blob: 7d822fae7765b2bfdf0978e8ecf1296875073e85 [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>
15#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/blkpg.h>
19#include <linux/buffer_head.h>
David Howells811d7362006-08-29 19:06:09 +010020#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mpage.h>
22#include <linux/mount.h>
23#include <linux/uio.h>
24#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070025#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29struct bdev_inode {
30 struct block_device bdev;
31 struct inode vfs_inode;
32};
33
Adrian Bunk4c54ac62008-02-18 13:48:31 +010034static const struct address_space_operations def_blk_aops;
35
Linus Torvalds1da177e2005-04-16 15:20:36 -070036static inline struct bdev_inode *BDEV_I(struct inode *inode)
37{
38 return container_of(inode, struct bdev_inode, vfs_inode);
39}
40
41inline struct block_device *I_BDEV(struct inode *inode)
42{
43 return &BDEV_I(inode)->bdev;
44}
45
46EXPORT_SYMBOL(I_BDEV);
47
48static sector_t max_block(struct block_device *bdev)
49{
50 sector_t retval = ~((sector_t)0);
51 loff_t sz = i_size_read(bdev->bd_inode);
52
53 if (sz) {
54 unsigned int size = block_size(bdev);
55 unsigned int sizebits = blksize_bits(size);
56 retval = (sz >> sizebits);
57 }
58 return retval;
59}
60
Peter Zijlstraf9a14392007-05-06 14:49:55 -070061/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070062static void kill_bdev(struct block_device *bdev)
63{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070064 if (bdev->bd_inode->i_mapping->nrpages == 0)
65 return;
66 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
68}
69
70int set_blocksize(struct block_device *bdev, int size)
71{
72 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070073 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 return -EINVAL;
75
76 /* Size cannot be smaller than the size supported by the device */
77 if (size < bdev_hardsect_size(bdev))
78 return -EINVAL;
79
80 /* Don't change the size if it is same as current */
81 if (bdev->bd_block_size != size) {
82 sync_blockdev(bdev);
83 bdev->bd_block_size = size;
84 bdev->bd_inode->i_blkbits = blksize_bits(size);
85 kill_bdev(bdev);
86 }
87 return 0;
88}
89
90EXPORT_SYMBOL(set_blocksize);
91
92int sb_set_blocksize(struct super_block *sb, int size)
93{
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 if (set_blocksize(sb->s_bdev, size))
95 return 0;
96 /* If we get here, we know size is power of two
97 * and it's value is between 512 and PAGE_SIZE */
98 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -080099 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 return sb->s_blocksize;
101}
102
103EXPORT_SYMBOL(sb_set_blocksize);
104
105int sb_min_blocksize(struct super_block *sb, int size)
106{
107 int minsize = bdev_hardsect_size(sb->s_bdev);
108 if (size < minsize)
109 size = minsize;
110 return sb_set_blocksize(sb, size);
111}
112
113EXPORT_SYMBOL(sb_min_blocksize);
114
115static int
116blkdev_get_block(struct inode *inode, sector_t iblock,
117 struct buffer_head *bh, int create)
118{
119 if (iblock >= max_block(I_BDEV(inode))) {
120 if (create)
121 return -EIO;
122
123 /*
124 * for reads, we're just trying to fill a partial page.
