blob: 94dfda24c06e5ea08b7df4883b640db573d137e7 [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 Bunk4c54ac62008-02-18 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" Sipek0f7fc9e2006-12-08 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
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200567/**
568 * bdgrab -- Grab a reference to an already referenced block device
569 * @bdev: Block device to grab a reference to.
570 */
571struct block_device *bdgrab(struct block_device *bdev)
572{
573 atomic_inc(&bdev->bd_inode->i_count);
574 return bdev;
575}
576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577long nr_blockdev_pages(void)
578{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700579 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 long ret = 0;
581 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700582 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 ret += bdev->bd_inode->i_mapping->nrpages;
584 }
585 spin_unlock(&bdev_lock);
586 return ret;
587}
588
589void bdput(struct block_device *bdev)
590{
591 iput(bdev->bd_inode);
592}
593
594EXPORT_SYMBOL(bdput);
595
596static struct block_device *bd_acquire(struct inode *inode)
597{
598 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 spin_lock(&bdev_lock);
601 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700602 if (bdev) {
603 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 spin_unlock(&bdev_lock);
605 return bdev;
606 }
607 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700608
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 bdev = bdget(inode->i_rdev);
610 if (bdev) {
611 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700612 if (!inode->i_bdev) {
613 /*
614 * We take an additional bd_inode->i_count for inode,
615 * and it's released in clear_inode() of inode.
616 * So, we can access it via ->i_mapping always
617 * without igrab().
618 */
619 atomic_inc(&bdev->bd_inode->i_count);
620 inode->i_bdev = bdev;
621 inode->i_mapping = bdev->bd_inode->i_mapping;
622 list_add(&inode->i_devices, &bdev->bd_inodes);
623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 spin_unlock(&bdev_lock);
625 }
626 return bdev;
627}
628
629/* Call when you free inode */
630
631void bd_forget(struct inode *inode)
632{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700633 struct block_device *bdev = NULL;
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700636 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800637 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700638 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700640 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700642
643 if (bdev)
644 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}
646
647int bd_claim(struct block_device *bdev, void *holder)
648{
649 int res;
650 spin_lock(&bdev_lock);
651
652 /* first decide result */
653 if (bdev->bd_holder == holder)
654 res = 0; /* already a holder */
655 else if (bdev->bd_holder != NULL)
656 res = -EBUSY; /* held by someone else */
657 else if (bdev->bd_contains == bdev)
658 res = 0; /* is a whole device which isn't held */
659
660 else if (bdev->bd_contains->bd_holder == bd_claim)
661 res = 0; /* is a partition of a device that is being partitioned */
662 else if (bdev->bd_contains->bd_holder != NULL)
663 res = -EBUSY; /* is a partition of a held device */
664 else
665 res = 0; /* is a partition of an un-held device */
666
667 /* now impose change */
668 if (res==0) {
669 /* note that for a whole device bd_holders
670 * will be incremented twice, and bd_holder will
671 * be set to bd_claim before being set to holder
672 */
673 bdev->bd_contains->bd_holders ++;
674 bdev->bd_contains->bd_holder = bd_claim;
675 bdev->bd_holders++;
676 bdev->bd_holder = holder;
677 }
678 spin_unlock(&bdev_lock);
679 return res;
680}
681
682EXPORT_SYMBOL(bd_claim);
683
684void bd_release(struct block_device *bdev)
685{
686 spin_lock(&bdev_lock);
687 if (!--bdev->bd_contains->bd_holders)
688 bdev->bd_contains->bd_holder = NULL;
689 if (!--bdev->bd_holders)
690 bdev->bd_holder = NULL;
691 spin_unlock(&bdev_lock);
692}
693
694EXPORT_SYMBOL(bd_release);
695
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800696#ifdef CONFIG_SYSFS
697/*
698 * Functions for bd_claim_by_kobject / bd_release_from_kobject
699 *
700 * If a kobject is passed to bd_claim_by_kobject()
701 * and the kobject has a parent directory,
702 * following symlinks are created:
703 * o from the kobject to the claimed bdev
704 * o from "holders" directory of the bdev to the parent of the kobject
705 * bd_release_from_kobject() removes these symlinks.
