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(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
213 bio_init(&bio, vecs, nr_pages);
214 bio_set_dev(&bio, bdev);
215 bio.bi_iter.bi_sector = pos >> 9;
216 bio.bi_write_hint = iocb->ki_hint;
217 bio.bi_private = current;
218 bio.bi_end_io = blkdev_bio_end_io_simple;
219 bio.bi_ioprio = iocb->ki_ioprio;
221 ret = bio_iov_iter_get_pages(&bio, iter);
224 ret = bio.bi_iter.bi_size;
226 if (iov_iter_rw(iter) == READ) {
227 bio.bi_opf = REQ_OP_READ;
228 if (iter_is_iovec(iter))
231 bio.bi_opf = dio_bio_write_op(iocb);
232 task_io_account_write(ret);
235 qc = submit_bio(&bio);
237 set_current_state(TASK_UNINTERRUPTIBLE);
238 if (!READ_ONCE(bio.bi_private))
240 if (!(iocb->ki_flags & IOCB_HIPRI) ||
241 !blk_poll(bdev_get_queue(bdev), qc))
244 __set_current_state(TASK_RUNNING);
246 bio_for_each_segment_all(bvec, &bio, i) {
247 if (should_dirty && !PageCompound(bvec->bv_page))
248 set_page_dirty_lock(bvec->bv_page);
249 put_page(bvec->bv_page);
252 if (vecs != inline_vecs)
255 if (unlikely(bio.bi_status))
256 ret = blk_status_to_errno(bio.bi_status);
266 struct task_struct *waiter;
271 bool should_dirty : 1;
276 static struct bio_set blkdev_dio_pool;
278 static void blkdev_bio_end_io(struct bio *bio)
280 struct blkdev_dio *dio = bio->bi_private;
281 bool should_dirty = dio->should_dirty;
283 if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
284 if (bio->bi_status && !dio->bio.bi_status)
285 dio->bio.bi_status = bio->bi_status;
288 struct kiocb *iocb = dio->iocb;
291 if (likely(!dio->bio.bi_status)) {
295 ret = blk_status_to_errno(dio->bio.bi_status);
298 dio->iocb->ki_complete(iocb, ret, 0);
301 struct task_struct *waiter = dio->waiter;
303 WRITE_ONCE(dio->waiter, NULL);
304 wake_up_process(waiter);
309 bio_check_pages_dirty(bio);
311 struct bio_vec *bvec;
314 bio_for_each_segment_all(bvec, bio, i)
315 put_page(bvec->bv_page);
321 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
323 struct file *file = iocb->ki_filp;
324 struct inode *inode = bdev_file_inode(file);
325 struct block_device *bdev = I_BDEV(inode);
326 struct blk_plug plug;
327 struct blkdev_dio *dio;
329 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
330 loff_t pos = iocb->ki_pos;
331 blk_qc_t qc = BLK_QC_T_NONE;
334 if ((pos | iov_iter_alignment(iter)) &
335 (bdev_logical_block_size(bdev) - 1))
338 bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, &blkdev_dio_pool);
339 bio_get(bio); /* extra ref for the completion handler */
341 dio = container_of(bio, struct blkdev_dio, bio);
342 dio->is_sync = is_sync = is_sync_kiocb(iocb);
344 dio->waiter = current;
349 dio->multi_bio = false;
350 dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
352 blk_start_plug(&plug);
354 bio_set_dev(bio, bdev);
355 bio->bi_iter.bi_sector = pos >> 9;
356 bio->bi_write_hint = iocb->ki_hint;
357 bio->bi_private = dio;
358 bio->bi_end_io = blkdev_bio_end_io;
359 bio->bi_ioprio = iocb->ki_ioprio;
361 ret = bio_iov_iter_get_pages(bio, iter);
363 bio->bi_status = BLK_STS_IOERR;
369 bio->bi_opf = REQ_OP_READ;
370 if (dio->should_dirty)
371 bio_set_pages_dirty(bio);
373 bio->bi_opf = dio_bio_write_op(iocb);
374 task_io_account_write(bio->bi_iter.bi_size);
377 dio->size += bio->bi_iter.bi_size;
378 pos += bio->bi_iter.bi_size;
380 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
382 qc = submit_bio(bio);
386 if (!dio->multi_bio) {
387 dio->multi_bio = true;
388 atomic_set(&dio->ref, 2);
390 atomic_inc(&dio->ref);
394 bio = bio_alloc(GFP_KERNEL, nr_pages);
396 blk_finish_plug(&plug);
402 set_current_state(TASK_UNINTERRUPTIBLE);
403 if (!READ_ONCE(dio->waiter))
406 if (!(iocb->ki_flags & IOCB_HIPRI) ||
407 !blk_poll(bdev_get_queue(bdev), qc))
410 __set_current_state(TASK_RUNNING);
413 ret = blk_status_to_errno(dio->bio.bi_status);
422 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
426 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
429 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
430 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
432 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
435 static __init int blkdev_init(void)
437 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
439 module_init(blkdev_init);
441 int __sync_blockdev(struct block_device *bdev, int wait)
446 return filemap_flush(bdev->bd_inode->i_mapping);
447 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
451 * Write out and wait upon all the dirty data associated with a block
452 * device via its mapping. Does not take the superblock lock.
454 int sync_blockdev(struct block_device *bdev)
456 return __sync_blockdev(bdev, 1);
458 EXPORT_SYMBOL(sync_blockdev);
461 * Write out and wait upon all dirty data associated with this
462 * device. Filesystem data as well as the underlying block
463 * device. Takes the superblock lock.
