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 int set_blocksize(struct block_device *bdev, int size)
109 /* Size must be a power of two, and between 512 and PAGE_SIZE */
110 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
113 /* Size cannot be smaller than the size supported by the device */
114 if (size < bdev_logical_block_size(bdev))
117 /* Don't change the size if it is same as current */
118 if (bdev->bd_block_size != size) {
120 bdev->bd_block_size = size;
121 bdev->bd_inode->i_blkbits = blksize_bits(size);
127 EXPORT_SYMBOL(set_blocksize);
129 int sb_set_blocksize(struct super_block *sb, int size)
131 if (set_blocksize(sb->s_bdev, size))
133 /* If we get here, we know size is power of two
134 * and it's value is between 512 and PAGE_SIZE */
135 sb->s_blocksize = size;
136 sb->s_blocksize_bits = blksize_bits(size);
137 return sb->s_blocksize;
140 EXPORT_SYMBOL(sb_set_blocksize);
142 int sb_min_blocksize(struct super_block *sb, int size)
144 int minsize = bdev_logical_block_size(sb->s_bdev);
147 return sb_set_blocksize(sb, size);
150 EXPORT_SYMBOL(sb_min_blocksize);
153 blkdev_get_block(struct inode *inode, sector_t iblock,
154 struct buffer_head *bh, int create)
156 bh->b_bdev = I_BDEV(inode);
157 bh->b_blocknr = iblock;
158 set_buffer_mapped(bh);
162 static struct inode *bdev_file_inode(struct file *file)
164 return file->f_mapping->host;
167 static unsigned int dio_bio_write_op(struct kiocb *iocb)
169 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
171 /* avoid the need for a I/O completion work item */
172 if (iocb->ki_flags & IOCB_DSYNC)
177 #define DIO_INLINE_BIO_VECS 4
179 static void blkdev_bio_end_io_simple(struct bio *bio)
181 struct task_struct *waiter = bio->bi_private;
183 WRITE_ONCE(bio->bi_private, NULL);
184 wake_up_process(waiter);
188 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
191 struct file *file = iocb->ki_filp;
192 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
193 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
194 loff_t pos = iocb->ki_pos;
195 bool should_dirty = false;
201 if ((pos | iov_iter_alignment(iter)) &
202 (bdev_logical_block_size(bdev) - 1))
205 if (nr_pages <= DIO_INLINE_BIO_VECS)
208 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
214 bio_init(&bio, vecs, nr_pages);
215 bio_set_dev(&bio, bdev);
216 bio.bi_iter.bi_sector = pos >> 9;
217 bio.bi_write_hint = iocb->ki_hint;
218 bio.bi_private = current;
219 bio.bi_end_io = blkdev_bio_end_io_simple;
220 bio.bi_ioprio = iocb->ki_ioprio;
222 ret = bio_iov_iter_get_pages(&bio, iter);
225 ret = bio.bi_iter.bi_size;
227 if (iov_iter_rw(iter) == READ) {
228 bio.bi_opf = REQ_OP_READ;
229 if (iter_is_iovec(iter))
232 bio.bi_opf = dio_bio_write_op(iocb);
233 task_io_account_write(ret);
235 if (iocb->ki_flags & IOCB_HIPRI)
236 bio.bi_opf |= REQ_HIPRI;
238 qc = submit_bio(&bio);
240 set_current_state(TASK_UNINTERRUPTIBLE);
241 if (!READ_ONCE(bio.bi_private))
243 if (!(iocb->ki_flags & IOCB_HIPRI) ||
244 !blk_poll(bdev_get_queue(bdev), qc))
247 __set_current_state(TASK_RUNNING);
249 bio_for_each_segment_all(bvec, &bio, i) {
250 if (should_dirty && !PageCompound(bvec->bv_page))
251 set_page_dirty_lock(bvec->bv_page);
252 put_page(bvec->bv_page);
255 if (unlikely(bio.bi_status))
256 ret = blk_status_to_errno(bio.bi_status);
259 if (vecs != inline_vecs)
270 struct task_struct *waiter;
275 bool should_dirty : 1;
280 static struct bio_set blkdev_dio_pool;
282 static void blkdev_bio_end_io(struct bio *bio)
284 struct blkdev_dio *dio = bio->bi_private;
285 bool should_dirty = dio->should_dirty;
287 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
288 if (bio->bi_status && !dio->bio.bi_status)
289 dio->bio.bi_status = bio->bi_status;
292 struct kiocb *iocb = dio->iocb;
295 if (likely(!dio->bio.bi_status)) {
299 ret = blk_status_to_errno(dio->bio.bi_status);
302 dio->iocb->ki_complete(iocb, ret, 0);
305 struct task_struct *waiter = dio->waiter;
307 WRITE_ONCE(dio->waiter, NULL);
308 wake_up_process(waiter);
313 bio_check_pages_dirty(bio);
315 struct bio_vec *bvec;
318 bio_for_each_segment_all(bvec, bio, i)
319 put_page(bvec->bv_page);
325 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
327 struct file *file = iocb->ki_filp;
328 struct inode *inode = bdev_file_inode(file);
329 struct block_device *bdev = I_BDEV(inode);
330 struct blk_plug plug;
331 struct blkdev_dio *dio;
333 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
334 loff_t pos = iocb->ki_pos;
335 blk_qc_t qc = BLK_QC_T_NONE;
338 if ((pos | iov_iter_alignment(iter)) &
339 (bdev_logical_block_size(bdev) - 1))
342 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
343 bio_get(bio); /* extra ref for the completion handler */
345 dio = container_of(bio, struct blkdev_dio, bio);
346 dio->is_sync = is_sync = is_sync_kiocb(iocb);
348 dio->waiter = current;
353 dio->multi_bio = false;
354 dio->should_dirty = is_read && iter_is_iovec(iter);
356 blk_start_plug(&plug);
358 bio_set_dev(bio, bdev);
359 bio->bi_iter.bi_sector = pos >> 9;
360 bio->bi_write_hint = iocb->ki_hint;
361 bio->bi_private = dio;
362 bio->bi_end_io = blkdev_bio_end_io;
363 bio->bi_ioprio = iocb->ki_ioprio;
365 ret = bio_iov_iter_get_pages(bio, iter);
367 bio->bi_status = BLK_STS_IOERR;
373 bio->bi_opf = REQ_OP_READ;
374 if (dio->should_dirty)
375 bio_set_pages_dirty(bio);
377 bio->bi_opf = dio_bio_write_op(iocb);
378 task_io_account_write(bio->bi_iter.bi_size);
381 dio->size += bio->bi_iter.bi_size;
382 pos += bio->bi_iter.bi_size;
384 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
386 qc = submit_bio(bio);
390 if (!dio->multi_bio) {
391 dio->multi_bio = true;
392 atomic_set(&dio->ref, 2);
394 atomic_inc(&dio->ref);
398 bio = bio_alloc(GFP_KERNEL, nr_pages);
400 blk_finish_plug(&plug);
406 set_current_state(TASK_UNINTERRUPTIBLE);
407 if (!READ_ONCE(dio->waiter))
410 if (!(iocb->ki_flags & IOCB_HIPRI) ||
411 !blk_poll(bdev_get_queue(bdev), qc))
414 __set_current_state(TASK_RUNNING);
417 ret = blk_status_to_errno(dio->bio.bi_status);
426 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
430 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
433 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
434 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
436 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
439 static __init int blkdev_init(void)
441 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
443 module_init(blkdev_init);
445 int __sync_blockdev(struct block_device *bdev, int wait)
450 return filemap_flush(bdev->bd_inode->i_mapping);
451 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
455 * Write out and wait upon all the dirty data associated with a block
456 * device via its mapping. Does not take the superblock lock.
