1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
9 #include <linux/init.h>
11 #include <linux/fcntl.h>
12 #include <linux/slab.h>
13 #include <linux/kmod.h>
14 #include <linux/major.h>
15 #include <linux/device_cgroup.h>
16 #include <linux/highmem.h>
17 #include <linux/blkdev.h>
18 #include <linux/backing-dev.h>
19 #include <linux/module.h>
20 #include <linux/blkpg.h>
21 #include <linux/magic.h>
22 #include <linux/dax.h>
23 #include <linux/buffer_head.h>
24 #include <linux/swap.h>
25 #include <linux/pagevec.h>
26 #include <linux/writeback.h>
27 #include <linux/mpage.h>
28 #include <linux/mount.h>
29 #include <linux/pseudo_fs.h>
30 #include <linux/uio.h>
31 #include <linux/namei.h>
32 #include <linux/log2.h>
33 #include <linux/cleancache.h>
34 #include <linux/task_io_accounting_ops.h>
35 #include <linux/falloc.h>
36 #include <linux/uaccess.h>
40 struct block_device bdev;
41 struct inode vfs_inode;
44 static const struct address_space_operations def_blk_aops;
46 static inline struct bdev_inode *BDEV_I(struct inode *inode)
48 return container_of(inode, struct bdev_inode, vfs_inode);
51 struct block_device *I_BDEV(struct inode *inode)
53 return &BDEV_I(inode)->bdev;
55 EXPORT_SYMBOL(I_BDEV);
57 static void bdev_write_inode(struct block_device *bdev)
59 struct inode *inode = bdev->bd_inode;
62 spin_lock(&inode->i_lock);
63 while (inode->i_state & I_DIRTY) {
64 spin_unlock(&inode->i_lock);
65 ret = write_inode_now(inode, true);
67 char name[BDEVNAME_SIZE];
68 pr_warn_ratelimited("VFS: Dirty inode writeback failed "
69 "for block device %s (err=%d).\n",
70 bdevname(bdev, name), ret);
72 spin_lock(&inode->i_lock);
74 spin_unlock(&inode->i_lock);
77 /* Kill _all_ buffers and pagecache , dirty or not.. */
78 void kill_bdev(struct block_device *bdev)
80 struct address_space *mapping = bdev->bd_inode->i_mapping;
82 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
86 truncate_inode_pages(mapping, 0);
88 EXPORT_SYMBOL(kill_bdev);
90 /* Invalidate clean unused buffers and pagecache. */
91 void invalidate_bdev(struct block_device *bdev)
93 struct address_space *mapping = bdev->bd_inode->i_mapping;
95 if (mapping->nrpages) {
97 lru_add_drain_all(); /* make sure all lru add caches are flushed */
98 invalidate_mapping_pages(mapping, 0, -1);
100 /* 99% of the time, we don't need to flush the cleancache on the bdev.
101 * But, for the strange corners, lets be cautious
103 cleancache_invalidate_inode(mapping);
105 EXPORT_SYMBOL(invalidate_bdev);
107 static void set_init_blocksize(struct block_device *bdev)
109 unsigned bsize = bdev_logical_block_size(bdev);
110 loff_t size = i_size_read(bdev->bd_inode);
112 while (bsize < PAGE_SIZE) {
117 bdev->bd_block_size = bsize;
118 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
121 int set_blocksize(struct block_device *bdev, int size)
123 /* Size must be a power of two, and between 512 and PAGE_SIZE */
124 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
127 /* Size cannot be smaller than the size supported by the device */
128 if (size < bdev_logical_block_size(bdev))
131 /* Don't change the size if it is same as current */
132 if (bdev->bd_block_size != size) {
134 bdev->bd_block_size = size;
135 bdev->bd_inode->i_blkbits = blksize_bits(size);
141 EXPORT_SYMBOL(set_blocksize);
143 int sb_set_blocksize(struct super_block *sb, int size)
145 if (set_blocksize(sb->s_bdev, size))
147 /* If we get here, we know size is power of two
148 * and it's value is between 512 and PAGE_SIZE */
149 sb->s_blocksize = size;
150 sb->s_blocksize_bits = blksize_bits(size);
151 return sb->s_blocksize;
154 EXPORT_SYMBOL(sb_set_blocksize);
156 int sb_min_blocksize(struct super_block *sb, int size)
158 int minsize = bdev_logical_block_size(sb->s_bdev);
161 return sb_set_blocksize(sb, size);
164 EXPORT_SYMBOL(sb_min_blocksize);
167 blkdev_get_block(struct inode *inode, sector_t iblock,
168 struct buffer_head *bh, int create)
170 bh->b_bdev = I_BDEV(inode);
171 bh->b_blocknr = iblock;
172 set_buffer_mapped(bh);
176 static struct inode *bdev_file_inode(struct file *file)
178 return file->f_mapping->host;
181 static unsigned int dio_bio_write_op(struct kiocb *iocb)
183 unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
185 /* avoid the need for a I/O completion work item */
186 if (iocb->ki_flags & IOCB_DSYNC)
191 #define DIO_INLINE_BIO_VECS 4
193 static void blkdev_bio_end_io_simple(struct bio *bio)
195 struct task_struct *waiter = bio->bi_private;
197 WRITE_ONCE(bio->bi_private, NULL);
198 blk_wake_io_task(waiter);
202 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
205 struct file *file = iocb->ki_filp;
206 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
207 struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs;
208 loff_t pos = iocb->ki_pos;
209 bool should_dirty = false;
214 if ((pos | iov_iter_alignment(iter)) &
215 (bdev_logical_block_size(bdev) - 1))
218 if (nr_pages <= DIO_INLINE_BIO_VECS)
221 vecs = kmalloc_array(nr_pages, sizeof(struct bio_vec),
227 bio_init(&bio, vecs, nr_pages);
228 bio_set_dev(&bio, bdev);
229 bio.bi_iter.bi_sector = pos >> 9;
230 bio.bi_write_hint = iocb->ki_hint;
231 bio.bi_private = current;
232 bio.bi_end_io = blkdev_bio_end_io_simple;
233 bio.bi_ioprio = iocb->ki_ioprio;
235 ret = bio_iov_iter_get_pages(&bio, iter);
238 ret = bio.bi_iter.bi_size;
240 if (iov_iter_rw(iter) == READ) {
241 bio.bi_opf = REQ_OP_READ;
242 if (iter_is_iovec(iter))
245 bio.bi_opf = dio_bio_write_op(iocb);
246 task_io_account_write(ret);
248 if (iocb->ki_flags & IOCB_HIPRI)
