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 /* Should we allow writing to mounted block devices? */
34 static bool bdev_allow_write_mounted = IS_ENABLED(CONFIG_BLK_DEV_WRITE_MOUNTED);
35
36 struct bdev_inode {
37 struct block_device bdev;
38 struct inode vfs_inode;
39 };
40
41 static inline struct bdev_inode *BDEV_I(struct inode *inode)
42 {
43 return container_of(inode, struct bdev_inode, vfs_inode);
44 }
45
46 static inline struct inode *BD_INODE(struct block_device *bdev)
47 {
48 return &container_of(bdev, struct bdev_inode, bdev)->vfs_inode;
49 }
50
51 struct block_device *I_BDEV(struct inode *inode)
52 {
53 return &BDEV_I(inode)->bdev;
54 }
55 EXPORT_SYMBOL(I_BDEV);
56
57 struct block_device *file_bdev(struct file *bdev_file)
58 {
59 return I_BDEV(bdev_file->f_mapping->host);
60 }
61 EXPORT_SYMBOL(file_bdev);
62
63 static void bdev_write_inode(struct block_device *bdev)
64 {
65 struct inode *inode = BD_INODE(bdev);
66 int ret;
67
68 spin_lock(&inode->i_lock);
69 while (inode->i_state & I_DIRTY) {
70 spin_unlock(&inode->i_lock);
71 ret = write_inode_now(inode, true);
72 if (ret)
73 pr_warn_ratelimited(
74 "VFS: Dirty inode writeback failed for block device %pg (err=%d).\n",
75 bdev, ret);
76 spin_lock(&inode->i_lock);
77 }
78 spin_unlock(&inode->i_lock);
79 }
80
81 /* Kill _all_ buffers and pagecache , dirty or not.. */
82 static void kill_bdev(struct block_device *bdev)
83 {
84 struct address_space *mapping = bdev->bd_mapping;
85
86 if (mapping_empty(mapping))
87 return;
88
89 invalidate_bh_lrus();
90 truncate_inode_pages(mapping, 0);
91 }
92
93 /* Invalidate clean unused buffers and pagecache. */
94 void invalidate_bdev(struct block_device *bdev)
95 {
96 struct address_space *mapping = bdev->bd_mapping;
97
98 if (mapping->nrpages) {
99 invalidate_bh_lrus();
100 lru_add_drain_all(); /* make sure all lru add caches are flushed */
101 invalidate_mapping_pages(mapping, 0, -1);
102 }
103 }
104 EXPORT_SYMBOL(invalidate_bdev);
105
106 /*
107 * Drop all buffers & page cache for given bdev range. This function bails
108 * with error if bdev has other exclusive owner (such as filesystem).
109 */
110 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
111 loff_t lstart, loff_t lend)
112 {
113 /*
114 * If we don't hold exclusive handle for the device, upgrade to it
115 * while we discard the buffer cache to avoid discarding buffers
116 * under live filesystem.
117 */
118 if (!(mode & BLK_OPEN_EXCL)) {
119 int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL);
120 if (err)
121 goto invalidate;
122 }
123
124 truncate_inode_pages_range(bdev->bd_mapping, lstart, lend);
125 if (!(mode & BLK_OPEN_EXCL))
126 bd_abort_claiming(bdev, truncate_bdev_range);
127 return 0;
128
129 invalidate:
130 /*
131 * Someone else has handle exclusively open. Try invalidating instead.
132 * The 'end' argument is inclusive so the rounding is safe.
133 */
134 return invalidate_inode_pages2_range(bdev->bd_mapping,
135 lstart >> PAGE_SHIFT,
136 lend >> PAGE_SHIFT);
137 }
138
139 static void set_init_blocksize(struct block_device *bdev)
140 {
141 unsigned int bsize = bdev_logical_block_size(bdev);
142 loff_t size = i_size_read(BD_INODE(bdev));
143
144 while (bsize < PAGE_SIZE) {
145 if (size & bsize)
146 break;
147 bsize <<= 1;
148 }
149 BD_INODE(bdev)->i_blkbits = blksize_bits(bsize);
150 }
151
152 int set_blocksize(struct file *file, int size)
153 {
154 struct inode *inode = file->f_mapping->host;
155 struct block_device *bdev = I_BDEV(inode);
156
157 /* Size must be a power of two, and between 512 and PAGE_SIZE */
158 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
159 return -EINVAL;
160
161 /* Size cannot be smaller than the size supported by the device */
162 if (size < bdev_logical_block_size(bdev))
163 return -EINVAL;
164
165 if (!file->private_data)
166 return -EINVAL;
167
168 /* Don't change the size if it is same as current */
169 if (inode->i_blkbits != blksize_bits(size)) {
170 sync_blockdev(bdev);
171 inode->i_blkbits = blksize_bits(size);
172 kill_bdev(bdev);
173 }
174 return 0;
175 }
176
177 EXPORT_SYMBOL(set_blocksize);
178
179 int sb_set_blocksize(struct super_block *sb, int size)
180 {
181 if (set_blocksize(sb->s_bdev_file, size))
182 return 0;
183 /* If we get here, we know size is power of two
184 * and it's value is between 512 and PAGE_SIZE */
185 sb->s_blocksize = size;
186 sb->s_blocksize_bits = blksize_bits(size);
187 return sb->s_blocksize;
188 }
189
190 EXPORT_SYMBOL(sb_set_blocksize);
191
192 int sb_min_blocksize(struct super_block *sb, int size)
193 {
194 int minsize = bdev_logical_block_size(sb->s_bdev);
195 if (size < minsize)
196 size = minsize;
197 return sb_set_blocksize(sb, size);
198 }
199
200 EXPORT_SYMBOL(sb_min_blocksize);
201
202 int sync_blockdev_nowait(struct block_device *bdev)
203 {
204 if (!bdev)
205 return 0;
206 return filemap_flush(bdev->bd_mapping);
207 }
208 EXPORT_SYMBOL_GPL(sync_blockdev_nowait);
209
210 /*
211 * Write out and wait upon all the dirty data associated with a block
212 * device via its mapping. Does not take the superblock lock.
