]>
Commit | Line | Data |
---|---|---|
afc51aaa DW |
1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* | |
3 | * Copyright (C) 2010 Red Hat, Inc. | |
598ecfba | 4 | * Copyright (C) 2016-2019 Christoph Hellwig. |
afc51aaa DW |
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> | |
598ecfba | 15 | #include <linux/list_sort.h> |
afc51aaa DW |
16 | #include <linux/swap.h> |
17 | #include <linux/bio.h> | |
18 | #include <linux/sched/signal.h> | |
19 | #include <linux/migrate.h> | |
9e91c572 | 20 | #include "trace.h" |
afc51aaa DW |
21 | |
22 | #include "../internal.h" | |
23 | ||
598ecfba CH |
24 | static struct bio_set iomap_ioend_bioset; |
25 | ||
afc51aaa DW |
26 | static struct iomap_page * |
27 | iomap_page_create(struct inode *inode, struct page *page) | |
28 | { | |
29 | struct iomap_page *iop = to_iomap_page(page); | |
30 | ||
31 | if (iop || i_blocksize(inode) == PAGE_SIZE) | |
32 | return iop; | |
33 | ||
34 | iop = kmalloc(sizeof(*iop), GFP_NOFS | __GFP_NOFAIL); | |
35 | atomic_set(&iop->read_count, 0); | |
36 | atomic_set(&iop->write_count, 0); | |
37 | bitmap_zero(iop->uptodate, PAGE_SIZE / SECTOR_SIZE); | |
38 | ||
39 | /* | |
40 | * migrate_page_move_mapping() assumes that pages with private data have | |
41 | * their count elevated by 1. | |
42 | */ | |
43 | get_page(page); | |
44 | set_page_private(page, (unsigned long)iop); | |
45 | SetPagePrivate(page); | |
46 | return iop; | |
47 | } | |
48 | ||
49 | static void | |
50 | iomap_page_release(struct page *page) | |
51 | { | |
52 | struct iomap_page *iop = to_iomap_page(page); | |
53 | ||
54 | if (!iop) | |
55 | return; | |
56 | WARN_ON_ONCE(atomic_read(&iop->read_count)); | |
57 | WARN_ON_ONCE(atomic_read(&iop->write_count)); | |
58 | ClearPagePrivate(page); | |
59 | set_page_private(page, 0); | |
60 | put_page(page); | |
61 | kfree(iop); | |
62 | } | |
63 | ||
64 | /* | |
65 | * Calculate the range inside the page that we actually need to read. | |
66 | */ | |
67 | static void | |
68 | iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop, | |
69 | loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp) | |
70 | { | |
71 | loff_t orig_pos = *pos; | |
72 | loff_t isize = i_size_read(inode); | |
73 | unsigned block_bits = inode->i_blkbits; | |
74 | unsigned block_size = (1 << block_bits); | |
75 | unsigned poff = offset_in_page(*pos); | |
76 | unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length); | |
77 | unsigned first = poff >> block_bits; | |
78 | unsigned last = (poff + plen - 1) >> block_bits; | |
79 | ||
80 | /* | |
81 | * If the block size is smaller than the page size we need to check the | |
82 | * per-block uptodate status and adjust the offset and length if needed | |
83 | * to avoid reading in already uptodate ranges. | |
84 | */ | |
85 | if (iop) { | |
86 | unsigned int i; | |
87 | ||
88 | /* move forward for each leading block marked uptodate */ | |
89 | for (i = first; i <= last; i++) { | |
90 | if (!test_bit(i, iop->uptodate)) | |
91 | break; | |
92 | *pos += block_size; | |
93 | poff += block_size; | |
94 | plen -= block_size; | |
95 | first++; | |
96 | } | |
97 | ||
98 | /* truncate len if we find any trailing uptodate block(s) */ | |
99 | for ( ; i <= last; i++) { | |
100 | if (test_bit(i, iop->uptodate)) { | |
101 | plen -= (last - i + 1) * block_size; | |
102 | last = i - 1; | |
103 | break; | |
104 | } | |
105 | } | |
106 | } | |
107 | ||
108 | /* | |
109 | * If the extent spans the block that contains the i_size we need to | |
110 | * handle both halves separately so that we properly zero data in the | |
111 | * page cache for blocks that are entirely outside of i_size. | |
112 | */ | |
113 | if (orig_pos <= isize && orig_pos + length > isize) { | |
114 | unsigned end = offset_in_page(isize - 1) >> block_bits; | |
115 | ||
116 | if (first <= end && last > end) | |
117 | plen -= (last - end) * block_size; | |
118 | } | |
119 | ||
120 | *offp = poff; | |
121 | *lenp = plen; | |
122 | } | |
123 | ||
124 | static void | |
125 | iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len) | |
126 | { | |
127 | struct iomap_page *iop = to_iomap_page(page); | |
128 | struct inode *inode = page->mapping->host; | |
129 | unsigned first = off >> inode->i_blkbits; | |
130 | unsigned last = (off + len - 1) >> inode->i_blkbits; | |
131 | unsigned int i; | |
132 | bool uptodate = true; | |
133 | ||
134 | if (iop) { | |
135 | for (i = 0; i < PAGE_SIZE / i_blocksize(inode); i++) { | |
136 | if (i >= first && i <= last) | |
137 | set_bit(i, iop->uptodate); | |
138 | else if (!test_bit(i, iop->uptodate)) | |
139 | uptodate = false; | |
140 | } | |
141 | } | |
142 | ||
143 | if (uptodate && !PageError(page)) | |
144 | SetPageUptodate(page); | |
145 | } | |
146 | ||
147 | static void | |
148 | iomap_read_finish(struct iomap_page *iop, struct page *page) | |
149 | { | |
150 | if (!iop || atomic_dec_and_test(&iop->read_count)) | |
151 | unlock_page(page); | |
152 | } | |
153 | ||
154 | static void | |
155 | iomap_read_page_end_io(struct bio_vec *bvec, int error) | |
156 | { | |
157 | struct page *page = bvec->bv_page; | |
158 | struct iomap_page *iop = to_iomap_page(page); | |
159 | ||
160 | if (unlikely(error)) { | |
161 | ClearPageUptodate(page); | |
162 | SetPageError(page); | |
163 | } else { | |
164 | iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len); | |
165 | } | |
166 | ||
167 | iomap_read_finish(iop, page); | |
168 | } | |
169 | ||
170 | static void | |
171 | iomap_read_end_io(struct bio *bio) | |
172 | { | |
173 | int error = blk_status_to_errno(bio->bi_status); | |
174 | struct bio_vec *bvec; | |
175 | struct bvec_iter_all iter_all; | |
176 | ||
177 | bio_for_each_segment_all(bvec, bio, iter_all) | |
178 | iomap_read_page_end_io(bvec, error); | |
179 | bio_put(bio); | |
180 | } | |
181 | ||
182 | struct iomap_readpage_ctx { | |
183 | struct page *cur_page; | |
184 | bool cur_page_in_bio; | |
185 | bool is_readahead; | |
186 | struct bio *bio; | |
187 | struct list_head *pages; | |
188 | }; | |
189 | ||
190 | static void | |
191 | iomap_read_inline_data(struct inode *inode, struct page *page, | |
192 | struct iomap *iomap) | |
193 | { | |
194 | size_t size = i_size_read(inode); | |
195 | void *addr; | |
196 | ||
197 | if (PageUptodate(page)) | |
198 | return; | |
199 | ||
200 | BUG_ON(page->index); | |
201 | BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data)); | |
202 | ||
203 | addr = kmap_atomic(page); | |
204 | memcpy(addr, iomap->inline_data, size); | |
205 | memset(addr + size, 0, PAGE_SIZE - size); | |
206 | kunmap_atomic(addr); | |
207 | SetPageUptodate(page); | |
208 | } | |
209 | ||
009d8d84 CH |
210 | static inline bool iomap_block_needs_zeroing(struct inode *inode, |
211 | struct iomap *iomap, loff_t pos) | |
212 | { | |
213 | return iomap->type != IOMAP_MAPPED || | |
214 | (iomap->flags & IOMAP_F_NEW) || | |
215 | pos >= i_size_read(inode); | |
216 | } | |
217 | ||
afc51aaa DW |
218 | static loff_t |
219 | iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
220 | struct iomap *iomap) | |
221 | { | |
222 | struct iomap_readpage_ctx *ctx = data; | |
223 | struct page *page = ctx->cur_page; | |
