blob: 335c38bb86eb1ed97f06e0edd02492e77edd8a9b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/smp_lock.h>
15#include <linux/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/blkpg.h>
19#include <linux/buffer_head.h>
David Howells811d7362006-08-29 19:06:09 +010020#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mpage.h>
22#include <linux/mount.h>
23#include <linux/uio.h>
24#include <linux/namei.h>
25#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020026#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28struct bdev_inode {
29 struct block_device bdev;
30 struct inode vfs_inode;
31};
32
33static inline struct bdev_inode *BDEV_I(struct inode *inode)
34{
35 return container_of(inode, struct bdev_inode, vfs_inode);
36}
37
38inline struct block_device *I_BDEV(struct inode *inode)
39{
40 return &BDEV_I(inode)->bdev;
41}
42
43EXPORT_SYMBOL(I_BDEV);
44
45static sector_t max_block(struct block_device *bdev)
46{
47 sector_t retval = ~((sector_t)0);
48 loff_t sz = i_size_read(bdev->bd_inode);
49
50 if (sz) {
51 unsigned int size = block_size(bdev);
52 unsigned int sizebits = blksize_bits(size);
53 retval = (sz >> sizebits);
54 }
55 return retval;
56}
57
58/* Kill _all_ buffers, dirty or not.. */
59static void kill_bdev(struct block_device *bdev)
60{
61 invalidate_bdev(bdev, 1);
62 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
63}
64
65int set_blocksize(struct block_device *bdev, int size)
66{
67 /* Size must be a power of two, and between 512 and PAGE_SIZE */
68 if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
69 return -EINVAL;
70
71 /* Size cannot be smaller than the size supported by the device */
72 if (size < bdev_hardsect_size(bdev))
73 return -EINVAL;
74
75 /* Don't change the size if it is same as current */
76 if (bdev->bd_block_size != size) {
77 sync_blockdev(bdev);
78 bdev->bd_block_size = size;
79 bdev->bd_inode->i_blkbits = blksize_bits(size);
80 kill_bdev(bdev);
81 }
82 return 0;
83}
84
85EXPORT_SYMBOL(set_blocksize);
86
87int sb_set_blocksize(struct super_block *sb, int size)
88{
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 if (set_blocksize(sb->s_bdev, size))
90 return 0;
91 /* If we get here, we know size is power of two
92 * and it's value is between 512 and PAGE_SIZE */
93 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -080094 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 return sb->s_blocksize;
96}
97
98EXPORT_SYMBOL(sb_set_blocksize);
99
100int sb_min_blocksize(struct super_block *sb, int size)
101{
102 int minsize = bdev_hardsect_size(sb->s_bdev);
103 if (size < minsize)
104 size = minsize;
105 return sb_set_blocksize(sb, size);
106}
107
108EXPORT_SYMBOL(sb_min_blocksize);
109
110static int
111blkdev_get_block(struct inode *inode, sector_t iblock,
112 struct buffer_head *bh, int create)
113{
114 if (iblock >= max_block(I_BDEV(inode))) {
115 if (create)
116 return -EIO;
117
118 /*
119 * for reads, we're just trying to fill a partial page.
120 * return a hole, they will have to call get_block again
121 * before they can fill it, and they will get -EIO at that
122 * time
123 */
124 return 0;
125 }
126 bh->b_bdev = I_BDEV(inode);
127 bh->b_blocknr = iblock;
128 set_buffer_mapped(bh);
129 return 0;
130}
131
132static int
133blkdev_get_blocks(struct inode *inode, sector_t iblock,
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800134 struct buffer_head *bh, int create)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135{
136 sector_t end_block = max_block(I_BDEV(inode));
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -0800137 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
139 if ((iblock + max_blocks) > end_block) {
140 max_blocks = end_block - iblock;
141 if ((long)max_blocks <= 0) {
142 if (create)
143 return -EIO; /* write fully beyond EOF */
144 /*
145 * It is a read which is fully beyond EOF. We return
146 * a !buffer_mapped buffer
147 */
148 max_blocks = 0;
149 }
150 }
151
152 bh->b_bdev = I_BDEV(inode);
153 bh->b_blocknr = iblock;
154 bh->b_size = max_blocks << inode->i_blkbits;
155 if (max_blocks)
156 set_buffer_mapped(bh);
157 return 0;
158}
159
160static ssize_t
161blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
162 loff_t offset, unsigned long nr_segs)
163{
164 struct file *file = iocb->ki_filp;
165 struct inode *inode = file->f_mapping->host;
166
167 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
168 iov, offset, nr_segs, blkdev_get_blocks, NULL);
169}
170
171static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
172{
173 return block_write_full_page(page, blkdev_get_block, wbc);
174}
175
176static int blkdev_readpage(struct file * file, struct page * page)
177{
178 return block_read_full_page(page, blkdev_get_block);
179}
180
181static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
182{
183 return block_prepare_write(page, from, to, blkdev_get_block);
184}
185
186static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
187{
188 return block_commit_write(page, from, to);
189}
190
191/*
192 * private llseek:
193 * for a block special file file->f_dentry->d_inode->i_size is zero
194 * so we compute the size by hand (just as in block_read/write above)
195 */
196static loff_t block_llseek(struct file *file, loff_t offset, int origin)
197{
198 struct inode *bd_inode = file->f_mapping->host;
199 loff_t size;
200 loff_t retval;
201
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800202 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 size = i_size_read(bd_inode);
204
205 switch (origin) {
206 case 2:
207 offset += size;
208 break;
209 case 1:
210 offset += file->f_pos;
211 }
212 retval = -EINVAL;
213 if (offset >= 0 && offset <= size) {
214 if (offset != file->f_pos) {
215 file->f_pos = offset;
216 }
217 retval = offset;
218 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800219 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 return retval;
221}
222
223/*
224 * Filp is never NULL; the only case when ->fsync() is called with
225 * NULL first argument is nfsd_sync_dir() and that's not a directory.