458 int sync_blockdev(struct block_device *bdev)
460 return __sync_blockdev(bdev, 1);
462 EXPORT_SYMBOL(sync_blockdev);
465 * Write out and wait upon all dirty data associated with this
466 * device. Filesystem data as well as the underlying block
467 * device. Takes the superblock lock.
469 int fsync_bdev(struct block_device *bdev)
471 struct super_block *sb = get_super(bdev);
473 int res = sync_filesystem(sb);
477 return sync_blockdev(bdev);
479 EXPORT_SYMBOL(fsync_bdev);
482 * freeze_bdev -- lock a filesystem and force it into a consistent state
483 * @bdev: blockdevice to lock
485 * If a superblock is found on this device, we take the s_umount semaphore
486 * on it to make sure nobody unmounts until the snapshot creation is done.
487 * The reference counter (bd_fsfreeze_count) guarantees that only the last
488 * unfreeze process can unfreeze the frozen filesystem actually when multiple
489 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
490 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
493 struct super_block *freeze_bdev(struct block_device *bdev)
495 struct super_block *sb;
498 mutex_lock(&bdev->bd_fsfreeze_mutex);
499 if (++bdev->bd_fsfreeze_count > 1) {
501 * We don't even need to grab a reference - the first call
502 * to freeze_bdev grab an active reference and only the last
503 * thaw_bdev drops it.
505 sb = get_super(bdev);
508 mutex_unlock(&bdev->bd_fsfreeze_mutex);
512 sb = get_active_super(bdev);
515 if (sb->s_op->freeze_super)
516 error = sb->s_op->freeze_super(sb);
518 error = freeze_super(sb);
520 deactivate_super(sb);
521 bdev->bd_fsfreeze_count--;
522 mutex_unlock(&bdev->bd_fsfreeze_mutex);
523 return ERR_PTR(error);
525 deactivate_super(sb);
528 mutex_unlock(&bdev->bd_fsfreeze_mutex);
529 return sb; /* thaw_bdev releases s->s_umount */
531 EXPORT_SYMBOL(freeze_bdev);
534 * thaw_bdev -- unlock filesystem
535 * @bdev: blockdevice to unlock
536 * @sb: associated superblock
538 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
540 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
544 mutex_lock(&bdev->bd_fsfreeze_mutex);
545 if (!bdev->bd_fsfreeze_count)
549 if (--bdev->bd_fsfreeze_count > 0)
555 if (sb->s_op->thaw_super)
556 error = sb->s_op->thaw_super(sb);
558 error = thaw_super(sb);
560 bdev->bd_fsfreeze_count++;
562 mutex_unlock(&bdev->bd_fsfreeze_mutex);
565 EXPORT_SYMBOL(thaw_bdev);
567 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
569 return block_write_full_page(page, blkdev_get_block, wbc);
572 static int blkdev_readpage(struct file * file, struct page * page)
574 return block_read_full_page(page, blkdev_get_block);
577 static int blkdev_readpages(struct file *file, struct address_space *mapping,
578 struct list_head *pages, unsigned nr_pages)
580 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
583 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
584 loff_t pos, unsigned len, unsigned flags,
585 struct page **pagep, void **fsdata)
587 return block_write_begin(mapping, pos, len, flags, pagep,
591 static int blkdev_write_end(struct file *file, struct address_space *mapping,
592 loff_t pos, unsigned len, unsigned copied,
593 struct page *page, void *fsdata)
596 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
606 * for a block special file file_inode(file)->i_size is zero
607 * so we compute the size by hand (just as in block_read/write above)
609 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
611 struct inode *bd_inode = bdev_file_inode(file);
614 inode_lock(bd_inode);
615 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
616 inode_unlock(bd_inode);
620 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
622 struct inode *bd_inode = bdev_file_inode(filp);
623 struct block_device *bdev = I_BDEV(bd_inode);
626 error = file_write_and_wait_range(filp, start, end);
631 * There is no need to serialise calls to blkdev_issue_flush with
632 * i_mutex and doing so causes performance issues with concurrent
633 * O_SYNC writers to a block device.
635 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
636 if (error == -EOPNOTSUPP)
641 EXPORT_SYMBOL(blkdev_fsync);
644 * bdev_read_page() - Start reading a page from a block device
645 * @bdev: The device to read the page from
646 * @sector: The offset on the device to read the page to (need not be aligned)
647 * @page: The page to read
649 * On entry, the page should be locked. It will be unlocked when the page
650 * has been read. If the block driver implements rw_page synchronously,
651 * that will be true on exit from this function, but it need not be.
