• 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,blk_mode_t mode,loff_t lstart,loff_t lend)101 int truncate_bdev_range(struct block_device *bdev, blk_mode_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 & BLK_OPEN_EXCL)) {
110 		int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL);
111 		if (err)
112 			goto invalidate;
113 	}
114 
115 	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
116 	if (!(mode & BLK_OPEN_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  * freeze_bdev - lock a filesystem and force it into a consistent state
216  * @bdev:	blockdevice to lock
217  *
218  * If a superblock is found on this device, we take the s_umount semaphore
219  * on it to make sure nobody unmounts until the snapshot creation is done.
220  * The reference counter (bd_fsfreeze_count) guarantees that only the last
221  * unfreeze process can unfreeze the frozen filesystem actually when multiple
222  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
223  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
224  * actually.
225  */
freeze_bdev(struct block_device * bdev)226 int freeze_bdev(struct block_device *bdev)
227 {
228 	struct super_block *sb;
229 	int error = 0;
230 
231 	mutex_lock(&bdev->bd_fsfreeze_mutex);
232 	if (++bdev->bd_fsfreeze_count > 1)
233 		goto done;
234 
235 	sb = get_active_super(bdev);
236 	if (!sb)
237 		goto sync;
238 	if (sb->s_op->freeze_super)
239 		error = sb->s_op->freeze_super(sb, FREEZE_HOLDER_USERSPACE);
240 	else
241 		error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
242 	deactivate_super(sb);
243 
244 	if (error) {
245 		bdev->bd_fsfreeze_count--;
246 		goto done;
247 	}
248 	bdev->bd_fsfreeze_sb = sb;
249 
250 sync:
251 	sync_blockdev(bdev);
252 done:
253 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
254 	return error;
255 }
256 EXPORT_SYMBOL(freeze_bdev);
257 
258 /**
259  * thaw_bdev - unlock filesystem
260  * @bdev:	blockdevice to unlock
261  *
262  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
263  */
thaw_bdev(struct block_device * bdev)264 int thaw_bdev(struct block_device *bdev)
265 {
266 	struct super_block *sb;
267 	int error = -EINVAL;
268 
269 	mutex_lock(&bdev->bd_fsfreeze_mutex);
270 	if (!bdev->bd_fsfreeze_count)
271 		goto out;
272 
273 	error = 0;
274 	if (--bdev->bd_fsfreeze_count > 0)
275 		goto out;
276 
277 	sb = bdev->bd_fsfreeze_sb;
278 	if (!sb)
279 		goto out;
280 
281 	if (sb->s_op->thaw_super)
282 		error = sb->s_op->thaw_super(sb, FREEZE_HOLDER_USERSPACE);
283 	else
284 		error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
285 	if (error)
286 		bdev->bd_fsfreeze_count++;
287 	else
288 		bdev->bd_fsfreeze_sb = NULL;
289 out:
290 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
291 	return error;
292 }
293 EXPORT_SYMBOL(thaw_bdev);
294 
295 /*
296  * pseudo-fs
297  */
298 
299 static  __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
300 static struct kmem_cache * bdev_cachep __read_mostly;
301 
bdev_alloc_inode(struct super_block * sb)302 static struct inode *bdev_alloc_inode(struct super_block *sb)
303 {
304 	struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
305 
306 	if (!ei)
307 		return NULL;
308 	memset(&ei->bdev, 0, sizeof(ei->bdev));
309 	return &ei->vfs_inode;
310 }
311 
bdev_free_inode(struct inode * inode)312 static void bdev_free_inode(struct inode *inode)
313 {
314 	struct block_device *bdev = I_BDEV(inode);
315 
316 	free_percpu(bdev->bd_stats);
317 	kfree(bdev->bd_meta_info);
318 
319 	if (!bdev_is_partition(bdev)) {
320 		if (bdev->bd_disk && bdev->bd_disk->bdi)
321 			bdi_put(bdev->bd_disk->bdi);
322 		kfree(bdev->bd_disk);
323 	}
324 
325 	if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
326 		blk_free_ext_minor(MINOR(bdev->bd_dev));
327 
328 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
329 }
330 
init_once(void * data)331 static void init_once(void *data)
332 {
333 	struct bdev_inode *ei = data;
