4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/init.h>
10 #include <linux/fcntl.h>
11 #include <linux/slab.h>
12 #include <linux/kmod.h>
13 #include <linux/major.h>
14 #include <linux/smp_lock.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/buffer_head.h>
21 #include <linux/pagevec.h>
22 #include <linux/writeback.h>
23 #include <linux/mpage.h>
24 #include <linux/mount.h>
25 #include <linux/uio.h>
26 #include <linux/namei.h>
27 #include <linux/log2.h>
28 #include <linux/kmemleak.h>
29 #include <asm/uaccess.h>
33 struct block_device bdev;
34 struct inode vfs_inode;
37 static const struct address_space_operations def_blk_aops;
39 static inline struct bdev_inode *BDEV_I(struct inode *inode)
41 return container_of(inode, struct bdev_inode, vfs_inode);
44 inline struct block_device *I_BDEV(struct inode *inode)
46 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
51 static sector_t max_block(struct block_device *bdev)
53 sector_t retval = ~((sector_t)0);
54 loff_t sz = i_size_read(bdev->bd_inode);
57 unsigned int size = block_size(bdev);
58 unsigned int sizebits = blksize_bits(size);
59 retval = (sz >> sizebits);
64 /* Kill _all_ buffers and pagecache , dirty or not.. */
65 static void kill_bdev(struct block_device *bdev)
67 if (bdev->bd_inode->i_mapping->nrpages == 0)
70 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
73 int set_blocksize(struct block_device *bdev, int size)
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
76 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
79 /* Size cannot be smaller than the size supported by the device */
80 if (size < bdev_logical_block_size(bdev))
83 /* Don't change the size if it is same as current */
84 if (bdev->bd_block_size != size) {
86 bdev->bd_block_size = size;
87 bdev->bd_inode->i_blkbits = blksize_bits(size);
93 EXPORT_SYMBOL(set_blocksize);
95 int sb_set_blocksize(struct super_block *sb, int size)
97 if (set_blocksize(sb->s_bdev, size))
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb->s_blocksize = size;
102 sb->s_blocksize_bits = blksize_bits(size);
103 return sb->s_blocksize;
106 EXPORT_SYMBOL(sb_set_blocksize);
108 int sb_min_blocksize(struct super_block *sb, int size)
110 int minsize = bdev_logical_block_size(sb->s_bdev);
113 return sb_set_blocksize(sb, size);
116 EXPORT_SYMBOL(sb_min_blocksize);
119 blkdev_get_block(struct inode *inode, sector_t iblock,
120 struct buffer_head *bh, int create)
122 if (iblock >= max_block(I_BDEV(inode))) {
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
134 bh->b_bdev = I_BDEV(inode);
135 bh->b_blocknr = iblock;
136 set_buffer_mapped(bh);
141 blkdev_get_blocks(struct inode *inode, sector_t iblock,
142 struct buffer_head *bh, int create)
144 sector_t end_block = max_block(I_BDEV(inode));
145 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
147 if ((iblock + max_blocks) > end_block) {
148 max_blocks = end_block - iblock;
149 if ((long)max_blocks <= 0) {
151 return -EIO; /* write fully beyond EOF */
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
160 bh->b_bdev = I_BDEV(inode);
161 bh->b_blocknr = iblock;
162 bh->b_size = max_blocks << inode->i_blkbits;
164 set_buffer_mapped(bh);
169 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
170 loff_t offset, unsigned long nr_segs)
172 struct file *file = iocb->ki_filp;
173 struct inode *inode = file->f_mapping->host;
175 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
176 iov, offset, nr_segs, blkdev_get_blocks, NULL);
179 int __sync_blockdev(struct block_device *bdev, int wait)
184 return filemap_flush(bdev->bd_inode->i_mapping);
185 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
192 int sync_blockdev(struct block_device *bdev)
194 return __sync_blockdev(bdev, 1);
196 EXPORT_SYMBOL(sync_blockdev);
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
203 int fsync_bdev(struct block_device *bdev)
205 struct super_block *sb = get_super(bdev);
207 int res = sync_filesystem(sb);
211 return sync_blockdev(bdev);
213 EXPORT_SYMBOL(fsync_bdev);
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
227 struct super_block *freeze_bdev(struct block_device *bdev)
229 struct super_block *sb;
232 mutex_lock(&bdev->bd_fsfreeze_mutex);
233 if (++bdev->bd_fsfreeze_count > 1) {
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
239 sb = get_super(bdev);
241 mutex_unlock(&bdev->bd_fsfreeze_mutex);
245 sb = get_active_super(bdev);
248 error = freeze_super(sb);
250 deactivate_super(sb);
251 bdev->bd_fsfreeze_count--;
252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
253 return ERR_PTR(error);
255 deactivate_super(sb);
258 mutex_unlock(&bdev->bd_fsfreeze_mutex);
259 return sb; /* thaw_bdev releases s->s_umount */
