[PATCH] remove the old bd_mutex lockdep annotation
[deliverable/linux.git] / fs / block_dev.c
1 /*
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/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
28 struct bdev_inode {
29 struct block_device bdev;
30 struct inode vfs_inode;
31 };
32
33 static inline struct bdev_inode *BDEV_I(struct inode *inode)
34 {
35 return container_of(inode, struct bdev_inode, vfs_inode);
36 }
37
38 inline struct block_device *I_BDEV(struct inode *inode)
39 {
40 return &BDEV_I(inode)->bdev;
41 }
42
43 EXPORT_SYMBOL(I_BDEV);
44
45 static 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.. */
59 static 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
65 int 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
85 EXPORT_SYMBOL(set_blocksize);
86
87 int 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
98 EXPORT_SYMBOL(sb_set_blocksize);
99
100 int 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
108 EXPORT_SYMBOL(sb_min_blocksize);
109
110 static int
111 blkdev_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
132 static int
133 blkdev_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
160 static ssize_t
161 blkdev_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 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
172 {
173 return block_write_full_page(page, blkdev_get_block, wbc);
174 }
175
176 static int blkdev_readpage(struct file * file, struct page * page)
177 {
178 return block_read_full_page(page, blkdev_get_block);
179 }
180
181 static 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
186 static 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 */
196 static 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
202 mutex_lock(&bd_inode->i_mutex);
203 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 }
219 mutex_unlock(&bd_inode->i_mutex);
220 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
228 static 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
237 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
238 static struct kmem_cache * bdev_cachep __read_mostly;
239
240 static struct inode *bdev_alloc_inode(struct super_block *sb)
241 {
242 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
243 if (!ei)
244 return NULL;
245 return &ei->vfs_inode;
246 }
247
248 static 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
256 static void init_once(void * foo, struct kmem_cache * 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));
265 mutex_init(&bdev->bd_mutex);
266 mutex_init(&bdev->bd_mount_mutex);
267 INIT_LIST_HEAD(&bdev->bd_inodes);
268 INIT_LIST_HEAD(&bdev->bd_list);
269 #ifdef CONFIG_SYSFS
270 INIT_LIST_HEAD(&bdev->bd_holder_list);
271 #endif
272 inode_init_once(&ei->vfs_inode);
273 }
274 }
275
276 static 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
283 static 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
295 static 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
303 static int bd_get_sb(struct file_system_type *fs_type,
304 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
305 {
306 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
307 }
308
309 static struct file_system_type bd_type = {
310 .name = "bdev",
311 .get_sb = bd_get_sb,
312 .kill_sb = kill_anon_super,
313 };
314
315 static struct vfsmount *bd_mnt __read_mostly;
316 struct super_block *blockdev_superblock;
317
318 void __init bdev_cache_init(void)
319 {
320 int err;
321 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
322 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
323 SLAB_MEM_SPREAD|SLAB_PANIC),
324 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 */
340 static inline unsigned long hash(dev_t dev)
341 {
342 return MAJOR(dev)+MINOR(dev);
343 }
344
345 static int bdev_test(struct inode *inode, void *data)
346 {
347 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
348 }
349
350 static int bdev_set(struct inode *inode, void *data)
351 {
352 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
353 return 0;
354 }
355
356 static LIST_HEAD(all_bdevs);
357
358 struct 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
391 EXPORT_SYMBOL(bdget);
392
393 long 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
407 void bdput(struct block_device *bdev)
408 {
409 iput(bdev->bd_inode);
410 }
411
412 EXPORT_SYMBOL(bdput);
413
414 static struct block_device *bd_acquire(struct inode *inode)
415 {
416 struct block_device *bdev;
417
418 spin_lock(&bdev_lock);
419 bdev = inode->i_bdev;
420 if (bdev) {
421 atomic_inc(&bdev->bd_inode->i_count);
422 spin_unlock(&bdev_lock);
423 return bdev;
424 }
425 spin_unlock(&bdev_lock);
426
427 bdev = bdget(inode->i_rdev);
428 if (bdev) {
429 spin_lock(&bdev_lock);
430 if (!inode->i_bdev) {
431 /*
432 * We take an additional bd_inode->i_count for inode,
433 * and it's released in clear_inode() of inode.
434 * So, we can access it via ->i_mapping always
435 * without igrab().
