blob: 26e5f50266205a7e7d2069bb92dd5600e77fe61c [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;
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{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700311 *pagep = NULL;
312 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
313 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
315
Nick Piggin6272b5a2007-10-16 01:25:04 -0700316static int blkdev_write_end(struct file *file, struct address_space *mapping,
317 loff_t pos, unsigned len, unsigned copied,
318 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700320 int ret;
321 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
322
323 unlock_page(page);
324 page_cache_release(page);
325
326 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
329/*
330 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800331 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 * so we compute the size by hand (just as in block_read/write above)
333 */
334static loff_t block_llseek(struct file *file, loff_t offset, int origin)
335{
336 struct inode *bd_inode = file->f_mapping->host;
337 loff_t size;
338 loff_t retval;
339
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800340 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 size = i_size_read(bd_inode);
342
343 switch (origin) {
344 case 2:
345 offset += size;
346 break;
347 case 1:
348 offset += file->f_pos;
349 }
350 retval = -EINVAL;
351 if (offset >= 0 && offset <= size) {
352 if (offset != file->f_pos) {
353 file->f_pos = offset;
354 }
355 retval = offset;
356 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800357 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 return retval;
359}
360
361/*
362 * Filp is never NULL; the only case when ->fsync() is called with
363 * NULL first argument is nfsd_sync_dir() and that's not a directory.
364 */
365
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700366int blkdev_fsync(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400368 struct inode *bd_inode = filp->f_mapping->host;
369 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100370 int error;
371
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400372 /*
373 * There is no need to serialise calls to blkdev_issue_flush with
374 * i_mutex and doing so causes performance issues with concurrent
375 * O_SYNC writers to a block device.
376 */
377 mutex_unlock(&bd_inode->i_mutex);
378
Jens Axboe7407cf32010-04-29 09:36:24 +0200379 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL, BLKDEV_IFL_WAIT);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100380 if (error == -EOPNOTSUPP)
381 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400382
383 mutex_lock(&bd_inode->i_mutex);
384
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100385 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700387EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389/*
390 * pseudo-fs
391 */
392
393static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800394static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396static struct inode *bdev_alloc_inode(struct super_block *sb)
397{
Christoph Lametere94b1762006-12-06 20:33:17 -0800398 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (!ei)
400 return NULL;
401 return &ei->vfs_inode;
402}
403
404static void bdev_destroy_inode(struct inode *inode)
405{
406 struct bdev_inode *bdi = BDEV_I(inode);
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 kmem_cache_free(bdev_cachep, bdi);
409}
410
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700411static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
413 struct bdev_inode *ei = (struct bdev_inode *) foo;
414 struct block_device *bdev = &ei->bdev;
415
Christoph Lametera35afb82007-05-16 22:10:57 -0700416 memset(bdev, 0, sizeof(*bdev));
417 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700418 INIT_LIST_HEAD(&bdev->bd_inodes);
419 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800420#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700421 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800422#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700423 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800424 /* Initialize mutex for freeze. */
425 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
428static inline void __bd_forget(struct inode *inode)
429{
430 list_del_init(&inode->i_devices);
431 inode->i_bdev = NULL;
432 inode->i_mapping = &inode->i_data;
433}
434
435static void bdev_clear_inode(struct inode *inode)
436{
437 struct block_device *bdev = &BDEV_I(inode)->bdev;
438 struct list_head *p;
439 spin_lock(&bdev_lock);
440 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
441 __bd_forget(list_entry(p, struct inode, i_devices));
442 }
443 list_del_init(&bdev->bd_list);
444 spin_unlock(&bdev_lock);
445}
446
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800447static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 .statfs = simple_statfs,
449 .alloc_inode = bdev_alloc_inode,
450 .destroy_inode = bdev_destroy_inode,
451 .drop_inode = generic_delete_inode,
452 .clear_inode = bdev_clear_inode,
453};
454
David Howells454e2392006-06-23 02:02:57 -0700455static int bd_get_sb(struct file_system_type *fs_type,
456 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
David Howells454e2392006-06-23 02:02:57 -0700458 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
460
461static struct file_system_type bd_type = {
462 .name = "bdev",
463 .get_sb = bd_get_sb,
464 .kill_sb = kill_anon_super,
465};
466
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800467struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469void __init bdev_cache_init(void)
470{
471 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800472 struct vfsmount *bd_mnt;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800475 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
476 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900477 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 err = register_filesystem(&bd_type);
479 if (err)
480 panic("Cannot register bdev pseudo-fs");
481 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (IS_ERR(bd_mnt))
483 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100484 /*
485 * This vfsmount structure is only used to obtain the
486 * blockdev_superblock, so tell kmemleak not to report it.
487 */
488 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
490}
491
492/*
493 * Most likely _very_ bad one - but then it's hardly critical for small
494 * /dev and can be fixed when somebody will need really large one.
495 * Keep in mind that it will be fed through icache hash function too.