465 int fsync_bdev(struct block_device *bdev)
467 struct super_block *sb = get_super(bdev);
469 int res = sync_filesystem(sb);
473 return sync_blockdev(bdev);
475 EXPORT_SYMBOL(fsync_bdev);
478 * freeze_bdev -- lock a filesystem and force it into a consistent state
479 * @bdev: blockdevice to lock
481 * If a superblock is found on this device, we take the s_umount semaphore
482 * on it to make sure nobody unmounts until the snapshot creation is done.
483 * The reference counter (bd_fsfreeze_count) guarantees that only the last
484 * unfreeze process can unfreeze the frozen filesystem actually when multiple
485 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
486 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
489 struct super_block *freeze_bdev(struct block_device *bdev)
491 struct super_block *sb;
494 mutex_lock(&bdev->bd_fsfreeze_mutex);
495 if (++bdev->bd_fsfreeze_count > 1) {
497 * We don't even need to grab a reference - the first call
498 * to freeze_bdev grab an active reference and only the last
499 * thaw_bdev drops it.
501 sb = get_super(bdev);
504 mutex_unlock(&bdev->bd_fsfreeze_mutex);
508 sb = get_active_super(bdev);
511 if (sb->s_op->freeze_super)
512 error = sb->s_op->freeze_super(sb);
514 error = freeze_super(sb);
516 deactivate_super(sb);
517 bdev->bd_fsfreeze_count--;
518 mutex_unlock(&bdev->bd_fsfreeze_mutex);
519 return ERR_PTR(error);
521 deactivate_super(sb);
524 mutex_unlock(&bdev->bd_fsfreeze_mutex);
525 return sb; /* thaw_bdev releases s->s_umount */
527 EXPORT_SYMBOL(freeze_bdev);
530 * thaw_bdev -- unlock filesystem
531 * @bdev: blockdevice to unlock
532 * @sb: associated superblock
534 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
536 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
540 mutex_lock(&bdev->bd_fsfreeze_mutex);
541 if (!bdev->bd_fsfreeze_count)
545 if (--bdev->bd_fsfreeze_count > 0)
551 if (sb->s_op->thaw_super)
552 error = sb->s_op->thaw_super(sb);
554 error = thaw_super(sb);
556 bdev->bd_fsfreeze_count++;
558 mutex_unlock(&bdev->bd_fsfreeze_mutex);
561 EXPORT_SYMBOL(thaw_bdev);
563 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
565 return block_write_full_page(page, blkdev_get_block, wbc);
568 static int blkdev_readpage(struct file * file, struct page * page)
570 return block_read_full_page(page, blkdev_get_block);
573 static int blkdev_readpages(struct file *file, struct address_space *mapping,
574 struct list_head *pages, unsigned nr_pages)
576 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
579 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
580 loff_t pos, unsigned len, unsigned flags,
581 struct page **pagep, void **fsdata)
583 return block_write_begin(mapping, pos, len, flags, pagep,
587 static int blkdev_write_end(struct file *file, struct address_space *mapping,
588 loff_t pos, unsigned len, unsigned copied,
589 struct page *page, void *fsdata)
592 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
602 * for a block special file file_inode(file)->i_size is zero
603 * so we compute the size by hand (just as in block_read/write above)
605 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
607 struct inode *bd_inode = bdev_file_inode(file);
610 inode_lock(bd_inode);
611 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
612 inode_unlock(bd_inode);
616 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
618 struct inode *bd_inode = bdev_file_inode(filp);
619 struct block_device *bdev = I_BDEV(bd_inode);
622 error = file_write_and_wait_range(filp, start, end);
627 * There is no need to serialise calls to blkdev_issue_flush with
628 * i_mutex and doing so causes performance issues with concurrent
629 * O_SYNC writers to a block device.
631 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
632 if (error == -EOPNOTSUPP)
637 EXPORT_SYMBOL(blkdev_fsync);
640 * bdev_read_page() - Start reading a page from a block device
641 * @bdev: The device to read the page from
642 * @sector: The offset on the device to read the page to (need not be aligned)
643 * @page: The page to read
645 * On entry, the page should be locked. It will be unlocked when the page
646 * has been read. If the block driver implements rw_page synchronously,
647 * that will be true on exit from this function, but it need not be.
649 * Errors returned by this function are usually "soft", eg out of memory, or
650 * queue full; callers should try a different route to read this page rather
651 * than propagate an error back up the stack.
653 * Return: negative errno if an error occurs, 0 if submission was successful.
655 int bdev_read_page(struct block_device *bdev, sector_t sector,
658 const struct block_device_operations *ops = bdev->bd_disk->fops;
659 int result = -EOPNOTSUPP;
661 if (!ops->rw_page || bdev_get_integrity(bdev))
664 result = blk_queue_enter(bdev->bd_queue, 0);
667 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
668 blk_queue_exit(bdev->bd_queue);
671 EXPORT_SYMBOL_GPL(bdev_read_page);
674 * bdev_write_page() - Start writing a page to a block device
675 * @bdev: The device to write the page to
676 * @sector: The offset on the device to write the page to (need not be aligned)
677 * @page: The page to write
678 * @wbc: The writeback_control for the write
680 * On entry, the page should be locked and not currently under writeback.
681 * On exit, if the write started successfully, the page will be unlocked and
682 * under writeback. If the write failed already (eg the driver failed to
683 * queue the page to the device), the page will still be locked. If the
684 * caller is a ->writepage implementation, it will need to unlock the page.
686 * Errors returned by this function are usually "soft", eg out of memory, or
687 * queue full; callers should try a different route to write this page rather
688 * than propagate an error back up the stack.
690 * Return: negative errno if an error occurs, 0 if submission was successful.
