1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2010 Red Hat, Inc.
4 * Copyright (C) 2016-2023 Christoph Hellwig.
5 */
6 #include <linux/module.h>
7 #include <linux/compiler.h>
8 #include <linux/fs.h>
9 #include <linux/iomap.h>
10 #include <linux/pagemap.h>
11 #include <linux/uio.h>
12 #include <linux/buffer_head.h>
13 #include <linux/dax.h>
14 #include <linux/writeback.h>
15 #include <linux/list_sort.h>
16 #include <linux/swap.h>
17 #include <linux/bio.h>
18 #include <linux/sched/signal.h>
19 #include <linux/migrate.h>
20 #include "trace.h"
21
22 #include "../internal.h"
23
24 #define IOEND_BATCH_SIZE 4096
25
26 typedef int (*iomap_punch_t)(struct inode *inode, loff_t offset, loff_t length);
27 /*
28 * Structure allocated for each folio to track per-block uptodate, dirty state
29 * and I/O completions.
30 */
31 struct iomap_folio_state {
32 spinlock_t state_lock;
33 unsigned int read_bytes_pending;
34 atomic_t write_bytes_pending;
35
36 /*
37 * Each block has two bits in this bitmap:
38 * Bits [0..blocks_per_folio) has the uptodate status.
39 * Bits [b_p_f...(2*b_p_f)) has the dirty status.
40 */
41 unsigned long state[];
42 };
43
44 static struct bio_set iomap_ioend_bioset;
45
46 static inline bool ifs_is_fully_uptodate(struct folio *folio,
47 struct iomap_folio_state *ifs)
48 {
49 struct inode *inode = folio->mapping->host;
50
51 return bitmap_full(ifs->state, i_blocks_per_folio(inode, folio));
52 }
53
54 static inline bool ifs_block_is_uptodate(struct iomap_folio_state *ifs,
55 unsigned int block)
56 {
57 return test_bit(block, ifs->state);
58 }
59
60 static bool ifs_set_range_uptodate(struct folio *folio,
61 struct iomap_folio_state *ifs, size_t off, size_t len)
62 {
63 struct inode *inode = folio->mapping->host;
64 unsigned int first_blk = off >> inode->i_blkbits;
65 unsigned int last_blk = (off + len - 1) >> inode->i_blkbits;
66 unsigned int nr_blks = last_blk - first_blk + 1;
67
68 bitmap_set(ifs->state, first_blk, nr_blks);
69 return ifs_is_fully_uptodate(folio, ifs);
70 }
71
72 static void iomap_set_range_uptodate(struct folio *folio, size_t off,
73 size_t len)
74 {
75 struct iomap_folio_state *ifs = folio->private;
76 unsigned long flags;
77 bool uptodate = true;
78
79 if (ifs) {
80 spin_lock_irqsave(&ifs->state_lock, flags);
81 uptodate = ifs_set_range_uptodate(folio, ifs, off, len);
82 spin_unlock_irqrestore(&ifs->state_lock, flags);
83 }
84
85 if (uptodate)
86 folio_mark_uptodate(folio);
87 }
88
89 static inline bool ifs_block_is_dirty(struct folio *folio,
90 struct iomap_folio_state *ifs, int block)
91 {
92 struct inode *inode = folio->mapping->host;
93 unsigned int blks_per_folio = i_blocks_per_folio(inode, folio);
94
95 return test_bit(block + blks_per_folio, ifs->state);
96 }
97
98 static unsigned ifs_find_dirty_range(struct folio *folio,
99 struct iomap_folio_state *ifs, u64 *range_start, u64 range_end)
100 {
101 struct inode *inode = folio->mapping->host;
102 unsigned start_blk =
103 offset_in_folio(folio, *range_start) >> inode->i_blkbits;
104 unsigned end_blk = min_not_zero(
105 offset_in_folio(folio, range_end) >> inode->i_blkbits,
106 i_blocks_per_folio(inode, folio));
107 unsigned nblks = 1;
108
109 while (!ifs_block_is_dirty(folio, ifs, start_blk))
110 if (++start_blk == end_blk)
111 return 0;
112
113 while (start_blk + nblks < end_blk) {
114 if (!ifs_block_is_dirty(folio, ifs, start_blk + nblks))
115 break;
116 nblks++;
117 }
118
119 *range_start = folio_pos(folio) + (start_blk << inode->i_blkbits);
120 return nblks << inode->i_blkbits;
121 }
122
123 static unsigned iomap_find_dirty_range(struct folio *folio, u64 *range_start,
124 u64 range_end)
125 {
126 struct iomap_folio_state *ifs = folio->private;
127
128 if (*range_start >= range_end)
129 return 0;
130
131 if (ifs)
132 return ifs_find_dirty_range(folio, ifs, range_start, range_end);
133 return range_end - *range_start;
134 }
135
136 static void ifs_clear_range_dirty(struct folio *folio,
137 struct iomap_folio_state *ifs, size_t off, size_t len)
138 {
139 struct inode *inode = folio->mapping->host;
140 unsigned int blks_per_folio = i_blocks_per_folio(inode, folio);
141 unsigned int first_blk = (off >> inode->i_blkbits);
142 unsigned int last_blk = (off + len - 1) >> inode->i_blkbits;
143 unsigned int nr_blks = last_blk - first_blk + 1;
144 unsigned long flags;
145
146 spin_lock_irqsave(&ifs->state_lock, flags);
147 bitmap_clear(ifs->state, first_blk + blks_per_folio, nr_blks);
148 spin_unlock_irqrestore(&ifs->state_lock, flags);
149 }
150
151 static void iomap_clear_range_dirty(struct folio *folio, size_t off, size_t len)
152 {
153 struct iomap_folio_state *ifs = folio->private;
154
155 if (ifs)
156 ifs_clear_range_dirty(folio, ifs, off, len);
157 }
158
159 static void ifs_set_range_dirty(struct folio *folio,
160 struct iomap_folio_state *ifs, size_t off, size_t len)
161 {
162 struct inode *inode = folio->mapping->host;
163 unsigned int blks_per_folio = i_blocks_per_folio(inode, folio);
164 unsigned int first_blk = (off >> inode->i_blkbits);
165 unsigned int last_blk = (off + len - 1) >> inode->i_blkbits;
166 unsigned int nr_blks = last_blk - first_blk + 1;
167 unsigned long flags;
168
169 spin_lock_irqsave(&ifs->state_lock, flags);
170 bitmap_set(ifs->state, first_blk + blks_per_folio, nr_blks);
171 spin_unlock_irqrestore(&ifs->state_lock, flags);
172 }
173
174 static void iomap_set_range_dirty(struct folio *folio, size_t off, size_t len)
175 {
176 struct iomap_folio_state *ifs = folio->private;
177
178 if (ifs)
179 ifs_set_range_dirty(folio, ifs, off, len);
180 }
181
182 static struct iomap_folio_state *ifs_alloc(struct inode *inode,
183 struct folio *folio, unsigned int flags)
184 {
185 struct iomap_folio_state *ifs = folio->private;
186 unsigned int nr_blocks = i_blocks_per_folio(inode, folio);
187 gfp_t gfp;
188
189 if (ifs || nr_blocks <= 1)
190 return ifs;
191
192 if (flags & IOMAP_NOWAIT)
193 gfp = GFP_NOWAIT;
194 else
195 gfp = GFP_NOFS | __GFP_NOFAIL;
196
197 /*
198 * ifs->state tracks two sets of state flags when the
199 * filesystem block size is smaller than the folio size.
200 * The first state tracks per-block uptodate and the
201 * second tracks per-block dirty state.
202 */
203 ifs = kzalloc(struct_size(ifs, state,
204 BITS_TO_LONGS(2 * nr_blocks)), gfp);
205 if (!ifs)
206 return ifs;
207
208 spin_lock_init(&ifs->state_lock);
209 if (folio_test_uptodate(folio))
210 bitmap_set(ifs->state, 0, nr_blocks);
211 if (folio_test_dirty(folio))
212 bitmap_set(ifs->state, nr_blocks, nr_blocks);
213 folio_attach_private(folio, ifs);
214
215 return ifs;
216 }
217
218 static void ifs_free(struct folio *folio)
219 {
220 struct iomap_folio_state *ifs = folio_detach_private(folio);
221
222 if (!ifs)
223 return;
224 WARN_ON_ONCE(ifs->read_bytes_pending != 0);
225 WARN_ON_ONCE(atomic_read(&ifs->write_bytes_pending));
226 WARN_ON_ONCE(ifs_is_fully_uptodate(folio, ifs) !=
227 folio_test_uptodate(folio));
228 kfree(ifs);
229 }
230
231 /*
232 * Calculate the range inside the folio that we actually need to read.
233 */
234 static void iomap_adjust_read_range(struct inode *inode, struct folio *folio,
235 loff_t *pos, loff_t length, size_t *offp, size_t *lenp)
236 {
237 struct iomap_folio_state *ifs = folio->private;
238 loff_t orig_pos = *pos;
239 loff_t isize = i_size_read(inode);
240 unsigned block_bits = inode->i_blkbits;
241 unsigned block_size = (1 << block_bits);
242 size_t poff = offset_in_folio(folio, *pos);
243 size_t plen = min_t(loff_t, folio_size(folio) - poff, length);
244 size_t orig_plen = plen;
245 unsigned first = poff >> block_bits;
246 unsigned last = (poff + plen - 1) >> block_bits;
247
248 /*
249 * If the block size is smaller than the page size, we need to check the
250 * per-block uptodate status and adjust the offset and length if needed
251 * to avoid reading in already uptodate ranges.
