blob: 9cf4b926f8e47a509f0994ad0e696d79205335bb [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 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100219 * If a superblock is found on this device, we take the s_umount semaphore
220 * on it to make sure nobody unmounts until the snapshot creation is done.
221 * The reference counter (bd_fsfreeze_count) guarantees that only the last
222 * unfreeze process can unfreeze the frozen filesystem actually when multiple
223 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
224 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
225 * actually.
226 */
227struct super_block *freeze_bdev(struct block_device *bdev)
228{
229 struct super_block *sb;
230 int error = 0;
231
232 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200233 if (++bdev->bd_fsfreeze_count > 1) {
234 /*
235 * We don't even need to grab a reference - the first call
236 * to freeze_bdev grab an active reference and only the last
237 * thaw_bdev drops it.
238 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100239 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200240 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100241 mutex_unlock(&bdev->bd_fsfreeze_mutex);
242 return sb;
243 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100244
Christoph Hellwig45042302009-08-03 23:28:35 +0200245 sb = get_active_super(bdev);
246 if (!sb)
247 goto out;
248 if (sb->s_flags & MS_RDONLY) {
249 deactivate_locked_super(sb);
250 mutex_unlock(&bdev->bd_fsfreeze_mutex);
251 return sb;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100252 }
253
Christoph Hellwig45042302009-08-03 23:28:35 +0200254 sb->s_frozen = SB_FREEZE_WRITE;
255 smp_wmb();
256
257 sync_filesystem(sb);
258
259 sb->s_frozen = SB_FREEZE_TRANS;
260 smp_wmb();
261
262 sync_blockdev(sb->s_bdev);
263
264 if (sb->s_op->freeze_fs) {
265 error = sb->s_op->freeze_fs(sb);
266 if (error) {
267 printk(KERN_ERR
268 "VFS:Filesystem freeze failed\n");
269 sb->s_frozen = SB_UNFROZEN;
270 deactivate_locked_super(sb);
271 bdev->bd_fsfreeze_count--;
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
273 return ERR_PTR(error);
274 }
275 }
276 up_write(&sb->s_umount);
277
278 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100279 sync_blockdev(bdev);
280 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200281 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100282}
283EXPORT_SYMBOL(freeze_bdev);
284
285/**
286 * thaw_bdev -- unlock filesystem
287 * @bdev: blockdevice to unlock
288 * @sb: associated superblock
289 *
290 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
291 */
292int thaw_bdev(struct block_device *bdev, struct super_block *sb)
293{
Christoph Hellwig45042302009-08-03 23:28:35 +0200294 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100295
296 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200297 if (!bdev->bd_fsfreeze_count)
298 goto out_unlock;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100299
Christoph Hellwig45042302009-08-03 23:28:35 +0200300 error = 0;
301 if (--bdev->bd_fsfreeze_count > 0)
302 goto out_unlock;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100303
Christoph Hellwig45042302009-08-03 23:28:35 +0200304 if (!sb)
305 goto out_unlock;
306
307 BUG_ON(sb->s_bdev != bdev);
308 down_write(&sb->s_umount);
309 if (sb->s_flags & MS_RDONLY)
310 goto out_deactivate;
311
312 if (sb->s_op->unfreeze_fs) {
313 error = sb->s_op->unfreeze_fs(sb);
314 if (error) {
315 printk(KERN_ERR
316 "VFS:Filesystem thaw failed\n");
317 sb->s_frozen = SB_FREEZE_TRANS;
318 bdev->bd_fsfreeze_count++;
319 mutex_unlock(&bdev->bd_fsfreeze_mutex);
320 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100321 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100322 }
323
Christoph Hellwig45042302009-08-03 23:28:35 +0200324 sb->s_frozen = SB_UNFROZEN;
325 smp_wmb();
326 wake_up(&sb->s_wait_unfrozen);
327
328out_deactivate:
329 if (sb)
330 deactivate_locked_super(sb);
331out_unlock:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
333 return 0;
334}
335EXPORT_SYMBOL(thaw_bdev);
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
338{
339 return block_write_full_page(page, blkdev_get_block, wbc);
340}
341
342static int blkdev_readpage(struct file * file, struct page * page)
343{
344 return block_read_full_page(page, blkdev_get_block);
345}
346
Nick Piggin6272b5a2007-10-16 01:25:04 -0700347static int blkdev_write_begin(struct file *file, struct address_space *mapping,
348 loff_t pos, unsigned len, unsigned flags,
349 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700351 *pagep = NULL;
352 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
353 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
Nick Piggin6272b5a2007-10-16 01:25:04 -0700356static int blkdev_write_end(struct file *file, struct address_space *mapping,
357 loff_t pos, unsigned len, unsigned copied,
358 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700360 int ret;
361 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
362
363 unlock_page(page);
364 page_cache_release(page);
365
366 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
369/*
370 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800371 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 * so we compute the size by hand (just as in block_read/write above)
373 */
374static loff_t block_llseek(struct file *file, loff_t offset, int origin)
375{
376 struct inode *bd_inode = file->f_mapping->host;
377 loff_t size;
378 loff_t retval;
379
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800380 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 size = i_size_read(bd_inode);
382
383 switch (origin) {
384 case 2:
385 offset += size;
386 break;
387 case 1:
388 offset += file->f_pos;
389 }
390 retval = -EINVAL;
391 if (offset >= 0 && offset <= size) {
392 if (offset != file->f_pos) {
393 file->f_pos = offset;
394 }
395 retval = offset;
396 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800397 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 return retval;
399}
400
401/*
402 * Filp is never NULL; the only case when ->fsync() is called with
403 * NULL first argument is nfsd_sync_dir() and that's not a directory.
