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/part_stat.h>
28 #include <linux/uaccess.h>
29 #include <linux/stat.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 
BDEV_I(struct inode * inode)38 static inline struct bdev_inode *BDEV_I(struct inode *inode)
39 {
40 	return container_of(inode, struct bdev_inode, vfs_inode);
41 }
42 
I_BDEV(struct inode * inode)43 struct block_device *I_BDEV(struct inode *inode)
44 {
45 	return &BDEV_I(inode)->bdev;
46 }
47 EXPORT_SYMBOL(I_BDEV);
48 
bdev_write_inode(struct block_device * bdev)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 			pr_warn_ratelimited(
60 	"VFS: Dirty inode writeback failed for block device %pg (err=%d).\n",
61 				bdev, ret);
62 		spin_lock(&inode->i_lock);
63 	}
64 	spin_unlock(&inode->i_lock);
65 }
66 
67 /* Kill _all_ buffers and pagecache , dirty or not.. */
kill_bdev(struct block_device * bdev)68 static void kill_bdev(struct block_device *bdev)
69 {
70 	struct address_space *mapping = bdev->bd_inode->i_mapping;
71 
72 	if (mapping_empty(mapping))
73 		return;
74 
75 	invalidate_bh_lrus();
76 	truncate_inode_pages(mapping, 0);
77 }
78 
79 /* Invalidate clean unused buffers and pagecache. */
invalidate_bdev(struct block_device * bdev)80 void invalidate_bdev(struct block_device *bdev)
81 {
82 	struct address_space *mapping = bdev->bd_inode->i_mapping;
83 
84 	if (mapping->nrpages) {
85 		invalidate_bh_lrus();
86 		lru_add_drain_all();	/* make sure all lru add caches are flushed */
87 		invalidate_mapping_pages(mapping, 0, -1);
88 	}
89 }
90 EXPORT_SYMBOL(invalidate_bdev);
91 
92 /*
93  * Drop all buffers & page cache for given bdev range. This function bails
94  * with error if bdev has other exclusive owner (such as filesystem).
95  */
truncate_bdev_range(struct block_device * bdev,blk_mode_t mode,loff_t lstart,loff_t lend)96 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
97 			loff_t lstart, loff_t lend)
98 {
99 	/*
100 	 * If we don't hold exclusive handle for the device, upgrade to it
101 	 * while we discard the buffer cache to avoid discarding buffers
102 	 * under live filesystem.
103 	 */
104 	if (!(mode & BLK_OPEN_EXCL)) {
105 		int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL);
106 		if (err)
107 			goto invalidate;
108 	}
109 
110 	truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
111 	if (!(mode & BLK_OPEN_EXCL))
112 		bd_abort_claiming(bdev, truncate_bdev_range);
113 	return 0;
114 
115 invalidate:
116 	/*
117 	 * Someone else has handle exclusively open. Try invalidating instead.
118 	 * The 'end' argument is inclusive so the rounding is safe.
119 	 */
120 	return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
121 					     lstart >> PAGE_SHIFT,
122 					     lend >> PAGE_SHIFT);
123 }
124 
set_init_blocksize(struct block_device * bdev)125 static void set_init_blocksize(struct block_device *bdev)
126 {
127 	unsigned int bsize = bdev_logical_block_size(bdev);
128 	loff_t size = i_size_read(bdev->bd_inode);
129 
130 	while (bsize < PAGE_SIZE) {
131 		if (size & bsize)
132 			break;
133 		bsize <<= 1;
134 	}
135 	bdev->bd_inode->i_blkbits = blksize_bits(bsize);
136 }
137 
set_blocksize(struct block_device * bdev,int size)138 int set_blocksize(struct block_device *bdev, int size)
139 {
140 	/* Size must be a power of two, and between 512 and PAGE_SIZE */
141 	if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
142 		return -EINVAL;
143 
144 	/* Size cannot be smaller than the size supported by the device */
145 	if (size < bdev_logical_block_size(bdev))
146 		return -EINVAL;
147 
148 	/* Don't change the size if it is same as current */
149 	if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
150 		sync_blockdev(bdev);
151 		bdev->bd_inode->i_blkbits = blksize_bits(size);
152 		kill_bdev(bdev);
153 	}
154 	return 0;
155 }
156 
157 EXPORT_SYMBOL(set_blocksize);
158 
sb_set_blocksize(struct super_block * sb,int size)159 int sb_set_blocksize(struct super_block *sb, int size)
160 {
161 	if (set_blocksize(sb->s_bdev, size))
162 		return 0;
163 	/* If we get here, we know size is power of two
164 	 * and it's value is between 512 and PAGE_SIZE */
165 	sb->s_blocksize = size;
166 	sb->s_blocksize_bits = blksize_bits(size);
167 	return sb->s_blocksize;
168 }
169 
170 EXPORT_SYMBOL(sb_set_blocksize);
171 
sb_min_blocksize(struct super_block * sb,int size)172 int sb_min_blocksize(struct super_block *sb, int size)
173 {
174 	int minsize = bdev_logical_block_size(sb->s_bdev);
175 	if (size < minsize)
176 		size = minsize;
177 	return sb_set_blocksize(sb, size);
178 }
179 
180 EXPORT_SYMBOL(sb_min_blocksize);
181 
sync_blockdev_nowait(struct block_device * bdev)182 int sync_blockdev_nowait(struct block_device *bdev)
183 {
184 	if (!bdev)
185 		return 0;
186 	return filemap_flush(bdev->bd_inode->i_mapping);
187 }
188 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
189 
190 /*
191  * Write out and wait upon all the dirty data associated with a block
192  * device via its mapping.  Does not take the superblock lock.