125 * return a hole, they will have to call get_block again
126 * before they can fill it, and they will get -EIO at that
127 * time
128 */
129 return 0;
130 }
131 bh->b_bdev = I_BDEV(inode);
132 bh->b_blocknr = iblock;
133 set_buffer_mapped(bh);
134 return 0;
135}
136
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800137static int
138blkdev_get_blocks(struct inode *inode, sector_t iblock,
139 struct buffer_head *bh, int create)
140{
141 sector_t end_block = max_block(I_BDEV(inode));
142 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
143
144 if ((iblock + max_blocks) > end_block) {
145 max_blocks = end_block - iblock;
146 if ((long)max_blocks <= 0) {
147 if (create)
148 return -EIO; /* write fully beyond EOF */
149 /*
150 * It is a read which is fully beyond EOF. We return
151 * a !buffer_mapped buffer
152 */
153 max_blocks = 0;
154 }
155 }
156
157 bh->b_bdev = I_BDEV(inode);
158 bh->b_blocknr = iblock;
159 bh->b_size = max_blocks << inode->i_blkbits;
160 if (max_blocks)
161 set_buffer_mapped(bh);
162 return 0;
163}
164
165static ssize_t
166blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
167 loff_t offset, unsigned long nr_segs)
168{
169 struct file *file = iocb->ki_filp;
170 struct inode *inode = file->f_mapping->host;
171
172 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
173 iov, offset, nr_segs, blkdev_get_blocks, NULL);
174}
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
177{
178 return block_write_full_page(page, blkdev_get_block, wbc);
179}
180
181static int blkdev_readpage(struct file * file, struct page * page)
182{
183 return block_read_full_page(page, blkdev_get_block);
184}
185
Nick Piggin6272b5a2007-10-16 01:25:04 -0700186static int blkdev_write_begin(struct file *file, struct address_space *mapping,
187 loff_t pos, unsigned len, unsigned flags,
188 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700190 *pagep = NULL;
191 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
192 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
Nick Piggin6272b5a2007-10-16 01:25:04 -0700195static int blkdev_write_end(struct file *file, struct address_space *mapping,
196 loff_t pos, unsigned len, unsigned copied,
197 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700199 int ret;
200 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
201
202 unlock_page(page);
203 page_cache_release(page);
204
205 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
208/*
209 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800210 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 * so we compute the size by hand (just as in block_read/write above)
212 */
213static loff_t block_llseek(struct file *file, loff_t offset, int origin)
214{
215 struct inode *bd_inode = file->f_mapping->host;
216 loff_t size;
217 loff_t retval;
218
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800219 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 size = i_size_read(bd_inode);
221
222 switch (origin) {
223 case 2:
224 offset += size;
225 break;
226 case 1:
227 offset += file->f_pos;
228 }
229 retval = -EINVAL;
230 if (offset >= 0 && offset <= size) {
231 if (offset != file->f_pos) {
232 file->f_pos = offset;
233 }
234 retval = offset;
235 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800236 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 return retval;
238}
239
240/*
241 * Filp is never NULL; the only case when ->fsync() is called with
242 * NULL first argument is nfsd_sync_dir() and that's not a directory.
243 */
244
245static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
246{
247 return sync_blockdev(I_BDEV(filp->f_mapping->host));
248}
249
250/*
251 * pseudo-fs
252 */
253
254static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800255static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257static struct inode *bdev_alloc_inode(struct super_block *sb)
258{
Christoph Lametere94b1762006-12-06 20:33:17 -0800259 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 if (!ei)
261 return NULL;
262 return &ei->vfs_inode;
263}
264
265static void bdev_destroy_inode(struct inode *inode)
266{
267 struct bdev_inode *bdi = BDEV_I(inode);
268
269 bdi->bdev.bd_inode_backing_dev_info = NULL;
270 kmem_cache_free(bdev_cachep, bdi);
271}
272
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700273static void init_once(struct kmem_cache * cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
275 struct bdev_inode *ei = (struct bdev_inode *) foo;
276 struct block_device *bdev = &ei->bdev;
277
Christoph Lametera35afb82007-05-16 22:10:57 -0700278 memset(bdev, 0, sizeof(*bdev));
279 mutex_init(&bdev->bd_mutex);
280 sema_init(&bdev->bd_mount_sem, 1);
281 INIT_LIST_HEAD(&bdev->bd_inodes);
282 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800283#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700284 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800285#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700286 inode_init_once(&ei->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287}
288
289static inline void __bd_forget(struct inode *inode)
290{
291 list_del_init(&inode->i_devices);
292 inode->i_bdev = NULL;
293 inode->i_mapping = &inode->i_data;
294}
295
296static void bdev_clear_inode(struct inode *inode)
297{
298 struct block_device *bdev = &BDEV_I(inode)->bdev;
299 struct list_head *p;
300 spin_lock(&bdev_lock);
301 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
302 __bd_forget(list_entry(p, struct inode, i_devices));
303 }
304 list_del_init(&bdev->bd_list);
305 spin_unlock(&bdev_lock);
306}
307
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800308static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 .statfs = simple_statfs,
310 .alloc_inode = bdev_alloc_inode,
311 .destroy_inode = bdev_destroy_inode,
312 .drop_inode = generic_delete_inode,
313 .clear_inode = bdev_clear_inode,
314};
315
David Howells454e2392006-06-23 02:02:57 -0700316static int bd_get_sb(struct file_system_type *fs_type,
317 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
David Howells454e2392006-06-23 02:02:57 -0700319 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320}
321
322static struct file_system_type bd_type = {
323 .name = "bdev",
324 .get_sb = bd_get_sb,
325 .kill_sb = kill_anon_super,
326};
327
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800328static struct vfsmount *bd_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329struct super_block *blockdev_superblock;
330
331void __init bdev_cache_init(void)
332{
333 int err;
334 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800335 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
336 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900337 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 err = register_filesystem(&bd_type);
339 if (err)
340 panic("Cannot register bdev pseudo-fs");
341 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (IS_ERR(bd_mnt))
343 panic("Cannot create bdev pseudo-fs");
344 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
345}
346
347/*
348 * Most likely _very_ bad one - but then it's hardly critical for small
349 * /dev and can be fixed when somebody will need really large one.