252 */
253 if (ifs) {
254 unsigned int i;
255
256 /* move forward for each leading block marked uptodate */
257 for (i = first; i <= last; i++) {
258 if (!ifs_block_is_uptodate(ifs, i))
259 break;
260 *pos += block_size;
261 poff += block_size;
262 plen -= block_size;
263 first++;
264 }
265
266 /* truncate len if we find any trailing uptodate block(s) */
267 for ( ; i <= last; i++) {
268 if (ifs_block_is_uptodate(ifs, i)) {
269 plen -= (last - i + 1) * block_size;
270 last = i - 1;
271 break;
272 }
273 }
274 }
275
276 /*
277 * If the extent spans the block that contains the i_size, we need to
278 * handle both halves separately so that we properly zero data in the
279 * page cache for blocks that are entirely outside of i_size.
280 */
281 if (orig_pos <= isize && orig_pos + orig_plen > isize) {
282 unsigned end = offset_in_folio(folio, isize - 1) >> block_bits;
283
284 if (first <= end && last > end)
285 plen -= (last - end) * block_size;
286 }
287
288 *offp = poff;
289 *lenp = plen;
290 }
291
292 static void iomap_finish_folio_read(struct folio *folio, size_t off,
293 size_t len, int error)
294 {
295 struct iomap_folio_state *ifs = folio->private;
296 bool uptodate = !error;
297 bool finished = true;
298
299 if (ifs) {
300 unsigned long flags;
301
302 spin_lock_irqsave(&ifs->state_lock, flags);
303 if (!error)
304 uptodate = ifs_set_range_uptodate(folio, ifs, off, len);
305 ifs->read_bytes_pending -= len;
306 finished = !ifs->read_bytes_pending;
307 spin_unlock_irqrestore(&ifs->state_lock, flags);
308 }
309
310 if (finished)
311 folio_end_read(folio, uptodate);
312 }
313
314 static void iomap_read_end_io(struct bio *bio)
315 {
316 int error = blk_status_to_errno(bio->bi_status);
317 struct folio_iter fi;
318
319 bio_for_each_folio_all(fi, bio)
320 iomap_finish_folio_read(fi.folio, fi.offset, fi.length, error);
321 bio_put(bio);
322 }
323
324 struct iomap_readpage_ctx {
325 struct folio *cur_folio;
326 bool cur_folio_in_bio;
327 struct bio *bio;
328 struct readahead_control *rac;
329 };
330
331 /**
332 * iomap_read_inline_data - copy inline data into the page cache
333 * @iter: iteration structure
334 * @folio: folio to copy to
335 *
336 * Copy the inline data in @iter into @folio and zero out the rest of the folio.
337 * Only a single IOMAP_INLINE extent is allowed at the end of each file.
338 * Returns zero for success to complete the read, or the usual negative errno.
339 */
340 static int iomap_read_inline_data(const struct iomap_iter *iter,
341 struct folio *folio)
342 {
343 const struct iomap *iomap = iomap_iter_srcmap(iter);
344 size_t size = i_size_read(iter->inode) - iomap->offset;
345 size_t offset = offset_in_folio(folio, iomap->offset);
346
347 if (folio_test_uptodate(folio))
348 return 0;
349
350 if (WARN_ON_ONCE(size > iomap->length))
351 return -EIO;
352 if (offset > 0)
353 ifs_alloc(iter->inode, folio, iter->flags);
354
355 folio_fill_tail(folio, offset, iomap->inline_data, size);
356 iomap_set_range_uptodate(folio, offset, folio_size(folio) - offset);
357 return 0;
358 }
359
360 static inline bool iomap_block_needs_zeroing(const struct iomap_iter *iter,
361 loff_t pos)
362 {
363 const struct iomap *srcmap = iomap_iter_srcmap(iter);
364
365 return srcmap->type != IOMAP_MAPPED ||
366 (srcmap->flags & IOMAP_F_NEW) ||
367 pos >= i_size_read(iter->inode);
368 }
369
370 static loff_t iomap_readpage_iter(const struct iomap_iter *iter,
371 struct iomap_readpage_ctx *ctx, loff_t offset)
372 {
373 const struct iomap *iomap = &iter->iomap;
374 loff_t pos = iter->pos + offset;
375 loff_t length = iomap_length(iter) - offset;
376 struct folio *folio = ctx->cur_folio;
377 struct iomap_folio_state *ifs;
378 loff_t orig_pos = pos;
379 size_t poff, plen;
380 sector_t sector;
381
382 if (iomap->type == IOMAP_INLINE)
383 return iomap_read_inline_data(iter, folio);
384
385 /* zero post-eof blocks as the page may be mapped */
386 ifs = ifs_alloc(iter->inode, folio, iter->flags);
387 iomap_adjust_read_range(iter->inode, folio, &pos, length, &poff, &plen);
388 if (plen == 0)
389 goto done;
390
391 if (iomap_block_needs_zeroing(iter, pos)) {
392 folio_zero_range(folio, poff, plen);
393 iomap_set_range_uptodate(folio, poff, plen);
394 goto done;
395 }
396
397 ctx->cur_folio_in_bio = true;
398 if (ifs) {
399 spin_lock_irq(&ifs->state_lock);
400 ifs->read_bytes_pending += plen;
401 spin_unlock_irq(&ifs->state_lock);
402 }
403
404 sector = iomap_sector(iomap, pos);
405 if (!ctx->bio ||
406 bio_end_sector(ctx->bio) != sector ||
407 !bio_add_folio(ctx->bio, folio, plen, poff)) {
408 gfp_t gfp = mapping_gfp_constraint(folio->mapping, GFP_KERNEL);
409 gfp_t orig_gfp = gfp;
410 unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE);
411
412 if (ctx->bio)
413 submit_bio(ctx->bio);
414
415 if (ctx->rac) /* same as readahead_gfp_mask */
416 gfp |= __GFP_NORETRY | __GFP_NOWARN;
417 ctx->bio = bio_alloc(iomap->bdev, bio_max_segs(nr_vecs),
418 REQ_OP_READ, gfp);
419 /*
420 * If the bio_alloc fails, try it again for a single page to
421 * avoid having to deal with partial page reads. This emulates
422 * what do_mpage_read_folio does.
423 */
424 if (!ctx->bio) {
425 ctx->bio = bio_alloc(iomap->bdev, 1, REQ_OP_READ,
426 orig_gfp);
427 }
428 if (ctx->rac)
429 ctx->bio->bi_opf |= REQ_RAHEAD;
430 ctx->bio->bi_iter.bi_sector = sector;
431 ctx->bio->bi_end_io = iomap_read_end_io;
432 bio_add_folio_nofail(ctx->bio, folio, plen, poff);
433 }
434
435 done:
436 /*
437 * Move the caller beyond our range so that it keeps making progress.
438 * For that, we have to include any leading non-uptodate ranges, but
439 * we can skip trailing ones as they will be handled in the next
440 * iteration.
441 */
442 return pos - orig_pos + plen;
443 }
444
445 static loff_t iomap_read_folio_iter(const struct iomap_iter *iter,
446 struct iomap_readpage_ctx *ctx)
447 {
448 struct folio *folio = ctx->cur_folio;
449 size_t offset = offset_in_folio(folio, iter->pos);
450 loff_t length = min_t(loff_t, folio_size(folio) - offset,
451 iomap_length(iter));
452 loff_t done, ret;
453
454 for (done = 0; done < length; done += ret) {
455 ret = iomap_readpage_iter(iter, ctx, done);
456 if (ret <= 0)
457 return ret;
458 }
459
460 return done;
461 }
462
463 int iomap_read_folio(struct folio *folio, const struct iomap_ops *ops)
464 {
465 struct iomap_iter iter = {
466 .inode = folio->mapping->host,
467 .pos = folio_pos(folio),
468 .len = folio_size(folio),
469 };
470 struct iomap_readpage_ctx ctx = {
471 .cur_folio = folio,
472 };
473 int ret;
474
475 trace_iomap_readpage(iter.inode, 1);
476
477 while ((ret = iomap_iter(&iter, ops)) > 0)
478 iter.processed = iomap_read_folio_iter(&iter, &ctx);
479
480 if (ctx.bio) {
481 submit_bio(ctx.bio);
482 WARN_ON_ONCE(!ctx.cur_folio_in_bio);
483 } else {
484 WARN_ON_ONCE(ctx.cur_folio_in_bio);
485 folio_unlock(folio);
486 }
487
488 /*
489 * Just like mpage_readahead and block_read_full_folio, we always
490 * return 0 and just set the folio error flag on errors. This
491 * should be cleaned up throughout the stack eventually.
492 */
493 return 0;
494 }
495 EXPORT_SYMBOL_GPL(iomap_read_folio);
496
497 static loff_t iomap_readahead_iter(const struct iomap_iter *iter,
498 struct iomap_readpage_ctx *ctx)
499 {
500 loff_t length = iomap_length(iter);
501 loff_t done, ret;
502
503 for (done = 0; done < length; done += ret) {
504 if (ctx->cur_folio &&
505 offset_in_folio(ctx->cur_folio, iter->pos + done) == 0) {
506 if (!ctx->cur_folio_in_bio)
507 folio_unlock(ctx->cur_folio);
508 ctx->cur_folio = NULL;
509 }
510 if (!ctx->cur_folio) {
511 ctx->cur_folio = readahead_folio(ctx->rac);
512 ctx->cur_folio_in_bio = false;
513 }
514 ret = iomap_readpage_iter(iter, ctx, done);
515 if (ret <= 0)
516 return ret;
517 }
518
519 return done;
520 }
521
522 /**
523 * iomap_readahead - Attempt to read pages from a file.
