blob: 2dfc6cdcebbe03a3c4ac76480771a9eac94ae27f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/smp_lock.h>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070015#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/highmem.h>
17#include <linux/blkdev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/buffer_head.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010021#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010022#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/mpage.h>
24#include <linux/mount.h>
25#include <linux/uio.h>
26#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070027#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31struct bdev_inode {
32 struct block_device bdev;
33 struct inode vfs_inode;
34};
35
Adrian Bunk4c54ac62008-02-18 13:48:31 +010036static const struct address_space_operations def_blk_aops;
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038static inline struct bdev_inode *BDEV_I(struct inode *inode)
39{
40 return container_of(inode, struct bdev_inode, vfs_inode);
41}
42
43inline struct block_device *I_BDEV(struct inode *inode)
44{
45 return &BDEV_I(inode)->bdev;
46}
47
48EXPORT_SYMBOL(I_BDEV);
49
50static sector_t max_block(struct block_device *bdev)
51{
52 sector_t retval = ~((sector_t)0);
53 loff_t sz = i_size_read(bdev->bd_inode);
54
55 if (sz) {
56 unsigned int size = block_size(bdev);
57 unsigned int sizebits = blksize_bits(size);
58 retval = (sz >> sizebits);
59 }
60 return retval;
61}
62
Peter Zijlstraf9a14392007-05-06 14:49:55 -070063/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070064static void kill_bdev(struct block_device *bdev)
65{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070066 if (bdev->bd_inode->i_mapping->nrpages == 0)
67 return;
68 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
70}
71
72int set_blocksize(struct block_device *bdev, int size)
73{
74 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070075 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 return -EINVAL;
77
78 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -040079 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 return -EINVAL;
81
82 /* Don't change the size if it is same as current */
83 if (bdev->bd_block_size != size) {
84 sync_blockdev(bdev);
85 bdev->bd_block_size = size;
86 bdev->bd_inode->i_blkbits = blksize_bits(size);
87 kill_bdev(bdev);
88 }
89 return 0;
90}
91
92EXPORT_SYMBOL(set_blocksize);
93
94int sb_set_blocksize(struct super_block *sb, int size)
95{
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 if (set_blocksize(sb->s_bdev, size))
97 return 0;
98 /* If we get here, we know size is power of two
99 * and it's value is between 512 and PAGE_SIZE */
100 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800101 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 return sb->s_blocksize;
103}
104
105EXPORT_SYMBOL(sb_set_blocksize);
106
107int sb_min_blocksize(struct super_block *sb, int size)
108{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400109 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 if (size < minsize)
111 size = minsize;
112 return sb_set_blocksize(sb, size);
113}
114
115EXPORT_SYMBOL(sb_min_blocksize);
116
117static int
118blkdev_get_block(struct inode *inode, sector_t iblock,
119 struct buffer_head *bh, int create)
120{
121 if (iblock >= max_block(I_BDEV(inode))) {
122 if (create)
123 return -EIO;
124
125 /*
126 * for reads, we're just trying to fill a partial page.
127 * return a hole, they will have to call get_block again
128 * before they can fill it, and they will get -EIO at that
129 * time
130 */
131 return 0;
132 }
133 bh->b_bdev = I_BDEV(inode);
134 bh->b_blocknr = iblock;
135 set_buffer_mapped(bh);
136 return 0;
137}
138
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800139static int
140blkdev_get_blocks(struct inode *inode, sector_t iblock,
141 struct buffer_head *bh, int create)
142{
143 sector_t end_block = max_block(I_BDEV(inode));
144 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
145
146 if ((iblock + max_blocks) > end_block) {
147 max_blocks = end_block - iblock;
148 if ((long)max_blocks <= 0) {
149 if (create)
150 return -EIO; /* write fully beyond EOF */
151 /*
152 * It is a read which is fully beyond EOF. We return
153 * a !buffer_mapped buffer
154 */
155 max_blocks = 0;
156 }
157 }
158
159 bh->b_bdev = I_BDEV(inode);
160 bh->b_blocknr = iblock;
161 bh->b_size = max_blocks << inode->i_blkbits;
162 if (max_blocks)
163 set_buffer_mapped(bh);
164 return 0;
165}
166
167static ssize_t
168blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
169 loff_t offset, unsigned long nr_segs)
170{
171 struct file *file = iocb->ki_filp;
172 struct inode *inode = file->f_mapping->host;
173
174 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
175 iov, offset, nr_segs, blkdev_get_blocks, NULL);
176}
177
Nick Piggin585d3bc2009-02-25 10:44:19 +0100178/*
179 * Write out and wait upon all the dirty data associated with a block
180 * device via its mapping. Does not take the superblock lock.
181 */
182int sync_blockdev(struct block_device *bdev)
183{
184 int ret = 0;
185
186 if (bdev)
187 ret = filemap_write_and_wait(bdev->bd_inode->i_mapping);
188 return ret;
189}
190EXPORT_SYMBOL(sync_blockdev);
191
192/*
193 * Write out and wait upon all dirty data associated with this
194 * device. Filesystem data as well as the underlying block
195 * device. Takes the superblock lock.
