projects/linux·blob v6.15 · refs

926 lines · 22.8 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1991, 1992 Linus Torvalds
4 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> 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>
17#include <linux/fs.h>
18#include <linux/iomap.h>
19#include <linux/module.h>
20#include <linux/io_uring/cmd.h>
21#include "blk.h"
22
23static inline struct inode *bdev_file_inode(struct file *file)
24{
25 return file->f_mapping->host;
26}
27
28static blk_opf_t dio_bio_write_op(struct kiocb *iocb)
29{
30 blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
31
32 /* avoid the need for a I/O completion work item */
33 if (iocb_is_dsync(iocb))
34 opf |= REQ_FUA;
35 return opf;
36}
37
38static bool blkdev_dio_invalid(struct block_device *bdev, struct kiocb *iocb,
39 struct iov_iter *iter)
40{
41 return iocb->ki_pos & (bdev_logical_block_size(bdev) - 1) ||
42 !bdev_iter_is_aligned(bdev, iter);
43}
44
45#define DIO_INLINE_BIO_VECS 4
46
47static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
48 struct iov_iter *iter, struct block_device *bdev,
49 unsigned int nr_pages)
50{
51 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
52 loff_t pos = iocb->ki_pos;
53 bool should_dirty = false;
54 struct bio bio;
55 ssize_t ret;
56
57 WARN_ON_ONCE(iocb->ki_flags & IOCB_HAS_METADATA);
58 if (nr_pages <= DIO_INLINE_BIO_VECS)
59 vecs = inline_vecs;
60 else {
61 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
62 GFP_KERNEL);
63 if (!vecs)
64 return -ENOMEM;
65 }
66
67 if (iov_iter_rw(iter) == READ) {
68 bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ);
69 if (user_backed_iter(iter))
70 should_dirty = true;
71 } else {
72 bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb));
73 }
74 bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
75 bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
76 bio.bi_ioprio = iocb->ki_ioprio;
77 if (iocb->ki_flags & IOCB_ATOMIC)
78 bio.bi_opf |= REQ_ATOMIC;
79
80 ret = bio_iov_iter_get_pages(&bio, iter);
81 if (unlikely(ret))
82 goto out;
83 ret = bio.bi_iter.bi_size;
84
85 if (iov_iter_rw(iter) == WRITE)
86 task_io_account_write(ret);
87
88 if (iocb->ki_flags & IOCB_NOWAIT)
89 bio.bi_opf |= REQ_NOWAIT;
90
91 submit_bio_wait(&bio);
92
93 bio_release_pages(&bio, should_dirty);
94 if (unlikely(bio.bi_status))
95 ret = blk_status_to_errno(bio.bi_status);
96
97out:
98 if (vecs != inline_vecs)
99 kfree(vecs);
100
101 bio_uninit(&bio);
102
103 return ret;
104}
105
106enum {
107 DIO_SHOULD_DIRTY = 1,
108 DIO_IS_SYNC = 2,
109};
110
111struct blkdev_dio {
112 union {
113 struct kiocb *iocb;
114 struct task_struct *waiter;
115 };
116 size_t size;
117 atomic_t ref;
118 unsigned int flags;
119 struct bio bio ____cacheline_aligned_in_smp;
120};
121
122static struct bio_set blkdev_dio_pool;
123
124static void blkdev_bio_end_io(struct bio *bio)
125{
126 struct blkdev_dio *dio = bio->bi_private;
127 bool should_dirty = dio->flags & DIO_SHOULD_DIRTY;
128 bool is_sync = dio->flags & DIO_IS_SYNC;
129
130 if (bio->bi_status && !dio->bio.bi_status)
131 dio->bio.bi_status = bio->bi_status;
132
133 if (!is_sync && (dio->iocb->ki_flags & IOCB_HAS_METADATA))
134 bio_integrity_unmap_user(bio);
135
136 if (atomic_dec_and_test(&dio->ref)) {
137 if (!is_sync) {
138 struct kiocb *iocb = dio->iocb;
139 ssize_t ret;
140
141 WRITE_ONCE(iocb->private, NULL);
142
143 if (likely(!dio->bio.bi_status)) {