193  */
sync_blockdev(struct block_device * bdev)194 int sync_blockdev(struct block_device *bdev)
195 {
196 	if (!bdev)
197 		return 0;
198 	return filemap_write_and_wait(bdev->bd_inode->i_mapping);
199 }
200 EXPORT_SYMBOL(sync_blockdev);
201 
sync_blockdev_range(struct block_device * bdev,loff_t lstart,loff_t lend)202 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend)
203 {
204 	return filemap_write_and_wait_range(bdev->bd_inode->i_mapping,
205 			lstart, lend);
206 }
207 EXPORT_SYMBOL(sync_blockdev_range);
208 
209 /**
210  * freeze_bdev - lock a filesystem and force it into a consistent state
211  * @bdev:	blockdevice to lock
212  *
213  * If a superblock is found on this device, we take the s_umount semaphore
214  * on it to make sure nobody unmounts until the snapshot creation is done.
215  * The reference counter (bd_fsfreeze_count) guarantees that only the last
216  * unfreeze process can unfreeze the frozen filesystem actually when multiple
217  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
218  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
219  * actually.
220  */
freeze_bdev(struct block_device * bdev)221 int freeze_bdev(struct block_device *bdev)
222 {
223 	struct super_block *sb;
224 	int error = 0;
225 
226 	mutex_lock(&bdev->bd_fsfreeze_mutex);
227 	if (++bdev->bd_fsfreeze_count > 1)
228 		goto done;
229 
230 	sb = get_active_super(bdev);
231 	if (!sb)
232 		goto sync;
233 	if (sb->s_op->freeze_super)
234 		error = sb->s_op->freeze_super(sb, FREEZE_HOLDER_USERSPACE);
235 	else
236 		error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
237 	deactivate_super(sb);
238 
239 	if (error) {
240 		bdev->bd_fsfreeze_count--;
241 		goto done;
242 	}
243 	bdev->bd_fsfreeze_sb = sb;
244 
245 sync:
246 	sync_blockdev(bdev);
247 done:
248 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
249 	return error;
250 }
251 EXPORT_SYMBOL(freeze_bdev);
252 
253 /**
254  * thaw_bdev - unlock filesystem
255  * @bdev:	blockdevice to unlock
256  *
257  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
258  */
thaw_bdev(struct block_device * bdev)259 int thaw_bdev(struct block_device *bdev)
260 {
261 	struct super_block *sb;
262 	int error = -EINVAL;
263 
264 	mutex_lock(&bdev->bd_fsfreeze_mutex);
265 	if (!bdev->bd_fsfreeze_count)
266 		goto out;
267 
268 	error = 0;
269 	if (--bdev->bd_fsfreeze_count > 0)
270 		goto out;
271 
272 	sb = bdev->bd_fsfreeze_sb;
273 	if (!sb)
274 		goto out;
275 
276 	if (sb->s_op->thaw_super)
277 		error = sb->s_op->thaw_super(sb, FREEZE_HOLDER_USERSPACE);
278 	else
279 		error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
280 	if (error)
281 		bdev->bd_fsfreeze_count++;
282 	else
283 		bdev->bd_fsfreeze_sb = NULL;
284 out:
285 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
286 	return error;
287 }
288 EXPORT_SYMBOL(thaw_bdev);
289 
290 /*
291  * pseudo-fs
292  */
293 
294 static  __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
295 static struct kmem_cache * bdev_cachep __read_mostly;
296 
bdev_alloc_inode(struct super_block * sb)297 static struct inode *bdev_alloc_inode(struct super_block *sb)
298 {
299 	struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
300 
301 	if (!ei)
302 		return NULL;
303 	memset(&ei->bdev, 0, sizeof(ei->bdev));
304 	return &ei->vfs_inode;
305 }
306 
bdev_free_inode(struct inode * inode)307 static void bdev_free_inode(struct inode *inode)
308 {
309 	struct block_device *bdev = I_BDEV(inode);
310 
311 	free_percpu(bdev->bd_stats);
312 	kfree(bdev->bd_meta_info);
313 
314 	if (!bdev_is_partition(bdev)) {
315 		if (bdev->bd_disk && bdev->bd_disk->bdi)
316 			bdi_put(bdev->bd_disk->bdi);
317 		kfree(bdev->bd_disk);
318 	}
319 
320 	if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
321 		blk_free_ext_minor(MINOR(bdev->bd_dev));
322 
323 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
324 }
325 
init_once(void * data)326 static void init_once(void *data)
327 {
