fops.c

  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/blk-integrity.h>
 11#include <linux/buffer_head.h>
 12#include <linux/mpage.h>
 13#include <linux/uio.h>
 14#include <linux/namei.h>
 15#include <linux/task_io_accounting_ops.h>
 16#include <linux/falloc.h>
 17#include <linux/suspend.h>
 18#include <linux/fs.h>
 19#include <linux/iomap.h>
 20#include <linux/module.h>
 21#include <linux/io_uring/cmd.h>
 22#include "blk.h"
 23
 24static inline struct inode *bdev_file_inode(struct file *file)
 25{
 26	return file->f_mapping->host;
 27}
 28
 29static blk_opf_t dio_bio_write_op(struct kiocb *iocb)
 30{
 31	blk_opf_t opf = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
 32
 33	/* avoid the need for a I/O completion work item */
 34	if (iocb_is_dsync(iocb))
 35		opf |= REQ_FUA;
 36	return opf;
 37}
 38
 39static bool blkdev_dio_invalid(struct block_device *bdev, struct kiocb *iocb,
 40				struct iov_iter *iter)
 41{
 42	return (iocb->ki_pos | iov_iter_count(iter)) &
 43			(bdev_logical_block_size(bdev) - 1);
 44}
 45
 46static inline int blkdev_iov_iter_get_pages(struct bio *bio,
 47		struct iov_iter *iter, struct block_device *bdev)
 48{
 49	return bio_iov_iter_get_pages(bio, iter,
 50			bdev_logical_block_size(bdev) - 1);
 51}
 52
 53#define DIO_INLINE_BIO_VECS 4
 54
 55static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
 56		struct iov_iter *iter, struct block_device *bdev,
 57		unsigned int nr_pages)
 58{
 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;
 64
 65	if (nr_pages <= DIO_INLINE_BIO_VECS)
 66		vecs = inline_vecs;
 67	else {
 68		vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
 69				     GFP_KERNEL);
 70		if (!vecs)
 71			return -ENOMEM;
 72	}
 73
 74	if (iov_iter_rw(iter) == READ) {
 75		bio_init(&bio, bdev, vecs, nr_pages, REQ_OP_READ);
 76		if (user_backed_iter(iter))
 77			should_dirty = true;
 78	} else {
 79		bio_init(&bio, bdev, vecs, nr_pages, dio_bio_write_op(iocb));
 80	}
 81	bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 82	bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
 83	bio.bi_write_stream = iocb->ki_write_stream;
 84	bio.bi_ioprio = iocb->ki_ioprio;
 85	if (iocb->ki_flags & IOCB_ATOMIC)
 86		bio.bi_opf |= REQ_ATOMIC;
 87
 88	ret = blkdev_iov_iter_get_pages(&bio, iter, bdev);
 89	if (unlikely(ret))
 90		goto out;
 91	ret = bio.bi_iter.bi_size;
 92
 93	if (iov_iter_rw(iter) == WRITE)
 94		task_io_account_write(ret);
 95
 96	if (iocb->ki_flags & IOCB_NOWAIT)
 97		bio.bi_opf |= REQ_NOWAIT;
 98
 99	submit_bio_wait(&bio);
100
101	bio_release_pages(&bio, should_dirty);
102	if (unlikely(bio.bi_status))
103		ret = blk_status_to_errno(bio.bi_status);
104
105out:
106	if (vecs != inline_vecs)
107		kfree(vecs);
108
109	bio_uninit(&bio);
110
111	return ret;
112}
113
114enum {
115	DIO_SHOULD_DIRTY	= 1,
116	DIO_IS_SYNC		= 2,
117};
118
119struct blkdev_dio {
120	union {
121		struct kiocb		*iocb;
122		struct task_struct	*waiter;
123	};
124	size_t			size;
125	atomic_t		ref;
126	unsigned int		flags;
