blob: ea1480a16f517d85d12de02bfde8afa53a43037c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#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>
David Howells811d7362006-08-29 19:06:09 +010020#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mpage.h>
22#include <linux/mount.h>
23#include <linux/uio.h>
24#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070025#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
29struct bdev_inode {
30 struct block_device bdev;
31 struct inode vfs_inode;
32};
33
34static inline struct bdev_inode *BDEV_I(struct inode *inode)
35{
36 return container_of(inode, struct bdev_inode, vfs_inode);
37}
38
39inline struct block_device *I_BDEV(struct inode *inode)
40{
41 return &BDEV_I(inode)->bdev;
42}
43
44EXPORT_SYMBOL(I_BDEV);
45
46static sector_t max_block(struct block_device *bdev)
47{
48 sector_t retval = ~((sector_t)0);
49 loff_t sz = i_size_read(bdev->bd_inode);
50
51 if (sz) {
52 unsigned int size = block_size(bdev);
53 unsigned int sizebits = blksize_bits(size);
54 retval = (sz >> sizebits);
55 }
56 return retval;
57}
58
Peter Zijlstraf9a14392007-05-06 14:49:55 -070059/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070060static void kill_bdev(struct block_device *bdev)
61{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070062 if (bdev->bd_inode->i_mapping->nrpages == 0)
63 return;
64 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
66}
67
68int set_blocksize(struct block_device *bdev, int size)
69{
70 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070071 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 return -EINVAL;
73
74 /* Size cannot be smaller than the size supported by the device */
75 if (size < bdev_hardsect_size(bdev))
76 return -EINVAL;
77
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);
84 }
85 return 0;
86}
87
88EXPORT_SYMBOL(set_blocksize);
89
90int sb_set_blocksize(struct super_block *sb, int size)
91{
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 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;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -080097 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 return sb->s_blocksize;
99}
100
101EXPORT_SYMBOL(sb_set_blocksize);
102
103int sb_min_blocksize(struct super_block *sb, int size)
104{
105 int minsize = bdev_hardsect_size(sb->s_bdev);
106 if (size < minsize)
107 size = minsize;
108 return sb_set_blocksize(sb, size);
109}
110
111EXPORT_SYMBOL(sb_min_blocksize);
112
113static int
114blkdev_get_block(struct inode *inode, sector_t iblock,
115 struct buffer_head *bh, int create)
116{
117 if (iblock >= max_block(I_BDEV(inode))) {
118 if (create)
119 return -EIO;
120
121 /*
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
126 */
127 return 0;
128 }
129 bh->b_bdev = I_BDEV(inode);
130 bh->b_blocknr = iblock;
131 set_buffer_mapped(bh);
132 return 0;
133}
134
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800135static int
136blkdev_get_blocks(struct inode *inode, sector_t iblock,
137 struct buffer_head *bh, int create)
138{
139 sector_t end_block = max_block(I_BDEV(inode));
140 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
141
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 */
147 /*
148 * It is a read which is fully beyond EOF. We return
149 * a !buffer_mapped buffer
150 */
151 max_blocks = 0;
152 }
153 }
154
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;
161}
162
163static ssize_t
164blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
165 loff_t offset, unsigned long nr_segs)
166{
167 struct file *file = iocb->ki_filp;
168 struct inode *inode = file->f_mapping->host;
169
170 return blockdev_direct_IO_no_locking(rw, iocb, inode, I_BDEV(inode),
171 iov, offset, nr_segs, blkdev_get_blocks, NULL);
172}
173
174#if 0
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800175static int blk_end_aio(struct bio *bio, unsigned int bytes_done, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176{
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800177 struct kiocb *iocb = bio->bi_private;
178 atomic_t *bio_count = &iocb->ki_bio_count;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800180 if (bio_data_dir(bio) == READ)
181 bio_check_pages_dirty(bio);
182 else {
183 bio_release_pages(bio);
184 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
186
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800187 /* iocb->ki_nbytes stores error code from LLDD */
188 if (error)
189 iocb->ki_nbytes = -EIO;
190
191 if (atomic_dec_and_test(bio_count)) {
Andrew Morton790816d2007-01-22 20:40:35 -0800192 if ((long)iocb->ki_nbytes < 0)
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800193 aio_complete(iocb, iocb->ki_nbytes, 0);
194 else
195 aio_complete(iocb, iocb->ki_left, 0);
196 }
197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 return 0;
199}
200
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800201#define VEC_SIZE 16
202struct pvec {
203 unsigned short nr;
204 unsigned short idx;
205 struct page *page[VEC_SIZE];
206};
207
208#define PAGES_SPANNED(addr, len) \
209 (DIV_ROUND_UP((addr) + (len), PAGE_SIZE) - (addr) / PAGE_SIZE);
210
211/*
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.
