Commit | Line | Data |
---|---|---|
cd82cca7 CH |
1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* | |
3 | * Copyright (C) 1991, 1992 Linus Torvalds | |
4 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * Copyright (C) 2016 - 2020 Christoph Hellwig | |
6 | */ | |
7 | #include <linux/init.h> | |
8 | #include <linux/mm.h> | |
9 | #include <linux/blkdev.h> | |
10 | #include <linux/buffer_head.h> | |
11 | #include <linux/mpage.h> | |
12 | #include <linux/uio.h> | |
13 | #include <linux/namei.h> | |
14 | #include <linux/task_io_accounting_ops.h> | |
15 | #include <linux/falloc.h> | |
16 | #include <linux/suspend.h> | |
f278eb3d | 17 | #include <linux/fs.h> |
8581fd40 | 18 | #include <linux/module.h> |
cd82cca7 CH |
19 | #include "blk.h" |
20 | ||
fac7c6d5 | 21 | static inline struct inode *bdev_file_inode(struct file *file) |
cd82cca7 CH |
22 | { |
23 | return file->f_mapping->host; | |
24 | } | |
25 | ||
26 | static int blkdev_get_block(struct inode *inode, sector_t iblock, | |
27 | struct buffer_head *bh, int create) | |
28 | { | |
29 | bh->b_bdev = I_BDEV(inode); | |
30 | bh->b_blocknr = iblock; | |
31 | set_buffer_mapped(bh); | |
32 | return 0; | |
33 | } | |
34 | ||
35 | static unsigned int dio_bio_write_op(struct kiocb *iocb) | |
36 | { | |
37 | unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE; | |
38 | ||
39 | /* avoid the need for a I/O completion work item */ | |
40 | if (iocb->ki_flags & IOCB_DSYNC) | |
41 | op |= REQ_FUA; | |
42 | return op; | |
43 | } | |
44 | ||
45 | #define DIO_INLINE_BIO_VECS 4 | |
46 | ||
47 | static void blkdev_bio_end_io_simple(struct bio *bio) | |
48 | { | |
49 | struct task_struct *waiter = bio->bi_private; | |
50 | ||
51 | WRITE_ONCE(bio->bi_private, NULL); | |
52 | blk_wake_io_task(waiter); | |
53 | } | |
54 | ||
55 | static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb, | |
56 | struct iov_iter *iter, unsigned int nr_pages) | |
57 | { | |
fac7c6d5 | 58 | struct block_device *bdev = iocb->ki_filp->private_data; |
cd82cca7 CH |
59 | struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs; |
60 | loff_t pos = iocb->ki_pos; | |
61 | bool should_dirty = false; | |
62 | struct bio bio; | |
63 | ssize_t ret; | |
cd82cca7 CH |
64 | |
65 | if ((pos | iov_iter_alignment(iter)) & | |
66 | (bdev_logical_block_size(bdev) - 1)) | |
67 | return -EINVAL; | |
68 | ||
69 | if (nr_pages <= DIO_INLINE_BIO_VECS) | |
70 | vecs = inline_vecs; | |
71 | else { | |
72 | vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec), | |
73 | GFP_KERNEL); | |
74 | if (!vecs) | |
75 | return -ENOMEM; | |
76 | } | |
77 | ||
49add496 CH |
78 | if (iov_iter_rw(iter) == READ) { |
79 | bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ); | |
80 | if (iter_is_iovec(iter)) | |
81 | should_dirty = true; | |
82 | } else { | |
83 | bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb)); | |
84 | } | |
6549a874 | 85 | bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT; |
cd82cca7 CH |
86 | bio.bi_private = current; |
87 | bio.bi_end_io = blkdev_bio_end_io_simple; | |
88 | bio.bi_ioprio = iocb->ki_ioprio; | |
89 | ||