404 */
405
406static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
407{
408 return sync_blockdev(I_BDEV(filp->f_mapping->host));
409}
410
411/*
412 * pseudo-fs
413 */
414
415static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800416static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
418static struct inode *bdev_alloc_inode(struct super_block *sb)
419{
Christoph Lametere94b1762006-12-06 20:33:17 -0800420 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (!ei)
422 return NULL;
423 return &ei->vfs_inode;
424}
425
426static void bdev_destroy_inode(struct inode *inode)
427{
428 struct bdev_inode *bdi = BDEV_I(inode);
429
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 kmem_cache_free(bdev_cachep, bdi);
431}
432
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700433static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
435 struct bdev_inode *ei = (struct bdev_inode *) foo;
436 struct block_device *bdev = &ei->bdev;
437
Christoph Lametera35afb82007-05-16 22:10:57 -0700438 memset(bdev, 0, sizeof(*bdev));
439 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700440 INIT_LIST_HEAD(&bdev->bd_inodes);
441 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800442#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700443 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800444#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700445 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800446 /* Initialize mutex for freeze. */
447 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448}
449
450static inline void __bd_forget(struct inode *inode)
451{
452 list_del_init(&inode->i_devices);
453 inode->i_bdev = NULL;
454 inode->i_mapping = &inode->i_data;
455}
456
457static void bdev_clear_inode(struct inode *inode)
458{
459 struct block_device *bdev = &BDEV_I(inode)->bdev;
460 struct list_head *p;
461 spin_lock(&bdev_lock);
462 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
463 __bd_forget(list_entry(p, struct inode, i_devices));
464 }
465 list_del_init(&bdev->bd_list);
466 spin_unlock(&bdev_lock);
467}
468
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800469static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 .statfs = simple_statfs,
471 .alloc_inode = bdev_alloc_inode,
472 .destroy_inode = bdev_destroy_inode,
473 .drop_inode = generic_delete_inode,
474 .clear_inode = bdev_clear_inode,
475};
476
David Howells454e2392006-06-23 02:02:57 -0700477static int bd_get_sb(struct file_system_type *fs_type,
478 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
David Howells454e2392006-06-23 02:02:57 -0700480 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
483static struct file_system_type bd_type = {
484 .name = "bdev",
485 .get_sb = bd_get_sb,
486 .kill_sb = kill_anon_super,
487};
488
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800489struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
491void __init bdev_cache_init(void)
492{
493 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800494 struct vfsmount *bd_mnt;
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800497 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
498 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900499 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 err = register_filesystem(&bd_type);
501 if (err)
502 panic("Cannot register bdev pseudo-fs");
503 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (IS_ERR(bd_mnt))
505 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100506 /*
507 * This vfsmount structure is only used to obtain the
508 * blockdev_superblock, so tell kmemleak not to report it.
509 */
510 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
512}
513
514/*
515 * Most likely _very_ bad one - but then it's hardly critical for small
516 * /dev and can be fixed when somebody will need really large one.
517 * Keep in mind that it will be fed through icache hash function too.
518 */
519static inline unsigned long hash(dev_t dev)
520{
521 return MAJOR(dev)+MINOR(dev);
522}
523
524static int bdev_test(struct inode *inode, void *data)
525{
526 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
527}
528
529static int bdev_set(struct inode *inode, void *data)
530{
531 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
532 return 0;
533}
534
535static LIST_HEAD(all_bdevs);
536
537struct block_device *bdget(dev_t dev)
538{
539 struct block_device *bdev;
540 struct inode *inode;
541
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800542 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 bdev_test, bdev_set, &dev);
544
545 if (!inode)
546 return NULL;
547
548 bdev = &BDEV_I(inode)->bdev;
549
550 if (inode->i_state & I_NEW) {
551 bdev->bd_contains = NULL;
552 bdev->bd_inode = inode;
553 bdev->bd_block_size = (1 << inode->i_blkbits);
554 bdev->bd_part_count = 0;
555 bdev->bd_invalidated = 0;
556 inode->i_mode = S_IFBLK;
557 inode->i_rdev = dev;
558 inode->i_bdev = bdev;
559 inode->i_data.a_ops = &def_blk_aops;
560 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
561 inode->i_data.backing_dev_info = &default_backing_dev_info;
562 spin_lock(&bdev_lock);
563 list_add(&bdev->bd_list, &all_bdevs);
564 spin_unlock(&bdev_lock);
565 unlock_new_inode(inode);
566 }
567 return bdev;
568}
569
570EXPORT_SYMBOL(bdget);
571
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200572/**
573 * bdgrab -- Grab a reference to an already referenced block device
574 * @bdev: Block device to grab a reference to.
