blob: f5958f413bd1135787e2dc731f92b562c452d390 [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
303static struct super_block *bd_get_sb(struct file_system_type *fs_type,
304 int flags, const char *dev_name, void *data)
305{
306 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
307}
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;
417 spin_lock(&bdev_lock);
418 bdev = inode->i_bdev;
419 if (bdev && igrab(bdev->bd_inode)) {
420 spin_unlock(&bdev_lock);
421 return bdev;
422 }
423 spin_unlock(&bdev_lock);
424 bdev = bdget(inode->i_rdev);
425 if (bdev) {
426 spin_lock(&bdev_lock);
427 if (inode->i_bdev)
428 __bd_forget(inode);
429 inode->i_bdev = bdev;
430 inode->i_mapping = bdev->bd_inode->i_mapping;
431 list_add(&inode->i_devices, &bdev->bd_inodes);
432 spin_unlock(&bdev_lock);
433 }
434 return bdev;
435}
436
437/* Call when you free inode */
438
439void bd_forget(struct inode *inode)
440{
441 spin_lock(&bdev_lock);
442 if (inode->i_bdev)
443 __bd_forget(inode);
444 spin_unlock(&bdev_lock);
445}
446
447int bd_claim(struct block_device *bdev, void *holder)
448{
449 int res;
450 spin_lock(&bdev_lock);
451
452 /* first decide result */
453 if (bdev->bd_holder == holder)
454 res = 0; /* already a holder */
455 else if (bdev->bd_holder != NULL)
456 res = -EBUSY; /* held by someone else */
457 else if (bdev->bd_contains == bdev)
458 res = 0; /* is a whole device which isn't held */
459
460 else if (bdev->bd_contains->bd_holder == bd_claim)
461 res = 0; /* is a partition of a device that is being partitioned */
462 else if (bdev->bd_contains->bd_holder != NULL)
463 res = -EBUSY; /* is a partition of a held device */
464 else
465 res = 0; /* is a partition of an un-held device */
466
467 /* now impose change */
468 if (res==0) {
469 /* note that for a whole device bd_holders
470 * will be incremented twice, and bd_holder will
471 * be set to bd_claim before being set to holder
472 */
473 bdev->bd_contains->bd_holders ++;
474 bdev->bd_contains->bd_holder = bd_claim;
475 bdev->bd_holders++;
476 bdev->bd_holder = holder;
477 }
478 spin_unlock(&bdev_lock);
479 return res;
480}
481
482EXPORT_SYMBOL(bd_claim);
483
484void bd_release(struct block_device *bdev)
485{
486 spin_lock(&bdev_lock);
487 if (!--bdev->bd_contains->bd_holders)
488 bdev->bd_contains->bd_holder = NULL;
489 if (!--bdev->bd_holders)
490 bdev->bd_holder = NULL;
491 spin_unlock(&bdev_lock);
492}
493
494EXPORT_SYMBOL(bd_release);
495
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800496#ifdef CONFIG_SYSFS
497/*
498 * Functions for bd_claim_by_kobject / bd_release_from_kobject
499 *
500 * If a kobject is passed to bd_claim_by_kobject()
501 * and the kobject has a parent directory,
502 * following symlinks are created:
503 * o from the kobject to the claimed bdev
504 * o from "holders" directory of the bdev to the parent of the kobject
505 * bd_release_from_kobject() removes these symlinks.
506 *
507 * Example:
508 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
509 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
510 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
511 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
512 */
513
514static struct kobject *bdev_get_kobj(struct block_device *bdev)
515{
516 if (bdev->bd_contains != bdev)
517 return kobject_get(&bdev->bd_part->kobj);
518 else
519 return kobject_get(&bdev->bd_disk->kobj);
520}
521
522static struct kobject *bdev_get_holder(struct block_device *bdev)
523{
524 if (bdev->bd_contains != bdev)
525 return kobject_get(bdev->bd_part->holder_dir);
526 else
527 return kobject_get(bdev->bd_disk->holder_dir);
528}
529
530static void add_symlink(struct kobject *from, struct kobject *to)
531{
532 if (!from || !to)
533 return;
534 sysfs_create_link(from, to, kobject_name(to));
535}
536
537static void del_symlink(struct kobject *from, struct kobject *to)
538{
539 if (!from || !to)
540 return;
541 sysfs_remove_link(from, kobject_name(to));
542}
543
544/*
545 * 'struct bd_holder' contains pointers to kobjects symlinked by
546 * bd_claim_by_kobject.