350 * Keep in mind that it will be fed through icache hash function too.
351 */
352static inline unsigned long hash(dev_t dev)
353{
354 return MAJOR(dev)+MINOR(dev);
355}
356
357static int bdev_test(struct inode *inode, void *data)
358{
359 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
360}
361
362static int bdev_set(struct inode *inode, void *data)
363{
364 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
365 return 0;
366}
367
368static LIST_HEAD(all_bdevs);
369
370struct block_device *bdget(dev_t dev)
371{
372 struct block_device *bdev;
373 struct inode *inode;
374
375 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
376 bdev_test, bdev_set, &dev);
377
378 if (!inode)
379 return NULL;
380
381 bdev = &BDEV_I(inode)->bdev;
382
383 if (inode->i_state & I_NEW) {
384 bdev->bd_contains = NULL;
385 bdev->bd_inode = inode;
386 bdev->bd_block_size = (1 << inode->i_blkbits);
387 bdev->bd_part_count = 0;
388 bdev->bd_invalidated = 0;
389 inode->i_mode = S_IFBLK;
390 inode->i_rdev = dev;
391 inode->i_bdev = bdev;
392 inode->i_data.a_ops = &def_blk_aops;
393 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
394 inode->i_data.backing_dev_info = &default_backing_dev_info;
395 spin_lock(&bdev_lock);
396 list_add(&bdev->bd_list, &all_bdevs);
397 spin_unlock(&bdev_lock);
398 unlock_new_inode(inode);
399 }
400 return bdev;
401}
402
403EXPORT_SYMBOL(bdget);
404
405long nr_blockdev_pages(void)
406{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700407 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 long ret = 0;
409 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700410 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 ret += bdev->bd_inode->i_mapping->nrpages;
412 }
413 spin_unlock(&bdev_lock);
414 return ret;
415}
416
417void bdput(struct block_device *bdev)
418{
419 iput(bdev->bd_inode);
420}
421
422EXPORT_SYMBOL(bdput);
423
424static struct block_device *bd_acquire(struct inode *inode)
425{
426 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 spin_lock(&bdev_lock);
429 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700430 if (bdev) {
431 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 spin_unlock(&bdev_lock);
433 return bdev;
434 }
435 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 bdev = bdget(inode->i_rdev);
438 if (bdev) {
439 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700440 if (!inode->i_bdev) {
441 /*
442 * We take an additional bd_inode->i_count for inode,
443 * and it's released in clear_inode() of inode.
444 * So, we can access it via ->i_mapping always
445 * without igrab().
446 */
447 atomic_inc(&bdev->bd_inode->i_count);
448 inode->i_bdev = bdev;
449 inode->i_mapping = bdev->bd_inode->i_mapping;
450 list_add(&inode->i_devices, &bdev->bd_inodes);
451 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 spin_unlock(&bdev_lock);
453 }
454 return bdev;
455}
456
457/* Call when you free inode */
458
459void bd_forget(struct inode *inode)
460{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700461 struct block_device *bdev = NULL;
462
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700464 if (inode->i_bdev) {
465 if (inode->i_sb != blockdev_superblock)
466 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700470
471 if (bdev)
472 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473}
474
475int bd_claim(struct block_device *bdev, void *holder)
476{
477 int res;
478 spin_lock(&bdev_lock);
479
480 /* first decide result */
481 if (bdev->bd_holder == holder)
482 res = 0; /* already a holder */
483 else if (bdev->bd_holder != NULL)
484 res = -EBUSY; /* held by someone else */
485 else if (bdev->bd_contains == bdev)
486 res = 0; /* is a whole device which isn't held */
487
488 else if (bdev->bd_contains->bd_holder == bd_claim)
489 res = 0; /* is a partition of a device that is being partitioned */
490 else if (bdev->bd_contains->bd_holder != NULL)
491 res = -EBUSY; /* is a partition of a held device */
492 else
493 res = 0; /* is a partition of an un-held device */
494
495 /* now impose change */
496 if (res==0) {
497 /* note that for a whole device bd_holders
498 * will be incremented twice, and bd_holder will
499 * be set to bd_claim before being set to holder
500 */
501 bdev->bd_contains->bd_holders ++;
502 bdev->bd_contains->bd_holder = bd_claim;
503 bdev->bd_holders++;
504 bdev->bd_holder = holder;
505 }
506 spin_unlock(&bdev_lock);
507 return res;
508}
509
510EXPORT_SYMBOL(bd_claim);
511
512void bd_release(struct block_device *bdev)
513{
514 spin_lock(&bdev_lock);
515 if (!--bdev->bd_contains->bd_holders)
516 bdev->bd_contains->bd_holder = NULL;
517 if (!--bdev->bd_holders)
518 bdev->bd_holder = NULL;
519 spin_unlock(&bdev_lock);
520}
521
522EXPORT_SYMBOL(bd_release);
523
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800524#ifdef CONFIG_SYSFS
525/*
526 * Functions for bd_claim_by_kobject / bd_release_from_kobject
527 *
528 * If a kobject is passed to bd_claim_by_kobject()
529 * and the kobject has a parent directory,
530 * following symlinks are created:
531 * o from the kobject to the claimed bdev
532 * o from "holders" directory of the bdev to the parent of the kobject
533 * bd_release_from_kobject() removes these symlinks.