575 */
576struct block_device *bdgrab(struct block_device *bdev)
577{
578 atomic_inc(&bdev->bd_inode->i_count);
579 return bdev;
580}
581
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582long nr_blockdev_pages(void)
583{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700584 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 long ret = 0;
586 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700587 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 ret += bdev->bd_inode->i_mapping->nrpages;
589 }
590 spin_unlock(&bdev_lock);
591 return ret;
592}
593
594void bdput(struct block_device *bdev)
595{
596 iput(bdev->bd_inode);
597}
598
599EXPORT_SYMBOL(bdput);
600
601static struct block_device *bd_acquire(struct inode *inode)
602{
603 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700604
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 spin_lock(&bdev_lock);
606 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700607 if (bdev) {
608 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 spin_unlock(&bdev_lock);
610 return bdev;
611 }
612 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 bdev = bdget(inode->i_rdev);
615 if (bdev) {
616 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700617 if (!inode->i_bdev) {
618 /*
619 * We take an additional bd_inode->i_count for inode,
620 * and it's released in clear_inode() of inode.
621 * So, we can access it via ->i_mapping always
622 * without igrab().
623 */
624 atomic_inc(&bdev->bd_inode->i_count);
625 inode->i_bdev = bdev;
626 inode->i_mapping = bdev->bd_inode->i_mapping;
627 list_add(&inode->i_devices, &bdev->bd_inodes);
628 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 spin_unlock(&bdev_lock);
630 }
631 return bdev;
632}
633
634/* Call when you free inode */
635
636void bd_forget(struct inode *inode)
637{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700638 struct block_device *bdev = NULL;
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700641 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800642 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700643 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700645 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700647
648 if (bdev)
649 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
651
652int bd_claim(struct block_device *bdev, void *holder)
653{
654 int res;
655 spin_lock(&bdev_lock);
656
657 /* first decide result */
658 if (bdev->bd_holder == holder)
659 res = 0; /* already a holder */
660 else if (bdev->bd_holder != NULL)
661 res = -EBUSY; /* held by someone else */
662 else if (bdev->bd_contains == bdev)
663 res = 0; /* is a whole device which isn't held */
664
665 else if (bdev->bd_contains->bd_holder == bd_claim)
666 res = 0; /* is a partition of a device that is being partitioned */
667 else if (bdev->bd_contains->bd_holder != NULL)
668 res = -EBUSY; /* is a partition of a held device */
669 else
670 res = 0; /* is a partition of an un-held device */
671
672 /* now impose change */
673 if (res==0) {
674 /* note that for a whole device bd_holders
675 * will be incremented twice, and bd_holder will
676 * be set to bd_claim before being set to holder
677 */
678 bdev->bd_contains->bd_holders ++;
679 bdev->bd_contains->bd_holder = bd_claim;
680 bdev->bd_holders++;
681 bdev->bd_holder = holder;
682 }
683 spin_unlock(&bdev_lock);
684 return res;
685}
686
687EXPORT_SYMBOL(bd_claim);
688
689void bd_release(struct block_device *bdev)
690{
691 spin_lock(&bdev_lock);
692 if (!--bdev->bd_contains->bd_holders)
693 bdev->bd_contains->bd_holder = NULL;
694 if (!--bdev->bd_holders)
695 bdev->bd_holder = NULL;
696 spin_unlock(&bdev_lock);
697}
698
699EXPORT_SYMBOL(bd_release);
700
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800701#ifdef CONFIG_SYSFS
702/*
703 * Functions for bd_claim_by_kobject / bd_release_from_kobject
704 *
705 * If a kobject is passed to bd_claim_by_kobject()
706 * and the kobject has a parent directory,
707 * following symlinks are created:
708 * o from the kobject to the claimed bdev
709 * o from "holders" directory of the bdev to the parent of the kobject
710 * bd_release_from_kobject() removes these symlinks.
