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; |
5467abba | 104 | btrfs_bio_end_io_t end_io_func; |
7aab8b32 | 105 | |
1d854e4f QW |
106 | /* |
107 | * This is for metadata read, to provide the extra needed verification | |
108 | * info. This has to be provided for submit_one_bio(), as | |
109 | * submit_one_bio() can submit a bio if it ends at stripe boundary. If | |
110 | * no such parent_check is provided, the metadata can hit false alert at | |
111 | * endio time. | |
112 | */ | |
113 | struct btrfs_tree_parent_check *parent_check; | |
114 | ||
ee5f017d DS |
115 | /* |
116 | * Tell writepage not to lock the state bits for this range, it still | |
117 | * does the unlocking. | |
771ed689 | 118 | */ |
8ec8519b | 119 | bool extent_locked; |
ffbd517d | 120 | |
ee5f017d | 121 | /* Tell the submit_bio code to use REQ_SYNC */ |
8ec8519b | 122 | bool sync_io; |
d1310b2e CM |
123 | }; |
124 | ||
722c82ac | 125 | static void submit_one_bio(struct btrfs_bio_ctrl *bio_ctrl) |
bb58eb9e | 126 | { |
722c82ac | 127 | struct bio *bio; |
7aa51232 | 128 | struct bio_vec *bv; |
35a8d7da | 129 | struct inode *inode; |
722c82ac CH |
130 | int mirror_num; |
131 | ||
132 | if (!bio_ctrl->bio) | |
133 | return; | |
bb58eb9e | 134 | |
722c82ac | 135 | bio = bio_ctrl->bio; |
7aa51232 | 136 | bv = bio_first_bvec_all(bio); |
35a8d7da | 137 | inode = bv->bv_page->mapping->host; |
722c82ac | 138 | mirror_num = bio_ctrl->mirror_num; |
bb58eb9e | 139 | |
e0eefe07 QW |
140 | /* Caller should ensure the bio has at least some range added */ |
141 | ASSERT(bio->bi_iter.bi_size); | |
c9583ada | 142 | |
7aa51232 | 143 | btrfs_bio(bio)->file_offset = page_offset(bv->bv_page) + bv->bv_offset; |
bb58eb9e | 144 | |
35a8d7da | 145 | if (!is_data_inode(inode)) { |
1d854e4f QW |
146 | if (btrfs_op(bio) != BTRFS_MAP_WRITE) { |
147 | /* | |
148 | * For metadata read, we should have the parent_check, | |
149 | * and copy it to bbio for metadata verification. | |
150 | */ | |
151 | ASSERT(bio_ctrl->parent_check); | |
152 | memcpy(&btrfs_bio(bio)->parent_check, | |
153 | bio_ctrl->parent_check, | |
154 | sizeof(struct btrfs_tree_parent_check)); | |
155 | } | |
35a8d7da | 156 | bio->bi_opf |= REQ_META; |
1d854e4f | 157 | } |
390ed29b | 158 | |
35a8d7da CH |
159 | if (btrfs_op(bio) == BTRFS_MAP_READ && |
160 | bio_ctrl->compress_type != BTRFS_COMPRESS_NONE) | |
161 | btrfs_submit_compressed_read(inode, bio, mirror_num); | |
162 | else | |
163 | btrfs_submit_bio(bio, mirror_num); | |
164 | ||
917f32a2 | 165 | /* The bio is owned by the end_io handler now */ |
722c82ac | 166 | bio_ctrl->bio = NULL; |
3065976b QW |
167 | } |
168 | ||
f4340622 | 169 | /* |
ee5f017d | 170 | * Submit or fail the current bio in the bio_ctrl structure. |
f4340622 | 171 | */ |
ee5f017d | 172 | static void submit_write_bio(struct btrfs_bio_ctrl *bio_ctrl, int ret) |
bb58eb9e | 173 | { |
ee5f017d | 174 | struct bio *bio = bio_ctrl->bio; |
bb58eb9e | 175 | |
9845e5dd CH |
176 | if (!bio) |
177 | return; | |
178 | ||
179 | if (ret) { | |
180 | ASSERT(ret < 0); | |
917f32a2 CH |
181 | btrfs_bio_end_io(btrfs_bio(bio), errno_to_blk_status(ret)); |
182 | /* The bio is owned by the end_io handler now */ | |
ee5f017d | 183 | bio_ctrl->bio = NULL; |
9845e5dd | 184 | } else { |
ee5f017d | 185 | submit_one_bio(bio_ctrl); |
bb58eb9e QW |
186 | } |
187 | } | |
e2932ee0 | 188 | |
a62a3bd9 JB |
189 | int __init extent_buffer_init_cachep(void) |
190 | { | |
837e1972 | 191 | extent_buffer_cache = kmem_cache_create("btrfs_extent_buffer", |
9601e3f6 | 192 | sizeof(struct extent_buffer), 0, |
fba4b697 | 193 | SLAB_MEM_SPREAD, NULL); |
a62a3bd9 | 194 | if (!extent_buffer_cache) |
6f0d04f8 | 195 | return -ENOMEM; |
b208c2f7 | 196 | |
d1310b2e | 197 | return 0; |
d1310b2e CM |
198 | } |
199 | ||
a62a3bd9 | 200 | void __cold extent_buffer_free_cachep(void) |
d1310b2e | 201 | { |
8c0a8537 KS |
202 | /* |
203 | * Make sure all delayed rcu free are flushed before we | |
204 | * destroy caches. | |
205 | */ | |
206 | rcu_barrier(); | |
5598e900 | 207 | kmem_cache_destroy(extent_buffer_cache); |
d1310b2e CM |
208 | } |
209 | ||
bd1fa4f0 | 210 | void extent_range_clear_dirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 211 | { |
09cbfeaf KS |
212 | unsigned long index = start >> PAGE_SHIFT; |
213 | unsigned long end_index = end >> PAGE_SHIFT; | |
4adaa611 CM |
214 | struct page *page; |
215 | ||
216 | while (index <= end_index) { | |
217 | page = find_get_page(inode->i_mapping, index); | |
218 | BUG_ON(!page); /* Pages should be in the extent_io_tree */ | |
219 | clear_page_dirty_for_io(page); | |
09cbfeaf | 220 | put_page(page); |
4adaa611 CM |
221 | index++; |
222 | } | |
4adaa611 CM |
223 | } |
224 | ||
f6311572 | 225 | void extent_range_redirty_for_io(struct inode *inode, u64 start, u64 end) |
4adaa611 | 226 | { |
ebf55c88 | 227 | struct address_space *mapping = inode->i_mapping; |
09cbfeaf KS |
228 | unsigned long index = start >> PAGE_SHIFT; |
229 | unsigned long end_index = end >> PAGE_SHIFT; | |
ebf55c88 | 230 | struct folio *folio; |
4adaa611 CM |
231 | |
232 | while (index <= end_index) { | |
ebf55c88 MWO |
233 | folio = filemap_get_folio(mapping, index); |
234 | filemap_dirty_folio(mapping, folio); | |
235 | folio_account_redirty(folio); | |
236 | index += folio_nr_pages(folio); | |
237 | folio_put(folio); | |
4adaa611 | 238 | } |
4adaa611 CM |
239 | } |
240 | ||
ed8f13bf QW |
241 | /* |
242 | * Process one page for __process_pages_contig(). | |
243 | * | |
244 | * Return >0 if we hit @page == @locked_page. | |
245 | * Return 0 if we updated the page status. | |
246 | * Return -EGAIN if the we need to try again. | |
247 | * (For PAGE_LOCK case but got dirty page or page not belong to mapping) | |
248 | */ | |
e38992be QW |
249 | static int process_one_page(struct btrfs_fs_info *fs_info, |
250 | struct address_space *mapping, | |
ed8f13bf | 251 | struct page *page, struct page *locked_page, |
e38992be | 252 | unsigned long page_ops, u64 start, u64 end) |
ed8f13bf | 253 | { |
e38992be QW |
254 | u32 len; |
255 | ||
256 | ASSERT(end + 1 - start != 0 && end + 1 - start < U32_MAX); | |
257 | len = end + 1 - start; | |
258 | ||
ed8f13bf | 259 | if (page_ops & PAGE_SET_ORDERED) |
b945a463 | 260 | btrfs_page_clamp_set_ordered(fs_info, page, start, len); |
ed8f13bf | 261 | if (page_ops & PAGE_SET_ERROR) |
e38992be | 262 | btrfs_page_clamp_set_error(fs_info, page, start, len); |
ed8f13bf | 263 | if (page_ops & PAGE_START_WRITEBACK) { |
e38992be QW |
264 | btrfs_page_clamp_clear_dirty(fs_info, page, start, len); |
265 | btrfs_page_clamp_set_writeback(fs_info, page, start, len); | |
ed8f13bf QW |
266 | } |
267 | if (page_ops & PAGE_END_WRITEBACK) | |
e38992be | 268 | btrfs_page_clamp_clear_writeback(fs_info, page, start, len); |
a33a8e9a QW |
269 | |
270 | if (page == locked_page) | |
271 | return 1; | |
272 | ||
ed8f13bf | 273 | if (page_ops & PAGE_LOCK) { |
1e1de387 QW |
274 | int ret; |
275 | ||
276 | ret = btrfs_page_start_writer_lock(fs_info, page, start, len); | |
277 | if (ret) | |
278 | return ret; | |
ed8f13bf | 279 | if (!PageDirty(page) || page->mapping != mapping) { |
1e1de387 | 280 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
281 | return -EAGAIN; |
282 | } | |
283 | } | |
284 | if (page_ops & PAGE_UNLOCK) | |
1e1de387 | 285 | btrfs_page_end_writer_lock(fs_info, page, start, len); |
ed8f13bf QW |
286 | return 0; |
287 | } | |
288 | ||
da2c7009 LB |
289 | static int __process_pages_contig(struct address_space *mapping, |
290 | struct page *locked_page, | |
98af9ab1 | 291 | u64 start, u64 end, unsigned long page_ops, |
ed8f13bf QW |
292 | u64 *processed_end) |
293 | { | |
e38992be | 294 | struct btrfs_fs_info *fs_info = btrfs_sb(mapping->host->i_sb); |
ed8f13bf QW |
295 | pgoff_t start_index = start >> PAGE_SHIFT; |
296 | pgoff_t end_index = end >> PAGE_SHIFT; | |
297 | pgoff_t index = start_index; | |
ed8f13bf | 298 | unsigned long pages_processed = 0; |
04c6b79a | 299 | struct folio_batch fbatch; |
ed8f13bf QW |
300 | int err = 0; |
301 | int i; | |
302 | ||
303 | if (page_ops & PAGE_LOCK) { | |
304 | ASSERT(page_ops == PAGE_LOCK); | |
305 | ASSERT(processed_end && *processed_end == start); | |
306 | } | |
307 | ||
04c6b79a | 308 | if ((page_ops & PAGE_SET_ERROR) && start_index <= end_index) |
ed8f13bf QW |
309 | mapping_set_error(mapping, -EIO); |
310 | ||
04c6b79a VMO |
311 | folio_batch_init(&fbatch); |
312 | while (index <= end_index) { | |
313 | int found_folios; | |
314 | ||
315 | found_folios = filemap_get_folios_contig(mapping, &index, | |
316 | end_index, &fbatch); | |
ed8f13bf | 317 | |
04c6b79a | 318 | if (found_folios == 0) { |
ed8f13bf QW |
319 | /* |
320 | * Only if we're going to lock these pages, we can find | |
321 | * nothing at @index. | |
322 | */ | |
323 | ASSERT(page_ops & PAGE_LOCK); | |
324 | err = -EAGAIN; | |
325 | goto out; | |
326 | } | |
327 | ||
04c6b79a | 328 | for (i = 0; i < found_folios; i++) { |
ed8f13bf | 329 | int process_ret; |
04c6b79a | 330 | struct folio *folio = fbatch.folios[i]; |
e38992be | 331 | process_ret = process_one_page(fs_info, mapping, |
04c6b79a | 332 | &folio->page, locked_page, page_ops, |
e38992be | 333 | start, end); |
ed8f13bf | 334 | if (process_ret < 0) { |
ed8f13bf | 335 | err = -EAGAIN; |
04c6b79a | 336 | folio_batch_release(&fbatch); |
ed8f13bf QW |
337 | goto out; |
338 | } | |
04c6b79a | 339 | pages_processed += folio_nr_pages(folio); |
ed8f13bf | 340 | } |
04c6b79a | 341 | folio_batch_release(&fbatch); |
ed8f13bf QW |
342 | cond_resched(); |
343 | } | |
344 | out: | |
345 | if (err && processed_end) { | |
346 | /* | |
347 | * Update @processed_end. I know this is awful since it has | |
348 | * two different return value patterns (inclusive vs exclusive). | |
349 | * | |
350 | * But the exclusive pattern is necessary if @start is 0, or we | |
351 | * underflow and check against processed_end won't work as | |
352 | * expected. | |
353 | */ | |
354 | if (pages_processed) | |
355 | *processed_end = min(end, | |
356 | ((u64)(start_index + pages_processed) << PAGE_SHIFT) - 1); | |
357 | else | |
358 | *processed_end = start; | |
359 | } | |
360 | return err; | |
361 | } | |
da2c7009 | 362 | |
143bede5 JM |
363 | static noinline void __unlock_for_delalloc(struct inode *inode, |
364 | struct page *locked_page, | |
365 | u64 start, u64 end) | |
c8b97818 | 366 | { |
09cbfeaf KS |
367 | unsigned long index = start >> PAGE_SHIFT; |
368 | unsigned long end_index = end >> PAGE_SHIFT; | |
c8b97818 | 369 | |
76c0021d | 370 | ASSERT(locked_page); |
c8b97818 | 371 | if (index == locked_page->index && end_index == index) |
143bede5 | 372 | return; |
c8b97818 | 373 | |
98af9ab1 | 374 | __process_pages_contig(inode->i_mapping, locked_page, start, end, |
76c0021d | 375 | PAGE_UNLOCK, NULL); |
c8b97818 CM |
376 | } |
377 | ||
378 | static noinline int lock_delalloc_pages(struct inode *inode, | |
379 | struct page *locked_page, | |
380 | u64 delalloc_start, | |
381 | u64 delalloc_end) | |
382 | { | |
09cbfeaf | 383 | unsigned long index = delalloc_start >> PAGE_SHIFT; |
09cbfeaf | 384 | unsigned long end_index = delalloc_end >> PAGE_SHIFT; |
98af9ab1 | 385 | u64 processed_end = delalloc_start; |
c8b97818 | 386 | int ret; |
c8b97818 | 387 | |
76c0021d | 388 | ASSERT(locked_page); |
c8b97818 CM |
389 | if (index == locked_page->index && index == end_index) |
390 | return 0; | |
391 | ||
98af9ab1 QW |
392 | ret = __process_pages_contig(inode->i_mapping, locked_page, delalloc_start, |
393 | delalloc_end, PAGE_LOCK, &processed_end); | |
394 | if (ret == -EAGAIN && processed_end > delalloc_start) | |
76c0021d | 395 | __unlock_for_delalloc(inode, locked_page, delalloc_start, |
98af9ab1 | 396 | processed_end); |
c8b97818 CM |
397 | return ret; |
398 | } | |
399 | ||
400 | /* | |
3522e903 | 401 | * Find and lock a contiguous range of bytes in the file marked as delalloc, no |
2749f7ef | 402 | * more than @max_bytes. |
c8b97818 | 403 | * |
2749f7ef QW |
404 | * @start: The original start bytenr to search. |
405 | * Will store the extent range start bytenr. | |
406 | * @end: The original end bytenr of the search range | |
407 | * Will store the extent range end bytenr. | |
408 | * | |
409 | * Return true if we find a delalloc range which starts inside the original | |
410 | * range, and @start/@end will store the delalloc range start/end. | |
411 | * | |
412 | * Return false if we can't find any delalloc range which starts inside the | |
413 | * original range, and @start/@end will be the non-delalloc range start/end. | |
c8b97818 | 414 | */ |
ce9f967f | 415 | EXPORT_FOR_TESTS |
3522e903 | 416 | noinline_for_stack bool find_lock_delalloc_range(struct inode *inode, |
294e30fe | 417 | struct page *locked_page, u64 *start, |
917aacec | 418 | u64 *end) |
c8b97818 | 419 | { |
f7b12a62 | 420 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
9978059b | 421 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
2749f7ef QW |
422 | const u64 orig_start = *start; |
423 | const u64 orig_end = *end; | |
f7b12a62 NA |
424 | /* The sanity tests may not set a valid fs_info. */ |
425 | u64 max_bytes = fs_info ? fs_info->max_extent_size : BTRFS_MAX_EXTENT_SIZE; | |
c8b97818 CM |
426 | u64 delalloc_start; |
427 | u64 delalloc_end; | |
3522e903 | 428 | bool found; |
9655d298 | 429 | struct extent_state *cached_state = NULL; |
c8b97818 CM |
430 | int ret; |
431 | int loops = 0; | |
432 | ||
2749f7ef QW |
433 | /* Caller should pass a valid @end to indicate the search range end */ |
434 | ASSERT(orig_end > orig_start); | |
435 | ||
436 | /* The range should at least cover part of the page */ | |
437 | ASSERT(!(orig_start >= page_offset(locked_page) + PAGE_SIZE || | |
438 | orig_end <= page_offset(locked_page))); | |
c8b97818 CM |
439 | again: |
440 | /* step one, find a bunch of delalloc bytes starting at start */ | |
441 | delalloc_start = *start; | |
442 | delalloc_end = 0; | |
083e75e7 JB |
443 | found = btrfs_find_delalloc_range(tree, &delalloc_start, &delalloc_end, |
444 | max_bytes, &cached_state); | |
2749f7ef | 445 | if (!found || delalloc_end <= *start || delalloc_start > orig_end) { |
c8b97818 | 446 | *start = delalloc_start; |
2749f7ef QW |
447 | |
448 | /* @delalloc_end can be -1, never go beyond @orig_end */ | |
449 | *end = min(delalloc_end, orig_end); | |
c2a128d2 | 450 | free_extent_state(cached_state); |
3522e903 | 451 | return false; |
c8b97818 CM |
452 | } |
453 | ||
70b99e69 CM |
454 | /* |
455 | * start comes from the offset of locked_page. We have to lock | |
456 | * pages in order, so we can't process delalloc bytes before | |
457 | * locked_page | |
458 | */ | |
d397712b | 459 | if (delalloc_start < *start) |
70b99e69 | 460 | delalloc_start = *start; |
70b99e69 | 461 | |
c8b97818 CM |
462 | /* |
463 | * make sure to limit the number of pages we try to lock down | |
c8b97818 | 464 | */ |
7bf811a5 JB |
465 | if (delalloc_end + 1 - delalloc_start > max_bytes) |
466 | delalloc_end = delalloc_start + max_bytes - 1; | |
d397712b | 467 | |
c8b97818 CM |
468 | /* step two, lock all the pages after the page that has start */ |
469 | ret = lock_delalloc_pages(inode, locked_page, | |
470 | delalloc_start, delalloc_end); | |
9bfd61d9 | 471 | ASSERT(!ret || ret == -EAGAIN); |
c8b97818 CM |
472 | if (ret == -EAGAIN) { |
473 | /* some of the pages are gone, lets avoid looping by | |
474 | * shortening the size of the delalloc range we're searching | |
475 | */ | |
9655d298 | 476 | free_extent_state(cached_state); |
7d788742 | 477 | cached_state = NULL; |
c8b97818 | 478 | if (!loops) { |
09cbfeaf | 479 | max_bytes = PAGE_SIZE; |
c8b97818 CM |
480 | loops = 1; |
481 | goto again; | |
482 | } else { | |
3522e903 | 483 | found = false; |
c8b97818 CM |
484 | goto out_failed; |
485 | } | |
486 | } | |
c8b97818 CM |
487 | |
488 | /* step three, lock the state bits for the whole range */ | |
570eb97b | 489 | lock_extent(tree, delalloc_start, delalloc_end, &cached_state); |
c8b97818 CM |
490 | |
491 | /* then test to make sure it is all still delalloc */ | |
492 | ret = test_range_bit(tree, delalloc_start, delalloc_end, | |
9655d298 | 493 | EXTENT_DELALLOC, 1, cached_state); |
c8b97818 | 494 | if (!ret) { |
570eb97b JB |
495 | unlock_extent(tree, delalloc_start, delalloc_end, |
496 | &cached_state); | |
c8b97818 CM |
497 | __unlock_for_delalloc(inode, locked_page, |
498 | delalloc_start, delalloc_end); | |
499 | cond_resched(); | |
500 | goto again; | |
501 | } | |
9655d298 | 502 | free_extent_state(cached_state); |
c8b97818 CM |
503 | *start = delalloc_start; |
504 | *end = delalloc_end; | |
505 | out_failed: | |
506 | return found; | |
507 | } | |
508 | ||
ad7ff17b | 509 | void extent_clear_unlock_delalloc(struct btrfs_inode *inode, u64 start, u64 end, |
74e9194a | 510 | struct page *locked_page, |
f97e27e9 | 511 | u32 clear_bits, unsigned long page_ops) |
873695b3 | 512 | { |
bd015294 | 513 | clear_extent_bit(&inode->io_tree, start, end, clear_bits, NULL); |
873695b3 | 514 | |
ad7ff17b | 515 | __process_pages_contig(inode->vfs_inode.i_mapping, locked_page, |
98af9ab1 | 516 | start, end, page_ops, NULL); |
873695b3 LB |
517 | } |
518 | ||
150e4b05 QW |
519 | static void end_page_read(struct page *page, bool uptodate, u64 start, u32 len) |
520 | { | |
521 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
522 | ||
523 | ASSERT(page_offset(page) <= start && | |
524 | start + len <= page_offset(page) + PAGE_SIZE); | |
525 | ||
150e4b05 | 526 | if (uptodate) { |
14605409 BB |
527 | if (fsverity_active(page->mapping->host) && |
528 | !PageError(page) && | |
529 | !PageUptodate(page) && | |
530 | start < i_size_read(page->mapping->host) && | |
531 | !fsverity_verify_page(page)) { | |
532 | btrfs_page_set_error(fs_info, page, start, len); | |
533 | } else { | |
534 | btrfs_page_set_uptodate(fs_info, page, start, len); | |
535 | } | |
150e4b05 QW |
536 | } else { |
537 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
538 | btrfs_page_set_error(fs_info, page, start, len); | |
539 | } | |
540 | ||
fbca46eb | 541 | if (!btrfs_is_subpage(fs_info, page)) |
150e4b05 | 542 | unlock_page(page); |
3d078efa | 543 | else |
150e4b05 QW |
544 | btrfs_subpage_end_reader(fs_info, page, start, len); |
545 | } | |
546 | ||
d1310b2e CM |
547 | /* lots and lots of room for performance fixes in the end_bio funcs */ |
548 | ||
b5227c07 | 549 | void end_extent_writepage(struct page *page, int err, u64 start, u64 end) |
87826df0 | 550 | { |
38a39ac7 | 551 | struct btrfs_inode *inode; |
25c1252a | 552 | const bool uptodate = (err == 0); |
3e2426bd | 553 | int ret = 0; |
87826df0 | 554 | |
38a39ac7 QW |
555 | ASSERT(page && page->mapping); |
556 | inode = BTRFS_I(page->mapping->host); | |
557 | btrfs_writepage_endio_finish_ordered(inode, page, start, end, uptodate); | |
87826df0 | 558 | |
87826df0 | 559 | if (!uptodate) { |
963e4db8 QW |
560 | const struct btrfs_fs_info *fs_info = inode->root->fs_info; |
561 | u32 len; | |
562 | ||
563 | ASSERT(end + 1 - start <= U32_MAX); | |
564 | len = end + 1 - start; | |
565 | ||
566 | btrfs_page_clear_uptodate(fs_info, page, start, len); | |
567 | btrfs_page_set_error(fs_info, page, start, len); | |
bff5baf8 | 568 | ret = err < 0 ? err : -EIO; |
5dca6eea | 569 | mapping_set_error(page->mapping, ret); |
87826df0 | 570 | } |
87826df0 JM |
571 | } |
572 | ||
d1310b2e CM |
573 | /* |
574 | * after a writepage IO is done, we need to: | |
575 | * clear the uptodate bits on error | |
576 | * clear the writeback bits in the extent tree for this IO | |
577 | * end_page_writeback if the page has no more pending IO | |
578 | * | |
579 | * Scheduling is not allowed, so the extent state tree is expected | |
580 | * to have one and only one object corresponding to this IO. | |
581 | */ | |
917f32a2 | 582 | static void end_bio_extent_writepage(struct btrfs_bio *bbio) |
d1310b2e | 583 | { |
917f32a2 | 584 | struct bio *bio = &bbio->bio; |
4e4cbee9 | 585 | int error = blk_status_to_errno(bio->bi_status); |
2c30c71b | 586 | struct bio_vec *bvec; |
d1310b2e CM |
587 | u64 start; |
588 | u64 end; | |
6dc4f100 | 589 | struct bvec_iter_all iter_all; |
d1310b2e | 590 | |
c09abff8 | 591 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 592 | bio_for_each_segment_all(bvec, bio, iter_all) { |
d1310b2e | 593 | struct page *page = bvec->bv_page; |
0b246afa JM |
594 | struct inode *inode = page->mapping->host; |
595 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); | |
321a02db QW |
596 | const u32 sectorsize = fs_info->sectorsize; |
597 | ||
598 | /* Our read/write should always be sector aligned. */ | |
599 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
600 | btrfs_err(fs_info, | |
601 | "partial page write in btrfs with offset %u and length %u", | |
602 | bvec->bv_offset, bvec->bv_len); | |
603 | else if (!IS_ALIGNED(bvec->bv_len, sectorsize)) | |
604 | btrfs_info(fs_info, | |
605 | "incomplete page write with offset %u and length %u", | |
606 | bvec->bv_offset, bvec->bv_len); | |
607 | ||
608 | start = page_offset(page) + bvec->bv_offset; | |
609 | end = start + bvec->bv_len - 1; | |
d1310b2e | 610 | |
4e4cbee9 | 611 | end_extent_writepage(page, error, start, end); |
9047e317 QW |
612 | |
613 | btrfs_page_clear_writeback(fs_info, page, start, bvec->bv_len); | |
2c30c71b | 614 | } |
2b1f55b0 | 615 | |
d1310b2e | 616 | bio_put(bio); |
d1310b2e CM |
617 | } |
618 | ||
94e8c95c QW |
619 | /* |
620 | * Record previously processed extent range | |
621 | * | |
622 | * For endio_readpage_release_extent() to handle a full extent range, reducing | |
623 | * the extent io operations. | |
624 | */ | |
625 | struct processed_extent { | |
626 | struct btrfs_inode *inode; | |
627 | /* Start of the range in @inode */ | |
628 | u64 start; | |
2e626e56 | 629 | /* End of the range in @inode */ |
94e8c95c QW |
630 | u64 end; |
631 | bool uptodate; | |
632 | }; | |
633 | ||
634 | /* | |
635 | * Try to release processed extent range | |
636 | * | |
637 | * May not release the extent range right now if the current range is | |
638 | * contiguous to processed extent. | |
639 | * | |
640 | * Will release processed extent when any of @inode, @uptodate, the range is | |
641 | * no longer contiguous to the processed range. | |
642 | * | |
643 | * Passing @inode == NULL will force processed extent to be released. | |
644 | */ | |
645 | static void endio_readpage_release_extent(struct processed_extent *processed, | |
646 | struct btrfs_inode *inode, u64 start, u64 end, | |
647 | bool uptodate) | |
883d0de4 MX |
648 | { |
649 | struct extent_state *cached = NULL; | |
94e8c95c QW |
650 | struct extent_io_tree *tree; |
651 | ||
652 | /* The first extent, initialize @processed */ | |
653 | if (!processed->inode) | |
654 | goto update; | |
883d0de4 | 655 | |
94e8c95c QW |
656 | /* |
657 | * Contiguous to processed extent, just uptodate the end. | |
658 | * | |
659 | * Several things to notice: | |
660 | * | |
661 | * - bio can be merged as long as on-disk bytenr is contiguous | |
662 | * This means we can have page belonging to other inodes, thus need to | |
663 | * check if the inode still matches. | |
664 | * - bvec can contain range beyond current page for multi-page bvec | |
665 | * Thus we need to do processed->end + 1 >= start check | |
666 | */ | |
667 | if (processed->inode == inode && processed->uptodate == uptodate && | |
668 | processed->end + 1 >= start && end >= processed->end) { | |
669 | processed->end = end; | |
670 | return; | |
671 | } | |
672 | ||
673 | tree = &processed->inode->io_tree; | |
674 | /* | |
675 | * Now we don't have range contiguous to the processed range, release | |
676 | * the processed range now. | |
677 | */ | |
48acc47d | 678 | unlock_extent(tree, processed->start, processed->end, &cached); |
94e8c95c QW |
679 | |
680 | update: | |
681 | /* Update processed to current range */ | |
682 | processed->inode = inode; | |
683 | processed->start = start; | |
684 | processed->end = end; | |
685 | processed->uptodate = uptodate; | |
883d0de4 MX |
686 | } |
687 | ||
92082d40 QW |
688 | static void begin_page_read(struct btrfs_fs_info *fs_info, struct page *page) |
689 | { | |
690 | ASSERT(PageLocked(page)); | |
fbca46eb | 691 | if (!btrfs_is_subpage(fs_info, page)) |
92082d40 QW |
692 | return; |
693 | ||
694 | ASSERT(PagePrivate(page)); | |
695 | btrfs_subpage_start_reader(fs_info, page, page_offset(page), PAGE_SIZE); | |
696 | } | |
697 | ||
d9bb77d5 | 698 | /* |
01cd3909 | 699 | * Find extent buffer for a givne bytenr. |
d9bb77d5 QW |
700 | * |
701 | * This is for end_bio_extent_readpage(), thus we can't do any unsafe locking | |
702 | * in endio context. | |
703 | */ | |
704 | static struct extent_buffer *find_extent_buffer_readpage( | |
705 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
706 | { | |
707 | struct extent_buffer *eb; | |
708 | ||
709 | /* | |
710 | * For regular sectorsize, we can use page->private to grab extent | |
711 | * buffer | |
712 | */ | |
fbca46eb | 713 | if (fs_info->nodesize >= PAGE_SIZE) { |
d9bb77d5 QW |
714 | ASSERT(PagePrivate(page) && page->private); |
715 | return (struct extent_buffer *)page->private; | |
716 | } | |
717 | ||
01cd3909 DS |
718 | /* For subpage case, we need to lookup buffer radix tree */ |
719 | rcu_read_lock(); | |
720 | eb = radix_tree_lookup(&fs_info->buffer_radix, | |
721 | bytenr >> fs_info->sectorsize_bits); | |
722 | rcu_read_unlock(); | |
d9bb77d5 QW |
723 | ASSERT(eb); |
724 | return eb; | |
725 | } | |
726 | ||
d1310b2e CM |
727 | /* |
728 | * after a readpage IO is done, we need to: | |
729 | * clear the uptodate bits on error | |
730 | * set the uptodate bits if things worked | |
731 | * set the page up to date if all extents in the tree are uptodate | |
732 | * clear the lock bit in the extent tree | |
733 | * unlock the page if there are no other extents locked for it | |
734 | * | |
735 | * Scheduling is not allowed, so the extent state tree is expected | |
736 | * to have one and only one object corresponding to this IO. | |
737 | */ | |
917f32a2 | 738 | static void end_bio_extent_readpage(struct btrfs_bio *bbio) |
d1310b2e | 739 | { |
917f32a2 | 740 | struct bio *bio = &bbio->bio; |
2c30c71b | 741 | struct bio_vec *bvec; |
94e8c95c | 742 | struct processed_extent processed = { 0 }; |
7ffd27e3 QW |
743 | /* |
744 | * The offset to the beginning of a bio, since one bio can never be | |
745 | * larger than UINT_MAX, u32 here is enough. | |
746 | */ | |
747 | u32 bio_offset = 0; | |
5cf1ab56 | 748 | int mirror; |
6dc4f100 | 749 | struct bvec_iter_all iter_all; |
d1310b2e | 750 | |
c09abff8 | 751 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 752 | bio_for_each_segment_all(bvec, bio, iter_all) { |
