• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 <linux/stat.h>
31 #include "../fs/internal.h"
32 #include "blk.h"
33 
34 struct bdev_inode {
35 	struct block_device bdev;
36 	struct inode vfs_inode;
37 };
38 
BDEV_I(struct inode * inode)39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
40 {
41 	return container_of(inode, struct bdev_inode, vfs_inode);
42 }
43 
I_BDEV(struct inode * inode)44 struct block_device *I_BDEV(struct inode *inode)
45 {
46 	return &BDEV_I(inode)->bdev;
47 }
48 EXPORT_SYMBOL(I_BDEV);
49 
bdev_write_inode(struct block_device * bdev)50 static void bdev_write_inode(struct block_device *bdev)
51 {
52 	struct inode *inode = bdev->bd_inode;
53 	int ret;
54 
55 	spin_lock(&inode->i_lock);
56 	while (inode->i_state & I_DIRTY) {
57 		spin_unlock(&inode->i_lock);
58 		ret = write_inode_now(inode, true);
59 		if (ret)
60 			pr_warn_ratelimited(
61 	"VFS: Dirty inode writeback failed for block device %pg (err=%d).\n",
62 				bdev, ret);
63 		spin_lock(&inode->i_lock);
64 	}
65 	spin_unlock(&inode->i_lock);
66 }
67 
68 /* Kill _all_ buffers and pagecache , dirty or not.. */
kill_bdev(struct block_device * bdev)69 static void kill_bdev(struct block_device *bdev)
70 {
71 	struct address_space *mapping = bdev->bd_inode->i_mapping;
72 
73 	if (mapping_empty(mapping))
74 		return;
75 
76 	invalidate_bh_lrus();
77 	truncate_inode_pages(mapping, 0);
78 }
79 
80 /* Invalidate clean unused buffers and pagecache. */
invalidate_bdev(struct block_device * bdev)81 void invalidate_bdev(struct block_device *bdev)
82 {
83 	struct address_space *mapping = bdev->bd_inode->i_mapping;
84 
85 	if (mapping->nrpages) {
86 		invalidate_bh_lrus();
87 		lru_add_drain_all();	/* make sure all lru add caches are flushed */
88 		invalidate_mapping_pages(mapping, 0, -1);
89 	}
90 	/* 99% of the time, we don't need to flush the cleancache on the bdev.
91 	 * But, for the strange corners, lets be cautious
92 	 */
93 	cleancache_invalidate_inode(mapping);
94 }
95 EXPORT_SYMBOL(invalidate_bdev);
96 
97 /*
98  * Drop all buffers & page cache for given bdev range. This function bails
99  * with error if bdev has other exclusive owner (such as filesystem).
100  */
truncate_bdev_range(struct block_device * bdev,fmode_t mode,loff_t lstart,loff_t lend)101 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
102 			loff_t lstart, loff_t lend)
103 {
104 	/*
105 	 * If we don't hold exclusive handle for the device, upgrade to it
106 	 * while we discard the buffer cache to avoid discarding buffers
107 	 * under live filesystem.
108 	 */
109 	if (!(mode & FMODE_EXCL)) {
110 		int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
111 		if (err)
112 			goto invalidate;
113 	}
114 
115 	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
116 	if (!(mode & FMODE_EXCL))
117 		bd_abort_claiming(bdev, truncate_bdev_range);
118 	return 0;
119 
120 invalidate:
121 	/*
122 	 * Someone else has handle exclusively open. Try invalidating instead.
123 	 * The 'end' argument is inclusive so the rounding is safe.