653 * Errors returned by this function are usually "soft", eg out of memory, or
654 * queue full; callers should try a different route to read this page rather
655 * than propagate an error back up the stack.
657 * Return: negative errno if an error occurs, 0 if submission was successful.
659 int bdev_read_page(struct block_device *bdev, sector_t sector,
662 const struct block_device_operations *ops = bdev->bd_disk->fops;
663 int result = -EOPNOTSUPP;
665 if (!ops->rw_page || bdev_get_integrity(bdev))
668 result = blk_queue_enter(bdev->bd_queue, 0);
671 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
673 blk_queue_exit(bdev->bd_queue);
676 EXPORT_SYMBOL_GPL(bdev_read_page);
679 * bdev_write_page() - Start writing a page to a block device
680 * @bdev: The device to write the page to
681 * @sector: The offset on the device to write the page to (need not be aligned)
682 * @page: The page to write
683 * @wbc: The writeback_control for the write
685 * On entry, the page should be locked and not currently under writeback.
686 * On exit, if the write started successfully, the page will be unlocked and
687 * under writeback. If the write failed already (eg the driver failed to
688 * queue the page to the device), the page will still be locked. If the
689 * caller is a ->writepage implementation, it will need to unlock the page.
691 * Errors returned by this function are usually "soft", eg out of memory, or
692 * queue full; callers should try a different route to write this page rather
693 * than propagate an error back up the stack.
695 * Return: negative errno if an error occurs, 0 if submission was successful.
697 int bdev_write_page(struct block_device *bdev, sector_t sector,
698 struct page *page, struct writeback_control *wbc)
701 const struct block_device_operations *ops = bdev->bd_disk->fops;
703 if (!ops->rw_page || bdev_get_integrity(bdev))
705 result = blk_queue_enter(bdev->bd_queue, 0);
709 set_page_writeback(page);
710 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
713 end_page_writeback(page);
715 clean_page_buffers(page);
718 blk_queue_exit(bdev->bd_queue);
721 EXPORT_SYMBOL_GPL(bdev_write_page);
727 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
728 static struct kmem_cache * bdev_cachep __read_mostly;
730 static struct inode *bdev_alloc_inode(struct super_block *sb)
732 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
735 return &ei->vfs_inode;
738 static void bdev_i_callback(struct rcu_head *head)
740 struct inode *inode = container_of(head, struct inode, i_rcu);
741 struct bdev_inode *bdi = BDEV_I(inode);
743 kmem_cache_free(bdev_cachep, bdi);
746 static void bdev_destroy_inode(struct inode *inode)
748 call_rcu(&inode->i_rcu, bdev_i_callback);
751 static void init_once(void *foo)
753 struct bdev_inode *ei = (struct bdev_inode *) foo;
754 struct block_device *bdev = &ei->bdev;
756 memset(bdev, 0, sizeof(*bdev));
757 mutex_init(&bdev->bd_mutex);
758 INIT_LIST_HEAD(&bdev->bd_list);
760 INIT_LIST_HEAD(&bdev->bd_holder_disks);
762 bdev->bd_bdi = &noop_backing_dev_info;
763 inode_init_once(&ei->vfs_inode);
764 /* Initialize mutex for freeze. */
765 mutex_init(&bdev->bd_fsfreeze_mutex);
768 static void bdev_evict_inode(struct inode *inode)
770 struct block_device *bdev = &BDEV_I(inode)->bdev;
771 truncate_inode_pages_final(&inode->i_data);
772 invalidate_inode_buffers(inode); /* is it needed here? */
774 spin_lock(&bdev_lock);
775 list_del_init(&bdev->bd_list);
776 spin_unlock(&bdev_lock);
777 /* Detach inode from wb early as bdi_put() may free bdi->wb */
778 inode_detach_wb(inode);
779 if (bdev->bd_bdi != &noop_backing_dev_info) {
780 bdi_put(bdev->bd_bdi);
781 bdev->bd_bdi = &noop_backing_dev_info;
785 static const struct super_operations bdev_sops = {
786 .statfs = simple_statfs,
787 .alloc_inode = bdev_alloc_inode,
788 .destroy_inode = bdev_destroy_inode,
789 .drop_inode = generic_delete_inode,
790 .evict_inode = bdev_evict_inode,
793 static struct dentry *bd_mount(struct file_system_type *fs_type,
794 int flags, const char *dev_name, void *data)
797 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
799 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
803 static struct file_system_type bd_type = {
806 .kill_sb = kill_anon_super,
809 struct super_block *blockdev_superblock __read_mostly;
810 EXPORT_SYMBOL_GPL(blockdev_superblock);
812 void __init bdev_cache_init(void)
815 static struct vfsmount *bd_mnt;
817 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
818 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
819 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
821 err = register_filesystem(&bd_type);
823 panic("Cannot register bdev pseudo-fs");
824 bd_mnt = kern_mount(&bd_type);
826 panic("Cannot create bdev pseudo-fs");
827 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
831 * Most likely _very_ bad one - but then it's hardly critical for small
832 * /dev and can be fixed when somebody will need really large one.
833 * Keep in mind that it will be fed through icache hash function too.
835 static inline unsigned long hash(dev_t dev)
837 return MAJOR(dev)+MINOR(dev);
840 static int bdev_test(struct inode *inode, void *data)
842 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
845 static int bdev_set(struct inode *inode, void *data)
847 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
851 static LIST_HEAD(all_bdevs);
854 * If there is a bdev inode for this device, unhash it so that it gets evicted
855 * as soon as last inode reference is dropped.
