xref: /linux/block/bdev.c (revision ea8a163e02d6925773129e2dd86e419e491b791d)
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 
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/slab.h>
11 #include <linux/kmod.h>
12 #include <linux/major.h>
13 #include <linux/device_cgroup.h>
14 #include <linux/blkdev.h>
15 #include <linux/blk-integrity.h>
16 #include <linux/backing-dev.h>
17 #include <linux/module.h>
18 #include <linux/blkpg.h>
19 #include <linux/magic.h>
20 #include <linux/buffer_head.h>
21 #include <linux/swap.h>
22 #include <linux/writeback.h>
23 #include <linux/mount.h>
24 #include <linux/pseudo_fs.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/cleancache.h>
28 #include <linux/part_stat.h>
29 #include <linux/uaccess.h>
30 #include "../fs/internal.h"
31 #include "blk.h"
32 
33 struct bdev_inode {
34 	struct block_device bdev;
35 	struct inode vfs_inode;
36 };
37 
38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
39 {
40 	return container_of(inode, struct bdev_inode, vfs_inode);
41 }
42 
43 struct block_device *I_BDEV(struct inode *inode)
44 {
45 	return &BDEV_I(inode)->bdev;
46 }
47 EXPORT_SYMBOL(I_BDEV);
48 
49 static void bdev_write_inode(struct block_device *bdev)
50 {
51 	struct inode *inode = bdev->bd_inode;
52 	int ret;
53 
54 	spin_lock(&inode->i_lock);
55 	while (inode->i_state & I_DIRTY) {
56 		spin_unlock(&inode->i_lock);
57 		ret = write_inode_now(inode, true);
58 		if (ret) {
59 			char name[BDEVNAME_SIZE];
60 			pr_warn_ratelimited("VFS: Dirty inode writeback failed "
61 					    "for block device %s (err=%d).\n",
62 					    bdevname(bdev, name), ret);
63 		}
64 		spin_lock(&inode->i_lock);
65 	}
66 	spin_unlock(&inode->i_lock);
67 }
68 
69 /* Kill _all_ buffers and pagecache , dirty or not.. */
70 static void kill_bdev(struct block_device *bdev)
71 {
72 	struct address_space *mapping = bdev->bd_inode->i_mapping;
73 
74 	if (mapping_empty(mapping))
75 		return;
76 
77 	invalidate_bh_lrus();
78 	truncate_inode_pages(mapping, 0);
79 }
80 
81 /* Invalidate clean unused buffers and pagecache. */
82 void invalidate_bdev(struct block_device *bdev)
83 {
84 	struct address_space *mapping = bdev->bd_inode->i_mapping;
85 
86 	if (mapping->nrpages) {
87 		invalidate_bh_lrus();
88 		lru_add_drain_all();	/* make sure all lru add caches are flushed */
89 		invalidate_mapping_pages(mapping, 0, -1);
90 	}
91 	/* 99% of the time, we don't need to flush the cleancache on the bdev.
92 	 * But, for the strange corners, lets be cautious
93 	 */
94 	cleancache_invalidate_inode(mapping);
95 }
96 EXPORT_SYMBOL(invalidate_bdev);
97 
98 /*
99  * Drop all buffers & page cache for given bdev range. This function bails
100  * with error if bdev has other exclusive owner (such as filesystem).
101  */
102 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
103 			loff_t lstart, loff_t lend)
104 {
105 	/*
106 	 * If we don't hold exclusive handle for the device, upgrade to it
107 	 * while we discard the buffer cache to avoid discarding buffers
108 	 * under live filesystem.
109 	 */
110 	if (!(mode & FMODE_EXCL)) {
111 		int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
112 		if (err)
113 			goto invalidate;
114 	}
115 
116 	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
117 	if (!(mode & FMODE_EXCL))
118 		bd_abort_claiming(bdev, truncate_bdev_range);
119 	return 0;
120 
121 invalidate:
122 	/*
123 	 * Someone else has handle exclusively open. Try invalidating instead.
124 	 * The 'end' argument is inclusive so the rounding is safe.