249 bio_set_polled(&bio, iocb);
251 qc = submit_bio(&bio);
253 set_current_state(TASK_UNINTERRUPTIBLE);
254 if (!READ_ONCE(bio.bi_private))
256 if (!(iocb->ki_flags & IOCB_HIPRI) ||
257 !blk_poll(bdev_get_queue(bdev), qc, true))
260 __set_current_state(TASK_RUNNING);
262 bio_release_pages(&bio, should_dirty);
263 if (unlikely(bio.bi_status))
264 ret = blk_status_to_errno(bio.bi_status);
267 if (vecs != inline_vecs)
278 struct task_struct *waiter;
283 bool should_dirty : 1;
288 static struct bio_set blkdev_dio_pool;
290 static int blkdev_iopoll(struct kiocb *kiocb, bool wait)
292 struct block_device *bdev = I_BDEV(kiocb->ki_filp->f_mapping->host);
293 struct request_queue *q = bdev_get_queue(bdev);
295 return blk_poll(q, READ_ONCE(kiocb->ki_cookie), wait);
298 static void blkdev_bio_end_io(struct bio *bio)
300 struct blkdev_dio *dio = bio->bi_private;
301 bool should_dirty = dio->should_dirty;
303 if (bio->bi_status && !dio->bio.bi_status)
304 dio->bio.bi_status = bio->bi_status;
306 if (!dio->multi_bio || atomic_dec_and_test(&dio->ref)) {
308 struct kiocb *iocb = dio->iocb;
311 if (likely(!dio->bio.bi_status)) {
315 ret = blk_status_to_errno(dio->bio.bi_status);
318 dio->iocb->ki_complete(iocb, ret, 0);
322 struct task_struct *waiter = dio->waiter;
324 WRITE_ONCE(dio->waiter, NULL);
325 blk_wake_io_task(waiter);
330 bio_check_pages_dirty(bio);
332 bio_release_pages(bio, false);
338 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
340 struct file *file = iocb->ki_filp;
341 struct inode *inode = bdev_file_inode(file);
342 struct block_device *bdev = I_BDEV(inode);
343 struct blk_plug plug;
344 struct blkdev_dio *dio;
346 bool is_poll = (iocb->ki_flags & IOCB_HIPRI) != 0;
347 bool is_read = (iov_iter_rw(iter) == READ), is_sync;
348 bool nowait = (iocb->ki_flags & IOCB_NOWAIT) != 0;
349 loff_t pos = iocb->ki_pos;
350 blk_qc_t qc = BLK_QC_T_NONE;
354 if ((pos | iov_iter_alignment(iter)) &
355 (bdev_logical_block_size(bdev) - 1))
363 bio = bio_alloc_bioset(gfp, nr_pages, &blkdev_dio_pool);
367 dio = container_of(bio, struct blkdev_dio, bio);
368 dio->is_sync = is_sync = is_sync_kiocb(iocb);
370 dio->waiter = current;
377 dio->multi_bio = false;
378 dio->should_dirty = is_read && iter_is_iovec(iter);
381 * Don't plug for HIPRI/polled IO, as those should go straight
385 blk_start_plug(&plug);
391 bio_set_dev(bio, bdev);
392 bio->bi_iter.bi_sector = pos >> 9;
393 bio->bi_write_hint = iocb->ki_hint;
394 bio->bi_private = dio;
395 bio->bi_end_io = blkdev_bio_end_io;
396 bio->bi_ioprio = iocb->ki_ioprio;
398 err = bio_iov_iter_get_pages(bio, iter);
402 bio->bi_status = BLK_STS_IOERR;
408 bio->bi_opf = REQ_OP_READ;
409 if (dio->should_dirty)
410 bio_set_pages_dirty(bio);
412 bio->bi_opf = dio_bio_write_op(iocb);
413 task_io_account_write(bio->bi_iter.bi_size);
417 * Tell underlying layer to not block for resource shortage.
418 * And if we would have blocked, return error inline instead
419 * of through the bio->bi_end_io() callback.
422 bio->bi_opf |= (REQ_NOWAIT | REQ_NOWAIT_INLINE);
424 dio->size += bio->bi_iter.bi_size;
425 pos += bio->bi_iter.bi_size;
427 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
431 if (iocb->ki_flags & IOCB_HIPRI) {
432 bio_set_polled(bio, iocb);
436 qc = submit_bio(bio);
437 if (qc == BLK_QC_T_EAGAIN) {
444 WRITE_ONCE(iocb->ki_cookie, qc);
448 if (!dio->multi_bio) {
450 * AIO needs an extra reference to ensure the dio
451 * structure which is embedded into the first bio
456 dio->multi_bio = true;
457 atomic_set(&dio->ref, 2);
459 atomic_inc(&dio->ref);
462 qc = submit_bio(bio);
463 if (qc == BLK_QC_T_EAGAIN) {
468 ret += bio->bi_iter.bi_size;
470 bio = bio_alloc(gfp, nr_pages);
479 blk_finish_plug(&plug);
485 set_current_state(TASK_UNINTERRUPTIBLE);
486 if (!READ_ONCE(dio->waiter))
489 if (!(iocb->ki_flags & IOCB_HIPRI) ||
490 !blk_poll(bdev_get_queue(bdev), qc, true))
493 __set_current_state(TASK_RUNNING);
497 ret = blk_status_to_errno(dio->bio.bi_status);
503 blk_finish_plug(&plug);
508 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
512 nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
515 if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
516 return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
518 return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
521 static __init int blkdev_init(void)
523 return bioset_init(&blkdev_dio_pool, 4, offsetof(struct blkdev_dio, bio), BIOSET_NEED_BVECS);
525 module_init(blkdev_init);
527 int __sync_blockdev(struct block_device *bdev, int wait)
532 return filemap_flush(bdev->bd_inode->i_mapping);
533 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
537 * Write out and wait upon all the dirty data associated with a block
538 * device via its mapping. Does not take the superblock lock.
540 int sync_blockdev(struct block_device *bdev)
542 return __sync_blockdev(bdev, 1);
544 EXPORT_SYMBOL(sync_blockdev);
547 * Write out and wait upon all dirty data associated with this
548 * device. Filesystem data as well as the underlying block
549 * device. Takes the superblock lock.
551 int fsync_bdev(struct block_device *bdev)
553 struct super_block *sb = get_super(bdev);
555 int res = sync_filesystem(sb);
559 return sync_blockdev(bdev);
561 EXPORT_SYMBOL(fsync_bdev);
564 * freeze_bdev -- lock a filesystem and force it into a consistent state
565 * @bdev: blockdevice to lock
567 * If a superblock is found on this device, we take the s_umount semaphore
568 * on it to make sure nobody unmounts until the snapshot creation is done.
