blob: 470c10ceb0fb3a33dc8f28e7e8b2d3535042075b [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>
David Howells811d7362006-08-29 19:06:09 +010021#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070026#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020028#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30struct bdev_inode {
31 struct block_device bdev;
32 struct inode vfs_inode;
33};
34
Adrian Bunk4c54ac62008-02-18 13:48:31 +010035static const struct address_space_operations def_blk_aops;
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037static inline struct bdev_inode *BDEV_I(struct inode *inode)
38{
39 return container_of(inode, struct bdev_inode, vfs_inode);
40}
41
42inline struct block_device *I_BDEV(struct inode *inode)
43{
44 return &BDEV_I(inode)->bdev;
45}
46
47EXPORT_SYMBOL(I_BDEV);
48
49static sector_t max_block(struct block_device *bdev)
50{
51 sector_t retval = ~((sector_t)0);
52 loff_t sz = i_size_read(bdev->bd_inode);
53
54 if (sz) {
55 unsigned int size = block_size(bdev);
56 unsigned int sizebits = blksize_bits(size);
57 retval = (sz >> sizebits);
58 }
59 return retval;
60}
61
Peter Zijlstraf9a14392007-05-06 14:49:55 -070062/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070063static void kill_bdev(struct block_device *bdev)
64{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070065 if (bdev->bd_inode->i_mapping->nrpages == 0)
66 return;
67 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
69}
70
71int set_blocksize(struct block_device *bdev, int size)
72{
73 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070074 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 return -EINVAL;
76
77 /* Size cannot be smaller than the size supported by the device */
78 if (size < bdev_hardsect_size(bdev))
79 return -EINVAL;
80
81 /* Don't change the size if it is same as current */
82 if (bdev->bd_block_size != size) {
83 sync_blockdev(bdev);
84 bdev->bd_block_size = size;
85 bdev->bd_inode->i_blkbits = blksize_bits(size);
86 kill_bdev(bdev);
87 }
88 return 0;
89}
90
91EXPORT_SYMBOL(set_blocksize);
92
93int sb_set_blocksize(struct super_block *sb, int size)
94{
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 if (set_blocksize(sb->s_bdev, size))
96 return 0;
97 /* If we get here, we know size is power of two
98 * and it's value is between 512 and PAGE_SIZE */
99 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800100 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return sb->s_blocksize;
102}
103
104EXPORT_SYMBOL(sb_set_blocksize);
105
106int sb_min_blocksize(struct super_block *sb, int size)
107{
108 int minsize = bdev_hardsect_size(sb->s_bdev);
109 if (size < minsize)
110 size = minsize;
111 return sb_set_blocksize(sb, size);
112}
113
114EXPORT_SYMBOL(sb_min_blocksize);
115
116static int
117blkdev_get_block(struct inode *inode, sector_t iblock,
118 struct buffer_head *bh, int create)
119{
120 if (iblock >= max_block(I_BDEV(inode))) {
121 if (create)
122 return -EIO;
123
124 /*
125 * for reads, we're just trying to fill a partial page.
126 * return a hole, they will have to call get_block again
127 * before they can fill it, and they will get -EIO at that
128 * time
129 */
130 return 0;
131 }
132 bh->b_bdev = I_BDEV(inode);
133 bh->b_blocknr = iblock;
134 set_buffer_mapped(bh);
135 return 0;
136}
137
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800138static int
139blkdev_get_blocks(struct inode *inode, sector_t iblock,
140 struct buffer_head *bh, int create)
141{
142 sector_t end_block = max_block(I_BDEV(inode));
143 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
144
145 if ((iblock + max_blocks) > end_block) {
146 max_blocks = end_block - iblock;
147 if ((long)max_blocks <= 0) {
148 if (create)
149 return -EIO; /* write fully beyond EOF */
150 /*
151 * It is a read which is fully beyond EOF. We return
152 * a !buffer_mapped buffer
153 */
154 max_blocks = 0;
155 }
156 }
157
158 bh->b_bdev = I_BDEV(inode);
159 bh->b_blocknr = iblock;
160 bh->b_size = max_blocks << inode->i_blkbits;
161 if (max_blocks)
162 set_buffer_mapped(bh);
163 return 0;
164}
165
166static ssize_t
167blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
168 loff_t offset, unsigned long nr_segs)
169{
170 struct file *file = iocb->ki_filp;
171 struct inode *inode = file->f_mapping->host;
172
173 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
174 iov, offset, nr_segs, blkdev_get_blocks, NULL);
175}
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
178{
179 return block_write_full_page(page, blkdev_get_block, wbc);
180}
181
182static int blkdev_readpage(struct file * file, struct page * page)
183{
184 return block_read_full_page(page, blkdev_get_block);
185}
186
Nick Piggin6272b5a2007-10-16 01:25:04 -0700187static int blkdev_write_begin(struct file *file, struct address_space *mapping,
188 loff_t pos, unsigned len, unsigned flags,
189 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700191 *pagep = NULL;
192 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
193 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194}
195