90 | ret = bio_iov_iter_get_pages(&bio, iter); | |
91 | if (unlikely(ret)) | |
92 | goto out; | |
93 | ret = bio.bi_iter.bi_size; | |
94 | ||
49add496 | 95 | if (iov_iter_rw(iter) == WRITE) |
cd82cca7 | 96 | task_io_account_write(ret); |
49add496 | 97 | |
cd82cca7 CH |
98 | if (iocb->ki_flags & IOCB_NOWAIT) |
99 | bio.bi_opf |= REQ_NOWAIT; | |
100 | if (iocb->ki_flags & IOCB_HIPRI) | |
101 | bio_set_polled(&bio, iocb); | |
102 | ||
3e08773c | 103 | submit_bio(&bio); |
cd82cca7 CH |
104 | for (;;) { |
105 | set_current_state(TASK_UNINTERRUPTIBLE); | |
106 | if (!READ_ONCE(bio.bi_private)) | |
107 | break; | |
5a72e899 | 108 | if (!(iocb->ki_flags & IOCB_HIPRI) || !bio_poll(&bio, NULL, 0)) |
cd82cca7 CH |
109 | blk_io_schedule(); |
110 | } | |
111 | __set_current_state(TASK_RUNNING); | |
112 | ||
113 | bio_release_pages(&bio, should_dirty); | |
114 | if (unlikely(bio.bi_status)) | |
115 | ret = blk_status_to_errno(bio.bi_status); | |
116 | ||
117 | out: | |
118 | if (vecs != inline_vecs) | |
119 | kfree(vecs); | |
120 | ||
121 | bio_uninit(&bio); | |
122 | ||
123 | return ret; | |
124 | } | |
125 | ||
09ce8744 | 126 | enum { |
e71aa913 PB |
127 | DIO_SHOULD_DIRTY = 1, |
128 | DIO_IS_SYNC = 2, | |
09ce8744 JA |
129 | }; |
130 | ||
cd82cca7 CH |
131 | struct blkdev_dio { |
132 | union { | |
133 | struct kiocb *iocb; | |
134 | struct task_struct *waiter; | |
135 | }; | |
136 | size_t size; | |
137 | atomic_t ref; | |
09ce8744 | 138 | unsigned int flags; |
6155631a | 139 | struct bio bio ____cacheline_aligned_in_smp; |
cd82cca7 CH |
140 | }; |
141 | ||
142 | static struct bio_set blkdev_dio_pool; | |
143 | ||
cd82cca7 CH |
144 | static void blkdev_bio_end_io(struct bio *bio) |
145 | { | |
146 | struct blkdev_dio *dio = bio->bi_private; | |
09ce8744 | 147 | bool should_dirty = dio->flags & DIO_SHOULD_DIRTY; |
cd82cca7 CH |
148 | |
149 | if (bio->bi_status && !dio->bio.bi_status) | |
150 | dio->bio.bi_status = bio->bi_status; | |
151 | ||
e71aa913 | 152 | if (atomic_dec_and_test(&dio->ref)) { |
09ce8744 | 153 | if (!(dio->flags & DIO_IS_SYNC)) { |
cd82cca7 CH |
154 | struct kiocb *iocb = dio->iocb; |
155 | ssize_t ret; | |
156 | ||
3e08773c CH |
157 | WRITE_ONCE(iocb->private, NULL); |
158 | ||
cd82cca7 CH |
159 | if (likely(!dio->bio.bi_status)) { |
160 | ret = dio->size; | |
161 | iocb->ki_pos += ret; | |
162 | } else { | |
163 | ret = blk_status_to_errno(dio->bio.bi_status); | |
164 | } | |
165 | ||
6b19b766 | 166 | dio->iocb->ki_complete(iocb, ret); |
e71aa913 | 167 | bio_put(&dio->bio); |
cd82cca7 CH |
168 | } else { |
169 | struct task_struct *waiter = dio->waiter; | |
170 | ||
171 | WRITE_ONCE(dio->waiter, NULL); | |
172 | blk_wake_io_task(waiter); | |
173 | } | |
174 | } | |
175 | ||
176 | if (should_dirty) { | |
177 | bio_check_pages_dirty(bio); | |
178 | } else { | |
179 | bio_release_pages(bio, false); | |
180 | bio_put(bio); | |
181 | } | |
182 | } | |
183 | ||
184 | static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, | |
185 | unsigned int nr_pages) | |
186 | { | |
fac7c6d5 | 187 | struct block_device *bdev = iocb->ki_filp->private_data; |
cd82cca7 CH |
