Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
c1d7c514 | 2 | |
d1310b2e CM |
3 | #include <linux/bitops.h> |
4 | #include <linux/slab.h> | |
5 | #include <linux/bio.h> | |
6 | #include <linux/mm.h> | |
d1310b2e CM |
7 | #include <linux/pagemap.h> |
8 | #include <linux/page-flags.h> | |
395cb57e | 9 | #include <linux/sched/mm.h> |
d1310b2e CM |
10 | #include <linux/spinlock.h> |
11 | #include <linux/blkdev.h> | |
12 | #include <linux/swap.h> | |
d1310b2e CM |
13 | #include <linux/writeback.h> |
14 | #include <linux/pagevec.h> | |
268bb0ce | 15 | #include <linux/prefetch.h> |
14605409 | 16 | #include <linux/fsverity.h> |
cea62800 | 17 | #include "misc.h" |
d1310b2e | 18 | #include "extent_io.h" |
9c7d3a54 | 19 | #include "extent-io-tree.h" |
d1310b2e | 20 | #include "extent_map.h" |
902b22f3 DW |
21 | #include "ctree.h" |
22 | #include "btrfs_inode.h" | |
103c1972 | 23 | #include "bio.h" |
21adbd5c | 24 | #include "check-integrity.h" |
0b32f4bb | 25 | #include "locking.h" |
606686ee | 26 | #include "rcu-string.h" |
fe09e16c | 27 | #include "backref.h" |
6af49dbd | 28 | #include "disk-io.h" |
760f991f | 29 | #include "subpage.h" |
d3575156 | 30 | #include "zoned.h" |
0bc09ca1 | 31 | #include "block-group.h" |
2a5232a8 | 32 | #include "compression.h" |
ec8eb376 | 33 | #include "fs.h" |
07e81dc9 | 34 | #include "accessors.h" |
7c8ede16 | 35 | #include "file-item.h" |
af142b6f | 36 | #include "file.h" |
77407dc0 | 37 | #include "dev-replace.h" |
7f0add25 | 38 | #include "super.h" |
98c8d683 | 39 | #include "transaction.h" |
d1310b2e | 40 | |
d1310b2e CM |
41 | static struct kmem_cache *extent_buffer_cache; |
42 | ||
6d49ba1b | 43 | #ifdef CONFIG_BTRFS_DEBUG |
a40246e8 JB |
44 | static inline void btrfs_leak_debug_add_eb(struct extent_buffer *eb) |
45 | { | |
46 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
47 | unsigned long flags; | |
48 | ||
49 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); | |
50 | list_add(&eb->leak_list, &fs_info->allocated_ebs); | |
51 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); | |
52 | } | |
53 | ||
a40246e8 JB |
54 | static inline void btrfs_leak_debug_del_eb(struct extent_buffer *eb) |
55 | { | |
56 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
57 | unsigned long flags; | |
58 | ||
59 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); | |
60 | list_del(&eb->leak_list); | |
61 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); | |
6d49ba1b ES |
62 | } |
63 | ||
3fd63727 | 64 | void btrfs_extent_buffer_leak_debug_check(struct btrfs_fs_info *fs_info) |
6d49ba1b | 65 | { |
6d49ba1b | 66 | struct extent_buffer *eb; |
3fd63727 | 67 | unsigned long flags; |
6d49ba1b | 68 | |
8c38938c JB |
69 | /* |
70 | * If we didn't get into open_ctree our allocated_ebs will not be | |
71 | * initialized, so just skip this. | |
72 | */ | |
73 | if (!fs_info->allocated_ebs.next) | |
74 | return; | |
75 | ||
b95b78e6 | 76 | WARN_ON(!list_empty(&fs_info->allocated_ebs)); |
3fd63727 JB |
77 | spin_lock_irqsave(&fs_info->eb_leak_lock, flags); |
78 | while (!list_empty(&fs_info->allocated_ebs)) { | |
79 | eb = list_first_entry(&fs_info->allocated_ebs, | |
80 | struct extent_buffer, leak_list); | |
8c38938c JB |
81 | pr_err( |
82 | "BTRFS: buffer leak start %llu len %lu refs %d bflags %lu owner %llu\n", | |
83 | eb->start, eb->len, atomic_read(&eb->refs), eb->bflags, | |
84 | btrfs_header_owner(eb)); | |
33ca832f JB |
85 | list_del(&eb->leak_list); |
86 | kmem_cache_free(extent_buffer_cache, eb); | |
87 | } | |
3fd63727 | 88 | spin_unlock_irqrestore(&fs_info->eb_leak_lock, flags); |
33ca832f | 89 | } |
6d49ba1b | 90 | #else |
a40246e8 | 91 | #define btrfs_leak_debug_add_eb(eb) do {} while (0) |
a40246e8 | 92 | #define btrfs_leak_debug_del_eb(eb) do {} while (0) |
4bef0848 | 93 | #endif |
d1310b2e | 94 | |
7aab8b32 CH |
95 | /* |
96 | * Structure to record info about the bio being assembled, and other info like | |
97 | * how many bytes are there before stripe/ordered extent boundary. | |
98 | */ | |
99 | struct btrfs_bio_ctrl { | |
100 | struct bio *bio; | |
722c82ac | 101 | int mirror_num; |
0f07003b | 102 | enum btrfs_compression_type compress_type; |
7aab8b32 | 103 | u32 len_to_oe_boundary; |
c000bc04 | 104 | blk_opf_t opf; |
5467abba | 105 | btrfs_bio_end_io_t end_io_func; |
72b505dc | 106 | struct writeback_control *wbc; |
7aab8b32 | 107 | |
1d854e4f QW |
108 | /* |
109 | * This is for metadata read, to provide the extra needed verification | |
110 | * info. This has to be provided for submit_one_bio(), as | |
111 | * submit_one_bio() can submit a bio if it ends at stripe boundary. If | |
112 | * no such parent_check is provided, the metadata can hit false alert at | |
113 | * endio time. | |
114 | */ | |
115 | struct btrfs_tree_parent_check *parent_check; | |
116 | ||
ee5f017d DS |
117 | /* |
118 | * Tell writepage not to lock the state bits for this range, it still | |
119 | * does the unlocking. | |
771ed689 | 120 | */ |
8ec8519b | 121 | bool extent_locked; |
d1310b2e CM |
122 | }; |
123 | ||
722c82ac | 124 | static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl) |
bb58eb9e | 125 | { |
722c82ac | 126 | struct bio *bio; |
7aa51232 | 127 | struct bio_vec *bv; |
35a8d7da | 128 | struct inode *inode; |
722c82ac CH |
129 | int mirror_num; |
130 | ||
131 | if (!bio_ctrl->bio) | |
132 | return; | |
bb58eb9e | 133 | |
722c82ac | 134 | bio = bio_ctrl->bio; |
7aa51232 | 135 | bv = bio_first_bvec_all(bio); |
35a8d7da | 136 | inode = bv->bv_page->mapping->host; |
722c82ac | 137 | mirror_num = bio_ctrl->mirror_num; |
bb58eb9e | 138 | |
e0eefe07 QW |
139 | /* Caller should ensure the bio has at least some range added */ |
140 | ASSERT(bio->bi_iter.bi_size); | |
c9583ada | 141 | |
35a8d7da | 142 | if (!is_data_inode(inode)) { |
1d854e4f QW |
143 | if (btrfs_op(bio) != BTRFS_MAP_WRITE) { |
144 | /* | |
145 | * For metadata read, we should have the parent_check, | |
146 | * and copy it to bbio for metadata verification. | |
147 | */ | |
148 | ASSERT(bio_ctrl->parent_check); | |
149 | memcpy(&btrfs_bio(bio)->parent_check, | |
150 | bio_ctrl->parent_check, | |
151 | sizeof(struct btrfs_tree_parent_check)); | |
152 | } | |
35a8d7da | 153 | bio->bi_opf |= REQ_META; |
1d854e4f | 154 | } |
390ed29b | 155 | |
35a8d7da CH |
156 | if (btrfs_op(bio) == BTRFS_MAP_READ && |
157 | bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) | |
544fe4a9 | 158 | btrfs_submit_compressed_read(bio, mirror_num); |
35a8d7da CH |
159 | else |
160 | btrfs_submit_bio(bio, mirror_num); | |
161 | ||
917f32a2 | 162 | /* The bio is owned by the end_io handler now */ |
722c82ac | 163 | bio_ctrl->bio = NULL; |
3065976b QW |
164 | } |
165 | ||
f4340622 | 166 | /* |
ee5f017d | 167 | * Submit or fail the current bio in the bio_ctrl structure. |
f4340622 | 168 | */ |
ee5f017d | 169 | static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret) |
bb58eb9e | 170 | { |
ee5f017d | 171 | struct bio *bio = bio_ctrl->bio; |
bb58eb9e | 172 | |
9845e5dd CH |
173 | if (!bio) |
174 | return; | |
175 | ||
176 | if (ret) { | |
177 | ASSERT(ret < 0); | |
917f32a2 CH |
178 | btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); |
179 | /* The bio is owned by the end_io handler now */ | |
ee5f017d | 180 | bio_ctrl->bio = NULL; |
9845e5dd | 181 | } else { |
ee5f017d | 182 | submit_one_bio(bio_ctrl); |
bb58eb9e QW |
183 | } |
184 | } | |
e2932ee0 | 185 | |
a62a3bd9 JB |
186 | int __init extent_buffer_init_cachep(void) |
187 | { | |
837e1972 | 188 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 189 | sizeof(struct extent_buffer), 0, |
fba4b697 | 190 | SLAB_MEM_SPREAD, NULL); |
a62a3bd9 | 191 | if (!extent_buffer_cache) |
6f0d04f8 | 192 | return -ENOMEM; |
b208c2f7 | 193 | |
d1310b2e | 194 | return 0; |
d1310b2e CM |
195 | } |
196 | ||
a62a3bd9 | 197 | void __cold extent_buffer_free_cachep(void) |
d1310b2e | 198 | { |
8c0a8537 KS |
199 | /* |
200 | * Make sure all delayed rcu free are flushed before we | |
201 | * destroy caches. | |
202 | */ | |
203 | rcu_barrier(); | |
5598e900 | 204 | kmem_cache_destroy(extent_buffer_cache); |
d1310b2e CM |
205 | } |
206 | ||
bd1fa4f0 | 207 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 208 | { |
09cbfeaf KS |
209 | unsigned long index = start >> PAGE_SHIFT; |
210 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
211 | struct page *page; |
212 | ||
213 | while (index <= end_index) { | |
214 | page = find_get_page(inode->i_mapping, index); | |
215 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
216 | clear_page_dirty_for_io(page); | |
09cbfeaf | 217 | put_page(page); |
4adaa611 CM |
218 | index++; |
219 | } | |
4adaa611 CM |
220 | } |
221 | ||
f6311572 | 222 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 223 | { |
ebf55c88 | 224 | struct address_space *mapping = inode->i_mapping; |
09cbfeaf KS |
225 | unsigned long index = start >> PAGE_SHIFT; |
226 | unsigned long end_index = end >> PAGE_SHIFT; | |
ebf55c88 | 227 | struct folio *folio; |
4adaa611 CM |
228 | |
229 | while (index <= end_index) { | |
ebf55c88 MWO |
230 | folio = filemap_get_folio(mapping, index); |
231 | filemap_dirty_folio(mapping, folio); | |
232 | folio_account_redirty(folio); | |
233 | index += folio_nr_pages(folio); | |
234 | folio_put(folio); | |
4adaa611 | 235 | } |
4adaa611 CM |
236 | } |
237 | ||
ed8f13bf QW |
238 | /* |
239 | * Process one page for __process_pages_contig(). | |
240 | * | |
241 | * Return >0 if we hit @page == @locked_page. | |
242 | * Return 0 if we updated the page status. | |
243 | * Return -EGAIN if the we need to try again. | |
244 | * (For PAGE_LOCK case but got dirty page or page not belong to mapping) | |
245 | */ | |
e38992be QW |
246 | static int process_one_page(struct btrfs_fs_info *fs_info, |
247 | struct address_space *mapping, | |
ed8f13bf | 248 | struct page *page, struct page *locked_page, |
e38992be | 249 | unsigned long page_ops, u64 start, u64 end) |
ed8f13bf | 250 | { |
e38992be QW |
251 | u32 len; |
252 | ||
253 | ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX); | |
254 | len = end + 1 - start; | |
255 | ||
ed8f13bf | 256 | if (page_ops & PAGE_SET_ORDERED) |
b945a463 | 257 | btrfs_page_clamp_set_ordered(fs_info, page, start, len); |
ed8f13bf | 258 | if (page_ops & PAGE_SET_ERROR) |
e38992be | 259 | btrfs_page_clamp_set_error(fs_info, page, start, len); |
ed8f13bf | 260 | if (page_ops & PAGE_START_WRITEBACK) { |
e38992be QW |
261 | btrfs_page_clamp_clear_dirty(fs_info, page, start, len); |
262 | btrfs_page_clamp_set_writeback(fs_info, page, start, len); | |
ed8f13bf QW |
263 | } |
264 | if (page_ops & PAGE_END_WRITEBACK) | |
e38992be | 265 | btrfs_page_clamp_clear_writeback(fs_info, page, start, len); |
a33a8e9a QW |
266 | |
267 | if (page == locked_page) | |
268 | return 1; | |
269 | ||
ed8f13bf | 270 | if (page_ops & PAGE_LOCK) { |
1e1de387 QW |
271 | int ret; |
272 | ||
273 | ret = btrfs_page_start_writer_lock(fs_info, page, start, len); | |
274 | if (ret) | |
275 | return ret; | |
ed8f13bf | 276 | if (!PageDirty(page) || page->mapping != mapping) { |
1e1de387 | 277 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
278 | return -EAGAIN; |
279 | } | |
280 | } | |
281 | if (page_ops & PAGE_UNLOCK) | |
1e1de387 | 282 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
283 | return 0; |
284 | } | |
285 | ||
da2c7009 LB |
286 | static int __process_pages_contig(struct address_space *mapping, |
287 | struct page *locked_page, | |
98af9ab1 | 288 | u64 start, u64 end, unsigned long page_ops, |
ed8f13bf QW |
289 | u64 *processed_end) |
290 | { | |
e38992be | 291 | struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); |
ed8f13bf QW |
292 | pgoff_t start_index = start >> PAGE_SHIFT; |
293 | pgoff_t end_index = end >> PAGE_SHIFT; | |
294 | pgoff_t index = start_index; | |
ed8f13bf | 295 | unsigned long pages_processed = 0; |
04c6b79a | 296 | struct folio_batch fbatch; |
ed8f13bf QW |
297 | int err = 0; |
298 | int i; | |
299 | ||
300 | if (page_ops & PAGE_LOCK) { | |
301 | ASSERT(page_ops == PAGE_LOCK); | |
302 | ASSERT(processed_end && *processed_end == start); | |
303 | } | |
304 | ||
04c6b79a | 305 | if ((page_ops & PAGE_SET_ERROR) && start_index <= end_index) |
ed8f13bf QW |
306 | mapping_set_error(mapping, -EIO); |
307 | ||
04c6b79a VMO |
308 | folio_batch_init(&fbatch); |
309 | while (index <= end_index) { | |
310 | int found_folios; | |
311 | ||
312 | found_folios = filemap_get_folios_contig(mapping, &index, | |
313 | end_index, &fbatch); | |
ed8f13bf | 314 | |
04c6b79a | 315 | if (found_folios == 0) { |
ed8f13bf QW |
316 | /* |
317 | * Only if we're going to lock these pages, we can find | |
318 | * nothing at @index. | |
319 | */ | |
320 | ASSERT(page_ops & PAGE_LOCK); | |
321 | err = -EAGAIN; | |
322 | goto out; | |
323 | } | |
324 | ||
04c6b79a | 325 | for (i = 0; i < found_folios; i++) { |
ed8f13bf | 326 | int process_ret; |
04c6b79a | 327 | struct folio *folio = fbatch.folios[i]; |
e38992be | 328 | process_ret = process_one_page(fs_info, mapping, |
04c6b79a | 329 | &folio->page, locked_page, page_ops, |
e38992be | 330 | start, end); |
ed8f13bf | 331 | if (process_ret < 0) { |
ed8f13bf | 332 | err = -EAGAIN; |
04c6b79a | 333 | folio_batch_release(&fbatch); |
ed8f13bf QW |
334 | goto out; |
335 | } | |
04c6b79a | 336 | pages_processed += folio_nr_pages(folio); |
ed8f13bf | 337 | } |
04c6b79a | 338 | folio_batch_release(&fbatch); |
ed8f13bf QW |
339 | cond_resched(); |
340 | } | |
341 | out: | |
342 | if (err && processed_end) { | |
343 | /* | |
344 | * Update @processed_end. I know this is awful since it has | |
345 | * two different return value patterns (inclusive vs exclusive). | |
346 | * | |
347 | * But the exclusive pattern is necessary if @start is 0, or we | |
348 | * underflow and check against processed_end won't work as | |
349 | * expected. | |
350 | */ | |
351 | if (pages_processed) | |
352 | *processed_end = min(end, | |
353 | ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1); | |
354 | else | |
355 | *processed_end = start; | |
356 | } | |
357 | return err; | |
358 | } | |
da2c7009 | 359 | |
143bede5 JM |
360 | static noinline void __unlock_for_delalloc(struct inode *inode, |
361 | struct page *locked_page, | |
362 | u64 start, u64 end) | |
c8b97818 | 363 | { |
09cbfeaf KS |
364 | unsigned long index = start >> PAGE_SHIFT; |
365 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 366 | |
76c0021d | 367 | ASSERT(locked_page); |
c8b97818 | 368 | if (index == locked_page->index && end_index == index) |
143bede5 | 369 | return; |
c8b97818 | 370 | |
98af9ab1 | 371 | __process_pages_contig(inode->i_mapping, locked_page, start, end, |
76c0021d | 372 | PAGE_UNLOCK, NULL); |
c8b97818 CM |
373 | } |
374 | ||
375 | static noinline int lock_delalloc_pages(struct inode *inode, | |
376 | struct page *locked_page, | |
377 | u64 delalloc_start, | |
378 | u64 delalloc_end) | |
379 | { | |
09cbfeaf | 380 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
09cbfeaf | 381 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
98af9ab1 | 382 | u64 processed_end = delalloc_start; |
c8b97818 | 383 | int ret; |
c8b97818 | 384 | |
76c0021d | 385 | ASSERT(locked_page); |
c8b97818 CM |
386 | if (index == locked_page->index && index == end_index) |
387 | return 0; | |
388 | ||
98af9ab1 QW |
389 | ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start, |
390 | delalloc_end, PAGE_LOCK, &processed_end); | |
391 | if (ret == -EAGAIN && processed_end > delalloc_start) | |
76c0021d | 392 | __unlock_for_delalloc(inode, locked_page, delalloc_start, |
98af9ab1 | 393 | processed_end); |
c8b97818 CM |
394 | return ret; |
395 | } | |
396 | ||
397 | /* | |
3522e903 | 398 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
2749f7ef | 399 | * more than @max_bytes. |
c8b97818 | 400 | * |
2749f7ef QW |
401 | * @start: The original start bytenr to search. |
402 | * Will store the extent range start bytenr. | |
403 | * @end: The original end bytenr of the search range | |
404 | * Will store the extent range end bytenr. | |
405 | * | |
406 | * Return true if we find a delalloc range which starts inside the original | |
407 | * range, and @start/@end will store the delalloc range start/end. | |
408 | * | |
409 | * Return false if we can't find any delalloc range which starts inside the | |
410 | * original range, and @start/@end will be the non-delalloc range start/end. | |
c8b97818 | 411 | */ |
ce9f967f | 412 | EXPORT_FOR_TESTS |
3522e903 | 413 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 414 | struct page *locked_page, u64 *start, |
917aacec | 415 | u64 *end) |
c8b97818 | 416 | { |
f7b12a62 | 417 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9978059b | 418 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
2749f7ef QW |
419 | const u64 orig_start = *start; |
420 | const u64 orig_end = *end; | |
f7b12a62 NA |
421 | /* The sanity tests may not set a valid fs_info. */ |
422 | u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE; | |
c8b97818 CM |
423 | u64 delalloc_start; |
424 | u64 delalloc_end; | |
3522e903 | 425 | bool found; |
9655d298 | 426 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
427 | int ret; |
428 | int loops = 0; | |
429 | ||
2749f7ef QW |
430 | /* Caller should pass a valid @end to indicate the search range end */ |
431 | ASSERT(orig_end > orig_start); | |
432 | ||
433 | /* The range should at least cover part of the page */ | |
434 | ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE || | |
435 | orig_end <= page_offset(locked_page))); | |
c8b97818 CM |
436 | again: |
437 | /* step one, find a bunch of delalloc bytes starting at start */ | |
438 | delalloc_start = *start; | |
439 | delalloc_end = 0; | |
083e75e7 JB |
440 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
441 | max_bytes, &cached_state); | |
2749f7ef | 442 | if (!found || delalloc_end <= *start || delalloc_start > orig_end) { |
c8b97818 | 443 | *start = delalloc_start; |
2749f7ef QW |
444 | |
445 | /* @delalloc_end can be -1, never go beyond @orig_end */ | |
446 | *end = min(delalloc_end, orig_end); | |
c2a128d2 | 447 | free_extent_state(cached_state); |
3522e903 | 448 | return false; |
c8b97818 CM |
449 | } |
450 | ||
70b99e69 CM |
451 | /* |
452 | * start comes from the offset of locked_page. We have to lock | |
453 | * pages in order, so we can't process delalloc bytes before | |
454 | * locked_page | |
455 | */ | |
d397712b | 456 | if (delalloc_start < *start) |
70b99e69 | 457 | delalloc_start = *start; |
70b99e69 | 458 | |
c8b97818 CM |
459 | /* |
460 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 461 | */ |
7bf811a5 JB |
462 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
463 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 464 | |
c8b97818 CM |
465 | /* step two, lock all the pages after the page that has start */ |
466 | ret = lock_delalloc_pages(inode, locked_page, | |
467 | delalloc_start, delalloc_end); | |
9bfd61d9 | 468 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
469 | if (ret == -EAGAIN) { |
470 | /* some of the pages are gone, lets avoid looping by | |
471 | * shortening the size of the delalloc range we're searching | |
472 | */ | |
9655d298 | 473 | free_extent_state(cached_state); |
7d788742 | 474 | cached_state = NULL; |
c8b97818 | 475 | if (!loops) { |
09cbfeaf | 476 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
477 | loops = 1; |
478 | goto again; | |
479 | } else { | |
3522e903 | 480 | found = false; |
c8b97818 CM |
481 | goto out_failed; |
482 | } | |
483 | } | |
c8b97818 CM |
484 | |
485 | /* step three, lock the state bits for the whole range */ | |
570eb97b | 486 | lock_extent(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
487 | |
488 | /* then test to make sure it is all still delalloc */ | |
489 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 490 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 491 | if (!ret) { |
570eb97b JB |
492 | unlock_extent(tree, delalloc_start, delalloc_end, |
493 | &cached_state); | |
c8b97818 CM |
494 | __unlock_for_delalloc(inode, locked_page, |
495 | delalloc_start, delalloc_end); | |
496 | cond_resched(); | |
497 | goto again; | |
498 | } | |
9655d298 | 499 | free_extent_state(cached_state); |
c8b97818 CM |
500 | *start = delalloc_start; |
501 | *end = delalloc_end; | |
502 | out_failed: | |
503 | return found; | |
504 | } | |
505 | ||
ad7ff17b | 506 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 507 | struct page *locked_page, |
f97e27e9 | 508 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 509 | { |
bd015294 | 510 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL); |
873695b3 | 511 | |
ad7ff17b | 512 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
98af9ab1 | 513 | start, end, page_ops, NULL); |
873695b3 LB |
514 | } |
515 | ||
150e4b05 QW |
516 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) |
517 | { | |
518 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
519 | ||
520 | ASSERT(page_offset(page) <= start && | |
521 | start + len <= page_offset(page) + PAGE_SIZE); | |
522 | ||
150e4b05 | 523 | if (uptodate) { |
14605409 BB |
524 | if (fsverity_active(page->mapping->host) && |
525 | !PageError(page) && | |
526 | !PageUptodate(page) && | |
527 | start < i_size_read(page->mapping->host) && | |
528 | !fsverity_verify_page(page)) { | |
529 | btrfs_page_set_error(fs_info, page, start, len); | |
530 | } else { | |
531 | btrfs_page_set_uptodate(fs_info, page, start, len); | |
532 | } | |
150e4b05 QW |
533 | } else { |
534 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
535 | btrfs_page_set_error(fs_info, page, start, len); | |
536 | } | |
537 | ||
fbca46eb | 538 | if (!btrfs_is_subpage(fs_info, page)) |
150e4b05 | 539 | unlock_page(page); |
3d078efa | 540 | else |
150e4b05 QW |
541 | btrfs_subpage_end_reader(fs_info, page, start, len); |
542 | } | |
543 | ||
d1310b2e CM |
544 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
545 | ||
b5227c07 | 546 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 547 | { |
38a39ac7 | 548 | struct btrfs_inode *inode; |
25c1252a | 549 | const bool uptodate = (err == 0); |
3e2426bd | 550 | int ret = 0; |
87826df0 | 551 | |
38a39ac7 QW |
552 | ASSERT(page && page->mapping); |
553 | inode = BTRFS_I(page->mapping->host); | |
554 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 555 | |
87826df0 | 556 | if (!uptodate) { |
963e4db8 QW |
557 | const struct btrfs_fs_info *fs_info = inode->root->fs_info; |
558 | u32 len; | |
559 | ||
560 | ASSERT(end + 1 - start <= U32_MAX); | |
561 | len = end + 1 - start; | |
562 | ||
563 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
564 | btrfs_page_set_error(fs_info, page, start, len); | |
bff5baf8 | 565 | ret = err < 0 ? err : -EIO; |
5dca6eea | 566 | mapping_set_error(page->mapping, ret); |
87826df0 | 567 | } |
87826df0 JM |
568 | } |
569 | ||
d1310b2e CM |
570 | /* |
571 | * after a writepage IO is done, we need to: | |
572 | * clear the uptodate bits on error | |
573 | * clear the writeback bits in the extent tree for this IO | |
574 | * end_page_writeback if the page has no more pending IO | |
575 | * | |
576 | * Scheduling is not allowed, so the extent state tree is expected | |
577 | * to have one and only one object corresponding to this IO. | |
578 | */ | |
917f32a2 | 579 | static void end_bio_extent_writepage(struct btrfs_bio *bbio) |
d1310b2e | 580 | { |
917f32a2 | 581 | struct bio *bio = &bbio->bio; |
4e4cbee9 | 582 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 583 | struct bio_vec *bvec; |
d1310b2e CM |
584 | u64 start; |
585 | u64 end; | |
6dc4f100 | 586 | struct bvec_iter_all iter_all; |
d1310b2e | 587 | |
c09abff8 | 588 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 589 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 590 | struct page *page = bvec->bv_page; |
0b246afa JM |
591 | struct inode *inode = page->mapping->host; |
592 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
593 | const u32 sectorsize = fs_info->sectorsize; |
594 | ||
595 | /* Our read/write should always be sector aligned. */ | |
596 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
597 | btrfs_err(fs_info, | |
598 | "partial page write in btrfs with offset %u and length %u", | |
599 | bvec->bv_offset, bvec->bv_len); | |
600 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
601 | btrfs_info(fs_info, | |
602 | "incomplete page write with offset %u and length %u", | |
603 | bvec->bv_offset, bvec->bv_len); | |
604 | ||
605 | start = page_offset(page) + bvec->bv_offset; | |
606 | end = start + bvec->bv_len - 1; | |
d1310b2e | 607 | |
4e4cbee9 | 608 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
609 | |
610 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 611 | } |
2b1f55b0 | 612 | |
d1310b2e | 613 | bio_put(bio); |
d1310b2e CM |
614 | } |
615 | ||
94e8c95c QW |
616 | /* |
617 | * Record previously processed extent range | |
618 | * | |
619 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
620 | * the extent io operations. | |
621 | */ | |
622 | struct processed_extent { | |
623 | struct btrfs_inode *inode; | |
624 | /* Start of the range in @inode */ | |
625 | u64 start; | |
2e626e56 | 626 | /* End of the range in @inode */ |
94e8c95c QW |
627 | u64 end; |
628 | bool uptodate; | |
629 | }; | |
630 | ||
631 | /* | |
632 | * Try to release processed extent range | |
633 | * | |
634 | * May not release the extent range right now if the current range is | |
635 | * contiguous to processed extent. | |
636 | * | |
637 | * Will release processed extent when any of @inode, @uptodate, the range is | |
638 | * no longer contiguous to the processed range. | |
639 | * | |
640 | * Passing @inode == NULL will force processed extent to be released. | |
641 | */ | |
642 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
643 | struct btrfs_inode *inode, u64 start, u64 end, | |
644 | bool uptodate) | |
883d0de4 MX |
645 | { |
646 | struct extent_state *cached = NULL; | |
94e8c95c QW |
647 | struct extent_io_tree *tree; |
648 | ||
649 | /* The first extent, initialize @processed */ | |
650 | if (!processed->inode) | |
651 | goto update; | |
883d0de4 | 652 | |
94e8c95c QW |
653 | /* |
654 | * Contiguous to processed extent, just uptodate the end. | |
655 | * | |
656 | * Several things to notice: | |
657 | * | |
