projects/linux·blob v3.18-rc3 · refs

1757 lines · 44.0 KiB

1/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE
6 */
7
8#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
14#include <linux/device_cgroup.h>
15#include <linux/highmem.h>
16#include <linux/blkdev.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/pagevec.h>
23#include <linux/writeback.h>
24#include <linux/mpage.h>
25#include <linux/mount.h>
26#include <linux/uio.h>
27#include <linux/namei.h>
28#include <linux/log2.h>
29#include <linux/cleancache.h>
30#include <linux/aio.h>
31#include <asm/uaccess.h>
32#include "internal.h"
33
34struct bdev_inode {
35 struct block_device bdev;
36 struct inode vfs_inode;
37};
38
39static const struct address_space_operations def_blk_aops;
40
41static inline struct bdev_inode *BDEV_I(struct inode *inode)
42{
43 return container_of(inode, struct bdev_inode, vfs_inode);
44}
45
46inline struct block_device *I_BDEV(struct inode *inode)
47{
48 return &BDEV_I(inode)->bdev;
49}
50EXPORT_SYMBOL(I_BDEV);
51
52/*
53 * Move the inode from its current bdi to a new bdi. Make sure the inode
54 * is clean before moving so that it doesn't linger on the old bdi.
55 */
56static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
58{
59 while (true) {
60 spin_lock(&inode->i_lock);
61 if (!(inode->i_state & I_DIRTY)) {
62 inode->i_data.backing_dev_info = dst;
63 spin_unlock(&inode->i_lock);
64 return;
65 }
66 spin_unlock(&inode->i_lock);
67 WARN_ON_ONCE(write_inode_now(inode, true));
68 }
69}
70
71/* Kill _all_ buffers and pagecache , dirty or not.. */
72void kill_bdev(struct block_device *bdev)
73{
74 struct address_space *mapping = bdev->bd_inode->i_mapping;
75
76 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
77 return;
78
79 invalidate_bh_lrus();
80 truncate_inode_pages(mapping, 0);
81}
82EXPORT_SYMBOL(kill_bdev);
83
84/* Invalidate clean unused buffers and pagecache. */
85void invalidate_bdev(struct block_device *bdev)
86{
87 struct address_space *mapping = bdev->bd_inode->i_mapping;
88
89 if (mapping->nrpages == 0)
90 return;
91
92 invalidate_bh_lrus();
93 lru_add_drain_all(); /* make sure all lru add caches are flushed */
94 invalidate_mapping_pages(mapping, 0, -1);
95 /* 99% of the time, we don't need to flush the cleancache on the bdev.
96 * But, for the strange corners, lets be cautious
97 */
98 cleancache_invalidate_inode(mapping);
99}
100EXPORT_SYMBOL(invalidate_bdev);
101
102int set_blocksize(struct block_device *bdev, int size)
103{
104 /* Size must be a power of two, and between 512 and PAGE_SIZE */
105 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
106 return -EINVAL;
107
108 /* Size cannot be smaller than the size supported by the device */
109 if (size < bdev_logical_block_size(bdev))
110 return -EINVAL;
111
112 /* Don't change the size if it is same as current */
113 if (bdev->bd_block_size != size) {
114 sync_blockdev(bdev);
115 bdev->bd_block_size = size;
116 bdev->bd_inode->i_blkbits = blksize_bits(size);
117 kill_bdev(bdev);
118 }
119 return 0;
120}
121
122EXPORT_SYMBOL(set_blocksize);
123
124int sb_set_blocksize(struct super_block *sb, int size)
125{
126 if (set_blocksize(sb->s_bdev, size))
127 return 0;
128 /* If we get here, we know size is power of two
129 * and it's value is between 512 and PAGE_SIZE */
130 sb->s_blocksize = size;
131 sb->s_blocksize_bits = blksize_bits(size);
132 return sb->s_blocksize;
133}
134
135EXPORT_SYMBOL(sb_set_blocksize);
136
137int sb_min_blocksize(struct super_block *sb, int size)
138{
139 int minsize = bdev_logical_block_size(sb->s_bdev);
140 if (size < minsize)
141 size = minsize;
142 return sb_set_blocksize(sb, size);
143}
144
145EXPORT_SYMBOL(sb_min_blocksize);
146
147static int
148blkdev_get_block(struct inode *inode, sector_t iblock,
149 struct buffer_head *bh, int create)
150{
151 bh->b_bdev = I_BDEV(inode);
152 bh->b_blocknr = iblock;
153 set_buffer_mapped(bh);
154 return 0;
155}
156
157static ssize_t
158blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
159 loff_t offset)
160{
161 struct file *file = iocb->ki_filp;
162 struct inode *inode = file->f_mapping->host;
163
164 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
165 offset, blkdev_get_block,
166 NULL, NULL, 0);
167}
168
169int __sync_blockdev(struct block_device *bdev, int wait)
170{
171 if (!bdev)
172 return 0;
173 if (!wait)
174 return filemap_flush(bdev->bd_inode->i_mapping);
175 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
176}
177
178/*
179 * Write out and wait upon all the dirty data associated with a block
180 * device via its mapping. Does not take the superblock lock.
181 */
182int sync_blockdev(struct block_device *bdev)
183{
184 return __sync_blockdev(bdev, 1);
185}
186EXPORT_SYMBOL(sync_blockdev);
187
188/*
189 * Write out and wait upon all dirty data associated with this
190 * device. Filesystem data as well as the underlying block
191 * device. Takes the superblock lock.
192 */
193int fsync_bdev(struct block_device *bdev)
194{
195 struct super_block *sb = get_super(bdev);
196 if (sb) {
197 int res = sync_filesystem(sb);
198 drop_super(sb);
199 return res;
200 }
201 return sync_blockdev(bdev);
202}
203EXPORT_SYMBOL(fsync_bdev);
204
205/**
206 * freeze_bdev -- lock a filesystem and force it into a consistent state
207 * @bdev: blockdevice to lock
208 *
209 * If a superblock is found on this device, we take the s_umount semaphore
210 * on it to make sure nobody unmounts until the snapshot creation is done.
211 * The reference counter (bd_fsfreeze_count) guarantees that only the last
212 * unfreeze process can unfreeze the frozen filesystem actually when multiple
213 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
214 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
215 * actually.
216 */
217struct super_block *freeze_bdev(struct block_device *bdev)
218{
219 struct super_block *sb;
220 int error = 0;
221
222 mutex_lock(&bdev->bd_fsfreeze_mutex);
223 if (++bdev->bd_fsfreeze_count > 1) {
224 /*
225 * We don't even need to grab a reference - the first call
226 * to freeze_bdev grab an active reference and only the last
227 * thaw_bdev drops it.
