blob: 50e8c8582faa06a38bae5beb66c4f3ec8af929f3 [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
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200175 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
176 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800177}
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;
Josef Bacik18e9e512010-03-23 10:34:56 -0400248 error = freeze_super(sb);
249 if (error) {
250 deactivate_super(sb);
251 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200252 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400253 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400255 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200256 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100257 sync_blockdev(bdev);
258 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200259 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100260}
261EXPORT_SYMBOL(freeze_bdev);
262
263/**
264 * thaw_bdev -- unlock filesystem
265 * @bdev: blockdevice to unlock
266 * @sb: associated superblock
267 *
268 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
269 */
270int thaw_bdev(struct block_device *bdev, struct super_block *sb)
271{
Christoph Hellwig45042302009-08-03 23:28:35 +0200272 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100273
274 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200275 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400276 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100277
Christoph Hellwig45042302009-08-03 23:28:35 +0200278 error = 0;
279 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400280 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100281
Christoph Hellwig45042302009-08-03 23:28:35 +0200282 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400283 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200284
Josef Bacik18e9e512010-03-23 10:34:56 -0400285 error = thaw_super(sb);
286 if (error) {
287 bdev->bd_fsfreeze_count++;
288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
289 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100290 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400291out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100292 mutex_unlock(&bdev->bd_fsfreeze_mutex);
293 return 0;
294}
295EXPORT_SYMBOL(thaw_bdev);
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
298{
299 return block_write_full_page(page, blkdev_get_block, wbc);
300}
301
302static int blkdev_readpage(struct file * file, struct page * page)
303{
304 return block_read_full_page(page, blkdev_get_block);
305}
306
Nick Piggin6272b5a2007-10-16 01:25:04 -0700307static int blkdev_write_begin(struct file *file, struct address_space *mapping,
308 loff_t pos, unsigned len, unsigned flags,
309 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200311 return block_write_begin(mapping, pos, len, flags, pagep,
312 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313}
314
Nick Piggin6272b5a2007-10-16 01:25:04 -0700315static int blkdev_write_end(struct file *file, struct address_space *mapping,
316 loff_t pos, unsigned len, unsigned copied,
317 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700319 int ret;
320 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
321
322 unlock_page(page);
323 page_cache_release(page);
324
325 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
328/*
329 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800330 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 * so we compute the size by hand (just as in block_read/write above)
332 */
333static loff_t block_llseek(struct file *file, loff_t offset, int origin)
334{
335 struct inode *bd_inode = file->f_mapping->host;
336 loff_t size;
337 loff_t retval;
338
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800339 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 size = i_size_read(bd_inode);
341
342 switch (origin) {
343 case 2:
344 offset += size;
345 break;
346 case 1:
347 offset += file->f_pos;
348 }
349 retval = -EINVAL;
350 if (offset >= 0 && offset <= size) {
351 if (offset != file->f_pos) {
352 file->f_pos = offset;
353 }
354 retval = offset;
355 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800356 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 return retval;
358}
359
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200360int blkdev_fsync(struct file *filp, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400362 struct inode *bd_inode = filp->f_mapping->host;
363 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100364 int error;
365
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400366 /*
367 * There is no need to serialise calls to blkdev_issue_flush with
368 * i_mutex and doing so causes performance issues with concurrent
369 * O_SYNC writers to a block device.
370 */
371 mutex_unlock(&bd_inode->i_mutex);
372
Jens Axboe7407cf32010-04-29 09:36:24 +0200373 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL, BLKDEV_IFL_WAIT);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100374 if (error == -EOPNOTSUPP)
375 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400376
377 mutex_lock(&bd_inode->i_mutex);
378
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100379 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700381EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383/*
384 * pseudo-fs
385 */
386
387static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800388static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390static struct inode *bdev_alloc_inode(struct super_block *sb)
391{
Christoph Lametere94b1762006-12-06 20:33:17 -0800392 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (!ei)
394 return NULL;
395 return &ei->vfs_inode;
396}
397
398static void bdev_destroy_inode(struct inode *inode)
399{
400 struct bdev_inode *bdi = BDEV_I(inode);
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 kmem_cache_free(bdev_cachep, bdi);
403}
404
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700405static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
407 struct bdev_inode *ei = (struct bdev_inode *) foo;
408 struct block_device *bdev = &ei->bdev;
409
Christoph Lametera35afb82007-05-16 22:10:57 -0700410 memset(bdev, 0, sizeof(*bdev));
411 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700412 INIT_LIST_HEAD(&bdev->bd_inodes);
413 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800414#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700415 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800416#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700417 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800418 /* Initialize mutex for freeze. */
419 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
422static inline void __bd_forget(struct inode *inode)
423{
424 list_del_init(&inode->i_devices);
425 inode->i_bdev = NULL;
426 inode->i_mapping = &inode->i_data;
427}
428
Al Virob57922d2010-06-07 14:34:48 -0400429static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
431 struct block_device *bdev = &BDEV_I(inode)->bdev;
432 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400433 truncate_inode_pages(&inode->i_data, 0);
434 invalidate_inode_buffers(inode); /* is it needed here? */
435 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 spin_lock(&bdev_lock);
437 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
438 __bd_forget(list_entry(p, struct inode, i_devices));
439 }
440 list_del_init(&bdev->bd_list);
441 spin_unlock(&bdev_lock);
442}
443
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800444static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 .statfs = simple_statfs,
446 .alloc_inode = bdev_alloc_inode,
447 .destroy_inode = bdev_destroy_inode,
448 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400449 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450};
451
David Howells454e2392006-06-23 02:02:57 -0700452static int bd_get_sb(struct file_system_type *fs_type,
453 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454{
David Howells454e2392006-06-23 02:02:57 -0700455 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
458static struct file_system_type bd_type = {
459 .name = "bdev",
460 .get_sb = bd_get_sb,
461 .kill_sb = kill_anon_super,
462};
463
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800464struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466void __init bdev_cache_init(void)
467{
468 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800469 struct vfsmount *bd_mnt;
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800472 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
473 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900474 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 err = register_filesystem(&bd_type);
476 if (err)
477 panic("Cannot register bdev pseudo-fs");
478 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 if (IS_ERR(bd_mnt))
480 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100481 /*
482 * This vfsmount structure is only used to obtain the
483 * blockdev_superblock, so tell kmemleak not to report it.
