projects/linux·blob v2.6.26 · refs

1313 lines · 30.6 KiB

1/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE
6 */
7
8#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>
14#include <linux/smp_lock.h>
15#include <linux/device_cgroup.h>
16#include <linux/highmem.h>
17#include <linux/blkdev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/buffer_head.h>
21#include <linux/writeback.h>
22#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
26#include <linux/log2.h>
27#include <asm/uaccess.h>
28#include "internal.h"
29
30struct bdev_inode {
31 struct block_device bdev;
32 struct inode vfs_inode;
33};
34
35static const struct address_space_operations def_blk_aops;
36
37static inline struct bdev_inode *BDEV_I(struct inode *inode)
38{
39 return container_of(inode, struct bdev_inode, vfs_inode);
40}
41
42inline struct block_device *I_BDEV(struct inode *inode)
43{
44 return &BDEV_I(inode)->bdev;
45}
46
47EXPORT_SYMBOL(I_BDEV);
48
49static sector_t max_block(struct block_device *bdev)
50{
51 sector_t retval = ~((sector_t)0);
52 loff_t sz = i_size_read(bdev->bd_inode);
53
54 if (sz) {
55 unsigned int size = block_size(bdev);
56 unsigned int sizebits = blksize_bits(size);
57 retval = (sz >> sizebits);
58 }
59 return retval;
60}
61
62/* Kill _all_ buffers and pagecache , dirty or not.. */
63static void kill_bdev(struct block_device *bdev)
64{
65 if (bdev->bd_inode->i_mapping->nrpages == 0)
66 return;
67 invalidate_bh_lrus();
68 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
69}
70
71int set_blocksize(struct block_device *bdev, int size)
72{
73 /* Size must be a power of two, and between 512 and PAGE_SIZE */
74 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
75 return -EINVAL;
76
77 /* Size cannot be smaller than the size supported by the device */
78 if (size < bdev_hardsect_size(bdev))
79 return -EINVAL;
80
81 /* Don't change the size if it is same as current */
82 if (bdev->bd_block_size != size) {
83 sync_blockdev(bdev);
84 bdev->bd_block_size = size;
85 bdev->bd_inode->i_blkbits = blksize_bits(size);
86 kill_bdev(bdev);
87 }
88 return 0;
89}
90
91EXPORT_SYMBOL(set_blocksize);
92
93int sb_set_blocksize(struct super_block *sb, int size)
94{
95 if (set_blocksize(sb->s_bdev, size))
96 return 0;
97 /* If we get here, we know size is power of two
98 * and it's value is between 512 and PAGE_SIZE */
99 sb->s_blocksize = size;
100 sb->s_blocksize_bits = blksize_bits(size);
101 return sb->s_blocksize;
102}
103
104EXPORT_SYMBOL(sb_set_blocksize);
105
106int sb_min_blocksize(struct super_block *sb, int size)
107{
108 int minsize = bdev_hardsect_size(sb->s_bdev);
109 if (size < minsize)
110 size = minsize;
111 return sb_set_blocksize(sb, size);
112}
113
114EXPORT_SYMBOL(sb_min_blocksize);
115
116static int
117blkdev_get_block(struct inode *inode, sector_t iblock,
118 struct buffer_head *bh, int create)
119{
120 if (iblock >= max_block(I_BDEV(inode))) {
121 if (create)
122 return -EIO;
123
124 /*
125 * for reads, we're just trying to fill a partial page.