188 | struct blk_plug plug; |
189 | struct blkdev_dio *dio; | |
190 | struct bio *bio; | |
cd82cca7 | 191 | bool is_read = (iov_iter_rw(iter) == READ), is_sync; |
b77c88c2 | 192 | unsigned int opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); |
cd82cca7 | 193 | loff_t pos = iocb->ki_pos; |
cd82cca7 CH |
194 | int ret = 0; |
195 | ||
196 | if ((pos | iov_iter_alignment(iter)) & | |
197 | (bdev_logical_block_size(bdev) - 1)) | |
198 | return -EINVAL; | |
199 | ||
b77c88c2 | 200 | bio = bio_alloc_kiocb(iocb, bdev, nr_pages, opf, &blkdev_dio_pool); |
cd82cca7 CH |
201 | |
202 | dio = container_of(bio, struct blkdev_dio, bio); | |
e71aa913 PB |
203 | atomic_set(&dio->ref, 1); |
204 | /* | |
205 | * Grab an extra reference to ensure the dio structure which is embedded | |
206 | * into the first bio stays around. | |
207 | */ | |
208 | bio_get(bio); | |
209 | ||
09ce8744 JA |
210 | is_sync = is_sync_kiocb(iocb); |
211 | if (is_sync) { | |
212 | dio->flags = DIO_IS_SYNC; | |
cd82cca7 | 213 | dio->waiter = current; |
cd82cca7 | 214 | } else { |
09ce8744 | 215 | dio->flags = 0; |
cd82cca7 CH |
216 | dio->iocb = iocb; |
217 | } | |
218 | ||
219 | dio->size = 0; | |
09ce8744 JA |
220 | if (is_read && iter_is_iovec(iter)) |
221 | dio->flags |= DIO_SHOULD_DIRTY; | |
cd82cca7 | 222 | |
25d207dc | 223 | blk_start_plug(&plug); |
cd82cca7 CH |
224 | |
225 | for (;;) { | |
6549a874 | 226 | bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; |
cd82cca7 CH |
227 | bio->bi_private = dio; |
228 | bio->bi_end_io = blkdev_bio_end_io; | |
229 | bio->bi_ioprio = iocb->ki_ioprio; | |
230 | ||
231 | ret = bio_iov_iter_get_pages(bio, iter); | |
232 | if (unlikely(ret)) { | |
233 | bio->bi_status = BLK_STS_IOERR; | |
234 | bio_endio(bio); | |
235 | break; | |
236 | } | |
237 | ||
238 | if (is_read) { | |
09ce8744 | 239 | if (dio->flags & DIO_SHOULD_DIRTY) |
cd82cca7 CH |
240 | bio_set_pages_dirty(bio); |
241 | } else { | |
cd82cca7 CH |
242 | task_io_account_write(bio->bi_iter.bi_size); |
243 | } | |
244 | if (iocb->ki_flags & IOCB_NOWAIT) | |
245 | bio->bi_opf |= REQ_NOWAIT; | |
246 | ||
247 | dio->size += bio->bi_iter.bi_size; | |
248 | pos += bio->bi_iter.bi_size; | |
249 | ||
250 | nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS); | |
251 | if (!nr_pages) { | |
3e08773c | 252 | submit_bio(bio); |
cd82cca7 CH |
253 | break; |
254 | } | |
e71aa913 | 255 | atomic_inc(&dio->ref); |
cd82cca7 | 256 | submit_bio(bio); |
07888c66 | 257 | bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL); |
cd82cca7 CH |
258 | } |
259 | ||
25d207dc | 260 | blk_finish_plug(&plug); |
cd82cca7 CH |
261 | |
262 | if (!is_sync) | |
263 | return -EIOCBQUEUED; | |
264 | ||
265 | for (;;) { | |
266 | set_current_state(TASK_UNINTERRUPTIBLE); | |
267 | if (!READ_ONCE(dio->waiter)) | |
268 | break; | |
25d207dc | 269 | blk_io_schedule(); |
cd82cca7 CH |
270 | } |
271 | __set_current_state(TASK_RUNNING); | |
272 | ||
273 | if (!ret) | |
274 | ret = blk_status_to_errno(dio->bio.bi_status); | |
275 | if (likely(!ret)) | |
276 | ret = dio->size; | |
277 | ||
278 | bio_put(&dio->bio); | |