484 */
485 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
487}
488
489/*
490 * Most likely _very_ bad one - but then it's hardly critical for small
491 * /dev and can be fixed when somebody will need really large one.
492 * Keep in mind that it will be fed through icache hash function too.
493 */
494static inline unsigned long hash(dev_t dev)
495{
496 return MAJOR(dev)+MINOR(dev);
497}
498
499static int bdev_test(struct inode *inode, void *data)
500{
501 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
502}
503
504static int bdev_set(struct inode *inode, void *data)
505{
506 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
507 return 0;
508}
509
510static LIST_HEAD(all_bdevs);
511
512struct block_device *bdget(dev_t dev)
513{
514 struct block_device *bdev;
515 struct inode *inode;
516
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800517 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 bdev_test, bdev_set, &dev);
519
520 if (!inode)
521 return NULL;
522
523 bdev = &BDEV_I(inode)->bdev;
524
525 if (inode->i_state & I_NEW) {
526 bdev->bd_contains = NULL;
527 bdev->bd_inode = inode;
528 bdev->bd_block_size = (1 << inode->i_blkbits);
529 bdev->bd_part_count = 0;
530 bdev->bd_invalidated = 0;
531 inode->i_mode = S_IFBLK;
532 inode->i_rdev = dev;
533 inode->i_bdev = bdev;
534 inode->i_data.a_ops = &def_blk_aops;
535 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
536 inode->i_data.backing_dev_info = &default_backing_dev_info;
537 spin_lock(&bdev_lock);
538 list_add(&bdev->bd_list, &all_bdevs);
539 spin_unlock(&bdev_lock);
540 unlock_new_inode(inode);
541 }
542 return bdev;
543}
544
545EXPORT_SYMBOL(bdget);
546
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200547/**
548 * bdgrab -- Grab a reference to an already referenced block device
549 * @bdev: Block device to grab a reference to.
550 */
551struct block_device *bdgrab(struct block_device *bdev)
552{
553 atomic_inc(&bdev->bd_inode->i_count);
554 return bdev;
555}
556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557long nr_blockdev_pages(void)
558{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700559 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 long ret = 0;
561 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700562 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 ret += bdev->bd_inode->i_mapping->nrpages;
564 }
565 spin_unlock(&bdev_lock);
566 return ret;
567}
568
569void bdput(struct block_device *bdev)
570{
571 iput(bdev->bd_inode);
572}
573
574EXPORT_SYMBOL(bdput);
575
576static struct block_device *bd_acquire(struct inode *inode)
577{
578 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 spin_lock(&bdev_lock);
581 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700582 if (bdev) {
583 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 spin_unlock(&bdev_lock);
585 return bdev;
586 }
587 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 bdev = bdget(inode->i_rdev);
590 if (bdev) {
591 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700592 if (!inode->i_bdev) {
593 /*
594 * We take an additional bd_inode->i_count for inode,
595 * and it's released in clear_inode() of inode.
596 * So, we can access it via ->i_mapping always
597 * without igrab().
598 */
599 atomic_inc(&bdev->bd_inode->i_count);
600 inode->i_bdev = bdev;
601 inode->i_mapping = bdev->bd_inode->i_mapping;
602 list_add(&inode->i_devices, &bdev->bd_inodes);
603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 spin_unlock(&bdev_lock);
605 }
606 return bdev;
607}
608
609/* Call when you free inode */
610
611void bd_forget(struct inode *inode)
612{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700613 struct block_device *bdev = NULL;
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700616 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800617 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700618 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700622
623 if (bdev)
624 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900627/**
628 * bd_may_claim - test whether a block device can be claimed
629 * @bdev: block device of interest
630 * @whole: whole block device containing @bdev, may equal @bdev
631 * @holder: holder trying to claim @bdev
632 *
633 * Test whther @bdev can be claimed by @holder.
634 *
635 * CONTEXT:
636 * spin_lock(&bdev_lock).
637 *
638 * RETURNS:
639 * %true if @bdev can be claimed, %false otherwise.
640 */
641static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
642 void *holder)
643{
644 if (bdev->bd_holder == holder)
645 return true; /* already a holder */
646 else if (bdev->bd_holder != NULL)
647 return false; /* held by someone else */
648 else if (bdev->bd_contains == bdev)
649 return true; /* is a whole device which isn't held */
650
651 else if (whole->bd_holder == bd_claim)
652 return true; /* is a partition of a device that is being partitioned */
653 else if (whole->bd_holder != NULL)
654 return false; /* is a partition of a held device */
655 else
656 return true; /* is a partition of an un-held device */
657}
658
659/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900660 * bd_prepare_to_claim - prepare to claim a block device
661 * @bdev: block device of interest
662 * @whole: the whole device containing @bdev, may equal @bdev
663 * @holder: holder trying to claim @bdev
664 *
665 * Prepare to claim @bdev. This function fails if @bdev is already
666 * claimed by another holder and waits if another claiming is in
667 * progress. This function doesn't actually claim. On successful
668 * return, the caller has ownership of bd_claiming and bd_holder[s].