547 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
548 */
549struct bd_holder {
550 struct list_head list; /* chain of holders of the bdev */
551 int count; /* references from the holder */
552 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
553 struct kobject *hdev; /* e.g. "/block/dm-0" */
554 struct kobject *hdir; /* e.g. "/block/sda/holders" */
555 struct kobject *sdev; /* e.g. "/block/sda" */
556};
557
558/*
559 * Get references of related kobjects at once.
560 * Returns 1 on success. 0 on failure.
561 *
562 * Should call bd_holder_release_dirs() after successful use.
563 */
564static int bd_holder_grab_dirs(struct block_device *bdev,
565 struct bd_holder *bo)
566{
567 if (!bdev || !bo)
568 return 0;
569
570 bo->sdir = kobject_get(bo->sdir);
571 if (!bo->sdir)
572 return 0;
573
574 bo->hdev = kobject_get(bo->sdir->parent);
575 if (!bo->hdev)
576 goto fail_put_sdir;
577
578 bo->sdev = bdev_get_kobj(bdev);
579 if (!bo->sdev)
580 goto fail_put_hdev;
581
582 bo->hdir = bdev_get_holder(bdev);
583 if (!bo->hdir)
584 goto fail_put_sdev;
585
586 return 1;
587
588fail_put_sdev:
589 kobject_put(bo->sdev);
590fail_put_hdev:
591 kobject_put(bo->hdev);
592fail_put_sdir:
593 kobject_put(bo->sdir);
594
595 return 0;
596}
597
598/* Put references of related kobjects at once. */
599static void bd_holder_release_dirs(struct bd_holder *bo)
600{
601 kobject_put(bo->hdir);
602 kobject_put(bo->sdev);
603 kobject_put(bo->hdev);
604 kobject_put(bo->sdir);
605}
606
607static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
608{
609 struct bd_holder *bo;
610
611 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
612 if (!bo)
613 return NULL;
614
615 bo->count = 1;
616 bo->sdir = kobj;
617
618 return bo;
619}
620
621static void free_bd_holder(struct bd_holder *bo)
622{
623 kfree(bo);
624}
625
626/**
627 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
628 *
629 * @bdev: block device to be bd_claimed
630 * @bo: preallocated and initialized by alloc_bd_holder()
631 *
632 * If there is no matching entry with @bo in @bdev->bd_holder_list,
633 * add @bo to the list, create symlinks.
634 *
635 * Returns 1 if @bo was added to the list.
636 * Returns 0 if @bo wasn't used by any reason and should be freed.
637 */
638static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
639{
640 struct bd_holder *tmp;
641
642 if (!bo)
643 return 0;
644
645 list_for_each_entry(tmp, &bdev->bd_holder_list, list) {
646 if (tmp->sdir == bo->sdir) {
647 tmp->count++;
648 return 0;
649 }
650 }
651
652 if (!bd_holder_grab_dirs(bdev, bo))
653 return 0;
654
655 add_symlink(bo->sdir, bo->sdev);
656 add_symlink(bo->hdir, bo->hdev);
657 list_add_tail(&bo->list, &bdev->bd_holder_list);
658 return 1;
659}
660
661/**
662 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
663 *
664 * @bdev: block device to be bd_claimed
665 * @kobj: holder's kobject
666 *
667 * If there is matching entry with @kobj in @bdev->bd_holder_list
668 * and no other bd_claim() from the same kobject,
669 * remove the struct bd_holder from the list, delete symlinks for it.
670 *
671 * Returns a pointer to the struct bd_holder when it's removed from the list
672 * and ready to be freed.
673 * Returns NULL if matching claim isn't found or there is other bd_claim()
674 * by the same kobject.