328 	struct bdev_inode *ei = data;
329 
330 	inode_init_once(&ei->vfs_inode);
331 }
332 
bdev_evict_inode(struct inode * inode)333 static void bdev_evict_inode(struct inode *inode)
334 {
335 	truncate_inode_pages_final(&inode->i_data);
336 	invalidate_inode_buffers(inode); /* is it needed here? */
337 	clear_inode(inode);
338 }
339 
340 static const struct super_operations bdev_sops = {
341 	.statfs = simple_statfs,
342 	.alloc_inode = bdev_alloc_inode,
343 	.free_inode = bdev_free_inode,
344 	.drop_inode = generic_delete_inode,
345 	.evict_inode = bdev_evict_inode,
346 };
347 
bd_init_fs_context(struct fs_context * fc)348 static int bd_init_fs_context(struct fs_context *fc)
349 {
350 	struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
351 	if (!ctx)
352 		return -ENOMEM;
353 	fc->s_iflags |= SB_I_CGROUPWB;
354 	ctx->ops = &bdev_sops;
355 	return 0;
356 }
357 
358 static struct file_system_type bd_type = {
359 	.name		= "bdev",
360 	.init_fs_context = bd_init_fs_context,
361 	.kill_sb	= kill_anon_super,
362 };
363 
364 struct super_block *blockdev_superblock __read_mostly;
365 EXPORT_SYMBOL_GPL(blockdev_superblock);
366 
bdev_cache_init(void)367 void __init bdev_cache_init(void)
368 {
369 	int err;
370 	static struct vfsmount *bd_mnt;
371 
372 	bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
373 			0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
374 				SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
375 			init_once);
376 	err = register_filesystem(&bd_type);
377 	if (err)
378 		panic("Cannot register bdev pseudo-fs");
379 	bd_mnt = kern_mount(&bd_type);
380 	if (IS_ERR(bd_mnt))
381 		panic("Cannot create bdev pseudo-fs");
382 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
383 }
384 
bdev_alloc(struct gendisk * disk,u8 partno)385 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
386 {
387 	struct block_device *bdev;
388 	struct inode *inode;
389 
390 	inode = new_inode(blockdev_superblock);
391 	if (!inode)
392 		return NULL;
393 	inode->i_mode = S_IFBLK;
394 	inode->i_rdev = 0;
395 	inode->i_data.a_ops = &def_blk_aops;
396 	mapping_set_gfp_mask(&inode->i_data, GFP_USER);
397 
398 	bdev = I_BDEV(inode);
399 	mutex_init(&bdev->bd_fsfreeze_mutex);
400 	spin_lock_init(&bdev->bd_size_lock);
401 	mutex_init(&bdev->bd_holder_lock);
402 	bdev->bd_partno = partno;
403 	bdev->bd_inode = inode;
404 	bdev->bd_queue = disk->queue;
405 	if (partno)
406 		bdev->bd_has_submit_bio = disk->part0->bd_has_submit_bio;
407 	else
408 		bdev->bd_has_submit_bio = false;
409 	bdev->bd_stats = alloc_percpu(struct disk_stats);
410 	if (!bdev->bd_stats) {
411 		iput(inode);
412 		return NULL;
413 	}
414 	bdev->bd_disk = disk;
415 	return bdev;
416 }
417 
bdev_set_nr_sectors(struct block_device * bdev,sector_t sectors)418 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
419 {
420 	spin_lock(&bdev->bd_size_lock);
421 	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
422 	bdev->bd_nr_sectors = sectors;
423 	spin_unlock(&bdev->bd_size_lock);
424 }
425 
bdev_add(struct block_device * bdev,dev_t dev)426 void bdev_add(struct block_device *bdev, dev_t dev)
427 {
428 	if (bdev_stable_writes(bdev))
429 		mapping_set_stable_writes(bdev->bd_inode->i_mapping);
430 	bdev->bd_dev = dev;
431 	bdev->bd_inode->i_rdev = dev;
432 	bdev->bd_inode->i_ino = dev;
433 	insert_inode_hash(bdev->bd_inode);
434 }
435 
nr_blockdev_pages(void)436 long nr_blockdev_pages(void)
437 {
438 	struct inode *inode;
439 	long ret = 0;
440 
441 	spin_lock(&blockdev_superblock->s_inode_list_lock);
442 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
443 		ret += inode->i_mapping->nrpages;
444 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
445 
446 	return ret;
447 }
448 
449 /**
450  * bd_may_claim - test whether a block device can be claimed
451  * @bdev: block device of interest
452  * @holder: holder trying to claim @bdev
453  * @hops: holder ops
454  *
455  * Test whether @bdev can be claimed by @holder.
