Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2010 Red Hat, Inc.
4 : * Copyright (C) 2016-2019 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 : /*
25 : * Structure allocated for each page or THP when block size < page size
26 : * to track sub-page uptodate status and I/O completions.
27 : */
28 : struct iomap_page {
29 : atomic_t read_bytes_pending;
30 : atomic_t write_bytes_pending;
31 : spinlock_t uptodate_lock;
32 : unsigned long uptodate[];
33 : };
34 :
35 0 : static inline struct iomap_page *to_iomap_page(struct page *page)
36 : {
37 : /*
38 : * per-block data is stored in the head page. Callers should
39 : * not be dealing with tail pages (and if they are, they can
40 : * call thp_head() first.
41 : */
42 0 : VM_BUG_ON_PGFLAGS(PageTail(page), page);
43 :
44 0 : if (page_has_private(page))
45 0 : return (struct iomap_page *)page_private(page);
46 : return NULL;
47 : }
48 :
49 : static struct bio_set iomap_ioend_bioset;
50 :
51 : static struct iomap_page *
52 0 : iomap_page_create(struct inode *inode, struct page *page)
53 : {
54 0 : struct iomap_page *iop = to_iomap_page(page);
55 0 : unsigned int nr_blocks = i_blocks_per_page(inode, page);
56 :
57 0 : if (iop || nr_blocks <= 1)
58 : return iop;
59 :
60 0 : iop = kzalloc(struct_size(iop, uptodate, BITS_TO_LONGS(nr_blocks)),
61 : GFP_NOFS | __GFP_NOFAIL);
62 0 : spin_lock_init(&iop->uptodate_lock);
63 0 : if (PageUptodate(page))
64 0 : bitmap_fill(iop->uptodate, nr_blocks);
65 0 : attach_page_private(page, iop);
66 0 : return iop;
67 : }
68 :
69 : static void
70 0 : iomap_page_release(struct page *page)
71 : {
72 0 : struct iomap_page *iop = detach_page_private(page);
73 0 : unsigned int nr_blocks = i_blocks_per_page(page->mapping->host, page);
74 :
75 0 : if (!iop)
76 : return;
77 0 : WARN_ON_ONCE(atomic_read(&iop->read_bytes_pending));
78 0 : WARN_ON_ONCE(atomic_read(&iop->write_bytes_pending));
79 0 : WARN_ON_ONCE(bitmap_full(iop->uptodate, nr_blocks) !=
80 : PageUptodate(page));
81 0 : kfree(iop);
82 : }
83 :
84 : /*
85 : * Calculate the range inside the page that we actually need to read.
86 : */
87 : static void
88 0 : iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop,
89 : loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp)
90 : {
91 0 : loff_t orig_pos = *pos;
92 0 : loff_t isize = i_size_read(inode);
93 0 : unsigned block_bits = inode->i_blkbits;
94 0 : unsigned block_size = (1 << block_bits);
95 0 : unsigned poff = offset_in_page(*pos);
96 0 : unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
97 0 : unsigned first = poff >> block_bits;
98 0 : unsigned last = (poff + plen - 1) >> block_bits;
99 :
100 : /*
101 : * If the block size is smaller than the page size we need to check the
102 : * per-block uptodate status and adjust the offset and length if needed
103 : * to avoid reading in already uptodate ranges.
104 : */
105 0 : if (iop) {
106 : unsigned int i;
107 :
108 : /* move forward for each leading block marked uptodate */
109 0 : for (i = first; i <= last; i++) {
110 0 : if (!test_bit(i, iop->uptodate))
111 : break;
112 0 : *pos += block_size;
113 0 : poff += block_size;
114 0 : plen -= block_size;
115 0 : first++;
116 : }
117 :
118 : /* truncate len if we find any trailing uptodate block(s) */
119 0 : for ( ; i <= last; i++) {
120 0 : if (test_bit(i, iop->uptodate)) {
121 0 : plen -= (last - i + 1) * block_size;
122 0 : last = i - 1;
123 0 : break;
124 : }
125 : }
126 : }
127 :
128 : /*
129 : * If the extent spans the block that contains the i_size we need to
130 : * handle both halves separately so that we properly zero data in the
131 : * page cache for blocks that are entirely outside of i_size.
132 : */
133 0 : if (orig_pos <= isize && orig_pos + length > isize) {
134 0 : unsigned end = offset_in_page(isize - 1) >> block_bits;
135 :
136 0 : if (first <= end && last > end)
137 0 : plen -= (last - end) * block_size;
138 : }
139 :
140 0 : *offp = poff;
141 0 : *lenp = plen;
142 0 : }
143 :
144 : static void
145 0 : iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len)
146 : {
147 0 : struct iomap_page *iop = to_iomap_page(page);
148 0 : struct inode *inode = page->mapping->host;
149 0 : unsigned first = off >> inode->i_blkbits;
150 0 : unsigned last = (off + len - 1) >> inode->i_blkbits;
151 0 : unsigned long flags;
152 :
153 0 : spin_lock_irqsave(&iop->uptodate_lock, flags);
154 0 : bitmap_set(iop->uptodate, first, last - first + 1);
155 0 : if (bitmap_full(iop->uptodate, i_blocks_per_page(inode, page)))
156 0 : SetPageUptodate(page);
157 0 : spin_unlock_irqrestore(&iop->uptodate_lock, flags);
158 0 : }
159 :
160 : static void
161 0 : iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len)
162 : {
163 0 : if (PageError(page))
164 : return;
165 :
166 0 : if (page_has_private(page))
167 0 : iomap_iop_set_range_uptodate(page, off, len);
168 : else
169 0 : SetPageUptodate(page);
170 : }
171 :
172 : static void
173 0 : iomap_read_page_end_io(struct bio_vec *bvec, int error)
174 : {
175 0 : struct page *page = bvec->bv_page;
176 0 : struct iomap_page *iop = to_iomap_page(page);
177 :
178 0 : if (unlikely(error)) {
179 0 : ClearPageUptodate(page);
180 0 : SetPageError(page);
181 : } else {
182 0 : iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len);
183 : }
184 :
185 0 : if (!iop || atomic_sub_and_test(bvec->bv_len, &iop->read_bytes_pending))
186 0 : unlock_page(page);
187 0 : }
188 :
189 : static void
190 0 : iomap_read_end_io(struct bio *bio)
191 : {
192 0 : int error = blk_status_to_errno(bio->bi_status);
193 0 : struct bio_vec *bvec;
194 0 : struct bvec_iter_all iter_all;
195 :
196 0 : bio_for_each_segment_all(bvec, bio, iter_all)
197 0 : iomap_read_page_end_io(bvec, error);
198 0 : bio_put(bio);
199 0 : }
200 :
201 : struct iomap_readpage_ctx {
202 : struct page *cur_page;
203 : bool cur_page_in_bio;
204 : struct bio *bio;
205 : struct readahead_control *rac;
206 : };
207 :
208 : static void
209 0 : iomap_read_inline_data(struct inode *inode, struct page *page,