675 */
676static struct bd_holder *del_bd_holder(struct block_device *bdev,
677 struct kobject *kobj)
678{
679 struct bd_holder *bo;
680
681 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
682 if (bo->sdir == kobj) {
683 bo->count--;
684 BUG_ON(bo->count < 0);
685 if (!bo->count) {
686 list_del(&bo->list);
687 del_symlink(bo->sdir, bo->sdev);
688 del_symlink(bo->hdir, bo->hdev);
689 bd_holder_release_dirs(bo);
690 return bo;
691 }
692 break;
693 }
694 }
695
696 return NULL;
697}
698
699/**
700 * bd_claim_by_kobject - bd_claim() with additional kobject signature
701 *
702 * @bdev: block device to be claimed
703 * @holder: holder's signature
704 * @kobj: holder's kobject
705 *
706 * Do bd_claim() and if it succeeds, create sysfs symlinks between
707 * the bdev and the holder's kobject.
708 * Use bd_release_from_kobject() when relesing the claimed bdev.
709 *
710 * Returns 0 on success. (same as bd_claim())
711 * Returns errno on failure.
712 */
713static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
714 struct kobject *kobj)
715{
716 int res;
717 struct bd_holder *bo;
718
719 if (!kobj)
720 return -EINVAL;
721
722 bo = alloc_bd_holder(kobj);
723 if (!bo)
724 return -ENOMEM;
725
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800726 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800727 res = bd_claim(bdev, holder);
728 if (res || !add_bd_holder(bdev, bo))
729 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800730 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800731
732 return res;
733}
734
735/**
736 * bd_release_from_kobject - bd_release() with additional kobject signature
737 *
738 * @bdev: block device to be released
739 * @kobj: holder's kobject
740 *
741 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
742 */
743static void bd_release_from_kobject(struct block_device *bdev,
744 struct kobject *kobj)
745{
746 struct bd_holder *bo;
747
748 if (!kobj)
749 return;
750
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800751 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800752 bd_release(bdev);
753 if ((bo = del_bd_holder(bdev, kobj)))
754 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800755 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800756}
757
758/**
759 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
760 *
761 * @bdev: block device to be claimed
762 * @holder: holder's signature
763 * @disk: holder's gendisk
764 *
765 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
766 */
767int bd_claim_by_disk(struct block_device *bdev, void *holder,
768 struct gendisk *disk)
769{
770 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
771}
772EXPORT_SYMBOL_GPL(bd_claim_by_disk);
773
774/**
775 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
776 *
777 * @bdev: block device to be claimed
778 * @disk: holder's gendisk
779 *
780 * Call bd_release_from_kobject() and put @disk->slave_dir.
781 */
782void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
783{
784 bd_release_from_kobject(bdev, disk->slave_dir);
785 kobject_put(disk->slave_dir);
786}
787EXPORT_SYMBOL_GPL(bd_release_from_disk);
788#endif
789
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790/*
791 * Tries to open block device by device number. Use it ONLY if you
792 * really do not have anything better - i.e. when you are behind a
793 * truly sucky interface and all you are given is a device number. _Never_
794 * to be used for internal purposes. If you ever need it - reconsider
795 * your API.
796 */
797struct block_device *open_by_devnum(dev_t dev, unsigned mode)
798{
799 struct block_device *bdev = bdget(dev);
800 int err = -ENOMEM;
801 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
802 if (bdev)
803 err = blkdev_get(bdev, mode, flags);
804 return err ? ERR_PTR(err) : bdev;
805}
806
807EXPORT_SYMBOL(open_by_devnum);
808
809/*
810 * This routine checks whether a removable media has been changed,
811 * and invalidates all buffer-cache-entries in that case. This
812 * is a relatively slow routine, so we have to try to minimize using
813 * it. Thus it is called only upon a 'mount' or 'open'. This
814 * is the best way of combining speed and utility, I think.