261 EXPORT_SYMBOL(freeze_bdev);
264 * thaw_bdev -- unlock filesystem
265 * @bdev: blockdevice to unlock
266 * @sb: associated superblock
268 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
270 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
274 mutex_lock(&bdev->bd_fsfreeze_mutex);
275 if (!bdev->bd_fsfreeze_count)
279 if (--bdev->bd_fsfreeze_count > 0)
285 error = thaw_super(sb);
287 bdev->bd_fsfreeze_count++;
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
295 EXPORT_SYMBOL(thaw_bdev);
297 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
299 return block_write_full_page(page, blkdev_get_block, wbc);
302 static int blkdev_readpage(struct file * file, struct page * page)
304 return block_read_full_page(page, blkdev_get_block);
307 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned flags,
309 struct page **pagep, void **fsdata)
312 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
316 static int blkdev_write_end(struct file *file, struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned copied,
318 struct page *page, void *fsdata)
321 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
324 page_cache_release(page);
331 * for a block special file file->f_path.dentry->d_inode->i_size is zero
332 * so we compute the size by hand (just as in block_read/write above)
334 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
336 struct inode *bd_inode = file->f_mapping->host;
340 mutex_lock(&bd_inode->i_mutex);
341 size = i_size_read(bd_inode);
348 offset += file->f_pos;
351 if (offset >= 0 && offset <= size) {
352 if (offset != file->f_pos) {
353 file->f_pos = offset;
357 mutex_unlock(&bd_inode->i_mutex);
362 * Filp is never NULL; the only case when ->fsync() is called with
363 * NULL first argument is nfsd_sync_dir() and that's not a directory.
366 int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
368 struct inode *bd_inode = filp->f_mapping->host;
369 struct block_device *bdev = I_BDEV(bd_inode);
373 * There is no need to serialise calls to blkdev_issue_flush with
374 * i_mutex and doing so causes performance issues with concurrent
375 * O_SYNC writers to a block device.
377 mutex_unlock(&bd_inode->i_mutex);
379 error = blkdev_issue_flush(bdev, NULL);
380 if (error == -EOPNOTSUPP)
383 mutex_lock(&bd_inode->i_mutex);
387 EXPORT_SYMBOL(blkdev_fsync);
393 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
394 static struct kmem_cache * bdev_cachep __read_mostly;
396 static struct inode *bdev_alloc_inode(struct super_block *sb)
398 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
401 return &ei->vfs_inode;
404 static void bdev_destroy_inode(struct inode *inode)
406 struct bdev_inode *bdi = BDEV_I(inode);
408 kmem_cache_free(bdev_cachep, bdi);
411 static void init_once(void *foo)
413 struct bdev_inode *ei = (struct bdev_inode *) foo;
414 struct block_device *bdev = &ei->bdev;
416 memset(bdev, 0, sizeof(*bdev));
417 mutex_init(&bdev->bd_mutex);
418 INIT_LIST_HEAD(&bdev->bd_inodes);
419 INIT_LIST_HEAD(&bdev->bd_list);
421 INIT_LIST_HEAD(&bdev->bd_holder_list);
423 inode_init_once(&ei->vfs_inode);
424 /* Initialize mutex for freeze. */
425 mutex_init(&bdev->bd_fsfreeze_mutex);
428 static inline void __bd_forget(struct inode *inode)
430 list_del_init(&inode->i_devices);
431 inode->i_bdev = NULL;
432 inode->i_mapping = &inode->i_data;
435 static void bdev_clear_inode(struct inode *inode)
437 struct block_device *bdev = &BDEV_I(inode)->bdev;
439 spin_lock(&bdev_lock);
440 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
441 __bd_forget(list_entry(p, struct inode, i_devices));
443 list_del_init(&bdev->bd_list);
444 spin_unlock(&bdev_lock);
447 static const struct super_operations bdev_sops = {
448 .statfs = simple_statfs,
449 .alloc_inode = bdev_alloc_inode,
450 .destroy_inode = bdev_destroy_inode,
451 .drop_inode = generic_delete_inode,
452 .clear_inode = bdev_clear_inode,
455 static int bd_get_sb(struct file_system_type *fs_type,
456 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
458 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
461 static struct file_system_type bd_type = {
464 .kill_sb = kill_anon_super,
467 struct super_block *blockdev_superblock __read_mostly;
469 void __init bdev_cache_init(void)
472 struct vfsmount *bd_mnt;
474 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
475 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
476 SLAB_MEM_SPREAD|SLAB_PANIC),
478 err = register_filesystem(&bd_type);
480 panic("Cannot register bdev pseudo-fs");
481 bd_mnt = kern_mount(&bd_type);
483 panic("Cannot create bdev pseudo-fs");
485 * This vfsmount structure is only used to obtain the
486 * blockdev_superblock, so tell kmemleak not to report it.
