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; | |
8ac9f7c1 | 36 | static struct bio_set btrfs_bioset; |
d1310b2e | 37 | |
27a3507d FM |
38 | static inline bool extent_state_in_tree(const struct extent_state *state) |
39 | { | |
40 | return !RB_EMPTY_NODE(&state->rb_node); | |
41 | } | |
42 | ||
6d49ba1b | 43 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e | 44 | static LIST_HEAD(states); |
d397712b | 45 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b | 46 | |
3fd63727 JB |
47 | static inline void btrfs_leak_debug_add(spinlock_t *lock, |
48 | struct list_head *new, | |
49 | struct list_head *head) | |
6d49ba1b ES |
50 | { |
51 | unsigned long flags; | |
52 | ||
3fd63727 | 53 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 54 | list_add(new, head); |
3fd63727 | 55 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
56 | } |
57 | ||
3fd63727 JB |
58 | static inline void btrfs_leak_debug_del(spinlock_t *lock, |
59 | struct list_head *entry) | |
6d49ba1b ES |
60 | { |
61 | unsigned long flags; | |
62 | ||
3fd63727 | 63 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 64 | list_del(entry); |
3fd63727 | 65 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
66 | } |
67 | ||
3fd63727 | 68 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 69 | { |
6d49ba1b | 70 | struct extent_buffer *eb; |
3fd63727 | 71 | unsigned long flags; |
6d49ba1b | 72 | |
8c38938c JB |
73 | /* |
74 | * If we didn't get into open_ctree our allocated_ebs will not be | |
75 | * initialized, so just skip this. | |
76 | */ | |
77 | if (!fs_info->allocated_ebs.next) | |
78 | return; | |
79 | ||
b95b78e6 | 80 | WARN_ON(!list_empty(&fs_info->allocated_ebs)); |
3fd63727 JB |
81 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
82 | while (!list_empty(&fs_info->allocated_ebs)) { | |
83 | eb = list_first_entry(&fs_info->allocated_ebs, | |
84 | struct extent_buffer, leak_list); | |
8c38938c JB |
85 | pr_err( |
86 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
87 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
88 | btrfs_header_owner(eb)); | |
33ca832f JB |
89 | list_del(&eb->leak_list); |
90 | kmem_cache_free(extent_buffer_cache, eb); | |
91 | } | |
3fd63727 | 92 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f JB |
93 | } |
94 | ||
95 | static inline void btrfs_extent_state_leak_debug_check(void) | |
96 | { | |
97 | struct extent_state *state; | |
98 | ||
6d49ba1b ES |
99 | while (!list_empty(&states)) { |
100 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 101 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
102 | state->start, state->end, state->state, |
103 | extent_state_in_tree(state), | |
b7ac31b7 | 104 | refcount_read(&state->refs)); |
6d49ba1b ES |
105 | list_del(&state->leak_list); |
106 | kmem_cache_free(extent_state_cache, state); | |
107 | } | |
6d49ba1b | 108 | } |
8d599ae1 | 109 | |
a5dee37d JB |
110 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
111 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 112 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 113 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 114 | { |
65a680f6 NB |
115 | struct inode *inode = tree->private_data; |
116 | u64 isize; | |
117 | ||
118 | if (!inode || !is_data_inode(inode)) | |
119 | return; | |
120 | ||
121 | isize = i_size_read(inode); | |
122 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
123 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
124 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
125 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
126 | } | |
8d599ae1 | 127 | } |
6d49ba1b | 128 | #else |
3fd63727 JB |
129 | #define btrfs_leak_debug_add(lock, new, head) do {} while (0) |
130 | #define btrfs_leak_debug_del(lock, entry) do {} while (0) | |
33ca832f | 131 | #define btrfs_extent_state_leak_debug_check() do {} while (0) |
8d599ae1 | 132 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 133 | #endif |
d1310b2e | 134 | |
d1310b2e CM |
135 | struct tree_entry { |
136 | u64 start; | |
137 | u64 end; | |
d1310b2e CM |
138 | struct rb_node rb_node; |
139 | }; | |
140 | ||
7aab8b32 CH |
141 | /* |
142 | * Structure to record info about the bio being assembled, and other info like | |
143 | * how many bytes are there before stripe/ordered extent boundary. | |
144 | */ | |
145 | struct btrfs_bio_ctrl { | |
146 | struct bio *bio; | |
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 | ||
cb3a12d9 DS |
181 | static void submit_one_bio(struct bio *bio, int mirror_num, |
182 | enum btrfs_compression_type compress_type) | |
bb58eb9e | 183 | { |
bb58eb9e | 184 | struct extent_io_tree *tree = bio->bi_private; |
bb58eb9e QW |
185 | |
186 | bio->bi_private = NULL; | |
187 | ||
e0eefe07 QW |
188 | /* Caller should ensure the bio has at least some range added */ |
189 | ASSERT(bio->bi_iter.bi_size); | |
c9583ada | 190 | |
908930f3 | 191 | if (is_data_inode(tree->private_data)) |
c9583ada | 192 | btrfs_submit_data_bio(tree->private_data, bio, mirror_num, |
cb3a12d9 | 193 | compress_type); |
908930f3 | 194 | else |
abf48d58 | 195 | btrfs_submit_metadata_bio(tree->private_data, bio, mirror_num); |
c9583ada QW |
196 | /* |
197 | * Above submission hooks will handle the error by ending the bio, | |
198 | * which will do the cleanup properly. So here we should not return | |
199 | * any error, or the caller of submit_extent_page() will do cleanup | |
200 | * again, causing problems. | |
201 | */ | |
bb58eb9e QW |
202 | } |
203 | ||
3065976b QW |
204 | /* Cleanup unsubmitted bios */ |
205 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
206 | { | |
390ed29b QW |
207 | struct bio *bio = epd->bio_ctrl.bio; |
208 | ||
209 | if (bio) { | |
210 | bio->bi_status = errno_to_blk_status(ret); | |
211 | bio_endio(bio); | |
212 | epd->bio_ctrl.bio = NULL; | |
3065976b QW |
213 | } |
214 | } | |
215 | ||
f4340622 QW |
216 | /* |
217 | * Submit bio from extent page data via submit_one_bio | |
218 | * | |
219 | * Return 0 if everything is OK. | |
220 | * Return <0 for error. | |
221 | */ | |
c9583ada | 222 | static void flush_write_bio(struct extent_page_data *epd) |
bb58eb9e | 223 | { |
390ed29b | 224 | struct bio *bio = epd->bio_ctrl.bio; |
bb58eb9e | 225 | |
390ed29b | 226 | if (bio) { |
c9583ada | 227 | submit_one_bio(bio, 0, 0); |
f4340622 QW |
228 | /* |
229 | * Clean up of epd->bio is handled by its endio function. | |
230 | * And endio is either triggered by successful bio execution | |
231 | * or the error handler of submit bio hook. | |
232 | * So at this point, no matter what happened, we don't need | |
233 | * to clean up epd->bio. | |
234 | */ | |
390ed29b | 235 | epd->bio_ctrl.bio = NULL; |
bb58eb9e QW |
236 | } |
237 | } | |
e2932ee0 | 238 | |
6f0d04f8 | 239 | int __init extent_state_cache_init(void) |
d1310b2e | 240 | { |
837e1972 | 241 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 242 | sizeof(struct extent_state), 0, |
fba4b697 | 243 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
244 | if (!extent_state_cache) |
245 | return -ENOMEM; | |
6f0d04f8 JB |
246 | return 0; |
247 | } | |
d1310b2e | 248 | |
6f0d04f8 JB |
249 | int __init extent_io_init(void) |
250 | { | |
837e1972 | 251 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 252 | sizeof(struct extent_buffer), 0, |
fba4b697 | 253 | SLAB_MEM_SPREAD, NULL); |
d1310b2e | 254 | if (!extent_buffer_cache) |
6f0d04f8 | 255 | return -ENOMEM; |
9be3395b | 256 | |
8ac9f7c1 | 257 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
c3a3b19b | 258 | offsetof(struct btrfs_bio, bio), |
8ac9f7c1 | 259 | BIOSET_NEED_BVECS)) |
9be3395b | 260 | goto free_buffer_cache; |
b208c2f7 | 261 | |
8ac9f7c1 | 262 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
263 | goto free_bioset; |
264 | ||
d1310b2e CM |
265 | return 0; |
266 | ||
b208c2f7 | 267 | free_bioset: |
8ac9f7c1 | 268 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 269 | |
9be3395b CM |
270 | free_buffer_cache: |
271 | kmem_cache_destroy(extent_buffer_cache); | |
272 | extent_buffer_cache = NULL; | |
6f0d04f8 JB |
273 | return -ENOMEM; |
274 | } | |
9be3395b | 275 | |
6f0d04f8 JB |
276 | void __cold extent_state_cache_exit(void) |
277 | { | |
278 | btrfs_extent_state_leak_debug_check(); | |
d1310b2e | 279 | kmem_cache_destroy(extent_state_cache); |
d1310b2e CM |
280 | } |
281 | ||
e67c718b | 282 | void __cold extent_io_exit(void) |
d1310b2e | 283 | { |
8c0a8537 KS |
284 | /* |
285 | * Make sure all delayed rcu free are flushed before we | |
286 | * destroy caches. | |
287 | */ | |
288 | rcu_barrier(); | |
5598e900 | 289 | kmem_cache_destroy(extent_buffer_cache); |
8ac9f7c1 | 290 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
291 | } |
292 | ||
41a2ee75 JB |
293 | /* |
294 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
295 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
296 | * the tree lock and get the inode lock when setting delalloc. These two things | |
297 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
298 | * two locking patterns mixed up. | |
299 | */ | |
300 | static struct lock_class_key file_extent_tree_class; | |
301 | ||
c258d6e3 | 302 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
303 | struct extent_io_tree *tree, unsigned int owner, |
304 | void *private_data) | |
d1310b2e | 305 | { |
c258d6e3 | 306 | tree->fs_info = fs_info; |
6bef4d31 | 307 | tree->state = RB_ROOT; |
d1310b2e | 308 | tree->dirty_bytes = 0; |
70dec807 | 309 | spin_lock_init(&tree->lock); |
c6100a4b | 310 | tree->private_data = private_data; |
43eb5f29 | 311 | tree->owner = owner; |
41a2ee75 JB |
312 | if (owner == IO_TREE_INODE_FILE_EXTENT) |
313 | lockdep_set_class(&tree->lock, &file_extent_tree_class); | |
d1310b2e | 314 | } |
d1310b2e | 315 | |
41e7acd3 NB |
316 | void extent_io_tree_release(struct extent_io_tree *tree) |
317 | { | |
318 | spin_lock(&tree->lock); | |
319 | /* | |
320 | * Do a single barrier for the waitqueue_active check here, the state | |
321 | * of the waitqueue should not change once extent_io_tree_release is | |
322 | * called. | |
323 | */ | |
324 | smp_mb(); | |
325 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
326 | struct rb_node *node; | |
327 | struct extent_state *state; | |
328 | ||
329 | node = rb_first(&tree->state); | |
330 | state = rb_entry(node, struct extent_state, rb_node); | |
331 | rb_erase(&state->rb_node, &tree->state); | |
332 | RB_CLEAR_NODE(&state->rb_node); | |
333 | /* | |
334 | * btree io trees aren't supposed to have tasks waiting for | |
335 | * changes in the flags of extent states ever. | |
336 | */ | |
337 | ASSERT(!waitqueue_active(&state->wq)); | |
338 | free_extent_state(state); | |
339 | ||
340 | cond_resched_lock(&tree->lock); | |
341 | } | |
342 | spin_unlock(&tree->lock); | |
343 | } | |
344 | ||
b2950863 | 345 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
346 | { |
347 | struct extent_state *state; | |
d1310b2e | 348 | |
3ba7ab22 MH |
349 | /* |
350 | * The given mask might be not appropriate for the slab allocator, | |
351 | * drop the unsupported bits | |
352 | */ | |
353 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 354 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 355 | if (!state) |
d1310b2e CM |
356 | return state; |
357 | state->state = 0; | |
47dc196a | 358 | state->failrec = NULL; |
27a3507d | 359 | RB_CLEAR_NODE(&state->rb_node); |
3fd63727 | 360 | btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states); |
b7ac31b7 | 361 | refcount_set(&state->refs, 1); |
d1310b2e | 362 | init_waitqueue_head(&state->wq); |
143bede5 | 363 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
364 | return state; |
365 | } | |
d1310b2e | 366 | |
4845e44f | 367 | void free_extent_state(struct extent_state *state) |
d1310b2e | 368 | { |
d1310b2e CM |
369 | if (!state) |
370 | return; | |
b7ac31b7 | 371 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 372 | WARN_ON(extent_state_in_tree(state)); |
3fd63727 | 373 | btrfs_leak_debug_del(&leak_lock, &state->leak_list); |
143bede5 | 374 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
375 | kmem_cache_free(extent_state_cache, state); |
376 | } | |
377 | } | |
d1310b2e | 378 | |
f2071b21 FM |
379 | static struct rb_node *tree_insert(struct rb_root *root, |
380 | struct rb_node *search_start, | |
381 | u64 offset, | |
12cfbad9 FDBM |
382 | struct rb_node *node, |
383 | struct rb_node ***p_in, | |
384 | struct rb_node **parent_in) | |
d1310b2e | 385 | { |
f2071b21 | 386 | struct rb_node **p; |
d397712b | 387 | struct rb_node *parent = NULL; |
d1310b2e CM |
388 | struct tree_entry *entry; |
389 | ||
12cfbad9 FDBM |
390 | if (p_in && parent_in) { |
391 | p = *p_in; | |
392 | parent = *parent_in; | |
393 | goto do_insert; | |
394 | } | |
395 | ||
f2071b21 | 396 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 397 | while (*p) { |
d1310b2e CM |
398 | parent = *p; |
399 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
400 | ||
401 | if (offset < entry->start) | |
402 | p = &(*p)->rb_left; | |
403 | else if (offset > entry->end) | |
404 | p = &(*p)->rb_right; | |
405 | else | |
406 | return parent; | |
407 | } | |
408 | ||
12cfbad9 | 409 | do_insert: |
d1310b2e CM |
410 | rb_link_node(node, parent, p); |
411 | rb_insert_color(node, root); | |
412 | return NULL; | |
413 | } | |
414 | ||
8666e638 | 415 | /** |
3bed2da1 NB |
416 | * Search @tree for an entry that contains @offset. Such entry would have |
417 | * entry->start <= offset && entry->end >= offset. | |
8666e638 | 418 | * |
3bed2da1 NB |
419 | * @tree: the tree to search |
420 | * @offset: offset that should fall within an entry in @tree | |
421 | * @next_ret: pointer to the first entry whose range ends after @offset | |
422 | * @prev_ret: pointer to the first entry whose range begins before @offset | |
423 | * @p_ret: pointer where new node should be anchored (used when inserting an | |
424 | * entry in the tree) | |
425 | * @parent_ret: points to entry which would have been the parent of the entry, | |
8666e638 NB |
426 | * containing @offset |
427 | * | |
428 | * This function returns a pointer to the entry that contains @offset byte | |
429 | * address. If no such entry exists, then NULL is returned and the other | |
430 | * pointer arguments to the function are filled, otherwise the found entry is | |
431 | * returned and other pointers are left untouched. | |
432 | */ | |
80ea96b1 | 433 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 434 | struct rb_node **next_ret, |
352646c7 | 435 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
436 | struct rb_node ***p_ret, |
437 | struct rb_node **parent_ret) | |
d1310b2e | 438 | { |
80ea96b1 | 439 | struct rb_root *root = &tree->state; |
12cfbad9 | 440 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
441 | struct rb_node *prev = NULL; |
442 | struct rb_node *orig_prev = NULL; | |
443 | struct tree_entry *entry; | |
444 | struct tree_entry *prev_entry = NULL; | |
445 | ||
12cfbad9 FDBM |
446 | while (*n) { |
447 | prev = *n; | |
448 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
449 | prev_entry = entry; |
450 | ||
451 | if (offset < entry->start) | |
12cfbad9 | 452 | n = &(*n)->rb_left; |
d1310b2e | 453 | else if (offset > entry->end) |
12cfbad9 | 454 | n = &(*n)->rb_right; |
d397712b | 455 | else |
12cfbad9 | 456 | return *n; |
d1310b2e CM |
457 | } |
458 | ||
12cfbad9 FDBM |
459 | if (p_ret) |
460 | *p_ret = n; | |
461 | if (parent_ret) | |
462 | *parent_ret = prev; | |
463 | ||
352646c7 | 464 | if (next_ret) { |
d1310b2e | 465 | orig_prev = prev; |
d397712b | 466 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
467 | prev = rb_next(prev); |
468 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
469 | } | |
352646c7 | 470 | *next_ret = prev; |
d1310b2e CM |
471 | prev = orig_prev; |
472 | } | |
473 | ||
352646c7 | 474 | if (prev_ret) { |
d1310b2e | 475 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 476 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
477 | prev = rb_prev(prev); |
478 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
479 | } | |
352646c7 | 480 | *prev_ret = prev; |
d1310b2e CM |
481 | } |
482 | return NULL; | |
483 | } | |
484 | ||
12cfbad9 FDBM |
485 | static inline struct rb_node * |
486 | tree_search_for_insert(struct extent_io_tree *tree, | |
487 | u64 offset, | |
488 | struct rb_node ***p_ret, | |
489 | struct rb_node **parent_ret) | |
d1310b2e | 490 | { |
352646c7 | 491 | struct rb_node *next= NULL; |
d1310b2e | 492 | struct rb_node *ret; |
70dec807 | 493 | |
352646c7 | 494 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 495 | if (!ret) |
352646c7 | 496 | return next; |
d1310b2e CM |
497 | return ret; |
498 | } | |
499 | ||
12cfbad9 FDBM |
500 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
501 | u64 offset) | |
502 | { | |
503 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
504 | } | |
505 | ||
d1310b2e CM |
506 | /* |
507 | * utility function to look for merge candidates inside a given range. | |
508 | * Any extents with matching state are merged together into a single | |
509 | * extent in the tree. Extents with EXTENT_IO in their state field | |
510 | * are not merged because the end_io handlers need to be able to do | |
511 | * operations on them without sleeping (or doing allocations/splits). | |
512 | * | |
513 | * This should be called with the tree lock held. | |
514 | */ | |
1bf85046 JM |
515 | static void merge_state(struct extent_io_tree *tree, |
516 | struct extent_state *state) | |
d1310b2e CM |
517 | { |
518 | struct extent_state *other; | |
519 | struct rb_node *other_node; | |
520 | ||
8882679e | 521 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 522 | return; |
d1310b2e CM |
523 | |
524 | other_node = rb_prev(&state->rb_node); | |
525 | if (other_node) { | |
526 | other = rb_entry(other_node, struct extent_state, rb_node); | |
527 | if (other->end == state->start - 1 && | |
528 | other->state == state->state) { | |
5c848198 NB |
529 | if (tree->private_data && |
530 | is_data_inode(tree->private_data)) | |
531 | btrfs_merge_delalloc_extent(tree->private_data, | |
532 | state, other); | |
d1310b2e | 533 | state->start = other->start; |
d1310b2e | 534 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 535 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
536 | free_extent_state(other); |
537 | } | |
538 | } | |
539 | other_node = rb_next(&state->rb_node); | |
540 | if (other_node) { | |
541 | other = rb_entry(other_node, struct extent_state, rb_node); | |
542 | if (other->start == state->end + 1 && | |
543 | other->state == state->state) { | |
5c848198 NB |
544 | if (tree->private_data && |
545 | is_data_inode(tree->private_data)) | |
546 | btrfs_merge_delalloc_extent(tree->private_data, | |
547 | state, other); | |
df98b6e2 | 548 | state->end = other->end; |
df98b6e2 | 549 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 550 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 551 | free_extent_state(other); |
d1310b2e CM |
552 | } |
553 | } | |
d1310b2e CM |
554 | } |
555 | ||
3150b699 | 556 | static void set_state_bits(struct extent_io_tree *tree, |
f97e27e9 | 557 | struct extent_state *state, u32 *bits, |
d38ed27f | 558 | struct extent_changeset *changeset); |
3150b699 | 559 | |
d1310b2e CM |
560 | /* |
561 | * insert an extent_state struct into the tree. 'bits' are set on the | |
562 | * struct before it is inserted. | |
563 | * | |
564 | * This may return -EEXIST if the extent is already there, in which case the | |
565 | * state struct is freed. | |
566 | * | |
567 | * The tree lock is not taken internally. This is a utility function and | |
568 | * probably isn't what you want to call (see set/clear_extent_bit). | |
569 | */ | |
570 | static int insert_state(struct extent_io_tree *tree, | |
571 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
572 | struct rb_node ***p, |
573 | struct rb_node **parent, | |
f97e27e9 | 574 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e CM |
575 | { |
576 | struct rb_node *node; | |
577 | ||
2792237d DS |
578 | if (end < start) { |
579 | btrfs_err(tree->fs_info, | |
580 | "insert state: end < start %llu %llu", end, start); | |
581 | WARN_ON(1); | |
582 | } | |
d1310b2e CM |
583 | state->start = start; |
584 | state->end = end; | |
9ed74f2d | 585 | |
d38ed27f | 586 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 587 | |
f2071b21 | 588 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
589 | if (node) { |
590 | struct extent_state *found; | |
591 | found = rb_entry(node, struct extent_state, rb_node); | |
2792237d DS |
592 | btrfs_err(tree->fs_info, |
593 | "found node %llu %llu on insert of %llu %llu", | |
c1c9ff7c | 594 | found->start, found->end, start, end); |
d1310b2e CM |
595 | return -EEXIST; |
596 | } | |
597 | merge_state(tree, state); | |
598 | return 0; | |
599 | } | |
600 | ||
601 | /* | |
602 | * split a given extent state struct in two, inserting the preallocated | |
603 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
604 | * offset inside 'orig' where it should be split. | |
605 | * | |
606 | * Before calling, | |
607 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
608 | * are two extent state structs in the tree: | |
609 | * prealloc: [orig->start, split - 1] | |
610 | * orig: [ split, orig->end ] | |
611 | * | |
612 | * The tree locks are not taken by this function. They need to be held | |
613 | * by the caller. | |
614 | */ | |
615 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
616 | struct extent_state *prealloc, u64 split) | |
617 | { | |
618 | struct rb_node *node; | |
9ed74f2d | 619 | |
abbb55f4 NB |
620 | if (tree->private_data && is_data_inode(tree->private_data)) |
621 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 622 | |
d1310b2e CM |
623 | prealloc->start = orig->start; |
624 | prealloc->end = split - 1; | |
625 | prealloc->state = orig->state; | |
626 | orig->start = split; | |
627 | ||
f2071b21 FM |
628 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
629 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 630 | if (node) { |
d1310b2e CM |
631 | free_extent_state(prealloc); |
632 | return -EEXIST; | |
633 | } | |
634 | return 0; | |
635 | } | |
636 | ||
cdc6a395 LZ |
637 | static struct extent_state *next_state(struct extent_state *state) |
638 | { | |
639 | struct rb_node *next = rb_next(&state->rb_node); | |
640 | if (next) | |
641 | return rb_entry(next, struct extent_state, rb_node); | |
642 | else | |
643 | return NULL; | |
644 | } | |
645 | ||
d1310b2e CM |
646 | /* |
647 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 648 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
649 | * |
650 | * If no bits are set on the state struct after clearing things, the | |
651 | * struct is freed and removed from the tree | |
652 | */ | |
cdc6a395 LZ |
653 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
654 | struct extent_state *state, | |
f97e27e9 | 655 | u32 *bits, int wake, |
fefdc557 | 656 | struct extent_changeset *changeset) |
d1310b2e | 657 | { |
cdc6a395 | 658 | struct extent_state *next; |
f97e27e9 | 659 | u32 bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 660 | int ret; |
d1310b2e | 661 | |
0ca1f7ce | 662 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
663 | u64 range = state->end - state->start + 1; |
664 | WARN_ON(range > tree->dirty_bytes); | |
665 | tree->dirty_bytes -= range; | |
666 | } | |
a36bb5f9 NB |
667 | |
668 | if (tree->private_data && is_data_inode(tree->private_data)) | |
669 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
670 | ||
57599c7e DS |
671 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
672 | BUG_ON(ret < 0); | |
32c00aff | 673 | state->state &= ~bits_to_clear; |
d1310b2e CM |
674 | if (wake) |
675 | wake_up(&state->wq); | |
0ca1f7ce | 676 | if (state->state == 0) { |
cdc6a395 | 677 | next = next_state(state); |
27a3507d | 678 | if (extent_state_in_tree(state)) { |
d1310b2e | 679 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 680 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
681 | free_extent_state(state); |
682 | } else { | |
683 | WARN_ON(1); | |
684 | } | |
685 | } else { | |
686 | merge_state(tree, state); | |
cdc6a395 | 687 | next = next_state(state); |
d1310b2e | 688 | } |
cdc6a395 | 689 | return next; |
d1310b2e CM |
690 | } |
691 | ||
8233767a XG |
692 | static struct extent_state * |
693 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
694 | { | |
695 | if (!prealloc) | |
696 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
697 | ||
698 | return prealloc; | |
699 | } | |
700 | ||
48a3b636 | 701 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 702 | { |
29b665cc | 703 | btrfs_panic(tree->fs_info, err, |
05912a3c | 704 | "locking error: extent tree was modified by another thread while locked"); |
c2d904e0 JM |
705 | } |
706 | ||
d1310b2e CM |
707 | /* |
708 | * clear some bits on a range in the tree. This may require splitting | |
709 | * or inserting elements in the tree, so the gfp mask is used to | |
710 | * indicate which allocations or sleeping are allowed. | |
711 | * | |
712 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
713 | * the given range from the tree regardless of state (ie for truncate). | |
714 | * | |
715 | * the range [start, end] is inclusive. | |
716 | * | |
6763af84 | 717 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 718 | */ |
66b0c887 | 719 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 QW |
720 | u32 bits, int wake, int delete, |
721 | struct extent_state **cached_state, | |
722 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
723 | { |
724 | struct extent_state *state; | |
2c64c53d | 725 | struct extent_state *cached; |
d1310b2e CM |
726 | struct extent_state *prealloc = NULL; |
727 | struct rb_node *node; | |
5c939df5 | 728 | u64 last_end; |
d1310b2e | 729 | int err; |
2ac55d41 | 730 | int clear = 0; |
d1310b2e | 731 | |
a5dee37d | 732 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 733 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 734 | |
7ee9e440 JB |
735 | if (bits & EXTENT_DELALLOC) |
736 | bits |= EXTENT_NORESERVE; | |
737 | ||
0ca1f7ce YZ |
738 | if (delete) |
739 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 740 | |
8882679e | 741 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 742 | clear = 1; |
d1310b2e | 743 | again: |
d0164adc | 744 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
745 | /* |
746 | * Don't care for allocation failure here because we might end | |
747 | * up not needing the pre-allocated extent state at all, which | |
748 | * is the case if we only have in the tree extent states that | |
749 | * cover our input range and don't cover too any other range. | |
750 | * If we end up needing a new extent state we allocate it later. | |
751 | */ | |
d1310b2e | 752 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
753 | } |
754 | ||
cad321ad | 755 | spin_lock(&tree->lock); |
2c64c53d CM |
756 | if (cached_state) { |
757 | cached = *cached_state; | |
2ac55d41 JB |
758 | |
759 | if (clear) { | |
760 | *cached_state = NULL; | |
761 | cached_state = NULL; | |
762 | } | |
763 | ||
27a3507d FM |
764 | if (cached && extent_state_in_tree(cached) && |
765 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 766 | if (clear) |
b7ac31b7 | 767 | refcount_dec(&cached->refs); |
2c64c53d | 768 | state = cached; |
42daec29 | 769 | goto hit_next; |
2c64c53d | 770 | } |
2ac55d41 JB |
771 | if (clear) |
772 | free_extent_state(cached); | |
2c64c53d | 773 | } |
d1310b2e CM |
774 | /* |
775 | * this search will find the extents that end after | |
776 | * our range starts | |
777 | */ | |
80ea96b1 | 778 | node = tree_search(tree, start); |
d1310b2e CM |
779 | if (!node) |
780 | goto out; | |
781 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 782 | hit_next: |
d1310b2e CM |
783 | if (state->start > end) |
784 | goto out; | |
785 | WARN_ON(state->end < start); | |
5c939df5 | 786 | last_end = state->end; |
d1310b2e | 787 | |
0449314a | 788 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
789 | if (!(state->state & bits)) { |
790 | state = next_state(state); | |
0449314a | 791 | goto next; |
cdc6a395 | 792 | } |
0449314a | 793 | |
d1310b2e CM |
794 | /* |
795 | * | ---- desired range ---- | | |
796 | * | state | or | |
797 | * | ------------- state -------------- | | |
798 | * | |
799 | * We need to split the extent we found, and may flip | |
800 | * bits on second half. | |
801 | * | |
802 | * If the extent we found extends past our range, we | |
803 | * just split and search again. It'll get split again | |
804 | * the next time though. | |
805 | * | |
806 | * If the extent we found is inside our range, we clear | |
807 | * the desired bit on it. | |
808 | */ | |
809 | ||
810 | if (state->start < start) { | |
8233767a XG |
811 | prealloc = alloc_extent_state_atomic(prealloc); |
812 | BUG_ON(!prealloc); | |
d1310b2e | 813 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
814 | if (err) |
815 | extent_io_tree_panic(tree, err); | |
816 | ||
d1310b2e CM |
817 | prealloc = NULL; |
818 | if (err) | |
819 | goto out; | |
820 | if (state->end <= end) { | |
fefdc557 QW |
821 | state = clear_state_bit(tree, state, &bits, wake, |
822 | changeset); | |
d1ac6e41 | 823 | goto next; |
d1310b2e CM |
824 | } |
825 | goto search_again; | |
826 | } | |
827 | /* | |
828 | * | ---- desired range ---- | | |
829 | * | state | | |
830 | * We need to split the extent, and clear the bit | |
831 | * on the first half | |
832 | */ | |
833 | if (state->start <= end && state->end > end) { | |
8233767a XG |
834 | prealloc = alloc_extent_state_atomic(prealloc); |
835 | BUG_ON(!prealloc); | |
d1310b2e | 836 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
837 | if (err) |
838 | extent_io_tree_panic(tree, err); | |
839 | ||
d1310b2e CM |
840 | if (wake) |
841 | wake_up(&state->wq); | |
42daec29 | 842 | |
fefdc557 | 843 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 844 | |
d1310b2e CM |
845 | prealloc = NULL; |
846 | goto out; | |
847 | } | |
42daec29 | 848 | |
fefdc557 | 849 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 850 | next: |
5c939df5 YZ |
851 | if (last_end == (u64)-1) |
852 | goto out; | |
853 | start = last_end + 1; | |
cdc6a395 | 854 | if (start <= end && state && !need_resched()) |
692e5759 | 855 | goto hit_next; |
d1310b2e CM |
856 | |
857 | search_again: | |
858 | if (start > end) | |
859 | goto out; | |
cad321ad | 860 | spin_unlock(&tree->lock); |
d0164adc | 861 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
862 | cond_resched(); |
863 | goto again; | |
7ab5cb2a DS |
864 | |
865 | out: | |
866 | spin_unlock(&tree->lock); | |
867 | if (prealloc) | |
868 | free_extent_state(prealloc); | |
869 | ||
870 | return 0; | |
871 | ||
d1310b2e | 872 | } |
d1310b2e | 873 | |
143bede5 JM |
874 | static void wait_on_state(struct extent_io_tree *tree, |
875 | struct extent_state *state) | |
641f5219 CH |
876 | __releases(tree->lock) |
877 | __acquires(tree->lock) | |
d1310b2e CM |
878 | { |
879 | DEFINE_WAIT(wait); | |
880 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 881 | spin_unlock(&tree->lock); |
d1310b2e | 882 | schedule(); |
cad321ad | 883 | spin_lock(&tree->lock); |
d1310b2e | 884 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
885 | } |
886 | ||
887 | /* | |
888 | * waits for one or more bits to clear on a range in the state tree. | |
889 | * The range [start, end] is inclusive. | |
890 | * The tree lock is taken by this function | |
891 | */ | |
41074888 | 892 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 893 | u32 bits) |
d1310b2e CM |
894 | { |
895 | struct extent_state *state; | |
896 | struct rb_node *node; | |
897 | ||
a5dee37d | 898 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 899 | |
cad321ad | 900 | spin_lock(&tree->lock); |
d1310b2e CM |
901 | again: |
902 | while (1) { | |
903 | /* | |
904 | * this search will find all the extents that end after | |
905 | * our range starts | |
906 | */ | |
80ea96b1 | 907 | node = tree_search(tree, start); |
c50d3e71 | 908 | process_node: |
d1310b2e CM |
909 | if (!node) |
910 | break; | |
911 | ||
912 | state = rb_entry(node, struct extent_state, rb_node); | |
913 | ||
914 | if (state->start > end) | |
915 | goto out; | |
916 | ||
917 | if (state->state & bits) { | |
918 | start = state->start; | |
b7ac31b7 | 919 | refcount_inc(&state->refs); |
d1310b2e CM |
920 | wait_on_state(tree, state); |
921 | free_extent_state(state); | |
922 | goto again; | |
923 | } | |
924 | start = state->end + 1; | |
925 | ||
926 | if (start > end) | |
927 | break; | |
928 | ||
c50d3e71 FM |
929 | if (!cond_resched_lock(&tree->lock)) { |
930 | node = rb_next(node); | |
931 | goto process_node; | |
932 | } | |
d1310b2e CM |
933 | } |
934 | out: | |
cad321ad | 935 | spin_unlock(&tree->lock); |
d1310b2e | 936 | } |
d1310b2e | 937 | |
1bf85046 | 938 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 939 | struct extent_state *state, |
f97e27e9 | 940 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e | 941 | { |
f97e27e9 | 942 | u32 bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 943 | int ret; |
9ed74f2d | 944 | |
e06a1fc9 NB |
945 | if (tree->private_data && is_data_inode(tree->private_data)) |
946 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
947 | ||
0ca1f7ce | 948 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
949 | u64 range = state->end - state->start + 1; |
950 | tree->dirty_bytes += range; | |
951 | } | |
57599c7e DS |
952 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
953 | BUG_ON(ret < 0); | |
0ca1f7ce | 954 | state->state |= bits_to_set; |
d1310b2e CM |
955 | } |
956 | ||
e38e2ed7 FM |
957 | static void cache_state_if_flags(struct extent_state *state, |
958 | struct extent_state **cached_ptr, | |
9ee49a04 | 959 | unsigned flags) |
2c64c53d CM |
960 | { |
961 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 962 | if (!flags || (state->state & flags)) { |
2c64c53d | 963 | *cached_ptr = state; |
b7ac31b7 | 964 | refcount_inc(&state->refs); |
2c64c53d CM |
965 | } |
966 | } | |
967 | } | |
968 | ||
e38e2ed7 FM |
969 | static void cache_state(struct extent_state *state, |
970 | struct extent_state **cached_ptr) | |
971 | { | |
972 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 973 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
974 | } |
975 | ||
d1310b2e | 976 | /* |
1edbb734 CM |
977 | * set some bits on a range in the tree. This may require allocations or |
978 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 979 | * |
1edbb734 CM |
980 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
981 | * part of the range already has the desired bits set. The start of the | |
982 | * existing range is returned in failed_start in this case. | |
d1310b2e | 983 | * |
1edbb734 | 984 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 985 | */ |
f97e27e9 QW |
986 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits, |
987 | u32 exclusive_bits, u64 *failed_start, | |
1cab5e72 NB |
988 | struct extent_state **cached_state, gfp_t mask, |
989 | struct extent_changeset *changeset) | |
d1310b2e CM |
990 | { |
991 | struct extent_state *state; | |
992 | struct extent_state *prealloc = NULL; | |
993 | struct rb_node *node; | |
12cfbad9 FDBM |
994 | struct rb_node **p; |
995 | struct rb_node *parent; | |
d1310b2e | 996 | int err = 0; |
d1310b2e CM |
997 | u64 last_start; |
998 | u64 last_end; | |
42daec29 | 999 | |
a5dee37d | 1000 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 1001 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 1002 | |
3f6bb4ae QW |
1003 | if (exclusive_bits) |
1004 | ASSERT(failed_start); | |
1005 | else | |
1006 | ASSERT(failed_start == NULL); | |
d1310b2e | 1007 | again: |
d0164adc | 1008 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
1009 | /* |
1010 | * Don't care for allocation failure here because we might end | |
1011 | * up not needing the pre-allocated extent state at all, which | |
1012 | * is the case if we only have in the tree extent states that | |