228 */
229 sb = get_super(bdev);
230 drop_super(sb);
231 mutex_unlock(&bdev->bd_fsfreeze_mutex);
232 return sb;
233 }
234
235 sb = get_active_super(bdev);
236 if (!sb)
237 goto out;
238 error = freeze_super(sb);
239 if (error) {
240 deactivate_super(sb);
241 bdev->bd_fsfreeze_count--;
242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
243 return ERR_PTR(error);
244 }
245 deactivate_super(sb);
246 out:
247 sync_blockdev(bdev);
248 mutex_unlock(&bdev->bd_fsfreeze_mutex);
249 return sb; /* thaw_bdev releases s->s_umount */
250}
251EXPORT_SYMBOL(freeze_bdev);
252
253/**
254 * thaw_bdev -- unlock filesystem
255 * @bdev: blockdevice to unlock
256 * @sb: associated superblock
257 *
258 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
259 */
260int thaw_bdev(struct block_device *bdev, struct super_block *sb)
261{
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 if (!sb)
273 goto out;
274
275 error = thaw_super(sb);
276 if (error) {
277 bdev->bd_fsfreeze_count++;
278 mutex_unlock(&bdev->bd_fsfreeze_mutex);
279 return error;
280 }
281out:
282 mutex_unlock(&bdev->bd_fsfreeze_mutex);
283 return 0;
284}
285EXPORT_SYMBOL(thaw_bdev);
286
287static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
288{
289 return block_write_full_page(page, blkdev_get_block, wbc);
290}
291
292static int blkdev_readpage(struct file * file, struct page * page)
293{
294 return block_read_full_page(page, blkdev_get_block);
295}
296
297static int blkdev_readpages(struct file *file, struct address_space *mapping,
298 struct list_head *pages, unsigned nr_pages)
299{
300 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
301}
302
303static int blkdev_write_begin(struct file *file, struct address_space *mapping,
304 loff_t pos, unsigned len, unsigned flags,
305 struct page **pagep, void **fsdata)
306{
307 return block_write_begin(mapping, pos, len, flags, pagep,
308 blkdev_get_block);
309}
310
311static int blkdev_write_end(struct file *file, struct address_space *mapping,
312 loff_t pos, unsigned len, unsigned copied,
313 struct page *page, void *fsdata)
314{
315 int ret;
316 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
317
318 unlock_page(page);
319 page_cache_release(page);
320
321 return ret;
322}
323
324/*
325 * private llseek:
326 * for a block special file file_inode(file)->i_size is zero
327 * so we compute the size by hand (just as in block_read/write above)
328 */
329static loff_t block_llseek(struct file *file, loff_t offset, int whence)
330{
331 struct inode *bd_inode = file->f_mapping->host;
332 loff_t retval;
333
334 mutex_lock(&bd_inode->i_mutex);
335 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
336 mutex_unlock(&bd_inode->i_mutex);
337 return retval;
338}
339
340int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
341{
342 struct inode *bd_inode = filp->f_mapping->host;
343 struct block_device *bdev = I_BDEV(bd_inode);
344 int error;
345
346 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
347 if (error)
348 return error;
349
350 /*
351 * There is no need to serialise calls to blkdev_issue_flush with
352 * i_mutex and doing so causes performance issues with concurrent
353 * O_SYNC writers to a block device.
354 */
355 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
356 if (error == -EOPNOTSUPP)
357 error = 0;
358
359 return error;
360}
361EXPORT_SYMBOL(blkdev_fsync);
362
363/**
364 * bdev_read_page() - Start reading a page from a block device
365 * @bdev: The device to read the page from
366 * @sector: The offset on the device to read the page to (need not be aligned)
367 * @page: The page to read
368 *
369 * On entry, the page should be locked. It will be unlocked when the page
370 * has been read. If the block driver implements rw_page synchronously,
371 * that will be true on exit from this function, but it need not be.
372 *
373 * Errors returned by this function are usually "soft", eg out of memory, or
374 * queue full; callers should try a different route to read this page rather
375 * than propagate an error back up the stack.
376 *
377 * Return: negative errno if an error occurs, 0 if submission was successful.
378 */
379int bdev_read_page(struct block_device *bdev, sector_t sector,
380 struct page *page)
381{
382 const struct block_device_operations *ops = bdev->bd_disk->fops;
383 if (!ops->rw_page)
384 return -EOPNOTSUPP;
385 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
386}
387EXPORT_SYMBOL_GPL(bdev_read_page);
388
389/**
390 * bdev_write_page() - Start writing a page to a block device
391 * @bdev: The device to write the page to
392 * @sector: The offset on the device to write the page to (need not be aligned)
393 * @page: The page to write
394 * @wbc: The writeback_control for the write
395 *
396 * On entry, the page should be locked and not currently under writeback.
397 * On exit, if the write started successfully, the page will be unlocked and
398 * under writeback. If the write failed already (eg the driver failed to
399 * queue the page to the device), the page will still be locked. If the
400 * caller is a ->writepage implementation, it will need to unlock the page.
401 *
402 * Errors returned by this function are usually "soft", eg out of memory, or
403 * queue full; callers should try a different route to write this page rather
404 * than propagate an error back up the stack.
405 *
406 * Return: negative errno if an error occurs, 0 if submission was successful.