569 * The reference counter (bd_fsfreeze_count) guarantees that only the last
570 * unfreeze process can unfreeze the frozen filesystem actually when multiple
571 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
572 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
575 struct super_block *freeze_bdev(struct block_device *bdev)
577 struct super_block *sb;
580 mutex_lock(&bdev->bd_fsfreeze_mutex);
581 if (++bdev->bd_fsfreeze_count > 1) {
583 * We don't even need to grab a reference - the first call
584 * to freeze_bdev grab an active reference and only the last
585 * thaw_bdev drops it.
587 sb = get_super(bdev);
590 mutex_unlock(&bdev->bd_fsfreeze_mutex);
594 sb = get_active_super(bdev);
597 if (sb->s_op->freeze_super)
598 error = sb->s_op->freeze_super(sb);
600 error = freeze_super(sb);
602 deactivate_super(sb);
603 bdev->bd_fsfreeze_count--;
604 mutex_unlock(&bdev->bd_fsfreeze_mutex);
605 return ERR_PTR(error);
607 deactivate_super(sb);
610 mutex_unlock(&bdev->bd_fsfreeze_mutex);
611 return sb; /* thaw_bdev releases s->s_umount */
613 EXPORT_SYMBOL(freeze_bdev);
616 * thaw_bdev -- unlock filesystem
617 * @bdev: blockdevice to unlock
618 * @sb: associated superblock
620 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
622 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
626 mutex_lock(&bdev->bd_fsfreeze_mutex);
627 if (!bdev->bd_fsfreeze_count)
631 if (--bdev->bd_fsfreeze_count > 0)
637 if (sb->s_op->thaw_super)
638 error = sb->s_op->thaw_super(sb);
640 error = thaw_super(sb);
642 bdev->bd_fsfreeze_count++;
644 mutex_unlock(&bdev->bd_fsfreeze_mutex);
647 EXPORT_SYMBOL(thaw_bdev);
649 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
651 return block_write_full_page(page, blkdev_get_block, wbc);
654 static int blkdev_readpage(struct file * file, struct page * page)
656 return block_read_full_page(page, blkdev_get_block);
659 static int blkdev_readpages(struct file *file, struct address_space *mapping,
660 struct list_head *pages, unsigned nr_pages)
662 return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
665 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
666 loff_t pos, unsigned len, unsigned flags,
667 struct page **pagep, void **fsdata)
669 return block_write_begin(mapping, pos, len, flags, pagep,
673 static int blkdev_write_end(struct file *file, struct address_space *mapping,
674 loff_t pos, unsigned len, unsigned copied,
675 struct page *page, void *fsdata)
678 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
688 * for a block special file file_inode(file)->i_size is zero
689 * so we compute the size by hand (just as in block_read/write above)
691 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
693 struct inode *bd_inode = bdev_file_inode(file);
696 inode_lock(bd_inode);
697 retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
698 inode_unlock(bd_inode);
702 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
704 struct inode *bd_inode = bdev_file_inode(filp);
705 struct block_device *bdev = I_BDEV(bd_inode);
708 error = file_write_and_wait_range(filp, start, end);
713 * There is no need to serialise calls to blkdev_issue_flush with
714 * i_mutex and doing so causes performance issues with concurrent
715 * O_SYNC writers to a block device.
717 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
718 if (error == -EOPNOTSUPP)
723 EXPORT_SYMBOL(blkdev_fsync);
726 * bdev_read_page() - Start reading a page from a block device
727 * @bdev: The device to read the page from
728 * @sector: The offset on the device to read the page to (need not be aligned)
729 * @page: The page to read
731 * On entry, the page should be locked. It will be unlocked when the page
732 * has been read. If the block driver implements rw_page synchronously,
733 * that will be true on exit from this function, but it need not be.
735 * Errors returned by this function are usually "soft", eg out of memory, or
736 * queue full; callers should try a different route to read this page rather
737 * than propagate an error back up the stack.
739 * Return: negative errno if an error occurs, 0 if submission was successful.
741 int bdev_read_page(struct block_device *bdev, sector_t sector,
744 const struct block_device_operations *ops = bdev->bd_disk->fops;
745 int result = -EOPNOTSUPP;
747 if (!ops->rw_page || bdev_get_integrity(bdev))
750 result = blk_queue_enter(bdev->bd_queue, 0);
753 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
755 blk_queue_exit(bdev->bd_queue);
758 EXPORT_SYMBOL_GPL(bdev_read_page);
761 * bdev_write_page() - Start writing a page to a block device
762 * @bdev: The device to write the page to
763 * @sector: The offset on the device to write the page to (need not be aligned)
764 * @page: The page to write
765 * @wbc: The writeback_control for the write
767 * On entry, the page should be locked and not currently under writeback.
768 * On exit, if the write started successfully, the page will be unlocked and
769 * under writeback. If the write failed already (eg the driver failed to
770 * queue the page to the device), the page will still be locked. If the
771 * caller is a ->writepage implementation, it will need to unlock the page.
773 * Errors returned by this function are usually "soft", eg out of memory, or
774 * queue full; callers should try a different route to write this page rather
775 * than propagate an error back up the stack.
777 * Return: negative errno if an error occurs, 0 if submission was successful.
779 int bdev_write_page(struct block_device *bdev, sector_t sector,
780 struct page *page, struct writeback_control *wbc)
783 const struct block_device_operations *ops = bdev->bd_disk->fops;
785 if (!ops->rw_page || bdev_get_integrity(bdev))
787 result = blk_queue_enter(bdev->bd_queue, 0);
791 set_page_writeback(page);
792 result = ops->rw_page(bdev, sector + get_start_sect(bdev), page,
795 end_page_writeback(page);
797 clean_page_buffers(page);
800 blk_queue_exit(bdev->bd_queue);
803 EXPORT_SYMBOL_GPL(bdev_write_page);
809 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
810 static struct kmem_cache * bdev_cachep __read_mostly;
812 static struct inode *bdev_alloc_inode(struct super_block *sb)
814 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
817 return &ei->vfs_inode;
820 static void bdev_free_inode(struct inode *inode)
822 kmem_cache_free(bdev_cachep, BDEV_I(inode));
825 static void init_once(void *foo)
827 struct bdev_inode *ei = (struct bdev_inode *) foo;
828 struct block_device *bdev = &ei->bdev;
830 memset(bdev, 0, sizeof(*bdev));
831 mutex_init(&bdev->bd_mutex);
832 INIT_LIST_HEAD(&bdev->bd_list);
834 INIT_LIST_HEAD(&bdev->bd_holder_disks);
836 bdev->bd_bdi = &noop_backing_dev_info;
837 inode_init_once(&ei->vfs_inode);
838 /* Initialize mutex for freeze. */
839 mutex_init(&bdev->bd_fsfreeze_mutex);
842 static void bdev_evict_inode(struct inode *inode)
844 struct block_device *bdev = &BDEV_I(inode)->bdev;
845 truncate_inode_pages_final(&inode->i_data);
846 invalidate_inode_buffers(inode); /* is it needed here? */
848 spin_lock(&bdev_lock);
849 list_del_init(&bdev->bd_list);
850 spin_unlock(&bdev_lock);
851 /* Detach inode from wb early as bdi_put() may free bdi->wb */
852 inode_detach_wb(inode);
853 if (bdev->bd_bdi != &noop_backing_dev_info) {
854 bdi_put(bdev->bd_bdi);
855 bdev->bd_bdi = &noop_backing_dev_info;
859 static const struct super_operations bdev_sops = {
860 .statfs = simple_statfs,
861 .alloc_inode = bdev_alloc_inode,
862 .free_inode = bdev_free_inode,
863 .drop_inode = generic_delete_inode,
864 .evict_inode = bdev_evict_inode,
867 static int bd_init_fs_context(struct fs_context *fc)
869 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC);
872 fc->s_iflags |= SB_I_CGROUPWB;
873 ctx->ops = &bdev_sops;
877 static struct file_system_type bd_type = {
879 .init_fs_context = bd_init_fs_context,
880 .kill_sb = kill_anon_super,
883 struct super_block *blockdev_superblock __read_mostly;
884 EXPORT_SYMBOL_GPL(blockdev_superblock);
886 void __init bdev_cache_init(void)
889 static struct vfsmount *bd_mnt;
891 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
892 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
893 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
895 err = register_filesystem(&bd_type);
897 panic("Cannot register bdev pseudo-fs");
898 bd_mnt = kern_mount(&bd_type);
900 panic("Cannot create bdev pseudo-fs");
901 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
905 * Most likely _very_ bad one - but then it's hardly critical for small
906 * /dev and can be fixed when somebody will need really large one.