857 void bdev_unhash_inode(dev_t dev)
861 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
863 remove_inode_hash(inode);
868 struct block_device *bdget(dev_t dev)
870 struct block_device *bdev;
873 inode = iget5_locked(blockdev_superblock, hash(dev),
874 bdev_test, bdev_set, &dev);
879 bdev = &BDEV_I(inode)->bdev;
881 if (inode->i_state & I_NEW) {
882 bdev->bd_contains = NULL;
883 bdev->bd_super = NULL;
884 bdev->bd_inode = inode;
885 bdev->bd_block_size = i_blocksize(inode);
886 bdev->bd_part_count = 0;
887 bdev->bd_invalidated = 0;
888 inode->i_mode = S_IFBLK;
890 inode->i_bdev = bdev;
891 inode->i_data.a_ops = &def_blk_aops;
892 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
893 spin_lock(&bdev_lock);
894 list_add(&bdev->bd_list, &all_bdevs);
895 spin_unlock(&bdev_lock);
896 unlock_new_inode(inode);
901 EXPORT_SYMBOL(bdget);
904 * bdgrab -- Grab a reference to an already referenced block device
905 * @bdev: Block device to grab a reference to.
907 struct block_device *bdgrab(struct block_device *bdev)
909 ihold(bdev->bd_inode);
912 EXPORT_SYMBOL(bdgrab);
914 long nr_blockdev_pages(void)
916 struct block_device *bdev;
918 spin_lock(&bdev_lock);
919 list_for_each_entry(bdev, &all_bdevs, bd_list) {
920 ret += bdev->bd_inode->i_mapping->nrpages;
922 spin_unlock(&bdev_lock);
926 void bdput(struct block_device *bdev)
928 iput(bdev->bd_inode);
931 EXPORT_SYMBOL(bdput);
933 static struct block_device *bd_acquire(struct inode *inode)
935 struct block_device *bdev;
937 spin_lock(&bdev_lock);
938 bdev = inode->i_bdev;
939 if (bdev && !inode_unhashed(bdev->bd_inode)) {
941 spin_unlock(&bdev_lock);
944 spin_unlock(&bdev_lock);
947 * i_bdev references block device inode that was already shut down
948 * (corresponding device got removed). Remove the reference and look
949 * up block device inode again just in case new device got
950 * reestablished under the same device number.
955 bdev = bdget(inode->i_rdev);
957 spin_lock(&bdev_lock);
958 if (!inode->i_bdev) {
960 * We take an additional reference to bd_inode,
961 * and it's released in clear_inode() of inode.
962 * So, we can access it via ->i_mapping always
966 inode->i_bdev = bdev;
967 inode->i_mapping = bdev->bd_inode->i_mapping;
969 spin_unlock(&bdev_lock);
974 /* Call when you free inode */
976 void bd_forget(struct inode *inode)
978 struct block_device *bdev = NULL;
980 spin_lock(&bdev_lock);
981 if (!sb_is_blkdev_sb(inode->i_sb))
982 bdev = inode->i_bdev;
983 inode->i_bdev = NULL;
984 inode->i_mapping = &inode->i_data;
985 spin_unlock(&bdev_lock);
992 * bd_may_claim - test whether a block device can be claimed
993 * @bdev: block device of interest
994 * @whole: whole block device containing @bdev, may equal @bdev
995 * @holder: holder trying to claim @bdev
997 * Test whether @bdev can be claimed by @holder.
1000 * spin_lock(&bdev_lock).
1003 * %true if @bdev can be claimed, %false otherwise.
1005 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1008 if (bdev->bd_holder == holder)
1009 return true; /* already a holder */
1010 else if (bdev->bd_holder != NULL)
1011 return false; /* held by someone else */
1012 else if (whole == bdev)
1013 return true; /* is a whole device which isn't held */
1015 else if (whole->bd_holder == bd_may_claim)
1016 return true; /* is a partition of a device that is being partitioned */
1017 else if (whole->bd_holder != NULL)
1018 return false; /* is a partition of a held device */
1020 return true; /* is a partition of an un-held device */
1024 * bd_prepare_to_claim - prepare to claim a block device
1025 * @bdev: block device of interest
1026 * @whole: the whole device containing @bdev, may equal @bdev
1027 * @holder: holder trying to claim @bdev
1029 * Prepare to claim @bdev. This function fails if @bdev is already
1030 * claimed by another holder and waits if another claiming is in
1031 * progress. This function doesn't actually claim. On successful
1032 * return, the caller has ownership of bd_claiming and bd_holder[s].
1035 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1036 * it multiple times.
1039 * 0 if @bdev can be claimed, -EBUSY otherwise.
1041 static int bd_prepare_to_claim(struct block_device *bdev,
1042 struct block_device *whole, void *holder)
1045 /* if someone else claimed, fail */
1046 if (!bd_may_claim(bdev, whole, holder))
1049 /* if claiming is already in progress, wait for it to finish */
1050 if (whole->bd_claiming) {
1051 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1054 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1055 spin_unlock(&bdev_lock);
1057 finish_wait(wq, &wait);
1058 spin_lock(&bdev_lock);
1066 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1068 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1073 * Now that we hold gendisk reference we make sure bdev we looked up is
1074 * not stale. If it is, it means device got removed and created before
1075 * we looked up gendisk and we fail open in such case. Associating
1076 * unhashed bdev with newly created gendisk could lead to two bdevs
1077 * (and thus two independent caches) being associated with one device
1080 if (inode_unhashed(bdev->bd_inode)) {
1081 put_disk_and_module(disk);
1088 * bd_start_claiming - start claiming a block device
1089 * @bdev: block device of interest
1090 * @holder: holder trying to claim @bdev
1092 * @bdev is about to be opened exclusively. Check @bdev can be opened
1093 * exclusively and mark that an exclusive open is in progress. Each
1094 * successful call to this function must be matched with a call to
1095 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1098 * This function is used to gain exclusive access to the block device
1099 * without actually causing other exclusive open attempts to fail. It
1100 * should be used when the open sequence itself requires exclusive
1101 * access but may subsequently fail.
1107 * Pointer to the block device containing @bdev on success, ERR_PTR()
1110 static struct block_device *bd_start_claiming(struct block_device *bdev,
1113 struct gendisk *disk;
1114 struct block_device *whole;
1120 * @bdev might not have been initialized properly yet, look up
1121 * and grab the outer block device the hard way.