711 *
712 * Example:
713 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
714 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
715 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
716 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
717 */
718
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700719static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800720{
721 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700722 return 0;
723 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800724}
725
726static void del_symlink(struct kobject *from, struct kobject *to)
727{
728 if (!from || !to)
729 return;
730 sysfs_remove_link(from, kobject_name(to));
731}
732
733/*
734 * 'struct bd_holder' contains pointers to kobjects symlinked by
735 * bd_claim_by_kobject.
736 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
737 */
738struct bd_holder {
739 struct list_head list; /* chain of holders of the bdev */
740 int count; /* references from the holder */
741 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
742 struct kobject *hdev; /* e.g. "/block/dm-0" */
743 struct kobject *hdir; /* e.g. "/block/sda/holders" */
744 struct kobject *sdev; /* e.g. "/block/sda" */
745};
746
747/*
748 * Get references of related kobjects at once.
749 * Returns 1 on success. 0 on failure.
750 *
751 * Should call bd_holder_release_dirs() after successful use.
752 */
753static int bd_holder_grab_dirs(struct block_device *bdev,
754 struct bd_holder *bo)
755{
756 if (!bdev || !bo)
757 return 0;
758
759 bo->sdir = kobject_get(bo->sdir);
760 if (!bo->sdir)
761 return 0;
762
763 bo->hdev = kobject_get(bo->sdir->parent);
764 if (!bo->hdev)
765 goto fail_put_sdir;
766
Tejun Heo0762b8b2008-08-25 19:56:12 +0900767 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800768 if (!bo->sdev)
769 goto fail_put_hdev;
770
Tejun Heo4c465012008-08-25 19:56:11 +0900771 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800772 if (!bo->hdir)
773 goto fail_put_sdev;
774
775 return 1;
776
777fail_put_sdev:
778 kobject_put(bo->sdev);
779fail_put_hdev:
780 kobject_put(bo->hdev);
781fail_put_sdir:
782 kobject_put(bo->sdir);
783
784 return 0;
785}
786
787/* Put references of related kobjects at once. */
788static void bd_holder_release_dirs(struct bd_holder *bo)
789{
790 kobject_put(bo->hdir);
791 kobject_put(bo->sdev);
792 kobject_put(bo->hdev);
793 kobject_put(bo->sdir);
794}
795
796static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
797{
798 struct bd_holder *bo;
799
800 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
801 if (!bo)
802 return NULL;
803
804 bo->count = 1;
805 bo->sdir = kobj;
806
807 return bo;
808}
809
810static void free_bd_holder(struct bd_holder *bo)
811{
812 kfree(bo);
813}
814
815/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500816 * find_bd_holder - find matching struct bd_holder from the block device
817 *
818 * @bdev: struct block device to be searched
819 * @bo: target struct bd_holder
820 *
821 * Returns matching entry with @bo in @bdev->bd_holder_list.
822 * If found, increment the reference count and return the pointer.
823 * If not found, returns NULL.
824 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800825static struct bd_holder *find_bd_holder(struct block_device *bdev,
826 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500827{
828 struct bd_holder *tmp;
829
830 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
831 if (tmp->sdir == bo->sdir) {
832 tmp->count++;
833 return tmp;
834 }
835
836 return NULL;
837}
838
839/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800840 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
841 *
842 * @bdev: block device to be bd_claimed
843 * @bo: preallocated and initialized by alloc_bd_holder()
844 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500845 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800846 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500847 * Returns 0 if symlinks are created.
848 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800849 */
850static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
851{
Johannes Weiner4e916722007-07-15 23:41:25 -0700852 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800853
854 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700855 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800856
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800857 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700858 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800859
Johannes Weiner4e916722007-07-15 23:41:25 -0700860 err = add_symlink(bo->sdir, bo->sdev);
861 if (err)
862 return err;
863
864 err = add_symlink(bo->hdir, bo->hdev);
865 if (err) {
866 del_symlink(bo->sdir, bo->sdev);
867 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700868 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700869
870 list_add_tail(&bo->list, &bdev->bd_holder_list);
871 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800872}
873
874/**
875 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
876 *
877 * @bdev: block device to be bd_claimed
878 * @kobj: holder's kobject
879 *
880 * If there is matching entry with @kobj in @bdev->bd_holder_list
881 * and no other bd_claim() from the same kobject,
882 * remove the struct bd_holder from the list, delete symlinks for it.
883 *
884 * Returns a pointer to the struct bd_holder when it's removed from the list
885 * and ready to be freed.
886 * Returns NULL if matching claim isn't found or there is other bd_claim()
887 * by the same kobject.
