• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/block_dev.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
7  */
8 
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/pseudo_fs.h>
29 #include <linux/uio.h>
30 #include <linux/namei.h>
31 #include <linux/log2.h>
32 #include <linux/cleancache.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
37 #include "internal.h"
38 
39 struct bdev_inode {
40 	struct block_device bdev;
41 	struct inode vfs_inode;
42 };
43 
44 static const struct address_space_operations def_blk_aops;
45 
BDEV_I(struct inode * inode)46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
47 {
48 	return container_of(inode, struct bdev_inode, vfs_inode);
49 }
50 
I_BDEV(struct inode * inode)51 struct block_device *I_BDEV(struct inode *inode)
52 {
53 	return &BDEV_I(inode)->bdev;
54 }
55 EXPORT_SYMBOL(I_BDEV);
56 
bdev_write_inode(struct block_device * bdev)57 static void bdev_write_inode(struct block_device *bdev)
58 {
59 	struct inode *inode = bdev->bd_inode;
60 	int ret;
61 
62 	spin_lock(&inode->i_lock);
63 	while (inode->i_state & I_DIRTY) {
64 		spin_unlock(&inode->i_lock);
65 		ret = write_inode_now(inode, true);
66 		if (ret) {
67 			char name[BDEVNAME_SIZE];
68 			pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 					    "for block device %s (err=%d).\n",
70 					    bdevname(bdev, name), ret);
71 		}
72 		spin_lock(&inode->i_lock);
73 	}
74 	spin_unlock(&inode->i_lock);
75 }
76 
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
kill_bdev(struct block_device * bdev)78 static void kill_bdev(struct block_device *bdev)
79 {
80 	struct address_space *mapping = bdev->bd_inode->i_mapping;
81 
82 	if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
83 		return;
84 
85 	invalidate_bh_lrus();
86 	truncate_inode_pages(mapping, 0);
87 }
88 
89 /* Invalidate clean unused buffers and pagecache. */
invalidate_bdev(struct block_device * bdev)90 void invalidate_bdev(struct block_device *bdev)
91 {
92 	struct address_space *mapping = bdev->bd_inode->i_mapping;
93 
94 	if (mapping->nrpages) {
95 		invalidate_bh_lrus();
96 		lru_add_drain_all();	/* make sure all lru add caches are flushed */
97 		invalidate_mapping_pages(mapping, 0, -1);
98 	}
99 	/* 99% of the time, we don't need to flush the cleancache on the bdev.
100 	 * But, for the strange corners, lets be cautious
101 	 */
102 	cleancache_invalidate_inode(mapping);
103 }
104 EXPORT_SYMBOL(invalidate_bdev);
105 
106 /*
107  * Drop all buffers & page cache for given bdev range. This function bails
108  * with error if bdev has other exclusive owner (such as filesystem).
109  */
truncate_bdev_range(struct block_device * bdev,fmode_t mode,loff_t lstart,loff_t lend)110 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
111 			loff_t lstart, loff_t lend)
112 {
113 	struct block_device *claimed_bdev = NULL;
114 	int err;
115 
116 	/*
117 	 * If we don't hold exclusive handle for the device, upgrade to it
118 	 * while we discard the buffer cache to avoid discarding buffers
119 	 * under live filesystem.
120 	 */
121 	if (!(mode & FMODE_EXCL)) {
122 		claimed_bdev = bdev->bd_contains;
123 		err = bd_prepare_to_claim(bdev, claimed_bdev,
124 					  truncate_bdev_range);
125 		if (err)
126 			goto invalidate;
127 	}
128 	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
129 	if (claimed_bdev)
130 		bd_abort_claiming(bdev, claimed_bdev, truncate_bdev_range);
131 	return 0;
132 
133 invalidate:
134 	/*
135 	 * Someone else has handle exclusively open. Try invalidating instead.
136 	 * The 'end' argument is inclusive so the rounding is safe.
137 	 */
138 	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
139 					     lstart >> PAGE_SHIFT,
140 					     lend >> PAGE_SHIFT);
141 }
142 EXPORT_SYMBOL(truncate_bdev_range);
143 
set_init_blocksize(struct block_device * bdev)144 static void set_init_blocksize(struct block_device *bdev)
145 {
146 	unsigned int bsize = bdev_logical_block_size(bdev);
147 	loff_t size = i_size_read(bdev->bd_inode);
148 
149 	while (bsize < PAGE_SIZE) {
150 		if (size & bsize)
151 			break;
152 		bsize <<= 1;
153 	}
154 	bdev->bd_inode->i_blkbits = blksize_bits(bsize);
155 }
156 
set_blocksize(struct block_device * bdev,int size)157 int set_blocksize(struct block_device *bdev, int size)
158 {
159 	/* Size must be a power of two, and between 512 and PAGE_SIZE */
160 	if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
161 		return -EINVAL;
162 
163 	/* Size cannot be smaller than the size supported by the device */
164 	if (size < bdev_logical_block_size(bdev))
165 		return -EINVAL;
166 
167 	/* Don't change the size if it is same as current */
168 	if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
169 		sync_blockdev(bdev);
170 		bdev->bd_inode->i_blkbits = blksize_bits(size);
171 		kill_bdev(bdev);
172 	}
173 	return 0;
174 }
175 
176 EXPORT_SYMBOL(set_blocksize);
177 
sb_set_blocksize(struct super_block * sb,int size)178 int sb_set_blocksize(struct super_block *sb, int size)
179 {
180 	if (set_blocksize(sb->s_bdev, size))
181 		return 0;
182 	/* If we get here, we know size is power of two
183 	 * and it's value is between 512 and PAGE_SIZE */
184 	sb->s_blocksize = size;
185 	sb->s_blocksize_bits = blksize_bits(size);
186 	return sb->s_blocksize;
187 }
188 
189 EXPORT_SYMBOL(sb_set_blocksize);
190 
sb_min_blocksize(struct super_block * sb,int size)191 int sb_min_blocksize(struct super_block *sb, int size)
192 {
193 	int minsize = bdev_logical_block_size(sb->s_bdev);
194 	if (size < minsize)
195 		size = minsize;
196 	return sb_set_blocksize(sb, size);
197 }
198 
199 EXPORT_SYMBOL(sb_min_blocksize);
200 
201 static int
blkdev_get_block(struct inode * inode,sector_t iblock,struct buffer_head * bh,int create)202 blkdev_get_block(struct inode *inode, sector_t iblock,
203 		struct buffer_head *bh, int create)
204 {
205 	bh->b_bdev = I_BDEV(inode);
206 	bh->b_blocknr = iblock;
207 	set_buffer_mapped(bh);
208 	return 0;
209 }
210 
bdev_file_inode(struct file * file)211 static struct inode *bdev_file_inode(struct file *file)
212 {
213 	return file->f_mapping->host;
214 }
215 
dio_bio_write_op(struct kiocb * iocb)216 static unsigned int dio_bio_write_op(struct kiocb *iocb)
217 {
218 	unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
219 
220 	/* avoid the need for a I/O completion work item */
221 	if (iocb->ki_flags & IOCB_DSYNC)
222 		op |= REQ_FUA;
223 	return op;
224 }
225 
226 #define DIO_INLINE_BIO_VECS 4
227 
blkdev_bio_end_io_simple(struct bio * bio)228 static void blkdev_bio_end_io_simple(struct bio *bio)
229 {
230 	struct task_struct *waiter = bio->bi_private;
231 
232 	WRITE_ONCE(bio->bi_private, NULL);
233 	blk_wake_io_task(waiter);
234 }
235 
236 static ssize_t
__blkdev_direct_IO_simple(struct kiocb * iocb,struct iov_iter * iter,int nr_pages)237 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
238 		int nr_pages)
239 {
240 	struct file *file = iocb->ki_filp;
241 	struct block_device *bdev = I_BDEV(bdev_file_inode(file));
242 	struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
243 	loff_t pos = iocb->ki_pos;
244 	bool should_dirty = false;
245 	struct bio bio;
246 	ssize_t ret;
247 	blk_qc_t qc;
248 
249 	if ((pos | iov_iter_alignment(iter)) &
250 	    (bdev_logical_block_size(bdev) - 1))
251 		return -EINVAL;
252 
253 	if (nr_pages <= DIO_INLINE_BIO_VECS)
254 		vecs = inline_vecs;
255 	else {
256 		vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
257 				     GFP_KERNEL);
258 		if (!vecs)
259 			return -ENOMEM;
260 	}
261 
262 	bio_init(&bio, vecs, nr_pages);
263 	bio_set_dev(&bio, bdev);
264 	bio.bi_iter.bi_sector = pos >> 9;
265 	bio.bi_write_hint = iocb->ki_hint;
266 	bio.bi_private = current;
267 	bio.bi_end_io = blkdev_bio_end_io_simple;
268 	bio.bi_ioprio = iocb->ki_ioprio;
269 
270 	ret = bio_iov_iter_get_pages(&bio, iter);
271 	if (unlikely(ret))
272 		goto out;
273 	ret = bio.bi_iter.bi_size;
274 
275 	if (iov_iter_rw(iter) == READ) {
276 		bio.bi_opf = REQ_OP_READ;
