Drop some headers from mm.h
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / block_dev.c
blob993f78c55221c4c4083ac72114afceadfafd12fe
1 /*
2 * linux/fs/block_dev.c
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/writeback.h>
21 #include <linux/mpage.h>
22 #include <linux/mount.h>
23 #include <linux/uio.h>
24 #include <linux/namei.h>
25 #include <linux/log2.h>
26 #include <asm/uaccess.h>
27 #include "internal.h"
29 struct bdev_inode {
30 struct block_device bdev;
31 struct inode vfs_inode;
34 static inline struct bdev_inode *BDEV_I(struct inode *inode)
36 return container_of(inode, struct bdev_inode, vfs_inode);
39 inline struct block_device *I_BDEV(struct inode *inode)
41 return &BDEV_I(inode)->bdev;
44 EXPORT_SYMBOL(I_BDEV);
46 static sector_t max_block(struct block_device *bdev)
48 sector_t retval = ~((sector_t)0);
49 loff_t sz = i_size_read(bdev->bd_inode);
51 if (sz) {
52 unsigned int size = block_size(bdev);
53 unsigned int sizebits = blksize_bits(size);
54 retval = (sz >> sizebits);
56 return retval;
59 /* Kill _all_ buffers and pagecache , dirty or not.. */
60 static void kill_bdev(struct block_device *bdev)
62 if (bdev->bd_inode->i_mapping->nrpages == 0)
63 return;
64 invalidate_bh_lrus();
65 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
68 int set_blocksize(struct block_device *bdev, int size)
70 /* Size must be a power of two, and between 512 and PAGE_SIZE */
71 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
72 return -EINVAL;
74 /* Size cannot be smaller than the size supported by the device */
75 if (size < bdev_hardsect_size(bdev))
76 return -EINVAL;
78 /* Don't change the size if it is same as current */
79 if (bdev->bd_block_size != size) {
80 sync_blockdev(bdev);
81 bdev->bd_block_size = size;
82 bdev->bd_inode->i_blkbits = blksize_bits(size);
83 kill_bdev(bdev);
85 return 0;
88 EXPORT_SYMBOL(set_blocksize);
90 int sb_set_blocksize(struct super_block *sb, int size)
92 if (set_blocksize(sb->s_bdev, size))
93 return 0;
94 /* If we get here, we know size is power of two
95 * and it's value is between 512 and PAGE_SIZE */
96 sb->s_blocksize = size;
97 sb->s_blocksize_bits = blksize_bits(size);
98 return sb->s_blocksize;
101 EXPORT_SYMBOL(sb_set_blocksize);
103 int sb_min_blocksize(struct super_block *sb, int size)
105 int minsize = bdev_hardsect_size(sb->s_bdev);
106 if (size < minsize)
107 size = minsize;
108 return sb_set_blocksize(sb, size);
111 EXPORT_SYMBOL(sb_min_blocksize);
113 static int
114 blkdev_get_block(struct inode *inode, sector_t iblock,
115 struct buffer_head *bh, int create)
117 if (iblock >= max_block(I_BDEV(inode))) {
118 if (create)
119 return -EIO;
122 * for reads, we're just trying to fill a partial page.
123 * return a hole, they will have to call get_block again
124 * before they can fill it, and they will get -EIO at that
125 * time
127 return 0;
129 bh->b_bdev = I_BDEV(inode);
130 bh->b_blocknr = iblock;
131 set_buffer_mapped(bh);
132 return 0;
135 static int
136 blkdev_get_blocks(struct inode *inode, sector_t iblock,
137 struct buffer_head *bh, int create)
139 sector_t end_block = max_block(I_BDEV(inode));
140 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
142 if ((iblock + max_blocks) > end_block) {
143 max_blocks = end_block - iblock;
144 if ((long)max_blocks <= 0) {
145 if (create)
146 return -EIO; /* write fully beyond EOF */
148 * It is a read which is fully beyond EOF. We return
149 * a !buffer_mapped buffer
151 max_blocks = 0;
155 bh->b_bdev = I_BDEV(inode);
156 bh->b_blocknr = iblock;
157 bh->b_size = max_blocks << inode->i_blkbits;
158 if (max_blocks)
159 set_buffer_mapped(bh);
160 return 0;
163 static ssize_t
164 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
165 loff_t offset, unsigned long nr_segs)
167 struct file *file = iocb->ki_filp;
168 struct inode *inode = file->f_mapping->host;
170 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
171 iov, offset, nr_segs, blkdev_get_blocks, NULL);
174 #if 0
175 static void blk_end_aio(struct bio *bio, int error)
177 struct kiocb *iocb = bio->bi_private;
178 atomic_t *bio_count = &iocb->ki_bio_count;
180 if (bio_data_dir(bio) == READ)
181 bio_check_pages_dirty(bio);
182 else {
183 bio_release_pages(bio);
184 bio_put(bio);
187 /* iocb->ki_nbytes stores error code from LLDD */
188 if (error)
189 iocb->ki_nbytes = -EIO;
191 if (atomic_dec_and_test(bio_count)) {
192 if ((long)iocb->ki_nbytes < 0)
193 aio_complete(iocb, iocb->ki_nbytes, 0);
194 else
195 aio_complete(iocb, iocb->ki_left, 0);
198 return 0;
201 #define VEC_SIZE 16
202 struct pvec {
203 unsigned short nr;
204 unsigned short idx;
205 struct page *page[VEC_SIZE];
208 #define PAGES_SPANNED(addr, len) \
209 (DIV_ROUND_UP((addr) + (len), PAGE_SIZE) - (addr) / PAGE_SIZE);
212 * get page pointer for user addr, we internally cache struct page array for
213 * (addr, count) range in pvec to avoid frequent call to get_user_pages. If
214 * internal page list is exhausted, a batch count of up to VEC_SIZE is used
215 * to get next set of page struct.