692 int bdev_write_page(struct block_device *bdev, sector_t sector,
693 struct page *page, struct writeback_control *wbc)
696 const struct block_device_operations *ops = bdev->bd_disk->fops;
698 if (!ops->rw_page || bdev_get_integrity(bdev))
700 result = blk_queue_enter(bdev->bd_queue, 0);
704 set_page_writeback(page);
705 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
707 end_page_writeback(page);
709 clean_page_buffers(page);
712 blk_queue_exit(bdev->bd_queue);
715 EXPORT_SYMBOL_GPL(bdev_write_page);
721 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
722 static struct kmem_cache * bdev_cachep __read_mostly;
724 static struct inode *bdev_alloc_inode(struct super_block *sb)
726 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
729 return &ei->vfs_inode;
732 static void bdev_i_callback(struct rcu_head *head)
734 struct inode *inode = container_of(head, struct inode, i_rcu);
735 struct bdev_inode *bdi = BDEV_I(inode);
737 kmem_cache_free(bdev_cachep, bdi);
740 static void bdev_destroy_inode(struct inode *inode)
742 call_rcu(&inode->i_rcu, bdev_i_callback);
745 static void init_once(void *foo)
747 struct bdev_inode *ei = (struct bdev_inode *) foo;
748 struct block_device *bdev = &ei->bdev;
750 memset(bdev, 0, sizeof(*bdev));
751 mutex_init(&bdev->bd_mutex);
752 INIT_LIST_HEAD(&bdev->bd_list);
754 INIT_LIST_HEAD(&bdev->bd_holder_disks);
756 bdev->bd_bdi = &noop_backing_dev_info;
757 inode_init_once(&ei->vfs_inode);
758 /* Initialize mutex for freeze. */
759 mutex_init(&bdev->bd_fsfreeze_mutex);
762 static void bdev_evict_inode(struct inode *inode)
764 struct block_device *bdev = &BDEV_I(inode)->bdev;
765 truncate_inode_pages_final(&inode->i_data);
766 invalidate_inode_buffers(inode); /* is it needed here? */
768 spin_lock(&bdev_lock);
769 list_del_init(&bdev->bd_list);
770 spin_unlock(&bdev_lock);
771 /* Detach inode from wb early as bdi_put() may free bdi->wb */
772 inode_detach_wb(inode);
773 if (bdev->bd_bdi != &noop_backing_dev_info) {
774 bdi_put(bdev->bd_bdi);
775 bdev->bd_bdi = &noop_backing_dev_info;
779 static const struct super_operations bdev_sops = {
780 .statfs = simple_statfs,
781 .alloc_inode = bdev_alloc_inode,
782 .destroy_inode = bdev_destroy_inode,
783 .drop_inode = generic_delete_inode,
784 .evict_inode = bdev_evict_inode,
787 static struct dentry *bd_mount(struct file_system_type *fs_type,
788 int flags, const char *dev_name, void *data)
791 dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
793 dent->d_sb->s_iflags |= SB_I_CGROUPWB;
797 static struct file_system_type bd_type = {
800 .kill_sb = kill_anon_super,
803 struct super_block *blockdev_superblock __read_mostly;
804 EXPORT_SYMBOL_GPL(blockdev_superblock);
806 void __init bdev_cache_init(void)
809 static struct vfsmount *bd_mnt;
811 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
812 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
813 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
815 err = register_filesystem(&bd_type);
817 panic("Cannot register bdev pseudo-fs");
818 bd_mnt = kern_mount(&bd_type);
820 panic("Cannot create bdev pseudo-fs");
821 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
825 * Most likely _very_ bad one - but then it's hardly critical for small
826 * /dev and can be fixed when somebody will need really large one.
827 * Keep in mind that it will be fed through icache hash function too.
829 static inline unsigned long hash(dev_t dev)
831 return MAJOR(dev)+MINOR(dev);
834 static int bdev_test(struct inode *inode, void *data)
836 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
839 static int bdev_set(struct inode *inode, void *data)
841 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
845 static LIST_HEAD(all_bdevs);
848 * If there is a bdev inode for this device, unhash it so that it gets evicted
849 * as soon as last inode reference is dropped.
851 void bdev_unhash_inode(dev_t dev)
855 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
857 remove_inode_hash(inode);
862 struct block_device *bdget(dev_t dev)
864 struct block_device *bdev;
867 inode = iget5_locked(blockdev_superblock, hash(dev),
868 bdev_test, bdev_set, &dev);
873 bdev = &BDEV_I(inode)->bdev;
875 if (inode->i_state & I_NEW) {
876 bdev->bd_contains = NULL;
877 bdev->bd_super = NULL;
878 bdev->bd_inode = inode;
879 bdev->bd_block_size = i_blocksize(inode);
880 bdev->bd_part_count = 0;
881 bdev->bd_invalidated = 0;
882 inode->i_mode = S_IFBLK;
884 inode->i_bdev = bdev;
885 inode->i_data.a_ops = &def_blk_aops;
886 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
887 spin_lock(&bdev_lock);
888 list_add(&bdev->bd_list, &all_bdevs);
889 spin_unlock(&bdev_lock);
890 unlock_new_inode(inode);
895 EXPORT_SYMBOL(bdget);
898 * bdgrab -- Grab a reference to an already referenced block device
899 * @bdev: Block device to grab a reference to.
901 struct block_device *bdgrab(struct block_device *bdev)
903 ihold(bdev->bd_inode);
906 EXPORT_SYMBOL(bdgrab);
908 long nr_blockdev_pages(void)
910 struct block_device *bdev;
912 spin_lock(&bdev_lock);
913 list_for_each_entry(bdev, &all_bdevs, bd_list) {
914 ret += bdev->bd_inode->i_mapping->nrpages;
916 spin_unlock(&bdev_lock);
920 void bdput(struct block_device *bdev)
922 iput(bdev->bd_inode);
925 EXPORT_SYMBOL(bdput);
927 static struct block_device *bd_acquire(struct inode *inode)
929 struct block_device *bdev;
931 spin_lock(&bdev_lock);
932 bdev = inode->i_bdev;
933 if (bdev && !inode_unhashed(bdev->bd_inode)) {
935 spin_unlock(&bdev_lock);
938 spin_unlock(&bdev_lock);
941 * i_bdev references block device inode that was already shut down
942 * (corresponding device got removed). Remove the reference and look
943 * up block device inode again just in case new device got
944 * reestablished under the same device number.