658 | * - bio can be merged as long as on-disk bytenr is contiguous | |
659 | * This means we can have page belonging to other inodes, thus need to | |
660 | * check if the inode still matches. | |
661 | * - bvec can contain range beyond current page for multi-page bvec | |
662 | * Thus we need to do processed->end + 1 >= start check | |
663 | */ | |
664 | if (processed->inode == inode && processed->uptodate == uptodate && | |
665 | processed->end + 1 >= start && end >= processed->end) { | |
666 | processed->end = end; | |
667 | return; | |
668 | } | |
669 | ||
670 | tree = &processed->inode->io_tree; | |
671 | /* | |
672 | * Now we don't have range contiguous to the processed range, release | |
673 | * the processed range now. | |
674 | */ | |
48acc47d | 675 | unlock_extent(tree, processed->start, processed->end, &cached); |
94e8c95c QW |
676 | |
677 | update: | |
678 | /* Update processed to current range */ | |
679 | processed->inode = inode; | |
680 | processed->start = start; | |
681 | processed->end = end; | |
682 | processed->uptodate = uptodate; | |
883d0de4 MX |
683 | } |
684 | ||
92082d40 QW |
685 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
686 | { | |
687 | ASSERT(PageLocked(page)); | |
fbca46eb | 688 | if (!btrfs_is_subpage(fs_info, page)) |
92082d40 QW |
689 | return; |
690 | ||
691 | ASSERT(PagePrivate(page)); | |
692 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
693 | } | |
694 | ||
d9bb77d5 | 695 | /* |
01cd3909 | 696 | * Find extent buffer for a givne bytenr. |
d9bb77d5 QW |
697 | * |
698 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
699 | * in endio context. | |
700 | */ | |
701 | static struct extent_buffer *find_extent_buffer_readpage( | |
702 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
703 | { | |
704 | struct extent_buffer *eb; | |
705 | ||
706 | /* | |
707 | * For regular sectorsize, we can use page->private to grab extent | |
708 | * buffer | |
709 | */ | |
fbca46eb | 710 | if (fs_info->nodesize >= PAGE_SIZE) { |
d9bb77d5 QW |
711 | ASSERT(PagePrivate(page) && page->private); |
712 | return (struct extent_buffer *)page->private; | |
713 | } | |
714 | ||
01cd3909 DS |
715 | /* For subpage case, we need to lookup buffer radix tree */ |
716 | rcu_read_lock(); | |
717 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
718 | bytenr >> fs_info->sectorsize_bits); | |
719 | rcu_read_unlock(); | |
d9bb77d5 QW |
720 | ASSERT(eb); |
721 | return eb; | |
722 | } | |
723 | ||
d1310b2e CM |
724 | /* |
725 | * after a readpage IO is done, we need to: | |
726 | * clear the uptodate bits on error | |
727 | * set the uptodate bits if things worked | |
728 | * set the page up to date if all extents in the tree are uptodate | |
729 | * clear the lock bit in the extent tree | |
730 | * unlock the page if there are no other extents locked for it | |
731 | * | |
732 | * Scheduling is not allowed, so the extent state tree is expected | |
733 | * to have one and only one object corresponding to this IO. | |
734 | */ | |
917f32a2 | 735 | static void end_bio_extent_readpage(struct btrfs_bio *bbio) |
d1310b2e | 736 | { |
917f32a2 | 737 | struct bio *bio = &bbio->bio; |
2c30c71b | 738 | struct bio_vec *bvec; |
94e8c95c | 739 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
740 | /* |
741 | * The offset to the beginning of a bio, since one bio can never be | |
742 | * larger than UINT_MAX, u32 here is enough. | |
743 | */ | |
744 | u32 bio_offset = 0; | |
5cf1ab56 | 745 | int mirror; |
6dc4f100 | 746 | struct bvec_iter_all iter_all; |
d1310b2e | 747 | |
c09abff8 | 748 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 749 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 750 | bool uptodate = !bio->bi_status; |
d1310b2e | 751 | struct page *page = bvec->bv_page; |
a71754fc | 752 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 753 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 QW |
754 | const u32 sectorsize = fs_info->sectorsize; |
755 | u64 start; | |
756 | u64 end; | |
757 | u32 len; | |
507903b8 | 758 | |
ab8d0fc4 JM |
759 | btrfs_debug(fs_info, |
760 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 761 | bio->bi_iter.bi_sector, bio->bi_status, |
c3a3b19b | 762 | bbio->mirror_num); |
902b22f3 | 763 | |
8b8bbd46 QW |
764 | /* |
765 | * We always issue full-sector reads, but if some block in a | |
766 | * page fails to read, blk_update_request() will advance | |
767 | * bv_offset and adjust bv_len to compensate. Print a warning | |
768 | * for unaligned offsets, and an error if they don't add up to | |
769 | * a full sector. | |
770 | */ | |
771 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
772 | btrfs_err(fs_info, | |
773 | "partial page read in btrfs with offset %u and length %u", | |
774 | bvec->bv_offset, bvec->bv_len); | |
775 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
776 | sectorsize)) | |
777 | btrfs_info(fs_info, | |
778 | "incomplete page read with offset %u and length %u", | |
779 | bvec->bv_offset, bvec->bv_len); | |
780 | ||
781 | start = page_offset(page) + bvec->bv_offset; | |
782 | end = start + bvec->bv_len - 1; | |
facc8a22 | 783 | len = bvec->bv_len; |
d1310b2e | 784 | |
c3a3b19b | 785 | mirror = bbio->mirror_num; |
7609afac CH |
786 | if (uptodate && !is_data_inode(inode) && |
787 | btrfs_validate_metadata_buffer(bbio, page, start, end, mirror)) | |
788 | uptodate = false; | |
ea466794 | 789 | |
883d0de4 | 790 | if (likely(uptodate)) { |
a71754fc | 791 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 792 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 793 | |
c28ea613 QW |
794 | /* |
795 | * Zero out the remaining part if this range straddles | |
796 | * i_size. | |
797 | * | |
798 | * Here we should only zero the range inside the bvec, | |
799 | * not touch anything else. | |
800 | * | |
801 | * NOTE: i_size is exclusive while end is inclusive. | |
802 | */ | |
803 | if (page->index == end_index && i_size <= end) { | |
804 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 805 | offset_in_page(start)); |
c28ea613 QW |
806 | |
807 | zero_user_segment(page, zero_start, | |
808 | offset_in_page(end) + 1); | |
809 | } | |
7609afac | 810 | } else if (!is_data_inode(inode)) { |
97861cd1 CH |
811 | struct extent_buffer *eb; |
812 | ||
813 | eb = find_extent_buffer_readpage(fs_info, page, start); | |
814 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
815 | eb->read_mirror = mirror; | |
816 | atomic_dec(&eb->io_pages); | |
70dec807 | 817 | } |
97861cd1 | 818 | |
7609afac CH |
819 | /* Update page status and unlock. */ |
820 | end_page_read(page, uptodate, start, len); | |
821 | endio_readpage_release_extent(&processed, BTRFS_I(inode), | |
822 | start, end, PageUptodate(page)); | |
97861cd1 | 823 | |
7ffd27e3 QW |
824 | ASSERT(bio_offset + len > bio_offset); |
825 | bio_offset += len; | |
883d0de4 | 826 | |
2c30c71b | 827 | } |
94e8c95c QW |
828 | /* Release the last extent */ |
829 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
d1310b2e | 830 | bio_put(bio); |
d1310b2e CM |
831 | } |
832 | ||
43dd529a | 833 | /* |
dd137dd1 STD |
834 | * Populate every free slot in a provided array with pages. |
835 | * | |
836 | * @nr_pages: number of pages to allocate | |
837 | * @page_array: the array to fill with pages; any existing non-null entries in | |
838 | * the array will be skipped | |
839 | * | |
840 | * Return: 0 if all pages were able to be allocated; | |
841 | * -ENOMEM otherwise, and the caller is responsible for freeing all | |
842 | * non-null page pointers in the array. | |
843 | */ | |
844 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array) | |
845 | { | |
91d6ac1d | 846 | unsigned int allocated; |
dd137dd1 | 847 | |
91d6ac1d STD |
848 | for (allocated = 0; allocated < nr_pages;) { |
849 | unsigned int last = allocated; | |
dd137dd1 | 850 | |
91d6ac1d STD |
851 | allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array); |
852 | ||
395cb57e STD |
853 | if (allocated == nr_pages) |
854 | return 0; | |
855 | ||
91d6ac1d STD |
856 | /* |
857 | * During this iteration, no page could be allocated, even | |
858 | * though alloc_pages_bulk_array() falls back to alloc_page() | |
859 | * if it could not bulk-allocate. So we must be out of memory. | |
860 | */ | |
861 | if (allocated == last) | |
dd137dd1 | 862 | return -ENOMEM; |
395cb57e STD |
863 | |
864 | memalloc_retry_wait(GFP_NOFS); | |
dd137dd1 STD |
865 | } |
866 | return 0; | |
867 | } | |
868 | ||
43dd529a DS |
869 | /* |
870 | * Attempt to add a page to bio. | |
953651eb | 871 | * |
43dd529a DS |
872 | * @bio_ctrl: record both the bio, and its bio_flags |
873 | * @page: page to add to the bio | |
874 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
875 | * a contiguous page to the previous one | |
876 | * @size: portion of page that we want to write | |
877 | * @pg_offset: starting offset in the page | |
953651eb NA |
878 | * |
879 | * Attempt to add a page to bio considering stripe alignment etc. | |
880 | * | |
e0eefe07 QW |
881 | * Return >= 0 for the number of bytes added to the bio. |
882 | * Can return 0 if the current bio is already at stripe/zone boundary. | |
883 | * Return <0 for error. | |
953651eb | 884 | */ |
e0eefe07 QW |
885 | static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
886 | struct page *page, | |
887 | u64 disk_bytenr, unsigned int size, | |
c9bc621f | 888 | unsigned int pg_offset) |
953651eb | 889 | { |
390ed29b QW |
890 | struct bio *bio = bio_ctrl->bio; |
891 | u32 bio_size = bio->bi_iter.bi_size; | |
e0eefe07 | 892 | u32 real_size; |
953651eb | 893 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
4a445b7b | 894 | bool contig = false; |
953651eb | 895 | |
390ed29b QW |
896 | ASSERT(bio); |
897 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
2380220e | 898 | ASSERT(bio_ctrl->len_to_oe_boundary); |
4a445b7b QW |
899 | |
900 | if (bio->bi_iter.bi_size == 0) { | |
901 | /* We can always add a page into an empty bio. */ | |
902 | contig = true; | |
903 | } else if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE) { | |
904 | struct bio_vec *bvec = bio_last_bvec_all(bio); | |
905 | ||
906 | /* | |
907 | * The contig check requires the following conditions to be met: | |
908 | * 1) The pages are belonging to the same inode | |
909 | * This is implied by the call chain. | |
910 | * | |
911 | * 2) The range has adjacent logical bytenr | |
912 | * | |
913 | * 3) The range has adjacent file offset | |
914 | * This is required for the usage of btrfs_bio->file_offset. | |
915 | */ | |
916 | if (bio_end_sector(bio) == sector && | |
917 | page_offset(bvec->bv_page) + bvec->bv_offset + | |
918 | bvec->bv_len == page_offset(page) + pg_offset) | |
919 | contig = true; | |
920 | } else { | |
921 | /* | |
922 | * For compression, all IO should have its logical bytenr | |
923 | * set to the starting bytenr of the compressed extent. | |
924 | */ | |
953651eb | 925 | contig = bio->bi_iter.bi_sector == sector; |
4a445b7b QW |
926 | } |
927 | ||
953651eb | 928 | if (!contig) |
e0eefe07 | 929 | return 0; |
953651eb | 930 | |
2380220e | 931 | real_size = min(bio_ctrl->len_to_oe_boundary - bio_size, size); |
e0eefe07 QW |
932 | |
933 | /* | |
934 | * If real_size is 0, never call bio_add_*_page(), as even size is 0, | |
935 | * bio will still execute its endio function on the page! | |
936 | */ | |
937 | if (real_size == 0) | |
938 | return 0; | |
953651eb | 939 | |
d5e4377d | 940 | return bio_add_page(bio, page, real_size, pg_offset); |
953651eb NA |
941 | } |
942 | ||
2380220e QW |
943 | static void calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
944 | struct btrfs_inode *inode, u64 file_offset) | |
390ed29b | 945 | { |
390ed29b | 946 | struct btrfs_ordered_extent *ordered; |
390ed29b QW |
947 | |
948 | /* | |
2380220e QW |
949 | * Limit the extent to the ordered boundary for Zone Append. |
950 | * Compressed bios aren't submitted directly, so it doesn't apply to | |
951 | * them. | |
390ed29b | 952 | */ |
2380220e | 953 | if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE && |
921603c7 | 954 | btrfs_use_zone_append(btrfs_bio(bio_ctrl->bio))) { |
2380220e QW |
955 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
956 | if (ordered) { | |
957 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
72fcf1a4 CH |
958 | ordered->file_offset + |
959 | ordered->disk_num_bytes - file_offset); | |
2380220e QW |
960 | btrfs_put_ordered_extent(ordered); |
961 | return; | |
962 | } | |
390ed29b QW |
963 | } |
964 | ||
2380220e | 965 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
390ed29b QW |
966 | } |
967 | ||
d5e4377d CH |
968 | static void alloc_new_bio(struct btrfs_inode *inode, |
969 | struct btrfs_bio_ctrl *bio_ctrl, | |
d5e4377d CH |
970 | u64 disk_bytenr, u32 offset, u64 file_offset, |
971 | enum btrfs_compression_type compress_type) | |
e0eefe07 QW |
972 | { |
973 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
974 | struct bio *bio; | |
e0eefe07 | 975 | |
c000bc04 CH |
976 | bio = btrfs_bio_alloc(BIO_MAX_VECS, bio_ctrl->opf, inode, |
977 | bio_ctrl->end_io_func, NULL); | |
e0eefe07 QW |
978 | /* |
979 | * For compressed page range, its disk_bytenr is always @disk_bytenr | |
980 | * passed in, no matter if we have added any range into previous bio. | |
981 | */ | |
cb3a12d9 | 982 | if (compress_type != BTRFS_COMPRESS_NONE) |
cd8e0cca | 983 | bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
e0eefe07 | 984 | else |
cd8e0cca | 985 | bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT; |
0d495430 | 986 | btrfs_bio(bio)->file_offset = file_offset; |
e0eefe07 | 987 | bio_ctrl->bio = bio; |
0f07003b | 988 | bio_ctrl->compress_type = compress_type; |
2380220e | 989 | calc_bio_boundaries(bio_ctrl, inode, file_offset); |
e0eefe07 | 990 | |
72b505dc | 991 | if (bio_ctrl->wbc) { |
50f1cff3 | 992 | /* |
d5e4377d CH |
993 | * Pick the last added device to support cgroup writeback. For |
994 | * multi-device file systems this means blk-cgroup policies have | |
995 | * to always be set on the last added/replaced device. | |
996 | * This is a bit odd but has been like that for a long time. | |
50f1cff3 | 997 | */ |
d5e4377d | 998 | bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev); |
72b505dc | 999 | wbc_init_bio(bio_ctrl->wbc, bio); |
e0eefe07 | 1000 | } |
e0eefe07 QW |
1001 | } |
1002 | ||
4b81ba48 | 1003 | /* |
0c64c33c | 1004 | * @disk_bytenr: logical bytenr where the write will be |
209ecde5 | 1005 | * @page: page to add to the bio |
0c64c33c | 1006 | * @size: portion of page that we want to write to |
b8b3d625 DS |
1007 | * @pg_offset: offset of the new bio or to check whether we are adding |
1008 | * a contiguous page to the previous one | |
cb3a12d9 | 1009 | * @compress_type: compress type for current bio |
814b6f91 QW |
1010 | * |
1011 | * The will either add the page into the existing @bio_ctrl->bio, or allocate a | |
1012 | * new one in @bio_ctrl->bio. | |
1013 | * The mirror number for this IO should already be initizlied in | |
1014 | * @bio_ctrl->mirror_num. | |
4b81ba48 | 1015 | */ |
72b505dc | 1016 | static int submit_extent_page(struct btrfs_bio_ctrl *bio_ctrl, |
209ecde5 | 1017 | u64 disk_bytenr, struct page *page, |
6c5a4e2c | 1018 | size_t size, unsigned long pg_offset, |
eb8d0c6d | 1019 | enum btrfs_compression_type compress_type) |
d1310b2e | 1020 | { |
e1326f03 | 1021 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
e0eefe07 | 1022 | unsigned int cur = pg_offset; |
d1310b2e | 1023 | |
390ed29b | 1024 | ASSERT(bio_ctrl); |
5c2b1fd7 | 1025 | |
390ed29b QW |
1026 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
1027 | pg_offset + size <= PAGE_SIZE); | |
5467abba QW |
1028 | |
1029 | ASSERT(bio_ctrl->end_io_func); | |
1030 | ||
e0eefe07 QW |
1031 | while (cur < pg_offset + size) { |
1032 | u32 offset = cur - pg_offset; | |
1033 | int added; | |
1034 | ||
1035 | /* Allocate new bio if needed */ | |
1036 | if (!bio_ctrl->bio) { | |
72b505dc | 1037 | alloc_new_bio(inode, bio_ctrl, disk_bytenr, |
d5e4377d CH |
1038 | offset, page_offset(page) + cur, |
1039 | compress_type); | |
e0eefe07 QW |
1040 | } |
1041 | /* | |
1042 | * We must go through btrfs_bio_add_page() to ensure each | |
1043 | * page range won't cross various boundaries. | |
1044 | */ | |
cb3a12d9 | 1045 | if (compress_type != BTRFS_COMPRESS_NONE) |
e0eefe07 | 1046 | added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, |
c9bc621f | 1047 | size - offset, pg_offset + offset); |
e0eefe07 QW |
1048 | else |
1049 | added = btrfs_bio_add_page(bio_ctrl, page, | |
1050 | disk_bytenr + offset, size - offset, | |
c9bc621f | 1051 | pg_offset + offset); |
e0eefe07 QW |
1052 | |
1053 | /* Metadata page range should never be split */ | |
1054 | if (!is_data_inode(&inode->vfs_inode)) | |
1055 | ASSERT(added == 0 || added == size - offset); | |
1056 | ||
1057 | /* At least we added some page, update the account */ | |
72b505dc CH |
1058 | if (bio_ctrl->wbc && added) |
1059 | wbc_account_cgroup_owner(bio_ctrl->wbc, page, added); | |
e0eefe07 QW |
1060 | |
1061 | /* We have reached boundary, submit right now */ | |
1062 | if (added < size - offset) { | |
1063 | /* The bio should contain some page(s) */ | |
1064 | ASSERT(bio_ctrl->bio->bi_iter.bi_size); | |
722c82ac | 1065 | submit_one_bio(bio_ctrl); |
d1310b2e | 1066 | } |
e0eefe07 | 1067 | cur += added; |
d1310b2e | 1068 | } |
e0eefe07 | 1069 | return 0; |
d1310b2e CM |
1070 | } |
1071 | ||
760f991f QW |
1072 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
1073 | struct page *page, | |
1074 | struct btrfs_subpage *prealloc) | |
d1310b2e | 1075 | { |
760f991f QW |
1076 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1077 | int ret = 0; | |
1078 | ||
0d01e247 QW |
1079 | /* |
1080 | * If the page is mapped to btree inode, we should hold the private | |
1081 | * lock to prevent race. | |
1082 | * For cloned or dummy extent buffers, their pages are not mapped and | |
1083 | * will not race with any other ebs. | |
1084 | */ | |
1085 | if (page->mapping) | |
1086 | lockdep_assert_held(&page->mapping->private_lock); | |
1087 | ||
fbca46eb | 1088 | if (fs_info->nodesize >= PAGE_SIZE) { |
760f991f QW |
1089 | if (!PagePrivate(page)) |
1090 | attach_page_private(page, eb); | |
1091 | else | |
1092 | WARN_ON(page->private != (unsigned long)eb); | |
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | /* Already mapped, just free prealloc */ | |
1097 | if (PagePrivate(page)) { | |
1098 | btrfs_free_subpage(prealloc); | |
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | if (prealloc) | |
1103 | /* Has preallocated memory for subpage */ | |
1104 | attach_page_private(page, prealloc); | |
d1b89bc0 | 1105 | else |
760f991f QW |
1106 | /* Do new allocation to attach subpage */ |
1107 | ret = btrfs_attach_subpage(fs_info, page, | |
1108 | BTRFS_SUBPAGE_METADATA); | |
1109 | return ret; | |
d1310b2e CM |
1110 | } |
1111 | ||
32443de3 | 1112 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 1113 | { |
32443de3 QW |
1114 | struct btrfs_fs_info *fs_info; |
1115 | ||
1116 | ASSERT(page->mapping); | |
1117 | ||
1118 | if (PagePrivate(page)) | |
1119 | return 0; | |
1120 | ||
1121 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
1122 | ||
fbca46eb | 1123 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
1124 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); |
1125 | ||
1126 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | void clear_page_extent_mapped(struct page *page) | |
1131 | { | |
1132 | struct btrfs_fs_info *fs_info; | |
1133 | ||
1134 | ASSERT(page->mapping); | |
1135 | ||
d1b89bc0 | 1136 | if (!PagePrivate(page)) |
32443de3 QW |
1137 | return; |
1138 | ||
1139 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
fbca46eb | 1140 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
1141 | return btrfs_detach_subpage(fs_info, page); |
1142 | ||
1143 | detach_page_private(page); | |
d1310b2e CM |
1144 | } |
1145 | ||
125bac01 MX |
1146 | static struct extent_map * |
1147 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 1148 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
1149 | { |
1150 | struct extent_map *em; | |
1151 | ||
1152 | if (em_cached && *em_cached) { | |
1153 | em = *em_cached; | |
cbc0e928 | 1154 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 1155 | start < extent_map_end(em)) { |
490b54d6 | 1156 | refcount_inc(&em->refs); |
125bac01 MX |
1157 | return em; |
1158 | } | |
1159 | ||
1160 | free_extent_map(em); | |
1161 | *em_cached = NULL; | |
1162 | } | |
1163 | ||
1a5ee1e6 | 1164 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
c0347550 | 1165 | if (em_cached && !IS_ERR(em)) { |
125bac01 | 1166 | BUG_ON(*em_cached); |
490b54d6 | 1167 | refcount_inc(&em->refs); |
125bac01 MX |
1168 | *em_cached = em; |
1169 | } | |
1170 | return em; | |
1171 | } | |
d1310b2e CM |
1172 | /* |
1173 | * basic readpage implementation. Locked extent state structs are inserted | |
1174 | * into the tree that are removed when the IO is done (by the end_io | |
1175 | * handlers) | |
79787eaa | 1176 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 1177 | * return 0 on success, otherwise return error |
d1310b2e | 1178 | */ |
7aab8b32 | 1179 | static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
c000bc04 | 1180 | struct btrfs_bio_ctrl *bio_ctrl, u64 *prev_em_start) |
d1310b2e CM |
1181 | { |
1182 | struct inode *inode = page->mapping->host; | |
92082d40 | 1183 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 1184 | u64 start = page_offset(page); |
8eec8296 | 1185 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
1186 | u64 cur = start; |
1187 | u64 extent_offset; | |
1188 | u64 last_byte = i_size_read(inode); | |
1189 | u64 block_start; | |
d1310b2e | 1190 | struct extent_map *em; |
baf863b9 | 1191 | int ret = 0; |
306e16ce | 1192 | size_t pg_offset = 0; |
d1310b2e CM |
1193 | size_t iosize; |
1194 | size_t blocksize = inode->i_sb->s_blocksize; | |
f657a31c | 1195 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 1196 | |
32443de3 QW |
1197 | ret = set_page_extent_mapped(page); |
1198 | if (ret < 0) { | |
570eb97b | 1199 | unlock_extent(tree, start, end, NULL); |
92082d40 QW |
1200 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
1201 | unlock_page(page); | |
32443de3 QW |
1202 | goto out; |
1203 | } | |
d1310b2e | 1204 | |
09cbfeaf | 1205 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 1206 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
1207 | |
1208 | if (zero_offset) { | |
09cbfeaf | 1209 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 1210 | memzero_page(page, zero_offset, iosize); |
c8b97818 CM |
1211 | } |
1212 | } | |
5467abba | 1213 | bio_ctrl->end_io_func = end_bio_extent_readpage; |
92082d40 | 1214 | begin_page_read(fs_info, page); |
d1310b2e | 1215 | while (cur <= end) { |
a140453b | 1216 | enum btrfs_compression_type compress_type = BTRFS_COMPRESS_NONE; |
005efedf | 1217 | bool force_bio_submit = false; |
0c64c33c | 1218 | u64 disk_bytenr; |
c8f2f24b | 1219 | |
6a404910 | 1220 | ASSERT(IS_ALIGNED(cur, fs_info->sectorsize)); |
d1310b2e | 1221 | if (cur >= last_byte) { |
09cbfeaf | 1222 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 1223 | memzero_page(page, pg_offset, iosize); |
2c8f5e8c | 1224 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
92082d40 | 1225 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
1226 | break; |
1227 | } | |
125bac01 | 1228 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 1229 | end - cur + 1, em_cached); |
c0347550 | 1230 | if (IS_ERR(em)) { |
570eb97b | 1231 | unlock_extent(tree, cur, end, NULL); |
92082d40 | 1232 | end_page_read(page, false, cur, end + 1 - cur); |
bbf0ea7e | 1233 | ret = PTR_ERR(em); |
d1310b2e CM |
1234 | break; |
1235 | } | |
d1310b2e CM |
1236 | extent_offset = cur - em->start; |
1237 | BUG_ON(extent_map_end(em) <= cur); | |
1238 | BUG_ON(end < cur); | |
1239 | ||
7f6ca7f2 | 1240 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
a140453b | 1241 | compress_type = em->compress_type; |
c8b97818 | 1242 | |
d1310b2e | 1243 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
fda2832f | 1244 | iosize = ALIGN(iosize, blocksize); |
a140453b | 1245 | if (compress_type != BTRFS_COMPRESS_NONE) |
0c64c33c | 1246 | disk_bytenr = em->block_start; |
949b3273 | 1247 | else |
0c64c33c | 1248 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 1249 | block_start = em->block_start; |
d899e052 YZ |
1250 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
1251 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
1252 | |
1253 | /* | |
1254 | * If we have a file range that points to a compressed extent | |
260db43c | 1255 | * and it's followed by a consecutive file range that points |
005efedf FM |
1256 | * to the same compressed extent (possibly with a different |
1257 | * offset and/or length, so it either points to the whole extent | |
1258 | * or only part of it), we must make sure we do not submit a | |
1259 | * single bio to populate the pages for the 2 ranges because | |
1260 | * this makes the compressed extent read zero out the pages | |
1261 | * belonging to the 2nd range. Imagine the following scenario: | |
1262 | * | |
1263 | * File layout | |
1264 | * [0 - 8K] [8K - 24K] | |
1265 | * | | | |
1266 | * | | | |
1267 | * points to extent X, points to extent X, | |
1268 | * offset 4K, length of 8K offset 0, length 16K | |
1269 | * | |
1270 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
1271 | * | |
1272 | * If the bio to read the compressed extent covers both ranges, | |
1273 | * it will decompress extent X into the pages belonging to the | |
1274 | * first range and then it will stop, zeroing out the remaining | |
1275 | * pages that belong to the other range that points to extent X. | |
1276 | * So here we make sure we submit 2 bios, one for the first | |
1277 | * range and another one for the third range. Both will target | |
1278 | * the same physical extent from disk, but we can't currently | |
1279 | * make the compressed bio endio callback populate the pages | |
1280 | * for both ranges because each compressed bio is tightly | |
1281 | * coupled with a single extent map, and each range can have | |
1282 | * an extent map with a different offset value relative to the | |
1283 | * uncompressed data of our extent and different lengths. This | |