488 kmemleak_not_leak(bd_mnt);
489 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
493 * Most likely _very_ bad one - but then it's hardly critical for small
494 * /dev and can be fixed when somebody will need really large one.
495 * Keep in mind that it will be fed through icache hash function too.
497 static inline unsigned long hash(dev_t dev)
499 return MAJOR(dev)+MINOR(dev);
502 static int bdev_test(struct inode *inode, void *data)
504 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
507 static int bdev_set(struct inode *inode, void *data)
509 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
513 static LIST_HEAD(all_bdevs);
515 struct block_device *bdget(dev_t dev)
517 struct block_device *bdev;
520 inode = iget5_locked(blockdev_superblock, hash(dev),
521 bdev_test, bdev_set, &dev);
526 bdev = &BDEV_I(inode)->bdev;
528 if (inode->i_state & I_NEW) {
529 bdev->bd_contains = NULL;
530 bdev->bd_inode = inode;
531 bdev->bd_block_size = (1 << inode->i_blkbits);
532 bdev->bd_part_count = 0;
533 bdev->bd_invalidated = 0;
534 inode->i_mode = S_IFBLK;
536 inode->i_bdev = bdev;
537 inode->i_data.a_ops = &def_blk_aops;
538 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
539 inode->i_data.backing_dev_info = &default_backing_dev_info;
540 spin_lock(&bdev_lock);
541 list_add(&bdev->bd_list, &all_bdevs);
542 spin_unlock(&bdev_lock);
543 unlock_new_inode(inode);
548 EXPORT_SYMBOL(bdget);
551 * bdgrab -- Grab a reference to an already referenced block device
552 * @bdev: Block device to grab a reference to.
554 struct block_device *bdgrab(struct block_device *bdev)
556 atomic_inc(&bdev->bd_inode->i_count);
560 long nr_blockdev_pages(void)
562 struct block_device *bdev;
564 spin_lock(&bdev_lock);
565 list_for_each_entry(bdev, &all_bdevs, bd_list) {
566 ret += bdev->bd_inode->i_mapping->nrpages;
568 spin_unlock(&bdev_lock);
572 void bdput(struct block_device *bdev)
574 iput(bdev->bd_inode);
577 EXPORT_SYMBOL(bdput);
579 static struct block_device *bd_acquire(struct inode *inode)
581 struct block_device *bdev;
583 spin_lock(&bdev_lock);
584 bdev = inode->i_bdev;
586 atomic_inc(&bdev->bd_inode->i_count);
587 spin_unlock(&bdev_lock);
590 spin_unlock(&bdev_lock);
592 bdev = bdget(inode->i_rdev);
594 spin_lock(&bdev_lock);
595 if (!inode->i_bdev) {
597 * We take an additional bd_inode->i_count for inode,
598 * and it's released in clear_inode() of inode.
