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/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 <asm/uaccess.h>
34 struct block_device bdev;
35 struct inode vfs_inode;
38 static const struct address_space_operations def_blk_aops;
40 static inline struct bdev_inode *BDEV_I(struct inode *inode)
42 return container_of(inode, struct bdev_inode, vfs_inode);
45 inline struct block_device *I_BDEV(struct inode *inode)
47 return &BDEV_I(inode)->bdev;
49 EXPORT_SYMBOL(I_BDEV);
51 static void bdev_write_inode(struct inode *inode)
53 spin_lock(&inode->i_lock);
54 while (inode->i_state & I_DIRTY) {
55 spin_unlock(&inode->i_lock);
56 WARN_ON_ONCE(write_inode_now(inode, true));
57 spin_lock(&inode->i_lock);
59 spin_unlock(&inode->i_lock);
62 /* Kill _all_ buffers and pagecache , dirty or not.. */
63 void kill_bdev(struct block_device *bdev)
65 struct address_space *mapping = bdev->bd_inode->i_mapping;
67 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
71 truncate_inode_pages(mapping, 0);
73 EXPORT_SYMBOL(kill_bdev);
75 /* Invalidate clean unused buffers and pagecache. */
76 void invalidate_bdev(struct block_device *bdev)
78 struct address_space *mapping = bdev->bd_inode->i_mapping;
80 if (mapping->nrpages == 0)
84 lru_add_drain_all(); /* make sure all lru add caches are flushed */
85 invalidate_mapping_pages(mapping, 0, -1);
86 /* 99% of the time, we don't need to flush the cleancache on the bdev.
87 * But, for the strange corners, lets be cautious
89 cleancache_invalidate_inode(mapping);
91 EXPORT_SYMBOL(invalidate_bdev);
93 int set_blocksize(struct block_device *bdev, int size)
95 /* Size must be a power of two, and between 512 and PAGE_SIZE */
96 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
99 /* Size cannot be smaller than the size supported by the device */
100 if (size < bdev_logical_block_size(bdev))
103 /* Don't change the size if it is same as current */
104 if (bdev->bd_block_size != size) {
106 bdev->bd_block_size = size;
107 bdev->bd_inode->i_blkbits = blksize_bits(size);
113 EXPORT_SYMBOL(set_blocksize);
115 int sb_set_blocksize(struct super_block *sb, int size)
117 if (set_blocksize(sb->s_bdev, size))
119 /* If we get here, we know size is power of two
120 * and it's value is between 512 and PAGE_SIZE */
121 sb->s_blocksize = size;
122 sb->s_blocksize_bits = blksize_bits(size);
123 return sb->s_blocksize;
126 EXPORT_SYMBOL(sb_set_blocksize);
128 int sb_min_blocksize(struct super_block *sb, int size)
130 int minsize = bdev_logical_block_size(sb->s_bdev);
133 return sb_set_blocksize(sb, size);
136 EXPORT_SYMBOL(sb_min_blocksize);
139 blkdev_get_block(struct inode *inode, sector_t iblock,
140 struct buffer_head *bh, int create)
142 bh->b_bdev = I_BDEV(inode);
143 bh->b_blocknr = iblock;
144 set_buffer_mapped(bh);
149 blkdev_direct_IO(int rw, struct kiocb *iocb, struct iov_iter *iter,
152 struct file *file = iocb->ki_filp;
153 struct inode *inode = file->f_mapping->host;
155 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iter,
156 offset, blkdev_get_block,
160 int __sync_blockdev(struct block_device *bdev, int wait)
165 return filemap_flush(bdev->bd_inode->i_mapping);
166 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
170 * Write out and wait upon all the dirty data associated with a block
171 * device via its mapping. Does not take the superblock lock.
173 int sync_blockdev(struct block_device *bdev)
175 return __sync_blockdev(bdev, 1);
177 EXPORT_SYMBOL(sync_blockdev);
180 * Write out and wait upon all dirty data associated with this
181 * device. Filesystem data as well as the underlying block
182 * device. Takes the superblock lock.
184 int fsync_bdev(struct block_device *bdev)
186 struct super_block *sb = get_super(bdev);
188 int res = sync_filesystem(sb);
192 return sync_blockdev(bdev);
194 EXPORT_SYMBOL(fsync_bdev);
197 * freeze_bdev -- lock a filesystem and force it into a consistent state
198 * @bdev: blockdevice to lock
200 * If a superblock is found on this device, we take the s_umount semaphore
201 * on it to make sure nobody unmounts until the snapshot creation is done.
202 * The reference counter (bd_fsfreeze_count) guarantees that only the last
203 * unfreeze process can unfreeze the frozen filesystem actually when multiple
204 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
205 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
208 struct super_block *freeze_bdev(struct block_device *bdev)
210 struct super_block *sb;
213 mutex_lock(&bdev->bd_fsfreeze_mutex);
214 if (++bdev->bd_fsfreeze_count > 1) {
216 * We don't even need to grab a reference - the first call
217 * to freeze_bdev grab an active reference and only the last
218 * thaw_bdev drops it.