534 *
535 * Example:
536 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
537 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
538 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
539 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
540 */
541
542static struct kobject *bdev_get_kobj(struct block_device *bdev)
543{
544 if (bdev->bd_contains != bdev)
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200545 return kobject_get(&bdev->bd_part->dev.kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800546 else
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200547 return kobject_get(&bdev->bd_disk->dev.kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800548}
549
550static struct kobject *bdev_get_holder(struct block_device *bdev)
551{
552 if (bdev->bd_contains != bdev)
553 return kobject_get(bdev->bd_part->holder_dir);
554 else
555 return kobject_get(bdev->bd_disk->holder_dir);
556}
557
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700558static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800559{
560 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700561 return 0;
562 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800563}
564
565static void del_symlink(struct kobject *from, struct kobject *to)
566{
567 if (!from || !to)
568 return;
569 sysfs_remove_link(from, kobject_name(to));
570}
571
572/*
573 * 'struct bd_holder' contains pointers to kobjects symlinked by
574 * bd_claim_by_kobject.
575 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
576 */
577struct bd_holder {
578 struct list_head list; /* chain of holders of the bdev */
579 int count; /* references from the holder */
580 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
581 struct kobject *hdev; /* e.g. "/block/dm-0" */
582 struct kobject *hdir; /* e.g. "/block/sda/holders" */
583 struct kobject *sdev; /* e.g. "/block/sda" */
584};
585
586/*
587 * Get references of related kobjects at once.
588 * Returns 1 on success. 0 on failure.
589 *
590 * Should call bd_holder_release_dirs() after successful use.
591 */
592static int bd_holder_grab_dirs(struct block_device *bdev,
593 struct bd_holder *bo)
594{
595 if (!bdev || !bo)
596 return 0;
597
598 bo->sdir = kobject_get(bo->sdir);
599 if (!bo->sdir)
600 return 0;
601
602 bo->hdev = kobject_get(bo->sdir->parent);
603 if (!bo->hdev)
604 goto fail_put_sdir;
605
606 bo->sdev = bdev_get_kobj(bdev);
607 if (!bo->sdev)
608 goto fail_put_hdev;
609
610 bo->hdir = bdev_get_holder(bdev);
611 if (!bo->hdir)
612 goto fail_put_sdev;
613
614 return 1;
615
616fail_put_sdev:
617 kobject_put(bo->sdev);
618fail_put_hdev:
619 kobject_put(bo->hdev);
620fail_put_sdir:
621 kobject_put(bo->sdir);
622
623 return 0;
624}
625
626/* Put references of related kobjects at once. */
627static void bd_holder_release_dirs(struct bd_holder *bo)
628{
629 kobject_put(bo->hdir);
630 kobject_put(bo->sdev);
631 kobject_put(bo->hdev);
632 kobject_put(bo->sdir);
633}
634
635static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
636{
637 struct bd_holder *bo;
638
639 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
640 if (!bo)
641 return NULL;
642
643 bo->count = 1;
644 bo->sdir = kobj;
645
646 return bo;
647}
648
649static void free_bd_holder(struct bd_holder *bo)
650{
651 kfree(bo);
652}
653
654/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500655 * find_bd_holder - find matching struct bd_holder from the block device
656 *
657 * @bdev: struct block device to be searched
658 * @bo: target struct bd_holder
659 *
660 * Returns matching entry with @bo in @bdev->bd_holder_list.
661 * If found, increment the reference count and return the pointer.
662 * If not found, returns NULL.