496 */
497static inline unsigned long hash(dev_t dev)
498{
499 return MAJOR(dev)+MINOR(dev);
500}
501
502static int bdev_test(struct inode *inode, void *data)
503{
504 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
505}
506
507static int bdev_set(struct inode *inode, void *data)
508{
509 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
510 return 0;
511}
512
513static LIST_HEAD(all_bdevs);
514
515struct block_device *bdget(dev_t dev)
516{
517 struct block_device *bdev;
518 struct inode *inode;
519
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800520 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 bdev_test, bdev_set, &dev);
522
523 if (!inode)
524 return NULL;
525
526 bdev = &BDEV_I(inode)->bdev;
527
528 if (inode->i_state & I_NEW) {
529 bdev->bd_contains = NULL;
530 bdev->bd_inode = inode;
531 bdev->bd_block_size = (1 << inode->i_blkbits);
532 bdev->bd_part_count = 0;
533 bdev->bd_invalidated = 0;
534 inode->i_mode = S_IFBLK;
535 inode->i_rdev = dev;
536 inode->i_bdev = bdev;
537 inode->i_data.a_ops = &def_blk_aops;
538 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
539 inode->i_data.backing_dev_info = &default_backing_dev_info;
540 spin_lock(&bdev_lock);
541 list_add(&bdev->bd_list, &all_bdevs);
542 spin_unlock(&bdev_lock);
543 unlock_new_inode(inode);
544 }
545 return bdev;
546}
547
548EXPORT_SYMBOL(bdget);
549
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200550/**
551 * bdgrab -- Grab a reference to an already referenced block device
552 * @bdev: Block device to grab a reference to.
553 */
554struct block_device *bdgrab(struct block_device *bdev)
555{
556 atomic_inc(&bdev->bd_inode->i_count);
557 return bdev;
558}
559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560long nr_blockdev_pages(void)
561{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700562 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 long ret = 0;
564 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700565 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 ret += bdev->bd_inode->i_mapping->nrpages;
567 }
568 spin_unlock(&bdev_lock);
569 return ret;
570}
571
572void bdput(struct block_device *bdev)
573{
574 iput(bdev->bd_inode);
575}
576
577EXPORT_SYMBOL(bdput);
578
579static struct block_device *bd_acquire(struct inode *inode)
580{
581 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 spin_lock(&bdev_lock);
584 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700585 if (bdev) {
586 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 spin_unlock(&bdev_lock);
588 return bdev;
589 }
590 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 bdev = bdget(inode->i_rdev);
593 if (bdev) {
594 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700595 if (!inode->i_bdev) {
596 /*
597 * We take an additional bd_inode->i_count for inode,
598 * and it's released in clear_inode() of inode.
599 * So, we can access it via ->i_mapping always
600 * without igrab().
601 */
602 atomic_inc(&bdev->bd_inode->i_count);
603 inode->i_bdev = bdev;
604 inode->i_mapping = bdev->bd_inode->i_mapping;
605 list_add(&inode->i_devices, &bdev->bd_inodes);
606 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 spin_unlock(&bdev_lock);
608 }
609 return bdev;
610}
611
612/* Call when you free inode */
613
614void bd_forget(struct inode *inode)
615{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700616 struct block_device *bdev = NULL;
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700619 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800620 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700621 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700625
626 if (bdev)
627 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900630/**
631 * bd_may_claim - test whether a block device can be claimed
632 * @bdev: block device of interest
633 * @whole: whole block device containing @bdev, may equal @bdev
634 * @holder: holder trying to claim @bdev
635 *
636 * Test whther @bdev can be claimed by @holder.
637 *
638 * CONTEXT:
639 * spin_lock(&bdev_lock).
640 *
641 * RETURNS:
642 * %true if @bdev can be claimed, %false otherwise.
643 */
644static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
645 void *holder)
646{
647 if (bdev->bd_holder == holder)
648 return true; /* already a holder */
649 else if (bdev->bd_holder != NULL)
650 return false; /* held by someone else */
651 else if (bdev->bd_contains == bdev)
652 return true; /* is a whole device which isn't held */
653
654 else if (whole->bd_holder == bd_claim)
655 return true; /* is a partition of a device that is being partitioned */
656 else if (whole->bd_holder != NULL)
657 return false; /* is a partition of a held device */
658 else
659 return true; /* is a partition of an un-held device */
660}
661
662/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900663 * bd_prepare_to_claim - prepare to claim a block device
664 * @bdev: block device of interest
665 * @whole: the whole device containing @bdev, may equal @bdev
666 * @holder: holder trying to claim @bdev
667 *
668 * Prepare to claim @bdev. This function fails if @bdev is already
669 * claimed by another holder and waits if another claiming is in
670 * progress. This function doesn't actually claim. On successful
671 * return, the caller has ownership of bd_claiming and bd_holder[s].
672 *
673 * CONTEXT:
674 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
675 * it multiple times.
676 *
677 * RETURNS:
678 * 0 if @bdev can be claimed, -EBUSY otherwise.