213 */
214 int sync_blockdev(struct block_device *bdev)
215 {
216 if (!bdev)
217 return 0;
218 return filemap_write_and_wait(bdev->bd_mapping);
219 }
220 EXPORT_SYMBOL(sync_blockdev);
221
222 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend)
223 {
224 return filemap_write_and_wait_range(bdev->bd_mapping,
225 lstart, lend);
226 }
227 EXPORT_SYMBOL(sync_blockdev_range);
228
229 /**
230 * bdev_freeze - lock a filesystem and force it into a consistent state
231 * @bdev: blockdevice to lock
232 *
233 * If a superblock is found on this device, we take the s_umount semaphore
234 * on it to make sure nobody unmounts until the snapshot creation is done.
235 * The reference counter (bd_fsfreeze_count) guarantees that only the last
236 * unfreeze process can unfreeze the frozen filesystem actually when multiple
237 * freeze requests arrive simultaneously. It counts up in bdev_freeze() and
238 * count down in bdev_thaw(). When it becomes 0, thaw_bdev() will unfreeze
239 * actually.
240 *
241 * Return: On success zero is returned, negative error code on failure.
242 */
243 int bdev_freeze(struct block_device *bdev)
244 {
245 int error = 0;
246
247 mutex_lock(&bdev->bd_fsfreeze_mutex);
248
249 if (atomic_inc_return(&bdev->bd_fsfreeze_count) > 1) {
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
251 return 0;
252 }
253
254 mutex_lock(&bdev->bd_holder_lock);
255 if (bdev->bd_holder_ops && bdev->bd_holder_ops->freeze) {
256 error = bdev->bd_holder_ops->freeze(bdev);
257 lockdep_assert_not_held(&bdev->bd_holder_lock);
258 } else {
259 mutex_unlock(&bdev->bd_holder_lock);
260 error = sync_blockdev(bdev);
261 }
262
263 if (error)
264 atomic_dec(&bdev->bd_fsfreeze_count);
265
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267 return error;
268 }
269 EXPORT_SYMBOL(bdev_freeze);
270
271 /**
272 * bdev_thaw - unlock filesystem
273 * @bdev: blockdevice to unlock
274 *
275 * Unlocks the filesystem and marks it writeable again after bdev_freeze().
276 *
277 * Return: On success zero is returned, negative error code on failure.
278 */
279 int bdev_thaw(struct block_device *bdev)
280 {
281 int error = -EINVAL, nr_freeze;
282
283 mutex_lock(&bdev->bd_fsfreeze_mutex);
284
285 /*
286 * If this returns < 0 it means that @bd_fsfreeze_count was
287 * already 0 and no decrement was performed.
288 */
289 nr_freeze = atomic_dec_if_positive(&bdev->bd_fsfreeze_count);
290 if (nr_freeze < 0)
291 goto out;
292
293 error = 0;
294 if (nr_freeze > 0)
295 goto out;
296
297 mutex_lock(&bdev->bd_holder_lock);
298 if (bdev->bd_holder_ops && bdev->bd_holder_ops->thaw) {
299 error = bdev->bd_holder_ops->thaw(bdev);
300 lockdep_assert_not_held(&bdev->bd_holder_lock);
301 } else {
302 mutex_unlock(&bdev->bd_holder_lock);
303 }
304
305 if (error)
306 atomic_inc(&bdev->bd_fsfreeze_count);
307 out:
308 mutex_unlock(&bdev->bd_fsfreeze_mutex);
309 return error;
310 }
311 EXPORT_SYMBOL(bdev_thaw);
312
313 /*
314 * pseudo-fs
315 */
316
317 static __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
318 static struct kmem_cache *bdev_cachep __ro_after_init;
319
320 static struct inode *bdev_alloc_inode(struct super_block *sb)
321 {
322 struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL);
323
324 if (!ei)
325 return NULL;
326 memset(&ei->bdev, 0, sizeof(ei->bdev));
327 return &ei->vfs_inode;
328 }
329
330 static void bdev_free_inode(struct inode *inode)
331 {
332 struct block_device *bdev = I_BDEV(inode);
333
334 free_percpu(bdev->bd_stats);
335 kfree(bdev->bd_meta_info);
336
337 if (!bdev_is_partition(bdev)) {
338 if (bdev->bd_disk && bdev->bd_disk->bdi)
339 bdi_put(bdev->bd_disk->bdi);
340 kfree(bdev->bd_disk);
341 }
342
343 if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR)
344 blk_free_ext_minor(MINOR(bdev->bd_dev));
345
346 kmem_cache_free(bdev_cachep, BDEV_I(inode));
347 }
348
349 static void init_once(void *data)
350 {
351 struct bdev_inode *ei = data;
352
353 inode_init_once(&ei->vfs_inode);
354 }
355
356 static void bdev_evict_inode(struct inode *inode)
357 {
358 truncate_inode_pages_final(&inode->i_data);
359 invalidate_inode_buffers(inode); /* is it needed here? */
360 clear_inode(inode);
361 }
362
363 static const struct super_operations bdev_sops = {
364 .statfs = simple_statfs,
365 .alloc_inode = bdev_alloc_inode,
366 .free_inode = bdev_free_inode,
367 .drop_inode = generic_delete_inode,
368 .evict_inode = bdev_evict_inode,
369 };
370
371 static int bd_init_fs_context(struct fs_context *fc)
