ia64/linux-2.6.18-xen.hg

view fs/block_dev.c @ 452:c7ed6fe5dca0

kexec: dont initialise regions in reserve_memory()

There is no need to initialise efi_memmap_res and boot_param_res in
reserve_memory() for the initial xen domain as it is done in
machine_kexec_setup_resources() using values from the kexec hypercall.

Signed-off-by: Simon Horman <horms@verge.net.au>
author Keir Fraser <keir.fraser@citrix.com>
date Thu Feb 28 10:55:18 2008 +0000 (2008-02-28)
parents 831230e53067
children
line source
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 */
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/mpage.h>
21 #include <linux/mount.h>
22 #include <linux/uio.h>
23 #include <linux/namei.h>
24 #include <asm/uaccess.h>
26 struct bdev_inode {
27 struct block_device bdev;
28 struct inode vfs_inode;
29 };
31 static inline struct bdev_inode *BDEV_I(struct inode *inode)
32 {
33 return container_of(inode, struct bdev_inode, vfs_inode);
34 }
36 inline struct block_device *I_BDEV(struct inode *inode)
37 {
38 return &BDEV_I(inode)->bdev;
39 }
41 EXPORT_SYMBOL(I_BDEV);
43 static sector_t max_block(struct block_device *bdev)
44 {
45 sector_t retval = ~((sector_t)0);
46 loff_t sz = i_size_read(bdev->bd_inode);
48 if (sz) {
49 unsigned int size = block_size(bdev);
50 unsigned int sizebits = blksize_bits(size);
51 retval = (sz >> sizebits);
52 }
53 return retval;
54 }
56 /* Kill _all_ buffers, dirty or not.. */
57 static void kill_bdev(struct block_device *bdev)
58 {
59 invalidate_bdev(bdev, 1);
60 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
61 }
63 int set_blocksize(struct block_device *bdev, int size)
64 {
65 /* Size must be a power of two, and between 512 and PAGE_SIZE */
66 if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
67 return -EINVAL;
69 /* Size cannot be smaller than the size supported by the device */
70 if (size < bdev_hardsect_size(bdev))
71 return -EINVAL;
73 /* Don't change the size if it is same as current */
74 if (bdev->bd_block_size != size) {
75 sync_blockdev(bdev);
76 bdev->bd_block_size = size;
77 bdev->bd_inode->i_blkbits = blksize_bits(size);
78 kill_bdev(bdev);
79 }
80 return 0;
81 }
83 EXPORT_SYMBOL(set_blocksize);
85 int sb_set_blocksize(struct super_block *sb, int size)
86 {
87 if (set_blocksize(sb->s_bdev, size))
88 return 0;
89 /* If we get here, we know size is power of two
90 * and it's value is between 512 and PAGE_SIZE */
91 sb->s_blocksize = size;
92 sb->s_blocksize_bits = blksize_bits(size);
93 return sb->s_blocksize;
94 }
96 EXPORT_SYMBOL(sb_set_blocksize);
98 int sb_min_blocksize(struct super_block *sb, int size)
99 {
100 int minsize = bdev_hardsect_size(sb->s_bdev);
101 if (size < minsize)
102 size = minsize;
103 return sb_set_blocksize(sb, size);
104 }
106 EXPORT_SYMBOL(sb_min_blocksize);
108 static int
109 blkdev_get_block(struct inode *inode, sector_t iblock,
110 struct buffer_head *bh, int create)
111 {
112 if (iblock >= max_block(I_BDEV(inode))) {
113 if (create)
114 return -EIO;
116 /*
117 * for reads, we're just trying to fill a partial page.