334 
335 	inode_init_once(&ei->vfs_inode);
336 }
337 
bdev_evict_inode(struct inode * inode)338 static void bdev_evict_inode(struct inode *inode)
339 {
340 	truncate_inode_pages_final(&inode->i_data);
341 	invalidate_inode_buffers(inode); /* is it needed here? */
342 	clear_inode(inode);
343 }
344 
345 static const struct super_operations bdev_sops = {
346 	.statfs = simple_statfs,
347 	.alloc_inode = bdev_alloc_inode,
348 	.free_inode = bdev_free_inode,
349 	.drop_inode = generic_delete_inode,
350 	.evict_inode = bdev_evict_inode,
351 };
352 
bd_init_fs_context(struct fs_context * fc)353 static int bd_init_fs_context(struct fs_context *fc)
354 {
355 	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
356 	if (!ctx)
357 		return -ENOMEM;
358 	fc->s_iflags |= SB_I_CGROUPWB;
359 	ctx->ops = &bdev_sops;
360 	return 0;
361 }
362 
363 static struct file_system_type bd_type = {
364 	.name		= "bdev",
365 	.init_fs_context = bd_init_fs_context,
366 	.kill_sb	= kill_anon_super,
367 };
368 
369 struct super_block *blockdev_superblock __read_mostly;
370 EXPORT_SYMBOL_GPL(blockdev_superblock);
371 
bdev_cache_init(void)372 void __init bdev_cache_init(void)
373 {
374 	int err;
375 	static struct vfsmount *bd_mnt;
376 
377 	bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
378 			0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
379 				SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
380 			init_once);
381 	err = register_filesystem(&bd_type);
382 	if (err)
383 		panic("Cannot register bdev pseudo-fs");
384 	bd_mnt = kern_mount(&bd_type);
385 	if (IS_ERR(bd_mnt))
386 		panic("Cannot create bdev pseudo-fs");
387 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
388 }
389 
bdev_alloc(struct gendisk * disk,u8 partno)390 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
391 {
392 	struct block_device *bdev;
393 	struct inode *inode;
394 
395 	inode = new_inode(blockdev_superblock);
396 	if (!inode)
397 		return NULL;
398 	inode->i_mode = S_IFBLK;
399 	inode->i_rdev = 0;
400 	inode->i_data.a_ops = &def_blk_aops;
401 	mapping_set_gfp_mask(&inode->i_data, GFP_USER);
402 
403 	bdev = I_BDEV(inode);
404 	mutex_init(&bdev->bd_fsfreeze_mutex);
405 	spin_lock_init(&bdev->bd_size_lock);
406 	mutex_init(&bdev->bd_holder_lock);
407 	bdev->bd_partno = partno;
408 	bdev->bd_inode = inode;
409 	bdev->bd_queue = disk->queue;
410 	if (partno)
411 		bdev->bd_has_submit_bio = disk->part0->bd_has_submit_bio;
412 	else
413 		bdev->bd_has_submit_bio = false;
414 	bdev->bd_stats = alloc_percpu(struct disk_stats);
415 	if (!bdev->bd_stats) {
416 		iput(inode);
417 		return NULL;
418 	}
419 	bdev->bd_disk = disk;
420 	return bdev;
421 }
422 
bdev_set_nr_sectors(struct block_device * bdev,sector_t sectors)423 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
424 {
425 	spin_lock(&bdev->bd_size_lock);
426 	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
427 	bdev->bd_nr_sectors = sectors;
428 	spin_unlock(&bdev->bd_size_lock);
429 }
430 
bdev_add(struct block_device * bdev,dev_t dev)431 void bdev_add(struct block_device *bdev, dev_t dev)
432 {
433 	if (bdev_stable_writes(bdev))
434 		mapping_set_stable_writes(bdev->bd_inode->i_mapping);
435 	bdev->bd_dev = dev;
436 	bdev->bd_inode->i_rdev = dev;
437 	bdev->bd_inode->i_ino = dev;
438 	insert_inode_hash(bdev->bd_inode);
439 }
440 
nr_blockdev_pages(void)441 long nr_blockdev_pages(void)
442 {
443 	struct inode *inode;
444 	long ret = 0;
445 
446 	spin_lock(&blockdev_superblock->s_inode_list_lock);
447 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
448 		ret += inode->i_mapping->nrpages;
449 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
450 
451 	return ret;
452 }
453 
454 /**
455  * bd_may_claim - test whether a block device can be claimed
456  * @bdev: block device of interest
457  * @holder: holder trying to claim @bdev
458  * @hops: holder ops
459  *
460  * Test whether @bdev can be claimed by @holder.