124 	 */
125 	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
126 					     lstart >> PAGE_SHIFT,
127 					     lend >> PAGE_SHIFT);
128 }
129 
set_init_blocksize(struct block_device * bdev)130 static void set_init_blocksize(struct block_device *bdev)
131 {
132 	unsigned int bsize = bdev_logical_block_size(bdev);
133 	loff_t size = i_size_read(bdev->bd_inode);
134 
135 	while (bsize < PAGE_SIZE) {
136 		if (size & bsize)
137 			break;
138 		bsize <<= 1;
139 	}
140 	bdev->bd_inode->i_blkbits = blksize_bits(bsize);
141 }
142 
set_blocksize(struct block_device * bdev,int size)143 int set_blocksize(struct block_device *bdev, int size)
144 {
145 	/* Size must be a power of two, and between 512 and PAGE_SIZE */
146 	if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
147 		return -EINVAL;
148 
149 	/* Size cannot be smaller than the size supported by the device */
150 	if (size < bdev_logical_block_size(bdev))
151 		return -EINVAL;
152 
153 	/* Don't change the size if it is same as current */
154 	if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
155 		sync_blockdev(bdev);
156 		bdev->bd_inode->i_blkbits = blksize_bits(size);
157 		kill_bdev(bdev);
158 	}
159 	return 0;
160 }
161 
162 EXPORT_SYMBOL(set_blocksize);
163 
sb_set_blocksize(struct super_block * sb,int size)164 int sb_set_blocksize(struct super_block *sb, int size)
165 {
166 	if (set_blocksize(sb->s_bdev, size))
167 		return 0;
168 	/* If we get here, we know size is power of two
169 	 * and it's value is between 512 and PAGE_SIZE */
170 	sb->s_blocksize = size;
171 	sb->s_blocksize_bits = blksize_bits(size);
172 	return sb->s_blocksize;
173 }
174 
175 EXPORT_SYMBOL(sb_set_blocksize);
176 
sb_min_blocksize(struct super_block * sb,int size)177 int sb_min_blocksize(struct super_block *sb, int size)
178 {
179 	int minsize = bdev_logical_block_size(sb->s_bdev);
180 	if (size < minsize)
181 		size = minsize;
182 	return sb_set_blocksize(sb, size);
183 }
184 
185 EXPORT_SYMBOL(sb_min_blocksize);
186 
sync_blockdev_nowait(struct block_device * bdev)187 int sync_blockdev_nowait(struct block_device *bdev)
188 {
189 	if (!bdev)
190 		return 0;
191 	return filemap_flush(bdev->bd_inode->i_mapping);
192 }
193 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
194 
195 /*
196  * Write out and wait upon all the dirty data associated with a block
197  * device via its mapping.  Does not take the superblock lock.
198  */
sync_blockdev(struct block_device * bdev)199 int sync_blockdev(struct block_device *bdev)
200 {
201 	if (!bdev)
202 		return 0;
203 	return filemap_write_and_wait(bdev->bd_inode->i_mapping);
204 }
205 EXPORT_SYMBOL(sync_blockdev);
206 
sync_blockdev_range(struct block_device * bdev,loff_t lstart,loff_t lend)207 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend)
208 {
209 	return filemap_write_and_wait_range(bdev->bd_inode->i_mapping,
210 			lstart, lend);
211 }
212 EXPORT_SYMBOL(sync_blockdev_range);
213 
214 /*
215  * Write out and wait upon all dirty data associated with this
216  * device.   Filesystem data as well as the underlying block
217  * device.  Takes the superblock lock.
218  */
fsync_bdev(struct block_device * bdev)219 int fsync_bdev(struct block_device *bdev)
220 {
221 	struct super_block *sb = get_super(bdev);
222 	if (sb) {
223 		int res = sync_filesystem(sb);
224 		drop_super(sb);
225 		return res;
226 	}
227 	return sync_blockdev(bdev);
228 }
229 EXPORT_SYMBOL(fsync_bdev);
230 
231 /**
232  * freeze_bdev  --  lock a filesystem and force it into a consistent state
233  * @bdev:	blockdevice to lock
234  *
235  * If a superblock is found on this device, we take the s_umount semaphore
236  * on it to make sure nobody unmounts until the snapshot creation is done.
237  * The reference counter (bd_fsfreeze_count) guarantees that only the last
238  * unfreeze process can unfreeze the frozen filesystem actually when multiple
239  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
240  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
241  * actually.
242  */
freeze_bdev(struct block_device * bdev)243 int freeze_bdev(struct block_device *bdev)
244 {
245 	struct super_block *sb;
246 	int error = 0;
247 
248 	mutex_lock(&bdev->bd_fsfreeze_mutex);
249 	if (++bdev->bd_fsfreeze_count > 1)
250 		goto done;
251 
252 	sb = get_active_super(bdev);
253 	if (!sb)
254 		goto sync;
255 	if (sb->s_op->freeze_super)
256 		error = sb->s_op->freeze_super(sb);
257 	else
258 		error = freeze_super(sb);
259 	deactivate_super(sb);
260 
261 	if (error) {
262 		bdev->bd_fsfreeze_count--;
263 		goto done;
264 	}
265 	bdev->bd_fsfreeze_sb = sb;
266 
267 sync:
268 	sync_blockdev(bdev);
269 done:
270 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
271 	return error;
272 }
273 EXPORT_SYMBOL(freeze_bdev);
274 
275 /**
276  * thaw_bdev  -- unlock filesystem
277  * @bdev:	blockdevice to unlock
278  *
279  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
280  */
thaw_bdev(struct block_device * bdev)281 int thaw_bdev(struct block_device *bdev)
282 {
283 	struct super_block *sb;
284 	int error = -EINVAL;
285 
286 	mutex_lock(&bdev->bd_fsfreeze_mutex);
287 	if (!bdev->bd_fsfreeze_count)
288 		goto out;
289 
290 	error = 0;
291 	if (--bdev->bd_fsfreeze_count > 0)
292 		goto out;
293 
294 	sb = bdev->bd_fsfreeze_sb;
295 	if (!sb)
296 		goto out;
297 
298 	if (sb->s_op->thaw_super)
299 		error = sb->s_op->thaw_super(sb);
300 	else
301 		error = thaw_super(sb);
302 	if (error)
303 		bdev->bd_fsfreeze_count++;
304 	else
305 		bdev->bd_fsfreeze_sb = NULL;
306 out:
307 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
308 	return error;
309 }
310 EXPORT_SYMBOL(thaw_bdev);
311 
312 /**
313  * bdev_read_page() - Start reading a page from a block device
314  * @bdev: The device to read the page from
315  * @sector: The offset on the device to read the page to (need not be aligned)
316  * @page: The page to read
317  *
318  * On entry, the page should be locked.  It will be unlocked when the page
319  * has been read.  If the block driver implements rw_page synchronously,
320  * that will be true on exit from this function, but it need not be.