125 	 */
126 	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
127 					     lstart >> PAGE_SHIFT,
128 					     lend >> PAGE_SHIFT);
129 }
130 
131 static void set_init_blocksize(struct block_device *bdev)
132 {
133 	unsigned int bsize = bdev_logical_block_size(bdev);
134 	loff_t size = i_size_read(bdev->bd_inode);
135 
136 	while (bsize < PAGE_SIZE) {
137 		if (size & bsize)
138 			break;
139 		bsize <<= 1;
140 	}
141 	bdev->bd_inode->i_blkbits = blksize_bits(bsize);
142 }
143 
144 int set_blocksize(struct block_device *bdev, int size)
145 {
146 	/* Size must be a power of two, and between 512 and PAGE_SIZE */
147 	if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
148 		return -EINVAL;
149 
150 	/* Size cannot be smaller than the size supported by the device */
151 	if (size < bdev_logical_block_size(bdev))
152 		return -EINVAL;
153 
154 	/* Don't change the size if it is same as current */
155 	if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
156 		sync_blockdev(bdev);
157 		bdev->bd_inode->i_blkbits = blksize_bits(size);
158 		kill_bdev(bdev);
159 	}
160 	return 0;
161 }
162 
163 EXPORT_SYMBOL(set_blocksize);
164 
165 int sb_set_blocksize(struct super_block *sb, int size)
166 {
167 	if (set_blocksize(sb->s_bdev, size))
168 		return 0;
169 	/* If we get here, we know size is power of two
170 	 * and it's value is between 512 and PAGE_SIZE */
171 	sb->s_blocksize = size;
172 	sb->s_blocksize_bits = blksize_bits(size);
173 	return sb->s_blocksize;
174 }
175 
176 EXPORT_SYMBOL(sb_set_blocksize);
177 
178 int sb_min_blocksize(struct super_block *sb, int size)
179 {
180 	int minsize = bdev_logical_block_size(sb->s_bdev);
181 	if (size < minsize)
182 		size = minsize;
183 	return sb_set_blocksize(sb, size);
184 }
185 
186 EXPORT_SYMBOL(sb_min_blocksize);
187 
188 int sync_blockdev_nowait(struct block_device *bdev)
189 {
190 	if (!bdev)
191 		return 0;
192 	return filemap_flush(bdev->bd_inode->i_mapping);
193 }
194 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
195 
196 /*
197  * Write out and wait upon all the dirty data associated with a block
198  * device via its mapping.  Does not take the superblock lock.
199  */
200 int sync_blockdev(struct block_device *bdev)
201 {
202 	if (!bdev)
203 		return 0;
204 	return filemap_write_and_wait(bdev->bd_inode->i_mapping);
205 }
206 EXPORT_SYMBOL(sync_blockdev);
207 
208 /*
209  * Write out and wait upon all dirty data associated with this
210  * device.   Filesystem data as well as the underlying block
211  * device.  Takes the superblock lock.
212  */
213 int fsync_bdev(struct block_device *bdev)
214 {
215 	struct super_block *sb = get_super(bdev);
216 	if (sb) {
217 		int res = sync_filesystem(sb);
218 		drop_super(sb);
219 		return res;
220 	}
221 	return sync_blockdev(bdev);
222 }
223 EXPORT_SYMBOL(fsync_bdev);
224 
225 /**
226  * freeze_bdev  --  lock a filesystem and force it into a consistent state
227  * @bdev:	blockdevice to lock
228  *
229  * If a superblock is found on this device, we take the s_umount semaphore
230  * on it to make sure nobody unmounts until the snapshot creation is done.
231  * The reference counter (bd_fsfreeze_count) guarantees that only the last
232  * unfreeze process can unfreeze the frozen filesystem actually when multiple
233  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
234  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
235  * actually.