217 static struct page *blk_get_page(unsigned long addr, size_t count, int rw,
218 struct pvec *pvec)
220 int ret, nr_pages;
221 if (pvec->idx == pvec->nr) {
222 nr_pages = PAGES_SPANNED(addr, count);
223 nr_pages = min(nr_pages, VEC_SIZE);
224 down_read(&current->mm->mmap_sem);
225 ret = get_user_pages(current, current->mm, addr, nr_pages,
226 rw == READ, 0, pvec->page, NULL);
227 up_read(&current->mm->mmap_sem);
228 if (ret < 0)
229 return ERR_PTR(ret);
230 pvec->nr = ret;
231 pvec->idx = 0;
233 return pvec->page[pvec->idx++];
236 /* return a page back to pvec array */
237 static void blk_unget_page(struct page *page, struct pvec *pvec)
239 pvec->page[--pvec->idx] = page;
242 static ssize_t
243 blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
244 loff_t pos, unsigned long nr_segs)
246 struct inode *inode = iocb->ki_filp->f_mapping->host;
247 unsigned blkbits = blksize_bits(bdev_hardsect_size(I_BDEV(inode)));
248 unsigned blocksize_mask = (1 << blkbits) - 1;
249 unsigned long seg = 0; /* iov segment iterator */
250 unsigned long nvec; /* number of bio vec needed */
251 unsigned long cur_off; /* offset into current page */
252 unsigned long cur_len; /* I/O len of current page, up to PAGE_SIZE */
254 unsigned long addr; /* user iovec address */
255 size_t count; /* user iovec len */
256 size_t nbytes = iocb->ki_nbytes = iocb->ki_left; /* total xfer size */
257 loff_t size; /* size of block device */
258 struct bio *bio;
259 atomic_t *bio_count = &iocb->ki_bio_count;
260 struct page *page;
261 struct pvec pvec;
263 pvec.nr = 0;
264 pvec.idx = 0;
266 if (pos & blocksize_mask)
267 return -EINVAL;
269 size = i_size_read(inode);
270 if (pos + nbytes > size) {
271 nbytes = size - pos;
272 iocb->ki_left = nbytes;
276 * check first non-zero iov alignment, the remaining
277 * iov alignment is checked inside bio loop below.
279 do {
280 addr = (unsigned long) iov[seg].iov_base;
281 count = min(iov[seg].iov_len, nbytes);
282 if (addr & blocksize_mask || count & blocksize_mask)
283 return -EINVAL;
284 } while (!count && ++seg < nr_segs);
285 atomic_set(bio_count, 1);
287 while (nbytes) {
288 /* roughly estimate number of bio vec needed */
289 nvec = (nbytes + PAGE_SIZE - 1) / PAGE_SIZE;
290 nvec = max(nvec, nr_segs - seg);
291 nvec = min(nvec, (unsigned long) BIO_MAX_PAGES);
293 /* bio_alloc should not fail with GFP_KERNEL flag */
294 bio = bio_alloc(GFP_KERNEL, nvec);
295 bio->bi_bdev = I_BDEV(inode);
296 bio->bi_end_io = blk_end_aio;
297 bio->bi_private = iocb;
298 bio->bi_sector = pos >> blkbits;
299 same_bio:
300 cur_off = addr & ~PAGE_MASK;
301 cur_len = PAGE_SIZE - cur_off;
302 if (count < cur_len)
303 cur_len = count;
305 page = blk_get_page(addr, count, rw, &pvec);
306 if (unlikely(IS_ERR(page)))
307 goto backout;
309 if (bio_add_page(bio, page, cur_len, cur_off)) {
310 pos += cur_len;
311 addr += cur_len;
312 count -= cur_len;
313 nbytes -= cur_len;
315 if (count)
316 goto same_bio;
317 while (++seg < nr_segs) {
318 addr = (unsigned long) iov[seg].iov_base;
319 count = iov[seg].iov_len;
320 if (!count)
321 continue;
322 if (unlikely(addr & blocksize_mask ||
323 count & blocksize_mask)) {
324 page = ERR_PTR(-EINVAL);
325 goto backout;
327 count = min(count, nbytes);
328 goto same_bio;
330 } else {
331 blk_unget_page(page, &pvec);
334 /* bio is ready, submit it */
335 if (rw == READ)
336 bio_set_pages_dirty(bio);
337 atomic_inc(bio_count);
338 submit_bio(rw, bio);
341 completion:
342 iocb->ki_left -= nbytes;
343 nbytes = iocb->ki_left;
344 iocb->ki_pos += nbytes;
346 blk_run_address_space(inode->i_mapping);
347 if (atomic_dec_and_test(bio_count))
348 aio_complete(iocb, nbytes, 0);
350 return -EIOCBQUEUED;
352 backout:
354 * back out nbytes count constructed so far for this bio,
355 * we will throw away current bio.