226 */
227
228static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
229{
230 return sync_blockdev(I_BDEV(filp->f_mapping->host));
231}
232
233/*
234 * pseudo-fs
235 */
236
237static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800238static kmem_cache_t * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240static struct inode *bdev_alloc_inode(struct super_block *sb)
241{
242 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
243 if (!ei)
244 return NULL;
245 return &ei->vfs_inode;
246}
247
248static void bdev_destroy_inode(struct inode *inode)
249{
250 struct bdev_inode *bdi = BDEV_I(inode);
251
252 bdi->bdev.bd_inode_backing_dev_info = NULL;
253 kmem_cache_free(bdev_cachep, bdi);
254}
255
256static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
257{
258 struct bdev_inode *ei = (struct bdev_inode *) foo;
259 struct block_device *bdev = &ei->bdev;
260
261 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
262 SLAB_CTOR_CONSTRUCTOR)
263 {
264 memset(bdev, 0, sizeof(*bdev));
Arjan van de Venc039e312006-03-23 03:00:28 -0800265 mutex_init(&bdev->bd_mutex);
266 mutex_init(&bdev->bd_mount_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 INIT_LIST_HEAD(&bdev->bd_inodes);
268 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800269#ifdef CONFIG_SYSFS
270 INIT_LIST_HEAD(&bdev->bd_holder_list);
271#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 inode_init_once(&ei->vfs_inode);
273 }
274}
275
276static inline void __bd_forget(struct inode *inode)
277{
278 list_del_init(&inode->i_devices);
279 inode->i_bdev = NULL;
280 inode->i_mapping = &inode->i_data;
281}
282
283static void bdev_clear_inode(struct inode *inode)
284{
285 struct block_device *bdev = &BDEV_I(inode)->bdev;
286 struct list_head *p;
287 spin_lock(&bdev_lock);
288 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
289 __bd_forget(list_entry(p, struct inode, i_devices));
290 }
291 list_del_init(&bdev->bd_list);
292 spin_unlock(&bdev_lock);
293}
294
295static struct super_operations bdev_sops = {
296 .statfs = simple_statfs,
297 .alloc_inode = bdev_alloc_inode,
298 .destroy_inode = bdev_destroy_inode,
299 .drop_inode = generic_delete_inode,
300 .clear_inode = bdev_clear_inode,
301};
302
David Howells454e2392006-06-23 02:02:57 -0700303static int bd_get_sb(struct file_system_type *fs_type,
304 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
David Howells454e2392006-06-23 02:02:57 -0700306 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
309static struct file_system_type bd_type = {
310 .name = "bdev",
311 .get_sb = bd_get_sb,
312 .kill_sb = kill_anon_super,
313};
314
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800315static struct vfsmount *bd_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316struct super_block *blockdev_superblock;
317
318void __init bdev_cache_init(void)
319{
320 int err;
321 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800322 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
323 SLAB_MEM_SPREAD|SLAB_PANIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 init_once, NULL);
325 err = register_filesystem(&bd_type);
326 if (err)
327 panic("Cannot register bdev pseudo-fs");
328 bd_mnt = kern_mount(&bd_type);
329 err = PTR_ERR(bd_mnt);
330 if (IS_ERR(bd_mnt))
331 panic("Cannot create bdev pseudo-fs");
332 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
333}
334
335/*
336 * Most likely _very_ bad one - but then it's hardly critical for small
337 * /dev and can be fixed when somebody will need really large one.
338 * Keep in mind that it will be fed through icache hash function too.
