blob: 88a776fa0ef6e29bc67322f3184cb04f31b2248d [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
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700274static void init_once(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
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700543static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800544{
545 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700546 return 0;
547 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800548}
549
550static void del_symlink(struct kobject *from, struct kobject *to)
551{
552 if (!from || !to)
553 return;
554 sysfs_remove_link(from, kobject_name(to));
555}
556
557/*
558 * 'struct bd_holder' contains pointers to kobjects symlinked by
559 * bd_claim_by_kobject.
560 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
561 */
562struct bd_holder {
563 struct list_head list; /* chain of holders of the bdev */
564 int count; /* references from the holder */
565 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
566 struct kobject *hdev; /* e.g. "/block/dm-0" */
567 struct kobject *hdir; /* e.g. "/block/sda/holders" */
568 struct kobject *sdev; /* e.g. "/block/sda" */
569};
570
571/*
572 * Get references of related kobjects at once.
573 * Returns 1 on success. 0 on failure.
574 *
575 * Should call bd_holder_release_dirs() after successful use.
576 */
577static int bd_holder_grab_dirs(struct block_device *bdev,
578 struct bd_holder *bo)
579{
580 if (!bdev || !bo)
581 return 0;
582
583 bo->sdir = kobject_get(bo->sdir);
584 if (!bo->sdir)
585 return 0;
586
587 bo->hdev = kobject_get(bo->sdir->parent);
588 if (!bo->hdev)
589 goto fail_put_sdir;
590
Tejun Heo0762b8b2008-08-25 19:56:12 +0900591 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800592 if (!bo->sdev)
593 goto fail_put_hdev;
594
Tejun Heo4c465012008-08-25 19:56:11 +0900595 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800596 if (!bo->hdir)
597 goto fail_put_sdev;
598
599 return 1;
600
601fail_put_sdev:
602 kobject_put(bo->sdev);
603fail_put_hdev:
604 kobject_put(bo->hdev);
605fail_put_sdir:
606 kobject_put(bo->sdir);
607
608 return 0;
609}
610
611/* Put references of related kobjects at once. */
612static void bd_holder_release_dirs(struct bd_holder *bo)
613{
614 kobject_put(bo->hdir);
615 kobject_put(bo->sdev);
616 kobject_put(bo->hdev);
617 kobject_put(bo->sdir);
618}
619
620static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
621{
622 struct bd_holder *bo;
623
624 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
625 if (!bo)
626 return NULL;
627
628 bo->count = 1;
629 bo->sdir = kobj;
630
631 return bo;
632}
633
634static void free_bd_holder(struct bd_holder *bo)
635{
636 kfree(bo);
637}
638
639/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500640 * find_bd_holder - find matching struct bd_holder from the block device
641 *
642 * @bdev: struct block device to be searched
643 * @bo: target struct bd_holder
644 *
645 * Returns matching entry with @bo in @bdev->bd_holder_list.
646 * If found, increment the reference count and return the pointer.
647 * If not found, returns NULL.
648 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800649static struct bd_holder *find_bd_holder(struct block_device *bdev,
650 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500651{
652 struct bd_holder *tmp;
653
654 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
655 if (tmp->sdir == bo->sdir) {
656 tmp->count++;
657 return tmp;
658 }
659
660 return NULL;
661}
662
663/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800664 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
665 *
666 * @bdev: block device to be bd_claimed
667 * @bo: preallocated and initialized by alloc_bd_holder()
668 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500669 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800670 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500671 * Returns 0 if symlinks are created.
672 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800673 */
674static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
675{
Johannes Weiner4e916722007-07-15 23:41:25 -0700676 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800677
678 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700679 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800680
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800681 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700682 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800683
Johannes Weiner4e916722007-07-15 23:41:25 -0700684 err = add_symlink(bo->sdir, bo->sdev);
685 if (err)
686 return err;
687
688 err = add_symlink(bo->hdir, bo->hdev);
689 if (err) {
690 del_symlink(bo->sdir, bo->sdev);
691 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700692 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700693
694 list_add_tail(&bo->list, &bdev->bd_holder_list);
695 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800696}
697
698/**
699 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
700 *
701 * @bdev: block device to be bd_claimed
702 * @kobj: holder's kobject
703 *
704 * If there is matching entry with @kobj in @bdev->bd_holder_list
705 * and no other bd_claim() from the same kobject,
706 * remove the struct bd_holder from the list, delete symlinks for it.
707 *
708 * Returns a pointer to the struct bd_holder when it's removed from the list
709 * and ready to be freed.
710 * Returns NULL if matching claim isn't found or there is other bd_claim()
711 * by the same kobject.