150e4b05 | 753 | bool uptodate = !bio->bi_status; |
d1310b2e | 754 | struct page *page = bvec->bv_page; |
a71754fc | 755 | struct inode *inode = page->mapping->host; |
ab8d0fc4 | 756 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
7ffd27e3 QW |
757 | const u32 sectorsize = fs_info->sectorsize; |
758 | u64 start; | |
759 | u64 end; | |
760 | u32 len; | |
507903b8 | 761 | |
ab8d0fc4 JM |
762 | btrfs_debug(fs_info, |
763 | "end_bio_extent_readpage: bi_sector=%llu, err=%d, mirror=%u", | |
1201b58b | 764 | bio->bi_iter.bi_sector, bio->bi_status, |
c3a3b19b | 765 | bbio->mirror_num); |
902b22f3 | 766 | |
8b8bbd46 QW |
767 | /* |
768 | * We always issue full-sector reads, but if some block in a | |
769 | * page fails to read, blk_update_request() will advance | |
770 | * bv_offset and adjust bv_len to compensate. Print a warning | |
771 | * for unaligned offsets, and an error if they don't add up to | |
772 | * a full sector. | |
773 | */ | |
774 | if (!IS_ALIGNED(bvec->bv_offset, sectorsize)) | |
775 | btrfs_err(fs_info, | |
776 | "partial page read in btrfs with offset %u and length %u", | |
777 | bvec->bv_offset, bvec->bv_len); | |
778 | else if (!IS_ALIGNED(bvec->bv_offset + bvec->bv_len, | |
779 | sectorsize)) | |
780 | btrfs_info(fs_info, | |
781 | "incomplete page read with offset %u and length %u", | |
782 | bvec->bv_offset, bvec->bv_len); | |
783 | ||
784 | start = page_offset(page) + bvec->bv_offset; | |
785 | end = start + bvec->bv_len - 1; | |
facc8a22 | 786 | len = bvec->bv_len; |
d1310b2e | 787 | |
c3a3b19b | 788 | mirror = bbio->mirror_num; |
7609afac CH |
789 | if (uptodate && !is_data_inode(inode) && |
790 | btrfs_validate_metadata_buffer(bbio, page, start, end, mirror)) | |
791 | uptodate = false; | |
ea466794 | 792 | |
883d0de4 | 793 | if (likely(uptodate)) { |
a71754fc | 794 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 795 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
a71754fc | 796 | |
c28ea613 QW |
797 | /* |
798 | * Zero out the remaining part if this range straddles | |
799 | * i_size. | |
800 | * | |
801 | * Here we should only zero the range inside the bvec, | |
802 | * not touch anything else. | |
803 | * | |
804 | * NOTE: i_size is exclusive while end is inclusive. | |
805 | */ | |
806 | if (page->index == end_index && i_size <= end) { | |
807 | u32 zero_start = max(offset_in_page(i_size), | |
d2dcc8ed | 808 | offset_in_page(start)); |
c28ea613 QW |
809 | |
810 | zero_user_segment(page, zero_start, | |
811 | offset_in_page(end) + 1); | |
812 | } | |
7609afac | 813 | } else if (!is_data_inode(inode)) { |
97861cd1 CH |
814 | struct extent_buffer *eb; |
815 | ||
816 | eb = find_extent_buffer_readpage(fs_info, page, start); | |
817 | set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
818 | eb->read_mirror = mirror; | |
819 | atomic_dec(&eb->io_pages); | |
70dec807 | 820 | } |
97861cd1 | 821 | |
7609afac CH |
822 | /* Update page status and unlock. */ |
823 | end_page_read(page, uptodate, start, len); | |
824 | endio_readpage_release_extent(&processed, BTRFS_I(inode), | |
825 | start, end, PageUptodate(page)); | |
97861cd1 | 826 | |
7ffd27e3 QW |
827 | ASSERT(bio_offset + len > bio_offset); |
828 | bio_offset += len; | |
883d0de4 | 829 | |
2c30c71b | 830 | } |
94e8c95c QW |
831 | /* Release the last extent */ |
832 | endio_readpage_release_extent(&processed, NULL, 0, 0, false); | |
d1310b2e | 833 | bio_put(bio); |
d1310b2e CM |
834 | } |
835 | ||
43dd529a | 836 | /* |
dd137dd1 STD |
837 | * Populate every free slot in a provided array with pages. |
838 | * | |
839 | * @nr_pages: number of pages to allocate | |
840 | * @page_array: the array to fill with pages; any existing non-null entries in | |
841 | * the array will be skipped | |
842 | * | |
843 | * Return: 0 if all pages were able to be allocated; | |
844 | * -ENOMEM otherwise, and the caller is responsible for freeing all | |
845 | * non-null page pointers in the array. | |
846 | */ | |
847 | int btrfs_alloc_page_array(unsigned int nr_pages, struct page **page_array) | |
848 | { | |
91d6ac1d | 849 | unsigned int allocated; |
dd137dd1 | 850 | |
91d6ac1d STD |
851 | for (allocated = 0; allocated < nr_pages;) { |
852 | unsigned int last = allocated; | |
dd137dd1 | 853 | |
91d6ac1d STD |
854 | allocated = alloc_pages_bulk_array(GFP_NOFS, nr_pages, page_array); |
855 | ||
395cb57e STD |
856 | if (allocated == nr_pages) |
857 | return 0; | |
858 | ||
91d6ac1d STD |
859 | /* |
860 | * During this iteration, no page could be allocated, even | |
861 | * though alloc_pages_bulk_array() falls back to alloc_page() | |
862 | * if it could not bulk-allocate. So we must be out of memory. | |
863 | */ | |
864 | if (allocated == last) | |
dd137dd1 | 865 | return -ENOMEM; |
395cb57e STD |
866 | |
867 | memalloc_retry_wait(GFP_NOFS); | |
dd137dd1 STD |
868 | } |
869 | return 0; | |
870 | } | |
871 | ||
43dd529a DS |
872 | /* |
873 | * Attempt to add a page to bio. | |
953651eb | 874 | * |
43dd529a DS |
875 | * @bio_ctrl: record both the bio, and its bio_flags |
876 | * @page: page to add to the bio | |
877 | * @disk_bytenr: offset of the new bio or to check whether we are adding | |
878 | * a contiguous page to the previous one | |
879 | * @size: portion of page that we want to write | |
880 | * @pg_offset: starting offset in the page | |
881 | * @compress_type: compression type of the current bio to see if we can merge them | |
953651eb NA |
882 | * |
883 | * Attempt to add a page to bio considering stripe alignment etc. | |
884 | * | |
e0eefe07 QW |
885 | * Return >= 0 for the number of bytes added to the bio. |
886 | * Can return 0 if the current bio is already at stripe/zone boundary. | |
887 | * Return <0 for error. | |
953651eb | 888 | */ |
e0eefe07 QW |
889 | static int btrfs_bio_add_page(struct btrfs_bio_ctrl *bio_ctrl, |
890 | struct page *page, | |
891 | u64 disk_bytenr, unsigned int size, | |
892 | unsigned int pg_offset, | |
cb3a12d9 | 893 | enum btrfs_compression_type compress_type) |
953651eb | 894 | { |
390ed29b QW |
895 | struct bio *bio = bio_ctrl->bio; |
896 | u32 bio_size = bio->bi_iter.bi_size; | |
e0eefe07 | 897 | u32 real_size; |
953651eb | 898 | const sector_t sector = disk_bytenr >> SECTOR_SHIFT; |
4a445b7b | 899 | bool contig = false; |
953651eb | 900 | |
390ed29b QW |
901 | ASSERT(bio); |
902 | /* The limit should be calculated when bio_ctrl->bio is allocated */ | |
2380220e | 903 | ASSERT(bio_ctrl->len_to_oe_boundary); |
0f07003b | 904 | if (bio_ctrl->compress_type != compress_type) |
e0eefe07 | 905 | return 0; |
953651eb | 906 | |
4a445b7b QW |
907 | |
908 | if (bio->bi_iter.bi_size == 0) { | |
909 | /* We can always add a page into an empty bio. */ | |
910 | contig = true; | |
911 | } else if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE) { | |
912 | struct bio_vec *bvec = bio_last_bvec_all(bio); | |
913 | ||
914 | /* | |
915 | * The contig check requires the following conditions to be met: | |
916 | * 1) The pages are belonging to the same inode | |
917 | * This is implied by the call chain. | |
918 | * | |
919 | * 2) The range has adjacent logical bytenr | |
920 | * | |
921 | * 3) The range has adjacent file offset | |
922 | * This is required for the usage of btrfs_bio->file_offset. | |
923 | */ | |
924 | if (bio_end_sector(bio) == sector && | |
925 | page_offset(bvec->bv_page) + bvec->bv_offset + | |
926 | bvec->bv_len == page_offset(page) + pg_offset) | |
927 | contig = true; | |
928 | } else { | |
929 | /* | |
930 | * For compression, all IO should have its logical bytenr | |
931 | * set to the starting bytenr of the compressed extent. | |
932 | */ | |
953651eb | 933 | contig = bio->bi_iter.bi_sector == sector; |
4a445b7b QW |
934 | } |
935 | ||
953651eb | 936 | if (!contig) |
e0eefe07 | 937 | return 0; |
953651eb | 938 | |
2380220e | 939 | real_size = min(bio_ctrl->len_to_oe_boundary - bio_size, size); |
e0eefe07 QW |
940 | |
941 | /* | |
942 | * If real_size is 0, never call bio_add_*_page(), as even size is 0, | |
943 | * bio will still execute its endio function on the page! | |
944 | */ | |
945 | if (real_size == 0) | |
946 | return 0; | |
953651eb | 947 | |
d5e4377d | 948 | return bio_add_page(bio, page, real_size, pg_offset); |
953651eb NA |
949 | } |
950 | ||
2380220e QW |
951 | static void calc_bio_boundaries(struct btrfs_bio_ctrl *bio_ctrl, |
952 | struct btrfs_inode *inode, u64 file_offset) | |
390ed29b | 953 | { |
390ed29b | 954 | struct btrfs_ordered_extent *ordered; |
390ed29b | 955 | u64 logical = (bio_ctrl->bio->bi_iter.bi_sector << SECTOR_SHIFT); |
390ed29b QW |
956 | |
957 | /* | |
2380220e QW |
958 | * Limit the extent to the ordered boundary for Zone Append. |
959 | * Compressed bios aren't submitted directly, so it doesn't apply to | |
960 | * them. | |
390ed29b | 961 | */ |
2380220e | 962 | if (bio_ctrl->compress_type == BTRFS_COMPRESS_NONE && |
d5e4377d | 963 | btrfs_use_zone_append(inode, logical)) { |
2380220e QW |
964 | ordered = btrfs_lookup_ordered_extent(inode, file_offset); |
965 | if (ordered) { | |
966 | bio_ctrl->len_to_oe_boundary = min_t(u32, U32_MAX, | |
967 | ordered->disk_bytenr + | |
968 | ordered->disk_num_bytes - logical); | |
969 | btrfs_put_ordered_extent(ordered); | |
970 | return; | |
971 | } | |
390ed29b QW |
972 | } |
973 | ||
2380220e | 974 | bio_ctrl->len_to_oe_boundary = U32_MAX; |
390ed29b QW |
975 | } |
976 | ||
d5e4377d CH |
977 | static void alloc_new_bio(struct btrfs_inode *inode, |
978 | struct btrfs_bio_ctrl *bio_ctrl, | |
979 | struct writeback_control *wbc, blk_opf_t opf, | |
980 | u64 disk_bytenr, u32 offset, u64 file_offset, | |
981 | enum btrfs_compression_type compress_type) | |
e0eefe07 QW |
982 | { |
983 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
984 | struct bio *bio; | |
e0eefe07 | 985 | |
d0e5cb2b CH |
986 | bio = btrfs_bio_alloc(BIO_MAX_VECS, opf, inode, bio_ctrl->end_io_func, |
987 | NULL); | |
e0eefe07 QW |
988 | /* |
989 | * For compressed page range, its disk_bytenr is always @disk_bytenr | |
990 | * passed in, no matter if we have added any range into previous bio. | |
991 | */ | |
cb3a12d9 | 992 | if (compress_type != BTRFS_COMPRESS_NONE) |
cd8e0cca | 993 | bio->bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT; |
e0eefe07 | 994 | else |
cd8e0cca | 995 | bio->bi_iter.bi_sector = (disk_bytenr + offset) >> SECTOR_SHIFT; |
e0eefe07 | 996 | bio_ctrl->bio = bio; |
0f07003b | 997 | bio_ctrl->compress_type = compress_type; |
2380220e | 998 | calc_bio_boundaries(bio_ctrl, inode, file_offset); |
e0eefe07 | 999 | |
50f1cff3 CH |
1000 | if (wbc) { |
1001 | /* | |
d5e4377d CH |
1002 | * Pick the last added device to support cgroup writeback. For |
1003 | * multi-device file systems this means blk-cgroup policies have | |
1004 | * to always be set on the last added/replaced device. | |
1005 | * This is a bit odd but has been like that for a long time. | |
50f1cff3 | 1006 | */ |
d5e4377d | 1007 | bio_set_dev(bio, fs_info->fs_devices->latest_dev->bdev); |
50f1cff3 | 1008 | wbc_init_bio(wbc, bio); |
e0eefe07 | 1009 | } |
e0eefe07 QW |
1010 | } |
1011 | ||
4b81ba48 DS |
1012 | /* |
1013 | * @opf: bio REQ_OP_* and REQ_* flags as one value | |
b8b3d625 | 1014 | * @wbc: optional writeback control for io accounting |
0c64c33c | 1015 | * @disk_bytenr: logical bytenr where the write will be |
209ecde5 | 1016 | * @page: page to add to the bio |
0c64c33c | 1017 | * @size: portion of page that we want to write to |
b8b3d625 DS |
1018 | * @pg_offset: offset of the new bio or to check whether we are adding |
1019 | * a contiguous page to the previous one | |
cb3a12d9 | 1020 | * @compress_type: compress type for current bio |
814b6f91 QW |
1021 | * |
1022 | * The will either add the page into the existing @bio_ctrl->bio, or allocate a | |
1023 | * new one in @bio_ctrl->bio. | |
1024 | * The mirror number for this IO should already be initizlied in | |
1025 | * @bio_ctrl->mirror_num. | |
4b81ba48 | 1026 | */ |
bf9486d6 | 1027 | static int submit_extent_page(blk_opf_t opf, |
da2f0f74 | 1028 | struct writeback_control *wbc, |
390ed29b | 1029 | struct btrfs_bio_ctrl *bio_ctrl, |
209ecde5 | 1030 | u64 disk_bytenr, struct page *page, |
6c5a4e2c | 1031 | size_t size, unsigned long pg_offset, |
cb3a12d9 | 1032 | enum btrfs_compression_type compress_type, |
005efedf | 1033 | bool force_bio_submit) |
d1310b2e | 1034 | { |
e1326f03 | 1035 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
e0eefe07 | 1036 | unsigned int cur = pg_offset; |
d1310b2e | 1037 | |
390ed29b | 1038 | ASSERT(bio_ctrl); |
5c2b1fd7 | 1039 | |
390ed29b QW |
1040 | ASSERT(pg_offset < PAGE_SIZE && size <= PAGE_SIZE && |
1041 | pg_offset + size <= PAGE_SIZE); | |
5467abba QW |
1042 | |
1043 | ASSERT(bio_ctrl->end_io_func); | |
1044 | ||
722c82ac CH |
1045 | if (force_bio_submit) |
1046 | submit_one_bio(bio_ctrl); | |
e0eefe07 QW |
1047 | |
1048 | while (cur < pg_offset + size) { | |
1049 | u32 offset = cur - pg_offset; | |
1050 | int added; | |
1051 | ||
1052 | /* Allocate new bio if needed */ | |
1053 | if (!bio_ctrl->bio) { | |
d5e4377d CH |
1054 | alloc_new_bio(inode, bio_ctrl, wbc, opf, disk_bytenr, |
1055 | offset, page_offset(page) + cur, | |
1056 | compress_type); | |
e0eefe07 QW |
1057 | } |
1058 | /* | |
1059 | * We must go through btrfs_bio_add_page() to ensure each | |
1060 | * page range won't cross various boundaries. | |
1061 | */ | |
cb3a12d9 | 1062 | if (compress_type != BTRFS_COMPRESS_NONE) |
e0eefe07 QW |
1063 | added = btrfs_bio_add_page(bio_ctrl, page, disk_bytenr, |
1064 | size - offset, pg_offset + offset, | |
cb3a12d9 | 1065 | compress_type); |
e0eefe07 QW |
1066 | else |
1067 | added = btrfs_bio_add_page(bio_ctrl, page, | |
1068 | disk_bytenr + offset, size - offset, | |
cb3a12d9 | 1069 | pg_offset + offset, compress_type); |
e0eefe07 QW |
1070 | |
1071 | /* Metadata page range should never be split */ | |
1072 | if (!is_data_inode(&inode->vfs_inode)) | |
1073 | ASSERT(added == 0 || added == size - offset); | |
1074 | ||
1075 | /* At least we added some page, update the account */ | |
1076 | if (wbc && added) | |
1077 | wbc_account_cgroup_owner(wbc, page, added); | |
1078 | ||
1079 | /* We have reached boundary, submit right now */ | |
1080 | if (added < size - offset) { | |
1081 | /* The bio should contain some page(s) */ | |
1082 | ASSERT(bio_ctrl->bio->bi_iter.bi_size); | |
722c82ac | 1083 | submit_one_bio(bio_ctrl); |
d1310b2e | 1084 | } |
e0eefe07 | 1085 | cur += added; |
d1310b2e | 1086 | } |
e0eefe07 | 1087 | return 0; |
d1310b2e CM |
1088 | } |
1089 | ||
760f991f QW |
1090 | static int attach_extent_buffer_page(struct extent_buffer *eb, |
1091 | struct page *page, | |
1092 | struct btrfs_subpage *prealloc) | |
d1310b2e | 1093 | { |
760f991f QW |
1094 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1095 | int ret = 0; | |
1096 | ||
0d01e247 QW |
1097 | /* |
1098 | * If the page is mapped to btree inode, we should hold the private | |
1099 | * lock to prevent race. | |
1100 | * For cloned or dummy extent buffers, their pages are not mapped and | |
1101 | * will not race with any other ebs. | |
1102 | */ | |
1103 | if (page->mapping) | |
1104 | lockdep_assert_held(&page->mapping->private_lock); | |
1105 | ||
fbca46eb | 1106 | if (fs_info->nodesize >= PAGE_SIZE) { |
760f991f QW |
1107 | if (!PagePrivate(page)) |
1108 | attach_page_private(page, eb); | |
1109 | else | |
1110 | WARN_ON(page->private != (unsigned long)eb); | |
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | /* Already mapped, just free prealloc */ | |
1115 | if (PagePrivate(page)) { | |
1116 | btrfs_free_subpage(prealloc); | |
1117 | return 0; | |
1118 | } | |
1119 | ||
1120 | if (prealloc) | |
1121 | /* Has preallocated memory for subpage */ | |
1122 | attach_page_private(page, prealloc); | |
d1b89bc0 | 1123 | else |
760f991f QW |
1124 | /* Do new allocation to attach subpage */ |
1125 | ret = btrfs_attach_subpage(fs_info, page, | |
1126 | BTRFS_SUBPAGE_METADATA); | |
1127 | return ret; | |
d1310b2e CM |
1128 | } |
1129 | ||
32443de3 | 1130 | int set_page_extent_mapped(struct page *page) |
d1310b2e | 1131 | { |
32443de3 QW |
1132 | struct btrfs_fs_info *fs_info; |
1133 | ||
1134 | ASSERT(page->mapping); | |
1135 | ||
1136 | if (PagePrivate(page)) | |
1137 | return 0; | |
1138 | ||
1139 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
1140 | ||
fbca46eb | 1141 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
1142 | return btrfs_attach_subpage(fs_info, page, BTRFS_SUBPAGE_DATA); |
1143 | ||
1144 | attach_page_private(page, (void *)EXTENT_PAGE_PRIVATE); | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | void clear_page_extent_mapped(struct page *page) | |
1149 | { | |
1150 | struct btrfs_fs_info *fs_info; | |
1151 | ||
1152 | ASSERT(page->mapping); | |
1153 | ||
d1b89bc0 | 1154 | if (!PagePrivate(page)) |
32443de3 QW |
1155 | return; |
1156 | ||
1157 | fs_info = btrfs_sb(page->mapping->host->i_sb); | |
fbca46eb | 1158 | if (btrfs_is_subpage(fs_info, page)) |
32443de3 QW |
1159 | return btrfs_detach_subpage(fs_info, page); |
1160 | ||
1161 | detach_page_private(page); | |
d1310b2e CM |
1162 | } |
1163 | ||
125bac01 MX |
1164 | static struct extent_map * |
1165 | __get_extent_map(struct inode *inode, struct page *page, size_t pg_offset, | |
1a5ee1e6 | 1166 | u64 start, u64 len, struct extent_map **em_cached) |
125bac01 MX |
1167 | { |
1168 | struct extent_map *em; | |
1169 | ||
1170 | if (em_cached && *em_cached) { | |
1171 | em = *em_cached; | |
cbc0e928 | 1172 | if (extent_map_in_tree(em) && start >= em->start && |
125bac01 | 1173 | start < extent_map_end(em)) { |
490b54d6 | 1174 | refcount_inc(&em->refs); |
125bac01 MX |
1175 | return em; |
1176 | } | |
1177 | ||
1178 | free_extent_map(em); | |
1179 | *em_cached = NULL; | |
1180 | } | |
1181 | ||
1a5ee1e6 | 1182 | em = btrfs_get_extent(BTRFS_I(inode), page, pg_offset, start, len); |
c0347550 | 1183 | if (em_cached && !IS_ERR(em)) { |
125bac01 | 1184 | BUG_ON(*em_cached); |
490b54d6 | 1185 | refcount_inc(&em->refs); |
125bac01 MX |
1186 | *em_cached = em; |
1187 | } | |
1188 | return em; | |
1189 | } | |
d1310b2e CM |
1190 | /* |
1191 | * basic readpage implementation. Locked extent state structs are inserted | |
1192 | * into the tree that are removed when the IO is done (by the end_io | |
1193 | * handlers) | |
79787eaa | 1194 | * XXX JDM: This needs looking at to ensure proper page locking |
baf863b9 | 1195 | * return 0 on success, otherwise return error |
d1310b2e | 1196 | */ |
7aab8b32 | 1197 | static int btrfs_do_readpage(struct page *page, struct extent_map **em_cached, |
390ed29b | 1198 | struct btrfs_bio_ctrl *bio_ctrl, |
bf9486d6 | 1199 | blk_opf_t read_flags, u64 *prev_em_start) |
d1310b2e CM |
1200 | { |
1201 | struct inode *inode = page->mapping->host; | |
92082d40 | 1202 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4eee4fa4 | 1203 | u64 start = page_offset(page); |
8eec8296 | 1204 | const u64 end = start + PAGE_SIZE - 1; |
d1310b2e CM |
1205 | u64 cur = start; |
1206 | u64 extent_offset; | |
1207 | u64 last_byte = i_size_read(inode); | |
1208 | u64 block_start; | |
d1310b2e | 1209 | struct extent_map *em; |
baf863b9 | 1210 | int ret = 0; |
306e16ce | 1211 | size_t pg_offset = 0; |
d1310b2e CM |
1212 | size_t iosize; |
1213 | size_t blocksize = inode->i_sb->s_blocksize; | |
f657a31c | 1214 | struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree; |
ae6957eb | 1215 | |
32443de3 QW |
1216 | ret = set_page_extent_mapped(page); |
1217 | if (ret < 0) { | |
570eb97b | 1218 | unlock_extent(tree, start, end, NULL); |
92082d40 QW |
1219 | btrfs_page_set_error(fs_info, page, start, PAGE_SIZE); |
1220 | unlock_page(page); | |
32443de3 QW |
1221 | goto out; |
1222 | } | |
d1310b2e | 1223 | |
09cbfeaf | 1224 | if (page->index == last_byte >> PAGE_SHIFT) { |
7073017a | 1225 | size_t zero_offset = offset_in_page(last_byte); |
c8b97818 CM |
1226 | |
1227 | if (zero_offset) { | |
09cbfeaf | 1228 | iosize = PAGE_SIZE - zero_offset; |
d048b9c2 | 1229 | memzero_page(page, zero_offset, iosize); |
c8b97818 CM |
1230 | } |
1231 | } | |
5467abba | 1232 | bio_ctrl->end_io_func = end_bio_extent_readpage; |
92082d40 | 1233 | begin_page_read(fs_info, page); |
d1310b2e | 1234 | while (cur <= end) { |
4c37a793 | 1235 | unsigned long this_bio_flag = 0; |
005efedf | 1236 | bool force_bio_submit = false; |
0c64c33c | 1237 | u64 disk_bytenr; |
c8f2f24b | 1238 | |
6a404910 | 1239 | ASSERT(IS_ALIGNED(cur, fs_info->sectorsize)); |
d1310b2e | 1240 | if (cur >= last_byte) { |
09cbfeaf | 1241 | iosize = PAGE_SIZE - pg_offset; |
d048b9c2 | 1242 | memzero_page(page, pg_offset, iosize); |
2c8f5e8c | 1243 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
92082d40 | 1244 | end_page_read(page, true, cur, iosize); |
d1310b2e CM |
1245 | break; |
1246 | } | |
125bac01 | 1247 | em = __get_extent_map(inode, page, pg_offset, cur, |
1a5ee1e6 | 1248 | end - cur + 1, em_cached); |
c0347550 | 1249 | if (IS_ERR(em)) { |
570eb97b | 1250 | unlock_extent(tree, cur, end, NULL); |
92082d40 | 1251 | end_page_read(page, false, cur, end + 1 - cur); |
bbf0ea7e | 1252 | ret = PTR_ERR(em); |
d1310b2e CM |
1253 | break; |
1254 | } | |
d1310b2e CM |
1255 | extent_offset = cur - em->start; |
1256 | BUG_ON(extent_map_end(em) <= cur); | |
1257 | BUG_ON(end < cur); | |
1258 | ||
7f6ca7f2 DS |
1259 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) |
1260 | this_bio_flag = em->compress_type; | |
c8b97818 | 1261 | |
d1310b2e | 1262 | iosize = min(extent_map_end(em) - cur, end - cur + 1); |
fda2832f | 1263 | iosize = ALIGN(iosize, blocksize); |
2a5232a8 | 1264 | if (this_bio_flag != BTRFS_COMPRESS_NONE) |
0c64c33c | 1265 | disk_bytenr = em->block_start; |
949b3273 | 1266 | else |
0c64c33c | 1267 | disk_bytenr = em->block_start + extent_offset; |
d1310b2e | 1268 | block_start = em->block_start; |
d899e052 YZ |
1269 | if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) |
1270 | block_start = EXTENT_MAP_HOLE; | |
005efedf FM |
1271 | |
1272 | /* | |
1273 | * If we have a file range that points to a compressed extent | |
260db43c | 1274 | * and it's followed by a consecutive file range that points |
005efedf FM |
1275 | * to the same compressed extent (possibly with a different |
1276 | * offset and/or length, so it either points to the whole extent | |
1277 | * or only part of it), we must make sure we do not submit a | |
1278 | * single bio to populate the pages for the 2 ranges because | |
1279 | * this makes the compressed extent read zero out the pages | |
1280 | * belonging to the 2nd range. Imagine the following scenario: | |
1281 | * | |
1282 | * File layout | |
1283 | * [0 - 8K] [8K - 24K] | |
1284 | * | | | |
1285 | * | | | |
1286 | * points to extent X, points to extent X, | |
1287 | * offset 4K, length of 8K offset 0, length 16K | |
1288 | * | |
1289 | * [extent X, compressed length = 4K uncompressed length = 16K] | |
1290 | * | |
1291 | * If the bio to read the compressed extent covers both ranges, | |
1292 | * it will decompress extent X into the pages belonging to the | |
1293 | * first range and then it will stop, zeroing out the remaining | |
1294 | * pages that belong to the other range that points to extent X. | |
1295 | * So here we make sure we submit 2 bios, one for the first | |
1296 | * range and another one for the third range. Both will target | |
1297 | * the same physical extent from disk, but we can't currently | |
1298 | * make the compressed bio endio callback populate the pages | |
1299 | * for both ranges because each compressed bio is tightly | |
1300 | * coupled with a single extent map, and each range can have | |
1301 | * an extent map with a different offset value relative to the | |
1302 | * uncompressed data of our extent and different lengths. This | |
1303 | * is a corner case so we prioritize correctness over | |
1304 | * non-optimal behavior (submitting 2 bios for the same extent). | |
1305 | */ | |
1306 | if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) && | |
1307 | prev_em_start && *prev_em_start != (u64)-1 && | |
8e928218 | 1308 | *prev_em_start != em->start) |
005efedf FM |
1309 | force_bio_submit = true; |
1310 | ||
1311 | if (prev_em_start) | |
8e928218 | 1312 | *prev_em_start = em->start; |
005efedf | 1313 | |
d1310b2e CM |
1314 | free_extent_map(em); |
1315 | em = NULL; | |
1316 | ||
1317 | /* we've found a hole, just zero and go on */ | |
1318 | if (block_start == EXTENT_MAP_HOLE) { | |
d048b9c2 | 1319 | memzero_page(page, pg_offset, iosize); |
d1310b2e | 1320 | |
2c8f5e8c | 1321 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
92082d40 | 1322 | end_page_read(page, true, cur, iosize); |
d1310b2e | 1323 | cur = cur + iosize; |
306e16ce | 1324 | pg_offset += iosize; |
d1310b2e CM |
1325 | continue; |
1326 | } | |
1327 | /* the get_extent function already copied into the page */ | |
70dec807 | 1328 | if (block_start == EXTENT_MAP_INLINE) { |
570eb97b | 1329 | unlock_extent(tree, cur, cur + iosize - 1, NULL); |
52b029f4 | 1330 | end_page_read(page, true, cur, iosize); |
70dec807 | 1331 | cur = cur + iosize; |
306e16ce | 1332 | pg_offset += iosize; |
70dec807 CM |
1333 | continue; |
1334 | } | |
d1310b2e | 1335 | |
0ceb34bf | 1336 | ret = submit_extent_page(REQ_OP_READ | read_flags, NULL, |
209ecde5 | 1337 | bio_ctrl, disk_bytenr, page, iosize, |
5467abba QW |
1338 | pg_offset, this_bio_flag, |
1339 | force_bio_submit); | |
ad3fc794 | 1340 | if (ret) { |
10f7f6f8 QW |
1341 | /* |
1342 | * We have to unlock the remaining range, or the page | |
1343 | * will never be unlocked. | |
1344 | */ | |
570eb97b | 1345 | unlock_extent(tree, cur, end, NULL); |
10f7f6f8 | 1346 | end_page_read(page, false, cur, end + 1 - cur); |
baf863b9 | 1347 | goto out; |
edd33c99 | 1348 | } |
d1310b2e | 1349 | cur = cur + iosize; |
306e16ce | 1350 | pg_offset += iosize; |
d1310b2e | 1351 | } |
90a887c9 | 1352 | out: |
baf863b9 | 1353 | return ret; |
d1310b2e CM |
1354 | } |
1355 | ||
fdaf9a58 | 1356 | int btrfs_read_folio(struct file *file, struct folio *folio) |
7aab8b32 | 1357 | { |
fdaf9a58 | 1358 | struct page *page = &folio->page; |
7aab8b32 CH |
1359 | struct btrfs_inode *inode = BTRFS_I(page->mapping->host); |
1360 | u64 start = page_offset(page); | |
1361 | u64 end = start + PAGE_SIZE - 1; | |
1362 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; | |
1363 | int ret; | |
1364 | ||
1365 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); | |
1366 | ||
1367 | ret = btrfs_do_readpage(page, NULL, &bio_ctrl, 0, NULL); | |
1368 | /* | |
1369 | * If btrfs_do_readpage() failed we will want to submit the assembled | |
1370 | * bio to do the cleanup. | |
1371 | */ | |
722c82ac | 1372 | submit_one_bio(&bio_ctrl); |
7aab8b32 CH |
1373 | return ret; |
