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/backing-dev.h>
18 #include <linux/module.h>
19 #include <linux/blkpg.h>
20 #include <linux/magic.h>
21 #include <linux/buffer_head.h>
22 #include <linux/swap.h>
23 #include <linux/pagevec.h>
24 #include <linux/writeback.h>
25 #include <linux/mpage.h>
26 #include <linux/mount.h>
27 #include <linux/uio.h>
28 #include <linux/namei.h>
29 #include <linux/log2.h>
30 #include <linux/cleancache.h>
31 #include <linux/dax.h>
32 #include <asm/uaccess.h>
36 struct block_device bdev;
37 struct inode vfs_inode;
40 static const struct address_space_operations def_blk_aops;
42 static inline struct bdev_inode *BDEV_I(struct inode *inode)
44 return container_of(inode, struct bdev_inode, vfs_inode);
47 struct block_device *I_BDEV(struct inode *inode)
49 return &BDEV_I(inode)->bdev;
51 EXPORT_SYMBOL(I_BDEV);
53 void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
61 printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
65 static void bdev_write_inode(struct block_device *bdev)
67 struct inode *inode = bdev->bd_inode;
70 spin_lock(&inode->i_lock);
71 while (inode->i_state & I_DIRTY) {
72 spin_unlock(&inode->i_lock);
73 ret = write_inode_now(inode, true);
75 char name[BDEVNAME_SIZE];
76 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
77 "for block device %s (err=%d).\n",
78 bdevname(bdev, name), ret);
80 spin_lock(&inode->i_lock);
82 spin_unlock(&inode->i_lock);
85 /* Kill _all_ buffers and pagecache , dirty or not.. */
86 void kill_bdev(struct block_device *bdev)
88 struct address_space *mapping = bdev->bd_inode->i_mapping;
90 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
94 truncate_inode_pages(mapping, 0);
96 EXPORT_SYMBOL(kill_bdev);
98 /* Invalidate clean unused buffers and pagecache. */
99 void invalidate_bdev(struct block_device *bdev)
101 struct address_space *mapping = bdev->bd_inode->i_mapping;
103 if (mapping->nrpages == 0)
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 /* 99% of the time, we don't need to flush the cleancache on the bdev.
110 * But, for the strange corners, lets be cautious
112 cleancache_invalidate_inode(mapping);
114 EXPORT_SYMBOL(invalidate_bdev);
116 int set_blocksize(struct block_device *bdev, int size)
118 /* Size must be a power of two, and between 512 and PAGE_SIZE */
119 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
122 /* Size cannot be smaller than the size supported by the device */
123 if (size < bdev_logical_block_size(bdev))
126 /* Don't change the size if it is same as current */
127 if (bdev->bd_block_size != size) {
129 bdev->bd_block_size = size;
130 bdev->bd_inode->i_blkbits = blksize_bits(size);
136 EXPORT_SYMBOL(set_blocksize);
138 int sb_set_blocksize(struct super_block *sb, int size)
140 if (set_blocksize(sb->s_bdev, size))
142 /* If we get here, we know size is power of two
143 * and it's value is between 512 and PAGE_SIZE */
144 sb->s_blocksize = size;
145 sb->s_blocksize_bits = blksize_bits(size);
146 return sb->s_blocksize;
149 EXPORT_SYMBOL(sb_set_blocksize);
151 int sb_min_blocksize(struct super_block *sb, int size)
153 int minsize = bdev_logical_block_size(sb->s_bdev);
156 return sb_set_blocksize(sb, size);
159 EXPORT_SYMBOL(sb_min_blocksize);
162 blkdev_get_block(struct inode *inode, sector_t iblock,
163 struct buffer_head *bh, int create)
165 bh->b_bdev = I_BDEV(inode);
166 bh->b_blocknr = iblock;
167 set_buffer_mapped(bh);
171 static struct inode *bdev_file_inode(struct file *file)
173 return file->f_mapping->host;
177 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, loff_t offset)
179 struct file *file = iocb->ki_filp;
180 struct inode *inode = bdev_file_inode(file);
183 return dax_do_io(iocb, inode, iter, offset, blkdev_get_block,
184 NULL, DIO_SKIP_DIO_COUNT);
185 return __blockdev_direct_IO(iocb, inode, I_BDEV(inode), iter, offset,
186 blkdev_get_block, NULL, NULL,
190 int __sync_blockdev(struct block_device *bdev, int wait)
195 return filemap_flush(bdev->bd_inode->i_mapping);
196 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
200 * Write out and wait upon all the dirty data associated with a block
201 * device via its mapping. Does not take the superblock lock.
203 int sync_blockdev(struct block_device *bdev)
205 return __sync_blockdev(bdev, 1);
207 EXPORT_SYMBOL(sync_blockdev);
210 * Write out and wait upon all dirty data associated with this
211 * device. Filesystem data as well as the underlying block
212 * device. Takes the superblock lock.
214 int fsync_bdev(struct block_device *bdev)
216 struct super_block *sb = get_super(bdev);
218 int res = sync_filesystem(sb);
222 return sync_blockdev(bdev);
224 EXPORT_SYMBOL(fsync_bdev);
227 * freeze_bdev -- lock a filesystem and force it into a consistent state
228 * @bdev: blockdevice to lock
230 * If a superblock is found on this device, we take the s_umount semaphore
231 * on it to make sure nobody unmounts until the snapshot creation is done.
232 * The reference counter (bd_fsfreeze_count) guarantees that only the last
233 * unfreeze process can unfreeze the frozen filesystem actually when multiple
234 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
235 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
238 struct super_block *freeze_bdev(struct block_device *bdev)
240 struct super_block *sb;
243 mutex_lock(&bdev->bd_fsfreeze_mutex);
244 if (++bdev->bd_fsfreeze_count > 1) {
246 * We don't even need to grab a reference - the first call
247 * to freeze_bdev grab an active reference and only the last
248 * thaw_bdev drops it.
