blob: 3a6d4fb2a329ceb5795bb18eec19ca71f330832b [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>
Catalin Marinas2e1483c2009-06-11 13:24:13 +010028#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32struct bdev_inode {
33 struct block_device bdev;
34 struct inode vfs_inode;
35};
36
Adrian Bunk4c54ac622008-02-18 13:48:31 +010037static const struct address_space_operations def_blk_aops;
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039static inline struct bdev_inode *BDEV_I(struct inode *inode)
40{
41 return container_of(inode, struct bdev_inode, vfs_inode);
42}
43
44inline struct block_device *I_BDEV(struct inode *inode)
45{
46 return &BDEV_I(inode)->bdev;
47}
48
49EXPORT_SYMBOL(I_BDEV);
50
51static sector_t max_block(struct block_device *bdev)
52{
53 sector_t retval = ~((sector_t)0);
54 loff_t sz = i_size_read(bdev->bd_inode);
55
56 if (sz) {
57 unsigned int size = block_size(bdev);
58 unsigned int sizebits = blksize_bits(size);
59 retval = (sz >> sizebits);
60 }
61 return retval;
62}
63
Peter Zijlstraf9a14392007-05-06 14:49:55 -070064/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070065static void kill_bdev(struct block_device *bdev)
66{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070067 if (bdev->bd_inode->i_mapping->nrpages == 0)
68 return;
69 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
71}
72
73int set_blocksize(struct block_device *bdev, int size)
74{
75 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070076 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 return -EINVAL;
78
79 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -040080 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 return -EINVAL;
82
83 /* Don't change the size if it is same as current */
84 if (bdev->bd_block_size != size) {
85 sync_blockdev(bdev);
86 bdev->bd_block_size = size;
87 bdev->bd_inode->i_blkbits = blksize_bits(size);
88 kill_bdev(bdev);
89 }
90 return 0;
91}
92
93EXPORT_SYMBOL(set_blocksize);
94
95int sb_set_blocksize(struct super_block *sb, int size)
96{
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 if (set_blocksize(sb->s_bdev, size))
98 return 0;
99 /* If we get here, we know size is power of two
100 * and it's value is between 512 and PAGE_SIZE */
101 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800102 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 return sb->s_blocksize;
104}
105
106EXPORT_SYMBOL(sb_set_blocksize);
107
108int sb_min_blocksize(struct super_block *sb, int size)
109{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400110 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (size < minsize)
112 size = minsize;
113 return sb_set_blocksize(sb, size);
114}
115
116EXPORT_SYMBOL(sb_min_blocksize);
117
118static int
119blkdev_get_block(struct inode *inode, sector_t iblock,
120 struct buffer_head *bh, int create)
121{
122 if (iblock >= max_block(I_BDEV(inode))) {
123 if (create)
124 return -EIO;
125
126 /*
127 * for reads, we're just trying to fill a partial page.
128 * return a hole, they will have to call get_block again
129 * before they can fill it, and they will get -EIO at that
130 * time
131 */
132 return 0;
133 }
134 bh->b_bdev = I_BDEV(inode);
135 bh->b_blocknr = iblock;
136 set_buffer_mapped(bh);
137 return 0;
138}
139
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800140static int
141blkdev_get_blocks(struct inode *inode, sector_t iblock,
142 struct buffer_head *bh, int create)
143{
144 sector_t end_block = max_block(I_BDEV(inode));
145 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
146
147 if ((iblock + max_blocks) > end_block) {
148 max_blocks = end_block - iblock;
149 if ((long)max_blocks <= 0) {
150 if (create)
151 return -EIO; /* write fully beyond EOF */
152 /*
153 * It is a read which is fully beyond EOF. We return
154 * a !buffer_mapped buffer
155 */
156 max_blocks = 0;
157 }
158 }
159
160 bh->b_bdev = I_BDEV(inode);
161 bh->b_blocknr = iblock;
162 bh->b_size = max_blocks << inode->i_blkbits;
163 if (max_blocks)
164 set_buffer_mapped(bh);
165 return 0;
166}
167
168static ssize_t
169blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
170 loff_t offset, unsigned long nr_segs)
171{
172 struct file *file = iocb->ki_filp;
173 struct inode *inode = file->f_mapping->host;
174
175 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
176 iov, offset, nr_segs, blkdev_get_blocks, NULL);
177}
178
Jan Kara5cee5812009-04-27 16:43:51 +0200179int __sync_blockdev(struct block_device *bdev, int wait)
180{
181 if (!bdev)
182 return 0;
183 if (!wait)
184 return filemap_flush(bdev->bd_inode->i_mapping);
185 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
186}
187
Nick Piggin585d3bc2009-02-25 10:44:19 +0100188/*
189 * Write out and wait upon all the dirty data associated with a block
190 * device via its mapping. Does not take the superblock lock.
191 */
192int sync_blockdev(struct block_device *bdev)
193{
Jan Kara5cee5812009-04-27 16:43:51 +0200194 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100195}
196EXPORT_SYMBOL(sync_blockdev);
197
198/*
199 * Write out and wait upon all dirty data associated with this
200 * device. Filesystem data as well as the underlying block
201 * device. Takes the superblock lock.