1284 | * is a corner case so we prioritize correctness over | |
1285 | * non-optimal behavior (submitting 2 bios for the same extent). | |
1286 | */ | |
1287 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
1288 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 1289 | *prev_em_start != em->start) |
005efedf FM |
1290 | force_bio_submit = true; |
1291 | ||
1292 | if (prev_em_start) | |
8e928218 | 1293 | *prev_em_start = em->start; |
005efedf | 1294 | |
d1310b2e CM |
1295 | free_extent_map(em); |
1296 | em = NULL; | |
1297 | ||
1298 | /* we've found a hole, just zero and go on */ | |
1299 | if (block_start == EXTENT_MAP_HOLE) { | |
d048b9c2 | 1300 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 1301 | |
2c8f5e8c | 1302 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
92082d40 | 1303 | end_page_read(page, true, cur, iosize); |
d1310b2e | 1304 | cur = cur + iosize; |
306e16ce | 1305 | pg_offset += iosize; |
d1310b2e CM |
1306 | continue; |
1307 | } | |
1308 | /* the get_extent function already copied into the page */ | |
70dec807 | 1309 | if (block_start == EXTENT_MAP_INLINE) { |
570eb97b | 1310 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
52b029f4 | 1311 | end_page_read(page, true, cur, iosize); |
70dec807 | 1312 | cur = cur + iosize; |
306e16ce | 1313 | pg_offset += iosize; |
70dec807 CM |
1314 | continue; |
1315 | } | |
d1310b2e | 1316 | |
a140453b | 1317 | if (bio_ctrl->compress_type != compress_type) |
c9bc621f CH |
1318 | submit_one_bio(bio_ctrl); |
1319 | ||
eb8d0c6d CH |
1320 | if (force_bio_submit) |
1321 | submit_one_bio(bio_ctrl); | |
72b505dc | 1322 | ret = submit_extent_page(bio_ctrl, disk_bytenr, page, iosize, |
a140453b | 1323 | pg_offset, compress_type); |
ad3fc794 | 1324 | if (ret) { |
10f7f6f8 QW |
1325 | /* |
1326 | * We have to unlock the remaining range, or the page | |
1327 | * will never be unlocked. | |
1328 | */ | |
570eb97b | 1329 | unlock_extent(tree, cur, end, NULL); |
10f7f6f8 | 1330 | end_page_read(page, false, cur, end + 1 - cur); |
baf863b9 | 1331 | goto out; |
edd33c99 | 1332 | } |
d1310b2e | 1333 | cur = cur + iosize; |
306e16ce | 1334 | pg_offset += iosize; |
d1310b2e | 1335 | } |
90a887c9 | 1336 | out: |
baf863b9 | 1337 | return ret; |
d1310b2e CM |
1338 | } |
1339 | ||
fdaf9a58 | 1340 | int btrfs_read_folio(struct file *file, struct folio *folio) |
7aab8b32 | 1341 | { |
fdaf9a58 | 1342 | struct page *page = &folio->page; |
7aab8b32 CH |
1343 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
1344 | u64 start = page_offset(page); | |
1345 | u64 end = start + PAGE_SIZE - 1; | |
c000bc04 | 1346 | struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ }; |
7aab8b32 CH |
1347 | int ret; |
1348 | ||
1349 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); | |
1350 | ||
c000bc04 | 1351 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, NULL); |
7aab8b32 CH |
1352 | /* |
1353 | * If btrfs_do_readpage() failed we will want to submit the assembled | |
1354 | * bio to do the cleanup. | |
1355 | */ | |
722c82ac | 1356 | submit_one_bio(&bio_ctrl); |
7aab8b32 CH |
1357 | return ret; |
1358 | } | |
1359 | ||
b6660e80 | 1360 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
1361 | u64 start, u64 end, |
1362 | struct extent_map **em_cached, | |
1363 | struct btrfs_bio_ctrl *bio_ctrl, | |
1364 | u64 *prev_em_start) | |
9974090b | 1365 | { |
23d31bd4 | 1366 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
1367 | int index; |
1368 | ||
b272ae22 | 1369 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
1370 | |
1371 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 1372 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
c000bc04 | 1373 | prev_em_start); |
09cbfeaf | 1374 | put_page(pages[index]); |
9974090b MX |
1375 | } |
1376 | } | |
1377 | ||
d1310b2e | 1378 | /* |
40f76580 CM |
1379 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
1380 | * | |
5eaad97a | 1381 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
1382 | * to write the page (copy into inline extent). In this case the IO has |
1383 | * been started and the page is already unlocked. | |
1384 | * | |
1385 | * This returns 0 if all went well (page still locked) | |
1386 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 1387 | */ |
cd4c0bf9 | 1388 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
83f1b680 | 1389 | struct page *page, struct writeback_control *wbc) |
40f76580 | 1390 | { |
2749f7ef | 1391 | const u64 page_end = page_offset(page) + PAGE_SIZE - 1; |
cf3075fb | 1392 | u64 delalloc_start = page_offset(page); |
40f76580 | 1393 | u64 delalloc_to_write = 0; |
83f1b680 QW |
1394 | /* How many pages are started by btrfs_run_delalloc_range() */ |
1395 | unsigned long nr_written = 0; | |
40f76580 CM |
1396 | int ret; |
1397 | int page_started = 0; | |
1398 | ||
2749f7ef QW |
1399 | while (delalloc_start < page_end) { |
1400 | u64 delalloc_end = page_end; | |
1401 | bool found; | |
40f76580 | 1402 | |
cd4c0bf9 | 1403 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 1404 | &delalloc_start, |
917aacec | 1405 | &delalloc_end); |
3522e903 | 1406 | if (!found) { |
40f76580 CM |
1407 | delalloc_start = delalloc_end + 1; |
1408 | continue; | |
1409 | } | |
cd4c0bf9 | 1410 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
83f1b680 | 1411 | delalloc_end, &page_started, &nr_written, wbc); |
40f76580 | 1412 | if (ret) { |
963e4db8 QW |
1413 | btrfs_page_set_error(inode->root->fs_info, page, |
1414 | page_offset(page), PAGE_SIZE); | |
7361b4ae | 1415 | return ret; |
40f76580 CM |
1416 | } |
1417 | /* | |
ea1754a0 KS |
1418 | * delalloc_end is already one less than the total length, so |
1419 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
1420 | */ |
1421 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 1422 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
1423 | delalloc_start = delalloc_end + 1; |
1424 | } | |
1425 | if (wbc->nr_to_write < delalloc_to_write) { | |
1426 | int thresh = 8192; | |
1427 | ||
1428 | if (delalloc_to_write < thresh * 2) | |
1429 | thresh = delalloc_to_write; | |
1430 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
1431 | thresh); | |
1432 | } | |
1433 | ||
83f1b680 | 1434 | /* Did btrfs_run_dealloc_range() already unlock and start the IO? */ |
40f76580 CM |
1435 | if (page_started) { |
1436 | /* | |
83f1b680 QW |
1437 | * We've unlocked the page, so we can't update the mapping's |
1438 | * writeback index, just update nr_to_write. | |
40f76580 | 1439 | */ |
83f1b680 | 1440 | wbc->nr_to_write -= nr_written; |
40f76580 CM |
1441 | return 1; |
1442 | } | |
1443 | ||
b69d1ee9 | 1444 | return 0; |
40f76580 CM |
1445 | } |
1446 | ||
c5ef5c6c QW |
1447 | /* |
1448 | * Find the first byte we need to write. | |
1449 | * | |
1450 | * For subpage, one page can contain several sectors, and | |
1451 | * __extent_writepage_io() will just grab all extent maps in the page | |
1452 | * range and try to submit all non-inline/non-compressed extents. | |
1453 | * | |
1454 | * This is a big problem for subpage, we shouldn't re-submit already written | |
1455 | * data at all. | |
1456 | * This function will lookup subpage dirty bit to find which range we really | |
1457 | * need to submit. | |
1458 | * | |
1459 | * Return the next dirty range in [@start, @end). | |
1460 | * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE. | |
1461 | */ | |
1462 | static void find_next_dirty_byte(struct btrfs_fs_info *fs_info, | |
1463 | struct page *page, u64 *start, u64 *end) | |
1464 | { | |
1465 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
72a69cd0 | 1466 | struct btrfs_subpage_info *spi = fs_info->subpage_info; |
c5ef5c6c QW |
1467 | u64 orig_start = *start; |
1468 | /* Declare as unsigned long so we can use bitmap ops */ | |
c5ef5c6c | 1469 | unsigned long flags; |
72a69cd0 | 1470 | int range_start_bit; |
c5ef5c6c QW |
1471 | int range_end_bit; |
1472 | ||
1473 | /* | |
1474 | * For regular sector size == page size case, since one page only | |
1475 | * contains one sector, we return the page offset directly. | |
1476 | */ | |
fbca46eb | 1477 | if (!btrfs_is_subpage(fs_info, page)) { |
c5ef5c6c QW |
1478 | *start = page_offset(page); |
1479 | *end = page_offset(page) + PAGE_SIZE; | |
1480 | return; | |
1481 | } | |
1482 | ||
72a69cd0 QW |
1483 | range_start_bit = spi->dirty_offset + |
1484 | (offset_in_page(orig_start) >> fs_info->sectorsize_bits); | |
1485 | ||
c5ef5c6c QW |
1486 | /* We should have the page locked, but just in case */ |
1487 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
1488 | bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit, |
1489 | spi->dirty_offset + spi->bitmap_nr_bits); | |
c5ef5c6c QW |
1490 | spin_unlock_irqrestore(&subpage->lock, flags); |
1491 | ||
72a69cd0 QW |
1492 | range_start_bit -= spi->dirty_offset; |
1493 | range_end_bit -= spi->dirty_offset; | |
1494 | ||
c5ef5c6c QW |
1495 | *start = page_offset(page) + range_start_bit * fs_info->sectorsize; |
1496 | *end = page_offset(page) + range_end_bit * fs_info->sectorsize; | |
1497 | } | |
1498 | ||
40f76580 CM |
1499 | /* |
1500 | * helper for __extent_writepage. This calls the writepage start hooks, | |
1501 | * and does the loop to map the page into extents and bios. | |
1502 | * | |
1503 | * We return 1 if the IO is started and the page is unlocked, | |
1504 | * 0 if all went well (page still locked) | |
1505 | * < 0 if there were errors (page still locked) | |
1506 | */ | |
d4580fe2 | 1507 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 | 1508 | struct page *page, |
ee5f017d | 1509 | struct btrfs_bio_ctrl *bio_ctrl, |
40f76580 | 1510 | loff_t i_size, |
57e5ffeb | 1511 | int *nr_ret) |
d1310b2e | 1512 | { |
6bc5636a | 1513 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a129ffb8 QW |
1514 | u64 cur = page_offset(page); |
1515 | u64 end = cur + PAGE_SIZE - 1; | |
d1310b2e | 1516 | u64 extent_offset; |
d1310b2e | 1517 | u64 block_start; |
d1310b2e | 1518 | struct extent_map *em; |
44e5801f | 1519 | int saved_ret = 0; |
40f76580 CM |
1520 | int ret = 0; |
1521 | int nr = 0; | |
44e5801f | 1522 | bool has_error = false; |
40f76580 | 1523 | bool compressed; |
c8b97818 | 1524 | |
a129ffb8 | 1525 | ret = btrfs_writepage_cow_fixup(page); |
d75855b4 NB |
1526 | if (ret) { |
1527 | /* Fixup worker will requeue */ | |
72b505dc | 1528 | redirty_page_for_writepage(bio_ctrl->wbc, page); |
d75855b4 NB |
1529 | unlock_page(page); |
1530 | return 1; | |
247e743c CM |
1531 | } |
1532 | ||
11c8349b CM |
1533 | /* |
1534 | * we don't want to touch the inode after unlocking the page, | |
1535 | * so we update the mapping writeback index now | |
1536 | */ | |
72b505dc | 1537 | bio_ctrl->wbc->nr_to_write--; |
771ed689 | 1538 | |
ee5f017d | 1539 | bio_ctrl->end_io_func = end_bio_extent_writepage; |
d1310b2e | 1540 | while (cur <= end) { |
0c64c33c | 1541 | u64 disk_bytenr; |
40f76580 | 1542 | u64 em_end; |
c5ef5c6c QW |
1543 | u64 dirty_range_start = cur; |
1544 | u64 dirty_range_end; | |
6bc5636a | 1545 | u32 iosize; |
58409edd | 1546 | |
40f76580 | 1547 | if (cur >= i_size) { |
38a39ac7 | 1548 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
25c1252a | 1549 | end, true); |
cc1d0d93 QW |
1550 | /* |
1551 | * This range is beyond i_size, thus we don't need to | |
1552 | * bother writing back. | |
1553 | * But we still need to clear the dirty subpage bit, or | |
1554 | * the next time the page gets dirtied, we will try to | |
1555 | * writeback the sectors with subpage dirty bits, | |
1556 | * causing writeback without ordered extent. | |
1557 | */ | |
1558 | btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur); | |
d1310b2e CM |
1559 | break; |
1560 | } | |
c5ef5c6c QW |
1561 | |
1562 | find_next_dirty_byte(fs_info, page, &dirty_range_start, | |
1563 | &dirty_range_end); | |
1564 | if (cur < dirty_range_start) { | |
1565 | cur = dirty_range_start; | |
1566 | continue; | |
1567 | } | |
1568 | ||
d4580fe2 | 1569 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c0347550 | 1570 | if (IS_ERR(em)) { |
c5ef5c6c | 1571 | btrfs_page_set_error(fs_info, page, cur, end - cur + 1); |
61391d56 | 1572 | ret = PTR_ERR_OR_ZERO(em); |
44e5801f QW |
1573 | has_error = true; |
1574 | if (!saved_ret) | |
1575 | saved_ret = ret; | |
d1310b2e CM |
1576 | break; |
1577 | } | |
1578 | ||
1579 | extent_offset = cur - em->start; | |
40f76580 | 1580 | em_end = extent_map_end(em); |
6bc5636a QW |
1581 | ASSERT(cur <= em_end); |
1582 | ASSERT(cur < end); | |
1583 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
1584 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 1585 | block_start = em->block_start; |
c8b97818 | 1586 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
1587 | disk_bytenr = em->block_start + extent_offset; |
1588 | ||
c5ef5c6c QW |
1589 | /* |
1590 | * Note that em_end from extent_map_end() and dirty_range_end from | |
1591 | * find_next_dirty_byte() are all exclusive | |
1592 | */ | |
1593 | iosize = min(min(em_end, end + 1), dirty_range_end) - cur; | |
d1310b2e CM |
1594 | free_extent_map(em); |
1595 | em = NULL; | |
1596 | ||
c8b97818 CM |
1597 | /* |
1598 | * compressed and inline extents are written through other | |
1599 | * paths in the FS | |
1600 | */ | |
1601 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 1602 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 1603 | if (compressed) |
c8b97818 | 1604 | nr++; |
c8b04030 | 1605 | else |
38a39ac7 | 1606 | btrfs_writepage_endio_finish_ordered(inode, |
25c1252a | 1607 | page, cur, cur + iosize - 1, true); |
cc1d0d93 | 1608 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); |
c8b97818 | 1609 | cur += iosize; |
d1310b2e CM |
1610 | continue; |
1611 | } | |
c8b97818 | 1612 | |
d2a91064 | 1613 | btrfs_set_range_writeback(inode, cur, cur + iosize - 1); |
58409edd | 1614 | if (!PageWriteback(page)) { |
d4580fe2 | 1615 | btrfs_err(inode->root->fs_info, |
58409edd DS |
1616 | "page %lu not writeback, cur %llu end %llu", |
1617 | page->index, cur, end); | |
d1310b2e | 1618 | } |
7f3c74fb | 1619 | |
c5ef5c6c QW |
1620 | /* |
1621 | * Although the PageDirty bit is cleared before entering this | |
1622 | * function, subpage dirty bit is not cleared. | |
1623 | * So clear subpage dirty bit here so next time we won't submit | |
1624 | * page for range already written to disk. | |
1625 | */ | |
1626 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); | |
1627 | ||
72b505dc | 1628 | ret = submit_extent_page(bio_ctrl, disk_bytenr, page, |
c000bc04 | 1629 | iosize, cur - page_offset(page), 0); |
fe01aa65 | 1630 | if (ret) { |
44e5801f QW |
1631 | has_error = true; |
1632 | if (!saved_ret) | |
1633 | saved_ret = ret; | |
1634 | ||
c5ef5c6c | 1635 | btrfs_page_set_error(fs_info, page, cur, iosize); |
fe01aa65 | 1636 | if (PageWriteback(page)) |
c5ef5c6c QW |
1637 | btrfs_page_clear_writeback(fs_info, page, cur, |
1638 | iosize); | |
fe01aa65 | 1639 | } |
d1310b2e | 1640 | |
6bc5636a | 1641 | cur += iosize; |
d1310b2e CM |
1642 | nr++; |
1643 | } | |
cc1d0d93 QW |
1644 | /* |
1645 | * If we finish without problem, we should not only clear page dirty, | |
1646 | * but also empty subpage dirty bits | |
1647 | */ | |
44e5801f | 1648 | if (!has_error) |
cc1d0d93 | 1649 | btrfs_page_assert_not_dirty(fs_info, page); |
44e5801f QW |
1650 | else |
1651 | ret = saved_ret; | |
40f76580 | 1652 | *nr_ret = nr; |
40f76580 CM |
1653 | return ret; |
1654 | } | |
1655 | ||
1656 | /* | |
1657 | * the writepage semantics are similar to regular writepage. extent | |
1658 | * records are inserted to lock ranges in the tree, and as dirty areas | |
1659 | * are found, they are marked writeback. Then the lock bits are removed | |
1660 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
1661 | * |
1662 | * Return 0 if everything goes well. | |
1663 | * Return <0 for error. | |
40f76580 | 1664 | */ |
72b505dc | 1665 | static int __extent_writepage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl) |
40f76580 | 1666 | { |
8e1dec8e | 1667 | struct folio *folio = page_folio(page); |
40f76580 | 1668 | struct inode *inode = page->mapping->host; |
e55a0de1 | 1669 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
cf3075fb QW |
1670 | const u64 page_start = page_offset(page); |
1671 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
40f76580 CM |
1672 | int ret; |
1673 | int nr = 0; | |
eb70d222 | 1674 | size_t pg_offset; |
40f76580 | 1675 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 1676 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 | 1677 | |
72b505dc | 1678 | trace___extent_writepage(page, inode, bio_ctrl->wbc); |
40f76580 CM |
1679 | |
1680 | WARN_ON(!PageLocked(page)); | |
1681 | ||
963e4db8 QW |
1682 | btrfs_page_clear_error(btrfs_sb(inode->i_sb), page, |
1683 | page_offset(page), PAGE_SIZE); | |
40f76580 | 1684 | |
7073017a | 1685 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
1686 | if (page->index > end_index || |
1687 | (page->index == end_index && !pg_offset)) { | |
8e1dec8e MWO |
1688 | folio_invalidate(folio, 0, folio_size(folio)); |
1689 | folio_unlock(folio); | |
40f76580 CM |
1690 | return 0; |
1691 | } | |
1692 | ||
21a8935e | 1693 | if (page->index == end_index) |
d048b9c2 | 1694 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 | 1695 | |
32443de3 QW |
1696 | ret = set_page_extent_mapped(page); |
1697 | if (ret < 0) { | |
1698 | SetPageError(page); | |
1699 | goto done; | |
1700 | } | |
40f76580 | 1701 | |
ee5f017d | 1702 | if (!bio_ctrl->extent_locked) { |
72b505dc | 1703 | ret = writepage_delalloc(BTRFS_I(inode), page, bio_ctrl->wbc); |
7789a55a | 1704 | if (ret == 1) |
169d2c87 | 1705 | return 0; |
7789a55a NB |
1706 | if (ret) |
1707 | goto done; | |
1708 | } | |
40f76580 | 1709 | |
72b505dc | 1710 | ret = __extent_writepage_io(BTRFS_I(inode), page, bio_ctrl, i_size, &nr); |
40f76580 | 1711 | if (ret == 1) |
169d2c87 | 1712 | return 0; |
40f76580 | 1713 | |
d1310b2e CM |
1714 | done: |
1715 | if (nr == 0) { | |
1716 | /* make sure the mapping tag for page dirty gets cleared */ | |
1717 | set_page_writeback(page); | |
1718 | end_page_writeback(page); | |
1719 | } | |
963e4db8 QW |
1720 | /* |
1721 | * Here we used to have a check for PageError() and then set @ret and | |
1722 | * call end_extent_writepage(). | |
1723 | * | |
1724 | * But in fact setting @ret here will cause different error paths | |
1725 | * between subpage and regular sectorsize. | |
1726 | * | |
1727 | * For regular page size, we never submit current page, but only add | |
1728 | * current page to current bio. | |
1729 | * The bio submission can only happen in next page. | |
1730 | * Thus if we hit the PageError() branch, @ret is already set to | |
1731 | * non-zero value and will not get updated for regular sectorsize. | |
1732 | * | |
1733 | * But for subpage case, it's possible we submit part of current page, | |
1734 | * thus can get PageError() set by submitted bio of the same page, | |
1735 | * while our @ret is still 0. | |
1736 | * | |
1737 | * So here we unify the behavior and don't set @ret. | |
1738 | * Error can still be properly passed to higher layer as page will | |
1739 | * be set error, here we just don't handle the IO failure. | |
1740 | * | |
1741 | * NOTE: This is just a hotfix for subpage. | |
1742 | * The root fix will be properly ending ordered extent when we hit | |
1743 | * an error during writeback. | |
1744 | * | |
1745 | * But that needs a bigger refactoring, as we not only need to grab the | |
1746 | * submitted OE, but also need to know exactly at which bytenr we hit | |
1747 | * the error. | |
1748 | * Currently the full page based __extent_writepage_io() is not | |
1749 | * capable of that. | |
1750 | */ | |
1751 | if (PageError(page)) | |
cf3075fb | 1752 | end_extent_writepage(page, ret, page_start, page_end); |
ee5f017d | 1753 | if (bio_ctrl->extent_locked) { |
72b505dc CH |
1754 | struct writeback_control *wbc = bio_ctrl->wbc; |
1755 | ||
e55a0de1 | 1756 | /* |
ee5f017d | 1757 | * If bio_ctrl->extent_locked, it's from extent_write_locked_range(), |
e55a0de1 QW |
1758 | * the page can either be locked by lock_page() or |
1759 | * process_one_page(). | |
1760 | * Let btrfs_page_unlock_writer() handle both cases. | |
1761 | */ | |
1762 | ASSERT(wbc); | |
1763 | btrfs_page_unlock_writer(fs_info, page, wbc->range_start, | |
1764 | wbc->range_end + 1 - wbc->range_start); | |
1765 | } else { | |
1766 | unlock_page(page); | |
1767 | } | |
3065976b | 1768 | ASSERT(ret <= 0); |
40f76580 | 1769 | return ret; |
d1310b2e CM |
1770 | } |
1771 | ||
fd8b2b61 | 1772 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 1773 | { |
74316201 N |
1774 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
1775 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
1776 | } |
1777 | ||
18dfa711 FM |
1778 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
1779 | { | |
1780 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
1781 | smp_mb__after_atomic(); | |
1782 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
1783 | } | |
1784 | ||
2e3c2513 | 1785 | /* |
a3efb2f0 | 1786 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 1787 | * |
a3efb2f0 QW |
1788 | * May try to flush write bio if we can't get the lock. |
1789 | * | |
1790 | * Return 0 if the extent buffer doesn't need to be submitted. | |
1791 | * (E.g. the extent buffer is not dirty) | |
1792 | * Return >0 is the extent buffer is submitted to bio. | |
1793 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 1794 | */ |
9df76fb5 | 1795 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
ee5f017d | 1796 | struct btrfs_bio_ctrl *bio_ctrl) |
0b32f4bb | 1797 | { |
9df76fb5 | 1798 | struct btrfs_fs_info *fs_info = eb->fs_info; |
c9583ada | 1799 | int i, num_pages; |
0b32f4bb JB |
1800 | int flush = 0; |
1801 | int ret = 0; | |
1802 | ||
1803 | if (!btrfs_try_tree_write_lock(eb)) { | |
ee5f017d | 1804 | submit_write_bio(bio_ctrl, 0); |
2e3c2513 | 1805 | flush = 1; |
0b32f4bb JB |
1806 | btrfs_tree_lock(eb); |
1807 | } | |
1808 | ||
1809 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
1810 | btrfs_tree_unlock(eb); | |
72b505dc | 1811 | if (bio_ctrl->wbc->sync_mode != WB_SYNC_ALL) |
0b32f4bb JB |
1812 | return 0; |
1813 | if (!flush) { | |
ee5f017d | 1814 | submit_write_bio(bio_ctrl, 0); |
0b32f4bb JB |
1815 | flush = 1; |
1816 | } | |
a098d8e8 CM |
1817 | while (1) { |
1818 | wait_on_extent_buffer_writeback(eb); | |
1819 | btrfs_tree_lock(eb); | |
1820 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
1821 | break; | |
0b32f4bb | 1822 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
1823 | } |
1824 | } | |
1825 | ||
51561ffe JB |
1826 | /* |
1827 | * We need to do this to prevent races in people who check if the eb is | |
1828 | * under IO since we can end up having no IO bits set for a short period | |
1829 | * of time. | |
1830 | */ | |
1831 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
1832 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
1833 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 1834 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 1835 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
1836 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
1837 | -eb->len, | |
1838 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 1839 | ret = 1; |
51561ffe JB |
1840 | } else { |
1841 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
1842 | } |
1843 | ||
1844 | btrfs_tree_unlock(eb); | |
1845 | ||
f3156df9 QW |
1846 | /* |
1847 | * Either we don't need to submit any tree block, or we're submitting | |
1848 | * subpage eb. | |
1849 | * Subpage metadata doesn't use page locking at all, so we can skip | |
1850 | * the page locking. | |
1851 | */ | |
fbca46eb | 1852 | if (!ret || fs_info->nodesize < PAGE_SIZE) |
0b32f4bb JB |
1853 | return ret; |
1854 | ||
65ad0104 | 1855 | num_pages = num_extent_pages(eb); |
0b32f4bb | 1856 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 1857 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
1858 | |
1859 | if (!trylock_page(p)) { | |
1860 | if (!flush) { | |
ee5f017d | 1861 | submit_write_bio(bio_ctrl, 0); |
0b32f4bb JB |
1862 | flush = 1; |
1863 | } | |
1864 | lock_page(p); | |
1865 | } | |
1866 | } | |
1867 | ||
2e3c2513 | 1868 | return ret; |
0b32f4bb JB |
1869 | } |
1870 | ||
5a2c6075 | 1871 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 1872 | { |
5a2c6075 | 1873 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 1874 | |
5a2c6075 | 1875 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
1876 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
1877 | return; | |
1878 | ||
c2e39305 JB |
1879 | /* |
1880 | * A read may stumble upon this buffer later, make sure that it gets an | |
1881 | * error and knows there was an error. | |
1882 | */ | |
1883 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
1884 | ||
68b85589 JB |
1885 | /* |
1886 | * We need to set the mapping with the io error as well because a write | |
1887 | * error will flip the file system readonly, and then syncfs() will | |
1888 | * return a 0 because we are readonly if we don't modify the err seq for | |
1889 | * the superblock. | |
1890 | */ | |
1891 | mapping_set_error(page->mapping, -EIO); | |
1892 | ||
656f30db FM |
1893 | /* |
1894 | * If writeback for a btree extent that doesn't belong to a log tree | |
1895 | * failed, increment the counter transaction->eb_write_errors. | |
1896 | * We do this because while the transaction is running and before it's | |
1897 | * committing (when we call filemap_fdata[write|wait]_range against | |
1898 | * the btree inode), we might have | |
1899 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
1900 | * returns an error or an error happens during writeback, when we're | |
1901 | * committing the transaction we wouldn't know about it, since the pages | |
1902 | * can be no longer dirty nor marked anymore for writeback (if a | |
1903 | * subsequent modification to the extent buffer didn't happen before the | |
1904 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
1905 | * able to find the pages tagged with SetPageError at transaction | |
1906 | * commit time. So if this happens we must abort the transaction, | |
1907 | * otherwise we commit a super block with btree roots that point to | |
1908 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