277 		if (iter_is_iovec(iter))
278 			should_dirty = true;
279 	} else {
280 		bio.bi_opf = dio_bio_write_op(iocb);
281 		task_io_account_write(ret);
282 	}
283 	if (iocb->ki_flags & IOCB_NOWAIT)
284 		bio.bi_opf |= REQ_NOWAIT;
285 	if (iocb->ki_flags & IOCB_HIPRI)
286 		bio_set_polled(&bio, iocb);
287 
288 	qc = submit_bio(&bio);
289 	for (;;) {
290 		set_current_state(TASK_UNINTERRUPTIBLE);
291 		if (!READ_ONCE(bio.bi_private))
292 			break;
293 		if (!(iocb->ki_flags & IOCB_HIPRI) ||
294 		    !blk_poll(bdev_get_queue(bdev), qc, true))
295 			blk_io_schedule();
296 	}
297 	__set_current_state(TASK_RUNNING);
298 
299 	bio_release_pages(&bio, should_dirty);
300 	if (unlikely(bio.bi_status))
301 		ret = blk_status_to_errno(bio.bi_status);
302 
303 out:
304 	if (vecs != inline_vecs)
305 		kfree(vecs);
306 
307 	bio_uninit(&bio);
308 
309 	return ret;
310 }
311 
312 struct blkdev_dio {
313 	union {
314 		struct kiocb		*iocb;
315 		struct task_struct	*waiter;
316 	};
317 	size_t			size;
318 	atomic_t		ref;
319 	bool			multi_bio : 1;
320 	bool			should_dirty : 1;
321 	bool			is_sync : 1;
322 	struct bio		bio;
323 };
324 
325 static struct bio_set blkdev_dio_pool;
326 
blkdev_iopoll(struct kiocb * kiocb,bool wait)327 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
328 {
329 	struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
330 	struct request_queue *q = bdev_get_queue(bdev);
331 
332 	return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
333 }
334 
blkdev_bio_end_io(struct bio * bio)335 static void blkdev_bio_end_io(struct bio *bio)
336 {
337 	struct blkdev_dio *dio = bio->bi_private;
338 	bool should_dirty = dio->should_dirty;
339 
340 	if (bio->bi_status && !dio->bio.bi_status)
341 		dio->bio.bi_status = bio->bi_status;
342 
343 	if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
344 		if (!dio->is_sync) {
345 			struct kiocb *iocb = dio->iocb;
346 			ssize_t ret;
347 
348 			if (likely(!dio->bio.bi_status)) {
349 				ret = dio->size;
350 				iocb->ki_pos += ret;
351 			} else {
352 				ret = blk_status_to_errno(dio->bio.bi_status);
353 			}
354 
355 			dio->iocb->ki_complete(iocb, ret, 0);
356 			if (dio->multi_bio)
357 				bio_put(&dio->bio);
358 		} else {
359 			struct task_struct *waiter = dio->waiter;
360 
361 			WRITE_ONCE(dio->waiter, NULL);
362 			blk_wake_io_task(waiter);
363 		}
364 	}
365 
366 	if (should_dirty) {
367 		bio_check_pages_dirty(bio);
368 	} else {
369 		bio_release_pages(bio, false);
370 		bio_put(bio);
371 	}
372 }
373 
374 static ssize_t
__blkdev_direct_IO(struct kiocb * iocb,struct iov_iter * iter,int nr_pages)375 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
376 {
377 	struct file *file = iocb->ki_filp;
378 	struct inode *inode = bdev_file_inode(file);
379 	struct block_device *bdev = I_BDEV(inode);
380 	struct blk_plug plug;
381 	struct blkdev_dio *dio;
382 	struct bio *bio;
383 	bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
384 	bool is_read = (iov_iter_rw(iter) == READ), is_sync;
385 	loff_t pos = iocb->ki_pos;
386 	blk_qc_t qc = BLK_QC_T_NONE;
387 	int ret = 0;
388 
389 	if ((pos | iov_iter_alignment(iter)) &
390 	    (bdev_logical_block_size(bdev) - 1))
391 		return -EINVAL;
392 
393 	bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
394 
395 	dio = container_of(bio, struct blkdev_dio, bio);
396 	dio->is_sync = is_sync = is_sync_kiocb(iocb);
397 	if (dio->is_sync) {
398 		dio->waiter = current;
399 		bio_get(bio);
400 	} else {
401 		dio->iocb = iocb;
402 	}
403 
404 	dio->size = 0;
405 	dio->multi_bio = false;
406 	dio->should_dirty = is_read && iter_is_iovec(iter);
407 
408 	/*
409 	 * Don't plug for HIPRI/polled IO, as those should go straight
410 	 * to issue
411 	 */
412 	if (!is_poll)
413 		blk_start_plug(&plug);
414 
415 	for (;;) {
416 		bio_set_dev(bio, bdev);
417 		bio->bi_iter.bi_sector = pos >> 9;
418 		bio->bi_write_hint = iocb->ki_hint;
419 		bio->bi_private = dio;
420 		bio->bi_end_io = blkdev_bio_end_io;
421 		bio->bi_ioprio = iocb->ki_ioprio;
422 
423 		ret = bio_iov_iter_get_pages(bio, iter);
424 		if (unlikely(ret)) {
425 			bio->bi_status = BLK_STS_IOERR;
426 			bio_endio(bio);
427 			break;
428 		}
429 
430 		if (is_read) {
431 			bio->bi_opf = REQ_OP_READ;
432 			if (dio->should_dirty)
433 				bio_set_pages_dirty(bio);
434 		} else {
435 			bio->bi_opf = dio_bio_write_op(iocb);
436 			task_io_account_write(bio->bi_iter.bi_size);
437 		}
438 		if (iocb->ki_flags & IOCB_NOWAIT)
439 			bio->bi_opf |= REQ_NOWAIT;
440 
441 		dio->size += bio->bi_iter.bi_size;
442 		pos += bio->bi_iter.bi_size;
443 
444 		nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
445 		if (!nr_pages) {
446 			bool polled = false;
447 
448 			if (iocb->ki_flags & IOCB_HIPRI) {
449 				bio_set_polled(bio, iocb);
450 				polled = true;
451 			}
452 
453 			qc = submit_bio(bio);
454 
455 			if (polled)
456 				WRITE_ONCE(iocb->ki_cookie, qc);
457 			break;
458 		}
459 
460 		if (!dio->multi_bio) {
461 			/*
462 			 * AIO needs an extra reference to ensure the dio
463 			 * structure which is embedded into the first bio
464 			 * stays around.
465 			 */
466 			if (!is_sync)
467 				bio_get(bio);
468 			dio->multi_bio = true;
469 			atomic_set(&dio->ref, 2);
470 		} else {
471 			atomic_inc(&dio->ref);
472 		}
473 
474 		submit_bio(bio);
475 		bio = bio_alloc(GFP_KERNEL, nr_pages);
476 	}
477 
478 	if (!is_poll)
479 		blk_finish_plug(&plug);
480 
481 	if (!is_sync)
482 		return -EIOCBQUEUED;
483 
484 	for (;;) {
485 		set_current_state(TASK_UNINTERRUPTIBLE);
486 		if (!READ_ONCE(dio->waiter))
487 			break;
488 
489 		if (!(iocb->ki_flags & IOCB_HIPRI) ||
490 		    !blk_poll(bdev_get_queue(bdev), qc, true))
491 			blk_io_schedule();
492 	}
493 	__set_current_state(TASK_RUNNING);
494 
495 	if (!ret)
496 		ret = blk_status_to_errno(dio->bio.bi_status);
497 	if (likely(!ret))
498 		ret = dio->size;
499 
500 	bio_put(&dio->bio);
501 	return ret;
502 }
503 
504 static ssize_t
blkdev_direct_IO(struct kiocb * iocb,struct iov_iter * iter)505 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
506 {
507 	int nr_pages;
508 
509 	nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
510 	if (!nr_pages)
511 		return 0;
512 	if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
513 		return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
514 
515 	return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
516 }
517 
blkdev_init(void)518 static __init int blkdev_init(void)
519 {
520 	return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
521 }
522 module_init(blkdev_init);
523 
__sync_blockdev(struct block_device * bdev,int wait)524 int __sync_blockdev(struct block_device *bdev, int wait)
525 {
526 	if (!bdev)
527 		return 0;
528 	if (!wait)
529 		return filemap_flush(bdev->bd_inode->i_mapping);
530 	return filemap_write_and_wait(bdev->bd_inode->i_mapping);
531 }
532 
533 /*
534  * Write out and wait upon all the dirty data associated with a block
535  * device via its mapping.  Does not take the superblock lock.
536  */
sync_blockdev(struct block_device * bdev)537 int sync_blockdev(struct block_device *bdev)
538 {
539 	return __sync_blockdev(bdev, 1);
540 }
541 EXPORT_SYMBOL(sync_blockdev);
542 
543 /*
544  * Write out and wait upon all dirty data associated with this
545  * device.   Filesystem data as well as the underlying block
546  * device.  Takes the superblock lock.
547  */
fsync_bdev(struct block_device * bdev)548 int fsync_bdev(struct block_device *bdev)
549 {
550 	struct super_block *sb = get_super(bdev);
551 	if (sb) {
552 		int res = sync_filesystem(sb);
553 		drop_super(sb);
554 		return res;
555 	}
556 	return sync_blockdev(bdev);
557 }
558 EXPORT_SYMBOL(fsync_bdev);
559 
560 /**
561  * freeze_bdev  --  lock a filesystem and force it into a consistent state
562  * @bdev:	blockdevice to lock
563  *
564  * If a superblock is found on this device, we take the s_umount semaphore
565  * on it to make sure nobody unmounts until the snapshot creation is done.