202 */
203int fsync_bdev(struct block_device *bdev)
204{
205 struct super_block *sb = get_super(bdev);
206 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200207 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100208 drop_super(sb);
209 return res;
210 }
211 return sync_blockdev(bdev);
212}
Al Viro47e44912009-04-01 07:07:16 -0400213EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100214
215/**
216 * freeze_bdev -- lock a filesystem and force it into a consistent state
217 * @bdev: blockdevice to lock
218 *
219 * This takes the block device bd_mount_sem to make sure no new mounts
220 * happen on bdev until thaw_bdev() is called.
221 * If a superblock is found on this device, we take the s_umount semaphore
222 * on it to make sure nobody unmounts until the snapshot creation is done.
223 * The reference counter (bd_fsfreeze_count) guarantees that only the last
224 * unfreeze process can unfreeze the frozen filesystem actually when multiple
225 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
226 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
227 * actually.
228 */
229struct super_block *freeze_bdev(struct block_device *bdev)
230{
231 struct super_block *sb;
232 int error = 0;
233
234 mutex_lock(&bdev->bd_fsfreeze_mutex);
235 if (bdev->bd_fsfreeze_count > 0) {
236 bdev->bd_fsfreeze_count++;
237 sb = get_super(bdev);
238 mutex_unlock(&bdev->bd_fsfreeze_mutex);
239 return sb;
240 }
241 bdev->bd_fsfreeze_count++;
242
243 down(&bdev->bd_mount_sem);
244 sb = get_super(bdev);
245 if (sb && !(sb->s_flags & MS_RDONLY)) {
246 sb->s_frozen = SB_FREEZE_WRITE;
247 smp_wmb();
248
Jan Kara60b06802009-04-27 16:43:53 +0200249 sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100250
251 sb->s_frozen = SB_FREEZE_TRANS;
252 smp_wmb();
253
254 sync_blockdev(sb->s_bdev);
255
256 if (sb->s_op->freeze_fs) {
257 error = sb->s_op->freeze_fs(sb);
258 if (error) {
259 printk(KERN_ERR
260 "VFS:Filesystem freeze failed\n");
261 sb->s_frozen = SB_UNFROZEN;
262 drop_super(sb);
263 up(&bdev->bd_mount_sem);
264 bdev->bd_fsfreeze_count--;
265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
266 return ERR_PTR(error);
267 }
268 }
269 }
270
271 sync_blockdev(bdev);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273
274 return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */
275}
276EXPORT_SYMBOL(freeze_bdev);
277
278/**
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
282 *
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 */
285int thaw_bdev(struct block_device *bdev, struct super_block *sb)
286{
287 int error = 0;
288
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
290 if (!bdev->bd_fsfreeze_count) {
291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
292 return -EINVAL;
293 }
294
295 bdev->bd_fsfreeze_count--;
296 if (bdev->bd_fsfreeze_count > 0) {
297 if (sb)
298 drop_super(sb);
299 mutex_unlock(&bdev->bd_fsfreeze_mutex);
300 return 0;
301 }
302
303 if (sb) {
304 BUG_ON(sb->s_bdev != bdev);
305 if (!(sb->s_flags & MS_RDONLY)) {
306 if (sb->s_op->unfreeze_fs) {
307 error = sb->s_op->unfreeze_fs(sb);
308 if (error) {
309 printk(KERN_ERR
310 "VFS:Filesystem thaw failed\n");
311 sb->s_frozen = SB_FREEZE_TRANS;
312 bdev->bd_fsfreeze_count++;
313 mutex_unlock(&bdev->bd_fsfreeze_mutex);
314 return error;
315 }
316 }
317 sb->s_frozen = SB_UNFROZEN;
318 smp_wmb();
319 wake_up(&sb->s_wait_unfrozen);
320 }
321 drop_super(sb);
322 }
323
324 up(&bdev->bd_mount_sem);
325 mutex_unlock(&bdev->bd_fsfreeze_mutex);
326 return 0;
327}
328EXPORT_SYMBOL(thaw_bdev);
329
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
331{
332 return block_write_full_page(page, blkdev_get_block, wbc);
333}
334
335static int blkdev_readpage(struct file * file, struct page * page)
336{
337 return block_read_full_page(page, blkdev_get_block);
338}
339
Nick Piggin6272b5a2007-10-16 01:25:04 -0700340static int blkdev_write_begin(struct file *file, struct address_space *mapping,
341 loff_t pos, unsigned len, unsigned flags,
342 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700344 *pagep = NULL;
345 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
346 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
Nick Piggin6272b5a2007-10-16 01:25:04 -0700349static int blkdev_write_end(struct file *file, struct address_space *mapping,
350 loff_t pos, unsigned len, unsigned copied,
351 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700353 int ret;
354 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
355
356 unlock_page(page);
357 page_cache_release(page);
358
359 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360}
361
362/*
363 * private llseek:
Josef "Jeff" Sipek0f7fc9e42006-12-08 02:36:35 -0800364 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 * so we compute the size by hand (just as in block_read/write above)
366 */
367static loff_t block_llseek(struct file *file, loff_t offset, int origin)
368{
369 struct inode *bd_inode = file->f_mapping->host;
370 loff_t size;
371 loff_t retval;
372
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800373 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 size = i_size_read(bd_inode);
375
376 switch (origin) {
377 case 2:
378 offset += size;
379 break;
380 case 1:
381 offset += file->f_pos;
382 }
383 retval = -EINVAL;
384 if (offset >= 0 && offset <= size) {
385 if (offset != file->f_pos) {
386 file->f_pos = offset;
387 }
388 retval = offset;
389 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800390 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 return retval;
392}
393
394/*
395 * Filp is never NULL; the only case when ->fsync() is called with
396 * NULL first argument is nfsd_sync_dir() and that's not a directory.
