3459 | } |
3460 | ||
32443de3 | 3461 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 3462 | { |
32443de3 QW |
3463 | struct btrfs_fs_info *fs_info; |
3464 | ||
3465 | ASSERT(page->mapping); | |
3466 | ||
3467 | if (PagePrivate(page)) | |
3468 | return 0; | |
3469 | ||
3470 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3471 | ||
fbca46eb | 3472 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3473 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); |
3474 | ||
3475 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
3476 | return 0; | |
3477 | } | |
3478 | ||
3479 | void clear_page_extent_mapped(struct page *page) | |
3480 | { | |
3481 | struct btrfs_fs_info *fs_info; | |
3482 | ||
3483 | ASSERT(page->mapping); | |
3484 | ||
d1b89bc0 | 3485 | if (!PagePrivate(page)) |
32443de3 QW |
3486 | return; |
3487 | ||
3488 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
fbca46eb | 3489 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3490 | return btrfs_detach_subpage(fs_info, page); |
3491 | ||
3492 | detach_page_private(page); | |
d1310b2e CM |
3493 | } |
3494 | ||
125bac01 MX |
3495 | static struct extent_map * |
3496 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3497 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3498 | { |
3499 | struct extent_map *em; | |
3500 | ||
3501 | if (em_cached && *em_cached) { | |
3502 | em = *em_cached; | |
cbc0e928 | 3503 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3504 | start < extent_map_end(em)) { |
490b54d6 | 3505 | refcount_inc(&em->refs); |
125bac01 MX |
3506 | return em; |
3507 | } | |
3508 | ||
3509 | free_extent_map(em); | |
3510 | *em_cached = NULL; | |
3511 | } | |
3512 | ||
1a5ee1e6 | 3513 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
c0347550 | 3514 | if (em_cached && !IS_ERR(em)) { |
125bac01 | 3515 | BUG_ON(*em_cached); |
490b54d6 | 3516 | refcount_inc(&em->refs); |
125bac01 MX |
3517 | *em_cached = em; |
3518 | } | |
3519 | return em; | |
3520 | } | |
d1310b2e CM |
3521 | /* |
3522 | * basic readpage implementation. Locked extent state structs are inserted | |
3523 | * into the tree that are removed when the IO is done (by the end_io | |
3524 | * handlers) | |
79787eaa | 3525 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3526 | * return 0 on success, otherwise return error |
d1310b2e | 3527 | */ |
7aab8b32 | 3528 | static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
390ed29b | 3529 | struct btrfs_bio_ctrl *bio_ctrl, |
bf9486d6 | 3530 | blk_opf_t read_flags, u64 *prev_em_start) |
d1310b2e CM |
3531 | { |
3532 | struct inode *inode = page->mapping->host; | |
92082d40 | 3533 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 3534 | u64 start = page_offset(page); |
8eec8296 | 3535 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3536 | u64 cur = start; |
3537 | u64 extent_offset; | |
3538 | u64 last_byte = i_size_read(inode); | |
3539 | u64 block_start; | |
d1310b2e | 3540 | struct extent_map *em; |
baf863b9 | 3541 | int ret = 0; |
306e16ce | 3542 | size_t pg_offset = 0; |
d1310b2e CM |
3543 | size_t iosize; |
3544 | size_t blocksize = inode->i_sb->s_blocksize; | |
f657a31c | 3545 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3546 | |
32443de3 QW |
3547 | ret = set_page_extent_mapped(page); |
3548 | if (ret < 0) { | |
3549 | unlock_extent(tree, start, end); | |
92082d40 QW |
3550 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
3551 | unlock_page(page); | |
32443de3 QW |
3552 | goto out; |
3553 | } | |
d1310b2e | 3554 | |
09cbfeaf | 3555 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 3556 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3557 | |
3558 | if (zero_offset) { | |
09cbfeaf | 3559 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 3560 | memzero_page(page, zero_offset, iosize); |
c8b97818 CM |
3561 | } |
3562 | } | |
92082d40 | 3563 | begin_page_read(fs_info, page); |
d1310b2e | 3564 | while (cur <= end) { |
4c37a793 | 3565 | unsigned long this_bio_flag = 0; |
005efedf | 3566 | bool force_bio_submit = false; |
0c64c33c | 3567 | u64 disk_bytenr; |
c8f2f24b | 3568 | |
6a404910 | 3569 | ASSERT(IS_ALIGNED(cur, fs_info->sectorsize)); |
d1310b2e | 3570 | if (cur >= last_byte) { |
507903b8 AJ |
3571 | struct extent_state *cached = NULL; |
3572 | ||
09cbfeaf | 3573 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 3574 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3575 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3576 | &cached, GFP_NOFS); |
7f042a83 | 3577 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3578 | cur + iosize - 1, &cached); |
92082d40 | 3579 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
3580 | break; |
3581 | } | |
125bac01 | 3582 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3583 | end - cur + 1, em_cached); |
c0347550 | 3584 | if (IS_ERR(em)) { |
7f042a83 | 3585 | unlock_extent(tree, cur, end); |
92082d40 | 3586 | end_page_read(page, false, cur, end + 1 - cur); |
bbf0ea7e | 3587 | ret = PTR_ERR(em); |
d1310b2e CM |
3588 | break; |
3589 | } | |
d1310b2e CM |
3590 | extent_offset = cur - em->start; |
3591 | BUG_ON(extent_map_end(em) <= cur); | |
3592 | BUG_ON(end < cur); | |
3593 | ||
7f6ca7f2 DS |
3594 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
3595 | this_bio_flag = em->compress_type; | |
c8b97818 | 3596 | |
d1310b2e | 3597 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
fda2832f | 3598 | iosize = ALIGN(iosize, blocksize); |
2a5232a8 | 3599 | if (this_bio_flag != BTRFS_COMPRESS_NONE) |
0c64c33c | 3600 | disk_bytenr = em->block_start; |
949b3273 | 3601 | else |
0c64c33c | 3602 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 3603 | block_start = em->block_start; |
d899e052 YZ |
3604 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3605 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3606 | |
3607 | /* | |
3608 | * If we have a file range that points to a compressed extent | |
260db43c | 3609 | * and it's followed by a consecutive file range that points |
005efedf FM |
3610 | * to the same compressed extent (possibly with a different |
3611 | * offset and/or length, so it either points to the whole extent | |
3612 | * or only part of it), we must make sure we do not submit a | |
3613 | * single bio to populate the pages for the 2 ranges because | |
3614 | * this makes the compressed extent read zero out the pages | |
3615 | * belonging to the 2nd range. Imagine the following scenario: | |
3616 | * | |
3617 | * File layout | |
3618 | * [0 - 8K] [8K - 24K] | |
3619 | * | | | |
3620 | * | | | |
3621 | * points to extent X, points to extent X, | |
3622 | * offset 4K, length of 8K offset 0, length 16K | |
3623 | * | |
3624 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3625 | * | |
3626 | * If the bio to read the compressed extent covers both ranges, | |
3627 | * it will decompress extent X into the pages belonging to the | |
3628 | * first range and then it will stop, zeroing out the remaining | |
3629 | * pages that belong to the other range that points to extent X. | |
3630 | * So here we make sure we submit 2 bios, one for the first | |
3631 | * range and another one for the third range. Both will target | |
3632 | * the same physical extent from disk, but we can't currently | |
3633 | * make the compressed bio endio callback populate the pages | |
3634 | * for both ranges because each compressed bio is tightly | |
3635 | * coupled with a single extent map, and each range can have | |
3636 | * an extent map with a different offset value relative to the | |
3637 | * uncompressed data of our extent and different lengths. This | |
3638 | * is a corner case so we prioritize correctness over | |
3639 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3640 | */ | |
3641 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3642 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3643 | *prev_em_start != em->start) |
005efedf FM |
3644 | force_bio_submit = true; |
3645 | ||
3646 | if (prev_em_start) | |
8e928218 | 3647 | *prev_em_start = em->start; |
005efedf | 3648 | |
d1310b2e CM |
3649 | free_extent_map(em); |
3650 | em = NULL; | |
3651 | ||
3652 | /* we've found a hole, just zero and go on */ | |
3653 | if (block_start == EXTENT_MAP_HOLE) { | |
507903b8 AJ |
3654 | struct extent_state *cached = NULL; |
3655 | ||
d048b9c2 | 3656 | memzero_page(page, pg_offset, iosize); |
d1310b2e CM |
3657 | |
3658 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3659 | &cached, GFP_NOFS); |
7f042a83 | 3660 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3661 | cur + iosize - 1, &cached); |
92082d40 | 3662 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3663 | cur = cur + iosize; |
306e16ce | 3664 | pg_offset += iosize; |
d1310b2e CM |
3665 | continue; |
3666 | } | |
3667 | /* the get_extent function already copied into the page */ | |
70dec807 | 3668 | if (block_start == EXTENT_MAP_INLINE) { |
7f042a83 | 3669 | unlock_extent(tree, cur, cur + iosize - 1); |
52b029f4 | 3670 | end_page_read(page, true, cur, iosize); |
70dec807 | 3671 | cur = cur + iosize; |
306e16ce | 3672 | pg_offset += iosize; |
70dec807 CM |
3673 | continue; |
3674 | } | |
d1310b2e | 3675 | |
0ceb34bf | 3676 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
390ed29b | 3677 | bio_ctrl, page, disk_bytenr, iosize, |
722c82ac CH |
3678 | pg_offset, end_bio_extent_readpage, |
3679 | this_bio_flag, force_bio_submit); | |
ad3fc794 | 3680 | if (ret) { |
10f7f6f8 QW |
3681 | /* |
3682 | * We have to unlock the remaining range, or the page | |
3683 | * will never be unlocked. | |
3684 | */ | |
3685 | unlock_extent(tree, cur, end); | |
3686 | end_page_read(page, false, cur, end + 1 - cur); | |
baf863b9 | 3687 | goto out; |
edd33c99 | 3688 | } |
d1310b2e | 3689 | cur = cur + iosize; |
306e16ce | 3690 | pg_offset += iosize; |
d1310b2e | 3691 | } |
90a887c9 | 3692 | out: |
baf863b9 | 3693 | return ret; |
d1310b2e CM |
3694 | } |
3695 | ||
fdaf9a58 | 3696 | int btrfs_read_folio(struct file *file, struct folio *folio) |
7aab8b32 | 3697 | { |
fdaf9a58 | 3698 | struct page *page = &folio->page; |
7aab8b32 CH |
3699 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
3700 | u64 start = page_offset(page); | |
3701 | u64 end = start + PAGE_SIZE - 1; | |
3702 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; | |
3703 | int ret; | |
3704 | ||
3705 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); | |
3706 | ||
3707 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); | |
3708 | /* | |
3709 | * If btrfs_do_readpage() failed we will want to submit the assembled | |
3710 | * bio to do the cleanup. | |
3711 | */ | |
722c82ac | 3712 | submit_one_bio(&bio_ctrl); |
7aab8b32 CH |
3713 | return ret; |
3714 | } | |
3715 | ||
b6660e80 | 3716 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
3717 | u64 start, u64 end, |
3718 | struct extent_map **em_cached, | |
3719 | struct btrfs_bio_ctrl *bio_ctrl, | |
3720 | u64 *prev_em_start) | |
9974090b | 3721 | { |
23d31bd4 | 3722 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3723 | int index; |
3724 | ||
b272ae22 | 3725 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3726 | |
3727 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 3728 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
0f208812 | 3729 | REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3730 | put_page(pages[index]); |
9974090b MX |
3731 | } |
3732 | } | |
3733 | ||
d1310b2e | 3734 | /* |
40f76580 CM |
3735 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3736 | * | |
5eaad97a | 3737 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3738 | * to write the page (copy into inline extent). In this case the IO has |
3739 | * been started and the page is already unlocked. | |
3740 | * | |
3741 | * This returns 0 if all went well (page still locked) | |
3742 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3743 | */ |
cd4c0bf9 | 3744 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
83f1b680 | 3745 | struct page *page, struct writeback_control *wbc) |
40f76580 | 3746 | { |
2749f7ef | 3747 | const u64 page_end = page_offset(page) + PAGE_SIZE - 1; |
cf3075fb | 3748 | u64 delalloc_start = page_offset(page); |
40f76580 | 3749 | u64 delalloc_to_write = 0; |
83f1b680 QW |
3750 | /* How many pages are started by btrfs_run_delalloc_range() */ |
3751 | unsigned long nr_written = 0; | |
40f76580 CM |
3752 | int ret; |
3753 | int page_started = 0; | |
3754 | ||
2749f7ef QW |
3755 | while (delalloc_start < page_end) { |
3756 | u64 delalloc_end = page_end; | |
3757 | bool found; | |
40f76580 | 3758 | |
cd4c0bf9 | 3759 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3760 | &delalloc_start, |
917aacec | 3761 | &delalloc_end); |
3522e903 | 3762 | if (!found) { |
40f76580 CM |
3763 | delalloc_start = delalloc_end + 1; |
3764 | continue; | |
3765 | } | |
cd4c0bf9 | 3766 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
83f1b680 | 3767 | delalloc_end, &page_started, &nr_written, wbc); |
40f76580 | 3768 | if (ret) { |
963e4db8 QW |
3769 | btrfs_page_set_error(inode->root->fs_info, page, |
3770 | page_offset(page), PAGE_SIZE); | |
7361b4ae | 3771 | return ret; |
40f76580 CM |
3772 | } |
3773 | /* | |
ea1754a0 KS |
3774 | * delalloc_end is already one less than the total length, so |
3775 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3776 | */ |
3777 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3778 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3779 | delalloc_start = delalloc_end + 1; |
3780 | } | |
3781 | if (wbc->nr_to_write < delalloc_to_write) { | |
3782 | int thresh = 8192; | |
3783 | ||
3784 | if (delalloc_to_write < thresh * 2) | |
3785 | thresh = delalloc_to_write; | |
3786 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3787 | thresh); | |
3788 | } | |
3789 | ||
83f1b680 | 3790 | /* Did btrfs_run_dealloc_range() already unlock and start the IO? */ |
40f76580 CM |
3791 | if (page_started) { |
3792 | /* | |
83f1b680 QW |
3793 | * We've unlocked the page, so we can't update the mapping's |
3794 | * writeback index, just update nr_to_write. | |
40f76580 | 3795 | */ |
83f1b680 | 3796 | wbc->nr_to_write -= nr_written; |
40f76580 CM |
3797 | return 1; |
3798 | } | |
3799 | ||
b69d1ee9 | 3800 | return 0; |
40f76580 CM |
3801 | } |
3802 | ||
c5ef5c6c QW |
3803 | /* |
3804 | * Find the first byte we need to write. | |
3805 | * | |
3806 | * For subpage, one page can contain several sectors, and | |
3807 | * __extent_writepage_io() will just grab all extent maps in the page | |
3808 | * range and try to submit all non-inline/non-compressed extents. | |
3809 | * | |
3810 | * This is a big problem for subpage, we shouldn't re-submit already written | |
3811 | * data at all. | |
3812 | * This function will lookup subpage dirty bit to find which range we really | |
3813 | * need to submit. | |
3814 | * | |
3815 | * Return the next dirty range in [@start, @end). | |
3816 | * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE. | |
3817 | */ | |
3818 | static void find_next_dirty_byte(struct btrfs_fs_info *fs_info, | |
3819 | struct page *page, u64 *start, u64 *end) | |
3820 | { | |
3821 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
72a69cd0 | 3822 | struct btrfs_subpage_info *spi = fs_info->subpage_info; |
c5ef5c6c QW |
3823 | u64 orig_start = *start; |
3824 | /* Declare as unsigned long so we can use bitmap ops */ | |
c5ef5c6c | 3825 | unsigned long flags; |
72a69cd0 | 3826 | int range_start_bit; |
c5ef5c6c QW |
3827 | int range_end_bit; |
3828 | ||
3829 | /* | |
3830 | * For regular sector size == page size case, since one page only | |
3831 | * contains one sector, we return the page offset directly. | |
3832 | */ | |
fbca46eb | 3833 | if (!btrfs_is_subpage(fs_info, page)) { |
c5ef5c6c QW |
3834 | *start = page_offset(page); |
3835 | *end = page_offset(page) + PAGE_SIZE; | |
3836 | return; | |
3837 | } | |
3838 | ||
72a69cd0 QW |
3839 | range_start_bit = spi->dirty_offset + |
3840 | (offset_in_page(orig_start) >> fs_info->sectorsize_bits); | |
3841 | ||
c5ef5c6c QW |
3842 | /* We should have the page locked, but just in case */ |
3843 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
3844 | bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit, |
3845 | spi->dirty_offset + spi->bitmap_nr_bits); | |
c5ef5c6c QW |
3846 | spin_unlock_irqrestore(&subpage->lock, flags); |
3847 | ||
72a69cd0 QW |
3848 | range_start_bit -= spi->dirty_offset; |
3849 | range_end_bit -= spi->dirty_offset; | |
3850 | ||
c5ef5c6c QW |
3851 | *start = page_offset(page) + range_start_bit * fs_info->sectorsize; |
3852 | *end = page_offset(page) + range_end_bit * fs_info->sectorsize; | |
3853 | } | |
3854 | ||
40f76580 CM |
3855 | /* |
3856 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3857 | * and does the loop to map the page into extents and bios. | |
3858 | * | |
3859 | * We return 1 if the IO is started and the page is unlocked, | |
3860 | * 0 if all went well (page still locked) | |
3861 | * < 0 if there were errors (page still locked) | |
3862 | */ | |
d4580fe2 | 3863 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3864 | struct page *page, |
3865 | struct writeback_control *wbc, | |
3866 | struct extent_page_data *epd, | |
3867 | loff_t i_size, | |
57e5ffeb | 3868 | int *nr_ret) |
d1310b2e | 3869 | { |
6bc5636a | 3870 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a129ffb8 QW |
3871 | u64 cur = page_offset(page); |
3872 | u64 end = cur + PAGE_SIZE - 1; | |
d1310b2e | 3873 | u64 extent_offset; |
d1310b2e | 3874 | u64 block_start; |
d1310b2e | 3875 | struct extent_map *em; |
44e5801f | 3876 | int saved_ret = 0; |
40f76580 CM |
3877 | int ret = 0; |
3878 | int nr = 0; | |
bf9486d6 BVA |
3879 | enum req_op op = REQ_OP_WRITE; |
3880 | const blk_opf_t write_flags = wbc_to_write_flags(wbc); | |
44e5801f | 3881 | bool has_error = false; |
40f76580 | 3882 | bool compressed; |
c8b97818 | 3883 | |
a129ffb8 | 3884 | ret = btrfs_writepage_cow_fixup(page); |
d75855b4 NB |
3885 | if (ret) { |
3886 | /* Fixup worker will requeue */ | |
5ab58055 | 3887 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3888 | unlock_page(page); |
3889 | return 1; | |
247e743c CM |
3890 | } |
3891 | ||
11c8349b CM |
3892 | /* |
3893 | * we don't want to touch the inode after unlocking the page, | |
3894 | * so we update the mapping writeback index now | |
3895 | */ | |
572f3dad | 3896 | wbc->nr_to_write--; |
771ed689 | 3897 | |
d1310b2e | 3898 | while (cur <= end) { |
0c64c33c | 3899 | u64 disk_bytenr; |
40f76580 | 3900 | u64 em_end; |
c5ef5c6c QW |
3901 | u64 dirty_range_start = cur; |
3902 | u64 dirty_range_end; | |
6bc5636a | 3903 | u32 iosize; |
58409edd | 3904 | |
40f76580 | 3905 | if (cur >= i_size) { |
38a39ac7 | 3906 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
25c1252a | 3907 | end, true); |
cc1d0d93 QW |
3908 | /* |
3909 | * This range is beyond i_size, thus we don't need to | |
3910 | * bother writing back. | |
3911 | * But we still need to clear the dirty subpage bit, or | |
3912 | * the next time the page gets dirtied, we will try to | |
3913 | * writeback the sectors with subpage dirty bits, | |
3914 | * causing writeback without ordered extent. | |