Nick Piggin6272b5a2007-10-16 01:25:04 -0700196static int blkdev_write_end(struct file *file, struct address_space *mapping,
197 loff_t pos, unsigned len, unsigned copied,
198 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700200 int ret;
201 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
202
203 unlock_page(page);
204 page_cache_release(page);
205
206 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207}
208
209/*
210 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800211 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * so we compute the size by hand (just as in block_read/write above)
213 */
214static loff_t block_llseek(struct file *file, loff_t offset, int origin)
215{
216 struct inode *bd_inode = file->f_mapping->host;
217 loff_t size;
218 loff_t retval;
219
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800220 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 size = i_size_read(bd_inode);
222
223 switch (origin) {
224 case 2:
225 offset += size;
226 break;
227 case 1:
228 offset += file->f_pos;
229 }
230 retval = -EINVAL;
231 if (offset >= 0 && offset <= size) {
232 if (offset != file->f_pos) {
233 file->f_pos = offset;
234 }
235 retval = offset;
236 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800237 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return retval;
239}
240
241/*
242 * Filp is never NULL; the only case when ->fsync() is called with
243 * NULL first argument is nfsd_sync_dir() and that's not a directory.
244 */
245
246static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
247{
248 return sync_blockdev(I_BDEV(filp->f_mapping->host));
249}
250
251/*
252 * pseudo-fs
253 */
254
255static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800256static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258static struct inode *bdev_alloc_inode(struct super_block *sb)
259{
Christoph Lametere94b1762006-12-06 20:33:17 -0800260 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 if (!ei)
262 return NULL;
263 return &ei->vfs_inode;
264}
265
266static void bdev_destroy_inode(struct inode *inode)
267{
268 struct bdev_inode *bdi = BDEV_I(inode);
269
270 bdi->bdev.bd_inode_backing_dev_info = NULL;
271 kmem_cache_free(bdev_cachep, bdi);
272}
273
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700274static void init_once(struct kmem_cache * cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct bdev_inode *ei = (struct bdev_inode *) foo;
277 struct block_device *bdev = &ei->bdev;
278
Christoph Lametera35afb82007-05-16 22:10:57 -0700279 memset(bdev, 0, sizeof(*bdev));
280 mutex_init(&bdev->bd_mutex);
281 sema_init(&bdev->bd_mount_sem, 1);
282 INIT_LIST_HEAD(&bdev->bd_inodes);
283 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800284#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700285 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800286#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700287 inode_init_once(&ei->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288}
289
290static inline void __bd_forget(struct inode *inode)
291{
292 list_del_init(&inode->i_devices);
293 inode->i_bdev = NULL;
294 inode->i_mapping = &inode->i_data;
295}
296
297static void bdev_clear_inode(struct inode *inode)
298{
299 struct block_device *bdev = &BDEV_I(inode)->bdev;
300 struct list_head *p;
301 spin_lock(&bdev_lock);
302 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
303 __bd_forget(list_entry(p, struct inode, i_devices));
304 }
305 list_del_init(&bdev->bd_list);
306 spin_unlock(&bdev_lock);
307}
308
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800309static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 .statfs = simple_statfs,
311 .alloc_inode = bdev_alloc_inode,
312 .destroy_inode = bdev_destroy_inode,
313 .drop_inode = generic_delete_inode,
314 .clear_inode = bdev_clear_inode,
315};
316
David Howells454e2392006-06-23 02:02:57 -0700317static int bd_get_sb(struct file_system_type *fs_type,
318 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319{
David Howells454e2392006-06-23 02:02:57 -0700320 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323static struct file_system_type bd_type = {
324 .name = "bdev",
325 .get_sb = bd_get_sb,
326 .kill_sb = kill_anon_super,
327};
328
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800329static struct vfsmount *bd_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330struct super_block *blockdev_superblock;
331
332void __init bdev_cache_init(void)
333{
334 int err;
335 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800336 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
337 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900338 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 err = register_filesystem(&bd_type);
340 if (err)
341 panic("Cannot register bdev pseudo-fs");
342 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (IS_ERR(bd_mnt))
344 panic("Cannot create bdev pseudo-fs");
345 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
346}
347
348/*
349 * Most likely _very_ bad one - but then it's hardly critical for small
350 * /dev and can be fixed when somebody will need really large one.