436 */
437 atomic_inc(&bdev->bd_inode->i_count);
438 inode->i_bdev = bdev;
439 inode->i_mapping = bdev->bd_inode->i_mapping;
440 list_add(&inode->i_devices, &bdev->bd_inodes);
441 }
442 spin_unlock(&bdev_lock);
443 }
444 return bdev;
445 }
446
447 /* Call when you free inode */
448
449 void bd_forget(struct inode *inode)
450 {
451 struct block_device *bdev = NULL;
452
453 spin_lock(&bdev_lock);
454 if (inode->i_bdev) {
455 if (inode->i_sb != blockdev_superblock)
456 bdev = inode->i_bdev;
457 __bd_forget(inode);
458 }
459 spin_unlock(&bdev_lock);
460
461 if (bdev)
462 iput(bdev->bd_inode);
463 }
464
465 int bd_claim(struct block_device *bdev, void *holder)
466 {
467 int res;
468 spin_lock(&bdev_lock);
469
470 /* first decide result */
471 if (bdev->bd_holder == holder)
472 res = 0; /* already a holder */
473 else if (bdev->bd_holder != NULL)
474 res = -EBUSY; /* held by someone else */
475 else if (bdev->bd_contains == bdev)
476 res = 0; /* is a whole device which isn't held */
477
478 else if (bdev->bd_contains->bd_holder == bd_claim)
479 res = 0; /* is a partition of a device that is being partitioned */
480 else if (bdev->bd_contains->bd_holder != NULL)
481 res = -EBUSY; /* is a partition of a held device */
482 else
483 res = 0; /* is a partition of an un-held device */
484
485 /* now impose change */
486 if (res==0) {
487 /* note that for a whole device bd_holders
488 * will be incremented twice, and bd_holder will
489 * be set to bd_claim before being set to holder
490 */
491 bdev->bd_contains->bd_holders ++;
492 bdev->bd_contains->bd_holder = bd_claim;
493 bdev->bd_holders++;
494 bdev->bd_holder = holder;
495 }
496 spin_unlock(&bdev_lock);
497 return res;
498 }
499
500 EXPORT_SYMBOL(bd_claim);
501
502 void bd_release(struct block_device *bdev)
503 {
504 spin_lock(&bdev_lock);
505 if (!--bdev->bd_contains->bd_holders)
506 bdev->bd_contains->bd_holder = NULL;
507 if (!--bdev->bd_holders)
508 bdev->bd_holder = NULL;
509 spin_unlock(&bdev_lock);
510 }
511
512 EXPORT_SYMBOL(bd_release);
513
514 #ifdef CONFIG_SYSFS
515 /*
516 * Functions for bd_claim_by_kobject / bd_release_from_kobject
517 *
518 * If a kobject is passed to bd_claim_by_kobject()
519 * and the kobject has a parent directory,
520 * following symlinks are created:
521 * o from the kobject to the claimed bdev
522 * o from "holders" directory of the bdev to the parent of the kobject
523 * bd_release_from_kobject() removes these symlinks.
524 *
525 * Example:
526 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
527 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
528 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
529 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
530 */
531
532 static struct kobject *bdev_get_kobj(struct block_device *bdev)
533 {
534 if (bdev->bd_contains != bdev)
535 return kobject_get(&bdev->bd_part->kobj);
536 else
537 return kobject_get(&bdev->bd_disk->kobj);
538 }
539
540 static struct kobject *bdev_get_holder(struct block_device *bdev)
541 {
542 if (bdev->bd_contains != bdev)
543 return kobject_get(bdev->bd_part->holder_dir);
544 else
545 return kobject_get(bdev->bd_disk->holder_dir);
546 }
547
548 static int add_symlink(struct kobject *from, struct kobject *to)
549 {
550 if (!from || !to)
551 return 0;
552 return sysfs_create_link(from, to, kobject_name(to));
553 }
554
555 static void del_symlink(struct kobject *from, struct kobject *to)
556 {
557 if (!from || !to)
558 return;
559 sysfs_remove_link(from, kobject_name(to));
560 }
561
562 /*
563 * 'struct bd_holder' contains pointers to kobjects symlinked by
564 * bd_claim_by_kobject.
