]>
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 | ||
ab08b01e CH |
24 | /* |
25 | * Structure allocated for each page when block size < PAGE_SIZE to track | |
26 | * sub-page uptodate status and I/O completions. | |
27 | */ | |
28 | struct iomap_page { | |
29 | atomic_t read_count; | |
30 | atomic_t write_count; | |
1cea335d | 31 | spinlock_t uptodate_lock; |
ab08b01e CH |
32 | DECLARE_BITMAP(uptodate, PAGE_SIZE / 512); |
33 | }; | |
34 | ||
35 | static inline struct iomap_page *to_iomap_page(struct page *page) | |
36 | { | |
37 | if (page_has_private(page)) | |
38 | return (struct iomap_page *)page_private(page); | |
39 | return NULL; | |
40 | } | |
41 | ||
598ecfba CH |
42 | static struct bio_set iomap_ioend_bioset; |
43 | ||
afc51aaa DW |
44 | static struct iomap_page * |
45 | iomap_page_create(struct inode *inode, struct page *page) | |
46 | { | |
47 | struct iomap_page *iop = to_iomap_page(page); | |
48 | ||
49 | if (iop || i_blocksize(inode) == PAGE_SIZE) | |
50 | return iop; | |
51 | ||
52 | iop = kmalloc(sizeof(*iop), GFP_NOFS | __GFP_NOFAIL); | |
53 | atomic_set(&iop->read_count, 0); | |
54 | atomic_set(&iop->write_count, 0); | |
1cea335d | 55 | spin_lock_init(&iop->uptodate_lock); |
afc51aaa DW |
56 | bitmap_zero(iop->uptodate, PAGE_SIZE / SECTOR_SIZE); |
57 | ||
58 | /* | |
59 | * migrate_page_move_mapping() assumes that pages with private data have | |
60 | * their count elevated by 1. | |
61 | */ | |
58aeb731 | 62 | attach_page_private(page, iop); |
afc51aaa DW |
63 | return iop; |
64 | } | |
65 | ||
66 | static void | |
67 | iomap_page_release(struct page *page) | |
68 | { | |
58aeb731 | 69 | struct iomap_page *iop = detach_page_private(page); |
afc51aaa DW |
70 | |
71 | if (!iop) | |
72 | return; | |
73 | WARN_ON_ONCE(atomic_read(&iop->read_count)); | |
74 | WARN_ON_ONCE(atomic_read(&iop->write_count)); | |
afc51aaa DW |
75 | kfree(iop); |
76 | } | |
77 | ||
78 | /* | |
79 | * Calculate the range inside the page that we actually need to read. | |
80 | */ | |
81 | static void | |
82 | iomap_adjust_read_range(struct inode *inode, struct iomap_page *iop, | |
83 | loff_t *pos, loff_t length, unsigned *offp, unsigned *lenp) | |
84 | { | |
85 | loff_t orig_pos = *pos; | |
86 | loff_t isize = i_size_read(inode); | |
87 | unsigned block_bits = inode->i_blkbits; | |
88 | unsigned block_size = (1 << block_bits); | |
89 | unsigned poff = offset_in_page(*pos); | |
90 | unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length); | |
91 | unsigned first = poff >> block_bits; | |
92 | unsigned last = (poff + plen - 1) >> block_bits; | |
93 | ||
94 | /* | |
95 | * If the block size is smaller than the page size we need to check the | |
96 | * per-block uptodate status and adjust the offset and length if needed | |
97 | * to avoid reading in already uptodate ranges. | |
98 | */ | |
99 | if (iop) { | |
100 | unsigned int i; | |
101 | ||
102 | /* move forward for each leading block marked uptodate */ | |
103 | for (i = first; i <= last; i++) { | |
104 | if (!test_bit(i, iop->uptodate)) | |
105 | break; | |
106 | *pos += block_size; | |
107 | poff += block_size; | |
108 | plen -= block_size; | |
109 | first++; | |
110 | } | |
111 | ||
112 | /* truncate len if we find any trailing uptodate block(s) */ | |
113 | for ( ; i <= last; i++) { | |
114 | if (test_bit(i, iop->uptodate)) { | |
115 | plen -= (last - i + 1) * block_size; | |
116 | last = i - 1; | |
117 | break; | |
118 | } | |
119 | } | |
120 | } | |
121 | ||
122 | /* | |
123 | * If the extent spans the block that contains the i_size we need to | |
124 | * handle both halves separately so that we properly zero data in the | |
125 | * page cache for blocks that are entirely outside of i_size. | |
126 | */ | |
127 | if (orig_pos <= isize && orig_pos + length > isize) { | |
128 | unsigned end = offset_in_page(isize - 1) >> block_bits; | |
129 | ||
130 | if (first <= end && last > end) | |
131 | plen -= (last - end) * block_size; | |
132 | } | |
133 | ||
134 | *offp = poff; | |
135 | *lenp = plen; | |
136 | } | |
137 | ||
138 | static void | |
1cea335d | 139 | iomap_iop_set_range_uptodate(struct page *page, unsigned off, unsigned len) |
afc51aaa DW |
140 | { |
141 | struct iomap_page *iop = to_iomap_page(page); | |
142 | struct inode *inode = page->mapping->host; | |
143 | unsigned first = off >> inode->i_blkbits; | |
144 | unsigned last = (off + len - 1) >> inode->i_blkbits; | |
afc51aaa | 145 | bool uptodate = true; |
1cea335d CH |
146 | unsigned long flags; |
147 | unsigned int i; | |
afc51aaa | 148 | |
1cea335d CH |
149 | spin_lock_irqsave(&iop->uptodate_lock, flags); |
150 | for (i = 0; i < PAGE_SIZE / i_blocksize(inode); i++) { | |
151 | if (i >= first && i <= last) | |
152 | set_bit(i, iop->uptodate); | |
153 | else if (!test_bit(i, iop->uptodate)) | |
154 | uptodate = false; | |
afc51aaa DW |
155 | } |
156 | ||
1cea335d CH |
157 | if (uptodate) |
158 | SetPageUptodate(page); | |
159 | spin_unlock_irqrestore(&iop->uptodate_lock, flags); | |
160 | } | |
161 | ||
162 | static void | |
163 | iomap_set_range_uptodate(struct page *page, unsigned off, unsigned len) | |
164 | { | |
165 | if (PageError(page)) | |
166 | return; | |
167 | ||
168 | if (page_has_private(page)) | |
169 | iomap_iop_set_range_uptodate(page, off, len); | |
170 | else | |
afc51aaa DW |
171 | SetPageUptodate(page); |
172 | } | |
173 | ||
174 | static void | |
175 | iomap_read_finish(struct iomap_page *iop, struct page *page) | |
176 | { | |
177 | if (!iop || atomic_dec_and_test(&iop->read_count)) | |
178 | unlock_page(page); | |
179 | } | |
180 | ||
181 | static void | |
182 | iomap_read_page_end_io(struct bio_vec *bvec, int error) | |
183 | { | |
184 | struct page *page = bvec->bv_page; | |
185 | struct iomap_page *iop = to_iomap_page(page); | |
186 | ||
187 | if (unlikely(error)) { | |
188 | ClearPageUptodate(page); | |
189 | SetPageError(page); | |
190 | } else { | |
191 | iomap_set_range_uptodate(page, bvec->bv_offset, bvec->bv_len); | |
192 | } | |
193 | ||
194 | iomap_read_finish(iop, page); | |
195 | } | |
196 | ||
197 | static void | |
198 | iomap_read_end_io(struct bio *bio) | |
199 | { | |
200 | int error = blk_status_to_errno(bio->bi_status); | |
201 | struct bio_vec *bvec; | |
202 | struct bvec_iter_all iter_all; | |
203 | ||
204 | bio_for_each_segment_all(bvec, bio, iter_all) | |
205 | iomap_read_page_end_io(bvec, error); | |
206 | bio_put(bio); | |
207 | } | |
208 | ||
209 | struct iomap_readpage_ctx { | |
210 | struct page *cur_page; | |
211 | bool cur_page_in_bio; | |
afc51aaa | 212 | struct bio *bio; |
9d24a13a | 213 | struct readahead_control *rac; |
afc51aaa DW |
214 | }; |
215 | ||
216 | static void | |
217 | iomap_read_inline_data(struct inode *inode, struct page *page, | |
218 | struct iomap *iomap) | |
219 | { | |
220 | size_t size = i_size_read(inode); | |
221 | void *addr; | |
222 | ||
223 | if (PageUptodate(page)) | |
224 | return; | |
225 | ||
226 | BUG_ON(page->index); | |
227 | BUG_ON(size > PAGE_SIZE - offset_in_page(iomap->inline_data)); | |
228 | ||
229 | addr = kmap_atomic(page); | |
230 | memcpy(addr, iomap->inline_data, size); | |
231 | memset(addr + size, 0, PAGE_SIZE - size); | |
232 | kunmap_atomic(addr); | |
233 | SetPageUptodate(page); | |
234 | } | |
235 | ||
009d8d84 CH |
236 | static inline bool iomap_block_needs_zeroing(struct inode *inode, |
237 | struct iomap *iomap, loff_t pos) | |