3915 | */ | |
3916 | btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur); | |
d1310b2e CM |
3917 | break; |
3918 | } | |
c5ef5c6c QW |
3919 | |
3920 | find_next_dirty_byte(fs_info, page, &dirty_range_start, | |
3921 | &dirty_range_end); | |
3922 | if (cur < dirty_range_start) { | |
3923 | cur = dirty_range_start; | |
3924 | continue; | |
3925 | } | |
3926 | ||
d4580fe2 | 3927 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c0347550 | 3928 | if (IS_ERR(em)) { |
c5ef5c6c | 3929 | btrfs_page_set_error(fs_info, page, cur, end - cur + 1); |
61391d56 | 3930 | ret = PTR_ERR_OR_ZERO(em); |
44e5801f QW |
3931 | has_error = true; |
3932 | if (!saved_ret) | |
3933 | saved_ret = ret; | |
d1310b2e CM |
3934 | break; |
3935 | } | |
3936 | ||
3937 | extent_offset = cur - em->start; | |
40f76580 | 3938 | em_end = extent_map_end(em); |
6bc5636a QW |
3939 | ASSERT(cur <= em_end); |
3940 | ASSERT(cur < end); | |
3941 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
3942 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 3943 | block_start = em->block_start; |
c8b97818 | 3944 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
3945 | disk_bytenr = em->block_start + extent_offset; |
3946 | ||
c5ef5c6c QW |
3947 | /* |
3948 | * Note that em_end from extent_map_end() and dirty_range_end from | |
3949 | * find_next_dirty_byte() are all exclusive | |
3950 | */ | |
3951 | iosize = min(min(em_end, end + 1), dirty_range_end) - cur; | |
d8e3fb10 | 3952 | |
e380adfc | 3953 | if (btrfs_use_zone_append(inode, em->block_start)) |
bf9486d6 | 3954 | op = REQ_OP_ZONE_APPEND; |
d8e3fb10 | 3955 | |
d1310b2e CM |
3956 | free_extent_map(em); |
3957 | em = NULL; | |
3958 | ||
c8b97818 CM |
3959 | /* |
3960 | * compressed and inline extents are written through other | |
3961 | * paths in the FS | |
3962 | */ | |
3963 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3964 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 3965 | if (compressed) |
c8b97818 | 3966 | nr++; |
c8b04030 | 3967 | else |
38a39ac7 | 3968 | btrfs_writepage_endio_finish_ordered(inode, |
25c1252a | 3969 | page, cur, cur + iosize - 1, true); |
cc1d0d93 | 3970 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); |
c8b97818 | 3971 | cur += iosize; |
d1310b2e CM |
3972 | continue; |
3973 | } | |
c8b97818 | 3974 | |
d2a91064 | 3975 | btrfs_set_range_writeback(inode, cur, cur + iosize - 1); |
58409edd | 3976 | if (!PageWriteback(page)) { |
d4580fe2 | 3977 | btrfs_err(inode->root->fs_info, |
58409edd DS |
3978 | "page %lu not writeback, cur %llu end %llu", |
3979 | page->index, cur, end); | |
d1310b2e | 3980 | } |
7f3c74fb | 3981 | |
c5ef5c6c QW |
3982 | /* |
3983 | * Although the PageDirty bit is cleared before entering this | |
3984 | * function, subpage dirty bit is not cleared. | |
3985 | * So clear subpage dirty bit here so next time we won't submit | |
3986 | * page for range already written to disk. | |
3987 | */ | |
3988 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); | |
3989 | ||
bf9486d6 | 3990 | ret = submit_extent_page(op | write_flags, wbc, |
390ed29b | 3991 | &epd->bio_ctrl, page, |
d8e3fb10 | 3992 | disk_bytenr, iosize, |
390ed29b | 3993 | cur - page_offset(page), |
58409edd | 3994 | end_bio_extent_writepage, |
722c82ac | 3995 | 0, false); |
fe01aa65 | 3996 | if (ret) { |
44e5801f QW |
3997 | has_error = true; |
3998 | if (!saved_ret) | |
3999 | saved_ret = ret; | |
4000 | ||
c5ef5c6c | 4001 | btrfs_page_set_error(fs_info, page, cur, iosize); |
fe01aa65 | 4002 | if (PageWriteback(page)) |
c5ef5c6c QW |
4003 | btrfs_page_clear_writeback(fs_info, page, cur, |
4004 | iosize); | |
fe01aa65 | 4005 | } |
d1310b2e | 4006 | |
6bc5636a | 4007 | cur += iosize; |
d1310b2e CM |
4008 | nr++; |
4009 | } | |
cc1d0d93 QW |
4010 | /* |
4011 | * If we finish without problem, we should not only clear page dirty, | |
4012 | * but also empty subpage dirty bits | |
4013 | */ | |
44e5801f | 4014 | if (!has_error) |
cc1d0d93 | 4015 | btrfs_page_assert_not_dirty(fs_info, page); |
44e5801f QW |
4016 | else |
4017 | ret = saved_ret; | |
40f76580 | 4018 | *nr_ret = nr; |
40f76580 CM |
4019 | return ret; |
4020 | } | |
4021 | ||
4022 | /* | |
4023 | * the writepage semantics are similar to regular writepage. extent | |
4024 | * records are inserted to lock ranges in the tree, and as dirty areas | |
4025 | * are found, they are marked writeback. Then the lock bits are removed | |
4026 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
4027 | * |
4028 | * Return 0 if everything goes well. | |
4029 | * Return <0 for error. | |
40f76580 CM |
4030 | */ |
4031 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 4032 | struct extent_page_data *epd) |
40f76580 | 4033 | { |
8e1dec8e | 4034 | struct folio *folio = page_folio(page); |
40f76580 | 4035 | struct inode *inode = page->mapping->host; |
e55a0de1 | 4036 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
cf3075fb QW |
4037 | const u64 page_start = page_offset(page); |
4038 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
40f76580 CM |
4039 | int ret; |
4040 | int nr = 0; | |
eb70d222 | 4041 | size_t pg_offset; |
40f76580 | 4042 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 4043 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 | 4044 | |
40f76580 CM |
4045 | trace___extent_writepage(page, inode, wbc); |
4046 | ||
4047 | WARN_ON(!PageLocked(page)); | |
4048 | ||
963e4db8 QW |
4049 | btrfs_page_clear_error(btrfs_sb(inode->i_sb), page, |
4050 | page_offset(page), PAGE_SIZE); | |
40f76580 | 4051 | |
7073017a | 4052 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
4053 | if (page->index > end_index || |
4054 | (page->index == end_index && !pg_offset)) { | |
8e1dec8e MWO |
4055 | folio_invalidate(folio, 0, folio_size(folio)); |
4056 | folio_unlock(folio); | |
40f76580 CM |
4057 | return 0; |
4058 | } | |
4059 | ||
21a8935e | 4060 | if (page->index == end_index) |
d048b9c2 | 4061 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 | 4062 | |
32443de3 QW |
4063 | ret = set_page_extent_mapped(page); |
4064 | if (ret < 0) { | |
4065 | SetPageError(page); | |
4066 | goto done; | |
4067 | } | |
40f76580 | 4068 | |
7789a55a | 4069 | if (!epd->extent_locked) { |
83f1b680 | 4070 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc); |
7789a55a | 4071 | if (ret == 1) |
169d2c87 | 4072 | return 0; |
7789a55a NB |
4073 | if (ret) |
4074 | goto done; | |
4075 | } | |
40f76580 | 4076 | |
d4580fe2 | 4077 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
83f1b680 | 4078 | &nr); |
40f76580 | 4079 | if (ret == 1) |
169d2c87 | 4080 | return 0; |
40f76580 | 4081 | |
d1310b2e CM |
4082 | done: |
4083 | if (nr == 0) { | |
4084 | /* make sure the mapping tag for page dirty gets cleared */ | |
4085 | set_page_writeback(page); | |
4086 | end_page_writeback(page); | |
4087 | } | |
963e4db8 QW |
4088 | /* |
4089 | * Here we used to have a check for PageError() and then set @ret and | |
4090 | * call end_extent_writepage(). | |
4091 | * | |
4092 | * But in fact setting @ret here will cause different error paths | |
4093 | * between subpage and regular sectorsize. | |
4094 | * | |
4095 | * For regular page size, we never submit current page, but only add | |
4096 | * current page to current bio. | |
4097 | * The bio submission can only happen in next page. | |
4098 | * Thus if we hit the PageError() branch, @ret is already set to | |
4099 | * non-zero value and will not get updated for regular sectorsize. | |
4100 | * | |
4101 | * But for subpage case, it's possible we submit part of current page, | |
4102 | * thus can get PageError() set by submitted bio of the same page, | |
4103 | * while our @ret is still 0. | |
4104 | * | |
4105 | * So here we unify the behavior and don't set @ret. | |
4106 | * Error can still be properly passed to higher layer as page will | |
4107 | * be set error, here we just don't handle the IO failure. | |
4108 | * | |
4109 | * NOTE: This is just a hotfix for subpage. | |
4110 | * The root fix will be properly ending ordered extent when we hit | |
4111 | * an error during writeback. | |
4112 | * | |
4113 | * But that needs a bigger refactoring, as we not only need to grab the | |
4114 | * submitted OE, but also need to know exactly at which bytenr we hit | |
4115 | * the error. | |
4116 | * Currently the full page based __extent_writepage_io() is not | |
4117 | * capable of that. | |
4118 | */ | |
4119 | if (PageError(page)) | |
cf3075fb | 4120 | end_extent_writepage(page, ret, page_start, page_end); |
e55a0de1 QW |
4121 | if (epd->extent_locked) { |
4122 | /* | |
4123 | * If epd->extent_locked, it's from extent_write_locked_range(), | |
4124 | * the page can either be locked by lock_page() or | |
4125 | * process_one_page(). | |
4126 | * Let btrfs_page_unlock_writer() handle both cases. | |
4127 | */ | |
4128 | ASSERT(wbc); | |
4129 | btrfs_page_unlock_writer(fs_info, page, wbc->range_start, | |
4130 | wbc->range_end + 1 - wbc->range_start); | |
4131 | } else { | |
4132 | unlock_page(page); | |
4133 | } | |
3065976b | 4134 | ASSERT(ret <= 0); |
40f76580 | 4135 | return ret; |
d1310b2e CM |
4136 | } |
4137 | ||
fd8b2b61 | 4138 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 4139 | { |
74316201 N |
4140 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
4141 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
4142 | } |
4143 | ||
18dfa711 FM |
4144 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
4145 | { | |
4146 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4147 | smp_mb__after_atomic(); | |
4148 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
4149 | } | |
4150 | ||
2e3c2513 | 4151 | /* |
a3efb2f0 | 4152 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 4153 | * |
a3efb2f0 QW |
4154 | * May try to flush write bio if we can't get the lock. |
4155 | * | |
4156 | * Return 0 if the extent buffer doesn't need to be submitted. | |
4157 | * (E.g. the extent buffer is not dirty) | |
4158 | * Return >0 is the extent buffer is submitted to bio. | |
4159 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 4160 | */ |
9df76fb5 | 4161 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 4162 | struct extent_page_data *epd) |
0b32f4bb | 4163 | { |
9df76fb5 | 4164 | struct btrfs_fs_info *fs_info = eb->fs_info; |
c9583ada | 4165 | int i, num_pages; |
0b32f4bb JB |
4166 | int flush = 0; |
4167 | int ret = 0; | |
4168 | ||
4169 | if (!btrfs_try_tree_write_lock(eb)) { | |
9845e5dd | 4170 | submit_write_bio(epd, 0); |
2e3c2513 | 4171 | flush = 1; |
0b32f4bb JB |
4172 | btrfs_tree_lock(eb); |
4173 | } | |
4174 | ||
4175 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
4176 | btrfs_tree_unlock(eb); | |
4177 | if (!epd->sync_io) | |
4178 | return 0; | |
4179 | if (!flush) { | |
9845e5dd | 4180 | submit_write_bio(epd, 0); |
0b32f4bb JB |
4181 | flush = 1; |
4182 | } | |
a098d8e8 CM |
4183 | while (1) { |
4184 | wait_on_extent_buffer_writeback(eb); | |
4185 | btrfs_tree_lock(eb); | |
4186 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
4187 | break; | |
0b32f4bb | 4188 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
4189 | } |
4190 | } | |
4191 | ||
51561ffe JB |
4192 | /* |
4193 | * We need to do this to prevent races in people who check if the eb is | |
4194 | * under IO since we can end up having no IO bits set for a short period | |
4195 | * of time. | |
4196 | */ | |
4197 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
4198 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
4199 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 4200 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 4201 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
4202 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4203 | -eb->len, | |
4204 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 4205 | ret = 1; |
51561ffe JB |
4206 | } else { |
4207 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
4208 | } |
4209 | ||
4210 | btrfs_tree_unlock(eb); | |
4211 | ||
f3156df9 QW |
4212 | /* |
4213 | * Either we don't need to submit any tree block, or we're submitting | |
4214 | * subpage eb. | |
4215 | * Subpage metadata doesn't use page locking at all, so we can skip | |
4216 | * the page locking. | |
4217 | */ | |
fbca46eb | 4218 | if (!ret || fs_info->nodesize < PAGE_SIZE) |
0b32f4bb JB |
4219 | return ret; |
4220 | ||
65ad0104 | 4221 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4222 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4223 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4224 | |
4225 | if (!trylock_page(p)) { | |
4226 | if (!flush) { | |
9845e5dd | 4227 | submit_write_bio(epd, 0); |
0b32f4bb JB |
4228 | flush = 1; |
4229 | } | |
4230 | lock_page(p); | |
4231 | } | |
4232 | } | |
4233 | ||
2e3c2513 | 4234 | return ret; |
0b32f4bb JB |
4235 | } |
4236 | ||
5a2c6075 | 4237 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 4238 | { |
5a2c6075 | 4239 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 4240 | |
5a2c6075 | 4241 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
4242 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
4243 | return; | |
4244 | ||
c2e39305 JB |
4245 | /* |
4246 | * A read may stumble upon this buffer later, make sure that it gets an | |
4247 | * error and knows there was an error. | |
4248 | */ | |
4249 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4250 | ||
68b85589 JB |
4251 | /* |
4252 | * We need to set the mapping with the io error as well because a write | |
4253 | * error will flip the file system readonly, and then syncfs() will | |
4254 | * return a 0 because we are readonly if we don't modify the err seq for | |
4255 | * the superblock. | |
4256 | */ | |
4257 | mapping_set_error(page->mapping, -EIO); | |
4258 | ||
eb5b64f1 DZ |
4259 | /* |
4260 | * If we error out, we should add back the dirty_metadata_bytes | |
4261 | * to make it consistent. | |
4262 | */ | |
eb5b64f1 DZ |
4263 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4264 | eb->len, fs_info->dirty_metadata_batch); | |
4265 | ||
656f30db FM |
4266 | /* |
4267 | * If writeback for a btree extent that doesn't belong to a log tree | |
4268 | * failed, increment the counter transaction->eb_write_errors. | |
4269 | * We do this because while the transaction is running and before it's | |
4270 | * committing (when we call filemap_fdata[write|wait]_range against | |
4271 | * the btree inode), we might have | |
4272 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
4273 | * returns an error or an error happens during writeback, when we're | |
4274 | * committing the transaction we wouldn't know about it, since the pages | |
4275 | * can be no longer dirty nor marked anymore for writeback (if a | |
4276 | * subsequent modification to the extent buffer didn't happen before the | |
4277 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
4278 | * able to find the pages tagged with SetPageError at transaction | |
4279 | * commit time. So if this happens we must abort the transaction, | |
4280 | * otherwise we commit a super block with btree roots that point to | |
4281 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
4282 | * or the content of some node/leaf from a past generation that got | |
4283 | * cowed or deleted and is no longer valid. | |
4284 | * | |
4285 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
4286 | * not be enough - we need to distinguish between log tree extents vs | |
4287 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
4288 | * will catch and clear such errors in the mapping - and that call might | |
4289 | * be from a log sync and not from a transaction commit. Also, checking | |
4290 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
4291 | * not done and would not be reliable - the eb might have been released | |
4292 | * from memory and reading it back again means that flag would not be | |
4293 | * set (since it's a runtime flag, not persisted on disk). | |
4294 | * | |
4295 | * Using the flags below in the btree inode also makes us achieve the | |
4296 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
4297 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
4298 | * is called, the writeback for all dirty pages had already finished | |
4299 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
4300 | * filemap_fdatawait_range() would return success, as it could not know | |
4301 | * that writeback errors happened (the pages were no longer tagged for | |
4302 | * writeback). | |
4303 | */ | |
4304 | switch (eb->log_index) { | |
4305 | case -1: | |
5a2c6075 | 4306 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
4307 | break; |
4308 | case 0: | |
5a2c6075 | 4309 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
4310 | break; |
4311 | case 1: | |
5a2c6075 | 4312 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
4313 | break; |
4314 | default: | |
4315 | BUG(); /* unexpected, logic error */ | |
4316 | } | |
4317 | } | |
4318 | ||
2f3186d8 QW |
4319 | /* |
4320 | * The endio specific version which won't touch any unsafe spinlock in endio | |
4321 | * context. | |
4322 | */ | |
4323 | static struct extent_buffer *find_extent_buffer_nolock( | |
4324 | struct btrfs_fs_info *fs_info, u64 start) | |
4325 | { | |
4326 | struct extent_buffer *eb; | |
4327 | ||
4328 | rcu_read_lock(); | |
01cd3909 DS |
4329 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4330 | start >> fs_info->sectorsize_bits); | |
2f3186d8 QW |
4331 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4332 | rcu_read_unlock(); | |
4333 | return eb; | |
4334 | } | |
4335 | rcu_read_unlock(); | |
4336 | return NULL; | |
4337 | } | |
4338 | ||
4339 | /* | |
4340 | * The endio function for subpage extent buffer write. | |
4341 | * | |
4342 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
4343 | * after all extent buffers in the page has finished their writeback. | |
4344 | */ | |
917f32a2 | 4345 | static void end_bio_subpage_eb_writepage(struct btrfs_bio *bbio) |
2f3186d8 | 4346 | { |
917f32a2 | 4347 | struct bio *bio = &bbio->bio; |
fa04c165 | 4348 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
4349 | struct bio_vec *bvec; |
4350 | struct bvec_iter_all iter_all; | |
4351 | ||
fa04c165 | 4352 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
fbca46eb | 4353 | ASSERT(fs_info->nodesize < PAGE_SIZE); |
fa04c165 | 4354 | |
2f3186d8 QW |
4355 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
4356 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
4357 | struct page *page = bvec->bv_page; | |
4358 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
4359 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
4360 | u64 cur_bytenr = bvec_start; | |
4361 | ||
4362 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
4363 | ||
4364 | /* Iterate through all extent buffers in the range */ | |
4365 | while (cur_bytenr <= bvec_end) { | |
4366 | struct extent_buffer *eb; | |
4367 | int done; | |
4368 | ||
4369 | /* | |
4370 | * Here we can't use find_extent_buffer(), as it may | |
4371 | * try to lock eb->refs_lock, which is not safe in endio | |
4372 | * context. | |
4373 | */ | |
4374 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
4375 | ASSERT(eb); | |