461  *
462  * RETURNS:
463  * %true if @bdev can be claimed, %false otherwise.
464  */
bd_may_claim(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)465 static bool bd_may_claim(struct block_device *bdev, void *holder,
466 		const struct blk_holder_ops *hops)
467 {
468 	struct block_device *whole = bdev_whole(bdev);
469 
470 	lockdep_assert_held(&bdev_lock);
471 
472 	if (bdev->bd_holder) {
473 		/*
474 		 * The same holder can always re-claim.
475 		 */
476 		if (bdev->bd_holder == holder) {
477 			if (WARN_ON_ONCE(bdev->bd_holder_ops != hops))
478 				return false;
479 			return true;
480 		}
481 		return false;
482 	}
483 
484 	/*
485 	 * If the whole devices holder is set to bd_may_claim, a partition on
486 	 * the device is claimed, but not the whole device.
487 	 */
488 	if (whole != bdev &&
489 	    whole->bd_holder && whole->bd_holder != bd_may_claim)
490 		return false;
491 	return true;
492 }
493 
494 /**
495  * bd_prepare_to_claim - claim a block device
496  * @bdev: block device of interest
497  * @holder: holder trying to claim @bdev
498  * @hops: holder ops.
499  *
500  * Claim @bdev.  This function fails if @bdev is already claimed by another
501  * holder and waits if another claiming is in progress. return, the caller
502  * has ownership of bd_claiming and bd_holder[s].
503  *
504  * RETURNS:
505  * 0 if @bdev can be claimed, -EBUSY otherwise.
506  */
bd_prepare_to_claim(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)507 int bd_prepare_to_claim(struct block_device *bdev, void *holder,
508 		const struct blk_holder_ops *hops)
509 {
510 	struct block_device *whole = bdev_whole(bdev);
511 
512 	if (WARN_ON_ONCE(!holder))
513 		return -EINVAL;
514 retry:
515 	mutex_lock(&bdev_lock);
516 	/* if someone else claimed, fail */
517 	if (!bd_may_claim(bdev, holder, hops)) {
518 		mutex_unlock(&bdev_lock);
519 		return -EBUSY;
520 	}
521 
522 	/* if claiming is already in progress, wait for it to finish */
523 	if (whole->bd_claiming) {
524 		wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
525 		DEFINE_WAIT(wait);
526 
527 		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
528 		mutex_unlock(&bdev_lock);
529 		schedule();
530 		finish_wait(wq, &wait);
531 		goto retry;
532 	}
533 
534 	/* yay, all mine */
535 	whole->bd_claiming = holder;
536 	mutex_unlock(&bdev_lock);
537 	return 0;
538 }
539 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
540 
bd_clear_claiming(struct block_device * whole,void * holder)541 static void bd_clear_claiming(struct block_device *whole, void *holder)
542 {
543 	lockdep_assert_held(&bdev_lock);
544 	/* tell others that we're done */
545 	BUG_ON(whole->bd_claiming != holder);
546 	whole->bd_claiming = NULL;
547 	wake_up_bit(&whole->bd_claiming, 0);
548 }
549 
550 /**
551  * bd_finish_claiming - finish claiming of a block device
552  * @bdev: block device of interest
553  * @holder: holder that has claimed @bdev
554  * @hops: block device holder operations
555  *
556  * Finish exclusive open of a block device. Mark the device as exlusively
557  * open by the holder and wake up all waiters for exclusive open to finish.