456  *
457  * RETURNS:
458  * %true if @bdev can be claimed, %false otherwise.
459  */
bd_may_claim(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)460 static bool bd_may_claim(struct block_device *bdev, void *holder,
461 		const struct blk_holder_ops *hops)
462 {
463 	struct block_device *whole = bdev_whole(bdev);
464 
465 	lockdep_assert_held(&bdev_lock);
466 
467 	if (bdev->bd_holder) {
468 		/*
469 		 * The same holder can always re-claim.
470 		 */
471 		if (bdev->bd_holder == holder) {
472 			if (WARN_ON_ONCE(bdev->bd_holder_ops != hops))
473 				return false;
474 			return true;
475 		}
476 		return false;
477 	}
478 
479 	/*
480 	 * If the whole devices holder is set to bd_may_claim, a partition on
481 	 * the device is claimed, but not the whole device.
482 	 */
483 	if (whole != bdev &&
484 	    whole->bd_holder && whole->bd_holder != bd_may_claim)
485 		return false;
486 	return true;
487 }
488 
489 /**
490  * bd_prepare_to_claim - claim a block device
491  * @bdev: block device of interest
492  * @holder: holder trying to claim @bdev
493  * @hops: holder ops.
494  *
495  * Claim @bdev.  This function fails if @bdev is already claimed by another
496  * holder and waits if another claiming is in progress. return, the caller
497  * has ownership of bd_claiming and bd_holder[s].
498  *
499  * RETURNS:
500  * 0 if @bdev can be claimed, -EBUSY otherwise.
501  */
bd_prepare_to_claim(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)502 int bd_prepare_to_claim(struct block_device *bdev, void *holder,
503 		const struct blk_holder_ops *hops)
504 {
505 	struct block_device *whole = bdev_whole(bdev);
506 
507 	if (WARN_ON_ONCE(!holder))
508 		return -EINVAL;
509 retry:
510 	mutex_lock(&bdev_lock);
511 	/* if someone else claimed, fail */
512 	if (!bd_may_claim(bdev, holder, hops)) {
513 		mutex_unlock(&bdev_lock);
514 		return -EBUSY;
515 	}
516 
517 	/* if claiming is already in progress, wait for it to finish */
518 	if (whole->bd_claiming) {
519 		wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
520 		DEFINE_WAIT(wait);
521 
522 		prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
523 		mutex_unlock(&bdev_lock);
524 		schedule();
525 		finish_wait(wq, &wait);
526 		goto retry;
527 	}
528 
529 	/* yay, all mine */
530 	whole->bd_claiming = holder;
531 	mutex_unlock(&bdev_lock);
532 	return 0;
533 }
534 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
535 
bd_clear_claiming(struct block_device * whole,void * holder)536 static void bd_clear_claiming(struct block_device *whole, void *holder)
537 {
538 	lockdep_assert_held(&bdev_lock);
539 	/* tell others that we're done */
540 	BUG_ON(whole->bd_claiming != holder);
541 	whole->bd_claiming = NULL;
542 	wake_up_bit(&whole->bd_claiming, 0);
543 }
544 
545 /**
546  * bd_finish_claiming - finish claiming of a block device
547  * @bdev: block device of interest
548  * @holder: holder that has claimed @bdev
549  * @hops: block device holder operations
550  *
551  * Finish exclusive open of a block device. Mark the device as exlusively
552  * open by the holder and wake up all waiters for exclusive open to finish.
