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 | ||
c0111c44 QW |
2730 | static blk_status_t submit_data_read_repair(struct inode *inode, |
2731 | struct bio *failed_bio, | |
2732 | u32 bio_offset, struct page *page, | |
2733 | unsigned int pgoff, | |
2734 | u64 start, u64 end, | |
2735 | int failed_mirror, | |
2736 | unsigned int error_bitmap) | |
150e4b05 QW |
2737 | { |
2738 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
2739 | const u32 sectorsize = fs_info->sectorsize; | |
2740 | const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits; | |
2741 | int error = 0; | |
2742 | int i; | |
2743 | ||
2744 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); | |
2745 | ||
c0111c44 QW |
2746 | /* This repair is only for data */ |
2747 | ASSERT(is_data_inode(inode)); | |
2748 | ||
150e4b05 QW |
2749 | /* We're here because we had some read errors or csum mismatch */ |
2750 | ASSERT(error_bitmap); | |
2751 | ||
2752 | /* | |
2753 | * We only get called on buffered IO, thus page must be mapped and bio | |
2754 | * must not be cloned. | |
2755 | */ | |
2756 | ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED)); | |
2757 | ||
2758 | /* Iterate through all the sectors in the range */ | |
2759 | for (i = 0; i < nr_bits; i++) { | |
2760 | const unsigned int offset = i * sectorsize; | |
2761 | struct extent_state *cached = NULL; | |
2762 | bool uptodate = false; | |
2763 | int ret; | |
2764 | ||
2765 | if (!(error_bitmap & (1U << i))) { | |
2766 | /* | |
2767 | * This sector has no error, just end the page read | |
2768 | * and unlock the range. | |
2769 | */ | |
2770 | uptodate = true; | |
2771 | goto next; | |
2772 | } | |
2773 | ||
2774 | ret = btrfs_repair_one_sector(inode, failed_bio, | |
2775 | bio_offset + offset, | |
2776 | page, pgoff + offset, start + offset, | |
c0111c44 | 2777 | failed_mirror, btrfs_submit_data_bio); |
150e4b05 QW |
2778 | if (!ret) { |
2779 | /* | |
2780 | * We have submitted the read repair, the page release | |
2781 | * will be handled by the endio function of the | |
2782 | * submitted repair bio. | |
2783 | * Thus we don't need to do any thing here. | |
2784 | */ | |
2785 | continue; | |
2786 | } | |
2787 | /* | |
2788 | * Repair failed, just record the error but still continue. | |
2789 | * Or the remaining sectors will not be properly unlocked. | |
2790 | */ | |
2791 | if (!error) | |
2792 | error = ret; | |
2793 | next: | |
2794 | end_page_read(page, uptodate, start + offset, sectorsize); | |
2795 | if (uptodate) | |
2796 | set_extent_uptodate(&BTRFS_I(inode)->io_tree, | |
2797 | start + offset, | |
2798 | start + offset + sectorsize - 1, | |
2799 | &cached, GFP_ATOMIC); | |
2800 | unlock_extent_cached_atomic(&BTRFS_I(inode)->io_tree, | |
2801 | start + offset, | |
2802 | start + offset + sectorsize - 1, | |
2803 | &cached); | |
2804 | } | |
2805 | return errno_to_blk_status(error); | |
4a54c8c1 JS |
2806 | } |
2807 | ||
d1310b2e CM |
2808 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2809 | ||
b5227c07 | 2810 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 2811 | { |
38a39ac7 | 2812 | struct btrfs_inode *inode; |
25c1252a | 2813 | const bool uptodate = (err == 0); |
3e2426bd | 2814 | int ret = 0; |
87826df0 | 2815 | |
38a39ac7 QW |
2816 | ASSERT(page && page->mapping); |
2817 | inode = BTRFS_I(page->mapping->host); | |
2818 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 2819 | |
87826df0 | 2820 | if (!uptodate) { |
963e4db8 QW |
2821 | const struct btrfs_fs_info *fs_info = inode->root->fs_info; |
2822 | u32 len; | |
2823 | ||
2824 | ASSERT(end + 1 - start <= U32_MAX); | |
2825 | len = end + 1 - start; | |
2826 | ||
2827 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2828 | btrfs_page_set_error(fs_info, page, start, len); | |
bff5baf8 | 2829 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2830 | mapping_set_error(page->mapping, ret); |
87826df0 | 2831 | } |
87826df0 JM |
2832 | } |
2833 | ||
d1310b2e CM |
2834 | /* |
2835 | * after a writepage IO is done, we need to: | |
2836 | * clear the uptodate bits on error | |
2837 | * clear the writeback bits in the extent tree for this IO | |
2838 | * end_page_writeback if the page has no more pending IO | |
2839 | * | |
2840 | * Scheduling is not allowed, so the extent state tree is expected | |
2841 | * to have one and only one object corresponding to this IO. | |
2842 | */ | |
4246a0b6 | 2843 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2844 | { |
4e4cbee9 | 2845 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2846 | struct bio_vec *bvec; |
d1310b2e CM |
2847 | u64 start; |
2848 | u64 end; | |
6dc4f100 | 2849 | struct bvec_iter_all iter_all; |
d8e3fb10 | 2850 | bool first_bvec = true; |
d1310b2e | 2851 | |
c09abff8 | 2852 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2853 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2854 | struct page *page = bvec->bv_page; |
0b246afa JM |
2855 | struct inode *inode = page->mapping->host; |
2856 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
2857 | const u32 sectorsize = fs_info->sectorsize; |
2858 | ||
2859 | /* Our read/write should always be sector aligned. */ | |
2860 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2861 | btrfs_err(fs_info, | |
2862 | "partial page write in btrfs with offset %u and length %u", | |
2863 | bvec->bv_offset, bvec->bv_len); | |
2864 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
2865 | btrfs_info(fs_info, | |
2866 | "incomplete page write with offset %u and length %u", | |
2867 | bvec->bv_offset, bvec->bv_len); | |
2868 | ||
2869 | start = page_offset(page) + bvec->bv_offset; | |
2870 | end = start + bvec->bv_len - 1; | |
d1310b2e | 2871 | |
d8e3fb10 NA |
2872 | if (first_bvec) { |
2873 | btrfs_record_physical_zoned(inode, start, bio); | |
2874 | first_bvec = false; | |
2875 | } | |
2876 | ||
4e4cbee9 | 2877 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
2878 | |
2879 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 2880 | } |
2b1f55b0 | 2881 | |
d1310b2e | 2882 | bio_put(bio); |
d1310b2e CM |
2883 | } |
2884 | ||
94e8c95c QW |
2885 | /* |
2886 | * Record previously processed extent range | |
2887 | * | |
2888 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
2889 | * the extent io operations. | |
2890 | */ | |
2891 | struct processed_extent { | |
2892 | struct btrfs_inode *inode; | |
2893 | /* Start of the range in @inode */ | |
2894 | u64 start; | |
2e626e56 | 2895 | /* End of the range in @inode */ |
94e8c95c QW |
2896 | u64 end; |
2897 | bool uptodate; | |
2898 | }; | |
2899 | ||
2900 | /* | |
2901 | * Try to release processed extent range | |
2902 | * | |
2903 | * May not release the extent range right now if the current range is | |
2904 | * contiguous to processed extent. | |
2905 | * | |
2906 | * Will release processed extent when any of @inode, @uptodate, the range is | |
2907 | * no longer contiguous to the processed range. | |
2908 | * | |
2909 | * Passing @inode == NULL will force processed extent to be released. | |
2910 | */ | |
2911 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
2912 | struct btrfs_inode *inode, u64 start, u64 end, | |
2913 | bool uptodate) | |
883d0de4 MX |
2914 | { |
2915 | struct extent_state *cached = NULL; | |
94e8c95c QW |
2916 | struct extent_io_tree *tree; |
2917 | ||
2918 | /* The first extent, initialize @processed */ | |
2919 | if (!processed->inode) | |
2920 | goto update; | |
883d0de4 | 2921 | |
94e8c95c QW |
2922 | /* |
2923 | * Contiguous to processed extent, just uptodate the end. | |
2924 | * | |
2925 | * Several things to notice: | |
2926 | * | |
2927 | * - bio can be merged as long as on-disk bytenr is contiguous | |
2928 | * This means we can have page belonging to other inodes, thus need to | |
2929 | * check if the inode still matches. | |
2930 | * - bvec can contain range beyond current page for multi-page bvec | |
2931 | * Thus we need to do processed->end + 1 >= start check | |
2932 | */ | |
2933 | if (processed->inode == inode && processed->uptodate == uptodate && | |
2934 | processed->end + 1 >= start && end >= processed->end) { | |
2935 | processed->end = end; | |
2936 | return; | |
2937 | } | |
2938 | ||
2939 | tree = &processed->inode->io_tree; | |
2940 | /* | |
2941 | * Now we don't have range contiguous to the processed range, release | |
2942 | * the processed range now. | |
2943 | */ | |
2944 | if (processed->uptodate && tree->track_uptodate) | |
2945 | set_extent_uptodate(tree, processed->start, processed->end, | |
2946 | &cached, GFP_ATOMIC); | |
2947 | unlock_extent_cached_atomic(tree, processed->start, processed->end, | |
2948 | &cached); | |
2949 | ||
2950 | update: | |
2951 | /* Update processed to current range */ | |
2952 | processed->inode = inode; | |
2953 | processed->start = start; | |
2954 | processed->end = end; | |
2955 | processed->uptodate = uptodate; | |
883d0de4 MX |
2956 | } |
2957 | ||
92082d40 QW |
2958 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
2959 | { | |
2960 | ASSERT(PageLocked(page)); | |
fbca46eb | 2961 | if (!btrfs_is_subpage(fs_info, page)) |
92082d40 QW |
2962 | return; |
2963 | ||
2964 | ASSERT(PagePrivate(page)); | |
2965 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
2966 | } | |
2967 | ||
d9bb77d5 | 2968 | /* |
8ee92268 | 2969 | * Find extent buffer for a given bytenr. |
d9bb77d5 QW |
2970 | * |
2971 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
2972 | * in endio context. | |
2973 | */ | |
2974 | static struct extent_buffer *find_extent_buffer_readpage( | |
2975 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
2976 | { | |
2977 | struct extent_buffer *eb; | |
2978 | ||
2979 | /* | |
2980 | * For regular sectorsize, we can use page->private to grab extent | |
2981 | * buffer | |
2982 | */ | |
fbca46eb | 2983 | if (fs_info->nodesize >= PAGE_SIZE) { |
d9bb77d5 QW |
2984 | ASSERT(PagePrivate(page) && page->private); |
2985 | return (struct extent_buffer *)page->private; | |
2986 | } | |
2987 | ||
8ee92268 GN |
2988 | /* For subpage case, we need to lookup extent buffer xarray */ |
2989 | eb = xa_load(&fs_info->extent_buffers, | |
2990 | bytenr >> fs_info->sectorsize_bits); | |
d9bb77d5 QW |
2991 | ASSERT(eb); |
2992 | return eb; | |
2993 | } | |
2994 | ||
d1310b2e CM |
2995 | /* |
2996 | * after a readpage IO is done, we need to: | |
2997 | * clear the uptodate bits on error | |
2998 | * set the uptodate bits if things worked | |
2999 | * set the page up to date if all extents in the tree are uptodate | |
3000 | * clear the lock bit in the extent tree | |
3001 | * unlock the page if there are no other extents locked for it | |
3002 | * | |
3003 | * Scheduling is not allowed, so the extent state tree is expected | |
3004 | * to have one and only one object corresponding to this IO. | |
3005 | */ | |
4246a0b6 | 3006 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 3007 | { |
2c30c71b | 3008 | struct bio_vec *bvec; |
c3a3b19b | 3009 | struct btrfs_bio *bbio = btrfs_bio(bio); |
7870d082 | 3010 | struct extent_io_tree *tree, *failure_tree; |
94e8c95c | 3011 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
3012 | /* |
3013 | * The offset to the beginning of a bio, since one bio can never be | |
3014 | * larger than UINT_MAX, u32 here is enough. | |
3015 | */ | |
3016 | u32 bio_offset = 0; | |
5cf1ab56 | 3017 | int mirror; |
d1310b2e | 3018 | int ret; |
6dc4f100 | 3019 | struct bvec_iter_all iter_all; |
d1310b2e | 3020 | |
c09abff8 | 3021 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 3022 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 3023 | bool uptodate = !bio->bi_status; |
d1310b2e | 3024 | struct page *page = bvec->bv_page; |
a71754fc | 3025 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 3026 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 | 3027 | const u32 sectorsize = fs_info->sectorsize; |
150e4b05 | 3028 | unsigned int error_bitmap = (unsigned int)-1; |
7ffd27e3 QW |
3029 | u64 start; |
3030 | u64 end; | |
3031 | u32 len; | |
507903b8 | 3032 | |
ab8d0fc4 JM |
3033 | btrfs_debug(fs_info, |
3034 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 3035 | bio->bi_iter.bi_sector, bio->bi_status, |
c3a3b19b | 3036 | bbio->mirror_num); |
a71754fc | 3037 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 3038 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 3039 | |
8b8bbd46 QW |
3040 | /* |
3041 | * We always issue full-sector reads, but if some block in a | |
3042 | * page fails to read, blk_update_request() will advance | |
3043 | * bv_offset and adjust bv_len to compensate. Print a warning | |
3044 | * for unaligned offsets, and an error if they don't add up to | |
3045 | * a full sector. | |
3046 | */ | |
3047 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
3048 | btrfs_err(fs_info, | |
3049 | "partial page read in btrfs with offset %u and length %u", | |
3050 | bvec->bv_offset, bvec->bv_len); | |
3051 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
3052 | sectorsize)) | |
3053 | btrfs_info(fs_info, | |
3054 | "incomplete page read with offset %u and length %u", | |
3055 | bvec->bv_offset, bvec->bv_len); | |
3056 | ||
3057 | start = page_offset(page) + bvec->bv_offset; | |
3058 | end = start + bvec->bv_len - 1; | |
facc8a22 | 3059 | len = bvec->bv_len; |
d1310b2e | 3060 | |
c3a3b19b | 3061 | mirror = bbio->mirror_num; |
78e62c02 | 3062 | if (likely(uptodate)) { |
150e4b05 | 3063 | if (is_data_inode(inode)) { |
c3a3b19b | 3064 | error_bitmap = btrfs_verify_data_csum(bbio, |
5e295768 | 3065 | bio_offset, page, start, end); |
150e4b05 QW |
3066 | ret = error_bitmap; |
3067 | } else { | |
c3a3b19b | 3068 | ret = btrfs_validate_metadata_buffer(bbio, |
8e1dc982 | 3069 | page, start, end, mirror); |
150e4b05 | 3070 | } |
5ee0844d | 3071 | if (ret) |
150e4b05 | 3072 | uptodate = false; |
5ee0844d | 3073 | else |
7870d082 JB |
3074 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
3075 | failure_tree, tree, start, | |
3076 | page, | |
3077 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 3078 | } |
ea466794 | 3079 | |
f2a09da9 MX |
3080 | if (likely(uptodate)) |
3081 | goto readpage_ok; | |
3082 | ||
be17b3af | 3083 | if (is_data_inode(inode)) { |
510671d2 JB |
3084 | /* |
3085 | * If we failed to submit the IO at all we'll have a | |
3086 | * mirror_num == 0, in which case we need to just mark | |
3087 | * the page with an error and unlock it and carry on. | |
3088 | */ | |
3089 | if (mirror == 0) | |
3090 | goto readpage_ok; | |
3091 | ||
f4a8e656 | 3092 | /* |
c0111c44 | 3093 | * submit_data_read_repair() will handle all the good |
150e4b05 | 3094 | * and bad sectors, we just continue to the next bvec. |
f4a8e656 | 3095 | */ |
c0111c44 QW |
3096 | submit_data_read_repair(inode, bio, bio_offset, page, |
3097 | start - page_offset(page), | |
3098 | start, end, mirror, | |
3099 | error_bitmap); | |
150e4b05 QW |
3100 | |
3101 | ASSERT(bio_offset + len > bio_offset); | |
3102 | bio_offset += len; | |
3103 | continue; | |
78e62c02 NB |
3104 | } else { |
3105 | struct extent_buffer *eb; | |
3106 | ||
d9bb77d5 | 3107 | eb = find_extent_buffer_readpage(fs_info, page, start); |
78e62c02 NB |
3108 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
3109 | eb->read_mirror = mirror; | |
3110 | atomic_dec(&eb->io_pages); | |
7e38326f | 3111 | } |
f2a09da9 | 3112 | readpage_ok: |
883d0de4 | 3113 | if (likely(uptodate)) { |
a71754fc | 3114 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3115 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 3116 | |
c28ea613 QW |
3117 | /* |
3118 | * Zero out the remaining part if this range straddles | |
3119 | * i_size. | |
3120 | * | |
3121 | * Here we should only zero the range inside the bvec, | |
3122 | * not touch anything else. | |
3123 | * | |
3124 | * NOTE: i_size is exclusive while end is inclusive. | |
3125 | */ | |
3126 | if (page->index == end_index && i_size <= end) { | |
3127 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 3128 | offset_in_page(start)); |
c28ea613 QW |
3129 | |
3130 | zero_user_segment(page, zero_start, | |
3131 | offset_in_page(end) + 1); | |
3132 | } | |
70dec807 | 3133 | } |
7ffd27e3 QW |
3134 | ASSERT(bio_offset + len > bio_offset); |
3135 | bio_offset += len; | |
883d0de4 | 3136 | |
e09caaf9 | 3137 | /* Update page status and unlock */ |
92082d40 | 3138 | end_page_read(page, uptodate, start, len); |
94e8c95c | 3139 | endio_readpage_release_extent(&processed, BTRFS_I(inode), |
14605409 | 3140 | start, end, PageUptodate(page)); |
2c30c71b | 3141 | } |
94e8c95c QW |
3142 | /* Release the last extent */ |
3143 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
c3a3b19b | 3144 | btrfs_bio_free_csum(bbio); |
d1310b2e | 3145 | bio_put(bio); |
d1310b2e CM |
3146 | } |
3147 | ||
dd137dd1 STD |
3148 | /** |
3149 | * Populate every free slot in a provided array with pages. | |
3150 | * | |
3151 | * @nr_pages: number of pages to allocate | |
3152 | * @page_array: the array to fill with pages; any existing non-null entries in | |
3153 | * the array will be skipped | |
3154 | * | |
3155 | * Return: 0 if all pages were able to be allocated; | |
3156 | * -ENOMEM otherwise, and the caller is responsible for freeing all | |
3157 | * non-null page pointers in the array. | |
3158 | */ | |
3159 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array) | |
3160 | { | |
91d6ac1d | 3161 | unsigned int allocated; |
dd137dd1 | 3162 | |
91d6ac1d STD |
3163 | for (allocated = 0; allocated < nr_pages;) { |
3164 | unsigned int last = allocated; | |
dd137dd1 | 3165 | |
91d6ac1d STD |
3166 | allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array); |
3167 | ||
395cb57e STD |
3168 | if (allocated == nr_pages) |
3169 | return 0; | |
3170 | ||
91d6ac1d STD |
3171 | /* |
3172 | * During this iteration, no page could be allocated, even | |
3173 | * though alloc_pages_bulk_array() falls back to alloc_page() | |
3174 | * if it could not bulk-allocate. So we must be out of memory. | |
3175 | */ | |
3176 | if (allocated == last) | |
dd137dd1 | 3177 | return -ENOMEM; |
395cb57e STD |
3178 | |
3179 | memalloc_retry_wait(GFP_NOFS); | |
dd137dd1 STD |
3180 | } |
3181 | return 0; | |
3182 | } | |
3183 | ||
9be3395b | 3184 | /* |
184f999e DS |
3185 | * Initialize the members up to but not including 'bio'. Use after allocating a |
3186 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
3187 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 3188 | */ |
c3a3b19b | 3189 | static inline void btrfs_bio_init(struct btrfs_bio *bbio) |
d1310b2e | 3190 | { |
c3a3b19b | 3191 | memset(bbio, 0, offsetof(struct btrfs_bio, bio)); |
184f999e | 3192 | } |
d1310b2e | 3193 | |
9be3395b | 3194 | /* |
cd8e0cca QW |
3195 | * Allocate a btrfs_io_bio, with @nr_iovecs as maximum number of iovecs. |
3196 | * | |
3197 | * The bio allocation is backed by bioset and does not fail. | |
9be3395b | 3198 | */ |
c3a3b19b | 3199 | struct bio *btrfs_bio_alloc(unsigned int nr_iovecs) |
d1310b2e CM |
3200 | { |
3201 | struct bio *bio; | |
d1310b2e | 3202 | |
cd8e0cca | 3203 | ASSERT(0 < nr_iovecs && nr_iovecs <= BIO_MAX_VECS); |
609be106 | 3204 | bio = bio_alloc_bioset(NULL, nr_iovecs, 0, GFP_NOFS, &btrfs_bioset); |
c3a3b19b | 3205 | btrfs_bio_init(btrfs_bio(bio)); |
d1310b2e CM |
3206 | return bio; |
3207 | } | |
3208 | ||
110ac0e5 | 3209 | struct bio *btrfs_bio_clone(struct block_device *bdev, struct bio *bio) |
9be3395b | 3210 | { |
c3a3b19b | 3211 | struct btrfs_bio *bbio; |
23ea8e5a | 3212 | struct bio *new; |
9be3395b | 3213 | |
6e707bcd | 3214 | /* Bio allocation backed by a bioset does not fail */ |
110ac0e5 | 3215 | new = bio_alloc_clone(bdev, bio, GFP_NOFS, &btrfs_bioset); |
c3a3b19b QW |
3216 | bbio = btrfs_bio(new); |
3217 | btrfs_bio_init(bbio); | |
3218 | bbio->iter = bio->bi_iter; | |
23ea8e5a MX |
3219 | return new; |
3220 | } | |
9be3395b | 3221 | |
21dda654 | 3222 | struct bio *btrfs_bio_clone_partial(struct bio *orig, u64 offset, u64 size) |
2f8e9140 LB |
3223 | { |
3224 | struct bio *bio; | |
c3a3b19b | 3225 | struct btrfs_bio *bbio; |
2f8e9140 | 3226 | |
21dda654 CK |
3227 | ASSERT(offset <= UINT_MAX && size <= UINT_MAX); |
3228 | ||
2f8e9140 | 3229 | /* this will never fail when it's backed by a bioset */ |
abfc426d | 3230 | bio = bio_alloc_clone(orig->bi_bdev, orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3231 | ASSERT(bio); |
3232 | ||
c3a3b19b QW |
3233 | bbio = btrfs_bio(bio); |
3234 | btrfs_bio_init(bbio); | |
2f8e9140 LB |
3235 | |
3236 | bio_trim(bio, offset >> 9, size >> 9); | |
c3a3b19b | 3237 | bbio->iter = bio->bi_iter; |
2f8e9140 LB |
3238 | return bio; |
3239 | } | |
9be3395b | 3240 | |
953651eb NA |
3241 | /** |
3242 | * Attempt to add a page to bio | |
3243 | * | |
be8d1a2a | 3244 | * @bio_ctrl: record both the bio, and its bio_flags |
953651eb NA |
3245 | * @page: page to add to the bio |
3246 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
3247 | * a contiguous page to the previous one | |
953651eb | 3248 | * @size: portion of page that we want to write |
be8d1a2a | 3249 | * @pg_offset: starting offset in the page |
cb3a12d9 | 3250 | * @compress_type: compression type of the current bio to see if we can merge them |
953651eb NA |
3251 | * |
3252 | * Attempt to add a page to bio considering stripe alignment etc. | |
3253 | * | |
e0eefe07 QW |
3254 | * Return >= 0 for the number of bytes added to the bio. |
3255 | * Can return 0 if the current bio is already at stripe/zone boundary. | |
3256 | * Return <0 for error. | |
953651eb | 3257 | */ |
e0eefe07 QW |
3258 | static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
3259 | struct page *page, | |
3260 | u64 disk_bytenr, unsigned int size, | |
3261 | unsigned int pg_offset, | |
cb3a12d9 | 3262 | enum btrfs_compression_type compress_type) |
953651eb | 3263 | { |
390ed29b QW |
3264 | struct bio *bio = bio_ctrl->bio; |
3265 | u32 bio_size = bio->bi_iter.bi_size; | |
e0eefe07 | 3266 | u32 real_size; |
953651eb NA |
3267 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
3268 | bool contig; | |
e1326f03 | 3269 | int ret; |
953651eb | 3270 | |
390ed29b QW |
3271 | ASSERT(bio); |
3272 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
3273 | ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary); | |
0f07003b | 3274 | if (bio_ctrl->compress_type != compress_type) |
e0eefe07 | 3275 | return 0; |
953651eb | 3276 | |
0f07003b | 3277 | if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) |
953651eb NA |
3278 | contig = bio->bi_iter.bi_sector == sector; |
3279 | else | |
3280 | contig = bio_end_sector(bio) == sector; | |
3281 | if (!contig) | |
e0eefe07 | 3282 | return 0; |
953651eb | 3283 | |
e0eefe07 QW |
3284 | real_size = min(bio_ctrl->len_to_oe_boundary, |
3285 | bio_ctrl->len_to_stripe_boundary) - bio_size; | |
3286 | real_size = min(real_size, size); | |
3287 | ||
3288 | /* | |
3289 | * If real_size is 0, never call bio_add_*_page(), as even size is 0, | |
3290 | * bio will still execute its endio function on the page! | |
3291 | */ | |
3292 | if (real_size == 0) | |
3293 | return 0; | |
953651eb | 3294 | |
390ed29b | 3295 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
e0eefe07 | 3296 | ret = bio_add_zone_append_page(bio, page, real_size, pg_offset); |
390ed29b | 3297 | else |
e0eefe07 | 3298 | ret = bio_add_page(bio, page, real_size, pg_offset); |
e1326f03 | 3299 | |
e0eefe07 | 3300 | return ret; |
953651eb NA |
3301 | } |