250 sb = get_super(bdev);
252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
256 sb = get_active_super(bdev);
259 if (sb->s_op->freeze_super)
260 error = sb->s_op->freeze_super(sb);
262 error = freeze_super(sb);
264 deactivate_super(sb);
265 bdev->bd_fsfreeze_count--;
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267 return ERR_PTR(error);
269 deactivate_super(sb);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return sb; /* thaw_bdev releases s->s_umount */
275 EXPORT_SYMBOL(freeze_bdev);
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
288 mutex_lock(&bdev->bd_fsfreeze_mutex);
289 if (!bdev->bd_fsfreeze_count)
293 if (--bdev->bd_fsfreeze_count > 0)
299 if (sb->s_op->thaw_super)
300 error = sb->s_op->thaw_super(sb);
302 error = thaw_super(sb);
304 bdev->bd_fsfreeze_count++;
305 mutex_unlock(&bdev->bd_fsfreeze_mutex);
309 mutex_unlock(&bdev->bd_fsfreeze_mutex);
312 EXPORT_SYMBOL(thaw_bdev);
314 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
316 return block_write_full_page(page, blkdev_get_block, wbc);
319 static int blkdev_readpage(struct file * file, struct page * page)
321 return block_read_full_page(page, blkdev_get_block);
324 static int blkdev_readpages(struct file *file, struct address_space *mapping,
325 struct list_head *pages, unsigned nr_pages)
327 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
330 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
331 loff_t pos, unsigned len, unsigned flags,
332 struct page **pagep, void **fsdata)
334 return block_write_begin(mapping, pos, len, flags, pagep,
338 static int blkdev_write_end(struct file *file, struct address_space *mapping,
339 loff_t pos, unsigned len, unsigned copied,
340 struct page *page, void *fsdata)
343 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
353 * for a block special file file_inode(file)->i_size is zero
354 * so we compute the size by hand (just as in block_read/write above)
356 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
358 struct inode *bd_inode = bdev_file_inode(file);
361 inode_lock(bd_inode);
362 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
363 inode_unlock(bd_inode);
367 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
369 struct inode *bd_inode = bdev_file_inode(filp);
370 struct block_device *bdev = I_BDEV(bd_inode);
373 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
378 * There is no need to serialise calls to blkdev_issue_flush with
379 * i_mutex and doing so causes performance issues with concurrent
380 * O_SYNC writers to a block device.
382 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
383 if (error == -EOPNOTSUPP)
388 EXPORT_SYMBOL(blkdev_fsync);
391 * bdev_read_page() - Start reading a page from a block device
392 * @bdev: The device to read the page from
393 * @sector: The offset on the device to read the page to (need not be aligned)
394 * @page: The page to read
396 * On entry, the page should be locked. It will be unlocked when the page
397 * has been read. If the block driver implements rw_page synchronously,
398 * that will be true on exit from this function, but it need not be.
400 * Errors returned by this function are usually "soft", eg out of memory, or
401 * queue full; callers should try a different route to read this page rather
402 * than propagate an error back up the stack.
404 * Return: negative errno if an error occurs, 0 if submission was successful.
406 int bdev_read_page(struct block_device *bdev, sector_t sector,
409 const struct block_device_operations *ops = bdev->bd_disk->fops;
410 int result = -EOPNOTSUPP;
412 if (!ops->rw_page || bdev_get_integrity(bdev))
415 result = blk_queue_enter(bdev->bd_queue, false);
418 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, READ);
419 blk_queue_exit(bdev->bd_queue);
422 EXPORT_SYMBOL_GPL(bdev_read_page);
425 * bdev_write_page() - Start writing a page to a block device
426 * @bdev: The device to write the page to
427 * @sector: The offset on the device to write the page to (need not be aligned)
428 * @page: The page to write
429 * @wbc: The writeback_control for the write
431 * On entry, the page should be locked and not currently under writeback.
432 * On exit, if the write started successfully, the page will be unlocked and
433 * under writeback. If the write failed already (eg the driver failed to
434 * queue the page to the device), the page will still be locked. If the
435 * caller is a ->writepage implementation, it will need to unlock the page.
437 * Errors returned by this function are usually "soft", eg out of memory, or
438 * queue full; callers should try a different route to write this page rather
439 * than propagate an error back up the stack.
441 * Return: negative errno if an error occurs, 0 if submission was successful.
443 int bdev_write_page(struct block_device *bdev, sector_t sector,
444 struct page *page, struct writeback_control *wbc)
447 int rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE;
448 const struct block_device_operations *ops = bdev->bd_disk->fops;
450 if (!ops->rw_page || bdev_get_integrity(bdev))
452 result = blk_queue_enter(bdev->bd_queue, false);
456 set_page_writeback(page);
457 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, rw);
459 end_page_writeback(page);
462 blk_queue_exit(bdev->bd_queue);
465 EXPORT_SYMBOL_GPL(bdev_write_page);
468 * bdev_direct_access() - Get the address for directly-accessibly memory
469 * @bdev: The device containing the memory
470 * @dax: control and output parameters for ->direct_access
472 * If a block device is made up of directly addressable memory, this function
473 * will tell the caller the PFN and the address of the memory. The address
474 * may be directly dereferenced within the kernel without the need to call
475 * ioremap(), kmap() or similar. The PFN is suitable for inserting into
478 * Return: negative errno if an error occurs, otherwise the number of bytes
479 * accessible at this address.
481 long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
483 sector_t sector = dax->sector;
484 long avail, size = dax->size;
485 const struct block_device_operations *ops = bdev->bd_disk->fops;
488 * The device driver is allowed to sleep, in order to make the
489 * memory directly accessible.