220 sb = get_super(bdev);
222 mutex_unlock(&bdev->bd_fsfreeze_mutex);
226 sb = get_active_super(bdev);
229 if (sb->s_op->freeze_super)
230 error = sb->s_op->freeze_super(sb);
232 error = freeze_super(sb);
234 deactivate_super(sb);
235 bdev->bd_fsfreeze_count--;
236 mutex_unlock(&bdev->bd_fsfreeze_mutex);
237 return ERR_PTR(error);
239 deactivate_super(sb);
242 mutex_unlock(&bdev->bd_fsfreeze_mutex);
243 return sb; /* thaw_bdev releases s->s_umount */
245 EXPORT_SYMBOL(freeze_bdev);
248 * thaw_bdev -- unlock filesystem
249 * @bdev: blockdevice to unlock
250 * @sb: associated superblock
252 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
254 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
258 mutex_lock(&bdev->bd_fsfreeze_mutex);
259 if (!bdev->bd_fsfreeze_count)
263 if (--bdev->bd_fsfreeze_count > 0)
269 if (sb->s_op->thaw_super)
270 error = sb->s_op->thaw_super(sb);
272 error = thaw_super(sb);
274 bdev->bd_fsfreeze_count++;
275 mutex_unlock(&bdev->bd_fsfreeze_mutex);
279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
282 EXPORT_SYMBOL(thaw_bdev);
284 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
286 return block_write_full_page(page, blkdev_get_block, wbc);
289 static int blkdev_readpage(struct file * file, struct page * page)
291 return block_read_full_page(page, blkdev_get_block);
294 static int blkdev_readpages(struct file *file, struct address_space *mapping,
295 struct list_head *pages, unsigned nr_pages)
297 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
300 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
301 loff_t pos, unsigned len, unsigned flags,
302 struct page **pagep, void **fsdata)
304 return block_write_begin(mapping, pos, len, flags, pagep,
308 static int blkdev_write_end(struct file *file, struct address_space *mapping,
309 loff_t pos, unsigned len, unsigned copied,
310 struct page *page, void *fsdata)
313 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
316 page_cache_release(page);
323 * for a block special file file_inode(file)->i_size is zero
324 * so we compute the size by hand (just as in block_read/write above)
326 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
328 struct inode *bd_inode = file->f_mapping->host;
331 mutex_lock(&bd_inode->i_mutex);
332 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
333 mutex_unlock(&bd_inode->i_mutex);
337 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
339 struct inode *bd_inode = filp->f_mapping->host;
340 struct block_device *bdev = I_BDEV(bd_inode);
343 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
348 * There is no need to serialise calls to blkdev_issue_flush with
349 * i_mutex and doing so causes performance issues with concurrent
350 * O_SYNC writers to a block device.
352 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
353 if (error == -EOPNOTSUPP)
358 EXPORT_SYMBOL(blkdev_fsync);
361 * bdev_read_page() - Start reading a page from a block device
362 * @bdev: The device to read the page from
363 * @sector: The offset on the device to read the page to (need not be aligned)
364 * @page: The page to read
366 * On entry, the page should be locked. It will be unlocked when the page
367 * has been read. If the block driver implements rw_page synchronously,
368 * that will be true on exit from this function, but it need not be.
370 * Errors returned by this function are usually "soft", eg out of memory, or
371 * queue full; callers should try a different route to read this page rather
372 * than propagate an error back up the stack.
374 * Return: negative errno if an error occurs, 0 if submission was successful.
376 int bdev_read_page(struct block_device *bdev, sector_t sector,
379 const struct block_device_operations *ops = bdev->bd_disk->fops;
382 return ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
384 EXPORT_SYMBOL_GPL(bdev_read_page);
387 * bdev_write_page() - Start writing a page to a block device
388 * @bdev: The device to write the page to
389 * @sector: The offset on the device to write the page to (need not be aligned)
390 * @page: The page to write
391 * @wbc: The writeback_control for the write
393 * On entry, the page should be locked and not currently under writeback.
394 * On exit, if the write started successfully, the page will be unlocked and
395 * under writeback. If the write failed already (eg the driver failed to
396 * queue the page to the device), the page will still be locked. If the
397 * caller is a ->writepage implementation, it will need to unlock the page.
399 * Errors returned by this function are usually "soft", eg out of memory, or
400 * queue full; callers should try a different route to write this page rather
401 * than propagate an error back up the stack.
403 * Return: negative errno if an error occurs, 0 if submission was successful.
405 int bdev_write_page(struct block_device *bdev, sector_t sector,
406 struct page *page, struct writeback_control *wbc)
409 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
410 const struct block_device_operations *ops = bdev->bd_disk->fops;
413 set_page_writeback(page);
414 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
416 end_page_writeback(page);
421 EXPORT_SYMBOL_GPL(bdev_write_page);
424 * bdev_direct_access() - Get the address for directly-accessibly memory
425 * @bdev: The device containing the memory
426 * @sector: The offset within the device
427 * @addr: Where to put the address of the memory
428 * @pfn: The Page Frame Number for the memory
429 * @size: The number of bytes requested
431 * If a block device is made up of directly addressable memory, this function
432 * will tell the caller the PFN and the address of the memory. The address
433 * may be directly dereferenced within the kernel without the need to call
434 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
437 * Return: negative errno if an error occurs, otherwise the number of bytes
438 * accessible at this address.
