projects/linux·blob v2.6.21 · refs

1484 lines · 35.0 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/highmem.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/blkpg.h>
19#include <linux/buffer_head.h>
20#include <linux/writeback.h>
21#include <linux/mpage.h>
22#include <linux/mount.h>
23#include <linux/uio.h>
24#include <linux/namei.h>
25#include <asm/uaccess.h>
26#include "internal.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{
89 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;
94 sb->s_blocksize_bits = blksize_bits(size);
95 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,
134 struct buffer_head *bh, int create)
135{
136 sector_t end_block = max_block(I_BDEV(inode));
137 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
138
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
171#if 0
172static int blk_end_aio(struct bio *bio, unsigned int bytes_done, int error)
173{
174 struct kiocb *iocb = bio->bi_private;
175 atomic_t *bio_count = &iocb->ki_bio_count;
176
177 if (bio_data_dir(bio) == READ)
178 bio_check_pages_dirty(bio);
179 else {
180 bio_release_pages(bio);
181 bio_put(bio);
182 }
183
184 /* iocb->ki_nbytes stores error code from LLDD */
185 if (error)
186 iocb->ki_nbytes = -EIO;
187
188 if (atomic_dec_and_test(bio_count)) {
189 if ((long)iocb->ki_nbytes < 0)
190 aio_complete(iocb, iocb->ki_nbytes, 0);
191 else
192 aio_complete(iocb, iocb->ki_left, 0);
193 }
194
195 return 0;
196}
197
198#define VEC_SIZE 16
199struct pvec {
200 unsigned short nr;
201 unsigned short idx;
202 struct page *page[VEC_SIZE];
203};
204
205#define PAGES_SPANNED(addr, len) \
206 (DIV_ROUND_UP((addr) + (len), PAGE_SIZE) - (addr) / PAGE_SIZE);
207
208/*
209 * get page pointer for user addr, we internally cache struct page array for
210 * (addr, count) range in pvec to avoid frequent call to get_user_pages. If
211 * internal page list is exhausted, a batch count of up to VEC_SIZE is used
212 * to get next set of page struct.
213 */
214static struct page *blk_get_page(unsigned long addr, size_t count, int rw,
215 struct pvec *pvec)
216{
217 int ret, nr_pages;
218 if (pvec->idx == pvec->nr) {
219 nr_pages = PAGES_SPANNED(addr, count);
220 nr_pages = min(nr_pages, VEC_SIZE);
221 down_read(&current->mm->mmap_sem);
222 ret = get_user_pages(current, current->mm, addr, nr_pages,
223 rw == READ, 0, pvec->page, NULL);
224 up_read(&current->mm->mmap_sem);
225 if (ret < 0)
226 return ERR_PTR(ret);
227 pvec->nr = ret;
228 pvec->idx = 0;
229 }
230 return pvec->page[pvec->idx++];
231}
232
233/* return a page back to pvec array */
234static void blk_unget_page(struct page *page, struct pvec *pvec)
235{
236 pvec->page[--pvec->idx] = page;
237}
238
239static ssize_t
240blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
241 loff_t pos, unsigned long nr_segs)
242{
243 struct inode *inode = iocb->ki_filp->f_mapping->host;
244 unsigned blkbits = blksize_bits(bdev_hardsect_size(I_BDEV(inode)));
245 unsigned blocksize_mask = (1 << blkbits) - 1;
246 unsigned long seg = 0; /* iov segment iterator */
247 unsigned long nvec; /* number of bio vec needed */
248 unsigned long cur_off; /* offset into current page */
249 unsigned long cur_len; /* I/O len of current page, up to PAGE_SIZE */
250
251 unsigned long addr; /* user iovec address */
252 size_t count; /* user iovec len */
253 size_t nbytes = iocb->ki_nbytes = iocb->ki_left; /* total xfer size */
254 loff_t size; /* size of block device */
255 struct bio *bio;
256 atomic_t *bio_count = &iocb->ki_bio_count;
257 struct page *page;
258 struct pvec pvec;
259
260 pvec.nr = 0;
261 pvec.idx = 0;
262
263 if (pos & blocksize_mask)
264 return -EINVAL;
265
266 size = i_size_read(inode);
267 if (pos + nbytes > size) {
268 nbytes = size - pos;
269 iocb->ki_left = nbytes;
270 }
271
272 /*
273 * check first non-zero iov alignment, the remaining
274 * iov alignment is checked inside bio loop below.