216 */
217static struct page *blk_get_page(unsigned long addr, size_t count, int rw,
218 struct pvec *pvec)
219{
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;
232 }
233 return pvec->page[pvec->idx++];
234}
235
Chen, Kenneth Wcda92052007-01-22 20:40:43 -0800236/* return a page back to pvec array */
237static void blk_unget_page(struct page *page, struct pvec *pvec)
238{
239 pvec->page[--pvec->idx] = page;
240}
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242static ssize_t
243blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800244 loff_t pos, unsigned long nr_segs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800246 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800254 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;
262
263 pvec.nr = 0;
264 pvec.idx = 0;
265
266 if (pos & blocksize_mask)
267 return -EINVAL;
268
269 size = i_size_read(inode);
270 if (pos + nbytes > size) {
271 nbytes = size - pos;
272 iocb->ki_left = nbytes;
273 }
274
275 /*
276 * check first non-zero iov alignment, the remaining
277 * iov alignment is checked inside bio loop below.
278 */
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);
286
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);
292
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;
299same_bio:
300 cur_off = addr & ~PAGE_MASK;
301 cur_len = PAGE_SIZE - cur_off;
302 if (count < cur_len)
303 cur_len = count;
304
305 page = blk_get_page(addr, count, rw, &pvec);
306 if (unlikely(IS_ERR(page)))
307 goto backout;
308
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;
314
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;
326 }
327 count = min(count, nbytes);
328 goto same_bio;
329 }
Chen, Kenneth Wcda92052007-01-22 20:40:43 -0800330 } else {
331 blk_unget_page(page, &pvec);
Chen, Kenneth We61c9012006-12-13 00:34:36 -0800332 }
333
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);
339 }
340
341completion:
342 iocb->ki_left -= nbytes;
343 nbytes = iocb->ki_left;
344 iocb->ki_pos += nbytes;
345
346 blk_run_address_space(inode->i_mapping);
347 if (atomic_dec_and_test(bio_count))
348 aio_complete(iocb, nbytes, 0);
349
350 return -EIOCBQUEUED;
351
352backout:
353 /*
354 * back out nbytes count constructed so far for this bio,
355 * we will throw away current bio.
356 */
357 nbytes += bio->bi_size;
358 bio_release_pages(bio);
359 bio_put(bio);
360
361 /*
362 * if no bio was submmitted, return the error code.
363 * otherwise, proceed with pending I/O completion.
364 */
365 if (atomic_read(bio_count) == 1)
366 return PTR_ERR(page);
367 goto completion;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368}
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800369#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
371static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
372{
373 return block_write_full_page(page, blkdev_get_block, wbc);
374}
375
376static int blkdev_readpage(struct file * file, struct page * page)
377{
378 return block_read_full_page(page, blkdev_get_block);
379}
380
381static int blkdev_prepare_write(struct file *file, struct page *page, unsigned from, unsigned to)
382{
383 return block_prepare_write(page, from, to, blkdev_get_block);
384}
385
386static int blkdev_commit_write(struct file *file, struct page *page, unsigned from, unsigned to)
387{
388 return block_commit_write(page, from, to);
389}
390
391/*
392 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800393 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 * so we compute the size by hand (just as in block_read/write above)
395 */
396static loff_t block_llseek(struct file *file, loff_t offset, int origin)
397{
398 struct inode *bd_inode = file->f_mapping->host;
399 loff_t size;
400 loff_t retval;
401
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800402 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 size = i_size_read(bd_inode);
404
405 switch (origin) {
406 case 2:
407 offset += size;
408 break;
409 case 1:
410 offset += file->f_pos;
411 }
412 retval = -EINVAL;
413 if (offset >= 0 && offset <= size) {
414 if (offset != file->f_pos) {
415 file->f_pos = offset;
416 }
417 retval = offset;
418 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800419 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 return retval;
421}
422
423/*
424 * Filp is never NULL; the only case when ->fsync() is called with
425 * NULL first argument is nfsd_sync_dir() and that's not a directory.