321  *
322  * Errors returned by this function are usually "soft", eg out of memory, or
323  * queue full; callers should try a different route to read this page rather
324  * than propagate an error back up the stack.
325  *
326  * Return: negative errno if an error occurs, 0 if submission was successful.
327  */
bdev_read_page(struct block_device * bdev,sector_t sector,struct page * page)328 int bdev_read_page(struct block_device *bdev, sector_t sector,
329 			struct page *page)
330 {
331 	const struct block_device_operations *ops = bdev->bd_disk->fops;
332 	int result = -EOPNOTSUPP;
333 
334 	if (!ops->rw_page || bdev_get_integrity(bdev))
335 		return result;
336 
337 	result = blk_queue_enter(bdev_get_queue(bdev), 0);
338 	if (result)
339 		return result;
340 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
341 			      REQ_OP_READ);
342 	blk_queue_exit(bdev_get_queue(bdev));
343 	return result;
344 }
345 
346 /**
347  * bdev_write_page() - Start writing a page to a block device
348  * @bdev: The device to write the page to
349  * @sector: The offset on the device to write the page to (need not be aligned)
350  * @page: The page to write
351  * @wbc: The writeback_control for the write
352  *
353  * On entry, the page should be locked and not currently under writeback.
354  * On exit, if the write started successfully, the page will be unlocked and
355  * under writeback.  If the write failed already (eg the driver failed to
356  * queue the page to the device), the page will still be locked.  If the
357  * caller is a ->writepage implementation, it will need to unlock the page.
358  *
359  * Errors returned by this function are usually "soft", eg out of memory, or
360  * queue full; callers should try a different route to write this page rather
361  * than propagate an error back up the stack.
362  *
363  * Return: negative errno if an error occurs, 0 if submission was successful.
364  */
bdev_write_page(struct block_device * bdev,sector_t sector,struct page * page,struct writeback_control * wbc)365 int bdev_write_page(struct block_device *bdev, sector_t sector,
366 			struct page *page, struct writeback_control *wbc)
367 {
368 	int result;
369 	const struct block_device_operations *ops = bdev->bd_disk->fops;
370 
371 	if (!ops->rw_page || bdev_get_integrity(bdev))
372 		return -EOPNOTSUPP;
373 	result = blk_queue_enter(bdev_get_queue(bdev), 0);
374 	if (result)
375 		return result;
376 
377 	set_page_writeback(page);
378 	result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
379 			      REQ_OP_WRITE);
380 	if (result) {
381 		end_page_writeback(page);
382 	} else {
383 		clean_page_buffers(page);
384 		unlock_page(page);
385 	}
386 	blk_queue_exit(bdev_get_queue(bdev));
387 	return result;
388 }
389 
390 /*
391  * pseudo-fs
392  */
393 
394 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
395 static struct kmem_cache * bdev_cachep __read_mostly;
396 
bdev_alloc_inode(struct super_block * sb)397 static struct inode *bdev_alloc_inode(struct super_block *sb)
398 {
399 	struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
400 
401 	if (!ei)
402 		return NULL;
403 	memset(&ei->bdev, 0, sizeof(ei->bdev));
404 	return &ei->vfs_inode;
405 }
406 
bdev_free_inode(struct inode * inode)407 static void bdev_free_inode(struct inode *inode)
408 {
409 	struct block_device *bdev = I_BDEV(inode);
410 
411 	free_percpu(bdev->bd_stats);
412 	kfree(bdev->bd_meta_info);
413 
414 	if (!bdev_is_partition(bdev)) {
415 		if (bdev->bd_disk && bdev->bd_disk->bdi)
416 			bdi_put(bdev->bd_disk->bdi);
417 		kfree(bdev->bd_disk);
418 	}
419 
420 	if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
421 		blk_free_ext_minor(MINOR(bdev->bd_dev));
422 
423 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
424 }
425 
init_once(void * data)426 static void init_once(void *data)
427 {
428 	struct bdev_inode *ei = data;
429 
430 	inode_init_once(&ei->vfs_inode);
431 }
432 
bdev_evict_inode(struct inode * inode)433 static void bdev_evict_inode(struct inode *inode)
434 {
435 	truncate_inode_pages_final(&inode->i_data);
436 	invalidate_inode_buffers(inode); /* is it needed here? */
437 	clear_inode(inode);
438 }
439 
440 static const struct super_operations bdev_sops = {
441 	.statfs = simple_statfs,
442 	.alloc_inode = bdev_alloc_inode,