407 */
408int bdev_write_page(struct block_device *bdev, sector_t sector,
409 struct page *page, struct writeback_control *wbc)
410{
411 int result;
412 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
413 const struct block_device_operations *ops = bdev->bd_disk->fops;
414 if (!ops->rw_page)
415 return -EOPNOTSUPP;
416 set_page_writeback(page);
417 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
418 if (result)
419 end_page_writeback(page);
420 else
421 unlock_page(page);
422 return result;
423}
424EXPORT_SYMBOL_GPL(bdev_write_page);
425
426/*
427 * pseudo-fs
428 */
429
430static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
431static struct kmem_cache * bdev_cachep __read_mostly;
432
433static struct inode *bdev_alloc_inode(struct super_block *sb)
434{
435 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
436 if (!ei)
437 return NULL;
438 return &ei->vfs_inode;
439}
440
441static void bdev_i_callback(struct rcu_head *head)
442{
443 struct inode *inode = container_of(head, struct inode, i_rcu);
444 struct bdev_inode *bdi = BDEV_I(inode);
445
446 kmem_cache_free(bdev_cachep, bdi);
447}
448
449static void bdev_destroy_inode(struct inode *inode)
450{
451 call_rcu(&inode->i_rcu, bdev_i_callback);
452}
453
454static void init_once(void *foo)
455{
456 struct bdev_inode *ei = (struct bdev_inode *) foo;
457 struct block_device *bdev = &ei->bdev;
458
459 memset(bdev, 0, sizeof(*bdev));
460 mutex_init(&bdev->bd_mutex);
461 INIT_LIST_HEAD(&bdev->bd_inodes);
462 INIT_LIST_HEAD(&bdev->bd_list);
463#ifdef CONFIG_SYSFS
464 INIT_LIST_HEAD(&bdev->bd_holder_disks);
465#endif
466 inode_init_once(&ei->vfs_inode);
467 /* Initialize mutex for freeze. */
468 mutex_init(&bdev->bd_fsfreeze_mutex);
469}
470
471static inline void __bd_forget(struct inode *inode)
472{
473 list_del_init(&inode->i_devices);
474 inode->i_bdev = NULL;
475 inode->i_mapping = &inode->i_data;
476}
477
478static void bdev_evict_inode(struct inode *inode)
479{
480 struct block_device *bdev = &BDEV_I(inode)->bdev;
481 struct list_head *p;
482 truncate_inode_pages_final(&inode->i_data);
483 invalidate_inode_buffers(inode); /* is it needed here? */
484 clear_inode(inode);
485 spin_lock(&bdev_lock);
486 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
487 __bd_forget(list_entry(p, struct inode, i_devices));
488 }
489 list_del_init(&bdev->bd_list);
490 spin_unlock(&bdev_lock);
491}
492
493static const struct super_operations bdev_sops = {
494 .statfs = simple_statfs,
495 .alloc_inode = bdev_alloc_inode,
496 .destroy_inode = bdev_destroy_inode,
497 .drop_inode = generic_delete_inode,
498 .evict_inode = bdev_evict_inode,
499};
500
501static struct dentry *bd_mount(struct file_system_type *fs_type,
502 int flags, const char *dev_name, void *data)
503{
504 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
505}
506
507static struct file_system_type bd_type = {
508 .name = "bdev",
509 .mount = bd_mount,
510 .kill_sb = kill_anon_super,
511};
512
513static struct super_block *blockdev_superblock __read_mostly;
514
515void __init bdev_cache_init(void)
516{
517 int err;
518 static struct vfsmount *bd_mnt;
519
520 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
521 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
522 SLAB_MEM_SPREAD|SLAB_PANIC),
523 init_once);
524 err = register_filesystem(&bd_type);
525 if (err)
526 panic("Cannot register bdev pseudo-fs");
527 bd_mnt = kern_mount(&bd_type);
528 if (IS_ERR(bd_mnt))
529 panic("Cannot create bdev pseudo-fs");
530 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
531}
532
533/*
534 * Most likely _very_ bad one - but then it's hardly critical for small
535 * /dev and can be fixed when somebody will need really large one.
536 * Keep in mind that it will be fed through icache hash function too.
537 */
538static inline unsigned long hash(dev_t dev)
539{
540 return MAJOR(dev)+MINOR(dev);
541}
542
543static int bdev_test(struct inode *inode, void *data)
544{
545 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
546}
547
548static int bdev_set(struct inode *inode, void *data)
549{
550 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
551 return 0;
552}
553
554static LIST_HEAD(all_bdevs);
555
556struct block_device *bdget(dev_t dev)
557{
558 struct block_device *bdev;
559 struct inode *inode;
560
561 inode = iget5_locked(blockdev_superblock, hash(dev),
562 bdev_test, bdev_set, &dev);
563
564 if (!inode)
565 return NULL;
566
567 bdev = &BDEV_I(inode)->bdev;
568
569 if (inode->i_state & I_NEW) {
570 bdev->bd_contains = NULL;
571 bdev->bd_super = NULL;
572 bdev->bd_inode = inode;
573 bdev->bd_block_size = (1 << inode->i_blkbits);
574 bdev->bd_part_count = 0;
575 bdev->bd_invalidated = 0;
576 inode->i_mode = S_IFBLK;
577 inode->i_rdev = dev;
578 inode->i_bdev = bdev;
579 inode->i_data.a_ops = &def_blk_aops;
580 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
581 inode->i_data.backing_dev_info = &default_backing_dev_info;
582 spin_lock(&bdev_lock);
583 list_add(&bdev->bd_list, &all_bdevs);
584 spin_unlock(&bdev_lock);
585 unlock_new_inode(inode);
586 }
587 return bdev;
588}
589
590EXPORT_SYMBOL(bdget);
591
592/**
593 * bdgrab -- Grab a reference to an already referenced block device
594 * @bdev: Block device to grab a reference to.
595 */
596struct block_device *bdgrab(struct block_device *bdev)
597{
598 ihold(bdev->bd_inode);
599 return bdev;
600}
601EXPORT_SYMBOL(bdgrab);
602
603long nr_blockdev_pages(void)
604{
605 struct block_device *bdev;
606 long ret = 0;
607 spin_lock(&bdev_lock);
608 list_for_each_entry(bdev, &all_bdevs, bd_list) {
609 ret += bdev->bd_inode->i_mapping->nrpages;
610 }
611 spin_unlock(&bdev_lock);
612 return ret;
613}
614
615void bdput(struct block_device *bdev)
616{
617 iput(bdev->bd_inode);
618}
619
620EXPORT_SYMBOL(bdput);
621
622static struct block_device *bd_acquire(struct inode *inode)
623{
624 struct block_device *bdev;
625
626 spin_lock(&bdev_lock);
627 bdev = inode->i_bdev;
628 if (bdev) {
629 ihold(bdev->bd_inode);
630 spin_unlock(&bdev_lock);
631 return bdev;
632 }
633 spin_unlock(&bdev_lock);
634
635 bdev = bdget(inode->i_rdev);
636 if (bdev) {
637 spin_lock(&bdev_lock);
638 if (!inode->i_bdev) {
639 /*
640 * We take an additional reference to bd_inode,
641 * and it's released in clear_inode() of inode.
642 * So, we can access it via ->i_mapping always
643 * without igrab().
644 */
645 ihold(bdev->bd_inode);
646 inode->i_bdev = bdev;
647 inode->i_mapping = bdev->bd_inode->i_mapping;
648 list_add(&inode->i_devices, &bdev->bd_inodes);
649 }
650 spin_unlock(&bdev_lock);
651 }
652 return bdev;
653}
654
655int sb_is_blkdev_sb(struct super_block *sb)
656{
657 return sb == blockdev_superblock;
658}
659
660/* Call when you free inode */
661
662void bd_forget(struct inode *inode)
663{
664 struct block_device *bdev = NULL;
665
666 spin_lock(&bdev_lock);
667 if (!sb_is_blkdev_sb(inode->i_sb))
668 bdev = inode->i_bdev;
669 __bd_forget(inode);
670 spin_unlock(&bdev_lock);
671
672 if (bdev)
673 iput(bdev->bd_inode);
674}
675
676/**
677 * bd_may_claim - test whether a block device can be claimed
678 * @bdev: block device of interest
679 * @whole: whole block device containing @bdev, may equal @bdev
680 * @holder: holder trying to claim @bdev
681 *
682 * Test whether @bdev can be claimed by @holder.