907 * Keep in mind that it will be fed through icache hash function too.
909 static inline unsigned long hash(dev_t dev)
911 return MAJOR(dev)+MINOR(dev);
914 static int bdev_test(struct inode *inode, void *data)
916 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
919 static int bdev_set(struct inode *inode, void *data)
921 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
925 static LIST_HEAD(all_bdevs);
928 * If there is a bdev inode for this device, unhash it so that it gets evicted
929 * as soon as last inode reference is dropped.
931 void bdev_unhash_inode(dev_t dev)
935 inode = ilookup5(blockdev_superblock, hash(dev), bdev_test, &dev);
937 remove_inode_hash(inode);
942 struct block_device *bdget(dev_t dev)
944 struct block_device *bdev;
947 inode = iget5_locked(blockdev_superblock, hash(dev),
948 bdev_test, bdev_set, &dev);
953 bdev = &BDEV_I(inode)->bdev;
955 if (inode->i_state & I_NEW) {
956 bdev->bd_contains = NULL;
957 bdev->bd_super = NULL;
958 bdev->bd_inode = inode;
959 bdev->bd_block_size = i_blocksize(inode);
960 bdev->bd_part_count = 0;
961 bdev->bd_invalidated = 0;
962 inode->i_mode = S_IFBLK;
964 inode->i_bdev = bdev;
965 inode->i_data.a_ops = &def_blk_aops;
966 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
967 spin_lock(&bdev_lock);
968 list_add(&bdev->bd_list, &all_bdevs);
969 spin_unlock(&bdev_lock);
970 unlock_new_inode(inode);
975 EXPORT_SYMBOL(bdget);
978 * bdgrab -- Grab a reference to an already referenced block device
979 * @bdev: Block device to grab a reference to.
981 struct block_device *bdgrab(struct block_device *bdev)
983 ihold(bdev->bd_inode);
986 EXPORT_SYMBOL(bdgrab);
988 long nr_blockdev_pages(void)
990 struct block_device *bdev;
992 spin_lock(&bdev_lock);
993 list_for_each_entry(bdev, &all_bdevs, bd_list) {
994 ret += bdev->bd_inode->i_mapping->nrpages;
996 spin_unlock(&bdev_lock);
1000 void bdput(struct block_device *bdev)
1002 iput(bdev->bd_inode);
1005 EXPORT_SYMBOL(bdput);
1007 static struct block_device *bd_acquire(struct inode *inode)
1009 struct block_device *bdev;
1011 spin_lock(&bdev_lock);
1012 bdev = inode->i_bdev;
1013 if (bdev && !inode_unhashed(bdev->bd_inode)) {
1015 spin_unlock(&bdev_lock);
1018 spin_unlock(&bdev_lock);
1021 * i_bdev references block device inode that was already shut down
1022 * (corresponding device got removed). Remove the reference and look
1023 * up block device inode again just in case new device got
1024 * reestablished under the same device number.
1029 bdev = bdget(inode->i_rdev);
1031 spin_lock(&bdev_lock);
1032 if (!inode->i_bdev) {
1034 * We take an additional reference to bd_inode,
1035 * and it's released in clear_inode() of inode.
1036 * So, we can access it via ->i_mapping always
1040 inode->i_bdev = bdev;
1041 inode->i_mapping = bdev->bd_inode->i_mapping;
1043 spin_unlock(&bdev_lock);
1048 /* Call when you free inode */
1050 void bd_forget(struct inode *inode)
1052 struct block_device *bdev = NULL;
1054 spin_lock(&bdev_lock);
1055 if (!sb_is_blkdev_sb(inode->i_sb))
1056 bdev = inode->i_bdev;
1057 inode->i_bdev = NULL;
1058 inode->i_mapping = &inode->i_data;
1059 spin_unlock(&bdev_lock);
1066 * bd_may_claim - test whether a block device can be claimed
1067 * @bdev: block device of interest
1068 * @whole: whole block device containing @bdev, may equal @bdev
1069 * @holder: holder trying to claim @bdev
1071 * Test whether @bdev can be claimed by @holder.
1074 * spin_lock(&bdev_lock).
1077 * %true if @bdev can be claimed, %false otherwise.
1079 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1082 if (bdev->bd_holder == holder)
1083 return true; /* already a holder */
1084 else if (bdev->bd_holder != NULL)
1085 return false; /* held by someone else */
1086 else if (whole == bdev)
1087 return true; /* is a whole device which isn't held */
1089 else if (whole->bd_holder == bd_may_claim)
1090 return true; /* is a partition of a device that is being partitioned */
1091 else if (whole->bd_holder != NULL)
1092 return false; /* is a partition of a held device */
1094 return true; /* is a partition of an un-held device */
1098 * bd_prepare_to_claim - prepare to claim a block device
1099 * @bdev: block device of interest
1100 * @whole: the whole device containing @bdev, may equal @bdev
1101 * @holder: holder trying to claim @bdev
1103 * Prepare to claim @bdev. This function fails if @bdev is already
1104 * claimed by another holder and waits if another claiming is in
1105 * progress. This function doesn't actually claim. On successful
1106 * return, the caller has ownership of bd_claiming and bd_holder[s].