238 | { | |
239 | return iomap->type != IOMAP_MAPPED || | |
240 | (iomap->flags & IOMAP_F_NEW) || | |
241 | pos >= i_size_read(inode); | |
242 | } | |
243 | ||
afc51aaa DW |
244 | static loff_t |
245 | iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
c039b997 | 246 | struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
247 | { |
248 | struct iomap_readpage_ctx *ctx = data; | |
249 | struct page *page = ctx->cur_page; | |
250 | struct iomap_page *iop = iomap_page_create(inode, page); | |
251 | bool same_page = false, is_contig = false; | |
252 | loff_t orig_pos = pos; | |
253 | unsigned poff, plen; | |
254 | sector_t sector; | |
255 | ||
256 | if (iomap->type == IOMAP_INLINE) { | |
257 | WARN_ON_ONCE(pos); | |
258 | iomap_read_inline_data(inode, page, iomap); | |
259 | return PAGE_SIZE; | |
260 | } | |
261 | ||
262 | /* zero post-eof blocks as the page may be mapped */ | |
263 | iomap_adjust_read_range(inode, iop, &pos, length, &poff, &plen); | |
264 | if (plen == 0) | |
265 | goto done; | |
266 | ||
009d8d84 | 267 | if (iomap_block_needs_zeroing(inode, iomap, pos)) { |
afc51aaa DW |
268 | zero_user(page, poff, plen); |
269 | iomap_set_range_uptodate(page, poff, plen); | |
270 | goto done; | |
271 | } | |
272 | ||
273 | ctx->cur_page_in_bio = true; | |
274 | ||
275 | /* | |
276 | * Try to merge into a previous segment if we can. | |
277 | */ | |
278 | sector = iomap_sector(iomap, pos); | |
279 | if (ctx->bio && bio_end_sector(ctx->bio) == sector) | |
280 | is_contig = true; | |
281 | ||
282 | if (is_contig && | |
283 | __bio_try_merge_page(ctx->bio, page, plen, poff, &same_page)) { | |
284 | if (!same_page && iop) | |
285 | atomic_inc(&iop->read_count); | |
286 | goto done; | |
287 | } | |
288 | ||
289 | /* | |
290 | * If we start a new segment we need to increase the read count, and we | |
291 | * need to do so before submitting any previous full bio to make sure | |
292 | * that we don't prematurely unlock the page. | |
293 | */ | |
294 | if (iop) | |
295 | atomic_inc(&iop->read_count); | |
296 | ||
297 | if (!ctx->bio || !is_contig || bio_full(ctx->bio, plen)) { | |
298 | gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL); | |
457df33e | 299 | gfp_t orig_gfp = gfp; |
afc51aaa DW |
300 | int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT; |
301 | ||
302 | if (ctx->bio) | |
303 | submit_bio(ctx->bio); | |
304 | ||
9d24a13a | 305 | if (ctx->rac) /* same as readahead_gfp_mask */ |
afc51aaa DW |
306 | gfp |= __GFP_NORETRY | __GFP_NOWARN; |
307 | ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs)); | |
457df33e MWO |
308 | /* |
309 | * If the bio_alloc fails, try it again for a single page to | |
310 | * avoid having to deal with partial page reads. This emulates | |
311 | * what do_mpage_readpage does. | |
312 | */ | |
313 | if (!ctx->bio) | |
314 | ctx->bio = bio_alloc(orig_gfp, 1); | |
afc51aaa | 315 | ctx->bio->bi_opf = REQ_OP_READ; |
9d24a13a | 316 | if (ctx->rac) |
afc51aaa DW |
317 | ctx->bio->bi_opf |= REQ_RAHEAD; |
318 | ctx->bio->bi_iter.bi_sector = sector; | |
319 | bio_set_dev(ctx->bio, iomap->bdev); | |
320 | ctx->bio->bi_end_io = iomap_read_end_io; | |
321 | } | |
322 | ||
323 | bio_add_page(ctx->bio, page, plen, poff); | |
324 | done: | |
325 | /* | |
326 | * Move the caller beyond our range so that it keeps making progress. | |
327 | * For that we have to include any leading non-uptodate ranges, but | |
328 | * we can skip trailing ones as they will be handled in the next | |
329 | * iteration. | |
330 | */ | |
331 | return pos - orig_pos + plen; | |
332 | } | |
333 | ||
334 | int | |
335 | iomap_readpage(struct page *page, const struct iomap_ops *ops) | |
336 | { | |
337 | struct iomap_readpage_ctx ctx = { .cur_page = page }; | |
338 | struct inode *inode = page->mapping->host; | |
339 | unsigned poff; | |
340 | loff_t ret; | |
341 | ||
9e91c572 CH |
342 | trace_iomap_readpage(page->mapping->host, 1); |
343 | ||
afc51aaa DW |
344 | for (poff = 0; poff < PAGE_SIZE; poff += ret) { |
345 | ret = iomap_apply(inode, page_offset(page) + poff, | |
346 | PAGE_SIZE - poff, 0, ops, &ctx, | |
347 | iomap_readpage_actor); | |
348 | if (ret <= 0) { | |
349 | WARN_ON_ONCE(ret == 0); | |
350 | SetPageError(page); | |
351 | break; | |
352 | } | |
353 | } | |
354 | ||
355 | if (ctx.bio) { | |
356 | submit_bio(ctx.bio); | |
357 | WARN_ON_ONCE(!ctx.cur_page_in_bio); | |
358 | } else { | |
359 | WARN_ON_ONCE(ctx.cur_page_in_bio); | |
360 | unlock_page(page); | |
361 | } | |
362 | ||
363 | /* | |
d4388340 | 364 | * Just like mpage_readahead and block_read_full_page we always |
afc51aaa DW |
365 | * return 0 and just mark the page as PageError on errors. This |
366 | * should be cleaned up all through the stack eventually. | |
367 | */ | |
368 | return 0; | |
369 | } | |
370 | EXPORT_SYMBOL_GPL(iomap_readpage); | |
371 | ||
afc51aaa | 372 | static loff_t |
9d24a13a | 373 | iomap_readahead_actor(struct inode *inode, loff_t pos, loff_t length, |
c039b997 | 374 | void *data, struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
375 | { |
376 | struct iomap_readpage_ctx *ctx = data; | |
377 | loff_t done, ret; | |
378 | ||
379 | for (done = 0; done < length; done += ret) { | |
380 | if (ctx->cur_page && offset_in_page(pos + done) == 0) { | |
381 | if (!ctx->cur_page_in_bio) | |
382 | unlock_page(ctx->cur_page); | |
383 | put_page(ctx->cur_page); | |
384 | ctx->cur_page = NULL; | |
385 | } | |
386 | if (!ctx->cur_page) { | |
9d24a13a | 387 | ctx->cur_page = readahead_page(ctx->rac); |
afc51aaa DW |
388 | ctx->cur_page_in_bio = false; |
389 | } | |
390 | ret = iomap_readpage_actor(inode, pos + done, length - done, | |
c039b997 | 391 | ctx, iomap, srcmap); |
afc51aaa DW |
392 | } |
393 | ||
394 | return done; | |
395 | } | |
396 | ||
9d24a13a MWO |
397 | /** |
398 | * iomap_readahead - Attempt to read pages from a file. | |
399 | * @rac: Describes the pages to be read. | |
400 | * @ops: The operations vector for the filesystem. | |
401 | * | |
402 | * This function is for filesystems to call to implement their readahead | |
403 | * address_space operation. | |
404 | * | |
405 | * Context: The @ops callbacks may submit I/O (eg to read the addresses of | |
406 | * blocks from disc), and may wait for it. The caller may be trying to | |
407 | * access a different page, and so sleeping excessively should be avoided. | |
408 | * It may allocate memory, but should avoid costly allocations. This | |
409 | * function is called with memalloc_nofs set, so allocations will not cause | |
410 | * the filesystem to be reentered. | |
411 | */ | |
412 | void iomap_readahead(struct readahead_control *rac, const struct iomap_ops *ops) | |
afc51aaa | 413 | { |
9d24a13a MWO |
414 | struct inode *inode = rac->mapping->host; |
415 | loff_t pos = readahead_pos(rac); | |
416 | loff_t length = readahead_length(rac); | |
afc51aaa | 417 | struct iomap_readpage_ctx ctx = { |
9d24a13a | 418 | .rac = rac, |
afc51aaa | 419 | }; |
afc51aaa | 420 | |
9d24a13a | 421 | trace_iomap_readahead(inode, readahead_count(rac)); |
9e91c572 | 422 | |
afc51aaa | 423 | while (length > 0) { |
9d24a13a MWO |
424 | loff_t ret = iomap_apply(inode, pos, length, 0, ops, |
425 | &ctx, iomap_readahead_actor); | |
afc51aaa DW |
426 | if (ret <= 0) { |
427 | WARN_ON_ONCE(ret == 0); | |
9d24a13a | 428 | break; |
afc51aaa DW |
429 | } |
430 | pos += ret; | |
431 | length -= ret; | |