683 *
684 * CONTEXT:
685 * spin_lock(&bdev_lock).
686 *
687 * RETURNS:
688 * %true if @bdev can be claimed, %false otherwise.
689 */
690static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
691 void *holder)
692{
693 if (bdev->bd_holder == holder)
694 return true; /* already a holder */
695 else if (bdev->bd_holder != NULL)
696 return false; /* held by someone else */
697 else if (bdev->bd_contains == bdev)
698 return true; /* is a whole device which isn't held */
699
700 else if (whole->bd_holder == bd_may_claim)
701 return true; /* is a partition of a device that is being partitioned */
702 else if (whole->bd_holder != NULL)
703 return false; /* is a partition of a held device */
704 else
705 return true; /* is a partition of an un-held device */
706}
707
708/**
709 * bd_prepare_to_claim - prepare to claim a block device
710 * @bdev: block device of interest
711 * @whole: the whole device containing @bdev, may equal @bdev
712 * @holder: holder trying to claim @bdev
713 *
714 * Prepare to claim @bdev. This function fails if @bdev is already
715 * claimed by another holder and waits if another claiming is in
716 * progress. This function doesn't actually claim. On successful
717 * return, the caller has ownership of bd_claiming and bd_holder[s].
718 *
719 * CONTEXT:
720 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
721 * it multiple times.
722 *
723 * RETURNS:
724 * 0 if @bdev can be claimed, -EBUSY otherwise.
725 */
726static int bd_prepare_to_claim(struct block_device *bdev,
727 struct block_device *whole, void *holder)
728{
729retry:
730 /* if someone else claimed, fail */
731 if (!bd_may_claim(bdev, whole, holder))
732 return -EBUSY;
733
734 /* if claiming is already in progress, wait for it to finish */
735 if (whole->bd_claiming) {
736 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
737 DEFINE_WAIT(wait);
738
739 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
740 spin_unlock(&bdev_lock);
741 schedule();
742 finish_wait(wq, &wait);
743 spin_lock(&bdev_lock);
744 goto retry;
745 }
746
747 /* yay, all mine */
748 return 0;
749}
750
751/**
752 * bd_start_claiming - start claiming a block device
753 * @bdev: block device of interest
754 * @holder: holder trying to claim @bdev
755 *
756 * @bdev is about to be opened exclusively. Check @bdev can be opened
757 * exclusively and mark that an exclusive open is in progress. Each
758 * successful call to this function must be matched with a call to
759 * either bd_finish_claiming() or bd_abort_claiming() (which do not
760 * fail).
761 *
762 * This function is used to gain exclusive access to the block device
763 * without actually causing other exclusive open attempts to fail. It
764 * should be used when the open sequence itself requires exclusive
765 * access but may subsequently fail.
766 *
767 * CONTEXT:
768 * Might sleep.
769 *
770 * RETURNS:
771 * Pointer to the block device containing @bdev on success, ERR_PTR()
772 * value on failure.
773 */
774static struct block_device *bd_start_claiming(struct block_device *bdev,
775 void *holder)
776{
777 struct gendisk *disk;
778 struct block_device *whole;
779 int partno, err;
780
781 might_sleep();
782
783 /*
784 * @bdev might not have been initialized properly yet, look up
785 * and grab the outer block device the hard way.
786 */
787 disk = get_gendisk(bdev->bd_dev, &partno);
788 if (!disk)
789 return ERR_PTR(-ENXIO);
790
791 /*
792 * Normally, @bdev should equal what's returned from bdget_disk()
793 * if partno is 0; however, some drivers (floppy) use multiple
794 * bdev's for the same physical device and @bdev may be one of the
795 * aliases. Keep @bdev if partno is 0. This means claimer
796 * tracking is broken for those devices but it has always been that
797 * way.
798 */
799 if (partno)
800 whole = bdget_disk(disk, 0);
801 else
802 whole = bdgrab(bdev);
803
804 module_put(disk->fops->owner);
805 put_disk(disk);
806 if (!whole)
807 return ERR_PTR(-ENOMEM);
808
809 /* prepare to claim, if successful, mark claiming in progress */
810 spin_lock(&bdev_lock);
811
812 err = bd_prepare_to_claim(bdev, whole, holder);
813 if (err == 0) {
814 whole->bd_claiming = holder;
815 spin_unlock(&bdev_lock);
816 return whole;
817 } else {
818 spin_unlock(&bdev_lock);
819 bdput(whole);
820 return ERR_PTR(err);
821 }
822}
823
824#ifdef CONFIG_SYSFS
825struct bd_holder_disk {
826 struct list_head list;
827 struct gendisk *disk;
828 int refcnt;
829};
830
831static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
832 struct gendisk *disk)
833{
834 struct bd_holder_disk *holder;
835
836 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
837 if (holder->disk == disk)
838 return holder;
839 return NULL;
840}
841
842static int add_symlink(struct kobject *from, struct kobject *to)
843{
844 return sysfs_create_link(from, to, kobject_name(to));
845}
846
847static void del_symlink(struct kobject *from, struct kobject *to)
848{
849 sysfs_remove_link(from, kobject_name(to));
850}
851
852/**
853 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
854 * @bdev: the claimed slave bdev
855 * @disk: the holding disk
856 *
857 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
858 *
859 * This functions creates the following sysfs symlinks.
860 *
861 * - from "slaves" directory of the holder @disk to the claimed @bdev
862 * - from "holders" directory of the @bdev to the holder @disk
863 *
864 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
865 * passed to bd_link_disk_holder(), then:
866 *
867 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
868 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
869 *
870 * The caller must have claimed @bdev before calling this function and
871 * ensure that both @bdev and @disk are valid during the creation and
872 * lifetime of these symlinks.
873 *
874 * CONTEXT:
875 * Might sleep.
876 *
877 * RETURNS:
878 * 0 on success, -errno on failure.
879 */
880int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
881{
882 struct bd_holder_disk *holder;
883 int ret = 0;
884
885 mutex_lock(&bdev->bd_mutex);
886
887 WARN_ON_ONCE(!bdev->bd_holder);
888
889 /* FIXME: remove the following once add_disk() handles errors */
890 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
891 goto out_unlock;
892
893 holder = bd_find_holder_disk(bdev, disk);
894 if (holder) {
895 holder->refcnt++;
896 goto out_unlock;
897 }
898
899 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
900 if (!holder) {
901 ret = -ENOMEM;
902 goto out_unlock;
903 }
904
905 INIT_LIST_HEAD(&holder->list);
906 holder->disk = disk;
907 holder->refcnt = 1;
908
909 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
910 if (ret)
911 goto out_free;
912
913 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
914 if (ret)
915 goto out_del;
916 /*
917 * bdev could be deleted beneath us which would implicitly destroy
918 * the holder directory. Hold on to it.