210 : struct iomap *iomap)
211 : {
212 0 : size_t size = i_size_read(inode);
213 0 : void *addr;
214 :
215 0 : if (PageUptodate(page))
216 : return;
217 :
218 0 : BUG_ON(page->index);
219 0 : BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data));
220 :
221 0 : addr = kmap_atomic(page);
222 0 : memcpy(addr, iomap->inline_data, size);
223 0 : memset(addr + size, 0, PAGE_SIZE - size);
224 0 : kunmap_atomic(addr);
225 0 : SetPageUptodate(page);
226 : }
227 :
228 0 : static inline bool iomap_block_needs_zeroing(struct inode *inode,
229 : struct iomap *iomap, loff_t pos)
230 : {
231 0 : return iomap->type != IOMAP_MAPPED ||
232 0 : (iomap->flags & IOMAP_F_NEW) ||
233 0 : pos >= i_size_read(inode);
234 : }
235 :
236 : static loff_t
237 0 : iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
238 : struct iomap *iomap, struct iomap *srcmap)
239 : {
240 0 : struct iomap_readpage_ctx *ctx = data;
241 0 : struct page *page = ctx->cur_page;
242 0 : struct iomap_page *iop = iomap_page_create(inode, page);
243 0 : bool same_page = false, is_contig = false;
244 0 : loff_t orig_pos = pos;
245 0 : unsigned poff, plen;
246 0 : sector_t sector;
247 :
248 0 : if (iomap->type == IOMAP_INLINE) {
249 0 : WARN_ON_ONCE(pos);
250 0 : iomap_read_inline_data(inode, page, iomap);
251 0 : return PAGE_SIZE;
252 : }
253 :
254 : /* zero post-eof blocks as the page may be mapped */
255 0 : iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen);
256 0 : if (plen == 0)
257 0 : goto done;
258 :
259 0 : if (iomap_block_needs_zeroing(inode, iomap, pos)) {
260 0 : zero_user(page, poff, plen);
261 0 : iomap_set_range_uptodate(page, poff, plen);
262 0 : goto done;
263 : }
264 :
265 0 : ctx->cur_page_in_bio = true;
266 0 : if (iop)
267 0 : atomic_add(plen, &iop->read_bytes_pending);
268 :
269 : /* Try to merge into a previous segment if we can */
270 0 : sector = iomap_sector(iomap, pos);
271 0 : if (ctx->bio && bio_end_sector(ctx->bio) == sector) {
272 0 : if (__bio_try_merge_page(ctx->bio, page, plen, poff,
273 : &same_page))
274 0 : goto done;
275 : is_contig = true;
276 : }
277 :
278 0 : if (!is_contig || bio_full(ctx->bio, plen)) {
279 0 : gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
280 0 : gfp_t orig_gfp = gfp;
281 0 : unsigned int nr_vecs = DIV_ROUND_UP(length, PAGE_SIZE);
282 :
283 0 : if (ctx->bio)
284 0 : submit_bio(ctx->bio);
285 :
286 0 : if (ctx->rac) /* same as readahead_gfp_mask */
287 0 : gfp |= __GFP_NORETRY | __GFP_NOWARN;
288 0 : ctx->bio = bio_alloc(gfp, bio_max_segs(nr_vecs));
289 : /*
290 : * If the bio_alloc fails, try it again for a single page to
291 : * avoid having to deal with partial page reads. This emulates
292 : * what do_mpage_readpage does.
293 : */
294 0 : if (!ctx->bio)
295 0 : ctx->bio = bio_alloc(orig_gfp, 1);
296 0 : ctx->bio->bi_opf = REQ_OP_READ;
297 0 : if (ctx->rac)
298 0 : ctx->bio->bi_opf |= REQ_RAHEAD;
299 0 : ctx->bio->bi_iter.bi_sector = sector;
300 0 : bio_set_dev(ctx->bio, iomap->bdev);
301 0 : ctx->bio->bi_end_io = iomap_read_end_io;
302 : }
303 :
304 0 : bio_add_page(ctx->bio, page, plen, poff);
305 0 : done:
306 : /*
307 : * Move the caller beyond our range so that it keeps making progress.
308 : * For that we have to include any leading non-uptodate ranges, but
309 : * we can skip trailing ones as they will be handled in the next
310 : * iteration.
311 : */
312 0 : return pos - orig_pos + plen;
313 : }
314 :
315 : int
316 0 : iomap_readpage(struct page *page, const struct iomap_ops *ops)
317 : {
318 0 : struct iomap_readpage_ctx ctx = { .cur_page = page };
319 0 : struct inode *inode = page->mapping->host;
320 0 : unsigned poff;
321 0 : loff_t ret;
322 :
323 0 : trace_iomap_readpage(page->mapping->host, 1);
324 :
325 0 : for (poff = 0; poff < PAGE_SIZE; poff += ret) {
326 0 : ret = iomap_apply(inode, page_offset(page) + poff,
327 0 : PAGE_SIZE - poff, 0, ops, &ctx,
328 : iomap_readpage_actor);
329 0 : if (ret <= 0) {
330 0 : WARN_ON_ONCE(ret == 0);
331 0 : SetPageError(page);
332 : break;
333 : }
334 : }
335 :
336 0 : if (ctx.bio) {
337 0 : submit_bio(ctx.bio);
338 0 : WARN_ON_ONCE(!ctx.cur_page_in_bio);
339 : } else {
340 0 : WARN_ON_ONCE(ctx.cur_page_in_bio);
341 0 : unlock_page(page);
342 : }
343 :
344 : /*
345 : * Just like mpage_readahead and block_read_full_page we always
346 : * return 0 and just mark the page as PageError on errors. This
347 : * should be cleaned up all through the stack eventually.
348 : */
349 0 : return 0;
350 : }
351 : EXPORT_SYMBOL_GPL(iomap_readpage);
352 :
353 : static loff_t
354 0 : iomap_readahead_actor(struct inode *inode, loff_t pos, loff_t length,
355 : void *data, struct iomap *iomap, struct iomap *srcmap)
356 : {
357 0 : struct iomap_readpage_ctx *ctx = data;
358 0 : loff_t done, ret;
359 :
360 0 : for (done = 0; done < length; done += ret) {
361 0 : if (ctx->cur_page && offset_in_page(pos + done) == 0) {
362 0 : if (!ctx->cur_page_in_bio)
363 0 : unlock_page(ctx->cur_page);
364 0 : put_page(ctx->cur_page);
365 0 : ctx->cur_page = NULL;
366 : }
367 0 : if (!ctx->cur_page) {
368 0 : ctx->cur_page = readahead_page(ctx->rac);
369 0 : ctx->cur_page_in_bio = false;
370 : }
371 0 : ret = iomap_readpage_actor(inode, pos + done, length - done,
372 : ctx, iomap, srcmap);
373 : }
374 :
375 0 : return done;
376 : }
377 :
378 : /**
379 : * iomap_readahead - Attempt to read pages from a file.
380 : * @rac: Describes the pages to be read.
381 : * @ops: The operations vector for the filesystem.
382 : *
383 : * This function is for filesystems to call to implement their readahead
384 : * address_space operation.
385 : *
386 : * Context: The @ops callbacks may submit I/O (eg to read the addresses of
387 : * blocks from disc), and may wait for it. The caller may be trying to
388 : * access a different page, and so sleeping excessively should be avoided.
389 : * It may allocate memory, but should avoid costly allocations. This
390 : * function is called with memalloc_nofs set, so allocations will not cause
391 : * the filesystem to be reentered.
