4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
8 #include <linux/config.h>
9 #include <linux/init.h>
11 #include <linux/locks.h>
12 #include <linux/fcntl.h>
13 #include <linux/slab.h>
14 #include <linux/kmod.h>
15 #include <linux/major.h>
16 #include <linux/devfs_fs_kernel.h>
17 #include <linux/smp_lock.h>
18 #include <linux/iobuf.h>
19 #include <linux/highmem.h>
20 #include <linux/blkdev.h>
21 #include <linux/module.h>
23 #include <asm/uaccess.h>
25 static unsigned long max_block(kdev_t dev)
27 unsigned int retval = ~0U;
28 int major = MAJOR(dev);
30 if (blk_size[major]) {
31 int minor = MINOR(dev);
32 unsigned int blocks = blk_size[major][minor];
34 unsigned int size = block_size(dev);
35 unsigned int sizebits = blksize_bits(size);
36 blocks += (size-1) >> BLOCK_SIZE_BITS;
37 retval = blocks << (BLOCK_SIZE_BITS - sizebits);
38 if (sizebits > BLOCK_SIZE_BITS)
39 retval = blocks >> (sizebits - BLOCK_SIZE_BITS);
45 static loff_t blkdev_size(kdev_t dev)
47 unsigned int blocks = ~0U;
48 int major = MAJOR(dev);
50 if (blk_size[major]) {
51 int minor = MINOR(dev);
52 blocks = blk_size[major][minor];
54 return (loff_t) blocks << BLOCK_SIZE_BITS;
57 /* Kill _all_ buffers, dirty or not.. */
58 static void kill_bdev(struct block_device *bdev)
60 invalidate_bdev(bdev, 1);
61 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
64 int set_blocksize(kdev_t dev, int size)
67 struct block_device *bdev;
69 /* Size must be a power of two, and between 512 and PAGE_SIZE */
70 if (size > PAGE_SIZE || size < 512 || (size & (size-1)))
73 /* Size cannot be smaller than the size supported by the device */
74 if (size < get_hardsect_size(dev))
77 /* No blocksize array? Implies hardcoded BLOCK_SIZE */
78 if (!blksize_size[MAJOR(dev)]) {
79 if (size == BLOCK_SIZE)
84 oldsize = blksize_size[MAJOR(dev)][MINOR(dev)];
88 if (!oldsize && size == BLOCK_SIZE) {
89 blksize_size[MAJOR(dev)][MINOR(dev)] = size;
93 /* Ok, we're actually changing the blocksize.. */
96 blksize_size[MAJOR(dev)][MINOR(dev)] = size;
97 bdev->bd_inode->i_blkbits = blksize_bits(size);
103 int sb_set_blocksize(struct super_block *sb, int size)
106 if (set_blocksize(sb->s_dev, size) < 0)
108 sb->s_blocksize = size;
109 for (bits = 9, size >>= 9; size >>= 1; bits++)
111 sb->s_blocksize_bits = bits;
112 return sb->s_blocksize;
115 int sb_min_blocksize(struct super_block *sb, int size)
117 int minsize = get_hardsect_size(sb->s_dev);
120 return sb_set_blocksize(sb, size);
123 static int blkdev_get_block(struct inode * inode, long iblock, struct buffer_head * bh, int create)
125 if (iblock >= max_block(inode->i_rdev))
128 bh->b_dev = inode->i_rdev;
129 bh->b_blocknr = iblock;
130 bh->b_state |= 1UL << BH_Mapped;
134 static int blkdev_direct_IO(int rw, struct inode * inode, struct kiobuf * iobuf, unsigned long blocknr, int blocksize)
136 return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize, blkdev_get_block);
139 static int blkdev_writepage(struct page * page)
141 return block_write_full_page(page, blkdev_get_block);
144 static int blkdev_readpage(struct file * file, struct page * page)
146 return block_read_full_page(page, blkdev_get_block);
149 static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
151 return block_prepare_write(page, from, to, blkdev_get_block);
154 static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
156 return block_commit_write(page, from, to);
161 * for a block special file file->f_dentry->d_inode->i_size is zero
162 * so we compute the size by hand (just as in block_read/write above)
164 static loff_t block_llseek(struct file *file, loff_t offset, int origin)
167 loff_t size = file->f_dentry->d_inode->i_bdev->bd_inode->i_size;
175 offset += file->f_pos;
178 if (offset >= 0 && offset <= size) {
179 if (offset != file->f_pos) {
180 file->f_pos = offset;
182 file->f_version = ++event;
190 static int __block_fsync(struct inode * inode)
194 ret = filemap_fdatasync(inode->i_mapping);
195 err = sync_buffers(inode->i_rdev, 1);
198 err = filemap_fdatawait(inode->i_mapping);
206 * Filp may be NULL when we are called by an msync of a vma
207 * since the vma has no handle.
