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/dax.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/uio.h>
29 #include <linux/namei.h>
30 #include <linux/log2.h>
31 #include <linux/cleancache.h>
32 #include <linux/dax.h>
33 #include <linux/badblocks.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.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 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);
88 EXPORT_SYMBOL(kill_bdev);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device *bdev)
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
95 if (mapping->nrpages) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 static void set_init_blocksize(struct block_device *bdev)
109 unsigned bsize = bdev_logical_block_size(bdev);
110 loff_t size = i_size_read(bdev->bd_inode);
112 while (bsize < PAGE_SIZE) {
117 bdev->bd_block_size = bsize;
118 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
121 int set_blocksize(struct block_device *bdev, int size)
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
124 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
127 /* Size cannot be smaller than the size supported by the device */
128 if (size < bdev_logical_block_size(bdev))
131 /* Don't change the size if it is same as current */
132 if (bdev->bd_block_size != size) {
134 bdev->bd_block_size = size;
135 bdev->bd_inode->i_blkbits = blksize_bits(size);
141 EXPORT_SYMBOL(set_blocksize);
143 int sb_set_blocksize(struct super_block *sb, int size)
145 if (set_blocksize(sb->s_bdev, size))
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb->s_blocksize = size;
150 sb->s_blocksize_bits = blksize_bits(size);
151 return sb->s_blocksize;
154 EXPORT_SYMBOL(sb_set_blocksize);
156 int sb_min_blocksize(struct super_block *sb, int size)
158 int minsize = bdev_logical_block_size(sb->s_bdev);
161 return sb_set_blocksize(sb, size);
164 EXPORT_SYMBOL(sb_min_blocksize);
167 blkdev_get_block(struct inode *inode, sector_t iblock,
168 struct buffer_head *bh, int create)
170 bh->b_bdev = I_BDEV(inode);
171 bh->b_blocknr = iblock;
172 set_buffer_mapped(bh);
176 static struct inode *bdev_file_inode(struct file *file)
178 return file->f_mapping->host;
181 static unsigned int dio_bio_write_op(struct kiocb *iocb)
183 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
185 /* avoid the need for a I/O completion work item */
186 if (iocb->ki_flags & IOCB_DSYNC)
191 #define DIO_INLINE_BIO_VECS 4
193 static void blkdev_bio_end_io_simple(struct bio *bio)
195 struct task_struct *waiter = bio->bi_private;
197 WRITE_ONCE(bio->bi_private, NULL);
198 blk_wake_io_task(waiter);
202 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
205 struct file *file = iocb->ki_filp;
206 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
207 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
208 loff_t pos = iocb->ki_pos;
209 bool should_dirty = false;
214 struct bvec_iter_all iter_all;
216 if ((pos | iov_iter_alignment(iter)) &
217 (bdev_logical_block_size(bdev) - 1))
220 if (nr_pages <= DIO_INLINE_BIO_VECS)
223 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
229 bio_init(&bio, vecs, nr_pages);
230 bio_set_dev(&bio, bdev);
231 bio.bi_iter.bi_sector = pos >> 9;
232 bio.bi_write_hint = iocb->ki_hint;
233 bio.bi_private = current;
234 bio.bi_end_io = blkdev_bio_end_io_simple;
235 bio.bi_ioprio = iocb->ki_ioprio;
237 ret = bio_iov_iter_get_pages(&bio, iter);
240 ret = bio.bi_iter.bi_size;
242 if (iov_iter_rw(iter) == READ) {
243 bio.bi_opf = REQ_OP_READ;
244 if (iter_is_iovec(iter))
247 bio.bi_opf = dio_bio_write_op(iocb);
248 task_io_account_write(ret);
250 if (iocb->ki_flags & IOCB_HIPRI)
251 bio_set_polled(&bio, iocb);
253 qc = submit_bio(&bio);
255 set_current_state(TASK_UNINTERRUPTIBLE);
256 if (!READ_ONCE(bio.bi_private))
258 if (!(iocb->ki_flags & IOCB_HIPRI) ||
259 !blk_poll(bdev_get_queue(bdev), qc, true))
262 __set_current_state(TASK_RUNNING);
264 bio_for_each_segment_all(bvec, &bio, i, iter_all) {
265 if (should_dirty && !PageCompound(bvec->bv_page))
266 set_page_dirty_lock(bvec->bv_page);
267 if (!bio_flagged(&bio, BIO_NO_PAGE_REF))
268 put_page(bvec->bv_page);
271 if (unlikely(bio.bi_status))
272 ret = blk_status_to_errno(bio.bi_status);
275 if (vecs != inline_vecs)
286 struct task_struct *waiter;
291 bool should_dirty : 1;
296 static struct bio_set blkdev_dio_pool;
298 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
300 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
301 struct request_queue *q = bdev_get_queue(bdev);
303 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
306 static void blkdev_bio_end_io(struct bio *bio)
308 struct blkdev_dio *dio = bio->bi_private;
309 bool should_dirty = dio->should_dirty;
311 if (bio->bi_status && !dio->bio.bi_status)
312 dio->bio.bi_status = bio->bi_status;
314 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
316 struct kiocb *iocb = dio->iocb;
319 if (likely(!dio->bio.bi_status)) {
323 ret = blk_status_to_errno(dio->bio.bi_status);
326 dio->iocb->ki_complete(iocb, ret, 0);
330 struct task_struct *waiter = dio->waiter;
332 WRITE_ONCE(dio->waiter, NULL);
333 blk_wake_io_task(waiter);
338 bio_check_pages_dirty(bio);
340 if (!bio_flagged(bio, BIO_NO_PAGE_REF)) {
341 struct bvec_iter_all iter_all;
342 struct bio_vec *bvec;
345 bio_for_each_segment_all(bvec, bio, i, iter_all)
346 put_page(bvec->bv_page);
353 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
355 struct file *file = iocb->ki_filp;
356 struct inode *inode = bdev_file_inode(file);
357 struct block_device *bdev = I_BDEV(inode);
358 struct blk_plug plug;
359 struct blkdev_dio *dio;
361 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
362 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
363 loff_t pos = iocb->ki_pos;
364 blk_qc_t qc = BLK_QC_T_NONE;
367 if ((pos | iov_iter_alignment(iter)) &
368 (bdev_logical_block_size(bdev) - 1))
371 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
373 dio = container_of(bio, struct blkdev_dio, bio);
374 dio->is_sync = is_sync = is_sync_kiocb(iocb);
376 dio->waiter = current;
383 dio->multi_bio = false;
384 dio->should_dirty = is_read && iter_is_iovec(iter);