565 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
566 */
567 struct bd_holder {
568 struct list_head list; /* chain of holders of the bdev */
569 int count; /* references from the holder */
570 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
571 struct kobject *hdev; /* e.g. "/block/dm-0" */
572 struct kobject *hdir; /* e.g. "/block/sda/holders" */
573 struct kobject *sdev; /* e.g. "/block/sda" */
574 };
575
576 /*
577 * Get references of related kobjects at once.
578 * Returns 1 on success. 0 on failure.
579 *
580 * Should call bd_holder_release_dirs() after successful use.
581 */
582 static int bd_holder_grab_dirs(struct block_device *bdev,
583 struct bd_holder *bo)
584 {
585 if (!bdev || !bo)
586 return 0;
587
588 bo->sdir = kobject_get(bo->sdir);
589 if (!bo->sdir)
590 return 0;
591
592 bo->hdev = kobject_get(bo->sdir->parent);
593 if (!bo->hdev)
594 goto fail_put_sdir;
595
596 bo->sdev = bdev_get_kobj(bdev);
597 if (!bo->sdev)
598 goto fail_put_hdev;
599
600 bo->hdir = bdev_get_holder(bdev);
601 if (!bo->hdir)
602 goto fail_put_sdev;
603
604 return 1;
605
606 fail_put_sdev:
607 kobject_put(bo->sdev);
608 fail_put_hdev:
609 kobject_put(bo->hdev);
610 fail_put_sdir:
611 kobject_put(bo->sdir);
612
613 return 0;
614 }
615
616 /* Put references of related kobjects at once. */
617 static void bd_holder_release_dirs(struct bd_holder *bo)
618 {
619 kobject_put(bo->hdir);
620 kobject_put(bo->sdev);
621 kobject_put(bo->hdev);
622 kobject_put(bo->sdir);
623 }
624
625 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
626 {
627 struct bd_holder *bo;
628
629 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
630 if (!bo)
631 return NULL;
632
633 bo->count = 1;
634 bo->sdir = kobj;
635
636 return bo;
637 }
638
639 static void free_bd_holder(struct bd_holder *bo)
640 {
641 kfree(bo);
642 }
643
644 /**
645 * find_bd_holder - find matching struct bd_holder from the block device
646 *
647 * @bdev: struct block device to be searched
648 * @bo: target struct bd_holder
649 *
650 * Returns matching entry with @bo in @bdev->bd_holder_list.
651 * If found, increment the reference count and return the pointer.
652 * If not found, returns NULL.
653 */
654 static struct bd_holder *find_bd_holder(struct block_device *bdev,
655 struct bd_holder *bo)
656 {
657 struct bd_holder *tmp;
658
659 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
660 if (tmp->sdir == bo->sdir) {
661 tmp->count++;
662 return tmp;
663 }
664
665 return NULL;
666 }
667
668 /**
669 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
670 *
671 * @bdev: block device to be bd_claimed
672 * @bo: preallocated and initialized by alloc_bd_holder()
673 *
674 * Add @bo to @bdev->bd_holder_list, create symlinks.
675 *
676 * Returns 0 if symlinks are created.
677 * Returns -ve if something fails.
678 */
679 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
680 {
681 int ret;
682
683 if (!bo)
684 return -EINVAL;
685
686 if (!bd_holder_grab_dirs(bdev, bo))
687 return -EBUSY;
688
689 ret = add_symlink(bo->sdir, bo->sdev);
690 if (ret == 0) {
691 ret = add_symlink(bo->hdir, bo->hdev);
692 if (ret)
693 del_symlink(bo->sdir, bo->sdev);
694 }
695 if (ret == 0)
696 list_add_tail(&bo->list, &bdev->bd_holder_list);
697 return ret;
698 }
699
700 /**
701 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
702 *
703 * @bdev: block device to be bd_claimed
704 * @kobj: holder's kobject
705 *
706 * If there is matching entry with @kobj in @bdev->bd_holder_list
707 * and no other bd_claim() from the same kobject,
708 * remove the struct bd_holder from the list, delete symlinks for it.
709 *
710 * Returns a pointer to the struct bd_holder when it's removed from the list
711 * and ready to be freed.
712 * Returns NULL if matching claim isn't found or there is other bd_claim()
713 * by the same kobject.