440 long bdev_direct_access(struct block_device *bdev, sector_t sector,
441 void **addr, unsigned long *pfn, long size)
444 const struct block_device_operations *ops = bdev->bd_disk->fops;
448 if (!ops->direct_access)
450 if ((sector + DIV_ROUND_UP(size, 512)) >
451 part_nr_sects_read(bdev->bd_part))
453 sector += get_start_sect(bdev);
454 if (sector % (PAGE_SIZE / 512))
456 avail = ops->direct_access(bdev, sector, addr, pfn, size);
459 return min(avail, size);
461 EXPORT_SYMBOL_GPL(bdev_direct_access);
467 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
468 static struct kmem_cache * bdev_cachep __read_mostly;
470 static struct inode *bdev_alloc_inode(struct super_block *sb)
472 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
475 return &ei->vfs_inode;
478 static void bdev_i_callback(struct rcu_head *head)
480 struct inode *inode = container_of(head, struct inode, i_rcu);
481 struct bdev_inode *bdi = BDEV_I(inode);
483 kmem_cache_free(bdev_cachep, bdi);
486 static void bdev_destroy_inode(struct inode *inode)
488 call_rcu(&inode->i_rcu, bdev_i_callback);
491 static void init_once(void *foo)
493 struct bdev_inode *ei = (struct bdev_inode *) foo;
494 struct block_device *bdev = &ei->bdev;
496 memset(bdev, 0, sizeof(*bdev));
497 mutex_init(&bdev->bd_mutex);
498 INIT_LIST_HEAD(&bdev->bd_inodes);
499 INIT_LIST_HEAD(&bdev->bd_list);
501 INIT_LIST_HEAD(&bdev->bd_holder_disks);
503 inode_init_once(&ei->vfs_inode);
504 /* Initialize mutex for freeze. */
505 mutex_init(&bdev->bd_fsfreeze_mutex);
508 static inline void __bd_forget(struct inode *inode)
510 list_del_init(&inode->i_devices);
511 inode->i_bdev = NULL;
512 inode->i_mapping = &inode->i_data;
515 static void bdev_evict_inode(struct inode *inode)
517 struct block_device *bdev = &BDEV_I(inode)->bdev;
519 truncate_inode_pages_final(&inode->i_data);
520 invalidate_inode_buffers(inode); /* is it needed here? */
522 spin_lock(&bdev_lock);
523 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
524 __bd_forget(list_entry(p, struct inode, i_devices));
526 list_del_init(&bdev->bd_list);
527 spin_unlock(&bdev_lock);
530 static const struct super_operations bdev_sops = {
531 .statfs = simple_statfs,
532 .alloc_inode = bdev_alloc_inode,
533 .destroy_inode = bdev_destroy_inode,
534 .drop_inode = generic_delete_inode,
535 .evict_inode = bdev_evict_inode,
538 static struct dentry *bd_mount(struct file_system_type *fs_type,
539 int flags, const char *dev_name, void *data)
541 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
544 static struct file_system_type bd_type = {
547 .kill_sb = kill_anon_super,
550 static struct super_block *blockdev_superblock __read_mostly;
552 void __init bdev_cache_init(void)
555 static struct vfsmount *bd_mnt;
557 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
558 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
559 SLAB_MEM_SPREAD|SLAB_PANIC),
561 err = register_filesystem(&bd_type);
563 panic("Cannot register bdev pseudo-fs");
564 bd_mnt = kern_mount(&bd_type);
566 panic("Cannot create bdev pseudo-fs");
567 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
571 * Most likely _very_ bad one - but then it's hardly critical for small
572 * /dev and can be fixed when somebody will need really large one.
573 * Keep in mind that it will be fed through icache hash function too.
575 static inline unsigned long hash(dev_t dev)
577 return MAJOR(dev)+MINOR(dev);
580 static int bdev_test(struct inode *inode, void *data)
582 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
585 static int bdev_set(struct inode *inode, void *data)
587 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
591 static LIST_HEAD(all_bdevs);
593 struct block_device *bdget(dev_t dev)
595 struct block_device *bdev;
598 inode = iget5_locked(blockdev_superblock, hash(dev),
599 bdev_test, bdev_set, &dev);
604 bdev = &BDEV_I(inode)->bdev;
606 if (inode->i_state & I_NEW) {
607 bdev->bd_contains = NULL;
608 bdev->bd_super = NULL;
609 bdev->bd_inode = inode;
610 bdev->bd_block_size = (1 << inode->i_blkbits);
611 bdev->bd_part_count = 0;
612 bdev->bd_invalidated = 0;
613 inode->i_mode = S_IFBLK;
615 inode->i_bdev = bdev;
616 inode->i_data.a_ops = &def_blk_aops;
617 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
618 spin_lock(&bdev_lock);
619 list_add(&bdev->bd_list, &all_bdevs);
620 spin_unlock(&bdev_lock);
621 unlock_new_inode(inode);
626 EXPORT_SYMBOL(bdget);
629 * bdgrab -- Grab a reference to an already referenced block device
630 * @bdev: Block device to grab a reference to.