443 	.free_inode = bdev_free_inode,
444 	.drop_inode = generic_delete_inode,
445 	.evict_inode = bdev_evict_inode,
446 };
447 
bd_init_fs_context(struct fs_context * fc)448 static int bd_init_fs_context(struct fs_context *fc)
449 {
450 	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
451 	if (!ctx)
452 		return -ENOMEM;
453 	fc->s_iflags |= SB_I_CGROUPWB;
454 	ctx->ops = &bdev_sops;
455 	return 0;
456 }
457 
458 static struct file_system_type bd_type = {
459 	.name		= "bdev",
460 	.init_fs_context = bd_init_fs_context,
461 	.kill_sb	= kill_anon_super,
462 };
463 
464 struct super_block *blockdev_superblock __read_mostly;
465 EXPORT_SYMBOL_GPL(blockdev_superblock);
466 
bdev_cache_init(void)467 void __init bdev_cache_init(void)
468 {
469 	int err;
470 	static struct vfsmount *bd_mnt;
471 
472 	bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
473 			0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
474 				SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
475 			init_once);
476 	err = register_filesystem(&bd_type);
477 	if (err)
478 		panic("Cannot register bdev pseudo-fs");
479 	bd_mnt = kern_mount(&bd_type);
480 	if (IS_ERR(bd_mnt))
481 		panic("Cannot create bdev pseudo-fs");
482 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
483 }
484 
bdev_alloc(struct gendisk * disk,u8 partno)485 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
486 {
487 	struct block_device *bdev;
488 	struct inode *inode;
489 
490 	inode = new_inode(blockdev_superblock);
491 	if (!inode)
492 		return NULL;
493 	inode->i_mode = S_IFBLK;
494 	inode->i_rdev = 0;
495 	inode->i_data.a_ops = &def_blk_aops;
496 	mapping_set_gfp_mask(&inode->i_data, GFP_USER);
497 
498 	bdev = I_BDEV(inode);
499 	mutex_init(&bdev->bd_fsfreeze_mutex);
500 	spin_lock_init(&bdev->bd_size_lock);
501 	bdev->bd_partno = partno;
502 	bdev->bd_inode = inode;
503 	bdev->bd_queue = disk->queue;
504 	bdev->bd_stats = alloc_percpu(struct disk_stats);
505 	if (!bdev->bd_stats) {
506 		iput(inode);
507 		return NULL;
508 	}
509 	bdev->bd_disk = disk;
510 	return bdev;
511 }
512 
bdev_add(struct block_device * bdev,dev_t dev)513 void bdev_add(struct block_device *bdev, dev_t dev)
514 {
515 	if (bdev_stable_writes(bdev))
516 		mapping_set_stable_writes(bdev->bd_inode->i_mapping);
517 	bdev->bd_dev = dev;
518 	bdev->bd_inode->i_rdev = dev;
519 	bdev->bd_inode->i_ino = dev;
520 	insert_inode_hash(bdev->bd_inode);
521 }
522 
nr_blockdev_pages(void)523 long nr_blockdev_pages(void)
524 {
525 	struct inode *inode;
526 	long ret = 0;
527 
528 	spin_lock(&blockdev_superblock->s_inode_list_lock);
529 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
530 		ret += inode->i_mapping->nrpages;
531 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
532 
533 	return ret;
534 }
535 
536 /**
537  * bd_may_claim - test whether a block device can be claimed
538  * @bdev: block device of interest
539  * @whole: whole block device containing @bdev, may equal @bdev
540  * @holder: holder trying to claim @bdev
541  *
542  * Test whether @bdev can be claimed by @holder.
543  *
544  * CONTEXT:
545  * spin_lock(&bdev_lock).
546  *
547  * RETURNS:
548  * %true if @bdev can be claimed, %false otherwise.
549  */
bd_may_claim(struct block_device * bdev,struct block_device * whole,void * holder)550 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
551 			 void *holder)
552 {
553 	if (bdev->bd_holder == holder)
554 		return true;	 /* already a holder */
555 	else if (bdev->bd_holder != NULL)
556 		return false; 	 /* held by someone else */
557 	else if (whole == bdev)
558 		return true;  	 /* is a whole device which isn't held */
559 
560 	else if (whole->bd_holder == bd_may_claim)
561 		return true; 	 /* is a partition of a device that is being partitioned */
562 	else if (whole->bd_holder != NULL)
563 		return false;	 /* is a partition of a held device */
564 	else
565 		return true;	 /* is a partition of an un-held device */
566 }
567 
568 /**
569  * bd_prepare_to_claim - claim a block device
570  * @bdev: block device of interest
571  * @holder: holder trying to claim @bdev
572  *
573  * Claim @bdev.  This function fails if @bdev is already claimed by another
574  * holder and waits if another claiming is in progress. return, the caller
575  * has ownership of bd_claiming and bd_holder[s].