196 */
197int fsync_bdev(struct block_device *bdev)
198{
199 struct super_block *sb = get_super(bdev);
200 if (sb) {
201 int res = fsync_super(sb);
202 drop_super(sb);
203 return res;
204 }
205 return sync_blockdev(bdev);
206}
Al Viro47e44912009-04-01 07:07:16 -0400207EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100208
209/**
210 * freeze_bdev -- lock a filesystem and force it into a consistent state
211 * @bdev: blockdevice to lock
212 *
213 * This takes the block device bd_mount_sem to make sure no new mounts
214 * happen on bdev until thaw_bdev() is called.
215 * If a superblock is found on this device, we take the s_umount semaphore
216 * on it to make sure nobody unmounts until the snapshot creation is done.
217 * The reference counter (bd_fsfreeze_count) guarantees that only the last
218 * unfreeze process can unfreeze the frozen filesystem actually when multiple
219 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
220 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
221 * actually.
222 */
223struct super_block *freeze_bdev(struct block_device *bdev)
224{
225 struct super_block *sb;
226 int error = 0;
227
228 mutex_lock(&bdev->bd_fsfreeze_mutex);
229 if (bdev->bd_fsfreeze_count > 0) {
230 bdev->bd_fsfreeze_count++;
231 sb = get_super(bdev);
232 mutex_unlock(&bdev->bd_fsfreeze_mutex);
233 return sb;
234 }
235 bdev->bd_fsfreeze_count++;
236
237 down(&bdev->bd_mount_sem);
238 sb = get_super(bdev);
239 if (sb && !(sb->s_flags & MS_RDONLY)) {
240 sb->s_frozen = SB_FREEZE_WRITE;
241 smp_wmb();
242
243 __fsync_super(sb);
244
245 sb->s_frozen = SB_FREEZE_TRANS;
246 smp_wmb();
247
248 sync_blockdev(sb->s_bdev);
249
250 if (sb->s_op->freeze_fs) {
251 error = sb->s_op->freeze_fs(sb);
252 if (error) {
253 printk(KERN_ERR
254 "VFS:Filesystem freeze failed\n");
255 sb->s_frozen = SB_UNFROZEN;
256 drop_super(sb);
257 up(&bdev->bd_mount_sem);
258 bdev->bd_fsfreeze_count--;
259 mutex_unlock(&bdev->bd_fsfreeze_mutex);
260 return ERR_PTR(error);
261 }
262 }
263 }
264
265 sync_blockdev(bdev);
266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
267
268 return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */
269}
270EXPORT_SYMBOL(freeze_bdev);
271
272/**
273 * thaw_bdev -- unlock filesystem
274 * @bdev: blockdevice to unlock
275 * @sb: associated superblock
276 *
277 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
278 */
279int thaw_bdev(struct block_device *bdev, struct super_block *sb)
280{
281 int error = 0;
282
283 mutex_lock(&bdev->bd_fsfreeze_mutex);
284 if (!bdev->bd_fsfreeze_count) {
285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
286 return -EINVAL;
287 }
288
289 bdev->bd_fsfreeze_count--;
290 if (bdev->bd_fsfreeze_count > 0) {
291 if (sb)
292 drop_super(sb);
293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
294 return 0;
295 }
296
297 if (sb) {
298 BUG_ON(sb->s_bdev != bdev);
299 if (!(sb->s_flags & MS_RDONLY)) {
300 if (sb->s_op->unfreeze_fs) {
301 error = sb->s_op->unfreeze_fs(sb);
302 if (error) {
303 printk(KERN_ERR
304 "VFS:Filesystem thaw failed\n");
305 sb->s_frozen = SB_FREEZE_TRANS;
306 bdev->bd_fsfreeze_count++;
307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
308 return error;
309 }
310 }
311 sb->s_frozen = SB_UNFROZEN;
312 smp_wmb();
313 wake_up(&sb->s_wait_unfrozen);
314 }
315 drop_super(sb);
316 }
317
318 up(&bdev->bd_mount_sem);
319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
320 return 0;
321}
322EXPORT_SYMBOL(thaw_bdev);
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
325{
326 return block_write_full_page(page, blkdev_get_block, wbc);
327}
328
329static int blkdev_readpage(struct file * file, struct page * page)
330{
331 return block_read_full_page(page, blkdev_get_block);
332}
333
Nick Piggin6272b5a2007-10-16 01:25:04 -0700334static int blkdev_write_begin(struct file *file, struct address_space *mapping,
335 loff_t pos, unsigned len, unsigned flags,
336 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700338 *pagep = NULL;
339 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
340 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
Nick Piggin6272b5a2007-10-16 01:25:04 -0700343static int blkdev_write_end(struct file *file, struct address_space *mapping,
344 loff_t pos, unsigned len, unsigned copied,
345 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700347 int ret;
348 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
349
350 unlock_page(page);
351 page_cache_release(page);
352
353 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
356/*
357 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800358 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 * so we compute the size by hand (just as in block_read/write above)
360 */
361static loff_t block_llseek(struct file *file, loff_t offset, int origin)
362{
363 struct inode *bd_inode = file->f_mapping->host;
364 loff_t size;
365 loff_t retval;
366
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800367 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 size = i_size_read(bd_inode);
369
370 switch (origin) {
371 case 2:
372 offset += size;
373 break;
374 case 1:
375 offset += file->f_pos;
376 }
377 retval = -EINVAL;
378 if (offset >= 0 && offset <= size) {
379 if (offset != file->f_pos) {
380 file->f_pos = offset;
381 }
382 retval = offset;
383 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800384 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 return retval;
386}
387
388/*
389 * Filp is never NULL; the only case when ->fsync() is called with
390 * NULL first argument is nfsd_sync_dir() and that's not a directory.