236  */
237 int freeze_bdev(struct block_device *bdev)
238 {
239 	struct super_block *sb;
240 	int error = 0;
241 
242 	mutex_lock(&bdev->bd_fsfreeze_mutex);
243 	if (++bdev->bd_fsfreeze_count > 1)
244 		goto done;
245 
246 	sb = get_active_super(bdev);
247 	if (!sb)
248 		goto sync;
249 	if (sb->s_op->freeze_super)
250 		error = sb->s_op->freeze_super(sb);
251 	else
252 		error = freeze_super(sb);
253 	deactivate_super(sb);
254 
255 	if (error) {
256 		bdev->bd_fsfreeze_count--;
257 		goto done;
258 	}
259 	bdev->bd_fsfreeze_sb = sb;
260 
261 sync:
262 	sync_blockdev(bdev);
263 done:
264 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
265 	return error;
266 }
267 EXPORT_SYMBOL(freeze_bdev);
268 
269 /**
270  * thaw_bdev  -- unlock filesystem
271  * @bdev:	blockdevice to unlock
272  *
273  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
274  */
275 int thaw_bdev(struct block_device *bdev)
276 {
277 	struct super_block *sb;
278 	int error = -EINVAL;
279 
280 	mutex_lock(&bdev->bd_fsfreeze_mutex);
281 	if (!bdev->bd_fsfreeze_count)
282 		goto out;
283 
284 	error = 0;
285 	if (--bdev->bd_fsfreeze_count > 0)
286 		goto out;
287 
288 	sb = bdev->bd_fsfreeze_sb;
289 	if (!sb)
290 		goto out;
291 
292 	if (sb->s_op->thaw_super)
293 		error = sb->s_op->thaw_super(sb);
294 	else
295 		error = thaw_super(sb);
296 	if (error)
297 		bdev->bd_fsfreeze_count++;
298 	else
299 		bdev->bd_fsfreeze_sb = NULL;
300 out:
301 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
302 	return error;
303 }
304 EXPORT_SYMBOL(thaw_bdev);
305 
306 /**
307  * bdev_read_page() - Start reading a page from a block device
308  * @bdev: The device to read the page from
309  * @sector: The offset on the device to read the page to (need not be aligned)
310  * @page: The page to read
311  *
312  * On entry, the page should be locked.  It will be unlocked when the page
313  * has been read.  If the block driver implements rw_page synchronously,
314  * that will be true on exit from this function, but it need not be.
315  *
316  * Errors returned by this function are usually "soft", eg out of memory, or
317  * queue full; callers should try a different route to read this page rather
318  * than propagate an error back up the stack.
319  *
320  * Return: negative errno if an error occurs, 0 if submission was successful.
321  */
322 int bdev_read_page(struct block_device *bdev, sector_t sector,
323 			struct page *page)
324 {
325 	const struct block_device_operations *ops = bdev->bd_disk->fops;
326 	int result = -EOPNOTSUPP;
327 
328 	if (!ops->rw_page || bdev_get_integrity(bdev))
329 		return result;
330 
331 	result = blk_queue_enter(bdev_get_queue(bdev), 0);
332 	if (result)
333 		return result;
334 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
335 			      REQ_OP_READ);
336 	blk_queue_exit(bdev_get_queue(bdev));
337 	return result;
338 }
339 
340 /**
341  * bdev_write_page() - Start writing a page to a block device
342  * @bdev: The device to write the page to
343  * @sector: The offset on the device to write the page to (need not be aligned)
344  * @page: The page to write
345  * @wbc: The writeback_control for the write
346  *
347  * On entry, the page should be locked and not currently under writeback.
348  * On exit, if the write started successfully, the page will be unlocked and
349  * under writeback.  If the write failed already (eg the driver failed to
350  * queue the page to the device), the page will still be locked.  If the
351  * caller is a ->writepage implementation, it will need to unlock the page.
352  *
353  * Errors returned by this function are usually "soft", eg out of memory, or
354  * queue full; callers should try a different route to write this page rather
355  * than propagate an error back up the stack.
356  *
357  * Return: negative errno if an error occurs, 0 if submission was successful.
358  */
359 int bdev_write_page(struct block_device *bdev, sector_t sector,
360 			struct page *page, struct writeback_control *wbc)
361 {
362 	int result;
363 	const struct block_device_operations *ops = bdev->bd_disk->fops;
364 
365 	if (!ops->rw_page || bdev_get_integrity(bdev))
366 		return -EOPNOTSUPP;
367 	result = blk_queue_enter(bdev_get_queue(bdev), 0);
368 	if (result)
369 		return result;
370 
371 	set_page_writeback(page);
372 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
373 			      REQ_OP_WRITE);
374 	if (result) {
375 		end_page_writeback(page);
376 	} else {