275 */
276 do {
277 addr = (unsigned long) iov[seg].iov_base;
278 count = min(iov[seg].iov_len, nbytes);
279 if (addr & blocksize_mask || count & blocksize_mask)
280 return -EINVAL;
281 } while (!count && ++seg < nr_segs);
282 atomic_set(bio_count, 1);
283
284 while (nbytes) {
285 /* roughly estimate number of bio vec needed */
286 nvec = (nbytes + PAGE_SIZE - 1) / PAGE_SIZE;
287 nvec = max(nvec, nr_segs - seg);
288 nvec = min(nvec, (unsigned long) BIO_MAX_PAGES);
289
290 /* bio_alloc should not fail with GFP_KERNEL flag */
291 bio = bio_alloc(GFP_KERNEL, nvec);
292 bio->bi_bdev = I_BDEV(inode);
293 bio->bi_end_io = blk_end_aio;
294 bio->bi_private = iocb;
295 bio->bi_sector = pos >> blkbits;
296same_bio:
297 cur_off = addr & ~PAGE_MASK;
298 cur_len = PAGE_SIZE - cur_off;
299 if (count < cur_len)
300 cur_len = count;
301
302 page = blk_get_page(addr, count, rw, &pvec);
303 if (unlikely(IS_ERR(page)))
304 goto backout;
305
306 if (bio_add_page(bio, page, cur_len, cur_off)) {
307 pos += cur_len;
308 addr += cur_len;
309 count -= cur_len;
310 nbytes -= cur_len;
311
312 if (count)
313 goto same_bio;
314 while (++seg < nr_segs) {
315 addr = (unsigned long) iov[seg].iov_base;
316 count = iov[seg].iov_len;
317 if (!count)
318 continue;
319 if (unlikely(addr & blocksize_mask ||
320 count & blocksize_mask)) {
321 page = ERR_PTR(-EINVAL);
322 goto backout;
323 }
324 count = min(count, nbytes);
325 goto same_bio;
326 }
327 } else {
328 blk_unget_page(page, &pvec);
329 }
330
331 /* bio is ready, submit it */
332 if (rw == READ)
333 bio_set_pages_dirty(bio);
334 atomic_inc(bio_count);
335 submit_bio(rw, bio);
336 }
337
338completion:
339 iocb->ki_left -= nbytes;
340 nbytes = iocb->ki_left;
341 iocb->ki_pos += nbytes;
342
343 blk_run_address_space(inode->i_mapping);
344 if (atomic_dec_and_test(bio_count))
345 aio_complete(iocb, nbytes, 0);
346
347 return -EIOCBQUEUED;
348
349backout:
350 /*
351 * back out nbytes count constructed so far for this bio,
352 * we will throw away current bio.
353 */
354 nbytes += bio->bi_size;
355 bio_release_pages(bio);
356 bio_put(bio);
357
358 /*
359 * if no bio was submmitted, return the error code.
360 * otherwise, proceed with pending I/O completion.
361 */
362 if (atomic_read(bio_count) == 1)
363 return PTR_ERR(page);
364 goto completion;
365}
366#endif
367
368static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
369{
370 return block_write_full_page(page, blkdev_get_block, wbc);
371}
372
373static int blkdev_readpage(struct file * file, struct page * page)
374{
375 return block_read_full_page(page, blkdev_get_block);
376}
377
378static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
379{
380 return block_prepare_write(page, from, to, blkdev_get_block);
381}
382
383static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
384{
385 return block_commit_write(page, from, to);
386}
387
388/*
389 * private llseek:
390 * for a block special file file->f_path.dentry->d_inode->i_size is zero
391 * so we compute the size by hand (just as in block_read/write above)
392 */
393static loff_t block_llseek(struct file *file, loff_t offset, int origin)
394{
395 struct inode *bd_inode = file->f_mapping->host;
396 loff_t size;
397 loff_t retval;
398
399 mutex_lock(&bd_inode->i_mutex);
400 size = i_size_read(bd_inode);
401
402 switch (origin) {
403 case 2:
404 offset += size;
405 break;
406 case 1:
407 offset += file->f_pos;
408 }
409 retval = -EINVAL;
410 if (offset >= 0 && offset <= size) {
411 if (offset != file->f_pos) {
412 file->f_pos = offset;
413 }
414 retval = offset;
415 }
416 mutex_unlock(&bd_inode->i_mutex);
417 return retval;
418}
419
420/*
421 * Filp is never NULL; the only case when ->fsync() is called with
422 * NULL first argument is nfsd_sync_dir() and that's not a directory.