3302 | ||
390ed29b | 3303 | static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
939c7feb | 3304 | struct btrfs_inode *inode, u64 file_offset) |
390ed29b QW |
3305 | { |
3306 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3307 | struct btrfs_io_geometry geom; | |
3308 | struct btrfs_ordered_extent *ordered; | |
3309 | struct extent_map *em; | |
3310 | u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT); | |
3311 | int ret; | |
3312 | ||
3313 | /* | |
3314 | * Pages for compressed extent are never submitted to disk directly, | |
3315 | * thus it has no real boundary, just set them to U32_MAX. | |
3316 | * | |
3317 | * The split happens for real compressed bio, which happens in | |
3318 | * btrfs_submit_compressed_read/write(). | |
3319 | */ | |
0f07003b | 3320 | if (bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) { |
390ed29b QW |
3321 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3322 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3323 | return 0; | |
3324 | } | |
3325 | em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize); | |
3326 | if (IS_ERR(em)) | |
3327 | return PTR_ERR(em); | |
3328 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio), | |
3329 | logical, &geom); | |
3330 | free_extent_map(em); | |
3331 | if (ret < 0) { | |
3332 | return ret; | |
3333 | } | |
3334 | if (geom.len > U32_MAX) | |
3335 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3336 | else | |
3337 | bio_ctrl->len_to_stripe_boundary = (u32)geom.len; | |
3338 | ||
73672710 | 3339 | if (bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) { |
390ed29b QW |
3340 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
3341 | return 0; | |
3342 | } | |
3343 | ||
390ed29b | 3344 | /* Ordered extent not yet created, so we're good */ |
939c7feb | 3345 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
390ed29b QW |
3346 | if (!ordered) { |
3347 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3348 | return 0; | |
3349 | } | |
3350 | ||
3351 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
3352 | ordered->disk_bytenr + ordered->disk_num_bytes - logical); | |
3353 | btrfs_put_ordered_extent(ordered); | |
3354 | return 0; | |
3355 | } | |
3356 | ||
e0eefe07 QW |
3357 | static int alloc_new_bio(struct btrfs_inode *inode, |
3358 | struct btrfs_bio_ctrl *bio_ctrl, | |
3359 | struct writeback_control *wbc, | |
3360 | unsigned int opf, | |
3361 | bio_end_io_t end_io_func, | |
939c7feb | 3362 | u64 disk_bytenr, u32 offset, u64 file_offset, |
cb3a12d9 | 3363 | enum btrfs_compression_type compress_type) |
e0eefe07 QW |
3364 | { |
3365 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3366 | struct bio *bio; | |
3367 | int ret; | |
3368 | ||
c3a3b19b | 3369 | bio = btrfs_bio_alloc(BIO_MAX_VECS); |
e0eefe07 QW |
3370 | /* |
3371 | * For compressed page range, its disk_bytenr is always @disk_bytenr | |
3372 | * passed in, no matter if we have added any range into previous bio. | |
3373 | */ | |
cb3a12d9 | 3374 | if (compress_type != BTRFS_COMPRESS_NONE) |
cd8e0cca | 3375 | bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
e0eefe07 | 3376 | else |
cd8e0cca | 3377 | bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT; |
e0eefe07 | 3378 | bio_ctrl->bio = bio; |
0f07003b | 3379 | bio_ctrl->compress_type = compress_type; |
e0eefe07 QW |
3380 | bio->bi_end_io = end_io_func; |
3381 | bio->bi_private = &inode->io_tree; | |
e0eefe07 | 3382 | bio->bi_opf = opf; |
939c7feb NA |
3383 | ret = calc_bio_boundaries(bio_ctrl, inode, file_offset); |
3384 | if (ret < 0) | |
3385 | goto error; | |
e0eefe07 | 3386 | |
50f1cff3 CH |
3387 | if (wbc) { |
3388 | /* | |
3389 | * For Zone append we need the correct block_device that we are | |
3390 | * going to write to set in the bio to be able to respect the | |
3391 | * hardware limitation. Look it up here: | |
3392 | */ | |
3393 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) { | |
3394 | struct btrfs_device *dev; | |
3395 | ||
3396 | dev = btrfs_zoned_get_device(fs_info, disk_bytenr, | |
3397 | fs_info->sectorsize); | |
3398 | if (IS_ERR(dev)) { | |
3399 | ret = PTR_ERR(dev); | |
3400 | goto error; | |
3401 | } | |
e0eefe07 | 3402 | |
50f1cff3 CH |
3403 | bio_set_dev(bio, dev->bdev); |
3404 | } else { | |
3405 | /* | |
3406 | * Otherwise pick the last added device to support | |
3407 | * cgroup writeback. For multi-device file systems this | |
3408 | * means blk-cgroup policies have to always be set on the | |
3409 | * last added/replaced device. This is a bit odd but has | |
3410 | * been like that for a long time. | |
3411 | */ | |
3412 | bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev); | |
e0eefe07 | 3413 | } |
50f1cff3 CH |
3414 | wbc_init_bio(wbc, bio); |
3415 | } else { | |
3416 | ASSERT(bio_op(bio) != REQ_OP_ZONE_APPEND); | |
e0eefe07 QW |
3417 | } |
3418 | return 0; | |
3419 | error: | |
3420 | bio_ctrl->bio = NULL; | |
3421 | bio->bi_status = errno_to_blk_status(ret); | |
3422 | bio_endio(bio); | |
3423 | return ret; | |
3424 | } | |
3425 | ||
4b81ba48 DS |
3426 | /* |
3427 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3428 | * @wbc: optional writeback control for io accounting |
3429 | * @page: page to add to the bio | |
0c64c33c QW |
3430 | * @disk_bytenr: logical bytenr where the write will be |
3431 | * @size: portion of page that we want to write to | |
b8b3d625 DS |
3432 | * @pg_offset: offset of the new bio or to check whether we are adding |
3433 | * a contiguous page to the previous one | |
5c2b1fd7 | 3434 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3435 | * @end_io_func: end_io callback for new bio |
3436 | * @mirror_num: desired mirror to read/write | |
3437 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
cb3a12d9 | 3438 | * @compress_type: compress type for current bio |
4b81ba48 | 3439 | */ |
0ceb34bf | 3440 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3441 | struct writeback_control *wbc, |
390ed29b | 3442 | struct btrfs_bio_ctrl *bio_ctrl, |
0c64c33c | 3443 | struct page *page, u64 disk_bytenr, |
6c5a4e2c | 3444 | size_t size, unsigned long pg_offset, |
f188591e | 3445 | bio_end_io_t end_io_func, |
c8b97818 | 3446 | int mirror_num, |
cb3a12d9 | 3447 | enum btrfs_compression_type compress_type, |
005efedf | 3448 | bool force_bio_submit) |
d1310b2e CM |
3449 | { |
3450 | int ret = 0; | |
e1326f03 | 3451 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
e0eefe07 | 3452 | unsigned int cur = pg_offset; |
d1310b2e | 3453 | |
390ed29b | 3454 | ASSERT(bio_ctrl); |
5c2b1fd7 | 3455 | |
390ed29b QW |
3456 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
3457 | pg_offset + size <= PAGE_SIZE); | |
e0eefe07 | 3458 | if (force_bio_submit && bio_ctrl->bio) { |
0f07003b | 3459 | submit_one_bio(bio_ctrl->bio, mirror_num, bio_ctrl->compress_type); |
e0eefe07 | 3460 | bio_ctrl->bio = NULL; |
e0eefe07 QW |
3461 | } |
3462 | ||
3463 | while (cur < pg_offset + size) { | |
3464 | u32 offset = cur - pg_offset; | |
3465 | int added; | |
3466 | ||
3467 | /* Allocate new bio if needed */ | |
3468 | if (!bio_ctrl->bio) { | |
3469 | ret = alloc_new_bio(inode, bio_ctrl, wbc, opf, | |
3470 | end_io_func, disk_bytenr, offset, | |
939c7feb | 3471 | page_offset(page) + cur, |
cb3a12d9 | 3472 | compress_type); |
e0eefe07 QW |
3473 | if (ret < 0) |
3474 | return ret; | |
3475 | } | |
3476 | /* | |
3477 | * We must go through btrfs_bio_add_page() to ensure each | |
3478 | * page range won't cross various boundaries. | |
3479 | */ | |
cb3a12d9 | 3480 | if (compress_type != BTRFS_COMPRESS_NONE) |
e0eefe07 QW |
3481 | added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, |
3482 | size - offset, pg_offset + offset, | |
cb3a12d9 | 3483 | compress_type); |
e0eefe07 QW |
3484 | else |
3485 | added = btrfs_bio_add_page(bio_ctrl, page, | |
3486 | disk_bytenr + offset, size - offset, | |
cb3a12d9 | 3487 | pg_offset + offset, compress_type); |
e0eefe07 QW |
3488 | |
3489 | /* Metadata page range should never be split */ | |
3490 | if (!is_data_inode(&inode->vfs_inode)) | |
3491 | ASSERT(added == 0 || added == size - offset); | |
3492 | ||
3493 | /* At least we added some page, update the account */ | |
3494 | if (wbc && added) | |
3495 | wbc_account_cgroup_owner(wbc, page, added); | |
3496 | ||
3497 | /* We have reached boundary, submit right now */ | |
3498 | if (added < size - offset) { | |
3499 | /* The bio should contain some page(s) */ | |
3500 | ASSERT(bio_ctrl->bio->bi_iter.bi_size); | |
0f07003b | 3501 | submit_one_bio(bio_ctrl->bio, mirror_num, bio_ctrl->compress_type); |
390ed29b | 3502 | bio_ctrl->bio = NULL; |
d1310b2e | 3503 | } |
e0eefe07 | 3504 | cur += added; |
d1310b2e | 3505 | } |
e0eefe07 | 3506 | return 0; |
d1310b2e CM |
3507 | } |
3508 | ||
760f991f QW |
3509 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
3510 | struct page *page, | |
3511 | struct btrfs_subpage *prealloc) | |
d1310b2e | 3512 | { |
760f991f QW |
3513 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3514 | int ret = 0; | |
3515 | ||
0d01e247 QW |
3516 | /* |
3517 | * If the page is mapped to btree inode, we should hold the private | |
3518 | * lock to prevent race. | |
3519 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3520 | * will not race with any other ebs. | |
3521 | */ | |
3522 | if (page->mapping) | |
3523 | lockdep_assert_held(&page->mapping->private_lock); | |
3524 | ||
fbca46eb | 3525 | if (fs_info->nodesize >= PAGE_SIZE) { |
760f991f QW |
3526 | if (!PagePrivate(page)) |
3527 | attach_page_private(page, eb); | |
3528 | else | |
3529 | WARN_ON(page->private != (unsigned long)eb); | |
3530 | return 0; | |
3531 | } | |
3532 | ||
3533 | /* Already mapped, just free prealloc */ | |
3534 | if (PagePrivate(page)) { | |
3535 | btrfs_free_subpage(prealloc); | |
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | if (prealloc) | |
3540 | /* Has preallocated memory for subpage */ | |
3541 | attach_page_private(page, prealloc); | |
d1b89bc0 | 3542 | else |
760f991f QW |
3543 | /* Do new allocation to attach subpage */ |
3544 | ret = btrfs_attach_subpage(fs_info, page, | |
3545 | BTRFS_SUBPAGE_METADATA); | |
3546 | return ret; | |
d1310b2e CM |
3547 | } |
3548 | ||
32443de3 | 3549 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 3550 | { |
32443de3 QW |
3551 | struct btrfs_fs_info *fs_info; |
3552 | ||
3553 | ASSERT(page->mapping); | |
3554 | ||
3555 | if (PagePrivate(page)) | |
3556 | return 0; | |
3557 | ||
3558 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3559 | ||
fbca46eb | 3560 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3561 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); |
3562 | ||
3563 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
3564 | return 0; | |
3565 | } | |
3566 | ||
3567 | void clear_page_extent_mapped(struct page *page) | |
3568 | { | |
3569 | struct btrfs_fs_info *fs_info; | |
3570 | ||
3571 | ASSERT(page->mapping); | |
3572 | ||
d1b89bc0 | 3573 | if (!PagePrivate(page)) |
32443de3 QW |
3574 | return; |
3575 | ||
3576 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
fbca46eb | 3577 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
3578 | return btrfs_detach_subpage(fs_info, page); |
3579 | ||
3580 | detach_page_private(page); | |
d1310b2e CM |
3581 | } |
3582 | ||
125bac01 MX |
3583 | static struct extent_map * |
3584 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3585 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3586 | { |
3587 | struct extent_map *em; | |
3588 | ||
3589 | if (em_cached && *em_cached) { | |
3590 | em = *em_cached; | |
cbc0e928 | 3591 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3592 | start < extent_map_end(em)) { |
490b54d6 | 3593 | refcount_inc(&em->refs); |
125bac01 MX |
3594 | return em; |
3595 | } | |
3596 | ||
3597 | free_extent_map(em); | |
3598 | *em_cached = NULL; | |
3599 | } | |
3600 | ||
1a5ee1e6 | 3601 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
c0347550 | 3602 | if (em_cached && !IS_ERR(em)) { |
125bac01 | 3603 | BUG_ON(*em_cached); |
490b54d6 | 3604 | refcount_inc(&em->refs); |
125bac01 MX |
3605 | *em_cached = em; |
3606 | } | |
3607 | return em; | |
3608 | } | |
d1310b2e CM |
3609 | /* |
3610 | * basic readpage implementation. Locked extent state structs are inserted | |
3611 | * into the tree that are removed when the IO is done (by the end_io | |
3612 | * handlers) | |
79787eaa | 3613 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3614 | * return 0 on success, otherwise return error |
d1310b2e | 3615 | */ |
7aab8b32 | 3616 | static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
390ed29b | 3617 | struct btrfs_bio_ctrl *bio_ctrl, |
0f208812 | 3618 | unsigned int read_flags, u64 *prev_em_start) |
d1310b2e CM |
3619 | { |
3620 | struct inode *inode = page->mapping->host; | |
92082d40 | 3621 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 3622 | u64 start = page_offset(page); |
8eec8296 | 3623 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3624 | u64 cur = start; |
3625 | u64 extent_offset; | |
3626 | u64 last_byte = i_size_read(inode); | |
3627 | u64 block_start; | |
3628 | u64 cur_end; | |
d1310b2e | 3629 | struct extent_map *em; |
baf863b9 | 3630 | int ret = 0; |
306e16ce | 3631 | size_t pg_offset = 0; |
d1310b2e CM |
3632 | size_t iosize; |
3633 | size_t blocksize = inode->i_sb->s_blocksize; | |
f657a31c | 3634 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3635 | |
32443de3 QW |
3636 | ret = set_page_extent_mapped(page); |
3637 | if (ret < 0) { | |
3638 | unlock_extent(tree, start, end); | |
92082d40 QW |
3639 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
3640 | unlock_page(page); | |
32443de3 QW |
3641 | goto out; |
3642 | } | |
d1310b2e | 3643 | |
09cbfeaf | 3644 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 3645 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3646 | |
3647 | if (zero_offset) { | |
09cbfeaf | 3648 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 3649 | memzero_page(page, zero_offset, iosize); |
c8b97818 | 3650 | flush_dcache_page(page); |
c8b97818 CM |
3651 | } |
3652 | } | |
92082d40 | 3653 | begin_page_read(fs_info, page); |
d1310b2e | 3654 | while (cur <= end) { |
4c37a793 | 3655 | unsigned long this_bio_flag = 0; |
005efedf | 3656 | bool force_bio_submit = false; |
0c64c33c | 3657 | u64 disk_bytenr; |
c8f2f24b | 3658 | |
6a404910 | 3659 | ASSERT(IS_ALIGNED(cur, fs_info->sectorsize)); |
d1310b2e | 3660 | if (cur >= last_byte) { |
507903b8 AJ |
3661 | struct extent_state *cached = NULL; |
3662 | ||
09cbfeaf | 3663 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 3664 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3665 | flush_dcache_page(page); |
d1310b2e | 3666 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3667 | &cached, GFP_NOFS); |
7f042a83 | 3668 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3669 | cur + iosize - 1, &cached); |
92082d40 | 3670 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
3671 | break; |
3672 | } | |
125bac01 | 3673 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3674 | end - cur + 1, em_cached); |
c0347550 | 3675 | if (IS_ERR(em)) { |
7f042a83 | 3676 | unlock_extent(tree, cur, end); |
92082d40 | 3677 | end_page_read(page, false, cur, end + 1 - cur); |
bbf0ea7e | 3678 | ret = PTR_ERR(em); |
d1310b2e CM |
3679 | break; |
3680 | } | |
d1310b2e CM |
3681 | extent_offset = cur - em->start; |
3682 | BUG_ON(extent_map_end(em) <= cur); | |
3683 | BUG_ON(end < cur); | |
3684 | ||
7f6ca7f2 DS |
3685 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
3686 | this_bio_flag = em->compress_type; | |
c8b97818 | 3687 | |
d1310b2e CM |
3688 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3689 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3690 | iosize = ALIGN(iosize, blocksize); |
2a5232a8 | 3691 | if (this_bio_flag != BTRFS_COMPRESS_NONE) |
0c64c33c | 3692 | disk_bytenr = em->block_start; |
949b3273 | 3693 | else |
0c64c33c | 3694 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 3695 | block_start = em->block_start; |
d899e052 YZ |
3696 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3697 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3698 | |
3699 | /* | |
3700 | * If we have a file range that points to a compressed extent | |
260db43c | 3701 | * and it's followed by a consecutive file range that points |
005efedf FM |
3702 | * to the same compressed extent (possibly with a different |
3703 | * offset and/or length, so it either points to the whole extent | |
3704 | * or only part of it), we must make sure we do not submit a | |
3705 | * single bio to populate the pages for the 2 ranges because | |
3706 | * this makes the compressed extent read zero out the pages | |
3707 | * belonging to the 2nd range. Imagine the following scenario: | |
3708 | * | |
3709 | * File layout | |
3710 | * [0 - 8K] [8K - 24K] | |
3711 | * | | | |
3712 | * | | | |
3713 | * points to extent X, points to extent X, | |
3714 | * offset 4K, length of 8K offset 0, length 16K | |
3715 | * | |
3716 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3717 | * | |
3718 | * If the bio to read the compressed extent covers both ranges, | |
3719 | * it will decompress extent X into the pages belonging to the | |
3720 | * first range and then it will stop, zeroing out the remaining | |
3721 | * pages that belong to the other range that points to extent X. | |
3722 | * So here we make sure we submit 2 bios, one for the first | |
3723 | * range and another one for the third range. Both will target | |
3724 | * the same physical extent from disk, but we can't currently | |
3725 | * make the compressed bio endio callback populate the pages | |
3726 | * for both ranges because each compressed bio is tightly | |
3727 | * coupled with a single extent map, and each range can have | |
3728 | * an extent map with a different offset value relative to the | |
3729 | * uncompressed data of our extent and different lengths. This | |
3730 | * is a corner case so we prioritize correctness over | |
3731 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3732 | */ | |
3733 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3734 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3735 | *prev_em_start != em->start) |
005efedf FM |
3736 | force_bio_submit = true; |
3737 | ||
3738 | if (prev_em_start) | |
8e928218 | 3739 | *prev_em_start = em->start; |
005efedf | 3740 | |
d1310b2e CM |
3741 | free_extent_map(em); |
3742 | em = NULL; | |
3743 | ||
3744 | /* we've found a hole, just zero and go on */ | |
3745 | if (block_start == EXTENT_MAP_HOLE) { | |
507903b8 AJ |
3746 | struct extent_state *cached = NULL; |
3747 | ||
d048b9c2 | 3748 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3749 | flush_dcache_page(page); |
d1310b2e CM |
3750 | |
3751 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3752 | &cached, GFP_NOFS); |
7f042a83 | 3753 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3754 | cur + iosize - 1, &cached); |
92082d40 | 3755 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3756 | cur = cur + iosize; |
306e16ce | 3757 | pg_offset += iosize; |
d1310b2e CM |
3758 | continue; |
3759 | } | |
3760 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3761 | if (test_range_bit(tree, cur, cur_end, |
3762 | EXTENT_UPTODATE, 1, NULL)) { | |
7f042a83 | 3763 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3764 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3765 | cur = cur + iosize; |
306e16ce | 3766 | pg_offset += iosize; |
d1310b2e CM |
3767 | continue; |
3768 | } | |
70dec807 CM |
3769 | /* we have an inline extent but it didn't get marked up |
3770 | * to date. Error out | |
3771 | */ | |
3772 | if (block_start == EXTENT_MAP_INLINE) { | |
7f042a83 | 3773 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3774 | end_page_read(page, false, cur, iosize); |
70dec807 | 3775 | cur = cur + iosize; |
306e16ce | 3776 | pg_offset += iosize; |
70dec807 CM |
3777 | continue; |
3778 | } | |
d1310b2e | 3779 | |
0ceb34bf | 3780 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
390ed29b QW |
3781 | bio_ctrl, page, disk_bytenr, iosize, |
3782 | pg_offset, | |
fd513000 | 3783 | end_bio_extent_readpage, 0, |
005efedf FM |
3784 | this_bio_flag, |
3785 | force_bio_submit); | |
ad3fc794 | 3786 | if (ret) { |
10f7f6f8 QW |
3787 | /* |
3788 | * We have to unlock the remaining range, or the page | |
3789 | * will never be unlocked. | |
3790 | */ | |
3791 | unlock_extent(tree, cur, end); | |
3792 | end_page_read(page, false, cur, end + 1 - cur); | |
baf863b9 | 3793 | goto out; |
edd33c99 | 3794 | } |
d1310b2e | 3795 | cur = cur + iosize; |
306e16ce | 3796 | pg_offset += iosize; |
d1310b2e | 3797 | } |
90a887c9 | 3798 | out: |
baf863b9 | 3799 | return ret; |
d1310b2e CM |
3800 | } |
3801 | ||
fdaf9a58 | 3802 | int btrfs_read_folio(struct file *file, struct folio *folio) |
7aab8b32 | 3803 | { |
fdaf9a58 | 3804 | struct page *page = &folio->page; |
7aab8b32 CH |
3805 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
3806 | u64 start = page_offset(page); | |
3807 | u64 end = start + PAGE_SIZE - 1; | |
3808 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; | |
3809 | int ret; | |
3810 | ||
3811 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); | |
3812 | ||
3813 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); | |
3814 | /* | |
3815 | * If btrfs_do_readpage() failed we will want to submit the assembled | |
3816 | * bio to do the cleanup. | |
3817 | */ | |
3818 | if (bio_ctrl.bio) | |
0f07003b | 3819 | submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.compress_type); |
7aab8b32 CH |
3820 | return ret; |
3821 | } | |
3822 | ||
b6660e80 | 3823 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
3824 | u64 start, u64 end, |
3825 | struct extent_map **em_cached, | |
3826 | struct btrfs_bio_ctrl *bio_ctrl, | |
3827 | u64 *prev_em_start) | |
9974090b | 3828 | { |
23d31bd4 | 3829 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3830 | int index; |
3831 | ||
b272ae22 | 3832 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3833 | |
3834 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 3835 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
0f208812 | 3836 | REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3837 | put_page(pages[index]); |
9974090b MX |
3838 | } |
3839 | } | |
3840 | ||
d1310b2e | 3841 | /* |
40f76580 CM |
3842 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3843 | * | |
5eaad97a | 3844 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3845 | * to write the page (copy into inline extent). In this case the IO has |
3846 | * been started and the page is already unlocked. | |
3847 | * | |
3848 | * This returns 0 if all went well (page still locked) | |
3849 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3850 | */ |
cd4c0bf9 | 3851 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
83f1b680 | 3852 | struct page *page, struct writeback_control *wbc) |
40f76580 | 3853 | { |
2749f7ef | 3854 | const u64 page_end = page_offset(page) + PAGE_SIZE - 1; |
cf3075fb | 3855 | u64 delalloc_start = page_offset(page); |
40f76580 | 3856 | u64 delalloc_to_write = 0; |
83f1b680 QW |
3857 | /* How many pages are started by btrfs_run_delalloc_range() */ |
3858 | unsigned long nr_written = 0; | |
40f76580 CM |
3859 | int ret; |
3860 | int page_started = 0; | |
3861 | ||
2749f7ef QW |
