]>
Commit | Line | Data |
---|---|---|
ae259a9c CH |
1 | /* |
2 | * Copyright (C) 2010 Red Hat, Inc. | |
3 | * Copyright (c) 2016 Christoph Hellwig. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | */ | |
14 | #include <linux/module.h> | |
15 | #include <linux/compiler.h> | |
16 | #include <linux/fs.h> | |
17 | #include <linux/iomap.h> | |
18 | #include <linux/uaccess.h> | |
19 | #include <linux/gfp.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/swap.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/uio.h> | |
25 | #include <linux/backing-dev.h> | |
26 | #include <linux/buffer_head.h> | |
ff6a9292 | 27 | #include <linux/task_io_accounting_ops.h> |
9a286f0e | 28 | #include <linux/dax.h> |
ae259a9c CH |
29 | #include "internal.h" |
30 | ||
ae259a9c CH |
31 | /* |
32 | * Execute a iomap write on a segment of the mapping that spans a | |
33 | * contiguous range of pages that have identical block mapping state. | |
34 | * | |
35 | * This avoids the need to map pages individually, do individual allocations | |
36 | * for each page and most importantly avoid the need for filesystem specific | |
37 | * locking per page. Instead, all the operations are amortised over the entire | |
38 | * range of pages. It is assumed that the filesystems will lock whatever | |
39 | * resources they require in the iomap_begin call, and release them in the | |
40 | * iomap_end call. | |
41 | */ | |
befb503c | 42 | loff_t |
ae259a9c | 43 | iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags, |
8ff6daa1 | 44 | const struct iomap_ops *ops, void *data, iomap_actor_t actor) |
ae259a9c CH |
45 | { |
46 | struct iomap iomap = { 0 }; | |
47 | loff_t written = 0, ret; | |
48 | ||
49 | /* | |
50 | * Need to map a range from start position for length bytes. This can | |
51 | * span multiple pages - it is only guaranteed to return a range of a | |
52 | * single type of pages (e.g. all into a hole, all mapped or all | |
53 | * unwritten). Failure at this point has nothing to undo. | |
54 | * | |
55 | * If allocation is required for this range, reserve the space now so | |
56 | * that the allocation is guaranteed to succeed later on. Once we copy | |
57 | * the data into the page cache pages, then we cannot fail otherwise we | |
58 | * expose transient stale data. If the reserve fails, we can safely | |
59 | * back out at this point as there is nothing to undo. | |
60 | */ | |
61 | ret = ops->iomap_begin(inode, pos, length, flags, &iomap); | |
62 | if (ret) | |
63 | return ret; | |
64 | if (WARN_ON(iomap.offset > pos)) | |
65 | return -EIO; | |
66 | ||
67 | /* | |
68 | * Cut down the length to the one actually provided by the filesystem, | |
69 | * as it might not be able to give us the whole size that we requested. | |
70 | */ | |
71 | if (iomap.offset + iomap.length < pos + length) | |
72 | length = iomap.offset + iomap.length - pos; | |
73 | ||
74 | /* | |
75 | * Now that we have guaranteed that the space allocation will succeed. | |
76 | * we can do the copy-in page by page without having to worry about | |
77 | * failures exposing transient data. | |
78 | */ | |
79 | written = actor(inode, pos, length, data, &iomap); | |
80 | ||
81 | /* | |
82 | * Now the data has been copied, commit the range we've copied. This | |
83 | * should not fail unless the filesystem has had a fatal error. | |
84 | */ | |
f20ac7ab CH |
85 | if (ops->iomap_end) { |
86 | ret = ops->iomap_end(inode, pos, length, | |
87 | written > 0 ? written : 0, | |
88 | flags, &iomap); | |
89 | } | |
ae259a9c CH |
90 | |
91 | return written ? written : ret; | |
92 | } | |
93 | ||
94 | static void | |
95 | iomap_write_failed(struct inode *inode, loff_t pos, unsigned len) | |
96 | { | |
97 | loff_t i_size = i_size_read(inode); | |
98 | ||
99 | /* | |
100 | * Only truncate newly allocated pages beyoned EOF, even if the | |
101 | * write started inside the existing inode size. | |
102 | */ | |
103 | if (pos + len > i_size) | |