279 | return ret; | |
280 | } | |
281 | ||
54a88eb8 PB |
282 | static void blkdev_bio_end_io_async(struct bio *bio) |
283 | { | |
284 | struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio); | |
285 | struct kiocb *iocb = dio->iocb; | |
286 | ssize_t ret; | |
287 | ||
bb49c6fa SG |
288 | WRITE_ONCE(iocb->private, NULL); |
289 | ||
54a88eb8 PB |
290 | if (likely(!bio->bi_status)) { |
291 | ret = dio->size; | |
292 | iocb->ki_pos += ret; | |
293 | } else { | |
294 | ret = blk_status_to_errno(bio->bi_status); | |
295 | } | |
296 | ||
b6773cdb | 297 | iocb->ki_complete(iocb, ret); |
54a88eb8 PB |
298 | |
299 | if (dio->flags & DIO_SHOULD_DIRTY) { | |
300 | bio_check_pages_dirty(bio); | |
301 | } else { | |
302 | bio_release_pages(bio, false); | |
303 | bio_put(bio); | |
304 | } | |
305 | } | |
306 | ||
307 | static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, | |
308 | struct iov_iter *iter, | |
309 | unsigned int nr_pages) | |
310 | { | |
311 | struct block_device *bdev = iocb->ki_filp->private_data; | |
b77c88c2 CH |
312 | bool is_read = iov_iter_rw(iter) == READ; |
313 | unsigned int opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb); | |
54a88eb8 PB |
314 | struct blkdev_dio *dio; |
315 | struct bio *bio; | |
316 | loff_t pos = iocb->ki_pos; | |
317 | int ret = 0; | |
318 | ||
319 | if ((pos | iov_iter_alignment(iter)) & | |
320 | (bdev_logical_block_size(bdev) - 1)) | |
321 | return -EINVAL; | |
322 | ||
b77c88c2 | 323 | bio = bio_alloc_kiocb(iocb, bdev, nr_pages, opf, &blkdev_dio_pool); |
54a88eb8 PB |
324 | dio = container_of(bio, struct blkdev_dio, bio); |
325 | dio->flags = 0; | |
326 | dio->iocb = iocb; | |
54a88eb8 | 327 | bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT; |
54a88eb8 PB |
328 | bio->bi_end_io = blkdev_bio_end_io_async; |
329 | bio->bi_ioprio = iocb->ki_ioprio; | |
330 | ||
1bb6b810 PB |
331 | if (iov_iter_is_bvec(iter)) { |
332 | /* | |
333 | * Users don't rely on the iterator being in any particular | |
334 | * state for async I/O returning -EIOCBQUEUED, hence we can | |
335 | * avoid expensive iov_iter_advance(). Bypass | |
336 | * bio_iov_iter_get_pages() and set the bvec directly. | |
337 | */ | |
338 | bio_iov_bvec_set(bio, iter); | |
339 | } else { | |
340 | ret = bio_iov_iter_get_pages(bio, iter); | |
341 | if (unlikely(ret)) { | |
75feae73 | 342 | bio_put(bio); |
1bb6b810 PB |
343 | return ret; |
344 | } | |
54a88eb8 PB |
345 | } |
346 | dio->size = bio->bi_iter.bi_size; | |
347 | ||
b77c88c2 | 348 | if (is_read) { |
54a88eb8 PB |
349 | if (iter_is_iovec(iter)) { |
350 | dio->flags |= DIO_SHOULD_DIRTY; | |
351 | bio_set_pages_dirty(bio); | |
352 | } | |
353 | } else { | |
54a88eb8 PB |
354 | task_io_account_write(bio->bi_iter.bi_size); |
355 | } | |
356 | ||
54a88eb8 | 357 | if (iocb->ki_flags & IOCB_HIPRI) { |
842e39b0 | 358 | bio->bi_opf |= REQ_POLLED | REQ_NOWAIT; |
54a88eb8 PB |
359 | submit_bio(bio); |
360 | WRITE_ONCE(iocb->private, bio); | |
361 | } else { | |
842e39b0 PB |
362 | if (iocb->ki_flags & IOCB_NOWAIT) |
363 | bio->bi_opf |= REQ_NOWAIT; | |
54a88eb8 PB |
364 | submit_bio(bio); |
365 | } | |