126 * return a hole, they will have to call get_block again
127 * before they can fill it, and they will get -EIO at that
128 * time
129 */
130 return 0;
131 }
132 bh->b_bdev = I_BDEV(inode);
133 bh->b_blocknr = iblock;
134 set_buffer_mapped(bh);
135 return 0;
136}
137
138static int
139blkdev_get_blocks(struct inode *inode, sector_t iblock,
140 struct buffer_head *bh, int create)
141{
142 sector_t end_block = max_block(I_BDEV(inode));
143 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
144
145 if ((iblock + max_blocks) > end_block) {
146 max_blocks = end_block - iblock;
147 if ((long)max_blocks <= 0) {
148 if (create)
149 return -EIO; /* write fully beyond EOF */
150 /*
151 * It is a read which is fully beyond EOF. We return
152 * a !buffer_mapped buffer
153 */
154 max_blocks = 0;
155 }
156 }
157
158 bh->b_bdev = I_BDEV(inode);
159 bh->b_blocknr = iblock;
160 bh->b_size = max_blocks << inode->i_blkbits;
161 if (max_blocks)
162 set_buffer_mapped(bh);
163 return 0;
164}
165
166static ssize_t
167blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
168 loff_t offset, unsigned long nr_segs)
169{
170 struct file *file = iocb->ki_filp;
171 struct inode *inode = file->f_mapping->host;
172
173 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
174 iov, offset, nr_segs, blkdev_get_blocks, NULL);
175}
176
177static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
178{
179 return block_write_full_page(page, blkdev_get_block, wbc);
180}
181
182static int blkdev_readpage(struct file * file, struct page * page)
183{
184 return block_read_full_page(page, blkdev_get_block);
185}
186
187static int blkdev_write_begin(struct file *file, struct address_space *mapping,
188 loff_t pos, unsigned len, unsigned flags,
189 struct page **pagep, void **fsdata)
190{
191 *pagep = NULL;
192 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
193 blkdev_get_block);
194}
195
196static int blkdev_write_end(struct file *file, struct address_space *mapping,
197 loff_t pos, unsigned len, unsigned copied,
198 struct page *page, void *fsdata)
199{
200 int ret;
201 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
202
203 unlock_page(page);
204 page_cache_release(page);
205
206 return ret;
207}
208
209/*
210 * private llseek:
211 * for a block special file file->f_path.dentry->d_inode->i_size is zero
212 * so we compute the size by hand (just as in block_read/write above)
213 */
214static loff_t block_llseek(struct file *file, loff_t offset, int origin)
215{
216 struct inode *bd_inode = file->f_mapping->host;
217 loff_t size;
218 loff_t retval;
219
220 mutex_lock(&bd_inode->i_mutex);
221 size = i_size_read(bd_inode);
222
223 switch (origin) {
224 case 2:
225 offset += size;
226 break;
227 case 1:
228 offset += file->f_pos;
229 }
230 retval = -EINVAL;
231 if (offset >= 0 && offset <= size) {
232 if (offset != file->f_pos) {
233 file->f_pos = offset;
234 }
235 retval = offset;
236 }
237 mutex_unlock(&bd_inode->i_mutex);
238 return retval;
239}
240
241/*
242 * Filp is never NULL; the only case when ->fsync() is called with
243 * NULL first argument is nfsd_sync_dir() and that's not a directory.
244 */
245
246static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
247{
248 return sync_blockdev(I_BDEV(filp->f_mapping->host));
249}
250
251/*
252 * pseudo-fs
253 */
254
255static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
256static struct kmem_cache * bdev_cachep __read_mostly;
257
258static struct inode *bdev_alloc_inode(struct super_block *sb)
259{
260 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
261 if (!ei)
262 return NULL;
263 return &ei->vfs_inode;
264}
265
266static void bdev_destroy_inode(struct inode *inode)
267{
268 struct bdev_inode *bdi = BDEV_I(inode);
269
270 bdi->bdev.bd_inode_backing_dev_info = NULL;
271 kmem_cache_free(bdev_cachep, bdi);
272}
273
274static void init_once(struct kmem_cache * cachep, void *foo)
275{
276 struct bdev_inode *ei = (struct bdev_inode *) foo;
277 struct block_device *bdev = &ei->bdev;
278
279 memset(bdev, 0, sizeof(*bdev));
280 mutex_init(&bdev->bd_mutex);
281 sema_init(&bdev->bd_mount_sem, 1);
282 INIT_LIST_HEAD(&bdev->bd_inodes);
283 INIT_LIST_HEAD(&bdev->bd_list);
284#ifdef CONFIG_SYSFS
285 INIT_LIST_HEAD(&bdev->bd_holder_list);
286#endif
287 inode_init_once(&ei->vfs_inode);
288}
289
290static inline void __bd_forget(struct inode *inode)
291{
292 list_del_init(&inode->i_devices);
293 inode->i_bdev = NULL;
294 inode->i_mapping = &inode->i_data;
295}
296
297static void bdev_clear_inode(struct inode *inode)
298{
299 struct block_device *bdev = &BDEV_I(inode)->bdev;
300 struct list_head *p;
301 spin_lock(&bdev_lock);
302 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
303 __bd_forget(list_entry(p, struct inode, i_devices));
304 }
305 list_del_init(&bdev->bd_list);
306 spin_unlock(&bdev_lock);
307}
308
309static const struct super_operations bdev_sops = {
310 .statfs = simple_statfs,
311 .alloc_inode = bdev_alloc_inode,
312 .destroy_inode = bdev_destroy_inode,
313 .drop_inode = generic_delete_inode,
314 .clear_inode = bdev_clear_inode,
315};
316
317static int bd_get_sb(struct file_system_type *fs_type,
318 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
319{
320 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
321}
322
323static struct file_system_type bd_type = {
324 .name = "bdev",
325 .get_sb = bd_get_sb,
326 .kill_sb = kill_anon_super,
327};
328
329static struct vfsmount *bd_mnt __read_mostly;
330struct super_block *blockdev_superblock;
331
332void __init bdev_cache_init(void)
333{
334 int err;
335 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
336 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
337 SLAB_MEM_SPREAD|SLAB_PANIC),
338 init_once);
339 err = register_filesystem(&bd_type);
340 if (err)
341 panic("Cannot register bdev pseudo-fs");
342 bd_mnt = kern_mount(&bd_type);
343 if (IS_ERR(bd_mnt))
344 panic("Cannot create bdev pseudo-fs");
345 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
346}
347
348/*
349 * Most likely _very_ bad one - but then it's hardly critical for small
350 * /dev and can be fixed when somebody will need really large one.