127	struct bio		bio ____cacheline_aligned_in_smp;
128};
129
130static struct bio_set blkdev_dio_pool;
131
132static void blkdev_bio_end_io(struct bio *bio)
133{
134	struct blkdev_dio *dio = bio->bi_private;
135	bool should_dirty = dio->flags & DIO_SHOULD_DIRTY;
136	bool is_sync = dio->flags & DIO_IS_SYNC;
137
138	if (bio->bi_status && !dio->bio.bi_status)
139		dio->bio.bi_status = bio->bi_status;
140
141	if (bio_integrity(bio))
142		bio_integrity_unmap_user(bio);
143
144	if (atomic_dec_and_test(&dio->ref)) {
145		if (!is_sync) {
146			struct kiocb *iocb = dio->iocb;
147			ssize_t ret;
148
149			WRITE_ONCE(iocb->private, NULL);
150
151			if (likely(!dio->bio.bi_status)) {
152				ret = dio->size;
153				iocb->ki_pos += ret;
154			} else {
155				ret = blk_status_to_errno(dio->bio.bi_status);
156			}
157
158			dio->iocb->ki_complete(iocb, ret);
159			bio_put(&dio->bio);
160		} else {
161			struct task_struct *waiter = dio->waiter;
162
163			WRITE_ONCE(dio->waiter, NULL);
164			blk_wake_io_task(waiter);
165		}
166	}
167
168	if (should_dirty) {
169		bio_check_pages_dirty(bio);
170	} else {
171		bio_release_pages(bio, false);
172		bio_put(bio);
173	}
174}
175
176static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
177		struct block_device *bdev, unsigned int nr_pages)
178{
179	struct blk_plug plug;
180	struct blkdev_dio *dio;
181	struct bio *bio;
182	bool is_read = (iov_iter_rw(iter) == READ), is_sync;
183	blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
184	loff_t pos = iocb->ki_pos;
185	int ret = 0;
186
187	bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
188			       &blkdev_dio_pool);
189	dio = container_of(bio, struct blkdev_dio, bio);
190	atomic_set(&dio->ref, 1);
191	/*
192	 * Grab an extra reference to ensure the dio structure which is embedded
193	 * into the first bio stays around.
194	 */
195	bio_get(bio);
196
197	is_sync = is_sync_kiocb(iocb);
198	if (is_sync) {
199		dio->flags = DIO_IS_SYNC;
200		dio->waiter = current;
201	} else {
202		dio->flags = 0;
203		dio->iocb = iocb;
204	}
205
206	dio->size = 0;
207	if (is_read && user_backed_iter(iter))
208		dio->flags |= DIO_SHOULD_DIRTY;
209
210	blk_start_plug(&plug);
211
212	for (;;) {
213		bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
214		bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
215		bio->bi_write_stream = iocb->ki_write_stream;
216		bio->bi_private = dio;
217		bio->bi_end_io = blkdev_bio_end_io;
218		bio->bi_ioprio = iocb->ki_ioprio;
219
220		ret = blkdev_iov_iter_get_pages(bio, iter, bdev);
221		if (unlikely(ret)) {
222			bio->bi_status = BLK_STS_IOERR;
223			bio_endio(bio);
224			break;
225		}
226		if (iocb->ki_flags & IOCB_NOWAIT) {
227			/*
228			 * This is nonblocking IO, and we need to allocate
229			 * another bio if we have data left to map. As we
230			 * cannot guarantee that one of the sub bios will not
231			 * fail getting issued FOR NOWAIT and as error results
232			 * are coalesced across all of them, be safe and ask for
233			 * a retry of this from blocking context.