1123 disk = bdev_get_gendisk(bdev, &partno);
1125 return ERR_PTR(-ENXIO);
1128 * Normally, @bdev should equal what's returned from bdget_disk()
1129 * if partno is 0; however, some drivers (floppy) use multiple
1130 * bdev's for the same physical device and @bdev may be one of the
1131 * aliases. Keep @bdev if partno is 0. This means claimer
1132 * tracking is broken for those devices but it has always been that
1136 whole = bdget_disk(disk, 0);
1138 whole = bdgrab(bdev);
1140 put_disk_and_module(disk);
1142 return ERR_PTR(-ENOMEM);
1144 /* prepare to claim, if successful, mark claiming in progress */
1145 spin_lock(&bdev_lock);
1147 err = bd_prepare_to_claim(bdev, whole, holder);
1149 whole->bd_claiming = holder;
1150 spin_unlock(&bdev_lock);
1153 spin_unlock(&bdev_lock);
1155 return ERR_PTR(err);
1160 struct bd_holder_disk {
1161 struct list_head list;
1162 struct gendisk *disk;
1166 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1167 struct gendisk *disk)
1169 struct bd_holder_disk *holder;
1171 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1172 if (holder->disk == disk)
1177 static int add_symlink(struct kobject *from, struct kobject *to)
1179 return sysfs_create_link(from, to, kobject_name(to));
1182 static void del_symlink(struct kobject *from, struct kobject *to)
1184 sysfs_remove_link(from, kobject_name(to));
1188 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1189 * @bdev: the claimed slave bdev
1190 * @disk: the holding disk
1192 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1194 * This functions creates the following sysfs symlinks.
1196 * - from "slaves" directory of the holder @disk to the claimed @bdev
1197 * - from "holders" directory of the @bdev to the holder @disk
1199 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1200 * passed to bd_link_disk_holder(), then:
1202 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1203 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1205 * The caller must have claimed @bdev before calling this function and
1206 * ensure that both @bdev and @disk are valid during the creation and
1207 * lifetime of these symlinks.
1213 * 0 on success, -errno on failure.
1215 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1217 struct bd_holder_disk *holder;
1220 mutex_lock(&bdev->bd_mutex);
1222 WARN_ON_ONCE(!bdev->bd_holder);
1224 /* FIXME: remove the following once add_disk() handles errors */
1225 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1228 holder = bd_find_holder_disk(bdev, disk);
1234 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1240 INIT_LIST_HEAD(&holder->list);
1241 holder->disk = disk;
1244 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1248 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1252 * bdev could be deleted beneath us which would implicitly destroy
1253 * the holder directory. Hold on to it.
1255 kobject_get(bdev->bd_part->holder_dir);
1257 list_add(&holder->list, &bdev->bd_holder_disks);
1261 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1265 mutex_unlock(&bdev->bd_mutex);
1268 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1271 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1272 * @bdev: the calimed slave bdev
1273 * @disk: the holding disk
1275 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1280 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1282 struct bd_holder_disk *holder;
1284 mutex_lock(&bdev->bd_mutex);
1286 holder = bd_find_holder_disk(bdev, disk);
1288 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1289 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1290 del_symlink(bdev->bd_part->holder_dir,
1291 &disk_to_dev(disk)->kobj);
1292 kobject_put(bdev->bd_part->holder_dir);
1293 list_del_init(&holder->list);
1297 mutex_unlock(&bdev->bd_mutex);
1299 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1303 * flush_disk - invalidates all buffer-cache entries on a disk
1305 * @bdev: struct block device to be flushed
1306 * @kill_dirty: flag to guide handling of dirty inodes
1308 * Invalidates all buffer-cache entries on a disk. It should be called
1309 * when a disk has been changed -- either by a media change or online
1312 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1314 if (__invalidate_device(bdev, kill_dirty)) {
1315 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1316 "resized disk %s\n",
1317 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1322 if (disk_part_scan_enabled(bdev->bd_disk))
1323 bdev->bd_invalidated = 1;
1327 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1328 * @disk: struct gendisk to check
1329 * @bdev: struct bdev to adjust.
1330 * @verbose: if %true log a message about a size change if there is any
1332 * This routine checks to see if the bdev size does not match the disk size
1333 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1336 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1339 loff_t disk_size, bdev_size;
1341 disk_size = (loff_t)get_capacity(disk) << 9;
1342 bdev_size = i_size_read(bdev->bd_inode);
1343 if (disk_size != bdev_size) {
1346 "%s: detected capacity change from %lld to %lld\n",
1347 disk->disk_name, bdev_size, disk_size);
1349 i_size_write(bdev->bd_inode, disk_size);
1350 if (bdev_size > disk_size)
1351 flush_disk(bdev, false);
1356 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1357 * @disk: struct gendisk to be revalidated
1359 * This routine is a wrapper for lower-level driver's revalidate_disk
1360 * call-backs. It is used to do common pre and post operations needed
1361 * for all revalidate_disk operations.
1363 int revalidate_disk(struct gendisk *disk)
1365 struct block_device *bdev;
1368 if (disk->fops->revalidate_disk)
1369 ret = disk->fops->revalidate_disk(disk);
1370 bdev = bdget_disk(disk, 0);
1374 mutex_lock(&bdev->bd_mutex);
1375 check_disk_size_change(disk, bdev, ret == 0);
1376 bdev->bd_invalidated = 0;
1377 mutex_unlock(&bdev->bd_mutex);
1381 EXPORT_SYMBOL(revalidate_disk);
1384 * This routine checks whether a removable media has been changed,
1385 * and invalidates all buffer-cache-entries in that case. This
1386 * is a relatively slow routine, so we have to try to minimize using
1387 * it. Thus it is called only upon a 'mount' or 'open'. This
1388 * is the best way of combining speed and utility, I think.