397 */
398
399static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
400{
401 return sync_blockdev(I_BDEV(filp->f_mapping->host));
402}
403
404/*
405 * pseudo-fs
406 */
407
408static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800409static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411static struct inode *bdev_alloc_inode(struct super_block *sb)
412{
Christoph Lametere94b1762006-12-06 20:33:17 -0800413 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (!ei)
415 return NULL;
416 return &ei->vfs_inode;
417}
418
419static void bdev_destroy_inode(struct inode *inode)
420{
421 struct bdev_inode *bdi = BDEV_I(inode);
422
423 bdi->bdev.bd_inode_backing_dev_info = NULL;
424 kmem_cache_free(bdev_cachep, bdi);
425}
426
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700427static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
429 struct bdev_inode *ei = (struct bdev_inode *) foo;
430 struct block_device *bdev = &ei->bdev;
431
Christoph Lametera35afb82007-05-16 22:10:57 -0700432 memset(bdev, 0, sizeof(*bdev));
433 mutex_init(&bdev->bd_mutex);
434 sema_init(&bdev->bd_mount_sem, 1);
435 INIT_LIST_HEAD(&bdev->bd_inodes);
436 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800437#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700438 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800439#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700440 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800441 /* Initialize mutex for freeze. */
442 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443}
444
445static inline void __bd_forget(struct inode *inode)
446{
447 list_del_init(&inode->i_devices);
448 inode->i_bdev = NULL;
449 inode->i_mapping = &inode->i_data;
450}
451
452static void bdev_clear_inode(struct inode *inode)
453{
454 struct block_device *bdev = &BDEV_I(inode)->bdev;
455 struct list_head *p;
456 spin_lock(&bdev_lock);
457 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
458 __bd_forget(list_entry(p, struct inode, i_devices));
459 }
460 list_del_init(&bdev->bd_list);
461 spin_unlock(&bdev_lock);
462}
463
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800464static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 .statfs = simple_statfs,
466 .alloc_inode = bdev_alloc_inode,
467 .destroy_inode = bdev_destroy_inode,
468 .drop_inode = generic_delete_inode,
469 .clear_inode = bdev_clear_inode,
470};
471
David Howells454e2392006-06-23 02:02:57 -0700472static int bd_get_sb(struct file_system_type *fs_type,
473 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474{
David Howells454e2392006-06-23 02:02:57 -0700475 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
478static struct file_system_type bd_type = {
479 .name = "bdev",
480 .get_sb = bd_get_sb,
481 .kill_sb = kill_anon_super,
482};
483
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800484struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486void __init bdev_cache_init(void)
487{
488 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800489 struct vfsmount *bd_mnt;
490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800492 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
493 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900494 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 err = register_filesystem(&bd_type);
496 if (err)
497 panic("Cannot register bdev pseudo-fs");
498 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if (IS_ERR(bd_mnt))
500 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100501 /*
502 * This vfsmount structure is only used to obtain the
503 * blockdev_superblock, so tell kmemleak not to report it.
504 */
505 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
507}
508
509/*
510 * Most likely _very_ bad one - but then it's hardly critical for small
511 * /dev and can be fixed when somebody will need really large one.
512 * Keep in mind that it will be fed through icache hash function too.
513 */
514static inline unsigned long hash(dev_t dev)
515{
516 return MAJOR(dev)+MINOR(dev);
517}
518
519static int bdev_test(struct inode *inode, void *data)
520{
521 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
522}
523
524static int bdev_set(struct inode *inode, void *data)
525{
526 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
527 return 0;
528}
529
530static LIST_HEAD(all_bdevs);
531
532struct block_device *bdget(dev_t dev)
533{
534 struct block_device *bdev;
535 struct inode *inode;
536
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800537 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 bdev_test, bdev_set, &dev);
539
540 if (!inode)
541 return NULL;
542
543 bdev = &BDEV_I(inode)->bdev;
544
545 if (inode->i_state & I_NEW) {
546 bdev->bd_contains = NULL;
547 bdev->bd_inode = inode;
548 bdev->bd_block_size = (1 << inode->i_blkbits);
549 bdev->bd_part_count = 0;
550 bdev->bd_invalidated = 0;
551 inode->i_mode = S_IFBLK;
552 inode->i_rdev = dev;
553 inode->i_bdev = bdev;
554 inode->i_data.a_ops = &def_blk_aops;
555 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
556 inode->i_data.backing_dev_info = &default_backing_dev_info;
557 spin_lock(&bdev_lock);
558 list_add(&bdev->bd_list, &all_bdevs);
559 spin_unlock(&bdev_lock);
560 unlock_new_inode(inode);
561 }
562 return bdev;
563}
564
565EXPORT_SYMBOL(bdget);
566
567long nr_blockdev_pages(void)
568{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700569 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 long ret = 0;
571 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700572 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 ret += bdev->bd_inode->i_mapping->nrpages;
574 }
575 spin_unlock(&bdev_lock);
576 return ret;
577}
578
579void bdput(struct block_device *bdev)
580{
581 iput(bdev->bd_inode);
582}
583
584EXPORT_SYMBOL(bdput);
585
586static struct block_device *bd_acquire(struct inode *inode)
587{
588 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700589
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 spin_lock(&bdev_lock);
591 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700592 if (bdev) {
593 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 spin_unlock(&bdev_lock);
595 return bdev;
596 }
597 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700598
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 bdev = bdget(inode->i_rdev);
600 if (bdev) {
601 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700602 if (!inode->i_bdev) {
603 /*
604 * We take an additional bd_inode->i_count for inode,
605 * and it's released in clear_inode() of inode.