919 */
920 kobject_get(bdev->bd_part->holder_dir);
921
922 list_add(&holder->list, &bdev->bd_holder_disks);
923 goto out_unlock;
924
925out_del:
926 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
927out_free:
928 kfree(holder);
929out_unlock:
930 mutex_unlock(&bdev->bd_mutex);
931 return ret;
932}
933EXPORT_SYMBOL_GPL(bd_link_disk_holder);
934
935/**
936 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
937 * @bdev: the calimed slave bdev
938 * @disk: the holding disk
939 *
940 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
941 *
942 * CONTEXT:
943 * Might sleep.
944 */
945void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
946{
947 struct bd_holder_disk *holder;
948
949 mutex_lock(&bdev->bd_mutex);
950
951 holder = bd_find_holder_disk(bdev, disk);
952
953 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
954 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
955 del_symlink(bdev->bd_part->holder_dir,
956 &disk_to_dev(disk)->kobj);
957 kobject_put(bdev->bd_part->holder_dir);
958 list_del_init(&holder->list);
959 kfree(holder);
960 }
961
962 mutex_unlock(&bdev->bd_mutex);
963}
964EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
965#endif
966
967/**
968 * flush_disk - invalidates all buffer-cache entries on a disk
969 *
970 * @bdev: struct block device to be flushed
971 * @kill_dirty: flag to guide handling of dirty inodes
972 *
973 * Invalidates all buffer-cache entries on a disk. It should be called
974 * when a disk has been changed -- either by a media change or online
975 * resize.
976 */
977static void flush_disk(struct block_device *bdev, bool kill_dirty)
978{
979 if (__invalidate_device(bdev, kill_dirty)) {
980 char name[BDEVNAME_SIZE] = "";
981
982 if (bdev->bd_disk)
983 disk_name(bdev->bd_disk, 0, name);
984 printk(KERN_WARNING "VFS: busy inodes on changed media or "
985 "resized disk %s\n", name);
986 }
987
988 if (!bdev->bd_disk)
989 return;
990 if (disk_part_scan_enabled(bdev->bd_disk))
991 bdev->bd_invalidated = 1;
992}
993
994/**
995 * check_disk_size_change - checks for disk size change and adjusts bdev size.
996 * @disk: struct gendisk to check
997 * @bdev: struct bdev to adjust.
998 *
999 * This routine checks to see if the bdev size does not match the disk size
1000 * and adjusts it if it differs.
1001 */
1002void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1003{
1004 loff_t disk_size, bdev_size;
1005
1006 disk_size = (loff_t)get_capacity(disk) << 9;
1007 bdev_size = i_size_read(bdev->bd_inode);
1008 if (disk_size != bdev_size) {
1009 char name[BDEVNAME_SIZE];
1010
1011 disk_name(disk, 0, name);
1012 printk(KERN_INFO
1013 "%s: detected capacity change from %lld to %lld\n",
1014 name, bdev_size, disk_size);
1015 i_size_write(bdev->bd_inode, disk_size);
1016 flush_disk(bdev, false);
1017 }
1018}
1019EXPORT_SYMBOL(check_disk_size_change);
1020
1021/**
1022 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1023 * @disk: struct gendisk to be revalidated
1024 *
1025 * This routine is a wrapper for lower-level driver's revalidate_disk
1026 * call-backs. It is used to do common pre and post operations needed
1027 * for all revalidate_disk operations.
1028 */
1029int revalidate_disk(struct gendisk *disk)
1030{
1031 struct block_device *bdev;
1032 int ret = 0;
1033
1034 if (disk->fops->revalidate_disk)
1035 ret = disk->fops->revalidate_disk(disk);
1036
1037 bdev = bdget_disk(disk, 0);
1038 if (!bdev)
1039 return ret;
1040
1041 mutex_lock(&bdev->bd_mutex);
1042 check_disk_size_change(disk, bdev);
1043 bdev->bd_invalidated = 0;
1044 mutex_unlock(&bdev->bd_mutex);
1045 bdput(bdev);
1046 return ret;
1047}
1048EXPORT_SYMBOL(revalidate_disk);
1049
1050/*
1051 * This routine checks whether a removable media has been changed,
1052 * and invalidates all buffer-cache-entries in that case. This
1053 * is a relatively slow routine, so we have to try to minimize using
1054 * it. Thus it is called only upon a 'mount' or 'open'. This
1055 * is the best way of combining speed and utility, I think.
1056 * People changing diskettes in the middle of an operation deserve
1057 * to lose :-)
1058 */
1059int check_disk_change(struct block_device *bdev)
1060{
1061 struct gendisk *disk = bdev->bd_disk;
1062 const struct block_device_operations *bdops = disk->fops;
1063 unsigned int events;
1064
1065 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1066 DISK_EVENT_EJECT_REQUEST);
1067 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1068 return 0;
1069
1070 flush_disk(bdev, true);
1071 if (bdops->revalidate_disk)
1072 bdops->revalidate_disk(bdev->bd_disk);
1073 return 1;
1074}
1075
1076EXPORT_SYMBOL(check_disk_change);
1077
1078void bd_set_size(struct block_device *bdev, loff_t size)
1079{
1080 unsigned bsize = bdev_logical_block_size(bdev);
1081
1082 mutex_lock(&bdev->bd_inode->i_mutex);
1083 i_size_write(bdev->bd_inode, size);
1084 mutex_unlock(&bdev->bd_inode->i_mutex);
1085 while (bsize < PAGE_CACHE_SIZE) {
1086 if (size & bsize)
1087 break;
1088 bsize <<= 1;
1089 }
1090 bdev->bd_block_size = bsize;
1091 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1092}
1093EXPORT_SYMBOL(bd_set_size);
1094
1095static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1096
1097/*
1098 * bd_mutex locking:
1099 *
1100 * mutex_lock(part->bd_mutex)
1101 * mutex_lock_nested(whole->bd_mutex, 1)
1102 */
1103
1104static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1105{
1106 struct gendisk *disk;
1107 struct module *owner;
1108 int ret;
1109 int partno;
1110 int perm = 0;
1111
1112 if (mode & FMODE_READ)
1113 perm |= MAY_READ;
1114 if (mode & FMODE_WRITE)
1115 perm |= MAY_WRITE;
1116 /*
1117 * hooks: /n/, see "layering violations".