387 * Don't plug for HIPRI/polled IO, as those should go straight
391 blk_start_plug(&plug);
394 bio_set_dev(bio, bdev);
395 bio->bi_iter.bi_sector = pos >> 9;
396 bio->bi_write_hint = iocb->ki_hint;
397 bio->bi_private = dio;
398 bio->bi_end_io = blkdev_bio_end_io;
399 bio->bi_ioprio = iocb->ki_ioprio;
401 ret = bio_iov_iter_get_pages(bio, iter);
403 bio->bi_status = BLK_STS_IOERR;
409 bio->bi_opf = REQ_OP_READ;
410 if (dio->should_dirty)
411 bio_set_pages_dirty(bio);
413 bio->bi_opf = dio_bio_write_op(iocb);
414 task_io_account_write(bio->bi_iter.bi_size);
417 dio->size += bio->bi_iter.bi_size;
418 pos += bio->bi_iter.bi_size;
420 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
424 if (iocb->ki_flags & IOCB_HIPRI) {
425 bio_set_polled(bio, iocb);
429 qc = submit_bio(bio);
432 WRITE_ONCE(iocb->ki_cookie, qc);
436 if (!dio->multi_bio) {
438 * AIO needs an extra reference to ensure the dio
439 * structure which is embedded into the first bio
444 dio->multi_bio = true;
445 atomic_set(&dio->ref, 2);
447 atomic_inc(&dio->ref);
451 bio = bio_alloc(GFP_KERNEL, nr_pages);
455 blk_finish_plug(&plug);
461 set_current_state(TASK_UNINTERRUPTIBLE);
462 if (!READ_ONCE(dio->waiter))
465 if (!(iocb->ki_flags & IOCB_HIPRI) ||
466 !blk_poll(bdev_get_queue(bdev), qc, true))
469 __set_current_state(TASK_RUNNING);
472 ret = blk_status_to_errno(dio->bio.bi_status);
481 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
485 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
488 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
489 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
491 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
494 static __init int blkdev_init(void)
496 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
498 module_init(blkdev_init);
500 int __sync_blockdev(struct block_device *bdev, int wait)
505 return filemap_flush(bdev->bd_inode->i_mapping);
506 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
510 * Write out and wait upon all the dirty data associated with a block
511 * device via its mapping. Does not take the superblock lock.
513 int sync_blockdev(struct block_device *bdev)
515 return __sync_blockdev(bdev, 1);
517 EXPORT_SYMBOL(sync_blockdev);
520 * Write out and wait upon all dirty data associated with this
521 * device. Filesystem data as well as the underlying block
522 * device. Takes the superblock lock.
524 int fsync_bdev(struct block_device *bdev)
526 struct super_block *sb = get_super(bdev);
528 int res = sync_filesystem(sb);
532 return sync_blockdev(bdev);
534 EXPORT_SYMBOL(fsync_bdev);
537 * freeze_bdev -- lock a filesystem and force it into a consistent state
538 * @bdev: blockdevice to lock
540 * If a superblock is found on this device, we take the s_umount semaphore
541 * on it to make sure nobody unmounts until the snapshot creation is done.
542 * The reference counter (bd_fsfreeze_count) guarantees that only the last
543 * unfreeze process can unfreeze the frozen filesystem actually when multiple
544 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
545 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
548 struct super_block *freeze_bdev(struct block_device *bdev)
550 struct super_block *sb;
553 mutex_lock(&bdev->bd_fsfreeze_mutex);
554 if (++bdev->bd_fsfreeze_count > 1) {
556 * We don't even need to grab a reference - the first call
557 * to freeze_bdev grab an active reference and only the last
558 * thaw_bdev drops it.
560 sb = get_super(bdev);
563 mutex_unlock(&bdev->bd_fsfreeze_mutex);
567 sb = get_active_super(bdev);
570 if (sb->s_op->freeze_super)
571 error = sb->s_op->freeze_super(sb);
573 error = freeze_super(sb);
575 deactivate_super(sb);
576 bdev->bd_fsfreeze_count--;
577 mutex_unlock(&bdev->bd_fsfreeze_mutex);
578 return ERR_PTR(error);
580 deactivate_super(sb);
583 mutex_unlock(&bdev->bd_fsfreeze_mutex);
584 return sb; /* thaw_bdev releases s->s_umount */
586 EXPORT_SYMBOL(freeze_bdev);
589 * thaw_bdev -- unlock filesystem
590 * @bdev: blockdevice to unlock
591 * @sb: associated superblock
593 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
595 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
599 mutex_lock(&bdev->bd_fsfreeze_mutex);
600 if (!bdev->bd_fsfreeze_count)
604 if (--bdev->bd_fsfreeze_count > 0)
610 if (sb->s_op->thaw_super)
611 error = sb->s_op->thaw_super(sb);
613 error = thaw_super(sb);
615 bdev->bd_fsfreeze_count++;
617 mutex_unlock(&bdev->bd_fsfreeze_mutex);
620 EXPORT_SYMBOL(thaw_bdev);
622 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
624 return block_write_full_page(page, blkdev_get_block, wbc);
627 static int blkdev_readpage(struct file * file, struct page * page)
629 return block_read_full_page(page, blkdev_get_block);
632 static int blkdev_readpages(struct file *file, struct address_space *mapping,
633 struct list_head *pages, unsigned nr_pages)
635 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
638 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
639 loff_t pos, unsigned len, unsigned flags,
640 struct page **pagep, void **fsdata)
642 return block_write_begin(mapping, pos, len, flags, pagep,
646 static int blkdev_write_end(struct file *file, struct address_space *mapping,
647 loff_t pos, unsigned len, unsigned copied,
648 struct page *page, void *fsdata)
651 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
661 * for a block special file file_inode(file)->i_size is zero
662 * so we compute the size by hand (just as in block_read/write above)
664 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
666 struct inode *bd_inode = bdev_file_inode(file);
669 inode_lock(bd_inode);
670 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
671 inode_unlock(bd_inode);
675 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
677 struct inode *bd_inode = bdev_file_inode(filp);
678 struct block_device *bdev = I_BDEV(bd_inode);
681 error = file_write_and_wait_range(filp, start, end);
686 * There is no need to serialise calls to blkdev_issue_flush with
687 * i_mutex and doing so causes performance issues with concurrent
688 * O_SYNC writers to a block device.