1013 | * cover our input range and don't cover too any other range. | |
1014 | * If we end up needing a new extent state we allocate it later. | |
1015 | */ | |
d1310b2e | 1016 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
1017 | } |
1018 | ||
cad321ad | 1019 | spin_lock(&tree->lock); |
9655d298 CM |
1020 | if (cached_state && *cached_state) { |
1021 | state = *cached_state; | |
df98b6e2 | 1022 | if (state->start <= start && state->end > start && |
27a3507d | 1023 | extent_state_in_tree(state)) { |
9655d298 CM |
1024 | node = &state->rb_node; |
1025 | goto hit_next; | |
1026 | } | |
1027 | } | |
d1310b2e CM |
1028 | /* |
1029 | * this search will find all the extents that end after | |
1030 | * our range starts. | |
1031 | */ | |
12cfbad9 | 1032 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1033 | if (!node) { |
8233767a XG |
1034 | prealloc = alloc_extent_state_atomic(prealloc); |
1035 | BUG_ON(!prealloc); | |
12cfbad9 | 1036 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1037 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
1038 | if (err) |
1039 | extent_io_tree_panic(tree, err); | |
1040 | ||
c42ac0bc | 1041 | cache_state(prealloc, cached_state); |
d1310b2e | 1042 | prealloc = NULL; |
d1310b2e CM |
1043 | goto out; |
1044 | } | |
d1310b2e | 1045 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1046 | hit_next: |
d1310b2e CM |
1047 | last_start = state->start; |
1048 | last_end = state->end; | |
1049 | ||
1050 | /* | |
1051 | * | ---- desired range ---- | | |
1052 | * | state | | |
1053 | * | |
1054 | * Just lock what we found and keep going | |
1055 | */ | |
1056 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1057 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1058 | *failed_start = state->start; |
1059 | err = -EEXIST; | |
1060 | goto out; | |
1061 | } | |
42daec29 | 1062 | |
d38ed27f | 1063 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1064 | cache_state(state, cached_state); |
d1310b2e | 1065 | merge_state(tree, state); |
5c939df5 YZ |
1066 | if (last_end == (u64)-1) |
1067 | goto out; | |
1068 | start = last_end + 1; | |
d1ac6e41 LB |
1069 | state = next_state(state); |
1070 | if (start < end && state && state->start == start && | |
1071 | !need_resched()) | |
1072 | goto hit_next; | |
d1310b2e CM |
1073 | goto search_again; |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * | ---- desired range ---- | | |
1078 | * | state | | |
1079 | * or | |
1080 | * | ------------- state -------------- | | |
1081 | * | |
1082 | * We need to split the extent we found, and may flip bits on | |
1083 | * second half. | |
1084 | * | |
1085 | * If the extent we found extends past our | |
1086 | * range, we just split and search again. It'll get split | |
1087 | * again the next time though. | |
1088 | * | |
1089 | * If the extent we found is inside our range, we set the | |
1090 | * desired bit on it. | |
1091 | */ | |
1092 | if (state->start < start) { | |
1edbb734 | 1093 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1094 | *failed_start = start; |
1095 | err = -EEXIST; | |
1096 | goto out; | |
1097 | } | |
8233767a | 1098 | |
55ffaabe FM |
1099 | /* |
1100 | * If this extent already has all the bits we want set, then | |
1101 | * skip it, not necessary to split it or do anything with it. | |
1102 | */ | |
1103 | if ((state->state & bits) == bits) { | |
1104 | start = state->end + 1; | |
1105 | cache_state(state, cached_state); | |
1106 | goto search_again; | |
1107 | } | |
1108 | ||
8233767a XG |
1109 | prealloc = alloc_extent_state_atomic(prealloc); |
1110 | BUG_ON(!prealloc); | |
d1310b2e | 1111 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1112 | if (err) |
1113 | extent_io_tree_panic(tree, err); | |
1114 | ||
d1310b2e CM |
1115 | prealloc = NULL; |
1116 | if (err) | |
1117 | goto out; | |
1118 | if (state->end <= end) { | |
d38ed27f | 1119 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1120 | cache_state(state, cached_state); |
d1310b2e | 1121 | merge_state(tree, state); |
5c939df5 YZ |
1122 | if (last_end == (u64)-1) |
1123 | goto out; | |
1124 | start = last_end + 1; | |
d1ac6e41 LB |
1125 | state = next_state(state); |
1126 | if (start < end && state && state->start == start && | |
1127 | !need_resched()) | |
1128 | goto hit_next; | |
d1310b2e CM |
1129 | } |
1130 | goto search_again; | |
1131 | } | |
1132 | /* | |
1133 | * | ---- desired range ---- | | |
1134 | * | state | or | state | | |
1135 | * | |
1136 | * There's a hole, we need to insert something in it and | |
1137 | * ignore the extent we found. | |
1138 | */ | |
1139 | if (state->start > start) { | |
1140 | u64 this_end; | |
1141 | if (end < last_start) | |
1142 | this_end = end; | |
1143 | else | |
d397712b | 1144 | this_end = last_start - 1; |
8233767a XG |
1145 | |
1146 | prealloc = alloc_extent_state_atomic(prealloc); | |
1147 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1148 | |
1149 | /* | |
1150 | * Avoid to free 'prealloc' if it can be merged with | |
1151 | * the later extent. | |
1152 | */ | |
d1310b2e | 1153 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1154 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1155 | if (err) |
1156 | extent_io_tree_panic(tree, err); | |
1157 | ||
9ed74f2d JB |
1158 | cache_state(prealloc, cached_state); |
1159 | prealloc = NULL; | |
d1310b2e CM |
1160 | start = this_end + 1; |
1161 | goto search_again; | |
1162 | } | |
1163 | /* | |
1164 | * | ---- desired range ---- | | |
1165 | * | state | | |
1166 | * We need to split the extent, and set the bit | |
1167 | * on the first half | |
1168 | */ | |
1169 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1170 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1171 | *failed_start = start; |
1172 | err = -EEXIST; | |
1173 | goto out; | |
1174 | } | |
8233767a XG |
1175 | |
1176 | prealloc = alloc_extent_state_atomic(prealloc); | |
1177 | BUG_ON(!prealloc); | |
d1310b2e | 1178 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1179 | if (err) |
1180 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1181 | |
d38ed27f | 1182 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1183 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1184 | merge_state(tree, prealloc); |
1185 | prealloc = NULL; | |
1186 | goto out; | |
1187 | } | |
1188 | ||
b5a4ba14 DS |
1189 | search_again: |
1190 | if (start > end) | |
1191 | goto out; | |
1192 | spin_unlock(&tree->lock); | |
1193 | if (gfpflags_allow_blocking(mask)) | |
1194 | cond_resched(); | |
1195 | goto again; | |
d1310b2e CM |
1196 | |
1197 | out: | |
cad321ad | 1198 | spin_unlock(&tree->lock); |
d1310b2e CM |
1199 | if (prealloc) |
1200 | free_extent_state(prealloc); | |
1201 | ||
1202 | return err; | |
1203 | ||
d1310b2e | 1204 | } |
d1310b2e | 1205 | |
462d6fac | 1206 | /** |
10983f2e LB |
1207 | * convert_extent_bit - convert all bits in a given range from one bit to |
1208 | * another | |
462d6fac JB |
1209 | * @tree: the io tree to search |
1210 | * @start: the start offset in bytes | |
1211 | * @end: the end offset in bytes (inclusive) | |
1212 | * @bits: the bits to set in this range | |
1213 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1214 | * @cached_state: state that we're going to cache |
462d6fac JB |
1215 | * |
1216 | * This will go through and set bits for the given range. If any states exist | |
1217 | * already in this range they are set with the given bit and cleared of the | |
1218 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1219 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1220 | * boundary bits like LOCK. | |
210aa277 DS |
1221 | * |
1222 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1223 | */ |
1224 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 1225 | u32 bits, u32 clear_bits, |
210aa277 | 1226 | struct extent_state **cached_state) |
462d6fac JB |
1227 | { |
1228 | struct extent_state *state; | |
1229 | struct extent_state *prealloc = NULL; | |
1230 | struct rb_node *node; | |
12cfbad9 FDBM |
1231 | struct rb_node **p; |
1232 | struct rb_node *parent; | |
462d6fac JB |
1233 | int err = 0; |
1234 | u64 last_start; | |
1235 | u64 last_end; | |
c8fd3de7 | 1236 | bool first_iteration = true; |
462d6fac | 1237 | |
a5dee37d | 1238 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1239 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1240 | clear_bits); | |
8d599ae1 | 1241 | |
462d6fac | 1242 | again: |
210aa277 | 1243 | if (!prealloc) { |
c8fd3de7 FM |
1244 | /* |
1245 | * Best effort, don't worry if extent state allocation fails | |
1246 | * here for the first iteration. We might have a cached state | |
1247 | * that matches exactly the target range, in which case no | |
1248 | * extent state allocations are needed. We'll only know this | |
1249 | * after locking the tree. | |
1250 | */ | |
210aa277 | 1251 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1252 | if (!prealloc && !first_iteration) |
462d6fac JB |
1253 | return -ENOMEM; |
1254 | } | |
1255 | ||
1256 | spin_lock(&tree->lock); | |
e6138876 JB |
1257 | if (cached_state && *cached_state) { |
1258 | state = *cached_state; | |
1259 | if (state->start <= start && state->end > start && | |
27a3507d | 1260 | extent_state_in_tree(state)) { |
e6138876 JB |
1261 | node = &state->rb_node; |
1262 | goto hit_next; | |
1263 | } | |
1264 | } | |
1265 | ||
462d6fac JB |
1266 | /* |
1267 | * this search will find all the extents that end after | |
1268 | * our range starts. | |
1269 | */ | |
12cfbad9 | 1270 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1271 | if (!node) { |
1272 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1273 | if (!prealloc) { |
1274 | err = -ENOMEM; | |
1275 | goto out; | |
1276 | } | |
12cfbad9 | 1277 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1278 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1279 | if (err) |
1280 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1281 | cache_state(prealloc, cached_state); |
1282 | prealloc = NULL; | |
462d6fac JB |
1283 | goto out; |
1284 | } | |
1285 | state = rb_entry(node, struct extent_state, rb_node); | |
1286 | hit_next: | |
1287 | last_start = state->start; | |
1288 | last_end = state->end; | |
1289 | ||
1290 | /* | |
1291 | * | ---- desired range ---- | | |
1292 | * | state | | |
1293 | * | |
1294 | * Just lock what we found and keep going | |
1295 | */ | |
1296 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1297 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1298 | cache_state(state, cached_state); |
fefdc557 | 1299 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1300 | if (last_end == (u64)-1) |
1301 | goto out; | |
462d6fac | 1302 | start = last_end + 1; |
d1ac6e41 LB |
1303 | if (start < end && state && state->start == start && |
1304 | !need_resched()) | |
1305 | goto hit_next; | |
462d6fac JB |
1306 | goto search_again; |
1307 | } | |
1308 | ||
1309 | /* | |
1310 | * | ---- desired range ---- | | |
1311 | * | state | | |
1312 | * or | |
1313 | * | ------------- state -------------- | | |
1314 | * | |
1315 | * We need to split the extent we found, and may flip bits on | |
1316 | * second half. | |
1317 | * | |
1318 | * If the extent we found extends past our | |
1319 | * range, we just split and search again. It'll get split | |
1320 | * again the next time though. | |
1321 | * | |
1322 | * If the extent we found is inside our range, we set the | |
1323 | * desired bit on it. | |
1324 | */ | |
1325 | if (state->start < start) { | |
1326 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1327 | if (!prealloc) { |
1328 | err = -ENOMEM; | |
1329 | goto out; | |
1330 | } | |
462d6fac | 1331 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1332 | if (err) |
1333 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1334 | prealloc = NULL; |
1335 | if (err) | |
1336 | goto out; | |
1337 | if (state->end <= end) { | |
d38ed27f | 1338 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1339 | cache_state(state, cached_state); |
fefdc557 QW |
1340 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1341 | NULL); | |
462d6fac JB |
1342 | if (last_end == (u64)-1) |
1343 | goto out; | |
1344 | start = last_end + 1; | |
d1ac6e41 LB |
1345 | if (start < end && state && state->start == start && |
1346 | !need_resched()) | |
1347 | goto hit_next; | |
462d6fac JB |
1348 | } |
1349 | goto search_again; | |
1350 | } | |
1351 | /* | |
1352 | * | ---- desired range ---- | | |
1353 | * | state | or | state | | |
1354 | * | |
1355 | * There's a hole, we need to insert something in it and | |
1356 | * ignore the extent we found. | |
1357 | */ | |
1358 | if (state->start > start) { | |
1359 | u64 this_end; | |
1360 | if (end < last_start) | |
1361 | this_end = end; | |
1362 | else | |
1363 | this_end = last_start - 1; | |
1364 | ||
1365 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1366 | if (!prealloc) { |
1367 | err = -ENOMEM; | |
1368 | goto out; | |
1369 | } | |
462d6fac JB |
1370 | |
1371 | /* | |
1372 | * Avoid to free 'prealloc' if it can be merged with | |
1373 | * the later extent. | |
1374 | */ | |
1375 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1376 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1377 | if (err) |
1378 | extent_io_tree_panic(tree, err); | |
e6138876 | 1379 | cache_state(prealloc, cached_state); |
462d6fac JB |
1380 | prealloc = NULL; |
1381 | start = this_end + 1; | |
1382 | goto search_again; | |
1383 | } | |
1384 | /* | |
1385 | * | ---- desired range ---- | | |
1386 | * | state | | |
1387 | * We need to split the extent, and set the bit | |
1388 | * on the first half | |
1389 | */ | |
1390 | if (state->start <= end && state->end > end) { | |
1391 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1392 | if (!prealloc) { |
1393 | err = -ENOMEM; | |
1394 | goto out; | |
1395 | } | |
462d6fac JB |
1396 | |
1397 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1398 | if (err) |
1399 | extent_io_tree_panic(tree, err); | |
462d6fac | 1400 | |
d38ed27f | 1401 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1402 | cache_state(prealloc, cached_state); |
fefdc557 | 1403 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1404 | prealloc = NULL; |
1405 | goto out; | |
1406 | } | |
1407 | ||
462d6fac JB |
1408 | search_again: |
1409 | if (start > end) | |
1410 | goto out; | |
1411 | spin_unlock(&tree->lock); | |
210aa277 | 1412 | cond_resched(); |
c8fd3de7 | 1413 | first_iteration = false; |
462d6fac | 1414 | goto again; |
462d6fac JB |
1415 | |
1416 | out: | |
1417 | spin_unlock(&tree->lock); | |
1418 | if (prealloc) | |
1419 | free_extent_state(prealloc); | |
1420 | ||
1421 | return err; | |
462d6fac JB |
1422 | } |
1423 | ||
d1310b2e | 1424 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1425 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1426 | u32 bits, struct extent_changeset *changeset) |
d38ed27f QW |
1427 | { |
1428 | /* | |
1429 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1430 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1431 | * either fail with -EEXIST or changeset will record the whole | |
1432 | * range. | |
1433 | */ | |
1434 | BUG_ON(bits & EXTENT_LOCKED); | |
1435 | ||
1cab5e72 NB |
1436 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
1437 | changeset); | |
d38ed27f QW |
1438 | } |
1439 | ||
4ca73656 | 1440 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1441 | u32 bits) |
4ca73656 | 1442 | { |
1cab5e72 NB |
1443 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, |
1444 | GFP_NOWAIT, NULL); | |
4ca73656 NB |
1445 | } |
1446 | ||
fefdc557 | 1447 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1448 | u32 bits, int wake, int delete, |
ae0f1625 | 1449 | struct extent_state **cached) |
fefdc557 QW |
1450 | { |
1451 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1452 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1453 | } |
1454 | ||
fefdc557 | 1455 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1456 | u32 bits, struct extent_changeset *changeset) |
fefdc557 QW |
1457 | { |
1458 | /* | |
1459 | * Don't support EXTENT_LOCKED case, same reason as | |
1460 | * set_record_extent_bits(). | |
1461 | */ | |
1462 | BUG_ON(bits & EXTENT_LOCKED); | |
1463 | ||
f734c44a | 1464 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1465 | changeset); |
1466 | } | |
1467 | ||
d352ac68 CM |
1468 | /* |
1469 | * either insert or lock state struct between start and end use mask to tell | |
1470 | * us if waiting is desired. | |
1471 | */ | |
1edbb734 | 1472 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1473 | struct extent_state **cached_state) |
d1310b2e CM |
1474 | { |
1475 | int err; | |
1476 | u64 failed_start; | |
9ee49a04 | 1477 | |
d1310b2e | 1478 | while (1) { |
1cab5e72 NB |
1479 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, |
1480 | EXTENT_LOCKED, &failed_start, | |
1481 | cached_state, GFP_NOFS, NULL); | |
d0082371 | 1482 | if (err == -EEXIST) { |
d1310b2e CM |
1483 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1484 | start = failed_start; | |
d0082371 | 1485 | } else |
d1310b2e | 1486 | break; |
d1310b2e CM |
1487 | WARN_ON(start > end); |
1488 | } | |
1489 | return err; | |
1490 | } | |
d1310b2e | 1491 | |
d0082371 | 1492 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1493 | { |
1494 | int err; | |
1495 | u64 failed_start; | |
1496 | ||
1cab5e72 NB |
1497 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1498 | &failed_start, NULL, GFP_NOFS, NULL); | |
6643558d YZ |
1499 | if (err == -EEXIST) { |
1500 | if (failed_start > start) | |
1501 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1502 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1503 | return 0; |
6643558d | 1504 | } |
25179201 JB |
1505 | return 1; |
1506 | } | |
25179201 | 1507 | |
bd1fa4f0 | 1508 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1509 | { |
09cbfeaf KS |
1510 | unsigned long index = start >> PAGE_SHIFT; |
1511 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1512 | struct page *page; |
1513 | ||
1514 | while (index <= end_index) { | |
1515 | page = find_get_page(inode->i_mapping, index); | |
1516 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1517 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1518 | put_page(page); |
4adaa611 CM |
1519 | index++; |
1520 | } | |
4adaa611 CM |
1521 | } |
1522 | ||
f6311572 | 1523 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1524 | { |
ebf55c88 | 1525 | struct address_space *mapping = inode->i_mapping; |
09cbfeaf KS |
1526 | unsigned long index = start >> PAGE_SHIFT; |
1527 | unsigned long end_index = end >> PAGE_SHIFT; | |
ebf55c88 | 1528 | struct folio *folio; |
4adaa611 CM |
1529 | |
1530 | while (index <= end_index) { | |
ebf55c88 MWO |
1531 | folio = filemap_get_folio(mapping, index); |
1532 | filemap_dirty_folio(mapping, folio); | |
1533 | folio_account_redirty(folio); | |
1534 | index += folio_nr_pages(folio); | |
1535 | folio_put(folio); | |
4adaa611 | 1536 | } |
4adaa611 CM |
1537 | } |
1538 | ||
d352ac68 CM |
1539 | /* find the first state struct with 'bits' set after 'start', and |
1540 | * return it. tree->lock must be held. NULL will returned if | |
1541 | * nothing was found after 'start' | |
1542 | */ | |
48a3b636 | 1543 | static struct extent_state * |
f97e27e9 | 1544 | find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits) |
d7fc640e CM |
1545 | { |
1546 | struct rb_node *node; | |
1547 | struct extent_state *state; | |
1548 | ||
1549 | /* | |
1550 | * this search will find all the extents that end after | |
1551 | * our range starts. | |
1552 | */ | |
1553 | node = tree_search(tree, start); | |
d397712b | 1554 | if (!node) |
d7fc640e | 1555 | goto out; |
d7fc640e | 1556 | |
d397712b | 1557 | while (1) { |
d7fc640e | 1558 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1559 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1560 | return state; |
d397712b | 1561 | |
d7fc640e CM |
1562 | node = rb_next(node); |
1563 | if (!node) | |
1564 | break; | |
1565 | } | |
1566 | out: | |
1567 | return NULL; | |
1568 | } | |
d7fc640e | 1569 | |
69261c4b | 1570 | /* |
03509b78 | 1571 | * Find the first offset in the io tree with one or more @bits set. |
69261c4b | 1572 | * |
03509b78 QW |
1573 | * Note: If there are multiple bits set in @bits, any of them will match. |
1574 | * | |
1575 | * Return 0 if we find something, and update @start_ret and @end_ret. | |
1576 | * Return 1 if we found nothing. | |
69261c4b XG |
1577 | */ |
1578 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1579 | u64 *start_ret, u64 *end_ret, u32 bits, |
e6138876 | 1580 | struct extent_state **cached_state) |
69261c4b XG |
1581 | { |
1582 | struct extent_state *state; | |
1583 | int ret = 1; | |
1584 | ||
1585 | spin_lock(&tree->lock); | |
e6138876 JB |
1586 | if (cached_state && *cached_state) { |
1587 | state = *cached_state; | |
27a3507d | 1588 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1589 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1590 | if (state->state & bits) |
1591 | goto got_it; | |
e6138876 JB |
1592 | } |
1593 | free_extent_state(*cached_state); | |
1594 | *cached_state = NULL; | |
1595 | goto out; | |
1596 | } | |
1597 | free_extent_state(*cached_state); | |
1598 | *cached_state = NULL; | |
1599 | } | |
1600 | ||
69261c4b | 1601 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1602 | got_it: |
69261c4b | 1603 | if (state) { |
e38e2ed7 | 1604 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1605 | *start_ret = state->start; |
1606 | *end_ret = state->end; | |
1607 | ret = 0; | |
1608 | } | |
e6138876 | 1609 | out: |
69261c4b XG |
1610 | spin_unlock(&tree->lock); |
1611 | return ret; | |
1612 | } | |
1613 | ||
41a2ee75 | 1614 | /** |
3bed2da1 NB |
1615 | * Find a contiguous area of bits |
1616 | * | |
1617 | * @tree: io tree to check | |
1618 | * @start: offset to start the search from | |
1619 | * @start_ret: the first offset we found with the bits set | |
1620 | * @end_ret: the final contiguous range of the bits that were set | |
1621 | * @bits: bits to look for | |
41a2ee75 JB |
1622 | * |
1623 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1624 | * to set bits appropriately, and then merge them again. During this time it | |
1625 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1626 | * contiguous area for given bits. We will search to the first bit we find, and | |
1627 | * then walk down the tree until we find a non-contiguous area. The area | |
1628 | * returned will be the full contiguous area with the bits set. | |
1629 | */ | |
1630 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1631 | u64 *start_ret, u64 *end_ret, u32 bits) |
41a2ee75 JB |
1632 | { |
1633 | struct extent_state *state; | |
1634 | int ret = 1; | |
1635 | ||
1636 | spin_lock(&tree->lock); | |
1637 | state = find_first_extent_bit_state(tree, start, bits); | |
1638 | if (state) { | |
1639 | *start_ret = state->start; | |
1640 | *end_ret = state->end; | |
1641 | while ((state = next_state(state)) != NULL) { | |
1642 | if (state->start > (*end_ret + 1)) | |
1643 | break; | |
1644 | *end_ret = state->end; | |
1645 | } | |
1646 | ret = 0; | |
1647 | } | |
1648 | spin_unlock(&tree->lock); | |
1649 | return ret; | |
1650 | } | |
1651 | ||
45bfcfc1 | 1652 | /** |
3bed2da1 NB |
1653 | * Find the first range that has @bits not set. This range could start before |
1654 | * @start. | |
45bfcfc1 | 1655 | * |
3bed2da1 NB |
1656 | * @tree: the tree to search |
1657 | * @start: offset at/after which the found extent should start | |
1658 | * @start_ret: records the beginning of the range | |
1659 | * @end_ret: records the end of the range (inclusive) | |
1660 | * @bits: the set of bits which must be unset | |
45bfcfc1 NB |
1661 | * |
1662 | * Since unallocated range is also considered one which doesn't have the bits | |
1663 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1664 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1665 | * trim @end_ret to the appropriate size. | |
1666 | */ | |
1667 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1668 | u64 *start_ret, u64 *end_ret, u32 bits) |
45bfcfc1 NB |
1669 | { |
1670 | struct extent_state *state; | |
1671 | struct rb_node *node, *prev = NULL, *next; | |
1672 | ||
1673 | spin_lock(&tree->lock); | |
1674 | ||
1675 | /* Find first extent with bits cleared */ | |
1676 | while (1) { | |
1677 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
5750c375 NB |
1678 | if (!node && !next && !prev) { |
1679 | /* | |
1680 | * Tree is completely empty, send full range and let | |
1681 | * caller deal with it | |
1682 | */ | |
1683 | *start_ret = 0; | |
1684 | *end_ret = -1; | |
1685 | goto out; | |
1686 | } else if (!node && !next) { | |
1687 | /* | |
1688 | * We are past the last allocated chunk, set start at | |
1689 | * the end of the last extent. | |
1690 | */ | |
1691 | state = rb_entry(prev, struct extent_state, rb_node); | |
1692 | *start_ret = state->end + 1; | |
1693 | *end_ret = -1; | |
1694 | goto out; | |
1695 | } else if (!node) { | |
45bfcfc1 | 1696 | node = next; |
45bfcfc1 | 1697 | } |
1eaebb34 NB |
1698 | /* |
1699 | * At this point 'node' either contains 'start' or start is | |
1700 | * before 'node' | |
1701 | */ | |
45bfcfc1 | 1702 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1703 | |
1704 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1705 | if (state->state & bits) { | |
1706 | /* | |
1707 | * |--range with bits sets--| | |
1708 | * | | |
1709 | * start | |
1710 | */ | |
1711 | start = state->end + 1; | |
1712 | } else { | |
1713 | /* | |
1714 | * 'start' falls within a range that doesn't | |
1715 | * have the bits set, so take its start as | |
1716 | * the beginning of the desired range | |
1717 | * | |
1718 | * |--range with bits cleared----| | |
1719 | * | | |
1720 | * start | |
1721 | */ | |
1722 | *start_ret = state->start; | |
1723 | break; | |
1724 | } | |
45bfcfc1 | 1725 | } else { |
1eaebb34 NB |
1726 | /* |
1727 | * |---prev range---|---hole/unset---|---node range---| | |
1728 | * | | |
1729 | * start | |
1730 | * | |
1731 | * or | |
1732 | * | |
1733 | * |---hole/unset--||--first node--| | |
1734 | * 0 | | |
1735 | * start | |
1736 | */ | |
1737 | if (prev) { | |
1738 | state = rb_entry(prev, struct extent_state, | |
1739 | rb_node); | |
1740 | *start_ret = state->end + 1; | |
1741 | } else { | |
1742 | *start_ret = 0; | |
1743 | } | |
45bfcfc1 NB |
1744 | break; |
1745 | } | |
1746 | } | |
1747 | ||
1748 | /* | |
1749 | * Find the longest stretch from start until an entry which has the | |
1750 | * bits set | |
1751 | */ | |
1752 | while (1) { | |
1753 | state = rb_entry(node, struct extent_state, rb_node); | |
1754 | if (state->end >= start && !(state->state & bits)) { | |
1755 | *end_ret = state->end; | |
1756 | } else { | |
1757 | *end_ret = state->start - 1; | |
1758 | break; | |
1759 | } | |
1760 | ||
1761 | node = rb_next(node); | |
1762 | if (!node) | |
1763 | break; | |
1764 | } | |
1765 | out: | |
1766 | spin_unlock(&tree->lock); | |
1767 | } | |
1768 | ||
d352ac68 CM |
1769 | /* |
1770 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1771 | * more than 'max_bytes'. start and end are used to return the range, | |
1772 | * | |
3522e903 | 1773 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1774 | */ |
083e75e7 JB |
1775 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1776 | u64 *end, u64 max_bytes, | |
1777 | struct extent_state **cached_state) | |
d1310b2e CM |
1778 | { |
1779 | struct rb_node *node; | |
1780 | struct extent_state *state; | |
1781 | u64 cur_start = *start; | |
3522e903 | 1782 | bool found = false; |
d1310b2e CM |
1783 | u64 total_bytes = 0; |
1784 | ||
cad321ad | 1785 | spin_lock(&tree->lock); |
c8b97818 | 1786 | |
d1310b2e CM |
1787 | /* |
1788 | * this search will find all the extents that end after | |
1789 | * our range starts. | |
1790 | */ | |
80ea96b1 | 1791 | node = tree_search(tree, cur_start); |
2b114d1d | 1792 | if (!node) { |
3522e903 | 1793 | *end = (u64)-1; |
d1310b2e CM |
1794 | goto out; |
1795 | } | |
1796 | ||
d397712b | 1797 | while (1) { |
d1310b2e | 1798 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1799 | if (found && (state->start != cur_start || |
1800 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1801 | goto out; |
1802 | } | |
1803 | if (!(state->state & EXTENT_DELALLOC)) { | |
1804 | if (!found) | |
1805 | *end = state->end; | |
1806 | goto out; | |
1807 | } | |
c2a128d2 | 1808 | if (!found) { |
d1310b2e | 1809 | *start = state->start; |
c2a128d2 | 1810 | *cached_state = state; |
b7ac31b7 | 1811 | refcount_inc(&state->refs); |
c2a128d2 | 1812 | } |
3522e903 | 1813 | found = true; |
d1310b2e CM |
1814 | *end = state->end; |
1815 | cur_start = state->end + 1; | |
1816 | node = rb_next(node); | |
d1310b2e | 1817 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1818 | if (total_bytes >= max_bytes) |
573aecaf | 1819 | break; |
573aecaf | 1820 | if (!node) |
d1310b2e CM |
1821 | break; |
1822 | } | |
1823 | out: | |
cad321ad | 1824 | spin_unlock(&tree->lock); |
d1310b2e CM |
1825 | return found; |
1826 | } | |
1827 | ||
ed8f13bf QW |
1828 | /* |
1829 | * Process one page for __process_pages_contig(). | |
1830 | * | |
1831 | * Return >0 if we hit @page == @locked_page. | |
1832 | * Return 0 if we updated the page status. | |
1833 | * Return -EGAIN if the we need to try again. | |
1834 | * (For PAGE_LOCK case but got dirty page or page not belong to mapping) | |
1835 | */ | |
e38992be QW |
1836 | static int process_one_page(struct btrfs_fs_info *fs_info, |
1837 | struct address_space *mapping, | |
ed8f13bf | 1838 | struct page *page, struct page *locked_page, |
e38992be | 1839 | unsigned long page_ops, u64 start, u64 end) |
ed8f13bf | 1840 | { |
e38992be QW |
1841 | u32 len; |
1842 | ||
1843 | ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX); | |
1844 | len = end + 1 - start; | |
1845 | ||
ed8f13bf | 1846 | if (page_ops & PAGE_SET_ORDERED) |
b945a463 | 1847 | btrfs_page_clamp_set_ordered(fs_info, page, start, len); |
ed8f13bf | 1848 | if (page_ops & PAGE_SET_ERROR) |
e38992be | 1849 | btrfs_page_clamp_set_error(fs_info, page, start, len); |
ed8f13bf | 1850 | if (page_ops & PAGE_START_WRITEBACK) { |
e38992be QW |
1851 | btrfs_page_clamp_clear_dirty(fs_info, page, start, len); |
1852 | btrfs_page_clamp_set_writeback(fs_info, page, start, len); | |
ed8f13bf QW |
1853 | } |
1854 | if (page_ops & PAGE_END_WRITEBACK) | |
e38992be | 1855 | btrfs_page_clamp_clear_writeback(fs_info, page, start, len); |
a33a8e9a QW |
1856 | |
1857 | if (page == locked_page) | |
1858 | return 1; | |
1859 | ||
ed8f13bf | 1860 | if (page_ops & PAGE_LOCK) { |
1e1de387 QW |
1861 | int ret; |
1862 | ||
1863 | ret = btrfs_page_start_writer_lock(fs_info, page, start, len); | |
1864 | if (ret) | |
1865 | return ret; | |
ed8f13bf | 1866 | if (!PageDirty(page) || page->mapping != mapping) { |
1e1de387 | 1867 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1868 | return -EAGAIN; |
1869 | } | |
1870 | } | |
1871 | if (page_ops & PAGE_UNLOCK) | |
1e1de387 | 1872 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1873 | return 0; |
1874 | } | |
1875 | ||
da2c7009 LB |
1876 | static int __process_pages_contig(struct address_space *mapping, |
1877 | struct page *locked_page, | |
98af9ab1 | 1878 | u64 start, u64 end, unsigned long page_ops, |
ed8f13bf QW |
1879 | u64 *processed_end) |
1880 | { | |
e38992be | 1881 | struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); |
ed8f13bf QW |
1882 | pgoff_t start_index = start >> PAGE_SHIFT; |
1883 | pgoff_t end_index = end >> PAGE_SHIFT; | |
1884 | pgoff_t index = start_index; | |
1885 | unsigned long nr_pages = end_index - start_index + 1; | |
1886 | unsigned long pages_processed = 0; | |
1887 | struct page *pages[16]; | |
1888 | int err = 0; | |
1889 | int i; | |
1890 | ||
1891 | if (page_ops & PAGE_LOCK) { | |
1892 | ASSERT(page_ops == PAGE_LOCK); | |
1893 | ASSERT(processed_end && *processed_end == start); | |
1894 | } | |
1895 | ||
1896 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) | |
1897 | mapping_set_error(mapping, -EIO); | |
1898 | ||
1899 | while (nr_pages > 0) { | |
1900 | int found_pages; | |
1901 | ||
1902 | found_pages = find_get_pages_contig(mapping, index, | |
1903 | min_t(unsigned long, | |
1904 | nr_pages, ARRAY_SIZE(pages)), pages); | |
1905 | if (found_pages == 0) { | |
1906 | /* | |
1907 | * Only if we're going to lock these pages, we can find | |
1908 | * nothing at @index. | |
1909 | */ | |
1910 | ASSERT(page_ops & PAGE_LOCK); | |
1911 | err = -EAGAIN; | |
1912 | goto out; | |
1913 | } | |
1914 | ||
1915 | for (i = 0; i < found_pages; i++) { | |
1916 | int process_ret; | |
1917 | ||
e38992be QW |
1918 | process_ret = process_one_page(fs_info, mapping, |
1919 | pages[i], locked_page, page_ops, | |
1920 | start, end); | |
ed8f13bf QW |
1921 | if (process_ret < 0) { |
1922 | for (; i < found_pages; i++) | |
1923 | put_page(pages[i]); | |
1924 | err = -EAGAIN; | |
1925 | goto out; | |
1926 | } | |
1927 | put_page(pages[i]); | |
1928 | pages_processed++; | |
1929 | } | |
1930 | nr_pages -= found_pages; | |
1931 | index += found_pages; | |
1932 | cond_resched(); | |
1933 | } | |
1934 | out: | |
1935 | if (err && processed_end) { | |
1936 | /* | |
1937 | * Update @processed_end. I know this is awful since it has | |
1938 | * two different return value patterns (inclusive vs exclusive). | |
1939 | * | |
1940 | * But the exclusive pattern is necessary if @start is 0, or we | |
1941 | * underflow and check against processed_end won't work as | |
1942 | * expected. | |
1943 | */ | |
1944 | if (pages_processed) | |
1945 | *processed_end = min(end, | |
1946 | ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1); | |
1947 | else | |
1948 | *processed_end = start; | |
1949 | } | |
1950 | return err; | |
1951 | } | |
da2c7009 | 1952 | |
143bede5 JM |
1953 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1954 | struct page *locked_page, | |
1955 | u64 start, u64 end) | |
c8b97818 | 1956 | { |
09cbfeaf KS |
1957 | unsigned long index = start >> PAGE_SHIFT; |
1958 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1959 | |
76c0021d | 1960 | ASSERT(locked_page); |
c8b97818 | 1961 | if (index == locked_page->index && end_index == index) |
143bede5 | 1962 | return; |
c8b97818 | 1963 | |
98af9ab1 | 1964 | __process_pages_contig(inode->i_mapping, locked_page, start, end, |
76c0021d | 1965 | PAGE_UNLOCK, NULL); |
c8b97818 CM |
1966 | } |
1967 | ||
1968 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1969 | struct page *locked_page, | |
1970 | u64 delalloc_start, | |
1971 | u64 delalloc_end) | |
1972 | { | |
09cbfeaf | 1973 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
09cbfeaf | 1974 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
98af9ab1 | 1975 | u64 processed_end = delalloc_start; |
c8b97818 | 1976 | int ret; |
c8b97818 | 1977 | |
76c0021d | 1978 | ASSERT(locked_page); |
c8b97818 CM |
1979 | if (index == locked_page->index && index == end_index) |
1980 | return 0; | |
1981 | ||
98af9ab1 QW |
1982 | ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start, |
1983 | delalloc_end, PAGE_LOCK, &processed_end); | |
1984 | if (ret == -EAGAIN && processed_end > delalloc_start) | |
76c0021d | 1985 | __unlock_for_delalloc(inode, locked_page, delalloc_start, |
98af9ab1 | 1986 | processed_end); |
c8b97818 CM |
1987 | return ret; |
1988 | } | |
1989 | ||
1990 | /* | |
3522e903 | 1991 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
2749f7ef | 1992 | * more than @max_bytes. |
c8b97818 | 1993 | * |
2749f7ef QW |
1994 | * @start: The original start bytenr to search. |
1995 | * Will store the extent range start bytenr. | |
1996 | * @end: The original end bytenr of the search range | |
1997 | * Will store the extent range end bytenr. | |
1998 | * | |
1999 | * Return true if we find a delalloc range which starts inside the original | |
2000 | * range, and @start/@end will store the delalloc range start/end. | |
2001 | * | |
2002 | * Return false if we can't find any delalloc range which starts inside the | |