423 */
424
425static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
426{
427 return sync_blockdev(I_BDEV(filp->f_mapping->host));
428}
429
430/*
431 * pseudo-fs
432 */
433
434static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
435static struct kmem_cache * bdev_cachep __read_mostly;
436
437static struct inode *bdev_alloc_inode(struct super_block *sb)
438{
439 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
440 if (!ei)
441 return NULL;
442 return &ei->vfs_inode;
443}
444
445static void bdev_destroy_inode(struct inode *inode)
446{
447 struct bdev_inode *bdi = BDEV_I(inode);
448
449 bdi->bdev.bd_inode_backing_dev_info = NULL;
450 kmem_cache_free(bdev_cachep, bdi);
451}
452
453static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
454{
455 struct bdev_inode *ei = (struct bdev_inode *) foo;
456 struct block_device *bdev = &ei->bdev;
457
458 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
459 SLAB_CTOR_CONSTRUCTOR)
460 {
461 memset(bdev, 0, sizeof(*bdev));
462 mutex_init(&bdev->bd_mutex);
463 sema_init(&bdev->bd_mount_sem, 1);
464 INIT_LIST_HEAD(&bdev->bd_inodes);
465 INIT_LIST_HEAD(&bdev->bd_list);
466#ifdef CONFIG_SYSFS
467 INIT_LIST_HEAD(&bdev->bd_holder_list);
468#endif
469 inode_init_once(&ei->vfs_inode);
470 }
471}
472
473static inline void __bd_forget(struct inode *inode)
474{
475 list_del_init(&inode->i_devices);
476 inode->i_bdev = NULL;
477 inode->i_mapping = &inode->i_data;
478}
479
480static void bdev_clear_inode(struct inode *inode)
481{
482 struct block_device *bdev = &BDEV_I(inode)->bdev;
483 struct list_head *p;
484 spin_lock(&bdev_lock);
485 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
486 __bd_forget(list_entry(p, struct inode, i_devices));
487 }
488 list_del_init(&bdev->bd_list);
489 spin_unlock(&bdev_lock);
490}
491
492static const struct super_operations bdev_sops = {
493 .statfs = simple_statfs,
494 .alloc_inode = bdev_alloc_inode,
495 .destroy_inode = bdev_destroy_inode,
496 .drop_inode = generic_delete_inode,
497 .clear_inode = bdev_clear_inode,
498};
499
500static int bd_get_sb(struct file_system_type *fs_type,
501 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
502{
503 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
504}
505
506static struct file_system_type bd_type = {
507 .name = "bdev",
508 .get_sb = bd_get_sb,
509 .kill_sb = kill_anon_super,
510};
511
512static struct vfsmount *bd_mnt __read_mostly;
513struct super_block *blockdev_superblock;
514
515void __init bdev_cache_init(void)
516{
517 int err;
518 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
519 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
520 SLAB_MEM_SPREAD|SLAB_PANIC),
521 init_once, NULL);
522 err = register_filesystem(&bd_type);
523 if (err)
524 panic("Cannot register bdev pseudo-fs");
525 bd_mnt = kern_mount(&bd_type);
526 err = PTR_ERR(bd_mnt);
527 if (IS_ERR(bd_mnt))
528 panic("Cannot create bdev pseudo-fs");
529 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
530}
531
532/*
533 * Most likely _very_ bad one - but then it's hardly critical for small
534 * /dev and can be fixed when somebody will need really large one.
535 * Keep in mind that it will be fed through icache hash function too.
536 */
537static inline unsigned long hash(dev_t dev)
538{
539 return MAJOR(dev)+MINOR(dev);
540}
541
542static int bdev_test(struct inode *inode, void *data)
543{
544 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
545}
546
547static int bdev_set(struct inode *inode, void *data)
548{
549 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
550 return 0;
551}
552
553static LIST_HEAD(all_bdevs);
554
555struct block_device *bdget(dev_t dev)
556{
557 struct block_device *bdev;
558 struct inode *inode;
559
560 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
561 bdev_test, bdev_set, &dev);
562
563 if (!inode)
564 return NULL;
565
566 bdev = &BDEV_I(inode)->bdev;
567
568 if (inode->i_state & I_NEW) {
569 bdev->bd_contains = NULL;
570 bdev->bd_inode = inode;
571 bdev->bd_block_size = (1 << inode->i_blkbits);
572 bdev->bd_part_count = 0;
573 bdev->bd_invalidated = 0;
574 inode->i_mode = S_IFBLK;
575 inode->i_rdev = dev;
576 inode->i_bdev = bdev;
577 inode->i_data.a_ops = &def_blk_aops;
578 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
579 inode->i_data.backing_dev_info = &default_backing_dev_info;
580 spin_lock(&bdev_lock);
581 list_add(&bdev->bd_list, &all_bdevs);
582 spin_unlock(&bdev_lock);
583 unlock_new_inode(inode);
584 }
585 return bdev;
586}
587
588EXPORT_SYMBOL(bdget);
589
590long nr_blockdev_pages(void)
591{
592 struct list_head *p;
593 long ret = 0;
594 spin_lock(&bdev_lock);
595 list_for_each(p, &all_bdevs) {
596 struct block_device *bdev;
597 bdev = list_entry(p, struct block_device, bd_list);
598 ret += bdev->bd_inode->i_mapping->nrpages;
599 }
600 spin_unlock(&bdev_lock);
601 return ret;
602}
603
604void bdput(struct block_device *bdev)
605{
606 iput(bdev->bd_inode);
607}
608
609EXPORT_SYMBOL(bdput);
610
611static struct block_device *bd_acquire(struct inode *inode)
612{
613 struct block_device *bdev;
614
615 spin_lock(&bdev_lock);
616 bdev = inode->i_bdev;
617 if (bdev) {
618 atomic_inc(&bdev->bd_inode->i_count);
619 spin_unlock(&bdev_lock);
620 return bdev;
621 }
622 spin_unlock(&bdev_lock);
623
624 bdev = bdget(inode->i_rdev);
625 if (bdev) {
626 spin_lock(&bdev_lock);
627 if (!inode->i_bdev) {
628 /*
629 * We take an additional bd_inode->i_count for inode,
630 * and it's released in clear_inode() of inode.
631 * So, we can access it via ->i_mapping always
632 * without igrab().
