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> | |
d1310b2e CM |
9 | #include <linux/spinlock.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/swap.h> | |
d1310b2e CM |
12 | #include <linux/writeback.h> |
13 | #include <linux/pagevec.h> | |
268bb0ce | 14 | #include <linux/prefetch.h> |
90a887c9 | 15 | #include <linux/cleancache.h> |
cea62800 | 16 | #include "misc.h" |
d1310b2e | 17 | #include "extent_io.h" |
9c7d3a54 | 18 | #include "extent-io-tree.h" |
d1310b2e | 19 | #include "extent_map.h" |
902b22f3 DW |
20 | #include "ctree.h" |
21 | #include "btrfs_inode.h" | |
4a54c8c1 | 22 | #include "volumes.h" |
21adbd5c | 23 | #include "check-integrity.h" |
0b32f4bb | 24 | #include "locking.h" |
606686ee | 25 | #include "rcu-string.h" |
fe09e16c | 26 | #include "backref.h" |
6af49dbd | 27 | #include "disk-io.h" |
760f991f | 28 | #include "subpage.h" |
d3575156 | 29 | #include "zoned.h" |
0bc09ca1 | 30 | #include "block-group.h" |
d1310b2e | 31 | |
d1310b2e CM |
32 | static struct kmem_cache *extent_state_cache; |
33 | static struct kmem_cache *extent_buffer_cache; | |
8ac9f7c1 | 34 | static struct bio_set btrfs_bioset; |
d1310b2e | 35 | |
27a3507d FM |
36 | static inline bool extent_state_in_tree(const struct extent_state *state) |
37 | { | |
38 | return !RB_EMPTY_NODE(&state->rb_node); | |
39 | } | |
40 | ||
6d49ba1b | 41 | #ifdef CONFIG_BTRFS_DEBUG |
d1310b2e | 42 | static LIST_HEAD(states); |
d397712b | 43 | static DEFINE_SPINLOCK(leak_lock); |
6d49ba1b | 44 | |
3fd63727 JB |
45 | static inline void btrfs_leak_debug_add(spinlock_t *lock, |
46 | struct list_head *new, | |
47 | struct list_head *head) | |
6d49ba1b ES |
48 | { |
49 | unsigned long flags; | |
50 | ||
3fd63727 | 51 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 52 | list_add(new, head); |
3fd63727 | 53 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
54 | } |
55 | ||
3fd63727 JB |
56 | static inline void btrfs_leak_debug_del(spinlock_t *lock, |
57 | struct list_head *entry) | |
6d49ba1b ES |
58 | { |
59 | unsigned long flags; | |
60 | ||
3fd63727 | 61 | spin_lock_irqsave(lock, flags); |
6d49ba1b | 62 | list_del(entry); |
3fd63727 | 63 | spin_unlock_irqrestore(lock, flags); |
6d49ba1b ES |
64 | } |
65 | ||
3fd63727 | 66 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 67 | { |
6d49ba1b | 68 | struct extent_buffer *eb; |
3fd63727 | 69 | unsigned long flags; |
6d49ba1b | 70 | |
8c38938c JB |
71 | /* |
72 | * If we didn't get into open_ctree our allocated_ebs will not be | |
73 | * initialized, so just skip this. | |
74 | */ | |
75 | if (!fs_info->allocated_ebs.next) | |
76 | return; | |
77 | ||
3fd63727 JB |
78 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
79 | while (!list_empty(&fs_info->allocated_ebs)) { | |
80 | eb = list_first_entry(&fs_info->allocated_ebs, | |
81 | struct extent_buffer, leak_list); | |
8c38938c JB |
82 | pr_err( |
83 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
84 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
85 | btrfs_header_owner(eb)); | |
33ca832f JB |
86 | list_del(&eb->leak_list); |
87 | kmem_cache_free(extent_buffer_cache, eb); | |
88 | } | |
3fd63727 | 89 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f JB |
90 | } |
91 | ||
92 | static inline void btrfs_extent_state_leak_debug_check(void) | |
93 | { | |
94 | struct extent_state *state; | |
95 | ||
6d49ba1b ES |
96 | while (!list_empty(&states)) { |
97 | state = list_entry(states.next, struct extent_state, leak_list); | |
9ee49a04 | 98 | pr_err("BTRFS: state leak: start %llu end %llu state %u in tree %d refs %d\n", |
27a3507d FM |
99 | state->start, state->end, state->state, |
100 | extent_state_in_tree(state), | |
b7ac31b7 | 101 | refcount_read(&state->refs)); |
6d49ba1b ES |
102 | list_del(&state->leak_list); |
103 | kmem_cache_free(extent_state_cache, state); | |
104 | } | |
6d49ba1b | 105 | } |
8d599ae1 | 106 | |
a5dee37d JB |
107 | #define btrfs_debug_check_extent_io_range(tree, start, end) \ |
108 | __btrfs_debug_check_extent_io_range(__func__, (tree), (start), (end)) | |
8d599ae1 | 109 | static inline void __btrfs_debug_check_extent_io_range(const char *caller, |
a5dee37d | 110 | struct extent_io_tree *tree, u64 start, u64 end) |
8d599ae1 | 111 | { |
65a680f6 NB |
112 | struct inode *inode = tree->private_data; |
113 | u64 isize; | |
114 | ||
115 | if (!inode || !is_data_inode(inode)) | |
116 | return; | |
117 | ||
118 | isize = i_size_read(inode); | |
119 | if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) { | |
120 | btrfs_debug_rl(BTRFS_I(inode)->root->fs_info, | |
121 | "%s: ino %llu isize %llu odd range [%llu,%llu]", | |
122 | caller, btrfs_ino(BTRFS_I(inode)), isize, start, end); | |
123 | } | |
8d599ae1 | 124 | } |
6d49ba1b | 125 | #else |
3fd63727 JB |
126 | #define btrfs_leak_debug_add(lock, new, head) do {} while (0) |
127 | #define btrfs_leak_debug_del(lock, entry) do {} while (0) | |
33ca832f | 128 | #define btrfs_extent_state_leak_debug_check() do {} while (0) |
8d599ae1 | 129 | #define btrfs_debug_check_extent_io_range(c, s, e) do {} while (0) |
4bef0848 | 130 | #endif |
d1310b2e | 131 | |
d1310b2e CM |
132 | struct tree_entry { |
133 | u64 start; | |
134 | u64 end; | |
d1310b2e CM |
135 | struct rb_node rb_node; |
136 | }; | |
137 | ||
138 | struct extent_page_data { | |
390ed29b | 139 | struct btrfs_bio_ctrl bio_ctrl; |
771ed689 CM |
140 | /* tells writepage not to lock the state bits for this range |
141 | * it still does the unlocking | |
142 | */ | |
ffbd517d CM |
143 | unsigned int extent_locked:1; |
144 | ||
70fd7614 | 145 | /* tells the submit_bio code to use REQ_SYNC */ |
ffbd517d | 146 | unsigned int sync_io:1; |
d1310b2e CM |
147 | }; |
148 | ||
f97e27e9 | 149 | static int add_extent_changeset(struct extent_state *state, u32 bits, |
d38ed27f QW |
150 | struct extent_changeset *changeset, |
151 | int set) | |
152 | { | |
153 | int ret; | |
154 | ||
155 | if (!changeset) | |
57599c7e | 156 | return 0; |
d38ed27f | 157 | if (set && (state->state & bits) == bits) |
57599c7e | 158 | return 0; |
fefdc557 | 159 | if (!set && (state->state & bits) == 0) |
57599c7e | 160 | return 0; |
d38ed27f | 161 | changeset->bytes_changed += state->end - state->start + 1; |
53d32359 | 162 | ret = ulist_add(&changeset->range_changed, state->start, state->end, |
d38ed27f | 163 | GFP_ATOMIC); |
57599c7e | 164 | return ret; |
d38ed27f QW |
165 | } |
166 | ||
c1be9c1a NB |
167 | int __must_check submit_one_bio(struct bio *bio, int mirror_num, |
168 | unsigned long bio_flags) | |
bb58eb9e QW |
169 | { |
170 | blk_status_t ret = 0; | |
bb58eb9e | 171 | struct extent_io_tree *tree = bio->bi_private; |
bb58eb9e QW |
172 | |
173 | bio->bi_private = NULL; | |
174 | ||
908930f3 NB |
175 | if (is_data_inode(tree->private_data)) |
176 | ret = btrfs_submit_data_bio(tree->private_data, bio, mirror_num, | |
177 | bio_flags); | |
178 | else | |
1b36294a NB |
179 | ret = btrfs_submit_metadata_bio(tree->private_data, bio, |
180 | mirror_num, bio_flags); | |
bb58eb9e QW |
181 | |
182 | return blk_status_to_errno(ret); | |
183 | } | |
184 | ||
3065976b QW |
185 | /* Cleanup unsubmitted bios */ |
186 | static void end_write_bio(struct extent_page_data *epd, int ret) | |
187 | { | |
390ed29b QW |
188 | struct bio *bio = epd->bio_ctrl.bio; |
189 | ||
190 | if (bio) { | |
191 | bio->bi_status = errno_to_blk_status(ret); | |
192 | bio_endio(bio); | |
193 | epd->bio_ctrl.bio = NULL; | |
3065976b QW |
194 | } |
195 | } | |
196 | ||
f4340622 QW |
197 | /* |
198 | * Submit bio from extent page data via submit_one_bio | |
199 | * | |
200 | * Return 0 if everything is OK. | |
201 | * Return <0 for error. | |
202 | */ | |
203 | static int __must_check flush_write_bio(struct extent_page_data *epd) | |
bb58eb9e | 204 | { |
f4340622 | 205 | int ret = 0; |
390ed29b | 206 | struct bio *bio = epd->bio_ctrl.bio; |
bb58eb9e | 207 | |
390ed29b QW |
208 | if (bio) { |
209 | ret = submit_one_bio(bio, 0, 0); | |
f4340622 QW |
210 | /* |
211 | * Clean up of epd->bio is handled by its endio function. | |
212 | * And endio is either triggered by successful bio execution | |
213 | * or the error handler of submit bio hook. | |
214 | * So at this point, no matter what happened, we don't need | |
215 | * to clean up epd->bio. | |
216 | */ | |
390ed29b | 217 | epd->bio_ctrl.bio = NULL; |
bb58eb9e | 218 | } |
f4340622 | 219 | return ret; |
bb58eb9e | 220 | } |
e2932ee0 | 221 | |
6f0d04f8 | 222 | int __init extent_state_cache_init(void) |
d1310b2e | 223 | { |
837e1972 | 224 | extent_state_cache = kmem_cache_create("btrfs_extent_state", |
9601e3f6 | 225 | sizeof(struct extent_state), 0, |
fba4b697 | 226 | SLAB_MEM_SPREAD, NULL); |
d1310b2e CM |
227 | if (!extent_state_cache) |
228 | return -ENOMEM; | |
6f0d04f8 JB |
229 | return 0; |
230 | } | |
d1310b2e | 231 | |
6f0d04f8 JB |
232 | int __init extent_io_init(void) |
233 | { | |
837e1972 | 234 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 235 | sizeof(struct extent_buffer), 0, |
fba4b697 | 236 | SLAB_MEM_SPREAD, NULL); |
d1310b2e | 237 | if (!extent_buffer_cache) |
6f0d04f8 | 238 | return -ENOMEM; |
9be3395b | 239 | |
8ac9f7c1 KO |
240 | if (bioset_init(&btrfs_bioset, BIO_POOL_SIZE, |
241 | offsetof(struct btrfs_io_bio, bio), | |
242 | BIOSET_NEED_BVECS)) | |
9be3395b | 243 | goto free_buffer_cache; |
b208c2f7 | 244 | |
8ac9f7c1 | 245 | if (bioset_integrity_create(&btrfs_bioset, BIO_POOL_SIZE)) |
b208c2f7 DW |
246 | goto free_bioset; |
247 | ||
d1310b2e CM |
248 | return 0; |
249 | ||
b208c2f7 | 250 | free_bioset: |
8ac9f7c1 | 251 | bioset_exit(&btrfs_bioset); |
b208c2f7 | 252 | |
9be3395b CM |
253 | free_buffer_cache: |
254 | kmem_cache_destroy(extent_buffer_cache); | |
255 | extent_buffer_cache = NULL; | |
6f0d04f8 JB |
256 | return -ENOMEM; |
257 | } | |
9be3395b | 258 | |
6f0d04f8 JB |
259 | void __cold extent_state_cache_exit(void) |
260 | { | |
261 | btrfs_extent_state_leak_debug_check(); | |
d1310b2e | 262 | kmem_cache_destroy(extent_state_cache); |
d1310b2e CM |
263 | } |
264 | ||
e67c718b | 265 | void __cold extent_io_exit(void) |
d1310b2e | 266 | { |
8c0a8537 KS |
267 | /* |
268 | * Make sure all delayed rcu free are flushed before we | |
269 | * destroy caches. | |
270 | */ | |
271 | rcu_barrier(); | |
5598e900 | 272 | kmem_cache_destroy(extent_buffer_cache); |
8ac9f7c1 | 273 | bioset_exit(&btrfs_bioset); |
d1310b2e CM |
274 | } |
275 | ||
41a2ee75 JB |
276 | /* |
277 | * For the file_extent_tree, we want to hold the inode lock when we lookup and | |
278 | * update the disk_i_size, but lockdep will complain because our io_tree we hold | |
279 | * the tree lock and get the inode lock when setting delalloc. These two things | |
280 | * are unrelated, so make a class for the file_extent_tree so we don't get the | |
281 | * two locking patterns mixed up. | |
282 | */ | |
283 | static struct lock_class_key file_extent_tree_class; | |
284 | ||
c258d6e3 | 285 | void extent_io_tree_init(struct btrfs_fs_info *fs_info, |
43eb5f29 QW |
286 | struct extent_io_tree *tree, unsigned int owner, |
287 | void *private_data) | |
d1310b2e | 288 | { |
c258d6e3 | 289 | tree->fs_info = fs_info; |
6bef4d31 | 290 | tree->state = RB_ROOT; |
d1310b2e | 291 | tree->dirty_bytes = 0; |
70dec807 | 292 | spin_lock_init(&tree->lock); |
c6100a4b | 293 | tree->private_data = private_data; |
43eb5f29 | 294 | tree->owner = owner; |
41a2ee75 JB |
295 | if (owner == IO_TREE_INODE_FILE_EXTENT) |
296 | lockdep_set_class(&tree->lock, &file_extent_tree_class); | |
d1310b2e | 297 | } |
d1310b2e | 298 | |
41e7acd3 NB |
299 | void extent_io_tree_release(struct extent_io_tree *tree) |
300 | { | |
301 | spin_lock(&tree->lock); | |
302 | /* | |
303 | * Do a single barrier for the waitqueue_active check here, the state | |
304 | * of the waitqueue should not change once extent_io_tree_release is | |
305 | * called. | |
306 | */ | |
307 | smp_mb(); | |
308 | while (!RB_EMPTY_ROOT(&tree->state)) { | |
309 | struct rb_node *node; | |
310 | struct extent_state *state; | |
311 | ||
312 | node = rb_first(&tree->state); | |
313 | state = rb_entry(node, struct extent_state, rb_node); | |
314 | rb_erase(&state->rb_node, &tree->state); | |
315 | RB_CLEAR_NODE(&state->rb_node); | |
316 | /* | |
317 | * btree io trees aren't supposed to have tasks waiting for | |
318 | * changes in the flags of extent states ever. | |
319 | */ | |
320 | ASSERT(!waitqueue_active(&state->wq)); | |
321 | free_extent_state(state); | |
322 | ||
323 | cond_resched_lock(&tree->lock); | |
324 | } | |
325 | spin_unlock(&tree->lock); | |
326 | } | |
327 | ||
b2950863 | 328 | static struct extent_state *alloc_extent_state(gfp_t mask) |
d1310b2e CM |
329 | { |
330 | struct extent_state *state; | |
d1310b2e | 331 | |
3ba7ab22 MH |
332 | /* |
333 | * The given mask might be not appropriate for the slab allocator, | |
334 | * drop the unsupported bits | |
335 | */ | |
336 | mask &= ~(__GFP_DMA32|__GFP_HIGHMEM); | |
d1310b2e | 337 | state = kmem_cache_alloc(extent_state_cache, mask); |
2b114d1d | 338 | if (!state) |
d1310b2e CM |
339 | return state; |
340 | state->state = 0; | |
47dc196a | 341 | state->failrec = NULL; |
27a3507d | 342 | RB_CLEAR_NODE(&state->rb_node); |
3fd63727 | 343 | btrfs_leak_debug_add(&leak_lock, &state->leak_list, &states); |
b7ac31b7 | 344 | refcount_set(&state->refs, 1); |
d1310b2e | 345 | init_waitqueue_head(&state->wq); |
143bede5 | 346 | trace_alloc_extent_state(state, mask, _RET_IP_); |
d1310b2e CM |
347 | return state; |
348 | } | |
d1310b2e | 349 | |
4845e44f | 350 | void free_extent_state(struct extent_state *state) |
d1310b2e | 351 | { |
d1310b2e CM |
352 | if (!state) |
353 | return; | |
b7ac31b7 | 354 | if (refcount_dec_and_test(&state->refs)) { |
27a3507d | 355 | WARN_ON(extent_state_in_tree(state)); |
3fd63727 | 356 | btrfs_leak_debug_del(&leak_lock, &state->leak_list); |
143bede5 | 357 | trace_free_extent_state(state, _RET_IP_); |
d1310b2e CM |
358 | kmem_cache_free(extent_state_cache, state); |
359 | } | |
360 | } | |
d1310b2e | 361 | |
f2071b21 FM |
362 | static struct rb_node *tree_insert(struct rb_root *root, |
363 | struct rb_node *search_start, | |
364 | u64 offset, | |
12cfbad9 FDBM |
365 | struct rb_node *node, |
366 | struct rb_node ***p_in, | |
367 | struct rb_node **parent_in) | |
d1310b2e | 368 | { |
f2071b21 | 369 | struct rb_node **p; |
d397712b | 370 | struct rb_node *parent = NULL; |
d1310b2e CM |
371 | struct tree_entry *entry; |
372 | ||
12cfbad9 FDBM |
373 | if (p_in && parent_in) { |
374 | p = *p_in; | |
375 | parent = *parent_in; | |
376 | goto do_insert; | |
377 | } | |
378 | ||
f2071b21 | 379 | p = search_start ? &search_start : &root->rb_node; |
d397712b | 380 | while (*p) { |
d1310b2e CM |
381 | parent = *p; |
382 | entry = rb_entry(parent, struct tree_entry, rb_node); | |
383 | ||
384 | if (offset < entry->start) | |
385 | p = &(*p)->rb_left; | |
386 | else if (offset > entry->end) | |
387 | p = &(*p)->rb_right; | |
388 | else | |
389 | return parent; | |
390 | } | |
391 | ||
12cfbad9 | 392 | do_insert: |
d1310b2e CM |
393 | rb_link_node(node, parent, p); |
394 | rb_insert_color(node, root); | |
395 | return NULL; | |
396 | } | |
397 | ||
8666e638 | 398 | /** |
3bed2da1 NB |
399 | * Search @tree for an entry that contains @offset. Such entry would have |
400 | * entry->start <= offset && entry->end >= offset. | |
8666e638 | 401 | * |
3bed2da1 NB |
402 | * @tree: the tree to search |
403 | * @offset: offset that should fall within an entry in @tree | |
404 | * @next_ret: pointer to the first entry whose range ends after @offset | |
405 | * @prev_ret: pointer to the first entry whose range begins before @offset | |
406 | * @p_ret: pointer where new node should be anchored (used when inserting an | |
407 | * entry in the tree) | |
408 | * @parent_ret: points to entry which would have been the parent of the entry, | |
8666e638 NB |
409 | * containing @offset |
410 | * | |
411 | * This function returns a pointer to the entry that contains @offset byte | |
412 | * address. If no such entry exists, then NULL is returned and the other | |
413 | * pointer arguments to the function are filled, otherwise the found entry is | |
414 | * returned and other pointers are left untouched. | |
415 | */ | |
80ea96b1 | 416 | static struct rb_node *__etree_search(struct extent_io_tree *tree, u64 offset, |
12cfbad9 | 417 | struct rb_node **next_ret, |
352646c7 | 418 | struct rb_node **prev_ret, |
12cfbad9 FDBM |
419 | struct rb_node ***p_ret, |
420 | struct rb_node **parent_ret) | |
d1310b2e | 421 | { |
80ea96b1 | 422 | struct rb_root *root = &tree->state; |
12cfbad9 | 423 | struct rb_node **n = &root->rb_node; |
d1310b2e CM |
424 | struct rb_node *prev = NULL; |
425 | struct rb_node *orig_prev = NULL; | |
426 | struct tree_entry *entry; | |
427 | struct tree_entry *prev_entry = NULL; | |
428 | ||
12cfbad9 FDBM |
429 | while (*n) { |
430 | prev = *n; | |
431 | entry = rb_entry(prev, struct tree_entry, rb_node); | |
d1310b2e CM |
432 | prev_entry = entry; |
433 | ||
434 | if (offset < entry->start) | |
12cfbad9 | 435 | n = &(*n)->rb_left; |
d1310b2e | 436 | else if (offset > entry->end) |
12cfbad9 | 437 | n = &(*n)->rb_right; |
d397712b | 438 | else |
12cfbad9 | 439 | return *n; |
d1310b2e CM |
440 | } |
441 | ||
12cfbad9 FDBM |
442 | if (p_ret) |
443 | *p_ret = n; | |
444 | if (parent_ret) | |
445 | *parent_ret = prev; | |
446 | ||
352646c7 | 447 | if (next_ret) { |
d1310b2e | 448 | orig_prev = prev; |
d397712b | 449 | while (prev && offset > prev_entry->end) { |
d1310b2e CM |
450 | prev = rb_next(prev); |
451 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
452 | } | |
352646c7 | 453 | *next_ret = prev; |
d1310b2e CM |
454 | prev = orig_prev; |
455 | } | |
456 | ||
352646c7 | 457 | if (prev_ret) { |
d1310b2e | 458 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); |
d397712b | 459 | while (prev && offset < prev_entry->start) { |
d1310b2e CM |
460 | prev = rb_prev(prev); |
461 | prev_entry = rb_entry(prev, struct tree_entry, rb_node); | |
462 | } | |
352646c7 | 463 | *prev_ret = prev; |
d1310b2e CM |
464 | } |
465 | return NULL; | |
466 | } | |
467 | ||
12cfbad9 FDBM |
468 | static inline struct rb_node * |
469 | tree_search_for_insert(struct extent_io_tree *tree, | |
470 | u64 offset, | |
471 | struct rb_node ***p_ret, | |
472 | struct rb_node **parent_ret) | |
d1310b2e | 473 | { |
352646c7 | 474 | struct rb_node *next= NULL; |
d1310b2e | 475 | struct rb_node *ret; |
70dec807 | 476 | |
352646c7 | 477 | ret = __etree_search(tree, offset, &next, NULL, p_ret, parent_ret); |
d397712b | 478 | if (!ret) |
352646c7 | 479 | return next; |
d1310b2e CM |
480 | return ret; |
481 | } | |
482 | ||
12cfbad9 FDBM |
483 | static inline struct rb_node *tree_search(struct extent_io_tree *tree, |
484 | u64 offset) | |
485 | { | |
486 | return tree_search_for_insert(tree, offset, NULL, NULL); | |
487 | } | |
488 | ||
d1310b2e CM |
489 | /* |
490 | * utility function to look for merge candidates inside a given range. | |
491 | * Any extents with matching state are merged together into a single | |
492 | * extent in the tree. Extents with EXTENT_IO in their state field | |
493 | * are not merged because the end_io handlers need to be able to do | |
494 | * operations on them without sleeping (or doing allocations/splits). | |
495 | * | |
496 | * This should be called with the tree lock held. | |
497 | */ | |
1bf85046 JM |
498 | static void merge_state(struct extent_io_tree *tree, |
499 | struct extent_state *state) | |
d1310b2e CM |
500 | { |
501 | struct extent_state *other; | |
502 | struct rb_node *other_node; | |
503 | ||
8882679e | 504 | if (state->state & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
1bf85046 | 505 | return; |
d1310b2e CM |
506 | |
507 | other_node = rb_prev(&state->rb_node); | |
508 | if (other_node) { | |
509 | other = rb_entry(other_node, struct extent_state, rb_node); | |
510 | if (other->end == state->start - 1 && | |
511 | other->state == state->state) { | |
5c848198 NB |
512 | if (tree->private_data && |
513 | is_data_inode(tree->private_data)) | |
514 | btrfs_merge_delalloc_extent(tree->private_data, | |
515 | state, other); | |
d1310b2e | 516 | state->start = other->start; |
d1310b2e | 517 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 518 | RB_CLEAR_NODE(&other->rb_node); |
d1310b2e CM |
519 | free_extent_state(other); |
520 | } | |
521 | } | |
522 | other_node = rb_next(&state->rb_node); | |
523 | if (other_node) { | |
524 | other = rb_entry(other_node, struct extent_state, rb_node); | |
525 | if (other->start == state->end + 1 && | |
526 | other->state == state->state) { | |
5c848198 NB |
527 | if (tree->private_data && |
528 | is_data_inode(tree->private_data)) | |
529 | btrfs_merge_delalloc_extent(tree->private_data, | |
530 | state, other); | |
df98b6e2 | 531 | state->end = other->end; |
df98b6e2 | 532 | rb_erase(&other->rb_node, &tree->state); |
27a3507d | 533 | RB_CLEAR_NODE(&other->rb_node); |
df98b6e2 | 534 | free_extent_state(other); |
d1310b2e CM |
535 | } |
536 | } | |
d1310b2e CM |
537 | } |
538 | ||
3150b699 | 539 | static void set_state_bits(struct extent_io_tree *tree, |
f97e27e9 | 540 | struct extent_state *state, u32 *bits, |
d38ed27f | 541 | struct extent_changeset *changeset); |
3150b699 | 542 | |
d1310b2e CM |
543 | /* |
544 | * insert an extent_state struct into the tree. 'bits' are set on the | |
545 | * struct before it is inserted. | |
546 | * | |
547 | * This may return -EEXIST if the extent is already there, in which case the | |
548 | * state struct is freed. | |
549 | * | |
550 | * The tree lock is not taken internally. This is a utility function and | |
551 | * probably isn't what you want to call (see set/clear_extent_bit). | |
552 | */ | |
553 | static int insert_state(struct extent_io_tree *tree, | |
554 | struct extent_state *state, u64 start, u64 end, | |
12cfbad9 FDBM |
555 | struct rb_node ***p, |
556 | struct rb_node **parent, | |
f97e27e9 | 557 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e CM |
558 | { |
559 | struct rb_node *node; | |
560 | ||
2792237d DS |
561 | if (end < start) { |
562 | btrfs_err(tree->fs_info, | |
563 | "insert state: end < start %llu %llu", end, start); | |
564 | WARN_ON(1); | |
565 | } | |
d1310b2e CM |
566 | state->start = start; |
567 | state->end = end; | |
9ed74f2d | 568 | |
d38ed27f | 569 | set_state_bits(tree, state, bits, changeset); |
3150b699 | 570 | |
f2071b21 | 571 | node = tree_insert(&tree->state, NULL, end, &state->rb_node, p, parent); |
d1310b2e CM |
572 | if (node) { |
573 | struct extent_state *found; | |
574 | found = rb_entry(node, struct extent_state, rb_node); | |
2792237d DS |
575 | btrfs_err(tree->fs_info, |
576 | "found node %llu %llu on insert of %llu %llu", | |
c1c9ff7c | 577 | found->start, found->end, start, end); |
d1310b2e CM |
578 | return -EEXIST; |
579 | } | |
580 | merge_state(tree, state); | |
581 | return 0; | |
582 | } | |
583 | ||
584 | /* | |
585 | * split a given extent state struct in two, inserting the preallocated | |
586 | * struct 'prealloc' as the newly created second half. 'split' indicates an | |
587 | * offset inside 'orig' where it should be split. | |
588 | * | |
589 | * Before calling, | |
590 | * the tree has 'orig' at [orig->start, orig->end]. After calling, there | |
591 | * are two extent state structs in the tree: | |
592 | * prealloc: [orig->start, split - 1] | |
593 | * orig: [ split, orig->end ] | |
594 | * | |
595 | * The tree locks are not taken by this function. They need to be held | |
596 | * by the caller. | |
597 | */ | |
598 | static int split_state(struct extent_io_tree *tree, struct extent_state *orig, | |
599 | struct extent_state *prealloc, u64 split) | |
600 | { | |
601 | struct rb_node *node; | |
9ed74f2d | 602 | |
abbb55f4 NB |
603 | if (tree->private_data && is_data_inode(tree->private_data)) |
604 | btrfs_split_delalloc_extent(tree->private_data, orig, split); | |
9ed74f2d | 605 | |
d1310b2e CM |
606 | prealloc->start = orig->start; |
607 | prealloc->end = split - 1; | |
608 | prealloc->state = orig->state; | |
609 | orig->start = split; | |
610 | ||
f2071b21 FM |
611 | node = tree_insert(&tree->state, &orig->rb_node, prealloc->end, |
612 | &prealloc->rb_node, NULL, NULL); | |
d1310b2e | 613 | if (node) { |
d1310b2e CM |
614 | free_extent_state(prealloc); |
615 | return -EEXIST; | |
616 | } | |
617 | return 0; | |
618 | } | |
619 | ||
cdc6a395 LZ |
620 | static struct extent_state *next_state(struct extent_state *state) |
621 | { | |
622 | struct rb_node *next = rb_next(&state->rb_node); | |
623 | if (next) | |
624 | return rb_entry(next, struct extent_state, rb_node); | |
625 | else | |
626 | return NULL; | |
627 | } | |
628 | ||
d1310b2e CM |
629 | /* |
630 | * utility function to clear some bits in an extent state struct. | |
52042d8e | 631 | * it will optionally wake up anyone waiting on this state (wake == 1). |
d1310b2e CM |
632 | * |
633 | * If no bits are set on the state struct after clearing things, the | |
634 | * struct is freed and removed from the tree | |
635 | */ | |
cdc6a395 LZ |
636 | static struct extent_state *clear_state_bit(struct extent_io_tree *tree, |
637 | struct extent_state *state, | |
f97e27e9 | 638 | u32 *bits, int wake, |
fefdc557 | 639 | struct extent_changeset *changeset) |
d1310b2e | 640 | { |
cdc6a395 | 641 | struct extent_state *next; |
f97e27e9 | 642 | u32 bits_to_clear = *bits & ~EXTENT_CTLBITS; |
57599c7e | 643 | int ret; |
d1310b2e | 644 | |
0ca1f7ce | 645 | if ((bits_to_clear & EXTENT_DIRTY) && (state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
646 | u64 range = state->end - state->start + 1; |
647 | WARN_ON(range > tree->dirty_bytes); | |
648 | tree->dirty_bytes -= range; | |
649 | } | |
a36bb5f9 NB |
650 | |
651 | if (tree->private_data && is_data_inode(tree->private_data)) | |
652 | btrfs_clear_delalloc_extent(tree->private_data, state, bits); | |
653 | ||
57599c7e DS |
654 | ret = add_extent_changeset(state, bits_to_clear, changeset, 0); |
655 | BUG_ON(ret < 0); | |
32c00aff | 656 | state->state &= ~bits_to_clear; |
d1310b2e CM |
657 | if (wake) |
658 | wake_up(&state->wq); | |
0ca1f7ce | 659 | if (state->state == 0) { |
cdc6a395 | 660 | next = next_state(state); |
27a3507d | 661 | if (extent_state_in_tree(state)) { |
d1310b2e | 662 | rb_erase(&state->rb_node, &tree->state); |
27a3507d | 663 | RB_CLEAR_NODE(&state->rb_node); |
d1310b2e CM |
664 | free_extent_state(state); |
665 | } else { | |
666 | WARN_ON(1); | |
667 | } | |
668 | } else { | |
669 | merge_state(tree, state); | |
cdc6a395 | 670 | next = next_state(state); |
d1310b2e | 671 | } |
cdc6a395 | 672 | return next; |
d1310b2e CM |
673 | } |
674 | ||
8233767a XG |
675 | static struct extent_state * |
676 | alloc_extent_state_atomic(struct extent_state *prealloc) | |
677 | { | |
678 | if (!prealloc) | |
679 | prealloc = alloc_extent_state(GFP_ATOMIC); | |
680 | ||
681 | return prealloc; | |
682 | } | |
683 | ||
48a3b636 | 684 | static void extent_io_tree_panic(struct extent_io_tree *tree, int err) |
c2d904e0 | 685 | { |
29b665cc | 686 | btrfs_panic(tree->fs_info, err, |
05912a3c | 687 | "locking error: extent tree was modified by another thread while locked"); |
c2d904e0 JM |
688 | } |
689 | ||
d1310b2e CM |
690 | /* |
691 | * clear some bits on a range in the tree. This may require splitting | |
692 | * or inserting elements in the tree, so the gfp mask is used to | |
693 | * indicate which allocations or sleeping are allowed. | |
694 | * | |
695 | * pass 'wake' == 1 to kick any sleepers, and 'delete' == 1 to remove | |
696 | * the given range from the tree regardless of state (ie for truncate). | |
697 | * | |
698 | * the range [start, end] is inclusive. | |
699 | * | |
6763af84 | 700 | * This takes the tree lock, and returns 0 on success and < 0 on error. |
d1310b2e | 701 | */ |
66b0c887 | 702 | int __clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 QW |
703 | u32 bits, int wake, int delete, |
704 | struct extent_state **cached_state, | |
705 | gfp_t mask, struct extent_changeset *changeset) | |
d1310b2e CM |
706 | { |
707 | struct extent_state *state; | |
2c64c53d | 708 | struct extent_state *cached; |
d1310b2e CM |
709 | struct extent_state *prealloc = NULL; |
710 | struct rb_node *node; | |
5c939df5 | 711 | u64 last_end; |
d1310b2e | 712 | int err; |
2ac55d41 | 713 | int clear = 0; |
d1310b2e | 714 | |
a5dee37d | 715 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 716 | trace_btrfs_clear_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 717 | |
7ee9e440 JB |
718 | if (bits & EXTENT_DELALLOC) |
719 | bits |= EXTENT_NORESERVE; | |
720 | ||
0ca1f7ce YZ |
721 | if (delete) |
722 | bits |= ~EXTENT_CTLBITS; | |
0ca1f7ce | 723 | |
8882679e | 724 | if (bits & (EXTENT_LOCKED | EXTENT_BOUNDARY)) |
2ac55d41 | 725 | clear = 1; |
d1310b2e | 726 | again: |
d0164adc | 727 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
c7bc6319 FM |
728 | /* |
729 | * Don't care for allocation failure here because we might end | |
730 | * up not needing the pre-allocated extent state at all, which | |
731 | * is the case if we only have in the tree extent states that | |
732 | * cover our input range and don't cover too any other range. | |
733 | * If we end up needing a new extent state we allocate it later. | |
734 | */ | |
d1310b2e | 735 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
736 | } |
737 | ||
cad321ad | 738 | spin_lock(&tree->lock); |
2c64c53d CM |
739 | if (cached_state) { |
740 | cached = *cached_state; | |
2ac55d41 JB |
741 | |
742 | if (clear) { | |
743 | *cached_state = NULL; | |
744 | cached_state = NULL; | |
745 | } | |
746 | ||
27a3507d FM |
747 | if (cached && extent_state_in_tree(cached) && |
748 | cached->start <= start && cached->end > start) { | |
2ac55d41 | 749 | if (clear) |
b7ac31b7 | 750 | refcount_dec(&cached->refs); |
2c64c53d | 751 | state = cached; |
42daec29 | 752 | goto hit_next; |
2c64c53d | 753 | } |
2ac55d41 JB |
754 | if (clear) |
755 | free_extent_state(cached); | |
2c64c53d | 756 | } |
d1310b2e CM |
757 | /* |
758 | * this search will find the extents that end after | |
759 | * our range starts | |
760 | */ | |
80ea96b1 | 761 | node = tree_search(tree, start); |
d1310b2e CM |
762 | if (!node) |
763 | goto out; | |
764 | state = rb_entry(node, struct extent_state, rb_node); | |
2c64c53d | 765 | hit_next: |
d1310b2e CM |
766 | if (state->start > end) |
767 | goto out; | |
768 | WARN_ON(state->end < start); | |
5c939df5 | 769 | last_end = state->end; |
d1310b2e | 770 | |
0449314a | 771 | /* the state doesn't have the wanted bits, go ahead */ |
cdc6a395 LZ |
772 | if (!(state->state & bits)) { |
773 | state = next_state(state); | |
0449314a | 774 | goto next; |
cdc6a395 | 775 | } |
0449314a | 776 | |
d1310b2e CM |
777 | /* |
778 | * | ---- desired range ---- | | |
779 | * | state | or | |
780 | * | ------------- state -------------- | | |
781 | * | |
782 | * We need to split the extent we found, and may flip | |
783 | * bits on second half. | |
784 | * | |
785 | * If the extent we found extends past our range, we | |
786 | * just split and search again. It'll get split again | |
787 | * the next time though. | |
788 | * | |
789 | * If the extent we found is inside our range, we clear | |
790 | * the desired bit on it. | |
791 | */ | |
792 | ||
793 | if (state->start < start) { | |
8233767a XG |
794 | prealloc = alloc_extent_state_atomic(prealloc); |
795 | BUG_ON(!prealloc); | |
d1310b2e | 796 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
797 | if (err) |
798 | extent_io_tree_panic(tree, err); | |
799 | ||
d1310b2e CM |
800 | prealloc = NULL; |
801 | if (err) | |
802 | goto out; | |
803 | if (state->end <= end) { | |
fefdc557 QW |
804 | state = clear_state_bit(tree, state, &bits, wake, |
805 | changeset); | |
d1ac6e41 | 806 | goto next; |
d1310b2e CM |
807 | } |
808 | goto search_again; | |
809 | } | |
810 | /* | |
811 | * | ---- desired range ---- | | |
812 | * | state | | |
813 | * We need to split the extent, and clear the bit | |
814 | * on the first half | |
815 | */ | |
816 | if (state->start <= end && state->end > end) { | |
8233767a XG |
817 | prealloc = alloc_extent_state_atomic(prealloc); |
818 | BUG_ON(!prealloc); | |
d1310b2e | 819 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
820 | if (err) |
821 | extent_io_tree_panic(tree, err); | |
822 | ||
d1310b2e CM |
823 | if (wake) |
824 | wake_up(&state->wq); | |
42daec29 | 825 | |
fefdc557 | 826 | clear_state_bit(tree, prealloc, &bits, wake, changeset); |
9ed74f2d | 827 | |
d1310b2e CM |
828 | prealloc = NULL; |
829 | goto out; | |
830 | } | |
42daec29 | 831 | |
fefdc557 | 832 | state = clear_state_bit(tree, state, &bits, wake, changeset); |
0449314a | 833 | next: |
5c939df5 YZ |
834 | if (last_end == (u64)-1) |
835 | goto out; | |
836 | start = last_end + 1; | |
cdc6a395 | 837 | if (start <= end && state && !need_resched()) |
692e5759 | 838 | goto hit_next; |
d1310b2e CM |
839 | |
840 | search_again: | |
841 | if (start > end) | |
842 | goto out; | |
cad321ad | 843 | spin_unlock(&tree->lock); |
d0164adc | 844 | if (gfpflags_allow_blocking(mask)) |
d1310b2e CM |
845 | cond_resched(); |
846 | goto again; | |
7ab5cb2a DS |
847 | |
848 | out: | |
849 | spin_unlock(&tree->lock); | |
850 | if (prealloc) | |
851 | free_extent_state(prealloc); | |
852 | ||
853 | return 0; | |
854 | ||
d1310b2e | 855 | } |
d1310b2e | 856 | |
143bede5 JM |
857 | static void wait_on_state(struct extent_io_tree *tree, |
858 | struct extent_state *state) | |
641f5219 CH |
859 | __releases(tree->lock) |
860 | __acquires(tree->lock) | |
d1310b2e CM |
861 | { |
862 | DEFINE_WAIT(wait); | |
863 | prepare_to_wait(&state->wq, &wait, TASK_UNINTERRUPTIBLE); | |
cad321ad | 864 | spin_unlock(&tree->lock); |
d1310b2e | 865 | schedule(); |
cad321ad | 866 | spin_lock(&tree->lock); |
d1310b2e | 867 | finish_wait(&state->wq, &wait); |
d1310b2e CM |
868 | } |
869 | ||
870 | /* | |
871 | * waits for one or more bits to clear on a range in the state tree. | |
872 | * The range [start, end] is inclusive. | |
873 | * The tree lock is taken by this function | |
874 | */ | |
41074888 | 875 | static void wait_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 876 | u32 bits) |
d1310b2e CM |
877 | { |
878 | struct extent_state *state; | |
879 | struct rb_node *node; | |
880 | ||
a5dee37d | 881 | btrfs_debug_check_extent_io_range(tree, start, end); |
8d599ae1 | 882 | |
cad321ad | 883 | spin_lock(&tree->lock); |
d1310b2e CM |
884 | again: |
885 | while (1) { | |
886 | /* | |
887 | * this search will find all the extents that end after | |
888 | * our range starts | |
889 | */ | |
80ea96b1 | 890 | node = tree_search(tree, start); |
c50d3e71 | 891 | process_node: |
d1310b2e CM |
892 | if (!node) |
893 | break; | |
894 | ||
