blob: 4230252fd6895c97cb81541741f9e8e584e04322 [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>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070014#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#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>
Nick Piggin585d3bc2009-02-25 10:44:19 +010020#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010021#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mpage.h>
23#include <linux/mount.h>
24#include <linux/uio.h>
25#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070026#include <linux/log2.h>
Catalin Marinas2e1483c2009-06-11 13:24:13 +010027#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020029#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
31struct bdev_inode {
32 struct block_device bdev;
33 struct inode vfs_inode;
34};
35
Adrian Bunk4c54ac62008-02-18 13:48:31 +010036static const struct address_space_operations def_blk_aops;
37
Linus Torvalds1da177e2005-04-16 15:20:36 -070038static inline struct bdev_inode *BDEV_I(struct inode *inode)
39{
40 return container_of(inode, struct bdev_inode, vfs_inode);
41}
42
43inline struct block_device *I_BDEV(struct inode *inode)
44{
45 return &BDEV_I(inode)->bdev;
46}
47
48EXPORT_SYMBOL(I_BDEV);
49
Dave Chinnera5491e02010-10-21 11:49:26 +110050/*
51 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
52 * we need to move it onto the dirty list of @dst so that the inode is always
53 * on the right list.
54 */
55static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
57{
58 spin_lock(&inode_lock);
59 inode->i_data.backing_dev_info = dst;
60 if (inode->i_state & I_DIRTY)
Nick Piggin7ccf19a2010-10-21 11:49:30 +110061 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Dave Chinnera5491e02010-10-21 11:49:26 +110062 spin_unlock(&inode_lock);
63}
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065static sector_t max_block(struct block_device *bdev)
66{
67 sector_t retval = ~((sector_t)0);
68 loff_t sz = i_size_read(bdev->bd_inode);
69
70 if (sz) {
71 unsigned int size = block_size(bdev);
72 unsigned int sizebits = blksize_bits(size);
73 retval = (sz >> sizebits);
74 }
75 return retval;
76}
77
Peter Zijlstraf9a14392007-05-06 14:49:55 -070078/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070079static void kill_bdev(struct block_device *bdev)
80{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070081 if (bdev->bd_inode->i_mapping->nrpages == 0)
82 return;
83 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
85}
86
87int set_blocksize(struct block_device *bdev, int size)
88{
89 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070090 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070091 return -EINVAL;
92
93 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -040094 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 return -EINVAL;
96
97 /* Don't change the size if it is same as current */
98 if (bdev->bd_block_size != size) {
99 sync_blockdev(bdev);
100 bdev->bd_block_size = size;
101 bdev->bd_inode->i_blkbits = blksize_bits(size);
102 kill_bdev(bdev);
103 }
104 return 0;
105}
106
107EXPORT_SYMBOL(set_blocksize);
108
109int sb_set_blocksize(struct super_block *sb, int size)
110{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 if (set_blocksize(sb->s_bdev, size))
112 return 0;
113 /* If we get here, we know size is power of two
114 * and it's value is between 512 and PAGE_SIZE */
115 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800116 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return sb->s_blocksize;
118}
119
120EXPORT_SYMBOL(sb_set_blocksize);
121
122int sb_min_blocksize(struct super_block *sb, int size)
123{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400124 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 if (size < minsize)
126 size = minsize;
127 return sb_set_blocksize(sb, size);
128}
129
130EXPORT_SYMBOL(sb_min_blocksize);
131
132static int
133blkdev_get_block(struct inode *inode, sector_t iblock,
134 struct buffer_head *bh, int create)
135{
136 if (iblock >= max_block(I_BDEV(inode))) {
137 if (create)
138 return -EIO;
139
140 /*
141 * for reads, we're just trying to fill a partial page.