669 *
670 * CONTEXT:
671 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
672 * it multiple times.
673 *
674 * RETURNS:
675 * 0 if @bdev can be claimed, -EBUSY otherwise.
676 */
677static int bd_prepare_to_claim(struct block_device *bdev,
678 struct block_device *whole, void *holder)
679{
680retry:
681 /* if someone else claimed, fail */
682 if (!bd_may_claim(bdev, whole, holder))
683 return -EBUSY;
684
Tejun Heoe75aa852010-08-04 17:59:39 +0200685 /* if claiming is already in progress, wait for it to finish */
686 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900687 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
688 DEFINE_WAIT(wait);
689
690 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
691 spin_unlock(&bdev_lock);
692 schedule();
693 finish_wait(wq, &wait);
694 spin_lock(&bdev_lock);
695 goto retry;
696 }
697
698 /* yay, all mine */
699 return 0;
700}
701
702/**
703 * bd_start_claiming - start claiming a block device
704 * @bdev: block device of interest
705 * @holder: holder trying to claim @bdev
706 *
707 * @bdev is about to be opened exclusively. Check @bdev can be opened
708 * exclusively and mark that an exclusive open is in progress. Each
709 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000710 * either bd_finish_claiming() or bd_abort_claiming() (which do not
711 * fail).
712 *
713 * This function is used to gain exclusive access to the block device
714 * without actually causing other exclusive open attempts to fail. It
715 * should be used when the open sequence itself requires exclusive
716 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900717 *
718 * CONTEXT:
719 * Might sleep.
720 *
721 * RETURNS:
722 * Pointer to the block device containing @bdev on success, ERR_PTR()
723 * value on failure.
724 */
725static struct block_device *bd_start_claiming(struct block_device *bdev,
726 void *holder)
727{
728 struct gendisk *disk;
729 struct block_device *whole;
730 int partno, err;
731
732 might_sleep();
733
734 /*
735 * @bdev might not have been initialized properly yet, look up
736 * and grab the outer block device the hard way.
737 */
738 disk = get_gendisk(bdev->bd_dev, &partno);
739 if (!disk)
740 return ERR_PTR(-ENXIO);
741
742 whole = bdget_disk(disk, 0);
Nick Piggincf342572010-05-26 01:50:21 +1000743 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900744 put_disk(disk);
745 if (!whole)
746 return ERR_PTR(-ENOMEM);
747
748 /* prepare to claim, if successful, mark claiming in progress */
749 spin_lock(&bdev_lock);
750
751 err = bd_prepare_to_claim(bdev, whole, holder);
752 if (err == 0) {
753 whole->bd_claiming = holder;
754 spin_unlock(&bdev_lock);
755 return whole;
756 } else {
757 spin_unlock(&bdev_lock);
758 bdput(whole);
759 return ERR_PTR(err);
760 }
761}
762
763/* releases bdev_lock */
764static void __bd_abort_claiming(struct block_device *whole, void *holder)
765{
766 BUG_ON(whole->bd_claiming != holder);
767 whole->bd_claiming = NULL;
768 wake_up_bit(&whole->bd_claiming, 0);
769
770 spin_unlock(&bdev_lock);
771 bdput(whole);
772}
773
774/**
775 * bd_abort_claiming - abort claiming a block device
776 * @whole: whole block device returned by bd_start_claiming()
777 * @holder: holder trying to claim @bdev
778 *
779 * Abort a claiming block started by bd_start_claiming(). Note that
780 * @whole is not the block device to be claimed but the whole device
781 * returned by bd_start_claiming().
782 *
783 * CONTEXT:
784 * Grabs and releases bdev_lock.
785 */
786static void bd_abort_claiming(struct block_device *whole, void *holder)
787{
788 spin_lock(&bdev_lock);
789 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
790}
791
Nick Pigginb0018362010-05-26 01:51:19 +1000792/* increment holders when we have a legitimate claim. requires bdev_lock */
793static void __bd_claim(struct block_device *bdev, struct block_device *whole,
794 void *holder)
795{
796 /* note that for a whole device bd_holders
797 * will be incremented twice, and bd_holder will
798 * be set to bd_claim before being set to holder
799 */
800 whole->bd_holders++;
801 whole->bd_holder = bd_claim;
802 bdev->bd_holders++;
803 bdev->bd_holder = holder;
804}
805
806/**
807 * bd_finish_claiming - finish claiming a block device
808 * @bdev: block device of interest (passed to bd_start_claiming())
809 * @whole: whole block device returned by bd_start_claiming()
810 * @holder: holder trying to claim @bdev
811 *
812 * Finish a claiming block started by bd_start_claiming().
813 *
814 * CONTEXT:
815 * Grabs and releases bdev_lock.