118 * return a hole, they will have to call get_block again
119 * before they can fill it, and they will get -EIO at that
120 * time
121 */
122 return 0;
123 }
124 bh->b_bdev = I_BDEV(inode);
125 bh->b_blocknr = iblock;
126 set_buffer_mapped(bh);
127 return 0;
128 }
130 static int
131 blkdev_get_blocks(struct inode *inode, sector_t iblock,
132 struct buffer_head *bh, int create)
133 {
134 sector_t end_block = max_block(I_BDEV(inode));
135 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
137 if ((iblock + max_blocks) > end_block) {
138 max_blocks = end_block - iblock;
139 if ((long)max_blocks <= 0) {
140 if (create)
141 return -EIO; /* write fully beyond EOF */
142 /*
143 * It is a read which is fully beyond EOF. We return
144 * a !buffer_mapped buffer
145 */
146 max_blocks = 0;
147 }
148 }
150 bh->b_bdev = I_BDEV(inode);
151 bh->b_blocknr = iblock;
152 bh->b_size = max_blocks << inode->i_blkbits;
153 if (max_blocks)
154 set_buffer_mapped(bh);
155 return 0;
156 }
158 static ssize_t
159 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
160 loff_t offset, unsigned long nr_segs)
161 {
162 struct file *file = iocb->ki_filp;
163 struct inode *inode = file->f_mapping->host;
165 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
166 iov, offset, nr_segs, blkdev_get_blocks, NULL);
167 }
169 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
170 {
171 return block_write_full_page(page, blkdev_get_block, wbc);
172 }
174 static int blkdev_readpage(struct file * file, struct page * page)
175 {
176 return block_read_full_page(page, blkdev_get_block);
177 }
179 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
180 {
181 return block_prepare_write(page, from, to, blkdev_get_block);
182 }
184 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
185 {
186 return block_commit_write(page, from, to);
187 }
189 /*
190 * private llseek:
191 * for a block special file file->f_dentry->d_inode->i_size is zero
192 * so we compute the size by hand (just as in block_read/write above)
193 */
194 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
195 {
196 struct inode *bd_inode = file->f_mapping->host;
197 loff_t size;
198 loff_t retval;
200 mutex_lock(&bd_inode->i_mutex);
201 size = i_size_read(bd_inode);
203 switch (origin) {
204 case 2:
205 offset += size;
206 break;
207 case 1:
208 offset += file->f_pos;
209 }
210 retval = -EINVAL;
211 if (offset >= 0 && offset <= size) {
212 if (offset != file->f_pos) {
213 file->f_pos = offset;
214 }
215 retval = offset;
216 }
217 mutex_unlock(&bd_inode->i_mutex);
218 return retval;
219 }
221 /*
222 * Filp is never NULL; the only case when ->fsync() is called with
223 * NULL first argument is nfsd_sync_dir() and that's not a directory.
224 */
226 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
227 {
228 return sync_blockdev(I_BDEV(filp->f_mapping->host));
229 }
231 /*
232 * pseudo-fs
233 */
235 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
236 static kmem_cache_t * bdev_cachep __read_mostly;
238 static struct inode *bdev_alloc_inode(struct super_block *sb)
239 {
240 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, SLAB_KERNEL);
241 if (!ei)
242 return NULL;
243 return &ei->vfs_inode;
244 }
246 static void bdev_destroy_inode(struct inode *inode)
247 {
248 struct bdev_inode *bdi = BDEV_I(inode);
250 bdi->bdev.bd_inode_backing_dev_info = NULL;
251 kmem_cache_free(bdev_cachep, bdi);
252 }
254 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
255 {
256 struct bdev_inode *ei = (struct bdev_inode *) foo;
257 struct block_device *bdev = &ei->bdev;
259 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
260 SLAB_CTOR_CONSTRUCTOR)
261 {
262 memset(bdev, 0, sizeof(*bdev));
263 mutex_init(&bdev->bd_mutex);
264 mutex_init(&bdev->bd_mount_mutex);
265 INIT_LIST_HEAD(&bdev->bd_inodes);
266 INIT_LIST_HEAD(&bdev->bd_list);
267 #ifdef CONFIG_SYSFS
268 INIT_LIST_HEAD(&bdev->bd_holder_list);
269 #endif
270 inode_init_once(&ei->vfs_inode);
271 }
272 }
274 static inline void __bd_forget(struct inode *inode)
275 {
276 list_del_init(&inode->i_devices);
277 inode->i_bdev = NULL;
278 inode->i_mapping = &inode->i_data;
279 }
281 static void bdev_clear_inode(struct inode *inode)
282 {
283 struct block_device *bdev = &BDEV_I(inode)->bdev;
284 struct list_head *p;
285 spin_lock(&bdev_lock);
286 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
287 __bd_forget(list_entry(p, struct inode, i_devices));
288 }
289 list_del_init(&bdev->bd_list);
290 spin_unlock(&bdev_lock);
291 }
293 static struct super_operations bdev_sops = {
294 .statfs = simple_statfs,
295 .alloc_inode = bdev_alloc_inode,
296 .destroy_inode = bdev_destroy_inode,
297 .drop_inode = generic_delete_inode,
298 .clear_inode = bdev_clear_inode,
299 };
301 static int bd_get_sb(struct file_system_type *fs_type,
302 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
303 {
304 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
305 }
307 static struct file_system_type bd_type = {
308 .name = "bdev",
309 .get_sb = bd_get_sb,
310 .kill_sb = kill_anon_super,
311 };
313 static struct vfsmount *bd_mnt __read_mostly;
314 struct super_block *blockdev_superblock;
316 void __init bdev_cache_init(void)
317 {
318 int err;
319 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
320 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
321 SLAB_MEM_SPREAD|SLAB_PANIC),
322 init_once, NULL);
323 err = register_filesystem(&bd_type);
324 if (err)
325 panic("Cannot register bdev pseudo-fs");
326 bd_mnt = kern_mount(&bd_type);
327 err = PTR_ERR(bd_mnt);
328 if (IS_ERR(bd_mnt))
329 panic("Cannot create bdev pseudo-fs");
330 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
331 }
333 /*
334 * Most likely _very_ bad one - but then it's hardly critical for small
335 * /dev and can be fixed when somebody will need really large one.