210 static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
212 struct inode * inode = dentry->d_inode;
214 return __block_fsync(inode);
221 static struct super_block *bd_read_super(struct super_block *sb, void *data, int silent)
223 static struct super_operations sops = {};
224 struct inode *root = new_inode(sb);
227 root->i_mode = S_IFDIR | S_IRUSR | S_IWUSR;
228 root->i_uid = root->i_gid = 0;
229 root->i_atime = root->i_mtime = root->i_ctime = CURRENT_TIME;
230 sb->s_maxbytes = ~0ULL;
231 sb->s_blocksize = 1024;
232 sb->s_blocksize_bits = 10;
233 sb->s_magic = 0x62646576;
235 sb->s_root = d_alloc(NULL, &(const struct qstr) { "bdev:", 5, 0 });
240 sb->s_root->d_sb = sb;
241 sb->s_root->d_parent = sb->s_root;
242 d_instantiate(sb->s_root, root);
246 static DECLARE_FSTYPE(bd_type, "bdev", bd_read_super, FS_NOMOUNT);
248 static struct vfsmount *bd_mnt;
251 * bdev cache handling - shamelessly stolen from inode.c
252 * We use smaller hashtable, though.
256 #define HASH_SIZE (1UL << HASH_BITS)
257 #define HASH_MASK (HASH_SIZE-1)
258 static struct list_head bdev_hashtable[HASH_SIZE];
259 static spinlock_t bdev_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
260 static kmem_cache_t * bdev_cachep;
262 #define alloc_bdev() \
263 ((struct block_device *) kmem_cache_alloc(bdev_cachep, SLAB_KERNEL))
264 #define destroy_bdev(bdev) kmem_cache_free(bdev_cachep, (bdev))
266 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
268 struct block_device * bdev = (struct block_device *) foo;
270 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
271 SLAB_CTOR_CONSTRUCTOR)
273 memset(bdev, 0, sizeof(*bdev));
274 sema_init(&bdev->bd_sem, 1);
275 INIT_LIST_HEAD(&bdev->bd_inodes);
279 void __init bdev_cache_init(void)
282 struct list_head *head = bdev_hashtable;
286 INIT_LIST_HEAD(head);
291 bdev_cachep = kmem_cache_create("bdev_cache",
292 sizeof(struct block_device),
293 0, SLAB_HWCACHE_ALIGN, init_once,
296 panic("Cannot create bdev_cache SLAB cache");
297 err = register_filesystem(&bd_type);
299 panic("Cannot register bdev pseudo-fs");
300 bd_mnt = kern_mount(&bd_type);
301 err = PTR_ERR(bd_mnt);
303 panic("Cannot create bdev pseudo-fs");
307 * Most likely _very_ bad one - but then it's hardly critical for small
308 * /dev and can be fixed when somebody will need really large one.