432 | } | |
9d24a13a | 433 | |
afc51aaa DW |
434 | if (ctx.bio) |
435 | submit_bio(ctx.bio); | |
436 | if (ctx.cur_page) { | |
437 | if (!ctx.cur_page_in_bio) | |
438 | unlock_page(ctx.cur_page); | |
439 | put_page(ctx.cur_page); | |
440 | } | |
afc51aaa | 441 | } |
9d24a13a | 442 | EXPORT_SYMBOL_GPL(iomap_readahead); |
afc51aaa DW |
443 | |
444 | /* | |
445 | * iomap_is_partially_uptodate checks whether blocks within a page are | |
446 | * uptodate or not. | |
447 | * | |
448 | * Returns true if all blocks which correspond to a file portion | |
449 | * we want to read within the page are uptodate. | |
450 | */ | |
451 | int | |
452 | iomap_is_partially_uptodate(struct page *page, unsigned long from, | |
453 | unsigned long count) | |
454 | { | |
455 | struct iomap_page *iop = to_iomap_page(page); | |
456 | struct inode *inode = page->mapping->host; | |
457 | unsigned len, first, last; | |
458 | unsigned i; | |
459 | ||
460 | /* Limit range to one page */ | |
461 | len = min_t(unsigned, PAGE_SIZE - from, count); | |
462 | ||
463 | /* First and last blocks in range within page */ | |
464 | first = from >> inode->i_blkbits; | |
465 | last = (from + len - 1) >> inode->i_blkbits; | |
466 | ||
467 | if (iop) { | |
468 | for (i = first; i <= last; i++) | |
469 | if (!test_bit(i, iop->uptodate)) | |
470 | return 0; | |
471 | return 1; | |
472 | } | |
473 | ||
474 | return 0; | |
475 | } | |
476 | EXPORT_SYMBOL_GPL(iomap_is_partially_uptodate); | |
477 | ||
478 | int | |
479 | iomap_releasepage(struct page *page, gfp_t gfp_mask) | |
480 | { | |
1ac99452 MWO |
481 | trace_iomap_releasepage(page->mapping->host, page_offset(page), |
482 | PAGE_SIZE); | |
9e91c572 | 483 | |
afc51aaa DW |
484 | /* |
485 | * mm accommodates an old ext3 case where clean pages might not have had | |
486 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
487 | * ->releasepage() via shrink_active_list(), skip those here. | |
488 | */ | |
489 | if (PageDirty(page) || PageWriteback(page)) | |
490 | return 0; | |
491 | iomap_page_release(page); | |
492 | return 1; | |
493 | } | |
494 | EXPORT_SYMBOL_GPL(iomap_releasepage); | |
495 | ||
496 | void | |
497 | iomap_invalidatepage(struct page *page, unsigned int offset, unsigned int len) | |
498 | { | |
1ac99452 | 499 | trace_iomap_invalidatepage(page->mapping->host, offset, len); |
9e91c572 | 500 | |
afc51aaa DW |
501 | /* |
502 | * If we are invalidating the entire page, clear the dirty state from it | |
503 | * and release it to avoid unnecessary buildup of the LRU. | |
504 | */ | |
505 | if (offset == 0 && len == PAGE_SIZE) { | |
506 | WARN_ON_ONCE(PageWriteback(page)); | |
507 | cancel_dirty_page(page); | |
508 | iomap_page_release(page); | |
509 | } | |
510 | } | |
511 | EXPORT_SYMBOL_GPL(iomap_invalidatepage); | |
512 | ||
513 | #ifdef CONFIG_MIGRATION | |
514 | int | |
515 | iomap_migrate_page(struct address_space *mapping, struct page *newpage, | |
516 | struct page *page, enum migrate_mode mode) | |
517 | { | |
518 | int ret; | |
519 | ||
26473f83 | 520 | ret = migrate_page_move_mapping(mapping, newpage, page, 0); |
afc51aaa DW |
521 | if (ret != MIGRATEPAGE_SUCCESS) |
522 | return ret; | |
523 | ||
58aeb731 GJ |
524 | if (page_has_private(page)) |
525 | attach_page_private(newpage, detach_page_private(page)); | |
afc51aaa DW |
526 | |
527 | if (mode != MIGRATE_SYNC_NO_COPY) | |
528 | migrate_page_copy(newpage, page); | |
529 | else | |
530 | migrate_page_states(newpage, page); | |
531 | return MIGRATEPAGE_SUCCESS; | |
532 | } | |
533 | EXPORT_SYMBOL_GPL(iomap_migrate_page); | |
534 | #endif /* CONFIG_MIGRATION */ | |
535 | ||
32a38a49 CH |
536 | enum { |
537 | IOMAP_WRITE_F_UNSHARE = (1 << 0), | |
538 | }; | |
539 | ||
afc51aaa DW |
540 | static void |
541 | iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) | |
542 | { | |
543 | loff_t i_size = i_size_read(inode); | |
544 | ||
545 | /* | |
546 | * Only truncate newly allocated pages beyoned EOF, even if the | |
547 | * write started inside the existing inode size. | |
548 | */ | |
549 | if (pos + len > i_size) | |
550 | truncate_pagecache_range(inode, max(pos, i_size), pos + len); | |
551 | } | |
552 | ||
553 | static int | |
d3b40439 CH |
554 | iomap_read_page_sync(loff_t block_start, struct page *page, unsigned poff, |
555 | unsigned plen, struct iomap *iomap) | |
afc51aaa DW |
556 | { |
557 | struct bio_vec bvec; | |
558 | struct bio bio; | |
559 | ||
afc51aaa DW |
560 | bio_init(&bio, &bvec, 1); |
561 | bio.bi_opf = REQ_OP_READ; | |
562 | bio.bi_iter.bi_sector = iomap_sector(iomap, block_start); | |
563 | bio_set_dev(&bio, iomap->bdev); | |
564 | __bio_add_page(&bio, page, plen, poff); | |
565 | return submit_bio_wait(&bio); | |
566 | } | |
567 | ||
568 | static int | |
32a38a49 | 569 | __iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, int flags, |
c039b997 | 570 | struct page *page, struct iomap *srcmap) |
afc51aaa DW |
571 | { |
572 | struct iomap_page *iop = iomap_page_create(inode, page); | |
573 | loff_t block_size = i_blocksize(inode); | |
6cc19c5f NB |
574 | loff_t block_start = round_down(pos, block_size); |
575 | loff_t block_end = round_up(pos + len, block_size); | |
afc51aaa | 576 | unsigned from = offset_in_page(pos), to = from + len, poff, plen; |
afc51aaa DW |
577 | |
578 | if (PageUptodate(page)) | |
579 | return 0; | |
e6e7ca92 | 580 | ClearPageError(page); |
afc51aaa DW |
581 | |
582 | do { | |
583 | iomap_adjust_read_range(inode, iop, &block_start, | |
584 | block_end - block_start, &poff, &plen); | |
585 | if (plen == 0) | |
586 | break; | |
587 | ||
32a38a49 CH |
588 | if (!(flags & IOMAP_WRITE_F_UNSHARE) && |
589 | (from <= poff || from >= poff + plen) && | |
d3b40439 CH |
590 | (to <= poff || to >= poff + plen)) |
591 | continue; | |
592 | ||
c039b997 | 593 | if (iomap_block_needs_zeroing(inode, srcmap, block_start)) { |
32a38a49 CH |
594 | if (WARN_ON_ONCE(flags & IOMAP_WRITE_F_UNSHARE)) |
595 | return -EIO; | |
d3b40439 | 596 | zero_user_segments(page, poff, from, to, poff + plen); |
14284fed MWO |
597 | } else { |
598 | int status = iomap_read_page_sync(block_start, page, | |
599 | poff, plen, srcmap); | |
600 | if (status) | |
601 | return status; | |
afc51aaa | 602 | } |
14284fed | 603 | iomap_set_range_uptodate(page, poff, plen); |
afc51aaa DW |
604 | } while ((block_start += plen) < block_end); |
605 | ||
d3b40439 | 606 | return 0; |
afc51aaa DW |
607 | } |
608 | ||
609 | static int | |
610 | iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, | |
c039b997 | 611 | struct page **pagep, struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
612 | { |
613 | const struct iomap_page_ops *page_ops = iomap->page_ops; | |
afc51aaa DW |
614 | struct page *page; |
615 | int status = 0; | |
616 | ||
617 | BUG_ON(pos + len > iomap->offset + iomap->length); | |
c039b997 GR |
618 | if (srcmap != iomap) |
619 | BUG_ON(pos + len > srcmap->offset + srcmap->length); | |
afc51aaa DW |
620 | |
621 | if (fatal_signal_pending(current)) | |
622 | return -EINTR; | |
623 | ||
624 | if (page_ops && page_ops->page_prepare) { | |
625 | status = page_ops->page_prepare(inode, pos, len, iomap); | |
626 | if (status) | |
627 | return status; | |
628 | } | |
629 | ||
dcd6158d CH |
630 | page = grab_cache_page_write_begin(inode->i_mapping, pos >> PAGE_SHIFT, |
631 | AOP_FLAG_NOFS); | |
afc51aaa DW |
632 | if (!page) { |
633 | status = -ENOMEM; | |