599 * So, we can access it via ->i_mapping always
602 atomic_inc(&bdev->bd_inode->i_count);
603 inode->i_bdev = bdev;
604 inode->i_mapping = bdev->bd_inode->i_mapping;
605 list_add(&inode->i_devices, &bdev->bd_inodes);
607 spin_unlock(&bdev_lock);
612 /* Call when you free inode */
614 void bd_forget(struct inode *inode)
616 struct block_device *bdev = NULL;
618 spin_lock(&bdev_lock);
620 if (!sb_is_blkdev_sb(inode->i_sb))
621 bdev = inode->i_bdev;
624 spin_unlock(&bdev_lock);
627 iput(bdev->bd_inode);
630 int bd_claim(struct block_device *bdev, void *holder)
633 spin_lock(&bdev_lock);
635 /* first decide result */
636 if (bdev->bd_holder == holder)
637 res = 0; /* already a holder */
638 else if (bdev->bd_holder != NULL)
639 res = -EBUSY; /* held by someone else */
640 else if (bdev->bd_contains == bdev)
641 res = 0; /* is a whole device which isn't held */
643 else if (bdev->bd_contains->bd_holder == bd_claim)
644 res = 0; /* is a partition of a device that is being partitioned */
645 else if (bdev->bd_contains->bd_holder != NULL)
646 res = -EBUSY; /* is a partition of a held device */
648 res = 0; /* is a partition of an un-held device */
650 /* now impose change */
652 /* note that for a whole device bd_holders
653 * will be incremented twice, and bd_holder will
654 * be set to bd_claim before being set to holder
656 bdev->bd_contains->bd_holders ++;
657 bdev->bd_contains->bd_holder = bd_claim;
659 bdev->bd_holder = holder;
661 spin_unlock(&bdev_lock);
665 EXPORT_SYMBOL(bd_claim);
667 void bd_release(struct block_device *bdev)
669 spin_lock(&bdev_lock);
670 if (!--bdev->bd_contains->bd_holders)
671 bdev->bd_contains->bd_holder = NULL;
672 if (!--bdev->bd_holders)
673 bdev->bd_holder = NULL;
674 spin_unlock(&bdev_lock);
677 EXPORT_SYMBOL(bd_release);
681 * Functions for bd_claim_by_kobject / bd_release_from_kobject
683 * If a kobject is passed to bd_claim_by_kobject()
684 * and the kobject has a parent directory,
685 * following symlinks are created:
686 * o from the kobject to the claimed bdev
687 * o from "holders" directory of the bdev to the parent of the kobject
688 * bd_release_from_kobject() removes these symlinks.
691 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
692 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
693 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
694 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
697 static int add_symlink(struct kobject *from, struct kobject *to)
701 return sysfs_create_link(from, to, kobject_name(to));
704 static void del_symlink(struct kobject *from, struct kobject *to)
708 sysfs_remove_link(from, kobject_name(to));
712 * 'struct bd_holder' contains pointers to kobjects symlinked by
713 * bd_claim_by_kobject.
714 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
717 struct list_head list; /* chain of holders of the bdev */
718 int count; /* references from the holder */
719 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
720 struct kobject *hdev; /* e.g. "/block/dm-0" */
721 struct kobject *hdir; /* e.g. "/block/sda/holders" */
722 struct kobject *sdev; /* e.g. "/block/sda" */
726 * Get references of related kobjects at once.
727 * Returns 1 on success. 0 on failure.
729 * Should call bd_holder_release_dirs() after successful use.
731 static int bd_holder_grab_dirs(struct block_device *bdev,
732 struct bd_holder *bo)
737 bo->sdir = kobject_get(bo->sdir);
741 bo->hdev = kobject_get(bo->sdir->parent);
745 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
749 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
756 kobject_put(bo->sdev);
758 kobject_put(bo->hdev);
760 kobject_put(bo->sdir);
765 /* Put references of related kobjects at once. */
766 static void bd_holder_release_dirs(struct bd_holder *bo)
768 kobject_put(bo->hdir);
769 kobject_put(bo->sdev);
770 kobject_put(bo->hdev);
771 kobject_put(bo->sdir);
774 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
776 struct bd_holder *bo;
778 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
788 static void free_bd_holder(struct bd_holder *bo)
794 * find_bd_holder - find matching struct bd_holder from the block device
796 * @bdev: struct block device to be searched
797 * @bo: target struct bd_holder
799 * Returns matching entry with @bo in @bdev->bd_holder_list.
800 * If found, increment the reference count and return the pointer.
801 * If not found, returns NULL.
803 static struct bd_holder *find_bd_holder(struct block_device *bdev,
804 struct bd_holder *bo)
806 struct bd_holder *tmp;
808 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
809 if (tmp->sdir == bo->sdir) {
818 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
820 * @bdev: block device to be bd_claimed
821 * @bo: preallocated and initialized by alloc_bd_holder()
823 * Add @bo to @bdev->bd_holder_list, create symlinks.