372 {
373 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
374 if (!ctx)
375 return -ENOMEM;
376 fc->s_iflags |= SB_I_CGROUPWB;
377 ctx->ops = &bdev_sops;
378 return 0;
379 }
380
381 static struct file_system_type bd_type = {
382 .name = "bdev",
383 .init_fs_context = bd_init_fs_context,
384 .kill_sb = kill_anon_super,
385 };
386
387 struct super_block *blockdev_superblock __ro_after_init;
388 static struct vfsmount *blockdev_mnt __ro_after_init;
389 EXPORT_SYMBOL_GPL(blockdev_superblock);
390
391 void __init bdev_cache_init(void)
392 {
393 int err;
394
395 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
396 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
397 SLAB_ACCOUNT|SLAB_PANIC),
398 init_once);
399 err = register_filesystem(&bd_type);
400 if (err)
401 panic("Cannot register bdev pseudo-fs");
402 blockdev_mnt = kern_mount(&bd_type);
403 if (IS_ERR(blockdev_mnt))
404 panic("Cannot create bdev pseudo-fs");
405 blockdev_superblock = blockdev_mnt->mnt_sb; /* For writeback */
406 }
407
408 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
409 {
410 struct block_device *bdev;
411 struct inode *inode;
412
413 inode = new_inode(blockdev_superblock);
414 if (!inode)
415 return NULL;
416 inode->i_mode = S_IFBLK;
417 inode->i_rdev = 0;
418 inode->i_data.a_ops = &def_blk_aops;
419 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
420
421 bdev = I_BDEV(inode);
422 mutex_init(&bdev->bd_fsfreeze_mutex);
423 spin_lock_init(&bdev->bd_size_lock);
424 mutex_init(&bdev->bd_holder_lock);
425 atomic_set(&bdev->__bd_flags, partno);
426 bdev->bd_mapping = &inode->i_data;
427 bdev->bd_queue = disk->queue;
428 if (partno && bdev_test_flag(disk->part0, BD_HAS_SUBMIT_BIO))
429 bdev_set_flag(bdev, BD_HAS_SUBMIT_BIO);
430 bdev->bd_stats = alloc_percpu(struct disk_stats);
431 if (!bdev->bd_stats) {
432 iput(inode);
433 return NULL;
434 }
435 bdev->bd_disk = disk;
436 return bdev;
437 }
438
439 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
440 {
441 spin_lock(&bdev->bd_size_lock);
442 i_size_write(BD_INODE(bdev), (loff_t)sectors << SECTOR_SHIFT);
443 bdev->bd_nr_sectors = sectors;
444 spin_unlock(&bdev->bd_size_lock);
445 }
446
447 void bdev_add(struct block_device *bdev, dev_t dev)
448 {
449 struct inode *inode = BD_INODE(bdev);
450 if (bdev_stable_writes(bdev))
451 mapping_set_stable_writes(bdev->bd_mapping);
452 bdev->bd_dev = dev;
453 inode->i_rdev = dev;
454 inode->i_ino = dev;
455 insert_inode_hash(inode);
456 }
457
458 void bdev_unhash(struct block_device *bdev)
459 {
460 remove_inode_hash(BD_INODE(bdev));
461 }
462
463 void bdev_drop(struct block_device *bdev)
464 {
465 iput(BD_INODE(bdev));
466 }
467
468 long nr_blockdev_pages(void)
469 {
470 struct inode *inode;
471 long ret = 0;
472
473 spin_lock(&blockdev_superblock->s_inode_list_lock);
474 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
475 ret += inode->i_mapping->nrpages;
476 spin_unlock(&blockdev_superblock->s_inode_list_lock);
477
478 return ret;
479 }
480
481 /**
482 * bd_may_claim - test whether a block device can be claimed
483 * @bdev: block device of interest
484 * @holder: holder trying to claim @bdev
485 * @hops: holder ops
486 *
487 * Test whether @bdev can be claimed by @holder.
488 *
489 * RETURNS:
490 * %true if @bdev can be claimed, %false otherwise.
491 */
492 static bool bd_may_claim(struct block_device *bdev, void *holder,
493 const struct blk_holder_ops *hops)
494 {
495 struct block_device *whole = bdev_whole(bdev);
496
497 lockdep_assert_held(&bdev_lock);
498
499 if (bdev->bd_holder) {
500 /*
501 * The same holder can always re-claim.
502 */
503 if (bdev->bd_holder == holder) {
504 if (WARN_ON_ONCE(bdev->bd_holder_ops != hops))
505 return false;
506 return true;
507 }
508 return false;
509 }
510
511 /*
512 * If the whole devices holder is set to bd_may_claim, a partition on
513 * the device is claimed, but not the whole device.
514 */
515 if (whole != bdev &&
516 whole->bd_holder && whole->bd_holder != bd_may_claim)
517 return false;
518 return true;
519 }
520
521 /**
522 * bd_prepare_to_claim - claim a block device
523 * @bdev: block device of interest
524 * @holder: holder trying to claim @bdev
525 * @hops: holder ops.
526 *
527 * Claim @bdev. This function fails if @bdev is already claimed by another
528 * holder and waits if another claiming is in progress. return, the caller
529 * has ownership of bd_claiming and bd_holder[s].