895 | state = rb_entry(node, struct extent_state, rb_node); | |
896 | ||
897 | if (state->start > end) | |
898 | goto out; | |
899 | ||
900 | if (state->state & bits) { | |
901 | start = state->start; | |
b7ac31b7 | 902 | refcount_inc(&state->refs); |
d1310b2e CM |
903 | wait_on_state(tree, state); |
904 | free_extent_state(state); | |
905 | goto again; | |
906 | } | |
907 | start = state->end + 1; | |
908 | ||
909 | if (start > end) | |
910 | break; | |
911 | ||
c50d3e71 FM |
912 | if (!cond_resched_lock(&tree->lock)) { |
913 | node = rb_next(node); | |
914 | goto process_node; | |
915 | } | |
d1310b2e CM |
916 | } |
917 | out: | |
cad321ad | 918 | spin_unlock(&tree->lock); |
d1310b2e | 919 | } |
d1310b2e | 920 | |
1bf85046 | 921 | static void set_state_bits(struct extent_io_tree *tree, |
d1310b2e | 922 | struct extent_state *state, |
f97e27e9 | 923 | u32 *bits, struct extent_changeset *changeset) |
d1310b2e | 924 | { |
f97e27e9 | 925 | u32 bits_to_set = *bits & ~EXTENT_CTLBITS; |
57599c7e | 926 | int ret; |
9ed74f2d | 927 | |
e06a1fc9 NB |
928 | if (tree->private_data && is_data_inode(tree->private_data)) |
929 | btrfs_set_delalloc_extent(tree->private_data, state, bits); | |
930 | ||
0ca1f7ce | 931 | if ((bits_to_set & EXTENT_DIRTY) && !(state->state & EXTENT_DIRTY)) { |
d1310b2e CM |
932 | u64 range = state->end - state->start + 1; |
933 | tree->dirty_bytes += range; | |
934 | } | |
57599c7e DS |
935 | ret = add_extent_changeset(state, bits_to_set, changeset, 1); |
936 | BUG_ON(ret < 0); | |
0ca1f7ce | 937 | state->state |= bits_to_set; |
d1310b2e CM |
938 | } |
939 | ||
e38e2ed7 FM |
940 | static void cache_state_if_flags(struct extent_state *state, |
941 | struct extent_state **cached_ptr, | |
9ee49a04 | 942 | unsigned flags) |
2c64c53d CM |
943 | { |
944 | if (cached_ptr && !(*cached_ptr)) { | |
e38e2ed7 | 945 | if (!flags || (state->state & flags)) { |
2c64c53d | 946 | *cached_ptr = state; |
b7ac31b7 | 947 | refcount_inc(&state->refs); |
2c64c53d CM |
948 | } |
949 | } | |
950 | } | |
951 | ||
e38e2ed7 FM |
952 | static void cache_state(struct extent_state *state, |
953 | struct extent_state **cached_ptr) | |
954 | { | |
955 | return cache_state_if_flags(state, cached_ptr, | |
8882679e | 956 | EXTENT_LOCKED | EXTENT_BOUNDARY); |
e38e2ed7 FM |
957 | } |
958 | ||
d1310b2e | 959 | /* |
1edbb734 CM |
960 | * set some bits on a range in the tree. This may require allocations or |
961 | * sleeping, so the gfp mask is used to indicate what is allowed. | |
d1310b2e | 962 | * |
1edbb734 CM |
963 | * If any of the exclusive bits are set, this will fail with -EEXIST if some |
964 | * part of the range already has the desired bits set. The start of the | |
965 | * existing range is returned in failed_start in this case. | |
d1310b2e | 966 | * |
1edbb734 | 967 | * [start, end] is inclusive This takes the tree lock. |
d1310b2e | 968 | */ |
f97e27e9 QW |
969 | int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, u32 bits, |
970 | u32 exclusive_bits, u64 *failed_start, | |
1cab5e72 NB |
971 | struct extent_state **cached_state, gfp_t mask, |
972 | struct extent_changeset *changeset) | |
d1310b2e CM |
973 | { |
974 | struct extent_state *state; | |
975 | struct extent_state *prealloc = NULL; | |
976 | struct rb_node *node; | |
12cfbad9 FDBM |
977 | struct rb_node **p; |
978 | struct rb_node *parent; | |
d1310b2e | 979 | int err = 0; |
d1310b2e CM |
980 | u64 last_start; |
981 | u64 last_end; | |
42daec29 | 982 | |
a5dee37d | 983 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 | 984 | trace_btrfs_set_extent_bit(tree, start, end - start + 1, bits); |
8d599ae1 | 985 | |
3f6bb4ae QW |
986 | if (exclusive_bits) |
987 | ASSERT(failed_start); | |
988 | else | |
989 | ASSERT(failed_start == NULL); | |
d1310b2e | 990 | again: |
d0164adc | 991 | if (!prealloc && gfpflags_allow_blocking(mask)) { |
059f791c DS |
992 | /* |
993 | * Don't care for allocation failure here because we might end | |
994 | * up not needing the pre-allocated extent state at all, which | |
995 | * is the case if we only have in the tree extent states that | |
996 | * cover our input range and don't cover too any other range. | |
997 | * If we end up needing a new extent state we allocate it later. | |
998 | */ | |
d1310b2e | 999 | prealloc = alloc_extent_state(mask); |
d1310b2e CM |
1000 | } |
1001 | ||
cad321ad | 1002 | spin_lock(&tree->lock); |
9655d298 CM |
1003 | if (cached_state && *cached_state) { |
1004 | state = *cached_state; | |
df98b6e2 | 1005 | if (state->start <= start && state->end > start && |
27a3507d | 1006 | extent_state_in_tree(state)) { |
9655d298 CM |
1007 | node = &state->rb_node; |
1008 | goto hit_next; | |
1009 | } | |
1010 | } | |
d1310b2e CM |
1011 | /* |
1012 | * this search will find all the extents that end after | |
1013 | * our range starts. | |
1014 | */ | |
12cfbad9 | 1015 | node = tree_search_for_insert(tree, start, &p, &parent); |
d1310b2e | 1016 | if (!node) { |
8233767a XG |
1017 | prealloc = alloc_extent_state_atomic(prealloc); |
1018 | BUG_ON(!prealloc); | |
12cfbad9 | 1019 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1020 | &p, &parent, &bits, changeset); |
c2d904e0 JM |
1021 | if (err) |
1022 | extent_io_tree_panic(tree, err); | |
1023 | ||
c42ac0bc | 1024 | cache_state(prealloc, cached_state); |
d1310b2e | 1025 | prealloc = NULL; |
d1310b2e CM |
1026 | goto out; |
1027 | } | |
d1310b2e | 1028 | state = rb_entry(node, struct extent_state, rb_node); |
40431d6c | 1029 | hit_next: |
d1310b2e CM |
1030 | last_start = state->start; |
1031 | last_end = state->end; | |
1032 | ||
1033 | /* | |
1034 | * | ---- desired range ---- | | |
1035 | * | state | | |
1036 | * | |
1037 | * Just lock what we found and keep going | |
1038 | */ | |
1039 | if (state->start == start && state->end <= end) { | |
1edbb734 | 1040 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1041 | *failed_start = state->start; |
1042 | err = -EEXIST; | |
1043 | goto out; | |
1044 | } | |
42daec29 | 1045 | |
d38ed27f | 1046 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1047 | cache_state(state, cached_state); |
d1310b2e | 1048 | merge_state(tree, state); |
5c939df5 YZ |
1049 | if (last_end == (u64)-1) |
1050 | goto out; | |
1051 | start = last_end + 1; | |
d1ac6e41 LB |
1052 | state = next_state(state); |
1053 | if (start < end && state && state->start == start && | |
1054 | !need_resched()) | |
1055 | goto hit_next; | |
d1310b2e CM |
1056 | goto search_again; |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * | ---- desired range ---- | | |
1061 | * | state | | |
1062 | * or | |
1063 | * | ------------- state -------------- | | |
1064 | * | |
1065 | * We need to split the extent we found, and may flip bits on | |
1066 | * second half. | |
1067 | * | |
1068 | * If the extent we found extends past our | |
1069 | * range, we just split and search again. It'll get split | |
1070 | * again the next time though. | |
1071 | * | |
1072 | * If the extent we found is inside our range, we set the | |
1073 | * desired bit on it. | |
1074 | */ | |
1075 | if (state->start < start) { | |
1edbb734 | 1076 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1077 | *failed_start = start; |
1078 | err = -EEXIST; | |
1079 | goto out; | |
1080 | } | |
8233767a | 1081 | |
55ffaabe FM |
1082 | /* |
1083 | * If this extent already has all the bits we want set, then | |
1084 | * skip it, not necessary to split it or do anything with it. | |
1085 | */ | |
1086 | if ((state->state & bits) == bits) { | |
1087 | start = state->end + 1; | |
1088 | cache_state(state, cached_state); | |
1089 | goto search_again; | |
1090 | } | |
1091 | ||
8233767a XG |
1092 | prealloc = alloc_extent_state_atomic(prealloc); |
1093 | BUG_ON(!prealloc); | |
d1310b2e | 1094 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1095 | if (err) |
1096 | extent_io_tree_panic(tree, err); | |
1097 | ||
d1310b2e CM |
1098 | prealloc = NULL; |
1099 | if (err) | |
1100 | goto out; | |
1101 | if (state->end <= end) { | |
d38ed27f | 1102 | set_state_bits(tree, state, &bits, changeset); |
2c64c53d | 1103 | cache_state(state, cached_state); |
d1310b2e | 1104 | merge_state(tree, state); |
5c939df5 YZ |
1105 | if (last_end == (u64)-1) |
1106 | goto out; | |
1107 | start = last_end + 1; | |
d1ac6e41 LB |
1108 | state = next_state(state); |
1109 | if (start < end && state && state->start == start && | |
1110 | !need_resched()) | |
1111 | goto hit_next; | |
d1310b2e CM |
1112 | } |
1113 | goto search_again; | |
1114 | } | |
1115 | /* | |
1116 | * | ---- desired range ---- | | |
1117 | * | state | or | state | | |
1118 | * | |
1119 | * There's a hole, we need to insert something in it and | |
1120 | * ignore the extent we found. | |
1121 | */ | |
1122 | if (state->start > start) { | |
1123 | u64 this_end; | |
1124 | if (end < last_start) | |
1125 | this_end = end; | |
1126 | else | |
d397712b | 1127 | this_end = last_start - 1; |
8233767a XG |
1128 | |
1129 | prealloc = alloc_extent_state_atomic(prealloc); | |
1130 | BUG_ON(!prealloc); | |
c7f895a2 XG |
1131 | |
1132 | /* | |
1133 | * Avoid to free 'prealloc' if it can be merged with | |
1134 | * the later extent. | |
1135 | */ | |
d1310b2e | 1136 | err = insert_state(tree, prealloc, start, this_end, |
d38ed27f | 1137 | NULL, NULL, &bits, changeset); |
c2d904e0 JM |
1138 | if (err) |
1139 | extent_io_tree_panic(tree, err); | |
1140 | ||
9ed74f2d JB |
1141 | cache_state(prealloc, cached_state); |
1142 | prealloc = NULL; | |
d1310b2e CM |
1143 | start = this_end + 1; |
1144 | goto search_again; | |
1145 | } | |
1146 | /* | |
1147 | * | ---- desired range ---- | | |
1148 | * | state | | |
1149 | * We need to split the extent, and set the bit | |
1150 | * on the first half | |
1151 | */ | |
1152 | if (state->start <= end && state->end > end) { | |
1edbb734 | 1153 | if (state->state & exclusive_bits) { |
d1310b2e CM |
1154 | *failed_start = start; |
1155 | err = -EEXIST; | |
1156 | goto out; | |
1157 | } | |
8233767a XG |
1158 | |
1159 | prealloc = alloc_extent_state_atomic(prealloc); | |
1160 | BUG_ON(!prealloc); | |
d1310b2e | 1161 | err = split_state(tree, state, prealloc, end + 1); |
c2d904e0 JM |
1162 | if (err) |
1163 | extent_io_tree_panic(tree, err); | |
d1310b2e | 1164 | |
d38ed27f | 1165 | set_state_bits(tree, prealloc, &bits, changeset); |
2c64c53d | 1166 | cache_state(prealloc, cached_state); |
d1310b2e CM |
1167 | merge_state(tree, prealloc); |
1168 | prealloc = NULL; | |
1169 | goto out; | |
1170 | } | |
1171 | ||
b5a4ba14 DS |
1172 | search_again: |
1173 | if (start > end) | |
1174 | goto out; | |
1175 | spin_unlock(&tree->lock); | |
1176 | if (gfpflags_allow_blocking(mask)) | |
1177 | cond_resched(); | |
1178 | goto again; | |
d1310b2e CM |
1179 | |
1180 | out: | |
cad321ad | 1181 | spin_unlock(&tree->lock); |
d1310b2e CM |
1182 | if (prealloc) |
1183 | free_extent_state(prealloc); | |
1184 | ||
1185 | return err; | |
1186 | ||
d1310b2e | 1187 | } |
d1310b2e | 1188 | |
462d6fac | 1189 | /** |
10983f2e LB |
1190 | * convert_extent_bit - convert all bits in a given range from one bit to |
1191 | * another | |
462d6fac JB |
1192 | * @tree: the io tree to search |
1193 | * @start: the start offset in bytes | |
1194 | * @end: the end offset in bytes (inclusive) | |
1195 | * @bits: the bits to set in this range | |
1196 | * @clear_bits: the bits to clear in this range | |
e6138876 | 1197 | * @cached_state: state that we're going to cache |
462d6fac JB |
1198 | * |
1199 | * This will go through and set bits for the given range. If any states exist | |
1200 | * already in this range they are set with the given bit and cleared of the | |
1201 | * clear_bits. This is only meant to be used by things that are mergeable, ie | |
1202 | * converting from say DELALLOC to DIRTY. This is not meant to be used with | |
1203 | * boundary bits like LOCK. | |
210aa277 DS |
1204 | * |
1205 | * All allocations are done with GFP_NOFS. | |
462d6fac JB |
1206 | */ |
1207 | int convert_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 1208 | u32 bits, u32 clear_bits, |
210aa277 | 1209 | struct extent_state **cached_state) |
462d6fac JB |
1210 | { |
1211 | struct extent_state *state; | |
1212 | struct extent_state *prealloc = NULL; | |
1213 | struct rb_node *node; | |
12cfbad9 FDBM |
1214 | struct rb_node **p; |
1215 | struct rb_node *parent; | |
462d6fac JB |
1216 | int err = 0; |
1217 | u64 last_start; | |
1218 | u64 last_end; | |
c8fd3de7 | 1219 | bool first_iteration = true; |
462d6fac | 1220 | |
a5dee37d | 1221 | btrfs_debug_check_extent_io_range(tree, start, end); |
a1d19847 QW |
1222 | trace_btrfs_convert_extent_bit(tree, start, end - start + 1, bits, |
1223 | clear_bits); | |
8d599ae1 | 1224 | |
462d6fac | 1225 | again: |
210aa277 | 1226 | if (!prealloc) { |
c8fd3de7 FM |
1227 | /* |
1228 | * Best effort, don't worry if extent state allocation fails | |
1229 | * here for the first iteration. We might have a cached state | |
1230 | * that matches exactly the target range, in which case no | |
1231 | * extent state allocations are needed. We'll only know this | |
1232 | * after locking the tree. | |
1233 | */ | |
210aa277 | 1234 | prealloc = alloc_extent_state(GFP_NOFS); |
c8fd3de7 | 1235 | if (!prealloc && !first_iteration) |
462d6fac JB |
1236 | return -ENOMEM; |
1237 | } | |
1238 | ||
1239 | spin_lock(&tree->lock); | |
e6138876 JB |
1240 | if (cached_state && *cached_state) { |
1241 | state = *cached_state; | |
1242 | if (state->start <= start && state->end > start && | |
27a3507d | 1243 | extent_state_in_tree(state)) { |
e6138876 JB |
1244 | node = &state->rb_node; |
1245 | goto hit_next; | |
1246 | } | |
1247 | } | |
1248 | ||
462d6fac JB |
1249 | /* |
1250 | * this search will find all the extents that end after | |
1251 | * our range starts. | |
1252 | */ | |
12cfbad9 | 1253 | node = tree_search_for_insert(tree, start, &p, &parent); |
462d6fac JB |
1254 | if (!node) { |
1255 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1256 | if (!prealloc) { |
1257 | err = -ENOMEM; | |
1258 | goto out; | |
1259 | } | |
12cfbad9 | 1260 | err = insert_state(tree, prealloc, start, end, |
d38ed27f | 1261 | &p, &parent, &bits, NULL); |
c2d904e0 JM |
1262 | if (err) |
1263 | extent_io_tree_panic(tree, err); | |
c42ac0bc FDBM |
1264 | cache_state(prealloc, cached_state); |
1265 | prealloc = NULL; | |
462d6fac JB |
1266 | goto out; |
1267 | } | |
1268 | state = rb_entry(node, struct extent_state, rb_node); | |
1269 | hit_next: | |
1270 | last_start = state->start; | |
1271 | last_end = state->end; | |
1272 | ||
1273 | /* | |
1274 | * | ---- desired range ---- | | |
1275 | * | state | | |
1276 | * | |
1277 | * Just lock what we found and keep going | |
1278 | */ | |
1279 | if (state->start == start && state->end <= end) { | |
d38ed27f | 1280 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1281 | cache_state(state, cached_state); |
fefdc557 | 1282 | state = clear_state_bit(tree, state, &clear_bits, 0, NULL); |
462d6fac JB |
1283 | if (last_end == (u64)-1) |
1284 | goto out; | |
462d6fac | 1285 | start = last_end + 1; |
d1ac6e41 LB |
1286 | if (start < end && state && state->start == start && |
1287 | !need_resched()) | |
1288 | goto hit_next; | |
462d6fac JB |
1289 | goto search_again; |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * | ---- desired range ---- | | |
1294 | * | state | | |
1295 | * or | |
1296 | * | ------------- state -------------- | | |
1297 | * | |
1298 | * We need to split the extent we found, and may flip bits on | |
1299 | * second half. | |
1300 | * | |
1301 | * If the extent we found extends past our | |
1302 | * range, we just split and search again. It'll get split | |
1303 | * again the next time though. | |
1304 | * | |
1305 | * If the extent we found is inside our range, we set the | |
1306 | * desired bit on it. | |
1307 | */ | |
1308 | if (state->start < start) { | |
1309 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1310 | if (!prealloc) { |
1311 | err = -ENOMEM; | |
1312 | goto out; | |
1313 | } | |
462d6fac | 1314 | err = split_state(tree, state, prealloc, start); |
c2d904e0 JM |
1315 | if (err) |
1316 | extent_io_tree_panic(tree, err); | |
462d6fac JB |
1317 | prealloc = NULL; |
1318 | if (err) | |
1319 | goto out; | |
1320 | if (state->end <= end) { | |
d38ed27f | 1321 | set_state_bits(tree, state, &bits, NULL); |
e6138876 | 1322 | cache_state(state, cached_state); |
fefdc557 QW |
1323 | state = clear_state_bit(tree, state, &clear_bits, 0, |
1324 | NULL); | |
462d6fac JB |
1325 | if (last_end == (u64)-1) |
1326 | goto out; | |
1327 | start = last_end + 1; | |
d1ac6e41 LB |
1328 | if (start < end && state && state->start == start && |
1329 | !need_resched()) | |
1330 | goto hit_next; | |
462d6fac JB |
1331 | } |
1332 | goto search_again; | |
1333 | } | |
1334 | /* | |
1335 | * | ---- desired range ---- | | |
1336 | * | state | or | state | | |
1337 | * | |
1338 | * There's a hole, we need to insert something in it and | |
1339 | * ignore the extent we found. | |
1340 | */ | |
1341 | if (state->start > start) { | |
1342 | u64 this_end; | |
1343 | if (end < last_start) | |
1344 | this_end = end; | |
1345 | else | |
1346 | this_end = last_start - 1; | |
1347 | ||
1348 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1349 | if (!prealloc) { |
1350 | err = -ENOMEM; | |
1351 | goto out; | |
1352 | } | |
462d6fac JB |
1353 | |
1354 | /* | |
1355 | * Avoid to free 'prealloc' if it can be merged with | |
1356 | * the later extent. | |
1357 | */ | |
1358 | err = insert_state(tree, prealloc, start, this_end, | |
d38ed27f | 1359 | NULL, NULL, &bits, NULL); |
c2d904e0 JM |
1360 | if (err) |
1361 | extent_io_tree_panic(tree, err); | |
e6138876 | 1362 | cache_state(prealloc, cached_state); |
462d6fac JB |
1363 | prealloc = NULL; |
1364 | start = this_end + 1; | |
1365 | goto search_again; | |
1366 | } | |
1367 | /* | |
1368 | * | ---- desired range ---- | | |
1369 | * | state | | |
1370 | * We need to split the extent, and set the bit | |
1371 | * on the first half | |
1372 | */ | |
1373 | if (state->start <= end && state->end > end) { | |
1374 | prealloc = alloc_extent_state_atomic(prealloc); | |
1cf4ffdb LB |
1375 | if (!prealloc) { |
1376 | err = -ENOMEM; | |
1377 | goto out; | |
1378 | } | |
462d6fac JB |
1379 | |
1380 | err = split_state(tree, state, prealloc, end + 1); | |
c2d904e0 JM |
1381 | if (err) |
1382 | extent_io_tree_panic(tree, err); | |
462d6fac | 1383 | |
d38ed27f | 1384 | set_state_bits(tree, prealloc, &bits, NULL); |
e6138876 | 1385 | cache_state(prealloc, cached_state); |
fefdc557 | 1386 | clear_state_bit(tree, prealloc, &clear_bits, 0, NULL); |
462d6fac JB |
1387 | prealloc = NULL; |
1388 | goto out; | |
1389 | } | |
1390 | ||
462d6fac JB |
1391 | search_again: |
1392 | if (start > end) | |
1393 | goto out; | |
1394 | spin_unlock(&tree->lock); | |
210aa277 | 1395 | cond_resched(); |
c8fd3de7 | 1396 | first_iteration = false; |
462d6fac | 1397 | goto again; |
462d6fac JB |
1398 | |
1399 | out: | |
1400 | spin_unlock(&tree->lock); | |
1401 | if (prealloc) | |
1402 | free_extent_state(prealloc); | |
1403 | ||
1404 | return err; | |
462d6fac JB |
1405 | } |
1406 | ||
d1310b2e | 1407 | /* wrappers around set/clear extent bit */ |
d38ed27f | 1408 | int set_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1409 | u32 bits, struct extent_changeset *changeset) |
d38ed27f QW |
1410 | { |
1411 | /* | |
1412 | * We don't support EXTENT_LOCKED yet, as current changeset will | |
1413 | * record any bits changed, so for EXTENT_LOCKED case, it will | |
1414 | * either fail with -EEXIST or changeset will record the whole | |
1415 | * range. | |
1416 | */ | |
1417 | BUG_ON(bits & EXTENT_LOCKED); | |
1418 | ||
1cab5e72 NB |
1419 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, GFP_NOFS, |
1420 | changeset); | |
d38ed27f QW |
1421 | } |
1422 | ||
4ca73656 | 1423 | int set_extent_bits_nowait(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1424 | u32 bits) |
4ca73656 | 1425 | { |
1cab5e72 NB |
1426 | return set_extent_bit(tree, start, end, bits, 0, NULL, NULL, |
1427 | GFP_NOWAIT, NULL); | |
4ca73656 NB |
1428 | } |
1429 | ||
fefdc557 | 1430 | int clear_extent_bit(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1431 | u32 bits, int wake, int delete, |
ae0f1625 | 1432 | struct extent_state **cached) |
fefdc557 QW |
1433 | { |
1434 | return __clear_extent_bit(tree, start, end, bits, wake, delete, | |
ae0f1625 | 1435 | cached, GFP_NOFS, NULL); |
fefdc557 QW |
1436 | } |
1437 | ||
fefdc557 | 1438 | int clear_record_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
f97e27e9 | 1439 | u32 bits, struct extent_changeset *changeset) |
fefdc557 QW |
1440 | { |
1441 | /* | |
1442 | * Don't support EXTENT_LOCKED case, same reason as | |
1443 | * set_record_extent_bits(). | |
1444 | */ | |
1445 | BUG_ON(bits & EXTENT_LOCKED); | |
1446 | ||
f734c44a | 1447 | return __clear_extent_bit(tree, start, end, bits, 0, 0, NULL, GFP_NOFS, |
fefdc557 QW |
1448 | changeset); |
1449 | } | |
1450 | ||
d352ac68 CM |
1451 | /* |
1452 | * either insert or lock state struct between start and end use mask to tell | |
1453 | * us if waiting is desired. | |
1454 | */ | |
1edbb734 | 1455 | int lock_extent_bits(struct extent_io_tree *tree, u64 start, u64 end, |
ff13db41 | 1456 | struct extent_state **cached_state) |
d1310b2e CM |
1457 | { |
1458 | int err; | |
1459 | u64 failed_start; | |
9ee49a04 | 1460 | |
d1310b2e | 1461 | while (1) { |
1cab5e72 NB |
1462 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, |
1463 | EXTENT_LOCKED, &failed_start, | |
1464 | cached_state, GFP_NOFS, NULL); | |
d0082371 | 1465 | if (err == -EEXIST) { |
d1310b2e CM |
1466 | wait_extent_bit(tree, failed_start, end, EXTENT_LOCKED); |
1467 | start = failed_start; | |
d0082371 | 1468 | } else |
d1310b2e | 1469 | break; |
d1310b2e CM |
1470 | WARN_ON(start > end); |
1471 | } | |
1472 | return err; | |
1473 | } | |
d1310b2e | 1474 | |
d0082371 | 1475 | int try_lock_extent(struct extent_io_tree *tree, u64 start, u64 end) |
25179201 JB |
1476 | { |
1477 | int err; | |
1478 | u64 failed_start; | |
1479 | ||
1cab5e72 NB |
1480 | err = set_extent_bit(tree, start, end, EXTENT_LOCKED, EXTENT_LOCKED, |
1481 | &failed_start, NULL, GFP_NOFS, NULL); | |
6643558d YZ |
1482 | if (err == -EEXIST) { |
1483 | if (failed_start > start) | |
1484 | clear_extent_bit(tree, start, failed_start - 1, | |
ae0f1625 | 1485 | EXTENT_LOCKED, 1, 0, NULL); |
25179201 | 1486 | return 0; |
6643558d | 1487 | } |
25179201 JB |
1488 | return 1; |
1489 | } | |
25179201 | 1490 | |
bd1fa4f0 | 1491 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1492 | { |
09cbfeaf KS |
1493 | unsigned long index = start >> PAGE_SHIFT; |
1494 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1495 | struct page *page; |
1496 | ||
1497 | while (index <= end_index) { | |
1498 | page = find_get_page(inode->i_mapping, index); | |
1499 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
1500 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1501 | put_page(page); |
4adaa611 CM |
1502 | index++; |
1503 | } | |
4adaa611 CM |
1504 | } |
1505 | ||
f6311572 | 1506 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 1507 | { |
09cbfeaf KS |
1508 | unsigned long index = start >> PAGE_SHIFT; |
1509 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
1510 | struct page *page; |
1511 | ||
1512 | while (index <= end_index) { | |
1513 | page = find_get_page(inode->i_mapping, index); | |
1514 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
4adaa611 | 1515 | __set_page_dirty_nobuffers(page); |
8d38633c | 1516 | account_page_redirty(page); |
09cbfeaf | 1517 | put_page(page); |
4adaa611 CM |
1518 | index++; |
1519 | } | |
4adaa611 CM |
1520 | } |
1521 | ||
d352ac68 CM |
1522 | /* find the first state struct with 'bits' set after 'start', and |
1523 | * return it. tree->lock must be held. NULL will returned if | |
1524 | * nothing was found after 'start' | |
1525 | */ | |
48a3b636 | 1526 | static struct extent_state * |
f97e27e9 | 1527 | find_first_extent_bit_state(struct extent_io_tree *tree, u64 start, u32 bits) |
d7fc640e CM |
1528 | { |
1529 | struct rb_node *node; | |
1530 | struct extent_state *state; | |
1531 | ||
1532 | /* | |
1533 | * this search will find all the extents that end after | |
1534 | * our range starts. | |
1535 | */ | |
1536 | node = tree_search(tree, start); | |
d397712b | 1537 | if (!node) |
d7fc640e | 1538 | goto out; |
d7fc640e | 1539 | |
d397712b | 1540 | while (1) { |
d7fc640e | 1541 | state = rb_entry(node, struct extent_state, rb_node); |
d397712b | 1542 | if (state->end >= start && (state->state & bits)) |
d7fc640e | 1543 | return state; |
d397712b | 1544 | |
d7fc640e CM |
1545 | node = rb_next(node); |
1546 | if (!node) | |
1547 | break; | |
1548 | } | |
1549 | out: | |
1550 | return NULL; | |
1551 | } | |
d7fc640e | 1552 | |
69261c4b | 1553 | /* |
03509b78 | 1554 | * Find the first offset in the io tree with one or more @bits set. |
69261c4b | 1555 | * |
03509b78 QW |
1556 | * Note: If there are multiple bits set in @bits, any of them will match. |
1557 | * | |
1558 | * Return 0 if we find something, and update @start_ret and @end_ret. | |
1559 | * Return 1 if we found nothing. | |
69261c4b XG |
1560 | */ |
1561 | int find_first_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1562 | u64 *start_ret, u64 *end_ret, u32 bits, |
e6138876 | 1563 | struct extent_state **cached_state) |
69261c4b XG |
1564 | { |
1565 | struct extent_state *state; | |
1566 | int ret = 1; | |
1567 | ||
1568 | spin_lock(&tree->lock); | |
e6138876 JB |
1569 | if (cached_state && *cached_state) { |
1570 | state = *cached_state; | |
27a3507d | 1571 | if (state->end == start - 1 && extent_state_in_tree(state)) { |
9688e9a9 | 1572 | while ((state = next_state(state)) != NULL) { |
e6138876 JB |
1573 | if (state->state & bits) |
1574 | goto got_it; | |
e6138876 JB |
1575 | } |
1576 | free_extent_state(*cached_state); | |
1577 | *cached_state = NULL; | |
1578 | goto out; | |
1579 | } | |
1580 | free_extent_state(*cached_state); | |
1581 | *cached_state = NULL; | |
1582 | } | |
1583 | ||
69261c4b | 1584 | state = find_first_extent_bit_state(tree, start, bits); |
e6138876 | 1585 | got_it: |
69261c4b | 1586 | if (state) { |
e38e2ed7 | 1587 | cache_state_if_flags(state, cached_state, 0); |
69261c4b XG |
1588 | *start_ret = state->start; |
1589 | *end_ret = state->end; | |
1590 | ret = 0; | |
1591 | } | |
e6138876 | 1592 | out: |
69261c4b XG |
1593 | spin_unlock(&tree->lock); |
1594 | return ret; | |
1595 | } | |
1596 | ||
41a2ee75 | 1597 | /** |
3bed2da1 NB |
1598 | * Find a contiguous area of bits |
1599 | * | |
1600 | * @tree: io tree to check | |
1601 | * @start: offset to start the search from | |
1602 | * @start_ret: the first offset we found with the bits set | |
1603 | * @end_ret: the final contiguous range of the bits that were set | |
1604 | * @bits: bits to look for | |
41a2ee75 JB |
1605 | * |
1606 | * set_extent_bit and clear_extent_bit can temporarily split contiguous ranges | |
1607 | * to set bits appropriately, and then merge them again. During this time it | |
1608 | * will drop the tree->lock, so use this helper if you want to find the actual | |
1609 | * contiguous area for given bits. We will search to the first bit we find, and | |
1610 | * then walk down the tree until we find a non-contiguous area. The area | |
1611 | * returned will be the full contiguous area with the bits set. | |
1612 | */ | |
1613 | int find_contiguous_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1614 | u64 *start_ret, u64 *end_ret, u32 bits) |
41a2ee75 JB |
1615 | { |
1616 | struct extent_state *state; | |
1617 | int ret = 1; | |
1618 | ||
1619 | spin_lock(&tree->lock); | |
1620 | state = find_first_extent_bit_state(tree, start, bits); | |
1621 | if (state) { | |
1622 | *start_ret = state->start; | |
1623 | *end_ret = state->end; | |
1624 | while ((state = next_state(state)) != NULL) { | |
1625 | if (state->start > (*end_ret + 1)) | |
1626 | break; | |
1627 | *end_ret = state->end; | |
1628 | } | |
1629 | ret = 0; | |
1630 | } | |
1631 | spin_unlock(&tree->lock); | |
1632 | return ret; | |
1633 | } | |
1634 | ||
45bfcfc1 | 1635 | /** |
3bed2da1 NB |
1636 | * Find the first range that has @bits not set. This range could start before |
1637 | * @start. | |
45bfcfc1 | 1638 | * |
3bed2da1 NB |
1639 | * @tree: the tree to search |
1640 | * @start: offset at/after which the found extent should start | |
1641 | * @start_ret: records the beginning of the range | |
1642 | * @end_ret: records the end of the range (inclusive) | |
1643 | * @bits: the set of bits which must be unset | |
45bfcfc1 NB |
1644 | * |
1645 | * Since unallocated range is also considered one which doesn't have the bits | |
1646 | * set it's possible that @end_ret contains -1, this happens in case the range | |
1647 | * spans (last_range_end, end of device]. In this case it's up to the caller to | |
1648 | * trim @end_ret to the appropriate size. | |
1649 | */ | |
1650 | void find_first_clear_extent_bit(struct extent_io_tree *tree, u64 start, | |
f97e27e9 | 1651 | u64 *start_ret, u64 *end_ret, u32 bits) |
45bfcfc1 NB |
1652 | { |
1653 | struct extent_state *state; | |
1654 | struct rb_node *node, *prev = NULL, *next; | |
1655 | ||
1656 | spin_lock(&tree->lock); | |
1657 | ||
1658 | /* Find first extent with bits cleared */ | |
1659 | while (1) { | |
1660 | node = __etree_search(tree, start, &next, &prev, NULL, NULL); | |
5750c375 NB |
1661 | if (!node && !next && !prev) { |
1662 | /* | |
1663 | * Tree is completely empty, send full range and let | |
1664 | * caller deal with it | |
1665 | */ | |
1666 | *start_ret = 0; | |
1667 | *end_ret = -1; | |
1668 | goto out; | |
1669 | } else if (!node && !next) { | |
1670 | /* | |
1671 | * We are past the last allocated chunk, set start at | |
1672 | * the end of the last extent. | |
1673 | */ | |
1674 | state = rb_entry(prev, struct extent_state, rb_node); | |
1675 | *start_ret = state->end + 1; | |
1676 | *end_ret = -1; | |
1677 | goto out; | |
1678 | } else if (!node) { | |
45bfcfc1 | 1679 | node = next; |
45bfcfc1 | 1680 | } |
1eaebb34 NB |
1681 | /* |
1682 | * At this point 'node' either contains 'start' or start is | |
1683 | * before 'node' | |
1684 | */ | |
45bfcfc1 | 1685 | state = rb_entry(node, struct extent_state, rb_node); |
1eaebb34 NB |
1686 | |
1687 | if (in_range(start, state->start, state->end - state->start + 1)) { | |
1688 | if (state->state & bits) { | |
1689 | /* | |
1690 | * |--range with bits sets--| | |
1691 | * | | |
1692 | * start | |
1693 | */ | |
1694 | start = state->end + 1; | |
1695 | } else { | |
1696 | /* | |
1697 | * 'start' falls within a range that doesn't | |
1698 | * have the bits set, so take its start as | |
1699 | * the beginning of the desired range | |
1700 | * | |
1701 | * |--range with bits cleared----| | |
1702 | * | | |
1703 | * start | |
1704 | */ | |
1705 | *start_ret = state->start; | |
1706 | break; | |
1707 | } | |
45bfcfc1 | 1708 | } else { |
1eaebb34 NB |
1709 | /* |
1710 | * |---prev range---|---hole/unset---|---node range---| | |
1711 | * | | |
1712 | * start | |
1713 | * | |
1714 | * or | |
1715 | * | |
1716 | * |---hole/unset--||--first node--| | |
1717 | * 0 | | |
1718 | * start | |
1719 | */ | |
1720 | if (prev) { | |
1721 | state = rb_entry(prev, struct extent_state, | |
1722 | rb_node); | |
1723 | *start_ret = state->end + 1; | |
1724 | } else { | |
1725 | *start_ret = 0; | |
1726 | } | |
45bfcfc1 NB |
1727 | break; |
1728 | } | |
1729 | } | |
1730 | ||
1731 | /* | |
1732 | * Find the longest stretch from start until an entry which has the | |
1733 | * bits set | |
1734 | */ | |
1735 | while (1) { | |
1736 | state = rb_entry(node, struct extent_state, rb_node); | |
1737 | if (state->end >= start && !(state->state & bits)) { | |
1738 | *end_ret = state->end; | |
1739 | } else { | |
1740 | *end_ret = state->start - 1; | |
1741 | break; | |
1742 | } | |
1743 | ||
1744 | node = rb_next(node); | |
1745 | if (!node) | |
1746 | break; | |
1747 | } | |
1748 | out: | |
1749 | spin_unlock(&tree->lock); | |
1750 | } | |
1751 | ||
d352ac68 CM |
1752 | /* |
1753 | * find a contiguous range of bytes in the file marked as delalloc, not | |
1754 | * more than 'max_bytes'. start and end are used to return the range, | |
1755 | * | |
3522e903 | 1756 | * true is returned if we find something, false if nothing was in the tree |
d352ac68 | 1757 | */ |
083e75e7 JB |
1758 | bool btrfs_find_delalloc_range(struct extent_io_tree *tree, u64 *start, |
1759 | u64 *end, u64 max_bytes, | |
1760 | struct extent_state **cached_state) | |
d1310b2e CM |
1761 | { |
1762 | struct rb_node *node; | |
1763 | struct extent_state *state; | |
1764 | u64 cur_start = *start; | |
3522e903 | 1765 | bool found = false; |
d1310b2e CM |
1766 | u64 total_bytes = 0; |
1767 | ||
cad321ad | 1768 | spin_lock(&tree->lock); |
c8b97818 | 1769 | |
d1310b2e CM |
1770 | /* |
1771 | * this search will find all the extents that end after | |
1772 | * our range starts. | |
1773 | */ | |
80ea96b1 | 1774 | node = tree_search(tree, cur_start); |
2b114d1d | 1775 | if (!node) { |
3522e903 | 1776 | *end = (u64)-1; |
d1310b2e CM |
1777 | goto out; |
1778 | } | |
1779 | ||
d397712b | 1780 | while (1) { |
d1310b2e | 1781 | state = rb_entry(node, struct extent_state, rb_node); |
5b21f2ed ZY |
1782 | if (found && (state->start != cur_start || |
1783 | (state->state & EXTENT_BOUNDARY))) { | |
d1310b2e CM |
1784 | goto out; |
1785 | } | |
1786 | if (!(state->state & EXTENT_DELALLOC)) { | |
1787 | if (!found) | |
1788 | *end = state->end; | |
1789 | goto out; | |
1790 | } | |
c2a128d2 | 1791 | if (!found) { |
d1310b2e | 1792 | *start = state->start; |
c2a128d2 | 1793 | *cached_state = state; |
b7ac31b7 | 1794 | refcount_inc(&state->refs); |
c2a128d2 | 1795 | } |
3522e903 | 1796 | found = true; |
d1310b2e CM |
1797 | *end = state->end; |
1798 | cur_start = state->end + 1; | |
1799 | node = rb_next(node); | |
d1310b2e | 1800 | total_bytes += state->end - state->start + 1; |
7bf811a5 | 1801 | if (total_bytes >= max_bytes) |
573aecaf | 1802 | break; |
573aecaf | 1803 | if (!node) |
d1310b2e CM |
1804 | break; |
1805 | } | |
1806 | out: | |
cad321ad | 1807 | spin_unlock(&tree->lock); |
d1310b2e CM |
1808 | return found; |
1809 | } | |
1810 | ||
ed8f13bf QW |
1811 | /* |
1812 | * Process one page for __process_pages_contig(). | |
1813 | * | |
1814 | * Return >0 if we hit @page == @locked_page. | |
1815 | * Return 0 if we updated the page status. | |
1816 | * Return -EGAIN if the we need to try again. | |
1817 | * (For PAGE_LOCK case but got dirty page or page not belong to mapping) | |
1818 | */ | |
e38992be QW |
1819 | static int process_one_page(struct btrfs_fs_info *fs_info, |
1820 | struct address_space *mapping, | |