712 */
713static struct bd_holder *del_bd_holder(struct block_device *bdev,
714 struct kobject *kobj)
715{
716 struct bd_holder *bo;
717
718 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
719 if (bo->sdir == kobj) {
720 bo->count--;
721 BUG_ON(bo->count < 0);
722 if (!bo->count) {
723 list_del(&bo->list);
724 del_symlink(bo->sdir, bo->sdev);
725 del_symlink(bo->hdir, bo->hdev);
726 bd_holder_release_dirs(bo);
727 return bo;
728 }
729 break;
730 }
731 }
732
733 return NULL;
734}
735
736/**
737 * bd_claim_by_kobject - bd_claim() with additional kobject signature
738 *
739 * @bdev: block device to be claimed
740 * @holder: holder's signature
741 * @kobj: holder's kobject
742 *
743 * Do bd_claim() and if it succeeds, create sysfs symlinks between
744 * the bdev and the holder's kobject.
745 * Use bd_release_from_kobject() when relesing the claimed bdev.
746 *
747 * Returns 0 on success. (same as bd_claim())
748 * Returns errno on failure.
749 */
750static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
751 struct kobject *kobj)
752{
Johannes Weiner4e916722007-07-15 23:41:25 -0700753 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500754 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800755
756 if (!kobj)
757 return -EINVAL;
758
759 bo = alloc_bd_holder(kobj);
760 if (!bo)
761 return -ENOMEM;
762
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800763 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500764
Johannes Weiner4e916722007-07-15 23:41:25 -0700765 err = bd_claim(bdev, holder);
766 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700767 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700768
769 found = find_bd_holder(bdev, bo);
770 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700771 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700772
773 err = add_bd_holder(bdev, bo);
774 if (err)
775 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700776 else
777 bo = NULL;
778fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800779 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700780 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700781 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800782}
783
784/**
785 * bd_release_from_kobject - bd_release() with additional kobject signature
786 *
787 * @bdev: block device to be released
788 * @kobj: holder's kobject
789 *
790 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
791 */
792static void bd_release_from_kobject(struct block_device *bdev,
793 struct kobject *kobj)
794{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800795 if (!kobj)
796 return;
797
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800798 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800799 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700800 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800801 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800802}
803
804/**
805 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
806 *
807 * @bdev: block device to be claimed
808 * @holder: holder's signature
809 * @disk: holder's gendisk
810 *
811 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
812 */
813int bd_claim_by_disk(struct block_device *bdev, void *holder,
814 struct gendisk *disk)
815{
816 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
817}
818EXPORT_SYMBOL_GPL(bd_claim_by_disk);
819
820/**
821 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
822 *
823 * @bdev: block device to be claimed
824 * @disk: holder's gendisk
825 *
826 * Call bd_release_from_kobject() and put @disk->slave_dir.
827 */
828void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
829{
830 bd_release_from_kobject(bdev, disk->slave_dir);
831 kobject_put(disk->slave_dir);
832}
833EXPORT_SYMBOL_GPL(bd_release_from_disk);
834#endif
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836/*
837 * Tries to open block device by device number. Use it ONLY if you
838 * really do not have anything better - i.e. when you are behind a
839 * truly sucky interface and all you are given is a device number. _Never_
840 * to be used for internal purposes. If you ever need it - reconsider
841 * your API.
842 */
Al Viroaeb5d722008-09-02 15:28:45 -0400843struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844{
845 struct block_device *bdev = bdget(dev);
846 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -0400848 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return err ? ERR_PTR(err) : bdev;
850}
851
852EXPORT_SYMBOL(open_by_devnum);
853
Andrew Patterson0c002c22008-09-04 14:27:20 -0600854/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600855 * flush_disk - invalidates all buffer-cache entries on a disk
856 *
857 * @bdev: struct block device to be flushed
858 *
859 * Invalidates all buffer-cache entries on a disk. It should be called
860 * when a disk has been changed -- either by a media change or online
861 * resize.
862 */
863static void flush_disk(struct block_device *bdev)
864{
865 if (__invalidate_device(bdev)) {
866 char name[BDEVNAME_SIZE] = "";
867
868 if (bdev->bd_disk)
869 disk_name(bdev->bd_disk, 0, name);
870 printk(KERN_WARNING "VFS: busy inodes on changed media or "
871 "resized disk %s\n", name);
872 }
873
874 if (!bdev->bd_disk)
875 return;
876 if (disk_partitionable(bdev->bd_disk))
877 bdev->bd_invalidated = 1;
878}
879
880/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200881 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600882 * @disk: struct gendisk to check
883 * @bdev: struct bdev to adjust.