357 nbytes += bio->bi_size;
358 bio_release_pages(bio);
359 bio_put(bio);
362 * if no bio was submmitted, return the error code.
363 * otherwise, proceed with pending I/O completion.
365 if (atomic_read(bio_count) == 1)
366 return PTR_ERR(page);
367 goto completion;
369 #endif
371 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
373 return block_write_full_page(page, blkdev_get_block, wbc);
376 static int blkdev_readpage(struct file * file, struct page * page)
378 return block_read_full_page(page, blkdev_get_block);
381 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
382 loff_t pos, unsigned len, unsigned flags,
383 struct page **pagep, void **fsdata)
385 *pagep = NULL;
386 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
387 blkdev_get_block);
390 static int blkdev_write_end(struct file *file, struct address_space *mapping,
391 loff_t pos, unsigned len, unsigned copied,
392 struct page *page, void *fsdata)
394 int ret;
395 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
397 unlock_page(page);
398 page_cache_release(page);
400 return ret;
404 * private llseek:
405 * for a block special file file->f_path.dentry->d_inode->i_size is zero
406 * so we compute the size by hand (just as in block_read/write above)
408 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
410 struct inode *bd_inode = file->f_mapping->host;
411 loff_t size;
412 loff_t retval;
414 mutex_lock(&bd_inode->i_mutex);
415 size = i_size_read(bd_inode);
417 switch (origin) {
418 case 2:
419 offset += size;
420 break;
421 case 1:
422 offset += file->f_pos;
424 retval = -EINVAL;
425 if (offset >= 0 && offset <= size) {
426 if (offset != file->f_pos) {
427 file->f_pos = offset;
429 retval = offset;
431 mutex_unlock(&bd_inode->i_mutex);
432 return retval;
436 * Filp is never NULL; the only case when ->fsync() is called with
437 * NULL first argument is nfsd_sync_dir() and that's not a directory.
440 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
442 return sync_blockdev(I_BDEV(filp->f_mapping->host));
446 * pseudo-fs
449 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
450 static struct kmem_cache * bdev_cachep __read_mostly;
452 static struct inode *bdev_alloc_inode(struct super_block *sb)
454 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
455 if (!ei)
456 return NULL;
457 return &ei->vfs_inode;
460 static void bdev_destroy_inode(struct inode *inode)
462 struct bdev_inode *bdi = BDEV_I(inode);
464 bdi->bdev.bd_inode_backing_dev_info = NULL;
465 kmem_cache_free(bdev_cachep, bdi);
468 static void init_once(struct kmem_cache * cachep, void *foo)
470 struct bdev_inode *ei = (struct bdev_inode *) foo;
471 struct block_device *bdev = &ei->bdev;
473 memset(bdev, 0, sizeof(*bdev));
474 mutex_init(&bdev->bd_mutex);
475 sema_init(&bdev->bd_mount_sem, 1);
476 INIT_LIST_HEAD(&bdev->bd_inodes);
477 INIT_LIST_HEAD(&bdev->bd_list);
478 #ifdef CONFIG_SYSFS
479 INIT_LIST_HEAD(&bdev->bd_holder_list);
480 #endif
481 inode_init_once(&ei->vfs_inode);
484 static inline void __bd_forget(struct inode *inode)
486 list_del_init(&inode->i_devices);
487 inode->i_bdev = NULL;
488 inode->i_mapping = &inode->i_data;
491 static void bdev_clear_inode(struct inode *inode)
493 struct block_device *bdev = &BDEV_I(inode)->bdev;
494 struct list_head *p;
495 spin_lock(&bdev_lock);
496 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
497 __bd_forget(list_entry(p, struct inode, i_devices));
499 list_del_init(&bdev->bd_list);
500 spin_unlock(&bdev_lock);
503 static const struct super_operations bdev_sops = {
504 .statfs = simple_statfs,
505 .alloc_inode = bdev_alloc_inode,
506 .destroy_inode = bdev_destroy_inode,
507 .drop_inode = generic_delete_inode,
508 .clear_inode = bdev_clear_inode,
511 static int bd_get_sb(struct file_system_type *fs_type,
512 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
514 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
517 static struct file_system_type bd_type = {
518 .name = "bdev",
519 .get_sb = bd_get_sb,
520 .kill_sb = kill_anon_super,
523 static struct vfsmount *bd_mnt __read_mostly;
524 struct super_block *blockdev_superblock;
526 void __init bdev_cache_init(void)
528 int err;
529 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
530 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
531 SLAB_MEM_SPREAD|SLAB_PANIC),
532 init_once);
533 err = register_filesystem(&bd_type);
534 if (err)
535 panic("Cannot register bdev pseudo-fs");
536 bd_mnt = kern_mount(&bd_type);
537 err = PTR_ERR(bd_mnt);
538 if (IS_ERR(bd_mnt))
539 panic("Cannot create bdev pseudo-fs");
540 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
544 * Most likely _very_ bad one - but then it's hardly critical for small
545 * /dev and can be fixed when somebody will need really large one.
546 * Keep in mind that it will be fed through icache hash function too.