ed8f13bf | 1821 | struct page *page, struct page *locked_page, |
e38992be | 1822 | unsigned long page_ops, u64 start, u64 end) |
ed8f13bf | 1823 | { |
e38992be QW |
1824 | u32 len; |
1825 | ||
1826 | ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX); | |
1827 | len = end + 1 - start; | |
1828 | ||
ed8f13bf | 1829 | if (page_ops & PAGE_SET_ORDERED) |
b945a463 | 1830 | btrfs_page_clamp_set_ordered(fs_info, page, start, len); |
ed8f13bf QW |
1831 | |
1832 | if (page == locked_page) | |
1833 | return 1; | |
1834 | ||
1835 | if (page_ops & PAGE_SET_ERROR) | |
e38992be | 1836 | btrfs_page_clamp_set_error(fs_info, page, start, len); |
ed8f13bf | 1837 | if (page_ops & PAGE_START_WRITEBACK) { |
e38992be QW |
1838 | btrfs_page_clamp_clear_dirty(fs_info, page, start, len); |
1839 | btrfs_page_clamp_set_writeback(fs_info, page, start, len); | |
ed8f13bf QW |
1840 | } |
1841 | if (page_ops & PAGE_END_WRITEBACK) | |
e38992be | 1842 | btrfs_page_clamp_clear_writeback(fs_info, page, start, len); |
ed8f13bf | 1843 | if (page_ops & PAGE_LOCK) { |
1e1de387 QW |
1844 | int ret; |
1845 | ||
1846 | ret = btrfs_page_start_writer_lock(fs_info, page, start, len); | |
1847 | if (ret) | |
1848 | return ret; | |
ed8f13bf | 1849 | if (!PageDirty(page) || page->mapping != mapping) { |
1e1de387 | 1850 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1851 | return -EAGAIN; |
1852 | } | |
1853 | } | |
1854 | if (page_ops & PAGE_UNLOCK) | |
1e1de387 | 1855 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
1856 | return 0; |
1857 | } | |
1858 | ||
da2c7009 LB |
1859 | static int __process_pages_contig(struct address_space *mapping, |
1860 | struct page *locked_page, | |
98af9ab1 | 1861 | u64 start, u64 end, unsigned long page_ops, |
ed8f13bf QW |
1862 | u64 *processed_end) |
1863 | { | |
e38992be | 1864 | struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); |
ed8f13bf QW |
1865 | pgoff_t start_index = start >> PAGE_SHIFT; |
1866 | pgoff_t end_index = end >> PAGE_SHIFT; | |
1867 | pgoff_t index = start_index; | |
1868 | unsigned long nr_pages = end_index - start_index + 1; | |
1869 | unsigned long pages_processed = 0; | |
1870 | struct page *pages[16]; | |
1871 | int err = 0; | |
1872 | int i; | |
1873 | ||
1874 | if (page_ops & PAGE_LOCK) { | |
1875 | ASSERT(page_ops == PAGE_LOCK); | |
1876 | ASSERT(processed_end && *processed_end == start); | |
1877 | } | |
1878 | ||
1879 | if ((page_ops & PAGE_SET_ERROR) && nr_pages > 0) | |
1880 | mapping_set_error(mapping, -EIO); | |
1881 | ||
1882 | while (nr_pages > 0) { | |
1883 | int found_pages; | |
1884 | ||
1885 | found_pages = find_get_pages_contig(mapping, index, | |
1886 | min_t(unsigned long, | |
1887 | nr_pages, ARRAY_SIZE(pages)), pages); | |
1888 | if (found_pages == 0) { | |
1889 | /* | |
1890 | * Only if we're going to lock these pages, we can find | |
1891 | * nothing at @index. | |
1892 | */ | |
1893 | ASSERT(page_ops & PAGE_LOCK); | |
1894 | err = -EAGAIN; | |
1895 | goto out; | |
1896 | } | |
1897 | ||
1898 | for (i = 0; i < found_pages; i++) { | |
1899 | int process_ret; | |
1900 | ||
e38992be QW |
1901 | process_ret = process_one_page(fs_info, mapping, |
1902 | pages[i], locked_page, page_ops, | |
1903 | start, end); | |
ed8f13bf QW |
1904 | if (process_ret < 0) { |
1905 | for (; i < found_pages; i++) | |
1906 | put_page(pages[i]); | |
1907 | err = -EAGAIN; | |
1908 | goto out; | |
1909 | } | |
1910 | put_page(pages[i]); | |
1911 | pages_processed++; | |
1912 | } | |
1913 | nr_pages -= found_pages; | |
1914 | index += found_pages; | |
1915 | cond_resched(); | |
1916 | } | |
1917 | out: | |
1918 | if (err && processed_end) { | |
1919 | /* | |
1920 | * Update @processed_end. I know this is awful since it has | |
1921 | * two different return value patterns (inclusive vs exclusive). | |
1922 | * | |
1923 | * But the exclusive pattern is necessary if @start is 0, or we | |
1924 | * underflow and check against processed_end won't work as | |
1925 | * expected. | |
1926 | */ | |
1927 | if (pages_processed) | |
1928 | *processed_end = min(end, | |
1929 | ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1); | |
1930 | else | |
1931 | *processed_end = start; | |
1932 | } | |
1933 | return err; | |
1934 | } | |
da2c7009 | 1935 | |
143bede5 JM |
1936 | static noinline void __unlock_for_delalloc(struct inode *inode, |
1937 | struct page *locked_page, | |
1938 | u64 start, u64 end) | |
c8b97818 | 1939 | { |
09cbfeaf KS |
1940 | unsigned long index = start >> PAGE_SHIFT; |
1941 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 1942 | |
76c0021d | 1943 | ASSERT(locked_page); |
c8b97818 | 1944 | if (index == locked_page->index && end_index == index) |
143bede5 | 1945 | return; |
c8b97818 | 1946 | |
98af9ab1 | 1947 | __process_pages_contig(inode->i_mapping, locked_page, start, end, |
76c0021d | 1948 | PAGE_UNLOCK, NULL); |
c8b97818 CM |
1949 | } |
1950 | ||
1951 | static noinline int lock_delalloc_pages(struct inode *inode, | |
1952 | struct page *locked_page, | |
1953 | u64 delalloc_start, | |
1954 | u64 delalloc_end) | |
1955 | { | |
09cbfeaf | 1956 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
09cbfeaf | 1957 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
98af9ab1 | 1958 | u64 processed_end = delalloc_start; |
c8b97818 | 1959 | int ret; |
c8b97818 | 1960 | |
76c0021d | 1961 | ASSERT(locked_page); |
c8b97818 CM |
1962 | if (index == locked_page->index && index == end_index) |
1963 | return 0; | |
1964 | ||
98af9ab1 QW |
1965 | ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start, |
1966 | delalloc_end, PAGE_LOCK, &processed_end); | |
1967 | if (ret == -EAGAIN && processed_end > delalloc_start) | |
76c0021d | 1968 | __unlock_for_delalloc(inode, locked_page, delalloc_start, |
98af9ab1 | 1969 | processed_end); |
c8b97818 CM |
1970 | return ret; |
1971 | } | |
1972 | ||
1973 | /* | |
3522e903 LF |
1974 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
1975 | * more than @max_bytes. @Start and @end are used to return the range, | |
c8b97818 | 1976 | * |
3522e903 LF |
1977 | * Return: true if we find something |
1978 | * false if nothing was in the tree | |
c8b97818 | 1979 | */ |
ce9f967f | 1980 | EXPORT_FOR_TESTS |
3522e903 | 1981 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 1982 | struct page *locked_page, u64 *start, |
917aacec | 1983 | u64 *end) |
c8b97818 | 1984 | { |
9978059b | 1985 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
917aacec | 1986 | u64 max_bytes = BTRFS_MAX_EXTENT_SIZE; |
c8b97818 CM |
1987 | u64 delalloc_start; |
1988 | u64 delalloc_end; | |
3522e903 | 1989 | bool found; |
9655d298 | 1990 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
1991 | int ret; |
1992 | int loops = 0; | |
1993 | ||
1994 | again: | |
1995 | /* step one, find a bunch of delalloc bytes starting at start */ | |
1996 | delalloc_start = *start; | |
1997 | delalloc_end = 0; | |
083e75e7 JB |
1998 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
1999 | max_bytes, &cached_state); | |
70b99e69 | 2000 | if (!found || delalloc_end <= *start) { |
c8b97818 CM |
2001 | *start = delalloc_start; |
2002 | *end = delalloc_end; | |
c2a128d2 | 2003 | free_extent_state(cached_state); |
3522e903 | 2004 | return false; |
c8b97818 CM |
2005 | } |
2006 | ||
70b99e69 CM |
2007 | /* |
2008 | * start comes from the offset of locked_page. We have to lock | |
2009 | * pages in order, so we can't process delalloc bytes before | |
2010 | * locked_page | |
2011 | */ | |
d397712b | 2012 | if (delalloc_start < *start) |
70b99e69 | 2013 | delalloc_start = *start; |
70b99e69 | 2014 | |
c8b97818 CM |
2015 | /* |
2016 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 2017 | */ |
7bf811a5 JB |
2018 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
2019 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 2020 | |
c8b97818 CM |
2021 | /* step two, lock all the pages after the page that has start */ |
2022 | ret = lock_delalloc_pages(inode, locked_page, | |
2023 | delalloc_start, delalloc_end); | |
9bfd61d9 | 2024 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
2025 | if (ret == -EAGAIN) { |
2026 | /* some of the pages are gone, lets avoid looping by | |
2027 | * shortening the size of the delalloc range we're searching | |
2028 | */ | |
9655d298 | 2029 | free_extent_state(cached_state); |
7d788742 | 2030 | cached_state = NULL; |
c8b97818 | 2031 | if (!loops) { |
09cbfeaf | 2032 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
2033 | loops = 1; |
2034 | goto again; | |
2035 | } else { | |
3522e903 | 2036 | found = false; |
c8b97818 CM |
2037 | goto out_failed; |
2038 | } | |
2039 | } | |
c8b97818 CM |
2040 | |
2041 | /* step three, lock the state bits for the whole range */ | |
ff13db41 | 2042 | lock_extent_bits(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
2043 | |
2044 | /* then test to make sure it is all still delalloc */ | |
2045 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 2046 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 2047 | if (!ret) { |
9655d298 | 2048 | unlock_extent_cached(tree, delalloc_start, delalloc_end, |
e43bbe5e | 2049 | &cached_state); |
c8b97818 CM |
2050 | __unlock_for_delalloc(inode, locked_page, |
2051 | delalloc_start, delalloc_end); | |
2052 | cond_resched(); | |
2053 | goto again; | |
2054 | } | |
9655d298 | 2055 | free_extent_state(cached_state); |
c8b97818 CM |
2056 | *start = delalloc_start; |
2057 | *end = delalloc_end; | |
2058 | out_failed: | |
2059 | return found; | |
2060 | } | |
2061 | ||
ad7ff17b | 2062 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 2063 | struct page *locked_page, |
f97e27e9 | 2064 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 2065 | { |
ad7ff17b | 2066 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, 1, 0, NULL); |
873695b3 | 2067 | |
ad7ff17b | 2068 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
98af9ab1 | 2069 | start, end, page_ops, NULL); |
873695b3 LB |
2070 | } |
2071 | ||
d352ac68 CM |
2072 | /* |
2073 | * count the number of bytes in the tree that have a given bit(s) | |
2074 | * set. This can be fairly slow, except for EXTENT_DIRTY which is | |
2075 | * cached. The total number found is returned. | |
2076 | */ | |
d1310b2e CM |
2077 | u64 count_range_bits(struct extent_io_tree *tree, |
2078 | u64 *start, u64 search_end, u64 max_bytes, | |
f97e27e9 | 2079 | u32 bits, int contig) |
d1310b2e CM |
2080 | { |
2081 | struct rb_node *node; | |
2082 | struct extent_state *state; | |
2083 | u64 cur_start = *start; | |
2084 | u64 total_bytes = 0; | |
ec29ed5b | 2085 | u64 last = 0; |
d1310b2e CM |
2086 | int found = 0; |
2087 | ||
fae7f21c | 2088 | if (WARN_ON(search_end <= cur_start)) |
d1310b2e | 2089 | return 0; |
d1310b2e | 2090 | |
cad321ad | 2091 | spin_lock(&tree->lock); |
d1310b2e CM |
2092 | if (cur_start == 0 && bits == EXTENT_DIRTY) { |
2093 | total_bytes = tree->dirty_bytes; | |
2094 | goto out; | |
2095 | } | |
2096 | /* | |
2097 | * this search will find all the extents that end after | |
2098 | * our range starts. | |
2099 | */ | |
80ea96b1 | 2100 | node = tree_search(tree, cur_start); |
d397712b | 2101 | if (!node) |
d1310b2e | 2102 | goto out; |
d1310b2e | 2103 | |
d397712b | 2104 | while (1) { |
d1310b2e CM |
2105 | state = rb_entry(node, struct extent_state, rb_node); |
2106 | if (state->start > search_end) | |
2107 | break; | |
ec29ed5b CM |
2108 | if (contig && found && state->start > last + 1) |
2109 | break; | |
2110 | if (state->end >= cur_start && (state->state & bits) == bits) { | |
d1310b2e CM |
2111 | total_bytes += min(search_end, state->end) + 1 - |
2112 | max(cur_start, state->start); | |
2113 | if (total_bytes >= max_bytes) | |
2114 | break; | |
2115 | if (!found) { | |
af60bed2 | 2116 | *start = max(cur_start, state->start); |
d1310b2e CM |
2117 | found = 1; |
2118 | } | |
ec29ed5b CM |
2119 | last = state->end; |
2120 | } else if (contig && found) { | |
2121 | break; | |
d1310b2e CM |
2122 | } |
2123 | node = rb_next(node); | |
2124 | if (!node) | |
2125 | break; | |
2126 | } | |
2127 | out: | |
cad321ad | 2128 | spin_unlock(&tree->lock); |
d1310b2e CM |
2129 | return total_bytes; |
2130 | } | |
b2950863 | 2131 | |
d352ac68 CM |
2132 | /* |
2133 | * set the private field for a given byte offset in the tree. If there isn't | |
2134 | * an extent_state there already, this does nothing. | |
2135 | */ | |
b3f167aa JB |
2136 | int set_state_failrec(struct extent_io_tree *tree, u64 start, |
2137 | struct io_failure_record *failrec) | |
d1310b2e CM |
2138 | { |
2139 | struct rb_node *node; | |
2140 | struct extent_state *state; | |
2141 | int ret = 0; | |
2142 | ||
cad321ad | 2143 | spin_lock(&tree->lock); |
d1310b2e CM |
2144 | /* |
2145 | * this search will find all the extents that end after | |
2146 | * our range starts. | |
2147 | */ | |
80ea96b1 | 2148 | node = tree_search(tree, start); |
2b114d1d | 2149 | if (!node) { |
d1310b2e CM |
2150 | ret = -ENOENT; |
2151 | goto out; | |
2152 | } | |
2153 | state = rb_entry(node, struct extent_state, rb_node); | |
2154 | if (state->start != start) { | |
2155 | ret = -ENOENT; | |
2156 | goto out; | |
2157 | } | |
47dc196a | 2158 | state->failrec = failrec; |
d1310b2e | 2159 | out: |
cad321ad | 2160 | spin_unlock(&tree->lock); |
d1310b2e CM |
2161 | return ret; |
2162 | } | |
2163 | ||
2279a270 | 2164 | struct io_failure_record *get_state_failrec(struct extent_io_tree *tree, u64 start) |
d1310b2e CM |
2165 | { |
2166 | struct rb_node *node; | |
2167 | struct extent_state *state; | |
2279a270 | 2168 | struct io_failure_record *failrec; |
d1310b2e | 2169 | |
cad321ad | 2170 | spin_lock(&tree->lock); |
d1310b2e CM |
2171 | /* |
2172 | * this search will find all the extents that end after | |
2173 | * our range starts. | |
2174 | */ | |
80ea96b1 | 2175 | node = tree_search(tree, start); |
2b114d1d | 2176 | if (!node) { |
2279a270 | 2177 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2178 | goto out; |
2179 | } | |
2180 | state = rb_entry(node, struct extent_state, rb_node); | |
2181 | if (state->start != start) { | |
2279a270 | 2182 | failrec = ERR_PTR(-ENOENT); |
d1310b2e CM |
2183 | goto out; |
2184 | } | |
2279a270 NB |
2185 | |
2186 | failrec = state->failrec; | |
d1310b2e | 2187 | out: |
cad321ad | 2188 | spin_unlock(&tree->lock); |
2279a270 | 2189 | return failrec; |
d1310b2e CM |
2190 | } |
2191 | ||
2192 | /* | |
2193 | * searches a range in the state tree for a given mask. | |
70dec807 | 2194 | * If 'filled' == 1, this returns 1 only if every extent in the tree |
d1310b2e CM |
2195 | * has the bits set. Otherwise, 1 is returned if any bit in the |
2196 | * range is found set. | |
2197 | */ | |
2198 | int test_range_bit(struct extent_io_tree *tree, u64 start, u64 end, | |
f97e27e9 | 2199 | u32 bits, int filled, struct extent_state *cached) |
d1310b2e CM |
2200 | { |
2201 | struct extent_state *state = NULL; | |
2202 | struct rb_node *node; | |
2203 | int bitset = 0; | |
d1310b2e | 2204 | |
cad321ad | 2205 | spin_lock(&tree->lock); |
27a3507d | 2206 | if (cached && extent_state_in_tree(cached) && cached->start <= start && |
df98b6e2 | 2207 | cached->end > start) |
9655d298 CM |
2208 | node = &cached->rb_node; |
2209 | else | |
2210 | node = tree_search(tree, start); | |
d1310b2e CM |
2211 | while (node && start <= end) { |
2212 | state = rb_entry(node, struct extent_state, rb_node); | |
2213 | ||
2214 | if (filled && state->start > start) { | |
2215 | bitset = 0; | |
2216 | break; | |
2217 | } | |
2218 | ||
2219 | if (state->start > end) | |
2220 | break; | |
2221 | ||
2222 | if (state->state & bits) { | |
2223 | bitset = 1; | |
2224 | if (!filled) | |
2225 | break; | |
2226 | } else if (filled) { | |
2227 | bitset = 0; | |
2228 | break; | |
2229 | } | |
46562cec CM |
2230 | |
2231 | if (state->end == (u64)-1) | |
2232 | break; | |
2233 | ||
d1310b2e CM |
2234 | start = state->end + 1; |
2235 | if (start > end) | |
2236 | break; | |
2237 | node = rb_next(node); | |
2238 | if (!node) { | |
2239 | if (filled) | |
2240 | bitset = 0; | |
2241 | break; | |
2242 | } | |
2243 | } | |
cad321ad | 2244 | spin_unlock(&tree->lock); |
d1310b2e CM |
2245 | return bitset; |
2246 | } | |
d1310b2e CM |
2247 | |
2248 | /* | |
2249 | * helper function to set a given page up to date if all the | |
2250 | * extents in the tree for that page are up to date | |
2251 | */ | |
143bede5 | 2252 | static void check_page_uptodate(struct extent_io_tree *tree, struct page *page) |
d1310b2e | 2253 | { |
4eee4fa4 | 2254 | u64 start = page_offset(page); |
09cbfeaf | 2255 | u64 end = start + PAGE_SIZE - 1; |
9655d298 | 2256 | if (test_range_bit(tree, start, end, EXTENT_UPTODATE, 1, NULL)) |
d1310b2e | 2257 | SetPageUptodate(page); |
d1310b2e CM |
2258 | } |
2259 | ||
7870d082 JB |
2260 | int free_io_failure(struct extent_io_tree *failure_tree, |
2261 | struct extent_io_tree *io_tree, | |
2262 | struct io_failure_record *rec) | |
4a54c8c1 JS |
2263 | { |
2264 | int ret; | |
2265 | int err = 0; | |
4a54c8c1 | 2266 | |
47dc196a | 2267 | set_state_failrec(failure_tree, rec->start, NULL); |
4a54c8c1 JS |
2268 | ret = clear_extent_bits(failure_tree, rec->start, |
2269 | rec->start + rec->len - 1, | |
91166212 | 2270 | EXTENT_LOCKED | EXTENT_DIRTY); |
4a54c8c1 JS |
2271 | if (ret) |
2272 | err = ret; | |
2273 | ||
7870d082 | 2274 | ret = clear_extent_bits(io_tree, rec->start, |
53b381b3 | 2275 | rec->start + rec->len - 1, |
91166212 | 2276 | EXTENT_DAMAGED); |
53b381b3 DW |
2277 | if (ret && !err) |
2278 | err = ret; | |
4a54c8c1 JS |
2279 | |
2280 | kfree(rec); | |
2281 | return err; | |
2282 | } | |
2283 | ||
4a54c8c1 JS |
2284 | /* |
2285 | * this bypasses the standard btrfs submit functions deliberately, as | |
2286 | * the standard behavior is to write all copies in a raid setup. here we only | |
2287 | * want to write the one bad copy. so we do the mapping for ourselves and issue | |
2288 | * submit_bio directly. | |
3ec706c8 | 2289 | * to avoid any synchronization issues, wait for the data after writing, which |
4a54c8c1 JS |
2290 | * actually prevents the read that triggered the error from finishing. |
2291 | * currently, there can be no more than two copies of every data bit. thus, | |
2292 | * exactly one rewrite is required. | |
2293 | */ | |
6ec656bc JB |
2294 | int repair_io_failure(struct btrfs_fs_info *fs_info, u64 ino, u64 start, |
2295 | u64 length, u64 logical, struct page *page, | |
2296 | unsigned int pg_offset, int mirror_num) | |
4a54c8c1 JS |
2297 | { |
2298 | struct bio *bio; | |
2299 | struct btrfs_device *dev; | |
4a54c8c1 JS |
2300 | u64 map_length = 0; |
2301 | u64 sector; | |
2302 | struct btrfs_bio *bbio = NULL; | |
2303 | int ret; | |
2304 | ||
1751e8a6 | 2305 | ASSERT(!(fs_info->sb->s_flags & SB_RDONLY)); |
4a54c8c1 JS |
2306 | BUG_ON(!mirror_num); |
2307 | ||
f7ef5287 NA |
2308 | if (btrfs_is_zoned(fs_info)) |
2309 | return btrfs_repair_one_zone(fs_info, logical); | |
2310 | ||
c5e4c3d7 | 2311 | bio = btrfs_io_bio_alloc(1); |
4f024f37 | 2312 | bio->bi_iter.bi_size = 0; |
4a54c8c1 JS |
2313 | map_length = length; |
2314 | ||
b5de8d0d FM |
2315 | /* |
2316 | * Avoid races with device replace and make sure our bbio has devices | |
2317 | * associated to its stripes that don't go away while we are doing the | |
2318 | * read repair operation. | |
2319 | */ | |
2320 | btrfs_bio_counter_inc_blocked(fs_info); | |
e4ff5fb5 | 2321 | if (btrfs_is_parity_mirror(fs_info, logical, length)) { |
c725328c LB |
2322 | /* |
2323 | * Note that we don't use BTRFS_MAP_WRITE because it's supposed | |
2324 | * to update all raid stripes, but here we just want to correct | |
2325 | * bad stripe, thus BTRFS_MAP_READ is abused to only get the bad | |
2326 | * stripe's dev and sector. | |
2327 | */ | |
2328 | ret = btrfs_map_block(fs_info, BTRFS_MAP_READ, logical, | |
2329 | &map_length, &bbio, 0); | |
2330 | if (ret) { | |
2331 | btrfs_bio_counter_dec(fs_info); | |
2332 | bio_put(bio); | |
2333 | return -EIO; | |
2334 | } | |
2335 | ASSERT(bbio->mirror_num == 1); | |
2336 | } else { | |
2337 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, | |
2338 | &map_length, &bbio, mirror_num); | |
2339 | if (ret) { | |
2340 | btrfs_bio_counter_dec(fs_info); | |
2341 | bio_put(bio); | |
2342 | return -EIO; | |
2343 | } | |
2344 | BUG_ON(mirror_num != bbio->mirror_num); | |
4a54c8c1 | 2345 | } |
c725328c LB |
2346 | |
2347 | sector = bbio->stripes[bbio->mirror_num - 1].physical >> 9; | |
4f024f37 | 2348 | bio->bi_iter.bi_sector = sector; |
c725328c | 2349 | dev = bbio->stripes[bbio->mirror_num - 1].dev; |
6e9606d2 | 2350 | btrfs_put_bbio(bbio); |
ebbede42 AJ |
2351 | if (!dev || !dev->bdev || |
2352 | !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state)) { | |
b5de8d0d | 2353 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2354 | bio_put(bio); |
2355 | return -EIO; | |
2356 | } | |
74d46992 | 2357 | bio_set_dev(bio, dev->bdev); |
70fd7614 | 2358 | bio->bi_opf = REQ_OP_WRITE | REQ_SYNC; |
ffdd2018 | 2359 | bio_add_page(bio, page, length, pg_offset); |
4a54c8c1 | 2360 | |
4e49ea4a | 2361 | if (btrfsic_submit_bio_wait(bio)) { |
4a54c8c1 | 2362 | /* try to remap that extent elsewhere? */ |
b5de8d0d | 2363 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 | 2364 | bio_put(bio); |
442a4f63 | 2365 | btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_WRITE_ERRS); |
4a54c8c1 JS |
2366 | return -EIO; |
2367 | } | |
2368 | ||
b14af3b4 DS |
2369 | btrfs_info_rl_in_rcu(fs_info, |
2370 | "read error corrected: ino %llu off %llu (dev %s sector %llu)", | |
6ec656bc | 2371 | ino, start, |
1203b681 | 2372 | rcu_str_deref(dev->name), sector); |
b5de8d0d | 2373 | btrfs_bio_counter_dec(fs_info); |
4a54c8c1 JS |
2374 | bio_put(bio); |
2375 | return 0; | |
2376 | } | |
2377 | ||
2b48966a | 2378 | int btrfs_repair_eb_io_failure(const struct extent_buffer *eb, int mirror_num) |
ea466794 | 2379 | { |
20a1fbf9 | 2380 | struct btrfs_fs_info *fs_info = eb->fs_info; |
ea466794 | 2381 | u64 start = eb->start; |
cc5e31a4 | 2382 | int i, num_pages = num_extent_pages(eb); |
d95603b2 | 2383 | int ret = 0; |
ea466794 | 2384 | |
bc98a42c | 2385 | if (sb_rdonly(fs_info->sb)) |
908960c6 ID |
2386 | return -EROFS; |
2387 | ||
ea466794 | 2388 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2389 | struct page *p = eb->pages[i]; |
1203b681 | 2390 | |
6ec656bc | 2391 | ret = repair_io_failure(fs_info, 0, start, PAGE_SIZE, start, p, |
1203b681 | 2392 | start - page_offset(p), mirror_num); |
ea466794 JB |
2393 | if (ret) |
2394 | break; | |
09cbfeaf | 2395 | start += PAGE_SIZE; |
ea466794 JB |
2396 | } |
2397 | ||
2398 | return ret; | |
2399 | } | |
2400 | ||
4a54c8c1 JS |
2401 | /* |
2402 | * each time an IO finishes, we do a fast check in the IO failure tree | |
2403 | * to see if we need to process or clean up an io_failure_record | |
2404 | */ | |
7870d082 JB |
2405 | int clean_io_failure(struct btrfs_fs_info *fs_info, |
2406 | struct extent_io_tree *failure_tree, | |
2407 | struct extent_io_tree *io_tree, u64 start, | |
2408 | struct page *page, u64 ino, unsigned int pg_offset) | |
4a54c8c1 JS |
2409 | { |
2410 | u64 private; | |
4a54c8c1 | 2411 | struct io_failure_record *failrec; |
4a54c8c1 JS |
2412 | struct extent_state *state; |
2413 | int num_copies; | |
4a54c8c1 | 2414 | int ret; |
4a54c8c1 JS |
2415 | |
2416 | private = 0; | |
7870d082 JB |
2417 | ret = count_range_bits(failure_tree, &private, (u64)-1, 1, |
2418 | EXTENT_DIRTY, 0); | |
4a54c8c1 JS |
2419 | if (!ret) |
2420 | return 0; | |
2421 | ||
2279a270 NB |
2422 | failrec = get_state_failrec(failure_tree, start); |
2423 | if (IS_ERR(failrec)) | |
4a54c8c1 JS |
2424 | return 0; |
2425 | ||
4a54c8c1 JS |
2426 | BUG_ON(!failrec->this_mirror); |
2427 | ||
bc98a42c | 2428 | if (sb_rdonly(fs_info->sb)) |
908960c6 | 2429 | goto out; |
4a54c8c1 | 2430 | |
7870d082 JB |
2431 | spin_lock(&io_tree->lock); |
2432 | state = find_first_extent_bit_state(io_tree, | |
4a54c8c1 JS |
2433 | failrec->start, |
2434 | EXTENT_LOCKED); | |
7870d082 | 2435 | spin_unlock(&io_tree->lock); |
4a54c8c1 | 2436 | |
883d0de4 MX |
2437 | if (state && state->start <= failrec->start && |
2438 | state->end >= failrec->start + failrec->len - 1) { | |
3ec706c8 SB |
2439 | num_copies = btrfs_num_copies(fs_info, failrec->logical, |
2440 | failrec->len); | |
4a54c8c1 | 2441 | if (num_copies > 1) { |
7870d082 JB |
2442 | repair_io_failure(fs_info, ino, start, failrec->len, |
2443 | failrec->logical, page, pg_offset, | |
2444 | failrec->failed_mirror); | |
4a54c8c1 JS |
2445 | } |
2446 | } | |
2447 | ||
2448 | out: | |
7870d082 | 2449 | free_io_failure(failure_tree, io_tree, failrec); |
4a54c8c1 | 2450 | |
454ff3de | 2451 | return 0; |
4a54c8c1 JS |
2452 | } |
2453 | ||
f612496b MX |
2454 | /* |
2455 | * Can be called when | |
2456 | * - hold extent lock | |
2457 | * - under ordered extent | |
2458 | * - the inode is freeing | |
2459 | */ | |
7ab7956e | 2460 | void btrfs_free_io_failure_record(struct btrfs_inode *inode, u64 start, u64 end) |
f612496b | 2461 | { |
7ab7956e | 2462 | struct extent_io_tree *failure_tree = &inode->io_failure_tree; |
f612496b MX |
2463 | struct io_failure_record *failrec; |
2464 | struct extent_state *state, *next; | |
2465 | ||
2466 | if (RB_EMPTY_ROOT(&failure_tree->state)) | |
2467 | return; | |
2468 | ||
2469 | spin_lock(&failure_tree->lock); | |
2470 | state = find_first_extent_bit_state(failure_tree, start, EXTENT_DIRTY); | |
2471 | while (state) { | |
2472 | if (state->start > end) | |
2473 | break; | |
2474 | ||
2475 | ASSERT(state->end <= end); | |
2476 | ||
2477 | next = next_state(state); | |
2478 | ||
47dc196a | 2479 | failrec = state->failrec; |
f612496b MX |
2480 | free_extent_state(state); |
2481 | kfree(failrec); | |
2482 | ||
2483 | state = next; | |
2484 | } | |
2485 | spin_unlock(&failure_tree->lock); | |
2486 | } | |
2487 | ||
3526302f | 2488 | static struct io_failure_record *btrfs_get_io_failure_record(struct inode *inode, |
150e4b05 | 2489 | u64 start) |
4a54c8c1 | 2490 | { |
ab8d0fc4 | 2491 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2492 | struct io_failure_record *failrec; |
4a54c8c1 | 2493 | struct extent_map *em; |
4a54c8c1 JS |
2494 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
2495 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; | |
2496 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; | |
150e4b05 | 2497 | const u32 sectorsize = fs_info->sectorsize; |
4a54c8c1 | 2498 | int ret; |
4a54c8c1 JS |
2499 | u64 logical; |
2500 | ||
2279a270 | 2501 | failrec = get_state_failrec(failure_tree, start); |
3526302f | 2502 | if (!IS_ERR(failrec)) { |
ab8d0fc4 | 2503 | btrfs_debug(fs_info, |
1245835d QW |
2504 | "Get IO Failure Record: (found) logical=%llu, start=%llu, len=%llu", |
2505 | failrec->logical, failrec->start, failrec->len); | |
4a54c8c1 JS |
2506 | /* |
2507 | * when data can be on disk more than twice, add to failrec here | |
2508 | * (e.g. with a list for failed_mirror) to make | |
2509 | * clean_io_failure() clean all those errors at once. | |
2510 | */ | |
3526302f NB |
2511 | |
2512 | return failrec; | |
4a54c8c1 | 2513 | } |
2fe6303e | 2514 | |
3526302f NB |
2515 | failrec = kzalloc(sizeof(*failrec), GFP_NOFS); |
2516 | if (!failrec) | |
2517 | return ERR_PTR(-ENOMEM); | |
2fe6303e | 2518 | |
3526302f | 2519 | failrec->start = start; |
150e4b05 | 2520 | failrec->len = sectorsize; |
3526302f NB |
2521 | failrec->this_mirror = 0; |
2522 | failrec->bio_flags = 0; | |
3526302f NB |
2523 | |
2524 | read_lock(&em_tree->lock); | |
2525 | em = lookup_extent_mapping(em_tree, start, failrec->len); | |
2526 | if (!em) { | |
2527 | read_unlock(&em_tree->lock); | |
2528 | kfree(failrec); | |
2529 | return ERR_PTR(-EIO); | |
2530 | } | |
2531 | ||
2532 | if (em->start > start || em->start + em->len <= start) { | |
2533 | free_extent_map(em); | |
2534 | em = NULL; | |
2535 | } | |
2536 | read_unlock(&em_tree->lock); | |
2537 | if (!em) { | |
2538 | kfree(failrec); | |
2539 | return ERR_PTR(-EIO); | |
2540 | } | |
2541 | ||
2542 | logical = start - em->start; | |
2543 | logical = em->block_start + logical; | |
2544 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { | |
2545 | logical = em->block_start; | |
2546 | failrec->bio_flags = EXTENT_BIO_COMPRESSED; | |
2547 | extent_set_compress_type(&failrec->bio_flags, em->compress_type); | |
2548 | } | |
2549 | ||
2550 | btrfs_debug(fs_info, | |
2551 | "Get IO Failure Record: (new) logical=%llu, start=%llu, len=%llu", | |
2552 | logical, start, failrec->len); | |
2553 | ||
2554 | failrec->logical = logical; | |
2555 | free_extent_map(em); | |
2556 | ||
2557 | /* Set the bits in the private failure tree */ | |
150e4b05 | 2558 | ret = set_extent_bits(failure_tree, start, start + sectorsize - 1, |
3526302f NB |
2559 | EXTENT_LOCKED | EXTENT_DIRTY); |
2560 | if (ret >= 0) { | |
2561 | ret = set_state_failrec(failure_tree, start, failrec); | |
2562 | /* Set the bits in the inode's tree */ | |
150e4b05 QW |
2563 | ret = set_extent_bits(tree, start, start + sectorsize - 1, |
2564 | EXTENT_DAMAGED); | |
3526302f NB |
2565 | } else if (ret < 0) { |
2566 | kfree(failrec); | |
2567 | return ERR_PTR(ret); | |
2568 | } | |
2569 | ||
2570 | return failrec; | |
2fe6303e MX |
2571 | } |
2572 | ||
1245835d | 2573 | static bool btrfs_check_repairable(struct inode *inode, |
ce06d3ec OS |
2574 | struct io_failure_record *failrec, |
2575 | int failed_mirror) | |
2fe6303e | 2576 | { |
ab8d0fc4 | 2577 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e MX |
2578 | int num_copies; |
2579 | ||
ab8d0fc4 | 2580 | num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len); |
4a54c8c1 JS |
2581 | if (num_copies == 1) { |
2582 | /* | |
2583 | * we only have a single copy of the data, so don't bother with | |
2584 | * all the retry and error correction code that follows. no | |
2585 | * matter what the error is, it is very likely to persist. | |
2586 | */ | |
ab8d0fc4 JM |
2587 | btrfs_debug(fs_info, |
2588 | "Check Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d", | |
2589 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2590 | return false; |
4a54c8c1 JS |
2591 | } |
2592 | ||
1245835d QW |
2593 | /* The failure record should only contain one sector */ |
2594 | ASSERT(failrec->len == fs_info->sectorsize); | |
2595 | ||
4a54c8c1 | 2596 | /* |
1245835d QW |
2597 | * There are two premises: |
2598 | * a) deliver good data to the caller | |
2599 | * b) correct the bad sectors on disk | |
2600 | * | |
2601 | * Since we're only doing repair for one sector, we only need to get | |
2602 | * a good copy of the failed sector and if we succeed, we have setup | |
2603 | * everything for repair_io_failure to do the rest for us. | |
4a54c8c1 | 2604 | */ |
1245835d QW |
2605 | failrec->failed_mirror = failed_mirror; |
2606 | failrec->this_mirror++; | |
2607 | if (failrec->this_mirror == failed_mirror) | |
4a54c8c1 | 2608 | failrec->this_mirror++; |
4a54c8c1 | 2609 | |
facc8a22 | 2610 | if (failrec->this_mirror > num_copies) { |
ab8d0fc4 JM |
2611 | btrfs_debug(fs_info, |
2612 | "Check Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d", | |
2613 | num_copies, failrec->this_mirror, failed_mirror); | |
c3cfb656 | 2614 | return false; |
4a54c8c1 JS |
2615 | } |
2616 | ||
c3cfb656 | 2617 | return true; |
2fe6303e MX |
2618 | } |
2619 | ||
150e4b05 QW |
2620 | int btrfs_repair_one_sector(struct inode *inode, |
2621 | struct bio *failed_bio, u32 bio_offset, | |
2622 | struct page *page, unsigned int pgoff, | |
2623 | u64 start, int failed_mirror, | |
2624 | submit_bio_hook_t *submit_bio_hook) | |
2fe6303e MX |
2625 | { |
2626 | struct io_failure_record *failrec; | |
77d5d689 | 2627 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
2fe6303e | 2628 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 2629 | struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree; |
77d5d689 | 2630 | struct btrfs_io_bio *failed_io_bio = btrfs_io_bio(failed_bio); |
7ffd27e3 | 2631 | const int icsum = bio_offset >> fs_info->sectorsize_bits; |
77d5d689 OS |
2632 | struct bio *repair_bio; |
2633 | struct btrfs_io_bio *repair_io_bio; | |
4e4cbee9 | 2634 | blk_status_t status; |
2fe6303e | 2635 | |
77d5d689 OS |
2636 | btrfs_debug(fs_info, |
2637 | "repair read error: read error at %llu", start); | |
2fe6303e | 2638 | |
1f7ad75b | 2639 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); |
2fe6303e | 2640 | |
150e4b05 | 2641 | failrec = btrfs_get_io_failure_record(inode, start); |
3526302f | 2642 | if (IS_ERR(failrec)) |
150e4b05 | 2643 | return PTR_ERR(failrec); |
2fe6303e | 2644 | |
1245835d QW |
2645 | |
2646 | if (!btrfs_check_repairable(inode, failrec, failed_mirror)) { | |
7870d082 | 2647 | free_io_failure(failure_tree, tree, failrec); |
150e4b05 | 2648 | return -EIO; |
2fe6303e MX |
2649 | } |
2650 | ||
77d5d689 OS |
2651 | repair_bio = btrfs_io_bio_alloc(1); |
2652 | repair_io_bio = btrfs_io_bio(repair_bio); | |
2653 | repair_bio->bi_opf = REQ_OP_READ; | |
77d5d689 OS |
2654 | repair_bio->bi_end_io = failed_bio->bi_end_io; |
2655 | repair_bio->bi_iter.bi_sector = failrec->logical >> 9; | |
2656 | repair_bio->bi_private = failed_bio->bi_private; | |
2fe6303e | 2657 | |
77d5d689 | 2658 | if (failed_io_bio->csum) { |
223486c2 | 2659 | const u32 csum_size = fs_info->csum_size; |
77d5d689 OS |
2660 | |
2661 | repair_io_bio->csum = repair_io_bio->csum_inline; | |
2662 | memcpy(repair_io_bio->csum, | |
2663 | failed_io_bio->csum + csum_size * icsum, csum_size); | |
2664 | } | |
2fe6303e | 2665 | |
77d5d689 OS |
2666 | bio_add_page(repair_bio, page, failrec->len, pgoff); |
2667 | repair_io_bio->logical = failrec->start; | |
2668 | repair_io_bio->iter = repair_bio->bi_iter; | |
4a54c8c1 | 2669 | |
ab8d0fc4 | 2670 | btrfs_debug(btrfs_sb(inode->i_sb), |
1245835d QW |
2671 | "repair read error: submitting new read to mirror %d", |
2672 | failrec->this_mirror); | |
4a54c8c1 | 2673 | |
77d5d689 OS |
2674 | status = submit_bio_hook(inode, repair_bio, failrec->this_mirror, |
2675 | failrec->bio_flags); | |
4e4cbee9 | 2676 | if (status) { |
7870d082 | 2677 | free_io_failure(failure_tree, tree, failrec); |
77d5d689 | 2678 | bio_put(repair_bio); |
6c387ab2 | 2679 | } |
150e4b05 QW |
2680 | return blk_status_to_errno(status); |
2681 | } | |
2682 | ||
2683 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) | |
2684 | { | |
2685 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
2686 | ||
2687 | ASSERT(page_offset(page) <= start && | |
2688 | start + len <= page_offset(page) + PAGE_SIZE); | |
2689 | ||
2690 | /* | |
2691 | * For subapge metadata case, all btrfs_page_* helpers need page to | |
2692 | * have page::private populated. | |
2693 | * But we can have rare case where the last eb in the page is only | |
2694 | * referred by the IO, and it gets released immedately after it's | |