1909 | * or the content of some node/leaf from a past generation that got | |
1910 | * cowed or deleted and is no longer valid. | |
1911 | * | |
1912 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
1913 | * not be enough - we need to distinguish between log tree extents vs | |
1914 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
1915 | * will catch and clear such errors in the mapping - and that call might | |
1916 | * be from a log sync and not from a transaction commit. Also, checking | |
1917 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
1918 | * not done and would not be reliable - the eb might have been released | |
1919 | * from memory and reading it back again means that flag would not be | |
1920 | * set (since it's a runtime flag, not persisted on disk). | |
1921 | * | |
1922 | * Using the flags below in the btree inode also makes us achieve the | |
1923 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
1924 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
1925 | * is called, the writeback for all dirty pages had already finished | |
1926 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
1927 | * filemap_fdatawait_range() would return success, as it could not know | |
1928 | * that writeback errors happened (the pages were no longer tagged for | |
1929 | * writeback). | |
1930 | */ | |
1931 | switch (eb->log_index) { | |
1932 | case -1: | |
5a2c6075 | 1933 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
1934 | break; |
1935 | case 0: | |
5a2c6075 | 1936 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
1937 | break; |
1938 | case 1: | |
5a2c6075 | 1939 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
1940 | break; |
1941 | default: | |
1942 | BUG(); /* unexpected, logic error */ | |
1943 | } | |
1944 | } | |
1945 | ||
2f3186d8 QW |
1946 | /* |
1947 | * The endio specific version which won't touch any unsafe spinlock in endio | |
1948 | * context. | |
1949 | */ | |
1950 | static struct extent_buffer *find_extent_buffer_nolock( | |
1951 | struct btrfs_fs_info *fs_info, u64 start) | |
1952 | { | |
1953 | struct extent_buffer *eb; | |
1954 | ||
1955 | rcu_read_lock(); | |
01cd3909 DS |
1956 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
1957 | start >> fs_info->sectorsize_bits); | |
2f3186d8 QW |
1958 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
1959 | rcu_read_unlock(); | |
1960 | return eb; | |
1961 | } | |
1962 | rcu_read_unlock(); | |
1963 | return NULL; | |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * The endio function for subpage extent buffer write. | |
1968 | * | |
1969 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
1970 | * after all extent buffers in the page has finished their writeback. | |
1971 | */ | |
917f32a2 | 1972 | static void end_bio_subpage_eb_writepage(struct btrfs_bio *bbio) |
2f3186d8 | 1973 | { |
917f32a2 | 1974 | struct bio *bio = &bbio->bio; |
fa04c165 | 1975 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
1976 | struct bio_vec *bvec; |
1977 | struct bvec_iter_all iter_all; | |
1978 | ||
fa04c165 | 1979 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
fbca46eb | 1980 | ASSERT(fs_info->nodesize < PAGE_SIZE); |
fa04c165 | 1981 | |
2f3186d8 QW |
1982 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
1983 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
1984 | struct page *page = bvec->bv_page; | |
1985 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
1986 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
1987 | u64 cur_bytenr = bvec_start; | |
1988 | ||
1989 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
1990 | ||
1991 | /* Iterate through all extent buffers in the range */ | |
1992 | while (cur_bytenr <= bvec_end) { | |
1993 | struct extent_buffer *eb; | |
1994 | int done; | |
1995 | ||
1996 | /* | |
1997 | * Here we can't use find_extent_buffer(), as it may | |
1998 | * try to lock eb->refs_lock, which is not safe in endio | |
1999 | * context. | |
2000 | */ | |
2001 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
2002 | ASSERT(eb); | |
2003 | ||
2004 | cur_bytenr = eb->start + eb->len; | |
2005 | ||
2006 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
2007 | done = atomic_dec_and_test(&eb->io_pages); | |
2008 | ASSERT(done); | |
2009 | ||
2010 | if (bio->bi_status || | |
2011 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
2012 | ClearPageUptodate(page); | |
2013 | set_btree_ioerr(page, eb); | |
2014 | } | |
2015 | ||
2016 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
2017 | eb->len); | |
2018 | end_extent_buffer_writeback(eb); | |
2019 | /* | |
2020 | * free_extent_buffer() will grab spinlock which is not | |
2021 | * safe in endio context. Thus here we manually dec | |
2022 | * the ref. | |
2023 | */ | |
2024 | atomic_dec(&eb->refs); | |
2025 | } | |
2026 | } | |
2027 | bio_put(bio); | |
2028 | } | |
2029 | ||
917f32a2 | 2030 | static void end_bio_extent_buffer_writepage(struct btrfs_bio *bbio) |
0b32f4bb | 2031 | { |
917f32a2 | 2032 | struct bio *bio = &bbio->bio; |
2c30c71b | 2033 | struct bio_vec *bvec; |
0b32f4bb | 2034 | struct extent_buffer *eb; |
2b070cfe | 2035 | int done; |
6dc4f100 | 2036 | struct bvec_iter_all iter_all; |
0b32f4bb | 2037 | |
c09abff8 | 2038 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2039 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
2040 | struct page *page = bvec->bv_page; |
2041 | ||
0b32f4bb JB |
2042 | eb = (struct extent_buffer *)page->private; |
2043 | BUG_ON(!eb); | |
2044 | done = atomic_dec_and_test(&eb->io_pages); | |
2045 | ||
4e4cbee9 | 2046 | if (bio->bi_status || |
4246a0b6 | 2047 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 2048 | ClearPageUptodate(page); |
5a2c6075 | 2049 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
2050 | } |
2051 | ||
2052 | end_page_writeback(page); | |
2053 | ||
2054 | if (!done) | |
2055 | continue; | |
2056 | ||
2057 | end_extent_buffer_writeback(eb); | |
2c30c71b | 2058 | } |
0b32f4bb JB |
2059 | |
2060 | bio_put(bio); | |
0b32f4bb JB |
2061 | } |
2062 | ||
fa04c165 QW |
2063 | static void prepare_eb_write(struct extent_buffer *eb) |
2064 | { | |
2065 | u32 nritems; | |
2066 | unsigned long start; | |
2067 | unsigned long end; | |
2068 | ||
2069 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
2070 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
2071 | ||
2072 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
2073 | nritems = btrfs_header_nritems(eb); | |
2074 | if (btrfs_header_level(eb) > 0) { | |
e23efd8e | 2075 | end = btrfs_node_key_ptr_offset(eb, nritems); |
fa04c165 QW |
2076 | memzero_extent_buffer(eb, end, eb->len - end); |
2077 | } else { | |
2078 | /* | |
2079 | * Leaf: | |
2080 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
2081 | */ | |
42c9419a | 2082 | start = btrfs_item_nr_offset(eb, nritems); |
8009adf3 | 2083 | end = btrfs_item_nr_offset(eb, 0); |
3a3178c7 JB |
2084 | if (nritems == 0) |
2085 | end += BTRFS_LEAF_DATA_SIZE(eb->fs_info); | |
2086 | else | |
2087 | end += btrfs_item_offset(eb, nritems - 1); | |
fa04c165 QW |
2088 | memzero_extent_buffer(eb, start, end - start); |
2089 | } | |
2090 | } | |
2091 | ||
35b6ddfa QW |
2092 | /* |
2093 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
2094 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
2095 | */ |
2096 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
ee5f017d | 2097 | struct btrfs_bio_ctrl *bio_ctrl) |
35b6ddfa QW |
2098 | { |
2099 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
2100 | struct page *page = eb->pages[0]; | |
35b6ddfa QW |
2101 | bool no_dirty_ebs = false; |
2102 | int ret; | |
2103 | ||
fa04c165 QW |
2104 | prepare_eb_write(eb); |
2105 | ||
35b6ddfa QW |
2106 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
2107 | lock_page(page); | |
2108 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
2109 | ||
2110 | /* Check if this is the last dirty bit to update nr_written */ | |
2111 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
2112 | eb->start, eb->len); | |
2113 | if (no_dirty_ebs) | |
2114 | clear_page_dirty_for_io(page); | |
2115 | ||
ee5f017d | 2116 | bio_ctrl->end_io_func = end_bio_subpage_eb_writepage; |
5467abba | 2117 | |
72b505dc | 2118 | ret = submit_extent_page(bio_ctrl, eb->start, page, eb->len, |
eb8d0c6d | 2119 | eb->start - page_offset(page), 0); |
35b6ddfa QW |
2120 | if (ret) { |
2121 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
2122 | set_btree_ioerr(page, eb); | |
2123 | unlock_page(page); | |
2124 | ||
2125 | if (atomic_dec_and_test(&eb->io_pages)) | |
2126 | end_extent_buffer_writeback(eb); | |
2127 | return -EIO; | |
2128 | } | |
2129 | unlock_page(page); | |
2130 | /* | |
2131 | * Submission finished without problem, if no range of the page is | |
2132 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
2133 | */ | |
2134 | if (no_dirty_ebs) | |
72b505dc | 2135 | bio_ctrl->wbc->nr_to_write--; |
35b6ddfa QW |
2136 | return ret; |
2137 | } | |
2138 | ||
0e378df1 | 2139 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
ee5f017d | 2140 | struct btrfs_bio_ctrl *bio_ctrl) |
0b32f4bb | 2141 | { |
0c64c33c | 2142 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 2143 | int i, num_pages; |
d7dbe9e7 | 2144 | int ret = 0; |
0b32f4bb | 2145 | |
fa04c165 | 2146 | prepare_eb_write(eb); |
35b6ddfa | 2147 | |
ee5f017d | 2148 | bio_ctrl->end_io_func = end_bio_extent_buffer_writepage; |
5467abba | 2149 | |
fa04c165 | 2150 | num_pages = num_extent_pages(eb); |
0b32f4bb | 2151 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2152 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
2153 | |
2154 | clear_page_dirty_for_io(p); | |
2155 | set_page_writeback(p); | |
72b505dc | 2156 | ret = submit_extent_page(bio_ctrl, disk_bytenr, p, |
eb8d0c6d | 2157 | PAGE_SIZE, 0, 0); |
0b32f4bb | 2158 | if (ret) { |
5a2c6075 | 2159 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
2160 | if (PageWriteback(p)) |
2161 | end_page_writeback(p); | |
0b32f4bb JB |
2162 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
2163 | end_extent_buffer_writeback(eb); | |
2164 | ret = -EIO; | |
2165 | break; | |
2166 | } | |
0c64c33c | 2167 | disk_bytenr += PAGE_SIZE; |
72b505dc | 2168 | bio_ctrl->wbc->nr_to_write--; |
0b32f4bb JB |
2169 | unlock_page(p); |
2170 | } | |
2171 | ||
2172 | if (unlikely(ret)) { | |
2173 | for (; i < num_pages; i++) { | |
bbf65cf0 | 2174 | struct page *p = eb->pages[i]; |
81465028 | 2175 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
2176 | unlock_page(p); |
2177 | } | |
2178 | } | |
2179 | ||
2180 | return ret; | |
2181 | } | |
2182 | ||
c4aec299 QW |
2183 | /* |
2184 | * Submit one subpage btree page. | |
2185 | * | |
2186 | * The main difference to submit_eb_page() is: | |
2187 | * - Page locking | |
2188 | * For subpage, we don't rely on page locking at all. | |
2189 | * | |
2190 | * - Flush write bio | |
2191 | * We only flush bio if we may be unable to fit current extent buffers into | |
2192 | * current bio. | |
2193 | * | |
2194 | * Return >=0 for the number of submitted extent buffers. | |
2195 | * Return <0 for fatal error. | |
2196 | */ | |
72b505dc | 2197 | static int submit_eb_subpage(struct page *page, struct btrfs_bio_ctrl *bio_ctrl) |
c4aec299 QW |
2198 | { |
2199 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
2200 | int submitted = 0; | |
2201 | u64 page_start = page_offset(page); | |
2202 | int bit_start = 0; | |
c4aec299 QW |
2203 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; |
2204 | int ret; | |
2205 | ||
2206 | /* Lock and write each dirty extent buffers in the range */ | |
72a69cd0 | 2207 | while (bit_start < fs_info->subpage_info->bitmap_nr_bits) { |
c4aec299 QW |
2208 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; |
2209 | struct extent_buffer *eb; | |
2210 | unsigned long flags; | |
2211 | u64 start; | |
2212 | ||
2213 | /* | |
2214 | * Take private lock to ensure the subpage won't be detached | |
2215 | * in the meantime. | |
2216 | */ | |
2217 | spin_lock(&page->mapping->private_lock); | |
2218 | if (!PagePrivate(page)) { | |
2219 | spin_unlock(&page->mapping->private_lock); | |
2220 | break; | |
2221 | } | |
2222 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
2223 | if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset, |
2224 | subpage->bitmaps)) { | |
c4aec299 QW |
2225 | spin_unlock_irqrestore(&subpage->lock, flags); |
2226 | spin_unlock(&page->mapping->private_lock); | |
2227 | bit_start++; | |
2228 | continue; | |
2229 | } | |
2230 | ||
2231 | start = page_start + bit_start * fs_info->sectorsize; | |
2232 | bit_start += sectors_per_node; | |
2233 | ||
2234 | /* | |
2235 | * Here we just want to grab the eb without touching extra | |
2236 | * spin locks, so call find_extent_buffer_nolock(). | |
2237 | */ | |
2238 | eb = find_extent_buffer_nolock(fs_info, start); | |
2239 | spin_unlock_irqrestore(&subpage->lock, flags); | |
2240 | spin_unlock(&page->mapping->private_lock); | |
2241 | ||
2242 | /* | |
2243 | * The eb has already reached 0 refs thus find_extent_buffer() | |
2244 | * doesn't return it. We don't need to write back such eb | |
2245 | * anyway. | |
2246 | */ | |
2247 | if (!eb) | |
2248 | continue; | |
2249 | ||
72b505dc | 2250 | ret = lock_extent_buffer_for_io(eb, bio_ctrl); |
c4aec299 QW |
2251 | if (ret == 0) { |
2252 | free_extent_buffer(eb); | |
2253 | continue; | |
2254 | } | |
2255 | if (ret < 0) { | |
2256 | free_extent_buffer(eb); | |
2257 | goto cleanup; | |
2258 | } | |
72b505dc | 2259 | ret = write_one_subpage_eb(eb, bio_ctrl); |
c4aec299 QW |
2260 | free_extent_buffer(eb); |
2261 | if (ret < 0) | |
2262 | goto cleanup; | |
2263 | submitted++; | |
2264 | } | |
2265 | return submitted; | |
2266 | ||
2267 | cleanup: | |
2268 | /* We hit error, end bio for the submitted extent buffers */ | |
ee5f017d | 2269 | submit_write_bio(bio_ctrl, ret); |
c4aec299 QW |
2270 | return ret; |
2271 | } | |
2272 | ||
f91e0d0c QW |
2273 | /* |
2274 | * Submit all page(s) of one extent buffer. | |
2275 | * | |
2276 | * @page: the page of one extent buffer | |
2277 | * @eb_context: to determine if we need to submit this page, if current page | |
2278 | * belongs to this eb, we don't need to submit | |
2279 | * | |
2280 | * The caller should pass each page in their bytenr order, and here we use | |
2281 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
2282 | * | |
2283 | * If we have, we just skip until we hit a new page that doesn't belong to | |
2284 | * current @eb_context. | |
2285 | * | |
2286 | * If not, we submit all the page(s) of the extent buffer. | |
2287 | * | |
2288 | * Return >0 if we have submitted the extent buffer successfully. | |
2289 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
2290 | * previous call. | |
2291 | * Return <0 for fatal error. | |
2292 | */ | |
72b505dc | 2293 | static int submit_eb_page(struct page *page, struct btrfs_bio_ctrl *bio_ctrl, |
f91e0d0c QW |
2294 | struct extent_buffer **eb_context) |
2295 | { | |
2296 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 2297 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
2298 | struct extent_buffer *eb; |
2299 | int ret; | |
2300 | ||
2301 | if (!PagePrivate(page)) | |
2302 | return 0; | |
2303 | ||
fbca46eb | 2304 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
72b505dc | 2305 | return submit_eb_subpage(page, bio_ctrl); |
c4aec299 | 2306 | |
f91e0d0c QW |
2307 | spin_lock(&mapping->private_lock); |
2308 | if (!PagePrivate(page)) { | |
2309 | spin_unlock(&mapping->private_lock); | |
2310 | return 0; | |
2311 | } | |
2312 | ||
2313 | eb = (struct extent_buffer *)page->private; | |
2314 | ||
2315 | /* | |
2316 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
2317 | * crashing the machine for something we can survive anyway. | |
2318 | */ | |
2319 | if (WARN_ON(!eb)) { | |
2320 | spin_unlock(&mapping->private_lock); | |
2321 | return 0; | |
2322 | } | |
2323 | ||
2324 | if (eb == *eb_context) { | |
2325 | spin_unlock(&mapping->private_lock); | |
2326 | return 0; | |
2327 | } | |
2328 | ret = atomic_inc_not_zero(&eb->refs); | |
2329 | spin_unlock(&mapping->private_lock); | |
2330 | if (!ret) | |
2331 | return 0; | |
2332 | ||
0bc09ca1 NA |
2333 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
2334 | /* | |
2335 | * If for_sync, this hole will be filled with | |
2336 | * trasnsaction commit. | |
2337 | */ | |
72b505dc CH |
2338 | if (bio_ctrl->wbc->sync_mode == WB_SYNC_ALL && |
2339 | !bio_ctrl->wbc->for_sync) | |
0bc09ca1 NA |
2340 | ret = -EAGAIN; |
2341 | else | |
2342 | ret = 0; | |
2343 | free_extent_buffer(eb); | |
2344 | return ret; | |
2345 | } | |
2346 | ||
f91e0d0c QW |
2347 | *eb_context = eb; |
2348 | ||
72b505dc | 2349 | ret = lock_extent_buffer_for_io(eb, bio_ctrl); |
f91e0d0c | 2350 | if (ret <= 0) { |
0bc09ca1 NA |
2351 | btrfs_revert_meta_write_pointer(cache, eb); |
2352 | if (cache) | |
2353 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
2354 | free_extent_buffer(eb); |
2355 | return ret; | |
2356 | } | |
be1a1d7a | 2357 | if (cache) { |
d3e29967 NB |
2358 | /* |
2359 | * Implies write in zoned mode. Mark the last eb in a block group. | |
2360 | */ | |
56fbb0a4 | 2361 | btrfs_schedule_zone_finish_bg(cache, eb); |
d3e29967 | 2362 | btrfs_put_block_group(cache); |
be1a1d7a | 2363 | } |
72b505dc | 2364 | ret = write_one_eb(eb, bio_ctrl); |
f91e0d0c QW |
2365 | free_extent_buffer(eb); |
2366 | if (ret < 0) | |
2367 | return ret; | |
2368 | return 1; | |
2369 | } | |
2370 | ||
0b32f4bb JB |
2371 | int btree_write_cache_pages(struct address_space *mapping, |
2372 | struct writeback_control *wbc) | |
2373 | { | |
f91e0d0c | 2374 | struct extent_buffer *eb_context = NULL; |
ee5f017d | 2375 | struct btrfs_bio_ctrl bio_ctrl = { |
72b505dc | 2376 | .wbc = wbc, |
c000bc04 | 2377 | .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc), |
0b32f4bb | 2378 | .extent_locked = 0, |
0b32f4bb | 2379 | }; |
b3ff8f1d | 2380 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
2381 | int ret = 0; |
2382 | int done = 0; | |
2383 | int nr_to_write_done = 0; | |
51c5cd3b VMO |
2384 | struct folio_batch fbatch; |
2385 | unsigned int nr_folios; | |
0b32f4bb JB |
2386 | pgoff_t index; |
2387 | pgoff_t end; /* Inclusive */ | |
2388 | int scanned = 0; | |
10bbd235 | 2389 | xa_mark_t tag; |
0b32f4bb | 2390 | |
51c5cd3b | 2391 | folio_batch_init(&fbatch); |
0b32f4bb JB |
2392 | if (wbc->range_cyclic) { |
2393 | index = mapping->writeback_index; /* Start from prev offset */ | |
2394 | end = -1; | |
556755a8 JB |
2395 | /* |
2396 | * Start from the beginning does not need to cycle over the | |
2397 | * range, mark it as scanned. | |
2398 | */ | |
2399 | scanned = (index == 0); | |
0b32f4bb | 2400 | } else { |
09cbfeaf KS |
2401 | index = wbc->range_start >> PAGE_SHIFT; |
2402 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
2403 | scanned = 1; |
2404 | } | |
2405 | if (wbc->sync_mode == WB_SYNC_ALL) | |
2406 | tag = PAGECACHE_TAG_TOWRITE; | |
2407 | else | |
2408 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 2409 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
2410 | retry: |
2411 | if (wbc->sync_mode == WB_SYNC_ALL) | |
2412 | tag_pages_for_writeback(mapping, index, end); | |
2413 | while (!done && !nr_to_write_done && (index <= end) && | |
51c5cd3b VMO |
2414 | (nr_folios = filemap_get_folios_tag(mapping, &index, end, |
2415 | tag, &fbatch))) { | |
0b32f4bb JB |
2416 | unsigned i; |
2417 | ||
51c5cd3b VMO |
2418 | for (i = 0; i < nr_folios; i++) { |
2419 | struct folio *folio = fbatch.folios[i]; | |
0b32f4bb | 2420 | |
72b505dc | 2421 | ret = submit_eb_page(&folio->page, &bio_ctrl, &eb_context); |
f91e0d0c | 2422 | if (ret == 0) |
0b32f4bb | 2423 | continue; |
f91e0d0c | 2424 | if (ret < 0) { |
0b32f4bb | 2425 | done = 1; |
0b32f4bb JB |
2426 | break; |
2427 | } | |
0b32f4bb JB |
2428 | |
2429 | /* | |
2430 | * the filesystem may choose to bump up nr_to_write. | |
2431 | * We have to make sure to honor the new nr_to_write | |
2432 | * at any time | |
2433 | */ | |
2434 | nr_to_write_done = wbc->nr_to_write <= 0; | |
2435 | } | |
51c5cd3b | 2436 | folio_batch_release(&fbatch); |
0b32f4bb JB |
2437 | cond_resched(); |
2438 | } | |
2439 | if (!scanned && !done) { | |
2440 | /* | |
2441 | * We hit the last page and there is more work to be done: wrap | |
2442 | * back to the start of the file | |
2443 | */ | |
2444 | scanned = 1; | |
2445 | index = 0; | |
2446 | goto retry; | |
2447 | } | |
b3ff8f1d QW |
2448 | /* |
2449 | * If something went wrong, don't allow any metadata write bio to be | |
2450 | * submitted. | |
2451 | * | |
2452 | * This would prevent use-after-free if we had dirty pages not | |
2453 | * cleaned up, which can still happen by fuzzed images. | |
2454 | * | |
2455 | * - Bad extent tree | |
2456 | * Allowing existing tree block to be allocated for other trees. | |
2457 | * | |
2458 | * - Log tree operations | |
2459 | * Exiting tree blocks get allocated to log tree, bumps its | |
2460 | * generation, then get cleaned in tree re-balance. | |
2461 | * Such tree block will not be written back, since it's clean, | |
2462 | * thus no WRITTEN flag set. | |
2463 | * And after log writes back, this tree block is not traced by | |
2464 | * any dirty extent_io_tree. | |
2465 | * | |
2466 | * - Offending tree block gets re-dirtied from its original owner | |
2467 | * Since it has bumped generation, no WRITTEN flag, it can be | |
2468 | * reused without COWing. This tree block will not be traced | |
2469 | * by btrfs_transaction::dirty_pages. | |
2470 | * | |
2471 | * Now such dirty tree block will not be cleaned by any dirty | |
2472 | * extent io tree. Thus we don't want to submit such wild eb | |
2473 | * if the fs already has error. | |
9845e5dd | 2474 | * |
c9583ada QW |
2475 | * We can get ret > 0 from submit_extent_page() indicating how many ebs |
2476 | * were submitted. Reset it to 0 to avoid false alerts for the caller. | |
2477 | */ | |
2478 | if (ret > 0) | |
2479 | ret = 0; | |
9845e5dd CH |
2480 | if (!ret && BTRFS_FS_ERROR(fs_info)) |
2481 | ret = -EROFS; | |
ee5f017d | 2482 | submit_write_bio(&bio_ctrl, ret); |
9845e5dd CH |
2483 | |
2484 | btrfs_zoned_meta_io_unlock(fs_info); | |
0b32f4bb JB |
2485 | return ret; |
2486 | } | |
2487 | ||
43dd529a | 2488 | /* |
3bed2da1 NB |
2489 | * Walk the list of dirty pages of the given address space and write all of them. |
2490 | * | |
ee5f017d DS |
2491 | * @mapping: address space structure to write |
2492 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2493 | * @bio_ctrl: holds context for the write, namely the bio | |
d1310b2e CM |
2494 | * |
2495 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2496 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2497 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2498 | * and msync() need to guarantee that all the data which was dirty at the time | |
2499 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2500 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2501 | * existing IO to complete. | |
2502 | */ | |
4242b64a | 2503 | static int extent_write_cache_pages(struct address_space *mapping, |
ee5f017d | 2504 | struct btrfs_bio_ctrl *bio_ctrl) |
d1310b2e | 2505 | { |
72b505dc | 2506 | struct writeback_control *wbc = bio_ctrl->wbc; |
7fd1a3f7 | 2507 | struct inode *inode = mapping->host; |
d1310b2e CM |
2508 | int ret = 0; |
2509 | int done = 0; | |
f85d7d6c | 2510 | int nr_to_write_done = 0; |
9f50fd2e VMO |
2511 | struct folio_batch fbatch; |
2512 | unsigned int nr_folios; | |
d1310b2e CM |
2513 | pgoff_t index; |
2514 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
2515 | pgoff_t done_index; |
2516 | int range_whole = 0; | |
d1310b2e | 2517 | int scanned = 0; |
10bbd235 | 2518 | xa_mark_t tag; |
d1310b2e | 2519 | |
7fd1a3f7 JB |
2520 | /* |
2521 | * We have to hold onto the inode so that ordered extents can do their | |
2522 | * work when the IO finishes. The alternative to this is failing to add | |
2523 | * an ordered extent if the igrab() fails there and that is a huge pain | |
2524 | * to deal with, so instead just hold onto the inode throughout the | |
2525 | * writepages operation. If it fails here we are freeing up the inode | |
2526 | * anyway and we'd rather not waste our time writing out stuff that is | |
2527 | * going to be truncated anyway. | |
2528 | */ | |
2529 | if (!igrab(inode)) | |
2530 | return 0; | |
2531 | ||
9f50fd2e | 2532 | folio_batch_init(&fbatch); |
d1310b2e CM |
2533 | if (wbc->range_cyclic) { |
2534 | index = mapping->writeback_index; /* Start from prev offset */ | |
2535 | end = -1; | |
556755a8 JB |
2536 | /* |
2537 | * Start from the beginning does not need to cycle over the | |
2538 | * range, mark it as scanned. | |
2539 | */ | |
2540 | scanned = (index == 0); | |
d1310b2e | 2541 | } else { |
09cbfeaf KS |
2542 | index = wbc->range_start >> PAGE_SHIFT; |
2543 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
2544 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2545 | range_whole = 1; | |
d1310b2e CM |
2546 | scanned = 1; |
2547 | } | |
3cd24c69 EL |
2548 | |
2549 | /* | |
2550 | * We do the tagged writepage as long as the snapshot flush bit is set | |
2551 | * and we are the first one who do the filemap_flush() on this inode. | |
2552 | * | |
2553 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
2554 | * not race in and drop the bit. | |
2555 | */ | |
2556 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
2557 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
2558 | &BTRFS_I(inode)->runtime_flags)) | |
2559 | wbc->tagged_writepages = 1; | |
2560 | ||
2561 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
2562 | tag = PAGECACHE_TAG_TOWRITE; |
2563 | else | |
2564 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 2565 | retry: |
3cd24c69 | 2566 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 2567 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 2568 | done_index = index; |
f85d7d6c | 2569 | while (!done && !nr_to_write_done && (index <= end) && |
9f50fd2e VMO |
2570 | (nr_folios = filemap_get_folios_tag(mapping, &index, |
2571 | end, tag, &fbatch))) { | |
d1310b2e CM |
2572 | unsigned i; |
2573 | ||
9f50fd2e VMO |
2574 | for (i = 0; i < nr_folios; i++) { |
2575 | struct folio *folio = fbatch.folios[i]; | |