576  *
577  * RETURNS:
578  * 0 if @bdev can be claimed, -EBUSY otherwise.
579  */
bd_prepare_to_claim(struct block_device * bdev,void * holder)580 int bd_prepare_to_claim(struct block_device *bdev, void *holder)
581 {
582 	struct block_device *whole = bdev_whole(bdev);
583 
584 	if (WARN_ON_ONCE(!holder))
585 		return -EINVAL;
586 retry:
587 	spin_lock(&bdev_lock);
588 	/* if someone else claimed, fail */
589 	if (!bd_may_claim(bdev, whole, holder)) {
590 		spin_unlock(&bdev_lock);
591 		return -EBUSY;
592 	}
593 
594 	/* if claiming is already in progress, wait for it to finish */
595 	if (whole->bd_claiming) {
596 		wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
597 		DEFINE_WAIT(wait);
598 
599 		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
600 		spin_unlock(&bdev_lock);
601 		schedule();
602 		finish_wait(wq, &wait);
603 		goto retry;
604 	}
605 
606 	/* yay, all mine */
607 	whole->bd_claiming = holder;
608 	spin_unlock(&bdev_lock);
609 	return 0;
610 }
611 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
612 
bd_clear_claiming(struct block_device * whole,void * holder)613 static void bd_clear_claiming(struct block_device *whole, void *holder)
614 {
615 	lockdep_assert_held(&bdev_lock);
616 	/* tell others that we're done */
617 	BUG_ON(whole->bd_claiming != holder);
618 	whole->bd_claiming = NULL;
619 	wake_up_bit(&whole->bd_claiming, 0);
620 }
621 
622 /**
623  * bd_finish_claiming - finish claiming of a block device
624  * @bdev: block device of interest
625  * @holder: holder that has claimed @bdev
626  *
627  * Finish exclusive open of a block device. Mark the device as exlusively
628  * open by the holder and wake up all waiters for exclusive open to finish.
629  */
bd_finish_claiming(struct block_device * bdev,void * holder)630 static void bd_finish_claiming(struct block_device *bdev, void *holder)
631 {
632 	struct block_device *whole = bdev_whole(bdev);
633 
634 	spin_lock(&bdev_lock);
635 	BUG_ON(!bd_may_claim(bdev, whole, holder));
636 	/*
637 	 * Note that for a whole device bd_holders will be incremented twice,
638 	 * and bd_holder will be set to bd_may_claim before being set to holder
639 	 */
640 	whole->bd_holders++;
641 	whole->bd_holder = bd_may_claim;
642 	bdev->bd_holders++;
643 	bdev->bd_holder = holder;
644 	bd_clear_claiming(whole, holder);
645 	spin_unlock(&bdev_lock);
646 }
647 
648 /**
649  * bd_abort_claiming - abort claiming of a block device
650  * @bdev: block device of interest
651  * @holder: holder that has claimed @bdev
652  *
653  * Abort claiming of a block device when the exclusive open failed. This can be
654  * also used when exclusive open is not actually desired and we just needed
655  * to block other exclusive openers for a while.
656  */
bd_abort_claiming(struct block_device * bdev,void * holder)657 void bd_abort_claiming(struct block_device *bdev, void *holder)
658 {
659 	spin_lock(&bdev_lock);
660 	bd_clear_claiming(bdev_whole(bdev), holder);
661 	spin_unlock(&bdev_lock);
662 }
663 EXPORT_SYMBOL(bd_abort_claiming);
664 
blkdev_flush_mapping(struct block_device * bdev)665 static void blkdev_flush_mapping(struct block_device *bdev)
666 {
667 	WARN_ON_ONCE(bdev->bd_holders);
668 	sync_blockdev(bdev);
669 	kill_bdev(bdev);
670 	bdev_write_inode(bdev);
671 }
672 
blkdev_get_whole(struct block_device * bdev,fmode_t mode)673 static int blkdev_get_whole(struct block_device *bdev, fmode_t mode)
674 {
675 	struct gendisk *disk = bdev->bd_disk;
676 	int ret;
677 
678 	if (disk->fops->open) {
679 		ret = disk->fops->open(bdev, mode);
680 		if (ret) {
681 			/* avoid ghost partitions on a removed medium */
682 			if (ret == -ENOMEDIUM &&
683 			     test_bit(GD_NEED_PART_SCAN, &disk->state))
684 				bdev_disk_changed(disk, true);
685 			return ret;
686 		}
687 	}
688 
689 	if (!atomic_read(&bdev->bd_openers))
690 		set_init_blocksize(bdev);
691 	if (test_bit(GD_NEED_PART_SCAN, &disk->state))
692 		bdev_disk_changed(disk, false);
693 	atomic_inc(&bdev->bd_openers);
694 	return 0;
695 }
696 
blkdev_put_whole(struct block_device * bdev,fmode_t mode)697 static void blkdev_put_whole(struct block_device *bdev, fmode_t mode)
698 {
699 	if (atomic_dec_and_test(&bdev->bd_openers))
700 		blkdev_flush_mapping(bdev);
701 	if (bdev->bd_disk->fops->release)