104 | truncate_pagecache_range(inode, max(pos, i_size), pos + len); | |
105 | } | |
106 | ||
107 | static int | |
108 | iomap_write_begin(struct inode *inode, loff_t pos, unsigned len, unsigned flags, | |
109 | struct page **pagep, struct iomap *iomap) | |
110 | { | |
111 | pgoff_t index = pos >> PAGE_SHIFT; | |
112 | struct page *page; | |
113 | int status = 0; | |
114 | ||
115 | BUG_ON(pos + len > iomap->offset + iomap->length); | |
116 | ||
d1908f52 MH |
117 | if (fatal_signal_pending(current)) |
118 | return -EINTR; | |
119 | ||
ae259a9c CH |
120 | page = grab_cache_page_write_begin(inode->i_mapping, index, flags); |
121 | if (!page) | |
122 | return -ENOMEM; | |
123 | ||
124 | status = __block_write_begin_int(page, pos, len, NULL, iomap); | |
125 | if (unlikely(status)) { | |
126 | unlock_page(page); | |
127 | put_page(page); | |
128 | page = NULL; | |
129 | ||
130 | iomap_write_failed(inode, pos, len); | |
131 | } | |
132 | ||
133 | *pagep = page; | |
134 | return status; | |
135 | } | |
136 | ||
137 | static int | |
138 | iomap_write_end(struct inode *inode, loff_t pos, unsigned len, | |
139 | unsigned copied, struct page *page) | |
140 | { | |
141 | int ret; | |
142 | ||
143 | ret = generic_write_end(NULL, inode->i_mapping, pos, len, | |
144 | copied, page, NULL); | |
145 | if (ret < len) | |
146 | iomap_write_failed(inode, pos, len); | |
147 | return ret; | |
148 | } | |
149 | ||
150 | static loff_t | |
151 | iomap_write_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
152 | struct iomap *iomap) | |
153 | { | |
154 | struct iov_iter *i = data; | |
155 | long status = 0; | |
156 | ssize_t written = 0; | |
157 | unsigned int flags = AOP_FLAG_NOFS; | |
158 | ||
159 | /* | |
160 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
161 | */ | |
162 | if (!iter_is_iovec(i)) | |
163 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
164 | ||
165 | do { | |
166 | struct page *page; | |
167 | unsigned long offset; /* Offset into pagecache page */ | |
168 | unsigned long bytes; /* Bytes to write to page */ | |
169 | size_t copied; /* Bytes copied from user */ | |
170 | ||
171 | offset = (pos & (PAGE_SIZE - 1)); | |
172 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
173 | iov_iter_count(i)); | |
174 | again: | |
175 | if (bytes > length) | |
176 | bytes = length; | |
177 | ||
178 | /* | |
179 | * Bring in the user page that we will copy from _first_. | |
180 | * Otherwise there's a nasty deadlock on copying from the | |
181 | * same page as we're writing to, without it being marked | |
182 | * up-to-date. | |
183 | * | |
184 | * Not only is this an optimisation, but it is also required | |
185 | * to check that the address is actually valid, when atomic | |
186 | * usercopies are used, below. | |
187 | */ | |
188 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
189 | status = -EFAULT; | |
190 | break; | |
191 | } | |
192 | ||
193 | status = iomap_write_begin(inode, pos, bytes, flags, &page, | |
194 | iomap); | |
195 | if (unlikely(status)) | |
196 | break; | |
197 | ||
198 | if (mapping_writably_mapped(inode->i_mapping)) | |
199 | flush_dcache_page(page); | |
200 | ||
ae259a9c | 201 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
ae259a9c CH |
202 | |
203 | flush_dcache_page(page); | |
ae259a9c CH |
204 | |
205 | status = iomap_write_end(inode, pos, bytes, copied, page); | |
206 | if (unlikely(status < 0)) | |
207 | break; | |
208 | copied = status; | |
209 | ||
210 | cond_resched(); | |
211 | ||
212 | iov_iter_advance(i, copied); | |
213 | if (unlikely(copied == 0)) { | |
214 | /* | |
215 | * If we were unable to copy any data at all, we must | |
216 | * fall back to a single segment length write. | |
217 | * | |
218 | * If we didn't fallback here, we could livelock | |
219 | * because not all segments in the iov can be copied at | |
220 | * once without a pagefault. | |
221 | */ | |
222 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