351 * Keep in mind that it will be fed through icache hash function too.
352 */
353static inline unsigned long hash(dev_t dev)
354{
355 return MAJOR(dev)+MINOR(dev);
356}
357
358static int bdev_test(struct inode *inode, void *data)
359{
360 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
361}
362
363static int bdev_set(struct inode *inode, void *data)
364{
365 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
366 return 0;
367}
368
369static LIST_HEAD(all_bdevs);
370
371struct block_device *bdget(dev_t dev)
372{
373 struct block_device *bdev;
374 struct inode *inode;
375
376 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
377 bdev_test, bdev_set, &dev);
378
379 if (!inode)
380 return NULL;
381
382 bdev = &BDEV_I(inode)->bdev;
383
384 if (inode->i_state & I_NEW) {
385 bdev->bd_contains = NULL;
386 bdev->bd_inode = inode;
387 bdev->bd_block_size = (1 << inode->i_blkbits);
388 bdev->bd_part_count = 0;
389 bdev->bd_invalidated = 0;
390 inode->i_mode = S_IFBLK;
391 inode->i_rdev = dev;
392 inode->i_bdev = bdev;
393 inode->i_data.a_ops = &def_blk_aops;
394 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
395 inode->i_data.backing_dev_info = &default_backing_dev_info;
396 spin_lock(&bdev_lock);
397 list_add(&bdev->bd_list, &all_bdevs);
398 spin_unlock(&bdev_lock);
399 unlock_new_inode(inode);
400 }
401 return bdev;
402}
403
404EXPORT_SYMBOL(bdget);
405
406long nr_blockdev_pages(void)
407{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700408 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 long ret = 0;
410 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700411 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 ret += bdev->bd_inode->i_mapping->nrpages;
413 }
414 spin_unlock(&bdev_lock);
415 return ret;
416}
417
418void bdput(struct block_device *bdev)
419{
420 iput(bdev->bd_inode);
421}
422
423EXPORT_SYMBOL(bdput);
424
425static struct block_device *bd_acquire(struct inode *inode)
426{
427 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 spin_lock(&bdev_lock);
430 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700431 if (bdev) {
432 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 spin_unlock(&bdev_lock);
434 return bdev;
435 }
436 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 bdev = bdget(inode->i_rdev);
439 if (bdev) {
440 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700441 if (!inode->i_bdev) {
442 /*
443 * We take an additional bd_inode->i_count for inode,
444 * and it's released in clear_inode() of inode.
445 * So, we can access it via ->i_mapping always
446 * without igrab().
447 */
448 atomic_inc(&bdev->bd_inode->i_count);
449 inode->i_bdev = bdev;
450 inode->i_mapping = bdev->bd_inode->i_mapping;
451 list_add(&inode->i_devices, &bdev->bd_inodes);
452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 spin_unlock(&bdev_lock);
454 }
455 return bdev;
456}
457
458/* Call when you free inode */
459
460void bd_forget(struct inode *inode)
461{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700462 struct block_device *bdev = NULL;
463
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700465 if (inode->i_bdev) {
466 if (inode->i_sb != blockdev_superblock)
467 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700471
472 if (bdev)
473 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
476int bd_claim(struct block_device *bdev, void *holder)
477{
478 int res;
479 spin_lock(&bdev_lock);
480
481 /* first decide result */
482 if (bdev->bd_holder == holder)
483 res = 0; /* already a holder */
484 else if (bdev->bd_holder != NULL)
485 res = -EBUSY; /* held by someone else */
486 else if (bdev->bd_contains == bdev)
487 res = 0; /* is a whole device which isn't held */
488
489 else if (bdev->bd_contains->bd_holder == bd_claim)
490 res = 0; /* is a partition of a device that is being partitioned */
491 else if (bdev->bd_contains->bd_holder != NULL)
492 res = -EBUSY; /* is a partition of a held device */
493 else
494 res = 0; /* is a partition of an un-held device */
495
496 /* now impose change */
497 if (res==0) {
498 /* note that for a whole device bd_holders
499 * will be incremented twice, and bd_holder will
500 * be set to bd_claim before being set to holder
501 */
502 bdev->bd_contains->bd_holders ++;
503 bdev->bd_contains->bd_holder = bd_claim;
504 bdev->bd_holders++;
505 bdev->bd_holder = holder;
506 }
507 spin_unlock(&bdev_lock);
508 return res;
509}
510
511EXPORT_SYMBOL(bd_claim);
512
513void bd_release(struct block_device *bdev)
514{
515 spin_lock(&bdev_lock);
516 if (!--bdev->bd_contains->bd_holders)
517 bdev->bd_contains->bd_holder = NULL;
518 if (!--bdev->bd_holders)
519 bdev->bd_holder = NULL;
520 spin_unlock(&bdev_lock);
521}
522
523EXPORT_SYMBOL(bd_release);
524
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800525#ifdef CONFIG_SYSFS
526/*
527 * Functions for bd_claim_by_kobject / bd_release_from_kobject
528 *
529 * If a kobject is passed to bd_claim_by_kobject()
530 * and the kobject has a parent directory,
531 * following symlinks are created:
532 * o from the kobject to the claimed bdev
533 * o from "holders" directory of the bdev to the parent of the kobject
534 * bd_release_from_kobject() removes these symlinks.