1118 */
1119 if (!for_part) {
1120 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1121 if (ret != 0) {
1122 bdput(bdev);
1123 return ret;
1124 }
1125 }
1126
1127 restart:
1128
1129 ret = -ENXIO;
1130 disk = get_gendisk(bdev->bd_dev, &partno);
1131 if (!disk)
1132 goto out;
1133 owner = disk->fops->owner;
1134
1135 disk_block_events(disk);
1136 mutex_lock_nested(&bdev->bd_mutex, for_part);
1137 if (!bdev->bd_openers) {
1138 bdev->bd_disk = disk;
1139 bdev->bd_queue = disk->queue;
1140 bdev->bd_contains = bdev;
1141 if (!partno) {
1142 struct backing_dev_info *bdi;
1143
1144 ret = -ENXIO;
1145 bdev->bd_part = disk_get_part(disk, partno);
1146 if (!bdev->bd_part)
1147 goto out_clear;
1148
1149 ret = 0;
1150 if (disk->fops->open) {
1151 ret = disk->fops->open(bdev, mode);
1152 if (ret == -ERESTARTSYS) {
1153 /* Lost a race with 'disk' being
1154 * deleted, try again.
1155 * See md.c
1156 */
1157 disk_put_part(bdev->bd_part);
1158 bdev->bd_part = NULL;
1159 bdev->bd_disk = NULL;
1160 bdev->bd_queue = NULL;
1161 mutex_unlock(&bdev->bd_mutex);
1162 disk_unblock_events(disk);
1163 put_disk(disk);
1164 module_put(owner);
1165 goto restart;
1166 }
1167 }
1168
1169 if (!ret) {
1170 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1171 bdi = blk_get_backing_dev_info(bdev);
1172 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1173 }
1174
1175 /*
1176 * If the device is invalidated, rescan partition
1177 * if open succeeded or failed with -ENOMEDIUM.
1178 * The latter is necessary to prevent ghost
1179 * partitions on a removed medium.
1180 */
1181 if (bdev->bd_invalidated) {
1182 if (!ret)
1183 rescan_partitions(disk, bdev);
1184 else if (ret == -ENOMEDIUM)
1185 invalidate_partitions(disk, bdev);
1186 }
1187 if (ret)
1188 goto out_clear;
1189 } else {
1190 struct block_device *whole;
1191 whole = bdget_disk(disk, 0);
1192 ret = -ENOMEM;
1193 if (!whole)
1194 goto out_clear;
1195 BUG_ON(for_part);
1196 ret = __blkdev_get(whole, mode, 1);
1197 if (ret)
1198 goto out_clear;
1199 bdev->bd_contains = whole;
1200 bdev_inode_switch_bdi(bdev->bd_inode,
1201 whole->bd_inode->i_data.backing_dev_info);
1202 bdev->bd_part = disk_get_part(disk, partno);
1203 if (!(disk->flags & GENHD_FL_UP) ||
1204 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1205 ret = -ENXIO;
1206 goto out_clear;
1207 }
1208 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1209 }
1210 } else {
1211 if (bdev->bd_contains == bdev) {
1212 ret = 0;
1213 if (bdev->bd_disk->fops->open)
1214 ret = bdev->bd_disk->fops->open(bdev, mode);
1215 /* the same as first opener case, read comment there */
1216 if (bdev->bd_invalidated) {
1217 if (!ret)
1218 rescan_partitions(bdev->bd_disk, bdev);
1219 else if (ret == -ENOMEDIUM)
1220 invalidate_partitions(bdev->bd_disk, bdev);
1221 }
1222 if (ret)
1223 goto out_unlock_bdev;
1224 }
1225 /* only one opener holds refs to the module and disk */
1226 put_disk(disk);
1227 module_put(owner);
1228 }
1229 bdev->bd_openers++;
1230 if (for_part)
1231 bdev->bd_part_count++;
1232 mutex_unlock(&bdev->bd_mutex);
1233 disk_unblock_events(disk);
1234 return 0;
1235
1236 out_clear:
1237 disk_put_part(bdev->bd_part);
1238 bdev->bd_disk = NULL;
1239 bdev->bd_part = NULL;
1240 bdev->bd_queue = NULL;
1241 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
1242 if (bdev != bdev->bd_contains)
1243 __blkdev_put(bdev->bd_contains, mode, 1);
1244 bdev->bd_contains = NULL;
1245 out_unlock_bdev:
1246 mutex_unlock(&bdev->bd_mutex);
1247 disk_unblock_events(disk);
1248 put_disk(disk);
1249 module_put(owner);
1250 out:
1251 bdput(bdev);
1252
1253 return ret;
1254}
1255
1256/**
1257 * blkdev_get - open a block device
1258 * @bdev: block_device to open
1259 * @mode: FMODE_* mask
1260 * @holder: exclusive holder identifier
1261 *
1262 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1263 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1264 * @holder is invalid. Exclusive opens may nest for the same @holder.
1265 *
1266 * On success, the reference count of @bdev is unchanged. On failure,
1267 * @bdev is put.
1268 *
1269 * CONTEXT:
1270 * Might sleep.
1271 *
1272 * RETURNS:
1273 * 0 on success, -errno on failure.
1274 */
1275int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1276{
1277 struct block_device *whole = NULL;
1278 int res;
1279
1280 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1281
1282 if ((mode & FMODE_EXCL) && holder) {
1283 whole = bd_start_claiming(bdev, holder);
1284 if (IS_ERR(whole)) {
1285 bdput(bdev);
1286 return PTR_ERR(whole);
1287 }
1288 }
1289
1290 res = __blkdev_get(bdev, mode, 0);
1291
1292 if (whole) {
1293 struct gendisk *disk = whole->bd_disk;
1294
1295 /* finish claiming */
1296 mutex_lock(&bdev->bd_mutex);
1297 spin_lock(&bdev_lock);
1298
1299 if (!res) {
1300 BUG_ON(!bd_may_claim(bdev, whole, holder));
1301 /*
1302 * Note that for a whole device bd_holders
1303 * will be incremented twice, and bd_holder
1304 * will be set to bd_may_claim before being
1305 * set to holder
1306 */
1307 whole->bd_holders++;
1308 whole->bd_holder = bd_may_claim;
1309 bdev->bd_holders++;
1310 bdev->bd_holder = holder;
1311 }
1312
1313 /* tell others that we're done */
1314 BUG_ON(whole->bd_claiming != holder);
1315 whole->bd_claiming = NULL;
1316 wake_up_bit(&whole->bd_claiming, 0);
1317
1318 spin_unlock(&bdev_lock);
1319
1320 /*
1321 * Block event polling for write claims if requested. Any
1322 * write holder makes the write_holder state stick until
1323 * all are released. This is good enough and tracking
1324 * individual writeable reference is too fragile given the
1325 * way @mode is used in blkdev_get/put().