690 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
691 if (error == -EOPNOTSUPP)
696 EXPORT_SYMBOL(blkdev_fsync);
699 * bdev_read_page() - Start reading a page from a block device
700 * @bdev: The device to read the page from
701 * @sector: The offset on the device to read the page to (need not be aligned)
702 * @page: The page to read
704 * On entry, the page should be locked. It will be unlocked when the page
705 * has been read. If the block driver implements rw_page synchronously,
706 * that will be true on exit from this function, but it need not be.
708 * Errors returned by this function are usually "soft", eg out of memory, or
709 * queue full; callers should try a different route to read this page rather
710 * than propagate an error back up the stack.
712 * Return: negative errno if an error occurs, 0 if submission was successful.
714 int bdev_read_page(struct block_device *bdev, sector_t sector,
717 const struct block_device_operations *ops = bdev->bd_disk->fops;
718 int result = -EOPNOTSUPP;
720 if (!ops->rw_page || bdev_get_integrity(bdev))
723 result = blk_queue_enter(bdev->bd_queue, 0);
726 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
728 blk_queue_exit(bdev->bd_queue);
731 EXPORT_SYMBOL_GPL(bdev_read_page);
734 * bdev_write_page() - Start writing a page to a block device
735 * @bdev: The device to write the page to
736 * @sector: The offset on the device to write the page to (need not be aligned)
737 * @page: The page to write
738 * @wbc: The writeback_control for the write
740 * On entry, the page should be locked and not currently under writeback.
741 * On exit, if the write started successfully, the page will be unlocked and
742 * under writeback. If the write failed already (eg the driver failed to
743 * queue the page to the device), the page will still be locked. If the
744 * caller is a ->writepage implementation, it will need to unlock the page.
746 * Errors returned by this function are usually "soft", eg out of memory, or
747 * queue full; callers should try a different route to write this page rather
748 * than propagate an error back up the stack.
750 * Return: negative errno if an error occurs, 0 if submission was successful.
752 int bdev_write_page(struct block_device *bdev, sector_t sector,
753 struct page *page, struct writeback_control *wbc)
756 const struct block_device_operations *ops = bdev->bd_disk->fops;
758 if (!ops->rw_page || bdev_get_integrity(bdev))
760 result = blk_queue_enter(bdev->bd_queue, 0);
764 set_page_writeback(page);
765 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
768 end_page_writeback(page);
770 clean_page_buffers(page);
773 blk_queue_exit(bdev->bd_queue);
776 EXPORT_SYMBOL_GPL(bdev_write_page);
782 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
783 static struct kmem_cache * bdev_cachep __read_mostly;
785 static struct inode *bdev_alloc_inode(struct super_block *sb)
787 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
790 return &ei->vfs_inode;
793 static void bdev_free_inode(struct inode *inode)
795 kmem_cache_free(bdev_cachep, BDEV_I(inode));
798 static void init_once(void *foo)
800 struct bdev_inode *ei = (struct bdev_inode *) foo;
801 struct block_device *bdev = &ei->bdev;
803 memset(bdev, 0, sizeof(*bdev));
804 mutex_init(&bdev->bd_mutex);
805 INIT_LIST_HEAD(&bdev->bd_list);
807 INIT_LIST_HEAD(&bdev->bd_holder_disks);
809 bdev->bd_bdi = &noop_backing_dev_info;
810 inode_init_once(&ei->vfs_inode);
811 /* Initialize mutex for freeze. */
812 mutex_init(&bdev->bd_fsfreeze_mutex);
815 static void bdev_evict_inode(struct inode *inode)
817 struct block_device *bdev = &BDEV_I(inode)->bdev;
818 truncate_inode_pages_final(&inode->i_data);
819 invalidate_inode_buffers(inode); /* is it needed here? */
821 spin_lock(&bdev_lock);
822 list_del_init(&bdev->bd_list);
823 spin_unlock(&bdev_lock);
824 /* Detach inode from wb early as bdi_put() may free bdi->wb */
825 inode_detach_wb(inode);
826 if (bdev->bd_bdi != &noop_backing_dev_info) {
827 bdi_put(bdev->bd_bdi);
828 bdev->bd_bdi = &noop_backing_dev_info;
832 static const struct super_operations bdev_sops = {
833 .statfs = simple_statfs,
834 .alloc_inode = bdev_alloc_inode,
835 .free_inode = bdev_free_inode,
836 .drop_inode = generic_delete_inode,
837 .evict_inode = bdev_evict_inode,
840 static struct dentry *bd_mount(struct file_system_type *fs_type,
841 int flags, const char *dev_name, void *data)
844 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
846 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
850 static struct file_system_type bd_type = {
853 .kill_sb = kill_anon_super,
856 struct super_block *blockdev_superblock __read_mostly;
857 EXPORT_SYMBOL_GPL(blockdev_superblock);
859 void __init bdev_cache_init(void)
862 static struct vfsmount *bd_mnt;
864 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
865 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
866 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
868 err = register_filesystem(&bd_type);
870 panic("Cannot register bdev pseudo-fs");
871 bd_mnt = kern_mount(&bd_type);
873 panic("Cannot create bdev pseudo-fs");
874 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
878 * Most likely _very_ bad one - but then it's hardly critical for small
879 * /dev and can be fixed when somebody will need really large one.
880 * Keep in mind that it will be fed through icache hash function too.
882 static inline unsigned long hash(dev_t dev)
884 return MAJOR(dev)+MINOR(dev);
887 static int bdev_test(struct inode *inode, void *data)
889 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
892 static int bdev_set(struct inode *inode, void *data)
894 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
898 static LIST_HEAD(all_bdevs);
901 * If there is a bdev inode for this device, unhash it so that it gets evicted
902 * as soon as last inode reference is dropped.