1374 | } | |
1375 | ||
b6660e80 | 1376 | static inline void contiguous_readpages(struct page *pages[], int nr_pages, |
390ed29b QW |
1377 | u64 start, u64 end, |
1378 | struct extent_map **em_cached, | |
1379 | struct btrfs_bio_ctrl *bio_ctrl, | |
1380 | u64 *prev_em_start) | |
9974090b | 1381 | { |
23d31bd4 | 1382 | struct btrfs_inode *inode = BTRFS_I(pages[0]->mapping->host); |
9974090b MX |
1383 | int index; |
1384 | ||
b272ae22 | 1385 | btrfs_lock_and_flush_ordered_range(inode, start, end, NULL); |
9974090b MX |
1386 | |
1387 | for (index = 0; index < nr_pages; index++) { | |
390ed29b | 1388 | btrfs_do_readpage(pages[index], em_cached, bio_ctrl, |
0f208812 | 1389 | REQ_RAHEAD, prev_em_start); |
09cbfeaf | 1390 | put_page(pages[index]); |
9974090b MX |
1391 | } |
1392 | } | |
1393 | ||
d1310b2e | 1394 | /* |
40f76580 CM |
1395 | * helper for __extent_writepage, doing all of the delayed allocation setup. |
1396 | * | |
5eaad97a | 1397 | * This returns 1 if btrfs_run_delalloc_range function did all the work required |
40f76580 CM |
1398 | * to write the page (copy into inline extent). In this case the IO has |
1399 | * been started and the page is already unlocked. | |
1400 | * | |
1401 | * This returns 0 if all went well (page still locked) | |
1402 | * This returns < 0 if there were errors (page still locked) | |
d1310b2e | 1403 | */ |
cd4c0bf9 | 1404 | static noinline_for_stack int writepage_delalloc(struct btrfs_inode *inode, |
83f1b680 | 1405 | struct page *page, struct writeback_control *wbc) |
40f76580 | 1406 | { |
2749f7ef | 1407 | const u64 page_end = page_offset(page) + PAGE_SIZE - 1; |
cf3075fb | 1408 | u64 delalloc_start = page_offset(page); |
40f76580 | 1409 | u64 delalloc_to_write = 0; |
83f1b680 QW |
1410 | /* How many pages are started by btrfs_run_delalloc_range() */ |
1411 | unsigned long nr_written = 0; | |
40f76580 CM |
1412 | int ret; |
1413 | int page_started = 0; | |
1414 | ||
2749f7ef QW |
1415 | while (delalloc_start < page_end) { |
1416 | u64 delalloc_end = page_end; | |
1417 | bool found; | |
40f76580 | 1418 | |
cd4c0bf9 | 1419 | found = find_lock_delalloc_range(&inode->vfs_inode, page, |
40f76580 | 1420 | &delalloc_start, |
917aacec | 1421 | &delalloc_end); |
3522e903 | 1422 | if (!found) { |
40f76580 CM |
1423 | delalloc_start = delalloc_end + 1; |
1424 | continue; | |
1425 | } | |
cd4c0bf9 | 1426 | ret = btrfs_run_delalloc_range(inode, page, delalloc_start, |
83f1b680 | 1427 | delalloc_end, &page_started, &nr_written, wbc); |
40f76580 | 1428 | if (ret) { |
963e4db8 QW |
1429 | btrfs_page_set_error(inode->root->fs_info, page, |
1430 | page_offset(page), PAGE_SIZE); | |
7361b4ae | 1431 | return ret; |
40f76580 CM |
1432 | } |
1433 | /* | |
ea1754a0 KS |
1434 | * delalloc_end is already one less than the total length, so |
1435 | * we don't subtract one from PAGE_SIZE | |
40f76580 CM |
1436 | */ |
1437 | delalloc_to_write += (delalloc_end - delalloc_start + | |
ea1754a0 | 1438 | PAGE_SIZE) >> PAGE_SHIFT; |
40f76580 CM |
1439 | delalloc_start = delalloc_end + 1; |
1440 | } | |
1441 | if (wbc->nr_to_write < delalloc_to_write) { | |
1442 | int thresh = 8192; | |
1443 | ||
1444 | if (delalloc_to_write < thresh * 2) | |
1445 | thresh = delalloc_to_write; | |
1446 | wbc->nr_to_write = min_t(u64, delalloc_to_write, | |
1447 | thresh); | |
1448 | } | |
1449 | ||
83f1b680 | 1450 | /* Did btrfs_run_dealloc_range() already unlock and start the IO? */ |
40f76580 CM |
1451 | if (page_started) { |
1452 | /* | |
83f1b680 QW |
1453 | * We've unlocked the page, so we can't update the mapping's |
1454 | * writeback index, just update nr_to_write. | |
40f76580 | 1455 | */ |
83f1b680 | 1456 | wbc->nr_to_write -= nr_written; |
40f76580 CM |
1457 | return 1; |
1458 | } | |
1459 | ||
b69d1ee9 | 1460 | return 0; |
40f76580 CM |
1461 | } |
1462 | ||
c5ef5c6c QW |
1463 | /* |
1464 | * Find the first byte we need to write. | |
1465 | * | |
1466 | * For subpage, one page can contain several sectors, and | |
1467 | * __extent_writepage_io() will just grab all extent maps in the page | |
1468 | * range and try to submit all non-inline/non-compressed extents. | |
1469 | * | |
1470 | * This is a big problem for subpage, we shouldn't re-submit already written | |
1471 | * data at all. | |
1472 | * This function will lookup subpage dirty bit to find which range we really | |
1473 | * need to submit. | |
1474 | * | |
1475 | * Return the next dirty range in [@start, @end). | |
1476 | * If no dirty range is found, @start will be page_offset(page) + PAGE_SIZE. | |
1477 | */ | |
1478 | static void find_next_dirty_byte(struct btrfs_fs_info *fs_info, | |
1479 | struct page *page, u64 *start, u64 *end) | |
1480 | { | |
1481 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; | |
72a69cd0 | 1482 | struct btrfs_subpage_info *spi = fs_info->subpage_info; |
c5ef5c6c QW |
1483 | u64 orig_start = *start; |
1484 | /* Declare as unsigned long so we can use bitmap ops */ | |
c5ef5c6c | 1485 | unsigned long flags; |
72a69cd0 | 1486 | int range_start_bit; |
c5ef5c6c QW |
1487 | int range_end_bit; |
1488 | ||
1489 | /* | |
1490 | * For regular sector size == page size case, since one page only | |
1491 | * contains one sector, we return the page offset directly. | |
1492 | */ | |
fbca46eb | 1493 | if (!btrfs_is_subpage(fs_info, page)) { |
c5ef5c6c QW |
1494 | *start = page_offset(page); |
1495 | *end = page_offset(page) + PAGE_SIZE; | |
1496 | return; | |
1497 | } | |
1498 | ||
72a69cd0 QW |
1499 | range_start_bit = spi->dirty_offset + |
1500 | (offset_in_page(orig_start) >> fs_info->sectorsize_bits); | |
1501 | ||
c5ef5c6c QW |
1502 | /* We should have the page locked, but just in case */ |
1503 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
1504 | bitmap_next_set_region(subpage->bitmaps, &range_start_bit, &range_end_bit, |
1505 | spi->dirty_offset + spi->bitmap_nr_bits); | |
c5ef5c6c QW |
1506 | spin_unlock_irqrestore(&subpage->lock, flags); |
1507 | ||
72a69cd0 QW |
1508 | range_start_bit -= spi->dirty_offset; |
1509 | range_end_bit -= spi->dirty_offset; | |
1510 | ||
c5ef5c6c QW |
1511 | *start = page_offset(page) + range_start_bit * fs_info->sectorsize; |
1512 | *end = page_offset(page) + range_end_bit * fs_info->sectorsize; | |
1513 | } | |
1514 | ||
40f76580 CM |
1515 | /* |
1516 | * helper for __extent_writepage. This calls the writepage start hooks, | |
1517 | * and does the loop to map the page into extents and bios. | |
1518 | * | |
1519 | * We return 1 if the IO is started and the page is unlocked, | |
1520 | * 0 if all went well (page still locked) | |
1521 | * < 0 if there were errors (page still locked) | |
1522 | */ | |
d4580fe2 | 1523 | static noinline_for_stack int __extent_writepage_io(struct btrfs_inode *inode, |
40f76580 CM |
1524 | struct page *page, |
1525 | struct writeback_control *wbc, | |
ee5f017d | 1526 | struct btrfs_bio_ctrl *bio_ctrl, |
40f76580 | 1527 | loff_t i_size, |
57e5ffeb | 1528 | int *nr_ret) |
d1310b2e | 1529 | { |
6bc5636a | 1530 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
a129ffb8 QW |
1531 | u64 cur = page_offset(page); |
1532 | u64 end = cur + PAGE_SIZE - 1; | |
d1310b2e | 1533 | u64 extent_offset; |
d1310b2e | 1534 | u64 block_start; |
d1310b2e | 1535 | struct extent_map *em; |
44e5801f | 1536 | int saved_ret = 0; |
40f76580 CM |
1537 | int ret = 0; |
1538 | int nr = 0; | |
bf9486d6 BVA |
1539 | enum req_op op = REQ_OP_WRITE; |
1540 | const blk_opf_t write_flags = wbc_to_write_flags(wbc); | |
44e5801f | 1541 | bool has_error = false; |
40f76580 | 1542 | bool compressed; |
c8b97818 | 1543 | |
a129ffb8 | 1544 | ret = btrfs_writepage_cow_fixup(page); |
d75855b4 NB |
1545 | if (ret) { |
1546 | /* Fixup worker will requeue */ | |
5ab58055 | 1547 | redirty_page_for_writepage(wbc, page); |
d75855b4 NB |
1548 | unlock_page(page); |
1549 | return 1; | |
247e743c CM |
1550 | } |
1551 | ||
11c8349b CM |
1552 | /* |
1553 | * we don't want to touch the inode after unlocking the page, | |
1554 | * so we update the mapping writeback index now | |
1555 | */ | |
572f3dad | 1556 | wbc->nr_to_write--; |
771ed689 | 1557 | |
ee5f017d | 1558 | bio_ctrl->end_io_func = end_bio_extent_writepage; |
d1310b2e | 1559 | while (cur <= end) { |
0c64c33c | 1560 | u64 disk_bytenr; |
40f76580 | 1561 | u64 em_end; |
c5ef5c6c QW |
1562 | u64 dirty_range_start = cur; |
1563 | u64 dirty_range_end; | |
6bc5636a | 1564 | u32 iosize; |
58409edd | 1565 | |
40f76580 | 1566 | if (cur >= i_size) { |
38a39ac7 | 1567 | btrfs_writepage_endio_finish_ordered(inode, page, cur, |
25c1252a | 1568 | end, true); |
cc1d0d93 QW |
1569 | /* |
1570 | * This range is beyond i_size, thus we don't need to | |
1571 | * bother writing back. | |
1572 | * But we still need to clear the dirty subpage bit, or | |
1573 | * the next time the page gets dirtied, we will try to | |
1574 | * writeback the sectors with subpage dirty bits, | |
1575 | * causing writeback without ordered extent. | |
1576 | */ | |
1577 | btrfs_page_clear_dirty(fs_info, page, cur, end + 1 - cur); | |
d1310b2e CM |
1578 | break; |
1579 | } | |
c5ef5c6c QW |
1580 | |
1581 | find_next_dirty_byte(fs_info, page, &dirty_range_start, | |
1582 | &dirty_range_end); | |
1583 | if (cur < dirty_range_start) { | |
1584 | cur = dirty_range_start; | |
1585 | continue; | |
1586 | } | |
1587 | ||
d4580fe2 | 1588 | em = btrfs_get_extent(inode, NULL, 0, cur, end - cur + 1); |
c0347550 | 1589 | if (IS_ERR(em)) { |
c5ef5c6c | 1590 | btrfs_page_set_error(fs_info, page, cur, end - cur + 1); |
61391d56 | 1591 | ret = PTR_ERR_OR_ZERO(em); |
44e5801f QW |
1592 | has_error = true; |
1593 | if (!saved_ret) | |
1594 | saved_ret = ret; | |
d1310b2e CM |
1595 | break; |
1596 | } | |
1597 | ||
1598 | extent_offset = cur - em->start; | |
40f76580 | 1599 | em_end = extent_map_end(em); |
6bc5636a QW |
1600 | ASSERT(cur <= em_end); |
1601 | ASSERT(cur < end); | |
1602 | ASSERT(IS_ALIGNED(em->start, fs_info->sectorsize)); | |
1603 | ASSERT(IS_ALIGNED(em->len, fs_info->sectorsize)); | |
d1310b2e | 1604 | block_start = em->block_start; |
c8b97818 | 1605 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
6bc5636a QW |
1606 | disk_bytenr = em->block_start + extent_offset; |
1607 | ||
c5ef5c6c QW |
1608 | /* |
1609 | * Note that em_end from extent_map_end() and dirty_range_end from | |
1610 | * find_next_dirty_byte() are all exclusive | |
1611 | */ | |
1612 | iosize = min(min(em_end, end + 1), dirty_range_end) - cur; | |
d1310b2e CM |
1613 | free_extent_map(em); |
1614 | em = NULL; | |
1615 | ||
c8b97818 CM |
1616 | /* |
1617 | * compressed and inline extents are written through other | |
1618 | * paths in the FS | |
1619 | */ | |
1620 | if (compressed || block_start == EXTENT_MAP_HOLE || | |
d1310b2e | 1621 | block_start == EXTENT_MAP_INLINE) { |
c8b04030 | 1622 | if (compressed) |
c8b97818 | 1623 | nr++; |
c8b04030 | 1624 | else |
38a39ac7 | 1625 | btrfs_writepage_endio_finish_ordered(inode, |
25c1252a | 1626 | page, cur, cur + iosize - 1, true); |
cc1d0d93 | 1627 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); |
c8b97818 | 1628 | cur += iosize; |
d1310b2e CM |
1629 | continue; |
1630 | } | |
c8b97818 | 1631 | |
d2a91064 | 1632 | btrfs_set_range_writeback(inode, cur, cur + iosize - 1); |
58409edd | 1633 | if (!PageWriteback(page)) { |
d4580fe2 | 1634 | btrfs_err(inode->root->fs_info, |
58409edd DS |
1635 | "page %lu not writeback, cur %llu end %llu", |
1636 | page->index, cur, end); | |
d1310b2e | 1637 | } |
7f3c74fb | 1638 | |
c5ef5c6c QW |
1639 | /* |
1640 | * Although the PageDirty bit is cleared before entering this | |
1641 | * function, subpage dirty bit is not cleared. | |
1642 | * So clear subpage dirty bit here so next time we won't submit | |
1643 | * page for range already written to disk. | |
1644 | */ | |
1645 | btrfs_page_clear_dirty(fs_info, page, cur, iosize); | |
1646 | ||
bf9486d6 | 1647 | ret = submit_extent_page(op | write_flags, wbc, |
ee5f017d | 1648 | bio_ctrl, disk_bytenr, |
209ecde5 | 1649 | page, iosize, |
390ed29b | 1650 | cur - page_offset(page), |
722c82ac | 1651 | 0, false); |
fe01aa65 | 1652 | if (ret) { |
44e5801f QW |
1653 | has_error = true; |
1654 | if (!saved_ret) | |
1655 | saved_ret = ret; | |
1656 | ||
c5ef5c6c | 1657 | btrfs_page_set_error(fs_info, page, cur, iosize); |
fe01aa65 | 1658 | if (PageWriteback(page)) |
c5ef5c6c QW |
1659 | btrfs_page_clear_writeback(fs_info, page, cur, |
1660 | iosize); | |
fe01aa65 | 1661 | } |
d1310b2e | 1662 | |
6bc5636a | 1663 | cur += iosize; |
d1310b2e CM |
1664 | nr++; |
1665 | } | |
cc1d0d93 QW |
1666 | /* |
1667 | * If we finish without problem, we should not only clear page dirty, | |
1668 | * but also empty subpage dirty bits | |
1669 | */ | |
44e5801f | 1670 | if (!has_error) |
cc1d0d93 | 1671 | btrfs_page_assert_not_dirty(fs_info, page); |
44e5801f QW |
1672 | else |
1673 | ret = saved_ret; | |
40f76580 | 1674 | *nr_ret = nr; |
40f76580 CM |
1675 | return ret; |
1676 | } | |
1677 | ||
1678 | /* | |
1679 | * the writepage semantics are similar to regular writepage. extent | |
1680 | * records are inserted to lock ranges in the tree, and as dirty areas | |
1681 | * are found, they are marked writeback. Then the lock bits are removed | |
1682 | * and the end_io handler clears the writeback ranges | |
3065976b QW |
1683 | * |
1684 | * Return 0 if everything goes well. | |
1685 | * Return <0 for error. | |
40f76580 CM |
1686 | */ |
1687 | static int __extent_writepage(struct page *page, struct writeback_control *wbc, | |
ee5f017d | 1688 | struct btrfs_bio_ctrl *bio_ctrl) |
40f76580 | 1689 | { |
8e1dec8e | 1690 | struct folio *folio = page_folio(page); |
40f76580 | 1691 | struct inode *inode = page->mapping->host; |
e55a0de1 | 1692 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
cf3075fb QW |
1693 | const u64 page_start = page_offset(page); |
1694 | const u64 page_end = page_start + PAGE_SIZE - 1; | |
40f76580 CM |
1695 | int ret; |
1696 | int nr = 0; | |
eb70d222 | 1697 | size_t pg_offset; |
40f76580 | 1698 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 1699 | unsigned long end_index = i_size >> PAGE_SHIFT; |
40f76580 | 1700 | |
40f76580 CM |
1701 | trace___extent_writepage(page, inode, wbc); |
1702 | ||
1703 | WARN_ON(!PageLocked(page)); | |
1704 | ||
963e4db8 QW |
1705 | btrfs_page_clear_error(btrfs_sb(inode->i_sb), page, |
1706 | page_offset(page), PAGE_SIZE); | |
40f76580 | 1707 | |
7073017a | 1708 | pg_offset = offset_in_page(i_size); |
40f76580 CM |
1709 | if (page->index > end_index || |
1710 | (page->index == end_index && !pg_offset)) { | |
8e1dec8e MWO |
1711 | folio_invalidate(folio, 0, folio_size(folio)); |
1712 | folio_unlock(folio); | |
40f76580 CM |
1713 | return 0; |
1714 | } | |
1715 | ||
21a8935e | 1716 | if (page->index == end_index) |
d048b9c2 | 1717 | memzero_page(page, pg_offset, PAGE_SIZE - pg_offset); |
40f76580 | 1718 | |
32443de3 QW |
1719 | ret = set_page_extent_mapped(page); |
1720 | if (ret < 0) { | |
1721 | SetPageError(page); | |
1722 | goto done; | |
1723 | } | |
40f76580 | 1724 | |
ee5f017d | 1725 | if (!bio_ctrl->extent_locked) { |
83f1b680 | 1726 | ret = writepage_delalloc(BTRFS_I(inode), page, wbc); |
7789a55a | 1727 | if (ret == 1) |
169d2c87 | 1728 | return 0; |
7789a55a NB |
1729 | if (ret) |
1730 | goto done; | |
1731 | } | |
40f76580 | 1732 | |
ee5f017d | 1733 | ret = __extent_writepage_io(BTRFS_I(inode), page, wbc, bio_ctrl, i_size, |
83f1b680 | 1734 | &nr); |
40f76580 | 1735 | if (ret == 1) |
169d2c87 | 1736 | return 0; |
40f76580 | 1737 | |
d1310b2e CM |
1738 | done: |
1739 | if (nr == 0) { | |
1740 | /* make sure the mapping tag for page dirty gets cleared */ | |
1741 | set_page_writeback(page); | |
1742 | end_page_writeback(page); | |
1743 | } | |
963e4db8 QW |
1744 | /* |
1745 | * Here we used to have a check for PageError() and then set @ret and | |
1746 | * call end_extent_writepage(). | |
1747 | * | |
1748 | * But in fact setting @ret here will cause different error paths | |
1749 | * between subpage and regular sectorsize. | |
1750 | * | |
1751 | * For regular page size, we never submit current page, but only add | |
1752 | * current page to current bio. | |
1753 | * The bio submission can only happen in next page. | |
1754 | * Thus if we hit the PageError() branch, @ret is already set to | |
1755 | * non-zero value and will not get updated for regular sectorsize. | |
1756 | * | |
1757 | * But for subpage case, it's possible we submit part of current page, | |
1758 | * thus can get PageError() set by submitted bio of the same page, | |
1759 | * while our @ret is still 0. | |
1760 | * | |
1761 | * So here we unify the behavior and don't set @ret. | |
1762 | * Error can still be properly passed to higher layer as page will | |
1763 | * be set error, here we just don't handle the IO failure. | |
1764 | * | |
1765 | * NOTE: This is just a hotfix for subpage. | |
1766 | * The root fix will be properly ending ordered extent when we hit | |
1767 | * an error during writeback. | |
1768 | * | |
1769 | * But that needs a bigger refactoring, as we not only need to grab the | |
1770 | * submitted OE, but also need to know exactly at which bytenr we hit | |
1771 | * the error. | |
1772 | * Currently the full page based __extent_writepage_io() is not | |
1773 | * capable of that. | |
1774 | */ | |
1775 | if (PageError(page)) | |
cf3075fb | 1776 | end_extent_writepage(page, ret, page_start, page_end); |
ee5f017d | 1777 | if (bio_ctrl->extent_locked) { |
e55a0de1 | 1778 | /* |
ee5f017d | 1779 | * If bio_ctrl->extent_locked, it's from extent_write_locked_range(), |
e55a0de1 QW |
1780 | * the page can either be locked by lock_page() or |
1781 | * process_one_page(). | |
1782 | * Let btrfs_page_unlock_writer() handle both cases. | |
1783 | */ | |
1784 | ASSERT(wbc); | |
1785 | btrfs_page_unlock_writer(fs_info, page, wbc->range_start, | |
1786 | wbc->range_end + 1 - wbc->range_start); | |
1787 | } else { | |
1788 | unlock_page(page); | |
1789 | } | |
3065976b | 1790 | ASSERT(ret <= 0); |
40f76580 | 1791 | return ret; |
d1310b2e CM |
1792 | } |
1793 | ||
fd8b2b61 | 1794 | void wait_on_extent_buffer_writeback(struct extent_buffer *eb) |
0b32f4bb | 1795 | { |
74316201 N |
1796 | wait_on_bit_io(&eb->bflags, EXTENT_BUFFER_WRITEBACK, |
1797 | TASK_UNINTERRUPTIBLE); | |
0b32f4bb JB |
1798 | } |
1799 | ||
18dfa711 FM |
1800 | static void end_extent_buffer_writeback(struct extent_buffer *eb) |
1801 | { | |
1802 | clear_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
1803 | smp_mb__after_atomic(); | |
1804 | wake_up_bit(&eb->bflags, EXTENT_BUFFER_WRITEBACK); | |
1805 | } | |
1806 | ||
2e3c2513 | 1807 | /* |
a3efb2f0 | 1808 | * Lock extent buffer status and pages for writeback. |
2e3c2513 | 1809 | * |
a3efb2f0 QW |
1810 | * May try to flush write bio if we can't get the lock. |
1811 | * | |
1812 | * Return 0 if the extent buffer doesn't need to be submitted. | |
1813 | * (E.g. the extent buffer is not dirty) | |
1814 | * Return >0 is the extent buffer is submitted to bio. | |
1815 | * Return <0 if something went wrong, no page is locked. | |
2e3c2513 | 1816 | */ |
9df76fb5 | 1817 | static noinline_for_stack int lock_extent_buffer_for_io(struct extent_buffer *eb, |
ee5f017d | 1818 | struct btrfs_bio_ctrl *bio_ctrl) |
0b32f4bb | 1819 | { |
9df76fb5 | 1820 | struct btrfs_fs_info *fs_info = eb->fs_info; |
c9583ada | 1821 | int i, num_pages; |
0b32f4bb JB |
1822 | int flush = 0; |
1823 | int ret = 0; | |
1824 | ||
1825 | if (!btrfs_try_tree_write_lock(eb)) { | |
ee5f017d | 1826 | submit_write_bio(bio_ctrl, 0); |
2e3c2513 | 1827 | flush = 1; |
0b32f4bb JB |
1828 | btrfs_tree_lock(eb); |
1829 | } | |
1830 | ||
1831 | if (test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) { | |
1832 | btrfs_tree_unlock(eb); | |
ee5f017d | 1833 | if (!bio_ctrl->sync_io) |
0b32f4bb JB |
1834 | return 0; |
1835 | if (!flush) { | |
ee5f017d | 1836 | submit_write_bio(bio_ctrl, 0); |
0b32f4bb JB |
1837 | flush = 1; |
1838 | } | |
a098d8e8 CM |
1839 | while (1) { |
1840 | wait_on_extent_buffer_writeback(eb); | |
1841 | btrfs_tree_lock(eb); | |
1842 | if (!test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)) | |
1843 | break; | |
0b32f4bb | 1844 | btrfs_tree_unlock(eb); |
0b32f4bb JB |
1845 | } |
1846 | } | |
1847 | ||
51561ffe JB |
1848 | /* |
1849 | * We need to do this to prevent races in people who check if the eb is | |
1850 | * under IO since we can end up having no IO bits set for a short period | |
1851 | * of time. | |
1852 | */ | |
1853 | spin_lock(&eb->refs_lock); | |
0b32f4bb JB |
1854 | if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) { |
1855 | set_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags); | |
51561ffe | 1856 | spin_unlock(&eb->refs_lock); |
0b32f4bb | 1857 | btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN); |
104b4e51 NB |
1858 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, |
1859 | -eb->len, | |
1860 | fs_info->dirty_metadata_batch); | |
0b32f4bb | 1861 | ret = 1; |
51561ffe JB |
1862 | } else { |
1863 | spin_unlock(&eb->refs_lock); | |
0b32f4bb JB |
1864 | } |
1865 | ||
1866 | btrfs_tree_unlock(eb); | |
1867 | ||
f3156df9 QW |
1868 | /* |
1869 | * Either we don't need to submit any tree block, or we're submitting | |
1870 | * subpage eb. | |
1871 | * Subpage metadata doesn't use page locking at all, so we can skip | |
1872 | * the page locking. | |
1873 | */ | |
fbca46eb | 1874 | if (!ret || fs_info->nodesize < PAGE_SIZE) |
0b32f4bb JB |
1875 | return ret; |
1876 | ||
65ad0104 | 1877 | num_pages = num_extent_pages(eb); |
0b32f4bb | 1878 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 1879 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
1880 | |
1881 | if (!trylock_page(p)) { | |
1882 | if (!flush) { | |
ee5f017d | 1883 | submit_write_bio(bio_ctrl, 0); |
0b32f4bb JB |
1884 | flush = 1; |
1885 | } | |
1886 | lock_page(p); | |
1887 | } | |
1888 | } | |
1889 | ||
2e3c2513 | 1890 | return ret; |
0b32f4bb JB |
1891 | } |
1892 | ||
5a2c6075 | 1893 | static void set_btree_ioerr(struct page *page, struct extent_buffer *eb) |
656f30db | 1894 | { |
5a2c6075 | 1895 | struct btrfs_fs_info *fs_info = eb->fs_info; |
656f30db | 1896 | |
5a2c6075 | 1897 | btrfs_page_set_error(fs_info, page, eb->start, eb->len); |
656f30db FM |
1898 | if (test_and_set_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) |
1899 | return; | |
1900 | ||
c2e39305 JB |
1901 | /* |
1902 | * A read may stumble upon this buffer later, make sure that it gets an | |
1903 | * error and knows there was an error. | |
1904 | */ | |
1905 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
1906 | ||
68b85589 JB |
1907 | /* |
1908 | * We need to set the mapping with the io error as well because a write | |
1909 | * error will flip the file system readonly, and then syncfs() will | |
1910 | * return a 0 because we are readonly if we don't modify the err seq for | |
1911 | * the superblock. | |
1912 | */ | |
1913 | mapping_set_error(page->mapping, -EIO); | |
1914 | ||
656f30db FM |
1915 | /* |
1916 | * If writeback for a btree extent that doesn't belong to a log tree | |
1917 | * failed, increment the counter transaction->eb_write_errors. | |
1918 | * We do this because while the transaction is running and before it's | |
1919 | * committing (when we call filemap_fdata[write|wait]_range against | |
1920 | * the btree inode), we might have | |
1921 | * btree_inode->i_mapping->a_ops->writepages() called by the VM - if it | |
1922 | * returns an error or an error happens during writeback, when we're | |
1923 | * committing the transaction we wouldn't know about it, since the pages | |
1924 | * can be no longer dirty nor marked anymore for writeback (if a | |
1925 | * subsequent modification to the extent buffer didn't happen before the | |
1926 | * transaction commit), which makes filemap_fdata[write|wait]_range not | |
1927 | * able to find the pages tagged with SetPageError at transaction | |
1928 | * commit time. So if this happens we must abort the transaction, | |
1929 | * otherwise we commit a super block with btree roots that point to | |
1930 | * btree nodes/leafs whose content on disk is invalid - either garbage | |
1931 | * or the content of some node/leaf from a past generation that got | |
1932 | * cowed or deleted and is no longer valid. | |
1933 | * | |
1934 | * Note: setting AS_EIO/AS_ENOSPC in the btree inode's i_mapping would | |
1935 | * not be enough - we need to distinguish between log tree extents vs | |
1936 | * non-log tree extents, and the next filemap_fdatawait_range() call | |
1937 | * will catch and clear such errors in the mapping - and that call might | |
1938 | * be from a log sync and not from a transaction commit. Also, checking | |
1939 | * for the eb flag EXTENT_BUFFER_WRITE_ERR at transaction commit time is | |
1940 | * not done and would not be reliable - the eb might have been released | |
1941 | * from memory and reading it back again means that flag would not be | |
1942 | * set (since it's a runtime flag, not persisted on disk). | |
1943 | * | |
1944 | * Using the flags below in the btree inode also makes us achieve the | |
1945 | * goal of AS_EIO/AS_ENOSPC when writepages() returns success, started | |
1946 | * writeback for all dirty pages and before filemap_fdatawait_range() | |
1947 | * is called, the writeback for all dirty pages had already finished | |
1948 | * with errors - because we were not using AS_EIO/AS_ENOSPC, | |
1949 | * filemap_fdatawait_range() would return success, as it could not know | |
1950 | * that writeback errors happened (the pages were no longer tagged for | |
1951 | * writeback). | |
1952 | */ | |
1953 | switch (eb->log_index) { | |
1954 | case -1: | |
5a2c6075 | 1955 | set_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags); |
656f30db FM |
1956 | break; |
1957 | case 0: | |
5a2c6075 | 1958 | set_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
656f30db FM |
1959 | break; |
1960 | case 1: | |
5a2c6075 | 1961 | set_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
656f30db FM |
1962 | break; |
1963 | default: | |
1964 | BUG(); /* unexpected, logic error */ | |
1965 | } | |
1966 | } | |
1967 | ||
2f3186d8 QW |
1968 | /* |
1969 | * The endio specific version which won't touch any unsafe spinlock in endio | |
1970 | * context. | |
1971 | */ | |
1972 | static struct extent_buffer *find_extent_buffer_nolock( | |
1973 | struct btrfs_fs_info *fs_info, u64 start) | |
1974 | { | |
1975 | struct extent_buffer *eb; | |
1976 | ||
1977 | rcu_read_lock(); | |
01cd3909 DS |
1978 | eb = radix_tree_lookup(&fs_info->buffer_radix, |
1979 | start >> fs_info->sectorsize_bits); | |
2f3186d8 QW |
1980 | if (eb && atomic_inc_not_zero(&eb->refs)) { |