144 ret = dio->size;
145 iocb->ki_pos += ret;
146 } else {
147 ret = blk_status_to_errno(dio->bio.bi_status);
148 }
149
150 dio->iocb->ki_complete(iocb, ret);
151 bio_put(&dio->bio);
152 } else {
153 struct task_struct *waiter = dio->waiter;
154
155 WRITE_ONCE(dio->waiter, NULL);
156 blk_wake_io_task(waiter);
157 }
158 }
159
160 if (should_dirty) {
161 bio_check_pages_dirty(bio);
162 } else {
163 bio_release_pages(bio, false);
164 bio_put(bio);
165 }
166}
167
168static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
169 struct block_device *bdev, unsigned int nr_pages)
170{
171 struct blk_plug plug;
172 struct blkdev_dio *dio;
173 struct bio *bio;
174 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
175 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
176 loff_t pos = iocb->ki_pos;
177 int ret = 0;
178
179 if (iocb->ki_flags & IOCB_ALLOC_CACHE)
180 opf |= REQ_ALLOC_CACHE;
181 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
182 &blkdev_dio_pool);
183 dio = container_of(bio, struct blkdev_dio, bio);
184 atomic_set(&dio->ref, 1);
185 /*
186 * Grab an extra reference to ensure the dio structure which is embedded
187 * into the first bio stays around.
188 */
189 bio_get(bio);
190
191 is_sync = is_sync_kiocb(iocb);
192 if (is_sync) {
193 dio->flags = DIO_IS_SYNC;
194 dio->waiter = current;
195 } else {
196 dio->flags = 0;
197 dio->iocb = iocb;
198 }
199
200 dio->size = 0;
201 if (is_read && user_backed_iter(iter))
202 dio->flags |= DIO_SHOULD_DIRTY;
203
204 blk_start_plug(&plug);
205
206 for (;;) {
207 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
208 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
209 bio->bi_private = dio;
210 bio->bi_end_io = blkdev_bio_end_io;
211 bio->bi_ioprio = iocb->ki_ioprio;
212
213 ret = bio_iov_iter_get_pages(bio, iter);
214 if (unlikely(ret)) {
215 bio->bi_status = BLK_STS_IOERR;
216 bio_endio(bio);
217 break;
218 }
219 if (iocb->ki_flags & IOCB_NOWAIT) {
220 /*
221 * This is nonblocking IO, and we need to allocate
222 * another bio if we have data left to map. As we
223 * cannot guarantee that one of the sub bios will not
224 * fail getting issued FOR NOWAIT and as error results
225 * are coalesced across all of them, be safe and ask for
226 * a retry of this from blocking context.
227 */
228 if (unlikely(iov_iter_count(iter))) {
229 ret = -EAGAIN;
230 goto fail;
231 }
232 bio->bi_opf |= REQ_NOWAIT;
233 }
234 if (!is_sync && (iocb->ki_flags & IOCB_HAS_METADATA)) {
235 ret = bio_integrity_map_iter(bio, iocb->private);
236 if (unlikely(ret))
237 goto fail;
238 }
239
240 if (is_read) {
241 if (dio->flags & DIO_SHOULD_DIRTY)
242 bio_set_pages_dirty(bio);
243 } else {
244 task_io_account_write(bio->bi_iter.bi_size);
245 }
246 dio->size += bio->bi_iter.bi_size;
247 pos += bio->bi_iter.bi_size;
248
249 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
250 if (!nr_pages) {
251 submit_bio(bio);
252 break;
253 }
254 atomic_inc(&dio->ref);
255 submit_bio(bio);
256 bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL);
257 }
258
259 blk_finish_plug(&plug);
260
261 if (!is_sync)
262 return -EIOCBQUEUED;
263
264 for (;;) {
265 set_current_state(TASK_UNINTERRUPTIBLE);
266 if (!READ_ONCE(dio->waiter))
267 break;
268 blk_io_schedule();
269 }
270 __set_current_state(TASK_RUNNING);
271
272 if (!ret)
273 ret = blk_status_to_errno(dio->bio.bi_status);
274 if (likely(!ret))