904 void bdev_unhash_inode(dev_t dev)
908 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
910 remove_inode_hash(inode);
915 struct block_device *bdget(dev_t dev)
917 struct block_device *bdev;
920 inode = iget5_locked(blockdev_superblock, hash(dev),
921 bdev_test, bdev_set, &dev);
926 bdev = &BDEV_I(inode)->bdev;
928 if (inode->i_state & I_NEW) {
929 bdev->bd_contains = NULL;
930 bdev->bd_super = NULL;
931 bdev->bd_inode = inode;
932 bdev->bd_block_size = i_blocksize(inode);
933 bdev->bd_part_count = 0;
934 bdev->bd_invalidated = 0;
935 inode->i_mode = S_IFBLK;
937 inode->i_bdev = bdev;
938 inode->i_data.a_ops = &def_blk_aops;
939 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
940 spin_lock(&bdev_lock);
941 list_add(&bdev->bd_list, &all_bdevs);
942 spin_unlock(&bdev_lock);
943 unlock_new_inode(inode);
948 EXPORT_SYMBOL(bdget);
951 * bdgrab -- Grab a reference to an already referenced block device
952 * @bdev: Block device to grab a reference to.
954 struct block_device *bdgrab(struct block_device *bdev)
956 ihold(bdev->bd_inode);
959 EXPORT_SYMBOL(bdgrab);
961 long nr_blockdev_pages(void)
963 struct block_device *bdev;
965 spin_lock(&bdev_lock);
966 list_for_each_entry(bdev, &all_bdevs, bd_list) {
967 ret += bdev->bd_inode->i_mapping->nrpages;
969 spin_unlock(&bdev_lock);
973 void bdput(struct block_device *bdev)
975 iput(bdev->bd_inode);
978 EXPORT_SYMBOL(bdput);
980 static struct block_device *bd_acquire(struct inode *inode)
982 struct block_device *bdev;
984 spin_lock(&bdev_lock);
985 bdev = inode->i_bdev;
986 if (bdev && !inode_unhashed(bdev->bd_inode)) {
988 spin_unlock(&bdev_lock);
991 spin_unlock(&bdev_lock);
994 * i_bdev references block device inode that was already shut down
995 * (corresponding device got removed). Remove the reference and look
996 * up block device inode again just in case new device got
997 * reestablished under the same device number.
1002 bdev = bdget(inode->i_rdev);
1004 spin_lock(&bdev_lock);
1005 if (!inode->i_bdev) {
1007 * We take an additional reference to bd_inode,
1008 * and it's released in clear_inode() of inode.
1009 * So, we can access it via ->i_mapping always
1013 inode->i_bdev = bdev;
1014 inode->i_mapping = bdev->bd_inode->i_mapping;
1016 spin_unlock(&bdev_lock);
1021 /* Call when you free inode */
1023 void bd_forget(struct inode *inode)
1025 struct block_device *bdev = NULL;
1027 spin_lock(&bdev_lock);
1028 if (!sb_is_blkdev_sb(inode->i_sb))
1029 bdev = inode->i_bdev;
1030 inode->i_bdev = NULL;
1031 inode->i_mapping = &inode->i_data;
1032 spin_unlock(&bdev_lock);
1039 * bd_may_claim - test whether a block device can be claimed
1040 * @bdev: block device of interest
1041 * @whole: whole block device containing @bdev, may equal @bdev
1042 * @holder: holder trying to claim @bdev
1044 * Test whether @bdev can be claimed by @holder.
1047 * spin_lock(&bdev_lock).
1050 * %true if @bdev can be claimed, %false otherwise.
1052 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1055 if (bdev->bd_holder == holder)
1056 return true; /* already a holder */
1057 else if (bdev->bd_holder != NULL)
1058 return false; /* held by someone else */
1059 else if (whole == bdev)
1060 return true; /* is a whole device which isn't held */
1062 else if (whole->bd_holder == bd_may_claim)
1063 return true; /* is a partition of a device that is being partitioned */
1064 else if (whole->bd_holder != NULL)
1065 return false; /* is a partition of a held device */
1067 return true; /* is a partition of an un-held device */
1071 * bd_prepare_to_claim - prepare to claim a block device
1072 * @bdev: block device of interest
1073 * @whole: the whole device containing @bdev, may equal @bdev
1074 * @holder: holder trying to claim @bdev
1076 * Prepare to claim @bdev. This function fails if @bdev is already
1077 * claimed by another holder and waits if another claiming is in
1078 * progress. This function doesn't actually claim. On successful
1079 * return, the caller has ownership of bd_claiming and bd_holder[s].
1082 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1083 * it multiple times.
1086 * 0 if @bdev can be claimed, -EBUSY otherwise.
1088 static int bd_prepare_to_claim(struct block_device *bdev,
1089 struct block_device *whole, void *holder)
1092 /* if someone else claimed, fail */
1093 if (!bd_may_claim(bdev, whole, holder))
1096 /* if claiming is already in progress, wait for it to finish */
1097 if (whole->bd_claiming) {
1098 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1101 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1102 spin_unlock(&bdev_lock);
1104 finish_wait(wq, &wait);
1105 spin_lock(&bdev_lock);
1113 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1115 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1120 * Now that we hold gendisk reference we make sure bdev we looked up is
1121 * not stale. If it is, it means device got removed and created before
1122 * we looked up gendisk and we fail open in such case. Associating
1123 * unhashed bdev with newly created gendisk could lead to two bdevs
1124 * (and thus two independent caches) being associated with one device
1127 if (inode_unhashed(bdev->bd_inode)) {
1128 put_disk_and_module(disk);
1135 * bd_start_claiming - start claiming a block device
1136 * @bdev: block device of interest
1137 * @holder: holder trying to claim @bdev
1139 * @bdev is about to be opened exclusively. Check @bdev can be opened
1140 * exclusively and mark that an exclusive open is in progress. Each
1141 * successful call to this function must be matched with a call to
1142 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1145 * This function is used to gain exclusive access to the block device
1146 * without actually causing other exclusive open attempts to fail. It
1147 * should be used when the open sequence itself requires exclusive
1148 * access but may subsequently fail.