632 struct block_device *bdgrab(struct block_device *bdev)
634 ihold(bdev->bd_inode);
637 EXPORT_SYMBOL(bdgrab);
639 long nr_blockdev_pages(void)
641 struct block_device *bdev;
643 spin_lock(&bdev_lock);
644 list_for_each_entry(bdev, &all_bdevs, bd_list) {
645 ret += bdev->bd_inode->i_mapping->nrpages;
647 spin_unlock(&bdev_lock);
651 void bdput(struct block_device *bdev)
653 iput(bdev->bd_inode);
656 EXPORT_SYMBOL(bdput);
658 static struct block_device *bd_acquire(struct inode *inode)
660 struct block_device *bdev;
662 spin_lock(&bdev_lock);
663 bdev = inode->i_bdev;
665 ihold(bdev->bd_inode);
666 spin_unlock(&bdev_lock);
669 spin_unlock(&bdev_lock);
671 bdev = bdget(inode->i_rdev);
673 spin_lock(&bdev_lock);
674 if (!inode->i_bdev) {
676 * We take an additional reference to bd_inode,
677 * and it's released in clear_inode() of inode.
678 * So, we can access it via ->i_mapping always
681 ihold(bdev->bd_inode);
682 inode->i_bdev = bdev;
683 inode->i_mapping = bdev->bd_inode->i_mapping;
684 list_add(&inode->i_devices, &bdev->bd_inodes);
686 spin_unlock(&bdev_lock);
691 int sb_is_blkdev_sb(struct super_block *sb)
693 return sb == blockdev_superblock;
696 /* Call when you free inode */
698 void bd_forget(struct inode *inode)
700 struct block_device *bdev = NULL;
702 spin_lock(&bdev_lock);
703 if (!sb_is_blkdev_sb(inode->i_sb))
704 bdev = inode->i_bdev;
706 spin_unlock(&bdev_lock);
709 iput(bdev->bd_inode);
713 * bd_may_claim - test whether a block device can be claimed
714 * @bdev: block device of interest
715 * @whole: whole block device containing @bdev, may equal @bdev
716 * @holder: holder trying to claim @bdev
718 * Test whether @bdev can be claimed by @holder.
721 * spin_lock(&bdev_lock).
724 * %true if @bdev can be claimed, %false otherwise.
726 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
729 if (bdev->bd_holder == holder)
730 return true; /* already a holder */
731 else if (bdev->bd_holder != NULL)
732 return false; /* held by someone else */
733 else if (bdev->bd_contains == bdev)
734 return true; /* is a whole device which isn't held */
736 else if (whole->bd_holder == bd_may_claim)
737 return true; /* is a partition of a device that is being partitioned */
738 else if (whole->bd_holder != NULL)
739 return false; /* is a partition of a held device */
741 return true; /* is a partition of an un-held device */
745 * bd_prepare_to_claim - prepare to claim a block device
746 * @bdev: block device of interest
747 * @whole: the whole device containing @bdev, may equal @bdev
748 * @holder: holder trying to claim @bdev
750 * Prepare to claim @bdev. This function fails if @bdev is already
751 * claimed by another holder and waits if another claiming is in
752 * progress. This function doesn't actually claim. On successful
753 * return, the caller has ownership of bd_claiming and bd_holder[s].
756 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
760 * 0 if @bdev can be claimed, -EBUSY otherwise.
762 static int bd_prepare_to_claim(struct block_device *bdev,
763 struct block_device *whole, void *holder)
766 /* if someone else claimed, fail */
767 if (!bd_may_claim(bdev, whole, holder))
770 /* if claiming is already in progress, wait for it to finish */
771 if (whole->bd_claiming) {
772 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
775 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
776 spin_unlock(&bdev_lock);
778 finish_wait(wq, &wait);
779 spin_lock(&bdev_lock);
788 * bd_start_claiming - start claiming a block device
789 * @bdev: block device of interest
790 * @holder: holder trying to claim @bdev
792 * @bdev is about to be opened exclusively. Check @bdev can be opened
793 * exclusively and mark that an exclusive open is in progress. Each
794 * successful call to this function must be matched with a call to
795 * either bd_finish_claiming() or bd_abort_claiming() (which do not
798 * This function is used to gain exclusive access to the block device
799 * without actually causing other exclusive open attempts to fail. It
800 * should be used when the open sequence itself requires exclusive
801 * access but may subsequently fail.
807 * Pointer to the block device containing @bdev on success, ERR_PTR()
810 static struct block_device *bd_start_claiming(struct block_device *bdev,
813 struct gendisk *disk;
814 struct block_device *whole;
820 * @bdev might not have been initialized properly yet, look up
821 * and grab the outer block device the hard way.
823 disk = get_gendisk(bdev->bd_dev, &partno);
825 return ERR_PTR(-ENXIO);
828 * Normally, @bdev should equal what's returned from bdget_disk()
829 * if partno is 0; however, some drivers (floppy) use multiple
830 * bdev's for the same physical device and @bdev may be one of the
831 * aliases. Keep @bdev if partno is 0. This means claimer
832 * tracking is broken for those devices but it has always been that
836 whole = bdget_disk(disk, 0);
838 whole = bdgrab(bdev);
840 module_put(disk->fops->owner);
843 return ERR_PTR(-ENOMEM);
845 /* prepare to claim, if successful, mark claiming in progress */
846 spin_lock(&bdev_lock);
848 err = bd_prepare_to_claim(bdev, whole, holder);
850 whole->bd_claiming = holder;
851 spin_unlock(&bdev_lock);
854 spin_unlock(&bdev_lock);
861 struct bd_holder_disk {
862 struct list_head list;
863 struct gendisk *disk;
867 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
868 struct gendisk *disk)
870 struct bd_holder_disk *holder;
872 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
873 if (holder->disk == disk)
878 static int add_symlink(struct kobject *from, struct kobject *to)
880 return sysfs_create_link(from, to, kobject_name(to));
883 static void del_symlink(struct kobject *from, struct kobject *to)
885 sysfs_remove_link(from, kobject_name(to));
889 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
890 * @bdev: the claimed slave bdev
891 * @disk: the holding disk
893 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
895 * This functions creates the following sysfs symlinks.