706 *
707 * Example:
708 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
709 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
710 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
711 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
712 */
713
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700714static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800715{
716 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700717 return 0;
718 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800719}
720
721static void del_symlink(struct kobject *from, struct kobject *to)
722{
723 if (!from || !to)
724 return;
725 sysfs_remove_link(from, kobject_name(to));
726}
727
728/*
729 * 'struct bd_holder' contains pointers to kobjects symlinked by
730 * bd_claim_by_kobject.
731 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
732 */
733struct bd_holder {
734 struct list_head list; /* chain of holders of the bdev */
735 int count; /* references from the holder */
736 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
737 struct kobject *hdev; /* e.g. "/block/dm-0" */
738 struct kobject *hdir; /* e.g. "/block/sda/holders" */
739 struct kobject *sdev; /* e.g. "/block/sda" */
740};
741
742/*
743 * Get references of related kobjects at once.
744 * Returns 1 on success. 0 on failure.
745 *
746 * Should call bd_holder_release_dirs() after successful use.
747 */
748static int bd_holder_grab_dirs(struct block_device *bdev,
749 struct bd_holder *bo)
750{
751 if (!bdev || !bo)
752 return 0;
753
754 bo->sdir = kobject_get(bo->sdir);
755 if (!bo->sdir)
756 return 0;
757
758 bo->hdev = kobject_get(bo->sdir->parent);
759 if (!bo->hdev)
760 goto fail_put_sdir;
761
Tejun Heo0762b8b2008-08-25 19:56:12 +0900762 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800763 if (!bo->sdev)
764 goto fail_put_hdev;
765
Tejun Heo4c465012008-08-25 19:56:11 +0900766 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800767 if (!bo->hdir)
768 goto fail_put_sdev;
769
770 return 1;
771
772fail_put_sdev:
773 kobject_put(bo->sdev);
774fail_put_hdev:
775 kobject_put(bo->hdev);
776fail_put_sdir:
777 kobject_put(bo->sdir);
778
779 return 0;
780}
781
782/* Put references of related kobjects at once. */
783static void bd_holder_release_dirs(struct bd_holder *bo)
784{
785 kobject_put(bo->hdir);
786 kobject_put(bo->sdev);
787 kobject_put(bo->hdev);
788 kobject_put(bo->sdir);
789}
790
791static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
792{
793 struct bd_holder *bo;
794
795 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
796 if (!bo)
797 return NULL;
798
799 bo->count = 1;
800 bo->sdir = kobj;
801
802 return bo;
803}
804
805static void free_bd_holder(struct bd_holder *bo)
806{
807 kfree(bo);
808}
809
810/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500811 * find_bd_holder - find matching struct bd_holder from the block device
812 *
813 * @bdev: struct block device to be searched
814 * @bo: target struct bd_holder
815 *
816 * Returns matching entry with @bo in @bdev->bd_holder_list.
817 * If found, increment the reference count and return the pointer.
818 * If not found, returns NULL.
819 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800820static struct bd_holder *find_bd_holder(struct block_device *bdev,
821 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500822{
823 struct bd_holder *tmp;
824
825 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
826 if (tmp->sdir == bo->sdir) {
827 tmp->count++;
828 return tmp;
829 }
830
831 return NULL;
832}
833
834/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800835 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
836 *
837 * @bdev: block device to be bd_claimed
838 * @bo: preallocated and initialized by alloc_bd_holder()
839 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500840 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800841 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500842 * Returns 0 if symlinks are created.