679 */
680static int bd_prepare_to_claim(struct block_device *bdev,
681 struct block_device *whole, void *holder)
682{
683retry:
684 /* if someone else claimed, fail */
685 if (!bd_may_claim(bdev, whole, holder))
686 return -EBUSY;
687
688 /* if someone else is claiming, wait for it to finish */
689 if (whole->bd_claiming && whole->bd_claiming != holder) {
690 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
691 DEFINE_WAIT(wait);
692
693 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
694 spin_unlock(&bdev_lock);
695 schedule();
696 finish_wait(wq, &wait);
697 spin_lock(&bdev_lock);
698 goto retry;
699 }
700
701 /* yay, all mine */
702 return 0;
703}
704
705/**
706 * bd_start_claiming - start claiming a block device
707 * @bdev: block device of interest
708 * @holder: holder trying to claim @bdev
709 *
710 * @bdev is about to be opened exclusively. Check @bdev can be opened
711 * exclusively and mark that an exclusive open is in progress. Each
712 * successful call to this function must be matched with a call to
713 * either bd_claim() or bd_abort_claiming(). If this function
714 * succeeds, the matching bd_claim() is guaranteed to succeed.
715 *
716 * CONTEXT:
717 * Might sleep.
718 *
719 * RETURNS:
720 * Pointer to the block device containing @bdev on success, ERR_PTR()
721 * value on failure.
722 */
723static struct block_device *bd_start_claiming(struct block_device *bdev,
724 void *holder)
725{
726 struct gendisk *disk;
727 struct block_device *whole;
728 int partno, err;
729
730 might_sleep();
731
732 /*
733 * @bdev might not have been initialized properly yet, look up
734 * and grab the outer block device the hard way.
735 */
736 disk = get_gendisk(bdev->bd_dev, &partno);
737 if (!disk)
738 return ERR_PTR(-ENXIO);
739
740 whole = bdget_disk(disk, 0);
741 put_disk(disk);
742 if (!whole)
743 return ERR_PTR(-ENOMEM);
744
745 /* prepare to claim, if successful, mark claiming in progress */
746 spin_lock(&bdev_lock);
747
748 err = bd_prepare_to_claim(bdev, whole, holder);
749 if (err == 0) {
750 whole->bd_claiming = holder;
751 spin_unlock(&bdev_lock);
752 return whole;
753 } else {
754 spin_unlock(&bdev_lock);
755 bdput(whole);
756 return ERR_PTR(err);
757 }
758}
759
760/* releases bdev_lock */
761static void __bd_abort_claiming(struct block_device *whole, void *holder)
762{
763 BUG_ON(whole->bd_claiming != holder);
764 whole->bd_claiming = NULL;
765 wake_up_bit(&whole->bd_claiming, 0);
766
767 spin_unlock(&bdev_lock);
768 bdput(whole);
769}
770
771/**
772 * bd_abort_claiming - abort claiming a block device
773 * @whole: whole block device returned by bd_start_claiming()
774 * @holder: holder trying to claim @bdev
775 *
776 * Abort a claiming block started by bd_start_claiming(). Note that
777 * @whole is not the block device to be claimed but the whole device
778 * returned by bd_start_claiming().
779 *
780 * CONTEXT:
781 * Grabs and releases bdev_lock.
782 */
783static void bd_abort_claiming(struct block_device *whole, void *holder)
784{
785 spin_lock(&bdev_lock);
786 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
787}
788
789/**
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900790 * bd_claim - claim a block device
791 * @bdev: block device to claim
792 * @holder: holder trying to claim @bdev
793 *
Tejun Heo6b4517a2010-04-07 18:53:59 +0900794 * Try to claim @bdev which must have been opened successfully. This
795 * function may be called with or without preceding
796 * blk_start_claiming(). In the former case, this function is always
797 * successful and terminates the claiming block.
798 *
799 * CONTEXT:
800 * Might sleep.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900801 *
802 * RETURNS:
803 * 0 if successful, -EBUSY if @bdev is already claimed.
804 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805int bd_claim(struct block_device *bdev, void *holder)
806{
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900807 struct block_device *whole = bdev->bd_contains;
Tejun Heo6b4517a2010-04-07 18:53:59 +0900808 int res;
809
810 might_sleep();
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900811
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 spin_lock(&bdev_lock);
813
Tejun Heo6b4517a2010-04-07 18:53:59 +0900814 res = bd_prepare_to_claim(bdev, whole, holder);
815 if (res == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 /* note that for a whole device bd_holders
817 * will be incremented twice, and bd_holder will
818 * be set to bd_claim before being set to holder
819 */
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900820 whole->bd_holders++;
821 whole->bd_holder = bd_claim;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 bdev->bd_holders++;
823 bdev->bd_holder = holder;
824 }
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900825
Tejun Heo6b4517a2010-04-07 18:53:59 +0900826 if (whole->bd_claiming)
827 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
828 else
829 spin_unlock(&bdev_lock);
830
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return res;
832}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833EXPORT_SYMBOL(bd_claim);
834
835void bd_release(struct block_device *bdev)
836{
837 spin_lock(&bdev_lock);
838 if (!--bdev->bd_contains->bd_holders)
839 bdev->bd_contains->bd_holder = NULL;
840 if (!--bdev->bd_holders)
841 bdev->bd_holder = NULL;
842 spin_unlock(&bdev_lock);
843}
844
845EXPORT_SYMBOL(bd_release);
846
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800847#ifdef CONFIG_SYSFS
848/*
849 * Functions for bd_claim_by_kobject / bd_release_from_kobject
850 *
851 * If a kobject is passed to bd_claim_by_kobject()
852 * and the kobject has a parent directory,
853 * following symlinks are created:
854 * o from the kobject to the claimed bdev
855 * o from "holders" directory of the bdev to the parent of the kobject
856 * bd_release_from_kobject() removes these symlinks.