949 bdev = bdget(inode->i_rdev);
951 spin_lock(&bdev_lock);
952 if (!inode->i_bdev) {
954 * We take an additional reference to bd_inode,
955 * and it's released in clear_inode() of inode.
956 * So, we can access it via ->i_mapping always
960 inode->i_bdev = bdev;
961 inode->i_mapping = bdev->bd_inode->i_mapping;
963 spin_unlock(&bdev_lock);
968 /* Call when you free inode */
970 void bd_forget(struct inode *inode)
972 struct block_device *bdev = NULL;
974 spin_lock(&bdev_lock);
975 if (!sb_is_blkdev_sb(inode->i_sb))
976 bdev = inode->i_bdev;
977 inode->i_bdev = NULL;
978 inode->i_mapping = &inode->i_data;
979 spin_unlock(&bdev_lock);
986 * bd_may_claim - test whether a block device can be claimed
987 * @bdev: block device of interest
988 * @whole: whole block device containing @bdev, may equal @bdev
989 * @holder: holder trying to claim @bdev
991 * Test whether @bdev can be claimed by @holder.
994 * spin_lock(&bdev_lock).
997 * %true if @bdev can be claimed, %false otherwise.
999 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1002 if (bdev->bd_holder == holder)
1003 return true; /* already a holder */
1004 else if (bdev->bd_holder != NULL)
1005 return false; /* held by someone else */
1006 else if (whole == bdev)
1007 return true; /* is a whole device which isn't held */
1009 else if (whole->bd_holder == bd_may_claim)
1010 return true; /* is a partition of a device that is being partitioned */
1011 else if (whole->bd_holder != NULL)
1012 return false; /* is a partition of a held device */
1014 return true; /* is a partition of an un-held device */
1018 * bd_prepare_to_claim - prepare to claim a block device
1019 * @bdev: block device of interest
1020 * @whole: the whole device containing @bdev, may equal @bdev
1021 * @holder: holder trying to claim @bdev
1023 * Prepare to claim @bdev. This function fails if @bdev is already
1024 * claimed by another holder and waits if another claiming is in
1025 * progress. This function doesn't actually claim. On successful
1026 * return, the caller has ownership of bd_claiming and bd_holder[s].
1029 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1030 * it multiple times.
1033 * 0 if @bdev can be claimed, -EBUSY otherwise.
1035 static int bd_prepare_to_claim(struct block_device *bdev,
1036 struct block_device *whole, void *holder)
1039 /* if someone else claimed, fail */
1040 if (!bd_may_claim(bdev, whole, holder))
1043 /* if claiming is already in progress, wait for it to finish */
1044 if (whole->bd_claiming) {
1045 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1048 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1049 spin_unlock(&bdev_lock);
1051 finish_wait(wq, &wait);
1052 spin_lock(&bdev_lock);
1060 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1062 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1067 * Now that we hold gendisk reference we make sure bdev we looked up is
1068 * not stale. If it is, it means device got removed and created before
1069 * we looked up gendisk and we fail open in such case. Associating
1070 * unhashed bdev with newly created gendisk could lead to two bdevs
1071 * (and thus two independent caches) being associated with one device
1074 if (inode_unhashed(bdev->bd_inode)) {
1075 put_disk_and_module(disk);
1082 * bd_start_claiming - start claiming a block device
1083 * @bdev: block device of interest
1084 * @holder: holder trying to claim @bdev
1086 * @bdev is about to be opened exclusively. Check @bdev can be opened
1087 * exclusively and mark that an exclusive open is in progress. Each
1088 * successful call to this function must be matched with a call to
1089 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1092 * This function is used to gain exclusive access to the block device
1093 * without actually causing other exclusive open attempts to fail. It
1094 * should be used when the open sequence itself requires exclusive
1095 * access but may subsequently fail.
1101 * Pointer to the block device containing @bdev on success, ERR_PTR()
1104 static struct block_device *bd_start_claiming(struct block_device *bdev,
1107 struct gendisk *disk;
1108 struct block_device *whole;
1114 * @bdev might not have been initialized properly yet, look up
1115 * and grab the outer block device the hard way.
1117 disk = bdev_get_gendisk(bdev, &partno);
1119 return ERR_PTR(-ENXIO);
1122 * Normally, @bdev should equal what's returned from bdget_disk()
1123 * if partno is 0; however, some drivers (floppy) use multiple
1124 * bdev's for the same physical device and @bdev may be one of the
1125 * aliases. Keep @bdev if partno is 0. This means claimer
1126 * tracking is broken for those devices but it has always been that
1130 whole = bdget_disk(disk, 0);
1132 whole = bdgrab(bdev);
1134 put_disk_and_module(disk);
1136 return ERR_PTR(-ENOMEM);
1138 /* prepare to claim, if successful, mark claiming in progress */
1139 spin_lock(&bdev_lock);
1141 err = bd_prepare_to_claim(bdev, whole, holder);
1143 whole->bd_claiming = holder;
1144 spin_unlock(&bdev_lock);
1147 spin_unlock(&bdev_lock);
1149 return ERR_PTR(err);
1154 struct bd_holder_disk {
1155 struct list_head list;
1156 struct gendisk *disk;
1160 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1161 struct gendisk *disk)
1163 struct bd_holder_disk *holder;
1165 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1166 if (holder->disk == disk)
1171 static int add_symlink(struct kobject *from, struct kobject *to)
1173 return sysfs_create_link(from, to, kobject_name(to));
1176 static void del_symlink(struct kobject *from, struct kobject *to)
1178 sysfs_remove_link(from, kobject_name(to));
1182 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1183 * @bdev: the claimed slave bdev
1184 * @disk: the holding disk
1186 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1188 * This functions creates the following sysfs symlinks.