634 | goto out_no_page; | |
635 | } | |
636 | ||
c039b997 GR |
637 | if (srcmap->type == IOMAP_INLINE) |
638 | iomap_read_inline_data(inode, page, srcmap); | |
afc51aaa | 639 | else if (iomap->flags & IOMAP_F_BUFFER_HEAD) |
c039b997 | 640 | status = __block_write_begin_int(page, pos, len, NULL, srcmap); |
afc51aaa | 641 | else |
32a38a49 | 642 | status = __iomap_write_begin(inode, pos, len, flags, page, |
c039b997 | 643 | srcmap); |
afc51aaa DW |
644 | |
645 | if (unlikely(status)) | |
646 | goto out_unlock; | |
647 | ||
648 | *pagep = page; | |
649 | return 0; | |
650 | ||
651 | out_unlock: | |
652 | unlock_page(page); | |
653 | put_page(page); | |
654 | iomap_write_failed(inode, pos, len); | |
655 | ||
656 | out_no_page: | |
657 | if (page_ops && page_ops->page_done) | |
658 | page_ops->page_done(inode, pos, 0, NULL, iomap); | |
659 | return status; | |
660 | } | |
661 | ||
662 | int | |
663 | iomap_set_page_dirty(struct page *page) | |
664 | { | |
665 | struct address_space *mapping = page_mapping(page); | |
666 | int newly_dirty; | |
667 | ||
668 | if (unlikely(!mapping)) | |
669 | return !TestSetPageDirty(page); | |
670 | ||
671 | /* | |
672 | * Lock out page->mem_cgroup migration to keep PageDirty | |
673 | * synchronized with per-memcg dirty page counters. | |
674 | */ | |
675 | lock_page_memcg(page); | |
676 | newly_dirty = !TestSetPageDirty(page); | |
677 | if (newly_dirty) | |
678 | __set_page_dirty(page, mapping, 0); | |
679 | unlock_page_memcg(page); | |
680 | ||
681 | if (newly_dirty) | |
682 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
683 | return newly_dirty; | |
684 | } | |
685 | EXPORT_SYMBOL_GPL(iomap_set_page_dirty); | |
686 | ||
687 | static int | |
688 | __iomap_write_end(struct inode *inode, loff_t pos, unsigned len, | |
c12d6fa8 | 689 | unsigned copied, struct page *page) |
afc51aaa DW |
690 | { |
691 | flush_dcache_page(page); | |
692 | ||
693 | /* | |
694 | * The blocks that were entirely written will now be uptodate, so we | |
695 | * don't have to worry about a readpage reading them and overwriting a | |
696 | * partial write. However if we have encountered a short write and only | |
697 | * partially written into a block, it will not be marked uptodate, so a | |
698 | * readpage might come in and destroy our partial write. | |
699 | * | |
700 | * Do the simplest thing, and just treat any short write to a non | |
701 | * uptodate page as a zero-length write, and force the caller to redo | |
702 | * the whole thing. | |
703 | */ | |
704 | if (unlikely(copied < len && !PageUptodate(page))) | |
705 | return 0; | |
706 | iomap_set_range_uptodate(page, offset_in_page(pos), len); | |
707 | iomap_set_page_dirty(page); | |
708 | return copied; | |
709 | } | |
710 | ||
711 | static int | |
712 | iomap_write_end_inline(struct inode *inode, struct page *page, | |
713 | struct iomap *iomap, loff_t pos, unsigned copied) | |
714 | { | |
715 | void *addr; | |
716 | ||
717 | WARN_ON_ONCE(!PageUptodate(page)); | |
718 | BUG_ON(pos + copied > PAGE_SIZE - offset_in_page(iomap->inline_data)); | |
719 | ||
720 | addr = kmap_atomic(page); | |
721 | memcpy(iomap->inline_data + pos, addr + pos, copied); | |
722 | kunmap_atomic(addr); | |
723 | ||
724 | mark_inode_dirty(inode); | |
725 | return copied; | |
726 | } | |
727 | ||
728 | static int | |
c039b997 GR |
729 | iomap_write_end(struct inode *inode, loff_t pos, unsigned len, unsigned copied, |
730 | struct page *page, struct iomap *iomap, struct iomap *srcmap) | |
afc51aaa DW |
731 | { |
732 | const struct iomap_page_ops *page_ops = iomap->page_ops; | |
733 | loff_t old_size = inode->i_size; | |
734 | int ret; | |
735 | ||
c039b997 | 736 | if (srcmap->type == IOMAP_INLINE) { |
afc51aaa | 737 | ret = iomap_write_end_inline(inode, page, iomap, pos, copied); |
c039b997 | 738 | } else if (srcmap->flags & IOMAP_F_BUFFER_HEAD) { |
afc51aaa DW |
739 | ret = block_write_end(NULL, inode->i_mapping, pos, len, copied, |
740 | page, NULL); | |
741 | } else { | |
c12d6fa8 | 742 | ret = __iomap_write_end(inode, pos, len, copied, page); |
afc51aaa DW |
743 | } |
744 | ||
745 | /* | |
746 | * Update the in-memory inode size after copying the data into the page | |
747 | * cache. It's up to the file system to write the updated size to disk, | |
748 | * preferably after I/O completion so that no stale data is exposed. | |
749 | */ | |
750 | if (pos + ret > old_size) { | |
751 | i_size_write(inode, pos + ret); | |
752 | iomap->flags |= IOMAP_F_SIZE_CHANGED; | |
753 | } | |
754 | unlock_page(page); | |
755 | ||
756 | if (old_size < pos) | |
757 | pagecache_isize_extended(inode, old_size, pos); | |
758 | if (page_ops && page_ops->page_done) | |
759 | page_ops->page_done(inode, pos, ret, page, iomap); | |
760 | put_page(page); | |
761 | ||
762 | if (ret < len) | |
763 | iomap_write_failed(inode, pos, len); | |
764 | return ret; | |
765 | } | |
766 | ||
767 | static loff_t | |
768 | iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
c039b997 | 769 | struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
770 | { |
771 | struct iov_iter *i = data; | |
772 | long status = 0; | |
773 | ssize_t written = 0; | |
afc51aaa DW |
774 | |
775 | do { | |
776 | struct page *page; | |
777 | unsigned long offset; /* Offset into pagecache page */ | |
778 | unsigned long bytes; /* Bytes to write to page */ | |
779 | size_t copied; /* Bytes copied from user */ | |
780 | ||
781 | offset = offset_in_page(pos); | |
782 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
783 | iov_iter_count(i)); | |
784 | again: | |
785 | if (bytes > length) | |
786 | bytes = length; | |
787 | ||
788 | /* | |
789 | * Bring in the user page that we will copy from _first_. | |
790 | * Otherwise there's a nasty deadlock on copying from the | |
791 | * same page as we're writing to, without it being marked | |
792 | * up-to-date. | |
793 | * | |
794 | * Not only is this an optimisation, but it is also required | |
795 | * to check that the address is actually valid, when atomic | |
796 | * usercopies are used, below. | |
797 | */ | |
798 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
799 | status = -EFAULT; | |
800 | break; | |
801 | } | |
802 | ||
c039b997 GR |
803 | status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, |
804 | srcmap); | |
afc51aaa DW |
805 | if (unlikely(status)) |
806 | break; | |
807 | ||
808 | if (mapping_writably_mapped(inode->i_mapping)) | |
809 | flush_dcache_page(page); | |
810 | ||
811 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
812 | ||
813 | flush_dcache_page(page); | |
814 | ||
c039b997 GR |
815 | status = iomap_write_end(inode, pos, bytes, copied, page, iomap, |
816 | srcmap); | |
afc51aaa DW |
817 | if (unlikely(status < 0)) |
818 | break; | |
819 | copied = status; | |
820 | ||
821 | cond_resched(); | |
822 | ||
823 | iov_iter_advance(i, copied); | |
824 | if (unlikely(copied == 0)) { | |
825 | /* | |
826 | * If we were unable to copy any data at all, we must | |
827 | * fall back to a single segment length write. | |
828 | * | |
829 | * If we didn't fallback here, we could livelock | |
830 | * because not all segments in the iov can be copied at | |
831 | * once without a pagefault. | |
832 | */ | |
833 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
834 | iov_iter_single_seg_count(i)); | |
835 | goto again; | |
836 | } | |
837 | pos += copied; | |
838 | written += copied; | |