524 * @rac: Describes the pages to be read.
525 * @ops: The operations vector for the filesystem.
526 *
527 * This function is for filesystems to call to implement their readahead
528 * address_space operation.
529 *
530 * Context: The @ops callbacks may submit I/O (eg to read the addresses of
531 * blocks from disc), and may wait for it. The caller may be trying to
532 * access a different page, and so sleeping excessively should be avoided.
533 * It may allocate memory, but should avoid costly allocations. This
534 * function is called with memalloc_nofs set, so allocations will not cause
535 * the filesystem to be reentered.
536 */
537 void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops)
538 {
539 struct iomap_iter iter = {
540 .inode = rac->mapping->host,
541 .pos = readahead_pos(rac),
542 .len = readahead_length(rac),
543 };
544 struct iomap_readpage_ctx ctx = {
545 .rac = rac,
546 };
547
548 trace_iomap_readahead(rac->mapping->host, readahead_count(rac));
549
550 while (iomap_iter(&iter, ops) > 0)
551 iter.processed = iomap_readahead_iter(&iter, &ctx);
552
553 if (ctx.bio)
554 submit_bio(ctx.bio);
555 if (ctx.cur_folio) {
556 if (!ctx.cur_folio_in_bio)
557 folio_unlock(ctx.cur_folio);
558 }
559 }
560 EXPORT_SYMBOL_GPL(iomap_readahead);
561
562 /*
563 * iomap_is_partially_uptodate checks whether blocks within a folio are
564 * uptodate or not.
565 *
566 * Returns true if all blocks which correspond to the specified part
567 * of the folio are uptodate.
568 */
569 bool iomap_is_partially_uptodate(struct folio *folio, size_t from, size_t count)
570 {
571 struct iomap_folio_state *ifs = folio->private;
572 struct inode *inode = folio->mapping->host;
573 unsigned first, last, i;
574
575 if (!ifs)
576 return false;
577
578 /* Caller's range may extend past the end of this folio */
579 count = min(folio_size(folio) - from, count);
580
581 /* First and last blocks in range within folio */
582 first = from >> inode->i_blkbits;
583 last = (from + count - 1) >> inode->i_blkbits;
584
585 for (i = first; i <= last; i++)
586 if (!ifs_block_is_uptodate(ifs, i))
587 return false;
588 return true;
589 }
590 EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate);
591
592 /**
593 * iomap_get_folio - get a folio reference for writing
594 * @iter: iteration structure
595 * @pos: start offset of write
596 * @len: Suggested size of folio to create.
597 *
598 * Returns a locked reference to the folio at @pos, or an error pointer if the
599 * folio could not be obtained.
600 */
601 struct folio *iomap_get_folio(struct iomap_iter *iter, loff_t pos, size_t len)
602 {
603 fgf_t fgp = FGP_WRITEBEGIN | FGP_NOFS;
604
605 if (iter->flags & IOMAP_NOWAIT)
606 fgp |= FGP_NOWAIT;
607 fgp |= fgf_set_order(len);
608
609 return __filemap_get_folio(iter->inode->i_mapping, pos >> PAGE_SHIFT,
610 fgp, mapping_gfp_mask(iter->inode->i_mapping));
611 }
612 EXPORT_SYMBOL_GPL(iomap_get_folio);
613
614 bool iomap_release_folio(struct folio *folio, gfp_t gfp_flags)
615 {
616 trace_iomap_release_folio(folio->mapping->host, folio_pos(folio),
617 folio_size(folio));
618
619 /*
620 * If the folio is dirty, we refuse to release our metadata because
621 * it may be partially dirty. Once we track per-block dirty state,
622 * we can release the metadata if every block is dirty.
623 */
624 if (folio_test_dirty(folio))
625 return false;
626 ifs_free(folio);
627 return true;
628 }
629 EXPORT_SYMBOL_GPL(iomap_release_folio);
630
631 void iomap_invalidate_folio(struct folio *folio, size_t offset, size_t len)
632 {
633 trace_iomap_invalidate_folio(folio->mapping->host,
634 folio_pos(folio) + offset, len);
635
636 /*
637 * If we're invalidating the entire folio, clear the dirty state
638 * from it and release it to avoid unnecessary buildup of the LRU.
639 */
640 if (offset == 0 && len == folio_size(folio)) {
641 WARN_ON_ONCE(folio_test_writeback(folio));
642 folio_cancel_dirty(folio);
643 ifs_free(folio);
644 }
645 }
646 EXPORT_SYMBOL_GPL(iomap_invalidate_folio);
647
648 bool iomap_dirty_folio(struct address_space *mapping, struct folio *folio)
649 {
650 struct inode *inode = mapping->host;
651 size_t len = folio_size(folio);
652
653 ifs_alloc(inode, folio, 0);
654 iomap_set_range_dirty(folio, 0, len);
655 return filemap_dirty_folio(mapping, folio);
656 }
657 EXPORT_SYMBOL_GPL(iomap_dirty_folio);
658
659 static void
660 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
661 {
662 loff_t i_size = i_size_read(inode);
663
664 /*
665 * Only truncate newly allocated pages beyoned EOF, even if the
666 * write started inside the existing inode size.
667 */
668 if (pos + len > i_size)
669 truncate_pagecache_range(inode, max(pos, i_size),
670 pos + len - 1);
671 }
672
673 static int iomap_read_folio_sync(loff_t block_start, struct folio *folio,
674 size_t poff, size_t plen, const struct iomap *iomap)
675 {
676 struct bio_vec bvec;
677 struct bio bio;
678
679 bio_init(&bio, iomap->bdev, &bvec, 1, REQ_OP_READ);
680 bio.bi_iter.bi_sector = iomap_sector(iomap, block_start);
681 bio_add_folio_nofail(&bio, folio, plen, poff);
682 return submit_bio_wait(&bio);
683 }
684
685 static int __iomap_write_begin(const struct iomap_iter *iter, loff_t pos,
686 size_t len, struct folio *folio)
687 {
688 const struct iomap *srcmap = iomap_iter_srcmap(iter);
689 struct iomap_folio_state *ifs;
690 loff_t block_size = i_blocksize(iter->inode);
691 loff_t block_start = round_down(pos, block_size);
692 loff_t block_end = round_up(pos + len, block_size);
693 unsigned int nr_blocks = i_blocks_per_folio(iter->inode, folio);
694 size_t from = offset_in_folio(folio, pos), to = from + len;
695 size_t poff, plen;
696
697 /*
698 * If the write or zeroing completely overlaps the current folio, then
699 * entire folio will be dirtied so there is no need for
700 * per-block state tracking structures to be attached to this folio.
701 * For the unshare case, we must read in the ondisk contents because we
702 * are not changing pagecache contents.
703 */
704 if (!(iter->flags & IOMAP_UNSHARE) && pos <= folio_pos(folio) &&
705 pos + len >= folio_pos(folio) + folio_size(folio))
706 return 0;
707
708 ifs = ifs_alloc(iter->inode, folio, iter->flags);
709 if ((iter->flags & IOMAP_NOWAIT) && !ifs && nr_blocks > 1)
710 return -EAGAIN;
711
712 if (folio_test_uptodate(folio))
713 return 0;
714
715 do {
716 iomap_adjust_read_range(iter->inode, folio, &block_start,
717 block_end - block_start, &poff, &plen);
718 if (plen == 0)
719 break;
720
721 if (!(iter->flags & IOMAP_UNSHARE) &&
722 (from <= poff || from >= poff + plen) &&
723 (to <= poff || to >= poff + plen))
724 continue;
725
726 if (iomap_block_needs_zeroing(iter, block_start)) {
727 if (WARN_ON_ONCE(iter->flags & IOMAP_UNSHARE))
728 return -EIO;
729 folio_zero_segments(folio, poff, from, to, poff + plen);
730 } else {
731 int status;
732
733 if (iter->flags & IOMAP_NOWAIT)
734 return -EAGAIN;
735
736 status = iomap_read_folio_sync(block_start, folio,
737 poff, plen, srcmap);
738 if (status)
739 return status;
740 }
741 iomap_set_range_uptodate(folio, poff, plen);
742 } while ((block_start += plen) < block_end);
743
744 return 0;
745 }
746
747 static struct folio *__iomap_get_folio(struct iomap_iter *iter, loff_t pos,
748 size_t len)
749 {
750 const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops;
751
752 if (folio_ops && folio_ops->get_folio)
753 return folio_ops->get_folio(iter, pos, len);
754 else
755 return iomap_get_folio(iter, pos, len);
756 }
757
758 static void __iomap_put_folio(struct iomap_iter *iter, loff_t pos, size_t ret,
759 struct folio *folio)
760 {
761 const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops;
762
763 if (folio_ops && folio_ops->put_folio) {
764 folio_ops->put_folio(iter->inode, pos, ret, folio);
765 } else {
766 folio_unlock(folio);
767 folio_put(folio);
768 }
769 }
770
771 static int iomap_write_begin_inline(const struct iomap_iter *iter,
772 struct folio *folio)
773 {
774 /* needs more work for the tailpacking case; disable for now */
775 if (WARN_ON_ONCE(iomap_iter_srcmap(iter)->offset != 0))
776 return -EIO;
777 return iomap_read_inline_data(iter, folio);
778 }
779
780 static int iomap_write_begin(struct iomap_iter *iter, loff_t pos,
781 size_t len, struct folio **foliop)
782 {
783 const struct iomap_folio_ops *folio_ops = iter->iomap.folio_ops;
784 const struct iomap *srcmap = iomap_iter_srcmap(iter);
785 struct folio *folio;
786 int status = 0;
787
788 BUG_ON(pos + len > iter->iomap.offset + iter->iomap.length);
789 if (srcmap != &iter->iomap)
790 BUG_ON(pos + len > srcmap->offset + srcmap->length);
791
792 if (fatal_signal_pending(current))
793 return -EINTR;
794
795 if (!mapping_large_folio_support(iter->inode->i_mapping))
796 len = min_t(size_t, len, PAGE_SIZE - offset_in_page(pos));
797
798 folio = __iomap_get_folio(iter, pos, len);
799 if (IS_ERR(folio))
800 return PTR_ERR(folio);
801
802 /*
803 * Now we have a locked folio, before we do anything with it we need to
804 * check that the iomap we have cached is not stale. The inode extent
805 * mapping can change due to concurrent IO in flight (e.g.