1190 * - from "slaves" directory of the holder @disk to the claimed @bdev
1191 * - from "holders" directory of the @bdev to the holder @disk
1193 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1194 * passed to bd_link_disk_holder(), then:
1196 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1197 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1199 * The caller must have claimed @bdev before calling this function and
1200 * ensure that both @bdev and @disk are valid during the creation and
1201 * lifetime of these symlinks.
1207 * 0 on success, -errno on failure.
1209 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1211 struct bd_holder_disk *holder;
1214 mutex_lock(&bdev->bd_mutex);
1216 WARN_ON_ONCE(!bdev->bd_holder);
1218 /* FIXME: remove the following once add_disk() handles errors */
1219 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1222 holder = bd_find_holder_disk(bdev, disk);
1228 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1234 INIT_LIST_HEAD(&holder->list);
1235 holder->disk = disk;
1238 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1242 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1246 * bdev could be deleted beneath us which would implicitly destroy
1247 * the holder directory. Hold on to it.
1249 kobject_get(bdev->bd_part->holder_dir);
1251 list_add(&holder->list, &bdev->bd_holder_disks);
1255 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1259 mutex_unlock(&bdev->bd_mutex);
1262 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1265 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1266 * @bdev: the calimed slave bdev
1267 * @disk: the holding disk
1269 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1274 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1276 struct bd_holder_disk *holder;
1278 mutex_lock(&bdev->bd_mutex);
1280 holder = bd_find_holder_disk(bdev, disk);
1282 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1283 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1284 del_symlink(bdev->bd_part->holder_dir,
1285 &disk_to_dev(disk)->kobj);
1286 kobject_put(bdev->bd_part->holder_dir);
1287 list_del_init(&holder->list);
1291 mutex_unlock(&bdev->bd_mutex);
1293 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1297 * flush_disk - invalidates all buffer-cache entries on a disk
1299 * @bdev: struct block device to be flushed
1300 * @kill_dirty: flag to guide handling of dirty inodes
1302 * Invalidates all buffer-cache entries on a disk. It should be called
1303 * when a disk has been changed -- either by a media change or online
1306 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1308 if (__invalidate_device(bdev, kill_dirty)) {
1309 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1310 "resized disk %s\n",
1311 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1316 if (disk_part_scan_enabled(bdev->bd_disk))
1317 bdev->bd_invalidated = 1;
1321 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1322 * @disk: struct gendisk to check
1323 * @bdev: struct bdev to adjust.
1324 * @verbose: if %true log a message about a size change if there is any
1326 * This routine checks to see if the bdev size does not match the disk size
1327 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1330 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1333 loff_t disk_size, bdev_size;
1335 disk_size = (loff_t)get_capacity(disk) << 9;
1336 bdev_size = i_size_read(bdev->bd_inode);
1337 if (disk_size != bdev_size) {
1340 "%s: detected capacity change from %lld to %lld\n",
1341 disk->disk_name, bdev_size, disk_size);
1343 i_size_write(bdev->bd_inode, disk_size);
1344 if (bdev_size > disk_size)
1345 flush_disk(bdev, false);
1350 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1351 * @disk: struct gendisk to be revalidated
1353 * This routine is a wrapper for lower-level driver's revalidate_disk
1354 * call-backs. It is used to do common pre and post operations needed
1355 * for all revalidate_disk operations.
1357 int revalidate_disk(struct gendisk *disk)
1359 struct block_device *bdev;
1362 if (disk->fops->revalidate_disk)
1363 ret = disk->fops->revalidate_disk(disk);
1364 bdev = bdget_disk(disk, 0);
1368 mutex_lock(&bdev->bd_mutex);
1369 check_disk_size_change(disk, bdev, ret == 0);
1370 bdev->bd_invalidated = 0;
1371 mutex_unlock(&bdev->bd_mutex);
1375 EXPORT_SYMBOL(revalidate_disk);
1378 * This routine checks whether a removable media has been changed,
1379 * and invalidates all buffer-cache-entries in that case. This
1380 * is a relatively slow routine, so we have to try to minimize using
1381 * it. Thus it is called only upon a 'mount' or 'open'. This
1382 * is the best way of combining speed and utility, I think.
1383 * People changing diskettes in the middle of an operation deserve
1386 int check_disk_change(struct block_device *bdev)
1388 struct gendisk *disk = bdev->bd_disk;
1389 const struct block_device_operations *bdops = disk->fops;
1390 unsigned int events;
1392 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1393 DISK_EVENT_EJECT_REQUEST);
1394 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1397 flush_disk(bdev, true);
1398 if (bdops->revalidate_disk)
1399 bdops->revalidate_disk(bdev->bd_disk);
1403 EXPORT_SYMBOL(check_disk_change);
1405 void bd_set_size(struct block_device *bdev, loff_t size)
1407 unsigned bsize = bdev_logical_block_size(bdev);
1409 inode_lock(bdev->bd_inode);
1410 i_size_write(bdev->bd_inode, size);
1411 inode_unlock(bdev->bd_inode);
1412 while (bsize < PAGE_SIZE) {
1417 bdev->bd_block_size = bsize;
1418 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1420 EXPORT_SYMBOL(bd_set_size);
1422 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1427 * mutex_lock(part->bd_mutex)
1428 * mutex_lock_nested(whole->bd_mutex, 1)
1431 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1433 struct gendisk *disk;
1437 bool first_open = false;
1439 if (mode & FMODE_READ)
1441 if (mode & FMODE_WRITE)
1444 * hooks: /n/, see "layering violations".