884 *
885 * This routine checks to see if the bdev size does not match the disk size
886 * and adjusts it if it differs.
887 */
888void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
889{
890 loff_t disk_size, bdev_size;
891
892 disk_size = (loff_t)get_capacity(disk) << 9;
893 bdev_size = i_size_read(bdev->bd_inode);
894 if (disk_size != bdev_size) {
895 char name[BDEVNAME_SIZE];
896
897 disk_name(disk, 0, name);
898 printk(KERN_INFO
899 "%s: detected capacity change from %lld to %lld\n",
900 name, bdev_size, disk_size);
901 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -0600902 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600903 }
904}
905EXPORT_SYMBOL(check_disk_size_change);
906
907/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200908 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -0600909 * @disk: struct gendisk to be revalidated
910 *
911 * This routine is a wrapper for lower-level driver's revalidate_disk
912 * call-backs. It is used to do common pre and post operations needed
913 * for all revalidate_disk operations.
914 */
915int revalidate_disk(struct gendisk *disk)
916{
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600917 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -0600918 int ret = 0;
919
920 if (disk->fops->revalidate_disk)
921 ret = disk->fops->revalidate_disk(disk);
922
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600923 bdev = bdget_disk(disk, 0);
924 if (!bdev)
925 return ret;
926
927 mutex_lock(&bdev->bd_mutex);
928 check_disk_size_change(disk, bdev);
929 mutex_unlock(&bdev->bd_mutex);
930 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -0600931 return ret;
932}
933EXPORT_SYMBOL(revalidate_disk);
934
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935/*
936 * This routine checks whether a removable media has been changed,
937 * and invalidates all buffer-cache-entries in that case. This
938 * is a relatively slow routine, so we have to try to minimize using
939 * it. Thus it is called only upon a 'mount' or 'open'. This
940 * is the best way of combining speed and utility, I think.
941 * People changing diskettes in the middle of an operation deserve
942 * to lose :-)
943 */
944int check_disk_change(struct block_device *bdev)
945{
946 struct gendisk *disk = bdev->bd_disk;
947 struct block_device_operations * bdops = disk->fops;
948
949 if (!bdops->media_changed)
950 return 0;
951 if (!bdops->media_changed(bdev->bd_disk))
952 return 0;
953
Andrew Patterson56ade442008-09-04 14:27:40 -0600954 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 if (bdops->revalidate_disk)
956 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 return 1;
958}
959
960EXPORT_SYMBOL(check_disk_change);
961
962void bd_set_size(struct block_device *bdev, loff_t size)
963{
964 unsigned bsize = bdev_hardsect_size(bdev);
965
966 bdev->bd_inode->i_size = size;
967 while (bsize < PAGE_CACHE_SIZE) {
968 if (size & bsize)
969 break;
970 bsize <<= 1;
971 }
972 bdev->bd_block_size = bsize;
973 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
974}
975EXPORT_SYMBOL(bd_set_size);
976
Al Viro9a1c3542008-02-22 20:40:24 -0500977static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -0800978
Peter Zijlstra6d740cd2007-02-20 13:58:18 -0800979/*
980 * bd_mutex locking:
981 *
982 * mutex_lock(part->bd_mutex)
983 * mutex_lock_nested(whole->bd_mutex, 1)
984 */
985
Al Viro572c4892007-10-08 13:24:05 -0400986static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 struct gendisk *disk;
Tejun Heoe71bf0d2008-09-03 09:03:02 +0200989 struct hd_struct *part = NULL;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -0700990 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +0200991 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -0400992 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Al Viro572c4892007-10-08 13:24:05 -0400994 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -0400995 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -0400996 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -0400997 perm |= MAY_WRITE;
998 /*
999 * hooks: /n/, see "layering violations".