535 *
536 * Example:
537 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
538 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
539 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
540 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
541 */
542
543static struct kobject *bdev_get_kobj(struct block_device *bdev)
544{
545 if (bdev->bd_contains != bdev)
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200546 return kobject_get(&bdev->bd_part->dev.kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800547 else
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200548 return kobject_get(&bdev->bd_disk->dev.kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800549}
550
551static struct kobject *bdev_get_holder(struct block_device *bdev)
552{
553 if (bdev->bd_contains != bdev)
554 return kobject_get(bdev->bd_part->holder_dir);
555 else
556 return kobject_get(bdev->bd_disk->holder_dir);
557}
558
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700559static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800560{
561 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700562 return 0;
563 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800564}
565
566static void del_symlink(struct kobject *from, struct kobject *to)
567{
568 if (!from || !to)
569 return;
570 sysfs_remove_link(from, kobject_name(to));
571}
572
573/*
574 * 'struct bd_holder' contains pointers to kobjects symlinked by
575 * bd_claim_by_kobject.
576 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
577 */
578struct bd_holder {
579 struct list_head list; /* chain of holders of the bdev */
580 int count; /* references from the holder */
581 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
582 struct kobject *hdev; /* e.g. "/block/dm-0" */
583 struct kobject *hdir; /* e.g. "/block/sda/holders" */
584 struct kobject *sdev; /* e.g. "/block/sda" */
585};
586
587/*
588 * Get references of related kobjects at once.
589 * Returns 1 on success. 0 on failure.
590 *
591 * Should call bd_holder_release_dirs() after successful use.
592 */
593static int bd_holder_grab_dirs(struct block_device *bdev,
594 struct bd_holder *bo)
595{
596 if (!bdev || !bo)
597 return 0;
598
599 bo->sdir = kobject_get(bo->sdir);
600 if (!bo->sdir)
601 return 0;
602
603 bo->hdev = kobject_get(bo->sdir->parent);
604 if (!bo->hdev)
605 goto fail_put_sdir;
606
607 bo->sdev = bdev_get_kobj(bdev);
608 if (!bo->sdev)
609 goto fail_put_hdev;
610
611 bo->hdir = bdev_get_holder(bdev);
612 if (!bo->hdir)
613 goto fail_put_sdev;
614
615 return 1;
616
617fail_put_sdev:
618 kobject_put(bo->sdev);
619fail_put_hdev:
620 kobject_put(bo->hdev);
621fail_put_sdir:
622 kobject_put(bo->sdir);
623
624 return 0;
625}
626
627/* Put references of related kobjects at once. */
628static void bd_holder_release_dirs(struct bd_holder *bo)
629{
630 kobject_put(bo->hdir);
631 kobject_put(bo->sdev);
632 kobject_put(bo->hdev);
633 kobject_put(bo->sdir);
634}
635
636static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
637{
638 struct bd_holder *bo;
639
640 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
641 if (!bo)
642 return NULL;
643
644 bo->count = 1;
645 bo->sdir = kobj;
646
647 return bo;
648}
649
650static void free_bd_holder(struct bd_holder *bo)
651{
652 kfree(bo);
653}
654
655/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500656 * find_bd_holder - find matching struct bd_holder from the block device
657 *
658 * @bdev: struct block device to be searched
659 * @bo: target struct bd_holder
660 *
661 * Returns matching entry with @bo in @bdev->bd_holder_list.
662 * If found, increment the reference count and return the pointer.
663 * If not found, returns NULL.
664 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800665static struct bd_holder *find_bd_holder(struct block_device *bdev,
666 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500667{
668 struct bd_holder *tmp;
669
670 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
671 if (tmp->sdir == bo->sdir) {
672 tmp->count++;
673 return tmp;
674 }
675
676 return NULL;
677}
678
679/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800680 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
681 *
682 * @bdev: block device to be bd_claimed
683 * @bo: preallocated and initialized by alloc_bd_holder()
684 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500685 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800686 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500687 * Returns 0 if symlinks are created.