816 */
817static void bd_finish_claiming(struct block_device *bdev,
818 struct block_device *whole, void *holder)
819{
820 spin_lock(&bdev_lock);
Nick Pigginb0018362010-05-26 01:51:19 +1000821 BUG_ON(!bd_may_claim(bdev, whole, holder));
822 __bd_claim(bdev, whole, holder);
823 __bd_abort_claiming(whole, holder); /* not actually an abort */
824}
825
Tejun Heo6b4517a2010-04-07 18:53:59 +0900826/**
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900827 * bd_claim - claim a block device
828 * @bdev: block device to claim
829 * @holder: holder trying to claim @bdev
830 *
Nick Pigginb0018362010-05-26 01:51:19 +1000831 * Try to claim @bdev which must have been opened successfully.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900832 *
833 * CONTEXT:
834 * Might sleep.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900835 *
836 * RETURNS:
837 * 0 if successful, -EBUSY if @bdev is already claimed.
838 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839int bd_claim(struct block_device *bdev, void *holder)
840{
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900841 struct block_device *whole = bdev->bd_contains;
Tejun Heo6b4517a2010-04-07 18:53:59 +0900842 int res;
843
844 might_sleep();
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 spin_lock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900847 res = bd_prepare_to_claim(bdev, whole, holder);
Nick Pigginb0018362010-05-26 01:51:19 +1000848 if (res == 0)
849 __bd_claim(bdev, whole, holder);
850 spin_unlock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 return res;
853}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854EXPORT_SYMBOL(bd_claim);
855
856void bd_release(struct block_device *bdev)
857{
858 spin_lock(&bdev_lock);
859 if (!--bdev->bd_contains->bd_holders)
860 bdev->bd_contains->bd_holder = NULL;
861 if (!--bdev->bd_holders)
862 bdev->bd_holder = NULL;
863 spin_unlock(&bdev_lock);
864}
865
866EXPORT_SYMBOL(bd_release);
867
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800868#ifdef CONFIG_SYSFS
869/*
870 * Functions for bd_claim_by_kobject / bd_release_from_kobject
871 *
872 * If a kobject is passed to bd_claim_by_kobject()
873 * and the kobject has a parent directory,
874 * following symlinks are created:
875 * o from the kobject to the claimed bdev
876 * o from "holders" directory of the bdev to the parent of the kobject
877 * bd_release_from_kobject() removes these symlinks.
878 *
879 * Example:
880 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
881 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
882 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
883 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
884 */
885
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700886static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800887{
888 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700889 return 0;
890 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800891}
892
893static void del_symlink(struct kobject *from, struct kobject *to)
894{
895 if (!from || !to)
896 return;
897 sysfs_remove_link(from, kobject_name(to));
898}
899
900/*
901 * 'struct bd_holder' contains pointers to kobjects symlinked by
902 * bd_claim_by_kobject.
903 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
904 */
905struct bd_holder {
906 struct list_head list; /* chain of holders of the bdev */
907 int count; /* references from the holder */
908 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
909 struct kobject *hdev; /* e.g. "/block/dm-0" */
910 struct kobject *hdir; /* e.g. "/block/sda/holders" */
911 struct kobject *sdev; /* e.g. "/block/sda" */
912};
913
914/*
915 * Get references of related kobjects at once.
916 * Returns 1 on success. 0 on failure.
917 *
918 * Should call bd_holder_release_dirs() after successful use.
919 */
920static int bd_holder_grab_dirs(struct block_device *bdev,
921 struct bd_holder *bo)
922{
923 if (!bdev || !bo)
924 return 0;
925
926 bo->sdir = kobject_get(bo->sdir);
927 if (!bo->sdir)
928 return 0;
929
930 bo->hdev = kobject_get(bo->sdir->parent);
931 if (!bo->hdev)
932 goto fail_put_sdir;
933
Tejun Heo0762b8b2008-08-25 19:56:12 +0900934 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800935 if (!bo->sdev)
936 goto fail_put_hdev;
937
Tejun Heo4c465012008-08-25 19:56:11 +0900938 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800939 if (!bo->hdir)
940 goto fail_put_sdev;
941
942 return 1;
943
944fail_put_sdev:
945 kobject_put(bo->sdev);
946fail_put_hdev:
947 kobject_put(bo->hdev);
948fail_put_sdir:
949 kobject_put(bo->sdir);
950
951 return 0;
952}
953
954/* Put references of related kobjects at once. */
955static void bd_holder_release_dirs(struct bd_holder *bo)
956{
957 kobject_put(bo->hdir);
958 kobject_put(bo->sdev);
959 kobject_put(bo->hdev);
960 kobject_put(bo->sdir);
961}
962
963static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
964{
965 struct bd_holder *bo;
966
967 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
968 if (!bo)
969 return NULL;
970
971 bo->count = 1;
972 bo->sdir = kobj;
973
974 return bo;
975}
976
977static void free_bd_holder(struct bd_holder *bo)
978{
979 kfree(bo);
980}
981
982/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500983 * find_bd_holder - find matching struct bd_holder from the block device
984 *
985 * @bdev: struct block device to be searched
986 * @bo: target struct bd_holder
987 *
988 * Returns matching entry with @bo in @bdev->bd_holder_list.
989 * If found, increment the reference count and return the pointer.
990 * If not found, returns NULL.
991 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800992static struct bd_holder *find_bd_holder(struct block_device *bdev,
993 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500994{
995 struct bd_holder *tmp;
996
997 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
998 if (tmp->sdir == bo->sdir) {
999 tmp->count++;
1000 return tmp;
1001 }
1002
1003 return NULL;
1004}
1005
1006/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001007 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1008 *
1009 * @bdev: block device to be bd_claimed
1010 * @bo: preallocated and initialized by alloc_bd_holder()
1011 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001012 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001013 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001014 * Returns 0 if symlinks are created.