566  * The reference counter (bd_fsfreeze_count) guarantees that only the last
567  * unfreeze process can unfreeze the frozen filesystem actually when multiple
568  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
569  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
570  * actually.
571  */
freeze_bdev(struct block_device * bdev)572 struct super_block *freeze_bdev(struct block_device *bdev)
573 {
574 	struct super_block *sb;
575 	int error = 0;
576 
577 	mutex_lock(&bdev->bd_fsfreeze_mutex);
578 	if (++bdev->bd_fsfreeze_count > 1) {
579 		/*
580 		 * We don't even need to grab a reference - the first call
581 		 * to freeze_bdev grab an active reference and only the last
582 		 * thaw_bdev drops it.
583 		 */
584 		sb = get_super(bdev);
585 		if (sb)
586 			drop_super(sb);
587 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
588 		return sb;
589 	}
590 
591 	sb = get_active_super(bdev);
592 	if (!sb)
593 		goto out;
594 	if (sb->s_op->freeze_super)
595 		error = sb->s_op->freeze_super(sb);
596 	else
597 		error = freeze_super(sb);
598 	if (error) {
599 		deactivate_super(sb);
600 		bdev->bd_fsfreeze_count--;
601 		mutex_unlock(&bdev->bd_fsfreeze_mutex);
602 		return ERR_PTR(error);
603 	}
604 	deactivate_super(sb);
605  out:
606 	sync_blockdev(bdev);
607 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
608 	return sb;	/* thaw_bdev releases s->s_umount */
609 }
610 EXPORT_SYMBOL(freeze_bdev);
611 
612 /**
613  * thaw_bdev  -- unlock filesystem
614  * @bdev:	blockdevice to unlock
615  * @sb:		associated superblock
616  *
617  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
618  */
thaw_bdev(struct block_device * bdev,struct super_block * sb)619 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
620 {
621 	int error = -EINVAL;
622 
623 	mutex_lock(&bdev->bd_fsfreeze_mutex);
624 	if (!bdev->bd_fsfreeze_count)
625 		goto out;
626 
627 	error = 0;
628 	if (--bdev->bd_fsfreeze_count > 0)
629 		goto out;
630 
631 	if (!sb)
632 		goto out;
633 
634 	if (sb->s_op->thaw_super)
635 		error = sb->s_op->thaw_super(sb);
636 	else
637 		error = thaw_super(sb);
638 	if (error)
639 		bdev->bd_fsfreeze_count++;
640 out:
641 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
642 	return error;
643 }
644 EXPORT_SYMBOL(thaw_bdev);
645 
blkdev_writepage(struct page * page,struct writeback_control * wbc)646 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
647 {
648 	return block_write_full_page(page, blkdev_get_block, wbc);
649 }
650 
blkdev_readpage(struct file * file,struct page * page)651 static int blkdev_readpage(struct file * file, struct page * page)
652 {
653 	return block_read_full_page(page, blkdev_get_block);
654 }
655 
blkdev_readahead(struct readahead_control * rac)656 static void blkdev_readahead(struct readahead_control *rac)
657 {
658 	mpage_readahead(rac, blkdev_get_block);
659 }
660 
blkdev_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** pagep,void ** fsdata)661 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
662 			loff_t pos, unsigned len, unsigned flags,
663 			struct page **pagep, void **fsdata)
664 {
665 	return block_write_begin(mapping, pos, len, flags, pagep,
666 				 blkdev_get_block);
667 }
668 
blkdev_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)669 static int blkdev_write_end(struct file *file, struct address_space *mapping,
670 			loff_t pos, unsigned len, unsigned copied,
671 			struct page *page, void *fsdata)
672 {
673 	int ret;
674 	ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
675 
676 	unlock_page(page);
677 	put_page(page);
678 
679 	return ret;
680 }
681 
682 /*
683  * private llseek:
684  * for a block special file file_inode(file)->i_size is zero
685  * so we compute the size by hand (just as in block_read/write above)
686  */
block_llseek(struct file * file,loff_t offset,int whence)687 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
688 {
689 	struct inode *bd_inode = bdev_file_inode(file);
690 	loff_t retval;
691 
692 	inode_lock(bd_inode);
693 	retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
694 	inode_unlock(bd_inode);
695 	return retval;
696 }
697 
blkdev_fsync(struct file * filp,loff_t start,loff_t end,int datasync)698 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
699 {
700 	struct inode *bd_inode = bdev_file_inode(filp);
701 	struct block_device *bdev = I_BDEV(bd_inode);
702 	int error;
703 
704 	error = file_write_and_wait_range(filp, start, end);
705 	if (error)
706 		return error;
707 
708 	/*
709 	 * There is no need to serialise calls to blkdev_issue_flush with
710 	 * i_mutex and doing so causes performance issues with concurrent
711 	 * O_SYNC writers to a block device.
712 	 */
713 	error = blkdev_issue_flush(bdev, GFP_KERNEL);
714 	if (error == -EOPNOTSUPP)
715 		error = 0;
716 
717 	return error;
718 }
719 EXPORT_SYMBOL(blkdev_fsync);
720 
721 /**
722  * bdev_read_page() - Start reading a page from a block device
723  * @bdev: The device to read the page from
724  * @sector: The offset on the device to read the page to (need not be aligned)
725  * @page: The page to read
726  *
727  * On entry, the page should be locked.  It will be unlocked when the page
728  * has been read.  If the block driver implements rw_page synchronously,
729  * that will be true on exit from this function, but it need not be.
730  *
731  * Errors returned by this function are usually "soft", eg out of memory, or
732  * queue full; callers should try a different route to read this page rather
733  * than propagate an error back up the stack.
734  *
735  * Return: negative errno if an error occurs, 0 if submission was successful.
736  */
bdev_read_page(struct block_device * bdev,sector_t sector,struct page * page)737 int bdev_read_page(struct block_device *bdev, sector_t sector,
738 			struct page *page)
739 {
740 	const struct block_device_operations *ops = bdev->bd_disk->fops;
741 	int result = -EOPNOTSUPP;
742 
743 	if (!ops->rw_page || bdev_get_integrity(bdev))
744 		return result;
745 
746 	result = blk_queue_enter(bdev->bd_disk->queue, 0);
747 	if (result)
748 		return result;
749 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
750 			      REQ_OP_READ);
751 	blk_queue_exit(bdev->bd_disk->queue);
752 	return result;
753 }
754 
755 /**
756  * bdev_write_page() - Start writing a page to a block device
757  * @bdev: The device to write the page to
758  * @sector: The offset on the device to write the page to (need not be aligned)
759  * @page: The page to write
760  * @wbc: The writeback_control for the write
761  *
762  * On entry, the page should be locked and not currently under writeback.
763  * On exit, if the write started successfully, the page will be unlocked and
764  * under writeback.  If the write failed already (eg the driver failed to
765  * queue the page to the device), the page will still be locked.  If the
766  * caller is a ->writepage implementation, it will need to unlock the page.
767  *
768  * Errors returned by this function are usually "soft", eg out of memory, or
769  * queue full; callers should try a different route to write this page rather
770  * than propagate an error back up the stack.
771  *
772  * Return: negative errno if an error occurs, 0 if submission was successful.
773  */
bdev_write_page(struct block_device * bdev,sector_t sector,struct page * page,struct writeback_control * wbc)774 int bdev_write_page(struct block_device *bdev, sector_t sector,
775 			struct page *page, struct writeback_control *wbc)
776 {
777 	int result;
778 	const struct block_device_operations *ops = bdev->bd_disk->fops;
779 
780 	if (!ops->rw_page || bdev_get_integrity(bdev))
781 		return -EOPNOTSUPP;
782 	result = blk_queue_enter(bdev->bd_disk->queue, 0);
783 	if (result)
784 		return result;
785 
786 	set_page_writeback(page);
787 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
788 			      REQ_OP_WRITE);
789 	if (result) {
790 		end_page_writeback(page);
791 	} else {
792 		clean_page_buffers(page);
793 		unlock_page(page);
794 	}
795 	blk_queue_exit(bdev->bd_disk->queue);
796 	return result;
797 }
798 
799 /*
800  * pseudo-fs
801  */
802 
803 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
804 static struct kmem_cache * bdev_cachep __read_mostly;
805 
bdev_alloc_inode(struct super_block * sb)806 static struct inode *bdev_alloc_inode(struct super_block *sb)
807 {
808 	struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
809 	if (!ei)
810 		return NULL;
811 	return &ei->vfs_inode;
812 }
813 
bdev_free_inode(struct inode * inode)814 static void bdev_free_inode(struct inode *inode)
815 {
816 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
817 }
818 
init_once(void * foo)819 static void init_once(void *foo)
820 {
821 	struct bdev_inode *ei = (struct bdev_inode *) foo;
822 	struct block_device *bdev = &ei->bdev;
823 
824 	memset(bdev, 0, sizeof(*bdev));
825 	mutex_init(&bdev->bd_mutex);
826 #ifdef CONFIG_SYSFS
827 	INIT_LIST_HEAD(&bdev->bd_holder_disks);
828 #endif
829 	bdev->bd_bdi = &noop_backing_dev_info;
830 	inode_init_once(&ei->vfs_inode);
831 	/* Initialize mutex for freeze. */
832 	mutex_init(&bdev->bd_fsfreeze_mutex);
833 }
834 
bdev_evict_inode(struct inode * inode)835 static void bdev_evict_inode(struct inode *inode)
836 {
837 	struct block_device *bdev = &BDEV_I(inode)->bdev;
838 	truncate_inode_pages_final(&inode->i_data);
839 	invalidate_inode_buffers(inode); /* is it needed here? */
840 	clear_inode(inode);
841 	/* Detach inode from wb early as bdi_put() may free bdi->wb */
842 	inode_detach_wb(inode);
843 	if (bdev->bd_bdi != &noop_backing_dev_info) {
844 		bdi_put(bdev->bd_bdi);
845 		bdev->bd_bdi = &noop_backing_dev_info;
846 	}
847 }
848 
849 static const struct super_operations bdev_sops = {
850 	.statfs = simple_statfs,
851 	.alloc_inode = bdev_alloc_inode,
852 	.free_inode = bdev_free_inode,
853 	.drop_inode = generic_delete_inode,
854 	.evict_inode = bdev_evict_inode,
855 };
856 
bd_init_fs_context(struct fs_context * fc)857 static int bd_init_fs_context(struct fs_context *fc)
858 {
859 	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
860 	if (!ctx)
861 		return -ENOMEM;
862 	fc->s_iflags |= SB_I_CGROUPWB;
863 	ctx->ops = &bdev_sops;
864 	return 0;
865 }
866 
867 static struct file_system_type bd_type = {
868 	.name		= "bdev",
869 	.init_fs_context = bd_init_fs_context,
870 	.kill_sb	= kill_anon_super,
871 };
872 
873 struct super_block *blockdev_superblock __read_mostly;
874 EXPORT_SYMBOL_GPL(blockdev_superblock);
875 
bdev_cache_init(void)876 void __init bdev_cache_init(void)
877 {
878 	int err;
879 	static struct vfsmount *bd_mnt;
880 
881 	bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
882 			0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
883 				SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
884 			init_once);
885 	err = register_filesystem(&bd_type);
886 	if (err)
887 		panic("Cannot register bdev pseudo-fs");
888 	bd_mnt = kern_mount(&bd_type);
889 	if (IS_ERR(bd_mnt))
890 		panic("Cannot create bdev pseudo-fs");
891 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
892 }
893 
894 /*
895  * Most likely _very_ bad one - but then it's hardly critical for small
896  * /dev and can be fixed when somebody will need really large one.