553  */
bd_finish_claiming(struct block_device * bdev,void * holder,const struct blk_holder_ops * hops)554 static void bd_finish_claiming(struct block_device *bdev, void *holder,
555 		const struct blk_holder_ops *hops)
556 {
557 	struct block_device *whole = bdev_whole(bdev);
558 
559 	mutex_lock(&bdev_lock);
560 	BUG_ON(!bd_may_claim(bdev, holder, hops));
561 	/*
562 	 * Note that for a whole device bd_holders will be incremented twice,
563 	 * and bd_holder will be set to bd_may_claim before being set to holder
564 	 */
565 	whole->bd_holders++;
566 	whole->bd_holder = bd_may_claim;
567 	bdev->bd_holders++;
568 	mutex_lock(&bdev->bd_holder_lock);
569 	bdev->bd_holder = holder;
570 	bdev->bd_holder_ops = hops;
571 	mutex_unlock(&bdev->bd_holder_lock);
572 	bd_clear_claiming(whole, holder);
573 	mutex_unlock(&bdev_lock);
574 }
575 
576 /**
577  * bd_abort_claiming - abort claiming of a block device
578  * @bdev: block device of interest
579  * @holder: holder that has claimed @bdev
580  *
581  * Abort claiming of a block device when the exclusive open failed. This can be
582  * also used when exclusive open is not actually desired and we just needed
583  * to block other exclusive openers for a while.
584  */
bd_abort_claiming(struct block_device * bdev,void * holder)585 void bd_abort_claiming(struct block_device *bdev, void *holder)
586 {
587 	mutex_lock(&bdev_lock);
588 	bd_clear_claiming(bdev_whole(bdev), holder);
589 	mutex_unlock(&bdev_lock);
590 }
591 EXPORT_SYMBOL(bd_abort_claiming);
592 
bd_end_claim(struct block_device * bdev,void * holder)593 static void bd_end_claim(struct block_device *bdev, void *holder)
594 {
595 	struct block_device *whole = bdev_whole(bdev);
596 	bool unblock = false;
597 
598 	/*
599 	 * Release a claim on the device.  The holder fields are protected with
600 	 * bdev_lock.  open_mutex is used to synchronize disk_holder unlinking.
601 	 */
602 	mutex_lock(&bdev_lock);
603 	WARN_ON_ONCE(bdev->bd_holder != holder);
604 	WARN_ON_ONCE(--bdev->bd_holders < 0);
605 	WARN_ON_ONCE(--whole->bd_holders < 0);
606 	if (!bdev->bd_holders) {
607 		mutex_lock(&bdev->bd_holder_lock);
608 		bdev->bd_holder = NULL;
609 		bdev->bd_holder_ops = NULL;
610 		mutex_unlock(&bdev->bd_holder_lock);
611 		if (bdev->bd_write_holder)
612 			unblock = true;
613 	}
614 	if (!whole->bd_holders)
615 		whole->bd_holder = NULL;
616 	mutex_unlock(&bdev_lock);
617 
618 	/*
619 	 * If this was the last claim, remove holder link and unblock evpoll if
620 	 * it was a write holder.
621 	 */
622 	if (unblock) {
623 		disk_unblock_events(bdev->bd_disk);
624 		bdev->bd_write_holder = false;
625 	}
626 }
627 
blkdev_flush_mapping(struct block_device * bdev)628 static void blkdev_flush_mapping(struct block_device *bdev)
629 {
630 	WARN_ON_ONCE(bdev->bd_holders);
631 	sync_blockdev(bdev);
632 	kill_bdev(bdev);
633 	bdev_write_inode(bdev);
634 }
635 
blkdev_get_whole(struct block_device * bdev,blk_mode_t mode)636 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode)
637 {
638 	struct gendisk *disk = bdev->bd_disk;
639 	int ret;
640 
641 	if (disk->fops->open) {
642 		ret = disk->fops->open(disk, mode);
643 		if (ret) {
644 			/* avoid ghost partitions on a removed medium */
645 			if (ret == -ENOMEDIUM &&
646 			     test_bit(GD_NEED_PART_SCAN, &disk->state))
647 				bdev_disk_changed(disk, true);
648 			return ret;
649 		}
650 	}
651 
652 	if (!atomic_read(&bdev->bd_openers))
653 		set_init_blocksize(bdev);
654 	if (test_bit(GD_NEED_PART_SCAN, &disk->state))
655 		bdev_disk_changed(disk, false);
656 	atomic_inc(&bdev->bd_openers);
657 	return 0;
658 }
659 
blkdev_put_whole(struct block_device * bdev)660 static void blkdev_put_whole(struct block_device *bdev)
661 {
662 	if (atomic_dec_and_test(&bdev->bd_openers))
663 		blkdev_flush_mapping(bdev);
664 	if (bdev->bd_disk->fops->release)
665 		bdev->bd_disk->fops->release(bdev->bd_disk);
666 }
667 
blkdev_get_part(struct block_device * part,blk_mode_t mode)668 static int blkdev_get_part(struct block_device *part, blk_mode_t mode)
669 {
670 	struct gendisk *disk = part->bd_disk;
671 	int ret;
672 
673 	ret = blkdev_get_whole(bdev_whole(part), mode);