888 */
889static struct bd_holder *del_bd_holder(struct block_device *bdev,
890 struct kobject *kobj)
891{
892 struct bd_holder *bo;
893
894 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
895 if (bo->sdir == kobj) {
896 bo->count--;
897 BUG_ON(bo->count < 0);
898 if (!bo->count) {
899 list_del(&bo->list);
900 del_symlink(bo->sdir, bo->sdev);
901 del_symlink(bo->hdir, bo->hdev);
902 bd_holder_release_dirs(bo);
903 return bo;
904 }
905 break;
906 }
907 }
908
909 return NULL;
910}
911
912/**
913 * bd_claim_by_kobject - bd_claim() with additional kobject signature
914 *
915 * @bdev: block device to be claimed
916 * @holder: holder's signature
917 * @kobj: holder's kobject
918 *
919 * Do bd_claim() and if it succeeds, create sysfs symlinks between
920 * the bdev and the holder's kobject.
921 * Use bd_release_from_kobject() when relesing the claimed bdev.
922 *
923 * Returns 0 on success. (same as bd_claim())
924 * Returns errno on failure.
925 */
926static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
927 struct kobject *kobj)
928{
Johannes Weiner4e916722007-07-15 23:41:25 -0700929 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500930 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800931
932 if (!kobj)
933 return -EINVAL;
934
935 bo = alloc_bd_holder(kobj);
936 if (!bo)
937 return -ENOMEM;
938
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800939 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500940
Johannes Weiner4e916722007-07-15 23:41:25 -0700941 err = bd_claim(bdev, holder);
942 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700943 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700944
945 found = find_bd_holder(bdev, bo);
946 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700947 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700948
949 err = add_bd_holder(bdev, bo);
950 if (err)
951 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700952 else
953 bo = NULL;
954fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800955 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700956 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700957 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800958}
959
960/**
961 * bd_release_from_kobject - bd_release() with additional kobject signature
962 *
963 * @bdev: block device to be released
964 * @kobj: holder's kobject
965 *
966 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
967 */
968static void bd_release_from_kobject(struct block_device *bdev,
969 struct kobject *kobj)
970{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800971 if (!kobj)
972 return;
973
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800974 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800975 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700976 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800977 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800978}
979
980/**
981 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
982 *
983 * @bdev: block device to be claimed
984 * @holder: holder's signature
985 * @disk: holder's gendisk
986 *
987 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
988 */
989int bd_claim_by_disk(struct block_device *bdev, void *holder,
990 struct gendisk *disk)
991{
992 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
993}
994EXPORT_SYMBOL_GPL(bd_claim_by_disk);
995
996/**
997 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
998 *
999 * @bdev: block device to be claimed
1000 * @disk: holder's gendisk
1001 *
1002 * Call bd_release_from_kobject() and put @disk->slave_dir.
1003 */
1004void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1005{
1006 bd_release_from_kobject(bdev, disk->slave_dir);
1007 kobject_put(disk->slave_dir);
1008}
1009EXPORT_SYMBOL_GPL(bd_release_from_disk);
1010#endif
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012/*
1013 * Tries to open block device by device number. Use it ONLY if you
1014 * really do not have anything better - i.e. when you are behind a
1015 * truly sucky interface and all you are given is a device number. _Never_
1016 * to be used for internal purposes. If you ever need it - reconsider
1017 * your API.
1018 */
Al Viroaeb5d722008-09-02 15:28:45 -04001019struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020{
1021 struct block_device *bdev = bdget(dev);
1022 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001024 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 return err ? ERR_PTR(err) : bdev;
1026}
1027
1028EXPORT_SYMBOL(open_by_devnum);
1029
Andrew Patterson0c002c22008-09-04 14:27:20 -06001030/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001031 * flush_disk - invalidates all buffer-cache entries on a disk
1032 *
1033 * @bdev: struct block device to be flushed
1034 *
1035 * Invalidates all buffer-cache entries on a disk. It should be called
1036 * when a disk has been changed -- either by a media change or online
1037 * resize.
1038 */
1039static void flush_disk(struct block_device *bdev)
1040{
1041 if (__invalidate_device(bdev)) {
1042 char name[BDEVNAME_SIZE] = "";
1043
1044 if (bdev->bd_disk)
1045 disk_name(bdev->bd_disk, 0, name);
1046 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1047 "resized disk %s\n", name);
1048 }
1049
1050 if (!bdev->bd_disk)
1051 return;
1052 if (disk_partitionable(bdev->bd_disk))
1053 bdev->bd_invalidated = 1;
1054}
1055
1056/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001057 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001058 * @disk: struct gendisk to check
1059 * @bdev: struct bdev to adjust.
1060 *
1061 * This routine checks to see if the bdev size does not match the disk size
1062 * and adjusts it if it differs.
