blob: 2980eabe577981c0852a9286eea9be4474143687 [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),
Paul Mundt20c2df82007-07-20 10:11:58 +0900520 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 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{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700591 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 long ret = 0;
593 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700594 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 ret += bdev->bd_inode->i_mapping->nrpages;
596 }
597 spin_unlock(&bdev_lock);
598 return ret;
599}
600
601void bdput(struct block_device *bdev)
602{
603 iput(bdev->bd_inode);
604}
605
606EXPORT_SYMBOL(bdput);
607
608static struct block_device *bd_acquire(struct inode *inode)
609{
610 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 spin_lock(&bdev_lock);
613 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700614 if (bdev) {
615 atomic_inc(&bdev->bd_inode->i_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 spin_unlock(&bdev_lock);
617 return bdev;
618 }
619 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 bdev = bdget(inode->i_rdev);
622 if (bdev) {
623 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700624 if (!inode->i_bdev) {
625 /*
626 * We take an additional bd_inode->i_count for inode,
627 * and it's released in clear_inode() of inode.
628 * So, we can access it via ->i_mapping always
629 * without igrab().
630 */
631 atomic_inc(&bdev->bd_inode->i_count);
632 inode->i_bdev = bdev;
633 inode->i_mapping = bdev->bd_inode->i_mapping;
634 list_add(&inode->i_devices, &bdev->bd_inodes);
635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 spin_unlock(&bdev_lock);
637 }
638 return bdev;
639}
640
641/* Call when you free inode */
642
643void bd_forget(struct inode *inode)
644{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700645 struct block_device *bdev = NULL;
646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700648 if (inode->i_bdev) {
649 if (inode->i_sb != blockdev_superblock)
650 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700654
655 if (bdev)
656 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659int bd_claim(struct block_device *bdev, void *holder)
660{
661 int res;
662 spin_lock(&bdev_lock);
663
664 /* first decide result */
665 if (bdev->bd_holder == holder)
666 res = 0; /* already a holder */
667 else if (bdev->bd_holder != NULL)
668 res = -EBUSY; /* held by someone else */
669 else if (bdev->bd_contains == bdev)
670 res = 0; /* is a whole device which isn't held */
671
672 else if (bdev->bd_contains->bd_holder == bd_claim)
673 res = 0; /* is a partition of a device that is being partitioned */
674 else if (bdev->bd_contains->bd_holder != NULL)
675 res = -EBUSY; /* is a partition of a held device */
676 else
677 res = 0; /* is a partition of an un-held device */
678
679 /* now impose change */
680 if (res==0) {
681 /* note that for a whole device bd_holders
682 * will be incremented twice, and bd_holder will
683 * be set to bd_claim before being set to holder
684 */
685 bdev->bd_contains->bd_holders ++;
686 bdev->bd_contains->bd_holder = bd_claim;
687 bdev->bd_holders++;
688 bdev->bd_holder = holder;
689 }
690 spin_unlock(&bdev_lock);
691 return res;
692}
693
694EXPORT_SYMBOL(bd_claim);
695
696void bd_release(struct block_device *bdev)
697{
698 spin_lock(&bdev_lock);
699 if (!--bdev->bd_contains->bd_holders)
700 bdev->bd_contains->bd_holder = NULL;
701 if (!--bdev->bd_holders)
702 bdev->bd_holder = NULL;
703 spin_unlock(&bdev_lock);
704}
705
706EXPORT_SYMBOL(bd_release);
707
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800708#ifdef CONFIG_SYSFS
709/*
710 * Functions for bd_claim_by_kobject / bd_release_from_kobject
711 *
712 * If a kobject is passed to bd_claim_by_kobject()
713 * and the kobject has a parent directory,
714 * following symlinks are created:
715 * o from the kobject to the claimed bdev
716 * o from "holders" directory of the bdev to the parent of the kobject
717 * bd_release_from_kobject() removes these symlinks.