224 | struct iomap_page *iop = iomap_page_create(inode, page); | |
225 | bool same_page = false, is_contig = false; | |
226 | loff_t orig_pos = pos; | |
227 | unsigned poff, plen; | |
228 | sector_t sector; | |
229 | ||
230 | if (iomap->type == IOMAP_INLINE) { | |
231 | WARN_ON_ONCE(pos); | |
232 | iomap_read_inline_data(inode, page, iomap); | |
233 | return PAGE_SIZE; | |
234 | } | |
235 | ||
236 | /* zero post-eof blocks as the page may be mapped */ | |
237 | iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen); | |
238 | if (plen == 0) | |
239 | goto done; | |
240 | ||
009d8d84 | 241 | if (iomap_block_needs_zeroing(inode, iomap, pos)) { |
afc51aaa DW |
242 | zero_user(page, poff, plen); |
243 | iomap_set_range_uptodate(page, poff, plen); | |
244 | goto done; | |
245 | } | |
246 | ||
247 | ctx->cur_page_in_bio = true; | |
248 | ||
249 | /* | |
250 | * Try to merge into a previous segment if we can. | |
251 | */ | |
252 | sector = iomap_sector(iomap, pos); | |
253 | if (ctx->bio && bio_end_sector(ctx->bio) == sector) | |
254 | is_contig = true; | |
255 | ||
256 | if (is_contig && | |
257 | __bio_try_merge_page(ctx->bio, page, plen, poff, &same_page)) { | |
258 | if (!same_page && iop) | |
259 | atomic_inc(&iop->read_count); | |
260 | goto done; | |
261 | } | |
262 | ||
263 | /* | |
264 | * If we start a new segment we need to increase the read count, and we | |
265 | * need to do so before submitting any previous full bio to make sure | |
266 | * that we don't prematurely unlock the page. | |
267 | */ | |
268 | if (iop) | |
269 | atomic_inc(&iop->read_count); | |
270 | ||
271 | if (!ctx->bio || !is_contig || bio_full(ctx->bio, plen)) { | |
272 | gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); | |
273 | int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
274 | ||
275 | if (ctx->bio) | |
276 | submit_bio(ctx->bio); | |
277 | ||
278 | if (ctx->is_readahead) /* same as readahead_gfp_mask */ | |
279 | gfp |= __GFP_NORETRY | __GFP_NOWARN; | |
280 | ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs)); | |
281 | ctx->bio->bi_opf = REQ_OP_READ; | |
282 | if (ctx->is_readahead) | |
283 | ctx->bio->bi_opf |= REQ_RAHEAD; | |
284 | ctx->bio->bi_iter.bi_sector = sector; | |
285 | bio_set_dev(ctx->bio, iomap->bdev); | |
286 | ctx->bio->bi_end_io = iomap_read_end_io; | |
287 | } | |
288 | ||
289 | bio_add_page(ctx->bio, page, plen, poff); | |
290 | done: | |
291 | /* | |
292 | * Move the caller beyond our range so that it keeps making progress. | |
293 | * For that we have to include any leading non-uptodate ranges, but | |
294 | * we can skip trailing ones as they will be handled in the next | |
295 | * iteration. | |
296 | */ | |
297 | return pos - orig_pos + plen; | |
298 | } | |
299 | ||
300 | int | |
301 | iomap_readpage(struct page *page, const struct iomap_ops *ops) | |
302 | { | |
303 | struct iomap_readpage_ctx ctx = { .cur_page = page }; | |
304 | struct inode *inode = page->mapping->host; | |
305 | unsigned poff; | |
306 | loff_t ret; | |
307 | ||
9e91c572 CH |
308 | trace_iomap_readpage(page->mapping->host, 1); |
309 | ||
afc51aaa DW |
310 | for (poff = 0; poff < PAGE_SIZE; poff += ret) { |
311 | ret = iomap_apply(inode, page_offset(page) + poff, | |
312 | PAGE_SIZE - poff, 0, ops, &ctx, | |
313 | iomap_readpage_actor); | |
314 | if (ret <= 0) { | |
315 | WARN_ON_ONCE(ret == 0); | |
316 | SetPageError(page); | |
317 | break; | |
318 | } | |
319 | } | |
320 | ||
321 | if (ctx.bio) { | |
322 | submit_bio(ctx.bio); | |
323 | WARN_ON_ONCE(!ctx.cur_page_in_bio); | |
324 | } else { | |
325 | WARN_ON_ONCE(ctx.cur_page_in_bio); | |
326 | unlock_page(page); | |
327 | } | |
328 | ||
329 | /* | |
330 | * Just like mpage_readpages and block_read_full_page we always | |
331 | * return 0 and just mark the page as PageError on errors. This | |
332 | * should be cleaned up all through the stack eventually. | |
333 | */ | |
334 | return 0; | |
335 | } | |
336 | EXPORT_SYMBOL_GPL(iomap_readpage); | |
337 | ||
338 | static struct page * | |
339 | iomap_next_page(struct inode *inode, struct list_head *pages, loff_t pos, | |
340 | loff_t length, loff_t *done) | |
341 | { | |
342 | while (!list_empty(pages)) { | |
343 | struct page *page = lru_to_page(pages); | |
344 | ||
345 | if (page_offset(page) >= (u64)pos + length) | |
346 | break; | |
347 | ||
348 | list_del(&page->lru); | |
349 | if (!add_to_page_cache_lru(page, inode->i_mapping, page->index, | |
350 | GFP_NOFS)) | |
351 | return page; | |
352 | ||
353 | /* | |
354 | * If we already have a page in the page cache at index we are | |
355 | * done. Upper layers don't care if it is uptodate after the | |
356 | * readpages call itself as every page gets checked again once | |
357 | * actually needed. | |
358 | */ | |
359 | *done += PAGE_SIZE; | |
360 | put_page(page); | |
361 | } | |
362 | ||
363 | return NULL; | |
364 | } | |
365 | ||
366 | static loff_t | |
367 | iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length, | |
368 | void *data, struct iomap *iomap) | |
369 | { | |
370 | struct iomap_readpage_ctx *ctx = data; | |
371 | loff_t done, ret; | |
372 | ||
373 | for (done = 0; done < length; done += ret) { | |
374 | if (ctx->cur_page && offset_in_page(pos + done) == 0) { | |
375 | if (!ctx->cur_page_in_bio) | |
376 | unlock_page(ctx->cur_page); | |
377 | put_page(ctx->cur_page); | |
378 | ctx->cur_page = NULL; | |
379 | } | |
380 | if (!ctx->cur_page) { | |
381 | ctx->cur_page = iomap_next_page(inode, ctx->pages, | |
382 | pos, length, &done); | |
383 | if (!ctx->cur_page) | |
384 | break; | |
385 | ctx->cur_page_in_bio = false; | |
386 | } | |
387 | ret = iomap_readpage_actor(inode, pos + done, length - done, | |
388 | ctx, iomap); | |
389 | } | |
390 | ||
391 | return done; | |
392 | } | |
393 | ||
394 | int | |
395 | iomap_readpages(struct address_space *mapping, struct list_head *pages, | |
396 | unsigned nr_pages, const struct iomap_ops *ops) | |
397 | { | |
398 | struct iomap_readpage_ctx ctx = { | |
399 | .pages = pages, | |
400 | .is_readahead = true, | |
401 | }; | |
402 | loff_t pos = page_offset(list_entry(pages->prev, struct page, lru)); | |
403 | loff_t last = page_offset(list_entry(pages->next, struct page, lru)); | |
404 | loff_t length = last - pos + PAGE_SIZE, ret = 0; | |
405 | ||
9e91c572 CH |
406 | trace_iomap_readpages(mapping->host, nr_pages); |
407 | ||
afc51aaa DW |
408 | while (length > 0) { |
409 | ret = iomap_apply(mapping->host, pos, length, 0, ops, | |
410 | &ctx, iomap_readpages_actor); | |
411 | if (ret <= 0) { | |
412 | WARN_ON_ONCE(ret == 0); | |
413 | goto done; | |
414 | } | |
415 | pos += ret; | |
416 | length -= ret; | |
417 | } | |
418 | ret = 0; | |
419 | done: | |
420 | if (ctx.bio) | |
421 | submit_bio(ctx.bio); | |
422 | if (ctx.cur_page) { | |
423 | if (!ctx.cur_page_in_bio) | |
424 | unlock_page(ctx.cur_page); | |
425 | put_page(ctx.cur_page); | |
426 | } | |
427 | ||
428 | /* | |
429 | * Check that we didn't lose a page due to the arcance calling | |
430 | * conventions.. | |
431 | */ | |
432 | WARN_ON_ONCE(!ret && !list_empty(ctx.pages)); | |
433 | return ret; | |
434 | } | |
435 | EXPORT_SYMBOL_GPL(iomap_readpages); | |
436 | ||
437 | /* | |