377 		clean_page_buffers(page);
378 		unlock_page(page);
379 	}
380 	blk_queue_exit(bdev_get_queue(bdev));
381 	return result;
382 }
383 
384 /*
385  * pseudo-fs
386  */
387 
388 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
389 static struct kmem_cache * bdev_cachep __read_mostly;
390 
391 static struct inode *bdev_alloc_inode(struct super_block *sb)
392 {
393 	struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
394 
395 	if (!ei)
396 		return NULL;
397 	memset(&ei->bdev, 0, sizeof(ei->bdev));
398 	return &ei->vfs_inode;
399 }
400 
401 static void bdev_free_inode(struct inode *inode)
402 {
403 	struct block_device *bdev = I_BDEV(inode);
404 
405 	free_percpu(bdev->bd_stats);
406 	kfree(bdev->bd_meta_info);
407 
408 	if (!bdev_is_partition(bdev)) {
409 		if (bdev->bd_disk && bdev->bd_disk->bdi)
410 			bdi_put(bdev->bd_disk->bdi);
411 		kfree(bdev->bd_disk);
412 	}
413 
414 	if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
415 		blk_free_ext_minor(MINOR(bdev->bd_dev));
416 
417 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
418 }
419 
420 static void init_once(void *data)
421 {
422 	struct bdev_inode *ei = data;
423 
424 	inode_init_once(&ei->vfs_inode);
425 }
426 
427 static void bdev_evict_inode(struct inode *inode)
428 {
429 	truncate_inode_pages_final(&inode->i_data);
430 	invalidate_inode_buffers(inode); /* is it needed here? */
431 	clear_inode(inode);
432 }
433 
434 static const struct super_operations bdev_sops = {
435 	.statfs = simple_statfs,
436 	.alloc_inode = bdev_alloc_inode,
437 	.free_inode = bdev_free_inode,
438 	.drop_inode = generic_delete_inode,
439 	.evict_inode = bdev_evict_inode,
440 };
441 
442 static int bd_init_fs_context(struct fs_context *fc)
443 {
444 	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
445 	if (!ctx)
446 		return -ENOMEM;
447 	fc->s_iflags |= SB_I_CGROUPWB;
448 	ctx->ops = &bdev_sops;
449 	return 0;
450 }
451 
452 static struct file_system_type bd_type = {
453 	.name		= "bdev",
454 	.init_fs_context = bd_init_fs_context,
455 	.kill_sb	= kill_anon_super,
456 };
457 
458 struct super_block *blockdev_superblock __read_mostly;
459 EXPORT_SYMBOL_GPL(blockdev_superblock);
460 
461 void __init bdev_cache_init(void)
462 {
463 	int err;
464 	static struct vfsmount *bd_mnt;
465 
466 	bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
467 			0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
468 				SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
469 			init_once);
470 	err = register_filesystem(&bd_type);
471 	if (err)
472 		panic("Cannot register bdev pseudo-fs");
473 	bd_mnt = kern_mount(&bd_type);
474 	if (IS_ERR(bd_mnt))
475 		panic("Cannot create bdev pseudo-fs");
476 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
477 }
478 
479 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
480 {
481 	struct block_device *bdev;
482 	struct inode *inode;
483 
484 	inode = new_inode(blockdev_superblock);
485 	if (!inode)
486 		return NULL;
487 	inode->i_mode = S_IFBLK;
488 	inode->i_rdev = 0;
489 	inode->i_data.a_ops = &def_blk_aops;
490 	mapping_set_gfp_mask(&inode->i_data, GFP_USER);
491 
492 	bdev = I_BDEV(inode);
493 	mutex_init(&bdev->bd_fsfreeze_mutex);
494 	spin_lock_init(&bdev->bd_size_lock);
495 	bdev->bd_partno = partno;
496 	bdev->bd_inode = inode;
497 	bdev->bd_queue = disk->queue;
498 	bdev->bd_stats = alloc_percpu(struct disk_stats);
499 	if (!bdev->bd_stats) {
500 		iput(inode);
501 		return NULL;
502 	}
503 	bdev->bd_disk = disk;
504 	return bdev;
505 }
506 
507 void bdev_add(struct block_device *bdev, dev_t dev)
508 {
509 	bdev->bd_dev = dev;
510 	bdev->bd_inode->i_rdev = dev;
511 	bdev->bd_inode->i_ino = dev;
512 	insert_inode_hash(bdev->bd_inode);
513 }
514 
515 long nr_blockdev_pages(void)
516 {
517 	struct inode *inode;
518 	long ret = 0;
519 
520 	spin_lock(&blockdev_superblock->s_inode_list_lock);
521 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
522 		ret += inode->i_mapping->nrpages;
523 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
524 
525 	return ret;
526 }
527 
528 /**
529  * bd_may_claim - test whether a block device can be claimed
530  * @bdev: block device of interest
531  * @whole: whole block device containing @bdev, may equal @bdev
532  * @holder: holder trying to claim @bdev
533  *
534  * Test whether @bdev can be claimed by @holder.
535  *
536  * CONTEXT:
537  * spin_lock(&bdev_lock).
538  *
539  * RETURNS:
540  * %true if @bdev can be claimed, %false otherwise.