843 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800844 */
845static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
846{
Johannes Weiner4e916722007-07-15 23:41:25 -0700847 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800848
849 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700850 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800851
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800852 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700853 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800854
Johannes Weiner4e916722007-07-15 23:41:25 -0700855 err = add_symlink(bo->sdir, bo->sdev);
856 if (err)
857 return err;
858
859 err = add_symlink(bo->hdir, bo->hdev);
860 if (err) {
861 del_symlink(bo->sdir, bo->sdev);
862 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700863 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700864
865 list_add_tail(&bo->list, &bdev->bd_holder_list);
866 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800867}
868
869/**
870 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
871 *
872 * @bdev: block device to be bd_claimed
873 * @kobj: holder's kobject
874 *
875 * If there is matching entry with @kobj in @bdev->bd_holder_list
876 * and no other bd_claim() from the same kobject,
877 * remove the struct bd_holder from the list, delete symlinks for it.
878 *
879 * Returns a pointer to the struct bd_holder when it's removed from the list
880 * and ready to be freed.
881 * Returns NULL if matching claim isn't found or there is other bd_claim()
882 * by the same kobject.
883 */
884static struct bd_holder *del_bd_holder(struct block_device *bdev,
885 struct kobject *kobj)
886{
887 struct bd_holder *bo;
888
889 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
890 if (bo->sdir == kobj) {
891 bo->count--;
892 BUG_ON(bo->count < 0);
893 if (!bo->count) {
894 list_del(&bo->list);
895 del_symlink(bo->sdir, bo->sdev);
896 del_symlink(bo->hdir, bo->hdev);
897 bd_holder_release_dirs(bo);
898 return bo;
899 }
900 break;
901 }
902 }
903
904 return NULL;
905}
906
907/**
908 * bd_claim_by_kobject - bd_claim() with additional kobject signature
909 *
910 * @bdev: block device to be claimed
911 * @holder: holder's signature
912 * @kobj: holder's kobject
913 *
914 * Do bd_claim() and if it succeeds, create sysfs symlinks between
915 * the bdev and the holder's kobject.
916 * Use bd_release_from_kobject() when relesing the claimed bdev.
917 *
918 * Returns 0 on success. (same as bd_claim())
919 * Returns errno on failure.
920 */
921static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
922 struct kobject *kobj)
923{
Johannes Weiner4e916722007-07-15 23:41:25 -0700924 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500925 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800926
927 if (!kobj)
928 return -EINVAL;
929
930 bo = alloc_bd_holder(kobj);
931 if (!bo)
932 return -ENOMEM;
933
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800934 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500935
Johannes Weiner4e916722007-07-15 23:41:25 -0700936 err = bd_claim(bdev, holder);
937 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700938 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700939
940 found = find_bd_holder(bdev, bo);
941 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700942 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700943
944 err = add_bd_holder(bdev, bo);
945 if (err)
946 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700947 else
948 bo = NULL;
949fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800950 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700951 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700952 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800953}
954
955/**
956 * bd_release_from_kobject - bd_release() with additional kobject signature
957 *
958 * @bdev: block device to be released
959 * @kobj: holder's kobject
960 *
961 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
962 */
963static void bd_release_from_kobject(struct block_device *bdev,
964 struct kobject *kobj)
965{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800966 if (!kobj)
967 return;
968
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800969 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800970 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700971 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800972 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800973}
974
975/**
976 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
977 *
978 * @bdev: block device to be claimed
979 * @holder: holder's signature
980 * @disk: holder's gendisk
981 *
982 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
983 */
984int bd_claim_by_disk(struct block_device *bdev, void *holder,
985 struct gendisk *disk)
986{
987 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
988}
989EXPORT_SYMBOL_GPL(bd_claim_by_disk);
990
991/**
992 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
993 *
994 * @bdev: block device to be claimed
995 * @disk: holder's gendisk
996 *
997 * Call bd_release_from_kobject() and put @disk->slave_dir.
998 */
999void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1000{
1001 bd_release_from_kobject(bdev, disk->slave_dir);
1002 kobject_put(disk->slave_dir);
1003}
1004EXPORT_SYMBOL_GPL(bd_release_from_disk);
1005#endif
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007/*
1008 * Tries to open block device by device number. Use it ONLY if you
1009 * really do not have anything better - i.e. when you are behind a
1010 * truly sucky interface and all you are given is a device number. _Never_
1011 * to be used for internal purposes. If you ever need it - reconsider
1012 * your API.