1981 | rcu_read_unlock(); | |
1982 | return eb; | |
1983 | } | |
1984 | rcu_read_unlock(); | |
1985 | return NULL; | |
1986 | } | |
1987 | ||
1988 | /* | |
1989 | * The endio function for subpage extent buffer write. | |
1990 | * | |
1991 | * Unlike end_bio_extent_buffer_writepage(), we only call end_page_writeback() | |
1992 | * after all extent buffers in the page has finished their writeback. | |
1993 | */ | |
917f32a2 | 1994 | static void end_bio_subpage_eb_writepage(struct btrfs_bio *bbio) |
2f3186d8 | 1995 | { |
917f32a2 | 1996 | struct bio *bio = &bbio->bio; |
fa04c165 | 1997 | struct btrfs_fs_info *fs_info; |
2f3186d8 QW |
1998 | struct bio_vec *bvec; |
1999 | struct bvec_iter_all iter_all; | |
2000 | ||
fa04c165 | 2001 | fs_info = btrfs_sb(bio_first_page_all(bio)->mapping->host->i_sb); |
fbca46eb | 2002 | ASSERT(fs_info->nodesize < PAGE_SIZE); |
fa04c165 | 2003 | |
2f3186d8 QW |
2004 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2005 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
2006 | struct page *page = bvec->bv_page; | |
2007 | u64 bvec_start = page_offset(page) + bvec->bv_offset; | |
2008 | u64 bvec_end = bvec_start + bvec->bv_len - 1; | |
2009 | u64 cur_bytenr = bvec_start; | |
2010 | ||
2011 | ASSERT(IS_ALIGNED(bvec->bv_len, fs_info->nodesize)); | |
2012 | ||
2013 | /* Iterate through all extent buffers in the range */ | |
2014 | while (cur_bytenr <= bvec_end) { | |
2015 | struct extent_buffer *eb; | |
2016 | int done; | |
2017 | ||
2018 | /* | |
2019 | * Here we can't use find_extent_buffer(), as it may | |
2020 | * try to lock eb->refs_lock, which is not safe in endio | |
2021 | * context. | |
2022 | */ | |
2023 | eb = find_extent_buffer_nolock(fs_info, cur_bytenr); | |
2024 | ASSERT(eb); | |
2025 | ||
2026 | cur_bytenr = eb->start + eb->len; | |
2027 | ||
2028 | ASSERT(test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags)); | |
2029 | done = atomic_dec_and_test(&eb->io_pages); | |
2030 | ASSERT(done); | |
2031 | ||
2032 | if (bio->bi_status || | |
2033 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { | |
2034 | ClearPageUptodate(page); | |
2035 | set_btree_ioerr(page, eb); | |
2036 | } | |
2037 | ||
2038 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, | |
2039 | eb->len); | |
2040 | end_extent_buffer_writeback(eb); | |
2041 | /* | |
2042 | * free_extent_buffer() will grab spinlock which is not | |
2043 | * safe in endio context. Thus here we manually dec | |
2044 | * the ref. | |
2045 | */ | |
2046 | atomic_dec(&eb->refs); | |
2047 | } | |
2048 | } | |
2049 | bio_put(bio); | |
2050 | } | |
2051 | ||
917f32a2 | 2052 | static void end_bio_extent_buffer_writepage(struct btrfs_bio *bbio) |
0b32f4bb | 2053 | { |
917f32a2 | 2054 | struct bio *bio = &bbio->bio; |
2c30c71b | 2055 | struct bio_vec *bvec; |
0b32f4bb | 2056 | struct extent_buffer *eb; |
2b070cfe | 2057 | int done; |
6dc4f100 | 2058 | struct bvec_iter_all iter_all; |
0b32f4bb | 2059 | |
c09abff8 | 2060 | ASSERT(!bio_flagged(bio, BIO_CLONED)); |
2b070cfe | 2061 | bio_for_each_segment_all(bvec, bio, iter_all) { |
0b32f4bb JB |
2062 | struct page *page = bvec->bv_page; |
2063 | ||
0b32f4bb JB |
2064 | eb = (struct extent_buffer *)page->private; |
2065 | BUG_ON(!eb); | |
2066 | done = atomic_dec_and_test(&eb->io_pages); | |
2067 | ||
4e4cbee9 | 2068 | if (bio->bi_status || |
4246a0b6 | 2069 | test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags)) { |
0b32f4bb | 2070 | ClearPageUptodate(page); |
5a2c6075 | 2071 | set_btree_ioerr(page, eb); |
0b32f4bb JB |
2072 | } |
2073 | ||
2074 | end_page_writeback(page); | |
2075 | ||
2076 | if (!done) | |
2077 | continue; | |
2078 | ||
2079 | end_extent_buffer_writeback(eb); | |
2c30c71b | 2080 | } |
0b32f4bb JB |
2081 | |
2082 | bio_put(bio); | |
0b32f4bb JB |
2083 | } |
2084 | ||
fa04c165 QW |
2085 | static void prepare_eb_write(struct extent_buffer *eb) |
2086 | { | |
2087 | u32 nritems; | |
2088 | unsigned long start; | |
2089 | unsigned long end; | |
2090 | ||
2091 | clear_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags); | |
2092 | atomic_set(&eb->io_pages, num_extent_pages(eb)); | |
2093 | ||
2094 | /* Set btree blocks beyond nritems with 0 to avoid stale content */ | |
2095 | nritems = btrfs_header_nritems(eb); | |
2096 | if (btrfs_header_level(eb) > 0) { | |
e23efd8e | 2097 | end = btrfs_node_key_ptr_offset(eb, nritems); |
fa04c165 QW |
2098 | memzero_extent_buffer(eb, end, eb->len - end); |
2099 | } else { | |
2100 | /* | |
2101 | * Leaf: | |
2102 | * header 0 1 2 .. N ... data_N .. data_2 data_1 data_0 | |
2103 | */ | |
42c9419a | 2104 | start = btrfs_item_nr_offset(eb, nritems); |
8009adf3 | 2105 | end = btrfs_item_nr_offset(eb, 0); |
3a3178c7 JB |
2106 | if (nritems == 0) |
2107 | end += BTRFS_LEAF_DATA_SIZE(eb->fs_info); | |
2108 | else | |
2109 | end += btrfs_item_offset(eb, nritems - 1); | |
fa04c165 QW |
2110 | memzero_extent_buffer(eb, start, end - start); |
2111 | } | |
2112 | } | |
2113 | ||
35b6ddfa QW |
2114 | /* |
2115 | * Unlike the work in write_one_eb(), we rely completely on extent locking. | |
2116 | * Page locking is only utilized at minimum to keep the VMM code happy. | |
35b6ddfa QW |
2117 | */ |
2118 | static int write_one_subpage_eb(struct extent_buffer *eb, | |
2119 | struct writeback_control *wbc, | |
ee5f017d | 2120 | struct btrfs_bio_ctrl *bio_ctrl) |
35b6ddfa QW |
2121 | { |
2122 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
2123 | struct page *page = eb->pages[0]; | |
353767e4 | 2124 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
35b6ddfa QW |
2125 | bool no_dirty_ebs = false; |
2126 | int ret; | |
2127 | ||
fa04c165 QW |
2128 | prepare_eb_write(eb); |
2129 | ||
35b6ddfa QW |
2130 | /* clear_page_dirty_for_io() in subpage helper needs page locked */ |
2131 | lock_page(page); | |
2132 | btrfs_subpage_set_writeback(fs_info, page, eb->start, eb->len); | |
2133 | ||
2134 | /* Check if this is the last dirty bit to update nr_written */ | |
2135 | no_dirty_ebs = btrfs_subpage_clear_and_test_dirty(fs_info, page, | |
2136 | eb->start, eb->len); | |
2137 | if (no_dirty_ebs) | |
2138 | clear_page_dirty_for_io(page); | |
2139 | ||
ee5f017d | 2140 | bio_ctrl->end_io_func = end_bio_subpage_eb_writepage; |
5467abba | 2141 | |
390ed29b | 2142 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
ee5f017d | 2143 | bio_ctrl, eb->start, page, eb->len, |
5467abba | 2144 | eb->start - page_offset(page), 0, false); |
35b6ddfa QW |
2145 | if (ret) { |
2146 | btrfs_subpage_clear_writeback(fs_info, page, eb->start, eb->len); | |
2147 | set_btree_ioerr(page, eb); | |
2148 | unlock_page(page); | |
2149 | ||
2150 | if (atomic_dec_and_test(&eb->io_pages)) | |
2151 | end_extent_buffer_writeback(eb); | |
2152 | return -EIO; | |
2153 | } | |
2154 | unlock_page(page); | |
2155 | /* | |
2156 | * Submission finished without problem, if no range of the page is | |
2157 | * dirty anymore, we have submitted a page. Update nr_written in wbc. | |
2158 | */ | |
2159 | if (no_dirty_ebs) | |
572f3dad | 2160 | wbc->nr_to_write--; |
35b6ddfa QW |
2161 | return ret; |
2162 | } | |
2163 | ||
0e378df1 | 2164 | static noinline_for_stack int write_one_eb(struct extent_buffer *eb, |
0b32f4bb | 2165 | struct writeback_control *wbc, |
ee5f017d | 2166 | struct btrfs_bio_ctrl *bio_ctrl) |
0b32f4bb | 2167 | { |
0c64c33c | 2168 | u64 disk_bytenr = eb->start; |
cc5e31a4 | 2169 | int i, num_pages; |
353767e4 | 2170 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
d7dbe9e7 | 2171 | int ret = 0; |
0b32f4bb | 2172 | |
fa04c165 | 2173 | prepare_eb_write(eb); |
35b6ddfa | 2174 | |
ee5f017d | 2175 | bio_ctrl->end_io_func = end_bio_extent_buffer_writepage; |
5467abba | 2176 | |
fa04c165 | 2177 | num_pages = num_extent_pages(eb); |
0b32f4bb | 2178 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 2179 | struct page *p = eb->pages[i]; |
0b32f4bb JB |
2180 | |
2181 | clear_page_dirty_for_io(p); | |
2182 | set_page_writeback(p); | |
0ceb34bf | 2183 | ret = submit_extent_page(REQ_OP_WRITE | write_flags, wbc, |
ee5f017d | 2184 | bio_ctrl, disk_bytenr, p, |
5467abba | 2185 | PAGE_SIZE, 0, 0, false); |
0b32f4bb | 2186 | if (ret) { |
5a2c6075 | 2187 | set_btree_ioerr(p, eb); |
fe01aa65 TK |
2188 | if (PageWriteback(p)) |
2189 | end_page_writeback(p); | |
0b32f4bb JB |
2190 | if (atomic_sub_and_test(num_pages - i, &eb->io_pages)) |
2191 | end_extent_buffer_writeback(eb); | |
2192 | ret = -EIO; | |
2193 | break; | |
2194 | } | |
0c64c33c | 2195 | disk_bytenr += PAGE_SIZE; |
572f3dad | 2196 | wbc->nr_to_write--; |
0b32f4bb JB |
2197 | unlock_page(p); |
2198 | } | |
2199 | ||
2200 | if (unlikely(ret)) { | |
2201 | for (; i < num_pages; i++) { | |
bbf65cf0 | 2202 | struct page *p = eb->pages[i]; |
81465028 | 2203 | clear_page_dirty_for_io(p); |
0b32f4bb JB |
2204 | unlock_page(p); |
2205 | } | |
2206 | } | |
2207 | ||
2208 | return ret; | |
2209 | } | |
2210 | ||
c4aec299 QW |
2211 | /* |
2212 | * Submit one subpage btree page. | |
2213 | * | |
2214 | * The main difference to submit_eb_page() is: | |
2215 | * - Page locking | |
2216 | * For subpage, we don't rely on page locking at all. | |
2217 | * | |
2218 | * - Flush write bio | |
2219 | * We only flush bio if we may be unable to fit current extent buffers into | |
2220 | * current bio. | |
2221 | * | |
2222 | * Return >=0 for the number of submitted extent buffers. | |
2223 | * Return <0 for fatal error. | |
2224 | */ | |
2225 | static int submit_eb_subpage(struct page *page, | |
2226 | struct writeback_control *wbc, | |
ee5f017d | 2227 | struct btrfs_bio_ctrl *bio_ctrl) |
c4aec299 QW |
2228 | { |
2229 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
2230 | int submitted = 0; | |
2231 | u64 page_start = page_offset(page); | |
2232 | int bit_start = 0; | |
c4aec299 QW |
2233 | int sectors_per_node = fs_info->nodesize >> fs_info->sectorsize_bits; |
2234 | int ret; | |
2235 | ||
2236 | /* Lock and write each dirty extent buffers in the range */ | |
72a69cd0 | 2237 | while (bit_start < fs_info->subpage_info->bitmap_nr_bits) { |
c4aec299 QW |
2238 | struct btrfs_subpage *subpage = (struct btrfs_subpage *)page->private; |
2239 | struct extent_buffer *eb; | |
2240 | unsigned long flags; | |
2241 | u64 start; | |
2242 | ||
2243 | /* | |
2244 | * Take private lock to ensure the subpage won't be detached | |
2245 | * in the meantime. | |
2246 | */ | |
2247 | spin_lock(&page->mapping->private_lock); | |
2248 | if (!PagePrivate(page)) { | |
2249 | spin_unlock(&page->mapping->private_lock); | |
2250 | break; | |
2251 | } | |
2252 | spin_lock_irqsave(&subpage->lock, flags); | |
72a69cd0 QW |
2253 | if (!test_bit(bit_start + fs_info->subpage_info->dirty_offset, |
2254 | subpage->bitmaps)) { | |
c4aec299 QW |
2255 | spin_unlock_irqrestore(&subpage->lock, flags); |
2256 | spin_unlock(&page->mapping->private_lock); | |
2257 | bit_start++; | |
2258 | continue; | |
2259 | } | |
2260 | ||
2261 | start = page_start + bit_start * fs_info->sectorsize; | |
2262 | bit_start += sectors_per_node; | |
2263 | ||
2264 | /* | |
2265 | * Here we just want to grab the eb without touching extra | |
2266 | * spin locks, so call find_extent_buffer_nolock(). | |
2267 | */ | |
2268 | eb = find_extent_buffer_nolock(fs_info, start); | |
2269 | spin_unlock_irqrestore(&subpage->lock, flags); | |
2270 | spin_unlock(&page->mapping->private_lock); | |
2271 | ||
2272 | /* | |
2273 | * The eb has already reached 0 refs thus find_extent_buffer() | |
2274 | * doesn't return it. We don't need to write back such eb | |
2275 | * anyway. | |
2276 | */ | |
2277 | if (!eb) | |
2278 | continue; | |
2279 | ||
ee5f017d | 2280 | ret = lock_extent_buffer_for_io(eb, bio_ctrl); |
c4aec299 QW |
2281 | if (ret == 0) { |
2282 | free_extent_buffer(eb); | |
2283 | continue; | |
2284 | } | |
2285 | if (ret < 0) { | |
2286 | free_extent_buffer(eb); | |
2287 | goto cleanup; | |
2288 | } | |
ee5f017d | 2289 | ret = write_one_subpage_eb(eb, wbc, bio_ctrl); |
c4aec299 QW |
2290 | free_extent_buffer(eb); |
2291 | if (ret < 0) | |
2292 | goto cleanup; | |
2293 | submitted++; | |
2294 | } | |
2295 | return submitted; | |
2296 | ||
2297 | cleanup: | |
2298 | /* We hit error, end bio for the submitted extent buffers */ | |
ee5f017d | 2299 | submit_write_bio(bio_ctrl, ret); |
c4aec299 QW |
2300 | return ret; |
2301 | } | |
2302 | ||
f91e0d0c QW |
2303 | /* |
2304 | * Submit all page(s) of one extent buffer. | |
2305 | * | |
2306 | * @page: the page of one extent buffer | |
2307 | * @eb_context: to determine if we need to submit this page, if current page | |
2308 | * belongs to this eb, we don't need to submit | |
2309 | * | |
2310 | * The caller should pass each page in their bytenr order, and here we use | |
2311 | * @eb_context to determine if we have submitted pages of one extent buffer. | |
2312 | * | |
2313 | * If we have, we just skip until we hit a new page that doesn't belong to | |
2314 | * current @eb_context. | |
2315 | * | |
2316 | * If not, we submit all the page(s) of the extent buffer. | |
2317 | * | |
2318 | * Return >0 if we have submitted the extent buffer successfully. | |
2319 | * Return 0 if we don't need to submit the page, as it's already submitted by | |
2320 | * previous call. | |
2321 | * Return <0 for fatal error. | |
2322 | */ | |
2323 | static int submit_eb_page(struct page *page, struct writeback_control *wbc, | |
ee5f017d | 2324 | struct btrfs_bio_ctrl *bio_ctrl, |
f91e0d0c QW |
2325 | struct extent_buffer **eb_context) |
2326 | { | |
2327 | struct address_space *mapping = page->mapping; | |
0bc09ca1 | 2328 | struct btrfs_block_group *cache = NULL; |
f91e0d0c QW |
2329 | struct extent_buffer *eb; |
2330 | int ret; | |
2331 | ||
2332 | if (!PagePrivate(page)) | |
2333 | return 0; | |
2334 | ||
fbca46eb | 2335 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
ee5f017d | 2336 | return submit_eb_subpage(page, wbc, bio_ctrl); |
c4aec299 | 2337 | |
f91e0d0c QW |
2338 | spin_lock(&mapping->private_lock); |
2339 | if (!PagePrivate(page)) { | |
2340 | spin_unlock(&mapping->private_lock); | |
2341 | return 0; | |
2342 | } | |
2343 | ||
2344 | eb = (struct extent_buffer *)page->private; | |
2345 | ||
2346 | /* | |
2347 | * Shouldn't happen and normally this would be a BUG_ON but no point | |
2348 | * crashing the machine for something we can survive anyway. | |
2349 | */ | |
2350 | if (WARN_ON(!eb)) { | |
2351 | spin_unlock(&mapping->private_lock); | |
2352 | return 0; | |
2353 | } | |
2354 | ||
2355 | if (eb == *eb_context) { | |
2356 | spin_unlock(&mapping->private_lock); | |
2357 | return 0; | |
2358 | } | |
2359 | ret = atomic_inc_not_zero(&eb->refs); | |
2360 | spin_unlock(&mapping->private_lock); | |
2361 | if (!ret) | |
2362 | return 0; | |
2363 | ||
0bc09ca1 NA |
2364 | if (!btrfs_check_meta_write_pointer(eb->fs_info, eb, &cache)) { |
2365 | /* | |
2366 | * If for_sync, this hole will be filled with | |
2367 | * trasnsaction commit. | |
2368 | */ | |
2369 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
2370 | ret = -EAGAIN; | |
2371 | else | |
2372 | ret = 0; | |
2373 | free_extent_buffer(eb); | |
2374 | return ret; | |
2375 | } | |
2376 | ||
f91e0d0c QW |
2377 | *eb_context = eb; |
2378 | ||
ee5f017d | 2379 | ret = lock_extent_buffer_for_io(eb, bio_ctrl); |
f91e0d0c | 2380 | if (ret <= 0) { |
0bc09ca1 NA |
2381 | btrfs_revert_meta_write_pointer(cache, eb); |
2382 | if (cache) | |
2383 | btrfs_put_block_group(cache); | |
f91e0d0c QW |
2384 | free_extent_buffer(eb); |
2385 | return ret; | |
2386 | } | |
be1a1d7a | 2387 | if (cache) { |
d3e29967 NB |
2388 | /* |
2389 | * Implies write in zoned mode. Mark the last eb in a block group. | |
2390 | */ | |
56fbb0a4 | 2391 | btrfs_schedule_zone_finish_bg(cache, eb); |
d3e29967 | 2392 | btrfs_put_block_group(cache); |
be1a1d7a | 2393 | } |
ee5f017d | 2394 | ret = write_one_eb(eb, wbc, bio_ctrl); |
f91e0d0c QW |
2395 | free_extent_buffer(eb); |
2396 | if (ret < 0) | |
2397 | return ret; | |
2398 | return 1; | |
2399 | } | |
2400 | ||
0b32f4bb JB |
2401 | int btree_write_cache_pages(struct address_space *mapping, |
2402 | struct writeback_control *wbc) | |
2403 | { | |
f91e0d0c | 2404 | struct extent_buffer *eb_context = NULL; |
ee5f017d | 2405 | struct btrfs_bio_ctrl bio_ctrl = { |
0b32f4bb | 2406 | .extent_locked = 0, |
ee5f017d | 2407 | .sync_io = (wbc->sync_mode == WB_SYNC_ALL), |
0b32f4bb | 2408 | }; |
b3ff8f1d | 2409 | struct btrfs_fs_info *fs_info = BTRFS_I(mapping->host)->root->fs_info; |
0b32f4bb JB |
2410 | int ret = 0; |
2411 | int done = 0; | |
2412 | int nr_to_write_done = 0; | |
2413 | struct pagevec pvec; | |
2414 | int nr_pages; | |
2415 | pgoff_t index; | |
2416 | pgoff_t end; /* Inclusive */ | |
2417 | int scanned = 0; | |
10bbd235 | 2418 | xa_mark_t tag; |
0b32f4bb | 2419 | |
86679820 | 2420 | pagevec_init(&pvec); |
0b32f4bb JB |
2421 | if (wbc->range_cyclic) { |
2422 | index = mapping->writeback_index; /* Start from prev offset */ | |
2423 | end = -1; | |
556755a8 JB |
2424 | /* |
2425 | * Start from the beginning does not need to cycle over the | |
2426 | * range, mark it as scanned. | |
2427 | */ | |
2428 | scanned = (index == 0); | |
0b32f4bb | 2429 | } else { |
09cbfeaf KS |
2430 | index = wbc->range_start >> PAGE_SHIFT; |
2431 | end = wbc->range_end >> PAGE_SHIFT; | |
0b32f4bb JB |
2432 | scanned = 1; |
2433 | } | |
2434 | if (wbc->sync_mode == WB_SYNC_ALL) | |
2435 | tag = PAGECACHE_TAG_TOWRITE; | |
2436 | else | |
2437 | tag = PAGECACHE_TAG_DIRTY; | |
0bc09ca1 | 2438 | btrfs_zoned_meta_io_lock(fs_info); |
0b32f4bb JB |
2439 | retry: |
2440 | if (wbc->sync_mode == WB_SYNC_ALL) | |
2441 | tag_pages_for_writeback(mapping, index, end); | |
2442 | while (!done && !nr_to_write_done && (index <= end) && | |
4006f437 | 2443 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2444 | tag))) { |
0b32f4bb JB |
2445 | unsigned i; |
2446 | ||
0b32f4bb JB |
2447 | for (i = 0; i < nr_pages; i++) { |
2448 | struct page *page = pvec.pages[i]; | |
2449 | ||
ee5f017d | 2450 | ret = submit_eb_page(page, wbc, &bio_ctrl, &eb_context); |
f91e0d0c | 2451 | if (ret == 0) |
0b32f4bb | 2452 | continue; |
f91e0d0c | 2453 | if (ret < 0) { |
0b32f4bb | 2454 | done = 1; |
0b32f4bb JB |
2455 | break; |
2456 | } | |
0b32f4bb JB |
2457 | |
2458 | /* | |
2459 | * the filesystem may choose to bump up nr_to_write. | |
2460 | * We have to make sure to honor the new nr_to_write | |
2461 | * at any time | |
2462 | */ | |
2463 | nr_to_write_done = wbc->nr_to_write <= 0; | |
2464 | } | |
2465 | pagevec_release(&pvec); | |
2466 | cond_resched(); | |
2467 | } | |
2468 | if (!scanned && !done) { | |
2469 | /* | |
2470 | * We hit the last page and there is more work to be done: wrap | |
2471 | * back to the start of the file | |
2472 | */ | |
2473 | scanned = 1; | |
2474 | index = 0; | |
2475 | goto retry; | |
2476 | } | |
b3ff8f1d QW |
2477 | /* |
2478 | * If something went wrong, don't allow any metadata write bio to be | |
2479 | * submitted. | |
2480 | * | |
2481 | * This would prevent use-after-free if we had dirty pages not | |
2482 | * cleaned up, which can still happen by fuzzed images. | |
2483 | * | |
2484 | * - Bad extent tree | |
2485 | * Allowing existing tree block to be allocated for other trees. | |
2486 | * | |
2487 | * - Log tree operations | |
2488 | * Exiting tree blocks get allocated to log tree, bumps its | |
2489 | * generation, then get cleaned in tree re-balance. | |
2490 | * Such tree block will not be written back, since it's clean, | |
2491 | * thus no WRITTEN flag set. | |
2492 | * And after log writes back, this tree block is not traced by | |
2493 | * any dirty extent_io_tree. | |
2494 | * | |
2495 | * - Offending tree block gets re-dirtied from its original owner | |
2496 | * Since it has bumped generation, no WRITTEN flag, it can be | |
2497 | * reused without COWing. This tree block will not be traced | |
2498 | * by btrfs_transaction::dirty_pages. | |
2499 | * | |
2500 | * Now such dirty tree block will not be cleaned by any dirty | |
2501 | * extent io tree. Thus we don't want to submit such wild eb | |
2502 | * if the fs already has error. | |
9845e5dd | 2503 | * |
c9583ada QW |
2504 | * We can get ret > 0 from submit_extent_page() indicating how many ebs |
2505 | * were submitted. Reset it to 0 to avoid false alerts for the caller. | |
2506 | */ | |
2507 | if (ret > 0) | |
2508 | ret = 0; | |
9845e5dd CH |
2509 | if (!ret && BTRFS_FS_ERROR(fs_info)) |
2510 | ret = -EROFS; | |
ee5f017d | 2511 | submit_write_bio(&bio_ctrl, ret); |
9845e5dd CH |
2512 | |
2513 | btrfs_zoned_meta_io_unlock(fs_info); | |
0b32f4bb JB |
2514 | return ret; |
2515 | } | |
2516 | ||
43dd529a | 2517 | /* |
3bed2da1 NB |
2518 | * Walk the list of dirty pages of the given address space and write all of them. |
2519 | * | |
ee5f017d DS |
2520 | * @mapping: address space structure to write |
2521 | * @wbc: subtract the number of written pages from *@wbc->nr_to_write | |
2522 | * @bio_ctrl: holds context for the write, namely the bio | |
d1310b2e CM |
2523 | * |
2524 | * If a page is already under I/O, write_cache_pages() skips it, even | |
2525 | * if it's dirty. This is desirable behaviour for memory-cleaning writeback, | |
2526 | * but it is INCORRECT for data-integrity system calls such as fsync(). fsync() | |
2527 | * and msync() need to guarantee that all the data which was dirty at the time | |
2528 | * the call was made get new I/O started against them. If wbc->sync_mode is | |
2529 | * WB_SYNC_ALL then we were called for data integrity and we must wait for | |
2530 | * existing IO to complete. | |
2531 | */ | |
4242b64a | 2532 | static int extent_write_cache_pages(struct address_space *mapping, |
4bef0848 | 2533 | struct writeback_control *wbc, |
ee5f017d | 2534 | struct btrfs_bio_ctrl *bio_ctrl) |
d1310b2e | 2535 | { |
7fd1a3f7 | 2536 | struct inode *inode = mapping->host; |
d1310b2e CM |
2537 | int ret = 0; |
2538 | int done = 0; | |
f85d7d6c | 2539 | int nr_to_write_done = 0; |
d1310b2e CM |
2540 | struct pagevec pvec; |
2541 | int nr_pages; | |
2542 | pgoff_t index; | |
2543 | pgoff_t end; /* Inclusive */ | |
a9132667 LB |
2544 | pgoff_t done_index; |
2545 | int range_whole = 0; | |
d1310b2e | 2546 | int scanned = 0; |
10bbd235 | 2547 | xa_mark_t tag; |
d1310b2e | 2548 | |
7fd1a3f7 JB |
2549 | /* |
2550 | * We have to hold onto the inode so that ordered extents can do their | |
2551 | * work when the IO finishes. The alternative to this is failing to add | |
2552 | * an ordered extent if the igrab() fails there and that is a huge pain | |
2553 | * to deal with, so instead just hold onto the inode throughout the | |
2554 | * writepages operation. If it fails here we are freeing up the inode | |
2555 | * anyway and we'd rather not waste our time writing out stuff that is | |
2556 | * going to be truncated anyway. | |
2557 | */ | |
2558 | if (!igrab(inode)) | |
2559 | return 0; | |
2560 | ||
86679820 | 2561 | pagevec_init(&pvec); |
d1310b2e CM |
2562 | if (wbc->range_cyclic) { |
2563 | index = mapping->writeback_index; /* Start from prev offset */ | |
2564 | end = -1; | |
556755a8 JB |
2565 | /* |
2566 | * Start from the beginning does not need to cycle over the | |
2567 | * range, mark it as scanned. | |
2568 | */ | |
2569 | scanned = (index == 0); | |
d1310b2e | 2570 | } else { |
09cbfeaf KS |
2571 | index = wbc->range_start >> PAGE_SHIFT; |
2572 | end = wbc->range_end >> PAGE_SHIFT; | |
a9132667 LB |
2573 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2574 | range_whole = 1; | |
d1310b2e CM |
2575 | scanned = 1; |
2576 | } | |
3cd24c69 EL |
2577 | |
2578 | /* | |
2579 | * We do the tagged writepage as long as the snapshot flush bit is set | |
2580 | * and we are the first one who do the filemap_flush() on this inode. | |
2581 | * | |
2582 | * The nr_to_write == LONG_MAX is needed to make sure other flushers do | |
2583 | * not race in and drop the bit. | |
2584 | */ | |
2585 | if (range_whole && wbc->nr_to_write == LONG_MAX && | |
2586 | test_and_clear_bit(BTRFS_INODE_SNAPSHOT_FLUSH, | |
2587 | &BTRFS_I(inode)->runtime_flags)) | |
2588 | wbc->tagged_writepages = 1; | |
2589 | ||
2590 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) | |
f7aaa06b JB |
2591 | tag = PAGECACHE_TAG_TOWRITE; |
2592 | else | |
2593 | tag = PAGECACHE_TAG_DIRTY; | |
d1310b2e | 2594 | retry: |
3cd24c69 | 2595 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
f7aaa06b | 2596 | tag_pages_for_writeback(mapping, index, end); |
a9132667 | 2597 | done_index = index; |
f85d7d6c | 2598 | while (!done && !nr_to_write_done && (index <= end) && |
67fd707f JK |
2599 | (nr_pages = pagevec_lookup_range_tag(&pvec, mapping, |
2600 | &index, end, tag))) { | |
d1310b2e CM |
2601 | unsigned i; |
2602 | ||
d1310b2e CM |
2603 | for (i = 0; i < nr_pages; i++) { |
2604 | struct page *page = pvec.pages[i]; | |
2605 | ||
f7bddf1e | 2606 | done_index = page->index + 1; |
d1310b2e | 2607 | /* |
b93b0163 MW |
2608 | * At this point we hold neither the i_pages lock nor |
2609 | * the page lock: the page may be truncated or | |
2610 | * invalidated (changing page->mapping to NULL), | |
2611 | * or even swizzled back from swapper_space to | |
2612 | * tmpfs file mapping | |
d1310b2e | 2613 | */ |
c8f2f24b | 2614 | if (!trylock_page(page)) { |
ee5f017d | 2615 | submit_write_bio(bio_ctrl, 0); |
c8f2f24b | 2616 | lock_page(page); |
01d658f2 | 2617 | } |
d1310b2e CM |
2618 | |
2619 | if (unlikely(page->mapping != mapping)) { | |
2620 | unlock_page(page); | |
2621 | continue; | |
2622 | } | |
2623 | ||
d2c3f4f6 | 2624 | if (wbc->sync_mode != WB_SYNC_NONE) { |
c9583ada | 2625 | if (PageWriteback(page)) |
ee5f017d | 2626 | submit_write_bio(bio_ctrl, 0); |
d1310b2e | 2627 | wait_on_page_writeback(page); |
d2c3f4f6 | 2628 | } |
d1310b2e CM |
2629 | |
2630 | if (PageWriteback(page) || | |
2631 | !clear_page_dirty_for_io(page)) { | |
2632 | unlock_page(page); | |
2633 | continue; | |
2634 | } | |
2635 | ||
ee5f017d | 2636 | ret = __extent_writepage(page, wbc, bio_ctrl); |
a9132667 | 2637 | if (ret < 0) { |
a9132667 LB |
2638 | done = 1; |
2639 | break; | |
2640 | } | |
f85d7d6c CM |
2641 | |
2642 | /* | |
2643 | * the filesystem may choose to bump up nr_to_write. | |
2644 | * We have to make sure to honor the new nr_to_write | |
2645 | * at any time | |
2646 | */ | |
2647 | nr_to_write_done = wbc->nr_to_write <= 0; | |
d1310b2e CM |
2648 | } |
2649 | pagevec_release(&pvec); | |
2650 | cond_resched(); | |
2651 | } | |
894b36e3 | 2652 | if (!scanned && !done) { |
d1310b2e CM |
2653 | /* |
2654 | * We hit the last page and there is more work to be done: wrap | |
2655 | * back to the start of the file | |
2656 | */ | |
2657 | scanned = 1; | |
2658 | index = 0; | |
42ffb0bf JB |
2659 | |
2660 | /* | |