897  * Keep in mind that it will be fed through icache hash function too.
898  */
hash(dev_t dev)899 static inline unsigned long hash(dev_t dev)
900 {
901 	return MAJOR(dev)+MINOR(dev);
902 }
903 
bdev_test(struct inode * inode,void * data)904 static int bdev_test(struct inode *inode, void *data)
905 {
906 	return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
907 }
908 
bdev_set(struct inode * inode,void * data)909 static int bdev_set(struct inode *inode, void *data)
910 {
911 	BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
912 	return 0;
913 }
914 
bdget(dev_t dev)915 static struct block_device *bdget(dev_t dev)
916 {
917 	struct block_device *bdev;
918 	struct inode *inode;
919 
920 	inode = iget5_locked(blockdev_superblock, hash(dev),
921 			bdev_test, bdev_set, &dev);
922 
923 	if (!inode)
924 		return NULL;
925 
926 	bdev = &BDEV_I(inode)->bdev;
927 
928 	if (inode->i_state & I_NEW) {
929 		spin_lock_init(&bdev->bd_size_lock);
930 		bdev->bd_contains = NULL;
931 		bdev->bd_super = NULL;
932 		bdev->bd_inode = inode;
933 		bdev->bd_part_count = 0;
934 		inode->i_mode = S_IFBLK;
935 		inode->i_rdev = dev;
936 		inode->i_bdev = bdev;
937 		inode->i_data.a_ops = &def_blk_aops;
938 		mapping_set_gfp_mask(&inode->i_data, GFP_USER);
939 		unlock_new_inode(inode);
940 	}
941 	return bdev;
942 }
943 
944 /**
945  * bdgrab -- Grab a reference to an already referenced block device
946  * @bdev:	Block device to grab a reference to.
947  */
bdgrab(struct block_device * bdev)948 struct block_device *bdgrab(struct block_device *bdev)
949 {
950 	ihold(bdev->bd_inode);
951 	return bdev;
952 }
953 EXPORT_SYMBOL(bdgrab);
954 
bdget_part(struct hd_struct * part)955 struct block_device *bdget_part(struct hd_struct *part)
956 {
957 	return bdget(part_devt(part));
958 }
959 
nr_blockdev_pages(void)960 long nr_blockdev_pages(void)
961 {
962 	struct inode *inode;
963 	long ret = 0;
964 
965 	spin_lock(&blockdev_superblock->s_inode_list_lock);
966 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
967 		ret += inode->i_mapping->nrpages;
968 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
969 
970 	return ret;
971 }
972 
bdput(struct block_device * bdev)973 void bdput(struct block_device *bdev)
974 {
975 	iput(bdev->bd_inode);
976 }
977 
978 EXPORT_SYMBOL(bdput);
979 
bd_acquire(struct inode * inode)980 static struct block_device *bd_acquire(struct inode *inode)
981 {
982 	struct block_device *bdev;
983 
984 	spin_lock(&bdev_lock);
985 	bdev = inode->i_bdev;
986 	if (bdev && !inode_unhashed(bdev->bd_inode)) {
987 		bdgrab(bdev);
988 		spin_unlock(&bdev_lock);
989 		return bdev;
990 	}
991 	spin_unlock(&bdev_lock);
992 
993 	/*
994 	 * i_bdev references block device inode that was already shut down
995 	 * (corresponding device got removed).  Remove the reference and look
996 	 * up block device inode again just in case new device got
997 	 * reestablished under the same device number.
998 	 */
999 	if (bdev)
1000 		bd_forget(inode);
1001 
1002 	bdev = bdget(inode->i_rdev);
1003 	if (bdev) {
1004 		spin_lock(&bdev_lock);
1005 		if (!inode->i_bdev) {
1006 			/*
1007 			 * We take an additional reference to bd_inode,
1008 			 * and it's released in clear_inode() of inode.
1009 			 * So, we can access it via ->i_mapping always
1010 			 * without igrab().
1011 			 */
1012 			bdgrab(bdev);
1013 			inode->i_bdev = bdev;
1014 			inode->i_mapping = bdev->bd_inode->i_mapping;
1015 		}
1016 		spin_unlock(&bdev_lock);
1017 	}
1018 	return bdev;
1019 }
1020 
1021 /* Call when you free inode */
1022 
bd_forget(struct inode * inode)1023 void bd_forget(struct inode *inode)
1024 {
1025 	struct block_device *bdev = NULL;
1026 
1027 	spin_lock(&bdev_lock);
1028 	if (!sb_is_blkdev_sb(inode->i_sb))
1029 		bdev = inode->i_bdev;
1030 	inode->i_bdev = NULL;
1031 	inode->i_mapping = &inode->i_data;
1032 	spin_unlock(&bdev_lock);
1033 
1034 	if (bdev)
1035 		bdput(bdev);
1036 }
1037 
1038 /**
1039  * bd_may_claim - test whether a block device can be claimed
1040  * @bdev: block device of interest
1041  * @whole: whole block device containing @bdev, may equal @bdev
1042  * @holder: holder trying to claim @bdev
1043  *
1044  * Test whether @bdev can be claimed by @holder.
1045  *
1046  * CONTEXT:
1047  * spin_lock(&bdev_lock).
1048  *
1049  * RETURNS:
1050  * %true if @bdev can be claimed, %false otherwise.
1051  */
bd_may_claim(struct block_device * bdev,struct block_device * whole,void * holder)1052 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1053 			 void *holder)
1054 {
1055 	if (bdev->bd_holder == holder)
1056 		return true;	 /* already a holder */
1057 	else if (bdev->bd_holder != NULL)
1058 		return false; 	 /* held by someone else */
1059 	else if (whole == bdev)
1060 		return true;  	 /* is a whole device which isn't held */
1061 
1062 	else if (whole->bd_holder == bd_may_claim)
1063 		return true; 	 /* is a partition of a device that is being partitioned */
1064 	else if (whole->bd_holder != NULL)
1065 		return false;	 /* is a partition of a held device */
1066 	else
1067 		return true;	 /* is a partition of an un-held device */
1068 }
1069 
1070 /**
1071  * bd_prepare_to_claim - claim a block device
1072  * @bdev: block device of interest
1073  * @whole: the whole device containing @bdev, may equal @bdev
1074  * @holder: holder trying to claim @bdev
1075  *
1076  * Claim @bdev.  This function fails if @bdev is already claimed by another
1077  * holder and waits if another claiming is in progress. return, the caller
1078  * has ownership of bd_claiming and bd_holder[s].
1079  *
1080  * RETURNS:
1081  * 0 if @bdev can be claimed, -EBUSY otherwise.