275 ret = dio->size;
276
277 bio_put(&dio->bio);
278 return ret;
279fail:
280 bio_release_pages(bio, false);
281 bio_clear_flag(bio, BIO_REFFED);
282 bio_put(bio);
283 blk_finish_plug(&plug);
284 return ret;
285}
286
287static void blkdev_bio_end_io_async(struct bio *bio)
288{
289 struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio);
290 struct kiocb *iocb = dio->iocb;
291 ssize_t ret;
292
293 WRITE_ONCE(iocb->private, NULL);
294
295 if (likely(!bio->bi_status)) {
296 ret = dio->size;
297 iocb->ki_pos += ret;
298 } else {
299 ret = blk_status_to_errno(bio->bi_status);
300 }
301
302 if (iocb->ki_flags & IOCB_HAS_METADATA)
303 bio_integrity_unmap_user(bio);
304
305 iocb->ki_complete(iocb, ret);
306
307 if (dio->flags & DIO_SHOULD_DIRTY) {
308 bio_check_pages_dirty(bio);
309 } else {
310 bio_release_pages(bio, false);
311 bio_put(bio);
312 }
313}
314
315static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
316 struct iov_iter *iter,
317 struct block_device *bdev,
318 unsigned int nr_pages)
319{
320 bool is_read = iov_iter_rw(iter) == READ;
321 blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
322 struct blkdev_dio *dio;
323 struct bio *bio;
324 loff_t pos = iocb->ki_pos;
325 int ret = 0;
326
327 if (iocb->ki_flags & IOCB_ALLOC_CACHE)
328 opf |= REQ_ALLOC_CACHE;
329 bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
330 &blkdev_dio_pool);
331 dio = container_of(bio, struct blkdev_dio, bio);
332 dio->flags = 0;
333 dio->iocb = iocb;
334 bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
335 bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
336 bio->bi_end_io = blkdev_bio_end_io_async;
337 bio->bi_ioprio = iocb->ki_ioprio;
338
339 if (iov_iter_is_bvec(iter)) {
340 /*
341 * Users don't rely on the iterator being in any particular
342 * state for async I/O returning -EIOCBQUEUED, hence we can
343 * avoid expensive iov_iter_advance(). Bypass
344 * bio_iov_iter_get_pages() and set the bvec directly.
345 */
346 bio_iov_bvec_set(bio, iter);
347 } else {
348 ret = bio_iov_iter_get_pages(bio, iter);
349 if (unlikely(ret))
350 goto out_bio_put;
351 }
352 dio->size = bio->bi_iter.bi_size;
353
354 if (is_read) {
355 if (user_backed_iter(iter)) {
356 dio->flags |= DIO_SHOULD_DIRTY;
357 bio_set_pages_dirty(bio);
358 }
359 } else {
360 task_io_account_write(bio->bi_iter.bi_size);
361 }
362
363 if (iocb->ki_flags & IOCB_HAS_METADATA) {
364 ret = bio_integrity_map_iter(bio, iocb->private);
365 WRITE_ONCE(iocb->private, NULL);
366 if (unlikely(ret))
367 goto out_bio_put;
368 }
369
370 if (iocb->ki_flags & IOCB_ATOMIC)
371 bio->bi_opf |= REQ_ATOMIC;
372
373 if (iocb->ki_flags & IOCB_NOWAIT)
374 bio->bi_opf |= REQ_NOWAIT;
375
376 if (iocb->ki_flags & IOCB_HIPRI) {
377 bio->bi_opf |= REQ_POLLED;
378 submit_bio(bio);
379 WRITE_ONCE(iocb->private, bio);
380 } else {
381 submit_bio(bio);
382 }
383 return -EIOCBQUEUED;
384
385out_bio_put:
386 bio_put(bio);
387 return ret;
388}
389
390static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
391{
392 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
393 unsigned int nr_pages;
394
395 if (!iov_iter_count(iter))
396 return 0;
397
398 if (blkdev_dio_invalid(bdev, iocb, iter))
399 return -EINVAL;
400
401 nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
402 if (likely(nr_pages <= BIO_MAX_VECS)) {
403 if (is_sync_kiocb(iocb))
404 return __blkdev_direct_IO_simple(iocb, iter, bdev,