2695 | * read and verified. | |
2696 | * | |
2697 | * This can detach the page private completely. | |
2698 | * In that case, we can just skip the page status update completely, | |
2699 | * as the page has no eb anymore. | |
2700 | */ | |
2701 | if (fs_info->sectorsize < PAGE_SIZE && unlikely(!PagePrivate(page))) { | |
2702 | ASSERT(!is_data_inode(page->mapping->host)); | |
2703 | return; | |
2704 | } | |
2705 | if (uptodate) { | |
2706 | btrfs_page_set_uptodate(fs_info, page, start, len); | |
2707 | } else { | |
2708 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
2709 | btrfs_page_set_error(fs_info, page, start, len); | |
2710 | } | |
2711 | ||
2712 | if (fs_info->sectorsize == PAGE_SIZE) | |
2713 | unlock_page(page); | |
2714 | else if (is_data_inode(page->mapping->host)) | |
2715 | /* | |
2716 | * For subpage data, unlock the page if we're the last reader. | |
2717 | * For subpage metadata, page lock is not utilized for read. | |
2718 | */ | |
2719 | btrfs_subpage_end_reader(fs_info, page, start, len); | |
2720 | } | |
2721 | ||
2722 | static blk_status_t submit_read_repair(struct inode *inode, | |
2723 | struct bio *failed_bio, u32 bio_offset, | |
2724 | struct page *page, unsigned int pgoff, | |
2725 | u64 start, u64 end, int failed_mirror, | |
2726 | unsigned int error_bitmap, | |
2727 | submit_bio_hook_t *submit_bio_hook) | |
2728 | { | |
2729 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
2730 | const u32 sectorsize = fs_info->sectorsize; | |
2731 | const int nr_bits = (end + 1 - start) >> fs_info->sectorsize_bits; | |
2732 | int error = 0; | |
2733 | int i; | |
2734 | ||
2735 | BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE); | |
2736 | ||
2737 | /* We're here because we had some read errors or csum mismatch */ | |
2738 | ASSERT(error_bitmap); | |
2739 | ||
2740 | /* | |
2741 | * We only get called on buffered IO, thus page must be mapped and bio | |
2742 | * must not be cloned. | |
2743 | */ | |
2744 | ASSERT(page->mapping && !bio_flagged(failed_bio, BIO_CLONED)); | |
2745 | ||
2746 | /* Iterate through all the sectors in the range */ | |
2747 | for (i = 0; i < nr_bits; i++) { | |
2748 | const unsigned int offset = i * sectorsize; | |
2749 | struct extent_state *cached = NULL; | |
2750 | bool uptodate = false; | |
2751 | int ret; | |
2752 | ||
2753 | if (!(error_bitmap & (1U << i))) { | |
2754 | /* | |
2755 | * This sector has no error, just end the page read | |
2756 | * and unlock the range. | |
2757 | */ | |
2758 | uptodate = true; | |
2759 | goto next; | |
2760 | } | |
2761 | ||
2762 | ret = btrfs_repair_one_sector(inode, failed_bio, | |
2763 | bio_offset + offset, | |
2764 | page, pgoff + offset, start + offset, | |
2765 | failed_mirror, submit_bio_hook); | |
2766 | if (!ret) { | |
2767 | /* | |
2768 | * We have submitted the read repair, the page release | |
2769 | * will be handled by the endio function of the | |
2770 | * submitted repair bio. | |
2771 | * Thus we don't need to do any thing here. | |
2772 | */ | |
2773 | continue; | |
2774 | } | |
2775 | /* | |
2776 | * Repair failed, just record the error but still continue. | |
2777 | * Or the remaining sectors will not be properly unlocked. | |
2778 | */ | |
2779 | if (!error) | |
2780 | error = ret; | |
2781 | next: | |
2782 | end_page_read(page, uptodate, start + offset, sectorsize); | |
2783 | if (uptodate) | |
2784 | set_extent_uptodate(&BTRFS_I(inode)->io_tree, | |
2785 | start + offset, | |
2786 | start + offset + sectorsize - 1, | |
2787 | &cached, GFP_ATOMIC); | |
2788 | unlock_extent_cached_atomic(&BTRFS_I(inode)->io_tree, | |
2789 | start + offset, | |
2790 | start + offset + sectorsize - 1, | |
2791 | &cached); | |
2792 | } | |
2793 | return errno_to_blk_status(error); | |
4a54c8c1 JS |
2794 | } |
2795 | ||
d1310b2e CM |
2796 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
2797 | ||
b5227c07 | 2798 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 2799 | { |
38a39ac7 | 2800 | struct btrfs_inode *inode; |
87826df0 | 2801 | int uptodate = (err == 0); |
3e2426bd | 2802 | int ret = 0; |
87826df0 | 2803 | |
38a39ac7 QW |
2804 | ASSERT(page && page->mapping); |
2805 | inode = BTRFS_I(page->mapping->host); | |
2806 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 2807 | |
87826df0 | 2808 | if (!uptodate) { |
87826df0 JM |
2809 | ClearPageUptodate(page); |
2810 | SetPageError(page); | |
bff5baf8 | 2811 | ret = err < 0 ? err : -EIO; |
5dca6eea | 2812 | mapping_set_error(page->mapping, ret); |
87826df0 | 2813 | } |
87826df0 JM |
2814 | } |
2815 | ||
d1310b2e CM |
2816 | /* |
2817 | * after a writepage IO is done, we need to: | |
2818 | * clear the uptodate bits on error | |
2819 | * clear the writeback bits in the extent tree for this IO | |
2820 | * end_page_writeback if the page has no more pending IO | |
2821 | * | |
2822 | * Scheduling is not allowed, so the extent state tree is expected | |
2823 | * to have one and only one object corresponding to this IO. | |
2824 | */ | |
4246a0b6 | 2825 | static void end_bio_extent_writepage(struct bio *bio) |
d1310b2e | 2826 | { |
4e4cbee9 | 2827 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 2828 | struct bio_vec *bvec; |
d1310b2e CM |
2829 | u64 start; |
2830 | u64 end; | |
6dc4f100 | 2831 | struct bvec_iter_all iter_all; |
d8e3fb10 | 2832 | bool first_bvec = true; |
d1310b2e | 2833 | |
c09abff8 | 2834 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2835 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 2836 | struct page *page = bvec->bv_page; |
0b246afa JM |
2837 | struct inode *inode = page->mapping->host; |
2838 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
2839 | const u32 sectorsize = fs_info->sectorsize; |
2840 | ||
2841 | /* Our read/write should always be sector aligned. */ | |
2842 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
2843 | btrfs_err(fs_info, | |
2844 | "partial page write in btrfs with offset %u and length %u", | |
2845 | bvec->bv_offset, bvec->bv_len); | |
2846 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
2847 | btrfs_info(fs_info, | |
2848 | "incomplete page write with offset %u and length %u", | |
2849 | bvec->bv_offset, bvec->bv_len); | |
2850 | ||
2851 | start = page_offset(page) + bvec->bv_offset; | |
2852 | end = start + bvec->bv_len - 1; | |
d1310b2e | 2853 | |
d8e3fb10 NA |
2854 | if (first_bvec) { |
2855 | btrfs_record_physical_zoned(inode, start, bio); | |
2856 | first_bvec = false; | |
2857 | } | |
2858 | ||
4e4cbee9 | 2859 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
2860 | |
2861 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 2862 | } |
2b1f55b0 | 2863 | |
d1310b2e | 2864 | bio_put(bio); |
d1310b2e CM |
2865 | } |
2866 | ||
94e8c95c QW |
2867 | /* |
2868 | * Record previously processed extent range | |
2869 | * | |
2870 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
2871 | * the extent io operations. | |
2872 | */ | |
2873 | struct processed_extent { | |
2874 | struct btrfs_inode *inode; | |
2875 | /* Start of the range in @inode */ | |
2876 | u64 start; | |
2e626e56 | 2877 | /* End of the range in @inode */ |
94e8c95c QW |
2878 | u64 end; |
2879 | bool uptodate; | |
2880 | }; | |
2881 | ||
2882 | /* | |
2883 | * Try to release processed extent range | |
2884 | * | |
2885 | * May not release the extent range right now if the current range is | |
2886 | * contiguous to processed extent. | |
2887 | * | |
2888 | * Will release processed extent when any of @inode, @uptodate, the range is | |
2889 | * no longer contiguous to the processed range. | |
2890 | * | |
2891 | * Passing @inode == NULL will force processed extent to be released. | |
2892 | */ | |
2893 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
2894 | struct btrfs_inode *inode, u64 start, u64 end, | |
2895 | bool uptodate) | |
883d0de4 MX |
2896 | { |
2897 | struct extent_state *cached = NULL; | |
94e8c95c QW |
2898 | struct extent_io_tree *tree; |
2899 | ||
2900 | /* The first extent, initialize @processed */ | |
2901 | if (!processed->inode) | |
2902 | goto update; | |
883d0de4 | 2903 | |
94e8c95c QW |
2904 | /* |
2905 | * Contiguous to processed extent, just uptodate the end. | |
2906 | * | |
2907 | * Several things to notice: | |
2908 | * | |
2909 | * - bio can be merged as long as on-disk bytenr is contiguous | |
2910 | * This means we can have page belonging to other inodes, thus need to | |
2911 | * check if the inode still matches. | |
2912 | * - bvec can contain range beyond current page for multi-page bvec | |
2913 | * Thus we need to do processed->end + 1 >= start check | |
2914 | */ | |
2915 | if (processed->inode == inode && processed->uptodate == uptodate && | |
2916 | processed->end + 1 >= start && end >= processed->end) { | |
2917 | processed->end = end; | |
2918 | return; | |
2919 | } | |
2920 | ||
2921 | tree = &processed->inode->io_tree; | |
2922 | /* | |
2923 | * Now we don't have range contiguous to the processed range, release | |
2924 | * the processed range now. | |
2925 | */ | |
2926 | if (processed->uptodate && tree->track_uptodate) | |
2927 | set_extent_uptodate(tree, processed->start, processed->end, | |
2928 | &cached, GFP_ATOMIC); | |
2929 | unlock_extent_cached_atomic(tree, processed->start, processed->end, | |
2930 | &cached); | |
2931 | ||
2932 | update: | |
2933 | /* Update processed to current range */ | |
2934 | processed->inode = inode; | |
2935 | processed->start = start; | |
2936 | processed->end = end; | |
2937 | processed->uptodate = uptodate; | |
883d0de4 MX |
2938 | } |
2939 | ||
92082d40 QW |
2940 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
2941 | { | |
2942 | ASSERT(PageLocked(page)); | |
2943 | if (fs_info->sectorsize == PAGE_SIZE) | |
2944 | return; | |
2945 | ||
2946 | ASSERT(PagePrivate(page)); | |
2947 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
2948 | } | |
2949 | ||
d9bb77d5 QW |
2950 | /* |
2951 | * Find extent buffer for a givne bytenr. | |
2952 | * | |
2953 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
2954 | * in endio context. | |
2955 | */ | |
2956 | static struct extent_buffer *find_extent_buffer_readpage( | |
2957 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
2958 | { | |
2959 | struct extent_buffer *eb; | |
2960 | ||
2961 | /* | |
2962 | * For regular sectorsize, we can use page->private to grab extent | |
2963 | * buffer | |
2964 | */ | |
2965 | if (fs_info->sectorsize == PAGE_SIZE) { | |
2966 | ASSERT(PagePrivate(page) && page->private); | |
2967 | return (struct extent_buffer *)page->private; | |
2968 | } | |
2969 | ||
2970 | /* For subpage case, we need to lookup buffer radix tree */ | |
2971 | rcu_read_lock(); | |
2972 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
2973 | bytenr >> fs_info->sectorsize_bits); | |
2974 | rcu_read_unlock(); | |
2975 | ASSERT(eb); | |
2976 | return eb; | |
2977 | } | |
2978 | ||
d1310b2e CM |
2979 | /* |
2980 | * after a readpage IO is done, we need to: | |
2981 | * clear the uptodate bits on error | |
2982 | * set the uptodate bits if things worked | |
2983 | * set the page up to date if all extents in the tree are uptodate | |
2984 | * clear the lock bit in the extent tree | |
2985 | * unlock the page if there are no other extents locked for it | |
2986 | * | |
2987 | * Scheduling is not allowed, so the extent state tree is expected | |
2988 | * to have one and only one object corresponding to this IO. | |
2989 | */ | |
4246a0b6 | 2990 | static void end_bio_extent_readpage(struct bio *bio) |
d1310b2e | 2991 | { |
2c30c71b | 2992 | struct bio_vec *bvec; |
facc8a22 | 2993 | struct btrfs_io_bio *io_bio = btrfs_io_bio(bio); |
7870d082 | 2994 | struct extent_io_tree *tree, *failure_tree; |
94e8c95c | 2995 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
2996 | /* |
2997 | * The offset to the beginning of a bio, since one bio can never be | |
2998 | * larger than UINT_MAX, u32 here is enough. | |
2999 | */ | |
3000 | u32 bio_offset = 0; | |
5cf1ab56 | 3001 | int mirror; |
d1310b2e | 3002 | int ret; |
6dc4f100 | 3003 | struct bvec_iter_all iter_all; |
d1310b2e | 3004 | |
c09abff8 | 3005 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 3006 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 3007 | bool uptodate = !bio->bi_status; |
d1310b2e | 3008 | struct page *page = bvec->bv_page; |
a71754fc | 3009 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 3010 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 | 3011 | const u32 sectorsize = fs_info->sectorsize; |
150e4b05 | 3012 | unsigned int error_bitmap = (unsigned int)-1; |
7ffd27e3 QW |
3013 | u64 start; |
3014 | u64 end; | |
3015 | u32 len; | |
507903b8 | 3016 | |
ab8d0fc4 JM |
3017 | btrfs_debug(fs_info, |
3018 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 3019 | bio->bi_iter.bi_sector, bio->bi_status, |
ab8d0fc4 | 3020 | io_bio->mirror_num); |
a71754fc | 3021 | tree = &BTRFS_I(inode)->io_tree; |
7870d082 | 3022 | failure_tree = &BTRFS_I(inode)->io_failure_tree; |
902b22f3 | 3023 | |
8b8bbd46 QW |
3024 | /* |
3025 | * We always issue full-sector reads, but if some block in a | |
3026 | * page fails to read, blk_update_request() will advance | |
3027 | * bv_offset and adjust bv_len to compensate. Print a warning | |
3028 | * for unaligned offsets, and an error if they don't add up to | |
3029 | * a full sector. | |
3030 | */ | |
3031 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
3032 | btrfs_err(fs_info, | |
3033 | "partial page read in btrfs with offset %u and length %u", | |
3034 | bvec->bv_offset, bvec->bv_len); | |
3035 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
3036 | sectorsize)) | |
3037 | btrfs_info(fs_info, | |
3038 | "incomplete page read with offset %u and length %u", | |
3039 | bvec->bv_offset, bvec->bv_len); | |
3040 | ||
3041 | start = page_offset(page) + bvec->bv_offset; | |
3042 | end = start + bvec->bv_len - 1; | |
facc8a22 | 3043 | len = bvec->bv_len; |
d1310b2e | 3044 | |
9be3395b | 3045 | mirror = io_bio->mirror_num; |
78e62c02 | 3046 | if (likely(uptodate)) { |
150e4b05 QW |
3047 | if (is_data_inode(inode)) { |
3048 | error_bitmap = btrfs_verify_data_csum(io_bio, | |
5e295768 | 3049 | bio_offset, page, start, end); |
150e4b05 QW |
3050 | ret = error_bitmap; |
3051 | } else { | |
9a446d6a | 3052 | ret = btrfs_validate_metadata_buffer(io_bio, |
8e1dc982 | 3053 | page, start, end, mirror); |
150e4b05 | 3054 | } |
5ee0844d | 3055 | if (ret) |
150e4b05 | 3056 | uptodate = false; |
5ee0844d | 3057 | else |
7870d082 JB |
3058 | clean_io_failure(BTRFS_I(inode)->root->fs_info, |
3059 | failure_tree, tree, start, | |
3060 | page, | |
3061 | btrfs_ino(BTRFS_I(inode)), 0); | |
d1310b2e | 3062 | } |
ea466794 | 3063 | |
f2a09da9 MX |
3064 | if (likely(uptodate)) |
3065 | goto readpage_ok; | |
3066 | ||
be17b3af | 3067 | if (is_data_inode(inode)) { |
f4a8e656 | 3068 | /* |
150e4b05 QW |
3069 | * btrfs_submit_read_repair() will handle all the good |
3070 | * and bad sectors, we just continue to the next bvec. | |
f4a8e656 | 3071 | */ |
150e4b05 QW |
3072 | submit_read_repair(inode, bio, bio_offset, page, |
3073 | start - page_offset(page), start, | |
3074 | end, mirror, error_bitmap, | |
3075 | btrfs_submit_data_bio); | |
3076 | ||
3077 | ASSERT(bio_offset + len > bio_offset); | |
3078 | bio_offset += len; | |
3079 | continue; | |
78e62c02 NB |
3080 | } else { |
3081 | struct extent_buffer *eb; | |
3082 | ||
d9bb77d5 | 3083 | eb = find_extent_buffer_readpage(fs_info, page, start); |
78e62c02 NB |
3084 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
3085 | eb->read_mirror = mirror; | |
3086 | atomic_dec(&eb->io_pages); | |
3087 | if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, | |
3088 | &eb->bflags)) | |
3089 | btree_readahead_hook(eb, -EIO); | |
7e38326f | 3090 | } |
f2a09da9 | 3091 | readpage_ok: |
883d0de4 | 3092 | if (likely(uptodate)) { |
a71754fc | 3093 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3094 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 3095 | |
c28ea613 QW |
3096 | /* |
3097 | * Zero out the remaining part if this range straddles | |
3098 | * i_size. | |
3099 | * | |
3100 | * Here we should only zero the range inside the bvec, | |
3101 | * not touch anything else. | |
3102 | * | |
3103 | * NOTE: i_size is exclusive while end is inclusive. | |
3104 | */ | |
3105 | if (page->index == end_index && i_size <= end) { | |
3106 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 3107 | offset_in_page(start)); |
c28ea613 QW |
3108 | |
3109 | zero_user_segment(page, zero_start, | |
3110 | offset_in_page(end) + 1); | |
3111 | } | |
70dec807 | 3112 | } |
7ffd27e3 QW |
3113 | ASSERT(bio_offset + len > bio_offset); |
3114 | bio_offset += len; | |
883d0de4 | 3115 | |
e09caaf9 | 3116 | /* Update page status and unlock */ |
92082d40 | 3117 | end_page_read(page, uptodate, start, len); |
94e8c95c QW |
3118 | endio_readpage_release_extent(&processed, BTRFS_I(inode), |
3119 | start, end, uptodate); | |
2c30c71b | 3120 | } |
94e8c95c QW |
3121 | /* Release the last extent */ |
3122 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
b3a0dd50 | 3123 | btrfs_io_bio_free_csum(io_bio); |
d1310b2e | 3124 | bio_put(bio); |
d1310b2e CM |
3125 | } |
3126 | ||
9be3395b | 3127 | /* |
184f999e DS |
3128 | * Initialize the members up to but not including 'bio'. Use after allocating a |
3129 | * new bio by bio_alloc_bioset as it does not initialize the bytes outside of | |
3130 | * 'bio' because use of __GFP_ZERO is not supported. | |
9be3395b | 3131 | */ |
184f999e | 3132 | static inline void btrfs_io_bio_init(struct btrfs_io_bio *btrfs_bio) |
d1310b2e | 3133 | { |
184f999e DS |
3134 | memset(btrfs_bio, 0, offsetof(struct btrfs_io_bio, bio)); |
3135 | } | |
d1310b2e | 3136 | |
9be3395b | 3137 | /* |
6e707bcd DS |
3138 | * The following helpers allocate a bio. As it's backed by a bioset, it'll |
3139 | * never fail. We're returning a bio right now but you can call btrfs_io_bio | |
3140 | * for the appropriate container_of magic | |
9be3395b | 3141 | */ |
e749af44 | 3142 | struct bio *btrfs_bio_alloc(u64 first_byte) |
d1310b2e CM |
3143 | { |
3144 | struct bio *bio; | |
d1310b2e | 3145 | |
a8affc03 | 3146 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &btrfs_bioset); |
c821e7f3 | 3147 | bio->bi_iter.bi_sector = first_byte >> 9; |
184f999e | 3148 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
d1310b2e CM |
3149 | return bio; |
3150 | } | |
3151 | ||
8b6c1d56 | 3152 | struct bio *btrfs_bio_clone(struct bio *bio) |
9be3395b | 3153 | { |
23ea8e5a MX |
3154 | struct btrfs_io_bio *btrfs_bio; |
3155 | struct bio *new; | |
9be3395b | 3156 | |
6e707bcd | 3157 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 3158 | new = bio_clone_fast(bio, GFP_NOFS, &btrfs_bioset); |
6e707bcd | 3159 | btrfs_bio = btrfs_io_bio(new); |
184f999e | 3160 | btrfs_io_bio_init(btrfs_bio); |
6e707bcd | 3161 | btrfs_bio->iter = bio->bi_iter; |
23ea8e5a MX |
3162 | return new; |
3163 | } | |
9be3395b | 3164 | |
c5e4c3d7 | 3165 | struct bio *btrfs_io_bio_alloc(unsigned int nr_iovecs) |
9be3395b | 3166 | { |
facc8a22 MX |
3167 | struct bio *bio; |
3168 | ||
6e707bcd | 3169 | /* Bio allocation backed by a bioset does not fail */ |
8ac9f7c1 | 3170 | bio = bio_alloc_bioset(GFP_NOFS, nr_iovecs, &btrfs_bioset); |
184f999e | 3171 | btrfs_io_bio_init(btrfs_io_bio(bio)); |
facc8a22 | 3172 | return bio; |
9be3395b CM |
3173 | } |
3174 | ||
e477094f | 3175 | struct bio *btrfs_bio_clone_partial(struct bio *orig, int offset, int size) |
2f8e9140 LB |
3176 | { |
3177 | struct bio *bio; | |
3178 | struct btrfs_io_bio *btrfs_bio; | |
3179 | ||
3180 | /* this will never fail when it's backed by a bioset */ | |
8ac9f7c1 | 3181 | bio = bio_clone_fast(orig, GFP_NOFS, &btrfs_bioset); |
2f8e9140 LB |
3182 | ASSERT(bio); |
3183 | ||
3184 | btrfs_bio = btrfs_io_bio(bio); | |
184f999e | 3185 | btrfs_io_bio_init(btrfs_bio); |
2f8e9140 LB |
3186 | |
3187 | bio_trim(bio, offset >> 9, size >> 9); | |
17347cec | 3188 | btrfs_bio->iter = bio->bi_iter; |
2f8e9140 LB |
3189 | return bio; |
3190 | } | |
9be3395b | 3191 | |
953651eb NA |
3192 | /** |
3193 | * Attempt to add a page to bio | |
3194 | * | |
3195 | * @bio: destination bio | |
3196 | * @page: page to add to the bio | |
3197 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
3198 | * a contiguous page to the previous one | |
3199 | * @pg_offset: starting offset in the page | |
3200 | * @size: portion of page that we want to write | |
3201 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3202 | * @bio_flags: flags of the current bio to see if we can merge them | |
3203 | * @return: true if page was added, false otherwise | |
3204 | * | |
3205 | * Attempt to add a page to bio considering stripe alignment etc. | |
3206 | * | |
3207 | * Return true if successfully page added. Otherwise, return false. | |
3208 | */ | |
390ed29b QW |
3209 | static bool btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
3210 | struct page *page, | |
953651eb NA |
3211 | u64 disk_bytenr, unsigned int size, |
3212 | unsigned int pg_offset, | |
953651eb NA |
3213 | unsigned long bio_flags) |
3214 | { | |
390ed29b QW |
3215 | struct bio *bio = bio_ctrl->bio; |
3216 | u32 bio_size = bio->bi_iter.bi_size; | |
953651eb NA |
3217 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
3218 | bool contig; | |
e1326f03 | 3219 | int ret; |
953651eb | 3220 | |
390ed29b QW |
3221 | ASSERT(bio); |
3222 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
3223 | ASSERT(bio_ctrl->len_to_oe_boundary && bio_ctrl->len_to_stripe_boundary); | |
3224 | if (bio_ctrl->bio_flags != bio_flags) | |
953651eb NA |
3225 | return false; |
3226 | ||
390ed29b | 3227 | if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED) |
953651eb NA |
3228 | contig = bio->bi_iter.bi_sector == sector; |
3229 | else | |
3230 | contig = bio_end_sector(bio) == sector; | |
3231 | if (!contig) | |
3232 | return false; | |
3233 | ||
390ed29b QW |
3234 | if (bio_size + size > bio_ctrl->len_to_oe_boundary || |
3235 | bio_size + size > bio_ctrl->len_to_stripe_boundary) | |
953651eb NA |
3236 | return false; |
3237 | ||
390ed29b | 3238 | if (bio_op(bio) == REQ_OP_ZONE_APPEND) |
e1326f03 | 3239 | ret = bio_add_zone_append_page(bio, page, size, pg_offset); |
390ed29b | 3240 | else |
e1326f03 NA |
3241 | ret = bio_add_page(bio, page, size, pg_offset); |
3242 | ||
3243 | return ret == size; | |
953651eb NA |
3244 | } |
3245 | ||
390ed29b QW |
3246 | static int calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
3247 | struct btrfs_inode *inode) | |
3248 | { | |
3249 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
3250 | struct btrfs_io_geometry geom; | |
3251 | struct btrfs_ordered_extent *ordered; | |
3252 | struct extent_map *em; | |
3253 | u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT); | |
3254 | int ret; | |
3255 | ||
3256 | /* | |
3257 | * Pages for compressed extent are never submitted to disk directly, | |
3258 | * thus it has no real boundary, just set them to U32_MAX. | |
3259 | * | |
3260 | * The split happens for real compressed bio, which happens in | |
3261 | * btrfs_submit_compressed_read/write(). | |
3262 | */ | |
3263 | if (bio_ctrl->bio_flags & EXTENT_BIO_COMPRESSED) { | |
3264 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3265 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3266 | return 0; | |
3267 | } | |
3268 | em = btrfs_get_chunk_map(fs_info, logical, fs_info->sectorsize); | |
3269 | if (IS_ERR(em)) | |
3270 | return PTR_ERR(em); | |
3271 | ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio_ctrl->bio), | |
3272 | logical, &geom); | |
3273 | free_extent_map(em); | |
3274 | if (ret < 0) { | |
3275 | return ret; | |
3276 | } | |
3277 | if (geom.len > U32_MAX) | |
3278 | bio_ctrl->len_to_stripe_boundary = U32_MAX; | |
3279 | else | |
3280 | bio_ctrl->len_to_stripe_boundary = (u32)geom.len; | |
3281 | ||
3282 | if (!btrfs_is_zoned(fs_info) || | |
3283 | bio_op(bio_ctrl->bio) != REQ_OP_ZONE_APPEND) { | |
3284 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3285 | return 0; | |
3286 | } | |
3287 | ||
3288 | ASSERT(fs_info->max_zone_append_size > 0); | |
3289 | /* Ordered extent not yet created, so we're good */ | |
3290 | ordered = btrfs_lookup_ordered_extent(inode, logical); | |
3291 | if (!ordered) { | |
3292 | bio_ctrl->len_to_oe_boundary = U32_MAX; | |
3293 | return 0; | |
3294 | } | |
3295 | ||
3296 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
3297 | ordered->disk_bytenr + ordered->disk_num_bytes - logical); | |
3298 | btrfs_put_ordered_extent(ordered); | |
3299 | return 0; | |
3300 | } | |
3301 | ||
4b81ba48 DS |
3302 | /* |
3303 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 DS |
3304 | * @wbc: optional writeback control for io accounting |
3305 | * @page: page to add to the bio | |
0c64c33c QW |
3306 | * @disk_bytenr: logical bytenr where the write will be |
3307 | * @size: portion of page that we want to write to | |
b8b3d625 DS |
3308 | * @pg_offset: offset of the new bio or to check whether we are adding |
3309 | * a contiguous page to the previous one | |
5c2b1fd7 | 3310 | * @bio_ret: must be valid pointer, newly allocated bio will be stored there |
b8b3d625 DS |
3311 | * @end_io_func: end_io callback for new bio |
3312 | * @mirror_num: desired mirror to read/write | |
3313 | * @prev_bio_flags: flags of previous bio to see if we can merge the current one | |
3314 | * @bio_flags: flags of the current bio to see if we can merge them | |
4b81ba48 | 3315 | */ |
0ceb34bf | 3316 | static int submit_extent_page(unsigned int opf, |
da2f0f74 | 3317 | struct writeback_control *wbc, |
390ed29b | 3318 | struct btrfs_bio_ctrl *bio_ctrl, |
0c64c33c | 3319 | struct page *page, u64 disk_bytenr, |
6c5a4e2c | 3320 | size_t size, unsigned long pg_offset, |
f188591e | 3321 | bio_end_io_t end_io_func, |
c8b97818 | 3322 | int mirror_num, |
005efedf FM |
3323 | unsigned long bio_flags, |
3324 | bool force_bio_submit) | |
d1310b2e CM |
3325 | { |
3326 | int ret = 0; | |
3327 | struct bio *bio; | |
e940e9a7 | 3328 | size_t io_size = min_t(size_t, size, PAGE_SIZE); |
e1326f03 NA |
3329 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
3330 | struct extent_io_tree *tree = &inode->io_tree; | |
3331 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
d1310b2e | 3332 | |
390ed29b | 3333 | ASSERT(bio_ctrl); |
5c2b1fd7 | 3334 | |
390ed29b QW |
3335 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
3336 | pg_offset + size <= PAGE_SIZE); | |
3337 | if (bio_ctrl->bio) { | |
3338 | bio = bio_ctrl->bio; | |
953651eb | 3339 | if (force_bio_submit || |
390ed29b QW |
3340 | !btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, io_size, |
3341 | pg_offset, bio_flags)) { | |
3342 | ret = submit_one_bio(bio, mirror_num, bio_ctrl->bio_flags); | |
3343 | bio_ctrl->bio = NULL; | |
3344 | if (ret < 0) | |
79787eaa | 3345 | return ret; |
d1310b2e | 3346 | } else { |
da2f0f74 | 3347 | if (wbc) |
e940e9a7 | 3348 | wbc_account_cgroup_owner(wbc, page, io_size); |
d1310b2e CM |
3349 | return 0; |
3350 | } | |
3351 | } | |
c8b97818 | 3352 | |
0c64c33c | 3353 | bio = btrfs_bio_alloc(disk_bytenr); |
e940e9a7 | 3354 | bio_add_page(bio, page, io_size, pg_offset); |
d1310b2e CM |
3355 | bio->bi_end_io = end_io_func; |
3356 | bio->bi_private = tree; | |
e6959b93 | 3357 | bio->bi_write_hint = page->mapping->host->i_write_hint; |
4b81ba48 | 3358 | bio->bi_opf = opf; |
da2f0f74 | 3359 | if (wbc) { |
429aebc0 DS |
3360 | struct block_device *bdev; |
3361 | ||
e1326f03 | 3362 | bdev = fs_info->fs_devices->latest_bdev; |
429aebc0 | 3363 | bio_set_dev(bio, bdev); |
da2f0f74 | 3364 | wbc_init_bio(wbc, bio); |
e940e9a7 | 3365 | wbc_account_cgroup_owner(wbc, page, io_size); |
da2f0f74 | 3366 | } |
e1326f03 | 3367 | if (btrfs_is_zoned(fs_info) && bio_op(bio) == REQ_OP_ZONE_APPEND) { |
e7ff9e6b | 3368 | struct btrfs_device *device; |
e1326f03 | 3369 | |
e7ff9e6b JT |
3370 | device = btrfs_zoned_get_device(fs_info, disk_bytenr, io_size); |
3371 | if (IS_ERR(device)) | |
3372 | return PTR_ERR(device); | |
e1326f03 | 3373 | |
e7ff9e6b | 3374 | btrfs_io_bio(bio)->device = device; |
e1326f03 | 3375 | } |
70dec807 | 3376 | |
390ed29b QW |
3377 | bio_ctrl->bio = bio; |
3378 | bio_ctrl->bio_flags = bio_flags; | |
3379 | ret = calc_bio_boundaries(bio_ctrl, inode); | |
d1310b2e CM |
3380 | |
3381 | return ret; | |
3382 | } | |
3383 | ||
760f991f QW |
3384 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
3385 | struct page *page, | |
3386 | struct btrfs_subpage *prealloc) | |
d1310b2e | 3387 | { |
760f991f QW |
3388 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3389 | int ret = 0; | |
3390 | ||
0d01e247 QW |
3391 | /* |
3392 | * If the page is mapped to btree inode, we should hold the private | |
3393 | * lock to prevent race. | |
3394 | * For cloned or dummy extent buffers, their pages are not mapped and | |
3395 | * will not race with any other ebs. | |
3396 | */ | |
3397 | if (page->mapping) | |
3398 | lockdep_assert_held(&page->mapping->private_lock); | |
3399 | ||
760f991f QW |
3400 | if (fs_info->sectorsize == PAGE_SIZE) { |
3401 | if (!PagePrivate(page)) | |
3402 | attach_page_private(page, eb); | |
3403 | else | |
3404 | WARN_ON(page->private != (unsigned long)eb); | |
3405 | return 0; | |
3406 | } | |
3407 | ||
3408 | /* Already mapped, just free prealloc */ | |
3409 | if (PagePrivate(page)) { | |
3410 | btrfs_free_subpage(prealloc); | |
3411 | return 0; | |
3412 | } | |
3413 | ||
3414 | if (prealloc) | |
3415 | /* Has preallocated memory for subpage */ | |
3416 | attach_page_private(page, prealloc); | |
d1b89bc0 | 3417 | else |
760f991f QW |
3418 | /* Do new allocation to attach subpage */ |
3419 | ret = btrfs_attach_subpage(fs_info, page, | |
3420 | BTRFS_SUBPAGE_METADATA); | |
3421 | return ret; | |
d1310b2e CM |
3422 | } |
3423 | ||
32443de3 | 3424 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 3425 | { |
32443de3 QW |
3426 | struct btrfs_fs_info *fs_info; |
3427 | ||
3428 | ASSERT(page->mapping); | |
3429 | ||
3430 | if (PagePrivate(page)) | |
3431 | return 0; | |
3432 | ||
3433 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3434 | ||
3435 | if (fs_info->sectorsize < PAGE_SIZE) | |
3436 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); | |
3437 | ||
3438 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
3439 | return 0; | |
3440 | } | |
3441 | ||
3442 | void clear_page_extent_mapped(struct page *page) | |
3443 | { | |
3444 | struct btrfs_fs_info *fs_info; | |
3445 | ||
3446 | ASSERT(page->mapping); | |
3447 | ||
d1b89bc0 | 3448 | if (!PagePrivate(page)) |
32443de3 QW |
3449 | return; |
3450 | ||
3451 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
3452 | if (fs_info->sectorsize < PAGE_SIZE) | |
3453 | return btrfs_detach_subpage(fs_info, page); | |
3454 | ||
3455 | detach_page_private(page); | |
d1310b2e CM |
3456 | } |
3457 | ||
125bac01 MX |
3458 | static struct extent_map * |
3459 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 3460 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
3461 | { |
3462 | struct extent_map *em; | |
3463 | ||
3464 | if (em_cached && *em_cached) { | |
3465 | em = *em_cached; | |
cbc0e928 | 3466 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 3467 | start < extent_map_end(em)) { |
490b54d6 | 3468 | refcount_inc(&em->refs); |
125bac01 MX |
3469 | return em; |
3470 | } | |
3471 | ||
3472 | free_extent_map(em); | |
3473 | *em_cached = NULL; | |
3474 | } | |
3475 | ||
1a5ee1e6 | 3476 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
125bac01 MX |
3477 | if (em_cached && !IS_ERR_OR_NULL(em)) { |
3478 | BUG_ON(*em_cached); | |
490b54d6 | 3479 | refcount_inc(&em->refs); |
125bac01 MX |
3480 | *em_cached = em; |
3481 | } | |
3482 | return em; | |
3483 | } | |
d1310b2e CM |
3484 | /* |
3485 | * basic readpage implementation. Locked extent state structs are inserted | |
3486 | * into the tree that are removed when the IO is done (by the end_io | |
3487 | * handlers) | |
79787eaa | 3488 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 3489 | * return 0 on success, otherwise return error |
d1310b2e | 3490 | */ |
0f208812 | 3491 | int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
390ed29b | 3492 | struct btrfs_bio_ctrl *bio_ctrl, |
0f208812 | 3493 | unsigned int read_flags, u64 *prev_em_start) |
d1310b2e CM |
3494 | { |
3495 | struct inode *inode = page->mapping->host; | |
92082d40 | 3496 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 3497 | u64 start = page_offset(page); |
8eec8296 | 3498 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3499 | u64 cur = start; |
3500 | u64 extent_offset; | |
3501 | u64 last_byte = i_size_read(inode); | |
3502 | u64 block_start; | |
3503 | u64 cur_end; | |
d1310b2e | 3504 | struct extent_map *em; |
baf863b9 | 3505 | int ret = 0; |
d1310b2e | 3506 | int nr = 0; |
306e16ce | 3507 | size_t pg_offset = 0; |
d1310b2e CM |
3508 | size_t iosize; |
3509 | size_t blocksize = inode->i_sb->s_blocksize; | |
7f042a83 | 3510 | unsigned long this_bio_flag = 0; |
f657a31c | 3511 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 3512 | |
32443de3 QW |
3513 | ret = set_page_extent_mapped(page); |
3514 | if (ret < 0) { | |
3515 | unlock_extent(tree, start, end); | |
92082d40 QW |
3516 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
3517 | unlock_page(page); | |
32443de3 QW |
3518 | goto out; |
3519 | } | |
d1310b2e | 3520 | |
90a887c9 DM |
3521 | if (!PageUptodate(page)) { |
3522 | if (cleancache_get_page(page) == 0) { | |
3523 | BUG_ON(blocksize != PAGE_SIZE); | |
9974090b | 3524 | unlock_extent(tree, start, end); |
92082d40 | 3525 | unlock_page(page); |
90a887c9 DM |
3526 | goto out; |
3527 | } | |
3528 | } | |
3529 | ||
09cbfeaf | 3530 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 3531 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
3532 | |
3533 | if (zero_offset) { | |
09cbfeaf | 3534 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 3535 | memzero_page(page, zero_offset, iosize); |
c8b97818 | 3536 | flush_dcache_page(page); |
c8b97818 CM |
3537 | } |
3538 | } | |
92082d40 | 3539 | begin_page_read(fs_info, page); |
d1310b2e | 3540 | while (cur <= end) { |
005efedf | 3541 | bool force_bio_submit = false; |
0c64c33c | 3542 | u64 disk_bytenr; |
c8f2f24b | 3543 | |
d1310b2e | 3544 | if (cur >= last_byte) { |
507903b8 AJ |
3545 | struct extent_state *cached = NULL; |
3546 | ||
09cbfeaf | 3547 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 3548 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3549 | flush_dcache_page(page); |
d1310b2e | 3550 | set_extent_uptodate(tree, cur, cur + iosize - 1, |
507903b8 | 3551 | &cached, GFP_NOFS); |
7f042a83 | 3552 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3553 | cur + iosize - 1, &cached); |
92082d40 | 3554 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