234			 */
235			if (unlikely(iov_iter_count(iter))) {
236				ret = -EAGAIN;
237				goto fail;
238			}
239			bio->bi_opf |= REQ_NOWAIT;
240		}
241		if (iocb->ki_flags & IOCB_HAS_METADATA) {
242			ret = bio_integrity_map_iter(bio, iocb->private);
243			if (unlikely(ret))
244				goto fail;
245		}
246
247		if (is_read) {
248			if (dio->flags & DIO_SHOULD_DIRTY)
249				bio_set_pages_dirty(bio);
250		} else {
251			task_io_account_write(bio->bi_iter.bi_size);
252		}
253		dio->size += bio->bi_iter.bi_size;
254		pos += bio->bi_iter.bi_size;
255
256		nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS);
257		if (!nr_pages) {
258			submit_bio(bio);
259			break;
260		}
261		atomic_inc(&dio->ref);
262		submit_bio(bio);
263		bio = bio_alloc(bdev, nr_pages, opf, GFP_KERNEL);
264	}
265
266	blk_finish_plug(&plug);
267
268	if (!is_sync)
269		return -EIOCBQUEUED;
270
271	for (;;) {
272		set_current_state(TASK_UNINTERRUPTIBLE);
273		if (!READ_ONCE(dio->waiter))
274			break;
275		blk_io_schedule();
276	}
277	__set_current_state(TASK_RUNNING);
278
279	if (!ret)
280		ret = blk_status_to_errno(dio->bio.bi_status);
281	if (likely(!ret))
282		ret = dio->size;
283
284	bio_put(&dio->bio);
285	return ret;
286fail:
287	bio_release_pages(bio, false);
288	bio_clear_flag(bio, BIO_REFFED);
289	bio_put(bio);
290	blk_finish_plug(&plug);
291	return ret;
292}
293
294static void blkdev_bio_end_io_async(struct bio *bio)
295{
296	struct blkdev_dio *dio = container_of(bio, struct blkdev_dio, bio);
297	struct kiocb *iocb = dio->iocb;
298	ssize_t ret;
299
300	WRITE_ONCE(iocb->private, NULL);
301
302	if (likely(!bio->bi_status)) {
303		ret = dio->size;
304		iocb->ki_pos += ret;
305	} else {
306		ret = blk_status_to_errno(bio->bi_status);
307	}
308
309	if (bio_integrity(bio))
310		bio_integrity_unmap_user(bio);
311
312	iocb->ki_complete(iocb, ret);
313
314	if (dio->flags & DIO_SHOULD_DIRTY) {
315		bio_check_pages_dirty(bio);
316	} else {
317		bio_release_pages(bio, false);
318		bio_put(bio);
319	}
320}
321
322static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
323					struct iov_iter *iter,
324					struct block_device *bdev,
325					unsigned int nr_pages)
326{
327	bool is_read = iov_iter_rw(iter) == READ;
328	blk_opf_t opf = is_read ? REQ_OP_READ : dio_bio_write_op(iocb);
329	struct blkdev_dio *dio;
330	struct bio *bio;
331	loff_t pos = iocb->ki_pos;
332	int ret = 0;
333
334	bio = bio_alloc_bioset(bdev, nr_pages, opf, GFP_KERNEL,
335			       &blkdev_dio_pool);
336	dio = container_of(bio, struct blkdev_dio, bio);
337	dio->flags = 0;
338	dio->iocb = iocb;
339	bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
340	bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
341	bio->bi_write_stream = iocb->ki_write_stream;
342	bio->bi_end_io = blkdev_bio_end_io_async;
343	bio->bi_ioprio = iocb->ki_ioprio;
344
345	if (iov_iter_is_bvec(iter)) {
346		/*
347		 * Users don't rely on the iterator being in any particular
348		 * state for async I/O returning -EIOCBQUEUED, hence we can
349		 * avoid expensive iov_iter_advance(). Bypass
350		 * bio_iov_iter_get_pages() and set the bvec directly.
