1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/buffer_head.h>
23 #include <linux/swap.h>
24 #include <linux/pagevec.h>
25 #include <linux/writeback.h>
26 #include <linux/mpage.h>
27 #include <linux/mount.h>
28 #include <linux/pseudo_fs.h>
29 #include <linux/uio.h>
30 #include <linux/namei.h>
31 #include <linux/log2.h>
32 #include <linux/cleancache.h>
33 #include <linux/task_io_accounting_ops.h>
34 #include <linux/falloc.h>
35 #include <linux/uaccess.h>
36 #include <linux/suspend.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 static void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
89 /* Invalidate clean unused buffers and pagecache. */
90 void invalidate_bdev(struct block_device *bdev)
92 struct address_space *mapping = bdev->bd_inode->i_mapping;
94 if (mapping->nrpages) {
96 lru_add_drain_all(); /* make sure all lru add caches are flushed */
97 invalidate_mapping_pages(mapping, 0, -1);
99 /* 99% of the time, we don't need to flush the cleancache on the bdev.
100 * But, for the strange corners, lets be cautious
102 cleancache_invalidate_inode(mapping);
104 EXPORT_SYMBOL(invalidate_bdev);
107 * Drop all buffers & page cache for given bdev range. This function bails
108 * with error if bdev has other exclusive owner (such as filesystem).
110 int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
111 loff_t lstart, loff_t lend)
113 struct block_device *claimed_bdev = NULL;
117 * If we don't hold exclusive handle for the device, upgrade to it
118 * while we discard the buffer cache to avoid discarding buffers
119 * under live filesystem.
121 if (!(mode & FMODE_EXCL)) {
122 claimed_bdev = bdev->bd_contains;
123 err = bd_prepare_to_claim(bdev, claimed_bdev,
124 truncate_bdev_range);
128 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
130 bd_abort_claiming(bdev, claimed_bdev, truncate_bdev_range);
133 EXPORT_SYMBOL(truncate_bdev_range);
135 static void set_init_blocksize(struct block_device *bdev)
137 bdev->bd_inode->i_blkbits = blksize_bits(bdev_logical_block_size(bdev));
140 int set_blocksize(struct block_device *bdev, int size)
142 /* Size must be a power of two, and between 512 and PAGE_SIZE */
143 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
146 /* Size cannot be smaller than the size supported by the device */
147 if (size < bdev_logical_block_size(bdev))
150 /* Don't change the size if it is same as current */
151 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) {
153 bdev->bd_inode->i_blkbits = blksize_bits(size);
159 EXPORT_SYMBOL(set_blocksize);
161 int sb_set_blocksize(struct super_block *sb, int size)
163 if (set_blocksize(sb->s_bdev, size))
165 /* If we get here, we know size is power of two
166 * and it's value is between 512 and PAGE_SIZE */
167 sb->s_blocksize = size;
168 sb->s_blocksize_bits = blksize_bits(size);
169 return sb->s_blocksize;
172 EXPORT_SYMBOL(sb_set_blocksize);
174 int sb_min_blocksize(struct super_block *sb, int size)
176 int minsize = bdev_logical_block_size(sb->s_bdev);
179 return sb_set_blocksize(sb, size);
182 EXPORT_SYMBOL(sb_min_blocksize);
185 blkdev_get_block(struct inode *inode, sector_t iblock,
186 struct buffer_head *bh, int create)
188 bh->b_bdev = I_BDEV(inode);
189 bh->b_blocknr = iblock;
190 set_buffer_mapped(bh);
194 static struct inode *bdev_file_inode(struct file *file)
196 return file->f_mapping->host;
199 static unsigned int dio_bio_write_op(struct kiocb *iocb)
201 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
203 /* avoid the need for a I/O completion work item */
204 if (iocb->ki_flags & IOCB_DSYNC)
209 #define DIO_INLINE_BIO_VECS 4
211 static void blkdev_bio_end_io_simple(struct bio *bio)
213 struct task_struct *waiter = bio->bi_private;
215 WRITE_ONCE(bio->bi_private, NULL);
216 blk_wake_io_task(waiter);
220 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
223 struct file *file = iocb->ki_filp;
224 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
225 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
226 loff_t pos = iocb->ki_pos;
227 bool should_dirty = false;
232 if ((pos | iov_iter_alignment(iter)) &
233 (bdev_logical_block_size(bdev) - 1))
236 if (nr_pages <= DIO_INLINE_BIO_VECS)
239 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
245 bio_init(&bio, vecs, nr_pages);
246 bio_set_dev(&bio, bdev);
247 bio.bi_iter.bi_sector = pos >> 9;
248 bio.bi_write_hint = iocb->ki_hint;
249 bio.bi_private = current;
250 bio.bi_end_io = blkdev_bio_end_io_simple;
251 bio.bi_ioprio = iocb->ki_ioprio;
253 ret = bio_iov_iter_get_pages(&bio, iter);
256 ret = bio.bi_iter.bi_size;
258 if (iov_iter_rw(iter) == READ) {
259 bio.bi_opf = REQ_OP_READ;
260 if (iter_is_iovec(iter))
263 bio.bi_opf = dio_bio_write_op(iocb);
264 task_io_account_write(ret);
266 if (iocb->ki_flags & IOCB_HIPRI)
267 bio_set_polled(&bio, iocb);
269 qc = submit_bio(&bio);
271 set_current_state(TASK_UNINTERRUPTIBLE);