391 */
392
393static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
394{
395 return sync_blockdev(I_BDEV(filp->f_mapping->host));
396}
397
398/*
399 * pseudo-fs
400 */
401
402static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800403static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405static struct inode *bdev_alloc_inode(struct super_block *sb)
406{
Christoph Lametere94b1762006-12-06 20:33:17 -0800407 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (!ei)
409 return NULL;
410 return &ei->vfs_inode;
411}
412
413static void bdev_destroy_inode(struct inode *inode)
414{
415 struct bdev_inode *bdi = BDEV_I(inode);
416
417 bdi->bdev.bd_inode_backing_dev_info = NULL;
418 kmem_cache_free(bdev_cachep, bdi);
419}
420
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700421static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
423 struct bdev_inode *ei = (struct bdev_inode *) foo;
424 struct block_device *bdev = &ei->bdev;
425
Christoph Lametera35afb82007-05-16 22:10:57 -0700426 memset(bdev, 0, sizeof(*bdev));
427 mutex_init(&bdev->bd_mutex);
428 sema_init(&bdev->bd_mount_sem, 1);
429 INIT_LIST_HEAD(&bdev->bd_inodes);
430 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800431#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700432 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800433#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700434 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800435 /* Initialize mutex for freeze. */
436 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
439static inline void __bd_forget(struct inode *inode)
440{
441 list_del_init(&inode->i_devices);
442 inode->i_bdev = NULL;
443 inode->i_mapping = &inode->i_data;
444}
445
446static void bdev_clear_inode(struct inode *inode)
447{
448 struct block_device *bdev = &BDEV_I(inode)->bdev;
449 struct list_head *p;
450 spin_lock(&bdev_lock);
451 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
452 __bd_forget(list_entry(p, struct inode, i_devices));
453 }
454 list_del_init(&bdev->bd_list);
455 spin_unlock(&bdev_lock);
456}
457
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800458static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 .statfs = simple_statfs,
460 .alloc_inode = bdev_alloc_inode,
461 .destroy_inode = bdev_destroy_inode,
462 .drop_inode = generic_delete_inode,
463 .clear_inode = bdev_clear_inode,
464};
465
David Howells454e2392006-06-23 02:02:57 -0700466static int bd_get_sb(struct file_system_type *fs_type,
467 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
David Howells454e2392006-06-23 02:02:57 -0700469 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472static struct file_system_type bd_type = {
473 .name = "bdev",
474 .get_sb = bd_get_sb,
475 .kill_sb = kill_anon_super,
476};
477
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800478struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480void __init bdev_cache_init(void)
481{
482 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800483 struct vfsmount *bd_mnt;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800486 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
487 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900488 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 err = register_filesystem(&bd_type);
490 if (err)
491 panic("Cannot register bdev pseudo-fs");
492 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (IS_ERR(bd_mnt))
494 panic("Cannot create bdev pseudo-fs");
495 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
496}
497
498/*
499 * Most likely _very_ bad one - but then it's hardly critical for small
500 * /dev and can be fixed when somebody will need really large one.
501 * Keep in mind that it will be fed through icache hash function too.
502 */
503static inline unsigned long hash(dev_t dev)
504{
505 return MAJOR(dev)+MINOR(dev);
506}
507
508static int bdev_test(struct inode *inode, void *data)
509{
510 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
511}
512
513static int bdev_set(struct inode *inode, void *data)
514{
515 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
516 return 0;
517}
518
519static LIST_HEAD(all_bdevs);
520
521struct block_device *bdget(dev_t dev)
522{
523 struct block_device *bdev;
524 struct inode *inode;
525
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800526 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 bdev_test, bdev_set, &dev);
528
529 if (!inode)
530 return NULL;
531
532 bdev = &BDEV_I(inode)->bdev;
533
534 if (inode->i_state & I_NEW) {
535 bdev->bd_contains = NULL;
536 bdev->bd_inode = inode;
537 bdev->bd_block_size = (1 << inode->i_blkbits);
538 bdev->bd_part_count = 0;
539 bdev->bd_invalidated = 0;
540 inode->i_mode = S_IFBLK;
541 inode->i_rdev = dev;
542 inode->i_bdev = bdev;
543 inode->i_data.a_ops = &def_blk_aops;
544 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
545 inode->i_data.backing_dev_info = &default_backing_dev_info;
546 spin_lock(&bdev_lock);
547 list_add(&bdev->bd_list, &all_bdevs);
548 spin_unlock(&bdev_lock);
549 unlock_new_inode(inode);
550 }
551 return bdev;
552}
553
554EXPORT_SYMBOL(bdget);
555
556long nr_blockdev_pages(void)
557{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700558 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 long ret = 0;
560 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700561 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 ret += bdev->bd_inode->i_mapping->nrpages;
563 }
564 spin_unlock(&bdev_lock);
565 return ret;
566}
567
568void bdput(struct block_device *bdev)
569{
570 iput(bdev->bd_inode);
571}
572
573EXPORT_SYMBOL(bdput);
574
575static struct block_device *bd_acquire(struct inode *inode)
576{
577 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 spin_lock(&bdev_lock);
580 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700581 if (bdev) {
582 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 spin_unlock(&bdev_lock);
584 return bdev;
585 }
586 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700587
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 bdev = bdget(inode->i_rdev);
589 if (bdev) {
590 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700591 if (!inode->i_bdev) {
592 /*
593 * We take an additional bd_inode->i_count for inode,
594 * and it's released in clear_inode() of inode.