633 */
634 atomic_inc(&bdev->bd_inode->i_count);
635 inode->i_bdev = bdev;
636 inode->i_mapping = bdev->bd_inode->i_mapping;
637 list_add(&inode->i_devices, &bdev->bd_inodes);
638 }
639 spin_unlock(&bdev_lock);
640 }
641 return bdev;
642}
643
644/* Call when you free inode */
645
646void bd_forget(struct inode *inode)
647{
648 struct block_device *bdev = NULL;
649
650 spin_lock(&bdev_lock);
651 if (inode->i_bdev) {
652 if (inode->i_sb != blockdev_superblock)
653 bdev = inode->i_bdev;
654 __bd_forget(inode);
655 }
656 spin_unlock(&bdev_lock);
657
658 if (bdev)
659 iput(bdev->bd_inode);
660}
661
662int bd_claim(struct block_device *bdev, void *holder)
663{
664 int res;
665 spin_lock(&bdev_lock);
666
667 /* first decide result */
668 if (bdev->bd_holder == holder)
669 res = 0; /* already a holder */
670 else if (bdev->bd_holder != NULL)
671 res = -EBUSY; /* held by someone else */
672 else if (bdev->bd_contains == bdev)
673 res = 0; /* is a whole device which isn't held */
674
675 else if (bdev->bd_contains->bd_holder == bd_claim)
676 res = 0; /* is a partition of a device that is being partitioned */
677 else if (bdev->bd_contains->bd_holder != NULL)
678 res = -EBUSY; /* is a partition of a held device */
679 else
680 res = 0; /* is a partition of an un-held device */
681
682 /* now impose change */
683 if (res==0) {
684 /* note that for a whole device bd_holders
685 * will be incremented twice, and bd_holder will
686 * be set to bd_claim before being set to holder
687 */
688 bdev->bd_contains->bd_holders ++;
689 bdev->bd_contains->bd_holder = bd_claim;
690 bdev->bd_holders++;
691 bdev->bd_holder = holder;
692 }
693 spin_unlock(&bdev_lock);
694 return res;
695}
696
697EXPORT_SYMBOL(bd_claim);
698
699void bd_release(struct block_device *bdev)
700{
701 spin_lock(&bdev_lock);
702 if (!--bdev->bd_contains->bd_holders)
703 bdev->bd_contains->bd_holder = NULL;
704 if (!--bdev->bd_holders)
705 bdev->bd_holder = NULL;
706 spin_unlock(&bdev_lock);
707}
708
709EXPORT_SYMBOL(bd_release);
710
711#ifdef CONFIG_SYSFS
712/*
713 * Functions for bd_claim_by_kobject / bd_release_from_kobject
714 *
715 * If a kobject is passed to bd_claim_by_kobject()
716 * and the kobject has a parent directory,
717 * following symlinks are created:
718 * o from the kobject to the claimed bdev
719 * o from "holders" directory of the bdev to the parent of the kobject
720 * bd_release_from_kobject() removes these symlinks.
721 *
722 * Example:
723 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
724 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
725 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
726 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
727 */
728
729static struct kobject *bdev_get_kobj(struct block_device *bdev)
730{
731 if (bdev->bd_contains != bdev)
732 return kobject_get(&bdev->bd_part->kobj);
733 else
734 return kobject_get(&bdev->bd_disk->kobj);
735}
736
737static struct kobject *bdev_get_holder(struct block_device *bdev)
738{
739 if (bdev->bd_contains != bdev)
740 return kobject_get(bdev->bd_part->holder_dir);
741 else
742 return kobject_get(bdev->bd_disk->holder_dir);
743}
744
745static int add_symlink(struct kobject *from, struct kobject *to)
746{
747 if (!from || !to)
748 return 0;
749 return sysfs_create_link(from, to, kobject_name(to));
750}
751
752static void del_symlink(struct kobject *from, struct kobject *to)
753{
754 if (!from || !to)
755 return;
756 sysfs_remove_link(from, kobject_name(to));
757}
758
759/*
760 * 'struct bd_holder' contains pointers to kobjects symlinked by
761 * bd_claim_by_kobject.
762 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
763 */
764struct bd_holder {
765 struct list_head list; /* chain of holders of the bdev */
766 int count; /* references from the holder */
767 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
768 struct kobject *hdev; /* e.g. "/block/dm-0" */
769 struct kobject *hdir; /* e.g. "/block/sda/holders" */
770 struct kobject *sdev; /* e.g. "/block/sda" */
771};
772
773/*
774 * Get references of related kobjects at once.
775 * Returns 1 on success. 0 on failure.
776 *
777 * Should call bd_holder_release_dirs() after successful use.