688 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800689 */
690static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
691{
Johannes Weiner4e916722007-07-15 23:41:25 -0700692 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800693
694 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700695 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800696
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800697 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700698 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800699
Johannes Weiner4e916722007-07-15 23:41:25 -0700700 err = add_symlink(bo->sdir, bo->sdev);
701 if (err)
702 return err;
703
704 err = add_symlink(bo->hdir, bo->hdev);
705 if (err) {
706 del_symlink(bo->sdir, bo->sdev);
707 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700708 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700709
710 list_add_tail(&bo->list, &bdev->bd_holder_list);
711 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800712}
713
714/**
715 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
716 *
717 * @bdev: block device to be bd_claimed
718 * @kobj: holder's kobject
719 *
720 * If there is matching entry with @kobj in @bdev->bd_holder_list
721 * and no other bd_claim() from the same kobject,
722 * remove the struct bd_holder from the list, delete symlinks for it.
723 *
724 * Returns a pointer to the struct bd_holder when it's removed from the list
725 * and ready to be freed.
726 * Returns NULL if matching claim isn't found or there is other bd_claim()
727 * by the same kobject.
728 */
729static struct bd_holder *del_bd_holder(struct block_device *bdev,
730 struct kobject *kobj)
731{
732 struct bd_holder *bo;
733
734 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
735 if (bo->sdir == kobj) {
736 bo->count--;
737 BUG_ON(bo->count < 0);
738 if (!bo->count) {
739 list_del(&bo->list);
740 del_symlink(bo->sdir, bo->sdev);
741 del_symlink(bo->hdir, bo->hdev);
742 bd_holder_release_dirs(bo);
743 return bo;
744 }
745 break;
746 }
747 }
748
749 return NULL;
750}
751
752/**
753 * bd_claim_by_kobject - bd_claim() with additional kobject signature
754 *
755 * @bdev: block device to be claimed
756 * @holder: holder's signature
757 * @kobj: holder's kobject
758 *
759 * Do bd_claim() and if it succeeds, create sysfs symlinks between
760 * the bdev and the holder's kobject.
761 * Use bd_release_from_kobject() when relesing the claimed bdev.
762 *
763 * Returns 0 on success. (same as bd_claim())
764 * Returns errno on failure.
765 */
766static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
767 struct kobject *kobj)
768{
Johannes Weiner4e916722007-07-15 23:41:25 -0700769 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500770 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800771
772 if (!kobj)
773 return -EINVAL;
774
775 bo = alloc_bd_holder(kobj);
776 if (!bo)
777 return -ENOMEM;
778
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800779 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500780
Johannes Weiner4e916722007-07-15 23:41:25 -0700781 err = bd_claim(bdev, holder);
782 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700783 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700784
785 found = find_bd_holder(bdev, bo);
786 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700787 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700788
789 err = add_bd_holder(bdev, bo);
790 if (err)
791 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700792 else
793 bo = NULL;
794fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800795 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700796 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700797 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800798}
799
800/**
801 * bd_release_from_kobject - bd_release() with additional kobject signature
802 *
803 * @bdev: block device to be released
804 * @kobj: holder's kobject
805 *
806 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
807 */
808static void bd_release_from_kobject(struct block_device *bdev,
809 struct kobject *kobj)
810{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800811 if (!kobj)
812 return;
813
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800814 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800815 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700816 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800817 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800818}
819
820/**
821 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
822 *
823 * @bdev: block device to be claimed
824 * @holder: holder's signature
825 * @disk: holder's gendisk
826 *
827 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
828 */
829int bd_claim_by_disk(struct block_device *bdev, void *holder,
830 struct gendisk *disk)
831{
832 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
833}
834EXPORT_SYMBOL_GPL(bd_claim_by_disk);
835
836/**
837 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
838 *
839 * @bdev: block device to be claimed
840 * @disk: holder's gendisk
841 *
842 * Call bd_release_from_kobject() and put @disk->slave_dir.
843 */
844void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
845{
846 bd_release_from_kobject(bdev, disk->slave_dir);
847 kobject_put(disk->slave_dir);
848}
849EXPORT_SYMBOL_GPL(bd_release_from_disk);
850#endif
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852/*
853 * Tries to open block device by device number. Use it ONLY if you
854 * really do not have anything better - i.e. when you are behind a
855 * truly sucky interface and all you are given is a device number. _Never_
856 * to be used for internal purposes. If you ever need it - reconsider
857 * your API.