595 * So, we can access it via ->i_mapping always
596 * without igrab().
597 */
598 atomic_inc(&bdev->bd_inode->i_count);
599 inode->i_bdev = bdev;
600 inode->i_mapping = bdev->bd_inode->i_mapping;
601 list_add(&inode->i_devices, &bdev->bd_inodes);
602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 spin_unlock(&bdev_lock);
604 }
605 return bdev;
606}
607
608/* Call when you free inode */
609
610void bd_forget(struct inode *inode)
611{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700612 struct block_device *bdev = NULL;
613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700615 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800616 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700617 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700621
622 if (bdev)
623 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624}
625
626int bd_claim(struct block_device *bdev, void *holder)
627{
628 int res;
629 spin_lock(&bdev_lock);
630
631 /* first decide result */
632 if (bdev->bd_holder == holder)
633 res = 0; /* already a holder */
634 else if (bdev->bd_holder != NULL)
635 res = -EBUSY; /* held by someone else */
636 else if (bdev->bd_contains == bdev)
637 res = 0; /* is a whole device which isn't held */
638
639 else if (bdev->bd_contains->bd_holder == bd_claim)
640 res = 0; /* is a partition of a device that is being partitioned */
641 else if (bdev->bd_contains->bd_holder != NULL)
642 res = -EBUSY; /* is a partition of a held device */
643 else
644 res = 0; /* is a partition of an un-held device */
645
646 /* now impose change */
647 if (res==0) {
648 /* note that for a whole device bd_holders
649 * will be incremented twice, and bd_holder will
650 * be set to bd_claim before being set to holder
651 */
652 bdev->bd_contains->bd_holders ++;
653 bdev->bd_contains->bd_holder = bd_claim;
654 bdev->bd_holders++;
655 bdev->bd_holder = holder;
656 }
657 spin_unlock(&bdev_lock);
658 return res;
659}
660
661EXPORT_SYMBOL(bd_claim);
662
663void bd_release(struct block_device *bdev)
664{
665 spin_lock(&bdev_lock);
666 if (!--bdev->bd_contains->bd_holders)
667 bdev->bd_contains->bd_holder = NULL;
668 if (!--bdev->bd_holders)
669 bdev->bd_holder = NULL;
670 spin_unlock(&bdev_lock);
671}
672
673EXPORT_SYMBOL(bd_release);
674
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800675#ifdef CONFIG_SYSFS
676/*
677 * Functions for bd_claim_by_kobject / bd_release_from_kobject
678 *
679 * If a kobject is passed to bd_claim_by_kobject()
680 * and the kobject has a parent directory,
681 * following symlinks are created:
682 * o from the kobject to the claimed bdev
683 * o from "holders" directory of the bdev to the parent of the kobject
684 * bd_release_from_kobject() removes these symlinks.
685 *
686 * Example:
687 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
688 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
689 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
690 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
691 */
692
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700693static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800694{
695 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700696 return 0;
697 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800698}
699
700static void del_symlink(struct kobject *from, struct kobject *to)
701{
702 if (!from || !to)
703 return;
704 sysfs_remove_link(from, kobject_name(to));
705}
706
707/*
708 * 'struct bd_holder' contains pointers to kobjects symlinked by
709 * bd_claim_by_kobject.
710 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
711 */
712struct bd_holder {
713 struct list_head list; /* chain of holders of the bdev */
714 int count; /* references from the holder */
715 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
716 struct kobject *hdev; /* e.g. "/block/dm-0" */
717 struct kobject *hdir; /* e.g. "/block/sda/holders" */
718 struct kobject *sdev; /* e.g. "/block/sda" */
719};
720
721/*
722 * Get references of related kobjects at once.
723 * Returns 1 on success. 0 on failure.
724 *
725 * Should call bd_holder_release_dirs() after successful use.
726 */
727static int bd_holder_grab_dirs(struct block_device *bdev,
728 struct bd_holder *bo)
729{
730 if (!bdev || !bo)
731 return 0;
732
733 bo->sdir = kobject_get(bo->sdir);
734 if (!bo->sdir)
735 return 0;
736
737 bo->hdev = kobject_get(bo->sdir->parent);
738 if (!bo->hdev)
739 goto fail_put_sdir;
740
Tejun Heo0762b8b2008-08-25 19:56:12 +0900741 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800742 if (!bo->sdev)
743 goto fail_put_hdev;
744
Tejun Heo4c465012008-08-25 19:56:11 +0900745 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800746 if (!bo->hdir)
747 goto fail_put_sdev;
748
749 return 1;
750
751fail_put_sdev:
752 kobject_put(bo->sdev);
753fail_put_hdev:
754 kobject_put(bo->hdev);
755fail_put_sdir:
756 kobject_put(bo->sdir);
757
758 return 0;
759}
760
761/* Put references of related kobjects at once. */
762static void bd_holder_release_dirs(struct bd_holder *bo)
763{
764 kobject_put(bo->hdir);
765 kobject_put(bo->sdev);
766 kobject_put(bo->hdev);
767 kobject_put(bo->sdir);
768}
769
770static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
771{
772 struct bd_holder *bo;
773
774 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
775 if (!bo)
776 return NULL;
777
778 bo->count = 1;
779 bo->sdir = kobj;
780
781 return bo;
782}
783
784static void free_bd_holder(struct bd_holder *bo)
785{
786 kfree(bo);
787}
788
789/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500790 * find_bd_holder - find matching struct bd_holder from the block device
791 *
792 * @bdev: struct block device to be searched
793 * @bo: target struct bd_holder
794 *
795 * Returns matching entry with @bo in @bdev->bd_holder_list.