674 	if (ret)
675 		return ret;
676 
677 	ret = -ENXIO;
678 	if (!bdev_nr_sectors(part))
679 		goto out_blkdev_put;
680 
681 	if (!atomic_read(&part->bd_openers)) {
682 		disk->open_partitions++;
683 		set_init_blocksize(part);
684 	}
685 	atomic_inc(&part->bd_openers);
686 	return 0;
687 
688 out_blkdev_put:
689 	blkdev_put_whole(bdev_whole(part));
690 	return ret;
691 }
692 
blkdev_put_part(struct block_device * part)693 static void blkdev_put_part(struct block_device *part)
694 {
695 	struct block_device *whole = bdev_whole(part);
696 
697 	if (atomic_dec_and_test(&part->bd_openers)) {
698 		blkdev_flush_mapping(part);
699 		whole->bd_disk->open_partitions--;
700 	}
701 	blkdev_put_whole(whole);
702 }
703 
blkdev_get_no_open(dev_t dev)704 struct block_device *blkdev_get_no_open(dev_t dev)
705 {
706 	struct block_device *bdev;
707 	struct inode *inode;
708 
709 	inode = ilookup(blockdev_superblock, dev);
710 	if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
711 		blk_request_module(dev);
712 		inode = ilookup(blockdev_superblock, dev);
713 		if (inode)
714 			pr_warn_ratelimited(
715 "block device autoloading is deprecated and will be removed.\n");
716 	}
717 	if (!inode)
718 		return NULL;
719 
720 	/* switch from the inode reference to a device mode one: */
721 	bdev = &BDEV_I(inode)->bdev;
722 	if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
723 		bdev = NULL;
724 	iput(inode);
725 	return bdev;
726 }
727 
blkdev_put_no_open(struct block_device * bdev)728 void blkdev_put_no_open(struct block_device *bdev)
729 {
730 	put_device(&bdev->bd_device);
731 }
732 
733 /**
734  * blkdev_get_by_dev - open a block device by device number
735  * @dev: device number of block device to open
736  * @mode: open mode (BLK_OPEN_*)
737  * @holder: exclusive holder identifier
738  * @hops: holder operations
739  *
740  * Open the block device described by device number @dev. If @holder is not
741  * %NULL, the block device is opened with exclusive access.  Exclusive opens may
742  * nest for the same @holder.
743  *
744  * Use this interface ONLY if you really do not have anything better - i.e. when
745  * you are behind a truly sucky interface and all you are given is a device
746  * number.  Everything else should use blkdev_get_by_path().
747  *
748  * CONTEXT:
749  * Might sleep.
750  *
751  * RETURNS:
752  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
753  */
blkdev_get_by_dev(dev_t dev,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)754 struct block_device *blkdev_get_by_dev(dev_t dev, blk_mode_t mode, void *holder,
755 		const struct blk_holder_ops *hops)
756 {
757 	bool unblock_events = true;
758 	struct block_device *bdev;
759 	struct gendisk *disk;
760 	int ret;
761 
762 	ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
763 			MAJOR(dev), MINOR(dev),
764 			((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) |
765 			((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0));
766 	if (ret)
767 		return ERR_PTR(ret);
768 
769 	bdev = blkdev_get_no_open(dev);
770 	if (!bdev)
771 		return ERR_PTR(-ENXIO);
772 	disk = bdev->bd_disk;
773 
774 	if (holder) {
775 		mode |= BLK_OPEN_EXCL;
776 		ret = bd_prepare_to_claim(bdev, holder, hops);
777 		if (ret)
778 			goto put_blkdev;
779 	} else {
780 		if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) {
781 			ret = -EIO;
782 			goto put_blkdev;
783 		}
784 	}
785 
786 	disk_block_events(disk);
787 
788 	mutex_lock(&disk->open_mutex);
789 	ret = -ENXIO;
790 	if (!disk_live(disk))
791 		goto abort_claiming;
792 	if (!try_module_get(disk->fops->owner))
793 		goto abort_claiming;
794 	if (bdev_is_partition(bdev))
795 		ret = blkdev_get_part(bdev, mode);
796 	else
797 		ret = blkdev_get_whole(bdev, mode);
798 	if (ret)
799 		goto put_module;
800 	if (holder) {
801 		bd_finish_claiming(bdev, holder, hops);
802 
803 		/*
804 		 * Block event polling for write claims if requested.  Any write
805 		 * holder makes the write_holder state stick until all are
806 		 * released.  This is good enough and tracking individual
807 		 * writeable reference is too fragile given the way @mode is
808 		 * used in blkdev_get/put().