558  */
bd_finish_claiming(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)559 static void bd_finish_claiming(struct block_device *bdev, void *holder,
560 		const struct blk_holder_ops *hops)
561 {
562 	struct block_device *whole = bdev_whole(bdev);
563 
564 	mutex_lock(&bdev_lock);
565 	BUG_ON(!bd_may_claim(bdev, holder, hops));
566 	/*
567 	 * Note that for a whole device bd_holders will be incremented twice,
568 	 * and bd_holder will be set to bd_may_claim before being set to holder
569 	 */
570 	whole->bd_holders++;
571 	whole->bd_holder = bd_may_claim;
572 	bdev->bd_holders++;
573 	mutex_lock(&bdev->bd_holder_lock);
574 	bdev->bd_holder = holder;
575 	bdev->bd_holder_ops = hops;
576 	mutex_unlock(&bdev->bd_holder_lock);
577 	bd_clear_claiming(whole, holder);
578 	mutex_unlock(&bdev_lock);
579 }
580 
581 /**
582  * bd_abort_claiming - abort claiming of a block device
583  * @bdev: block device of interest
584  * @holder: holder that has claimed @bdev
585  *
586  * Abort claiming of a block device when the exclusive open failed. This can be
587  * also used when exclusive open is not actually desired and we just needed
588  * to block other exclusive openers for a while.
589  */
bd_abort_claiming(struct block_device * bdev,void * holder)590 void bd_abort_claiming(struct block_device *bdev, void *holder)
591 {
592 	mutex_lock(&bdev_lock);
593 	bd_clear_claiming(bdev_whole(bdev), holder);
594 	mutex_unlock(&bdev_lock);
595 }
596 EXPORT_SYMBOL(bd_abort_claiming);
597 
bd_end_claim(struct block_device * bdev,void * holder)598 static void bd_end_claim(struct block_device *bdev, void *holder)
599 {
600 	struct block_device *whole = bdev_whole(bdev);
601 	bool unblock = false;
602 
603 	/*
604 	 * Release a claim on the device.  The holder fields are protected with
605 	 * bdev_lock.  open_mutex is used to synchronize disk_holder unlinking.
606 	 */
607 	mutex_lock(&bdev_lock);
608 	WARN_ON_ONCE(bdev->bd_holder != holder);
609 	WARN_ON_ONCE(--bdev->bd_holders < 0);
610 	WARN_ON_ONCE(--whole->bd_holders < 0);
611 	if (!bdev->bd_holders) {
612 		mutex_lock(&bdev->bd_holder_lock);
613 		bdev->bd_holder = NULL;
614 		bdev->bd_holder_ops = NULL;
615 		mutex_unlock(&bdev->bd_holder_lock);
616 		if (bdev->bd_write_holder)
617 			unblock = true;
618 	}
619 	if (!whole->bd_holders)
620 		whole->bd_holder = NULL;
621 	mutex_unlock(&bdev_lock);
622 
623 	/*
624 	 * If this was the last claim, remove holder link and unblock evpoll if
625 	 * it was a write holder.
626 	 */
627 	if (unblock) {
628 		disk_unblock_events(bdev->bd_disk);
629 		bdev->bd_write_holder = false;
630 	}
631 }
632 
blkdev_flush_mapping(struct block_device * bdev)633 static void blkdev_flush_mapping(struct block_device *bdev)
634 {
635 	WARN_ON_ONCE(bdev->bd_holders);
636 	sync_blockdev(bdev);
637 	kill_bdev(bdev);
638 	bdev_write_inode(bdev);
639 }
640 
blkdev_get_whole(struct block_device * bdev,blk_mode_t mode)641 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode)
642 {
643 	struct gendisk *disk = bdev->bd_disk;
644 	int ret;
645 
646 	if (disk->fops->open) {
647 		ret = disk->fops->open(disk, mode);
648 		if (ret) {
649 			/* avoid ghost partitions on a removed medium */
650 			if (ret == -ENOMEDIUM &&