1389 * People changing diskettes in the middle of an operation deserve
1392 int check_disk_change(struct block_device *bdev)
1394 struct gendisk *disk = bdev->bd_disk;
1395 const struct block_device_operations *bdops = disk->fops;
1396 unsigned int events;
1398 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1399 DISK_EVENT_EJECT_REQUEST);
1400 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1403 flush_disk(bdev, true);
1404 if (bdops->revalidate_disk)
1405 bdops->revalidate_disk(bdev->bd_disk);
1409 EXPORT_SYMBOL(check_disk_change);
1411 void bd_set_size(struct block_device *bdev, loff_t size)
1413 unsigned bsize = bdev_logical_block_size(bdev);
1415 inode_lock(bdev->bd_inode);
1416 i_size_write(bdev->bd_inode, size);
1417 inode_unlock(bdev->bd_inode);
1418 while (bsize < PAGE_SIZE) {
1423 bdev->bd_block_size = bsize;
1424 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1426 EXPORT_SYMBOL(bd_set_size);
1428 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1433 * mutex_lock(part->bd_mutex)
1434 * mutex_lock_nested(whole->bd_mutex, 1)
1437 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1439 struct gendisk *disk;
1443 bool first_open = false;
1445 if (mode & FMODE_READ)
1447 if (mode & FMODE_WRITE)
1450 * hooks: /n/, see "layering violations".
1453 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1463 disk = bdev_get_gendisk(bdev, &partno);
1467 disk_block_events(disk);
1468 mutex_lock_nested(&bdev->bd_mutex, for_part);
1469 if (!bdev->bd_openers) {
1471 bdev->bd_disk = disk;
1472 bdev->bd_queue = disk->queue;
1473 bdev->bd_contains = bdev;
1474 bdev->bd_partno = partno;
1478 bdev->bd_part = disk_get_part(disk, partno);
1483 if (disk->fops->open) {
1484 ret = disk->fops->open(bdev, mode);
1485 if (ret == -ERESTARTSYS) {
1486 /* Lost a race with 'disk' being
1487 * deleted, try again.
1490 disk_put_part(bdev->bd_part);
1491 bdev->bd_part = NULL;
1492 bdev->bd_disk = NULL;
1493 bdev->bd_queue = NULL;
1494 mutex_unlock(&bdev->bd_mutex);
1495 disk_unblock_events(disk);
1496 put_disk_and_module(disk);
1502 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1505 * If the device is invalidated, rescan partition
1506 * if open succeeded or failed with -ENOMEDIUM.
1507 * The latter is necessary to prevent ghost
1508 * partitions on a removed medium.
1510 if (bdev->bd_invalidated) {
1512 rescan_partitions(disk, bdev);
1513 else if (ret == -ENOMEDIUM)
1514 invalidate_partitions(disk, bdev);
1520 struct block_device *whole;
1521 whole = bdget_disk(disk, 0);
1526 ret = __blkdev_get(whole, mode, 1);
1529 bdev->bd_contains = whole;
1530 bdev->bd_part = disk_get_part(disk, partno);
1531 if (!(disk->flags & GENHD_FL_UP) ||
1532 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1536 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1539 if (bdev->bd_bdi == &noop_backing_dev_info)
1540 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1542 if (bdev->bd_contains == bdev) {
1544 if (bdev->bd_disk->fops->open)
1545 ret = bdev->bd_disk->fops->open(bdev, mode);
1546 /* the same as first opener case, read comment there */
1547 if (bdev->bd_invalidated) {
1549 rescan_partitions(bdev->bd_disk, bdev);
1550 else if (ret == -ENOMEDIUM)
1551 invalidate_partitions(bdev->bd_disk, bdev);
1554 goto out_unlock_bdev;
1559 bdev->bd_part_count++;
1560 mutex_unlock(&bdev->bd_mutex);
1561 disk_unblock_events(disk);
1562 /* only one opener holds refs to the module and disk */
1564 put_disk_and_module(disk);
1568 disk_put_part(bdev->bd_part);
1569 bdev->bd_disk = NULL;
1570 bdev->bd_part = NULL;
1571 bdev->bd_queue = NULL;
1572 if (bdev != bdev->bd_contains)
1573 __blkdev_put(bdev->bd_contains, mode, 1);
1574 bdev->bd_contains = NULL;
1576 mutex_unlock(&bdev->bd_mutex);
1577 disk_unblock_events(disk);
1578 put_disk_and_module(disk);
1586 * blkdev_get - open a block device
1587 * @bdev: block_device to open
1588 * @mode: FMODE_* mask
1589 * @holder: exclusive holder identifier
1591 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1592 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1593 * @holder is invalid. Exclusive opens may nest for the same @holder.
1595 * On success, the reference count of @bdev is unchanged. On failure,
1602 * 0 on success, -errno on failure.
1604 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1606 struct block_device *whole = NULL;
1609 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1611 if ((mode & FMODE_EXCL) && holder) {
1612 whole = bd_start_claiming(bdev, holder);
1613 if (IS_ERR(whole)) {
1615 return PTR_ERR(whole);
1619 res = __blkdev_get(bdev, mode, 0);
1622 struct gendisk *disk = whole->bd_disk;
1624 /* finish claiming */
1625 mutex_lock(&bdev->bd_mutex);
1626 spin_lock(&bdev_lock);
1629 BUG_ON(!bd_may_claim(bdev, whole, holder));
1631 * Note that for a whole device bd_holders
1632 * will be incremented twice, and bd_holder
1633 * will be set to bd_may_claim before being
1636 whole->bd_holders++;
1637 whole->bd_holder = bd_may_claim;
1639 bdev->bd_holder = holder;
1642 /* tell others that we're done */
1643 BUG_ON(whole->bd_claiming != holder);
1644 whole->bd_claiming = NULL;
1645 wake_up_bit(&whole->bd_claiming, 0);
1647 spin_unlock(&bdev_lock);
1650 * Block event polling for write claims if requested. Any
1651 * write holder makes the write_holder state stick until
1652 * all are released. This is good enough and tracking
1653 * individual writeable reference is too fragile given the
1654 * way @mode is used in blkdev_get/put().