839 | length -= copied; | |
840 | ||
841 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
842 | } while (iov_iter_count(i) && length); | |
843 | ||
844 | return written ? written : status; | |
845 | } | |
846 | ||
847 | ssize_t | |
848 | iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, | |
849 | const struct iomap_ops *ops) | |
850 | { | |
851 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
852 | loff_t pos = iocb->ki_pos, ret = 0, written = 0; | |
853 | ||
854 | while (iov_iter_count(iter)) { | |
855 | ret = iomap_apply(inode, pos, iov_iter_count(iter), | |
856 | IOMAP_WRITE, ops, iter, iomap_write_actor); | |
857 | if (ret <= 0) | |
858 | break; | |
859 | pos += ret; | |
860 | written += ret; | |
861 | } | |
862 | ||
863 | return written ? written : ret; | |
864 | } | |
865 | EXPORT_SYMBOL_GPL(iomap_file_buffered_write); | |
866 | ||
afc51aaa | 867 | static loff_t |
3590c4d8 | 868 | iomap_unshare_actor(struct inode *inode, loff_t pos, loff_t length, void *data, |
c039b997 | 869 | struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
870 | { |
871 | long status = 0; | |
d4ff3b2e | 872 | loff_t written = 0; |
afc51aaa | 873 | |
3590c4d8 CH |
874 | /* don't bother with blocks that are not shared to start with */ |
875 | if (!(iomap->flags & IOMAP_F_SHARED)) | |
876 | return length; | |
877 | /* don't bother with holes or unwritten extents */ | |
c039b997 | 878 | if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) |
3590c4d8 CH |
879 | return length; |
880 | ||
afc51aaa | 881 | do { |
32a38a49 CH |
882 | unsigned long offset = offset_in_page(pos); |
883 | unsigned long bytes = min_t(loff_t, PAGE_SIZE - offset, length); | |
884 | struct page *page; | |
afc51aaa | 885 | |
32a38a49 | 886 | status = iomap_write_begin(inode, pos, bytes, |
c039b997 | 887 | IOMAP_WRITE_F_UNSHARE, &page, iomap, srcmap); |
afc51aaa DW |
888 | if (unlikely(status)) |
889 | return status; | |
890 | ||
c039b997 GR |
891 | status = iomap_write_end(inode, pos, bytes, bytes, page, iomap, |
892 | srcmap); | |
afc51aaa DW |
893 | if (unlikely(status <= 0)) { |
894 | if (WARN_ON_ONCE(status == 0)) | |
895 | return -EIO; | |
896 | return status; | |
897 | } | |
898 | ||
899 | cond_resched(); | |
900 | ||
901 | pos += status; | |
902 | written += status; | |
903 | length -= status; | |
904 | ||
905 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
906 | } while (length); | |
907 | ||
908 | return written; | |
909 | } | |
910 | ||
911 | int | |
3590c4d8 | 912 | iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len, |
afc51aaa DW |
913 | const struct iomap_ops *ops) |
914 | { | |
915 | loff_t ret; | |
916 | ||
917 | while (len) { | |
918 | ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, | |
3590c4d8 | 919 | iomap_unshare_actor); |
afc51aaa DW |
920 | if (ret <= 0) |
921 | return ret; | |
922 | pos += ret; | |
923 | len -= ret; | |
924 | } | |
925 | ||
926 | return 0; | |
927 | } | |
3590c4d8 | 928 | EXPORT_SYMBOL_GPL(iomap_file_unshare); |
afc51aaa DW |
929 | |
930 | static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset, | |
c039b997 | 931 | unsigned bytes, struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
932 | { |
933 | struct page *page; | |
934 | int status; | |
935 | ||
c039b997 | 936 | status = iomap_write_begin(inode, pos, bytes, 0, &page, iomap, srcmap); |
afc51aaa DW |
937 | if (status) |
938 | return status; | |
939 | ||
940 | zero_user(page, offset, bytes); | |
941 | mark_page_accessed(page); | |
942 | ||
c039b997 | 943 | return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap); |
afc51aaa DW |
944 | } |
945 | ||
afc51aaa DW |
946 | static loff_t |
947 | iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count, | |
c039b997 | 948 | void *data, struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
949 | { |
950 | bool *did_zero = data; | |
951 | loff_t written = 0; | |
952 | int status; | |
953 | ||
954 | /* already zeroed? we're done. */ | |
c039b997 | 955 | if (srcmap->type == IOMAP_HOLE || srcmap->type == IOMAP_UNWRITTEN) |
afc51aaa DW |
956 | return count; |
957 | ||
958 | do { | |
959 | unsigned offset, bytes; | |
960 | ||
961 | offset = offset_in_page(pos); | |
962 | bytes = min_t(loff_t, PAGE_SIZE - offset, count); | |
963 | ||
964 | if (IS_DAX(inode)) | |
4f3b4f16 | 965 | status = dax_iomap_zero(pos, offset, bytes, iomap); |
afc51aaa | 966 | else |
c039b997 GR |
967 | status = iomap_zero(inode, pos, offset, bytes, iomap, |
968 | srcmap); | |
afc51aaa DW |
969 | if (status < 0) |
970 | return status; | |
971 | ||
972 | pos += bytes; | |
973 | count -= bytes; | |
974 | written += bytes; | |
975 | if (did_zero) | |
976 | *did_zero = true; | |
977 | } while (count > 0); | |
978 | ||
979 | return written; | |
980 | } | |
981 | ||
982 | int | |
983 | iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, | |
984 | const struct iomap_ops *ops) | |
985 | { | |
986 | loff_t ret; | |
987 | ||
988 | while (len > 0) { | |
989 | ret = iomap_apply(inode, pos, len, IOMAP_ZERO, | |
990 | ops, did_zero, iomap_zero_range_actor); | |
991 | if (ret <= 0) | |
992 | return ret; | |
993 | ||
994 | pos += ret; | |
995 | len -= ret; | |
996 | } | |
997 | ||
998 | return 0; | |
999 | } | |
1000 | EXPORT_SYMBOL_GPL(iomap_zero_range); | |
1001 | ||
1002 | int | |
1003 | iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, | |
1004 | const struct iomap_ops *ops) | |
1005 | { | |
1006 | unsigned int blocksize = i_blocksize(inode); | |
1007 | unsigned int off = pos & (blocksize - 1); | |
1008 | ||
1009 | /* Block boundary? Nothing to do */ | |
1010 | if (!off) | |
1011 | return 0; | |
1012 | return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); | |
1013 | } | |
1014 | EXPORT_SYMBOL_GPL(iomap_truncate_page); | |
1015 | ||
1016 | static loff_t | |
1017 | iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, | |
c039b997 | 1018 | void *data, struct iomap *iomap, struct iomap *srcmap) |
afc51aaa DW |
1019 | { |
1020 | struct page *page = data; | |
1021 | int ret; | |
1022 | ||
1023 | if (iomap->flags & IOMAP_F_BUFFER_HEAD) { | |
1024 | ret = __block_write_begin_int(page, pos, length, NULL, iomap); | |
1025 | if (ret) | |
1026 | return ret; | |
1027 | block_commit_write(page, 0, length); | |
1028 | } else { | |
1029 | WARN_ON_ONCE(!PageUptodate(page)); | |
1030 | iomap_page_create(inode, page); | |
1031 | set_page_dirty(page); | |
1032 | } | |
1033 | ||
1034 | return length; | |
1035 | } | |
1036 | ||
1037 | vm_fault_t iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) | |
1038 | { | |
1039 | struct page *page = vmf->page; | |
1040 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
1041 | unsigned long length; | |
243145bc | 1042 | loff_t offset; |
afc51aaa DW |
1043 | ssize_t ret; |
1044 | ||
1045 | lock_page(page); | |
243145bc AG |
1046 | ret = page_mkwrite_check_truncate(page, inode); |
1047 | if (ret < 0) | |
afc51aaa | 1048 | goto out_unlock; |
243145bc | 1049 | length = ret; |
afc51aaa | 1050 | |
243145bc | 1051 | offset = page_offset(page); |
afc51aaa DW |
1052 | while (length > 0) { |
1053 | ret = iomap_apply(inode, offset, length, | |
1054 | IOMAP_WRITE | IOMAP_FAULT, ops, page, | |
1055 | iomap_page_mkwrite_actor); | |
1056 | if (unlikely(ret <= 0)) | |
1057 | goto out_unlock; | |
1058 | offset += ret; | |
1059 | length -= ret; | |
1060 | } | |
1061 | ||
1062 | wait_for_stable_page(page); | |