714 */
715 static struct bd_holder *del_bd_holder(struct block_device *bdev,
716 struct kobject *kobj)
717 {
718 struct bd_holder *bo;
719
720 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
721 if (bo->sdir == kobj) {
722 bo->count--;
723 BUG_ON(bo->count < 0);
724 if (!bo->count) {
725 list_del(&bo->list);
726 del_symlink(bo->sdir, bo->sdev);
727 del_symlink(bo->hdir, bo->hdev);
728 bd_holder_release_dirs(bo);
729 return bo;
730 }
731 break;
732 }
733 }
734
735 return NULL;
736 }
737
738 /**
739 * bd_claim_by_kobject - bd_claim() with additional kobject signature
740 *
741 * @bdev: block device to be claimed
742 * @holder: holder's signature
743 * @kobj: holder's kobject
744 *
745 * Do bd_claim() and if it succeeds, create sysfs symlinks between
746 * the bdev and the holder's kobject.
747 * Use bd_release_from_kobject() when relesing the claimed bdev.
748 *
749 * Returns 0 on success. (same as bd_claim())
750 * Returns errno on failure.
751 */
752 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
753 struct kobject *kobj)
754 {
755 int res;
756 struct bd_holder *bo, *found;
757
758 if (!kobj)
759 return -EINVAL;
760
761 bo = alloc_bd_holder(kobj);
762 if (!bo)
763 return -ENOMEM;
764
765 mutex_lock(&bdev->bd_mutex);
766 res = bd_claim(bdev, holder);
767 if (res == 0) {
768 found = find_bd_holder(bdev, bo);
769 if (found == NULL) {
770 res = add_bd_holder(bdev, bo);
771 if (res)
772 bd_release(bdev);
773 }
774 }
775
776 if (res || found)
777 free_bd_holder(bo);
778 mutex_unlock(&bdev->bd_mutex);
779
780 return res;
781 }
782
783 /**
784 * bd_release_from_kobject - bd_release() with additional kobject signature
785 *
786 * @bdev: block device to be released
787 * @kobj: holder's kobject
788 *
789 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
790 */
791 static void bd_release_from_kobject(struct block_device *bdev,
792 struct kobject *kobj)
793 {
794 struct bd_holder *bo;
795
796 if (!kobj)
797 return;
798
799 mutex_lock(&bdev->bd_mutex);
800 bd_release(bdev);
801 if ((bo = del_bd_holder(bdev, kobj)))
802 free_bd_holder(bo);
803 mutex_unlock(&bdev->bd_mutex);
804 }
805
806 /**
807 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
808 *
809 * @bdev: block device to be claimed
810 * @holder: holder's signature
811 * @disk: holder's gendisk
812 *
813 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
814 */
815 int bd_claim_by_disk(struct block_device *bdev, void *holder,
816 struct gendisk *disk)
817 {
818 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
819 }
820 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
821
822 /**
823 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
824 *
825 * @bdev: block device to be claimed
826 * @disk: holder's gendisk
827 *
828 * Call bd_release_from_kobject() and put @disk->slave_dir.
829 */
830 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
831 {
832 bd_release_from_kobject(bdev, disk->slave_dir);
833 kobject_put(disk->slave_dir);
834 }
835 EXPORT_SYMBOL_GPL(bd_release_from_disk);
836 #endif
837
838 /*
839 * Tries to open block device by device number. Use it ONLY if you
840 * really do not have anything better - i.e. when you are behind a
841 * truly sucky interface and all you are given is a device number. _Never_
842 * to be used for internal purposes. If you ever need it - reconsider
843 * your API.
844 */
845 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
846 {
847 struct block_device *bdev = bdget(dev);
848 int err = -ENOMEM;
849 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
850 if (bdev)
851 err = blkdev_get(bdev, mode, flags);
852 return err ? ERR_PTR(err) : bdev;
853 }
854
855 EXPORT_SYMBOL(open_by_devnum);
856
857 /*
858 * This routine checks whether a removable media has been changed,
859 * and invalidates all buffer-cache-entries in that case. This
860 * is a relatively slow routine, so we have to try to minimize using
861 * it. Thus it is called only upon a 'mount' or 'open'. This
862 * is the best way of combining speed and utility, I think.