2661 | * If we're looping we could run into a page that is locked by a | |
2662 | * writer and that writer could be waiting on writeback for a | |
2663 | * page in our current bio, and thus deadlock, so flush the | |
2664 | * write bio here. | |
2665 | */ | |
ee5f017d | 2666 | submit_write_bio(bio_ctrl, 0); |
c9583ada | 2667 | goto retry; |
d1310b2e | 2668 | } |
a9132667 LB |
2669 | |
2670 | if (wbc->range_cyclic || (wbc->nr_to_write > 0 && range_whole)) | |
2671 | mapping->writeback_index = done_index; | |
2672 | ||
e55cf7ca | 2673 | btrfs_add_delayed_iput(BTRFS_I(inode)); |
894b36e3 | 2674 | return ret; |
d1310b2e | 2675 | } |
d1310b2e | 2676 | |
2bd0fc93 QW |
2677 | /* |
2678 | * Submit the pages in the range to bio for call sites which delalloc range has | |
2679 | * already been ran (aka, ordered extent inserted) and all pages are still | |
2680 | * locked. | |
2681 | */ | |
2682 | int extent_write_locked_range(struct inode *inode, u64 start, u64 end) | |
771ed689 | 2683 | { |
2bd0fc93 QW |
2684 | bool found_error = false; |
2685 | int first_error = 0; | |
771ed689 CM |
2686 | int ret = 0; |
2687 | struct address_space *mapping = inode->i_mapping; | |
2688 | struct page *page; | |
2bd0fc93 | 2689 | u64 cur = start; |
66448b9d QW |
2690 | unsigned long nr_pages; |
2691 | const u32 sectorsize = btrfs_sb(inode->i_sb)->sectorsize; | |
ee5f017d | 2692 | struct btrfs_bio_ctrl bio_ctrl = { |
771ed689 | 2693 | .extent_locked = 1, |
2bd0fc93 | 2694 | .sync_io = 1, |
771ed689 CM |
2695 | }; |
2696 | struct writeback_control wbc_writepages = { | |
2bd0fc93 | 2697 | .sync_mode = WB_SYNC_ALL, |
771ed689 CM |
2698 | .range_start = start, |
2699 | .range_end = end + 1, | |
ec39f769 CM |
2700 | /* We're called from an async helper function */ |
2701 | .punt_to_cgroup = 1, | |
2702 | .no_cgroup_owner = 1, | |
771ed689 CM |
2703 | }; |
2704 | ||
66448b9d QW |
2705 | ASSERT(IS_ALIGNED(start, sectorsize) && IS_ALIGNED(end + 1, sectorsize)); |
2706 | nr_pages = (round_up(end, PAGE_SIZE) - round_down(start, PAGE_SIZE)) >> | |
2707 | PAGE_SHIFT; | |
2708 | wbc_writepages.nr_to_write = nr_pages * 2; | |
2709 | ||
dbb70bec | 2710 | wbc_attach_fdatawrite_inode(&wbc_writepages, inode); |
2bd0fc93 | 2711 | while (cur <= end) { |
66448b9d QW |
2712 | u64 cur_end = min(round_down(cur, PAGE_SIZE) + PAGE_SIZE - 1, end); |
2713 | ||
2bd0fc93 QW |
2714 | page = find_get_page(mapping, cur >> PAGE_SHIFT); |
2715 | /* | |
2716 | * All pages in the range are locked since | |
2717 | * btrfs_run_delalloc_range(), thus there is no way to clear | |
2718 | * the page dirty flag. | |
2719 | */ | |
66448b9d | 2720 | ASSERT(PageLocked(page)); |
2bd0fc93 QW |
2721 | ASSERT(PageDirty(page)); |
2722 | clear_page_dirty_for_io(page); | |
ee5f017d | 2723 | ret = __extent_writepage(page, &wbc_writepages, &bio_ctrl); |
2bd0fc93 QW |
2724 | ASSERT(ret <= 0); |
2725 | if (ret < 0) { | |
2726 | found_error = true; | |
2727 | first_error = ret; | |
771ed689 | 2728 | } |
09cbfeaf | 2729 | put_page(page); |
66448b9d | 2730 | cur = cur_end + 1; |
771ed689 CM |
2731 | } |
2732 | ||
ee5f017d | 2733 | submit_write_bio(&bio_ctrl, found_error ? ret : 0); |
dbb70bec CM |
2734 | |
2735 | wbc_detach_inode(&wbc_writepages); | |
2bd0fc93 QW |
2736 | if (found_error) |
2737 | return first_error; | |
771ed689 CM |
2738 | return ret; |
2739 | } | |
d1310b2e | 2740 | |
8ae225a8 | 2741 | int extent_writepages(struct address_space *mapping, |
d1310b2e CM |
2742 | struct writeback_control *wbc) |
2743 | { | |
35156d85 | 2744 | struct inode *inode = mapping->host; |
d1310b2e | 2745 | int ret = 0; |
ee5f017d | 2746 | struct btrfs_bio_ctrl bio_ctrl = { |
771ed689 | 2747 | .extent_locked = 0, |
ee5f017d | 2748 | .sync_io = (wbc->sync_mode == WB_SYNC_ALL), |
d1310b2e CM |
2749 | }; |
2750 | ||
35156d85 JT |
2751 | /* |
2752 | * Allow only a single thread to do the reloc work in zoned mode to | |
2753 | * protect the write pointer updates. | |
2754 | */ | |
869f4cdc | 2755 | btrfs_zoned_data_reloc_lock(BTRFS_I(inode)); |
ee5f017d DS |
2756 | ret = extent_write_cache_pages(mapping, wbc, &bio_ctrl); |
2757 | submit_write_bio(&bio_ctrl, ret); | |
19ab78ca | 2758 | btrfs_zoned_data_reloc_unlock(BTRFS_I(inode)); |
d1310b2e CM |
2759 | return ret; |
2760 | } | |
d1310b2e | 2761 | |
ba206a02 | 2762 | void extent_readahead(struct readahead_control *rac) |
d1310b2e | 2763 | { |
390ed29b | 2764 | struct btrfs_bio_ctrl bio_ctrl = { 0 }; |
67c9684f | 2765 | struct page *pagepool[16]; |
125bac01 | 2766 | struct extent_map *em_cached = NULL; |
808f80b4 | 2767 | u64 prev_em_start = (u64)-1; |
ba206a02 | 2768 | int nr; |
d1310b2e | 2769 | |
ba206a02 | 2770 | while ((nr = readahead_page_batch(rac, pagepool))) { |
32c0a6bc MWO |
2771 | u64 contig_start = readahead_pos(rac); |
2772 | u64 contig_end = contig_start + readahead_batch_length(rac) - 1; | |
e65ef21e | 2773 | |
ba206a02 | 2774 | contiguous_readpages(pagepool, nr, contig_start, contig_end, |
390ed29b | 2775 | &em_cached, &bio_ctrl, &prev_em_start); |
d1310b2e | 2776 | } |
67c9684f | 2777 | |
125bac01 MX |
2778 | if (em_cached) |
2779 | free_extent_map(em_cached); | |
722c82ac | 2780 | submit_one_bio(&bio_ctrl); |
d1310b2e | 2781 | } |
d1310b2e CM |
2782 | |
2783 | /* | |
895586eb MWO |
2784 | * basic invalidate_folio code, this waits on any locked or writeback |
2785 | * ranges corresponding to the folio, and then deletes any extent state | |
d1310b2e CM |
2786 | * records from the tree |
2787 | */ | |
895586eb MWO |
2788 | int extent_invalidate_folio(struct extent_io_tree *tree, |
2789 | struct folio *folio, size_t offset) | |
d1310b2e | 2790 | { |
2ac55d41 | 2791 | struct extent_state *cached_state = NULL; |
895586eb MWO |
2792 | u64 start = folio_pos(folio); |
2793 | u64 end = start + folio_size(folio) - 1; | |
2794 | size_t blocksize = folio->mapping->host->i_sb->s_blocksize; | |
d1310b2e | 2795 | |
829ddec9 QW |
2796 | /* This function is only called for the btree inode */ |
2797 | ASSERT(tree->owner == IO_TREE_BTREE_INODE_IO); | |
2798 | ||
fda2832f | 2799 | start += ALIGN(offset, blocksize); |
d1310b2e CM |
2800 | if (start > end) |
2801 | return 0; | |
2802 | ||
570eb97b | 2803 | lock_extent(tree, start, end, &cached_state); |
895586eb | 2804 | folio_wait_writeback(folio); |
829ddec9 QW |
2805 | |
2806 | /* | |
2807 | * Currently for btree io tree, only EXTENT_LOCKED is utilized, | |
2808 | * so here we only need to unlock the extent range to free any | |
2809 | * existing extent state. | |
2810 | */ | |
570eb97b | 2811 | unlock_extent(tree, start, end, &cached_state); |
d1310b2e CM |
2812 | return 0; |
2813 | } | |
d1310b2e | 2814 | |
7b13b7b1 | 2815 | /* |
f913cff3 | 2816 | * a helper for release_folio, this tests for areas of the page that |
7b13b7b1 CM |
2817 | * are locked or under IO and drops the related state bits if it is safe |
2818 | * to drop the page. | |
2819 | */ | |
29c68b2d | 2820 | static int try_release_extent_state(struct extent_io_tree *tree, |
48a3b636 | 2821 | struct page *page, gfp_t mask) |
7b13b7b1 | 2822 | { |
4eee4fa4 | 2823 | u64 start = page_offset(page); |
09cbfeaf | 2824 | u64 end = start + PAGE_SIZE - 1; |
7b13b7b1 CM |
2825 | int ret = 1; |
2826 | ||
8882679e | 2827 | if (test_range_bit(tree, start, end, EXTENT_LOCKED, 0, NULL)) { |
7b13b7b1 | 2828 | ret = 0; |
8882679e | 2829 | } else { |
b71fb16b JB |
2830 | u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM | |
2831 | EXTENT_DELALLOC_NEW | EXTENT_CTLBITS); | |
2832 | ||
11ef160f | 2833 | /* |
2766ff61 FM |
2834 | * At this point we can safely clear everything except the |
2835 | * locked bit, the nodatasum bit and the delalloc new bit. | |
2836 | * The delalloc new bit will be cleared by ordered extent | |
2837 | * completion. | |
11ef160f | 2838 | */ |
bd015294 | 2839 | ret = __clear_extent_bit(tree, start, end, clear_bits, NULL, |
b71fb16b | 2840 | mask, NULL); |
e3f24cc5 CM |
2841 | |
2842 | /* if clear_extent_bit failed for enomem reasons, | |
2843 | * we can't allow the release to continue. | |
2844 | */ | |
2845 | if (ret < 0) | |
2846 | ret = 0; | |
2847 | else | |
2848 | ret = 1; | |
7b13b7b1 CM |
2849 | } |
2850 | return ret; | |
2851 | } | |
7b13b7b1 | 2852 | |
d1310b2e | 2853 | /* |
f913cff3 | 2854 | * a helper for release_folio. As long as there are no locked extents |
d1310b2e CM |
2855 | * in the range corresponding to the page, both state records and extent |
2856 | * map records are removed | |
2857 | */ | |
477a30ba | 2858 | int try_release_extent_mapping(struct page *page, gfp_t mask) |
d1310b2e CM |
2859 | { |
2860 | struct extent_map *em; | |
4eee4fa4 | 2861 | u64 start = page_offset(page); |
09cbfeaf | 2862 | u64 end = start + PAGE_SIZE - 1; |
bd3599a0 FM |
2863 | struct btrfs_inode *btrfs_inode = BTRFS_I(page->mapping->host); |
2864 | struct extent_io_tree *tree = &btrfs_inode->io_tree; | |
2865 | struct extent_map_tree *map = &btrfs_inode->extent_tree; | |
7b13b7b1 | 2866 | |
d0164adc | 2867 | if (gfpflags_allow_blocking(mask) && |
ee22184b | 2868 | page->mapping->host->i_size > SZ_16M) { |
39b5637f | 2869 | u64 len; |
70dec807 | 2870 | while (start <= end) { |
fbc2bd7e FM |
2871 | struct btrfs_fs_info *fs_info; |
2872 | u64 cur_gen; | |
2873 | ||
39b5637f | 2874 | len = end - start + 1; |
890871be | 2875 | write_lock(&map->lock); |
39b5637f | 2876 | em = lookup_extent_mapping(map, start, len); |
285190d9 | 2877 | if (!em) { |
890871be | 2878 | write_unlock(&map->lock); |
70dec807 CM |
2879 | break; |
2880 | } | |
7f3c74fb CM |
2881 | if (test_bit(EXTENT_FLAG_PINNED, &em->flags) || |
2882 | em->start != start) { | |
890871be | 2883 | write_unlock(&map->lock); |
70dec807 CM |
2884 | free_extent_map(em); |
2885 | break; | |
2886 | } | |
3d6448e6 FM |
2887 | if (test_range_bit(tree, em->start, |
2888 | extent_map_end(em) - 1, | |
2889 | EXTENT_LOCKED, 0, NULL)) | |
2890 | goto next; | |
2891 | /* | |
2892 | * If it's not in the list of modified extents, used | |
2893 | * by a fast fsync, we can remove it. If it's being | |
2894 | * logged we can safely remove it since fsync took an | |
2895 | * extra reference on the em. | |
2896 | */ | |
2897 | if (list_empty(&em->list) || | |
fbc2bd7e FM |
2898 | test_bit(EXTENT_FLAG_LOGGING, &em->flags)) |
2899 | goto remove_em; | |
2900 | /* | |
2901 | * If it's in the list of modified extents, remove it | |
2902 | * only if its generation is older then the current one, | |
2903 | * in which case we don't need it for a fast fsync. | |
2904 | * Otherwise don't remove it, we could be racing with an | |
2905 | * ongoing fast fsync that could miss the new extent. | |
2906 | */ | |
2907 | fs_info = btrfs_inode->root->fs_info; | |
2908 | spin_lock(&fs_info->trans_lock); | |
2909 | cur_gen = fs_info->generation; | |
2910 | spin_unlock(&fs_info->trans_lock); | |
2911 | if (em->generation >= cur_gen) | |
2912 | goto next; | |
2913 | remove_em: | |
5e548b32 FM |
2914 | /* |
2915 | * We only remove extent maps that are not in the list of | |
2916 | * modified extents or that are in the list but with a | |
2917 | * generation lower then the current generation, so there | |
2918 | * is no need to set the full fsync flag on the inode (it | |
2919 | * hurts the fsync performance for workloads with a data | |
2920 | * size that exceeds or is close to the system's memory). | |
2921 | */ | |
fbc2bd7e FM |
2922 | remove_extent_mapping(map, em); |
2923 | /* once for the rb tree */ | |
2924 | free_extent_map(em); | |
3d6448e6 | 2925 | next: |
70dec807 | 2926 | start = extent_map_end(em); |
890871be | 2927 | write_unlock(&map->lock); |
70dec807 CM |
2928 | |
2929 | /* once for us */ | |
d1310b2e | 2930 | free_extent_map(em); |
9f47eb54 PM |
2931 | |
2932 | cond_resched(); /* Allow large-extent preemption. */ | |
d1310b2e | 2933 | } |
d1310b2e | 2934 | } |
29c68b2d | 2935 | return try_release_extent_state(tree, page, mask); |
d1310b2e | 2936 | } |
d1310b2e | 2937 | |
4751832d QW |
2938 | /* |
2939 | * To cache previous fiemap extent | |
2940 | * | |
2941 | * Will be used for merging fiemap extent | |
2942 | */ | |
2943 | struct fiemap_cache { | |
2944 | u64 offset; | |
2945 | u64 phys; | |
2946 | u64 len; | |
2947 | u32 flags; | |
2948 | bool cached; | |
2949 | }; | |
2950 | ||
2951 | /* | |
2952 | * Helper to submit fiemap extent. | |
2953 | * | |
2954 | * Will try to merge current fiemap extent specified by @offset, @phys, | |
2955 | * @len and @flags with cached one. | |
2956 | * And only when we fails to merge, cached one will be submitted as | |
2957 | * fiemap extent. | |
2958 | * | |
2959 | * Return value is the same as fiemap_fill_next_extent(). | |
2960 | */ | |
2961 | static int emit_fiemap_extent(struct fiemap_extent_info *fieinfo, | |
2962 | struct fiemap_cache *cache, | |
2963 | u64 offset, u64 phys, u64 len, u32 flags) | |
2964 | { | |
2965 | int ret = 0; | |
2966 | ||
ac3c0d36 FM |
2967 | /* Set at the end of extent_fiemap(). */ |
2968 | ASSERT((flags & FIEMAP_EXTENT_LAST) == 0); | |
2969 | ||
4751832d QW |
2970 | if (!cache->cached) |
2971 | goto assign; | |
2972 | ||
2973 | /* | |
2974 | * Sanity check, extent_fiemap() should have ensured that new | |
52042d8e | 2975 | * fiemap extent won't overlap with cached one. |
4751832d QW |
2976 | * Not recoverable. |
2977 | * | |
2978 | * NOTE: Physical address can overlap, due to compression | |
2979 | */ | |
2980 | if (cache->offset + cache->len > offset) { | |
2981 | WARN_ON(1); | |
2982 | return -EINVAL; | |
2983 | } | |
2984 | ||
2985 | /* | |
2986 | * Only merges fiemap extents if | |
2987 | * 1) Their logical addresses are continuous | |
2988 | * | |
2989 | * 2) Their physical addresses are continuous | |
2990 | * So truly compressed (physical size smaller than logical size) | |
2991 | * extents won't get merged with each other | |
2992 | * | |
ac3c0d36 | 2993 | * 3) Share same flags |
4751832d QW |
2994 | */ |
2995 | if (cache->offset + cache->len == offset && | |
2996 | cache->phys + cache->len == phys && | |
ac3c0d36 | 2997 | cache->flags == flags) { |
4751832d | 2998 | cache->len += len; |
ac3c0d36 | 2999 | return 0; |
4751832d QW |
3000 | } |
3001 | ||
3002 | /* Not mergeable, need to submit cached one */ | |
3003 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, | |
3004 | cache->len, cache->flags); | |
3005 | cache->cached = false; | |
3006 | if (ret) | |
3007 | return ret; | |
3008 | assign: | |
3009 | cache->cached = true; | |
3010 | cache->offset = offset; | |
3011 | cache->phys = phys; | |
3012 | cache->len = len; | |
3013 | cache->flags = flags; | |
ac3c0d36 FM |
3014 | |
3015 | return 0; | |
4751832d QW |
3016 | } |
3017 | ||
3018 | /* | |
848c23b7 | 3019 | * Emit last fiemap cache |
4751832d | 3020 | * |
848c23b7 QW |
3021 | * The last fiemap cache may still be cached in the following case: |
3022 | * 0 4k 8k | |
3023 | * |<- Fiemap range ->| | |
3024 | * |<------------ First extent ----------->| | |
3025 | * | |
3026 | * In this case, the first extent range will be cached but not emitted. | |
3027 | * So we must emit it before ending extent_fiemap(). | |
4751832d | 3028 | */ |
5c5aff98 | 3029 | static int emit_last_fiemap_cache(struct fiemap_extent_info *fieinfo, |
848c23b7 | 3030 | struct fiemap_cache *cache) |
4751832d QW |
3031 | { |
3032 | int ret; | |
3033 | ||
3034 | if (!cache->cached) | |
3035 | return 0; | |
3036 | ||
4751832d QW |
3037 | ret = fiemap_fill_next_extent(fieinfo, cache->offset, cache->phys, |
3038 | cache->len, cache->flags); | |
3039 | cache->cached = false; | |
3040 | if (ret > 0) | |
3041 | ret = 0; | |
3042 | return ret; | |
3043 | } | |
3044 | ||
ac3c0d36 | 3045 | static int fiemap_next_leaf_item(struct btrfs_inode *inode, struct btrfs_path *path) |
1506fcc8 | 3046 | { |
ac3c0d36 FM |
3047 | struct extent_buffer *clone; |
3048 | struct btrfs_key key; | |
3049 | int slot; | |
3050 | int ret; | |
3051 | ||
3052 | path->slots[0]++; | |
3053 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) | |
3054 | return 0; | |
3055 | ||
3056 | ret = btrfs_next_leaf(inode->root, path); | |
3057 | if (ret != 0) | |
3058 | return ret; | |
3059 | ||
3060 | /* | |
3061 | * Don't bother with cloning if there are no more file extent items for | |
3062 | * our inode. | |
3063 | */ | |
3064 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3065 | if (key.objectid != btrfs_ino(inode) || key.type != BTRFS_EXTENT_DATA_KEY) | |
3066 | return 1; | |
3067 | ||
3068 | /* See the comment at fiemap_search_slot() about why we clone. */ | |
3069 | clone = btrfs_clone_extent_buffer(path->nodes[0]); | |
3070 | if (!clone) | |
3071 | return -ENOMEM; | |
3072 | ||
3073 | slot = path->slots[0]; | |
3074 | btrfs_release_path(path); | |
3075 | path->nodes[0] = clone; | |
3076 | path->slots[0] = slot; | |
3077 | ||
3078 | return 0; | |
3079 | } | |
3080 | ||
3081 | /* | |
3082 | * Search for the first file extent item that starts at a given file offset or | |
3083 | * the one that starts immediately before that offset. | |
3084 | * Returns: 0 on success, < 0 on error, 1 if not found. | |
3085 | */ | |
3086 | static int fiemap_search_slot(struct btrfs_inode *inode, struct btrfs_path *path, | |
3087 | u64 file_offset) | |
3088 | { | |
3089 | const u64 ino = btrfs_ino(inode); | |
facee0a0 | 3090 | struct btrfs_root *root = inode->root; |
ac3c0d36 FM |
3091 | struct extent_buffer *clone; |
3092 | struct btrfs_key key; | |
3093 | int slot; | |
3094 | int ret; | |
1506fcc8 | 3095 | |
ac3c0d36 FM |
3096 | key.objectid = ino; |
3097 | key.type = BTRFS_EXTENT_DATA_KEY; | |
3098 | key.offset = file_offset; | |
3099 | ||
3100 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
3101 | if (ret < 0) | |
3102 | return ret; | |
3103 | ||
3104 | if (ret > 0 && path->slots[0] > 0) { | |
3105 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0] - 1); | |
3106 | if (key.objectid == ino && key.type == BTRFS_EXTENT_DATA_KEY) | |
3107 | path->slots[0]--; | |
3108 | } | |
3109 | ||
3110 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { | |
3111 | ret = btrfs_next_leaf(root, path); | |
3112 | if (ret != 0) | |
3113 | return ret; | |
3114 | ||
3115 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
3116 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
3117 | return 1; | |
5911c8fe DS |
3118 | } |
3119 | ||
15c7745c | 3120 | /* |
ac3c0d36 FM |
3121 | * We clone the leaf and use it during fiemap. This is because while |
3122 | * using the leaf we do expensive things like checking if an extent is | |
3123 | * shared, which can take a long time. In order to prevent blocking | |
3124 | * other tasks for too long, we use a clone of the leaf. We have locked | |
3125 | * the file range in the inode's io tree, so we know none of our file | |
3126 | * extent items can change. This way we avoid blocking other tasks that | |
3127 | * want to insert items for other inodes in the same leaf or b+tree | |
3128 | * rebalance operations (triggered for example when someone is trying | |
3129 | * to push items into this leaf when trying to insert an item in a | |
3130 | * neighbour leaf). | |
3131 | * We also need the private clone because holding a read lock on an | |
3132 | * extent buffer of the subvolume's b+tree will make lockdep unhappy | |
3133 | * when we call fiemap_fill_next_extent(), because that may cause a page | |
3134 | * fault when filling the user space buffer with fiemap data. | |
15c7745c | 3135 | */ |
ac3c0d36 FM |
3136 | clone = btrfs_clone_extent_buffer(path->nodes[0]); |
3137 | if (!clone) | |
3138 | return -ENOMEM; | |
3139 | ||
3140 | slot = path->slots[0]; | |
3141 | btrfs_release_path(path); | |
3142 | path->nodes[0] = clone; | |
3143 | path->slots[0] = slot; | |
3144 | ||
3145 | return 0; | |
3146 | } | |
3147 | ||
3148 | /* | |
3149 | * Process a range which is a hole or a prealloc extent in the inode's subvolume | |
3150 | * btree. If @disk_bytenr is 0, we are dealing with a hole, otherwise a prealloc | |
3151 | * extent. The end offset (@end) is inclusive. | |
3152 | */ | |
3153 | static int fiemap_process_hole(struct btrfs_inode *inode, | |
3154 | struct fiemap_extent_info *fieinfo, | |
3155 | struct fiemap_cache *cache, | |
b3e744fe | 3156 | struct extent_state **delalloc_cached_state, |
61dbb952 | 3157 | struct btrfs_backref_share_check_ctx *backref_ctx, |
ac3c0d36 FM |
3158 | u64 disk_bytenr, u64 extent_offset, |
3159 | u64 extent_gen, | |
ac3c0d36 FM |
3160 | u64 start, u64 end) |
3161 | { | |
3162 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
ac3c0d36 FM |
3163 | u64 cur_offset = start; |
3164 | u64 last_delalloc_end = 0; | |
3165 | u32 prealloc_flags = FIEMAP_EXTENT_UNWRITTEN; | |
3166 | bool checked_extent_shared = false; | |
3167 | int ret; | |
4d479cf0 | 3168 | |
ec29ed5b | 3169 | /* |
ac3c0d36 FM |
3170 | * There can be no delalloc past i_size, so don't waste time looking for |
3171 | * it beyond i_size. | |
ec29ed5b | 3172 | */ |
ac3c0d36 FM |
3173 | while (cur_offset < end && cur_offset < i_size) { |
3174 | u64 delalloc_start; | |
3175 | u64 delalloc_end; | |
3176 | u64 prealloc_start; | |
3177 | u64 prealloc_len = 0; | |
3178 | bool delalloc; | |
3179 | ||
3180 | delalloc = btrfs_find_delalloc_in_range(inode, cur_offset, end, | |
b3e744fe | 3181 | delalloc_cached_state, |
ac3c0d36 FM |
3182 | &delalloc_start, |
3183 | &delalloc_end); | |
3184 | if (!delalloc) | |
3185 | break; | |
2d324f59 | 3186 | |
ec29ed5b | 3187 | /* |
ac3c0d36 FM |
3188 | * If this is a prealloc extent we have to report every section |
3189 | * of it that has no delalloc. | |
ec29ed5b | 3190 | */ |
ac3c0d36 FM |
3191 | if (disk_bytenr != 0) { |
3192 | if (last_delalloc_end == 0) { | |
3193 | prealloc_start = start; | |
3194 | prealloc_len = delalloc_start - start; | |
3195 | } else { | |
3196 | prealloc_start = last_delalloc_end + 1; | |
3197 | prealloc_len = delalloc_start - prealloc_start; | |
3198 | } | |
3199 | } | |
3200 | ||
3201 | if (prealloc_len > 0) { | |
3202 | if (!checked_extent_shared && fieinfo->fi_extents_max) { | |
ceb707da | 3203 | ret = btrfs_is_data_extent_shared(inode, |
84a7949d FM |
3204 | disk_bytenr, |
3205 | extent_gen, | |
3206 | backref_ctx); | |
ac3c0d36 FM |
3207 | if (ret < 0) |
3208 | return ret; | |
3209 | else if (ret > 0) | |
3210 | prealloc_flags |= FIEMAP_EXTENT_SHARED; | |
3211 | ||
3212 | checked_extent_shared = true; | |
3213 | } | |
3214 | ret = emit_fiemap_extent(fieinfo, cache, prealloc_start, | |
3215 | disk_bytenr + extent_offset, | |
3216 | prealloc_len, prealloc_flags); | |
3217 | if (ret) | |
3218 | return ret; | |
3219 | extent_offset += prealloc_len; | |
3220 | } | |
3221 | ||
3222 | ret = emit_fiemap_extent(fieinfo, cache, delalloc_start, 0, | |
3223 | delalloc_end + 1 - delalloc_start, | |
3224 | FIEMAP_EXTENT_DELALLOC | | |
3225 | FIEMAP_EXTENT_UNKNOWN); | |
3226 | if (ret) | |
3227 | return ret; | |
3228 | ||
3229 | last_delalloc_end = delalloc_end; | |
3230 | cur_offset = delalloc_end + 1; | |
3231 | extent_offset += cur_offset - delalloc_start; | |
3232 | cond_resched(); | |
3233 | } | |
3234 | ||
3235 | /* | |
3236 | * Either we found no delalloc for the whole prealloc extent or we have | |
3237 | * a prealloc extent that spans i_size or starts at or after i_size. | |
3238 | */ | |
3239 | if (disk_bytenr != 0 && last_delalloc_end < end) { | |
3240 | u64 prealloc_start; | |
3241 | u64 prealloc_len; | |
3242 | ||
3243 | if (last_delalloc_end == 0) { | |
3244 | prealloc_start = start; | |
3245 | prealloc_len = end + 1 - start; | |
3246 | } else { | |
3247 | prealloc_start = last_delalloc_end + 1; | |
3248 | prealloc_len = end + 1 - prealloc_start; | |
3249 | } | |
3250 | ||
3251 | if (!checked_extent_shared && fieinfo->fi_extents_max) { | |
ceb707da FM |
3252 | ret = btrfs_is_data_extent_shared(inode, |
3253 | disk_bytenr, | |
84a7949d | 3254 | extent_gen, |
61dbb952 | 3255 | backref_ctx); |
ac3c0d36 FM |
3256 | if (ret < 0) |
3257 | return ret; | |
3258 | else if (ret > 0) | |
3259 | prealloc_flags |= FIEMAP_EXTENT_SHARED; | |
3260 | } | |
3261 | ret = emit_fiemap_extent(fieinfo, cache, prealloc_start, | |
3262 | disk_bytenr + extent_offset, | |
3263 | prealloc_len, prealloc_flags); | |
3264 | if (ret) | |
3265 | return ret; | |
3266 | } | |
3267 | ||
3268 | return 0; | |
3269 | } | |
3270 | ||
3271 | static int fiemap_find_last_extent_offset(struct btrfs_inode *inode, | |
3272 | struct btrfs_path *path, | |
3273 | u64 *last_extent_end_ret) | |
3274 | { | |
3275 | const u64 ino = btrfs_ino(inode); | |
3276 | struct btrfs_root *root = inode->root; | |
3277 | struct extent_buffer *leaf; | |
3278 | struct btrfs_file_extent_item *ei; | |
3279 | struct btrfs_key key; | |
3280 | u64 disk_bytenr; | |
3281 | int ret; | |
3282 | ||
3283 | /* | |
3284 | * Lookup the last file extent. We're not using i_size here because | |
3285 | * there might be preallocation past i_size. | |
3286 | */ | |
3287 | ret = btrfs_lookup_file_extent(NULL, root, path, ino, (u64)-1, 0); | |
3288 | /* There can't be a file extent item at offset (u64)-1 */ | |
3289 | ASSERT(ret != 0); | |
3290 | if (ret < 0) | |
3291 | return ret; | |
3292 | ||
3293 | /* | |
3294 | * For a non-existing key, btrfs_search_slot() always leaves us at a | |
3295 | * slot > 0, except if the btree is empty, which is impossible because | |
3296 | * at least it has the inode item for this inode and all the items for | |
3297 | * the root inode 256. | |
3298 | */ | |
3299 | ASSERT(path->slots[0] > 0); | |
3300 | path->slots[0]--; | |
3301 | leaf = path->nodes[0]; | |
3302 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3303 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) { | |
3304 | /* No file extent items in the subvolume tree. */ | |
3305 | *last_extent_end_ret = 0; | |
3306 | return 0; | |
975f84fe | 3307 | } |
975f84fe | 3308 | |
ec29ed5b | 3309 | /* |
ac3c0d36 FM |
3310 | * For an inline extent, the disk_bytenr is where inline data starts at, |
3311 | * so first check if we have an inline extent item before checking if we | |
3312 | * have an implicit hole (disk_bytenr == 0). | |
ec29ed5b | 3313 | */ |
ac3c0d36 FM |
3314 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item); |
3315 | if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_INLINE) { | |
3316 | *last_extent_end_ret = btrfs_file_extent_end(path); | |
3317 | return 0; | |
ec29ed5b CM |
3318 | } |
3319 | ||
ac3c0d36 FM |
3320 | /* |
3321 | * Find the last file extent item that is not a hole (when NO_HOLES is | |
3322 | * not enabled). This should take at most 2 iterations in the worst | |
3323 | * case: we have one hole file extent item at slot 0 of a leaf and | |
3324 | * another hole file extent item as the last item in the previous leaf. | |
3325 | * This is because we merge file extent items that represent holes. | |