272 if (!READ_ONCE(bio.bi_private))
274 if (!(iocb->ki_flags & IOCB_HIPRI) ||
275 !blk_poll(bdev_get_queue(bdev), qc, true))
278 __set_current_state(TASK_RUNNING);
280 bio_release_pages(&bio, should_dirty);
281 if (unlikely(bio.bi_status))
282 ret = blk_status_to_errno(bio.bi_status);
285 if (vecs != inline_vecs)
296 struct task_struct *waiter;
301 bool should_dirty : 1;
306 static struct bio_set blkdev_dio_pool;
308 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
310 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
311 struct request_queue *q = bdev_get_queue(bdev);
313 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
316 static void blkdev_bio_end_io(struct bio *bio)
318 struct blkdev_dio *dio = bio->bi_private;
319 bool should_dirty = dio->should_dirty;
321 if (bio->bi_status && !dio->bio.bi_status)
322 dio->bio.bi_status = bio->bi_status;
324 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
326 struct kiocb *iocb = dio->iocb;
329 if (likely(!dio->bio.bi_status)) {
333 ret = blk_status_to_errno(dio->bio.bi_status);
336 dio->iocb->ki_complete(iocb, ret, 0);
340 struct task_struct *waiter = dio->waiter;
342 WRITE_ONCE(dio->waiter, NULL);
343 blk_wake_io_task(waiter);
348 bio_check_pages_dirty(bio);
350 bio_release_pages(bio, false);
356 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
358 struct file *file = iocb->ki_filp;
359 struct inode *inode = bdev_file_inode(file);
360 struct block_device *bdev = I_BDEV(inode);
361 struct blk_plug plug;
362 struct blkdev_dio *dio;
364 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
365 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
366 loff_t pos = iocb->ki_pos;
367 blk_qc_t qc = BLK_QC_T_NONE;
370 if ((pos | iov_iter_alignment(iter)) &
371 (bdev_logical_block_size(bdev) - 1))
374 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
376 dio = container_of(bio, struct blkdev_dio, bio);
377 dio->is_sync = is_sync = is_sync_kiocb(iocb);
379 dio->waiter = current;
386 dio->multi_bio = false;
387 dio->should_dirty = is_read && iter_is_iovec(iter);
390 * Don't plug for HIPRI/polled IO, as those should go straight
394 blk_start_plug(&plug);
397 bio_set_dev(bio, bdev);
398 bio->bi_iter.bi_sector = pos >> 9;
399 bio->bi_write_hint = iocb->ki_hint;
400 bio->bi_private = dio;
401 bio->bi_end_io = blkdev_bio_end_io;
402 bio->bi_ioprio = iocb->ki_ioprio;
404 ret = bio_iov_iter_get_pages(bio, iter);
406 bio->bi_status = BLK_STS_IOERR;
412 bio->bi_opf = REQ_OP_READ;
413 if (dio->should_dirty)
414 bio_set_pages_dirty(bio);
416 bio->bi_opf = dio_bio_write_op(iocb);
417 task_io_account_write(bio->bi_iter.bi_size);
420 dio->size += bio->bi_iter.bi_size;
421 pos += bio->bi_iter.bi_size;
423 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
427 if (iocb->ki_flags & IOCB_HIPRI) {
428 bio_set_polled(bio, iocb);
432 qc = submit_bio(bio);
435 WRITE_ONCE(iocb->ki_cookie, qc);
439 if (!dio->multi_bio) {
441 * AIO needs an extra reference to ensure the dio
442 * structure which is embedded into the first bio
447 dio->multi_bio = true;
448 atomic_set(&dio->ref, 2);
450 atomic_inc(&dio->ref);
454 bio = bio_alloc(GFP_KERNEL, nr_pages);
458 blk_finish_plug(&plug);
464 set_current_state(TASK_UNINTERRUPTIBLE);
465 if (!READ_ONCE(dio->waiter))
468 if (!(iocb->ki_flags & IOCB_HIPRI) ||
469 !blk_poll(bdev_get_queue(bdev), qc, true))
472 __set_current_state(TASK_RUNNING);
475 ret = blk_status_to_errno(dio->bio.bi_status);
484 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
488 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
491 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
492 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
494 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
497 static __init int blkdev_init(void)
499 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
501 module_init(blkdev_init);
503 int __sync_blockdev(struct block_device *bdev, int wait)
508 return filemap_flush(bdev->bd_inode->i_mapping);
509 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
513 * Write out and wait upon all the dirty data associated with a block
514 * device via its mapping. Does not take the superblock lock.
516 int sync_blockdev(struct block_device *bdev)
518 return __sync_blockdev(bdev, 1);
520 EXPORT_SYMBOL(sync_blockdev);
523 * Write out and wait upon all dirty data associated with this
524 * device. Filesystem data as well as the underlying block
525 * device. Takes the superblock lock.
527 int fsync_bdev(struct block_device *bdev)
529 struct super_block *sb = get_super(bdev);
531 int res = sync_filesystem(sb);
535 return sync_blockdev(bdev);
537 EXPORT_SYMBOL(fsync_bdev);
540 * freeze_bdev -- lock a filesystem and force it into a consistent state
541 * @bdev: blockdevice to lock
543 * If a superblock is found on this device, we take the s_umount semaphore
544 * on it to make sure nobody unmounts until the snapshot creation is done.