3862 | while (delalloc_start < page_end) { |
3863 | u64 delalloc_end = page_end; | |
3864 | bool found; | |
40f76580 | 3865 | |
cd4c0bf9 | 3866 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3867 | &delalloc_start, |
917aacec | 3868 | &delalloc_end); |
3522e903 | 3869 | if (!found) { |
40f76580 CM |
3870 | delalloc_start = delalloc_end + 1; |
3871 | continue; | |
3872 | } | |
cd4c0bf9 | 3873 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
83f1b680 | 3874 | delalloc_end, &page_started, &nr_written, wbc); |
40f76580 | 3875 | if (ret) { |
963e4db8 QW |
3876 | btrfs_page_set_error(inode->root->fs_info, page, |
3877 | page_offset(page), PAGE_SIZE); | |
7361b4ae | 3878 | return ret; |
40f76580 CM |
3879 | } |
3880 | /* | |
ea1754a0 KS |
3881 | * delalloc_end is already one less than the total length, so |
3882 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3883 | */ |
3884 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3885 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3886 | delalloc_start = delalloc_end + 1; |
3887 | } | |
3888 | if (wbc->nr_to_write < delalloc_to_write) { | |
3889 | int thresh = 8192; | |
3890 | ||
3891 | if (delalloc_to_write < thresh * 2) | |
3892 | thresh = delalloc_to_write; | |
3893 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3894 | thresh); | |
3895 | } | |
3896 | ||
83f1b680 | 3897 | /* Did btrfs_run_dealloc_range() already unlock and start the IO? */ |
40f76580 CM |
3898 | if (page_started) { |
3899 | /* | |
83f1b680 QW |
3900 | * We've unlocked the page, so we can't update the mapping's |
3901 | * writeback index, just update nr_to_write. | |
40f76580 | 3902 | */ |
83f1b680 | 3903 | wbc->nr_to_write -= nr_written; |
40f76580 CM |
3904 | return 1; |
3905 | } | |
3906 | ||
b69d1ee9 | 3907 | return 0; |
40f76580 CM |
3908 | } |
3909 | ||
c5ef5c6c QW |
3910 | /* |
3911 | * Find the first byte we need to write. | |
3912 | * | |
3913 | * For subpage, one page can contain several sectors, and | |
3914 | * __extent_writepage_io() will just grab all extent maps in the page | |
3915 | * range and try to submit all non-inline/non-compressed extents. | |
3916 | * | |
3917 | * This is a big problem for subpage, we shouldn't re-submit already written | |
3918 | * data at all. | |
3919 | * This function will lookup subpage dirty bit to find which range we really | |
3920 | * need to submit. | |
3921 | * | |
3922 | * Return the next dirty range in [@start, @end). | |
3923 | * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE. | |
3924 | */ | |
3925 | static void find_next_dirty_byte(struct btrfs_fs_info *fs_info, | |
3926 | struct page *page, u64 *start, u64 *end) | |
3927 | { | |
3928 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
72a69cd0 | 3929 | struct btrfs_subpage_info *spi = fs_info->subpage_info; |
c5ef5c6c QW |
3930 | u64 orig_start = *start; |
3931 | /* Declare as unsigned long so we can use bitmap ops */ | |
c5ef5c6c | 3932 | unsigned long flags; |
72a69cd0 | 3933 | int range_start_bit; |
c5ef5c6c QW |
3934 | int range_end_bit; |
3935 | ||
3936 | /* | |
3937 | * For regular sector size == page size case, since one page only | |
3938 | * contains one sector, we return the page offset directly. | |
3939 | */ | |
fbca46eb | 3940 | if (!btrfs_is_subpage(fs_info, page)) { |
c5ef5c6c QW |
3941 | *start = page_offset(page); |
3942 | *end = page_offset(page) + PAGE_SIZE; | |
3943 | return; | |
3944 | } | |
3945 | ||
72a69cd0 QW |
3946 | range_start_bit = spi->dirty_offset + |
3947 | (offset_in_page(orig_start) >> fs_info->sectorsize_bits); | |
3948 | ||
c5ef5c6c QW |
3949 | /* We should have the page locked, but just in case */ |
3950 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
3951 | bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit, |
3952 | spi->dirty_offset + spi->bitmap_nr_bits); | |
c5ef5c6c QW |
3953 | spin_unlock_irqrestore(&subpage->lock, flags); |
3954 | ||
72a69cd0 QW |
3955 | range_start_bit -= spi->dirty_offset; |
3956 | range_end_bit -= spi->dirty_offset; | |
3957 | ||
c5ef5c6c QW |
3958 | *start = page_offset(page) + range_start_bit * fs_info->sectorsize; |
3959 | *end = page_offset(page) + range_end_bit * fs_info->sectorsize; | |
3960 | } | |
3961 | ||
40f76580 CM |
3962 | /* |
3963 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3964 | * and does the loop to map the page into extents and bios. | |
3965 | * | |
3966 | * We return 1 if the IO is started and the page is unlocked, | |
3967 | * 0 if all went well (page still locked) | |
3968 | * < 0 if there were errors (page still locked) | |
3969 | */ | |
d4580fe2 | 3970 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3971 | struct page *page, |
3972 | struct writeback_control *wbc, | |
3973 | struct extent_page_data *epd, | |
3974 | loff_t i_size, | |
57e5ffeb | 3975 | int *nr_ret) |
d1310b2e | 3976 | { |
6bc5636a | 3977 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a129ffb8 QW |
3978 | u64 cur = page_offset(page); |
3979 | u64 end = cur + PAGE_SIZE - 1; | |
d1310b2e | 3980 | u64 extent_offset; |
d1310b2e | 3981 | u64 block_start; |
d1310b2e | 3982 | struct extent_map *em; |
44e5801f | 3983 | int saved_ret = 0; |
40f76580 CM |
3984 | int ret = 0; |
3985 | int nr = 0; | |
d8e3fb10 | 3986 | u32 opf = REQ_OP_WRITE; |
57e5ffeb | 3987 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
44e5801f | 3988 | bool has_error = false; |
40f76580 | 3989 | bool compressed; |
c8b97818 | 3990 | |
a129ffb8 | 3991 | ret = btrfs_writepage_cow_fixup(page); |
d75855b4 NB |
3992 | if (ret) { |
3993 | /* Fixup worker will requeue */ | |
5ab58055 | 3994 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3995 | unlock_page(page); |
3996 | return 1; | |
247e743c CM |
3997 | } |
3998 | ||
11c8349b CM |
3999 | /* |
4000 | * we don't want to touch the inode after unlocking the page, | |
4001 | * so we update the mapping writeback index now | |
4002 | */ | |
572f3dad | 4003 | wbc->nr_to_write--; |
771ed689 | 4004 | |
d1310b2e | 4005 | while (cur <= end) { |
0c64c33c | 4006 | u64 disk_bytenr; |
40f76580 | 4007 | u64 em_end; |
c5ef5c6c QW |
4008 | u64 dirty_range_start = cur; |
4009 | u64 dirty_range_end; | |
6bc5636a | 4010 | u32 iosize; |
58409edd | 4011 | |
40f76580 | 4012 | if (cur >= i_size) { |
38a39ac7 | 4013 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
25c1252a | 4014 | end, true); |
cc1d0d93 QW |
4015 | /* |
4016 | * This range is beyond i_size, thus we don't need to | |
4017 | * bother writing back. | |
4018 | * But we still need to clear the dirty subpage bit, or | |
4019 | * the next time the page gets dirtied, we will try to | |
4020 | * writeback the sectors with subpage dirty bits, | |
4021 | * causing writeback without ordered extent. | |
4022 | */ | |
4023 | btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur); | |
d1310b2e CM |
4024 | break; |
4025 | } | |
c5ef5c6c QW |
4026 | |
4027 | find_next_dirty_byte(fs_info, page, &dirty_range_start, | |
4028 | &dirty_range_end); | |
4029 | if (cur < dirty_range_start) { | |
4030 | cur = dirty_range_start; | |
4031 | continue; | |
4032 | } | |
4033 | ||
d4580fe2 | 4034 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c0347550 | 4035 | if (IS_ERR(em)) { |
c5ef5c6c | 4036 | btrfs_page_set_error(fs_info, page, cur, end - cur + 1); |
61391d56 | 4037 | ret = PTR_ERR_OR_ZERO(em); |
44e5801f QW |
4038 | has_error = true; |
4039 | if (!saved_ret) | |
4040 | saved_ret = ret; | |
d1310b2e CM |
4041 | break; |
4042 | } | |
4043 | ||
4044 | extent_offset = cur - em->start; | |
40f76580 | 4045 | em_end = extent_map_end(em); |
6bc5636a QW |
4046 | ASSERT(cur <= em_end); |
4047 | ASSERT(cur < end); | |
4048 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
4049 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 4050 | block_start = em->block_start; |
c8b97818 | 4051 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
4052 | disk_bytenr = em->block_start + extent_offset; |
4053 | ||
c5ef5c6c QW |
4054 | /* |
4055 | * Note that em_end from extent_map_end() and dirty_range_end from | |
4056 | * find_next_dirty_byte() are all exclusive | |
4057 | */ | |
4058 | iosize = min(min(em_end, end + 1), dirty_range_end) - cur; | |
d8e3fb10 | 4059 | |
e380adfc | 4060 | if (btrfs_use_zone_append(inode, em->block_start)) |
d8e3fb10 NA |
4061 | opf = REQ_OP_ZONE_APPEND; |
4062 | ||
d1310b2e CM |
4063 | free_extent_map(em); |
4064 | em = NULL; | |
4065 | ||
c8b97818 CM |
4066 | /* |
4067 | * compressed and inline extents are written through other | |
4068 | * paths in the FS | |
4069 | */ | |
4070 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 4071 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 4072 | if (compressed) |
c8b97818 | 4073 | nr++; |
c8b04030 | 4074 | else |
38a39ac7 | 4075 | btrfs_writepage_endio_finish_ordered(inode, |
25c1252a | 4076 | page, cur, cur + iosize - 1, true); |
cc1d0d93 | 4077 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); |
c8b97818 | 4078 | cur += iosize; |
d1310b2e CM |
4079 | continue; |
4080 | } | |
c8b97818 | 4081 | |
d2a91064 | 4082 | btrfs_set_range_writeback(inode, cur, cur + iosize - 1); |
58409edd | 4083 | if (!PageWriteback(page)) { |
d4580fe2 | 4084 | btrfs_err(inode->root->fs_info, |
58409edd DS |
4085 | "page %lu not writeback, cur %llu end %llu", |
4086 | page->index, cur, end); | |
d1310b2e | 4087 | } |
7f3c74fb | 4088 | |
c5ef5c6c QW |
4089 | /* |
4090 | * Although the PageDirty bit is cleared before entering this | |
4091 | * function, subpage dirty bit is not cleared. | |
4092 | * So clear subpage dirty bit here so next time we won't submit | |
4093 | * page for range already written to disk. | |
4094 | */ | |
4095 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); | |
4096 | ||
390ed29b QW |
4097 | ret = submit_extent_page(opf | write_flags, wbc, |
4098 | &epd->bio_ctrl, page, | |
d8e3fb10 | 4099 | disk_bytenr, iosize, |
390ed29b | 4100 | cur - page_offset(page), |
58409edd | 4101 | end_bio_extent_writepage, |
390ed29b | 4102 | 0, 0, false); |
fe01aa65 | 4103 | if (ret) { |
44e5801f QW |
4104 | has_error = true; |
4105 | if (!saved_ret) | |
4106 | saved_ret = ret; | |
4107 | ||
c5ef5c6c | 4108 | btrfs_page_set_error(fs_info, page, cur, iosize); |
fe01aa65 | 4109 | if (PageWriteback(page)) |
c5ef5c6c QW |
4110 | btrfs_page_clear_writeback(fs_info, page, cur, |
4111 | iosize); | |
fe01aa65 | 4112 | } |
d1310b2e | 4113 | |
6bc5636a | 4114 | cur += iosize; |
d1310b2e CM |
4115 | nr++; |
4116 | } | |
cc1d0d93 QW |
4117 | /* |
4118 | * If we finish without problem, we should not only clear page dirty, | |
4119 | * but also empty subpage dirty bits | |
4120 | */ | |
44e5801f | 4121 | if (!has_error) |
cc1d0d93 | 4122 | btrfs_page_assert_not_dirty(fs_info, page); |
44e5801f QW |
4123 | else |
4124 | ret = saved_ret; | |
40f76580 | 4125 | *nr_ret = nr; |
40f76580 CM |
4126 | return ret; |
4127 | } | |
4128 | ||
4129 | /* | |
4130 | * the writepage semantics are similar to regular writepage. extent | |
4131 | * records are inserted to lock ranges in the tree, and as dirty areas | |
4132 | * are found, they are marked writeback. Then the lock bits are removed | |
4133 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
4134 | * |
4135 | * Return 0 if everything goes well. | |
4136 | * Return <0 for error. | |
40f76580 CM |
4137 | */ |
4138 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 4139 | struct extent_page_data *epd) |
40f76580 | 4140 | { |
8e1dec8e | 4141 | struct folio *folio = page_folio(page); |
40f76580 | 4142 | struct inode *inode = page->mapping->host; |
e55a0de1 | 4143 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
cf3075fb QW |
4144 | const u64 page_start = page_offset(page); |
4145 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
40f76580 CM |
4146 | int ret; |
4147 | int nr = 0; | |
eb70d222 | 4148 | size_t pg_offset; |
40f76580 | 4149 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 4150 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 | 4151 | |
40f76580 CM |
4152 | trace___extent_writepage(page, inode, wbc); |
4153 | ||
4154 | WARN_ON(!PageLocked(page)); | |
4155 | ||
963e4db8 QW |
4156 | btrfs_page_clear_error(btrfs_sb(inode->i_sb), page, |
4157 | page_offset(page), PAGE_SIZE); | |
40f76580 | 4158 | |
7073017a | 4159 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
4160 | if (page->index > end_index || |
4161 | (page->index == end_index && !pg_offset)) { | |
8e1dec8e MWO |
4162 | folio_invalidate(folio, 0, folio_size(folio)); |
4163 | folio_unlock(folio); | |
40f76580 CM |
4164 | return 0; |
4165 | } | |
4166 | ||
4167 | if (page->index == end_index) { | |
d048b9c2 | 4168 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 CM |
4169 | flush_dcache_page(page); |
4170 | } | |
4171 | ||
32443de3 QW |
4172 | ret = set_page_extent_mapped(page); |
4173 | if (ret < 0) { | |
4174 | SetPageError(page); | |
4175 | goto done; | |
4176 | } | |
40f76580 | 4177 | |
7789a55a | 4178 | if (!epd->extent_locked) { |
83f1b680 | 4179 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc); |
7789a55a | 4180 | if (ret == 1) |
169d2c87 | 4181 | return 0; |
7789a55a NB |
4182 | if (ret) |
4183 | goto done; | |
4184 | } | |
40f76580 | 4185 | |
d4580fe2 | 4186 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
83f1b680 | 4187 | &nr); |
40f76580 | 4188 | if (ret == 1) |
169d2c87 | 4189 | return 0; |
40f76580 | 4190 | |
d1310b2e CM |
4191 | done: |
4192 | if (nr == 0) { | |
4193 | /* make sure the mapping tag for page dirty gets cleared */ | |
4194 | set_page_writeback(page); | |
4195 | end_page_writeback(page); | |
4196 | } | |
963e4db8 QW |
4197 | /* |
4198 | * Here we used to have a check for PageError() and then set @ret and | |
4199 | * call end_extent_writepage(). | |
4200 | * | |
4201 | * But in fact setting @ret here will cause different error paths | |
4202 | * between subpage and regular sectorsize. | |
4203 | * | |
4204 | * For regular page size, we never submit current page, but only add | |
4205 | * current page to current bio. | |
4206 | * The bio submission can only happen in next page. | |
4207 | * Thus if we hit the PageError() branch, @ret is already set to | |
4208 | * non-zero value and will not get updated for regular sectorsize. | |
4209 | * | |
4210 | * But for subpage case, it's possible we submit part of current page, | |
4211 | * thus can get PageError() set by submitted bio of the same page, | |
4212 | * while our @ret is still 0. | |
4213 | * | |
4214 | * So here we unify the behavior and don't set @ret. | |
4215 | * Error can still be properly passed to higher layer as page will | |
4216 | * be set error, here we just don't handle the IO failure. | |
4217 | * | |
4218 | * NOTE: This is just a hotfix for subpage. | |
4219 | * The root fix will be properly ending ordered extent when we hit | |
4220 | * an error during writeback. | |
4221 | * | |
4222 | * But that needs a bigger refactoring, as we not only need to grab the | |
4223 | * submitted OE, but also need to know exactly at which bytenr we hit | |
4224 | * the error. | |
4225 | * Currently the full page based __extent_writepage_io() is not | |
4226 | * capable of that. | |
4227 | */ | |
4228 | if (PageError(page)) | |
cf3075fb | 4229 | end_extent_writepage(page, ret, page_start, page_end); |
e55a0de1 QW |
4230 | if (epd->extent_locked) { |
4231 | /* | |
4232 | * If epd->extent_locked, it's from extent_write_locked_range(), | |
4233 | * the page can either be locked by lock_page() or | |
4234 | * process_one_page(). | |
4235 | * Let btrfs_page_unlock_writer() handle both cases. | |
4236 | */ | |
4237 | ASSERT(wbc); | |
4238 | btrfs_page_unlock_writer(fs_info, page, wbc->range_start, | |
4239 | wbc->range_end + 1 - wbc->range_start); | |
4240 | } else { | |
4241 | unlock_page(page); | |
4242 | } | |
3065976b | 4243 | ASSERT(ret <= 0); |
40f76580 | 4244 | return ret; |
d1310b2e CM |
4245 | } |
4246 | ||
fd8b2b61 | 4247 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 4248 | { |
74316201 N |
4249 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
4250 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
4251 | } |
4252 | ||
18dfa711 FM |
4253 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
4254 | { | |
4255 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
4256 | smp_mb__after_atomic(); | |
4257 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
4258 | } | |
4259 | ||
2e3c2513 | 4260 | /* |
a3efb2f0 | 4261 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 4262 | * |
a3efb2f0 QW |
4263 | * May try to flush write bio if we can't get the lock. |
4264 | * | |
4265 | * Return 0 if the extent buffer doesn't need to be submitted. | |
4266 | * (E.g. the extent buffer is not dirty) | |
4267 | * Return >0 is the extent buffer is submitted to bio. | |
4268 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 4269 | */ |
9df76fb5 | 4270 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 4271 | struct extent_page_data *epd) |
0b32f4bb | 4272 | { |
9df76fb5 | 4273 | struct btrfs_fs_info *fs_info = eb->fs_info; |
c9583ada | 4274 | int i, num_pages; |
0b32f4bb JB |
4275 | int flush = 0; |
4276 | int ret = 0; | |
4277 | ||
4278 | if (!btrfs_try_tree_write_lock(eb)) { | |
c9583ada | 4279 | flush_write_bio(epd); |
2e3c2513 | 4280 | flush = 1; |
0b32f4bb JB |
4281 | btrfs_tree_lock(eb); |
4282 | } | |
4283 | ||
4284 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
4285 | btrfs_tree_unlock(eb); | |
4286 | if (!epd->sync_io) | |
4287 | return 0; | |
4288 | if (!flush) { | |
c9583ada | 4289 | flush_write_bio(epd); |
0b32f4bb JB |
4290 | flush = 1; |
4291 | } | |
a098d8e8 CM |
4292 | while (1) { |
4293 | wait_on_extent_buffer_writeback(eb); | |
4294 | btrfs_tree_lock(eb); | |
4295 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
4296 | break; | |
0b32f4bb | 4297 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
4298 | } |
4299 | } | |
4300 | ||
51561ffe JB |
4301 | /* |
4302 | * We need to do this to prevent races in people who check if the eb is | |
4303 | * under IO since we can end up having no IO bits set for a short period | |
4304 | * of time. | |
4305 | */ | |
4306 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
4307 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
4308 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 4309 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 4310 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
4311 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4312 | -eb->len, | |
4313 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 4314 | ret = 1; |
51561ffe JB |
4315 | } else { |
4316 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
4317 | } |
4318 | ||
4319 | btrfs_tree_unlock(eb); | |
4320 | ||
f3156df9 QW |
4321 | /* |
4322 | * Either we don't need to submit any tree block, or we're submitting | |
4323 | * subpage eb. | |
4324 | * Subpage metadata doesn't use page locking at all, so we can skip | |
4325 | * the page locking. | |
4326 | */ | |
fbca46eb | 4327 | if (!ret || fs_info->nodesize < PAGE_SIZE) |
0b32f4bb JB |
4328 | return ret; |
4329 | ||
65ad0104 | 4330 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4331 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4332 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4333 | |
4334 | if (!trylock_page(p)) { | |
4335 | if (!flush) { | |
c9583ada | 4336 | flush_write_bio(epd); |
0b32f4bb JB |
4337 | flush = 1; |
4338 | } | |
4339 | lock_page(p); | |
4340 | } | |
4341 | } | |
4342 | ||
2e3c2513 | 4343 | return ret; |
0b32f4bb JB |
4344 | } |
4345 | ||
5a2c6075 | 4346 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 4347 | { |
5a2c6075 | 4348 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 4349 | |
5a2c6075 | 4350 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
4351 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
4352 | return; | |
4353 | ||
c2e39305 JB |
4354 | /* |
4355 | * A read may stumble upon this buffer later, make sure that it gets an | |
4356 | * error and knows there was an error. | |
4357 | */ | |
4358 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
4359 | ||
68b85589 JB |
4360 | /* |
4361 | * We need to set the mapping with the io error as well because a write | |
4362 | * error will flip the file system readonly, and then syncfs() will | |
4363 | * return a 0 because we are readonly if we don't modify the err seq for | |
4364 | * the superblock. | |
4365 | */ | |
4366 | mapping_set_error(page->mapping, -EIO); | |
4367 | ||
eb5b64f1 DZ |
4368 | /* |
4369 | * If we error out, we should add back the dirty_metadata_bytes | |
4370 | * to make it consistent. | |
4371 | */ | |
eb5b64f1 DZ |
4372 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4373 | eb->len, fs_info->dirty_metadata_batch); | |
4374 | ||
656f30db FM |
4375 | /* |
4376 | * If writeback for a btree extent that doesn't belong to a log tree | |
4377 | * failed, increment the counter transaction->eb_write_errors. | |
4378 | * We do this because while the transaction is running and before it's | |
4379 | * committing (when we call filemap_fdata[write|wait]_range against | |
4380 | * the btree inode), we might have | |
4381 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
4382 | * returns an error or an error happens during writeback, when we're | |
4383 | * committing the transaction we wouldn't know about it, since the pages | |
4384 | * can be no longer dirty nor marked anymore for writeback (if a | |
4385 | * subsequent modification to the extent buffer didn't happen before the | |
4386 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
4387 | * able to find the pages tagged with SetPageError at transaction | |
4388 | * commit time. So if this happens we must abort the transaction, | |
4389 | * otherwise we commit a super block with btree roots that point to | |
4390 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
4391 | * or the content of some node/leaf from a past generation that got | |
4392 | * cowed or deleted and is no longer valid. | |
4393 | * | |
4394 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
4395 | * not be enough - we need to distinguish between log tree extents vs | |
4396 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
4397 | * will catch and clear such errors in the mapping - and that call might | |
4398 | * be from a log sync and not from a transaction commit. Also, checking | |
4399 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
4400 | * not done and would not be reliable - the eb might have been released | |
4401 | * from memory and reading it back again means that flag would not be | |
4402 | * set (since it's a runtime flag, not persisted on disk). | |
4403 | * | |
4404 | * Using the flags below in the btree inode also makes us achieve the | |
4405 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
4406 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
4407 | * is called, the writeback for all dirty pages had already finished | |
4408 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
4409 | * filemap_fdatawait_range() would return success, as it could not know | |
4410 | * that writeback errors happened (the pages were no longer tagged for | |
4411 | * writeback). | |
4412 | */ | |
4413 | switch (eb->log_index) { | |
4414 | case -1: | |
5a2c6075 | 4415 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
4416 | break; |
4417 | case 0: | |