606 * So, we can access it via ->i_mapping always
607 * without igrab().
608 */
609 atomic_inc(&bdev->bd_inode->i_count);
610 inode->i_bdev = bdev;
611 inode->i_mapping = bdev->bd_inode->i_mapping;
612 list_add(&inode->i_devices, &bdev->bd_inodes);
613 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 spin_unlock(&bdev_lock);
615 }
616 return bdev;
617}
618
619/* Call when you free inode */
620
621void bd_forget(struct inode *inode)
622{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700623 struct block_device *bdev = NULL;
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700626 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800627 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700628 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700630 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700632
633 if (bdev)
634 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}
636
637int bd_claim(struct block_device *bdev, void *holder)
638{
639 int res;
640 spin_lock(&bdev_lock);
641
642 /* first decide result */
643 if (bdev->bd_holder == holder)
644 res = 0; /* already a holder */
645 else if (bdev->bd_holder != NULL)
646 res = -EBUSY; /* held by someone else */
647 else if (bdev->bd_contains == bdev)
648 res = 0; /* is a whole device which isn't held */
649
650 else if (bdev->bd_contains->bd_holder == bd_claim)
651 res = 0; /* is a partition of a device that is being partitioned */
652 else if (bdev->bd_contains->bd_holder != NULL)
653 res = -EBUSY; /* is a partition of a held device */
654 else
655 res = 0; /* is a partition of an un-held device */
656
657 /* now impose change */
658 if (res==0) {
659 /* note that for a whole device bd_holders
660 * will be incremented twice, and bd_holder will
661 * be set to bd_claim before being set to holder
662 */
663 bdev->bd_contains->bd_holders ++;
664 bdev->bd_contains->bd_holder = bd_claim;
665 bdev->bd_holders++;
666 bdev->bd_holder = holder;
667 }
668 spin_unlock(&bdev_lock);
669 return res;
670}
671
672EXPORT_SYMBOL(bd_claim);
673
674void bd_release(struct block_device *bdev)
675{
676 spin_lock(&bdev_lock);
677 if (!--bdev->bd_contains->bd_holders)
678 bdev->bd_contains->bd_holder = NULL;
679 if (!--bdev->bd_holders)
680 bdev->bd_holder = NULL;
681 spin_unlock(&bdev_lock);
682}
683
684EXPORT_SYMBOL(bd_release);
685
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800686#ifdef CONFIG_SYSFS
687/*
688 * Functions for bd_claim_by_kobject / bd_release_from_kobject
689 *
690 * If a kobject is passed to bd_claim_by_kobject()
691 * and the kobject has a parent directory,
692 * following symlinks are created:
693 * o from the kobject to the claimed bdev
694 * o from "holders" directory of the bdev to the parent of the kobject
695 * bd_release_from_kobject() removes these symlinks.
696 *
697 * Example:
698 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
699 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
700 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
701 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
702 */
703
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700704static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800705{
706 if (!from || !to)
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700707 return 0;
708 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800709}
710
711static void del_symlink(struct kobject *from, struct kobject *to)
712{
713 if (!from || !to)
714 return;
715 sysfs_remove_link(from, kobject_name(to));
716}
717
718/*
719 * 'struct bd_holder' contains pointers to kobjects symlinked by
720 * bd_claim_by_kobject.
721 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
722 */
723struct bd_holder {
724 struct list_head list; /* chain of holders of the bdev */
725 int count; /* references from the holder */
726 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
727 struct kobject *hdev; /* e.g. "/block/dm-0" */
728 struct kobject *hdir; /* e.g. "/block/sda/holders" */
729 struct kobject *sdev; /* e.g. "/block/sda" */
730};
731
732/*
733 * Get references of related kobjects at once.
734 * Returns 1 on success. 0 on failure.
735 *
736 * Should call bd_holder_release_dirs() after successful use.
737 */
738static int bd_holder_grab_dirs(struct block_device *bdev,
739 struct bd_holder *bo)
740{
741 if (!bdev || !bo)
742 return 0;
743
744 bo->sdir = kobject_get(bo->sdir);
745 if (!bo->sdir)
746 return 0;
747
748 bo->hdev = kobject_get(bo->sdir->parent);
749 if (!bo->hdev)
750 goto fail_put_sdir;
751
Tejun Heo0762b8b2008-08-25 19:56:12 +0900752 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800753 if (!bo->sdev)
754 goto fail_put_hdev;
755
Tejun Heo4c465012008-08-25 19:56:11 +0900756 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800757 if (!bo->hdir)
758 goto fail_put_sdev;
759
760 return 1;
761
762fail_put_sdev:
763 kobject_put(bo->sdev);
764fail_put_hdev:
765 kobject_put(bo->hdev);
766fail_put_sdir:
767 kobject_put(bo->sdir);
768
769 return 0;
770}
771
772/* Put references of related kobjects at once. */
773static void bd_holder_release_dirs(struct bd_holder *bo)
774{
775 kobject_put(bo->hdir);
776 kobject_put(bo->sdev);
777 kobject_put(bo->hdev);
778 kobject_put(bo->sdir);
779}
780
781static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
782{
783 struct bd_holder *bo;
784
785 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
786 if (!bo)
787 return NULL;
788
789 bo->count = 1;
790 bo->sdir = kobj;
791
792 return bo;
793}
794
795static void free_bd_holder(struct bd_holder *bo)
796{
797 kfree(bo);
798}
799
800/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500801 * find_bd_holder - find matching struct bd_holder from the block device
802 *
803 * @bdev: struct block device to be searched
804 * @bo: target struct bd_holder
805 *
806 * Returns matching entry with @bo in @bdev->bd_holder_list.