1082  */
bd_prepare_to_claim(struct block_device * bdev,struct block_device * whole,void * holder)1083 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1084 		void *holder)
1085 {
1086 retry:
1087 	spin_lock(&bdev_lock);
1088 	/* if someone else claimed, fail */
1089 	if (!bd_may_claim(bdev, whole, holder)) {
1090 		spin_unlock(&bdev_lock);
1091 		return -EBUSY;
1092 	}
1093 
1094 	/* if claiming is already in progress, wait for it to finish */
1095 	if (whole->bd_claiming) {
1096 		wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1097 		DEFINE_WAIT(wait);
1098 
1099 		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1100 		spin_unlock(&bdev_lock);
1101 		schedule();
1102 		finish_wait(wq, &wait);
1103 		goto retry;
1104 	}
1105 
1106 	/* yay, all mine */
1107 	whole->bd_claiming = holder;
1108 	spin_unlock(&bdev_lock);
1109 	return 0;
1110 }
1111 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1112 
bdev_get_gendisk(struct block_device * bdev,int * partno)1113 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1114 {
1115 	struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1116 
1117 	if (!disk)
1118 		return NULL;
1119 	/*
1120 	 * Now that we hold gendisk reference we make sure bdev we looked up is
1121 	 * not stale. If it is, it means device got removed and created before
1122 	 * we looked up gendisk and we fail open in such case. Associating
1123 	 * unhashed bdev with newly created gendisk could lead to two bdevs
1124 	 * (and thus two independent caches) being associated with one device
1125 	 * which is bad.
1126 	 */
1127 	if (inode_unhashed(bdev->bd_inode)) {
1128 		put_disk_and_module(disk);
1129 		return NULL;
1130 	}
1131 	return disk;
1132 }
1133 
bd_clear_claiming(struct block_device * whole,void * holder)1134 static void bd_clear_claiming(struct block_device *whole, void *holder)
1135 {
1136 	lockdep_assert_held(&bdev_lock);
1137 	/* tell others that we're done */
1138 	BUG_ON(whole->bd_claiming != holder);
1139 	whole->bd_claiming = NULL;
1140 	wake_up_bit(&whole->bd_claiming, 0);
1141 }
1142 
1143 /**
1144  * bd_finish_claiming - finish claiming of a block device
1145  * @bdev: block device of interest
1146  * @whole: whole block device
1147  * @holder: holder that has claimed @bdev
1148  *
1149  * Finish exclusive open of a block device. Mark the device as exlusively
1150  * open by the holder and wake up all waiters for exclusive open to finish.
1151  */
bd_finish_claiming(struct block_device * bdev,struct block_device * whole,void * holder)1152 static void bd_finish_claiming(struct block_device *bdev,
1153 		struct block_device *whole, void *holder)
1154 {
1155 	spin_lock(&bdev_lock);
1156 	BUG_ON(!bd_may_claim(bdev, whole, holder));
1157 	/*
1158 	 * Note that for a whole device bd_holders will be incremented twice,
1159 	 * and bd_holder will be set to bd_may_claim before being set to holder
1160 	 */
1161 	whole->bd_holders++;
1162 	whole->bd_holder = bd_may_claim;
1163 	bdev->bd_holders++;
1164 	bdev->bd_holder = holder;
1165 	bd_clear_claiming(whole, holder);
1166 	spin_unlock(&bdev_lock);
1167 }
1168 
1169 /**
1170  * bd_abort_claiming - abort claiming of a block device
1171  * @bdev: block device of interest
1172  * @whole: whole block device
1173  * @holder: holder that has claimed @bdev
1174  *
1175  * Abort claiming of a block device when the exclusive open failed. This can be
1176  * also used when exclusive open is not actually desired and we just needed
1177  * to block other exclusive openers for a while.
1178  */
bd_abort_claiming(struct block_device * bdev,struct block_device * whole,void * holder)1179 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1180 		       void *holder)
1181 {
1182 	spin_lock(&bdev_lock);
1183 	bd_clear_claiming(whole, holder);
1184 	spin_unlock(&bdev_lock);
1185 }
1186 EXPORT_SYMBOL(bd_abort_claiming);
1187 
1188 #ifdef CONFIG_SYSFS
1189 struct bd_holder_disk {
1190 	struct list_head	list;
1191 	struct gendisk		*disk;
1192 	int			refcnt;
1193 };
1194 
bd_find_holder_disk(struct block_device * bdev,struct gendisk * disk)1195 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1196 						  struct gendisk *disk)
1197 {
1198 	struct bd_holder_disk *holder;
1199 
1200 	list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1201 		if (holder->disk == disk)
1202 			return holder;
1203 	return NULL;
1204 }
1205 
add_symlink(struct kobject * from,struct kobject * to)1206 static int add_symlink(struct kobject *from, struct kobject *to)
1207 {
1208 	return sysfs_create_link(from, to, kobject_name(to));
1209 }
1210 
del_symlink(struct kobject * from,struct kobject * to)1211 static void del_symlink(struct kobject *from, struct kobject *to)
1212 {
1213 	sysfs_remove_link(from, kobject_name(to));
1214 }
1215 
1216 /**
1217  * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1218  * @bdev: the claimed slave bdev
1219  * @disk: the holding disk
1220  *
1221  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1222  *
1223  * This functions creates the following sysfs symlinks.
1224  *
1225  * - from "slaves" directory of the holder @disk to the claimed @bdev
1226  * - from "holders" directory of the @bdev to the holder @disk
1227  *
1228  * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1229  * passed to bd_link_disk_holder(), then:
1230  *
1231  *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1232  *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1233  *
1234  * The caller must have claimed @bdev before calling this function and
1235  * ensure that both @bdev and @disk are valid during the creation and
1236  * lifetime of these symlinks.
1237  *
1238  * CONTEXT:
1239  * Might sleep.
1240  *
1241  * RETURNS:
1242  * 0 on success, -errno on failure.
1243  */
bd_link_disk_holder(struct block_device * bdev,struct gendisk * disk)1244 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1245 {
1246 	struct bd_holder_disk *holder;
1247 	int ret = 0;
1248 
1249 	mutex_lock(&bdev->bd_mutex);
1250 
1251 	WARN_ON_ONCE(!bdev->bd_holder);
1252 
1253 	/* FIXME: remove the following once add_disk() handles errors */
1254 	if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1255 		goto out_unlock;
1256 
1257 	holder = bd_find_holder_disk(bdev, disk);
1258 	if (holder) {
1259 		holder->refcnt++;
1260 		goto out_unlock;
1261 	}
1262 
1263 	holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1264 	if (!holder) {
1265 		ret = -ENOMEM;
1266 		goto out_unlock;
1267 	}
1268 
1269 	INIT_LIST_HEAD(&holder->list);
1270 	holder->disk = disk;
1271 	holder->refcnt = 1;
1272 
1273 	ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1274 	if (ret)
1275 		goto out_free;
1276 
1277 	ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1278 	if (ret)
1279 		goto out_del;
1280 	/*
1281 	 * bdev could be deleted beneath us which would implicitly destroy
1282 	 * the holder directory.  Hold on to it.
1283 	 */
1284 	kobject_get(bdev->bd_part->holder_dir);
1285 
1286 	list_add(&holder->list, &bdev->bd_holder_disks);
1287 	goto out_unlock;
1288 
1289 out_del:
1290 	del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1291 out_free:
1292 	kfree(holder);
1293 out_unlock:
1294 	mutex_unlock(&bdev->bd_mutex);
1295 	return ret;
1296 }
1297 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1298 
1299 /**
1300  * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1301  * @bdev: the calimed slave bdev
1302  * @disk: the holding disk
1303  *
1304  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1305  *
1306  * CONTEXT:
1307  * Might sleep.
1308  */
bd_unlink_disk_holder(struct block_device * bdev,struct gendisk * disk)1309 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1310 {
1311 	struct bd_holder_disk *holder;
1312 
1313 	mutex_lock(&bdev->bd_mutex);
1314 
1315 	holder = bd_find_holder_disk(bdev, disk);
1316 
1317 	if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1318 		del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1319 		del_symlink(bdev->bd_part->holder_dir,
1320 			    &disk_to_dev(disk)->kobj);
1321 		kobject_put(bdev->bd_part->holder_dir);
1322 		list_del_init(&holder->list);
1323 		kfree(holder);
1324 	}
1325 
1326 	mutex_unlock(&bdev->bd_mutex);
1327 }
1328 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1329 #endif
1330 
1331 /**
1332  * check_disk_size_change - checks for disk size change and adjusts bdev size.
1333  * @disk: struct gendisk to check
1334  * @bdev: struct bdev to adjust.
1335  * @verbose: if %true log a message about a size change if there is any
1336  *
1337  * This routine checks to see if the bdev size does not match the disk size
1338  * and adjusts it if it differs. When shrinking the bdev size, its all caches
1339  * are freed.
1340  */
check_disk_size_change(struct gendisk * disk,struct block_device * bdev,bool verbose)1341 static void check_disk_size_change(struct gendisk *disk,
1342 		struct block_device *bdev, bool verbose)
1343 {
1344 	loff_t disk_size, bdev_size;
1345 
1346 	spin_lock(&bdev->bd_size_lock);
1347 	disk_size = (loff_t)get_capacity(disk) << 9;
1348 	bdev_size = i_size_read(bdev->bd_inode);
1349 	if (disk_size != bdev_size) {
1350 		if (verbose) {
1351 			printk(KERN_INFO
1352 			       "%s: detected capacity change from %lld to %lld\n",
1353 			       disk->disk_name, bdev_size, disk_size);
1354 		}
1355 		i_size_write(bdev->bd_inode, disk_size);
1356 	}
1357 	spin_unlock(&bdev->bd_size_lock);
1358 
1359 	if (bdev_size > disk_size) {
1360 		if (__invalidate_device(bdev, false))
1361 			pr_warn("VFS: busy inodes on resized disk %s\n",
1362 				disk->disk_name);
1363 	}
1364 }
1365 
1366 /**
1367  * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1368  * @disk: struct gendisk to check
1369  * @verbose: if %true log a message about a size change if there is any
1370  *
1371  * This routine checks to see if the bdev size does not match the disk size
1372  * and adjusts it if it differs. When shrinking the bdev size, its all caches
1373  * are freed.