541  */
542 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
543 			 void *holder)
544 {
545 	if (bdev->bd_holder == holder)
546 		return true;	 /* already a holder */
547 	else if (bdev->bd_holder != NULL)
548 		return false; 	 /* held by someone else */
549 	else if (whole == bdev)
550 		return true;  	 /* is a whole device which isn't held */
551 
552 	else if (whole->bd_holder == bd_may_claim)
553 		return true; 	 /* is a partition of a device that is being partitioned */
554 	else if (whole->bd_holder != NULL)
555 		return false;	 /* is a partition of a held device */
556 	else
557 		return true;	 /* is a partition of an un-held device */
558 }
559 
560 /**
561  * bd_prepare_to_claim - claim a block device
562  * @bdev: block device of interest
563  * @holder: holder trying to claim @bdev
564  *
565  * Claim @bdev.  This function fails if @bdev is already claimed by another
566  * holder and waits if another claiming is in progress. return, the caller
567  * has ownership of bd_claiming and bd_holder[s].
568  *
569  * RETURNS:
570  * 0 if @bdev can be claimed, -EBUSY otherwise.
571  */
572 int bd_prepare_to_claim(struct block_device *bdev, void *holder)
573 {
574 	struct block_device *whole = bdev_whole(bdev);
575 
576 	if (WARN_ON_ONCE(!holder))
577 		return -EINVAL;
578 retry:
579 	spin_lock(&bdev_lock);
580 	/* if someone else claimed, fail */
581 	if (!bd_may_claim(bdev, whole, holder)) {
582 		spin_unlock(&bdev_lock);
583 		return -EBUSY;
584 	}
585 
586 	/* if claiming is already in progress, wait for it to finish */
587 	if (whole->bd_claiming) {
588 		wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
589 		DEFINE_WAIT(wait);
590 
591 		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
592 		spin_unlock(&bdev_lock);
593 		schedule();
594 		finish_wait(wq, &wait);
595 		goto retry;
596 	}
597 
598 	/* yay, all mine */
599 	whole->bd_claiming = holder;
600 	spin_unlock(&bdev_lock);
601 	return 0;
602 }
603 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
604 
605 static void bd_clear_claiming(struct block_device *whole, void *holder)
606 {
607 	lockdep_assert_held(&bdev_lock);
608 	/* tell others that we're done */
609 	BUG_ON(whole->bd_claiming != holder);
610 	whole->bd_claiming = NULL;
611 	wake_up_bit(&whole->bd_claiming, 0);
612 }
613 
614 /**
615  * bd_finish_claiming - finish claiming of a block device
616  * @bdev: block device of interest
617  * @holder: holder that has claimed @bdev
618  *
619  * Finish exclusive open of a block device. Mark the device as exlusively
620  * open by the holder and wake up all waiters for exclusive open to finish.
621  */
622 static void bd_finish_claiming(struct block_device *bdev, void *holder)
623 {
624 	struct block_device *whole = bdev_whole(bdev);
625 
626 	spin_lock(&bdev_lock);
627 	BUG_ON(!bd_may_claim(bdev, whole, holder));
628 	/*
629 	 * Note that for a whole device bd_holders will be incremented twice,
630 	 * and bd_holder will be set to bd_may_claim before being set to holder
631 	 */
632 	whole->bd_holders++;
633 	whole->bd_holder = bd_may_claim;
634 	bdev->bd_holders++;
635 	bdev->bd_holder = holder;
636 	bd_clear_claiming(whole, holder);
637 	spin_unlock(&bdev_lock);
638 }
639 
640 /**
641  * bd_abort_claiming - abort claiming of a block device
642  * @bdev: block device of interest
643  * @holder: holder that has claimed @bdev
644  *
645  * Abort claiming of a block device when the exclusive open failed. This can be
646  * also used when exclusive open is not actually desired and we just needed
647  * to block other exclusive openers for a while.