426 */
427
428static int block_fsync(struct file *filp, struct dentry *dentry, int datasync)
429{
430 return sync_blockdev(I_BDEV(filp->f_mapping->host));
431}
432
433/*
434 * pseudo-fs
435 */
436
437static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800438static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
440static struct inode *bdev_alloc_inode(struct super_block *sb)
441{
Christoph Lametere94b1762006-12-06 20:33:17 -0800442 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 if (!ei)
444 return NULL;
445 return &ei->vfs_inode;
446}
447
448static void bdev_destroy_inode(struct inode *inode)
449{
450 struct bdev_inode *bdi = BDEV_I(inode);
451
452 bdi->bdev.bd_inode_backing_dev_info = NULL;
453 kmem_cache_free(bdev_cachep, bdi);
454}
455
Christoph Lametere18b8902006-12-06 20:33:20 -0800456static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
458 struct bdev_inode *ei = (struct bdev_inode *) foo;
459 struct block_device *bdev = &ei->bdev;
460
Christoph Lametera35afb82007-05-16 22:10:57 -0700461 memset(bdev, 0, sizeof(*bdev));
462 mutex_init(&bdev->bd_mutex);
463 sema_init(&bdev->bd_mount_sem, 1);
464 INIT_LIST_HEAD(&bdev->bd_inodes);
465 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800466#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700467 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800468#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700469 inode_init_once(&ei->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472static inline void __bd_forget(struct inode *inode)
473{
474 list_del_init(&inode->i_devices);
475 inode->i_bdev = NULL;
476 inode->i_mapping = &inode->i_data;
477}
478
479static void bdev_clear_inode(struct inode *inode)
480{
481 struct block_device *bdev = &BDEV_I(inode)->bdev;
482 struct list_head *p;
483 spin_lock(&bdev_lock);
484 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
485 __bd_forget(list_entry(p, struct inode, i_devices));
486 }
487 list_del_init(&bdev->bd_list);
488 spin_unlock(&bdev_lock);
489}
490
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800491static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 .statfs = simple_statfs,
493 .alloc_inode = bdev_alloc_inode,
494 .destroy_inode = bdev_destroy_inode,
495 .drop_inode = generic_delete_inode,
496 .clear_inode = bdev_clear_inode,
497};
498
David Howells454e2392006-06-23 02:02:57 -0700499static int bd_get_sb(struct file_system_type *fs_type,
500 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
David Howells454e2392006-06-23 02:02:57 -0700502 return get_sb_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503}
504
505static struct file_system_type bd_type = {
506 .name = "bdev",
507 .get_sb = bd_get_sb,
508 .kill_sb = kill_anon_super,
509};
510
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800511static struct vfsmount *bd_mnt __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512struct super_block *blockdev_superblock;
513
514void __init bdev_cache_init(void)
515{
516 int err;
517 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800518 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
519 SLAB_MEM_SPREAD|SLAB_PANIC),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 init_once, NULL);
521 err = register_filesystem(&bd_type);
522 if (err)
523 panic("Cannot register bdev pseudo-fs");
524 bd_mnt = kern_mount(&bd_type);
525 err = PTR_ERR(bd_mnt);
526 if (IS_ERR(bd_mnt))
527 panic("Cannot create bdev pseudo-fs");
528 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
529}
530
531/*
532 * Most likely _very_ bad one - but then it's hardly critical for small
533 * /dev and can be fixed when somebody will need really large one.
534 * Keep in mind that it will be fed through icache hash function too.
535 */
536static inline unsigned long hash(dev_t dev)
537{
538 return MAJOR(dev)+MINOR(dev);
539}
540
541static int bdev_test(struct inode *inode, void *data)
542{
543 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
544}
545
546static int bdev_set(struct inode *inode, void *data)
547{
548 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
549 return 0;
550}
551
552static LIST_HEAD(all_bdevs);
553
554struct block_device *bdget(dev_t dev)
555{
556 struct block_device *bdev;
557 struct inode *inode;
558
559 inode = iget5_locked(bd_mnt->mnt_sb, hash(dev),
560 bdev_test, bdev_set, &dev);
561
562 if (!inode)
563 return NULL;
564
565 bdev = &BDEV_I(inode)->bdev;
566
567 if (inode->i_state & I_NEW) {
568 bdev->bd_contains = NULL;
569 bdev->bd_inode = inode;
570 bdev->bd_block_size = (1 << inode->i_blkbits);
571 bdev->bd_part_count = 0;
572 bdev->bd_invalidated = 0;
573 inode->i_mode = S_IFBLK;
574 inode->i_rdev = dev;
575 inode->i_bdev = bdev;
576 inode->i_data.a_ops = &def_blk_aops;
577 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
578 inode->i_data.backing_dev_info = &default_backing_dev_info;
579 spin_lock(&bdev_lock);
580 list_add(&bdev->bd_list, &all_bdevs);
581 spin_unlock(&bdev_lock);
582 unlock_new_inode(inode);
583 }
584 return bdev;
585}
586
587EXPORT_SYMBOL(bdget);
588
589long nr_blockdev_pages(void)
590{
591 struct list_head *p;
592 long ret = 0;
593 spin_lock(&bdev_lock);
594 list_for_each(p, &all_bdevs) {
595 struct block_device *bdev;
596 bdev = list_entry(p, struct block_device, bd_list);
597 ret += bdev->bd_inode->i_mapping->nrpages;
598 }
599 spin_unlock(&bdev_lock);
600 return ret;
601}
602
603void bdput(struct block_device *bdev)
604{
605 iput(bdev->bd_inode);
606}
607
608EXPORT_SYMBOL(bdput);
609
610static struct block_device *bd_acquire(struct inode *inode)
611{
612 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 spin_lock(&bdev_lock);
615 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700616 if (bdev) {
617 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 spin_unlock(&bdev_lock);
619 return bdev;
620 }
621 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700622
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 bdev = bdget(inode->i_rdev);
624 if (bdev) {
625 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700626 if (!inode->i_bdev) {
627 /*
628 * We take an additional bd_inode->i_count for inode,
629 * and it's released in clear_inode() of inode.
630 * So, we can access it via ->i_mapping always
631 * without igrab().