366 | return -EIOCBQUEUED; | |
367 | } | |
368 | ||
cd82cca7 CH |
369 | static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
370 | { | |
371 | unsigned int nr_pages; | |
372 | ||
373 | if (!iov_iter_count(iter)) | |
374 | return 0; | |
375 | ||
376 | nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1); | |
54a88eb8 PB |
377 | if (likely(nr_pages <= BIO_MAX_VECS)) { |
378 | if (is_sync_kiocb(iocb)) | |
379 | return __blkdev_direct_IO_simple(iocb, iter, nr_pages); | |
380 | return __blkdev_direct_IO_async(iocb, iter, nr_pages); | |
381 | } | |
cd82cca7 CH |
382 | return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages)); |
383 | } | |
384 | ||
385 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) | |
386 | { | |
387 | return block_write_full_page(page, blkdev_get_block, wbc); | |
388 | } | |
389 | ||
390 | static int blkdev_readpage(struct file * file, struct page * page) | |
391 | { | |
392 | return block_read_full_page(page, blkdev_get_block); | |
393 | } | |
394 | ||
395 | static void blkdev_readahead(struct readahead_control *rac) | |
396 | { | |
397 | mpage_readahead(rac, blkdev_get_block); | |
398 | } | |
399 | ||
400 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, | |
401 | loff_t pos, unsigned len, unsigned flags, struct page **pagep, | |
402 | void **fsdata) | |
403 | { | |
404 | return block_write_begin(mapping, pos, len, flags, pagep, | |
405 | blkdev_get_block); | |
406 | } | |
407 | ||
408 | static int blkdev_write_end(struct file *file, struct address_space *mapping, | |
409 | loff_t pos, unsigned len, unsigned copied, struct page *page, | |
410 | void *fsdata) | |
411 | { | |
412 | int ret; | |
413 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
414 | ||
415 | unlock_page(page); | |
416 | put_page(page); | |
417 | ||
418 | return ret; | |
419 | } | |
420 | ||
421 | static int blkdev_writepages(struct address_space *mapping, | |
422 | struct writeback_control *wbc) | |
423 | { | |
424 | return generic_writepages(mapping, wbc); | |
425 | } | |
426 | ||
427 | const struct address_space_operations def_blk_aops = { | |
e621900a | 428 | .dirty_folio = block_dirty_folio, |
7ba13abb | 429 | .invalidate_folio = block_invalidate_folio, |
cd82cca7 CH |
430 | .readpage = blkdev_readpage, |
431 | .readahead = blkdev_readahead, | |
432 | .writepage = blkdev_writepage, | |
433 | .write_begin = blkdev_write_begin, | |
434 | .write_end = blkdev_write_end, | |
435 | .writepages = blkdev_writepages, | |
436 | .direct_IO = blkdev_direct_IO, | |
437 | .migratepage = buffer_migrate_page_norefs, | |
438 | .is_dirty_writeback = buffer_check_dirty_writeback, | |
439 | }; | |
440 | ||
441 | /* | |
442 | * for a block special file file_inode(file)->i_size is zero | |
443 | * so we compute the size by hand (just as in block_read/write above) | |
444 | */ | |
445 | static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence) | |
446 | { | |
447 | struct inode *bd_inode = bdev_file_inode(file); | |
448 | loff_t retval; | |
449 | ||
450 | inode_lock(bd_inode); | |
451 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); | |