807 * If found, increment the reference count and return the pointer.
808 * If not found, returns NULL.
809 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800810static struct bd_holder *find_bd_holder(struct block_device *bdev,
811 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500812{
813 struct bd_holder *tmp;
814
815 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
816 if (tmp->sdir == bo->sdir) {
817 tmp->count++;
818 return tmp;
819 }
820
821 return NULL;
822}
823
824/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800825 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
826 *
827 * @bdev: block device to be bd_claimed
828 * @bo: preallocated and initialized by alloc_bd_holder()
829 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500830 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800831 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500832 * Returns 0 if symlinks are created.
833 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800834 */
835static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
836{
Johannes Weiner4e916722007-07-15 23:41:25 -0700837 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800838
839 if (!bo)
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700840 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800841
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800842 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700843 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800844
Johannes Weiner4e916722007-07-15 23:41:25 -0700845 err = add_symlink(bo->sdir, bo->sdev);
846 if (err)
847 return err;
848
849 err = add_symlink(bo->hdir, bo->hdev);
850 if (err) {
851 del_symlink(bo->sdir, bo->sdev);
852 return err;
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700853 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700854
855 list_add_tail(&bo->list, &bdev->bd_holder_list);
856 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800857}
858
859/**
860 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
861 *
862 * @bdev: block device to be bd_claimed
863 * @kobj: holder's kobject
864 *
865 * If there is matching entry with @kobj in @bdev->bd_holder_list
866 * and no other bd_claim() from the same kobject,
867 * remove the struct bd_holder from the list, delete symlinks for it.
868 *
869 * Returns a pointer to the struct bd_holder when it's removed from the list
870 * and ready to be freed.
871 * Returns NULL if matching claim isn't found or there is other bd_claim()
872 * by the same kobject.
873 */
874static struct bd_holder *del_bd_holder(struct block_device *bdev,
875 struct kobject *kobj)
876{
877 struct bd_holder *bo;
878
879 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
880 if (bo->sdir == kobj) {
881 bo->count--;
882 BUG_ON(bo->count < 0);
883 if (!bo->count) {
884 list_del(&bo->list);
885 del_symlink(bo->sdir, bo->sdev);
886 del_symlink(bo->hdir, bo->hdev);
887 bd_holder_release_dirs(bo);
888 return bo;
889 }
890 break;
891 }
892 }
893
894 return NULL;
895}
896
897/**
898 * bd_claim_by_kobject - bd_claim() with additional kobject signature
899 *
900 * @bdev: block device to be claimed
901 * @holder: holder's signature
902 * @kobj: holder's kobject
903 *
904 * Do bd_claim() and if it succeeds, create sysfs symlinks between
905 * the bdev and the holder's kobject.
906 * Use bd_release_from_kobject() when relesing the claimed bdev.
907 *
908 * Returns 0 on success. (same as bd_claim())
909 * Returns errno on failure.
910 */
911static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
912 struct kobject *kobj)
913{
Johannes Weiner4e916722007-07-15 23:41:25 -0700914 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500915 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800916
917 if (!kobj)
918 return -EINVAL;
919
920 bo = alloc_bd_holder(kobj);
921 if (!bo)
922 return -ENOMEM;
923
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800924 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500925
Johannes Weiner4e916722007-07-15 23:41:25 -0700926 err = bd_claim(bdev, holder);
927 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700928 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700929
930 found = find_bd_holder(bdev, bo);
931 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700932 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700933
934 err = add_bd_holder(bdev, bo);
935 if (err)
936 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700937 else
938 bo = NULL;
939fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800940 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700941 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700942 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800943}
944
945/**
946 * bd_release_from_kobject - bd_release() with additional kobject signature
947 *
948 * @bdev: block device to be released
949 * @kobj: holder's kobject
950 *
951 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
952 */
953static void bd_release_from_kobject(struct block_device *bdev,
954 struct kobject *kobj)
955{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800956 if (!kobj)
957 return;
958
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800959 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800960 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700961 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800962 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800963}
964
965/**
966 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
967 *
968 * @bdev: block device to be claimed
969 * @holder: holder's signature
970 * @disk: holder's gendisk
971 *
972 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
973 */
974int bd_claim_by_disk(struct block_device *bdev, void *holder,
975 struct gendisk *disk)
976{
977 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
978}
979EXPORT_SYMBOL_GPL(bd_claim_by_disk);
980
981/**
982 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
983 *
984 * @bdev: block device to be claimed
985 * @disk: holder's gendisk
986 *
987 * Call bd_release_from_kobject() and put @disk->slave_dir.