1109 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
1110 * it multiple times.
1113 * 0 if @bdev can be claimed, -EBUSY otherwise.
1115 static int bd_prepare_to_claim(struct block_device *bdev,
1116 struct block_device *whole, void *holder)
1119 /* if someone else claimed, fail */
1120 if (!bd_may_claim(bdev, whole, holder))
1123 /* if claiming is already in progress, wait for it to finish */
1124 if (whole->bd_claiming) {
1125 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1128 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1129 spin_unlock(&bdev_lock);
1131 finish_wait(wq, &wait);
1132 spin_lock(&bdev_lock);
1140 static struct gendisk *bdev_get_gendisk(struct block_device *bdev, int *partno)
1142 struct gendisk *disk = get_gendisk(bdev->bd_dev, partno);
1147 * Now that we hold gendisk reference we make sure bdev we looked up is
1148 * not stale. If it is, it means device got removed and created before
1149 * we looked up gendisk and we fail open in such case. Associating
1150 * unhashed bdev with newly created gendisk could lead to two bdevs
1151 * (and thus two independent caches) being associated with one device
1154 if (inode_unhashed(bdev->bd_inode)) {
1155 put_disk_and_module(disk);
1162 * bd_start_claiming - start claiming a block device
1163 * @bdev: block device of interest
1164 * @holder: holder trying to claim @bdev
1166 * @bdev is about to be opened exclusively. Check @bdev can be opened
1167 * exclusively and mark that an exclusive open is in progress. Each
1168 * successful call to this function must be matched with a call to
1169 * either bd_finish_claiming() or bd_abort_claiming() (which do not
1172 * This function is used to gain exclusive access to the block device
1173 * without actually causing other exclusive open attempts to fail. It
1174 * should be used when the open sequence itself requires exclusive
1175 * access but may subsequently fail.
1181 * Pointer to the block device containing @bdev on success, ERR_PTR()
1184 static struct block_device *bd_start_claiming(struct block_device *bdev,
1187 struct gendisk *disk;
1188 struct block_device *whole;
1194 * @bdev might not have been initialized properly yet, look up
1195 * and grab the outer block device the hard way.
1197 disk = bdev_get_gendisk(bdev, &partno);
1199 return ERR_PTR(-ENXIO);
1202 * Normally, @bdev should equal what's returned from bdget_disk()
1203 * if partno is 0; however, some drivers (floppy) use multiple
1204 * bdev's for the same physical device and @bdev may be one of the
1205 * aliases. Keep @bdev if partno is 0. This means claimer
1206 * tracking is broken for those devices but it has always been that
1210 whole = bdget_disk(disk, 0);
1212 whole = bdgrab(bdev);
1214 put_disk_and_module(disk);
1216 return ERR_PTR(-ENOMEM);
1218 /* prepare to claim, if successful, mark claiming in progress */
1219 spin_lock(&bdev_lock);
1221 err = bd_prepare_to_claim(bdev, whole, holder);
1223 whole->bd_claiming = holder;
1224 spin_unlock(&bdev_lock);
1227 spin_unlock(&bdev_lock);
1229 return ERR_PTR(err);
1234 struct bd_holder_disk {
1235 struct list_head list;
1236 struct gendisk *disk;
1240 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1241 struct gendisk *disk)
1243 struct bd_holder_disk *holder;
1245 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1246 if (holder->disk == disk)
1251 static int add_symlink(struct kobject *from, struct kobject *to)
1253 return sysfs_create_link(from, to, kobject_name(to));
1256 static void del_symlink(struct kobject *from, struct kobject *to)
1258 sysfs_remove_link(from, kobject_name(to));
1262 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1263 * @bdev: the claimed slave bdev
1264 * @disk: the holding disk
1266 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1268 * This functions creates the following sysfs symlinks.
1270 * - from "slaves" directory of the holder @disk to the claimed @bdev
1271 * - from "holders" directory of the @bdev to the holder @disk
1273 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1274 * passed to bd_link_disk_holder(), then:
1276 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
1277 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1279 * The caller must have claimed @bdev before calling this function and
1280 * ensure that both @bdev and @disk are valid during the creation and
1281 * lifetime of these symlinks.
1287 * 0 on success, -errno on failure.
1289 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1291 struct bd_holder_disk *holder;
1294 mutex_lock(&bdev->bd_mutex);
1296 WARN_ON_ONCE(!bdev->bd_holder);
1298 /* FIXME: remove the following once add_disk() handles errors */
1299 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1302 holder = bd_find_holder_disk(bdev, disk);
1308 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1314 INIT_LIST_HEAD(&holder->list);
1315 holder->disk = disk;
1318 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1322 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1326 * bdev could be deleted beneath us which would implicitly destroy
1327 * the holder directory. Hold on to it.
1329 kobject_get(bdev->bd_part->holder_dir);
1331 list_add(&holder->list, &bdev->bd_holder_disks);
1335 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1339 mutex_unlock(&bdev->bd_mutex);
1342 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1345 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1346 * @bdev: the calimed slave bdev
1347 * @disk: the holding disk
1349 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1354 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1356 struct bd_holder_disk *holder;
1358 mutex_lock(&bdev->bd_mutex);
1360 holder = bd_find_holder_disk(bdev, disk);
1362 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1363 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1364 del_symlink(bdev->bd_part->holder_dir,
1365 &disk_to_dev(disk)->kobj);
1366 kobject_put(bdev->bd_part->holder_dir);
1367 list_del_init(&holder->list);
1371 mutex_unlock(&bdev->bd_mutex);
1373 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1377 * flush_disk - invalidates all buffer-cache entries on a disk
1379 * @bdev: struct block device to be flushed
1380 * @kill_dirty: flag to guide handling of dirty inodes
1382 * Invalidates all buffer-cache entries on a disk. It should be called
1383 * when a disk has been changed -- either by a media change or online
1386 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1388 if (__invalidate_device(bdev, kill_dirty)) {
1389 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1390 "resized disk %s\n",
1391 bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1396 if (disk_part_scan_enabled(bdev->bd_disk))
1397 bdev->bd_invalidated = 1;
1401 * check_disk_size_change - checks for disk size change and adjusts bdev size.
1402 * @disk: struct gendisk to check
1403 * @bdev: struct bdev to adjust.
1404 * @verbose: if %true log a message about a size change if there is any
1406 * This routine checks to see if the bdev size does not match the disk size
1407 * and adjusts it if it differs. When shrinking the bdev size, its all caches
1410 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev,
1413 loff_t disk_size, bdev_size;
1415 disk_size = (loff_t)get_capacity(disk) << 9;
1416 bdev_size = i_size_read(bdev->bd_inode);
1417 if (disk_size != bdev_size) {
1420 "%s: detected capacity change from %lld to %lld\n",
1421 disk->disk_name, bdev_size, disk_size);
1423 i_size_write(bdev->bd_inode, disk_size);
1424 if (bdev_size > disk_size)
1425 flush_disk(bdev, false);
1430 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1431 * @disk: struct gendisk to be revalidated
1433 * This routine is a wrapper for lower-level driver's revalidate_disk
1434 * call-backs. It is used to do common pre and post operations needed
1435 * for all revalidate_disk operations.