339 */
340static inline unsigned long hash(dev_t dev)
341{
342 return MAJOR(dev)+MINOR(dev);
343}
344
345static int bdev_test(struct inode *inode, void *data)
346{
347 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
348}
349
350static int bdev_set(struct inode *inode, void *data)
351{
352 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
353 return 0;
354}
355
356static LIST_HEAD(all_bdevs);
357
358struct block_device *bdget(dev_t dev)
359{
360 struct block_device *bdev;
361 struct inode *inode;
362
363 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
364 bdev_test, bdev_set, &dev);
365
366 if (!inode)
367 return NULL;
368
369 bdev = &BDEV_I(inode)->bdev;
370
371 if (inode->i_state & I_NEW) {
372 bdev->bd_contains = NULL;
373 bdev->bd_inode = inode;
374 bdev->bd_block_size = (1 << inode->i_blkbits);
375 bdev->bd_part_count = 0;
376 bdev->bd_invalidated = 0;
377 inode->i_mode = S_IFBLK;
378 inode->i_rdev = dev;
379 inode->i_bdev = bdev;
380 inode->i_data.a_ops = &def_blk_aops;
381 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
382 inode->i_data.backing_dev_info = &default_backing_dev_info;
383 spin_lock(&bdev_lock);
384 list_add(&bdev->bd_list, &all_bdevs);
385 spin_unlock(&bdev_lock);
386 unlock_new_inode(inode);
387 }
388 return bdev;
389}
390
391EXPORT_SYMBOL(bdget);
392
393long nr_blockdev_pages(void)
394{
395 struct list_head *p;
396 long ret = 0;
397 spin_lock(&bdev_lock);
398 list_for_each(p, &all_bdevs) {
399 struct block_device *bdev;
400 bdev = list_entry(p, struct block_device, bd_list);
401 ret += bdev->bd_inode->i_mapping->nrpages;
402 }
403 spin_unlock(&bdev_lock);
404 return ret;
405}
406
407void bdput(struct block_device *bdev)
408{
409 iput(bdev->bd_inode);
410}
411
412EXPORT_SYMBOL(bdput);
413
414static struct block_device *bd_acquire(struct inode *inode)
415{
416 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 spin_lock(&bdev_lock);
419 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700420 if (bdev) {
421 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 spin_unlock(&bdev_lock);
423 return bdev;
424 }
425 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 bdev = bdget(inode->i_rdev);
428 if (bdev) {
429 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700430 if (!inode->i_bdev) {
431 /*
432 * We take an additional bd_inode->i_count for inode,
433 * and it's released in clear_inode() of inode.
434 * So, we can access it via ->i_mapping always
435 * without igrab().
436 */
437 atomic_inc(&bdev->bd_inode->i_count);
438 inode->i_bdev = bdev;
439 inode->i_mapping = bdev->bd_inode->i_mapping;
440 list_add(&inode->i_devices, &bdev->bd_inodes);
441 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 spin_unlock(&bdev_lock);
443 }
444 return bdev;
445}
446
447/* Call when you free inode */
448
449void bd_forget(struct inode *inode)
450{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700451 struct block_device *bdev = NULL;
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700454 if (inode->i_bdev) {
455 if (inode->i_sb != blockdev_superblock)
456 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700458 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700460
461 if (bdev)
462 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
465int bd_claim(struct block_device *bdev, void *holder)
466{
467 int res;
468 spin_lock(&bdev_lock);
469
470 /* first decide result */
471 if (bdev->bd_holder == holder)
472 res = 0; /* already a holder */
473 else if (bdev->bd_holder != NULL)
474 res = -EBUSY; /* held by someone else */
475 else if (bdev->bd_contains == bdev)
476 res = 0; /* is a whole device which isn't held */
477
478 else if (bdev->bd_contains->bd_holder == bd_claim)
479 res = 0; /* is a partition of a device that is being partitioned */
480 else if (bdev->bd_contains->bd_holder != NULL)
481 res = -EBUSY; /* is a partition of a held device */
482 else
483 res = 0; /* is a partition of an un-held device */
484
485 /* now impose change */
486 if (res==0) {
487 /* note that for a whole device bd_holders
488 * will be incremented twice, and bd_holder will
489 * be set to bd_claim before being set to holder
490 */
491 bdev->bd_contains->bd_holders ++;
492 bdev->bd_contains->bd_holder = bd_claim;
493 bdev->bd_holders++;
494 bdev->bd_holder = holder;
495 }
496 spin_unlock(&bdev_lock);
497 return res;
498}
499
500EXPORT_SYMBOL(bd_claim);
501
502void bd_release(struct block_device *bdev)
503{
504 spin_lock(&bdev_lock);
505 if (!--bdev->bd_contains->bd_holders)
506 bdev->bd_contains->bd_holder = NULL;
507 if (!--bdev->bd_holders)
508 bdev->bd_holder = NULL;
509 spin_unlock(&bdev_lock);
510}
511
512EXPORT_SYMBOL(bd_release);
513
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800514#ifdef CONFIG_SYSFS
515/*
516 * Functions for bd_claim_by_kobject / bd_release_from_kobject
517 *
518 * If a kobject is passed to bd_claim_by_kobject()
519 * and the kobject has a parent directory,
520 * following symlinks are created:
521 * o from the kobject to the claimed bdev
522 * o from "holders" directory of the bdev to the parent of the kobject
523 * bd_release_from_kobject() removes these symlinks.