351 * Keep in mind that it will be fed through icache hash function too.
352 */
353static inline unsigned long hash(dev_t dev)
354{
355 return MAJOR(dev)+MINOR(dev);
356}
357
358static int bdev_test(struct inode *inode, void *data)
359{
360 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
361}
362
363static int bdev_set(struct inode *inode, void *data)
364{
365 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
366 return 0;
367}
368
369static LIST_HEAD(all_bdevs);
370
371struct block_device *bdget(dev_t dev)
372{
373 struct block_device *bdev;
374 struct inode *inode;
375
376 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
377 bdev_test, bdev_set, &dev);
378
379 if (!inode)
380 return NULL;
381
382 bdev = &BDEV_I(inode)->bdev;
383
384 if (inode->i_state & I_NEW) {
385 bdev->bd_contains = NULL;
386 bdev->bd_inode = inode;
387 bdev->bd_block_size = (1 << inode->i_blkbits);
388 bdev->bd_part_count = 0;
389 bdev->bd_invalidated = 0;
390 inode->i_mode = S_IFBLK;
391 inode->i_rdev = dev;
392 inode->i_bdev = bdev;
393 inode->i_data.a_ops = &def_blk_aops;
394 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
395 inode->i_data.backing_dev_info = &default_backing_dev_info;
396 spin_lock(&bdev_lock);
397 list_add(&bdev->bd_list, &all_bdevs);
398 spin_unlock(&bdev_lock);
399 unlock_new_inode(inode);
400 }
401 return bdev;
402}
403
404EXPORT_SYMBOL(bdget);
405
406long nr_blockdev_pages(void)
407{
408 struct block_device *bdev;
409 long ret = 0;
410 spin_lock(&bdev_lock);
411 list_for_each_entry(bdev, &all_bdevs, bd_list) {
412 ret += bdev->bd_inode->i_mapping->nrpages;
413 }
414 spin_unlock(&bdev_lock);
415 return ret;
416}
417
418void bdput(struct block_device *bdev)
419{
420 iput(bdev->bd_inode);
421}
422
423EXPORT_SYMBOL(bdput);
424
425static struct block_device *bd_acquire(struct inode *inode)
426{
427 struct block_device *bdev;
428
429 spin_lock(&bdev_lock);
430 bdev = inode->i_bdev;
431 if (bdev) {
432 atomic_inc(&bdev->bd_inode->i_count);
433 spin_unlock(&bdev_lock);
434 return bdev;
435 }
436 spin_unlock(&bdev_lock);
437
438 bdev = bdget(inode->i_rdev);
439 if (bdev) {
440 spin_lock(&bdev_lock);
441 if (!inode->i_bdev) {
442 /*
443 * We take an additional bd_inode->i_count for inode,
444 * and it's released in clear_inode() of inode.
445 * So, we can access it via ->i_mapping always
446 * without igrab().
447 */
448 atomic_inc(&bdev->bd_inode->i_count);
449 inode->i_bdev = bdev;
450 inode->i_mapping = bdev->bd_inode->i_mapping;
451 list_add(&inode->i_devices, &bdev->bd_inodes);
452 }
453 spin_unlock(&bdev_lock);
454 }
455 return bdev;
456}
457
458/* Call when you free inode */
459
460void bd_forget(struct inode *inode)
461{
462 struct block_device *bdev = NULL;
463
464 spin_lock(&bdev_lock);
465 if (inode->i_bdev) {
466 if (inode->i_sb != blockdev_superblock)
467 bdev = inode->i_bdev;
468 __bd_forget(inode);
469 }
470 spin_unlock(&bdev_lock);
471
472 if (bdev)
473 iput(bdev->bd_inode);
474}
475
476int bd_claim(struct block_device *bdev, void *holder)
477{
478 int res;
479 spin_lock(&bdev_lock);
480
481 /* first decide result */
482 if (bdev->bd_holder == holder)
483 res = 0; /* already a holder */
484 else if (bdev->bd_holder != NULL)
485 res = -EBUSY; /* held by someone else */
486 else if (bdev->bd_contains == bdev)
487 res = 0; /* is a whole device which isn't held */
488
489 else if (bdev->bd_contains->bd_holder == bd_claim)
490 res = 0; /* is a partition of a device that is being partitioned */
491 else if (bdev->bd_contains->bd_holder != NULL)
492 res = -EBUSY; /* is a partition of a held device */
493 else
494 res = 0; /* is a partition of an un-held device */
495
496 /* now impose change */
497 if (res==0) {
498 /* note that for a whole device bd_holders
499 * will be incremented twice, and bd_holder will
500 * be set to bd_claim before being set to holder
501 */
502 bdev->bd_contains->bd_holders ++;
503 bdev->bd_contains->bd_holder = bd_claim;
504 bdev->bd_holders++;
505 bdev->bd_holder = holder;
506 }
507 spin_unlock(&bdev_lock);
508 return res;
509}
510
511EXPORT_SYMBOL(bd_claim);
512
513void bd_release(struct block_device *bdev)
514{
515 spin_lock(&bdev_lock);
516 if (!--bdev->bd_contains->bd_holders)
517 bdev->bd_contains->bd_holder = NULL;
518 if (!--bdev->bd_holders)
519 bdev->bd_holder = NULL;
520 spin_unlock(&bdev_lock);
521}
522
523EXPORT_SYMBOL(bd_release);
524
525#ifdef CONFIG_SYSFS
526/*
527 * Functions for bd_claim_by_kobject / bd_release_from_kobject
528 *
529 * If a kobject is passed to bd_claim_by_kobject()
530 * and the kobject has a parent directory,
531 * following symlinks are created:
532 * o from the kobject to the claimed bdev
533 * o from "holders" directory of the bdev to the parent of the kobject
534 * bd_release_from_kobject() removes these symlinks.