336 * Keep in mind that it will be fed through icache hash function too.
337 */
338 static inline unsigned long hash(dev_t dev)
339 {
340 return MAJOR(dev)+MINOR(dev);
341 }
343 static int bdev_test(struct inode *inode, void *data)
344 {
345 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
346 }
348 static int bdev_set(struct inode *inode, void *data)
349 {
350 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
351 return 0;
352 }
354 static LIST_HEAD(all_bdevs);
356 struct block_device *bdget(dev_t dev)
357 {
358 struct block_device *bdev;
359 struct inode *inode;
361 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
362 bdev_test, bdev_set, &dev);
364 if (!inode)
365 return NULL;
367 bdev = &BDEV_I(inode)->bdev;
369 if (inode->i_state & I_NEW) {
370 bdev->bd_contains = NULL;
371 bdev->bd_inode = inode;
372 bdev->bd_block_size = (1 << inode->i_blkbits);
373 bdev->bd_part_count = 0;
374 bdev->bd_invalidated = 0;
375 inode->i_mode = S_IFBLK;
376 inode->i_rdev = dev;
377 inode->i_bdev = bdev;
378 inode->i_data.a_ops = &def_blk_aops;
379 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
380 inode->i_data.backing_dev_info = &default_backing_dev_info;
381 spin_lock(&bdev_lock);
382 list_add(&bdev->bd_list, &all_bdevs);
383 spin_unlock(&bdev_lock);
384 unlock_new_inode(inode);
385 }
386 return bdev;
387 }
389 EXPORT_SYMBOL(bdget);
391 long nr_blockdev_pages(void)
392 {
393 struct list_head *p;
394 long ret = 0;
395 spin_lock(&bdev_lock);
396 list_for_each(p, &all_bdevs) {
397 struct block_device *bdev;
398 bdev = list_entry(p, struct block_device, bd_list);
399 ret += bdev->bd_inode->i_mapping->nrpages;
400 }
401 spin_unlock(&bdev_lock);
402 return ret;
403 }
405 void bdput(struct block_device *bdev)
406 {
407 iput(bdev->bd_inode);
408 }
410 EXPORT_SYMBOL(bdput);
412 static struct block_device *bd_acquire(struct inode *inode)
413 {
414 struct block_device *bdev;
416 spin_lock(&bdev_lock);
417 bdev = inode->i_bdev;
418 if (bdev) {
419 atomic_inc(&bdev->bd_inode->i_count);
420 spin_unlock(&bdev_lock);
421 return bdev;
422 }
423 spin_unlock(&bdev_lock);
425 bdev = bdget(inode->i_rdev);
426 if (bdev) {
427 spin_lock(&bdev_lock);
428 if (!inode->i_bdev) {
429 /*
430 * We take an additional bd_inode->i_count for inode,
431 * and it's released in clear_inode() of inode.
432 * So, we can access it via ->i_mapping always
433 * without igrab().
434 */
435 atomic_inc(&bdev->bd_inode->i_count);
436 inode->i_bdev = bdev;
437 inode->i_mapping = bdev->bd_inode->i_mapping;
438 list_add(&inode->i_devices, &bdev->bd_inodes);
439 }
440 spin_unlock(&bdev_lock);
441 }
442 return bdev;
443 }
445 /* Call when you free inode */
447 void bd_forget(struct inode *inode)
448 {
449 struct block_device *bdev = NULL;
451 spin_lock(&bdev_lock);
452 if (inode->i_bdev) {
453 if (inode->i_sb != blockdev_superblock)
454 bdev = inode->i_bdev;
455 __bd_forget(inode);
456 }
457 spin_unlock(&bdev_lock);
459 if (bdev)
460 iput(bdev->bd_inode);
461 }
463 int bd_claim(struct block_device *bdev, void *holder)
464 {
465 int res;
466 spin_lock(&bdev_lock);
468 /* first decide result */
469 if (bdev->bd_holder == holder)
470 res = 0; /* already a holder */
471 else if (bdev->bd_holder != NULL)
472 res = -EBUSY; /* held by someone else */
473 else if (bdev->bd_contains == bdev)
474 res = 0; /* is a whole device which isn't held */
476 else if (bdev->bd_contains->bd_holder == bd_claim)
477 res = 0; /* is a partition of a device that is being partitioned */
478 else if (bdev->bd_contains->bd_holder != NULL)
479 res = -EBUSY; /* is a partition of a held device */
480 else
481 res = 0; /* is a partition of an un-held device */
483 /* now impose change */
484 if (res==0) {
485 /* note that for a whole device bd_holders
486 * will be incremented twice, and bd_holder will
487 * be set to bd_claim before being set to holder
488 */
489 bdev->bd_contains->bd_holders ++;
490 bdev->bd_contains->bd_holder = bd_claim;
491 bdev->bd_holders++;
492 bdev->bd_holder = holder;
493 }
494 spin_unlock(&bdev_lock);
495 return res;
496 }
498 EXPORT_SYMBOL(bd_claim);
500 void bd_release(struct block_device *bdev)
501 {
502 spin_lock(&bdev_lock);
503 if (!--bdev->bd_contains->bd_holders)
504 bdev->bd_contains->bd_holder = NULL;
505 if (!--bdev->bd_holders)
506 bdev->bd_holder = NULL;
507 spin_unlock(&bdev_lock);
508 }
510 EXPORT_SYMBOL(bd_release);
512 #ifdef CONFIG_SYSFS
513 /*
514 * Functions for bd_claim_by_kobject / bd_release_from_kobject
515 *
516 * If a kobject is passed to bd_claim_by_kobject()
517 * and the kobject has a parent directory,
518 * following symlinks are created:
519 * o from the kobject to the claimed bdev
520 * o from "holders" directory of the bdev to the parent of the kobject
521 * bd_release_from_kobject() removes these symlinks.