1374  */
revalidate_disk_size(struct gendisk * disk,bool verbose)1375 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1376 {
1377 	struct block_device *bdev;
1378 
1379 	/*
1380 	 * Hidden disks don't have associated bdev so there's no point in
1381 	 * revalidating them.
1382 	 */
1383 	if (disk->flags & GENHD_FL_HIDDEN)
1384 		return;
1385 
1386 	bdev = bdget_disk(disk, 0);
1387 	if (bdev) {
1388 		check_disk_size_change(disk, bdev, verbose);
1389 		bdput(bdev);
1390 	}
1391 }
1392 EXPORT_SYMBOL(revalidate_disk_size);
1393 
bd_set_nr_sectors(struct block_device * bdev,sector_t sectors)1394 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1395 {
1396 	spin_lock(&bdev->bd_size_lock);
1397 	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1398 	spin_unlock(&bdev->bd_size_lock);
1399 }
1400 EXPORT_SYMBOL(bd_set_nr_sectors);
1401 
1402 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1403 
bdev_disk_changed(struct block_device * bdev,bool invalidate)1404 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1405 {
1406 	struct gendisk *disk = bdev->bd_disk;
1407 	int ret;
1408 
1409 	lockdep_assert_held(&bdev->bd_mutex);
1410 
1411 	if (!(disk->flags & GENHD_FL_UP))
1412 		return -ENXIO;
1413 
1414 rescan:
1415 	ret = blk_drop_partitions(bdev);
1416 	if (ret)
1417 		return ret;
1418 
1419 	clear_bit(GD_NEED_PART_SCAN, &disk->state);
1420 
1421 	/*
1422 	 * Historically we only set the capacity to zero for devices that
1423 	 * support partitions (independ of actually having partitions created).
1424 	 * Doing that is rather inconsistent, but changing it broke legacy
1425 	 * udisks polling for legacy ide-cdrom devices.  Use the crude check
1426 	 * below to get the sane behavior for most device while not breaking
1427 	 * userspace for this particular setup.
1428 	 */
1429 	if (invalidate) {
1430 		if (disk_part_scan_enabled(disk) ||
1431 		    !(disk->flags & GENHD_FL_REMOVABLE))
1432 			set_capacity(disk, 0);
1433 	} else {
1434 		if (disk->fops->revalidate_disk)
1435 			disk->fops->revalidate_disk(disk);
1436 	}
1437 
1438 	check_disk_size_change(disk, bdev, !invalidate);
1439 
1440 	if (get_capacity(disk)) {
1441 		ret = blk_add_partitions(disk, bdev);
1442 		if (ret == -EAGAIN)
1443 			goto rescan;
1444 	} else if (invalidate) {
1445 		/*
1446 		 * Tell userspace that the media / partition table may have
1447 		 * changed.
1448 		 */
1449 		kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1450 	}
1451 
1452 	return ret;
1453 }
1454 /*
1455  * Only exported for for loop and dasd for historic reasons.  Don't use in new
1456  * code!
1457  */
1458 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1459 
1460 /*
1461  * bd_mutex locking:
1462  *
1463  *  mutex_lock(part->bd_mutex)
1464  *    mutex_lock_nested(whole->bd_mutex, 1)
1465  */
1466 
__blkdev_get(struct block_device * bdev,fmode_t mode,void * holder,int for_part)1467 static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1468 		int for_part)
1469 {
1470 	struct block_device *whole = NULL, *claiming = NULL;
1471 	struct gendisk *disk;
1472 	int ret;
1473 	int partno;
1474 	bool first_open = false, unblock_events = true, need_restart;
1475 
1476  restart:
1477 	need_restart = false;
1478 	ret = -ENXIO;
1479 	disk = bdev_get_gendisk(bdev, &partno);
1480 	if (!disk)
1481 		goto out;
1482 
1483 	if (partno) {
1484 		whole = bdget_disk(disk, 0);
1485 		if (!whole) {
1486 			ret = -ENOMEM;
1487 			goto out_put_disk;
1488 		}
1489 	}
1490 
1491 	if (!for_part && (mode & FMODE_EXCL)) {
1492 		WARN_ON_ONCE(!holder);
1493 		if (whole)
1494 			claiming = whole;
1495 		else
1496 			claiming = bdev;
1497 		ret = bd_prepare_to_claim(bdev, claiming, holder);
1498 		if (ret)
1499 			goto out_put_whole;
1500 	}
1501 
1502 	disk_block_events(disk);
1503 	mutex_lock_nested(&bdev->bd_mutex, for_part);
1504 	if (!bdev->bd_openers) {
1505 		first_open = true;
1506 		bdev->bd_disk = disk;
1507 		bdev->bd_contains = bdev;
1508 		bdev->bd_partno = partno;
1509 
1510 		if (!partno) {
1511 			ret = -ENXIO;
1512 			bdev->bd_part = disk_get_part(disk, partno);
1513 			if (!bdev->bd_part)
1514 				goto out_clear;
1515 
1516 			ret = 0;
1517 			if (disk->fops->open) {
1518 				ret = disk->fops->open(bdev, mode);
1519 				/*
1520 				 * If we lost a race with 'disk' being deleted,
1521 				 * try again.  See md.c
1522 				 */
1523 				if (ret == -ERESTARTSYS)
1524 					need_restart = true;
1525 			}
1526 
1527 			if (!ret) {
1528 				bd_set_nr_sectors(bdev, get_capacity(disk));
1529 				set_init_blocksize(bdev);
1530 			}
1531 
1532 			/*
1533 			 * If the device is invalidated, rescan partition
1534 			 * if open succeeded or failed with -ENOMEDIUM.
1535 			 * The latter is necessary to prevent ghost
1536 			 * partitions on a removed medium.
1537 			 */
1538 			if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1539 			    (!ret || ret == -ENOMEDIUM))
1540 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1541 
1542 			if (ret)
1543 				goto out_clear;
1544 		} else {
1545 			BUG_ON(for_part);
1546 			ret = __blkdev_get(whole, mode, NULL, 1);
1547 			if (ret)
1548 				goto out_clear;
1549 			bdev->bd_contains = bdgrab(whole);
1550 			bdev->bd_part = disk_get_part(disk, partno);
1551 			if (!(disk->flags & GENHD_FL_UP) ||
1552 			    !bdev->bd_part || !bdev->bd_part->nr_sects) {
1553 				ret = -ENXIO;
1554 				goto out_clear;
1555 			}
1556 			bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1557 			set_init_blocksize(bdev);
1558 		}
1559 
1560 		if (bdev->bd_bdi == &noop_backing_dev_info)
1561 			bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1562 	} else {
1563 		if (bdev->bd_contains == bdev) {
1564 			ret = 0;
1565 			if (bdev->bd_disk->fops->open)
1566 				ret = bdev->bd_disk->fops->open(bdev, mode);
1567 			/* the same as first opener case, read comment there */
1568 			if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1569 			    (!ret || ret == -ENOMEDIUM))
1570 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1571 			if (ret)
1572 				goto out_unlock_bdev;
1573 		}
1574 	}
1575 	bdev->bd_openers++;
1576 	if (for_part)
1577 		bdev->bd_part_count++;
1578 	if (claiming)
1579 		bd_finish_claiming(bdev, claiming, holder);
1580 
1581 	/*
1582 	 * Block event polling for write claims if requested.  Any write holder
1583 	 * makes the write_holder state stick until all are released.  This is
1584 	 * good enough and tracking individual writeable reference is too
1585 	 * fragile given the way @mode is used in blkdev_get/put().
1586 	 */
1587 	if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1588 	    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1589 		bdev->bd_write_holder = true;
1590 		unblock_events = false;
1591 	}
1592 	mutex_unlock(&bdev->bd_mutex);
1593 
1594 	if (unblock_events)
1595 		disk_unblock_events(disk);
1596 
1597 	/* only one opener holds refs to the module and disk */
1598 	if (!first_open)
1599 		put_disk_and_module(disk);
1600 	if (whole)
1601 		bdput(whole);
1602 	return 0;
1603 
1604  out_clear:
1605 	disk_put_part(bdev->bd_part);
1606 	bdev->bd_disk = NULL;
1607 	bdev->bd_part = NULL;
1608 	if (bdev != bdev->bd_contains)
1609 		__blkdev_put(bdev->bd_contains, mode, 1);
1610 	bdev->bd_contains = NULL;
1611  out_unlock_bdev:
1612 	if (claiming)
1613 		bd_abort_claiming(bdev, claiming, holder);
1614 	mutex_unlock(&bdev->bd_mutex);
1615 	disk_unblock_events(disk);
1616  out_put_whole:
1617  	if (whole)
1618 		bdput(whole);
1619  out_put_disk:
1620 	put_disk_and_module(disk);
1621 	if (need_restart)
1622 		goto restart;
1623  out:
1624 	return ret;
1625 }
1626 
1627 /**
1628  * blkdev_get - open a block device
1629  * @bdev: block_device to open
1630  * @mode: FMODE_* mask
1631  * @holder: exclusive holder identifier
1632  *
1633  * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1634  * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1635  * @holder is invalid.  Exclusive opens may nest for the same @holder.
1636  *
1637  * On success, the reference count of @bdev is unchanged.  On failure,
1638  * @bdev is put.