1063 */
1064void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1065{
1066 loff_t disk_size, bdev_size;
1067
1068 disk_size = (loff_t)get_capacity(disk) << 9;
1069 bdev_size = i_size_read(bdev->bd_inode);
1070 if (disk_size != bdev_size) {
1071 char name[BDEVNAME_SIZE];
1072
1073 disk_name(disk, 0, name);
1074 printk(KERN_INFO
1075 "%s: detected capacity change from %lld to %lld\n",
1076 name, bdev_size, disk_size);
1077 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001078 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001079 }
1080}
1081EXPORT_SYMBOL(check_disk_size_change);
1082
1083/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001084 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001085 * @disk: struct gendisk to be revalidated
1086 *
1087 * This routine is a wrapper for lower-level driver's revalidate_disk
1088 * call-backs. It is used to do common pre and post operations needed
1089 * for all revalidate_disk operations.
1090 */
1091int revalidate_disk(struct gendisk *disk)
1092{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001093 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001094 int ret = 0;
1095
1096 if (disk->fops->revalidate_disk)
1097 ret = disk->fops->revalidate_disk(disk);
1098
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001099 bdev = bdget_disk(disk, 0);
1100 if (!bdev)
1101 return ret;
1102
1103 mutex_lock(&bdev->bd_mutex);
1104 check_disk_size_change(disk, bdev);
1105 mutex_unlock(&bdev->bd_mutex);
1106 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001107 return ret;
1108}
1109EXPORT_SYMBOL(revalidate_disk);
1110
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111/*
1112 * This routine checks whether a removable media has been changed,
1113 * and invalidates all buffer-cache-entries in that case. This
1114 * is a relatively slow routine, so we have to try to minimize using
1115 * it. Thus it is called only upon a 'mount' or 'open'. This
1116 * is the best way of combining speed and utility, I think.
1117 * People changing diskettes in the middle of an operation deserve
1118 * to lose :-)
1119 */
1120int check_disk_change(struct block_device *bdev)
1121{
1122 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001123 const struct block_device_operations *bdops = disk->fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124
1125 if (!bdops->media_changed)
1126 return 0;
1127 if (!bdops->media_changed(bdev->bd_disk))
1128 return 0;
1129
Andrew Patterson56ade442008-09-04 14:27:40 -06001130 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (bdops->revalidate_disk)
1132 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return 1;
1134}
1135
1136EXPORT_SYMBOL(check_disk_change);
1137
1138void bd_set_size(struct block_device *bdev, loff_t size)
1139{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001140 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142 bdev->bd_inode->i_size = size;
1143 while (bsize < PAGE_CACHE_SIZE) {
1144 if (size & bsize)
1145 break;
1146 bsize <<= 1;
1147 }
1148 bdev->bd_block_size = bsize;
1149 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1150}
1151EXPORT_SYMBOL(bd_set_size);
1152
Al Viro9a1c3542008-02-22 20:40:24 -05001153static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001154
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001155/*
1156 * bd_mutex locking:
1157 *
1158 * mutex_lock(part->bd_mutex)
1159 * mutex_lock_nested(whole->bd_mutex, 1)
1160 */
1161
Al Viro572c4892007-10-08 13:24:05 -04001162static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001165 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001166 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001167 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
Al Viro572c4892007-10-08 13:24:05 -04001169 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001170 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001171 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001172 perm |= MAY_WRITE;
1173 /*
1174 * hooks: /n/, see "layering violations".
1175 */
1176 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001177 if (ret != 0) {
1178 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001179 return ret;
Al Viro82666022008-08-01 05:32:04 -04001180 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 lock_kernel();
NeilBrownd3374822009-01-09 08:31:10 +11001183 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001184
Tejun Heo89f97492008-11-05 10:21:06 +01001185 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001186 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001187 if (!disk)
1188 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
NeilBrown6796bf52006-12-08 02:36:16 -08001190 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 if (!bdev->bd_openers) {
1192 bdev->bd_disk = disk;
1193 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001194 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001196
1197 ret = -ENXIO;
1198 bdev->bd_part = disk_get_part(disk, partno);
1199 if (!bdev->bd_part)
1200 goto out_clear;
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001203 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001204 if (ret == -ERESTARTSYS) {
1205 /* Lost a race with 'disk' being
1206 * deleted, try again.