142 * return a hole, they will have to call get_block again
143 * before they can fill it, and they will get -EIO at that
144 * time
145 */
146 return 0;
147 }
148 bh->b_bdev = I_BDEV(inode);
149 bh->b_blocknr = iblock;
150 set_buffer_mapped(bh);
151 return 0;
152}
153
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800154static int
155blkdev_get_blocks(struct inode *inode, sector_t iblock,
156 struct buffer_head *bh, int create)
157{
158 sector_t end_block = max_block(I_BDEV(inode));
159 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
160
161 if ((iblock + max_blocks) > end_block) {
162 max_blocks = end_block - iblock;
163 if ((long)max_blocks <= 0) {
164 if (create)
165 return -EIO; /* write fully beyond EOF */
166 /*
167 * It is a read which is fully beyond EOF. We return
168 * a !buffer_mapped buffer
169 */
170 max_blocks = 0;
171 }
172 }
173
174 bh->b_bdev = I_BDEV(inode);
175 bh->b_blocknr = iblock;
176 bh->b_size = max_blocks << inode->i_blkbits;
177 if (max_blocks)
178 set_buffer_mapped(bh);
179 return 0;
180}
181
182static ssize_t
183blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
184 loff_t offset, unsigned long nr_segs)
185{
186 struct file *file = iocb->ki_filp;
187 struct inode *inode = file->f_mapping->host;
188
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200189 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
190 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800191}
192
Jan Kara5cee5812009-04-27 16:43:51 +0200193int __sync_blockdev(struct block_device *bdev, int wait)
194{
195 if (!bdev)
196 return 0;
197 if (!wait)
198 return filemap_flush(bdev->bd_inode->i_mapping);
199 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
200}
201
Nick Piggin585d3bc2009-02-25 10:44:19 +0100202/*
203 * Write out and wait upon all the dirty data associated with a block
204 * device via its mapping. Does not take the superblock lock.
205 */
206int sync_blockdev(struct block_device *bdev)
207{
Jan Kara5cee5812009-04-27 16:43:51 +0200208 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100209}
210EXPORT_SYMBOL(sync_blockdev);
211
212/*
213 * Write out and wait upon all dirty data associated with this
214 * device. Filesystem data as well as the underlying block
215 * device. Takes the superblock lock.
216 */
217int fsync_bdev(struct block_device *bdev)
218{
219 struct super_block *sb = get_super(bdev);
220 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200221 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100222 drop_super(sb);
223 return res;
224 }
225 return sync_blockdev(bdev);
226}
Al Viro47e44912009-04-01 07:07:16 -0400227EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100228
229/**
230 * freeze_bdev -- lock a filesystem and force it into a consistent state
231 * @bdev: blockdevice to lock
232 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100233 * If a superblock is found on this device, we take the s_umount semaphore
234 * on it to make sure nobody unmounts until the snapshot creation is done.
235 * The reference counter (bd_fsfreeze_count) guarantees that only the last
236 * unfreeze process can unfreeze the frozen filesystem actually when multiple
237 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
238 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
239 * actually.
240 */
241struct super_block *freeze_bdev(struct block_device *bdev)
242{
243 struct super_block *sb;
244 int error = 0;
245
246 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200247 if (++bdev->bd_fsfreeze_count > 1) {
248 /*
249 * We don't even need to grab a reference - the first call
250 * to freeze_bdev grab an active reference and only the last
251 * thaw_bdev drops it.