778 */
779static int bd_holder_grab_dirs(struct block_device *bdev,
780 struct bd_holder *bo)
781{
782 if (!bdev || !bo)
783 return 0;
784
785 bo->sdir = kobject_get(bo->sdir);
786 if (!bo->sdir)
787 return 0;
788
789 bo->hdev = kobject_get(bo->sdir->parent);
790 if (!bo->hdev)
791 goto fail_put_sdir;
792
793 bo->sdev = bdev_get_kobj(bdev);
794 if (!bo->sdev)
795 goto fail_put_hdev;
796
797 bo->hdir = bdev_get_holder(bdev);
798 if (!bo->hdir)
799 goto fail_put_sdev;
800
801 return 1;
802
803fail_put_sdev:
804 kobject_put(bo->sdev);
805fail_put_hdev:
806 kobject_put(bo->hdev);
807fail_put_sdir:
808 kobject_put(bo->sdir);
809
810 return 0;
811}
812
813/* Put references of related kobjects at once. */
814static void bd_holder_release_dirs(struct bd_holder *bo)
815{
816 kobject_put(bo->hdir);
817 kobject_put(bo->sdev);
818 kobject_put(bo->hdev);
819 kobject_put(bo->sdir);
820}
821
822static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
823{
824 struct bd_holder *bo;
825
826 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
827 if (!bo)
828 return NULL;
829
830 bo->count = 1;
831 bo->sdir = kobj;
832
833 return bo;
834}
835
836static void free_bd_holder(struct bd_holder *bo)
837{
838 kfree(bo);
839}
840
841/**
842 * find_bd_holder - find matching struct bd_holder from the block device
843 *
844 * @bdev: struct block device to be searched
845 * @bo: target struct bd_holder
846 *
847 * Returns matching entry with @bo in @bdev->bd_holder_list.
848 * If found, increment the reference count and return the pointer.
849 * If not found, returns NULL.
850 */
851static struct bd_holder *find_bd_holder(struct block_device *bdev,
852 struct bd_holder *bo)
853{
854 struct bd_holder *tmp;
855
856 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
857 if (tmp->sdir == bo->sdir) {
858 tmp->count++;
859 return tmp;
860 }
861
862 return NULL;
863}
864
865/**
866 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
867 *
868 * @bdev: block device to be bd_claimed
869 * @bo: preallocated and initialized by alloc_bd_holder()
870 *
871 * Add @bo to @bdev->bd_holder_list, create symlinks.
872 *
873 * Returns 0 if symlinks are created.
874 * Returns -ve if something fails.
875 */
876static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
877{
878 int ret;
879
880 if (!bo)
881 return -EINVAL;
882
883 if (!bd_holder_grab_dirs(bdev, bo))
884 return -EBUSY;
885
886 ret = add_symlink(bo->sdir, bo->sdev);
887 if (ret == 0) {
888 ret = add_symlink(bo->hdir, bo->hdev);
889 if (ret)
890 del_symlink(bo->sdir, bo->sdev);
891 }
892 if (ret == 0)
893 list_add_tail(&bo->list, &bdev->bd_holder_list);
894 return ret;
895}
896
897/**
898 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
899 *
900 * @bdev: block device to be bd_claimed
901 * @kobj: holder's kobject
902 *
903 * If there is matching entry with @kobj in @bdev->bd_holder_list
904 * and no other bd_claim() from the same kobject,
905 * remove the struct bd_holder from the list, delete symlinks for it.
906 *
907 * Returns a pointer to the struct bd_holder when it's removed from the list
908 * and ready to be freed.
909 * Returns NULL if matching claim isn't found or there is other bd_claim()
910 * by the same kobject.
911 */
912static struct bd_holder *del_bd_holder(struct block_device *bdev,
913 struct kobject *kobj)
914{
915 struct bd_holder *bo;
916
917 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
918 if (bo->sdir == kobj) {
919 bo->count--;
920 BUG_ON(bo->count < 0);
921 if (!bo->count) {
922 list_del(&bo->list);
923 del_symlink(bo->sdir, bo->sdev);
924 del_symlink(bo->hdir, bo->hdev);
925 bd_holder_release_dirs(bo);
926 return bo;
927 }
928 break;
929 }
930 }
931
932 return NULL;
933}
934
935/**
936 * bd_claim_by_kobject - bd_claim() with additional kobject signature
937 *
938 * @bdev: block device to be claimed
939 * @holder: holder's signature
940 * @kobj: holder's kobject
941 *
942 * Do bd_claim() and if it succeeds, create sysfs symlinks between
943 * the bdev and the holder's kobject.
944 * Use bd_release_from_kobject() when relesing the claimed bdev.
945 *
946 * Returns 0 on success. (same as bd_claim())
947 * Returns errno on failure.
948 */
949static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
950 struct kobject *kobj)
951{
952 int res;
953 struct bd_holder *bo, *found;
954
955 if (!kobj)
956 return -EINVAL;
957
958 bo = alloc_bd_holder(kobj);
959 if (!bo)
960 return -ENOMEM;
961
962 mutex_lock(&bdev->bd_mutex);
963 res = bd_claim(bdev, holder);
964 if (res == 0) {
965 found = find_bd_holder(bdev, bo);
966 if (found == NULL) {
967 res = add_bd_holder(bdev, bo);
968 if (res)
969 bd_release(bdev);
970 }
971 }
972
973 if (res || found)
974 free_bd_holder(bo);
975 mutex_unlock(&bdev->bd_mutex);
976
977 return res;
978}
979
980/**
981 * bd_release_from_kobject - bd_release() with additional kobject signature
982 *
983 * @bdev: block device to be released
984 * @kobj: holder's kobject
985 *
986 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
987 */
988static void bd_release_from_kobject(struct block_device *bdev,
989 struct kobject *kobj)
990{
991 struct bd_holder *bo;
992
993 if (!kobj)
994 return;
995
996 mutex_lock(&bdev->bd_mutex);
997 bd_release(bdev);
998 if ((bo = del_bd_holder(bdev, kobj)))
999 free_bd_holder(bo);
1000 mutex_unlock(&bdev->bd_mutex);
1001}
1002
1003/**
1004 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1005 *
1006 * @bdev: block device to be claimed
1007 * @holder: holder's signature
1008 * @disk: holder's gendisk
1009 *
1010 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1011 */
1012int bd_claim_by_disk(struct block_device *bdev, void *holder,
1013 struct gendisk *disk)
1014{
1015 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1016}
1017EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1018
1019/**
1020 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1021 *
1022 * @bdev: block device to be claimed
1023 * @disk: holder's gendisk
1024 *
1025 * Call bd_release_from_kobject() and put @disk->slave_dir.