815 * People changing diskettes in the middle of an operation deserve
816 * to lose :-)
817 */
818int check_disk_change(struct block_device *bdev)
819{
820 struct gendisk *disk = bdev->bd_disk;
821 struct block_device_operations * bdops = disk->fops;
822
823 if (!bdops->media_changed)
824 return 0;
825 if (!bdops->media_changed(bdev->bd_disk))
826 return 0;
827
Christoph Hellwig2ef41632005-05-05 16:15:59 -0700828 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 printk("VFS: busy inodes on changed media.\n");
830
831 if (bdops->revalidate_disk)
832 bdops->revalidate_disk(bdev->bd_disk);
833 if (bdev->bd_disk->minors > 1)
834 bdev->bd_invalidated = 1;
835 return 1;
836}
837
838EXPORT_SYMBOL(check_disk_change);
839
840void bd_set_size(struct block_device *bdev, loff_t size)
841{
842 unsigned bsize = bdev_hardsect_size(bdev);
843
844 bdev->bd_inode->i_size = size;
845 while (bsize < PAGE_CACHE_SIZE) {
846 if (size & bsize)
847 break;
848 bsize <<= 1;
849 }
850 bdev->bd_block_size = bsize;
851 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
852}
853EXPORT_SYMBOL(bd_set_size);
854
855static int do_open(struct block_device *bdev, struct file *file)
856{
857 struct module *owner = NULL;
858 struct gendisk *disk;
859 int ret = -ENXIO;
860 int part;
861
862 file->f_mapping = bdev->bd_inode->i_mapping;
863 lock_kernel();
864 disk = get_gendisk(bdev->bd_dev, &part);
865 if (!disk) {
866 unlock_kernel();
867 bdput(bdev);
868 return ret;
869 }
870 owner = disk->fops->owner;
871
Arjan van de Venc039e312006-03-23 03:00:28 -0800872 mutex_lock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (!bdev->bd_openers) {
874 bdev->bd_disk = disk;
875 bdev->bd_contains = bdev;
876 if (!part) {
877 struct backing_dev_info *bdi;
878 if (disk->fops->open) {
879 ret = disk->fops->open(bdev->bd_inode, file);
880 if (ret)
881 goto out_first;
882 }
883 if (!bdev->bd_openers) {
884 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
885 bdi = blk_get_backing_dev_info(bdev);
886 if (bdi == NULL)
887 bdi = &default_backing_dev_info;
888 bdev->bd_inode->i_data.backing_dev_info = bdi;
889 }
890 if (bdev->bd_invalidated)
891 rescan_partitions(disk, bdev);
892 } else {
893 struct hd_struct *p;
894 struct block_device *whole;
895 whole = bdget_disk(disk, 0);
896 ret = -ENOMEM;
897 if (!whole)
898 goto out_first;
899 ret = blkdev_get(whole, file->f_mode, file->f_flags);
900 if (ret)
901 goto out_first;
902 bdev->bd_contains = whole;
Arjan van de Venc039e312006-03-23 03:00:28 -0800903 mutex_lock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 whole->bd_part_count++;
905 p = disk->part[part - 1];
906 bdev->bd_inode->i_data.backing_dev_info =
907 whole->bd_inode->i_data.backing_dev_info;
908 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
909 whole->bd_part_count--;
Arjan van de Venc039e312006-03-23 03:00:28 -0800910 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 ret = -ENXIO;
912 goto out_first;
913 }
914 kobject_get(&p->kobj);
915 bdev->bd_part = p;
916 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Arjan van de Venc039e312006-03-23 03:00:28 -0800917 mutex_unlock(&whole->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 }
919 } else {
920 put_disk(disk);
921 module_put(owner);
922 if (bdev->bd_contains == bdev) {
923 if (bdev->bd_disk->fops->open) {
924 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
925 if (ret)
926 goto out;
927 }
928 if (bdev->bd_invalidated)
929 rescan_partitions(bdev->bd_disk, bdev);
930 } else {
Arjan van de Venc039e312006-03-23 03:00:28 -0800931 mutex_lock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 bdev->bd_contains->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -0800933 mutex_unlock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 }
935 }
936 bdev->bd_openers++;
Arjan van de Venc039e312006-03-23 03:00:28 -0800937 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 unlock_kernel();
939 return 0;
940
941out_first:
942 bdev->bd_disk = NULL;
943 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
944 if (bdev != bdev->bd_contains)
945 blkdev_put(bdev->bd_contains);
946 bdev->bd_contains = NULL;
947 put_disk(disk);
948 module_put(owner);
949out:
Arjan van de Venc039e312006-03-23 03:00:28 -0800950 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 unlock_kernel();
952 if (ret)
953 bdput(bdev);
954 return ret;
955}
956
957int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
958{
959 /*
960 * This crockload is due to bad choice of ->open() type.