857 *
858 * Example:
859 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
860 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
861 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
862 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
863 */
864
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700865static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800866{
867 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700868 return 0;
869 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800870}
871
872static void del_symlink(struct kobject *from, struct kobject *to)
873{
874 if (!from || !to)
875 return;
876 sysfs_remove_link(from, kobject_name(to));
877}
878
879/*
880 * 'struct bd_holder' contains pointers to kobjects symlinked by
881 * bd_claim_by_kobject.
882 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
883 */
884struct bd_holder {
885 struct list_head list; /* chain of holders of the bdev */
886 int count; /* references from the holder */
887 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
888 struct kobject *hdev; /* e.g. "/block/dm-0" */
889 struct kobject *hdir; /* e.g. "/block/sda/holders" */
890 struct kobject *sdev; /* e.g. "/block/sda" */
891};
892
893/*
894 * Get references of related kobjects at once.
895 * Returns 1 on success. 0 on failure.
896 *
897 * Should call bd_holder_release_dirs() after successful use.
898 */
899static int bd_holder_grab_dirs(struct block_device *bdev,
900 struct bd_holder *bo)
901{
902 if (!bdev || !bo)
903 return 0;
904
905 bo->sdir = kobject_get(bo->sdir);
906 if (!bo->sdir)
907 return 0;
908
909 bo->hdev = kobject_get(bo->sdir->parent);
910 if (!bo->hdev)
911 goto fail_put_sdir;
912
Tejun Heo0762b8b2008-08-25 19:56:12 +0900913 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800914 if (!bo->sdev)
915 goto fail_put_hdev;
916
Tejun Heo4c465012008-08-25 19:56:11 +0900917 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800918 if (!bo->hdir)
919 goto fail_put_sdev;
920
921 return 1;
922
923fail_put_sdev:
924 kobject_put(bo->sdev);
925fail_put_hdev:
926 kobject_put(bo->hdev);
927fail_put_sdir:
928 kobject_put(bo->sdir);
929
930 return 0;
931}
932
933/* Put references of related kobjects at once. */
934static void bd_holder_release_dirs(struct bd_holder *bo)
935{
936 kobject_put(bo->hdir);
937 kobject_put(bo->sdev);
938 kobject_put(bo->hdev);
939 kobject_put(bo->sdir);
940}
941
942static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
943{
944 struct bd_holder *bo;
945
946 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
947 if (!bo)
948 return NULL;
949
950 bo->count = 1;
951 bo->sdir = kobj;
952
953 return bo;
954}
955
956static void free_bd_holder(struct bd_holder *bo)
957{
958 kfree(bo);
959}
960
961/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500962 * find_bd_holder - find matching struct bd_holder from the block device
963 *
964 * @bdev: struct block device to be searched
965 * @bo: target struct bd_holder
966 *
967 * Returns matching entry with @bo in @bdev->bd_holder_list.
968 * If found, increment the reference count and return the pointer.
969 * If not found, returns NULL.
970 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800971static struct bd_holder *find_bd_holder(struct block_device *bdev,
972 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500973{
974 struct bd_holder *tmp;
975
976 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
977 if (tmp->sdir == bo->sdir) {
978 tmp->count++;
979 return tmp;
980 }
981
982 return NULL;
983}
984
985/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800986 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
987 *
988 * @bdev: block device to be bd_claimed
989 * @bo: preallocated and initialized by alloc_bd_holder()
990 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500991 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800992 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500993 * Returns 0 if symlinks are created.
994 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800995 */
996static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
997{
Johannes Weiner4e916722007-07-15 23:41:25 -0700998 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800999
1000 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001001 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001002
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001003 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001004 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001005
Johannes Weiner4e916722007-07-15 23:41:25 -07001006 err = add_symlink(bo->sdir, bo->sdev);
1007 if (err)
1008 return err;
1009
1010 err = add_symlink(bo->hdir, bo->hdev);
1011 if (err) {
1012 del_symlink(bo->sdir, bo->sdev);
1013 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001014 }
Johannes Weiner4e916722007-07-15 23:41:25 -07001015
1016 list_add_tail(&bo->list, &bdev->bd_holder_list);
1017 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001018}
1019
1020/**
1021 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1022 *
1023 * @bdev: block device to be bd_claimed
1024 * @kobj: holder's kobject
1025 *
1026 * If there is matching entry with @kobj in @bdev->bd_holder_list
1027 * and no other bd_claim() from the same kobject,
1028 * remove the struct bd_holder from the list, delete symlinks for it.
1029 *
1030 * Returns a pointer to the struct bd_holder when it's removed from the list
1031 * and ready to be freed.
1032 * Returns NULL if matching claim isn't found or there is other bd_claim()
1033 * by the same kobject.