702 		bdev->bd_disk->fops->release(bdev->bd_disk, mode);
703 }
704 
blkdev_get_part(struct block_device * part,fmode_t mode)705 static int blkdev_get_part(struct block_device *part, fmode_t mode)
706 {
707 	struct gendisk *disk = part->bd_disk;
708 	int ret;
709 
710 	if (atomic_read(&part->bd_openers))
711 		goto done;
712 
713 	ret = blkdev_get_whole(bdev_whole(part), mode);
714 	if (ret)
715 		return ret;
716 
717 	ret = -ENXIO;
718 	if (!bdev_nr_sectors(part))
719 		goto out_blkdev_put;
720 
721 	disk->open_partitions++;
722 	set_init_blocksize(part);
723 done:
724 	atomic_inc(&part->bd_openers);
725 	return 0;
726 
727 out_blkdev_put:
728 	blkdev_put_whole(bdev_whole(part), mode);
729 	return ret;
730 }
731 
blkdev_put_part(struct block_device * part,fmode_t mode)732 static void blkdev_put_part(struct block_device *part, fmode_t mode)
733 {
734 	struct block_device *whole = bdev_whole(part);
735 
736 	if (!atomic_dec_and_test(&part->bd_openers))
737 		return;
738 	blkdev_flush_mapping(part);
739 	whole->bd_disk->open_partitions--;
740 	blkdev_put_whole(whole, mode);
741 }
742 
blkdev_get_no_open(dev_t dev)743 struct block_device *blkdev_get_no_open(dev_t dev)
744 {
745 	struct block_device *bdev;
746 	struct inode *inode;
747 
748 	inode = ilookup(blockdev_superblock, dev);
749 	if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
750 		blk_request_module(dev);
751 		inode = ilookup(blockdev_superblock, dev);
752 		if (inode)
753 			pr_warn_ratelimited(
754 "block device autoloading is deprecated and will be removed.\n");
755 	}
756 	if (!inode)
757 		return NULL;
758 
759 	/* switch from the inode reference to a device mode one: */
760 	bdev = &BDEV_I(inode)->bdev;
761 	if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
762 		bdev = NULL;
763 	iput(inode);
764 	return bdev;
765 }
766 
blkdev_put_no_open(struct block_device * bdev)767 void blkdev_put_no_open(struct block_device *bdev)
768 {
769 	put_device(&bdev->bd_device);
770 }
771 
772 /**
773  * blkdev_get_by_dev - open a block device by device number
774  * @dev: device number of block device to open
775  * @mode: FMODE_* mask
776  * @holder: exclusive holder identifier
777  *
778  * Open the block device described by device number @dev. If @mode includes
779  * %FMODE_EXCL, the block device is opened with exclusive access.  Specifying
780  * %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may nest for
781  * the same @holder.
782  *
783  * Use this interface ONLY if you really do not have anything better - i.e. when
784  * you are behind a truly sucky interface and all you are given is a device
785  * number.  Everything else should use blkdev_get_by_path().
786  *
787  * CONTEXT:
788  * Might sleep.
789  *
790  * RETURNS:
791  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
792  */
blkdev_get_by_dev(dev_t dev,fmode_t mode,void * holder)793 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
794 {
795 	bool unblock_events = true;
796 	struct block_device *bdev;
797 	struct gendisk *disk;
798 	int ret;
799 
800 	ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
801 			MAJOR(dev), MINOR(dev),
802 			((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
803 			((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
804 	if (ret)
805 		return ERR_PTR(ret);
806 
807 	bdev = blkdev_get_no_open(dev);
808 	if (!bdev)
809 		return ERR_PTR(-ENXIO);
810 	disk = bdev->bd_disk;
811 
812 	if (mode & FMODE_EXCL) {
813 		ret = bd_prepare_to_claim(bdev, holder);
814 		if (ret)
815 			goto put_blkdev;
816 	}
817 
818 	disk_block_events(disk);
819 
820 	mutex_lock(&disk->open_mutex);
821 	ret = -ENXIO;
822 	if (!disk_live(disk))
823 		goto abort_claiming;
824 	if (!try_module_get(disk->fops->owner))
825 		goto abort_claiming;
826 	if (bdev_is_partition(bdev))
827 		ret = blkdev_get_part(bdev, mode);
828 	else
829 		ret = blkdev_get_whole(bdev, mode);
830 	if (ret)
831 		goto put_module;
832 	if (mode & FMODE_EXCL) {
833 		bd_finish_claiming(bdev, holder);
834 
835 		/*
836 		 * Block event polling for write claims if requested.  Any write
837 		 * holder makes the write_holder state stick until all are
838 		 * released.  This is good enough and tracking individual
839 		 * writeable reference is too fragile given the way @mode is
840 		 * used in blkdev_get/put().