1154 * Pointer to the block device containing @bdev on success, ERR_PTR()
1157 static struct block_device *bd_start_claiming(struct block_device *bdev,
1160 struct gendisk *disk;
1161 struct block_device *whole;
1167 * @bdev might not have been initialized properly yet, look up
1168 * and grab the outer block device the hard way.
1170 disk = bdev_get_gendisk(bdev, &partno);
1172 return ERR_PTR(-ENXIO);
1175 * Normally, @bdev should equal what's returned from bdget_disk()
1176 * if partno is 0; however, some drivers (floppy) use multiple
1177 * bdev's for the same physical device and @bdev may be one of the
1178 * aliases. Keep @bdev if partno is 0. This means claimer
1179 * tracking is broken for those devices but it has always been that
1183 whole = bdget_disk(disk, 0);
1185 whole = bdgrab(bdev);
1187 put_disk_and_module(disk);
1189 return ERR_PTR(-ENOMEM);
1191 /* prepare to claim, if successful, mark claiming in progress */
1192 spin_lock(&bdev_lock);
1194 err = bd_prepare_to_claim(bdev, whole, holder);
1196 whole->bd_claiming = holder;
1197 spin_unlock(&bdev_lock);
1200 spin_unlock(&bdev_lock);
1202 return ERR_PTR(err);
1207 struct bd_holder_disk {
1208 struct list_head list;
1209 struct gendisk *disk;
1213 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1214 struct gendisk *disk)
1216 struct bd_holder_disk *holder;
1218 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1219 if (holder->disk == disk)
1224 static int add_symlink(struct kobject *from, struct kobject *to)
1226 return sysfs_create_link(from, to, kobject_name(to));
1229 static void del_symlink(struct kobject *from, struct kobject *to)
1231 sysfs_remove_link(from, kobject_name(to));
1235 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1236 * @bdev: the claimed slave bdev
1237 * @disk: the holding disk
1239 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1241 * This functions creates the following sysfs symlinks.
1243 * - from "slaves" directory of the holder @disk to the claimed @bdev
1244 * - from "holders" directory of the @bdev to the holder @disk
1246 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1247 * passed to bd_link_disk_holder(), then:
1249 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1250 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1252 * The caller must have claimed @bdev before calling this function and
1253 * ensure that both @bdev and @disk are valid during the creation and
1254 * lifetime of these symlinks.
1260 * 0 on success, -errno on failure.
1262 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1264 struct bd_holder_disk *holder;
1267 mutex_lock(&bdev->bd_mutex);
1269 WARN_ON_ONCE(!bdev->bd_holder);
1271 /* FIXME: remove the following once add_disk() handles errors */
1272 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1275 holder = bd_find_holder_disk(bdev, disk);
1281 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1287 INIT_LIST_HEAD(&holder->list);
1288 holder->disk = disk;
1291 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1295 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1299 * bdev could be deleted beneath us which would implicitly destroy
1300 * the holder directory. Hold on to it.
1302 kobject_get(bdev->bd_part->holder_dir);
1304 list_add(&holder->list, &bdev->bd_holder_disks);
1308 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1312 mutex_unlock(&bdev->bd_mutex);
1315 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1318 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1319 * @bdev: the calimed slave bdev
1320 * @disk: the holding disk
1322 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1327 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1329 struct bd_holder_disk *holder;
1331 mutex_lock(&bdev->bd_mutex);
1333 holder = bd_find_holder_disk(bdev, disk);
1335 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1336 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1337 del_symlink(bdev->bd_part->holder_dir,
1338 &disk_to_dev(disk)->kobj);
1339 kobject_put(bdev->bd_part->holder_dir);
1340 list_del_init(&holder->list);
1344 mutex_unlock(&bdev->bd_mutex);
1346 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1350 * flush_disk - invalidates all buffer-cache entries on a disk
1352 * @bdev: struct block device to be flushed
1353 * @kill_dirty: flag to guide handling of dirty inodes
1355 * Invalidates all buffer-cache entries on a disk. It should be called
1356 * when a disk has been changed -- either by a media change or online
1359 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1361 if (__invalidate_device(bdev, kill_dirty)) {
1362 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1363 "resized disk %s\n",
1364 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1369 if (disk_part_scan_enabled(bdev->bd_disk))
1370 bdev->bd_invalidated = 1;
1374 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1375 * @disk: struct gendisk to check
1376 * @bdev: struct bdev to adjust.
1377 * @verbose: if %true log a message about a size change if there is any
1379 * This routine checks to see if the bdev size does not match the disk size
1380 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1383 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1386 loff_t disk_size, bdev_size;
1388 disk_size = (loff_t)get_capacity(disk) << 9;
1389 bdev_size = i_size_read(bdev->bd_inode);
1390 if (disk_size != bdev_size) {
1393 "%s: detected capacity change from %lld to %lld\n",
1394 disk->disk_name, bdev_size, disk_size);
1396 i_size_write(bdev->bd_inode, disk_size);
1397 if (bdev_size > disk_size)
1398 flush_disk(bdev, false);
1403 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1404 * @disk: struct gendisk to be revalidated
1406 * This routine is a wrapper for lower-level driver's revalidate_disk
1407 * call-backs. It is used to do common pre and post operations needed
1408 * for all revalidate_disk operations.
1410 int revalidate_disk(struct gendisk *disk)
1412 struct block_device *bdev;
1415 if (disk->fops->revalidate_disk)
1416 ret = disk->fops->revalidate_disk(disk);
1417 bdev = bdget_disk(disk, 0);
1421 mutex_lock(&bdev->bd_mutex);
1422 check_disk_size_change(disk, bdev, ret == 0);
1423 bdev->bd_invalidated = 0;
1424 mutex_unlock(&bdev->bd_mutex);
1428 EXPORT_SYMBOL(revalidate_disk);
1431 * This routine checks whether a removable media has been changed,
1432 * and invalidates all buffer-cache-entries in that case. This
1433 * is a relatively slow routine, so we have to try to minimize using
1434 * it. Thus it is called only upon a 'mount' or 'open'. This
1435 * is the best way of combining speed and utility, I think.