1000 */
1001 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
Al Viro82666022008-08-01 05:32:04 -04001002 if (ret != 0) {
1003 bdput(bdev);
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001004 return ret;
Al Viro82666022008-08-01 05:32:04 -04001005 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001006
1007 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 lock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001010
Tejun Heocf771cb2008-09-03 09:01:09 +02001011 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001012 if (!disk)
1013 goto out_unlock_kernel;
1014 part = disk_get_part(disk, partno);
1015 if (!part)
1016 goto out_unlock_kernel;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
NeilBrown6796bf52006-12-08 02:36:16 -08001018 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 if (!bdev->bd_openers) {
1020 bdev->bd_disk = disk;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001021 bdev->bd_part = part;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001023 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 struct backing_dev_info *bdi;
1025 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001026 ret = disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001028 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030 if (!bdev->bd_openers) {
1031 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1032 bdi = blk_get_backing_dev_info(bdev);
1033 if (bdi == NULL)
1034 bdi = &default_backing_dev_info;
1035 bdev->bd_inode->i_data.backing_dev_info = bdi;
1036 }
1037 if (bdev->bd_invalidated)
1038 rescan_partitions(disk, bdev);
1039 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 struct block_device *whole;
1041 whole = bdget_disk(disk, 0);
1042 ret = -ENOMEM;
1043 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001044 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001045 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001046 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001048 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 bdev->bd_inode->i_data.backing_dev_info =
1051 whole->bd_inode->i_data.backing_dev_info;
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001052 if (!(disk->flags & GENHD_FL_UP) ||
1053 !part || !part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001055 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 }
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001057 bd_set_size(bdev, (loff_t)part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 }
1059 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001060 disk_put_part(part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001062 module_put(disk->fops->owner);
1063 part = NULL;
1064 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (bdev->bd_contains == bdev) {
1066 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001067 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001069 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
1071 if (bdev->bd_invalidated)
1072 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 }
1074 }
1075 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001076 if (for_part)
1077 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001078 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 unlock_kernel();
1080 return 0;
1081
Tejun Heo0762b8b2008-08-25 19:56:12 +09001082 out_clear:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001084 bdev->bd_part = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1086 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001087 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001089 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001090 mutex_unlock(&bdev->bd_mutex);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001091 out_unlock_kernel:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 unlock_kernel();
Tejun Heo0762b8b2008-08-25 19:56:12 +09001093
1094 disk_put_part(part);
1095 if (disk)
1096 module_put(disk->fops->owner);
1097 put_disk(disk);
1098 bdput(bdev);
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 return ret;
1101}
1102
Al Viro572c4892007-10-08 13:24:05 -04001103int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104{
Al Viro572c4892007-10-08 13:24:05 -04001105 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001106}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107EXPORT_SYMBOL(blkdev_get);
1108
1109static int blkdev_open(struct inode * inode, struct file * filp)
1110{
1111 struct block_device *bdev;
1112 int res;
1113
1114 /*
1115 * Preserve backwards compatibility and allow large file access
1116 * even if userspace doesn't ask for it explicitly. Some mkfs
1117 * binary needs it. We might want to drop this workaround
1118 * during an unstable branch.
1119 */
1120 filp->f_flags |= O_LARGEFILE;
1121
Al Viro572c4892007-10-08 13:24:05 -04001122 if (filp->f_flags & O_NDELAY)
1123 filp->f_mode |= FMODE_NDELAY;
1124 if (filp->f_flags & O_EXCL)
1125 filp->f_mode |= FMODE_EXCL;
1126 if ((filp->f_flags & O_ACCMODE) == 3)
1127 filp->f_mode |= FMODE_WRITE_IOCTL;
1128
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001130 if (bdev == NULL)
1131 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132
Al Viro572c4892007-10-08 13:24:05 -04001133 filp->f_mapping = bdev->bd_inode->i_mapping;
1134
1135 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (res)
1137 return res;
1138
Al Viro572c4892007-10-08 13:24:05 -04001139 if (!(filp->f_mode & FMODE_EXCL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return 0;
1141
1142 if (!(res = bd_claim(bdev, filp)))
1143 return 0;
1144
Al Viro9a1c3542008-02-22 20:40:24 -05001145 blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 return res;
1147}
1148
Al Viro9a1c3542008-02-22 20:40:24 -05001149static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001150{
1151 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001152 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001153 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001154
NeilBrown6796bf52006-12-08 02:36:16 -08001155 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001156 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001157 if (for_part)
1158 bdev->bd_part_count--;
1159
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001160 if (!--bdev->bd_openers) {
1161 sync_blockdev(bdev);
1162 kill_bdev(bdev);
1163 }
1164 if (bdev->bd_contains == bdev) {
1165 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001166 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001167 }
1168 if (!bdev->bd_openers) {
1169 struct module *owner = disk->fops->owner;
1170
1171 put_disk(disk);
1172 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001173 disk_put_part(bdev->bd_part);
1174 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001175 bdev->bd_disk = NULL;
1176 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001177 if (bdev != bdev->bd_contains)