438 | * iomap_is_partially_uptodate checks whether blocks within a page are | |
439 | * uptodate or not. | |
440 | * | |
441 | * Returns true if all blocks which correspond to a file portion | |
442 | * we want to read within the page are uptodate. | |
443 | */ | |
444 | int | |
445 | iomap_is_partially_uptodate(struct page *page, unsigned long from, | |
446 | unsigned long count) | |
447 | { | |
448 | struct iomap_page *iop = to_iomap_page(page); | |
449 | struct inode *inode = page->mapping->host; | |
450 | unsigned len, first, last; | |
451 | unsigned i; | |
452 | ||
453 | /* Limit range to one page */ | |
454 | len = min_t(unsigned, PAGE_SIZE - from, count); | |
455 | ||
456 | /* First and last blocks in range within page */ | |
457 | first = from >> inode->i_blkbits; | |
458 | last = (from + len - 1) >> inode->i_blkbits; | |
459 | ||
460 | if (iop) { | |
461 | for (i = first; i <= last; i++) | |
462 | if (!test_bit(i, iop->uptodate)) | |
463 | return 0; | |
464 | return 1; | |
465 | } | |
466 | ||
467 | return 0; | |
468 | } | |
469 | EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); | |
470 | ||
471 | int | |
472 | iomap_releasepage(struct page *page, gfp_t gfp_mask) | |
473 | { | |
9e91c572 CH |
474 | trace_iomap_releasepage(page->mapping->host, page, 0, 0); |
475 | ||
afc51aaa DW |
476 | /* |
477 | * mm accommodates an old ext3 case where clean pages might not have had | |
478 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
479 | * ->releasepage() via shrink_active_list(), skip those here. | |
480 | */ | |
481 | if (PageDirty(page) || PageWriteback(page)) | |
482 | return 0; | |
483 | iomap_page_release(page); | |
484 | return 1; | |
485 | } | |
486 | EXPORT_SYMBOL_GPL(iomap_releasepage); | |
487 | ||
488 | void | |
489 | iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len) | |
490 | { | |
9e91c572 CH |
491 | trace_iomap_invalidatepage(page->mapping->host, page, offset, len); |
492 | ||
afc51aaa DW |
493 | /* |
494 | * If we are invalidating the entire page, clear the dirty state from it | |
495 | * and release it to avoid unnecessary buildup of the LRU. | |
496 | */ | |
497 | if (offset == 0 && len == PAGE_SIZE) { | |
498 | WARN_ON_ONCE(PageWriteback(page)); | |
499 | cancel_dirty_page(page); | |
500 | iomap_page_release(page); | |
501 | } | |
502 | } | |
503 | EXPORT_SYMBOL_GPL(iomap_invalidatepage); | |
504 | ||
505 | #ifdef CONFIG_MIGRATION | |
506 | int | |
507 | iomap_migrate_page(struct address_space *mapping, struct page *newpage, | |
508 | struct page *page, enum migrate_mode mode) | |
509 | { | |
510 | int ret; | |
511 | ||
26473f83 | 512 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); |
afc51aaa DW |
513 | if (ret != MIGRATEPAGE_SUCCESS) |
514 | return ret; | |
515 | ||
516 | if (page_has_private(page)) { | |
517 | ClearPagePrivate(page); | |
518 | get_page(newpage); | |
519 | set_page_private(newpage, page_private(page)); | |
520 | set_page_private(page, 0); | |
521 | put_page(page); | |
522 | SetPagePrivate(newpage); | |
523 | } | |
524 | ||
525 | if (mode != MIGRATE_SYNC_NO_COPY) | |
526 | migrate_page_copy(newpage, page); | |
527 | else | |
528 | migrate_page_states(newpage, page); | |
529 | return MIGRATEPAGE_SUCCESS; | |
530 | } | |
531 | EXPORT_SYMBOL_GPL(iomap_migrate_page); | |
532 | #endif /* CONFIG_MIGRATION */ | |
533 | ||
534 | static void | |
535 | iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) | |
536 | { | |
537 | loff_t i_size = i_size_read(inode); | |
538 | ||
539 | /* | |
540 | * Only truncate newly allocated pages beyoned EOF, even if the | |
541 | * write started inside the existing inode size. | |
542 | */ | |
543 | if (pos + len > i_size) | |
544 | truncate_pagecache_range(inode, max(pos, i_size), pos + len); | |
545 | } | |
546 | ||
547 | static int | |
548 | iomap_read_page_sync(struct inode *inode, loff_t block_start, struct page *page, | |
549 | unsigned poff, unsigned plen, unsigned from, unsigned to, | |
550 | struct iomap *iomap) | |
551 | { | |
552 | struct bio_vec bvec; | |
553 | struct bio bio; | |
554 | ||
009d8d84 | 555 | if (iomap_block_needs_zeroing(inode, iomap, block_start)) { |
afc51aaa DW |
556 | zero_user_segments(page, poff, from, to, poff + plen); |
557 | iomap_set_range_uptodate(page, poff, plen); | |
558 | return 0; | |
559 | } | |
560 | ||
561 | bio_init(&bio, &bvec, 1); | |
562 | bio.bi_opf = REQ_OP_READ; | |
563 | bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); | |
564 | bio_set_dev(&bio, iomap->bdev); | |
565 | __bio_add_page(&bio, page, plen, poff); | |
566 | return submit_bio_wait(&bio); | |
567 | } | |
568 | ||
569 | static int | |
570 | __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, | |
571 | struct page *page, struct iomap *iomap) | |
572 | { | |
573 | struct iomap_page *iop = iomap_page_create(inode, page); | |
574 | loff_t block_size = i_blocksize(inode); | |
575 | loff_t block_start = pos & ~(block_size - 1); | |
576 | loff_t block_end = (pos + len + block_size - 1) & ~(block_size - 1); | |
577 | unsigned from = offset_in_page(pos), to = from + len, poff, plen; | |
578 | int status = 0; | |
579 | ||
580 | if (PageUptodate(page)) | |
581 | return 0; | |
582 | ||
583 | do { | |
584 | iomap_adjust_read_range(inode, iop, &block_start, | |
585 | block_end - block_start, &poff, &plen); | |
586 | if (plen == 0) | |
587 | break; | |
588 | ||
589 | if ((from > poff && from < poff + plen) || | |
590 | (to > poff && to < poff + plen)) { | |
591 | status = iomap_read_page_sync(inode, block_start, page, | |
592 | poff, plen, from, to, iomap); | |
593 | if (status) | |
594 | break; | |
595 | } | |
596 | ||
597 | } while ((block_start += plen) < block_end); | |
598 | ||
599 | return status; | |
600 | } | |
601 | ||
602 | static int | |
603 | iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, | |
604 | struct page **pagep, struct iomap *iomap) | |
605 | { | |
606 | const struct iomap_page_ops *page_ops = iomap->page_ops; | |
607 | pgoff_t index = pos >> PAGE_SHIFT; | |
608 | struct page *page; | |
609 | int status = 0; | |
610 | ||
611 | BUG_ON(pos + len > iomap->offset + iomap->length); | |
612 | ||
613 | if (fatal_signal_pending(current)) | |
614 | return -EINTR; | |
615 | ||
616 | if (page_ops && page_ops->page_prepare) { | |
617 | status = page_ops->page_prepare(inode, pos, len, iomap); | |
618 | if (status) | |
619 | return status; | |
620 | } | |
621 | ||
622 | page = grab_cache_page_write_begin(inode->i_mapping, index, flags); | |
623 | if (!page) { | |
624 | status = -ENOMEM; | |
625 | goto out_no_page; | |
626 | } | |
627 | ||
628 | if (iomap->type == IOMAP_INLINE) | |
629 | iomap_read_inline_data(inode, page, iomap); | |
630 | else if (iomap->flags & IOMAP_F_BUFFER_HEAD) | |
631 | status = __block_write_begin_int(page, pos, len, NULL, iomap); | |
632 | else | |
633 | status = __iomap_write_begin(inode, pos, len, page, iomap); | |
634 | ||
635 | if (unlikely(status)) | |
636 | goto out_unlock; | |
637 | ||
638 | *pagep = page; | |
639 | return 0; | |
640 | ||
641 | out_unlock: | |
642 | unlock_page(page); | |
643 | put_page(page); | |
644 | iomap_write_failed(inode, pos, len); | |
645 | ||
646 | out_no_page: | |