548 static inline unsigned long hash(dev_t dev)
550 return MAJOR(dev)+MINOR(dev);
553 static int bdev_test(struct inode *inode, void *data)
555 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
558 static int bdev_set(struct inode *inode, void *data)
560 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
561 return 0;
564 static LIST_HEAD(all_bdevs);
566 struct block_device *bdget(dev_t dev)
568 struct block_device *bdev;
569 struct inode *inode;
571 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
572 bdev_test, bdev_set, &dev);
574 if (!inode)
575 return NULL;
577 bdev = &BDEV_I(inode)->bdev;
579 if (inode->i_state & I_NEW) {
580 bdev->bd_contains = NULL;
581 bdev->bd_inode = inode;
582 bdev->bd_block_size = (1 << inode->i_blkbits);
583 bdev->bd_part_count = 0;
584 bdev->bd_invalidated = 0;
585 inode->i_mode = S_IFBLK;
586 inode->i_rdev = dev;
587 inode->i_bdev = bdev;
588 inode->i_data.a_ops = &def_blk_aops;
589 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
590 inode->i_data.backing_dev_info = &default_backing_dev_info;
591 spin_lock(&bdev_lock);
592 list_add(&bdev->bd_list, &all_bdevs);
593 spin_unlock(&bdev_lock);
594 unlock_new_inode(inode);
596 return bdev;
599 EXPORT_SYMBOL(bdget);
601 long nr_blockdev_pages(void)
603 struct block_device *bdev;
604 long ret = 0;
605 spin_lock(&bdev_lock);
606 list_for_each_entry(bdev, &all_bdevs, bd_list) {
607 ret += bdev->bd_inode->i_mapping->nrpages;
609 spin_unlock(&bdev_lock);
610 return ret;
613 void bdput(struct block_device *bdev)
615 iput(bdev->bd_inode);
618 EXPORT_SYMBOL(bdput);
620 static struct block_device *bd_acquire(struct inode *inode)
622 struct block_device *bdev;
624 spin_lock(&bdev_lock);
625 bdev = inode->i_bdev;
626 if (bdev) {
627 atomic_inc(&bdev->bd_inode->i_count);
628 spin_unlock(&bdev_lock);
629 return bdev;
631 spin_unlock(&bdev_lock);
633 bdev = bdget(inode->i_rdev);
634 if (bdev) {
635 spin_lock(&bdev_lock);
636 if (!inode->i_bdev) {
638 * We take an additional bd_inode->i_count for inode,
639 * and it's released in clear_inode() of inode.
640 * So, we can access it via ->i_mapping always
641 * without igrab().
643 atomic_inc(&bdev->bd_inode->i_count);
644 inode->i_bdev = bdev;
645 inode->i_mapping = bdev->bd_inode->i_mapping;
646 list_add(&inode->i_devices, &bdev->bd_inodes);
648 spin_unlock(&bdev_lock);
650 return bdev;
653 /* Call when you free inode */
655 void bd_forget(struct inode *inode)
657 struct block_device *bdev = NULL;
659 spin_lock(&bdev_lock);
660 if (inode->i_bdev) {
661 if (inode->i_sb != blockdev_superblock)
662 bdev = inode->i_bdev;
663 __bd_forget(inode);
665 spin_unlock(&bdev_lock);
667 if (bdev)
668 iput(bdev->bd_inode);
671 int bd_claim(struct block_device *bdev, void *holder)
673 int res;
674 spin_lock(&bdev_lock);
676 /* first decide result */
677 if (bdev->bd_holder == holder)
678 res = 0; /* already a holder */
679 else if (bdev->bd_holder != NULL)
680 res = -EBUSY; /* held by someone else */
681 else if (bdev->bd_contains == bdev)
682 res = 0; /* is a whole device which isn't held */
684 else if (bdev->bd_contains->bd_holder == bd_claim)
685 res = 0; /* is a partition of a device that is being partitioned */
686 else if (bdev->bd_contains->bd_holder != NULL)
687 res = -EBUSY; /* is a partition of a held device */
688 else
689 res = 0; /* is a partition of an un-held device */
691 /* now impose change */
692 if (res==0) {
693 /* note that for a whole device bd_holders
694 * will be incremented twice, and bd_holder will
695 * be set to bd_claim before being set to holder
697 bdev->bd_contains->bd_holders ++;
698 bdev->bd_contains->bd_holder = bd_claim;
699 bdev->bd_holders++;
700 bdev->bd_holder = holder;
702 spin_unlock(&bdev_lock);
703 return res;
706 EXPORT_SYMBOL(bd_claim);
708 void bd_release(struct block_device *bdev)
710 spin_lock(&bdev_lock);
711 if (!--bdev->bd_contains->bd_holders)
712 bdev->bd_contains->bd_holder = NULL;
713 if (!--bdev->bd_holders)
714 bdev->bd_holder = NULL;
715 spin_unlock(&bdev_lock);
718 EXPORT_SYMBOL(bd_release);
720 #ifdef CONFIG_SYSFS
722 * Functions for bd_claim_by_kobject / bd_release_from_kobject
724 * If a kobject is passed to bd_claim_by_kobject()
725 * and the kobject has a parent directory,
726 * following symlinks are created:
727 * o from the kobject to the claimed bdev
728 * o from "holders" directory of the bdev to the parent of the kobject
729 * bd_release_from_kobject() removes these symlinks.