4376 | ||
4377 | cur_bytenr = eb->start + eb->len; | |
4378 | ||
4379 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
4380 | done = atomic_dec_and_test(&eb->io_pages); | |
4381 | ASSERT(done); | |
4382 | ||
4383 | if (bio->bi_status || | |
4384 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
4385 | ClearPageUptodate(page); | |
4386 | set_btree_ioerr(page, eb); | |
4387 | } | |
4388 | ||
4389 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
4390 | eb->len); | |
4391 | end_extent_buffer_writeback(eb); | |
4392 | /* | |
4393 | * free_extent_buffer() will grab spinlock which is not | |
4394 | * safe in endio context. Thus here we manually dec | |
4395 | * the ref. | |
4396 | */ | |
4397 | atomic_dec(&eb->refs); | |
4398 | } | |
4399 | } | |
4400 | bio_put(bio); | |
4401 | } | |
4402 | ||
917f32a2 | 4403 | static void end_bio_extent_buffer_writepage(struct btrfs_bio *bbio) |
0b32f4bb | 4404 | { |
917f32a2 | 4405 | struct bio *bio = &bbio->bio; |
2c30c71b | 4406 | struct bio_vec *bvec; |
0b32f4bb | 4407 | struct extent_buffer *eb; |
2b070cfe | 4408 | int done; |
6dc4f100 | 4409 | struct bvec_iter_all iter_all; |
0b32f4bb | 4410 | |
c09abff8 | 4411 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 4412 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
4413 | struct page *page = bvec->bv_page; |
4414 | ||
0b32f4bb JB |
4415 | eb = (struct extent_buffer *)page->private; |
4416 | BUG_ON(!eb); | |
4417 | done = atomic_dec_and_test(&eb->io_pages); | |
4418 | ||
4e4cbee9 | 4419 | if (bio->bi_status || |
4246a0b6 | 4420 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 4421 | ClearPageUptodate(page); |
5a2c6075 | 4422 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
4423 | } |
4424 | ||
4425 | end_page_writeback(page); | |
4426 | ||
4427 | if (!done) | |
4428 | continue; | |
4429 | ||
4430 | end_extent_buffer_writeback(eb); | |
2c30c71b | 4431 | } |
0b32f4bb JB |
4432 | |
4433 | bio_put(bio); | |
0b32f4bb JB |
4434 | } |
4435 | ||
fa04c165 QW |
4436 | static void prepare_eb_write(struct extent_buffer *eb) |
4437 | { | |
4438 | u32 nritems; | |
4439 | unsigned long start; | |
4440 | unsigned long end; | |
4441 | ||
4442 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
4443 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
4444 | ||
4445 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
4446 | nritems = btrfs_header_nritems(eb); | |
4447 | if (btrfs_header_level(eb) > 0) { | |
4448 | end = btrfs_node_key_ptr_offset(nritems); | |
4449 | memzero_extent_buffer(eb, end, eb->len - end); | |
4450 | } else { | |
4451 | /* | |
4452 | * Leaf: | |
4453 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
4454 | */ | |
4455 | start = btrfs_item_nr_offset(nritems); | |
4456 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); | |
4457 | memzero_extent_buffer(eb, start, end - start); | |
4458 | } | |
4459 | } | |
4460 | ||
35b6ddfa QW |
4461 | /* |
4462 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
4463 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
4464 | */ |
4465 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
4466 | struct writeback_control *wbc, | |
4467 | struct extent_page_data *epd) | |
4468 | { | |
4469 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4470 | struct page *page = eb->pages[0]; | |
353767e4 | 4471 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
35b6ddfa QW |
4472 | bool no_dirty_ebs = false; |
4473 | int ret; | |
4474 | ||
fa04c165 QW |
4475 | prepare_eb_write(eb); |
4476 | ||
35b6ddfa QW |
4477 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
4478 | lock_page(page); | |
4479 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
4480 | ||
4481 | /* Check if this is the last dirty bit to update nr_written */ | |
4482 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
4483 | eb->start, eb->len); | |
4484 | if (no_dirty_ebs) | |
4485 | clear_page_dirty_for_io(page); | |
4486 | ||
390ed29b QW |
4487 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
4488 | &epd->bio_ctrl, page, eb->start, eb->len, | |
4489 | eb->start - page_offset(page), | |
722c82ac | 4490 | end_bio_subpage_eb_writepage, 0, false); |
35b6ddfa QW |
4491 | if (ret) { |
4492 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
4493 | set_btree_ioerr(page, eb); | |
4494 | unlock_page(page); | |
4495 | ||
4496 | if (atomic_dec_and_test(&eb->io_pages)) | |
4497 | end_extent_buffer_writeback(eb); | |
4498 | return -EIO; | |
4499 | } | |
4500 | unlock_page(page); | |
4501 | /* | |
4502 | * Submission finished without problem, if no range of the page is | |
4503 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
4504 | */ | |
4505 | if (no_dirty_ebs) | |
572f3dad | 4506 | wbc->nr_to_write--; |
35b6ddfa QW |
4507 | return ret; |
4508 | } | |
4509 | ||
0e378df1 | 4510 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
4511 | struct writeback_control *wbc, |
4512 | struct extent_page_data *epd) | |
4513 | { | |
0c64c33c | 4514 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 4515 | int i, num_pages; |
353767e4 | 4516 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
d7dbe9e7 | 4517 | int ret = 0; |
0b32f4bb | 4518 | |
fa04c165 | 4519 | prepare_eb_write(eb); |
35b6ddfa | 4520 | |
fa04c165 | 4521 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4522 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4523 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4524 | |
4525 | clear_page_dirty_for_io(p); | |
4526 | set_page_writeback(p); | |
0ceb34bf | 4527 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
390ed29b QW |
4528 | &epd->bio_ctrl, p, disk_bytenr, |
4529 | PAGE_SIZE, 0, | |
1f7ad75b | 4530 | end_bio_extent_buffer_writepage, |
722c82ac | 4531 | 0, false); |
0b32f4bb | 4532 | if (ret) { |
5a2c6075 | 4533 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
4534 | if (PageWriteback(p)) |
4535 | end_page_writeback(p); | |
0b32f4bb JB |
4536 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
4537 | end_extent_buffer_writeback(eb); | |
4538 | ret = -EIO; | |
4539 | break; | |
4540 | } | |
0c64c33c | 4541 | disk_bytenr += PAGE_SIZE; |
572f3dad | 4542 | wbc->nr_to_write--; |
0b32f4bb JB |
4543 | unlock_page(p); |
4544 | } | |
4545 | ||
4546 | if (unlikely(ret)) { | |
4547 | for (; i < num_pages; i++) { | |
bbf65cf0 | 4548 | struct page *p = eb->pages[i]; |
81465028 | 4549 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
4550 | unlock_page(p); |
4551 | } | |
4552 | } | |
4553 | ||
4554 | return ret; | |
4555 | } | |
4556 | ||
c4aec299 QW |
4557 | /* |
4558 | * Submit one subpage btree page. | |
4559 | * | |
4560 | * The main difference to submit_eb_page() is: | |
4561 | * - Page locking | |
4562 | * For subpage, we don't rely on page locking at all. | |
4563 | * | |
4564 | * - Flush write bio | |
4565 | * We only flush bio if we may be unable to fit current extent buffers into | |
4566 | * current bio. | |
4567 | * | |
4568 | * Return >=0 for the number of submitted extent buffers. | |
4569 | * Return <0 for fatal error. | |
4570 | */ | |
4571 | static int submit_eb_subpage(struct page *page, | |
4572 | struct writeback_control *wbc, | |
4573 | struct extent_page_data *epd) | |
4574 | { | |
4575 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
4576 | int submitted = 0; | |
4577 | u64 page_start = page_offset(page); | |
4578 | int bit_start = 0; | |
c4aec299 QW |
4579 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; |
4580 | int ret; | |
4581 | ||
4582 | /* Lock and write each dirty extent buffers in the range */ | |
72a69cd0 | 4583 | while (bit_start < fs_info->subpage_info->bitmap_nr_bits) { |
c4aec299 QW |
4584 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; |
4585 | struct extent_buffer *eb; | |
4586 | unsigned long flags; | |
4587 | u64 start; | |
4588 | ||
4589 | /* | |
4590 | * Take private lock to ensure the subpage won't be detached | |
4591 | * in the meantime. | |
4592 | */ | |
4593 | spin_lock(&page->mapping->private_lock); | |
4594 | if (!PagePrivate(page)) { | |
4595 | spin_unlock(&page->mapping->private_lock); | |
4596 | break; | |
4597 | } | |
4598 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
4599 | if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset, |
4600 | subpage->bitmaps)) { | |
c4aec299 QW |
4601 | spin_unlock_irqrestore(&subpage->lock, flags); |
4602 | spin_unlock(&page->mapping->private_lock); | |
4603 | bit_start++; | |
4604 | continue; | |
4605 | } | |
4606 | ||
4607 | start = page_start + bit_start * fs_info->sectorsize; | |
4608 | bit_start += sectors_per_node; | |
4609 | ||
4610 | /* | |
4611 | * Here we just want to grab the eb without touching extra | |
4612 | * spin locks, so call find_extent_buffer_nolock(). | |
4613 | */ | |
4614 | eb = find_extent_buffer_nolock(fs_info, start); | |
4615 | spin_unlock_irqrestore(&subpage->lock, flags); | |
4616 | spin_unlock(&page->mapping->private_lock); | |
4617 | ||
4618 | /* | |
4619 | * The eb has already reached 0 refs thus find_extent_buffer() | |
4620 | * doesn't return it. We don't need to write back such eb | |
4621 | * anyway. | |
4622 | */ | |
4623 | if (!eb) | |
4624 | continue; | |
4625 | ||
4626 | ret = lock_extent_buffer_for_io(eb, epd); | |
4627 | if (ret == 0) { | |
4628 | free_extent_buffer(eb); | |
4629 | continue; | |
4630 | } | |
4631 | if (ret < 0) { | |
4632 | free_extent_buffer(eb); | |
4633 | goto cleanup; | |
4634 | } | |
fa04c165 | 4635 | ret = write_one_subpage_eb(eb, wbc, epd); |
c4aec299 QW |
4636 | free_extent_buffer(eb); |
4637 | if (ret < 0) | |
4638 | goto cleanup; | |
4639 | submitted++; | |
4640 | } | |
4641 | return submitted; | |
4642 | ||
4643 | cleanup: | |
4644 | /* We hit error, end bio for the submitted extent buffers */ | |
9845e5dd | 4645 | submit_write_bio(epd, ret); |
c4aec299 QW |
4646 | return ret; |
4647 | } | |
4648 | ||
f91e0d0c QW |
4649 | /* |
4650 | * Submit all page(s) of one extent buffer. | |
4651 | * | |
4652 | * @page: the page of one extent buffer | |
4653 | * @eb_context: to determine if we need to submit this page, if current page | |
4654 | * belongs to this eb, we don't need to submit | |
4655 | * | |
4656 | * The caller should pass each page in their bytenr order, and here we use | |
4657 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
4658 | * | |
4659 | * If we have, we just skip until we hit a new page that doesn't belong to | |
4660 | * current @eb_context. | |
4661 | * | |
4662 | * If not, we submit all the page(s) of the extent buffer. | |
4663 | * | |
4664 | * Return >0 if we have submitted the extent buffer successfully. | |
4665 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
4666 | * previous call. | |
4667 | * Return <0 for fatal error. | |
4668 | */ | |
4669 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
4670 | struct extent_page_data *epd, | |
4671 | struct extent_buffer **eb_context) | |
4672 | { | |
4673 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 4674 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
4675 | struct extent_buffer *eb; |
4676 | int ret; | |
4677 | ||
4678 | if (!PagePrivate(page)) | |
4679 | return 0; | |
4680 | ||
fbca46eb | 4681 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
c4aec299 QW |
4682 | return submit_eb_subpage(page, wbc, epd); |
4683 | ||
f91e0d0c QW |
4684 | spin_lock(&mapping->private_lock); |
4685 | if (!PagePrivate(page)) { | |
4686 | spin_unlock(&mapping->private_lock); | |
4687 | return 0; | |
4688 | } | |
4689 | ||
4690 | eb = (struct extent_buffer *)page->private; | |
4691 | ||
4692 | /* | |
4693 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
4694 | * crashing the machine for something we can survive anyway. | |
4695 | */ | |
4696 | if (WARN_ON(!eb)) { | |
4697 | spin_unlock(&mapping->private_lock); | |
4698 | return 0; | |
4699 | } | |
4700 | ||
4701 | if (eb == *eb_context) { | |
4702 | spin_unlock(&mapping->private_lock); | |
4703 | return 0; | |
4704 | } | |
4705 | ret = atomic_inc_not_zero(&eb->refs); | |
4706 | spin_unlock(&mapping->private_lock); | |
4707 | if (!ret) | |
4708 | return 0; | |
4709 | ||
0bc09ca1 NA |
4710 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
4711 | /* | |
4712 | * If for_sync, this hole will be filled with | |
4713 | * trasnsaction commit. | |
4714 | */ | |
4715 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
4716 | ret = -EAGAIN; | |
4717 | else | |
4718 | ret = 0; | |
4719 | free_extent_buffer(eb); | |
4720 | return ret; | |
4721 | } | |
4722 | ||
f91e0d0c QW |
4723 | *eb_context = eb; |
4724 | ||
4725 | ret = lock_extent_buffer_for_io(eb, epd); | |
4726 | if (ret <= 0) { | |
0bc09ca1 NA |
4727 | btrfs_revert_meta_write_pointer(cache, eb); |
4728 | if (cache) | |
4729 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
4730 | free_extent_buffer(eb); |
4731 | return ret; | |
4732 | } | |
be1a1d7a | 4733 | if (cache) { |
d3e29967 NB |
4734 | /* |
4735 | * Implies write in zoned mode. Mark the last eb in a block group. | |
4736 | */ | |
56fbb0a4 | 4737 | btrfs_schedule_zone_finish_bg(cache, eb); |
d3e29967 | 4738 | btrfs_put_block_group(cache); |
be1a1d7a | 4739 | } |
f91e0d0c QW |
4740 | ret = write_one_eb(eb, wbc, epd); |
4741 | free_extent_buffer(eb); | |
4742 | if (ret < 0) | |
4743 | return ret; | |
4744 | return 1; | |
4745 | } | |
4746 | ||
0b32f4bb JB |
4747 | int btree_write_cache_pages(struct address_space *mapping, |
4748 | struct writeback_control *wbc) | |
4749 | { | |
f91e0d0c | 4750 | struct extent_buffer *eb_context = NULL; |
0b32f4bb | 4751 | struct extent_page_data epd = { |
390ed29b | 4752 | .bio_ctrl = { 0 }, |
0b32f4bb JB |
4753 | .extent_locked = 0, |
4754 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4755 | }; | |
b3ff8f1d | 4756 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4757 | int ret = 0; |
4758 | int done = 0; | |
4759 | int nr_to_write_done = 0; | |
4760 | struct pagevec pvec; | |
4761 | int nr_pages; | |
4762 | pgoff_t index; | |
4763 | pgoff_t end; /* Inclusive */ | |
4764 | int scanned = 0; | |
10bbd235 | 4765 | xa_mark_t tag; |
0b32f4bb | 4766 | |
86679820 | 4767 | pagevec_init(&pvec); |
0b32f4bb JB |
4768 | if (wbc->range_cyclic) { |
4769 | index = mapping->writeback_index; /* Start from prev offset */ | |
4770 | end = -1; | |
556755a8 JB |
4771 | /* |
4772 | * Start from the beginning does not need to cycle over the | |
4773 | * range, mark it as scanned. | |
4774 | */ | |
4775 | scanned = (index == 0); | |
0b32f4bb | 4776 | } else { |
09cbfeaf KS |
4777 | index = wbc->range_start >> PAGE_SHIFT; |
4778 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4779 | scanned = 1; |
4780 | } | |
4781 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4782 | tag = PAGECACHE_TAG_TOWRITE; | |
4783 | else | |
4784 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 4785 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
4786 | retry: |
4787 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4788 | tag_pages_for_writeback(mapping, index, end); | |
4789 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4790 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4791 | tag))) { |
0b32f4bb JB |
4792 | unsigned i; |
4793 | ||
0b32f4bb JB |
4794 | for (i = 0; i < nr_pages; i++) { |
4795 | struct page *page = pvec.pages[i]; | |
4796 | ||
f91e0d0c QW |
4797 | ret = submit_eb_page(page, wbc, &epd, &eb_context); |
4798 | if (ret == 0) | |
0b32f4bb | 4799 | continue; |
f91e0d0c | 4800 | if (ret < 0) { |
0b32f4bb | 4801 | done = 1; |
0b32f4bb JB |
4802 | break; |
4803 | } | |
0b32f4bb JB |
4804 | |
4805 | /* | |
4806 | * the filesystem may choose to bump up nr_to_write. | |
4807 | * We have to make sure to honor the new nr_to_write | |
4808 | * at any time | |
4809 | */ | |
4810 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4811 | } | |
4812 | pagevec_release(&pvec); | |
4813 | cond_resched(); | |
4814 | } | |
4815 | if (!scanned && !done) { | |
4816 | /* | |
4817 | * We hit the last page and there is more work to be done: wrap | |
4818 | * back to the start of the file | |
4819 | */ | |
4820 | scanned = 1; | |
4821 | index = 0; | |
4822 | goto retry; | |
4823 | } | |
b3ff8f1d QW |
4824 | /* |
4825 | * If something went wrong, don't allow any metadata write bio to be | |
4826 | * submitted. | |
4827 | * | |
4828 | * This would prevent use-after-free if we had dirty pages not | |
4829 | * cleaned up, which can still happen by fuzzed images. | |
4830 | * | |
4831 | * - Bad extent tree | |
4832 | * Allowing existing tree block to be allocated for other trees. | |
4833 | * | |
4834 | * - Log tree operations | |
4835 | * Exiting tree blocks get allocated to log tree, bumps its | |
4836 | * generation, then get cleaned in tree re-balance. | |
4837 | * Such tree block will not be written back, since it's clean, | |
4838 | * thus no WRITTEN flag set. | |
4839 | * And after log writes back, this tree block is not traced by | |
4840 | * any dirty extent_io_tree. | |
4841 | * | |
4842 | * - Offending tree block gets re-dirtied from its original owner | |
4843 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4844 | * reused without COWing. This tree block will not be traced | |
4845 | * by btrfs_transaction::dirty_pages. | |
4846 | * | |
4847 | * Now such dirty tree block will not be cleaned by any dirty | |
4848 | * extent io tree. Thus we don't want to submit such wild eb | |
4849 | * if the fs already has error. | |
9845e5dd | 4850 | * |
c9583ada QW |
4851 | * We can get ret > 0 from submit_extent_page() indicating how many ebs |
4852 | * were submitted. Reset it to 0 to avoid false alerts for the caller. | |
4853 | */ | |
4854 | if (ret > 0) | |
4855 | ret = 0; | |
9845e5dd CH |
4856 | if (!ret && BTRFS_FS_ERROR(fs_info)) |
4857 | ret = -EROFS; | |
4858 | submit_write_bio(&epd, ret); | |
4859 | ||
4860 | btrfs_zoned_meta_io_unlock(fs_info); | |
0b32f4bb JB |
4861 | return ret; |
4862 | } | |
4863 | ||
d1310b2e | 4864 | /** |
3bed2da1 NB |
4865 | * Walk the list of dirty pages of the given address space and write all of them. |
4866 | * | |
d1310b2e | 4867 | * @mapping: address space structure to write |
3bed2da1 NB |
4868 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
4869 | * @epd: holds context for the write, namely the bio | |
d1310b2e CM |
4870 | * |
4871 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4872 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4873 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4874 | * and msync() need to guarantee that all the data which was dirty at the time | |
4875 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4876 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4877 | * existing IO to complete. | |
4878 | */ | |
4242b64a | 4879 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4880 | struct writeback_control *wbc, |
aab6e9ed | 4881 | struct extent_page_data *epd) |
d1310b2e | 4882 | { |
7fd1a3f7 | 4883 | struct inode *inode = mapping->host; |
d1310b2e CM |
4884 | int ret = 0; |
4885 | int done = 0; | |
f85d7d6c | 4886 | int nr_to_write_done = 0; |
d1310b2e CM |
4887 | struct pagevec pvec; |
4888 | int nr_pages; | |
4889 | pgoff_t index; | |
4890 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4891 | pgoff_t done_index; |