545 * The reference counter (bd_fsfreeze_count) guarantees that only the last
546 * unfreeze process can unfreeze the frozen filesystem actually when multiple
547 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
548 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
551 int freeze_bdev(struct block_device *bdev)
553 struct super_block *sb;
556 mutex_lock(&bdev->bd_fsfreeze_mutex);
557 if (++bdev->bd_fsfreeze_count > 1)
560 sb = get_active_super(bdev);
563 if (sb->s_op->freeze_super)
564 error = sb->s_op->freeze_super(sb);
566 error = freeze_super(sb);
567 deactivate_super(sb);
570 bdev->bd_fsfreeze_count--;
573 bdev->bd_fsfreeze_sb = sb;
578 mutex_unlock(&bdev->bd_fsfreeze_mutex);
581 EXPORT_SYMBOL(freeze_bdev);
584 * thaw_bdev -- unlock filesystem
585 * @bdev: blockdevice to unlock
587 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
589 int thaw_bdev(struct block_device *bdev)
591 struct super_block *sb;
594 mutex_lock(&bdev->bd_fsfreeze_mutex);
595 if (!bdev->bd_fsfreeze_count)
599 if (--bdev->bd_fsfreeze_count > 0)
602 sb = bdev->bd_fsfreeze_sb;
606 if (sb->s_op->thaw_super)
607 error = sb->s_op->thaw_super(sb);
609 error = thaw_super(sb);
611 bdev->bd_fsfreeze_count++;
613 mutex_unlock(&bdev->bd_fsfreeze_mutex);
616 EXPORT_SYMBOL(thaw_bdev);
618 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
620 return block_write_full_page(page, blkdev_get_block, wbc);
623 static int blkdev_readpage(struct file * file, struct page * page)
625 return block_read_full_page(page, blkdev_get_block);
628 static void blkdev_readahead(struct readahead_control *rac)
630 mpage_readahead(rac, blkdev_get_block);
633 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
634 loff_t pos, unsigned len, unsigned flags,
635 struct page **pagep, void **fsdata)
637 return block_write_begin(mapping, pos, len, flags, pagep,
641 static int blkdev_write_end(struct file *file, struct address_space *mapping,
642 loff_t pos, unsigned len, unsigned copied,
643 struct page *page, void *fsdata)
646 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
656 * for a block special file file_inode(file)->i_size is zero
657 * so we compute the size by hand (just as in block_read/write above)
659 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
661 struct inode *bd_inode = bdev_file_inode(file);
664 inode_lock(bd_inode);
665 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
666 inode_unlock(bd_inode);
670 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
672 struct inode *bd_inode = bdev_file_inode(filp);
673 struct block_device *bdev = I_BDEV(bd_inode);
676 error = file_write_and_wait_range(filp, start, end);
681 * There is no need to serialise calls to blkdev_issue_flush with
682 * i_mutex and doing so causes performance issues with concurrent
683 * O_SYNC writers to a block device.
685 error = blkdev_issue_flush(bdev, GFP_KERNEL);
686 if (error == -EOPNOTSUPP)
691 EXPORT_SYMBOL(blkdev_fsync);
694 * bdev_read_page() - Start reading a page from a block device
695 * @bdev: The device to read the page from
696 * @sector: The offset on the device to read the page to (need not be aligned)
697 * @page: The page to read
699 * On entry, the page should be locked. It will be unlocked when the page
700 * has been read. If the block driver implements rw_page synchronously,
701 * that will be true on exit from this function, but it need not be.
703 * Errors returned by this function are usually "soft", eg out of memory, or
704 * queue full; callers should try a different route to read this page rather
705 * than propagate an error back up the stack.
707 * Return: negative errno if an error occurs, 0 if submission was successful.
709 int bdev_read_page(struct block_device *bdev, sector_t sector,
712 const struct block_device_operations *ops = bdev->bd_disk->fops;
713 int result = -EOPNOTSUPP;
715 if (!ops->rw_page || bdev_get_integrity(bdev))
718 result = blk_queue_enter(bdev->bd_disk->queue, 0);
721 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
723 blk_queue_exit(bdev->bd_disk->queue);
728 * bdev_write_page() - Start writing a page to a block device
729 * @bdev: The device to write the page to
730 * @sector: The offset on the device to write the page to (need not be aligned)
731 * @page: The page to write
732 * @wbc: The writeback_control for the write
734 * On entry, the page should be locked and not currently under writeback.
735 * On exit, if the write started successfully, the page will be unlocked and
736 * under writeback. If the write failed already (eg the driver failed to
737 * queue the page to the device), the page will still be locked. If the
738 * caller is a ->writepage implementation, it will need to unlock the page.
740 * Errors returned by this function are usually "soft", eg out of memory, or
741 * queue full; callers should try a different route to write this page rather
742 * than propagate an error back up the stack.
744 * Return: negative errno if an error occurs, 0 if submission was successful.
746 int bdev_write_page(struct block_device *bdev, sector_t sector,
747 struct page *page, struct writeback_control *wbc)
750 const struct block_device_operations *ops = bdev->bd_disk->fops;
752 if (!ops->rw_page || bdev_get_integrity(bdev))
754 result = blk_queue_enter(bdev->bd_disk->queue, 0);
758 set_page_writeback(page);
759 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
762 end_page_writeback(page);
764 clean_page_buffers(page);
767 blk_queue_exit(bdev->bd_disk->queue);
775 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
776 static struct kmem_cache * bdev_cachep __read_mostly;
778 static struct inode *bdev_alloc_inode(struct super_block *sb)
780 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
783 return &ei->vfs_inode;
786 static void bdev_free_inode(struct inode *inode)
788 kmem_cache_free(bdev_cachep, BDEV_I(inode));
791 static void init_once(void *foo)
793 struct bdev_inode *ei = (struct bdev_inode *) foo;
794 struct block_device *bdev = &ei->bdev;
796 memset(bdev, 0, sizeof(*bdev));
797 mutex_init(&bdev->bd_mutex);
799 INIT_LIST_HEAD(&bdev->bd_holder_disks);
801 bdev->bd_bdi = &noop_backing_dev_info;
802 inode_init_once(&ei->vfs_inode);
803 /* Initialize mutex for freeze. */
804 mutex_init(&bdev->bd_fsfreeze_mutex);
807 static void bdev_evict_inode(struct inode *inode)
809 struct block_device *bdev = &BDEV_I(inode)->bdev;
810 truncate_inode_pages_final(&inode->i_data);
811 invalidate_inode_buffers(inode); /* is it needed here? */
813 /* Detach inode from wb early as bdi_put() may free bdi->wb */
814 inode_detach_wb(inode);
815 if (bdev->bd_bdi != &noop_backing_dev_info) {
816 bdi_put(bdev->bd_bdi);
817 bdev->bd_bdi = &noop_backing_dev_info;
821 static const struct super_operations bdev_sops = {
822 .statfs = simple_statfs,
823 .alloc_inode = bdev_alloc_inode,
824 .free_inode = bdev_free_inode,
825 .drop_inode = generic_delete_inode,
826 .evict_inode = bdev_evict_inode,
829 static int bd_init_fs_context(struct fs_context *fc)
831 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
834 fc->s_iflags |= SB_I_CGROUPWB;
835 ctx->ops = &bdev_sops;
839 static struct file_system_type bd_type = {
841 .init_fs_context = bd_init_fs_context,
842 .kill_sb = kill_anon_super,
845 struct super_block *blockdev_superblock __read_mostly;
846 EXPORT_SYMBOL_GPL(blockdev_superblock);
848 void __init bdev_cache_init(void)
851 static struct vfsmount *bd_mnt;
853 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
854 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
855 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
857 err = register_filesystem(&bd_type);
859 panic("Cannot register bdev pseudo-fs");
860 bd_mnt = kern_mount(&bd_type);
862 panic("Cannot create bdev pseudo-fs");
863 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
866 static struct block_device *bdget(dev_t dev)
868 struct block_device *bdev;
871 inode = iget_locked(blockdev_superblock, dev);
875 bdev = &BDEV_I(inode)->bdev;
877 if (inode->i_state & I_NEW) {
878 spin_lock_init(&bdev->bd_size_lock);
879 bdev->bd_contains = NULL;
880 bdev->bd_super = NULL;
881 bdev->bd_inode = inode;
882 bdev->bd_part_count = 0;
884 inode->i_mode = S_IFBLK;
886 inode->i_bdev = bdev;
887 inode->i_data.a_ops = &def_blk_aops;
888 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
889 unlock_new_inode(inode);
895 * bdgrab -- Grab a reference to an already referenced block device
896 * @bdev: Block device to grab a reference to.