d1310b2e | 2576 | |
9f50fd2e | 2577 | done_index = folio->index + folio_nr_pages(folio); |
d1310b2e | 2578 | /* |
b93b0163 MW |
2579 | * At this point we hold neither the i_pages lock nor |
2580 | * the page lock: the page may be truncated or | |
2581 | * invalidated (changing page->mapping to NULL), | |
2582 | * or even swizzled back from swapper_space to | |
2583 | * tmpfs file mapping | |
d1310b2e | 2584 | */ |
9f50fd2e | 2585 | if (!folio_trylock(folio)) { |
ee5f017d | 2586 | submit_write_bio(bio_ctrl, 0); |
9f50fd2e | 2587 | folio_lock(folio); |
01d658f2 | 2588 | } |
d1310b2e | 2589 | |
9f50fd2e VMO |
2590 | if (unlikely(folio->mapping != mapping)) { |
2591 | folio_unlock(folio); | |
d1310b2e CM |
2592 | continue; |
2593 | } | |
2594 | ||
d2c3f4f6 | 2595 | if (wbc->sync_mode != WB_SYNC_NONE) { |
9f50fd2e | 2596 | if (folio_test_writeback(folio)) |
ee5f017d | 2597 | submit_write_bio(bio_ctrl, 0); |
9f50fd2e | 2598 | folio_wait_writeback(folio); |
d2c3f4f6 | 2599 | } |
d1310b2e | 2600 | |
9f50fd2e VMO |
2601 | if (folio_test_writeback(folio) || |
2602 | !folio_clear_dirty_for_io(folio)) { | |
2603 | folio_unlock(folio); | |
d1310b2e CM |
2604 | continue; |
2605 | } | |
2606 | ||
72b505dc | 2607 | ret = __extent_writepage(&folio->page, bio_ctrl); |
a9132667 | 2608 | if (ret < 0) { |
a9132667 LB |
2609 | done = 1; |
2610 | break; | |
2611 | } | |
f85d7d6c CM |
2612 | |
2613 | /* | |
2614 | * the filesystem may choose to bump up nr_to_write. | |
2615 | * We have to make sure to honor the new nr_to_write | |
2616 | * at any time | |
2617 | */ | |
2618 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e | 2619 | } |
9f50fd2e | 2620 | folio_batch_release(&fbatch); |
d1310b2e CM |
2621 | cond_resched(); |
2622 | } | |
894b36e3 | 2623 | if (!scanned && !done) { |
d1310b2e CM |
2624 | /* |
2625 | * We hit the last page and there is more work to be done: wrap | |
2626 | * back to the start of the file | |
2627 | */ | |
2628 | scanned = 1; | |
2629 | index = 0; | |
42ffb0bf JB |
2630 | |
2631 | /* | |
2632 | * If we're looping we could run into a page that is locked by a | |
2633 | * writer and that writer could be waiting on writeback for a | |
2634 | * page in our current bio, and thus deadlock, so flush the | |
2635 | * write bio here. | |
2636 | */ | |
ee5f017d | 2637 | submit_write_bio(bio_ctrl, 0); |
c9583ada | 2638 | goto retry; |
d1310b2e | 2639 | } |
a9132667 LB |
2640 | |
2641 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
2642 | mapping->writeback_index = done_index; | |
2643 | ||
e55cf7ca | 2644 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
894b36e3 | 2645 | return ret; |
d1310b2e | 2646 | } |
d1310b2e | 2647 | |
2bd0fc93 QW |
2648 | /* |
2649 | * Submit the pages in the range to bio for call sites which delalloc range has | |
2650 | * already been ran (aka, ordered extent inserted) and all pages are still | |
2651 | * locked. | |
2652 | */ | |
2653 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end) | |
771ed689 | 2654 | { |
2bd0fc93 QW |
2655 | bool found_error = false; |
2656 | int first_error = 0; | |
771ed689 CM |
2657 | int ret = 0; |
2658 | struct address_space *mapping = inode->i_mapping; | |
2659 | struct page *page; | |
2bd0fc93 | 2660 | u64 cur = start; |
66448b9d QW |
2661 | unsigned long nr_pages; |
2662 | const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize; | |
771ed689 | 2663 | struct writeback_control wbc_writepages = { |
2bd0fc93 | 2664 | .sync_mode = WB_SYNC_ALL, |
771ed689 CM |
2665 | .range_start = start, |
2666 | .range_end = end + 1, | |
ec39f769 CM |
2667 | /* We're called from an async helper function */ |
2668 | .punt_to_cgroup = 1, | |
2669 | .no_cgroup_owner = 1, | |
771ed689 | 2670 | }; |
c000bc04 | 2671 | struct btrfs_bio_ctrl bio_ctrl = { |
72b505dc | 2672 | .wbc = &wbc_writepages, |
c000bc04 CH |
2673 | .opf = REQ_OP_WRITE | wbc_to_write_flags(&wbc_writepages), |
2674 | .extent_locked = 1, | |
c000bc04 | 2675 | }; |
771ed689 | 2676 | |
66448b9d QW |
2677 | ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize)); |
2678 | nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >> | |
2679 | PAGE_SHIFT; | |
2680 | wbc_writepages.nr_to_write = nr_pages * 2; | |
2681 | ||
dbb70bec | 2682 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
2bd0fc93 | 2683 | while (cur <= end) { |
66448b9d QW |
2684 | u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end); |
2685 | ||
2bd0fc93 QW |
2686 | page = find_get_page(mapping, cur >> PAGE_SHIFT); |
2687 | /* | |
2688 | * All pages in the range are locked since | |
2689 | * btrfs_run_delalloc_range(), thus there is no way to clear | |
2690 | * the page dirty flag. | |
2691 | */ | |
66448b9d | 2692 | ASSERT(PageLocked(page)); |
2bd0fc93 QW |
2693 | ASSERT(PageDirty(page)); |
2694 | clear_page_dirty_for_io(page); | |
72b505dc | 2695 | ret = __extent_writepage(page, &bio_ctrl); |
2bd0fc93 QW |
2696 | ASSERT(ret <= 0); |
2697 | if (ret < 0) { | |
2698 | found_error = true; | |
2699 | first_error = ret; | |
771ed689 | 2700 | } |
09cbfeaf | 2701 | put_page(page); |
66448b9d | 2702 | cur = cur_end + 1; |
771ed689 CM |
2703 | } |
2704 | ||
ee5f017d | 2705 | submit_write_bio(&bio_ctrl, found_error ? ret : 0); |
dbb70bec CM |
2706 | |
2707 | wbc_detach_inode(&wbc_writepages); | |
2bd0fc93 QW |
2708 | if (found_error) |
2709 | return first_error; | |
771ed689 CM |
2710 | return ret; |
2711 | } | |
d1310b2e | 2712 | |
8ae225a8 | 2713 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
2714 | struct writeback_control *wbc) |
2715 | { | |
35156d85 | 2716 | struct inode *inode = mapping->host; |
d1310b2e | 2717 | int ret = 0; |
ee5f017d | 2718 | struct btrfs_bio_ctrl bio_ctrl = { |
72b505dc | 2719 | .wbc = wbc, |
c000bc04 | 2720 | .opf = REQ_OP_WRITE | wbc_to_write_flags(wbc), |
771ed689 | 2721 | .extent_locked = 0, |
d1310b2e CM |
2722 | }; |
2723 | ||
35156d85 JT |
2724 | /* |
2725 | * Allow only a single thread to do the reloc work in zoned mode to | |
2726 | * protect the write pointer updates. | |
2727 | */ | |
869f4cdc | 2728 | btrfs_zoned_data_reloc_lock(BTRFS_I(inode)); |
72b505dc | 2729 | ret = extent_write_cache_pages(mapping, &bio_ctrl); |
ee5f017d | 2730 | submit_write_bio(&bio_ctrl, ret); |
19ab78ca | 2731 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
d1310b2e CM |
2732 | return ret; |
2733 | } | |
d1310b2e | 2734 | |
ba206a02 | 2735 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 2736 | { |
c000bc04 | 2737 | struct btrfs_bio_ctrl bio_ctrl = { .opf = REQ_OP_READ | REQ_RAHEAD }; |
67c9684f | 2738 | struct page *pagepool[16]; |
125bac01 | 2739 | struct extent_map *em_cached = NULL; |
808f80b4 | 2740 | u64 prev_em_start = (u64)-1; |
ba206a02 | 2741 | int nr; |
d1310b2e | 2742 | |
ba206a02 | 2743 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
2744 | u64 contig_start = readahead_pos(rac); |
2745 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 2746 | |
ba206a02 | 2747 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 2748 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 2749 | } |
67c9684f | 2750 | |
125bac01 MX |
2751 | if (em_cached) |
2752 | free_extent_map(em_cached); | |
722c82ac | 2753 | submit_one_bio(&bio_ctrl); |
d1310b2e | 2754 | } |
d1310b2e CM |
2755 | |
2756 | /* | |
895586eb MWO |
2757 | * basic invalidate_folio code, this waits on any locked or writeback |
2758 | * ranges corresponding to the folio, and then deletes any extent state | |
d1310b2e CM |
2759 | * records from the tree |
2760 | */ | |
895586eb MWO |
2761 | int extent_invalidate_folio(struct extent_io_tree *tree, |
2762 | struct folio *folio, size_t offset) | |
d1310b2e | 2763 | { |
2ac55d41 | 2764 | struct extent_state *cached_state = NULL; |
895586eb MWO |
2765 | u64 start = folio_pos(folio); |
2766 | u64 end = start + folio_size(folio) - 1; | |
2767 | size_t blocksize = folio->mapping->host->i_sb->s_blocksize; | |
d1310b2e | 2768 | |
829ddec9 QW |
2769 | /* This function is only called for the btree inode */ |
2770 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
2771 | ||
fda2832f | 2772 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
2773 | if (start > end) |
2774 | return 0; | |
2775 | ||
570eb97b | 2776 | lock_extent(tree, start, end, &cached_state); |
895586eb | 2777 | folio_wait_writeback(folio); |
829ddec9 QW |
2778 | |
2779 | /* | |
2780 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
2781 | * so here we only need to unlock the extent range to free any | |
2782 | * existing extent state. | |
2783 | */ | |
570eb97b | 2784 | unlock_extent(tree, start, end, &cached_state); |
d1310b2e CM |
2785 | return 0; |
2786 | } | |
d1310b2e | 2787 | |
7b13b7b1 | 2788 | /* |
f913cff3 | 2789 | * a helper for release_folio, this tests for areas of the page that |
7b13b7b1 CM |
2790 | * are locked or under IO and drops the related state bits if it is safe |
2791 | * to drop the page. | |
2792 | */ | |
29c68b2d | 2793 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 2794 | struct page *page, gfp_t mask) |
7b13b7b1 | 2795 | { |
4eee4fa4 | 2796 | u64 start = page_offset(page); |
09cbfeaf | 2797 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
2798 | int ret = 1; |
2799 | ||
8882679e | 2800 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 2801 | ret = 0; |
8882679e | 2802 | } else { |
b71fb16b JB |
2803 | u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM | |
2804 | EXTENT_DELALLOC_NEW | EXTENT_CTLBITS); | |
2805 | ||
11ef160f | 2806 | /* |
2766ff61 FM |
2807 | * At this point we can safely clear everything except the |
2808 | * locked bit, the nodatasum bit and the delalloc new bit. | |
2809 | * The delalloc new bit will be cleared by ordered extent | |
2810 | * completion. | |
11ef160f | 2811 | */ |
bd015294 | 2812 | ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, |
b71fb16b | 2813 | mask, NULL); |
e3f24cc5 CM |
2814 | |
2815 | /* if clear_extent_bit failed for enomem reasons, | |
2816 | * we can't allow the release to continue. | |
2817 | */ | |
2818 | if (ret < 0) | |
2819 | ret = 0; | |
2820 | else | |
2821 | ret = 1; | |
7b13b7b1 CM |
2822 | } |
2823 | return ret; | |
2824 | } | |
7b13b7b1 | 2825 | |
d1310b2e | 2826 | /* |
f913cff3 | 2827 | * a helper for release_folio. As long as there are no locked extents |
d1310b2e CM |
2828 | * in the range corresponding to the page, both state records and extent |
2829 | * map records are removed | |
2830 | */ | |
477a30ba | 2831 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
2832 | { |
2833 | struct extent_map *em; | |
4eee4fa4 | 2834 | u64 start = page_offset(page); |
09cbfeaf | 2835 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
2836 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
2837 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
2838 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 2839 | |
d0164adc | 2840 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 2841 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 2842 | u64 len; |
70dec807 | 2843 | while (start <= end) { |
fbc2bd7e FM |
2844 | struct btrfs_fs_info *fs_info; |
2845 | u64 cur_gen; | |
2846 | ||
39b5637f | 2847 | len = end - start + 1; |
890871be | 2848 | write_lock(&map->lock); |
39b5637f | 2849 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 2850 | if (!em) { |
890871be | 2851 | write_unlock(&map->lock); |
70dec807 CM |
2852 | break; |
2853 | } | |
7f3c74fb CM |
2854 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2855 | em->start != start) { | |
890871be | 2856 | write_unlock(&map->lock); |
70dec807 CM |
2857 | free_extent_map(em); |
2858 | break; | |
2859 | } | |
3d6448e6 FM |
2860 | if (test_range_bit(tree, em->start, |
2861 | extent_map_end(em) - 1, | |
2862 | EXTENT_LOCKED, 0, NULL)) | |
2863 | goto next; | |
2864 | /* | |
2865 | * If it's not in the list of modified extents, used | |
2866 | * by a fast fsync, we can remove it. If it's being | |
2867 | * logged we can safely remove it since fsync took an | |
2868 | * extra reference on the em. | |
2869 | */ | |
2870 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
2871 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
2872 | goto remove_em; | |
2873 | /* | |
2874 | * If it's in the list of modified extents, remove it | |
2875 | * only if its generation is older then the current one, | |
2876 | * in which case we don't need it for a fast fsync. | |
2877 | * Otherwise don't remove it, we could be racing with an | |
2878 | * ongoing fast fsync that could miss the new extent. | |
2879 | */ | |
2880 | fs_info = btrfs_inode->root->fs_info; | |
2881 | spin_lock(&fs_info->trans_lock); | |
2882 | cur_gen = fs_info->generation; | |
2883 | spin_unlock(&fs_info->trans_lock); | |
2884 | if (em->generation >= cur_gen) | |
2885 | goto next; | |
2886 | remove_em: | |
5e548b32 FM |
2887 | /* |
2888 | * We only remove extent maps that are not in the list of | |
2889 | * modified extents or that are in the list but with a | |
2890 | * generation lower then the current generation, so there | |
2891 | * is no need to set the full fsync flag on the inode (it | |
2892 | * hurts the fsync performance for workloads with a data | |
2893 | * size that exceeds or is close to the system's memory). | |
2894 | */ | |
fbc2bd7e FM |
2895 | remove_extent_mapping(map, em); |
2896 | /* once for the rb tree */ | |
2897 | free_extent_map(em); | |
3d6448e6 | 2898 | next: |
70dec807 | 2899 | start = extent_map_end(em); |
890871be | 2900 | write_unlock(&map->lock); |
70dec807 CM |
2901 | |
2902 | /* once for us */ | |
d1310b2e | 2903 | free_extent_map(em); |
9f47eb54 PM |
2904 | |
2905 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 2906 | } |
d1310b2e | 2907 | } |
29c68b2d | 2908 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 2909 | } |
d1310b2e | 2910 | |
4751832d QW |
2911 | /* |
2912 | * To cache previous fiemap extent | |
2913 | * | |
2914 | * Will be used for merging fiemap extent | |
2915 | */ | |
2916 | struct fiemap_cache { | |
2917 | u64 offset; | |
2918 | u64 phys; | |
2919 | u64 len; | |
2920 | u32 flags; | |
2921 | bool cached; | |
2922 | }; | |
2923 | ||
2924 | /* | |
2925 | * Helper to submit fiemap extent. | |
2926 | * | |
2927 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
2928 | * @len and @flags with cached one. | |
2929 | * And only when we fails to merge, cached one will be submitted as | |
2930 | * fiemap extent. | |
2931 | * | |
2932 | * Return value is the same as fiemap_fill_next_extent(). | |
2933 | */ | |
2934 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
2935 | struct fiemap_cache *cache, | |
2936 | u64 offset, u64 phys, u64 len, u32 flags) | |
2937 | { | |
2938 | int ret = 0; | |
2939 | ||
ac3c0d36 FM |
2940 | /* Set at the end of extent_fiemap(). */ |
2941 | ASSERT((flags & FIEMAP_EXTENT_LAST) == 0); | |
2942 | ||
4751832d QW |
2943 | if (!cache->cached) |
2944 | goto assign; | |
2945 | ||
2946 | /* | |
2947 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 2948 | * fiemap extent won't overlap with cached one. |
4751832d QW |
2949 | * Not recoverable. |
2950 | * | |
2951 | * NOTE: Physical address can overlap, due to compression | |
2952 | */ | |
2953 | if (cache->offset + cache->len > offset) { | |
2954 | WARN_ON(1); | |
2955 | return -EINVAL; | |
2956 | } | |
2957 | ||
2958 | /* | |
2959 | * Only merges fiemap extents if | |
2960 | * 1) Their logical addresses are continuous | |
2961 | * | |
2962 | * 2) Their physical addresses are continuous | |
2963 | * So truly compressed (physical size smaller than logical size) | |
2964 | * extents won't get merged with each other | |
2965 | * | |
ac3c0d36 | 2966 | * 3) Share same flags |
4751832d QW |
2967 | */ |
2968 | if (cache->offset + cache->len == offset && | |
2969 | cache->phys + cache->len == phys && | |
ac3c0d36 | 2970 | cache->flags == flags) { |
4751832d | 2971 | cache->len += len; |
ac3c0d36 | 2972 | return 0; |
4751832d QW |
2973 | } |
2974 | ||
2975 | /* Not mergeable, need to submit cached one */ | |
2976 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
2977 | cache->len, cache->flags); | |
2978 | cache->cached = false; | |
2979 | if (ret) | |
2980 | return ret; | |
2981 | assign: | |
2982 | cache->cached = true; | |
2983 | cache->offset = offset; | |
2984 | cache->phys = phys; | |
2985 | cache->len = len; | |
2986 | cache->flags = flags; | |
ac3c0d36 FM |
2987 | |
2988 | return 0; | |
4751832d QW |
2989 | } |
2990 | ||
2991 | /* | |
848c23b7 | 2992 | * Emit last fiemap cache |
4751832d | 2993 | * |
848c23b7 QW |
2994 | * The last fiemap cache may still be cached in the following case: |
2995 | * 0 4k 8k | |
2996 | * |<- Fiemap range ->| | |
2997 | * |<------------ First extent ----------->| | |
2998 | * | |
2999 | * In this case, the first extent range will be cached but not emitted. | |
3000 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 3001 | */ |
5c5aff98 | 3002 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 3003 | struct fiemap_cache *cache) |
4751832d QW |
3004 | { |
3005 | int ret; | |
3006 | ||
3007 | if (!cache->cached) | |
3008 | return 0; | |
3009 | ||
4751832d QW |
3010 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
3011 | cache->len, cache->flags); | |
3012 | cache->cached = false; | |
3013 | if (ret > 0) | |
3014 | ret = 0; | |
3015 | return ret; | |
3016 | } | |
3017 | ||
ac3c0d36 | 3018 | static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path) |
1506fcc8 | 3019 | { |
ac3c0d36 FM |
3020 | struct extent_buffer *clone; |
3021 | struct btrfs_key key; | |
3022 | int slot; | |
3023 | int ret; | |
3024 | ||
3025 | path->slots[0]++; | |
3026 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) | |
3027 | return 0; | |
3028 | ||
3029 | ret = btrfs_next_leaf(inode->root, path); | |
3030 | if (ret != 0) | |
3031 | return ret; | |
3032 | ||
3033 | /* | |
3034 | * Don't bother with cloning if there are no more file extent items for | |
3035 | * our inode. | |
3036 | */ | |
3037 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3038 | if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY) | |
3039 | return 1; | |
3040 | ||
3041 | /* See the comment at fiemap_search_slot() about why we clone. */ | |
3042 | clone = btrfs_clone_extent_buffer(path->nodes[0]); | |
3043 | if (!clone) | |
3044 | return -ENOMEM; | |
3045 | ||
3046 | slot = path->slots[0]; | |
3047 | btrfs_release_path(path); | |
3048 | path->nodes[0] = clone; | |
3049 | path->slots[0] = slot; | |
3050 | ||
3051 | return 0; | |
3052 | } | |
3053 | ||
3054 | /* | |
3055 | * Search for the first file extent item that starts at a given file offset or | |
3056 | * the one that starts immediately before that offset. | |
3057 | * Returns: 0 on success, < 0 on error, 1 if not found. | |
3058 | */ | |
3059 | static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path, | |
3060 | u64 file_offset) | |
3061 | { | |
3062 | const u64 ino = btrfs_ino(inode); | |
facee0a0 | 3063 | struct btrfs_root *root = inode->root; |
ac3c0d36 FM |
3064 | struct extent_buffer *clone; |
3065 | struct btrfs_key key; | |
3066 | int slot; | |
3067 | int ret; | |
1506fcc8 | 3068 | |
ac3c0d36 FM |
3069 | key.objectid = ino; |
3070 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3071 | key.offset = file_offset; | |
3072 | ||
3073 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3074 | if (ret < 0) | |
3075 | return ret; | |
3076 | ||
3077 | if (ret > 0 && path->slots[0] > 0) { | |
3078 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3079 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3080 | path->slots[0]--; | |
3081 | } | |
3082 | ||
3083 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
3084 | ret = btrfs_next_leaf(root, path); | |
3085 | if (ret != 0) | |
3086 | return ret; | |
3087 | ||
3088 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3089 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
3090 | return 1; | |
5911c8fe DS |
3091 | } |
3092 | ||
15c7745c | 3093 | /* |
ac3c0d36 FM |
3094 | * We clone the leaf and use it during fiemap. This is because while |
3095 | * using the leaf we do expensive things like checking if an extent is | |
3096 | * shared, which can take a long time. In order to prevent blocking | |
3097 | * other tasks for too long, we use a clone of the leaf. We have locked | |
3098 | * the file range in the inode's io tree, so we know none of our file | |
3099 | * extent items can change. This way we avoid blocking other tasks that | |
3100 | * want to insert items for other inodes in the same leaf or b+tree | |
3101 | * rebalance operations (triggered for example when someone is trying | |
3102 | * to push items into this leaf when trying to insert an item in a | |
3103 | * neighbour leaf). | |
3104 | * We also need the private clone because holding a read lock on an | |
3105 | * extent buffer of the subvolume's b+tree will make lockdep unhappy | |
3106 | * when we call fiemap_fill_next_extent(), because that may cause a page | |
3107 | * fault when filling the user space buffer with fiemap data. | |
15c7745c | 3108 | */ |
ac3c0d36 FM |
3109 | clone = btrfs_clone_extent_buffer(path->nodes[0]); |
3110 | if (!clone) | |
3111 | return -ENOMEM; | |
3112 | ||
3113 | slot = path->slots[0]; | |
3114 | btrfs_release_path(path); | |
3115 | path->nodes[0] = clone; | |
3116 | path->slots[0] = slot; | |
3117 | ||
3118 | return 0; | |
3119 | } | |
3120 | ||
3121 | /* | |
3122 | * Process a range which is a hole or a prealloc extent in the inode's subvolume | |
3123 | * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc | |
3124 | * extent. The end offset (@end) is inclusive. | |
3125 | */ | |
3126 | static int fiemap_process_hole(struct btrfs_inode *inode, | |
3127 | struct fiemap_extent_info *fieinfo, | |
3128 | struct fiemap_cache *cache, | |
b3e744fe | 3129 | struct extent_state **delalloc_cached_state, |
61dbb952 | 3130 | struct btrfs_backref_share_check_ctx *backref_ctx, |
ac3c0d36 FM |
3131 | u64 disk_bytenr, u64 extent_offset, |
3132 | u64 extent_gen, | |
ac3c0d36 FM |
3133 | u64 start, u64 end) |
3134 | { | |
3135 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
ac3c0d36 FM |
3136 | u64 cur_offset = start; |
3137 | u64 last_delalloc_end = 0; | |
3138 | u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN; | |
3139 | bool checked_extent_shared = false; | |
3140 | int ret; | |
4d479cf0 | 3141 | |
ec29ed5b | 3142 | /* |
ac3c0d36 FM |
3143 | * There can be no delalloc past i_size, so don't waste time looking for |
3144 | * it beyond i_size. | |
ec29ed5b | 3145 | */ |
ac3c0d36 FM |
3146 | while (cur_offset < end && cur_offset < i_size) { |
3147 | u64 delalloc_start; | |
3148 | u64 delalloc_end; | |
3149 | u64 prealloc_start; | |
3150 | u64 prealloc_len = 0; | |
3151 | bool delalloc; | |
3152 | ||
3153 | delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end, | |
b3e744fe | 3154 | delalloc_cached_state, |
ac3c0d36 FM |
3155 | &delalloc_start, |
3156 | &delalloc_end); | |
3157 | if (!delalloc) | |
3158 | break; | |
2d324f59 | 3159 | |
ec29ed5b | 3160 | /* |
ac3c0d36 FM |
3161 | * If this is a prealloc extent we have to report every section |
3162 | * of it that has no delalloc. | |
ec29ed5b | 3163 | */ |
ac3c0d36 FM |
3164 | if (disk_bytenr != 0) { |
3165 | if (last_delalloc_end == 0) { | |
3166 | prealloc_start = start; | |
3167 | prealloc_len = delalloc_start - start; | |
3168 | } else { | |
3169 | prealloc_start = last_delalloc_end + 1; | |
3170 | prealloc_len = delalloc_start - prealloc_start; | |
3171 | } | |
3172 | } | |
3173 | ||
3174 | if (prealloc_len > 0) { | |
3175 | if (!checked_extent_shared && fieinfo->fi_extents_max) { | |
ceb707da | 3176 | ret = btrfs_is_data_extent_shared(inode, |
84a7949d FM |
3177 | disk_bytenr, |
3178 | extent_gen, | |
3179 | backref_ctx); | |
ac3c0d36 FM |
3180 | if (ret < 0) |
3181 | return ret; | |
3182 | else if (ret > 0) | |
3183 | prealloc_flags |= FIEMAP_EXTENT_SHARED; | |
3184 | ||
3185 | checked_extent_shared = true; | |
3186 | } | |
3187 | ret = emit_fiemap_extent(fieinfo, cache, prealloc_start, | |
3188 | disk_bytenr + extent_offset, | |
3189 | prealloc_len, prealloc_flags); | |
3190 | if (ret) | |
3191 | return ret; | |
3192 | extent_offset += prealloc_len; | |
3193 | } | |
3194 | ||
3195 | ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0, | |
3196 | delalloc_end + 1 - delalloc_start, | |
3197 | FIEMAP_EXTENT_DELALLOC | | |
3198 | FIEMAP_EXTENT_UNKNOWN); | |
3199 | if (ret) | |
3200 | return ret; | |
3201 | ||
3202 | last_delalloc_end = delalloc_end; | |
3203 | cur_offset = delalloc_end + 1; | |
3204 | extent_offset += cur_offset - delalloc_start; | |
3205 | cond_resched(); | |
3206 | } | |
3207 | ||
3208 | /* | |
3209 | * Either we found no delalloc for the whole prealloc extent or we have | |
3210 | * a prealloc extent that spans i_size or starts at or after i_size. | |
3211 | */ | |
3212 | if (disk_bytenr != 0 && last_delalloc_end < end) { | |
3213 | u64 prealloc_start; | |
3214 | u64 prealloc_len; | |
3215 | ||
3216 | if (last_delalloc_end == 0) { | |
3217 | prealloc_start = start; | |
3218 | prealloc_len = end + 1 - start; | |
3219 | } else { | |
3220 | prealloc_start = last_delalloc_end + 1; | |
3221 | prealloc_len = end + 1 - prealloc_start; | |
3222 | } | |
3223 | ||
3224 | if (!checked_extent_shared && fieinfo->fi_extents_max) { | |
ceb707da FM |
3225 | ret = btrfs_is_data_extent_shared(inode, |
3226 | disk_bytenr, | |
84a7949d | 3227 | extent_gen, |
61dbb952 | 3228 | backref_ctx); |
ac3c0d36 FM |
3229 | if (ret < 0) |
3230 | return ret; | |
3231 | else if (ret > 0) | |
3232 | prealloc_flags |= FIEMAP_EXTENT_SHARED; | |
3233 | } | |
3234 | ret = emit_fiemap_extent(fieinfo, cache, prealloc_start, | |
3235 | disk_bytenr + extent_offset, | |
3236 | prealloc_len, prealloc_flags); | |
3237 | if (ret) | |
3238 | return ret; | |
3239 | } | |
3240 | ||
3241 | return 0; | |
3242 | } | |
3243 | ||
3244 | static int fiemap_find_last_extent_offset(struct btrfs_inode *inode, | |
3245 | struct btrfs_path *path, | |
3246 | u64 *last_extent_end_ret) | |
3247 | { | |
3248 | const u64 ino = btrfs_ino(inode); | |
3249 | struct btrfs_root *root = inode->root; | |
3250 | struct extent_buffer *leaf; | |
3251 | struct btrfs_file_extent_item *ei; | |
3252 | struct btrfs_key key; | |
3253 | u64 disk_bytenr; | |
3254 | int ret; | |