4892 | int range_whole = 0; | |
d1310b2e | 4893 | int scanned = 0; |
10bbd235 | 4894 | xa_mark_t tag; |
d1310b2e | 4895 | |
7fd1a3f7 JB |
4896 | /* |
4897 | * We have to hold onto the inode so that ordered extents can do their | |
4898 | * work when the IO finishes. The alternative to this is failing to add | |
4899 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4900 | * to deal with, so instead just hold onto the inode throughout the | |
4901 | * writepages operation. If it fails here we are freeing up the inode | |
4902 | * anyway and we'd rather not waste our time writing out stuff that is | |
4903 | * going to be truncated anyway. | |
4904 | */ | |
4905 | if (!igrab(inode)) | |
4906 | return 0; | |
4907 | ||
86679820 | 4908 | pagevec_init(&pvec); |
d1310b2e CM |
4909 | if (wbc->range_cyclic) { |
4910 | index = mapping->writeback_index; /* Start from prev offset */ | |
4911 | end = -1; | |
556755a8 JB |
4912 | /* |
4913 | * Start from the beginning does not need to cycle over the | |
4914 | * range, mark it as scanned. | |
4915 | */ | |
4916 | scanned = (index == 0); | |
d1310b2e | 4917 | } else { |
09cbfeaf KS |
4918 | index = wbc->range_start >> PAGE_SHIFT; |
4919 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4920 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4921 | range_whole = 1; | |
d1310b2e CM |
4922 | scanned = 1; |
4923 | } | |
3cd24c69 EL |
4924 | |
4925 | /* | |
4926 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4927 | * and we are the first one who do the filemap_flush() on this inode. | |
4928 | * | |
4929 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4930 | * not race in and drop the bit. | |
4931 | */ | |
4932 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4933 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4934 | &BTRFS_I(inode)->runtime_flags)) | |
4935 | wbc->tagged_writepages = 1; | |
4936 | ||
4937 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4938 | tag = PAGECACHE_TAG_TOWRITE; |
4939 | else | |
4940 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4941 | retry: |
3cd24c69 | 4942 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4943 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4944 | done_index = index; |
f85d7d6c | 4945 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4946 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4947 | &index, end, tag))) { | |
d1310b2e CM |
4948 | unsigned i; |
4949 | ||
d1310b2e CM |
4950 | for (i = 0; i < nr_pages; i++) { |
4951 | struct page *page = pvec.pages[i]; | |
4952 | ||
f7bddf1e | 4953 | done_index = page->index + 1; |
d1310b2e | 4954 | /* |
b93b0163 MW |
4955 | * At this point we hold neither the i_pages lock nor |
4956 | * the page lock: the page may be truncated or | |
4957 | * invalidated (changing page->mapping to NULL), | |
4958 | * or even swizzled back from swapper_space to | |
4959 | * tmpfs file mapping | |
d1310b2e | 4960 | */ |
c8f2f24b | 4961 | if (!trylock_page(page)) { |
9845e5dd | 4962 | submit_write_bio(epd, 0); |
c8f2f24b | 4963 | lock_page(page); |
01d658f2 | 4964 | } |
d1310b2e CM |
4965 | |
4966 | if (unlikely(page->mapping != mapping)) { | |
4967 | unlock_page(page); | |
4968 | continue; | |
4969 | } | |
4970 | ||
d2c3f4f6 | 4971 | if (wbc->sync_mode != WB_SYNC_NONE) { |
c9583ada | 4972 | if (PageWriteback(page)) |
9845e5dd | 4973 | submit_write_bio(epd, 0); |
d1310b2e | 4974 | wait_on_page_writeback(page); |
d2c3f4f6 | 4975 | } |
d1310b2e CM |
4976 | |
4977 | if (PageWriteback(page) || | |
4978 | !clear_page_dirty_for_io(page)) { | |
4979 | unlock_page(page); | |
4980 | continue; | |
4981 | } | |
4982 | ||
aab6e9ed | 4983 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 4984 | if (ret < 0) { |
a9132667 LB |
4985 | done = 1; |
4986 | break; | |
4987 | } | |
f85d7d6c CM |
4988 | |
4989 | /* | |
4990 | * the filesystem may choose to bump up nr_to_write. | |
4991 | * We have to make sure to honor the new nr_to_write | |
4992 | * at any time | |
4993 | */ | |
4994 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4995 | } |
4996 | pagevec_release(&pvec); | |
4997 | cond_resched(); | |
4998 | } | |
894b36e3 | 4999 | if (!scanned && !done) { |
d1310b2e CM |
5000 | /* |
5001 | * We hit the last page and there is more work to be done: wrap | |
5002 | * back to the start of the file | |
5003 | */ | |
5004 | scanned = 1; | |
5005 | index = 0; | |
42ffb0bf JB |
5006 | |
5007 | /* | |
5008 | * If we're looping we could run into a page that is locked by a | |
5009 | * writer and that writer could be waiting on writeback for a | |
5010 | * page in our current bio, and thus deadlock, so flush the | |
5011 | * write bio here. | |
5012 | */ | |
9845e5dd | 5013 | submit_write_bio(epd, 0); |
c9583ada | 5014 | goto retry; |
d1310b2e | 5015 | } |
a9132667 LB |
5016 | |
5017 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
5018 | mapping->writeback_index = done_index; | |
5019 | ||
7fd1a3f7 | 5020 | btrfs_add_delayed_iput(inode); |
894b36e3 | 5021 | return ret; |
d1310b2e | 5022 | } |
d1310b2e | 5023 | |
2bd0fc93 QW |
5024 | /* |
5025 | * Submit the pages in the range to bio for call sites which delalloc range has | |
5026 | * already been ran (aka, ordered extent inserted) and all pages are still | |
5027 | * locked. | |
5028 | */ | |
5029 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end) | |
771ed689 | 5030 | { |
2bd0fc93 QW |
5031 | bool found_error = false; |
5032 | int first_error = 0; | |
771ed689 CM |
5033 | int ret = 0; |
5034 | struct address_space *mapping = inode->i_mapping; | |
5035 | struct page *page; | |
2bd0fc93 | 5036 | u64 cur = start; |
66448b9d QW |
5037 | unsigned long nr_pages; |
5038 | const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize; | |
771ed689 | 5039 | struct extent_page_data epd = { |
390ed29b | 5040 | .bio_ctrl = { 0 }, |
771ed689 | 5041 | .extent_locked = 1, |
2bd0fc93 | 5042 | .sync_io = 1, |
771ed689 CM |
5043 | }; |
5044 | struct writeback_control wbc_writepages = { | |
2bd0fc93 | 5045 | .sync_mode = WB_SYNC_ALL, |
771ed689 CM |
5046 | .range_start = start, |
5047 | .range_end = end + 1, | |
ec39f769 CM |
5048 | /* We're called from an async helper function */ |
5049 | .punt_to_cgroup = 1, | |
5050 | .no_cgroup_owner = 1, | |
771ed689 CM |
5051 | }; |
5052 | ||
66448b9d QW |
5053 | ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize)); |
5054 | nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >> | |
5055 | PAGE_SHIFT; | |
5056 | wbc_writepages.nr_to_write = nr_pages * 2; | |
5057 | ||
dbb70bec | 5058 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
2bd0fc93 | 5059 | while (cur <= end) { |
66448b9d QW |
5060 | u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end); |
5061 | ||
2bd0fc93 QW |
5062 | page = find_get_page(mapping, cur >> PAGE_SHIFT); |
5063 | /* | |
5064 | * All pages in the range are locked since | |
5065 | * btrfs_run_delalloc_range(), thus there is no way to clear | |
5066 | * the page dirty flag. | |
5067 | */ | |
66448b9d | 5068 | ASSERT(PageLocked(page)); |
2bd0fc93 QW |
5069 | ASSERT(PageDirty(page)); |
5070 | clear_page_dirty_for_io(page); | |
5071 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
5072 | ASSERT(ret <= 0); | |
5073 | if (ret < 0) { | |
5074 | found_error = true; | |
5075 | first_error = ret; | |
771ed689 | 5076 | } |
09cbfeaf | 5077 | put_page(page); |
66448b9d | 5078 | cur = cur_end + 1; |
771ed689 CM |
5079 | } |
5080 | ||
9845e5dd | 5081 | submit_write_bio(&epd, found_error ? ret : 0); |
dbb70bec CM |
5082 | |
5083 | wbc_detach_inode(&wbc_writepages); | |
2bd0fc93 QW |
5084 | if (found_error) |
5085 | return first_error; | |
771ed689 CM |
5086 | return ret; |
5087 | } | |
d1310b2e | 5088 | |
8ae225a8 | 5089 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
5090 | struct writeback_control *wbc) |
5091 | { | |
35156d85 | 5092 | struct inode *inode = mapping->host; |
d1310b2e CM |
5093 | int ret = 0; |
5094 | struct extent_page_data epd = { | |
390ed29b | 5095 | .bio_ctrl = { 0 }, |
771ed689 | 5096 | .extent_locked = 0, |
ffbd517d | 5097 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
5098 | }; |
5099 | ||
35156d85 JT |
5100 | /* |
5101 | * Allow only a single thread to do the reloc work in zoned mode to | |
5102 | * protect the write pointer updates. | |
5103 | */ | |
869f4cdc | 5104 | btrfs_zoned_data_reloc_lock(BTRFS_I(inode)); |
935db853 | 5105 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
9845e5dd | 5106 | submit_write_bio(&epd, ret); |
19ab78ca | 5107 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
d1310b2e CM |
5108 | return ret; |
5109 | } | |
d1310b2e | 5110 | |
ba206a02 | 5111 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 5112 | { |
390ed29b | 5113 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
67c9684f | 5114 | struct page *pagepool[16]; |
125bac01 | 5115 | struct extent_map *em_cached = NULL; |
808f80b4 | 5116 | u64 prev_em_start = (u64)-1; |
ba206a02 | 5117 | int nr; |
d1310b2e | 5118 | |
ba206a02 | 5119 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
5120 | u64 contig_start = readahead_pos(rac); |
5121 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 5122 | |
ba206a02 | 5123 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 5124 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 5125 | } |
67c9684f | 5126 | |
125bac01 MX |
5127 | if (em_cached) |
5128 | free_extent_map(em_cached); | |
722c82ac | 5129 | submit_one_bio(&bio_ctrl); |
d1310b2e | 5130 | } |
d1310b2e CM |
5131 | |
5132 | /* | |
895586eb MWO |
5133 | * basic invalidate_folio code, this waits on any locked or writeback |
5134 | * ranges corresponding to the folio, and then deletes any extent state | |
d1310b2e CM |
5135 | * records from the tree |
5136 | */ | |
895586eb MWO |
5137 | int extent_invalidate_folio(struct extent_io_tree *tree, |
5138 | struct folio *folio, size_t offset) | |
d1310b2e | 5139 | { |
2ac55d41 | 5140 | struct extent_state *cached_state = NULL; |
895586eb MWO |
5141 | u64 start = folio_pos(folio); |
5142 | u64 end = start + folio_size(folio) - 1; | |
5143 | size_t blocksize = folio->mapping->host->i_sb->s_blocksize; | |
d1310b2e | 5144 | |
829ddec9 QW |
5145 | /* This function is only called for the btree inode */ |
5146 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
5147 | ||
fda2832f | 5148 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
5149 | if (start > end) |
5150 | return 0; | |
5151 | ||
ff13db41 | 5152 | lock_extent_bits(tree, start, end, &cached_state); |
895586eb | 5153 | folio_wait_writeback(folio); |
829ddec9 QW |
5154 | |
5155 | /* | |
5156 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
5157 | * so here we only need to unlock the extent range to free any | |
5158 | * existing extent state. | |
5159 | */ | |
5160 | unlock_extent_cached(tree, start, end, &cached_state); | |
d1310b2e CM |
5161 | return 0; |
5162 | } | |
d1310b2e | 5163 | |
7b13b7b1 | 5164 | /* |
f913cff3 | 5165 | * a helper for release_folio, this tests for areas of the page that |
7b13b7b1 CM |
5166 | * are locked or under IO and drops the related state bits if it is safe |
5167 | * to drop the page. | |
5168 | */ | |
29c68b2d | 5169 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 5170 | struct page *page, gfp_t mask) |
7b13b7b1 | 5171 | { |
4eee4fa4 | 5172 | u64 start = page_offset(page); |
09cbfeaf | 5173 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
5174 | int ret = 1; |
5175 | ||
8882679e | 5176 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 5177 | ret = 0; |
8882679e | 5178 | } else { |
11ef160f | 5179 | /* |
2766ff61 FM |
5180 | * At this point we can safely clear everything except the |
5181 | * locked bit, the nodatasum bit and the delalloc new bit. | |
5182 | * The delalloc new bit will be cleared by ordered extent | |
5183 | * completion. | |
11ef160f | 5184 | */ |
66b0c887 | 5185 | ret = __clear_extent_bit(tree, start, end, |
2766ff61 FM |
5186 | ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW), |
5187 | 0, 0, NULL, mask, NULL); | |
e3f24cc5 CM |
5188 | |
5189 | /* if clear_extent_bit failed for enomem reasons, | |
5190 | * we can't allow the release to continue. | |
5191 | */ | |
5192 | if (ret < 0) | |
5193 | ret = 0; | |
5194 | else | |
5195 | ret = 1; | |
7b13b7b1 CM |
5196 | } |
5197 | return ret; | |
5198 | } | |
7b13b7b1 | 5199 | |
d1310b2e | 5200 | /* |
f913cff3 | 5201 | * a helper for release_folio. As long as there are no locked extents |
d1310b2e CM |
5202 | * in the range corresponding to the page, both state records and extent |
5203 | * map records are removed | |
5204 | */ | |
477a30ba | 5205 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
5206 | { |
5207 | struct extent_map *em; | |
4eee4fa4 | 5208 | u64 start = page_offset(page); |
09cbfeaf | 5209 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
5210 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
5211 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
5212 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 5213 | |
d0164adc | 5214 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 5215 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 5216 | u64 len; |
70dec807 | 5217 | while (start <= end) { |
fbc2bd7e FM |
5218 | struct btrfs_fs_info *fs_info; |
5219 | u64 cur_gen; | |
5220 | ||
39b5637f | 5221 | len = end - start + 1; |
890871be | 5222 | write_lock(&map->lock); |
39b5637f | 5223 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 5224 | if (!em) { |
890871be | 5225 | write_unlock(&map->lock); |
70dec807 CM |
5226 | break; |
5227 | } | |
7f3c74fb CM |
5228 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
5229 | em->start != start) { | |
890871be | 5230 | write_unlock(&map->lock); |
70dec807 CM |
5231 | free_extent_map(em); |
5232 | break; | |
5233 | } | |
3d6448e6 FM |
5234 | if (test_range_bit(tree, em->start, |
5235 | extent_map_end(em) - 1, | |
5236 | EXTENT_LOCKED, 0, NULL)) | |
5237 | goto next; | |
5238 | /* | |
5239 | * If it's not in the list of modified extents, used | |
5240 | * by a fast fsync, we can remove it. If it's being | |
5241 | * logged we can safely remove it since fsync took an | |
5242 | * extra reference on the em. | |
5243 | */ | |
5244 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
5245 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
5246 | goto remove_em; | |
5247 | /* | |
5248 | * If it's in the list of modified extents, remove it | |
5249 | * only if its generation is older then the current one, | |
5250 | * in which case we don't need it for a fast fsync. | |
5251 | * Otherwise don't remove it, we could be racing with an | |
5252 | * ongoing fast fsync that could miss the new extent. | |
5253 | */ | |
5254 | fs_info = btrfs_inode->root->fs_info; | |
5255 | spin_lock(&fs_info->trans_lock); | |
5256 | cur_gen = fs_info->generation; | |
5257 | spin_unlock(&fs_info->trans_lock); | |
5258 | if (em->generation >= cur_gen) | |
5259 | goto next; | |
5260 | remove_em: | |
5e548b32 FM |
5261 | /* |
5262 | * We only remove extent maps that are not in the list of | |
5263 | * modified extents or that are in the list but with a | |
5264 | * generation lower then the current generation, so there | |
5265 | * is no need to set the full fsync flag on the inode (it | |
5266 | * hurts the fsync performance for workloads with a data | |
5267 | * size that exceeds or is close to the system's memory). | |
5268 | */ | |
fbc2bd7e FM |
5269 | remove_extent_mapping(map, em); |
5270 | /* once for the rb tree */ | |
5271 | free_extent_map(em); | |
3d6448e6 | 5272 | next: |
70dec807 | 5273 | start = extent_map_end(em); |
890871be | 5274 | write_unlock(&map->lock); |
70dec807 CM |
5275 | |
5276 | /* once for us */ | |
d1310b2e | 5277 | free_extent_map(em); |
9f47eb54 PM |
5278 | |
5279 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 5280 | } |
d1310b2e | 5281 | } |
29c68b2d | 5282 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 5283 | } |
d1310b2e | 5284 | |
ec29ed5b CM |
5285 | /* |
5286 | * helper function for fiemap, which doesn't want to see any holes. | |
5287 | * This maps until we find something past 'last' | |
5288 | */ | |
f1bbde8d | 5289 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 5290 | u64 offset, u64 last) |
ec29ed5b | 5291 | { |
f1bbde8d | 5292 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
5293 | struct extent_map *em; |
5294 | u64 len; | |
5295 | ||
5296 | if (offset >= last) | |
5297 | return NULL; | |
5298 | ||
67871254 | 5299 | while (1) { |
ec29ed5b CM |
5300 | len = last - offset; |
5301 | if (len == 0) | |
5302 | break; | |
fda2832f | 5303 | len = ALIGN(len, sectorsize); |
f1bbde8d | 5304 | em = btrfs_get_extent_fiemap(inode, offset, len); |
6b5b7a41 | 5305 | if (IS_ERR(em)) |
ec29ed5b CM |
5306 | return em; |
5307 | ||
5308 | /* if this isn't a hole return it */ | |
4a2d25cd | 5309 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 5310 | return em; |
ec29ed5b CM |
5311 | |
5312 | /* this is a hole, advance to the next extent */ | |
5313 | offset = extent_map_end(em); | |
5314 | free_extent_map(em); | |
5315 | if (offset >= last) | |
5316 | break; | |
5317 | } | |
5318 | return NULL; | |
5319 | } | |
5320 | ||
4751832d QW |
5321 | /* |
5322 | * To cache previous fiemap extent | |
5323 | * | |
5324 | * Will be used for merging fiemap extent | |
5325 | */ | |
5326 | struct fiemap_cache { | |
5327 | u64 offset; | |
5328 | u64 phys; | |
5329 | u64 len; | |
5330 | u32 flags; | |
5331 | bool cached; | |
5332 | }; | |
5333 | ||
5334 | /* | |
5335 | * Helper to submit fiemap extent. | |
5336 | * | |
5337 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
5338 | * @len and @flags with cached one. | |
5339 | * And only when we fails to merge, cached one will be submitted as | |
5340 | * fiemap extent. | |
5341 | * | |
5342 | * Return value is the same as fiemap_fill_next_extent(). | |
5343 | */ | |
5344 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
5345 | struct fiemap_cache *cache, | |
5346 | u64 offset, u64 phys, u64 len, u32 flags) | |
5347 | { | |
5348 | int ret = 0; | |
5349 | ||
5350 | if (!cache->cached) | |
5351 | goto assign; | |
5352 | ||
5353 | /* | |
5354 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 5355 | * fiemap extent won't overlap with cached one. |
4751832d QW |
5356 | * Not recoverable. |
5357 | * | |
5358 | * NOTE: Physical address can overlap, due to compression | |
5359 | */ | |
5360 | if (cache->offset + cache->len > offset) { | |
5361 | WARN_ON(1); | |
5362 | return -EINVAL; | |
5363 | } | |
5364 | ||
5365 | /* | |
5366 | * Only merges fiemap extents if | |
5367 | * 1) Their logical addresses are continuous | |
5368 | * | |
5369 | * 2) Their physical addresses are continuous | |
5370 | * So truly compressed (physical size smaller than logical size) | |
5371 | * extents won't get merged with each other | |
5372 | * | |
5373 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
5374 | * So regular extent won't get merged with prealloc extent | |
5375 | */ | |
5376 | if (cache->offset + cache->len == offset && | |
5377 | cache->phys + cache->len == phys && | |
5378 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