530 *
531 * RETURNS:
532 * 0 if @bdev can be claimed, -EBUSY otherwise.
533 */
534 int bd_prepare_to_claim(struct block_device *bdev, void *holder,
535 const struct blk_holder_ops *hops)
536 {
537 struct block_device *whole = bdev_whole(bdev);
538
539 if (WARN_ON_ONCE(!holder))
540 return -EINVAL;
541 retry:
542 mutex_lock(&bdev_lock);
543 /* if someone else claimed, fail */
544 if (!bd_may_claim(bdev, holder, hops)) {
545 mutex_unlock(&bdev_lock);
546 return -EBUSY;
547 }
548
549 /* if claiming is already in progress, wait for it to finish */
550 if (whole->bd_claiming) {
551 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
552 DEFINE_WAIT(wait);
553
554 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
555 mutex_unlock(&bdev_lock);
556 schedule();
557 finish_wait(wq, &wait);
558 goto retry;
559 }
560
561 /* yay, all mine */
562 whole->bd_claiming = holder;
563 mutex_unlock(&bdev_lock);
564 return 0;
565 }
566 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
567
568 static void bd_clear_claiming(struct block_device *whole, void *holder)
569 {
570 lockdep_assert_held(&bdev_lock);
571 /* tell others that we're done */
572 BUG_ON(whole->bd_claiming != holder);
573 whole->bd_claiming = NULL;
574 wake_up_bit(&whole->bd_claiming, 0);
575 }
576
577 /**
578 * bd_finish_claiming - finish claiming of a block device
579 * @bdev: block device of interest
580 * @holder: holder that has claimed @bdev
581 * @hops: block device holder operations
582 *
583 * Finish exclusive open of a block device. Mark the device as exlusively
584 * open by the holder and wake up all waiters for exclusive open to finish.
585 */
586 static void bd_finish_claiming(struct block_device *bdev, void *holder,
587 const struct blk_holder_ops *hops)
588 {
589 struct block_device *whole = bdev_whole(bdev);
590
591 mutex_lock(&bdev_lock);
592 BUG_ON(!bd_may_claim(bdev, holder, hops));
593 /*
594 * Note that for a whole device bd_holders will be incremented twice,
595 * and bd_holder will be set to bd_may_claim before being set to holder
596 */
597 whole->bd_holders++;
598 whole->bd_holder = bd_may_claim;
599 bdev->bd_holders++;
600 mutex_lock(&bdev->bd_holder_lock);
601 bdev->bd_holder = holder;
602 bdev->bd_holder_ops = hops;
603 mutex_unlock(&bdev->bd_holder_lock);
604 bd_clear_claiming(whole, holder);
605 mutex_unlock(&bdev_lock);
606 }
607
608 /**
609 * bd_abort_claiming - abort claiming of a block device
610 * @bdev: block device of interest
611 * @holder: holder that has claimed @bdev
612 *
613 * Abort claiming of a block device when the exclusive open failed. This can be
614 * also used when exclusive open is not actually desired and we just needed
615 * to block other exclusive openers for a while.
616 */
617 void bd_abort_claiming(struct block_device *bdev, void *holder)
618 {
619 mutex_lock(&bdev_lock);
620 bd_clear_claiming(bdev_whole(bdev), holder);
621 mutex_unlock(&bdev_lock);
622 }
623 EXPORT_SYMBOL(bd_abort_claiming);
624
625 static void bd_end_claim(struct block_device *bdev, void *holder)
626 {
627 struct block_device *whole = bdev_whole(bdev);
628 bool unblock = false;
629
630 /*
631 * Release a claim on the device. The holder fields are protected with
632 * bdev_lock. open_mutex is used to synchronize disk_holder unlinking.
633 */
634 mutex_lock(&bdev_lock);
635 WARN_ON_ONCE(bdev->bd_holder != holder);
636 WARN_ON_ONCE(--bdev->bd_holders < 0);
637 WARN_ON_ONCE(--whole->bd_holders < 0);
638 if (!bdev->bd_holders) {
639 mutex_lock(&bdev->bd_holder_lock);
640 bdev->bd_holder = NULL;
641 bdev->bd_holder_ops = NULL;
642 mutex_unlock(&bdev->bd_holder_lock);
643 if (bdev_test_flag(bdev, BD_WRITE_HOLDER))
644 unblock = true;
645 }
646 if (!whole->bd_holders)
647 whole->bd_holder = NULL;
648 mutex_unlock(&bdev_lock);
649
650 /*
651 * If this was the last claim, remove holder link and unblock evpoll if
652 * it was a write holder.
653 */
654 if (unblock) {
655 disk_unblock_events(bdev->bd_disk);
656 bdev_clear_flag(bdev, BD_WRITE_HOLDER);
657 }
658 }
659
660 static void blkdev_flush_mapping(struct block_device *bdev)
661 {
662 WARN_ON_ONCE(bdev->bd_holders);
663 sync_blockdev(bdev);
664 kill_bdev(bdev);
665 bdev_write_inode(bdev);
666 }
667
668 static void blkdev_put_whole(struct block_device *bdev)
669 {
670 if (atomic_dec_and_test(&bdev->bd_openers))
671 blkdev_flush_mapping(bdev);
672 if (bdev->bd_disk->fops->release)
673 bdev->bd_disk->fops->release(bdev->bd_disk);
674 }
675
676 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode)
677 {
678 struct gendisk *disk = bdev->bd_disk;
679 int ret;
680
681 if (disk->fops->open) {
682 ret = disk->fops->open(disk, mode);
683 if (ret) {
684 /* avoid ghost partitions on a removed medium */
685 if (ret == -ENOMEDIUM &&
686 test_bit(GD_NEED_PART_SCAN, &disk->state))
687 bdev_disk_changed(disk, true);
688 return ret;
689 }
690 }
691
692 if (!atomic_read(&bdev->bd_openers))
693 set_init_blocksize(bdev);
694 atomic_inc(&bdev->bd_openers);
695 if (test_bit(GD_NEED_PART_SCAN, &disk->state)) {
696 /*
697 * Only return scanning errors if we are called from contexts
698 * that explicitly want them, e.g. the BLKRRPART ioctl.