1656 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1657 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1658 bdev->bd_write_holder = true;
1659 disk_block_events(disk);
1662 mutex_unlock(&bdev->bd_mutex);
1668 EXPORT_SYMBOL(blkdev_get);
1671 * blkdev_get_by_path - open a block device by name
1672 * @path: path to the block device to open
1673 * @mode: FMODE_* mask
1674 * @holder: exclusive holder identifier
1676 * Open the blockdevice described by the device file at @path. @mode
1677 * and @holder are identical to blkdev_get().
1679 * On success, the returned block_device has reference count of one.
1685 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1687 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1690 struct block_device *bdev;
1693 bdev = lookup_bdev(path);
1697 err = blkdev_get(bdev, mode, holder);
1699 return ERR_PTR(err);
1701 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1702 blkdev_put(bdev, mode);
1703 return ERR_PTR(-EACCES);
1708 EXPORT_SYMBOL(blkdev_get_by_path);
1711 * blkdev_get_by_dev - open a block device by device number
1712 * @dev: device number of block device to open
1713 * @mode: FMODE_* mask
1714 * @holder: exclusive holder identifier
1716 * Open the blockdevice described by device number @dev. @mode and
1717 * @holder are identical to blkdev_get().
1719 * Use it ONLY if you really do not have anything better - i.e. when
1720 * you are behind a truly sucky interface and all you are given is a
1721 * device number. _Never_ to be used for internal purposes. If you
1722 * ever need it - reconsider your API.
1724 * On success, the returned block_device has reference count of one.
1730 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1732 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1734 struct block_device *bdev;
1739 return ERR_PTR(-ENOMEM);
1741 err = blkdev_get(bdev, mode, holder);
1743 return ERR_PTR(err);
1747 EXPORT_SYMBOL(blkdev_get_by_dev);
1749 static int blkdev_open(struct inode * inode, struct file * filp)
1751 struct block_device *bdev;
1754 * Preserve backwards compatibility and allow large file access
1755 * even if userspace doesn't ask for it explicitly. Some mkfs
1756 * binary needs it. We might want to drop this workaround
1757 * during an unstable branch.
1759 filp->f_flags |= O_LARGEFILE;
1761 filp->f_mode |= FMODE_NOWAIT;
1763 if (filp->f_flags & O_NDELAY)
1764 filp->f_mode |= FMODE_NDELAY;
1765 if (filp->f_flags & O_EXCL)
1766 filp->f_mode |= FMODE_EXCL;
1767 if ((filp->f_flags & O_ACCMODE) == 3)
1768 filp->f_mode |= FMODE_WRITE_IOCTL;
1770 bdev = bd_acquire(inode);
1774 filp->f_mapping = bdev->bd_inode->i_mapping;
1775 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1777 return blkdev_get(bdev, filp->f_mode, filp);
1780 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1782 struct gendisk *disk = bdev->bd_disk;
1783 struct block_device *victim = NULL;
1785 mutex_lock_nested(&bdev->bd_mutex, for_part);
1787 bdev->bd_part_count--;
1789 if (!--bdev->bd_openers) {
1790 WARN_ON_ONCE(bdev->bd_holders);
1791 sync_blockdev(bdev);
1794 bdev_write_inode(bdev);
1796 if (bdev->bd_contains == bdev) {
1797 if (disk->fops->release)
1798 disk->fops->release(disk, mode);
1800 if (!bdev->bd_openers) {
1801 disk_put_part(bdev->bd_part);
1802 bdev->bd_part = NULL;
1803 bdev->bd_disk = NULL;
1804 if (bdev != bdev->bd_contains)
1805 victim = bdev->bd_contains;
1806 bdev->bd_contains = NULL;
1808 put_disk_and_module(disk);
1810 mutex_unlock(&bdev->bd_mutex);
1813 __blkdev_put(victim, mode, 1);
1816 void blkdev_put(struct block_device *bdev, fmode_t mode)
1818 mutex_lock(&bdev->bd_mutex);
1820 if (mode & FMODE_EXCL) {
1824 * Release a claim on the device. The holder fields
1825 * are protected with bdev_lock. bd_mutex is to
1826 * synchronize disk_holder unlinking.
1828 spin_lock(&bdev_lock);
1830 WARN_ON_ONCE(--bdev->bd_holders < 0);
1831 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1833 /* bd_contains might point to self, check in a separate step */
1834 if ((bdev_free = !bdev->bd_holders))
1835 bdev->bd_holder = NULL;
1836 if (!bdev->bd_contains->bd_holders)
1837 bdev->bd_contains->bd_holder = NULL;
1839 spin_unlock(&bdev_lock);
1842 * If this was the last claim, remove holder link and
1843 * unblock evpoll if it was a write holder.
1845 if (bdev_free && bdev->bd_write_holder) {
1846 disk_unblock_events(bdev->bd_disk);
1847 bdev->bd_write_holder = false;
1852 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1853 * event. This is to ensure detection of media removal commanded
1854 * from userland - e.g. eject(1).
1856 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1858 mutex_unlock(&bdev->bd_mutex);
1860 __blkdev_put(bdev, mode, 0);
1862 EXPORT_SYMBOL(blkdev_put);
1864 static int blkdev_close(struct inode * inode, struct file * filp)
1866 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1867 blkdev_put(bdev, filp->f_mode);
1871 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1873 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1874 fmode_t mode = file->f_mode;
1877 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1878 * to updated it before every ioctl.
1880 if (file->f_flags & O_NDELAY)
1881 mode |= FMODE_NDELAY;
1883 mode &= ~FMODE_NDELAY;
1885 return blkdev_ioctl(bdev, mode, cmd, arg);
1889 * Write data to the block device. Only intended for the block device itself
1890 * and the raw driver which basically is a fake block device.