806 * IOMAP_UNWRITTEN state can change and memory reclaim could have
807 * reclaimed a previously partially written page at this index after IO
808 * completion before this write reaches this file offset) and hence we
809 * could do the wrong thing here (zero a page range incorrectly or fail
810 * to zero) and corrupt data.
811 */
812 if (folio_ops && folio_ops->iomap_valid) {
813 bool iomap_valid = folio_ops->iomap_valid(iter->inode,
814 &iter->iomap);
815 if (!iomap_valid) {
816 iter->iomap.flags |= IOMAP_F_STALE;
817 status = 0;
818 goto out_unlock;
819 }
820 }
821
822 if (pos + len > folio_pos(folio) + folio_size(folio))
823 len = folio_pos(folio) + folio_size(folio) - pos;
824
825 if (srcmap->type == IOMAP_INLINE)
826 status = iomap_write_begin_inline(iter, folio);
827 else if (srcmap->flags & IOMAP_F_BUFFER_HEAD)
828 status = __block_write_begin_int(folio, pos, len, NULL, srcmap);
829 else
830 status = __iomap_write_begin(iter, pos, len, folio);
831
832 if (unlikely(status))
833 goto out_unlock;
834
835 *foliop = folio;
836 return 0;
837
838 out_unlock:
839 __iomap_put_folio(iter, pos, 0, folio);
840
841 return status;
842 }
843
844 static bool __iomap_write_end(struct inode *inode, loff_t pos, size_t len,
845 size_t copied, struct folio *folio)
846 {
847 flush_dcache_folio(folio);
848
849 /*
850 * The blocks that were entirely written will now be uptodate, so we
851 * don't have to worry about a read_folio reading them and overwriting a
852 * partial write. However, if we've encountered a short write and only
853 * partially written into a block, it will not be marked uptodate, so a
854 * read_folio might come in and destroy our partial write.
855 *
856 * Do the simplest thing and just treat any short write to a
857 * non-uptodate page as a zero-length write, and force the caller to
858 * redo the whole thing.
859 */
860 if (unlikely(copied < len && !folio_test_uptodate(folio)))
861 return false;
862 iomap_set_range_uptodate(folio, offset_in_folio(folio, pos), len);
863 iomap_set_range_dirty(folio, offset_in_folio(folio, pos), copied);
864 filemap_dirty_folio(inode->i_mapping, folio);
865 return true;
866 }
867
868 static void iomap_write_end_inline(const struct iomap_iter *iter,
869 struct folio *folio, loff_t pos, size_t copied)
870 {
871 const struct iomap *iomap = &iter->iomap;
872 void *addr;
873
874 WARN_ON_ONCE(!folio_test_uptodate(folio));
875 BUG_ON(!iomap_inline_data_valid(iomap));
876
877 flush_dcache_folio(folio);
878 addr = kmap_local_folio(folio, pos);
879 memcpy(iomap_inline_data(iomap, pos), addr, copied);
880 kunmap_local(addr);
881
882 mark_inode_dirty(iter->inode);
883 }
884
885 /*
886 * Returns true if all copied bytes have been written to the pagecache,
887 * otherwise return false.
888 */
889 static bool iomap_write_end(struct iomap_iter *iter, loff_t pos, size_t len,
890 size_t copied, struct folio *folio)
891 {
892 const struct iomap *srcmap = iomap_iter_srcmap(iter);
893
894 if (srcmap->type == IOMAP_INLINE) {
895 iomap_write_end_inline(iter, folio, pos, copied);
896 return true;
897 }
898
899 if (srcmap->flags & IOMAP_F_BUFFER_HEAD) {
900 size_t bh_written;
901
902 bh_written = block_write_end(NULL, iter->inode->i_mapping, pos,
903 len, copied, &folio->page, NULL);
904 WARN_ON_ONCE(bh_written != copied && bh_written != 0);
905 return bh_written == copied;
906 }
907
908 return __iomap_write_end(iter->inode, pos, len, copied, folio);
909 }
910
911 static loff_t iomap_write_iter(struct iomap_iter *iter, struct iov_iter *i)
912 {
913 loff_t length = iomap_length(iter);
914 loff_t pos = iter->pos;
915 ssize_t total_written = 0;
916 long status = 0;
917 struct address_space *mapping = iter->inode->i_mapping;
918 size_t chunk = mapping_max_folio_size(mapping);
919 unsigned int bdp_flags = (iter->flags & IOMAP_NOWAIT) ? BDP_ASYNC : 0;
920
921 do {
922 struct folio *folio;
923 loff_t old_size;
924 size_t offset; /* Offset into folio */
925 size_t bytes; /* Bytes to write to folio */
926 size_t copied; /* Bytes copied from user */
927 size_t written; /* Bytes have been written */
928
929 bytes = iov_iter_count(i);
930 retry:
931 offset = pos & (chunk - 1);
932 bytes = min(chunk - offset, bytes);
933 status = balance_dirty_pages_ratelimited_flags(mapping,
934 bdp_flags);
935 if (unlikely(status))
936 break;
937
938 if (bytes > length)
939 bytes = length;
940
941 /*
942 * Bring in the user page that we'll copy from _first_.
943 * Otherwise there's a nasty deadlock on copying from the
944 * same page as we're writing to, without it being marked
945 * up-to-date.
946 *
947 * For async buffered writes the assumption is that the user
948 * page has already been faulted in. This can be optimized by
949 * faulting the user page.
950 */
951 if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
952 status = -EFAULT;
953 break;
954 }
955
956 status = iomap_write_begin(iter, pos, bytes, &folio);
957 if (unlikely(status)) {
958 iomap_write_failed(iter->inode, pos, bytes);
959 break;
960 }
961 if (iter->iomap.flags & IOMAP_F_STALE)
962 break;
963
964 offset = offset_in_folio(folio, pos);
965 if (bytes > folio_size(folio) - offset)
966 bytes = folio_size(folio) - offset;
967
968 if (mapping_writably_mapped(mapping))
969 flush_dcache_folio(folio);
970
971 copied = copy_folio_from_iter_atomic(folio, offset, bytes, i);
972 written = iomap_write_end(iter, pos, bytes, copied, folio) ?
973 copied : 0;
974
975 /*
976 * Update the in-memory inode size after copying the data into
977 * the page cache. It's up to the file system to write the
978 * updated size to disk, preferably after I/O completion so that
979 * no stale data is exposed. Only once that's done can we
980 * unlock and release the folio.
981 */
982 old_size = iter->inode->i_size;
983 if (pos + written > old_size) {
984 i_size_write(iter->inode, pos + written);
985 iter->iomap.flags |= IOMAP_F_SIZE_CHANGED;
986 }
987 __iomap_put_folio(iter, pos, written, folio);
988
989 if (old_size < pos)
990 pagecache_isize_extended(iter->inode, old_size, pos);
991
992 cond_resched();
993 if (unlikely(written == 0)) {
994 /*
995 * A short copy made iomap_write_end() reject the
996 * thing entirely. Might be memory poisoning
997 * halfway through, might be a race with munmap,
998 * might be severe memory pressure.