223 | iov_iter_single_seg_count(i)); | |
224 | goto again; | |
225 | } | |
226 | pos += copied; | |
227 | written += copied; | |
228 | length -= copied; | |
229 | ||
230 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
231 | } while (iov_iter_count(i) && length); | |
232 | ||
233 | return written ? written : status; | |
234 | } | |
235 | ||
236 | ssize_t | |
237 | iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *iter, | |
8ff6daa1 | 238 | const struct iomap_ops *ops) |
ae259a9c CH |
239 | { |
240 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
241 | loff_t pos = iocb->ki_pos, ret = 0, written = 0; | |
242 | ||
243 | while (iov_iter_count(iter)) { | |
244 | ret = iomap_apply(inode, pos, iov_iter_count(iter), | |
245 | IOMAP_WRITE, ops, iter, iomap_write_actor); | |
246 | if (ret <= 0) | |
247 | break; | |
248 | pos += ret; | |
249 | written += ret; | |
250 | } | |
251 | ||
252 | return written ? written : ret; | |
253 | } | |
254 | EXPORT_SYMBOL_GPL(iomap_file_buffered_write); | |
255 | ||
5f4e5752 CH |
256 | static struct page * |
257 | __iomap_read_page(struct inode *inode, loff_t offset) | |
258 | { | |
259 | struct address_space *mapping = inode->i_mapping; | |
260 | struct page *page; | |
261 | ||
262 | page = read_mapping_page(mapping, offset >> PAGE_SHIFT, NULL); | |
263 | if (IS_ERR(page)) | |
264 | return page; | |
265 | if (!PageUptodate(page)) { | |
266 | put_page(page); | |
267 | return ERR_PTR(-EIO); | |
268 | } | |
269 | return page; | |
270 | } | |
271 | ||
272 | static loff_t | |
273 | iomap_dirty_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
274 | struct iomap *iomap) | |
275 | { | |
276 | long status = 0; | |
277 | ssize_t written = 0; | |
278 | ||
279 | do { | |
280 | struct page *page, *rpage; | |
281 | unsigned long offset; /* Offset into pagecache page */ | |
282 | unsigned long bytes; /* Bytes to write to page */ | |
283 | ||
284 | offset = (pos & (PAGE_SIZE - 1)); | |
285 | bytes = min_t(unsigned long, PAGE_SIZE - offset, length); | |
286 | ||
287 | rpage = __iomap_read_page(inode, pos); | |
288 | if (IS_ERR(rpage)) | |
289 | return PTR_ERR(rpage); | |
290 | ||
291 | status = iomap_write_begin(inode, pos, bytes, | |
292 | AOP_FLAG_NOFS | AOP_FLAG_UNINTERRUPTIBLE, | |
293 | &page, iomap); | |
294 | put_page(rpage); | |
295 | if (unlikely(status)) | |
296 | return status; | |
297 | ||
298 | WARN_ON_ONCE(!PageUptodate(page)); | |
299 | ||
300 | status = iomap_write_end(inode, pos, bytes, bytes, page); | |
301 | if (unlikely(status <= 0)) { | |
302 | if (WARN_ON_ONCE(status == 0)) | |
303 | return -EIO; | |
304 | return status; | |
305 | } | |
306 | ||
307 | cond_resched(); | |
308 | ||
309 | pos += status; | |
310 | written += status; | |
311 | length -= status; | |
312 | ||
313 | balance_dirty_pages_ratelimited(inode->i_mapping); | |
314 | } while (length); | |
315 | ||
316 | return written; | |
317 | } | |
318 | ||
319 | int | |
320 | iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len, | |
8ff6daa1 | 321 | const struct iomap_ops *ops) |
5f4e5752 CH |
322 | { |
323 | loff_t ret; | |
324 | ||
325 | while (len) { | |
326 | ret = iomap_apply(inode, pos, len, IOMAP_WRITE, ops, NULL, | |
327 | iomap_dirty_actor); | |
328 | if (ret <= 0) | |
329 | return ret; | |
330 | pos += ret; | |
331 | len -= ret; | |
332 | } | |
333 | ||
334 | return 0; | |
335 | } | |
336 | EXPORT_SYMBOL_GPL(iomap_file_dirty); | |
337 | ||
ae259a9c CH |
338 | static int iomap_zero(struct inode *inode, loff_t pos, unsigned offset, |
339 | unsigned bytes, struct iomap *iomap) | |
340 | { | |
341 | struct page *page; | |
342 | int status; | |
343 | ||
344 | status = iomap_write_begin(inode, pos, bytes, | |
345 | AOP_FLAG_UNINTERRUPTIBLE | AOP_FLAG_NOFS, &page, iomap); | |
346 | if (status) | |
347 | return status; | |
348 | ||
349 | zero_user(page, offset, bytes); | |
350 | mark_page_accessed(page); | |
351 | ||