3326 | */ | |
3327 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3328 | while (disk_bytenr == 0) { | |
3329 | ret = btrfs_previous_item(root, path, ino, BTRFS_EXTENT_DATA_KEY); | |
3330 | if (ret < 0) { | |
3331 | return ret; | |
3332 | } else if (ret > 0) { | |
3333 | /* No file extent items that are not holes. */ | |
3334 | *last_extent_end_ret = 0; | |
3335 | return 0; | |
3336 | } | |
3337 | leaf = path->nodes[0]; | |
3338 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
3339 | struct btrfs_file_extent_item); | |
3340 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3341 | } | |
ec29ed5b | 3342 | |
ac3c0d36 FM |
3343 | *last_extent_end_ret = btrfs_file_extent_end(path); |
3344 | return 0; | |
3345 | } | |
3346 | ||
3347 | int extent_fiemap(struct btrfs_inode *inode, struct fiemap_extent_info *fieinfo, | |
3348 | u64 start, u64 len) | |
3349 | { | |
3350 | const u64 ino = btrfs_ino(inode); | |
3351 | struct extent_state *cached_state = NULL; | |
b3e744fe | 3352 | struct extent_state *delalloc_cached_state = NULL; |
ac3c0d36 | 3353 | struct btrfs_path *path; |
ac3c0d36 | 3354 | struct fiemap_cache cache = { 0 }; |
61dbb952 | 3355 | struct btrfs_backref_share_check_ctx *backref_ctx; |
ac3c0d36 FM |
3356 | u64 last_extent_end; |
3357 | u64 prev_extent_end; | |
3358 | u64 lockstart; | |
3359 | u64 lockend; | |
3360 | bool stopped = false; | |
3361 | int ret; | |
3362 | ||
84a7949d | 3363 | backref_ctx = btrfs_alloc_backref_share_check_ctx(); |
ac3c0d36 | 3364 | path = btrfs_alloc_path(); |
84a7949d | 3365 | if (!backref_ctx || !path) { |
ac3c0d36 | 3366 | ret = -ENOMEM; |
1506fcc8 YS |
3367 | goto out; |
3368 | } | |
975f84fe | 3369 | |
ceb707da FM |
3370 | lockstart = round_down(start, inode->root->fs_info->sectorsize); |
3371 | lockend = round_up(start + len, inode->root->fs_info->sectorsize); | |
ac3c0d36 | 3372 | prev_extent_end = lockstart; |
ea8efc74 | 3373 | |
519b7e13 | 3374 | btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED); |
570eb97b | 3375 | lock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
ea8efc74 | 3376 | |
ac3c0d36 FM |
3377 | ret = fiemap_find_last_extent_offset(inode, path, &last_extent_end); |
3378 | if (ret < 0) | |
3379 | goto out_unlock; | |
3380 | btrfs_release_path(path); | |
1506fcc8 | 3381 | |
ac3c0d36 FM |
3382 | path->reada = READA_FORWARD; |
3383 | ret = fiemap_search_slot(inode, path, lockstart); | |
3384 | if (ret < 0) { | |
3385 | goto out_unlock; | |
3386 | } else if (ret > 0) { | |
ea8efc74 | 3387 | /* |
ac3c0d36 FM |
3388 | * No file extent item found, but we may have delalloc between |
3389 | * the current offset and i_size. So check for that. | |
ea8efc74 | 3390 | */ |
ac3c0d36 FM |
3391 | ret = 0; |
3392 | goto check_eof_delalloc; | |
3393 | } | |
3394 | ||
3395 | while (prev_extent_end < lockend) { | |
3396 | struct extent_buffer *leaf = path->nodes[0]; | |
3397 | struct btrfs_file_extent_item *ei; | |
3398 | struct btrfs_key key; | |
3399 | u64 extent_end; | |
3400 | u64 extent_len; | |
3401 | u64 extent_offset = 0; | |
3402 | u64 extent_gen; | |
3403 | u64 disk_bytenr = 0; | |
3404 | u64 flags = 0; | |
3405 | int extent_type; | |
3406 | u8 compression; | |
3407 | ||
3408 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
3409 | if (key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY) | |
3410 | break; | |
3411 | ||
3412 | extent_end = btrfs_file_extent_end(path); | |
1506fcc8 | 3413 | |
ea8efc74 | 3414 | /* |
ac3c0d36 FM |
3415 | * The first iteration can leave us at an extent item that ends |
3416 | * before our range's start. Move to the next item. | |
ea8efc74 | 3417 | */ |
ac3c0d36 FM |
3418 | if (extent_end <= lockstart) |
3419 | goto next_item; | |
fe09e16c | 3420 | |
877c1476 FM |
3421 | backref_ctx->curr_leaf_bytenr = leaf->start; |
3422 | ||
ac3c0d36 FM |
3423 | /* We have in implicit hole (NO_HOLES feature enabled). */ |
3424 | if (prev_extent_end < key.offset) { | |
3425 | const u64 range_end = min(key.offset, lockend) - 1; | |
b8f164e3 | 3426 | |
ac3c0d36 | 3427 | ret = fiemap_process_hole(inode, fieinfo, &cache, |
b3e744fe | 3428 | &delalloc_cached_state, |
61dbb952 | 3429 | backref_ctx, 0, 0, 0, |
ac3c0d36 FM |
3430 | prev_extent_end, range_end); |
3431 | if (ret < 0) { | |
3432 | goto out_unlock; | |
3433 | } else if (ret > 0) { | |
3434 | /* fiemap_fill_next_extent() told us to stop. */ | |
3435 | stopped = true; | |
3436 | break; | |
3437 | } | |
1506fcc8 | 3438 | |
ac3c0d36 FM |
3439 | /* We've reached the end of the fiemap range, stop. */ |
3440 | if (key.offset >= lockend) { | |
3441 | stopped = true; | |
3442 | break; | |
3443 | } | |
1506fcc8 YS |
3444 | } |
3445 | ||
ac3c0d36 FM |
3446 | extent_len = extent_end - key.offset; |
3447 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
3448 | struct btrfs_file_extent_item); | |
3449 | compression = btrfs_file_extent_compression(leaf, ei); | |
3450 | extent_type = btrfs_file_extent_type(leaf, ei); | |
3451 | extent_gen = btrfs_file_extent_generation(leaf, ei); | |
3452 | ||
3453 | if (extent_type != BTRFS_FILE_EXTENT_INLINE) { | |
3454 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, ei); | |
3455 | if (compression == BTRFS_COMPRESS_NONE) | |
3456 | extent_offset = btrfs_file_extent_offset(leaf, ei); | |
ec29ed5b | 3457 | } |
ac3c0d36 FM |
3458 | |
3459 | if (compression != BTRFS_COMPRESS_NONE) | |
3460 | flags |= FIEMAP_EXTENT_ENCODED; | |
3461 | ||
3462 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { | |
3463 | flags |= FIEMAP_EXTENT_DATA_INLINE; | |
3464 | flags |= FIEMAP_EXTENT_NOT_ALIGNED; | |
3465 | ret = emit_fiemap_extent(fieinfo, &cache, key.offset, 0, | |
3466 | extent_len, flags); | |
3467 | } else if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) { | |
3468 | ret = fiemap_process_hole(inode, fieinfo, &cache, | |
b3e744fe | 3469 | &delalloc_cached_state, |
61dbb952 | 3470 | backref_ctx, |
ac3c0d36 | 3471 | disk_bytenr, extent_offset, |
84a7949d FM |
3472 | extent_gen, key.offset, |
3473 | extent_end - 1); | |
ac3c0d36 FM |
3474 | } else if (disk_bytenr == 0) { |
3475 | /* We have an explicit hole. */ | |
3476 | ret = fiemap_process_hole(inode, fieinfo, &cache, | |
b3e744fe | 3477 | &delalloc_cached_state, |
61dbb952 | 3478 | backref_ctx, 0, 0, 0, |
ac3c0d36 FM |
3479 | key.offset, extent_end - 1); |
3480 | } else { | |
3481 | /* We have a regular extent. */ | |
3482 | if (fieinfo->fi_extents_max) { | |
ceb707da | 3483 | ret = btrfs_is_data_extent_shared(inode, |
ac3c0d36 FM |
3484 | disk_bytenr, |
3485 | extent_gen, | |
61dbb952 | 3486 | backref_ctx); |
ac3c0d36 FM |
3487 | if (ret < 0) |
3488 | goto out_unlock; | |
3489 | else if (ret > 0) | |
3490 | flags |= FIEMAP_EXTENT_SHARED; | |
3491 | } | |
3492 | ||
3493 | ret = emit_fiemap_extent(fieinfo, &cache, key.offset, | |
3494 | disk_bytenr + extent_offset, | |
3495 | extent_len, flags); | |
975f84fe | 3496 | } |
ac3c0d36 FM |
3497 | |
3498 | if (ret < 0) { | |
3499 | goto out_unlock; | |
3500 | } else if (ret > 0) { | |
3501 | /* fiemap_fill_next_extent() told us to stop. */ | |
3502 | stopped = true; | |
3503 | break; | |
26e726af | 3504 | } |
09fbc1c8 | 3505 | |
ac3c0d36 FM |
3506 | prev_extent_end = extent_end; |
3507 | next_item: | |
09fbc1c8 FM |
3508 | if (fatal_signal_pending(current)) { |
3509 | ret = -EINTR; | |
ac3c0d36 | 3510 | goto out_unlock; |
09fbc1c8 | 3511 | } |
ac3c0d36 FM |
3512 | |
3513 | ret = fiemap_next_leaf_item(inode, path); | |
3514 | if (ret < 0) { | |
3515 | goto out_unlock; | |
3516 | } else if (ret > 0) { | |
3517 | /* No more file extent items for this inode. */ | |
3518 | break; | |
3519 | } | |
3520 | cond_resched(); | |
1506fcc8 | 3521 | } |
5911c8fe | 3522 | |
ac3c0d36 FM |
3523 | check_eof_delalloc: |
3524 | /* | |
3525 | * Release (and free) the path before emitting any final entries to | |
3526 | * fiemap_fill_next_extent() to keep lockdep happy. This is because | |
3527 | * once we find no more file extent items exist, we may have a | |
3528 | * non-cloned leaf, and fiemap_fill_next_extent() can trigger page | |
3529 | * faults when copying data to the user space buffer. | |
3530 | */ | |
3531 | btrfs_free_path(path); | |
3532 | path = NULL; | |
3533 | ||
3534 | if (!stopped && prev_extent_end < lockend) { | |
b3e744fe FM |
3535 | ret = fiemap_process_hole(inode, fieinfo, &cache, |
3536 | &delalloc_cached_state, backref_ctx, | |
84a7949d | 3537 | 0, 0, 0, prev_extent_end, lockend - 1); |
ac3c0d36 FM |
3538 | if (ret < 0) |
3539 | goto out_unlock; | |
3540 | prev_extent_end = lockend; | |
3541 | } | |
3542 | ||
3543 | if (cache.cached && cache.offset + cache.len >= last_extent_end) { | |
3544 | const u64 i_size = i_size_read(&inode->vfs_inode); | |
3545 | ||
3546 | if (prev_extent_end < i_size) { | |
3547 | u64 delalloc_start; | |
3548 | u64 delalloc_end; | |
3549 | bool delalloc; | |
3550 | ||
3551 | delalloc = btrfs_find_delalloc_in_range(inode, | |
3552 | prev_extent_end, | |
3553 | i_size - 1, | |
b3e744fe | 3554 | &delalloc_cached_state, |
ac3c0d36 FM |
3555 | &delalloc_start, |
3556 | &delalloc_end); | |
3557 | if (!delalloc) | |
3558 | cache.flags |= FIEMAP_EXTENT_LAST; | |
3559 | } else { | |
3560 | cache.flags |= FIEMAP_EXTENT_LAST; | |
3561 | } | |
3562 | } | |
3563 | ||
3564 | ret = emit_last_fiemap_cache(fieinfo, &cache); | |
3565 | ||
3566 | out_unlock: | |
570eb97b | 3567 | unlock_extent(&inode->io_tree, lockstart, lockend, &cached_state); |
519b7e13 | 3568 | btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED); |
ac3c0d36 | 3569 | out: |
b3e744fe | 3570 | free_extent_state(delalloc_cached_state); |
84a7949d | 3571 | btrfs_free_backref_share_ctx(backref_ctx); |
e02d48ea | 3572 | btrfs_free_path(path); |
1506fcc8 YS |
3573 | return ret; |
3574 | } | |
3575 | ||
727011e0 CM |
3576 | static void __free_extent_buffer(struct extent_buffer *eb) |
3577 | { | |
727011e0 CM |
3578 | kmem_cache_free(extent_buffer_cache, eb); |
3579 | } | |
3580 | ||
2b48966a | 3581 | int extent_buffer_under_io(const struct extent_buffer *eb) |
db7f3436 JB |
3582 | { |
3583 | return (atomic_read(&eb->io_pages) || | |
3584 | test_bit(EXTENT_BUFFER_WRITEBACK, &eb->bflags) || | |
3585 | test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
3586 | } | |
3587 | ||
8ff8466d | 3588 | static bool page_range_has_eb(struct btrfs_fs_info *fs_info, struct page *page) |
db7f3436 | 3589 | { |
8ff8466d | 3590 | struct btrfs_subpage *subpage; |
db7f3436 | 3591 | |
8ff8466d | 3592 | lockdep_assert_held(&page->mapping->private_lock); |
db7f3436 | 3593 | |
8ff8466d QW |
3594 | if (PagePrivate(page)) { |
3595 | subpage = (struct btrfs_subpage *)page->private; | |
3596 | if (atomic_read(&subpage->eb_refs)) | |
3597 | return true; | |
3d078efa QW |
3598 | /* |
3599 | * Even there is no eb refs here, we may still have | |
3600 | * end_page_read() call relying on page::private. | |
3601 | */ | |
3602 | if (atomic_read(&subpage->readers)) | |
3603 | return true; | |
8ff8466d QW |
3604 | } |
3605 | return false; | |
3606 | } | |
db7f3436 | 3607 | |
8ff8466d QW |
3608 | static void detach_extent_buffer_page(struct extent_buffer *eb, struct page *page) |
3609 | { | |
3610 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
3611 | const bool mapped = !test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); | |
3612 | ||
3613 | /* | |
3614 | * For mapped eb, we're going to change the page private, which should | |
3615 | * be done under the private_lock. | |
3616 | */ | |
3617 | if (mapped) | |
3618 | spin_lock(&page->mapping->private_lock); | |
3619 | ||
3620 | if (!PagePrivate(page)) { | |
5d2361db | 3621 | if (mapped) |
8ff8466d QW |
3622 | spin_unlock(&page->mapping->private_lock); |
3623 | return; | |
3624 | } | |
3625 | ||
fbca46eb | 3626 | if (fs_info->nodesize >= PAGE_SIZE) { |
5d2361db FL |
3627 | /* |
3628 | * We do this since we'll remove the pages after we've | |
3629 | * removed the eb from the radix tree, so we could race | |
3630 | * and have this page now attached to the new eb. So | |
3631 | * only clear page_private if it's still connected to | |
3632 | * this eb. | |
3633 | */ | |
3634 | if (PagePrivate(page) && | |
3635 | page->private == (unsigned long)eb) { | |
3636 | BUG_ON(test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); | |
3637 | BUG_ON(PageDirty(page)); | |
3638 | BUG_ON(PageWriteback(page)); | |
db7f3436 | 3639 | /* |
5d2361db FL |
3640 | * We need to make sure we haven't be attached |
3641 | * to a new eb. | |
db7f3436 | 3642 | */ |
d1b89bc0 | 3643 | detach_page_private(page); |
db7f3436 | 3644 | } |
5d2361db FL |
3645 | if (mapped) |
3646 | spin_unlock(&page->mapping->private_lock); | |
8ff8466d QW |
3647 | return; |
3648 | } | |
3649 | ||
3650 | /* | |
3651 | * For subpage, we can have dummy eb with page private. In this case, | |
3652 | * we can directly detach the private as such page is only attached to | |
3653 | * one dummy eb, no sharing. | |
3654 | */ | |
3655 | if (!mapped) { | |
3656 | btrfs_detach_subpage(fs_info, page); | |
3657 | return; | |
3658 | } | |
3659 | ||
3660 | btrfs_page_dec_eb_refs(fs_info, page); | |
3661 | ||
3662 | /* | |
3663 | * We can only detach the page private if there are no other ebs in the | |
3d078efa | 3664 | * page range and no unfinished IO. |
8ff8466d QW |
3665 | */ |
3666 | if (!page_range_has_eb(fs_info, page)) | |
3667 | btrfs_detach_subpage(fs_info, page); | |
3668 | ||
3669 | spin_unlock(&page->mapping->private_lock); | |
3670 | } | |
3671 | ||
3672 | /* Release all pages attached to the extent buffer */ | |
3673 | static void btrfs_release_extent_buffer_pages(struct extent_buffer *eb) | |
3674 | { | |
3675 | int i; | |
3676 | int num_pages; | |
3677 | ||
3678 | ASSERT(!extent_buffer_under_io(eb)); | |
3679 | ||
3680 | num_pages = num_extent_pages(eb); | |
3681 | for (i = 0; i < num_pages; i++) { | |
3682 | struct page *page = eb->pages[i]; | |
3683 | ||
3684 | if (!page) | |
3685 | continue; | |
3686 | ||
3687 | detach_extent_buffer_page(eb, page); | |
5d2361db | 3688 | |
01327610 | 3689 | /* One for when we allocated the page */ |
09cbfeaf | 3690 | put_page(page); |
d64766fd | 3691 | } |
db7f3436 JB |
3692 | } |
3693 | ||
3694 | /* | |
3695 | * Helper for releasing the extent buffer. | |
3696 | */ | |
3697 | static inline void btrfs_release_extent_buffer(struct extent_buffer *eb) | |
3698 | { | |
55ac0139 | 3699 | btrfs_release_extent_buffer_pages(eb); |
a40246e8 | 3700 | btrfs_leak_debug_del_eb(eb); |
db7f3436 JB |
3701 | __free_extent_buffer(eb); |
3702 | } | |
3703 | ||
f28491e0 JB |
3704 | static struct extent_buffer * |
3705 | __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, | |
23d79d81 | 3706 | unsigned long len) |
d1310b2e CM |
3707 | { |
3708 | struct extent_buffer *eb = NULL; | |
3709 | ||
d1b5c567 | 3710 | eb = kmem_cache_zalloc(extent_buffer_cache, GFP_NOFS|__GFP_NOFAIL); |
d1310b2e CM |
3711 | eb->start = start; |
3712 | eb->len = len; | |
f28491e0 | 3713 | eb->fs_info = fs_info; |
196d59ab | 3714 | init_rwsem(&eb->lock); |
b4ce94de | 3715 | |
a40246e8 | 3716 | btrfs_leak_debug_add_eb(eb); |
d3575156 | 3717 | INIT_LIST_HEAD(&eb->release_list); |
6d49ba1b | 3718 | |
3083ee2e | 3719 | spin_lock_init(&eb->refs_lock); |
d1310b2e | 3720 | atomic_set(&eb->refs, 1); |
0b32f4bb | 3721 | atomic_set(&eb->io_pages, 0); |
727011e0 | 3722 | |
deb67895 | 3723 | ASSERT(len <= BTRFS_MAX_METADATA_BLOCKSIZE); |
d1310b2e CM |
3724 | |
3725 | return eb; | |
3726 | } | |
3727 | ||
2b48966a | 3728 | struct extent_buffer *btrfs_clone_extent_buffer(const struct extent_buffer *src) |
815a51c7 | 3729 | { |
cc5e31a4 | 3730 | int i; |
815a51c7 | 3731 | struct extent_buffer *new; |
cc5e31a4 | 3732 | int num_pages = num_extent_pages(src); |
dd137dd1 | 3733 | int ret; |
815a51c7 | 3734 | |
3f556f78 | 3735 | new = __alloc_extent_buffer(src->fs_info, src->start, src->len); |
815a51c7 JS |
3736 | if (new == NULL) |
3737 | return NULL; | |
3738 | ||
62c053fb QW |
3739 | /* |
3740 | * Set UNMAPPED before calling btrfs_release_extent_buffer(), as | |
3741 | * btrfs_release_extent_buffer() have different behavior for | |
3742 | * UNMAPPED subpage extent buffer. | |
3743 | */ | |
3744 | set_bit(EXTENT_BUFFER_UNMAPPED, &new->bflags); | |
3745 | ||
dd137dd1 STD |
3746 | ret = btrfs_alloc_page_array(num_pages, new->pages); |
3747 | if (ret) { | |
3748 | btrfs_release_extent_buffer(new); | |
3749 | return NULL; | |
3750 | } | |
3751 | ||
815a51c7 | 3752 | for (i = 0; i < num_pages; i++) { |
760f991f | 3753 | int ret; |
dd137dd1 | 3754 | struct page *p = new->pages[i]; |
760f991f | 3755 | |
760f991f QW |
3756 | ret = attach_extent_buffer_page(new, p, NULL); |
3757 | if (ret < 0) { | |
760f991f QW |
3758 | btrfs_release_extent_buffer(new); |
3759 | return NULL; | |
3760 | } | |
815a51c7 | 3761 | WARN_ON(PageDirty(p)); |
fba1acf9 | 3762 | copy_page(page_address(p), page_address(src->pages[i])); |
815a51c7 | 3763 | } |
92d83e94 | 3764 | set_extent_buffer_uptodate(new); |
815a51c7 JS |
3765 | |
3766 | return new; | |
3767 | } | |
3768 | ||
0f331229 OS |
3769 | struct extent_buffer *__alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
3770 | u64 start, unsigned long len) | |
815a51c7 JS |
3771 | { |
3772 | struct extent_buffer *eb; | |
cc5e31a4 DS |
3773 | int num_pages; |
3774 | int i; | |
dd137dd1 | 3775 | int ret; |
815a51c7 | 3776 | |
3f556f78 | 3777 | eb = __alloc_extent_buffer(fs_info, start, len); |
815a51c7 JS |
3778 | if (!eb) |
3779 | return NULL; | |
3780 | ||
65ad0104 | 3781 | num_pages = num_extent_pages(eb); |
dd137dd1 STD |
3782 | ret = btrfs_alloc_page_array(num_pages, eb->pages); |
3783 | if (ret) | |
3784 | goto err; | |
3785 | ||
815a51c7 | 3786 | for (i = 0; i < num_pages; i++) { |
dd137dd1 | 3787 | struct page *p = eb->pages[i]; |
09bc1f0f | 3788 | |
dd137dd1 | 3789 | ret = attach_extent_buffer_page(eb, p, NULL); |
09bc1f0f QW |
3790 | if (ret < 0) |
3791 | goto err; | |
815a51c7 | 3792 | } |
dd137dd1 | 3793 | |
815a51c7 JS |
3794 | set_extent_buffer_uptodate(eb); |
3795 | btrfs_set_header_nritems(eb, 0); | |
b0132a3b | 3796 | set_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags); |
815a51c7 JS |
3797 | |
3798 | return eb; | |
3799 | err: | |
dd137dd1 STD |
3800 | for (i = 0; i < num_pages; i++) { |
3801 | if (eb->pages[i]) { | |
3802 | detach_extent_buffer_page(eb, eb->pages[i]); | |
3803 | __free_page(eb->pages[i]); | |
3804 | } | |
09bc1f0f | 3805 | } |
815a51c7 JS |
3806 | __free_extent_buffer(eb); |
3807 | return NULL; | |
3808 | } | |
3809 | ||
0f331229 | 3810 | struct extent_buffer *alloc_dummy_extent_buffer(struct btrfs_fs_info *fs_info, |
da17066c | 3811 | u64 start) |
0f331229 | 3812 | { |
da17066c | 3813 | return __alloc_dummy_extent_buffer(fs_info, start, fs_info->nodesize); |
0f331229 OS |
3814 | } |
3815 | ||
0b32f4bb JB |
3816 | static void check_buffer_tree_ref(struct extent_buffer *eb) |
3817 | { | |
242e18c7 | 3818 | int refs; |
6bf9cd2e BB |
3819 | /* |
3820 | * The TREE_REF bit is first set when the extent_buffer is added | |
3821 | * to the radix tree. It is also reset, if unset, when a new reference | |
3822 | * is created by find_extent_buffer. | |
0b32f4bb | 3823 | * |
6bf9cd2e BB |
3824 | * It is only cleared in two cases: freeing the last non-tree |
3825 | * reference to the extent_buffer when its STALE bit is set or | |
f913cff3 | 3826 | * calling release_folio when the tree reference is the only reference. |
0b32f4bb | 3827 | * |
6bf9cd2e | 3828 | * In both cases, care is taken to ensure that the extent_buffer's |
f913cff3 | 3829 | * pages are not under io. However, release_folio can be concurrently |
6bf9cd2e BB |
3830 | * called with creating new references, which is prone to race |
3831 | * conditions between the calls to check_buffer_tree_ref in those | |
3832 | * codepaths and clearing TREE_REF in try_release_extent_buffer. | |
0b32f4bb | 3833 | * |
6bf9cd2e BB |
3834 | * The actual lifetime of the extent_buffer in the radix tree is |
3835 | * adequately protected by the refcount, but the TREE_REF bit and | |
3836 | * its corresponding reference are not. To protect against this | |
3837 | * class of races, we call check_buffer_tree_ref from the codepaths | |
3838 | * which trigger io after they set eb->io_pages. Note that once io is | |
3839 | * initiated, TREE_REF can no longer be cleared, so that is the | |
3840 | * moment at which any such race is best fixed. | |
0b32f4bb | 3841 | */ |
242e18c7 CM |
3842 | refs = atomic_read(&eb->refs); |
3843 | if (refs >= 2 && test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
3844 | return; | |
3845 | ||
594831c4 JB |
3846 | spin_lock(&eb->refs_lock); |
3847 | if (!test_and_set_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) | |
0b32f4bb | 3848 | atomic_inc(&eb->refs); |
594831c4 | 3849 | spin_unlock(&eb->refs_lock); |
0b32f4bb JB |
3850 | } |
3851 | ||
2457aec6 MG |
3852 | static void mark_extent_buffer_accessed(struct extent_buffer *eb, |
3853 | struct page *accessed) | |
5df4235e | 3854 | { |
cc5e31a4 | 3855 | int num_pages, i; |
5df4235e | 3856 | |
0b32f4bb JB |
3857 | check_buffer_tree_ref(eb); |
3858 | ||
65ad0104 | 3859 | num_pages = num_extent_pages(eb); |
5df4235e | 3860 | for (i = 0; i < num_pages; i++) { |
fb85fc9a DS |
3861 | struct page *p = eb->pages[i]; |
3862 | ||
2457aec6 MG |
3863 | if (p != accessed) |
3864 | mark_page_accessed(p); | |
5df4235e JB |
3865 | } |
3866 | } | |
3867 | ||
f28491e0 JB |
3868 | struct extent_buffer *find_extent_buffer(struct btrfs_fs_info *fs_info, |
3869 | u64 start) | |
452c75c3 CS |
3870 | { |
3871 | struct extent_buffer *eb; | |
3872 | ||
2f3186d8 QW |
3873 | eb = find_extent_buffer_nolock(fs_info, start); |
3874 | if (!eb) | |
3875 | return NULL; | |
3876 | /* | |
3877 | * Lock our eb's refs_lock to avoid races with free_extent_buffer(). | |
3878 | * When we get our eb it might be flagged with EXTENT_BUFFER_STALE and | |
3879 | * another task running free_extent_buffer() might have seen that flag | |
3880 | * set, eb->refs == 2, that the buffer isn't under IO (dirty and | |
3881 | * writeback flags not set) and it's still in the tree (flag | |
3882 | * EXTENT_BUFFER_TREE_REF set), therefore being in the process of | |
3883 | * decrementing the extent buffer's reference count twice. So here we | |
3884 | * could race and increment the eb's reference count, clear its stale | |
3885 | * flag, mark it as dirty and drop our reference before the other task | |
3886 | * finishes executing free_extent_buffer, which would later result in | |
3887 | * an attempt to free an extent buffer that is dirty. | |
3888 | */ | |
3889 | if (test_bit(EXTENT_BUFFER_STALE, &eb->bflags)) { | |
3890 | spin_lock(&eb->refs_lock); | |
3891 | spin_unlock(&eb->refs_lock); | |
452c75c3 | 3892 | } |
2f3186d8 QW |
3893 | mark_extent_buffer_accessed(eb, NULL); |
3894 | return eb; | |
452c75c3 CS |
3895 | } |
3896 | ||
faa2dbf0 JB |
3897 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
3898 | struct extent_buffer *alloc_test_extent_buffer(struct btrfs_fs_info *fs_info, | |
da17066c | 3899 | u64 start) |
faa2dbf0 JB |
3900 | { |
3901 | struct extent_buffer *eb, *exists = NULL; | |
3902 | int ret; | |
3903 | ||
3904 | eb = find_extent_buffer(fs_info, start); | |
3905 | if (eb) | |
3906 | return eb; | |
da17066c | 3907 | eb = alloc_dummy_extent_buffer(fs_info, start); |
faa2dbf0 | 3908 | if (!eb) |
b6293c82 | 3909 | return ERR_PTR(-ENOMEM); |
faa2dbf0 | 3910 | eb->fs_info = fs_info; |
01cd3909 DS |
3911 | again: |
3912 | ret = radix_tree_preload(GFP_NOFS); | |
3913 | if (ret) { | |
3914 | exists = ERR_PTR(ret); | |
3915 | goto free_eb; | |
3916 | } | |
3917 | spin_lock(&fs_info->buffer_lock); | |
3918 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
3919 | start >> fs_info->sectorsize_bits, eb); | |
3920 | spin_unlock(&fs_info->buffer_lock); | |
3921 | radix_tree_preload_end(); | |
3922 | if (ret == -EEXIST) { | |
3923 | exists = find_extent_buffer(fs_info, start); | |
3924 | if (exists) | |
faa2dbf0 | 3925 | goto free_eb; |
01cd3909 DS |
3926 | else |
3927 | goto again; | |
3928 | } | |
faa2dbf0 JB |
3929 | check_buffer_tree_ref(eb); |
3930 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); | |
3931 | ||
faa2dbf0 JB |
3932 | return eb; |
3933 | free_eb: | |
3934 | btrfs_release_extent_buffer(eb); | |
3935 | return exists; | |
3936 | } | |
3937 | #endif | |
3938 | ||
81982210 QW |
3939 | static struct extent_buffer *grab_extent_buffer( |
3940 | struct btrfs_fs_info *fs_info, struct page *page) | |
c0f0a9e7 QW |
3941 | { |
3942 | struct extent_buffer *exists; | |
3943 | ||
81982210 QW |
3944 | /* |
3945 | * For subpage case, we completely rely on radix tree to ensure we | |
3946 | * don't try to insert two ebs for the same bytenr. So here we always | |
3947 | * return NULL and just continue. | |
3948 | */ | |
fbca46eb | 3949 | if (fs_info->nodesize < PAGE_SIZE) |
81982210 QW |
3950 | return NULL; |
3951 | ||
c0f0a9e7 QW |
3952 | /* Page not yet attached to an extent buffer */ |
3953 | if (!PagePrivate(page)) | |
3954 | return NULL; | |
3955 | ||
3956 | /* | |
3957 | * We could have already allocated an eb for this page and attached one | |
3958 | * so lets see if we can get a ref on the existing eb, and if we can we | |
3959 | * know it's good and we can just return that one, else we know we can | |
3960 | * just overwrite page->private. | |
3961 | */ | |
3962 | exists = (struct extent_buffer *)page->private; | |
3963 | if (atomic_inc_not_zero(&exists->refs)) | |
3964 | return exists; | |
3965 | ||
3966 | WARN_ON(PageDirty(page)); | |
3967 | detach_page_private(page); | |
3968 | return NULL; | |
3969 | } | |
3970 | ||
fbca46eb QW |
3971 | static int check_eb_alignment(struct btrfs_fs_info *fs_info, u64 start) |
3972 | { | |
3973 | if (!IS_ALIGNED(start, fs_info->sectorsize)) { | |
3974 | btrfs_err(fs_info, "bad tree block start %llu", start); | |
3975 | return -EINVAL; | |
3976 | } | |
3977 | ||
3978 | if (fs_info->nodesize < PAGE_SIZE && | |
3979 | offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) { | |
3980 | btrfs_err(fs_info, | |
3981 | "tree block crosses page boundary, start %llu nodesize %u", | |
3982 | start, fs_info->nodesize); | |
3983 | return -EINVAL; | |
3984 | } | |
3985 | if (fs_info->nodesize >= PAGE_SIZE && | |
1280d2d1 | 3986 | !PAGE_ALIGNED(start)) { |
fbca46eb QW |
3987 | btrfs_err(fs_info, |
3988 | "tree block is not page aligned, start %llu nodesize %u", | |
3989 | start, fs_info->nodesize); | |
3990 | return -EINVAL; | |
3991 | } | |
3992 | return 0; | |
3993 | } | |
3994 | ||
f28491e0 | 3995 | struct extent_buffer *alloc_extent_buffer(struct btrfs_fs_info *fs_info, |
3fbaf258 | 3996 | u64 start, u64 owner_root, int level) |