495 if (!ops->direct_access)
497 if ((sector + DIV_ROUND_UP(size, 512)) >
498 part_nr_sects_read(bdev->bd_part))
500 sector += get_start_sect(bdev);
501 if (sector % (PAGE_SIZE / 512))
503 avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn);
506 if (avail > 0 && avail & ~PAGE_MASK)
508 return min(avail, size);
510 EXPORT_SYMBOL_GPL(bdev_direct_access);
513 * bdev_dax_supported() - Check if the device supports dax for filesystem
514 * @sb: The superblock of the device
515 * @blocksize: The block size of the device
517 * This is a library function for filesystems to check if the block device
518 * can be mounted with dax option.
520 * Return: negative errno if unsupported, 0 if supported.
522 int bdev_dax_supported(struct super_block *sb, int blocksize)
524 struct blk_dax_ctl dax = {
530 if (blocksize != PAGE_SIZE) {
531 vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
535 err = bdev_direct_access(sb->s_bdev, &dax);
539 vfs_msg(sb, KERN_ERR,
540 "error: device does not support dax");
543 vfs_msg(sb, KERN_ERR,
544 "error: unaligned partition for dax");
547 vfs_msg(sb, KERN_ERR,
548 "error: dax access failed (%d)", err);
555 EXPORT_SYMBOL_GPL(bdev_dax_supported);
561 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
562 static struct kmem_cache * bdev_cachep __read_mostly;
564 static struct inode *bdev_alloc_inode(struct super_block *sb)
566 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
569 return &ei->vfs_inode;
572 static void bdev_i_callback(struct rcu_head *head)
574 struct inode *inode = container_of(head, struct inode, i_rcu);
575 struct bdev_inode *bdi = BDEV_I(inode);
577 kmem_cache_free(bdev_cachep, bdi);
580 static void bdev_destroy_inode(struct inode *inode)
582 call_rcu(&inode->i_rcu, bdev_i_callback);
585 static void init_once(void *foo)
587 struct bdev_inode *ei = (struct bdev_inode *) foo;
588 struct block_device *bdev = &ei->bdev;
590 memset(bdev, 0, sizeof(*bdev));
591 mutex_init(&bdev->bd_mutex);
592 INIT_LIST_HEAD(&bdev->bd_inodes);
593 INIT_LIST_HEAD(&bdev->bd_list);
595 INIT_LIST_HEAD(&bdev->bd_holder_disks);
597 inode_init_once(&ei->vfs_inode);
598 /* Initialize mutex for freeze. */
599 mutex_init(&bdev->bd_fsfreeze_mutex);
602 static inline void __bd_forget(struct inode *inode)
604 list_del_init(&inode->i_devices);
605 inode->i_bdev = NULL;
606 inode->i_mapping = &inode->i_data;
609 static void bdev_evict_inode(struct inode *inode)
611 struct block_device *bdev = &BDEV_I(inode)->bdev;
613 truncate_inode_pages_final(&inode->i_data);
614 invalidate_inode_buffers(inode); /* is it needed here? */
616 spin_lock(&bdev_lock);
617 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
618 __bd_forget(list_entry(p, struct inode, i_devices));
620 list_del_init(&bdev->bd_list);
621 spin_unlock(&bdev_lock);
624 static const struct super_operations bdev_sops = {
625 .statfs = simple_statfs,
626 .alloc_inode = bdev_alloc_inode,
627 .destroy_inode = bdev_destroy_inode,
628 .drop_inode = generic_delete_inode,
629 .evict_inode = bdev_evict_inode,
632 static struct dentry *bd_mount(struct file_system_type *fs_type,
633 int flags, const char *dev_name, void *data)
636 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
638 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
642 static struct file_system_type bd_type = {
645 .kill_sb = kill_anon_super,
648 struct super_block *blockdev_superblock __read_mostly;
649 EXPORT_SYMBOL_GPL(blockdev_superblock);
651 void __init bdev_cache_init(void)
654 static struct vfsmount *bd_mnt;
656 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
657 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
658 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
660 err = register_filesystem(&bd_type);
662 panic("Cannot register bdev pseudo-fs");
663 bd_mnt = kern_mount(&bd_type);
665 panic("Cannot create bdev pseudo-fs");
666 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
670 * Most likely _very_ bad one - but then it's hardly critical for small
671 * /dev and can be fixed when somebody will need really large one.
672 * Keep in mind that it will be fed through icache hash function too.
674 static inline unsigned long hash(dev_t dev)
676 return MAJOR(dev)+MINOR(dev);
679 static int bdev_test(struct inode *inode, void *data)
681 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
684 static int bdev_set(struct inode *inode, void *data)
686 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
690 static LIST_HEAD(all_bdevs);
692 struct block_device *bdget(dev_t dev)
694 struct block_device *bdev;
697 inode = iget5_locked(blockdev_superblock, hash(dev),
698 bdev_test, bdev_set, &dev);
703 bdev = &BDEV_I(inode)->bdev;
705 if (inode->i_state & I_NEW) {
706 bdev->bd_contains = NULL;
707 bdev->bd_super = NULL;
708 bdev->bd_inode = inode;
709 bdev->bd_block_size = (1 << inode->i_blkbits);
710 bdev->bd_part_count = 0;
711 bdev->bd_invalidated = 0;
712 inode->i_mode = S_IFBLK;
714 inode->i_bdev = bdev;
715 inode->i_data.a_ops = &def_blk_aops;
716 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
717 spin_lock(&bdev_lock);
718 list_add(&bdev->bd_list, &all_bdevs);
719 spin_unlock(&bdev_lock);
720 unlock_new_inode(inode);
725 EXPORT_SYMBOL(bdget);
728 * bdgrab -- Grab a reference to an already referenced block device
729 * @bdev: Block device to grab a reference to.