651 			     test_bit(GD_NEED_PART_SCAN, &disk->state))
652 				bdev_disk_changed(disk, true);
653 			return ret;
654 		}
655 	}
656 
657 	if (!atomic_read(&bdev->bd_openers))
658 		set_init_blocksize(bdev);
659 	if (test_bit(GD_NEED_PART_SCAN, &disk->state))
660 		bdev_disk_changed(disk, false);
661 	atomic_inc(&bdev->bd_openers);
662 	return 0;
663 }
664 
blkdev_put_whole(struct block_device * bdev)665 static void blkdev_put_whole(struct block_device *bdev)
666 {
667 	if (atomic_dec_and_test(&bdev->bd_openers))
668 		blkdev_flush_mapping(bdev);
669 	if (bdev->bd_disk->fops->release)
670 		bdev->bd_disk->fops->release(bdev->bd_disk);
671 }
672 
blkdev_get_part(struct block_device * part,blk_mode_t mode)673 static int blkdev_get_part(struct block_device *part, blk_mode_t mode)
674 {
675 	struct gendisk *disk = part->bd_disk;
676 	int ret;
677 
678 	ret = blkdev_get_whole(bdev_whole(part), mode);
679 	if (ret)
680 		return ret;
681 
682 	ret = -ENXIO;
683 	if (!bdev_nr_sectors(part))
684 		goto out_blkdev_put;
685 
686 	if (!atomic_read(&part->bd_openers)) {
687 		disk->open_partitions++;
688 		set_init_blocksize(part);
689 	}
690 	atomic_inc(&part->bd_openers);
691 	return 0;
692 
693 out_blkdev_put:
694 	blkdev_put_whole(bdev_whole(part));
695 	return ret;
696 }
697 
blkdev_put_part(struct block_device * part)698 static void blkdev_put_part(struct block_device *part)
699 {
700 	struct block_device *whole = bdev_whole(part);
701 
702 	if (atomic_dec_and_test(&part->bd_openers)) {
703 		blkdev_flush_mapping(part);
704 		whole->bd_disk->open_partitions--;
705 	}
706 	blkdev_put_whole(whole);
707 }
708 
blkdev_get_no_open(dev_t dev)709 struct block_device *blkdev_get_no_open(dev_t dev)
710 {
711 	struct block_device *bdev;
712 	struct inode *inode;
713 
714 	inode = ilookup(blockdev_superblock, dev);
715 	if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
716 		blk_request_module(dev);
717 		inode = ilookup(blockdev_superblock, dev);
718 		if (inode)
719 			pr_warn_ratelimited(
720 "block device autoloading is deprecated and will be removed.\n");
721 	}
722 	if (!inode)
723 		return NULL;
724 
725 	/* switch from the inode reference to a device mode one: */
726 	bdev = &BDEV_I(inode)->bdev;
727 	if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
728 		bdev = NULL;
729 	iput(inode);
730 	return bdev;
731 }
732 
blkdev_put_no_open(struct block_device * bdev)733 void blkdev_put_no_open(struct block_device *bdev)
734 {
735 	put_device(&bdev->bd_device);
736 }
737 
738 /**
739  * blkdev_get_by_dev - open a block device by device number
740  * @dev: device number of block device to open
741  * @mode: open mode (BLK_OPEN_*)
742  * @holder: exclusive holder identifier
743  * @hops: holder operations
744  *
745  * Open the block device described by device number @dev. If @holder is not
746  * %NULL, the block device is opened with exclusive access.  Exclusive opens may
747  * nest for the same @holder.
748  *
749  * Use this interface ONLY if you really do not have anything better - i.e. when
750  * you are behind a truly sucky interface and all you are given is a device
751  * number.  Everything else should use blkdev_get_by_path().
752  *
753  * CONTEXT:
754  * Might sleep.