988 */
989void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
990{
991 bd_release_from_kobject(bdev, disk->slave_dir);
992 kobject_put(disk->slave_dir);
993}
994EXPORT_SYMBOL_GPL(bd_release_from_disk);
995#endif
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997/*
998 * Tries to open block device by device number. Use it ONLY if you
999 * really do not have anything better - i.e. when you are behind a
1000 * truly sucky interface and all you are given is a device number. _Never_
1001 * to be used for internal purposes. If you ever need it - reconsider
1002 * your API.
1003 */
Al Viroaeb5d722008-09-02 15:28:45 -04001004struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005{
1006 struct block_device *bdev = bdget(dev);
1007 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001009 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 return err ? ERR_PTR(err) : bdev;
1011}
1012
1013EXPORT_SYMBOL(open_by_devnum);
1014
Andrew Patterson0c002c22008-09-04 14:27:20 -06001015/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001016 * flush_disk - invalidates all buffer-cache entries on a disk
1017 *
1018 * @bdev: struct block device to be flushed
1019 *
1020 * Invalidates all buffer-cache entries on a disk. It should be called
1021 * when a disk has been changed -- either by a media change or online
1022 * resize.
1023 */
1024static void flush_disk(struct block_device *bdev)
1025{
1026 if (__invalidate_device(bdev)) {
1027 char name[BDEVNAME_SIZE] = "";
1028
1029 if (bdev->bd_disk)
1030 disk_name(bdev->bd_disk, 0, name);
1031 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1032 "resized disk %s\n", name);
1033 }
1034
1035 if (!bdev->bd_disk)
1036 return;
1037 if (disk_partitionable(bdev->bd_disk))
1038 bdev->bd_invalidated = 1;
1039}
1040
1041/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001042 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001043 * @disk: struct gendisk to check
1044 * @bdev: struct bdev to adjust.
1045 *
1046 * This routine checks to see if the bdev size does not match the disk size
1047 * and adjusts it if it differs.
1048 */
1049void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1050{
1051 loff_t disk_size, bdev_size;
1052
1053 disk_size = (loff_t)get_capacity(disk) << 9;
1054 bdev_size = i_size_read(bdev->bd_inode);
1055 if (disk_size != bdev_size) {
1056 char name[BDEVNAME_SIZE];
1057
1058 disk_name(disk, 0, name);
1059 printk(KERN_INFO
1060 "%s: detected capacity change from %lld to %lld\n",
1061 name, bdev_size, disk_size);
1062 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001063 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001064 }
1065}
1066EXPORT_SYMBOL(check_disk_size_change);
1067
1068/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001069 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001070 * @disk: struct gendisk to be revalidated
1071 *
1072 * This routine is a wrapper for lower-level driver's revalidate_disk
1073 * call-backs. It is used to do common pre and post operations needed
1074 * for all revalidate_disk operations.
1075 */
1076int revalidate_disk(struct gendisk *disk)
1077{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001078 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001079 int ret = 0;
1080
1081 if (disk->fops->revalidate_disk)
1082 ret = disk->fops->revalidate_disk(disk);
1083
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001084 bdev = bdget_disk(disk, 0);
1085 if (!bdev)
1086 return ret;
1087
1088 mutex_lock(&bdev->bd_mutex);
1089 check_disk_size_change(disk, bdev);
1090 mutex_unlock(&bdev->bd_mutex);
1091 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001092 return ret;
1093}
1094EXPORT_SYMBOL(revalidate_disk);
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096/*
1097 * This routine checks whether a removable media has been changed,
1098 * and invalidates all buffer-cache-entries in that case. This
1099 * is a relatively slow routine, so we have to try to minimize using
1100 * it. Thus it is called only upon a 'mount' or 'open'. This
1101 * is the best way of combining speed and utility, I think.
1102 * People changing diskettes in the middle of an operation deserve
1103 * to lose :-)
1104 */
1105int check_disk_change(struct block_device *bdev)
1106{
1107 struct gendisk *disk = bdev->bd_disk;
1108 struct block_device_operations * bdops = disk->fops;
1109
1110 if (!bdops->media_changed)
1111 return 0;
1112 if (!bdops->media_changed(bdev->bd_disk))
1113 return 0;
1114
Andrew Patterson56ade442008-09-04 14:27:40 -06001115 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (bdops->revalidate_disk)
1117 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 return 1;
1119}
1120
1121EXPORT_SYMBOL(check_disk_change);
1122
1123void bd_set_size(struct block_device *bdev, loff_t size)
1124{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001125 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 bdev->bd_inode->i_size = size;
1128 while (bsize < PAGE_CACHE_SIZE) {
1129 if (size & bsize)
1130 break;
1131 bsize <<= 1;
1132 }
1133 bdev->bd_block_size = bsize;
1134 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1135}
1136EXPORT_SYMBOL(bd_set_size);
1137
Al Viro9a1c3542008-02-22 20:40:24 -05001138static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001139
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001140/*
1141 * bd_mutex locking:
1142 *
1143 * mutex_lock(part->bd_mutex)
1144 * mutex_lock_nested(whole->bd_mutex, 1)
1145 */
1146
Al Viro572c4892007-10-08 13:24:05 -04001147static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001150 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001151 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001152 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153
Al Viro572c4892007-10-08 13:24:05 -04001154 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001155 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001156 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001157 perm |= MAY_WRITE;
1158 /*
1159 * hooks: /n/, see "layering violations".