2003 | * original range, and @start/@end will be the non-delalloc range start/end. | |
c8b97818 | 2004 | */ |
ce9f967f | 2005 | EXPORT_FOR_TESTS |
3522e903 | 2006 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 2007 | struct page *locked_page, u64 *start, |
917aacec | 2008 | u64 *end) |
c8b97818 | 2009 | { |
9978059b | 2010 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
2749f7ef QW |
2011 | const u64 orig_start = *start; |
2012 | const u64 orig_end = *end; | |
917aacec | 2013 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
2014 | u64 delalloc_start; |
2015 | u64 delalloc_end; | |
3522e903 | 2016 | bool found; |
9655d298 | 2017 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
2018 | int ret; |
2019 | int loops = 0; | |
2020 | ||
2749f7ef QW |
2021 | /* Caller should pass a valid @end to indicate the search range end */ |
2022 | ASSERT(orig_end > orig_start); | |
2023 | ||
2024 | /* The range should at least cover part of the page */ | |
2025 | ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE || | |
2026 | orig_end <= page_offset(locked_page))); | |
c8b97818 CM |
2027 | again: |
2028 | /* step one, find a bunch of delalloc bytes starting at start */ | |
2029 | delalloc_start = *start; | |
2030 | delalloc_end = 0; | |
083e75e7 JB |
2031 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
2032 | max_bytes, &cached_state); | |
2749f7ef | 2033 | if (!found || delalloc_end <= *start || delalloc_start > orig_end) { |
c8b97818 | 2034 | *start = delalloc_start; |
2749f7ef QW |
2035 | |
2036 | /* @delalloc_end can be -1, never go beyond @orig_end */ | |
2037 | *end = min(delalloc_end, orig_end); | |
c2a128d2 | 2038 | free_extent_state(cached_state); |
3522e903 | 2039 | return false; |
c8b97818 CM |
2040 | } |
2041 | ||
70b99e69 CM |
2042 | /* |
2043 | * start comes from the offset of locked_page. We have to lock | |
2044 | * pages in order, so we can't process delalloc bytes before | |
2045 | * locked_page | |
2046 | */ | |
d397712b | 2047 | if (delalloc_start < *start) |
70b99e69 | 2048 | delalloc_start = *start; |
70b99e69 | 2049 | |
c8b97818 CM |
2050 | /* |
2051 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 2052 | */ |
7bf811a5 JB |
2053 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
2054 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 2055 | |
c8b97818 CM |
2056 | /* step two, lock all the pages after the page that has start */ |
2057 | ret = lock_delalloc_pages(inode, locked_page, | |
2058 | delalloc_start, delalloc_end); | |
9bfd61d9 | 2059 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
2060 | if (ret == -EAGAIN) { |
2061 | /* some of the pages are gone, lets avoid looping by | |
2062 | * shortening the size of the delalloc range we're searching | |
2063 | */ | |
9655d298 | 2064 | free_extent_state(cached_state); |
7d788742 | 2065 | cached_state = NULL; |
c8b97818 | 2066 | if (!loops) { |
09cbfeaf | 2067 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
2068 | loops = 1; |
2069 | goto again; | |
2070 | } else { | |
3522e903 | 2071 | found = false; |
c8b97818 CM |
2072 | goto out_failed; |
2073 | } | |
2074 | } | |
c8b97818 CM |
2075 | |
2076 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 2077 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
2078 | |
2079 | /* then test to make sure it is all still delalloc */ | |
2080 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 2081 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 2082 | if (!ret) { |
9655d298 | 2083 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 2084 | &cached_state); |
c8b97818 CM |
2085 | __unlock_for_delalloc(inode, locked_page, |
2086 | delalloc_start, delalloc_end); | |
2087 | cond_resched(); | |
2088 | goto again; | |
2089 | } | |
9655d298 | 2090 | free_extent_state(cached_state); |
c8b97818 CM |
2091 | *start = delalloc_start; |
2092 | *end = delalloc_end; | |
2093 | out_failed: | |
2094 | return found; | |
2095 | } | |
2096 | ||
ad7ff17b | 2097 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 2098 | struct page *locked_page, |
f97e27e9 | 2099 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 2100 | { |
ad7ff17b | 2101 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL); |
873695b3 | 2102 | |
ad7ff17b | 2103 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
98af9ab1 | 2104 | start, end, page_ops, NULL); |
873695b3 LB |
2105 | } |
2106 | ||
d352ac68 CM |
2107 | /* |
2108 | * count the number of bytes in the tree that have a given bit(s) | |
2109 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2110 | * cached. The total number found is returned. | |
2111 | */ | |
d1310b2e CM |
2112 | u64 count_range_bits(struct extent_io_tree *tree, |
2113 | u64 *start, u64 search_end, u64 max_bytes, | |
f97e27e9 | 2114 | u32 bits, int contig) |
d1310b2e CM |
2115 | { |
2116 | struct rb_node *node; | |
2117 | struct extent_state *state; | |
2118 | u64 cur_start = *start; | |
2119 | u64 total_bytes = 0; | |
ec29ed5b | 2120 | u64 last = 0; |
d1310b2e CM |
2121 | int found = 0; |
2122 | ||
fae7f21c | 2123 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2124 | return 0; |
d1310b2e | 2125 | |
cad321ad | 2126 | spin_lock(&tree->lock); |
d1310b2e CM |
2127 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2128 | total_bytes = tree->dirty_bytes; | |
2129 | goto out; | |
2130 | } | |
2131 | /* | |
2132 | * this search will find all the extents that end after | |
2133 | * our range starts. | |
2134 | */ | |
80ea96b1 | 2135 | node = tree_search(tree, cur_start); |
d397712b | 2136 | if (!node) |
d1310b2e | 2137 | goto out; |
d1310b2e | 2138 | |
d397712b | 2139 | while (1) { |
d1310b2e CM |
2140 | state = rb_entry(node, struct extent_state, rb_node); |
2141 | if (state->start > search_end) | |
2142 | break; | |
ec29ed5b CM |
2143 | if (contig && found && state->start > last + 1) |
2144 | break; | |
2145 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2146 | total_bytes += min(search_end, state->end) + 1 - |
2147 | max(cur_start, state->start); | |
2148 | if (total_bytes >= max_bytes) | |
2149 | break; | |
2150 | if (!found) { | |
af60bed2 | 2151 | *start = max(cur_start, state->start); |
d1310b2e CM |
2152 | found = 1; |
2153 | } | |
ec29ed5b CM |
2154 | last = state->end; |
2155 | } else if (contig && found) { | |
2156 | break; | |
d1310b2e CM |
2157 | } |
2158 | node = rb_next(node); | |
2159 | if (!node) | |
2160 | break; | |
2161 | } | |
2162 | out: | |
cad321ad | 2163 | spin_unlock(&tree->lock); |
d1310b2e CM |
2164 | return total_bytes; |
2165 | } | |
b2950863 | 2166 | |
d352ac68 CM |
2167 | /* |
2168 | * set the private field for a given byte offset in the tree. If there isn't | |
2169 | * an extent_state there already, this does nothing. | |
2170 | */ | |
b3f167aa JB |
2171 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2172 | struct io_failure_record *failrec) | |
d1310b2e CM |
2173 | { |
2174 | struct rb_node *node; | |
2175 | struct extent_state *state; | |
2176 | int ret = 0; | |
2177 | ||
cad321ad | 2178 | spin_lock(&tree->lock); |
d1310b2e CM |
2179 | /* |
2180 | * this search will find all the extents that end after | |
2181 | * our range starts. | |
2182 | */ | |
80ea96b1 | 2183 | node = tree_search(tree, start); |
2b114d1d | 2184 | if (!node) { |
d1310b2e CM |
2185 | ret = -ENOENT; |
2186 | goto out; | |
2187 | } | |
2188 | state = rb_entry(node, struct extent_state, rb_node); | |
2189 | if (state->start != start) { | |
2190 | ret = -ENOENT; | |
2191 | goto out; | |
2192 | } | |
47dc196a | 2193 | state->failrec = failrec; |
d1310b2e | 2194 | out: |
cad321ad | 2195 | spin_unlock(&tree->lock); |
d1310b2e CM |
2196 | return ret; |
2197 | } | |
2198 | ||
2279a270 | 2199 | struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start) |
d1310b2e CM |
2200 | { |
2201 | struct rb_node *node; | |
2202 | struct extent_state *state; | |
2279a270 | 2203 | struct io_failure_record *failrec; |
d1310b2e | 2204 | |
cad321ad | 2205 | spin_lock(&tree->lock); |
d1310b2e CM |
2206 | /* |
2207 | * this search will find all the extents that end after | |
2208 | * our range starts. | |
2209 | */ | |
80ea96b1 | 2210 | node = tree_search(tree, start); |
2b114d1d | 2211 | if (!node) { |
2279a270 | 2212 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2213 | goto out; |
2214 | } | |
2215 | state = rb_entry(node, struct extent_state, rb_node); | |
2216 | if (state->start != start) { | |
2279a270 | 2217 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2218 | goto out; |
2219 | } | |
2279a270 NB |
2220 | |
2221 | failrec = state->failrec; | |
d1310b2e | 2222 | out: |
cad321ad | 2223 | spin_unlock(&tree->lock); |
2279a270 | 2224 | return failrec; |
d1310b2e CM |
2225 | } |
2226 | ||
2227 | /* | |
2228 | * searches a range in the state tree for a given mask. | |
70dec807 | 2229 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2230 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2231 | * range is found set. | |
2232 | */ | |
2233 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 2234 | u32 bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2235 | { |
2236 | struct extent_state *state = NULL; | |
2237 | struct rb_node *node; | |
2238 | int bitset = 0; | |
d1310b2e | 2239 | |
cad321ad | 2240 | spin_lock(&tree->lock); |
27a3507d | 2241 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2242 | cached->end > start) |
9655d298 CM |
2243 | node = &cached->rb_node; |
2244 | else | |
2245 | node = tree_search(tree, start); | |
d1310b2e CM |
2246 | while (node && start <= end) { |
2247 | state = rb_entry(node, struct extent_state, rb_node); | |
2248 | ||
2249 | if (filled && state->start > start) { | |
2250 | bitset = 0; | |
2251 | break; | |
2252 | } | |
2253 | ||
2254 | if (state->start > end) | |
2255 | break; | |
2256 | ||
2257 | if (state->state & bits) { | |
2258 | bitset = 1; | |
2259 | if (!filled) | |
2260 | break; | |
2261 | } else if (filled) { | |
2262 | bitset = 0; | |
2263 | break; | |
2264 | } | |
46562cec CM |
2265 | |
2266 | if (state->end == (u64)-1) | |
2267 | break; | |
2268 | ||
d1310b2e CM |
2269 | start = state->end + 1; |
2270 | if (start > end) | |
2271 | break; | |
2272 | node = rb_next(node); | |
2273 | if (!node) { | |
2274 | if (filled) | |
2275 | bitset = 0; | |
2276 | break; | |
2277 | } | |
2278 | } | |
cad321ad | 2279 | spin_unlock(&tree->lock); |
d1310b2e CM |
2280 | return bitset; |
2281 | } | |
d1310b2e | 2282 | |
7870d082 JB |
2283 | int free_io_failure(struct extent_io_tree *failure_tree, |
2284 | struct extent_io_tree *io_tree, | |
2285 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2286 | { |
2287 | int ret; | |
2288 | int err = 0; | |
4a54c8c1 | 2289 | |
47dc196a | 2290 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2291 | ret = clear_extent_bits(failure_tree, rec->start, |
2292 | rec->start + rec->len - 1, | |
91166212 | 2293 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2294 | if (ret) |
2295 | err = ret; | |
2296 | ||
7870d082 | 2297 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2298 | rec->start + rec->len - 1, |
91166212 | 2299 | EXTENT_DAMAGED); |
53b381b3 DW |
2300 | if (ret && !err) |
2301 | err = ret; | |
4a54c8c1 JS |
2302 | |
2303 | kfree(rec); | |
2304 | return err; | |
2305 | } | |
2306 | ||
4a54c8c1 JS |
2307 | /* |
2308 | * this bypasses the standard btrfs submit functions deliberately, as | |
2309 | * the standard behavior is to write all copies in a raid setup. here we only | |
2310 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2311 | * submit_bio directly. | |
3ec706c8 | 2312 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2313 | * actually prevents the read that triggered the error from finishing. |
2314 | * currently, there can be no more than two copies of every data bit. thus, | |
2315 | * exactly one rewrite is required. | |
2316 | */ | |
38d5e541 QW |
2317 | static int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2318 | u64 length, u64 logical, struct page *page, | |
2319 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 | 2320 | { |
4a54c8c1 | 2321 | struct btrfs_device *dev; |
e9458bfe CH |
2322 | struct bio_vec bvec; |
2323 | struct bio bio; | |
4a54c8c1 JS |
2324 | u64 map_length = 0; |
2325 | u64 sector; | |
4c664611 | 2326 | struct btrfs_io_context *bioc = NULL; |
e9458bfe | 2327 | int ret = 0; |
4a54c8c1 | 2328 | |
1751e8a6 | 2329 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2330 | BUG_ON(!mirror_num); |
2331 | ||
554aed7d JT |
2332 | if (btrfs_repair_one_zone(fs_info, logical)) |
2333 | return 0; | |
f7ef5287 | 2334 | |
4a54c8c1 JS |
2335 | map_length = length; |
2336 | ||
b5de8d0d | 2337 | /* |
4c664611 | 2338 | * Avoid races with device replace and make sure our bioc has devices |
b5de8d0d FM |
2339 | * associated to its stripes that don't go away while we are doing the |
2340 | * read repair operation. | |
2341 | */ | |
2342 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2343 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2344 | /* |
2345 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2346 | * to update all raid stripes, but here we just want to correct | |
2347 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2348 | * stripe's dev and sector. | |
2349 | */ | |
2350 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
4c664611 | 2351 | &map_length, &bioc, 0); |
e9458bfe CH |
2352 | if (ret) |
2353 | goto out_counter_dec; | |
4c664611 | 2354 | ASSERT(bioc->mirror_num == 1); |
c725328c LB |
2355 | } else { |
2356 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
4c664611 | 2357 | &map_length, &bioc, mirror_num); |
e9458bfe CH |
2358 | if (ret) |
2359 | goto out_counter_dec; | |
4c664611 | 2360 | BUG_ON(mirror_num != bioc->mirror_num); |
4a54c8c1 | 2361 | } |
c725328c | 2362 | |
4c664611 | 2363 | sector = bioc->stripes[bioc->mirror_num - 1].physical >> 9; |
4c664611 QW |
2364 | dev = bioc->stripes[bioc->mirror_num - 1].dev; |
2365 | btrfs_put_bioc(bioc); | |
e9458bfe | 2366 | |
ebbede42 AJ |
2367 | if (!dev || !dev->bdev || |
2368 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
e9458bfe CH |
2369 | ret = -EIO; |
2370 | goto out_counter_dec; | |
4a54c8c1 | 2371 | } |
4a54c8c1 | 2372 | |
e9458bfe CH |
2373 | bio_init(&bio, dev->bdev, &bvec, 1, REQ_OP_WRITE | REQ_SYNC); |
2374 | bio.bi_iter.bi_sector = sector; | |
2375 | __bio_add_page(&bio, page, length, pg_offset); | |
2376 | ||
2377 | btrfsic_check_bio(&bio); | |
2378 | ret = submit_bio_wait(&bio); | |
2379 | if (ret) { | |
4a54c8c1 | 2380 | /* try to remap that extent elsewhere? */ |
442a4f63 | 2381 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
e9458bfe | 2382 | goto out_bio_uninit; |
4a54c8c1 JS |
2383 | } |
2384 | ||
b14af3b4 DS |
2385 | btrfs_info_rl_in_rcu(fs_info, |
2386 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2387 | ino, start, |
1203b681 | 2388 | rcu_str_deref(dev->name), sector); |
e9458bfe CH |
2389 | ret = 0; |
2390 | ||
2391 | out_bio_uninit: | |
2392 | bio_uninit(&bio); | |
2393 | out_counter_dec: | |
b5de8d0d | 2394 | btrfs_bio_counter_dec(fs_info); |
e9458bfe | 2395 | return ret; |
4a54c8c1 JS |
2396 | } |
2397 | ||
2b48966a | 2398 | int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num) |
ea466794 | 2399 | { |
20a1fbf9 | 2400 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2401 | u64 start = eb->start; |
cc5e31a4 | 2402 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2403 | int ret = 0; |
ea466794 | 2404 | |
bc98a42c | 2405 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2406 | return -EROFS; |
2407 | ||
ea466794 | 2408 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2409 | struct page *p = eb->pages[i]; |
1203b681 | 2410 | |
6ec656bc | 2411 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2412 | start - page_offset(p), mirror_num); |
ea466794 JB |
2413 | if (ret) |
2414 | break; | |
09cbfeaf | 2415 | start += PAGE_SIZE; |
ea466794 JB |
2416 | } |
2417 | ||
2418 | return ret; | |
2419 | } | |
2420 | ||
4a54c8c1 JS |
2421 | /* |
2422 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2423 | * to see if we need to process or clean up an io_failure_record | |
2424 | */ | |
7870d082 JB |
2425 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2426 | struct extent_io_tree *failure_tree, | |
2427 | struct extent_io_tree *io_tree, u64 start, | |
2428 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2429 | { |
2430 | u64 private; | |
4a54c8c1 | 2431 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2432 | struct extent_state *state; |
2433 | int num_copies; | |
4a54c8c1 | 2434 | int ret; |
4a54c8c1 JS |
2435 | |
2436 | private = 0; | |
7870d082 JB |
2437 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2438 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2439 | if (!ret) |
2440 | return 0; | |
2441 | ||
2279a270 NB |
2442 | failrec = get_state_failrec(failure_tree, start); |
2443 | if (IS_ERR(failrec)) | |
4a54c8c1 JS |
2444 | return 0; |
2445 | ||
4a54c8c1 JS |
2446 | BUG_ON(!failrec->this_mirror); |
2447 | ||
bc98a42c | 2448 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2449 | goto out; |
4a54c8c1 | 2450 | |
7870d082 JB |
2451 | spin_lock(&io_tree->lock); |
2452 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2453 | failrec->start, |
2454 | EXTENT_LOCKED); | |
7870d082 | 2455 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2456 | |
883d0de4 MX |
2457 | if (state && state->start <= failrec->start && |
2458 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2459 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2460 | failrec->len); | |
4a54c8c1 | 2461 | if (num_copies > 1) { |
7870d082 JB |
2462 | repair_io_failure(fs_info, ino, start, failrec->len, |
2463 | failrec->logical, page, pg_offset, | |
2464 | failrec->failed_mirror); | |
4a54c8c1 JS |
2465 | } |
2466 | } | |
2467 | ||
2468 | out: | |
7870d082 | 2469 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2470 | |
454ff3de | 2471 | return 0; |
4a54c8c1 JS |
2472 | } |
2473 | ||
f612496b MX |
2474 | /* |
2475 | * Can be called when | |
2476 | * - hold extent lock | |
2477 | * - under ordered extent | |
2478 | * - the inode is freeing | |
2479 | */ | |
7ab7956e | 2480 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2481 | { |
7ab7956e | 2482 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2483 | struct io_failure_record *failrec; |
2484 | struct extent_state *state, *next; | |
2485 | ||
2486 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2487 | return; | |
2488 | ||
2489 | spin_lock(&failure_tree->lock); | |
2490 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2491 | while (state) { | |
2492 | if (state->start > end) | |
2493 | break; | |
2494 | ||
2495 | ASSERT(state->end <= end); | |
2496 | ||
2497 | next = next_state(state); | |
2498 | ||
47dc196a | 2499 | failrec = state->failrec; |
f612496b MX |
2500 | free_extent_state(state); |
2501 | kfree(failrec); | |
2502 | ||
2503 | state = next; | |
2504 | } | |
2505 | spin_unlock(&failure_tree->lock); | |
2506 | } | |
2507 | ||
3526302f | 2508 | static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode, |
150e4b05 | 2509 | u64 start) |
4a54c8c1 | 2510 | { |
ab8d0fc4 | 2511 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2512 | struct io_failure_record *failrec; |
4a54c8c1 | 2513 | struct extent_map *em; |
4a54c8c1 JS |
2514 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2515 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2516 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
150e4b05 | 2517 | const u32 sectorsize = fs_info->sectorsize; |
4a54c8c1 | 2518 | int ret; |
4a54c8c1 JS |
2519 | u64 logical; |
2520 | ||
2279a270 | 2521 | failrec = get_state_failrec(failure_tree, start); |
3526302f | 2522 | if (!IS_ERR(failrec)) { |
ab8d0fc4 | 2523 | btrfs_debug(fs_info, |
1245835d QW |
2524 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu", |
2525 | failrec->logical, failrec->start, failrec->len); | |
4a54c8c1 JS |
2526 | /* |
2527 | * when data can be on disk more than twice, add to failrec here | |
2528 | * (e.g. with a list for failed_mirror) to make | |
2529 | * clean_io_failure() clean all those errors at once. | |
2530 | */ | |
3526302f NB |
2531 | |
2532 | return failrec; | |
4a54c8c1 | 2533 | } |
2fe6303e | 2534 | |
3526302f NB |
2535 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); |
2536 | if (!failrec) | |
2537 | return ERR_PTR(-ENOMEM); | |
2fe6303e | 2538 | |
3526302f | 2539 | failrec->start = start; |
150e4b05 | 2540 | failrec->len = sectorsize; |
3526302f | 2541 | failrec->this_mirror = 0; |
0ff40013 | 2542 | failrec->compress_type = BTRFS_COMPRESS_NONE; |
3526302f NB |
2543 | |
2544 | read_lock(&em_tree->lock); | |
2545 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2546 | if (!em) { | |
2547 | read_unlock(&em_tree->lock); | |
2548 | kfree(failrec); | |
2549 | return ERR_PTR(-EIO); | |
2550 | } | |
2551 | ||
2552 | if (em->start > start || em->start + em->len <= start) { | |
2553 | free_extent_map(em); | |
2554 | em = NULL; | |
2555 | } | |
2556 | read_unlock(&em_tree->lock); | |
2557 | if (!em) { | |
2558 | kfree(failrec); | |
2559 | return ERR_PTR(-EIO); | |
2560 | } | |
2561 | ||
2562 | logical = start - em->start; | |
2563 | logical = em->block_start + logical; | |
2564 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2565 | logical = em->block_start; | |
0ff40013 | 2566 | failrec->compress_type = em->compress_type; |
3526302f NB |
2567 | } |
2568 | ||
2569 | btrfs_debug(fs_info, | |
2570 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2571 | logical, start, failrec->len); | |
2572 | ||
2573 | failrec->logical = logical; | |
2574 | free_extent_map(em); | |
2575 | ||
2576 | /* Set the bits in the private failure tree */ | |
150e4b05 | 2577 | ret = set_extent_bits(failure_tree, start, start + sectorsize - 1, |
3526302f NB |
2578 | EXTENT_LOCKED | EXTENT_DIRTY); |
2579 | if (ret >= 0) { | |
2580 | ret = set_state_failrec(failure_tree, start, failrec); | |
2581 | /* Set the bits in the inode's tree */ | |
150e4b05 QW |
2582 | ret = set_extent_bits(tree, start, start + sectorsize - 1, |
2583 | EXTENT_DAMAGED); | |
3526302f NB |
2584 | } else if (ret < 0) { |
2585 | kfree(failrec); | |
2586 | return ERR_PTR(ret); | |
2587 | } | |
2588 | ||
2589 | return failrec; | |
2fe6303e MX |
2590 | } |
2591 | ||
1245835d | 2592 | static bool btrfs_check_repairable(struct inode *inode, |
ce06d3ec OS |
2593 | struct io_failure_record *failrec, |
2594 | int failed_mirror) | |
2fe6303e | 2595 | { |
ab8d0fc4 | 2596 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2597 | int num_copies; |
2598 | ||
ab8d0fc4 | 2599 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2600 | if (num_copies == 1) { |
2601 | /* | |
2602 | * we only have a single copy of the data, so don't bother with | |
2603 | * all the retry and error correction code that follows. no | |
2604 | * matter what the error is, it is very likely to persist. | |
2605 | */ | |
ab8d0fc4 JM |
2606 | btrfs_debug(fs_info, |
2607 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2608 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2609 | return false; |
4a54c8c1 JS |
2610 | } |
2611 | ||
1245835d QW |
2612 | /* The failure record should only contain one sector */ |
2613 | ASSERT(failrec->len == fs_info->sectorsize); | |
2614 | ||
4a54c8c1 | 2615 | /* |
1245835d QW |
2616 | * There are two premises: |
2617 | * a) deliver good data to the caller | |
2618 | * b) correct the bad sectors on disk | |
2619 | * | |
2620 | * Since we're only doing repair for one sector, we only need to get | |
2621 | * a good copy of the failed sector and if we succeed, we have setup | |
2622 | * everything for repair_io_failure to do the rest for us. | |
4a54c8c1 | 2623 | */ |
510671d2 | 2624 | ASSERT(failed_mirror); |
1245835d QW |
2625 | failrec->failed_mirror = failed_mirror; |
2626 | failrec->this_mirror++; | |
2627 | if (failrec->this_mirror == failed_mirror) | |
4a54c8c1 | 2628 | failrec->this_mirror++; |
4a54c8c1 | 2629 | |
facc8a22 | 2630 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2631 | btrfs_debug(fs_info, |
2632 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2633 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2634 | return false; |
4a54c8c1 JS |
2635 | } |
2636 | ||
c3cfb656 | 2637 | return true; |
2fe6303e MX |
2638 | } |
2639 | ||
150e4b05 QW |
2640 | int btrfs_repair_one_sector(struct inode *inode, |
2641 | struct bio *failed_bio, u32 bio_offset, | |
2642 | struct page *page, unsigned int pgoff, | |
2643 | u64 start, int failed_mirror, | |
2644 | submit_bio_hook_t *submit_bio_hook) | |
2fe6303e MX |
2645 | { |
2646 | struct io_failure_record *failrec; | |
77d5d689 | 2647 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2648 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2649 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
c3a3b19b | 2650 | struct btrfs_bio *failed_bbio = btrfs_bio(failed_bio); |
7ffd27e3 | 2651 | const int icsum = bio_offset >> fs_info->sectorsize_bits; |
77d5d689 | 2652 | struct bio *repair_bio; |
c3a3b19b | 2653 | struct btrfs_bio *repair_bbio; |
2fe6303e | 2654 | |
77d5d689 OS |
2655 | btrfs_debug(fs_info, |
2656 | "repair read error: read error at %llu", start); | |
2fe6303e | 2657 | |
1f7ad75b | 2658 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e | 2659 | |
150e4b05 | 2660 | failrec = btrfs_get_io_failure_record(inode, start); |
3526302f | 2661 | if (IS_ERR(failrec)) |
150e4b05 | 2662 | return PTR_ERR(failrec); |
2fe6303e | 2663 | |
1245835d QW |
2664 | |
2665 | if (!btrfs_check_repairable(inode, failrec, failed_mirror)) { | |
7870d082 | 2666 | free_io_failure(failure_tree, tree, failrec); |
150e4b05 | 2667 | return -EIO; |
2fe6303e MX |
2668 | } |
2669 | ||
c3a3b19b QW |
2670 | repair_bio = btrfs_bio_alloc(1); |
2671 | repair_bbio = btrfs_bio(repair_bio); | |
00d82525 | 2672 | repair_bbio->file_offset = start; |
77d5d689 | 2673 | repair_bio->bi_opf = REQ_OP_READ; |
77d5d689 OS |
2674 | repair_bio->bi_end_io = failed_bio->bi_end_io; |
2675 | repair_bio->bi_iter.bi_sector = failrec->logical >> 9; | |
2676 | repair_bio->bi_private = failed_bio->bi_private; | |
2fe6303e | 2677 | |
c3a3b19b | 2678 | if (failed_bbio->csum) { |
223486c2 | 2679 | const u32 csum_size = fs_info->csum_size; |
77d5d689 | 2680 | |
c3a3b19b QW |
2681 | repair_bbio->csum = repair_bbio->csum_inline; |
2682 | memcpy(repair_bbio->csum, | |
2683 | failed_bbio->csum + csum_size * icsum, csum_size); | |
77d5d689 | 2684 | } |
2fe6303e | 2685 | |
77d5d689 | 2686 | bio_add_page(repair_bio, page, failrec->len, pgoff); |
c3a3b19b | 2687 | repair_bbio->iter = repair_bio->bi_iter; |
4a54c8c1 | 2688 | |
ab8d0fc4 | 2689 | btrfs_debug(btrfs_sb(inode->i_sb), |
1245835d QW |
2690 | "repair read error: submitting new read to mirror %d", |
2691 | failrec->this_mirror); | |
4a54c8c1 | 2692 | |
8cbc3001 JB |
2693 | /* |
2694 | * At this point we have a bio, so any errors from submit_bio_hook() | |
2695 | * will be handled by the endio on the repair_bio, so we can't return an | |
2696 | * error here. | |
2697 | */ | |
0ff40013 | 2698 | submit_bio_hook(inode, repair_bio, failrec->this_mirror, failrec->compress_type); |
8cbc3001 | 2699 | return BLK_STS_OK; |
150e4b05 QW |
2700 | } |
2701 | ||
2702 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) | |
2703 | { | |
2704 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
2705 | ||
2706 | ASSERT(page_offset(page) <= start && | |
2707 | start + len <= page_offset(page) + PAGE_SIZE); | |
2708 | ||
150e4b05 | 2709 | if (uptodate) { |
14605409 BB |
2710 | if (fsverity_active(page->mapping->host) && |
2711 | !PageError(page) && | |
2712 | !PageUptodate(page) && | |
2713 | start < i_size_read(page->mapping->host) && | |
2714 | !fsverity_verify_page(page)) { | |
2715 | btrfs_page_set_error(fs_info, page, start, len); | |
2716 | } else { | |
2717 | btrfs_page_set_uptodate(fs_info, page, start, len); | |
2718 | } | |
150e4b05 QW |
2719 | } else { |
2720 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2721 | btrfs_page_set_error(fs_info, page, start, len); | |
2722 | } | |
2723 | ||
fbca46eb | 2724 | if (!btrfs_is_subpage(fs_info, page)) |
150e4b05 | 2725 | unlock_page(page); |
3d078efa | 2726 | else |
150e4b05 QW |
2727 | btrfs_subpage_end_reader(fs_info, page, start, len); |
2728 | } | |
2729 | ||
fd5a6f63 QW |
2730 | static void submit_data_read_repair(struct inode *inode, struct bio *failed_bio, |
2731 | u32 bio_offset, const struct bio_vec *bvec, | |
2732 | int failed_mirror, unsigned int error_bitmap) | |
150e4b05 | 2733 | { |
fd5a6f63 | 2734 | const unsigned int pgoff = bvec->bv_offset; |
150e4b05 | 2735 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
fd5a6f63 QW |
2736 | struct page *page = bvec->bv_page; |
2737 | const u64 start = page_offset(bvec->bv_page) + bvec->bv_offset; | |
2738 | const u64 end = start + bvec->bv_len - 1; | |
150e4b05 QW |
2739 | const u32 sectorsize = fs_info->sectorsize; |
2740 | const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits; | |
150e4b05 QW |
2741 | int i; |
2742 | ||
2743 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); | |
2744 | ||
c0111c44 QW |
2745 | /* This repair is only for data */ |
2746 | ASSERT(is_data_inode(inode)); | |
2747 | ||
150e4b05 QW |
2748 | /* We're here because we had some read errors or csum mismatch */ |
2749 | ASSERT(error_bitmap); | |
2750 | ||
2751 | /* | |
2752 | * We only get called on buffered IO, thus page must be mapped and bio | |
2753 | * must not be cloned. | |
2754 | */ | |
2755 | ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED)); | |
2756 | ||
2757 | /* Iterate through all the sectors in the range */ | |
2758 | for (i = 0; i < nr_bits; i++) { | |
2759 | const unsigned int offset = i * sectorsize; | |
2760 | struct extent_state *cached = NULL; | |
2761 | bool uptodate = false; | |
2762 | int ret; | |
2763 | ||
2764 | if (!(error_bitmap & (1U << i))) { | |
2765 | /* | |
2766 | * This sector has no error, just end the page read | |
2767 | * and unlock the range. | |
2768 | */ | |
2769 | uptodate = true; | |
2770 | goto next; | |
2771 | } | |
2772 | ||
2773 | ret = btrfs_repair_one_sector(inode, failed_bio, | |
2774 | bio_offset + offset, | |
2775 | page, pgoff + offset, start + offset, | |
c0111c44 | 2776 | failed_mirror, btrfs_submit_data_bio); |
150e4b05 QW |
2777 | if (!ret) { |
2778 | /* | |
2779 | * We have submitted the read repair, the page release | |
2780 | * will be handled by the endio function of the | |
2781 | * submitted repair bio. | |
2782 | * Thus we don't need to do any thing here. | |
2783 | */ | |
2784 | continue; | |
2785 | } | |
2786 | /* | |
fd5a6f63 QW |
2787 | * Continue on failed repair, otherwise the remaining sectors |
2788 | * will not be properly unlocked. | |
150e4b05 | 2789 | */ |
150e4b05 QW |
2790 | next: |
2791 | end_page_read(page, uptodate, start + offset, sectorsize); | |
2792 | if (uptodate) | |
2793 | set_extent_uptodate(&BTRFS_I(inode)->io_tree, | |
2794 | start + offset, | |
2795 | start + offset + sectorsize - 1, | |
2796 | &cached, GFP_ATOMIC); | |
2797 | unlock_extent_cached_atomic(&BTRFS_I(inode)->io_tree, | |
2798 | start + offset, | |
2799 | start + offset + sectorsize - 1, | |
2800 | &cached); | |
2801 | } | |
4a54c8c1 JS |
2802 | } |
2803 | ||
d1310b2e CM |
2804 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2805 | ||
b5227c07 | 2806 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 2807 | { |
38a39ac7 | 2808 | struct btrfs_inode *inode; |
25c1252a | 2809 | const bool uptodate = (err == 0); |
3e2426bd | 2810 | int ret = 0; |
87826df0 | 2811 | |
38a39ac7 QW |
2812 | ASSERT(page && page->mapping); |
2813 | inode = BTRFS_I(page->mapping->host); | |
2814 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 2815 | |
87826df0 | 2816 | if (!uptodate) { |
963e4db8 QW |
2817 | const struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2818 | u32 len; | |
2819 | ||
2820 | ASSERT(end + 1 - start <= U32_MAX); | |
2821 | len = end + 1 - start; | |
2822 | ||
2823 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2824 | btrfs_page_set_error(fs_info, page, start, len); | |
bff5baf8 | 2825 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2826 | mapping_set_error(page->mapping, ret); |
87826df0 | 2827 | } |
87826df0 JM |
2828 | } |
2829 | ||
d1310b2e CM |
2830 | /* |
2831 | * after a writepage IO is done, we need to: | |
2832 | * clear the uptodate bits on error | |
2833 | * clear the writeback bits in the extent tree for this IO | |
2834 | * end_page_writeback if the page has no more pending IO | |
2835 | * | |
2836 | * Scheduling is not allowed, so the extent state tree is expected | |
2837 | * to have one and only one object corresponding to this IO. | |
2838 | */ | |
4246a0b6 | 2839 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2840 | { |
4e4cbee9 | 2841 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2842 | struct bio_vec *bvec; |
d1310b2e CM |
2843 | u64 start; |
2844 | u64 end; | |
6dc4f100 | 2845 | struct bvec_iter_all iter_all; |
d8e3fb10 | 2846 | bool first_bvec = true; |
d1310b2e | 2847 | |
c09abff8 | 2848 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2849 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2850 | struct page *page = bvec->bv_page; |
0b246afa JM |
2851 | struct inode *inode = page->mapping->host; |
2852 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
2853 | const u32 sectorsize = fs_info->sectorsize; |
2854 | ||
2855 | /* Our read/write should always be sector aligned. */ | |
2856 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2857 | btrfs_err(fs_info, | |
2858 | "partial page write in btrfs with offset %u and length %u", | |
2859 | bvec->bv_offset, bvec->bv_len); | |
2860 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
2861 | btrfs_info(fs_info, | |
2862 | "incomplete page write with offset %u and length %u", | |
2863 | bvec->bv_offset, bvec->bv_len); | |
2864 | ||
2865 | start = page_offset(page) + bvec->bv_offset; | |
2866 | end = start + bvec->bv_len - 1; | |
d1310b2e | 2867 | |
d8e3fb10 NA |
2868 | if (first_bvec) { |
2869 | btrfs_record_physical_zoned(inode, start, bio); | |
2870 | first_bvec = false; | |
2871 | } | |
2872 | ||
4e4cbee9 | 2873 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
2874 | |
2875 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 2876 | } |
2b1f55b0 | 2877 | |
d1310b2e | 2878 | bio_put(bio); |
d1310b2e CM |
2879 | } |
2880 | ||
94e8c95c QW |
2881 | /* |
2882 | * Record previously processed extent range | |
2883 | * | |
2884 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
2885 | * the extent io operations. | |
2886 | */ | |
2887 | struct processed_extent { | |
2888 | struct btrfs_inode *inode; | |
2889 | /* Start of the range in @inode */ | |
2890 | u64 start; | |
2e626e56 | 2891 | /* End of the range in @inode */ |
94e8c95c QW |
2892 | u64 end; |
2893 | bool uptodate; | |
2894 | }; | |
2895 | ||
2896 | /* | |
2897 | * Try to release processed extent range | |
2898 | * | |
2899 | * May not release the extent range right now if the current range is | |
2900 | * contiguous to processed extent. | |
2901 | * | |
2902 | * Will release processed extent when any of @inode, @uptodate, the range is | |
2903 | * no longer contiguous to the processed range. | |
2904 | * | |
2905 | * Passing @inode == NULL will force processed extent to be released. | |
2906 | */ | |
2907 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
2908 | struct btrfs_inode *inode, u64 start, u64 end, | |
2909 | bool uptodate) | |
883d0de4 MX |
2910 | { |
2911 | struct extent_state *cached = NULL; | |
94e8c95c QW |
2912 | struct extent_io_tree *tree; |
2913 | ||
2914 | /* The first extent, initialize @processed */ | |
2915 | if (!processed->inode) | |
2916 | goto update; | |
883d0de4 | 2917 | |
94e8c95c QW |
2918 | /* |
2919 | * Contiguous to processed extent, just uptodate the end. | |
2920 | * | |
2921 | * Several things to notice: | |
2922 | * | |
2923 | * - bio can be merged as long as on-disk bytenr is contiguous | |