392 : */
393 0 : void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops)
394 : {
395 0 : struct inode *inode = rac->mapping->host;
396 0 : loff_t pos = readahead_pos(rac);
397 0 : loff_t length = readahead_length(rac);
398 0 : struct iomap_readpage_ctx ctx = {
399 : .rac = rac,
400 : };
401 :
402 0 : trace_iomap_readahead(inode, readahead_count(rac));
403 :
404 0 : while (length > 0) {
405 0 : loff_t ret = iomap_apply(inode, pos, length, 0, ops,
406 : &ctx, iomap_readahead_actor);
407 0 : if (ret <= 0) {
408 0 : WARN_ON_ONCE(ret == 0);
409 : break;
410 : }
411 0 : pos += ret;
412 0 : length -= ret;
413 : }
414 :
415 0 : if (ctx.bio)
416 0 : submit_bio(ctx.bio);
417 0 : if (ctx.cur_page) {
418 0 : if (!ctx.cur_page_in_bio)
419 0 : unlock_page(ctx.cur_page);
420 0 : put_page(ctx.cur_page);
421 : }
422 0 : }
423 : EXPORT_SYMBOL_GPL(iomap_readahead);
424 :
425 : /*
426 : * iomap_is_partially_uptodate checks whether blocks within a page are
427 : * uptodate or not.
428 : *
429 : * Returns true if all blocks which correspond to a file portion
430 : * we want to read within the page are uptodate.
431 : */
432 : int
433 0 : iomap_is_partially_uptodate(struct page *page, unsigned long from,
434 : unsigned long count)
435 : {
436 0 : struct iomap_page *iop = to_iomap_page(page);
437 0 : struct inode *inode = page->mapping->host;
438 0 : unsigned len, first, last;
439 0 : unsigned i;
440 :
441 : /* Limit range to one page */
442 0 : len = min_t(unsigned, PAGE_SIZE - from, count);
443 :
444 : /* First and last blocks in range within page */
445 0 : first = from >> inode->i_blkbits;
446 0 : last = (from + len - 1) >> inode->i_blkbits;
447 :
448 0 : if (iop) {
449 0 : for (i = first; i <= last; i++)
450 0 : if (!test_bit(i, iop->uptodate))
451 : return 0;
452 : return 1;
453 : }
454 :
455 : return 0;
456 : }
457 : EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate);
458 :
459 : int
460 0 : iomap_releasepage(struct page *page, gfp_t gfp_mask)
461 : {
462 0 : trace_iomap_releasepage(page->mapping->host, page_offset(page),
463 : PAGE_SIZE);
464 :
465 : /*
466 : * mm accommodates an old ext3 case where clean pages might not have had
467 : * the dirty bit cleared. Thus, it can send actual dirty pages to
468 : * ->releasepage() via shrink_active_list(), skip those here.
469 : */
470 0 : if (PageDirty(page) || PageWriteback(page))
471 0 : return 0;
472 0 : iomap_page_release(page);
473 0 : return 1;
474 : }
475 : EXPORT_SYMBOL_GPL(iomap_releasepage);
476 :
477 : void
478 0 : iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len)
479 : {
480 0 : trace_iomap_invalidatepage(page->mapping->host, offset, len);
481 :
482 : /*
483 : * If we are invalidating the entire page, clear the dirty state from it
484 : * and release it to avoid unnecessary buildup of the LRU.
485 : */
486 0 : if (offset == 0 && len == PAGE_SIZE) {
487 0 : WARN_ON_ONCE(PageWriteback(page));
488 0 : cancel_dirty_page(page);
489 0 : iomap_page_release(page);
490 : }
491 0 : }
492 : EXPORT_SYMBOL_GPL(iomap_invalidatepage);
493 :
494 : #ifdef CONFIG_MIGRATION
495 : int
496 0 : iomap_migrate_page(struct address_space *mapping, struct page *newpage,
497 : struct page *page, enum migrate_mode mode)
498 : {
499 0 : int ret;
500 :
501 0 : ret = migrate_page_move_mapping(mapping, newpage, page, 0);
502 0 : if (ret != MIGRATEPAGE_SUCCESS)
503 : return ret;
504 :
505 0 : if (page_has_private(page))
506 0 : attach_page_private(newpage, detach_page_private(page));
507 :
508 0 : if (mode != MIGRATE_SYNC_NO_COPY)
509 0 : migrate_page_copy(newpage, page);
510 : else
511 0 : migrate_page_states(newpage, page);
512 : return MIGRATEPAGE_SUCCESS;
513 : }
514 : EXPORT_SYMBOL_GPL(iomap_migrate_page);
515 : #endif /* CONFIG_MIGRATION */
516 :
517 : enum {
518 : IOMAP_WRITE_F_UNSHARE = (1 << 0),
519 : };
520 :
521 : static void
522 0 : iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
523 : {
524 0 : loff_t i_size = i_size_read(inode);
525 :
526 : /*
527 : * Only truncate newly allocated pages beyoned EOF, even if the
528 : * write started inside the existing inode size.
529 : */
530 0 : if (pos + len > i_size)
531 0 : truncate_pagecache_range(inode, max(pos, i_size), pos + len);
532 0 : }
533 :
534 : static int
535 0 : iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff,
536 : unsigned plen, struct iomap *iomap)
537 : {
538 0 : struct bio_vec bvec;
539 0 : struct bio bio;
540 :
541 0 : bio_init(&bio, &bvec, 1);
542 0 : bio.bi_opf = REQ_OP_READ;
543 0 : bio.bi_iter.bi_sector = iomap_sector(iomap, block_start);
544 0 : bio_set_dev(&bio, iomap->bdev);
545 0 : __bio_add_page(&bio, page, plen, poff);
546 0 : return submit_bio_wait(&bio);
547 : }
548 :
549 : static int
550 0 : __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags,
551 : struct page *page, struct iomap *srcmap)
552 : {
553 0 : struct iomap_page *iop = iomap_page_create(inode, page);
554 0 : loff_t block_size = i_blocksize(inode);
555 0 : loff_t block_start = round_down(pos, block_size);
556 0 : loff_t block_end = round_up(pos + len, block_size);
557 0 : unsigned from = offset_in_page(pos), to = from + len, poff, plen;
558 :
559 0 : if (PageUptodate(page))
560 : return 0;
561 0 : ClearPageError(page);
562 :
563 0 : do {
564 0 : iomap_adjust_read_range(inode, iop, &block_start,
565 : block_end - block_start, &poff, &plen);
566 0 : if (plen == 0)
567 : break;
568 :
569 0 : if (!(flags & IOMAP_WRITE_F_UNSHARE) &&
570 0 : (from <= poff || from >= poff + plen) &&
571 0 : (to <= poff || to >= poff + plen))
572 0 : continue;
573 :
574 0 : if (iomap_block_needs_zeroing(inode, srcmap, block_start)) {
575 0 : if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE))
576 : return -EIO;
577 0 : zero_user_segments(page, poff, from, to, poff + plen);
578 : } else {
579 0 : int status = iomap_read_page_sync(block_start, page,
580 : poff, plen, srcmap);
581 0 : if (status)
582 0 : return status;
583 : }
584 0 : iomap_set_range_uptodate(page, poff, plen);
585 0 : } while ((block_start += plen) < block_end);
586 :
587 : return 0;
588 : }
589 :
590 : static int
591 0 : iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags,
592 : struct page **pagep, struct iomap *iomap, struct iomap *srcmap)
593 : {
594 0 : const struct iomap_page_ops *page_ops = iomap->page_ops;
595 0 : struct page *page;
596 0 : int status = 0;
597 :
598 0 : BUG_ON(pos + len > iomap->offset + iomap->length);
599 0 : if (srcmap != iomap)
600 0 : BUG_ON(pos + len > srcmap->offset + srcmap->length);
601 :
602 0 : if (fatal_signal_pending(current))
603 : return -EINTR;
604 :
605 0 : if (page_ops && page_ops->page_prepare) {
606 0 : status = page_ops->page_prepare(inode, pos, len, iomap);
607 0 : if (status)
608 : return status;
609 : }
610 :
611 0 : page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT,
612 : AOP_FLAG_NOFS);
613 0 : if (!page) {
614 0 : status = -ENOMEM;
615 0 : goto out_no_page;
616 : }
617 :
618 0 : if (srcmap->type == IOMAP_INLINE)
619 0 : iomap_read_inline_data(inode, page, srcmap);
620 0 : else if (iomap->flags & IOMAP_F_BUFFER_HEAD)
621 0 : status = __block_write_begin_int(page, pos, len, NULL, srcmap);
622 : else
623 0 : status = __iomap_write_begin(inode, pos, len, flags, page,
624 : srcmap);
625 :
626 0 : if (unlikely(status))
627 0 : goto out_unlock;
628 :
629 0 : *pagep = page;
630 0 : return 0;
631 :
632 0 : out_unlock:
633 0 : unlock_page(page);
634 0 : put_page(page);
635 0 : iomap_write_failed(inode, pos, len);
636 :
637 0 : out_no_page:
638 0 : if (page_ops && page_ops->page_done)
639 0 : page_ops->page_done(inode, pos, 0, NULL, iomap);
640 : return status;
641 : }
642 :
643 : int
644 0 : iomap_set_page_dirty(struct page *page)
645 : {
646 0 : struct address_space *mapping = page_mapping(page);
647 0 : int newly_dirty;
648 :
649 0 : if (unlikely(!mapping))
650 0 : return !TestSetPageDirty(page);
651 :
652 : /*
653 : * Lock out page's memcg migration to keep PageDirty
654 : * synchronized with per-memcg dirty page counters.