522 *
523 * Example:
524 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
525 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
526 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
527 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
528 */
530 static struct kobject *bdev_get_kobj(struct block_device *bdev)
531 {
532 if (bdev->bd_contains != bdev)
533 return kobject_get(&bdev->bd_part->kobj);
534 else
535 return kobject_get(&bdev->bd_disk->kobj);
536 }
538 static struct kobject *bdev_get_holder(struct block_device *bdev)
539 {
540 if (bdev->bd_contains != bdev)
541 return kobject_get(bdev->bd_part->holder_dir);
542 else
543 return kobject_get(bdev->bd_disk->holder_dir);
544 }
546 static void add_symlink(struct kobject *from, struct kobject *to)
547 {
548 if (!from || !to)
549 return;
550 sysfs_create_link(from, to, kobject_name(to));
551 }
553 static void del_symlink(struct kobject *from, struct kobject *to)
554 {
555 if (!from || !to)
556 return;
557 sysfs_remove_link(from, kobject_name(to));
558 }
560 /*
561 * 'struct bd_holder' contains pointers to kobjects symlinked by
562 * bd_claim_by_kobject.
563 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
564 */
565 struct bd_holder {
566 struct list_head list; /* chain of holders of the bdev */
567 int count; /* references from the holder */
568 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
569 struct kobject *hdev; /* e.g. "/block/dm-0" */
570 struct kobject *hdir; /* e.g. "/block/sda/holders" */
571 struct kobject *sdev; /* e.g. "/block/sda" */
572 };
574 /*
575 * Get references of related kobjects at once.
576 * Returns 1 on success. 0 on failure.
577 *
578 * Should call bd_holder_release_dirs() after successful use.
579 */
580 static int bd_holder_grab_dirs(struct block_device *bdev,
581 struct bd_holder *bo)
582 {
583 if (!bdev || !bo)
584 return 0;
586 bo->sdir = kobject_get(bo->sdir);
587 if (!bo->sdir)
588 return 0;
590 bo->hdev = kobject_get(bo->sdir->parent);
591 if (!bo->hdev)
592 goto fail_put_sdir;
594 bo->sdev = bdev_get_kobj(bdev);
595 if (!bo->sdev)
596 goto fail_put_hdev;
598 bo->hdir = bdev_get_holder(bdev);
599 if (!bo->hdir)
600 goto fail_put_sdev;
602 return 1;
604 fail_put_sdev:
605 kobject_put(bo->sdev);
606 fail_put_hdev:
607 kobject_put(bo->hdev);
608 fail_put_sdir:
609 kobject_put(bo->sdir);
611 return 0;
612 }
614 /* Put references of related kobjects at once. */
615 static void bd_holder_release_dirs(struct bd_holder *bo)
616 {
617 kobject_put(bo->hdir);
618 kobject_put(bo->sdev);
619 kobject_put(bo->hdev);
620 kobject_put(bo->sdir);
621 }
623 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
624 {
625 struct bd_holder *bo;
627 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
628 if (!bo)
629 return NULL;
631 bo->count = 1;
632 bo->sdir = kobj;
634 return bo;
635 }
637 static void free_bd_holder(struct bd_holder *bo)
638 {
639 kfree(bo);
640 }
642 /**
643 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
644 *
645 * @bdev: block device to be bd_claimed
646 * @bo: preallocated and initialized by alloc_bd_holder()
647 *
648 * If there is no matching entry with @bo in @bdev->bd_holder_list,
649 * add @bo to the list, create symlinks.
650 *
651 * Returns 1 if @bo was added to the list.
652 * Returns 0 if @bo wasn't used by any reason and should be freed.