796 * If found, increment the reference count and return the pointer.
797 * If not found, returns NULL.
798 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800799static struct bd_holder *find_bd_holder(struct block_device *bdev,
800 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500801{
802 struct bd_holder *tmp;
803
804 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
805 if (tmp->sdir == bo->sdir) {
806 tmp->count++;
807 return tmp;
808 }
809
810 return NULL;
811}
812
813/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800814 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
815 *
816 * @bdev: block device to be bd_claimed
817 * @bo: preallocated and initialized by alloc_bd_holder()
818 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500819 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800820 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500821 * Returns 0 if symlinks are created.
822 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800823 */
824static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
825{
Johannes Weiner4e916722007-07-15 23:41:25 -0700826 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800827
828 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700829 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800830
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800831 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700832 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800833
Johannes Weiner4e916722007-07-15 23:41:25 -0700834 err = add_symlink(bo->sdir, bo->sdev);
835 if (err)
836 return err;
837
838 err = add_symlink(bo->hdir, bo->hdev);
839 if (err) {
840 del_symlink(bo->sdir, bo->sdev);
841 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700842 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700843
844 list_add_tail(&bo->list, &bdev->bd_holder_list);
845 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800846}
847
848/**
849 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
850 *
851 * @bdev: block device to be bd_claimed
852 * @kobj: holder's kobject
853 *
854 * If there is matching entry with @kobj in @bdev->bd_holder_list
855 * and no other bd_claim() from the same kobject,
856 * remove the struct bd_holder from the list, delete symlinks for it.
857 *
858 * Returns a pointer to the struct bd_holder when it's removed from the list
859 * and ready to be freed.
860 * Returns NULL if matching claim isn't found or there is other bd_claim()
861 * by the same kobject.
862 */
863static struct bd_holder *del_bd_holder(struct block_device *bdev,
864 struct kobject *kobj)
865{
866 struct bd_holder *bo;
867
868 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
869 if (bo->sdir == kobj) {
870 bo->count--;
871 BUG_ON(bo->count < 0);
872 if (!bo->count) {
873 list_del(&bo->list);
874 del_symlink(bo->sdir, bo->sdev);
875 del_symlink(bo->hdir, bo->hdev);
876 bd_holder_release_dirs(bo);
877 return bo;
878 }
879 break;
880 }
881 }
882
883 return NULL;
884}
885
886/**
887 * bd_claim_by_kobject - bd_claim() with additional kobject signature
888 *
889 * @bdev: block device to be claimed
890 * @holder: holder's signature
891 * @kobj: holder's kobject
892 *
893 * Do bd_claim() and if it succeeds, create sysfs symlinks between
894 * the bdev and the holder's kobject.
895 * Use bd_release_from_kobject() when relesing the claimed bdev.
896 *
897 * Returns 0 on success. (same as bd_claim())
898 * Returns errno on failure.
899 */
900static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
901 struct kobject *kobj)
902{
Johannes Weiner4e916722007-07-15 23:41:25 -0700903 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500904 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800905
906 if (!kobj)
907 return -EINVAL;
908
909 bo = alloc_bd_holder(kobj);
910 if (!bo)
911 return -ENOMEM;
912
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800913 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500914
Johannes Weiner4e916722007-07-15 23:41:25 -0700915 err = bd_claim(bdev, holder);
916 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700917 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700918
919 found = find_bd_holder(bdev, bo);
920 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700921 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700922
923 err = add_bd_holder(bdev, bo);
924 if (err)
925 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700926 else
927 bo = NULL;
928fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800929 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700930 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700931 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800932}
933
934/**
935 * bd_release_from_kobject - bd_release() with additional kobject signature
936 *
937 * @bdev: block device to be released
938 * @kobj: holder's kobject
939 *
940 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
941 */
942static void bd_release_from_kobject(struct block_device *bdev,
943 struct kobject *kobj)
944{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800945 if (!kobj)
946 return;
947
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800948 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800949 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700950 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800951 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800952}
953
954/**
955 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
956 *
957 * @bdev: block device to be claimed
958 * @holder: holder's signature
959 * @disk: holder's gendisk
960 *
961 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
962 */
963int bd_claim_by_disk(struct block_device *bdev, void *holder,
964 struct gendisk *disk)
965{
966 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
967}
968EXPORT_SYMBOL_GPL(bd_claim_by_disk);
969
970/**
971 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
972 *
973 * @bdev: block device to be claimed
974 * @disk: holder's gendisk
975 *
976 * Call bd_release_from_kobject() and put @disk->slave_dir.