535 *
536 * Example:
537 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
538 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
539 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
540 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
541 */
542
543static struct kobject *bdev_get_kobj(struct block_device *bdev)
544{
545 if (bdev->bd_contains != bdev)
546 return kobject_get(&bdev->bd_part->dev.kobj);
547 else
548 return kobject_get(&bdev->bd_disk->dev.kobj);
549}
550
551static struct kobject *bdev_get_holder(struct block_device *bdev)
552{
553 if (bdev->bd_contains != bdev)
554 return kobject_get(bdev->bd_part->holder_dir);
555 else
556 return kobject_get(bdev->bd_disk->holder_dir);
557}
558
559static int add_symlink(struct kobject *from, struct kobject *to)
560{
561 if (!from || !to)
562 return 0;
563 return sysfs_create_link(from, to, kobject_name(to));
564}
565
566static void del_symlink(struct kobject *from, struct kobject *to)
567{
568 if (!from || !to)
569 return;
570 sysfs_remove_link(from, kobject_name(to));
571}
572
573/*
574 * 'struct bd_holder' contains pointers to kobjects symlinked by
575 * bd_claim_by_kobject.
576 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
577 */
578struct bd_holder {
579 struct list_head list; /* chain of holders of the bdev */
580 int count; /* references from the holder */
581 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
582 struct kobject *hdev; /* e.g. "/block/dm-0" */
583 struct kobject *hdir; /* e.g. "/block/sda/holders" */
584 struct kobject *sdev; /* e.g. "/block/sda" */
585};
586
587/*
588 * Get references of related kobjects at once.
589 * Returns 1 on success. 0 on failure.
590 *
591 * Should call bd_holder_release_dirs() after successful use.
592 */
593static int bd_holder_grab_dirs(struct block_device *bdev,
594 struct bd_holder *bo)
595{
596 if (!bdev || !bo)
597 return 0;
598
599 bo->sdir = kobject_get(bo->sdir);
600 if (!bo->sdir)
601 return 0;
602
603 bo->hdev = kobject_get(bo->sdir->parent);
604 if (!bo->hdev)
605 goto fail_put_sdir;
606
607 bo->sdev = bdev_get_kobj(bdev);
608 if (!bo->sdev)
609 goto fail_put_hdev;
610
611 bo->hdir = bdev_get_holder(bdev);
612 if (!bo->hdir)
613 goto fail_put_sdev;
614
615 return 1;
616
617fail_put_sdev:
618 kobject_put(bo->sdev);
619fail_put_hdev:
620 kobject_put(bo->hdev);
621fail_put_sdir:
622 kobject_put(bo->sdir);
623
624 return 0;
625}
626
627/* Put references of related kobjects at once. */
628static void bd_holder_release_dirs(struct bd_holder *bo)
629{
630 kobject_put(bo->hdir);
631 kobject_put(bo->sdev);
632 kobject_put(bo->hdev);
633 kobject_put(bo->sdir);
634}
635
636static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
637{
638 struct bd_holder *bo;
639
640 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
641 if (!bo)
642 return NULL;
643
644 bo->count = 1;
645 bo->sdir = kobj;
646
647 return bo;
648}
649
650static void free_bd_holder(struct bd_holder *bo)
651{
652 kfree(bo);
653}
654
655/**
656 * find_bd_holder - find matching struct bd_holder from the block device
657 *
658 * @bdev: struct block device to be searched
659 * @bo: target struct bd_holder
660 *
661 * Returns matching entry with @bo in @bdev->bd_holder_list.
662 * If found, increment the reference count and return the pointer.
663 * If not found, returns NULL.