632 */
633 atomic_inc(&bdev->bd_inode->i_count);
634 inode->i_bdev = bdev;
635 inode->i_mapping = bdev->bd_inode->i_mapping;
636 list_add(&inode->i_devices, &bdev->bd_inodes);
637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 spin_unlock(&bdev_lock);
639 }
640 return bdev;
641}
642
643/* Call when you free inode */
644
645void bd_forget(struct inode *inode)
646{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700647 struct block_device *bdev = NULL;
648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700650 if (inode->i_bdev) {
651 if (inode->i_sb != blockdev_superblock)
652 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700656
657 if (bdev)
658 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
661int bd_claim(struct block_device *bdev, void *holder)
662{
663 int res;
664 spin_lock(&bdev_lock);
665
666 /* first decide result */
667 if (bdev->bd_holder == holder)
668 res = 0; /* already a holder */
669 else if (bdev->bd_holder != NULL)
670 res = -EBUSY; /* held by someone else */
671 else if (bdev->bd_contains == bdev)
672 res = 0; /* is a whole device which isn't held */
673
674 else if (bdev->bd_contains->bd_holder == bd_claim)
675 res = 0; /* is a partition of a device that is being partitioned */
676 else if (bdev->bd_contains->bd_holder != NULL)
677 res = -EBUSY; /* is a partition of a held device */
678 else
679 res = 0; /* is a partition of an un-held device */
680
681 /* now impose change */
682 if (res==0) {
683 /* note that for a whole device bd_holders
684 * will be incremented twice, and bd_holder will
685 * be set to bd_claim before being set to holder
686 */
687 bdev->bd_contains->bd_holders ++;
688 bdev->bd_contains->bd_holder = bd_claim;
689 bdev->bd_holders++;
690 bdev->bd_holder = holder;
691 }
692 spin_unlock(&bdev_lock);
693 return res;
694}
695
696EXPORT_SYMBOL(bd_claim);
697
698void bd_release(struct block_device *bdev)
699{
700 spin_lock(&bdev_lock);
701 if (!--bdev->bd_contains->bd_holders)
702 bdev->bd_contains->bd_holder = NULL;
703 if (!--bdev->bd_holders)
704 bdev->bd_holder = NULL;
705 spin_unlock(&bdev_lock);
706}
707
708EXPORT_SYMBOL(bd_release);
709
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800710#ifdef CONFIG_SYSFS
711/*
712 * Functions for bd_claim_by_kobject / bd_release_from_kobject
713 *
714 * If a kobject is passed to bd_claim_by_kobject()
715 * and the kobject has a parent directory,
716 * following symlinks are created:
717 * o from the kobject to the claimed bdev
718 * o from "holders" directory of the bdev to the parent of the kobject
719 * bd_release_from_kobject() removes these symlinks.
720 *
721 * Example:
722 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
723 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
724 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
725 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
726 */
727
728static struct kobject *bdev_get_kobj(struct block_device *bdev)
729{
730 if (bdev->bd_contains != bdev)
731 return kobject_get(&bdev->bd_part->kobj);
732 else
733 return kobject_get(&bdev->bd_disk->kobj);
734}
735
736static struct kobject *bdev_get_holder(struct block_device *bdev)
737{
738 if (bdev->bd_contains != bdev)
739 return kobject_get(bdev->bd_part->holder_dir);
740 else
741 return kobject_get(bdev->bd_disk->holder_dir);
742}
743
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700744static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800745{
746 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700747 return 0;
748 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800749}
750
751static void del_symlink(struct kobject *from, struct kobject *to)
752{
753 if (!from || !to)
754 return;
755 sysfs_remove_link(from, kobject_name(to));
756}
757
758/*
759 * 'struct bd_holder' contains pointers to kobjects symlinked by
760 * bd_claim_by_kobject.
761 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
762 */
763struct bd_holder {
764 struct list_head list; /* chain of holders of the bdev */
765 int count; /* references from the holder */
766 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
767 struct kobject *hdev; /* e.g. "/block/dm-0" */
768 struct kobject *hdir; /* e.g. "/block/sda/holders" */
769 struct kobject *sdev; /* e.g. "/block/sda" */
770};
771
772/*
773 * Get references of related kobjects at once.
774 * Returns 1 on success. 0 on failure.
775 *
776 * Should call bd_holder_release_dirs() after successful use.
777 */
778static int bd_holder_grab_dirs(struct block_device *bdev,
779 struct bd_holder *bo)
780{
781 if (!bdev || !bo)
782 return 0;
783
784 bo->sdir = kobject_get(bo->sdir);
785 if (!bo->sdir)
786 return 0;
787
788 bo->hdev = kobject_get(bo->sdir->parent);
789 if (!bo->hdev)
790 goto fail_put_sdir;
791
792 bo->sdev = bdev_get_kobj(bdev);
793 if (!bo->sdev)
794 goto fail_put_hdev;
795
796 bo->hdir = bdev_get_holder(bdev);
797 if (!bo->hdir)
798 goto fail_put_sdev;
799
800 return 1;
801
802fail_put_sdev:
803 kobject_put(bo->sdev);
804fail_put_hdev:
805 kobject_put(bo->hdev);
806fail_put_sdir:
807 kobject_put(bo->sdir);
808
809 return 0;
810}
811
812/* Put references of related kobjects at once. */
813static void bd_holder_release_dirs(struct bd_holder *bo)
814{
815 kobject_put(bo->hdir);
816 kobject_put(bo->sdev);
817 kobject_put(bo->hdev);
818 kobject_put(bo->sdir);
819}
820
821static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
822{
823 struct bd_holder *bo;
824
825 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
826 if (!bo)
827 return NULL;
828
829 bo->count = 1;
830 bo->sdir = kobj;
831
832 return bo;
833}
834
835static void free_bd_holder(struct bd_holder *bo)
836{
837 kfree(bo);
838}
839
840/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500841 * find_bd_holder - find matching struct bd_holder from the block device
842 *
843 * @bdev: struct block device to be searched
844 * @bo: target struct bd_holder
845 *
846 * Returns matching entry with @bo in @bdev->bd_holder_list.