1034 */
1035static struct bd_holder *del_bd_holder(struct block_device *bdev,
1036 struct kobject *kobj)
1037{
1038 struct bd_holder *bo;
1039
1040 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1041 if (bo->sdir == kobj) {
1042 bo->count--;
1043 BUG_ON(bo->count < 0);
1044 if (!bo->count) {
1045 list_del(&bo->list);
1046 del_symlink(bo->sdir, bo->sdev);
1047 del_symlink(bo->hdir, bo->hdev);
1048 bd_holder_release_dirs(bo);
1049 return bo;
1050 }
1051 break;
1052 }
1053 }
1054
1055 return NULL;
1056}
1057
1058/**
1059 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1060 *
1061 * @bdev: block device to be claimed
1062 * @holder: holder's signature
1063 * @kobj: holder's kobject
1064 *
1065 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1066 * the bdev and the holder's kobject.
1067 * Use bd_release_from_kobject() when relesing the claimed bdev.
1068 *
1069 * Returns 0 on success. (same as bd_claim())
1070 * Returns errno on failure.
1071 */
1072static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1073 struct kobject *kobj)
1074{
Johannes Weiner4e916722007-07-15 23:41:25 -07001075 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001076 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001077
1078 if (!kobj)
1079 return -EINVAL;
1080
1081 bo = alloc_bd_holder(kobj);
1082 if (!bo)
1083 return -ENOMEM;
1084
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001085 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001086
Johannes Weiner4e916722007-07-15 23:41:25 -07001087 err = bd_claim(bdev, holder);
1088 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -07001089 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001090
1091 found = find_bd_holder(bdev, bo);
1092 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -07001093 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001094
1095 err = add_bd_holder(bdev, bo);
1096 if (err)
1097 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001098 else
1099 bo = NULL;
1100fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001101 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -07001102 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -07001103 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001104}
1105
1106/**
1107 * bd_release_from_kobject - bd_release() with additional kobject signature
1108 *
1109 * @bdev: block device to be released
1110 * @kobj: holder's kobject
1111 *
1112 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1113 */
1114static void bd_release_from_kobject(struct block_device *bdev,
1115 struct kobject *kobj)
1116{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001117 if (!kobj)
1118 return;
1119
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001120 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001121 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001122 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001123 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001124}
1125
1126/**
1127 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1128 *
1129 * @bdev: block device to be claimed
1130 * @holder: holder's signature
1131 * @disk: holder's gendisk
1132 *
1133 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1134 */
1135int bd_claim_by_disk(struct block_device *bdev, void *holder,
1136 struct gendisk *disk)
1137{
1138 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1139}
1140EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1141
1142/**
1143 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1144 *
1145 * @bdev: block device to be claimed
1146 * @disk: holder's gendisk
1147 *
1148 * Call bd_release_from_kobject() and put @disk->slave_dir.
1149 */
1150void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1151{
1152 bd_release_from_kobject(bdev, disk->slave_dir);
1153 kobject_put(disk->slave_dir);
1154}
1155EXPORT_SYMBOL_GPL(bd_release_from_disk);
1156#endif
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158/*
1159 * Tries to open block device by device number. Use it ONLY if you
1160 * really do not have anything better - i.e. when you are behind a
1161 * truly sucky interface and all you are given is a device number. _Never_
1162 * to be used for internal purposes. If you ever need it - reconsider
1163 * your API.
1164 */
Al Viroaeb5d722008-09-02 15:28:45 -04001165struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166{
1167 struct block_device *bdev = bdget(dev);
1168 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001170 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 return err ? ERR_PTR(err) : bdev;
1172}
1173
1174EXPORT_SYMBOL(open_by_devnum);
1175
Andrew Patterson0c002c22008-09-04 14:27:20 -06001176/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001177 * flush_disk - invalidates all buffer-cache entries on a disk
1178 *
1179 * @bdev: struct block device to be flushed
1180 *
1181 * Invalidates all buffer-cache entries on a disk. It should be called
1182 * when a disk has been changed -- either by a media change or online
1183 * resize.
1184 */
1185static void flush_disk(struct block_device *bdev)
1186{
1187 if (__invalidate_device(bdev)) {
1188 char name[BDEVNAME_SIZE] = "";
1189
1190 if (bdev->bd_disk)
1191 disk_name(bdev->bd_disk, 0, name);
1192 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1193 "resized disk %s\n", name);
1194 }
1195
1196 if (!bdev->bd_disk)
1197 return;
1198 if (disk_partitionable(bdev->bd_disk))
1199 bdev->bd_invalidated = 1;
1200}
1201
1202/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001203 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001204 * @disk: struct gendisk to check
1205 * @bdev: struct bdev to adjust.
1206 *
1207 * This routine checks to see if the bdev size does not match the disk size
1208 * and adjusts it if it differs.