731 * Example:
732 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
733 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
734 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
735 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
738 static struct kobject *bdev_get_kobj(struct block_device *bdev)
740 if (bdev->bd_contains != bdev)
741 return kobject_get(&bdev->bd_part->kobj);
742 else
743 return kobject_get(&bdev->bd_disk->kobj);
746 static struct kobject *bdev_get_holder(struct block_device *bdev)
748 if (bdev->bd_contains != bdev)
749 return kobject_get(bdev->bd_part->holder_dir);
750 else
751 return kobject_get(bdev->bd_disk->holder_dir);
754 static int add_symlink(struct kobject *from, struct kobject *to)
756 if (!from || !to)
757 return 0;
758 return sysfs_create_link(from, to, kobject_name(to));
761 static void del_symlink(struct kobject *from, struct kobject *to)
763 if (!from || !to)
764 return;
765 sysfs_remove_link(from, kobject_name(to));
769 * 'struct bd_holder' contains pointers to kobjects symlinked by
770 * bd_claim_by_kobject.
771 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
773 struct bd_holder {
774 struct list_head list; /* chain of holders of the bdev */
775 int count; /* references from the holder */
776 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
777 struct kobject *hdev; /* e.g. "/block/dm-0" */
778 struct kobject *hdir; /* e.g. "/block/sda/holders" */
779 struct kobject *sdev; /* e.g. "/block/sda" */
783 * Get references of related kobjects at once.
784 * Returns 1 on success. 0 on failure.
786 * Should call bd_holder_release_dirs() after successful use.
788 static int bd_holder_grab_dirs(struct block_device *bdev,
789 struct bd_holder *bo)
791 if (!bdev || !bo)
792 return 0;
794 bo->sdir = kobject_get(bo->sdir);
795 if (!bo->sdir)
796 return 0;
798 bo->hdev = kobject_get(bo->sdir->parent);
799 if (!bo->hdev)
800 goto fail_put_sdir;
802 bo->sdev = bdev_get_kobj(bdev);
803 if (!bo->sdev)
804 goto fail_put_hdev;
806 bo->hdir = bdev_get_holder(bdev);
807 if (!bo->hdir)
808 goto fail_put_sdev;
810 return 1;
812 fail_put_sdev:
813 kobject_put(bo->sdev);
814 fail_put_hdev:
815 kobject_put(bo->hdev);
816 fail_put_sdir:
817 kobject_put(bo->sdir);
819 return 0;
822 /* Put references of related kobjects at once. */
823 static void bd_holder_release_dirs(struct bd_holder *bo)
825 kobject_put(bo->hdir);
826 kobject_put(bo->sdev);
827 kobject_put(bo->hdev);
828 kobject_put(bo->sdir);
831 static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
833 struct bd_holder *bo;
835 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
836 if (!bo)
837 return NULL;
839 bo->count = 1;
840 bo->sdir = kobj;
842 return bo;
845 static void free_bd_holder(struct bd_holder *bo)
847 kfree(bo);
851 * find_bd_holder - find matching struct bd_holder from the block device
853 * @bdev: struct block device to be searched
854 * @bo: target struct bd_holder
856 * Returns matching entry with @bo in @bdev->bd_holder_list.
857 * If found, increment the reference count and return the pointer.
858 * If not found, returns NULL.
860 static struct bd_holder *find_bd_holder(struct block_device *bdev,
861 struct bd_holder *bo)
863 struct bd_holder *tmp;
865 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
866 if (tmp->sdir == bo->sdir) {
867 tmp->count++;
868 return tmp;
871 return NULL;
875 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
877 * @bdev: block device to be bd_claimed
878 * @bo: preallocated and initialized by alloc_bd_holder()
880 * Add @bo to @bdev->bd_holder_list, create symlinks.
882 * Returns 0 if symlinks are created.
883 * Returns -ve if something fails.
885 static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
887 int err;
889 if (!bo)
890 return -EINVAL;
892 if (!bd_holder_grab_dirs(bdev, bo))
893 return -EBUSY;
895 err = add_symlink(bo->sdir, bo->sdev);
896 if (err)
897 return err;
899 err = add_symlink(bo->hdir, bo->hdev);
900 if (err) {
901 del_symlink(bo->sdir, bo->sdev);
902 return err;
905 list_add_tail(&bo->list, &bdev->bd_holder_list);
906 return 0;
910 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
912 * @bdev: block device to be bd_claimed
913 * @kobj: holder's kobject
915 * If there is matching entry with @kobj in @bdev->bd_holder_list
916 * and no other bd_claim() from the same kobject,
917 * remove the struct bd_holder from the list, delete symlinks for it.
919 * Returns a pointer to the struct bd_holder when it's removed from the list
920 * and ready to be freed.
921 * Returns NULL if matching claim isn't found or there is other bd_claim()
922 * by the same kobject.