977 */
978void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
979{
980 bd_release_from_kobject(bdev, disk->slave_dir);
981 kobject_put(disk->slave_dir);
982}
983EXPORT_SYMBOL_GPL(bd_release_from_disk);
984#endif
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986/*
987 * Tries to open block device by device number. Use it ONLY if you
988 * really do not have anything better - i.e. when you are behind a
989 * truly sucky interface and all you are given is a device number. _Never_
990 * to be used for internal purposes. If you ever need it - reconsider
991 * your API.
992 */
Al Viroaeb5d722008-09-02 15:28:45 -0400993struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 struct block_device *bdev = bdget(dev);
996 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -0400998 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 return err ? ERR_PTR(err) : bdev;
1000}
1001
1002EXPORT_SYMBOL(open_by_devnum);
1003
Andrew Patterson0c002c22008-09-04 14:27:20 -06001004/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001005 * flush_disk - invalidates all buffer-cache entries on a disk
1006 *
1007 * @bdev: struct block device to be flushed
1008 *
1009 * Invalidates all buffer-cache entries on a disk. It should be called
1010 * when a disk has been changed -- either by a media change or online
1011 * resize.
1012 */
1013static void flush_disk(struct block_device *bdev)
1014{
1015 if (__invalidate_device(bdev)) {
1016 char name[BDEVNAME_SIZE] = "";
1017
1018 if (bdev->bd_disk)
1019 disk_name(bdev->bd_disk, 0, name);
1020 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1021 "resized disk %s\n", name);
1022 }
1023
1024 if (!bdev->bd_disk)
1025 return;
1026 if (disk_partitionable(bdev->bd_disk))
1027 bdev->bd_invalidated = 1;
1028}
1029
1030/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001031 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001032 * @disk: struct gendisk to check
1033 * @bdev: struct bdev to adjust.
1034 *
1035 * This routine checks to see if the bdev size does not match the disk size
1036 * and adjusts it if it differs.
1037 */
1038void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1039{
1040 loff_t disk_size, bdev_size;
1041
1042 disk_size = (loff_t)get_capacity(disk) << 9;
1043 bdev_size = i_size_read(bdev->bd_inode);
1044 if (disk_size != bdev_size) {
1045 char name[BDEVNAME_SIZE];
1046
1047 disk_name(disk, 0, name);
1048 printk(KERN_INFO
1049 "%s: detected capacity change from %lld to %lld\n",
1050 name, bdev_size, disk_size);
1051 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001052 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001053 }
1054}
1055EXPORT_SYMBOL(check_disk_size_change);
1056
1057/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001058 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001059 * @disk: struct gendisk to be revalidated
1060 *
1061 * This routine is a wrapper for lower-level driver's revalidate_disk
1062 * call-backs. It is used to do common pre and post operations needed
1063 * for all revalidate_disk operations.
1064 */
1065int revalidate_disk(struct gendisk *disk)
1066{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001067 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001068 int ret = 0;
1069
1070 if (disk->fops->revalidate_disk)
1071 ret = disk->fops->revalidate_disk(disk);
1072
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001073 bdev = bdget_disk(disk, 0);
1074 if (!bdev)
1075 return ret;
1076
1077 mutex_lock(&bdev->bd_mutex);
1078 check_disk_size_change(disk, bdev);
1079 mutex_unlock(&bdev->bd_mutex);
1080 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001081 return ret;
1082}
1083EXPORT_SYMBOL(revalidate_disk);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085/*
1086 * This routine checks whether a removable media has been changed,
1087 * and invalidates all buffer-cache-entries in that case. This
1088 * is a relatively slow routine, so we have to try to minimize using
1089 * it. Thus it is called only upon a 'mount' or 'open'. This
1090 * is the best way of combining speed and utility, I think.
1091 * People changing diskettes in the middle of an operation deserve
1092 * to lose :-)
1093 */
1094int check_disk_change(struct block_device *bdev)
1095{
1096 struct gendisk *disk = bdev->bd_disk;
1097 struct block_device_operations * bdops = disk->fops;
1098
1099 if (!bdops->media_changed)
1100 return 0;
1101 if (!bdops->media_changed(bdev->bd_disk))
1102 return 0;
1103
Andrew Patterson56ade442008-09-04 14:27:40 -06001104 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (bdops->revalidate_disk)
1106 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 return 1;
1108}
1109
1110EXPORT_SYMBOL(check_disk_change);
1111
1112void bd_set_size(struct block_device *bdev, loff_t size)
1113{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001114 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 bdev->bd_inode->i_size = size;
1117 while (bsize < PAGE_CACHE_SIZE) {
1118 if (size & bsize)
1119 break;
1120 bsize <<= 1;
1121 }
1122 bdev->bd_block_size = bsize;
1123 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1124}
1125EXPORT_SYMBOL(bd_set_size);
1126
Al Viro9a1c3542008-02-22 20:40:24 -05001127static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001128
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001129/*
1130 * bd_mutex locking:
1131 *
1132 * mutex_lock(part->bd_mutex)
1133 * mutex_lock_nested(whole->bd_mutex, 1)
1134 */
1135
Al Viro572c4892007-10-08 13:24:05 -04001136static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001139 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001140 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001141 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142
Al Viro572c4892007-10-08 13:24:05 -04001143 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001144 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001145 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001146 perm |= MAY_WRITE;
1147 /*
1148 * hooks: /n/, see "layering violations".