718 *
719 * Example:
720 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
721 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
722 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
723 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
724 */
725
726static struct kobject *bdev_get_kobj(struct block_device *bdev)
727{
728 if (bdev->bd_contains != bdev)
729 return kobject_get(&bdev->bd_part->kobj);
730 else
731 return kobject_get(&bdev->bd_disk->kobj);
732}
733
734static struct kobject *bdev_get_holder(struct block_device *bdev)
735{
736 if (bdev->bd_contains != bdev)
737 return kobject_get(bdev->bd_part->holder_dir);
738 else
739 return kobject_get(bdev->bd_disk->holder_dir);
740}
741
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700742static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800743{
744 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700745 return 0;
746 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800747}
748
749static void del_symlink(struct kobject *from, struct kobject *to)
750{
751 if (!from || !to)
752 return;
753 sysfs_remove_link(from, kobject_name(to));
754}
755
756/*
757 * 'struct bd_holder' contains pointers to kobjects symlinked by
758 * bd_claim_by_kobject.
759 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
760 */
761struct bd_holder {
762 struct list_head list; /* chain of holders of the bdev */
763 int count; /* references from the holder */
764 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
765 struct kobject *hdev; /* e.g. "/block/dm-0" */
766 struct kobject *hdir; /* e.g. "/block/sda/holders" */
767 struct kobject *sdev; /* e.g. "/block/sda" */
768};
769
770/*
771 * Get references of related kobjects at once.
772 * Returns 1 on success. 0 on failure.
773 *
774 * Should call bd_holder_release_dirs() after successful use.
775 */
776static int bd_holder_grab_dirs(struct block_device *bdev,
777 struct bd_holder *bo)
778{
779 if (!bdev || !bo)
780 return 0;
781
782 bo->sdir = kobject_get(bo->sdir);
783 if (!bo->sdir)
784 return 0;
785
786 bo->hdev = kobject_get(bo->sdir->parent);
787 if (!bo->hdev)
788 goto fail_put_sdir;
789
790 bo->sdev = bdev_get_kobj(bdev);
791 if (!bo->sdev)
792 goto fail_put_hdev;
793
794 bo->hdir = bdev_get_holder(bdev);
795 if (!bo->hdir)
796 goto fail_put_sdev;
797
798 return 1;
799
800fail_put_sdev:
801 kobject_put(bo->sdev);
802fail_put_hdev:
803 kobject_put(bo->hdev);
804fail_put_sdir:
805 kobject_put(bo->sdir);
806
807 return 0;
808}
809
810/* Put references of related kobjects at once. */
811static void bd_holder_release_dirs(struct bd_holder *bo)
812{
813 kobject_put(bo->hdir);
814 kobject_put(bo->sdev);
815 kobject_put(bo->hdev);
816 kobject_put(bo->sdir);
817}
818
819static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
820{
821 struct bd_holder *bo;
822
823 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
824 if (!bo)
825 return NULL;
826
827 bo->count = 1;
828 bo->sdir = kobj;
829
830 return bo;
831}
832
833static void free_bd_holder(struct bd_holder *bo)
834{
835 kfree(bo);
836}
837
838/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500839 * find_bd_holder - find matching struct bd_holder from the block device
840 *
841 * @bdev: struct block device to be searched
842 * @bo: target struct bd_holder
843 *
844 * Returns matching entry with @bo in @bdev->bd_holder_list.
845 * If found, increment the reference count and return the pointer.
846 * If not found, returns NULL.
847 */
Andrew Morton36a561d2006-10-30 22:07:03 -0800848static struct bd_holder *find_bd_holder(struct block_device *bdev,
849 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500850{
851 struct bd_holder *tmp;
852
853 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
854 if (tmp->sdir == bo->sdir) {
855 tmp->count++;
856 return tmp;
857 }
858
859 return NULL;
860}
861
862/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800863 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
864 *
865 * @bdev: block device to be bd_claimed
866 * @bo: preallocated and initialized by alloc_bd_holder()
867 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500868 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800869 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500870 * Returns 0 if symlinks are created.