897 * - from "slaves" directory of the holder @disk to the claimed @bdev
898 * - from "holders" directory of the @bdev to the holder @disk
900 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
901 * passed to bd_link_disk_holder(), then:
903 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
904 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
906 * The caller must have claimed @bdev before calling this function and
907 * ensure that both @bdev and @disk are valid during the creation and
908 * lifetime of these symlinks.
914 * 0 on success, -errno on failure.
916 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
918 struct bd_holder_disk *holder;
921 mutex_lock(&bdev->bd_mutex);
923 WARN_ON_ONCE(!bdev->bd_holder);
925 /* FIXME: remove the following once add_disk() handles errors */
926 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
929 holder = bd_find_holder_disk(bdev, disk);
935 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
941 INIT_LIST_HEAD(&holder->list);
945 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
949 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
953 * bdev could be deleted beneath us which would implicitly destroy
954 * the holder directory. Hold on to it.
956 kobject_get(bdev->bd_part->holder_dir);
958 list_add(&holder->list, &bdev->bd_holder_disks);
962 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
966 mutex_unlock(&bdev->bd_mutex);
969 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
972 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
973 * @bdev: the calimed slave bdev
974 * @disk: the holding disk
976 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
981 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
983 struct bd_holder_disk *holder;
985 mutex_lock(&bdev->bd_mutex);
987 holder = bd_find_holder_disk(bdev, disk);
989 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
990 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
991 del_symlink(bdev->bd_part->holder_dir,
992 &disk_to_dev(disk)->kobj);
993 kobject_put(bdev->bd_part->holder_dir);
994 list_del_init(&holder->list);
998 mutex_unlock(&bdev->bd_mutex);
1000 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1004 * flush_disk - invalidates all buffer-cache entries on a disk
1006 * @bdev: struct block device to be flushed
1007 * @kill_dirty: flag to guide handling of dirty inodes
1009 * Invalidates all buffer-cache entries on a disk. It should be called
1010 * when a disk has been changed -- either by a media change or online
1013 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1015 if (__invalidate_device(bdev, kill_dirty)) {
1016 char name[BDEVNAME_SIZE] = "";
1019 disk_name(bdev->bd_disk, 0, name);
1020 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1021 "resized disk %s\n", name);
1026 if (disk_part_scan_enabled(bdev->bd_disk))
1027 bdev->bd_invalidated = 1;
1031 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1032 * @disk: struct gendisk to check
1033 * @bdev: struct bdev to adjust.
1035 * This routine checks to see if the bdev size does not match the disk size
1036 * and adjusts it if it differs.
1038 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1040 loff_t disk_size, bdev_size;
1042 disk_size = (loff_t)get_capacity(disk) << 9;
1043 bdev_size = i_size_read(bdev->bd_inode);
1044 if (disk_size != bdev_size) {
1045 char name[BDEVNAME_SIZE];
1047 disk_name(disk, 0, name);
1049 "%s: detected capacity change from %lld to %lld\n",
1050 name, bdev_size, disk_size);
1051 i_size_write(bdev->bd_inode, disk_size);
1052 flush_disk(bdev, false);
1055 EXPORT_SYMBOL(check_disk_size_change);
1058 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1059 * @disk: struct gendisk to be revalidated
1061 * This routine is a wrapper for lower-level driver's revalidate_disk
1062 * call-backs. It is used to do common pre and post operations needed
1063 * for all revalidate_disk operations.
1065 int revalidate_disk(struct gendisk *disk)
1067 struct block_device *bdev;
1070 if (disk->fops->revalidate_disk)
1071 ret = disk->fops->revalidate_disk(disk);
1073 bdev = bdget_disk(disk, 0);
1077 mutex_lock(&bdev->bd_mutex);
1078 check_disk_size_change(disk, bdev);
1079 bdev->bd_invalidated = 0;
1080 mutex_unlock(&bdev->bd_mutex);
1084 EXPORT_SYMBOL(revalidate_disk);
1087 * This routine checks whether a removable media has been changed,
1088 * and invalidates all buffer-cache-entries in that case. This
1089 * is a relatively slow routine, so we have to try to minimize using
1090 * it. Thus it is called only upon a 'mount' or 'open'. This
1091 * is the best way of combining speed and utility, I think.