405 nr_pages);
406 return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages);
407 } else if (iocb->ki_flags & IOCB_ATOMIC) {
408 return -EINVAL;
409 }
410 return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages));
411}
412
413static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
414 unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
415{
416 struct block_device *bdev = I_BDEV(inode);
417 loff_t isize = i_size_read(inode);
418
419 if (offset >= isize)
420 return -EIO;
421
422 iomap->bdev = bdev;
423 iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev));
424 iomap->type = IOMAP_MAPPED;
425 iomap->addr = iomap->offset;
426 iomap->length = isize - iomap->offset;
427 iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */
428 return 0;
429}
430
431static const struct iomap_ops blkdev_iomap_ops = {
432 .iomap_begin = blkdev_iomap_begin,
433};
434
435#ifdef CONFIG_BUFFER_HEAD
436static int blkdev_get_block(struct inode *inode, sector_t iblock,
437 struct buffer_head *bh, int create)
438{
439 bh->b_bdev = I_BDEV(inode);
440 bh->b_blocknr = iblock;
441 set_buffer_mapped(bh);
442 return 0;
443}
444
445/*
446 * We cannot call mpage_writepages() as it does not take the buffer lock.
447 * We must use block_write_full_folio() directly which holds the buffer
448 * lock. The buffer lock provides the synchronisation with writeback
449 * that filesystems rely on when they use the blockdev's mapping.
450 */
451static int blkdev_writepages(struct address_space *mapping,
452 struct writeback_control *wbc)
453{
454 struct blk_plug plug;
455 int err;
456
457 blk_start_plug(&plug);
458 err = write_cache_pages(mapping, wbc, block_write_full_folio,
459 blkdev_get_block);
460 blk_finish_plug(&plug);
461
462 return err;
463}
464
465static int blkdev_read_folio(struct file *file, struct folio *folio)
466{
467 return block_read_full_folio(folio, blkdev_get_block);
468}
469
470static void blkdev_readahead(struct readahead_control *rac)
471{
472 mpage_readahead(rac, blkdev_get_block);
473}
474
475static int blkdev_write_begin(struct file *file, struct address_space *mapping,
476 loff_t pos, unsigned len, struct folio **foliop, void **fsdata)
477{
478 return block_write_begin(mapping, pos, len, foliop, blkdev_get_block);
479}
480
481static int blkdev_write_end(struct file *file, struct address_space *mapping,
482 loff_t pos, unsigned len, unsigned copied, struct folio *folio,
483 void *fsdata)
484{
485 int ret;
486 ret = block_write_end(file, mapping, pos, len, copied, folio, fsdata);
487
488 folio_unlock(folio);
489 folio_put(folio);
490
491 return ret;
492}
493
494const struct address_space_operations def_blk_aops = {
495 .dirty_folio = block_dirty_folio,
496 .invalidate_folio = block_invalidate_folio,
497 .read_folio = blkdev_read_folio,
498 .readahead = blkdev_readahead,
499 .writepages = blkdev_writepages,
500 .write_begin = blkdev_write_begin,
501 .write_end = blkdev_write_end,
502 .migrate_folio = buffer_migrate_folio_norefs,
503 .is_dirty_writeback = buffer_check_dirty_writeback,
504};
505#else /* CONFIG_BUFFER_HEAD */
506static int blkdev_read_folio(struct file *file, struct folio *folio)
507{
508 return iomap_read_folio(folio, &blkdev_iomap_ops);
509}
510
511static void blkdev_readahead(struct readahead_control *rac)
512{
513 iomap_readahead(rac, &blkdev_iomap_ops);
514}
515
516static int blkdev_map_blocks(struct iomap_writepage_ctx *wpc,