999 */
1000 iomap_write_failed(iter->inode, pos, bytes);
1001 iov_iter_revert(i, copied);
1002
1003 if (chunk > PAGE_SIZE)
1004 chunk /= 2;
1005 if (copied) {
1006 bytes = copied;
1007 goto retry;
1008 }
1009 } else {
1010 pos += written;
1011 total_written += written;
1012 length -= written;
1013 }
1014 } while (iov_iter_count(i) && length);
1015
1016 if (status == -EAGAIN) {
1017 iov_iter_revert(i, total_written);
1018 return -EAGAIN;
1019 }
1020 return total_written ? total_written : status;
1021 }
1022
1023 ssize_t
1024 iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *i,
1025 const struct iomap_ops *ops)
1026 {
1027 struct iomap_iter iter = {
1028 .inode = iocb->ki_filp->f_mapping->host,
1029 .pos = iocb->ki_pos,
1030 .len = iov_iter_count(i),
1031 .flags = IOMAP_WRITE,
1032 };
1033 ssize_t ret;
1034
1035 if (iocb->ki_flags & IOCB_NOWAIT)
1036 iter.flags |= IOMAP_NOWAIT;
1037
1038 while ((ret = iomap_iter(&iter, ops)) > 0)
1039 iter.processed = iomap_write_iter(&iter, i);
1040
1041 if (unlikely(iter.pos == iocb->ki_pos))
1042 return ret;
1043 ret = iter.pos - iocb->ki_pos;
1044 iocb->ki_pos = iter.pos;
1045 return ret;
1046 }
1047 EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
1048
1049 static int iomap_write_delalloc_ifs_punch(struct inode *inode,
1050 struct folio *folio, loff_t start_byte, loff_t end_byte,
1051 iomap_punch_t punch)
1052 {
1053 unsigned int first_blk, last_blk, i;
1054 loff_t last_byte;
1055 u8 blkbits = inode->i_blkbits;
1056 struct iomap_folio_state *ifs;
1057 int ret = 0;
1058
1059 /*
1060 * When we have per-block dirty tracking, there can be
1061 * blocks within a folio which are marked uptodate
1062 * but not dirty. In that case it is necessary to punch
1063 * out such blocks to avoid leaking any delalloc blocks.
1064 */
1065 ifs = folio->private;
1066 if (!ifs)
1067 return ret;
1068
1069 last_byte = min_t(loff_t, end_byte - 1,
1070 folio_pos(folio) + folio_size(folio) - 1);
1071 first_blk = offset_in_folio(folio, start_byte) >> blkbits;
1072 last_blk = offset_in_folio(folio, last_byte) >> blkbits;
1073 for (i = first_blk; i <= last_blk; i++) {
1074 if (!ifs_block_is_dirty(folio, ifs, i)) {
1075 ret = punch(inode, folio_pos(folio) + (i << blkbits),
1076 1 << blkbits);
1077 if (ret)
1078 return ret;
1079 }
1080 }
1081
1082 return ret;
1083 }
1084
1085
1086 static int iomap_write_delalloc_punch(struct inode *inode, struct folio *folio,
1087 loff_t *punch_start_byte, loff_t start_byte, loff_t end_byte,
1088 iomap_punch_t punch)
1089 {
1090 int ret = 0;
1091
1092 if (!folio_test_dirty(folio))
1093 return ret;
1094
1095 /* if dirty, punch up to offset */
1096 if (start_byte > *punch_start_byte) {
1097 ret = punch(inode, *punch_start_byte,
1098 start_byte - *punch_start_byte);
1099 if (ret)
1100 return ret;
1101 }
1102
1103 /* Punch non-dirty blocks within folio */
1104 ret = iomap_write_delalloc_ifs_punch(inode, folio, start_byte,
1105 end_byte, punch);
1106 if (ret)
1107 return ret;
1108
1109 /*
1110 * Make sure the next punch start is correctly bound to
1111 * the end of this data range, not the end of the folio.
1112 */
1113 *punch_start_byte = min_t(loff_t, end_byte,
1114 folio_pos(folio) + folio_size(folio));
1115
1116 return ret;
1117 }
1118
1119 /*
1120 * Scan the data range passed to us for dirty page cache folios. If we find a
1121 * dirty folio, punch out the preceding range and update the offset from which
1122 * the next punch will start from.
1123 *
1124 * We can punch out storage reservations under clean pages because they either
1125 * contain data that has been written back - in which case the delalloc punch
1126 * over that range is a no-op - or they have been read faults in which case they
1127 * contain zeroes and we can remove the delalloc backing range and any new
1128 * writes to those pages will do the normal hole filling operation...
1129 *
1130 * This makes the logic simple: we only need to keep the delalloc extents only
1131 * over the dirty ranges of the page cache.
1132 *
1133 * This function uses [start_byte, end_byte) intervals (i.e. open ended) to
1134 * simplify range iterations.
1135 */
1136 static int iomap_write_delalloc_scan(struct inode *inode,
1137 loff_t *punch_start_byte, loff_t start_byte, loff_t end_byte,
1138 iomap_punch_t punch)
1139 {
1140 while (start_byte < end_byte) {
1141 struct folio *folio;
1142 int ret;
1143
1144 /* grab locked page */
1145 folio = filemap_lock_folio(inode->i_mapping,
1146 start_byte >> PAGE_SHIFT);
1147 if (IS_ERR(folio)) {
1148 start_byte = ALIGN_DOWN(start_byte, PAGE_SIZE) +
1149 PAGE_SIZE;
1150 continue;
1151 }
1152
1153 ret = iomap_write_delalloc_punch(inode, folio, punch_start_byte,
1154 start_byte, end_byte, punch);
1155 if (ret) {
1156 folio_unlock(folio);
1157 folio_put(folio);
1158 return ret;
1159 }
1160
1161 /* move offset to start of next folio in range */
1162 start_byte = folio_next_index(folio) << PAGE_SHIFT;
1163 folio_unlock(folio);
1164 folio_put(folio);
1165 }
1166 return 0;
1167 }
1168
1169 /*
1170 * Punch out all the delalloc blocks in the range given except for those that
1171 * have dirty data still pending in the page cache - those are going to be
1172 * written and so must still retain the delalloc backing for writeback.
1173 *
1174 * As we are scanning the page cache for data, we don't need to reimplement the
1175 * wheel - mapping_seek_hole_data() does exactly what we need to identify the
1176 * start and end of data ranges correctly even for sub-folio block sizes. This
1177 * byte range based iteration is especially convenient because it means we
1178 * don't have to care about variable size folios, nor where the start or end of
1179 * the data range lies within a folio, if they lie within the same folio or even
1180 * if there are multiple discontiguous data ranges within the folio.
1181 *
1182 * It should be noted that mapping_seek_hole_data() is not aware of EOF, and so
1183 * can return data ranges that exist in the cache beyond EOF. e.g. a page fault
1184 * spanning EOF will initialise the post-EOF data to zeroes and mark it up to
1185 * date. A write page fault can then mark it dirty. If we then fail a write()
1186 * beyond EOF into that up to date cached range, we allocate a delalloc block
1187 * beyond EOF and then have to punch it out. Because the range is up to date,
1188 * mapping_seek_hole_data() will return it, and we will skip the punch because
1189 * the folio is dirty. THis is incorrect - we always need to punch out delalloc
1190 * beyond EOF in this case as writeback will never write back and covert that
1191 * delalloc block beyond EOF. Hence we limit the cached data scan range to EOF,
1192 * resulting in always punching out the range from the EOF to the end of the
1193 * range the iomap spans.
1194 *
1195 * Intervals are of the form [start_byte, end_byte) (i.e. open ended) because it
1196 * matches the intervals returned by mapping_seek_hole_data(). i.e. SEEK_DATA
1197 * returns the start of a data range (start_byte), and SEEK_HOLE(start_byte)
1198 * returns the end of the data range (data_end). Using closed intervals would
1199 * require sprinkling this code with magic "+ 1" and "- 1" arithmetic and expose
1200 * the code to subtle off-by-one bugs....
1201 */
1202 static int iomap_write_delalloc_release(struct inode *inode,
1203 loff_t start_byte, loff_t end_byte, iomap_punch_t punch)
1204 {
1205 loff_t punch_start_byte = start_byte;
1206 loff_t scan_end_byte = min(i_size_read(inode), end_byte);
1207 int error = 0;
1208
1209 /*
1210 * Lock the mapping to avoid races with page faults re-instantiating
1211 * folios and dirtying them via ->page_mkwrite whilst we walk the
1212 * cache and perform delalloc extent removal. Failing to do this can
1213 * leave dirty pages with no space reservation in the cache.
1214 */
1215 filemap_invalidate_lock(inode->i_mapping);
1216 while (start_byte < scan_end_byte) {
1217 loff_t data_end;
1218
1219 start_byte = mapping_seek_hole_data(inode->i_mapping,
1220 start_byte, scan_end_byte, SEEK_DATA);
1221 /*
1222 * If there is no more data to scan, all that is left is to
1223 * punch out the remaining range.
1224 */
1225 if (start_byte == -ENXIO || start_byte == scan_end_byte)
1226 break;
1227 if (start_byte < 0) {
1228 error = start_byte;
1229 goto out_unlock;
1230 }
1231 WARN_ON_ONCE(start_byte < punch_start_byte);
1232 WARN_ON_ONCE(start_byte > scan_end_byte);
1233
1234 /*
1235 * We find the end of this contiguous cached data range by
1236 * seeking from start_byte to the beginning of the next hole.
1237 */
1238 data_end = mapping_seek_hole_data(inode->i_mapping, start_byte,
1239 scan_end_byte, SEEK_HOLE);
1240 if (data_end < 0) {
1241 error = data_end;
1242 goto out_unlock;
1243 }
1244
1245 /*
1246 * If we race with post-direct I/O invalidation of the page cache,
1247 * there might be no data left at start_byte.