731 struct block_device *bdgrab(struct block_device *bdev)
733 ihold(bdev->bd_inode);
736 EXPORT_SYMBOL(bdgrab);
738 long nr_blockdev_pages(void)
740 struct block_device *bdev;
742 spin_lock(&bdev_lock);
743 list_for_each_entry(bdev, &all_bdevs, bd_list) {
744 ret += bdev->bd_inode->i_mapping->nrpages;
746 spin_unlock(&bdev_lock);
750 void bdput(struct block_device *bdev)
752 iput(bdev->bd_inode);
755 EXPORT_SYMBOL(bdput);
757 static struct block_device *bd_acquire(struct inode *inode)
759 struct block_device *bdev;
761 spin_lock(&bdev_lock);
762 bdev = inode->i_bdev;
765 spin_unlock(&bdev_lock);
768 spin_unlock(&bdev_lock);
770 bdev = bdget(inode->i_rdev);
772 spin_lock(&bdev_lock);
773 if (!inode->i_bdev) {
775 * We take an additional reference to bd_inode,
776 * and it's released in clear_inode() of inode.
777 * So, we can access it via ->i_mapping always
781 inode->i_bdev = bdev;
782 inode->i_mapping = bdev->bd_inode->i_mapping;
783 list_add(&inode->i_devices, &bdev->bd_inodes);
785 spin_unlock(&bdev_lock);
790 /* Call when you free inode */
792 void bd_forget(struct inode *inode)
794 struct block_device *bdev = NULL;
796 spin_lock(&bdev_lock);
797 if (!sb_is_blkdev_sb(inode->i_sb))
798 bdev = inode->i_bdev;
800 spin_unlock(&bdev_lock);
807 * bd_may_claim - test whether a block device can be claimed
808 * @bdev: block device of interest
809 * @whole: whole block device containing @bdev, may equal @bdev
810 * @holder: holder trying to claim @bdev
812 * Test whether @bdev can be claimed by @holder.
815 * spin_lock(&bdev_lock).
818 * %true if @bdev can be claimed, %false otherwise.
820 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
823 if (bdev->bd_holder == holder)
824 return true; /* already a holder */
825 else if (bdev->bd_holder != NULL)
826 return false; /* held by someone else */
827 else if (bdev->bd_contains == bdev)
828 return true; /* is a whole device which isn't held */
830 else if (whole->bd_holder == bd_may_claim)
831 return true; /* is a partition of a device that is being partitioned */
832 else if (whole->bd_holder != NULL)
833 return false; /* is a partition of a held device */
835 return true; /* is a partition of an un-held device */
839 * bd_prepare_to_claim - prepare to claim a block device
840 * @bdev: block device of interest
841 * @whole: the whole device containing @bdev, may equal @bdev
842 * @holder: holder trying to claim @bdev
844 * Prepare to claim @bdev. This function fails if @bdev is already
845 * claimed by another holder and waits if another claiming is in
846 * progress. This function doesn't actually claim. On successful
847 * return, the caller has ownership of bd_claiming and bd_holder[s].
850 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
854 * 0 if @bdev can be claimed, -EBUSY otherwise.
856 static int bd_prepare_to_claim(struct block_device *bdev,
857 struct block_device *whole, void *holder)
860 /* if someone else claimed, fail */
861 if (!bd_may_claim(bdev, whole, holder))
864 /* if claiming is already in progress, wait for it to finish */
865 if (whole->bd_claiming) {
866 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
869 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
870 spin_unlock(&bdev_lock);
872 finish_wait(wq, &wait);
873 spin_lock(&bdev_lock);
882 * bd_start_claiming - start claiming a block device
883 * @bdev: block device of interest
884 * @holder: holder trying to claim @bdev
886 * @bdev is about to be opened exclusively. Check @bdev can be opened
887 * exclusively and mark that an exclusive open is in progress. Each
888 * successful call to this function must be matched with a call to
889 * either bd_finish_claiming() or bd_abort_claiming() (which do not
892 * This function is used to gain exclusive access to the block device
893 * without actually causing other exclusive open attempts to fail. It
894 * should be used when the open sequence itself requires exclusive
895 * access but may subsequently fail.
901 * Pointer to the block device containing @bdev on success, ERR_PTR()
904 static struct block_device *bd_start_claiming(struct block_device *bdev,
907 struct gendisk *disk;
908 struct block_device *whole;
914 * @bdev might not have been initialized properly yet, look up
915 * and grab the outer block device the hard way.
917 disk = get_gendisk(bdev->bd_dev, &partno);
919 return ERR_PTR(-ENXIO);
922 * Normally, @bdev should equal what's returned from bdget_disk()
923 * if partno is 0; however, some drivers (floppy) use multiple
924 * bdev's for the same physical device and @bdev may be one of the
925 * aliases. Keep @bdev if partno is 0. This means claimer
926 * tracking is broken for those devices but it has always been that
930 whole = bdget_disk(disk, 0);
932 whole = bdgrab(bdev);
934 module_put(disk->fops->owner);
937 return ERR_PTR(-ENOMEM);
939 /* prepare to claim, if successful, mark claiming in progress */
940 spin_lock(&bdev_lock);
942 err = bd_prepare_to_claim(bdev, whole, holder);
944 whole->bd_claiming = holder;
945 spin_unlock(&bdev_lock);
948 spin_unlock(&bdev_lock);
955 struct bd_holder_disk {
956 struct list_head list;
957 struct gendisk *disk;
961 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
962 struct gendisk *disk)
964 struct bd_holder_disk *holder;
966 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
967 if (holder->disk == disk)
972 static int add_symlink(struct kobject *from, struct kobject *to)
974 return sysfs_create_link(from, to, kobject_name(to));
977 static void del_symlink(struct kobject *from, struct kobject *to)
979 sysfs_remove_link(from, kobject_name(to));
983 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
984 * @bdev: the claimed slave bdev
985 * @disk: the holding disk
987 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
989 * This functions creates the following sysfs symlinks.
991 * - from "slaves" directory of the holder @disk to the claimed @bdev
992 * - from "holders" directory of the @bdev to the holder @disk
994 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
995 * passed to bd_link_disk_holder(), then:
997 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
998 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1000 * The caller must have claimed @bdev before calling this function and
1001 * ensure that both @bdev and @disk are valid during the creation and
1002 * lifetime of these symlinks.