755  *
756  * RETURNS:
757  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
758  */
blkdev_get_by_dev(dev_t dev,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)759 struct block_device *blkdev_get_by_dev(dev_t dev, blk_mode_t mode, void *holder,
760 		const struct blk_holder_ops *hops)
761 {
762 	bool unblock_events = true;
763 	struct block_device *bdev;
764 	struct gendisk *disk;
765 	int ret;
766 
767 	ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
768 			MAJOR(dev), MINOR(dev),
769 			((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) |
770 			((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0));
771 	if (ret)
772 		return ERR_PTR(ret);
773 
774 	bdev = blkdev_get_no_open(dev);
775 	if (!bdev)
776 		return ERR_PTR(-ENXIO);
777 	disk = bdev->bd_disk;
778 
779 	if (holder) {
780 		mode |= BLK_OPEN_EXCL;
781 		ret = bd_prepare_to_claim(bdev, holder, hops);
782 		if (ret)
783 			goto put_blkdev;
784 	} else {
785 		if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) {
786 			ret = -EIO;
787 			goto put_blkdev;
788 		}
789 	}
790 
791 	disk_block_events(disk);
792 
793 	mutex_lock(&disk->open_mutex);
794 	ret = -ENXIO;
795 	if (!disk_live(disk))
796 		goto abort_claiming;
797 	if (!try_module_get(disk->fops->owner))
798 		goto abort_claiming;
799 	if (bdev_is_partition(bdev))
800 		ret = blkdev_get_part(bdev, mode);
801 	else
802 		ret = blkdev_get_whole(bdev, mode);
803 	if (ret)
804 		goto put_module;
805 	if (holder) {
806 		bd_finish_claiming(bdev, holder, hops);
807 
808 		/*
809 		 * Block event polling for write claims if requested.  Any write
810 		 * holder makes the write_holder state stick until all are
811 		 * released.  This is good enough and tracking individual
812 		 * writeable reference is too fragile given the way @mode is
813 		 * used in blkdev_get/put().
814 		 */
815 		if ((mode & BLK_OPEN_WRITE) && !bdev->bd_write_holder &&
816 		    (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
817 			bdev->bd_write_holder = true;
818 			unblock_events = false;
819 		}
820 	}
821 	mutex_unlock(&disk->open_mutex);
822 
823 	if (unblock_events)
824 		disk_unblock_events(disk);
825 	return bdev;
826 put_module:
827 	module_put(disk->fops->owner);
828 abort_claiming:
829 	if (holder)
830 		bd_abort_claiming(bdev, holder);
831 	mutex_unlock(&disk->open_mutex);
832 	disk_unblock_events(disk);
833 put_blkdev:
834 	blkdev_put_no_open(bdev);
835 	return ERR_PTR(ret);
836 }
837 EXPORT_SYMBOL(blkdev_get_by_dev);
838 
bdev_open_by_dev(dev_t dev,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)839 struct bdev_handle *bdev_open_by_dev(dev_t dev, blk_mode_t mode, void *holder,
840 				     const struct blk_holder_ops *hops)
841 {
842 	struct bdev_handle *handle = kmalloc(sizeof(*handle), GFP_KERNEL);
843 	struct block_device *bdev;
844 
845 	if (!handle)
846 		return ERR_PTR(-ENOMEM);
847 	bdev = blkdev_get_by_dev(dev, mode, holder, hops);
848 	if (IS_ERR(bdev)) {
849 		kfree(handle);
850 		return ERR_CAST(bdev);
851 	}
852 	handle->bdev = bdev;
853 	handle->holder = holder;
854 	return handle;
855 }
856 EXPORT_SYMBOL(bdev_open_by_dev);
857 
858 /**
859  * blkdev_get_by_path - open a block device by name
860  * @path: path to the block device to open
861  * @mode: open mode (BLK_OPEN_*)
862  * @holder: exclusive holder identifier
863  * @hops: holder operations
864  *
865  * Open the block device described by the device file at @path.  If @holder is
866  * not %NULL, the block device is opened with exclusive access.  Exclusive opens
867  * may nest for the same @holder.
868  *
869  * CONTEXT:
870  * Might sleep.