517 struct inode *inode, loff_t offset, unsigned int len)
518{
519 loff_t isize = i_size_read(inode);
520
521 if (WARN_ON_ONCE(offset >= isize))
522 return -EIO;
523 if (offset >= wpc->iomap.offset &&
524 offset < wpc->iomap.offset + wpc->iomap.length)
525 return 0;
526 return blkdev_iomap_begin(inode, offset, isize - offset,
527 IOMAP_WRITE, &wpc->iomap, NULL);
528}
529
530static const struct iomap_writeback_ops blkdev_writeback_ops = {
531 .map_blocks = blkdev_map_blocks,
532};
533
534static int blkdev_writepages(struct address_space *mapping,
535 struct writeback_control *wbc)
536{
537 struct iomap_writepage_ctx wpc = { };
538
539 return iomap_writepages(mapping, wbc, &wpc, &blkdev_writeback_ops);
540}
541
542const struct address_space_operations def_blk_aops = {
543 .dirty_folio = filemap_dirty_folio,
544 .release_folio = iomap_release_folio,
545 .invalidate_folio = iomap_invalidate_folio,
546 .read_folio = blkdev_read_folio,
547 .readahead = blkdev_readahead,
548 .writepages = blkdev_writepages,
549 .is_partially_uptodate = iomap_is_partially_uptodate,
550 .error_remove_folio = generic_error_remove_folio,
551 .migrate_folio = filemap_migrate_folio,
552};
553#endif /* CONFIG_BUFFER_HEAD */
554
555/*
556 * for a block special file file_inode(file)->i_size is zero
557 * so we compute the size by hand (just as in block_read/write above)
558 */
559static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
560{
561 struct inode *bd_inode = bdev_file_inode(file);
562 loff_t retval;
563
564 inode_lock(bd_inode);
565 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
566 inode_unlock(bd_inode);
567 return retval;
568}
569
570static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
571 int datasync)
572{
573 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
574 int error;
575
576 error = file_write_and_wait_range(filp, start, end);
577 if (error)
578 return error;
579
580 /*
581 * There is no need to serialise calls to blkdev_issue_flush with
582 * i_mutex and doing so causes performance issues with concurrent
583 * O_SYNC writers to a block device.
584 */
585 error = blkdev_issue_flush(bdev);
586 if (error == -EOPNOTSUPP)
587 error = 0;
588
589 return error;
590}
591
592/**
593 * file_to_blk_mode - get block open flags from file flags
594 * @file: file whose open flags should be converted
595 *
596 * Look at file open flags and generate corresponding block open flags from
597 * them. The function works both for file just being open (e.g. during ->open
598 * callback) and for file that is already open. This is actually non-trivial
599 * (see comment in the function).
600 */
601blk_mode_t file_to_blk_mode(struct file *file)
602{
603 blk_mode_t mode = 0;
604
605 if (file->f_mode & FMODE_READ)
606 mode |= BLK_OPEN_READ;
607 if (file->f_mode & FMODE_WRITE)
608 mode |= BLK_OPEN_WRITE;
609 /*
610 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data
611 * to determine whether the open was exclusive for already open files.
612 */
613 if (file->private_data)
614 mode |= BLK_OPEN_EXCL;
615 else if (file->f_flags & O_EXCL)
616 mode |= BLK_OPEN_EXCL;
617 if (file->f_flags & O_NDELAY)
618 mode |= BLK_OPEN_NDELAY;
619
620 /*
621 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy
622 * driver has historically allowed ioctls as if the file was opened for
623 * writing, but does not allow and actual reads or writes.