310 static inline unsigned long hash(dev_t dev)
312 unsigned long tmp = dev;
313 tmp = tmp + (tmp >> HASH_BITS) + (tmp >> HASH_BITS*2);
314 return tmp & HASH_MASK;
317 static struct block_device *bdfind(dev_t dev, struct list_head *head)
320 struct block_device *bdev;
321 for (p=head->next; p!=head; p=p->next) {
322 bdev = list_entry(p, struct block_device, bd_hash);
323 if (bdev->bd_dev != dev)
325 atomic_inc(&bdev->bd_count);
331 struct block_device *bdget(dev_t dev)
333 struct list_head * head = bdev_hashtable + hash(dev);
334 struct block_device *bdev, *new_bdev;
335 spin_lock(&bdev_lock);
336 bdev = bdfind(dev, head);
337 spin_unlock(&bdev_lock);
340 new_bdev = alloc_bdev();
342 struct inode *inode = new_inode(bd_mnt->mnt_sb);
344 kdev_t kdev = to_kdev_t(dev);
345 atomic_set(&new_bdev->bd_count,1);
346 new_bdev->bd_dev = dev;
347 new_bdev->bd_op = NULL;
348 new_bdev->bd_inode = inode;
349 inode->i_rdev = kdev;
351 inode->i_bdev = new_bdev;
352 inode->i_data.a_ops = &def_blk_aops;
353 inode->i_data.gfp_mask = GFP_USER;
354 inode->i_mode = S_IFBLK;
355 spin_lock(&bdev_lock);
356 bdev = bdfind(dev, head);
358 list_add(&new_bdev->bd_hash, head);
359 spin_unlock(&bdev_lock);
362 spin_unlock(&bdev_lock);
363 iput(new_bdev->bd_inode);
365 destroy_bdev(new_bdev);
370 static inline void __bd_forget(struct inode *inode)
372 list_del_init(&inode->i_devices);
373 inode->i_bdev = NULL;
374 inode->i_mapping = &inode->i_data;
377 void bdput(struct block_device *bdev)
379 if (atomic_dec_and_lock(&bdev->bd_count, &bdev_lock)) {
381 if (bdev->bd_openers)
383 list_del(&bdev->bd_hash);
384 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
385 __bd_forget(list_entry(p, struct inode, i_devices));
387 spin_unlock(&bdev_lock);
388 iput(bdev->bd_inode);
393 int bd_acquire(struct inode *inode)
395 struct block_device *bdev;
396 spin_lock(&bdev_lock);
398 atomic_inc(&inode->i_bdev->bd_count);
399 spin_unlock(&bdev_lock);
402 spin_unlock(&bdev_lock);
403 bdev = bdget(kdev_t_to_nr(inode->i_rdev));
406 spin_lock(&bdev_lock);
407 if (!inode->i_bdev) {
408 inode->i_bdev = bdev;
409 inode->i_mapping = bdev->bd_inode->i_mapping;
410 list_add(&inode->i_devices, &bdev->bd_inodes);
411 } else if (inode->i_bdev != bdev)
413 spin_unlock(&bdev_lock);
417 /* Call when you free inode */
419 void bd_forget(struct inode *inode)
421 spin_lock(&bdev_lock);
424 spin_unlock(&bdev_lock);
429 struct block_device_operations *bdops;
430 } blkdevs[MAX_BLKDEV];
432 int get_blkdev_list(char * p)
437 len = sprintf(p, "\nBlock devices:\n");
438 for (i = 0; i < MAX_BLKDEV ; i++) {
439 if (blkdevs[i].bdops) {
440 len += sprintf(p+len, "%3d %s\n", i, blkdevs[i].name);
447 Return the function table of a device.
448 Load the driver if needed.
450 const struct block_device_operations * get_blkfops(unsigned int major)
452 const struct block_device_operations *ret = NULL;
454 /* major 0 is used for non-device mounts */
455 if (major && major < MAX_BLKDEV) {
457 if (!blkdevs[major].bdops) {
459 sprintf(name, "block-major-%d", major);
460 request_module(name);
463 ret = blkdevs[major].bdops;
468 int register_blkdev(unsigned int major, const char * name, struct block_device_operations *bdops)
471 for (major = MAX_BLKDEV-1; major > 0; major--) {
472 if (blkdevs[major].bdops == NULL) {
473 blkdevs[major].name = name;
474 blkdevs[major].bdops = bdops;
480 if (major >= MAX_BLKDEV)
482 if (blkdevs[major].bdops && blkdevs[major].bdops != bdops)
484 blkdevs[major].name = name;
485 blkdevs[major].bdops = bdops;
489 int unregister_blkdev(unsigned int major, const char * name)
491 if (major >= MAX_BLKDEV)
493 if (!blkdevs[major].bdops)
495 if (strcmp(blkdevs[major].name, name))
497 blkdevs[major].name = NULL;
498 blkdevs[major].bdops = NULL;
503 * This routine checks whether a removable media has been changed,
504 * and invalidates all buffer-cache-entries in that case. This
505 * is a relatively slow routine, so we have to try to minimize using
506 * it. Thus it is called only upon a 'mount' or 'open'. This
507 * is the best way of combining speed and utility, I think.