351		 */
352		bio_iov_bvec_set(bio, iter);
353	} else {
354		ret = blkdev_iov_iter_get_pages(bio, iter, bdev);
355		if (unlikely(ret))
356			goto out_bio_put;
357	}
358	dio->size = bio->bi_iter.bi_size;
359
360	if (is_read) {
361		if (user_backed_iter(iter)) {
362			dio->flags |= DIO_SHOULD_DIRTY;
363			bio_set_pages_dirty(bio);
364		}
365	} else {
366		task_io_account_write(bio->bi_iter.bi_size);
367	}
368
369	if (iocb->ki_flags & IOCB_HAS_METADATA) {
370		ret = bio_integrity_map_iter(bio, iocb->private);
371		WRITE_ONCE(iocb->private, NULL);
372		if (unlikely(ret))
373			goto out_bio_put;
374	}
375
376	if (iocb->ki_flags & IOCB_ATOMIC)
377		bio->bi_opf |= REQ_ATOMIC;
378
379	if (iocb->ki_flags & IOCB_NOWAIT)
380		bio->bi_opf |= REQ_NOWAIT;
381
382	if (iocb->ki_flags & IOCB_HIPRI) {
383		bio->bi_opf |= REQ_POLLED;
384		submit_bio(bio);
385		WRITE_ONCE(iocb->private, bio);
386	} else {
387		submit_bio(bio);
388	}
389	return -EIOCBQUEUED;
390
391out_bio_put:
392	bio_put(bio);
393	return ret;
394}
395
396static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
397{
398	struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
399	unsigned int nr_pages;
400
401	if (!iov_iter_count(iter))
402		return 0;
403
404	if (blkdev_dio_invalid(bdev, iocb, iter))
405		return -EINVAL;
406
407	if (iov_iter_rw(iter) == WRITE) {
408		u16 max_write_streams = bdev_max_write_streams(bdev);
409
410		if (iocb->ki_write_stream) {
411			if (iocb->ki_write_stream > max_write_streams)
412				return -EINVAL;
413		} else if (max_write_streams) {
414			enum rw_hint write_hint =
415				file_inode(iocb->ki_filp)->i_write_hint;
416
417			/*
418			 * Just use the write hint as write stream for block
419			 * device writes.  This assumes no file system is
420			 * mounted that would use the streams differently.
421			 */
422			if (write_hint <= max_write_streams)
423				iocb->ki_write_stream = write_hint;
424		}
425	}
426
427	nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
428	if (likely(nr_pages <= BIO_MAX_VECS &&
429		   !(iocb->ki_flags & IOCB_HAS_METADATA))) {
430		if (is_sync_kiocb(iocb))
431			return __blkdev_direct_IO_simple(iocb, iter, bdev,
432							nr_pages);
433		return __blkdev_direct_IO_async(iocb, iter, bdev, nr_pages);
434	} else if (iocb->ki_flags & IOCB_ATOMIC) {
435		return -EINVAL;
436	}
437	return __blkdev_direct_IO(iocb, iter, bdev, bio_max_segs(nr_pages));
438}
439
440static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
441		unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
442{
443	struct block_device *bdev = I_BDEV(inode);
444	loff_t isize = i_size_read(inode);
445
446	if (offset >= isize)
447		return -EIO;
448
449	iomap->bdev = bdev;
450	iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev));
451	iomap->type = IOMAP_MAPPED;
452	iomap->addr = iomap->offset;
453	iomap->length = isize - iomap->offset;
454	iomap->flags |= IOMAP_F_BUFFER_HEAD; /* noop for !CONFIG_BUFFER_HEAD */
455	return 0;
456}
457
458static const struct iomap_ops blkdev_iomap_ops = {
459	.iomap_begin		= blkdev_iomap_begin,
460};
461
462#ifdef CONFIG_BUFFER_HEAD
463static int blkdev_get_block(struct inode *inode, sector_t iblock,
464		struct buffer_head *bh, int create)
465{
466	bh->b_bdev = I_BDEV(inode);
467	bh->b_blocknr = iblock;
468	set_buffer_mapped(bh);
469	return 0;
470}
471
472/*
473 * We cannot call mpage_writepages() as it does not take the buffer lock.
474 * We must use block_write_full_folio() directly which holds the buffer
475 * lock.  The buffer lock provides the synchronisation with writeback
476 * that filesystems rely on when they use the blockdev's mapping.