1008 * 0 on success, -errno on failure.
1010 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1012 struct bd_holder_disk *holder;
1015 mutex_lock(&bdev->bd_mutex);
1017 WARN_ON_ONCE(!bdev->bd_holder);
1019 /* FIXME: remove the following once add_disk() handles errors */
1020 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1023 holder = bd_find_holder_disk(bdev, disk);
1029 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1035 INIT_LIST_HEAD(&holder->list);
1036 holder->disk = disk;
1039 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1043 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1047 * bdev could be deleted beneath us which would implicitly destroy
1048 * the holder directory. Hold on to it.
1050 kobject_get(bdev->bd_part->holder_dir);
1052 list_add(&holder->list, &bdev->bd_holder_disks);
1056 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1060 mutex_unlock(&bdev->bd_mutex);
1063 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1066 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1067 * @bdev: the calimed slave bdev
1068 * @disk: the holding disk
1070 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1075 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1077 struct bd_holder_disk *holder;
1079 mutex_lock(&bdev->bd_mutex);
1081 holder = bd_find_holder_disk(bdev, disk);
1083 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1084 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1085 del_symlink(bdev->bd_part->holder_dir,
1086 &disk_to_dev(disk)->kobj);
1087 kobject_put(bdev->bd_part->holder_dir);
1088 list_del_init(&holder->list);
1092 mutex_unlock(&bdev->bd_mutex);
1094 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1098 * flush_disk - invalidates all buffer-cache entries on a disk
1100 * @bdev: struct block device to be flushed
1101 * @kill_dirty: flag to guide handling of dirty inodes
1103 * Invalidates all buffer-cache entries on a disk. It should be called
1104 * when a disk has been changed -- either by a media change or online
1107 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1109 if (__invalidate_device(bdev, kill_dirty)) {
1110 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1111 "resized disk %s\n",
1112 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1117 if (disk_part_scan_enabled(bdev->bd_disk))
1118 bdev->bd_invalidated = 1;
1122 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1123 * @disk: struct gendisk to check
1124 * @bdev: struct bdev to adjust.
1126 * This routine checks to see if the bdev size does not match the disk size
1127 * and adjusts it if it differs.
1129 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1131 loff_t disk_size, bdev_size;
1133 disk_size = (loff_t)get_capacity(disk) << 9;
1134 bdev_size = i_size_read(bdev->bd_inode);
1135 if (disk_size != bdev_size) {
1137 "%s: detected capacity change from %lld to %lld\n",
1138 disk->disk_name, bdev_size, disk_size);
1139 i_size_write(bdev->bd_inode, disk_size);
1140 flush_disk(bdev, false);
1143 EXPORT_SYMBOL(check_disk_size_change);
1146 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1147 * @disk: struct gendisk to be revalidated
1149 * This routine is a wrapper for lower-level driver's revalidate_disk
1150 * call-backs. It is used to do common pre and post operations needed
1151 * for all revalidate_disk operations.
1153 int revalidate_disk(struct gendisk *disk)
1155 struct block_device *bdev;
1158 if (disk->fops->revalidate_disk)
1159 ret = disk->fops->revalidate_disk(disk);
1160 blk_integrity_revalidate(disk);
1161 bdev = bdget_disk(disk, 0);
1165 mutex_lock(&bdev->bd_mutex);
1166 check_disk_size_change(disk, bdev);
1167 bdev->bd_invalidated = 0;
1168 mutex_unlock(&bdev->bd_mutex);
1172 EXPORT_SYMBOL(revalidate_disk);
1175 * This routine checks whether a removable media has been changed,
1176 * and invalidates all buffer-cache-entries in that case. This
1177 * is a relatively slow routine, so we have to try to minimize using
1178 * it. Thus it is called only upon a 'mount' or 'open'. This
1179 * is the best way of combining speed and utility, I think.
1180 * People changing diskettes in the middle of an operation deserve
1183 int check_disk_change(struct block_device *bdev)
1185 struct gendisk *disk = bdev->bd_disk;
1186 const struct block_device_operations *bdops = disk->fops;
1187 unsigned int events;
1189 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1190 DISK_EVENT_EJECT_REQUEST);
1191 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1194 flush_disk(bdev, true);
1195 if (bdops->revalidate_disk)
1196 bdops->revalidate_disk(bdev->bd_disk);
1200 EXPORT_SYMBOL(check_disk_change);
1202 void bd_set_size(struct block_device *bdev, loff_t size)
1204 unsigned bsize = bdev_logical_block_size(bdev);
1206 inode_lock(bdev->bd_inode);
1207 i_size_write(bdev->bd_inode, size);
1208 inode_unlock(bdev->bd_inode);
1209 while (bsize < PAGE_SIZE) {
1214 bdev->bd_block_size = bsize;
1215 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1217 EXPORT_SYMBOL(bd_set_size);
1219 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1224 * mutex_lock(part->bd_mutex)
1225 * mutex_lock_nested(whole->bd_mutex, 1)
1228 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1230 struct gendisk *disk;
1231 struct module *owner;
1236 if (mode & FMODE_READ)
1238 if (mode & FMODE_WRITE)
1241 * hooks: /n/, see "layering violations".
1244 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1254 disk = get_gendisk(bdev->bd_dev, &partno);
1257 owner = disk->fops->owner;
1259 disk_block_events(disk);
1260 mutex_lock_nested(&bdev->bd_mutex, for_part);
1261 if (!bdev->bd_openers) {
1262 bdev->bd_disk = disk;
1263 bdev->bd_queue = disk->queue;
1264 bdev->bd_contains = bdev;
1265 if (IS_ENABLED(CONFIG_BLK_DEV_DAX) && disk->fops->direct_access)
1266 bdev->bd_inode->i_flags = S_DAX;
1268 bdev->bd_inode->i_flags = 0;
1272 bdev->bd_part = disk_get_part(disk, partno);
1277 if (disk->fops->open) {
1278 ret = disk->fops->open(bdev, mode);
1279 if (ret == -ERESTARTSYS) {
1280 /* Lost a race with 'disk' being
1281 * deleted, try again.