1026 */
1027void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1028{
1029 bd_release_from_kobject(bdev, disk->slave_dir);
1030 kobject_put(disk->slave_dir);
1031}
1032EXPORT_SYMBOL_GPL(bd_release_from_disk);
1033#endif
1034
1035/*
1036 * Tries to open block device by device number. Use it ONLY if you
1037 * really do not have anything better - i.e. when you are behind a
1038 * truly sucky interface and all you are given is a device number. _Never_
1039 * to be used for internal purposes. If you ever need it - reconsider
1040 * your API.
1041 */
1042struct block_device *open_by_devnum(dev_t dev, unsigned mode)
1043{
1044 struct block_device *bdev = bdget(dev);
1045 int err = -ENOMEM;
1046 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
1047 if (bdev)
1048 err = blkdev_get(bdev, mode, flags);
1049 return err ? ERR_PTR(err) : bdev;
1050}
1051
1052EXPORT_SYMBOL(open_by_devnum);
1053
1054/*
1055 * This routine checks whether a removable media has been changed,
1056 * and invalidates all buffer-cache-entries in that case. This
1057 * is a relatively slow routine, so we have to try to minimize using
1058 * it. Thus it is called only upon a 'mount' or 'open'. This
1059 * is the best way of combining speed and utility, I think.
1060 * People changing diskettes in the middle of an operation deserve
1061 * to lose :-)
1062 */
1063int check_disk_change(struct block_device *bdev)
1064{
1065 struct gendisk *disk = bdev->bd_disk;
1066 struct block_device_operations * bdops = disk->fops;
1067
1068 if (!bdops->media_changed)
1069 return 0;
1070 if (!bdops->media_changed(bdev->bd_disk))
1071 return 0;
1072
1073 if (__invalidate_device(bdev))
1074 printk("VFS: busy inodes on changed media.\n");
1075
1076 if (bdops->revalidate_disk)
1077 bdops->revalidate_disk(bdev->bd_disk);
1078 if (bdev->bd_disk->minors > 1)
1079 bdev->bd_invalidated = 1;
1080 return 1;
1081}
1082
1083EXPORT_SYMBOL(check_disk_change);
1084
1085void bd_set_size(struct block_device *bdev, loff_t size)
1086{
1087 unsigned bsize = bdev_hardsect_size(bdev);
1088
1089 bdev->bd_inode->i_size = size;
1090 while (bsize < PAGE_CACHE_SIZE) {
1091 if (size & bsize)
1092 break;
1093 bsize <<= 1;
1094 }
1095 bdev->bd_block_size = bsize;
1096 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1097}
1098EXPORT_SYMBOL(bd_set_size);
1099
1100static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1101 int for_part);
1102static int __blkdev_put(struct block_device *bdev, int for_part);
1103
1104/*
1105 * bd_mutex locking:
1106 *
1107 * mutex_lock(part->bd_mutex)
1108 * mutex_lock_nested(whole->bd_mutex, 1)
1109 */
1110
1111static int do_open(struct block_device *bdev, struct file *file, int for_part)
1112{
1113 struct module *owner = NULL;
1114 struct gendisk *disk;
1115 int ret = -ENXIO;
1116 int part;
1117
1118 file->f_mapping = bdev->bd_inode->i_mapping;
1119 lock_kernel();
1120 disk = get_gendisk(bdev->bd_dev, &part);
1121 if (!disk) {
1122 unlock_kernel();
1123 bdput(bdev);
1124 return ret;
1125 }
1126 owner = disk->fops->owner;
1127
1128 mutex_lock_nested(&bdev->bd_mutex, for_part);
1129 if (!bdev->bd_openers) {
1130 bdev->bd_disk = disk;
1131 bdev->bd_contains = bdev;
1132 if (!part) {
1133 struct backing_dev_info *bdi;
1134 if (disk->fops->open) {
1135 ret = disk->fops->open(bdev->bd_inode, file);
1136 if (ret)
1137 goto out_first;
1138 }
1139 if (!bdev->bd_openers) {
1140 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1141 bdi = blk_get_backing_dev_info(bdev);
1142 if (bdi == NULL)
1143 bdi = &default_backing_dev_info;
1144 bdev->bd_inode->i_data.backing_dev_info = bdi;
1145 }
1146 if (bdev->bd_invalidated)
1147 rescan_partitions(disk, bdev);
1148 } else {
1149 struct hd_struct *p;
1150 struct block_device *whole;
1151 whole = bdget_disk(disk, 0);
1152 ret = -ENOMEM;
1153 if (!whole)
1154 goto out_first;
1155 BUG_ON(for_part);
1156 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
1157 if (ret)
1158 goto out_first;
1159 bdev->bd_contains = whole;
1160 p = disk->part[part - 1];
1161 bdev->bd_inode->i_data.backing_dev_info =
1162 whole->bd_inode->i_data.backing_dev_info;