2924 | * This means we can have page belonging to other inodes, thus need to | |
2925 | * check if the inode still matches. | |
2926 | * - bvec can contain range beyond current page for multi-page bvec | |
2927 | * Thus we need to do processed->end + 1 >= start check | |
2928 | */ | |
2929 | if (processed->inode == inode && processed->uptodate == uptodate && | |
2930 | processed->end + 1 >= start && end >= processed->end) { | |
2931 | processed->end = end; | |
2932 | return; | |
2933 | } | |
2934 | ||
2935 | tree = &processed->inode->io_tree; | |
2936 | /* | |
2937 | * Now we don't have range contiguous to the processed range, release | |
2938 | * the processed range now. | |
2939 | */ | |
2940 | if (processed->uptodate && tree->track_uptodate) | |
2941 | set_extent_uptodate(tree, processed->start, processed->end, | |
2942 | &cached, GFP_ATOMIC); | |
2943 | unlock_extent_cached_atomic(tree, processed->start, processed->end, | |
2944 | &cached); | |
2945 | ||
2946 | update: | |
2947 | /* Update processed to current range */ | |
2948 | processed->inode = inode; | |
2949 | processed->start = start; | |
2950 | processed->end = end; | |
2951 | processed->uptodate = uptodate; | |
883d0de4 MX |
2952 | } |
2953 | ||
92082d40 QW |
2954 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
2955 | { | |
2956 | ASSERT(PageLocked(page)); | |
fbca46eb | 2957 | if (!btrfs_is_subpage(fs_info, page)) |
92082d40 QW |
2958 | return; |
2959 | ||
2960 | ASSERT(PagePrivate(page)); | |
2961 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
2962 | } | |
2963 | ||
d9bb77d5 | 2964 | /* |
01cd3909 | 2965 | * Find extent buffer for a givne bytenr. |
d9bb77d5 QW |
2966 | * |
2967 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
2968 | * in endio context. | |
2969 | */ | |
2970 | static struct extent_buffer *find_extent_buffer_readpage( | |
2971 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
2972 | { | |
2973 | struct extent_buffer *eb; | |
2974 | ||
2975 | /* | |
2976 | * For regular sectorsize, we can use page->private to grab extent | |
2977 | * buffer | |
2978 | */ | |
fbca46eb | 2979 | if (fs_info->nodesize >= PAGE_SIZE) { |
d9bb77d5 QW |
2980 | ASSERT(PagePrivate(page) && page->private); |
2981 | return (struct extent_buffer *)page->private; | |
2982 | } | |
2983 | ||
01cd3909 DS |
2984 | /* For subpage case, we need to lookup buffer radix tree */ |
2985 | rcu_read_lock(); | |
2986 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
2987 | bytenr >> fs_info->sectorsize_bits); | |
2988 | rcu_read_unlock(); | |
d9bb77d5 QW |
2989 | ASSERT(eb); |
2990 | return eb; | |
2991 | } | |
2992 | ||
d1310b2e CM |
2993 | /* |
2994 | * after a readpage IO is done, we need to: | |
2995 | * clear the uptodate bits on error | |
2996 | * set the uptodate bits if things worked | |
2997 | * set the page up to date if all extents in the tree are uptodate | |
2998 | * clear the lock bit in the extent tree | |
2999 | * unlock the page if there are no other extents locked for it | |
3000 | * | |
3001 | * Scheduling is not allowed, so the extent state tree is expected | |
3002 | * to have one and only one object corresponding to this IO. | |
3003 | */ | |
4246a0b6 | 3004 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 3005 | { |
2c30c71b | 3006 | struct bio_vec *bvec; |
c3a3b19b | 3007 | struct btrfs_bio *bbio = btrfs_bio(bio); |
7870d082 | 3008 | struct extent_io_tree *tree, *failure_tree; |
94e8c95c | 3009 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
3010 | /* |
3011 | * The offset to the beginning of a bio, since one bio can never be | |
3012 | * larger than UINT_MAX, u32 here is enough. | |
3013 | */ | |
3014 | u32 bio_offset = 0; | |
5cf1ab56 | 3015 | int mirror; |
d1310b2e | 3016 | int ret; |
6dc4f100 | 3017 | struct bvec_iter_all iter_all; |
d1310b2e | 3018 | |
c09abff8 | 3019 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 3020 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 3021 | bool uptodate = !bio->bi_status; |
d1310b2e | 3022 | struct page *page = bvec->bv_page; |
a71754fc | 3023 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 3024 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 | 3025 | const u32 sectorsize = fs_info->sectorsize; |
150e4b05 | 3026 | unsigned int error_bitmap = (unsigned int)-1; |
7ffd27e3 QW |
3027 | u64 start; |
3028 | u64 end; | |
3029 | u32 len; | |
507903b8 | 3030 | |
ab8d0fc4 JM |
3031 | btrfs_debug(fs_info, |
3032 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 3033 | bio->bi_iter.bi_sector, bio->bi_status, |
c3a3b19b | 3034 | bbio->mirror_num); |
a71754fc | 3035 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 3036 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 3037 | |
8b8bbd46 QW |
3038 | /* |
3039 | * We always issue full-sector reads, but if some block in a | |
3040 | * page fails to read, blk_update_request() will advance | |
3041 | * bv_offset and adjust bv_len to compensate. Print a warning | |
3042 | * for unaligned offsets, and an error if they don't add up to | |
3043 | * a full sector. | |
3044 | */ | |
3045 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
3046 | btrfs_err(fs_info, | |
3047 | "partial page read in btrfs with offset %u and length %u", | |
3048 | bvec->bv_offset, bvec->bv_len); | |
3049 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
3050 | sectorsize)) | |
3051 | btrfs_info(fs_info, | |
3052 | "incomplete page read with offset %u and length %u", | |
3053 | bvec->bv_offset, bvec->bv_len); | |
3054 | ||
3055 | start = page_offset(page) + bvec->bv_offset; | |
3056 | end = start + bvec->bv_len - 1; | |
facc8a22 | 3057 | len = bvec->bv_len; |
d1310b2e | 3058 | |
c3a3b19b | 3059 | mirror = bbio->mirror_num; |
78e62c02 | 3060 | if (likely(uptodate)) { |
150e4b05 | 3061 | if (is_data_inode(inode)) { |
c3a3b19b | 3062 | error_bitmap = btrfs_verify_data_csum(bbio, |
5e295768 | 3063 | bio_offset, page, start, end); |
150e4b05 QW |
3064 | ret = error_bitmap; |
3065 | } else { | |
c3a3b19b | 3066 | ret = btrfs_validate_metadata_buffer(bbio, |
8e1dc982 | 3067 | page, start, end, mirror); |
150e4b05 | 3068 | } |
5ee0844d | 3069 | if (ret) |
150e4b05 | 3070 | uptodate = false; |
5ee0844d | 3071 | else |
7870d082 JB |
3072 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
3073 | failure_tree, tree, start, | |
3074 | page, | |
3075 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 3076 | } |
ea466794 | 3077 | |
f2a09da9 MX |
3078 | if (likely(uptodate)) |
3079 | goto readpage_ok; | |
3080 | ||
be17b3af | 3081 | if (is_data_inode(inode)) { |
510671d2 JB |
3082 | /* |
3083 | * If we failed to submit the IO at all we'll have a | |
3084 | * mirror_num == 0, in which case we need to just mark | |
3085 | * the page with an error and unlock it and carry on. | |
3086 | */ | |
3087 | if (mirror == 0) | |
3088 | goto readpage_ok; | |
3089 | ||
f4a8e656 | 3090 | /* |
c0111c44 | 3091 | * submit_data_read_repair() will handle all the good |
150e4b05 | 3092 | * and bad sectors, we just continue to the next bvec. |
f4a8e656 | 3093 | */ |
fd5a6f63 QW |
3094 | submit_data_read_repair(inode, bio, bio_offset, bvec, |
3095 | mirror, error_bitmap); | |
150e4b05 QW |
3096 | |
3097 | ASSERT(bio_offset + len > bio_offset); | |
3098 | bio_offset += len; | |
3099 | continue; | |
78e62c02 NB |
3100 | } else { |
3101 | struct extent_buffer *eb; | |
3102 | ||
d9bb77d5 | 3103 | eb = find_extent_buffer_readpage(fs_info, page, start); |
78e62c02 NB |
3104 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
3105 | eb->read_mirror = mirror; | |
3106 | atomic_dec(&eb->io_pages); | |
7e38326f | 3107 | } |
f2a09da9 | 3108 | readpage_ok: |
883d0de4 | 3109 | if (likely(uptodate)) { |
a71754fc | 3110 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3111 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 3112 | |
c28ea613 QW |
3113 | /* |
3114 | * Zero out the remaining part if this range straddles | |
3115 | * i_size. | |
3116 | * | |
3117 | * Here we should only zero the range inside the bvec, | |
3118 | * not touch anything else. | |
3119 | * | |
3120 | * NOTE: i_size is exclusive while end is inclusive. | |
3121 | */ | |
3122 | if (page->index == end_index && i_size <= end) { | |
3123 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 3124 | offset_in_page(start)); |
c28ea613 QW |
3125 | |
3126 | zero_user_segment(page, zero_start, | |
3127 | offset_in_page(end) + 1); | |
3128 | } | |
70dec807 | 3129 | } |
7ffd27e3 QW |
3130 | ASSERT(bio_offset + len > bio_offset); |
3131 | bio_offset += len; | |
883d0de4 | 3132 | |
e09caaf9 | 3133 | /* Update page status and unlock */ |
92082d40 | 3134 | end_page_read(page, uptodate, start, len); |
94e8c95c | 3135 | endio_readpage_release_extent(&processed, BTRFS_I(inode), |
14605409 | 3136 | start, end, PageUptodate(page)); |
2c30c71b | 3137 | } |
94e8c95c QW |
3138 | /* Release the last extent */ |
3139 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
c3a3b19b | 3140 | btrfs_bio_free_csum(bbio); |
d1310b2e | 3141 | bio_put(bio); |
d1310b2e CM |
3142 | } |
3143 | ||
dd137dd1 STD |
3144 | /** |
3145 | * Populate every free slot in a provided array with pages. | |
3146 | * | |
3147 | * @nr_pages: number of pages to allocate | |
3148 | * @page_array: the array to fill with pages; any existing non-null entries in | |
3149 | * the array will be skipped | |
3150 | * | |
3151 | * Return: 0 if all pages were able to be allocated; | |
3152 | * -ENOMEM otherwise, and the caller is responsible for freeing all | |
3153 | * non-null page pointers in the array. | |
3154 | */ | |
3155 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array) | |
3156 | { | |
91d6ac1d | 3157 | unsigned int allocated; |
dd137dd1 | 3158 | |
91d6ac1d STD |
3159 | for (allocated = 0; allocated < nr_pages;) { |
3160 | unsigned int last = allocated; | |
dd137dd1 | 3161 | |
91d6ac1d STD |
3162 | allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array); |
3163 | ||
395cb57e STD |
3164 | if (allocated == nr_pages) |
3165 | return 0; | |
3166 | ||
91d6ac1d STD |
3167 | /* |
3168 | * During this iteration, no page could be allocated, even | |
3169 | * though alloc_pages_bulk_array() falls back to alloc_page() | |
3170 | * if it could not bulk-allocate. So we must be out of memory. | |
3171 | */ | |
3172 | if (allocated == last) | |
dd137dd1 | 3173 | return -ENOMEM; |
395cb57e STD |
3174 | |
3175 | memalloc_retry_wait(GFP_NOFS); | |
dd137dd1 STD |
3176 | } |
3177 | return 0; | |
3178 | } | |
3179 | ||
9be3395b | 3180 | /* |
184f999e DS |
3181 | * Initialize the members up to but not including 'bio'. Use after allocating a |
3182 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
3183 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 3184 | */ |
c3a3b19b | 3185 | static inline void btrfs_bio_init(struct btrfs_bio *bbio) |
d1310b2e | 3186 | { |
c3a3b19b | 3187 | memset(bbio, 0, offsetof(struct btrfs_bio, bio)); |
184f999e | 3188 | } |
d1310b2e | 3189 | |
9be3395b | 3190 | /* |
cd8e0cca QW |
3191 | * Allocate a btrfs_io_bio, with @nr_iovecs as maximum number of iovecs. |
3192 | * | |
3193 | * The bio allocation is backed by bioset and does not fail. | |
9be3395b | 3194 | */ |
c3a3b19b | 3195 | struct bio *btrfs_bio_alloc(unsigned int nr_iovecs) |
d1310b2e CM |
3196 | { |
3197 | struct bio *bio; | |
d1310b2e | 3198 | |
cd8e0cca | 3199 | ASSERT(0 < nr_iovecs && nr_iovecs <= BIO_MAX_VECS); |
609be106 | 3200 | bio = bio_alloc_bioset(NULL, nr_iovecs, 0, GFP_NOFS, &btrfs_bioset); |
c3a3b19b | 3201 | btrfs_bio_init(btrfs_bio(bio)); |
d1310b2e CM |
3202 | return bio; |
3203 | } | |
3204 | ||
110ac0e5 | 3205 | struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio) |
9be3395b | 3206 | { |
c3a3b19b | 3207 | struct btrfs_bio *bbio; |
23ea8e5a | 3208 | struct bio *new; |
9be3395b | 3209 | |
6e707bcd | 3210 | /* Bio allocation backed by a bioset does not fail */ |
110ac0e5 | 3211 | new = bio_alloc_clone(bdev, bio, GFP_NOFS, &btrfs_bioset); |
c3a3b19b QW |
3212 | bbio = btrfs_bio(new); |
3213 | btrfs_bio_init(bbio); | |
3214 | bbio->iter = bio->bi_iter; | |
23ea8e5a MX |
3215 | return new; |
3216 | } | |
9be3395b | 3217 | |
21dda654 | 3218 | struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size) |
2f8e9140 LB |
3219 | { |
3220 | struct bio *bio; | |
c3a3b19b | 3221 | struct btrfs_bio *bbio; |
2f8e9140 | 3222 | |
21dda654 CK |
3223 | ASSERT(offset <= UINT_MAX && size <= UINT_MAX); |
3224 | ||
2f8e9140 | 3225 | /* this will never fail when it's backed by a bioset */ |
abfc426d | 3226 | bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3227 | ASSERT(bio); |
3228 | ||
c3a3b19b QW |
3229 | bbio = btrfs_bio(bio); |
3230 | btrfs_bio_init(bbio); | |
2f8e9140 LB |
3231 | |
3232 | bio_trim(bio, offset >> 9, size >> 9); | |
c3a3b19b | 3233 | bbio->iter = bio->bi_iter; |
2f8e9140 LB |
3234 | return bio; |
3235 | } | |
9be3395b | 3236 | |
953651eb NA |
3237 | /** |
3238 | * Attempt to add a page to bio | |
3239 | * | |
be8d1a2a | 3240 | * @bio_ctrl: record both the bio, and its bio_flags |
953651eb NA |
3241 | * @page: page to add to the bio |
3242 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
3243 | * a contiguous page to the previous one | |
953651eb | 3244 | * @size: portion of page that we want to write |
be8d1a2a | 3245 | * @pg_offset: starting offset in the page |
cb3a12d9 | 3246 | * @compress_type: compression type of the current bio to see if we can merge them |
953651eb NA |
3247 | * |
3248 | * Attempt to add a page to bio considering stripe alignment etc. | |
3249 | * | |
e0eefe07 QW |
3250 | * Return >= 0 for the number of bytes added to the bio. |
3251 | * Can return 0 if the current bio is already at stripe/zone boundary. | |
3252 | * Return <0 for error. | |
953651eb | 3253 | */ |
e0eefe07 QW |
3254 | static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
3255 | struct page *page, | |
3256 | u64 disk_bytenr, unsigned int size, | |
3257 | unsigned int pg_offset, | |
cb3a12d9 | 3258 | enum btrfs_compression_type compress_type) |
953651eb | 3259 | { |
390ed29b QW |
3260 | struct bio *bio = bio_ctrl->bio; |
3261 | u32 bio_size = bio->bi_iter.bi_size; | |
e0eefe07 | 3262 | u32 real_size; |
953651eb NA |
3263 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
3264 | bool contig; | |
e1326f03 | 3265 | int ret; |
953651eb | 3266 | |
390ed29b QW |
3267 | ASSERT(bio); |
3268 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
3269 | ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary); | |
0f07003b | 3270 | if (bio_ctrl->compress_type != compress_type) |
e0eefe07 | 3271 | return 0; |
953651eb | 3272 | |
0f07003b | 3273 | if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) |
953651eb NA |
3274 | contig = bio->bi_iter.bi_sector == sector; |
3275 | else | |
3276 | contig = bio_end_sector(bio) == sector; | |
3277 | if (!contig) | |
e0eefe07 | 3278 | return 0; |
953651eb | 3279 | |
e0eefe07 QW |
3280 | real_size = min(bio_ctrl->len_to_oe_boundary, |
3281 | bio_ctrl->len_to_stripe_boundary) - bio_size; | |
3282 | real_size = min(real_size, size); | |
3283 | ||
3284 | /* | |
3285 | * If real_size is 0, never call bio_add_*_page(), as even size is 0, | |
3286 | * bio will still execute its endio function on the page! | |
3287 | */ | |
3288 | if (real_size == 0) | |
3289 | return 0; | |
953651eb | 3290 | |
390ed29b | 3291 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
e0eefe07 | 3292 | ret = bio_add_zone_append_page(bio, page, real_size, pg_offset); |
390ed29b | 3293 | else |
e0eefe07 | 3294 | ret = bio_add_page(bio, page, real_size, pg_offset); |
e1326f03 | 3295 | |
e0eefe07 | 3296 | return ret; |
953651eb NA |
3297 | } |
3298 | ||
390ed29b | 3299 | static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
939c7feb | 3300 | struct btrfs_inode *inode, u64 file_offset) |
390ed29b QW |
3301 | { |
3302 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3303 | struct btrfs_io_geometry geom; | |
3304 | struct btrfs_ordered_extent *ordered; | |
3305 | struct extent_map *em; | |
3306 | u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT); | |
3307 | int ret; | |
3308 | ||
3309 | /* | |
3310 | * Pages for compressed extent are never submitted to disk directly, | |
3311 | * thus it has no real boundary, just set them to U32_MAX. | |
3312 | * | |
3313 | * The split happens for real compressed bio, which happens in | |
3314 | * btrfs_submit_compressed_read/write(). | |
3315 | */ | |
0f07003b | 3316 | if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) { |
390ed29b QW |
3317 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3318 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3319 | return 0; | |
3320 | } | |
3321 | em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize); | |
3322 | if (IS_ERR(em)) | |
3323 | return PTR_ERR(em); | |
3324 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio), | |
3325 | logical, &geom); | |
3326 | free_extent_map(em); | |
3327 | if (ret < 0) { | |
3328 | return ret; | |
3329 | } | |
3330 | if (geom.len > U32_MAX) | |
3331 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3332 | else | |
3333 | bio_ctrl->len_to_stripe_boundary = (u32)geom.len; | |
3334 | ||
73672710 | 3335 | if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) { |
390ed29b QW |
3336 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3337 | return 0; | |
3338 | } | |
3339 | ||
390ed29b | 3340 | /* Ordered extent not yet created, so we're good */ |
939c7feb | 3341 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
390ed29b QW |
3342 | if (!ordered) { |
3343 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3344 | return 0; | |
3345 | } | |
3346 | ||
3347 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
3348 | ordered->disk_bytenr + ordered->disk_num_bytes - logical); | |
3349 | btrfs_put_ordered_extent(ordered); | |
3350 | return 0; | |
3351 | } | |
3352 | ||
e0eefe07 QW |
3353 | static int alloc_new_bio(struct btrfs_inode *inode, |
3354 | struct btrfs_bio_ctrl *bio_ctrl, | |
3355 | struct writeback_control *wbc, | |
3356 | unsigned int opf, | |
3357 | bio_end_io_t end_io_func, | |
939c7feb | 3358 | u64 disk_bytenr, u32 offset, u64 file_offset, |
cb3a12d9 | 3359 | enum btrfs_compression_type compress_type) |
e0eefe07 QW |
3360 | { |
3361 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3362 | struct bio *bio; | |
3363 | int ret; | |
3364 | ||
c3a3b19b | 3365 | bio = btrfs_bio_alloc(BIO_MAX_VECS); |
e0eefe07 QW |
3366 | /* |
3367 | * For compressed page range, its disk_bytenr is always @disk_bytenr | |
3368 | * passed in, no matter if we have added any range into previous bio. | |
3369 | */ | |
cb3a12d9 | 3370 | if (compress_type != BTRFS_COMPRESS_NONE) |
cd8e0cca | 3371 | bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
e0eefe07 | 3372 | else |
cd8e0cca | 3373 | bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT; |
e0eefe07 | 3374 | bio_ctrl->bio = bio; |
0f07003b | 3375 | bio_ctrl->compress_type = compress_type; |
e0eefe07 QW |
3376 | bio->bi_end_io = end_io_func; |
3377 | bio->bi_private = &inode->io_tree; | |
e0eefe07 | 3378 | bio->bi_opf = opf; |
939c7feb NA |
3379 | ret = calc_bio_boundaries(bio_ctrl, inode, file_offset); |
3380 | if (ret < 0) | |
3381 | goto error; | |
e0eefe07 | 3382 | |
50f1cff3 CH |
3383 | if (wbc) { |
3384 | /* | |
3385 | * For Zone append we need the correct block_device that we are | |
3386 | * going to write to set in the bio to be able to respect the | |
3387 | * hardware limitation. Look it up here: | |
3388 | */ | |
3389 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { | |
3390 | struct btrfs_device *dev; | |
3391 | ||
3392 | dev = btrfs_zoned_get_device(fs_info, disk_bytenr, | |
3393 | fs_info->sectorsize); | |
3394 | if (IS_ERR(dev)) { | |
3395 | ret = PTR_ERR(dev); | |
3396 | goto error; | |
3397 | } | |
e0eefe07 | 3398 | |
50f1cff3 CH |
3399 | bio_set_dev(bio, dev->bdev); |
3400 | } else { | |
3401 | /* | |
3402 | * Otherwise pick the last added device to support | |
3403 | * cgroup writeback. For multi-device file systems this | |
3404 | * means blk-cgroup policies have to always be set on the | |
3405 | * last added/replaced device. This is a bit odd but has | |
3406 | * been like that for a long time. | |
3407 | */ | |
3408 | bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev); | |
e0eefe07 | 3409 | } |
50f1cff3 CH |
3410 | wbc_init_bio(wbc, bio); |
3411 | } else { | |
3412 | ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND); | |
e0eefe07 QW |
3413 | } |
3414 | return 0; | |
3415 | error: | |
3416 | bio_ctrl->bio = NULL; | |
3417 | bio->bi_status = errno_to_blk_status(ret); | |
3418 | bio_endio(bio); | |
3419 | return ret; | |
3420 | } | |
3421 | ||
4b81ba48 DS |
3422 | /* |
3423 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3424 | * @wbc: optional writeback control for io accounting |
3425 | * @page: page to add to the bio | |
0c64c33c QW |
3426 | * @disk_bytenr: logical bytenr where the write will be |
3427 | * @size: portion of page that we want to write to | |
b8b3d625 DS |
3428 | * @pg_offset: offset of the new bio or to check whether we are adding |
3429 | * a contiguous page to the previous one | |
5c2b1fd7 | 3430 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3431 | * @end_io_func: end_io callback for new bio |
3432 | * @mirror_num: desired mirror to read/write | |
3433 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
cb3a12d9 | 3434 | * @compress_type: compress type for current bio |
4b81ba48 | 3435 | */ |
0ceb34bf | 3436 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3437 | struct writeback_control *wbc, |
390ed29b | 3438 | struct btrfs_bio_ctrl *bio_ctrl, |
0c64c33c | 3439 | struct page *page, u64 disk_bytenr, |
6c5a4e2c | 3440 | size_t size, unsigned long pg_offset, |
f188591e | 3441 | bio_end_io_t end_io_func, |
c8b97818 | 3442 | int mirror_num, |
cb3a12d9 | 3443 | enum btrfs_compression_type compress_type, |
005efedf | 3444 | bool force_bio_submit) |
d1310b2e CM |
3445 | { |
3446 | int ret = 0; | |
e1326f03 | 3447 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
e0eefe07 | 3448 | unsigned int cur = pg_offset; |
d1310b2e | 3449 | |
390ed29b | 3450 | ASSERT(bio_ctrl); |
5c2b1fd7 | 3451 | |
390ed29b QW |
3452 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
3453 | pg_offset + size <= PAGE_SIZE); | |
e0eefe07 | 3454 | if (force_bio_submit && bio_ctrl->bio) { |
0f07003b | 3455 | submit_one_bio(bio_ctrl->bio, mirror_num, bio_ctrl->compress_type); |
e0eefe07 | 3456 | bio_ctrl->bio = NULL; |
e0eefe07 QW |
3457 | } |
3458 | ||
3459 | while (cur < pg_offset + size) { | |
3460 | u32 offset = cur - pg_offset; | |
3461 | int added; | |
3462 | ||
3463 | /* Allocate new bio if needed */ | |
3464 | if (!bio_ctrl->bio) { | |
3465 | ret = alloc_new_bio(inode, bio_ctrl, wbc, opf, | |
3466 | end_io_func, disk_bytenr, offset, | |
939c7feb | 3467 | page_offset(page) + cur, |
cb3a12d9 | 3468 | compress_type); |
e0eefe07 QW |
3469 | if (ret < 0) |
3470 | return ret; | |
3471 | } | |
3472 | /* | |
3473 | * We must go through btrfs_bio_add_page() to ensure each | |
3474 | * page range won't cross various boundaries. | |
3475 | */ | |
cb3a12d9 | 3476 | if (compress_type != BTRFS_COMPRESS_NONE) |
e0eefe07 QW |
3477 | added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, |
3478 | size - offset, pg_offset + offset, | |
cb3a12d9 | 3479 | compress_type); |
e0eefe07 QW |
3480 | else |
3481 | added = btrfs_bio_add_page(bio_ctrl, page, | |
3482 | disk_bytenr + offset, size - offset, | |
cb3a12d9 | 3483 | pg_offset + offset, compress_type); |
e0eefe07 QW |
3484 | |
3485 | /* Metadata page range should never be split */ | |
3486 | if (!is_data_inode(&inode->vfs_inode)) | |
3487 | ASSERT(added == 0 || added == size - offset); | |
3488 | ||
3489 | /* At least we added some page, update the account */ | |
3490 | if (wbc && added) | |
3491 | wbc_account_cgroup_owner(wbc, page, added); | |
3492 | ||
3493 | /* We have reached boundary, submit right now */ | |
3494 | if (added < size - offset) { | |
3495 | /* The bio should contain some page(s) */ | |
3496 | ASSERT(bio_ctrl->bio->bi_iter.bi_size); | |
0f07003b | 3497 | submit_one_bio(bio_ctrl->bio, mirror_num, bio_ctrl->compress_type); |
390ed29b | 3498 | bio_ctrl->bio = NULL; |
d1310b2e | 3499 | } |
e0eefe07 | 3500 | cur += added; |
d1310b2e | 3501 | } |
e0eefe07 | 3502 | return 0; |
d1310b2e CM |
3503 | } |
3504 | ||
760f991f QW |
3505 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
3506 | struct page *page, | |
3507 | struct btrfs_subpage *prealloc) | |
d1310b2e | 3508 | { |
760f991f QW |
3509 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3510 | int ret = 0; | |
3511 | ||
0d01e247 QW |
3512 | /* |
3513 | * If the page is mapped to btree inode, we should hold the private | |
3514 | * lock to prevent race. | |
3515 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3516 | * will not race with any other ebs. | |
3517 | */ | |
3518 | if (page->mapping) | |
3519 | lockdep_assert_held(&page->mapping->private_lock); | |
3520 | ||
fbca46eb | 3521 | if (fs_info->nodesize >= PAGE_SIZE) { |
760f991f QW |
3522 | if (!PagePrivate(page)) |
3523 | attach_page_private(page, eb); | |
3524 | else | |
3525 | WARN_ON(page->private != (unsigned long)eb); | |
3526 | return 0; | |
3527 | } | |
3528 | ||
3529 | /* Already mapped, just free prealloc */ | |
3530 | if (PagePrivate(page)) { | |
3531 | btrfs_free_subpage(prealloc); | |
3532 | return 0; | |
3533 | } | |
3534 | ||
3535 | if (prealloc) | |
3536 | /* Has preallocated memory for subpage */ | |
3537 | attach_page_private(page, prealloc); | |
d1b89bc0 | 3538 | else |
760f991f QW |
3539 | /* Do new allocation to attach subpage */ |
3540 | ret = btrfs_attach_subpage(fs_info, page, | |
3541 | BTRFS_SUBPAGE_METADATA); | |
3542 | return ret; | |
d1310b2e CM |
3543 | } |
3544 | ||
32443de3 | 3545 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 3546 | { |
32443de3 QW |
3547 | struct btrfs_fs_info *fs_info; |
3548 | ||
3549 | ASSERT(page->mapping); | |
3550 | ||
3551 | if (PagePrivate(page)) | |
3552 | return 0; | |
3553 | ||
3554 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3555 | ||
fbca46eb | 3556 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3557 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); |
3558 | ||
3559 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
3560 | return 0; | |
3561 | } | |
3562 | ||
3563 | void clear_page_extent_mapped(struct page *page) | |
3564 | { | |
3565 | struct btrfs_fs_info *fs_info; | |
3566 | ||
3567 | ASSERT(page->mapping); | |
3568 | ||
d1b89bc0 | 3569 | if (!PagePrivate(page)) |
32443de3 QW |
3570 | return; |
3571 | ||
3572 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
fbca46eb | 3573 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3574 | return btrfs_detach_subpage(fs_info, page); |
3575 | ||
3576 | detach_page_private(page); | |
d1310b2e CM |
3577 | } |
3578 | ||
125bac01 MX |
3579 | static struct extent_map * |
3580 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3581 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3582 | { |
3583 | struct extent_map *em; | |
3584 | ||
3585 | if (em_cached && *em_cached) { | |
3586 | em = *em_cached; | |
cbc0e928 | 3587 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3588 | start < extent_map_end(em)) { |
490b54d6 | 3589 | refcount_inc(&em->refs); |
125bac01 MX |
3590 | return em; |
3591 | } | |
3592 | ||
3593 | free_extent_map(em); | |
3594 | *em_cached = NULL; | |
3595 | } | |
3596 | ||
1a5ee1e6 | 3597 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
c0347550 | 3598 | if (em_cached && !IS_ERR(em)) { |
125bac01 | 3599 | BUG_ON(*em_cached); |
490b54d6 | 3600 | refcount_inc(&em->refs); |
125bac01 MX |
3601 | *em_cached = em; |
3602 | } | |
3603 | return em; | |
3604 | } | |
d1310b2e CM |
3605 | /* |
3606 | * basic readpage implementation. Locked extent state structs are inserted | |
3607 | * into the tree that are removed when the IO is done (by the end_io | |
3608 | * handlers) | |
79787eaa | 3609 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3610 | * return 0 on success, otherwise return error |
d1310b2e | 3611 | */ |
7aab8b32 | 3612 | static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
390ed29b | 3613 | struct btrfs_bio_ctrl *bio_ctrl, |
0f208812 | 3614 | unsigned int read_flags, u64 *prev_em_start) |
d1310b2e CM |
3615 | { |
3616 | struct inode *inode = page->mapping->host; | |
92082d40 | 3617 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 3618 | u64 start = page_offset(page); |
8eec8296 | 3619 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3620 | u64 cur = start; |
3621 | u64 extent_offset; | |
3622 | u64 last_byte = i_size_read(inode); | |
3623 | u64 block_start; | |
3624 | u64 cur_end; | |
d1310b2e | 3625 | struct extent_map *em; |
baf863b9 | 3626 | int ret = 0; |
306e16ce | 3627 | size_t pg_offset = 0; |
d1310b2e CM |
3628 | size_t iosize; |
3629 | size_t blocksize = inode->i_sb->s_blocksize; | |
f657a31c | 3630 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3631 | |
32443de3 QW |
3632 | ret = set_page_extent_mapped(page); |
3633 | if (ret < 0) { | |
3634 | unlock_extent(tree, start, end); | |
92082d40 QW |
3635 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
3636 | unlock_page(page); | |
32443de3 QW |
3637 | goto out; |
3638 | } | |
d1310b2e | 3639 | |
09cbfeaf | 3640 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 3641 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3642 | |
3643 | if (zero_offset) { | |
09cbfeaf | 3644 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 3645 | memzero_page(page, zero_offset, iosize); |
c8b97818 | 3646 | flush_dcache_page(page); |
c8b97818 CM |
3647 | } |
3648 | } | |
92082d40 | 3649 | begin_page_read(fs_info, page); |
d1310b2e | 3650 | while (cur <= end) { |
4c37a793 | 3651 | unsigned long this_bio_flag = 0; |
005efedf | 3652 | bool force_bio_submit = false; |
0c64c33c | 3653 | u64 disk_bytenr; |
c8f2f24b | 3654 | |
6a404910 | 3655 | ASSERT(IS_ALIGNED(cur, fs_info->sectorsize)); |
d1310b2e | 3656 | if (cur >= last_byte) { |
507903b8 AJ |
3657 | struct extent_state *cached = NULL; |
3658 | ||
09cbfeaf | 3659 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 3660 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3661 | flush_dcache_page(page); |
d1310b2e | 3662 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3663 | &cached, GFP_NOFS); |
7f042a83 | 3664 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3665 | cur + iosize - 1, &cached); |
92082d40 | 3666 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
3667 | break; |
3668 | } | |
125bac01 | 3669 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3670 | end - cur + 1, em_cached); |
c0347550 | 3671 | if (IS_ERR(em)) { |
7f042a83 | 3672 | unlock_extent(tree, cur, end); |
92082d40 | 3673 | end_page_read(page, false, cur, end + 1 - cur); |
bbf0ea7e | 3674 | ret = PTR_ERR(em); |
d1310b2e CM |
3675 | break; |
3676 | } | |
d1310b2e CM |
3677 | extent_offset = cur - em->start; |
3678 | BUG_ON(extent_map_end(em) <= cur); | |
3679 | BUG_ON(end < cur); | |
3680 | ||
7f6ca7f2 DS |
3681 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
3682 | this_bio_flag = em->compress_type; | |
c8b97818 | 3683 | |
d1310b2e CM |
3684 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3685 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3686 | iosize = ALIGN(iosize, blocksize); |
2a5232a8 | 3687 | if (this_bio_flag != BTRFS_COMPRESS_NONE) |
0c64c33c | 3688 | disk_bytenr = em->block_start; |
949b3273 | 3689 | else |
0c64c33c | 3690 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 3691 | block_start = em->block_start; |
d899e052 YZ |
3692 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3693 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3694 | |
3695 | /* | |
3696 | * If we have a file range that points to a compressed extent | |
260db43c | 3697 | * and it's followed by a consecutive file range that points |
005efedf FM |
3698 | * to the same compressed extent (possibly with a different |
3699 | * offset and/or length, so it either points to the whole extent | |
3700 | * or only part of it), we must make sure we do not submit a | |
3701 | * single bio to populate the pages for the 2 ranges because | |
3702 | * this makes the compressed extent read zero out the pages | |
3703 | * belonging to the 2nd range. Imagine the following scenario: | |
3704 | * | |
3705 | * File layout | |
3706 | * [0 - 8K] [8K - 24K] | |
3707 | * | | | |
3708 | * | | | |
3709 | * points to extent X, points to extent X, | |
3710 | * offset 4K, length of 8K offset 0, length 16K | |
3711 | * | |
3712 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3713 | * | |
3714 | * If the bio to read the compressed extent covers both ranges, | |
3715 | * it will decompress extent X into the pages belonging to the | |
3716 | * first range and then it will stop, zeroing out the remaining | |
3717 | * pages that belong to the other range that points to extent X. | |
3718 | * So here we make sure we submit 2 bios, one for the first | |
3719 | * range and another one for the third range. Both will target | |
3720 | * the same physical extent from disk, but we can't currently | |
3721 | * make the compressed bio endio callback populate the pages | |
3722 | * for both ranges because each compressed bio is tightly | |
3723 | * coupled with a single extent map, and each range can have | |
3724 | * an extent map with a different offset value relative to the | |
3725 | * uncompressed data of our extent and different lengths. This | |
3726 | * is a corner case so we prioritize correctness over | |
3727 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3728 | */ | |
3729 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3730 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3731 | *prev_em_start != em->start) |
005efedf FM |
3732 | force_bio_submit = true; |
3733 | ||
3734 | if (prev_em_start) | |
8e928218 | 3735 | *prev_em_start = em->start; |
005efedf | 3736 | |
d1310b2e CM |
3737 | free_extent_map(em); |
3738 | em = NULL; | |
3739 | ||
3740 | /* we've found a hole, just zero and go on */ | |
3741 | if (block_start == EXTENT_MAP_HOLE) { | |
507903b8 AJ |
3742 | struct extent_state *cached = NULL; |
3743 | ||
d048b9c2 | 3744 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3745 | flush_dcache_page(page); |
d1310b2e CM |
3746 | |
3747 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3748 | &cached, GFP_NOFS); |
7f042a83 | 3749 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3750 | cur + iosize - 1, &cached); |
92082d40 | 3751 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3752 | cur = cur + iosize; |
306e16ce | 3753 | pg_offset += iosize; |
d1310b2e CM |
3754 | continue; |
3755 | } | |
3756 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3757 | if (test_range_bit(tree, cur, cur_end, |
3758 | EXTENT_UPTODATE, 1, NULL)) { | |
7f042a83 | 3759 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3760 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3761 | cur = cur + iosize; |
306e16ce | 3762 | pg_offset += iosize; |
d1310b2e CM |
3763 | continue; |
3764 | } | |
70dec807 CM |
3765 | /* we have an inline extent but it didn't get marked up |
3766 | * to date. Error out | |
3767 | */ | |
3768 | if (block_start == EXTENT_MAP_INLINE) { | |
7f042a83 | 3769 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3770 | end_page_read(page, false, cur, iosize); |
70dec807 | 3771 | cur = cur + iosize; |
306e16ce | 3772 | pg_offset += iosize; |
70dec807 CM |
3773 | continue; |
3774 | } | |
d1310b2e | 3775 | |
0ceb34bf | 3776 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
390ed29b QW |
3777 | bio_ctrl, page, disk_bytenr, iosize, |
3778 | pg_offset, | |
fd513000 | 3779 | end_bio_extent_readpage, 0, |
005efedf FM |
3780 | this_bio_flag, |
3781 | force_bio_submit); | |
ad3fc794 | 3782 | if (ret) { |
10f7f6f8 QW |
3783 | /* |
3784 | * We have to unlock the remaining range, or the page | |
3785 | * will never be unlocked. | |