5379 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
5380 | cache->len += len; | |
5381 | cache->flags |= flags; | |
5382 | goto try_submit_last; | |
5383 | } | |
5384 | ||
5385 | /* Not mergeable, need to submit cached one */ | |
5386 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
5387 | cache->len, cache->flags); | |
5388 | cache->cached = false; | |
5389 | if (ret) | |
5390 | return ret; | |
5391 | assign: | |
5392 | cache->cached = true; | |
5393 | cache->offset = offset; | |
5394 | cache->phys = phys; | |
5395 | cache->len = len; | |
5396 | cache->flags = flags; | |
5397 | try_submit_last: | |
5398 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
5399 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
5400 | cache->phys, cache->len, cache->flags); | |
5401 | cache->cached = false; | |
5402 | } | |
5403 | return ret; | |
5404 | } | |
5405 | ||
5406 | /* | |
848c23b7 | 5407 | * Emit last fiemap cache |
4751832d | 5408 | * |
848c23b7 QW |
5409 | * The last fiemap cache may still be cached in the following case: |
5410 | * 0 4k 8k | |
5411 | * |<- Fiemap range ->| | |
5412 | * |<------------ First extent ----------->| | |
5413 | * | |
5414 | * In this case, the first extent range will be cached but not emitted. | |
5415 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 5416 | */ |
5c5aff98 | 5417 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 5418 | struct fiemap_cache *cache) |
4751832d QW |
5419 | { |
5420 | int ret; | |
5421 | ||
5422 | if (!cache->cached) | |
5423 | return 0; | |
5424 | ||
4751832d QW |
5425 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
5426 | cache->len, cache->flags); | |
5427 | cache->cached = false; | |
5428 | if (ret > 0) | |
5429 | ret = 0; | |
5430 | return ret; | |
5431 | } | |
5432 | ||
facee0a0 | 5433 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 5434 | u64 start, u64 len) |
1506fcc8 | 5435 | { |
975f84fe | 5436 | int ret = 0; |
15c7745c | 5437 | u64 off; |
1506fcc8 YS |
5438 | u64 max = start + len; |
5439 | u32 flags = 0; | |
975f84fe JB |
5440 | u32 found_type; |
5441 | u64 last; | |
ec29ed5b | 5442 | u64 last_for_get_extent = 0; |
1506fcc8 | 5443 | u64 disko = 0; |
facee0a0 | 5444 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 5445 | struct btrfs_key found_key; |
1506fcc8 | 5446 | struct extent_map *em = NULL; |
2ac55d41 | 5447 | struct extent_state *cached_state = NULL; |
975f84fe | 5448 | struct btrfs_path *path; |
facee0a0 | 5449 | struct btrfs_root *root = inode->root; |
4751832d | 5450 | struct fiemap_cache cache = { 0 }; |
12a824dc | 5451 | struct btrfs_backref_shared_cache *backref_cache; |
5911c8fe DS |
5452 | struct ulist *roots; |
5453 | struct ulist *tmp_ulist; | |
1506fcc8 | 5454 | int end = 0; |
ec29ed5b CM |
5455 | u64 em_start = 0; |
5456 | u64 em_len = 0; | |
5457 | u64 em_end = 0; | |
1506fcc8 | 5458 | |
12a824dc | 5459 | backref_cache = kzalloc(sizeof(*backref_cache), GFP_KERNEL); |
975f84fe | 5460 | path = btrfs_alloc_path(); |
5911c8fe DS |
5461 | roots = ulist_alloc(GFP_KERNEL); |
5462 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
12a824dc | 5463 | if (!backref_cache || !path || !roots || !tmp_ulist) { |
5911c8fe DS |
5464 | ret = -ENOMEM; |
5465 | goto out_free_ulist; | |
5466 | } | |
5467 | ||
15c7745c BB |
5468 | /* |
5469 | * We can't initialize that to 'start' as this could miss extents due | |
5470 | * to extent item merging | |
5471 | */ | |
5472 | off = 0; | |
facee0a0 NB |
5473 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
5474 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 5475 | |
ec29ed5b CM |
5476 | /* |
5477 | * lookup the last file extent. We're not using i_size here | |
5478 | * because there might be preallocation past i_size | |
5479 | */ | |
facee0a0 NB |
5480 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
5481 | 0); | |
975f84fe | 5482 | if (ret < 0) { |
5911c8fe | 5483 | goto out_free_ulist; |
2d324f59 LB |
5484 | } else { |
5485 | WARN_ON(!ret); | |
5486 | if (ret == 1) | |
5487 | ret = 0; | |
975f84fe | 5488 | } |
2d324f59 | 5489 | |
975f84fe | 5490 | path->slots[0]--; |
975f84fe | 5491 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 5492 | found_type = found_key.type; |
975f84fe | 5493 | |
ec29ed5b | 5494 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 5495 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 5496 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
5497 | /* have to trust i_size as the end */ |
5498 | last = (u64)-1; | |
5499 | last_for_get_extent = isize; | |
5500 | } else { | |
5501 | /* | |
5502 | * remember the start of the last extent. There are a | |
5503 | * bunch of different factors that go into the length of the | |
5504 | * extent, so its much less complex to remember where it started | |
5505 | */ | |
5506 | last = found_key.offset; | |
5507 | last_for_get_extent = last + 1; | |
975f84fe | 5508 | } |
fe09e16c | 5509 | btrfs_release_path(path); |
975f84fe | 5510 | |
ec29ed5b CM |
5511 | /* |
5512 | * we might have some extents allocated but more delalloc past those | |
5513 | * extents. so, we trust isize unless the start of the last extent is | |
5514 | * beyond isize | |
5515 | */ | |
5516 | if (last < isize) { | |
5517 | last = (u64)-1; | |
5518 | last_for_get_extent = isize; | |
5519 | } | |
5520 | ||
facee0a0 | 5521 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 5522 | &cached_state); |
ec29ed5b | 5523 | |
facee0a0 | 5524 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
5525 | if (!em) |
5526 | goto out; | |
5527 | if (IS_ERR(em)) { | |
5528 | ret = PTR_ERR(em); | |
5529 | goto out; | |
5530 | } | |
975f84fe | 5531 | |
1506fcc8 | 5532 | while (!end) { |
b76bb701 | 5533 | u64 offset_in_extent = 0; |
ea8efc74 CM |
5534 | |
5535 | /* break if the extent we found is outside the range */ | |
5536 | if (em->start >= max || extent_map_end(em) < off) | |
5537 | break; | |
5538 | ||
5539 | /* | |
5540 | * get_extent may return an extent that starts before our | |
5541 | * requested range. We have to make sure the ranges | |
5542 | * we return to fiemap always move forward and don't | |
5543 | * overlap, so adjust the offsets here | |
5544 | */ | |
5545 | em_start = max(em->start, off); | |
1506fcc8 | 5546 | |
ea8efc74 CM |
5547 | /* |
5548 | * record the offset from the start of the extent | |
b76bb701 JB |
5549 | * for adjusting the disk offset below. Only do this if the |
5550 | * extent isn't compressed since our in ram offset may be past | |
5551 | * what we have actually allocated on disk. | |
ea8efc74 | 5552 | */ |
b76bb701 JB |
5553 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
5554 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 5555 | em_end = extent_map_end(em); |
ea8efc74 | 5556 | em_len = em_end - em_start; |
1506fcc8 | 5557 | flags = 0; |
f0986318 FM |
5558 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
5559 | disko = em->block_start + offset_in_extent; | |
5560 | else | |
5561 | disko = 0; | |
1506fcc8 | 5562 | |
ea8efc74 CM |
5563 | /* |
5564 | * bump off for our next call to get_extent | |
5565 | */ | |
5566 | off = extent_map_end(em); | |
5567 | if (off >= max) | |
5568 | end = 1; | |
5569 | ||
f12eec9a | 5570 | if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
5571 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
5572 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 5573 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
5574 | flags |= (FIEMAP_EXTENT_DELALLOC | |
5575 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 | 5576 | } else if (fieinfo->fi_extents_max) { |
b8f164e3 | 5577 | u64 extent_gen; |
dc046b10 JB |
5578 | u64 bytenr = em->block_start - |
5579 | (em->start - em->orig_start); | |
fe09e16c | 5580 | |
b8f164e3 FM |
5581 | /* |
5582 | * If two extent maps are merged, then their generation | |
5583 | * is set to the maximum between their generations. | |
5584 | * Otherwise its generation matches the one we have in | |
5585 | * corresponding file extent item. If we have a merged | |
5586 | * extent map, don't use its generation to speedup the | |
5587 | * sharedness check below. | |
5588 | */ | |
5589 | if (test_bit(EXTENT_FLAG_MERGED, &em->flags)) | |
5590 | extent_gen = 0; | |
5591 | else | |
5592 | extent_gen = em->generation; | |
5593 | ||
fe09e16c LB |
5594 | /* |
5595 | * As btrfs supports shared space, this information | |
5596 | * can be exported to userspace tools via | |
dc046b10 JB |
5597 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
5598 | * then we're just getting a count and we can skip the | |
5599 | * lookup stuff. | |
fe09e16c | 5600 | */ |
8eedadda | 5601 | ret = btrfs_is_data_extent_shared(root, btrfs_ino(inode), |
b8f164e3 FM |
5602 | bytenr, extent_gen, |
5603 | roots, tmp_ulist, | |
12a824dc | 5604 | backref_cache); |
dc046b10 | 5605 | if (ret < 0) |
fe09e16c | 5606 | goto out_free; |
dc046b10 | 5607 | if (ret) |
fe09e16c | 5608 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 5609 | ret = 0; |
1506fcc8 YS |
5610 | } |
5611 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
5612 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
5613 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
5614 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 5615 | |
1506fcc8 YS |
5616 | free_extent_map(em); |
5617 | em = NULL; | |
ec29ed5b CM |
5618 | if ((em_start >= last) || em_len == (u64)-1 || |
5619 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
5620 | flags |= FIEMAP_EXTENT_LAST; |
5621 | end = 1; | |
5622 | } | |
5623 | ||
ec29ed5b | 5624 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 5625 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
5626 | if (IS_ERR(em)) { |
5627 | ret = PTR_ERR(em); | |
5628 | goto out; | |
5629 | } | |
5630 | if (!em) { | |
975f84fe JB |
5631 | flags |= FIEMAP_EXTENT_LAST; |
5632 | end = 1; | |
5633 | } | |
4751832d QW |
5634 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
5635 | em_len, flags); | |
26e726af CS |
5636 | if (ret) { |
5637 | if (ret == 1) | |
5638 | ret = 0; | |
ec29ed5b | 5639 | goto out_free; |
26e726af | 5640 | } |
09fbc1c8 FM |
5641 | |
5642 | if (fatal_signal_pending(current)) { | |
5643 | ret = -EINTR; | |
5644 | goto out_free; | |
5645 | } | |
1506fcc8 YS |
5646 | } |
5647 | out_free: | |
4751832d | 5648 | if (!ret) |
5c5aff98 | 5649 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
5650 | free_extent_map(em); |
5651 | out: | |
facee0a0 | 5652 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 5653 | &cached_state); |
5911c8fe DS |
5654 | |
5655 | out_free_ulist: | |
12a824dc | 5656 | kfree(backref_cache); |
e02d48ea | 5657 | btrfs_free_path(path); |
5911c8fe DS |
5658 | ulist_free(roots); |
5659 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
5660 | return ret; |
5661 | } | |
5662 | ||
727011e0 CM |
5663 | static void __free_extent_buffer(struct extent_buffer *eb) |
5664 | { | |
727011e0 CM |
5665 | kmem_cache_free(extent_buffer_cache, eb); |
5666 | } | |
5667 | ||
2b48966a | 5668 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
5669 | { |
5670 | return (atomic_read(&eb->io_pages) || | |
5671 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
5672 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5673 | } | |
5674 | ||
8ff8466d | 5675 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 5676 | { |
8ff8466d | 5677 | struct btrfs_subpage *subpage; |
db7f3436 | 5678 | |
8ff8466d | 5679 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 5680 | |
8ff8466d QW |
5681 | if (PagePrivate(page)) { |
5682 | subpage = (struct btrfs_subpage *)page->private; | |
5683 | if (atomic_read(&subpage->eb_refs)) | |
5684 | return true; | |
3d078efa QW |
5685 | /* |
5686 | * Even there is no eb refs here, we may still have | |
5687 | * end_page_read() call relying on page::private. | |
5688 | */ | |
5689 | if (atomic_read(&subpage->readers)) | |
5690 | return true; | |
8ff8466d QW |
5691 | } |
5692 | return false; | |
5693 | } | |
db7f3436 | 5694 | |
8ff8466d QW |
5695 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
5696 | { | |
5697 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5698 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
5699 | ||
5700 | /* | |
5701 | * For mapped eb, we're going to change the page private, which should | |
5702 | * be done under the private_lock. | |
5703 | */ | |
5704 | if (mapped) | |
5705 | spin_lock(&page->mapping->private_lock); | |
5706 | ||
5707 | if (!PagePrivate(page)) { | |
5d2361db | 5708 | if (mapped) |
8ff8466d QW |
5709 | spin_unlock(&page->mapping->private_lock); |
5710 | return; | |
5711 | } | |
5712 | ||
fbca46eb | 5713 | if (fs_info->nodesize >= PAGE_SIZE) { |
5d2361db FL |
5714 | /* |
5715 | * We do this since we'll remove the pages after we've | |
5716 | * removed the eb from the radix tree, so we could race | |
5717 | * and have this page now attached to the new eb. So | |
5718 | * only clear page_private if it's still connected to | |
5719 | * this eb. | |
5720 | */ | |
5721 | if (PagePrivate(page) && | |
5722 | page->private == (unsigned long)eb) { | |
5723 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5724 | BUG_ON(PageDirty(page)); | |
5725 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 5726 | /* |
5d2361db FL |
5727 | * We need to make sure we haven't be attached |
5728 | * to a new eb. | |
db7f3436 | 5729 | */ |
d1b89bc0 | 5730 | detach_page_private(page); |
db7f3436 | 5731 | } |
5d2361db FL |
5732 | if (mapped) |
5733 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
5734 | return; |
5735 | } | |
5736 | ||
5737 | /* | |
5738 | * For subpage, we can have dummy eb with page private. In this case, | |
5739 | * we can directly detach the private as such page is only attached to | |
5740 | * one dummy eb, no sharing. | |
5741 | */ | |
5742 | if (!mapped) { | |
5743 | btrfs_detach_subpage(fs_info, page); | |
5744 | return; | |
5745 | } | |
5746 | ||
5747 | btrfs_page_dec_eb_refs(fs_info, page); | |
5748 | ||
5749 | /* | |
5750 | * We can only detach the page private if there are no other ebs in the | |
3d078efa | 5751 | * page range and no unfinished IO. |
8ff8466d QW |
5752 | */ |
5753 | if (!page_range_has_eb(fs_info, page)) | |
5754 | btrfs_detach_subpage(fs_info, page); | |
5755 | ||
5756 | spin_unlock(&page->mapping->private_lock); | |
5757 | } | |
5758 | ||
5759 | /* Release all pages attached to the extent buffer */ | |
5760 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
5761 | { | |
5762 | int i; | |
5763 | int num_pages; | |
5764 | ||
5765 | ASSERT(!extent_buffer_under_io(eb)); | |
5766 | ||
5767 | num_pages = num_extent_pages(eb); | |
5768 | for (i = 0; i < num_pages; i++) { | |
5769 | struct page *page = eb->pages[i]; | |
5770 | ||
5771 | if (!page) | |
5772 | continue; | |
5773 | ||
5774 | detach_extent_buffer_page(eb, page); | |
5d2361db | 5775 | |
01327610 | 5776 | /* One for when we allocated the page */ |
09cbfeaf | 5777 | put_page(page); |
d64766fd | 5778 | } |
db7f3436 JB |
5779 | } |
5780 | ||
5781 | /* | |
5782 | * Helper for releasing the extent buffer. | |
5783 | */ | |
5784 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
5785 | { | |
55ac0139 | 5786 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 5787 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
5788 | __free_extent_buffer(eb); |
5789 | } | |
5790 | ||
f28491e0 JB |
5791 | static struct extent_buffer * |
5792 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 5793 | unsigned long len) |
d1310b2e CM |
5794 | { |
5795 | struct extent_buffer *eb = NULL; | |
5796 | ||
d1b5c567 | 5797 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
5798 | eb->start = start; |
5799 | eb->len = len; | |
f28491e0 | 5800 | eb->fs_info = fs_info; |
815a51c7 | 5801 | eb->bflags = 0; |
196d59ab | 5802 | init_rwsem(&eb->lock); |
b4ce94de | 5803 | |
3fd63727 JB |
5804 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5805 | &fs_info->allocated_ebs); | |
d3575156 | 5806 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 5807 | |
3083ee2e | 5808 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5809 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5810 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5811 | |
deb67895 | 5812 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
5813 | |
5814 | return eb; | |
5815 | } | |
5816 | ||
2b48966a | 5817 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 5818 | { |
cc5e31a4 | 5819 | int i; |
815a51c7 | 5820 | struct extent_buffer *new; |
cc5e31a4 | 5821 | int num_pages = num_extent_pages(src); |
dd137dd1 | 5822 | int ret; |
815a51c7 | 5823 | |
3f556f78 | 5824 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5825 | if (new == NULL) |
5826 | return NULL; | |
5827 | ||
62c053fb QW |
5828 | /* |
5829 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
5830 | * btrfs_release_extent_buffer() have different behavior for | |
5831 | * UNMAPPED subpage extent buffer. | |
5832 | */ | |
5833 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
5834 | ||
dd137dd1 STD |
5835 | memset(new->pages, 0, sizeof(*new->pages) * num_pages); |
5836 | ret = btrfs_alloc_page_array(num_pages, new->pages); | |
5837 | if (ret) { | |
5838 | btrfs_release_extent_buffer(new); | |
5839 | return NULL; | |
5840 | } | |
5841 | ||
815a51c7 | 5842 | for (i = 0; i < num_pages; i++) { |
760f991f | 5843 | int ret; |
dd137dd1 | 5844 | struct page *p = new->pages[i]; |
760f991f | 5845 | |
760f991f QW |
5846 | ret = attach_extent_buffer_page(new, p, NULL); |
5847 | if (ret < 0) { | |
760f991f QW |
5848 | btrfs_release_extent_buffer(new); |
5849 | return NULL; | |
5850 | } | |
815a51c7 | 5851 | WARN_ON(PageDirty(p)); |
fba1acf9 | 5852 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 5853 | } |
92d83e94 | 5854 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
5855 | |
5856 | return new; | |
5857 | } | |
5858 | ||
0f331229 OS |
5859 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5860 | u64 start, unsigned long len) | |
815a51c7 JS |
5861 | { |
5862 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5863 | int num_pages; |
5864 | int i; | |
dd137dd1 | 5865 | int ret; |
815a51c7 | 5866 | |
3f556f78 | 5867 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5868 | if (!eb) |
5869 | return NULL; | |
5870 | ||
65ad0104 | 5871 | num_pages = num_extent_pages(eb); |
dd137dd1 STD |
5872 | ret = btrfs_alloc_page_array(num_pages, eb->pages); |
5873 | if (ret) | |
5874 | goto err; | |
5875 | ||
815a51c7 | 5876 | for (i = 0; i < num_pages; i++) { |
dd137dd1 | 5877 | struct page *p = eb->pages[i]; |
09bc1f0f | 5878 | |
dd137dd1 | 5879 | ret = attach_extent_buffer_page(eb, p, NULL); |
09bc1f0f QW |
5880 | if (ret < 0) |
5881 | goto err; | |
815a51c7 | 5882 | } |
dd137dd1 | 5883 | |
815a51c7 JS |
5884 | set_extent_buffer_uptodate(eb); |
5885 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 5886 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
5887 | |
5888 | return eb; | |
5889 | err: | |
dd137dd1 STD |
5890 | for (i = 0; i < num_pages; i++) { |
5891 | if (eb->pages[i]) { | |
5892 | detach_extent_buffer_page(eb, eb->pages[i]); | |
5893 | __free_page(eb->pages[i]); | |
5894 | } | |