664 */
665static struct bd_holder *find_bd_holder(struct block_device *bdev,
666 struct bd_holder *bo)
667{
668 struct bd_holder *tmp;
669
670 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
671 if (tmp->sdir == bo->sdir) {
672 tmp->count++;
673 return tmp;
674 }
675
676 return NULL;
677}
678
679/**
680 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
681 *
682 * @bdev: block device to be bd_claimed
683 * @bo: preallocated and initialized by alloc_bd_holder()
684 *
685 * Add @bo to @bdev->bd_holder_list, create symlinks.
686 *
687 * Returns 0 if symlinks are created.
688 * Returns -ve if something fails.
689 */
690static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
691{
692 int err;
693
694 if (!bo)
695 return -EINVAL;
696
697 if (!bd_holder_grab_dirs(bdev, bo))
698 return -EBUSY;
699
700 err = add_symlink(bo->sdir, bo->sdev);
701 if (err)
702 return err;
703
704 err = add_symlink(bo->hdir, bo->hdev);
705 if (err) {
706 del_symlink(bo->sdir, bo->sdev);
707 return err;
708 }
709
710 list_add_tail(&bo->list, &bdev->bd_holder_list);
711 return 0;
712}
713
714/**
715 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
716 *
717 * @bdev: block device to be bd_claimed
718 * @kobj: holder's kobject
719 *
720 * If there is matching entry with @kobj in @bdev->bd_holder_list
721 * and no other bd_claim() from the same kobject,
722 * remove the struct bd_holder from the list, delete symlinks for it.
723 *
724 * Returns a pointer to the struct bd_holder when it's removed from the list
725 * and ready to be freed.
726 * Returns NULL if matching claim isn't found or there is other bd_claim()
727 * by the same kobject.
728 */
729static struct bd_holder *del_bd_holder(struct block_device *bdev,
730 struct kobject *kobj)
731{
732 struct bd_holder *bo;
733
734 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
735 if (bo->sdir == kobj) {
736 bo->count--;
737 BUG_ON(bo->count < 0);
738 if (!bo->count) {
739 list_del(&bo->list);
740 del_symlink(bo->sdir, bo->sdev);
741 del_symlink(bo->hdir, bo->hdev);
742 bd_holder_release_dirs(bo);
743 return bo;
744 }
745 break;
746 }
747 }
748
749 return NULL;
750}
751
752/**
753 * bd_claim_by_kobject - bd_claim() with additional kobject signature
754 *
755 * @bdev: block device to be claimed
756 * @holder: holder's signature
757 * @kobj: holder's kobject
758 *
759 * Do bd_claim() and if it succeeds, create sysfs symlinks between
760 * the bdev and the holder's kobject.
761 * Use bd_release_from_kobject() when relesing the claimed bdev.
762 *
763 * Returns 0 on success. (same as bd_claim())
764 * Returns errno on failure.
765 */
766static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
767 struct kobject *kobj)
768{
769 int err;
770 struct bd_holder *bo, *found;
771
772 if (!kobj)
773 return -EINVAL;
774
775 bo = alloc_bd_holder(kobj);
776 if (!bo)
777 return -ENOMEM;
778
779 mutex_lock(&bdev->bd_mutex);
780
781 err = bd_claim(bdev, holder);
782 if (err)
783 goto fail;
784
785 found = find_bd_holder(bdev, bo);
786 if (found)
787 goto fail;
788
789 err = add_bd_holder(bdev, bo);
790 if (err)
791 bd_release(bdev);
792 else
793 bo = NULL;
794fail:
795 mutex_unlock(&bdev->bd_mutex);
796 free_bd_holder(bo);
797 return err;
798}
799
800/**
801 * bd_release_from_kobject - bd_release() with additional kobject signature
802 *
803 * @bdev: block device to be released
804 * @kobj: holder's kobject
805 *
806 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
807 */
808static void bd_release_from_kobject(struct block_device *bdev,
809 struct kobject *kobj)
810{
811 if (!kobj)
812 return;
813
814 mutex_lock(&bdev->bd_mutex);
815 bd_release(bdev);
816 free_bd_holder(del_bd_holder(bdev, kobj));
817 mutex_unlock(&bdev->bd_mutex);
818}
819
820/**
821 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
822 *
823 * @bdev: block device to be claimed
824 * @holder: holder's signature
825 * @disk: holder's gendisk
826 *
827 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
828 */
829int bd_claim_by_disk(struct block_device *bdev, void *holder,
830 struct gendisk *disk)
831{
832 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
833}
834EXPORT_SYMBOL_GPL(bd_claim_by_disk);
835
836/**
837 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
838 *
839 * @bdev: block device to be claimed
840 * @disk: holder's gendisk
841 *
842 * Call bd_release_from_kobject() and put @disk->slave_dir.