1436 * People changing diskettes in the middle of an operation deserve
1439 int check_disk_change(struct block_device *bdev)
1441 struct gendisk *disk = bdev->bd_disk;
1442 const struct block_device_operations *bdops = disk->fops;
1443 unsigned int events;
1445 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1446 DISK_EVENT_EJECT_REQUEST);
1447 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1450 flush_disk(bdev, true);
1451 if (bdops->revalidate_disk)
1452 bdops->revalidate_disk(bdev->bd_disk);
1456 EXPORT_SYMBOL(check_disk_change);
1458 void bd_set_size(struct block_device *bdev, loff_t size)
1460 inode_lock(bdev->bd_inode);
1461 i_size_write(bdev->bd_inode, size);
1462 inode_unlock(bdev->bd_inode);
1464 EXPORT_SYMBOL(bd_set_size);
1466 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1471 * mutex_lock(part->bd_mutex)
1472 * mutex_lock_nested(whole->bd_mutex, 1)
1475 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1477 struct gendisk *disk;
1481 bool first_open = false;
1483 if (mode & FMODE_READ)
1485 if (mode & FMODE_WRITE)
1488 * hooks: /n/, see "layering violations".
1491 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1501 disk = bdev_get_gendisk(bdev, &partno);
1505 disk_block_events(disk);
1506 mutex_lock_nested(&bdev->bd_mutex, for_part);
1507 if (!bdev->bd_openers) {
1509 bdev->bd_disk = disk;
1510 bdev->bd_queue = disk->queue;
1511 bdev->bd_contains = bdev;
1512 bdev->bd_partno = partno;
1516 bdev->bd_part = disk_get_part(disk, partno);
1521 if (disk->fops->open) {
1522 ret = disk->fops->open(bdev, mode);
1523 if (ret == -ERESTARTSYS) {
1524 /* Lost a race with 'disk' being
1525 * deleted, try again.
1528 disk_put_part(bdev->bd_part);
1529 bdev->bd_part = NULL;
1530 bdev->bd_disk = NULL;
1531 bdev->bd_queue = NULL;
1532 mutex_unlock(&bdev->bd_mutex);
1533 disk_unblock_events(disk);
1534 put_disk_and_module(disk);
1540 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1541 set_init_blocksize(bdev);
1545 * If the device is invalidated, rescan partition
1546 * if open succeeded or failed with -ENOMEDIUM.
1547 * The latter is necessary to prevent ghost
1548 * partitions on a removed medium.
1550 if (bdev->bd_invalidated) {
1552 rescan_partitions(disk, bdev);
1553 else if (ret == -ENOMEDIUM)
1554 invalidate_partitions(disk, bdev);
1560 struct block_device *whole;
1561 whole = bdget_disk(disk, 0);
1566 ret = __blkdev_get(whole, mode, 1);
1569 bdev->bd_contains = whole;
1570 bdev->bd_part = disk_get_part(disk, partno);
1571 if (!(disk->flags & GENHD_FL_UP) ||
1572 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1576 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1577 set_init_blocksize(bdev);
1580 if (bdev->bd_bdi == &noop_backing_dev_info)
1581 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1583 if (bdev->bd_contains == bdev) {
1585 if (bdev->bd_disk->fops->open)
1586 ret = bdev->bd_disk->fops->open(bdev, mode);
1587 /* the same as first opener case, read comment there */
1588 if (bdev->bd_invalidated) {
1590 rescan_partitions(bdev->bd_disk, bdev);
1591 else if (ret == -ENOMEDIUM)
1592 invalidate_partitions(bdev->bd_disk, bdev);
1595 goto out_unlock_bdev;
1600 bdev->bd_part_count++;
1601 mutex_unlock(&bdev->bd_mutex);
1602 disk_unblock_events(disk);
1603 /* only one opener holds refs to the module and disk */
1605 put_disk_and_module(disk);
1609 disk_put_part(bdev->bd_part);
1610 bdev->bd_disk = NULL;
1611 bdev->bd_part = NULL;
1612 bdev->bd_queue = NULL;
1613 if (bdev != bdev->bd_contains)
1614 __blkdev_put(bdev->bd_contains, mode, 1);
1615 bdev->bd_contains = NULL;
1617 mutex_unlock(&bdev->bd_mutex);
1618 disk_unblock_events(disk);
1619 put_disk_and_module(disk);
1627 * blkdev_get - open a block device
1628 * @bdev: block_device to open
1629 * @mode: FMODE_* mask
1630 * @holder: exclusive holder identifier
1632 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1633 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1634 * @holder is invalid. Exclusive opens may nest for the same @holder.
1636 * On success, the reference count of @bdev is unchanged. On failure,
1643 * 0 on success, -errno on failure.
1645 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1647 struct block_device *whole = NULL;
1650 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1652 if ((mode & FMODE_EXCL) && holder) {
1653 whole = bd_start_claiming(bdev, holder);
1654 if (IS_ERR(whole)) {
1656 return PTR_ERR(whole);
1660 res = __blkdev_get(bdev, mode, 0);
1663 struct gendisk *disk = whole->bd_disk;
1665 /* finish claiming */
1666 mutex_lock(&bdev->bd_mutex);
1667 spin_lock(&bdev_lock);
1670 BUG_ON(!bd_may_claim(bdev, whole, holder));
1672 * Note that for a whole device bd_holders
1673 * will be incremented twice, and bd_holder
1674 * will be set to bd_may_claim before being
1677 whole->bd_holders++;
1678 whole->bd_holder = bd_may_claim;
1680 bdev->bd_holder = holder;
1683 /* tell others that we're done */
1684 BUG_ON(whole->bd_claiming != holder);
1685 whole->bd_claiming = NULL;
1686 wake_up_bit(&whole->bd_claiming, 0);
1688 spin_unlock(&bdev_lock);
1691 * Block event polling for write claims if requested. Any
1692 * write holder makes the write_holder state stick until
1693 * all are released. This is good enough and tracking
1694 * individual writeable reference is too fragile given the
1695 * way @mode is used in blkdev_get/put().