898 struct block_device *bdgrab(struct block_device *bdev)
900 ihold(bdev->bd_inode);
903 EXPORT_SYMBOL(bdgrab);
905 struct block_device *bdget_part(struct hd_struct *part)
907 return bdget(part_devt(part));
910 long nr_blockdev_pages(void)
915 spin_lock(&blockdev_superblock->s_inode_list_lock);
916 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list)
917 ret += inode->i_mapping->nrpages;
918 spin_unlock(&blockdev_superblock->s_inode_list_lock);
923 void bdput(struct block_device *bdev)
925 iput(bdev->bd_inode);
928 EXPORT_SYMBOL(bdput);
930 static struct block_device *bd_acquire(struct inode *inode)
932 struct block_device *bdev;
934 spin_lock(&bdev_lock);
935 bdev = inode->i_bdev;
936 if (bdev && !inode_unhashed(bdev->bd_inode)) {
938 spin_unlock(&bdev_lock);
941 spin_unlock(&bdev_lock);
944 * i_bdev references block device inode that was already shut down
945 * (corresponding device got removed). Remove the reference and look
946 * up block device inode again just in case new device got
947 * reestablished under the same device number.
952 bdev = bdget(inode->i_rdev);
954 spin_lock(&bdev_lock);
955 if (!inode->i_bdev) {
957 * We take an additional reference to bd_inode,
958 * and it's released in clear_inode() of inode.
959 * So, we can access it via ->i_mapping always
963 inode->i_bdev = bdev;
964 inode->i_mapping = bdev->bd_inode->i_mapping;
966 spin_unlock(&bdev_lock);
971 /* Call when you free inode */
973 void bd_forget(struct inode *inode)
975 struct block_device *bdev = NULL;
977 spin_lock(&bdev_lock);
978 if (!sb_is_blkdev_sb(inode->i_sb))
979 bdev = inode->i_bdev;
980 inode->i_bdev = NULL;
981 inode->i_mapping = &inode->i_data;
982 spin_unlock(&bdev_lock);
989 * bd_may_claim - test whether a block device can be claimed
990 * @bdev: block device of interest
991 * @whole: whole block device containing @bdev, may equal @bdev
992 * @holder: holder trying to claim @bdev
994 * Test whether @bdev can be claimed by @holder.
997 * spin_lock(&bdev_lock).
1000 * %true if @bdev can be claimed, %false otherwise.
1002 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1005 if (bdev->bd_holder == holder)
1006 return true; /* already a holder */
1007 else if (bdev->bd_holder != NULL)
1008 return false; /* held by someone else */
1009 else if (whole == bdev)
1010 return true; /* is a whole device which isn't held */
1012 else if (whole->bd_holder == bd_may_claim)
1013 return true; /* is a partition of a device that is being partitioned */
1014 else if (whole->bd_holder != NULL)
1015 return false; /* is a partition of a held device */
1017 return true; /* is a partition of an un-held device */
1021 * bd_prepare_to_claim - claim a block device
1022 * @bdev: block device of interest
1023 * @whole: the whole device containing @bdev, may equal @bdev
1024 * @holder: holder trying to claim @bdev
1026 * Claim @bdev. This function fails if @bdev is already claimed by another
1027 * holder and waits if another claiming is in progress. return, the caller
1028 * has ownership of bd_claiming and bd_holder[s].
1031 * 0 if @bdev can be claimed, -EBUSY otherwise.
1033 int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
1037 spin_lock(&bdev_lock);
1038 /* if someone else claimed, fail */
1039 if (!bd_may_claim(bdev, whole, holder)) {
1040 spin_unlock(&bdev_lock);
1044 /* if claiming is already in progress, wait for it to finish */
1045 if (whole->bd_claiming) {
1046 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1049 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1050 spin_unlock(&bdev_lock);
1052 finish_wait(wq, &wait);
1057 whole->bd_claiming = holder;
1058 spin_unlock(&bdev_lock);
1061 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
1063 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1065 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1070 * Now that we hold gendisk reference we make sure bdev we looked up is
1071 * not stale. If it is, it means device got removed and created before
1072 * we looked up gendisk and we fail open in such case. Associating
1073 * unhashed bdev with newly created gendisk could lead to two bdevs
1074 * (and thus two independent caches) being associated with one device
1077 if (inode_unhashed(bdev->bd_inode)) {
1078 put_disk_and_module(disk);
1084 static void bd_clear_claiming(struct block_device *whole, void *holder)
1086 lockdep_assert_held(&bdev_lock);
1087 /* tell others that we're done */
1088 BUG_ON(whole->bd_claiming != holder);
1089 whole->bd_claiming = NULL;
1090 wake_up_bit(&whole->bd_claiming, 0);
1094 * bd_finish_claiming - finish claiming of a block device
1095 * @bdev: block device of interest
1096 * @whole: whole block device
1097 * @holder: holder that has claimed @bdev
1099 * Finish exclusive open of a block device. Mark the device as exlusively
1100 * open by the holder and wake up all waiters for exclusive open to finish.