1248 */
1249 if (data_end == start_byte)
1250 continue;
1251
1252 WARN_ON_ONCE(data_end < start_byte);
1253 WARN_ON_ONCE(data_end > scan_end_byte);
1254
1255 error = iomap_write_delalloc_scan(inode, &punch_start_byte,
1256 start_byte, data_end, punch);
1257 if (error)
1258 goto out_unlock;
1259
1260 /* The next data search starts at the end of this one. */
1261 start_byte = data_end;
1262 }
1263
1264 if (punch_start_byte < end_byte)
1265 error = punch(inode, punch_start_byte,
1266 end_byte - punch_start_byte);
1267 out_unlock:
1268 filemap_invalidate_unlock(inode->i_mapping);
1269 return error;
1270 }
1271
1272 /*
1273 * When a short write occurs, the filesystem may need to remove reserved space
1274 * that was allocated in ->iomap_begin from it's ->iomap_end method. For
1275 * filesystems that use delayed allocation, we need to punch out delalloc
1276 * extents from the range that are not dirty in the page cache. As the write can
1277 * race with page faults, there can be dirty pages over the delalloc extent
1278 * outside the range of a short write but still within the delalloc extent
1279 * allocated for this iomap.
1280 *
1281 * This function uses [start_byte, end_byte) intervals (i.e. open ended) to
1282 * simplify range iterations.
1283 *
1284 * The punch() callback *must* only punch delalloc extents in the range passed
1285 * to it. It must skip over all other types of extents in the range and leave
1286 * them completely unchanged. It must do this punch atomically with respect to
1287 * other extent modifications.
1288 *
1289 * The punch() callback may be called with a folio locked to prevent writeback
1290 * extent allocation racing at the edge of the range we are currently punching.
1291 * The locked folio may or may not cover the range being punched, so it is not
1292 * safe for the punch() callback to lock folios itself.
1293 *
1294 * Lock order is:
1295 *
1296 * inode->i_rwsem (shared or exclusive)
1297 * inode->i_mapping->invalidate_lock (exclusive)
1298 * folio_lock()
1299 * ->punch
1300 * internal filesystem allocation lock
1301 */
1302 int iomap_file_buffered_write_punch_delalloc(struct inode *inode,
1303 struct iomap *iomap, loff_t pos, loff_t length,
1304 ssize_t written, iomap_punch_t punch)
1305 {
1306 loff_t start_byte;
1307 loff_t end_byte;
1308 unsigned int blocksize = i_blocksize(inode);
1309
1310 if (iomap->type != IOMAP_DELALLOC)
1311 return 0;
1312
1313 /* If we didn't reserve the blocks, we're not allowed to punch them. */
1314 if (!(iomap->flags & IOMAP_F_NEW))
1315 return 0;
1316
1317 /*
1318 * start_byte refers to the first unused block after a short write. If
1319 * nothing was written, round offset down to point at the first block in
1320 * the range.
1321 */
1322 if (unlikely(!written))
1323 start_byte = round_down(pos, blocksize);
1324 else
1325 start_byte = round_up(pos + written, blocksize);
1326 end_byte = round_up(pos + length, blocksize);
1327
1328 /* Nothing to do if we've written the entire delalloc extent */
1329 if (start_byte >= end_byte)
1330 return 0;
1331
1332 return iomap_write_delalloc_release(inode, start_byte, end_byte,
1333 punch);
1334 }
1335 EXPORT_SYMBOL_GPL(iomap_file_buffered_write_punch_delalloc);
1336
1337 static loff_t iomap_unshare_iter(struct iomap_iter *iter)
1338 {
1339 struct iomap *iomap = &iter->iomap;
1340 const struct iomap *srcmap = iomap_iter_srcmap(iter);
1341 loff_t pos = iter->pos;
1342 loff_t length = iomap_length(iter);
1343 loff_t written = 0;
1344
1345 /* don't bother with blocks that are not shared to start with */
1346 if (!(iomap->flags & IOMAP_F_SHARED))
1347 return length;
1348 /* don't bother with holes or unwritten extents */
1349 if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
1350 return length;
1351
1352 do {
1353 struct folio *folio;
1354 int status;
1355 size_t offset;
1356 size_t bytes = min_t(u64, SIZE_MAX, length);
1357 bool ret;
1358
1359 status = iomap_write_begin(iter, pos, bytes, &folio);
1360 if (unlikely(status))
1361 return status;
1362 if (iomap->flags & IOMAP_F_STALE)
1363 break;
1364
1365 offset = offset_in_folio(folio, pos);
1366 if (bytes > folio_size(folio) - offset)
1367 bytes = folio_size(folio) - offset;
1368
1369 ret = iomap_write_end(iter, pos, bytes, bytes, folio);
1370 __iomap_put_folio(iter, pos, bytes, folio);
1371 if (WARN_ON_ONCE(!ret))
1372 return -EIO;
1373
1374 cond_resched();
1375
1376 pos += bytes;
1377 written += bytes;
1378 length -= bytes;
1379
1380 balance_dirty_pages_ratelimited(iter->inode->i_mapping);
1381 } while (length > 0);
1382
1383 return written;
1384 }
1385
1386 int
1387 iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len,
1388 const struct iomap_ops *ops)
1389 {
1390 struct iomap_iter iter = {
1391 .inode = inode,
1392 .pos = pos,
1393 .flags = IOMAP_WRITE | IOMAP_UNSHARE,
1394 };
1395 loff_t size = i_size_read(inode);
1396 int ret;
1397
1398 if (pos < 0 || pos >= size)
1399 return 0;
1400
1401 iter.len = min(len, size - pos);
1402 while ((ret = iomap_iter(&iter, ops)) > 0)
1403 iter.processed = iomap_unshare_iter(&iter);
1404 return ret;
1405 }
1406 EXPORT_SYMBOL_GPL(iomap_file_unshare);
1407
1408 static loff_t iomap_zero_iter(struct iomap_iter *iter, bool *did_zero)
1409 {
1410 const struct iomap *srcmap = iomap_iter_srcmap(iter);
1411 loff_t pos = iter->pos;
1412 loff_t length = iomap_length(iter);
1413 loff_t written = 0;
1414
1415 /* already zeroed? we're done. */
1416 if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
1417 return length;
1418
1419 do {
1420 struct folio *folio;
1421 int status;
1422 size_t offset;
1423 size_t bytes = min_t(u64, SIZE_MAX, length);
1424 bool ret;
1425
1426 status = iomap_write_begin(iter, pos, bytes, &folio);
1427 if (status)
1428 return status;
1429 if (iter->iomap.flags & IOMAP_F_STALE)
1430 break;
1431
1432 offset = offset_in_folio(folio, pos);
1433 if (bytes > folio_size(folio) - offset)
1434 bytes = folio_size(folio) - offset;
1435
1436 folio_zero_range(folio, offset, bytes);
1437 folio_mark_accessed(folio);
1438
1439 ret = iomap_write_end(iter, pos, bytes, bytes, folio);
1440 __iomap_put_folio(iter, pos, bytes, folio);
1441 if (WARN_ON_ONCE(!ret))
1442 return -EIO;
1443
1444 pos += bytes;
1445 length -= bytes;
1446 written += bytes;
1447 } while (length > 0);
1448
1449 if (did_zero)
1450 *did_zero = true;
1451 return written;
1452 }
1453
1454 int
1455 iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
1456 const struct iomap_ops *ops)
1457 {
1458 struct iomap_iter iter = {
1459 .inode = inode,
1460 .pos = pos,
1461 .len = len,
1462 .flags = IOMAP_ZERO,
1463 };
1464 int ret;
1465
1466 while ((ret = iomap_iter(&iter, ops)) > 0)
1467 iter.processed = iomap_zero_iter(&iter, did_zero);
1468 return ret;
1469 }
1470 EXPORT_SYMBOL_GPL(iomap_zero_range);
1471
1472 int
1473 iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
1474 const struct iomap_ops *ops)
1475 {
1476 unsigned int blocksize = i_blocksize(inode);
1477 unsigned int off = pos & (blocksize - 1);
1478
1479 /* Block boundary? Nothing to do */
1480 if (!off)
1481 return 0;
1482 return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
1483 }
1484 EXPORT_SYMBOL_GPL(iomap_truncate_page);
1485
1486 static loff_t iomap_folio_mkwrite_iter(struct iomap_iter *iter,
1487 struct folio *folio)
1488 {
1489 loff_t length = iomap_length(iter);
1490 int ret;
1491
1492 if (iter->iomap.flags & IOMAP_F_BUFFER_HEAD) {
1493 ret = __block_write_begin_int(folio, iter->pos, length, NULL,
1494 &iter->iomap);
1495 if (ret)
1496 return ret;
1497 block_commit_write(&folio->page, 0, length);
1498 } else {
1499 WARN_ON_ONCE(!folio_test_uptodate(folio));
1500 folio_mark_dirty(folio);
1501 }
1502
1503 return length;
1504 }
1505
1506 vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops)
1507 {
1508 struct iomap_iter iter = {
1509 .inode = file_inode(vmf->vma->vm_file),
1510 .flags = IOMAP_WRITE | IOMAP_FAULT,
1511 };
1512 struct folio *folio = page_folio(vmf->page);
1513 ssize_t ret;
1514
1515 folio_lock(folio);
1516 ret = folio_mkwrite_check_truncate(folio, iter.inode);
1517 if (ret < 0)
1518 goto out_unlock;
1519 iter.pos = folio_pos(folio);
1520 iter.len = ret;
1521 while ((ret = iomap_iter(&iter, ops)) > 0)
1522 iter.processed = iomap_folio_mkwrite_iter(&iter, folio);
1523
1524 if (ret < 0)
1525 goto out_unlock;
1526 folio_wait_stable(folio);
1527 return VM_FAULT_LOCKED;
1528 out_unlock:
1529 folio_unlock(folio);
1530 return vmf_fs_error(ret);
1531 }
1532 EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
1533
1534 static void iomap_finish_folio_write(struct inode *inode, struct folio *folio,
1535 size_t len)
1536 {
1537 struct iomap_folio_state *ifs = folio->private;
1538
1539 WARN_ON_ONCE(i_blocks_per_folio(inode, folio) > 1 && !ifs);
1540 WARN_ON_ONCE(ifs && atomic_read(&ifs->write_bytes_pending) <= 0);
1541
1542 if (!ifs || atomic_sub_and_test(len, &ifs->write_bytes_pending))
1543 folio_end_writeback(folio);
1544 }
1545
1546 /*
1547 * We're now finished for good with this ioend structure. Update the page
1548 * state, release holds on bios, and finally free up memory. Do not use the
1549 * ioend after this.