825 * Returns 0 if symlinks are created.
826 * Returns -ve if something fails.
828 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
835 if (!bd_holder_grab_dirs(bdev, bo))
838 err = add_symlink(bo->sdir, bo->sdev);
842 err = add_symlink(bo->hdir, bo->hdev);
844 del_symlink(bo->sdir, bo->sdev);
848 list_add_tail(&bo->list, &bdev->bd_holder_list);
853 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
855 * @bdev: block device to be bd_claimed
856 * @kobj: holder's kobject
858 * If there is matching entry with @kobj in @bdev->bd_holder_list
859 * and no other bd_claim() from the same kobject,
860 * remove the struct bd_holder from the list, delete symlinks for it.
862 * Returns a pointer to the struct bd_holder when it's removed from the list
863 * and ready to be freed.
864 * Returns NULL if matching claim isn't found or there is other bd_claim()
865 * by the same kobject.
867 static struct bd_holder *del_bd_holder(struct block_device *bdev,
868 struct kobject *kobj)
870 struct bd_holder *bo;
872 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
873 if (bo->sdir == kobj) {
875 BUG_ON(bo->count < 0);
878 del_symlink(bo->sdir, bo->sdev);
879 del_symlink(bo->hdir, bo->hdev);
880 bd_holder_release_dirs(bo);
891 * bd_claim_by_kobject - bd_claim() with additional kobject signature
893 * @bdev: block device to be claimed
894 * @holder: holder's signature
895 * @kobj: holder's kobject
897 * Do bd_claim() and if it succeeds, create sysfs symlinks between
898 * the bdev and the holder's kobject.
899 * Use bd_release_from_kobject() when relesing the claimed bdev.
901 * Returns 0 on success. (same as bd_claim())
902 * Returns errno on failure.
904 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
905 struct kobject *kobj)
908 struct bd_holder *bo, *found;
913 bo = alloc_bd_holder(kobj);
917 mutex_lock(&bdev->bd_mutex);
919 err = bd_claim(bdev, holder);
923 found = find_bd_holder(bdev, bo);
927 err = add_bd_holder(bdev, bo);
933 mutex_unlock(&bdev->bd_mutex);
939 * bd_release_from_kobject - bd_release() with additional kobject signature
941 * @bdev: block device to be released
942 * @kobj: holder's kobject
944 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
946 static void bd_release_from_kobject(struct block_device *bdev,
947 struct kobject *kobj)
952 mutex_lock(&bdev->bd_mutex);
954 free_bd_holder(del_bd_holder(bdev, kobj));
955 mutex_unlock(&bdev->bd_mutex);
959 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
961 * @bdev: block device to be claimed
962 * @holder: holder's signature
963 * @disk: holder's gendisk
965 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
967 int bd_claim_by_disk(struct block_device *bdev, void *holder,
968 struct gendisk *disk)
970 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
972 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
975 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
977 * @bdev: block device to be claimed
978 * @disk: holder's gendisk
980 * Call bd_release_from_kobject() and put @disk->slave_dir.
982 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
984 bd_release_from_kobject(bdev, disk->slave_dir);
985 kobject_put(disk->slave_dir);
987 EXPORT_SYMBOL_GPL(bd_release_from_disk);
991 * Tries to open block device by device number. Use it ONLY if you
992 * really do not have anything better - i.e. when you are behind a
993 * truly sucky interface and all you are given is a device number. _Never_
994 * to be used for internal purposes. If you ever need it - reconsider
997 struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
999 struct block_device *bdev = bdget(dev);
1002 err = blkdev_get(bdev, mode);
1003 return err ? ERR_PTR(err) : bdev;
1006 EXPORT_SYMBOL(open_by_devnum);
1009 * flush_disk - invalidates all buffer-cache entries on a disk
1011 * @bdev: struct block device to be flushed
1013 * Invalidates all buffer-cache entries on a disk. It should be called
1014 * when a disk has been changed -- either by a media change or online
1017 static void flush_disk(struct block_device *bdev)
1019 if (__invalidate_device(bdev)) {
1020 char name[BDEVNAME_SIZE] = "";
1023 disk_name(bdev->bd_disk, 0, name);
1024 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1025 "resized disk %s\n", name);
1030 if (disk_partitionable(bdev->bd_disk))
1031 bdev->bd_invalidated = 1;
1035 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1036 * @disk: struct gendisk to check
1037 * @bdev: struct bdev to adjust.