1013 */
Al Viroaeb5d722008-09-02 15:28:45 -04001014struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015{
1016 struct block_device *bdev = bdget(dev);
1017 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001019 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 return err ? ERR_PTR(err) : bdev;
1021}
1022
1023EXPORT_SYMBOL(open_by_devnum);
1024
Andrew Patterson0c002c22008-09-04 14:27:20 -06001025/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001026 * flush_disk - invalidates all buffer-cache entries on a disk
1027 *
1028 * @bdev: struct block device to be flushed
1029 *
1030 * Invalidates all buffer-cache entries on a disk. It should be called
1031 * when a disk has been changed -- either by a media change or online
1032 * resize.
1033 */
1034static void flush_disk(struct block_device *bdev)
1035{
1036 if (__invalidate_device(bdev)) {
1037 char name[BDEVNAME_SIZE] = "";
1038
1039 if (bdev->bd_disk)
1040 disk_name(bdev->bd_disk, 0, name);
1041 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1042 "resized disk %s\n", name);
1043 }
1044
1045 if (!bdev->bd_disk)
1046 return;
1047 if (disk_partitionable(bdev->bd_disk))
1048 bdev->bd_invalidated = 1;
1049}
1050
1051/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001052 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001053 * @disk: struct gendisk to check
1054 * @bdev: struct bdev to adjust.
1055 *
1056 * This routine checks to see if the bdev size does not match the disk size
1057 * and adjusts it if it differs.
1058 */
1059void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1060{
1061 loff_t disk_size, bdev_size;
1062
1063 disk_size = (loff_t)get_capacity(disk) << 9;
1064 bdev_size = i_size_read(bdev->bd_inode);
1065 if (disk_size != bdev_size) {
1066 char name[BDEVNAME_SIZE];
1067
1068 disk_name(disk, 0, name);
1069 printk(KERN_INFO
1070 "%s: detected capacity change from %lld to %lld\n",
1071 name, bdev_size, disk_size);
1072 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001073 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001074 }
1075}
1076EXPORT_SYMBOL(check_disk_size_change);
1077
1078/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001079 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001080 * @disk: struct gendisk to be revalidated
1081 *
1082 * This routine is a wrapper for lower-level driver's revalidate_disk
1083 * call-backs. It is used to do common pre and post operations needed
1084 * for all revalidate_disk operations.
1085 */
1086int revalidate_disk(struct gendisk *disk)
1087{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001088 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001089 int ret = 0;
1090
1091 if (disk->fops->revalidate_disk)
1092 ret = disk->fops->revalidate_disk(disk);
1093
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001094 bdev = bdget_disk(disk, 0);
1095 if (!bdev)
1096 return ret;
1097
1098 mutex_lock(&bdev->bd_mutex);
1099 check_disk_size_change(disk, bdev);
1100 mutex_unlock(&bdev->bd_mutex);
1101 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001102 return ret;
1103}
1104EXPORT_SYMBOL(revalidate_disk);
1105
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106/*
1107 * This routine checks whether a removable media has been changed,
1108 * and invalidates all buffer-cache-entries in that case. This
1109 * is a relatively slow routine, so we have to try to minimize using
1110 * it. Thus it is called only upon a 'mount' or 'open'. This
1111 * is the best way of combining speed and utility, I think.