1015 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001016 */
1017static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1018{
Johannes Weiner4e916722007-07-15 23:41:25 -07001019 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001020
1021 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001022 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001023
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001024 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001025 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001026
Johannes Weiner4e916722007-07-15 23:41:25 -07001027 err = add_symlink(bo->sdir, bo->sdev);
1028 if (err)
1029 return err;
1030
1031 err = add_symlink(bo->hdir, bo->hdev);
1032 if (err) {
1033 del_symlink(bo->sdir, bo->sdev);
1034 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001035 }
Johannes Weiner4e916722007-07-15 23:41:25 -07001036
1037 list_add_tail(&bo->list, &bdev->bd_holder_list);
1038 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001039}
1040
1041/**
1042 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1043 *
1044 * @bdev: block device to be bd_claimed
1045 * @kobj: holder's kobject
1046 *
1047 * If there is matching entry with @kobj in @bdev->bd_holder_list
1048 * and no other bd_claim() from the same kobject,
1049 * remove the struct bd_holder from the list, delete symlinks for it.
1050 *
1051 * Returns a pointer to the struct bd_holder when it's removed from the list
1052 * and ready to be freed.
1053 * Returns NULL if matching claim isn't found or there is other bd_claim()
1054 * by the same kobject.
1055 */
1056static struct bd_holder *del_bd_holder(struct block_device *bdev,
1057 struct kobject *kobj)
1058{
1059 struct bd_holder *bo;
1060
1061 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1062 if (bo->sdir == kobj) {
1063 bo->count--;
1064 BUG_ON(bo->count < 0);
1065 if (!bo->count) {
1066 list_del(&bo->list);
1067 del_symlink(bo->sdir, bo->sdev);
1068 del_symlink(bo->hdir, bo->hdev);
1069 bd_holder_release_dirs(bo);
1070 return bo;
1071 }
1072 break;
1073 }
1074 }
1075
1076 return NULL;
1077}
1078
1079/**
1080 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1081 *
1082 * @bdev: block device to be claimed
1083 * @holder: holder's signature
1084 * @kobj: holder's kobject
1085 *
1086 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1087 * the bdev and the holder's kobject.
1088 * Use bd_release_from_kobject() when relesing the claimed bdev.
1089 *
1090 * Returns 0 on success. (same as bd_claim())
1091 * Returns errno on failure.
1092 */
1093static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1094 struct kobject *kobj)
1095{
Johannes Weiner4e916722007-07-15 23:41:25 -07001096 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001097 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001098
1099 if (!kobj)
1100 return -EINVAL;
1101
1102 bo = alloc_bd_holder(kobj);
1103 if (!bo)
1104 return -ENOMEM;
1105
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001106 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001107
Johannes Weiner4e916722007-07-15 23:41:25 -07001108 err = bd_claim(bdev, holder);
1109 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -07001110 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001111
1112 found = find_bd_holder(bdev, bo);
1113 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -07001114 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001115
1116 err = add_bd_holder(bdev, bo);
1117 if (err)
1118 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001119 else
1120 bo = NULL;
1121fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001122 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -07001123 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -07001124 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001125}
1126
1127/**
1128 * bd_release_from_kobject - bd_release() with additional kobject signature
1129 *
1130 * @bdev: block device to be released
1131 * @kobj: holder's kobject
1132 *
1133 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1134 */
1135static void bd_release_from_kobject(struct block_device *bdev,
1136 struct kobject *kobj)
1137{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001138 if (!kobj)
1139 return;
1140
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001141 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001142 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001143 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001144 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001145}
1146
1147/**
1148 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1149 *
1150 * @bdev: block device to be claimed
1151 * @holder: holder's signature
1152 * @disk: holder's gendisk
1153 *
1154 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1155 */
1156int bd_claim_by_disk(struct block_device *bdev, void *holder,
1157 struct gendisk *disk)
1158{
1159 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1160}
1161EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1162
1163/**
1164 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1165 *
1166 * @bdev: block device to be claimed
1167 * @disk: holder's gendisk
1168 *
1169 * Call bd_release_from_kobject() and put @disk->slave_dir.
1170 */
1171void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1172{
1173 bd_release_from_kobject(bdev, disk->slave_dir);
1174 kobject_put(disk->slave_dir);
1175}
1176EXPORT_SYMBOL_GPL(bd_release_from_disk);
1177#endif
1178
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179/*
1180 * Tries to open block device by device number. Use it ONLY if you
1181 * really do not have anything better - i.e. when you are behind a
1182 * truly sucky interface and all you are given is a device number. _Never_
1183 * to be used for internal purposes. If you ever need it - reconsider
1184 * your API.
1185 */
Al Viroaeb5d722008-09-02 15:28:45 -04001186struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187{
1188 struct block_device *bdev = bdget(dev);
1189 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001191 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return err ? ERR_PTR(err) : bdev;
1193}
1194
1195EXPORT_SYMBOL(open_by_devnum);
1196
Andrew Patterson0c002c22008-09-04 14:27:20 -06001197/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001198 * flush_disk - invalidates all buffer-cache entries on a disk
1199 *
1200 * @bdev: struct block device to be flushed
1201 *
1202 * Invalidates all buffer-cache entries on a disk. It should be called
1203 * when a disk has been changed -- either by a media change or online
1204 * resize.