843 */
844void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
845{
846 bd_release_from_kobject(bdev, disk->slave_dir);
847 kobject_put(disk->slave_dir);
848}
849EXPORT_SYMBOL_GPL(bd_release_from_disk);
850#endif
851
852/*
853 * Tries to open block device by device number. Use it ONLY if you
854 * really do not have anything better - i.e. when you are behind a
855 * truly sucky interface and all you are given is a device number. _Never_
856 * to be used for internal purposes. If you ever need it - reconsider
857 * your API.
858 */
859struct block_device *open_by_devnum(dev_t dev, unsigned mode)
860{
861 struct block_device *bdev = bdget(dev);
862 int err = -ENOMEM;
863 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
864 if (bdev)
865 err = blkdev_get(bdev, mode, flags);
866 return err ? ERR_PTR(err) : bdev;
867}
868
869EXPORT_SYMBOL(open_by_devnum);
870
871/*
872 * This routine checks whether a removable media has been changed,
873 * and invalidates all buffer-cache-entries in that case. This
874 * is a relatively slow routine, so we have to try to minimize using
875 * it. Thus it is called only upon a 'mount' or 'open'. This
876 * is the best way of combining speed and utility, I think.
877 * People changing diskettes in the middle of an operation deserve
878 * to lose :-)
879 */
880int check_disk_change(struct block_device *bdev)
881{
882 struct gendisk *disk = bdev->bd_disk;
883 struct block_device_operations * bdops = disk->fops;
884
885 if (!bdops->media_changed)
886 return 0;
887 if (!bdops->media_changed(bdev->bd_disk))
888 return 0;
889
890 if (__invalidate_device(bdev))
891 printk("VFS: busy inodes on changed media.\n");
892
893 if (bdops->revalidate_disk)
894 bdops->revalidate_disk(bdev->bd_disk);
895 if (bdev->bd_disk->minors > 1)
896 bdev->bd_invalidated = 1;
897 return 1;
898}
899
900EXPORT_SYMBOL(check_disk_change);
901
902void bd_set_size(struct block_device *bdev, loff_t size)
903{
904 unsigned bsize = bdev_hardsect_size(bdev);
905
906 bdev->bd_inode->i_size = size;
907 while (bsize < PAGE_CACHE_SIZE) {
908 if (size & bsize)
909 break;
910 bsize <<= 1;
911 }
912 bdev->bd_block_size = bsize;
913 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
914}
915EXPORT_SYMBOL(bd_set_size);
916
917static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
918 int for_part);
919static int __blkdev_put(struct block_device *bdev, int for_part);
920
921/*
922 * bd_mutex locking:
923 *
924 * mutex_lock(part->bd_mutex)
925 * mutex_lock_nested(whole->bd_mutex, 1)
926 */
927
928static int do_open(struct block_device *bdev, struct file *file, int for_part)
929{
930 struct module *owner = NULL;
931 struct gendisk *disk;
932 int ret;
933 int part;
934 int perm = 0;
935
936 if (file->f_mode & FMODE_READ)
937 perm |= MAY_READ;
938 if (file->f_mode & FMODE_WRITE)
939 perm |= MAY_WRITE;
940 /*
941 * hooks: /n/, see "layering violations".
942 */
943 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
944 if (ret != 0)
945 return ret;
946
947 ret = -ENXIO;
948 file->f_mapping = bdev->bd_inode->i_mapping;
949 lock_kernel();
950 disk = get_gendisk(bdev->bd_dev, &part);
951 if (!disk) {
952 unlock_kernel();
953 bdput(bdev);
954 return ret;
955 }
956 owner = disk->fops->owner;
957
958 mutex_lock_nested(&bdev->bd_mutex, for_part);
959 if (!bdev->bd_openers) {
960 bdev->bd_disk = disk;
961 bdev->bd_contains = bdev;
962 if (!part) {
963 struct backing_dev_info *bdi;
964 if (disk->fops->open) {
965 ret = disk->fops->open(bdev->bd_inode, file);
966 if (ret)
967 goto out_first;
968 }
969 if (!bdev->bd_openers) {
970 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
971 bdi = blk_get_backing_dev_info(bdev);
972 if (bdi == NULL)
973 bdi = &default_backing_dev_info;
974 bdev->bd_inode->i_data.backing_dev_info = bdi;
975 }
976 if (bdev->bd_invalidated)
977 rescan_partitions(disk, bdev);
978 } else {
979 struct hd_struct *p;
980 struct block_device *whole;
981 whole = bdget_disk(disk, 0);
982 ret = -ENOMEM;
983 if (!whole)
984 goto out_first;
985 BUG_ON(for_part);
986 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
987 if (ret)
988 goto out_first;
989 bdev->bd_contains = whole;
990 p = disk->part[part - 1];
991 bdev->bd_inode->i_data.backing_dev_info =
992 whole->bd_inode->i_data.backing_dev_info;
993 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
994 ret = -ENXIO;
995 goto out_first;
996 }
997 kobject_get(&p->dev.kobj);
998 bdev->bd_part = p;
999 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1000 }
1001 } else {
1002 put_disk(disk);
1003 module_put(owner);
1004 if (bdev->bd_contains == bdev) {
1005 if (bdev->bd_disk->fops->open) {
1006 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1007 if (ret)
1008 goto out;
1009 }
1010 if (bdev->bd_invalidated)
1011 rescan_partitions(bdev->bd_disk, bdev);
1012 }
1013 }
1014 bdev->bd_openers++;
1015 if (for_part)
1016 bdev->bd_part_count++;
1017 mutex_unlock(&bdev->bd_mutex);
1018 unlock_kernel();
1019 return 0;
1020
1021out_first:
1022 bdev->bd_disk = NULL;
1023 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1024 if (bdev != bdev->bd_contains)
1025 __blkdev_put(bdev->bd_contains, 1);
1026 bdev->bd_contains = NULL;
1027 put_disk(disk);
1028 module_put(owner);
1029out:
1030 mutex_unlock(&bdev->bd_mutex);
1031 unlock_kernel();
1032 if (ret)
1033 bdput(bdev);
1034 return ret;
1035}
1036
1037static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1038 int for_part)
1039{
1040 /*
1041 * This crockload is due to bad choice of ->open() type.