663 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800664static struct bd_holder *find_bd_holder(struct block_device *bdev,
665 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500666{
667 struct bd_holder *tmp;
668
669 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
670 if (tmp->sdir == bo->sdir) {
671 tmp->count++;
672 return tmp;
673 }
674
675 return NULL;
676}
677
678/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800679 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
680 *
681 * @bdev: block device to be bd_claimed
682 * @bo: preallocated and initialized by alloc_bd_holder()
683 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500684 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800685 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500686 * Returns 0 if symlinks are created.
687 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800688 */
689static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
690{
Johannes Weiner4e916722007-07-15 23:41:25 -0700691 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800692
693 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700694 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800695
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800696 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700697 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800698
Johannes Weiner4e916722007-07-15 23:41:25 -0700699 err = add_symlink(bo->sdir, bo->sdev);
700 if (err)
701 return err;
702
703 err = add_symlink(bo->hdir, bo->hdev);
704 if (err) {
705 del_symlink(bo->sdir, bo->sdev);
706 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700707 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700708
709 list_add_tail(&bo->list, &bdev->bd_holder_list);
710 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800711}
712
713/**
714 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
715 *
716 * @bdev: block device to be bd_claimed
717 * @kobj: holder's kobject
718 *
719 * If there is matching entry with @kobj in @bdev->bd_holder_list
720 * and no other bd_claim() from the same kobject,
721 * remove the struct bd_holder from the list, delete symlinks for it.
722 *
723 * Returns a pointer to the struct bd_holder when it's removed from the list
724 * and ready to be freed.
725 * Returns NULL if matching claim isn't found or there is other bd_claim()
726 * by the same kobject.
727 */
728static struct bd_holder *del_bd_holder(struct block_device *bdev,
729 struct kobject *kobj)
730{
731 struct bd_holder *bo;
732
733 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
734 if (bo->sdir == kobj) {
735 bo->count--;
736 BUG_ON(bo->count < 0);
737 if (!bo->count) {
738 list_del(&bo->list);
739 del_symlink(bo->sdir, bo->sdev);
740 del_symlink(bo->hdir, bo->hdev);
741 bd_holder_release_dirs(bo);
742 return bo;
743 }
744 break;
745 }
746 }
747
748 return NULL;
749}
750
751/**
752 * bd_claim_by_kobject - bd_claim() with additional kobject signature
753 *
754 * @bdev: block device to be claimed
755 * @holder: holder's signature
756 * @kobj: holder's kobject
757 *
758 * Do bd_claim() and if it succeeds, create sysfs symlinks between
759 * the bdev and the holder's kobject.
760 * Use bd_release_from_kobject() when relesing the claimed bdev.
761 *
762 * Returns 0 on success. (same as bd_claim())
763 * Returns errno on failure.
764 */
765static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
766 struct kobject *kobj)
767{
Johannes Weiner4e916722007-07-15 23:41:25 -0700768 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500769 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800770
771 if (!kobj)
772 return -EINVAL;
773
774 bo = alloc_bd_holder(kobj);
775 if (!bo)
776 return -ENOMEM;
777
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800778 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500779
Johannes Weiner4e916722007-07-15 23:41:25 -0700780 err = bd_claim(bdev, holder);
781 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700782 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700783
784 found = find_bd_holder(bdev, bo);
785 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700786 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700787
788 err = add_bd_holder(bdev, bo);
789 if (err)
790 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700791 else
792 bo = NULL;
793fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800794 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700795 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700796 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800797}
798
799/**
800 * bd_release_from_kobject - bd_release() with additional kobject signature
801 *
802 * @bdev: block device to be released
803 * @kobj: holder's kobject
804 *
805 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
806 */
807static void bd_release_from_kobject(struct block_device *bdev,
808 struct kobject *kobj)
809{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800810 if (!kobj)
811 return;
812
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800813 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800814 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700815 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800816 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800817}
818
819/**
820 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
821 *
822 * @bdev: block device to be claimed
823 * @holder: holder's signature
824 * @disk: holder's gendisk
825 *
826 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
827 */
828int bd_claim_by_disk(struct block_device *bdev, void *holder,
829 struct gendisk *disk)
830{
831 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
832}
833EXPORT_SYMBOL_GPL(bd_claim_by_disk);
834
835/**
836 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
837 *
838 * @bdev: block device to be claimed
839 * @disk: holder's gendisk
840 *
841 * Call bd_release_from_kobject() and put @disk->slave_dir.
842 */
843void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
844{
845 bd_release_from_kobject(bdev, disk->slave_dir);
846 kobject_put(disk->slave_dir);
847}
848EXPORT_SYMBOL_GPL(bd_release_from_disk);
849#endif
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851/*
852 * Tries to open block device by device number. Use it ONLY if you
853 * really do not have anything better - i.e. when you are behind a
854 * truly sucky interface and all you are given is a device number. _Never_
855 * to be used for internal purposes. If you ever need it - reconsider
856 * your API.