1039 * This routine checks to see if the bdev size does not match the disk size
1040 * and adjusts it if it differs.
1042 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1044 loff_t disk_size, bdev_size;
1046 disk_size = (loff_t)get_capacity(disk) << 9;
1047 bdev_size = i_size_read(bdev->bd_inode);
1048 if (disk_size != bdev_size) {
1049 char name[BDEVNAME_SIZE];
1051 disk_name(disk, 0, name);
1053 "%s: detected capacity change from %lld to %lld\n",
1054 name, bdev_size, disk_size);
1055 i_size_write(bdev->bd_inode, disk_size);
1059 EXPORT_SYMBOL(check_disk_size_change);
1062 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1063 * @disk: struct gendisk to be revalidated
1065 * This routine is a wrapper for lower-level driver's revalidate_disk
1066 * call-backs. It is used to do common pre and post operations needed
1067 * for all revalidate_disk operations.
1069 int revalidate_disk(struct gendisk *disk)
1071 struct block_device *bdev;
1074 if (disk->fops->revalidate_disk)
1075 ret = disk->fops->revalidate_disk(disk);
1077 bdev = bdget_disk(disk, 0);
1081 mutex_lock(&bdev->bd_mutex);
1082 check_disk_size_change(disk, bdev);
1083 mutex_unlock(&bdev->bd_mutex);
1087 EXPORT_SYMBOL(revalidate_disk);
1090 * This routine checks whether a removable media has been changed,
1091 * and invalidates all buffer-cache-entries in that case. This
1092 * is a relatively slow routine, so we have to try to minimize using
1093 * it. Thus it is called only upon a 'mount' or 'open'. This
1094 * is the best way of combining speed and utility, I think.
1095 * People changing diskettes in the middle of an operation deserve
1098 int check_disk_change(struct block_device *bdev)
1100 struct gendisk *disk = bdev->bd_disk;
1101 const struct block_device_operations *bdops = disk->fops;
1103 if (!bdops->media_changed)
1105 if (!bdops->media_changed(bdev->bd_disk))
1109 if (bdops->revalidate_disk)
1110 bdops->revalidate_disk(bdev->bd_disk);
1114 EXPORT_SYMBOL(check_disk_change);
1116 void bd_set_size(struct block_device *bdev, loff_t size)
1118 unsigned bsize = bdev_logical_block_size(bdev);
1120 bdev->bd_inode->i_size = size;
1121 while (bsize < PAGE_CACHE_SIZE) {
1126 bdev->bd_block_size = bsize;
1127 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1129 EXPORT_SYMBOL(bd_set_size);
1131 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1136 * mutex_lock(part->bd_mutex)
1137 * mutex_lock_nested(whole->bd_mutex, 1)
1140 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1142 struct gendisk *disk;
1147 if (mode & FMODE_READ)
1149 if (mode & FMODE_WRITE)
1152 * hooks: /n/, see "layering violations".
1154 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1164 disk = get_gendisk(bdev->bd_dev, &partno);
1166 goto out_unlock_kernel;
1168 mutex_lock_nested(&bdev->bd_mutex, for_part);
1169 if (!bdev->bd_openers) {
1170 bdev->bd_disk = disk;
1171 bdev->bd_contains = bdev;
1173 struct backing_dev_info *bdi;
1176 bdev->bd_part = disk_get_part(disk, partno);
1180 if (disk->fops->open) {
1181 ret = disk->fops->open(bdev, mode);
1182 if (ret == -ERESTARTSYS) {
1183 /* Lost a race with 'disk' being
1184 * deleted, try again.