1437 int revalidate_disk(struct gendisk *disk)
1441 if (disk->fops->revalidate_disk)
1442 ret = disk->fops->revalidate_disk(disk);
1445 * Hidden disks don't have associated bdev so there's no point in
1448 if (!(disk->flags & GENHD_FL_HIDDEN)) {
1449 struct block_device *bdev = bdget_disk(disk, 0);
1454 mutex_lock(&bdev->bd_mutex);
1455 check_disk_size_change(disk, bdev, ret == 0);
1456 bdev->bd_invalidated = 0;
1457 mutex_unlock(&bdev->bd_mutex);
1462 EXPORT_SYMBOL(revalidate_disk);
1465 * This routine checks whether a removable media has been changed,
1466 * and invalidates all buffer-cache-entries in that case. This
1467 * is a relatively slow routine, so we have to try to minimize using
1468 * it. Thus it is called only upon a 'mount' or 'open'. This
1469 * is the best way of combining speed and utility, I think.
1470 * People changing diskettes in the middle of an operation deserve
1473 int check_disk_change(struct block_device *bdev)
1475 struct gendisk *disk = bdev->bd_disk;
1476 const struct block_device_operations *bdops = disk->fops;
1477 unsigned int events;
1479 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1480 DISK_EVENT_EJECT_REQUEST);
1481 if (!(events & DISK_EVENT_MEDIA_CHANGE))
1484 flush_disk(bdev, true);
1485 if (bdops->revalidate_disk)
1486 bdops->revalidate_disk(bdev->bd_disk);
1490 EXPORT_SYMBOL(check_disk_change);
1492 void bd_set_size(struct block_device *bdev, loff_t size)
1494 inode_lock(bdev->bd_inode);
1495 i_size_write(bdev->bd_inode, size);
1496 inode_unlock(bdev->bd_inode);
1498 EXPORT_SYMBOL(bd_set_size);
1500 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1505 * mutex_lock(part->bd_mutex)
1506 * mutex_lock_nested(whole->bd_mutex, 1)
1509 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1511 struct gendisk *disk;
1515 bool first_open = false;
1517 if (mode & FMODE_READ)
1519 if (mode & FMODE_WRITE)
1522 * hooks: /n/, see "layering violations".
1525 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1535 disk = bdev_get_gendisk(bdev, &partno);
1539 disk_block_events(disk);
1540 mutex_lock_nested(&bdev->bd_mutex, for_part);
1541 if (!bdev->bd_openers) {
1543 bdev->bd_disk = disk;
1544 bdev->bd_queue = disk->queue;
1545 bdev->bd_contains = bdev;
1546 bdev->bd_partno = partno;
1550 bdev->bd_part = disk_get_part(disk, partno);
1555 if (disk->fops->open) {
1556 ret = disk->fops->open(bdev, mode);
1557 if (ret == -ERESTARTSYS) {
1558 /* Lost a race with 'disk' being
1559 * deleted, try again.
1562 disk_put_part(bdev->bd_part);
1563 bdev->bd_part = NULL;
1564 bdev->bd_disk = NULL;
1565 bdev->bd_queue = NULL;
1566 mutex_unlock(&bdev->bd_mutex);
1567 disk_unblock_events(disk);
1568 put_disk_and_module(disk);
1574 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1575 set_init_blocksize(bdev);
1579 * If the device is invalidated, rescan partition
1580 * if open succeeded or failed with -ENOMEDIUM.
1581 * The latter is necessary to prevent ghost
1582 * partitions on a removed medium.
1584 if (bdev->bd_invalidated) {
1586 rescan_partitions(disk, bdev);
1587 else if (ret == -ENOMEDIUM)
1588 invalidate_partitions(disk, bdev);
1594 struct block_device *whole;
1595 whole = bdget_disk(disk, 0);
1600 ret = __blkdev_get(whole, mode, 1);
1603 bdev->bd_contains = whole;
1604 bdev->bd_part = disk_get_part(disk, partno);
1605 if (!(disk->flags & GENHD_FL_UP) ||
1606 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1610 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1611 set_init_blocksize(bdev);
1614 if (bdev->bd_bdi == &noop_backing_dev_info)
1615 bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
1617 if (bdev->bd_contains == bdev) {
1619 if (bdev->bd_disk->fops->open)
1620 ret = bdev->bd_disk->fops->open(bdev, mode);
1621 /* the same as first opener case, read comment there */
1622 if (bdev->bd_invalidated) {
1624 rescan_partitions(bdev->bd_disk, bdev);
1625 else if (ret == -ENOMEDIUM)
1626 invalidate_partitions(bdev->bd_disk, bdev);
1629 goto out_unlock_bdev;
1634 bdev->bd_part_count++;
1635 mutex_unlock(&bdev->bd_mutex);
1636 disk_unblock_events(disk);
1637 /* only one opener holds refs to the module and disk */
1639 put_disk_and_module(disk);
1643 disk_put_part(bdev->bd_part);
1644 bdev->bd_disk = NULL;
1645 bdev->bd_part = NULL;
1646 bdev->bd_queue = NULL;
1647 if (bdev != bdev->bd_contains)
1648 __blkdev_put(bdev->bd_contains, mode, 1);
1649 bdev->bd_contains = NULL;
1651 mutex_unlock(&bdev->bd_mutex);
1652 disk_unblock_events(disk);
1653 put_disk_and_module(disk);
1661 * blkdev_get - open a block device
1662 * @bdev: block_device to open
1663 * @mode: FMODE_* mask
1664 * @holder: exclusive holder identifier
1666 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1667 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1668 * @holder is invalid. Exclusive opens may nest for the same @holder.
1670 * On success, the reference count of @bdev is unchanged. On failure,
1677 * 0 on success, -errno on failure.