1178 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001179 bdev->bd_contains = NULL;
1180 }
1181 unlock_kernel();
1182 mutex_unlock(&bdev->bd_mutex);
1183 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001184 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001185 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001186 return ret;
1187}
1188
Al Viro9a1c3542008-02-22 20:40:24 -05001189int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001190{
Al Viro9a1c3542008-02-22 20:40:24 -05001191 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001192}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001193EXPORT_SYMBOL(blkdev_put);
1194
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195static int blkdev_close(struct inode * inode, struct file * filp)
1196{
1197 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1198 if (bdev->bd_holder == filp)
1199 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001200 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201}
1202
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001203static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Al Viro56b26ad2008-09-19 03:17:36 -04001205 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1206 fmode_t mode = file->f_mode;
1207 if (file->f_flags & O_NDELAY)
1208 mode |= FMODE_NDELAY_NOW;
1209 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210}
1211
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001212static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 .readpage = blkdev_readpage,
1214 .writepage = blkdev_writepage,
1215 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001216 .write_begin = blkdev_write_begin,
1217 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 .writepages = generic_writepages,
1219 .direct_IO = blkdev_direct_IO,
1220};
1221
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001222const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 .open = blkdev_open,
1224 .release = blkdev_close,
1225 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001226 .read = do_sync_read,
1227 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001229 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 .mmap = generic_file_mmap,
1231 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001232 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233#ifdef CONFIG_COMPAT
1234 .compat_ioctl = compat_blkdev_ioctl,
1235#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001236 .splice_read = generic_file_splice_read,
1237 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238};
1239
1240int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1241{
1242 int res;
1243 mm_segment_t old_fs = get_fs();
1244 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001245 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 set_fs(old_fs);
1247 return res;
1248}
1249
1250EXPORT_SYMBOL(ioctl_by_bdev);
1251
1252/**
1253 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap496aa8a2008-10-16 07:46:23 +02001254 * @path: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001256 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 * namespace if possible and return it. Return ERR_PTR(error)
1258 * otherwise.
1259 */
Al Viro421748e2008-08-02 01:04:36 -04001260struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261{
1262 struct block_device *bdev;
1263 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001264 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 int error;
1266
Al Viro421748e2008-08-02 01:04:36 -04001267 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 return ERR_PTR(-EINVAL);
1269
Al Viro421748e2008-08-02 01:04:36 -04001270 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (error)
1272 return ERR_PTR(error);
1273
Al Viro421748e2008-08-02 01:04:36 -04001274 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 error = -ENOTBLK;
1276 if (!S_ISBLK(inode->i_mode))
1277 goto fail;
1278 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001279 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 goto fail;
1281 error = -ENOMEM;
1282 bdev = bd_acquire(inode);
1283 if (!bdev)
1284 goto fail;
1285out:
Al Viro421748e2008-08-02 01:04:36 -04001286 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 return bdev;
1288fail:
1289 bdev = ERR_PTR(error);
1290 goto out;
1291}
Al Virod5686b42008-08-01 05:00:11 -04001292EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
1294/**
Al Viro30c40d22008-02-22 19:50:45 -05001295 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 *
1297 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001298 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 * @holder: owner for exclusion
1300 *
1301 * Open the blockdevice described by the special file at @path, claim it
1302 * for the @holder.
1303 */
Al Viro30c40d22008-02-22 19:50:45 -05001304struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
1306 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 int error = 0;
1308
1309 bdev = lookup_bdev(path);
1310 if (IS_ERR(bdev))
1311 return bdev;
1312
Al Viro572c4892007-10-08 13:24:05 -04001313 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (error)
1315 return ERR_PTR(error);
1316 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001317 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 goto blkdev_put;
1319 error = bd_claim(bdev, holder);
1320 if (error)
1321 goto blkdev_put;
1322
1323 return bdev;
1324
1325blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001326 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 return ERR_PTR(error);
1328}
1329
Al Viro30c40d22008-02-22 19:50:45 -05001330EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331
1332/**
Al Viro30c40d22008-02-22 19:50:45 -05001333 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 *
1335 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001336 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 *
Al Viro30c40d22008-02-22 19:50:45 -05001338 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
Al Viro30c40d22008-02-22 19:50:45 -05001340void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001343 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
Al Viro30c40d22008-02-22 19:50:45 -05001346EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001347
1348int __invalidate_device(struct block_device *bdev)
1349{
1350 struct super_block *sb = get_super(bdev);
1351 int res = 0;
1352
1353 if (sb) {
1354 /*
1355 * no need to lock the super, get_super holds the
1356 * read mutex so the filesystem cannot go away
1357 * under us (->put_super runs with the write lock
1358 * hold).
1359 */
1360 shrink_dcache_sb(sb);
1361 res = invalidate_inodes(sb);
1362 drop_super(sb);
1363 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001364 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001365 return res;
1366}
1367EXPORT_SYMBOL(__invalidate_device);