5a2c6075 | 4418 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
4419 | break; |
4420 | case 1: | |
5a2c6075 | 4421 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
4422 | break; |
4423 | default: | |
4424 | BUG(); /* unexpected, logic error */ | |
4425 | } | |
4426 | } | |
4427 | ||
2f3186d8 QW |
4428 | /* |
4429 | * The endio specific version which won't touch any unsafe spinlock in endio | |
4430 | * context. | |
4431 | */ | |
4432 | static struct extent_buffer *find_extent_buffer_nolock( | |
4433 | struct btrfs_fs_info *fs_info, u64 start) | |
4434 | { | |
4435 | struct extent_buffer *eb; | |
4436 | ||
4437 | rcu_read_lock(); | |
8ee92268 GN |
4438 | eb = xa_load(&fs_info->extent_buffers, |
4439 | start >> fs_info->sectorsize_bits); | |
2f3186d8 QW |
4440 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
4441 | rcu_read_unlock(); | |
4442 | return eb; | |
4443 | } | |
4444 | rcu_read_unlock(); | |
4445 | return NULL; | |
4446 | } | |
4447 | ||
4448 | /* | |
4449 | * The endio function for subpage extent buffer write. | |
4450 | * | |
4451 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
4452 | * after all extent buffers in the page has finished their writeback. | |
4453 | */ | |
fa04c165 | 4454 | static void end_bio_subpage_eb_writepage(struct bio *bio) |
2f3186d8 | 4455 | { |
fa04c165 | 4456 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
4457 | struct bio_vec *bvec; |
4458 | struct bvec_iter_all iter_all; | |
4459 | ||
fa04c165 | 4460 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
fbca46eb | 4461 | ASSERT(fs_info->nodesize < PAGE_SIZE); |
fa04c165 | 4462 | |
2f3186d8 QW |
4463 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
4464 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
4465 | struct page *page = bvec->bv_page; | |
4466 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
4467 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
4468 | u64 cur_bytenr = bvec_start; | |
4469 | ||
4470 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
4471 | ||
4472 | /* Iterate through all extent buffers in the range */ | |
4473 | while (cur_bytenr <= bvec_end) { | |
4474 | struct extent_buffer *eb; | |
4475 | int done; | |
4476 | ||
4477 | /* | |
4478 | * Here we can't use find_extent_buffer(), as it may | |
4479 | * try to lock eb->refs_lock, which is not safe in endio | |
4480 | * context. | |
4481 | */ | |
4482 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
4483 | ASSERT(eb); | |
4484 | ||
4485 | cur_bytenr = eb->start + eb->len; | |
4486 | ||
4487 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
4488 | done = atomic_dec_and_test(&eb->io_pages); | |
4489 | ASSERT(done); | |
4490 | ||
4491 | if (bio->bi_status || | |
4492 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
4493 | ClearPageUptodate(page); | |
4494 | set_btree_ioerr(page, eb); | |
4495 | } | |
4496 | ||
4497 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
4498 | eb->len); | |
4499 | end_extent_buffer_writeback(eb); | |
4500 | /* | |
4501 | * free_extent_buffer() will grab spinlock which is not | |
4502 | * safe in endio context. Thus here we manually dec | |
4503 | * the ref. | |
4504 | */ | |
4505 | atomic_dec(&eb->refs); | |
4506 | } | |
4507 | } | |
4508 | bio_put(bio); | |
4509 | } | |
4510 | ||
4246a0b6 | 4511 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 4512 | { |
2c30c71b | 4513 | struct bio_vec *bvec; |
0b32f4bb | 4514 | struct extent_buffer *eb; |
2b070cfe | 4515 | int done; |
6dc4f100 | 4516 | struct bvec_iter_all iter_all; |
0b32f4bb | 4517 | |
c09abff8 | 4518 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 4519 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
4520 | struct page *page = bvec->bv_page; |
4521 | ||
0b32f4bb JB |
4522 | eb = (struct extent_buffer *)page->private; |
4523 | BUG_ON(!eb); | |
4524 | done = atomic_dec_and_test(&eb->io_pages); | |
4525 | ||
4e4cbee9 | 4526 | if (bio->bi_status || |
4246a0b6 | 4527 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 4528 | ClearPageUptodate(page); |
5a2c6075 | 4529 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
4530 | } |
4531 | ||
4532 | end_page_writeback(page); | |
4533 | ||
4534 | if (!done) | |
4535 | continue; | |
4536 | ||
4537 | end_extent_buffer_writeback(eb); | |
2c30c71b | 4538 | } |
0b32f4bb JB |
4539 | |
4540 | bio_put(bio); | |
0b32f4bb JB |
4541 | } |
4542 | ||
fa04c165 QW |
4543 | static void prepare_eb_write(struct extent_buffer *eb) |
4544 | { | |
4545 | u32 nritems; | |
4546 | unsigned long start; | |
4547 | unsigned long end; | |
4548 | ||
4549 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
4550 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
4551 | ||
4552 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
4553 | nritems = btrfs_header_nritems(eb); | |
4554 | if (btrfs_header_level(eb) > 0) { | |
4555 | end = btrfs_node_key_ptr_offset(nritems); | |
4556 | memzero_extent_buffer(eb, end, eb->len - end); | |
4557 | } else { | |
4558 | /* | |
4559 | * Leaf: | |
4560 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
4561 | */ | |
4562 | start = btrfs_item_nr_offset(nritems); | |
4563 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); | |
4564 | memzero_extent_buffer(eb, start, end - start); | |
4565 | } | |
4566 | } | |
4567 | ||
35b6ddfa QW |
4568 | /* |
4569 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
4570 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
4571 | */ |
4572 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
4573 | struct writeback_control *wbc, | |
4574 | struct extent_page_data *epd) | |
4575 | { | |
4576 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4577 | struct page *page = eb->pages[0]; | |
4578 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; | |
4579 | bool no_dirty_ebs = false; | |
4580 | int ret; | |
4581 | ||
fa04c165 QW |
4582 | prepare_eb_write(eb); |
4583 | ||
35b6ddfa QW |
4584 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
4585 | lock_page(page); | |
4586 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
4587 | ||
4588 | /* Check if this is the last dirty bit to update nr_written */ | |
4589 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
4590 | eb->start, eb->len); | |
4591 | if (no_dirty_ebs) | |
4592 | clear_page_dirty_for_io(page); | |
4593 | ||
390ed29b QW |
4594 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
4595 | &epd->bio_ctrl, page, eb->start, eb->len, | |
4596 | eb->start - page_offset(page), | |
fa04c165 | 4597 | end_bio_subpage_eb_writepage, 0, 0, false); |
35b6ddfa QW |
4598 | if (ret) { |
4599 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
4600 | set_btree_ioerr(page, eb); | |
4601 | unlock_page(page); | |
4602 | ||
4603 | if (atomic_dec_and_test(&eb->io_pages)) | |
4604 | end_extent_buffer_writeback(eb); | |
4605 | return -EIO; | |
4606 | } | |
4607 | unlock_page(page); | |
4608 | /* | |
4609 | * Submission finished without problem, if no range of the page is | |
4610 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
4611 | */ | |
4612 | if (no_dirty_ebs) | |
572f3dad | 4613 | wbc->nr_to_write--; |
35b6ddfa QW |
4614 | return ret; |
4615 | } | |
4616 | ||
0e378df1 | 4617 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
4618 | struct writeback_control *wbc, |
4619 | struct extent_page_data *epd) | |
4620 | { | |
0c64c33c | 4621 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 4622 | int i, num_pages; |
ff40adf7 | 4623 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 4624 | int ret = 0; |
0b32f4bb | 4625 | |
fa04c165 | 4626 | prepare_eb_write(eb); |
35b6ddfa | 4627 | |
fa04c165 | 4628 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4629 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4630 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4631 | |
4632 | clear_page_dirty_for_io(p); | |
4633 | set_page_writeback(p); | |
0ceb34bf | 4634 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
390ed29b QW |
4635 | &epd->bio_ctrl, p, disk_bytenr, |
4636 | PAGE_SIZE, 0, | |
1f7ad75b | 4637 | end_bio_extent_buffer_writepage, |
390ed29b | 4638 | 0, 0, false); |
0b32f4bb | 4639 | if (ret) { |
5a2c6075 | 4640 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
4641 | if (PageWriteback(p)) |
4642 | end_page_writeback(p); | |
0b32f4bb JB |
4643 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
4644 | end_extent_buffer_writeback(eb); | |
4645 | ret = -EIO; | |
4646 | break; | |
4647 | } | |
0c64c33c | 4648 | disk_bytenr += PAGE_SIZE; |
572f3dad | 4649 | wbc->nr_to_write--; |
0b32f4bb JB |
4650 | unlock_page(p); |
4651 | } | |
4652 | ||
4653 | if (unlikely(ret)) { | |
4654 | for (; i < num_pages; i++) { | |
bbf65cf0 | 4655 | struct page *p = eb->pages[i]; |
81465028 | 4656 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
4657 | unlock_page(p); |
4658 | } | |
4659 | } | |
4660 | ||
4661 | return ret; | |
4662 | } | |
4663 | ||
c4aec299 QW |
4664 | /* |
4665 | * Submit one subpage btree page. | |
4666 | * | |
4667 | * The main difference to submit_eb_page() is: | |
4668 | * - Page locking | |
4669 | * For subpage, we don't rely on page locking at all. | |
4670 | * | |
4671 | * - Flush write bio | |
4672 | * We only flush bio if we may be unable to fit current extent buffers into | |
4673 | * current bio. | |
4674 | * | |
4675 | * Return >=0 for the number of submitted extent buffers. | |
4676 | * Return <0 for fatal error. | |
4677 | */ | |
4678 | static int submit_eb_subpage(struct page *page, | |
4679 | struct writeback_control *wbc, | |
4680 | struct extent_page_data *epd) | |
4681 | { | |
4682 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
4683 | int submitted = 0; | |
4684 | u64 page_start = page_offset(page); | |
4685 | int bit_start = 0; | |
c4aec299 QW |
4686 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; |
4687 | int ret; | |
4688 | ||
4689 | /* Lock and write each dirty extent buffers in the range */ | |
72a69cd0 | 4690 | while (bit_start < fs_info->subpage_info->bitmap_nr_bits) { |
c4aec299 QW |
4691 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; |
4692 | struct extent_buffer *eb; | |
4693 | unsigned long flags; | |
4694 | u64 start; | |
4695 | ||
4696 | /* | |
4697 | * Take private lock to ensure the subpage won't be detached | |
4698 | * in the meantime. | |
4699 | */ | |
4700 | spin_lock(&page->mapping->private_lock); | |
4701 | if (!PagePrivate(page)) { | |
4702 | spin_unlock(&page->mapping->private_lock); | |
4703 | break; | |
4704 | } | |
4705 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
4706 | if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset, |
4707 | subpage->bitmaps)) { | |
c4aec299 QW |
4708 | spin_unlock_irqrestore(&subpage->lock, flags); |
4709 | spin_unlock(&page->mapping->private_lock); | |
4710 | bit_start++; | |
4711 | continue; | |
4712 | } | |
4713 | ||
4714 | start = page_start + bit_start * fs_info->sectorsize; | |
4715 | bit_start += sectors_per_node; | |
4716 | ||
4717 | /* | |
4718 | * Here we just want to grab the eb without touching extra | |
4719 | * spin locks, so call find_extent_buffer_nolock(). | |
4720 | */ | |
4721 | eb = find_extent_buffer_nolock(fs_info, start); | |
4722 | spin_unlock_irqrestore(&subpage->lock, flags); | |
4723 | spin_unlock(&page->mapping->private_lock); | |
4724 | ||
4725 | /* | |
4726 | * The eb has already reached 0 refs thus find_extent_buffer() | |
4727 | * doesn't return it. We don't need to write back such eb | |
4728 | * anyway. | |
4729 | */ | |
4730 | if (!eb) | |
4731 | continue; | |
4732 | ||
4733 | ret = lock_extent_buffer_for_io(eb, epd); | |
4734 | if (ret == 0) { | |
4735 | free_extent_buffer(eb); | |
4736 | continue; | |
4737 | } | |
4738 | if (ret < 0) { | |
4739 | free_extent_buffer(eb); | |
4740 | goto cleanup; | |
4741 | } | |
fa04c165 | 4742 | ret = write_one_subpage_eb(eb, wbc, epd); |
c4aec299 QW |
4743 | free_extent_buffer(eb); |
4744 | if (ret < 0) | |
4745 | goto cleanup; | |
4746 | submitted++; | |
4747 | } | |
4748 | return submitted; | |
4749 | ||
4750 | cleanup: | |
4751 | /* We hit error, end bio for the submitted extent buffers */ | |
4752 | end_write_bio(epd, ret); | |
4753 | return ret; | |
4754 | } | |
4755 | ||
f91e0d0c QW |
4756 | /* |
4757 | * Submit all page(s) of one extent buffer. | |
4758 | * | |
4759 | * @page: the page of one extent buffer | |
4760 | * @eb_context: to determine if we need to submit this page, if current page | |
4761 | * belongs to this eb, we don't need to submit | |
4762 | * | |
4763 | * The caller should pass each page in their bytenr order, and here we use | |
4764 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
4765 | * | |
4766 | * If we have, we just skip until we hit a new page that doesn't belong to | |
4767 | * current @eb_context. | |
4768 | * | |
4769 | * If not, we submit all the page(s) of the extent buffer. | |
4770 | * | |
4771 | * Return >0 if we have submitted the extent buffer successfully. | |
4772 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
4773 | * previous call. | |
4774 | * Return <0 for fatal error. | |
4775 | */ | |
4776 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
4777 | struct extent_page_data *epd, | |
4778 | struct extent_buffer **eb_context) | |
4779 | { | |
4780 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 4781 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
4782 | struct extent_buffer *eb; |
4783 | int ret; | |
4784 | ||
4785 | if (!PagePrivate(page)) | |
4786 | return 0; | |
4787 | ||
fbca46eb | 4788 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
c4aec299 QW |
4789 | return submit_eb_subpage(page, wbc, epd); |
4790 | ||
f91e0d0c QW |
4791 | spin_lock(&mapping->private_lock); |
4792 | if (!PagePrivate(page)) { | |
4793 | spin_unlock(&mapping->private_lock); | |
4794 | return 0; | |
4795 | } | |
4796 | ||
4797 | eb = (struct extent_buffer *)page->private; | |
4798 | ||
4799 | /* | |
4800 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
4801 | * crashing the machine for something we can survive anyway. | |
4802 | */ | |
4803 | if (WARN_ON(!eb)) { | |
4804 | spin_unlock(&mapping->private_lock); | |
4805 | return 0; | |
4806 | } | |
4807 | ||
4808 | if (eb == *eb_context) { | |
4809 | spin_unlock(&mapping->private_lock); | |
4810 | return 0; | |
4811 | } | |
4812 | ret = atomic_inc_not_zero(&eb->refs); | |
4813 | spin_unlock(&mapping->private_lock); | |
4814 | if (!ret) | |
4815 | return 0; | |
4816 | ||
0bc09ca1 NA |
4817 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
4818 | /* | |
4819 | * If for_sync, this hole will be filled with | |
4820 | * trasnsaction commit. | |
4821 | */ | |
4822 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
4823 | ret = -EAGAIN; | |
4824 | else | |
4825 | ret = 0; | |
4826 | free_extent_buffer(eb); | |
4827 | return ret; | |
4828 | } | |
4829 | ||
f91e0d0c QW |
4830 | *eb_context = eb; |
4831 | ||
4832 | ret = lock_extent_buffer_for_io(eb, epd); | |
4833 | if (ret <= 0) { | |
0bc09ca1 NA |
4834 | btrfs_revert_meta_write_pointer(cache, eb); |
4835 | if (cache) | |
4836 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
4837 | free_extent_buffer(eb); |
4838 | return ret; | |
4839 | } | |
be1a1d7a | 4840 | if (cache) { |
d3e29967 NB |
4841 | /* |
4842 | * Implies write in zoned mode. Mark the last eb in a block group. | |
4843 | */ | |
56fbb0a4 | 4844 | btrfs_schedule_zone_finish_bg(cache, eb); |
d3e29967 | 4845 | btrfs_put_block_group(cache); |
be1a1d7a | 4846 | } |
f91e0d0c QW |
4847 | ret = write_one_eb(eb, wbc, epd); |
4848 | free_extent_buffer(eb); | |
4849 | if (ret < 0) | |
4850 | return ret; | |
4851 | return 1; | |
4852 | } | |
4853 | ||
0b32f4bb JB |
4854 | int btree_write_cache_pages(struct address_space *mapping, |
4855 | struct writeback_control *wbc) | |
4856 | { | |
f91e0d0c | 4857 | struct extent_buffer *eb_context = NULL; |
0b32f4bb | 4858 | struct extent_page_data epd = { |
390ed29b | 4859 | .bio_ctrl = { 0 }, |
0b32f4bb JB |
4860 | .extent_locked = 0, |
4861 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4862 | }; | |
b3ff8f1d | 4863 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4864 | int ret = 0; |
4865 | int done = 0; | |
4866 | int nr_to_write_done = 0; | |
4867 | struct pagevec pvec; | |
4868 | int nr_pages; | |
4869 | pgoff_t index; | |
4870 | pgoff_t end; /* Inclusive */ | |
4871 | int scanned = 0; | |
10bbd235 | 4872 | xa_mark_t tag; |
0b32f4bb | 4873 | |
86679820 | 4874 | pagevec_init(&pvec); |
0b32f4bb JB |
4875 | if (wbc->range_cyclic) { |
4876 | index = mapping->writeback_index; /* Start from prev offset */ | |
4877 | end = -1; | |
556755a8 JB |
4878 | /* |
4879 | * Start from the beginning does not need to cycle over the | |
4880 | * range, mark it as scanned. | |
4881 | */ | |
4882 | scanned = (index == 0); | |
0b32f4bb | 4883 | } else { |
09cbfeaf KS |
4884 | index = wbc->range_start >> PAGE_SHIFT; |
4885 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4886 | scanned = 1; |
4887 | } | |
4888 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4889 | tag = PAGECACHE_TAG_TOWRITE; | |
4890 | else | |
4891 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 4892 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
4893 | retry: |
4894 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4895 | tag_pages_for_writeback(mapping, index, end); | |
4896 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4897 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4898 | tag))) { |
0b32f4bb JB |
4899 | unsigned i; |
4900 | ||
0b32f4bb JB |
4901 | for (i = 0; i < nr_pages; i++) { |
4902 | struct page *page = pvec.pages[i]; | |
4903 | ||
f91e0d0c QW |
4904 | ret = submit_eb_page(page, wbc, &epd, &eb_context); |
4905 | if (ret == 0) | |
0b32f4bb | 4906 | continue; |
f91e0d0c | 4907 | if (ret < 0) { |
0b32f4bb | 4908 | done = 1; |
0b32f4bb JB |
4909 | break; |
4910 | } | |
0b32f4bb JB |
4911 | |
4912 | /* | |
4913 | * the filesystem may choose to bump up nr_to_write. | |
4914 | * We have to make sure to honor the new nr_to_write | |
4915 | * at any time | |
4916 | */ | |
4917 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4918 | } | |
4919 | pagevec_release(&pvec); | |
4920 | cond_resched(); | |
4921 | } | |
4922 | if (!scanned && !done) { | |
4923 | /* | |
4924 | * We hit the last page and there is more work to be done: wrap | |
4925 | * back to the start of the file | |
4926 | */ | |
4927 | scanned = 1; | |
4928 | index = 0; | |
4929 | goto retry; | |
4930 | } | |
2b952eea QW |
4931 | if (ret < 0) { |
4932 | end_write_bio(&epd, ret); | |
0bc09ca1 | 4933 | goto out; |
2b952eea | 4934 | } |
b3ff8f1d QW |
4935 | /* |
4936 | * If something went wrong, don't allow any metadata write bio to be | |
4937 | * submitted. | |
4938 | * | |
4939 | * This would prevent use-after-free if we had dirty pages not | |
4940 | * cleaned up, which can still happen by fuzzed images. | |
4941 | * | |
4942 | * - Bad extent tree | |
4943 | * Allowing existing tree block to be allocated for other trees. | |
4944 | * | |
4945 | * - Log tree operations | |
4946 | * Exiting tree blocks get allocated to log tree, bumps its | |
4947 | * generation, then get cleaned in tree re-balance. | |
4948 | * Such tree block will not be written back, since it's clean, | |
4949 | * thus no WRITTEN flag set. | |
4950 | * And after log writes back, this tree block is not traced by | |
4951 | * any dirty extent_io_tree. | |
4952 | * | |
4953 | * - Offending tree block gets re-dirtied from its original owner | |
4954 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4955 | * reused without COWing. This tree block will not be traced | |
4956 | * by btrfs_transaction::dirty_pages. | |
4957 | * | |
4958 | * Now such dirty tree block will not be cleaned by any dirty | |
4959 | * extent io tree. Thus we don't want to submit such wild eb | |
4960 | * if the fs already has error. | |
4961 | */ | |
84961539 | 4962 | if (!BTRFS_FS_ERROR(fs_info)) { |
c9583ada | 4963 | flush_write_bio(&epd); |
b3ff8f1d | 4964 | } else { |
fbabd4a3 | 4965 | ret = -EROFS; |
b3ff8f1d QW |
4966 | end_write_bio(&epd, ret); |
4967 | } | |
0bc09ca1 NA |
4968 | out: |
4969 | btrfs_zoned_meta_io_unlock(fs_info); | |
c9583ada QW |
4970 | /* |
4971 | * We can get ret > 0 from submit_extent_page() indicating how many ebs | |
4972 | * were submitted. Reset it to 0 to avoid false alerts for the caller. | |
4973 | */ | |
4974 | if (ret > 0) | |
4975 | ret = 0; | |
0b32f4bb JB |
4976 | return ret; |
4977 | } | |
4978 | ||
d1310b2e | 4979 | /** |
3bed2da1 NB |
4980 | * Walk the list of dirty pages of the given address space and write all of them. |
4981 | * | |
d1310b2e | 4982 | * @mapping: address space structure to write |
3bed2da1 NB |
4983 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
4984 | * @epd: holds context for the write, namely the bio | |
d1310b2e CM |
4985 | * |
4986 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4987 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4988 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4989 | * and msync() need to guarantee that all the data which was dirty at the time | |
4990 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4991 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4992 | * existing IO to complete. | |
4993 | */ | |
4242b64a | 4994 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4995 | struct writeback_control *wbc, |
aab6e9ed | 4996 | struct extent_page_data *epd) |
d1310b2e | 4997 | { |
7fd1a3f7 | 4998 | struct inode *inode = mapping->host; |
d1310b2e CM |
4999 | int ret = 0; |
5000 | int done = 0; | |
f85d7d6c | 5001 | int nr_to_write_done = 0; |
d1310b2e CM |
5002 | struct pagevec pvec; |
5003 | int nr_pages; | |
5004 | pgoff_t index; | |
5005 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
5006 | pgoff_t done_index; |
5007 | int range_whole = 0; | |
d1310b2e | 5008 | int scanned = 0; |
10bbd235 | 5009 | xa_mark_t tag; |
d1310b2e | 5010 | |
7fd1a3f7 JB |
5011 | /* |
5012 | * We have to hold onto the inode so that ordered extents can do their | |
5013 | * work when the IO finishes. The alternative to this is failing to add | |
5014 | * an ordered extent if the igrab() fails there and that is a huge pain | |
5015 | * to deal with, so instead just hold onto the inode throughout the | |
5016 | * writepages operation. If it fails here we are freeing up the inode | |
5017 | * anyway and we'd rather not waste our time writing out stuff that is | |
5018 | * going to be truncated anyway. | |
5019 | */ | |
5020 | if (!igrab(inode)) | |
5021 | return 0; | |
5022 | ||
86679820 | 5023 | pagevec_init(&pvec); |
d1310b2e CM |
5024 | if (wbc->range_cyclic) { |
5025 | index = mapping->writeback_index; /* Start from prev offset */ | |
5026 | end = -1; | |
556755a8 JB |
5027 | /* |
5028 | * Start from the beginning does not need to cycle over the | |
5029 | * range, mark it as scanned. | |
5030 | */ | |
5031 | scanned = (index == 0); | |
d1310b2e | 5032 | } else { |
09cbfeaf KS |
5033 | index = wbc->range_start >> PAGE_SHIFT; |
5034 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
5035 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
5036 | range_whole = 1; | |
d1310b2e CM |
5037 | scanned = 1; |
5038 | } | |
3cd24c69 EL |
5039 | |
5040 | /* | |
5041 | * We do the tagged writepage as long as the snapshot flush bit is set | |
5042 | * and we are the first one who do the filemap_flush() on this inode. | |