452 | inode_unlock(bd_inode); | |
453 | return retval; | |
454 | } | |
455 | ||
456 | static int blkdev_fsync(struct file *filp, loff_t start, loff_t end, | |
457 | int datasync) | |
458 | { | |
fac7c6d5 | 459 | struct block_device *bdev = filp->private_data; |
cd82cca7 CH |
460 | int error; |
461 | ||
462 | error = file_write_and_wait_range(filp, start, end); | |
463 | if (error) | |
464 | return error; | |
465 | ||
466 | /* | |
467 | * There is no need to serialise calls to blkdev_issue_flush with | |
468 | * i_mutex and doing so causes performance issues with concurrent | |
469 | * O_SYNC writers to a block device. | |
470 | */ | |
471 | error = blkdev_issue_flush(bdev); | |
472 | if (error == -EOPNOTSUPP) | |
473 | error = 0; | |
474 | ||
475 | return error; | |
476 | } | |
477 | ||
478 | static int blkdev_open(struct inode *inode, struct file *filp) | |
479 | { | |
480 | struct block_device *bdev; | |
481 | ||
482 | /* | |
483 | * Preserve backwards compatibility and allow large file access | |
484 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
485 | * binary needs it. We might want to drop this workaround | |
486 | * during an unstable branch. | |
487 | */ | |
488 | filp->f_flags |= O_LARGEFILE; | |
489 | filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC; | |
490 | ||
491 | if (filp->f_flags & O_NDELAY) | |
492 | filp->f_mode |= FMODE_NDELAY; | |
493 | if (filp->f_flags & O_EXCL) | |
494 | filp->f_mode |= FMODE_EXCL; | |
495 | if ((filp->f_flags & O_ACCMODE) == 3) | |
496 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
497 | ||
498 | bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp); | |
499 | if (IS_ERR(bdev)) | |
500 | return PTR_ERR(bdev); | |
fac7c6d5 PB |
501 | |
502 | filp->private_data = bdev; | |
cd82cca7 CH |
503 | filp->f_mapping = bdev->bd_inode->i_mapping; |
504 | filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping); | |
505 | return 0; | |
506 | } | |
507 | ||
508 | static int blkdev_close(struct inode *inode, struct file *filp) | |
509 | { | |
fac7c6d5 | 510 | struct block_device *bdev = filp->private_data; |
cd82cca7 CH |
511 | |
512 | blkdev_put(bdev, filp->f_mode); | |
513 | return 0; | |
514 | } | |
515 | ||
cd82cca7 CH |
516 | /* |
517 | * Write data to the block device. Only intended for the block device itself | |
518 | * and the raw driver which basically is a fake block device. | |
519 | * | |
520 | * Does not take i_mutex for the write and thus is not for general purpose | |
521 | * use. | |
522 | */ | |
523 | static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
524 | { | |
fac7c6d5 PB |
525 | struct block_device *bdev = iocb->ki_filp->private_data; |
526 | struct inode *bd_inode = bdev->bd_inode; | |
138c1a38 | 527 | loff_t size = bdev_nr_bytes(bdev); |
cd82cca7 CH |
528 | struct blk_plug plug; |
529 | size_t shorted = 0; | |
530 | ssize_t ret; | |
531 | ||
fac7c6d5 | 532 | if (bdev_read_only(bdev)) |
cd82cca7 CH |
533 | return -EPERM; |
534 | ||
535 | if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev)) | |
536 | return -ETXTBSY; | |
537 | ||
538 | if (!iov_iter_count(from)) | |