809 		 */
810 		if ((mode & BLK_OPEN_WRITE) && !bdev->bd_write_holder &&
811 		    (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
812 			bdev->bd_write_holder = true;
813 			unblock_events = false;
814 		}
815 	}
816 	mutex_unlock(&disk->open_mutex);
817 
818 	if (unblock_events)
819 		disk_unblock_events(disk);
820 	return bdev;
821 put_module:
822 	module_put(disk->fops->owner);
823 abort_claiming:
824 	if (holder)
825 		bd_abort_claiming(bdev, holder);
826 	mutex_unlock(&disk->open_mutex);
827 	disk_unblock_events(disk);
828 put_blkdev:
829 	blkdev_put_no_open(bdev);
830 	return ERR_PTR(ret);
831 }
832 EXPORT_SYMBOL(blkdev_get_by_dev);
833 
834 /**
835  * blkdev_get_by_path - open a block device by name
836  * @path: path to the block device to open
837  * @mode: open mode (BLK_OPEN_*)
838  * @holder: exclusive holder identifier
839  * @hops: holder operations
840  *
841  * Open the block device described by the device file at @path.  If @holder is
842  * not %NULL, the block device is opened with exclusive access.  Exclusive opens
843  * may nest for the same @holder.
844  *
845  * CONTEXT:
846  * Might sleep.
847  *
848  * RETURNS:
849  * Reference to the block_device on success, ERR_PTR(-errno) on failure.
850  */
blkdev_get_by_path(const char * path,blk_mode_t mode,void * holder,const struct blk_holder_ops * hops)851 struct block_device *blkdev_get_by_path(const char *path, blk_mode_t mode,
852 		void *holder, const struct blk_holder_ops *hops)
853 {
854 	struct block_device *bdev;
855 	dev_t dev;
856 	int error;
857 
858 	error = lookup_bdev(path, &dev);
859 	if (error)
860 		return ERR_PTR(error);
861 
862 	bdev = blkdev_get_by_dev(dev, mode, holder, hops);
863 	if (!IS_ERR(bdev) && (mode & BLK_OPEN_WRITE) && bdev_read_only(bdev)) {
864 		blkdev_put(bdev, holder);
865 		return ERR_PTR(-EACCES);
866 	}
867 
868 	return bdev;
869 }
870 EXPORT_SYMBOL(blkdev_get_by_path);
871 
blkdev_put(struct block_device * bdev,void * holder)872 void blkdev_put(struct block_device *bdev, void *holder)
873 {
874 	struct gendisk *disk = bdev->bd_disk;
875 
876 	/*
877 	 * Sync early if it looks like we're the last one.  If someone else
878 	 * opens the block device between now and the decrement of bd_openers
879 	 * then we did a sync that we didn't need to, but that's not the end
880 	 * of the world and we want to avoid long (could be several minute)
881 	 * syncs while holding the mutex.
882 	 */
883 	if (atomic_read(&bdev->bd_openers) == 1)
884 		sync_blockdev(bdev);
885 
886 	mutex_lock(&disk->open_mutex);
887 	if (holder)
888 		bd_end_claim(bdev, holder);
889 
890 	/*
891 	 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
892 	 * event.  This is to ensure detection of media removal commanded
893 	 * from userland - e.g. eject(1).
894 	 */
895 	disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
896 
897 	if (bdev_is_partition(bdev))
898 		blkdev_put_part(bdev);
899 	else
900 		blkdev_put_whole(bdev);
901 	mutex_unlock(&disk->open_mutex);
902 
903 	module_put(disk->fops->owner);
904 	blkdev_put_no_open(bdev);
905 }
906 EXPORT_SYMBOL(blkdev_put);
907 
908 /**
909  * lookup_bdev() - Look up a struct block_device by name.
910  * @pathname: Name of the block device in the filesystem.
911  * @dev: Pointer to the block device's dev_t, if found.
912  *
913  * Lookup the block device's dev_t at @pathname in the current
914  * namespace if possible and return it in @dev.
915  *
916  * Context: May sleep.
917  * Return: 0 if succeeded, negative errno otherwise.