352 | return iomap_write_end(inode, pos, bytes, bytes, page); | |
353 | } | |
354 | ||
9a286f0e CH |
355 | static int iomap_dax_zero(loff_t pos, unsigned offset, unsigned bytes, |
356 | struct iomap *iomap) | |
357 | { | |
358 | sector_t sector = iomap->blkno + | |
359 | (((pos & ~(PAGE_SIZE - 1)) - iomap->offset) >> 9); | |
360 | ||
361 | return __dax_zero_page_range(iomap->bdev, sector, offset, bytes); | |
362 | } | |
363 | ||
ae259a9c CH |
364 | static loff_t |
365 | iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t count, | |
366 | void *data, struct iomap *iomap) | |
367 | { | |
368 | bool *did_zero = data; | |
369 | loff_t written = 0; | |
370 | int status; | |
371 | ||
372 | /* already zeroed? we're done. */ | |
373 | if (iomap->type == IOMAP_HOLE || iomap->type == IOMAP_UNWRITTEN) | |
374 | return count; | |
375 | ||
376 | do { | |
377 | unsigned offset, bytes; | |
378 | ||
379 | offset = pos & (PAGE_SIZE - 1); /* Within page */ | |
380 | bytes = min_t(unsigned, PAGE_SIZE - offset, count); | |
381 | ||
9a286f0e CH |
382 | if (IS_DAX(inode)) |
383 | status = iomap_dax_zero(pos, offset, bytes, iomap); | |
384 | else | |
385 | status = iomap_zero(inode, pos, offset, bytes, iomap); | |
ae259a9c CH |
386 | if (status < 0) |
387 | return status; | |
388 | ||
389 | pos += bytes; | |
390 | count -= bytes; | |
391 | written += bytes; | |
392 | if (did_zero) | |
393 | *did_zero = true; | |
394 | } while (count > 0); | |
395 | ||
396 | return written; | |
397 | } | |
398 | ||
399 | int | |
400 | iomap_zero_range(struct inode *inode, loff_t pos, loff_t len, bool *did_zero, | |
8ff6daa1 | 401 | const struct iomap_ops *ops) |
ae259a9c CH |
402 | { |
403 | loff_t ret; | |
404 | ||
405 | while (len > 0) { | |
406 | ret = iomap_apply(inode, pos, len, IOMAP_ZERO, | |
407 | ops, did_zero, iomap_zero_range_actor); | |
408 | if (ret <= 0) | |
409 | return ret; | |
410 | ||
411 | pos += ret; | |
412 | len -= ret; | |
413 | } | |
414 | ||
415 | return 0; | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(iomap_zero_range); | |
418 | ||
419 | int | |
420 | iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero, | |
8ff6daa1 | 421 | const struct iomap_ops *ops) |
ae259a9c | 422 | { |
93407472 FF |
423 | unsigned int blocksize = i_blocksize(inode); |
424 | unsigned int off = pos & (blocksize - 1); | |
ae259a9c CH |
425 | |
426 | /* Block boundary? Nothing to do */ | |
427 | if (!off) | |
428 | return 0; | |
429 | return iomap_zero_range(inode, pos, blocksize - off, did_zero, ops); | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(iomap_truncate_page); | |
432 | ||
433 | static loff_t | |
434 | iomap_page_mkwrite_actor(struct inode *inode, loff_t pos, loff_t length, | |
435 | void *data, struct iomap *iomap) | |
436 | { | |
437 | struct page *page = data; | |
438 | int ret; | |
439 | ||
c663e29f | 440 | ret = __block_write_begin_int(page, pos, length, NULL, iomap); |
ae259a9c CH |
441 | if (ret) |
442 | return ret; | |
443 | ||
444 | block_commit_write(page, 0, length); | |
445 | return length; | |
446 | } | |
447 | ||
11bac800 | 448 | int iomap_page_mkwrite(struct vm_fault *vmf, const struct iomap_ops *ops) |
ae259a9c CH |
449 | { |
450 | struct page *page = vmf->page; | |
11bac800 | 451 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ae259a9c CH |
452 | unsigned long length; |
453 | loff_t offset, size; | |
454 | ssize_t ret; | |
455 | ||
456 | lock_page(page); | |
457 | size = i_size_read(inode); | |
458 | if ((page->mapping != inode->i_mapping) || | |
459 | (page_offset(page) > size)) { | |
460 | /* We overload EFAULT to mean page got truncated */ | |
461 | ret = -EFAULT; | |
462 | goto out_unlock; | |
463 | } | |
464 | ||
465 | /* page is wholly or partially inside EOF */ | |
466 | if (((page->index + 1) << PAGE_SHIFT) > size) | |
467 | length = size & ~PAGE_MASK; | |
468 | else | |
469 | length = PAGE_SIZE; | |
470 | ||
471 | offset = page_offset(page); | |