655 : */
656 0 : lock_page_memcg(page);
657 0 : newly_dirty = !TestSetPageDirty(page);
658 0 : if (newly_dirty)
659 0 : __set_page_dirty(page, mapping, 0);
660 0 : unlock_page_memcg(page);
661 :
662 0 : if (newly_dirty)
663 0 : __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
664 : return newly_dirty;
665 : }
666 : EXPORT_SYMBOL_GPL(iomap_set_page_dirty);
667 :
668 0 : static size_t __iomap_write_end(struct inode *inode, loff_t pos, size_t len,
669 : size_t copied, struct page *page)
670 : {
671 0 : flush_dcache_page(page);
672 :
673 : /*
674 : * The blocks that were entirely written will now be uptodate, so we
675 : * don't have to worry about a readpage reading them and overwriting a
676 : * partial write. However if we have encountered a short write and only
677 : * partially written into a block, it will not be marked uptodate, so a
678 : * readpage might come in and destroy our partial write.
679 : *
680 : * Do the simplest thing, and just treat any short write to a non
681 : * uptodate page as a zero-length write, and force the caller to redo
682 : * the whole thing.
683 : */
684 0 : if (unlikely(copied < len && !PageUptodate(page)))
685 : return 0;
686 0 : iomap_set_range_uptodate(page, offset_in_page(pos), len);
687 0 : iomap_set_page_dirty(page);
688 0 : return copied;
689 : }
690 :
691 0 : static size_t iomap_write_end_inline(struct inode *inode, struct page *page,
692 : struct iomap *iomap, loff_t pos, size_t copied)
693 : {
694 0 : void *addr;
695 :
696 0 : WARN_ON_ONCE(!PageUptodate(page));
697 0 : BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data));
698 :
699 0 : flush_dcache_page(page);
700 0 : addr = kmap_atomic(page);
701 0 : memcpy(iomap->inline_data + pos, addr + pos, copied);
702 0 : kunmap_atomic(addr);
703 :
704 0 : mark_inode_dirty(inode);
705 0 : return copied;
706 : }
707 :
708 : /* Returns the number of bytes copied. May be 0. Cannot be an errno. */
709 0 : static size_t iomap_write_end(struct inode *inode, loff_t pos, size_t len,
710 : size_t copied, struct page *page, struct iomap *iomap,
711 : struct iomap *srcmap)
712 : {
713 0 : const struct iomap_page_ops *page_ops = iomap->page_ops;
714 0 : loff_t old_size = inode->i_size;
715 0 : size_t ret;
716 :
717 0 : if (srcmap->type == IOMAP_INLINE) {
718 0 : ret = iomap_write_end_inline(inode, page, iomap, pos, copied);
719 0 : } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) {
720 0 : ret = block_write_end(NULL, inode->i_mapping, pos, len, copied,
721 : page, NULL);
722 : } else {
723 0 : ret = __iomap_write_end(inode, pos, len, copied, page);
724 : }
725 :
726 : /*
727 : * Update the in-memory inode size after copying the data into the page
728 : * cache. It's up to the file system to write the updated size to disk,
729 : * preferably after I/O completion so that no stale data is exposed.
730 : */
731 0 : if (pos + ret > old_size) {
732 0 : i_size_write(inode, pos + ret);
733 0 : iomap->flags |= IOMAP_F_SIZE_CHANGED;
734 : }
735 0 : unlock_page(page);
736 :
737 0 : if (old_size < pos)
738 0 : pagecache_isize_extended(inode, old_size, pos);
739 0 : if (page_ops && page_ops->page_done)
740 0 : page_ops->page_done(inode, pos, ret, page, iomap);
741 0 : put_page(page);
742 :
743 0 : if (ret < len)
744 0 : iomap_write_failed(inode, pos, len);
745 0 : return ret;
746 : }
747 :
748 : static loff_t
749 0 : iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
750 : struct iomap *iomap, struct iomap *srcmap)
751 : {
752 0 : struct iov_iter *i = data;
753 0 : long status = 0;
754 0 : ssize_t written = 0;
755 :
756 0 : do {
757 0 : struct page *page;
758 0 : unsigned long offset; /* Offset into pagecache page */
759 0 : unsigned long bytes; /* Bytes to write to page */
760 0 : size_t copied; /* Bytes copied from user */
761 :
762 0 : offset = offset_in_page(pos);
763 0 : bytes = min_t(unsigned long, PAGE_SIZE - offset,
764 : iov_iter_count(i));
765 0 : again:
766 0 : if (bytes > length)
767 : bytes = length;
768 :
769 : /*
770 : * Bring in the user page that we will copy from _first_.
771 : * Otherwise there's a nasty deadlock on copying from the
772 : * same page as we're writing to, without it being marked
773 : * up-to-date.
774 : *
775 : * Not only is this an optimisation, but it is also required
776 : * to check that the address is actually valid, when atomic
777 : * usercopies are used, below.
778 : */
779 0 : if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
780 : status = -EFAULT;
781 0 : break;
782 : }
783 :
784 0 : status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap,
785 : srcmap);
786 0 : if (unlikely(status))
787 : break;
788 :
789 0 : if (mapping_writably_mapped(inode->i_mapping))
790 0 : flush_dcache_page(page);
791 :
792 0 : copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
793 :
794 0 : copied = iomap_write_end(inode, pos, bytes, copied, page, iomap,
795 : srcmap);
796 :
797 0 : cond_resched();
798 :
799 0 : iov_iter_advance(i, copied);
800 0 : if (unlikely(copied == 0)) {
801 : /*
802 : * If we were unable to copy any data at all, we must
803 : * fall back to a single segment length write.