1092 * People changing diskettes in the middle of an operation deserve
1095 int check_disk_change(struct block_device *bdev)
1097 struct gendisk *disk = bdev->bd_disk;
1098 const struct block_device_operations *bdops = disk->fops;
1099 unsigned int events;
1101 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1102 DISK_EVENT_EJECT_REQUEST);
1103 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1106 flush_disk(bdev, true);
1107 if (bdops->revalidate_disk)
1108 bdops->revalidate_disk(bdev->bd_disk);
1112 EXPORT_SYMBOL(check_disk_change);
1114 void bd_set_size(struct block_device *bdev, loff_t size)
1116 unsigned bsize = bdev_logical_block_size(bdev);
1118 mutex_lock(&bdev->bd_inode->i_mutex);
1119 i_size_write(bdev->bd_inode, size);
1120 mutex_unlock(&bdev->bd_inode->i_mutex);
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 void __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;
1143 struct module *owner;
1148 if (mode & FMODE_READ)
1150 if (mode & FMODE_WRITE)
1153 * hooks: /n/, see "layering violations".
1156 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1166 disk = get_gendisk(bdev->bd_dev, &partno);
1169 owner = disk->fops->owner;
1171 disk_block_events(disk);
1172 mutex_lock_nested(&bdev->bd_mutex, for_part);
1173 if (!bdev->bd_openers) {
1174 bdev->bd_disk = disk;
1175 bdev->bd_queue = disk->queue;
1176 bdev->bd_contains = bdev;
1179 bdev->bd_part = disk_get_part(disk, partno);
1184 if (disk->fops->open) {
1185 ret = disk->fops->open(bdev, mode);
1186 if (ret == -ERESTARTSYS) {
1187 /* Lost a race with 'disk' being
1188 * deleted, try again.
1191 disk_put_part(bdev->bd_part);
1192 bdev->bd_part = NULL;
1193 bdev->bd_disk = NULL;
1194 bdev->bd_queue = NULL;
1195 mutex_unlock(&bdev->bd_mutex);
1196 disk_unblock_events(disk);
1204 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1207 * If the device is invalidated, rescan partition
1208 * if open succeeded or failed with -ENOMEDIUM.
1209 * The latter is necessary to prevent ghost
1210 * partitions on a removed medium.
1212 if (bdev->bd_invalidated) {
1214 rescan_partitions(disk, bdev);
1215 else if (ret == -ENOMEDIUM)
1216 invalidate_partitions(disk, bdev);
1221 struct block_device *whole;
1222 whole = bdget_disk(disk, 0);
1227 ret = __blkdev_get(whole, mode, 1);
1230 bdev->bd_contains = whole;
1231 bdev->bd_part = disk_get_part(disk, partno);
1232 if (!(disk->flags & GENHD_FL_UP) ||
1233 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1237 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1240 if (bdev->bd_contains == bdev) {
1242 if (bdev->bd_disk->fops->open)
1243 ret = bdev->bd_disk->fops->open(bdev, mode);
1244 /* the same as first opener case, read comment there */
1245 if (bdev->bd_invalidated) {
1247 rescan_partitions(bdev->bd_disk, bdev);
1248 else if (ret == -ENOMEDIUM)
1249 invalidate_partitions(bdev->bd_disk, bdev);
1252 goto out_unlock_bdev;
1254 /* only one opener holds refs to the module and disk */
1260 bdev->bd_part_count++;
1261 mutex_unlock(&bdev->bd_mutex);
1262 disk_unblock_events(disk);
1266 disk_put_part(bdev->bd_part);
1267 bdev->bd_disk = NULL;
1268 bdev->bd_part = NULL;
1269 bdev->bd_queue = NULL;
1270 if (bdev != bdev->bd_contains)
1271 __blkdev_put(bdev->bd_contains, mode, 1);
1272 bdev->bd_contains = NULL;
1274 mutex_unlock(&bdev->bd_mutex);
1275 disk_unblock_events(disk);
1285 * blkdev_get - open a block device
1286 * @bdev: block_device to open
1287 * @mode: FMODE_* mask
1288 * @holder: exclusive holder identifier
1290 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1291 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1292 * @holder is invalid. Exclusive opens may nest for the same @holder.
1294 * On success, the reference count of @bdev is unchanged. On failure,
1301 * 0 on success, -errno on failure.
1303 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1305 struct block_device *whole = NULL;
1308 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1310 if ((mode & FMODE_EXCL) && holder) {
1311 whole = bd_start_claiming(bdev, holder);
1312 if (IS_ERR(whole)) {
1314 return PTR_ERR(whole);
1318 res = __blkdev_get(bdev, mode, 0);
1321 struct gendisk *disk = whole->bd_disk;
1323 /* finish claiming */
1324 mutex_lock(&bdev->bd_mutex);
1325 spin_lock(&bdev_lock);
1328 BUG_ON(!bd_may_claim(bdev, whole, holder));
1330 * Note that for a whole device bd_holders
1331 * will be incremented twice, and bd_holder
1332 * will be set to bd_may_claim before being
1335 whole->bd_holders++;
1336 whole->bd_holder = bd_may_claim;
1338 bdev->bd_holder = holder;
1341 /* tell others that we're done */
1342 BUG_ON(whole->bd_claiming != holder);
1343 whole->bd_claiming = NULL;
1344 wake_up_bit(&whole->bd_claiming, 0);
1346 spin_unlock(&bdev_lock);
1349 * Block event polling for write claims if requested. Any
1350 * write holder makes the write_holder state stick until
1351 * all are released. This is good enough and tracking
1352 * individual writeable reference is too fragile given the
1353 * way @mode is used in blkdev_get/put().