1639  *
1640  * CONTEXT:
1641  * Might sleep.
1642  *
1643  * RETURNS:
1644  * 0 on success, -errno on failure.
1645  */
blkdev_get(struct block_device * bdev,fmode_t mode,void * holder)1646 static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1647 {
1648 	int ret, perm = 0;
1649 
1650 	if (mode & FMODE_READ)
1651 		perm |= MAY_READ;
1652 	if (mode & FMODE_WRITE)
1653 		perm |= MAY_WRITE;
1654 	ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1655 	if (ret)
1656 		goto bdput;
1657 
1658 	ret =__blkdev_get(bdev, mode, holder, 0);
1659 	if (ret)
1660 		goto bdput;
1661 	return 0;
1662 
1663 bdput:
1664 	bdput(bdev);
1665 	return ret;
1666 }
1667 
1668 /**
1669  * blkdev_get_by_path - open a block device by name
1670  * @path: path to the block device to open
1671  * @mode: FMODE_* mask
1672  * @holder: exclusive holder identifier
1673  *
1674  * Open the blockdevice described by the device file at @path.  @mode
1675  * and @holder are identical to blkdev_get().
1676  *
1677  * On success, the returned block_device has reference count of one.
1678  *
1679  * CONTEXT:
1680  * Might sleep.
1681  *
1682  * RETURNS:
1683  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1684  */
blkdev_get_by_path(const char * path,fmode_t mode,void * holder)1685 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1686 					void *holder)
1687 {
1688 	struct block_device *bdev;
1689 	int err;
1690 
1691 	bdev = lookup_bdev(path);
1692 	if (IS_ERR(bdev))
1693 		return bdev;
1694 
1695 	err = blkdev_get(bdev, mode, holder);
1696 	if (err)
1697 		return ERR_PTR(err);
1698 
1699 	if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1700 		blkdev_put(bdev, mode);
1701 		return ERR_PTR(-EACCES);
1702 	}
1703 
1704 	return bdev;
1705 }
1706 EXPORT_SYMBOL(blkdev_get_by_path);
1707 
1708 /**
1709  * blkdev_get_by_dev - open a block device by device number
1710  * @dev: device number of block device to open
1711  * @mode: FMODE_* mask
1712  * @holder: exclusive holder identifier
1713  *
1714  * Open the blockdevice described by device number @dev.  @mode and
1715  * @holder are identical to blkdev_get().
1716  *
1717  * Use it ONLY if you really do not have anything better - i.e. when
1718  * you are behind a truly sucky interface and all you are given is a
1719  * device number.  _Never_ to be used for internal purposes.  If you
1720  * ever need it - reconsider your API.
1721  *
1722  * On success, the returned block_device has reference count of one.
1723  *
1724  * CONTEXT:
1725  * Might sleep.
1726  *
1727  * RETURNS:
1728  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1729  */
blkdev_get_by_dev(dev_t dev,fmode_t mode,void * holder)1730 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1731 {
1732 	struct block_device *bdev;
1733 	int err;
1734 
1735 	bdev = bdget(dev);
1736 	if (!bdev)
1737 		return ERR_PTR(-ENOMEM);
1738 
1739 	err = blkdev_get(bdev, mode, holder);
1740 	if (err)
1741 		return ERR_PTR(err);
1742 
1743 	return bdev;
1744 }
1745 EXPORT_SYMBOL(blkdev_get_by_dev);
1746 
blkdev_open(struct inode * inode,struct file * filp)1747 static int blkdev_open(struct inode * inode, struct file * filp)
1748 {
1749 	struct block_device *bdev;
1750 
1751 	/*
1752 	 * Preserve backwards compatibility and allow large file access
1753 	 * even if userspace doesn't ask for it explicitly. Some mkfs
1754 	 * binary needs it. We might want to drop this workaround
1755 	 * during an unstable branch.
1756 	 */
1757 	filp->f_flags |= O_LARGEFILE;
1758 
1759 	filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1760 
1761 	if (filp->f_flags & O_NDELAY)
1762 		filp->f_mode |= FMODE_NDELAY;
1763 	if (filp->f_flags & O_EXCL)
1764 		filp->f_mode |= FMODE_EXCL;
1765 	if ((filp->f_flags & O_ACCMODE) == 3)
1766 		filp->f_mode |= FMODE_WRITE_IOCTL;
1767 
1768 	bdev = bd_acquire(inode);
1769 	if (bdev == NULL)
1770 		return -ENOMEM;
1771 
1772 	filp->f_mapping = bdev->bd_inode->i_mapping;
1773 	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1774 
1775 	return blkdev_get(bdev, filp->f_mode, filp);
1776 }
1777 
__blkdev_put(struct block_device * bdev,fmode_t mode,int for_part)1778 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1779 {
1780 	struct gendisk *disk = bdev->bd_disk;
1781 	struct block_device *victim = NULL;
1782 
1783 	/*
1784 	 * Sync early if it looks like we're the last one.  If someone else
1785 	 * opens the block device between now and the decrement of bd_openers
1786 	 * then we did a sync that we didn't need to, but that's not the end
1787 	 * of the world and we want to avoid long (could be several minute)
1788 	 * syncs while holding the mutex.
1789 	 */
1790 	if (bdev->bd_openers == 1)
1791 		sync_blockdev(bdev);
1792 
1793 	mutex_lock_nested(&bdev->bd_mutex, for_part);
1794 	if (for_part)
1795 		bdev->bd_part_count--;
1796 
1797 	if (!--bdev->bd_openers) {
1798 		WARN_ON_ONCE(bdev->bd_holders);
1799 		sync_blockdev(bdev);
1800 		kill_bdev(bdev);
1801 
1802 		bdev_write_inode(bdev);
1803 	}
1804 	if (bdev->bd_contains == bdev) {
1805 		if (disk->fops->release)
1806 			disk->fops->release(disk, mode);
1807 	}
1808 	if (!bdev->bd_openers) {
1809 		disk_put_part(bdev->bd_part);
1810 		bdev->bd_part = NULL;
1811 		bdev->bd_disk = NULL;
1812 		if (bdev != bdev->bd_contains)
1813 			victim = bdev->bd_contains;
1814 		bdev->bd_contains = NULL;
1815 
1816 		put_disk_and_module(disk);
1817 	}
1818 	mutex_unlock(&bdev->bd_mutex);
1819 	bdput(bdev);
1820 	if (victim)
1821 		__blkdev_put(victim, mode, 1);
1822 }
1823 
blkdev_put(struct block_device * bdev,fmode_t mode)1824 void blkdev_put(struct block_device *bdev, fmode_t mode)
1825 {
1826 	mutex_lock(&bdev->bd_mutex);
1827 
1828 	if (mode & FMODE_EXCL) {
1829 		bool bdev_free;
1830 
1831 		/*
1832 		 * Release a claim on the device.  The holder fields
1833 		 * are protected with bdev_lock.  bd_mutex is to
1834 		 * synchronize disk_holder unlinking.
1835 		 */
1836 		spin_lock(&bdev_lock);
1837 
1838 		WARN_ON_ONCE(--bdev->bd_holders < 0);
1839 		WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1840 
1841 		/* bd_contains might point to self, check in a separate step */
1842 		if ((bdev_free = !bdev->bd_holders))
1843 			bdev->bd_holder = NULL;
1844 		if (!bdev->bd_contains->bd_holders)
1845 			bdev->bd_contains->bd_holder = NULL;
1846 
1847 		spin_unlock(&bdev_lock);
1848 
1849 		/*
1850 		 * If this was the last claim, remove holder link and
1851 		 * unblock evpoll if it was a write holder.
1852 		 */
1853 		if (bdev_free && bdev->bd_write_holder) {
1854 			disk_unblock_events(bdev->bd_disk);
1855 			bdev->bd_write_holder = false;
1856 		}
1857 	}
1858 
1859 	/*
1860 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1861 	 * event.  This is to ensure detection of media removal commanded
1862 	 * from userland - e.g. eject(1).
1863 	 */
1864 	disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1865 
1866 	mutex_unlock(&bdev->bd_mutex);
1867 
1868 	__blkdev_put(bdev, mode, 0);
1869 }
1870 EXPORT_SYMBOL(blkdev_put);
1871 
blkdev_close(struct inode * inode,struct file * filp)1872 static int blkdev_close(struct inode * inode, struct file * filp)
1873 {
1874 	struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1875 	blkdev_put(bdev, filp->f_mode);
1876 	return 0;
1877 }
1878 
block_ioctl(struct file * file,unsigned cmd,unsigned long arg)1879 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1880 {
1881 	struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1882 	fmode_t mode = file->f_mode;
1883 
1884 	/*
1885 	 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1886 	 * to updated it before every ioctl.
1887 	 */
1888 	if (file->f_flags & O_NDELAY)
1889 		mode |= FMODE_NDELAY;
1890 	else
1891 		mode &= ~FMODE_NDELAY;
1892 
1893 	return blkdev_ioctl(bdev, mode, cmd, arg);
1894 }
1895 
1896 /*
1897  * Write data to the block device.  Only intended for the block device itself
1898  * and the raw driver which basically is a fake block device.
1899  *
1900  * Does not take i_mutex for the write and thus is not for general purpose
1901  * use.