924 static struct bd_holder *del_bd_holder(struct block_device *bdev,
925 struct kobject *kobj)
927 struct bd_holder *bo;
929 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
930 if (bo->sdir == kobj) {
931 bo->count--;
932 BUG_ON(bo->count < 0);
933 if (!bo->count) {
934 list_del(&bo->list);
935 del_symlink(bo->sdir, bo->sdev);
936 del_symlink(bo->hdir, bo->hdev);
937 bd_holder_release_dirs(bo);
938 return bo;
940 break;
944 return NULL;
948 * bd_claim_by_kobject - bd_claim() with additional kobject signature
950 * @bdev: block device to be claimed
951 * @holder: holder's signature
952 * @kobj: holder's kobject
954 * Do bd_claim() and if it succeeds, create sysfs symlinks between
955 * the bdev and the holder's kobject.
956 * Use bd_release_from_kobject() when relesing the claimed bdev.
958 * Returns 0 on success. (same as bd_claim())
959 * Returns errno on failure.
961 static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
962 struct kobject *kobj)
964 int err;
965 struct bd_holder *bo, *found;
967 if (!kobj)
968 return -EINVAL;
970 bo = alloc_bd_holder(kobj);
971 if (!bo)
972 return -ENOMEM;
974 mutex_lock(&bdev->bd_mutex);
976 err = bd_claim(bdev, holder);
977 if (err)
978 goto fail;
980 found = find_bd_holder(bdev, bo);
981 if (found)
982 goto fail;
984 err = add_bd_holder(bdev, bo);
985 if (err)
986 bd_release(bdev);
987 else
988 bo = NULL;
989 fail:
990 mutex_unlock(&bdev->bd_mutex);
991 free_bd_holder(bo);
992 return err;
996 * bd_release_from_kobject - bd_release() with additional kobject signature
998 * @bdev: block device to be released
999 * @kobj: holder's kobject
1001 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1003 static void bd_release_from_kobject(struct block_device *bdev,
1004 struct kobject *kobj)
1006 if (!kobj)
1007 return;
1009 mutex_lock(&bdev->bd_mutex);
1010 bd_release(bdev);
1011 free_bd_holder(del_bd_holder(bdev, kobj));
1012 mutex_unlock(&bdev->bd_mutex);
1016 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1018 * @bdev: block device to be claimed
1019 * @holder: holder's signature
1020 * @disk: holder's gendisk
1022 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1024 int bd_claim_by_disk(struct block_device *bdev, void *holder,
1025 struct gendisk *disk)
1027 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1029 EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1032 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1034 * @bdev: block device to be claimed
1035 * @disk: holder's gendisk
1037 * Call bd_release_from_kobject() and put @disk->slave_dir.
1039 void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1041 bd_release_from_kobject(bdev, disk->slave_dir);
1042 kobject_put(disk->slave_dir);
1044 EXPORT_SYMBOL_GPL(bd_release_from_disk);
1045 #endif
1048 * Tries to open block device by device number. Use it ONLY if you
1049 * really do not have anything better - i.e. when you are behind a
1050 * truly sucky interface and all you are given is a device number. _Never_
1051 * to be used for internal purposes. If you ever need it - reconsider
1052 * your API.
1054 struct block_device *open_by_devnum(dev_t dev, unsigned mode)
1056 struct block_device *bdev = bdget(dev);
1057 int err = -ENOMEM;
1058 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
1059 if (bdev)
1060 err = blkdev_get(bdev, mode, flags);
1061 return err ? ERR_PTR(err) : bdev;
1064 EXPORT_SYMBOL(open_by_devnum);
1067 * This routine checks whether a removable media has been changed,
1068 * and invalidates all buffer-cache-entries in that case. This
1069 * is a relatively slow routine, so we have to try to minimize using
1070 * it. Thus it is called only upon a 'mount' or 'open'. This
1071 * is the best way of combining speed and utility, I think.
1072 * People changing diskettes in the middle of an operation deserve
1073 * to lose :-)
1075 int check_disk_change(struct block_device *bdev)
1077 struct gendisk *disk = bdev->bd_disk;
1078 struct block_device_operations * bdops = disk->fops;
1080 if (!bdops->media_changed)
1081 return 0;
1082 if (!bdops->media_changed(bdev->bd_disk))
1083 return 0;
1085 if (__invalidate_device(bdev))
1086 printk("VFS: busy inodes on changed media.\n");
1088 if (bdops->revalidate_disk)
1089 bdops->revalidate_disk(bdev->bd_disk);