804 : *
805 : * If we didn't fallback here, we could livelock
806 : * because not all segments in the iov can be copied at
807 : * once without a pagefault.
808 : */
809 0 : bytes = min_t(unsigned long, PAGE_SIZE - offset,
810 : iov_iter_single_seg_count(i));
811 0 : goto again;
812 : }
813 0 : pos += copied;
814 0 : written += copied;
815 0 : length -= copied;
816 :
817 0 : balance_dirty_pages_ratelimited(inode->i_mapping);
818 0 : } while (iov_iter_count(i) && length);
819 :
820 0 : return written ? written : status;
821 : }
822 :
823 : ssize_t
824 0 : iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter,
825 : const struct iomap_ops *ops)
826 : {
827 0 : struct inode *inode = iocb->ki_filp->f_mapping->host;
828 0 : loff_t pos = iocb->ki_pos, ret = 0, written = 0;
829 :
830 0 : while (iov_iter_count(iter)) {
831 0 : ret = iomap_apply(inode, pos, iov_iter_count(iter),
832 : IOMAP_WRITE, ops, iter, iomap_write_actor);
833 0 : if (ret <= 0)
834 : break;
835 0 : pos += ret;
836 0 : written += ret;
837 : }
838 :
839 0 : return written ? written : ret;
840 : }
841 : EXPORT_SYMBOL_GPL(iomap_file_buffered_write);
842 :
843 : static loff_t
844 0 : iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
845 : struct iomap *iomap, struct iomap *srcmap)
846 : {
847 0 : long status = 0;
848 0 : loff_t written = 0;
849 :
850 : /* don't bother with blocks that are not shared to start with */
851 0 : if (!(iomap->flags & IOMAP_F_SHARED))
852 : return length;
853 : /* don't bother with holes or unwritten extents */
854 0 : if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
855 : return length;
856 :
857 0 : do {
858 0 : unsigned long offset = offset_in_page(pos);
859 0 : unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length);
860 0 : struct page *page;
861 :
862 0 : status = iomap_write_begin(inode, pos, bytes,
863 : IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap);
864 0 : if (unlikely(status))
865 0 : return status;
866 :
867 0 : status = iomap_write_end(inode, pos, bytes, bytes, page, iomap,
868 : srcmap);
869 0 : if (WARN_ON_ONCE(status == 0))
870 : return -EIO;
871 :
872 0 : cond_resched();
873 :
874 0 : pos += status;
875 0 : written += status;
876 0 : length -= status;
877 :
878 0 : balance_dirty_pages_ratelimited(inode->i_mapping);
879 0 : } while (length);
880 :
881 : return written;
882 : }
883 :
884 : int
885 0 : iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len,
886 : const struct iomap_ops *ops)
887 : {
888 0 : loff_t ret;
889 :
890 0 : while (len) {
891 0 : ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL,
892 : iomap_unshare_actor);
893 0 : if (ret <= 0)
894 0 : return ret;
895 0 : pos += ret;
896 0 : len -= ret;
897 : }
898 :
899 : return 0;
900 : }
901 : EXPORT_SYMBOL_GPL(iomap_file_unshare);
902 :
903 0 : static s64 iomap_zero(struct inode *inode, loff_t pos, u64 length,
904 : struct iomap *iomap, struct iomap *srcmap)
905 : {
906 0 : struct page *page;
907 0 : int status;
908 0 : unsigned offset = offset_in_page(pos);
909 0 : unsigned bytes = min_t(u64, PAGE_SIZE - offset, length);
910 :
911 0 : status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap);
912 0 : if (status)
913 0 : return status;
914 :
915 0 : zero_user(page, offset, bytes);
916 0 : mark_page_accessed(page);
917 :
918 0 : return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap);
919 : }
920 :
921 0 : static loff_t iomap_zero_range_actor(struct inode *inode, loff_t pos,
922 : loff_t length, void *data, struct iomap *iomap,
923 : struct iomap *srcmap)
924 : {
925 0 : bool *did_zero = data;
926 0 : loff_t written = 0;
927 :
928 : /* already zeroed? we're done. */
929 0 : if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN)
930 : return length;
931 :
932 0 : do {
933 0 : s64 bytes;
934 :
935 0 : if (IS_DAX(inode))
936 : bytes = dax_iomap_zero(pos, length, iomap);
937 : else
938 0 : bytes = iomap_zero(inode, pos, length, iomap, srcmap);
939 0 : if (bytes < 0)
940 0 : return bytes;
941 :
942 0 : pos += bytes;
943 0 : length -= bytes;
944 0 : written += bytes;
945 0 : if (did_zero)
946 0 : *did_zero = true;
947 0 : } while (length > 0);
948 :
949 : return written;
950 : }
951 :
952 : int
953 0 : iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero,
954 : const struct iomap_ops *ops)
955 : {
956 0 : loff_t ret;
957 :
958 0 : while (len > 0) {
959 0 : ret = iomap_apply(inode, pos, len, IOMAP_ZERO,
960 : ops, did_zero, iomap_zero_range_actor);
961 0 : if (ret <= 0)
962 0 : return ret;
963 :
964 0 : pos += ret;
965 0 : len -= ret;
966 : }
967 :
968 : return 0;
969 : }
970 : EXPORT_SYMBOL_GPL(iomap_zero_range);
971 :
972 : int
973 0 : iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
974 : const struct iomap_ops *ops)
975 : {
976 0 : unsigned int blocksize = i_blocksize(inode);
977 0 : unsigned int off = pos & (blocksize - 1);
978 :
979 : /* Block boundary? Nothing to do */
980 0 : if (!off)
981 : return 0;
982 0 : return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
983 : }
984 : EXPORT_SYMBOL_GPL(iomap_truncate_page);
985 :
986 : static loff_t
987 0 : iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length,
988 : void *data, struct iomap *iomap, struct iomap *srcmap)
989 : {
990 0 : struct page *page = data;
991 0 : int ret;
992 :
993 0 : if (iomap->flags & IOMAP_F_BUFFER_HEAD) {
994 0 : ret = __block_write_begin_int(page, pos, length, NULL, iomap);
995 0 : if (ret)
996 0 : return ret;
997 0 : block_commit_write(page, 0, length);
998 : } else {
999 0 : WARN_ON_ONCE(!PageUptodate(page));
1000 0 : iomap_page_create(inode, page);
1001 0 : set_page_dirty(page);
1002 : }
1003 :
1004 : return length;
1005 : }
1006 :
1007 0 : vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops)
1008 : {
1009 0 : struct page *page = vmf->page;
1010 0 : struct inode *inode = file_inode(vmf->vma->vm_file);
1011 0 : unsigned long length;
1012 0 : loff_t offset;
1013 0 : ssize_t ret;
1014 :
1015 0 : lock_page(page);
1016 0 : ret = page_mkwrite_check_truncate(page, inode);
1017 0 : if (ret < 0)
1018 0 : goto out_unlock;
1019 0 : length = ret;
1020 :
1021 0 : offset = page_offset(page);
1022 0 : while (length > 0) {
1023 0 : ret = iomap_apply(inode, offset, length,
1024 : IOMAP_WRITE | IOMAP_FAULT, ops, page,
1025 : iomap_page_mkwrite_actor);
1026 0 : if (unlikely(ret <= 0))
1027 0 : goto out_unlock;
1028 0 : offset += ret;
1029 0 : length -= ret;
1030 : }
1031 :
1032 0 : wait_for_stable_page(page);
1033 0 : return VM_FAULT_LOCKED;
1034 0 : out_unlock:
1035 0 : unlock_page(page);
1036 0 : return block_page_mkwrite_return(ret);
1037 : }
1038 : EXPORT_SYMBOL_GPL(iomap_page_mkwrite);
1039 :
1040 : static void
1041 0 : iomap_finish_page_writeback(struct inode *inode, struct page *page,
1042 : int error, unsigned int len)
1043 : {
1044 0 : struct iomap_page *iop = to_iomap_page(page);
1045 :
1046 0 : if (error) {
1047 0 : SetPageError(page);
1048 0 : mapping_set_error(inode->i_mapping, -EIO);
1049 : }
1050 :
1051 0 : WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop);
1052 0 : WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) <= 0);
1053 :
1054 0 : if (!iop || atomic_sub_and_test(len, &iop->write_bytes_pending))
1055 0 : end_page_writeback(page);
1056 0 : }
1057 :
1058 : /*
1059 : * We're now finished for good with this ioend structure. Update the page
1060 : * state, release holds on bios, and finally free up memory. Do not use the
1061 : * ioend after this.