524 *
525 * Example:
526 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
527 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
528 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
529 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
530 */
531
532static struct kobject *bdev_get_kobj(struct block_device *bdev)
533{
534 if (bdev->bd_contains != bdev)
535 return kobject_get(&bdev->bd_part->kobj);
536 else
537 return kobject_get(&bdev->bd_disk->kobj);
538}
539
540static struct kobject *bdev_get_holder(struct block_device *bdev)
541{
542 if (bdev->bd_contains != bdev)
543 return kobject_get(bdev->bd_part->holder_dir);
544 else
545 return kobject_get(bdev->bd_disk->holder_dir);
546}
547
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700548static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800549{
550 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700551 return 0;
552 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800553}
554
555static void del_symlink(struct kobject *from, struct kobject *to)
556{
557 if (!from || !to)
558 return;
559 sysfs_remove_link(from, kobject_name(to));
560}
561
562/*
563 * 'struct bd_holder' contains pointers to kobjects symlinked by
564 * bd_claim_by_kobject.
565 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
566 */
567struct bd_holder {
568 struct list_head list; /* chain of holders of the bdev */
569 int count; /* references from the holder */
570 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
571 struct kobject *hdev; /* e.g. "/block/dm-0" */
572 struct kobject *hdir; /* e.g. "/block/sda/holders" */
573 struct kobject *sdev; /* e.g. "/block/sda" */
574};
575
576/*
577 * Get references of related kobjects at once.
578 * Returns 1 on success. 0 on failure.
579 *
580 * Should call bd_holder_release_dirs() after successful use.
581 */
582static int bd_holder_grab_dirs(struct block_device *bdev,
583 struct bd_holder *bo)
584{
585 if (!bdev || !bo)
586 return 0;
587
588 bo->sdir = kobject_get(bo->sdir);
589 if (!bo->sdir)
590 return 0;
591
592 bo->hdev = kobject_get(bo->sdir->parent);
593 if (!bo->hdev)
594 goto fail_put_sdir;
595
596 bo->sdev = bdev_get_kobj(bdev);
597 if (!bo->sdev)
598 goto fail_put_hdev;
599
600 bo->hdir = bdev_get_holder(bdev);
601 if (!bo->hdir)
602 goto fail_put_sdev;
603
604 return 1;
605
606fail_put_sdev:
607 kobject_put(bo->sdev);
608fail_put_hdev:
609 kobject_put(bo->hdev);
610fail_put_sdir:
611 kobject_put(bo->sdir);
612
613 return 0;
614}
615
616/* Put references of related kobjects at once. */
617static void bd_holder_release_dirs(struct bd_holder *bo)
618{
619 kobject_put(bo->hdir);
620 kobject_put(bo->sdev);
621 kobject_put(bo->hdev);
622 kobject_put(bo->sdir);
623}
624
625static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
626{
627 struct bd_holder *bo;
628
629 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
630 if (!bo)
631 return NULL;
632
633 bo->count = 1;
634 bo->sdir = kobj;
635
636 return bo;
637}
638
639static void free_bd_holder(struct bd_holder *bo)
640{
641 kfree(bo);
642}
643
644/**
645 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
646 *
647 * @bdev: block device to be bd_claimed
648 * @bo: preallocated and initialized by alloc_bd_holder()
649 *
650 * If there is no matching entry with @bo in @bdev->bd_holder_list,
651 * add @bo to the list, create symlinks.
652 *
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700653 * Returns 0 if symlinks are created or already there.
654 * Returns -ve if something fails and @bo can be freed.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800655 */
656static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
657{
658 struct bd_holder *tmp;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700659 int ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800660
661 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700662 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800663
664 list_for_each_entry(tmp, &bdev->bd_holder_list, list) {
665 if (tmp->sdir == bo->sdir) {
666 tmp->count++;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700667 /* We've already done what we need to do here. */
668 free_bd_holder(bo);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800669 return 0;
670 }
671 }
672
673 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700674 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800675
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700676 ret = add_symlink(bo->sdir, bo->sdev);
677 if (ret == 0) {
678 ret = add_symlink(bo->hdir, bo->hdev);
679 if (ret)
680 del_symlink(bo->sdir, bo->sdev);
681 }
682 if (ret == 0)
683 list_add_tail(&bo->list, &bdev->bd_holder_list);
684 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800685}
686
687/**
688 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
689 *
690 * @bdev: block device to be bd_claimed
691 * @kobj: holder's kobject
692 *
693 * If there is matching entry with @kobj in @bdev->bd_holder_list
694 * and no other bd_claim() from the same kobject,
695 * remove the struct bd_holder from the list, delete symlinks for it.
696 *
697 * Returns a pointer to the struct bd_holder when it's removed from the list
698 * and ready to be freed.