1063 | return VM_FAULT_LOCKED; | |
1064 | out_unlock: | |
1065 | unlock_page(page); | |
1066 | return block_page_mkwrite_return(ret); | |
1067 | } | |
1068 | EXPORT_SYMBOL_GPL(iomap_page_mkwrite); | |
598ecfba CH |
1069 | |
1070 | static void | |
48d64cd1 | 1071 | iomap_finish_page_writeback(struct inode *inode, struct page *page, |
598ecfba CH |
1072 | int error) |
1073 | { | |
48d64cd1 | 1074 | struct iomap_page *iop = to_iomap_page(page); |
598ecfba CH |
1075 | |
1076 | if (error) { | |
48d64cd1 | 1077 | SetPageError(page); |
598ecfba CH |
1078 | mapping_set_error(inode->i_mapping, -EIO); |
1079 | } | |
1080 | ||
1081 | WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); | |
1082 | WARN_ON_ONCE(iop && atomic_read(&iop->write_count) <= 0); | |
1083 | ||
1084 | if (!iop || atomic_dec_and_test(&iop->write_count)) | |
48d64cd1 | 1085 | end_page_writeback(page); |
598ecfba CH |
1086 | } |
1087 | ||
1088 | /* | |
1089 | * We're now finished for good with this ioend structure. Update the page | |
1090 | * state, release holds on bios, and finally free up memory. Do not use the | |
1091 | * ioend after this. | |
1092 | */ | |
1093 | static void | |
1094 | iomap_finish_ioend(struct iomap_ioend *ioend, int error) | |
1095 | { | |
1096 | struct inode *inode = ioend->io_inode; | |
1097 | struct bio *bio = &ioend->io_inline_bio; | |
1098 | struct bio *last = ioend->io_bio, *next; | |
1099 | u64 start = bio->bi_iter.bi_sector; | |
c275779f | 1100 | loff_t offset = ioend->io_offset; |
598ecfba CH |
1101 | bool quiet = bio_flagged(bio, BIO_QUIET); |
1102 | ||
1103 | for (bio = &ioend->io_inline_bio; bio; bio = next) { | |
1104 | struct bio_vec *bv; | |
1105 | struct bvec_iter_all iter_all; | |
1106 | ||
1107 | /* | |
1108 | * For the last bio, bi_private points to the ioend, so we | |
1109 | * need to explicitly end the iteration here. | |
1110 | */ | |
1111 | if (bio == last) | |
1112 | next = NULL; | |
1113 | else | |
1114 | next = bio->bi_private; | |
1115 | ||
1116 | /* walk each page on bio, ending page IO on them */ | |
1117 | bio_for_each_segment_all(bv, bio, iter_all) | |
48d64cd1 | 1118 | iomap_finish_page_writeback(inode, bv->bv_page, error); |
598ecfba CH |
1119 | bio_put(bio); |
1120 | } | |
c275779f | 1121 | /* The ioend has been freed by bio_put() */ |
598ecfba CH |
1122 | |
1123 | if (unlikely(error && !quiet)) { | |
1124 | printk_ratelimited(KERN_ERR | |
9cd0ed63 | 1125 | "%s: writeback error on inode %lu, offset %lld, sector %llu", |
c275779f | 1126 | inode->i_sb->s_id, inode->i_ino, offset, start); |
598ecfba CH |
1127 | } |
1128 | } | |
1129 | ||
1130 | void | |
1131 | iomap_finish_ioends(struct iomap_ioend *ioend, int error) | |
1132 | { | |
1133 | struct list_head tmp; | |
1134 | ||
1135 | list_replace_init(&ioend->io_list, &tmp); | |
1136 | iomap_finish_ioend(ioend, error); | |
1137 | ||
1138 | while (!list_empty(&tmp)) { | |
1139 | ioend = list_first_entry(&tmp, struct iomap_ioend, io_list); | |
1140 | list_del_init(&ioend->io_list); | |
1141 | iomap_finish_ioend(ioend, error); | |
1142 | } | |
1143 | } | |
1144 | EXPORT_SYMBOL_GPL(iomap_finish_ioends); | |
1145 | ||
1146 | /* | |
1147 | * We can merge two adjacent ioends if they have the same set of work to do. | |
1148 | */ | |
1149 | static bool | |
1150 | iomap_ioend_can_merge(struct iomap_ioend *ioend, struct iomap_ioend *next) | |
1151 | { | |
1152 | if (ioend->io_bio->bi_status != next->io_bio->bi_status) | |
1153 | return false; | |
1154 | if ((ioend->io_flags & IOMAP_F_SHARED) ^ | |
1155 | (next->io_flags & IOMAP_F_SHARED)) | |
1156 | return false; | |
1157 | if ((ioend->io_type == IOMAP_UNWRITTEN) ^ | |
1158 | (next->io_type == IOMAP_UNWRITTEN)) | |
1159 | return false; | |
1160 | if (ioend->io_offset + ioend->io_size != next->io_offset) | |
1161 | return false; | |
1162 | return true; | |
1163 | } | |
1164 | ||
1165 | void | |
1166 | iomap_ioend_try_merge(struct iomap_ioend *ioend, struct list_head *more_ioends, | |
1167 | void (*merge_private)(struct iomap_ioend *ioend, | |
1168 | struct iomap_ioend *next)) | |
1169 | { | |
1170 | struct iomap_ioend *next; | |
1171 | ||
1172 | INIT_LIST_HEAD(&ioend->io_list); | |
1173 | ||
1174 | while ((next = list_first_entry_or_null(more_ioends, struct iomap_ioend, | |
1175 | io_list))) { | |
1176 | if (!iomap_ioend_can_merge(ioend, next)) | |
1177 | break; | |
1178 | list_move_tail(&next->io_list, &ioend->io_list); | |
1179 | ioend->io_size += next->io_size; | |
1180 | if (next->io_private && merge_private) | |
1181 | merge_private(ioend, next); | |
1182 | } | |
1183 | } | |
1184 | EXPORT_SYMBOL_GPL(iomap_ioend_try_merge); | |
1185 | ||
1186 | static int | |
1187 | iomap_ioend_compare(void *priv, struct list_head *a, struct list_head *b) | |
1188 | { | |
b3d423ec CH |
1189 | struct iomap_ioend *ia = container_of(a, struct iomap_ioend, io_list); |
1190 | struct iomap_ioend *ib = container_of(b, struct iomap_ioend, io_list); | |
598ecfba | 1191 | |
598ecfba CH |
1192 | if (ia->io_offset < ib->io_offset) |
1193 | return -1; | |
b3d423ec | 1194 | if (ia->io_offset > ib->io_offset) |
598ecfba CH |
1195 | return 1; |
1196 | return 0; | |
1197 | } | |
1198 | ||
1199 | void | |
1200 | iomap_sort_ioends(struct list_head *ioend_list) | |
1201 | { | |
1202 | list_sort(NULL, ioend_list, iomap_ioend_compare); | |
1203 | } | |
1204 | EXPORT_SYMBOL_GPL(iomap_sort_ioends); | |
1205 | ||
1206 | static void iomap_writepage_end_bio(struct bio *bio) | |
1207 | { | |
1208 | struct iomap_ioend *ioend = bio->bi_private; | |
1209 | ||
1210 | iomap_finish_ioend(ioend, blk_status_to_errno(bio->bi_status)); | |
1211 | } | |
1212 | ||
1213 | /* | |
1214 | * Submit the final bio for an ioend. | |
1215 | * | |
1216 | * If @error is non-zero, it means that we have a situation where some part of | |
1217 | * the submission process has failed after we have marked paged for writeback | |
1218 | * and unlocked them. In this situation, we need to fail the bio instead of | |
1219 | * submitting it. This typically only happens on a filesystem shutdown. | |
1220 | */ | |
1221 | static int | |
1222 | iomap_submit_ioend(struct iomap_writepage_ctx *wpc, struct iomap_ioend *ioend, | |
1223 | int error) | |
1224 | { | |
1225 | ioend->io_bio->bi_private = ioend; | |
1226 | ioend->io_bio->bi_end_io = iomap_writepage_end_bio; | |
1227 | ||
1228 | if (wpc->ops->prepare_ioend) | |
1229 | error = wpc->ops->prepare_ioend(ioend, error); | |
1230 | if (error) { | |
1231 | /* | |
1232 | * If we are failing the IO now, just mark the ioend with an | |
1233 | * error and finish it. This will run IO completion immediately | |
1234 | * as there is only one reference to the ioend at this point in | |
1235 | * time. | |
1236 | */ | |
1237 | ioend->io_bio->bi_status = errno_to_blk_status(error); | |
1238 | bio_endio(ioend->io_bio); | |
1239 | return error; | |
1240 | } | |
1241 | ||
1242 | submit_bio(ioend->io_bio); | |
1243 | return 0; | |
1244 | } | |
1245 | ||
1246 | static struct iomap_ioend * | |
1247 | iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc, | |
1248 | loff_t offset, sector_t sector, struct writeback_control *wbc) | |
1249 | { | |
1250 | struct iomap_ioend *ioend; | |
1251 | struct bio *bio; | |
1252 | ||
1253 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &iomap_ioend_bioset); | |
1254 | bio_set_dev(bio, wpc->iomap.bdev); | |
1255 | bio->bi_iter.bi_sector = sector; | |