1697 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1698 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1699 bdev->bd_write_holder = true;
1700 disk_block_events(disk);
1703 mutex_unlock(&bdev->bd_mutex);
1709 EXPORT_SYMBOL(blkdev_get);
1712 * blkdev_get_by_path - open a block device by name
1713 * @path: path to the block device to open
1714 * @mode: FMODE_* mask
1715 * @holder: exclusive holder identifier
1717 * Open the blockdevice described by the device file at @path. @mode
1718 * and @holder are identical to blkdev_get().
1720 * On success, the returned block_device has reference count of one.
1726 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1728 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1731 struct block_device *bdev;
1734 bdev = lookup_bdev(path);
1738 err = blkdev_get(bdev, mode, holder);
1740 return ERR_PTR(err);
1742 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1743 blkdev_put(bdev, mode);
1744 return ERR_PTR(-EACCES);
1749 EXPORT_SYMBOL(blkdev_get_by_path);
1752 * blkdev_get_by_dev - open a block device by device number
1753 * @dev: device number of block device to open
1754 * @mode: FMODE_* mask
1755 * @holder: exclusive holder identifier
1757 * Open the blockdevice described by device number @dev. @mode and
1758 * @holder are identical to blkdev_get().
1760 * Use it ONLY if you really do not have anything better - i.e. when
1761 * you are behind a truly sucky interface and all you are given is a
1762 * device number. _Never_ to be used for internal purposes. If you
1763 * ever need it - reconsider your API.
1765 * On success, the returned block_device has reference count of one.
1771 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1773 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1775 struct block_device *bdev;
1780 return ERR_PTR(-ENOMEM);
1782 err = blkdev_get(bdev, mode, holder);
1784 return ERR_PTR(err);
1788 EXPORT_SYMBOL(blkdev_get_by_dev);
1790 static int blkdev_open(struct inode * inode, struct file * filp)
1792 struct block_device *bdev;
1795 * Preserve backwards compatibility and allow large file access
1796 * even if userspace doesn't ask for it explicitly. Some mkfs
1797 * binary needs it. We might want to drop this workaround
1798 * during an unstable branch.
1800 filp->f_flags |= O_LARGEFILE;
1802 filp->f_mode |= FMODE_NOWAIT;
1804 if (filp->f_flags & O_NDELAY)
1805 filp->f_mode |= FMODE_NDELAY;
1806 if (filp->f_flags & O_EXCL)
1807 filp->f_mode |= FMODE_EXCL;
1808 if ((filp->f_flags & O_ACCMODE) == 3)
1809 filp->f_mode |= FMODE_WRITE_IOCTL;
1811 bdev = bd_acquire(inode);
1815 filp->f_mapping = bdev->bd_inode->i_mapping;
1816 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1818 return blkdev_get(bdev, filp->f_mode, filp);
1821 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1823 struct gendisk *disk = bdev->bd_disk;
1824 struct block_device *victim = NULL;
1826 mutex_lock_nested(&bdev->bd_mutex, for_part);
1828 bdev->bd_part_count--;
1830 if (!--bdev->bd_openers) {
1831 WARN_ON_ONCE(bdev->bd_holders);
1832 sync_blockdev(bdev);
1835 bdev_write_inode(bdev);
1837 if (bdev->bd_contains == bdev) {
1838 if (disk->fops->release)
1839 disk->fops->release(disk, mode);
1841 if (!bdev->bd_openers) {
1842 disk_put_part(bdev->bd_part);
1843 bdev->bd_part = NULL;
1844 bdev->bd_disk = NULL;
1845 if (bdev != bdev->bd_contains)
1846 victim = bdev->bd_contains;
1847 bdev->bd_contains = NULL;
1849 put_disk_and_module(disk);
1851 mutex_unlock(&bdev->bd_mutex);
1854 __blkdev_put(victim, mode, 1);
1857 void blkdev_put(struct block_device *bdev, fmode_t mode)
1859 mutex_lock(&bdev->bd_mutex);
1861 if (mode & FMODE_EXCL) {
1865 * Release a claim on the device. The holder fields
1866 * are protected with bdev_lock. bd_mutex is to
1867 * synchronize disk_holder unlinking.
1869 spin_lock(&bdev_lock);
1871 WARN_ON_ONCE(--bdev->bd_holders < 0);
1872 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1874 /* bd_contains might point to self, check in a separate step */
1875 if ((bdev_free = !bdev->bd_holders))
1876 bdev->bd_holder = NULL;
1877 if (!bdev->bd_contains->bd_holders)
1878 bdev->bd_contains->bd_holder = NULL;
1880 spin_unlock(&bdev_lock);
1883 * If this was the last claim, remove holder link and
1884 * unblock evpoll if it was a write holder.
1886 if (bdev_free && bdev->bd_write_holder) {
1887 disk_unblock_events(bdev->bd_disk);
1888 bdev->bd_write_holder = false;
1893 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1894 * event. This is to ensure detection of media removal commanded
1895 * from userland - e.g. eject(1).
1897 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1899 mutex_unlock(&bdev->bd_mutex);
1901 __blkdev_put(bdev, mode, 0);
1903 EXPORT_SYMBOL(blkdev_put);
1905 static int blkdev_close(struct inode * inode, struct file * filp)
1907 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1908 blkdev_put(bdev, filp->f_mode);
1912 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1914 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1915 fmode_t mode = file->f_mode;
1918 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1919 * to updated it before every ioctl.
1921 if (file->f_flags & O_NDELAY)
1922 mode |= FMODE_NDELAY;
1924 mode &= ~FMODE_NDELAY;
1926 return blkdev_ioctl(bdev, mode, cmd, arg);
1930 * Write data to the block device. Only intended for the block device itself
1931 * and the raw driver which basically is a fake block device.