d1310b2e | 3997 | { |
da17066c | 3998 | unsigned long len = fs_info->nodesize; |
cc5e31a4 DS |
3999 | int num_pages; |
4000 | int i; | |
09cbfeaf | 4001 | unsigned long index = start >> PAGE_SHIFT; |
d1310b2e | 4002 | struct extent_buffer *eb; |
6af118ce | 4003 | struct extent_buffer *exists = NULL; |
d1310b2e | 4004 | struct page *p; |
f28491e0 | 4005 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
b40130b2 | 4006 | u64 lockdep_owner = owner_root; |
d1310b2e | 4007 | int uptodate = 1; |
19fe0a8b | 4008 | int ret; |
d1310b2e | 4009 | |
fbca46eb | 4010 | if (check_eb_alignment(fs_info, start)) |
c871b0f2 | 4011 | return ERR_PTR(-EINVAL); |
c871b0f2 | 4012 | |
e9306ad4 QW |
4013 | #if BITS_PER_LONG == 32 |
4014 | if (start >= MAX_LFS_FILESIZE) { | |
4015 | btrfs_err_rl(fs_info, | |
4016 | "extent buffer %llu is beyond 32bit page cache limit", start); | |
4017 | btrfs_err_32bit_limit(fs_info); | |
4018 | return ERR_PTR(-EOVERFLOW); | |
4019 | } | |
4020 | if (start >= BTRFS_32BIT_EARLY_WARN_THRESHOLD) | |
4021 | btrfs_warn_32bit_limit(fs_info); | |
4022 | #endif | |
4023 | ||
f28491e0 | 4024 | eb = find_extent_buffer(fs_info, start); |
452c75c3 | 4025 | if (eb) |
6af118ce | 4026 | return eb; |
6af118ce | 4027 | |
23d79d81 | 4028 | eb = __alloc_extent_buffer(fs_info, start, len); |
2b114d1d | 4029 | if (!eb) |
c871b0f2 | 4030 | return ERR_PTR(-ENOMEM); |
b40130b2 JB |
4031 | |
4032 | /* | |
4033 | * The reloc trees are just snapshots, so we need them to appear to be | |
4034 | * just like any other fs tree WRT lockdep. | |
4035 | */ | |
4036 | if (lockdep_owner == BTRFS_TREE_RELOC_OBJECTID) | |
4037 | lockdep_owner = BTRFS_FS_TREE_OBJECTID; | |
4038 | ||
4039 | btrfs_set_buffer_lockdep_class(lockdep_owner, eb, level); | |
d1310b2e | 4040 | |
65ad0104 | 4041 | num_pages = num_extent_pages(eb); |
727011e0 | 4042 | for (i = 0; i < num_pages; i++, index++) { |
760f991f QW |
4043 | struct btrfs_subpage *prealloc = NULL; |
4044 | ||
d1b5c567 | 4045 | p = find_or_create_page(mapping, index, GFP_NOFS|__GFP_NOFAIL); |
c871b0f2 LB |
4046 | if (!p) { |
4047 | exists = ERR_PTR(-ENOMEM); | |
6af118ce | 4048 | goto free_eb; |
c871b0f2 | 4049 | } |
4f2de97a | 4050 | |
760f991f QW |
4051 | /* |
4052 | * Preallocate page->private for subpage case, so that we won't | |
4053 | * allocate memory with private_lock hold. The memory will be | |
4054 | * freed by attach_extent_buffer_page() or freed manually if | |
4055 | * we exit earlier. | |
4056 | * | |
4057 | * Although we have ensured one subpage eb can only have one | |
4058 | * page, but it may change in the future for 16K page size | |
4059 | * support, so we still preallocate the memory in the loop. | |
4060 | */ | |
fbca46eb | 4061 | if (fs_info->nodesize < PAGE_SIZE) { |
651fb419 QW |
4062 | prealloc = btrfs_alloc_subpage(fs_info, BTRFS_SUBPAGE_METADATA); |
4063 | if (IS_ERR(prealloc)) { | |
4064 | ret = PTR_ERR(prealloc); | |
fdf250db QW |
4065 | unlock_page(p); |
4066 | put_page(p); | |
4067 | exists = ERR_PTR(ret); | |
4068 | goto free_eb; | |
4069 | } | |
760f991f QW |
4070 | } |
4071 | ||
4f2de97a | 4072 | spin_lock(&mapping->private_lock); |
81982210 | 4073 | exists = grab_extent_buffer(fs_info, p); |
c0f0a9e7 QW |
4074 | if (exists) { |
4075 | spin_unlock(&mapping->private_lock); | |
4076 | unlock_page(p); | |
4077 | put_page(p); | |
4078 | mark_extent_buffer_accessed(exists, p); | |
760f991f | 4079 | btrfs_free_subpage(prealloc); |
c0f0a9e7 | 4080 | goto free_eb; |
d1310b2e | 4081 | } |
760f991f QW |
4082 | /* Should not fail, as we have preallocated the memory */ |
4083 | ret = attach_extent_buffer_page(eb, p, prealloc); | |
4084 | ASSERT(!ret); | |
8ff8466d QW |
4085 | /* |
4086 | * To inform we have extra eb under allocation, so that | |
4087 | * detach_extent_buffer_page() won't release the page private | |
4088 | * when the eb hasn't yet been inserted into radix tree. | |
4089 | * | |
4090 | * The ref will be decreased when the eb released the page, in | |
4091 | * detach_extent_buffer_page(). | |
4092 | * Thus needs no special handling in error path. | |
4093 | */ | |
4094 | btrfs_page_inc_eb_refs(fs_info, p); | |
4f2de97a | 4095 | spin_unlock(&mapping->private_lock); |
760f991f | 4096 | |
1e5eb3d6 | 4097 | WARN_ON(btrfs_page_test_dirty(fs_info, p, eb->start, eb->len)); |
727011e0 | 4098 | eb->pages[i] = p; |
d1310b2e CM |
4099 | if (!PageUptodate(p)) |
4100 | uptodate = 0; | |
eb14ab8e CM |
4101 | |
4102 | /* | |
b16d011e NB |
4103 | * We can't unlock the pages just yet since the extent buffer |
4104 | * hasn't been properly inserted in the radix tree, this | |
f913cff3 | 4105 | * opens a race with btree_release_folio which can free a page |
b16d011e NB |
4106 | * while we are still filling in all pages for the buffer and |
4107 | * we could crash. | |
eb14ab8e | 4108 | */ |
d1310b2e CM |
4109 | } |
4110 | if (uptodate) | |
b4ce94de | 4111 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
01cd3909 DS |
4112 | again: |
4113 | ret = radix_tree_preload(GFP_NOFS); | |
4114 | if (ret) { | |
4115 | exists = ERR_PTR(ret); | |
4116 | goto free_eb; | |
4117 | } | |
4118 | ||
4119 | spin_lock(&fs_info->buffer_lock); | |
4120 | ret = radix_tree_insert(&fs_info->buffer_radix, | |
4121 | start >> fs_info->sectorsize_bits, eb); | |
4122 | spin_unlock(&fs_info->buffer_lock); | |
4123 | radix_tree_preload_end(); | |
4124 | if (ret == -EEXIST) { | |
4125 | exists = find_extent_buffer(fs_info, start); | |
4126 | if (exists) | |
452c75c3 | 4127 | goto free_eb; |
01cd3909 DS |
4128 | else |
4129 | goto again; | |
4130 | } | |
6af118ce | 4131 | /* add one reference for the tree */ |
0b32f4bb | 4132 | check_buffer_tree_ref(eb); |
34b41ace | 4133 | set_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags); |
eb14ab8e CM |
4134 | |
4135 | /* | |
b16d011e | 4136 | * Now it's safe to unlock the pages because any calls to |
f913cff3 | 4137 | * btree_release_folio will correctly detect that a page belongs to a |
b16d011e | 4138 | * live buffer and won't free them prematurely. |
eb14ab8e | 4139 | */ |
28187ae5 NB |
4140 | for (i = 0; i < num_pages; i++) |
4141 | unlock_page(eb->pages[i]); | |
d1310b2e CM |
4142 | return eb; |
4143 | ||
6af118ce | 4144 | free_eb: |
5ca64f45 | 4145 | WARN_ON(!atomic_dec_and_test(&eb->refs)); |
727011e0 CM |
4146 | for (i = 0; i < num_pages; i++) { |
4147 | if (eb->pages[i]) | |
4148 | unlock_page(eb->pages[i]); | |
4149 | } | |
eb14ab8e | 4150 | |
897ca6e9 | 4151 | btrfs_release_extent_buffer(eb); |
6af118ce | 4152 | return exists; |
d1310b2e | 4153 | } |
d1310b2e | 4154 | |
3083ee2e JB |
4155 | static inline void btrfs_release_extent_buffer_rcu(struct rcu_head *head) |
4156 | { | |
4157 | struct extent_buffer *eb = | |
4158 | container_of(head, struct extent_buffer, rcu_head); | |
4159 | ||
4160 | __free_extent_buffer(eb); | |
4161 | } | |
4162 | ||
f7a52a40 | 4163 | static int release_extent_buffer(struct extent_buffer *eb) |
5ce48d0f | 4164 | __releases(&eb->refs_lock) |
3083ee2e | 4165 | { |
07e21c4d NB |
4166 | lockdep_assert_held(&eb->refs_lock); |
4167 | ||
3083ee2e JB |
4168 | WARN_ON(atomic_read(&eb->refs) == 0); |
4169 | if (atomic_dec_and_test(&eb->refs)) { | |
34b41ace | 4170 | if (test_and_clear_bit(EXTENT_BUFFER_IN_TREE, &eb->bflags)) { |
f28491e0 | 4171 | struct btrfs_fs_info *fs_info = eb->fs_info; |
3083ee2e | 4172 | |
815a51c7 | 4173 | spin_unlock(&eb->refs_lock); |
3083ee2e | 4174 | |
01cd3909 DS |
4175 | spin_lock(&fs_info->buffer_lock); |
4176 | radix_tree_delete(&fs_info->buffer_radix, | |
4177 | eb->start >> fs_info->sectorsize_bits); | |
4178 | spin_unlock(&fs_info->buffer_lock); | |
34b41ace JB |
4179 | } else { |
4180 | spin_unlock(&eb->refs_lock); | |
815a51c7 | 4181 | } |
3083ee2e | 4182 | |
a40246e8 | 4183 | btrfs_leak_debug_del_eb(eb); |
3083ee2e | 4184 | /* Should be safe to release our pages at this point */ |
55ac0139 | 4185 | btrfs_release_extent_buffer_pages(eb); |
bcb7e449 | 4186 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
b0132a3b | 4187 | if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags))) { |
bcb7e449 JB |
4188 | __free_extent_buffer(eb); |
4189 | return 1; | |
4190 | } | |
4191 | #endif | |
3083ee2e | 4192 | call_rcu(&eb->rcu_head, btrfs_release_extent_buffer_rcu); |
e64860aa | 4193 | return 1; |
3083ee2e JB |
4194 | } |
4195 | spin_unlock(&eb->refs_lock); | |
e64860aa JB |
4196 | |
4197 | return 0; | |
3083ee2e JB |
4198 | } |
4199 | ||
d1310b2e CM |
4200 | void free_extent_buffer(struct extent_buffer *eb) |
4201 | { | |
242e18c7 | 4202 | int refs; |
d1310b2e CM |
4203 | if (!eb) |
4204 | return; | |
4205 | ||
e5677f05 | 4206 | refs = atomic_read(&eb->refs); |
242e18c7 | 4207 | while (1) { |
46cc775e NB |
4208 | if ((!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && refs <= 3) |
4209 | || (test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags) && | |
4210 | refs == 1)) | |
242e18c7 | 4211 | break; |
e5677f05 | 4212 | if (atomic_try_cmpxchg(&eb->refs, &refs, refs - 1)) |
242e18c7 CM |
4213 | return; |
4214 | } | |
4215 | ||
3083ee2e JB |
4216 | spin_lock(&eb->refs_lock); |
4217 | if (atomic_read(&eb->refs) == 2 && | |
4218 | test_bit(EXTENT_BUFFER_STALE, &eb->bflags) && | |
0b32f4bb | 4219 | !extent_buffer_under_io(eb) && |
3083ee2e JB |
4220 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4221 | atomic_dec(&eb->refs); | |
4222 | ||
4223 | /* | |
4224 | * I know this is terrible, but it's temporary until we stop tracking | |
4225 | * the uptodate bits and such for the extent buffers. | |
4226 | */ | |
f7a52a40 | 4227 | release_extent_buffer(eb); |
3083ee2e JB |
4228 | } |
4229 | ||
4230 | void free_extent_buffer_stale(struct extent_buffer *eb) | |
4231 | { | |
4232 | if (!eb) | |
d1310b2e CM |
4233 | return; |
4234 | ||
3083ee2e JB |
4235 | spin_lock(&eb->refs_lock); |
4236 | set_bit(EXTENT_BUFFER_STALE, &eb->bflags); | |
4237 | ||
0b32f4bb | 4238 | if (atomic_read(&eb->refs) == 2 && !extent_buffer_under_io(eb) && |
3083ee2e JB |
4239 | test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) |
4240 | atomic_dec(&eb->refs); | |
f7a52a40 | 4241 | release_extent_buffer(eb); |
d1310b2e | 4242 | } |
d1310b2e | 4243 | |
0d27797e QW |
4244 | static void btree_clear_page_dirty(struct page *page) |
4245 | { | |
4246 | ASSERT(PageDirty(page)); | |
4247 | ASSERT(PageLocked(page)); | |
4248 | clear_page_dirty_for_io(page); | |
4249 | xa_lock_irq(&page->mapping->i_pages); | |
4250 | if (!PageDirty(page)) | |
4251 | __xa_clear_mark(&page->mapping->i_pages, | |
4252 | page_index(page), PAGECACHE_TAG_DIRTY); | |
4253 | xa_unlock_irq(&page->mapping->i_pages); | |
4254 | } | |
4255 | ||
4256 | static void clear_subpage_extent_buffer_dirty(const struct extent_buffer *eb) | |
4257 | { | |
4258 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4259 | struct page *page = eb->pages[0]; | |
4260 | bool last; | |
4261 | ||
4262 | /* btree_clear_page_dirty() needs page locked */ | |
4263 | lock_page(page); | |
4264 | last = btrfs_subpage_clear_and_test_dirty(fs_info, page, eb->start, | |
4265 | eb->len); | |
4266 | if (last) | |
4267 | btree_clear_page_dirty(page); | |
4268 | unlock_page(page); | |
4269 | WARN_ON(atomic_read(&eb->refs) == 0); | |
4270 | } | |
4271 | ||
98c8d683 JB |
4272 | void btrfs_clear_buffer_dirty(struct btrfs_trans_handle *trans, |
4273 | struct extent_buffer *eb) | |
d1310b2e | 4274 | { |
98c8d683 | 4275 | struct btrfs_fs_info *fs_info = eb->fs_info; |
cc5e31a4 DS |
4276 | int i; |
4277 | int num_pages; | |
d1310b2e CM |
4278 | struct page *page; |
4279 | ||
98c8d683 JB |
4280 | btrfs_assert_tree_write_locked(eb); |
4281 | ||
4282 | if (trans && btrfs_header_generation(eb) != trans->transid) | |
4283 | return; | |
4284 | ||
4285 | if (!test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) | |
4286 | return; | |
4287 | ||
4288 | percpu_counter_add_batch(&fs_info->dirty_metadata_bytes, -eb->len, | |
4289 | fs_info->dirty_metadata_batch); | |
4290 | ||
fbca46eb | 4291 | if (eb->fs_info->nodesize < PAGE_SIZE) |
0d27797e QW |
4292 | return clear_subpage_extent_buffer_dirty(eb); |
4293 | ||
65ad0104 | 4294 | num_pages = num_extent_pages(eb); |
d1310b2e CM |
4295 | |
4296 | for (i = 0; i < num_pages; i++) { | |
fb85fc9a | 4297 | page = eb->pages[i]; |
b9473439 | 4298 | if (!PageDirty(page)) |
d2c3f4f6 | 4299 | continue; |
a61e6f29 | 4300 | lock_page(page); |
0d27797e | 4301 | btree_clear_page_dirty(page); |
bf0da8c1 | 4302 | ClearPageError(page); |
a61e6f29 | 4303 | unlock_page(page); |
d1310b2e | 4304 | } |
0b32f4bb | 4305 | WARN_ON(atomic_read(&eb->refs) == 0); |
d1310b2e | 4306 | } |
d1310b2e | 4307 | |
abb57ef3 | 4308 | bool set_extent_buffer_dirty(struct extent_buffer *eb) |
d1310b2e | 4309 | { |
cc5e31a4 DS |
4310 | int i; |
4311 | int num_pages; | |
abb57ef3 | 4312 | bool was_dirty; |
d1310b2e | 4313 | |
0b32f4bb JB |
4314 | check_buffer_tree_ref(eb); |
4315 | ||
b9473439 | 4316 | was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags); |
0b32f4bb | 4317 | |
65ad0104 | 4318 | num_pages = num_extent_pages(eb); |
3083ee2e | 4319 | WARN_ON(atomic_read(&eb->refs) == 0); |
0b32f4bb JB |
4320 | WARN_ON(!test_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)); |
4321 | ||
0d27797e | 4322 | if (!was_dirty) { |
fbca46eb | 4323 | bool subpage = eb->fs_info->nodesize < PAGE_SIZE; |
51995c39 | 4324 | |
0d27797e QW |
4325 | /* |
4326 | * For subpage case, we can have other extent buffers in the | |
4327 | * same page, and in clear_subpage_extent_buffer_dirty() we | |
4328 | * have to clear page dirty without subpage lock held. | |
4329 | * This can cause race where our page gets dirty cleared after | |
4330 | * we just set it. | |
4331 | * | |
4332 | * Thankfully, clear_subpage_extent_buffer_dirty() has locked | |
4333 | * its page for other reasons, we can use page lock to prevent | |
4334 | * the above race. | |
4335 | */ | |
4336 | if (subpage) | |
4337 | lock_page(eb->pages[0]); | |
4338 | for (i = 0; i < num_pages; i++) | |
4339 | btrfs_page_set_dirty(eb->fs_info, eb->pages[i], | |
4340 | eb->start, eb->len); | |
4341 | if (subpage) | |
4342 | unlock_page(eb->pages[0]); | |
4343 | } | |
51995c39 LB |
4344 | #ifdef CONFIG_BTRFS_DEBUG |
4345 | for (i = 0; i < num_pages; i++) | |
4346 | ASSERT(PageDirty(eb->pages[i])); | |
4347 | #endif | |
4348 | ||
b9473439 | 4349 | return was_dirty; |
d1310b2e | 4350 | } |
d1310b2e | 4351 | |
69ba3927 | 4352 | void clear_extent_buffer_uptodate(struct extent_buffer *eb) |
1259ab75 | 4353 | { |
251f2acc | 4354 | struct btrfs_fs_info *fs_info = eb->fs_info; |
1259ab75 | 4355 | struct page *page; |
cc5e31a4 | 4356 | int num_pages; |
251f2acc | 4357 | int i; |
1259ab75 | 4358 | |
b4ce94de | 4359 | clear_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 4360 | num_pages = num_extent_pages(eb); |
1259ab75 | 4361 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4362 | page = eb->pages[i]; |
fbca46eb QW |
4363 | if (!page) |
4364 | continue; | |
4365 | ||
4366 | /* | |
4367 | * This is special handling for metadata subpage, as regular | |
4368 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
4369 | */ | |
4370 | if (fs_info->nodesize >= PAGE_SIZE) | |
4371 | ClearPageUptodate(page); | |
4372 | else | |
4373 | btrfs_subpage_clear_uptodate(fs_info, page, eb->start, | |
4374 | eb->len); | |
1259ab75 | 4375 | } |
1259ab75 CM |
4376 | } |
4377 | ||
09c25a8c | 4378 | void set_extent_buffer_uptodate(struct extent_buffer *eb) |
d1310b2e | 4379 | { |
251f2acc | 4380 | struct btrfs_fs_info *fs_info = eb->fs_info; |
d1310b2e | 4381 | struct page *page; |
cc5e31a4 | 4382 | int num_pages; |
251f2acc | 4383 | int i; |
d1310b2e | 4384 | |
0b32f4bb | 4385 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
65ad0104 | 4386 | num_pages = num_extent_pages(eb); |
d1310b2e | 4387 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4388 | page = eb->pages[i]; |
fbca46eb QW |
4389 | |
4390 | /* | |
4391 | * This is special handling for metadata subpage, as regular | |
4392 | * btrfs_is_subpage() can not handle cloned/dummy metadata. | |
4393 | */ | |
4394 | if (fs_info->nodesize >= PAGE_SIZE) | |
4395 | SetPageUptodate(page); | |
4396 | else | |
4397 | btrfs_subpage_set_uptodate(fs_info, page, eb->start, | |
4398 | eb->len); | |
d1310b2e | 4399 | } |
d1310b2e | 4400 | } |
d1310b2e | 4401 | |
4012daf7 | 4402 | static int read_extent_buffer_subpage(struct extent_buffer *eb, int wait, |
947a6299 QW |
4403 | int mirror_num, |
4404 | struct btrfs_tree_parent_check *check) | |
4012daf7 QW |
4405 | { |
4406 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4407 | struct extent_io_tree *io_tree; | |
4408 | struct page *page = eb->pages[0]; | |
e5e886ba | 4409 | struct extent_state *cached_state = NULL; |
722c82ac CH |
4410 | struct btrfs_bio_ctrl bio_ctrl = { |
4411 | .mirror_num = mirror_num, | |
1d854e4f | 4412 | .parent_check = check, |
722c82ac | 4413 | }; |
4012daf7 QW |
4414 | int ret = 0; |
4415 | ||
4416 | ASSERT(!test_bit(EXTENT_BUFFER_UNMAPPED, &eb->bflags)); | |
4417 | ASSERT(PagePrivate(page)); | |
947a6299 | 4418 | ASSERT(check); |
4012daf7 QW |
4419 | io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree; |
4420 | ||
4421 | if (wait == WAIT_NONE) { | |
83ae4133 | 4422 | if (!try_lock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
e5e886ba | 4423 | &cached_state)) |
dc56219f | 4424 | return -EAGAIN; |
4012daf7 | 4425 | } else { |
e5e886ba JB |
4426 | ret = lock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
4427 | &cached_state); | |
4012daf7 QW |
4428 | if (ret < 0) |
4429 | return ret; | |
4430 | } | |
4431 | ||
4432 | ret = 0; | |
4433 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags) || | |
4434 | PageUptodate(page) || | |
4435 | btrfs_subpage_test_uptodate(fs_info, page, eb->start, eb->len)) { | |
4436 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); | |
e5e886ba JB |
4437 | unlock_extent(io_tree, eb->start, eb->start + eb->len - 1, |
4438 | &cached_state); | |
4012daf7 QW |
4439 | return ret; |
4440 | } | |
4441 | ||
4442 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); | |
4443 | eb->read_mirror = 0; | |
4444 | atomic_set(&eb->io_pages, 1); | |
4445 | check_buffer_tree_ref(eb); | |
5467abba QW |
4446 | bio_ctrl.end_io_func = end_bio_extent_readpage; |
4447 | ||
4012daf7 QW |
4448 | btrfs_subpage_clear_error(fs_info, page, eb->start, eb->len); |
4449 | ||
3d078efa | 4450 | btrfs_subpage_start_reader(fs_info, page, eb->start, eb->len); |
08a6f464 | 4451 | ret = submit_extent_page(REQ_OP_READ, NULL, &bio_ctrl, |
209ecde5 | 4452 | eb->start, page, eb->len, |
5467abba | 4453 | eb->start - page_offset(page), 0, true); |
4012daf7 QW |
4454 | if (ret) { |
4455 | /* | |
4456 | * In the endio function, if we hit something wrong we will | |
4457 | * increase the io_pages, so here we need to decrease it for | |
4458 | * error path. | |
4459 | */ | |
4460 | atomic_dec(&eb->io_pages); | |
4461 | } | |
722c82ac | 4462 | submit_one_bio(&bio_ctrl); |
e5e886ba JB |
4463 | if (ret || wait != WAIT_COMPLETE) { |
4464 | free_extent_state(cached_state); | |
4012daf7 | 4465 | return ret; |
e5e886ba | 4466 | } |
4012daf7 | 4467 | |
123a7f00 | 4468 | wait_extent_bit(io_tree, eb->start, eb->start + eb->len - 1, |
e5e886ba | 4469 | EXTENT_LOCKED, &cached_state); |
4012daf7 QW |
4470 | if (!test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
4471 | ret = -EIO; | |
4472 | return ret; | |
4473 | } | |
4474 | ||
947a6299 QW |
4475 | int read_extent_buffer_pages(struct extent_buffer *eb, int wait, int mirror_num, |
4476 | struct btrfs_tree_parent_check *check) | |
d1310b2e | 4477 | { |
cc5e31a4 | 4478 | int i; |
d1310b2e CM |
4479 | struct page *page; |
4480 | int err; | |
4481 | int ret = 0; | |
ce9adaa5 CM |
4482 | int locked_pages = 0; |
4483 | int all_uptodate = 1; | |
cc5e31a4 | 4484 | int num_pages; |
727011e0 | 4485 | unsigned long num_reads = 0; |
722c82ac CH |
4486 | struct btrfs_bio_ctrl bio_ctrl = { |
4487 | .mirror_num = mirror_num, | |
1d854e4f | 4488 | .parent_check = check, |
722c82ac | 4489 | }; |
a86c12c7 | 4490 | |
b4ce94de | 4491 | if (test_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags)) |
d1310b2e CM |
4492 | return 0; |
4493 | ||
651740a5 JB |
4494 | /* |
4495 | * We could have had EXTENT_BUFFER_UPTODATE cleared by the write | |
4496 | * operation, which could potentially still be in flight. In this case | |
4497 | * we simply want to return an error. | |
4498 | */ | |
4499 | if (unlikely(test_bit(EXTENT_BUFFER_WRITE_ERR, &eb->bflags))) | |
4500 | return -EIO; | |
4501 | ||
fbca46eb | 4502 | if (eb->fs_info->nodesize < PAGE_SIZE) |
947a6299 | 4503 | return read_extent_buffer_subpage(eb, wait, mirror_num, check); |
4012daf7 | 4504 | |
65ad0104 | 4505 | num_pages = num_extent_pages(eb); |
8436ea91 | 4506 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4507 | page = eb->pages[i]; |
bb82ab88 | 4508 | if (wait == WAIT_NONE) { |
2c4d8cb7 QW |
4509 | /* |
4510 | * WAIT_NONE is only utilized by readahead. If we can't | |
4511 | * acquire the lock atomically it means either the eb | |
4512 | * is being read out or under modification. | |
4513 | * Either way the eb will be or has been cached, | |
4514 | * readahead can exit safely. | |
4515 | */ | |
2db04966 | 4516 | if (!trylock_page(page)) |
ce9adaa5 | 4517 | goto unlock_exit; |
d1310b2e CM |
4518 | } else { |
4519 | lock_page(page); | |
4520 | } | |
ce9adaa5 | 4521 | locked_pages++; |
2571e739 LB |
4522 | } |
4523 | /* | |
4524 | * We need to firstly lock all pages to make sure that | |
4525 | * the uptodate bit of our pages won't be affected by | |
4526 | * clear_extent_buffer_uptodate(). | |
4527 | */ | |
8436ea91 | 4528 | for (i = 0; i < num_pages; i++) { |
2571e739 | 4529 | page = eb->pages[i]; |
727011e0 CM |
4530 | if (!PageUptodate(page)) { |
4531 | num_reads++; | |
ce9adaa5 | 4532 | all_uptodate = 0; |
727011e0 | 4533 | } |
ce9adaa5 | 4534 | } |
2571e739 | 4535 | |
ce9adaa5 | 4536 | if (all_uptodate) { |
8436ea91 | 4537 | set_bit(EXTENT_BUFFER_UPTODATE, &eb->bflags); |
ce9adaa5 CM |
4538 | goto unlock_exit; |
4539 | } | |
4540 | ||
656f30db | 4541 | clear_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags); |
5cf1ab56 | 4542 | eb->read_mirror = 0; |
0b32f4bb | 4543 | atomic_set(&eb->io_pages, num_reads); |
6bf9cd2e | 4544 | /* |
f913cff3 | 4545 | * It is possible for release_folio to clear the TREE_REF bit before we |
6bf9cd2e BB |
4546 | * set io_pages. See check_buffer_tree_ref for a more detailed comment. |
4547 | */ | |
4548 | check_buffer_tree_ref(eb); | |
5467abba | 4549 | bio_ctrl.end_io_func = end_bio_extent_readpage; |
8436ea91 | 4550 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4551 | page = eb->pages[i]; |
baf863b9 | 4552 | |
ce9adaa5 | 4553 | if (!PageUptodate(page)) { |
baf863b9 LB |
4554 | if (ret) { |
4555 | atomic_dec(&eb->io_pages); | |
4556 | unlock_page(page); | |
4557 | continue; | |
4558 | } | |
4559 | ||
f188591e | 4560 | ClearPageError(page); |
08a6f464 | 4561 | err = submit_extent_page(REQ_OP_READ, NULL, |
209ecde5 | 4562 | &bio_ctrl, page_offset(page), page, |
5467abba | 4563 | PAGE_SIZE, 0, 0, false); |
baf863b9 | 4564 | if (err) { |
baf863b9 | 4565 | /* |
0420177c NB |
4566 | * We failed to submit the bio so it's the |
4567 | * caller's responsibility to perform cleanup | |
4568 | * i.e unlock page/set error bit. | |
baf863b9 | 4569 | */ |
0420177c NB |
4570 | ret = err; |
4571 | SetPageError(page); | |
4572 | unlock_page(page); | |
baf863b9 LB |
4573 | atomic_dec(&eb->io_pages); |
4574 | } | |
d1310b2e CM |
4575 | } else { |
4576 | unlock_page(page); | |
4577 | } | |
4578 | } | |
4579 | ||
722c82ac | 4580 | submit_one_bio(&bio_ctrl); |
a86c12c7 | 4581 | |
bb82ab88 | 4582 | if (ret || wait != WAIT_COMPLETE) |
d1310b2e | 4583 | return ret; |
d397712b | 4584 | |
8436ea91 | 4585 | for (i = 0; i < num_pages; i++) { |
fb85fc9a | 4586 | page = eb->pages[i]; |
d1310b2e | 4587 | wait_on_page_locked(page); |
d397712b | 4588 | if (!PageUptodate(page)) |
d1310b2e | 4589 | ret = -EIO; |
d1310b2e | 4590 | } |
d397712b | 4591 | |
d1310b2e | 4592 | return ret; |
ce9adaa5 CM |
4593 | |
4594 | unlock_exit: | |
d397712b | 4595 | while (locked_pages > 0) { |
ce9adaa5 | 4596 | locked_pages--; |
8436ea91 JB |
4597 | page = eb->pages[locked_pages]; |
4598 | unlock_page(page); | |
ce9adaa5 CM |
4599 | } |
4600 | return ret; | |
d1310b2e | 4601 | } |
d1310b2e | 4602 | |
f98b6215 QW |
4603 | static bool report_eb_range(const struct extent_buffer *eb, unsigned long start, |
4604 | unsigned long len) | |
4605 | { | |
4606 | btrfs_warn(eb->fs_info, | |
4607 | "access to eb bytenr %llu len %lu out of range start %lu len %lu", | |
4608 | eb->start, eb->len, start, len); | |
4609 | WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG)); | |
4610 | ||
4611 | return true; | |
4612 | } | |
4613 | ||
4614 | /* | |
4615 | * Check if the [start, start + len) range is valid before reading/writing | |
4616 | * the eb. | |
4617 | * NOTE: @start and @len are offset inside the eb, not logical address. | |
4618 | * | |
4619 | * Caller should not touch the dst/src memory if this function returns error. | |
4620 | */ | |
4621 | static inline int check_eb_range(const struct extent_buffer *eb, | |
4622 | unsigned long start, unsigned long len) | |
4623 | { | |
4624 | unsigned long offset; | |
4625 | ||
4626 | /* start, start + len should not go beyond eb->len nor overflow */ | |
4627 | if (unlikely(check_add_overflow(start, len, &offset) || offset > eb->len)) | |
4628 | return report_eb_range(eb, start, len); | |
4629 | ||
4630 | return false; | |
4631 | } | |
4632 | ||
1cbb1f45 JM |
4633 | void read_extent_buffer(const struct extent_buffer *eb, void *dstv, |
4634 | unsigned long start, unsigned long len) | |
d1310b2e CM |
4635 | { |
4636 | size_t cur; | |
4637 | size_t offset; | |
4638 | struct page *page; | |
4639 | char *kaddr; | |
4640 | char *dst = (char *)dstv; | |
884b07d0 | 4641 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4642 | |
f98b6215 | 4643 | if (check_eb_range(eb, start, len)) |
f716abd5 | 4644 | return; |
d1310b2e | 4645 | |
884b07d0 | 4646 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4647 | |
d397712b | 4648 | while (len > 0) { |
fb85fc9a | 4649 | page = eb->pages[i]; |
d1310b2e | 4650 | |
09cbfeaf | 4651 | cur = min(len, (PAGE_SIZE - offset)); |
a6591715 | 4652 | kaddr = page_address(page); |
d1310b2e | 4653 | memcpy(dst, kaddr + offset, cur); |
d1310b2e CM |
4654 | |
4655 | dst += cur; | |
4656 | len -= cur; | |
4657 | offset = 0; | |
4658 | i++; | |
4659 | } | |
4660 | } | |
d1310b2e | 4661 | |
a48b73ec JB |
4662 | int read_extent_buffer_to_user_nofault(const struct extent_buffer *eb, |
4663 | void __user *dstv, | |