1205 */
1206static void flush_disk(struct block_device *bdev)
1207{
1208 if (__invalidate_device(bdev)) {
1209 char name[BDEVNAME_SIZE] = "";
1210
1211 if (bdev->bd_disk)
1212 disk_name(bdev->bd_disk, 0, name);
1213 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1214 "resized disk %s\n", name);
1215 }
1216
1217 if (!bdev->bd_disk)
1218 return;
1219 if (disk_partitionable(bdev->bd_disk))
1220 bdev->bd_invalidated = 1;
1221}
1222
1223/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001224 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001225 * @disk: struct gendisk to check
1226 * @bdev: struct bdev to adjust.
1227 *
1228 * This routine checks to see if the bdev size does not match the disk size
1229 * and adjusts it if it differs.
1230 */
1231void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1232{
1233 loff_t disk_size, bdev_size;
1234
1235 disk_size = (loff_t)get_capacity(disk) << 9;
1236 bdev_size = i_size_read(bdev->bd_inode);
1237 if (disk_size != bdev_size) {
1238 char name[BDEVNAME_SIZE];
1239
1240 disk_name(disk, 0, name);
1241 printk(KERN_INFO
1242 "%s: detected capacity change from %lld to %lld\n",
1243 name, bdev_size, disk_size);
1244 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001245 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001246 }
1247}
1248EXPORT_SYMBOL(check_disk_size_change);
1249
1250/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001251 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001252 * @disk: struct gendisk to be revalidated
1253 *
1254 * This routine is a wrapper for lower-level driver's revalidate_disk
1255 * call-backs. It is used to do common pre and post operations needed
1256 * for all revalidate_disk operations.
1257 */
1258int revalidate_disk(struct gendisk *disk)
1259{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001260 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001261 int ret = 0;
1262
1263 if (disk->fops->revalidate_disk)
1264 ret = disk->fops->revalidate_disk(disk);
1265
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001266 bdev = bdget_disk(disk, 0);
1267 if (!bdev)
1268 return ret;
1269
1270 mutex_lock(&bdev->bd_mutex);
1271 check_disk_size_change(disk, bdev);
1272 mutex_unlock(&bdev->bd_mutex);
1273 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001274 return ret;
1275}
1276EXPORT_SYMBOL(revalidate_disk);
1277
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278/*
1279 * This routine checks whether a removable media has been changed,
1280 * and invalidates all buffer-cache-entries in that case. This
1281 * is a relatively slow routine, so we have to try to minimize using
1282 * it. Thus it is called only upon a 'mount' or 'open'. This
1283 * is the best way of combining speed and utility, I think.
1284 * People changing diskettes in the middle of an operation deserve
1285 * to lose :-)
1286 */
1287int check_disk_change(struct block_device *bdev)
1288{
1289 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001290 const struct block_device_operations *bdops = disk->fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 if (!bdops->media_changed)
1293 return 0;
1294 if (!bdops->media_changed(bdev->bd_disk))
1295 return 0;
1296
Andrew Patterson56ade442008-09-04 14:27:40 -06001297 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if (bdops->revalidate_disk)
1299 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 return 1;
1301}
1302
1303EXPORT_SYMBOL(check_disk_change);
1304
1305void bd_set_size(struct block_device *bdev, loff_t size)
1306{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001307 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308
1309 bdev->bd_inode->i_size = size;
1310 while (bsize < PAGE_CACHE_SIZE) {
1311 if (size & bsize)
1312 break;
1313 bsize <<= 1;
1314 }
1315 bdev->bd_block_size = bsize;
1316 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1317}
1318EXPORT_SYMBOL(bd_set_size);
1319
Al Viro9a1c3542008-02-22 20:40:24 -05001320static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001321
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001322/*
1323 * bd_mutex locking:
1324 *
1325 * mutex_lock(part->bd_mutex)
1326 * mutex_lock_nested(whole->bd_mutex, 1)
1327 */
1328
Al Viro572c4892007-10-08 13:24:05 -04001329static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001332 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001333 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001334 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
Al Viro572c4892007-10-08 13:24:05 -04001336 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001337 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001338 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001339 perm |= MAY_WRITE;
1340 /*
1341 * hooks: /n/, see "layering violations".
1342 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001343 if (!for_part) {
1344 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1345 if (ret != 0) {
1346 bdput(bdev);
1347 return ret;
1348 }
Al Viro82666022008-08-01 05:32:04 -04001349 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001350
NeilBrownd3374822009-01-09 08:31:10 +11001351 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001352
Tejun Heo89f97492008-11-05 10:21:06 +01001353 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001354 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001355 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001356 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
NeilBrown6796bf52006-12-08 02:36:16 -08001358 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 if (!bdev->bd_openers) {
1360 bdev->bd_disk = disk;
1361 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001362 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001364
1365 ret = -ENXIO;
1366 bdev->bd_part = disk_get_part(disk, partno);
1367 if (!bdev->bd_part)
1368 goto out_clear;
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001371 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001372 if (ret == -ERESTARTSYS) {
1373 /* Lost a race with 'disk' being
1374 * deleted, try again.