1256 | bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); | |
1257 | bio->bi_write_hint = inode->i_write_hint; | |
1258 | wbc_init_bio(wbc, bio); | |
1259 | ||
1260 | ioend = container_of(bio, struct iomap_ioend, io_inline_bio); | |
1261 | INIT_LIST_HEAD(&ioend->io_list); | |
1262 | ioend->io_type = wpc->iomap.type; | |
1263 | ioend->io_flags = wpc->iomap.flags; | |
1264 | ioend->io_inode = inode; | |
1265 | ioend->io_size = 0; | |
1266 | ioend->io_offset = offset; | |
1267 | ioend->io_private = NULL; | |
1268 | ioend->io_bio = bio; | |
1269 | return ioend; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * Allocate a new bio, and chain the old bio to the new one. | |
1274 | * | |
1275 | * Note that we have to do perform the chaining in this unintuitive order | |
1276 | * so that the bi_private linkage is set up in the right direction for the | |
1277 | * traversal in iomap_finish_ioend(). | |
1278 | */ | |
1279 | static struct bio * | |
1280 | iomap_chain_bio(struct bio *prev) | |
1281 | { | |
1282 | struct bio *new; | |
1283 | ||
1284 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
1285 | bio_copy_dev(new, prev);/* also copies over blkcg information */ | |
1286 | new->bi_iter.bi_sector = bio_end_sector(prev); | |
1287 | new->bi_opf = prev->bi_opf; | |
1288 | new->bi_write_hint = prev->bi_write_hint; | |
1289 | ||
1290 | bio_chain(prev, new); | |
1291 | bio_get(prev); /* for iomap_finish_ioend */ | |
1292 | submit_bio(prev); | |
1293 | return new; | |
1294 | } | |
1295 | ||
1296 | static bool | |
1297 | iomap_can_add_to_ioend(struct iomap_writepage_ctx *wpc, loff_t offset, | |
1298 | sector_t sector) | |
1299 | { | |
1300 | if ((wpc->iomap.flags & IOMAP_F_SHARED) != | |
1301 | (wpc->ioend->io_flags & IOMAP_F_SHARED)) | |
1302 | return false; | |
1303 | if (wpc->iomap.type != wpc->ioend->io_type) | |
1304 | return false; | |
1305 | if (offset != wpc->ioend->io_offset + wpc->ioend->io_size) | |
1306 | return false; | |
1307 | if (sector != bio_end_sector(wpc->ioend->io_bio)) | |
1308 | return false; | |
1309 | return true; | |
1310 | } | |
1311 | ||
1312 | /* | |
1313 | * Test to see if we have an existing ioend structure that we could append to | |
1314 | * first, otherwise finish off the current ioend and start another. | |
1315 | */ | |
1316 | static void | |
1317 | iomap_add_to_ioend(struct inode *inode, loff_t offset, struct page *page, | |
1318 | struct iomap_page *iop, struct iomap_writepage_ctx *wpc, | |
1319 | struct writeback_control *wbc, struct list_head *iolist) | |
1320 | { | |
1321 | sector_t sector = iomap_sector(&wpc->iomap, offset); | |
1322 | unsigned len = i_blocksize(inode); | |
1323 | unsigned poff = offset & (PAGE_SIZE - 1); | |
1324 | bool merged, same_page = false; | |
1325 | ||
1326 | if (!wpc->ioend || !iomap_can_add_to_ioend(wpc, offset, sector)) { | |
1327 | if (wpc->ioend) | |
1328 | list_add(&wpc->ioend->io_list, iolist); | |
1329 | wpc->ioend = iomap_alloc_ioend(inode, wpc, offset, sector, wbc); | |
1330 | } | |
1331 | ||
1332 | merged = __bio_try_merge_page(wpc->ioend->io_bio, page, len, poff, | |
1333 | &same_page); | |
1334 | if (iop && !same_page) | |
1335 | atomic_inc(&iop->write_count); | |
1336 | ||
1337 | if (!merged) { | |
1338 | if (bio_full(wpc->ioend->io_bio, len)) { | |
1339 | wpc->ioend->io_bio = | |
1340 | iomap_chain_bio(wpc->ioend->io_bio); | |
1341 | } | |
1342 | bio_add_page(wpc->ioend->io_bio, page, len, poff); | |
1343 | } | |
1344 | ||
1345 | wpc->ioend->io_size += len; | |
1346 | wbc_account_cgroup_owner(wbc, page, len); | |
1347 | } | |
1348 | ||
1349 | /* | |
1350 | * We implement an immediate ioend submission policy here to avoid needing to | |
1351 | * chain multiple ioends and hence nest mempool allocations which can violate | |
1352 | * forward progress guarantees we need to provide. The current ioend we are | |
1353 | * adding blocks to is cached on the writepage context, and if the new block | |
1354 | * does not append to the cached ioend it will create a new ioend and cache that | |
1355 | * instead. | |
1356 | * | |
1357 | * If a new ioend is created and cached, the old ioend is returned and queued | |
1358 | * locally for submission once the entire page is processed or an error has been | |
1359 | * detected. While ioends are submitted immediately after they are completed, | |
1360 | * batching optimisations are provided by higher level block plugging. | |
1361 | * | |
1362 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
1363 | * writepage context that the caller will need to submit. | |
1364 | */ | |
1365 | static int | |
1366 | iomap_writepage_map(struct iomap_writepage_ctx *wpc, | |
1367 | struct writeback_control *wbc, struct inode *inode, | |
1368 | struct page *page, u64 end_offset) | |
1369 | { | |
1370 | struct iomap_page *iop = to_iomap_page(page); | |
1371 | struct iomap_ioend *ioend, *next; | |
1372 | unsigned len = i_blocksize(inode); | |
1373 | u64 file_offset; /* file offset of page */ | |
1374 | int error = 0, count = 0, i; | |
1375 | LIST_HEAD(submit_list); | |
1376 | ||
1377 | WARN_ON_ONCE(i_blocksize(inode) < PAGE_SIZE && !iop); | |
1378 | WARN_ON_ONCE(iop && atomic_read(&iop->write_count) != 0); | |
1379 | ||
1380 | /* | |
1381 | * Walk through the page to find areas to write back. If we run off the | |
1382 | * end of the current map or find the current map invalid, grab a new | |
1383 | * one. | |
1384 | */ | |
1385 | for (i = 0, file_offset = page_offset(page); | |
1386 | i < (PAGE_SIZE >> inode->i_blkbits) && file_offset < end_offset; | |
1387 | i++, file_offset += len) { | |
1388 | if (iop && !test_bit(i, iop->uptodate)) | |
1389 | continue; | |
1390 | ||
1391 | error = wpc->ops->map_blocks(wpc, inode, file_offset); | |
1392 | if (error) | |
1393 | break; | |
3e19e6f3 CH |
1394 | if (WARN_ON_ONCE(wpc->iomap.type == IOMAP_INLINE)) |
1395 | continue; | |
598ecfba CH |
1396 | if (wpc->iomap.type == IOMAP_HOLE) |
1397 | continue; | |
1398 | iomap_add_to_ioend(inode, file_offset, page, iop, wpc, wbc, | |
1399 | &submit_list); | |
1400 | count++; | |
1401 | } | |
1402 | ||
1403 | WARN_ON_ONCE(!wpc->ioend && !list_empty(&submit_list)); | |
1404 | WARN_ON_ONCE(!PageLocked(page)); | |
1405 | WARN_ON_ONCE(PageWriteback(page)); | |
1406 | ||
1407 | /* | |
1408 | * We cannot cancel the ioend directly here on error. We may have | |
1409 | * already set other pages under writeback and hence we have to run I/O | |
1410 | * completion to mark the error state of the pages under writeback | |
1411 | * appropriately. | |
1412 | */ | |
1413 | if (unlikely(error)) { | |
1414 | if (!count) { | |
1415 | /* | |
1416 | * If the current page hasn't been added to ioend, it | |
1417 | * won't be affected by I/O completions and we must | |
1418 | * discard and unlock it right here. | |
1419 | */ | |
1420 | if (wpc->ops->discard_page) | |
1421 | wpc->ops->discard_page(page); | |
1422 | ClearPageUptodate(page); | |
1423 | unlock_page(page); | |
1424 | goto done; | |
1425 | } | |
1426 | ||
1427 | /* | |
1428 | * If the page was not fully cleaned, we need to ensure that the | |
1429 | * higher layers come back to it correctly. That means we need | |
1430 | * to keep the page dirty, and for WB_SYNC_ALL writeback we need | |
1431 | * to ensure the PAGECACHE_TAG_TOWRITE index mark is not removed | |
1432 | * so another attempt to write this page in this writeback sweep | |