477 */
478static int blkdev_writepages(struct address_space *mapping,
479		struct writeback_control *wbc)
480{
481	struct folio *folio = NULL;
482	struct blk_plug plug;
483	int err;
484
485	blk_start_plug(&plug);
486	while ((folio = writeback_iter(mapping, wbc, folio, &err)))
487		err = block_write_full_folio(folio, wbc, blkdev_get_block);
488	blk_finish_plug(&plug);
489
490	return err;
491}
492
493static int blkdev_read_folio(struct file *file, struct folio *folio)
494{
495	return block_read_full_folio(folio, blkdev_get_block);
496}
497
498static void blkdev_readahead(struct readahead_control *rac)
499{
500	mpage_readahead(rac, blkdev_get_block);
501}
502
503static int blkdev_write_begin(const struct kiocb *iocb,
504			      struct address_space *mapping, loff_t pos,
505			      unsigned len, struct folio **foliop,
506			      void **fsdata)
507{
508	return block_write_begin(mapping, pos, len, foliop, blkdev_get_block);
509}
510
511static int blkdev_write_end(const struct kiocb *iocb,
512			    struct address_space *mapping,
513			    loff_t pos, unsigned len, unsigned copied,
514			    struct folio *folio, void *fsdata)
515{
516	int ret;
517	ret = block_write_end(pos, len, copied, folio);
518
519	folio_unlock(folio);
520	folio_put(folio);
521
522	return ret;
523}
524
525const struct address_space_operations def_blk_aops = {
526	.dirty_folio	= block_dirty_folio,
527	.invalidate_folio = block_invalidate_folio,
528	.read_folio	= blkdev_read_folio,
529	.readahead	= blkdev_readahead,
530	.writepages	= blkdev_writepages,
531	.write_begin	= blkdev_write_begin,
532	.write_end	= blkdev_write_end,
533	.migrate_folio	= buffer_migrate_folio_norefs,
534	.is_dirty_writeback = buffer_check_dirty_writeback,
535};
536#else /* CONFIG_BUFFER_HEAD */
537static int blkdev_read_folio(struct file *file, struct folio *folio)
538{
539	iomap_bio_read_folio(folio, &blkdev_iomap_ops);
540	return 0;
541}
542
543static void blkdev_readahead(struct readahead_control *rac)
544{
545	iomap_bio_readahead(rac, &blkdev_iomap_ops);
546}
547
548static ssize_t blkdev_writeback_range(struct iomap_writepage_ctx *wpc,
549		struct folio *folio, u64 offset, unsigned int len, u64 end_pos)
550{
551	loff_t isize = i_size_read(wpc->inode);
552
553	if (WARN_ON_ONCE(offset >= isize))
554		return -EIO;
555
556	if (offset < wpc->iomap.offset ||
557	    offset >= wpc->iomap.offset + wpc->iomap.length) {
558		int error;
559
560		error = blkdev_iomap_begin(wpc->inode, offset, isize - offset,
561				IOMAP_WRITE, &wpc->iomap, NULL);
562		if (error)
563			return error;
564	}
565
566	return iomap_add_to_ioend(wpc, folio, offset, end_pos, len);
567}
568
569static const struct iomap_writeback_ops blkdev_writeback_ops = {
570	.writeback_range	= blkdev_writeback_range,
571	.writeback_submit	= iomap_ioend_writeback_submit,
572};
573
574static int blkdev_writepages(struct address_space *mapping,
575		struct writeback_control *wbc)
576{
577	struct iomap_writepage_ctx wpc = {
578		.inode		= mapping->host,
579		.wbc		= wbc,
580		.ops		= &blkdev_writeback_ops
581	};
582
583	return iomap_writepages(&wpc);
584}
585
586const struct address_space_operations def_blk_aops = {
587	.dirty_folio	= filemap_dirty_folio,
588	.release_folio		= iomap_release_folio,
589	.invalidate_folio	= iomap_invalidate_folio,
590	.read_folio		= blkdev_read_folio,
591	.readahead		= blkdev_readahead,
592	.writepages		= blkdev_writepages,
593	.is_partially_uptodate  = iomap_is_partially_uptodate,
594	.error_remove_folio	= generic_error_remove_folio,
595	.migrate_folio		= filemap_migrate_folio,
596};
597#endif /* CONFIG_BUFFER_HEAD */
598
599/*
600 * for a block special file file_inode(file)->i_size is zero
601 * so we compute the size by hand (just as in block_read/write above)
602 */
603static loff_t blkdev_llseek(struct file *file, loff_t offset, int whence)
604{
605	struct inode *bd_inode = bdev_file_inode(file);
606	loff_t retval;
607
608	inode_lock(bd_inode);
609	retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
610	inode_unlock(bd_inode);
611	return retval;
612}
613
614static int blkdev_fsync(struct file *filp, loff_t start, loff_t end,
615		int datasync)
616{
617	struct block_device *bdev = I_BDEV(filp->f_mapping->host);
618	int error;
619
620	error = file_write_and_wait_range(filp, start, end);
621	if (error)
622		return error;
623
624	/*
625	 * There is no need to serialise calls to blkdev_issue_flush with
626	 * i_mutex and doing so causes performance issues with concurrent
627	 * O_SYNC writers to a block device.