1112 * People changing diskettes in the middle of an operation deserve
1113 * to lose :-)
1114 */
1115int check_disk_change(struct block_device *bdev)
1116{
1117 struct gendisk *disk = bdev->bd_disk;
1118 struct block_device_operations * bdops = disk->fops;
1119
1120 if (!bdops->media_changed)
1121 return 0;
1122 if (!bdops->media_changed(bdev->bd_disk))
1123 return 0;
1124
Andrew Patterson56ade442008-09-04 14:27:40 -06001125 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 if (bdops->revalidate_disk)
1127 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 return 1;
1129}
1130
1131EXPORT_SYMBOL(check_disk_change);
1132
1133void bd_set_size(struct block_device *bdev, loff_t size)
1134{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001135 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136
1137 bdev->bd_inode->i_size = size;
1138 while (bsize < PAGE_CACHE_SIZE) {
1139 if (size & bsize)
1140 break;
1141 bsize <<= 1;
1142 }
1143 bdev->bd_block_size = bsize;
1144 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1145}
1146EXPORT_SYMBOL(bd_set_size);
1147
Al Viro9a1c3542008-02-22 20:40:24 -05001148static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001149
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001150/*
1151 * bd_mutex locking:
1152 *
1153 * mutex_lock(part->bd_mutex)
1154 * mutex_lock_nested(whole->bd_mutex, 1)
1155 */
1156
Al Viro572c4892007-10-08 13:24:05 -04001157static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001160 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001161 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001162 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
Al Viro572c4892007-10-08 13:24:05 -04001164 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001165 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001166 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001167 perm |= MAY_WRITE;
1168 /*
1169 * hooks: /n/, see "layering violations".
1170 */
1171 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001172 if (ret != 0) {
1173 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001174 return ret;
Al Viro82666022008-08-01 05:32:04 -04001175 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 lock_kernel();
NeilBrownd3374822009-01-09 08:31:10 +11001178 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001179
Tejun Heo89f97492008-11-05 10:21:06 +01001180 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001181 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001182 if (!disk)
1183 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184
NeilBrown6796bf52006-12-08 02:36:16 -08001185 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 if (!bdev->bd_openers) {
1187 bdev->bd_disk = disk;
1188 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001189 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001191
1192 ret = -ENXIO;
1193 bdev->bd_part = disk_get_part(disk, partno);
1194 if (!bdev->bd_part)
1195 goto out_clear;
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001198 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001199 if (ret == -ERESTARTSYS) {
1200 /* Lost a race with 'disk' being
1201 * deleted, try again.
1202 * See md.c
1203 */
1204 disk_put_part(bdev->bd_part);
1205 bdev->bd_part = NULL;
1206 module_put(disk->fops->owner);
1207 put_disk(disk);
1208 bdev->bd_disk = NULL;
1209 mutex_unlock(&bdev->bd_mutex);
1210 goto restart;
1211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001213 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 }
1215 if (!bdev->bd_openers) {
1216 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1217 bdi = blk_get_backing_dev_info(bdev);
1218 if (bdi == NULL)
1219 bdi = &default_backing_dev_info;
1220 bdev->bd_inode->i_data.backing_dev_info = bdi;
1221 }
1222 if (bdev->bd_invalidated)
1223 rescan_partitions(disk, bdev);
1224 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 struct block_device *whole;
1226 whole = bdget_disk(disk, 0);
1227 ret = -ENOMEM;
1228 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001229 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001230 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001231 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001233 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bdev->bd_inode->i_data.backing_dev_info =
1236 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001237 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001238 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001239 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001241 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 }
Tejun Heo89f97492008-11-05 10:21:06 +01001243 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
1245 } else {
1246 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001247 module_put(disk->fops->owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001248 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (bdev->bd_contains == bdev) {
1250 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001251 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001253 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
1255 if (bdev->bd_invalidated)
1256 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 }
1258 }
1259 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001260 if (for_part)
1261 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001262 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 unlock_kernel();
1264 return 0;
1265
Tejun Heo0762b8b2008-08-25 19:56:12 +09001266 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001267 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001269 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1271 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001272 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001274 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001275 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001276 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001278
Tejun Heo0762b8b2008-08-25 19:56:12 +09001279 if (disk)
1280 module_put(disk->fops->owner);
1281 put_disk(disk);
1282 bdput(bdev);
1283
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return ret;
1285}
1286
Al Viro572c4892007-10-08 13:24:05 -04001287int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288{
Al Viro572c4892007-10-08 13:24:05 -04001289 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001290}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291EXPORT_SYMBOL(blkdev_get);
1292
1293static int blkdev_open(struct inode * inode, struct file * filp)
1294{
1295 struct block_device *bdev;
1296 int res;
1297
1298 /*
1299 * Preserve backwards compatibility and allow large file access
1300 * even if userspace doesn't ask for it explicitly. Some mkfs
1301 * binary needs it. We might want to drop this workaround
1302 * during an unstable branch.