1284 disk_put_part(bdev->bd_part);
1285 bdev->bd_part = NULL;
1286 bdev->bd_disk = NULL;
1287 bdev->bd_queue = NULL;
1288 mutex_unlock(&bdev->bd_mutex);
1289 disk_unblock_events(disk);
1297 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1298 if (!blkdev_dax_capable(bdev))
1299 bdev->bd_inode->i_flags &= ~S_DAX;
1303 * If the device is invalidated, rescan partition
1304 * if open succeeded or failed with -ENOMEDIUM.
1305 * The latter is necessary to prevent ghost
1306 * partitions on a removed medium.
1308 if (bdev->bd_invalidated) {
1310 rescan_partitions(disk, bdev);
1311 else if (ret == -ENOMEDIUM)
1312 invalidate_partitions(disk, bdev);
1318 struct block_device *whole;
1319 whole = bdget_disk(disk, 0);
1324 ret = __blkdev_get(whole, mode, 1);
1327 bdev->bd_contains = whole;
1328 bdev->bd_part = disk_get_part(disk, partno);
1329 if (!(disk->flags & GENHD_FL_UP) ||
1330 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1334 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1335 if (!blkdev_dax_capable(bdev))
1336 bdev->bd_inode->i_flags &= ~S_DAX;
1339 if (bdev->bd_contains == bdev) {
1341 if (bdev->bd_disk->fops->open)
1342 ret = bdev->bd_disk->fops->open(bdev, mode);
1343 /* the same as first opener case, read comment there */
1344 if (bdev->bd_invalidated) {
1346 rescan_partitions(bdev->bd_disk, bdev);
1347 else if (ret == -ENOMEDIUM)
1348 invalidate_partitions(bdev->bd_disk, bdev);
1351 goto out_unlock_bdev;
1353 /* only one opener holds refs to the module and disk */
1359 bdev->bd_part_count++;
1360 mutex_unlock(&bdev->bd_mutex);
1361 disk_unblock_events(disk);
1365 disk_put_part(bdev->bd_part);
1366 bdev->bd_disk = NULL;
1367 bdev->bd_part = NULL;
1368 bdev->bd_queue = NULL;
1369 if (bdev != bdev->bd_contains)
1370 __blkdev_put(bdev->bd_contains, mode, 1);
1371 bdev->bd_contains = NULL;
1373 mutex_unlock(&bdev->bd_mutex);
1374 disk_unblock_events(disk);
1384 * blkdev_get - open a block device
1385 * @bdev: block_device to open
1386 * @mode: FMODE_* mask
1387 * @holder: exclusive holder identifier
1389 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1390 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1391 * @holder is invalid. Exclusive opens may nest for the same @holder.
1393 * On success, the reference count of @bdev is unchanged. On failure,
1400 * 0 on success, -errno on failure.
1402 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1404 struct block_device *whole = NULL;
1407 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1409 if ((mode & FMODE_EXCL) && holder) {
1410 whole = bd_start_claiming(bdev, holder);
1411 if (IS_ERR(whole)) {
1413 return PTR_ERR(whole);
1417 res = __blkdev_get(bdev, mode, 0);
1420 struct gendisk *disk = whole->bd_disk;
1422 /* finish claiming */
1423 mutex_lock(&bdev->bd_mutex);
1424 spin_lock(&bdev_lock);
1427 BUG_ON(!bd_may_claim(bdev, whole, holder));
1429 * Note that for a whole device bd_holders
1430 * will be incremented twice, and bd_holder
1431 * will be set to bd_may_claim before being
1434 whole->bd_holders++;
1435 whole->bd_holder = bd_may_claim;
1437 bdev->bd_holder = holder;
1440 /* tell others that we're done */
1441 BUG_ON(whole->bd_claiming != holder);
1442 whole->bd_claiming = NULL;
1443 wake_up_bit(&whole->bd_claiming, 0);
1445 spin_unlock(&bdev_lock);
1448 * Block event polling for write claims if requested. Any
1449 * write holder makes the write_holder state stick until
1450 * all are released. This is good enough and tracking
1451 * individual writeable reference is too fragile given the
1452 * way @mode is used in blkdev_get/put().
1454 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1455 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1456 bdev->bd_write_holder = true;
1457 disk_block_events(disk);
1460 mutex_unlock(&bdev->bd_mutex);
1466 EXPORT_SYMBOL(blkdev_get);
1469 * blkdev_get_by_path - open a block device by name
1470 * @path: path to the block device to open
1471 * @mode: FMODE_* mask
1472 * @holder: exclusive holder identifier
1474 * Open the blockdevice described by the device file at @path. @mode
1475 * and @holder are identical to blkdev_get().
1477 * On success, the returned block_device has reference count of one.
1483 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1485 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1488 struct block_device *bdev;
1491 bdev = lookup_bdev(path);
1495 err = blkdev_get(bdev, mode, holder);
1497 return ERR_PTR(err);
1499 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1500 blkdev_put(bdev, mode);
1501 return ERR_PTR(-EACCES);
1506 EXPORT_SYMBOL(blkdev_get_by_path);
1509 * blkdev_get_by_dev - open a block device by device number
1510 * @dev: device number of block device to open
1511 * @mode: FMODE_* mask
1512 * @holder: exclusive holder identifier
1514 * Open the blockdevice described by device number @dev. @mode and
1515 * @holder are identical to blkdev_get().
1517 * Use it ONLY if you really do not have anything better - i.e. when
1518 * you are behind a truly sucky interface and all you are given is a
1519 * device number. _Never_ to be used for internal purposes. If you
1520 * ever need it - reconsider your API.
1522 * On success, the returned block_device has reference count of one.