624 */
625 if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY))
626 mode |= BLK_OPEN_WRITE_IOCTL;
627
628 return mode;
629}
630
631static int blkdev_open(struct inode *inode, struct file *filp)
632{
633 struct block_device *bdev;
634 blk_mode_t mode;
635 int ret;
636
637 mode = file_to_blk_mode(filp);
638 /* Use the file as the holder. */
639 if (mode & BLK_OPEN_EXCL)
640 filp->private_data = filp;
641 ret = bdev_permission(inode->i_rdev, mode, filp->private_data);
642 if (ret)
643 return ret;
644
645 bdev = blkdev_get_no_open(inode->i_rdev, true);
646 if (!bdev)
647 return -ENXIO;
648
649 if (bdev_can_atomic_write(bdev))
650 filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
651
652 ret = bdev_open(bdev, mode, filp->private_data, NULL, filp);
653 if (ret)
654 blkdev_put_no_open(bdev);
655 return ret;
656}
657
658static int blkdev_release(struct inode *inode, struct file *filp)
659{
660 bdev_release(filp);
661 return 0;
662}
663
664static ssize_t
665blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from)
666{
667 size_t count = iov_iter_count(from);
668 ssize_t written;
669
670 written = kiocb_invalidate_pages(iocb, count);
671 if (written) {
672 if (written == -EBUSY)
673 return 0;
674 return written;
675 }
676
677 written = blkdev_direct_IO(iocb, from);
678 if (written > 0) {
679 kiocb_invalidate_post_direct_write(iocb, count);
680 iocb->ki_pos += written;
681 count -= written;
682 }
683 if (written != -EIOCBQUEUED)
684 iov_iter_revert(from, count - iov_iter_count(from));
685 return written;
686}
687
688static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from)
689{
690 return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL);
691}
692
693/*
694 * Write data to the block device. Only intended for the block device itself
695 * and the raw driver which basically is a fake block device.
696 *
697 * Does not take i_mutex for the write and thus is not for general purpose
698 * use.
699 */
700static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
701{
702 struct file *file = iocb->ki_filp;
703 struct inode *bd_inode = bdev_file_inode(file);
704 struct block_device *bdev = I_BDEV(bd_inode);
705 bool atomic = iocb->ki_flags & IOCB_ATOMIC;
706 loff_t size = bdev_nr_bytes(bdev);
707 size_t shorted = 0;
708 ssize_t ret;
709
710 if (bdev_read_only(bdev))
711 return -EPERM;
712
713 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
714 return -ETXTBSY;
715
716 if (!iov_iter_count(from))
717 return 0;
718
719 if (iocb->ki_pos >= size)
720 return -ENOSPC;
721
722 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
723 return -EOPNOTSUPP;
724
725 if (atomic) {
726 ret = generic_atomic_write_valid(iocb, from);
727 if (ret)
728 return ret;
729 }
730
731 size -= iocb->ki_pos;
732 if (iov_iter_count(from) > size) {
733 if (atomic)
734 return -EINVAL;
735 shorted = iov_iter_count(from) - size;
736 iov_iter_truncate(from, size);
737 }
738
739 ret = file_update_time(file);
740 if (ret)
741 return ret;
742
743 if (iocb->ki_flags & IOCB_DIRECT) {
744 ret = blkdev_direct_write(iocb, from);
745 if (ret >= 0 && iov_iter_count(from))
746 ret = direct_write_fallback(iocb, from, ret,
747 blkdev_buffered_write(iocb, from));
748 } else {
749 /*
750 * Take i_rwsem and invalidate_lock to avoid racing with
751 * set_blocksize changing i_blkbits/folio order and punching
752 * out the pagecache.
753 */
754 inode_lock_shared(bd_inode);
755 ret = blkdev_buffered_write(iocb, from);
756 inode_unlock_shared(bd_inode);
757 }
758
759 if (ret > 0)
760 ret = generic_write_sync(iocb, ret);
761 iov_iter_reexpand(from, iov_iter_count(from) + shorted);
762 return ret;
763}
764
765static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
766{
767 struct inode *bd_inode = bdev_file_inode(iocb->ki_filp);
768 struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
769 loff_t size = bdev_nr_bytes(bdev);
770 loff_t pos = iocb->ki_pos;
771 size_t shorted = 0;
772 ssize_t ret = 0;
773 size_t count;
774
775 if (unlikely(pos + iov_iter_count(to) > size)) {
776 if (pos >= size)
777 return 0;
778 size -= pos;
779 shorted = iov_iter_count(to) - size;
780 iov_iter_truncate(to, size);
781 }
782
783 count = iov_iter_count(to);
784 if (!count)
785 goto reexpand; /* skip atime */
786
787 if (iocb->ki_flags & IOCB_DIRECT) {
788 ret = kiocb_write_and_wait(iocb, count);
789 if (ret < 0)
790 goto reexpand;
791 file_accessed(iocb->ki_filp);
792
793 ret = blkdev_direct_IO(iocb, to);
794 if (ret > 0) {
795 iocb->ki_pos += ret;
796 count -= ret;
797 }
798 if (ret != -EIOCBQUEUED)
799 iov_iter_revert(to, count - iov_iter_count(to));
800 if (ret < 0 || !count)
801 goto reexpand;
802 }
803
804 /*
805 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize
806 * changing i_blkbits/folio order and punching out the pagecache.