847 * If found, increment the reference count and return the pointer.
848 * If not found, returns NULL.
849 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800850static struct bd_holder *find_bd_holder(struct block_device *bdev,
851 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500852{
853 struct bd_holder *tmp;
854
855 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
856 if (tmp->sdir == bo->sdir) {
857 tmp->count++;
858 return tmp;
859 }
860
861 return NULL;
862}
863
864/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800865 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
866 *
867 * @bdev: block device to be bd_claimed
868 * @bo: preallocated and initialized by alloc_bd_holder()
869 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500870 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800871 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500872 * Returns 0 if symlinks are created.
873 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800874 */
875static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
876{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700877 int ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800878
879 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700880 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800881
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800882 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700883 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800884
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700885 ret = add_symlink(bo->sdir, bo->sdev);
886 if (ret == 0) {
887 ret = add_symlink(bo->hdir, bo->hdev);
888 if (ret)
889 del_symlink(bo->sdir, bo->sdev);
890 }
891 if (ret == 0)
892 list_add_tail(&bo->list, &bdev->bd_holder_list);
893 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800894}
895
896/**
897 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
898 *
899 * @bdev: block device to be bd_claimed
900 * @kobj: holder's kobject
901 *
902 * If there is matching entry with @kobj in @bdev->bd_holder_list
903 * and no other bd_claim() from the same kobject,
904 * remove the struct bd_holder from the list, delete symlinks for it.
905 *
906 * Returns a pointer to the struct bd_holder when it's removed from the list
907 * and ready to be freed.
908 * Returns NULL if matching claim isn't found or there is other bd_claim()
909 * by the same kobject.
910 */
911static struct bd_holder *del_bd_holder(struct block_device *bdev,
912 struct kobject *kobj)
913{
914 struct bd_holder *bo;
915
916 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
917 if (bo->sdir == kobj) {
918 bo->count--;
919 BUG_ON(bo->count < 0);
920 if (!bo->count) {
921 list_del(&bo->list);
922 del_symlink(bo->sdir, bo->sdev);
923 del_symlink(bo->hdir, bo->hdev);
924 bd_holder_release_dirs(bo);
925 return bo;
926 }
927 break;
928 }
929 }
930
931 return NULL;
932}
933
934/**
935 * bd_claim_by_kobject - bd_claim() with additional kobject signature
936 *
937 * @bdev: block device to be claimed
938 * @holder: holder's signature
939 * @kobj: holder's kobject
940 *
941 * Do bd_claim() and if it succeeds, create sysfs symlinks between
942 * the bdev and the holder's kobject.
943 * Use bd_release_from_kobject() when relesing the claimed bdev.
944 *
945 * Returns 0 on success. (same as bd_claim())
946 * Returns errno on failure.
947 */
948static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
949 struct kobject *kobj)
950{
951 int res;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500952 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800953
954 if (!kobj)
955 return -EINVAL;
956
957 bo = alloc_bd_holder(kobj);
958 if (!bo)
959 return -ENOMEM;
960
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800961 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800962 res = bd_claim(bdev, holder);
Jun'ichi Nomurabcb55162006-10-30 16:23:45 -0500963 if (res == 0) {
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500964 found = find_bd_holder(bdev, bo);
965 if (found == NULL) {
966 res = add_bd_holder(bdev, bo);
967 if (res)
968 bd_release(bdev);
969 }
Jun'ichi Nomurabcb55162006-10-30 16:23:45 -0500970 }
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500971
972 if (res || found)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800973 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800974 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800975
976 return res;
977}
978
979/**
980 * bd_release_from_kobject - bd_release() with additional kobject signature
981 *
982 * @bdev: block device to be released
983 * @kobj: holder's kobject
984 *
985 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
986 */
987static void bd_release_from_kobject(struct block_device *bdev,
988 struct kobject *kobj)
989{
990 struct bd_holder *bo;
991
992 if (!kobj)
993 return;
994
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800995 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800996 bd_release(bdev);
997 if ((bo = del_bd_holder(bdev, kobj)))
998 free_bd_holder(bo);
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800999 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001000}
1001
1002/**
1003 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1004 *
1005 * @bdev: block device to be claimed
1006 * @holder: holder's signature
1007 * @disk: holder's gendisk
1008 *
1009 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1010 */
1011int bd_claim_by_disk(struct block_device *bdev, void *holder,
1012 struct gendisk *disk)
1013{
1014 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1015}
1016EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1017
1018/**
1019 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1020 *
1021 * @bdev: block device to be claimed
1022 * @disk: holder's gendisk
1023 *
1024 * Call bd_release_from_kobject() and put @disk->slave_dir.