4664 | unsigned long start, unsigned long len) | |
550ac1d8 GH |
4665 | { |
4666 | size_t cur; | |
4667 | size_t offset; | |
4668 | struct page *page; | |
4669 | char *kaddr; | |
4670 | char __user *dst = (char __user *)dstv; | |
884b07d0 | 4671 | unsigned long i = get_eb_page_index(start); |
550ac1d8 GH |
4672 | int ret = 0; |
4673 | ||
4674 | WARN_ON(start > eb->len); | |
4675 | WARN_ON(start + len > eb->start + eb->len); | |
4676 | ||
884b07d0 | 4677 | offset = get_eb_offset_in_page(eb, start); |
550ac1d8 GH |
4678 | |
4679 | while (len > 0) { | |
fb85fc9a | 4680 | page = eb->pages[i]; |
550ac1d8 | 4681 | |
09cbfeaf | 4682 | cur = min(len, (PAGE_SIZE - offset)); |
550ac1d8 | 4683 | kaddr = page_address(page); |
a48b73ec | 4684 | if (copy_to_user_nofault(dst, kaddr + offset, cur)) { |
550ac1d8 GH |
4685 | ret = -EFAULT; |
4686 | break; | |
4687 | } | |
4688 | ||
4689 | dst += cur; | |
4690 | len -= cur; | |
4691 | offset = 0; | |
4692 | i++; | |
4693 | } | |
4694 | ||
4695 | return ret; | |
4696 | } | |
4697 | ||
1cbb1f45 JM |
4698 | int memcmp_extent_buffer(const struct extent_buffer *eb, const void *ptrv, |
4699 | unsigned long start, unsigned long len) | |
d1310b2e CM |
4700 | { |
4701 | size_t cur; | |
4702 | size_t offset; | |
4703 | struct page *page; | |
4704 | char *kaddr; | |
4705 | char *ptr = (char *)ptrv; | |
884b07d0 | 4706 | unsigned long i = get_eb_page_index(start); |
d1310b2e CM |
4707 | int ret = 0; |
4708 | ||
f98b6215 QW |
4709 | if (check_eb_range(eb, start, len)) |
4710 | return -EINVAL; | |
d1310b2e | 4711 | |
884b07d0 | 4712 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4713 | |
d397712b | 4714 | while (len > 0) { |
fb85fc9a | 4715 | page = eb->pages[i]; |
d1310b2e | 4716 | |
09cbfeaf | 4717 | cur = min(len, (PAGE_SIZE - offset)); |
d1310b2e | 4718 | |
a6591715 | 4719 | kaddr = page_address(page); |
d1310b2e | 4720 | ret = memcmp(ptr, kaddr + offset, cur); |
d1310b2e CM |
4721 | if (ret) |
4722 | break; | |
4723 | ||
4724 | ptr += cur; | |
4725 | len -= cur; | |
4726 | offset = 0; | |
4727 | i++; | |
4728 | } | |
4729 | return ret; | |
4730 | } | |
d1310b2e | 4731 | |
b8f95771 QW |
4732 | /* |
4733 | * Check that the extent buffer is uptodate. | |
4734 | * | |
4735 | * For regular sector size == PAGE_SIZE case, check if @page is uptodate. | |
4736 | * For subpage case, check if the range covered by the eb has EXTENT_UPTODATE. | |
4737 | */ | |
4738 | static void assert_eb_page_uptodate(const struct extent_buffer *eb, | |
4739 | struct page *page) | |
4740 | { | |
4741 | struct btrfs_fs_info *fs_info = eb->fs_info; | |
4742 | ||
a50e1fcb JB |
4743 | /* |
4744 | * If we are using the commit root we could potentially clear a page | |
4745 | * Uptodate while we're using the extent buffer that we've previously | |
4746 | * looked up. We don't want to complain in this case, as the page was | |
4747 | * valid before, we just didn't write it out. Instead we want to catch | |
4748 | * the case where we didn't actually read the block properly, which | |
4749 | * would have !PageUptodate && !PageError, as we clear PageError before | |
4750 | * reading. | |
4751 | */ | |
fbca46eb | 4752 | if (fs_info->nodesize < PAGE_SIZE) { |
a50e1fcb | 4753 | bool uptodate, error; |
b8f95771 QW |
4754 | |
4755 | uptodate = btrfs_subpage_test_uptodate(fs_info, page, | |
4756 | eb->start, eb->len); | |
a50e1fcb JB |
4757 | error = btrfs_subpage_test_error(fs_info, page, eb->start, eb->len); |
4758 | WARN_ON(!uptodate && !error); | |
b8f95771 | 4759 | } else { |
a50e1fcb | 4760 | WARN_ON(!PageUptodate(page) && !PageError(page)); |
b8f95771 QW |
4761 | } |
4762 | } | |
4763 | ||
2b48966a | 4764 | void write_extent_buffer_chunk_tree_uuid(const struct extent_buffer *eb, |
f157bf76 DS |
4765 | const void *srcv) |
4766 | { | |
4767 | char *kaddr; | |
4768 | ||
b8f95771 | 4769 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
4770 | kaddr = page_address(eb->pages[0]) + |
4771 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, | |
4772 | chunk_tree_uuid)); | |
4773 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
4774 | } |
4775 | ||
2b48966a | 4776 | void write_extent_buffer_fsid(const struct extent_buffer *eb, const void *srcv) |
f157bf76 DS |
4777 | { |
4778 | char *kaddr; | |
4779 | ||
b8f95771 | 4780 | assert_eb_page_uptodate(eb, eb->pages[0]); |
24880be5 DS |
4781 | kaddr = page_address(eb->pages[0]) + |
4782 | get_eb_offset_in_page(eb, offsetof(struct btrfs_header, fsid)); | |
4783 | memcpy(kaddr, srcv, BTRFS_FSID_SIZE); | |
f157bf76 DS |
4784 | } |
4785 | ||
2b48966a | 4786 | void write_extent_buffer(const struct extent_buffer *eb, const void *srcv, |
d1310b2e CM |
4787 | unsigned long start, unsigned long len) |
4788 | { | |
4789 | size_t cur; | |
4790 | size_t offset; | |
4791 | struct page *page; | |
4792 | char *kaddr; | |
4793 | char *src = (char *)srcv; | |
884b07d0 | 4794 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4795 | |
d3575156 NA |
4796 | WARN_ON(test_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags)); |
4797 | ||
f98b6215 QW |
4798 | if (check_eb_range(eb, start, len)) |
4799 | return; | |
d1310b2e | 4800 | |
884b07d0 | 4801 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4802 | |
d397712b | 4803 | while (len > 0) { |
fb85fc9a | 4804 | page = eb->pages[i]; |
b8f95771 | 4805 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 4806 | |
09cbfeaf | 4807 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 4808 | kaddr = page_address(page); |
d1310b2e | 4809 | memcpy(kaddr + offset, src, cur); |
d1310b2e CM |
4810 | |
4811 | src += cur; | |
4812 | len -= cur; | |
4813 | offset = 0; | |
4814 | i++; | |
4815 | } | |
4816 | } | |
d1310b2e | 4817 | |
2b48966a | 4818 | void memzero_extent_buffer(const struct extent_buffer *eb, unsigned long start, |
b159fa28 | 4819 | unsigned long len) |
d1310b2e CM |
4820 | { |
4821 | size_t cur; | |
4822 | size_t offset; | |
4823 | struct page *page; | |
4824 | char *kaddr; | |
884b07d0 | 4825 | unsigned long i = get_eb_page_index(start); |
d1310b2e | 4826 | |
f98b6215 QW |
4827 | if (check_eb_range(eb, start, len)) |
4828 | return; | |
d1310b2e | 4829 | |
884b07d0 | 4830 | offset = get_eb_offset_in_page(eb, start); |
d1310b2e | 4831 | |
d397712b | 4832 | while (len > 0) { |
fb85fc9a | 4833 | page = eb->pages[i]; |
b8f95771 | 4834 | assert_eb_page_uptodate(eb, page); |
d1310b2e | 4835 | |
09cbfeaf | 4836 | cur = min(len, PAGE_SIZE - offset); |
a6591715 | 4837 | kaddr = page_address(page); |
b159fa28 | 4838 | memset(kaddr + offset, 0, cur); |
d1310b2e CM |
4839 | |
4840 | len -= cur; | |
4841 | offset = 0; | |
4842 | i++; | |
4843 | } | |
4844 | } | |
d1310b2e | 4845 | |
2b48966a DS |
4846 | void copy_extent_buffer_full(const struct extent_buffer *dst, |
4847 | const struct extent_buffer *src) | |
58e8012c DS |
4848 | { |
4849 | int i; | |
cc5e31a4 | 4850 | int num_pages; |
58e8012c DS |
4851 | |
4852 | ASSERT(dst->len == src->len); | |
4853 | ||
fbca46eb | 4854 | if (dst->fs_info->nodesize >= PAGE_SIZE) { |
884b07d0 QW |
4855 | num_pages = num_extent_pages(dst); |
4856 | for (i = 0; i < num_pages; i++) | |
4857 | copy_page(page_address(dst->pages[i]), | |
4858 | page_address(src->pages[i])); | |
4859 | } else { | |
4860 | size_t src_offset = get_eb_offset_in_page(src, 0); | |
4861 | size_t dst_offset = get_eb_offset_in_page(dst, 0); | |
4862 | ||
fbca46eb | 4863 | ASSERT(src->fs_info->nodesize < PAGE_SIZE); |
884b07d0 QW |
4864 | memcpy(page_address(dst->pages[0]) + dst_offset, |
4865 | page_address(src->pages[0]) + src_offset, | |
4866 | src->len); | |
4867 | } | |
58e8012c DS |
4868 | } |
4869 | ||
2b48966a DS |
4870 | void copy_extent_buffer(const struct extent_buffer *dst, |
4871 | const struct extent_buffer *src, | |
d1310b2e CM |
4872 | unsigned long dst_offset, unsigned long src_offset, |
4873 | unsigned long len) | |
4874 | { | |
4875 | u64 dst_len = dst->len; | |
4876 | size_t cur; | |
4877 | size_t offset; | |
4878 | struct page *page; | |
4879 | char *kaddr; | |
884b07d0 | 4880 | unsigned long i = get_eb_page_index(dst_offset); |
d1310b2e | 4881 | |
f98b6215 QW |
4882 | if (check_eb_range(dst, dst_offset, len) || |
4883 | check_eb_range(src, src_offset, len)) | |
4884 | return; | |
4885 | ||
d1310b2e CM |
4886 | WARN_ON(src->len != dst_len); |
4887 | ||
884b07d0 | 4888 | offset = get_eb_offset_in_page(dst, dst_offset); |
d1310b2e | 4889 | |
d397712b | 4890 | while (len > 0) { |
fb85fc9a | 4891 | page = dst->pages[i]; |
b8f95771 | 4892 | assert_eb_page_uptodate(dst, page); |
d1310b2e | 4893 | |
09cbfeaf | 4894 | cur = min(len, (unsigned long)(PAGE_SIZE - offset)); |
d1310b2e | 4895 | |
a6591715 | 4896 | kaddr = page_address(page); |
d1310b2e | 4897 | read_extent_buffer(src, kaddr + offset, src_offset, cur); |
d1310b2e CM |
4898 | |
4899 | src_offset += cur; | |
4900 | len -= cur; | |
4901 | offset = 0; | |
4902 | i++; | |
4903 | } | |
4904 | } | |
d1310b2e | 4905 | |
3e1e8bb7 OS |
4906 | /* |
4907 | * eb_bitmap_offset() - calculate the page and offset of the byte containing the | |
4908 | * given bit number | |
4909 | * @eb: the extent buffer | |
4910 | * @start: offset of the bitmap item in the extent buffer | |
4911 | * @nr: bit number | |
4912 | * @page_index: return index of the page in the extent buffer that contains the | |
4913 | * given bit number | |
4914 | * @page_offset: return offset into the page given by page_index | |
4915 | * | |
4916 | * This helper hides the ugliness of finding the byte in an extent buffer which | |
4917 | * contains a given bit. | |
4918 | */ | |
2b48966a | 4919 | static inline void eb_bitmap_offset(const struct extent_buffer *eb, |
3e1e8bb7 OS |
4920 | unsigned long start, unsigned long nr, |
4921 | unsigned long *page_index, | |
4922 | size_t *page_offset) | |
4923 | { | |
3e1e8bb7 OS |
4924 | size_t byte_offset = BIT_BYTE(nr); |
4925 | size_t offset; | |
4926 | ||
4927 | /* | |
4928 | * The byte we want is the offset of the extent buffer + the offset of | |
4929 | * the bitmap item in the extent buffer + the offset of the byte in the | |
4930 | * bitmap item. | |
4931 | */ | |
884b07d0 | 4932 | offset = start + offset_in_page(eb->start) + byte_offset; |
3e1e8bb7 | 4933 | |
09cbfeaf | 4934 | *page_index = offset >> PAGE_SHIFT; |
7073017a | 4935 | *page_offset = offset_in_page(offset); |
3e1e8bb7 OS |
4936 | } |
4937 | ||
43dd529a DS |
4938 | /* |
4939 | * Determine whether a bit in a bitmap item is set. | |
4940 | * | |
4941 | * @eb: the extent buffer | |
4942 | * @start: offset of the bitmap item in the extent buffer | |
4943 | * @nr: bit number to test | |
3e1e8bb7 | 4944 | */ |
2b48966a | 4945 | int extent_buffer_test_bit(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
4946 | unsigned long nr) |
4947 | { | |
2fe1d551 | 4948 | u8 *kaddr; |
3e1e8bb7 OS |
4949 | struct page *page; |
4950 | unsigned long i; | |
4951 | size_t offset; | |
4952 | ||
4953 | eb_bitmap_offset(eb, start, nr, &i, &offset); | |
4954 | page = eb->pages[i]; | |
b8f95771 | 4955 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4956 | kaddr = page_address(page); |
4957 | return 1U & (kaddr[offset] >> (nr & (BITS_PER_BYTE - 1))); | |
4958 | } | |
4959 | ||
43dd529a DS |
4960 | /* |
4961 | * Set an area of a bitmap to 1. | |
4962 | * | |
4963 | * @eb: the extent buffer | |
4964 | * @start: offset of the bitmap item in the extent buffer | |
4965 | * @pos: bit number of the first bit | |
4966 | * @len: number of bits to set | |
3e1e8bb7 | 4967 | */ |
2b48966a | 4968 | void extent_buffer_bitmap_set(const struct extent_buffer *eb, unsigned long start, |
3e1e8bb7 OS |
4969 | unsigned long pos, unsigned long len) |
4970 | { | |
2fe1d551 | 4971 | u8 *kaddr; |
3e1e8bb7 OS |
4972 | struct page *page; |
4973 | unsigned long i; | |
4974 | size_t offset; | |
4975 | const unsigned int size = pos + len; | |
4976 | int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 4977 | u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
4978 | |
4979 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
4980 | page = eb->pages[i]; | |
b8f95771 | 4981 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4982 | kaddr = page_address(page); |
4983 | ||
4984 | while (len >= bits_to_set) { | |
4985 | kaddr[offset] |= mask_to_set; | |
4986 | len -= bits_to_set; | |
4987 | bits_to_set = BITS_PER_BYTE; | |
9c894696 | 4988 | mask_to_set = ~0; |
09cbfeaf | 4989 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
4990 | offset = 0; |
4991 | page = eb->pages[++i]; | |
b8f95771 | 4992 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
4993 | kaddr = page_address(page); |
4994 | } | |
4995 | } | |
4996 | if (len) { | |
4997 | mask_to_set &= BITMAP_LAST_BYTE_MASK(size); | |
4998 | kaddr[offset] |= mask_to_set; | |
4999 | } | |
5000 | } | |
5001 | ||
5002 | ||
43dd529a DS |
5003 | /* |
5004 | * Clear an area of a bitmap. | |
5005 | * | |
5006 | * @eb: the extent buffer | |
5007 | * @start: offset of the bitmap item in the extent buffer | |
5008 | * @pos: bit number of the first bit | |
5009 | * @len: number of bits to clear | |
3e1e8bb7 | 5010 | */ |
2b48966a DS |
5011 | void extent_buffer_bitmap_clear(const struct extent_buffer *eb, |
5012 | unsigned long start, unsigned long pos, | |
5013 | unsigned long len) | |
3e1e8bb7 | 5014 | { |
2fe1d551 | 5015 | u8 *kaddr; |
3e1e8bb7 OS |
5016 | struct page *page; |
5017 | unsigned long i; | |
5018 | size_t offset; | |
5019 | const unsigned int size = pos + len; | |
5020 | int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE); | |
2fe1d551 | 5021 | u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos); |
3e1e8bb7 OS |
5022 | |
5023 | eb_bitmap_offset(eb, start, pos, &i, &offset); | |
5024 | page = eb->pages[i]; | |
b8f95771 | 5025 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
5026 | kaddr = page_address(page); |
5027 | ||
5028 | while (len >= bits_to_clear) { | |
5029 | kaddr[offset] &= ~mask_to_clear; | |
5030 | len -= bits_to_clear; | |
5031 | bits_to_clear = BITS_PER_BYTE; | |
9c894696 | 5032 | mask_to_clear = ~0; |
09cbfeaf | 5033 | if (++offset >= PAGE_SIZE && len > 0) { |
3e1e8bb7 OS |
5034 | offset = 0; |
5035 | page = eb->pages[++i]; | |
b8f95771 | 5036 | assert_eb_page_uptodate(eb, page); |
3e1e8bb7 OS |
5037 | kaddr = page_address(page); |
5038 | } | |
5039 | } | |
5040 | if (len) { | |
5041 | mask_to_clear &= BITMAP_LAST_BYTE_MASK(size); | |
5042 | kaddr[offset] &= ~mask_to_clear; | |
5043 | } | |
5044 | } | |
5045 | ||
3387206f ST |
5046 | static inline bool areas_overlap(unsigned long src, unsigned long dst, unsigned long len) |
5047 | { | |
5048 | unsigned long distance = (src > dst) ? src - dst : dst - src; | |
5049 | return distance < len; | |
5050 | } | |
5051 | ||
d1310b2e CM |
5052 | static void copy_pages(struct page *dst_page, struct page *src_page, |
5053 | unsigned long dst_off, unsigned long src_off, | |
5054 | unsigned long len) | |
5055 | { | |
a6591715 | 5056 | char *dst_kaddr = page_address(dst_page); |
d1310b2e | 5057 | char *src_kaddr; |
727011e0 | 5058 | int must_memmove = 0; |
d1310b2e | 5059 | |
3387206f | 5060 | if (dst_page != src_page) { |
a6591715 | 5061 | src_kaddr = page_address(src_page); |
3387206f | 5062 | } else { |
d1310b2e | 5063 | src_kaddr = dst_kaddr; |
727011e0 CM |
5064 | if (areas_overlap(src_off, dst_off, len)) |
5065 | must_memmove = 1; | |
3387206f | 5066 | } |
d1310b2e | 5067 | |
727011e0 CM |
5068 | if (must_memmove) |
5069 | memmove(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
5070 | else | |
5071 | memcpy(dst_kaddr + dst_off, src_kaddr + src_off, len); | |
d1310b2e CM |
5072 | } |
5073 | ||
2b48966a DS |
5074 | void memcpy_extent_buffer(const struct extent_buffer *dst, |
5075 | unsigned long dst_offset, unsigned long src_offset, | |
5076 | unsigned long len) | |
d1310b2e CM |
5077 | { |
5078 | size_t cur; | |
5079 | size_t dst_off_in_page; | |
5080 | size_t src_off_in_page; | |
d1310b2e CM |
5081 | unsigned long dst_i; |
5082 | unsigned long src_i; | |
5083 | ||
f98b6215 QW |
5084 | if (check_eb_range(dst, dst_offset, len) || |
5085 | check_eb_range(dst, src_offset, len)) | |
5086 | return; | |
d1310b2e | 5087 | |
d397712b | 5088 | while (len > 0) { |
884b07d0 QW |
5089 | dst_off_in_page = get_eb_offset_in_page(dst, dst_offset); |
5090 | src_off_in_page = get_eb_offset_in_page(dst, src_offset); | |
d1310b2e | 5091 | |
884b07d0 QW |
5092 | dst_i = get_eb_page_index(dst_offset); |
5093 | src_i = get_eb_page_index(src_offset); | |
d1310b2e | 5094 | |
09cbfeaf | 5095 | cur = min(len, (unsigned long)(PAGE_SIZE - |
d1310b2e CM |
5096 | src_off_in_page)); |
5097 | cur = min_t(unsigned long, cur, | |
09cbfeaf | 5098 | (unsigned long)(PAGE_SIZE - dst_off_in_page)); |
d1310b2e | 5099 | |
fb85fc9a | 5100 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5101 | dst_off_in_page, src_off_in_page, cur); |
5102 | ||
5103 | src_offset += cur; | |
5104 | dst_offset += cur; | |
5105 | len -= cur; | |
5106 | } | |
5107 | } | |
d1310b2e | 5108 | |
2b48966a DS |
5109 | void memmove_extent_buffer(const struct extent_buffer *dst, |
5110 | unsigned long dst_offset, unsigned long src_offset, | |
5111 | unsigned long len) | |
d1310b2e CM |
5112 | { |
5113 | size_t cur; | |
5114 | size_t dst_off_in_page; | |
5115 | size_t src_off_in_page; | |
5116 | unsigned long dst_end = dst_offset + len - 1; | |
5117 | unsigned long src_end = src_offset + len - 1; | |
d1310b2e CM |
5118 | unsigned long dst_i; |
5119 | unsigned long src_i; | |
5120 | ||
f98b6215 QW |
5121 | if (check_eb_range(dst, dst_offset, len) || |
5122 | check_eb_range(dst, src_offset, len)) | |
5123 | return; | |
727011e0 | 5124 | if (dst_offset < src_offset) { |
d1310b2e CM |
5125 | memcpy_extent_buffer(dst, dst_offset, src_offset, len); |
5126 | return; | |
5127 | } | |
d397712b | 5128 | while (len > 0) { |
884b07d0 QW |
5129 | dst_i = get_eb_page_index(dst_end); |
5130 | src_i = get_eb_page_index(src_end); | |
d1310b2e | 5131 | |
884b07d0 QW |
5132 | dst_off_in_page = get_eb_offset_in_page(dst, dst_end); |
5133 | src_off_in_page = get_eb_offset_in_page(dst, src_end); | |
d1310b2e CM |
5134 | |
5135 | cur = min_t(unsigned long, len, src_off_in_page + 1); | |
5136 | cur = min(cur, dst_off_in_page + 1); | |
fb85fc9a | 5137 | copy_pages(dst->pages[dst_i], dst->pages[src_i], |
d1310b2e CM |
5138 | dst_off_in_page - cur + 1, |
5139 | src_off_in_page - cur + 1, cur); | |
5140 | ||
5141 | dst_end -= cur; | |
5142 | src_end -= cur; | |
5143 | len -= cur; | |
5144 | } | |
5145 | } | |
6af118ce | 5146 | |
01cd3909 | 5147 | #define GANG_LOOKUP_SIZE 16 |
d1e86e3f QW |
5148 | static struct extent_buffer *get_next_extent_buffer( |
5149 | struct btrfs_fs_info *fs_info, struct page *page, u64 bytenr) | |
5150 | { | |
01cd3909 DS |
5151 | struct extent_buffer *gang[GANG_LOOKUP_SIZE]; |
5152 | struct extent_buffer *found = NULL; | |
d1e86e3f | 5153 | u64 page_start = page_offset(page); |
01cd3909 | 5154 | u64 cur = page_start; |
d1e86e3f QW |
5155 | |
5156 | ASSERT(in_range(bytenr, page_start, PAGE_SIZE)); | |
d1e86e3f QW |
5157 | lockdep_assert_held(&fs_info->buffer_lock); |
5158 | ||
01cd3909 DS |
5159 | while (cur < page_start + PAGE_SIZE) { |
5160 | int ret; | |
5161 | int i; | |
5162 | ||
5163 | ret = radix_tree_gang_lookup(&fs_info->buffer_radix, | |
5164 | (void **)gang, cur >> fs_info->sectorsize_bits, | |
5165 | min_t(unsigned int, GANG_LOOKUP_SIZE, | |
5166 | PAGE_SIZE / fs_info->nodesize)); | |
5167 | if (ret == 0) | |
5168 | goto out; | |
5169 | for (i = 0; i < ret; i++) { | |
5170 | /* Already beyond page end */ | |
5171 | if (gang[i]->start >= page_start + PAGE_SIZE) | |
5172 | goto out; | |
5173 | /* Found one */ | |
5174 | if (gang[i]->start >= bytenr) { | |
5175 | found = gang[i]; | |
5176 | goto out; | |
5177 | } | |
5178 | } | |
5179 | cur = gang[ret - 1]->start + gang[ret - 1]->len; | |
d1e86e3f | 5180 | } |
01cd3909 DS |
5181 | out: |
5182 | return found; | |
d1e86e3f QW |
5183 | } |
5184 | ||
5185 | static int try_release_subpage_extent_buffer(struct page *page) | |
5186 | { | |
5187 | struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb); | |
5188 | u64 cur = page_offset(page); | |
5189 | const u64 end = page_offset(page) + PAGE_SIZE; | |
5190 | int ret; | |
5191 | ||
5192 | while (cur < end) { | |
5193 | struct extent_buffer *eb = NULL; | |
5194 | ||
5195 | /* | |
5196 | * Unlike try_release_extent_buffer() which uses page->private | |
5197 | * to grab buffer, for subpage case we rely on radix tree, thus | |
5198 | * we need to ensure radix tree consistency. | |
5199 | * | |
5200 | * We also want an atomic snapshot of the radix tree, thus go | |
5201 | * with spinlock rather than RCU. | |
5202 | */ | |
5203 | spin_lock(&fs_info->buffer_lock); | |
5204 | eb = get_next_extent_buffer(fs_info, page, cur); | |
5205 | if (!eb) { | |
5206 | /* No more eb in the page range after or at cur */ | |
5207 | spin_unlock(&fs_info->buffer_lock); | |
5208 | break; | |
5209 | } | |
5210 | cur = eb->start + eb->len; | |
5211 | ||
5212 | /* | |
5213 | * The same as try_release_extent_buffer(), to ensure the eb | |
5214 | * won't disappear out from under us. | |
5215 | */ | |
5216 | spin_lock(&eb->refs_lock); | |
5217 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { | |
5218 | spin_unlock(&eb->refs_lock); | |
5219 | spin_unlock(&fs_info->buffer_lock); | |
5220 | break; | |
5221 | } | |
5222 | spin_unlock(&fs_info->buffer_lock); | |
5223 | ||
5224 | /* | |
5225 | * If tree ref isn't set then we know the ref on this eb is a | |
5226 | * real ref, so just return, this eb will likely be freed soon | |
5227 | * anyway. | |
5228 | */ | |
5229 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { | |
5230 | spin_unlock(&eb->refs_lock); | |
5231 | break; | |
5232 | } | |
5233 | ||
5234 | /* | |
5235 | * Here we don't care about the return value, we will always | |
5236 | * check the page private at the end. And | |
5237 | * release_extent_buffer() will release the refs_lock. | |
5238 | */ | |
5239 | release_extent_buffer(eb); | |
5240 | } | |
5241 | /* | |
5242 | * Finally to check if we have cleared page private, as if we have | |
5243 | * released all ebs in the page, the page private should be cleared now. | |
5244 | */ | |
5245 | spin_lock(&page->mapping->private_lock); | |
5246 | if (!PagePrivate(page)) | |
5247 | ret = 1; | |
5248 | else | |
5249 | ret = 0; | |
5250 | spin_unlock(&page->mapping->private_lock); | |
5251 | return ret; | |
5252 | ||
5253 | } | |
5254 | ||
f7a52a40 | 5255 | int try_release_extent_buffer(struct page *page) |
19fe0a8b | 5256 | { |
6af118ce | 5257 | struct extent_buffer *eb; |
6af118ce | 5258 | |
fbca46eb | 5259 | if (btrfs_sb(page->mapping->host->i_sb)->nodesize < PAGE_SIZE) |
d1e86e3f QW |
5260 | return try_release_subpage_extent_buffer(page); |
5261 | ||
3083ee2e | 5262 | /* |
d1e86e3f QW |
5263 | * We need to make sure nobody is changing page->private, as we rely on |
5264 | * page->private as the pointer to extent buffer. | |
3083ee2e JB |
5265 | */ |
5266 | spin_lock(&page->mapping->private_lock); | |
5267 | if (!PagePrivate(page)) { | |
5268 | spin_unlock(&page->mapping->private_lock); | |
4f2de97a | 5269 | return 1; |
45f49bce | 5270 | } |
6af118ce | 5271 | |
3083ee2e JB |
5272 | eb = (struct extent_buffer *)page->private; |
5273 | BUG_ON(!eb); | |
19fe0a8b MX |
5274 | |
5275 | /* | |
3083ee2e JB |
5276 | * This is a little awful but should be ok, we need to make sure that |
5277 | * the eb doesn't disappear out from under us while we're looking at | |
5278 | * this page. | |
19fe0a8b | 5279 | */ |
3083ee2e | 5280 | spin_lock(&eb->refs_lock); |
0b32f4bb | 5281 | if (atomic_read(&eb->refs) != 1 || extent_buffer_under_io(eb)) { |
3083ee2e JB |
5282 | spin_unlock(&eb->refs_lock); |
5283 | spin_unlock(&page->mapping->private_lock); | |
5284 | return 0; | |
b9473439 | 5285 | } |
3083ee2e | 5286 | spin_unlock(&page->mapping->private_lock); |
897ca6e9 | 5287 | |
19fe0a8b | 5288 | /* |
3083ee2e JB |
5289 | * If tree ref isn't set then we know the ref on this eb is a real ref, |
5290 | * so just return, this page will likely be freed soon anyway. | |
19fe0a8b | 5291 | */ |
3083ee2e JB |
5292 | if (!test_and_clear_bit(EXTENT_BUFFER_TREE_REF, &eb->bflags)) { |
5293 | spin_unlock(&eb->refs_lock); | |
5294 | return 0; | |
b9473439 | 5295 | } |
19fe0a8b | 5296 | |
f7a52a40 | 5297 | return release_extent_buffer(eb); |
6af118ce | 5298 | } |
bfb484d9 JB |
5299 | |
5300 | /* | |
5301 | * btrfs_readahead_tree_block - attempt to readahead a child block | |
5302 | * @fs_info: the fs_info | |
5303 | * @bytenr: bytenr to read | |
3fbaf258 | 5304 | * @owner_root: objectid of the root that owns this eb |
bfb484d9 | 5305 | * @gen: generation for the uptodate check, can be 0 |
3fbaf258 | 5306 | * @level: level for the eb |
bfb484d9 JB |
5307 | * |
5308 | * Attempt to readahead a tree block at @bytenr. If @gen is 0 then we do a | |
5309 | * normal uptodate check of the eb, without checking the generation. If we have | |
5310 | * to read the block we will not block on anything. | |
5311 | */ | |
5312 | void btrfs_readahead_tree_block(struct btrfs_fs_info *fs_info, | |
3fbaf258 | 5313 | u64 bytenr, u64 owner_root, u64 gen, int level) |
bfb484d9 | 5314 | { |
947a6299 QW |
5315 | struct btrfs_tree_parent_check check = { |
5316 | .has_first_key = 0, | |
5317 | .level = level, | |
5318 | .transid = gen | |
5319 | }; | |
bfb484d9 JB |
5320 | struct extent_buffer *eb; |
5321 | int ret; | |
5322 | ||
3fbaf258 | 5323 | eb = btrfs_find_create_tree_block(fs_info, bytenr, owner_root, level); |
bfb484d9 JB |
5324 | if (IS_ERR(eb)) |
5325 | return; | |
5326 | ||
5327 | if (btrfs_buffer_uptodate(eb, gen, 1)) { | |
5328 | free_extent_buffer(eb); | |
5329 | return; | |
5330 | } | |
5331 | ||
947a6299 | 5332 | ret = read_extent_buffer_pages(eb, WAIT_NONE, 0, &check); |
bfb484d9 JB |
5333 | if (ret < 0) |
5334 | free_extent_buffer_stale(eb); | |
5335 | else | |
5336 | free_extent_buffer(eb); | |
5337 | } | |
5338 | ||
5339 | /* | |
5340 | * btrfs_readahead_node_child - readahead a node's child block | |
5341 | * @node: parent node we're reading from | |
5342 | * @slot: slot in the parent node for the child we want to read | |
5343 | * | |
5344 | * A helper for btrfs_readahead_tree_block, we simply read the bytenr pointed at | |
5345 | * the slot in the node provided. | |
5346 | */ | |
5347 | void btrfs_readahead_node_child(struct extent_buffer *node, int slot) | |
5348 | { | |
5349 | btrfs_readahead_tree_block(node->fs_info, | |
5350 | btrfs_node_blockptr(node, slot), | |
3fbaf258 JB |
5351 | btrfs_header_owner(node), |
5352 | btrfs_node_ptr_generation(node, slot), | |
5353 | btrfs_header_level(node) - 1); | |
bfb484d9 | 5354 | } |