508 * People changing diskettes in the middle of an operation deserve
511 int check_disk_change(kdev_t dev)
514 const struct block_device_operations * bdops = NULL;
518 bdops = blkdevs[i].bdops;
522 de = devfs_find_handle (NULL, NULL, i, MINOR (dev),
523 DEVFS_SPECIAL_BLK, 0);
525 bdops = devfs_get_ops (de);
526 devfs_put_ops (de); /* We're running in owner module */
531 if (bdops->check_media_change == NULL)
533 if (!bdops->check_media_change(dev))
536 if (invalidate_device(dev, 0))
537 printk("VFS: busy inodes on changed media.\n");
539 if (bdops->revalidate)
540 bdops->revalidate(dev);
544 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
547 mm_segment_t old_fs = get_fs();
549 if (!bdev->bd_op->ioctl)
552 res = bdev->bd_op->ioctl(bdev->bd_inode, NULL, cmd, arg);
557 static int do_open(struct block_device *bdev, struct inode *inode, struct file *file)
560 kdev_t dev = to_kdev_t(bdev->bd_dev);
565 bdev->bd_op = get_blkfops(MAJOR(dev));
568 if (bdev->bd_op->owner)
569 __MOD_INC_USE_COUNT(bdev->bd_op->owner);
570 if (bdev->bd_op->open)
571 ret = bdev->bd_op->open(inode, file);
574 bdev->bd_inode->i_size = blkdev_size(dev);
575 bdev->bd_inode->i_blkbits = blksize_bits(block_size(dev));
577 if (bdev->bd_op->owner)
578 __MOD_DEC_USE_COUNT(bdev->bd_op->owner);
579 if (!bdev->bd_openers)
590 int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags, int kind)
593 * This crockload is due to bad choice of ->open() type.
595 * For now, block device ->open() routine must _not_
596 * examine anything in 'inode' argument except ->i_rdev.
598 struct file fake_file = {};
599 struct dentry fake_dentry = {};
600 fake_file.f_mode = mode;
601 fake_file.f_flags = flags;
602 fake_file.f_dentry = &fake_dentry;
603 fake_dentry.d_inode = bdev->bd_inode;
605 return do_open(bdev, bdev->bd_inode, &fake_file);
608 int blkdev_open(struct inode * inode, struct file * filp)
610 struct block_device *bdev;
613 * Preserve backwards compatibility and allow large file access
614 * even if userspace doesn't ask for it explicitly. Some mkfs
615 * binary needs it. We might want to drop this workaround
616 * during an unstable branch.
618 filp->f_flags |= O_LARGEFILE;
621 bdev = inode->i_bdev;
623 return do_open(bdev, inode, filp);
626 int blkdev_put(struct block_device *bdev, int kind)
629 kdev_t rdev = to_kdev_t(bdev->bd_dev); /* this should become bdev */
630 struct inode *bd_inode = bdev->bd_inode;
634 if (kind == BDEV_FILE && bdev->bd_openers == 1)
635 __block_fsync(bd_inode);
636 else if (kind == BDEV_FS)
637 fsync_no_super(rdev);
638 if (!--bdev->bd_openers)
640 if (bdev->bd_op->release)
641 ret = bdev->bd_op->release(bd_inode, NULL);
642 if (bdev->bd_op->owner)
643 __MOD_DEC_USE_COUNT(bdev->bd_op->owner);
644 if (!bdev->bd_openers)
652 int blkdev_close(struct inode * inode, struct file * filp)
654 return blkdev_put(inode->i_bdev, BDEV_FILE);
657 static int blkdev_ioctl(struct inode *inode, struct file *file, unsigned cmd,
660 if (inode->i_bdev->bd_op->ioctl)
661 return inode->i_bdev->bd_op->ioctl(inode, file, cmd, arg);
665 struct address_space_operations def_blk_aops = {
666 readpage: blkdev_readpage,
667 writepage: blkdev_writepage,
668 sync_page: block_sync_page,
669 prepare_write: blkdev_prepare_write,
670 commit_write: blkdev_commit_write,
671 direct_IO: blkdev_direct_IO,
674 struct file_operations def_blk_fops = {
676 release: blkdev_close,
677 llseek: block_llseek,
678 read: generic_file_read,
679 write: generic_file_write,
680 mmap: generic_file_mmap,
685 const char * bdevname(kdev_t dev)
687 static char buffer[32];
688 const char * name = blkdevs[MAJOR(dev)].name;
691 name = "unknown-block";
693 sprintf(buffer, "%s(%d,%d)", name, MAJOR(dev), MINOR(dev));