252 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100253 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200254 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100255 mutex_unlock(&bdev->bd_fsfreeze_mutex);
256 return sb;
257 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100258
Christoph Hellwig45042302009-08-03 23:28:35 +0200259 sb = get_active_super(bdev);
260 if (!sb)
261 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400262 error = freeze_super(sb);
263 if (error) {
264 deactivate_super(sb);
265 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200266 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400267 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100268 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400269 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200270 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100271 sync_blockdev(bdev);
272 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200273 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100274}
275EXPORT_SYMBOL(freeze_bdev);
276
277/**
278 * thaw_bdev -- unlock filesystem
279 * @bdev: blockdevice to unlock
280 * @sb: associated superblock
281 *
282 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
283 */
284int thaw_bdev(struct block_device *bdev, struct super_block *sb)
285{
Christoph Hellwig45042302009-08-03 23:28:35 +0200286 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100287
288 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200289 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400290 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100291
Christoph Hellwig45042302009-08-03 23:28:35 +0200292 error = 0;
293 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400294 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100295
Christoph Hellwig45042302009-08-03 23:28:35 +0200296 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400297 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200298
Josef Bacik18e9e512010-03-23 10:34:56 -0400299 error = thaw_super(sb);
300 if (error) {
301 bdev->bd_fsfreeze_count++;
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
303 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100304 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400305out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100306 mutex_unlock(&bdev->bd_fsfreeze_mutex);
307 return 0;
308}
309EXPORT_SYMBOL(thaw_bdev);
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
312{
313 return block_write_full_page(page, blkdev_get_block, wbc);
314}
315
316static int blkdev_readpage(struct file * file, struct page * page)
317{
318 return block_read_full_page(page, blkdev_get_block);
319}
320
Nick Piggin6272b5a2007-10-16 01:25:04 -0700321static int blkdev_write_begin(struct file *file, struct address_space *mapping,
322 loff_t pos, unsigned len, unsigned flags,
323 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200325 return block_write_begin(mapping, pos, len, flags, pagep,
326 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327}
328
Nick Piggin6272b5a2007-10-16 01:25:04 -0700329static int blkdev_write_end(struct file *file, struct address_space *mapping,
330 loff_t pos, unsigned len, unsigned copied,
331 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700333 int ret;
334 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
335
336 unlock_page(page);
337 page_cache_release(page);
338
339 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342/*
343 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800344 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 * so we compute the size by hand (just as in block_read/write above)
346 */
347static loff_t block_llseek(struct file *file, loff_t offset, int origin)
348{
349 struct inode *bd_inode = file->f_mapping->host;
350 loff_t size;
351 loff_t retval;
352
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800353 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 size = i_size_read(bd_inode);
355
356 switch (origin) {
357 case 2:
358 offset += size;
359 break;
360 case 1:
361 offset += file->f_pos;
362 }
363 retval = -EINVAL;
364 if (offset >= 0 && offset <= size) {
365 if (offset != file->f_pos) {
366 file->f_pos = offset;
367 }
368 retval = offset;
369 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800370 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 return retval;
372}
373
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200374int blkdev_fsync(struct file *filp, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400376 struct inode *bd_inode = filp->f_mapping->host;
377 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100378 int error;
379
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400380 /*
381 * There is no need to serialise calls to blkdev_issue_flush with
382 * i_mutex and doing so causes performance issues with concurrent
383 * O_SYNC writers to a block device.
384 */
385 mutex_unlock(&bd_inode->i_mutex);
386
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200387 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100388 if (error == -EOPNOTSUPP)
389 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400390
391 mutex_lock(&bd_inode->i_mutex);
392
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100393 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700395EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397/*
398 * pseudo-fs
399 */
400
401static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800402static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
404static struct inode *bdev_alloc_inode(struct super_block *sb)
405{
Christoph Lametere94b1762006-12-06 20:33:17 -0800406 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 if (!ei)
408 return NULL;
409 return &ei->vfs_inode;
410}
411
412static void bdev_destroy_inode(struct inode *inode)
413{
414 struct bdev_inode *bdi = BDEV_I(inode);
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 kmem_cache_free(bdev_cachep, bdi);
417}
418
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700419static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
421 struct bdev_inode *ei = (struct bdev_inode *) foo;
422 struct block_device *bdev = &ei->bdev;
423
Christoph Lametera35afb82007-05-16 22:10:57 -0700424 memset(bdev, 0, sizeof(*bdev));
425 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700426 INIT_LIST_HEAD(&bdev->bd_inodes);
427 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800428#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700429 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800430#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700431 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800432 /* Initialize mutex for freeze. */
433 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
436static inline void __bd_forget(struct inode *inode)
437{
438 list_del_init(&inode->i_devices);
439 inode->i_bdev = NULL;
440 inode->i_mapping = &inode->i_data;
441}
442
Al Virob57922d2010-06-07 14:34:48 -0400443static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444{
445 struct block_device *bdev = &BDEV_I(inode)->bdev;
446 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400447 truncate_inode_pages(&inode->i_data, 0);
448 invalidate_inode_buffers(inode); /* is it needed here? */
449 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 spin_lock(&bdev_lock);
451 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
452 __bd_forget(list_entry(p, struct inode, i_devices));
453 }
454 list_del_init(&bdev->bd_list);
455 spin_unlock(&bdev_lock);
456}
457
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800458static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 .statfs = simple_statfs,
460 .alloc_inode = bdev_alloc_inode,
461 .destroy_inode = bdev_destroy_inode,
462 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400463 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464};
465
Al Viro51139ad2010-07-25 23:47:46 +0400466static struct dentry *bd_mount(struct file_system_type *fs_type,
467 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468{
Al Viro51139ad2010-07-25 23:47:46 +0400469 return mount_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470}
471
472static struct file_system_type bd_type = {
473 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400474 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 .kill_sb = kill_anon_super,
476};
477
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800478struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480void __init bdev_cache_init(void)
481{
482 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800483 struct vfsmount *bd_mnt;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800486 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
487 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900488 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 err = register_filesystem(&bd_type);
490 if (err)
491 panic("Cannot register bdev pseudo-fs");
492 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (IS_ERR(bd_mnt))
494 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100495 /*
496 * This vfsmount structure is only used to obtain the
497 * blockdev_superblock, so tell kmemleak not to report it.