1447 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1457 disk = bdev_get_gendisk(bdev, &partno);
1461 disk_block_events(disk);
1462 mutex_lock_nested(&bdev->bd_mutex, for_part);
1463 if (!bdev->bd_openers) {
1465 bdev->bd_disk = disk;
1466 bdev->bd_queue = disk->queue;
1467 bdev->bd_contains = bdev;
1468 bdev->bd_partno = partno;
1472 bdev->bd_part = disk_get_part(disk, partno);
1477 if (disk->fops->open) {
1478 ret = disk->fops->open(bdev, mode);
1479 if (ret == -ERESTARTSYS) {
1480 /* Lost a race with 'disk' being
1481 * deleted, try again.
1484 disk_put_part(bdev->bd_part);
1485 bdev->bd_part = NULL;
1486 bdev->bd_disk = NULL;
1487 bdev->bd_queue = NULL;
1488 mutex_unlock(&bdev->bd_mutex);
1489 disk_unblock_events(disk);
1490 put_disk_and_module(disk);
1496 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1499 * If the device is invalidated, rescan partition
1500 * if open succeeded or failed with -ENOMEDIUM.
1501 * The latter is necessary to prevent ghost
1502 * partitions on a removed medium.
1504 if (bdev->bd_invalidated) {
1506 rescan_partitions(disk, bdev);
1507 else if (ret == -ENOMEDIUM)
1508 invalidate_partitions(disk, bdev);
1514 struct block_device *whole;
1515 whole = bdget_disk(disk, 0);
1520 ret = __blkdev_get(whole, mode, 1);
1523 bdev->bd_contains = whole;
1524 bdev->bd_part = disk_get_part(disk, partno);
1525 if (!(disk->flags & GENHD_FL_UP) ||
1526 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1530 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1533 if (bdev->bd_bdi == &noop_backing_dev_info)
1534 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1536 if (bdev->bd_contains == bdev) {
1538 if (bdev->bd_disk->fops->open)
1539 ret = bdev->bd_disk->fops->open(bdev, mode);
1540 /* the same as first opener case, read comment there */
1541 if (bdev->bd_invalidated) {
1543 rescan_partitions(bdev->bd_disk, bdev);
1544 else if (ret == -ENOMEDIUM)
1545 invalidate_partitions(bdev->bd_disk, bdev);
1548 goto out_unlock_bdev;
1553 bdev->bd_part_count++;
1554 mutex_unlock(&bdev->bd_mutex);
1555 disk_unblock_events(disk);
1556 /* only one opener holds refs to the module and disk */
1558 put_disk_and_module(disk);
1562 disk_put_part(bdev->bd_part);
1563 bdev->bd_disk = NULL;
1564 bdev->bd_part = NULL;
1565 bdev->bd_queue = NULL;
1566 if (bdev != bdev->bd_contains)
1567 __blkdev_put(bdev->bd_contains, mode, 1);
1568 bdev->bd_contains = NULL;
1570 mutex_unlock(&bdev->bd_mutex);
1571 disk_unblock_events(disk);
1572 put_disk_and_module(disk);
1580 * blkdev_get - open a block device
1581 * @bdev: block_device to open
1582 * @mode: FMODE_* mask
1583 * @holder: exclusive holder identifier
1585 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1586 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1587 * @holder is invalid. Exclusive opens may nest for the same @holder.
1589 * On success, the reference count of @bdev is unchanged. On failure,
1596 * 0 on success, -errno on failure.
1598 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1600 struct block_device *whole = NULL;
1603 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1605 if ((mode & FMODE_EXCL) && holder) {
1606 whole = bd_start_claiming(bdev, holder);
1607 if (IS_ERR(whole)) {
1609 return PTR_ERR(whole);
1613 res = __blkdev_get(bdev, mode, 0);
1616 struct gendisk *disk = whole->bd_disk;
1618 /* finish claiming */
1619 mutex_lock(&bdev->bd_mutex);
1620 spin_lock(&bdev_lock);
1623 BUG_ON(!bd_may_claim(bdev, whole, holder));
1625 * Note that for a whole device bd_holders
1626 * will be incremented twice, and bd_holder
1627 * will be set to bd_may_claim before being
1630 whole->bd_holders++;
1631 whole->bd_holder = bd_may_claim;
1633 bdev->bd_holder = holder;
1636 /* tell others that we're done */
1637 BUG_ON(whole->bd_claiming != holder);
1638 whole->bd_claiming = NULL;
1639 wake_up_bit(&whole->bd_claiming, 0);
1641 spin_unlock(&bdev_lock);
1644 * Block event polling for write claims if requested. Any
1645 * write holder makes the write_holder state stick until
1646 * all are released. This is good enough and tracking
1647 * individual writeable reference is too fragile given the
1648 * way @mode is used in blkdev_get/put().
1650 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1651 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1652 bdev->bd_write_holder = true;
1653 disk_block_events(disk);
1656 mutex_unlock(&bdev->bd_mutex);
1662 EXPORT_SYMBOL(blkdev_get);
1665 * blkdev_get_by_path - open a block device by name
1666 * @path: path to the block device to open
1667 * @mode: FMODE_* mask
1668 * @holder: exclusive holder identifier
1670 * Open the blockdevice described by the device file at @path. @mode
1671 * and @holder are identical to blkdev_get().
1673 * On success, the returned block_device has reference count of one.