857 */
858struct block_device *open_by_devnum(dev_t dev, unsigned mode)
859{
860 struct block_device *bdev = bdget(dev);
861 int err = -ENOMEM;
862 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
863 if (bdev)
864 err = blkdev_get(bdev, mode, flags);
865 return err ? ERR_PTR(err) : bdev;
866}
867
868EXPORT_SYMBOL(open_by_devnum);
869
870/*
871 * This routine checks whether a removable media has been changed,
872 * and invalidates all buffer-cache-entries in that case. This
873 * is a relatively slow routine, so we have to try to minimize using
874 * it. Thus it is called only upon a 'mount' or 'open'. This
875 * is the best way of combining speed and utility, I think.
876 * People changing diskettes in the middle of an operation deserve
877 * to lose :-)
878 */
879int check_disk_change(struct block_device *bdev)
880{
881 struct gendisk *disk = bdev->bd_disk;
882 struct block_device_operations * bdops = disk->fops;
883
884 if (!bdops->media_changed)
885 return 0;
886 if (!bdops->media_changed(bdev->bd_disk))
887 return 0;
888
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700889 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 printk("VFS: busy inodes on changed media.\n");
891
892 if (bdops->revalidate_disk)
893 bdops->revalidate_disk(bdev->bd_disk);
894 if (bdev->bd_disk->minors > 1)
895 bdev->bd_invalidated = 1;
896 return 1;
897}
898
899EXPORT_SYMBOL(check_disk_change);
900
901void bd_set_size(struct block_device *bdev, loff_t size)
902{
903 unsigned bsize = bdev_hardsect_size(bdev);
904
905 bdev->bd_inode->i_size = size;
906 while (bsize < PAGE_CACHE_SIZE) {
907 if (size & bsize)
908 break;
909 bsize <<= 1;
910 }
911 bdev->bd_block_size = bsize;
912 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
913}
914EXPORT_SYMBOL(bd_set_size);
915
NeilBrown37be4122006-12-08 02:36:16 -0800916static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
917 int for_part);
Peter Zijlstrac48f70c2006-12-08 02:36:18 -0800918static int __blkdev_put(struct block_device *bdev, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -0800919
Peter Zijlstra6d740cd2007-02-20 13:58:18 -0800920/*
921 * bd_mutex locking:
922 *
923 * mutex_lock(part->bd_mutex)
924 * mutex_lock_nested(whole->bd_mutex, 1)
925 */
926
NeilBrown37be4122006-12-08 02:36:16 -0800927static int do_open(struct block_device *bdev, struct file *file, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
929 struct module *owner = NULL;
930 struct gendisk *disk;
931 int ret = -ENXIO;
932 int part;
933
934 file->f_mapping = bdev->bd_inode->i_mapping;
935 lock_kernel();
936 disk = get_gendisk(bdev->bd_dev, &part);
937 if (!disk) {
938 unlock_kernel();
939 bdput(bdev);
940 return ret;
941 }
942 owner = disk->fops->owner;
943
NeilBrown6796bf52006-12-08 02:36:16 -0800944 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 if (!bdev->bd_openers) {
946 bdev->bd_disk = disk;
947 bdev->bd_contains = bdev;
948 if (!part) {
949 struct backing_dev_info *bdi;
950 if (disk->fops->open) {
951 ret = disk->fops->open(bdev->bd_inode, file);
952 if (ret)
953 goto out_first;
954 }
955 if (!bdev->bd_openers) {
956 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
957 bdi = blk_get_backing_dev_info(bdev);
958 if (bdi == NULL)
959 bdi = &default_backing_dev_info;
960 bdev->bd_inode->i_data.backing_dev_info = bdi;
961 }
962 if (bdev->bd_invalidated)
963 rescan_partitions(disk, bdev);
964 } else {
965 struct hd_struct *p;
966 struct block_device *whole;
967 whole = bdget_disk(disk, 0);
968 ret = -ENOMEM;
969 if (!whole)
970 goto out_first;
NeilBrown37be4122006-12-08 02:36:16 -0800971 BUG_ON(for_part);
972 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 if (ret)
974 goto out_first;
975 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 p = disk->part[part - 1];
977 bdev->bd_inode->i_data.backing_dev_info =
978 whole->bd_inode->i_data.backing_dev_info;
979 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 ret = -ENXIO;
981 goto out_first;
982 }
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200983 kobject_get(&p->dev.kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 bdev->bd_part = p;
985 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 }
987 } else {
988 put_disk(disk);
989 module_put(owner);
990 if (bdev->bd_contains == bdev) {
991 if (bdev->bd_disk->fops->open) {
992 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
993 if (ret)
994 goto out;
995 }
996 if (bdev->bd_invalidated)
997 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 }
999 }
1000 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001001 if (for_part)
1002 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001003 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 unlock_kernel();
1005 return 0;
1006
1007out_first:
1008 bdev->bd_disk = NULL;
1009 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1010 if (bdev != bdev->bd_contains)
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001011 __blkdev_put(bdev->bd_contains, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 bdev->bd_contains = NULL;
1013 put_disk(disk);
1014 module_put(owner);
1015out:
Arjan van de Venc039e312006-03-23 03:00:28 -08001016 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 unlock_kernel();
1018 if (ret)
1019 bdput(bdev);
1020 return ret;
1021}
1022
NeilBrown37be4122006-12-08 02:36:16 -08001023static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1024 int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025{
1026 /*
1027 * This crockload is due to bad choice of ->open() type.
