blob: 028d9fb9c2d536dafad7563e36874c0e8c934d91 [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
8#include <linux/config.h>
9#include <linux/init.h>
10#include <linux/mm.h>
11#include <linux/fcntl.h>
12#include <linux/slab.h>
13#include <linux/kmod.h>
14#include <linux/major.h>
15#include <linux/devfs_fs_kernel.h>
16#include <linux/smp_lock.h>
17#include <linux/highmem.h>
18#include <linux/blkdev.h>
19#include <linux/module.h>
20#include <linux/blkpg.h>
21#include <linux/buffer_head.h>
22#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
26#include <asm/uaccess.h>
27
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
548static void add_symlink(struct kobject *from, struct kobject *to)
549{
550 if (!from || !to)
551 return;
552 sysfs_create_link(from, to, kobject_name(to));
553}
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 *
653 * Returns 1 if @bo was added to the list.
654 * Returns 0 if @bo wasn't used by any reason and should be freed.
655 */
656static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
657{
658 struct bd_holder *tmp;
659
660 if (!bo)
661 return 0;
662
663 list_for_each_entry(tmp, &bdev->bd_holder_list, list) {
664 if (tmp->sdir == bo->sdir) {
665 tmp->count++;
666 return 0;
667 }
668 }
669
670 if (!bd_holder_grab_dirs(bdev, bo))
671 return 0;
672
673 add_symlink(bo->sdir, bo->sdev);
674 add_symlink(bo->hdir, bo->hdev);
675 list_add_tail(&bo->list, &bdev->bd_holder_list);
676 return 1;
677}
678
679/**
680 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
681 *
682 * @bdev: block device to be bd_claimed
683 * @kobj: holder's kobject
684 *
685 * If there is matching entry with @kobj in @bdev->bd_holder_list
686 * and no other bd_claim() from the same kobject,
687 * remove the struct bd_holder from the list, delete symlinks for it.
688 *
689 * Returns a pointer to the struct bd_holder when it's removed from the list
690 * and ready to be freed.
691 * Returns NULL if matching claim isn't found or there is other bd_claim()
692 * by the same kobject.
693 */
694static struct bd_holder *del_bd_holder(struct block_device *bdev,
695 struct kobject *kobj)
696{
697 struct bd_holder *bo;
698
699 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
700 if (bo->sdir == kobj) {
701 bo->count--;
702 BUG_ON(bo->count < 0);
703 if (!bo->count) {
704 list_del(&bo->list);
705 del_symlink(bo->sdir, bo->sdev);
706 del_symlink(bo->hdir, bo->hdev);
707 bd_holder_release_dirs(bo);
708 return bo;
709 }
710 break;
711 }
712 }
713
714 return NULL;
715}
716
717/**
718 * bd_claim_by_kobject - bd_claim() with additional kobject signature
719 *
720 * @bdev: block device to be claimed
721 * @holder: holder's signature
722 * @kobj: holder's kobject
723 *
724 * Do bd_claim() and if it succeeds, create sysfs symlinks between
725 * the bdev and the holder's kobject.
726 * Use bd_release_from_kobject() when relesing the claimed bdev.
727 *
728 * Returns 0 on success. (same as bd_claim())
729 * Returns errno on failure.
730 */
731static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
732 struct kobject *kobj)
733{
734 int res;
735 struct bd_holder *bo;
736
737 if (!kobj)
738 return -EINVAL;
739
740 bo = alloc_bd_holder(kobj);
741 if (!bo)
742 return -ENOMEM;
743
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800744 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800745 res = bd_claim(bdev, holder);
746 if (res || !add_bd_holder(bdev, bo))
747 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800748 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800749
750 return res;
751}
752
753/**
754 * bd_release_from_kobject - bd_release() with additional kobject signature
755 *
756 * @bdev: block device to be released
757 * @kobj: holder's kobject
758 *
759 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
760 */
761static void bd_release_from_kobject(struct block_device *bdev,
762 struct kobject *kobj)
763{
764 struct bd_holder *bo;
765
766 if (!kobj)
767 return;
768
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800769 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800770 bd_release(bdev);
771 if ((bo = del_bd_holder(bdev, kobj)))
772 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800773 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800774}
775
776/**
777 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
778 *
779 * @bdev: block device to be claimed
780 * @holder: holder's signature
781 * @disk: holder's gendisk
782 *
783 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
784 */
785int bd_claim_by_disk(struct block_device *bdev, void *holder,
786 struct gendisk *disk)
787{
788 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
789}
790EXPORT_SYMBOL_GPL(bd_claim_by_disk);
791
792/**
793 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
794 *
795 * @bdev: block device to be claimed
796 * @disk: holder's gendisk
797 *
798 * Call bd_release_from_kobject() and put @disk->slave_dir.