1090 if (bdev->bd_disk->minors > 1)
1091 bdev->bd_invalidated = 1;
1092 return 1;
1095 EXPORT_SYMBOL(check_disk_change);
1097 void bd_set_size(struct block_device *bdev, loff_t size)
1099 unsigned bsize = bdev_hardsect_size(bdev);
1101 bdev->bd_inode->i_size = size;
1102 while (bsize < PAGE_CACHE_SIZE) {
1103 if (size & bsize)
1104 break;
1105 bsize <<= 1;
1107 bdev->bd_block_size = bsize;
1108 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1110 EXPORT_SYMBOL(bd_set_size);
1112 static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1113 int for_part);
1114 static int __blkdev_put(struct block_device *bdev, int for_part);
1117 * bd_mutex locking:
1119 * mutex_lock(part->bd_mutex)
1120 * mutex_lock_nested(whole->bd_mutex, 1)
1123 static int do_open(struct block_device *bdev, struct file *file, int for_part)
1125 struct module *owner = NULL;
1126 struct gendisk *disk;
1127 int ret = -ENXIO;
1128 int part;
1130 file->f_mapping = bdev->bd_inode->i_mapping;
1131 lock_kernel();
1132 disk = get_gendisk(bdev->bd_dev, &part);
1133 if (!disk) {
1134 unlock_kernel();
1135 bdput(bdev);
1136 return ret;
1138 owner = disk->fops->owner;
1140 mutex_lock_nested(&bdev->bd_mutex, for_part);
1141 if (!bdev->bd_openers) {
1142 bdev->bd_disk = disk;
1143 bdev->bd_contains = bdev;
1144 if (!part) {
1145 struct backing_dev_info *bdi;
1146 if (disk->fops->open) {
1147 ret = disk->fops->open(bdev->bd_inode, file);
1148 if (ret)
1149 goto out_first;
1151 if (!bdev->bd_openers) {
1152 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1153 bdi = blk_get_backing_dev_info(bdev);
1154 if (bdi == NULL)
1155 bdi = &default_backing_dev_info;
1156 bdev->bd_inode->i_data.backing_dev_info = bdi;
1158 if (bdev->bd_invalidated)
1159 rescan_partitions(disk, bdev);
1160 } else {
1161 struct hd_struct *p;
1162 struct block_device *whole;
1163 whole = bdget_disk(disk, 0);
1164 ret = -ENOMEM;
1165 if (!whole)
1166 goto out_first;
1167 BUG_ON(for_part);
1168 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
1169 if (ret)
1170 goto out_first;
1171 bdev->bd_contains = whole;
1172 p = disk->part[part - 1];
1173 bdev->bd_inode->i_data.backing_dev_info =
1174 whole->bd_inode->i_data.backing_dev_info;
1175 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
1176 ret = -ENXIO;
1177 goto out_first;
1179 kobject_get(&p->kobj);
1180 bdev->bd_part = p;
1181 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
1183 } else {
1184 put_disk(disk);
1185 module_put(owner);
1186 if (bdev->bd_contains == bdev) {
1187 if (bdev->bd_disk->fops->open) {
1188 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1189 if (ret)
1190 goto out;
1192 if (bdev->bd_invalidated)
1193 rescan_partitions(bdev->bd_disk, bdev);
1196 bdev->bd_openers++;
1197 if (for_part)
1198 bdev->bd_part_count++;
1199 mutex_unlock(&bdev->bd_mutex);
1200 unlock_kernel();
1201 return 0;
1203 out_first:
1204 bdev->bd_disk = NULL;
1205 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1206 if (bdev != bdev->bd_contains)
1207 __blkdev_put(bdev->bd_contains, 1);
1208 bdev->bd_contains = NULL;
1209 put_disk(disk);
1210 module_put(owner);
1211 out:
1212 mutex_unlock(&bdev->bd_mutex);
1213 unlock_kernel();
1214 if (ret)
1215 bdput(bdev);
1216 return ret;
1219 static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1220 int for_part)
1223 * This crockload is due to bad choice of ->open() type.
1224 * It will go away.
1225 * For now, block device ->open() routine must _not_
1226 * examine anything in 'inode' argument except ->i_rdev.
1228 struct file fake_file = {};
1229 struct dentry fake_dentry = {};
1230 fake_file.f_mode = mode;
1231 fake_file.f_flags = flags;
1232 fake_file.f_path.dentry = &fake_dentry;
1233 fake_dentry.d_inode = bdev->bd_inode;
1235 return do_open(bdev, &fake_file, for_part);
1238 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1240 return __blkdev_get(bdev, mode, flags, 0);
1242 EXPORT_SYMBOL(blkdev_get);
1244 static int blkdev_open(struct inode * inode, struct file * filp)
1246 struct block_device *bdev;
1247 int res;
1250 * Preserve backwards compatibility and allow large file access
1251 * even if userspace doesn't ask for it explicitly. Some mkfs
1252 * binary needs it. We might want to drop this workaround
1253 * during an unstable branch.