699 * Returns NULL if matching claim isn't found or there is other bd_claim()
700 * by the same kobject.
701 */
702static struct bd_holder *del_bd_holder(struct block_device *bdev,
703 struct kobject *kobj)
704{
705 struct bd_holder *bo;
706
707 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
708 if (bo->sdir == kobj) {
709 bo->count--;
710 BUG_ON(bo->count < 0);
711 if (!bo->count) {
712 list_del(&bo->list);
713 del_symlink(bo->sdir, bo->sdev);
714 del_symlink(bo->hdir, bo->hdev);
715 bd_holder_release_dirs(bo);
716 return bo;
717 }
718 break;
719 }
720 }
721
722 return NULL;
723}
724
725/**
726 * bd_claim_by_kobject - bd_claim() with additional kobject signature
727 *
728 * @bdev: block device to be claimed
729 * @holder: holder's signature
730 * @kobj: holder's kobject
731 *
732 * Do bd_claim() and if it succeeds, create sysfs symlinks between
733 * the bdev and the holder's kobject.
734 * Use bd_release_from_kobject() when relesing the claimed bdev.
735 *
736 * Returns 0 on success. (same as bd_claim())
737 * Returns errno on failure.
738 */
739static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
740 struct kobject *kobj)
741{
742 int res;
743 struct bd_holder *bo;
744
745 if (!kobj)
746 return -EINVAL;
747
748 bo = alloc_bd_holder(kobj);
749 if (!bo)
750 return -ENOMEM;
751
Ingo Molnar663d4402006-07-03 00:25:33 -0700752 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800753 res = bd_claim(bdev, holder);
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700754 if (res == 0)
755 res = add_bd_holder(bdev, bo);
756 if (res)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800757 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800758 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800759
760 return res;
761}
762
763/**
764 * bd_release_from_kobject - bd_release() with additional kobject signature
765 *
766 * @bdev: block device to be released
767 * @kobj: holder's kobject
768 *
769 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
770 */
771static void bd_release_from_kobject(struct block_device *bdev,
772 struct kobject *kobj)
773{
774 struct bd_holder *bo;
775
776 if (!kobj)
777 return;
778
Ingo Molnar663d4402006-07-03 00:25:33 -0700779 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800780 bd_release(bdev);
781 if ((bo = del_bd_holder(bdev, kobj)))
782 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800783 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800784}
785
786/**
787 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
788 *
789 * @bdev: block device to be claimed
790 * @holder: holder's signature
791 * @disk: holder's gendisk
792 *
793 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
794 */
795int bd_claim_by_disk(struct block_device *bdev, void *holder,
796 struct gendisk *disk)
797{
798 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
799}
800EXPORT_SYMBOL_GPL(bd_claim_by_disk);
801
802/**
803 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
804 *
805 * @bdev: block device to be claimed
806 * @disk: holder's gendisk
807 *
808 * Call bd_release_from_kobject() and put @disk->slave_dir.
809 */
810void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
811{
812 bd_release_from_kobject(bdev, disk->slave_dir);
813 kobject_put(disk->slave_dir);
814}
815EXPORT_SYMBOL_GPL(bd_release_from_disk);
816#endif
817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818/*
819 * Tries to open block device by device number. Use it ONLY if you
820 * really do not have anything better - i.e. when you are behind a
821 * truly sucky interface and all you are given is a device number. _Never_
822 * to be used for internal purposes. If you ever need it - reconsider
823 * your API.
824 */
825struct block_device *open_by_devnum(dev_t dev, unsigned mode)
826{
827 struct block_device *bdev = bdget(dev);
828 int err = -ENOMEM;
829 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
830 if (bdev)
831 err = blkdev_get(bdev, mode, flags);
832 return err ? ERR_PTR(err) : bdev;
833}
834
835EXPORT_SYMBOL(open_by_devnum);
836
Ingo Molnar663d4402006-07-03 00:25:33 -0700837static int
838blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags);
839
840struct block_device *open_partition_by_devnum(dev_t dev, unsigned mode)
841{
842 struct block_device *bdev = bdget(dev);
843 int err = -ENOMEM;
844 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
845 if (bdev)
846 err = blkdev_get_partition(bdev, mode, flags);
847 return err ? ERR_PTR(err) : bdev;
848}
849
850EXPORT_SYMBOL(open_partition_by_devnum);
851
852
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853/*
854 * This routine checks whether a removable media has been changed,
855 * and invalidates all buffer-cache-entries in that case. This
856 * is a relatively slow routine, so we have to try to minimize using
857 * it. Thus it is called only upon a 'mount' or 'open'. This
858 * is the best way of combining speed and utility, I think.