1550 */
1551 static u32
1552 iomap_finish_ioend(struct iomap_ioend *ioend, int error)
1553 {
1554 struct inode *inode = ioend->io_inode;
1555 struct bio *bio = &ioend->io_bio;
1556 struct folio_iter fi;
1557 u32 folio_count = 0;
1558
1559 if (error) {
1560 mapping_set_error(inode->i_mapping, error);
1561 if (!bio_flagged(bio, BIO_QUIET)) {
1562 pr_err_ratelimited(
1563 "%s: writeback error on inode %lu, offset %lld, sector %llu",
1564 inode->i_sb->s_id, inode->i_ino,
1565 ioend->io_offset, ioend->io_sector);
1566 }
1567 }
1568
1569 /* walk all folios in bio, ending page IO on them */
1570 bio_for_each_folio_all(fi, bio) {
1571 iomap_finish_folio_write(inode, fi.folio, fi.length);
1572 folio_count++;
1573 }
1574
1575 bio_put(bio); /* frees the ioend */
1576 return folio_count;
1577 }
1578
1579 /*
1580 * Ioend completion routine for merged bios. This can only be called from task
1581 * contexts as merged ioends can be of unbound length. Hence we have to break up
1582 * the writeback completions into manageable chunks to avoid long scheduler
1583 * holdoffs. We aim to keep scheduler holdoffs down below 10ms so that we get
1584 * good batch processing throughput without creating adverse scheduler latency
1585 * conditions.
1586 */
1587 void
1588 iomap_finish_ioends(struct iomap_ioend *ioend, int error)
1589 {
1590 struct list_head tmp;
1591 u32 completions;
1592
1593 might_sleep();
1594
1595 list_replace_init(&ioend->io_list, &tmp);
1596 completions = iomap_finish_ioend(ioend, error);
1597
1598 while (!list_empty(&tmp)) {
1599 if (completions > IOEND_BATCH_SIZE * 8) {
1600 cond_resched();
1601 completions = 0;
1602 }
1603 ioend = list_first_entry(&tmp, struct iomap_ioend, io_list);
1604 list_del_init(&ioend->io_list);
1605 completions += iomap_finish_ioend(ioend, error);
1606 }
1607 }
1608 EXPORT_SYMBOL_GPL(iomap_finish_ioends);
1609
1610 /*
1611 * We can merge two adjacent ioends if they have the same set of work to do.
1612 */
1613 static bool
1614 iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next)
1615 {
1616 if (ioend->io_bio.bi_status != next->io_bio.bi_status)
1617 return false;
1618 if ((ioend->io_flags & IOMAP_F_SHARED) ^
1619 (next->io_flags & IOMAP_F_SHARED))
1620 return false;
1621 if ((ioend->io_type == IOMAP_UNWRITTEN) ^
1622 (next->io_type == IOMAP_UNWRITTEN))
1623 return false;
1624 if (ioend->io_offset + ioend->io_size != next->io_offset)
1625 return false;
1626 /*
1627 * Do not merge physically discontiguous ioends. The filesystem
1628 * completion functions will have to iterate the physical
1629 * discontiguities even if we merge the ioends at a logical level, so
1630 * we don't gain anything by merging physical discontiguities here.
1631 *
1632 * We cannot use bio->bi_iter.bi_sector here as it is modified during
1633 * submission so does not point to the start sector of the bio at
1634 * completion.
1635 */
1636 if (ioend->io_sector + (ioend->io_size >> 9) != next->io_sector)
1637 return false;
1638 return true;
1639 }
1640
1641 void
1642 iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends)
1643 {
1644 struct iomap_ioend *next;
1645
1646 INIT_LIST_HEAD(&ioend->io_list);
1647
1648 while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend,
1649 io_list))) {
1650 if (!iomap_ioend_can_merge(ioend, next))
1651 break;
1652 list_move_tail(&next->io_list, &ioend->io_list);
1653 ioend->io_size += next->io_size;
1654 }
1655 }
1656 EXPORT_SYMBOL_GPL(iomap_ioend_try_merge);
1657
1658 static int
1659 iomap_ioend_compare(void *priv, const struct list_head *a,
1660 const struct list_head *b)
1661 {
1662 struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list);
1663 struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list);
1664
1665 if (ia->io_offset < ib->io_offset)
1666 return -1;
1667 if (ia->io_offset > ib->io_offset)
1668 return 1;
1669 return 0;
1670 }
1671
1672 void
1673 iomap_sort_ioends(struct list_head *ioend_list)
1674 {
1675 list_sort(NULL, ioend_list, iomap_ioend_compare);
1676 }
1677 EXPORT_SYMBOL_GPL(iomap_sort_ioends);
1678
1679 static void iomap_writepage_end_bio(struct bio *bio)
1680 {
1681 iomap_finish_ioend(iomap_ioend_from_bio(bio),
1682 blk_status_to_errno(bio->bi_status));
1683 }
1684
1685 /*
1686 * Submit the final bio for an ioend.
1687 *
1688 * If @error is non-zero, it means that we have a situation where some part of
1689 * the submission process has failed after we've marked pages for writeback.
1690 * We cannot cancel ioend directly in that case, so call the bio end I/O handler
1691 * with the error status here to run the normal I/O completion handler to clear
1692 * the writeback bit and let the file system proess the errors.
1693 */
1694 static int iomap_submit_ioend(struct iomap_writepage_ctx *wpc, int error)
1695 {
1696 if (!wpc->ioend)
1697 return error;
1698
1699 /*
1700 * Let the file systems prepare the I/O submission and hook in an I/O
1701 * comletion handler. This also needs to happen in case after a
1702 * failure happened so that the file system end I/O handler gets called
1703 * to clean up.
1704 */
1705 if (wpc->ops->prepare_ioend)
1706 error = wpc->ops->prepare_ioend(wpc->ioend, error);
1707
1708 if (error) {
1709 wpc->ioend->io_bio.bi_status = errno_to_blk_status(error);
1710 bio_endio(&wpc->ioend->io_bio);
1711 } else {
1712 submit_bio(&wpc->ioend->io_bio);
1713 }
1714
1715 wpc->ioend = NULL;
1716 return error;
1717 }
1718
1719 static struct iomap_ioend *iomap_alloc_ioend(struct iomap_writepage_ctx *wpc,
1720 struct writeback_control *wbc, struct inode *inode, loff_t pos)
1721 {
1722 struct iomap_ioend *ioend;
1723 struct bio *bio;
1724
1725 bio = bio_alloc_bioset(wpc->iomap.bdev, BIO_MAX_VECS,
1726 REQ_OP_WRITE | wbc_to_write_flags(wbc),
1727 GFP_NOFS, &iomap_ioend_bioset);
1728 bio->bi_iter.bi_sector = iomap_sector(&wpc->iomap, pos);
1729 bio->bi_end_io = iomap_writepage_end_bio;
1730 wbc_init_bio(wbc, bio);
1731 bio->bi_write_hint = inode->i_write_hint;
1732
1733 ioend = iomap_ioend_from_bio(bio);
1734 INIT_LIST_HEAD(&ioend->io_list);
1735 ioend->io_type = wpc->iomap.type;
1736 ioend->io_flags = wpc->iomap.flags;
1737 ioend->io_inode = inode;
1738 ioend->io_size = 0;
1739 ioend->io_offset = pos;
1740 ioend->io_sector = bio->bi_iter.bi_sector;
1741
1742 wpc->nr_folios = 0;
1743 return ioend;
1744 }
1745
1746 static bool iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t pos)
1747 {
1748 if ((wpc->iomap.flags & IOMAP_F_SHARED) !=
1749 (wpc->ioend->io_flags & IOMAP_F_SHARED))
1750 return false;
1751 if (wpc->iomap.type != wpc->ioend->io_type)
1752 return false;
1753 if (pos != wpc->ioend->io_offset + wpc->ioend->io_size)
1754 return false;
1755 if (iomap_sector(&wpc->iomap, pos) !=
1756 bio_end_sector(&wpc->ioend->io_bio))
1757 return false;
1758 /*
1759 * Limit ioend bio chain lengths to minimise IO completion latency. This
1760 * also prevents long tight loops ending page writeback on all the
1761 * folios in the ioend.
1762 */
1763 if (wpc->nr_folios >= IOEND_BATCH_SIZE)
1764 return false;
1765 return true;
1766 }
1767
1768 /*
1769 * Test to see if we have an existing ioend structure that we could append to
1770 * first; otherwise finish off the current ioend and start another.
1771 *
1772 * If a new ioend is created and cached, the old ioend is submitted to the block
1773 * layer instantly. Batching optimisations are provided by higher level block
1774 * plugging.
1775 *
1776 * At the end of a writeback pass, there will be a cached ioend remaining on the
1777 * writepage context that the caller will need to submit.