1062 : */
1063 : static void
1064 0 : iomap_finish_ioend(struct iomap_ioend *ioend, int error)
1065 : {
1066 0 : struct inode *inode = ioend->io_inode;
1067 0 : struct bio *bio = &ioend->io_inline_bio;
1068 0 : struct bio *last = ioend->io_bio, *next;
1069 0 : u64 start = bio->bi_iter.bi_sector;
1070 0 : loff_t offset = ioend->io_offset;
1071 0 : bool quiet = bio_flagged(bio, BIO_QUIET);
1072 :
1073 0 : for (bio = &ioend->io_inline_bio; bio; bio = next) {
1074 0 : struct bio_vec *bv;
1075 0 : struct bvec_iter_all iter_all;
1076 :
1077 : /*
1078 : * For the last bio, bi_private points to the ioend, so we
1079 : * need to explicitly end the iteration here.
1080 : */
1081 0 : if (bio == last)
1082 : next = NULL;
1083 : else
1084 0 : next = bio->bi_private;
1085 :
1086 : /* walk each page on bio, ending page IO on them */
1087 0 : bio_for_each_segment_all(bv, bio, iter_all)
1088 0 : iomap_finish_page_writeback(inode, bv->bv_page, error,
1089 : bv->bv_len);
1090 0 : bio_put(bio);
1091 : }
1092 : /* The ioend has been freed by bio_put() */
1093 :
1094 0 : if (unlikely(error && !quiet)) {
1095 0 : printk_ratelimited(KERN_ERR
1096 : "%s: writeback error on inode %lu, offset %lld, sector %llu",
1097 : inode->i_sb->s_id, inode->i_ino, offset, start);
1098 : }
1099 0 : }
1100 :
1101 : void
1102 0 : iomap_finish_ioends(struct iomap_ioend *ioend, int error)
1103 : {
1104 0 : struct list_head tmp;
1105 :
1106 0 : list_replace_init(&ioend->io_list, &tmp);
1107 0 : iomap_finish_ioend(ioend, error);
1108 :
1109 0 : while (!list_empty(&tmp)) {
1110 0 : ioend = list_first_entry(&tmp, struct iomap_ioend, io_list);
1111 0 : list_del_init(&ioend->io_list);
1112 0 : iomap_finish_ioend(ioend, error);
1113 : }
1114 0 : }
1115 : EXPORT_SYMBOL_GPL(iomap_finish_ioends);
1116 :
1117 : /*
1118 : * We can merge two adjacent ioends if they have the same set of work to do.
1119 : */
1120 : static bool
1121 0 : iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next)
1122 : {
1123 0 : if (ioend->io_bio->bi_status != next->io_bio->bi_status)
1124 : return false;
1125 0 : if ((ioend->io_flags & IOMAP_F_SHARED) ^
1126 0 : (next->io_flags & IOMAP_F_SHARED))
1127 : return false;
1128 0 : if ((ioend->io_type == IOMAP_UNWRITTEN) ^
1129 0 : (next->io_type == IOMAP_UNWRITTEN))
1130 : return false;
1131 0 : if (ioend->io_offset + ioend->io_size != next->io_offset)
1132 0 : return false;
1133 : return true;
1134 : }
1135 :
1136 : void
1137 0 : iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends,
1138 : void (*merge_private)(struct iomap_ioend *ioend,
1139 : struct iomap_ioend *next))
1140 : {
1141 0 : struct iomap_ioend *next;
1142 :
1143 0 : INIT_LIST_HEAD(&ioend->io_list);
1144 :
1145 0 : while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend,
1146 : io_list))) {
1147 0 : if (!iomap_ioend_can_merge(ioend, next))
1148 : break;
1149 0 : list_move_tail(&next->io_list, &ioend->io_list);
1150 0 : ioend->io_size += next->io_size;
1151 0 : if (next->io_private && merge_private)
1152 0 : merge_private(ioend, next);
1153 : }
1154 0 : }
1155 : EXPORT_SYMBOL_GPL(iomap_ioend_try_merge);
1156 :
1157 : static int
1158 0 : iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b)
1159 : {
1160 0 : struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list);
1161 0 : struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list);
1162 :
1163 0 : if (ia->io_offset < ib->io_offset)
1164 : return -1;
1165 0 : if (ia->io_offset > ib->io_offset)
1166 0 : return 1;
1167 : return 0;
1168 : }
1169 :
1170 : void
1171 0 : iomap_sort_ioends(struct list_head *ioend_list)
1172 : {
1173 0 : list_sort(NULL, ioend_list, iomap_ioend_compare);
1174 0 : }
1175 : EXPORT_SYMBOL_GPL(iomap_sort_ioends);
1176 :
1177 0 : static void iomap_writepage_end_bio(struct bio *bio)
1178 : {
1179 0 : struct iomap_ioend *ioend = bio->bi_private;
1180 :
1181 0 : iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status));
1182 0 : }
1183 :
1184 : /*
1185 : * Submit the final bio for an ioend.
1186 : *
1187 : * If @error is non-zero, it means that we have a situation where some part of
1188 : * the submission process has failed after we have marked paged for writeback
1189 : * and unlocked them. In this situation, we need to fail the bio instead of
1190 : * submitting it. This typically only happens on a filesystem shutdown.
1191 : */
1192 : static int
1193 0 : iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend,
1194 : int error)
1195 : {
1196 0 : ioend->io_bio->bi_private = ioend;
1197 0 : ioend->io_bio->bi_end_io = iomap_writepage_end_bio;
1198 :
1199 0 : if (wpc->ops->prepare_ioend)
1200 0 : error = wpc->ops->prepare_ioend(ioend, error);
1201 0 : if (error) {
1202 : /*
1203 : * If we are failing the IO now, just mark the ioend with an
1204 : * error and finish it. This will run IO completion immediately
1205 : * as there is only one reference to the ioend at this point in
1206 : * time.