3786 | */ | |
3787 | unlock_extent(tree, cur, end); | |
3788 | end_page_read(page, false, cur, end + 1 - cur); | |
baf863b9 | 3789 | goto out; |
edd33c99 | 3790 | } |
d1310b2e | 3791 | cur = cur + iosize; |
306e16ce | 3792 | pg_offset += iosize; |
d1310b2e | 3793 | } |
90a887c9 | 3794 | out: |
baf863b9 | 3795 | return ret; |
d1310b2e CM |
3796 | } |
3797 | ||
fdaf9a58 | 3798 | int btrfs_read_folio(struct file *file, struct folio *folio) |
7aab8b32 | 3799 | { |
fdaf9a58 | 3800 | struct page *page = &folio->page; |
7aab8b32 CH |
3801 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
3802 | u64 start = page_offset(page); | |
3803 | u64 end = start + PAGE_SIZE - 1; | |
3804 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; | |
3805 | int ret; | |
3806 | ||
3807 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); | |
3808 | ||
3809 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); | |
3810 | /* | |
3811 | * If btrfs_do_readpage() failed we will want to submit the assembled | |
3812 | * bio to do the cleanup. | |
3813 | */ | |
3814 | if (bio_ctrl.bio) | |
0f07003b | 3815 | submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.compress_type); |
7aab8b32 CH |
3816 | return ret; |
3817 | } | |
3818 | ||
b6660e80 | 3819 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
3820 | u64 start, u64 end, |
3821 | struct extent_map **em_cached, | |
3822 | struct btrfs_bio_ctrl *bio_ctrl, | |
3823 | u64 *prev_em_start) | |
9974090b | 3824 | { |
23d31bd4 | 3825 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3826 | int index; |
3827 | ||
b272ae22 | 3828 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3829 | |
3830 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 3831 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
0f208812 | 3832 | REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3833 | put_page(pages[index]); |
9974090b MX |
3834 | } |
3835 | } | |
3836 | ||
d1310b2e | 3837 | /* |
40f76580 CM |
3838 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3839 | * | |
5eaad97a | 3840 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3841 | * to write the page (copy into inline extent). In this case the IO has |
3842 | * been started and the page is already unlocked. | |
3843 | * | |
3844 | * This returns 0 if all went well (page still locked) | |
3845 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3846 | */ |
cd4c0bf9 | 3847 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
83f1b680 | 3848 | struct page *page, struct writeback_control *wbc) |
40f76580 | 3849 | { |
2749f7ef | 3850 | const u64 page_end = page_offset(page) + PAGE_SIZE - 1; |
cf3075fb | 3851 | u64 delalloc_start = page_offset(page); |
40f76580 | 3852 | u64 delalloc_to_write = 0; |
83f1b680 QW |
3853 | /* How many pages are started by btrfs_run_delalloc_range() */ |
3854 | unsigned long nr_written = 0; | |
40f76580 CM |
3855 | int ret; |
3856 | int page_started = 0; | |
3857 | ||
2749f7ef QW |
3858 | while (delalloc_start < page_end) { |
3859 | u64 delalloc_end = page_end; | |
3860 | bool found; | |
40f76580 | 3861 | |
cd4c0bf9 | 3862 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3863 | &delalloc_start, |
917aacec | 3864 | &delalloc_end); |
3522e903 | 3865 | if (!found) { |
40f76580 CM |
3866 | delalloc_start = delalloc_end + 1; |
3867 | continue; | |
3868 | } | |
cd4c0bf9 | 3869 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
83f1b680 | 3870 | delalloc_end, &page_started, &nr_written, wbc); |
40f76580 | 3871 | if (ret) { |
963e4db8 QW |
3872 | btrfs_page_set_error(inode->root->fs_info, page, |
3873 | page_offset(page), PAGE_SIZE); | |
7361b4ae | 3874 | return ret; |
40f76580 CM |
3875 | } |
3876 | /* | |
ea1754a0 KS |
3877 | * delalloc_end is already one less than the total length, so |
3878 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3879 | */ |
3880 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3881 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3882 | delalloc_start = delalloc_end + 1; |
3883 | } | |
3884 | if (wbc->nr_to_write < delalloc_to_write) { | |
3885 | int thresh = 8192; | |
3886 | ||
3887 | if (delalloc_to_write < thresh * 2) | |
3888 | thresh = delalloc_to_write; | |
3889 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3890 | thresh); | |
3891 | } | |
3892 | ||
83f1b680 | 3893 | /* Did btrfs_run_dealloc_range() already unlock and start the IO? */ |
40f76580 CM |
3894 | if (page_started) { |
3895 | /* | |
83f1b680 QW |
3896 | * We've unlocked the page, so we can't update the mapping's |
3897 | * writeback index, just update nr_to_write. | |
40f76580 | 3898 | */ |
83f1b680 | 3899 | wbc->nr_to_write -= nr_written; |
40f76580 CM |
3900 | return 1; |
3901 | } | |
3902 | ||
b69d1ee9 | 3903 | return 0; |
40f76580 CM |
3904 | } |
3905 | ||
c5ef5c6c QW |
3906 | /* |
3907 | * Find the first byte we need to write. | |
3908 | * | |
3909 | * For subpage, one page can contain several sectors, and | |
3910 | * __extent_writepage_io() will just grab all extent maps in the page | |
3911 | * range and try to submit all non-inline/non-compressed extents. | |
3912 | * | |
3913 | * This is a big problem for subpage, we shouldn't re-submit already written | |
3914 | * data at all. | |
3915 | * This function will lookup subpage dirty bit to find which range we really | |
3916 | * need to submit. | |
3917 | * | |
3918 | * Return the next dirty range in [@start, @end). | |
3919 | * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE. | |
3920 | */ | |
3921 | static void find_next_dirty_byte(struct btrfs_fs_info *fs_info, | |
3922 | struct page *page, u64 *start, u64 *end) | |
3923 | { | |
3924 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
72a69cd0 | 3925 | struct btrfs_subpage_info *spi = fs_info->subpage_info; |
c5ef5c6c QW |
3926 | u64 orig_start = *start; |
3927 | /* Declare as unsigned long so we can use bitmap ops */ | |
c5ef5c6c | 3928 | unsigned long flags; |
72a69cd0 | 3929 | int range_start_bit; |
c5ef5c6c QW |
3930 | int range_end_bit; |
3931 | ||
3932 | /* | |
3933 | * For regular sector size == page size case, since one page only | |
3934 | * contains one sector, we return the page offset directly. | |
3935 | */ | |
fbca46eb | 3936 | if (!btrfs_is_subpage(fs_info, page)) { |
c5ef5c6c QW |
3937 | *start = page_offset(page); |
3938 | *end = page_offset(page) + PAGE_SIZE; | |
3939 | return; | |
3940 | } | |
3941 | ||
72a69cd0 QW |
3942 | range_start_bit = spi->dirty_offset + |
3943 | (offset_in_page(orig_start) >> fs_info->sectorsize_bits); | |
3944 | ||
c5ef5c6c QW |
3945 | /* We should have the page locked, but just in case */ |
3946 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
3947 | bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit, |
3948 | spi->dirty_offset + spi->bitmap_nr_bits); | |
c5ef5c6c QW |
3949 | spin_unlock_irqrestore(&subpage->lock, flags); |
3950 | ||
72a69cd0 QW |
3951 | range_start_bit -= spi->dirty_offset; |
3952 | range_end_bit -= spi->dirty_offset; | |
3953 | ||
c5ef5c6c QW |
3954 | *start = page_offset(page) + range_start_bit * fs_info->sectorsize; |
3955 | *end = page_offset(page) + range_end_bit * fs_info->sectorsize; | |
3956 | } | |
3957 | ||
40f76580 CM |
3958 | /* |
3959 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3960 | * and does the loop to map the page into extents and bios. | |
3961 | * | |
3962 | * We return 1 if the IO is started and the page is unlocked, | |
3963 | * 0 if all went well (page still locked) | |
3964 | * < 0 if there were errors (page still locked) | |
3965 | */ | |
d4580fe2 | 3966 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3967 | struct page *page, |
3968 | struct writeback_control *wbc, | |
3969 | struct extent_page_data *epd, | |
3970 | loff_t i_size, | |
57e5ffeb | 3971 | int *nr_ret) |
d1310b2e | 3972 | { |
6bc5636a | 3973 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a129ffb8 QW |
3974 | u64 cur = page_offset(page); |
3975 | u64 end = cur + PAGE_SIZE - 1; | |
d1310b2e | 3976 | u64 extent_offset; |
d1310b2e | 3977 | u64 block_start; |
d1310b2e | 3978 | struct extent_map *em; |
44e5801f | 3979 | int saved_ret = 0; |
40f76580 CM |
3980 | int ret = 0; |
3981 | int nr = 0; | |
d8e3fb10 | 3982 | u32 opf = REQ_OP_WRITE; |
57e5ffeb | 3983 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
44e5801f | 3984 | bool has_error = false; |
40f76580 | 3985 | bool compressed; |
c8b97818 | 3986 | |
a129ffb8 | 3987 | ret = btrfs_writepage_cow_fixup(page); |
d75855b4 NB |
3988 | if (ret) { |
3989 | /* Fixup worker will requeue */ | |
5ab58055 | 3990 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3991 | unlock_page(page); |
3992 | return 1; | |
247e743c CM |
3993 | } |
3994 | ||
11c8349b CM |
3995 | /* |
3996 | * we don't want to touch the inode after unlocking the page, | |
3997 | * so we update the mapping writeback index now | |
3998 | */ | |
572f3dad | 3999 | wbc->nr_to_write--; |
771ed689 | 4000 | |
d1310b2e | 4001 | while (cur <= end) { |
0c64c33c | 4002 | u64 disk_bytenr; |
40f76580 | 4003 | u64 em_end; |
c5ef5c6c QW |
4004 | u64 dirty_range_start = cur; |
4005 | u64 dirty_range_end; | |
6bc5636a | 4006 | u32 iosize; |
58409edd | 4007 | |
40f76580 | 4008 | if (cur >= i_size) { |
38a39ac7 | 4009 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
25c1252a | 4010 | end, true); |
cc1d0d93 QW |
4011 | /* |
4012 | * This range is beyond i_size, thus we don't need to | |
4013 | * bother writing back. | |
4014 | * But we still need to clear the dirty subpage bit, or | |
4015 | * the next time the page gets dirtied, we will try to | |
4016 | * writeback the sectors with subpage dirty bits, | |
4017 | * causing writeback without ordered extent. | |
4018 | */ | |
4019 | btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur); | |
d1310b2e CM |
4020 | break; |
4021 | } | |
c5ef5c6c QW |
4022 | |
4023 | find_next_dirty_byte(fs_info, page, &dirty_range_start, | |
4024 | &dirty_range_end); | |
4025 | if (cur < dirty_range_start) { | |
4026 | cur = dirty_range_start; | |
4027 | continue; | |
4028 | } | |
4029 | ||
d4580fe2 | 4030 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c0347550 | 4031 | if (IS_ERR(em)) { |
c5ef5c6c | 4032 | btrfs_page_set_error(fs_info, page, cur, end - cur + 1); |
61391d56 | 4033 | ret = PTR_ERR_OR_ZERO(em); |
44e5801f QW |
4034 | has_error = true; |
4035 | if (!saved_ret) | |
4036 | saved_ret = ret; | |
d1310b2e CM |
4037 | break; |
4038 | } | |
4039 | ||
4040 | extent_offset = cur - em->start; | |
40f76580 | 4041 | em_end = extent_map_end(em); |
6bc5636a QW |
4042 | ASSERT(cur <= em_end); |
4043 | ASSERT(cur < end); | |
4044 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
4045 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 4046 | block_start = em->block_start; |
c8b97818 | 4047 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
4048 | disk_bytenr = em->block_start + extent_offset; |
4049 | ||
c5ef5c6c QW |
4050 | /* |
4051 | * Note that em_end from extent_map_end() and dirty_range_end from | |
4052 | * find_next_dirty_byte() are all exclusive | |
4053 | */ | |
4054 | iosize = min(min(em_end, end + 1), dirty_range_end) - cur; | |
d8e3fb10 | 4055 | |
e380adfc | 4056 | if (btrfs_use_zone_append(inode, em->block_start)) |
d8e3fb10 NA |
4057 | opf = REQ_OP_ZONE_APPEND; |
4058 | ||
d1310b2e CM |
4059 | free_extent_map(em); |
4060 | em = NULL; | |
4061 | ||
c8b97818 CM |
4062 | /* |
4063 | * compressed and inline extents are written through other | |
4064 | * paths in the FS | |
4065 | */ | |
4066 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 4067 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 4068 | if (compressed) |
c8b97818 | 4069 | nr++; |
c8b04030 | 4070 | else |
38a39ac7 | 4071 | btrfs_writepage_endio_finish_ordered(inode, |
25c1252a | 4072 | page, cur, cur + iosize - 1, true); |
cc1d0d93 | 4073 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); |
c8b97818 | 4074 | cur += iosize; |
d1310b2e CM |
4075 | continue; |
4076 | } | |
c8b97818 | 4077 | |
d2a91064 | 4078 | btrfs_set_range_writeback(inode, cur, cur + iosize - 1); |
58409edd | 4079 | if (!PageWriteback(page)) { |
d4580fe2 | 4080 | btrfs_err(inode->root->fs_info, |
58409edd DS |
4081 | "page %lu not writeback, cur %llu end %llu", |
4082 | page->index, cur, end); | |
d1310b2e | 4083 | } |
7f3c74fb | 4084 | |
c5ef5c6c QW |
4085 | /* |
4086 | * Although the PageDirty bit is cleared before entering this | |
4087 | * function, subpage dirty bit is not cleared. | |
4088 | * So clear subpage dirty bit here so next time we won't submit | |
4089 | * page for range already written to disk. | |
4090 | */ | |
4091 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); | |
4092 | ||
390ed29b QW |
4093 | ret = submit_extent_page(opf | write_flags, wbc, |
4094 | &epd->bio_ctrl, page, | |
d8e3fb10 | 4095 | disk_bytenr, iosize, |
390ed29b | 4096 | cur - page_offset(page), |
58409edd | 4097 | end_bio_extent_writepage, |
390ed29b | 4098 | 0, 0, false); |
fe01aa65 | 4099 | if (ret) { |
44e5801f QW |
4100 | has_error = true; |
4101 | if (!saved_ret) | |
4102 | saved_ret = ret; | |
4103 | ||
c5ef5c6c | 4104 | btrfs_page_set_error(fs_info, page, cur, iosize); |
fe01aa65 | 4105 | if (PageWriteback(page)) |
c5ef5c6c QW |
4106 | btrfs_page_clear_writeback(fs_info, page, cur, |
4107 | iosize); | |
fe01aa65 | 4108 | } |
d1310b2e | 4109 | |
6bc5636a | 4110 | cur += iosize; |
d1310b2e CM |
4111 | nr++; |
4112 | } | |
cc1d0d93 QW |
4113 | /* |
4114 | * If we finish without problem, we should not only clear page dirty, | |
4115 | * but also empty subpage dirty bits | |
4116 | */ | |
44e5801f | 4117 | if (!has_error) |
cc1d0d93 | 4118 | btrfs_page_assert_not_dirty(fs_info, page); |
44e5801f QW |
4119 | else |
4120 | ret = saved_ret; | |
40f76580 | 4121 | *nr_ret = nr; |
40f76580 CM |
4122 | return ret; |
4123 | } | |
4124 | ||
4125 | /* | |
4126 | * the writepage semantics are similar to regular writepage. extent | |
4127 | * records are inserted to lock ranges in the tree, and as dirty areas | |
4128 | * are found, they are marked writeback. Then the lock bits are removed | |
4129 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
4130 | * |
4131 | * Return 0 if everything goes well. | |
4132 | * Return <0 for error. | |
40f76580 CM |
4133 | */ |
4134 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 4135 | struct extent_page_data *epd) |
40f76580 | 4136 | { |
8e1dec8e | 4137 | struct folio *folio = page_folio(page); |
40f76580 | 4138 | struct inode *inode = page->mapping->host; |
e55a0de1 | 4139 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
cf3075fb QW |
4140 | const u64 page_start = page_offset(page); |
4141 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
40f76580 CM |
4142 | int ret; |
4143 | int nr = 0; | |
eb70d222 | 4144 | size_t pg_offset; |
40f76580 | 4145 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 4146 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 | 4147 | |
40f76580 CM |
4148 | trace___extent_writepage(page, inode, wbc); |
4149 | ||
4150 | WARN_ON(!PageLocked(page)); | |
4151 | ||
963e4db8 QW |
4152 | btrfs_page_clear_error(btrfs_sb(inode->i_sb), page, |
4153 | page_offset(page), PAGE_SIZE); | |
40f76580 | 4154 | |
7073017a | 4155 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
4156 | if (page->index > end_index || |
4157 | (page->index == end_index && !pg_offset)) { | |
8e1dec8e MWO |
4158 | folio_invalidate(folio, 0, folio_size(folio)); |
4159 | folio_unlock(folio); | |
40f76580 CM |
4160 | return 0; |
4161 | } | |
4162 | ||
4163 | if (page->index == end_index) { | |
d048b9c2 | 4164 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 CM |
4165 | flush_dcache_page(page); |
4166 | } | |
4167 | ||
32443de3 QW |
4168 | ret = set_page_extent_mapped(page); |
4169 | if (ret < 0) { | |
4170 | SetPageError(page); | |
4171 | goto done; | |
4172 | } | |
40f76580 | 4173 | |
7789a55a | 4174 | if (!epd->extent_locked) { |
83f1b680 | 4175 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc); |
7789a55a | 4176 | if (ret == 1) |
169d2c87 | 4177 | return 0; |
7789a55a NB |
4178 | if (ret) |
4179 | goto done; | |
4180 | } | |
40f76580 | 4181 | |
d4580fe2 | 4182 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
83f1b680 | 4183 | &nr); |
40f76580 | 4184 | if (ret == 1) |
169d2c87 | 4185 | return 0; |
40f76580 | 4186 | |
d1310b2e CM |
4187 | done: |
4188 | if (nr == 0) { | |
4189 | /* make sure the mapping tag for page dirty gets cleared */ | |
4190 | set_page_writeback(page); | |
4191 | end_page_writeback(page); | |
4192 | } | |
963e4db8 QW |
4193 | /* |
4194 | * Here we used to have a check for PageError() and then set @ret and | |
4195 | * call end_extent_writepage(). | |
4196 | * | |
4197 | * But in fact setting @ret here will cause different error paths | |
4198 | * between subpage and regular sectorsize. | |
4199 | * | |
4200 | * For regular page size, we never submit current page, but only add | |
4201 | * current page to current bio. | |
4202 | * The bio submission can only happen in next page. | |
4203 | * Thus if we hit the PageError() branch, @ret is already set to | |
4204 | * non-zero value and will not get updated for regular sectorsize. | |
4205 | * | |
4206 | * But for subpage case, it's possible we submit part of current page, | |
4207 | * thus can get PageError() set by submitted bio of the same page, | |
4208 | * while our @ret is still 0. | |
4209 | * | |
4210 | * So here we unify the behavior and don't set @ret. | |
4211 | * Error can still be properly passed to higher layer as page will | |
4212 | * be set error, here we just don't handle the IO failure. | |
4213 | * | |
4214 | * NOTE: This is just a hotfix for subpage. | |
4215 | * The root fix will be properly ending ordered extent when we hit | |
4216 | * an error during writeback. | |
4217 | * | |
4218 | * But that needs a bigger refactoring, as we not only need to grab the | |
4219 | * submitted OE, but also need to know exactly at which bytenr we hit | |
4220 | * the error. | |
4221 | * Currently the full page based __extent_writepage_io() is not | |
4222 | * capable of that. | |
4223 | */ | |
4224 | if (PageError(page)) | |
cf3075fb | 4225 | end_extent_writepage(page, ret, page_start, page_end); |
e55a0de1 QW |
4226 | if (epd->extent_locked) { |
4227 | /* | |
4228 | * If epd->extent_locked, it's from extent_write_locked_range(), | |
4229 | * the page can either be locked by lock_page() or | |
4230 | * process_one_page(). | |
4231 | * Let btrfs_page_unlock_writer() handle both cases. | |
4232 | */ | |
4233 | ASSERT(wbc); | |
4234 | btrfs_page_unlock_writer(fs_info, page, wbc->range_start, | |
4235 | wbc->range_end + 1 - wbc->range_start); | |
4236 | } else { | |
4237 | unlock_page(page); | |
4238 | } | |
3065976b | 4239 | ASSERT(ret <= 0); |
40f76580 | 4240 | return ret; |
d1310b2e CM |
4241 | } |
4242 | ||
fd8b2b61 | 4243 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 4244 | { |
74316201 N |
4245 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
4246 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
4247 | } |
4248 | ||
18dfa711 FM |
4249 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
4250 | { | |
4251 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4252 | smp_mb__after_atomic(); | |
4253 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
4254 | } | |
4255 | ||
2e3c2513 | 4256 | /* |
a3efb2f0 | 4257 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 4258 | * |
a3efb2f0 QW |
4259 | * May try to flush write bio if we can't get the lock. |
4260 | * | |
4261 | * Return 0 if the extent buffer doesn't need to be submitted. | |
4262 | * (E.g. the extent buffer is not dirty) | |
4263 | * Return >0 is the extent buffer is submitted to bio. | |
4264 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 4265 | */ |
9df76fb5 | 4266 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 4267 | struct extent_page_data *epd) |
0b32f4bb | 4268 | { |
9df76fb5 | 4269 | struct btrfs_fs_info *fs_info = eb->fs_info; |
c9583ada | 4270 | int i, num_pages; |
0b32f4bb JB |
4271 | int flush = 0; |
4272 | int ret = 0; | |
4273 | ||
4274 | if (!btrfs_try_tree_write_lock(eb)) { | |
c9583ada | 4275 | flush_write_bio(epd); |
2e3c2513 | 4276 | flush = 1; |
0b32f4bb JB |
4277 | btrfs_tree_lock(eb); |
4278 | } | |
4279 | ||
4280 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
4281 | btrfs_tree_unlock(eb); | |
4282 | if (!epd->sync_io) | |
4283 | return 0; | |
4284 | if (!flush) { | |
c9583ada | 4285 | flush_write_bio(epd); |
0b32f4bb JB |
4286 | flush = 1; |
4287 | } | |
a098d8e8 CM |
4288 | while (1) { |
4289 | wait_on_extent_buffer_writeback(eb); | |
4290 | btrfs_tree_lock(eb); | |
4291 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
4292 | break; | |
0b32f4bb | 4293 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
4294 | } |
4295 | } | |
4296 | ||
51561ffe JB |
4297 | /* |
4298 | * We need to do this to prevent races in people who check if the eb is | |
4299 | * under IO since we can end up having no IO bits set for a short period | |
4300 | * of time. | |
4301 | */ | |
4302 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
4303 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
4304 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 4305 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 4306 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
4307 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4308 | -eb->len, | |
4309 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 4310 | ret = 1; |
51561ffe JB |
4311 | } else { |
4312 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
4313 | } |
4314 | ||
4315 | btrfs_tree_unlock(eb); | |
4316 | ||
f3156df9 QW |
4317 | /* |
4318 | * Either we don't need to submit any tree block, or we're submitting | |
4319 | * subpage eb. | |
4320 | * Subpage metadata doesn't use page locking at all, so we can skip | |
4321 | * the page locking. | |
4322 | */ | |
fbca46eb | 4323 | if (!ret || fs_info->nodesize < PAGE_SIZE) |
0b32f4bb JB |
4324 | return ret; |
4325 | ||
65ad0104 | 4326 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4327 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4328 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4329 | |
4330 | if (!trylock_page(p)) { | |
4331 | if (!flush) { | |
c9583ada | 4332 | flush_write_bio(epd); |
0b32f4bb JB |
4333 | flush = 1; |
4334 | } | |
4335 | lock_page(p); | |
4336 | } | |
4337 | } | |
4338 | ||
2e3c2513 | 4339 | return ret; |
0b32f4bb JB |
4340 | } |
4341 | ||
5a2c6075 | 4342 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 4343 | { |
5a2c6075 | 4344 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 4345 | |
5a2c6075 | 4346 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
4347 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
4348 | return; | |
4349 | ||
c2e39305 JB |
4350 | /* |
4351 | * A read may stumble upon this buffer later, make sure that it gets an | |
4352 | * error and knows there was an error. | |
4353 | */ | |
4354 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4355 | ||
68b85589 JB |
4356 | /* |
4357 | * We need to set the mapping with the io error as well because a write | |
4358 | * error will flip the file system readonly, and then syncfs() will | |
4359 | * return a 0 because we are readonly if we don't modify the err seq for | |
4360 | * the superblock. | |
4361 | */ | |
4362 | mapping_set_error(page->mapping, -EIO); | |
4363 | ||
eb5b64f1 DZ |
4364 | /* |
4365 | * If we error out, we should add back the dirty_metadata_bytes | |
4366 | * to make it consistent. | |
4367 | */ | |
eb5b64f1 DZ |
4368 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4369 | eb->len, fs_info->dirty_metadata_batch); | |
4370 | ||
656f30db FM |
4371 | /* |
4372 | * If writeback for a btree extent that doesn't belong to a log tree | |
4373 | * failed, increment the counter transaction->eb_write_errors. | |
4374 | * We do this because while the transaction is running and before it's | |
4375 | * committing (when we call filemap_fdata[write|wait]_range against | |
4376 | * the btree inode), we might have | |
4377 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
4378 | * returns an error or an error happens during writeback, when we're | |
4379 | * committing the transaction we wouldn't know about it, since the pages | |
4380 | * can be no longer dirty nor marked anymore for writeback (if a | |
4381 | * subsequent modification to the extent buffer didn't happen before the | |
4382 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
4383 | * able to find the pages tagged with SetPageError at transaction | |
4384 | * commit time. So if this happens we must abort the transaction, | |
4385 | * otherwise we commit a super block with btree roots that point to | |
4386 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
4387 | * or the content of some node/leaf from a past generation that got | |
4388 | * cowed or deleted and is no longer valid. | |
4389 | * | |
4390 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
4391 | * not be enough - we need to distinguish between log tree extents vs | |
4392 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
4393 | * will catch and clear such errors in the mapping - and that call might | |
4394 | * be from a log sync and not from a transaction commit. Also, checking | |
4395 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
4396 | * not done and would not be reliable - the eb might have been released | |
4397 | * from memory and reading it back again means that flag would not be | |
4398 | * set (since it's a runtime flag, not persisted on disk). | |
4399 | * | |
4400 | * Using the flags below in the btree inode also makes us achieve the | |
4401 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
4402 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
4403 | * is called, the writeback for all dirty pages had already finished | |
4404 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
4405 | * filemap_fdatawait_range() would return success, as it could not know | |
4406 | * that writeback errors happened (the pages were no longer tagged for | |
4407 | * writeback). | |
4408 | */ | |
4409 | switch (eb->log_index) { | |
4410 | case -1: | |
5a2c6075 | 4411 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
4412 | break; |
4413 | case 0: | |
5a2c6075 | 4414 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
4415 | break; |
4416 | case 1: | |
5a2c6075 | 4417 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
4418 | break; |
4419 | default: | |
4420 | BUG(); /* unexpected, logic error */ | |
4421 | } | |
4422 | } | |
4423 | ||
2f3186d8 QW |
4424 | /* |
4425 | * The endio specific version which won't touch any unsafe spinlock in endio | |
4426 | * context. | |
4427 | */ | |
4428 | static struct extent_buffer *find_extent_buffer_nolock( | |
4429 | struct btrfs_fs_info *fs_info, u64 start) | |
4430 | { | |
4431 | struct extent_buffer *eb; | |
4432 | ||
4433 | rcu_read_lock(); | |
01cd3909 DS |
4434 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
4435 | start >> fs_info->sectorsize_bits); | |
2f3186d8 QW |
4436 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4437 | rcu_read_unlock(); | |
4438 | return eb; | |
4439 | } | |
4440 | rcu_read_unlock(); | |
4441 | return NULL; | |
4442 | } | |
4443 | ||
4444 | /* | |
4445 | * The endio function for subpage extent buffer write. | |
4446 | * | |
4447 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
4448 | * after all extent buffers in the page has finished their writeback. | |
4449 | */ | |
fa04c165 | 4450 | static void end_bio_subpage_eb_writepage(struct bio *bio) |
2f3186d8 | 4451 | { |
fa04c165 | 4452 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
4453 | struct bio_vec *bvec; |
4454 | struct bvec_iter_all iter_all; | |
4455 | ||
fa04c165 | 4456 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
fbca46eb | 4457 | ASSERT(fs_info->nodesize < PAGE_SIZE); |
fa04c165 | 4458 | |
2f3186d8 QW |
4459 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
4460 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
4461 | struct page *page = bvec->bv_page; | |
4462 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
4463 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
4464 | u64 cur_bytenr = bvec_start; | |
4465 | ||
4466 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
4467 | ||
4468 | /* Iterate through all extent buffers in the range */ | |
4469 | while (cur_bytenr <= bvec_end) { | |
4470 | struct extent_buffer *eb; | |
4471 | int done; | |
4472 | ||
4473 | /* | |
4474 | * Here we can't use find_extent_buffer(), as it may | |
4475 | * try to lock eb->refs_lock, which is not safe in endio | |
4476 | * context. | |
4477 | */ | |
4478 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
4479 | ASSERT(eb); | |
4480 | ||
4481 | cur_bytenr = eb->start + eb->len; | |
4482 | ||
4483 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
4484 | done = atomic_dec_and_test(&eb->io_pages); | |
4485 | ASSERT(done); | |
4486 | ||
4487 | if (bio->bi_status || | |
4488 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
4489 | ClearPageUptodate(page); | |
4490 | set_btree_ioerr(page, eb); | |
4491 | } | |
4492 | ||
4493 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
4494 | eb->len); | |
4495 | end_extent_buffer_writeback(eb); | |
4496 | /* | |
4497 | * free_extent_buffer() will grab spinlock which is not | |
4498 | * safe in endio context. Thus here we manually dec | |
4499 | * the ref. | |
4500 | */ | |
4501 | atomic_dec(&eb->refs); | |
4502 | } | |
4503 | } | |
4504 | bio_put(bio); | |
4505 | } | |
4506 | ||
4246a0b6 | 4507 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 4508 | { |
2c30c71b | 4509 | struct bio_vec *bvec; |
0b32f4bb | 4510 | struct extent_buffer *eb; |
2b070cfe | 4511 | int done; |
6dc4f100 | 4512 | struct bvec_iter_all iter_all; |
0b32f4bb | 4513 | |
c09abff8 | 4514 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 4515 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
4516 | struct page *page = bvec->bv_page; |
4517 | ||
0b32f4bb JB |
4518 | eb = (struct extent_buffer *)page->private; |
4519 | BUG_ON(!eb); | |
4520 | done = atomic_dec_and_test(&eb->io_pages); | |
4521 | ||
4e4cbee9 | 4522 | if (bio->bi_status || |
4246a0b6 | 4523 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 4524 | ClearPageUptodate(page); |
5a2c6075 | 4525 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
4526 | } |
4527 | ||
4528 | end_page_writeback(page); | |
4529 | ||
4530 | if (!done) | |
4531 | continue; | |
4532 | ||
4533 | end_extent_buffer_writeback(eb); | |
2c30c71b | 4534 | } |
0b32f4bb JB |
4535 | |
4536 | bio_put(bio); | |
0b32f4bb JB |
4537 | } |
4538 | ||
fa04c165 QW |
4539 | static void prepare_eb_write(struct extent_buffer *eb) |
4540 | { | |
4541 | u32 nritems; | |
4542 | unsigned long start; | |
4543 | unsigned long end; | |
4544 | ||
4545 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
4546 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
4547 | ||
4548 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
4549 | nritems = btrfs_header_nritems(eb); | |
4550 | if (btrfs_header_level(eb) > 0) { | |
4551 | end = btrfs_node_key_ptr_offset(nritems); | |
4552 | memzero_extent_buffer(eb, end, eb->len - end); | |
4553 | } else { | |
4554 | /* | |
4555 | * Leaf: | |
4556 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
4557 | */ | |
4558 | start = btrfs_item_nr_offset(nritems); | |
4559 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); | |
4560 | memzero_extent_buffer(eb, start, end - start); | |
4561 | } | |
4562 | } | |
4563 | ||
35b6ddfa QW |
4564 | /* |
4565 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
4566 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
4567 | */ |
4568 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
4569 | struct writeback_control *wbc, | |
4570 | struct extent_page_data *epd) | |
4571 | { | |
4572 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4573 | struct page *page = eb->pages[0]; | |
4574 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; | |
4575 | bool no_dirty_ebs = false; | |
4576 | int ret; | |
4577 | ||
fa04c165 QW |
4578 | prepare_eb_write(eb); |
4579 | ||
35b6ddfa QW |
4580 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
4581 | lock_page(page); | |
4582 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
4583 | ||
4584 | /* Check if this is the last dirty bit to update nr_written */ | |
4585 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
4586 | eb->start, eb->len); | |
4587 | if (no_dirty_ebs) | |
4588 | clear_page_dirty_for_io(page); | |
4589 | ||
390ed29b QW |
4590 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
4591 | &epd->bio_ctrl, page, eb->start, eb->len, | |
4592 | eb->start - page_offset(page), | |
fa04c165 | 4593 | end_bio_subpage_eb_writepage, 0, 0, false); |
35b6ddfa QW |
4594 | if (ret) { |
4595 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
4596 | set_btree_ioerr(page, eb); | |
4597 | unlock_page(page); | |
4598 | ||
4599 | if (atomic_dec_and_test(&eb->io_pages)) | |
4600 | end_extent_buffer_writeback(eb); | |
4601 | return -EIO; | |
4602 | } | |
4603 | unlock_page(page); | |
4604 | /* | |
4605 | * Submission finished without problem, if no range of the page is | |
4606 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
4607 | */ | |
4608 | if (no_dirty_ebs) | |
572f3dad | 4609 | wbc->nr_to_write--; |
35b6ddfa QW |
4610 | return ret; |
4611 | } | |
4612 | ||
0e378df1 | 4613 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
4614 | struct writeback_control *wbc, |
4615 | struct extent_page_data *epd) | |
4616 | { | |
0c64c33c | 4617 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 4618 | int i, num_pages; |
ff40adf7 | 4619 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 4620 | int ret = 0; |
0b32f4bb | 4621 | |
fa04c165 | 4622 | prepare_eb_write(eb); |
35b6ddfa | 4623 | |
fa04c165 | 4624 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4625 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4626 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4627 | |
4628 | clear_page_dirty_for_io(p); | |
4629 | set_page_writeback(p); | |
0ceb34bf | 4630 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
390ed29b QW |
4631 | &epd->bio_ctrl, p, disk_bytenr, |
4632 | PAGE_SIZE, 0, | |
1f7ad75b | 4633 | end_bio_extent_buffer_writepage, |
390ed29b | 4634 | 0, 0, false); |
0b32f4bb | 4635 | if (ret) { |
5a2c6075 | 4636 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
4637 | if (PageWriteback(p)) |
4638 | end_page_writeback(p); | |
0b32f4bb JB |
4639 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
4640 | end_extent_buffer_writeback(eb); | |
4641 | ret = -EIO; | |
4642 | break; | |
4643 | } | |
0c64c33c | 4644 | disk_bytenr += PAGE_SIZE; |
572f3dad | 4645 | wbc->nr_to_write--; |
0b32f4bb JB |
4646 | unlock_page(p); |
4647 | } | |
4648 | ||
4649 | if (unlikely(ret)) { | |
4650 | for (; i < num_pages; i++) { | |
bbf65cf0 | 4651 | struct page *p = eb->pages[i]; |
81465028 | 4652 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
4653 | unlock_page(p); |
4654 | } | |
4655 | } | |
4656 | ||
4657 | return ret; | |
4658 | } | |
4659 | ||
c4aec299 QW |
4660 | /* |
4661 | * Submit one subpage btree page. | |
4662 | * | |
4663 | * The main difference to submit_eb_page() is: | |
4664 | * - Page locking | |
4665 | * For subpage, we don't rely on page locking at all. | |
4666 | * | |
4667 | * - Flush write bio | |
4668 | * We only flush bio if we may be unable to fit current extent buffers into | |
4669 | * current bio. | |
4670 | * | |
4671 | * Return >=0 for the number of submitted extent buffers. | |
4672 | * Return <0 for fatal error. | |
4673 | */ | |
4674 | static int submit_eb_subpage(struct page *page, | |
4675 | struct writeback_control *wbc, | |
4676 | struct extent_page_data *epd) | |
4677 | { | |
4678 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
4679 | int submitted = 0; | |