799 */
800void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
801{
802 bd_release_from_kobject(bdev, disk->slave_dir);
803 kobject_put(disk->slave_dir);
804}
805EXPORT_SYMBOL_GPL(bd_release_from_disk);
806#endif
807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808/*
809 * Tries to open block device by device number. Use it ONLY if you
810 * really do not have anything better - i.e. when you are behind a
811 * truly sucky interface and all you are given is a device number. _Never_
812 * to be used for internal purposes. If you ever need it - reconsider
813 * your API.
814 */
815struct block_device *open_by_devnum(dev_t dev, unsigned mode)
816{
817 struct block_device *bdev = bdget(dev);
818 int err = -ENOMEM;
819 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
820 if (bdev)
821 err = blkdev_get(bdev, mode, flags);
822 return err ? ERR_PTR(err) : bdev;
823}
824
825EXPORT_SYMBOL(open_by_devnum);
826
827/*
828 * This routine checks whether a removable media has been changed,
829 * and invalidates all buffer-cache-entries in that case. This
830 * is a relatively slow routine, so we have to try to minimize using
831 * it. Thus it is called only upon a 'mount' or 'open'. This
832 * is the best way of combining speed and utility, I think.
833 * People changing diskettes in the middle of an operation deserve
834 * to lose :-)
835 */
836int check_disk_change(struct block_device *bdev)
837{
838 struct gendisk *disk = bdev->bd_disk;
839 struct block_device_operations * bdops = disk->fops;
840
841 if (!bdops->media_changed)
842 return 0;
843 if (!bdops->media_changed(bdev->bd_disk))
844 return 0;
845
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700846 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 printk("VFS: busy inodes on changed media.\n");
848
849 if (bdops->revalidate_disk)
850 bdops->revalidate_disk(bdev->bd_disk);
851 if (bdev->bd_disk->minors > 1)
852 bdev->bd_invalidated = 1;
853 return 1;
854}
855
856EXPORT_SYMBOL(check_disk_change);
857
858void bd_set_size(struct block_device *bdev, loff_t size)
859{
860 unsigned bsize = bdev_hardsect_size(bdev);
861
862 bdev->bd_inode->i_size = size;
863 while (bsize < PAGE_CACHE_SIZE) {
864 if (size & bsize)
865 break;
866 bsize <<= 1;
867 }
868 bdev->bd_block_size = bsize;
869 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
870}
871EXPORT_SYMBOL(bd_set_size);
872
873static int do_open(struct block_device *bdev, struct file *file)
874{
875 struct module *owner = NULL;
876 struct gendisk *disk;
877 int ret = -ENXIO;
878 int part;
879
880 file->f_mapping = bdev->bd_inode->i_mapping;
881 lock_kernel();
882 disk = get_gendisk(bdev->bd_dev, &part);
883 if (!disk) {
884 unlock_kernel();
885 bdput(bdev);
886 return ret;
887 }
888 owner = disk->fops->owner;
889
Arjan van de Venc039e312006-03-23 03:00:28 -0800890 mutex_lock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (!bdev->bd_openers) {
892 bdev->bd_disk = disk;
893 bdev->bd_contains = bdev;
894 if (!part) {
895 struct backing_dev_info *bdi;
896 if (disk->fops->open) {
897 ret = disk->fops->open(bdev->bd_inode, file);
898 if (ret)
899 goto out_first;
900 }
901 if (!bdev->bd_openers) {
902 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
903 bdi = blk_get_backing_dev_info(bdev);
904 if (bdi == NULL)
905 bdi = &default_backing_dev_info;
906 bdev->bd_inode->i_data.backing_dev_info = bdi;
907 }
908 if (bdev->bd_invalidated)
909 rescan_partitions(disk, bdev);
910 } else {
911 struct hd_struct *p;
912 struct block_device *whole;
913 whole = bdget_disk(disk, 0);
914 ret = -ENOMEM;
915 if (!whole)
916 goto out_first;
917 ret = blkdev_get(whole, file->f_mode, file->f_flags);
918 if (ret)
919 goto out_first;
920 bdev->bd_contains = whole;
Arjan van de Venc039e312006-03-23 03:00:28 -0800921 mutex_lock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 whole->bd_part_count++;
923 p = disk->part[part - 1];
924 bdev->bd_inode->i_data.backing_dev_info =
925 whole->bd_inode->i_data.backing_dev_info;
926 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
927 whole->bd_part_count--;
Arjan van de Venc039e312006-03-23 03:00:28 -0800928 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 ret = -ENXIO;
930 goto out_first;
931 }
932 kobject_get(&p->kobj);
933 bdev->bd_part = p;