1025 */
1026void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1027{
1028 bd_release_from_kobject(bdev, disk->slave_dir);
1029 kobject_put(disk->slave_dir);
1030}
1031EXPORT_SYMBOL_GPL(bd_release_from_disk);
1032#endif
1033
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034/*
1035 * Tries to open block device by device number. Use it ONLY if you
1036 * really do not have anything better - i.e. when you are behind a
1037 * truly sucky interface and all you are given is a device number. _Never_
1038 * to be used for internal purposes. If you ever need it - reconsider
1039 * your API.
1040 */
1041struct block_device *open_by_devnum(dev_t dev, unsigned mode)
1042{
1043 struct block_device *bdev = bdget(dev);
1044 int err = -ENOMEM;
1045 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
1046 if (bdev)
1047 err = blkdev_get(bdev, mode, flags);
1048 return err ? ERR_PTR(err) : bdev;
1049}
1050
1051EXPORT_SYMBOL(open_by_devnum);
1052
1053/*
1054 * This routine checks whether a removable media has been changed,
1055 * and invalidates all buffer-cache-entries in that case. This
1056 * is a relatively slow routine, so we have to try to minimize using
1057 * it. Thus it is called only upon a 'mount' or 'open'. This
1058 * is the best way of combining speed and utility, I think.
1059 * People changing diskettes in the middle of an operation deserve
1060 * to lose :-)
1061 */
1062int check_disk_change(struct block_device *bdev)
1063{
1064 struct gendisk *disk = bdev->bd_disk;
1065 struct block_device_operations * bdops = disk->fops;
1066
1067 if (!bdops->media_changed)
1068 return 0;
1069 if (!bdops->media_changed(bdev->bd_disk))
1070 return 0;
1071
Christoph Hellwig2ef41632005-05-05 16:15:59 -07001072 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 printk("VFS: busy inodes on changed media.\n");
1074
1075 if (bdops->revalidate_disk)
1076 bdops->revalidate_disk(bdev->bd_disk);
1077 if (bdev->bd_disk->minors > 1)
1078 bdev->bd_invalidated = 1;
1079 return 1;
1080}
1081
1082EXPORT_SYMBOL(check_disk_change);
1083
1084void bd_set_size(struct block_device *bdev, loff_t size)
1085{
1086 unsigned bsize = bdev_hardsect_size(bdev);
1087
1088 bdev->bd_inode->i_size = size;
1089 while (bsize < PAGE_CACHE_SIZE) {
1090 if (size & bsize)
1091 break;
1092 bsize <<= 1;
1093 }
1094 bdev->bd_block_size = bsize;
1095 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1096}
1097EXPORT_SYMBOL(bd_set_size);
1098
NeilBrown37be4122006-12-08 02:36:16 -08001099static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1100 int for_part);
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001101static int __blkdev_put(struct block_device *bdev, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001102
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001103/*
1104 * bd_mutex locking:
1105 *
1106 * mutex_lock(part->bd_mutex)
1107 * mutex_lock_nested(whole->bd_mutex, 1)
1108 */
1109
NeilBrown37be4122006-12-08 02:36:16 -08001110static int do_open(struct block_device *bdev, struct file *file, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111{
1112 struct module *owner = NULL;
1113 struct gendisk *disk;
1114 int ret = -ENXIO;
1115 int part;
1116
1117 file->f_mapping = bdev->bd_inode->i_mapping;
1118 lock_kernel();
1119 disk = get_gendisk(bdev->bd_dev, &part);
1120 if (!disk) {
1121 unlock_kernel();
1122 bdput(bdev);
1123 return ret;
1124 }
1125 owner = disk->fops->owner;
1126
NeilBrown6796bf52006-12-08 02:36:16 -08001127 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (!bdev->bd_openers) {
1129 bdev->bd_disk = disk;
1130 bdev->bd_contains = bdev;
1131 if (!part) {
1132 struct backing_dev_info *bdi;
1133 if (disk->fops->open) {
1134 ret = disk->fops->open(bdev->bd_inode, file);
1135 if (ret)
1136 goto out_first;
1137 }
1138 if (!bdev->bd_openers) {
1139 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1140 bdi = blk_get_backing_dev_info(bdev);
1141 if (bdi == NULL)
1142 bdi = &default_backing_dev_info;
1143 bdev->bd_inode->i_data.backing_dev_info = bdi;
1144 }
1145 if (bdev->bd_invalidated)
1146 rescan_partitions(disk, bdev);
1147 } else {
1148 struct hd_struct *p;
1149 struct block_device *whole;
1150 whole = bdget_disk(disk, 0);
1151 ret = -ENOMEM;
1152 if (!whole)
1153 goto out_first;
NeilBrown37be4122006-12-08 02:36:16 -08001154 BUG_ON(for_part);
1155 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (ret)
1157 goto out_first;
1158 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 p = disk->part[part - 1];
1160 bdev->bd_inode->i_data.backing_dev_info =
1161 whole->bd_inode->i_data.backing_dev_info;
1162 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 ret = -ENXIO;
1164 goto out_first;
1165 }
1166 kobject_get(&p->kobj);
1167 bdev->bd_part = p;
1168 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
1170 } else {
1171 put_disk(disk);
1172 module_put(owner);
1173 if (bdev->bd_contains == bdev) {