1042 * It will go away.
1043 * For now, block device ->open() routine must _not_
1044 * examine anything in 'inode' argument except ->i_rdev.
1045 */
1046 struct file fake_file = {};
1047 struct dentry fake_dentry = {};
1048 fake_file.f_mode = mode;
1049 fake_file.f_flags = flags;
1050 fake_file.f_path.dentry = &fake_dentry;
1051 fake_dentry.d_inode = bdev->bd_inode;
1052
1053 return do_open(bdev, &fake_file, for_part);
1054}
1055
1056int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1057{
1058 return __blkdev_get(bdev, mode, flags, 0);
1059}
1060EXPORT_SYMBOL(blkdev_get);
1061
1062static int blkdev_open(struct inode * inode, struct file * filp)
1063{
1064 struct block_device *bdev;
1065 int res;
1066
1067 /*
1068 * Preserve backwards compatibility and allow large file access
1069 * even if userspace doesn't ask for it explicitly. Some mkfs
1070 * binary needs it. We might want to drop this workaround
1071 * during an unstable branch.
1072 */
1073 filp->f_flags |= O_LARGEFILE;
1074
1075 bdev = bd_acquire(inode);
1076 if (bdev == NULL)
1077 return -ENOMEM;
1078
1079 res = do_open(bdev, filp, 0);
1080 if (res)
1081 return res;
1082
1083 if (!(filp->f_flags & O_EXCL) )
1084 return 0;
1085
1086 if (!(res = bd_claim(bdev, filp)))
1087 return 0;
1088
1089 blkdev_put(bdev);
1090 return res;
1091}
1092
1093static int __blkdev_put(struct block_device *bdev, int for_part)
1094{
1095 int ret = 0;
1096 struct inode *bd_inode = bdev->bd_inode;
1097 struct gendisk *disk = bdev->bd_disk;
1098 struct block_device *victim = NULL;
1099
1100 mutex_lock_nested(&bdev->bd_mutex, for_part);
1101 lock_kernel();
1102 if (for_part)
1103 bdev->bd_part_count--;
1104
1105 if (!--bdev->bd_openers) {
1106 sync_blockdev(bdev);
1107 kill_bdev(bdev);
1108 }
1109 if (bdev->bd_contains == bdev) {
1110 if (disk->fops->release)
1111 ret = disk->fops->release(bd_inode, NULL);
1112 }
1113 if (!bdev->bd_openers) {
1114 struct module *owner = disk->fops->owner;
1115
1116 put_disk(disk);
1117 module_put(owner);
1118
1119 if (bdev->bd_contains != bdev) {
1120 kobject_put(&bdev->bd_part->dev.kobj);
1121 bdev->bd_part = NULL;
1122 }
1123 bdev->bd_disk = NULL;
1124 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1125 if (bdev != bdev->bd_contains)
1126 victim = bdev->bd_contains;
1127 bdev->bd_contains = NULL;
1128 }
1129 unlock_kernel();
1130 mutex_unlock(&bdev->bd_mutex);
1131 bdput(bdev);
1132 if (victim)
1133 __blkdev_put(victim, 1);
1134 return ret;
1135}
1136
1137int blkdev_put(struct block_device *bdev)
1138{
1139 return __blkdev_put(bdev, 0);
1140}
1141EXPORT_SYMBOL(blkdev_put);
1142
1143static int blkdev_close(struct inode * inode, struct file * filp)
1144{
1145 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1146 if (bdev->bd_holder == filp)
1147 bd_release(bdev);
1148 return blkdev_put(bdev);
1149}
1150
1151static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1152{
1153 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1154}
1155
1156static const struct address_space_operations def_blk_aops = {
1157 .readpage = blkdev_readpage,
1158 .writepage = blkdev_writepage,
1159 .sync_page = block_sync_page,
1160 .write_begin = blkdev_write_begin,
1161 .write_end = blkdev_write_end,
1162 .writepages = generic_writepages,
1163 .direct_IO = blkdev_direct_IO,
1164};
1165
1166const struct file_operations def_blk_fops = {
1167 .open = blkdev_open,
1168 .release = blkdev_close,
1169 .llseek = block_llseek,
1170 .read = do_sync_read,
1171 .write = do_sync_write,
1172 .aio_read = generic_file_aio_read,