647 | if (page_ops && page_ops->page_done) | |
648 | page_ops->page_done(inode, pos, 0, NULL, iomap); | |
649 | return status; | |
650 | } | |
651 | ||
652 | int | |
653 | iomap_set_page_dirty(struct page *page) | |
654 | { | |
655 | struct address_space *mapping = page_mapping(page); | |
656 | int newly_dirty; | |
657 | ||
658 | if (unlikely(!mapping)) | |
659 | return !TestSetPageDirty(page); | |
660 | ||
661 | /* | |
662 | * Lock out page->mem_cgroup migration to keep PageDirty | |
663 | * synchronized with per-memcg dirty page counters. | |
664 | */ | |
665 | lock_page_memcg(page); | |
666 | newly_dirty = !TestSetPageDirty(page); | |
667 | if (newly_dirty) | |
668 | __set_page_dirty(page, mapping, 0); | |
669 | unlock_page_memcg(page); | |
670 | ||
671 | if (newly_dirty) | |
672 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
673 | return newly_dirty; | |
674 | } | |
675 | EXPORT_SYMBOL_GPL(iomap_set_page_dirty); | |
676 | ||
677 | static int | |
678 | __iomap_write_end(struct inode *inode, loff_t pos, unsigned len, | |
679 | unsigned copied, struct page *page, struct iomap *iomap) | |
680 | { | |
681 | flush_dcache_page(page); | |
682 | ||
683 | /* | |
684 | * The blocks that were entirely written will now be uptodate, so we | |
685 | * don't have to worry about a readpage reading them and overwriting a | |
686 | * partial write. However if we have encountered a short write and only | |
687 | * partially written into a block, it will not be marked uptodate, so a | |
688 | * readpage might come in and destroy our partial write. | |
689 | * | |
690 | * Do the simplest thing, and just treat any short write to a non | |
691 | * uptodate page as a zero-length write, and force the caller to redo | |
692 | * the whole thing. | |
693 | */ | |
694 | if (unlikely(copied < len && !PageUptodate(page))) | |
695 | return 0; | |
696 | iomap_set_range_uptodate(page, offset_in_page(pos), len); | |
697 | iomap_set_page_dirty(page); | |
698 | return copied; | |
699 | } | |
700 | ||
701 | static int | |
702 | iomap_write_end_inline(struct inode *inode, struct page *page, | |
703 | struct iomap *iomap, loff_t pos, unsigned copied) | |
704 | { | |
705 | void *addr; | |
706 | ||
707 | WARN_ON_ONCE(!PageUptodate(page)); | |
708 | BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data)); | |
709 | ||
710 | addr = kmap_atomic(page); | |
711 | memcpy(iomap->inline_data + pos, addr + pos, copied); | |
712 | kunmap_atomic(addr); | |
713 | ||
714 | mark_inode_dirty(inode); | |
715 | return copied; | |
716 | } | |
717 | ||
718 | static int | |
719 | iomap_write_end(struct inode *inode, loff_t pos, unsigned len, | |
720 | unsigned copied, struct page *page, struct iomap *iomap) | |
721 | { | |
722 | const struct iomap_page_ops *page_ops = iomap->page_ops; | |
723 | loff_t old_size = inode->i_size; | |
724 | int ret; | |
725 | ||
726 | if (iomap->type == IOMAP_INLINE) { | |
727 | ret = iomap_write_end_inline(inode, page, iomap, pos, copied); | |
728 | } else if (iomap->flags & IOMAP_F_BUFFER_HEAD) { | |
729 | ret = block_write_end(NULL, inode->i_mapping, pos, len, copied, | |
730 | page, NULL); | |
731 | } else { | |
732 | ret = __iomap_write_end(inode, pos, len, copied, page, iomap); | |
733 | } | |
734 | ||
735 | /* | |
736 | * Update the in-memory inode size after copying the data into the page | |
737 | * cache. It's up to the file system to write the updated size to disk, | |
738 | * preferably after I/O completion so that no stale data is exposed. | |
739 | */ | |
740 | if (pos + ret > old_size) { | |
741 | i_size_write(inode, pos + ret); | |
742 | iomap->flags |= IOMAP_F_SIZE_CHANGED; | |
743 | } | |
744 | unlock_page(page); | |
745 | ||
746 | if (old_size < pos) | |
747 | pagecache_isize_extended(inode, old_size, pos); | |
748 | if (page_ops && page_ops->page_done) | |
749 | page_ops->page_done(inode, pos, ret, page, iomap); | |
750 | put_page(page); | |
751 | ||
752 | if (ret < len) | |
753 | iomap_write_failed(inode, pos, len); | |
754 | return ret; | |
755 | } | |
756 | ||
757 | static loff_t | |
758 | iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
759 | struct iomap *iomap) | |
760 | { | |
761 | struct iov_iter *i = data; | |
762 | long status = 0; | |
763 | ssize_t written = 0; | |
764 | unsigned int flags = AOP_FLAG_NOFS; | |
765 | ||
766 | do { | |
767 | struct page *page; | |
768 | unsigned long offset; /* Offset into pagecache page */ | |
769 | unsigned long bytes; /* Bytes to write to page */ | |
770 | size_t copied; /* Bytes copied from user */ | |
771 | ||
772 | offset = offset_in_page(pos); | |
773 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
774 | iov_iter_count(i)); | |
775 | again: | |
776 | if (bytes > length) | |
777 | bytes = length; | |
778 | ||
779 | /* | |
780 | * Bring in the user page that we will copy from _first_. | |
781 | * Otherwise there's a nasty deadlock on copying from the | |
782 | * same page as we're writing to, without it being marked | |
783 | * up-to-date. | |
784 | * | |
785 | * Not only is this an optimisation, but it is also required | |
786 | * to check that the address is actually valid, when atomic | |
787 | * usercopies are used, below. | |
788 | */ | |
789 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
790 | status = -EFAULT; | |
791 | break; | |
792 | } | |
793 | ||
794 | status = iomap_write_begin(inode, pos, bytes, flags, &page, | |
795 | iomap); | |
796 | if (unlikely(status)) | |
797 | break; | |
798 | ||
799 | if (mapping_writably_mapped(inode->i_mapping)) | |
800 | flush_dcache_page(page); | |
801 | ||
802 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
803 | ||
804 | flush_dcache_page(page); | |
805 | ||
806 | status = iomap_write_end(inode, pos, bytes, copied, page, | |
807 | iomap); | |
808 | if (unlikely(status < 0)) | |
809 | break; | |
810 | copied = status; | |
811 | ||
812 | cond_resched(); | |
813 | ||
814 | iov_iter_advance(i, copied); | |
815 | if (unlikely(copied == 0)) { | |
816 | /* | |
817 | * If we were unable to copy any data at all, we must | |
818 | * fall back to a single segment length write. | |
819 | * | |
820 | * If we didn't fallback here, we could livelock | |
821 | * because not all segments in the iov can be copied at | |
822 | * once without a pagefault. | |
823 | */ | |
824 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
825 | iov_iter_single_seg_count(i)); | |
826 | goto again; | |
827 | } | |
828 | pos += copied; | |
829 | written += copied; | |
830 | length -= copied; | |
831 | ||
832 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
833 | } while (iov_iter_count(i) && length); | |
834 | ||
835 | return written ? written : status; | |
836 | } | |
837 | ||
838 | ssize_t | |
839 | iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, | |
840 | const struct iomap_ops *ops) | |
841 | { | |
842 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
843 | loff_t pos = iocb->ki_pos, ret = 0, written = 0; | |
844 | ||
845 | while (iov_iter_count(iter)) { | |
846 | ret = iomap_apply(inode, pos, iov_iter_count(iter), | |
847 | IOMAP_WRITE, ops, iter, iomap_write_actor); | |
848 | if (ret <= 0) | |
849 | break; | |