498 */
499 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
501}
502
503/*
504 * Most likely _very_ bad one - but then it's hardly critical for small
505 * /dev and can be fixed when somebody will need really large one.
506 * Keep in mind that it will be fed through icache hash function too.
507 */
508static inline unsigned long hash(dev_t dev)
509{
510 return MAJOR(dev)+MINOR(dev);
511}
512
513static int bdev_test(struct inode *inode, void *data)
514{
515 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
516}
517
518static int bdev_set(struct inode *inode, void *data)
519{
520 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
521 return 0;
522}
523
524static LIST_HEAD(all_bdevs);
525
526struct block_device *bdget(dev_t dev)
527{
528 struct block_device *bdev;
529 struct inode *inode;
530
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800531 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 bdev_test, bdev_set, &dev);
533
534 if (!inode)
535 return NULL;
536
537 bdev = &BDEV_I(inode)->bdev;
538
539 if (inode->i_state & I_NEW) {
540 bdev->bd_contains = NULL;
541 bdev->bd_inode = inode;
542 bdev->bd_block_size = (1 << inode->i_blkbits);
543 bdev->bd_part_count = 0;
544 bdev->bd_invalidated = 0;
545 inode->i_mode = S_IFBLK;
546 inode->i_rdev = dev;
547 inode->i_bdev = bdev;
548 inode->i_data.a_ops = &def_blk_aops;
549 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
550 inode->i_data.backing_dev_info = &default_backing_dev_info;
551 spin_lock(&bdev_lock);
552 list_add(&bdev->bd_list, &all_bdevs);
553 spin_unlock(&bdev_lock);
554 unlock_new_inode(inode);
555 }
556 return bdev;
557}
558
559EXPORT_SYMBOL(bdget);
560
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200561/**
562 * bdgrab -- Grab a reference to an already referenced block device
563 * @bdev: Block device to grab a reference to.
564 */
565struct block_device *bdgrab(struct block_device *bdev)
566{
Al Viro7de9c6e2010-10-23 11:11:40 -0400567 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200568 return bdev;
569}
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571long nr_blockdev_pages(void)
572{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700573 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 long ret = 0;
575 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700576 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 ret += bdev->bd_inode->i_mapping->nrpages;
578 }
579 spin_unlock(&bdev_lock);
580 return ret;
581}
582
583void bdput(struct block_device *bdev)
584{
585 iput(bdev->bd_inode);
586}
587
588EXPORT_SYMBOL(bdput);
589
590static struct block_device *bd_acquire(struct inode *inode)
591{
592 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 spin_lock(&bdev_lock);
595 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700596 if (bdev) {
Al Viro7de9c6e2010-10-23 11:11:40 -0400597 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 spin_unlock(&bdev_lock);
599 return bdev;
600 }
601 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 bdev = bdget(inode->i_rdev);
604 if (bdev) {
605 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700606 if (!inode->i_bdev) {
607 /*
Al Viro7de9c6e2010-10-23 11:11:40 -0400608 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700609 * and it's released in clear_inode() of inode.
610 * So, we can access it via ->i_mapping always
611 * without igrab().