653 */
654 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
655 {
656 struct bd_holder *tmp;
658 if (!bo)
659 return 0;
661 list_for_each_entry(tmp, &bdev->bd_holder_list, list) {
662 if (tmp->sdir == bo->sdir) {
663 tmp->count++;
664 return 0;
665 }
666 }
668 if (!bd_holder_grab_dirs(bdev, bo))
669 return 0;
671 add_symlink(bo->sdir, bo->sdev);
672 add_symlink(bo->hdir, bo->hdev);
673 list_add_tail(&bo->list, &bdev->bd_holder_list);
674 return 1;
675 }
677 /**
678 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
679 *
680 * @bdev: block device to be bd_claimed
681 * @kobj: holder's kobject
682 *
683 * If there is matching entry with @kobj in @bdev->bd_holder_list
684 * and no other bd_claim() from the same kobject,
685 * remove the struct bd_holder from the list, delete symlinks for it.
686 *
687 * Returns a pointer to the struct bd_holder when it's removed from the list
688 * and ready to be freed.
689 * Returns NULL if matching claim isn't found or there is other bd_claim()
690 * by the same kobject.
691 */
692 static struct bd_holder *del_bd_holder(struct block_device *bdev,
693 struct kobject *kobj)
694 {
695 struct bd_holder *bo;
697 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
698 if (bo->sdir == kobj) {
699 bo->count--;
700 BUG_ON(bo->count < 0);
701 if (!bo->count) {
702 list_del(&bo->list);
703 del_symlink(bo->sdir, bo->sdev);
704 del_symlink(bo->hdir, bo->hdev);
705 bd_holder_release_dirs(bo);
706 return bo;
707 }
708 break;
709 }
710 }
712 return NULL;
713 }
715 /**
716 * bd_claim_by_kobject - bd_claim() with additional kobject signature
717 *
718 * @bdev: block device to be claimed
719 * @holder: holder's signature
720 * @kobj: holder's kobject
721 *
722 * Do bd_claim() and if it succeeds, create sysfs symlinks between
723 * the bdev and the holder's kobject.
724 * Use bd_release_from_kobject() when relesing the claimed bdev.
725 *
726 * Returns 0 on success. (same as bd_claim())
727 * Returns errno on failure.
728 */
729 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
730 struct kobject *kobj)
731 {
732 int res;
733 struct bd_holder *bo;
735 if (!kobj)
736 return -EINVAL;
738 bo = alloc_bd_holder(kobj);
739 if (!bo)
740 return -ENOMEM;
742 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
743 res = bd_claim(bdev, holder);
744 if (res || !add_bd_holder(bdev, bo))
745 free_bd_holder(bo);
746 mutex_unlock(&bdev->bd_mutex);
748 return res;
749 }
751 /**
752 * bd_release_from_kobject - bd_release() with additional kobject signature
753 *
754 * @bdev: block device to be released
755 * @kobj: holder's kobject
756 *
757 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
758 */
759 static void bd_release_from_kobject(struct block_device *bdev,
760 struct kobject *kobj)
761 {
762 struct bd_holder *bo;
764 if (!kobj)
765 return;
767 mutex_lock_nested(&bdev->bd_mutex, BD_MUTEX_PARTITION);
768 bd_release(bdev);
769 if ((bo = del_bd_holder(bdev, kobj)))
770 free_bd_holder(bo);
771 mutex_unlock(&bdev->bd_mutex);
772 }
774 /**
775 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
776 *
777 * @bdev: block device to be claimed
778 * @holder: holder's signature
779 * @disk: holder's gendisk
780 *
781 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
782 */
783 int bd_claim_by_disk(struct block_device *bdev, void *holder,
784 struct gendisk *disk)
785 {
786 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
787 }
788 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
790 /**
791 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
792 *
793 * @bdev: block device to be claimed
794 * @disk: holder's gendisk
795 *
796 * Call bd_release_from_kobject() and put @disk->slave_dir.
797 */
798 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
799 {
800 bd_release_from_kobject(bdev, disk->slave_dir);
801 kobject_put(disk->slave_dir);
802 }
803 EXPORT_SYMBOL_GPL(bd_release_from_disk);
804 #endif
806 /*
807 * Tries to open block device by device number. Use it ONLY if you
808 * really do not have anything better - i.e. when you are behind a
809 * truly sucky interface and all you are given is a device number. _Never_
810 * to be used for internal purposes. If you ever need it - reconsider
811 * your API.
812 */
813 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
814 {
815 struct block_device *bdev = bdget(dev);
816 int err = -ENOMEM;
817 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
818 if (bdev)
819 err = blkdev_get(bdev, mode, flags);
820 return err ? ERR_PTR(err) : bdev;
821 }
823 EXPORT_SYMBOL(open_by_devnum);
825 static int
826 blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags);
828 struct block_device *open_partition_by_devnum(dev_t dev, unsigned mode)
829 {
830 struct block_device *bdev = bdget(dev);
831 int err = -ENOMEM;
832 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
833 if (bdev)
834 err = blkdev_get_partition(bdev, mode, flags);
835 return err ? ERR_PTR(err) : bdev;
836 }
838 EXPORT_SYMBOL(open_partition_by_devnum);
841 /*
842 * This routine checks whether a removable media has been changed,
843 * and invalidates all buffer-cache-entries in that case. This
844 * is a relatively slow routine, so we have to try to minimize using
845 * it. Thus it is called only upon a 'mount' or 'open'. This
846 * is the best way of combining speed and utility, I think.