850 | pos += ret; | |
851 | written += ret; | |
852 | } | |
853 | ||
854 | return written ? written : ret; | |
855 | } | |
856 | EXPORT_SYMBOL_GPL(iomap_file_buffered_write); | |
857 | ||
858 | static struct page * | |
859 | __iomap_read_page(struct inode *inode, loff_t offset) | |
860 | { | |
861 | struct address_space *mapping = inode->i_mapping; | |
862 | struct page *page; | |
863 | ||
864 | page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL); | |
865 | if (IS_ERR(page)) | |
866 | return page; | |
867 | if (!PageUptodate(page)) { | |
868 | put_page(page); | |
869 | return ERR_PTR(-EIO); | |
870 | } | |
871 | return page; | |
872 | } | |
873 | ||
874 | static loff_t | |
875 | iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
876 | struct iomap *iomap) | |
877 | { | |
878 | long status = 0; | |
879 | ssize_t written = 0; | |
880 | ||
881 | do { | |
882 | struct page *page, *rpage; | |
883 | unsigned long offset; /* Offset into pagecache page */ | |
884 | unsigned long bytes; /* Bytes to write to page */ | |
885 | ||
886 | offset = offset_in_page(pos); | |
887 | bytes = min_t(loff_t, PAGE_SIZE - offset, length); | |
888 | ||
889 | rpage = __iomap_read_page(inode, pos); | |
890 | if (IS_ERR(rpage)) | |
891 | return PTR_ERR(rpage); | |
892 | ||
893 | status = iomap_write_begin(inode, pos, bytes, | |
894 | AOP_FLAG_NOFS, &page, iomap); | |
895 | put_page(rpage); | |
896 | if (unlikely(status)) | |
897 | return status; | |
898 | ||
899 | WARN_ON_ONCE(!PageUptodate(page)); | |
900 | ||
901 | status = iomap_write_end(inode, pos, bytes, bytes, page, iomap); | |
902 | if (unlikely(status <= 0)) { | |
903 | if (WARN_ON_ONCE(status == 0)) | |
904 | return -EIO; | |
905 | return status; | |
906 | } | |
907 | ||
908 | cond_resched(); | |
909 | ||
910 | pos += status; | |
911 | written += status; | |
912 | length -= status; | |
913 | ||
914 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
915 | } while (length); | |
916 | ||
917 | return written; | |
918 | } | |
919 | ||
920 | int | |
921 | iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len, | |
922 | const struct iomap_ops *ops) | |
923 | { | |
924 | loff_t ret; | |
925 | ||
926 | while (len) { | |
927 | ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, | |
928 | iomap_dirty_actor); | |
929 | if (ret <= 0) | |
930 | return ret; | |
931 | pos += ret; | |
932 | len -= ret; | |
933 | } | |
934 | ||
935 | return 0; | |
936 | } | |
937 | EXPORT_SYMBOL_GPL(iomap_file_dirty); | |
938 | ||
939 | static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset, | |
940 | unsigned bytes, struct iomap *iomap) | |
941 | { | |
942 | struct page *page; | |
943 | int status; | |
944 | ||
945 | status = iomap_write_begin(inode, pos, bytes, AOP_FLAG_NOFS, &page, | |
946 | iomap); | |
947 | if (status) | |
948 | return status; | |
949 | ||
950 | zero_user(page, offset, bytes); | |
951 | mark_page_accessed(page); | |
952 | ||
953 | return iomap_write_end(inode, pos, bytes, bytes, page, iomap); | |
954 | } | |
955 | ||
956 | static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes, | |
957 | struct iomap *iomap) | |
958 | { | |
959 | return __dax_zero_page_range(iomap->bdev, iomap->dax_dev, | |
960 | iomap_sector(iomap, pos & PAGE_MASK), offset, bytes); | |
961 | } | |
962 | ||
963 | static loff_t | |
964 | iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count, | |
965 | void *data, struct iomap *iomap) | |
966 | { | |
967 | bool *did_zero = data; | |
968 | loff_t written = 0; | |
969 | int status; | |
970 | ||
971 | /* already zeroed? we're done. */ | |
972 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) | |
973 | return count; | |
974 | ||
975 | do { | |
976 | unsigned offset, bytes; | |
977 | ||
978 | offset = offset_in_page(pos); | |
979 | bytes = min_t(loff_t, PAGE_SIZE - offset, count); | |
980 | ||
981 | if (IS_DAX(inode)) | |
982 | status = iomap_dax_zero(pos, offset, bytes, iomap); | |
983 | else | |
984 | status = iomap_zero(inode, pos, offset, bytes, iomap); | |
985 | if (status < 0) | |
986 | return status; | |
987 | ||
988 | pos += bytes; | |
989 | count -= bytes; | |
990 | written += bytes; | |
991 | if (did_zero) | |
992 | *did_zero = true; | |
993 | } while (count > 0); | |
994 | ||
995 | return written; | |
996 | } | |
997 | ||
998 | int | |
999 | iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, | |
1000 | const struct iomap_ops *ops) | |
1001 | { | |
1002 | loff_t ret; | |
1003 | ||
1004 | while (len > 0) { | |
1005 | ret = iomap_apply(inode, pos, len, IOMAP_ZERO, | |
1006 | ops, did_zero, iomap_zero_range_actor); | |
1007 | if (ret <= 0) | |
1008 | return ret; | |
1009 | ||
1010 | pos += ret; | |
1011 | len -= ret; | |
1012 | } | |
1013 | ||
1014 | return 0; | |
1015 | } | |
1016 | EXPORT_SYMBOL_GPL(iomap_zero_range); | |
1017 | ||
1018 | int | |
1019 | iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, | |
1020 | const struct iomap_ops *ops) | |
1021 | { | |
1022 | unsigned int blocksize = i_blocksize(inode); | |
1023 | unsigned int off = pos & (blocksize - 1); | |
1024 | ||
1025 | /* Block boundary? Nothing to do */ | |
1026 | if (!off) | |
1027 | return 0; | |
1028 | return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); | |
1029 | } | |
1030 | EXPORT_SYMBOL_GPL(iomap_truncate_page); | |
1031 | ||
1032 | static loff_t | |
1033 | iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, | |
1034 | void *data, struct iomap *iomap) | |
1035 | { | |
1036 | struct page *page = data; | |
1037 | int ret; | |
1038 | ||
1039 | if (iomap->flags & IOMAP_F_BUFFER_HEAD) { | |
1040 | ret = __block_write_begin_int(page, pos, length, NULL, iomap); | |
1041 | if (ret) | |
1042 | return ret; | |
1043 | block_commit_write(page, 0, length); | |
1044 | } else { | |
1045 | WARN_ON_ONCE(!PageUptodate(page)); | |
1046 | iomap_page_create(inode, page); | |
1047 | set_page_dirty(page); | |
1048 | } | |
1049 | ||
1050 | return length; | |
1051 | } | |
1052 | ||
1053 | vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) | |
1054 | { | |
1055 | struct page *page = vmf->page; | |
1056 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
1057 | unsigned long length; | |
1058 | loff_t offset, size; | |
1059 | ssize_t ret; | |
1060 | ||
1061 | lock_page(page); | |
1062 | size = i_size_read(inode); | |
1063 | if ((page->mapping != inode->i_mapping) || | |
1064 | (page_offset(page) > size)) { | |
1065 | /* We overload EFAULT to mean page got truncated */ | |
1066 | ret = -EFAULT; | |
1067 | goto out_unlock; | |
1068 | } | |
1069 | ||
1070 | /* page is wholly or partially inside EOF */ | |
1071 | if (((page->index + 1) << PAGE_SHIFT) > size) | |
1072 | length = offset_in_page(size); | |
1073 | else | |
1074 | length = PAGE_SIZE; | |
1075 | ||
1076 | offset = page_offset(page); | |
1077 | while (length > 0) { | |
1078 | ret = iomap_apply(inode, offset, length, | |
1079 | IOMAP_WRITE | IOMAP_FAULT, ops, page, | |
1080 | iomap_page_mkwrite_actor); | |
1081 | if (unlikely(ret <= 0)) | |
1082 | goto out_unlock; | |
1083 | offset += ret; | |
1084 | length -= ret; | |
1085 | } | |
1086 | ||
1087 | wait_for_stable_page(page); | |
1088 | return VM_FAULT_LOCKED; | |
1089 | out_unlock: | |