1902  */
blkdev_write_iter(struct kiocb * iocb,struct iov_iter * from)1903 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1904 {
1905 	struct file *file = iocb->ki_filp;
1906 	struct inode *bd_inode = bdev_file_inode(file);
1907 	loff_t size = i_size_read(bd_inode);
1908 	struct blk_plug plug;
1909 	size_t shorted = 0;
1910 	ssize_t ret;
1911 
1912 	if (bdev_read_only(I_BDEV(bd_inode)))
1913 		return -EPERM;
1914 
1915 	if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1916 		return -ETXTBSY;
1917 
1918 	if (!iov_iter_count(from))
1919 		return 0;
1920 
1921 	if (iocb->ki_pos >= size)
1922 		return -ENOSPC;
1923 
1924 	if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1925 		return -EOPNOTSUPP;
1926 
1927 	size -= iocb->ki_pos;
1928 	if (iov_iter_count(from) > size) {
1929 		shorted = iov_iter_count(from) - size;
1930 		iov_iter_truncate(from, size);
1931 	}
1932 
1933 	blk_start_plug(&plug);
1934 	ret = __generic_file_write_iter(iocb, from);
1935 	if (ret > 0)
1936 		ret = generic_write_sync(iocb, ret);
1937 	iov_iter_reexpand(from, iov_iter_count(from) + shorted);
1938 	blk_finish_plug(&plug);
1939 	return ret;
1940 }
1941 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1942 
blkdev_read_iter(struct kiocb * iocb,struct iov_iter * to)1943 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1944 {
1945 	struct file *file = iocb->ki_filp;
1946 	struct inode *bd_inode = bdev_file_inode(file);
1947 	loff_t size = i_size_read(bd_inode);
1948 	loff_t pos = iocb->ki_pos;
1949 	size_t shorted = 0;
1950 	ssize_t ret;
1951 
1952 	if (pos >= size)
1953 		return 0;
1954 
1955 	size -= pos;
1956 	if (iov_iter_count(to) > size) {
1957 		shorted = iov_iter_count(to) - size;
1958 		iov_iter_truncate(to, size);
1959 	}
1960 
1961 	ret = generic_file_read_iter(iocb, to);
1962 	iov_iter_reexpand(to, iov_iter_count(to) + shorted);
1963 	return ret;
1964 }
1965 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1966 
blkdev_writepages(struct address_space * mapping,struct writeback_control * wbc)1967 static int blkdev_writepages(struct address_space *mapping,
1968 			     struct writeback_control *wbc)
1969 {
1970 	return generic_writepages(mapping, wbc);
1971 }
1972 
1973 static const struct address_space_operations def_blk_aops = {
1974 	.readpage	= blkdev_readpage,
1975 	.readahead	= blkdev_readahead,
1976 	.writepage	= blkdev_writepage,
1977 	.write_begin	= blkdev_write_begin,
1978 	.write_end	= blkdev_write_end,
1979 	.writepages	= blkdev_writepages,
1980 	.direct_IO	= blkdev_direct_IO,
1981 	.migratepage	= buffer_migrate_page_norefs,
1982 	.is_dirty_writeback = buffer_check_dirty_writeback,
1983 };
1984 
1985 #define	BLKDEV_FALLOC_FL_SUPPORTED					\
1986 		(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |		\
1987 		 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1988 
blkdev_fallocate(struct file * file,int mode,loff_t start,loff_t len)1989 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1990 			     loff_t len)
1991 {
1992 	struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1993 	loff_t end = start + len - 1;
1994 	loff_t isize;
1995 	int error;
1996 
1997 	/* Fail if we don't recognize the flags. */
1998 	if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1999 		return -EOPNOTSUPP;
2000 
2001 	/* Don't go off the end of the device. */
2002 	isize = i_size_read(bdev->bd_inode);
2003 	if (start >= isize)
2004 		return -EINVAL;
2005 	if (end >= isize) {
2006 		if (mode & FALLOC_FL_KEEP_SIZE) {
2007 			len = isize - start;
2008 			end = start + len - 1;
2009 		} else
2010 			return -EINVAL;
2011 	}
2012 
2013 	/*
2014 	 * Don't allow IO that isn't aligned to logical block size.
2015 	 */
2016 	if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2017 		return -EINVAL;
2018 
2019 	/* Invalidate the page cache, including dirty pages. */
2020 	error = truncate_bdev_range(bdev, file->f_mode, start, end);
2021 	if (error)
2022 		return error;
2023 
2024 	switch (mode) {
2025 	case FALLOC_FL_ZERO_RANGE:
2026 	case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2027 		error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2028 					    GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2029 		break;
2030 	case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2031 		error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2032 					     GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2033 		break;
2034 	case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2035 		error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2036 					     GFP_KERNEL, 0);
2037 		break;
2038 	default:
2039 		return -EOPNOTSUPP;
2040 	}
2041 	if (error)
2042 		return error;
2043 
2044 	/*
2045 	 * Invalidate the page cache again; if someone wandered in and dirtied
2046 	 * a page, we just discard it - userspace has no way of knowing whether
2047 	 * the write happened before or after discard completing...
2048 	 */
2049 	return truncate_bdev_range(bdev, file->f_mode, start, end);
2050 }
2051 
2052 const struct file_operations def_blk_fops = {
2053 	.open		= blkdev_open,
2054 	.release	= blkdev_close,
2055 	.llseek		= block_llseek,
2056 	.read_iter	= blkdev_read_iter,
2057 	.write_iter	= blkdev_write_iter,
2058 	.iopoll		= blkdev_iopoll,
2059 	.mmap		= generic_file_mmap,
2060 	.fsync		= blkdev_fsync,
2061 	.unlocked_ioctl	= block_ioctl,
2062 #ifdef CONFIG_COMPAT
2063 	.compat_ioctl	= compat_blkdev_ioctl,
2064 #endif
2065 	.splice_read	= generic_file_splice_read,
2066 	.splice_write	= iter_file_splice_write,
2067 	.fallocate	= blkdev_fallocate,
2068 };
2069 
2070 /**
2071  * lookup_bdev  - lookup a struct block_device by name
2072  * @pathname:	special file representing the block device
2073  *
2074  * Get a reference to the blockdevice at @pathname in the current
2075  * namespace if possible and return it.  Return ERR_PTR(error)
2076  * otherwise.
2077  */
lookup_bdev(const char * pathname)2078 struct block_device *lookup_bdev(const char *pathname)
2079 {
2080 	struct block_device *bdev;
2081 	struct inode *inode;
2082 	struct path path;
2083 	int error;
2084 
2085 	if (!pathname || !*pathname)
2086 		return ERR_PTR(-EINVAL);
2087 
2088 	error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2089 	if (error)
2090 		return ERR_PTR(error);
2091 
2092 	inode = d_backing_inode(path.dentry);
2093 	error = -ENOTBLK;
2094 	if (!S_ISBLK(inode->i_mode))
2095 		goto fail;
2096 	error = -EACCES;
2097 	if (!may_open_dev(&path))
2098 		goto fail;
2099 	error = -ENOMEM;
2100 	bdev = bd_acquire(inode);
2101 	if (!bdev)
2102 		goto fail;
2103 out:
2104 	path_put(&path);
2105 	return bdev;
2106 fail:
2107 	bdev = ERR_PTR(error);
2108 	goto out;
2109 }
2110 EXPORT_SYMBOL(lookup_bdev);
2111 
__invalidate_device(struct block_device * bdev,bool kill_dirty)2112 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2113 {
2114 	struct super_block *sb = get_super(bdev);
2115 	int res = 0;
2116 
2117 	if (sb) {
2118 		/*
2119 		 * no need to lock the super, get_super holds the
2120 		 * read mutex so the filesystem cannot go away
2121 		 * under us (->put_super runs with the write lock
2122 		 * hold).
2123 		 */
2124 		shrink_dcache_sb(sb);
2125 		res = invalidate_inodes(sb, kill_dirty);
2126 		drop_super(sb);
2127 	}
2128 	invalidate_bdev(bdev);
2129 	return res;
2130 }
2131 EXPORT_SYMBOL(__invalidate_device);
2132 
iterate_bdevs(void (* func)(struct block_device *,void *),void * arg)2133 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2134 {
2135 	struct inode *inode, *old_inode = NULL;
2136 
2137 	spin_lock(&blockdev_superblock->s_inode_list_lock);
2138 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2139 		struct address_space *mapping = inode->i_mapping;
2140 		struct block_device *bdev;
2141 
2142 		spin_lock(&inode->i_lock);
2143 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2144 		    mapping->nrpages == 0) {
2145 			spin_unlock(&inode->i_lock);
2146 			continue;
2147 		}
2148 		__iget(inode);
2149 		spin_unlock(&inode->i_lock);
2150 		spin_unlock(&blockdev_superblock->s_inode_list_lock);
2151 		/*
2152 		 * We hold a reference to 'inode' so it couldn't have been
2153 		 * removed from s_inodes list while we dropped the
2154 		 * s_inode_list_lock  We cannot iput the inode now as we can
2155 		 * be holding the last reference and we cannot iput it under
2156 		 * s_inode_list_lock. So we keep the reference and iput it
2157 		 * later.
2158 		 */
2159 		iput(old_inode);
2160 		old_inode = inode;
2161 		bdev = I_BDEV(inode);
2162 
2163 		mutex_lock(&bdev->bd_mutex);
2164 		if (bdev->bd_openers)
2165 			func(bdev, arg);
2166 		mutex_unlock(&bdev->bd_mutex);
2167 
2168 		spin_lock(&blockdev_superblock->s_inode_list_lock);
2169 	}
2170 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
2171 	iput(old_inode);
2172 }
2173