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