1303 */
1304 filp->f_flags |= O_LARGEFILE;
1305
Al Viro572c4892007-10-08 13:24:05 -04001306 if (filp->f_flags & O_NDELAY)
1307 filp->f_mode |= FMODE_NDELAY;
1308 if (filp->f_flags & O_EXCL)
1309 filp->f_mode |= FMODE_EXCL;
1310 if ((filp->f_flags & O_ACCMODE) == 3)
1311 filp->f_mode |= FMODE_WRITE_IOCTL;
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001314 if (bdev == NULL)
1315 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Al Viro572c4892007-10-08 13:24:05 -04001317 filp->f_mapping = bdev->bd_inode->i_mapping;
1318
1319 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 if (res)
1321 return res;
1322
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001323 if (filp->f_mode & FMODE_EXCL) {
1324 res = bd_claim(bdev, filp);
1325 if (res)
1326 goto out_blkdev_put;
1327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001329 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001331 out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001332 blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 return res;
1334}
1335
Al Viro9a1c3542008-02-22 20:40:24 -05001336static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001337{
1338 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001339 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001340 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001341
NeilBrown6796bf52006-12-08 02:36:16 -08001342 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001343 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001344 if (for_part)
1345 bdev->bd_part_count--;
1346
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001347 if (!--bdev->bd_openers) {
1348 sync_blockdev(bdev);
1349 kill_bdev(bdev);
1350 }
1351 if (bdev->bd_contains == bdev) {
1352 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001353 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001354 }
1355 if (!bdev->bd_openers) {
1356 struct module *owner = disk->fops->owner;
1357
1358 put_disk(disk);
1359 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001360 disk_put_part(bdev->bd_part);
1361 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001362 bdev->bd_disk = NULL;
1363 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001364 if (bdev != bdev->bd_contains)
1365 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001366 bdev->bd_contains = NULL;
1367 }
1368 unlock_kernel();
1369 mutex_unlock(&bdev->bd_mutex);
1370 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001371 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001372 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001373 return ret;
1374}
1375
Al Viro9a1c3542008-02-22 20:40:24 -05001376int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001377{
Al Viro9a1c3542008-02-22 20:40:24 -05001378 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001379}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001380EXPORT_SYMBOL(blkdev_put);
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382static int blkdev_close(struct inode * inode, struct file * filp)
1383{
1384 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1385 if (bdev->bd_holder == filp)
1386 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001387 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388}
1389
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001390static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391{
Al Viro56b26ad2008-09-19 03:17:36 -04001392 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1393 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001394
1395 /*
1396 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1397 * to updated it before every ioctl.
1398 */
Al Viro56b26ad2008-09-19 03:17:36 -04001399 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001400 mode |= FMODE_NDELAY;
1401 else
1402 mode &= ~FMODE_NDELAY;
1403
Al Viro56b26ad2008-09-19 03:17:36 -04001404 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405}
1406
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001407/*
1408 * Try to release a page associated with block device when the system
1409 * is under memory pressure.