5043 | * | |
5044 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
5045 | * not race in and drop the bit. | |
5046 | */ | |
5047 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
5048 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
5049 | &BTRFS_I(inode)->runtime_flags)) | |
5050 | wbc->tagged_writepages = 1; | |
5051 | ||
5052 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
5053 | tag = PAGECACHE_TAG_TOWRITE; |
5054 | else | |
5055 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 5056 | retry: |
3cd24c69 | 5057 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 5058 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 5059 | done_index = index; |
f85d7d6c | 5060 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
5061 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
5062 | &index, end, tag))) { | |
d1310b2e CM |
5063 | unsigned i; |
5064 | ||
d1310b2e CM |
5065 | for (i = 0; i < nr_pages; i++) { |
5066 | struct page *page = pvec.pages[i]; | |
5067 | ||
f7bddf1e | 5068 | done_index = page->index + 1; |
d1310b2e | 5069 | /* |
b93b0163 MW |
5070 | * At this point we hold neither the i_pages lock nor |
5071 | * the page lock: the page may be truncated or | |
5072 | * invalidated (changing page->mapping to NULL), | |
5073 | * or even swizzled back from swapper_space to | |
5074 | * tmpfs file mapping | |
d1310b2e | 5075 | */ |
c8f2f24b | 5076 | if (!trylock_page(page)) { |
c9583ada | 5077 | flush_write_bio(epd); |
c8f2f24b | 5078 | lock_page(page); |
01d658f2 | 5079 | } |
d1310b2e CM |
5080 | |
5081 | if (unlikely(page->mapping != mapping)) { | |
5082 | unlock_page(page); | |
5083 | continue; | |
5084 | } | |
5085 | ||
d2c3f4f6 | 5086 | if (wbc->sync_mode != WB_SYNC_NONE) { |
c9583ada QW |
5087 | if (PageWriteback(page)) |
5088 | flush_write_bio(epd); | |
d1310b2e | 5089 | wait_on_page_writeback(page); |
d2c3f4f6 | 5090 | } |
d1310b2e CM |
5091 | |
5092 | if (PageWriteback(page) || | |
5093 | !clear_page_dirty_for_io(page)) { | |
5094 | unlock_page(page); | |
5095 | continue; | |
5096 | } | |
5097 | ||
aab6e9ed | 5098 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 5099 | if (ret < 0) { |
a9132667 LB |
5100 | done = 1; |
5101 | break; | |
5102 | } | |
f85d7d6c CM |
5103 | |
5104 | /* | |
5105 | * the filesystem may choose to bump up nr_to_write. | |
5106 | * We have to make sure to honor the new nr_to_write | |
5107 | * at any time | |
5108 | */ | |
5109 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
5110 | } |
5111 | pagevec_release(&pvec); | |
5112 | cond_resched(); | |
5113 | } | |
894b36e3 | 5114 | if (!scanned && !done) { |
d1310b2e CM |
5115 | /* |
5116 | * We hit the last page and there is more work to be done: wrap | |
5117 | * back to the start of the file | |
5118 | */ | |
5119 | scanned = 1; | |
5120 | index = 0; | |
42ffb0bf JB |
5121 | |
5122 | /* | |
5123 | * If we're looping we could run into a page that is locked by a | |
5124 | * writer and that writer could be waiting on writeback for a | |
5125 | * page in our current bio, and thus deadlock, so flush the | |
5126 | * write bio here. | |
5127 | */ | |
c9583ada QW |
5128 | flush_write_bio(epd); |
5129 | goto retry; | |
d1310b2e | 5130 | } |
a9132667 LB |
5131 | |
5132 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
5133 | mapping->writeback_index = done_index; | |
5134 | ||
7fd1a3f7 | 5135 | btrfs_add_delayed_iput(inode); |
894b36e3 | 5136 | return ret; |
d1310b2e | 5137 | } |
d1310b2e | 5138 | |
0a9b0e53 | 5139 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
5140 | { |
5141 | int ret; | |
d1310b2e | 5142 | struct extent_page_data epd = { |
390ed29b | 5143 | .bio_ctrl = { 0 }, |
771ed689 | 5144 | .extent_locked = 0, |
ffbd517d | 5145 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 5146 | }; |
d1310b2e | 5147 | |
d1310b2e | 5148 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
5149 | ASSERT(ret <= 0); |
5150 | if (ret < 0) { | |
5151 | end_write_bio(&epd, ret); | |
5152 | return ret; | |
5153 | } | |
d1310b2e | 5154 | |
c9583ada | 5155 | flush_write_bio(&epd); |
d1310b2e CM |
5156 | return ret; |
5157 | } | |
d1310b2e | 5158 | |
2bd0fc93 QW |
5159 | /* |
5160 | * Submit the pages in the range to bio for call sites which delalloc range has | |
5161 | * already been ran (aka, ordered extent inserted) and all pages are still | |
5162 | * locked. | |
5163 | */ | |
5164 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end) | |
771ed689 | 5165 | { |
2bd0fc93 QW |
5166 | bool found_error = false; |
5167 | int first_error = 0; | |
771ed689 CM |
5168 | int ret = 0; |
5169 | struct address_space *mapping = inode->i_mapping; | |
5170 | struct page *page; | |
2bd0fc93 | 5171 | u64 cur = start; |
66448b9d QW |
5172 | unsigned long nr_pages; |
5173 | const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize; | |
771ed689 | 5174 | struct extent_page_data epd = { |
390ed29b | 5175 | .bio_ctrl = { 0 }, |
771ed689 | 5176 | .extent_locked = 1, |
2bd0fc93 | 5177 | .sync_io = 1, |
771ed689 CM |
5178 | }; |
5179 | struct writeback_control wbc_writepages = { | |
2bd0fc93 | 5180 | .sync_mode = WB_SYNC_ALL, |
771ed689 CM |
5181 | .range_start = start, |
5182 | .range_end = end + 1, | |
ec39f769 CM |
5183 | /* We're called from an async helper function */ |
5184 | .punt_to_cgroup = 1, | |
5185 | .no_cgroup_owner = 1, | |
771ed689 CM |
5186 | }; |
5187 | ||
66448b9d QW |
5188 | ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize)); |
5189 | nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >> | |
5190 | PAGE_SHIFT; | |
5191 | wbc_writepages.nr_to_write = nr_pages * 2; | |
5192 | ||
dbb70bec | 5193 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
2bd0fc93 | 5194 | while (cur <= end) { |
66448b9d QW |
5195 | u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end); |
5196 | ||
2bd0fc93 QW |
5197 | page = find_get_page(mapping, cur >> PAGE_SHIFT); |
5198 | /* | |
5199 | * All pages in the range are locked since | |
5200 | * btrfs_run_delalloc_range(), thus there is no way to clear | |
5201 | * the page dirty flag. | |
5202 | */ | |
66448b9d | 5203 | ASSERT(PageLocked(page)); |
2bd0fc93 QW |
5204 | ASSERT(PageDirty(page)); |
5205 | clear_page_dirty_for_io(page); | |
5206 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
5207 | ASSERT(ret <= 0); | |
5208 | if (ret < 0) { | |
5209 | found_error = true; | |
5210 | first_error = ret; | |
771ed689 | 5211 | } |
09cbfeaf | 5212 | put_page(page); |
66448b9d | 5213 | cur = cur_end + 1; |
771ed689 CM |
5214 | } |
5215 | ||
2bd0fc93 | 5216 | if (!found_error) |
c9583ada | 5217 | flush_write_bio(&epd); |
dbb70bec | 5218 | else |
02c6db4f | 5219 | end_write_bio(&epd, ret); |
dbb70bec CM |
5220 | |
5221 | wbc_detach_inode(&wbc_writepages); | |
2bd0fc93 QW |
5222 | if (found_error) |
5223 | return first_error; | |
771ed689 CM |
5224 | return ret; |
5225 | } | |
d1310b2e | 5226 | |
8ae225a8 | 5227 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
5228 | struct writeback_control *wbc) |
5229 | { | |
35156d85 | 5230 | struct inode *inode = mapping->host; |
d1310b2e CM |
5231 | int ret = 0; |
5232 | struct extent_page_data epd = { | |
390ed29b | 5233 | .bio_ctrl = { 0 }, |
771ed689 | 5234 | .extent_locked = 0, |
ffbd517d | 5235 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
5236 | }; |
5237 | ||
35156d85 JT |
5238 | /* |
5239 | * Allow only a single thread to do the reloc work in zoned mode to | |
5240 | * protect the write pointer updates. | |
5241 | */ | |
869f4cdc | 5242 | btrfs_zoned_data_reloc_lock(BTRFS_I(inode)); |
935db853 | 5243 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
5244 | ASSERT(ret <= 0); |
5245 | if (ret < 0) { | |
19ab78ca | 5246 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
a2a72fbd QW |
5247 | end_write_bio(&epd, ret); |
5248 | return ret; | |
5249 | } | |
c9583ada | 5250 | flush_write_bio(&epd); |
19ab78ca | 5251 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
d1310b2e CM |
5252 | return ret; |
5253 | } | |
d1310b2e | 5254 | |
ba206a02 | 5255 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 5256 | { |
390ed29b | 5257 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
67c9684f | 5258 | struct page *pagepool[16]; |
125bac01 | 5259 | struct extent_map *em_cached = NULL; |
808f80b4 | 5260 | u64 prev_em_start = (u64)-1; |
ba206a02 | 5261 | int nr; |
d1310b2e | 5262 | |
ba206a02 | 5263 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
5264 | u64 contig_start = readahead_pos(rac); |
5265 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 5266 | |
ba206a02 | 5267 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 5268 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 5269 | } |
67c9684f | 5270 | |
125bac01 MX |
5271 | if (em_cached) |
5272 | free_extent_map(em_cached); | |
5273 | ||
c9583ada | 5274 | if (bio_ctrl.bio) |
0f07003b | 5275 | submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.compress_type); |
d1310b2e | 5276 | } |
d1310b2e CM |
5277 | |
5278 | /* | |
895586eb MWO |
5279 | * basic invalidate_folio code, this waits on any locked or writeback |
5280 | * ranges corresponding to the folio, and then deletes any extent state | |
d1310b2e CM |
5281 | * records from the tree |
5282 | */ | |
895586eb MWO |
5283 | int extent_invalidate_folio(struct extent_io_tree *tree, |
5284 | struct folio *folio, size_t offset) | |
d1310b2e | 5285 | { |
2ac55d41 | 5286 | struct extent_state *cached_state = NULL; |
895586eb MWO |
5287 | u64 start = folio_pos(folio); |
5288 | u64 end = start + folio_size(folio) - 1; | |
5289 | size_t blocksize = folio->mapping->host->i_sb->s_blocksize; | |
d1310b2e | 5290 | |
829ddec9 QW |
5291 | /* This function is only called for the btree inode */ |
5292 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
5293 | ||
fda2832f | 5294 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
5295 | if (start > end) |
5296 | return 0; | |
5297 | ||
ff13db41 | 5298 | lock_extent_bits(tree, start, end, &cached_state); |
895586eb | 5299 | folio_wait_writeback(folio); |
829ddec9 QW |
5300 | |
5301 | /* | |
5302 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
5303 | * so here we only need to unlock the extent range to free any | |
5304 | * existing extent state. | |
5305 | */ | |
5306 | unlock_extent_cached(tree, start, end, &cached_state); | |
d1310b2e CM |
5307 | return 0; |
5308 | } | |
d1310b2e | 5309 | |
7b13b7b1 | 5310 | /* |
f913cff3 | 5311 | * a helper for release_folio, this tests for areas of the page that |
7b13b7b1 CM |
5312 | * are locked or under IO and drops the related state bits if it is safe |
5313 | * to drop the page. | |
5314 | */ | |
29c68b2d | 5315 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 5316 | struct page *page, gfp_t mask) |
7b13b7b1 | 5317 | { |
4eee4fa4 | 5318 | u64 start = page_offset(page); |
09cbfeaf | 5319 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
5320 | int ret = 1; |
5321 | ||
8882679e | 5322 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 5323 | ret = 0; |
8882679e | 5324 | } else { |
11ef160f | 5325 | /* |
2766ff61 FM |
5326 | * At this point we can safely clear everything except the |
5327 | * locked bit, the nodatasum bit and the delalloc new bit. | |
5328 | * The delalloc new bit will be cleared by ordered extent | |
5329 | * completion. | |
11ef160f | 5330 | */ |
66b0c887 | 5331 | ret = __clear_extent_bit(tree, start, end, |
2766ff61 FM |
5332 | ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW), |
5333 | 0, 0, NULL, mask, NULL); | |
e3f24cc5 CM |
5334 | |
5335 | /* if clear_extent_bit failed for enomem reasons, | |
5336 | * we can't allow the release to continue. | |
5337 | */ | |
5338 | if (ret < 0) | |
5339 | ret = 0; | |
5340 | else | |
5341 | ret = 1; | |
7b13b7b1 CM |
5342 | } |
5343 | return ret; | |
5344 | } | |
7b13b7b1 | 5345 | |
d1310b2e | 5346 | /* |
f913cff3 | 5347 | * a helper for release_folio. As long as there are no locked extents |
d1310b2e CM |
5348 | * in the range corresponding to the page, both state records and extent |
5349 | * map records are removed | |
5350 | */ | |
477a30ba | 5351 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
5352 | { |
5353 | struct extent_map *em; | |
4eee4fa4 | 5354 | u64 start = page_offset(page); |
09cbfeaf | 5355 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
5356 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
5357 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
5358 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 5359 | |
d0164adc | 5360 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 5361 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 5362 | u64 len; |
70dec807 | 5363 | while (start <= end) { |
fbc2bd7e FM |
5364 | struct btrfs_fs_info *fs_info; |
5365 | u64 cur_gen; | |
5366 | ||
39b5637f | 5367 | len = end - start + 1; |
890871be | 5368 | write_lock(&map->lock); |
39b5637f | 5369 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 5370 | if (!em) { |
890871be | 5371 | write_unlock(&map->lock); |
70dec807 CM |
5372 | break; |
5373 | } | |
7f3c74fb CM |
5374 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
5375 | em->start != start) { | |
890871be | 5376 | write_unlock(&map->lock); |
70dec807 CM |
5377 | free_extent_map(em); |
5378 | break; | |
5379 | } | |
3d6448e6 FM |
5380 | if (test_range_bit(tree, em->start, |
5381 | extent_map_end(em) - 1, | |
5382 | EXTENT_LOCKED, 0, NULL)) | |
5383 | goto next; | |
5384 | /* | |
5385 | * If it's not in the list of modified extents, used | |
5386 | * by a fast fsync, we can remove it. If it's being | |
5387 | * logged we can safely remove it since fsync took an | |
5388 | * extra reference on the em. | |
5389 | */ | |
5390 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
5391 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
5392 | goto remove_em; | |
5393 | /* | |
5394 | * If it's in the list of modified extents, remove it | |
5395 | * only if its generation is older then the current one, | |
5396 | * in which case we don't need it for a fast fsync. | |
5397 | * Otherwise don't remove it, we could be racing with an | |
5398 | * ongoing fast fsync that could miss the new extent. | |
5399 | */ | |
5400 | fs_info = btrfs_inode->root->fs_info; | |
5401 | spin_lock(&fs_info->trans_lock); | |
5402 | cur_gen = fs_info->generation; | |
5403 | spin_unlock(&fs_info->trans_lock); | |
5404 | if (em->generation >= cur_gen) | |
5405 | goto next; | |
5406 | remove_em: | |
5e548b32 FM |
5407 | /* |
5408 | * We only remove extent maps that are not in the list of | |
5409 | * modified extents or that are in the list but with a | |
5410 | * generation lower then the current generation, so there | |
5411 | * is no need to set the full fsync flag on the inode (it | |
5412 | * hurts the fsync performance for workloads with a data | |
5413 | * size that exceeds or is close to the system's memory). | |
5414 | */ | |
fbc2bd7e FM |
5415 | remove_extent_mapping(map, em); |
5416 | /* once for the rb tree */ | |
5417 | free_extent_map(em); | |
3d6448e6 | 5418 | next: |
70dec807 | 5419 | start = extent_map_end(em); |
890871be | 5420 | write_unlock(&map->lock); |
70dec807 CM |
5421 | |
5422 | /* once for us */ | |
d1310b2e | 5423 | free_extent_map(em); |
9f47eb54 PM |
5424 | |
5425 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 5426 | } |
d1310b2e | 5427 | } |
29c68b2d | 5428 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 5429 | } |
d1310b2e | 5430 | |
ec29ed5b CM |
5431 | /* |
5432 | * helper function for fiemap, which doesn't want to see any holes. | |
5433 | * This maps until we find something past 'last' | |
5434 | */ | |
f1bbde8d | 5435 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 5436 | u64 offset, u64 last) |
ec29ed5b | 5437 | { |
f1bbde8d | 5438 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
5439 | struct extent_map *em; |
5440 | u64 len; | |
5441 | ||
5442 | if (offset >= last) | |
5443 | return NULL; | |
5444 | ||
67871254 | 5445 | while (1) { |
ec29ed5b CM |
5446 | len = last - offset; |
5447 | if (len == 0) | |
5448 | break; | |
fda2832f | 5449 | len = ALIGN(len, sectorsize); |
f1bbde8d | 5450 | em = btrfs_get_extent_fiemap(inode, offset, len); |
6b5b7a41 | 5451 | if (IS_ERR(em)) |
ec29ed5b CM |
5452 | return em; |
5453 | ||
5454 | /* if this isn't a hole return it */ | |
4a2d25cd | 5455 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 5456 | return em; |
ec29ed5b CM |
5457 | |
5458 | /* this is a hole, advance to the next extent */ | |
5459 | offset = extent_map_end(em); | |
5460 | free_extent_map(em); | |
5461 | if (offset >= last) | |
5462 | break; | |
5463 | } | |
5464 | return NULL; | |
5465 | } | |
5466 | ||
4751832d QW |
5467 | /* |
5468 | * To cache previous fiemap extent | |
5469 | * | |
5470 | * Will be used for merging fiemap extent | |
5471 | */ | |
5472 | struct fiemap_cache { | |
5473 | u64 offset; | |
5474 | u64 phys; | |
5475 | u64 len; | |
5476 | u32 flags; | |
5477 | bool cached; | |
5478 | }; | |
5479 | ||
5480 | /* | |
5481 | * Helper to submit fiemap extent. | |
5482 | * | |
5483 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
5484 | * @len and @flags with cached one. | |
5485 | * And only when we fails to merge, cached one will be submitted as | |
5486 | * fiemap extent. | |
5487 | * | |
5488 | * Return value is the same as fiemap_fill_next_extent(). | |
5489 | */ | |
5490 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
5491 | struct fiemap_cache *cache, | |
5492 | u64 offset, u64 phys, u64 len, u32 flags) | |
5493 | { | |
5494 | int ret = 0; | |
5495 | ||
5496 | if (!cache->cached) | |
5497 | goto assign; | |
5498 | ||
5499 | /* | |
5500 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 5501 | * fiemap extent won't overlap with cached one. |
4751832d QW |
5502 | * Not recoverable. |
5503 | * | |
5504 | * NOTE: Physical address can overlap, due to compression | |
5505 | */ | |
5506 | if (cache->offset + cache->len > offset) { | |
5507 | WARN_ON(1); | |
5508 | return -EINVAL; | |
5509 | } | |
5510 | ||
5511 | /* | |
5512 | * Only merges fiemap extents if | |
5513 | * 1) Their logical addresses are continuous | |
5514 | * | |
5515 | * 2) Their physical addresses are continuous | |
5516 | * So truly compressed (physical size smaller than logical size) | |
5517 | * extents won't get merged with each other | |
5518 | * | |
5519 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
5520 | * So regular extent won't get merged with prealloc extent | |
5521 | */ | |
5522 | if (cache->offset + cache->len == offset && | |
5523 | cache->phys + cache->len == phys && | |
5524 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
5525 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
5526 | cache->len += len; | |
5527 | cache->flags |= flags; | |
5528 | goto try_submit_last; | |
5529 | } | |
5530 | ||
5531 | /* Not mergeable, need to submit cached one */ | |
5532 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
5533 | cache->len, cache->flags); | |
5534 | cache->cached = false; | |
5535 | if (ret) | |
5536 | return ret; | |
5537 | assign: | |
5538 | cache->cached = true; | |
5539 | cache->offset = offset; | |
5540 | cache->phys = phys; | |
5541 | cache->len = len; | |
5542 | cache->flags = flags; | |
5543 | try_submit_last: | |
5544 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
5545 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
5546 | cache->phys, cache->len, cache->flags); | |
5547 | cache->cached = false; | |
5548 | } | |
5549 | return ret; | |
5550 | } | |
5551 | ||
5552 | /* | |
848c23b7 | 5553 | * Emit last fiemap cache |
4751832d | 5554 | * |
848c23b7 QW |
5555 | * The last fiemap cache may still be cached in the following case: |
5556 | * 0 4k 8k | |
5557 | * |<- Fiemap range ->| | |
5558 | * |<------------ First extent ----------->| | |
5559 | * | |
5560 | * In this case, the first extent range will be cached but not emitted. | |
5561 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 5562 | */ |
5c5aff98 | 5563 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 5564 | struct fiemap_cache *cache) |
4751832d QW |
5565 | { |
5566 | int ret; | |
5567 | ||
5568 | if (!cache->cached) | |
5569 | return 0; | |
5570 | ||
4751832d QW |
5571 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
5572 | cache->len, cache->flags); | |
5573 | cache->cached = false; | |
5574 | if (ret > 0) | |
5575 | ret = 0; | |
5576 | return ret; | |
5577 | } | |
5578 | ||
facee0a0 | 5579 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 5580 | u64 start, u64 len) |
1506fcc8 | 5581 | { |
975f84fe | 5582 | int ret = 0; |
15c7745c | 5583 | u64 off; |
1506fcc8 YS |
5584 | u64 max = start + len; |
5585 | u32 flags = 0; | |
975f84fe JB |
5586 | u32 found_type; |
5587 | u64 last; | |
ec29ed5b | 5588 | u64 last_for_get_extent = 0; |
1506fcc8 | 5589 | u64 disko = 0; |
facee0a0 | 5590 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 5591 | struct btrfs_key found_key; |
1506fcc8 | 5592 | struct extent_map *em = NULL; |
2ac55d41 | 5593 | struct extent_state *cached_state = NULL; |
975f84fe | 5594 | struct btrfs_path *path; |
facee0a0 | 5595 | struct btrfs_root *root = inode->root; |
4751832d | 5596 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
5597 | struct ulist *roots; |
5598 | struct ulist *tmp_ulist; | |
1506fcc8 | 5599 | int end = 0; |
ec29ed5b CM |
5600 | u64 em_start = 0; |
5601 | u64 em_len = 0; | |
5602 | u64 em_end = 0; | |
1506fcc8 YS |
5603 | |
5604 | if (len == 0) | |
5605 | return -EINVAL; | |
5606 | ||
975f84fe JB |
5607 | path = btrfs_alloc_path(); |
5608 | if (!path) | |
5609 | return -ENOMEM; | |
975f84fe | 5610 | |
5911c8fe DS |
5611 | roots = ulist_alloc(GFP_KERNEL); |
5612 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
5613 | if (!roots || !tmp_ulist) { | |
5614 | ret = -ENOMEM; | |
5615 | goto out_free_ulist; | |
5616 | } | |
5617 | ||
15c7745c BB |
5618 | /* |
5619 | * We can't initialize that to 'start' as this could miss extents due | |
5620 | * to extent item merging | |
5621 | */ | |
5622 | off = 0; | |
facee0a0 NB |
5623 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
5624 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 5625 | |
ec29ed5b CM |
5626 | /* |
5627 | * lookup the last file extent. We're not using i_size here | |
5628 | * because there might be preallocation past i_size | |
5629 | */ | |
facee0a0 NB |
5630 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
5631 | 0); | |
975f84fe | 5632 | if (ret < 0) { |
5911c8fe | 5633 | goto out_free_ulist; |
2d324f59 LB |
5634 | } else { |
5635 | WARN_ON(!ret); | |
5636 | if (ret == 1) | |
5637 | ret = 0; | |
975f84fe | 5638 | } |
2d324f59 | 5639 | |
975f84fe | 5640 | path->slots[0]--; |
975f84fe | 5641 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 5642 | found_type = found_key.type; |
975f84fe | 5643 | |
ec29ed5b | 5644 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 5645 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 5646 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
5647 | /* have to trust i_size as the end */ |
5648 | last = (u64)-1; | |
5649 | last_for_get_extent = isize; | |
5650 | } else { | |
5651 | /* | |
5652 | * remember the start of the last extent. There are a | |
5653 | * bunch of different factors that go into the length of the | |
5654 | * extent, so its much less complex to remember where it started | |