4680 | u64 page_start = page_offset(page); | |
4681 | int bit_start = 0; | |
c4aec299 QW |
4682 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; |
4683 | int ret; | |
4684 | ||
4685 | /* Lock and write each dirty extent buffers in the range */ | |
72a69cd0 | 4686 | while (bit_start < fs_info->subpage_info->bitmap_nr_bits) { |
c4aec299 QW |
4687 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; |
4688 | struct extent_buffer *eb; | |
4689 | unsigned long flags; | |
4690 | u64 start; | |
4691 | ||
4692 | /* | |
4693 | * Take private lock to ensure the subpage won't be detached | |
4694 | * in the meantime. | |
4695 | */ | |
4696 | spin_lock(&page->mapping->private_lock); | |
4697 | if (!PagePrivate(page)) { | |
4698 | spin_unlock(&page->mapping->private_lock); | |
4699 | break; | |
4700 | } | |
4701 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
4702 | if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset, |
4703 | subpage->bitmaps)) { | |
c4aec299 QW |
4704 | spin_unlock_irqrestore(&subpage->lock, flags); |
4705 | spin_unlock(&page->mapping->private_lock); | |
4706 | bit_start++; | |
4707 | continue; | |
4708 | } | |
4709 | ||
4710 | start = page_start + bit_start * fs_info->sectorsize; | |
4711 | bit_start += sectors_per_node; | |
4712 | ||
4713 | /* | |
4714 | * Here we just want to grab the eb without touching extra | |
4715 | * spin locks, so call find_extent_buffer_nolock(). | |
4716 | */ | |
4717 | eb = find_extent_buffer_nolock(fs_info, start); | |
4718 | spin_unlock_irqrestore(&subpage->lock, flags); | |
4719 | spin_unlock(&page->mapping->private_lock); | |
4720 | ||
4721 | /* | |
4722 | * The eb has already reached 0 refs thus find_extent_buffer() | |
4723 | * doesn't return it. We don't need to write back such eb | |
4724 | * anyway. | |
4725 | */ | |
4726 | if (!eb) | |
4727 | continue; | |
4728 | ||
4729 | ret = lock_extent_buffer_for_io(eb, epd); | |
4730 | if (ret == 0) { | |
4731 | free_extent_buffer(eb); | |
4732 | continue; | |
4733 | } | |
4734 | if (ret < 0) { | |
4735 | free_extent_buffer(eb); | |
4736 | goto cleanup; | |
4737 | } | |
fa04c165 | 4738 | ret = write_one_subpage_eb(eb, wbc, epd); |
c4aec299 QW |
4739 | free_extent_buffer(eb); |
4740 | if (ret < 0) | |
4741 | goto cleanup; | |
4742 | submitted++; | |
4743 | } | |
4744 | return submitted; | |
4745 | ||
4746 | cleanup: | |
4747 | /* We hit error, end bio for the submitted extent buffers */ | |
4748 | end_write_bio(epd, ret); | |
4749 | return ret; | |
4750 | } | |
4751 | ||
f91e0d0c QW |
4752 | /* |
4753 | * Submit all page(s) of one extent buffer. | |
4754 | * | |
4755 | * @page: the page of one extent buffer | |
4756 | * @eb_context: to determine if we need to submit this page, if current page | |
4757 | * belongs to this eb, we don't need to submit | |
4758 | * | |
4759 | * The caller should pass each page in their bytenr order, and here we use | |
4760 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
4761 | * | |
4762 | * If we have, we just skip until we hit a new page that doesn't belong to | |
4763 | * current @eb_context. | |
4764 | * | |
4765 | * If not, we submit all the page(s) of the extent buffer. | |
4766 | * | |
4767 | * Return >0 if we have submitted the extent buffer successfully. | |
4768 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
4769 | * previous call. | |
4770 | * Return <0 for fatal error. | |
4771 | */ | |
4772 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
4773 | struct extent_page_data *epd, | |
4774 | struct extent_buffer **eb_context) | |
4775 | { | |
4776 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 4777 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
4778 | struct extent_buffer *eb; |
4779 | int ret; | |
4780 | ||
4781 | if (!PagePrivate(page)) | |
4782 | return 0; | |
4783 | ||
fbca46eb | 4784 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
c4aec299 QW |
4785 | return submit_eb_subpage(page, wbc, epd); |
4786 | ||
f91e0d0c QW |
4787 | spin_lock(&mapping->private_lock); |
4788 | if (!PagePrivate(page)) { | |
4789 | spin_unlock(&mapping->private_lock); | |
4790 | return 0; | |
4791 | } | |
4792 | ||
4793 | eb = (struct extent_buffer *)page->private; | |
4794 | ||
4795 | /* | |
4796 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
4797 | * crashing the machine for something we can survive anyway. | |
4798 | */ | |
4799 | if (WARN_ON(!eb)) { | |
4800 | spin_unlock(&mapping->private_lock); | |
4801 | return 0; | |
4802 | } | |
4803 | ||
4804 | if (eb == *eb_context) { | |
4805 | spin_unlock(&mapping->private_lock); | |
4806 | return 0; | |
4807 | } | |
4808 | ret = atomic_inc_not_zero(&eb->refs); | |
4809 | spin_unlock(&mapping->private_lock); | |
4810 | if (!ret) | |
4811 | return 0; | |
4812 | ||
0bc09ca1 NA |
4813 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
4814 | /* | |
4815 | * If for_sync, this hole will be filled with | |
4816 | * trasnsaction commit. | |
4817 | */ | |
4818 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
4819 | ret = -EAGAIN; | |
4820 | else | |
4821 | ret = 0; | |
4822 | free_extent_buffer(eb); | |
4823 | return ret; | |
4824 | } | |
4825 | ||
f91e0d0c QW |
4826 | *eb_context = eb; |
4827 | ||
4828 | ret = lock_extent_buffer_for_io(eb, epd); | |
4829 | if (ret <= 0) { | |
0bc09ca1 NA |
4830 | btrfs_revert_meta_write_pointer(cache, eb); |
4831 | if (cache) | |
4832 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
4833 | free_extent_buffer(eb); |
4834 | return ret; | |
4835 | } | |
be1a1d7a | 4836 | if (cache) { |
d3e29967 NB |
4837 | /* |
4838 | * Implies write in zoned mode. Mark the last eb in a block group. | |
4839 | */ | |
56fbb0a4 | 4840 | btrfs_schedule_zone_finish_bg(cache, eb); |
d3e29967 | 4841 | btrfs_put_block_group(cache); |
be1a1d7a | 4842 | } |
f91e0d0c QW |
4843 | ret = write_one_eb(eb, wbc, epd); |
4844 | free_extent_buffer(eb); | |
4845 | if (ret < 0) | |
4846 | return ret; | |
4847 | return 1; | |
4848 | } | |
4849 | ||
0b32f4bb JB |
4850 | int btree_write_cache_pages(struct address_space *mapping, |
4851 | struct writeback_control *wbc) | |
4852 | { | |
f91e0d0c | 4853 | struct extent_buffer *eb_context = NULL; |
0b32f4bb | 4854 | struct extent_page_data epd = { |
390ed29b | 4855 | .bio_ctrl = { 0 }, |
0b32f4bb JB |
4856 | .extent_locked = 0, |
4857 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4858 | }; | |
b3ff8f1d | 4859 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4860 | int ret = 0; |
4861 | int done = 0; | |
4862 | int nr_to_write_done = 0; | |
4863 | struct pagevec pvec; | |
4864 | int nr_pages; | |
4865 | pgoff_t index; | |
4866 | pgoff_t end; /* Inclusive */ | |
4867 | int scanned = 0; | |
10bbd235 | 4868 | xa_mark_t tag; |
0b32f4bb | 4869 | |
86679820 | 4870 | pagevec_init(&pvec); |
0b32f4bb JB |
4871 | if (wbc->range_cyclic) { |
4872 | index = mapping->writeback_index; /* Start from prev offset */ | |
4873 | end = -1; | |
556755a8 JB |
4874 | /* |
4875 | * Start from the beginning does not need to cycle over the | |
4876 | * range, mark it as scanned. | |
4877 | */ | |
4878 | scanned = (index == 0); | |
0b32f4bb | 4879 | } else { |
09cbfeaf KS |
4880 | index = wbc->range_start >> PAGE_SHIFT; |
4881 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4882 | scanned = 1; |
4883 | } | |
4884 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4885 | tag = PAGECACHE_TAG_TOWRITE; | |
4886 | else | |
4887 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 4888 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
4889 | retry: |
4890 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4891 | tag_pages_for_writeback(mapping, index, end); | |
4892 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4893 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4894 | tag))) { |
0b32f4bb JB |
4895 | unsigned i; |
4896 | ||
0b32f4bb JB |
4897 | for (i = 0; i < nr_pages; i++) { |
4898 | struct page *page = pvec.pages[i]; | |
4899 | ||
f91e0d0c QW |
4900 | ret = submit_eb_page(page, wbc, &epd, &eb_context); |
4901 | if (ret == 0) | |
0b32f4bb | 4902 | continue; |
f91e0d0c | 4903 | if (ret < 0) { |
0b32f4bb | 4904 | done = 1; |
0b32f4bb JB |
4905 | break; |
4906 | } | |
0b32f4bb JB |
4907 | |
4908 | /* | |
4909 | * the filesystem may choose to bump up nr_to_write. | |
4910 | * We have to make sure to honor the new nr_to_write | |
4911 | * at any time | |
4912 | */ | |
4913 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4914 | } | |
4915 | pagevec_release(&pvec); | |
4916 | cond_resched(); | |
4917 | } | |
4918 | if (!scanned && !done) { | |
4919 | /* | |
4920 | * We hit the last page and there is more work to be done: wrap | |
4921 | * back to the start of the file | |
4922 | */ | |
4923 | scanned = 1; | |
4924 | index = 0; | |
4925 | goto retry; | |
4926 | } | |
2b952eea QW |
4927 | if (ret < 0) { |
4928 | end_write_bio(&epd, ret); | |
0bc09ca1 | 4929 | goto out; |
2b952eea | 4930 | } |
b3ff8f1d QW |
4931 | /* |
4932 | * If something went wrong, don't allow any metadata write bio to be | |
4933 | * submitted. | |
4934 | * | |
4935 | * This would prevent use-after-free if we had dirty pages not | |
4936 | * cleaned up, which can still happen by fuzzed images. | |
4937 | * | |
4938 | * - Bad extent tree | |
4939 | * Allowing existing tree block to be allocated for other trees. | |
4940 | * | |
4941 | * - Log tree operations | |
4942 | * Exiting tree blocks get allocated to log tree, bumps its | |
4943 | * generation, then get cleaned in tree re-balance. | |
4944 | * Such tree block will not be written back, since it's clean, | |
4945 | * thus no WRITTEN flag set. | |
4946 | * And after log writes back, this tree block is not traced by | |
4947 | * any dirty extent_io_tree. | |
4948 | * | |
4949 | * - Offending tree block gets re-dirtied from its original owner | |
4950 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4951 | * reused without COWing. This tree block will not be traced | |
4952 | * by btrfs_transaction::dirty_pages. | |
4953 | * | |
4954 | * Now such dirty tree block will not be cleaned by any dirty | |
4955 | * extent io tree. Thus we don't want to submit such wild eb | |
4956 | * if the fs already has error. | |
4957 | */ | |
84961539 | 4958 | if (!BTRFS_FS_ERROR(fs_info)) { |
c9583ada | 4959 | flush_write_bio(&epd); |
b3ff8f1d | 4960 | } else { |
fbabd4a3 | 4961 | ret = -EROFS; |
b3ff8f1d QW |
4962 | end_write_bio(&epd, ret); |
4963 | } | |
0bc09ca1 NA |
4964 | out: |
4965 | btrfs_zoned_meta_io_unlock(fs_info); | |
c9583ada QW |
4966 | /* |
4967 | * We can get ret > 0 from submit_extent_page() indicating how many ebs | |
4968 | * were submitted. Reset it to 0 to avoid false alerts for the caller. | |
4969 | */ | |
4970 | if (ret > 0) | |
4971 | ret = 0; | |
0b32f4bb JB |
4972 | return ret; |
4973 | } | |
4974 | ||
d1310b2e | 4975 | /** |
3bed2da1 NB |
4976 | * Walk the list of dirty pages of the given address space and write all of them. |
4977 | * | |
d1310b2e | 4978 | * @mapping: address space structure to write |
3bed2da1 NB |
4979 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
4980 | * @epd: holds context for the write, namely the bio | |
d1310b2e CM |
4981 | * |
4982 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4983 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4984 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4985 | * and msync() need to guarantee that all the data which was dirty at the time | |
4986 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4987 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4988 | * existing IO to complete. | |
4989 | */ | |
4242b64a | 4990 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4991 | struct writeback_control *wbc, |
aab6e9ed | 4992 | struct extent_page_data *epd) |
d1310b2e | 4993 | { |
7fd1a3f7 | 4994 | struct inode *inode = mapping->host; |
d1310b2e CM |
4995 | int ret = 0; |
4996 | int done = 0; | |
f85d7d6c | 4997 | int nr_to_write_done = 0; |
d1310b2e CM |
4998 | struct pagevec pvec; |
4999 | int nr_pages; | |
5000 | pgoff_t index; | |
5001 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
5002 | pgoff_t done_index; |
5003 | int range_whole = 0; | |
d1310b2e | 5004 | int scanned = 0; |
10bbd235 | 5005 | xa_mark_t tag; |
d1310b2e | 5006 | |
7fd1a3f7 JB |
5007 | /* |
5008 | * We have to hold onto the inode so that ordered extents can do their | |
5009 | * work when the IO finishes. The alternative to this is failing to add | |
5010 | * an ordered extent if the igrab() fails there and that is a huge pain | |
5011 | * to deal with, so instead just hold onto the inode throughout the | |
5012 | * writepages operation. If it fails here we are freeing up the inode | |
5013 | * anyway and we'd rather not waste our time writing out stuff that is | |
5014 | * going to be truncated anyway. | |
5015 | */ | |
5016 | if (!igrab(inode)) | |
5017 | return 0; | |
5018 | ||
86679820 | 5019 | pagevec_init(&pvec); |
d1310b2e CM |
5020 | if (wbc->range_cyclic) { |
5021 | index = mapping->writeback_index; /* Start from prev offset */ | |
5022 | end = -1; | |
556755a8 JB |
5023 | /* |
5024 | * Start from the beginning does not need to cycle over the | |
5025 | * range, mark it as scanned. | |
5026 | */ | |
5027 | scanned = (index == 0); | |
d1310b2e | 5028 | } else { |
09cbfeaf KS |
5029 | index = wbc->range_start >> PAGE_SHIFT; |
5030 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
5031 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
5032 | range_whole = 1; | |
d1310b2e CM |
5033 | scanned = 1; |
5034 | } | |
3cd24c69 EL |
5035 | |
5036 | /* | |
5037 | * We do the tagged writepage as long as the snapshot flush bit is set | |
5038 | * and we are the first one who do the filemap_flush() on this inode. | |
5039 | * | |
5040 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
5041 | * not race in and drop the bit. | |
5042 | */ | |
5043 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
5044 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
5045 | &BTRFS_I(inode)->runtime_flags)) | |
5046 | wbc->tagged_writepages = 1; | |
5047 | ||
5048 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
5049 | tag = PAGECACHE_TAG_TOWRITE; |
5050 | else | |
5051 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 5052 | retry: |
3cd24c69 | 5053 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 5054 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 5055 | done_index = index; |
f85d7d6c | 5056 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
5057 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
5058 | &index, end, tag))) { | |
d1310b2e CM |
5059 | unsigned i; |
5060 | ||
d1310b2e CM |
5061 | for (i = 0; i < nr_pages; i++) { |
5062 | struct page *page = pvec.pages[i]; | |
5063 | ||
f7bddf1e | 5064 | done_index = page->index + 1; |
d1310b2e | 5065 | /* |
b93b0163 MW |
5066 | * At this point we hold neither the i_pages lock nor |
5067 | * the page lock: the page may be truncated or | |
5068 | * invalidated (changing page->mapping to NULL), | |
5069 | * or even swizzled back from swapper_space to | |
5070 | * tmpfs file mapping | |
d1310b2e | 5071 | */ |
c8f2f24b | 5072 | if (!trylock_page(page)) { |
c9583ada | 5073 | flush_write_bio(epd); |
c8f2f24b | 5074 | lock_page(page); |
01d658f2 | 5075 | } |
d1310b2e CM |
5076 | |
5077 | if (unlikely(page->mapping != mapping)) { | |
5078 | unlock_page(page); | |
5079 | continue; | |
5080 | } | |
5081 | ||
d2c3f4f6 | 5082 | if (wbc->sync_mode != WB_SYNC_NONE) { |
c9583ada QW |
5083 | if (PageWriteback(page)) |
5084 | flush_write_bio(epd); | |
d1310b2e | 5085 | wait_on_page_writeback(page); |
d2c3f4f6 | 5086 | } |
d1310b2e CM |
5087 | |
5088 | if (PageWriteback(page) || | |
5089 | !clear_page_dirty_for_io(page)) { | |
5090 | unlock_page(page); | |
5091 | continue; | |
5092 | } | |
5093 | ||
aab6e9ed | 5094 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 5095 | if (ret < 0) { |
a9132667 LB |
5096 | done = 1; |
5097 | break; | |
5098 | } | |
f85d7d6c CM |
5099 | |
5100 | /* | |
5101 | * the filesystem may choose to bump up nr_to_write. | |
5102 | * We have to make sure to honor the new nr_to_write | |
5103 | * at any time | |
5104 | */ | |
5105 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
5106 | } |
5107 | pagevec_release(&pvec); | |
5108 | cond_resched(); | |
5109 | } | |
894b36e3 | 5110 | if (!scanned && !done) { |
d1310b2e CM |
5111 | /* |
5112 | * We hit the last page and there is more work to be done: wrap | |
5113 | * back to the start of the file | |
5114 | */ | |
5115 | scanned = 1; | |
5116 | index = 0; | |
42ffb0bf JB |
5117 | |
5118 | /* | |
5119 | * If we're looping we could run into a page that is locked by a | |
5120 | * writer and that writer could be waiting on writeback for a | |
5121 | * page in our current bio, and thus deadlock, so flush the | |
5122 | * write bio here. | |
5123 | */ | |
c9583ada QW |
5124 | flush_write_bio(epd); |
5125 | goto retry; | |
d1310b2e | 5126 | } |
a9132667 LB |
5127 | |
5128 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
5129 | mapping->writeback_index = done_index; | |
5130 | ||
7fd1a3f7 | 5131 | btrfs_add_delayed_iput(inode); |
894b36e3 | 5132 | return ret; |
d1310b2e | 5133 | } |
d1310b2e | 5134 | |
0a9b0e53 | 5135 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
5136 | { |
5137 | int ret; | |
d1310b2e | 5138 | struct extent_page_data epd = { |
390ed29b | 5139 | .bio_ctrl = { 0 }, |
771ed689 | 5140 | .extent_locked = 0, |
ffbd517d | 5141 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 5142 | }; |
d1310b2e | 5143 | |
d1310b2e | 5144 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
5145 | ASSERT(ret <= 0); |
5146 | if (ret < 0) { | |
5147 | end_write_bio(&epd, ret); | |
5148 | return ret; | |
5149 | } | |
d1310b2e | 5150 | |
c9583ada | 5151 | flush_write_bio(&epd); |
d1310b2e CM |
5152 | return ret; |
5153 | } | |
d1310b2e | 5154 | |
2bd0fc93 QW |
5155 | /* |
5156 | * Submit the pages in the range to bio for call sites which delalloc range has | |
5157 | * already been ran (aka, ordered extent inserted) and all pages are still | |
5158 | * locked. | |
5159 | */ | |
5160 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end) | |
771ed689 | 5161 | { |
2bd0fc93 QW |
5162 | bool found_error = false; |
5163 | int first_error = 0; | |
771ed689 CM |
5164 | int ret = 0; |
5165 | struct address_space *mapping = inode->i_mapping; | |
5166 | struct page *page; | |
2bd0fc93 | 5167 | u64 cur = start; |
66448b9d QW |
5168 | unsigned long nr_pages; |
5169 | const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize; | |
771ed689 | 5170 | struct extent_page_data epd = { |
390ed29b | 5171 | .bio_ctrl = { 0 }, |
771ed689 | 5172 | .extent_locked = 1, |
2bd0fc93 | 5173 | .sync_io = 1, |
771ed689 CM |
5174 | }; |
5175 | struct writeback_control wbc_writepages = { | |
2bd0fc93 | 5176 | .sync_mode = WB_SYNC_ALL, |
771ed689 CM |
5177 | .range_start = start, |
5178 | .range_end = end + 1, | |
ec39f769 CM |
5179 | /* We're called from an async helper function */ |
5180 | .punt_to_cgroup = 1, | |
5181 | .no_cgroup_owner = 1, | |
771ed689 CM |
5182 | }; |
5183 | ||
66448b9d QW |
5184 | ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize)); |
5185 | nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >> | |
5186 | PAGE_SHIFT; | |
5187 | wbc_writepages.nr_to_write = nr_pages * 2; | |
5188 | ||
dbb70bec | 5189 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
2bd0fc93 | 5190 | while (cur <= end) { |
66448b9d QW |
5191 | u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end); |
5192 | ||
2bd0fc93 QW |
5193 | page = find_get_page(mapping, cur >> PAGE_SHIFT); |
5194 | /* | |
5195 | * All pages in the range are locked since | |
5196 | * btrfs_run_delalloc_range(), thus there is no way to clear | |
5197 | * the page dirty flag. | |
5198 | */ | |
66448b9d | 5199 | ASSERT(PageLocked(page)); |
2bd0fc93 QW |
5200 | ASSERT(PageDirty(page)); |
5201 | clear_page_dirty_for_io(page); | |
5202 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
5203 | ASSERT(ret <= 0); | |
5204 | if (ret < 0) { | |
5205 | found_error = true; | |
5206 | first_error = ret; | |
771ed689 | 5207 | } |
09cbfeaf | 5208 | put_page(page); |
66448b9d | 5209 | cur = cur_end + 1; |
771ed689 CM |
5210 | } |
5211 | ||
2bd0fc93 | 5212 | if (!found_error) |
c9583ada | 5213 | flush_write_bio(&epd); |
dbb70bec | 5214 | else |
02c6db4f | 5215 | end_write_bio(&epd, ret); |
dbb70bec CM |
5216 | |
5217 | wbc_detach_inode(&wbc_writepages); | |
2bd0fc93 QW |
5218 | if (found_error) |
5219 | return first_error; | |
771ed689 CM |
5220 | return ret; |
5221 | } | |
d1310b2e | 5222 | |
8ae225a8 | 5223 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
5224 | struct writeback_control *wbc) |
5225 | { | |
35156d85 | 5226 | struct inode *inode = mapping->host; |
d1310b2e CM |
5227 | int ret = 0; |
5228 | struct extent_page_data epd = { | |
390ed29b | 5229 | .bio_ctrl = { 0 }, |
771ed689 | 5230 | .extent_locked = 0, |
ffbd517d | 5231 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
5232 | }; |
5233 | ||
35156d85 JT |
5234 | /* |
5235 | * Allow only a single thread to do the reloc work in zoned mode to | |
5236 | * protect the write pointer updates. | |
5237 | */ | |
869f4cdc | 5238 | btrfs_zoned_data_reloc_lock(BTRFS_I(inode)); |
935db853 | 5239 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
5240 | ASSERT(ret <= 0); |
5241 | if (ret < 0) { | |
19ab78ca | 5242 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
a2a72fbd QW |
5243 | end_write_bio(&epd, ret); |
5244 | return ret; | |
5245 | } | |
c9583ada | 5246 | flush_write_bio(&epd); |
19ab78ca | 5247 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
d1310b2e CM |
5248 | return ret; |
5249 | } | |
d1310b2e | 5250 | |
ba206a02 | 5251 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 5252 | { |
390ed29b | 5253 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
67c9684f | 5254 | struct page *pagepool[16]; |
125bac01 | 5255 | struct extent_map *em_cached = NULL; |
808f80b4 | 5256 | u64 prev_em_start = (u64)-1; |
ba206a02 | 5257 | int nr; |
d1310b2e | 5258 | |
ba206a02 | 5259 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
5260 | u64 contig_start = readahead_pos(rac); |
5261 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 5262 | |
ba206a02 | 5263 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 5264 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 5265 | } |
67c9684f | 5266 | |
125bac01 MX |
5267 | if (em_cached) |
5268 | free_extent_map(em_cached); | |
5269 | ||
c9583ada | 5270 | if (bio_ctrl.bio) |
0f07003b | 5271 | submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.compress_type); |
d1310b2e | 5272 | } |
d1310b2e CM |
5273 | |
5274 | /* | |
895586eb MWO |
5275 | * basic invalidate_folio code, this waits on any locked or writeback |
5276 | * ranges corresponding to the folio, and then deletes any extent state | |
d1310b2e CM |
5277 | * records from the tree |
5278 | */ | |
895586eb MWO |
5279 | int extent_invalidate_folio(struct extent_io_tree *tree, |
5280 | struct folio *folio, size_t offset) | |
d1310b2e | 5281 | { |
2ac55d41 | 5282 | struct extent_state *cached_state = NULL; |
895586eb MWO |
5283 | u64 start = folio_pos(folio); |
5284 | u64 end = start + folio_size(folio) - 1; | |
5285 | size_t blocksize = folio->mapping->host->i_sb->s_blocksize; | |
d1310b2e | 5286 | |
829ddec9 QW |
5287 | /* This function is only called for the btree inode */ |
5288 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
5289 | ||
fda2832f | 5290 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
5291 | if (start > end) |
5292 | return 0; | |
5293 | ||
ff13db41 | 5294 | lock_extent_bits(tree, start, end, &cached_state); |
895586eb | 5295 | folio_wait_writeback(folio); |
829ddec9 QW |
5296 | |
5297 | /* | |
5298 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
5299 | * so here we only need to unlock the extent range to free any | |
5300 | * existing extent state. | |
5301 | */ | |
5302 | unlock_extent_cached(tree, start, end, &cached_state); | |
d1310b2e CM |
5303 | return 0; |
5304 | } | |
d1310b2e | 5305 | |
7b13b7b1 | 5306 | /* |
f913cff3 | 5307 | * a helper for release_folio, this tests for areas of the page that |
7b13b7b1 CM |
5308 | * are locked or under IO and drops the related state bits if it is safe |
5309 | * to drop the page. | |
5310 | */ | |
29c68b2d | 5311 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 5312 | struct page *page, gfp_t mask) |
7b13b7b1 | 5313 | { |
4eee4fa4 | 5314 | u64 start = page_offset(page); |
09cbfeaf | 5315 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
5316 | int ret = 1; |
5317 | ||
8882679e | 5318 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 5319 | ret = 0; |
8882679e | 5320 | } else { |
11ef160f | 5321 | /* |
2766ff61 FM |
5322 | * At this point we can safely clear everything except the |
5323 | * locked bit, the nodatasum bit and the delalloc new bit. | |
5324 | * The delalloc new bit will be cleared by ordered extent | |
5325 | * completion. | |
11ef160f | 5326 | */ |
66b0c887 | 5327 | ret = __clear_extent_bit(tree, start, end, |
2766ff61 FM |
5328 | ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW), |
5329 | 0, 0, NULL, mask, NULL); | |
e3f24cc5 CM |
5330 | |
5331 | /* if clear_extent_bit failed for enomem reasons, | |
5332 | * we can't allow the release to continue. | |
5333 | */ | |
5334 | if (ret < 0) | |
5335 | ret = 0; | |
5336 | else | |
5337 | ret = 1; | |
7b13b7b1 CM |
5338 | } |
5339 | return ret; | |
5340 | } | |
7b13b7b1 | 5341 | |
d1310b2e | 5342 | /* |
f913cff3 | 5343 | * a helper for release_folio. As long as there are no locked extents |
d1310b2e CM |
5344 | * in the range corresponding to the page, both state records and extent |
5345 | * map records are removed | |
5346 | */ | |
477a30ba | 5347 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
5348 | { |
5349 | struct extent_map *em; | |
4eee4fa4 | 5350 | u64 start = page_offset(page); |
09cbfeaf | 5351 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
5352 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
5353 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
5354 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 5355 | |
d0164adc | 5356 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 5357 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 5358 | u64 len; |
70dec807 | 5359 | while (start <= end) { |
fbc2bd7e FM |
5360 | struct btrfs_fs_info *fs_info; |
5361 | u64 cur_gen; | |
5362 | ||
39b5637f | 5363 | len = end - start + 1; |
890871be | 5364 | write_lock(&map->lock); |
39b5637f | 5365 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 5366 | if (!em) { |
890871be | 5367 | write_unlock(&map->lock); |
70dec807 CM |
5368 | break; |
5369 | } | |
7f3c74fb CM |
5370 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
5371 | em->start != start) { | |
890871be | 5372 | write_unlock(&map->lock); |
70dec807 CM |
5373 | free_extent_map(em); |
5374 | break; | |
5375 | } | |
3d6448e6 FM |
5376 | if (test_range_bit(tree, em->start, |
5377 | extent_map_end(em) - 1, | |
5378 | EXTENT_LOCKED, 0, NULL)) | |
5379 | goto next; | |
5380 | /* | |
5381 | * If it's not in the list of modified extents, used | |
5382 | * by a fast fsync, we can remove it. If it's being | |
5383 | * logged we can safely remove it since fsync took an | |
5384 | * extra reference on the em. | |
5385 | */ | |
5386 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
5387 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
5388 | goto remove_em; | |
5389 | /* | |
5390 | * If it's in the list of modified extents, remove it | |
5391 | * only if its generation is older then the current one, | |
5392 | * in which case we don't need it for a fast fsync. | |
5393 | * Otherwise don't remove it, we could be racing with an | |
5394 | * ongoing fast fsync that could miss the new extent. | |
5395 | */ | |
5396 | fs_info = btrfs_inode->root->fs_info; | |
5397 | spin_lock(&fs_info->trans_lock); | |
5398 | cur_gen = fs_info->generation; | |
5399 | spin_unlock(&fs_info->trans_lock); | |
5400 | if (em->generation >= cur_gen) | |
5401 | goto next; | |
5402 | remove_em: | |
5e548b32 FM |
5403 | /* |
5404 | * We only remove extent maps that are not in the list of | |
5405 | * modified extents or that are in the list but with a | |
5406 | * generation lower then the current generation, so there | |
5407 | * is no need to set the full fsync flag on the inode (it | |
5408 | * hurts the fsync performance for workloads with a data | |
5409 | * size that exceeds or is close to the system's memory). | |
5410 | */ | |
fbc2bd7e FM |
5411 | remove_extent_mapping(map, em); |
5412 | /* once for the rb tree */ | |
5413 | free_extent_map(em); | |
3d6448e6 | 5414 | next: |
70dec807 | 5415 | start = extent_map_end(em); |
890871be | 5416 | write_unlock(&map->lock); |
70dec807 CM |
5417 | |
5418 | /* once for us */ | |
d1310b2e | 5419 | free_extent_map(em); |
9f47eb54 PM |
5420 | |
5421 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 5422 | } |
d1310b2e | 5423 | } |
29c68b2d | 5424 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 5425 | } |
d1310b2e | 5426 | |
ec29ed5b CM |
5427 | /* |
5428 | * helper function for fiemap, which doesn't want to see any holes. | |
5429 | * This maps until we find something past 'last' | |
5430 | */ | |
f1bbde8d | 5431 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 5432 | u64 offset, u64 last) |
ec29ed5b | 5433 | { |
f1bbde8d | 5434 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
5435 | struct extent_map *em; |
5436 | u64 len; | |
5437 | ||
5438 | if (offset >= last) | |
5439 | return NULL; | |
5440 | ||
67871254 | 5441 | while (1) { |
ec29ed5b CM |
5442 | len = last - offset; |
5443 | if (len == 0) | |
5444 | break; | |
fda2832f | 5445 | len = ALIGN(len, sectorsize); |
f1bbde8d | 5446 | em = btrfs_get_extent_fiemap(inode, offset, len); |
6b5b7a41 | 5447 | if (IS_ERR(em)) |
ec29ed5b CM |
5448 | return em; |
5449 | ||
5450 | /* if this isn't a hole return it */ | |
4a2d25cd | 5451 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 5452 | return em; |
ec29ed5b CM |
5453 | |
5454 | /* this is a hole, advance to the next extent */ | |
5455 | offset = extent_map_end(em); | |
5456 | free_extent_map(em); | |
5457 | if (offset >= last) | |
5458 | break; | |
5459 | } | |
5460 | return NULL; | |
5461 | } | |
5462 | ||
4751832d QW |
5463 | /* |
5464 | * To cache previous fiemap extent | |
5465 | * | |
5466 | * Will be used for merging fiemap extent | |
5467 | */ | |
5468 | struct fiemap_cache { | |
5469 | u64 offset; | |
5470 | u64 phys; | |
5471 | u64 len; | |
5472 | u32 flags; | |
5473 | bool cached; | |
5474 | }; | |
5475 | ||
5476 | /* | |
5477 | * Helper to submit fiemap extent. | |
5478 | * | |
5479 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
5480 | * @len and @flags with cached one. | |
5481 | * And only when we fails to merge, cached one will be submitted as | |
5482 | * fiemap extent. | |
5483 | * | |
5484 | * Return value is the same as fiemap_fill_next_extent(). | |
5485 | */ | |
5486 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
5487 | struct fiemap_cache *cache, | |
5488 | u64 offset, u64 phys, u64 len, u32 flags) | |
5489 | { | |
5490 | int ret = 0; | |
5491 | ||
5492 | if (!cache->cached) | |
5493 | goto assign; | |
5494 | ||
5495 | /* | |
5496 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 5497 | * fiemap extent won't overlap with cached one. |
4751832d QW |
5498 | * Not recoverable. |
5499 | * | |
5500 | * NOTE: Physical address can overlap, due to compression | |
5501 | */ | |
5502 | if (cache->offset + cache->len > offset) { | |
5503 | WARN_ON(1); | |
5504 | return -EINVAL; | |
5505 | } | |
5506 | ||
5507 | /* | |
5508 | * Only merges fiemap extents if | |
5509 | * 1) Their logical addresses are continuous | |
5510 | * | |
5511 | * 2) Their physical addresses are continuous | |
5512 | * So truly compressed (physical size smaller than logical size) | |
5513 | * extents won't get merged with each other | |
5514 | * | |
5515 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
5516 | * So regular extent won't get merged with prealloc extent | |
5517 | */ | |
5518 | if (cache->offset + cache->len == offset && | |
5519 | cache->phys + cache->len == phys && | |
5520 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
5521 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
5522 | cache->len += len; | |
5523 | cache->flags |= flags; | |
5524 | goto try_submit_last; | |
5525 | } | |
5526 | ||
5527 | /* Not mergeable, need to submit cached one */ | |
5528 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
5529 | cache->len, cache->flags); | |
5530 | cache->cached = false; | |
5531 | if (ret) | |
5532 | return ret; | |
5533 | assign: | |
5534 | cache->cached = true; | |
5535 | cache->offset = offset; | |
5536 | cache->phys = phys; | |
5537 | cache->len = len; | |
5538 | cache->flags = flags; | |
5539 | try_submit_last: | |
5540 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
5541 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
5542 | cache->phys, cache->len, cache->flags); | |
5543 | cache->cached = false; | |
5544 | } | |
5545 | return ret; | |
5546 | } | |
5547 | ||
5548 | /* | |
848c23b7 | 5549 | * Emit last fiemap cache |
4751832d | 5550 | * |
848c23b7 QW |
5551 | * The last fiemap cache may still be cached in the following case: |
5552 | * 0 4k 8k | |
5553 | * |<- Fiemap range ->| | |
5554 | * |<------------ First extent ----------->| | |
5555 | * | |
5556 | * In this case, the first extent range will be cached but not emitted. | |
5557 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 5558 | */ |
5c5aff98 | 5559 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 5560 | struct fiemap_cache *cache) |
4751832d QW |
5561 | { |
5562 | int ret; | |
5563 | ||
5564 | if (!cache->cached) | |
5565 | return 0; | |
5566 | ||
4751832d QW |
5567 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
5568 | cache->len, cache->flags); | |
5569 | cache->cached = false; | |
5570 | if (ret > 0) | |
5571 | ret = 0; | |
5572 | return ret; | |
5573 | } | |
5574 | ||
facee0a0 | 5575 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 5576 | u64 start, u64 len) |
1506fcc8 | 5577 | { |
975f84fe | 5578 | int ret = 0; |
15c7745c | 5579 | u64 off; |
1506fcc8 YS |
5580 | u64 max = start + len; |
5581 | u32 flags = 0; | |
975f84fe JB |
5582 | u32 found_type; |
5583 | u64 last; | |
ec29ed5b | 5584 | u64 last_for_get_extent = 0; |
1506fcc8 | 5585 | u64 disko = 0; |
facee0a0 | 5586 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 5587 | struct btrfs_key found_key; |
1506fcc8 | 5588 | struct extent_map *em = NULL; |
2ac55d41 | 5589 | struct extent_state *cached_state = NULL; |
975f84fe | 5590 | struct btrfs_path *path; |
facee0a0 | 5591 | struct btrfs_root *root = inode->root; |
4751832d | 5592 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
5593 | struct ulist *roots; |
5594 | struct ulist *tmp_ulist; | |
1506fcc8 | 5595 | int end = 0; |
ec29ed5b CM |
5596 | u64 em_start = 0; |
5597 | u64 em_len = 0; | |
5598 | u64 em_end = 0; | |
1506fcc8 YS |
5599 | |
5600 | if (len == 0) | |
5601 | return -EINVAL; | |
5602 | ||
975f84fe JB |
5603 | path = btrfs_alloc_path(); |
5604 | if (!path) | |
5605 | return -ENOMEM; | |
975f84fe | 5606 | |
5911c8fe DS |
5607 | roots = ulist_alloc(GFP_KERNEL); |
5608 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
5609 | if (!roots || !tmp_ulist) { | |
5610 | ret = -ENOMEM; | |
5611 | goto out_free_ulist; | |
5612 | } | |
5613 | ||
15c7745c BB |
5614 | /* |
5615 | * We can't initialize that to 'start' as this could miss extents due | |
5616 | * to extent item merging | |
5617 | */ | |
5618 | off = 0; | |
facee0a0 NB |
5619 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
5620 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 5621 | |
ec29ed5b CM |
5622 | /* |
5623 | * lookup the last file extent. We're not using i_size here | |
5624 | * because there might be preallocation past i_size | |