1433 | * will be made. | |
1434 | */ | |
1435 | set_page_writeback_keepwrite(page); | |
1436 | } else { | |
1437 | clear_page_dirty_for_io(page); | |
1438 | set_page_writeback(page); | |
1439 | } | |
1440 | ||
1441 | unlock_page(page); | |
1442 | ||
1443 | /* | |
1444 | * Preserve the original error if there was one, otherwise catch | |
1445 | * submission errors here and propagate into subsequent ioend | |
1446 | * submissions. | |
1447 | */ | |
1448 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1449 | int error2; | |
1450 | ||
1451 | list_del_init(&ioend->io_list); | |
1452 | error2 = iomap_submit_ioend(wpc, ioend, error); | |
1453 | if (error2 && !error) | |
1454 | error = error2; | |
1455 | } | |
1456 | ||
1457 | /* | |
1458 | * We can end up here with no error and nothing to write only if we race | |
1459 | * with a partial page truncate on a sub-page block sized filesystem. | |
1460 | */ | |
1461 | if (!count) | |
1462 | end_page_writeback(page); | |
1463 | done: | |
1464 | mapping_set_error(page->mapping, error); | |
1465 | return error; | |
1466 | } | |
1467 | ||
1468 | /* | |
1469 | * Write out a dirty page. | |
1470 | * | |
1471 | * For delalloc space on the page we need to allocate space and flush it. | |
1472 | * For unwritten space on the page we need to start the conversion to | |
1473 | * regular allocated space. | |
1474 | */ | |
1475 | static int | |
1476 | iomap_do_writepage(struct page *page, struct writeback_control *wbc, void *data) | |
1477 | { | |
1478 | struct iomap_writepage_ctx *wpc = data; | |
1479 | struct inode *inode = page->mapping->host; | |
1480 | pgoff_t end_index; | |
1481 | u64 end_offset; | |
1482 | loff_t offset; | |
1483 | ||
1ac99452 | 1484 | trace_iomap_writepage(inode, page_offset(page), PAGE_SIZE); |
598ecfba CH |
1485 | |
1486 | /* | |
1487 | * Refuse to write the page out if we are called from reclaim context. | |
1488 | * | |
1489 | * This avoids stack overflows when called from deeply used stacks in | |
1490 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1491 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
1492 | * | |
1493 | * This should never happen except in the case of a VM regression so | |
1494 | * warn about it. | |
1495 | */ | |
1496 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == | |
1497 | PF_MEMALLOC)) | |
1498 | goto redirty; | |
1499 | ||
1500 | /* | |
1501 | * Given that we do not allow direct reclaim to call us, we should | |
1502 | * never be called in a recursive filesystem reclaim context. | |
1503 | */ | |
1504 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) | |
1505 | goto redirty; | |
1506 | ||
1507 | /* | |
1508 | * Is this page beyond the end of the file? | |
1509 | * | |
1510 | * The page index is less than the end_index, adjust the end_offset | |
1511 | * to the highest offset that this page should represent. | |
1512 | * ----------------------------------------------------- | |
1513 | * | file mapping | <EOF> | | |
1514 | * ----------------------------------------------------- | |
1515 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1516 | * ^--------------------------------^----------|-------- | |
1517 | * | desired writeback range | see else | | |
1518 | * ---------------------------------^------------------| | |
1519 | */ | |
1520 | offset = i_size_read(inode); | |
1521 | end_index = offset >> PAGE_SHIFT; | |
1522 | if (page->index < end_index) | |
1523 | end_offset = (loff_t)(page->index + 1) << PAGE_SHIFT; | |
1524 | else { | |
1525 | /* | |
1526 | * Check whether the page to write out is beyond or straddles | |
1527 | * i_size or not. | |
1528 | * ------------------------------------------------------- | |
1529 | * | file mapping | <EOF> | | |
1530 | * ------------------------------------------------------- | |
1531 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1532 | * ^--------------------------------^-----------|--------- | |
1533 | * | | Straddles | | |
1534 | * ---------------------------------^-----------|--------| | |
1535 | */ | |
1536 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); | |
1537 | ||
1538 | /* | |
1539 | * Skip the page if it is fully outside i_size, e.g. due to a | |
1540 | * truncate operation that is in progress. We must redirty the | |
1541 | * page so that reclaim stops reclaiming it. Otherwise | |
1542 | * iomap_vm_releasepage() is called on it and gets confused. | |
1543 | * | |
1544 | * Note that the end_index is unsigned long, it would overflow | |
1545 | * if the given offset is greater than 16TB on 32-bit system | |
1546 | * and if we do check the page is fully outside i_size or not | |
1547 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1548 | * will be evaluated to 0. Hence this page will be redirtied | |
1549 | * and be written out repeatedly which would result in an | |
1550 | * infinite loop, the user program that perform this operation | |
1551 | * will hang. Instead, we can verify this situation by checking | |
1552 | * if the page to write is totally beyond the i_size or if it's | |
1553 | * offset is just equal to the EOF. | |
1554 | */ | |
1555 | if (page->index > end_index || | |
1556 | (page->index == end_index && offset_into_page == 0)) | |
1557 | goto redirty; | |
1558 | ||
1559 | /* | |
1560 | * The page straddles i_size. It must be zeroed out on each | |
1561 | * and every writepage invocation because it may be mmapped. | |
1562 | * "A file is mapped in multiples of the page size. For a file | |
1563 | * that is not a multiple of the page size, the remaining | |
1564 | * memory is zeroed when mapped, and writes to that region are | |
1565 | * not written out to the file." | |
1566 | */ | |
1567 | zero_user_segment(page, offset_into_page, PAGE_SIZE); | |
1568 | ||
1569 | /* Adjust the end_offset to the end of file */ | |
1570 | end_offset = offset; | |
1571 | } | |
1572 | ||
1573 | return iomap_writepage_map(wpc, wbc, inode, page, end_offset); | |
1574 | ||
1575 | redirty: | |
1576 | redirty_page_for_writepage(wbc, page); | |
1577 | unlock_page(page); | |
1578 | return 0; | |
1579 | } | |
1580 | ||
1581 | int | |
1582 | iomap_writepage(struct page *page, struct writeback_control *wbc, | |
1583 | struct iomap_writepage_ctx *wpc, | |
1584 | const struct iomap_writeback_ops *ops) | |
1585 | { | |
1586 | int ret; | |
1587 | ||
1588 | wpc->ops = ops; | |
1589 | ret = iomap_do_writepage(page, wbc, wpc); | |
1590 | if (!wpc->ioend) | |
1591 | return ret; | |
1592 | return iomap_submit_ioend(wpc, wpc->ioend, ret); | |
1593 | } | |
1594 | EXPORT_SYMBOL_GPL(iomap_writepage); | |
1595 | ||
1596 | int | |
1597 | iomap_writepages(struct address_space *mapping, struct writeback_control *wbc, | |
1598 | struct iomap_writepage_ctx *wpc, | |
1599 | const struct iomap_writeback_ops *ops) | |
1600 | { | |
1601 | int ret; | |
1602 | ||
1603 | wpc->ops = ops; | |
1604 | ret = write_cache_pages(mapping, wbc, iomap_do_writepage, wpc); | |
1605 | if (!wpc->ioend) | |
1606 | return ret; | |
1607 | return iomap_submit_ioend(wpc, wpc->ioend, ret); | |
1608 | } | |
1609 | EXPORT_SYMBOL_GPL(iomap_writepages); | |
1610 | ||
1611 | static int __init iomap_init(void) | |
1612 | { | |
1613 | return bioset_init(&iomap_ioend_bioset, 4 * (PAGE_SIZE / SECTOR_SIZE), | |
1614 | offsetof(struct iomap_ioend, io_inline_bio), | |
1615 | BIOSET_NEED_BVECS); | |
1616 | } | |
1617 | fs_initcall(iomap_init); |