09bc1f0f | 5895 | } |
815a51c7 JS |
5896 | __free_extent_buffer(eb); |
5897 | return NULL; | |
5898 | } | |
5899 | ||
0f331229 | 5900 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 5901 | u64 start) |
0f331229 | 5902 | { |
da17066c | 5903 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
5904 | } |
5905 | ||
0b32f4bb JB |
5906 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
5907 | { | |
242e18c7 | 5908 | int refs; |
6bf9cd2e BB |
5909 | /* |
5910 | * The TREE_REF bit is first set when the extent_buffer is added | |
5911 | * to the radix tree. It is also reset, if unset, when a new reference | |
5912 | * is created by find_extent_buffer. | |
0b32f4bb | 5913 | * |
6bf9cd2e BB |
5914 | * It is only cleared in two cases: freeing the last non-tree |
5915 | * reference to the extent_buffer when its STALE bit is set or | |
f913cff3 | 5916 | * calling release_folio when the tree reference is the only reference. |
0b32f4bb | 5917 | * |
6bf9cd2e | 5918 | * In both cases, care is taken to ensure that the extent_buffer's |
f913cff3 | 5919 | * pages are not under io. However, release_folio can be concurrently |
6bf9cd2e BB |
5920 | * called with creating new references, which is prone to race |
5921 | * conditions between the calls to check_buffer_tree_ref in those | |
5922 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 5923 | * |
6bf9cd2e BB |
5924 | * The actual lifetime of the extent_buffer in the radix tree is |
5925 | * adequately protected by the refcount, but the TREE_REF bit and | |
5926 | * its corresponding reference are not. To protect against this | |
5927 | * class of races, we call check_buffer_tree_ref from the codepaths | |
5928 | * which trigger io after they set eb->io_pages. Note that once io is | |
5929 | * initiated, TREE_REF can no longer be cleared, so that is the | |
5930 | * moment at which any such race is best fixed. | |
0b32f4bb | 5931 | */ |
242e18c7 CM |
5932 | refs = atomic_read(&eb->refs); |
5933 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5934 | return; | |
5935 | ||
594831c4 JB |
5936 | spin_lock(&eb->refs_lock); |
5937 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 5938 | atomic_inc(&eb->refs); |
594831c4 | 5939 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
5940 | } |
5941 | ||
2457aec6 MG |
5942 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
5943 | struct page *accessed) | |
5df4235e | 5944 | { |
cc5e31a4 | 5945 | int num_pages, i; |
5df4235e | 5946 | |
0b32f4bb JB |
5947 | check_buffer_tree_ref(eb); |
5948 | ||
65ad0104 | 5949 | num_pages = num_extent_pages(eb); |
5df4235e | 5950 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
5951 | struct page *p = eb->pages[i]; |
5952 | ||
2457aec6 MG |
5953 | if (p != accessed) |
5954 | mark_page_accessed(p); | |
5df4235e JB |
5955 | } |
5956 | } | |
5957 | ||
f28491e0 JB |
5958 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
5959 | u64 start) | |
452c75c3 CS |
5960 | { |
5961 | struct extent_buffer *eb; | |
5962 | ||
2f3186d8 QW |
5963 | eb = find_extent_buffer_nolock(fs_info, start); |
5964 | if (!eb) | |
5965 | return NULL; | |
5966 | /* | |
5967 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
5968 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
5969 | * another task running free_extent_buffer() might have seen that flag | |
5970 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
5971 | * writeback flags not set) and it's still in the tree (flag | |
5972 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
5973 | * decrementing the extent buffer's reference count twice. So here we | |
5974 | * could race and increment the eb's reference count, clear its stale | |
5975 | * flag, mark it as dirty and drop our reference before the other task | |
5976 | * finishes executing free_extent_buffer, which would later result in | |
5977 | * an attempt to free an extent buffer that is dirty. | |
5978 | */ | |
5979 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
5980 | spin_lock(&eb->refs_lock); | |
5981 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 5982 | } |
2f3186d8 QW |
5983 | mark_extent_buffer_accessed(eb, NULL); |
5984 | return eb; | |
452c75c3 CS |
5985 | } |
5986 | ||
faa2dbf0 JB |
5987 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5988 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 5989 | u64 start) |
faa2dbf0 JB |
5990 | { |
5991 | struct extent_buffer *eb, *exists = NULL; | |
5992 | int ret; | |
5993 | ||
5994 | eb = find_extent_buffer(fs_info, start); | |
5995 | if (eb) | |
5996 | return eb; | |
da17066c | 5997 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 5998 | if (!eb) |
b6293c82 | 5999 | return ERR_PTR(-ENOMEM); |
faa2dbf0 | 6000 | eb->fs_info = fs_info; |
01cd3909 DS |
6001 | again: |
6002 | ret = radix_tree_preload(GFP_NOFS); | |
6003 | if (ret) { | |
6004 | exists = ERR_PTR(ret); | |
6005 | goto free_eb; | |
6006 | } | |
6007 | spin_lock(&fs_info->buffer_lock); | |
6008 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
6009 | start >> fs_info->sectorsize_bits, eb); | |
6010 | spin_unlock(&fs_info->buffer_lock); | |
6011 | radix_tree_preload_end(); | |
6012 | if (ret == -EEXIST) { | |
6013 | exists = find_extent_buffer(fs_info, start); | |
6014 | if (exists) | |
faa2dbf0 | 6015 | goto free_eb; |
01cd3909 DS |
6016 | else |
6017 | goto again; | |
6018 | } | |
faa2dbf0 JB |
6019 | check_buffer_tree_ref(eb); |
6020 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
6021 | ||
faa2dbf0 JB |
6022 | return eb; |
6023 | free_eb: | |
6024 | btrfs_release_extent_buffer(eb); | |
6025 | return exists; | |
6026 | } | |
6027 | #endif | |
6028 | ||
81982210 QW |
6029 | static struct extent_buffer *grab_extent_buffer( |
6030 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
6031 | { |
6032 | struct extent_buffer *exists; | |
6033 | ||
81982210 QW |
6034 | /* |
6035 | * For subpage case, we completely rely on radix tree to ensure we | |
6036 | * don't try to insert two ebs for the same bytenr. So here we always | |
6037 | * return NULL and just continue. | |
6038 | */ | |
fbca46eb | 6039 | if (fs_info->nodesize < PAGE_SIZE) |
81982210 QW |
6040 | return NULL; |
6041 | ||
c0f0a9e7 QW |
6042 | /* Page not yet attached to an extent buffer */ |
6043 | if (!PagePrivate(page)) | |
6044 | return NULL; | |
6045 | ||
6046 | /* | |
6047 | * We could have already allocated an eb for this page and attached one | |
6048 | * so lets see if we can get a ref on the existing eb, and if we can we | |
6049 | * know it's good and we can just return that one, else we know we can | |
6050 | * just overwrite page->private. | |
6051 | */ | |
6052 | exists = (struct extent_buffer *)page->private; | |
6053 | if (atomic_inc_not_zero(&exists->refs)) | |
6054 | return exists; | |
6055 | ||
6056 | WARN_ON(PageDirty(page)); | |
6057 | detach_page_private(page); | |
6058 | return NULL; | |
6059 | } | |
6060 | ||
fbca46eb QW |
6061 | static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start) |
6062 | { | |
6063 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { | |
6064 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
6065 | return -EINVAL; | |
6066 | } | |
6067 | ||
6068 | if (fs_info->nodesize < PAGE_SIZE && | |
6069 | offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) { | |
6070 | btrfs_err(fs_info, | |
6071 | "tree block crosses page boundary, start %llu nodesize %u", | |
6072 | start, fs_info->nodesize); | |
6073 | return -EINVAL; | |
6074 | } | |
6075 | if (fs_info->nodesize >= PAGE_SIZE && | |
1280d2d1 | 6076 | !PAGE_ALIGNED(start)) { |
fbca46eb QW |
6077 | btrfs_err(fs_info, |
6078 | "tree block is not page aligned, start %llu nodesize %u", | |
6079 | start, fs_info->nodesize); | |
6080 | return -EINVAL; | |
6081 | } | |
6082 | return 0; | |
6083 | } | |
6084 | ||
f28491e0 | 6085 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 6086 | u64 start, u64 owner_root, int level) |
d1310b2e | 6087 | { |
da17066c | 6088 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
6089 | int num_pages; |
6090 | int i; | |
09cbfeaf | 6091 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 6092 | struct extent_buffer *eb; |
6af118ce | 6093 | struct extent_buffer *exists = NULL; |
d1310b2e | 6094 | struct page *p; |
f28491e0 | 6095 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
b40130b2 | 6096 | u64 lockdep_owner = owner_root; |
d1310b2e | 6097 | int uptodate = 1; |
19fe0a8b | 6098 | int ret; |
d1310b2e | 6099 | |
fbca46eb | 6100 | if (check_eb_alignment(fs_info, start)) |
c871b0f2 | 6101 | return ERR_PTR(-EINVAL); |
c871b0f2 | 6102 | |
e9306ad4 QW |
6103 | #if BITS_PER_LONG == 32 |
6104 | if (start >= MAX_LFS_FILESIZE) { | |
6105 | btrfs_err_rl(fs_info, | |
6106 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
6107 | btrfs_err_32bit_limit(fs_info); | |
6108 | return ERR_PTR(-EOVERFLOW); | |
6109 | } | |
6110 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6111 | btrfs_warn_32bit_limit(fs_info); | |
6112 | #endif | |
6113 | ||
f28491e0 | 6114 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 6115 | if (eb) |
6af118ce | 6116 | return eb; |
6af118ce | 6117 | |
23d79d81 | 6118 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 6119 | if (!eb) |
c871b0f2 | 6120 | return ERR_PTR(-ENOMEM); |
b40130b2 JB |
6121 | |
6122 | /* | |
6123 | * The reloc trees are just snapshots, so we need them to appear to be | |
6124 | * just like any other fs tree WRT lockdep. | |
6125 | */ | |
6126 | if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID) | |
6127 | lockdep_owner = BTRFS_FS_TREE_OBJECTID; | |
6128 | ||
6129 | btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level); | |
d1310b2e | 6130 | |
65ad0104 | 6131 | num_pages = num_extent_pages(eb); |
727011e0 | 6132 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
6133 | struct btrfs_subpage *prealloc = NULL; |
6134 | ||
d1b5c567 | 6135 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
6136 | if (!p) { |
6137 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 6138 | goto free_eb; |
c871b0f2 | 6139 | } |
4f2de97a | 6140 | |
760f991f QW |
6141 | /* |
6142 | * Preallocate page->private for subpage case, so that we won't | |
6143 | * allocate memory with private_lock hold. The memory will be | |
6144 | * freed by attach_extent_buffer_page() or freed manually if | |
6145 | * we exit earlier. | |
6146 | * | |
6147 | * Although we have ensured one subpage eb can only have one | |
6148 | * page, but it may change in the future for 16K page size | |
6149 | * support, so we still preallocate the memory in the loop. | |
6150 | */ | |
fbca46eb | 6151 | if (fs_info->nodesize < PAGE_SIZE) { |
651fb419 QW |
6152 | prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA); |
6153 | if (IS_ERR(prealloc)) { | |
6154 | ret = PTR_ERR(prealloc); | |
fdf250db QW |
6155 | unlock_page(p); |
6156 | put_page(p); | |
6157 | exists = ERR_PTR(ret); | |
6158 | goto free_eb; | |
6159 | } | |
760f991f QW |
6160 | } |
6161 | ||
4f2de97a | 6162 | spin_lock(&mapping->private_lock); |
81982210 | 6163 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
6164 | if (exists) { |
6165 | spin_unlock(&mapping->private_lock); | |
6166 | unlock_page(p); | |
6167 | put_page(p); | |
6168 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 6169 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 6170 | goto free_eb; |
d1310b2e | 6171 | } |
760f991f QW |
6172 | /* Should not fail, as we have preallocated the memory */ |
6173 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
6174 | ASSERT(!ret); | |
8ff8466d QW |
6175 | /* |
6176 | * To inform we have extra eb under allocation, so that | |
6177 | * detach_extent_buffer_page() won't release the page private | |
6178 | * when the eb hasn't yet been inserted into radix tree. | |
6179 | * | |
6180 | * The ref will be decreased when the eb released the page, in | |
6181 | * detach_extent_buffer_page(). | |
6182 | * Thus needs no special handling in error path. | |
6183 | */ | |
6184 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 6185 | spin_unlock(&mapping->private_lock); |
760f991f | 6186 | |
1e5eb3d6 | 6187 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 6188 | eb->pages[i] = p; |
d1310b2e CM |
6189 | if (!PageUptodate(p)) |
6190 | uptodate = 0; | |
eb14ab8e CM |
6191 | |
6192 | /* | |
b16d011e NB |
6193 | * We can't unlock the pages just yet since the extent buffer |
6194 | * hasn't been properly inserted in the radix tree, this | |
f913cff3 | 6195 | * opens a race with btree_release_folio which can free a page |
b16d011e NB |
6196 | * while we are still filling in all pages for the buffer and |
6197 | * we could crash. | |
eb14ab8e | 6198 | */ |
d1310b2e CM |
6199 | } |
6200 | if (uptodate) | |
b4ce94de | 6201 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
01cd3909 DS |
6202 | again: |
6203 | ret = radix_tree_preload(GFP_NOFS); | |
6204 | if (ret) { | |
6205 | exists = ERR_PTR(ret); | |
6206 | goto free_eb; | |
6207 | } | |
6208 | ||
6209 | spin_lock(&fs_info->buffer_lock); | |
6210 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
6211 | start >> fs_info->sectorsize_bits, eb); | |
6212 | spin_unlock(&fs_info->buffer_lock); | |
6213 | radix_tree_preload_end(); | |
6214 | if (ret == -EEXIST) { | |
6215 | exists = find_extent_buffer(fs_info, start); | |
6216 | if (exists) | |
452c75c3 | 6217 | goto free_eb; |
01cd3909 DS |
6218 | else |
6219 | goto again; | |
6220 | } | |
6af118ce | 6221 | /* add one reference for the tree */ |
0b32f4bb | 6222 | check_buffer_tree_ref(eb); |
34b41ace | 6223 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
6224 | |
6225 | /* | |
b16d011e | 6226 | * Now it's safe to unlock the pages because any calls to |
f913cff3 | 6227 | * btree_release_folio will correctly detect that a page belongs to a |
b16d011e | 6228 | * live buffer and won't free them prematurely. |
eb14ab8e | 6229 | */ |
28187ae5 NB |
6230 | for (i = 0; i < num_pages; i++) |
6231 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
6232 | return eb; |
6233 | ||
6af118ce | 6234 | free_eb: |
5ca64f45 | 6235 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
6236 | for (i = 0; i < num_pages; i++) { |
6237 | if (eb->pages[i]) | |
6238 | unlock_page(eb->pages[i]); | |
6239 | } | |
eb14ab8e | 6240 | |
897ca6e9 | 6241 | btrfs_release_extent_buffer(eb); |
6af118ce | 6242 | return exists; |
d1310b2e | 6243 | } |
d1310b2e | 6244 | |
3083ee2e JB |
6245 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
6246 | { | |
6247 | struct extent_buffer *eb = | |
6248 | container_of(head, struct extent_buffer, rcu_head); | |
6249 | ||
6250 | __free_extent_buffer(eb); | |
6251 | } | |
6252 | ||
f7a52a40 | 6253 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 6254 | __releases(&eb->refs_lock) |
3083ee2e | 6255 | { |
07e21c4d NB |
6256 | lockdep_assert_held(&eb->refs_lock); |
6257 | ||
3083ee2e JB |
6258 | WARN_ON(atomic_read(&eb->refs) == 0); |
6259 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 6260 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 6261 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 6262 | |
815a51c7 | 6263 | spin_unlock(&eb->refs_lock); |
3083ee2e | 6264 | |
01cd3909 DS |
6265 | spin_lock(&fs_info->buffer_lock); |
6266 | radix_tree_delete(&fs_info->buffer_radix, | |
6267 | eb->start >> fs_info->sectorsize_bits); | |
6268 | spin_unlock(&fs_info->buffer_lock); | |
34b41ace JB |
6269 | } else { |
6270 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 6271 | } |
3083ee2e | 6272 | |
8c38938c | 6273 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 6274 | /* Should be safe to release our pages at this point */ |
55ac0139 | 6275 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 6276 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 6277 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
6278 | __free_extent_buffer(eb); |
6279 | return 1; | |
6280 | } | |
6281 | #endif | |
3083ee2e | 6282 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 6283 | return 1; |
3083ee2e JB |
6284 | } |
6285 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
6286 | |
6287 | return 0; | |
3083ee2e JB |
6288 | } |
6289 | ||
d1310b2e CM |
6290 | void free_extent_buffer(struct extent_buffer *eb) |
6291 | { | |
242e18c7 | 6292 | int refs; |
d1310b2e CM |
6293 | if (!eb) |
6294 | return; | |
6295 | ||
e5677f05 | 6296 | refs = atomic_read(&eb->refs); |
242e18c7 | 6297 | while (1) { |
46cc775e NB |
6298 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
6299 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
6300 | refs == 1)) | |
242e18c7 | 6301 | break; |
e5677f05 | 6302 | if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1)) |
242e18c7 CM |
6303 | return; |
6304 | } | |
6305 | ||
3083ee2e JB |
6306 | spin_lock(&eb->refs_lock); |
6307 | if (atomic_read(&eb->refs) == 2 && | |
6308 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 6309 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
6310 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6311 | atomic_dec(&eb->refs); | |
6312 | ||
6313 | /* | |
6314 | * I know this is terrible, but it's temporary until we stop tracking | |
6315 | * the uptodate bits and such for the extent buffers. | |
6316 | */ | |
f7a52a40 | 6317 | release_extent_buffer(eb); |
3083ee2e JB |
6318 | } |
6319 | ||
6320 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
6321 | { | |
6322 | if (!eb) | |
d1310b2e CM |
6323 | return; |
6324 | ||
3083ee2e JB |
6325 | spin_lock(&eb->refs_lock); |
6326 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
6327 | ||
0b32f4bb | 6328 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
6329 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6330 | atomic_dec(&eb->refs); | |
f7a52a40 | 6331 | release_extent_buffer(eb); |
d1310b2e | 6332 | } |
d1310b2e | 6333 | |
0d27797e QW |
6334 | static void btree_clear_page_dirty(struct page *page) |
6335 | { | |
6336 | ASSERT(PageDirty(page)); | |
6337 | ASSERT(PageLocked(page)); | |
6338 | clear_page_dirty_for_io(page); | |
6339 | xa_lock_irq(&page->mapping->i_pages); | |
6340 | if (!PageDirty(page)) | |
6341 | __xa_clear_mark(&page->mapping->i_pages, | |
6342 | page_index(page), PAGECACHE_TAG_DIRTY); | |
6343 | xa_unlock_irq(&page->mapping->i_pages); | |
6344 | } | |
6345 | ||
6346 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
6347 | { | |
6348 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6349 | struct page *page = eb->pages[0]; | |
6350 | bool last; | |
6351 | ||
6352 | /* btree_clear_page_dirty() needs page locked */ | |
6353 | lock_page(page); | |
6354 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
6355 | eb->len); | |
6356 | if (last) | |
6357 | btree_clear_page_dirty(page); | |
6358 | unlock_page(page); | |
6359 | WARN_ON(atomic_read(&eb->refs) == 0); | |
6360 | } | |
6361 | ||
2b48966a | 6362 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 6363 | { |
cc5e31a4 DS |
6364 | int i; |
6365 | int num_pages; | |
d1310b2e CM |
6366 | struct page *page; |
6367 | ||
fbca46eb | 6368 | if (eb->fs_info->nodesize < PAGE_SIZE) |
0d27797e QW |
6369 | return clear_subpage_extent_buffer_dirty(eb); |
6370 | ||
65ad0104 | 6371 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