1207 * See md.c
1208 */
1209 disk_put_part(bdev->bd_part);
1210 bdev->bd_part = NULL;
1211 module_put(disk->fops->owner);
1212 put_disk(disk);
1213 bdev->bd_disk = NULL;
1214 mutex_unlock(&bdev->bd_mutex);
1215 goto restart;
1216 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001218 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 }
1220 if (!bdev->bd_openers) {
1221 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1222 bdi = blk_get_backing_dev_info(bdev);
1223 if (bdi == NULL)
1224 bdi = &default_backing_dev_info;
1225 bdev->bd_inode->i_data.backing_dev_info = bdi;
1226 }
1227 if (bdev->bd_invalidated)
1228 rescan_partitions(disk, bdev);
1229 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 struct block_device *whole;
1231 whole = bdget_disk(disk, 0);
1232 ret = -ENOMEM;
1233 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001234 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001235 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001236 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001238 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 bdev->bd_inode->i_data.backing_dev_info =
1241 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001242 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001243 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001244 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001246 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
Tejun Heo89f97492008-11-05 10:21:06 +01001248 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 }
1250 } else {
1251 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001252 module_put(disk->fops->owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001253 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (bdev->bd_contains == bdev) {
1255 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001256 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001258 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 }
1260 if (bdev->bd_invalidated)
1261 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 }
1263 }
1264 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001265 if (for_part)
1266 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001267 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 unlock_kernel();
1269 return 0;
1270
Tejun Heo0762b8b2008-08-25 19:56:12 +09001271 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001272 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001274 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1276 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001277 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001279 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001280 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001281 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001283
Tejun Heo0762b8b2008-08-25 19:56:12 +09001284 if (disk)
1285 module_put(disk->fops->owner);
1286 put_disk(disk);
1287 bdput(bdev);
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return ret;
1290}
1291
Al Viro572c4892007-10-08 13:24:05 -04001292int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293{
Al Viro572c4892007-10-08 13:24:05 -04001294 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001295}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296EXPORT_SYMBOL(blkdev_get);
1297
1298static int blkdev_open(struct inode * inode, struct file * filp)
1299{
1300 struct block_device *bdev;
1301 int res;
1302
1303 /*
1304 * Preserve backwards compatibility and allow large file access
1305 * even if userspace doesn't ask for it explicitly. Some mkfs
1306 * binary needs it. We might want to drop this workaround
1307 * during an unstable branch.
1308 */
1309 filp->f_flags |= O_LARGEFILE;
1310
Al Viro572c4892007-10-08 13:24:05 -04001311 if (filp->f_flags & O_NDELAY)
1312 filp->f_mode |= FMODE_NDELAY;
1313 if (filp->f_flags & O_EXCL)
1314 filp->f_mode |= FMODE_EXCL;
1315 if ((filp->f_flags & O_ACCMODE) == 3)
1316 filp->f_mode |= FMODE_WRITE_IOCTL;
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001319 if (bdev == NULL)
1320 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Al Viro572c4892007-10-08 13:24:05 -04001322 filp->f_mapping = bdev->bd_inode->i_mapping;
1323
1324 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 if (res)
1326 return res;
1327
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001328 if (filp->f_mode & FMODE_EXCL) {
1329 res = bd_claim(bdev, filp);
1330 if (res)
1331 goto out_blkdev_put;
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001334 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001336 out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001337 blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 return res;
1339}
1340
Al Viro9a1c3542008-02-22 20:40:24 -05001341static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001342{
1343 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001344 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001345 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001346
NeilBrown6796bf52006-12-08 02:36:16 -08001347 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001348 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001349 if (for_part)
1350 bdev->bd_part_count--;
1351
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001352 if (!--bdev->bd_openers) {
1353 sync_blockdev(bdev);
1354 kill_bdev(bdev);
1355 }
1356 if (bdev->bd_contains == bdev) {
1357 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001358 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001359 }
1360 if (!bdev->bd_openers) {
1361 struct module *owner = disk->fops->owner;
1362
1363 put_disk(disk);
1364 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001365 disk_put_part(bdev->bd_part);
1366 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001367 bdev->bd_disk = NULL;
1368 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001369 if (bdev != bdev->bd_contains)
1370 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001371 bdev->bd_contains = NULL;
1372 }
1373 unlock_kernel();
1374 mutex_unlock(&bdev->bd_mutex);
1375 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001376 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001377 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001378 return ret;
1379}
1380
Al Viro9a1c3542008-02-22 20:40:24 -05001381int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001382{
Al Viro9a1c3542008-02-22 20:40:24 -05001383 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001384}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001385EXPORT_SYMBOL(blkdev_put);
1386
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387static int blkdev_close(struct inode * inode, struct file * filp)
1388{
1389 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1390 if (bdev->bd_holder == filp)
1391 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001392 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393}
1394
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001395static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
Al Viro56b26ad2008-09-19 03:17:36 -04001397 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1398 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001399
1400 /*
1401 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1402 * to updated it before every ioctl.
1403 */
Al Viro56b26ad2008-09-19 03:17:36 -04001404 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001405 mode |= FMODE_NDELAY;
1406 else
1407 mode &= ~FMODE_NDELAY;
1408
Al Viro56b26ad2008-09-19 03:17:36 -04001409 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410}
1411
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001412/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001413 * Write data to the block device. Only intended for the block device itself
1414 * and the raw driver which basically is a fake block device.
1415 *
1416 * Does not take i_mutex for the write and thus is not for general purpose
1417 * use.