1209 */
1210void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1211{
1212 loff_t disk_size, bdev_size;
1213
1214 disk_size = (loff_t)get_capacity(disk) << 9;
1215 bdev_size = i_size_read(bdev->bd_inode);
1216 if (disk_size != bdev_size) {
1217 char name[BDEVNAME_SIZE];
1218
1219 disk_name(disk, 0, name);
1220 printk(KERN_INFO
1221 "%s: detected capacity change from %lld to %lld\n",
1222 name, bdev_size, disk_size);
1223 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001224 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001225 }
1226}
1227EXPORT_SYMBOL(check_disk_size_change);
1228
1229/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001230 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001231 * @disk: struct gendisk to be revalidated
1232 *
1233 * This routine is a wrapper for lower-level driver's revalidate_disk
1234 * call-backs. It is used to do common pre and post operations needed
1235 * for all revalidate_disk operations.
1236 */
1237int revalidate_disk(struct gendisk *disk)
1238{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001239 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001240 int ret = 0;
1241
1242 if (disk->fops->revalidate_disk)
1243 ret = disk->fops->revalidate_disk(disk);
1244
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001245 bdev = bdget_disk(disk, 0);
1246 if (!bdev)
1247 return ret;
1248
1249 mutex_lock(&bdev->bd_mutex);
1250 check_disk_size_change(disk, bdev);
1251 mutex_unlock(&bdev->bd_mutex);
1252 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001253 return ret;
1254}
1255EXPORT_SYMBOL(revalidate_disk);
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257/*
1258 * This routine checks whether a removable media has been changed,
1259 * and invalidates all buffer-cache-entries in that case. This
1260 * is a relatively slow routine, so we have to try to minimize using
1261 * it. Thus it is called only upon a 'mount' or 'open'. This
1262 * is the best way of combining speed and utility, I think.
1263 * People changing diskettes in the middle of an operation deserve
1264 * to lose :-)
1265 */
1266int check_disk_change(struct block_device *bdev)
1267{
1268 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001269 const struct block_device_operations *bdops = disk->fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271 if (!bdops->media_changed)
1272 return 0;
1273 if (!bdops->media_changed(bdev->bd_disk))
1274 return 0;
1275
Andrew Patterson56ade442008-09-04 14:27:40 -06001276 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (bdops->revalidate_disk)
1278 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 return 1;
1280}
1281
1282EXPORT_SYMBOL(check_disk_change);
1283
1284void bd_set_size(struct block_device *bdev, loff_t size)
1285{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001286 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288 bdev->bd_inode->i_size = size;
1289 while (bsize < PAGE_CACHE_SIZE) {
1290 if (size & bsize)
1291 break;
1292 bsize <<= 1;
1293 }
1294 bdev->bd_block_size = bsize;
1295 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1296}
1297EXPORT_SYMBOL(bd_set_size);
1298
Al Viro9a1c3542008-02-22 20:40:24 -05001299static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001300
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001301/*
1302 * bd_mutex locking:
1303 *
1304 * mutex_lock(part->bd_mutex)
1305 * mutex_lock_nested(whole->bd_mutex, 1)
1306 */
1307
Al Viro572c4892007-10-08 13:24:05 -04001308static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001311 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001312 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001313 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Al Viro572c4892007-10-08 13:24:05 -04001315 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001316 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001317 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001318 perm |= MAY_WRITE;
1319 /*
1320 * hooks: /n/, see "layering violations".
1321 */
1322 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001323 if (ret != 0) {
1324 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001325 return ret;
Al Viro82666022008-08-01 05:32:04 -04001326 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001327
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 lock_kernel();
NeilBrownd3374822009-01-09 08:31:10 +11001329 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001330
Tejun Heo89f97492008-11-05 10:21:06 +01001331 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001332 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001333 if (!disk)
1334 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335
NeilBrown6796bf52006-12-08 02:36:16 -08001336 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 if (!bdev->bd_openers) {
1338 bdev->bd_disk = disk;
1339 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001340 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001342
1343 ret = -ENXIO;
1344 bdev->bd_part = disk_get_part(disk, partno);
1345 if (!bdev->bd_part)
1346 goto out_clear;
1347
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001349 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001350 if (ret == -ERESTARTSYS) {
1351 /* Lost a race with 'disk' being
1352 * deleted, try again.