847 * People changing diskettes in the middle of an operation deserve
848 * to lose :-)
849 */
850 int check_disk_change(struct block_device *bdev)
851 {
852 struct gendisk *disk = bdev->bd_disk;
853 struct block_device_operations * bdops = disk->fops;
855 if (!bdops->media_changed)
856 return 0;
857 if (!bdops->media_changed(bdev->bd_disk))
858 return 0;
860 if (__invalidate_device(bdev))
861 printk("VFS: busy inodes on changed media.\n");
863 if (bdops->revalidate_disk)
864 bdops->revalidate_disk(bdev->bd_disk);
865 if (bdev->bd_disk->minors > 1)
866 bdev->bd_invalidated = 1;
867 return 1;
868 }
870 EXPORT_SYMBOL(check_disk_change);
872 void bd_set_size(struct block_device *bdev, loff_t size)
873 {
874 unsigned bsize = bdev_hardsect_size(bdev);
876 bdev->bd_inode->i_size = size;
877 while (bsize < PAGE_CACHE_SIZE) {
878 if (size & bsize)
879 break;
880 bsize <<= 1;
881 }
882 bdev->bd_block_size = bsize;
883 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
884 }
885 EXPORT_SYMBOL(bd_set_size);
887 static int __blkdev_put(struct block_device *bdev, unsigned int subclass)
888 {
889 int ret = 0;
890 struct inode *bd_inode = bdev->bd_inode;
891 struct gendisk *disk = bdev->bd_disk;
893 mutex_lock_nested(&bdev->bd_mutex, subclass);
894 lock_kernel();
895 if (!--bdev->bd_openers) {
896 sync_blockdev(bdev);
897 kill_bdev(bdev);
898 }
899 if (bdev->bd_contains == bdev) {
900 if (disk->fops->release)
901 ret = disk->fops->release(bd_inode, NULL);
902 } else {
903 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
904 subclass + 1);
905 bdev->bd_contains->bd_part_count--;
906 mutex_unlock(&bdev->bd_contains->bd_mutex);
907 }
908 if (!bdev->bd_openers) {
909 struct module *owner = disk->fops->owner;
911 put_disk(disk);
912 module_put(owner);
914 if (bdev->bd_contains != bdev) {
915 kobject_put(&bdev->bd_part->kobj);
916 bdev->bd_part = NULL;
917 }
918 bdev->bd_disk = NULL;
919 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
920 if (bdev != bdev->bd_contains)
921 __blkdev_put(bdev->bd_contains, subclass + 1);
922 bdev->bd_contains = NULL;
923 }
924 unlock_kernel();
925 mutex_unlock(&bdev->bd_mutex);
926 bdput(bdev);
927 return ret;
928 }
930 int blkdev_put(struct block_device *bdev)
931 {
932 return __blkdev_put(bdev, BD_MUTEX_NORMAL);
933 }
934 EXPORT_SYMBOL(blkdev_put);
936 int blkdev_put_partition(struct block_device *bdev)
937 {
938 return __blkdev_put(bdev, BD_MUTEX_PARTITION);
939 }
940 EXPORT_SYMBOL(blkdev_put_partition);
942 static int
943 blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags);
945 static int
946 do_open(struct block_device *bdev, struct file *file, unsigned int subclass)
947 {
948 struct module *owner = NULL;
949 struct gendisk *disk;
950 int ret = -ENXIO;
951 int part;
953 file->f_mapping = bdev->bd_inode->i_mapping;
954 lock_kernel();
955 disk = get_gendisk(bdev->bd_dev, &part);
956 if (!disk) {
957 unlock_kernel();
958 bdput(bdev);
959 return ret;
960 }
961 owner = disk->fops->owner;
963 mutex_lock_nested(&bdev->bd_mutex, subclass);
965 if (!bdev->bd_openers) {
966 bdev->bd_disk = disk;
967 bdev->bd_contains = bdev;
968 if (!part) {
969 struct backing_dev_info *bdi;
970 if (disk->fops->open) {
971 ret = disk->fops->open(bdev->bd_inode, file);
972 if (ret)
973 goto out_first;
974 }
975 if (!bdev->bd_openers) {
976 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
977 bdi = blk_get_backing_dev_info(bdev);
978 if (bdi == NULL)
979 bdi = &default_backing_dev_info;
980 bdev->bd_inode->i_data.backing_dev_info = bdi;
981 }
982 if (bdev->bd_invalidated)
983 rescan_partitions(disk, bdev);
984 } else {
985 struct hd_struct *p;
986 struct block_device *whole;
987 whole = bdget_disk(disk, 0);
988 ret = -ENOMEM;
989 if (!whole)