858 */
859struct block_device *open_by_devnum(dev_t dev, unsigned mode)
860{
861 struct block_device *bdev = bdget(dev);
862 int err = -ENOMEM;
863 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
864 if (bdev)
865 err = blkdev_get(bdev, mode, flags);
866 return err ? ERR_PTR(err) : bdev;
867}
868
869EXPORT_SYMBOL(open_by_devnum);
870
871/*
872 * This routine checks whether a removable media has been changed,
873 * and invalidates all buffer-cache-entries in that case. This
874 * is a relatively slow routine, so we have to try to minimize using
875 * it. Thus it is called only upon a 'mount' or 'open'. This
876 * is the best way of combining speed and utility, I think.
877 * People changing diskettes in the middle of an operation deserve
878 * to lose :-)
879 */
880int check_disk_change(struct block_device *bdev)
881{
882 struct gendisk *disk = bdev->bd_disk;
883 struct block_device_operations * bdops = disk->fops;
884
885 if (!bdops->media_changed)
886 return 0;
887 if (!bdops->media_changed(bdev->bd_disk))
888 return 0;
889
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700890 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 printk("VFS: busy inodes on changed media.\n");
892
893 if (bdops->revalidate_disk)
894 bdops->revalidate_disk(bdev->bd_disk);
895 if (bdev->bd_disk->minors > 1)
896 bdev->bd_invalidated = 1;
897 return 1;
898}
899
900EXPORT_SYMBOL(check_disk_change);
901
902void bd_set_size(struct block_device *bdev, loff_t size)
903{
904 unsigned bsize = bdev_hardsect_size(bdev);
905
906 bdev->bd_inode->i_size = size;
907 while (bsize < PAGE_CACHE_SIZE) {
908 if (size & bsize)
909 break;
910 bsize <<= 1;
911 }
912 bdev->bd_block_size = bsize;
913 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
914}
915EXPORT_SYMBOL(bd_set_size);
916
NeilBrown37be4122006-12-08 02:36:16 -0800917static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
918 int for_part);
Peter Zijlstrac48f70c2006-12-08 02:36:18 -0800919static int __blkdev_put(struct block_device *bdev, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -0800920
Peter Zijlstra6d740cd2007-02-20 13:58:18 -0800921/*
922 * bd_mutex locking:
923 *
924 * mutex_lock(part->bd_mutex)
925 * mutex_lock_nested(whole->bd_mutex, 1)
926 */
927
NeilBrown37be4122006-12-08 02:36:16 -0800928static int do_open(struct block_device *bdev, struct file *file, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929{
930 struct module *owner = NULL;
931 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -0700932 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 int part;
934
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -0700935 ret = devcgroup_inode_permission(bdev->bd_inode, file->f_mode);
936 if (ret != 0)
937 return ret;
938
939 ret = -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 file->f_mapping = bdev->bd_inode->i_mapping;
941 lock_kernel();
942 disk = get_gendisk(bdev->bd_dev, &part);
943 if (!disk) {
944 unlock_kernel();
945 bdput(bdev);
946 return ret;
947 }
948 owner = disk->fops->owner;
949
NeilBrown6796bf52006-12-08 02:36:16 -0800950 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 if (!bdev->bd_openers) {
952 bdev->bd_disk = disk;
953 bdev->bd_contains = bdev;
954 if (!part) {
955 struct backing_dev_info *bdi;
956 if (disk->fops->open) {
957 ret = disk->fops->open(bdev->bd_inode, file);
958 if (ret)
959 goto out_first;
960 }
961 if (!bdev->bd_openers) {
962 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
963 bdi = blk_get_backing_dev_info(bdev);
964 if (bdi == NULL)
965 bdi = &default_backing_dev_info;
966 bdev->bd_inode->i_data.backing_dev_info = bdi;
967 }
968 if (bdev->bd_invalidated)
969 rescan_partitions(disk, bdev);
970 } else {
971 struct hd_struct *p;
972 struct block_device *whole;
973 whole = bdget_disk(disk, 0);
974 ret = -ENOMEM;
975 if (!whole)
976 goto out_first;
NeilBrown37be4122006-12-08 02:36:16 -0800977 BUG_ON(for_part);
978 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (ret)
980 goto out_first;
981 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 p = disk->part[part - 1];
983 bdev->bd_inode->i_data.backing_dev_info =
984 whole->bd_inode->i_data.backing_dev_info;
985 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 ret = -ENXIO;
987 goto out_first;
988 }
Kay Sieversedfaa7c2007-05-21 22:08:01 +0200989 kobject_get(&p->dev.kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 bdev->bd_part = p;
991 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 }
993 } else {
994 put_disk(disk);
995 module_put(owner);
996 if (bdev->bd_contains == bdev) {
997 if (bdev->bd_disk->fops->open) {
998 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
999 if (ret)
1000 goto out;
1001 }
1002 if (bdev->bd_invalidated)
1003 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 }
1006 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001007 if (for_part)
1008 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001009 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 unlock_kernel();
1011 return 0;
1012
1013out_first:
1014 bdev->bd_disk = NULL;
1015 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1016 if (bdev != bdev->bd_contains)
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001017 __blkdev_put(bdev->bd_contains, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 bdev->bd_contains = NULL;
1019 put_disk(disk);
1020 module_put(owner);
1021out:
Arjan van de Venc039e312006-03-23 03:00:28 -08001022 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 unlock_kernel();
1024 if (ret)
1025 bdput(bdev);
1026 return ret;
1027}
1028
NeilBrown37be4122006-12-08 02:36:16 -08001029static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1030 int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 /*
1033 * This crockload is due to bad choice of ->open() type.