1353 * See md.c
1354 */
1355 disk_put_part(bdev->bd_part);
1356 bdev->bd_part = NULL;
1357 module_put(disk->fops->owner);
1358 put_disk(disk);
1359 bdev->bd_disk = NULL;
1360 mutex_unlock(&bdev->bd_mutex);
1361 goto restart;
1362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001364 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 }
1366 if (!bdev->bd_openers) {
1367 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1368 bdi = blk_get_backing_dev_info(bdev);
1369 if (bdi == NULL)
1370 bdi = &default_backing_dev_info;
1371 bdev->bd_inode->i_data.backing_dev_info = bdi;
1372 }
1373 if (bdev->bd_invalidated)
1374 rescan_partitions(disk, bdev);
1375 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 struct block_device *whole;
1377 whole = bdget_disk(disk, 0);
1378 ret = -ENOMEM;
1379 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001380 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001381 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001382 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001384 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 bdev->bd_inode->i_data.backing_dev_info =
1387 whole->bd_inode->i_data.backing_dev_info;
Tejun Heo89f97492008-11-05 10:21:06 +01001388 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001389 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001390 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001392 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 }
Tejun Heo89f97492008-11-05 10:21:06 +01001394 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
1396 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001397 module_put(disk->fops->owner);
Neil Brown960cc0f2009-10-26 08:59:17 +01001398 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001399 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (bdev->bd_contains == bdev) {
1401 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001402 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001404 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 }
1406 if (bdev->bd_invalidated)
1407 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 }
1409 }
1410 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001411 if (for_part)
1412 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001413 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 unlock_kernel();
1415 return 0;
1416
Tejun Heo0762b8b2008-08-25 19:56:12 +09001417 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001418 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001420 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1422 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001423 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001425 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001426 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001427 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001429
Tejun Heo0762b8b2008-08-25 19:56:12 +09001430 if (disk)
1431 module_put(disk->fops->owner);
1432 put_disk(disk);
1433 bdput(bdev);
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return ret;
1436}
1437
Al Viro572c4892007-10-08 13:24:05 -04001438int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439{
Al Viro572c4892007-10-08 13:24:05 -04001440 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001441}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442EXPORT_SYMBOL(blkdev_get);
1443
1444static int blkdev_open(struct inode * inode, struct file * filp)
1445{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001446 struct block_device *whole = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 struct block_device *bdev;
1448 int res;
1449
1450 /*
1451 * Preserve backwards compatibility and allow large file access
1452 * even if userspace doesn't ask for it explicitly. Some mkfs
1453 * binary needs it. We might want to drop this workaround
1454 * during an unstable branch.
1455 */
1456 filp->f_flags |= O_LARGEFILE;
1457
Al Viro572c4892007-10-08 13:24:05 -04001458 if (filp->f_flags & O_NDELAY)
1459 filp->f_mode |= FMODE_NDELAY;
1460 if (filp->f_flags & O_EXCL)
1461 filp->f_mode |= FMODE_EXCL;
1462 if ((filp->f_flags & O_ACCMODE) == 3)
1463 filp->f_mode |= FMODE_WRITE_IOCTL;
1464
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001466 if (bdev == NULL)
1467 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
Tejun Heo6b4517a2010-04-07 18:53:59 +09001469 if (filp->f_mode & FMODE_EXCL) {
1470 whole = bd_start_claiming(bdev, filp);
1471 if (IS_ERR(whole)) {
1472 bdput(bdev);
1473 return PTR_ERR(whole);
1474 }
1475 }
1476
Al Viro572c4892007-10-08 13:24:05 -04001477 filp->f_mapping = bdev->bd_inode->i_mapping;
1478
1479 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480
Tejun Heo6b4517a2010-04-07 18:53:59 +09001481 if (whole) {
1482 if (res == 0)
1483 BUG_ON(bd_claim(bdev, filp) != 0);
1484 else
1485 bd_abort_claiming(whole, filp);
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001486 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return res;
1489}
1490
Al Viro9a1c3542008-02-22 20:40:24 -05001491static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001492{
1493 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001494 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001495 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001496
NeilBrown6796bf52006-12-08 02:36:16 -08001497 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001498 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001499 if (for_part)
1500 bdev->bd_part_count--;
1501
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001502 if (!--bdev->bd_openers) {
1503 sync_blockdev(bdev);
1504 kill_bdev(bdev);
1505 }
1506 if (bdev->bd_contains == bdev) {
1507 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001508 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001509 }
1510 if (!bdev->bd_openers) {
1511 struct module *owner = disk->fops->owner;
1512
1513 put_disk(disk);
1514 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001515 disk_put_part(bdev->bd_part);
1516 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001517 bdev->bd_disk = NULL;
1518 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001519 if (bdev != bdev->bd_contains)
1520 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001521 bdev->bd_contains = NULL;
1522 }
1523 unlock_kernel();
1524 mutex_unlock(&bdev->bd_mutex);
1525 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001526 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001527 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001528 return ret;
1529}
1530
Al Viro9a1c3542008-02-22 20:40:24 -05001531int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001532{
Al Viro9a1c3542008-02-22 20:40:24 -05001533 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001534}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535EXPORT_SYMBOL(blkdev_put);
1536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537static int blkdev_close(struct inode * inode, struct file * filp)
1538{
1539 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1540 if (bdev->bd_holder == filp)
1541 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001542 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543}
1544
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001545static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546{
Al Viro56b26ad2008-09-19 03:17:36 -04001547 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1548 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001549
1550 /*
1551 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1552 * to updated it before every ioctl.
1553 */
Al Viro56b26ad2008-09-19 03:17:36 -04001554 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001555 mode |= FMODE_NDELAY;
1556 else
1557 mode &= ~FMODE_NDELAY;
1558
Al Viro56b26ad2008-09-19 03:17:36 -04001559 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560}
1561
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001562/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001563 * Write data to the block device. Only intended for the block device itself
1564 * and the raw driver which basically is a fake block device.
1565 *
1566 * Does not take i_mutex for the write and thus is not for general purpose
1567 * use.