472 | while (length > 0) { | |
9484ab1b JK |
473 | ret = iomap_apply(inode, offset, length, |
474 | IOMAP_WRITE | IOMAP_FAULT, ops, page, | |
475 | iomap_page_mkwrite_actor); | |
ae259a9c CH |
476 | if (unlikely(ret <= 0)) |
477 | goto out_unlock; | |
478 | offset += ret; | |
479 | length -= ret; | |
480 | } | |
481 | ||
482 | set_page_dirty(page); | |
483 | wait_for_stable_page(page); | |
484 | return 0; | |
485 | out_unlock: | |
486 | unlock_page(page); | |
487 | return ret; | |
488 | } | |
489 | EXPORT_SYMBOL_GPL(iomap_page_mkwrite); | |
8be9f564 CH |
490 | |
491 | struct fiemap_ctx { | |
492 | struct fiemap_extent_info *fi; | |
493 | struct iomap prev; | |
494 | }; | |
495 | ||
496 | static int iomap_to_fiemap(struct fiemap_extent_info *fi, | |
497 | struct iomap *iomap, u32 flags) | |
498 | { | |
499 | switch (iomap->type) { | |
500 | case IOMAP_HOLE: | |
501 | /* skip holes */ | |
502 | return 0; | |
503 | case IOMAP_DELALLOC: | |
504 | flags |= FIEMAP_EXTENT_DELALLOC | FIEMAP_EXTENT_UNKNOWN; | |
505 | break; | |
506 | case IOMAP_UNWRITTEN: | |
507 | flags |= FIEMAP_EXTENT_UNWRITTEN; | |
508 | break; | |
509 | case IOMAP_MAPPED: | |
510 | break; | |
511 | } | |
512 | ||
17de0a9f CH |
513 | if (iomap->flags & IOMAP_F_MERGED) |
514 | flags |= FIEMAP_EXTENT_MERGED; | |
e43c460d DW |
515 | if (iomap->flags & IOMAP_F_SHARED) |
516 | flags |= FIEMAP_EXTENT_SHARED; | |
17de0a9f | 517 | |
8be9f564 CH |
518 | return fiemap_fill_next_extent(fi, iomap->offset, |
519 | iomap->blkno != IOMAP_NULL_BLOCK ? iomap->blkno << 9: 0, | |
17de0a9f | 520 | iomap->length, flags); |
8be9f564 CH |
521 | |
522 | } | |
523 | ||
524 | static loff_t | |
525 | iomap_fiemap_actor(struct inode *inode, loff_t pos, loff_t length, void *data, | |
526 | struct iomap *iomap) | |
527 | { | |
528 | struct fiemap_ctx *ctx = data; | |
529 | loff_t ret = length; | |
530 | ||
531 | if (iomap->type == IOMAP_HOLE) | |
532 | return length; | |
533 | ||
534 | ret = iomap_to_fiemap(ctx->fi, &ctx->prev, 0); | |
535 | ctx->prev = *iomap; | |
536 | switch (ret) { | |
537 | case 0: /* success */ | |
538 | return length; | |
539 | case 1: /* extent array full */ | |
540 | return 0; | |
541 | default: | |
542 | return ret; | |
543 | } | |
544 | } | |
545 | ||
546 | int iomap_fiemap(struct inode *inode, struct fiemap_extent_info *fi, | |
8ff6daa1 | 547 | loff_t start, loff_t len, const struct iomap_ops *ops) |
8be9f564 CH |
548 | { |
549 | struct fiemap_ctx ctx; | |
550 | loff_t ret; | |
551 | ||
552 | memset(&ctx, 0, sizeof(ctx)); | |
553 | ctx.fi = fi; | |
554 | ctx.prev.type = IOMAP_HOLE; | |
555 | ||
556 | ret = fiemap_check_flags(fi, FIEMAP_FLAG_SYNC); | |
557 | if (ret) | |
558 | return ret; | |
559 | ||
8896b8f6 DC |
560 | if (fi->fi_flags & FIEMAP_FLAG_SYNC) { |
561 | ret = filemap_write_and_wait(inode->i_mapping); | |
562 | if (ret) | |
563 | return ret; | |
564 | } | |
8be9f564 CH |
565 | |
566 | while (len > 0) { | |
d33fd776 | 567 | ret = iomap_apply(inode, start, len, IOMAP_REPORT, ops, &ctx, |
8be9f564 | 568 | iomap_fiemap_actor); |
ac2dc058 DC |
569 | /* inode with no (attribute) mapping will give ENOENT */ |
570 | if (ret == -ENOENT) | |
571 | break; | |
8be9f564 CH |
572 | if (ret < 0) |
573 | return ret; | |
574 | if (ret == 0) | |
575 | break; | |
576 | ||
577 | start += ret; | |
578 | len -= ret; | |
579 | } | |
580 | ||
581 | if (ctx.prev.type != IOMAP_HOLE) { | |
582 | ret = iomap_to_fiemap(fi, &ctx.prev, FIEMAP_EXTENT_LAST); | |
583 | if (ret < 0) | |
584 | return ret; | |
585 | } | |
586 | ||
587 | return 0; | |
588 | } | |
589 | EXPORT_SYMBOL_GPL(iomap_fiemap); | |
ff6a9292 CH |
590 | |
591 | /* | |
592 | * Private flags for iomap_dio, must not overlap with the public ones in | |
593 | * iomap.h: | |
594 | */ | |
595 | #define IOMAP_DIO_WRITE (1 << 30) | |
596 | #define IOMAP_DIO_DIRTY (1 << 31) | |
597 | ||
598 | struct iomap_dio { | |
599 | struct kiocb *iocb; | |
600 | iomap_dio_end_io_t *end_io; | |