1679 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1681 struct block_device *whole = NULL;
1684 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1686 if ((mode & FMODE_EXCL) && holder) {
1687 whole = bd_start_claiming(bdev, holder);
1688 if (IS_ERR(whole)) {
1690 return PTR_ERR(whole);
1694 res = __blkdev_get(bdev, mode, 0);
1697 struct gendisk *disk = whole->bd_disk;
1699 /* finish claiming */
1700 mutex_lock(&bdev->bd_mutex);
1701 spin_lock(&bdev_lock);
1704 BUG_ON(!bd_may_claim(bdev, whole, holder));
1706 * Note that for a whole device bd_holders
1707 * will be incremented twice, and bd_holder
1708 * will be set to bd_may_claim before being
1711 whole->bd_holders++;
1712 whole->bd_holder = bd_may_claim;
1714 bdev->bd_holder = holder;
1717 /* tell others that we're done */
1718 BUG_ON(whole->bd_claiming != holder);
1719 whole->bd_claiming = NULL;
1720 wake_up_bit(&whole->bd_claiming, 0);
1722 spin_unlock(&bdev_lock);
1725 * Block event polling for write claims if requested. Any
1726 * write holder makes the write_holder state stick until
1727 * all are released. This is good enough and tracking
1728 * individual writeable reference is too fragile given the
1729 * way @mode is used in blkdev_get/put().
1731 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1732 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1733 bdev->bd_write_holder = true;
1734 disk_block_events(disk);
1737 mutex_unlock(&bdev->bd_mutex);
1743 EXPORT_SYMBOL(blkdev_get);
1746 * blkdev_get_by_path - open a block device by name
1747 * @path: path to the block device to open
1748 * @mode: FMODE_* mask
1749 * @holder: exclusive holder identifier
1751 * Open the blockdevice described by the device file at @path. @mode
1752 * and @holder are identical to blkdev_get().
1754 * On success, the returned block_device has reference count of one.
1760 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1762 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1765 struct block_device *bdev;
1768 bdev = lookup_bdev(path);
1772 err = blkdev_get(bdev, mode, holder);
1774 return ERR_PTR(err);
1776 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1777 blkdev_put(bdev, mode);
1778 return ERR_PTR(-EACCES);
1783 EXPORT_SYMBOL(blkdev_get_by_path);
1786 * blkdev_get_by_dev - open a block device by device number
1787 * @dev: device number of block device to open
1788 * @mode: FMODE_* mask
1789 * @holder: exclusive holder identifier
1791 * Open the blockdevice described by device number @dev. @mode and
1792 * @holder are identical to blkdev_get().
1794 * Use it ONLY if you really do not have anything better - i.e. when
1795 * you are behind a truly sucky interface and all you are given is a
1796 * device number. _Never_ to be used for internal purposes. If you
1797 * ever need it - reconsider your API.
1799 * On success, the returned block_device has reference count of one.
1805 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1807 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1809 struct block_device *bdev;
1814 return ERR_PTR(-ENOMEM);
1816 err = blkdev_get(bdev, mode, holder);
1818 return ERR_PTR(err);
1822 EXPORT_SYMBOL(blkdev_get_by_dev);
1824 static int blkdev_open(struct inode * inode, struct file * filp)
1826 struct block_device *bdev;
1829 * Preserve backwards compatibility and allow large file access
1830 * even if userspace doesn't ask for it explicitly. Some mkfs
1831 * binary needs it. We might want to drop this workaround
1832 * during an unstable branch.
1834 filp->f_flags |= O_LARGEFILE;
1836 filp->f_mode |= FMODE_NOWAIT;
1838 if (filp->f_flags & O_NDELAY)
1839 filp->f_mode |= FMODE_NDELAY;
1840 if (filp->f_flags & O_EXCL)
1841 filp->f_mode |= FMODE_EXCL;
1842 if ((filp->f_flags & O_ACCMODE) == 3)
1843 filp->f_mode |= FMODE_WRITE_IOCTL;
1845 bdev = bd_acquire(inode);
1849 filp->f_mapping = bdev->bd_inode->i_mapping;
1850 filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
1852 return blkdev_get(bdev, filp->f_mode, filp);
1855 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1857 struct gendisk *disk = bdev->bd_disk;
1858 struct block_device *victim = NULL;
1860 mutex_lock_nested(&bdev->bd_mutex, for_part);
1862 bdev->bd_part_count--;
1864 if (!--bdev->bd_openers) {
1865 WARN_ON_ONCE(bdev->bd_holders);
1866 sync_blockdev(bdev);
1869 bdev_write_inode(bdev);
1871 if (bdev->bd_contains == bdev) {
1872 if (disk->fops->release)
1873 disk->fops->release(disk, mode);
1875 if (!bdev->bd_openers) {
1876 disk_put_part(bdev->bd_part);
1877 bdev->bd_part = NULL;
1878 bdev->bd_disk = NULL;
1879 if (bdev != bdev->bd_contains)
1880 victim = bdev->bd_contains;
1881 bdev->bd_contains = NULL;
1883 put_disk_and_module(disk);
1885 mutex_unlock(&bdev->bd_mutex);
1888 __blkdev_put(victim, mode, 1);
1891 void blkdev_put(struct block_device *bdev, fmode_t mode)
1893 mutex_lock(&bdev->bd_mutex);
1895 if (mode & FMODE_EXCL) {
1899 * Release a claim on the device. The holder fields
1900 * are protected with bdev_lock. bd_mutex is to
1901 * synchronize disk_holder unlinking.
1903 spin_lock(&bdev_lock);
1905 WARN_ON_ONCE(--bdev->bd_holders < 0);
1906 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1908 /* bd_contains might point to self, check in a separate step */
1909 if ((bdev_free = !bdev->bd_holders))
1910 bdev->bd_holder = NULL;
1911 if (!bdev->bd_contains->bd_holders)
1912 bdev->bd_contains->bd_holder = NULL;
1914 spin_unlock(&bdev_lock);
1917 * If this was the last claim, remove holder link and
1918 * unblock evpoll if it was a write holder.
1920 if (bdev_free && bdev->bd_write_holder) {
1921 disk_unblock_events(bdev->bd_disk);
1922 bdev->bd_write_holder = false;
1927 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1928 * event. This is to ensure detection of media removal commanded
1929 * from userland - e.g. eject(1).
1931 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1933 mutex_unlock(&bdev->bd_mutex);
1935 __blkdev_put(bdev, mode, 0);
1937 EXPORT_SYMBOL(blkdev_put);
1939 static int blkdev_close(struct inode * inode, struct file * filp)
1941 struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1942 blkdev_put(bdev, filp->f_mode);
1946 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1948 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1949 fmode_t mode = file->f_mode;
1952 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1953 * to updated it before every ioctl.
1955 if (file->f_flags & O_NDELAY)
1956 mode |= FMODE_NDELAY;
1958 mode &= ~FMODE_NDELAY;
1960 return blkdev_ioctl(bdev, mode, cmd, arg);
1964 * Write data to the block device. Only intended for the block device itself
1965 * and the raw driver which basically is a fake block device.