871 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800872 */
873static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
874{
Johannes Weiner4e916722007-07-15 23:41:25 -0700875 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800876
877 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700878 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800879
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800880 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700881 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800882
Johannes Weiner4e916722007-07-15 23:41:25 -0700883 err = add_symlink(bo->sdir, bo->sdev);
884 if (err)
885 return err;
886
887 err = add_symlink(bo->hdir, bo->hdev);
888 if (err) {
889 del_symlink(bo->sdir, bo->sdev);
890 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700891 }
Johannes Weiner4e916722007-07-15 23:41:25 -0700892
893 list_add_tail(&bo->list, &bdev->bd_holder_list);
894 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800895}
896
897/**
898 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
899 *
900 * @bdev: block device to be bd_claimed
901 * @kobj: holder's kobject
902 *
903 * If there is matching entry with @kobj in @bdev->bd_holder_list
904 * and no other bd_claim() from the same kobject,
905 * remove the struct bd_holder from the list, delete symlinks for it.
906 *
907 * Returns a pointer to the struct bd_holder when it's removed from the list
908 * and ready to be freed.
909 * Returns NULL if matching claim isn't found or there is other bd_claim()
910 * by the same kobject.
911 */
912static struct bd_holder *del_bd_holder(struct block_device *bdev,
913 struct kobject *kobj)
914{
915 struct bd_holder *bo;
916
917 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
918 if (bo->sdir == kobj) {
919 bo->count--;
920 BUG_ON(bo->count < 0);
921 if (!bo->count) {
922 list_del(&bo->list);
923 del_symlink(bo->sdir, bo->sdev);
924 del_symlink(bo->hdir, bo->hdev);
925 bd_holder_release_dirs(bo);
926 return bo;
927 }
928 break;
929 }
930 }
931
932 return NULL;
933}
934
935/**
936 * bd_claim_by_kobject - bd_claim() with additional kobject signature
937 *
938 * @bdev: block device to be claimed
939 * @holder: holder's signature
940 * @kobj: holder's kobject
941 *
942 * Do bd_claim() and if it succeeds, create sysfs symlinks between
943 * the bdev and the holder's kobject.
944 * Use bd_release_from_kobject() when relesing the claimed bdev.
945 *
946 * Returns 0 on success. (same as bd_claim())
947 * Returns errno on failure.
948 */
949static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
950 struct kobject *kobj)
951{
Johannes Weiner4e916722007-07-15 23:41:25 -0700952 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500953 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800954
955 if (!kobj)
956 return -EINVAL;
957
958 bo = alloc_bd_holder(kobj);
959 if (!bo)
960 return -ENOMEM;
961
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800962 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500963
Johannes Weiner4e916722007-07-15 23:41:25 -0700964 err = bd_claim(bdev, holder);
965 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -0700966 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700967
968 found = find_bd_holder(bdev, bo);
969 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -0700970 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -0700971
972 err = add_bd_holder(bdev, bo);
973 if (err)
974 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700975 else
976 bo = NULL;
977fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800978 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -0700979 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -0700980 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800981}
982
983/**
984 * bd_release_from_kobject - bd_release() with additional kobject signature
985 *
986 * @bdev: block device to be released
987 * @kobj: holder's kobject
988 *
989 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
990 */
991static void bd_release_from_kobject(struct block_device *bdev,
992 struct kobject *kobj)
993{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800994 if (!kobj)
995 return;
996
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800997 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800998 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -0700999 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001000 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001001}
1002
1003/**
1004 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1005 *
1006 * @bdev: block device to be claimed
1007 * @holder: holder's signature
1008 * @disk: holder's gendisk
1009 *
1010 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1011 */
1012int bd_claim_by_disk(struct block_device *bdev, void *holder,
1013 struct gendisk *disk)
1014{
1015 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1016}
1017EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1018
1019/**
1020 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1021 *
1022 * @bdev: block device to be claimed
1023 * @disk: holder's gendisk
1024 *
1025 * Call bd_release_from_kobject() and put @disk->slave_dir.