1149 */
1150 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001151 if (ret != 0) {
1152 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001153 return ret;
Al Viro82666022008-08-01 05:32:04 -04001154 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001155
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 lock_kernel();
NeilBrownd3374822009-01-09 08:31:10 +11001157 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001158
Tejun Heo89f97492008-11-05 10:21:06 +01001159 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001160 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001161 if (!disk)
1162 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
NeilBrown6796bf52006-12-08 02:36:16 -08001164 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 if (!bdev->bd_openers) {
1166 bdev->bd_disk = disk;
1167 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001168 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001170
1171 ret = -ENXIO;
1172 bdev->bd_part = disk_get_part(disk, partno);
1173 if (!bdev->bd_part)
1174 goto out_clear;
1175
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001177 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001178 if (ret == -ERESTARTSYS) {
1179 /* Lost a race with 'disk' being
1180 * deleted, try again.
1181 * See md.c
1182 */
1183 disk_put_part(bdev->bd_part);
1184 bdev->bd_part = NULL;
1185 module_put(disk->fops->owner);
1186 put_disk(disk);
1187 bdev->bd_disk = NULL;
1188 mutex_unlock(&bdev->bd_mutex);
1189 goto restart;
1190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001192 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 }
1194 if (!bdev->bd_openers) {
1195 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1196 bdi = blk_get_backing_dev_info(bdev);
1197 if (bdi == NULL)
1198 bdi = &default_backing_dev_info;
1199 bdev->bd_inode->i_data.backing_dev_info = bdi;
1200 }
1201 if (bdev->bd_invalidated)
1202 rescan_partitions(disk, bdev);
1203 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 struct block_device *whole;
1205 whole = bdget_disk(disk, 0);
1206 ret = -ENOMEM;
1207 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001208 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001209 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001210 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001212 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 bdev->bd_inode->i_data.backing_dev_info =
1215 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001216 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001217 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001218 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001220 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Tejun Heo89f97492008-11-05 10:21:06 +01001222 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 }
1224 } else {
1225 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001226 module_put(disk->fops->owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001227 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (bdev->bd_contains == bdev) {
1229 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001230 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001232 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 }
1234 if (bdev->bd_invalidated)
1235 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 }
1237 }
1238 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001239 if (for_part)
1240 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001241 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 unlock_kernel();
1243 return 0;
1244
Tejun Heo0762b8b2008-08-25 19:56:12 +09001245 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001246 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001248 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1250 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001251 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001253 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001254 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001255 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001257
Tejun Heo0762b8b2008-08-25 19:56:12 +09001258 if (disk)
1259 module_put(disk->fops->owner);
1260 put_disk(disk);
1261 bdput(bdev);
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return ret;
1264}
1265
Al Viro572c4892007-10-08 13:24:05 -04001266int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
Al Viro572c4892007-10-08 13:24:05 -04001268 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001269}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270EXPORT_SYMBOL(blkdev_get);
1271
1272static int blkdev_open(struct inode * inode, struct file * filp)
1273{
1274 struct block_device *bdev;
1275 int res;
1276
1277 /*
1278 * Preserve backwards compatibility and allow large file access
1279 * even if userspace doesn't ask for it explicitly. Some mkfs
1280 * binary needs it. We might want to drop this workaround
1281 * during an unstable branch.
1282 */
1283 filp->f_flags |= O_LARGEFILE;
1284
Al Viro572c4892007-10-08 13:24:05 -04001285 if (filp->f_flags & O_NDELAY)
1286 filp->f_mode |= FMODE_NDELAY;
1287 if (filp->f_flags & O_EXCL)
1288 filp->f_mode |= FMODE_EXCL;
1289 if ((filp->f_flags & O_ACCMODE) == 3)
1290 filp->f_mode |= FMODE_WRITE_IOCTL;
1291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001293 if (bdev == NULL)
1294 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
Al Viro572c4892007-10-08 13:24:05 -04001296 filp->f_mapping = bdev->bd_inode->i_mapping;
1297
1298 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 if (res)
1300 return res;
1301
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001302 if (filp->f_mode & FMODE_EXCL) {
1303 res = bd_claim(bdev, filp);
1304 if (res)
1305 goto out_blkdev_put;
1306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001308 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001310 out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001311 blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 return res;
1313}
1314
Al Viro9a1c3542008-02-22 20:40:24 -05001315static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001316{
1317 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001318 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001319 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001320
NeilBrown6796bf52006-12-08 02:36:16 -08001321 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001322 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001323 if (for_part)
1324 bdev->bd_part_count--;
1325
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001326 if (!--bdev->bd_openers) {
1327 sync_blockdev(bdev);
1328 kill_bdev(bdev);
1329 }
1330 if (bdev->bd_contains == bdev) {
1331 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001332 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001333 }
1334 if (!bdev->bd_openers) {
1335 struct module *owner = disk->fops->owner;
1336
1337 put_disk(disk);
1338 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001339 disk_put_part(bdev->bd_part);
1340 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001341 bdev->bd_disk = NULL;
1342 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001343 if (bdev != bdev->bd_contains)
1344 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001345 bdev->bd_contains = NULL;
1346 }
1347 unlock_kernel();
1348 mutex_unlock(&bdev->bd_mutex);
1349 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001350 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001351 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001352 return ret;
1353}
1354
Al Viro9a1c3542008-02-22 20:40:24 -05001355int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001356{
Al Viro9a1c3542008-02-22 20:40:24 -05001357 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001358}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001359EXPORT_SYMBOL(blkdev_put);
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361static int blkdev_close(struct inode * inode, struct file * filp)
1362{
1363 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1364 if (bdev->bd_holder == filp)
1365 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001366 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367}
1368
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001369static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370{
Al Viro56b26ad2008-09-19 03:17:36 -04001371 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1372 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001373
1374 /*
1375 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1376 * to updated it before every ioctl.