918  */
lookup_bdev(const char * pathname,dev_t * dev)919 int lookup_bdev(const char *pathname, dev_t *dev)
920 {
921 	struct inode *inode;
922 	struct path path;
923 	int error;
924 
925 	if (!pathname || !*pathname)
926 		return -EINVAL;
927 
928 	error = kern_path(pathname, LOOKUP_FOLLOW, &path);
929 	if (error)
930 		return error;
931 
932 	inode = d_backing_inode(path.dentry);
933 	error = -ENOTBLK;
934 	if (!S_ISBLK(inode->i_mode))
935 		goto out_path_put;
936 	error = -EACCES;
937 	if (!may_open_dev(&path))
938 		goto out_path_put;
939 
940 	*dev = inode->i_rdev;
941 	error = 0;
942 out_path_put:
943 	path_put(&path);
944 	return error;
945 }
946 EXPORT_SYMBOL(lookup_bdev);
947 
948 /**
949  * bdev_mark_dead - mark a block device as dead
950  * @bdev: block device to operate on
951  * @surprise: indicate a surprise removal
952  *
953  * Tell the file system that this devices or media is dead.  If @surprise is set
954  * to %true the device or media is already gone, if not we are preparing for an
955  * orderly removal.
956  *
957  * This calls into the file system, which then typicall syncs out all dirty data
958  * and writes back inodes and then invalidates any cached data in the inodes on
959  * the file system.  In addition we also invalidate the block device mapping.
960  */
bdev_mark_dead(struct block_device * bdev,bool surprise)961 void bdev_mark_dead(struct block_device *bdev, bool surprise)
962 {
963 	mutex_lock(&bdev->bd_holder_lock);
964 	if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead)
965 		bdev->bd_holder_ops->mark_dead(bdev, surprise);
966 	else
967 		sync_blockdev(bdev);
968 	mutex_unlock(&bdev->bd_holder_lock);
969 
970 	invalidate_bdev(bdev);
971 }
972 #ifdef CONFIG_DASD_MODULE
973 /*
974  * Drivers should not use this directly, but the DASD driver has historically
975  * had a shutdown to offline mode that doesn't actually remove the gendisk
976  * that otherwise looks a lot like a safe device removal.
977  */
978 EXPORT_SYMBOL_GPL(bdev_mark_dead);
979 #endif
980 
sync_bdevs(bool wait)981 void sync_bdevs(bool wait)
982 {
983 	struct inode *inode, *old_inode = NULL;
984 
985 	spin_lock(&blockdev_superblock->s_inode_list_lock);
986 	list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
987 		struct address_space *mapping = inode->i_mapping;
988 		struct block_device *bdev;
989 
990 		spin_lock(&inode->i_lock);
991 		if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
992 		    mapping->nrpages == 0) {
993 			spin_unlock(&inode->i_lock);
994 			continue;
995 		}
996 		__iget(inode);
997 		spin_unlock(&inode->i_lock);
998 		spin_unlock(&blockdev_superblock->s_inode_list_lock);
999 		/*
1000 		 * We hold a reference to 'inode' so it couldn't have been
1001 		 * removed from s_inodes list while we dropped the
1002 		 * s_inode_list_lock  We cannot iput the inode now as we can
1003 		 * be holding the last reference and we cannot iput it under
1004 		 * s_inode_list_lock. So we keep the reference and iput it
1005 		 * later.
1006 		 */
1007 		iput(old_inode);
1008 		old_inode = inode;
1009 		bdev = I_BDEV(inode);
1010 
1011 		mutex_lock(&bdev->bd_disk->open_mutex);
1012 		if (!atomic_read(&bdev->bd_openers)) {
1013 			; /* skip */
1014 		} else if (wait) {
1015 			/*
1016 			 * We keep the error status of individual mapping so
1017 			 * that applications can catch the writeback error using
1018 			 * fsync(2). See filemap_fdatawait_keep_errors() for
1019 			 * details.
1020 			 */
1021 			filemap_fdatawait_keep_errors(inode->i_mapping);
1022 		} else {
1023 			filemap_fdatawrite(inode->i_mapping);
1024 		}
1025 		mutex_unlock(&bdev->bd_disk->open_mutex);
1026 
1027 		spin_lock(&blockdev_superblock->s_inode_list_lock);
1028 	}
1029 	spin_unlock(&blockdev_superblock->s_inode_list_lock);
1030 	iput(old_inode);
1031 }
1032 
1033 /*
1034  * Handle STATX_DIOALIGN for block devices.
1035  *
1036  * Note that the inode passed to this is the inode of a block device node file,
1037  * not the block device's internal inode.  Therefore it is *not* valid to use
1038  * I_BDEV() here; the block device has to be looked up by i_rdev instead.
1039  */
bdev_statx_dioalign(struct inode * inode,struct kstat * stat)1040 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat)
1041 {
1042 	struct block_device *bdev;
1043 
1044 	bdev = blkdev_get_no_open(inode->i_rdev);
1045 	if (!bdev)
1046 		return;
1047 
1048 	stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1049 	stat->dio_offset_align = bdev_logical_block_size(bdev);
1050 	stat->result_mask |= STATX_DIOALIGN;
1051 
1052 	blkdev_put_no_open(bdev);
1053 }
1054