1026 */
1027void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1028{
1029 bd_release_from_kobject(bdev, disk->slave_dir);
1030 kobject_put(disk->slave_dir);
1031}
1032EXPORT_SYMBOL_GPL(bd_release_from_disk);
1033#endif
1034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035/*
1036 * Tries to open block device by device number. Use it ONLY if you
1037 * really do not have anything better - i.e. when you are behind a
1038 * truly sucky interface and all you are given is a device number. _Never_
1039 * to be used for internal purposes. If you ever need it - reconsider
1040 * your API.
1041 */
1042struct block_device *open_by_devnum(dev_t dev, unsigned mode)
1043{
1044 struct block_device *bdev = bdget(dev);
1045 int err = -ENOMEM;
1046 int flags = mode & FMODE_WRITE ? O_RDWR : O_RDONLY;
1047 if (bdev)
1048 err = blkdev_get(bdev, mode, flags);
1049 return err ? ERR_PTR(err) : bdev;
1050}
1051
1052EXPORT_SYMBOL(open_by_devnum);
1053
1054/*
1055 * This routine checks whether a removable media has been changed,
1056 * and invalidates all buffer-cache-entries in that case. This
1057 * is a relatively slow routine, so we have to try to minimize using
1058 * it. Thus it is called only upon a 'mount' or 'open'. This
1059 * is the best way of combining speed and utility, I think.
1060 * People changing diskettes in the middle of an operation deserve
1061 * to lose :-)
1062 */
1063int check_disk_change(struct block_device *bdev)
1064{
1065 struct gendisk *disk = bdev->bd_disk;
1066 struct block_device_operations * bdops = disk->fops;
1067
1068 if (!bdops->media_changed)
1069 return 0;
1070 if (!bdops->media_changed(bdev->bd_disk))
1071 return 0;
1072
Christoph Hellwig2ef41632005-05-05 16:15:59 -07001073 if (__invalidate_device(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 printk("VFS: busy inodes on changed media.\n");
1075
1076 if (bdops->revalidate_disk)
1077 bdops->revalidate_disk(bdev->bd_disk);
1078 if (bdev->bd_disk->minors > 1)
1079 bdev->bd_invalidated = 1;
1080 return 1;
1081}
1082
1083EXPORT_SYMBOL(check_disk_change);
1084
1085void bd_set_size(struct block_device *bdev, loff_t size)
1086{
1087 unsigned bsize = bdev_hardsect_size(bdev);
1088
1089 bdev->bd_inode->i_size = size;
1090 while (bsize < PAGE_CACHE_SIZE) {
1091 if (size & bsize)
1092 break;
1093 bsize <<= 1;
1094 }
1095 bdev->bd_block_size = bsize;
1096 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1097}
1098EXPORT_SYMBOL(bd_set_size);
1099
NeilBrown37be4122006-12-08 02:36:16 -08001100static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1101 int for_part);
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001102static int __blkdev_put(struct block_device *bdev, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001103
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001104/*
1105 * bd_mutex locking:
1106 *
1107 * mutex_lock(part->bd_mutex)
1108 * mutex_lock_nested(whole->bd_mutex, 1)
1109 */
1110
NeilBrown37be4122006-12-08 02:36:16 -08001111static int do_open(struct block_device *bdev, struct file *file, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112{
1113 struct module *owner = NULL;
1114 struct gendisk *disk;
1115 int ret = -ENXIO;
1116 int part;
1117
1118 file->f_mapping = bdev->bd_inode->i_mapping;
1119 lock_kernel();
1120 disk = get_gendisk(bdev->bd_dev, &part);
1121 if (!disk) {
1122 unlock_kernel();
1123 bdput(bdev);
1124 return ret;
1125 }
1126 owner = disk->fops->owner;
1127
NeilBrown6796bf52006-12-08 02:36:16 -08001128 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (!bdev->bd_openers) {
1130 bdev->bd_disk = disk;
1131 bdev->bd_contains = bdev;
1132 if (!part) {
1133 struct backing_dev_info *bdi;
1134 if (disk->fops->open) {
1135 ret = disk->fops->open(bdev->bd_inode, file);
1136 if (ret)