1187 disk_put_part(bdev->bd_part);
1188 bdev->bd_part = NULL;
1189 module_put(disk->fops->owner);
1191 bdev->bd_disk = NULL;
1192 mutex_unlock(&bdev->bd_mutex);
1198 if (!bdev->bd_openers) {
1199 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1200 bdi = blk_get_backing_dev_info(bdev);
1202 bdi = &default_backing_dev_info;
1203 bdev->bd_inode->i_data.backing_dev_info = bdi;
1205 if (bdev->bd_invalidated)
1206 rescan_partitions(disk, bdev);
1208 struct block_device *whole;
1209 whole = bdget_disk(disk, 0);
1214 ret = __blkdev_get(whole, mode, 1);
1217 bdev->bd_contains = whole;
1218 bdev->bd_inode->i_data.backing_dev_info =
1219 whole->bd_inode->i_data.backing_dev_info;
1220 bdev->bd_part = disk_get_part(disk, partno);
1221 if (!(disk->flags & GENHD_FL_UP) ||
1222 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1226 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1229 module_put(disk->fops->owner);
1232 if (bdev->bd_contains == bdev) {
1233 if (bdev->bd_disk->fops->open) {
1234 ret = bdev->bd_disk->fops->open(bdev, mode);
1236 goto out_unlock_bdev;
1238 if (bdev->bd_invalidated)
1239 rescan_partitions(bdev->bd_disk, bdev);
1244 bdev->bd_part_count++;
1245 mutex_unlock(&bdev->bd_mutex);
1250 disk_put_part(bdev->bd_part);
1251 bdev->bd_disk = NULL;
1252 bdev->bd_part = NULL;
1253 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1254 if (bdev != bdev->bd_contains)
1255 __blkdev_put(bdev->bd_contains, mode, 1);
1256 bdev->bd_contains = NULL;
1258 mutex_unlock(&bdev->bd_mutex);
1263 module_put(disk->fops->owner);
1270 int blkdev_get(struct block_device *bdev, fmode_t mode)
1272 return __blkdev_get(bdev, mode, 0);
1274 EXPORT_SYMBOL(blkdev_get);
1276 static int blkdev_open(struct inode * inode, struct file * filp)
1278 struct block_device *bdev;
1282 * Preserve backwards compatibility and allow large file access
1283 * even if userspace doesn't ask for it explicitly. Some mkfs
1284 * binary needs it. We might want to drop this workaround
1285 * during an unstable branch.
1287 filp->f_flags |= O_LARGEFILE;
1289 if (filp->f_flags & O_NDELAY)
1290 filp->f_mode |= FMODE_NDELAY;
1291 if (filp->f_flags & O_EXCL)
1292 filp->f_mode |= FMODE_EXCL;
1293 if ((filp->f_flags & O_ACCMODE) == 3)
1294 filp->f_mode |= FMODE_WRITE_IOCTL;
1296 bdev = bd_acquire(inode);
1300 filp->f_mapping = bdev->bd_inode->i_mapping;
1302 res = blkdev_get(bdev, filp->f_mode);
1306 if (filp->f_mode & FMODE_EXCL) {
1307 res = bd_claim(bdev, filp);
1309 goto out_blkdev_put;
1315 blkdev_put(bdev, filp->f_mode);
1319 static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1322 struct gendisk *disk = bdev->bd_disk;
1323 struct block_device *victim = NULL;
1325 mutex_lock_nested(&bdev->bd_mutex, for_part);
1328 bdev->bd_part_count--;
1330 if (!--bdev->bd_openers) {
1331 sync_blockdev(bdev);
1334 if (bdev->bd_contains == bdev) {
1335 if (disk->fops->release)
1336 ret = disk->fops->release(disk, mode);
1338 if (!bdev->bd_openers) {
1339 struct module *owner = disk->fops->owner;
1343 disk_put_part(bdev->bd_part);
1344 bdev->bd_part = NULL;
1345 bdev->bd_disk = NULL;
1346 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1347 if (bdev != bdev->bd_contains)
1348 victim = bdev->bd_contains;
1349 bdev->bd_contains = NULL;
1352 mutex_unlock(&bdev->bd_mutex);
1355 __blkdev_put(victim, mode, 1);
1359 int blkdev_put(struct block_device *bdev, fmode_t mode)
1361 return __blkdev_put(bdev, mode, 0);
1363 EXPORT_SYMBOL(blkdev_put);
1365 static int blkdev_close(struct inode * inode, struct file * filp)
1367 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1368 if (bdev->bd_holder == filp)
1370 return blkdev_put(bdev, filp->f_mode);
1373 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1375 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1376 fmode_t mode = file->f_mode;
1379 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1380 * to updated it before every ioctl.
1382 if (file->f_flags & O_NDELAY)
1383 mode |= FMODE_NDELAY;
1385 mode &= ~FMODE_NDELAY;
1387 return blkdev_ioctl(bdev, mode, cmd, arg);
1391 * Write data to the block device. Only intended for the block device itself
1392 * and the raw driver which basically is a fake block device.