1967 * Does not take i_mutex for the write and thus is not for general purpose
1970 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1972 struct file *file = iocb->ki_filp;
1973 struct inode *bd_inode = bdev_file_inode(file);
1974 loff_t size = i_size_read(bd_inode);
1975 struct blk_plug plug;
1978 if (bdev_read_only(I_BDEV(bd_inode)))
1981 if (!iov_iter_count(from))
1984 if (iocb->ki_pos >= size)
1987 if ((iocb->ki_flags & (IOCB_NOWAIT | IOCB_DIRECT)) == IOCB_NOWAIT)
1990 iov_iter_truncate(from, size - iocb->ki_pos);
1992 blk_start_plug(&plug);
1993 ret = __generic_file_write_iter(iocb, from);
1995 ret = generic_write_sync(iocb, ret);
1996 blk_finish_plug(&plug);
1999 EXPORT_SYMBOL_GPL(blkdev_write_iter);
2001 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
2003 struct file *file = iocb->ki_filp;
2004 struct inode *bd_inode = bdev_file_inode(file);
2005 loff_t size = i_size_read(bd_inode);
2006 loff_t pos = iocb->ki_pos;
2012 iov_iter_truncate(to, size);
2013 return generic_file_read_iter(iocb, to);
2015 EXPORT_SYMBOL_GPL(blkdev_read_iter);
2018 * Try to release a page associated with block device when the system
2019 * is under memory pressure.
2021 static int blkdev_releasepage(struct page *page, gfp_t wait)
2023 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2025 if (super && super->s_op->bdev_try_to_free_page)
2026 return super->s_op->bdev_try_to_free_page(super, page, wait);
2028 return try_to_free_buffers(page);
2031 static int blkdev_writepages(struct address_space *mapping,
2032 struct writeback_control *wbc)
2034 return generic_writepages(mapping, wbc);
2037 static const struct address_space_operations def_blk_aops = {
2038 .readpage = blkdev_readpage,
2039 .readpages = blkdev_readpages,
2040 .writepage = blkdev_writepage,
2041 .write_begin = blkdev_write_begin,
2042 .write_end = blkdev_write_end,
2043 .writepages = blkdev_writepages,
2044 .releasepage = blkdev_releasepage,
2045 .direct_IO = blkdev_direct_IO,
2046 .migratepage = buffer_migrate_page_norefs,
2047 .is_dirty_writeback = buffer_check_dirty_writeback,
2050 #define BLKDEV_FALLOC_FL_SUPPORTED \
2051 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
2052 FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2054 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2057 struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2058 struct address_space *mapping;
2059 loff_t end = start + len - 1;
2063 /* Fail if we don't recognize the flags. */
2064 if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2067 /* Don't go off the end of the device. */
2068 isize = i_size_read(bdev->bd_inode);
2072 if (mode & FALLOC_FL_KEEP_SIZE) {
2073 len = isize - start;
2074 end = start + len - 1;
2080 * Don't allow IO that isn't aligned to logical block size.
2082 if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2085 /* Invalidate the page cache, including dirty pages. */
2086 mapping = bdev->bd_inode->i_mapping;
2087 truncate_inode_pages_range(mapping, start, end);
2090 case FALLOC_FL_ZERO_RANGE:
2091 case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2092 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2093 GFP_KERNEL, BLKDEV_ZERO_NOUNMAP);
2095 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2096 error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2097 GFP_KERNEL, BLKDEV_ZERO_NOFALLBACK);
2099 case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2100 error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2110 * Invalidate again; if someone wandered in and dirtied a page,
2111 * the caller will be given -EBUSY. The third argument is
2112 * inclusive, so the rounding here is safe.
2114 return invalidate_inode_pages2_range(mapping,
2115 start >> PAGE_SHIFT,
2119 const struct file_operations def_blk_fops = {
2120 .open = blkdev_open,
2121 .release = blkdev_close,
2122 .llseek = block_llseek,
2123 .read_iter = blkdev_read_iter,
2124 .write_iter = blkdev_write_iter,
2125 .iopoll = blkdev_iopoll,
2126 .mmap = generic_file_mmap,
2127 .fsync = blkdev_fsync,
2128 .unlocked_ioctl = block_ioctl,
2129 #ifdef CONFIG_COMPAT
2130 .compat_ioctl = compat_blkdev_ioctl,
2132 .splice_read = generic_file_splice_read,
2133 .splice_write = iter_file_splice_write,
2134 .fallocate = blkdev_fallocate,
2137 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2140 mm_segment_t old_fs = get_fs();
2142 res = blkdev_ioctl(bdev, 0, cmd, arg);
2147 EXPORT_SYMBOL(ioctl_by_bdev);
2150 * lookup_bdev - lookup a struct block_device by name
2151 * @pathname: special file representing the block device
2153 * Get a reference to the blockdevice at @pathname in the current
2154 * namespace if possible and return it. Return ERR_PTR(error)
2157 struct block_device *lookup_bdev(const char *pathname)
2159 struct block_device *bdev;
2160 struct inode *inode;
2164 if (!pathname || !*pathname)
2165 return ERR_PTR(-EINVAL);
2167 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2169 return ERR_PTR(error);
2171 inode = d_backing_inode(path.dentry);
2173 if (!S_ISBLK(inode->i_mode))
2176 if (!may_open_dev(&path))
2179 bdev = bd_acquire(inode);
2186 bdev = ERR_PTR(error);
2189 EXPORT_SYMBOL(lookup_bdev);
2191 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2193 struct super_block *sb = get_super(bdev);
2198 * no need to lock the super, get_super holds the
2199 * read mutex so the filesystem cannot go away
2200 * under us (->put_super runs with the write lock
2203 shrink_dcache_sb(sb);
2204 res = invalidate_inodes(sb, kill_dirty);
2207 invalidate_bdev(bdev);
2210 EXPORT_SYMBOL(__invalidate_device);
2212 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2214 struct inode *inode, *old_inode = NULL;
2216 spin_lock(&blockdev_superblock->s_inode_list_lock);
2217 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2218 struct address_space *mapping = inode->i_mapping;
2219 struct block_device *bdev;
2221 spin_lock(&inode->i_lock);
2222 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2223 mapping->nrpages == 0) {
2224 spin_unlock(&inode->i_lock);
2228 spin_unlock(&inode->i_lock);
2229 spin_unlock(&blockdev_superblock->s_inode_list_lock);
2231 * We hold a reference to 'inode' so it couldn't have been
2232 * removed from s_inodes list while we dropped the
2233 * s_inode_list_lock We cannot iput the inode now as we can
2234 * be holding the last reference and we cannot iput it under
2235 * s_inode_list_lock. So we keep the reference and iput it
2240 bdev = I_BDEV(inode);
2242 mutex_lock(&bdev->bd_mutex);
2243 if (bdev->bd_openers)
2245 mutex_unlock(&bdev->bd_mutex);
2247 spin_lock(&blockdev_superblock->s_inode_list_lock);
2249 spin_unlock(&blockdev_superblock->s_inode_list_lock);