601 | loff_t i_size; | |
602 | loff_t size; | |
603 | atomic_t ref; | |
604 | unsigned flags; | |
605 | int error; | |
606 | ||
607 | union { | |
608 | /* used during submission and for synchronous completion: */ | |
609 | struct { | |
610 | struct iov_iter *iter; | |
611 | struct task_struct *waiter; | |
612 | struct request_queue *last_queue; | |
613 | blk_qc_t cookie; | |
614 | } submit; | |
615 | ||
616 | /* used for aio completion: */ | |
617 | struct { | |
618 | struct work_struct work; | |
619 | } aio; | |
620 | }; | |
621 | }; | |
622 | ||
623 | static ssize_t iomap_dio_complete(struct iomap_dio *dio) | |
624 | { | |
625 | struct kiocb *iocb = dio->iocb; | |
626 | ssize_t ret; | |
627 | ||
628 | if (dio->end_io) { | |
629 | ret = dio->end_io(iocb, | |
630 | dio->error ? dio->error : dio->size, | |
631 | dio->flags); | |
632 | } else { | |
633 | ret = dio->error; | |
634 | } | |
635 | ||
636 | if (likely(!ret)) { | |
637 | ret = dio->size; | |
638 | /* check for short read */ | |
639 | if (iocb->ki_pos + ret > dio->i_size && | |
640 | !(dio->flags & IOMAP_DIO_WRITE)) | |
641 | ret = dio->i_size - iocb->ki_pos; | |
642 | iocb->ki_pos += ret; | |
643 | } | |
644 | ||
645 | inode_dio_end(file_inode(iocb->ki_filp)); | |
646 | kfree(dio); | |
647 | ||
648 | return ret; | |
649 | } | |
650 | ||
651 | static void iomap_dio_complete_work(struct work_struct *work) | |
652 | { | |
653 | struct iomap_dio *dio = container_of(work, struct iomap_dio, aio.work); | |
654 | struct kiocb *iocb = dio->iocb; | |
655 | bool is_write = (dio->flags & IOMAP_DIO_WRITE); | |
656 | ssize_t ret; | |
657 | ||
658 | ret = iomap_dio_complete(dio); | |
659 | if (is_write && ret > 0) | |
660 | ret = generic_write_sync(iocb, ret); | |
661 | iocb->ki_complete(iocb, ret, 0); | |
662 | } | |
663 | ||
664 | /* | |
665 | * Set an error in the dio if none is set yet. We have to use cmpxchg | |
666 | * as the submission context and the completion context(s) can race to | |
667 | * update the error. | |
668 | */ | |
669 | static inline void iomap_dio_set_error(struct iomap_dio *dio, int ret) | |
670 | { | |
671 | cmpxchg(&dio->error, 0, ret); | |
672 | } | |
673 | ||
674 | static void iomap_dio_bio_end_io(struct bio *bio) | |
675 | { | |
676 | struct iomap_dio *dio = bio->bi_private; | |
677 | bool should_dirty = (dio->flags & IOMAP_DIO_DIRTY); | |
678 | ||
679 | if (bio->bi_error) | |
680 | iomap_dio_set_error(dio, bio->bi_error); | |
681 | ||
682 | if (atomic_dec_and_test(&dio->ref)) { | |
683 | if (is_sync_kiocb(dio->iocb)) { | |
684 | struct task_struct *waiter = dio->submit.waiter; | |
685 | ||
686 | WRITE_ONCE(dio->submit.waiter, NULL); | |
687 | wake_up_process(waiter); | |
688 | } else if (dio->flags & IOMAP_DIO_WRITE) { | |
689 | struct inode *inode = file_inode(dio->iocb->ki_filp); | |
690 | ||
691 | INIT_WORK(&dio->aio.work, iomap_dio_complete_work); | |
692 | queue_work(inode->i_sb->s_dio_done_wq, &dio->aio.work); | |
693 | } else { | |
694 | iomap_dio_complete_work(&dio->aio.work); | |
695 | } | |
696 | } | |
697 | ||
698 | if (should_dirty) { | |
699 | bio_check_pages_dirty(bio); | |
700 | } else { | |
701 | struct bio_vec *bvec; | |
702 | int i; | |
703 | ||
704 | bio_for_each_segment_all(bvec, bio, i) | |
705 | put_page(bvec->bv_page); | |
706 | bio_put(bio); | |
707 | } | |
708 | } | |
709 | ||
710 | static blk_qc_t | |
711 | iomap_dio_zero(struct iomap_dio *dio, struct iomap *iomap, loff_t pos, | |
712 | unsigned len) | |
713 | { | |
714 | struct page *page = ZERO_PAGE(0); | |
715 | struct bio *bio; | |
716 | ||
717 | bio = bio_alloc(GFP_KERNEL, 1); | |
718 | bio->bi_bdev = iomap->bdev; | |
719 | bio->bi_iter.bi_sector = | |
720 | iomap->blkno + ((pos - iomap->offset) >> 9); | |
721 | bio->bi_private = dio; | |
722 | bio->bi_end_io = iomap_dio_bio_end_io; | |
723 | ||
724 | get_page(page); | |
725 | if (bio_add_page(bio, page, len, 0) != len) | |
726 | BUG(); | |