1355 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1356 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1357 bdev->bd_write_holder = true;
1358 disk_block_events(disk);
1361 mutex_unlock(&bdev->bd_mutex);
1367 EXPORT_SYMBOL(blkdev_get);
1370 * blkdev_get_by_path - open a block device by name
1371 * @path: path to the block device to open
1372 * @mode: FMODE_* mask
1373 * @holder: exclusive holder identifier
1375 * Open the blockdevice described by the device file at @path. @mode
1376 * and @holder are identical to blkdev_get().
1378 * On success, the returned block_device has reference count of one.
1384 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1386 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1389 struct block_device *bdev;
1392 bdev = lookup_bdev(path);
1396 err = blkdev_get(bdev, mode, holder);
1398 return ERR_PTR(err);
1400 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1401 blkdev_put(bdev, mode);
1402 return ERR_PTR(-EACCES);
1407 EXPORT_SYMBOL(blkdev_get_by_path);
1410 * blkdev_get_by_dev - open a block device by device number
1411 * @dev: device number of block device to open
1412 * @mode: FMODE_* mask
1413 * @holder: exclusive holder identifier
1415 * Open the blockdevice described by device number @dev. @mode and
1416 * @holder are identical to blkdev_get().
1418 * Use it ONLY if you really do not have anything better - i.e. when
1419 * you are behind a truly sucky interface and all you are given is a
1420 * device number. _Never_ to be used for internal purposes. If you
1421 * ever need it - reconsider your API.
1423 * On success, the returned block_device has reference count of one.
1429 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1431 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1433 struct block_device *bdev;
1438 return ERR_PTR(-ENOMEM);
1440 err = blkdev_get(bdev, mode, holder);
1442 return ERR_PTR(err);
1446 EXPORT_SYMBOL(blkdev_get_by_dev);
1448 static int blkdev_open(struct inode * inode, struct file * filp)
1450 struct block_device *bdev;
1453 * Preserve backwards compatibility and allow large file access
1454 * even if userspace doesn't ask for it explicitly. Some mkfs
1455 * binary needs it. We might want to drop this workaround
1456 * during an unstable branch.
1458 filp->f_flags |= O_LARGEFILE;
1460 if (filp->f_flags & O_NDELAY)
1461 filp->f_mode |= FMODE_NDELAY;
1462 if (filp->f_flags & O_EXCL)
1463 filp->f_mode |= FMODE_EXCL;
1464 if ((filp->f_flags & O_ACCMODE) == 3)
1465 filp->f_mode |= FMODE_WRITE_IOCTL;
1467 bdev = bd_acquire(inode);
1471 filp->f_mapping = bdev->bd_inode->i_mapping;
1473 return blkdev_get(bdev, filp->f_mode, filp);
1476 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1478 struct gendisk *disk = bdev->bd_disk;
1479 struct block_device *victim = NULL;
1481 mutex_lock_nested(&bdev->bd_mutex, for_part);
1483 bdev->bd_part_count--;
1485 if (!--bdev->bd_openers) {
1486 WARN_ON_ONCE(bdev->bd_holders);
1487 sync_blockdev(bdev);
1490 * ->release can cause the queue to disappear, so flush all
1491 * dirty data before.
1493 bdev_write_inode(bdev->bd_inode);
1495 if (bdev->bd_contains == bdev) {
1496 if (disk->fops->release)
1497 disk->fops->release(disk, mode);
1499 if (!bdev->bd_openers) {
1500 struct module *owner = disk->fops->owner;
1502 disk_put_part(bdev->bd_part);
1503 bdev->bd_part = NULL;
1504 bdev->bd_disk = NULL;
1505 if (bdev != bdev->bd_contains)
1506 victim = bdev->bd_contains;
1507 bdev->bd_contains = NULL;
1512 mutex_unlock(&bdev->bd_mutex);
1515 __blkdev_put(victim, mode, 1);
1518 void blkdev_put(struct block_device *bdev, fmode_t mode)
1520 mutex_lock(&bdev->bd_mutex);
1522 if (mode & FMODE_EXCL) {
1526 * Release a claim on the device. The holder fields
1527 * are protected with bdev_lock. bd_mutex is to
1528 * synchronize disk_holder unlinking.
1530 spin_lock(&bdev_lock);
1532 WARN_ON_ONCE(--bdev->bd_holders < 0);
1533 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1535 /* bd_contains might point to self, check in a separate step */
1536 if ((bdev_free = !bdev->bd_holders))
1537 bdev->bd_holder = NULL;
1538 if (!bdev->bd_contains->bd_holders)
1539 bdev->bd_contains->bd_holder = NULL;
1541 spin_unlock(&bdev_lock);
1544 * If this was the last claim, remove holder link and
1545 * unblock evpoll if it was a write holder.
1547 if (bdev_free && bdev->bd_write_holder) {
1548 disk_unblock_events(bdev->bd_disk);
1549 bdev->bd_write_holder = false;
1554 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1555 * event. This is to ensure detection of media removal commanded
1556 * from userland - e.g. eject(1).