934 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Arjan van de Venc039e312006-03-23 03:00:28 -0800935 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 }
937 } else {
938 put_disk(disk);
939 module_put(owner);
940 if (bdev->bd_contains == bdev) {
941 if (bdev->bd_disk->fops->open) {
942 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
943 if (ret)
944 goto out;
945 }
946 if (bdev->bd_invalidated)
947 rescan_partitions(bdev->bd_disk, bdev);
948 } else {
Arjan van de Venc039e312006-03-23 03:00:28 -0800949 mutex_lock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 bdev->bd_contains->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -0800951 mutex_unlock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 }
953 }
954 bdev->bd_openers++;
Arjan van de Venc039e312006-03-23 03:00:28 -0800955 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 unlock_kernel();
957 return 0;
958
959out_first:
960 bdev->bd_disk = NULL;
961 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
962 if (bdev != bdev->bd_contains)
963 blkdev_put(bdev->bd_contains);
964 bdev->bd_contains = NULL;
965 put_disk(disk);
966 module_put(owner);
967out:
Arjan van de Venc039e312006-03-23 03:00:28 -0800968 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 unlock_kernel();
970 if (ret)
971 bdput(bdev);
972 return ret;
973}
974
975int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
976{
977 /*
978 * This crockload is due to bad choice of ->open() type.
979 * It will go away.
980 * For now, block device ->open() routine must _not_
981 * examine anything in 'inode' argument except ->i_rdev.
982 */
983 struct file fake_file = {};
984 struct dentry fake_dentry = {};
985 fake_file.f_mode = mode;
986 fake_file.f_flags = flags;
987 fake_file.f_dentry = &fake_dentry;
988 fake_dentry.d_inode = bdev->bd_inode;
989
990 return do_open(bdev, &fake_file);
991}
992
993EXPORT_SYMBOL(blkdev_get);
994
995static int blkdev_open(struct inode * inode, struct file * filp)
996{
997 struct block_device *bdev;
998 int res;
999
1000 /*
1001 * Preserve backwards compatibility and allow large file access
1002 * even if userspace doesn't ask for it explicitly. Some mkfs
1003 * binary needs it. We might want to drop this workaround
1004 * during an unstable branch.
1005 */
1006 filp->f_flags |= O_LARGEFILE;
1007
1008 bdev = bd_acquire(inode);
1009
1010 res = do_open(bdev, filp);
1011 if (res)
1012 return res;
1013
1014 if (!(filp->f_flags & O_EXCL) )
1015 return 0;
1016
1017 if (!(res = bd_claim(bdev, filp)))
1018 return 0;
1019
1020 blkdev_put(bdev);
1021 return res;
1022}
1023
1024int blkdev_put(struct block_device *bdev)
1025{
1026 int ret = 0;
1027 struct inode *bd_inode = bdev->bd_inode;
1028 struct gendisk *disk = bdev->bd_disk;
1029
Arjan van de Venc039e312006-03-23 03:00:28 -08001030 mutex_lock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 lock_kernel();
1032 if (!--bdev->bd_openers) {
1033 sync_blockdev(bdev);
1034 kill_bdev(bdev);
1035 }
1036 if (bdev->bd_contains == bdev) {
1037 if (disk->fops->release)
1038 ret = disk->fops->release(bd_inode, NULL);
1039 } else {
Arjan van de Venc039e312006-03-23 03:00:28 -08001040 mutex_lock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 bdev->bd_contains->bd_part_count--;
Arjan van de Venc039e312006-03-23 03:00:28 -08001042 mutex_unlock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 }
1044 if (!bdev->bd_openers) {
1045 struct module *owner = disk->fops->owner;
1046
1047 put_disk(disk);
1048 module_put(owner);
1049
1050 if (bdev->bd_contains != bdev) {
1051 kobject_put(&bdev->bd_part->kobj);
1052 bdev->bd_part = NULL;
1053 }
1054 bdev->bd_disk = NULL;
1055 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1056 if (bdev != bdev->bd_contains) {
1057 blkdev_put(bdev->bd_contains);
1058 }
1059 bdev->bd_contains = NULL;
1060 }
1061 unlock_kernel();
Arjan van de Venc039e312006-03-23 03:00:28 -08001062 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 bdput(bdev);
1064 return ret;
1065}
1066
1067EXPORT_SYMBOL(blkdev_put);
1068
1069static int blkdev_close(struct inode * inode, struct file * filp)
1070{
1071 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1072 if (bdev->bd_holder == filp)
1073 bd_release(bdev);
1074 return blkdev_put(bdev);
1075}
1076
1077static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
1078 size_t count, loff_t *ppos)
1079{
1080 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1081