1410 */
1411static int blkdev_releasepage(struct page *page, gfp_t wait)
1412{
1413 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1414
1415 if (super && super->s_op->bdev_try_to_free_page)
1416 return super->s_op->bdev_try_to_free_page(super, page, wait);
1417
1418 return try_to_free_buffers(page);
1419}
1420
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001421static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 .readpage = blkdev_readpage,
1423 .writepage = blkdev_writepage,
1424 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001425 .write_begin = blkdev_write_begin,
1426 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001428 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 .direct_IO = blkdev_direct_IO,
1430};
1431
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001432const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 .open = blkdev_open,
1434 .release = blkdev_close,
1435 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001436 .read = do_sync_read,
1437 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001439 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 .mmap = generic_file_mmap,
1441 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001442 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443#ifdef CONFIG_COMPAT
1444 .compat_ioctl = compat_blkdev_ioctl,
1445#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001446 .splice_read = generic_file_splice_read,
1447 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448};
1449
1450int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1451{
1452 int res;
1453 mm_segment_t old_fs = get_fs();
1454 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001455 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 set_fs(old_fs);
1457 return res;
1458}
1459
1460EXPORT_SYMBOL(ioctl_by_bdev);
1461
1462/**
1463 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001464 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001466 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 * namespace if possible and return it. Return ERR_PTR(error)
1468 * otherwise.
1469 */
Al Viro421748e2008-08-02 01:04:36 -04001470struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
1472 struct block_device *bdev;
1473 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001474 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 int error;
1476
Al Viro421748e2008-08-02 01:04:36 -04001477 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 return ERR_PTR(-EINVAL);
1479
Al Viro421748e2008-08-02 01:04:36 -04001480 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 if (error)
1482 return ERR_PTR(error);
1483
Al Viro421748e2008-08-02 01:04:36 -04001484 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 error = -ENOTBLK;
1486 if (!S_ISBLK(inode->i_mode))
1487 goto fail;
1488 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001489 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 goto fail;
1491 error = -ENOMEM;
1492 bdev = bd_acquire(inode);
1493 if (!bdev)
1494 goto fail;
1495out:
Al Viro421748e2008-08-02 01:04:36 -04001496 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 return bdev;
1498fail:
1499 bdev = ERR_PTR(error);
1500 goto out;
1501}
Al Virod5686b42008-08-01 05:00:11 -04001502EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504/**
Al Viro30c40d22008-02-22 19:50:45 -05001505 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 *
1507 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001508 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 * @holder: owner for exclusion
1510 *
1511 * Open the blockdevice described by the special file at @path, claim it
1512 * for the @holder.
1513 */
Al Viro30c40d22008-02-22 19:50:45 -05001514struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515{
1516 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 int error = 0;
1518
1519 bdev = lookup_bdev(path);
1520 if (IS_ERR(bdev))
1521 return bdev;
1522
Al Viro572c4892007-10-08 13:24:05 -04001523 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 if (error)
1525 return ERR_PTR(error);
1526 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001527 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 goto blkdev_put;
1529 error = bd_claim(bdev, holder);
1530 if (error)
1531 goto blkdev_put;
1532
1533 return bdev;
1534
1535blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001536 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 return ERR_PTR(error);
1538}
1539
Al Viro30c40d22008-02-22 19:50:45 -05001540EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
1542/**
Al Viro30c40d22008-02-22 19:50:45 -05001543 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 *
1545 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001546 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 *
Al Viro30c40d22008-02-22 19:50:45 -05001548 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 */
Al Viro30c40d22008-02-22 19:50:45 -05001550void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551{
1552 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001553 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554}
1555
Al Viro30c40d22008-02-22 19:50:45 -05001556EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001557
1558int __invalidate_device(struct block_device *bdev)
1559{
1560 struct super_block *sb = get_super(bdev);
1561 int res = 0;
1562
1563 if (sb) {
1564 /*
1565 * no need to lock the super, get_super holds the
1566 * read mutex so the filesystem cannot go away
1567 * under us (->put_super runs with the write lock
1568 * hold).
1569 */
1570 shrink_dcache_sb(sb);
1571 res = invalidate_inodes(sb);
1572 drop_super(sb);
1573 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001574 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001575 return res;
1576}
1577EXPORT_SYMBOL(__invalidate_device);