699 */
700 ret = bdev_disk_changed(disk, false);
701 if (ret && (mode & BLK_OPEN_STRICT_SCAN)) {
702 blkdev_put_whole(bdev);
703 return ret;
704 }
705 }
706 return 0;
707 }
708
709 static int blkdev_get_part(struct block_device *part, blk_mode_t mode)
710 {
711 struct gendisk *disk = part->bd_disk;
712 int ret;
713
714 ret = blkdev_get_whole(bdev_whole(part), mode);
715 if (ret)
716 return ret;
717
718 ret = -ENXIO;
719 if (!bdev_nr_sectors(part))
720 goto out_blkdev_put;
721
722 if (!atomic_read(&part->bd_openers)) {
723 disk->open_partitions++;
724 set_init_blocksize(part);
725 }
726 atomic_inc(&part->bd_openers);
727 return 0;
728
729 out_blkdev_put:
730 blkdev_put_whole(bdev_whole(part));
731 return ret;
732 }
733
734 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder)
735 {
736 int ret;
737
738 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
739 MAJOR(dev), MINOR(dev),
740 ((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) |
741 ((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0));
742 if (ret)
743 return ret;
744
745 /* Blocking writes requires exclusive opener */
746 if (mode & BLK_OPEN_RESTRICT_WRITES && !holder)
747 return -EINVAL;
748
749 /*
750 * We're using error pointers to indicate to ->release() when we
751 * failed to open that block device. Also this doesn't make sense.
752 */
753 if (WARN_ON_ONCE(IS_ERR(holder)))
754 return -EINVAL;
755
756 return 0;
757 }
758
759 static void blkdev_put_part(struct block_device *part)
760 {
761 struct block_device *whole = bdev_whole(part);
762
763 if (atomic_dec_and_test(&part->bd_openers)) {
764 blkdev_flush_mapping(part);
765 whole->bd_disk->open_partitions--;
766 }
767 blkdev_put_whole(whole);
768 }
769
770 struct block_device *blkdev_get_no_open(dev_t dev)
771 {
772 struct block_device *bdev;
773 struct inode *inode;
774
775 inode = ilookup(blockdev_superblock, dev);
776 if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) {
777 blk_request_module(dev);
778 inode = ilookup(blockdev_superblock, dev);
779 if (inode)
780 pr_warn_ratelimited(
781 "block device autoloading is deprecated and will be removed.\n");
782 }
783 if (!inode)
784 return NULL;
785
786 /* switch from the inode reference to a device mode one: */
787 bdev = &BDEV_I(inode)->bdev;
788 if (!kobject_get_unless_zero(&bdev->bd_device.kobj))
789 bdev = NULL;
790 iput(inode);
791 return bdev;
792 }
793
794 void blkdev_put_no_open(struct block_device *bdev)
795 {
796 put_device(&bdev->bd_device);
797 }
798
799 static bool bdev_writes_blocked(struct block_device *bdev)
800 {
801 return bdev->bd_writers < 0;
802 }
803
804 static void bdev_block_writes(struct block_device *bdev)
805 {
806 bdev->bd_writers--;
807 }
808
809 static void bdev_unblock_writes(struct block_device *bdev)
810 {
811 bdev->bd_writers++;
812 }
813
814 static bool bdev_may_open(struct block_device *bdev, blk_mode_t mode)
815 {
816 if (bdev_allow_write_mounted)
817 return true;
818 /* Writes blocked? */
819 if (mode & BLK_OPEN_WRITE && bdev_writes_blocked(bdev))
820 return false;
821 if (mode & BLK_OPEN_RESTRICT_WRITES && bdev->bd_writers > 0)
822 return false;
823 return true;
824 }
825
826 static void bdev_claim_write_access(struct block_device *bdev, blk_mode_t mode)
827 {
828 if (bdev_allow_write_mounted)
829 return;
830
831 /* Claim exclusive or shared write access. */
832 if (mode & BLK_OPEN_RESTRICT_WRITES)
833 bdev_block_writes(bdev);
834 else if (mode & BLK_OPEN_WRITE)
835 bdev->bd_writers++;
836 }
837
838 static inline bool bdev_unclaimed(const struct file *bdev_file)
839 {
840 return bdev_file->private_data == BDEV_I(bdev_file->f_mapping->host);
841 }
842
843 static void bdev_yield_write_access(struct file *bdev_file)
844 {
845 struct block_device *bdev;
846
847 if (bdev_allow_write_mounted)
848 return;
849
850 if (bdev_unclaimed(bdev_file))
851 return;
852
853 bdev = file_bdev(bdev_file);
854
855 if (bdev_file->f_mode & FMODE_WRITE_RESTRICTED)
856 bdev_unblock_writes(bdev);
857 else if (bdev_file->f_mode & FMODE_WRITE)
858 bdev->bd_writers--;
859 }
860
861 /**
862 * bdev_open - open a block device
863 * @bdev: block device to open
864 * @mode: open mode (BLK_OPEN_*)
865 * @holder: exclusive holder identifier
866 * @hops: holder operations
867 * @bdev_file: file for the block device
868 *
869 * Open the block device. If @holder is not %NULL, the block device is opened
870 * with exclusive access. Exclusive opens may nest for the same @holder.