1082 return generic_file_write_nolock(file, &local_iov, 1, ppos);
1083}
1084
1085static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
1086 size_t count, loff_t pos)
1087{
1088 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1089
1090 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
1091}
1092
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001093static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
1095 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1096}
1097
1098struct address_space_operations def_blk_aops = {
1099 .readpage = blkdev_readpage,
1100 .writepage = blkdev_writepage,
1101 .sync_page = block_sync_page,
1102 .prepare_write = blkdev_prepare_write,
1103 .commit_write = blkdev_commit_write,
1104 .writepages = generic_writepages,
1105 .direct_IO = blkdev_direct_IO,
1106};
1107
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001108const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 .open = blkdev_open,
1110 .release = blkdev_close,
1111 .llseek = block_llseek,
1112 .read = generic_file_read,
1113 .write = blkdev_file_write,
1114 .aio_read = generic_file_aio_read,
1115 .aio_write = blkdev_file_aio_write,
1116 .mmap = generic_file_mmap,
1117 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001118 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119#ifdef CONFIG_COMPAT
1120 .compat_ioctl = compat_blkdev_ioctl,
1121#endif
1122 .readv = generic_file_readv,
1123 .writev = generic_file_write_nolock,
1124 .sendfile = generic_file_sendfile,
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001125 .splice_read = generic_file_splice_read,
1126 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127};
1128
1129int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1130{
1131 int res;
1132 mm_segment_t old_fs = get_fs();
1133 set_fs(KERNEL_DS);
1134 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1135 set_fs(old_fs);
1136 return res;
1137}
1138
1139EXPORT_SYMBOL(ioctl_by_bdev);
1140
1141/**
1142 * lookup_bdev - lookup a struct block_device by name
1143 *
1144 * @path: special file representing the block device
1145 *
1146 * Get a reference to the blockdevice at @path in the current
1147 * namespace if possible and return it. Return ERR_PTR(error)
1148 * otherwise.
1149 */
1150struct block_device *lookup_bdev(const char *path)
1151{
1152 struct block_device *bdev;
1153 struct inode *inode;
1154 struct nameidata nd;
1155 int error;
1156
1157 if (!path || !*path)
1158 return ERR_PTR(-EINVAL);
1159
1160 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1161 if (error)
1162 return ERR_PTR(error);
1163
1164 inode = nd.dentry->d_inode;
1165 error = -ENOTBLK;
1166 if (!S_ISBLK(inode->i_mode))
1167 goto fail;
1168 error = -EACCES;
1169 if (nd.mnt->mnt_flags & MNT_NODEV)
1170 goto fail;
1171 error = -ENOMEM;
1172 bdev = bd_acquire(inode);
1173 if (!bdev)
1174 goto fail;
1175out:
1176 path_release(&nd);
1177 return bdev;
1178fail:
1179 bdev = ERR_PTR(error);
1180 goto out;
1181}
1182
1183/**
1184 * open_bdev_excl - open a block device by name and set it up for use
1185 *
1186 * @path: special file representing the block device
1187 * @flags: %MS_RDONLY for opening read-only
1188 * @holder: owner for exclusion
1189 *
1190 * Open the blockdevice described by the special file at @path, claim it
1191 * for the @holder.
1192 */
1193struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1194{
1195 struct block_device *bdev;
1196 mode_t mode = FMODE_READ;
1197 int error = 0;
1198
1199 bdev = lookup_bdev(path);
1200 if (IS_ERR(bdev))
1201 return bdev;
1202
1203 if (!(flags & MS_RDONLY))
1204 mode |= FMODE_WRITE;
1205 error = blkdev_get(bdev, mode, 0);
1206 if (error)
1207 return ERR_PTR(error);
1208 error = -EACCES;
1209 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1210 goto blkdev_put;
1211 error = bd_claim(bdev, holder);
1212 if (error)
1213 goto blkdev_put;
1214
1215 return bdev;
1216
1217blkdev_put:
1218 blkdev_put(bdev);
1219 return ERR_PTR(error);
1220}
1221
1222EXPORT_SYMBOL(open_bdev_excl);
1223
1224/**
1225 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1226 *
1227 * @bdev: blockdevice to close
1228 *
1229 * This is the counterpart to open_bdev_excl().
1230 */
1231void close_bdev_excl(struct block_device *bdev)
1232{
1233 bd_release(bdev);
1234 blkdev_put(bdev);
1235}
1236
1237EXPORT_SYMBOL(close_bdev_excl);