3255 | ||
3256 | /* | |
3257 | * Lookup the last file extent. We're not using i_size here because | |
3258 | * there might be preallocation past i_size. | |
3259 | */ | |
3260 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0); | |
3261 | /* There can't be a file extent item at offset (u64)-1 */ | |
3262 | ASSERT(ret != 0); | |
3263 | if (ret < 0) | |
3264 | return ret; | |
3265 | ||
3266 | /* | |
3267 | * For a non-existing key, btrfs_search_slot() always leaves us at a | |
3268 | * slot > 0, except if the btree is empty, which is impossible because | |
3269 | * at least it has the inode item for this inode and all the items for | |
3270 | * the root inode 256. | |
3271 | */ | |
3272 | ASSERT(path->slots[0] > 0); | |
3273 | path->slots[0]--; | |
3274 | leaf = path->nodes[0]; | |
3275 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3276 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
3277 | /* No file extent items in the subvolume tree. */ | |
3278 | *last_extent_end_ret = 0; | |
3279 | return 0; | |
975f84fe | 3280 | } |
975f84fe | 3281 | |
ec29ed5b | 3282 | /* |
ac3c0d36 FM |
3283 | * For an inline extent, the disk_bytenr is where inline data starts at, |
3284 | * so first check if we have an inline extent item before checking if we | |
3285 | * have an implicit hole (disk_bytenr == 0). | |
ec29ed5b | 3286 | */ |
ac3c0d36 FM |
3287 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); |
3288 | if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) { | |
3289 | *last_extent_end_ret = btrfs_file_extent_end(path); | |
3290 | return 0; | |
ec29ed5b CM |
3291 | } |
3292 | ||
ac3c0d36 FM |
3293 | /* |
3294 | * Find the last file extent item that is not a hole (when NO_HOLES is | |
3295 | * not enabled). This should take at most 2 iterations in the worst | |
3296 | * case: we have one hole file extent item at slot 0 of a leaf and | |
3297 | * another hole file extent item as the last item in the previous leaf. | |
3298 | * This is because we merge file extent items that represent holes. | |
3299 | */ | |
3300 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3301 | while (disk_bytenr == 0) { | |
3302 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
3303 | if (ret < 0) { | |
3304 | return ret; | |
3305 | } else if (ret > 0) { | |
3306 | /* No file extent items that are not holes. */ | |
3307 | *last_extent_end_ret = 0; | |
3308 | return 0; | |
3309 | } | |
3310 | leaf = path->nodes[0]; | |
3311 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
3312 | struct btrfs_file_extent_item); | |
3313 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3314 | } | |
ec29ed5b | 3315 | |
ac3c0d36 FM |
3316 | *last_extent_end_ret = btrfs_file_extent_end(path); |
3317 | return 0; | |
3318 | } | |
3319 | ||
3320 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, | |
3321 | u64 start, u64 len) | |
3322 | { | |
3323 | const u64 ino = btrfs_ino(inode); | |
3324 | struct extent_state *cached_state = NULL; | |
b3e744fe | 3325 | struct extent_state *delalloc_cached_state = NULL; |
ac3c0d36 | 3326 | struct btrfs_path *path; |
ac3c0d36 | 3327 | struct fiemap_cache cache = { 0 }; |
61dbb952 | 3328 | struct btrfs_backref_share_check_ctx *backref_ctx; |
ac3c0d36 FM |
3329 | u64 last_extent_end; |
3330 | u64 prev_extent_end; | |
3331 | u64 lockstart; | |
3332 | u64 lockend; | |
3333 | bool stopped = false; | |
3334 | int ret; | |
3335 | ||
84a7949d | 3336 | backref_ctx = btrfs_alloc_backref_share_check_ctx(); |
ac3c0d36 | 3337 | path = btrfs_alloc_path(); |
84a7949d | 3338 | if (!backref_ctx || !path) { |
ac3c0d36 | 3339 | ret = -ENOMEM; |
1506fcc8 YS |
3340 | goto out; |
3341 | } | |
975f84fe | 3342 | |
ceb707da FM |
3343 | lockstart = round_down(start, inode->root->fs_info->sectorsize); |
3344 | lockend = round_up(start + len, inode->root->fs_info->sectorsize); | |
ac3c0d36 | 3345 | prev_extent_end = lockstart; |
ea8efc74 | 3346 | |
519b7e13 | 3347 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
570eb97b | 3348 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
ea8efc74 | 3349 | |
ac3c0d36 FM |
3350 | ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end); |
3351 | if (ret < 0) | |
3352 | goto out_unlock; | |
3353 | btrfs_release_path(path); | |
1506fcc8 | 3354 | |
ac3c0d36 FM |
3355 | path->reada = READA_FORWARD; |
3356 | ret = fiemap_search_slot(inode, path, lockstart); | |
3357 | if (ret < 0) { | |
3358 | goto out_unlock; | |
3359 | } else if (ret > 0) { | |
ea8efc74 | 3360 | /* |
ac3c0d36 FM |
3361 | * No file extent item found, but we may have delalloc between |
3362 | * the current offset and i_size. So check for that. | |
ea8efc74 | 3363 | */ |
ac3c0d36 FM |
3364 | ret = 0; |
3365 | goto check_eof_delalloc; | |
3366 | } | |
3367 | ||
3368 | while (prev_extent_end < lockend) { | |
3369 | struct extent_buffer *leaf = path->nodes[0]; | |
3370 | struct btrfs_file_extent_item *ei; | |
3371 | struct btrfs_key key; | |
3372 | u64 extent_end; | |
3373 | u64 extent_len; | |
3374 | u64 extent_offset = 0; | |
3375 | u64 extent_gen; | |
3376 | u64 disk_bytenr = 0; | |
3377 | u64 flags = 0; | |
3378 | int extent_type; | |
3379 | u8 compression; | |
3380 | ||
3381 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3382 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
3383 | break; | |
3384 | ||
3385 | extent_end = btrfs_file_extent_end(path); | |
1506fcc8 | 3386 | |
ea8efc74 | 3387 | /* |
ac3c0d36 FM |
3388 | * The first iteration can leave us at an extent item that ends |
3389 | * before our range's start. Move to the next item. | |
ea8efc74 | 3390 | */ |
ac3c0d36 FM |
3391 | if (extent_end <= lockstart) |
3392 | goto next_item; | |
fe09e16c | 3393 | |
877c1476 FM |
3394 | backref_ctx->curr_leaf_bytenr = leaf->start; |
3395 | ||
ac3c0d36 FM |
3396 | /* We have in implicit hole (NO_HOLES feature enabled). */ |
3397 | if (prev_extent_end < key.offset) { | |
3398 | const u64 range_end = min(key.offset, lockend) - 1; | |
b8f164e3 | 3399 | |
ac3c0d36 | 3400 | ret = fiemap_process_hole(inode, fieinfo, &cache, |
b3e744fe | 3401 | &delalloc_cached_state, |
61dbb952 | 3402 | backref_ctx, 0, 0, 0, |
ac3c0d36 FM |
3403 | prev_extent_end, range_end); |
3404 | if (ret < 0) { | |
3405 | goto out_unlock; | |
3406 | } else if (ret > 0) { | |
3407 | /* fiemap_fill_next_extent() told us to stop. */ | |
3408 | stopped = true; | |
3409 | break; | |
3410 | } | |
1506fcc8 | 3411 | |
ac3c0d36 FM |
3412 | /* We've reached the end of the fiemap range, stop. */ |
3413 | if (key.offset >= lockend) { | |
3414 | stopped = true; | |
3415 | break; | |
3416 | } | |
1506fcc8 YS |
3417 | } |
3418 | ||
ac3c0d36 FM |
3419 | extent_len = extent_end - key.offset; |
3420 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
3421 | struct btrfs_file_extent_item); | |
3422 | compression = btrfs_file_extent_compression(leaf, ei); | |
3423 | extent_type = btrfs_file_extent_type(leaf, ei); | |
3424 | extent_gen = btrfs_file_extent_generation(leaf, ei); | |
3425 | ||
3426 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
3427 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3428 | if (compression == BTRFS_COMPRESS_NONE) | |
3429 | extent_offset = btrfs_file_extent_offset(leaf, ei); | |
ec29ed5b | 3430 | } |
ac3c0d36 FM |
3431 | |
3432 | if (compression != BTRFS_COMPRESS_NONE) | |
3433 | flags |= FIEMAP_EXTENT_ENCODED; | |
3434 | ||
3435 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
3436 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
3437 | flags |= FIEMAP_EXTENT_NOT_ALIGNED; | |
3438 | ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0, | |
3439 | extent_len, flags); | |
3440 | } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
3441 | ret = fiemap_process_hole(inode, fieinfo, &cache, | |
b3e744fe | 3442 | &delalloc_cached_state, |
61dbb952 | 3443 | backref_ctx, |
ac3c0d36 | 3444 | disk_bytenr, extent_offset, |
84a7949d FM |
3445 | extent_gen, key.offset, |
3446 | extent_end - 1); | |
ac3c0d36 FM |
3447 | } else if (disk_bytenr == 0) { |
3448 | /* We have an explicit hole. */ | |
3449 | ret = fiemap_process_hole(inode, fieinfo, &cache, | |
b3e744fe | 3450 | &delalloc_cached_state, |
61dbb952 | 3451 | backref_ctx, 0, 0, 0, |
ac3c0d36 FM |
3452 | key.offset, extent_end - 1); |
3453 | } else { | |
3454 | /* We have a regular extent. */ | |
3455 | if (fieinfo->fi_extents_max) { | |
ceb707da | 3456 | ret = btrfs_is_data_extent_shared(inode, |
ac3c0d36 FM |
3457 | disk_bytenr, |
3458 | extent_gen, | |
61dbb952 | 3459 | backref_ctx); |
ac3c0d36 FM |
3460 | if (ret < 0) |
3461 | goto out_unlock; | |
3462 | else if (ret > 0) | |
3463 | flags |= FIEMAP_EXTENT_SHARED; | |
3464 | } | |
3465 | ||
3466 | ret = emit_fiemap_extent(fieinfo, &cache, key.offset, | |
3467 | disk_bytenr + extent_offset, | |
3468 | extent_len, flags); | |
975f84fe | 3469 | } |
ac3c0d36 FM |
3470 | |
3471 | if (ret < 0) { | |
3472 | goto out_unlock; | |
3473 | } else if (ret > 0) { | |
3474 | /* fiemap_fill_next_extent() told us to stop. */ | |
3475 | stopped = true; | |
3476 | break; | |
26e726af | 3477 | } |
09fbc1c8 | 3478 | |
ac3c0d36 FM |
3479 | prev_extent_end = extent_end; |
3480 | next_item: | |
09fbc1c8 FM |
3481 | if (fatal_signal_pending(current)) { |
3482 | ret = -EINTR; | |
ac3c0d36 | 3483 | goto out_unlock; |
09fbc1c8 | 3484 | } |
ac3c0d36 FM |
3485 | |
3486 | ret = fiemap_next_leaf_item(inode, path); | |
3487 | if (ret < 0) { | |
3488 | goto out_unlock; | |
3489 | } else if (ret > 0) { | |
3490 | /* No more file extent items for this inode. */ | |
3491 | break; | |
3492 | } | |
3493 | cond_resched(); | |
1506fcc8 | 3494 | } |
5911c8fe | 3495 | |
ac3c0d36 FM |
3496 | check_eof_delalloc: |
3497 | /* | |
3498 | * Release (and free) the path before emitting any final entries to | |
3499 | * fiemap_fill_next_extent() to keep lockdep happy. This is because | |
3500 | * once we find no more file extent items exist, we may have a | |
3501 | * non-cloned leaf, and fiemap_fill_next_extent() can trigger page | |
3502 | * faults when copying data to the user space buffer. | |
3503 | */ | |
3504 | btrfs_free_path(path); | |
3505 | path = NULL; | |
3506 | ||
3507 | if (!stopped && prev_extent_end < lockend) { | |
b3e744fe FM |
3508 | ret = fiemap_process_hole(inode, fieinfo, &cache, |
3509 | &delalloc_cached_state, backref_ctx, | |
84a7949d | 3510 | 0, 0, 0, prev_extent_end, lockend - 1); |
ac3c0d36 FM |
3511 | if (ret < 0) |
3512 | goto out_unlock; | |
3513 | prev_extent_end = lockend; | |
3514 | } | |
3515 | ||
3516 | if (cache.cached && cache.offset + cache.len >= last_extent_end) { | |
3517 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
3518 | ||
3519 | if (prev_extent_end < i_size) { | |
3520 | u64 delalloc_start; | |
3521 | u64 delalloc_end; | |
3522 | bool delalloc; | |
3523 | ||
3524 | delalloc = btrfs_find_delalloc_in_range(inode, | |
3525 | prev_extent_end, | |
3526 | i_size - 1, | |
b3e744fe | 3527 | &delalloc_cached_state, |
ac3c0d36 FM |
3528 | &delalloc_start, |
3529 | &delalloc_end); | |
3530 | if (!delalloc) | |
3531 | cache.flags |= FIEMAP_EXTENT_LAST; | |
3532 | } else { | |
3533 | cache.flags |= FIEMAP_EXTENT_LAST; | |
3534 | } | |
3535 | } | |
3536 | ||
3537 | ret = emit_last_fiemap_cache(fieinfo, &cache); | |
3538 | ||
3539 | out_unlock: | |
570eb97b | 3540 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
519b7e13 | 3541 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
ac3c0d36 | 3542 | out: |
b3e744fe | 3543 | free_extent_state(delalloc_cached_state); |
84a7949d | 3544 | btrfs_free_backref_share_ctx(backref_ctx); |
e02d48ea | 3545 | btrfs_free_path(path); |
1506fcc8 YS |
3546 | return ret; |
3547 | } | |
3548 | ||
727011e0 CM |
3549 | static void __free_extent_buffer(struct extent_buffer *eb) |
3550 | { | |
727011e0 CM |
3551 | kmem_cache_free(extent_buffer_cache, eb); |
3552 | } | |
3553 | ||
2b48966a | 3554 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
3555 | { |
3556 | return (atomic_read(&eb->io_pages) || | |
3557 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
3558 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
3559 | } | |
3560 | ||
8ff8466d | 3561 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 3562 | { |
8ff8466d | 3563 | struct btrfs_subpage *subpage; |
db7f3436 | 3564 | |
8ff8466d | 3565 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 3566 | |
8ff8466d QW |
3567 | if (PagePrivate(page)) { |
3568 | subpage = (struct btrfs_subpage *)page->private; | |
3569 | if (atomic_read(&subpage->eb_refs)) | |
3570 | return true; | |
3d078efa QW |
3571 | /* |
3572 | * Even there is no eb refs here, we may still have | |
3573 | * end_page_read() call relying on page::private. | |
3574 | */ | |
3575 | if (atomic_read(&subpage->readers)) | |
3576 | return true; | |
8ff8466d QW |
3577 | } |
3578 | return false; | |
3579 | } | |
db7f3436 | 3580 | |
8ff8466d QW |
3581 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
3582 | { | |
3583 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
3584 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
3585 | ||
3586 | /* | |
3587 | * For mapped eb, we're going to change the page private, which should | |
3588 | * be done under the private_lock. | |
3589 | */ | |
3590 | if (mapped) | |
3591 | spin_lock(&page->mapping->private_lock); | |
3592 | ||
3593 | if (!PagePrivate(page)) { | |
5d2361db | 3594 | if (mapped) |
8ff8466d QW |
3595 | spin_unlock(&page->mapping->private_lock); |
3596 | return; | |
3597 | } | |
3598 | ||
fbca46eb | 3599 | if (fs_info->nodesize >= PAGE_SIZE) { |
5d2361db FL |
3600 | /* |
3601 | * We do this since we'll remove the pages after we've | |
3602 | * removed the eb from the radix tree, so we could race | |
3603 | * and have this page now attached to the new eb. So | |
3604 | * only clear page_private if it's still connected to | |
3605 | * this eb. | |
3606 | */ | |
3607 | if (PagePrivate(page) && | |
3608 | page->private == (unsigned long)eb) { | |
3609 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
3610 | BUG_ON(PageDirty(page)); | |
3611 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 3612 | /* |
5d2361db FL |
3613 | * We need to make sure we haven't be attached |
3614 | * to a new eb. | |
db7f3436 | 3615 | */ |
d1b89bc0 | 3616 | detach_page_private(page); |
db7f3436 | 3617 | } |
5d2361db FL |
3618 | if (mapped) |
3619 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
3620 | return; |
3621 | } | |
3622 | ||
3623 | /* | |
3624 | * For subpage, we can have dummy eb with page private. In this case, | |
3625 | * we can directly detach the private as such page is only attached to | |
3626 | * one dummy eb, no sharing. | |
3627 | */ | |
3628 | if (!mapped) { | |
3629 | btrfs_detach_subpage(fs_info, page); | |
3630 | return; | |
3631 | } | |
3632 | ||
3633 | btrfs_page_dec_eb_refs(fs_info, page); | |
3634 | ||
3635 | /* | |
3636 | * We can only detach the page private if there are no other ebs in the | |
3d078efa | 3637 | * page range and no unfinished IO. |
8ff8466d QW |
3638 | */ |
3639 | if (!page_range_has_eb(fs_info, page)) | |
3640 | btrfs_detach_subpage(fs_info, page); | |
3641 | ||
3642 | spin_unlock(&page->mapping->private_lock); | |
3643 | } | |
3644 | ||
3645 | /* Release all pages attached to the extent buffer */ | |
3646 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
3647 | { | |
3648 | int i; | |
3649 | int num_pages; | |
3650 | ||
3651 | ASSERT(!extent_buffer_under_io(eb)); | |
3652 | ||
3653 | num_pages = num_extent_pages(eb); | |
3654 | for (i = 0; i < num_pages; i++) { | |
3655 | struct page *page = eb->pages[i]; | |
3656 | ||
3657 | if (!page) | |
3658 | continue; | |
3659 | ||
3660 | detach_extent_buffer_page(eb, page); | |
5d2361db | 3661 | |
01327610 | 3662 | /* One for when we allocated the page */ |
09cbfeaf | 3663 | put_page(page); |
d64766fd | 3664 | } |
db7f3436 JB |
3665 | } |
3666 | ||
3667 | /* | |
3668 | * Helper for releasing the extent buffer. | |
3669 | */ | |
3670 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3671 | { | |
55ac0139 | 3672 | btrfs_release_extent_buffer_pages(eb); |
a40246e8 | 3673 | btrfs_leak_debug_del_eb(eb); |
db7f3436 JB |
3674 | __free_extent_buffer(eb); |
3675 | } | |
3676 | ||
f28491e0 JB |
3677 | static struct extent_buffer * |
3678 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 3679 | unsigned long len) |
d1310b2e CM |
3680 | { |
3681 | struct extent_buffer *eb = NULL; | |
3682 | ||
d1b5c567 | 3683 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
3684 | eb->start = start; |
3685 | eb->len = len; | |
f28491e0 | 3686 | eb->fs_info = fs_info; |
196d59ab | 3687 | init_rwsem(&eb->lock); |
b4ce94de | 3688 | |
a40246e8 | 3689 | btrfs_leak_debug_add_eb(eb); |
d3575156 | 3690 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 3691 | |
3083ee2e | 3692 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 3693 | atomic_set(&eb->refs, 1); |
0b32f4bb | 3694 | atomic_set(&eb->io_pages, 0); |
727011e0 | 3695 | |
deb67895 | 3696 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
3697 | |
3698 | return eb; | |
3699 | } | |
3700 | ||
2b48966a | 3701 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 3702 | { |
cc5e31a4 | 3703 | int i; |
815a51c7 | 3704 | struct extent_buffer *new; |
cc5e31a4 | 3705 | int num_pages = num_extent_pages(src); |
dd137dd1 | 3706 | int ret; |
815a51c7 | 3707 | |
3f556f78 | 3708 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
3709 | if (new == NULL) |
3710 | return NULL; | |
3711 | ||
62c053fb QW |
3712 | /* |
3713 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
3714 | * btrfs_release_extent_buffer() have different behavior for | |
3715 | * UNMAPPED subpage extent buffer. | |
3716 | */ | |
3717 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
3718 | ||
dd137dd1 STD |
3719 | ret = btrfs_alloc_page_array(num_pages, new->pages); |
3720 | if (ret) { | |
3721 | btrfs_release_extent_buffer(new); | |
3722 | return NULL; | |
3723 | } | |
3724 | ||
815a51c7 | 3725 | for (i = 0; i < num_pages; i++) { |
760f991f | 3726 | int ret; |
dd137dd1 | 3727 | struct page *p = new->pages[i]; |
760f991f | 3728 | |
760f991f QW |
3729 | ret = attach_extent_buffer_page(new, p, NULL); |
3730 | if (ret < 0) { | |
760f991f QW |
3731 | btrfs_release_extent_buffer(new); |
3732 | return NULL; | |
3733 | } | |
815a51c7 | 3734 | WARN_ON(PageDirty(p)); |
fba1acf9 | 3735 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 3736 | } |
92d83e94 | 3737 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
3738 | |
3739 | return new; | |
3740 | } | |
3741 | ||
0f331229 OS |
3742 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
3743 | u64 start, unsigned long len) | |
815a51c7 JS |
3744 | { |
3745 | struct extent_buffer *eb; | |
cc5e31a4 DS |
3746 | int num_pages; |
3747 | int i; | |
dd137dd1 | 3748 | int ret; |
815a51c7 | 3749 | |
3f556f78 | 3750 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
3751 | if (!eb) |
3752 | return NULL; | |
3753 | ||
65ad0104 | 3754 | num_pages = num_extent_pages(eb); |
dd137dd1 STD |
3755 | ret = btrfs_alloc_page_array(num_pages, eb->pages); |
3756 | if (ret) | |
3757 | goto err; | |
3758 | ||
815a51c7 | 3759 | for (i = 0; i < num_pages; i++) { |
dd137dd1 | 3760 | struct page *p = eb->pages[i]; |
09bc1f0f | 3761 | |
dd137dd1 | 3762 | ret = attach_extent_buffer_page(eb, p, NULL); |
09bc1f0f QW |
3763 | if (ret < 0) |
3764 | goto err; | |
815a51c7 | 3765 | } |
dd137dd1 | 3766 | |
815a51c7 JS |
3767 | set_extent_buffer_uptodate(eb); |
3768 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 3769 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
3770 | |
3771 | return eb; | |
3772 | err: | |
dd137dd1 STD |
3773 | for (i = 0; i < num_pages; i++) { |
3774 | if (eb->pages[i]) { | |
3775 | detach_extent_buffer_page(eb, eb->pages[i]); | |
3776 | __free_page(eb->pages[i]); | |
3777 | } | |
09bc1f0f | 3778 | } |
815a51c7 JS |
3779 | __free_extent_buffer(eb); |
3780 | return NULL; | |
3781 | } | |
3782 | ||
0f331229 | 3783 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 3784 | u64 start) |
0f331229 | 3785 | { |
da17066c | 3786 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
3787 | } |
3788 | ||
0b32f4bb JB |
3789 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
3790 | { | |
242e18c7 | 3791 | int refs; |
6bf9cd2e BB |
3792 | /* |
3793 | * The TREE_REF bit is first set when the extent_buffer is added | |
3794 | * to the radix tree. It is also reset, if unset, when a new reference | |
3795 | * is created by find_extent_buffer. | |
0b32f4bb | 3796 | * |
6bf9cd2e BB |
3797 | * It is only cleared in two cases: freeing the last non-tree |
3798 | * reference to the extent_buffer when its STALE bit is set or | |
f913cff3 | 3799 | * calling release_folio when the tree reference is the only reference. |
0b32f4bb | 3800 | * |
6bf9cd2e | 3801 | * In both cases, care is taken to ensure that the extent_buffer's |
f913cff3 | 3802 | * pages are not under io. However, release_folio can be concurrently |
6bf9cd2e BB |
3803 | * called with creating new references, which is prone to race |
3804 | * conditions between the calls to check_buffer_tree_ref in those | |
3805 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 3806 | * |
6bf9cd2e BB |
3807 | * The actual lifetime of the extent_buffer in the radix tree is |
3808 | * adequately protected by the refcount, but the TREE_REF bit and | |
3809 | * its corresponding reference are not. To protect against this | |
3810 | * class of races, we call check_buffer_tree_ref from the codepaths | |
3811 | * which trigger io after they set eb->io_pages. Note that once io is | |
3812 | * initiated, TREE_REF can no longer be cleared, so that is the | |
3813 | * moment at which any such race is best fixed. | |
0b32f4bb | 3814 | */ |
242e18c7 CM |
3815 | refs = atomic_read(&eb->refs); |
3816 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
3817 | return; | |
3818 | ||
594831c4 JB |
3819 | spin_lock(&eb->refs_lock); |
3820 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 3821 | atomic_inc(&eb->refs); |
594831c4 | 3822 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
3823 | } |
3824 | ||
2457aec6 MG |
3825 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
3826 | struct page *accessed) | |
5df4235e | 3827 | { |
cc5e31a4 | 3828 | int num_pages, i; |
5df4235e | 3829 | |
0b32f4bb JB |
3830 | check_buffer_tree_ref(eb); |
3831 | ||
65ad0104 | 3832 | num_pages = num_extent_pages(eb); |
5df4235e | 3833 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
3834 | struct page *p = eb->pages[i]; |
3835 | ||
2457aec6 MG |
3836 | if (p != accessed) |
3837 | mark_page_accessed(p); | |
5df4235e JB |
3838 | } |
3839 | } | |
3840 | ||
f28491e0 JB |
3841 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
3842 | u64 start) | |
452c75c3 CS |
3843 | { |
3844 | struct extent_buffer *eb; | |
3845 | ||
2f3186d8 QW |
3846 | eb = find_extent_buffer_nolock(fs_info, start); |
3847 | if (!eb) | |
3848 | return NULL; | |
3849 | /* | |
3850 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
3851 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
3852 | * another task running free_extent_buffer() might have seen that flag | |
3853 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
3854 | * writeback flags not set) and it's still in the tree (flag | |
3855 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
3856 | * decrementing the extent buffer's reference count twice. So here we | |
3857 | * could race and increment the eb's reference count, clear its stale | |
3858 | * flag, mark it as dirty and drop our reference before the other task | |
3859 | * finishes executing free_extent_buffer, which would later result in | |
3860 | * an attempt to free an extent buffer that is dirty. | |
3861 | */ | |
3862 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
3863 | spin_lock(&eb->refs_lock); | |
3864 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 3865 | } |
2f3186d8 QW |
3866 | mark_extent_buffer_accessed(eb, NULL); |
3867 | return eb; | |
452c75c3 CS |
3868 | } |
3869 | ||
faa2dbf0 JB |
3870 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
3871 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 3872 | u64 start) |
faa2dbf0 JB |
3873 | { |
3874 | struct extent_buffer *eb, *exists = NULL; | |
3875 | int ret; | |
3876 | ||
3877 | eb = find_extent_buffer(fs_info, start); | |
3878 | if (eb) | |
3879 | return eb; | |
da17066c | 3880 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 3881 | if (!eb) |
b6293c82 | 3882 | return ERR_PTR(-ENOMEM); |
faa2dbf0 | 3883 | eb->fs_info = fs_info; |
01cd3909 DS |
3884 | again: |
3885 | ret = radix_tree_preload(GFP_NOFS); | |
3886 | if (ret) { | |
3887 | exists = ERR_PTR(ret); | |
3888 | goto free_eb; | |
3889 | } | |
3890 | spin_lock(&fs_info->buffer_lock); | |
3891 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
3892 | start >> fs_info->sectorsize_bits, eb); | |
3893 | spin_unlock(&fs_info->buffer_lock); | |
3894 | radix_tree_preload_end(); | |
3895 | if (ret == -EEXIST) { | |
3896 | exists = find_extent_buffer(fs_info, start); | |
3897 | if (exists) | |
faa2dbf0 | 3898 | goto free_eb; |
01cd3909 DS |
3899 | else |
3900 | goto again; | |
3901 | } | |
faa2dbf0 JB |
3902 | check_buffer_tree_ref(eb); |
3903 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
3904 | ||
faa2dbf0 JB |
3905 | return eb; |
3906 | free_eb: | |
3907 | btrfs_release_extent_buffer(eb); | |
3908 | return exists; | |
3909 | } | |
3910 | #endif | |
3911 | ||
81982210 QW |
3912 | static struct extent_buffer *grab_extent_buffer( |
3913 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
3914 | { |
3915 | struct extent_buffer *exists; | |
3916 | ||
81982210 QW |
3917 | /* |
3918 | * For subpage case, we completely rely on radix tree to ensure we | |
3919 | * don't try to insert two ebs for the same bytenr. So here we always | |
3920 | * return NULL and just continue. | |
3921 | */ | |
fbca46eb | 3922 | if (fs_info->nodesize < PAGE_SIZE) |
81982210 QW |
3923 | return NULL; |
3924 | ||
c0f0a9e7 QW |
3925 | /* Page not yet attached to an extent buffer */ |
3926 | if (!PagePrivate(page)) | |
3927 | return NULL; | |
3928 | ||
3929 | /* | |
3930 | * We could have already allocated an eb for this page and attached one | |
3931 | * so lets see if we can get a ref on the existing eb, and if we can we | |