1034 * It will go away.
1035 * For now, block device ->open() routine must _not_
1036 * examine anything in 'inode' argument except ->i_rdev.
1037 */
1038 struct file fake_file = {};
1039 struct dentry fake_dentry = {};
1040 fake_file.f_mode = mode;
1041 fake_file.f_flags = flags;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001042 fake_file.f_path.dentry = &fake_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 fake_dentry.d_inode = bdev->bd_inode;
1044
NeilBrown37be4122006-12-08 02:36:16 -08001045 return do_open(bdev, &fake_file, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
NeilBrown37be4122006-12-08 02:36:16 -08001048int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1049{
1050 return __blkdev_get(bdev, mode, flags, 0);
1051}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052EXPORT_SYMBOL(blkdev_get);
1053
1054static int blkdev_open(struct inode * inode, struct file * filp)
1055{
1056 struct block_device *bdev;
1057 int res;
1058
1059 /*
1060 * Preserve backwards compatibility and allow large file access
1061 * even if userspace doesn't ask for it explicitly. Some mkfs
1062 * binary needs it. We might want to drop this workaround
1063 * during an unstable branch.
1064 */
1065 filp->f_flags |= O_LARGEFILE;
1066
1067 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001068 if (bdev == NULL)
1069 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
NeilBrown37be4122006-12-08 02:36:16 -08001071 res = do_open(bdev, filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (res)
1073 return res;
1074
1075 if (!(filp->f_flags & O_EXCL) )
1076 return 0;
1077
1078 if (!(res = bd_claim(bdev, filp)))
1079 return 0;
1080
1081 blkdev_put(bdev);
1082 return res;
1083}
1084
NeilBrown37be4122006-12-08 02:36:16 -08001085static int __blkdev_put(struct block_device *bdev, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001086{
1087 int ret = 0;
1088 struct inode *bd_inode = bdev->bd_inode;
1089 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001090 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001091
NeilBrown6796bf52006-12-08 02:36:16 -08001092 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001093 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001094 if (for_part)
1095 bdev->bd_part_count--;
1096
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001097 if (!--bdev->bd_openers) {
1098 sync_blockdev(bdev);
1099 kill_bdev(bdev);
1100 }
1101 if (bdev->bd_contains == bdev) {
1102 if (disk->fops->release)
1103 ret = disk->fops->release(bd_inode, NULL);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001104 }
1105 if (!bdev->bd_openers) {
1106 struct module *owner = disk->fops->owner;
1107
1108 put_disk(disk);
1109 module_put(owner);
1110
1111 if (bdev->bd_contains != bdev) {
Kay Sieversedfaa7c2007-05-21 22:08:01 +02001112 kobject_put(&bdev->bd_part->dev.kobj);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001113 bdev->bd_part = NULL;
1114 }
1115 bdev->bd_disk = NULL;
1116 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001117 if (bdev != bdev->bd_contains)
1118 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001119 bdev->bd_contains = NULL;
1120 }
1121 unlock_kernel();
1122 mutex_unlock(&bdev->bd_mutex);
1123 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001124 if (victim)
1125 __blkdev_put(victim, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001126 return ret;
1127}
1128
NeilBrown37be4122006-12-08 02:36:16 -08001129int blkdev_put(struct block_device *bdev)
1130{
1131 return __blkdev_put(bdev, 0);
1132}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001133EXPORT_SYMBOL(blkdev_put);
1134
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135static int blkdev_close(struct inode * inode, struct file * filp)
1136{
1137 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1138 if (bdev->bd_holder == filp)
1139 bd_release(bdev);
1140 return blkdev_put(bdev);
1141}
1142
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001143static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
1145 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1146}
1147
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001148static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 .readpage = blkdev_readpage,
1150 .writepage = blkdev_writepage,
1151 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001152 .write_begin = blkdev_write_begin,
1153 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 .writepages = generic_writepages,
1155 .direct_IO = blkdev_direct_IO,
1156};
1157
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001158const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 .open = blkdev_open,
1160 .release = blkdev_close,
1161 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001162 .read = do_sync_read,
1163 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001165 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 .mmap = generic_file_mmap,
1167 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001168 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169#ifdef CONFIG_COMPAT
1170 .compat_ioctl = compat_blkdev_ioctl,
1171#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001172 .splice_read = generic_file_splice_read,
1173 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174};
1175
1176int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1177{
1178 int res;
1179 mm_segment_t old_fs = get_fs();
1180 set_fs(KERNEL_DS);
1181 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1182 set_fs(old_fs);
1183 return res;
1184}
1185
1186EXPORT_SYMBOL(ioctl_by_bdev);
1187
1188/**
1189 * lookup_bdev - lookup a struct block_device by name
1190 *
1191 * @path: special file representing the block device
1192 *
1193 * Get a reference to the blockdevice at @path in the current
1194 * namespace if possible and return it. Return ERR_PTR(error)
1195 * otherwise.