1679 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1681 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1684 struct block_device *bdev;
1687 bdev = lookup_bdev(path);
1691 err = blkdev_get(bdev, mode, holder);
1693 return ERR_PTR(err);
1695 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1696 blkdev_put(bdev, mode);
1697 return ERR_PTR(-EACCES);
1702 EXPORT_SYMBOL(blkdev_get_by_path);
1705 * blkdev_get_by_dev - open a block device by device number
1706 * @dev: device number of block device to open
1707 * @mode: FMODE_* mask
1708 * @holder: exclusive holder identifier
1710 * Open the blockdevice described by device number @dev. @mode and
1711 * @holder are identical to blkdev_get().
1713 * Use it ONLY if you really do not have anything better - i.e. when
1714 * you are behind a truly sucky interface and all you are given is a
1715 * device number. _Never_ to be used for internal purposes. If you
1716 * ever need it - reconsider your API.
1718 * On success, the returned block_device has reference count of one.
1724 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1726 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1728 struct block_device *bdev;
1733 return ERR_PTR(-ENOMEM);
1735 err = blkdev_get(bdev, mode, holder);
1737 return ERR_PTR(err);
1741 EXPORT_SYMBOL(blkdev_get_by_dev);
1743 static int blkdev_open(struct inode * inode, struct file * filp)
1745 struct block_device *bdev;
1748 * Preserve backwards compatibility and allow large file access
1749 * even if userspace doesn't ask for it explicitly. Some mkfs
1750 * binary needs it. We might want to drop this workaround
1751 * during an unstable branch.
1753 filp->f_flags |= O_LARGEFILE;
1755 filp->f_mode |= FMODE_NOWAIT;
1757 if (filp->f_flags & O_NDELAY)
1758 filp->f_mode |= FMODE_NDELAY;
1759 if (filp->f_flags & O_EXCL)
1760 filp->f_mode |= FMODE_EXCL;
1761 if ((filp->f_flags & O_ACCMODE) == 3)
1762 filp->f_mode |= FMODE_WRITE_IOCTL;
1764 bdev = bd_acquire(inode);
1768 filp->f_mapping = bdev->bd_inode->i_mapping;
1769 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1771 return blkdev_get(bdev, filp->f_mode, filp);
1774 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1776 struct gendisk *disk = bdev->bd_disk;
1777 struct block_device *victim = NULL;
1779 mutex_lock_nested(&bdev->bd_mutex, for_part);
1781 bdev->bd_part_count--;
1783 if (!--bdev->bd_openers) {
1784 WARN_ON_ONCE(bdev->bd_holders);
1785 sync_blockdev(bdev);
1788 bdev_write_inode(bdev);
1790 if (bdev->bd_contains == bdev) {
1791 if (disk->fops->release)
1792 disk->fops->release(disk, mode);
1794 if (!bdev->bd_openers) {
1795 disk_put_part(bdev->bd_part);
1796 bdev->bd_part = NULL;
1797 bdev->bd_disk = NULL;
1798 if (bdev != bdev->bd_contains)
1799 victim = bdev->bd_contains;
1800 bdev->bd_contains = NULL;
1802 put_disk_and_module(disk);
1804 mutex_unlock(&bdev->bd_mutex);
1807 __blkdev_put(victim, mode, 1);
1810 void blkdev_put(struct block_device *bdev, fmode_t mode)
1812 mutex_lock(&bdev->bd_mutex);
1814 if (mode & FMODE_EXCL) {
1818 * Release a claim on the device. The holder fields
1819 * are protected with bdev_lock. bd_mutex is to
1820 * synchronize disk_holder unlinking.
1822 spin_lock(&bdev_lock);
1824 WARN_ON_ONCE(--bdev->bd_holders < 0);
1825 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1827 /* bd_contains might point to self, check in a separate step */
1828 if ((bdev_free = !bdev->bd_holders))
1829 bdev->bd_holder = NULL;
1830 if (!bdev->bd_contains->bd_holders)
1831 bdev->bd_contains->bd_holder = NULL;
1833 spin_unlock(&bdev_lock);
1836 * If this was the last claim, remove holder link and
1837 * unblock evpoll if it was a write holder.
1839 if (bdev_free && bdev->bd_write_holder) {
1840 disk_unblock_events(bdev->bd_disk);
1841 bdev->bd_write_holder = false;
1846 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1847 * event. This is to ensure detection of media removal commanded
1848 * from userland - e.g. eject(1).
1850 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1852 mutex_unlock(&bdev->bd_mutex);
1854 __blkdev_put(bdev, mode, 0);
1856 EXPORT_SYMBOL(blkdev_put);
1858 static int blkdev_close(struct inode * inode, struct file * filp)
1860 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1861 blkdev_put(bdev, filp->f_mode);
1865 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1867 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1868 fmode_t mode = file->f_mode;
1871 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1872 * to updated it before every ioctl.
1874 if (file->f_flags & O_NDELAY)
1875 mode |= FMODE_NDELAY;
1877 mode &= ~FMODE_NDELAY;
1879 return blkdev_ioctl(bdev, mode, cmd, arg);
1883 * Write data to the block device. Only intended for the block device itself
1884 * and the raw driver which basically is a fake block device.
1886 * Does not take i_mutex for the write and thus is not for general purpose
1889 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1891 struct file *file = iocb->ki_filp;
1892 struct inode *bd_inode = bdev_file_inode(file);
1893 loff_t size = i_size_read(bd_inode);
1894 struct blk_plug plug;
1897 if (bdev_read_only(I_BDEV(bd_inode)))
1900 if (!iov_iter_count(from))
1903 if (iocb->ki_pos >= size)
1906 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1909 iov_iter_truncate(from, size - iocb->ki_pos);
1911 blk_start_plug(&plug);
1912 ret = __generic_file_write_iter(iocb, from);
1914 ret = generic_write_sync(iocb, ret);
1915 blk_finish_plug(&plug);
1918 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1920 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1922 struct file *file = iocb->ki_filp;
1923 struct inode *bd_inode = bdev_file_inode(file);
1924 loff_t size = i_size_read(bd_inode);
1925 loff_t pos = iocb->ki_pos;
1931 iov_iter_truncate(to, size);
1932 return generic_file_read_iter(iocb, to);
1934 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1937 * Try to release a page associated with block device when the system
1938 * is under memory pressure.