6372 | |
6373 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 6374 | page = eb->pages[i]; |
b9473439 | 6375 | if (!PageDirty(page)) |
d2c3f4f6 | 6376 | continue; |
a61e6f29 | 6377 | lock_page(page); |
0d27797e | 6378 | btree_clear_page_dirty(page); |
bf0da8c1 | 6379 | ClearPageError(page); |
a61e6f29 | 6380 | unlock_page(page); |
d1310b2e | 6381 | } |
0b32f4bb | 6382 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 6383 | } |
d1310b2e | 6384 | |
abb57ef3 | 6385 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 6386 | { |
cc5e31a4 DS |
6387 | int i; |
6388 | int num_pages; | |
abb57ef3 | 6389 | bool was_dirty; |
d1310b2e | 6390 | |
0b32f4bb JB |
6391 | check_buffer_tree_ref(eb); |
6392 | ||
b9473439 | 6393 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 6394 | |
65ad0104 | 6395 | num_pages = num_extent_pages(eb); |
3083ee2e | 6396 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
6397 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
6398 | ||
0d27797e | 6399 | if (!was_dirty) { |
fbca46eb | 6400 | bool subpage = eb->fs_info->nodesize < PAGE_SIZE; |
51995c39 | 6401 | |
0d27797e QW |
6402 | /* |
6403 | * For subpage case, we can have other extent buffers in the | |
6404 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
6405 | * have to clear page dirty without subpage lock held. | |
6406 | * This can cause race where our page gets dirty cleared after | |
6407 | * we just set it. | |
6408 | * | |
6409 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
6410 | * its page for other reasons, we can use page lock to prevent | |
6411 | * the above race. | |
6412 | */ | |
6413 | if (subpage) | |
6414 | lock_page(eb->pages[0]); | |
6415 | for (i = 0; i < num_pages; i++) | |
6416 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
6417 | eb->start, eb->len); | |
6418 | if (subpage) | |
6419 | unlock_page(eb->pages[0]); | |
6420 | } | |
51995c39 LB |
6421 | #ifdef CONFIG_BTRFS_DEBUG |
6422 | for (i = 0; i < num_pages; i++) | |
6423 | ASSERT(PageDirty(eb->pages[i])); | |
6424 | #endif | |
6425 | ||
b9473439 | 6426 | return was_dirty; |
d1310b2e | 6427 | } |
d1310b2e | 6428 | |
69ba3927 | 6429 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 6430 | { |
251f2acc | 6431 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 6432 | struct page *page; |
cc5e31a4 | 6433 | int num_pages; |
251f2acc | 6434 | int i; |
1259ab75 | 6435 | |
b4ce94de | 6436 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6437 | num_pages = num_extent_pages(eb); |
1259ab75 | 6438 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6439 | page = eb->pages[i]; |
fbca46eb QW |
6440 | if (!page) |
6441 | continue; | |
6442 | ||
6443 | /* | |
6444 | * This is special handling for metadata subpage, as regular | |
6445 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6446 | */ | |
6447 | if (fs_info->nodesize >= PAGE_SIZE) | |
6448 | ClearPageUptodate(page); | |
6449 | else | |
6450 | btrfs_subpage_clear_uptodate(fs_info, page, eb->start, | |
6451 | eb->len); | |
1259ab75 | 6452 | } |
1259ab75 CM |
6453 | } |
6454 | ||
09c25a8c | 6455 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 6456 | { |
251f2acc | 6457 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 6458 | struct page *page; |
cc5e31a4 | 6459 | int num_pages; |
251f2acc | 6460 | int i; |
d1310b2e | 6461 | |
0b32f4bb | 6462 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6463 | num_pages = num_extent_pages(eb); |
d1310b2e | 6464 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6465 | page = eb->pages[i]; |
fbca46eb QW |
6466 | |
6467 | /* | |
6468 | * This is special handling for metadata subpage, as regular | |
6469 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6470 | */ | |
6471 | if (fs_info->nodesize >= PAGE_SIZE) | |
6472 | SetPageUptodate(page); | |
6473 | else | |
6474 | btrfs_subpage_set_uptodate(fs_info, page, eb->start, | |
6475 | eb->len); | |
d1310b2e | 6476 | } |
d1310b2e | 6477 | } |
d1310b2e | 6478 | |
4012daf7 QW |
6479 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
6480 | int mirror_num) | |
6481 | { | |
6482 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6483 | struct extent_io_tree *io_tree; | |
6484 | struct page *page = eb->pages[0]; | |
722c82ac CH |
6485 | struct btrfs_bio_ctrl bio_ctrl = { |
6486 | .mirror_num = mirror_num, | |
6487 | }; | |
4012daf7 QW |
6488 | int ret = 0; |
6489 | ||
6490 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
6491 | ASSERT(PagePrivate(page)); | |
6492 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
6493 | ||
6494 | if (wait == WAIT_NONE) { | |
dc56219f GR |
6495 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1)) |
6496 | return -EAGAIN; | |
4012daf7 QW |
6497 | } else { |
6498 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6499 | if (ret < 0) | |
6500 | return ret; | |
6501 | } | |
6502 | ||
6503 | ret = 0; | |
6504 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
6505 | PageUptodate(page) || | |
6506 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
6507 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
6508 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6509 | return ret; | |
6510 | } | |
6511 | ||
6512 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
6513 | eb->read_mirror = 0; | |
6514 | atomic_set(&eb->io_pages, 1); | |
6515 | check_buffer_tree_ref(eb); | |
6516 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); | |
6517 | ||
3d078efa | 6518 | btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len); |
08a6f464 | 6519 | ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl, |
390ed29b QW |
6520 | page, eb->start, eb->len, |
6521 | eb->start - page_offset(page), | |
722c82ac | 6522 | end_bio_extent_readpage, 0, true); |
4012daf7 QW |
6523 | if (ret) { |
6524 | /* | |
6525 | * In the endio function, if we hit something wrong we will | |
6526 | * increase the io_pages, so here we need to decrease it for | |
6527 | * error path. | |
6528 | */ | |
6529 | atomic_dec(&eb->io_pages); | |
6530 | } | |
722c82ac | 6531 | submit_one_bio(&bio_ctrl); |
4012daf7 QW |
6532 | if (ret || wait != WAIT_COMPLETE) |
6533 | return ret; | |
6534 | ||
6535 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED); | |
6536 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
6537 | ret = -EIO; | |
6538 | return ret; | |
6539 | } | |
6540 | ||
c2ccfbc6 | 6541 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 6542 | { |
cc5e31a4 | 6543 | int i; |
d1310b2e CM |
6544 | struct page *page; |
6545 | int err; | |
6546 | int ret = 0; | |
ce9adaa5 CM |
6547 | int locked_pages = 0; |
6548 | int all_uptodate = 1; | |
cc5e31a4 | 6549 | int num_pages; |
727011e0 | 6550 | unsigned long num_reads = 0; |
722c82ac CH |
6551 | struct btrfs_bio_ctrl bio_ctrl = { |
6552 | .mirror_num = mirror_num, | |
6553 | }; | |
a86c12c7 | 6554 | |
b4ce94de | 6555 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
6556 | return 0; |
6557 | ||
651740a5 JB |
6558 | /* |
6559 | * We could have had EXTENT_BUFFER_UPTODATE cleared by the write | |
6560 | * operation, which could potentially still be in flight. In this case | |
6561 | * we simply want to return an error. | |
6562 | */ | |
6563 | if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))) | |
6564 | return -EIO; | |
6565 | ||
fbca46eb | 6566 | if (eb->fs_info->nodesize < PAGE_SIZE) |
4012daf7 QW |
6567 | return read_extent_buffer_subpage(eb, wait, mirror_num); |
6568 | ||
65ad0104 | 6569 | num_pages = num_extent_pages(eb); |
8436ea91 | 6570 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6571 | page = eb->pages[i]; |
bb82ab88 | 6572 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
6573 | /* |
6574 | * WAIT_NONE is only utilized by readahead. If we can't | |
6575 | * acquire the lock atomically it means either the eb | |
6576 | * is being read out or under modification. | |
6577 | * Either way the eb will be or has been cached, | |
6578 | * readahead can exit safely. | |
6579 | */ | |
2db04966 | 6580 | if (!trylock_page(page)) |
ce9adaa5 | 6581 | goto unlock_exit; |
d1310b2e CM |
6582 | } else { |
6583 | lock_page(page); | |
6584 | } | |
ce9adaa5 | 6585 | locked_pages++; |
2571e739 LB |
6586 | } |
6587 | /* | |
6588 | * We need to firstly lock all pages to make sure that | |
6589 | * the uptodate bit of our pages won't be affected by | |
6590 | * clear_extent_buffer_uptodate(). | |
6591 | */ | |
8436ea91 | 6592 | for (i = 0; i < num_pages; i++) { |
2571e739 | 6593 | page = eb->pages[i]; |
727011e0 CM |
6594 | if (!PageUptodate(page)) { |
6595 | num_reads++; | |
ce9adaa5 | 6596 | all_uptodate = 0; |
727011e0 | 6597 | } |
ce9adaa5 | 6598 | } |
2571e739 | 6599 | |
ce9adaa5 | 6600 | if (all_uptodate) { |
8436ea91 | 6601 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
6602 | goto unlock_exit; |
6603 | } | |
6604 | ||
656f30db | 6605 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 6606 | eb->read_mirror = 0; |
0b32f4bb | 6607 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e | 6608 | /* |
f913cff3 | 6609 | * It is possible for release_folio to clear the TREE_REF bit before we |
6bf9cd2e BB |
6610 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. |
6611 | */ | |
6612 | check_buffer_tree_ref(eb); | |
8436ea91 | 6613 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6614 | page = eb->pages[i]; |
baf863b9 | 6615 | |
ce9adaa5 | 6616 | if (!PageUptodate(page)) { |
baf863b9 LB |
6617 | if (ret) { |
6618 | atomic_dec(&eb->io_pages); | |
6619 | unlock_page(page); | |
6620 | continue; | |
6621 | } | |
6622 | ||
f188591e | 6623 | ClearPageError(page); |
08a6f464 | 6624 | err = submit_extent_page(REQ_OP_READ, NULL, |
390ed29b QW |
6625 | &bio_ctrl, page, page_offset(page), |
6626 | PAGE_SIZE, 0, end_bio_extent_readpage, | |
722c82ac | 6627 | 0, false); |
baf863b9 | 6628 | if (err) { |
baf863b9 | 6629 | /* |
0420177c NB |
6630 | * We failed to submit the bio so it's the |
6631 | * caller's responsibility to perform cleanup | |
6632 | * i.e unlock page/set error bit. | |
baf863b9 | 6633 | */ |
0420177c NB |
6634 | ret = err; |
6635 | SetPageError(page); | |
6636 | unlock_page(page); | |
baf863b9 LB |
6637 | atomic_dec(&eb->io_pages); |
6638 | } | |
d1310b2e CM |
6639 | } else { |
6640 | unlock_page(page); | |
6641 | } | |
6642 | } | |
6643 | ||
722c82ac | 6644 | submit_one_bio(&bio_ctrl); |
a86c12c7 | 6645 | |
bb82ab88 | 6646 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 6647 | return ret; |
d397712b | 6648 | |
8436ea91 | 6649 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6650 | page = eb->pages[i]; |
d1310b2e | 6651 | wait_on_page_locked(page); |
d397712b | 6652 | if (!PageUptodate(page)) |
d1310b2e | 6653 | ret = -EIO; |
d1310b2e | 6654 | } |
d397712b | 6655 | |
d1310b2e | 6656 | return ret; |
ce9adaa5 CM |
6657 | |
6658 | unlock_exit: | |
d397712b | 6659 | while (locked_pages > 0) { |
ce9adaa5 | 6660 | locked_pages--; |
8436ea91 JB |
6661 | page = eb->pages[locked_pages]; |
6662 | unlock_page(page); | |
ce9adaa5 CM |
6663 | } |
6664 | return ret; | |
d1310b2e | 6665 | } |
d1310b2e | 6666 | |
f98b6215 QW |
6667 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
6668 | unsigned long len) | |
6669 | { | |
6670 | btrfs_warn(eb->fs_info, | |
6671 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
6672 | eb->start, eb->len, start, len); | |
6673 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6674 | ||
6675 | return true; | |
6676 | } | |
6677 | ||
6678 | /* | |
6679 | * Check if the [start, start + len) range is valid before reading/writing | |
6680 | * the eb. | |
6681 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
6682 | * | |
6683 | * Caller should not touch the dst/src memory if this function returns error. | |
6684 | */ | |
6685 | static inline int check_eb_range(const struct extent_buffer *eb, | |
6686 | unsigned long start, unsigned long len) | |
6687 | { | |
6688 | unsigned long offset; | |
6689 | ||
6690 | /* start, start + len should not go beyond eb->len nor overflow */ | |
6691 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
6692 | return report_eb_range(eb, start, len); | |
6693 | ||
6694 | return false; | |
6695 | } | |
6696 | ||
1cbb1f45 JM |
6697 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
6698 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6699 | { |
6700 | size_t cur; | |
6701 | size_t offset; | |
6702 | struct page *page; | |
6703 | char *kaddr; | |
6704 | char *dst = (char *)dstv; | |
884b07d0 | 6705 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6706 | |
f98b6215 | 6707 | if (check_eb_range(eb, start, len)) |
f716abd5 | 6708 | return; |
d1310b2e | 6709 | |
884b07d0 | 6710 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6711 | |
d397712b | 6712 | while (len > 0) { |
fb85fc9a | 6713 | page = eb->pages[i]; |
d1310b2e | 6714 | |
09cbfeaf | 6715 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 6716 | kaddr = page_address(page); |
d1310b2e | 6717 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
6718 | |
6719 | dst += cur; | |
6720 | len -= cur; | |
6721 | offset = 0; | |
6722 | i++; | |
6723 | } | |
6724 | } | |
d1310b2e | 6725 | |
a48b73ec JB |
6726 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
6727 | void __user *dstv, | |
6728 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
6729 | { |
6730 | size_t cur; | |
6731 | size_t offset; | |
6732 | struct page *page; | |
6733 | char *kaddr; | |
6734 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 6735 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
6736 | int ret = 0; |
6737 | ||
6738 | WARN_ON(start > eb->len); | |
6739 | WARN_ON(start + len > eb->start + eb->len); | |
6740 | ||
884b07d0 | 6741 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
6742 | |
6743 | while (len > 0) { | |
fb85fc9a | 6744 | page = eb->pages[i]; |
550ac1d8 | 6745 | |
09cbfeaf | 6746 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 6747 | kaddr = page_address(page); |
a48b73ec | 6748 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
6749 | ret = -EFAULT; |
6750 | break; | |
6751 | } | |
6752 | ||
6753 | dst += cur; | |
6754 | len -= cur; | |
6755 | offset = 0; | |
6756 | i++; | |
6757 | } | |
6758 | ||
6759 | return ret; | |
6760 | } | |
6761 | ||
1cbb1f45 JM |
6762 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
6763 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6764 | { |
6765 | size_t cur; | |
6766 | size_t offset; | |
6767 | struct page *page; | |
6768 | char *kaddr; | |
6769 | char *ptr = (char *)ptrv; | |
884b07d0 | 6770 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
6771 | int ret = 0; |
6772 | ||
f98b6215 QW |
6773 | if (check_eb_range(eb, start, len)) |
6774 | return -EINVAL; | |
d1310b2e | 6775 | |
884b07d0 | 6776 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6777 | |
d397712b | 6778 | while (len > 0) { |
fb85fc9a | 6779 | page = eb->pages[i]; |
d1310b2e | 6780 | |
09cbfeaf | 6781 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 6782 | |
a6591715 | 6783 | kaddr = page_address(page); |
d1310b2e | 6784 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
6785 | if (ret) |
6786 | break; | |
6787 | ||
6788 | ptr += cur; | |
6789 | len -= cur; | |
6790 | offset = 0; | |
6791 | i++; | |
6792 | } | |
6793 | return ret; | |
6794 | } | |
d1310b2e | 6795 | |
b8f95771 QW |
6796 | /* |
6797 | * Check that the extent buffer is uptodate. | |
6798 | * | |
6799 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
6800 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
6801 | */ | |
6802 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
6803 | struct page *page) | |
6804 | { | |
6805 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6806 | ||
a50e1fcb JB |
6807 | /* |
6808 | * If we are using the commit root we could potentially clear a page | |
6809 | * Uptodate while we're using the extent buffer that we've previously | |
6810 | * looked up. We don't want to complain in this case, as the page was | |
6811 | * valid before, we just didn't write it out. Instead we want to catch | |
6812 | * the case where we didn't actually read the block properly, which | |
6813 | * would have !PageUptodate && !PageError, as we clear PageError before | |
6814 | * reading. | |
6815 | */ | |
fbca46eb | 6816 | if (fs_info->nodesize < PAGE_SIZE) { |
a50e1fcb | 6817 | bool uptodate, error; |
b8f95771 QW |
6818 | |
6819 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
6820 | eb->start, eb->len); | |
a50e1fcb JB |
6821 | error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len); |
6822 | WARN_ON(!uptodate && !error); | |
b8f95771 | 6823 | } else { |
a50e1fcb | 6824 | WARN_ON(!PageUptodate(page) && !PageError(page)); |
b8f95771 QW |
6825 | } |
6826 | } | |
6827 | ||
2b48966a | 6828 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
6829 | const void *srcv) |
6830 | { | |
6831 | char *kaddr; | |
6832 | ||
b8f95771 | 6833 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6834 | kaddr = page_address(eb->pages[0]) + |
6835 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
6836 | chunk_tree_uuid)); | |
6837 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6838 | } |
6839 | ||
2b48966a | 6840 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
6841 | { |
6842 | char *kaddr; | |
6843 | ||
b8f95771 | 6844 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6845 | kaddr = page_address(eb->pages[0]) + |
6846 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
6847 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6848 | } |
6849 | ||
2b48966a | 6850 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
6851 | unsigned long start, unsigned long len) |
6852 | { | |
6853 | size_t cur; | |
6854 | size_t offset; | |
6855 | struct page *page; | |
6856 | char *kaddr; | |
6857 | char *src = (char *)srcv; | |
884b07d0 | 6858 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6859 | |
d3575156 NA |
6860 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
6861 | ||
f98b6215 QW |
6862 | if (check_eb_range(eb, start, len)) |
6863 | return; | |
d1310b2e | 6864 | |
884b07d0 | 6865 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6866 | |
d397712b | 6867 | while (len > 0) { |
fb85fc9a | 6868 | page = eb->pages[i]; |
b8f95771 | 6869 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6870 | |
09cbfeaf | 6871 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6872 | kaddr = page_address(page); |
d1310b2e | 6873 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
6874 | |
6875 | src += cur; | |
6876 | len -= cur; | |
6877 | offset = 0; | |
6878 | i++; | |
6879 | } | |
6880 | } | |
d1310b2e | 6881 | |
2b48966a | 6882 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 6883 | unsigned long len) |
d1310b2e CM |
6884 | { |
6885 | size_t cur; | |
6886 | size_t offset; | |
6887 | struct page *page; | |
6888 | char *kaddr; | |
884b07d0 | 6889 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6890 | |
f98b6215 QW |
6891 | if (check_eb_range(eb, start, len)) |
6892 | return; | |
d1310b2e | 6893 | |
884b07d0 | 6894 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6895 | |
d397712b | 6896 | while (len > 0) { |
fb85fc9a | 6897 | page = eb->pages[i]; |