1163 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1164 ret = -ENXIO;
1165 goto out_first;
1166 }
1167 kobject_get(&p->kobj);
1168 bdev->bd_part = p;
1169 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1170 }
1171 } else {
1172 put_disk(disk);
1173 module_put(owner);
1174 if (bdev->bd_contains == bdev) {
1175 if (bdev->bd_disk->fops->open) {
1176 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1177 if (ret)
1178 goto out;
1179 }
1180 if (bdev->bd_invalidated)
1181 rescan_partitions(bdev->bd_disk, bdev);
1182 }
1183 }
1184 bdev->bd_openers++;
1185 if (for_part)
1186 bdev->bd_part_count++;
1187 mutex_unlock(&bdev->bd_mutex);
1188 unlock_kernel();
1189 return 0;
1190
1191out_first:
1192 bdev->bd_disk = NULL;
1193 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1194 if (bdev != bdev->bd_contains)
1195 __blkdev_put(bdev->bd_contains, 1);
1196 bdev->bd_contains = NULL;
1197 put_disk(disk);
1198 module_put(owner);
1199out:
1200 mutex_unlock(&bdev->bd_mutex);
1201 unlock_kernel();
1202 if (ret)
1203 bdput(bdev);
1204 return ret;
1205}
1206
1207static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1208 int for_part)
1209{
1210 /*
1211 * This crockload is due to bad choice of ->open() type.
1212 * It will go away.
1213 * For now, block device ->open() routine must _not_
1214 * examine anything in 'inode' argument except ->i_rdev.
1215 */
1216 struct file fake_file = {};
1217 struct dentry fake_dentry = {};
1218 fake_file.f_mode = mode;
1219 fake_file.f_flags = flags;
1220 fake_file.f_path.dentry = &fake_dentry;
1221 fake_dentry.d_inode = bdev->bd_inode;
1222
1223 return do_open(bdev, &fake_file, for_part);
1224}
1225
1226int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1227{
1228 return __blkdev_get(bdev, mode, flags, 0);
1229}
1230EXPORT_SYMBOL(blkdev_get);
1231
1232static int blkdev_open(struct inode * inode, struct file * filp)
1233{
1234 struct block_device *bdev;
1235 int res;
1236
1237 /*
1238 * Preserve backwards compatibility and allow large file access
1239 * even if userspace doesn't ask for it explicitly. Some mkfs
1240 * binary needs it. We might want to drop this workaround
1241 * during an unstable branch.
1242 */
1243 filp->f_flags |= O_LARGEFILE;
1244
1245 bdev = bd_acquire(inode);
1246 if (bdev == NULL)
1247 return -ENOMEM;
1248
1249 res = do_open(bdev, filp, 0);
1250 if (res)
1251 return res;
1252
1253 if (!(filp->f_flags & O_EXCL) )
1254 return 0;
1255
1256 if (!(res = bd_claim(bdev, filp)))
1257 return 0;
1258
1259 blkdev_put(bdev);
1260 return res;
1261}
1262
1263static int __blkdev_put(struct block_device *bdev, int for_part)
1264{
1265 int ret = 0;
1266 struct inode *bd_inode = bdev->bd_inode;
1267 struct gendisk *disk = bdev->bd_disk;
1268 struct block_device *victim = NULL;
1269
1270 mutex_lock_nested(&bdev->bd_mutex, for_part);
1271 lock_kernel();
1272 if (for_part)
1273 bdev->bd_part_count--;
1274
1275 if (!--bdev->bd_openers) {
1276 sync_blockdev(bdev);
1277 kill_bdev(bdev);
1278 }
1279 if (bdev->bd_contains == bdev) {
1280 if (disk->fops->release)
1281 ret = disk->fops->release(bd_inode, NULL);
1282 }
1283 if (!bdev->bd_openers) {
1284 struct module *owner = disk->fops->owner;
1285
1286 put_disk(disk);
1287 module_put(owner);
1288
1289 if (bdev->bd_contains != bdev) {
1290 kobject_put(&bdev->bd_part->kobj);
1291 bdev->bd_part = NULL;
1292 }
1293 bdev->bd_disk = NULL;
1294 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1295 if (bdev != bdev->bd_contains)
1296 victim = bdev->bd_contains;
1297 bdev->bd_contains = NULL;
1298 }
1299 unlock_kernel();
1300 mutex_unlock(&bdev->bd_mutex);
1301 bdput(bdev);
1302 if (victim)
1303 __blkdev_put(victim, 1);
1304 return ret;
1305}
1306
1307int blkdev_put(struct block_device *bdev)
1308{
1309 return __blkdev_put(bdev, 0);
1310}
1311EXPORT_SYMBOL(blkdev_put);
1312
1313static int blkdev_close(struct inode * inode, struct file * filp)
1314{
1315 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1316 if (bdev->bd_holder == filp)
1317 bd_release(bdev);
1318 return blkdev_put(bdev);
1319}
1320
1321static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1322{
1323 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1324}