1137 goto out_first;
1138 }
1139 if (!bdev->bd_openers) {
1140 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1141 bdi = blk_get_backing_dev_info(bdev);
1142 if (bdi == NULL)
1143 bdi = &default_backing_dev_info;
1144 bdev->bd_inode->i_data.backing_dev_info = bdi;
1145 }
1146 if (bdev->bd_invalidated)
1147 rescan_partitions(disk, bdev);
1148 } else {
1149 struct hd_struct *p;
1150 struct block_device *whole;
1151 whole = bdget_disk(disk, 0);
1152 ret = -ENOMEM;
1153 if (!whole)
1154 goto out_first;
NeilBrown37be4122006-12-08 02:36:16 -08001155 BUG_ON(for_part);
1156 ret = __blkdev_get(whole, file->f_mode, file->f_flags, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (ret)
1158 goto out_first;
1159 bdev->bd_contains = whole;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 p = disk->part[part - 1];
1161 bdev->bd_inode->i_data.backing_dev_info =
1162 whole->bd_inode->i_data.backing_dev_info;
1163 if (!(disk->flags & GENHD_FL_UP) || !p || !p->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 ret = -ENXIO;
1165 goto out_first;
1166 }
1167 kobject_get(&p->kobj);
1168 bdev->bd_part = p;
1169 bd_set_size(bdev, (loff_t) p->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
1171 } else {
1172 put_disk(disk);
1173 module_put(owner);
1174 if (bdev->bd_contains == bdev) {
1175 if (bdev->bd_disk->fops->open) {
1176 ret = bdev->bd_disk->fops->open(bdev->bd_inode, file);
1177 if (ret)
1178 goto out;
1179 }
1180 if (bdev->bd_invalidated)
1181 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
1183 }
1184 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001185 if (for_part)
1186 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001187 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 unlock_kernel();
1189 return 0;
1190
1191out_first:
1192 bdev->bd_disk = NULL;
1193 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
1194 if (bdev != bdev->bd_contains)
Peter Zijlstrac48f70c2006-12-08 02:36:18 -08001195 __blkdev_put(bdev->bd_contains, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 bdev->bd_contains = NULL;
1197 put_disk(disk);
1198 module_put(owner);
1199out:
Arjan van de Venc039e312006-03-23 03:00:28 -08001200 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 unlock_kernel();
1202 if (ret)
1203 bdput(bdev);
1204 return ret;
1205}
1206
NeilBrown37be4122006-12-08 02:36:16 -08001207static int __blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags,
1208 int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
1210 /*
1211 * This crockload is due to bad choice of ->open() type.
1212 * It will go away.
1213 * For now, block device ->open() routine must _not_
1214 * examine anything in 'inode' argument except ->i_rdev.
1215 */
1216 struct file fake_file = {};
1217 struct dentry fake_dentry = {};
1218 fake_file.f_mode = mode;
1219 fake_file.f_flags = flags;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001220 fake_file.f_path.dentry = &fake_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 fake_dentry.d_inode = bdev->bd_inode;
1222
NeilBrown37be4122006-12-08 02:36:16 -08001223 return do_open(bdev, &fake_file, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
NeilBrown37be4122006-12-08 02:36:16 -08001226int blkdev_get(struct block_device *bdev, mode_t mode, unsigned flags)
1227{
1228 return __blkdev_get(bdev, mode, flags, 0);
1229}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230EXPORT_SYMBOL(blkdev_get);
1231
1232static int blkdev_open(struct inode * inode, struct file * filp)
1233{
1234 struct block_device *bdev;
1235 int res;
1236
1237 /*
1238 * Preserve backwards compatibility and allow large file access
1239 * even if userspace doesn't ask for it explicitly. Some mkfs
1240 * binary needs it. We might want to drop this workaround
1241 * during an unstable branch.