1394 * Does not take i_mutex for the write and thus is not for general purpose
1397 ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1398 unsigned long nr_segs, loff_t pos)
1400 struct file *file = iocb->ki_filp;
1403 BUG_ON(iocb->ki_pos != pos);
1405 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1406 if (ret > 0 || ret == -EIOCBQUEUED) {
1409 err = generic_write_sync(file, pos, ret);
1410 if (err < 0 && ret > 0)
1415 EXPORT_SYMBOL_GPL(blkdev_aio_write);
1418 * Try to release a page associated with block device when the system
1419 * is under memory pressure.
1421 static int blkdev_releasepage(struct page *page, gfp_t wait)
1423 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1425 if (super && super->s_op->bdev_try_to_free_page)
1426 return super->s_op->bdev_try_to_free_page(super, page, wait);
1428 return try_to_free_buffers(page);
1431 static const struct address_space_operations def_blk_aops = {
1432 .readpage = blkdev_readpage,
1433 .writepage = blkdev_writepage,
1434 .sync_page = block_sync_page,
1435 .write_begin = blkdev_write_begin,
1436 .write_end = blkdev_write_end,
1437 .writepages = generic_writepages,
1438 .releasepage = blkdev_releasepage,
1439 .direct_IO = blkdev_direct_IO,
1442 const struct file_operations def_blk_fops = {
1443 .open = blkdev_open,
1444 .release = blkdev_close,
1445 .llseek = block_llseek,
1446 .read = do_sync_read,
1447 .write = do_sync_write,
1448 .aio_read = generic_file_aio_read,
1449 .aio_write = blkdev_aio_write,
1450 .mmap = generic_file_mmap,
1451 .fsync = blkdev_fsync,
1452 .unlocked_ioctl = block_ioctl,
1453 #ifdef CONFIG_COMPAT
1454 .compat_ioctl = compat_blkdev_ioctl,
1456 .splice_read = generic_file_splice_read,
1457 .splice_write = generic_file_splice_write,
1460 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1463 mm_segment_t old_fs = get_fs();
1465 res = blkdev_ioctl(bdev, 0, cmd, arg);
1470 EXPORT_SYMBOL(ioctl_by_bdev);
1473 * lookup_bdev - lookup a struct block_device by name
1474 * @pathname: special file representing the block device
1476 * Get a reference to the blockdevice at @pathname in the current
1477 * namespace if possible and return it. Return ERR_PTR(error)
1480 struct block_device *lookup_bdev(const char *pathname)
1482 struct block_device *bdev;
1483 struct inode *inode;
1487 if (!pathname || !*pathname)
1488 return ERR_PTR(-EINVAL);
1490 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1492 return ERR_PTR(error);
1494 inode = path.dentry->d_inode;
1496 if (!S_ISBLK(inode->i_mode))
1499 if (path.mnt->mnt_flags & MNT_NODEV)
1502 bdev = bd_acquire(inode);
1509 bdev = ERR_PTR(error);
1512 EXPORT_SYMBOL(lookup_bdev);
1515 * open_bdev_exclusive - open a block device by name and set it up for use
1517 * @path: special file representing the block device
1518 * @mode: FMODE_... combination to pass be used
1519 * @holder: owner for exclusion
1521 * Open the blockdevice described by the special file at @path, claim it
1524 struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
1526 struct block_device *bdev;
1529 bdev = lookup_bdev(path);
1533 error = blkdev_get(bdev, mode);
1535 return ERR_PTR(error);
1537 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
1539 error = bd_claim(bdev, holder);
1546 blkdev_put(bdev, mode);
1547 return ERR_PTR(error);
1550 EXPORT_SYMBOL(open_bdev_exclusive);
1553 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
1555 * @bdev: blockdevice to close
1556 * @mode: mode, must match that used to open.
1558 * This is the counterpart to open_bdev_exclusive().
1560 void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
1563 blkdev_put(bdev, mode);
1566 EXPORT_SYMBOL(close_bdev_exclusive);
1568 int __invalidate_device(struct block_device *bdev)
1570 struct super_block *sb = get_super(bdev);
1575 * no need to lock the super, get_super holds the
1576 * read mutex so the filesystem cannot go away
1577 * under us (->put_super runs with the write lock
1580 shrink_dcache_sb(sb);
1581 res = invalidate_inodes(sb);
1584 invalidate_bdev(bdev);
1587 EXPORT_SYMBOL(__invalidate_device);