990 goto out_first;
991 ret = blkdev_get_whole(whole, file->f_mode, file->f_flags);
992 if (ret)
993 goto out_first;
994 bdev->bd_contains = whole;
995 mutex_lock_nested(&whole->bd_mutex, BD_MUTEX_WHOLE);
996 whole->bd_part_count++;
997 p = disk->part[part - 1];
998 bdev->bd_inode->i_data.backing_dev_info =
999 whole->bd_inode->i_data.backing_dev_info;
1000 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1001 whole->bd_part_count--;
1002 mutex_unlock(&whole->bd_mutex);
1003 ret = -ENXIO;
1004 goto out_first;
1006 kobject_get(&p->kobj);
1007 bdev->bd_part = p;
1008 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1009 mutex_unlock(&whole->bd_mutex);
1011 } else {
1012 put_disk(disk);
1013 module_put(owner);
1014 if (bdev->bd_contains == bdev) {
1015 if (bdev->bd_disk->fops->open) {
1016 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1017 if (ret)
1018 goto out;
1020 if (bdev->bd_invalidated)
1021 rescan_partitions(bdev->bd_disk, bdev);
1022 } else {
1023 mutex_lock_nested(&bdev->bd_contains->bd_mutex,
1024 BD_MUTEX_PARTITION);
1025 bdev->bd_contains->bd_part_count++;
1026 mutex_unlock(&bdev->bd_contains->bd_mutex);
1029 bdev->bd_openers++;
1030 mutex_unlock(&bdev->bd_mutex);
1031 unlock_kernel();
1032 return 0;
1034 out_first:
1035 bdev->bd_disk = NULL;
1036 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1037 if (bdev != bdev->bd_contains)
1038 __blkdev_put(bdev->bd_contains, BD_MUTEX_WHOLE);
1039 bdev->bd_contains = NULL;
1040 put_disk(disk);
1041 module_put(owner);
1042 out:
1043 mutex_unlock(&bdev->bd_mutex);
1044 unlock_kernel();
1045 if (ret)
1046 bdput(bdev);
1047 return ret;
1050 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1052 /*
1053 * This crockload is due to bad choice of ->open() type.
1054 * It will go away.
1055 * For now, block device ->open() routine must _not_
1056 * examine anything in 'inode' argument except ->i_rdev.
1057 */
1058 struct file fake_file = {};
1059 struct dentry fake_dentry = {};
1060 fake_file.f_mode = mode;
1061 fake_file.f_flags = flags;
1062 fake_file.f_dentry = &fake_dentry;
1063 fake_dentry.d_inode = bdev->bd_inode;
1065 return do_open(bdev, &fake_file, BD_MUTEX_NORMAL);
1068 EXPORT_SYMBOL(blkdev_get);
1070 static int
1071 blkdev_get_whole(struct block_device *bdev, mode_t mode, unsigned flags)
1073 /*
1074 * This crockload is due to bad choice of ->open() type.
1075 * It will go away.
1076 * For now, block device ->open() routine must _not_
1077 * examine anything in 'inode' argument except ->i_rdev.
1078 */
1079 struct file fake_file = {};
1080 struct dentry fake_dentry = {};
1081 fake_file.f_mode = mode;
1082 fake_file.f_flags = flags;
1083 fake_file.f_dentry = &fake_dentry;
1084 fake_dentry.d_inode = bdev->bd_inode;
1086 return do_open(bdev, &fake_file, BD_MUTEX_WHOLE);
1089 static int
1090 blkdev_get_partition(struct block_device *bdev, mode_t mode, unsigned flags)
1092 /*
1093 * This crockload is due to bad choice of ->open() type.
1094 * It will go away.
1095 * For now, block device ->open() routine must _not_
1096 * examine anything in 'inode' argument except ->i_rdev.
1097 */
1098 struct file fake_file = {};
1099 struct dentry fake_dentry = {};
1100 fake_file.f_mode = mode;
1101 fake_file.f_flags = flags;
1102 fake_file.f_dentry = &fake_dentry;
1103 fake_dentry.d_inode = bdev->bd_inode;
1105 return do_open(bdev, &fake_file, BD_MUTEX_PARTITION);
1108 static int blkdev_open(struct inode * inode, struct file * filp)
1110 struct block_device *bdev;
1111 int res;
1113 /*
1114 * Preserve backwards compatibility and allow large file access
1115 * even if userspace doesn't ask for it explicitly. Some mkfs
1116 * binary needs it. We might want to drop this workaround
1117 * during an unstable branch.