871 *
872 * CONTEXT:
873 * Might sleep.
874 *
875 * RETURNS:
876 * zero on success, -errno on failure.
877 */
878 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
879 const struct blk_holder_ops *hops, struct file *bdev_file)
880 {
881 bool unblock_events = true;
882 struct gendisk *disk = bdev->bd_disk;
883 int ret;
884
885 if (holder) {
886 mode |= BLK_OPEN_EXCL;
887 ret = bd_prepare_to_claim(bdev, holder, hops);
888 if (ret)
889 return ret;
890 } else {
891 if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL))
892 return -EIO;
893 }
894
895 disk_block_events(disk);
896
897 mutex_lock(&disk->open_mutex);
898 ret = -ENXIO;
899 if (!disk_live(disk))
900 goto abort_claiming;
901 if (!try_module_get(disk->fops->owner))
902 goto abort_claiming;
903 ret = -EBUSY;
904 if (!bdev_may_open(bdev, mode))
905 goto put_module;
906 if (bdev_is_partition(bdev))
907 ret = blkdev_get_part(bdev, mode);
908 else
909 ret = blkdev_get_whole(bdev, mode);
910 if (ret)
911 goto put_module;
912 bdev_claim_write_access(bdev, mode);
913 if (holder) {
914 bd_finish_claiming(bdev, holder, hops);
915
916 /*
917 * Block event polling for write claims if requested. Any write
918 * holder makes the write_holder state stick until all are
919 * released. This is good enough and tracking individual
920 * writeable reference is too fragile given the way @mode is
921 * used in blkdev_get/put().
922 */
923 if ((mode & BLK_OPEN_WRITE) &&
924 !bdev_test_flag(bdev, BD_WRITE_HOLDER) &&
925 (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) {
926 bdev_set_flag(bdev, BD_WRITE_HOLDER);
927 unblock_events = false;
928 }
929 }
930 mutex_unlock(&disk->open_mutex);
931
932 if (unblock_events)
933 disk_unblock_events(disk);
934
935 bdev_file->f_flags |= O_LARGEFILE;
936 bdev_file->f_mode |= FMODE_CAN_ODIRECT;
937 if (bdev_nowait(bdev))
938 bdev_file->f_mode |= FMODE_NOWAIT;
939 if (mode & BLK_OPEN_RESTRICT_WRITES)
940 bdev_file->f_mode |= FMODE_WRITE_RESTRICTED;
941 bdev_file->f_mapping = bdev->bd_mapping;
942 bdev_file->f_wb_err = filemap_sample_wb_err(bdev_file->f_mapping);
943 bdev_file->private_data = holder;
944
945 return 0;
946 put_module:
947 module_put(disk->fops->owner);
948 abort_claiming:
949 if (holder)
950 bd_abort_claiming(bdev, holder);
951 mutex_unlock(&disk->open_mutex);
952 disk_unblock_events(disk);
953 return ret;
954 }
955
956 /*
957 * If BLK_OPEN_WRITE_IOCTL is set then this is a historical quirk
958 * associated with the floppy driver where it has allowed ioctls if the
959 * file was opened for writing, but does not allow reads or writes.
960 * Make sure that this quirk is reflected in @f_flags.
961 *
962 * It can also happen if a block device is opened as O_RDWR | O_WRONLY.
963 */
964 static unsigned blk_to_file_flags(blk_mode_t mode)
965 {
966 unsigned int flags = 0;
967
968 if ((mode & (BLK_OPEN_READ | BLK_OPEN_WRITE)) ==
969 (BLK_OPEN_READ | BLK_OPEN_WRITE))
970 flags |= O_RDWR;
971 else if (mode & BLK_OPEN_WRITE_IOCTL)
972 flags |= O_RDWR | O_WRONLY;
973 else if (mode & BLK_OPEN_WRITE)
974 flags |= O_WRONLY;
975 else if (mode & BLK_OPEN_READ)
976 flags |= O_RDONLY; /* homeopathic, because O_RDONLY is 0 */
977 else
978 WARN_ON_ONCE(true);
979
980 if (mode & BLK_OPEN_NDELAY)
981 flags |= O_NDELAY;
982
983 return flags;
984 }
985
986 struct file *bdev_file_open_by_dev(dev_t dev, blk_mode_t mode, void *holder,
987 const struct blk_holder_ops *hops)
988 {
989 struct file *bdev_file;
990 struct block_device *bdev;
991 unsigned int flags;
992 int ret;
993
994 ret = bdev_permission(dev, mode, holder);
995 if (ret)
996 return ERR_PTR(ret);
997
998 bdev = blkdev_get_no_open(dev);
999 if (!bdev)
1000 return ERR_PTR(-ENXIO);
1001
1002 flags = blk_to_file_flags(mode);
1003 bdev_file = alloc_file_pseudo_noaccount(BD_INODE(bdev),
1004 blockdev_mnt, "", flags | O_LARGEFILE, &def_blk_fops);
1005 if (IS_ERR(bdev_file)) {
1006 blkdev_put_no_open(bdev);
1007 return bdev_file;
1008 }
1009 ihold(BD_INODE(bdev));
1010
1011 ret = bdev_open(bdev, mode, holder, hops, bdev_file);
1012 if (ret) {
1013 /* We failed to open the block device. Let ->release() know. */
1014 bdev_file->private_data = ERR_PTR(ret);
1015 fput(bdev_file);
1016 return ERR_PTR(ret);
1017 }
1018 return bdev_file;
1019 }
1020 EXPORT_SYMBOL(bdev_file_open_by_dev);
1021
1022 struct file *bdev_file_open_by_path(const char *path, blk_mode_t mode,
1023 void *holder,
1024 const struct blk_holder_ops *hops)
1025 {
1026 struct file *file;
1027 dev_t dev;
1028 int error;
1029
1030 error = lookup_bdev(path, &dev);