1102 static void bd_finish_claiming(struct block_device *bdev,
1103 struct block_device *whole, void *holder)
1105 spin_lock(&bdev_lock);
1106 BUG_ON(!bd_may_claim(bdev, whole, holder));
1108 * Note that for a whole device bd_holders will be incremented twice,
1109 * and bd_holder will be set to bd_may_claim before being set to holder
1111 whole->bd_holders++;
1112 whole->bd_holder = bd_may_claim;
1114 bdev->bd_holder = holder;
1115 bd_clear_claiming(whole, holder);
1116 spin_unlock(&bdev_lock);
1120 * bd_abort_claiming - abort claiming of a block device
1121 * @bdev: block device of interest
1122 * @whole: whole block device
1123 * @holder: holder that has claimed @bdev
1125 * Abort claiming of a block device when the exclusive open failed. This can be
1126 * also used when exclusive open is not actually desired and we just needed
1127 * to block other exclusive openers for a while.
1129 void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
1132 spin_lock(&bdev_lock);
1133 bd_clear_claiming(whole, holder);
1134 spin_unlock(&bdev_lock);
1136 EXPORT_SYMBOL(bd_abort_claiming);
1139 struct bd_holder_disk {
1140 struct list_head list;
1141 struct gendisk *disk;
1145 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1146 struct gendisk *disk)
1148 struct bd_holder_disk *holder;
1150 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1151 if (holder->disk == disk)
1156 static int add_symlink(struct kobject *from, struct kobject *to)
1158 return sysfs_create_link(from, to, kobject_name(to));
1161 static void del_symlink(struct kobject *from, struct kobject *to)
1163 sysfs_remove_link(from, kobject_name(to));
1167 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1168 * @bdev: the claimed slave bdev
1169 * @disk: the holding disk
1171 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1173 * This functions creates the following sysfs symlinks.
1175 * - from "slaves" directory of the holder @disk to the claimed @bdev
1176 * - from "holders" directory of the @bdev to the holder @disk
1178 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1179 * passed to bd_link_disk_holder(), then:
1181 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1182 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1184 * The caller must have claimed @bdev before calling this function and
1185 * ensure that both @bdev and @disk are valid during the creation and
1186 * lifetime of these symlinks.
1192 * 0 on success, -errno on failure.
1194 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1196 struct bd_holder_disk *holder;
1199 mutex_lock(&bdev->bd_mutex);
1201 WARN_ON_ONCE(!bdev->bd_holder);
1203 /* FIXME: remove the following once add_disk() handles errors */
1204 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1207 holder = bd_find_holder_disk(bdev, disk);
1213 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1219 INIT_LIST_HEAD(&holder->list);
1220 holder->disk = disk;
1223 ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
1227 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1231 * bdev could be deleted beneath us which would implicitly destroy
1232 * the holder directory. Hold on to it.
1234 kobject_get(bdev->bd_part->holder_dir);
1236 list_add(&holder->list, &bdev->bd_holder_disks);
1240 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1244 mutex_unlock(&bdev->bd_mutex);
1247 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1250 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1251 * @bdev: the calimed slave bdev
1252 * @disk: the holding disk
1254 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1259 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1261 struct bd_holder_disk *holder;
1263 mutex_lock(&bdev->bd_mutex);
1265 holder = bd_find_holder_disk(bdev, disk);
1267 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1268 del_symlink(disk->slave_dir, bdev_kobj(bdev));
1269 del_symlink(bdev->bd_part->holder_dir,
1270 &disk_to_dev(disk)->kobj);
1271 kobject_put(bdev->bd_part->holder_dir);
1272 list_del_init(&holder->list);
1276 mutex_unlock(&bdev->bd_mutex);
1278 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1282 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1283 * @disk: struct gendisk to check
1284 * @bdev: struct bdev to adjust.
1285 * @verbose: if %true log a message about a size change if there is any
1287 * This routine checks to see if the bdev size does not match the disk size
1288 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1291 static void check_disk_size_change(struct gendisk *disk,
1292 struct block_device *bdev, bool verbose)
1294 loff_t disk_size, bdev_size;
1296 spin_lock(&bdev->bd_size_lock);
1297 disk_size = (loff_t)get_capacity(disk) << 9;
1298 bdev_size = i_size_read(bdev->bd_inode);
1299 if (disk_size != bdev_size) {
1302 "%s: detected capacity change from %lld to %lld\n",
1303 disk->disk_name, bdev_size, disk_size);
1305 i_size_write(bdev->bd_inode, disk_size);
1307 spin_unlock(&bdev->bd_size_lock);
1311 * revalidate_disk_size - checks for disk size change and adjusts bdev size.
1312 * @disk: struct gendisk to check
1313 * @verbose: if %true log a message about a size change if there is any
1315 * This routine checks to see if the bdev size does not match the disk size
1316 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1319 void revalidate_disk_size(struct gendisk *disk, bool verbose)
1321 struct block_device *bdev;
1324 * Hidden disks don't have associated bdev so there's no point in
1325 * revalidating them.