841 		 */
842 		if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
843 		    (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
844 			bdev->bd_write_holder = true;
845 			unblock_events = false;
846 		}
847 	}
848 	mutex_unlock(&disk->open_mutex);
849 
850 	if (unblock_events)
851 		disk_unblock_events(disk);
852 	return bdev;
853 put_module:
854 	module_put(disk->fops->owner);
855 abort_claiming:
856 	if (mode & FMODE_EXCL)
857 		bd_abort_claiming(bdev, holder);
858 	mutex_unlock(&disk->open_mutex);
859 	disk_unblock_events(disk);
860 put_blkdev:
861 	blkdev_put_no_open(bdev);
862 	return ERR_PTR(ret);
863 }
864 EXPORT_SYMBOL(blkdev_get_by_dev);
865 
866 /**
867  * blkdev_get_by_path - open a block device by name
868  * @path: path to the block device to open
869  * @mode: FMODE_* mask
870  * @holder: exclusive holder identifier
871  *
872  * Open the block device described by the device file at @path.  If @mode
873  * includes %FMODE_EXCL, the block device is opened with exclusive access.
874  * Specifying %FMODE_EXCL with a %NULL @holder is invalid.  Exclusive opens may
875  * nest for the same @holder.
876  *
877  * CONTEXT:
878  * Might sleep.
879  *
880  * RETURNS:
881  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
882  */
blkdev_get_by_path(const char * path,fmode_t mode,void * holder)883 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
884 					void *holder)
885 {
886 	struct block_device *bdev;
887 	dev_t dev;
888 	int error;
889 
890 	error = lookup_bdev(path, &dev);
891 	if (error)
892 		return ERR_PTR(error);
893 
894 	bdev = blkdev_get_by_dev(dev, mode, holder);
895 	if (!IS_ERR(bdev) && (mode & FMODE_WRITE) && bdev_read_only(bdev)) {
896 		blkdev_put(bdev, mode);
897 		return ERR_PTR(-EACCES);
898 	}
899 
900 	return bdev;
901 }
902 EXPORT_SYMBOL(blkdev_get_by_path);
903 
blkdev_put(struct block_device * bdev,fmode_t mode)904 void blkdev_put(struct block_device *bdev, fmode_t mode)
905 {
906 	struct gendisk *disk = bdev->bd_disk;
907 
908 	/*
909 	 * Sync early if it looks like we're the last one.  If someone else
910 	 * opens the block device between now and the decrement of bd_openers
911 	 * then we did a sync that we didn't need to, but that's not the end
912 	 * of the world and we want to avoid long (could be several minute)
913 	 * syncs while holding the mutex.
914 	 */
915 	if (atomic_read(&bdev->bd_openers) == 1)
916 		sync_blockdev(bdev);
917 
918 	mutex_lock(&disk->open_mutex);
919 	if (mode & FMODE_EXCL) {
920 		struct block_device *whole = bdev_whole(bdev);
921 		bool bdev_free;
922 
923 		/*
924 		 * Release a claim on the device.  The holder fields
925 		 * are protected with bdev_lock.  open_mutex is to
926 		 * synchronize disk_holder unlinking.
927 		 */
928 		spin_lock(&bdev_lock);
929 
930 		WARN_ON_ONCE(--bdev->bd_holders < 0);
931 		WARN_ON_ONCE(--whole->bd_holders < 0);
932 
933 		if ((bdev_free = !bdev->bd_holders))
934 			bdev->bd_holder = NULL;
935 		if (!whole->bd_holders)
936 			whole->bd_holder = NULL;
937 
938 		spin_unlock(&bdev_lock);
939 
940 		/*
941 		 * If this was the last claim, remove holder link and
942 		 * unblock evpoll if it was a write holder.
943 		 */
944 		if (bdev_free && bdev->bd_write_holder) {
945 			disk_unblock_events(disk);
946 			bdev->bd_write_holder = false;
947 		}
948 	}
949 
950 	/*
951 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
952 	 * event.  This is to ensure detection of media removal commanded
953 	 * from userland - e.g. eject(1).
954 	 */
955 	disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
956 
957 	if (bdev_is_partition(bdev))
958 		blkdev_put_part(bdev, mode);
959 	else
960 		blkdev_put_whole(bdev, mode);
961 	mutex_unlock(&disk->open_mutex);
962 
963 	module_put(disk->fops->owner);
964 	blkdev_put_no_open(bdev);
965 }
966 EXPORT_SYMBOL(blkdev_put);
967 
968 /**
969  * lookup_bdev() - Look up a struct block_device by name.
970  * @pathname: Name of the block device in the filesystem.