1196 */
1197struct block_device *lookup_bdev(const char *path)
1198{
1199 struct block_device *bdev;
1200 struct inode *inode;
1201 struct nameidata nd;
1202 int error;
1203
1204 if (!path || !*path)
1205 return ERR_PTR(-EINVAL);
1206
1207 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1208 if (error)
1209 return ERR_PTR(error);
1210
Jan Blunck4ac91372008-02-14 19:34:32 -08001211 inode = nd.path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 error = -ENOTBLK;
1213 if (!S_ISBLK(inode->i_mode))
1214 goto fail;
1215 error = -EACCES;
Jan Blunck4ac91372008-02-14 19:34:32 -08001216 if (nd.path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 goto fail;
1218 error = -ENOMEM;
1219 bdev = bd_acquire(inode);
1220 if (!bdev)
1221 goto fail;
1222out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001223 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 return bdev;
1225fail:
1226 bdev = ERR_PTR(error);
1227 goto out;
1228}
1229
1230/**
1231 * open_bdev_excl - open a block device by name and set it up for use
1232 *
1233 * @path: special file representing the block device
1234 * @flags: %MS_RDONLY for opening read-only
1235 * @holder: owner for exclusion
1236 *
1237 * Open the blockdevice described by the special file at @path, claim it
1238 * for the @holder.
1239 */
1240struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1241{
1242 struct block_device *bdev;
1243 mode_t mode = FMODE_READ;
1244 int error = 0;
1245
1246 bdev = lookup_bdev(path);
1247 if (IS_ERR(bdev))
1248 return bdev;
1249
1250 if (!(flags & MS_RDONLY))
1251 mode |= FMODE_WRITE;
1252 error = blkdev_get(bdev, mode, 0);
1253 if (error)
1254 return ERR_PTR(error);
1255 error = -EACCES;
1256 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1257 goto blkdev_put;
1258 error = bd_claim(bdev, holder);
1259 if (error)
1260 goto blkdev_put;
1261
1262 return bdev;
1263
1264blkdev_put:
1265 blkdev_put(bdev);
1266 return ERR_PTR(error);
1267}
1268
1269EXPORT_SYMBOL(open_bdev_excl);
1270
1271/**
1272 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1273 *
1274 * @bdev: blockdevice to close
1275 *
1276 * This is the counterpart to open_bdev_excl().
1277 */
1278void close_bdev_excl(struct block_device *bdev)
1279{
1280 bd_release(bdev);
1281 blkdev_put(bdev);
1282}
1283
1284EXPORT_SYMBOL(close_bdev_excl);
David Howellsb71e8a42006-08-29 19:06:11 +01001285
1286int __invalidate_device(struct block_device *bdev)
1287{
1288 struct super_block *sb = get_super(bdev);
1289 int res = 0;
1290
1291 if (sb) {
1292 /*
1293 * no need to lock the super, get_super holds the
1294 * read mutex so the filesystem cannot go away
1295 * under us (->put_super runs with the write lock
1296 * hold).
1297 */
1298 shrink_dcache_sb(sb);
1299 res = invalidate_inodes(sb);
1300 drop_super(sb);
1301 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001302 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001303 return res;
1304}
1305EXPORT_SYMBOL(__invalidate_device);