612 */
Al Viro7de9c6e2010-10-23 11:11:40 -0400613 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700614 inode->i_bdev = bdev;
615 inode->i_mapping = bdev->bd_inode->i_mapping;
616 list_add(&inode->i_devices, &bdev->bd_inodes);
617 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 spin_unlock(&bdev_lock);
619 }
620 return bdev;
621}
622
623/* Call when you free inode */
624
625void bd_forget(struct inode *inode)
626{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700627 struct block_device *bdev = NULL;
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700630 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800631 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700632 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700636
637 if (bdev)
638 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
640
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900641/**
642 * bd_may_claim - test whether a block device can be claimed
643 * @bdev: block device of interest
644 * @whole: whole block device containing @bdev, may equal @bdev
645 * @holder: holder trying to claim @bdev
646 *
647 * Test whther @bdev can be claimed by @holder.
648 *
649 * CONTEXT:
650 * spin_lock(&bdev_lock).
651 *
652 * RETURNS:
653 * %true if @bdev can be claimed, %false otherwise.
654 */
655static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
656 void *holder)
657{
658 if (bdev->bd_holder == holder)
659 return true; /* already a holder */
660 else if (bdev->bd_holder != NULL)
661 return false; /* held by someone else */
662 else if (bdev->bd_contains == bdev)
663 return true; /* is a whole device which isn't held */
664
665 else if (whole->bd_holder == bd_claim)
666 return true; /* is a partition of a device that is being partitioned */
667 else if (whole->bd_holder != NULL)
668 return false; /* is a partition of a held device */
669 else
670 return true; /* is a partition of an un-held device */
671}
672
673/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900674 * bd_prepare_to_claim - prepare to claim a block device
675 * @bdev: block device of interest
676 * @whole: the whole device containing @bdev, may equal @bdev
677 * @holder: holder trying to claim @bdev
678 *
679 * Prepare to claim @bdev. This function fails if @bdev is already
680 * claimed by another holder and waits if another claiming is in
681 * progress. This function doesn't actually claim. On successful
682 * return, the caller has ownership of bd_claiming and bd_holder[s].
683 *
684 * CONTEXT:
685 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
686 * it multiple times.
687 *
688 * RETURNS:
689 * 0 if @bdev can be claimed, -EBUSY otherwise.
690 */
691static int bd_prepare_to_claim(struct block_device *bdev,
692 struct block_device *whole, void *holder)
693{
694retry:
695 /* if someone else claimed, fail */
696 if (!bd_may_claim(bdev, whole, holder))
697 return -EBUSY;
698
Tejun Heoe75aa852010-08-04 17:59:39 +0200699 /* if claiming is already in progress, wait for it to finish */
700 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900701 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
702 DEFINE_WAIT(wait);
703
704 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
705 spin_unlock(&bdev_lock);
706 schedule();
707 finish_wait(wq, &wait);
708 spin_lock(&bdev_lock);
709 goto retry;
710 }
711
712 /* yay, all mine */
713 return 0;
714}
715
716/**
717 * bd_start_claiming - start claiming a block device
718 * @bdev: block device of interest
719 * @holder: holder trying to claim @bdev
720 *
721 * @bdev is about to be opened exclusively. Check @bdev can be opened
722 * exclusively and mark that an exclusive open is in progress. Each
723 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000724 * either bd_finish_claiming() or bd_abort_claiming() (which do not
725 * fail).
726 *
727 * This function is used to gain exclusive access to the block device
728 * without actually causing other exclusive open attempts to fail. It
729 * should be used when the open sequence itself requires exclusive
730 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900731 *
732 * CONTEXT:
733 * Might sleep.
734 *
735 * RETURNS:
736 * Pointer to the block device containing @bdev on success, ERR_PTR()
737 * value on failure.
738 */
739static struct block_device *bd_start_claiming(struct block_device *bdev,
740 void *holder)
741{
742 struct gendisk *disk;
743 struct block_device *whole;
744 int partno, err;
745
746 might_sleep();
747
748 /*
749 * @bdev might not have been initialized properly yet, look up
750 * and grab the outer block device the hard way.