5655 | */ | |
5656 | last = found_key.offset; | |
5657 | last_for_get_extent = last + 1; | |
975f84fe | 5658 | } |
fe09e16c | 5659 | btrfs_release_path(path); |
975f84fe | 5660 | |
ec29ed5b CM |
5661 | /* |
5662 | * we might have some extents allocated but more delalloc past those | |
5663 | * extents. so, we trust isize unless the start of the last extent is | |
5664 | * beyond isize | |
5665 | */ | |
5666 | if (last < isize) { | |
5667 | last = (u64)-1; | |
5668 | last_for_get_extent = isize; | |
5669 | } | |
5670 | ||
facee0a0 | 5671 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 5672 | &cached_state); |
ec29ed5b | 5673 | |
facee0a0 | 5674 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
5675 | if (!em) |
5676 | goto out; | |
5677 | if (IS_ERR(em)) { | |
5678 | ret = PTR_ERR(em); | |
5679 | goto out; | |
5680 | } | |
975f84fe | 5681 | |
1506fcc8 | 5682 | while (!end) { |
b76bb701 | 5683 | u64 offset_in_extent = 0; |
ea8efc74 CM |
5684 | |
5685 | /* break if the extent we found is outside the range */ | |
5686 | if (em->start >= max || extent_map_end(em) < off) | |
5687 | break; | |
5688 | ||
5689 | /* | |
5690 | * get_extent may return an extent that starts before our | |
5691 | * requested range. We have to make sure the ranges | |
5692 | * we return to fiemap always move forward and don't | |
5693 | * overlap, so adjust the offsets here | |
5694 | */ | |
5695 | em_start = max(em->start, off); | |
1506fcc8 | 5696 | |
ea8efc74 CM |
5697 | /* |
5698 | * record the offset from the start of the extent | |
b76bb701 JB |
5699 | * for adjusting the disk offset below. Only do this if the |
5700 | * extent isn't compressed since our in ram offset may be past | |
5701 | * what we have actually allocated on disk. | |
ea8efc74 | 5702 | */ |
b76bb701 JB |
5703 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
5704 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 5705 | em_end = extent_map_end(em); |
ea8efc74 | 5706 | em_len = em_end - em_start; |
1506fcc8 | 5707 | flags = 0; |
f0986318 FM |
5708 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
5709 | disko = em->block_start + offset_in_extent; | |
5710 | else | |
5711 | disko = 0; | |
1506fcc8 | 5712 | |
ea8efc74 CM |
5713 | /* |
5714 | * bump off for our next call to get_extent | |
5715 | */ | |
5716 | off = extent_map_end(em); | |
5717 | if (off >= max) | |
5718 | end = 1; | |
5719 | ||
93dbfad7 | 5720 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
5721 | end = 1; |
5722 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 5723 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
5724 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
5725 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 5726 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
5727 | flags |= (FIEMAP_EXTENT_DELALLOC | |
5728 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
5729 | } else if (fieinfo->fi_extents_max) { |
5730 | u64 bytenr = em->block_start - | |
5731 | (em->start - em->orig_start); | |
fe09e16c | 5732 | |
fe09e16c LB |
5733 | /* |
5734 | * As btrfs supports shared space, this information | |
5735 | * can be exported to userspace tools via | |
dc046b10 JB |
5736 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
5737 | * then we're just getting a count and we can skip the | |
5738 | * lookup stuff. | |
fe09e16c | 5739 | */ |
facee0a0 | 5740 | ret = btrfs_check_shared(root, btrfs_ino(inode), |
5911c8fe | 5741 | bytenr, roots, tmp_ulist); |
dc046b10 | 5742 | if (ret < 0) |
fe09e16c | 5743 | goto out_free; |
dc046b10 | 5744 | if (ret) |
fe09e16c | 5745 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 5746 | ret = 0; |
1506fcc8 YS |
5747 | } |
5748 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
5749 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
5750 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
5751 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 5752 | |
1506fcc8 YS |
5753 | free_extent_map(em); |
5754 | em = NULL; | |
ec29ed5b CM |
5755 | if ((em_start >= last) || em_len == (u64)-1 || |
5756 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
5757 | flags |= FIEMAP_EXTENT_LAST; |
5758 | end = 1; | |
5759 | } | |
5760 | ||
ec29ed5b | 5761 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 5762 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
5763 | if (IS_ERR(em)) { |
5764 | ret = PTR_ERR(em); | |
5765 | goto out; | |
5766 | } | |
5767 | if (!em) { | |
975f84fe JB |
5768 | flags |= FIEMAP_EXTENT_LAST; |
5769 | end = 1; | |
5770 | } | |
4751832d QW |
5771 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
5772 | em_len, flags); | |
26e726af CS |
5773 | if (ret) { |
5774 | if (ret == 1) | |
5775 | ret = 0; | |
ec29ed5b | 5776 | goto out_free; |
26e726af | 5777 | } |
1506fcc8 YS |
5778 | } |
5779 | out_free: | |
4751832d | 5780 | if (!ret) |
5c5aff98 | 5781 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
5782 | free_extent_map(em); |
5783 | out: | |
facee0a0 | 5784 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 5785 | &cached_state); |
5911c8fe DS |
5786 | |
5787 | out_free_ulist: | |
e02d48ea | 5788 | btrfs_free_path(path); |
5911c8fe DS |
5789 | ulist_free(roots); |
5790 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
5791 | return ret; |
5792 | } | |
5793 | ||
727011e0 CM |
5794 | static void __free_extent_buffer(struct extent_buffer *eb) |
5795 | { | |
727011e0 CM |
5796 | kmem_cache_free(extent_buffer_cache, eb); |
5797 | } | |
5798 | ||
2b48966a | 5799 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
5800 | { |
5801 | return (atomic_read(&eb->io_pages) || | |
5802 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
5803 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5804 | } | |
5805 | ||
8ff8466d | 5806 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 5807 | { |
8ff8466d | 5808 | struct btrfs_subpage *subpage; |
db7f3436 | 5809 | |
8ff8466d | 5810 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 5811 | |
8ff8466d QW |
5812 | if (PagePrivate(page)) { |
5813 | subpage = (struct btrfs_subpage *)page->private; | |
5814 | if (atomic_read(&subpage->eb_refs)) | |
5815 | return true; | |
3d078efa QW |
5816 | /* |
5817 | * Even there is no eb refs here, we may still have | |
5818 | * end_page_read() call relying on page::private. | |
5819 | */ | |
5820 | if (atomic_read(&subpage->readers)) | |
5821 | return true; | |
8ff8466d QW |
5822 | } |
5823 | return false; | |
5824 | } | |
db7f3436 | 5825 | |
8ff8466d QW |
5826 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
5827 | { | |
5828 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5829 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
5830 | ||
5831 | /* | |
5832 | * For mapped eb, we're going to change the page private, which should | |
5833 | * be done under the private_lock. | |
5834 | */ | |
5835 | if (mapped) | |
5836 | spin_lock(&page->mapping->private_lock); | |
5837 | ||
5838 | if (!PagePrivate(page)) { | |
5d2361db | 5839 | if (mapped) |
8ff8466d QW |
5840 | spin_unlock(&page->mapping->private_lock); |
5841 | return; | |
5842 | } | |
5843 | ||
fbca46eb | 5844 | if (fs_info->nodesize >= PAGE_SIZE) { |
5d2361db FL |
5845 | /* |
5846 | * We do this since we'll remove the pages after we've | |
5847 | * removed the eb from the radix tree, so we could race | |
5848 | * and have this page now attached to the new eb. So | |
5849 | * only clear page_private if it's still connected to | |
5850 | * this eb. | |
5851 | */ | |
5852 | if (PagePrivate(page) && | |
5853 | page->private == (unsigned long)eb) { | |
5854 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5855 | BUG_ON(PageDirty(page)); | |
5856 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 5857 | /* |
5d2361db FL |
5858 | * We need to make sure we haven't be attached |
5859 | * to a new eb. | |
db7f3436 | 5860 | */ |
d1b89bc0 | 5861 | detach_page_private(page); |
db7f3436 | 5862 | } |
5d2361db FL |
5863 | if (mapped) |
5864 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
5865 | return; |
5866 | } | |
5867 | ||
5868 | /* | |
5869 | * For subpage, we can have dummy eb with page private. In this case, | |
5870 | * we can directly detach the private as such page is only attached to | |
5871 | * one dummy eb, no sharing. | |
5872 | */ | |
5873 | if (!mapped) { | |
5874 | btrfs_detach_subpage(fs_info, page); | |
5875 | return; | |
5876 | } | |
5877 | ||
5878 | btrfs_page_dec_eb_refs(fs_info, page); | |
5879 | ||
5880 | /* | |
5881 | * We can only detach the page private if there are no other ebs in the | |
3d078efa | 5882 | * page range and no unfinished IO. |
8ff8466d QW |
5883 | */ |
5884 | if (!page_range_has_eb(fs_info, page)) | |
5885 | btrfs_detach_subpage(fs_info, page); | |
5886 | ||
5887 | spin_unlock(&page->mapping->private_lock); | |
5888 | } | |
5889 | ||
5890 | /* Release all pages attached to the extent buffer */ | |
5891 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
5892 | { | |
5893 | int i; | |
5894 | int num_pages; | |
5895 | ||
5896 | ASSERT(!extent_buffer_under_io(eb)); | |
5897 | ||
5898 | num_pages = num_extent_pages(eb); | |
5899 | for (i = 0; i < num_pages; i++) { | |
5900 | struct page *page = eb->pages[i]; | |
5901 | ||
5902 | if (!page) | |
5903 | continue; | |
5904 | ||
5905 | detach_extent_buffer_page(eb, page); | |
5d2361db | 5906 | |
01327610 | 5907 | /* One for when we allocated the page */ |
09cbfeaf | 5908 | put_page(page); |
d64766fd | 5909 | } |
db7f3436 JB |
5910 | } |
5911 | ||
5912 | /* | |
5913 | * Helper for releasing the extent buffer. | |
5914 | */ | |
5915 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
5916 | { | |
55ac0139 | 5917 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 5918 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
5919 | __free_extent_buffer(eb); |
5920 | } | |
5921 | ||
f28491e0 JB |
5922 | static struct extent_buffer * |
5923 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 5924 | unsigned long len) |
d1310b2e CM |
5925 | { |
5926 | struct extent_buffer *eb = NULL; | |
5927 | ||
d1b5c567 | 5928 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
5929 | eb->start = start; |
5930 | eb->len = len; | |
f28491e0 | 5931 | eb->fs_info = fs_info; |
815a51c7 | 5932 | eb->bflags = 0; |
196d59ab | 5933 | init_rwsem(&eb->lock); |
b4ce94de | 5934 | |
3fd63727 JB |
5935 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5936 | &fs_info->allocated_ebs); | |
d3575156 | 5937 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 5938 | |
3083ee2e | 5939 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5940 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5941 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5942 | |
deb67895 | 5943 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
5944 | |
5945 | return eb; | |
5946 | } | |
5947 | ||
2b48966a | 5948 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 5949 | { |
cc5e31a4 | 5950 | int i; |
815a51c7 | 5951 | struct extent_buffer *new; |
cc5e31a4 | 5952 | int num_pages = num_extent_pages(src); |
dd137dd1 | 5953 | int ret; |
815a51c7 | 5954 | |
3f556f78 | 5955 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5956 | if (new == NULL) |
5957 | return NULL; | |
5958 | ||
62c053fb QW |
5959 | /* |
5960 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
5961 | * btrfs_release_extent_buffer() have different behavior for | |
5962 | * UNMAPPED subpage extent buffer. | |
5963 | */ | |
5964 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
5965 | ||
dd137dd1 STD |
5966 | memset(new->pages, 0, sizeof(*new->pages) * num_pages); |
5967 | ret = btrfs_alloc_page_array(num_pages, new->pages); | |
5968 | if (ret) { | |
5969 | btrfs_release_extent_buffer(new); | |
5970 | return NULL; | |
5971 | } | |
5972 | ||
815a51c7 | 5973 | for (i = 0; i < num_pages; i++) { |
760f991f | 5974 | int ret; |
dd137dd1 | 5975 | struct page *p = new->pages[i]; |
760f991f | 5976 | |
760f991f QW |
5977 | ret = attach_extent_buffer_page(new, p, NULL); |
5978 | if (ret < 0) { | |
760f991f QW |
5979 | btrfs_release_extent_buffer(new); |
5980 | return NULL; | |
5981 | } | |
815a51c7 | 5982 | WARN_ON(PageDirty(p)); |
fba1acf9 | 5983 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 5984 | } |
92d83e94 | 5985 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
5986 | |
5987 | return new; | |
5988 | } | |
5989 | ||
0f331229 OS |
5990 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5991 | u64 start, unsigned long len) | |
815a51c7 JS |
5992 | { |
5993 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5994 | int num_pages; |
5995 | int i; | |
dd137dd1 | 5996 | int ret; |
815a51c7 | 5997 | |
3f556f78 | 5998 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5999 | if (!eb) |
6000 | return NULL; | |
6001 | ||
65ad0104 | 6002 | num_pages = num_extent_pages(eb); |
dd137dd1 STD |
6003 | ret = btrfs_alloc_page_array(num_pages, eb->pages); |
6004 | if (ret) | |
6005 | goto err; | |
6006 | ||
815a51c7 | 6007 | for (i = 0; i < num_pages; i++) { |
dd137dd1 | 6008 | struct page *p = eb->pages[i]; |
09bc1f0f | 6009 | |
dd137dd1 | 6010 | ret = attach_extent_buffer_page(eb, p, NULL); |
09bc1f0f QW |
6011 | if (ret < 0) |
6012 | goto err; | |
815a51c7 | 6013 | } |
dd137dd1 | 6014 | |
815a51c7 JS |
6015 | set_extent_buffer_uptodate(eb); |
6016 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 6017 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
6018 | |
6019 | return eb; | |
6020 | err: | |
dd137dd1 STD |
6021 | for (i = 0; i < num_pages; i++) { |
6022 | if (eb->pages[i]) { | |
6023 | detach_extent_buffer_page(eb, eb->pages[i]); | |
6024 | __free_page(eb->pages[i]); | |
6025 | } | |
09bc1f0f | 6026 | } |
815a51c7 JS |
6027 | __free_extent_buffer(eb); |
6028 | return NULL; | |
6029 | } | |
6030 | ||
0f331229 | 6031 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 6032 | u64 start) |
0f331229 | 6033 | { |
da17066c | 6034 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
6035 | } |
6036 | ||
0b32f4bb JB |
6037 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
6038 | { | |
242e18c7 | 6039 | int refs; |
6bf9cd2e BB |
6040 | /* |
6041 | * The TREE_REF bit is first set when the extent_buffer is added | |
6042 | * to the radix tree. It is also reset, if unset, when a new reference | |
6043 | * is created by find_extent_buffer. | |
0b32f4bb | 6044 | * |
6bf9cd2e BB |
6045 | * It is only cleared in two cases: freeing the last non-tree |
6046 | * reference to the extent_buffer when its STALE bit is set or | |
f913cff3 | 6047 | * calling release_folio when the tree reference is the only reference. |
0b32f4bb | 6048 | * |
6bf9cd2e | 6049 | * In both cases, care is taken to ensure that the extent_buffer's |
f913cff3 | 6050 | * pages are not under io. However, release_folio can be concurrently |
6bf9cd2e BB |
6051 | * called with creating new references, which is prone to race |
6052 | * conditions between the calls to check_buffer_tree_ref in those | |
6053 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 6054 | * |
6bf9cd2e BB |
6055 | * The actual lifetime of the extent_buffer in the radix tree is |
6056 | * adequately protected by the refcount, but the TREE_REF bit and | |
6057 | * its corresponding reference are not. To protect against this | |
6058 | * class of races, we call check_buffer_tree_ref from the codepaths | |
6059 | * which trigger io after they set eb->io_pages. Note that once io is | |
6060 | * initiated, TREE_REF can no longer be cleared, so that is the | |
6061 | * moment at which any such race is best fixed. | |
0b32f4bb | 6062 | */ |
242e18c7 CM |
6063 | refs = atomic_read(&eb->refs); |
6064 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
6065 | return; | |
6066 | ||
594831c4 JB |
6067 | spin_lock(&eb->refs_lock); |
6068 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 6069 | atomic_inc(&eb->refs); |
594831c4 | 6070 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
6071 | } |
6072 | ||
2457aec6 MG |
6073 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
6074 | struct page *accessed) | |
5df4235e | 6075 | { |
cc5e31a4 | 6076 | int num_pages, i; |
5df4235e | 6077 | |
0b32f4bb JB |
6078 | check_buffer_tree_ref(eb); |
6079 | ||
65ad0104 | 6080 | num_pages = num_extent_pages(eb); |
5df4235e | 6081 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
6082 | struct page *p = eb->pages[i]; |
6083 | ||
2457aec6 MG |
6084 | if (p != accessed) |
6085 | mark_page_accessed(p); | |
5df4235e JB |
6086 | } |
6087 | } | |
6088 | ||
f28491e0 JB |
6089 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
6090 | u64 start) | |
452c75c3 CS |
6091 | { |
6092 | struct extent_buffer *eb; | |
6093 | ||
2f3186d8 QW |
6094 | eb = find_extent_buffer_nolock(fs_info, start); |
6095 | if (!eb) | |
6096 | return NULL; | |
6097 | /* | |
6098 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
6099 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
6100 | * another task running free_extent_buffer() might have seen that flag | |
6101 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
6102 | * writeback flags not set) and it's still in the tree (flag | |
6103 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
6104 | * decrementing the extent buffer's reference count twice. So here we | |
6105 | * could race and increment the eb's reference count, clear its stale | |
6106 | * flag, mark it as dirty and drop our reference before the other task | |
6107 | * finishes executing free_extent_buffer, which would later result in | |
6108 | * an attempt to free an extent buffer that is dirty. | |
6109 | */ | |
6110 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
6111 | spin_lock(&eb->refs_lock); | |
6112 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 6113 | } |
2f3186d8 QW |
6114 | mark_extent_buffer_accessed(eb, NULL); |
6115 | return eb; | |
452c75c3 CS |
6116 | } |
6117 | ||
faa2dbf0 JB |
6118 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
6119 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 6120 | u64 start) |
faa2dbf0 JB |
6121 | { |
6122 | struct extent_buffer *eb, *exists = NULL; | |
6123 | int ret; | |
6124 | ||
6125 | eb = find_extent_buffer(fs_info, start); | |
6126 | if (eb) | |
6127 | return eb; | |
da17066c | 6128 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 6129 | if (!eb) |
b6293c82 | 6130 | return ERR_PTR(-ENOMEM); |
faa2dbf0 | 6131 | eb->fs_info = fs_info; |
8ee92268 GN |
6132 | |
6133 | do { | |
6134 | ret = xa_insert(&fs_info->extent_buffers, | |
6135 | start >> fs_info->sectorsize_bits, | |
6136 | eb, GFP_NOFS); | |
6137 | if (ret == -ENOMEM) { | |
6138 | exists = ERR_PTR(ret); | |
faa2dbf0 | 6139 | goto free_eb; |
8ee92268 GN |
6140 | } |
6141 | if (ret == -EBUSY) { | |
6142 | exists = find_extent_buffer(fs_info, start); | |
6143 | if (exists) | |
6144 | goto free_eb; | |
6145 | } | |
6146 | } while (ret); | |
6147 | ||
faa2dbf0 JB |
6148 | check_buffer_tree_ref(eb); |
6149 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
6150 | ||
faa2dbf0 JB |
6151 | return eb; |
6152 | free_eb: | |
6153 | btrfs_release_extent_buffer(eb); | |
6154 | return exists; | |
6155 | } | |
6156 | #endif | |
6157 | ||
81982210 QW |
6158 | static struct extent_buffer *grab_extent_buffer( |
6159 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
6160 | { |
6161 | struct extent_buffer *exists; | |
6162 | ||
81982210 QW |
6163 | /* |
6164 | * For subpage case, we completely rely on radix tree to ensure we | |
6165 | * don't try to insert two ebs for the same bytenr. So here we always | |
6166 | * return NULL and just continue. | |
6167 | */ | |
fbca46eb | 6168 | if (fs_info->nodesize < PAGE_SIZE) |
81982210 QW |
6169 | return NULL; |
6170 | ||
c0f0a9e7 QW |
6171 | /* Page not yet attached to an extent buffer */ |
6172 | if (!PagePrivate(page)) | |
6173 | return NULL; | |
6174 | ||
6175 | /* | |
6176 | * We could have already allocated an eb for this page and attached one | |
6177 | * so lets see if we can get a ref on the existing eb, and if we can we | |
6178 | * know it's good and we can just return that one, else we know we can | |
6179 | * just overwrite page->private. | |
6180 | */ | |
6181 | exists = (struct extent_buffer *)page->private; | |
6182 | if (atomic_inc_not_zero(&exists->refs)) | |
6183 | return exists; | |
6184 | ||
6185 | WARN_ON(PageDirty(page)); | |
6186 | detach_page_private(page); | |
6187 | return NULL; | |
6188 | } | |
6189 | ||
fbca46eb QW |
6190 | static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start) |
6191 | { | |
6192 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { | |
6193 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
6194 | return -EINVAL; | |
6195 | } | |
6196 | ||
6197 | if (fs_info->nodesize < PAGE_SIZE && | |
6198 | offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) { | |
6199 | btrfs_err(fs_info, | |
6200 | "tree block crosses page boundary, start %llu nodesize %u", | |
6201 | start, fs_info->nodesize); | |
6202 | return -EINVAL; | |
6203 | } | |
6204 | if (fs_info->nodesize >= PAGE_SIZE && | |
6205 | !IS_ALIGNED(start, PAGE_SIZE)) { | |
6206 | btrfs_err(fs_info, | |
6207 | "tree block is not page aligned, start %llu nodesize %u", | |
6208 | start, fs_info->nodesize); | |
6209 | return -EINVAL; | |
6210 | } | |
6211 | return 0; | |
6212 | } | |
6213 | ||
f28491e0 | 6214 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 6215 | u64 start, u64 owner_root, int level) |
d1310b2e | 6216 | { |
da17066c | 6217 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
6218 | int num_pages; |
6219 | int i; | |
09cbfeaf | 6220 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 6221 | struct extent_buffer *eb; |
6af118ce | 6222 | struct extent_buffer *exists = NULL; |
d1310b2e | 6223 | struct page *p; |
f28491e0 | 6224 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 6225 | int uptodate = 1; |
19fe0a8b | 6226 | int ret; |
d1310b2e | 6227 | |
fbca46eb | 6228 | if (check_eb_alignment(fs_info, start)) |
c871b0f2 | 6229 | return ERR_PTR(-EINVAL); |
c871b0f2 | 6230 | |
e9306ad4 QW |
6231 | #if BITS_PER_LONG == 32 |
6232 | if (start >= MAX_LFS_FILESIZE) { | |
6233 | btrfs_err_rl(fs_info, | |
6234 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
6235 | btrfs_err_32bit_limit(fs_info); | |
6236 | return ERR_PTR(-EOVERFLOW); | |
6237 | } | |
6238 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
6239 | btrfs_warn_32bit_limit(fs_info); | |
6240 | #endif | |
6241 | ||
f28491e0 | 6242 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 6243 | if (eb) |
6af118ce | 6244 | return eb; |
6af118ce | 6245 | |
23d79d81 | 6246 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 6247 | if (!eb) |
c871b0f2 | 6248 | return ERR_PTR(-ENOMEM); |
e114c545 | 6249 | btrfs_set_buffer_lockdep_class(owner_root, eb, level); |
d1310b2e | 6250 | |
65ad0104 | 6251 | num_pages = num_extent_pages(eb); |
727011e0 | 6252 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
6253 | struct btrfs_subpage *prealloc = NULL; |
6254 | ||
d1b5c567 | 6255 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
6256 | if (!p) { |
6257 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 6258 | goto free_eb; |
c871b0f2 | 6259 | } |
4f2de97a | 6260 | |
760f991f QW |
6261 | /* |
6262 | * Preallocate page->private for subpage case, so that we won't | |
6263 | * allocate memory with private_lock hold. The memory will be | |