1326 */
1327 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1328 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1329 bdev->bd_write_holder = true;
1330 disk_block_events(disk);
1331 }
1332
1333 mutex_unlock(&bdev->bd_mutex);
1334 bdput(whole);
1335 }
1336
1337 return res;
1338}
1339EXPORT_SYMBOL(blkdev_get);
1340
1341/**
1342 * blkdev_get_by_path - open a block device by name
1343 * @path: path to the block device to open
1344 * @mode: FMODE_* mask
1345 * @holder: exclusive holder identifier
1346 *
1347 * Open the blockdevice described by the device file at @path. @mode
1348 * and @holder are identical to blkdev_get().
1349 *
1350 * On success, the returned block_device has reference count of one.
1351 *
1352 * CONTEXT:
1353 * Might sleep.
1354 *
1355 * RETURNS:
1356 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1357 */
1358struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1359 void *holder)
1360{
1361 struct block_device *bdev;
1362 int err;
1363
1364 bdev = lookup_bdev(path);
1365 if (IS_ERR(bdev))
1366 return bdev;
1367
1368 err = blkdev_get(bdev, mode, holder);
1369 if (err)
1370 return ERR_PTR(err);
1371
1372 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1373 blkdev_put(bdev, mode);
1374 return ERR_PTR(-EACCES);
1375 }
1376
1377 return bdev;
1378}
1379EXPORT_SYMBOL(blkdev_get_by_path);
1380
1381/**
1382 * blkdev_get_by_dev - open a block device by device number
1383 * @dev: device number of block device to open
1384 * @mode: FMODE_* mask
1385 * @holder: exclusive holder identifier
1386 *
1387 * Open the blockdevice described by device number @dev. @mode and
1388 * @holder are identical to blkdev_get().
1389 *
1390 * Use it ONLY if you really do not have anything better - i.e. when
1391 * you are behind a truly sucky interface and all you are given is a
1392 * device number. _Never_ to be used for internal purposes. If you
1393 * ever need it - reconsider your API.
1394 *
1395 * On success, the returned block_device has reference count of one.
1396 *
1397 * CONTEXT:
1398 * Might sleep.
1399 *
1400 * RETURNS:
1401 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1402 */
1403struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1404{
1405 struct block_device *bdev;
1406 int err;
1407
1408 bdev = bdget(dev);
1409 if (!bdev)
1410 return ERR_PTR(-ENOMEM);
1411
1412 err = blkdev_get(bdev, mode, holder);
1413 if (err)
1414 return ERR_PTR(err);
1415
1416 return bdev;
1417}
1418EXPORT_SYMBOL(blkdev_get_by_dev);
1419
1420static int blkdev_open(struct inode * inode, struct file * filp)
1421{
1422 struct block_device *bdev;
1423
1424 /*
1425 * Preserve backwards compatibility and allow large file access
1426 * even if userspace doesn't ask for it explicitly. Some mkfs
1427 * binary needs it. We might want to drop this workaround
1428 * during an unstable branch.
1429 */
1430 filp->f_flags |= O_LARGEFILE;
1431
1432 if (filp->f_flags & O_NDELAY)
1433 filp->f_mode |= FMODE_NDELAY;
1434 if (filp->f_flags & O_EXCL)
1435 filp->f_mode |= FMODE_EXCL;
1436 if ((filp->f_flags & O_ACCMODE) == 3)
1437 filp->f_mode |= FMODE_WRITE_IOCTL;
1438
1439 bdev = bd_acquire(inode);
1440 if (bdev == NULL)
1441 return -ENOMEM;
1442
1443 filp->f_mapping = bdev->bd_inode->i_mapping;
1444
1445 return blkdev_get(bdev, filp->f_mode, filp);
1446}
1447
1448static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1449{
1450 struct gendisk *disk = bdev->bd_disk;
1451 struct block_device *victim = NULL;
1452
1453 mutex_lock_nested(&bdev->bd_mutex, for_part);
1454 if (for_part)
1455 bdev->bd_part_count--;
1456
1457 if (!--bdev->bd_openers) {
1458 WARN_ON_ONCE(bdev->bd_holders);
1459 sync_blockdev(bdev);
1460 kill_bdev(bdev);
1461 /* ->release can cause the old bdi to disappear,
1462 * so must switch it out first
1463 */
1464 bdev_inode_switch_bdi(bdev->bd_inode,
1465 &default_backing_dev_info);
1466 }
1467 if (bdev->bd_contains == bdev) {
1468 if (disk->fops->release)
1469 disk->fops->release(disk, mode);
1470 }
1471 if (!bdev->bd_openers) {
1472 struct module *owner = disk->fops->owner;
1473
1474 disk_put_part(bdev->bd_part);
1475 bdev->bd_part = NULL;
1476 bdev->bd_disk = NULL;
1477 if (bdev != bdev->bd_contains)
1478 victim = bdev->bd_contains;
1479 bdev->bd_contains = NULL;
1480
1481 put_disk(disk);
1482 module_put(owner);
1483 }
1484 mutex_unlock(&bdev->bd_mutex);
1485 bdput(bdev);
1486 if (victim)
1487 __blkdev_put(victim, mode, 1);
1488}
1489
1490void blkdev_put(struct block_device *bdev, fmode_t mode)
1491{
1492 mutex_lock(&bdev->bd_mutex);
1493
1494 if (mode & FMODE_EXCL) {
1495 bool bdev_free;
1496
1497 /*
1498 * Release a claim on the device. The holder fields
1499 * are protected with bdev_lock. bd_mutex is to
1500 * synchronize disk_holder unlinking.
1501 */
1502 spin_lock(&bdev_lock);
1503
1504 WARN_ON_ONCE(--bdev->bd_holders < 0);
1505 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1506
1507 /* bd_contains might point to self, check in a separate step */
1508 if ((bdev_free = !bdev->bd_holders))
1509 bdev->bd_holder = NULL;
1510 if (!bdev->bd_contains->bd_holders)
1511 bdev->bd_contains->bd_holder = NULL;
1512
1513 spin_unlock(&bdev_lock);
1514
1515 /*
1516 * If this was the last claim, remove holder link and
1517 * unblock evpoll if it was a write holder.
1518 */
1519 if (bdev_free && bdev->bd_write_holder) {
1520 disk_unblock_events(bdev->bd_disk);
1521 bdev->bd_write_holder = false;
1522 }
1523 }
1524
1525 /*
1526 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1527 * event. This is to ensure detection of media removal commanded
1528 * from userland - e.g. eject(1).
1529 */
1530 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1531
1532 mutex_unlock(&bdev->bd_mutex);
1533
1534 __blkdev_put(bdev, mode, 0);
1535}
1536EXPORT_SYMBOL(blkdev_put);
1537
1538static int blkdev_close(struct inode * inode, struct file * filp)
1539{
1540 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1541 blkdev_put(bdev, filp->f_mode);
1542 return 0;
1543}
1544
1545static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1546{
1547 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1548 fmode_t mode = file->f_mode;
1549
1550 /*
1551 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1552 * to updated it before every ioctl.