1327 if (disk->flags & GENHD_FL_HIDDEN)
1330 bdev = bdget_disk(disk, 0);
1332 check_disk_size_change(disk, bdev, verbose);
1337 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
1339 spin_lock(&bdev->bd_size_lock);
1340 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
1341 spin_unlock(&bdev->bd_size_lock);
1343 EXPORT_SYMBOL(bd_set_nr_sectors);
1345 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1347 int bdev_disk_changed(struct block_device *bdev, bool invalidate)
1349 struct gendisk *disk = bdev->bd_disk;
1352 lockdep_assert_held(&bdev->bd_mutex);
1354 clear_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state);
1357 ret = blk_drop_partitions(bdev);
1362 * Historically we only set the capacity to zero for devices that
1363 * support partitions (independ of actually having partitions created).
1364 * Doing that is rather inconsistent, but changing it broke legacy
1365 * udisks polling for legacy ide-cdrom devices. Use the crude check
1366 * below to get the sane behavior for most device while not breaking
1367 * userspace for this particular setup.
1370 if (disk_part_scan_enabled(disk) ||
1371 !(disk->flags & GENHD_FL_REMOVABLE))
1372 set_capacity(disk, 0);
1374 if (disk->fops->revalidate_disk)
1375 disk->fops->revalidate_disk(disk);
1378 check_disk_size_change(disk, bdev, !invalidate);
1380 if (get_capacity(disk)) {
1381 ret = blk_add_partitions(disk, bdev);
1384 } else if (invalidate) {
1386 * Tell userspace that the media / partition table may have
1389 kobject_uevent(&disk_to_dev(disk)->kobj, KOBJ_CHANGE);
1395 * Only exported for for loop and dasd for historic reasons. Don't use in new
1398 EXPORT_SYMBOL_GPL(bdev_disk_changed);
1403 * mutex_lock(part->bd_mutex)
1404 * mutex_lock_nested(whole->bd_mutex, 1)
1407 static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
1410 struct block_device *whole = NULL, *claiming = NULL;
1411 struct gendisk *disk;
1414 bool first_open = false, unblock_events = true, need_restart;
1417 need_restart = false;
1419 disk = bdev_get_gendisk(bdev, &partno);
1424 whole = bdget_disk(disk, 0);
1431 if (!for_part && (mode & FMODE_EXCL)) {
1432 WARN_ON_ONCE(!holder);
1437 ret = bd_prepare_to_claim(bdev, claiming, holder);
1442 disk_block_events(disk);
1443 mutex_lock_nested(&bdev->bd_mutex, for_part);
1444 if (!bdev->bd_openers) {
1446 bdev->bd_disk = disk;
1447 bdev->bd_contains = bdev;
1448 bdev->bd_partno = partno;
1452 bdev->bd_part = disk_get_part(disk, partno);
1457 if (disk->fops->open) {
1458 ret = disk->fops->open(bdev, mode);
1460 * If we lost a race with 'disk' being deleted,
1461 * try again. See md.c
1463 if (ret == -ERESTARTSYS)
1464 need_restart = true;
1468 bd_set_nr_sectors(bdev, get_capacity(disk));
1469 set_init_blocksize(bdev);
1473 * If the device is invalidated, rescan partition
1474 * if open succeeded or failed with -ENOMEDIUM.
1475 * The latter is necessary to prevent ghost
1476 * partitions on a removed medium.
1478 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1479 (!ret || ret == -ENOMEDIUM))
1480 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1486 ret = __blkdev_get(whole, mode, NULL, 1);
1489 bdev->bd_contains = bdgrab(whole);
1490 bdev->bd_part = disk_get_part(disk, partno);
1491 if (!(disk->flags & GENHD_FL_UP) ||
1492 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1496 bd_set_nr_sectors(bdev, bdev->bd_part->nr_sects);
1497 set_init_blocksize(bdev);
1500 if (bdev->bd_bdi == &noop_backing_dev_info)
1501 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1503 if (bdev->bd_contains == bdev) {
1505 if (bdev->bd_disk->fops->open)
1506 ret = bdev->bd_disk->fops->open(bdev, mode);
1507 /* the same as first opener case, read comment there */
1508 if (test_bit(GD_NEED_PART_SCAN, &disk->state) &&
1509 (!ret || ret == -ENOMEDIUM))
1510 bdev_disk_changed(bdev, ret == -ENOMEDIUM);
1512 goto out_unlock_bdev;
1517 bdev->bd_part_count++;
1519 bd_finish_claiming(bdev, claiming, holder);
1522 * Block event polling for write claims if requested. Any write holder
1523 * makes the write_holder state stick until all are released. This is
1524 * good enough and tracking individual writeable reference is too
1525 * fragile given the way @mode is used in blkdev_get/put().
1527 if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1528 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1529 bdev->bd_write_holder = true;
1530 unblock_events = false;
1532 mutex_unlock(&bdev->bd_mutex);
1535 disk_unblock_events(disk);
1537 /* only one opener holds refs to the module and disk */
1539 put_disk_and_module(disk);
1545 disk_put_part(bdev->bd_part);
1546 bdev->bd_disk = NULL;
1547 bdev->bd_part = NULL;
1548 if (bdev != bdev->bd_contains)
1549 __blkdev_put(bdev->bd_contains, mode, 1);
1550 bdev->bd_contains = NULL;
1553 bd_abort_claiming(bdev, claiming, holder);
1554 mutex_unlock(&bdev->bd_mutex);
1555 disk_unblock_events(disk);
1560 put_disk_and_module(disk);
1568 * blkdev_get - open a block device
1569 * @bdev: block_device to open
1570 * @mode: FMODE_* mask
1571 * @holder: exclusive holder identifier
1573 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1574 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1575 * @holder is invalid. Exclusive opens may nest for the same @holder.
1577 * On success, the reference count of @bdev is unchanged. On failure,
1584 * 0 on success, -errno on failure.
1586 static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1590 if (mode & FMODE_READ)
1592 if (mode & FMODE_WRITE)
1594 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1598 ret =__blkdev_get(bdev, mode, holder, 0);
1609 * blkdev_get_by_path - open a block device by name
1610 * @path: path to the block device to open
1611 * @mode: FMODE_* mask
1612 * @holder: exclusive holder identifier
1614 * Open the blockdevice described by the device file at @path. @mode
1615 * and @holder are identical to blkdev_get().