971  * @dev: Pointer to the block device's dev_t, if found.
972  *
973  * Lookup the block device's dev_t at @pathname in the current
974  * namespace if possible and return it in @dev.
975  *
976  * Context: May sleep.
977  * Return: 0 if succeeded, negative errno otherwise.
978  */
lookup_bdev(const char * pathname,dev_t * dev)979 int lookup_bdev(const char *pathname, dev_t *dev)
980 {
981 	struct inode *inode;
982 	struct path path;
983 	int error;
984 
985 	if (!pathname || !*pathname)
986 		return -EINVAL;
987 
988 	error = kern_path(pathname, LOOKUP_FOLLOW, &path);
989 	if (error)
990 		return error;
991 
992 	inode = d_backing_inode(path.dentry);
993 	error = -ENOTBLK;
994 	if (!S_ISBLK(inode->i_mode))
995 		goto out_path_put;
996 	error = -EACCES;
997 	if (!may_open_dev(&path))
998 		goto out_path_put;
999 
1000 	*dev = inode->i_rdev;
1001 	error = 0;
1002 out_path_put:
1003 	path_put(&path);
1004 	return error;
1005 }
1006 EXPORT_SYMBOL(lookup_bdev);
1007 
__invalidate_device(struct block_device * bdev,bool kill_dirty)1008 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1009 {
1010 	struct super_block *sb = get_super(bdev);
1011 	int res = 0;
1012 
1013 	if (sb) {
1014 		/*
1015 		 * no need to lock the super, get_super holds the
1016 		 * read mutex so the filesystem cannot go away
1017 		 * under us (->put_super runs with the write lock
1018 		 * hold).
1019 		 */
1020 		shrink_dcache_sb(sb);
1021 		res = invalidate_inodes(sb, kill_dirty);
1022 		drop_super(sb);
1023 	}
1024 	invalidate_bdev(bdev);
1025 	return res;
1026 }
1027 EXPORT_SYMBOL(__invalidate_device);
1028 
sync_bdevs(bool wait)1029 void sync_bdevs(bool wait)
1030 {
1031 	struct inode *inode, *old_inode = NULL;
1032 
1033 	spin_lock(&blockdev_superblock->s_inode_list_lock);
1034 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1035 		struct address_space *mapping = inode->i_mapping;
1036 		struct block_device *bdev;
1037 
1038 		spin_lock(&inode->i_lock);
1039 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1040 		    mapping->nrpages == 0) {
1041 			spin_unlock(&inode->i_lock);
1042 			continue;
1043 		}
1044 		__iget(inode);
1045 		spin_unlock(&inode->i_lock);
1046 		spin_unlock(&blockdev_superblock->s_inode_list_lock);
1047 		/*
1048 		 * We hold a reference to 'inode' so it couldn't have been
1049 		 * removed from s_inodes list while we dropped the
1050 		 * s_inode_list_lock  We cannot iput the inode now as we can
1051 		 * be holding the last reference and we cannot iput it under
1052 		 * s_inode_list_lock. So we keep the reference and iput it
1053 		 * later.
1054 		 */
1055 		iput(old_inode);
1056 		old_inode = inode;
1057 		bdev = I_BDEV(inode);
1058 
1059 		mutex_lock(&bdev->bd_disk->open_mutex);
1060 		if (!atomic_read(&bdev->bd_openers)) {
1061 			; /* skip */
1062 		} else if (wait) {
1063 			/*
1064 			 * We keep the error status of individual mapping so
1065 			 * that applications can catch the writeback error using
1066 			 * fsync(2). See filemap_fdatawait_keep_errors() for
1067 			 * details.
1068 			 */
1069 			filemap_fdatawait_keep_errors(inode->i_mapping);
1070 		} else {
1071 			filemap_fdatawrite(inode->i_mapping);
1072 		}
1073 		mutex_unlock(&bdev->bd_disk->open_mutex);
1074 
1075 		spin_lock(&blockdev_superblock->s_inode_list_lock);
1076 	}
1077 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
1078 	iput(old_inode);
1079 }
1080 
1081 /*
1082  * Handle STATX_DIOALIGN for block devices.
1083  *
1084  * Note that the inode passed to this is the inode of a block device node file,
1085  * not the block device's internal inode.  Therefore it is *not* valid to use
1086  * I_BDEV() here; the block device has to be looked up by i_rdev instead.
1087  */
bdev_statx_dioalign(struct inode * inode,struct kstat * stat)1088 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat)
1089 {
1090 	struct block_device *bdev;
1091 
1092 	bdev = blkdev_get_no_open(inode->i_rdev);
1093 	if (!bdev)
1094 		return;
1095 
1096 	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1097 	stat->dio_offset_align = bdev_logical_block_size(bdev);
1098 	stat->result_mask |= STATX_DIOALIGN;
1099 
1100 	blkdev_put_no_open(bdev);
1101 }
1102