871  *
872  * RETURNS:
873  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
874  */
blkdev_get_by_path(const char * path,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)875 struct block_device *blkdev_get_by_path(const char *path, blk_mode_t mode,
876 		void *holder, const struct blk_holder_ops *hops)
877 {
878 	struct block_device *bdev;
879 	dev_t dev;
880 	int error;
881 
882 	error = lookup_bdev(path, &dev);
883 	if (error)
884 		return ERR_PTR(error);
885 
886 	bdev = blkdev_get_by_dev(dev, mode, holder, hops);
887 	if (!IS_ERR(bdev) && (mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
888 		blkdev_put(bdev, holder);
889 		return ERR_PTR(-EACCES);
890 	}
891 
892 	return bdev;
893 }
894 EXPORT_SYMBOL(blkdev_get_by_path);
895 
bdev_open_by_path(const char * path,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)896 struct bdev_handle *bdev_open_by_path(const char *path, blk_mode_t mode,
897 		void *holder, const struct blk_holder_ops *hops)
898 {
899 	struct bdev_handle *handle;
900 	dev_t dev;
901 	int error;
902 
903 	error = lookup_bdev(path, &dev);
904 	if (error)
905 		return ERR_PTR(error);
906 
907 	handle = bdev_open_by_dev(dev, mode, holder, hops);
908 	if (!IS_ERR(handle) && (mode & BLK_OPEN_WRITE) &&
909 	    bdev_read_only(handle->bdev)) {
910 		bdev_release(handle);
911 		return ERR_PTR(-EACCES);
912 	}
913 
914 	return handle;
915 }
916 EXPORT_SYMBOL(bdev_open_by_path);
917 
blkdev_put(struct block_device * bdev,void * holder)918 void blkdev_put(struct block_device *bdev, void *holder)
919 {
920 	struct gendisk *disk = bdev->bd_disk;
921 
922 	/*
923 	 * Sync early if it looks like we're the last one.  If someone else
924 	 * opens the block device between now and the decrement of bd_openers
925 	 * then we did a sync that we didn't need to, but that's not the end
926 	 * of the world and we want to avoid long (could be several minute)
927 	 * syncs while holding the mutex.
928 	 */
929 	if (atomic_read(&bdev->bd_openers) == 1)
930 		sync_blockdev(bdev);
931 
932 	mutex_lock(&disk->open_mutex);
933 	if (holder)
934 		bd_end_claim(bdev, holder);
935 
936 	/*
937 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
938 	 * event.  This is to ensure detection of media removal commanded
939 	 * from userland - e.g. eject(1).
940 	 */
941 	disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
942 
943 	if (bdev_is_partition(bdev))
944 		blkdev_put_part(bdev);
945 	else
946 		blkdev_put_whole(bdev);
947 	mutex_unlock(&disk->open_mutex);
948 
949 	module_put(disk->fops->owner);
950 	blkdev_put_no_open(bdev);
951 }
952 EXPORT_SYMBOL(blkdev_put);
953 
bdev_release(struct bdev_handle * handle)954 void bdev_release(struct bdev_handle *handle)
955 {
956 	blkdev_put(handle->bdev, handle->holder);
957 	kfree(handle);
958 }
959 EXPORT_SYMBOL(bdev_release);
960 
961 /**
962  * lookup_bdev() - Look up a struct block_device by name.
963  * @pathname: Name of the block device in the filesystem.
964  * @dev: Pointer to the block device's dev_t, if found.
965  *
966  * Lookup the block device's dev_t at @pathname in the current
967  * namespace if possible and return it in @dev.
968  *
969  * Context: May sleep.
970  * Return: 0 if succeeded, negative errno otherwise.
971  */
lookup_bdev(const char * pathname,dev_t * dev)972 int lookup_bdev(const char *pathname, dev_t *dev)
973 {
974 	struct inode *inode;
975 	struct path path;
976 	int error;
977 
978 	if (!pathname || !*pathname)
979 		return -EINVAL;
980 
981 	error = kern_path(pathname, LOOKUP_FOLLOW, &path);
982 	if (error)
983 		return error;
984 
985 	inode = d_backing_inode(path.dentry);
986 	error = -ENOTBLK;
987 	if (!S_ISBLK(inode->i_mode))
988 		goto out_path_put;
989 	error = -EACCES;
990 	if (!may_open_dev(&path))
991 		goto out_path_put;
992 
993 	*dev = inode->i_rdev;
994 	error = 0;
995 out_path_put:
996 	path_put(&path);
997 	return error;
998 }
999 EXPORT_SYMBOL(lookup_bdev);
1000 
1001 /**
1002  * bdev_mark_dead - mark a block device as dead
1003  * @bdev: block device to operate on
1004  * @surprise: indicate a surprise removal
1005  *
1006  * Tell the file system that this devices or media is dead.  If @surprise is set
1007  * to %true the device or media is already gone, if not we are preparing for an
1008  * orderly removal.