1617 * On success, the returned block_device has reference count of one.
1623 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1625 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1628 struct block_device *bdev;
1631 bdev = lookup_bdev(path);
1635 err = blkdev_get(bdev, mode, holder);
1637 return ERR_PTR(err);
1639 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1640 blkdev_put(bdev, mode);
1641 return ERR_PTR(-EACCES);
1646 EXPORT_SYMBOL(blkdev_get_by_path);
1649 * blkdev_get_by_dev - open a block device by device number
1650 * @dev: device number of block device to open
1651 * @mode: FMODE_* mask
1652 * @holder: exclusive holder identifier
1654 * Open the blockdevice described by device number @dev. @mode and
1655 * @holder are identical to blkdev_get().
1657 * Use it ONLY if you really do not have anything better - i.e. when
1658 * you are behind a truly sucky interface and all you are given is a
1659 * device number. _Never_ to be used for internal purposes. If you
1660 * ever need it - reconsider your API.
1662 * On success, the returned block_device has reference count of one.
1668 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1670 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1672 struct block_device *bdev;
1677 return ERR_PTR(-ENOMEM);
1679 err = blkdev_get(bdev, mode, holder);
1681 return ERR_PTR(err);
1685 EXPORT_SYMBOL(blkdev_get_by_dev);
1687 static int blkdev_open(struct inode * inode, struct file * filp)
1689 struct block_device *bdev;
1692 * Preserve backwards compatibility and allow large file access
1693 * even if userspace doesn't ask for it explicitly. Some mkfs
1694 * binary needs it. We might want to drop this workaround
1695 * during an unstable branch.
1697 filp->f_flags |= O_LARGEFILE;
1699 filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1701 if (filp->f_flags & O_NDELAY)
1702 filp->f_mode |= FMODE_NDELAY;
1703 if (filp->f_flags & O_EXCL)
1704 filp->f_mode |= FMODE_EXCL;
1705 if ((filp->f_flags & O_ACCMODE) == 3)
1706 filp->f_mode |= FMODE_WRITE_IOCTL;
1708 bdev = bd_acquire(inode);
1712 filp->f_mapping = bdev->bd_inode->i_mapping;
1713 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1715 return blkdev_get(bdev, filp->f_mode, filp);
1718 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1720 struct gendisk *disk = bdev->bd_disk;
1721 struct block_device *victim = NULL;
1724 * Sync early if it looks like we're the last one. If someone else
1725 * opens the block device between now and the decrement of bd_openers
1726 * then we did a sync that we didn't need to, but that's not the end
1727 * of the world and we want to avoid long (could be several minute)
1728 * syncs while holding the mutex.
1730 if (bdev->bd_openers == 1)
1731 sync_blockdev(bdev);
1733 mutex_lock_nested(&bdev->bd_mutex, for_part);
1735 bdev->bd_part_count--;
1737 if (!--bdev->bd_openers) {
1738 WARN_ON_ONCE(bdev->bd_holders);
1739 sync_blockdev(bdev);
1742 bdev_write_inode(bdev);
1744 if (bdev->bd_contains == bdev) {
1745 if (disk->fops->release)
1746 disk->fops->release(disk, mode);
1748 if (!bdev->bd_openers) {
1749 disk_put_part(bdev->bd_part);
1750 bdev->bd_part = NULL;
1751 bdev->bd_disk = NULL;
1752 if (bdev != bdev->bd_contains)
1753 victim = bdev->bd_contains;
1754 bdev->bd_contains = NULL;
1756 put_disk_and_module(disk);
1758 mutex_unlock(&bdev->bd_mutex);
1761 __blkdev_put(victim, mode, 1);
1764 void blkdev_put(struct block_device *bdev, fmode_t mode)
1766 mutex_lock(&bdev->bd_mutex);
1768 if (mode & FMODE_EXCL) {
1772 * Release a claim on the device. The holder fields
1773 * are protected with bdev_lock. bd_mutex is to
1774 * synchronize disk_holder unlinking.
1776 spin_lock(&bdev_lock);
1778 WARN_ON_ONCE(--bdev->bd_holders < 0);
1779 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1781 /* bd_contains might point to self, check in a separate step */
1782 if ((bdev_free = !bdev->bd_holders))
1783 bdev->bd_holder = NULL;
1784 if (!bdev->bd_contains->bd_holders)
1785 bdev->bd_contains->bd_holder = NULL;
1787 spin_unlock(&bdev_lock);
1790 * If this was the last claim, remove holder link and
1791 * unblock evpoll if it was a write holder.
1793 if (bdev_free && bdev->bd_write_holder) {
1794 disk_unblock_events(bdev->bd_disk);
1795 bdev->bd_write_holder = false;
1800 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1801 * event. This is to ensure detection of media removal commanded
1802 * from userland - e.g. eject(1).
1804 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1806 mutex_unlock(&bdev->bd_mutex);
1808 __blkdev_put(bdev, mode, 0);
1810 EXPORT_SYMBOL(blkdev_put);
1812 static int blkdev_close(struct inode * inode, struct file * filp)
1814 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1815 blkdev_put(bdev, filp->f_mode);
1819 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1821 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1822 fmode_t mode = file->f_mode;
1825 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1826 * to updated it before every ioctl.
1828 if (file->f_flags & O_NDELAY)
1829 mode |= FMODE_NDELAY;
1831 mode &= ~FMODE_NDELAY;
1833 return blkdev_ioctl(bdev, mode, cmd, arg);
1837 * Write data to the block device. Only intended for the block device itself
1838 * and the raw driver which basically is a fake block device.