807 */
808 inode_lock_shared(bd_inode);
809 ret = filemap_read(iocb, to, ret);
810 inode_unlock_shared(bd_inode);
811
812reexpand:
813 if (unlikely(shorted))
814 iov_iter_reexpand(to, iov_iter_count(to) + shorted);
815 return ret;
816}
817
818#define BLKDEV_FALLOC_FL_SUPPORTED \
819 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
820 FALLOC_FL_ZERO_RANGE)
821
822static long blkdev_fallocate(struct file *file, int mode, loff_t start,
823 loff_t len)
824{
825 struct inode *inode = bdev_file_inode(file);
826 struct block_device *bdev = I_BDEV(inode);
827 loff_t end = start + len - 1;
828 loff_t isize;
829 int error;
830
831 /* Fail if we don't recognize the flags. */
832 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
833 return -EOPNOTSUPP;
834
835 /* Don't go off the end of the device. */
836 isize = bdev_nr_bytes(bdev);
837 if (start >= isize)
838 return -EINVAL;
839 if (end >= isize) {
840 if (mode & FALLOC_FL_KEEP_SIZE) {
841 len = isize - start;
842 end = start + len - 1;
843 } else
844 return -EINVAL;
845 }
846
847 /*
848 * Don't allow IO that isn't aligned to logical block size.
849 */
850 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
851 return -EINVAL;
852
853 inode_lock(inode);
854 filemap_invalidate_lock(inode->i_mapping);
855
856 /*
857 * Invalidate the page cache, including dirty pages, for valid
858 * de-allocate mode calls to fallocate().
859 */
860 switch (mode) {
861 case FALLOC_FL_ZERO_RANGE:
862 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
863 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end);
864 if (error)
865 goto fail;
866
867 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
868 len >> SECTOR_SHIFT, GFP_KERNEL,
869 BLKDEV_ZERO_NOUNMAP);
870 break;
871 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
872 error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end);
873 if (error)
874 goto fail;
875
876 error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
877 len >> SECTOR_SHIFT, GFP_KERNEL,
878 BLKDEV_ZERO_NOFALLBACK);
879 break;
880 default:
881 error = -EOPNOTSUPP;
882 }
883
884 fail:
885 filemap_invalidate_unlock(inode->i_mapping);
886 inode_unlock(inode);
887 return error;
888}
889
890static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
891{
892 struct inode *bd_inode = bdev_file_inode(file);
893
894 if (bdev_read_only(I_BDEV(bd_inode)))
895 return generic_file_readonly_mmap(file, vma);
896
897 return generic_file_mmap(file, vma);
898}
899
900const struct file_operations def_blk_fops = {
901 .open = blkdev_open,
902 .release = blkdev_release,
903 .llseek = blkdev_llseek,
904 .read_iter = blkdev_read_iter,
905 .write_iter = blkdev_write_iter,
906 .iopoll = iocb_bio_iopoll,
907 .mmap = blkdev_mmap,
908 .fsync = blkdev_fsync,
909 .unlocked_ioctl = blkdev_ioctl,
910#ifdef CONFIG_COMPAT
911 .compat_ioctl = compat_blkdev_ioctl,
912#endif
913 .splice_read = filemap_splice_read,
914 .splice_write = iter_file_splice_write,
915 .fallocate = blkdev_fallocate,
916 .uring_cmd = blkdev_uring_cmd,
917 .fop_flags = FOP_BUFFER_RASYNC,
918};
919
920static __init int blkdev_init(void)
921{
922 return bioset_init(&blkdev_dio_pool, 4,
923 offsetof(struct blkdev_dio, bio),
924 BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
925}
926module_init(blkdev_init);