3555 | break; |
3556 | } | |
125bac01 | 3557 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 3558 | end - cur + 1, em_cached); |
c704005d | 3559 | if (IS_ERR_OR_NULL(em)) { |
7f042a83 | 3560 | unlock_extent(tree, cur, end); |
92082d40 | 3561 | end_page_read(page, false, cur, end + 1 - cur); |
d1310b2e CM |
3562 | break; |
3563 | } | |
d1310b2e CM |
3564 | extent_offset = cur - em->start; |
3565 | BUG_ON(extent_map_end(em) <= cur); | |
3566 | BUG_ON(end < cur); | |
3567 | ||
261507a0 | 3568 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) { |
4b384318 | 3569 | this_bio_flag |= EXTENT_BIO_COMPRESSED; |
261507a0 LZ |
3570 | extent_set_compress_type(&this_bio_flag, |
3571 | em->compress_type); | |
3572 | } | |
c8b97818 | 3573 | |
d1310b2e CM |
3574 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
3575 | cur_end = min(extent_map_end(em) - 1, end); | |
fda2832f | 3576 | iosize = ALIGN(iosize, blocksize); |
949b3273 | 3577 | if (this_bio_flag & EXTENT_BIO_COMPRESSED) |
0c64c33c | 3578 | disk_bytenr = em->block_start; |
949b3273 | 3579 | else |
0c64c33c | 3580 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 3581 | block_start = em->block_start; |
d899e052 YZ |
3582 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
3583 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
3584 | |
3585 | /* | |
3586 | * If we have a file range that points to a compressed extent | |
260db43c | 3587 | * and it's followed by a consecutive file range that points |
005efedf FM |
3588 | * to the same compressed extent (possibly with a different |
3589 | * offset and/or length, so it either points to the whole extent | |
3590 | * or only part of it), we must make sure we do not submit a | |
3591 | * single bio to populate the pages for the 2 ranges because | |
3592 | * this makes the compressed extent read zero out the pages | |
3593 | * belonging to the 2nd range. Imagine the following scenario: | |
3594 | * | |
3595 | * File layout | |
3596 | * [0 - 8K] [8K - 24K] | |
3597 | * | | | |
3598 | * | | | |
3599 | * points to extent X, points to extent X, | |
3600 | * offset 4K, length of 8K offset 0, length 16K | |
3601 | * | |
3602 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
3603 | * | |
3604 | * If the bio to read the compressed extent covers both ranges, | |
3605 | * it will decompress extent X into the pages belonging to the | |
3606 | * first range and then it will stop, zeroing out the remaining | |
3607 | * pages that belong to the other range that points to extent X. | |
3608 | * So here we make sure we submit 2 bios, one for the first | |
3609 | * range and another one for the third range. Both will target | |
3610 | * the same physical extent from disk, but we can't currently | |
3611 | * make the compressed bio endio callback populate the pages | |
3612 | * for both ranges because each compressed bio is tightly | |
3613 | * coupled with a single extent map, and each range can have | |
3614 | * an extent map with a different offset value relative to the | |
3615 | * uncompressed data of our extent and different lengths. This | |
3616 | * is a corner case so we prioritize correctness over | |
3617 | * non-optimal behavior (submitting 2 bios for the same extent). | |
3618 | */ | |
3619 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
3620 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 3621 | *prev_em_start != em->start) |
005efedf FM |
3622 | force_bio_submit = true; |
3623 | ||
3624 | if (prev_em_start) | |
8e928218 | 3625 | *prev_em_start = em->start; |
005efedf | 3626 | |
d1310b2e CM |
3627 | free_extent_map(em); |
3628 | em = NULL; | |
3629 | ||
3630 | /* we've found a hole, just zero and go on */ | |
3631 | if (block_start == EXTENT_MAP_HOLE) { | |
507903b8 AJ |
3632 | struct extent_state *cached = NULL; |
3633 | ||
d048b9c2 | 3634 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 3635 | flush_dcache_page(page); |
d1310b2e CM |
3636 | |
3637 | set_extent_uptodate(tree, cur, cur + iosize - 1, | |
507903b8 | 3638 | &cached, GFP_NOFS); |
7f042a83 | 3639 | unlock_extent_cached(tree, cur, |
e43bbe5e | 3640 | cur + iosize - 1, &cached); |
92082d40 | 3641 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3642 | cur = cur + iosize; |
306e16ce | 3643 | pg_offset += iosize; |
d1310b2e CM |
3644 | continue; |
3645 | } | |
3646 | /* the get_extent function already copied into the page */ | |
9655d298 CM |
3647 | if (test_range_bit(tree, cur, cur_end, |
3648 | EXTENT_UPTODATE, 1, NULL)) { | |
a1b32a59 | 3649 | check_page_uptodate(tree, page); |
7f042a83 | 3650 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3651 | end_page_read(page, true, cur, iosize); |
d1310b2e | 3652 | cur = cur + iosize; |
306e16ce | 3653 | pg_offset += iosize; |
d1310b2e CM |
3654 | continue; |
3655 | } | |
70dec807 CM |
3656 | /* we have an inline extent but it didn't get marked up |
3657 | * to date. Error out | |
3658 | */ | |
3659 | if (block_start == EXTENT_MAP_INLINE) { | |
7f042a83 | 3660 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3661 | end_page_read(page, false, cur, iosize); |
70dec807 | 3662 | cur = cur + iosize; |
306e16ce | 3663 | pg_offset += iosize; |
70dec807 CM |
3664 | continue; |
3665 | } | |
d1310b2e | 3666 | |
0ceb34bf | 3667 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
390ed29b QW |
3668 | bio_ctrl, page, disk_bytenr, iosize, |
3669 | pg_offset, | |
fd513000 | 3670 | end_bio_extent_readpage, 0, |
005efedf FM |
3671 | this_bio_flag, |
3672 | force_bio_submit); | |
c8f2f24b JB |
3673 | if (!ret) { |
3674 | nr++; | |
c8f2f24b | 3675 | } else { |
7f042a83 | 3676 | unlock_extent(tree, cur, cur + iosize - 1); |
92082d40 | 3677 | end_page_read(page, false, cur, iosize); |
baf863b9 | 3678 | goto out; |
edd33c99 | 3679 | } |
d1310b2e | 3680 | cur = cur + iosize; |
306e16ce | 3681 | pg_offset += iosize; |
d1310b2e | 3682 | } |
90a887c9 | 3683 | out: |
baf863b9 | 3684 | return ret; |
d1310b2e CM |
3685 | } |
3686 | ||
b6660e80 | 3687 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
3688 | u64 start, u64 end, |
3689 | struct extent_map **em_cached, | |
3690 | struct btrfs_bio_ctrl *bio_ctrl, | |
3691 | u64 *prev_em_start) | |
9974090b | 3692 | { |
23d31bd4 | 3693 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
3694 | int index; |
3695 | ||
b272ae22 | 3696 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
3697 | |
3698 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 3699 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
0f208812 | 3700 | REQ_RAHEAD, prev_em_start); |
09cbfeaf | 3701 | put_page(pages[index]); |
9974090b MX |
3702 | } |
3703 | } | |
3704 | ||
3d4b9496 | 3705 | static void update_nr_written(struct writeback_control *wbc, |
a9132667 | 3706 | unsigned long nr_written) |
11c8349b CM |
3707 | { |
3708 | wbc->nr_to_write -= nr_written; | |
11c8349b CM |
3709 | } |
3710 | ||
d1310b2e | 3711 | /* |
40f76580 CM |
3712 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
3713 | * | |
5eaad97a | 3714 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
3715 | * to write the page (copy into inline extent). In this case the IO has |
3716 | * been started and the page is already unlocked. | |
3717 | * | |
3718 | * This returns 0 if all went well (page still locked) | |
3719 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 3720 | */ |
cd4c0bf9 | 3721 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
8cc0237a NB |
3722 | struct page *page, struct writeback_control *wbc, |
3723 | u64 delalloc_start, unsigned long *nr_written) | |
40f76580 | 3724 | { |
09cbfeaf | 3725 | u64 page_end = delalloc_start + PAGE_SIZE - 1; |
3522e903 | 3726 | bool found; |
40f76580 CM |
3727 | u64 delalloc_to_write = 0; |
3728 | u64 delalloc_end = 0; | |
3729 | int ret; | |
3730 | int page_started = 0; | |
3731 | ||
40f76580 CM |
3732 | |
3733 | while (delalloc_end < page_end) { | |
cd4c0bf9 | 3734 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 3735 | &delalloc_start, |
917aacec | 3736 | &delalloc_end); |
3522e903 | 3737 | if (!found) { |
40f76580 CM |
3738 | delalloc_start = delalloc_end + 1; |
3739 | continue; | |
3740 | } | |
cd4c0bf9 | 3741 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
5eaad97a | 3742 | delalloc_end, &page_started, nr_written, wbc); |
40f76580 CM |
3743 | if (ret) { |
3744 | SetPageError(page); | |
5eaad97a NB |
3745 | /* |
3746 | * btrfs_run_delalloc_range should return < 0 for error | |
3747 | * but just in case, we use > 0 here meaning the IO is | |
3748 | * started, so we don't want to return > 0 unless | |
3749 | * things are going well. | |
40f76580 | 3750 | */ |
b69d1ee9 | 3751 | return ret < 0 ? ret : -EIO; |
40f76580 CM |
3752 | } |
3753 | /* | |
ea1754a0 KS |
3754 | * delalloc_end is already one less than the total length, so |
3755 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
3756 | */ |
3757 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 3758 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
3759 | delalloc_start = delalloc_end + 1; |
3760 | } | |
3761 | if (wbc->nr_to_write < delalloc_to_write) { | |
3762 | int thresh = 8192; | |
3763 | ||
3764 | if (delalloc_to_write < thresh * 2) | |
3765 | thresh = delalloc_to_write; | |
3766 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
3767 | thresh); | |
3768 | } | |
3769 | ||
3770 | /* did the fill delalloc function already unlock and start | |
3771 | * the IO? | |
3772 | */ | |
3773 | if (page_started) { | |
3774 | /* | |
3775 | * we've unlocked the page, so we can't update | |
3776 | * the mapping's writeback index, just update | |
3777 | * nr_to_write. | |
3778 | */ | |
3779 | wbc->nr_to_write -= *nr_written; | |
3780 | return 1; | |
3781 | } | |
3782 | ||
b69d1ee9 | 3783 | return 0; |
40f76580 CM |
3784 | } |
3785 | ||
3786 | /* | |
3787 | * helper for __extent_writepage. This calls the writepage start hooks, | |
3788 | * and does the loop to map the page into extents and bios. | |
3789 | * | |
3790 | * We return 1 if the IO is started and the page is unlocked, | |
3791 | * 0 if all went well (page still locked) | |
3792 | * < 0 if there were errors (page still locked) | |
3793 | */ | |
d4580fe2 | 3794 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
3795 | struct page *page, |
3796 | struct writeback_control *wbc, | |
3797 | struct extent_page_data *epd, | |
3798 | loff_t i_size, | |
3799 | unsigned long nr_written, | |
57e5ffeb | 3800 | int *nr_ret) |
d1310b2e | 3801 | { |
6bc5636a | 3802 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
d4580fe2 | 3803 | struct extent_io_tree *tree = &inode->io_tree; |
4eee4fa4 | 3804 | u64 start = page_offset(page); |
6bc5636a | 3805 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
3806 | u64 cur = start; |
3807 | u64 extent_offset; | |
d1310b2e | 3808 | u64 block_start; |
d1310b2e | 3809 | struct extent_map *em; |
40f76580 CM |
3810 | int ret = 0; |
3811 | int nr = 0; | |
d8e3fb10 | 3812 | u32 opf = REQ_OP_WRITE; |
57e5ffeb | 3813 | const unsigned int write_flags = wbc_to_write_flags(wbc); |
40f76580 | 3814 | bool compressed; |
c8b97818 | 3815 | |
6bc5636a | 3816 | ret = btrfs_writepage_cow_fixup(page, start, end); |
d75855b4 NB |
3817 | if (ret) { |
3818 | /* Fixup worker will requeue */ | |
5ab58055 | 3819 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
3820 | update_nr_written(wbc, nr_written); |
3821 | unlock_page(page); | |
3822 | return 1; | |
247e743c CM |
3823 | } |
3824 | ||
11c8349b CM |
3825 | /* |
3826 | * we don't want to touch the inode after unlocking the page, | |
3827 | * so we update the mapping writeback index now | |
3828 | */ | |
3d4b9496 | 3829 | update_nr_written(wbc, nr_written + 1); |
771ed689 | 3830 | |
d1310b2e | 3831 | while (cur <= end) { |
0c64c33c | 3832 | u64 disk_bytenr; |
40f76580 | 3833 | u64 em_end; |
6bc5636a | 3834 | u32 iosize; |
58409edd | 3835 | |
40f76580 | 3836 | if (cur >= i_size) { |
38a39ac7 QW |
3837 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
3838 | end, 1); | |
d1310b2e CM |
3839 | break; |
3840 | } | |
d4580fe2 | 3841 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c704005d | 3842 | if (IS_ERR_OR_NULL(em)) { |
d1310b2e | 3843 | SetPageError(page); |
61391d56 | 3844 | ret = PTR_ERR_OR_ZERO(em); |
d1310b2e CM |
3845 | break; |
3846 | } | |
3847 | ||
3848 | extent_offset = cur - em->start; | |
40f76580 | 3849 | em_end = extent_map_end(em); |
6bc5636a QW |
3850 | ASSERT(cur <= em_end); |
3851 | ASSERT(cur < end); | |
3852 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
3853 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 3854 | block_start = em->block_start; |
c8b97818 | 3855 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
3856 | disk_bytenr = em->block_start + extent_offset; |
3857 | ||
3858 | /* Note that em_end from extent_map_end() is exclusive */ | |
3859 | iosize = min(em_end, end + 1) - cur; | |
d8e3fb10 | 3860 | |
e380adfc | 3861 | if (btrfs_use_zone_append(inode, em->block_start)) |
d8e3fb10 NA |
3862 | opf = REQ_OP_ZONE_APPEND; |
3863 | ||
d1310b2e CM |
3864 | free_extent_map(em); |
3865 | em = NULL; | |
3866 | ||
c8b97818 CM |
3867 | /* |
3868 | * compressed and inline extents are written through other | |
3869 | * paths in the FS | |
3870 | */ | |
3871 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 3872 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 3873 | if (compressed) |
c8b97818 | 3874 | nr++; |
c8b04030 | 3875 | else |
38a39ac7 QW |
3876 | btrfs_writepage_endio_finish_ordered(inode, |
3877 | page, cur, cur + iosize - 1, 1); | |
c8b97818 | 3878 | cur += iosize; |
d1310b2e CM |
3879 | continue; |
3880 | } | |
c8b97818 | 3881 | |
5cdc84bf | 3882 | btrfs_set_range_writeback(tree, cur, cur + iosize - 1); |
58409edd | 3883 | if (!PageWriteback(page)) { |
d4580fe2 | 3884 | btrfs_err(inode->root->fs_info, |
58409edd DS |
3885 | "page %lu not writeback, cur %llu end %llu", |
3886 | page->index, cur, end); | |
d1310b2e | 3887 | } |
7f3c74fb | 3888 | |
390ed29b QW |
3889 | ret = submit_extent_page(opf | write_flags, wbc, |
3890 | &epd->bio_ctrl, page, | |
d8e3fb10 | 3891 | disk_bytenr, iosize, |
390ed29b | 3892 | cur - page_offset(page), |
58409edd | 3893 | end_bio_extent_writepage, |
390ed29b | 3894 | 0, 0, false); |
fe01aa65 | 3895 | if (ret) { |
58409edd | 3896 | SetPageError(page); |
fe01aa65 TK |
3897 | if (PageWriteback(page)) |
3898 | end_page_writeback(page); | |
3899 | } | |
d1310b2e | 3900 | |
6bc5636a | 3901 | cur += iosize; |
d1310b2e CM |
3902 | nr++; |
3903 | } | |
40f76580 | 3904 | *nr_ret = nr; |
40f76580 CM |
3905 | return ret; |
3906 | } | |
3907 | ||
3908 | /* | |
3909 | * the writepage semantics are similar to regular writepage. extent | |
3910 | * records are inserted to lock ranges in the tree, and as dirty areas | |
3911 | * are found, they are marked writeback. Then the lock bits are removed | |
3912 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
3913 | * |
3914 | * Return 0 if everything goes well. | |
3915 | * Return <0 for error. | |
40f76580 CM |
3916 | */ |
3917 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
aab6e9ed | 3918 | struct extent_page_data *epd) |
40f76580 CM |
3919 | { |
3920 | struct inode *inode = page->mapping->host; | |
40f76580 | 3921 | u64 start = page_offset(page); |
09cbfeaf | 3922 | u64 page_end = start + PAGE_SIZE - 1; |
40f76580 CM |
3923 | int ret; |
3924 | int nr = 0; | |
eb70d222 | 3925 | size_t pg_offset; |
40f76580 | 3926 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 3927 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 CM |
3928 | unsigned long nr_written = 0; |
3929 | ||
40f76580 CM |
3930 | trace___extent_writepage(page, inode, wbc); |
3931 | ||
3932 | WARN_ON(!PageLocked(page)); | |
3933 | ||
3934 | ClearPageError(page); | |
3935 | ||
7073017a | 3936 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
3937 | if (page->index > end_index || |
3938 | (page->index == end_index && !pg_offset)) { | |
09cbfeaf | 3939 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
40f76580 CM |
3940 | unlock_page(page); |
3941 | return 0; | |
3942 | } | |
3943 | ||
3944 | if (page->index == end_index) { | |
d048b9c2 | 3945 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 CM |
3946 | flush_dcache_page(page); |
3947 | } | |
3948 | ||
32443de3 QW |
3949 | ret = set_page_extent_mapped(page); |
3950 | if (ret < 0) { | |
3951 | SetPageError(page); | |
3952 | goto done; | |
3953 | } | |
40f76580 | 3954 | |
7789a55a | 3955 | if (!epd->extent_locked) { |
cd4c0bf9 NB |
3956 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc, start, |
3957 | &nr_written); | |
7789a55a | 3958 | if (ret == 1) |
169d2c87 | 3959 | return 0; |
7789a55a NB |
3960 | if (ret) |
3961 | goto done; | |
3962 | } | |
40f76580 | 3963 | |
d4580fe2 NB |
3964 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, epd, i_size, |
3965 | nr_written, &nr); | |
40f76580 | 3966 | if (ret == 1) |
169d2c87 | 3967 | return 0; |
40f76580 | 3968 | |
d1310b2e CM |
3969 | done: |
3970 | if (nr == 0) { | |
3971 | /* make sure the mapping tag for page dirty gets cleared */ | |
3972 | set_page_writeback(page); | |
3973 | end_page_writeback(page); | |
3974 | } | |
61391d56 FM |
3975 | if (PageError(page)) { |
3976 | ret = ret < 0 ? ret : -EIO; | |
3977 | end_extent_writepage(page, ret, start, page_end); | |
3978 | } | |
d1310b2e | 3979 | unlock_page(page); |
3065976b | 3980 | ASSERT(ret <= 0); |
40f76580 | 3981 | return ret; |
d1310b2e CM |
3982 | } |
3983 | ||
fd8b2b61 | 3984 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 3985 | { |
74316201 N |
3986 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
3987 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
3988 | } |
3989 | ||
18dfa711 FM |
3990 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
3991 | { | |
3992 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
3993 | smp_mb__after_atomic(); | |
3994 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
3995 | } | |
3996 | ||
2e3c2513 | 3997 | /* |
a3efb2f0 | 3998 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 3999 | * |
a3efb2f0 QW |
4000 | * May try to flush write bio if we can't get the lock. |
4001 | * | |
4002 | * Return 0 if the extent buffer doesn't need to be submitted. | |
4003 | * (E.g. the extent buffer is not dirty) | |
4004 | * Return >0 is the extent buffer is submitted to bio. | |
4005 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 4006 | */ |
9df76fb5 | 4007 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
0e378df1 | 4008 | struct extent_page_data *epd) |
0b32f4bb | 4009 | { |
9df76fb5 | 4010 | struct btrfs_fs_info *fs_info = eb->fs_info; |
2e3c2513 | 4011 | int i, num_pages, failed_page_nr; |
0b32f4bb JB |
4012 | int flush = 0; |
4013 | int ret = 0; | |
4014 | ||
4015 | if (!btrfs_try_tree_write_lock(eb)) { | |
f4340622 | 4016 | ret = flush_write_bio(epd); |
2e3c2513 QW |
4017 | if (ret < 0) |
4018 | return ret; | |
4019 | flush = 1; | |
0b32f4bb JB |
4020 | btrfs_tree_lock(eb); |
4021 | } | |
4022 | ||
4023 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
4024 | btrfs_tree_unlock(eb); | |
4025 | if (!epd->sync_io) | |
4026 | return 0; | |
4027 | if (!flush) { | |
f4340622 | 4028 | ret = flush_write_bio(epd); |
2e3c2513 QW |
4029 | if (ret < 0) |
4030 | return ret; | |
0b32f4bb JB |
4031 | flush = 1; |
4032 | } | |
a098d8e8 CM |
4033 | while (1) { |
4034 | wait_on_extent_buffer_writeback(eb); | |
4035 | btrfs_tree_lock(eb); | |
4036 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
4037 | break; | |
0b32f4bb | 4038 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
4039 | } |
4040 | } | |
4041 | ||
51561ffe JB |
4042 | /* |
4043 | * We need to do this to prevent races in people who check if the eb is | |
4044 | * under IO since we can end up having no IO bits set for a short period | |
4045 | * of time. | |
4046 | */ | |
4047 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
4048 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
4049 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 4050 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 4051 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
4052 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4053 | -eb->len, | |
4054 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 4055 | ret = 1; |
51561ffe JB |
4056 | } else { |
4057 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
4058 | } |
4059 | ||
4060 | btrfs_tree_unlock(eb); | |
4061 | ||
f3156df9 QW |
4062 | /* |
4063 | * Either we don't need to submit any tree block, or we're submitting | |
4064 | * subpage eb. | |
4065 | * Subpage metadata doesn't use page locking at all, so we can skip | |
4066 | * the page locking. | |
4067 | */ | |
4068 | if (!ret || fs_info->sectorsize < PAGE_SIZE) | |
0b32f4bb JB |
4069 | return ret; |
4070 | ||
65ad0104 | 4071 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4072 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4073 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4074 | |
4075 | if (!trylock_page(p)) { | |
4076 | if (!flush) { | |
18dfa711 FM |
4077 | int err; |
4078 | ||
4079 | err = flush_write_bio(epd); | |
4080 | if (err < 0) { | |
4081 | ret = err; | |
2e3c2513 QW |
4082 | failed_page_nr = i; |
4083 | goto err_unlock; | |
4084 | } | |
0b32f4bb JB |
4085 | flush = 1; |
4086 | } | |
4087 | lock_page(p); | |
4088 | } | |
4089 | } | |
4090 | ||
4091 | return ret; | |
2e3c2513 QW |
4092 | err_unlock: |
4093 | /* Unlock already locked pages */ | |
4094 | for (i = 0; i < failed_page_nr; i++) | |
4095 | unlock_page(eb->pages[i]); | |
18dfa711 FM |
4096 | /* |
4097 | * Clear EXTENT_BUFFER_WRITEBACK and wake up anyone waiting on it. | |
4098 | * Also set back EXTENT_BUFFER_DIRTY so future attempts to this eb can | |
4099 | * be made and undo everything done before. | |
4100 | */ | |
4101 | btrfs_tree_lock(eb); | |
4102 | spin_lock(&eb->refs_lock); | |
4103 | set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); | |
4104 | end_extent_buffer_writeback(eb); | |
4105 | spin_unlock(&eb->refs_lock); | |
4106 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, eb->len, | |
4107 | fs_info->dirty_metadata_batch); | |
4108 | btrfs_clear_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); | |
4109 | btrfs_tree_unlock(eb); | |
2e3c2513 | 4110 | return ret; |
0b32f4bb JB |
4111 | } |
4112 | ||
5a2c6075 | 4113 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 4114 | { |
5a2c6075 | 4115 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 4116 | |
5a2c6075 | 4117 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
4118 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
4119 | return; | |
4120 | ||
eb5b64f1 DZ |
4121 | /* |
4122 | * If we error out, we should add back the dirty_metadata_bytes | |
4123 | * to make it consistent. | |
4124 | */ | |
eb5b64f1 DZ |
4125 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
4126 | eb->len, fs_info->dirty_metadata_batch); | |
4127 | ||
656f30db FM |
4128 | /* |
4129 | * If writeback for a btree extent that doesn't belong to a log tree | |
4130 | * failed, increment the counter transaction->eb_write_errors. | |
4131 | * We do this because while the transaction is running and before it's | |
4132 | * committing (when we call filemap_fdata[write|wait]_range against | |
4133 | * the btree inode), we might have | |
4134 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
4135 | * returns an error or an error happens during writeback, when we're | |
4136 | * committing the transaction we wouldn't know about it, since the pages | |
4137 | * can be no longer dirty nor marked anymore for writeback (if a | |
4138 | * subsequent modification to the extent buffer didn't happen before the | |
4139 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
4140 | * able to find the pages tagged with SetPageError at transaction | |
4141 | * commit time. So if this happens we must abort the transaction, | |
4142 | * otherwise we commit a super block with btree roots that point to | |
4143 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
4144 | * or the content of some node/leaf from a past generation that got | |
4145 | * cowed or deleted and is no longer valid. | |
4146 | * | |
4147 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
4148 | * not be enough - we need to distinguish between log tree extents vs | |
4149 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
4150 | * will catch and clear such errors in the mapping - and that call might | |
4151 | * be from a log sync and not from a transaction commit. Also, checking | |
4152 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
4153 | * not done and would not be reliable - the eb might have been released | |
4154 | * from memory and reading it back again means that flag would not be | |
4155 | * set (since it's a runtime flag, not persisted on disk). | |
4156 | * | |
4157 | * Using the flags below in the btree inode also makes us achieve the | |
4158 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
4159 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
4160 | * is called, the writeback for all dirty pages had already finished | |
4161 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
4162 | * filemap_fdatawait_range() would return success, as it could not know | |
4163 | * that writeback errors happened (the pages were no longer tagged for | |
4164 | * writeback). | |
4165 | */ | |
4166 | switch (eb->log_index) { | |
4167 | case -1: | |
5a2c6075 | 4168 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
4169 | break; |
4170 | case 0: | |
5a2c6075 | 4171 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
4172 | break; |
4173 | case 1: | |
5a2c6075 | 4174 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
4175 | break; |
4176 | default: | |
4177 | BUG(); /* unexpected, logic error */ | |
4178 | } | |
4179 | } | |
4180 | ||
2f3186d8 QW |
4181 | /* |
4182 | * The endio specific version which won't touch any unsafe spinlock in endio | |
4183 | * context. | |
4184 | */ | |
4185 | static struct extent_buffer *find_extent_buffer_nolock( | |
4186 | struct btrfs_fs_info *fs_info, u64 start) | |
4187 | { | |
4188 | struct extent_buffer *eb; | |
4189 | ||
4190 | rcu_read_lock(); | |
4191 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
4192 | start >> fs_info->sectorsize_bits); | |
4193 | if (eb && atomic_inc_not_zero(&eb->refs)) { | |
4194 | rcu_read_unlock(); | |
4195 | return eb; | |
4196 | } | |
4197 | rcu_read_unlock(); | |
4198 | return NULL; | |
4199 | } | |
4200 | ||
4201 | /* | |
4202 | * The endio function for subpage extent buffer write. | |
4203 | * | |
4204 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
4205 | * after all extent buffers in the page has finished their writeback. | |
4206 | */ | |
fa04c165 | 4207 | static void end_bio_subpage_eb_writepage(struct bio *bio) |
2f3186d8 | 4208 | { |
fa04c165 | 4209 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
4210 | struct bio_vec *bvec; |
4211 | struct bvec_iter_all iter_all; | |
4212 | ||
fa04c165 QW |
4213 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
4214 | ASSERT(fs_info->sectorsize < PAGE_SIZE); | |
4215 | ||
2f3186d8 QW |
4216 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
4217 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
4218 | struct page *page = bvec->bv_page; | |
4219 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
4220 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
4221 | u64 cur_bytenr = bvec_start; | |
4222 | ||
4223 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
4224 | ||
4225 | /* Iterate through all extent buffers in the range */ | |
4226 | while (cur_bytenr <= bvec_end) { | |
4227 | struct extent_buffer *eb; | |
4228 | int done; | |
4229 | ||
4230 | /* | |
4231 | * Here we can't use find_extent_buffer(), as it may | |
4232 | * try to lock eb->refs_lock, which is not safe in endio | |
4233 | * context. | |
4234 | */ | |
4235 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
4236 | ASSERT(eb); | |
4237 | ||
4238 | cur_bytenr = eb->start + eb->len; | |
4239 | ||
4240 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
4241 | done = atomic_dec_and_test(&eb->io_pages); | |
4242 | ASSERT(done); | |
4243 | ||
4244 | if (bio->bi_status || | |
4245 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
4246 | ClearPageUptodate(page); | |
4247 | set_btree_ioerr(page, eb); | |
4248 | } | |
4249 | ||
4250 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
4251 | eb->len); | |
4252 | end_extent_buffer_writeback(eb); | |
4253 | /* | |
4254 | * free_extent_buffer() will grab spinlock which is not | |
4255 | * safe in endio context. Thus here we manually dec | |
4256 | * the ref. | |
4257 | */ | |
4258 | atomic_dec(&eb->refs); | |
4259 | } | |
4260 | } | |
4261 | bio_put(bio); | |
4262 | } | |
4263 | ||
4246a0b6 | 4264 | static void end_bio_extent_buffer_writepage(struct bio *bio) |
0b32f4bb | 4265 | { |
2c30c71b | 4266 | struct bio_vec *bvec; |
0b32f4bb | 4267 | struct extent_buffer *eb; |
2b070cfe | 4268 | int done; |
6dc4f100 | 4269 | struct bvec_iter_all iter_all; |
0b32f4bb | 4270 | |
c09abff8 | 4271 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 4272 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
4273 | struct page *page = bvec->bv_page; |
4274 | ||
0b32f4bb JB |
4275 | eb = (struct extent_buffer *)page->private; |
4276 | BUG_ON(!eb); | |
4277 | done = atomic_dec_and_test(&eb->io_pages); | |
4278 | ||
4e4cbee9 | 4279 | if (bio->bi_status || |
4246a0b6 | 4280 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 4281 | ClearPageUptodate(page); |
5a2c6075 | 4282 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
4283 | } |
4284 | ||
4285 | end_page_writeback(page); | |
4286 | ||
4287 | if (!done) | |
4288 | continue; | |
4289 | ||
4290 | end_extent_buffer_writeback(eb); | |
2c30c71b | 4291 | } |
0b32f4bb JB |
4292 | |
4293 | bio_put(bio); | |
0b32f4bb JB |
4294 | } |
4295 | ||
fa04c165 QW |
4296 | static void prepare_eb_write(struct extent_buffer *eb) |
4297 | { | |
4298 | u32 nritems; | |
4299 | unsigned long start; | |
4300 | unsigned long end; | |
4301 | ||
4302 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
4303 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
4304 | ||
4305 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
4306 | nritems = btrfs_header_nritems(eb); | |
4307 | if (btrfs_header_level(eb) > 0) { | |
4308 | end = btrfs_node_key_ptr_offset(nritems); | |
4309 | memzero_extent_buffer(eb, end, eb->len - end); | |
4310 | } else { | |
4311 | /* | |
4312 | * Leaf: | |
4313 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
4314 | */ | |
4315 | start = btrfs_item_nr_offset(nritems); | |
4316 | end = BTRFS_LEAF_DATA_OFFSET + leaf_data_end(eb); | |
4317 | memzero_extent_buffer(eb, start, end - start); | |
4318 | } | |
4319 | } | |
4320 | ||
35b6ddfa QW |
4321 | /* |
4322 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
4323 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
4324 | */ |
4325 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
4326 | struct writeback_control *wbc, | |
4327 | struct extent_page_data *epd) | |
4328 | { | |
4329 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4330 | struct page *page = eb->pages[0]; | |
4331 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; | |
4332 | bool no_dirty_ebs = false; | |
4333 | int ret; | |
4334 | ||
fa04c165 QW |
4335 | prepare_eb_write(eb); |
4336 | ||
35b6ddfa QW |
4337 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
4338 | lock_page(page); | |
4339 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
4340 | ||
4341 | /* Check if this is the last dirty bit to update nr_written */ | |
4342 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
4343 | eb->start, eb->len); | |
4344 | if (no_dirty_ebs) | |
4345 | clear_page_dirty_for_io(page); | |
4346 | ||
390ed29b QW |
4347 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
4348 | &epd->bio_ctrl, page, eb->start, eb->len, | |
4349 | eb->start - page_offset(page), | |
fa04c165 | 4350 | end_bio_subpage_eb_writepage, 0, 0, false); |
35b6ddfa QW |
4351 | if (ret) { |
4352 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
4353 | set_btree_ioerr(page, eb); | |
4354 | unlock_page(page); | |
4355 | ||
4356 | if (atomic_dec_and_test(&eb->io_pages)) | |
4357 | end_extent_buffer_writeback(eb); | |
4358 | return -EIO; | |
4359 | } | |
4360 | unlock_page(page); | |
4361 | /* | |
4362 | * Submission finished without problem, if no range of the page is | |
4363 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
4364 | */ | |
4365 | if (no_dirty_ebs) | |
4366 | update_nr_written(wbc, 1); | |
4367 | return ret; | |
4368 | } | |
4369 | ||
0e378df1 | 4370 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb JB |
4371 | struct writeback_control *wbc, |
4372 | struct extent_page_data *epd) | |
4373 | { | |
0c64c33c | 4374 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 4375 | int i, num_pages; |
ff40adf7 | 4376 | unsigned int write_flags = wbc_to_write_flags(wbc) | REQ_META; |
d7dbe9e7 | 4377 | int ret = 0; |
0b32f4bb | 4378 | |
fa04c165 | 4379 | prepare_eb_write(eb); |
35b6ddfa | 4380 | |
fa04c165 | 4381 | num_pages = num_extent_pages(eb); |
0b32f4bb | 4382 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4383 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
4384 | |
4385 | clear_page_dirty_for_io(p); | |
4386 | set_page_writeback(p); | |
0ceb34bf | 4387 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
390ed29b QW |
4388 | &epd->bio_ctrl, p, disk_bytenr, |
4389 | PAGE_SIZE, 0, | |
1f7ad75b | 4390 | end_bio_extent_buffer_writepage, |
390ed29b | 4391 | 0, 0, false); |
0b32f4bb | 4392 | if (ret) { |
5a2c6075 | 4393 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
4394 | if (PageWriteback(p)) |
4395 | end_page_writeback(p); | |
0b32f4bb JB |
4396 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
4397 | end_extent_buffer_writeback(eb); | |
4398 | ret = -EIO; | |
4399 | break; | |
4400 | } | |
0c64c33c | 4401 | disk_bytenr += PAGE_SIZE; |
3d4b9496 | 4402 | update_nr_written(wbc, 1); |
0b32f4bb JB |
4403 | unlock_page(p); |
4404 | } | |
4405 | ||
4406 | if (unlikely(ret)) { | |
4407 | for (; i < num_pages; i++) { | |
bbf65cf0 | 4408 | struct page *p = eb->pages[i]; |
81465028 | 4409 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
4410 | unlock_page(p); |
4411 | } | |
4412 | } | |
4413 | ||
4414 | return ret; | |
4415 | } | |
4416 | ||
c4aec299 QW |
4417 | /* |
4418 | * Submit one subpage btree page. | |
4419 | * | |
4420 | * The main difference to submit_eb_page() is: | |
4421 | * - Page locking | |
4422 | * For subpage, we don't rely on page locking at all. | |
4423 | * | |
4424 | * - Flush write bio | |