1568 */
1569ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1570 unsigned long nr_segs, loff_t pos)
1571{
1572 struct file *file = iocb->ki_filp;
1573 ssize_t ret;
1574
1575 BUG_ON(iocb->ki_pos != pos);
1576
1577 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1578 if (ret > 0 || ret == -EIOCBQUEUED) {
1579 ssize_t err;
1580
1581 err = generic_write_sync(file, pos, ret);
1582 if (err < 0 && ret > 0)
1583 ret = err;
1584 }
1585 return ret;
1586}
1587EXPORT_SYMBOL_GPL(blkdev_aio_write);
1588
1589/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001590 * Try to release a page associated with block device when the system
1591 * is under memory pressure.
1592 */
1593static int blkdev_releasepage(struct page *page, gfp_t wait)
1594{
1595 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1596
1597 if (super && super->s_op->bdev_try_to_free_page)
1598 return super->s_op->bdev_try_to_free_page(super, page, wait);
1599
1600 return try_to_free_buffers(page);
1601}
1602
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001603static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 .readpage = blkdev_readpage,
1605 .writepage = blkdev_writepage,
1606 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001607 .write_begin = blkdev_write_begin,
1608 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001610 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 .direct_IO = blkdev_direct_IO,
1612};
1613
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001614const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 .open = blkdev_open,
1616 .release = blkdev_close,
1617 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001618 .read = do_sync_read,
1619 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001621 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001623 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001624 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625#ifdef CONFIG_COMPAT
1626 .compat_ioctl = compat_blkdev_ioctl,
1627#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001628 .splice_read = generic_file_splice_read,
1629 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630};
1631
1632int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1633{
1634 int res;
1635 mm_segment_t old_fs = get_fs();
1636 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001637 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 set_fs(old_fs);
1639 return res;
1640}
1641
1642EXPORT_SYMBOL(ioctl_by_bdev);
1643
1644/**
1645 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001646 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001648 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 * namespace if possible and return it. Return ERR_PTR(error)
1650 * otherwise.
1651 */
Al Viro421748e2008-08-02 01:04:36 -04001652struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653{
1654 struct block_device *bdev;
1655 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001656 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 int error;
1658
Al Viro421748e2008-08-02 01:04:36 -04001659 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 return ERR_PTR(-EINVAL);
1661
Al Viro421748e2008-08-02 01:04:36 -04001662 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (error)
1664 return ERR_PTR(error);
1665
Al Viro421748e2008-08-02 01:04:36 -04001666 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 error = -ENOTBLK;
1668 if (!S_ISBLK(inode->i_mode))
1669 goto fail;
1670 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001671 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 goto fail;
1673 error = -ENOMEM;
1674 bdev = bd_acquire(inode);
1675 if (!bdev)
1676 goto fail;
1677out:
Al Viro421748e2008-08-02 01:04:36 -04001678 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 return bdev;
1680fail:
1681 bdev = ERR_PTR(error);
1682 goto out;
1683}
Al Virod5686b42008-08-01 05:00:11 -04001684EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
1686/**
Al Viro30c40d22008-02-22 19:50:45 -05001687 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 *
1689 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001690 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 * @holder: owner for exclusion
1692 *
1693 * Open the blockdevice described by the special file at @path, claim it
1694 * for the @holder.
1695 */
Al Viro30c40d22008-02-22 19:50:45 -05001696struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001698 struct block_device *bdev, *whole;
1699 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701 bdev = lookup_bdev(path);
1702 if (IS_ERR(bdev))
1703 return bdev;
1704
Tejun Heo6b4517a2010-04-07 18:53:59 +09001705 whole = bd_start_claiming(bdev, holder);
1706 if (IS_ERR(whole)) {
1707 bdput(bdev);
1708 return whole;
1709 }
1710
Al Viro572c4892007-10-08 13:24:05 -04001711 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (error)
Tejun Heo6b4517a2010-04-07 18:53:59 +09001713 goto out_abort_claiming;
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001716 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Tejun Heo6b4517a2010-04-07 18:53:59 +09001717 goto out_blkdev_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Tejun Heo6b4517a2010-04-07 18:53:59 +09001719 BUG_ON(bd_claim(bdev, holder) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 return bdev;
Tejun Heo6b4517a2010-04-07 18:53:59 +09001721
1722out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001723 blkdev_put(bdev, mode);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001724out_abort_claiming:
1725 bd_abort_claiming(whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 return ERR_PTR(error);
1727}
1728
Al Viro30c40d22008-02-22 19:50:45 -05001729EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
1731/**
Al Viro30c40d22008-02-22 19:50:45 -05001732 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 *
1734 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001735 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 *
Al Viro30c40d22008-02-22 19:50:45 -05001737 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 */
Al Viro30c40d22008-02-22 19:50:45 -05001739void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
1741 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001742 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
Al Viro30c40d22008-02-22 19:50:45 -05001745EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001746
1747int __invalidate_device(struct block_device *bdev)
1748{
1749 struct super_block *sb = get_super(bdev);
1750 int res = 0;
1751
1752 if (sb) {
1753 /*
1754 * no need to lock the super, get_super holds the
1755 * read mutex so the filesystem cannot go away
1756 * under us (->put_super runs with the write lock
1757 * hold).
1758 */
1759 shrink_dcache_sb(sb);
1760 res = invalidate_inodes(sb);
1761 drop_super(sb);
1762 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001763 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001764 return res;
1765}
1766EXPORT_SYMBOL(__invalidate_device);