648  */
649 void bd_abort_claiming(struct block_device *bdev, void *holder)
650 {
651 	spin_lock(&bdev_lock);
652 	bd_clear_claiming(bdev_whole(bdev), holder);
653 	spin_unlock(&bdev_lock);
654 }
655 EXPORT_SYMBOL(bd_abort_claiming);
656 
657 static void blkdev_flush_mapping(struct block_device *bdev)
658 {
659 	WARN_ON_ONCE(bdev->bd_holders);
660 	sync_blockdev(bdev);
661 	kill_bdev(bdev);
662 	bdev_write_inode(bdev);
663 }
664 
665 static int blkdev_get_whole(struct block_device *bdev, fmode_t mode)
666 {
667 	struct gendisk *disk = bdev->bd_disk;
668 	int ret = 0;
669 
670 	if (disk->fops->open) {
671 		ret = disk->fops->open(bdev, mode);
672 		if (ret) {
673 			/* avoid ghost partitions on a removed medium */
674 			if (ret == -ENOMEDIUM &&
675 			     test_bit(GD_NEED_PART_SCAN, &disk->state))
676 				bdev_disk_changed(disk, true);
677 			return ret;
678 		}
679 	}
680 
681 	if (!bdev->bd_openers)
682 		set_init_blocksize(bdev);
683 	if (test_bit(GD_NEED_PART_SCAN, &disk->state))
684 		bdev_disk_changed(disk, false);
685 	bdev->bd_openers++;
686 	return 0;;
687 }
688 
689 static void blkdev_put_whole(struct block_device *bdev, fmode_t mode)
690 {
691 	if (!--bdev->bd_openers)
692 		blkdev_flush_mapping(bdev);
693 	if (bdev->bd_disk->fops->release)
694 		bdev->bd_disk->fops->release(bdev->bd_disk, mode);
695 }
696 
697 static int blkdev_get_part(struct block_device *part, fmode_t mode)
698 {
699 	struct gendisk *disk = part->bd_disk;
700 	int ret;
701 
702 	if (part->bd_openers)
703 		goto done;
704 
705 	ret = blkdev_get_whole(bdev_whole(part), mode);
706 	if (ret)
707 		return ret;
708 
709 	ret = -ENXIO;
710 	if (!bdev_nr_sectors(part))
711 		goto out_blkdev_put;
712 
713 	disk->open_partitions++;
714 	set_init_blocksize(part);
715 done:
716 	part->bd_openers++;
717 	return 0;
718 
719 out_blkdev_put:
720 	blkdev_put_whole(bdev_whole(part), mode);
721 	return ret;
722 }
723 
724 static void blkdev_put_part(struct block_device *part, fmode_t mode)
725 {
726 	struct block_device *whole = bdev_whole(part);
727 
728 	if (--part->bd_openers)
729 		return;
730 	blkdev_flush_mapping(part);
731 	whole->bd_disk->open_partitions--;
732 	blkdev_put_whole(whole, mode);
733 }
734 
735 struct block_device *blkdev_get_no_open(dev_t dev)
736 {
737 	struct block_device *bdev;
738 	struct inode *inode;
739 
740 	inode = ilookup(blockdev_superblock, dev);
741 	if (!inode) {
742 		blk_request_module(dev);
743 		inode = ilookup(blockdev_superblock, dev);
744 		if (!inode)
745 			return NULL;
746 	}
747 
748 	/* switch from the inode reference to a device mode one: */
749 	bdev = &BDEV_I(inode)->bdev;
750 	if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
751 		bdev = NULL;
752 	iput(inode);
753 
754 	if (!bdev)
755 		return NULL;
756 	if ((bdev->bd_disk->flags & GENHD_FL_HIDDEN) ||
757 	    !try_module_get(bdev->bd_disk->fops->owner)) {
758 		put_device(&bdev->bd_device);
759 		return NULL;
760 	}
761 
762 	return bdev;
763 }
764 
765 void blkdev_put_no_open(struct block_device *bdev)
766 {
767 	module_put(bdev->bd_disk->fops->owner);
768 	put_device(&bdev->bd_device);
769 }
770 
771 /**
772  * blkdev_get_by_dev - open a block device by device number
773  * @dev: device number of block device to open
774  * @mode: FMODE_* mask
775  * @holder: exclusive holder identifier
776  *
777  * Open the block device described by device number @dev. If @mode includes
778  * %FMODE_EXCL, the block device is opened with exclusive access.  Specifying
779  * %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may nest for
780  * the same @holder.
781  *
782  * Use this interface ONLY if you really do not have anything better - i.e. when
783  * you are behind a truly sucky interface and all you are given is a device
784  * number.  Everything else should use blkdev_get_by_path().
785  *
786  * CONTEXT:
787  * Might sleep.