1090 | unlock_page(page); | |
1091 | return block_page_mkwrite_return(ret); | |
1092 | } | |
1093 | EXPORT_SYMBOL_GPL(iomap_page_mkwrite); | |
598ecfba CH |
1094 | |
1095 | static void | |
1096 | iomap_finish_page_writeback(struct inode *inode, struct bio_vec *bvec, | |
1097 | int error) | |
1098 | { | |
1099 | struct iomap_page *iop = to_iomap_page(bvec->bv_page); | |
1100 | ||
1101 | if (error) { | |
1102 | SetPageError(bvec->bv_page); | |
1103 | mapping_set_error(inode->i_mapping, -EIO); | |
1104 | } | |
1105 | ||
1106 | WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); | |
1107 | WARN_ON_ONCE(iop && atomic_read(&iop->write_count) <= 0); | |
1108 | ||
1109 | if (!iop || atomic_dec_and_test(&iop->write_count)) | |
1110 | end_page_writeback(bvec->bv_page); | |
1111 | } | |
1112 | ||
1113 | /* | |
1114 | * We're now finished for good with this ioend structure. Update the page | |
1115 | * state, release holds on bios, and finally free up memory. Do not use the | |
1116 | * ioend after this. | |
1117 | */ | |
1118 | static void | |
1119 | iomap_finish_ioend(struct iomap_ioend *ioend, int error) | |
1120 | { | |
1121 | struct inode *inode = ioend->io_inode; | |
1122 | struct bio *bio = &ioend->io_inline_bio; | |
1123 | struct bio *last = ioend->io_bio, *next; | |
1124 | u64 start = bio->bi_iter.bi_sector; | |
1125 | bool quiet = bio_flagged(bio, BIO_QUIET); | |
1126 | ||
1127 | for (bio = &ioend->io_inline_bio; bio; bio = next) { | |
1128 | struct bio_vec *bv; | |
1129 | struct bvec_iter_all iter_all; | |
1130 | ||
1131 | /* | |
1132 | * For the last bio, bi_private points to the ioend, so we | |
1133 | * need to explicitly end the iteration here. | |
1134 | */ | |
1135 | if (bio == last) | |
1136 | next = NULL; | |
1137 | else | |
1138 | next = bio->bi_private; | |
1139 | ||
1140 | /* walk each page on bio, ending page IO on them */ | |
1141 | bio_for_each_segment_all(bv, bio, iter_all) | |
1142 | iomap_finish_page_writeback(inode, bv, error); | |
1143 | bio_put(bio); | |
1144 | } | |
1145 | ||
1146 | if (unlikely(error && !quiet)) { | |
1147 | printk_ratelimited(KERN_ERR | |
1148 | "%s: writeback error on sector %llu", | |
1149 | inode->i_sb->s_id, start); | |
1150 | } | |
1151 | } | |
1152 | ||
1153 | void | |
1154 | iomap_finish_ioends(struct iomap_ioend *ioend, int error) | |
1155 | { | |
1156 | struct list_head tmp; | |
1157 | ||
1158 | list_replace_init(&ioend->io_list, &tmp); | |
1159 | iomap_finish_ioend(ioend, error); | |
1160 | ||
1161 | while (!list_empty(&tmp)) { | |
1162 | ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); | |
1163 | list_del_init(&ioend->io_list); | |
1164 | iomap_finish_ioend(ioend, error); | |
1165 | } | |
1166 | } | |
1167 | EXPORT_SYMBOL_GPL(iomap_finish_ioends); | |
1168 | ||
1169 | /* | |
1170 | * We can merge two adjacent ioends if they have the same set of work to do. | |
1171 | */ | |
1172 | static bool | |
1173 | iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) | |
1174 | { | |
1175 | if (ioend->io_bio->bi_status != next->io_bio->bi_status) | |
1176 | return false; | |
1177 | if ((ioend->io_flags & IOMAP_F_SHARED) ^ | |
1178 | (next->io_flags & IOMAP_F_SHARED)) | |
1179 | return false; | |
1180 | if ((ioend->io_type == IOMAP_UNWRITTEN) ^ | |
1181 | (next->io_type == IOMAP_UNWRITTEN)) | |
1182 | return false; | |
1183 | if (ioend->io_offset + ioend->io_size != next->io_offset) | |
1184 | return false; | |
1185 | return true; | |
1186 | } | |
1187 | ||
1188 | void | |
1189 | iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends, | |
1190 | void (*merge_private)(struct iomap_ioend *ioend, | |
1191 | struct iomap_ioend *next)) | |
1192 | { | |
1193 | struct iomap_ioend *next; | |
1194 | ||
1195 | INIT_LIST_HEAD(&ioend->io_list); | |
1196 | ||
1197 | while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, | |
1198 | io_list))) { | |
1199 | if (!iomap_ioend_can_merge(ioend, next)) | |
1200 | break; | |
1201 | list_move_tail(&next->io_list, &ioend->io_list); | |
1202 | ioend->io_size += next->io_size; | |
1203 | if (next->io_private && merge_private) | |
1204 | merge_private(ioend, next); | |
1205 | } | |
1206 | } | |
1207 | EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); | |
1208 | ||
1209 | static int | |
1210 | iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b) | |
1211 | { | |
1212 | struct iomap_ioend *ia, *ib; | |
1213 | ||
1214 | ia = container_of(a, struct iomap_ioend, io_list); | |
1215 | ib = container_of(b, struct iomap_ioend, io_list); | |
1216 | if (ia->io_offset < ib->io_offset) | |
1217 | return -1; | |
1218 | else if (ia->io_offset > ib->io_offset) | |
1219 | return 1; | |
1220 | return 0; | |
1221 | } | |
1222 | ||
1223 | void | |
1224 | iomap_sort_ioends(struct list_head *ioend_list) | |
1225 | { | |
1226 | list_sort(NULL, ioend_list, iomap_ioend_compare); | |
1227 | } | |
1228 | EXPORT_SYMBOL_GPL(iomap_sort_ioends); | |
1229 | ||
1230 | static void iomap_writepage_end_bio(struct bio *bio) | |
1231 | { | |
1232 | struct iomap_ioend *ioend = bio->bi_private; | |
1233 | ||
1234 | iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); | |
1235 | } | |
1236 | ||
1237 | /* | |
1238 | * Submit the final bio for an ioend. | |
1239 | * | |
1240 | * If @error is non-zero, it means that we have a situation where some part of | |
1241 | * the submission process has failed after we have marked paged for writeback | |
1242 | * and unlocked them. In this situation, we need to fail the bio instead of | |
1243 | * submitting it. This typically only happens on a filesystem shutdown. | |
1244 | */ | |
1245 | static int | |
1246 | iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, | |
1247 | int error) | |
1248 | { | |
1249 | ioend->io_bio->bi_private = ioend; | |
1250 | ioend->io_bio->bi_end_io = iomap_writepage_end_bio; | |
1251 | ||
1252 | if (wpc->ops->prepare_ioend) | |
1253 | error = wpc->ops->prepare_ioend(ioend, error); | |
1254 | if (error) { | |
1255 | /* | |
1256 | * If we are failing the IO now, just mark the ioend with an | |
1257 | * error and finish it. This will run IO completion immediately | |
1258 | * as there is only one reference to the ioend at this point in | |
1259 | * time. | |
1260 | */ | |
1261 | ioend->io_bio->bi_status = errno_to_blk_status(error); | |
1262 | bio_endio(ioend->io_bio); | |
1263 | return error; | |
1264 | } | |
1265 | ||
1266 | submit_bio(ioend->io_bio); | |
1267 | return 0; | |
1268 | } | |
1269 | ||
1270 | static struct iomap_ioend * | |
1271 | iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, | |
1272 | loff_t offset, sector_t sector, struct writeback_control *wbc) | |
1273 | { | |
1274 | struct iomap_ioend *ioend; | |
1275 | struct bio *bio; | |
1276 | ||
1277 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &iomap_ioend_bioset); | |
1278 | bio_set_dev(bio, wpc->iomap.bdev); | |
1279 | bio->bi_iter.bi_sector = sector; | |
1280 | bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); | |
1281 | bio->bi_write_hint = inode->i_write_hint; | |
1282 | wbc_init_bio(wbc, bio); | |
1283 | ||
1284 | ioend = container_of(bio, struct iomap_ioend, io_inline_bio); | |
1285 | INIT_LIST_HEAD(&ioend->io_list); | |