1174 if (bdev->bd_disk->fops->open) {
1175 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1176 if (ret)
1177 goto out;
1178 }
1179 if (bdev->bd_invalidated)
1180 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 }
1182 }
1183 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001184 if (for_part)
1185 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001186 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 unlock_kernel();
1188 return 0;
1189
1190out_first:
1191 bdev->bd_disk = NULL;
1192 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1193 if (bdev != bdev->bd_contains)
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001194 __blkdev_put(bdev->bd_contains, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 bdev->bd_contains = NULL;
1196 put_disk(disk);
1197 module_put(owner);
1198out:
Arjan van de Venc039e312006-03-23 03:00:28 -08001199 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 unlock_kernel();
1201 if (ret)
1202 bdput(bdev);
1203 return ret;
1204}
1205
NeilBrown37be4122006-12-08 02:36:16 -08001206static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1207 int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
1209 /*
1210 * This crockload is due to bad choice of ->open() type.
1211 * It will go away.
1212 * For now, block device ->open() routine must _not_
1213 * examine anything in 'inode' argument except ->i_rdev.
1214 */
1215 struct file fake_file = {};
1216 struct dentry fake_dentry = {};
1217 fake_file.f_mode = mode;
1218 fake_file.f_flags = flags;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001219 fake_file.f_path.dentry = &fake_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 fake_dentry.d_inode = bdev->bd_inode;
1221
NeilBrown37be4122006-12-08 02:36:16 -08001222 return do_open(bdev, &fake_file, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223}
1224
NeilBrown37be4122006-12-08 02:36:16 -08001225int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1226{
1227 return __blkdev_get(bdev, mode, flags, 0);
1228}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229EXPORT_SYMBOL(blkdev_get);
1230
1231static int blkdev_open(struct inode * inode, struct file * filp)
1232{
1233 struct block_device *bdev;
1234 int res;
1235
1236 /*
1237 * Preserve backwards compatibility and allow large file access
1238 * even if userspace doesn't ask for it explicitly. Some mkfs
1239 * binary needs it. We might want to drop this workaround
1240 * during an unstable branch.
1241 */
1242 filp->f_flags |= O_LARGEFILE;
1243
1244 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001245 if (bdev == NULL)
1246 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
NeilBrown37be4122006-12-08 02:36:16 -08001248 res = do_open(bdev, filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (res)
1250 return res;
1251
1252 if (!(filp->f_flags & O_EXCL) )
1253 return 0;
1254
1255 if (!(res = bd_claim(bdev, filp)))
1256 return 0;
1257
1258 blkdev_put(bdev);
1259 return res;
1260}
1261
NeilBrown37be4122006-12-08 02:36:16 -08001262static int __blkdev_put(struct block_device *bdev, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001263{
1264 int ret = 0;
1265 struct inode *bd_inode = bdev->bd_inode;
1266 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001267 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001268
NeilBrown6796bf52006-12-08 02:36:16 -08001269 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001270 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001271 if (for_part)
1272 bdev->bd_part_count--;
1273
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001274 if (!--bdev->bd_openers) {
1275 sync_blockdev(bdev);
1276 kill_bdev(bdev);
1277 }
1278 if (bdev->bd_contains == bdev) {
1279 if (disk->fops->release)
1280 ret = disk->fops->release(bd_inode, NULL);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001281 }
1282 if (!bdev->bd_openers) {
1283 struct module *owner = disk->fops->owner;
1284
1285 put_disk(disk);
1286 module_put(owner);
1287
1288 if (bdev->bd_contains != bdev) {
1289 kobject_put(&bdev->bd_part->kobj);
1290 bdev->bd_part = NULL;
1291 }
1292 bdev->bd_disk = NULL;
1293 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001294 if (bdev != bdev->bd_contains)
1295 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001296 bdev->bd_contains = NULL;
1297 }
1298 unlock_kernel();
1299 mutex_unlock(&bdev->bd_mutex);
1300 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001301 if (victim)
1302 __blkdev_put(victim, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001303 return ret;
1304}
1305
NeilBrown37be4122006-12-08 02:36:16 -08001306int blkdev_put(struct block_device *bdev)
1307{
1308 return __blkdev_put(bdev, 0);
1309}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001310EXPORT_SYMBOL(blkdev_put);
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312static int blkdev_close(struct inode * inode, struct file * filp)
1313{