5625 | */ | |
facee0a0 NB |
5626 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
5627 | 0); | |
975f84fe | 5628 | if (ret < 0) { |
5911c8fe | 5629 | goto out_free_ulist; |
2d324f59 LB |
5630 | } else { |
5631 | WARN_ON(!ret); | |
5632 | if (ret == 1) | |
5633 | ret = 0; | |
975f84fe | 5634 | } |
2d324f59 | 5635 | |
975f84fe | 5636 | path->slots[0]--; |
975f84fe | 5637 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 5638 | found_type = found_key.type; |
975f84fe | 5639 | |
ec29ed5b | 5640 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 5641 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 5642 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
5643 | /* have to trust i_size as the end */ |
5644 | last = (u64)-1; | |
5645 | last_for_get_extent = isize; | |
5646 | } else { | |
5647 | /* | |
5648 | * remember the start of the last extent. There are a | |
5649 | * bunch of different factors that go into the length of the | |
5650 | * extent, so its much less complex to remember where it started | |
5651 | */ | |
5652 | last = found_key.offset; | |
5653 | last_for_get_extent = last + 1; | |
975f84fe | 5654 | } |
fe09e16c | 5655 | btrfs_release_path(path); |
975f84fe | 5656 | |
ec29ed5b CM |
5657 | /* |
5658 | * we might have some extents allocated but more delalloc past those | |
5659 | * extents. so, we trust isize unless the start of the last extent is | |
5660 | * beyond isize | |
5661 | */ | |
5662 | if (last < isize) { | |
5663 | last = (u64)-1; | |
5664 | last_for_get_extent = isize; | |
5665 | } | |
5666 | ||
facee0a0 | 5667 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 5668 | &cached_state); |
ec29ed5b | 5669 | |
facee0a0 | 5670 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
5671 | if (!em) |
5672 | goto out; | |
5673 | if (IS_ERR(em)) { | |
5674 | ret = PTR_ERR(em); | |
5675 | goto out; | |
5676 | } | |
975f84fe | 5677 | |
1506fcc8 | 5678 | while (!end) { |
b76bb701 | 5679 | u64 offset_in_extent = 0; |
ea8efc74 CM |
5680 | |
5681 | /* break if the extent we found is outside the range */ | |
5682 | if (em->start >= max || extent_map_end(em) < off) | |
5683 | break; | |
5684 | ||
5685 | /* | |
5686 | * get_extent may return an extent that starts before our | |
5687 | * requested range. We have to make sure the ranges | |
5688 | * we return to fiemap always move forward and don't | |
5689 | * overlap, so adjust the offsets here | |
5690 | */ | |
5691 | em_start = max(em->start, off); | |
1506fcc8 | 5692 | |
ea8efc74 CM |
5693 | /* |
5694 | * record the offset from the start of the extent | |
b76bb701 JB |
5695 | * for adjusting the disk offset below. Only do this if the |
5696 | * extent isn't compressed since our in ram offset may be past | |
5697 | * what we have actually allocated on disk. | |
ea8efc74 | 5698 | */ |
b76bb701 JB |
5699 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
5700 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 5701 | em_end = extent_map_end(em); |
ea8efc74 | 5702 | em_len = em_end - em_start; |
1506fcc8 | 5703 | flags = 0; |
f0986318 FM |
5704 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
5705 | disko = em->block_start + offset_in_extent; | |
5706 | else | |
5707 | disko = 0; | |
1506fcc8 | 5708 | |
ea8efc74 CM |
5709 | /* |
5710 | * bump off for our next call to get_extent | |
5711 | */ | |
5712 | off = extent_map_end(em); | |
5713 | if (off >= max) | |
5714 | end = 1; | |
5715 | ||
93dbfad7 | 5716 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
5717 | end = 1; |
5718 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 5719 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
5720 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
5721 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 5722 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
5723 | flags |= (FIEMAP_EXTENT_DELALLOC | |
5724 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
5725 | } else if (fieinfo->fi_extents_max) { |
5726 | u64 bytenr = em->block_start - | |
5727 | (em->start - em->orig_start); | |
fe09e16c | 5728 | |
fe09e16c LB |
5729 | /* |
5730 | * As btrfs supports shared space, this information | |
5731 | * can be exported to userspace tools via | |
dc046b10 JB |
5732 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
5733 | * then we're just getting a count and we can skip the | |
5734 | * lookup stuff. | |
fe09e16c | 5735 | */ |
facee0a0 | 5736 | ret = btrfs_check_shared(root, btrfs_ino(inode), |
5911c8fe | 5737 | bytenr, roots, tmp_ulist); |
dc046b10 | 5738 | if (ret < 0) |
fe09e16c | 5739 | goto out_free; |
dc046b10 | 5740 | if (ret) |
fe09e16c | 5741 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 5742 | ret = 0; |
1506fcc8 YS |
5743 | } |
5744 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
5745 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
5746 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
5747 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 5748 | |
1506fcc8 YS |
5749 | free_extent_map(em); |
5750 | em = NULL; | |
ec29ed5b CM |
5751 | if ((em_start >= last) || em_len == (u64)-1 || |
5752 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
5753 | flags |= FIEMAP_EXTENT_LAST; |
5754 | end = 1; | |
5755 | } | |
5756 | ||
ec29ed5b | 5757 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 5758 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
5759 | if (IS_ERR(em)) { |
5760 | ret = PTR_ERR(em); | |
5761 | goto out; | |
5762 | } | |
5763 | if (!em) { | |
975f84fe JB |
5764 | flags |= FIEMAP_EXTENT_LAST; |
5765 | end = 1; | |
5766 | } | |
4751832d QW |
5767 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
5768 | em_len, flags); | |
26e726af CS |
5769 | if (ret) { |
5770 | if (ret == 1) | |
5771 | ret = 0; | |
ec29ed5b | 5772 | goto out_free; |
26e726af | 5773 | } |
1506fcc8 YS |
5774 | } |
5775 | out_free: | |
4751832d | 5776 | if (!ret) |
5c5aff98 | 5777 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
5778 | free_extent_map(em); |
5779 | out: | |
facee0a0 | 5780 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 5781 | &cached_state); |
5911c8fe DS |
5782 | |
5783 | out_free_ulist: | |
e02d48ea | 5784 | btrfs_free_path(path); |
5911c8fe DS |
5785 | ulist_free(roots); |
5786 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
5787 | return ret; |
5788 | } | |
5789 | ||
727011e0 CM |
5790 | static void __free_extent_buffer(struct extent_buffer *eb) |
5791 | { | |
727011e0 CM |
5792 | kmem_cache_free(extent_buffer_cache, eb); |
5793 | } | |
5794 | ||
2b48966a | 5795 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
5796 | { |
5797 | return (atomic_read(&eb->io_pages) || | |
5798 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
5799 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5800 | } | |
5801 | ||
8ff8466d | 5802 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 5803 | { |
8ff8466d | 5804 | struct btrfs_subpage *subpage; |
db7f3436 | 5805 | |
8ff8466d | 5806 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 5807 | |
8ff8466d QW |
5808 | if (PagePrivate(page)) { |
5809 | subpage = (struct btrfs_subpage *)page->private; | |
5810 | if (atomic_read(&subpage->eb_refs)) | |
5811 | return true; | |
3d078efa QW |
5812 | /* |
5813 | * Even there is no eb refs here, we may still have | |
5814 | * end_page_read() call relying on page::private. | |
5815 | */ | |
5816 | if (atomic_read(&subpage->readers)) | |
5817 | return true; | |
8ff8466d QW |
5818 | } |
5819 | return false; | |
5820 | } | |
db7f3436 | 5821 | |
8ff8466d QW |
5822 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
5823 | { | |
5824 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5825 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
5826 | ||
5827 | /* | |
5828 | * For mapped eb, we're going to change the page private, which should | |
5829 | * be done under the private_lock. | |
5830 | */ | |
5831 | if (mapped) | |
5832 | spin_lock(&page->mapping->private_lock); | |
5833 | ||
5834 | if (!PagePrivate(page)) { | |
5d2361db | 5835 | if (mapped) |
8ff8466d QW |
5836 | spin_unlock(&page->mapping->private_lock); |
5837 | return; | |
5838 | } | |
5839 | ||
fbca46eb | 5840 | if (fs_info->nodesize >= PAGE_SIZE) { |
5d2361db FL |
5841 | /* |
5842 | * We do this since we'll remove the pages after we've | |
5843 | * removed the eb from the radix tree, so we could race | |
5844 | * and have this page now attached to the new eb. So | |
5845 | * only clear page_private if it's still connected to | |
5846 | * this eb. | |
5847 | */ | |
5848 | if (PagePrivate(page) && | |
5849 | page->private == (unsigned long)eb) { | |
5850 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5851 | BUG_ON(PageDirty(page)); | |
5852 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 5853 | /* |
5d2361db FL |
5854 | * We need to make sure we haven't be attached |
5855 | * to a new eb. | |
db7f3436 | 5856 | */ |
d1b89bc0 | 5857 | detach_page_private(page); |
db7f3436 | 5858 | } |
5d2361db FL |
5859 | if (mapped) |
5860 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
5861 | return; |
5862 | } | |
5863 | ||
5864 | /* | |
5865 | * For subpage, we can have dummy eb with page private. In this case, | |
5866 | * we can directly detach the private as such page is only attached to | |
5867 | * one dummy eb, no sharing. | |
5868 | */ | |
5869 | if (!mapped) { | |
5870 | btrfs_detach_subpage(fs_info, page); | |
5871 | return; | |
5872 | } | |
5873 | ||
5874 | btrfs_page_dec_eb_refs(fs_info, page); | |
5875 | ||
5876 | /* | |
5877 | * We can only detach the page private if there are no other ebs in the | |
3d078efa | 5878 | * page range and no unfinished IO. |
8ff8466d QW |
5879 | */ |
5880 | if (!page_range_has_eb(fs_info, page)) | |
5881 | btrfs_detach_subpage(fs_info, page); | |
5882 | ||
5883 | spin_unlock(&page->mapping->private_lock); | |
5884 | } | |
5885 | ||
5886 | /* Release all pages attached to the extent buffer */ | |
5887 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
5888 | { | |
5889 | int i; | |
5890 | int num_pages; | |
5891 | ||
5892 | ASSERT(!extent_buffer_under_io(eb)); | |
5893 | ||
5894 | num_pages = num_extent_pages(eb); | |
5895 | for (i = 0; i < num_pages; i++) { | |
5896 | struct page *page = eb->pages[i]; | |
5897 | ||
5898 | if (!page) | |
5899 | continue; | |
5900 | ||
5901 | detach_extent_buffer_page(eb, page); | |
5d2361db | 5902 | |
01327610 | 5903 | /* One for when we allocated the page */ |
09cbfeaf | 5904 | put_page(page); |
d64766fd | 5905 | } |
db7f3436 JB |
5906 | } |
5907 | ||
5908 | /* | |
5909 | * Helper for releasing the extent buffer. | |
5910 | */ | |
5911 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
5912 | { | |
55ac0139 | 5913 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 5914 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
5915 | __free_extent_buffer(eb); |
5916 | } | |
5917 | ||
f28491e0 JB |
5918 | static struct extent_buffer * |
5919 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 5920 | unsigned long len) |
d1310b2e CM |
5921 | { |
5922 | struct extent_buffer *eb = NULL; | |
5923 | ||
d1b5c567 | 5924 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
5925 | eb->start = start; |
5926 | eb->len = len; | |
f28491e0 | 5927 | eb->fs_info = fs_info; |
815a51c7 | 5928 | eb->bflags = 0; |
196d59ab | 5929 | init_rwsem(&eb->lock); |
b4ce94de | 5930 | |
3fd63727 JB |
5931 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5932 | &fs_info->allocated_ebs); | |
d3575156 | 5933 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 5934 | |
3083ee2e | 5935 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5936 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5937 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5938 | |
deb67895 | 5939 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
5940 | |
5941 | return eb; | |
5942 | } | |
5943 | ||
2b48966a | 5944 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 5945 | { |
cc5e31a4 | 5946 | int i; |
815a51c7 | 5947 | struct extent_buffer *new; |
cc5e31a4 | 5948 | int num_pages = num_extent_pages(src); |
dd137dd1 | 5949 | int ret; |
815a51c7 | 5950 | |
3f556f78 | 5951 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5952 | if (new == NULL) |
5953 | return NULL; | |
5954 | ||
62c053fb QW |
5955 | /* |
5956 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
5957 | * btrfs_release_extent_buffer() have different behavior for | |
5958 | * UNMAPPED subpage extent buffer. | |
5959 | */ | |
5960 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
5961 | ||
dd137dd1 STD |
5962 | memset(new->pages, 0, sizeof(*new->pages) * num_pages); |
5963 | ret = btrfs_alloc_page_array(num_pages, new->pages); | |
5964 | if (ret) { | |
5965 | btrfs_release_extent_buffer(new); | |
5966 | return NULL; | |
5967 | } | |
5968 | ||
815a51c7 | 5969 | for (i = 0; i < num_pages; i++) { |
760f991f | 5970 | int ret; |
dd137dd1 | 5971 | struct page *p = new->pages[i]; |
760f991f | 5972 | |
760f991f QW |
5973 | ret = attach_extent_buffer_page(new, p, NULL); |
5974 | if (ret < 0) { | |
760f991f QW |
5975 | btrfs_release_extent_buffer(new); |
5976 | return NULL; | |
5977 | } | |
815a51c7 | 5978 | WARN_ON(PageDirty(p)); |
fba1acf9 | 5979 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 5980 | } |
92d83e94 | 5981 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
5982 | |
5983 | return new; | |
5984 | } | |
5985 | ||
0f331229 OS |
5986 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5987 | u64 start, unsigned long len) | |
815a51c7 JS |
5988 | { |
5989 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5990 | int num_pages; |
5991 | int i; | |
dd137dd1 | 5992 | int ret; |
815a51c7 | 5993 | |
3f556f78 | 5994 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5995 | if (!eb) |
5996 | return NULL; | |
5997 | ||
65ad0104 | 5998 | num_pages = num_extent_pages(eb); |
dd137dd1 STD |
5999 | ret = btrfs_alloc_page_array(num_pages, eb->pages); |
6000 | if (ret) | |
6001 | goto err; | |
6002 | ||
815a51c7 | 6003 | for (i = 0; i < num_pages; i++) { |
dd137dd1 | 6004 | struct page *p = eb->pages[i]; |
09bc1f0f | 6005 | |
dd137dd1 | 6006 | ret = attach_extent_buffer_page(eb, p, NULL); |
09bc1f0f QW |
6007 | if (ret < 0) |
6008 | goto err; | |
815a51c7 | 6009 | } |
dd137dd1 | 6010 | |
815a51c7 JS |
6011 | set_extent_buffer_uptodate(eb); |
6012 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 6013 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
6014 | |
6015 | return eb; | |
6016 | err: | |
dd137dd1 STD |
6017 | for (i = 0; i < num_pages; i++) { |
6018 | if (eb->pages[i]) { | |
6019 | detach_extent_buffer_page(eb, eb->pages[i]); | |
6020 | __free_page(eb->pages[i]); | |
6021 | } | |
09bc1f0f | 6022 | } |
815a51c7 JS |
6023 | __free_extent_buffer(eb); |
6024 | return NULL; | |
6025 | } | |
6026 | ||
0f331229 | 6027 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 6028 | u64 start) |
0f331229 | 6029 | { |
da17066c | 6030 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
6031 | } |
6032 | ||
0b32f4bb JB |
6033 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
6034 | { | |
242e18c7 | 6035 | int refs; |
6bf9cd2e BB |
6036 | /* |
6037 | * The TREE_REF bit is first set when the extent_buffer is added | |
6038 | * to the radix tree. It is also reset, if unset, when a new reference | |
6039 | * is created by find_extent_buffer. | |
0b32f4bb | 6040 | * |
6bf9cd2e BB |
6041 | * It is only cleared in two cases: freeing the last non-tree |
6042 | * reference to the extent_buffer when its STALE bit is set or | |
f913cff3 | 6043 | * calling release_folio when the tree reference is the only reference. |
0b32f4bb | 6044 | * |
6bf9cd2e | 6045 | * In both cases, care is taken to ensure that the extent_buffer's |
f913cff3 | 6046 | * pages are not under io. However, release_folio can be concurrently |
6bf9cd2e BB |
6047 | * called with creating new references, which is prone to race |
6048 | * conditions between the calls to check_buffer_tree_ref in those | |
6049 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 6050 | * |
6bf9cd2e BB |
6051 | * The actual lifetime of the extent_buffer in the radix tree is |
6052 | * adequately protected by the refcount, but the TREE_REF bit and | |
6053 | * its corresponding reference are not. To protect against this | |
6054 | * class of races, we call check_buffer_tree_ref from the codepaths | |
6055 | * which trigger io after they set eb->io_pages. Note that once io is | |
6056 | * initiated, TREE_REF can no longer be cleared, so that is the | |
6057 | * moment at which any such race is best fixed. | |
0b32f4bb | 6058 | */ |
242e18c7 CM |
6059 | refs = atomic_read(&eb->refs); |
6060 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
6061 | return; | |
6062 | ||
594831c4 JB |
6063 | spin_lock(&eb->refs_lock); |
6064 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 6065 | atomic_inc(&eb->refs); |
594831c4 | 6066 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
6067 | } |
6068 | ||
2457aec6 MG |
6069 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
6070 | struct page *accessed) | |
5df4235e | 6071 | { |
cc5e31a4 | 6072 | int num_pages, i; |
5df4235e | 6073 | |
0b32f4bb JB |
6074 | check_buffer_tree_ref(eb); |
6075 | ||
65ad0104 | 6076 | num_pages = num_extent_pages(eb); |
5df4235e | 6077 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
6078 | struct page *p = eb->pages[i]; |
6079 | ||
2457aec6 MG |
6080 | if (p != accessed) |
6081 | mark_page_accessed(p); | |
5df4235e JB |
6082 | } |
6083 | } | |
6084 | ||
f28491e0 JB |
6085 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
6086 | u64 start) | |
452c75c3 CS |
6087 | { |
6088 | struct extent_buffer *eb; | |
6089 | ||
2f3186d8 QW |
6090 | eb = find_extent_buffer_nolock(fs_info, start); |
6091 | if (!eb) | |
6092 | return NULL; | |
6093 | /* | |
6094 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
6095 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
6096 | * another task running free_extent_buffer() might have seen that flag | |
6097 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
6098 | * writeback flags not set) and it's still in the tree (flag | |
6099 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
6100 | * decrementing the extent buffer's reference count twice. So here we | |
6101 | * could race and increment the eb's reference count, clear its stale | |
6102 | * flag, mark it as dirty and drop our reference before the other task | |
6103 | * finishes executing free_extent_buffer, which would later result in | |
6104 | * an attempt to free an extent buffer that is dirty. | |
6105 | */ | |
6106 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
6107 | spin_lock(&eb->refs_lock); | |
6108 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 6109 | } |
2f3186d8 QW |
6110 | mark_extent_buffer_accessed(eb, NULL); |
6111 | return eb; | |
452c75c3 CS |
6112 | } |
6113 | ||
faa2dbf0 JB |
6114 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
6115 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 6116 | u64 start) |
faa2dbf0 JB |
6117 | { |
6118 | struct extent_buffer *eb, *exists = NULL; | |
6119 | int ret; | |
6120 | ||
6121 | eb = find_extent_buffer(fs_info, start); | |
6122 | if (eb) | |
6123 | return eb; | |
da17066c | 6124 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 6125 | if (!eb) |
b6293c82 | 6126 | return ERR_PTR(-ENOMEM); |
faa2dbf0 | 6127 | eb->fs_info = fs_info; |
01cd3909 DS |
6128 | again: |
6129 | ret = radix_tree_preload(GFP_NOFS); | |
6130 | if (ret) { | |
6131 | exists = ERR_PTR(ret); | |
6132 | goto free_eb; | |
6133 | } | |
6134 | spin_lock(&fs_info->buffer_lock); | |
6135 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
6136 | start >> fs_info->sectorsize_bits, eb); | |
6137 | spin_unlock(&fs_info->buffer_lock); | |
6138 | radix_tree_preload_end(); | |
6139 | if (ret == -EEXIST) { | |
6140 | exists = find_extent_buffer(fs_info, start); | |
6141 | if (exists) | |
faa2dbf0 | 6142 | goto free_eb; |
01cd3909 DS |
6143 | else |
6144 | goto again; | |
6145 | } | |
faa2dbf0 JB |
6146 | check_buffer_tree_ref(eb); |
6147 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
6148 | ||
faa2dbf0 JB |
6149 | return eb; |
6150 | free_eb: | |
6151 | btrfs_release_extent_buffer(eb); | |
6152 | return exists; | |
6153 | } | |
6154 | #endif | |
6155 | ||
81982210 QW |
6156 | static struct extent_buffer *grab_extent_buffer( |
6157 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
6158 | { |
6159 | struct extent_buffer *exists; | |
6160 | ||
81982210 QW |
6161 | /* |
6162 | * For subpage case, we completely rely on radix tree to ensure we | |
6163 | * don't try to insert two ebs for the same bytenr. So here we always | |
6164 | * return NULL and just continue. | |
6165 | */ | |
fbca46eb | 6166 | if (fs_info->nodesize < PAGE_SIZE) |
81982210 QW |
6167 | return NULL; |
6168 | ||
c0f0a9e7 QW |
6169 | /* Page not yet attached to an extent buffer */ |
6170 | if (!PagePrivate(page)) | |
6171 | return NULL; | |
6172 | ||
6173 | /* | |
6174 | * We could have already allocated an eb for this page and attached one | |
6175 | * so lets see if we can get a ref on the existing eb, and if we can we | |
6176 | * know it's good and we can just return that one, else we know we can | |
6177 | * just overwrite page->private. | |
6178 | */ | |
6179 | exists = (struct extent_buffer *)page->private; | |
6180 | if (atomic_inc_not_zero(&exists->refs)) | |
6181 | return exists; | |
6182 | ||
6183 | WARN_ON(PageDirty(page)); | |
6184 | detach_page_private(page); | |
6185 | return NULL; | |
6186 | } | |
6187 | ||
fbca46eb QW |
6188 | static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start) |
6189 | { | |
6190 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { | |
6191 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
6192 | return -EINVAL; | |
6193 | } | |
6194 | ||
6195 | if (fs_info->nodesize < PAGE_SIZE && | |
6196 | offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) { | |
6197 | btrfs_err(fs_info, | |
6198 | "tree block crosses page boundary, start %llu nodesize %u", | |
6199 | start, fs_info->nodesize); | |
6200 | return -EINVAL; | |
6201 | } | |
6202 | if (fs_info->nodesize >= PAGE_SIZE && | |
1280d2d1 | 6203 | !PAGE_ALIGNED(start)) { |
fbca46eb QW |
6204 | btrfs_err(fs_info, |
6205 | "tree block is not page aligned, start %llu nodesize %u", | |
6206 | start, fs_info->nodesize); | |
6207 | return -EINVAL; | |
6208 | } | |
6209 | return 0; | |
6210 | } | |
6211 | ||
f28491e0 | 6212 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 6213 | u64 start, u64 owner_root, int level) |
d1310b2e | 6214 | { |
da17066c | 6215 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
6216 | int num_pages; |
6217 | int i; | |
09cbfeaf | 6218 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 6219 | struct extent_buffer *eb; |
6af118ce | 6220 | struct extent_buffer *exists = NULL; |
d1310b2e | 6221 | struct page *p; |
f28491e0 | 6222 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 6223 | int uptodate = 1; |
19fe0a8b | 6224 | int ret; |
d1310b2e | 6225 | |
fbca46eb | 6226 | if (check_eb_alignment(fs_info, start)) |
c871b0f2 | 6227 | return ERR_PTR(-EINVAL); |
c871b0f2 | 6228 | |
e9306ad4 QW |
6229 | #if BITS_PER_LONG == 32 |
6230 | if (start >= MAX_LFS_FILESIZE) { | |
6231 | btrfs_err_rl(fs_info, | |
6232 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
6233 | btrfs_err_32bit_limit(fs_info); | |
6234 | return ERR_PTR(-EOVERFLOW); | |
6235 | } | |
6236 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6237 | btrfs_warn_32bit_limit(fs_info); | |
6238 | #endif | |
6239 | ||
f28491e0 | 6240 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 6241 | if (eb) |
6af118ce | 6242 | return eb; |
6af118ce | 6243 | |
23d79d81 | 6244 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 6245 | if (!eb) |
c871b0f2 | 6246 | return ERR_PTR(-ENOMEM); |
e114c545 | 6247 | btrfs_set_buffer_lockdep_class(owner_root, eb, level); |
d1310b2e | 6248 | |
65ad0104 | 6249 | num_pages = num_extent_pages(eb); |
727011e0 | 6250 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
6251 | struct btrfs_subpage *prealloc = NULL; |
6252 | ||
d1b5c567 | 6253 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
6254 | if (!p) { |
6255 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 6256 | goto free_eb; |
c871b0f2 | 6257 | } |
4f2de97a | 6258 | |
760f991f QW |
6259 | /* |
6260 | * Preallocate page->private for subpage case, so that we won't | |
6261 | * allocate memory with private_lock hold. The memory will be | |
6262 | * freed by attach_extent_buffer_page() or freed manually if | |
6263 | * we exit earlier. | |
6264 | * | |
6265 | * Although we have ensured one subpage eb can only have one | |
6266 | * page, but it may change in the future for 16K page size | |
6267 | * support, so we still preallocate the memory in the loop. | |
6268 | */ | |
fbca46eb | 6269 | if (fs_info->nodesize < PAGE_SIZE) { |
651fb419 QW |
6270 | prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA); |
6271 | if (IS_ERR(prealloc)) { | |
6272 | ret = PTR_ERR(prealloc); | |
fdf250db QW |
6273 | unlock_page(p); |
6274 | put_page(p); | |
6275 | exists = ERR_PTR(ret); | |
6276 | goto free_eb; | |
6277 | } | |
760f991f QW |
6278 | } |
6279 | ||
4f2de97a | 6280 | spin_lock(&mapping->private_lock); |
81982210 | 6281 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
6282 | if (exists) { |
6283 | spin_unlock(&mapping->private_lock); | |
6284 | unlock_page(p); | |
6285 | put_page(p); | |
6286 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 6287 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 6288 | goto free_eb; |
d1310b2e | 6289 | } |
760f991f QW |
6290 | /* Should not fail, as we have preallocated the memory */ |
6291 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
6292 | ASSERT(!ret); | |
8ff8466d QW |
6293 | /* |
6294 | * To inform we have extra eb under allocation, so that | |
6295 | * detach_extent_buffer_page() won't release the page private | |
6296 | * when the eb hasn't yet been inserted into radix tree. | |
6297 | * | |
6298 | * The ref will be decreased when the eb released the page, in | |
6299 | * detach_extent_buffer_page(). | |
6300 | * Thus needs no special handling in error path. | |
6301 | */ | |
6302 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 6303 | spin_unlock(&mapping->private_lock); |
760f991f | 6304 | |
1e5eb3d6 | 6305 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 6306 | eb->pages[i] = p; |
d1310b2e CM |
6307 | if (!PageUptodate(p)) |
6308 | uptodate = 0; | |
eb14ab8e CM |
6309 | |
6310 | /* | |
b16d011e NB |
6311 | * We can't unlock the pages just yet since the extent buffer |
6312 | * hasn't been properly inserted in the radix tree, this | |
f913cff3 | 6313 | * opens a race with btree_release_folio which can free a page |
b16d011e NB |
6314 | * while we are still filling in all pages for the buffer and |
6315 | * we could crash. | |
eb14ab8e | 6316 | */ |
d1310b2e CM |
6317 | } |
6318 | if (uptodate) | |
b4ce94de | 6319 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
01cd3909 DS |
6320 | again: |
6321 | ret = radix_tree_preload(GFP_NOFS); | |
6322 | if (ret) { | |
6323 | exists = ERR_PTR(ret); | |
6324 | goto free_eb; | |
6325 | } | |
6326 | ||
6327 | spin_lock(&fs_info->buffer_lock); | |
6328 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
6329 | start >> fs_info->sectorsize_bits, eb); | |
6330 | spin_unlock(&fs_info->buffer_lock); | |
6331 | radix_tree_preload_end(); | |
6332 | if (ret == -EEXIST) { | |
6333 | exists = find_extent_buffer(fs_info, start); | |
6334 | if (exists) | |
452c75c3 | 6335 | goto free_eb; |
01cd3909 DS |
6336 | else |
6337 | goto again; | |
6338 | } | |
6af118ce | 6339 | /* add one reference for the tree */ |
0b32f4bb | 6340 | check_buffer_tree_ref(eb); |
34b41ace | 6341 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
6342 | |
6343 | /* | |
b16d011e | 6344 | * Now it's safe to unlock the pages because any calls to |
f913cff3 | 6345 | * btree_release_folio will correctly detect that a page belongs to a |
b16d011e | 6346 | * live buffer and won't free them prematurely. |
eb14ab8e | 6347 | */ |
28187ae5 NB |
6348 | for (i = 0; i < num_pages; i++) |
6349 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
6350 | return eb; |
6351 | ||
6af118ce | 6352 | free_eb: |
5ca64f45 | 6353 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
6354 | for (i = 0; i < num_pages; i++) { |
6355 | if (eb->pages[i]) | |
6356 | unlock_page(eb->pages[i]); | |
6357 | } | |
eb14ab8e | 6358 | |
897ca6e9 | 6359 | btrfs_release_extent_buffer(eb); |
6af118ce | 6360 | return exists; |
d1310b2e | 6361 | } |
d1310b2e | 6362 | |
3083ee2e JB |
6363 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
6364 | { | |
6365 | struct extent_buffer *eb = | |
6366 | container_of(head, struct extent_buffer, rcu_head); | |
6367 | ||
6368 | __free_extent_buffer(eb); | |
6369 | } | |
6370 | ||
f7a52a40 | 6371 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 6372 | __releases(&eb->refs_lock) |
3083ee2e | 6373 | { |
07e21c4d NB |
6374 | lockdep_assert_held(&eb->refs_lock); |
6375 | ||
3083ee2e JB |
6376 | WARN_ON(atomic_read(&eb->refs) == 0); |
6377 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 6378 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 6379 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 6380 | |
815a51c7 | 6381 | spin_unlock(&eb->refs_lock); |
3083ee2e | 6382 | |
01cd3909 DS |
6383 | spin_lock(&fs_info->buffer_lock); |
6384 | radix_tree_delete(&fs_info->buffer_radix, | |
6385 | eb->start >> fs_info->sectorsize_bits); | |
6386 | spin_unlock(&fs_info->buffer_lock); | |
34b41ace JB |
6387 | } else { |
6388 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 6389 | } |
3083ee2e | 6390 | |
8c38938c | 6391 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 6392 | /* Should be safe to release our pages at this point */ |
55ac0139 | 6393 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 6394 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 6395 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
6396 | __free_extent_buffer(eb); |
6397 | return 1; | |
6398 | } | |
6399 | #endif | |
3083ee2e | 6400 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 6401 | return 1; |
3083ee2e JB |
6402 | } |
6403 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
6404 | |
6405 | return 0; | |
3083ee2e JB |
6406 | } |
6407 | ||
d1310b2e CM |
6408 | void free_extent_buffer(struct extent_buffer *eb) |
6409 | { | |
242e18c7 CM |
6410 | int refs; |
6411 | int old; | |
d1310b2e CM |
6412 | if (!eb) |
6413 | return; | |
6414 | ||
242e18c7 CM |
6415 | while (1) { |
6416 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
6417 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
6418 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
6419 | refs == 1)) | |
242e18c7 CM |
6420 | break; |
6421 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
6422 | if (old == refs) | |
6423 | return; | |
6424 | } | |
6425 | ||
3083ee2e JB |
6426 | spin_lock(&eb->refs_lock); |
6427 | if (atomic_read(&eb->refs) == 2 && | |
6428 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 6429 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
6430 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6431 | atomic_dec(&eb->refs); | |
6432 | ||
6433 | /* | |
6434 | * I know this is terrible, but it's temporary until we stop tracking | |
6435 | * the uptodate bits and such for the extent buffers. | |
6436 | */ | |
f7a52a40 | 6437 | release_extent_buffer(eb); |
3083ee2e JB |
6438 | } |
6439 | ||
6440 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
6441 | { | |
6442 | if (!eb) | |
d1310b2e CM |
6443 | return; |
6444 | ||
3083ee2e JB |
6445 | spin_lock(&eb->refs_lock); |
6446 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
6447 | ||
0b32f4bb | 6448 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
6449 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6450 | atomic_dec(&eb->refs); | |
f7a52a40 | 6451 | release_extent_buffer(eb); |
d1310b2e | 6452 | } |
d1310b2e | 6453 | |
0d27797e QW |
6454 | static void btree_clear_page_dirty(struct page *page) |
6455 | { | |
6456 | ASSERT(PageDirty(page)); | |
6457 | ASSERT(PageLocked(page)); | |
6458 | clear_page_dirty_for_io(page); | |
6459 | xa_lock_irq(&page->mapping->i_pages); | |
6460 | if (!PageDirty(page)) | |
6461 | __xa_clear_mark(&page->mapping->i_pages, | |
6462 | page_index(page), PAGECACHE_TAG_DIRTY); | |
6463 | xa_unlock_irq(&page->mapping->i_pages); | |
6464 | } | |
6465 | ||
6466 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
6467 | { | |
6468 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6469 | struct page *page = eb->pages[0]; | |
6470 | bool last; | |
6471 | ||
6472 | /* btree_clear_page_dirty() needs page locked */ | |
6473 | lock_page(page); | |
6474 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
6475 | eb->len); | |
6476 | if (last) | |
6477 | btree_clear_page_dirty(page); | |
6478 | unlock_page(page); | |
6479 | WARN_ON(atomic_read(&eb->refs) == 0); | |
6480 | } | |
6481 | ||
2b48966a | 6482 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 6483 | { |
cc5e31a4 DS |
6484 | int i; |
6485 | int num_pages; | |
d1310b2e CM |
6486 | struct page *page; |
6487 | ||
fbca46eb | 6488 | if (eb->fs_info->nodesize < PAGE_SIZE) |
0d27797e QW |
6489 | return clear_subpage_extent_buffer_dirty(eb); |
6490 | ||
65ad0104 | 6491 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
6492 | |
6493 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 6494 | page = eb->pages[i]; |
b9473439 | 6495 | if (!PageDirty(page)) |
d2c3f4f6 | 6496 | continue; |
a61e6f29 | 6497 | lock_page(page); |
0d27797e | 6498 | btree_clear_page_dirty(page); |
bf0da8c1 | 6499 | ClearPageError(page); |
a61e6f29 | 6500 | unlock_page(page); |
d1310b2e | 6501 | } |
0b32f4bb | 6502 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 6503 | } |
d1310b2e | 6504 | |
abb57ef3 | 6505 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 6506 | { |
cc5e31a4 DS |
6507 | int i; |
6508 | int num_pages; | |
abb57ef3 | 6509 | bool was_dirty; |
d1310b2e | 6510 | |
0b32f4bb JB |
6511 | check_buffer_tree_ref(eb); |
6512 | ||
b9473439 | 6513 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 6514 | |
65ad0104 | 6515 | num_pages = num_extent_pages(eb); |
3083ee2e | 6516 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
6517 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
6518 | ||
0d27797e | 6519 | if (!was_dirty) { |
fbca46eb | 6520 | bool subpage = eb->fs_info->nodesize < PAGE_SIZE; |
51995c39 | 6521 | |
0d27797e QW |
6522 | /* |
6523 | * For subpage case, we can have other extent buffers in the | |
6524 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
6525 | * have to clear page dirty without subpage lock held. | |
6526 | * This can cause race where our page gets dirty cleared after | |
6527 | * we just set it. | |
6528 | * | |
6529 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
6530 | * its page for other reasons, we can use page lock to prevent | |
6531 | * the above race. | |
6532 | */ | |
6533 | if (subpage) | |
6534 | lock_page(eb->pages[0]); | |
6535 | for (i = 0; i < num_pages; i++) | |
6536 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
6537 | eb->start, eb->len); | |
6538 | if (subpage) | |
6539 | unlock_page(eb->pages[0]); | |
6540 | } | |
51995c39 LB |
6541 | #ifdef CONFIG_BTRFS_DEBUG |
6542 | for (i = 0; i < num_pages; i++) | |
6543 | ASSERT(PageDirty(eb->pages[i])); | |
6544 | #endif | |
6545 | ||
b9473439 | 6546 | return was_dirty; |
d1310b2e | 6547 | } |
d1310b2e | 6548 | |
69ba3927 | 6549 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 6550 | { |
251f2acc | 6551 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 6552 | struct page *page; |
cc5e31a4 | 6553 | int num_pages; |
251f2acc | 6554 | int i; |
1259ab75 | 6555 | |
b4ce94de | 6556 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6557 | num_pages = num_extent_pages(eb); |