1940 static int blkdev_releasepage(struct page *page, gfp_t wait)
1942 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1944 if (super && super->s_op->bdev_try_to_free_page)
1945 return super->s_op->bdev_try_to_free_page(super, page, wait);
1947 return try_to_free_buffers(page);
1950 static int blkdev_writepages(struct address_space *mapping,
1951 struct writeback_control *wbc)
1953 return generic_writepages(mapping, wbc);
1956 static const struct address_space_operations def_blk_aops = {
1957 .readpage = blkdev_readpage,
1958 .readpages = blkdev_readpages,
1959 .writepage = blkdev_writepage,
1960 .write_begin = blkdev_write_begin,
1961 .write_end = blkdev_write_end,
1962 .writepages = blkdev_writepages,
1963 .releasepage = blkdev_releasepage,
1964 .direct_IO = blkdev_direct_IO,
1965 .is_dirty_writeback = buffer_check_dirty_writeback,
1968 #define BLKDEV_FALLOC_FL_SUPPORTED \
1969 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
1970 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
1972 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
1975 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1976 struct address_space *mapping;
1977 loff_t end = start + len - 1;
1981 /* Fail if we don't recognize the flags. */
1982 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
1985 /* Don't go off the end of the device. */
1986 isize = i_size_read(bdev->bd_inode);
1990 if (mode & FALLOC_FL_KEEP_SIZE) {
1991 len = isize - start;
1992 end = start + len - 1;
1998 * Don't allow IO that isn't aligned to logical block size.
2000 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2003 /* Invalidate the page cache, including dirty pages. */
2004 mapping = bdev->bd_inode->i_mapping;
2005 truncate_inode_pages_range(mapping, start, end);
2008 case FALLOC_FL_ZERO_RANGE:
2009 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2010 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2011 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2013 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2014 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2015 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2017 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2018 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2028 * Invalidate again; if someone wandered in and dirtied a page,
2029 * the caller will be given -EBUSY. The third argument is
2030 * inclusive, so the rounding here is safe.
2032 return invalidate_inode_pages2_range(mapping,
2033 start >> PAGE_SHIFT,
2037 const struct file_operations def_blk_fops = {
2038 .open = blkdev_open,
2039 .release = blkdev_close,
2040 .llseek = block_llseek,
2041 .read_iter = blkdev_read_iter,
2042 .write_iter = blkdev_write_iter,
2043 .mmap = generic_file_mmap,
2044 .fsync = blkdev_fsync,
2045 .unlocked_ioctl = block_ioctl,
2046 #ifdef CONFIG_COMPAT
2047 .compat_ioctl = compat_blkdev_ioctl,
2049 .splice_read = generic_file_splice_read,
2050 .splice_write = iter_file_splice_write,
2051 .fallocate = blkdev_fallocate,
2054 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2057 mm_segment_t old_fs = get_fs();
2059 res = blkdev_ioctl(bdev, 0, cmd, arg);
2064 EXPORT_SYMBOL(ioctl_by_bdev);
2067 * lookup_bdev - lookup a struct block_device by name
2068 * @pathname: special file representing the block device
2070 * Get a reference to the blockdevice at @pathname in the current
2071 * namespace if possible and return it. Return ERR_PTR(error)
2074 struct block_device *lookup_bdev(const char *pathname)
2076 struct block_device *bdev;
2077 struct inode *inode;
2081 if (!pathname || !*pathname)
2082 return ERR_PTR(-EINVAL);
2084 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2086 return ERR_PTR(error);
2088 inode = d_backing_inode(path.dentry);
2090 if (!S_ISBLK(inode->i_mode))
2093 if (!may_open_dev(&path))
2096 bdev = bd_acquire(inode);
2103 bdev = ERR_PTR(error);
2106 EXPORT_SYMBOL(lookup_bdev);
2108 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2110 struct super_block *sb = get_super(bdev);
2115 * no need to lock the super, get_super holds the
2116 * read mutex so the filesystem cannot go away
2117 * under us (->put_super runs with the write lock
2120 shrink_dcache_sb(sb);
2121 res = invalidate_inodes(sb, kill_dirty);
2124 invalidate_bdev(bdev);
2127 EXPORT_SYMBOL(__invalidate_device);
2129 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2131 struct inode *inode, *old_inode = NULL;
2133 spin_lock(&blockdev_superblock->s_inode_list_lock);
2134 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2135 struct address_space *mapping = inode->i_mapping;
2136 struct block_device *bdev;
2138 spin_lock(&inode->i_lock);
2139 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2140 mapping->nrpages == 0) {
2141 spin_unlock(&inode->i_lock);
2145 spin_unlock(&inode->i_lock);
2146 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2148 * We hold a reference to 'inode' so it couldn't have been
2149 * removed from s_inodes list while we dropped the
2150 * s_inode_list_lock We cannot iput the inode now as we can
2151 * be holding the last reference and we cannot iput it under
2152 * s_inode_list_lock. So we keep the reference and iput it
2157 bdev = I_BDEV(inode);
2159 mutex_lock(&bdev->bd_mutex);
2160 if (bdev->bd_openers)
2162 mutex_unlock(&bdev->bd_mutex);
2164 spin_lock(&blockdev_superblock->s_inode_list_lock);
2166 spin_unlock(&blockdev_superblock->s_inode_list_lock);