961 * It will go away.
962 * For now, block device ->open() routine must _not_
963 * examine anything in 'inode' argument except ->i_rdev.
964 */
965 struct file fake_file = {};
966 struct dentry fake_dentry = {};
967 fake_file.f_mode = mode;
968 fake_file.f_flags = flags;
969 fake_file.f_dentry = &fake_dentry;
970 fake_dentry.d_inode = bdev->bd_inode;
971
972 return do_open(bdev, &fake_file);
973}
974
975EXPORT_SYMBOL(blkdev_get);
976
977static int blkdev_open(struct inode * inode, struct file * filp)
978{
979 struct block_device *bdev;
980 int res;
981
982 /*
983 * Preserve backwards compatibility and allow large file access
984 * even if userspace doesn't ask for it explicitly. Some mkfs
985 * binary needs it. We might want to drop this workaround
986 * during an unstable branch.
987 */
988 filp->f_flags |= O_LARGEFILE;
989
990 bdev = bd_acquire(inode);
991
992 res = do_open(bdev, filp);
993 if (res)
994 return res;
995
996 if (!(filp->f_flags & O_EXCL) )
997 return 0;
998
999 if (!(res = bd_claim(bdev, filp)))
1000 return 0;
1001
1002 blkdev_put(bdev);
1003 return res;
1004}
1005
1006int blkdev_put(struct block_device *bdev)
1007{
1008 int ret = 0;
1009 struct inode *bd_inode = bdev->bd_inode;
1010 struct gendisk *disk = bdev->bd_disk;
1011
Arjan van de Venc039e312006-03-23 03:00:28 -08001012 mutex_lock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 lock_kernel();
1014 if (!--bdev->bd_openers) {
1015 sync_blockdev(bdev);
1016 kill_bdev(bdev);
1017 }
1018 if (bdev->bd_contains == bdev) {
1019 if (disk->fops->release)
1020 ret = disk->fops->release(bd_inode, NULL);
1021 } else {
Arjan van de Venc039e312006-03-23 03:00:28 -08001022 mutex_lock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 bdev->bd_contains->bd_part_count--;
Arjan van de Venc039e312006-03-23 03:00:28 -08001024 mutex_unlock(&bdev->bd_contains->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
1026 if (!bdev->bd_openers) {
1027 struct module *owner = disk->fops->owner;
1028
1029 put_disk(disk);
1030 module_put(owner);
1031
1032 if (bdev->bd_contains != bdev) {
1033 kobject_put(&bdev->bd_part->kobj);
1034 bdev->bd_part = NULL;
1035 }
1036 bdev->bd_disk = NULL;
1037 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1038 if (bdev != bdev->bd_contains) {
1039 blkdev_put(bdev->bd_contains);
1040 }
1041 bdev->bd_contains = NULL;
1042 }
1043 unlock_kernel();
Arjan van de Venc039e312006-03-23 03:00:28 -08001044 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 bdput(bdev);
1046 return ret;
1047}
1048
1049EXPORT_SYMBOL(blkdev_put);
1050
1051static int blkdev_close(struct inode * inode, struct file * filp)
1052{
1053 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1054 if (bdev->bd_holder == filp)
1055 bd_release(bdev);
1056 return blkdev_put(bdev);
1057}
1058
1059static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
1060 size_t count, loff_t *ppos)
1061{
1062 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1063
1064 return generic_file_write_nolock(file, &local_iov, 1, ppos);
1065}
1066
1067static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
1068 size_t count, loff_t pos)
1069{
1070 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1071
1072 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
1073}
1074
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001075static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076{
1077 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1078}
1079
1080struct address_space_operations def_blk_aops = {
1081 .readpage = blkdev_readpage,
1082 .writepage = blkdev_writepage,
1083 .sync_page = block_sync_page,
1084 .prepare_write = blkdev_prepare_write,