4425 | * We only flush bio if we may be unable to fit current extent buffers into | |
4426 | * current bio. | |
4427 | * | |
4428 | * Return >=0 for the number of submitted extent buffers. | |
4429 | * Return <0 for fatal error. | |
4430 | */ | |
4431 | static int submit_eb_subpage(struct page *page, | |
4432 | struct writeback_control *wbc, | |
4433 | struct extent_page_data *epd) | |
4434 | { | |
4435 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
4436 | int submitted = 0; | |
4437 | u64 page_start = page_offset(page); | |
4438 | int bit_start = 0; | |
4439 | const int nbits = BTRFS_SUBPAGE_BITMAP_SIZE; | |
4440 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; | |
4441 | int ret; | |
4442 | ||
4443 | /* Lock and write each dirty extent buffers in the range */ | |
4444 | while (bit_start < nbits) { | |
4445 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
4446 | struct extent_buffer *eb; | |
4447 | unsigned long flags; | |
4448 | u64 start; | |
4449 | ||
4450 | /* | |
4451 | * Take private lock to ensure the subpage won't be detached | |
4452 | * in the meantime. | |
4453 | */ | |
4454 | spin_lock(&page->mapping->private_lock); | |
4455 | if (!PagePrivate(page)) { | |
4456 | spin_unlock(&page->mapping->private_lock); | |
4457 | break; | |
4458 | } | |
4459 | spin_lock_irqsave(&subpage->lock, flags); | |
4460 | if (!((1 << bit_start) & subpage->dirty_bitmap)) { | |
4461 | spin_unlock_irqrestore(&subpage->lock, flags); | |
4462 | spin_unlock(&page->mapping->private_lock); | |
4463 | bit_start++; | |
4464 | continue; | |
4465 | } | |
4466 | ||
4467 | start = page_start + bit_start * fs_info->sectorsize; | |
4468 | bit_start += sectors_per_node; | |
4469 | ||
4470 | /* | |
4471 | * Here we just want to grab the eb without touching extra | |
4472 | * spin locks, so call find_extent_buffer_nolock(). | |
4473 | */ | |
4474 | eb = find_extent_buffer_nolock(fs_info, start); | |
4475 | spin_unlock_irqrestore(&subpage->lock, flags); | |
4476 | spin_unlock(&page->mapping->private_lock); | |
4477 | ||
4478 | /* | |
4479 | * The eb has already reached 0 refs thus find_extent_buffer() | |
4480 | * doesn't return it. We don't need to write back such eb | |
4481 | * anyway. | |
4482 | */ | |
4483 | if (!eb) | |
4484 | continue; | |
4485 | ||
4486 | ret = lock_extent_buffer_for_io(eb, epd); | |
4487 | if (ret == 0) { | |
4488 | free_extent_buffer(eb); | |
4489 | continue; | |
4490 | } | |
4491 | if (ret < 0) { | |
4492 | free_extent_buffer(eb); | |
4493 | goto cleanup; | |
4494 | } | |
fa04c165 | 4495 | ret = write_one_subpage_eb(eb, wbc, epd); |
c4aec299 QW |
4496 | free_extent_buffer(eb); |
4497 | if (ret < 0) | |
4498 | goto cleanup; | |
4499 | submitted++; | |
4500 | } | |
4501 | return submitted; | |
4502 | ||
4503 | cleanup: | |
4504 | /* We hit error, end bio for the submitted extent buffers */ | |
4505 | end_write_bio(epd, ret); | |
4506 | return ret; | |
4507 | } | |
4508 | ||
f91e0d0c QW |
4509 | /* |
4510 | * Submit all page(s) of one extent buffer. | |
4511 | * | |
4512 | * @page: the page of one extent buffer | |
4513 | * @eb_context: to determine if we need to submit this page, if current page | |
4514 | * belongs to this eb, we don't need to submit | |
4515 | * | |
4516 | * The caller should pass each page in their bytenr order, and here we use | |
4517 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
4518 | * | |
4519 | * If we have, we just skip until we hit a new page that doesn't belong to | |
4520 | * current @eb_context. | |
4521 | * | |
4522 | * If not, we submit all the page(s) of the extent buffer. | |
4523 | * | |
4524 | * Return >0 if we have submitted the extent buffer successfully. | |
4525 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
4526 | * previous call. | |
4527 | * Return <0 for fatal error. | |
4528 | */ | |
4529 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
4530 | struct extent_page_data *epd, | |
4531 | struct extent_buffer **eb_context) | |
4532 | { | |
4533 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 4534 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
4535 | struct extent_buffer *eb; |
4536 | int ret; | |
4537 | ||
4538 | if (!PagePrivate(page)) | |
4539 | return 0; | |
4540 | ||
c4aec299 QW |
4541 | if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) |
4542 | return submit_eb_subpage(page, wbc, epd); | |
4543 | ||
f91e0d0c QW |
4544 | spin_lock(&mapping->private_lock); |
4545 | if (!PagePrivate(page)) { | |
4546 | spin_unlock(&mapping->private_lock); | |
4547 | return 0; | |
4548 | } | |
4549 | ||
4550 | eb = (struct extent_buffer *)page->private; | |
4551 | ||
4552 | /* | |
4553 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
4554 | * crashing the machine for something we can survive anyway. | |
4555 | */ | |
4556 | if (WARN_ON(!eb)) { | |
4557 | spin_unlock(&mapping->private_lock); | |
4558 | return 0; | |
4559 | } | |
4560 | ||
4561 | if (eb == *eb_context) { | |
4562 | spin_unlock(&mapping->private_lock); | |
4563 | return 0; | |
4564 | } | |
4565 | ret = atomic_inc_not_zero(&eb->refs); | |
4566 | spin_unlock(&mapping->private_lock); | |
4567 | if (!ret) | |
4568 | return 0; | |
4569 | ||
0bc09ca1 NA |
4570 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
4571 | /* | |
4572 | * If for_sync, this hole will be filled with | |
4573 | * trasnsaction commit. | |
4574 | */ | |
4575 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
4576 | ret = -EAGAIN; | |
4577 | else | |
4578 | ret = 0; | |
4579 | free_extent_buffer(eb); | |
4580 | return ret; | |
4581 | } | |
4582 | ||
f91e0d0c QW |
4583 | *eb_context = eb; |
4584 | ||
4585 | ret = lock_extent_buffer_for_io(eb, epd); | |
4586 | if (ret <= 0) { | |
0bc09ca1 NA |
4587 | btrfs_revert_meta_write_pointer(cache, eb); |
4588 | if (cache) | |
4589 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
4590 | free_extent_buffer(eb); |
4591 | return ret; | |
4592 | } | |
0bc09ca1 NA |
4593 | if (cache) |
4594 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
4595 | ret = write_one_eb(eb, wbc, epd); |
4596 | free_extent_buffer(eb); | |
4597 | if (ret < 0) | |
4598 | return ret; | |
4599 | return 1; | |
4600 | } | |
4601 | ||
0b32f4bb JB |
4602 | int btree_write_cache_pages(struct address_space *mapping, |
4603 | struct writeback_control *wbc) | |
4604 | { | |
f91e0d0c | 4605 | struct extent_buffer *eb_context = NULL; |
0b32f4bb | 4606 | struct extent_page_data epd = { |
390ed29b | 4607 | .bio_ctrl = { 0 }, |
0b32f4bb JB |
4608 | .extent_locked = 0, |
4609 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, | |
4610 | }; | |
b3ff8f1d | 4611 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
4612 | int ret = 0; |
4613 | int done = 0; | |
4614 | int nr_to_write_done = 0; | |
4615 | struct pagevec pvec; | |
4616 | int nr_pages; | |
4617 | pgoff_t index; | |
4618 | pgoff_t end; /* Inclusive */ | |
4619 | int scanned = 0; | |
10bbd235 | 4620 | xa_mark_t tag; |
0b32f4bb | 4621 | |
86679820 | 4622 | pagevec_init(&pvec); |
0b32f4bb JB |
4623 | if (wbc->range_cyclic) { |
4624 | index = mapping->writeback_index; /* Start from prev offset */ | |
4625 | end = -1; | |
556755a8 JB |
4626 | /* |
4627 | * Start from the beginning does not need to cycle over the | |
4628 | * range, mark it as scanned. | |
4629 | */ | |
4630 | scanned = (index == 0); | |
0b32f4bb | 4631 | } else { |
09cbfeaf KS |
4632 | index = wbc->range_start >> PAGE_SHIFT; |
4633 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
4634 | scanned = 1; |
4635 | } | |
4636 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4637 | tag = PAGECACHE_TAG_TOWRITE; | |
4638 | else | |
4639 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 4640 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
4641 | retry: |
4642 | if (wbc->sync_mode == WB_SYNC_ALL) | |
4643 | tag_pages_for_writeback(mapping, index, end); | |
4644 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 4645 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 4646 | tag))) { |
0b32f4bb JB |
4647 | unsigned i; |
4648 | ||
0b32f4bb JB |
4649 | for (i = 0; i < nr_pages; i++) { |
4650 | struct page *page = pvec.pages[i]; | |
4651 | ||
f91e0d0c QW |
4652 | ret = submit_eb_page(page, wbc, &epd, &eb_context); |
4653 | if (ret == 0) | |
0b32f4bb | 4654 | continue; |
f91e0d0c | 4655 | if (ret < 0) { |
0b32f4bb | 4656 | done = 1; |
0b32f4bb JB |
4657 | break; |
4658 | } | |
0b32f4bb JB |
4659 | |
4660 | /* | |
4661 | * the filesystem may choose to bump up nr_to_write. | |
4662 | * We have to make sure to honor the new nr_to_write | |
4663 | * at any time | |
4664 | */ | |
4665 | nr_to_write_done = wbc->nr_to_write <= 0; | |
4666 | } | |
4667 | pagevec_release(&pvec); | |
4668 | cond_resched(); | |
4669 | } | |
4670 | if (!scanned && !done) { | |
4671 | /* | |
4672 | * We hit the last page and there is more work to be done: wrap | |
4673 | * back to the start of the file | |
4674 | */ | |
4675 | scanned = 1; | |
4676 | index = 0; | |
4677 | goto retry; | |
4678 | } | |
2b952eea QW |
4679 | if (ret < 0) { |
4680 | end_write_bio(&epd, ret); | |
0bc09ca1 | 4681 | goto out; |
2b952eea | 4682 | } |
b3ff8f1d QW |
4683 | /* |
4684 | * If something went wrong, don't allow any metadata write bio to be | |
4685 | * submitted. | |
4686 | * | |
4687 | * This would prevent use-after-free if we had dirty pages not | |
4688 | * cleaned up, which can still happen by fuzzed images. | |
4689 | * | |
4690 | * - Bad extent tree | |
4691 | * Allowing existing tree block to be allocated for other trees. | |
4692 | * | |
4693 | * - Log tree operations | |
4694 | * Exiting tree blocks get allocated to log tree, bumps its | |
4695 | * generation, then get cleaned in tree re-balance. | |
4696 | * Such tree block will not be written back, since it's clean, | |
4697 | * thus no WRITTEN flag set. | |
4698 | * And after log writes back, this tree block is not traced by | |
4699 | * any dirty extent_io_tree. | |
4700 | * | |
4701 | * - Offending tree block gets re-dirtied from its original owner | |
4702 | * Since it has bumped generation, no WRITTEN flag, it can be | |
4703 | * reused without COWing. This tree block will not be traced | |
4704 | * by btrfs_transaction::dirty_pages. | |
4705 | * | |
4706 | * Now such dirty tree block will not be cleaned by any dirty | |
4707 | * extent io tree. Thus we don't want to submit such wild eb | |
4708 | * if the fs already has error. | |
4709 | */ | |
4710 | if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { | |
4711 | ret = flush_write_bio(&epd); | |
4712 | } else { | |
fbabd4a3 | 4713 | ret = -EROFS; |
b3ff8f1d QW |
4714 | end_write_bio(&epd, ret); |
4715 | } | |
0bc09ca1 NA |
4716 | out: |
4717 | btrfs_zoned_meta_io_unlock(fs_info); | |
0b32f4bb JB |
4718 | return ret; |
4719 | } | |
4720 | ||
d1310b2e | 4721 | /** |
3bed2da1 NB |
4722 | * Walk the list of dirty pages of the given address space and write all of them. |
4723 | * | |
d1310b2e | 4724 | * @mapping: address space structure to write |
3bed2da1 NB |
4725 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write |
4726 | * @epd: holds context for the write, namely the bio | |
d1310b2e CM |
4727 | * |
4728 | * If a page is already under I/O, write_cache_pages() skips it, even | |
4729 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
4730 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
4731 | * and msync() need to guarantee that all the data which was dirty at the time | |
4732 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
4733 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
4734 | * existing IO to complete. | |
4735 | */ | |
4242b64a | 4736 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 4737 | struct writeback_control *wbc, |
aab6e9ed | 4738 | struct extent_page_data *epd) |
d1310b2e | 4739 | { |
7fd1a3f7 | 4740 | struct inode *inode = mapping->host; |
d1310b2e CM |
4741 | int ret = 0; |
4742 | int done = 0; | |
f85d7d6c | 4743 | int nr_to_write_done = 0; |
d1310b2e CM |
4744 | struct pagevec pvec; |
4745 | int nr_pages; | |
4746 | pgoff_t index; | |
4747 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
4748 | pgoff_t done_index; |
4749 | int range_whole = 0; | |
d1310b2e | 4750 | int scanned = 0; |
10bbd235 | 4751 | xa_mark_t tag; |
d1310b2e | 4752 | |
7fd1a3f7 JB |
4753 | /* |
4754 | * We have to hold onto the inode so that ordered extents can do their | |
4755 | * work when the IO finishes. The alternative to this is failing to add | |
4756 | * an ordered extent if the igrab() fails there and that is a huge pain | |
4757 | * to deal with, so instead just hold onto the inode throughout the | |
4758 | * writepages operation. If it fails here we are freeing up the inode | |
4759 | * anyway and we'd rather not waste our time writing out stuff that is | |
4760 | * going to be truncated anyway. | |
4761 | */ | |
4762 | if (!igrab(inode)) | |
4763 | return 0; | |
4764 | ||
86679820 | 4765 | pagevec_init(&pvec); |
d1310b2e CM |
4766 | if (wbc->range_cyclic) { |
4767 | index = mapping->writeback_index; /* Start from prev offset */ | |
4768 | end = -1; | |
556755a8 JB |
4769 | /* |
4770 | * Start from the beginning does not need to cycle over the | |
4771 | * range, mark it as scanned. | |
4772 | */ | |
4773 | scanned = (index == 0); | |
d1310b2e | 4774 | } else { |
09cbfeaf KS |
4775 | index = wbc->range_start >> PAGE_SHIFT; |
4776 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
4777 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
4778 | range_whole = 1; | |
d1310b2e CM |
4779 | scanned = 1; |
4780 | } | |
3cd24c69 EL |
4781 | |
4782 | /* | |
4783 | * We do the tagged writepage as long as the snapshot flush bit is set | |
4784 | * and we are the first one who do the filemap_flush() on this inode. | |
4785 | * | |
4786 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
4787 | * not race in and drop the bit. | |
4788 | */ | |
4789 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
4790 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
4791 | &BTRFS_I(inode)->runtime_flags)) | |
4792 | wbc->tagged_writepages = 1; | |
4793 | ||
4794 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
4795 | tag = PAGECACHE_TAG_TOWRITE; |
4796 | else | |
4797 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 4798 | retry: |
3cd24c69 | 4799 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 4800 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 4801 | done_index = index; |
f85d7d6c | 4802 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
4803 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
4804 | &index, end, tag))) { | |
d1310b2e CM |
4805 | unsigned i; |
4806 | ||
d1310b2e CM |
4807 | for (i = 0; i < nr_pages; i++) { |
4808 | struct page *page = pvec.pages[i]; | |
4809 | ||
f7bddf1e | 4810 | done_index = page->index + 1; |
d1310b2e | 4811 | /* |
b93b0163 MW |
4812 | * At this point we hold neither the i_pages lock nor |
4813 | * the page lock: the page may be truncated or | |
4814 | * invalidated (changing page->mapping to NULL), | |
4815 | * or even swizzled back from swapper_space to | |
4816 | * tmpfs file mapping | |
d1310b2e | 4817 | */ |
c8f2f24b | 4818 | if (!trylock_page(page)) { |
f4340622 QW |
4819 | ret = flush_write_bio(epd); |
4820 | BUG_ON(ret < 0); | |
c8f2f24b | 4821 | lock_page(page); |
01d658f2 | 4822 | } |
d1310b2e CM |
4823 | |
4824 | if (unlikely(page->mapping != mapping)) { | |
4825 | unlock_page(page); | |
4826 | continue; | |
4827 | } | |
4828 | ||
d2c3f4f6 | 4829 | if (wbc->sync_mode != WB_SYNC_NONE) { |
f4340622 QW |
4830 | if (PageWriteback(page)) { |
4831 | ret = flush_write_bio(epd); | |
4832 | BUG_ON(ret < 0); | |
4833 | } | |
d1310b2e | 4834 | wait_on_page_writeback(page); |
d2c3f4f6 | 4835 | } |
d1310b2e CM |
4836 | |
4837 | if (PageWriteback(page) || | |
4838 | !clear_page_dirty_for_io(page)) { | |
4839 | unlock_page(page); | |
4840 | continue; | |
4841 | } | |
4842 | ||
aab6e9ed | 4843 | ret = __extent_writepage(page, wbc, epd); |
a9132667 | 4844 | if (ret < 0) { |
a9132667 LB |
4845 | done = 1; |
4846 | break; | |
4847 | } | |
f85d7d6c CM |
4848 | |
4849 | /* | |
4850 | * the filesystem may choose to bump up nr_to_write. | |
4851 | * We have to make sure to honor the new nr_to_write | |
4852 | * at any time | |
4853 | */ | |
4854 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
4855 | } |
4856 | pagevec_release(&pvec); | |
4857 | cond_resched(); | |
4858 | } | |
894b36e3 | 4859 | if (!scanned && !done) { |
d1310b2e CM |
4860 | /* |
4861 | * We hit the last page and there is more work to be done: wrap | |
4862 | * back to the start of the file | |
4863 | */ | |
4864 | scanned = 1; | |
4865 | index = 0; | |
42ffb0bf JB |
4866 | |
4867 | /* | |
4868 | * If we're looping we could run into a page that is locked by a | |
4869 | * writer and that writer could be waiting on writeback for a | |
4870 | * page in our current bio, and thus deadlock, so flush the | |
4871 | * write bio here. | |
4872 | */ | |
4873 | ret = flush_write_bio(epd); | |
4874 | if (!ret) | |
4875 | goto retry; | |
d1310b2e | 4876 | } |
a9132667 LB |
4877 | |
4878 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
4879 | mapping->writeback_index = done_index; | |
4880 | ||
7fd1a3f7 | 4881 | btrfs_add_delayed_iput(inode); |
894b36e3 | 4882 | return ret; |
d1310b2e | 4883 | } |
d1310b2e | 4884 | |
0a9b0e53 | 4885 | int extent_write_full_page(struct page *page, struct writeback_control *wbc) |
d1310b2e CM |
4886 | { |
4887 | int ret; | |
d1310b2e | 4888 | struct extent_page_data epd = { |
390ed29b | 4889 | .bio_ctrl = { 0 }, |
771ed689 | 4890 | .extent_locked = 0, |
ffbd517d | 4891 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e | 4892 | }; |
d1310b2e | 4893 | |
d1310b2e | 4894 | ret = __extent_writepage(page, wbc, &epd); |
3065976b QW |
4895 | ASSERT(ret <= 0); |
4896 | if (ret < 0) { | |
4897 | end_write_bio(&epd, ret); | |
4898 | return ret; | |
4899 | } | |
d1310b2e | 4900 | |
3065976b QW |
4901 | ret = flush_write_bio(&epd); |
4902 | ASSERT(ret <= 0); | |
d1310b2e CM |
4903 | return ret; |
4904 | } | |
d1310b2e | 4905 | |
5e3ee236 | 4906 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end, |
771ed689 CM |
4907 | int mode) |
4908 | { | |
4909 | int ret = 0; | |
4910 | struct address_space *mapping = inode->i_mapping; | |
4911 | struct page *page; | |
09cbfeaf KS |
4912 | unsigned long nr_pages = (end - start + PAGE_SIZE) >> |
4913 | PAGE_SHIFT; | |
771ed689 CM |
4914 | |
4915 | struct extent_page_data epd = { | |
390ed29b | 4916 | .bio_ctrl = { 0 }, |
771ed689 | 4917 | .extent_locked = 1, |
ffbd517d | 4918 | .sync_io = mode == WB_SYNC_ALL, |
771ed689 CM |
4919 | }; |
4920 | struct writeback_control wbc_writepages = { | |
771ed689 | 4921 | .sync_mode = mode, |
771ed689 CM |
4922 | .nr_to_write = nr_pages * 2, |
4923 | .range_start = start, | |
4924 | .range_end = end + 1, | |
ec39f769 CM |
4925 | /* We're called from an async helper function */ |
4926 | .punt_to_cgroup = 1, | |
4927 | .no_cgroup_owner = 1, | |
771ed689 CM |
4928 | }; |
4929 | ||
dbb70bec | 4930 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
d397712b | 4931 | while (start <= end) { |
09cbfeaf | 4932 | page = find_get_page(mapping, start >> PAGE_SHIFT); |
771ed689 CM |
4933 | if (clear_page_dirty_for_io(page)) |
4934 | ret = __extent_writepage(page, &wbc_writepages, &epd); | |
4935 | else { | |
38a39ac7 QW |
4936 | btrfs_writepage_endio_finish_ordered(BTRFS_I(inode), |
4937 | page, start, start + PAGE_SIZE - 1, 1); | |
771ed689 CM |
4938 | unlock_page(page); |
4939 | } | |
09cbfeaf KS |
4940 | put_page(page); |
4941 | start += PAGE_SIZE; | |
771ed689 CM |
4942 | } |
4943 | ||
02c6db4f | 4944 | ASSERT(ret <= 0); |
dbb70bec CM |
4945 | if (ret == 0) |
4946 | ret = flush_write_bio(&epd); | |
4947 | else | |
02c6db4f | 4948 | end_write_bio(&epd, ret); |
dbb70bec CM |
4949 | |
4950 | wbc_detach_inode(&wbc_writepages); | |
771ed689 CM |
4951 | return ret; |
4952 | } | |
d1310b2e | 4953 | |
8ae225a8 | 4954 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
4955 | struct writeback_control *wbc) |
4956 | { | |
4957 | int ret = 0; | |
4958 | struct extent_page_data epd = { | |
390ed29b | 4959 | .bio_ctrl = { 0 }, |
771ed689 | 4960 | .extent_locked = 0, |
ffbd517d | 4961 | .sync_io = wbc->sync_mode == WB_SYNC_ALL, |
d1310b2e CM |
4962 | }; |
4963 | ||
935db853 | 4964 | ret = extent_write_cache_pages(mapping, wbc, &epd); |
a2a72fbd QW |
4965 | ASSERT(ret <= 0); |
4966 | if (ret < 0) { | |
4967 | end_write_bio(&epd, ret); | |
4968 | return ret; | |
4969 | } | |
4970 | ret = flush_write_bio(&epd); | |
d1310b2e CM |
4971 | return ret; |
4972 | } | |
d1310b2e | 4973 | |
ba206a02 | 4974 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 4975 | { |
390ed29b | 4976 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
67c9684f | 4977 | struct page *pagepool[16]; |
125bac01 | 4978 | struct extent_map *em_cached = NULL; |
808f80b4 | 4979 | u64 prev_em_start = (u64)-1; |
ba206a02 | 4980 | int nr; |
d1310b2e | 4981 | |
ba206a02 | 4982 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
4983 | u64 contig_start = readahead_pos(rac); |
4984 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 4985 | |
ba206a02 | 4986 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 4987 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 4988 | } |
67c9684f | 4989 | |
125bac01 MX |
4990 | if (em_cached) |
4991 | free_extent_map(em_cached); | |
4992 | ||
390ed29b QW |
4993 | if (bio_ctrl.bio) { |
4994 | if (submit_one_bio(bio_ctrl.bio, 0, bio_ctrl.bio_flags)) | |
ba206a02 MWO |
4995 | return; |
4996 | } | |
d1310b2e | 4997 | } |
d1310b2e CM |
4998 | |
4999 | /* | |
5000 | * basic invalidatepage code, this waits on any locked or writeback | |
5001 | * ranges corresponding to the page, and then deletes any extent state | |
5002 | * records from the tree | |
5003 | */ | |
5004 | int extent_invalidatepage(struct extent_io_tree *tree, | |
5005 | struct page *page, unsigned long offset) | |
5006 | { | |
2ac55d41 | 5007 | struct extent_state *cached_state = NULL; |
4eee4fa4 | 5008 | u64 start = page_offset(page); |
09cbfeaf | 5009 | u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
5010 | size_t blocksize = page->mapping->host->i_sb->s_blocksize; |
5011 | ||
829ddec9 QW |
5012 | /* This function is only called for the btree inode */ |
5013 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
5014 | ||
fda2832f | 5015 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
5016 | if (start > end) |
5017 | return 0; | |
5018 | ||
ff13db41 | 5019 | lock_extent_bits(tree, start, end, &cached_state); |
1edbb734 | 5020 | wait_on_page_writeback(page); |
829ddec9 QW |
5021 | |
5022 | /* | |
5023 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
5024 | * so here we only need to unlock the extent range to free any | |
5025 | * existing extent state. | |
5026 | */ | |
5027 | unlock_extent_cached(tree, start, end, &cached_state); | |
d1310b2e CM |
5028 | return 0; |
5029 | } | |
d1310b2e | 5030 | |
7b13b7b1 CM |
5031 | /* |
5032 | * a helper for releasepage, this tests for areas of the page that | |
5033 | * are locked or under IO and drops the related state bits if it is safe | |
5034 | * to drop the page. | |
5035 | */ | |
29c68b2d | 5036 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 5037 | struct page *page, gfp_t mask) |
7b13b7b1 | 5038 | { |
4eee4fa4 | 5039 | u64 start = page_offset(page); |
09cbfeaf | 5040 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
5041 | int ret = 1; |
5042 | ||
8882679e | 5043 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 5044 | ret = 0; |
8882679e | 5045 | } else { |
11ef160f | 5046 | /* |
2766ff61 FM |
5047 | * At this point we can safely clear everything except the |
5048 | * locked bit, the nodatasum bit and the delalloc new bit. | |
5049 | * The delalloc new bit will be cleared by ordered extent | |
5050 | * completion. | |
11ef160f | 5051 | */ |
66b0c887 | 5052 | ret = __clear_extent_bit(tree, start, end, |
2766ff61 FM |
5053 | ~(EXTENT_LOCKED | EXTENT_NODATASUM | EXTENT_DELALLOC_NEW), |
5054 | 0, 0, NULL, mask, NULL); | |
e3f24cc5 CM |
5055 | |
5056 | /* if clear_extent_bit failed for enomem reasons, | |
5057 | * we can't allow the release to continue. | |
5058 | */ | |
5059 | if (ret < 0) | |
5060 | ret = 0; | |
5061 | else | |
5062 | ret = 1; | |
7b13b7b1 CM |
5063 | } |
5064 | return ret; | |
5065 | } | |
7b13b7b1 | 5066 | |
d1310b2e CM |
5067 | /* |
5068 | * a helper for releasepage. As long as there are no locked extents | |
5069 | * in the range corresponding to the page, both state records and extent | |
5070 | * map records are removed | |
5071 | */ | |
477a30ba | 5072 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
5073 | { |
5074 | struct extent_map *em; | |
4eee4fa4 | 5075 | u64 start = page_offset(page); |
09cbfeaf | 5076 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
5077 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
5078 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
5079 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 5080 | |
d0164adc | 5081 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 5082 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 5083 | u64 len; |
70dec807 | 5084 | while (start <= end) { |
fbc2bd7e FM |
5085 | struct btrfs_fs_info *fs_info; |
5086 | u64 cur_gen; | |
5087 | ||
39b5637f | 5088 | len = end - start + 1; |
890871be | 5089 | write_lock(&map->lock); |
39b5637f | 5090 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 5091 | if (!em) { |
890871be | 5092 | write_unlock(&map->lock); |
70dec807 CM |
5093 | break; |
5094 | } | |
7f3c74fb CM |
5095 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
5096 | em->start != start) { | |
890871be | 5097 | write_unlock(&map->lock); |
70dec807 CM |
5098 | free_extent_map(em); |
5099 | break; | |
5100 | } | |
3d6448e6 FM |
5101 | if (test_range_bit(tree, em->start, |
5102 | extent_map_end(em) - 1, | |
5103 | EXTENT_LOCKED, 0, NULL)) | |
5104 | goto next; | |
5105 | /* | |
5106 | * If it's not in the list of modified extents, used | |
5107 | * by a fast fsync, we can remove it. If it's being | |
5108 | * logged we can safely remove it since fsync took an | |
5109 | * extra reference on the em. | |
5110 | */ | |
5111 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
5112 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
5113 | goto remove_em; | |
5114 | /* | |
5115 | * If it's in the list of modified extents, remove it | |
5116 | * only if its generation is older then the current one, | |
5117 | * in which case we don't need it for a fast fsync. | |
5118 | * Otherwise don't remove it, we could be racing with an | |
5119 | * ongoing fast fsync that could miss the new extent. | |
5120 | */ | |
5121 | fs_info = btrfs_inode->root->fs_info; | |
5122 | spin_lock(&fs_info->trans_lock); | |
5123 | cur_gen = fs_info->generation; | |
5124 | spin_unlock(&fs_info->trans_lock); | |
5125 | if (em->generation >= cur_gen) | |
5126 | goto next; | |
5127 | remove_em: | |
5e548b32 FM |
5128 | /* |
5129 | * We only remove extent maps that are not in the list of | |
5130 | * modified extents or that are in the list but with a | |
5131 | * generation lower then the current generation, so there | |
5132 | * is no need to set the full fsync flag on the inode (it | |
5133 | * hurts the fsync performance for workloads with a data | |
5134 | * size that exceeds or is close to the system's memory). | |
5135 | */ | |
fbc2bd7e FM |
5136 | remove_extent_mapping(map, em); |
5137 | /* once for the rb tree */ | |
5138 | free_extent_map(em); | |
3d6448e6 | 5139 | next: |
70dec807 | 5140 | start = extent_map_end(em); |
890871be | 5141 | write_unlock(&map->lock); |
70dec807 CM |
5142 | |
5143 | /* once for us */ | |
d1310b2e | 5144 | free_extent_map(em); |
9f47eb54 PM |
5145 | |
5146 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 5147 | } |
d1310b2e | 5148 | } |
29c68b2d | 5149 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 5150 | } |
d1310b2e | 5151 | |
ec29ed5b CM |
5152 | /* |
5153 | * helper function for fiemap, which doesn't want to see any holes. | |
5154 | * This maps until we find something past 'last' | |
5155 | */ | |
f1bbde8d | 5156 | static struct extent_map *get_extent_skip_holes(struct btrfs_inode *inode, |
e3350e16 | 5157 | u64 offset, u64 last) |
ec29ed5b | 5158 | { |
f1bbde8d | 5159 | u64 sectorsize = btrfs_inode_sectorsize(inode); |
ec29ed5b CM |
5160 | struct extent_map *em; |
5161 | u64 len; | |
5162 | ||
5163 | if (offset >= last) | |
5164 | return NULL; | |
5165 | ||
67871254 | 5166 | while (1) { |
ec29ed5b CM |
5167 | len = last - offset; |
5168 | if (len == 0) | |
5169 | break; | |
fda2832f | 5170 | len = ALIGN(len, sectorsize); |
f1bbde8d | 5171 | em = btrfs_get_extent_fiemap(inode, offset, len); |
c704005d | 5172 | if (IS_ERR_OR_NULL(em)) |
ec29ed5b CM |
5173 | return em; |
5174 | ||
5175 | /* if this isn't a hole return it */ | |
4a2d25cd | 5176 | if (em->block_start != EXTENT_MAP_HOLE) |
ec29ed5b | 5177 | return em; |
ec29ed5b CM |
5178 | |
5179 | /* this is a hole, advance to the next extent */ | |
5180 | offset = extent_map_end(em); | |
5181 | free_extent_map(em); | |
5182 | if (offset >= last) | |
5183 | break; | |
5184 | } | |
5185 | return NULL; | |
5186 | } | |
5187 | ||
4751832d QW |
5188 | /* |
5189 | * To cache previous fiemap extent | |
5190 | * | |
5191 | * Will be used for merging fiemap extent | |
5192 | */ | |
5193 | struct fiemap_cache { | |
5194 | u64 offset; | |
5195 | u64 phys; | |
5196 | u64 len; | |
5197 | u32 flags; | |
5198 | bool cached; | |
5199 | }; | |
5200 | ||
5201 | /* | |
5202 | * Helper to submit fiemap extent. | |
5203 | * | |
5204 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
5205 | * @len and @flags with cached one. | |
5206 | * And only when we fails to merge, cached one will be submitted as | |
5207 | * fiemap extent. | |
5208 | * | |
5209 | * Return value is the same as fiemap_fill_next_extent(). | |
5210 | */ | |
5211 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
5212 | struct fiemap_cache *cache, | |
5213 | u64 offset, u64 phys, u64 len, u32 flags) | |
5214 | { | |
5215 | int ret = 0; | |
5216 | ||
5217 | if (!cache->cached) | |
5218 | goto assign; | |
5219 | ||
5220 | /* | |
5221 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 5222 | * fiemap extent won't overlap with cached one. |
4751832d QW |
5223 | * Not recoverable. |
5224 | * | |
5225 | * NOTE: Physical address can overlap, due to compression | |
5226 | */ | |
5227 | if (cache->offset + cache->len > offset) { | |
5228 | WARN_ON(1); | |
5229 | return -EINVAL; | |
5230 | } | |
5231 | ||
5232 | /* | |
5233 | * Only merges fiemap extents if | |
5234 | * 1) Their logical addresses are continuous | |
5235 | * | |
5236 | * 2) Their physical addresses are continuous | |
5237 | * So truly compressed (physical size smaller than logical size) | |
5238 | * extents won't get merged with each other | |
5239 | * | |
5240 | * 3) Share same flags except FIEMAP_EXTENT_LAST | |
5241 | * So regular extent won't get merged with prealloc extent | |
5242 | */ | |
5243 | if (cache->offset + cache->len == offset && | |
5244 | cache->phys + cache->len == phys && | |
5245 | (cache->flags & ~FIEMAP_EXTENT_LAST) == | |
5246 | (flags & ~FIEMAP_EXTENT_LAST)) { | |
5247 | cache->len += len; | |
5248 | cache->flags |= flags; | |
5249 | goto try_submit_last; | |
5250 | } | |
5251 | ||
5252 | /* Not mergeable, need to submit cached one */ | |
5253 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
5254 | cache->len, cache->flags); | |
5255 | cache->cached = false; | |
5256 | if (ret) | |
5257 | return ret; | |
5258 | assign: | |
5259 | cache->cached = true; | |
5260 | cache->offset = offset; | |
5261 | cache->phys = phys; | |
5262 | cache->len = len; | |
5263 | cache->flags = flags; | |
5264 | try_submit_last: | |
5265 | if (cache->flags & FIEMAP_EXTENT_LAST) { | |
5266 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, | |
5267 | cache->phys, cache->len, cache->flags); | |
5268 | cache->cached = false; | |
5269 | } | |
5270 | return ret; | |
5271 | } | |
5272 | ||
5273 | /* | |
848c23b7 | 5274 | * Emit last fiemap cache |
4751832d | 5275 | * |
848c23b7 QW |
5276 | * The last fiemap cache may still be cached in the following case: |
5277 | * 0 4k 8k | |
5278 | * |<- Fiemap range ->| | |
5279 | * |<------------ First extent ----------->| | |
5280 | * | |
5281 | * In this case, the first extent range will be cached but not emitted. | |
5282 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 5283 | */ |
5c5aff98 | 5284 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 5285 | struct fiemap_cache *cache) |
4751832d QW |
5286 | { |
5287 | int ret; | |
5288 | ||
5289 | if (!cache->cached) | |
5290 | return 0; | |
5291 | ||
4751832d QW |
5292 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
5293 | cache->len, cache->flags); | |
5294 | cache->cached = false; | |
5295 | if (ret > 0) | |
5296 | ret = 0; | |
5297 | return ret; | |
5298 | } | |
5299 | ||
facee0a0 | 5300 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, |
bab16e21 | 5301 | u64 start, u64 len) |
1506fcc8 | 5302 | { |
975f84fe | 5303 | int ret = 0; |
15c7745c | 5304 | u64 off; |
1506fcc8 YS |
5305 | u64 max = start + len; |
5306 | u32 flags = 0; | |
975f84fe JB |
5307 | u32 found_type; |
5308 | u64 last; | |
ec29ed5b | 5309 | u64 last_for_get_extent = 0; |
1506fcc8 | 5310 | u64 disko = 0; |
facee0a0 | 5311 | u64 isize = i_size_read(&inode->vfs_inode); |
975f84fe | 5312 | struct btrfs_key found_key; |
1506fcc8 | 5313 | struct extent_map *em = NULL; |
2ac55d41 | 5314 | struct extent_state *cached_state = NULL; |
975f84fe | 5315 | struct btrfs_path *path; |
facee0a0 | 5316 | struct btrfs_root *root = inode->root; |
4751832d | 5317 | struct fiemap_cache cache = { 0 }; |
5911c8fe DS |
5318 | struct ulist *roots; |
5319 | struct ulist *tmp_ulist; | |
1506fcc8 | 5320 | int end = 0; |
ec29ed5b CM |
5321 | u64 em_start = 0; |
5322 | u64 em_len = 0; | |
5323 | u64 em_end = 0; | |
1506fcc8 YS |
5324 | |
5325 | if (len == 0) | |
5326 | return -EINVAL; | |
5327 | ||
975f84fe JB |
5328 | path = btrfs_alloc_path(); |
5329 | if (!path) | |
5330 | return -ENOMEM; | |
975f84fe | 5331 | |
5911c8fe DS |
5332 | roots = ulist_alloc(GFP_KERNEL); |
5333 | tmp_ulist = ulist_alloc(GFP_KERNEL); | |
5334 | if (!roots || !tmp_ulist) { | |
5335 | ret = -ENOMEM; | |
5336 | goto out_free_ulist; | |
5337 | } | |
5338 | ||
15c7745c BB |
5339 | /* |
5340 | * We can't initialize that to 'start' as this could miss extents due | |
5341 | * to extent item merging | |
5342 | */ | |
5343 | off = 0; | |
facee0a0 NB |
5344 | start = round_down(start, btrfs_inode_sectorsize(inode)); |
5345 | len = round_up(max, btrfs_inode_sectorsize(inode)) - start; | |