1242 */
1243 filp->f_flags |= O_LARGEFILE;
1244
1245 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001246 if (bdev == NULL)
1247 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248
NeilBrown37be4122006-12-08 02:36:16 -08001249 res = do_open(bdev, filp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 if (res)
1251 return res;
1252
1253 if (!(filp->f_flags & O_EXCL) )
1254 return 0;
1255
1256 if (!(res = bd_claim(bdev, filp)))
1257 return 0;
1258
1259 blkdev_put(bdev);
1260 return res;
1261}
1262
NeilBrown37be4122006-12-08 02:36:16 -08001263static int __blkdev_put(struct block_device *bdev, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001264{
1265 int ret = 0;
1266 struct inode *bd_inode = bdev->bd_inode;
1267 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001268 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001269
NeilBrown6796bf52006-12-08 02:36:16 -08001270 mutex_lock_nested(&bdev->bd_mutex, for_part);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001271 lock_kernel();
NeilBrown37be4122006-12-08 02:36:16 -08001272 if (for_part)
1273 bdev->bd_part_count--;
1274
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001275 if (!--bdev->bd_openers) {
1276 sync_blockdev(bdev);
1277 kill_bdev(bdev);
1278 }
1279 if (bdev->bd_contains == bdev) {
1280 if (disk->fops->release)
1281 ret = disk->fops->release(bd_inode, NULL);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001282 }
1283 if (!bdev->bd_openers) {
1284 struct module *owner = disk->fops->owner;
1285
1286 put_disk(disk);
1287 module_put(owner);
1288
1289 if (bdev->bd_contains != bdev) {
1290 kobject_put(&bdev->bd_part->kobj);
1291 bdev->bd_part = NULL;
1292 }
1293 bdev->bd_disk = NULL;
1294 bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
NeilBrown37be4122006-12-08 02:36:16 -08001295 if (bdev != bdev->bd_contains)
1296 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001297 bdev->bd_contains = NULL;
1298 }
1299 unlock_kernel();
1300 mutex_unlock(&bdev->bd_mutex);
1301 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001302 if (victim)
1303 __blkdev_put(victim, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001304 return ret;
1305}
1306
NeilBrown37be4122006-12-08 02:36:16 -08001307int blkdev_put(struct block_device *bdev)
1308{
1309 return __blkdev_put(bdev, 0);
1310}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001311EXPORT_SYMBOL(blkdev_put);
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313static int blkdev_close(struct inode * inode, struct file * filp)
1314{
1315 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1316 if (bdev->bd_holder == filp)
1317 bd_release(bdev);
1318 return blkdev_put(bdev);
1319}
1320
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001321static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322{
1323 return blkdev_ioctl(file->f_mapping->host, file, cmd, arg);
1324}
1325
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001326const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 .readpage = blkdev_readpage,
1328 .writepage = blkdev_writepage,
1329 .sync_page = block_sync_page,
1330 .prepare_write = blkdev_prepare_write,
1331 .commit_write = blkdev_commit_write,
1332 .writepages = generic_writepages,
1333 .direct_IO = blkdev_direct_IO,
1334};
1335
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001336const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 .open = blkdev_open,
1338 .release = blkdev_close,
1339 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001340 .read = do_sync_read,
1341 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 .aio_read = generic_file_aio_read,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001343 .aio_write = generic_file_aio_write_nolock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 .mmap = generic_file_mmap,
1345 .fsync = block_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001346 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347#ifdef CONFIG_COMPAT
1348 .compat_ioctl = compat_blkdev_ioctl,
1349#endif
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);