1085 .commit_write = blkdev_commit_write,
1086 .writepages = generic_writepages,
1087 .direct_IO = blkdev_direct_IO,
1088};
1089
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001090const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 .open = blkdev_open,
1092 .release = blkdev_close,
1093 .llseek = block_llseek,
1094 .read = generic_file_read,
1095 .write = blkdev_file_write,
1096 .aio_read = generic_file_aio_read,
1097 .aio_write = blkdev_file_aio_write,
1098 .mmap = generic_file_mmap,
1099 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001100 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101#ifdef CONFIG_COMPAT
1102 .compat_ioctl = compat_blkdev_ioctl,
1103#endif
1104 .readv = generic_file_readv,
1105 .writev = generic_file_write_nolock,
1106 .sendfile = generic_file_sendfile,
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001107 .splice_read = generic_file_splice_read,
1108 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109};
1110
1111int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1112{
1113 int res;
1114 mm_segment_t old_fs = get_fs();
1115 set_fs(KERNEL_DS);
1116 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1117 set_fs(old_fs);
1118 return res;
1119}
1120
1121EXPORT_SYMBOL(ioctl_by_bdev);
1122
1123/**
1124 * lookup_bdev - lookup a struct block_device by name
1125 *
1126 * @path: special file representing the block device
1127 *
1128 * Get a reference to the blockdevice at @path in the current
1129 * namespace if possible and return it. Return ERR_PTR(error)
1130 * otherwise.
1131 */
1132struct block_device *lookup_bdev(const char *path)
1133{
1134 struct block_device *bdev;
1135 struct inode *inode;
1136 struct nameidata nd;
1137 int error;
1138
1139 if (!path || !*path)
1140 return ERR_PTR(-EINVAL);
1141
1142 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1143 if (error)
1144 return ERR_PTR(error);
1145
1146 inode = nd.dentry->d_inode;
1147 error = -ENOTBLK;
1148 if (!S_ISBLK(inode->i_mode))
1149 goto fail;
1150 error = -EACCES;
1151 if (nd.mnt->mnt_flags & MNT_NODEV)
1152 goto fail;
1153 error = -ENOMEM;
1154 bdev = bd_acquire(inode);
1155 if (!bdev)
1156 goto fail;
1157out:
1158 path_release(&nd);
1159 return bdev;
1160fail:
1161 bdev = ERR_PTR(error);
1162 goto out;
1163}
1164
1165/**
1166 * open_bdev_excl - open a block device by name and set it up for use
1167 *
1168 * @path: special file representing the block device
1169 * @flags: %MS_RDONLY for opening read-only
1170 * @holder: owner for exclusion
1171 *
1172 * Open the blockdevice described by the special file at @path, claim it
1173 * for the @holder.
1174 */
1175struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1176{
1177 struct block_device *bdev;
1178 mode_t mode = FMODE_READ;
1179 int error = 0;
1180
1181 bdev = lookup_bdev(path);
1182 if (IS_ERR(bdev))
1183 return bdev;
1184
1185 if (!(flags & MS_RDONLY))
1186 mode |= FMODE_WRITE;
1187 error = blkdev_get(bdev, mode, 0);
1188 if (error)
1189 return ERR_PTR(error);
1190 error = -EACCES;
1191 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1192 goto blkdev_put;
1193 error = bd_claim(bdev, holder);
1194 if (error)
1195 goto blkdev_put;
1196
1197 return bdev;
1198
1199blkdev_put:
1200 blkdev_put(bdev);
1201 return ERR_PTR(error);
1202}
1203
1204EXPORT_SYMBOL(open_bdev_excl);
1205
1206/**
1207 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1208 *
1209 * @bdev: blockdevice to close
1210 *
1211 * This is the counterpart to open_bdev_excl().
1212 */
1213void close_bdev_excl(struct block_device *bdev)
1214{
1215 bd_release(bdev);
1216 blkdev_put(bdev);
1217}
1218
1219EXPORT_SYMBOL(close_bdev_excl);