1173 .aio_write = generic_file_aio_write_nolock,
1174 .mmap = generic_file_mmap,
1175 .fsync = block_fsync,
1176 .unlocked_ioctl = block_ioctl,
1177#ifdef CONFIG_COMPAT
1178 .compat_ioctl = compat_blkdev_ioctl,
1179#endif
1180 .splice_read = generic_file_splice_read,
1181 .splice_write = generic_file_splice_write,
1182};
1183
1184int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1185{
1186 int res;
1187 mm_segment_t old_fs = get_fs();
1188 set_fs(KERNEL_DS);
1189 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1190 set_fs(old_fs);
1191 return res;
1192}
1193
1194EXPORT_SYMBOL(ioctl_by_bdev);
1195
1196/**
1197 * lookup_bdev - lookup a struct block_device by name
1198 *
1199 * @path: special file representing the block device
1200 *
1201 * Get a reference to the blockdevice at @path in the current
1202 * namespace if possible and return it. Return ERR_PTR(error)
1203 * otherwise.
1204 */
1205struct block_device *lookup_bdev(const char *path)
1206{
1207 struct block_device *bdev;
1208 struct inode *inode;
1209 struct nameidata nd;
1210 int error;
1211
1212 if (!path || !*path)
1213 return ERR_PTR(-EINVAL);
1214
1215 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1216 if (error)
1217 return ERR_PTR(error);
1218
1219 inode = nd.path.dentry->d_inode;
1220 error = -ENOTBLK;
1221 if (!S_ISBLK(inode->i_mode))
1222 goto fail;
1223 error = -EACCES;
1224 if (nd.path.mnt->mnt_flags & MNT_NODEV)
1225 goto fail;
1226 error = -ENOMEM;
1227 bdev = bd_acquire(inode);
1228 if (!bdev)
1229 goto fail;
1230out:
1231 path_put(&nd.path);
1232 return bdev;
1233fail:
1234 bdev = ERR_PTR(error);
1235 goto out;
1236}
1237
1238/**
1239 * open_bdev_excl - open a block device by name and set it up for use
1240 *
1241 * @path: special file representing the block device
1242 * @flags: %MS_RDONLY for opening read-only
1243 * @holder: owner for exclusion
1244 *
1245 * Open the blockdevice described by the special file at @path, claim it
1246 * for the @holder.
1247 */
1248struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1249{
1250 struct block_device *bdev;
1251 mode_t mode = FMODE_READ;
1252 int error = 0;
1253
1254 bdev = lookup_bdev(path);
1255 if (IS_ERR(bdev))
1256 return bdev;
1257
1258 if (!(flags & MS_RDONLY))
1259 mode |= FMODE_WRITE;
1260 error = blkdev_get(bdev, mode, 0);
1261 if (error)
1262 return ERR_PTR(error);
1263 error = -EACCES;
1264 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1265 goto blkdev_put;
1266 error = bd_claim(bdev, holder);
1267 if (error)
1268 goto blkdev_put;
1269
1270 return bdev;
1271
1272blkdev_put:
1273 blkdev_put(bdev);
1274 return ERR_PTR(error);
1275}
1276
1277EXPORT_SYMBOL(open_bdev_excl);
1278
1279/**
1280 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1281 *
1282 * @bdev: blockdevice to close
1283 *
1284 * This is the counterpart to open_bdev_excl().
1285 */
1286void close_bdev_excl(struct block_device *bdev)
1287{
1288 bd_release(bdev);
1289 blkdev_put(bdev);
1290}
1291
1292EXPORT_SYMBOL(close_bdev_excl);
1293
1294int __invalidate_device(struct block_device *bdev)
1295{
1296 struct super_block *sb = get_super(bdev);
1297 int res = 0;
1298
1299 if (sb) {
1300 /*
1301 * no need to lock the super, get_super holds the
1302 * read mutex so the filesystem cannot go away
1303 * under us (->put_super runs with the write lock
1304 * hold).
1305 */
1306 shrink_dcache_sb(sb);
1307 res = invalidate_inodes(sb);
1308 drop_super(sb);
1309 }
1310 invalidate_bdev(bdev);
1311 return res;
1312}
1313EXPORT_SYMBOL(__invalidate_device);