1118 */
1119 filp->f_flags |= O_LARGEFILE;
1121 bdev = bd_acquire(inode);
1123 res = do_open(bdev, filp, BD_MUTEX_NORMAL);
1124 if (res)
1125 return res;
1127 if (!(filp->f_flags & O_EXCL) )
1128 return 0;
1130 if (!(res = bd_claim(bdev, filp)))
1131 return 0;
1133 blkdev_put(bdev);
1134 return res;
1137 static int blkdev_close(struct inode * inode, struct file * filp)
1139 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1140 if (bdev->bd_holder == filp)
1141 bd_release(bdev);
1142 return blkdev_put(bdev);
1145 static ssize_t blkdev_file_write(struct file *file, const char __user *buf,
1146 size_t count, loff_t *ppos)
1148 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1150 return generic_file_write_nolock(file, &local_iov, 1, ppos);
1153 static ssize_t blkdev_file_aio_write(struct kiocb *iocb, const char __user *buf,
1154 size_t count, loff_t pos)
1156 struct iovec local_iov = { .iov_base = (void __user *)buf, .iov_len = count };
1158 return generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
1161 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1163 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1166 const struct address_space_operations def_blk_aops = {
1167 .readpage = blkdev_readpage,
1168 .writepage = blkdev_writepage,
1169 .sync_page = block_sync_page,
1170 .prepare_write = blkdev_prepare_write,
1171 .commit_write = blkdev_commit_write,
1172 .writepages = generic_writepages,
1173 .direct_IO = blkdev_direct_IO,
1174 };
1176 const struct file_operations def_blk_fops = {
1177 .open = blkdev_open,
1178 .release = blkdev_close,
1179 .llseek = block_llseek,
1180 .read = generic_file_read,
1181 .write = blkdev_file_write,
1182 .aio_read = generic_file_aio_read,
1183 .aio_write = blkdev_file_aio_write,
1184 .mmap = generic_file_mmap,
1185 .fsync = block_fsync,
1186 .unlocked_ioctl = block_ioctl,
1187 #ifdef CONFIG_COMPAT
1188 .compat_ioctl = compat_blkdev_ioctl,
1189 #endif
1190 .readv = generic_file_readv,
1191 .writev = generic_file_write_nolock,
1192 .sendfile = generic_file_sendfile,
1193 .splice_read = generic_file_splice_read,
1194 .splice_write = generic_file_splice_write,
1195 };
1197 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1199 int res;
1200 mm_segment_t old_fs = get_fs();
1201 set_fs(KERNEL_DS);
1202 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1203 set_fs(old_fs);
1204 return res;
1207 EXPORT_SYMBOL(ioctl_by_bdev);
1209 /**
1210 * lookup_bdev - lookup a struct block_device by name
1212 * @path: special file representing the block device
1214 * Get a reference to the blockdevice at @path in the current
1215 * namespace if possible and return it. Return ERR_PTR(error)
1216 * otherwise.
1217 */
1218 struct block_device *lookup_bdev(const char *path)
1220 struct block_device *bdev;
1221 struct inode *inode;
1222 struct nameidata nd;
1223 int error;
1225 if (!path || !*path)
1226 return ERR_PTR(-EINVAL);
1228 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1229 if (error)
1230 return ERR_PTR(error);
1232 inode = nd.dentry->d_inode;
1233 error = -ENOTBLK;
1234 if (!S_ISBLK(inode->i_mode))
1235 goto fail;
1236 error = -EACCES;
1237 if (nd.mnt->mnt_flags & MNT_NODEV)
1238 goto fail;
1239 error = -ENOMEM;
1240 bdev = bd_acquire(inode);
1241 if (!bdev)
1242 goto fail;
1243 out:
1244 path_release(&nd);
1245 return bdev;
1246 fail:
1247 bdev = ERR_PTR(error);
1248 goto out;
1251 /**
1252 * open_bdev_excl - open a block device by name and set it up for use
1254 * @path: special file representing the block device
1255 * @flags: %MS_RDONLY for opening read-only
1256 * @holder: owner for exclusion
1258 * Open the blockdevice described by the special file at @path, claim it
1259 * for the @holder.
1260 */
1261 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1263 struct block_device *bdev;
1264 mode_t mode = FMODE_READ;
1265 int error = 0;
1267 bdev = lookup_bdev(path);
1268 if (IS_ERR(bdev))
1269 return bdev;
1271 if (!(flags & MS_RDONLY))
1272 mode |= FMODE_WRITE;
1273 error = blkdev_get(bdev, mode, 0);
1274 if (error)
1275 return ERR_PTR(error);
1276 error = -EACCES;
1277 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1278 goto blkdev_put;
1279 error = bd_claim(bdev, holder);
1280 if (error)
1281 goto blkdev_put;
1283 return bdev;
1285 blkdev_put:
1286 blkdev_put(bdev);
1287 return ERR_PTR(error);
1290 EXPORT_SYMBOL(open_bdev_excl);
1292 /**
1293 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1295 * @bdev: blockdevice to close
1297 * This is the counterpart to open_bdev_excl().
1298 */
1299 void close_bdev_excl(struct block_device *bdev)
1301 bd_release(bdev);
1302 blkdev_put(bdev);
1305 EXPORT_SYMBOL(close_bdev_excl);