3932 | * know it's good and we can just return that one, else we know we can | |
3933 | * just overwrite page->private. | |
3934 | */ | |
3935 | exists = (struct extent_buffer *)page->private; | |
3936 | if (atomic_inc_not_zero(&exists->refs)) | |
3937 | return exists; | |
3938 | ||
3939 | WARN_ON(PageDirty(page)); | |
3940 | detach_page_private(page); | |
3941 | return NULL; | |
3942 | } | |
3943 | ||
fbca46eb QW |
3944 | static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start) |
3945 | { | |
3946 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { | |
3947 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
3948 | return -EINVAL; | |
3949 | } | |
3950 | ||
3951 | if (fs_info->nodesize < PAGE_SIZE && | |
3952 | offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) { | |
3953 | btrfs_err(fs_info, | |
3954 | "tree block crosses page boundary, start %llu nodesize %u", | |
3955 | start, fs_info->nodesize); | |
3956 | return -EINVAL; | |
3957 | } | |
3958 | if (fs_info->nodesize >= PAGE_SIZE && | |
1280d2d1 | 3959 | !PAGE_ALIGNED(start)) { |
fbca46eb QW |
3960 | btrfs_err(fs_info, |
3961 | "tree block is not page aligned, start %llu nodesize %u", | |
3962 | start, fs_info->nodesize); | |
3963 | return -EINVAL; | |
3964 | } | |
3965 | return 0; | |
3966 | } | |
3967 | ||
f28491e0 | 3968 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 3969 | u64 start, u64 owner_root, int level) |
d1310b2e | 3970 | { |
da17066c | 3971 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
3972 | int num_pages; |
3973 | int i; | |
09cbfeaf | 3974 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 3975 | struct extent_buffer *eb; |
6af118ce | 3976 | struct extent_buffer *exists = NULL; |
d1310b2e | 3977 | struct page *p; |
f28491e0 | 3978 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
b40130b2 | 3979 | u64 lockdep_owner = owner_root; |
d1310b2e | 3980 | int uptodate = 1; |
19fe0a8b | 3981 | int ret; |
d1310b2e | 3982 | |
fbca46eb | 3983 | if (check_eb_alignment(fs_info, start)) |
c871b0f2 | 3984 | return ERR_PTR(-EINVAL); |
c871b0f2 | 3985 | |
e9306ad4 QW |
3986 | #if BITS_PER_LONG == 32 |
3987 | if (start >= MAX_LFS_FILESIZE) { | |
3988 | btrfs_err_rl(fs_info, | |
3989 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
3990 | btrfs_err_32bit_limit(fs_info); | |
3991 | return ERR_PTR(-EOVERFLOW); | |
3992 | } | |
3993 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
3994 | btrfs_warn_32bit_limit(fs_info); | |
3995 | #endif | |
3996 | ||
f28491e0 | 3997 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 3998 | if (eb) |
6af118ce | 3999 | return eb; |
6af118ce | 4000 | |
23d79d81 | 4001 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4002 | if (!eb) |
c871b0f2 | 4003 | return ERR_PTR(-ENOMEM); |
b40130b2 JB |
4004 | |
4005 | /* | |
4006 | * The reloc trees are just snapshots, so we need them to appear to be | |
4007 | * just like any other fs tree WRT lockdep. | |
4008 | */ | |
4009 | if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID) | |
4010 | lockdep_owner = BTRFS_FS_TREE_OBJECTID; | |
4011 | ||
4012 | btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level); | |
d1310b2e | 4013 | |
65ad0104 | 4014 | num_pages = num_extent_pages(eb); |
727011e0 | 4015 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
4016 | struct btrfs_subpage *prealloc = NULL; |
4017 | ||
d1b5c567 | 4018 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4019 | if (!p) { |
4020 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4021 | goto free_eb; |
c871b0f2 | 4022 | } |
4f2de97a | 4023 | |
760f991f QW |
4024 | /* |
4025 | * Preallocate page->private for subpage case, so that we won't | |
4026 | * allocate memory with private_lock hold. The memory will be | |
4027 | * freed by attach_extent_buffer_page() or freed manually if | |
4028 | * we exit earlier. | |
4029 | * | |
4030 | * Although we have ensured one subpage eb can only have one | |
4031 | * page, but it may change in the future for 16K page size | |
4032 | * support, so we still preallocate the memory in the loop. | |
4033 | */ | |
fbca46eb | 4034 | if (fs_info->nodesize < PAGE_SIZE) { |
651fb419 QW |
4035 | prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA); |
4036 | if (IS_ERR(prealloc)) { | |
4037 | ret = PTR_ERR(prealloc); | |
fdf250db QW |
4038 | unlock_page(p); |
4039 | put_page(p); | |
4040 | exists = ERR_PTR(ret); | |
4041 | goto free_eb; | |
4042 | } | |
760f991f QW |
4043 | } |
4044 | ||
4f2de97a | 4045 | spin_lock(&mapping->private_lock); |
81982210 | 4046 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
4047 | if (exists) { |
4048 | spin_unlock(&mapping->private_lock); | |
4049 | unlock_page(p); | |
4050 | put_page(p); | |
4051 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 4052 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 4053 | goto free_eb; |
d1310b2e | 4054 | } |
760f991f QW |
4055 | /* Should not fail, as we have preallocated the memory */ |
4056 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
4057 | ASSERT(!ret); | |
8ff8466d QW |
4058 | /* |
4059 | * To inform we have extra eb under allocation, so that | |
4060 | * detach_extent_buffer_page() won't release the page private | |
4061 | * when the eb hasn't yet been inserted into radix tree. | |
4062 | * | |
4063 | * The ref will be decreased when the eb released the page, in | |
4064 | * detach_extent_buffer_page(). | |
4065 | * Thus needs no special handling in error path. | |
4066 | */ | |
4067 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 4068 | spin_unlock(&mapping->private_lock); |
760f991f | 4069 | |
1e5eb3d6 | 4070 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 4071 | eb->pages[i] = p; |
d1310b2e CM |
4072 | if (!PageUptodate(p)) |
4073 | uptodate = 0; | |
eb14ab8e CM |
4074 | |
4075 | /* | |
b16d011e NB |
4076 | * We can't unlock the pages just yet since the extent buffer |
4077 | * hasn't been properly inserted in the radix tree, this | |
f913cff3 | 4078 | * opens a race with btree_release_folio which can free a page |
b16d011e NB |
4079 | * while we are still filling in all pages for the buffer and |
4080 | * we could crash. | |
eb14ab8e | 4081 | */ |
d1310b2e CM |
4082 | } |
4083 | if (uptodate) | |
b4ce94de | 4084 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
01cd3909 DS |
4085 | again: |
4086 | ret = radix_tree_preload(GFP_NOFS); | |
4087 | if (ret) { | |
4088 | exists = ERR_PTR(ret); | |
4089 | goto free_eb; | |
4090 | } | |
4091 | ||
4092 | spin_lock(&fs_info->buffer_lock); | |
4093 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4094 | start >> fs_info->sectorsize_bits, eb); | |
4095 | spin_unlock(&fs_info->buffer_lock); | |
4096 | radix_tree_preload_end(); | |
4097 | if (ret == -EEXIST) { | |
4098 | exists = find_extent_buffer(fs_info, start); | |
4099 | if (exists) | |
452c75c3 | 4100 | goto free_eb; |
01cd3909 DS |
4101 | else |
4102 | goto again; | |
4103 | } | |
6af118ce | 4104 | /* add one reference for the tree */ |
0b32f4bb | 4105 | check_buffer_tree_ref(eb); |
34b41ace | 4106 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4107 | |
4108 | /* | |
b16d011e | 4109 | * Now it's safe to unlock the pages because any calls to |
f913cff3 | 4110 | * btree_release_folio will correctly detect that a page belongs to a |
b16d011e | 4111 | * live buffer and won't free them prematurely. |
eb14ab8e | 4112 | */ |
28187ae5 NB |
4113 | for (i = 0; i < num_pages; i++) |
4114 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
4115 | return eb; |
4116 | ||
6af118ce | 4117 | free_eb: |
5ca64f45 | 4118 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
4119 | for (i = 0; i < num_pages; i++) { |
4120 | if (eb->pages[i]) | |
4121 | unlock_page(eb->pages[i]); | |
4122 | } | |
eb14ab8e | 4123 | |
897ca6e9 | 4124 | btrfs_release_extent_buffer(eb); |
6af118ce | 4125 | return exists; |
d1310b2e | 4126 | } |
d1310b2e | 4127 | |
3083ee2e JB |
4128 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4129 | { | |
4130 | struct extent_buffer *eb = | |
4131 | container_of(head, struct extent_buffer, rcu_head); | |
4132 | ||
4133 | __free_extent_buffer(eb); | |
4134 | } | |
4135 | ||
f7a52a40 | 4136 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 4137 | __releases(&eb->refs_lock) |
3083ee2e | 4138 | { |
07e21c4d NB |
4139 | lockdep_assert_held(&eb->refs_lock); |
4140 | ||
3083ee2e JB |
4141 | WARN_ON(atomic_read(&eb->refs) == 0); |
4142 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 4143 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 4144 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 4145 | |
815a51c7 | 4146 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4147 | |
01cd3909 DS |
4148 | spin_lock(&fs_info->buffer_lock); |
4149 | radix_tree_delete(&fs_info->buffer_radix, | |
4150 | eb->start >> fs_info->sectorsize_bits); | |
4151 | spin_unlock(&fs_info->buffer_lock); | |
34b41ace JB |
4152 | } else { |
4153 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 4154 | } |
3083ee2e | 4155 | |
a40246e8 | 4156 | btrfs_leak_debug_del_eb(eb); |
3083ee2e | 4157 | /* Should be safe to release our pages at this point */ |
55ac0139 | 4158 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 4159 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 4160 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
4161 | __free_extent_buffer(eb); |
4162 | return 1; | |
4163 | } | |
4164 | #endif | |
3083ee2e | 4165 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4166 | return 1; |
3083ee2e JB |
4167 | } |
4168 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4169 | |
4170 | return 0; | |
3083ee2e JB |
4171 | } |
4172 | ||
d1310b2e CM |
4173 | void free_extent_buffer(struct extent_buffer *eb) |
4174 | { | |
242e18c7 | 4175 | int refs; |
d1310b2e CM |
4176 | if (!eb) |
4177 | return; | |
4178 | ||
e5677f05 | 4179 | refs = atomic_read(&eb->refs); |
242e18c7 | 4180 | while (1) { |
46cc775e NB |
4181 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
4182 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
4183 | refs == 1)) | |
242e18c7 | 4184 | break; |
e5677f05 | 4185 | if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1)) |
242e18c7 CM |
4186 | return; |
4187 | } | |
4188 | ||
3083ee2e JB |
4189 | spin_lock(&eb->refs_lock); |
4190 | if (atomic_read(&eb->refs) == 2 && | |
4191 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4192 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4193 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4194 | atomic_dec(&eb->refs); | |
4195 | ||
4196 | /* | |
4197 | * I know this is terrible, but it's temporary until we stop tracking | |
4198 | * the uptodate bits and such for the extent buffers. | |
4199 | */ | |
f7a52a40 | 4200 | release_extent_buffer(eb); |
3083ee2e JB |
4201 | } |
4202 | ||
4203 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4204 | { | |
4205 | if (!eb) | |
d1310b2e CM |
4206 | return; |
4207 | ||
3083ee2e JB |
4208 | spin_lock(&eb->refs_lock); |
4209 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4210 | ||
0b32f4bb | 4211 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4212 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4213 | atomic_dec(&eb->refs); | |
f7a52a40 | 4214 | release_extent_buffer(eb); |
d1310b2e | 4215 | } |
d1310b2e | 4216 | |
0d27797e QW |
4217 | static void btree_clear_page_dirty(struct page *page) |
4218 | { | |
4219 | ASSERT(PageDirty(page)); | |
4220 | ASSERT(PageLocked(page)); | |
4221 | clear_page_dirty_for_io(page); | |
4222 | xa_lock_irq(&page->mapping->i_pages); | |
4223 | if (!PageDirty(page)) | |
4224 | __xa_clear_mark(&page->mapping->i_pages, | |
4225 | page_index(page), PAGECACHE_TAG_DIRTY); | |
4226 | xa_unlock_irq(&page->mapping->i_pages); | |
4227 | } | |
4228 | ||
4229 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
4230 | { | |
4231 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4232 | struct page *page = eb->pages[0]; | |
4233 | bool last; | |
4234 | ||
4235 | /* btree_clear_page_dirty() needs page locked */ | |
4236 | lock_page(page); | |
4237 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
4238 | eb->len); | |
4239 | if (last) | |
4240 | btree_clear_page_dirty(page); | |
4241 | unlock_page(page); | |
4242 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4243 | } | |
4244 | ||
98c8d683 JB |
4245 | void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans, |
4246 | struct extent_buffer *eb) | |
d1310b2e | 4247 | { |
98c8d683 | 4248 | struct btrfs_fs_info *fs_info = eb->fs_info; |
cc5e31a4 DS |
4249 | int i; |
4250 | int num_pages; | |
d1310b2e CM |
4251 | struct page *page; |
4252 | ||
98c8d683 JB |
4253 | btrfs_assert_tree_write_locked(eb); |
4254 | ||
4255 | if (trans && btrfs_header_generation(eb) != trans->transid) | |
4256 | return; | |
4257 | ||
4258 | if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) | |
4259 | return; | |
4260 | ||
4261 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, | |
4262 | fs_info->dirty_metadata_batch); | |
4263 | ||
fbca46eb | 4264 | if (eb->fs_info->nodesize < PAGE_SIZE) |
0d27797e QW |
4265 | return clear_subpage_extent_buffer_dirty(eb); |
4266 | ||
65ad0104 | 4267 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
4268 | |
4269 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 4270 | page = eb->pages[i]; |
b9473439 | 4271 | if (!PageDirty(page)) |
d2c3f4f6 | 4272 | continue; |
a61e6f29 | 4273 | lock_page(page); |
0d27797e | 4274 | btree_clear_page_dirty(page); |
bf0da8c1 | 4275 | ClearPageError(page); |
a61e6f29 | 4276 | unlock_page(page); |
d1310b2e | 4277 | } |
0b32f4bb | 4278 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4279 | } |
d1310b2e | 4280 | |
abb57ef3 | 4281 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4282 | { |
cc5e31a4 DS |
4283 | int i; |
4284 | int num_pages; | |
abb57ef3 | 4285 | bool was_dirty; |
d1310b2e | 4286 | |
0b32f4bb JB |
4287 | check_buffer_tree_ref(eb); |
4288 | ||
b9473439 | 4289 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4290 | |
65ad0104 | 4291 | num_pages = num_extent_pages(eb); |
3083ee2e | 4292 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4293 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4294 | ||
0d27797e | 4295 | if (!was_dirty) { |
fbca46eb | 4296 | bool subpage = eb->fs_info->nodesize < PAGE_SIZE; |
51995c39 | 4297 | |
0d27797e QW |
4298 | /* |
4299 | * For subpage case, we can have other extent buffers in the | |
4300 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
4301 | * have to clear page dirty without subpage lock held. | |
4302 | * This can cause race where our page gets dirty cleared after | |
4303 | * we just set it. | |
4304 | * | |
4305 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
4306 | * its page for other reasons, we can use page lock to prevent | |
4307 | * the above race. | |
4308 | */ | |
4309 | if (subpage) | |
4310 | lock_page(eb->pages[0]); | |
4311 | for (i = 0; i < num_pages; i++) | |
4312 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
4313 | eb->start, eb->len); | |
4314 | if (subpage) | |
4315 | unlock_page(eb->pages[0]); | |
4316 | } | |
51995c39 LB |
4317 | #ifdef CONFIG_BTRFS_DEBUG |
4318 | for (i = 0; i < num_pages; i++) | |
4319 | ASSERT(PageDirty(eb->pages[i])); | |
4320 | #endif | |
4321 | ||
b9473439 | 4322 | return was_dirty; |
d1310b2e | 4323 | } |
d1310b2e | 4324 | |
69ba3927 | 4325 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 4326 | { |
251f2acc | 4327 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 4328 | struct page *page; |
cc5e31a4 | 4329 | int num_pages; |
251f2acc | 4330 | int i; |
1259ab75 | 4331 | |
b4ce94de | 4332 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 4333 | num_pages = num_extent_pages(eb); |
1259ab75 | 4334 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4335 | page = eb->pages[i]; |
fbca46eb QW |
4336 | if (!page) |
4337 | continue; | |
4338 | ||
4339 | /* | |
4340 | * This is special handling for metadata subpage, as regular | |
4341 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
4342 | */ | |
4343 | if (fs_info->nodesize >= PAGE_SIZE) | |
4344 | ClearPageUptodate(page); | |
4345 | else | |
4346 | btrfs_subpage_clear_uptodate(fs_info, page, eb->start, | |
4347 | eb->len); | |
1259ab75 | 4348 | } |
1259ab75 CM |
4349 | } |
4350 | ||
09c25a8c | 4351 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4352 | { |
251f2acc | 4353 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 4354 | struct page *page; |
cc5e31a4 | 4355 | int num_pages; |
251f2acc | 4356 | int i; |
d1310b2e | 4357 | |
0b32f4bb | 4358 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 4359 | num_pages = num_extent_pages(eb); |
d1310b2e | 4360 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4361 | page = eb->pages[i]; |
fbca46eb QW |
4362 | |
4363 | /* | |
4364 | * This is special handling for metadata subpage, as regular | |
4365 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
4366 | */ | |
4367 | if (fs_info->nodesize >= PAGE_SIZE) | |
4368 | SetPageUptodate(page); | |
4369 | else | |
4370 | btrfs_subpage_set_uptodate(fs_info, page, eb->start, | |
4371 | eb->len); | |
d1310b2e | 4372 | } |
d1310b2e | 4373 | } |
d1310b2e | 4374 | |
4012daf7 | 4375 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
947a6299 QW |
4376 | int mirror_num, |
4377 | struct btrfs_tree_parent_check *check) | |
4012daf7 QW |
4378 | { |
4379 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4380 | struct extent_io_tree *io_tree; | |
4381 | struct page *page = eb->pages[0]; | |
e5e886ba | 4382 | struct extent_state *cached_state = NULL; |
722c82ac | 4383 | struct btrfs_bio_ctrl bio_ctrl = { |
c000bc04 | 4384 | .opf = REQ_OP_READ, |
722c82ac | 4385 | .mirror_num = mirror_num, |
1d854e4f | 4386 | .parent_check = check, |
722c82ac | 4387 | }; |
4012daf7 QW |
4388 | int ret = 0; |
4389 | ||
4390 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
4391 | ASSERT(PagePrivate(page)); | |
947a6299 | 4392 | ASSERT(check); |
4012daf7 QW |
4393 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
4394 | ||
4395 | if (wait == WAIT_NONE) { | |
83ae4133 | 4396 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
e5e886ba | 4397 | &cached_state)) |
dc56219f | 4398 | return -EAGAIN; |
4012daf7 | 4399 | } else { |
e5e886ba JB |
4400 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
4401 | &cached_state); | |
4012daf7 QW |
4402 | if (ret < 0) |
4403 | return ret; | |
4404 | } | |
4405 | ||
4406 | ret = 0; | |
4407 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
4408 | PageUptodate(page) || | |
4409 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
4410 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
e5e886ba JB |
4411 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
4412 | &cached_state); | |
4012daf7 QW |
4413 | return ret; |
4414 | } | |
4415 | ||
4416 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
4417 | eb->read_mirror = 0; | |
4418 | atomic_set(&eb->io_pages, 1); | |
4419 | check_buffer_tree_ref(eb); | |
5467abba QW |
4420 | bio_ctrl.end_io_func = end_bio_extent_readpage; |
4421 | ||
4012daf7 QW |
4422 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); |
4423 | ||
3d078efa | 4424 | btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len); |
72b505dc | 4425 | ret = submit_extent_page(&bio_ctrl, eb->start, page, eb->len, |
eb8d0c6d | 4426 | eb->start - page_offset(page), 0); |
4012daf7 QW |
4427 | if (ret) { |
4428 | /* | |
4429 | * In the endio function, if we hit something wrong we will | |
4430 | * increase the io_pages, so here we need to decrease it for | |
4431 | * error path. | |
4432 | */ | |
4433 | atomic_dec(&eb->io_pages); | |
4434 | } | |
722c82ac | 4435 | submit_one_bio(&bio_ctrl); |
e5e886ba JB |
4436 | if (ret || wait != WAIT_COMPLETE) { |
4437 | free_extent_state(cached_state); | |
4012daf7 | 4438 | return ret; |
e5e886ba | 4439 | } |
4012daf7 | 4440 | |
123a7f00 | 4441 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, |
e5e886ba | 4442 | EXTENT_LOCKED, &cached_state); |
4012daf7 QW |
4443 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4444 | ret = -EIO; | |
4445 | return ret; | |
4446 | } | |
4447 | ||
947a6299 QW |
4448 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num, |
4449 | struct btrfs_tree_parent_check *check) | |
d1310b2e | 4450 | { |
cc5e31a4 | 4451 | int i; |
d1310b2e CM |
4452 | struct page *page; |
4453 | int err; | |
4454 | int ret = 0; | |
ce9adaa5 CM |
4455 | int locked_pages = 0; |
4456 | int all_uptodate = 1; | |
cc5e31a4 | 4457 | int num_pages; |
727011e0 | 4458 | unsigned long num_reads = 0; |
722c82ac | 4459 | struct btrfs_bio_ctrl bio_ctrl = { |
c000bc04 | 4460 | .opf = REQ_OP_READ, |
722c82ac | 4461 | .mirror_num = mirror_num, |
1d854e4f | 4462 | .parent_check = check, |
722c82ac | 4463 | }; |
a86c12c7 | 4464 | |
b4ce94de | 4465 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4466 | return 0; |
4467 | ||
651740a5 JB |
4468 | /* |
4469 | * We could have had EXTENT_BUFFER_UPTODATE cleared by the write | |
4470 | * operation, which could potentially still be in flight. In this case | |
4471 | * we simply want to return an error. | |
4472 | */ | |
4473 | if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))) | |
4474 | return -EIO; | |
4475 | ||
fbca46eb | 4476 | if (eb->fs_info->nodesize < PAGE_SIZE) |
947a6299 | 4477 | return read_extent_buffer_subpage(eb, wait, mirror_num, check); |
4012daf7 | 4478 | |
65ad0104 | 4479 | num_pages = num_extent_pages(eb); |
8436ea91 | 4480 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4481 | page = eb->pages[i]; |
bb82ab88 | 4482 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
4483 | /* |
4484 | * WAIT_NONE is only utilized by readahead. If we can't | |
4485 | * acquire the lock atomically it means either the eb | |
4486 | * is being read out or under modification. | |
4487 | * Either way the eb will be or has been cached, | |
4488 | * readahead can exit safely. | |
4489 | */ | |
2db04966 | 4490 | if (!trylock_page(page)) |
ce9adaa5 | 4491 | goto unlock_exit; |
d1310b2e CM |
4492 | } else { |
4493 | lock_page(page); | |
4494 | } | |
ce9adaa5 | 4495 | locked_pages++; |
2571e739 LB |
4496 | } |
4497 | /* | |
4498 | * We need to firstly lock all pages to make sure that | |
4499 | * the uptodate bit of our pages won't be affected by | |
4500 | * clear_extent_buffer_uptodate(). | |
4501 | */ | |
8436ea91 | 4502 | for (i = 0; i < num_pages; i++) { |
2571e739 | 4503 | page = eb->pages[i]; |
727011e0 CM |
4504 | if (!PageUptodate(page)) { |
4505 | num_reads++; | |
ce9adaa5 | 4506 | all_uptodate = 0; |
727011e0 | 4507 | } |
ce9adaa5 | 4508 | } |
2571e739 | 4509 | |
ce9adaa5 | 4510 | if (all_uptodate) { |
8436ea91 | 4511 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4512 | goto unlock_exit; |
4513 | } | |
4514 | ||
656f30db | 4515 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 4516 | eb->read_mirror = 0; |
0b32f4bb | 4517 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e | 4518 | /* |
f913cff3 | 4519 | * It is possible for release_folio to clear the TREE_REF bit before we |
6bf9cd2e BB |
4520 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. |
4521 | */ | |
4522 | check_buffer_tree_ref(eb); | |
5467abba | 4523 | bio_ctrl.end_io_func = end_bio_extent_readpage; |
8436ea91 | 4524 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4525 | page = eb->pages[i]; |
baf863b9 | 4526 | |
ce9adaa5 | 4527 | if (!PageUptodate(page)) { |
baf863b9 LB |
4528 | if (ret) { |
4529 | atomic_dec(&eb->io_pages); | |
4530 | unlock_page(page); | |
4531 | continue; | |
4532 | } | |
4533 | ||
f188591e | 4534 | ClearPageError(page); |
72b505dc | 4535 | err = submit_extent_page(&bio_ctrl, |
c000bc04 CH |
4536 | page_offset(page), page, |
4537 | PAGE_SIZE, 0, 0); | |
baf863b9 | 4538 | if (err) { |
baf863b9 | 4539 | /* |
0420177c NB |
4540 | * We failed to submit the bio so it's the |
4541 | * caller's responsibility to perform cleanup | |
4542 | * i.e unlock page/set error bit. | |
baf863b9 | 4543 | */ |
0420177c NB |
4544 | ret = err; |
4545 | SetPageError(page); | |
4546 | unlock_page(page); | |
baf863b9 LB |
4547 | atomic_dec(&eb->io_pages); |
4548 | } | |
d1310b2e CM |
4549 | } else { |
4550 | unlock_page(page); | |
4551 | } | |
4552 | } | |
4553 | ||
722c82ac | 4554 | submit_one_bio(&bio_ctrl); |
a86c12c7 | 4555 | |
bb82ab88 | 4556 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4557 | return ret; |
d397712b | 4558 | |
8436ea91 | 4559 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4560 | page = eb->pages[i]; |
d1310b2e | 4561 | wait_on_page_locked(page); |
d397712b | 4562 | if (!PageUptodate(page)) |
d1310b2e | 4563 | ret = -EIO; |
d1310b2e | 4564 | } |
d397712b | 4565 | |
d1310b2e | 4566 | return ret; |
ce9adaa5 CM |
4567 | |
4568 | unlock_exit: | |
d397712b | 4569 | while (locked_pages > 0) { |
ce9adaa5 | 4570 | locked_pages--; |
8436ea91 JB |
4571 | page = eb->pages[locked_pages]; |
4572 | unlock_page(page); | |
ce9adaa5 CM |
4573 | } |
4574 | return ret; | |
d1310b2e | 4575 | } |
d1310b2e | 4576 | |
f98b6215 QW |
4577 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
4578 | unsigned long len) | |
4579 | { | |
4580 | btrfs_warn(eb->fs_info, | |
4581 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
4582 | eb->start, eb->len, start, len); | |
4583 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
4584 | ||
4585 | return true; | |
4586 | } | |
4587 | ||
4588 | /* | |
4589 | * Check if the [start, start + len) range is valid before reading/writing | |
4590 | * the eb. | |
4591 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
4592 | * | |
4593 | * Caller should not touch the dst/src memory if this function returns error. | |
4594 | */ | |
4595 | static inline int check_eb_range(const struct extent_buffer *eb, | |
4596 | unsigned long start, unsigned long len) | |
4597 | { | |
4598 | unsigned long offset; | |
4599 | ||
4600 | /* start, start + len should not go beyond eb->len nor overflow */ | |
4601 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
4602 | return report_eb_range(eb, start, len); | |
4603 | ||
4604 | return false; | |
4605 | } | |
4606 | ||
1cbb1f45 JM |
4607 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
4608 | unsigned long start, unsigned long len) | |
d1310b2e CM |
4609 | { |
4610 | size_t cur; | |