1207 : */
1208 0 : ioend->io_bio->bi_status = errno_to_blk_status(error);
1209 0 : bio_endio(ioend->io_bio);
1210 0 : return error;
1211 : }
1212 :
1213 0 : submit_bio(ioend->io_bio);
1214 0 : return 0;
1215 : }
1216 :
1217 : static struct iomap_ioend *
1218 0 : iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc,
1219 : loff_t offset, sector_t sector, struct writeback_control *wbc)
1220 : {
1221 0 : struct iomap_ioend *ioend;
1222 0 : struct bio *bio;
1223 :
1224 0 : bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_VECS, &iomap_ioend_bioset);
1225 0 : bio_set_dev(bio, wpc->iomap.bdev);
1226 0 : bio->bi_iter.bi_sector = sector;
1227 0 : bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
1228 0 : bio->bi_write_hint = inode->i_write_hint;
1229 0 : wbc_init_bio(wbc, bio);
1230 :
1231 0 : ioend = container_of(bio, struct iomap_ioend, io_inline_bio);
1232 0 : INIT_LIST_HEAD(&ioend->io_list);
1233 0 : ioend->io_type = wpc->iomap.type;
1234 0 : ioend->io_flags = wpc->iomap.flags;
1235 0 : ioend->io_inode = inode;
1236 0 : ioend->io_size = 0;
1237 0 : ioend->io_offset = offset;
1238 0 : ioend->io_private = NULL;
1239 0 : ioend->io_bio = bio;
1240 0 : return ioend;
1241 : }
1242 :
1243 : /*
1244 : * Allocate a new bio, and chain the old bio to the new one.
1245 : *
1246 : * Note that we have to do perform the chaining in this unintuitive order
1247 : * so that the bi_private linkage is set up in the right direction for the
1248 : * traversal in iomap_finish_ioend().
1249 : */
1250 : static struct bio *
1251 0 : iomap_chain_bio(struct bio *prev)
1252 : {
1253 0 : struct bio *new;
1254 :
1255 0 : new = bio_alloc(GFP_NOFS, BIO_MAX_VECS);
1256 0 : bio_copy_dev(new, prev);/* also copies over blkcg information */
1257 0 : new->bi_iter.bi_sector = bio_end_sector(prev);
1258 0 : new->bi_opf = prev->bi_opf;
1259 0 : new->bi_write_hint = prev->bi_write_hint;
1260 :
1261 0 : bio_chain(prev, new);
1262 0 : bio_get(prev); /* for iomap_finish_ioend */
1263 0 : submit_bio(prev);
1264 0 : return new;
1265 : }
1266 :
1267 : static bool
1268 0 : iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset,
1269 : sector_t sector)
1270 : {
1271 0 : if ((wpc->iomap.flags & IOMAP_F_SHARED) !=
1272 0 : (wpc->ioend->io_flags & IOMAP_F_SHARED))
1273 : return false;
1274 0 : if (wpc->iomap.type != wpc->ioend->io_type)
1275 : return false;
1276 0 : if (offset != wpc->ioend->io_offset + wpc->ioend->io_size)
1277 : return false;
1278 0 : if (sector != bio_end_sector(wpc->ioend->io_bio))
1279 0 : return false;
1280 : return true;
1281 : }
1282 :
1283 : /*
1284 : * Test to see if we have an existing ioend structure that we could append to
1285 : * first, otherwise finish off the current ioend and start another.
1286 : */
1287 : static void
1288 0 : iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page,
1289 : struct iomap_page *iop, struct iomap_writepage_ctx *wpc,
1290 : struct writeback_control *wbc, struct list_head *iolist)
1291 : {
1292 0 : sector_t sector = iomap_sector(&wpc->iomap, offset);
1293 0 : unsigned len = i_blocksize(inode);
1294 0 : unsigned poff = offset & (PAGE_SIZE - 1);
1295 0 : bool merged, same_page = false;
1296 :
1297 0 : if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) {
1298 0 : if (wpc->ioend)
1299 0 : list_add(&wpc->ioend->io_list, iolist);
1300 0 : wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc);
1301 : }
1302 :
1303 0 : merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff,
1304 : &same_page);
1305 0 : if (iop)
1306 0 : atomic_add(len, &iop->write_bytes_pending);
1307 :
1308 0 : if (!merged) {
1309 0 : if (bio_full(wpc->ioend->io_bio, len)) {
1310 0 : wpc->ioend->io_bio =
1311 0 : iomap_chain_bio(wpc->ioend->io_bio);
1312 : }
1313 0 : bio_add_page(wpc->ioend->io_bio, page, len, poff);
1314 : }
1315 :
1316 0 : wpc->ioend->io_size += len;
1317 0 : wbc_account_cgroup_owner(wbc, page, len);
1318 0 : }
1319 :
1320 : /*
1321 : * We implement an immediate ioend submission policy here to avoid needing to
1322 : * chain multiple ioends and hence nest mempool allocations which can violate
1323 : * forward progress guarantees we need to provide. The current ioend we are
1324 : * adding blocks to is cached on the writepage context, and if the new block
1325 : * does not append to the cached ioend it will create a new ioend and cache that
1326 : * instead.
1327 : *
1328 : * If a new ioend is created and cached, the old ioend is returned and queued
1329 : * locally for submission once the entire page is processed or an error has been
1330 : * detected. While ioends are submitted immediately after they are completed,
1331 : * batching optimisations are provided by higher level block plugging.
1332 : *
1333 : * At the end of a writeback pass, there will be a cached ioend remaining on the
1334 : * writepage context that the caller will need to submit.
1335 : */
1336 : static int
1337 0 : iomap_writepage_map(struct iomap_writepage_ctx *wpc,
1338 : struct writeback_control *wbc, struct inode *inode,
1339 : struct page *page, u64 end_offset)
1340 : {
1341 0 : struct iomap_page *iop = to_iomap_page(page);
1342 0 : struct iomap_ioend *ioend, *next;
1343 0 : unsigned len = i_blocksize(inode);
1344 0 : u64 file_offset; /* file offset of page */
1345 0 : int error = 0, count = 0, i;
1346 0 : LIST_HEAD(submit_list);
1347 :
1348 0 : WARN_ON_ONCE(i_blocks_per_page(inode, page) > 1 && !iop);
1349 0 : WARN_ON_ONCE(iop && atomic_read(&iop->write_bytes_pending) != 0);
1350 :
1351 : /*
1352 : * Walk through the page to find areas to write back. If we run off the
1353 : * end of the current map or find the current map invalid, grab a new
1354 : * one.
1355 : */
1356 0 : for (i = 0, file_offset = page_offset(page);
1357 0 : i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset;
1358 0 : i++, file_offset += len) {
1359 0 : if (iop && !test_bit(i, iop->uptodate))
1360 0 : continue;
1361 :
1362 0 : error = wpc->ops->map_blocks(wpc, inode, file_offset);
1363 0 : if (error)
1364 : break;
1365 0 : if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE))
1366 0 : continue;
1367 0 : if (wpc->iomap.type == IOMAP_HOLE)
1368 0 : continue;
1369 0 : iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc,
1370 : &submit_list);
1371 0 : count++;
1372 : }
1373 :
1374 0 : WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list));
1375 0 : WARN_ON_ONCE(!PageLocked(page));
1376 0 : WARN_ON_ONCE(PageWriteback(page));
1377 0 : WARN_ON_ONCE(PageDirty(page));
1378 :
1379 : /*
1380 : * We cannot cancel the ioend directly here on error. We may have
1381 : * already set other pages under writeback and hence we have to run I/O
1382 : * completion to mark the error state of the pages under writeback
1383 : * appropriately.