1528 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1530 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1532 struct block_device *bdev;
1537 return ERR_PTR(-ENOMEM);
1539 err = blkdev_get(bdev, mode, holder);
1541 return ERR_PTR(err);
1545 EXPORT_SYMBOL(blkdev_get_by_dev);
1547 static int blkdev_open(struct inode * inode, struct file * filp)
1549 struct block_device *bdev;
1552 * Preserve backwards compatibility and allow large file access
1553 * even if userspace doesn't ask for it explicitly. Some mkfs
1554 * binary needs it. We might want to drop this workaround
1555 * during an unstable branch.
1557 filp->f_flags |= O_LARGEFILE;
1559 if (filp->f_flags & O_NDELAY)
1560 filp->f_mode |= FMODE_NDELAY;
1561 if (filp->f_flags & O_EXCL)
1562 filp->f_mode |= FMODE_EXCL;
1563 if ((filp->f_flags & O_ACCMODE) == 3)
1564 filp->f_mode |= FMODE_WRITE_IOCTL;
1566 bdev = bd_acquire(inode);
1570 filp->f_mapping = bdev->bd_inode->i_mapping;
1572 return blkdev_get(bdev, filp->f_mode, filp);
1575 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1577 struct gendisk *disk = bdev->bd_disk;
1578 struct block_device *victim = NULL;
1580 mutex_lock_nested(&bdev->bd_mutex, for_part);
1582 bdev->bd_part_count--;
1584 if (!--bdev->bd_openers) {
1585 WARN_ON_ONCE(bdev->bd_holders);
1586 sync_blockdev(bdev);
1589 bdev_write_inode(bdev);
1591 * Detaching bdev inode from its wb in __destroy_inode()
1592 * is too late: the queue which embeds its bdi (along with
1593 * root wb) can be gone as soon as we put_disk() below.
1595 inode_detach_wb(bdev->bd_inode);
1597 if (bdev->bd_contains == bdev) {
1598 if (disk->fops->release)
1599 disk->fops->release(disk, mode);
1601 if (!bdev->bd_openers) {
1602 struct module *owner = disk->fops->owner;
1604 disk_put_part(bdev->bd_part);
1605 bdev->bd_part = NULL;
1606 bdev->bd_disk = NULL;
1607 if (bdev != bdev->bd_contains)
1608 victim = bdev->bd_contains;
1609 bdev->bd_contains = NULL;
1614 mutex_unlock(&bdev->bd_mutex);
1617 __blkdev_put(victim, mode, 1);
1620 void blkdev_put(struct block_device *bdev, fmode_t mode)
1622 mutex_lock(&bdev->bd_mutex);
1624 if (mode & FMODE_EXCL) {
1628 * Release a claim on the device. The holder fields
1629 * are protected with bdev_lock. bd_mutex is to
1630 * synchronize disk_holder unlinking.
1632 spin_lock(&bdev_lock);
1634 WARN_ON_ONCE(--bdev->bd_holders < 0);
1635 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1637 /* bd_contains might point to self, check in a separate step */
1638 if ((bdev_free = !bdev->bd_holders))
1639 bdev->bd_holder = NULL;
1640 if (!bdev->bd_contains->bd_holders)
1641 bdev->bd_contains->bd_holder = NULL;
1643 spin_unlock(&bdev_lock);
1646 * If this was the last claim, remove holder link and
1647 * unblock evpoll if it was a write holder.
1649 if (bdev_free && bdev->bd_write_holder) {
1650 disk_unblock_events(bdev->bd_disk);
1651 bdev->bd_write_holder = false;
1656 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1657 * event. This is to ensure detection of media removal commanded
1658 * from userland - e.g. eject(1).
1660 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1662 mutex_unlock(&bdev->bd_mutex);
1664 __blkdev_put(bdev, mode, 0);
1666 EXPORT_SYMBOL(blkdev_put);
1668 static int blkdev_close(struct inode * inode, struct file * filp)
1670 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1671 blkdev_put(bdev, filp->f_mode);
1675 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1677 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1678 fmode_t mode = file->f_mode;
1681 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1682 * to updated it before every ioctl.
1684 if (file->f_flags & O_NDELAY)
1685 mode |= FMODE_NDELAY;
1687 mode &= ~FMODE_NDELAY;
1689 return blkdev_ioctl(bdev, mode, cmd, arg);
1693 * Write data to the block device. Only intended for the block device itself
1694 * and the raw driver which basically is a fake block device.
1696 * Does not take i_mutex for the write and thus is not for general purpose
1699 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1701 struct file *file = iocb->ki_filp;
1702 struct inode *bd_inode = bdev_file_inode(file);
1703 loff_t size = i_size_read(bd_inode);
1704 struct blk_plug plug;
1707 if (bdev_read_only(I_BDEV(bd_inode)))
1710 if (!iov_iter_count(from))
1713 if (iocb->ki_pos >= size)
1716 iov_iter_truncate(from, size - iocb->ki_pos);
1718 blk_start_plug(&plug);
1719 ret = __generic_file_write_iter(iocb, from);
1722 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
1726 blk_finish_plug(&plug);
1729 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1731 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1733 struct file *file = iocb->ki_filp;
1734 struct inode *bd_inode = bdev_file_inode(file);
1735 loff_t size = i_size_read(bd_inode);
1736 loff_t pos = iocb->ki_pos;
1742 iov_iter_truncate(to, size);
1743 return generic_file_read_iter(iocb, to);
1745 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1748 * Try to release a page associated with block device when the system
1749 * is under memory pressure.