859 * People changing diskettes in the middle of an operation deserve
860 * to lose :-)
861 */
862int check_disk_change(struct block_device *bdev)
863{
864 struct gendisk *disk = bdev->bd_disk;
865 struct block_device_operations * bdops = disk->fops;
866
867 if (!bdops->media_changed)
868 return 0;
869 if (!bdops->media_changed(bdev->bd_disk))
870 return 0;
871
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700872 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 printk("VFS: busy inodes on changed media.\n");
874
875 if (bdops->revalidate_disk)
876 bdops->revalidate_disk(bdev->bd_disk);
877 if (bdev->bd_disk->minors > 1)
878 bdev->bd_invalidated = 1;
879 return 1;
880}
881
882EXPORT_SYMBOL(check_disk_change);
883
884void bd_set_size(struct block_device *bdev, loff_t size)
885{
886 unsigned bsize = bdev_hardsect_size(bdev);
887
888 bdev->bd_inode->i_size = size;
889 while (bsize < PAGE_CACHE_SIZE) {
890 if (size & bsize)
891 break;
892 bsize <<= 1;
893 }
894 bdev->bd_block_size = bsize;
895 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
896}
897EXPORT_SYMBOL(bd_set_size);
898
Peter Zijlstra6946bd62006-08-27 01:23:31 -0700899static int __blkdev_put(struct block_device *bdev, unsigned int subclass)
900{
901 int ret = 0;
902 struct inode *bd_inode = bdev->bd_inode;
903 struct gendisk *disk = bdev->bd_disk;
904
905 mutex_lock_nested(&bdev->bd_mutex, subclass);
906 lock_kernel();
907 if (!--bdev->bd_openers) {
908 sync_blockdev(bdev);
909 kill_bdev(bdev);
910 }
911 if (bdev->bd_contains == bdev) {
912 if (disk->fops->release)
913 ret = disk->fops->release(bd_inode, NULL);
914 } else {
915 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
916 subclass + 1);
917 bdev->bd_contains->bd_part_count--;
918 mutex_unlock(&bdev->bd_contains->bd_mutex);
919 }
920 if (!bdev->bd_openers) {
921 struct module *owner = disk->fops->owner;
922
923 put_disk(disk);
924 module_put(owner);
925
926 if (bdev->bd_contains != bdev) {
927 kobject_put(&bdev->bd_part->kobj);
928 bdev->bd_part = NULL;
929 }
930 bdev->bd_disk = NULL;
931 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
932 if (bdev != bdev->bd_contains)
933 __blkdev_put(bdev->bd_contains, subclass + 1);
934 bdev->bd_contains = NULL;
935 }
936 unlock_kernel();
937 mutex_unlock(&bdev->bd_mutex);
938 bdput(bdev);
939 return ret;
940}
941
942int blkdev_put(struct block_device *bdev)
943{
944 return __blkdev_put(bdev, BD_MUTEX_NORMAL);
945}
946EXPORT_SYMBOL(blkdev_put);
947
948int blkdev_put_partition(struct block_device *bdev)
949{
950 return __blkdev_put(bdev, BD_MUTEX_PARTITION);
951}
952EXPORT_SYMBOL(blkdev_put_partition);
953
Ingo Molnar663d4402006-07-03 00:25:33 -0700954static int
955blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags);
956
957static int
958do_open(struct block_device *bdev, struct file *file, unsigned int subclass)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959{
960 struct module *owner = NULL;
961 struct gendisk *disk;
962 int ret = -ENXIO;
963 int part;
964
965 file->f_mapping = bdev->bd_inode->i_mapping;
966 lock_kernel();
967 disk = get_gendisk(bdev->bd_dev, &part);
968 if (!disk) {
969 unlock_kernel();
970 bdput(bdev);
971 return ret;
972 }
973 owner = disk->fops->owner;
974
Ingo Molnar663d4402006-07-03 00:25:33 -0700975 mutex_lock_nested(&bdev->bd_mutex, subclass);
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (!bdev->bd_openers) {
978 bdev->bd_disk = disk;
979 bdev->bd_contains = bdev;
980 if (!part) {
981 struct backing_dev_info *bdi;
982 if (disk->fops->open) {
983 ret = disk->fops->open(bdev->bd_inode, file);
984 if (ret)
985 goto out_first;
986 }
987 if (!bdev->bd_openers) {
988 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
989 bdi = blk_get_backing_dev_info(bdev);
990 if (bdi == NULL)
991 bdi = &default_backing_dev_info;
992 bdev->bd_inode->i_data.backing_dev_info = bdi;
993 }
994 if (bdev->bd_invalidated)
995 rescan_partitions(disk, bdev);
996 } else {
997 struct hd_struct *p;
998 struct block_device *whole;
999 whole = bdget_disk(disk, 0);
1000 ret = -ENOMEM;
1001 if (!whole)
1002 goto out_first;
Ingo Molnar663d4402006-07-03 00:25:33 -07001003 ret = blkdev_get_whole(whole, file->f_mode, file->f_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (ret)
1005 goto out_first;
1006 bdev->bd_contains = whole;
Ingo Molnar663d4402006-07-03 00:25:33 -07001007 mutex_lock_nested(&whole->bd_mutex, BD_MUTEX_WHOLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 whole->bd_part_count++;