1286 | ioend->io_type = wpc->iomap.type; | |
1287 | ioend->io_flags = wpc->iomap.flags; | |
1288 | ioend->io_inode = inode; | |
1289 | ioend->io_size = 0; | |
1290 | ioend->io_offset = offset; | |
1291 | ioend->io_private = NULL; | |
1292 | ioend->io_bio = bio; | |
1293 | return ioend; | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * Allocate a new bio, and chain the old bio to the new one. | |
1298 | * | |
1299 | * Note that we have to do perform the chaining in this unintuitive order | |
1300 | * so that the bi_private linkage is set up in the right direction for the | |
1301 | * traversal in iomap_finish_ioend(). | |
1302 | */ | |
1303 | static struct bio * | |
1304 | iomap_chain_bio(struct bio *prev) | |
1305 | { | |
1306 | struct bio *new; | |
1307 | ||
1308 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
1309 | bio_copy_dev(new, prev);/* also copies over blkcg information */ | |
1310 | new->bi_iter.bi_sector = bio_end_sector(prev); | |
1311 | new->bi_opf = prev->bi_opf; | |
1312 | new->bi_write_hint = prev->bi_write_hint; | |
1313 | ||
1314 | bio_chain(prev, new); | |
1315 | bio_get(prev); /* for iomap_finish_ioend */ | |
1316 | submit_bio(prev); | |
1317 | return new; | |
1318 | } | |
1319 | ||
1320 | static bool | |
1321 | iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, | |
1322 | sector_t sector) | |
1323 | { | |
1324 | if ((wpc->iomap.flags & IOMAP_F_SHARED) != | |
1325 | (wpc->ioend->io_flags & IOMAP_F_SHARED)) | |
1326 | return false; | |
1327 | if (wpc->iomap.type != wpc->ioend->io_type) | |
1328 | return false; | |
1329 | if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) | |
1330 | return false; | |
1331 | if (sector != bio_end_sector(wpc->ioend->io_bio)) | |
1332 | return false; | |
1333 | return true; | |
1334 | } | |
1335 | ||
1336 | /* | |
1337 | * Test to see if we have an existing ioend structure that we could append to | |
1338 | * first, otherwise finish off the current ioend and start another. | |
1339 | */ | |
1340 | static void | |
1341 | iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, | |
1342 | struct iomap_page *iop, struct iomap_writepage_ctx *wpc, | |
1343 | struct writeback_control *wbc, struct list_head *iolist) | |
1344 | { | |
1345 | sector_t sector = iomap_sector(&wpc->iomap, offset); | |
1346 | unsigned len = i_blocksize(inode); | |
1347 | unsigned poff = offset & (PAGE_SIZE - 1); | |
1348 | bool merged, same_page = false; | |
1349 | ||
1350 | if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { | |
1351 | if (wpc->ioend) | |
1352 | list_add(&wpc->ioend->io_list, iolist); | |
1353 | wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); | |
1354 | } | |
1355 | ||
1356 | merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff, | |
1357 | &same_page); | |
1358 | if (iop && !same_page) | |
1359 | atomic_inc(&iop->write_count); | |
1360 | ||
1361 | if (!merged) { | |
1362 | if (bio_full(wpc->ioend->io_bio, len)) { | |
1363 | wpc->ioend->io_bio = | |
1364 | iomap_chain_bio(wpc->ioend->io_bio); | |
1365 | } | |
1366 | bio_add_page(wpc->ioend->io_bio, page, len, poff); | |
1367 | } | |
1368 | ||
1369 | wpc->ioend->io_size += len; | |
1370 | wbc_account_cgroup_owner(wbc, page, len); | |
1371 | } | |
1372 | ||
1373 | /* | |
1374 | * We implement an immediate ioend submission policy here to avoid needing to | |
1375 | * chain multiple ioends and hence nest mempool allocations which can violate | |
1376 | * forward progress guarantees we need to provide. The current ioend we are | |
1377 | * adding blocks to is cached on the writepage context, and if the new block | |
1378 | * does not append to the cached ioend it will create a new ioend and cache that | |
1379 | * instead. | |
1380 | * | |
1381 | * If a new ioend is created and cached, the old ioend is returned and queued | |
1382 | * locally for submission once the entire page is processed or an error has been | |
1383 | * detected. While ioends are submitted immediately after they are completed, | |
1384 | * batching optimisations are provided by higher level block plugging. | |
1385 | * | |
1386 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
1387 | * writepage context that the caller will need to submit. | |
1388 | */ | |
1389 | static int | |
1390 | iomap_writepage_map(struct iomap_writepage_ctx *wpc, | |
1391 | struct writeback_control *wbc, struct inode *inode, | |
1392 | struct page *page, u64 end_offset) | |
1393 | { | |
1394 | struct iomap_page *iop = to_iomap_page(page); | |
1395 | struct iomap_ioend *ioend, *next; | |
1396 | unsigned len = i_blocksize(inode); | |
1397 | u64 file_offset; /* file offset of page */ | |
1398 | int error = 0, count = 0, i; | |
1399 | LIST_HEAD(submit_list); | |
1400 | ||
1401 | WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); | |
1402 | WARN_ON_ONCE(iop && atomic_read(&iop->write_count) != 0); | |
1403 | ||
1404 | /* | |
1405 | * Walk through the page to find areas to write back. If we run off the | |
1406 | * end of the current map or find the current map invalid, grab a new | |
1407 | * one. | |
1408 | */ | |
1409 | for (i = 0, file_offset = page_offset(page); | |
1410 | i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; | |
1411 | i++, file_offset += len) { | |
1412 | if (iop && !test_bit(i, iop->uptodate)) | |
1413 | continue; | |
1414 | ||
1415 | error = wpc->ops->map_blocks(wpc, inode, file_offset); | |
1416 | if (error) | |
1417 | break; | |
3e19e6f3 CH |
1418 | if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) |
1419 | continue; | |
598ecfba CH |
1420 | if (wpc->iomap.type == IOMAP_HOLE) |
1421 | continue; | |
1422 | iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, | |
1423 | &submit_list); | |
1424 | count++; | |
1425 | } | |
1426 | ||
1427 | WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); | |
1428 | WARN_ON_ONCE(!PageLocked(page)); | |
1429 | WARN_ON_ONCE(PageWriteback(page)); | |
1430 | ||
1431 | /* | |
1432 | * We cannot cancel the ioend directly here on error. We may have | |
1433 | * already set other pages under writeback and hence we have to run I/O | |
1434 | * completion to mark the error state of the pages under writeback | |
1435 | * appropriately. | |
1436 | */ | |
1437 | if (unlikely(error)) { | |
1438 | if (!count) { | |
1439 | /* | |
1440 | * If the current page hasn't been added to ioend, it | |
1441 | * won't be affected by I/O completions and we must | |
1442 | * discard and unlock it right here. | |
1443 | */ | |
1444 | if (wpc->ops->discard_page) | |
1445 | wpc->ops->discard_page(page); | |
1446 | ClearPageUptodate(page); | |
1447 | unlock_page(page); | |
1448 | goto done; | |
1449 | } | |
1450 | ||
1451 | /* | |
1452 | * If the page was not fully cleaned, we need to ensure that the | |
1453 | * higher layers come back to it correctly. That means we need | |
1454 | * to keep the page dirty, and for WB_SYNC_ALL writeback we need | |
1455 | * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed | |
1456 | * so another attempt to write this page in this writeback sweep | |
1457 | * will be made. | |
1458 | */ | |
1459 | set_page_writeback_keepwrite(page); | |
1460 | } else { | |