1375 * See md.c
1376 */
1377 disk_put_part(bdev->bd_part);
1378 bdev->bd_part = NULL;
1379 module_put(disk->fops->owner);
1380 put_disk(disk);
1381 bdev->bd_disk = NULL;
1382 mutex_unlock(&bdev->bd_mutex);
1383 goto restart;
1384 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001386 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 }
1388 if (!bdev->bd_openers) {
1389 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1390 bdi = blk_get_backing_dev_info(bdev);
1391 if (bdi == NULL)
1392 bdi = &default_backing_dev_info;
1393 bdev->bd_inode->i_data.backing_dev_info = bdi;
1394 }
1395 if (bdev->bd_invalidated)
1396 rescan_partitions(disk, bdev);
1397 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 struct block_device *whole;
1399 whole = bdget_disk(disk, 0);
1400 ret = -ENOMEM;
1401 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001402 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001403 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001404 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001406 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 bdev->bd_inode->i_data.backing_dev_info =
1409 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001410 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001411 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001412 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001414 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 }
Tejun Heo89f97492008-11-05 10:21:06 +01001416 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 }
1418 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001419 module_put(disk->fops->owner);
Neil Brown960cc0f2009-10-26 08:59:17 +01001420 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001421 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if (bdev->bd_contains == bdev) {
1423 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001424 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001426 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 }
1428 if (bdev->bd_invalidated)
1429 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 }
1431 }
1432 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001433 if (for_part)
1434 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001435 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return 0;
1437
Tejun Heo0762b8b2008-08-25 19:56:12 +09001438 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001439 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001441 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1443 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001444 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001446 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001447 mutex_unlock(&bdev->bd_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001448 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001449 if (disk)
1450 module_put(disk->fops->owner);
1451 put_disk(disk);
1452 bdput(bdev);
1453
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 return ret;
1455}
1456
Al Viro572c4892007-10-08 13:24:05 -04001457int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Al Viro572c4892007-10-08 13:24:05 -04001459 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001460}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461EXPORT_SYMBOL(blkdev_get);
1462
1463static int blkdev_open(struct inode * inode, struct file * filp)
1464{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001465 struct block_device *whole = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 struct block_device *bdev;
1467 int res;
1468
1469 /*
1470 * Preserve backwards compatibility and allow large file access
1471 * even if userspace doesn't ask for it explicitly. Some mkfs
1472 * binary needs it. We might want to drop this workaround
1473 * during an unstable branch.
1474 */
1475 filp->f_flags |= O_LARGEFILE;
1476
Al Viro572c4892007-10-08 13:24:05 -04001477 if (filp->f_flags & O_NDELAY)
1478 filp->f_mode |= FMODE_NDELAY;
1479 if (filp->f_flags & O_EXCL)
1480 filp->f_mode |= FMODE_EXCL;
1481 if ((filp->f_flags & O_ACCMODE) == 3)
1482 filp->f_mode |= FMODE_WRITE_IOCTL;
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001485 if (bdev == NULL)
1486 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Tejun Heo6b4517a2010-04-07 18:53:59 +09001488 if (filp->f_mode & FMODE_EXCL) {
1489 whole = bd_start_claiming(bdev, filp);
1490 if (IS_ERR(whole)) {
1491 bdput(bdev);
1492 return PTR_ERR(whole);
1493 }
1494 }
1495
Al Viro572c4892007-10-08 13:24:05 -04001496 filp->f_mapping = bdev->bd_inode->i_mapping;
1497
1498 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
Tejun Heo6b4517a2010-04-07 18:53:59 +09001500 if (whole) {
1501 if (res == 0)
Nick Pigginb0018362010-05-26 01:51:19 +10001502 bd_finish_claiming(bdev, whole, filp);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001503 else
1504 bd_abort_claiming(whole, filp);
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001505 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 return res;
1508}
1509
Al Viro9a1c3542008-02-22 20:40:24 -05001510static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001511{
1512 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001513 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001514 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001515
NeilBrown6796bf52006-12-08 02:36:16 -08001516 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001517 if (for_part)
1518 bdev->bd_part_count--;
1519
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001520 if (!--bdev->bd_openers) {
1521 sync_blockdev(bdev);
1522 kill_bdev(bdev);
1523 }
1524 if (bdev->bd_contains == bdev) {
1525 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001526 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001527 }
1528 if (!bdev->bd_openers) {
1529 struct module *owner = disk->fops->owner;
1530
1531 put_disk(disk);
1532 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001533 disk_put_part(bdev->bd_part);
1534 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535 bdev->bd_disk = NULL;
1536 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001537 if (bdev != bdev->bd_contains)
1538 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001539 bdev->bd_contains = NULL;
1540 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001541 mutex_unlock(&bdev->bd_mutex);
1542 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001543 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001544 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001545 return ret;
1546}
1547
Al Viro9a1c3542008-02-22 20:40:24 -05001548int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001549{
Al Viro9a1c3542008-02-22 20:40:24 -05001550 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001551}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001552EXPORT_SYMBOL(blkdev_put);
1553
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554static int blkdev_close(struct inode * inode, struct file * filp)
1555{
1556 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1557 if (bdev->bd_holder == filp)
1558 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001559 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001562static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Al Viro56b26ad2008-09-19 03:17:36 -04001564 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1565 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001566
1567 /*
1568 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1569 * to updated it before every ioctl.
1570 */
Al Viro56b26ad2008-09-19 03:17:36 -04001571 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001572 mode |= FMODE_NDELAY;
1573 else
1574 mode &= ~FMODE_NDELAY;
1575
Al Viro56b26ad2008-09-19 03:17:36 -04001576 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577}
1578
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001579/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001580 * Write data to the block device. Only intended for the block device itself
1581 * and the raw driver which basically is a fake block device.
1582 *
1583 * Does not take i_mutex for the write and thus is not for general purpose
1584 * use.