1778 */
1779 static int iomap_add_to_ioend(struct iomap_writepage_ctx *wpc,
1780 struct writeback_control *wbc, struct folio *folio,
1781 struct inode *inode, loff_t pos, unsigned len)
1782 {
1783 struct iomap_folio_state *ifs = folio->private;
1784 size_t poff = offset_in_folio(folio, pos);
1785 int error;
1786
1787 if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, pos)) {
1788 new_ioend:
1789 error = iomap_submit_ioend(wpc, 0);
1790 if (error)
1791 return error;
1792 wpc->ioend = iomap_alloc_ioend(wpc, wbc, inode, pos);
1793 }
1794
1795 if (!bio_add_folio(&wpc->ioend->io_bio, folio, len, poff))
1796 goto new_ioend;
1797
1798 if (ifs)
1799 atomic_add(len, &ifs->write_bytes_pending);
1800 wpc->ioend->io_size += len;
1801 wbc_account_cgroup_owner(wbc, &folio->page, len);
1802 return 0;
1803 }
1804
1805 static int iomap_writepage_map_blocks(struct iomap_writepage_ctx *wpc,
1806 struct writeback_control *wbc, struct folio *folio,
1807 struct inode *inode, u64 pos, unsigned dirty_len,
1808 unsigned *count)
1809 {
1810 int error;
1811
1812 do {
1813 unsigned map_len;
1814
1815 error = wpc->ops->map_blocks(wpc, inode, pos, dirty_len);
1816 if (error)
1817 break;
1818 trace_iomap_writepage_map(inode, pos, dirty_len, &wpc->iomap);
1819
1820 map_len = min_t(u64, dirty_len,
1821 wpc->iomap.offset + wpc->iomap.length - pos);
1822 WARN_ON_ONCE(!folio->private && map_len < dirty_len);
1823
1824 switch (wpc->iomap.type) {
1825 case IOMAP_INLINE:
1826 WARN_ON_ONCE(1);
1827 error = -EIO;
1828 break;
1829 case IOMAP_HOLE:
1830 break;
1831 default:
1832 error = iomap_add_to_ioend(wpc, wbc, folio, inode, pos,
1833 map_len);
1834 if (!error)
1835 (*count)++;
1836 break;
1837 }
1838 dirty_len -= map_len;
1839 pos += map_len;
1840 } while (dirty_len && !error);
1841
1842 /*
1843 * We cannot cancel the ioend directly here on error. We may have
1844 * already set other pages under writeback and hence we have to run I/O
1845 * completion to mark the error state of the pages under writeback
1846 * appropriately.
1847 *
1848 * Just let the file system know what portion of the folio failed to
1849 * map.
1850 */
1851 if (error && wpc->ops->discard_folio)
1852 wpc->ops->discard_folio(folio, pos);
1853 return error;
1854 }
1855
1856 /*
1857 * Check interaction of the folio with the file end.
1858 *
1859 * If the folio is entirely beyond i_size, return false. If it straddles
1860 * i_size, adjust end_pos and zero all data beyond i_size.
1861 */
1862 static bool iomap_writepage_handle_eof(struct folio *folio, struct inode *inode,
1863 u64 *end_pos)
1864 {
1865 u64 isize = i_size_read(inode);
1866
1867 if (*end_pos > isize) {
1868 size_t poff = offset_in_folio(folio, isize);
1869 pgoff_t end_index = isize >> PAGE_SHIFT;
1870
1871 /*
1872 * If the folio is entirely ouside of i_size, skip it.
1873 *
1874 * This can happen due to a truncate operation that is in
1875 * progress and in that case truncate will finish it off once
1876 * we've dropped the folio lock.
1877 *
1878 * Note that the pgoff_t used for end_index is an unsigned long.
1879 * If the given offset is greater than 16TB on a 32-bit system,
1880 * then if we checked if the folio is fully outside i_size with
1881 * "if (folio->index >= end_index + 1)", "end_index + 1" would
1882 * overflow and evaluate to 0. Hence this folio would be
1883 * redirtied and written out repeatedly, which would result in
1884 * an infinite loop; the user program performing this operation
1885 * would hang. Instead, we can detect this situation by
1886 * checking if the folio is totally beyond i_size or if its
1887 * offset is just equal to the EOF.
1888 */
1889 if (folio->index > end_index ||
1890 (folio->index == end_index && poff == 0))
1891 return false;
1892
1893 /*
1894 * The folio straddles i_size.
1895 *
1896 * It must be zeroed out on each and every writepage invocation
1897 * because it may be mmapped:
1898 *
1899 * A file is mapped in multiples of the page size. For a
1900 * file that is not a multiple of the page size, the
1901 * remaining memory is zeroed when mapped, and writes to that
1902 * region are not written out to the file.
1903 *
1904 * Also adjust the writeback range to skip all blocks entirely
1905 * beyond i_size.
1906 */
1907 folio_zero_segment(folio, poff, folio_size(folio));
1908 *end_pos = round_up(isize, i_blocksize(inode));
1909 }
1910
1911 return true;
1912 }
1913
1914 static int iomap_writepage_map(struct iomap_writepage_ctx *wpc,
1915 struct writeback_control *wbc, struct folio *folio)
1916 {
1917 struct iomap_folio_state *ifs = folio->private;
1918 struct inode *inode = folio->mapping->host;
1919 u64 pos = folio_pos(folio);
1920 u64 end_pos = pos + folio_size(folio);
1921 unsigned count = 0;
1922 int error = 0;
1923 u32 rlen;
1924
1925 WARN_ON_ONCE(!folio_test_locked(folio));
1926 WARN_ON_ONCE(folio_test_dirty(folio));
1927 WARN_ON_ONCE(folio_test_writeback(folio));
1928
1929 trace_iomap_writepage(inode, pos, folio_size(folio));
1930
1931 if (!iomap_writepage_handle_eof(folio, inode, &end_pos)) {
1932 folio_unlock(folio);
1933 return 0;
1934 }
1935 WARN_ON_ONCE(end_pos <= pos);
1936
1937 if (i_blocks_per_folio(inode, folio) > 1) {
1938 if (!ifs) {
1939 ifs = ifs_alloc(inode, folio, 0);
1940 iomap_set_range_dirty(folio, 0, end_pos - pos);
1941 }
1942
1943 /*
1944 * Keep the I/O completion handler from clearing the writeback
1945 * bit until we have submitted all blocks by adding a bias to
1946 * ifs->write_bytes_pending, which is dropped after submitting
1947 * all blocks.
1948 */
1949 WARN_ON_ONCE(atomic_read(&ifs->write_bytes_pending) != 0);
1950 atomic_inc(&ifs->write_bytes_pending);
1951 }
1952
1953 /*
1954 * Set the writeback bit ASAP, as the I/O completion for the single
1955 * block per folio case happen hit as soon as we're submitting the bio.
1956 */
1957 folio_start_writeback(folio);
1958
1959 /*
1960 * Walk through the folio to find dirty areas to write back.
1961 */
1962 while ((rlen = iomap_find_dirty_range(folio, &pos, end_pos))) {
1963 error = iomap_writepage_map_blocks(wpc, wbc, folio, inode,
1964 pos, rlen, &count);
1965 if (error)
1966 break;
1967 pos += rlen;
1968 }
1969
1970 if (count)
1971 wpc->nr_folios++;
1972
1973 /*
1974 * We can have dirty bits set past end of file in page_mkwrite path
1975 * while mapping the last partial folio. Hence it's better to clear
1976 * all the dirty bits in the folio here.
1977 */
1978 iomap_clear_range_dirty(folio, 0, folio_size(folio));
1979
1980 /*
1981 * Usually the writeback bit is cleared by the I/O completion handler.
1982 * But we may end up either not actually writing any blocks, or (when
1983 * there are multiple blocks in a folio) all I/O might have finished
1984 * already at this point. In that case we need to clear the writeback
1985 * bit ourselves right after unlocking the page.
1986 */
1987 folio_unlock(folio);
1988 if (ifs) {
1989 if (atomic_dec_and_test(&ifs->write_bytes_pending))
1990 folio_end_writeback(folio);
1991 } else {
1992 if (!count)
1993 folio_end_writeback(folio);
1994 }
1995 mapping_set_error(inode->i_mapping, error);
1996 return error;
1997 }
1998
1999 int
2000 iomap_writepages(struct address_space *mapping, struct writeback_control *wbc,
2001 struct iomap_writepage_ctx *wpc,
2002 const struct iomap_writeback_ops *ops)
2003 {
2004 struct folio *folio = NULL;
2005 int error;
2006
2007 /*
2008 * Writeback from reclaim context should never happen except in the case
2009 * of a VM regression so warn about it and refuse to write the data.
2010 */
2011 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC | PF_KSWAPD)) ==
2012 PF_MEMALLOC))
2013 return -EIO;
2014
2015 wpc->ops = ops;
2016 while ((folio = writeback_iter(mapping, wbc, folio, &error)))
2017 error = iomap_writepage_map(wpc, wbc, folio);
2018 return iomap_submit_ioend(wpc, error);
2019 }
2020 EXPORT_SYMBOL_GPL(iomap_writepages);
2021
2022 static int __init iomap_init(void)
2023 {
2024 return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
2025 offsetof(struct iomap_ioend, io_bio),
2026 BIOSET_NEED_BVECS);
2027 }
2028 fs_initcall(iomap_init);
2029
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.