1255 filp->f_flags |= O_LARGEFILE;
1257 bdev = bd_acquire(inode);
1258 if (bdev == NULL)
1259 return -ENOMEM;
1261 res = do_open(bdev, filp, 0);
1262 if (res)
1263 return res;
1265 if (!(filp->f_flags & O_EXCL) )
1266 return 0;
1268 if (!(res = bd_claim(bdev, filp)))
1269 return 0;
1271 blkdev_put(bdev);
1272 return res;
1275 static int __blkdev_put(struct block_device *bdev, int for_part)
1277 int ret = 0;
1278 struct inode *bd_inode = bdev->bd_inode;
1279 struct gendisk *disk = bdev->bd_disk;
1280 struct block_device *victim = NULL;
1282 mutex_lock_nested(&bdev->bd_mutex, for_part);
1283 lock_kernel();
1284 if (for_part)
1285 bdev->bd_part_count--;
1287 if (!--bdev->bd_openers) {
1288 sync_blockdev(bdev);
1289 kill_bdev(bdev);
1291 if (bdev->bd_contains == bdev) {
1292 if (disk->fops->release)
1293 ret = disk->fops->release(bd_inode, NULL);
1295 if (!bdev->bd_openers) {
1296 struct module *owner = disk->fops->owner;
1298 put_disk(disk);
1299 module_put(owner);
1301 if (bdev->bd_contains != bdev) {
1302 kobject_put(&bdev->bd_part->kobj);
1303 bdev->bd_part = NULL;
1305 bdev->bd_disk = NULL;
1306 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1307 if (bdev != bdev->bd_contains)
1308 victim = bdev->bd_contains;
1309 bdev->bd_contains = NULL;
1311 unlock_kernel();
1312 mutex_unlock(&bdev->bd_mutex);
1313 bdput(bdev);
1314 if (victim)
1315 __blkdev_put(victim, 1);
1316 return ret;
1319 int blkdev_put(struct block_device *bdev)
1321 return __blkdev_put(bdev, 0);
1323 EXPORT_SYMBOL(blkdev_put);
1325 static int blkdev_close(struct inode * inode, struct file * filp)
1327 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1328 if (bdev->bd_holder == filp)
1329 bd_release(bdev);
1330 return blkdev_put(bdev);
1333 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1335 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1338 const struct address_space_operations def_blk_aops = {
1339 .readpage = blkdev_readpage,
1340 .writepage = blkdev_writepage,
1341 .sync_page = block_sync_page,
1342 .write_begin = blkdev_write_begin,
1343 .write_end = blkdev_write_end,
1344 .writepages = generic_writepages,
1345 .direct_IO = blkdev_direct_IO,
1348 const struct file_operations def_blk_fops = {
1349 .open = blkdev_open,
1350 .release = blkdev_close,
1351 .llseek = block_llseek,
1352 .read = do_sync_read,
1353 .write = do_sync_write,
1354 .aio_read = generic_file_aio_read,
1355 .aio_write = generic_file_aio_write_nolock,
1356 .mmap = generic_file_mmap,
1357 .fsync = block_fsync,
1358 .unlocked_ioctl = block_ioctl,
1359 #ifdef CONFIG_COMPAT
1360 .compat_ioctl = compat_blkdev_ioctl,
1361 #endif
1362 .splice_read = generic_file_splice_read,
1363 .splice_write = generic_file_splice_write,
1366 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1368 int res;
1369 mm_segment_t old_fs = get_fs();
1370 set_fs(KERNEL_DS);
1371 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1372 set_fs(old_fs);
1373 return res;
1376 EXPORT_SYMBOL(ioctl_by_bdev);
1379 * lookup_bdev - lookup a struct block_device by name
1381 * @path: special file representing the block device
1383 * Get a reference to the blockdevice at @path in the current
1384 * namespace if possible and return it. Return ERR_PTR(error)
1385 * otherwise.
1387 struct block_device *lookup_bdev(const char *path)
1389 struct block_device *bdev;
1390 struct inode *inode;
1391 struct nameidata nd;
1392 int error;
1394 if (!path || !*path)
1395 return ERR_PTR(-EINVAL);
1397 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1398 if (error)
1399 return ERR_PTR(error);
1401 inode = nd.dentry->d_inode;
1402 error = -ENOTBLK;
1403 if (!S_ISBLK(inode->i_mode))
1404 goto fail;
1405 error = -EACCES;
1406 if (nd.mnt->mnt_flags & MNT_NODEV)
1407 goto fail;
1408 error = -ENOMEM;
1409 bdev = bd_acquire(inode);
1410 if (!bdev)
1411 goto fail;
1412 out:
1413 path_release(&nd);
1414 return bdev;
1415 fail:
1416 bdev = ERR_PTR(error);
1417 goto out;
1421 * open_bdev_excl - open a block device by name and set it up for use
1423 * @path: special file representing the block device
1424 * @flags: %MS_RDONLY for opening read-only
1425 * @holder: owner for exclusion
1427 * Open the blockdevice described by the special file at @path, claim it
1428 * for the @holder.
1430 struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1432 struct block_device *bdev;
1433 mode_t mode = FMODE_READ;
1434 int error = 0;
1436 bdev = lookup_bdev(path);
1437 if (IS_ERR(bdev))
1438 return bdev;
1440 if (!(flags & MS_RDONLY))
1441 mode |= FMODE_WRITE;
1442 error = blkdev_get(bdev, mode, 0);
1443 if (error)
1444 return ERR_PTR(error);
1445 error = -EACCES;
1446 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1447 goto blkdev_put;
1448 error = bd_claim(bdev, holder);
1449 if (error)
1450 goto blkdev_put;
1452 return bdev;
1454 blkdev_put:
1455 blkdev_put(bdev);
1456 return ERR_PTR(error);
1459 EXPORT_SYMBOL(open_bdev_excl);
1462 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1464 * @bdev: blockdevice to close
1466 * This is the counterpart to open_bdev_excl().
1468 void close_bdev_excl(struct block_device *bdev)
1470 bd_release(bdev);
1471 blkdev_put(bdev);
1474 EXPORT_SYMBOL(close_bdev_excl);
1476 int __invalidate_device(struct block_device *bdev)
1478 struct super_block *sb = get_super(bdev);
1479 int res = 0;
1481 if (sb) {
1483 * no need to lock the super, get_super holds the
1484 * read mutex so the filesystem cannot go away
1485 * under us (->put_super runs with the write lock
1486 * hold).
1488 shrink_dcache_sb(sb);
1489 res = invalidate_inodes(sb);
1490 drop_super(sb);
1492 invalidate_bdev(bdev);
1493 return res;
1495 EXPORT_SYMBOL(__invalidate_device);