1009  *
1010  * This calls into the file system, which then typicall syncs out all dirty data
1011  * and writes back inodes and then invalidates any cached data in the inodes on
1012  * the file system.  In addition we also invalidate the block device mapping.
1013  */
bdev_mark_dead(struct block_device * bdev,bool surprise)1014 void bdev_mark_dead(struct block_device *bdev, bool surprise)
1015 {
1016 	mutex_lock(&bdev->bd_holder_lock);
1017 	if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead)
1018 		bdev->bd_holder_ops->mark_dead(bdev, surprise);
1019 	else
1020 		sync_blockdev(bdev);
1021 	mutex_unlock(&bdev->bd_holder_lock);
1022 
1023 	invalidate_bdev(bdev);
1024 }
1025 #ifdef CONFIG_DASD_MODULE
1026 /*
1027  * Drivers should not use this directly, but the DASD driver has historically
1028  * had a shutdown to offline mode that doesn't actually remove the gendisk
1029  * that otherwise looks a lot like a safe device removal.
1030  */
1031 EXPORT_SYMBOL_GPL(bdev_mark_dead);
1032 #endif
1033 
sync_bdevs(bool wait)1034 void sync_bdevs(bool wait)
1035 {
1036 	struct inode *inode, *old_inode = NULL;
1037 
1038 	spin_lock(&blockdev_superblock->s_inode_list_lock);
1039 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1040 		struct address_space *mapping = inode->i_mapping;
1041 		struct block_device *bdev;
1042 
1043 		spin_lock(&inode->i_lock);
1044 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1045 		    mapping->nrpages == 0) {
1046 			spin_unlock(&inode->i_lock);
1047 			continue;
1048 		}
1049 		__iget(inode);
1050 		spin_unlock(&inode->i_lock);
1051 		spin_unlock(&blockdev_superblock->s_inode_list_lock);
1052 		/*
1053 		 * We hold a reference to 'inode' so it couldn't have been
1054 		 * removed from s_inodes list while we dropped the
1055 		 * s_inode_list_lock  We cannot iput the inode now as we can
1056 		 * be holding the last reference and we cannot iput it under
1057 		 * s_inode_list_lock. So we keep the reference and iput it
1058 		 * later.
1059 		 */
1060 		iput(old_inode);
1061 		old_inode = inode;
1062 		bdev = I_BDEV(inode);
1063 
1064 		mutex_lock(&bdev->bd_disk->open_mutex);
1065 		if (!atomic_read(&bdev->bd_openers)) {
1066 			; /* skip */
1067 		} else if (wait) {
1068 			/*
1069 			 * We keep the error status of individual mapping so
1070 			 * that applications can catch the writeback error using
1071 			 * fsync(2). See filemap_fdatawait_keep_errors() for
1072 			 * details.
1073 			 */
1074 			filemap_fdatawait_keep_errors(inode->i_mapping);
1075 		} else {
1076 			filemap_fdatawrite(inode->i_mapping);
1077 		}
1078 		mutex_unlock(&bdev->bd_disk->open_mutex);
1079 
1080 		spin_lock(&blockdev_superblock->s_inode_list_lock);
1081 	}
1082 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
1083 	iput(old_inode);
1084 }
1085 
1086 /*
1087  * Handle STATX_DIOALIGN for block devices.
1088  *
1089  * Note that the inode passed to this is the inode of a block device node file,
1090  * not the block device's internal inode.  Therefore it is *not* valid to use
1091  * I_BDEV() here; the block device has to be looked up by i_rdev instead.
1092  */
bdev_statx_dioalign(struct inode * inode,struct kstat * stat)1093 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat)
1094 {
1095 	struct block_device *bdev;
1096 
1097 	bdev = blkdev_get_no_open(inode->i_rdev);
1098 	if (!bdev)
1099 		return;
1100 
1101 	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1102 	stat->dio_offset_align = bdev_logical_block_size(bdev);
1103 	stat->result_mask |= STATX_DIOALIGN;
1104 
1105 	blkdev_put_no_open(bdev);
1106 }
1107