1031 if (error)
1032 return ERR_PTR(error);
1033
1034 file = bdev_file_open_by_dev(dev, mode, holder, hops);
1035 if (!IS_ERR(file) && (mode & BLK_OPEN_WRITE)) {
1036 if (bdev_read_only(file_bdev(file))) {
1037 fput(file);
1038 file = ERR_PTR(-EACCES);
1039 }
1040 }
1041
1042 return file;
1043 }
1044 EXPORT_SYMBOL(bdev_file_open_by_path);
1045
1046 static inline void bd_yield_claim(struct file *bdev_file)
1047 {
1048 struct block_device *bdev = file_bdev(bdev_file);
1049 void *holder = bdev_file->private_data;
1050
1051 lockdep_assert_held(&bdev->bd_disk->open_mutex);
1052
1053 if (WARN_ON_ONCE(IS_ERR_OR_NULL(holder)))
1054 return;
1055
1056 if (!bdev_unclaimed(bdev_file))
1057 bd_end_claim(bdev, holder);
1058 }
1059
1060 void bdev_release(struct file *bdev_file)
1061 {
1062 struct block_device *bdev = file_bdev(bdev_file);
1063 void *holder = bdev_file->private_data;
1064 struct gendisk *disk = bdev->bd_disk;
1065
1066 /* We failed to open that block device. */
1067 if (IS_ERR(holder))
1068 goto put_no_open;
1069
1070 /*
1071 * Sync early if it looks like we're the last one. If someone else
1072 * opens the block device between now and the decrement of bd_openers
1073 * then we did a sync that we didn't need to, but that's not the end
1074 * of the world and we want to avoid long (could be several minute)
1075 * syncs while holding the mutex.
1076 */
1077 if (atomic_read(&bdev->bd_openers) == 1)
1078 sync_blockdev(bdev);
1079
1080 mutex_lock(&disk->open_mutex);
1081 bdev_yield_write_access(bdev_file);
1082
1083 if (holder)
1084 bd_yield_claim(bdev_file);
1085
1086 /*
1087 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1088 * event. This is to ensure detection of media removal commanded
1089 * from userland - e.g. eject(1).
1090 */
1091 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
1092
1093 if (bdev_is_partition(bdev))
1094 blkdev_put_part(bdev);
1095 else
1096 blkdev_put_whole(bdev);
1097 mutex_unlock(&disk->open_mutex);
1098
1099 module_put(disk->fops->owner);
1100 put_no_open:
1101 blkdev_put_no_open(bdev);
1102 }
1103
1104 /**
1105 * bdev_fput - yield claim to the block device and put the file
1106 * @bdev_file: open block device
1107 *
1108 * Yield claim on the block device and put the file. Ensure that the
1109 * block device can be reclaimed before the file is closed which is a
1110 * deferred operation.
1111 */
1112 void bdev_fput(struct file *bdev_file)
1113 {
1114 if (WARN_ON_ONCE(bdev_file->f_op != &def_blk_fops))
1115 return;
1116
1117 if (bdev_file->private_data) {
1118 struct block_device *bdev = file_bdev(bdev_file);
1119 struct gendisk *disk = bdev->bd_disk;
1120
1121 mutex_lock(&disk->open_mutex);
1122 bdev_yield_write_access(bdev_file);
1123 bd_yield_claim(bdev_file);
1124 /*
1125 * Tell release we already gave up our hold on the
1126 * device and if write restrictions are available that
1127 * we already gave up write access to the device.
1128 */
1129 bdev_file->private_data = BDEV_I(bdev_file->f_mapping->host);
1130 mutex_unlock(&disk->open_mutex);
1131 }
1132
1133 fput(bdev_file);
1134 }
1135 EXPORT_SYMBOL(bdev_fput);
1136
1137 /**
1138 * lookup_bdev() - Look up a struct block_device by name.
1139 * @pathname: Name of the block device in the filesystem.
1140 * @dev: Pointer to the block device's dev_t, if found.
1141 *
1142 * Lookup the block device's dev_t at @pathname in the current
1143 * namespace if possible and return it in @dev.
1144 *
1145 * Context: May sleep.
1146 * Return: 0 if succeeded, negative errno otherwise.
1147 */
1148 int lookup_bdev(const char *pathname, dev_t *dev)
1149 {
1150 struct inode *inode;
1151 struct path path;
1152 int error;
1153
1154 if (!pathname || !*pathname)
1155 return -EINVAL;
1156
1157 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1158 if (error)
1159 return error;
1160
1161 inode = d_backing_inode(path.dentry);
1162 error = -ENOTBLK;
1163 if (!S_ISBLK(inode->i_mode))
1164 goto out_path_put;
1165 error = -EACCES;
1166 if (!may_open_dev(&path))
1167 goto out_path_put;
1168
1169 *dev = inode->i_rdev;
1170 error = 0;
1171 out_path_put:
1172 path_put(&path);
1173 return error;
1174 }
1175 EXPORT_SYMBOL(lookup_bdev);
1176
1177 /**
1178 * bdev_mark_dead - mark a block device as dead
1179 * @bdev: block device to operate on
1180 * @surprise: indicate a surprise removal
1181 *
1182 * Tell the file system that this devices or media is dead. If @surprise is set
1183 * to %true the device or media is already gone, if not we are preparing for an
1184 * orderly removal.