788  *
789  * RETURNS:
790  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
791  */
792 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
793 {
794 	bool unblock_events = true;
795 	struct block_device *bdev;
796 	struct gendisk *disk;
797 	int ret;
798 
799 	ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
800 			MAJOR(dev), MINOR(dev),
801 			((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
802 			((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
803 	if (ret)
804 		return ERR_PTR(ret);
805 
806 	bdev = blkdev_get_no_open(dev);
807 	if (!bdev)
808 		return ERR_PTR(-ENXIO);
809 	disk = bdev->bd_disk;
810 
811 	if (mode & FMODE_EXCL) {
812 		ret = bd_prepare_to_claim(bdev, holder);
813 		if (ret)
814 			goto put_blkdev;
815 	}
816 
817 	disk_block_events(disk);
818 
819 	mutex_lock(&disk->open_mutex);
820 	ret = -ENXIO;
821 	if (!disk_live(disk))
822 		goto abort_claiming;
823 	if (bdev_is_partition(bdev))
824 		ret = blkdev_get_part(bdev, mode);
825 	else
826 		ret = blkdev_get_whole(bdev, mode);
827 	if (ret)
828 		goto abort_claiming;
829 	if (mode & FMODE_EXCL) {
830 		bd_finish_claiming(bdev, holder);
831 
832 		/*
833 		 * Block event polling for write claims if requested.  Any write
834 		 * holder makes the write_holder state stick until all are
835 		 * released.  This is good enough and tracking individual
836 		 * writeable reference is too fragile given the way @mode is
837 		 * used in blkdev_get/put().
838 		 */
839 		if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
840 		    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
841 			bdev->bd_write_holder = true;
842 			unblock_events = false;
843 		}
844 	}
845 	mutex_unlock(&disk->open_mutex);
846 
847 	if (unblock_events)
848 		disk_unblock_events(disk);
849 	return bdev;
850 
851 abort_claiming:
852 	if (mode & FMODE_EXCL)
853 		bd_abort_claiming(bdev, holder);
854 	mutex_unlock(&disk->open_mutex);
855 	disk_unblock_events(disk);
856 put_blkdev:
857 	blkdev_put_no_open(bdev);
858 	return ERR_PTR(ret);
859 }
860 EXPORT_SYMBOL(blkdev_get_by_dev);
861 
862 /**
863  * blkdev_get_by_path - open a block device by name
864  * @path: path to the block device to open
865  * @mode: FMODE_* mask
866  * @holder: exclusive holder identifier
867  *
868  * Open the block device described by the device file at @path.  If @mode
869  * includes %FMODE_EXCL, the block device is opened with exclusive access.
870  * Specifying %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may
871  * nest for the same @holder.
872  *
873  * CONTEXT:
874  * Might sleep.
875  *
876  * RETURNS:
877  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
878  */
879 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
880 					void *holder)
881 {
882 	struct block_device *bdev;
883 	dev_t dev;
884 	int error;
885 
886 	error = lookup_bdev(path, &dev);
887 	if (error)
888 		return ERR_PTR(error);
889 
890 	bdev = blkdev_get_by_dev(dev, mode, holder);
891 	if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
892 		blkdev_put(bdev, mode);
893 		return ERR_PTR(-EACCES);
894 	}
895 
896 	return bdev;
897 }
898 EXPORT_SYMBOL(blkdev_get_by_path);
899 
900 void blkdev_put(struct block_device *bdev, fmode_t mode)
901 {
902 	struct gendisk *disk = bdev->bd_disk;
903 
904 	/*
905 	 * Sync early if it looks like we're the last one.  If someone else
906 	 * opens the block device between now and the decrement of bd_openers
907 	 * then we did a sync that we didn't need to, but that's not the end
908 	 * of the world and we want to avoid long (could be several minute)
909 	 * syncs while holding the mutex.
910 	 */
911 	if (bdev->bd_openers == 1)
912 		sync_blockdev(bdev);
913 
914 	mutex_lock(&disk->open_mutex);
915 	if (mode & FMODE_EXCL) {
916 		struct block_device *whole = bdev_whole(bdev);
917 		bool bdev_free;
918 
919 		/*
920 		 * Release a claim on the device.  The holder fields
921 		 * are protected with bdev_lock.  open_mutex is to
922 		 * synchronize disk_holder unlinking.
923 		 */
924 		spin_lock(&bdev_lock);
925 
926 		WARN_ON_ONCE(--bdev->bd_holders < 0);
927 		WARN_ON_ONCE(--whole->bd_holders < 0);
928 
929 		if ((bdev_free = !bdev->bd_holders))
930 			bdev->bd_holder = NULL;
931 		if (!whole->bd_holders)
932 			whole->bd_holder = NULL;
933 
934 		spin_unlock(&bdev_lock);
935 
936 		/*
937 		 * If this was the last claim, remove holder link and
938 		 * unblock evpoll if it was a write holder.