4611 | size_t offset; | |
4612 | struct page *page; | |
4613 | char *kaddr; | |
4614 | char *dst = (char *)dstv; | |
884b07d0 | 4615 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4616 | |
f98b6215 | 4617 | if (check_eb_range(eb, start, len)) |
f716abd5 | 4618 | return; |
d1310b2e | 4619 | |
884b07d0 | 4620 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4621 | |
d397712b | 4622 | while (len > 0) { |
fb85fc9a | 4623 | page = eb->pages[i]; |
d1310b2e | 4624 | |
09cbfeaf | 4625 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 4626 | kaddr = page_address(page); |
d1310b2e | 4627 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4628 | |
4629 | dst += cur; | |
4630 | len -= cur; | |
4631 | offset = 0; | |
4632 | i++; | |
4633 | } | |
4634 | } | |
d1310b2e | 4635 | |
a48b73ec JB |
4636 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
4637 | void __user *dstv, | |
4638 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
4639 | { |
4640 | size_t cur; | |
4641 | size_t offset; | |
4642 | struct page *page; | |
4643 | char *kaddr; | |
4644 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 4645 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
4646 | int ret = 0; |
4647 | ||
4648 | WARN_ON(start > eb->len); | |
4649 | WARN_ON(start + len > eb->start + eb->len); | |
4650 | ||
884b07d0 | 4651 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
4652 | |
4653 | while (len > 0) { | |
fb85fc9a | 4654 | page = eb->pages[i]; |
550ac1d8 | 4655 | |
09cbfeaf | 4656 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 4657 | kaddr = page_address(page); |
a48b73ec | 4658 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
4659 | ret = -EFAULT; |
4660 | break; | |
4661 | } | |
4662 | ||
4663 | dst += cur; | |
4664 | len -= cur; | |
4665 | offset = 0; | |
4666 | i++; | |
4667 | } | |
4668 | ||
4669 | return ret; | |
4670 | } | |
4671 | ||
1cbb1f45 JM |
4672 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
4673 | unsigned long start, unsigned long len) | |
d1310b2e CM |
4674 | { |
4675 | size_t cur; | |
4676 | size_t offset; | |
4677 | struct page *page; | |
4678 | char *kaddr; | |
4679 | char *ptr = (char *)ptrv; | |
884b07d0 | 4680 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
4681 | int ret = 0; |
4682 | ||
f98b6215 QW |
4683 | if (check_eb_range(eb, start, len)) |
4684 | return -EINVAL; | |
d1310b2e | 4685 | |
884b07d0 | 4686 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4687 | |
d397712b | 4688 | while (len > 0) { |
fb85fc9a | 4689 | page = eb->pages[i]; |
d1310b2e | 4690 | |
09cbfeaf | 4691 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 4692 | |
a6591715 | 4693 | kaddr = page_address(page); |
d1310b2e | 4694 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4695 | if (ret) |
4696 | break; | |
4697 | ||
4698 | ptr += cur; | |
4699 | len -= cur; | |
4700 | offset = 0; | |
4701 | i++; | |
4702 | } | |
4703 | return ret; | |
4704 | } | |
d1310b2e | 4705 | |
b8f95771 QW |
4706 | /* |
4707 | * Check that the extent buffer is uptodate. | |
4708 | * | |
4709 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
4710 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
4711 | */ | |
4712 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
4713 | struct page *page) | |
4714 | { | |
4715 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4716 | ||
a50e1fcb JB |
4717 | /* |
4718 | * If we are using the commit root we could potentially clear a page | |
4719 | * Uptodate while we're using the extent buffer that we've previously | |
4720 | * looked up. We don't want to complain in this case, as the page was | |
4721 | * valid before, we just didn't write it out. Instead we want to catch | |
4722 | * the case where we didn't actually read the block properly, which | |
4723 | * would have !PageUptodate && !PageError, as we clear PageError before | |
4724 | * reading. | |
4725 | */ | |
fbca46eb | 4726 | if (fs_info->nodesize < PAGE_SIZE) { |
a50e1fcb | 4727 | bool uptodate, error; |
b8f95771 QW |
4728 | |
4729 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
4730 | eb->start, eb->len); | |
a50e1fcb JB |
4731 | error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len); |
4732 | WARN_ON(!uptodate && !error); | |
b8f95771 | 4733 | } else { |
a50e1fcb | 4734 | WARN_ON(!PageUptodate(page) && !PageError(page)); |
b8f95771 QW |
4735 | } |
4736 | } | |
4737 | ||
2b48966a | 4738 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
4739 | const void *srcv) |
4740 | { | |
4741 | char *kaddr; | |
4742 | ||
b8f95771 | 4743 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
4744 | kaddr = page_address(eb->pages[0]) + |
4745 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
4746 | chunk_tree_uuid)); | |
4747 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
4748 | } |
4749 | ||
2b48966a | 4750 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
4751 | { |
4752 | char *kaddr; | |
4753 | ||
b8f95771 | 4754 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
4755 | kaddr = page_address(eb->pages[0]) + |
4756 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
4757 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
4758 | } |
4759 | ||
2b48966a | 4760 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
4761 | unsigned long start, unsigned long len) |
4762 | { | |
4763 | size_t cur; | |
4764 | size_t offset; | |
4765 | struct page *page; | |
4766 | char *kaddr; | |
4767 | char *src = (char *)srcv; | |
884b07d0 | 4768 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4769 | |
d3575156 NA |
4770 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
4771 | ||
f98b6215 QW |
4772 | if (check_eb_range(eb, start, len)) |
4773 | return; | |
d1310b2e | 4774 | |
884b07d0 | 4775 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4776 | |
d397712b | 4777 | while (len > 0) { |
fb85fc9a | 4778 | page = eb->pages[i]; |
b8f95771 | 4779 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 4780 | |
09cbfeaf | 4781 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 4782 | kaddr = page_address(page); |
d1310b2e | 4783 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4784 | |
4785 | src += cur; | |
4786 | len -= cur; | |
4787 | offset = 0; | |
4788 | i++; | |
4789 | } | |
4790 | } | |
d1310b2e | 4791 | |
2b48966a | 4792 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 4793 | unsigned long len) |
d1310b2e CM |
4794 | { |
4795 | size_t cur; | |
4796 | size_t offset; | |
4797 | struct page *page; | |
4798 | char *kaddr; | |
884b07d0 | 4799 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4800 | |
f98b6215 QW |
4801 | if (check_eb_range(eb, start, len)) |
4802 | return; | |
d1310b2e | 4803 | |
884b07d0 | 4804 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4805 | |
d397712b | 4806 | while (len > 0) { |
fb85fc9a | 4807 | page = eb->pages[i]; |
b8f95771 | 4808 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 4809 | |
09cbfeaf | 4810 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 4811 | kaddr = page_address(page); |
b159fa28 | 4812 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
4813 | |
4814 | len -= cur; | |
4815 | offset = 0; | |
4816 | i++; | |
4817 | } | |
4818 | } | |
d1310b2e | 4819 | |
2b48966a DS |
4820 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
4821 | const struct extent_buffer *src) | |
58e8012c DS |
4822 | { |
4823 | int i; | |
cc5e31a4 | 4824 | int num_pages; |
58e8012c DS |
4825 | |
4826 | ASSERT(dst->len == src->len); | |
4827 | ||
fbca46eb | 4828 | if (dst->fs_info->nodesize >= PAGE_SIZE) { |
884b07d0 QW |
4829 | num_pages = num_extent_pages(dst); |
4830 | for (i = 0; i < num_pages; i++) | |
4831 | copy_page(page_address(dst->pages[i]), | |
4832 | page_address(src->pages[i])); | |
4833 | } else { | |
4834 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
4835 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
4836 | ||
fbca46eb | 4837 | ASSERT(src->fs_info->nodesize < PAGE_SIZE); |
884b07d0 QW |
4838 | memcpy(page_address(dst->pages[0]) + dst_offset, |
4839 | page_address(src->pages[0]) + src_offset, | |
4840 | src->len); | |
4841 | } | |
58e8012c DS |
4842 | } |
4843 | ||
2b48966a DS |
4844 | void copy_extent_buffer(const struct extent_buffer *dst, |
4845 | const struct extent_buffer *src, | |
d1310b2e CM |
4846 | unsigned long dst_offset, unsigned long src_offset, |
4847 | unsigned long len) | |
4848 | { | |
4849 | u64 dst_len = dst->len; | |
4850 | size_t cur; | |
4851 | size_t offset; | |
4852 | struct page *page; | |
4853 | char *kaddr; | |
884b07d0 | 4854 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 4855 | |
f98b6215 QW |
4856 | if (check_eb_range(dst, dst_offset, len) || |
4857 | check_eb_range(src, src_offset, len)) | |
4858 | return; | |
4859 | ||
d1310b2e CM |
4860 | WARN_ON(src->len != dst_len); |
4861 | ||
884b07d0 | 4862 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 4863 | |
d397712b | 4864 | while (len > 0) { |
fb85fc9a | 4865 | page = dst->pages[i]; |
b8f95771 | 4866 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 4867 | |
09cbfeaf | 4868 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 4869 | |
a6591715 | 4870 | kaddr = page_address(page); |
d1310b2e | 4871 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4872 | |
4873 | src_offset += cur; | |
4874 | len -= cur; | |
4875 | offset = 0; | |
4876 | i++; | |
4877 | } | |
4878 | } | |
d1310b2e | 4879 | |
3e1e8bb7 OS |
4880 | /* |
4881 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
4882 | * given bit number | |
4883 | * @eb: the extent buffer | |
4884 | * @start: offset of the bitmap item in the extent buffer | |
4885 | * @nr: bit number | |
4886 | * @page_index: return index of the page in the extent buffer that contains the | |
4887 | * given bit number | |
4888 | * @page_offset: return offset into the page given by page_index | |
4889 | * | |
4890 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
4891 | * contains a given bit. | |
4892 | */ | |
2b48966a | 4893 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
4894 | unsigned long start, unsigned long nr, |
4895 | unsigned long *page_index, | |
4896 | size_t *page_offset) | |
4897 | { | |
3e1e8bb7 OS |
4898 | size_t byte_offset = BIT_BYTE(nr); |
4899 | size_t offset; | |
4900 | ||
4901 | /* | |
4902 | * The byte we want is the offset of the extent buffer + the offset of | |
4903 | * the bitmap item in the extent buffer + the offset of the byte in the | |
4904 | * bitmap item. | |
4905 | */ | |
884b07d0 | 4906 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 4907 | |
09cbfeaf | 4908 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 4909 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
4910 | } |
4911 | ||
43dd529a DS |
4912 | /* |
4913 | * Determine whether a bit in a bitmap item is set. | |
4914 | * | |
4915 | * @eb: the extent buffer | |
4916 | * @start: offset of the bitmap item in the extent buffer | |
4917 | * @nr: bit number to test | |
3e1e8bb7 | 4918 | */ |
2b48966a | 4919 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
4920 | unsigned long nr) |
4921 | { | |
2fe1d551 | 4922 | u8 *kaddr; |
3e1e8bb7 OS |
4923 | struct page *page; |
4924 | unsigned long i; | |
4925 | size_t offset; | |
4926 | ||
4927 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
4928 | page = eb->pages[i]; | |
b8f95771 | 4929 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4930 | kaddr = page_address(page); |
4931 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
4932 | } | |
4933 | ||
43dd529a DS |
4934 | /* |
4935 | * Set an area of a bitmap to 1. | |
4936 | * | |
4937 | * @eb: the extent buffer | |
4938 | * @start: offset of the bitmap item in the extent buffer | |
4939 | * @pos: bit number of the first bit | |
4940 | * @len: number of bits to set | |
3e1e8bb7 | 4941 | */ |
2b48966a | 4942 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
4943 | unsigned long pos, unsigned long len) |
4944 | { | |
2fe1d551 | 4945 | u8 *kaddr; |
3e1e8bb7 OS |
4946 | struct page *page; |
4947 | unsigned long i; | |
4948 | size_t offset; | |
4949 | const unsigned int size = pos + len; | |
4950 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 4951 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
4952 | |
4953 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
4954 | page = eb->pages[i]; | |
b8f95771 | 4955 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4956 | kaddr = page_address(page); |
4957 | ||
4958 | while (len >= bits_to_set) { | |
4959 | kaddr[offset] |= mask_to_set; | |
4960 | len -= bits_to_set; | |
4961 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 4962 | mask_to_set = ~0; |
09cbfeaf | 4963 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
4964 | offset = 0; |
4965 | page = eb->pages[++i]; | |
b8f95771 | 4966 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4967 | kaddr = page_address(page); |
4968 | } | |
4969 | } | |
4970 | if (len) { | |
4971 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
4972 | kaddr[offset] |= mask_to_set; | |
4973 | } | |
4974 | } | |
4975 | ||
4976 | ||
43dd529a DS |
4977 | /* |
4978 | * Clear an area of a bitmap. | |
4979 | * | |
4980 | * @eb: the extent buffer | |
4981 | * @start: offset of the bitmap item in the extent buffer | |
4982 | * @pos: bit number of the first bit | |
4983 | * @len: number of bits to clear | |
3e1e8bb7 | 4984 | */ |
2b48966a DS |
4985 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
4986 | unsigned long start, unsigned long pos, | |
4987 | unsigned long len) | |
3e1e8bb7 | 4988 | { |
2fe1d551 | 4989 | u8 *kaddr; |
3e1e8bb7 OS |
4990 | struct page *page; |
4991 | unsigned long i; | |
4992 | size_t offset; | |
4993 | const unsigned int size = pos + len; | |
4994 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 4995 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
4996 | |
4997 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
4998 | page = eb->pages[i]; | |
b8f95771 | 4999 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
5000 | kaddr = page_address(page); |
5001 | ||
5002 | while (len >= bits_to_clear) { | |
5003 | kaddr[offset] &= ~mask_to_clear; | |
5004 | len -= bits_to_clear; | |
5005 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5006 | mask_to_clear = ~0; |
09cbfeaf | 5007 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5008 | offset = 0; |
5009 | page = eb->pages[++i]; | |
b8f95771 | 5010 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
5011 | kaddr = page_address(page); |
5012 | } | |
5013 | } | |
5014 | if (len) { | |
5015 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5016 | kaddr[offset] &= ~mask_to_clear; | |
5017 | } | |
5018 | } | |
5019 | ||
3387206f ST |
5020 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5021 | { | |
5022 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5023 | return distance < len; | |
5024 | } | |
5025 | ||
d1310b2e CM |
5026 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5027 | unsigned long dst_off, unsigned long src_off, | |
5028 | unsigned long len) | |
5029 | { | |
a6591715 | 5030 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5031 | char *src_kaddr; |
727011e0 | 5032 | int must_memmove = 0; |
d1310b2e | 5033 | |
3387206f | 5034 | if (dst_page != src_page) { |
a6591715 | 5035 | src_kaddr = page_address(src_page); |
3387206f | 5036 | } else { |
d1310b2e | 5037 | src_kaddr = dst_kaddr; |
727011e0 CM |
5038 | if (areas_overlap(src_off, dst_off, len)) |
5039 | must_memmove = 1; | |
3387206f | 5040 | } |
d1310b2e | 5041 | |
727011e0 CM |
5042 | if (must_memmove) |
5043 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5044 | else | |
5045 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5046 | } |
5047 | ||
2b48966a DS |
5048 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
5049 | unsigned long dst_offset, unsigned long src_offset, | |
5050 | unsigned long len) | |
d1310b2e CM |
5051 | { |
5052 | size_t cur; | |
5053 | size_t dst_off_in_page; | |
5054 | size_t src_off_in_page; | |
d1310b2e CM |
5055 | unsigned long dst_i; |
5056 | unsigned long src_i; | |
5057 | ||
f98b6215 QW |
5058 | if (check_eb_range(dst, dst_offset, len) || |
5059 | check_eb_range(dst, src_offset, len)) | |
5060 | return; | |
d1310b2e | 5061 | |
d397712b | 5062 | while (len > 0) { |
884b07d0 QW |
5063 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
5064 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 5065 | |
884b07d0 QW |
5066 | dst_i = get_eb_page_index(dst_offset); |
5067 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 5068 | |
09cbfeaf | 5069 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5070 | src_off_in_page)); |
5071 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5072 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5073 | |
fb85fc9a | 5074 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5075 | dst_off_in_page, src_off_in_page, cur); |
5076 | ||
5077 | src_offset += cur; | |
5078 | dst_offset += cur; | |
5079 | len -= cur; | |
5080 | } | |
5081 | } | |
d1310b2e | 5082 | |
2b48966a DS |
5083 | void memmove_extent_buffer(const struct extent_buffer *dst, |
5084 | unsigned long dst_offset, unsigned long src_offset, | |
5085 | unsigned long len) | |
d1310b2e CM |
5086 | { |
5087 | size_t cur; | |
5088 | size_t dst_off_in_page; | |
5089 | size_t src_off_in_page; | |
5090 | unsigned long dst_end = dst_offset + len - 1; | |
5091 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
5092 | unsigned long dst_i; |
5093 | unsigned long src_i; | |
5094 | ||
f98b6215 QW |
5095 | if (check_eb_range(dst, dst_offset, len) || |
5096 | check_eb_range(dst, src_offset, len)) | |
5097 | return; | |
727011e0 | 5098 | if (dst_offset < src_offset) { |
d1310b2e CM |
5099 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5100 | return; | |
5101 | } | |
d397712b | 5102 | while (len > 0) { |
884b07d0 QW |
5103 | dst_i = get_eb_page_index(dst_end); |
5104 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 5105 | |
884b07d0 QW |
5106 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
5107 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
5108 | |
5109 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5110 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5111 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5112 | dst_off_in_page - cur + 1, |
5113 | src_off_in_page - cur + 1, cur); | |
5114 | ||
5115 | dst_end -= cur; | |
5116 | src_end -= cur; | |
5117 | len -= cur; | |
5118 | } | |
5119 | } | |
6af118ce | 5120 | |
01cd3909 | 5121 | #define GANG_LOOKUP_SIZE 16 |
d1e86e3f QW |
5122 | static struct extent_buffer *get_next_extent_buffer( |
5123 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
5124 | { | |
01cd3909 DS |
5125 | struct extent_buffer *gang[GANG_LOOKUP_SIZE]; |
5126 | struct extent_buffer *found = NULL; | |
d1e86e3f | 5127 | u64 page_start = page_offset(page); |
01cd3909 | 5128 | u64 cur = page_start; |
d1e86e3f QW |
5129 | |
5130 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
d1e86e3f QW |
5131 | lockdep_assert_held(&fs_info->buffer_lock); |
5132 | ||
01cd3909 DS |
5133 | while (cur < page_start + PAGE_SIZE) { |
5134 | int ret; | |
5135 | int i; | |
5136 | ||
5137 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, | |
5138 | (void **)gang, cur >> fs_info->sectorsize_bits, | |
5139 | min_t(unsigned int, GANG_LOOKUP_SIZE, | |
5140 | PAGE_SIZE / fs_info->nodesize)); | |
5141 | if (ret == 0) | |
5142 | goto out; | |
5143 | for (i = 0; i < ret; i++) { | |
5144 | /* Already beyond page end */ | |
5145 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
5146 | goto out; | |
5147 | /* Found one */ | |
5148 | if (gang[i]->start >= bytenr) { | |
5149 | found = gang[i]; | |
5150 | goto out; | |
5151 | } | |
5152 | } | |
5153 | cur = gang[ret - 1]->start + gang[ret - 1]->len; | |
d1e86e3f | 5154 | } |
01cd3909 DS |
5155 | out: |
5156 | return found; | |
d1e86e3f QW |
5157 | } |
5158 | ||
5159 | static int try_release_subpage_extent_buffer(struct page *page) | |
5160 | { | |
5161 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
5162 | u64 cur = page_offset(page); | |
5163 | const u64 end = page_offset(page) + PAGE_SIZE; | |
5164 | int ret; | |
5165 | ||
5166 | while (cur < end) { | |
5167 | struct extent_buffer *eb = NULL; | |
5168 | ||
5169 | /* | |
5170 | * Unlike try_release_extent_buffer() which uses page->private | |
5171 | * to grab buffer, for subpage case we rely on radix tree, thus | |
5172 | * we need to ensure radix tree consistency. | |
5173 | * | |
5174 | * We also want an atomic snapshot of the radix tree, thus go | |
5175 | * with spinlock rather than RCU. | |
5176 | */ | |
5177 | spin_lock(&fs_info->buffer_lock); | |
5178 | eb = get_next_extent_buffer(fs_info, page, cur); | |
5179 | if (!eb) { | |
5180 | /* No more eb in the page range after or at cur */ | |
5181 | spin_unlock(&fs_info->buffer_lock); | |
5182 | break; | |
5183 | } | |
5184 | cur = eb->start + eb->len; | |
5185 | ||
5186 | /* | |
5187 | * The same as try_release_extent_buffer(), to ensure the eb | |
5188 | * won't disappear out from under us. | |
5189 | */ | |
5190 | spin_lock(&eb->refs_lock); | |
5191 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
5192 | spin_unlock(&eb->refs_lock); | |
5193 | spin_unlock(&fs_info->buffer_lock); | |
5194 | break; | |
5195 | } | |
5196 | spin_unlock(&fs_info->buffer_lock); | |
5197 | ||
5198 | /* | |
5199 | * If tree ref isn't set then we know the ref on this eb is a | |
5200 | * real ref, so just return, this eb will likely be freed soon | |
5201 | * anyway. | |
5202 | */ | |
5203 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
5204 | spin_unlock(&eb->refs_lock); | |
5205 | break; | |
5206 | } | |
5207 | ||
5208 | /* | |
5209 | * Here we don't care about the return value, we will always | |
5210 | * check the page private at the end. And | |
5211 | * release_extent_buffer() will release the refs_lock. | |
5212 | */ | |
5213 | release_extent_buffer(eb); | |
5214 | } | |
5215 | /* | |
5216 | * Finally to check if we have cleared page private, as if we have | |
5217 | * released all ebs in the page, the page private should be cleared now. | |
5218 | */ | |
5219 | spin_lock(&page->mapping->private_lock); | |
5220 | if (!PagePrivate(page)) | |
5221 | ret = 1; | |
5222 | else | |
5223 | ret = 0; | |
5224 | spin_unlock(&page->mapping->private_lock); | |
5225 | return ret; | |
5226 | ||
5227 | } | |
5228 | ||
f7a52a40 | 5229 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5230 | { |
6af118ce | 5231 | struct extent_buffer *eb; |
6af118ce | 5232 | |
fbca46eb | 5233 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
d1e86e3f QW |
5234 | return try_release_subpage_extent_buffer(page); |
5235 | ||
3083ee2e | 5236 | /* |
d1e86e3f QW |
5237 | * We need to make sure nobody is changing page->private, as we rely on |
5238 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
5239 | */ |
5240 | spin_lock(&page->mapping->private_lock); | |
5241 | if (!PagePrivate(page)) { | |
5242 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5243 | return 1; |
45f49bce | 5244 | } |
6af118ce | 5245 | |
3083ee2e JB |
5246 | eb = (struct extent_buffer *)page->private; |
5247 | BUG_ON(!eb); | |
19fe0a8b MX |
5248 | |
5249 | /* | |
3083ee2e JB |
5250 | * This is a little awful but should be ok, we need to make sure that |
5251 | * the eb doesn't disappear out from under us while we're looking at | |
5252 | * this page. | |
19fe0a8b | 5253 | */ |
3083ee2e | 5254 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5255 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5256 | spin_unlock(&eb->refs_lock); |
5257 | spin_unlock(&page->mapping->private_lock); | |
5258 | return 0; | |
b9473439 | 5259 | } |
3083ee2e | 5260 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5261 | |
19fe0a8b | 5262 | /* |
3083ee2e JB |
5263 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5264 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5265 | */ |
3083ee2e JB |
5266 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5267 | spin_unlock(&eb->refs_lock); | |
5268 | return 0; | |
b9473439 | 5269 | } |
19fe0a8b | 5270 | |
f7a52a40 | 5271 | return release_extent_buffer(eb); |
6af118ce | 5272 | } |
bfb484d9 JB |
5273 | |
5274 | /* | |
5275 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
5276 | * @fs_info: the fs_info | |
5277 | * @bytenr: bytenr to read | |
3fbaf258 | 5278 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 5279 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 5280 | * @level: level for the eb |
bfb484d9 JB |
5281 | * |
5282 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
5283 | * normal uptodate check of the eb, without checking the generation. If we have | |
5284 | * to read the block we will not block on anything. | |
5285 | */ | |
5286 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 5287 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 | 5288 | { |
947a6299 QW |
5289 | struct btrfs_tree_parent_check check = { |
5290 | .has_first_key = 0, | |
5291 | .level = level, | |
5292 | .transid = gen | |
5293 | }; | |
bfb484d9 JB |
5294 | struct extent_buffer *eb; |
5295 | int ret; | |
5296 | ||
3fbaf258 | 5297 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
5298 | if (IS_ERR(eb)) |
5299 | return; | |
5300 | ||
5301 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
5302 | free_extent_buffer(eb); | |
5303 | return; | |
5304 | } | |
5305 | ||
947a6299 | 5306 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check); |
bfb484d9 JB |
5307 | if (ret < 0) |
5308 | free_extent_buffer_stale(eb); | |
5309 | else | |
5310 | free_extent_buffer(eb); | |
5311 | } | |
5312 | ||
5313 | /* | |
5314 | * btrfs_readahead_node_child - readahead a node's child block | |
5315 | * @node: parent node we're reading from | |
5316 | * @slot: slot in the parent node for the child we want to read | |
5317 | * | |
5318 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
5319 | * the slot in the node provided. | |
5320 | */ | |
5321 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
5322 | { | |
5323 | btrfs_readahead_tree_block(node->fs_info, | |
5324 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
5325 | btrfs_header_owner(node), |
5326 | btrfs_node_ptr_generation(node, slot), | |
5327 | btrfs_header_level(node) - 1); | |
bfb484d9 | 5328 | } |