b8f95771 | 6898 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6899 | |
09cbfeaf | 6900 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6901 | kaddr = page_address(page); |
b159fa28 | 6902 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
6903 | |
6904 | len -= cur; | |
6905 | offset = 0; | |
6906 | i++; | |
6907 | } | |
6908 | } | |
d1310b2e | 6909 | |
2b48966a DS |
6910 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
6911 | const struct extent_buffer *src) | |
58e8012c DS |
6912 | { |
6913 | int i; | |
cc5e31a4 | 6914 | int num_pages; |
58e8012c DS |
6915 | |
6916 | ASSERT(dst->len == src->len); | |
6917 | ||
fbca46eb | 6918 | if (dst->fs_info->nodesize >= PAGE_SIZE) { |
884b07d0 QW |
6919 | num_pages = num_extent_pages(dst); |
6920 | for (i = 0; i < num_pages; i++) | |
6921 | copy_page(page_address(dst->pages[i]), | |
6922 | page_address(src->pages[i])); | |
6923 | } else { | |
6924 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
6925 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
6926 | ||
fbca46eb | 6927 | ASSERT(src->fs_info->nodesize < PAGE_SIZE); |
884b07d0 QW |
6928 | memcpy(page_address(dst->pages[0]) + dst_offset, |
6929 | page_address(src->pages[0]) + src_offset, | |
6930 | src->len); | |
6931 | } | |
58e8012c DS |
6932 | } |
6933 | ||
2b48966a DS |
6934 | void copy_extent_buffer(const struct extent_buffer *dst, |
6935 | const struct extent_buffer *src, | |
d1310b2e CM |
6936 | unsigned long dst_offset, unsigned long src_offset, |
6937 | unsigned long len) | |
6938 | { | |
6939 | u64 dst_len = dst->len; | |
6940 | size_t cur; | |
6941 | size_t offset; | |
6942 | struct page *page; | |
6943 | char *kaddr; | |
884b07d0 | 6944 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 6945 | |
f98b6215 QW |
6946 | if (check_eb_range(dst, dst_offset, len) || |
6947 | check_eb_range(src, src_offset, len)) | |
6948 | return; | |
6949 | ||
d1310b2e CM |
6950 | WARN_ON(src->len != dst_len); |
6951 | ||
884b07d0 | 6952 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 6953 | |
d397712b | 6954 | while (len > 0) { |
fb85fc9a | 6955 | page = dst->pages[i]; |
b8f95771 | 6956 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 6957 | |
09cbfeaf | 6958 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 6959 | |
a6591715 | 6960 | kaddr = page_address(page); |
d1310b2e | 6961 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
6962 | |
6963 | src_offset += cur; | |
6964 | len -= cur; | |
6965 | offset = 0; | |
6966 | i++; | |
6967 | } | |
6968 | } | |
d1310b2e | 6969 | |
3e1e8bb7 OS |
6970 | /* |
6971 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
6972 | * given bit number | |
6973 | * @eb: the extent buffer | |
6974 | * @start: offset of the bitmap item in the extent buffer | |
6975 | * @nr: bit number | |
6976 | * @page_index: return index of the page in the extent buffer that contains the | |
6977 | * given bit number | |
6978 | * @page_offset: return offset into the page given by page_index | |
6979 | * | |
6980 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
6981 | * contains a given bit. | |
6982 | */ | |
2b48966a | 6983 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
6984 | unsigned long start, unsigned long nr, |
6985 | unsigned long *page_index, | |
6986 | size_t *page_offset) | |
6987 | { | |
3e1e8bb7 OS |
6988 | size_t byte_offset = BIT_BYTE(nr); |
6989 | size_t offset; | |
6990 | ||
6991 | /* | |
6992 | * The byte we want is the offset of the extent buffer + the offset of | |
6993 | * the bitmap item in the extent buffer + the offset of the byte in the | |
6994 | * bitmap item. | |
6995 | */ | |
884b07d0 | 6996 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 6997 | |
09cbfeaf | 6998 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 6999 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
7000 | } |
7001 | ||
7002 | /** | |
7003 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
7004 | * @eb: the extent buffer | |
7005 | * @start: offset of the bitmap item in the extent buffer | |
7006 | * @nr: bit number to test | |
7007 | */ | |
2b48966a | 7008 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7009 | unsigned long nr) |
7010 | { | |
2fe1d551 | 7011 | u8 *kaddr; |
3e1e8bb7 OS |
7012 | struct page *page; |
7013 | unsigned long i; | |
7014 | size_t offset; | |
7015 | ||
7016 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
7017 | page = eb->pages[i]; | |
b8f95771 | 7018 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7019 | kaddr = page_address(page); |
7020 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
7021 | } | |
7022 | ||
7023 | /** | |
7024 | * extent_buffer_bitmap_set - set an area of a bitmap | |
7025 | * @eb: the extent buffer | |
7026 | * @start: offset of the bitmap item in the extent buffer | |
7027 | * @pos: bit number of the first bit | |
7028 | * @len: number of bits to set | |
7029 | */ | |
2b48966a | 7030 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7031 | unsigned long pos, unsigned long len) |
7032 | { | |
2fe1d551 | 7033 | u8 *kaddr; |
3e1e8bb7 OS |
7034 | struct page *page; |
7035 | unsigned long i; | |
7036 | size_t offset; | |
7037 | const unsigned int size = pos + len; | |
7038 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7039 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7040 | |
7041 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7042 | page = eb->pages[i]; | |
b8f95771 | 7043 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7044 | kaddr = page_address(page); |
7045 | ||
7046 | while (len >= bits_to_set) { | |
7047 | kaddr[offset] |= mask_to_set; | |
7048 | len -= bits_to_set; | |
7049 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 7050 | mask_to_set = ~0; |
09cbfeaf | 7051 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7052 | offset = 0; |
7053 | page = eb->pages[++i]; | |
b8f95771 | 7054 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7055 | kaddr = page_address(page); |
7056 | } | |
7057 | } | |
7058 | if (len) { | |
7059 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
7060 | kaddr[offset] |= mask_to_set; | |
7061 | } | |
7062 | } | |
7063 | ||
7064 | ||
7065 | /** | |
7066 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
7067 | * @eb: the extent buffer | |
7068 | * @start: offset of the bitmap item in the extent buffer | |
7069 | * @pos: bit number of the first bit | |
7070 | * @len: number of bits to clear | |
7071 | */ | |
2b48966a DS |
7072 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
7073 | unsigned long start, unsigned long pos, | |
7074 | unsigned long len) | |
3e1e8bb7 | 7075 | { |
2fe1d551 | 7076 | u8 *kaddr; |
3e1e8bb7 OS |
7077 | struct page *page; |
7078 | unsigned long i; | |
7079 | size_t offset; | |
7080 | const unsigned int size = pos + len; | |
7081 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7082 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7083 | |
7084 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7085 | page = eb->pages[i]; | |
b8f95771 | 7086 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7087 | kaddr = page_address(page); |
7088 | ||
7089 | while (len >= bits_to_clear) { | |
7090 | kaddr[offset] &= ~mask_to_clear; | |
7091 | len -= bits_to_clear; | |
7092 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 7093 | mask_to_clear = ~0; |
09cbfeaf | 7094 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7095 | offset = 0; |
7096 | page = eb->pages[++i]; | |
b8f95771 | 7097 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7098 | kaddr = page_address(page); |
7099 | } | |
7100 | } | |
7101 | if (len) { | |
7102 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
7103 | kaddr[offset] &= ~mask_to_clear; | |
7104 | } | |
7105 | } | |
7106 | ||
3387206f ST |
7107 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
7108 | { | |
7109 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
7110 | return distance < len; | |
7111 | } | |
7112 | ||
d1310b2e CM |
7113 | static void copy_pages(struct page *dst_page, struct page *src_page, |
7114 | unsigned long dst_off, unsigned long src_off, | |
7115 | unsigned long len) | |
7116 | { | |
a6591715 | 7117 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 7118 | char *src_kaddr; |
727011e0 | 7119 | int must_memmove = 0; |
d1310b2e | 7120 | |
3387206f | 7121 | if (dst_page != src_page) { |
a6591715 | 7122 | src_kaddr = page_address(src_page); |
3387206f | 7123 | } else { |
d1310b2e | 7124 | src_kaddr = dst_kaddr; |
727011e0 CM |
7125 | if (areas_overlap(src_off, dst_off, len)) |
7126 | must_memmove = 1; | |
3387206f | 7127 | } |
d1310b2e | 7128 | |
727011e0 CM |
7129 | if (must_memmove) |
7130 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
7131 | else | |
7132 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
7133 | } |
7134 | ||
2b48966a DS |
7135 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
7136 | unsigned long dst_offset, unsigned long src_offset, | |
7137 | unsigned long len) | |
d1310b2e CM |
7138 | { |
7139 | size_t cur; | |
7140 | size_t dst_off_in_page; | |
7141 | size_t src_off_in_page; | |
d1310b2e CM |
7142 | unsigned long dst_i; |
7143 | unsigned long src_i; | |
7144 | ||
f98b6215 QW |
7145 | if (check_eb_range(dst, dst_offset, len) || |
7146 | check_eb_range(dst, src_offset, len)) | |
7147 | return; | |
d1310b2e | 7148 | |
d397712b | 7149 | while (len > 0) { |
884b07d0 QW |
7150 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
7151 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 7152 | |
884b07d0 QW |
7153 | dst_i = get_eb_page_index(dst_offset); |
7154 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 7155 | |
09cbfeaf | 7156 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
7157 | src_off_in_page)); |
7158 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 7159 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 7160 | |
fb85fc9a | 7161 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7162 | dst_off_in_page, src_off_in_page, cur); |
7163 | ||
7164 | src_offset += cur; | |
7165 | dst_offset += cur; | |
7166 | len -= cur; | |
7167 | } | |
7168 | } | |
d1310b2e | 7169 | |
2b48966a DS |
7170 | void memmove_extent_buffer(const struct extent_buffer *dst, |
7171 | unsigned long dst_offset, unsigned long src_offset, | |
7172 | unsigned long len) | |
d1310b2e CM |
7173 | { |
7174 | size_t cur; | |
7175 | size_t dst_off_in_page; | |
7176 | size_t src_off_in_page; | |
7177 | unsigned long dst_end = dst_offset + len - 1; | |
7178 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
7179 | unsigned long dst_i; |
7180 | unsigned long src_i; | |
7181 | ||
f98b6215 QW |
7182 | if (check_eb_range(dst, dst_offset, len) || |
7183 | check_eb_range(dst, src_offset, len)) | |
7184 | return; | |
727011e0 | 7185 | if (dst_offset < src_offset) { |
d1310b2e CM |
7186 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
7187 | return; | |
7188 | } | |
d397712b | 7189 | while (len > 0) { |
884b07d0 QW |
7190 | dst_i = get_eb_page_index(dst_end); |
7191 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 7192 | |
884b07d0 QW |
7193 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
7194 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
7195 | |
7196 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
7197 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 7198 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7199 | dst_off_in_page - cur + 1, |
7200 | src_off_in_page - cur + 1, cur); | |
7201 | ||
7202 | dst_end -= cur; | |
7203 | src_end -= cur; | |
7204 | len -= cur; | |
7205 | } | |
7206 | } | |
6af118ce | 7207 | |
01cd3909 | 7208 | #define GANG_LOOKUP_SIZE 16 |
d1e86e3f QW |
7209 | static struct extent_buffer *get_next_extent_buffer( |
7210 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
7211 | { | |
01cd3909 DS |
7212 | struct extent_buffer *gang[GANG_LOOKUP_SIZE]; |
7213 | struct extent_buffer *found = NULL; | |
d1e86e3f | 7214 | u64 page_start = page_offset(page); |
01cd3909 | 7215 | u64 cur = page_start; |
d1e86e3f QW |
7216 | |
7217 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
d1e86e3f QW |
7218 | lockdep_assert_held(&fs_info->buffer_lock); |
7219 | ||
01cd3909 DS |
7220 | while (cur < page_start + PAGE_SIZE) { |
7221 | int ret; | |
7222 | int i; | |
7223 | ||
7224 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, | |
7225 | (void **)gang, cur >> fs_info->sectorsize_bits, | |
7226 | min_t(unsigned int, GANG_LOOKUP_SIZE, | |
7227 | PAGE_SIZE / fs_info->nodesize)); | |
7228 | if (ret == 0) | |
7229 | goto out; | |
7230 | for (i = 0; i < ret; i++) { | |
7231 | /* Already beyond page end */ | |
7232 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
7233 | goto out; | |
7234 | /* Found one */ | |
7235 | if (gang[i]->start >= bytenr) { | |
7236 | found = gang[i]; | |
7237 | goto out; | |
7238 | } | |
7239 | } | |
7240 | cur = gang[ret - 1]->start + gang[ret - 1]->len; | |
d1e86e3f | 7241 | } |
01cd3909 DS |
7242 | out: |
7243 | return found; | |
d1e86e3f QW |
7244 | } |
7245 | ||
7246 | static int try_release_subpage_extent_buffer(struct page *page) | |
7247 | { | |
7248 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7249 | u64 cur = page_offset(page); | |
7250 | const u64 end = page_offset(page) + PAGE_SIZE; | |
7251 | int ret; | |
7252 | ||
7253 | while (cur < end) { | |
7254 | struct extent_buffer *eb = NULL; | |
7255 | ||
7256 | /* | |
7257 | * Unlike try_release_extent_buffer() which uses page->private | |
7258 | * to grab buffer, for subpage case we rely on radix tree, thus | |
7259 | * we need to ensure radix tree consistency. | |
7260 | * | |
7261 | * We also want an atomic snapshot of the radix tree, thus go | |
7262 | * with spinlock rather than RCU. | |
7263 | */ | |
7264 | spin_lock(&fs_info->buffer_lock); | |
7265 | eb = get_next_extent_buffer(fs_info, page, cur); | |
7266 | if (!eb) { | |
7267 | /* No more eb in the page range after or at cur */ | |
7268 | spin_unlock(&fs_info->buffer_lock); | |
7269 | break; | |
7270 | } | |
7271 | cur = eb->start + eb->len; | |
7272 | ||
7273 | /* | |
7274 | * The same as try_release_extent_buffer(), to ensure the eb | |
7275 | * won't disappear out from under us. | |
7276 | */ | |
7277 | spin_lock(&eb->refs_lock); | |
7278 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
7279 | spin_unlock(&eb->refs_lock); | |
7280 | spin_unlock(&fs_info->buffer_lock); | |
7281 | break; | |
7282 | } | |
7283 | spin_unlock(&fs_info->buffer_lock); | |
7284 | ||
7285 | /* | |
7286 | * If tree ref isn't set then we know the ref on this eb is a | |
7287 | * real ref, so just return, this eb will likely be freed soon | |
7288 | * anyway. | |
7289 | */ | |
7290 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
7291 | spin_unlock(&eb->refs_lock); | |
7292 | break; | |
7293 | } | |
7294 | ||
7295 | /* | |
7296 | * Here we don't care about the return value, we will always | |
7297 | * check the page private at the end. And | |
7298 | * release_extent_buffer() will release the refs_lock. | |
7299 | */ | |
7300 | release_extent_buffer(eb); | |
7301 | } | |
7302 | /* | |
7303 | * Finally to check if we have cleared page private, as if we have | |
7304 | * released all ebs in the page, the page private should be cleared now. | |
7305 | */ | |
7306 | spin_lock(&page->mapping->private_lock); | |
7307 | if (!PagePrivate(page)) | |
7308 | ret = 1; | |
7309 | else | |
7310 | ret = 0; | |
7311 | spin_unlock(&page->mapping->private_lock); | |
7312 | return ret; | |
7313 | ||
7314 | } | |
7315 | ||
f7a52a40 | 7316 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 7317 | { |
6af118ce | 7318 | struct extent_buffer *eb; |
6af118ce | 7319 | |
fbca46eb | 7320 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
d1e86e3f QW |
7321 | return try_release_subpage_extent_buffer(page); |
7322 | ||
3083ee2e | 7323 | /* |
d1e86e3f QW |
7324 | * We need to make sure nobody is changing page->private, as we rely on |
7325 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
7326 | */ |
7327 | spin_lock(&page->mapping->private_lock); | |
7328 | if (!PagePrivate(page)) { | |
7329 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 7330 | return 1; |
45f49bce | 7331 | } |
6af118ce | 7332 | |
3083ee2e JB |
7333 | eb = (struct extent_buffer *)page->private; |
7334 | BUG_ON(!eb); | |
19fe0a8b MX |
7335 | |
7336 | /* | |
3083ee2e JB |
7337 | * This is a little awful but should be ok, we need to make sure that |
7338 | * the eb doesn't disappear out from under us while we're looking at | |
7339 | * this page. | |
19fe0a8b | 7340 | */ |
3083ee2e | 7341 | spin_lock(&eb->refs_lock); |
0b32f4bb | 7342 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
7343 | spin_unlock(&eb->refs_lock); |
7344 | spin_unlock(&page->mapping->private_lock); | |
7345 | return 0; | |
b9473439 | 7346 | } |
3083ee2e | 7347 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 7348 | |
19fe0a8b | 7349 | /* |
3083ee2e JB |
7350 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
7351 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 7352 | */ |
3083ee2e JB |
7353 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
7354 | spin_unlock(&eb->refs_lock); | |
7355 | return 0; | |
b9473439 | 7356 | } |
19fe0a8b | 7357 | |
f7a52a40 | 7358 | return release_extent_buffer(eb); |
6af118ce | 7359 | } |
bfb484d9 JB |
7360 | |
7361 | /* | |
7362 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
7363 | * @fs_info: the fs_info | |
7364 | * @bytenr: bytenr to read | |
3fbaf258 | 7365 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 7366 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 7367 | * @level: level for the eb |
bfb484d9 JB |
7368 | * |
7369 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
7370 | * normal uptodate check of the eb, without checking the generation. If we have | |
7371 | * to read the block we will not block on anything. | |
7372 | */ | |
7373 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 7374 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
7375 | { |
7376 | struct extent_buffer *eb; | |
7377 | int ret; | |
7378 | ||
3fbaf258 | 7379 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
7380 | if (IS_ERR(eb)) |
7381 | return; | |
7382 | ||
7383 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
7384 | free_extent_buffer(eb); | |
7385 | return; | |
7386 | } | |
7387 | ||
7388 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
7389 | if (ret < 0) | |
7390 | free_extent_buffer_stale(eb); | |
7391 | else | |
7392 | free_extent_buffer(eb); | |
7393 | } | |
7394 | ||
7395 | /* | |
7396 | * btrfs_readahead_node_child - readahead a node's child block | |
7397 | * @node: parent node we're reading from | |
7398 | * @slot: slot in the parent node for the child we want to read | |
7399 | * | |
7400 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
7401 | * the slot in the node provided. | |
7402 | */ | |
7403 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
7404 | { | |
7405 | btrfs_readahead_tree_block(node->fs_info, | |
7406 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
7407 | btrfs_header_owner(node), |
7408 | btrfs_node_ptr_generation(node, slot), | |
7409 | btrfs_header_level(node) - 1); | |
bfb484d9 | 7410 | } |