539 | return 0; | |
540 | ||
541 | if (iocb->ki_pos >= size) | |
542 | return -ENOSPC; | |
543 | ||
544 | if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT) | |
545 | return -EOPNOTSUPP; | |
546 | ||
547 | size -= iocb->ki_pos; | |
548 | if (iov_iter_count(from) > size) { | |
549 | shorted = iov_iter_count(from) - size; | |
550 | iov_iter_truncate(from, size); | |
551 | } | |
552 | ||
553 | blk_start_plug(&plug); | |
554 | ret = __generic_file_write_iter(iocb, from); | |
555 | if (ret > 0) | |
556 | ret = generic_write_sync(iocb, ret); | |
557 | iov_iter_reexpand(from, iov_iter_count(from) + shorted); | |
558 | blk_finish_plug(&plug); | |
559 | return ret; | |
560 | } | |
561 | ||
562 | static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
563 | { | |
fac7c6d5 | 564 | struct block_device *bdev = iocb->ki_filp->private_data; |
138c1a38 | 565 | loff_t size = bdev_nr_bytes(bdev); |
cd82cca7 CH |
566 | loff_t pos = iocb->ki_pos; |
567 | size_t shorted = 0; | |
ceaa7625 | 568 | ssize_t ret = 0; |
3e1f941d | 569 | size_t count; |
cd82cca7 | 570 | |
3e1f941d | 571 | if (unlikely(pos + iov_iter_count(to) > size)) { |
6450fe1f PB |
572 | if (pos >= size) |
573 | return 0; | |
574 | size -= pos; | |
3e1f941d ID |
575 | shorted = iov_iter_count(to) - size; |
576 | iov_iter_truncate(to, size); | |
cd82cca7 CH |
577 | } |
578 | ||
3e1f941d ID |
579 | count = iov_iter_count(to); |
580 | if (!count) | |
581 | goto reexpand; /* skip atime */ | |
582 | ||
ceaa7625 JA |
583 | if (iocb->ki_flags & IOCB_DIRECT) { |
584 | struct address_space *mapping = iocb->ki_filp->f_mapping; | |
585 | ||
586 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
3e1f941d ID |
587 | if (filemap_range_needs_writeback(mapping, pos, |
588 | pos + count - 1)) { | |
589 | ret = -EAGAIN; | |
590 | goto reexpand; | |
591 | } | |
ceaa7625 | 592 | } else { |
3e1f941d ID |
593 | ret = filemap_write_and_wait_range(mapping, pos, |
594 | pos + count - 1); | |
ceaa7625 | 595 | if (ret < 0) |
3e1f941d | 596 | goto reexpand; |
ceaa7625 JA |
597 | } |
598 | ||
599 | file_accessed(iocb->ki_filp); | |
600 | ||
601 | ret = blkdev_direct_IO(iocb, to); | |
602 | if (ret >= 0) { | |
603 | iocb->ki_pos += ret; | |
604 | count -= ret; | |
605 | } | |
3e1f941d | 606 | iov_iter_revert(to, count - iov_iter_count(to)); |
ceaa7625 | 607 | if (ret < 0 || !count) |
3e1f941d | 608 | goto reexpand; |
ceaa7625 JA |
609 | } |
610 | ||
611 | ret = filemap_read(iocb, to, ret); | |
6450fe1f | 612 | |
3e1f941d | 613 | reexpand: |
6450fe1f PB |
614 | if (unlikely(shorted)) |
615 | iov_iter_reexpand(to, iov_iter_count(to) + shorted); | |
cd82cca7 CH |
616 | return ret; |
617 | } | |
618 | ||
619 | #define BLKDEV_FALLOC_FL_SUPPORTED \ | |
620 | (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \ | |
621 | FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE) | |
622 | ||
623 | static long blkdev_fallocate(struct file *file, int mode, loff_t start, | |
624 | loff_t len) | |
625 | { | |
f278eb3d ML |
626 | struct inode *inode = bdev_file_inode(file); |
627 | struct block_device *bdev = I_BDEV(inode); | |
cd82cca7 CH |