1259ab75 | 6558 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6559 | page = eb->pages[i]; |
fbca46eb QW |
6560 | if (!page) |
6561 | continue; | |
6562 | ||
6563 | /* | |
6564 | * This is special handling for metadata subpage, as regular | |
6565 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6566 | */ | |
6567 | if (fs_info->nodesize >= PAGE_SIZE) | |
6568 | ClearPageUptodate(page); | |
6569 | else | |
6570 | btrfs_subpage_clear_uptodate(fs_info, page, eb->start, | |
6571 | eb->len); | |
1259ab75 | 6572 | } |
1259ab75 CM |
6573 | } |
6574 | ||
09c25a8c | 6575 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 6576 | { |
251f2acc | 6577 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 6578 | struct page *page; |
cc5e31a4 | 6579 | int num_pages; |
251f2acc | 6580 | int i; |
d1310b2e | 6581 | |
0b32f4bb | 6582 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6583 | num_pages = num_extent_pages(eb); |
d1310b2e | 6584 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6585 | page = eb->pages[i]; |
fbca46eb QW |
6586 | |
6587 | /* | |
6588 | * This is special handling for metadata subpage, as regular | |
6589 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6590 | */ | |
6591 | if (fs_info->nodesize >= PAGE_SIZE) | |
6592 | SetPageUptodate(page); | |
6593 | else | |
6594 | btrfs_subpage_set_uptodate(fs_info, page, eb->start, | |
6595 | eb->len); | |
d1310b2e | 6596 | } |
d1310b2e | 6597 | } |
d1310b2e | 6598 | |
4012daf7 QW |
6599 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
6600 | int mirror_num) | |
6601 | { | |
6602 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6603 | struct extent_io_tree *io_tree; | |
6604 | struct page *page = eb->pages[0]; | |
390ed29b | 6605 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
4012daf7 QW |
6606 | int ret = 0; |
6607 | ||
6608 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
6609 | ASSERT(PagePrivate(page)); | |
6610 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
6611 | ||
6612 | if (wait == WAIT_NONE) { | |
dc56219f GR |
6613 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1)) |
6614 | return -EAGAIN; | |
4012daf7 QW |
6615 | } else { |
6616 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6617 | if (ret < 0) | |
6618 | return ret; | |
6619 | } | |
6620 | ||
6621 | ret = 0; | |
6622 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
6623 | PageUptodate(page) || | |
6624 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
6625 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
6626 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6627 | return ret; | |
6628 | } | |
6629 | ||
6630 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
6631 | eb->read_mirror = 0; | |
6632 | atomic_set(&eb->io_pages, 1); | |
6633 | check_buffer_tree_ref(eb); | |
6634 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); | |
6635 | ||
3d078efa | 6636 | btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len); |
390ed29b QW |
6637 | ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl, |
6638 | page, eb->start, eb->len, | |
6639 | eb->start - page_offset(page), | |
6640 | end_bio_extent_readpage, mirror_num, 0, | |
4012daf7 QW |
6641 | true); |
6642 | if (ret) { | |
6643 | /* | |
6644 | * In the endio function, if we hit something wrong we will | |
6645 | * increase the io_pages, so here we need to decrease it for | |
6646 | * error path. | |
6647 | */ | |
6648 | atomic_dec(&eb->io_pages); | |
6649 | } | |
390ed29b | 6650 | if (bio_ctrl.bio) { |
c9583ada | 6651 | submit_one_bio(bio_ctrl.bio, mirror_num, 0); |
390ed29b | 6652 | bio_ctrl.bio = NULL; |
4012daf7 QW |
6653 | } |
6654 | if (ret || wait != WAIT_COMPLETE) | |
6655 | return ret; | |
6656 | ||
6657 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED); | |
6658 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
6659 | ret = -EIO; | |
6660 | return ret; | |
6661 | } | |
6662 | ||
c2ccfbc6 | 6663 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 6664 | { |
cc5e31a4 | 6665 | int i; |
d1310b2e CM |
6666 | struct page *page; |
6667 | int err; | |
6668 | int ret = 0; | |
ce9adaa5 CM |
6669 | int locked_pages = 0; |
6670 | int all_uptodate = 1; | |
cc5e31a4 | 6671 | int num_pages; |
727011e0 | 6672 | unsigned long num_reads = 0; |
390ed29b | 6673 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
a86c12c7 | 6674 | |
b4ce94de | 6675 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
6676 | return 0; |
6677 | ||
651740a5 JB |
6678 | /* |
6679 | * We could have had EXTENT_BUFFER_UPTODATE cleared by the write | |
6680 | * operation, which could potentially still be in flight. In this case | |
6681 | * we simply want to return an error. | |
6682 | */ | |
6683 | if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))) | |
6684 | return -EIO; | |
6685 | ||
fbca46eb | 6686 | if (eb->fs_info->nodesize < PAGE_SIZE) |
4012daf7 QW |
6687 | return read_extent_buffer_subpage(eb, wait, mirror_num); |
6688 | ||
65ad0104 | 6689 | num_pages = num_extent_pages(eb); |
8436ea91 | 6690 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6691 | page = eb->pages[i]; |
bb82ab88 | 6692 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
6693 | /* |
6694 | * WAIT_NONE is only utilized by readahead. If we can't | |
6695 | * acquire the lock atomically it means either the eb | |
6696 | * is being read out or under modification. | |
6697 | * Either way the eb will be or has been cached, | |
6698 | * readahead can exit safely. | |
6699 | */ | |
2db04966 | 6700 | if (!trylock_page(page)) |
ce9adaa5 | 6701 | goto unlock_exit; |
d1310b2e CM |
6702 | } else { |
6703 | lock_page(page); | |
6704 | } | |
ce9adaa5 | 6705 | locked_pages++; |
2571e739 LB |
6706 | } |
6707 | /* | |
6708 | * We need to firstly lock all pages to make sure that | |
6709 | * the uptodate bit of our pages won't be affected by | |
6710 | * clear_extent_buffer_uptodate(). | |
6711 | */ | |
8436ea91 | 6712 | for (i = 0; i < num_pages; i++) { |
2571e739 | 6713 | page = eb->pages[i]; |
727011e0 CM |
6714 | if (!PageUptodate(page)) { |
6715 | num_reads++; | |
ce9adaa5 | 6716 | all_uptodate = 0; |
727011e0 | 6717 | } |
ce9adaa5 | 6718 | } |
2571e739 | 6719 | |
ce9adaa5 | 6720 | if (all_uptodate) { |
8436ea91 | 6721 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
6722 | goto unlock_exit; |
6723 | } | |
6724 | ||
656f30db | 6725 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 6726 | eb->read_mirror = 0; |
0b32f4bb | 6727 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e | 6728 | /* |
f913cff3 | 6729 | * It is possible for release_folio to clear the TREE_REF bit before we |
6bf9cd2e BB |
6730 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. |
6731 | */ | |
6732 | check_buffer_tree_ref(eb); | |
8436ea91 | 6733 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6734 | page = eb->pages[i]; |
baf863b9 | 6735 | |
ce9adaa5 | 6736 | if (!PageUptodate(page)) { |
baf863b9 LB |
6737 | if (ret) { |
6738 | atomic_dec(&eb->io_pages); | |
6739 | unlock_page(page); | |
6740 | continue; | |
6741 | } | |
6742 | ||
f188591e | 6743 | ClearPageError(page); |
0420177c | 6744 | err = submit_extent_page(REQ_OP_READ | REQ_META, NULL, |
390ed29b QW |
6745 | &bio_ctrl, page, page_offset(page), |
6746 | PAGE_SIZE, 0, end_bio_extent_readpage, | |
6747 | mirror_num, 0, false); | |
baf863b9 | 6748 | if (err) { |
baf863b9 | 6749 | /* |
0420177c NB |
6750 | * We failed to submit the bio so it's the |
6751 | * caller's responsibility to perform cleanup | |
6752 | * i.e unlock page/set error bit. | |
baf863b9 | 6753 | */ |
0420177c NB |
6754 | ret = err; |
6755 | SetPageError(page); | |
6756 | unlock_page(page); | |
baf863b9 LB |
6757 | atomic_dec(&eb->io_pages); |
6758 | } | |
d1310b2e CM |
6759 | } else { |
6760 | unlock_page(page); | |
6761 | } | |
6762 | } | |
6763 | ||
390ed29b | 6764 | if (bio_ctrl.bio) { |
0f07003b | 6765 | submit_one_bio(bio_ctrl.bio, mirror_num, bio_ctrl.compress_type); |
390ed29b | 6766 | bio_ctrl.bio = NULL; |
355808c2 | 6767 | } |
a86c12c7 | 6768 | |
bb82ab88 | 6769 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 6770 | return ret; |
d397712b | 6771 | |
8436ea91 | 6772 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6773 | page = eb->pages[i]; |
d1310b2e | 6774 | wait_on_page_locked(page); |
d397712b | 6775 | if (!PageUptodate(page)) |
d1310b2e | 6776 | ret = -EIO; |
d1310b2e | 6777 | } |
d397712b | 6778 | |
d1310b2e | 6779 | return ret; |
ce9adaa5 CM |
6780 | |
6781 | unlock_exit: | |
d397712b | 6782 | while (locked_pages > 0) { |
ce9adaa5 | 6783 | locked_pages--; |
8436ea91 JB |
6784 | page = eb->pages[locked_pages]; |
6785 | unlock_page(page); | |
ce9adaa5 CM |
6786 | } |
6787 | return ret; | |
d1310b2e | 6788 | } |
d1310b2e | 6789 | |
f98b6215 QW |
6790 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
6791 | unsigned long len) | |
6792 | { | |
6793 | btrfs_warn(eb->fs_info, | |
6794 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
6795 | eb->start, eb->len, start, len); | |
6796 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6797 | ||
6798 | return true; | |
6799 | } | |
6800 | ||
6801 | /* | |
6802 | * Check if the [start, start + len) range is valid before reading/writing | |
6803 | * the eb. | |
6804 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
6805 | * | |
6806 | * Caller should not touch the dst/src memory if this function returns error. | |
6807 | */ | |
6808 | static inline int check_eb_range(const struct extent_buffer *eb, | |
6809 | unsigned long start, unsigned long len) | |
6810 | { | |
6811 | unsigned long offset; | |
6812 | ||
6813 | /* start, start + len should not go beyond eb->len nor overflow */ | |
6814 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
6815 | return report_eb_range(eb, start, len); | |
6816 | ||
6817 | return false; | |
6818 | } | |
6819 | ||
1cbb1f45 JM |
6820 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
6821 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6822 | { |
6823 | size_t cur; | |
6824 | size_t offset; | |
6825 | struct page *page; | |
6826 | char *kaddr; | |
6827 | char *dst = (char *)dstv; | |
884b07d0 | 6828 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6829 | |
f98b6215 | 6830 | if (check_eb_range(eb, start, len)) |
f716abd5 | 6831 | return; |
d1310b2e | 6832 | |
884b07d0 | 6833 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6834 | |
d397712b | 6835 | while (len > 0) { |
fb85fc9a | 6836 | page = eb->pages[i]; |
d1310b2e | 6837 | |
09cbfeaf | 6838 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 6839 | kaddr = page_address(page); |
d1310b2e | 6840 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
6841 | |
6842 | dst += cur; | |
6843 | len -= cur; | |
6844 | offset = 0; | |
6845 | i++; | |
6846 | } | |
6847 | } | |
d1310b2e | 6848 | |
a48b73ec JB |
6849 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
6850 | void __user *dstv, | |
6851 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
6852 | { |
6853 | size_t cur; | |
6854 | size_t offset; | |
6855 | struct page *page; | |
6856 | char *kaddr; | |
6857 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 6858 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
6859 | int ret = 0; |
6860 | ||
6861 | WARN_ON(start > eb->len); | |
6862 | WARN_ON(start + len > eb->start + eb->len); | |
6863 | ||
884b07d0 | 6864 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
6865 | |
6866 | while (len > 0) { | |
fb85fc9a | 6867 | page = eb->pages[i]; |
550ac1d8 | 6868 | |
09cbfeaf | 6869 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 6870 | kaddr = page_address(page); |
a48b73ec | 6871 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
6872 | ret = -EFAULT; |
6873 | break; | |
6874 | } | |
6875 | ||
6876 | dst += cur; | |
6877 | len -= cur; | |
6878 | offset = 0; | |
6879 | i++; | |
6880 | } | |
6881 | ||
6882 | return ret; | |
6883 | } | |
6884 | ||
1cbb1f45 JM |
6885 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
6886 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6887 | { |
6888 | size_t cur; | |
6889 | size_t offset; | |
6890 | struct page *page; | |
6891 | char *kaddr; | |
6892 | char *ptr = (char *)ptrv; | |
884b07d0 | 6893 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
6894 | int ret = 0; |
6895 | ||
f98b6215 QW |
6896 | if (check_eb_range(eb, start, len)) |
6897 | return -EINVAL; | |
d1310b2e | 6898 | |
884b07d0 | 6899 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6900 | |
d397712b | 6901 | while (len > 0) { |
fb85fc9a | 6902 | page = eb->pages[i]; |
d1310b2e | 6903 | |
09cbfeaf | 6904 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 6905 | |
a6591715 | 6906 | kaddr = page_address(page); |
d1310b2e | 6907 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
6908 | if (ret) |
6909 | break; | |
6910 | ||
6911 | ptr += cur; | |
6912 | len -= cur; | |
6913 | offset = 0; | |
6914 | i++; | |
6915 | } | |
6916 | return ret; | |
6917 | } | |
d1310b2e | 6918 | |
b8f95771 QW |
6919 | /* |
6920 | * Check that the extent buffer is uptodate. | |
6921 | * | |
6922 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
6923 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
6924 | */ | |
6925 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
6926 | struct page *page) | |
6927 | { | |
6928 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6929 | ||
a50e1fcb JB |
6930 | /* |
6931 | * If we are using the commit root we could potentially clear a page | |
6932 | * Uptodate while we're using the extent buffer that we've previously | |
6933 | * looked up. We don't want to complain in this case, as the page was | |
6934 | * valid before, we just didn't write it out. Instead we want to catch | |
6935 | * the case where we didn't actually read the block properly, which | |
6936 | * would have !PageUptodate && !PageError, as we clear PageError before | |
6937 | * reading. | |
6938 | */ | |
fbca46eb | 6939 | if (fs_info->nodesize < PAGE_SIZE) { |
a50e1fcb | 6940 | bool uptodate, error; |
b8f95771 QW |
6941 | |
6942 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
6943 | eb->start, eb->len); | |
a50e1fcb JB |
6944 | error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len); |
6945 | WARN_ON(!uptodate && !error); | |
b8f95771 | 6946 | } else { |
a50e1fcb | 6947 | WARN_ON(!PageUptodate(page) && !PageError(page)); |
b8f95771 QW |
6948 | } |
6949 | } | |
6950 | ||
2b48966a | 6951 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
6952 | const void *srcv) |
6953 | { | |
6954 | char *kaddr; | |
6955 | ||
b8f95771 | 6956 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6957 | kaddr = page_address(eb->pages[0]) + |
6958 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
6959 | chunk_tree_uuid)); | |
6960 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6961 | } |
6962 | ||
2b48966a | 6963 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
6964 | { |
6965 | char *kaddr; | |
6966 | ||
b8f95771 | 6967 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6968 | kaddr = page_address(eb->pages[0]) + |
6969 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
6970 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6971 | } |
6972 | ||
2b48966a | 6973 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
6974 | unsigned long start, unsigned long len) |
6975 | { | |
6976 | size_t cur; | |
6977 | size_t offset; | |
6978 | struct page *page; | |
6979 | char *kaddr; | |
6980 | char *src = (char *)srcv; | |
884b07d0 | 6981 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6982 | |
d3575156 NA |
6983 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
6984 | ||
f98b6215 QW |
6985 | if (check_eb_range(eb, start, len)) |
6986 | return; | |
d1310b2e | 6987 | |
884b07d0 | 6988 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6989 | |
d397712b | 6990 | while (len > 0) { |
fb85fc9a | 6991 | page = eb->pages[i]; |
b8f95771 | 6992 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6993 | |
09cbfeaf | 6994 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6995 | kaddr = page_address(page); |
d1310b2e | 6996 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
6997 | |
6998 | src += cur; | |
6999 | len -= cur; | |
7000 | offset = 0; | |
7001 | i++; | |
7002 | } | |
7003 | } | |
d1310b2e | 7004 | |
2b48966a | 7005 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 7006 | unsigned long len) |
d1310b2e CM |
7007 | { |
7008 | size_t cur; | |
7009 | size_t offset; | |
7010 | struct page *page; | |
7011 | char *kaddr; | |
884b07d0 | 7012 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 7013 | |
f98b6215 QW |
7014 | if (check_eb_range(eb, start, len)) |
7015 | return; | |
d1310b2e | 7016 | |
884b07d0 | 7017 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 7018 | |
d397712b | 7019 | while (len > 0) { |
fb85fc9a | 7020 | page = eb->pages[i]; |
b8f95771 | 7021 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 7022 | |
09cbfeaf | 7023 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 7024 | kaddr = page_address(page); |
b159fa28 | 7025 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
7026 | |
7027 | len -= cur; | |
7028 | offset = 0; | |
7029 | i++; | |
7030 | } | |
7031 | } | |
d1310b2e | 7032 | |
2b48966a DS |
7033 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
7034 | const struct extent_buffer *src) | |
58e8012c DS |
7035 | { |
7036 | int i; | |
cc5e31a4 | 7037 | int num_pages; |
58e8012c DS |
7038 | |
7039 | ASSERT(dst->len == src->len); | |
7040 | ||
fbca46eb | 7041 | if (dst->fs_info->nodesize >= PAGE_SIZE) { |
884b07d0 QW |
7042 | num_pages = num_extent_pages(dst); |
7043 | for (i = 0; i < num_pages; i++) | |
7044 | copy_page(page_address(dst->pages[i]), | |
7045 | page_address(src->pages[i])); | |
7046 | } else { | |
7047 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
7048 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
7049 | ||
fbca46eb | 7050 | ASSERT(src->fs_info->nodesize < PAGE_SIZE); |
884b07d0 QW |
7051 | memcpy(page_address(dst->pages[0]) + dst_offset, |
7052 | page_address(src->pages[0]) + src_offset, | |
7053 | src->len); | |
7054 | } | |
58e8012c DS |
7055 | } |
7056 | ||
2b48966a DS |
7057 | void copy_extent_buffer(const struct extent_buffer *dst, |
7058 | const struct extent_buffer *src, | |
d1310b2e CM |
7059 | unsigned long dst_offset, unsigned long src_offset, |
7060 | unsigned long len) | |
7061 | { | |
7062 | u64 dst_len = dst->len; | |
7063 | size_t cur; | |
7064 | size_t offset; | |
7065 | struct page *page; | |
7066 | char *kaddr; | |
884b07d0 | 7067 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 7068 | |
f98b6215 QW |
7069 | if (check_eb_range(dst, dst_offset, len) || |
7070 | check_eb_range(src, src_offset, len)) | |
7071 | return; | |
7072 | ||
d1310b2e CM |
7073 | WARN_ON(src->len != dst_len); |
7074 | ||
884b07d0 | 7075 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 7076 | |
d397712b | 7077 | while (len > 0) { |
fb85fc9a | 7078 | page = dst->pages[i]; |
b8f95771 | 7079 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 7080 | |
09cbfeaf | 7081 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 7082 | |
a6591715 | 7083 | kaddr = page_address(page); |
d1310b2e | 7084 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
7085 | |
7086 | src_offset += cur; | |
7087 | len -= cur; | |
7088 | offset = 0; | |
7089 | i++; | |
7090 | } | |
7091 | } | |
d1310b2e | 7092 | |
3e1e8bb7 OS |
7093 | /* |
7094 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
7095 | * given bit number | |
7096 | * @eb: the extent buffer | |
7097 | * @start: offset of the bitmap item in the extent buffer | |
7098 | * @nr: bit number | |
7099 | * @page_index: return index of the page in the extent buffer that contains the | |
7100 | * given bit number | |
7101 | * @page_offset: return offset into the page given by page_index | |
7102 | * | |
7103 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
7104 | * contains a given bit. | |
7105 | */ | |
2b48966a | 7106 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
7107 | unsigned long start, unsigned long nr, |
7108 | unsigned long *page_index, | |
7109 | size_t *page_offset) | |
7110 | { | |
3e1e8bb7 OS |
7111 | size_t byte_offset = BIT_BYTE(nr); |
7112 | size_t offset; | |
7113 | ||
7114 | /* | |
7115 | * The byte we want is the offset of the extent buffer + the offset of | |
7116 | * the bitmap item in the extent buffer + the offset of the byte in the | |
7117 | * bitmap item. | |
7118 | */ | |
884b07d0 | 7119 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 7120 | |
09cbfeaf | 7121 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 7122 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
7123 | } |
7124 | ||
7125 | /** | |
7126 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
7127 | * @eb: the extent buffer | |
7128 | * @start: offset of the bitmap item in the extent buffer | |
7129 | * @nr: bit number to test | |
7130 | */ | |
2b48966a | 7131 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7132 | unsigned long nr) |
7133 | { | |
2fe1d551 | 7134 | u8 *kaddr; |
3e1e8bb7 OS |
7135 | struct page *page; |
7136 | unsigned long i; | |
7137 | size_t offset; | |
7138 | ||
7139 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
7140 | page = eb->pages[i]; | |
b8f95771 | 7141 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7142 | kaddr = page_address(page); |
7143 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
7144 | } | |
7145 | ||
7146 | /** | |
7147 | * extent_buffer_bitmap_set - set an area of a bitmap | |
7148 | * @eb: the extent buffer | |
7149 | * @start: offset of the bitmap item in the extent buffer | |
7150 | * @pos: bit number of the first bit | |
7151 | * @len: number of bits to set | |
7152 | */ | |
2b48966a | 7153 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7154 | unsigned long pos, unsigned long len) |
7155 | { | |
2fe1d551 | 7156 | u8 *kaddr; |
3e1e8bb7 OS |
7157 | struct page *page; |
7158 | unsigned long i; | |
7159 | size_t offset; | |
7160 | const unsigned int size = pos + len; | |
7161 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7162 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7163 | |
7164 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7165 | page = eb->pages[i]; | |
b8f95771 | 7166 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7167 | kaddr = page_address(page); |
7168 | ||
7169 | while (len >= bits_to_set) { | |
7170 | kaddr[offset] |= mask_to_set; | |
7171 | len -= bits_to_set; | |
7172 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 7173 | mask_to_set = ~0; |
09cbfeaf | 7174 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7175 | offset = 0; |
7176 | page = eb->pages[++i]; | |
b8f95771 | 7177 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7178 | kaddr = page_address(page); |
7179 | } | |
7180 | } | |
7181 | if (len) { | |
7182 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
7183 | kaddr[offset] |= mask_to_set; | |
7184 | } | |
7185 | } | |
7186 | ||
7187 | ||
7188 | /** | |
7189 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
7190 | * @eb: the extent buffer | |
7191 | * @start: offset of the bitmap item in the extent buffer | |
7192 | * @pos: bit number of the first bit | |
7193 | * @len: number of bits to clear | |
7194 | */ | |
2b48966a DS |
7195 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
7196 | unsigned long start, unsigned long pos, | |
7197 | unsigned long len) | |
3e1e8bb7 | 7198 | { |
2fe1d551 | 7199 | u8 *kaddr; |
3e1e8bb7 OS |
7200 | struct page *page; |
7201 | unsigned long i; | |
7202 | size_t offset; | |
7203 | const unsigned int size = pos + len; | |
7204 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7205 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7206 | |
7207 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7208 | page = eb->pages[i]; | |
b8f95771 | 7209 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7210 | kaddr = page_address(page); |
7211 | ||
7212 | while (len >= bits_to_clear) { | |
7213 | kaddr[offset] &= ~mask_to_clear; | |
7214 | len -= bits_to_clear; | |
7215 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 7216 | mask_to_clear = ~0; |
09cbfeaf | 7217 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7218 | offset = 0; |
7219 | page = eb->pages[++i]; | |
b8f95771 | 7220 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7221 | kaddr = page_address(page); |
7222 | } | |
7223 | } | |
7224 | if (len) { | |
7225 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
7226 | kaddr[offset] &= ~mask_to_clear; | |
7227 | } | |
7228 | } | |
7229 | ||
3387206f ST |
7230 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
7231 | { | |
7232 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
7233 | return distance < len; | |
7234 | } | |
7235 | ||
d1310b2e CM |
7236 | static void copy_pages(struct page *dst_page, struct page *src_page, |
7237 | unsigned long dst_off, unsigned long src_off, | |
7238 | unsigned long len) | |
7239 | { | |
a6591715 | 7240 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 7241 | char *src_kaddr; |
727011e0 | 7242 | int must_memmove = 0; |
d1310b2e | 7243 | |
3387206f | 7244 | if (dst_page != src_page) { |
a6591715 | 7245 | src_kaddr = page_address(src_page); |
3387206f | 7246 | } else { |
d1310b2e | 7247 | src_kaddr = dst_kaddr; |
727011e0 CM |
7248 | if (areas_overlap(src_off, dst_off, len)) |
7249 | must_memmove = 1; | |
3387206f | 7250 | } |
d1310b2e | 7251 | |
727011e0 CM |
7252 | if (must_memmove) |
7253 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
7254 | else | |
7255 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
7256 | } |
7257 | ||
2b48966a DS |
7258 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
7259 | unsigned long dst_offset, unsigned long src_offset, | |
7260 | unsigned long len) | |
d1310b2e CM |
7261 | { |
7262 | size_t cur; | |
7263 | size_t dst_off_in_page; | |
7264 | size_t src_off_in_page; | |
d1310b2e CM |
7265 | unsigned long dst_i; |
7266 | unsigned long src_i; | |
7267 | ||
f98b6215 QW |
7268 | if (check_eb_range(dst, dst_offset, len) || |
7269 | check_eb_range(dst, src_offset, len)) | |
7270 | return; | |
d1310b2e | 7271 | |
d397712b | 7272 | while (len > 0) { |
884b07d0 QW |
7273 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
7274 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 7275 | |
884b07d0 QW |
7276 | dst_i = get_eb_page_index(dst_offset); |
7277 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 7278 | |
09cbfeaf | 7279 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
7280 | src_off_in_page)); |
7281 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 7282 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 7283 | |
fb85fc9a | 7284 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7285 | dst_off_in_page, src_off_in_page, cur); |
7286 | ||
7287 | src_offset += cur; | |
7288 | dst_offset += cur; | |
7289 | len -= cur; | |
7290 | } | |
7291 | } | |
d1310b2e | 7292 | |
2b48966a DS |
7293 | void memmove_extent_buffer(const struct extent_buffer *dst, |
7294 | unsigned long dst_offset, unsigned long src_offset, | |
7295 | unsigned long len) | |
d1310b2e CM |
7296 | { |
7297 | size_t cur; | |
7298 | size_t dst_off_in_page; | |
7299 | size_t src_off_in_page; | |
7300 | unsigned long dst_end = dst_offset + len - 1; | |
7301 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
7302 | unsigned long dst_i; |
7303 | unsigned long src_i; | |
7304 | ||
f98b6215 QW |
7305 | if (check_eb_range(dst, dst_offset, len) || |
7306 | check_eb_range(dst, src_offset, len)) | |
7307 | return; | |
727011e0 | 7308 | if (dst_offset < src_offset) { |
d1310b2e CM |
7309 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
7310 | return; | |
7311 | } | |
d397712b | 7312 | while (len > 0) { |
884b07d0 QW |
7313 | dst_i = get_eb_page_index(dst_end); |
7314 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 7315 | |
884b07d0 QW |
7316 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
7317 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
7318 | |
7319 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
7320 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 7321 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7322 | dst_off_in_page - cur + 1, |
7323 | src_off_in_page - cur + 1, cur); | |
7324 | ||
7325 | dst_end -= cur; | |
7326 | src_end -= cur; | |
7327 | len -= cur; | |
7328 | } | |
7329 | } | |
6af118ce | 7330 | |
01cd3909 | 7331 | #define GANG_LOOKUP_SIZE 16 |
d1e86e3f QW |
7332 | static struct extent_buffer *get_next_extent_buffer( |
7333 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
7334 | { | |
01cd3909 DS |
7335 | struct extent_buffer *gang[GANG_LOOKUP_SIZE]; |
7336 | struct extent_buffer *found = NULL; | |
d1e86e3f | 7337 | u64 page_start = page_offset(page); |
01cd3909 | 7338 | u64 cur = page_start; |
d1e86e3f QW |
7339 | |
7340 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
d1e86e3f QW |
7341 | lockdep_assert_held(&fs_info->buffer_lock); |
7342 | ||
01cd3909 DS |
7343 | while (cur < page_start + PAGE_SIZE) { |
7344 | int ret; | |
7345 | int i; | |
7346 | ||
7347 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, | |
7348 | (void **)gang, cur >> fs_info->sectorsize_bits, | |
7349 | min_t(unsigned int, GANG_LOOKUP_SIZE, | |
7350 | PAGE_SIZE / fs_info->nodesize)); | |
7351 | if (ret == 0) | |
7352 | goto out; | |
7353 | for (i = 0; i < ret; i++) { | |
7354 | /* Already beyond page end */ | |
7355 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
7356 | goto out; | |
7357 | /* Found one */ | |
7358 | if (gang[i]->start >= bytenr) { | |
7359 | found = gang[i]; | |
7360 | goto out; | |
7361 | } | |
7362 | } | |
7363 | cur = gang[ret - 1]->start + gang[ret - 1]->len; | |
d1e86e3f | 7364 | } |
01cd3909 DS |
7365 | out: |
7366 | return found; | |
d1e86e3f QW |
7367 | } |
7368 | ||
7369 | static int try_release_subpage_extent_buffer(struct page *page) | |
7370 | { | |
7371 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7372 | u64 cur = page_offset(page); | |
7373 | const u64 end = page_offset(page) + PAGE_SIZE; | |
7374 | int ret; | |
7375 | ||
7376 | while (cur < end) { | |
7377 | struct extent_buffer *eb = NULL; | |
7378 | ||
7379 | /* | |
7380 | * Unlike try_release_extent_buffer() which uses page->private | |
7381 | * to grab buffer, for subpage case we rely on radix tree, thus | |
7382 | * we need to ensure radix tree consistency. | |
7383 | * | |
7384 | * We also want an atomic snapshot of the radix tree, thus go | |
7385 | * with spinlock rather than RCU. | |
7386 | */ | |
7387 | spin_lock(&fs_info->buffer_lock); | |
7388 | eb = get_next_extent_buffer(fs_info, page, cur); | |
7389 | if (!eb) { | |
7390 | /* No more eb in the page range after or at cur */ | |
7391 | spin_unlock(&fs_info->buffer_lock); | |
7392 | break; | |
7393 | } | |
7394 | cur = eb->start + eb->len; | |
7395 | ||
7396 | /* | |
7397 | * The same as try_release_extent_buffer(), to ensure the eb | |
7398 | * won't disappear out from under us. | |
7399 | */ | |
7400 | spin_lock(&eb->refs_lock); | |
7401 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
7402 | spin_unlock(&eb->refs_lock); | |
7403 | spin_unlock(&fs_info->buffer_lock); | |
7404 | break; | |
7405 | } | |
7406 | spin_unlock(&fs_info->buffer_lock); | |
7407 | ||
7408 | /* | |
7409 | * If tree ref isn't set then we know the ref on this eb is a | |
7410 | * real ref, so just return, this eb will likely be freed soon | |
7411 | * anyway. | |
7412 | */ | |
7413 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
7414 | spin_unlock(&eb->refs_lock); | |
7415 | break; | |
7416 | } | |
7417 | ||
7418 | /* | |
7419 | * Here we don't care about the return value, we will always | |
7420 | * check the page private at the end. And | |
7421 | * release_extent_buffer() will release the refs_lock. | |
7422 | */ | |
7423 | release_extent_buffer(eb); | |
7424 | } | |
7425 | /* | |
7426 | * Finally to check if we have cleared page private, as if we have | |
7427 | * released all ebs in the page, the page private should be cleared now. | |
7428 | */ | |
7429 | spin_lock(&page->mapping->private_lock); | |
7430 | if (!PagePrivate(page)) | |
7431 | ret = 1; | |
7432 | else | |
7433 | ret = 0; | |
7434 | spin_unlock(&page->mapping->private_lock); | |
7435 | return ret; | |
7436 | ||
7437 | } | |
7438 | ||
f7a52a40 | 7439 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 7440 | { |
6af118ce | 7441 | struct extent_buffer *eb; |
6af118ce | 7442 | |
fbca46eb | 7443 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
d1e86e3f QW |
7444 | return try_release_subpage_extent_buffer(page); |
7445 | ||
3083ee2e | 7446 | /* |
d1e86e3f QW |
7447 | * We need to make sure nobody is changing page->private, as we rely on |
7448 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
7449 | */ |
7450 | spin_lock(&page->mapping->private_lock); | |
7451 | if (!PagePrivate(page)) { | |
7452 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 7453 | return 1; |
45f49bce | 7454 | } |
6af118ce | 7455 | |
3083ee2e JB |
7456 | eb = (struct extent_buffer *)page->private; |
7457 | BUG_ON(!eb); | |
19fe0a8b MX |
7458 | |
7459 | /* | |
3083ee2e JB |
7460 | * This is a little awful but should be ok, we need to make sure that |
7461 | * the eb doesn't disappear out from under us while we're looking at | |
7462 | * this page. | |
19fe0a8b | 7463 | */ |
3083ee2e | 7464 | spin_lock(&eb->refs_lock); |
0b32f4bb | 7465 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
7466 | spin_unlock(&eb->refs_lock); |
7467 | spin_unlock(&page->mapping->private_lock); | |
7468 | return 0; | |
b9473439 | 7469 | } |
3083ee2e | 7470 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 7471 | |
19fe0a8b | 7472 | /* |
3083ee2e JB |
7473 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
7474 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 7475 | */ |
3083ee2e JB |
7476 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
7477 | spin_unlock(&eb->refs_lock); | |
7478 | return 0; | |
b9473439 | 7479 | } |
19fe0a8b | 7480 | |
f7a52a40 | 7481 | return release_extent_buffer(eb); |
6af118ce | 7482 | } |
bfb484d9 JB |
7483 | |
7484 | /* | |
7485 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
7486 | * @fs_info: the fs_info | |
7487 | * @bytenr: bytenr to read | |
3fbaf258 | 7488 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 7489 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 7490 | * @level: level for the eb |
bfb484d9 JB |
7491 | * |
7492 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
7493 | * normal uptodate check of the eb, without checking the generation. If we have | |
7494 | * to read the block we will not block on anything. | |
7495 | */ | |
7496 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 7497 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
7498 | { |
7499 | struct extent_buffer *eb; | |
7500 | int ret; | |
7501 | ||
3fbaf258 | 7502 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
7503 | if (IS_ERR(eb)) |
7504 | return; | |
7505 | ||
7506 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
7507 | free_extent_buffer(eb); | |
7508 | return; | |
7509 | } | |
7510 | ||
7511 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
7512 | if (ret < 0) | |
7513 | free_extent_buffer_stale(eb); | |
7514 | else | |
7515 | free_extent_buffer(eb); | |
7516 | } | |
7517 | ||
7518 | /* | |
7519 | * btrfs_readahead_node_child - readahead a node's child block | |
7520 | * @node: parent node we're reading from | |
7521 | * @slot: slot in the parent node for the child we want to read | |
7522 | * | |
7523 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
7524 | * the slot in the node provided. | |
7525 | */ | |
7526 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
7527 | { | |
7528 | btrfs_readahead_tree_block(node->fs_info, | |
7529 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
7530 | btrfs_header_owner(node), |
7531 | btrfs_node_ptr_generation(node, slot), | |
7532 | btrfs_header_level(node) - 1); | |
bfb484d9 | 7533 | } |