4d479cf0 | 5346 | |
ec29ed5b CM |
5347 | /* |
5348 | * lookup the last file extent. We're not using i_size here | |
5349 | * because there might be preallocation past i_size | |
5350 | */ | |
facee0a0 NB |
5351 | ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode), -1, |
5352 | 0); | |
975f84fe | 5353 | if (ret < 0) { |
5911c8fe | 5354 | goto out_free_ulist; |
2d324f59 LB |
5355 | } else { |
5356 | WARN_ON(!ret); | |
5357 | if (ret == 1) | |
5358 | ret = 0; | |
975f84fe | 5359 | } |
2d324f59 | 5360 | |
975f84fe | 5361 | path->slots[0]--; |
975f84fe | 5362 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); |
962a298f | 5363 | found_type = found_key.type; |
975f84fe | 5364 | |
ec29ed5b | 5365 | /* No extents, but there might be delalloc bits */ |
facee0a0 | 5366 | if (found_key.objectid != btrfs_ino(inode) || |
975f84fe | 5367 | found_type != BTRFS_EXTENT_DATA_KEY) { |
ec29ed5b CM |
5368 | /* have to trust i_size as the end */ |
5369 | last = (u64)-1; | |
5370 | last_for_get_extent = isize; | |
5371 | } else { | |
5372 | /* | |
5373 | * remember the start of the last extent. There are a | |
5374 | * bunch of different factors that go into the length of the | |
5375 | * extent, so its much less complex to remember where it started | |
5376 | */ | |
5377 | last = found_key.offset; | |
5378 | last_for_get_extent = last + 1; | |
975f84fe | 5379 | } |
fe09e16c | 5380 | btrfs_release_path(path); |
975f84fe | 5381 | |
ec29ed5b CM |
5382 | /* |
5383 | * we might have some extents allocated but more delalloc past those | |
5384 | * extents. so, we trust isize unless the start of the last extent is | |
5385 | * beyond isize | |
5386 | */ | |
5387 | if (last < isize) { | |
5388 | last = (u64)-1; | |
5389 | last_for_get_extent = isize; | |
5390 | } | |
5391 | ||
facee0a0 | 5392 | lock_extent_bits(&inode->io_tree, start, start + len - 1, |
d0082371 | 5393 | &cached_state); |
ec29ed5b | 5394 | |
facee0a0 | 5395 | em = get_extent_skip_holes(inode, start, last_for_get_extent); |
1506fcc8 YS |
5396 | if (!em) |
5397 | goto out; | |
5398 | if (IS_ERR(em)) { | |
5399 | ret = PTR_ERR(em); | |
5400 | goto out; | |
5401 | } | |
975f84fe | 5402 | |
1506fcc8 | 5403 | while (!end) { |
b76bb701 | 5404 | u64 offset_in_extent = 0; |
ea8efc74 CM |
5405 | |
5406 | /* break if the extent we found is outside the range */ | |
5407 | if (em->start >= max || extent_map_end(em) < off) | |
5408 | break; | |
5409 | ||
5410 | /* | |
5411 | * get_extent may return an extent that starts before our | |
5412 | * requested range. We have to make sure the ranges | |
5413 | * we return to fiemap always move forward and don't | |
5414 | * overlap, so adjust the offsets here | |
5415 | */ | |
5416 | em_start = max(em->start, off); | |
1506fcc8 | 5417 | |
ea8efc74 CM |
5418 | /* |
5419 | * record the offset from the start of the extent | |
b76bb701 JB |
5420 | * for adjusting the disk offset below. Only do this if the |
5421 | * extent isn't compressed since our in ram offset may be past | |
5422 | * what we have actually allocated on disk. | |
ea8efc74 | 5423 | */ |
b76bb701 JB |
5424 | if (!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
5425 | offset_in_extent = em_start - em->start; | |
ec29ed5b | 5426 | em_end = extent_map_end(em); |
ea8efc74 | 5427 | em_len = em_end - em_start; |
1506fcc8 | 5428 | flags = 0; |
f0986318 FM |
5429 | if (em->block_start < EXTENT_MAP_LAST_BYTE) |
5430 | disko = em->block_start + offset_in_extent; | |
5431 | else | |
5432 | disko = 0; | |
1506fcc8 | 5433 | |
ea8efc74 CM |
5434 | /* |
5435 | * bump off for our next call to get_extent | |
5436 | */ | |
5437 | off = extent_map_end(em); | |
5438 | if (off >= max) | |
5439 | end = 1; | |
5440 | ||
93dbfad7 | 5441 | if (em->block_start == EXTENT_MAP_LAST_BYTE) { |
1506fcc8 YS |
5442 | end = 1; |
5443 | flags |= FIEMAP_EXTENT_LAST; | |
93dbfad7 | 5444 | } else if (em->block_start == EXTENT_MAP_INLINE) { |
1506fcc8 YS |
5445 | flags |= (FIEMAP_EXTENT_DATA_INLINE | |
5446 | FIEMAP_EXTENT_NOT_ALIGNED); | |
93dbfad7 | 5447 | } else if (em->block_start == EXTENT_MAP_DELALLOC) { |
1506fcc8 YS |
5448 | flags |= (FIEMAP_EXTENT_DELALLOC | |
5449 | FIEMAP_EXTENT_UNKNOWN); | |
dc046b10 JB |
5450 | } else if (fieinfo->fi_extents_max) { |
5451 | u64 bytenr = em->block_start - | |
5452 | (em->start - em->orig_start); | |
fe09e16c | 5453 | |
fe09e16c LB |
5454 | /* |
5455 | * As btrfs supports shared space, this information | |
5456 | * can be exported to userspace tools via | |
dc046b10 JB |
5457 | * flag FIEMAP_EXTENT_SHARED. If fi_extents_max == 0 |
5458 | * then we're just getting a count and we can skip the | |
5459 | * lookup stuff. | |
fe09e16c | 5460 | */ |
facee0a0 | 5461 | ret = btrfs_check_shared(root, btrfs_ino(inode), |
5911c8fe | 5462 | bytenr, roots, tmp_ulist); |
dc046b10 | 5463 | if (ret < 0) |
fe09e16c | 5464 | goto out_free; |
dc046b10 | 5465 | if (ret) |
fe09e16c | 5466 | flags |= FIEMAP_EXTENT_SHARED; |
dc046b10 | 5467 | ret = 0; |
1506fcc8 YS |
5468 | } |
5469 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) | |
5470 | flags |= FIEMAP_EXTENT_ENCODED; | |
0d2b2372 JB |
5471 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
5472 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
1506fcc8 | 5473 | |
1506fcc8 YS |
5474 | free_extent_map(em); |
5475 | em = NULL; | |
ec29ed5b CM |
5476 | if ((em_start >= last) || em_len == (u64)-1 || |
5477 | (last == (u64)-1 && isize <= em_end)) { | |
1506fcc8 YS |
5478 | flags |= FIEMAP_EXTENT_LAST; |
5479 | end = 1; | |
5480 | } | |
5481 | ||
ec29ed5b | 5482 | /* now scan forward to see if this is really the last extent. */ |
facee0a0 | 5483 | em = get_extent_skip_holes(inode, off, last_for_get_extent); |
ec29ed5b CM |
5484 | if (IS_ERR(em)) { |
5485 | ret = PTR_ERR(em); | |
5486 | goto out; | |
5487 | } | |
5488 | if (!em) { | |
975f84fe JB |
5489 | flags |= FIEMAP_EXTENT_LAST; |
5490 | end = 1; | |
5491 | } | |
4751832d QW |
5492 | ret = emit_fiemap_extent(fieinfo, &cache, em_start, disko, |
5493 | em_len, flags); | |
26e726af CS |
5494 | if (ret) { |
5495 | if (ret == 1) | |
5496 | ret = 0; | |
ec29ed5b | 5497 | goto out_free; |
26e726af | 5498 | } |
1506fcc8 YS |
5499 | } |
5500 | out_free: | |
4751832d | 5501 | if (!ret) |
5c5aff98 | 5502 | ret = emit_last_fiemap_cache(fieinfo, &cache); |
1506fcc8 YS |
5503 | free_extent_map(em); |
5504 | out: | |
facee0a0 | 5505 | unlock_extent_cached(&inode->io_tree, start, start + len - 1, |
e43bbe5e | 5506 | &cached_state); |
5911c8fe DS |
5507 | |
5508 | out_free_ulist: | |
e02d48ea | 5509 | btrfs_free_path(path); |
5911c8fe DS |
5510 | ulist_free(roots); |
5511 | ulist_free(tmp_ulist); | |
1506fcc8 YS |
5512 | return ret; |
5513 | } | |
5514 | ||
727011e0 CM |
5515 | static void __free_extent_buffer(struct extent_buffer *eb) |
5516 | { | |
727011e0 CM |
5517 | kmem_cache_free(extent_buffer_cache, eb); |
5518 | } | |
5519 | ||
2b48966a | 5520 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
5521 | { |
5522 | return (atomic_read(&eb->io_pages) || | |
5523 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
5524 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5525 | } | |
5526 | ||
8ff8466d | 5527 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 5528 | { |
8ff8466d | 5529 | struct btrfs_subpage *subpage; |
db7f3436 | 5530 | |
8ff8466d | 5531 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 5532 | |
8ff8466d QW |
5533 | if (PagePrivate(page)) { |
5534 | subpage = (struct btrfs_subpage *)page->private; | |
5535 | if (atomic_read(&subpage->eb_refs)) | |
5536 | return true; | |
5537 | } | |
5538 | return false; | |
5539 | } | |
db7f3436 | 5540 | |
8ff8466d QW |
5541 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
5542 | { | |
5543 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
5544 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
5545 | ||
5546 | /* | |
5547 | * For mapped eb, we're going to change the page private, which should | |
5548 | * be done under the private_lock. | |
5549 | */ | |
5550 | if (mapped) | |
5551 | spin_lock(&page->mapping->private_lock); | |
5552 | ||
5553 | if (!PagePrivate(page)) { | |
5d2361db | 5554 | if (mapped) |
8ff8466d QW |
5555 | spin_unlock(&page->mapping->private_lock); |
5556 | return; | |
5557 | } | |
5558 | ||
5559 | if (fs_info->sectorsize == PAGE_SIZE) { | |
5d2361db FL |
5560 | /* |
5561 | * We do this since we'll remove the pages after we've | |
5562 | * removed the eb from the radix tree, so we could race | |
5563 | * and have this page now attached to the new eb. So | |
5564 | * only clear page_private if it's still connected to | |
5565 | * this eb. | |
5566 | */ | |
5567 | if (PagePrivate(page) && | |
5568 | page->private == (unsigned long)eb) { | |
5569 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
5570 | BUG_ON(PageDirty(page)); | |
5571 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 5572 | /* |
5d2361db FL |
5573 | * We need to make sure we haven't be attached |
5574 | * to a new eb. | |
db7f3436 | 5575 | */ |
d1b89bc0 | 5576 | detach_page_private(page); |
db7f3436 | 5577 | } |
5d2361db FL |
5578 | if (mapped) |
5579 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
5580 | return; |
5581 | } | |
5582 | ||
5583 | /* | |
5584 | * For subpage, we can have dummy eb with page private. In this case, | |
5585 | * we can directly detach the private as such page is only attached to | |
5586 | * one dummy eb, no sharing. | |
5587 | */ | |
5588 | if (!mapped) { | |
5589 | btrfs_detach_subpage(fs_info, page); | |
5590 | return; | |
5591 | } | |
5592 | ||
5593 | btrfs_page_dec_eb_refs(fs_info, page); | |
5594 | ||
5595 | /* | |
5596 | * We can only detach the page private if there are no other ebs in the | |
5597 | * page range. | |
5598 | */ | |
5599 | if (!page_range_has_eb(fs_info, page)) | |
5600 | btrfs_detach_subpage(fs_info, page); | |
5601 | ||
5602 | spin_unlock(&page->mapping->private_lock); | |
5603 | } | |
5604 | ||
5605 | /* Release all pages attached to the extent buffer */ | |
5606 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
5607 | { | |
5608 | int i; | |
5609 | int num_pages; | |
5610 | ||
5611 | ASSERT(!extent_buffer_under_io(eb)); | |
5612 | ||
5613 | num_pages = num_extent_pages(eb); | |
5614 | for (i = 0; i < num_pages; i++) { | |
5615 | struct page *page = eb->pages[i]; | |
5616 | ||
5617 | if (!page) | |
5618 | continue; | |
5619 | ||
5620 | detach_extent_buffer_page(eb, page); | |
5d2361db | 5621 | |
01327610 | 5622 | /* One for when we allocated the page */ |
09cbfeaf | 5623 | put_page(page); |
d64766fd | 5624 | } |
db7f3436 JB |
5625 | } |
5626 | ||
5627 | /* | |
5628 | * Helper for releasing the extent buffer. | |
5629 | */ | |
5630 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
5631 | { | |
55ac0139 | 5632 | btrfs_release_extent_buffer_pages(eb); |
8c38938c | 5633 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
db7f3436 JB |
5634 | __free_extent_buffer(eb); |
5635 | } | |
5636 | ||
f28491e0 JB |
5637 | static struct extent_buffer * |
5638 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 5639 | unsigned long len) |
d1310b2e CM |
5640 | { |
5641 | struct extent_buffer *eb = NULL; | |
5642 | ||
d1b5c567 | 5643 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
5644 | eb->start = start; |
5645 | eb->len = len; | |
f28491e0 | 5646 | eb->fs_info = fs_info; |
815a51c7 | 5647 | eb->bflags = 0; |
196d59ab | 5648 | init_rwsem(&eb->lock); |
b4ce94de | 5649 | |
3fd63727 JB |
5650 | btrfs_leak_debug_add(&fs_info->eb_leak_lock, &eb->leak_list, |
5651 | &fs_info->allocated_ebs); | |
d3575156 | 5652 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 5653 | |
3083ee2e | 5654 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 5655 | atomic_set(&eb->refs, 1); |
0b32f4bb | 5656 | atomic_set(&eb->io_pages, 0); |
727011e0 | 5657 | |
deb67895 | 5658 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
5659 | |
5660 | return eb; | |
5661 | } | |
5662 | ||
2b48966a | 5663 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 5664 | { |
cc5e31a4 | 5665 | int i; |
815a51c7 JS |
5666 | struct page *p; |
5667 | struct extent_buffer *new; | |
cc5e31a4 | 5668 | int num_pages = num_extent_pages(src); |
815a51c7 | 5669 | |
3f556f78 | 5670 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
5671 | if (new == NULL) |
5672 | return NULL; | |
5673 | ||
62c053fb QW |
5674 | /* |
5675 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
5676 | * btrfs_release_extent_buffer() have different behavior for | |
5677 | * UNMAPPED subpage extent buffer. | |
5678 | */ | |
5679 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
5680 | ||
815a51c7 | 5681 | for (i = 0; i < num_pages; i++) { |
760f991f QW |
5682 | int ret; |
5683 | ||
9ec72677 | 5684 | p = alloc_page(GFP_NOFS); |
db7f3436 JB |
5685 | if (!p) { |
5686 | btrfs_release_extent_buffer(new); | |
5687 | return NULL; | |
5688 | } | |
760f991f QW |
5689 | ret = attach_extent_buffer_page(new, p, NULL); |
5690 | if (ret < 0) { | |
5691 | put_page(p); | |
5692 | btrfs_release_extent_buffer(new); | |
5693 | return NULL; | |
5694 | } | |
815a51c7 | 5695 | WARN_ON(PageDirty(p)); |
815a51c7 | 5696 | new->pages[i] = p; |
fba1acf9 | 5697 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 5698 | } |
92d83e94 | 5699 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
5700 | |
5701 | return new; | |
5702 | } | |
5703 | ||
0f331229 OS |
5704 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
5705 | u64 start, unsigned long len) | |
815a51c7 JS |
5706 | { |
5707 | struct extent_buffer *eb; | |
cc5e31a4 DS |
5708 | int num_pages; |
5709 | int i; | |
815a51c7 | 5710 | |
3f556f78 | 5711 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
5712 | if (!eb) |
5713 | return NULL; | |
5714 | ||
65ad0104 | 5715 | num_pages = num_extent_pages(eb); |
815a51c7 | 5716 | for (i = 0; i < num_pages; i++) { |
09bc1f0f QW |
5717 | int ret; |
5718 | ||
9ec72677 | 5719 | eb->pages[i] = alloc_page(GFP_NOFS); |
815a51c7 JS |
5720 | if (!eb->pages[i]) |
5721 | goto err; | |
09bc1f0f QW |
5722 | ret = attach_extent_buffer_page(eb, eb->pages[i], NULL); |
5723 | if (ret < 0) | |
5724 | goto err; | |
815a51c7 JS |
5725 | } |
5726 | set_extent_buffer_uptodate(eb); | |
5727 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 5728 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
5729 | |
5730 | return eb; | |
5731 | err: | |
09bc1f0f QW |
5732 | for (; i > 0; i--) { |
5733 | detach_extent_buffer_page(eb, eb->pages[i - 1]); | |
84167d19 | 5734 | __free_page(eb->pages[i - 1]); |
09bc1f0f | 5735 | } |
815a51c7 JS |
5736 | __free_extent_buffer(eb); |
5737 | return NULL; | |
5738 | } | |
5739 | ||
0f331229 | 5740 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 5741 | u64 start) |
0f331229 | 5742 | { |
da17066c | 5743 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
5744 | } |
5745 | ||
0b32f4bb JB |
5746 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
5747 | { | |
242e18c7 | 5748 | int refs; |
6bf9cd2e BB |
5749 | /* |
5750 | * The TREE_REF bit is first set when the extent_buffer is added | |
5751 | * to the radix tree. It is also reset, if unset, when a new reference | |
5752 | * is created by find_extent_buffer. | |
0b32f4bb | 5753 | * |
6bf9cd2e BB |
5754 | * It is only cleared in two cases: freeing the last non-tree |
5755 | * reference to the extent_buffer when its STALE bit is set or | |
5756 | * calling releasepage when the tree reference is the only reference. | |
0b32f4bb | 5757 | * |
6bf9cd2e BB |
5758 | * In both cases, care is taken to ensure that the extent_buffer's |
5759 | * pages are not under io. However, releasepage can be concurrently | |
5760 | * called with creating new references, which is prone to race | |
5761 | * conditions between the calls to check_buffer_tree_ref in those | |
5762 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 5763 | * |
6bf9cd2e BB |
5764 | * The actual lifetime of the extent_buffer in the radix tree is |
5765 | * adequately protected by the refcount, but the TREE_REF bit and | |
5766 | * its corresponding reference are not. To protect against this | |
5767 | * class of races, we call check_buffer_tree_ref from the codepaths | |
5768 | * which trigger io after they set eb->io_pages. Note that once io is | |
5769 | * initiated, TREE_REF can no longer be cleared, so that is the | |
5770 | * moment at which any such race is best fixed. | |
0b32f4bb | 5771 | */ |
242e18c7 CM |
5772 | refs = atomic_read(&eb->refs); |
5773 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
5774 | return; | |
5775 | ||
594831c4 JB |
5776 | spin_lock(&eb->refs_lock); |
5777 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 5778 | atomic_inc(&eb->refs); |
594831c4 | 5779 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
5780 | } |
5781 | ||
2457aec6 MG |
5782 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
5783 | struct page *accessed) | |
5df4235e | 5784 | { |
cc5e31a4 | 5785 | int num_pages, i; |
5df4235e | 5786 | |
0b32f4bb JB |
5787 | check_buffer_tree_ref(eb); |
5788 | ||
65ad0104 | 5789 | num_pages = num_extent_pages(eb); |
5df4235e | 5790 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
5791 | struct page *p = eb->pages[i]; |
5792 | ||
2457aec6 MG |
5793 | if (p != accessed) |
5794 | mark_page_accessed(p); | |
5df4235e JB |
5795 | } |
5796 | } | |
5797 | ||
f28491e0 JB |
5798 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
5799 | u64 start) | |
452c75c3 CS |
5800 | { |
5801 | struct extent_buffer *eb; | |
5802 | ||
2f3186d8 QW |
5803 | eb = find_extent_buffer_nolock(fs_info, start); |
5804 | if (!eb) | |
5805 | return NULL; | |
5806 | /* | |
5807 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
5808 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
5809 | * another task running free_extent_buffer() might have seen that flag | |
5810 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
5811 | * writeback flags not set) and it's still in the tree (flag | |
5812 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
5813 | * decrementing the extent buffer's reference count twice. So here we | |
5814 | * could race and increment the eb's reference count, clear its stale | |
5815 | * flag, mark it as dirty and drop our reference before the other task | |
5816 | * finishes executing free_extent_buffer, which would later result in | |
5817 | * an attempt to free an extent buffer that is dirty. | |
5818 | */ | |
5819 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
5820 | spin_lock(&eb->refs_lock); | |
5821 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 5822 | } |
2f3186d8 QW |
5823 | mark_extent_buffer_accessed(eb, NULL); |
5824 | return eb; | |
452c75c3 CS |
5825 | } |
5826 | ||
faa2dbf0 JB |
5827 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
5828 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 5829 | u64 start) |
faa2dbf0 JB |
5830 | { |
5831 | struct extent_buffer *eb, *exists = NULL; | |
5832 | int ret; | |
5833 | ||
5834 | eb = find_extent_buffer(fs_info, start); | |
5835 | if (eb) | |
5836 | return eb; | |
da17066c | 5837 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 5838 | if (!eb) |
b6293c82 | 5839 | return ERR_PTR(-ENOMEM); |
faa2dbf0 JB |
5840 | eb->fs_info = fs_info; |
5841 | again: | |
e1860a77 | 5842 | ret = radix_tree_preload(GFP_NOFS); |
b6293c82 DC |
5843 | if (ret) { |
5844 | exists = ERR_PTR(ret); | |
faa2dbf0 | 5845 | goto free_eb; |
b6293c82 | 5846 | } |
faa2dbf0 JB |
5847 | spin_lock(&fs_info->buffer_lock); |
5848 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 5849 | start >> fs_info->sectorsize_bits, eb); |
faa2dbf0 JB |
5850 | spin_unlock(&fs_info->buffer_lock); |
5851 | radix_tree_preload_end(); | |
5852 | if (ret == -EEXIST) { | |
5853 | exists = find_extent_buffer(fs_info, start); | |
5854 | if (exists) | |
5855 | goto free_eb; | |
5856 | else | |
5857 | goto again; | |
5858 | } | |
5859 | check_buffer_tree_ref(eb); | |
5860 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
5861 | ||
faa2dbf0 JB |
5862 | return eb; |
5863 | free_eb: | |
5864 | btrfs_release_extent_buffer(eb); | |
5865 | return exists; | |
5866 | } | |
5867 | #endif | |
5868 | ||
81982210 QW |
5869 | static struct extent_buffer *grab_extent_buffer( |
5870 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
5871 | { |
5872 | struct extent_buffer *exists; | |
5873 | ||
81982210 QW |
5874 | /* |
5875 | * For subpage case, we completely rely on radix tree to ensure we | |
5876 | * don't try to insert two ebs for the same bytenr. So here we always | |
5877 | * return NULL and just continue. | |
5878 | */ | |
5879 | if (fs_info->sectorsize < PAGE_SIZE) | |
5880 | return NULL; | |
5881 | ||
c0f0a9e7 QW |
5882 | /* Page not yet attached to an extent buffer */ |
5883 | if (!PagePrivate(page)) | |
5884 | return NULL; | |
5885 | ||
5886 | /* | |
5887 | * We could have already allocated an eb for this page and attached one | |
5888 | * so lets see if we can get a ref on the existing eb, and if we can we | |
5889 | * know it's good and we can just return that one, else we know we can | |
5890 | * just overwrite page->private. | |
5891 | */ | |
5892 | exists = (struct extent_buffer *)page->private; | |
5893 | if (atomic_inc_not_zero(&exists->refs)) | |
5894 | return exists; | |
5895 | ||
5896 | WARN_ON(PageDirty(page)); | |
5897 | detach_page_private(page); | |
5898 | return NULL; | |
5899 | } | |
5900 | ||
f28491e0 | 5901 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 5902 | u64 start, u64 owner_root, int level) |
d1310b2e | 5903 | { |
da17066c | 5904 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
5905 | int num_pages; |
5906 | int i; | |
09cbfeaf | 5907 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 5908 | struct extent_buffer *eb; |
6af118ce | 5909 | struct extent_buffer *exists = NULL; |
d1310b2e | 5910 | struct page *p; |
f28491e0 | 5911 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
d1310b2e | 5912 | int uptodate = 1; |
19fe0a8b | 5913 | int ret; |
d1310b2e | 5914 | |
da17066c | 5915 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { |
c871b0f2 LB |
5916 | btrfs_err(fs_info, "bad tree block start %llu", start); |
5917 | return ERR_PTR(-EINVAL); | |
5918 | } | |
5919 | ||
e9306ad4 QW |
5920 | #if BITS_PER_LONG == 32 |
5921 | if (start >= MAX_LFS_FILESIZE) { | |
5922 | btrfs_err_rl(fs_info, | |
5923 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
5924 | btrfs_err_32bit_limit(fs_info); | |
5925 | return ERR_PTR(-EOVERFLOW); | |
5926 | } | |
5927 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
5928 | btrfs_warn_32bit_limit(fs_info); | |
5929 | #endif | |
5930 | ||
1aaac38c QW |
5931 | if (fs_info->sectorsize < PAGE_SIZE && |
5932 | offset_in_page(start) + len > PAGE_SIZE) { | |
5933 | btrfs_err(fs_info, | |
5934 | "tree block crosses page boundary, start %llu nodesize %lu", | |
5935 | start, len); | |
5936 | return ERR_PTR(-EINVAL); | |
5937 | } | |
5938 | ||
f28491e0 | 5939 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 5940 | if (eb) |
6af118ce | 5941 | return eb; |
6af118ce | 5942 | |
23d79d81 | 5943 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 5944 | if (!eb) |
c871b0f2 | 5945 | return ERR_PTR(-ENOMEM); |
e114c545 | 5946 | btrfs_set_buffer_lockdep_class(owner_root, eb, level); |
d1310b2e | 5947 | |
65ad0104 | 5948 | num_pages = num_extent_pages(eb); |
727011e0 | 5949 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
5950 | struct btrfs_subpage *prealloc = NULL; |
5951 | ||
d1b5c567 | 5952 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
5953 | if (!p) { |
5954 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 5955 | goto free_eb; |
c871b0f2 | 5956 | } |
4f2de97a | 5957 | |
760f991f QW |
5958 | /* |
5959 | * Preallocate page->private for subpage case, so that we won't | |
5960 | * allocate memory with private_lock hold. The memory will be | |
5961 | * freed by attach_extent_buffer_page() or freed manually if | |
5962 | * we exit earlier. | |
5963 | * | |
5964 | * Although we have ensured one subpage eb can only have one | |
5965 | * page, but it may change in the future for 16K page size | |
5966 | * support, so we still preallocate the memory in the loop. | |
5967 | */ | |
5968 | ret = btrfs_alloc_subpage(fs_info, &prealloc, | |
5969 | BTRFS_SUBPAGE_METADATA); | |
5970 | if (ret < 0) { | |
5971 | unlock_page(p); | |
5972 | put_page(p); | |
5973 | exists = ERR_PTR(ret); | |
5974 | goto free_eb; | |
5975 | } | |
5976 | ||
4f2de97a | 5977 | spin_lock(&mapping->private_lock); |
81982210 | 5978 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
5979 | if (exists) { |
5980 | spin_unlock(&mapping->private_lock); | |
5981 | unlock_page(p); | |
5982 | put_page(p); | |
5983 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 5984 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 5985 | goto free_eb; |
d1310b2e | 5986 | } |
760f991f QW |
5987 | /* Should not fail, as we have preallocated the memory */ |
5988 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
5989 | ASSERT(!ret); | |
8ff8466d QW |
5990 | /* |
5991 | * To inform we have extra eb under allocation, so that | |
5992 | * detach_extent_buffer_page() won't release the page private | |
5993 | * when the eb hasn't yet been inserted into radix tree. | |
5994 | * | |
5995 | * The ref will be decreased when the eb released the page, in | |
5996 | * detach_extent_buffer_page(). | |
5997 | * Thus needs no special handling in error path. | |
5998 | */ | |
5999 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 6000 | spin_unlock(&mapping->private_lock); |
760f991f | 6001 | |
1e5eb3d6 | 6002 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 6003 | eb->pages[i] = p; |
d1310b2e CM |
6004 | if (!PageUptodate(p)) |
6005 | uptodate = 0; | |
eb14ab8e CM |
6006 | |
6007 | /* | |
b16d011e NB |
6008 | * We can't unlock the pages just yet since the extent buffer |
6009 | * hasn't been properly inserted in the radix tree, this | |
6010 | * opens a race with btree_releasepage which can free a page | |
6011 | * while we are still filling in all pages for the buffer and | |
6012 | * we could crash. | |
eb14ab8e | 6013 | */ |
d1310b2e CM |
6014 | } |
6015 | if (uptodate) | |
b4ce94de | 6016 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
115391d2 | 6017 | again: |
e1860a77 | 6018 | ret = radix_tree_preload(GFP_NOFS); |
c871b0f2 LB |
6019 | if (ret) { |
6020 | exists = ERR_PTR(ret); | |
19fe0a8b | 6021 | goto free_eb; |
c871b0f2 | 6022 | } |
19fe0a8b | 6023 | |
f28491e0 JB |
6024 | spin_lock(&fs_info->buffer_lock); |
6025 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
478ef886 | 6026 | start >> fs_info->sectorsize_bits, eb); |
f28491e0 | 6027 | spin_unlock(&fs_info->buffer_lock); |
452c75c3 | 6028 | radix_tree_preload_end(); |
19fe0a8b | 6029 | if (ret == -EEXIST) { |
f28491e0 | 6030 | exists = find_extent_buffer(fs_info, start); |
452c75c3 CS |
6031 | if (exists) |
6032 | goto free_eb; | |
6033 | else | |
115391d2 | 6034 | goto again; |
6af118ce | 6035 | } |
6af118ce | 6036 | /* add one reference for the tree */ |
0b32f4bb | 6037 | check_buffer_tree_ref(eb); |
34b41ace | 6038 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
6039 | |
6040 | /* | |
b16d011e NB |
6041 | * Now it's safe to unlock the pages because any calls to |
6042 | * btree_releasepage will correctly detect that a page belongs to a | |
6043 | * live buffer and won't free them prematurely. | |
eb14ab8e | 6044 | */ |
28187ae5 NB |
6045 | for (i = 0; i < num_pages; i++) |
6046 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
6047 | return eb; |
6048 | ||
6af118ce | 6049 | free_eb: |
5ca64f45 | 6050 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
6051 | for (i = 0; i < num_pages; i++) { |
6052 | if (eb->pages[i]) | |
6053 | unlock_page(eb->pages[i]); | |
6054 | } | |
eb14ab8e | 6055 | |
897ca6e9 | 6056 | btrfs_release_extent_buffer(eb); |
6af118ce | 6057 | return exists; |
d1310b2e | 6058 | } |
d1310b2e | 6059 | |
3083ee2e JB |
6060 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
6061 | { | |
6062 | struct extent_buffer *eb = | |
6063 | container_of(head, struct extent_buffer, rcu_head); | |
6064 | ||
6065 | __free_extent_buffer(eb); | |
6066 | } | |
6067 | ||
f7a52a40 | 6068 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 6069 | __releases(&eb->refs_lock) |
3083ee2e | 6070 | { |
07e21c4d NB |
6071 | lockdep_assert_held(&eb->refs_lock); |
6072 | ||
3083ee2e JB |
6073 | WARN_ON(atomic_read(&eb->refs) == 0); |
6074 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 6075 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 6076 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 6077 | |
815a51c7 | 6078 | spin_unlock(&eb->refs_lock); |
3083ee2e | 6079 | |
f28491e0 JB |
6080 | spin_lock(&fs_info->buffer_lock); |
6081 | radix_tree_delete(&fs_info->buffer_radix, | |
478ef886 | 6082 | eb->start >> fs_info->sectorsize_bits); |
f28491e0 | 6083 | spin_unlock(&fs_info->buffer_lock); |
34b41ace JB |
6084 | } else { |
6085 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 6086 | } |
3083ee2e | 6087 | |
8c38938c | 6088 | btrfs_leak_debug_del(&eb->fs_info->eb_leak_lock, &eb->leak_list); |
3083ee2e | 6089 | /* Should be safe to release our pages at this point */ |
55ac0139 | 6090 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 6091 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 6092 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
6093 | __free_extent_buffer(eb); |
6094 | return 1; | |
6095 | } | |
6096 | #endif | |
3083ee2e | 6097 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 6098 | return 1; |
3083ee2e JB |
6099 | } |
6100 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
6101 | |
6102 | return 0; | |
3083ee2e JB |
6103 | } |
6104 | ||
d1310b2e CM |
6105 | void free_extent_buffer(struct extent_buffer *eb) |
6106 | { | |
242e18c7 CM |
6107 | int refs; |
6108 | int old; | |
d1310b2e CM |
6109 | if (!eb) |
6110 | return; | |
6111 | ||
242e18c7 CM |
6112 | while (1) { |
6113 | refs = atomic_read(&eb->refs); | |
46cc775e NB |
6114 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
6115 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
6116 | refs == 1)) | |
242e18c7 CM |
6117 | break; |
6118 | old = atomic_cmpxchg(&eb->refs, refs, refs - 1); | |
6119 | if (old == refs) | |
6120 | return; | |
6121 | } | |
6122 | ||
3083ee2e JB |
6123 | spin_lock(&eb->refs_lock); |
6124 | if (atomic_read(&eb->refs) == 2 && | |
6125 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 6126 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
6127 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6128 | atomic_dec(&eb->refs); | |
6129 | ||
6130 | /* | |
6131 | * I know this is terrible, but it's temporary until we stop tracking | |
6132 | * the uptodate bits and such for the extent buffers. | |
6133 | */ | |
f7a52a40 | 6134 | release_extent_buffer(eb); |
3083ee2e JB |
6135 | } |
6136 | ||
6137 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
6138 | { | |
6139 | if (!eb) | |
d1310b2e CM |
6140 | return; |
6141 | ||
3083ee2e JB |
6142 | spin_lock(&eb->refs_lock); |
6143 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
6144 | ||
0b32f4bb | 6145 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
6146 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
6147 | atomic_dec(&eb->refs); | |
f7a52a40 | 6148 | release_extent_buffer(eb); |
d1310b2e | 6149 | } |
d1310b2e | 6150 | |
0d27797e QW |
6151 | static void btree_clear_page_dirty(struct page *page) |
6152 | { | |
6153 | ASSERT(PageDirty(page)); | |
6154 | ASSERT(PageLocked(page)); | |
6155 | clear_page_dirty_for_io(page); | |
6156 | xa_lock_irq(&page->mapping->i_pages); | |
6157 | if (!PageDirty(page)) | |
6158 | __xa_clear_mark(&page->mapping->i_pages, | |
6159 | page_index(page), PAGECACHE_TAG_DIRTY); | |
6160 | xa_unlock_irq(&page->mapping->i_pages); | |
6161 | } | |
6162 | ||
6163 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
6164 | { | |
6165 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6166 | struct page *page = eb->pages[0]; | |
6167 | bool last; | |
6168 | ||
6169 | /* btree_clear_page_dirty() needs page locked */ | |
6170 | lock_page(page); | |
6171 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
6172 | eb->len); | |
6173 | if (last) | |
6174 | btree_clear_page_dirty(page); | |
6175 | unlock_page(page); | |
6176 | WARN_ON(atomic_read(&eb->refs) == 0); | |
6177 | } | |
6178 | ||
2b48966a | 6179 | void clear_extent_buffer_dirty(const struct extent_buffer *eb) |
d1310b2e | 6180 | { |
cc5e31a4 DS |
6181 | int i; |
6182 | int num_pages; | |
d1310b2e CM |
6183 | struct page *page; |
6184 | ||
0d27797e QW |
6185 | if (eb->fs_info->sectorsize < PAGE_SIZE) |
6186 | return clear_subpage_extent_buffer_dirty(eb); | |
6187 | ||
65ad0104 | 6188 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
6189 | |
6190 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 6191 | page = eb->pages[i]; |
b9473439 | 6192 | if (!PageDirty(page)) |
d2c3f4f6 | 6193 | continue; |
a61e6f29 | 6194 | lock_page(page); |
0d27797e | 6195 | btree_clear_page_dirty(page); |
bf0da8c1 | 6196 | ClearPageError(page); |
a61e6f29 | 6197 | unlock_page(page); |
d1310b2e | 6198 | } |
0b32f4bb | 6199 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 6200 | } |
d1310b2e | 6201 | |
abb57ef3 | 6202 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 6203 | { |
cc5e31a4 DS |
6204 | int i; |
6205 | int num_pages; | |
abb57ef3 | 6206 | bool was_dirty; |
d1310b2e | 6207 | |
0b32f4bb JB |
6208 | check_buffer_tree_ref(eb); |
6209 | ||
b9473439 | 6210 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 6211 | |
65ad0104 | 6212 | num_pages = num_extent_pages(eb); |
3083ee2e | 6213 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
6214 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
6215 | ||
0d27797e QW |
6216 | if (!was_dirty) { |
6217 | bool subpage = eb->fs_info->sectorsize < PAGE_SIZE; | |
51995c39 | 6218 | |
0d27797e QW |
6219 | /* |
6220 | * For subpage case, we can have other extent buffers in the | |
6221 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
6222 | * have to clear page dirty without subpage lock held. | |
6223 | * This can cause race where our page gets dirty cleared after | |
6224 | * we just set it. | |
6225 | * | |
6226 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
6227 | * its page for other reasons, we can use page lock to prevent | |
6228 | * the above race. | |
6229 | */ | |
6230 | if (subpage) | |
6231 | lock_page(eb->pages[0]); | |
6232 | for (i = 0; i < num_pages; i++) | |
6233 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
6234 | eb->start, eb->len); | |
6235 | if (subpage) | |
6236 | unlock_page(eb->pages[0]); | |
6237 | } | |
51995c39 LB |
6238 | #ifdef CONFIG_BTRFS_DEBUG |