1840 * Does not take i_mutex for the write and thus is not for general purpose
1843 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1845 struct file *file = iocb->ki_filp;
1846 struct inode *bd_inode = bdev_file_inode(file);
1847 loff_t size = i_size_read(bd_inode);
1848 struct blk_plug plug;
1851 if (bdev_read_only(I_BDEV(bd_inode)))
1854 if (IS_SWAPFILE(bd_inode) && !is_hibernate_resume_dev(bd_inode->i_rdev))
1857 if (!iov_iter_count(from))
1860 if (iocb->ki_pos >= size)
1863 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1866 iov_iter_truncate(from, size - iocb->ki_pos);
1868 blk_start_plug(&plug);
1869 ret = __generic_file_write_iter(iocb, from);
1871 ret = generic_write_sync(iocb, ret);
1872 blk_finish_plug(&plug);
1875 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1877 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1879 struct file *file = iocb->ki_filp;
1880 struct inode *bd_inode = bdev_file_inode(file);
1881 loff_t size = i_size_read(bd_inode);
1882 loff_t pos = iocb->ki_pos;
1888 iov_iter_truncate(to, size);
1889 return generic_file_read_iter(iocb, to);
1891 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1894 * Try to release a page associated with block device when the system
1895 * is under memory pressure.
1897 static int blkdev_releasepage(struct page *page, gfp_t wait)
1899 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1901 if (super && super->s_op->bdev_try_to_free_page)
1902 return super->s_op->bdev_try_to_free_page(super, page, wait);
1904 return try_to_free_buffers(page);
1907 static int blkdev_writepages(struct address_space *mapping,
1908 struct writeback_control *wbc)
1910 return generic_writepages(mapping, wbc);
1913 static const struct address_space_operations def_blk_aops = {
1914 .readpage = blkdev_readpage,
1915 .readahead = blkdev_readahead,
1916 .writepage = blkdev_writepage,
1917 .write_begin = blkdev_write_begin,
1918 .write_end = blkdev_write_end,
1919 .writepages = blkdev_writepages,
1920 .releasepage = blkdev_releasepage,
1921 .direct_IO = blkdev_direct_IO,
1922 .migratepage = buffer_migrate_page_norefs,
1923 .is_dirty_writeback = buffer_check_dirty_writeback,
1926 #define BLKDEV_FALLOC_FL_SUPPORTED \
1927 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1928 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1930 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1933 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1934 loff_t end = start + len - 1;
1938 /* Fail if we don't recognize the flags. */
1939 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1942 /* Don't go off the end of the device. */
1943 isize = i_size_read(bdev->bd_inode);
1947 if (mode & FALLOC_FL_KEEP_SIZE) {
1948 len = isize - start;
1949 end = start + len - 1;
1955 * Don't allow IO that isn't aligned to logical block size.
1957 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
1960 /* Invalidate the page cache, including dirty pages. */
1961 error = truncate_bdev_range(bdev, file->f_mode, start, end);
1966 case FALLOC_FL_ZERO_RANGE:
1967 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
1968 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1969 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
1971 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
1972 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
1973 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
1975 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
1976 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
1986 * Invalidate again; if someone wandered in and dirtied a page,
1987 * the caller will be given -EBUSY. The third argument is
1988 * inclusive, so the rounding here is safe.
1990 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
1991 start >> PAGE_SHIFT,
1995 const struct file_operations def_blk_fops = {
1996 .open = blkdev_open,
1997 .release = blkdev_close,
1998 .llseek = block_llseek,
1999 .read_iter = blkdev_read_iter,
2000 .write_iter = blkdev_write_iter,
2001 .iopoll = blkdev_iopoll,
2002 .mmap = generic_file_mmap,
2003 .fsync = blkdev_fsync,
2004 .unlocked_ioctl = block_ioctl,
2005 #ifdef CONFIG_COMPAT
2006 .compat_ioctl = compat_blkdev_ioctl,
2008 .splice_read = generic_file_splice_read,
2009 .splice_write = iter_file_splice_write,
2010 .fallocate = blkdev_fallocate,
2014 * lookup_bdev - lookup a struct block_device by name
2015 * @pathname: special file representing the block device
2017 * Get a reference to the blockdevice at @pathname in the current
2018 * namespace if possible and return it. Return ERR_PTR(error)
2021 struct block_device *lookup_bdev(const char *pathname)
2023 struct block_device *bdev;
2024 struct inode *inode;
2028 if (!pathname || !*pathname)
2029 return ERR_PTR(-EINVAL);
2031 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2033 return ERR_PTR(error);
2035 inode = d_backing_inode(path.dentry);
2037 if (!S_ISBLK(inode->i_mode))
2040 if (!may_open_dev(&path))
2043 bdev = bd_acquire(inode);
2050 bdev = ERR_PTR(error);
2053 EXPORT_SYMBOL(lookup_bdev);
2055 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2057 struct super_block *sb = get_super(bdev);
2062 * no need to lock the super, get_super holds the
2063 * read mutex so the filesystem cannot go away
2064 * under us (->put_super runs with the write lock
2067 shrink_dcache_sb(sb);
2068 res = invalidate_inodes(sb, kill_dirty);
2071 invalidate_bdev(bdev);
2074 EXPORT_SYMBOL(__invalidate_device);
2076 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2078 struct inode *inode, *old_inode = NULL;
2080 spin_lock(&blockdev_superblock->s_inode_list_lock);
2081 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2082 struct address_space *mapping = inode->i_mapping;
2083 struct block_device *bdev;
2085 spin_lock(&inode->i_lock);
2086 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2087 mapping->nrpages == 0) {
2088 spin_unlock(&inode->i_lock);
2092 spin_unlock(&inode->i_lock);
2093 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2095 * We hold a reference to 'inode' so it couldn't have been
2096 * removed from s_inodes list while we dropped the
2097 * s_inode_list_lock We cannot iput the inode now as we can
2098 * be holding the last reference and we cannot iput it under
2099 * s_inode_list_lock. So we keep the reference and iput it
2104 bdev = I_BDEV(inode);
2106 mutex_lock(&bdev->bd_mutex);
2107 if (bdev->bd_openers)
2109 mutex_unlock(&bdev->bd_mutex);
2111 spin_lock(&blockdev_superblock->s_inode_list_lock);
2113 spin_unlock(&blockdev_superblock->s_inode_list_lock);