1558 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1560 mutex_unlock(&bdev->bd_mutex);
1562 __blkdev_put(bdev, mode, 0);
1564 EXPORT_SYMBOL(blkdev_put);
1566 static int blkdev_close(struct inode * inode, struct file * filp)
1568 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1569 blkdev_put(bdev, filp->f_mode);
1573 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1575 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1576 fmode_t mode = file->f_mode;
1579 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1580 * to updated it before every ioctl.
1582 if (file->f_flags & O_NDELAY)
1583 mode |= FMODE_NDELAY;
1585 mode &= ~FMODE_NDELAY;
1587 return blkdev_ioctl(bdev, mode, cmd, arg);
1591 * Write data to the block device. Only intended for the block device itself
1592 * and the raw driver which basically is a fake block device.
1594 * Does not take i_mutex for the write and thus is not for general purpose
1597 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1599 struct file *file = iocb->ki_filp;
1600 struct blk_plug plug;
1603 blk_start_plug(&plug);
1604 ret = __generic_file_write_iter(iocb, from);
1607 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1611 blk_finish_plug(&plug);
1614 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1616 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1618 struct file *file = iocb->ki_filp;
1619 struct inode *bd_inode = file->f_mapping->host;
1620 loff_t size = i_size_read(bd_inode);
1621 loff_t pos = iocb->ki_pos;
1627 iov_iter_truncate(to, size);
1628 return generic_file_read_iter(iocb, to);
1630 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1633 * Try to release a page associated with block device when the system
1634 * is under memory pressure.
1636 static int blkdev_releasepage(struct page *page, gfp_t wait)
1638 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1640 if (super && super->s_op->bdev_try_to_free_page)
1641 return super->s_op->bdev_try_to_free_page(super, page, wait);
1643 return try_to_free_buffers(page);
1646 static const struct address_space_operations def_blk_aops = {
1647 .readpage = blkdev_readpage,
1648 .readpages = blkdev_readpages,
1649 .writepage = blkdev_writepage,
1650 .write_begin = blkdev_write_begin,
1651 .write_end = blkdev_write_end,
1652 .writepages = generic_writepages,
1653 .releasepage = blkdev_releasepage,
1654 .direct_IO = blkdev_direct_IO,
1655 .is_dirty_writeback = buffer_check_dirty_writeback,
1658 const struct file_operations def_blk_fops = {
1659 .open = blkdev_open,
1660 .release = blkdev_close,
1661 .llseek = block_llseek,
1662 .read_iter = blkdev_read_iter,
1663 .write_iter = blkdev_write_iter,
1664 .mmap = generic_file_mmap,
1665 .fsync = blkdev_fsync,
1666 .unlocked_ioctl = block_ioctl,
1667 #ifdef CONFIG_COMPAT
1668 .compat_ioctl = compat_blkdev_ioctl,
1670 .splice_read = generic_file_splice_read,
1671 .splice_write = iter_file_splice_write,
1674 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1677 mm_segment_t old_fs = get_fs();
1679 res = blkdev_ioctl(bdev, 0, cmd, arg);
1684 EXPORT_SYMBOL(ioctl_by_bdev);
1687 * lookup_bdev - lookup a struct block_device by name
1688 * @pathname: special file representing the block device
1690 * Get a reference to the blockdevice at @pathname in the current
1691 * namespace if possible and return it. Return ERR_PTR(error)
1694 struct block_device *lookup_bdev(const char *pathname)
1696 struct block_device *bdev;
1697 struct inode *inode;
1701 if (!pathname || !*pathname)
1702 return ERR_PTR(-EINVAL);
1704 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1706 return ERR_PTR(error);
1708 inode = path.dentry->d_inode;
1710 if (!S_ISBLK(inode->i_mode))
1713 if (path.mnt->mnt_flags & MNT_NODEV)
1716 bdev = bd_acquire(inode);
1723 bdev = ERR_PTR(error);
1726 EXPORT_SYMBOL(lookup_bdev);
1728 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1730 struct super_block *sb = get_super(bdev);
1735 * no need to lock the super, get_super holds the
1736 * read mutex so the filesystem cannot go away
1737 * under us (->put_super runs with the write lock
1740 shrink_dcache_sb(sb);
1741 res = invalidate_inodes(sb, kill_dirty);
1744 invalidate_bdev(bdev);
1747 EXPORT_SYMBOL(__invalidate_device);
1749 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1751 struct inode *inode, *old_inode = NULL;
1753 spin_lock(&inode_sb_list_lock);
1754 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1755 struct address_space *mapping = inode->i_mapping;
1757 spin_lock(&inode->i_lock);
1758 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1759 mapping->nrpages == 0) {
1760 spin_unlock(&inode->i_lock);
1764 spin_unlock(&inode->i_lock);
1765 spin_unlock(&inode_sb_list_lock);
1767 * We hold a reference to 'inode' so it couldn't have been
1768 * removed from s_inodes list while we dropped the
1769 * inode_sb_list_lock. We cannot iput the inode now as we can
1770 * be holding the last reference and we cannot iput it under
1771 * inode_sb_list_lock. So we keep the reference and iput it
1777 func(I_BDEV(inode), arg);
1779 spin_lock(&inode_sb_list_lock);
1781 spin_unlock(&inode_sb_list_lock);