1892 * Does not take i_mutex for the write and thus is not for general purpose
1895 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1897 struct file *file = iocb->ki_filp;
1898 struct inode *bd_inode = bdev_file_inode(file);
1899 loff_t size = i_size_read(bd_inode);
1900 struct blk_plug plug;
1903 if (bdev_read_only(I_BDEV(bd_inode)))
1906 if (!iov_iter_count(from))
1909 if (iocb->ki_pos >= size)
1912 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1915 iov_iter_truncate(from, size - iocb->ki_pos);
1917 blk_start_plug(&plug);
1918 ret = __generic_file_write_iter(iocb, from);
1920 ret = generic_write_sync(iocb, ret);
1921 blk_finish_plug(&plug);
1924 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1926 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1928 struct file *file = iocb->ki_filp;
1929 struct inode *bd_inode = bdev_file_inode(file);
1930 loff_t size = i_size_read(bd_inode);
1931 loff_t pos = iocb->ki_pos;
1937 iov_iter_truncate(to, size);
1938 return generic_file_read_iter(iocb, to);
1940 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1943 * Try to release a page associated with block device when the system
1944 * is under memory pressure.
1946 static int blkdev_releasepage(struct page *page, gfp_t wait)
1948 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1950 if (super && super->s_op->bdev_try_to_free_page)
1951 return super->s_op->bdev_try_to_free_page(super, page, wait);
1953 return try_to_free_buffers(page);
1956 static int blkdev_writepages(struct address_space *mapping,
1957 struct writeback_control *wbc)
1959 return generic_writepages(mapping, wbc);
1962 static const struct address_space_operations def_blk_aops = {
1963 .readpage = blkdev_readpage,
1964 .readpages = blkdev_readpages,
1965 .writepage = blkdev_writepage,
1966 .write_begin = blkdev_write_begin,
1967 .write_end = blkdev_write_end,
1968 .writepages = blkdev_writepages,
1969 .releasepage = blkdev_releasepage,
1970 .direct_IO = blkdev_direct_IO,
1971 .is_dirty_writeback = buffer_check_dirty_writeback,
1974 #define BLKDEV_FALLOC_FL_SUPPORTED \
1975 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1976 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1978 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1981 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1982 struct address_space *mapping;
1983 loff_t end = start + len - 1;
1987 /* Fail if we don't recognize the flags. */
1988 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1991 /* Don't go off the end of the device. */
1992 isize = i_size_read(bdev->bd_inode);
1996 if (mode & FALLOC_FL_KEEP_SIZE) {
1997 len = isize - start;
1998 end = start + len - 1;
2004 * Don't allow IO that isn't aligned to logical block size.
2006 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2009 /* Invalidate the page cache, including dirty pages. */
2010 mapping = bdev->bd_inode->i_mapping;
2011 truncate_inode_pages_range(mapping, start, end);
2014 case FALLOC_FL_ZERO_RANGE:
2015 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2016 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2017 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2019 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2020 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2021 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2023 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2024 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2034 * Invalidate again; if someone wandered in and dirtied a page,
2035 * the caller will be given -EBUSY. The third argument is
2036 * inclusive, so the rounding here is safe.
2038 return invalidate_inode_pages2_range(mapping,
2039 start >> PAGE_SHIFT,
2043 const struct file_operations def_blk_fops = {
2044 .open = blkdev_open,
2045 .release = blkdev_close,
2046 .llseek = block_llseek,
2047 .read_iter = blkdev_read_iter,
2048 .write_iter = blkdev_write_iter,
2049 .mmap = generic_file_mmap,
2050 .fsync = blkdev_fsync,
2051 .unlocked_ioctl = block_ioctl,
2052 #ifdef CONFIG_COMPAT
2053 .compat_ioctl = compat_blkdev_ioctl,
2055 .splice_read = generic_file_splice_read,
2056 .splice_write = iter_file_splice_write,
2057 .fallocate = blkdev_fallocate,
2060 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2063 mm_segment_t old_fs = get_fs();
2065 res = blkdev_ioctl(bdev, 0, cmd, arg);
2070 EXPORT_SYMBOL(ioctl_by_bdev);
2073 * lookup_bdev - lookup a struct block_device by name
2074 * @pathname: special file representing the block device
2076 * Get a reference to the blockdevice at @pathname in the current
2077 * namespace if possible and return it. Return ERR_PTR(error)
2080 struct block_device *lookup_bdev(const char *pathname)
2082 struct block_device *bdev;
2083 struct inode *inode;
2087 if (!pathname || !*pathname)
2088 return ERR_PTR(-EINVAL);
2090 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2092 return ERR_PTR(error);
2094 inode = d_backing_inode(path.dentry);
2096 if (!S_ISBLK(inode->i_mode))
2099 if (!may_open_dev(&path))
2102 bdev = bd_acquire(inode);
2109 bdev = ERR_PTR(error);
2112 EXPORT_SYMBOL(lookup_bdev);
2114 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2116 struct super_block *sb = get_super(bdev);
2121 * no need to lock the super, get_super holds the
2122 * read mutex so the filesystem cannot go away
2123 * under us (->put_super runs with the write lock
2126 shrink_dcache_sb(sb);
2127 res = invalidate_inodes(sb, kill_dirty);
2130 invalidate_bdev(bdev);
2133 EXPORT_SYMBOL(__invalidate_device);
2135 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2137 struct inode *inode, *old_inode = NULL;
2139 spin_lock(&blockdev_superblock->s_inode_list_lock);
2140 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2141 struct address_space *mapping = inode->i_mapping;
2142 struct block_device *bdev;
2144 spin_lock(&inode->i_lock);
2145 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2146 mapping->nrpages == 0) {
2147 spin_unlock(&inode->i_lock);
2151 spin_unlock(&inode->i_lock);
2152 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2154 * We hold a reference to 'inode' so it couldn't have been
2155 * removed from s_inodes list while we dropped the
2156 * s_inode_list_lock We cannot iput the inode now as we can
2157 * be holding the last reference and we cannot iput it under
2158 * s_inode_list_lock. So we keep the reference and iput it
2163 bdev = I_BDEV(inode);
2165 mutex_lock(&bdev->bd_mutex);
2166 if (bdev->bd_openers)
2168 mutex_unlock(&bdev->bd_mutex);
2170 spin_lock(&blockdev_superblock->s_inode_list_lock);
2172 spin_unlock(&blockdev_superblock->s_inode_list_lock);