6264 | * freed by attach_extent_buffer_page() or freed manually if | |
6265 | * we exit earlier. | |
6266 | * | |
6267 | * Although we have ensured one subpage eb can only have one | |
6268 | * page, but it may change in the future for 16K page size | |
6269 | * support, so we still preallocate the memory in the loop. | |
6270 | */ | |
fbca46eb | 6271 | if (fs_info->nodesize < PAGE_SIZE) { |
651fb419 QW |
6272 | prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA); |
6273 | if (IS_ERR(prealloc)) { | |
6274 | ret = PTR_ERR(prealloc); | |
fdf250db QW |
6275 | unlock_page(p); |
6276 | put_page(p); | |
6277 | exists = ERR_PTR(ret); | |
6278 | goto free_eb; | |
6279 | } | |
760f991f QW |
6280 | } |
6281 | ||
4f2de97a | 6282 | spin_lock(&mapping->private_lock); |
81982210 | 6283 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
6284 | if (exists) { |
6285 | spin_unlock(&mapping->private_lock); | |
6286 | unlock_page(p); | |
6287 | put_page(p); | |
6288 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 6289 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 6290 | goto free_eb; |
d1310b2e | 6291 | } |
760f991f QW |
6292 | /* Should not fail, as we have preallocated the memory */ |
6293 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
6294 | ASSERT(!ret); | |
8ff8466d QW |
6295 | /* |
6296 | * To inform we have extra eb under allocation, so that | |
6297 | * detach_extent_buffer_page() won't release the page private | |
6298 | * when the eb hasn't yet been inserted into radix tree. | |
6299 | * | |
6300 | * The ref will be decreased when the eb released the page, in | |
6301 | * detach_extent_buffer_page(). | |
6302 | * Thus needs no special handling in error path. | |
6303 | */ | |
6304 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 6305 | spin_unlock(&mapping->private_lock); |
760f991f | 6306 | |
1e5eb3d6 | 6307 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 6308 | eb->pages[i] = p; |
d1310b2e CM |
6309 | if (!PageUptodate(p)) |
6310 | uptodate = 0; | |
eb14ab8e CM |
6311 | |
6312 | /* | |
b16d011e NB |
6313 | * We can't unlock the pages just yet since the extent buffer |
6314 | * hasn't been properly inserted in the radix tree, this | |
f913cff3 | 6315 | * opens a race with btree_release_folio which can free a page |
b16d011e NB |
6316 | * while we are still filling in all pages for the buffer and |
6317 | * we could crash. | |
eb14ab8e | 6318 | */ |
d1310b2e CM |
6319 | } |
6320 | if (uptodate) | |
b4ce94de | 6321 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
8ee92268 GN |
6322 | |
6323 | do { | |
6324 | ret = xa_insert(&fs_info->extent_buffers, | |
6325 | start >> fs_info->sectorsize_bits, | |
6326 | eb, GFP_NOFS); | |
6327 | if (ret == -ENOMEM) { | |
6328 | exists = ERR_PTR(ret); | |
452c75c3 | 6329 | goto free_eb; |
8ee92268 GN |
6330 | } |
6331 | if (ret == -EBUSY) { | |
6332 | exists = find_extent_buffer(fs_info, start); | |
6333 | if (exists) | |
6334 | goto free_eb; | |
6335 | } | |
6336 | } while (ret); | |
6337 | ||
6af118ce | 6338 | /* add one reference for the tree */ |
0b32f4bb | 6339 | check_buffer_tree_ref(eb); |
34b41ace | 6340 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
6341 | |
6342 | /* | |
b16d011e | 6343 | * Now it's safe to unlock the pages because any calls to |
f913cff3 | 6344 | * btree_release_folio will correctly detect that a page belongs to a |
b16d011e | 6345 | * live buffer and won't free them prematurely. |
eb14ab8e | 6346 | */ |
28187ae5 NB |
6347 | for (i = 0; i < num_pages; i++) |
6348 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
6349 | return eb; |
6350 | ||
6af118ce | 6351 | free_eb: |
5ca64f45 | 6352 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
6353 | for (i = 0; i < num_pages; i++) { |
6354 | if (eb->pages[i]) | |
6355 | unlock_page(eb->pages[i]); | |
6356 | } | |
eb14ab8e | 6357 | |
897ca6e9 | 6358 | btrfs_release_extent_buffer(eb); |
6af118ce | 6359 | return exists; |
d1310b2e | 6360 | } |
d1310b2e | 6361 | |
3083ee2e JB |
6362 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
6363 | { | |
6364 | struct extent_buffer *eb = | |
6365 | container_of(head, struct extent_buffer, rcu_head); | |
6366 | ||
6367 | __free_extent_buffer(eb); | |
6368 | } | |
6369 | ||
f7a52a40 | 6370 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 6371 | __releases(&eb->refs_lock) |
3083ee2e | 6372 | { |
07e21c4d NB |
6373 | lockdep_assert_held(&eb->refs_lock); |
6374 | ||
3083ee2e JB |
6375 | WARN_ON(atomic_read(&eb->refs) == 0); |
6376 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 6377 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 6378 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 6379 | |
815a51c7 | 6380 | spin_unlock(&eb->refs_lock); |
3083ee2e | 6381 | |
8ee92268 GN |
6382 | xa_erase(&fs_info->extent_buffers, |
6383 | eb->start >> fs_info->sectorsize_bits); | |
34b41ace JB |
6384 | } else { |
6385 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 6386 | } |
3083ee2e | 6387 | |
8c38938c | 6388 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 6389 | /* Should be safe to release our pages at this point */ |
55ac0139 | 6390 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 6391 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 6392 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
6393 | __free_extent_buffer(eb); |
6394 | return 1; | |
6395 | } | |
6396 | #endif | |
3083ee2e | 6397 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 6398 | return 1; |
3083ee2e JB |
6399 | } |
6400 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
6401 | |
6402 | return 0; | |
3083ee2e JB |
6403 | } |
6404 | ||
d1310b2e CM |
6405 | void free_extent_buffer(struct extent_buffer *eb) |
6406 | { | |
242e18c7 CM |
6407 | int refs; |
6408 | int old; | |
d1310b2e CM |
6409 | if (!eb) |
6410 | return; | |
6411 | ||
242e18c7 CM |
6412 | while (1) { |
6413 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
6414 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
6415 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
6416 | refs == 1)) | |
242e18c7 CM |
6417 | break; |
6418 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
6419 | if (old == refs) | |
6420 | return; | |
6421 | } | |
6422 | ||
3083ee2e JB |
6423 | spin_lock(&eb->refs_lock); |
6424 | if (atomic_read(&eb->refs) == 2 && | |
6425 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 6426 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
6427 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6428 | atomic_dec(&eb->refs); | |
6429 | ||
6430 | /* | |
6431 | * I know this is terrible, but it's temporary until we stop tracking | |
6432 | * the uptodate bits and such for the extent buffers. | |
6433 | */ | |
f7a52a40 | 6434 | release_extent_buffer(eb); |
3083ee2e JB |
6435 | } |
6436 | ||
6437 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
6438 | { | |
6439 | if (!eb) | |
d1310b2e CM |
6440 | return; |
6441 | ||
3083ee2e JB |
6442 | spin_lock(&eb->refs_lock); |
6443 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
6444 | ||
0b32f4bb | 6445 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
6446 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6447 | atomic_dec(&eb->refs); | |
f7a52a40 | 6448 | release_extent_buffer(eb); |
d1310b2e | 6449 | } |
d1310b2e | 6450 | |
0d27797e QW |
6451 | static void btree_clear_page_dirty(struct page *page) |
6452 | { | |
6453 | ASSERT(PageDirty(page)); | |
6454 | ASSERT(PageLocked(page)); | |
6455 | clear_page_dirty_for_io(page); | |
6456 | xa_lock_irq(&page->mapping->i_pages); | |
6457 | if (!PageDirty(page)) | |
6458 | __xa_clear_mark(&page->mapping->i_pages, | |
6459 | page_index(page), PAGECACHE_TAG_DIRTY); | |
6460 | xa_unlock_irq(&page->mapping->i_pages); | |
6461 | } | |
6462 | ||
6463 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
6464 | { | |
6465 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6466 | struct page *page = eb->pages[0]; | |
6467 | bool last; | |
6468 | ||
6469 | /* btree_clear_page_dirty() needs page locked */ | |
6470 | lock_page(page); | |
6471 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
6472 | eb->len); | |
6473 | if (last) | |
6474 | btree_clear_page_dirty(page); | |
6475 | unlock_page(page); | |
6476 | WARN_ON(atomic_read(&eb->refs) == 0); | |
6477 | } | |
6478 | ||
2b48966a | 6479 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 6480 | { |
cc5e31a4 DS |
6481 | int i; |
6482 | int num_pages; | |
d1310b2e CM |
6483 | struct page *page; |
6484 | ||
fbca46eb | 6485 | if (eb->fs_info->nodesize < PAGE_SIZE) |
0d27797e QW |
6486 | return clear_subpage_extent_buffer_dirty(eb); |
6487 | ||
65ad0104 | 6488 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
6489 | |
6490 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 6491 | page = eb->pages[i]; |
b9473439 | 6492 | if (!PageDirty(page)) |
d2c3f4f6 | 6493 | continue; |
a61e6f29 | 6494 | lock_page(page); |
0d27797e | 6495 | btree_clear_page_dirty(page); |
bf0da8c1 | 6496 | ClearPageError(page); |
a61e6f29 | 6497 | unlock_page(page); |
d1310b2e | 6498 | } |
0b32f4bb | 6499 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 6500 | } |
d1310b2e | 6501 | |
abb57ef3 | 6502 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 6503 | { |
cc5e31a4 DS |
6504 | int i; |
6505 | int num_pages; | |
abb57ef3 | 6506 | bool was_dirty; |
d1310b2e | 6507 | |
0b32f4bb JB |
6508 | check_buffer_tree_ref(eb); |
6509 | ||
b9473439 | 6510 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 6511 | |
65ad0104 | 6512 | num_pages = num_extent_pages(eb); |
3083ee2e | 6513 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
6514 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
6515 | ||
0d27797e | 6516 | if (!was_dirty) { |
fbca46eb | 6517 | bool subpage = eb->fs_info->nodesize < PAGE_SIZE; |
51995c39 | 6518 | |
0d27797e QW |
6519 | /* |
6520 | * For subpage case, we can have other extent buffers in the | |
6521 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
6522 | * have to clear page dirty without subpage lock held. | |
6523 | * This can cause race where our page gets dirty cleared after | |
6524 | * we just set it. | |
6525 | * | |
6526 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
6527 | * its page for other reasons, we can use page lock to prevent | |
6528 | * the above race. | |
6529 | */ | |
6530 | if (subpage) | |
6531 | lock_page(eb->pages[0]); | |
6532 | for (i = 0; i < num_pages; i++) | |
6533 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
6534 | eb->start, eb->len); | |
6535 | if (subpage) | |
6536 | unlock_page(eb->pages[0]); | |
6537 | } | |
51995c39 LB |
6538 | #ifdef CONFIG_BTRFS_DEBUG |
6539 | for (i = 0; i < num_pages; i++) | |
6540 | ASSERT(PageDirty(eb->pages[i])); | |
6541 | #endif | |
6542 | ||
b9473439 | 6543 | return was_dirty; |
d1310b2e | 6544 | } |
d1310b2e | 6545 | |
69ba3927 | 6546 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 6547 | { |
251f2acc | 6548 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 6549 | struct page *page; |
cc5e31a4 | 6550 | int num_pages; |
251f2acc | 6551 | int i; |
1259ab75 | 6552 | |
b4ce94de | 6553 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6554 | num_pages = num_extent_pages(eb); |
1259ab75 | 6555 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6556 | page = eb->pages[i]; |
fbca46eb QW |
6557 | if (!page) |
6558 | continue; | |
6559 | ||
6560 | /* | |
6561 | * This is special handling for metadata subpage, as regular | |
6562 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6563 | */ | |
6564 | if (fs_info->nodesize >= PAGE_SIZE) | |
6565 | ClearPageUptodate(page); | |
6566 | else | |
6567 | btrfs_subpage_clear_uptodate(fs_info, page, eb->start, | |
6568 | eb->len); | |
1259ab75 | 6569 | } |
1259ab75 CM |
6570 | } |
6571 | ||
09c25a8c | 6572 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 6573 | { |
251f2acc | 6574 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 6575 | struct page *page; |
cc5e31a4 | 6576 | int num_pages; |
251f2acc | 6577 | int i; |
d1310b2e | 6578 | |
0b32f4bb | 6579 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6580 | num_pages = num_extent_pages(eb); |
d1310b2e | 6581 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6582 | page = eb->pages[i]; |
fbca46eb QW |
6583 | |
6584 | /* | |
6585 | * This is special handling for metadata subpage, as regular | |
6586 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
6587 | */ | |
6588 | if (fs_info->nodesize >= PAGE_SIZE) | |
6589 | SetPageUptodate(page); | |
6590 | else | |
6591 | btrfs_subpage_set_uptodate(fs_info, page, eb->start, | |
6592 | eb->len); | |
d1310b2e | 6593 | } |
d1310b2e | 6594 | } |
d1310b2e | 6595 | |
4012daf7 QW |
6596 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
6597 | int mirror_num) | |
6598 | { | |
6599 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6600 | struct extent_io_tree *io_tree; | |
6601 | struct page *page = eb->pages[0]; | |
390ed29b | 6602 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
4012daf7 QW |
6603 | int ret = 0; |
6604 | ||
6605 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
6606 | ASSERT(PagePrivate(page)); | |
6607 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
6608 | ||
6609 | if (wait == WAIT_NONE) { | |
dc56219f GR |
6610 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1)) |
6611 | return -EAGAIN; | |
4012daf7 QW |
6612 | } else { |
6613 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6614 | if (ret < 0) | |
6615 | return ret; | |
6616 | } | |
6617 | ||
6618 | ret = 0; | |
6619 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
6620 | PageUptodate(page) || | |
6621 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
6622 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
6623 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6624 | return ret; | |
6625 | } | |
6626 | ||
6627 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
6628 | eb->read_mirror = 0; | |
6629 | atomic_set(&eb->io_pages, 1); | |
6630 | check_buffer_tree_ref(eb); | |
6631 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); | |
6632 | ||
3d078efa | 6633 | btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len); |
390ed29b QW |
6634 | ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl, |
6635 | page, eb->start, eb->len, | |
6636 | eb->start - page_offset(page), | |
6637 | end_bio_extent_readpage, mirror_num, 0, | |
4012daf7 QW |
6638 | true); |
6639 | if (ret) { | |
6640 | /* | |
6641 | * In the endio function, if we hit something wrong we will | |
6642 | * increase the io_pages, so here we need to decrease it for | |
6643 | * error path. | |
6644 | */ | |
6645 | atomic_dec(&eb->io_pages); | |
6646 | } | |
390ed29b | 6647 | if (bio_ctrl.bio) { |
c9583ada | 6648 | submit_one_bio(bio_ctrl.bio, mirror_num, 0); |
390ed29b | 6649 | bio_ctrl.bio = NULL; |
4012daf7 QW |
6650 | } |
6651 | if (ret || wait != WAIT_COMPLETE) | |
6652 | return ret; | |
6653 | ||
6654 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED); | |
6655 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
6656 | ret = -EIO; | |
6657 | return ret; | |
6658 | } | |
6659 | ||
c2ccfbc6 | 6660 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 6661 | { |
cc5e31a4 | 6662 | int i; |
d1310b2e CM |
6663 | struct page *page; |
6664 | int err; | |
6665 | int ret = 0; | |
ce9adaa5 CM |
6666 | int locked_pages = 0; |
6667 | int all_uptodate = 1; | |
cc5e31a4 | 6668 | int num_pages; |
727011e0 | 6669 | unsigned long num_reads = 0; |
390ed29b | 6670 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
a86c12c7 | 6671 | |
b4ce94de | 6672 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
6673 | return 0; |
6674 | ||
651740a5 JB |
6675 | /* |
6676 | * We could have had EXTENT_BUFFER_UPTODATE cleared by the write | |
6677 | * operation, which could potentially still be in flight. In this case | |
6678 | * we simply want to return an error. | |
6679 | */ | |
6680 | if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))) | |
6681 | return -EIO; | |
6682 | ||
fbca46eb | 6683 | if (eb->fs_info->nodesize < PAGE_SIZE) |
4012daf7 QW |
6684 | return read_extent_buffer_subpage(eb, wait, mirror_num); |
6685 | ||
65ad0104 | 6686 | num_pages = num_extent_pages(eb); |
8436ea91 | 6687 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6688 | page = eb->pages[i]; |
bb82ab88 | 6689 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
6690 | /* |
6691 | * WAIT_NONE is only utilized by readahead. If we can't | |
6692 | * acquire the lock atomically it means either the eb | |
6693 | * is being read out or under modification. | |
6694 | * Either way the eb will be or has been cached, | |
6695 | * readahead can exit safely. | |
6696 | */ | |
2db04966 | 6697 | if (!trylock_page(page)) |
ce9adaa5 | 6698 | goto unlock_exit; |
d1310b2e CM |
6699 | } else { |
6700 | lock_page(page); | |
6701 | } | |
ce9adaa5 | 6702 | locked_pages++; |
2571e739 LB |
6703 | } |
6704 | /* | |
6705 | * We need to firstly lock all pages to make sure that | |
6706 | * the uptodate bit of our pages won't be affected by | |
6707 | * clear_extent_buffer_uptodate(). | |
6708 | */ | |
8436ea91 | 6709 | for (i = 0; i < num_pages; i++) { |
2571e739 | 6710 | page = eb->pages[i]; |
727011e0 CM |
6711 | if (!PageUptodate(page)) { |
6712 | num_reads++; | |
ce9adaa5 | 6713 | all_uptodate = 0; |
727011e0 | 6714 | } |
ce9adaa5 | 6715 | } |
2571e739 | 6716 | |
ce9adaa5 | 6717 | if (all_uptodate) { |
8436ea91 | 6718 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
6719 | goto unlock_exit; |
6720 | } | |
6721 | ||
656f30db | 6722 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 6723 | eb->read_mirror = 0; |
0b32f4bb | 6724 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e | 6725 | /* |
f913cff3 | 6726 | * It is possible for release_folio to clear the TREE_REF bit before we |
6bf9cd2e BB |
6727 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. |
6728 | */ | |
6729 | check_buffer_tree_ref(eb); | |
8436ea91 | 6730 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6731 | page = eb->pages[i]; |
baf863b9 | 6732 | |
ce9adaa5 | 6733 | if (!PageUptodate(page)) { |
baf863b9 LB |
6734 | if (ret) { |
6735 | atomic_dec(&eb->io_pages); | |
6736 | unlock_page(page); | |
6737 | continue; | |
6738 | } | |
6739 | ||
f188591e | 6740 | ClearPageError(page); |
0420177c | 6741 | err = submit_extent_page(REQ_OP_READ | REQ_META, NULL, |
390ed29b QW |
6742 | &bio_ctrl, page, page_offset(page), |
6743 | PAGE_SIZE, 0, end_bio_extent_readpage, | |
6744 | mirror_num, 0, false); | |
baf863b9 | 6745 | if (err) { |
baf863b9 | 6746 | /* |
0420177c NB |
6747 | * We failed to submit the bio so it's the |
6748 | * caller's responsibility to perform cleanup | |
6749 | * i.e unlock page/set error bit. | |
baf863b9 | 6750 | */ |
0420177c NB |
6751 | ret = err; |
6752 | SetPageError(page); | |
6753 | unlock_page(page); | |
baf863b9 LB |
6754 | atomic_dec(&eb->io_pages); |
6755 | } | |
d1310b2e CM |
6756 | } else { |
6757 | unlock_page(page); | |
6758 | } | |
6759 | } | |
6760 | ||
390ed29b | 6761 | if (bio_ctrl.bio) { |
0f07003b | 6762 | submit_one_bio(bio_ctrl.bio, mirror_num, bio_ctrl.compress_type); |
390ed29b | 6763 | bio_ctrl.bio = NULL; |
355808c2 | 6764 | } |
a86c12c7 | 6765 | |
bb82ab88 | 6766 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 6767 | return ret; |
d397712b | 6768 | |
8436ea91 | 6769 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6770 | page = eb->pages[i]; |
d1310b2e | 6771 | wait_on_page_locked(page); |
d397712b | 6772 | if (!PageUptodate(page)) |
d1310b2e | 6773 | ret = -EIO; |
d1310b2e | 6774 | } |
d397712b | 6775 | |
d1310b2e | 6776 | return ret; |
ce9adaa5 CM |
6777 | |
6778 | unlock_exit: | |
d397712b | 6779 | while (locked_pages > 0) { |
ce9adaa5 | 6780 | locked_pages--; |
8436ea91 JB |
6781 | page = eb->pages[locked_pages]; |
6782 | unlock_page(page); | |
ce9adaa5 CM |
6783 | } |
6784 | return ret; | |
d1310b2e | 6785 | } |
d1310b2e | 6786 | |
f98b6215 QW |
6787 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
6788 | unsigned long len) | |
6789 | { | |
6790 | btrfs_warn(eb->fs_info, | |
6791 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
6792 | eb->start, eb->len, start, len); | |
6793 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6794 | ||
6795 | return true; | |
6796 | } | |
6797 | ||
6798 | /* | |
6799 | * Check if the [start, start + len) range is valid before reading/writing | |
6800 | * the eb. | |
6801 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
6802 | * | |
6803 | * Caller should not touch the dst/src memory if this function returns error. | |
6804 | */ | |
6805 | static inline int check_eb_range(const struct extent_buffer *eb, | |
6806 | unsigned long start, unsigned long len) | |
6807 | { | |
6808 | unsigned long offset; | |
6809 | ||
6810 | /* start, start + len should not go beyond eb->len nor overflow */ | |
6811 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
6812 | return report_eb_range(eb, start, len); | |
6813 | ||
6814 | return false; | |
6815 | } | |
6816 | ||
1cbb1f45 JM |
6817 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
6818 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6819 | { |
6820 | size_t cur; | |
6821 | size_t offset; | |
6822 | struct page *page; | |
6823 | char *kaddr; | |
6824 | char *dst = (char *)dstv; | |
884b07d0 | 6825 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6826 | |
f98b6215 | 6827 | if (check_eb_range(eb, start, len)) |
f716abd5 | 6828 | return; |
d1310b2e | 6829 | |
884b07d0 | 6830 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6831 | |
d397712b | 6832 | while (len > 0) { |
fb85fc9a | 6833 | page = eb->pages[i]; |
d1310b2e | 6834 | |
09cbfeaf | 6835 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 6836 | kaddr = page_address(page); |
d1310b2e | 6837 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
6838 | |
6839 | dst += cur; | |
6840 | len -= cur; | |
6841 | offset = 0; | |
6842 | i++; | |
6843 | } | |
6844 | } | |
d1310b2e | 6845 | |
a48b73ec JB |
6846 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
6847 | void __user *dstv, | |
6848 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
6849 | { |
6850 | size_t cur; | |
6851 | size_t offset; | |
6852 | struct page *page; | |
6853 | char *kaddr; | |
6854 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 6855 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
6856 | int ret = 0; |
6857 | ||
6858 | WARN_ON(start > eb->len); | |
6859 | WARN_ON(start + len > eb->start + eb->len); | |
6860 | ||
884b07d0 | 6861 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
6862 | |
6863 | while (len > 0) { | |
fb85fc9a | 6864 | page = eb->pages[i]; |
550ac1d8 | 6865 | |
09cbfeaf | 6866 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 6867 | kaddr = page_address(page); |
a48b73ec | 6868 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
6869 | ret = -EFAULT; |
6870 | break; | |
6871 | } | |
6872 | ||
6873 | dst += cur; | |
6874 | len -= cur; | |
6875 | offset = 0; | |
6876 | i++; | |
6877 | } | |
6878 | ||
6879 | return ret; | |
6880 | } | |
6881 | ||
1cbb1f45 JM |
6882 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
6883 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6884 | { |
6885 | size_t cur; | |
6886 | size_t offset; | |
6887 | struct page *page; | |
6888 | char *kaddr; | |