863 * People changing diskettes in the middle of an operation deserve
864 * to lose :-)
865 */
866 int check_disk_change(struct block_device *bdev)
867 {
868 struct gendisk *disk = bdev->bd_disk;
869 struct block_device_operations * bdops = disk->fops;
870
871 if (!bdops->media_changed)
872 return 0;
873 if (!bdops->media_changed(bdev->bd_disk))
874 return 0;
875
876 if (__invalidate_device(bdev))
877 printk("VFS: busy inodes on changed media.\n");
878
879 if (bdops->revalidate_disk)
880 bdops->revalidate_disk(bdev->bd_disk);
881 if (bdev->bd_disk->minors > 1)
882 bdev->bd_invalidated = 1;
883 return 1;
884 }
885
886 EXPORT_SYMBOL(check_disk_change);
887
888 void bd_set_size(struct block_device *bdev, loff_t size)
889 {
890 unsigned bsize = bdev_hardsect_size(bdev);
891
892 bdev->bd_inode->i_size = size;
893 while (bsize < PAGE_CACHE_SIZE) {
894 if (size & bsize)
895 break;
896 bsize <<= 1;
897 }
898 bdev->bd_block_size = bsize;
899 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
900 }
901 EXPORT_SYMBOL(bd_set_size);
902
903 static int do_open(struct block_device *bdev, struct file *file)
904 {
905 struct module *owner = NULL;
906 struct gendisk *disk;
907 int ret = -ENXIO;
908 int part;
909
910 file->f_mapping = bdev->bd_inode->i_mapping;
911 lock_kernel();
912 disk = get_gendisk(bdev->bd_dev, &part);
913 if (!disk) {
914 unlock_kernel();
915 bdput(bdev);
916 return ret;
917 }
918 owner = disk->fops->owner;
919
920 mutex_lock(&bdev->bd_mutex);
921 if (!bdev->bd_openers) {
922 bdev->bd_disk = disk;
923 bdev->bd_contains = bdev;
924 if (!part) {
925 struct backing_dev_info *bdi;
926 if (disk->fops->open) {
927 ret = disk->fops->open(bdev->bd_inode, file);
928 if (ret)
929 goto out_first;
930 }
931 if (!bdev->bd_openers) {
932 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
933 bdi = blk_get_backing_dev_info(bdev);
934 if (bdi == NULL)
935 bdi = &default_backing_dev_info;
936 bdev->bd_inode->i_data.backing_dev_info = bdi;
937 }
938 if (bdev->bd_invalidated)
939 rescan_partitions(disk, bdev);
940 } else {
941 struct hd_struct *p;
942 struct block_device *whole;
943 whole = bdget_disk(disk, 0);
944 ret = -ENOMEM;
945 if (!whole)
946 goto out_first;
947 ret = blkdev_get(whole, file->f_mode, file->f_flags);
948 if (ret)
949 goto out_first;
950 bdev->bd_contains = whole;
951 mutex_lock(&whole->bd_mutex);
952 whole->bd_part_count++;
953 p = disk->part[part - 1];
954 bdev->bd_inode->i_data.backing_dev_info =
955 whole->bd_inode->i_data.backing_dev_info;
956 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
957 whole->bd_part_count--;
958 mutex_unlock(&whole->bd_mutex);
959 ret = -ENXIO;
960 goto out_first;
961 }
962 kobject_get(&p->kobj);
963 bdev->bd_part = p;
964 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
965 mutex_unlock(&whole->bd_mutex);
966 }
967 } else {
968 put_disk(disk);
969 module_put(owner);
970 if (bdev->bd_contains == bdev) {
971 if (bdev->bd_disk->fops->open) {
972 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
973 if (ret)
974 goto out;
975 }
976 if (bdev->bd_invalidated)
977 rescan_partitions(bdev->bd_disk, bdev);
978 } else {
979 mutex_lock(&bdev->bd_contains->bd_mutex);
980 bdev->bd_contains->bd_part_count++;
981 mutex_unlock(&bdev->bd_contains->bd_mutex);
982 }
983 }
984 bdev->bd_openers++;
985 mutex_unlock(&bdev->bd_mutex);
986 unlock_kernel();
987 return 0;
988
989 out_first:
990 bdev->bd_disk = NULL;
991 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
992 if (bdev != bdev->bd_contains)
993 blkdev_put(bdev->bd_contains);
994 bdev->bd_contains = NULL;
995 put_disk(disk);
996 module_put(owner);
997 out:
998 mutex_unlock(&bdev->bd_mutex);
999 unlock_kernel();
1000 if (ret)
1001 bdput(bdev);
1002 return ret;
1003 }
1004
1005 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1006 {
1007 /*
1008 * This crockload is due to bad choice of ->open() type.