1028 * It will go away.
1029 * For now, block device ->open() routine must _not_
1030 * examine anything in 'inode' argument except ->i_rdev.
1031 */
1032 struct file fake_file = {};
1033 struct dentry fake_dentry = {};
1034 fake_file.f_mode = mode;
1035 fake_file.f_flags = flags;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001036 fake_file.f_path.dentry = &fake_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 fake_dentry.d_inode = bdev->bd_inode;
1038
NeilBrown37be4122006-12-08 02:36:16 -08001039 return do_open(bdev, &fake_file, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040}
1041
NeilBrown37be4122006-12-08 02:36:16 -08001042int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1043{
1044 return __blkdev_get(bdev, mode, flags, 0);
1045}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046EXPORT_SYMBOL(blkdev_get);
1047
1048static int blkdev_open(struct inode * inode, struct file * filp)
1049{
1050 struct block_device *bdev;
1051 int res;
1052
1053 /*
1054 * Preserve backwards compatibility and allow large file access
1055 * even if userspace doesn't ask for it explicitly. Some mkfs
1056 * binary needs it. We might want to drop this workaround
1057 * during an unstable branch.
1058 */
1059 filp->f_flags |= O_LARGEFILE;
1060
1061 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001062 if (bdev == NULL)
1063 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
NeilBrown37be4122006-12-08 02:36:16 -08001065 res = do_open(bdev, filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 if (res)
1067 return res;
1068
1069 if (!(filp->f_flags & O_EXCL) )
1070 return 0;
1071
1072 if (!(res = bd_claim(bdev, filp)))
1073 return 0;
1074
1075 blkdev_put(bdev);
1076 return res;
1077}
1078
NeilBrown37be4122006-12-08 02:36:16 -08001079static int __blkdev_put(struct block_device *bdev, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001080{
1081 int ret = 0;
1082 struct inode *bd_inode = bdev->bd_inode;
1083 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001084 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001085
NeilBrown6796bf52006-12-08 02:36:16 -08001086 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001087 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001088 if (for_part)
1089 bdev->bd_part_count--;
1090
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001091 if (!--bdev->bd_openers) {
1092 sync_blockdev(bdev);
1093 kill_bdev(bdev);
1094 }
1095 if (bdev->bd_contains == bdev) {
1096 if (disk->fops->release)
1097 ret = disk->fops->release(bd_inode, NULL);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001098 }
1099 if (!bdev->bd_openers) {
1100 struct module *owner = disk->fops->owner;
1101
1102 put_disk(disk);
1103 module_put(owner);
1104
1105 if (bdev->bd_contains != bdev) {
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001106 kobject_put(&bdev->bd_part->dev.kobj);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001107 bdev->bd_part = NULL;
1108 }
1109 bdev->bd_disk = NULL;
1110 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001111 if (bdev != bdev->bd_contains)
1112 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001113 bdev->bd_contains = NULL;
1114 }
1115 unlock_kernel();
1116 mutex_unlock(&bdev->bd_mutex);
1117 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001118 if (victim)
1119 __blkdev_put(victim, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001120 return ret;
1121}
1122
NeilBrown37be4122006-12-08 02:36:16 -08001123int blkdev_put(struct block_device *bdev)
1124{
1125 return __blkdev_put(bdev, 0);
1126}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001127EXPORT_SYMBOL(blkdev_put);
1128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129static int blkdev_close(struct inode * inode, struct file * filp)
1130{
1131 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1132 if (bdev->bd_holder == filp)
1133 bd_release(bdev);
1134 return blkdev_put(bdev);
1135}
1136
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001137static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138{
1139 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1140}
1141
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001142static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 .readpage = blkdev_readpage,
1144 .writepage = blkdev_writepage,
1145 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001146 .write_begin = blkdev_write_begin,
1147 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 .writepages = generic_writepages,
1149 .direct_IO = blkdev_direct_IO,
1150};
1151
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001152const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 .open = blkdev_open,
1154 .release = blkdev_close,
1155 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001156 .read = do_sync_read,
1157 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001159 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 .mmap = generic_file_mmap,
1161 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001162 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163#ifdef CONFIG_COMPAT
1164 .compat_ioctl = compat_blkdev_ioctl,
1165#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001166 .splice_read = generic_file_splice_read,
1167 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168};
1169
1170int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1171{
1172 int res;
1173 mm_segment_t old_fs = get_fs();
1174 set_fs(KERNEL_DS);
1175 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1176 set_fs(old_fs);
1177 return res;
1178}
1179
1180EXPORT_SYMBOL(ioctl_by_bdev);
1181
1182/**
1183 * lookup_bdev - lookup a struct block_device by name
1184 *
1185 * @path: special file representing the block device
1186 *
1187 * Get a reference to the blockdevice at @path in the current
1188 * namespace if possible and return it. Return ERR_PTR(error)
1189 * otherwise.