628	 */
629	error = blkdev_issue_flush(bdev);
630	if (error == -EOPNOTSUPP)
631		error = 0;
632
633	return error;
634}
635
636/**
637 * file_to_blk_mode - get block open flags from file flags
638 * @file: file whose open flags should be converted
639 *
640 * Look at file open flags and generate corresponding block open flags from
641 * them. The function works both for file just being open (e.g. during ->open
642 * callback) and for file that is already open. This is actually non-trivial
643 * (see comment in the function).
644 */
645blk_mode_t file_to_blk_mode(struct file *file)
646{
647	blk_mode_t mode = 0;
648
649	if (file->f_mode & FMODE_READ)
650		mode |= BLK_OPEN_READ;
651	if (file->f_mode & FMODE_WRITE)
652		mode |= BLK_OPEN_WRITE;
653	/*
654	 * do_dentry_open() clears O_EXCL from f_flags, use file->private_data
655	 * to determine whether the open was exclusive for already open files.
656	 */
657	if (file->private_data)
658		mode |= BLK_OPEN_EXCL;
659	else if (file->f_flags & O_EXCL)
660		mode |= BLK_OPEN_EXCL;
661	if (file->f_flags & O_NDELAY)
662		mode |= BLK_OPEN_NDELAY;
663
664	/*
665	 * If all bits in O_ACCMODE set (aka O_RDWR | O_WRONLY), the floppy
666	 * driver has historically allowed ioctls as if the file was opened for
667	 * writing, but does not allow and actual reads or writes.
668	 */
669	if ((file->f_flags & O_ACCMODE) == (O_RDWR | O_WRONLY))
670		mode |= BLK_OPEN_WRITE_IOCTL;
671
672	return mode;
673}
674
675static int blkdev_open(struct inode *inode, struct file *filp)
676{
677	struct block_device *bdev;
678	blk_mode_t mode;
679	int ret;
680
681	mode = file_to_blk_mode(filp);
682	/* Use the file as the holder. */
683	if (mode & BLK_OPEN_EXCL)
684		filp->private_data = filp;
685	ret = bdev_permission(inode->i_rdev, mode, filp->private_data);
686	if (ret)
687		return ret;
688
689	bdev = blkdev_get_no_open(inode->i_rdev, true);
690	if (!bdev)
691		return -ENXIO;
692
693	if (bdev_can_atomic_write(bdev))
694		filp->f_mode |= FMODE_CAN_ATOMIC_WRITE;
695	if (blk_get_integrity(bdev->bd_disk))
696		filp->f_mode |= FMODE_HAS_METADATA;
697
698	ret = bdev_open(bdev, mode, filp->private_data, NULL, filp);
699	if (ret)
700		blkdev_put_no_open(bdev);
701	return ret;
702}
703
704static int blkdev_release(struct inode *inode, struct file *filp)
705{
706	bdev_release(filp);
707	return 0;
708}
709
710static ssize_t
711blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from)
712{
713	size_t count = iov_iter_count(from);
714	ssize_t written;
715
716	written = kiocb_invalidate_pages(iocb, count);
717	if (written) {
718		if (written == -EBUSY)
719			return 0;
720		return written;
721	}
722
723	written = blkdev_direct_IO(iocb, from);
724	if (written > 0) {
725		kiocb_invalidate_post_direct_write(iocb, count);
726		iocb->ki_pos += written;
727		count -= written;
728	}
729	if (written != -EIOCBQUEUED)
730		iov_iter_revert(from, count - iov_iter_count(from));
731	return written;
732}
733
734static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from)
735{
736	return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops, NULL,
737			NULL);
738}
739
740/*
741 * Write data to the block device.  Only intended for the block device itself
742 * and the raw driver which basically is a fake block device.