6239 | for (i = 0; i < num_pages; i++) | |
6240 | ASSERT(PageDirty(eb->pages[i])); | |
6241 | #endif | |
6242 | ||
b9473439 | 6243 | return was_dirty; |
d1310b2e | 6244 | } |
d1310b2e | 6245 | |
69ba3927 | 6246 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 6247 | { |
251f2acc | 6248 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 6249 | struct page *page; |
cc5e31a4 | 6250 | int num_pages; |
251f2acc | 6251 | int i; |
1259ab75 | 6252 | |
b4ce94de | 6253 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6254 | num_pages = num_extent_pages(eb); |
1259ab75 | 6255 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6256 | page = eb->pages[i]; |
33958dc6 | 6257 | if (page) |
251f2acc QW |
6258 | btrfs_page_clear_uptodate(fs_info, page, |
6259 | eb->start, eb->len); | |
1259ab75 | 6260 | } |
1259ab75 CM |
6261 | } |
6262 | ||
09c25a8c | 6263 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 6264 | { |
251f2acc | 6265 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 6266 | struct page *page; |
cc5e31a4 | 6267 | int num_pages; |
251f2acc | 6268 | int i; |
d1310b2e | 6269 | |
0b32f4bb | 6270 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 6271 | num_pages = num_extent_pages(eb); |
d1310b2e | 6272 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6273 | page = eb->pages[i]; |
251f2acc | 6274 | btrfs_page_set_uptodate(fs_info, page, eb->start, eb->len); |
d1310b2e | 6275 | } |
d1310b2e | 6276 | } |
d1310b2e | 6277 | |
4012daf7 QW |
6278 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
6279 | int mirror_num) | |
6280 | { | |
6281 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6282 | struct extent_io_tree *io_tree; | |
6283 | struct page *page = eb->pages[0]; | |
390ed29b | 6284 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
4012daf7 QW |
6285 | int ret = 0; |
6286 | ||
6287 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
6288 | ASSERT(PagePrivate(page)); | |
6289 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; | |
6290 | ||
6291 | if (wait == WAIT_NONE) { | |
dc56219f GR |
6292 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1)) |
6293 | return -EAGAIN; | |
4012daf7 QW |
6294 | } else { |
6295 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6296 | if (ret < 0) | |
6297 | return ret; | |
6298 | } | |
6299 | ||
6300 | ret = 0; | |
6301 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
6302 | PageUptodate(page) || | |
6303 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
6304 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
6305 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1); | |
6306 | return ret; | |
6307 | } | |
6308 | ||
6309 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
6310 | eb->read_mirror = 0; | |
6311 | atomic_set(&eb->io_pages, 1); | |
6312 | check_buffer_tree_ref(eb); | |
6313 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); | |
6314 | ||
390ed29b QW |
6315 | ret = submit_extent_page(REQ_OP_READ | REQ_META, NULL, &bio_ctrl, |
6316 | page, eb->start, eb->len, | |
6317 | eb->start - page_offset(page), | |
6318 | end_bio_extent_readpage, mirror_num, 0, | |
4012daf7 QW |
6319 | true); |
6320 | if (ret) { | |
6321 | /* | |
6322 | * In the endio function, if we hit something wrong we will | |
6323 | * increase the io_pages, so here we need to decrease it for | |
6324 | * error path. | |
6325 | */ | |
6326 | atomic_dec(&eb->io_pages); | |
6327 | } | |
390ed29b | 6328 | if (bio_ctrl.bio) { |
4012daf7 QW |
6329 | int tmp; |
6330 | ||
390ed29b QW |
6331 | tmp = submit_one_bio(bio_ctrl.bio, mirror_num, 0); |
6332 | bio_ctrl.bio = NULL; | |
4012daf7 QW |
6333 | if (tmp < 0) |
6334 | return tmp; | |
6335 | } | |
6336 | if (ret || wait != WAIT_COMPLETE) | |
6337 | return ret; | |
6338 | ||
6339 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, EXTENT_LOCKED); | |
6340 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) | |
6341 | ret = -EIO; | |
6342 | return ret; | |
6343 | } | |
6344 | ||
c2ccfbc6 | 6345 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num) |
d1310b2e | 6346 | { |
cc5e31a4 | 6347 | int i; |
d1310b2e CM |
6348 | struct page *page; |
6349 | int err; | |
6350 | int ret = 0; | |
ce9adaa5 CM |
6351 | int locked_pages = 0; |
6352 | int all_uptodate = 1; | |
cc5e31a4 | 6353 | int num_pages; |
727011e0 | 6354 | unsigned long num_reads = 0; |
390ed29b | 6355 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
a86c12c7 | 6356 | |
b4ce94de | 6357 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
6358 | return 0; |
6359 | ||
4012daf7 QW |
6360 | if (eb->fs_info->sectorsize < PAGE_SIZE) |
6361 | return read_extent_buffer_subpage(eb, wait, mirror_num); | |
6362 | ||
65ad0104 | 6363 | num_pages = num_extent_pages(eb); |
8436ea91 | 6364 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6365 | page = eb->pages[i]; |
bb82ab88 | 6366 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
6367 | /* |
6368 | * WAIT_NONE is only utilized by readahead. If we can't | |
6369 | * acquire the lock atomically it means either the eb | |
6370 | * is being read out or under modification. | |
6371 | * Either way the eb will be or has been cached, | |
6372 | * readahead can exit safely. | |
6373 | */ | |
2db04966 | 6374 | if (!trylock_page(page)) |
ce9adaa5 | 6375 | goto unlock_exit; |
d1310b2e CM |
6376 | } else { |
6377 | lock_page(page); | |
6378 | } | |
ce9adaa5 | 6379 | locked_pages++; |
2571e739 LB |
6380 | } |
6381 | /* | |
6382 | * We need to firstly lock all pages to make sure that | |
6383 | * the uptodate bit of our pages won't be affected by | |
6384 | * clear_extent_buffer_uptodate(). | |
6385 | */ | |
8436ea91 | 6386 | for (i = 0; i < num_pages; i++) { |
2571e739 | 6387 | page = eb->pages[i]; |
727011e0 CM |
6388 | if (!PageUptodate(page)) { |
6389 | num_reads++; | |
ce9adaa5 | 6390 | all_uptodate = 0; |
727011e0 | 6391 | } |
ce9adaa5 | 6392 | } |
2571e739 | 6393 | |
ce9adaa5 | 6394 | if (all_uptodate) { |
8436ea91 | 6395 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
6396 | goto unlock_exit; |
6397 | } | |
6398 | ||
656f30db | 6399 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 6400 | eb->read_mirror = 0; |
0b32f4bb | 6401 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e BB |
6402 | /* |
6403 | * It is possible for releasepage to clear the TREE_REF bit before we | |
6404 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. | |
6405 | */ | |
6406 | check_buffer_tree_ref(eb); | |
8436ea91 | 6407 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6408 | page = eb->pages[i]; |
baf863b9 | 6409 | |
ce9adaa5 | 6410 | if (!PageUptodate(page)) { |
baf863b9 LB |
6411 | if (ret) { |
6412 | atomic_dec(&eb->io_pages); | |
6413 | unlock_page(page); | |
6414 | continue; | |
6415 | } | |
6416 | ||
f188591e | 6417 | ClearPageError(page); |
0420177c | 6418 | err = submit_extent_page(REQ_OP_READ | REQ_META, NULL, |
390ed29b QW |
6419 | &bio_ctrl, page, page_offset(page), |
6420 | PAGE_SIZE, 0, end_bio_extent_readpage, | |
6421 | mirror_num, 0, false); | |
baf863b9 | 6422 | if (err) { |
baf863b9 | 6423 | /* |
0420177c NB |
6424 | * We failed to submit the bio so it's the |
6425 | * caller's responsibility to perform cleanup | |
6426 | * i.e unlock page/set error bit. | |
baf863b9 | 6427 | */ |
0420177c NB |
6428 | ret = err; |
6429 | SetPageError(page); | |
6430 | unlock_page(page); | |
baf863b9 LB |
6431 | atomic_dec(&eb->io_pages); |
6432 | } | |
d1310b2e CM |
6433 | } else { |
6434 | unlock_page(page); | |
6435 | } | |
6436 | } | |
6437 | ||
390ed29b QW |
6438 | if (bio_ctrl.bio) { |
6439 | err = submit_one_bio(bio_ctrl.bio, mirror_num, bio_ctrl.bio_flags); | |
6440 | bio_ctrl.bio = NULL; | |
79787eaa JM |
6441 | if (err) |
6442 | return err; | |
355808c2 | 6443 | } |
a86c12c7 | 6444 | |
bb82ab88 | 6445 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 6446 | return ret; |
d397712b | 6447 | |
8436ea91 | 6448 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 6449 | page = eb->pages[i]; |
d1310b2e | 6450 | wait_on_page_locked(page); |
d397712b | 6451 | if (!PageUptodate(page)) |
d1310b2e | 6452 | ret = -EIO; |
d1310b2e | 6453 | } |
d397712b | 6454 | |
d1310b2e | 6455 | return ret; |
ce9adaa5 CM |
6456 | |
6457 | unlock_exit: | |
d397712b | 6458 | while (locked_pages > 0) { |
ce9adaa5 | 6459 | locked_pages--; |
8436ea91 JB |
6460 | page = eb->pages[locked_pages]; |
6461 | unlock_page(page); | |
ce9adaa5 CM |
6462 | } |
6463 | return ret; | |
d1310b2e | 6464 | } |
d1310b2e | 6465 | |
f98b6215 QW |
6466 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
6467 | unsigned long len) | |
6468 | { | |
6469 | btrfs_warn(eb->fs_info, | |
6470 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
6471 | eb->start, eb->len, start, len); | |
6472 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
6473 | ||
6474 | return true; | |
6475 | } | |
6476 | ||
6477 | /* | |
6478 | * Check if the [start, start + len) range is valid before reading/writing | |
6479 | * the eb. | |
6480 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
6481 | * | |
6482 | * Caller should not touch the dst/src memory if this function returns error. | |
6483 | */ | |
6484 | static inline int check_eb_range(const struct extent_buffer *eb, | |
6485 | unsigned long start, unsigned long len) | |
6486 | { | |
6487 | unsigned long offset; | |
6488 | ||
6489 | /* start, start + len should not go beyond eb->len nor overflow */ | |
6490 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
6491 | return report_eb_range(eb, start, len); | |
6492 | ||
6493 | return false; | |
6494 | } | |
6495 | ||
1cbb1f45 JM |
6496 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
6497 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6498 | { |
6499 | size_t cur; | |
6500 | size_t offset; | |
6501 | struct page *page; | |
6502 | char *kaddr; | |
6503 | char *dst = (char *)dstv; | |
884b07d0 | 6504 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6505 | |
f98b6215 | 6506 | if (check_eb_range(eb, start, len)) |
f716abd5 | 6507 | return; |
d1310b2e | 6508 | |
884b07d0 | 6509 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6510 | |
d397712b | 6511 | while (len > 0) { |
fb85fc9a | 6512 | page = eb->pages[i]; |
d1310b2e | 6513 | |
09cbfeaf | 6514 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 6515 | kaddr = page_address(page); |
d1310b2e | 6516 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
6517 | |
6518 | dst += cur; | |
6519 | len -= cur; | |
6520 | offset = 0; | |
6521 | i++; | |
6522 | } | |
6523 | } | |
d1310b2e | 6524 | |
a48b73ec JB |
6525 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
6526 | void __user *dstv, | |
6527 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
6528 | { |
6529 | size_t cur; | |
6530 | size_t offset; | |
6531 | struct page *page; | |
6532 | char *kaddr; | |
6533 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 6534 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
6535 | int ret = 0; |
6536 | ||
6537 | WARN_ON(start > eb->len); | |
6538 | WARN_ON(start + len > eb->start + eb->len); | |
6539 | ||
884b07d0 | 6540 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
6541 | |
6542 | while (len > 0) { | |
fb85fc9a | 6543 | page = eb->pages[i]; |
550ac1d8 | 6544 | |
09cbfeaf | 6545 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 6546 | kaddr = page_address(page); |
a48b73ec | 6547 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
6548 | ret = -EFAULT; |
6549 | break; | |
6550 | } | |
6551 | ||
6552 | dst += cur; | |
6553 | len -= cur; | |
6554 | offset = 0; | |
6555 | i++; | |
6556 | } | |
6557 | ||
6558 | return ret; | |
6559 | } | |
6560 | ||
1cbb1f45 JM |
6561 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
6562 | unsigned long start, unsigned long len) | |
d1310b2e CM |
6563 | { |
6564 | size_t cur; | |
6565 | size_t offset; | |
6566 | struct page *page; | |
6567 | char *kaddr; | |
6568 | char *ptr = (char *)ptrv; | |
884b07d0 | 6569 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
6570 | int ret = 0; |
6571 | ||
f98b6215 QW |
6572 | if (check_eb_range(eb, start, len)) |
6573 | return -EINVAL; | |
d1310b2e | 6574 | |
884b07d0 | 6575 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6576 | |
d397712b | 6577 | while (len > 0) { |
fb85fc9a | 6578 | page = eb->pages[i]; |
d1310b2e | 6579 | |
09cbfeaf | 6580 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 6581 | |
a6591715 | 6582 | kaddr = page_address(page); |
d1310b2e | 6583 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
6584 | if (ret) |
6585 | break; | |
6586 | ||
6587 | ptr += cur; | |
6588 | len -= cur; | |
6589 | offset = 0; | |
6590 | i++; | |
6591 | } | |
6592 | return ret; | |
6593 | } | |
d1310b2e | 6594 | |
b8f95771 QW |
6595 | /* |
6596 | * Check that the extent buffer is uptodate. | |
6597 | * | |
6598 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
6599 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
6600 | */ | |
6601 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
6602 | struct page *page) | |
6603 | { | |
6604 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
6605 | ||
6606 | if (fs_info->sectorsize < PAGE_SIZE) { | |
6607 | bool uptodate; | |
6608 | ||
6609 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
6610 | eb->start, eb->len); | |
6611 | WARN_ON(!uptodate); | |
6612 | } else { | |
6613 | WARN_ON(!PageUptodate(page)); | |
6614 | } | |
6615 | } | |
6616 | ||
2b48966a | 6617 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
6618 | const void *srcv) |
6619 | { | |
6620 | char *kaddr; | |
6621 | ||
b8f95771 | 6622 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6623 | kaddr = page_address(eb->pages[0]) + |
6624 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
6625 | chunk_tree_uuid)); | |
6626 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6627 | } |
6628 | ||
2b48966a | 6629 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
6630 | { |
6631 | char *kaddr; | |
6632 | ||
b8f95771 | 6633 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
6634 | kaddr = page_address(eb->pages[0]) + |
6635 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
6636 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
6637 | } |
6638 | ||
2b48966a | 6639 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
6640 | unsigned long start, unsigned long len) |
6641 | { | |
6642 | size_t cur; | |
6643 | size_t offset; | |
6644 | struct page *page; | |
6645 | char *kaddr; | |
6646 | char *src = (char *)srcv; | |
884b07d0 | 6647 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6648 | |
d3575156 NA |
6649 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
6650 | ||
f98b6215 QW |
6651 | if (check_eb_range(eb, start, len)) |
6652 | return; | |
d1310b2e | 6653 | |
884b07d0 | 6654 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6655 | |
d397712b | 6656 | while (len > 0) { |
fb85fc9a | 6657 | page = eb->pages[i]; |
b8f95771 | 6658 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6659 | |
09cbfeaf | 6660 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6661 | kaddr = page_address(page); |
d1310b2e | 6662 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
6663 | |
6664 | src += cur; | |
6665 | len -= cur; | |
6666 | offset = 0; | |
6667 | i++; | |
6668 | } | |
6669 | } | |
d1310b2e | 6670 | |
2b48966a | 6671 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 6672 | unsigned long len) |
d1310b2e CM |
6673 | { |
6674 | size_t cur; | |
6675 | size_t offset; | |
6676 | struct page *page; | |
6677 | char *kaddr; | |
884b07d0 | 6678 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 6679 | |
f98b6215 QW |
6680 | if (check_eb_range(eb, start, len)) |
6681 | return; | |
d1310b2e | 6682 | |
884b07d0 | 6683 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 6684 | |
d397712b | 6685 | while (len > 0) { |
fb85fc9a | 6686 | page = eb->pages[i]; |
b8f95771 | 6687 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 6688 | |
09cbfeaf | 6689 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 6690 | kaddr = page_address(page); |
b159fa28 | 6691 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
6692 | |
6693 | len -= cur; | |
6694 | offset = 0; | |
6695 | i++; | |
6696 | } | |
6697 | } | |
d1310b2e | 6698 | |
2b48966a DS |
6699 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
6700 | const struct extent_buffer *src) | |
58e8012c DS |
6701 | { |
6702 | int i; | |
cc5e31a4 | 6703 | int num_pages; |
58e8012c DS |
6704 | |
6705 | ASSERT(dst->len == src->len); | |
6706 | ||
884b07d0 QW |
6707 | if (dst->fs_info->sectorsize == PAGE_SIZE) { |
6708 | num_pages = num_extent_pages(dst); | |
6709 | for (i = 0; i < num_pages; i++) | |
6710 | copy_page(page_address(dst->pages[i]), | |
6711 | page_address(src->pages[i])); | |
6712 | } else { | |
6713 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
6714 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
6715 | ||
6716 | ASSERT(src->fs_info->sectorsize < PAGE_SIZE); | |
6717 | memcpy(page_address(dst->pages[0]) + dst_offset, | |
6718 | page_address(src->pages[0]) + src_offset, | |
6719 | src->len); | |
6720 | } | |
58e8012c DS |
6721 | } |
6722 | ||
2b48966a DS |
6723 | void copy_extent_buffer(const struct extent_buffer *dst, |
6724 | const struct extent_buffer *src, | |
d1310b2e CM |
6725 | unsigned long dst_offset, unsigned long src_offset, |
6726 | unsigned long len) | |
6727 | { | |
6728 | u64 dst_len = dst->len; | |
6729 | size_t cur; | |
6730 | size_t offset; | |
6731 | struct page *page; | |
6732 | char *kaddr; | |
884b07d0 | 6733 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 6734 | |
f98b6215 QW |
6735 | if (check_eb_range(dst, dst_offset, len) || |
6736 | check_eb_range(src, src_offset, len)) | |
6737 | return; | |
6738 | ||
d1310b2e CM |
6739 | WARN_ON(src->len != dst_len); |
6740 | ||
884b07d0 | 6741 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 6742 | |
d397712b | 6743 | while (len > 0) { |
fb85fc9a | 6744 | page = dst->pages[i]; |
b8f95771 | 6745 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 6746 | |
09cbfeaf | 6747 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 6748 | |
a6591715 | 6749 | kaddr = page_address(page); |
d1310b2e | 6750 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
6751 | |
6752 | src_offset += cur; | |
6753 | len -= cur; | |
6754 | offset = 0; | |
6755 | i++; | |
6756 | } | |
6757 | } | |
d1310b2e | 6758 | |
3e1e8bb7 OS |
6759 | /* |
6760 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
6761 | * given bit number | |
6762 | * @eb: the extent buffer | |
6763 | * @start: offset of the bitmap item in the extent buffer | |
6764 | * @nr: bit number | |
6765 | * @page_index: return index of the page in the extent buffer that contains the | |
6766 | * given bit number | |
6767 | * @page_offset: return offset into the page given by page_index | |
6768 | * | |
6769 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
6770 | * contains a given bit. | |
6771 | */ | |
2b48966a | 6772 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
6773 | unsigned long start, unsigned long nr, |
6774 | unsigned long *page_index, | |
6775 | size_t *page_offset) | |
6776 | { | |
3e1e8bb7 OS |
6777 | size_t byte_offset = BIT_BYTE(nr); |
6778 | size_t offset; | |
6779 | ||
6780 | /* | |
6781 | * The byte we want is the offset of the extent buffer + the offset of | |
6782 | * the bitmap item in the extent buffer + the offset of the byte in the | |
6783 | * bitmap item. | |
6784 | */ | |
884b07d0 | 6785 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 6786 | |
09cbfeaf | 6787 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 6788 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
6789 | } |
6790 | ||
6791 | /** | |
6792 | * extent_buffer_test_bit - determine whether a bit in a bitmap item is set | |
6793 | * @eb: the extent buffer | |
6794 | * @start: offset of the bitmap item in the extent buffer | |
6795 | * @nr: bit number to test | |
6796 | */ | |
2b48966a | 6797 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
6798 | unsigned long nr) |
6799 | { | |
2fe1d551 | 6800 | u8 *kaddr; |
3e1e8bb7 OS |
6801 | struct page *page; |
6802 | unsigned long i; | |
6803 | size_t offset; | |
6804 | ||
6805 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
6806 | page = eb->pages[i]; | |
b8f95771 | 6807 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
6808 | kaddr = page_address(page); |
6809 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
6810 | } | |
6811 | ||
6812 | /** | |
6813 | * extent_buffer_bitmap_set - set an area of a bitmap | |
6814 | * @eb: the extent buffer | |
6815 | * @start: offset of the bitmap item in the extent buffer | |
6816 | * @pos: bit number of the first bit | |
6817 | * @len: number of bits to set | |
6818 | */ | |
2b48966a | 6819 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
6820 | unsigned long pos, unsigned long len) |
6821 | { | |
2fe1d551 | 6822 | u8 *kaddr; |
3e1e8bb7 OS |
6823 | struct page *page; |
6824 | unsigned long i; | |
6825 | size_t offset; | |
6826 | const unsigned int size = pos + len; | |
6827 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 6828 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
6829 | |
6830 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
6831 | page = eb->pages[i]; | |
b8f95771 | 6832 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
6833 | kaddr = page_address(page); |
6834 | ||
6835 | while (len >= bits_to_set) { | |
6836 | kaddr[offset] |= mask_to_set; | |
6837 | len -= bits_to_set; | |
6838 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 6839 | mask_to_set = ~0; |
09cbfeaf | 6840 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
6841 | offset = 0; |
6842 | page = eb->pages[++i]; | |
b8f95771 | 6843 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
6844 | kaddr = page_address(page); |
6845 | } | |
6846 | } | |
6847 | if (len) { | |
6848 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
6849 | kaddr[offset] |= mask_to_set; | |
6850 | } | |
6851 | } | |
6852 | ||
6853 | ||
6854 | /** | |
6855 | * extent_buffer_bitmap_clear - clear an area of a bitmap | |
6856 | * @eb: the extent buffer | |
6857 | * @start: offset of the bitmap item in the extent buffer | |
6858 | * @pos: bit number of the first bit | |
6859 | * @len: number of bits to clear | |
6860 | */ | |
2b48966a DS |
6861 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
6862 | unsigned long start, unsigned long pos, | |
6863 | unsigned long len) | |
3e1e8bb7 | 6864 | { |
2fe1d551 | 6865 | u8 *kaddr; |
3e1e8bb7 OS |
6866 | struct page *page; |
6867 | unsigned long i; | |
6868 | size_t offset; | |
6869 | const unsigned int size = pos + len; | |
6870 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 6871 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
6872 | |
6873 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
6874 | page = eb->pages[i]; | |
b8f95771 | 6875 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
6876 | kaddr = page_address(page); |
6877 | ||
6878 | while (len >= bits_to_clear) { | |
6879 | kaddr[offset] &= ~mask_to_clear; | |
6880 | len -= bits_to_clear; | |
6881 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 6882 | mask_to_clear = ~0; |
09cbfeaf | 6883 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
6884 | offset = 0; |
6885 | page = eb->pages[++i]; | |
b8f95771 | 6886 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
6887 | kaddr = page_address(page); |
6888 | } | |
6889 | } | |
6890 | if (len) { | |
6891 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
6892 | kaddr[offset] &= ~mask_to_clear; | |
6893 | } | |
6894 | } | |
6895 | ||
3387206f ST |
6896 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
6897 | { | |
6898 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
6899 | return distance < len; | |
6900 | } | |
6901 | ||
d1310b2e CM |
6902 | static void copy_pages(struct page *dst_page, struct page *src_page, |
6903 | unsigned long dst_off, unsigned long src_off, | |
6904 | unsigned long len) | |
6905 | { | |
a6591715 | 6906 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 6907 | char *src_kaddr; |
727011e0 | 6908 | int must_memmove = 0; |
d1310b2e | 6909 | |
3387206f | 6910 | if (dst_page != src_page) { |
a6591715 | 6911 | src_kaddr = page_address(src_page); |
3387206f | 6912 | } else { |
d1310b2e | 6913 | src_kaddr = dst_kaddr; |
727011e0 CM |
6914 | if (areas_overlap(src_off, dst_off, len)) |
6915 | must_memmove = 1; | |
3387206f | 6916 | } |
d1310b2e | 6917 | |
727011e0 CM |
6918 | if (must_memmove) |
6919 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
6920 | else | |
6921 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
6922 | } |
6923 | ||
2b48966a DS |
6924 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
6925 | unsigned long dst_offset, unsigned long src_offset, | |
6926 | unsigned long len) | |
d1310b2e CM |
6927 | { |
6928 | size_t cur; | |
6929 | size_t dst_off_in_page; | |
6930 | size_t src_off_in_page; | |
d1310b2e CM |
6931 | unsigned long dst_i; |
6932 | unsigned long src_i; | |
6933 | ||
f98b6215 QW |
6934 | if (check_eb_range(dst, dst_offset, len) || |
6935 | check_eb_range(dst, src_offset, len)) | |
6936 | return; | |
d1310b2e | 6937 | |
d397712b | 6938 | while (len > 0) { |
884b07d0 QW |
6939 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
6940 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 6941 | |
884b07d0 QW |
6942 | dst_i = get_eb_page_index(dst_offset); |
6943 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 6944 | |
09cbfeaf | 6945 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
6946 | src_off_in_page)); |
6947 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 6948 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 6949 | |
fb85fc9a | 6950 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6951 | dst_off_in_page, src_off_in_page, cur); |
6952 | ||
6953 | src_offset += cur; | |
6954 | dst_offset += cur; | |
6955 | len -= cur; | |
6956 | } | |
6957 | } | |
d1310b2e | 6958 | |
2b48966a DS |
6959 | void memmove_extent_buffer(const struct extent_buffer *dst, |
6960 | unsigned long dst_offset, unsigned long src_offset, | |
6961 | unsigned long len) | |
d1310b2e CM |
6962 | { |
6963 | size_t cur; | |
6964 | size_t dst_off_in_page; | |
6965 | size_t src_off_in_page; | |
6966 | unsigned long dst_end = dst_offset + len - 1; | |
6967 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
6968 | unsigned long dst_i; |
6969 | unsigned long src_i; | |
6970 | ||
f98b6215 QW |
6971 | if (check_eb_range(dst, dst_offset, len) || |
6972 | check_eb_range(dst, src_offset, len)) | |
6973 | return; | |
727011e0 | 6974 | if (dst_offset < src_offset) { |
d1310b2e CM |
6975 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
6976 | return; | |
6977 | } | |
d397712b | 6978 | while (len > 0) { |
884b07d0 QW |
6979 | dst_i = get_eb_page_index(dst_end); |
6980 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 6981 | |
884b07d0 QW |
6982 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
6983 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
6984 | |
6985 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
6986 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 6987 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
6988 | dst_off_in_page - cur + 1, |
6989 | src_off_in_page - cur + 1, cur); | |
6990 | ||
6991 | dst_end -= cur; | |
6992 | src_end -= cur; | |
6993 | len -= cur; | |
6994 | } | |
6995 | } | |
6af118ce | 6996 | |
d1e86e3f QW |
6997 | static struct extent_buffer *get_next_extent_buffer( |
6998 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
6999 | { | |
7000 | struct extent_buffer *gang[BTRFS_SUBPAGE_BITMAP_SIZE]; | |
7001 | struct extent_buffer *found = NULL; | |
7002 | u64 page_start = page_offset(page); | |
7003 | int ret; | |
7004 | int i; | |
7005 | ||
7006 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
7007 | ASSERT(PAGE_SIZE / fs_info->nodesize <= BTRFS_SUBPAGE_BITMAP_SIZE); | |
7008 | lockdep_assert_held(&fs_info->buffer_lock); | |
7009 | ||
7010 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, (void **)gang, | |
7011 | bytenr >> fs_info->sectorsize_bits, | |
7012 | PAGE_SIZE / fs_info->nodesize); | |
7013 | for (i = 0; i < ret; i++) { | |
7014 | /* Already beyond page end */ | |
7015 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
7016 | break; | |
7017 | /* Found one */ | |
7018 | if (gang[i]->start >= bytenr) { | |
7019 | found = gang[i]; | |
7020 | break; | |
7021 | } | |
7022 | } | |
7023 | return found; | |
7024 | } | |
7025 | ||
7026 | static int try_release_subpage_extent_buffer(struct page *page) | |
7027 | { | |
7028 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
7029 | u64 cur = page_offset(page); | |
7030 | const u64 end = page_offset(page) + PAGE_SIZE; | |
7031 | int ret; | |
7032 | ||
7033 | while (cur < end) { | |
7034 | struct extent_buffer *eb = NULL; | |
7035 | ||
7036 | /* | |
7037 | * Unlike try_release_extent_buffer() which uses page->private | |
7038 | * to grab buffer, for subpage case we rely on radix tree, thus | |
7039 | * we need to ensure radix tree consistency. | |
7040 | * | |
7041 | * We also want an atomic snapshot of the radix tree, thus go | |
7042 | * with spinlock rather than RCU. | |
7043 | */ | |
7044 | spin_lock(&fs_info->buffer_lock); | |
7045 | eb = get_next_extent_buffer(fs_info, page, cur); | |
7046 | if (!eb) { | |
7047 | /* No more eb in the page range after or at cur */ | |
7048 | spin_unlock(&fs_info->buffer_lock); | |
7049 | break; | |
7050 | } | |
7051 | cur = eb->start + eb->len; | |
7052 | ||
7053 | /* | |
7054 | * The same as try_release_extent_buffer(), to ensure the eb | |
7055 | * won't disappear out from under us. | |
7056 | */ | |
7057 | spin_lock(&eb->refs_lock); | |
7058 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
7059 | spin_unlock(&eb->refs_lock); | |
7060 | spin_unlock(&fs_info->buffer_lock); | |
7061 | break; | |
7062 | } | |
7063 | spin_unlock(&fs_info->buffer_lock); | |
7064 | ||
7065 | /* | |
7066 | * If tree ref isn't set then we know the ref on this eb is a | |
7067 | * real ref, so just return, this eb will likely be freed soon | |
7068 | * anyway. | |
7069 | */ | |
7070 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
7071 | spin_unlock(&eb->refs_lock); | |
7072 | break; | |
7073 | } | |
7074 | ||
7075 | /* | |
7076 | * Here we don't care about the return value, we will always | |
7077 | * check the page private at the end. And | |
7078 | * release_extent_buffer() will release the refs_lock. | |
7079 | */ | |
7080 | release_extent_buffer(eb); | |
7081 | } | |
7082 | /* | |
7083 | * Finally to check if we have cleared page private, as if we have | |
7084 | * released all ebs in the page, the page private should be cleared now. | |
7085 | */ | |
7086 | spin_lock(&page->mapping->private_lock); | |
7087 | if (!PagePrivate(page)) | |
7088 | ret = 1; | |
7089 | else | |
7090 | ret = 0; | |
7091 | spin_unlock(&page->mapping->private_lock); | |
7092 | return ret; | |
7093 | ||
7094 | } | |
7095 | ||
f7a52a40 | 7096 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 7097 | { |
6af118ce | 7098 | struct extent_buffer *eb; |
6af118ce | 7099 | |
d1e86e3f QW |
7100 | if (btrfs_sb(page->mapping->host->i_sb)->sectorsize < PAGE_SIZE) |
7101 | return try_release_subpage_extent_buffer(page); | |
7102 | ||
3083ee2e | 7103 | /* |
d1e86e3f QW |
7104 | * We need to make sure nobody is changing page->private, as we rely on |
7105 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
7106 | */ |
7107 | spin_lock(&page->mapping->private_lock); | |
7108 | if (!PagePrivate(page)) { | |
7109 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 7110 | return 1; |
45f49bce | 7111 | } |
6af118ce | 7112 | |
3083ee2e JB |
7113 | eb = (struct extent_buffer *)page->private; |
7114 | BUG_ON(!eb); | |
19fe0a8b MX |
7115 | |
7116 | /* | |
3083ee2e JB |
7117 | * This is a little awful but should be ok, we need to make sure that |
7118 | * the eb doesn't disappear out from under us while we're looking at | |
7119 | * this page. | |
19fe0a8b | 7120 | */ |
3083ee2e | 7121 | spin_lock(&eb->refs_lock); |
0b32f4bb | 7122 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
7123 | spin_unlock(&eb->refs_lock); |
7124 | spin_unlock(&page->mapping->private_lock); | |
7125 | return 0; | |
b9473439 | 7126 | } |
3083ee2e | 7127 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 7128 | |
19fe0a8b | 7129 | /* |
3083ee2e JB |
7130 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
7131 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 7132 | */ |
3083ee2e JB |
7133 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
7134 | spin_unlock(&eb->refs_lock); | |
7135 | return 0; | |
b9473439 | 7136 | } |
19fe0a8b | 7137 | |
f7a52a40 | 7138 | return release_extent_buffer(eb); |
6af118ce | 7139 | } |
bfb484d9 JB |
7140 | |
7141 | /* | |
7142 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
7143 | * @fs_info: the fs_info | |
7144 | * @bytenr: bytenr to read | |
3fbaf258 | 7145 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 7146 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 7147 | * @level: level for the eb |
bfb484d9 JB |
7148 | * |
7149 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
7150 | * normal uptodate check of the eb, without checking the generation. If we have | |
7151 | * to read the block we will not block on anything. | |
7152 | */ | |
7153 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 7154 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 JB |
7155 | { |
7156 | struct extent_buffer *eb; | |
7157 | int ret; | |
7158 | ||
3fbaf258 | 7159 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
7160 | if (IS_ERR(eb)) |
7161 | return; | |
7162 | ||
7163 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
7164 | free_extent_buffer(eb); | |
7165 | return; | |
7166 | } | |
7167 | ||
7168 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0); | |
7169 | if (ret < 0) | |
7170 | free_extent_buffer_stale(eb); | |
7171 | else | |
7172 | free_extent_buffer(eb); | |
7173 | } | |
7174 | ||
7175 | /* | |
7176 | * btrfs_readahead_node_child - readahead a node's child block | |
7177 | * @node: parent node we're reading from | |
7178 | * @slot: slot in the parent node for the child we want to read | |
7179 | * | |
7180 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
7181 | * the slot in the node provided. | |
7182 | */ | |
7183 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
7184 | { | |
7185 | btrfs_readahead_tree_block(node->fs_info, | |
7186 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
7187 | btrfs_header_owner(node), |
7188 | btrfs_node_ptr_generation(node, slot), | |
7189 | btrfs_header_level(node) - 1); | |
bfb484d9 | 7190 | } |