1190 */
1191struct block_device *lookup_bdev(const char *path)
1192{
1193 struct block_device *bdev;
1194 struct inode *inode;
1195 struct nameidata nd;
1196 int error;
1197
1198 if (!path || !*path)
1199 return ERR_PTR(-EINVAL);
1200
1201 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1202 if (error)
1203 return ERR_PTR(error);
1204
Jan Blunck4ac91372008-02-14 19:34:32 -08001205 inode = nd.path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 error = -ENOTBLK;
1207 if (!S_ISBLK(inode->i_mode))
1208 goto fail;
1209 error = -EACCES;
Jan Blunck4ac91372008-02-14 19:34:32 -08001210 if (nd.path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 goto fail;
1212 error = -ENOMEM;
1213 bdev = bd_acquire(inode);
1214 if (!bdev)
1215 goto fail;
1216out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001217 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 return bdev;
1219fail:
1220 bdev = ERR_PTR(error);
1221 goto out;
1222}
1223
1224/**
1225 * open_bdev_excl - open a block device by name and set it up for use
1226 *
1227 * @path: special file representing the block device
1228 * @flags: %MS_RDONLY for opening read-only
1229 * @holder: owner for exclusion
1230 *
1231 * Open the blockdevice described by the special file at @path, claim it
1232 * for the @holder.
1233 */
1234struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1235{
1236 struct block_device *bdev;
1237 mode_t mode = FMODE_READ;
1238 int error = 0;
1239
1240 bdev = lookup_bdev(path);
1241 if (IS_ERR(bdev))
1242 return bdev;
1243
1244 if (!(flags & MS_RDONLY))
1245 mode |= FMODE_WRITE;
1246 error = blkdev_get(bdev, mode, 0);
1247 if (error)
1248 return ERR_PTR(error);
1249 error = -EACCES;
1250 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1251 goto blkdev_put;
1252 error = bd_claim(bdev, holder);
1253 if (error)
1254 goto blkdev_put;
1255
1256 return bdev;
1257
1258blkdev_put:
1259 blkdev_put(bdev);
1260 return ERR_PTR(error);
1261}
1262
1263EXPORT_SYMBOL(open_bdev_excl);
1264
1265/**
1266 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1267 *
1268 * @bdev: blockdevice to close
1269 *
1270 * This is the counterpart to open_bdev_excl().
1271 */
1272void close_bdev_excl(struct block_device *bdev)
1273{
1274 bd_release(bdev);
1275 blkdev_put(bdev);
1276}
1277
1278EXPORT_SYMBOL(close_bdev_excl);
David Howellsb71e8a42006-08-29 19:06:11 +01001279
1280int __invalidate_device(struct block_device *bdev)
1281{
1282 struct super_block *sb = get_super(bdev);
1283 int res = 0;
1284
1285 if (sb) {
1286 /*
1287 * no need to lock the super, get_super holds the
1288 * read mutex so the filesystem cannot go away
1289 * under us (->put_super runs with the write lock
1290 * hold).
1291 */
1292 shrink_dcache_sb(sb);
1293 res = invalidate_inodes(sb);
1294 drop_super(sb);
1295 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001296 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001297 return res;
1298}
1299EXPORT_SYMBOL(__invalidate_device);