5cc60aee | 727 | bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC | REQ_IDLE); |
ff6a9292 CH |
728 | |
729 | atomic_inc(&dio->ref); | |
730 | return submit_bio(bio); | |
731 | } | |
732 | ||
733 | static loff_t | |
734 | iomap_dio_actor(struct inode *inode, loff_t pos, loff_t length, | |
735 | void *data, struct iomap *iomap) | |
736 | { | |
737 | struct iomap_dio *dio = data; | |
93407472 FF |
738 | unsigned int blkbits = blksize_bits(bdev_logical_block_size(iomap->bdev)); |
739 | unsigned int fs_block_size = i_blocksize(inode), pad; | |
740 | unsigned int align = iov_iter_alignment(dio->submit.iter); | |
ff6a9292 CH |
741 | struct iov_iter iter; |
742 | struct bio *bio; | |
743 | bool need_zeroout = false; | |
744 | int nr_pages, ret; | |
745 | ||
746 | if ((pos | length | align) & ((1 << blkbits) - 1)) | |
747 | return -EINVAL; | |
748 | ||
749 | switch (iomap->type) { | |
750 | case IOMAP_HOLE: | |
751 | if (WARN_ON_ONCE(dio->flags & IOMAP_DIO_WRITE)) | |
752 | return -EIO; | |
753 | /*FALLTHRU*/ | |
754 | case IOMAP_UNWRITTEN: | |
755 | if (!(dio->flags & IOMAP_DIO_WRITE)) { | |
756 | iov_iter_zero(length, dio->submit.iter); | |
757 | dio->size += length; | |
758 | return length; | |
759 | } | |
760 | dio->flags |= IOMAP_DIO_UNWRITTEN; | |
761 | need_zeroout = true; | |
762 | break; | |
763 | case IOMAP_MAPPED: | |
764 | if (iomap->flags & IOMAP_F_SHARED) | |
765 | dio->flags |= IOMAP_DIO_COW; | |
766 | if (iomap->flags & IOMAP_F_NEW) | |
767 | need_zeroout = true; | |
768 | break; | |
769 | default: | |
770 | WARN_ON_ONCE(1); | |
771 | return -EIO; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Operate on a partial iter trimmed to the extent we were called for. | |
776 | * We'll update the iter in the dio once we're done with this extent. | |
777 | */ | |
778 | iter = *dio->submit.iter; | |
779 | iov_iter_truncate(&iter, length); | |
780 | ||
781 | nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES); | |
782 | if (nr_pages <= 0) | |
783 | return nr_pages; | |
784 | ||
785 | if (need_zeroout) { | |
786 | /* zero out from the start of the block to the write offset */ | |
787 | pad = pos & (fs_block_size - 1); | |
788 | if (pad) | |
789 | iomap_dio_zero(dio, iomap, pos - pad, pad); | |
790 | } | |
791 | ||
792 | do { | |
793 | if (dio->error) | |
794 | return 0; | |
795 | ||
796 | bio = bio_alloc(GFP_KERNEL, nr_pages); | |
797 | bio->bi_bdev = iomap->bdev; | |
798 | bio->bi_iter.bi_sector = | |
799 | iomap->blkno + ((pos - iomap->offset) >> 9); | |
800 | bio->bi_private = dio; | |
801 | bio->bi_end_io = iomap_dio_bio_end_io; | |
802 | ||
803 | ret = bio_iov_iter_get_pages(bio, &iter); | |
804 | if (unlikely(ret)) { | |
805 | bio_put(bio); | |
806 | return ret; | |
807 | } | |
808 | ||
809 | if (dio->flags & IOMAP_DIO_WRITE) { | |
5cc60aee | 810 | bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC | REQ_IDLE); |
ff6a9292 CH |
811 | task_io_account_write(bio->bi_iter.bi_size); |
812 | } else { | |
813 | bio_set_op_attrs(bio, REQ_OP_READ, 0); | |
814 | if (dio->flags & IOMAP_DIO_DIRTY) | |
815 | bio_set_pages_dirty(bio); | |
816 | } | |
817 | ||
818 | dio->size += bio->bi_iter.bi_size; | |
819 | pos += bio->bi_iter.bi_size; | |
820 | ||
821 | nr_pages = iov_iter_npages(&iter, BIO_MAX_PAGES); | |
822 | ||
823 | atomic_inc(&dio->ref); | |
824 | ||
825 | dio->submit.last_queue = bdev_get_queue(iomap->bdev); | |
826 | dio->submit.cookie = submit_bio(bio); | |
827 | } while (nr_pages); | |
828 | ||
829 | if (need_zeroout) { | |
830 | /* zero out from the end of the write to the end of the block */ | |
831 | pad = pos & (fs_block_size - 1); | |
832 | if (pad) | |
833 | iomap_dio_zero(dio, iomap, pos, fs_block_size - pad); | |
834 | } | |
835 | ||
836 | iov_iter_advance(dio->submit.iter, length); | |
837 | return length; | |
838 | } | |
839 | ||
840 | ssize_t | |
8ff6daa1 CH |
841 | iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter, |