1377 */
Al Viro56b26ad2008-09-19 03:17:36 -04001378 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001379 mode |= FMODE_NDELAY;
1380 else
1381 mode &= ~FMODE_NDELAY;
1382
Al Viro56b26ad2008-09-19 03:17:36 -04001383 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384}
1385
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001386/*
1387 * Try to release a page associated with block device when the system
1388 * is under memory pressure.
1389 */
1390static int blkdev_releasepage(struct page *page, gfp_t wait)
1391{
1392 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1393
1394 if (super && super->s_op->bdev_try_to_free_page)
1395 return super->s_op->bdev_try_to_free_page(super, page, wait);
1396
1397 return try_to_free_buffers(page);
1398}
1399
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001400static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 .readpage = blkdev_readpage,
1402 .writepage = blkdev_writepage,
1403 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001404 .write_begin = blkdev_write_begin,
1405 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001407 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 .direct_IO = blkdev_direct_IO,
1409};
1410
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001411const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 .open = blkdev_open,
1413 .release = blkdev_close,
1414 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001415 .read = do_sync_read,
1416 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001418 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 .mmap = generic_file_mmap,
1420 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001421 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422#ifdef CONFIG_COMPAT
1423 .compat_ioctl = compat_blkdev_ioctl,
1424#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001425 .splice_read = generic_file_splice_read,
1426 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427};
1428
1429int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1430{
1431 int res;
1432 mm_segment_t old_fs = get_fs();
1433 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001434 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 set_fs(old_fs);
1436 return res;
1437}
1438
1439EXPORT_SYMBOL(ioctl_by_bdev);
1440
1441/**
1442 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001443 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001445 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 * namespace if possible and return it. Return ERR_PTR(error)
1447 * otherwise.
1448 */
Al Viro421748e2008-08-02 01:04:36 -04001449struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
1451 struct block_device *bdev;
1452 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001453 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 int error;
1455
Al Viro421748e2008-08-02 01:04:36 -04001456 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 return ERR_PTR(-EINVAL);
1458
Al Viro421748e2008-08-02 01:04:36 -04001459 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 if (error)
1461 return ERR_PTR(error);
1462
Al Viro421748e2008-08-02 01:04:36 -04001463 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464 error = -ENOTBLK;
1465 if (!S_ISBLK(inode->i_mode))
1466 goto fail;
1467 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001468 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 goto fail;
1470 error = -ENOMEM;
1471 bdev = bd_acquire(inode);
1472 if (!bdev)
1473 goto fail;
1474out:
Al Viro421748e2008-08-02 01:04:36 -04001475 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 return bdev;
1477fail:
1478 bdev = ERR_PTR(error);
1479 goto out;
1480}
Al Virod5686b42008-08-01 05:00:11 -04001481EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
1483/**
Al Viro30c40d22008-02-22 19:50:45 -05001484 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 *
1486 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001487 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * @holder: owner for exclusion
1489 *
1490 * Open the blockdevice described by the special file at @path, claim it
1491 * for the @holder.
1492 */
Al Viro30c40d22008-02-22 19:50:45 -05001493struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
1495 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 int error = 0;
1497
1498 bdev = lookup_bdev(path);
1499 if (IS_ERR(bdev))
1500 return bdev;
1501
Al Viro572c4892007-10-08 13:24:05 -04001502 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 if (error)
1504 return ERR_PTR(error);
1505 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001506 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 goto blkdev_put;
1508 error = bd_claim(bdev, holder);
1509 if (error)
1510 goto blkdev_put;
1511
1512 return bdev;
1513
1514blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001515 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return ERR_PTR(error);
1517}
1518
Al Viro30c40d22008-02-22 19:50:45 -05001519EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521/**
Al Viro30c40d22008-02-22 19:50:45 -05001522 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001525 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
Al Viro30c40d22008-02-22 19:50:45 -05001527 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 */
Al Viro30c40d22008-02-22 19:50:45 -05001529void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
1531 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001532 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533}
1534
Al Viro30c40d22008-02-22 19:50:45 -05001535EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001536
1537int __invalidate_device(struct block_device *bdev)
1538{
1539 struct super_block *sb = get_super(bdev);
1540 int res = 0;
1541
1542 if (sb) {
1543 /*
1544 * no need to lock the super, get_super holds the
1545 * read mutex so the filesystem cannot go away
1546 * under us (->put_super runs with the write lock
1547 * hold).
1548 */
1549 shrink_dcache_sb(sb);
1550 res = invalidate_inodes(sb);
1551 drop_super(sb);
1552 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001553 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001554 return res;
1555}
1556EXPORT_SYMBOL(__invalidate_device);