939 		 */
940 		if (bdev_free && bdev->bd_write_holder) {
941 			disk_unblock_events(disk);
942 			bdev->bd_write_holder = false;
943 		}
944 	}
945 
946 	/*
947 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
948 	 * event.  This is to ensure detection of media removal commanded
949 	 * from userland - e.g. eject(1).
950 	 */
951 	disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
952 
953 	if (bdev_is_partition(bdev))
954 		blkdev_put_part(bdev, mode);
955 	else
956 		blkdev_put_whole(bdev, mode);
957 	mutex_unlock(&disk->open_mutex);
958 
959 	blkdev_put_no_open(bdev);
960 }
961 EXPORT_SYMBOL(blkdev_put);
962 
963 /**
964  * lookup_bdev  - lookup a struct block_device by name
965  * @pathname:	special file representing the block device
966  * @dev:	return value of the block device's dev_t
967  *
968  * Lookup the block device's dev_t at @pathname in the current
969  * namespace if possible and return it by @dev.
970  *
971  * RETURNS:
972  * 0 if succeeded, errno otherwise.
973  */
974 int lookup_bdev(const char *pathname, dev_t *dev)
975 {
976 	struct inode *inode;
977 	struct path path;
978 	int error;
979 
980 	if (!pathname || !*pathname)
981 		return -EINVAL;
982 
983 	error = kern_path(pathname, LOOKUP_FOLLOW, &path);
984 	if (error)
985 		return error;
986 
987 	inode = d_backing_inode(path.dentry);
988 	error = -ENOTBLK;
989 	if (!S_ISBLK(inode->i_mode))
990 		goto out_path_put;
991 	error = -EACCES;
992 	if (!may_open_dev(&path))
993 		goto out_path_put;
994 
995 	*dev = inode->i_rdev;
996 	error = 0;
997 out_path_put:
998 	path_put(&path);
999 	return error;
1000 }
1001 EXPORT_SYMBOL(lookup_bdev);
1002 
1003 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1004 {
1005 	struct super_block *sb = get_super(bdev);
1006 	int res = 0;
1007 
1008 	if (sb) {
1009 		/*
1010 		 * no need to lock the super, get_super holds the
1011 		 * read mutex so the filesystem cannot go away
1012 		 * under us (->put_super runs with the write lock
1013 		 * hold).
1014 		 */
1015 		shrink_dcache_sb(sb);
1016 		res = invalidate_inodes(sb, kill_dirty);
1017 		drop_super(sb);
1018 	}
1019 	invalidate_bdev(bdev);
1020 	return res;
1021 }
1022 EXPORT_SYMBOL(__invalidate_device);
1023 
1024 void sync_bdevs(bool wait)
1025 {
1026 	struct inode *inode, *old_inode = NULL;
1027 
1028 	spin_lock(&blockdev_superblock->s_inode_list_lock);
1029 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1030 		struct address_space *mapping = inode->i_mapping;
1031 		struct block_device *bdev;
1032 
1033 		spin_lock(&inode->i_lock);
1034 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1035 		    mapping->nrpages == 0) {
1036 			spin_unlock(&inode->i_lock);
1037 			continue;
1038 		}
1039 		__iget(inode);
1040 		spin_unlock(&inode->i_lock);
1041 		spin_unlock(&blockdev_superblock->s_inode_list_lock);
1042 		/*
1043 		 * We hold a reference to 'inode' so it couldn't have been
1044 		 * removed from s_inodes list while we dropped the
1045 		 * s_inode_list_lock  We cannot iput the inode now as we can
1046 		 * be holding the last reference and we cannot iput it under
1047 		 * s_inode_list_lock. So we keep the reference and iput it
1048 		 * later.
1049 		 */
1050 		iput(old_inode);
1051 		old_inode = inode;
1052 		bdev = I_BDEV(inode);
1053 
1054 		mutex_lock(&bdev->bd_disk->open_mutex);
1055 		if (!bdev->bd_openers) {
1056 			; /* skip */
1057 		} else if (wait) {
1058 			/*
1059 			 * We keep the error status of individual mapping so
1060 			 * that applications can catch the writeback error using
1061 			 * fsync(2). See filemap_fdatawait_keep_errors() for
1062 			 * details.
1063 			 */
1064 			filemap_fdatawait_keep_errors(inode->i_mapping);
1065 		} else {
1066 			filemap_fdatawrite(inode->i_mapping);
1067 		}
1068 		mutex_unlock(&bdev->bd_disk->open_mutex);
1069 
1070 		spin_lock(&blockdev_superblock->s_inode_list_lock);
1071 	}
1072 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
1073 	iput(old_inode);
1074 }
1075