1160 */
1161 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001162 if (ret != 0) {
1163 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001164 return ret;
Al Viro82666022008-08-01 05:32:04 -04001165 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 lock_kernel();
NeilBrownd3374822009-01-09 08:31:10 +11001168 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001169
Tejun Heo89f97492008-11-05 10:21:06 +01001170 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001171 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001172 if (!disk)
1173 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
NeilBrown6796bf52006-12-08 02:36:16 -08001175 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 if (!bdev->bd_openers) {
1177 bdev->bd_disk = disk;
1178 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001179 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001181
1182 ret = -ENXIO;
1183 bdev->bd_part = disk_get_part(disk, partno);
1184 if (!bdev->bd_part)
1185 goto out_clear;
1186
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001188 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001189 if (ret == -ERESTARTSYS) {
1190 /* Lost a race with 'disk' being
1191 * deleted, try again.
1192 * See md.c
1193 */
1194 disk_put_part(bdev->bd_part);
1195 bdev->bd_part = NULL;
1196 module_put(disk->fops->owner);
1197 put_disk(disk);
1198 bdev->bd_disk = NULL;
1199 mutex_unlock(&bdev->bd_mutex);
1200 goto restart;
1201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001203 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 }
1205 if (!bdev->bd_openers) {
1206 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1207 bdi = blk_get_backing_dev_info(bdev);
1208 if (bdi == NULL)
1209 bdi = &default_backing_dev_info;
1210 bdev->bd_inode->i_data.backing_dev_info = bdi;
1211 }
1212 if (bdev->bd_invalidated)
1213 rescan_partitions(disk, bdev);
1214 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 struct block_device *whole;
1216 whole = bdget_disk(disk, 0);
1217 ret = -ENOMEM;
1218 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001219 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001220 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001221 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001223 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 bdev->bd_inode->i_data.backing_dev_info =
1226 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001227 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001228 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001229 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001231 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 }
Tejun Heo89f97492008-11-05 10:21:06 +01001233 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235 } else {
1236 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001237 module_put(disk->fops->owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001238 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (bdev->bd_contains == bdev) {
1240 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001241 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001243 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245 if (bdev->bd_invalidated)
1246 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248 }
1249 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001250 if (for_part)
1251 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001252 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 unlock_kernel();
1254 return 0;
1255
Tejun Heo0762b8b2008-08-25 19:56:12 +09001256 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001257 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001259 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1261 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001262 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001264 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001265 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001266 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001268
Tejun Heo0762b8b2008-08-25 19:56:12 +09001269 if (disk)
1270 module_put(disk->fops->owner);
1271 put_disk(disk);
1272 bdput(bdev);
1273
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 return ret;
1275}
1276
Al Viro572c4892007-10-08 13:24:05 -04001277int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278{
Al Viro572c4892007-10-08 13:24:05 -04001279 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001280}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281EXPORT_SYMBOL(blkdev_get);
1282
1283static int blkdev_open(struct inode * inode, struct file * filp)
1284{
1285 struct block_device *bdev;
1286 int res;
1287
1288 /*
1289 * Preserve backwards compatibility and allow large file access
1290 * even if userspace doesn't ask for it explicitly. Some mkfs
1291 * binary needs it. We might want to drop this workaround
1292 * during an unstable branch.
1293 */
1294 filp->f_flags |= O_LARGEFILE;
1295
Al Viro572c4892007-10-08 13:24:05 -04001296 if (filp->f_flags & O_NDELAY)
1297 filp->f_mode |= FMODE_NDELAY;
1298 if (filp->f_flags & O_EXCL)
1299 filp->f_mode |= FMODE_EXCL;
1300 if ((filp->f_flags & O_ACCMODE) == 3)
1301 filp->f_mode |= FMODE_WRITE_IOCTL;
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001304 if (bdev == NULL)
1305 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Al Viro572c4892007-10-08 13:24:05 -04001307 filp->f_mapping = bdev->bd_inode->i_mapping;
1308
1309 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 if (res)
1311 return res;
1312
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001313 if (filp->f_mode & FMODE_EXCL) {
1314 res = bd_claim(bdev, filp);
1315 if (res)
1316 goto out_blkdev_put;
1317 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001319 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001321 out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001322 blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return res;
1324}
1325
Al Viro9a1c3542008-02-22 20:40:24 -05001326static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001327{
1328 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001329 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001330 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001331
NeilBrown6796bf52006-12-08 02:36:16 -08001332 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001333 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001334 if (for_part)
1335 bdev->bd_part_count--;
1336
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001337 if (!--bdev->bd_openers) {
1338 sync_blockdev(bdev);
1339 kill_bdev(bdev);
1340 }
1341 if (bdev->bd_contains == bdev) {
1342 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001343 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001344 }
1345 if (!bdev->bd_openers) {
1346 struct module *owner = disk->fops->owner;
1347
1348 put_disk(disk);
1349 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001350 disk_put_part(bdev->bd_part);
1351 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001352 bdev->bd_disk = NULL;
1353 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001354 if (bdev != bdev->bd_contains)
1355 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001356 bdev->bd_contains = NULL;
1357 }
1358 unlock_kernel();
1359 mutex_unlock(&bdev->bd_mutex);
1360 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001361 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001362 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001363 return ret;
1364}
1365
Al Viro9a1c3542008-02-22 20:40:24 -05001366int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001367{
Al Viro9a1c3542008-02-22 20:40:24 -05001368 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001369}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001370EXPORT_SYMBOL(blkdev_put);
1371
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372static int blkdev_close(struct inode * inode, struct file * filp)
1373{
1374 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1375 if (bdev->bd_holder == filp)
1376 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001377 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001380static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381{
Al Viro56b26ad2008-09-19 03:17:36 -04001382 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1383 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001384
1385 /*
1386 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1387 * to updated it before every ioctl.