743 *
744 * Does not take i_mutex for the write and thus is not for general purpose
745 * use.
746 */
747static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
748{
749	struct file *file = iocb->ki_filp;
750	struct inode *bd_inode = bdev_file_inode(file);
751	struct block_device *bdev = I_BDEV(bd_inode);
752	bool atomic = iocb->ki_flags & IOCB_ATOMIC;
753	loff_t size = bdev_nr_bytes(bdev);
754	size_t shorted = 0;
755	ssize_t ret;
756
757	if (bdev_read_only(bdev))
758		return -EPERM;
759
760	if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
761		return -ETXTBSY;
762
763	if (!iov_iter_count(from))
764		return 0;
765
766	if (iocb->ki_pos >= size)
767		return -ENOSPC;
768
769	if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
770		return -EOPNOTSUPP;
771
772	if (atomic) {
773		ret = generic_atomic_write_valid(iocb, from);
774		if (ret)
775			return ret;
776	}
777
778	size -= iocb->ki_pos;
779	if (iov_iter_count(from) > size) {
780		if (atomic)
781			return -EINVAL;
782		shorted = iov_iter_count(from) - size;
783		iov_iter_truncate(from, size);
784	}
785
786	ret = file_update_time(file);
787	if (ret)
788		return ret;
789
790	if (iocb->ki_flags & IOCB_DIRECT) {
791		ret = blkdev_direct_write(iocb, from);
792		if (ret >= 0 && iov_iter_count(from))
793			ret = direct_write_fallback(iocb, from, ret,
794					blkdev_buffered_write(iocb, from));
795	} else {
796		/*
797		 * Take i_rwsem and invalidate_lock to avoid racing with
798		 * set_blocksize changing i_blkbits/folio order and punching
799		 * out the pagecache.
800		 */
801		inode_lock_shared(bd_inode);
802		ret = blkdev_buffered_write(iocb, from);
803		inode_unlock_shared(bd_inode);
804	}
805
806	if (ret > 0)
807		ret = generic_write_sync(iocb, ret);
808	iov_iter_reexpand(from, iov_iter_count(from) + shorted);
809	return ret;
810}
811
812static ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
813{
814	struct inode *bd_inode = bdev_file_inode(iocb->ki_filp);
815	struct block_device *bdev = I_BDEV(iocb->ki_filp->f_mapping->host);
816	loff_t size = bdev_nr_bytes(bdev);
817	loff_t pos = iocb->ki_pos;
818	size_t shorted = 0;
819	ssize_t ret = 0;
820	size_t count;
821
822	if (unlikely(pos + iov_iter_count(to) > size)) {
823		if (pos >= size)
824			return 0;
825		size -= pos;
826		shorted = iov_iter_count(to) - size;
827		iov_iter_truncate(to, size);
828	}
829
830	count = iov_iter_count(to);
831	if (!count)
832		goto reexpand; /* skip atime */
833
834	if (iocb->ki_flags & IOCB_DIRECT) {
835		ret = kiocb_write_and_wait(iocb, count);
836		if (ret < 0)
837			goto reexpand;
838		file_accessed(iocb->ki_filp);
839
840		ret = blkdev_direct_IO(iocb, to);
841		if (ret > 0) {
842			iocb->ki_pos += ret;
843			count -= ret;
844		}
845		if (ret != -EIOCBQUEUED)
846			iov_iter_revert(to, count - iov_iter_count(to));
847		if (ret < 0 || !count)
848			goto reexpand;
849	}
850
851	/*
852	 * Take i_rwsem and invalidate_lock to avoid racing with set_blocksize
853	 * changing i_blkbits/folio order and punching out the pagecache.