1185 *
1186 * This calls into the file system, which then typicall syncs out all dirty data
1187 * and writes back inodes and then invalidates any cached data in the inodes on
1188 * the file system. In addition we also invalidate the block device mapping.
1189 */
1190 void bdev_mark_dead(struct block_device *bdev, bool surprise)
1191 {
1192 mutex_lock(&bdev->bd_holder_lock);
1193 if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead)
1194 bdev->bd_holder_ops->mark_dead(bdev, surprise);
1195 else {
1196 mutex_unlock(&bdev->bd_holder_lock);
1197 sync_blockdev(bdev);
1198 }
1199
1200 invalidate_bdev(bdev);
1201 }
1202 /*
1203 * New drivers should not use this directly. There are some drivers however
1204 * that needs this for historical reasons. For example, the DASD driver has
1205 * historically had a shutdown to offline mode that doesn't actually remove the
1206 * gendisk that otherwise looks a lot like a safe device removal.
1207 */
1208 EXPORT_SYMBOL_GPL(bdev_mark_dead);
1209
1210 void sync_bdevs(bool wait)
1211 {
1212 struct inode *inode, *old_inode = NULL;
1213
1214 spin_lock(&blockdev_superblock->s_inode_list_lock);
1215 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1216 struct address_space *mapping = inode->i_mapping;
1217 struct block_device *bdev;
1218
1219 spin_lock(&inode->i_lock);
1220 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1221 mapping->nrpages == 0) {
1222 spin_unlock(&inode->i_lock);
1223 continue;
1224 }
1225 __iget(inode);
1226 spin_unlock(&inode->i_lock);
1227 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1228 /*
1229 * We hold a reference to 'inode' so it couldn't have been
1230 * removed from s_inodes list while we dropped the
1231 * s_inode_list_lock We cannot iput the inode now as we can
1232 * be holding the last reference and we cannot iput it under
1233 * s_inode_list_lock. So we keep the reference and iput it
1234 * later.
1235 */
1236 iput(old_inode);
1237 old_inode = inode;
1238 bdev = I_BDEV(inode);
1239
1240 mutex_lock(&bdev->bd_disk->open_mutex);
1241 if (!atomic_read(&bdev->bd_openers)) {
1242 ; /* skip */
1243 } else if (wait) {
1244 /*
1245 * We keep the error status of individual mapping so
1246 * that applications can catch the writeback error using
1247 * fsync(2). See filemap_fdatawait_keep_errors() for
1248 * details.
1249 */
1250 filemap_fdatawait_keep_errors(inode->i_mapping);
1251 } else {
1252 filemap_fdatawrite(inode->i_mapping);
1253 }
1254 mutex_unlock(&bdev->bd_disk->open_mutex);
1255
1256 spin_lock(&blockdev_superblock->s_inode_list_lock);
1257 }
1258 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1259 iput(old_inode);
1260 }
1261
1262 /*
1263 * Handle STATX_{DIOALIGN, WRITE_ATOMIC} for block devices.
1264 */
1265 void bdev_statx(struct path *path, struct kstat *stat,
1266 u32 request_mask)
1267 {
1268 struct inode *backing_inode;
1269 struct block_device *bdev;
1270
1271 if (!(request_mask & (STATX_DIOALIGN | STATX_WRITE_ATOMIC)))
1272 return;
1273
1274 backing_inode = d_backing_inode(path->dentry);
1275
1276 /*
1277 * Note that backing_inode is the inode of a block device node file,
1278 * not the block device's internal inode. Therefore it is *not* valid
1279 * to use I_BDEV() here; the block device has to be looked up by i_rdev
1280 * instead.
1281 */
1282 bdev = blkdev_get_no_open(backing_inode->i_rdev);
1283 if (!bdev)
1284 return;
1285
1286 if (request_mask & STATX_DIOALIGN) {
1287 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
1288 stat->dio_offset_align = bdev_logical_block_size(bdev);
1289 stat->result_mask |= STATX_DIOALIGN;
1290 }
1291
1292 if (request_mask & STATX_WRITE_ATOMIC && bdev_can_atomic_write(bdev)) {
1293 struct request_queue *bd_queue = bdev->bd_queue;
1294
1295 generic_fill_statx_atomic_writes(stat,
1296 queue_atomic_write_unit_min_bytes(bd_queue),
1297 queue_atomic_write_unit_max_bytes(bd_queue));
1298 }
1299
1300 blkdev_put_no_open(bdev);
1301 }
1302
1303 bool disk_live(struct gendisk *disk)
1304 {
1305 return !inode_unhashed(BD_INODE(disk->part0));
1306 }
1307 EXPORT_SYMBOL_GPL(disk_live);
1308
1309 unsigned int block_size(struct block_device *bdev)
1310 {
1311 return 1 << BD_INODE(bdev)->i_blkbits;
1312 }
1313 EXPORT_SYMBOL_GPL(block_size);
1314
1315 static int __init setup_bdev_allow_write_mounted(char *str)
1316 {
1317 if (kstrtobool(str, &bdev_allow_write_mounted))
1318 pr_warn("Invalid option string for bdev_allow_write_mounted:"
1319 " '%s'\n", str);
1320 return 1;
1321 }
1322 __setup("bdev_allow_write_mounted=", setup_bdev_allow_write_mounted);
1323
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.