1314 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1315 if (bdev->bd_holder == filp)
1316 bd_release(bdev);
1317 return blkdev_put(bdev);
1318}
1319
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001320static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
1322 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1323}
1324
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001325const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 .readpage = blkdev_readpage,
1327 .writepage = blkdev_writepage,
1328 .sync_page = block_sync_page,
1329 .prepare_write = blkdev_prepare_write,
1330 .commit_write = blkdev_commit_write,
1331 .writepages = generic_writepages,
1332 .direct_IO = blkdev_direct_IO,
1333};
1334
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001335const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 .open = blkdev_open,
1337 .release = blkdev_close,
1338 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001339 .read = do_sync_read,
1340 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001342 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 .mmap = generic_file_mmap,
1344 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001345 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346#ifdef CONFIG_COMPAT
1347 .compat_ioctl = compat_blkdev_ioctl,
1348#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 .sendfile = generic_file_sendfile,
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001350 .splice_read = generic_file_splice_read,
1351 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352};
1353
1354int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1355{
1356 int res;
1357 mm_segment_t old_fs = get_fs();
1358 set_fs(KERNEL_DS);
1359 res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
1360 set_fs(old_fs);
1361 return res;
1362}
1363
1364EXPORT_SYMBOL(ioctl_by_bdev);
1365
1366/**
1367 * lookup_bdev - lookup a struct block_device by name
1368 *
1369 * @path: special file representing the block device
1370 *
1371 * Get a reference to the blockdevice at @path in the current
1372 * namespace if possible and return it. Return ERR_PTR(error)
1373 * otherwise.
1374 */
1375struct block_device *lookup_bdev(const char *path)
1376{
1377 struct block_device *bdev;
1378 struct inode *inode;
1379 struct nameidata nd;
1380 int error;
1381
1382 if (!path || !*path)
1383 return ERR_PTR(-EINVAL);
1384
1385 error = path_lookup(path, LOOKUP_FOLLOW, &nd);
1386 if (error)
1387 return ERR_PTR(error);
1388
1389 inode = nd.dentry->d_inode;
1390 error = -ENOTBLK;
1391 if (!S_ISBLK(inode->i_mode))
1392 goto fail;
1393 error = -EACCES;
1394 if (nd.mnt->mnt_flags & MNT_NODEV)
1395 goto fail;
1396 error = -ENOMEM;
1397 bdev = bd_acquire(inode);
1398 if (!bdev)
1399 goto fail;
1400out:
1401 path_release(&nd);
1402 return bdev;
1403fail:
1404 bdev = ERR_PTR(error);
1405 goto out;
1406}
1407
1408/**
1409 * open_bdev_excl - open a block device by name and set it up for use
1410 *
1411 * @path: special file representing the block device
1412 * @flags: %MS_RDONLY for opening read-only
1413 * @holder: owner for exclusion
1414 *
1415 * Open the blockdevice described by the special file at @path, claim it
1416 * for the @holder.
1417 */
1418struct block_device *open_bdev_excl(const char *path, int flags, void *holder)
1419{
1420 struct block_device *bdev;
1421 mode_t mode = FMODE_READ;
1422 int error = 0;
1423
1424 bdev = lookup_bdev(path);
1425 if (IS_ERR(bdev))
1426 return bdev;
1427
1428 if (!(flags & MS_RDONLY))
1429 mode |= FMODE_WRITE;
1430 error = blkdev_get(bdev, mode, 0);
1431 if (error)
1432 return ERR_PTR(error);
1433 error = -EACCES;
1434 if (!(flags & MS_RDONLY) && bdev_read_only(bdev))
1435 goto blkdev_put;
1436 error = bd_claim(bdev, holder);
1437 if (error)
1438 goto blkdev_put;
1439
1440 return bdev;
1441
1442blkdev_put:
1443 blkdev_put(bdev);
1444 return ERR_PTR(error);
1445}
1446
1447EXPORT_SYMBOL(open_bdev_excl);
1448
1449/**
1450 * close_bdev_excl - release a blockdevice openen by open_bdev_excl()
1451 *
1452 * @bdev: blockdevice to close
1453 *
1454 * This is the counterpart to open_bdev_excl().
1455 */
1456void close_bdev_excl(struct block_device *bdev)
1457{
1458 bd_release(bdev);
1459 blkdev_put(bdev);
1460}
1461
1462EXPORT_SYMBOL(close_bdev_excl);
David Howellsb71e8a42006-08-29 19:06:11 +01001463
1464int __invalidate_device(struct block_device *bdev)
1465{
1466 struct super_block *sb = get_super(bdev);
1467 int res = 0;
1468
1469 if (sb) {
1470 /*
1471 * no need to lock the super, get_super holds the
1472 * read mutex so the filesystem cannot go away
1473 * under us (->put_super runs with the write lock
1474 * hold).
1475 */
1476 shrink_dcache_sb(sb);
1477 res = invalidate_inodes(sb);
1478 drop_super(sb);
1479 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001480 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001481 return res;
1482}
1483EXPORT_SYMBOL(__invalidate_device);