628 | loff_t end = start + len - 1; |
629 | loff_t isize; | |
630 | int error; | |
631 | ||
632 | /* Fail if we don't recognize the flags. */ | |
633 | if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED) | |
634 | return -EOPNOTSUPP; | |
635 | ||
636 | /* Don't go off the end of the device. */ | |
2a93ad8f | 637 | isize = bdev_nr_bytes(bdev); |
cd82cca7 CH |
638 | if (start >= isize) |
639 | return -EINVAL; | |
640 | if (end >= isize) { | |
641 | if (mode & FALLOC_FL_KEEP_SIZE) { | |
642 | len = isize - start; | |
643 | end = start + len - 1; | |
644 | } else | |
645 | return -EINVAL; | |
646 | } | |
647 | ||
648 | /* | |
649 | * Don't allow IO that isn't aligned to logical block size. | |
650 | */ | |
651 | if ((start | len) & (bdev_logical_block_size(bdev) - 1)) | |
652 | return -EINVAL; | |
653 | ||
f278eb3d ML |
654 | filemap_invalidate_lock(inode->i_mapping); |
655 | ||
cd82cca7 CH |
656 | /* Invalidate the page cache, including dirty pages. */ |
657 | error = truncate_bdev_range(bdev, file->f_mode, start, end); | |
658 | if (error) | |
f278eb3d | 659 | goto fail; |
cd82cca7 CH |
660 | |
661 | switch (mode) { | |
662 | case FALLOC_FL_ZERO_RANGE: | |
663 | case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE: | |
6549a874 PB |
664 | error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, |
665 | len >> SECTOR_SHIFT, GFP_KERNEL, | |
666 | BLKDEV_ZERO_NOUNMAP); | |
cd82cca7 CH |
667 | break; |
668 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE: | |
6549a874 PB |
669 | error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT, |
670 | len >> SECTOR_SHIFT, GFP_KERNEL, | |
671 | BLKDEV_ZERO_NOFALLBACK); | |
cd82cca7 CH |
672 | break; |
673 | case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE: | |
6549a874 PB |
674 | error = blkdev_issue_discard(bdev, start >> SECTOR_SHIFT, |
675 | len >> SECTOR_SHIFT, GFP_KERNEL, 0); | |
cd82cca7 CH |
676 | break; |
677 | default: | |
f278eb3d | 678 | error = -EOPNOTSUPP; |
cd82cca7 | 679 | } |
cd82cca7 | 680 | |
f278eb3d ML |
681 | fail: |
682 | filemap_invalidate_unlock(inode->i_mapping); | |
683 | return error; | |
cd82cca7 CH |
684 | } |
685 | ||
686 | const struct file_operations def_blk_fops = { | |
687 | .open = blkdev_open, | |
688 | .release = blkdev_close, | |
689 | .llseek = blkdev_llseek, | |
690 | .read_iter = blkdev_read_iter, | |
691 | .write_iter = blkdev_write_iter, | |
3e08773c | 692 | .iopoll = iocb_bio_iopoll, |
cd82cca7 CH |
693 | .mmap = generic_file_mmap, |
694 | .fsync = blkdev_fsync, | |
8a709512 | 695 | .unlocked_ioctl = blkdev_ioctl, |
cd82cca7 CH |
696 | #ifdef CONFIG_COMPAT |
697 | .compat_ioctl = compat_blkdev_ioctl, | |
698 | #endif | |
699 | .splice_read = generic_file_splice_read, | |
700 | .splice_write = iter_file_splice_write, | |
701 | .fallocate = blkdev_fallocate, | |
702 | }; | |
703 | ||
704 | static __init int blkdev_init(void) | |
705 | { | |
706 | return bioset_init(&blkdev_dio_pool, 4, | |
707 | offsetof(struct blkdev_dio, bio), | |
708 | BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE); | |
709 | } | |
710 | module_init(blkdev_init); |