1461 | clear_page_dirty_for_io(page); | |
1462 | set_page_writeback(page); | |
1463 | } | |
1464 | ||
1465 | unlock_page(page); | |
1466 | ||
1467 | /* | |
1468 | * Preserve the original error if there was one, otherwise catch | |
1469 | * submission errors here and propagate into subsequent ioend | |
1470 | * submissions. | |
1471 | */ | |
1472 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1473 | int error2; | |
1474 | ||
1475 | list_del_init(&ioend->io_list); | |
1476 | error2 = iomap_submit_ioend(wpc, ioend, error); | |
1477 | if (error2 && !error) | |
1478 | error = error2; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * We can end up here with no error and nothing to write only if we race | |
1483 | * with a partial page truncate on a sub-page block sized filesystem. | |
1484 | */ | |
1485 | if (!count) | |
1486 | end_page_writeback(page); | |
1487 | done: | |
1488 | mapping_set_error(page->mapping, error); | |
1489 | return error; | |
1490 | } | |
1491 | ||
1492 | /* | |
1493 | * Write out a dirty page. | |
1494 | * | |
1495 | * For delalloc space on the page we need to allocate space and flush it. | |
1496 | * For unwritten space on the page we need to start the conversion to | |
1497 | * regular allocated space. | |
1498 | */ | |
1499 | static int | |
1500 | iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) | |
1501 | { | |
1502 | struct iomap_writepage_ctx *wpc = data; | |
1503 | struct inode *inode = page->mapping->host; | |
1504 | pgoff_t end_index; | |
1505 | u64 end_offset; | |
1506 | loff_t offset; | |
1507 | ||
1508 | trace_iomap_writepage(inode, page, 0, 0); | |
1509 | ||
1510 | /* | |
1511 | * Refuse to write the page out if we are called from reclaim context. | |
1512 | * | |
1513 | * This avoids stack overflows when called from deeply used stacks in | |
1514 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1515 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
1516 | * | |
1517 | * This should never happen except in the case of a VM regression so | |
1518 | * warn about it. | |
1519 | */ | |
1520 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == | |
1521 | PF_MEMALLOC)) | |
1522 | goto redirty; | |
1523 | ||
1524 | /* | |
1525 | * Given that we do not allow direct reclaim to call us, we should | |
1526 | * never be called in a recursive filesystem reclaim context. | |
1527 | */ | |
1528 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) | |
1529 | goto redirty; | |
1530 | ||
1531 | /* | |
1532 | * Is this page beyond the end of the file? | |
1533 | * | |
1534 | * The page index is less than the end_index, adjust the end_offset | |
1535 | * to the highest offset that this page should represent. | |
1536 | * ----------------------------------------------------- | |
1537 | * | file mapping | <EOF> | | |
1538 | * ----------------------------------------------------- | |
1539 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1540 | * ^--------------------------------^----------|-------- | |
1541 | * | desired writeback range | see else | | |
1542 | * ---------------------------------^------------------| | |
1543 | */ | |
1544 | offset = i_size_read(inode); | |
1545 | end_index = offset >> PAGE_SHIFT; | |
1546 | if (page->index < end_index) | |
1547 | end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; | |
1548 | else { | |
1549 | /* | |
1550 | * Check whether the page to write out is beyond or straddles | |
1551 | * i_size or not. | |
1552 | * ------------------------------------------------------- | |
1553 | * | file mapping | <EOF> | | |
1554 | * ------------------------------------------------------- | |
1555 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1556 | * ^--------------------------------^-----------|--------- | |
1557 | * | | Straddles | | |
1558 | * ---------------------------------^-----------|--------| | |
1559 | */ | |
1560 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); | |
1561 | ||
1562 | /* | |
1563 | * Skip the page if it is fully outside i_size, e.g. due to a | |
1564 | * truncate operation that is in progress. We must redirty the | |
1565 | * page so that reclaim stops reclaiming it. Otherwise | |
1566 | * iomap_vm_releasepage() is called on it and gets confused. | |
1567 | * | |
1568 | * Note that the end_index is unsigned long, it would overflow | |
1569 | * if the given offset is greater than 16TB on 32-bit system | |
1570 | * and if we do check the page is fully outside i_size or not | |
1571 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1572 | * will be evaluated to 0. Hence this page will be redirtied | |
1573 | * and be written out repeatedly which would result in an | |
1574 | * infinite loop, the user program that perform this operation | |
1575 | * will hang. Instead, we can verify this situation by checking | |
1576 | * if the page to write is totally beyond the i_size or if it's | |
1577 | * offset is just equal to the EOF. | |
1578 | */ | |
1579 | if (page->index > end_index || | |
1580 | (page->index == end_index && offset_into_page == 0)) | |
1581 | goto redirty; | |
1582 | ||
1583 | /* | |
1584 | * The page straddles i_size. It must be zeroed out on each | |
1585 | * and every writepage invocation because it may be mmapped. | |
1586 | * "A file is mapped in multiples of the page size. For a file | |
1587 | * that is not a multiple of the page size, the remaining | |
1588 | * memory is zeroed when mapped, and writes to that region are | |
1589 | * not written out to the file." | |
1590 | */ | |
1591 | zero_user_segment(page, offset_into_page, PAGE_SIZE); | |
1592 | ||
1593 | /* Adjust the end_offset to the end of file */ | |
1594 | end_offset = offset; | |
1595 | } | |
1596 | ||
1597 | return iomap_writepage_map(wpc, wbc, inode, page, end_offset); | |
1598 | ||
1599 | redirty: | |
1600 | redirty_page_for_writepage(wbc, page); | |
1601 | unlock_page(page); | |
1602 | return 0; | |
1603 | } | |
1604 | ||
1605 | int | |
1606 | iomap_writepage(struct page *page, struct writeback_control *wbc, | |
1607 | struct iomap_writepage_ctx *wpc, | |
1608 | const struct iomap_writeback_ops *ops) | |
1609 | { | |
1610 | int ret; | |
1611 | ||
1612 | wpc->ops = ops; | |
1613 | ret = iomap_do_writepage(page, wbc, wpc); | |
1614 | if (!wpc->ioend) | |
1615 | return ret; | |
1616 | return iomap_submit_ioend(wpc, wpc->ioend, ret); | |
1617 | } | |
1618 | EXPORT_SYMBOL_GPL(iomap_writepage); | |
1619 | ||
1620 | int | |
1621 | iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, | |
1622 | struct iomap_writepage_ctx *wpc, | |
1623 | const struct iomap_writeback_ops *ops) | |
1624 | { | |
1625 | int ret; | |
1626 | ||
1627 | wpc->ops = ops; | |
1628 | ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); | |
1629 | if (!wpc->ioend) | |
1630 | return ret; | |
1631 | return iomap_submit_ioend(wpc, wpc->ioend, ret); | |
1632 | } | |
1633 | EXPORT_SYMBOL_GPL(iomap_writepages); | |
1634 | ||
1635 | static int __init iomap_init(void) | |
1636 | { | |
1637 | return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), | |
1638 | offsetof(struct iomap_ioend, io_inline_bio), | |
1639 | BIOSET_NEED_BVECS); | |
1640 | } | |
1641 | fs_initcall(iomap_init); |