1933 * Does not take i_mutex for the write and thus is not for general purpose
1936 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1938 struct file *file = iocb->ki_filp;
1939 struct inode *bd_inode = bdev_file_inode(file);
1940 loff_t size = i_size_read(bd_inode);
1941 struct blk_plug plug;
1944 if (bdev_read_only(I_BDEV(bd_inode)))
1947 if (!iov_iter_count(from))
1950 if (iocb->ki_pos >= size)
1953 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1956 iov_iter_truncate(from, size - iocb->ki_pos);
1958 blk_start_plug(&plug);
1959 ret = __generic_file_write_iter(iocb, from);
1961 ret = generic_write_sync(iocb, ret);
1962 blk_finish_plug(&plug);
1965 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1967 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1969 struct file *file = iocb->ki_filp;
1970 struct inode *bd_inode = bdev_file_inode(file);
1971 loff_t size = i_size_read(bd_inode);
1972 loff_t pos = iocb->ki_pos;
1978 iov_iter_truncate(to, size);
1979 return generic_file_read_iter(iocb, to);
1981 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1984 * Try to release a page associated with block device when the system
1985 * is under memory pressure.
1987 static int blkdev_releasepage(struct page *page, gfp_t wait)
1989 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1991 if (super && super->s_op->bdev_try_to_free_page)
1992 return super->s_op->bdev_try_to_free_page(super, page, wait);
1994 return try_to_free_buffers(page);
1997 static int blkdev_writepages(struct address_space *mapping,
1998 struct writeback_control *wbc)
2000 return generic_writepages(mapping, wbc);
2003 static const struct address_space_operations def_blk_aops = {
2004 .readpage = blkdev_readpage,
2005 .readpages = blkdev_readpages,
2006 .writepage = blkdev_writepage,
2007 .write_begin = blkdev_write_begin,
2008 .write_end = blkdev_write_end,
2009 .writepages = blkdev_writepages,
2010 .releasepage = blkdev_releasepage,
2011 .direct_IO = blkdev_direct_IO,
2012 .migratepage = buffer_migrate_page_norefs,
2013 .is_dirty_writeback = buffer_check_dirty_writeback,
2016 #define BLKDEV_FALLOC_FL_SUPPORTED \
2017 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2018 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2020 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2023 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2024 struct address_space *mapping;
2025 loff_t end = start + len - 1;
2029 /* Fail if we don't recognize the flags. */
2030 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2033 /* Don't go off the end of the device. */
2034 isize = i_size_read(bdev->bd_inode);
2038 if (mode & FALLOC_FL_KEEP_SIZE) {
2039 len = isize - start;
2040 end = start + len - 1;
2046 * Don't allow IO that isn't aligned to logical block size.
2048 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2051 /* Invalidate the page cache, including dirty pages. */
2052 mapping = bdev->bd_inode->i_mapping;
2053 truncate_inode_pages_range(mapping, start, end);
2056 case FALLOC_FL_ZERO_RANGE:
2057 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2058 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2059 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2061 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2062 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2063 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2065 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2066 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2076 * Invalidate again; if someone wandered in and dirtied a page,
2077 * the caller will be given -EBUSY. The third argument is
2078 * inclusive, so the rounding here is safe.
2080 return invalidate_inode_pages2_range(mapping,
2081 start >> PAGE_SHIFT,
2085 const struct file_operations def_blk_fops = {
2086 .open = blkdev_open,
2087 .release = blkdev_close,
2088 .llseek = block_llseek,
2089 .read_iter = blkdev_read_iter,
2090 .write_iter = blkdev_write_iter,
2091 .iopoll = blkdev_iopoll,
2092 .mmap = generic_file_mmap,
2093 .fsync = blkdev_fsync,
2094 .unlocked_ioctl = block_ioctl,
2095 #ifdef CONFIG_COMPAT
2096 .compat_ioctl = compat_blkdev_ioctl,
2098 .splice_read = generic_file_splice_read,
2099 .splice_write = iter_file_splice_write,
2100 .fallocate = blkdev_fallocate,
2103 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2106 mm_segment_t old_fs = get_fs();
2108 res = blkdev_ioctl(bdev, 0, cmd, arg);
2113 EXPORT_SYMBOL(ioctl_by_bdev);
2116 * lookup_bdev - lookup a struct block_device by name
2117 * @pathname: special file representing the block device
2119 * Get a reference to the blockdevice at @pathname in the current
2120 * namespace if possible and return it. Return ERR_PTR(error)
2123 struct block_device *lookup_bdev(const char *pathname)
2125 struct block_device *bdev;
2126 struct inode *inode;
2130 if (!pathname || !*pathname)
2131 return ERR_PTR(-EINVAL);
2133 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2135 return ERR_PTR(error);
2137 inode = d_backing_inode(path.dentry);
2139 if (!S_ISBLK(inode->i_mode))
2142 if (!may_open_dev(&path))
2145 bdev = bd_acquire(inode);
2152 bdev = ERR_PTR(error);
2155 EXPORT_SYMBOL(lookup_bdev);
2157 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2159 struct super_block *sb = get_super(bdev);
2164 * no need to lock the super, get_super holds the
2165 * read mutex so the filesystem cannot go away
2166 * under us (->put_super runs with the write lock
2169 shrink_dcache_sb(sb);
2170 res = invalidate_inodes(sb, kill_dirty);
2173 invalidate_bdev(bdev);
2176 EXPORT_SYMBOL(__invalidate_device);
2178 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2180 struct inode *inode, *old_inode = NULL;
2182 spin_lock(&blockdev_superblock->s_inode_list_lock);
2183 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2184 struct address_space *mapping = inode->i_mapping;
2185 struct block_device *bdev;
2187 spin_lock(&inode->i_lock);
2188 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2189 mapping->nrpages == 0) {
2190 spin_unlock(&inode->i_lock);
2194 spin_unlock(&inode->i_lock);
2195 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2197 * We hold a reference to 'inode' so it couldn't have been
2198 * removed from s_inodes list while we dropped the
2199 * s_inode_list_lock We cannot iput the inode now as we can
2200 * be holding the last reference and we cannot iput it under
2201 * s_inode_list_lock. So we keep the reference and iput it
2206 bdev = I_BDEV(inode);
2208 mutex_lock(&bdev->bd_mutex);
2209 if (bdev->bd_openers)
2211 mutex_unlock(&bdev->bd_mutex);
2213 spin_lock(&blockdev_superblock->s_inode_list_lock);
2215 spin_unlock(&blockdev_superblock->s_inode_list_lock);