1009 p = disk->part[part - 1];
1010 bdev->bd_inode->i_data.backing_dev_info =
1011 whole->bd_inode->i_data.backing_dev_info;
1012 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1013 whole->bd_part_count--;
Arjan van de Venc039e312006-03-23 03:00:28 -08001014 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 ret = -ENXIO;
1016 goto out_first;
1017 }
1018 kobject_get(&p->kobj);
1019 bdev->bd_part = p;
1020 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Arjan van de Venc039e312006-03-23 03:00:28 -08001021 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 }
1023 } else {
1024 put_disk(disk);
1025 module_put(owner);
1026 if (bdev->bd_contains == bdev) {
1027 if (bdev->bd_disk->fops->open) {
1028 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1029 if (ret)
1030 goto out;
1031 }
1032 if (bdev->bd_invalidated)
1033 rescan_partitions(bdev->bd_disk, bdev);
1034 } else {
Ingo Molnar663d4402006-07-03 00:25:33 -07001035 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
Jason Baron87d7c8a2006-09-29 02:01:01 -07001036 BD_MUTEX_WHOLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 bdev->bd_contains->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001038 mutex_unlock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 }
1040 }
1041 bdev->bd_openers++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001042 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 unlock_kernel();
1044 return 0;
1045
1046out_first:
1047 bdev->bd_disk = NULL;
1048 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1049 if (bdev != bdev->bd_contains)
Peter Zijlstra6946bd62006-08-27 01:23:31 -07001050 __blkdev_put(bdev->bd_contains, BD_MUTEX_WHOLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 bdev->bd_contains = NULL;
1052 put_disk(disk);
1053 module_put(owner);
1054out:
Arjan van de Venc039e312006-03-23 03:00:28 -08001055 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 unlock_kernel();
1057 if (ret)
1058 bdput(bdev);
1059 return ret;
1060}
1061
1062int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1063{
1064 /*
1065 * This crockload is due to bad choice of ->open() type.
1066 * It will go away.
1067 * For now, block device ->open() routine must _not_
1068 * examine anything in 'inode' argument except ->i_rdev.
1069 */
1070 struct file fake_file = {};
1071 struct dentry fake_dentry = {};
1072 fake_file.f_mode = mode;
1073 fake_file.f_flags = flags;
1074 fake_file.f_dentry = &fake_dentry;
1075 fake_dentry.d_inode = bdev->bd_inode;
1076
Ingo Molnar663d4402006-07-03 00:25:33 -07001077 return do_open(bdev, &fake_file, BD_MUTEX_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
1080EXPORT_SYMBOL(blkdev_get);
1081
Ingo Molnar663d4402006-07-03 00:25:33 -07001082static int
1083blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags)
1084{
1085 /*
1086 * This crockload is due to bad choice of ->open() type.
1087 * It will go away.
1088 * For now, block device ->open() routine must _not_
1089 * examine anything in 'inode' argument except ->i_rdev.
1090 */
1091 struct file fake_file = {};
1092 struct dentry fake_dentry = {};
1093 fake_file.f_mode = mode;
1094 fake_file.f_flags = flags;
1095 fake_file.f_dentry = &fake_dentry;
1096 fake_dentry.d_inode = bdev->bd_inode;
1097
1098 return do_open(bdev, &fake_file, BD_MUTEX_WHOLE);
1099}
1100
1101static int
1102blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags)
1103{
1104 /*
1105 * This crockload is due to bad choice of ->open() type.
1106 * It will go away.
1107 * For now, block device ->open() routine must _not_
1108 * examine anything in 'inode' argument except ->i_rdev.
1109 */
1110 struct file fake_file = {};
1111 struct dentry fake_dentry = {};
1112 fake_file.f_mode = mode;
1113 fake_file.f_flags = flags;
1114 fake_file.f_dentry = &fake_dentry;
1115 fake_dentry.d_inode = bdev->bd_inode;
1116
1117 return do_open(bdev, &fake_file, BD_MUTEX_PARTITION);
1118}
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120static int blkdev_open(struct inode * inode, struct file * filp)
1121{
1122 struct block_device *bdev;
1123 int res;
1124
1125 /*
1126 * Preserve backwards compatibility and allow large file access
1127 * even if userspace doesn't ask for it explicitly. Some mkfs
1128 * binary needs it. We might want to drop this workaround
1129 * during an unstable branch.