1418 */
1419ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1420 unsigned long nr_segs, loff_t pos)
1421{
1422 struct file *file = iocb->ki_filp;
1423 ssize_t ret;
1424
1425 BUG_ON(iocb->ki_pos != pos);
1426
1427 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1428 if (ret > 0 || ret == -EIOCBQUEUED) {
1429 ssize_t err;
1430
1431 err = generic_write_sync(file, pos, ret);
1432 if (err < 0 && ret > 0)
1433 ret = err;
1434 }
1435 return ret;
1436}
1437EXPORT_SYMBOL_GPL(blkdev_aio_write);
1438
1439/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001440 * Try to release a page associated with block device when the system
1441 * is under memory pressure.
1442 */
1443static int blkdev_releasepage(struct page *page, gfp_t wait)
1444{
1445 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1446
1447 if (super && super->s_op->bdev_try_to_free_page)
1448 return super->s_op->bdev_try_to_free_page(super, page, wait);
1449
1450 return try_to_free_buffers(page);
1451}
1452
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001453static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 .readpage = blkdev_readpage,
1455 .writepage = blkdev_writepage,
1456 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001457 .write_begin = blkdev_write_begin,
1458 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001460 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 .direct_IO = blkdev_direct_IO,
1462};
1463
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001464const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 .open = blkdev_open,
1466 .release = blkdev_close,
1467 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001468 .read = do_sync_read,
1469 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001471 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 .mmap = generic_file_mmap,
1473 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001474 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475#ifdef CONFIG_COMPAT
1476 .compat_ioctl = compat_blkdev_ioctl,
1477#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001478 .splice_read = generic_file_splice_read,
1479 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480};
1481
1482int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1483{
1484 int res;
1485 mm_segment_t old_fs = get_fs();
1486 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001487 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 set_fs(old_fs);
1489 return res;
1490}
1491
1492EXPORT_SYMBOL(ioctl_by_bdev);
1493
1494/**
1495 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001496 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001498 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 * namespace if possible and return it. Return ERR_PTR(error)
1500 * otherwise.
1501 */
Al Viro421748e2008-08-02 01:04:36 -04001502struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
1504 struct block_device *bdev;
1505 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001506 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 int error;
1508
Al Viro421748e2008-08-02 01:04:36 -04001509 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 return ERR_PTR(-EINVAL);
1511
Al Viro421748e2008-08-02 01:04:36 -04001512 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 if (error)
1514 return ERR_PTR(error);
1515
Al Viro421748e2008-08-02 01:04:36 -04001516 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 error = -ENOTBLK;
1518 if (!S_ISBLK(inode->i_mode))
1519 goto fail;
1520 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001521 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 goto fail;
1523 error = -ENOMEM;
1524 bdev = bd_acquire(inode);
1525 if (!bdev)
1526 goto fail;
1527out:
Al Viro421748e2008-08-02 01:04:36 -04001528 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 return bdev;
1530fail:
1531 bdev = ERR_PTR(error);
1532 goto out;
1533}
Al Virod5686b42008-08-01 05:00:11 -04001534EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536/**
Al Viro30c40d22008-02-22 19:50:45 -05001537 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 *
1539 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001540 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 * @holder: owner for exclusion
1542 *
1543 * Open the blockdevice described by the special file at @path, claim it
1544 * for the @holder.
1545 */
Al Viro30c40d22008-02-22 19:50:45 -05001546struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547{
1548 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 int error = 0;
1550
1551 bdev = lookup_bdev(path);
1552 if (IS_ERR(bdev))
1553 return bdev;
1554
Al Viro572c4892007-10-08 13:24:05 -04001555 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (error)
1557 return ERR_PTR(error);
1558 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001559 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 goto blkdev_put;
1561 error = bd_claim(bdev, holder);
1562 if (error)
1563 goto blkdev_put;
1564
1565 return bdev;
1566
1567blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001568 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 return ERR_PTR(error);
1570}
1571
Al Viro30c40d22008-02-22 19:50:45 -05001572EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
1574/**
Al Viro30c40d22008-02-22 19:50:45 -05001575 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 *
1577 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001578 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 *
Al Viro30c40d22008-02-22 19:50:45 -05001580 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 */
Al Viro30c40d22008-02-22 19:50:45 -05001582void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583{
1584 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001585 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586}
1587
Al Viro30c40d22008-02-22 19:50:45 -05001588EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001589
1590int __invalidate_device(struct block_device *bdev)
1591{
1592 struct super_block *sb = get_super(bdev);
1593 int res = 0;
1594
1595 if (sb) {
1596 /*
1597 * no need to lock the super, get_super holds the
1598 * read mutex so the filesystem cannot go away
1599 * under us (->put_super runs with the write lock
1600 * hold).
1601 */
1602 shrink_dcache_sb(sb);
1603 res = invalidate_inodes(sb);
1604 drop_super(sb);
1605 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001606 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001607 return res;
1608}
1609EXPORT_SYMBOL(__invalidate_device);