6889 | char *ptr = (char *)ptrv; | |
884b07d0 | 6890 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
6891 | int ret = 0; |
6892 | ||
f98b6215 QW |
6893 | if (check_eb_range(eb, start, len)) |
6894 | return -EINVAL; | |
d1310b2e | 6895 | |
884b07d0 | 6896 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6897 | |
d397712b | 6898 | while (len > 0) { |
fb85fc9a | 6899 | page = eb->pages[i]; |
d1310b2e | 6900 | |
09cbfeaf | 6901 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 6902 | |
a6591715 | 6903 | kaddr = page_address(page); |
d1310b2e | 6904 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
6905 | if (ret) |
6906 | break; | |
6907 | ||
6908 | ptr += cur; | |
6909 | len -= cur; | |
6910 | offset = 0; | |
6911 | i++; | |
6912 | } | |
6913 | return ret; | |
6914 | } | |
d1310b2e | 6915 | |
b8f95771 QW |
6916 | /* |
6917 | * Check that the extent buffer is uptodate. | |
6918 | * | |
6919 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
6920 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
6921 | */ | |
6922 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
6923 | struct page *page) | |
6924 | { | |
6925 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6926 | ||
a50e1fcb JB |
6927 | /* |
6928 | * If we are using the commit root we could potentially clear a page | |
6929 | * Uptodate while we're using the extent buffer that we've previously | |
6930 | * looked up. We don't want to complain in this case, as the page was | |
6931 | * valid before, we just didn't write it out. Instead we want to catch | |
6932 | * the case where we didn't actually read the block properly, which | |
6933 | * would have !PageUptodate && !PageError, as we clear PageError before | |
6934 | * reading. | |
6935 | */ | |
fbca46eb | 6936 | if (fs_info->nodesize < PAGE_SIZE) { |
a50e1fcb | 6937 | bool uptodate, error; |
b8f95771 QW |
6938 | |
6939 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
6940 | eb->start, eb->len); | |
a50e1fcb JB |
6941 | error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len); |
6942 | WARN_ON(!uptodate && !error); | |
b8f95771 | 6943 | } else { |
a50e1fcb | 6944 | WARN_ON(!PageUptodate(page) && !PageError(page)); |
b8f95771 QW |
6945 | } |
6946 | } | |
6947 | ||
2b48966a | 6948 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
6949 | const void *srcv) |
6950 | { | |
6951 | char *kaddr; | |
6952 | ||
b8f95771 | 6953 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6954 | kaddr = page_address(eb->pages[0]) + |
6955 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
6956 | chunk_tree_uuid)); | |
6957 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6958 | } |
6959 | ||
2b48966a | 6960 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
6961 | { |
6962 | char *kaddr; | |
6963 | ||
b8f95771 | 6964 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6965 | kaddr = page_address(eb->pages[0]) + |
6966 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
6967 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6968 | } |
6969 | ||
2b48966a | 6970 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
6971 | unsigned long start, unsigned long len) |
6972 | { | |
6973 | size_t cur; | |
6974 | size_t offset; | |
6975 | struct page *page; | |
6976 | char *kaddr; | |
6977 | char *src = (char *)srcv; | |
884b07d0 | 6978 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6979 | |
d3575156 NA |
6980 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
6981 | ||
f98b6215 QW |
6982 | if (check_eb_range(eb, start, len)) |
6983 | return; | |
d1310b2e | 6984 | |
884b07d0 | 6985 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6986 | |
d397712b | 6987 | while (len > 0) { |
fb85fc9a | 6988 | page = eb->pages[i]; |
b8f95771 | 6989 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6990 | |
09cbfeaf | 6991 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6992 | kaddr = page_address(page); |
d1310b2e | 6993 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
6994 | |
6995 | src += cur; | |
6996 | len -= cur; | |
6997 | offset = 0; | |
6998 | i++; | |
6999 | } | |
7000 | } | |
d1310b2e | 7001 | |
2b48966a | 7002 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 7003 | unsigned long len) |
d1310b2e CM |
7004 | { |
7005 | size_t cur; | |
7006 | size_t offset; | |
7007 | struct page *page; | |
7008 | char *kaddr; | |
884b07d0 | 7009 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 7010 | |
f98b6215 QW |
7011 | if (check_eb_range(eb, start, len)) |
7012 | return; | |
d1310b2e | 7013 | |
884b07d0 | 7014 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 7015 | |
d397712b | 7016 | while (len > 0) { |
fb85fc9a | 7017 | page = eb->pages[i]; |
b8f95771 | 7018 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 7019 | |
09cbfeaf | 7020 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 7021 | kaddr = page_address(page); |
b159fa28 | 7022 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
7023 | |
7024 | len -= cur; | |
7025 | offset = 0; | |
7026 | i++; | |
7027 | } | |
7028 | } | |
d1310b2e | 7029 | |
2b48966a DS |
7030 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
7031 | const struct extent_buffer *src) | |
58e8012c DS |
7032 | { |
7033 | int i; | |
cc5e31a4 | 7034 | int num_pages; |
58e8012c DS |
7035 | |
7036 | ASSERT(dst->len == src->len); | |
7037 | ||
fbca46eb | 7038 | if (dst->fs_info->nodesize >= PAGE_SIZE) { |
884b07d0 QW |
7039 | num_pages = num_extent_pages(dst); |
7040 | for (i = 0; i < num_pages; i++) | |
7041 | copy_page(page_address(dst->pages[i]), | |
7042 | page_address(src->pages[i])); | |
7043 | } else { | |
7044 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
7045 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
7046 | ||
fbca46eb | 7047 | ASSERT(src->fs_info->nodesize < PAGE_SIZE); |
884b07d0 QW |
7048 | memcpy(page_address(dst->pages[0]) + dst_offset, |
7049 | page_address(src->pages[0]) + src_offset, | |
7050 | src->len); | |
7051 | } | |
58e8012c DS |
7052 | } |
7053 | ||
2b48966a DS |
7054 | void copy_extent_buffer(const struct extent_buffer *dst, |
7055 | const struct extent_buffer *src, | |
d1310b2e CM |
7056 | unsigned long dst_offset, unsigned long src_offset, |
7057 | unsigned long len) | |
7058 | { | |
7059 | u64 dst_len = dst->len; | |
7060 | size_t cur; | |
7061 | size_t offset; | |
7062 | struct page *page; | |
7063 | char *kaddr; | |
884b07d0 | 7064 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 7065 | |
f98b6215 QW |
7066 | if (check_eb_range(dst, dst_offset, len) || |
7067 | check_eb_range(src, src_offset, len)) | |
7068 | return; | |
7069 | ||
d1310b2e CM |
7070 | WARN_ON(src->len != dst_len); |
7071 | ||
884b07d0 | 7072 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 7073 | |
d397712b | 7074 | while (len > 0) { |
fb85fc9a | 7075 | page = dst->pages[i]; |
b8f95771 | 7076 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 7077 | |
09cbfeaf | 7078 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 7079 | |
a6591715 | 7080 | kaddr = page_address(page); |
d1310b2e | 7081 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
7082 | |
7083 | src_offset += cur; | |
7084 | len -= cur; | |
7085 | offset = 0; | |
7086 | i++; | |
7087 | } | |
7088 | } | |
d1310b2e | 7089 | |
3e1e8bb7 OS |
7090 | /* |
7091 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
7092 | * given bit number | |
7093 | * @eb: the extent buffer | |
7094 | * @start: offset of the bitmap item in the extent buffer | |
7095 | * @nr: bit number | |
7096 | * @page_index: return index of the page in the extent buffer that contains the | |
7097 | * given bit number | |
7098 | * @page_offset: return offset into the page given by page_index | |
7099 | * | |
7100 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
7101 | * contains a given bit. | |
7102 | */ | |
2b48966a | 7103 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
7104 | unsigned long start, unsigned long nr, |
7105 | unsigned long *page_index, | |
7106 | size_t *page_offset) | |
7107 | { | |
3e1e8bb7 OS |
7108 | size_t byte_offset = BIT_BYTE(nr); |
7109 | size_t offset; | |
7110 | ||
7111 | /* | |
7112 | * The byte we want is the offset of the extent buffer + the offset of | |
7113 | * the bitmap item in the extent buffer + the offset of the byte in the | |
7114 | * bitmap item. | |
7115 | */ | |
884b07d0 | 7116 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 7117 | |
09cbfeaf | 7118 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 7119 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
7120 | } |
7121 | ||
7122 | /** | |
7123 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
7124 | * @eb: the extent buffer | |
7125 | * @start: offset of the bitmap item in the extent buffer | |
7126 | * @nr: bit number to test | |
7127 | */ | |
2b48966a | 7128 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7129 | unsigned long nr) |
7130 | { | |
2fe1d551 | 7131 | u8 *kaddr; |
3e1e8bb7 OS |
7132 | struct page *page; |
7133 | unsigned long i; | |
7134 | size_t offset; | |
7135 | ||
7136 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
7137 | page = eb->pages[i]; | |
b8f95771 | 7138 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7139 | kaddr = page_address(page); |
7140 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
7141 | } | |
7142 | ||
7143 | /** | |
7144 | * extent_buffer_bitmap_set - set an area of a bitmap | |
7145 | * @eb: the extent buffer | |
7146 | * @start: offset of the bitmap item in the extent buffer | |
7147 | * @pos: bit number of the first bit | |
7148 | * @len: number of bits to set | |
7149 | */ | |
2b48966a | 7150 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
7151 | unsigned long pos, unsigned long len) |
7152 | { | |
2fe1d551 | 7153 | u8 *kaddr; |
3e1e8bb7 OS |
7154 | struct page *page; |
7155 | unsigned long i; | |
7156 | size_t offset; | |
7157 | const unsigned int size = pos + len; | |
7158 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7159 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7160 | |
7161 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7162 | page = eb->pages[i]; | |
b8f95771 | 7163 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7164 | kaddr = page_address(page); |
7165 | ||
7166 | while (len >= bits_to_set) { | |
7167 | kaddr[offset] |= mask_to_set; | |
7168 | len -= bits_to_set; | |
7169 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 7170 | mask_to_set = ~0; |
09cbfeaf | 7171 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7172 | offset = 0; |
7173 | page = eb->pages[++i]; | |
b8f95771 | 7174 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7175 | kaddr = page_address(page); |
7176 | } | |
7177 | } | |
7178 | if (len) { | |
7179 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
7180 | kaddr[offset] |= mask_to_set; | |
7181 | } | |
7182 | } | |
7183 | ||
7184 | ||
7185 | /** | |
7186 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
7187 | * @eb: the extent buffer | |
7188 | * @start: offset of the bitmap item in the extent buffer | |
7189 | * @pos: bit number of the first bit | |
7190 | * @len: number of bits to clear | |
7191 | */ | |
2b48966a DS |
7192 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
7193 | unsigned long start, unsigned long pos, | |
7194 | unsigned long len) | |
3e1e8bb7 | 7195 | { |
2fe1d551 | 7196 | u8 *kaddr; |
3e1e8bb7 OS |
7197 | struct page *page; |
7198 | unsigned long i; | |
7199 | size_t offset; | |
7200 | const unsigned int size = pos + len; | |
7201 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 7202 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
7203 | |
7204 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
7205 | page = eb->pages[i]; | |
b8f95771 | 7206 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7207 | kaddr = page_address(page); |
7208 | ||
7209 | while (len >= bits_to_clear) { | |
7210 | kaddr[offset] &= ~mask_to_clear; | |
7211 | len -= bits_to_clear; | |
7212 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 7213 | mask_to_clear = ~0; |
09cbfeaf | 7214 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
7215 | offset = 0; |
7216 | page = eb->pages[++i]; | |
b8f95771 | 7217 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
7218 | kaddr = page_address(page); |
7219 | } | |
7220 | } | |
7221 | if (len) { | |
7222 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
7223 | kaddr[offset] &= ~mask_to_clear; | |
7224 | } | |
7225 | } | |
7226 | ||
3387206f ST |
7227 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
7228 | { | |
7229 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
7230 | return distance < len; | |
7231 | } | |
7232 | ||
d1310b2e CM |
7233 | static void copy_pages(struct page *dst_page, struct page *src_page, |
7234 | unsigned long dst_off, unsigned long src_off, | |
7235 | unsigned long len) | |
7236 | { | |
a6591715 | 7237 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 7238 | char *src_kaddr; |
727011e0 | 7239 | int must_memmove = 0; |
d1310b2e | 7240 | |
3387206f | 7241 | if (dst_page != src_page) { |
a6591715 | 7242 | src_kaddr = page_address(src_page); |
3387206f | 7243 | } else { |
d1310b2e | 7244 | src_kaddr = dst_kaddr; |
727011e0 CM |
7245 | if (areas_overlap(src_off, dst_off, len)) |
7246 | must_memmove = 1; | |
3387206f | 7247 | } |
d1310b2e | 7248 | |
727011e0 CM |
7249 | if (must_memmove) |
7250 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
7251 | else | |
7252 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
7253 | } |
7254 | ||
2b48966a DS |
7255 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
7256 | unsigned long dst_offset, unsigned long src_offset, | |
7257 | unsigned long len) | |
d1310b2e CM |
7258 | { |
7259 | size_t cur; | |
7260 | size_t dst_off_in_page; | |
7261 | size_t src_off_in_page; | |
d1310b2e CM |
7262 | unsigned long dst_i; |
7263 | unsigned long src_i; | |
7264 | ||
f98b6215 QW |
7265 | if (check_eb_range(dst, dst_offset, len) || |
7266 | check_eb_range(dst, src_offset, len)) | |
7267 | return; | |
d1310b2e | 7268 | |
d397712b | 7269 | while (len > 0) { |
884b07d0 QW |
7270 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
7271 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 7272 | |
884b07d0 QW |
7273 | dst_i = get_eb_page_index(dst_offset); |
7274 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 7275 | |
09cbfeaf | 7276 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
7277 | src_off_in_page)); |
7278 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 7279 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 7280 | |
fb85fc9a | 7281 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7282 | dst_off_in_page, src_off_in_page, cur); |
7283 | ||
7284 | src_offset += cur; | |
7285 | dst_offset += cur; | |
7286 | len -= cur; | |
7287 | } | |
7288 | } | |
d1310b2e | 7289 | |
2b48966a DS |
7290 | void memmove_extent_buffer(const struct extent_buffer *dst, |
7291 | unsigned long dst_offset, unsigned long src_offset, | |
7292 | unsigned long len) | |
d1310b2e CM |
7293 | { |
7294 | size_t cur; | |
7295 | size_t dst_off_in_page; | |
7296 | size_t src_off_in_page; | |
7297 | unsigned long dst_end = dst_offset + len - 1; | |
7298 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
7299 | unsigned long dst_i; |
7300 | unsigned long src_i; | |
7301 | ||
f98b6215 QW |
7302 | if (check_eb_range(dst, dst_offset, len) || |
7303 | check_eb_range(dst, src_offset, len)) | |
7304 | return; | |
727011e0 | 7305 | if (dst_offset < src_offset) { |
d1310b2e CM |
7306 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
7307 | return; | |
7308 | } | |
d397712b | 7309 | while (len > 0) { |
884b07d0 QW |
7310 | dst_i = get_eb_page_index(dst_end); |
7311 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 7312 | |
884b07d0 QW |
7313 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
7314 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
7315 | |
7316 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
7317 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 7318 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
7319 | dst_off_in_page - cur + 1, |
7320 | src_off_in_page - cur + 1, cur); | |
7321 | ||
7322 | dst_end -= cur; | |
7323 | src_end -= cur; | |
7324 | len -= cur; | |
7325 | } | |
7326 | } | |
6af118ce | 7327 | |
d1e86e3f QW |
7328 | static struct extent_buffer *get_next_extent_buffer( |
7329 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
7330 | { | |
8ee92268 GN |
7331 | struct extent_buffer *eb; |
7332 | unsigned long index; | |
d1e86e3f | 7333 | u64 page_start = page_offset(page); |
d1e86e3f QW |
7334 | |
7335 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
d1e86e3f QW |
7336 | lockdep_assert_held(&fs_info->buffer_lock); |
7337 | ||
8ee92268 GN |
7338 | xa_for_each_start(&fs_info->extent_buffers, index, eb, |
7339 | page_start >> fs_info->sectorsize_bits) { | |
7340 | if (in_range(eb->start, page_start, PAGE_SIZE)) | |
7341 | return eb; | |
7342 | else if (eb->start >= page_start + PAGE_SIZE) | |
7343 | /* Already beyond page end */ | |
7344 | return NULL; | |
d1e86e3f | 7345 | } |
8ee92268 | 7346 | return NULL; |
d1e86e3f QW |
7347 | } |
7348 | ||
7349 | static int try_release_subpage_extent_buffer(struct page *page) | |
7350 | { | |
7351 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7352 | u64 cur = page_offset(page); | |
7353 | const u64 end = page_offset(page) + PAGE_SIZE; | |
7354 | int ret; | |
7355 | ||
7356 | while (cur < end) { | |
7357 | struct extent_buffer *eb = NULL; | |
7358 | ||
7359 | /* | |
7360 | * Unlike try_release_extent_buffer() which uses page->private | |
7361 | * to grab buffer, for subpage case we rely on radix tree, thus | |
7362 | * we need to ensure radix tree consistency. | |
7363 | * | |
7364 | * We also want an atomic snapshot of the radix tree, thus go | |
7365 | * with spinlock rather than RCU. | |
7366 | */ | |
7367 | spin_lock(&fs_info->buffer_lock); | |
7368 | eb = get_next_extent_buffer(fs_info, page, cur); | |
7369 | if (!eb) { | |
7370 | /* No more eb in the page range after or at cur */ | |
7371 | spin_unlock(&fs_info->buffer_lock); | |
7372 | break; | |
7373 | } | |
7374 | cur = eb->start + eb->len; | |
7375 | ||
7376 | /* | |
7377 | * The same as try_release_extent_buffer(), to ensure the eb | |
7378 | * won't disappear out from under us. | |
7379 | */ | |
7380 | spin_lock(&eb->refs_lock); | |
7381 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
7382 | spin_unlock(&eb->refs_lock); | |
7383 | spin_unlock(&fs_info->buffer_lock); | |
7384 | break; | |
7385 | } | |
7386 | spin_unlock(&fs_info->buffer_lock); | |
7387 | ||
7388 | /* | |
7389 | * If tree ref isn't set then we know the ref on this eb is a | |
7390 | * real ref, so just return, this eb will likely be freed soon | |
7391 | * anyway. | |
7392 | */ | |
7393 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
7394 | spin_unlock(&eb->refs_lock); | |
7395 | break; | |
7396 | } | |
7397 | ||
7398 | /* | |
7399 | * Here we don't care about the return value, we will always | |
7400 | * check the page private at the end. And | |
7401 | * release_extent_buffer() will release the refs_lock. | |
7402 | */ | |
7403 | release_extent_buffer(eb); | |
7404 | } | |
7405 | /* | |
7406 | * Finally to check if we have cleared page private, as if we have | |
7407 | * released all ebs in the page, the page private should be cleared now. | |
7408 | */ | |
7409 | spin_lock(&page->mapping->private_lock); | |
7410 | if (!PagePrivate(page)) | |
7411 | ret = 1; | |
7412 | else | |
7413 | ret = 0; | |
7414 | spin_unlock(&page->mapping->private_lock); | |
7415 | return ret; | |
7416 | ||
7417 | } | |
7418 | ||
f7a52a40 | 7419 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 7420 | { |
6af118ce | 7421 | struct extent_buffer *eb; |
6af118ce | 7422 | |
fbca46eb | 7423 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
d1e86e3f QW |
7424 | return try_release_subpage_extent_buffer(page); |
7425 | ||
3083ee2e | 7426 | /* |
d1e86e3f QW |
7427 | * We need to make sure nobody is changing page->private, as we rely on |
7428 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
7429 | */ |
7430 | spin_lock(&page->mapping->private_lock); | |
7431 | if (!PagePrivate(page)) { | |
7432 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 7433 | return 1; |
45f49bce | 7434 | } |
6af118ce | 7435 | |
3083ee2e JB |
7436 | eb = (struct extent_buffer *)page->private; |
7437 | BUG_ON(!eb); | |
19fe0a8b MX |
7438 | |
7439 | /* | |
3083ee2e JB |
7440 | * This is a little awful but should be ok, we need to make sure that |
7441 | * the eb doesn't disappear out from under us while we're looking at | |
7442 | * this page. | |
19fe0a8b | 7443 | */ |
3083ee2e | 7444 | spin_lock(&eb->refs_lock); |
0b32f4bb | 7445 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
7446 | spin_unlock(&eb->refs_lock); |
7447 | spin_unlock(&page->mapping->private_lock); | |
7448 | return 0; | |
b9473439 | 7449 | } |
3083ee2e | 7450 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 7451 | |
19fe0a8b | 7452 | /* |
3083ee2e JB |
7453 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
7454 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 7455 | */ |
3083ee2e JB |
7456 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
7457 | spin_unlock(&eb->refs_lock); | |
7458 | return 0; | |
b9473439 | 7459 | } |
19fe0a8b | 7460 | |
f7a52a40 | 7461 | return release_extent_buffer(eb); |
6af118ce | 7462 | } |
bfb484d9 JB |
7463 | |
7464 | /* | |
7465 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
7466 | * @fs_info: the fs_info | |
7467 | * @bytenr: bytenr to read | |
3fbaf258 | 7468 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 7469 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 7470 | * @level: level for the eb |
bfb484d9 JB |
7471 | * |
7472 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
7473 | * normal uptodate check of the eb, without checking the generation. If we have | |
7474 | * to read the block we will not block on anything. | |
7475 | */ | |
7476 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 7477 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
7478 | { |
7479 | struct extent_buffer *eb; | |
7480 | int ret; | |
7481 | ||
3fbaf258 | 7482 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
7483 | if (IS_ERR(eb)) |
7484 | return; | |
7485 | ||
7486 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
7487 | free_extent_buffer(eb); | |
7488 | return; | |
7489 | } | |
7490 | ||
7491 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
7492 | if (ret < 0) | |
7493 | free_extent_buffer_stale(eb); | |
7494 | else | |
7495 | free_extent_buffer(eb); | |
7496 | } | |
7497 | ||
7498 | /* | |
7499 | * btrfs_readahead_node_child - readahead a node's child block | |
7500 | * @node: parent node we're reading from | |
7501 | * @slot: slot in the parent node for the child we want to read | |
7502 | * | |
7503 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
7504 | * the slot in the node provided. | |
7505 | */ | |
7506 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
7507 | { | |
7508 | btrfs_readahead_tree_block(node->fs_info, | |
7509 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
7510 | btrfs_header_owner(node), |
7511 | btrfs_node_ptr_generation(node, slot), | |
7512 | btrfs_header_level(node) - 1); | |
bfb484d9 | 7513 | } |