1553 */
1554 if (file->f_flags & O_NDELAY)
1555 mode |= FMODE_NDELAY;
1556 else
1557 mode &= ~FMODE_NDELAY;
1558
1559 return blkdev_ioctl(bdev, mode, cmd, arg);
1560}
1561
1562/*
1563 * Write data to the block device. Only intended for the block device itself
1564 * and the raw driver which basically is a fake block device.
1565 *
1566 * Does not take i_mutex for the write and thus is not for general purpose
1567 * use.
1568 */
1569ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1570{
1571 struct file *file = iocb->ki_filp;
1572 struct blk_plug plug;
1573 ssize_t ret;
1574
1575 blk_start_plug(&plug);
1576 ret = __generic_file_write_iter(iocb, from);
1577 if (ret > 0) {
1578 ssize_t err;
1579 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1580 if (err < 0)
1581 ret = err;
1582 }
1583 blk_finish_plug(&plug);
1584 return ret;
1585}
1586EXPORT_SYMBOL_GPL(blkdev_write_iter);
1587
1588ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1589{
1590 struct file *file = iocb->ki_filp;
1591 struct inode *bd_inode = file->f_mapping->host;
1592 loff_t size = i_size_read(bd_inode);
1593 loff_t pos = iocb->ki_pos;
1594
1595 if (pos >= size)
1596 return 0;
1597
1598 size -= pos;
1599 iov_iter_truncate(to, size);
1600 return generic_file_read_iter(iocb, to);
1601}
1602EXPORT_SYMBOL_GPL(blkdev_read_iter);
1603
1604/*
1605 * Try to release a page associated with block device when the system
1606 * is under memory pressure.
1607 */
1608static int blkdev_releasepage(struct page *page, gfp_t wait)
1609{
1610 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1611
1612 if (super && super->s_op->bdev_try_to_free_page)
1613 return super->s_op->bdev_try_to_free_page(super, page, wait);
1614
1615 return try_to_free_buffers(page);
1616}
1617
1618static const struct address_space_operations def_blk_aops = {
1619 .readpage = blkdev_readpage,
1620 .readpages = blkdev_readpages,
1621 .writepage = blkdev_writepage,
1622 .write_begin = blkdev_write_begin,
1623 .write_end = blkdev_write_end,
1624 .writepages = generic_writepages,
1625 .releasepage = blkdev_releasepage,
1626 .direct_IO = blkdev_direct_IO,
1627 .is_dirty_writeback = buffer_check_dirty_writeback,
1628};
1629
1630const struct file_operations def_blk_fops = {
1631 .open = blkdev_open,
1632 .release = blkdev_close,
1633 .llseek = block_llseek,
1634 .read = new_sync_read,
1635 .write = new_sync_write,
1636 .read_iter = blkdev_read_iter,
1637 .write_iter = blkdev_write_iter,
1638 .mmap = generic_file_mmap,
1639 .fsync = blkdev_fsync,
1640 .unlocked_ioctl = block_ioctl,
1641#ifdef CONFIG_COMPAT
1642 .compat_ioctl = compat_blkdev_ioctl,
1643#endif
1644 .splice_read = generic_file_splice_read,
1645 .splice_write = iter_file_splice_write,
1646};
1647
1648int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1649{
1650 int res;
1651 mm_segment_t old_fs = get_fs();
1652 set_fs(KERNEL_DS);
1653 res = blkdev_ioctl(bdev, 0, cmd, arg);
1654 set_fs(old_fs);
1655 return res;
1656}
1657
1658EXPORT_SYMBOL(ioctl_by_bdev);
1659
1660/**
1661 * lookup_bdev - lookup a struct block_device by name
1662 * @pathname: special file representing the block device
1663 *
1664 * Get a reference to the blockdevice at @pathname in the current
1665 * namespace if possible and return it. Return ERR_PTR(error)
1666 * otherwise.
1667 */
1668struct block_device *lookup_bdev(const char *pathname)
1669{
1670 struct block_device *bdev;
1671 struct inode *inode;
1672 struct path path;
1673 int error;
1674
1675 if (!pathname || !*pathname)
1676 return ERR_PTR(-EINVAL);
1677
1678 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1679 if (error)
1680 return ERR_PTR(error);
1681
1682 inode = path.dentry->d_inode;
1683 error = -ENOTBLK;
1684 if (!S_ISBLK(inode->i_mode))
1685 goto fail;
1686 error = -EACCES;
1687 if (path.mnt->mnt_flags & MNT_NODEV)
1688 goto fail;
1689 error = -ENOMEM;
1690 bdev = bd_acquire(inode);
1691 if (!bdev)
1692 goto fail;
1693out:
1694 path_put(&path);
1695 return bdev;
1696fail:
1697 bdev = ERR_PTR(error);
1698 goto out;
1699}
1700EXPORT_SYMBOL(lookup_bdev);
1701
1702int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1703{
1704 struct super_block *sb = get_super(bdev);
1705 int res = 0;
1706
1707 if (sb) {
1708 /*
1709 * no need to lock the super, get_super holds the
1710 * read mutex so the filesystem cannot go away
1711 * under us (->put_super runs with the write lock
1712 * hold).
1713 */
1714 shrink_dcache_sb(sb);
1715 res = invalidate_inodes(sb, kill_dirty);
1716 drop_super(sb);
1717 }
1718 invalidate_bdev(bdev);
1719 return res;
1720}
1721EXPORT_SYMBOL(__invalidate_device);
1722
1723void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1724{
1725 struct inode *inode, *old_inode = NULL;
1726
1727 spin_lock(&inode_sb_list_lock);
1728 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1729 struct address_space *mapping = inode->i_mapping;
1730
1731 spin_lock(&inode->i_lock);
1732 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1733 mapping->nrpages == 0) {
1734 spin_unlock(&inode->i_lock);
1735 continue;
1736 }
1737 __iget(inode);
1738 spin_unlock(&inode->i_lock);
1739 spin_unlock(&inode_sb_list_lock);
1740 /*
1741 * We hold a reference to 'inode' so it couldn't have been
1742 * removed from s_inodes list while we dropped the
1743 * inode_sb_list_lock. We cannot iput the inode now as we can
1744 * be holding the last reference and we cannot iput it under
1745 * inode_sb_list_lock. So we keep the reference and iput it
1746 * later.
1747 */
1748 iput(old_inode);
1749 old_inode = inode;
1750
1751 func(I_BDEV(inode), arg);
1752
1753 spin_lock(&inode_sb_list_lock);
1754 }
1755 spin_unlock(&inode_sb_list_lock);
1756 iput(old_inode);
1757}