1585 */
1586ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1587 unsigned long nr_segs, loff_t pos)
1588{
1589 struct file *file = iocb->ki_filp;
1590 ssize_t ret;
1591
1592 BUG_ON(iocb->ki_pos != pos);
1593
1594 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1595 if (ret > 0 || ret == -EIOCBQUEUED) {
1596 ssize_t err;
1597
1598 err = generic_write_sync(file, pos, ret);
1599 if (err < 0 && ret > 0)
1600 ret = err;
1601 }
1602 return ret;
1603}
1604EXPORT_SYMBOL_GPL(blkdev_aio_write);
1605
1606/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001607 * Try to release a page associated with block device when the system
1608 * is under memory pressure.
1609 */
1610static int blkdev_releasepage(struct page *page, gfp_t wait)
1611{
1612 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1613
1614 if (super && super->s_op->bdev_try_to_free_page)
1615 return super->s_op->bdev_try_to_free_page(super, page, wait);
1616
1617 return try_to_free_buffers(page);
1618}
1619
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001620static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 .readpage = blkdev_readpage,
1622 .writepage = blkdev_writepage,
1623 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001624 .write_begin = blkdev_write_begin,
1625 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001627 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 .direct_IO = blkdev_direct_IO,
1629};
1630
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001631const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 .open = blkdev_open,
1633 .release = blkdev_close,
1634 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001635 .read = do_sync_read,
1636 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001638 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001640 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001641 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642#ifdef CONFIG_COMPAT
1643 .compat_ioctl = compat_blkdev_ioctl,
1644#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001645 .splice_read = generic_file_splice_read,
1646 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647};
1648
1649int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1650{
1651 int res;
1652 mm_segment_t old_fs = get_fs();
1653 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001654 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 set_fs(old_fs);
1656 return res;
1657}
1658
1659EXPORT_SYMBOL(ioctl_by_bdev);
1660
1661/**
1662 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001663 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001665 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 * namespace if possible and return it. Return ERR_PTR(error)
1667 * otherwise.
1668 */
Al Viro421748e2008-08-02 01:04:36 -04001669struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 struct block_device *bdev;
1672 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001673 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 int error;
1675
Al Viro421748e2008-08-02 01:04:36 -04001676 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return ERR_PTR(-EINVAL);
1678
Al Viro421748e2008-08-02 01:04:36 -04001679 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 if (error)
1681 return ERR_PTR(error);
1682
Al Viro421748e2008-08-02 01:04:36 -04001683 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 error = -ENOTBLK;
1685 if (!S_ISBLK(inode->i_mode))
1686 goto fail;
1687 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001688 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 goto fail;
1690 error = -ENOMEM;
1691 bdev = bd_acquire(inode);
1692 if (!bdev)
1693 goto fail;
1694out:
Al Viro421748e2008-08-02 01:04:36 -04001695 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 return bdev;
1697fail:
1698 bdev = ERR_PTR(error);
1699 goto out;
1700}
Al Virod5686b42008-08-01 05:00:11 -04001701EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
1703/**
Al Viro30c40d22008-02-22 19:50:45 -05001704 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 *
1706 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001707 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 * @holder: owner for exclusion
1709 *
1710 * Open the blockdevice described by the special file at @path, claim it
1711 * for the @holder.
1712 */
Al Viro30c40d22008-02-22 19:50:45 -05001713struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001715 struct block_device *bdev, *whole;
1716 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 bdev = lookup_bdev(path);
1719 if (IS_ERR(bdev))
1720 return bdev;
1721
Tejun Heo6b4517a2010-04-07 18:53:59 +09001722 whole = bd_start_claiming(bdev, holder);
1723 if (IS_ERR(whole)) {
1724 bdput(bdev);
1725 return whole;
1726 }
1727
Al Viro572c4892007-10-08 13:24:05 -04001728 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (error)
Tejun Heo6b4517a2010-04-07 18:53:59 +09001730 goto out_abort_claiming;
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001733 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Tejun Heo6b4517a2010-04-07 18:53:59 +09001734 goto out_blkdev_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
Nick Pigginb0018362010-05-26 01:51:19 +10001736 bd_finish_claiming(bdev, whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 return bdev;
Tejun Heo6b4517a2010-04-07 18:53:59 +09001738
1739out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001740 blkdev_put(bdev, mode);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001741out_abort_claiming:
1742 bd_abort_claiming(whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return ERR_PTR(error);
1744}
1745
Al Viro30c40d22008-02-22 19:50:45 -05001746EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748/**
Al Viro30c40d22008-02-22 19:50:45 -05001749 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 *
1751 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001752 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 *
Al Viro30c40d22008-02-22 19:50:45 -05001754 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 */
Al Viro30c40d22008-02-22 19:50:45 -05001756void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757{
1758 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001759 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760}
1761
Al Viro30c40d22008-02-22 19:50:45 -05001762EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001763
1764int __invalidate_device(struct block_device *bdev)
1765{
1766 struct super_block *sb = get_super(bdev);
1767 int res = 0;
1768
1769 if (sb) {
1770 /*
1771 * no need to lock the super, get_super holds the
1772 * read mutex so the filesystem cannot go away
1773 * under us (->put_super runs with the write lock
1774 * hold).
1775 */
1776 shrink_dcache_sb(sb);
1777 res = invalidate_inodes(sb);
1778 drop_super(sb);
1779 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001780 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001781 return res;
1782}
1783EXPORT_SYMBOL(__invalidate_device);