1009 * It will go away.
1010 * For now, block device ->open() routine must _not_
1011 * examine anything in 'inode' argument except ->i_rdev.
1012 */
1013 struct file fake_file = {};
1014 struct dentry fake_dentry = {};
1015 fake_file.f_mode = mode;
1016 fake_file.f_flags = flags;
1017 fake_file.f_dentry = &fake_dentry;
1018 fake_dentry.d_inode = bdev->bd_inode;
1019
1020 return do_open(bdev, &fake_file);
1021 }
1022
1023 EXPORT_SYMBOL(blkdev_get);
1024
1025 static int blkdev_open(struct inode * inode, struct file * filp)
1026 {
1027 struct block_device *bdev;
1028 int res;
1029
1030 /*
1031 * Preserve backwards compatibility and allow large file access
1032 * even if userspace doesn't ask for it explicitly. Some mkfs
1033 * binary needs it. We might want to drop this workaround
1034 * during an unstable branch.
1035 */
1036 filp->f_flags |= O_LARGEFILE;
1037
1038 bdev = bd_acquire(inode);
1039 if (bdev == NULL)
1040 return -ENOMEM;
1041
1042 res = do_open(bdev, filp);
1043 if (res)
1044 return res;
1045
1046 if (!(filp->f_flags & O_EXCL) )
1047 return 0;
1048
1049 if (!(res = bd_claim(bdev, filp)))
1050 return 0;
1051
1052 blkdev_put(bdev);
1053 return res;
1054 }
1055
1056 int blkdev_put(struct block_device *bdev)
1057 {
1058 int ret = 0;
1059 struct inode *bd_inode = bdev->bd_inode;
1060 struct gendisk *disk = bdev->bd_disk;
1061
1062 mutex_lock(&bdev->bd_mutex);
1063 lock_kernel();
1064 if (!--bdev->bd_openers) {
1065 sync_blockdev(bdev);
1066 kill_bdev(bdev);
1067 }
1068 if (bdev->bd_contains == bdev) {
1069 if (disk->fops->release)
1070 ret = disk->fops->release(bd_inode, NULL);
1071 } else {
1072 mutex_lock(&bdev->bd_contains->bd_mutex);
1073 bdev->bd_contains->bd_part_count--;
1074 mutex_unlock(&bdev->bd_contains->bd_mutex);
1075 }
1076 if (!bdev->bd_openers) {
1077 struct module *owner = disk->fops->owner;
1078
1079 put_disk(disk);
1080 module_put(owner);
1081
1082 if (bdev->bd_contains != bdev) {
1083 kobject_put(&bdev->bd_part->kobj);
1084 bdev->bd_part = NULL;
1085 }
1086 bdev->bd_disk = NULL;
1087 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1088 if (bdev != bdev->bd_contains) {
1089 blkdev_put(bdev->bd_contains);
1090 }
1091 bdev->bd_contains = NULL;
1092 }
1093 unlock_kernel();
1094 mutex_unlock(&bdev->bd_mutex);
1095 bdput(bdev);
1096 return ret;
1097 }
1098
1099 EXPORT_SYMBOL(blkdev_put);
1100
1101 static int blkdev_close(struct inode * inode, struct file * filp)
1102 {
1103 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1104 if (bdev->bd_holder == filp)
1105 bd_release(bdev);
1106 return blkdev_put(bdev);
1107 }
1108
1109 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1110 {
1111 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1112 }
1113
1114 const struct address_space_operations def_blk_aops = {
1115 .readpage = blkdev_readpage,
1116 .writepage = blkdev_writepage,
1117 .sync_page = block_sync_page,
1118 .prepare_write = blkdev_prepare_write,
1119 .commit_write = blkdev_commit_write,
1120 .writepages = generic_writepages,
1121 .direct_IO = blkdev_direct_IO,
1122 };
1123
1124 const struct file_operations def_blk_fops = {
1125 .open = blkdev_open,
1126 .release = blkdev_close,
1127 .llseek = block_llseek,
1128 .read = do_sync_read,
1129 .write = do_sync_write,
1130 .aio_read = generic_file_aio_read,
1131 .aio_write = generic_file_aio_write_nolock,
1132 .mmap = generic_file_mmap,
1133 .fsync = block_fsync,
1134 .unlocked_ioctl = block_ioctl,
1135 #ifdef CONFIG_COMPAT
1136 .compat_ioctl = compat_blkdev_ioctl,
1137 #endif
1138 .sendfile = generic_file_sendfile,
1139 .splice_read = generic_file_splice_read,
1140 .splice_write = generic_file_splice_write,
1141 };
1142
1143 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1144 {
1145 int res;
1146 mm_segment_t old_fs = get_fs();
1147 set_fs(KERNEL_DS);
1148 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1149 set_fs(old_fs);
1150 return res;
1151 }
1152
1153 EXPORT_SYMBOL(ioctl_by_bdev);
1154
1155 /**
1156 * lookup_bdev - lookup a struct block_device by name
1157 *
1158 * @path: special file representing the block device
1159 *
1160 * Get a reference to the blockdevice at @path in the current
1161 * namespace if possible and return it. Return ERR_PTR(error)
1162 * otherwise.