854	 */
855	inode_lock_shared(bd_inode);
856	ret = filemap_read(iocb, to, ret);
857	inode_unlock_shared(bd_inode);
858
859reexpand:
860	if (unlikely(shorted))
861		iov_iter_reexpand(to, iov_iter_count(to) + shorted);
862	return ret;
863}
864
865#define	BLKDEV_FALLOC_FL_SUPPORTED					\
866		(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |		\
867		 FALLOC_FL_ZERO_RANGE | FALLOC_FL_WRITE_ZEROES)
868
869static long blkdev_fallocate(struct file *file, int mode, loff_t start,
870			     loff_t len)
871{
872	struct inode *inode = bdev_file_inode(file);
873	struct block_device *bdev = I_BDEV(inode);
874	loff_t end = start + len - 1;
875	loff_t isize;
876	unsigned int flags;
877	int error;
878
879	/* Fail if we don't recognize the flags. */
880	if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
881		return -EOPNOTSUPP;
882	/*
883	 * Don't allow writing zeroes if the device does not enable the
884	 * unmap write zeroes operation.
885	 */
886	if ((mode & FALLOC_FL_WRITE_ZEROES) &&
887	    !bdev_write_zeroes_unmap_sectors(bdev))
888		return -EOPNOTSUPP;
889
890	/* Don't go off the end of the device. */
891	isize = bdev_nr_bytes(bdev);
892	if (start >= isize)
893		return -EINVAL;
894	if (end >= isize) {
895		if (mode & FALLOC_FL_KEEP_SIZE) {
896			len = isize - start;
897			end = start + len - 1;
898		} else
899			return -EINVAL;
900	}
901
902	/*
903	 * Don't allow IO that isn't aligned to logical block size.
904	 */
905	if ((start | len) & (bdev_logical_block_size(bdev) - 1))
906		return -EINVAL;
907
908	inode_lock(inode);
909	filemap_invalidate_lock(inode->i_mapping);
910
911	switch (mode) {
912	case FALLOC_FL_ZERO_RANGE:
913	case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
914		flags = BLKDEV_ZERO_NOUNMAP;
915		break;
916	case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
917		flags = BLKDEV_ZERO_NOFALLBACK;
918		break;
919	case FALLOC_FL_WRITE_ZEROES:
920		flags = 0;
921		break;
922	default:
923		error = -EOPNOTSUPP;
924		goto fail;
925	}
926
927	/*
928	 * Invalidate the page cache, including dirty pages, for valid
929	 * de-allocate mode calls to fallocate().
930	 */
931	error = truncate_bdev_range(bdev, file_to_blk_mode(file), start, end);
932	if (error)
933		goto fail;
934
935	error = blkdev_issue_zeroout(bdev, start >> SECTOR_SHIFT,
936				     len >> SECTOR_SHIFT, GFP_KERNEL, flags);
937 fail:
938	filemap_invalidate_unlock(inode->i_mapping);
939	inode_unlock(inode);
940	return error;
941}
942
943static int blkdev_mmap_prepare(struct vm_area_desc *desc)
944{
945	struct file *file = desc->file;
946
947	if (bdev_read_only(I_BDEV(bdev_file_inode(file))))
948		return generic_file_readonly_mmap_prepare(desc);
949
950	return generic_file_mmap_prepare(desc);
951}
952
953const struct file_operations def_blk_fops = {
954	.open		= blkdev_open,
955	.release	= blkdev_release,
956	.llseek		= blkdev_llseek,
957	.read_iter	= blkdev_read_iter,
958	.write_iter	= blkdev_write_iter,
959	.iopoll		= iocb_bio_iopoll,
960	.mmap_prepare	= blkdev_mmap_prepare,
961	.fsync		= blkdev_fsync,
962	.unlocked_ioctl	= blkdev_ioctl,
963#ifdef CONFIG_COMPAT
964	.compat_ioctl	= compat_blkdev_ioctl,
965#endif
966	.splice_read	= filemap_splice_read,
967	.splice_write	= iter_file_splice_write,
968	.fallocate	= blkdev_fallocate,
969	.uring_cmd	= blkdev_uring_cmd,
970	.fop_flags	= FOP_BUFFER_RASYNC,
971};
972
973static __init int blkdev_init(void)
974{
975	return bioset_init(&blkdev_dio_pool, 4,
976				offsetof(struct blkdev_dio, bio),
977				BIOSET_NEED_BVECS|BIOSET_PERCPU_CACHE);
978}
979module_init(blkdev_init);