1751 static int blkdev_releasepage(struct page *page, gfp_t wait)
1753 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1755 if (super && super->s_op->bdev_try_to_free_page)
1756 return super->s_op->bdev_try_to_free_page(super, page, wait);
1758 return try_to_free_buffers(page);
1761 static int blkdev_writepages(struct address_space *mapping,
1762 struct writeback_control *wbc)
1764 if (dax_mapping(mapping)) {
1765 struct block_device *bdev = I_BDEV(mapping->host);
1767 return dax_writeback_mapping_range(mapping, bdev, wbc);
1769 return generic_writepages(mapping, wbc);
1772 static const struct address_space_operations def_blk_aops = {
1773 .readpage = blkdev_readpage,
1774 .readpages = blkdev_readpages,
1775 .writepage = blkdev_writepage,
1776 .write_begin = blkdev_write_begin,
1777 .write_end = blkdev_write_end,
1778 .writepages = blkdev_writepages,
1779 .releasepage = blkdev_releasepage,
1780 .direct_IO = blkdev_direct_IO,
1781 .is_dirty_writeback = buffer_check_dirty_writeback,
1784 #ifdef CONFIG_FS_DAX
1786 * In the raw block case we do not need to contend with truncation nor
1787 * unwritten file extents. Without those concerns there is no need for
1788 * additional locking beyond the mmap_sem context that these routines
1789 * are already executing under.
1791 * Note, there is no protection if the block device is dynamically
1792 * resized (partition grow/shrink) during a fault. A stable block device
1793 * size is already not enforced in the blkdev_direct_IO path.
1795 * For DAX, it is the responsibility of the block device driver to
1796 * ensure the whole-disk device size is stable while requests are in
1799 * Finally, unlike the filemap_page_mkwrite() case there is no
1800 * filesystem superblock to sync against freezing. We still include a
1801 * pfn_mkwrite callback for dax drivers to receive write fault
1804 static int blkdev_dax_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1806 return __dax_fault(vma, vmf, blkdev_get_block);
1809 static int blkdev_dax_pfn_mkwrite(struct vm_area_struct *vma,
1810 struct vm_fault *vmf)
1812 return dax_pfn_mkwrite(vma, vmf);
1815 static int blkdev_dax_pmd_fault(struct vm_area_struct *vma, unsigned long addr,
1816 pmd_t *pmd, unsigned int flags)
1818 return __dax_pmd_fault(vma, addr, pmd, flags, blkdev_get_block);
1821 static const struct vm_operations_struct blkdev_dax_vm_ops = {
1822 .fault = blkdev_dax_fault,
1823 .pmd_fault = blkdev_dax_pmd_fault,
1824 .pfn_mkwrite = blkdev_dax_pfn_mkwrite,
1827 static const struct vm_operations_struct blkdev_default_vm_ops = {
1828 .fault = filemap_fault,
1829 .map_pages = filemap_map_pages,
1832 static int blkdev_mmap(struct file *file, struct vm_area_struct *vma)
1834 struct inode *bd_inode = bdev_file_inode(file);
1836 file_accessed(file);
1837 if (IS_DAX(bd_inode)) {
1838 vma->vm_ops = &blkdev_dax_vm_ops;
1839 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
1841 vma->vm_ops = &blkdev_default_vm_ops;
1847 #define blkdev_mmap generic_file_mmap
1850 const struct file_operations def_blk_fops = {
1851 .open = blkdev_open,
1852 .release = blkdev_close,
1853 .llseek = block_llseek,
1854 .read_iter = blkdev_read_iter,
1855 .write_iter = blkdev_write_iter,
1856 .mmap = blkdev_mmap,
1857 .fsync = blkdev_fsync,
1858 .unlocked_ioctl = block_ioctl,
1859 #ifdef CONFIG_COMPAT
1860 .compat_ioctl = compat_blkdev_ioctl,
1862 .splice_read = generic_file_splice_read,
1863 .splice_write = iter_file_splice_write,
1866 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1869 mm_segment_t old_fs = get_fs();
1871 res = blkdev_ioctl(bdev, 0, cmd, arg);
1876 EXPORT_SYMBOL(ioctl_by_bdev);
1879 * lookup_bdev - lookup a struct block_device by name
1880 * @pathname: special file representing the block device
1882 * Get a reference to the blockdevice at @pathname in the current
1883 * namespace if possible and return it. Return ERR_PTR(error)
1886 struct block_device *lookup_bdev(const char *pathname)
1888 struct block_device *bdev;
1889 struct inode *inode;
1893 if (!pathname || !*pathname)
1894 return ERR_PTR(-EINVAL);
1896 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
1898 return ERR_PTR(error);
1900 inode = d_backing_inode(path.dentry);
1902 if (!S_ISBLK(inode->i_mode))
1905 if (path.mnt->mnt_flags & MNT_NODEV)
1908 bdev = bd_acquire(inode);
1915 bdev = ERR_PTR(error);
1918 EXPORT_SYMBOL(lookup_bdev);
1920 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
1922 struct super_block *sb = get_super(bdev);
1927 * no need to lock the super, get_super holds the
1928 * read mutex so the filesystem cannot go away
1929 * under us (->put_super runs with the write lock
1932 shrink_dcache_sb(sb);
1933 res = invalidate_inodes(sb, kill_dirty);
1936 invalidate_bdev(bdev);
1939 EXPORT_SYMBOL(__invalidate_device);
1941 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1943 struct inode *inode, *old_inode = NULL;
1945 spin_lock(&blockdev_superblock->s_inode_list_lock);
1946 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1947 struct address_space *mapping = inode->i_mapping;
1949 spin_lock(&inode->i_lock);
1950 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1951 mapping->nrpages == 0) {
1952 spin_unlock(&inode->i_lock);
1956 spin_unlock(&inode->i_lock);
1957 spin_unlock(&blockdev_superblock->s_inode_list_lock);
1959 * We hold a reference to 'inode' so it couldn't have been
1960 * removed from s_inodes list while we dropped the
1961 * s_inode_list_lock We cannot iput the inode now as we can
1962 * be holding the last reference and we cannot iput it under
1963 * s_inode_list_lock. So we keep the reference and iput it
1969 func(I_BDEV(inode), arg);
1971 spin_lock(&blockdev_superblock->s_inode_list_lock);
1973 spin_unlock(&blockdev_superblock->s_inode_list_lock);