1130 */
1131 filp->f_flags |= O_LARGEFILE;
1132
1133 bdev = bd_acquire(inode);
1134
Ingo Molnar663d4402006-07-03 00:25:33 -07001135 res = do_open(bdev, filp, BD_MUTEX_NORMAL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 if (res)
1137 return res;
1138
1139 if (!(filp->f_flags & O_EXCL) )
1140 return 0;
1141
1142 if (!(res = bd_claim(bdev, filp)))
1143 return 0;
1144
1145 blkdev_put(bdev);
1146 return res;
1147}
1148
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149static int blkdev_close(struct inode * inode, struct file * filp)
1150{
1151 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1152 if (bdev->bd_holder == filp)
1153 bd_release(bdev);
1154 return blkdev_put(bdev);
1155}
1156
1157static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
1158 size_t count, loff_t *ppos)
1159{
1160 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1161
1162 return generic_file_write_nolock(file, &local_iov, 1, ppos);
1163}
1164
1165static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
1166 size_t count, loff_t pos)
1167{
1168 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1169
1170 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
1171}
1172
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001173static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174{
1175 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1176}
1177
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001178const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 .readpage = blkdev_readpage,
1180 .writepage = blkdev_writepage,
1181 .sync_page = block_sync_page,
1182 .prepare_write = blkdev_prepare_write,
1183 .commit_write = blkdev_commit_write,
1184 .writepages = generic_writepages,
1185 .direct_IO = blkdev_direct_IO,
1186};
1187
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001188const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 .open = blkdev_open,
1190 .release = blkdev_close,
1191 .llseek = block_llseek,
1192 .read = generic_file_read,
1193 .write = blkdev_file_write,
1194 .aio_read = generic_file_aio_read,
1195 .aio_write = blkdev_file_aio_write,
1196 .mmap = generic_file_mmap,
1197 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001198 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199#ifdef CONFIG_COMPAT
1200 .compat_ioctl = compat_blkdev_ioctl,
1201#endif
1202 .readv = generic_file_readv,
1203 .writev = generic_file_write_nolock,
1204 .sendfile = generic_file_sendfile,
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001205 .splice_read = generic_file_splice_read,
1206 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207};
1208
1209int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1210{
1211 int res;
1212 mm_segment_t old_fs = get_fs();
1213 set_fs(KERNEL_DS);
1214 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1215 set_fs(old_fs);
1216 return res;
1217}
1218
1219EXPORT_SYMBOL(ioctl_by_bdev);
1220
1221/**
1222 * lookup_bdev - lookup a struct block_device by name
1223 *
1224 * @path: special file representing the block device
1225 *
1226 * Get a reference to the blockdevice at @path in the current
1227 * namespace if possible and return it. Return ERR_PTR(error)
1228 * otherwise.
1229 */
1230struct block_device *lookup_bdev(const char *path)
1231{
1232 struct block_device *bdev;
1233 struct inode *inode;
1234 struct nameidata nd;
1235 int error;
1236
1237 if (!path || !*path)
1238 return ERR_PTR(-EINVAL);
1239
1240 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1241 if (error)
1242 return ERR_PTR(error);
1243
1244 inode = nd.dentry->d_inode;
1245 error = -ENOTBLK;
1246 if (!S_ISBLK(inode->i_mode))
1247 goto fail;
1248 error = -EACCES;
1249 if (nd.mnt->mnt_flags & MNT_NODEV)
1250 goto fail;
1251 error = -ENOMEM;
1252 bdev = bd_acquire(inode);
1253 if (!bdev)
1254 goto fail;
1255out:
1256 path_release(&nd);
1257 return bdev;
1258fail:
1259 bdev = ERR_PTR(error);
1260 goto out;
1261}
1262
1263/**
1264 * open_bdev_excl - open a block device by name and set it up for use
1265 *
1266 * @path: special file representing the block device
1267 * @flags: %MS_RDONLY for opening read-only
1268 * @holder: owner for exclusion
1269 *
1270 * Open the blockdevice described by the special file at @path, claim it
1271 * for the @holder.
1272 */
1273struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1274{
1275 struct block_device *bdev;
1276 mode_t mode = FMODE_READ;
1277 int error = 0;
1278
1279 bdev = lookup_bdev(path);
1280 if (IS_ERR(bdev))
1281 return bdev;
1282
1283 if (!(flags & MS_RDONLY))
1284 mode |= FMODE_WRITE;
1285 error = blkdev_get(bdev, mode, 0);
1286 if (error)
1287 return ERR_PTR(error);
1288 error = -EACCES;
1289 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1290 goto blkdev_put;
1291 error = bd_claim(bdev, holder);
1292 if (error)
1293 goto blkdev_put;
1294
1295 return bdev;
1296
1297blkdev_put:
1298 blkdev_put(bdev);
1299 return ERR_PTR(error);
1300}
1301
1302EXPORT_SYMBOL(open_bdev_excl);
1303
1304/**
1305 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1306 *
1307 * @bdev: blockdevice to close
1308 *
1309 * This is the counterpart to open_bdev_excl().
1310 */
1311void close_bdev_excl(struct block_device *bdev)
1312{
1313 bd_release(bdev);
1314 blkdev_put(bdev);
1315}
1316
1317EXPORT_SYMBOL(close_bdev_excl);