842 | const struct iomap_ops *ops, iomap_dio_end_io_t end_io) | |
ff6a9292 CH |
843 | { |
844 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
845 | struct inode *inode = file_inode(iocb->ki_filp); | |
846 | size_t count = iov_iter_count(iter); | |
847 | loff_t pos = iocb->ki_pos, end = iocb->ki_pos + count - 1, ret = 0; | |
848 | unsigned int flags = IOMAP_DIRECT; | |
849 | struct blk_plug plug; | |
850 | struct iomap_dio *dio; | |
851 | ||
852 | lockdep_assert_held(&inode->i_rwsem); | |
853 | ||
854 | if (!count) | |
855 | return 0; | |
856 | ||
857 | dio = kmalloc(sizeof(*dio), GFP_KERNEL); | |
858 | if (!dio) | |
859 | return -ENOMEM; | |
860 | ||
861 | dio->iocb = iocb; | |
862 | atomic_set(&dio->ref, 1); | |
863 | dio->size = 0; | |
864 | dio->i_size = i_size_read(inode); | |
865 | dio->end_io = end_io; | |
866 | dio->error = 0; | |
867 | dio->flags = 0; | |
868 | ||
869 | dio->submit.iter = iter; | |
870 | if (is_sync_kiocb(iocb)) { | |
871 | dio->submit.waiter = current; | |
872 | dio->submit.cookie = BLK_QC_T_NONE; | |
873 | dio->submit.last_queue = NULL; | |
874 | } | |
875 | ||
876 | if (iov_iter_rw(iter) == READ) { | |
877 | if (pos >= dio->i_size) | |
878 | goto out_free_dio; | |
879 | ||
880 | if (iter->type == ITER_IOVEC) | |
881 | dio->flags |= IOMAP_DIO_DIRTY; | |
882 | } else { | |
883 | dio->flags |= IOMAP_DIO_WRITE; | |
884 | flags |= IOMAP_WRITE; | |
885 | } | |
886 | ||
887 | if (mapping->nrpages) { | |
888 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, end); | |
889 | if (ret) | |
890 | goto out_free_dio; | |
891 | ||
892 | ret = invalidate_inode_pages2_range(mapping, | |
893 | iocb->ki_pos >> PAGE_SHIFT, end >> PAGE_SHIFT); | |
894 | WARN_ON_ONCE(ret); | |
895 | ret = 0; | |
896 | } | |
897 | ||
898 | inode_dio_begin(inode); | |
899 | ||
900 | blk_start_plug(&plug); | |
901 | do { | |
902 | ret = iomap_apply(inode, pos, count, flags, ops, dio, | |
903 | iomap_dio_actor); | |
904 | if (ret <= 0) { | |
905 | /* magic error code to fall back to buffered I/O */ | |
906 | if (ret == -ENOTBLK) | |
907 | ret = 0; | |
908 | break; | |
909 | } | |
910 | pos += ret; | |
911 | } while ((count = iov_iter_count(iter)) > 0); | |
912 | blk_finish_plug(&plug); | |
913 | ||
914 | if (ret < 0) | |
915 | iomap_dio_set_error(dio, ret); | |
916 | ||
917 | if (ret >= 0 && iov_iter_rw(iter) == WRITE && !is_sync_kiocb(iocb) && | |
918 | !inode->i_sb->s_dio_done_wq) { | |
919 | ret = sb_init_dio_done_wq(inode->i_sb); | |
920 | if (ret < 0) | |
921 | iomap_dio_set_error(dio, ret); | |
922 | } | |
923 | ||
924 | if (!atomic_dec_and_test(&dio->ref)) { | |
925 | if (!is_sync_kiocb(iocb)) | |
926 | return -EIOCBQUEUED; | |
927 | ||
928 | for (;;) { | |
929 | set_current_state(TASK_UNINTERRUPTIBLE); | |
930 | if (!READ_ONCE(dio->submit.waiter)) | |
931 | break; | |
932 | ||
933 | if (!(iocb->ki_flags & IOCB_HIPRI) || | |
934 | !dio->submit.last_queue || | |
5cc60aee LT |
935 | !blk_mq_poll(dio->submit.last_queue, |
936 | dio->submit.cookie)) | |
ff6a9292 CH |
937 | io_schedule(); |
938 | } | |
939 | __set_current_state(TASK_RUNNING); | |
940 | } | |
941 | ||
942 | /* | |
943 | * Try again to invalidate clean pages which might have been cached by | |
944 | * non-direct readahead, or faulted in by get_user_pages() if the source | |
945 | * of the write was an mmap'ed region of the file we're writing. Either | |
946 | * one is a pretty crazy thing to do, so we don't support it 100%. If | |
947 | * this invalidation fails, tough, the write still worked... | |
948 | */ | |
949 | if (iov_iter_rw(iter) == WRITE && mapping->nrpages) { | |
950 | ret = invalidate_inode_pages2_range(mapping, | |
951 | iocb->ki_pos >> PAGE_SHIFT, end >> PAGE_SHIFT); | |
952 | WARN_ON_ONCE(ret); | |
953 | } | |
954 | ||
955 | return iomap_dio_complete(dio); | |
956 | ||
957 | out_free_dio: | |
958 | kfree(dio); | |
959 | return ret; | |
960 | } | |
961 | EXPORT_SYMBOL_GPL(iomap_dio_rw); |