1388 */
Al Viro56b26ad2008-09-19 03:17:36 -04001389 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001390 mode |= FMODE_NDELAY;
1391 else
1392 mode &= ~FMODE_NDELAY;
1393
Al Viro56b26ad2008-09-19 03:17:36 -04001394 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395}
1396
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001397/*
1398 * Try to release a page associated with block device when the system
1399 * is under memory pressure.
1400 */
1401static int blkdev_releasepage(struct page *page, gfp_t wait)
1402{
1403 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1404
1405 if (super && super->s_op->bdev_try_to_free_page)
1406 return super->s_op->bdev_try_to_free_page(super, page, wait);
1407
1408 return try_to_free_buffers(page);
1409}
1410
Adrian Bunk4c54ac622008-02-18 13:48:31 +01001411static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 .readpage = blkdev_readpage,
1413 .writepage = blkdev_writepage,
1414 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001415 .write_begin = blkdev_write_begin,
1416 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001418 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 .direct_IO = blkdev_direct_IO,
1420};
1421
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001422const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 .open = blkdev_open,
1424 .release = blkdev_close,
1425 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001426 .read = do_sync_read,
1427 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001429 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 .mmap = generic_file_mmap,
1431 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001432 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433#ifdef CONFIG_COMPAT
1434 .compat_ioctl = compat_blkdev_ioctl,
1435#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001436 .splice_read = generic_file_splice_read,
1437 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438};
1439
1440int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1441{
1442 int res;
1443 mm_segment_t old_fs = get_fs();
1444 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001445 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 set_fs(old_fs);
1447 return res;
1448}
1449
1450EXPORT_SYMBOL(ioctl_by_bdev);
1451
1452/**
1453 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001454 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001456 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 * namespace if possible and return it. Return ERR_PTR(error)
1458 * otherwise.
1459 */
Al Viro421748e2008-08-02 01:04:36 -04001460struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
1462 struct block_device *bdev;
1463 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001464 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 int error;
1466
Al Viro421748e2008-08-02 01:04:36 -04001467 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return ERR_PTR(-EINVAL);
1469
Al Viro421748e2008-08-02 01:04:36 -04001470 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 if (error)
1472 return ERR_PTR(error);
1473
Al Viro421748e2008-08-02 01:04:36 -04001474 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 error = -ENOTBLK;
1476 if (!S_ISBLK(inode->i_mode))
1477 goto fail;
1478 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001479 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 goto fail;
1481 error = -ENOMEM;
1482 bdev = bd_acquire(inode);
1483 if (!bdev)
1484 goto fail;
1485out:
Al Viro421748e2008-08-02 01:04:36 -04001486 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 return bdev;
1488fail:
1489 bdev = ERR_PTR(error);
1490 goto out;
1491}
Al Virod5686b42008-08-01 05:00:11 -04001492EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494/**
Al Viro30c40d22008-02-22 19:50:45 -05001495 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 *
1497 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001498 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 * @holder: owner for exclusion
1500 *
1501 * Open the blockdevice described by the special file at @path, claim it
1502 * for the @holder.
1503 */
Al Viro30c40d22008-02-22 19:50:45 -05001504struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
1506 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 int error = 0;
1508
1509 bdev = lookup_bdev(path);
1510 if (IS_ERR(bdev))
1511 return bdev;
1512
Al Viro572c4892007-10-08 13:24:05 -04001513 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 if (error)
1515 return ERR_PTR(error);
1516 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001517 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 goto blkdev_put;
1519 error = bd_claim(bdev, holder);
1520 if (error)
1521 goto blkdev_put;
1522
1523 return bdev;
1524
1525blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001526 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 return ERR_PTR(error);
1528}
1529
Al Viro30c40d22008-02-22 19:50:45 -05001530EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531
1532/**
Al Viro30c40d22008-02-22 19:50:45 -05001533 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 *
1535 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001536 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 *
Al Viro30c40d22008-02-22 19:50:45 -05001538 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 */
Al Viro30c40d22008-02-22 19:50:45 -05001540void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
1542 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001543 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544}
1545
Al Viro30c40d22008-02-22 19:50:45 -05001546EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001547
1548int __invalidate_device(struct block_device *bdev)
1549{
1550 struct super_block *sb = get_super(bdev);
1551 int res = 0;
1552
1553 if (sb) {
1554 /*
1555 * no need to lock the super, get_super holds the
1556 * read mutex so the filesystem cannot go away
1557 * under us (->put_super runs with the write lock
1558 * hold).
1559 */
1560 shrink_dcache_sb(sb);
1561 res = invalidate_inodes(sb);
1562 drop_super(sb);
1563 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001564 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001565 return res;
1566}
1567EXPORT_SYMBOL(__invalidate_device);