1384 : */
1385 0 : if (unlikely(error)) {
1386 : /*
1387 : * Let the filesystem know what portion of the current page
1388 : * failed to map. If the page wasn't been added to ioend, it
1389 : * won't be affected by I/O completion and we must unlock it
1390 : * now.
1391 : */
1392 0 : if (wpc->ops->discard_page)
1393 0 : wpc->ops->discard_page(page, file_offset);
1394 0 : if (!count) {
1395 0 : ClearPageUptodate(page);
1396 0 : unlock_page(page);
1397 0 : goto done;
1398 : }
1399 : }
1400 :
1401 0 : set_page_writeback(page);
1402 0 : unlock_page(page);
1403 :
1404 : /*
1405 : * Preserve the original error if there was one, otherwise catch
1406 : * submission errors here and propagate into subsequent ioend
1407 : * submissions.
1408 : */
1409 0 : list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
1410 0 : int error2;
1411 :
1412 0 : list_del_init(&ioend->io_list);
1413 0 : error2 = iomap_submit_ioend(wpc, ioend, error);
1414 0 : if (error2 && !error)
1415 0 : error = error2;
1416 : }
1417 :
1418 : /*
1419 : * We can end up here with no error and nothing to write only if we race
1420 : * with a partial page truncate on a sub-page block sized filesystem.
1421 : */
1422 0 : if (!count)
1423 0 : end_page_writeback(page);
1424 0 : done:
1425 0 : mapping_set_error(page->mapping, error);
1426 0 : return error;
1427 : }
1428 :
1429 : /*
1430 : * Write out a dirty page.
1431 : *
1432 : * For delalloc space on the page we need to allocate space and flush it.
1433 : * For unwritten space on the page we need to start the conversion to
1434 : * regular allocated space.
1435 : */
1436 : static int
1437 0 : iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data)
1438 : {
1439 0 : struct iomap_writepage_ctx *wpc = data;
1440 0 : struct inode *inode = page->mapping->host;
1441 0 : pgoff_t end_index;
1442 0 : u64 end_offset;
1443 0 : loff_t offset;
1444 :
1445 0 : trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE);
1446 :
1447 : /*
1448 : * Refuse to write the page out if we are called from reclaim context.
1449 : *
1450 : * This avoids stack overflows when called from deeply used stacks in
1451 : * random callers for direct reclaim or memcg reclaim. We explicitly
1452 : * allow reclaim from kswapd as the stack usage there is relatively low.
1453 : *
1454 : * This should never happen except in the case of a VM regression so
1455 : * warn about it.
1456 : */
1457 0 : if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1458 : PF_MEMALLOC))
1459 0 : goto redirty;
1460 :
1461 : /*
1462 : * Is this page beyond the end of the file?
1463 : *
1464 : * The page index is less than the end_index, adjust the end_offset
1465 : * to the highest offset that this page should represent.
1466 : * -----------------------------------------------------
1467 : * | file mapping | <EOF> |
1468 : * -----------------------------------------------------
1469 : * | Page ... | Page N-2 | Page N-1 | Page N | |
1470 : * ^--------------------------------^----------|--------
1471 : * | desired writeback range | see else |
1472 : * ---------------------------------^------------------|
1473 : */
1474 0 : offset = i_size_read(inode);
1475 0 : end_index = offset >> PAGE_SHIFT;
1476 0 : if (page->index < end_index)
1477 0 : end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT;
1478 : else {
1479 : /*
1480 : * Check whether the page to write out is beyond or straddles
1481 : * i_size or not.
1482 : * -------------------------------------------------------
1483 : * | file mapping | <EOF> |
1484 : * -------------------------------------------------------
1485 : * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1486 : * ^--------------------------------^-----------|---------
1487 : * | | Straddles |
1488 : * ---------------------------------^-----------|--------|
1489 : */
1490 0 : unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1491 :
1492 : /*
1493 : * Skip the page if it is fully outside i_size, e.g. due to a
1494 : * truncate operation that is in progress. We must redirty the
1495 : * page so that reclaim stops reclaiming it. Otherwise
1496 : * iomap_vm_releasepage() is called on it and gets confused.
1497 : *
1498 : * Note that the end_index is unsigned long, it would overflow
1499 : * if the given offset is greater than 16TB on 32-bit system
1500 : * and if we do check the page is fully outside i_size or not
1501 : * via "if (page->index >= end_index + 1)" as "end_index + 1"
1502 : * will be evaluated to 0. Hence this page will be redirtied
1503 : * and be written out repeatedly which would result in an
1504 : * infinite loop, the user program that perform this operation
1505 : * will hang. Instead, we can verify this situation by checking
1506 : * if the page to write is totally beyond the i_size or if it's
1507 : * offset is just equal to the EOF.
1508 : */
1509 0 : if (page->index > end_index ||
1510 0 : (page->index == end_index && offset_into_page == 0))
1511 0 : goto redirty;
1512 :
1513 : /*
1514 : * The page straddles i_size. It must be zeroed out on each
1515 : * and every writepage invocation because it may be mmapped.
1516 : * "A file is mapped in multiples of the page size. For a file
1517 : * that is not a multiple of the page size, the remaining
1518 : * memory is zeroed when mapped, and writes to that region are
1519 : * not written out to the file."
1520 : */
1521 0 : zero_user_segment(page, offset_into_page, PAGE_SIZE);
1522 :
1523 : /* Adjust the end_offset to the end of file */
1524 0 : end_offset = offset;
1525 : }
1526 :
1527 0 : return iomap_writepage_map(wpc, wbc, inode, page, end_offset);
1528 :
1529 0 : redirty:
1530 0 : redirty_page_for_writepage(wbc, page);
1531 0 : unlock_page(page);
1532 0 : return 0;
1533 : }
1534 :
1535 : int
1536 0 : iomap_writepage(struct page *page, struct writeback_control *wbc,
1537 : struct iomap_writepage_ctx *wpc,
1538 : const struct iomap_writeback_ops *ops)
1539 : {
1540 0 : int ret;
1541 :
1542 0 : wpc->ops = ops;
1543 0 : ret = iomap_do_writepage(page, wbc, wpc);
1544 0 : if (!wpc->ioend)
1545 : return ret;
1546 0 : return iomap_submit_ioend(wpc, wpc->ioend, ret);
1547 : }
1548 : EXPORT_SYMBOL_GPL(iomap_writepage);
1549 :
1550 : int
1551 0 : iomap_writepages(struct address_space *mapping, struct writeback_control *wbc,
1552 : struct iomap_writepage_ctx *wpc,
1553 : const struct iomap_writeback_ops *ops)
1554 : {
1555 0 : int ret;
1556 :
1557 0 : wpc->ops = ops;
1558 0 : ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc);
1559 0 : if (!wpc->ioend)
1560 : return ret;
1561 0 : return iomap_submit_ioend(wpc, wpc->ioend, ret);
1562 : }
1563 : EXPORT_SYMBOL_GPL(iomap_writepages);
1564 :
1565 1 : static int __init iomap_init(void)
1566 : {
1567 1 : return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE),
1568 : offsetof(struct iomap_ioend, io_inline_bio),
1569 : BIOSET_NEED_BVECS);
1570 : }
1571 : fs_initcall(iomap_init);
|