1325
1326const struct address_space_operations def_blk_aops = {
1327 .readpage = blkdev_readpage,
1328 .writepage = blkdev_writepage,
1329 .sync_page = block_sync_page,
1330 .prepare_write = blkdev_prepare_write,
1331 .commit_write = blkdev_commit_write,
1332 .writepages = generic_writepages,
1333 .direct_IO = blkdev_direct_IO,
1334};
1335
1336const struct file_operations def_blk_fops = {
1337 .open = blkdev_open,
1338 .release = blkdev_close,
1339 .llseek = block_llseek,
1340 .read = do_sync_read,
1341 .write = do_sync_write,
1342 .aio_read = generic_file_aio_read,
1343 .aio_write = generic_file_aio_write_nolock,
1344 .mmap = generic_file_mmap,
1345 .fsync = block_fsync,
1346 .unlocked_ioctl = block_ioctl,
1347#ifdef CONFIG_COMPAT
1348 .compat_ioctl = compat_blkdev_ioctl,
1349#endif
1350 .sendfile = generic_file_sendfile,
1351 .splice_read = generic_file_splice_read,
1352 .splice_write = generic_file_splice_write,
1353};
1354
1355int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1356{
1357 int res;
1358 mm_segment_t old_fs = get_fs();
1359 set_fs(KERNEL_DS);
1360 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1361 set_fs(old_fs);
1362 return res;
1363}
1364
1365EXPORT_SYMBOL(ioctl_by_bdev);
1366
1367/**
1368 * lookup_bdev - lookup a struct block_device by name
1369 *
1370 * @path: special file representing the block device
1371 *
1372 * Get a reference to the blockdevice at @path in the current
1373 * namespace if possible and return it. Return ERR_PTR(error)
1374 * otherwise.
1375 */
1376struct block_device *lookup_bdev(const char *path)
1377{
1378 struct block_device *bdev;
1379 struct inode *inode;
1380 struct nameidata nd;
1381 int error;
1382
1383 if (!path || !*path)
1384 return ERR_PTR(-EINVAL);
1385
1386 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1387 if (error)
1388 return ERR_PTR(error);
1389
1390 inode = nd.dentry->d_inode;
1391 error = -ENOTBLK;
1392 if (!S_ISBLK(inode->i_mode))
1393 goto fail;
1394 error = -EACCES;
1395 if (nd.mnt->mnt_flags & MNT_NODEV)
1396 goto fail;
1397 error = -ENOMEM;
1398 bdev = bd_acquire(inode);
1399 if (!bdev)
1400 goto fail;
1401out:
1402 path_release(&nd);
1403 return bdev;
1404fail:
1405 bdev = ERR_PTR(error);
1406 goto out;
1407}
1408
1409/**
1410 * open_bdev_excl - open a block device by name and set it up for use
1411 *
1412 * @path: special file representing the block device
1413 * @flags: %MS_RDONLY for opening read-only
1414 * @holder: owner for exclusion
1415 *
1416 * Open the blockdevice described by the special file at @path, claim it
1417 * for the @holder.
1418 */
1419struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1420{
1421 struct block_device *bdev;
1422 mode_t mode = FMODE_READ;
1423 int error = 0;
1424
1425 bdev = lookup_bdev(path);
1426 if (IS_ERR(bdev))
1427 return bdev;
1428
1429 if (!(flags & MS_RDONLY))
1430 mode |= FMODE_WRITE;
1431 error = blkdev_get(bdev, mode, 0);
1432 if (error)
1433 return ERR_PTR(error);
1434 error = -EACCES;
1435 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1436 goto blkdev_put;
1437 error = bd_claim(bdev, holder);
1438 if (error)
1439 goto blkdev_put;
1440
1441 return bdev;
1442
1443blkdev_put:
1444 blkdev_put(bdev);
1445 return ERR_PTR(error);
1446}
1447
1448EXPORT_SYMBOL(open_bdev_excl);
1449
1450/**
1451 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1452 *
1453 * @bdev: blockdevice to close
1454 *
1455 * This is the counterpart to open_bdev_excl().
1456 */
1457void close_bdev_excl(struct block_device *bdev)
1458{
1459 bd_release(bdev);
1460 blkdev_put(bdev);
1461}
1462
1463EXPORT_SYMBOL(close_bdev_excl);
1464
1465int __invalidate_device(struct block_device *bdev)
1466{
1467 struct super_block *sb = get_super(bdev);
1468 int res = 0;
1469
1470 if (sb) {
1471 /*
1472 * no need to lock the super, get_super holds the
1473 * read mutex so the filesystem cannot go away
1474 * under us (->put_super runs with the write lock
1475 * hold).
1476 */
1477 shrink_dcache_sb(sb);
1478 res = invalidate_inodes(sb);
1479 drop_super(sb);
1480 }
1481 invalidate_bdev(bdev, 0);
1482 return res;
1483}
1484EXPORT_SYMBOL(__invalidate_device);