1163 */
1164 struct block_device *lookup_bdev(const char *path)
1165 {
1166 struct block_device *bdev;
1167 struct inode *inode;
1168 struct nameidata nd;
1169 int error;
1170
1171 if (!path || !*path)
1172 return ERR_PTR(-EINVAL);
1173
1174 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1175 if (error)
1176 return ERR_PTR(error);
1177
1178 inode = nd.dentry->d_inode;
1179 error = -ENOTBLK;
1180 if (!S_ISBLK(inode->i_mode))
1181 goto fail;
1182 error = -EACCES;
1183 if (nd.mnt->mnt_flags & MNT_NODEV)
1184 goto fail;
1185 error = -ENOMEM;
1186 bdev = bd_acquire(inode);
1187 if (!bdev)
1188 goto fail;
1189 out:
1190 path_release(&nd);
1191 return bdev;
1192 fail:
1193 bdev = ERR_PTR(error);
1194 goto out;
1195 }
1196
1197 /**
1198 * open_bdev_excl - open a block device by name and set it up for use
1199 *
1200 * @path: special file representing the block device
1201 * @flags: %MS_RDONLY for opening read-only
1202 * @holder: owner for exclusion
1203 *
1204 * Open the blockdevice described by the special file at @path, claim it
1205 * for the @holder.
1206 */
1207 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1208 {
1209 struct block_device *bdev;
1210 mode_t mode = FMODE_READ;
1211 int error = 0;
1212
1213 bdev = lookup_bdev(path);
1214 if (IS_ERR(bdev))
1215 return bdev;
1216
1217 if (!(flags & MS_RDONLY))
1218 mode |= FMODE_WRITE;
1219 error = blkdev_get(bdev, mode, 0);
1220 if (error)
1221 return ERR_PTR(error);
1222 error = -EACCES;
1223 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1224 goto blkdev_put;
1225 error = bd_claim(bdev, holder);
1226 if (error)
1227 goto blkdev_put;
1228
1229 return bdev;
1230
1231 blkdev_put:
1232 blkdev_put(bdev);
1233 return ERR_PTR(error);
1234 }
1235
1236 EXPORT_SYMBOL(open_bdev_excl);
1237
1238 /**
1239 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1240 *
1241 * @bdev: blockdevice to close
1242 *
1243 * This is the counterpart to open_bdev_excl().
1244 */
1245 void close_bdev_excl(struct block_device *bdev)
1246 {
1247 bd_release(bdev);
1248 blkdev_put(bdev);
1249 }
1250
1251 EXPORT_SYMBOL(close_bdev_excl);
1252
1253 int __invalidate_device(struct block_device *bdev)
1254 {
1255 struct super_block *sb = get_super(bdev);
1256 int res = 0;
1257
1258 if (sb) {
1259 /*
1260 * no need to lock the super, get_super holds the
1261 * read mutex so the filesystem cannot go away
1262 * under us (->put_super runs with the write lock
1263 * hold).
1264 */
1265 shrink_dcache_sb(sb);
1266 res = invalidate_inodes(sb);
1267 drop_super(sb);
1268 }
1269 invalidate_bdev(bdev, 0);
1270 return res;
1271 }
1272 EXPORT_SYMBOL(__invalidate_device);
This page took 0.059243 seconds and 5 git commands to generate.