751 */
752 disk = get_gendisk(bdev->bd_dev, &partno);
753 if (!disk)
754 return ERR_PTR(-ENXIO);
755
756 whole = bdget_disk(disk, 0);
Nick Piggincf342572010-05-26 01:50:21 +1000757 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900758 put_disk(disk);
759 if (!whole)
760 return ERR_PTR(-ENOMEM);
761
762 /* prepare to claim, if successful, mark claiming in progress */
763 spin_lock(&bdev_lock);
764
765 err = bd_prepare_to_claim(bdev, whole, holder);
766 if (err == 0) {
767 whole->bd_claiming = holder;
768 spin_unlock(&bdev_lock);
769 return whole;
770 } else {
771 spin_unlock(&bdev_lock);
772 bdput(whole);
773 return ERR_PTR(err);
774 }
775}
776
777/* releases bdev_lock */
778static void __bd_abort_claiming(struct block_device *whole, void *holder)
779{
780 BUG_ON(whole->bd_claiming != holder);
781 whole->bd_claiming = NULL;
782 wake_up_bit(&whole->bd_claiming, 0);
783
784 spin_unlock(&bdev_lock);
785 bdput(whole);
786}
787
788/**
789 * bd_abort_claiming - abort claiming a block device
790 * @whole: whole block device returned by bd_start_claiming()
791 * @holder: holder trying to claim @bdev
792 *
793 * Abort a claiming block started by bd_start_claiming(). Note that
794 * @whole is not the block device to be claimed but the whole device
795 * returned by bd_start_claiming().
796 *
797 * CONTEXT:
798 * Grabs and releases bdev_lock.
799 */
800static void bd_abort_claiming(struct block_device *whole, void *holder)
801{
802 spin_lock(&bdev_lock);
803 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
804}
805
Nick Pigginb0018362010-05-26 01:51:19 +1000806/* increment holders when we have a legitimate claim. requires bdev_lock */
807static void __bd_claim(struct block_device *bdev, struct block_device *whole,
808 void *holder)
809{
810 /* note that for a whole device bd_holders
811 * will be incremented twice, and bd_holder will
812 * be set to bd_claim before being set to holder
813 */
814 whole->bd_holders++;
815 whole->bd_holder = bd_claim;
816 bdev->bd_holders++;
817 bdev->bd_holder = holder;
818}
819
820/**
821 * bd_finish_claiming - finish claiming a block device
822 * @bdev: block device of interest (passed to bd_start_claiming())
823 * @whole: whole block device returned by bd_start_claiming()
824 * @holder: holder trying to claim @bdev
825 *
826 * Finish a claiming block started by bd_start_claiming().
827 *
828 * CONTEXT:
829 * Grabs and releases bdev_lock.
830 */
831static void bd_finish_claiming(struct block_device *bdev,
832 struct block_device *whole, void *holder)
833{
834 spin_lock(&bdev_lock);
Nick Pigginb0018362010-05-26 01:51:19 +1000835 BUG_ON(!bd_may_claim(bdev, whole, holder));
836 __bd_claim(bdev, whole, holder);
837 __bd_abort_claiming(whole, holder); /* not actually an abort */
838}
839
Tejun Heo6b4517a2010-04-07 18:53:59 +0900840/**
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900841 * bd_claim - claim a block device
842 * @bdev: block device to claim
843 * @holder: holder trying to claim @bdev
844 *
Nick Pigginb0018362010-05-26 01:51:19 +1000845 * Try to claim @bdev which must have been opened successfully.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900846 *
847 * CONTEXT:
848 * Might sleep.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900849 *
850 * RETURNS:
851 * 0 if successful, -EBUSY if @bdev is already claimed.
852 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853int bd_claim(struct block_device *bdev, void *holder)
854{
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900855 struct block_device *whole = bdev->bd_contains;
Tejun Heo6b4517a2010-04-07 18:53:59 +0900856 int res;
857
858 might_sleep();
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 spin_lock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900861 res = bd_prepare_to_claim(bdev, whole, holder);
Nick Pigginb0018362010-05-26 01:51:19 +1000862 if (res == 0)
863 __bd_claim(bdev, whole, holder);
864 spin_unlock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900865
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 return res;
867}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868EXPORT_SYMBOL(bd_claim);
869
870void bd_release(struct block_device *bdev)
871{
872 spin_lock(&bdev_lock);
873 if (!--bdev->bd_contains->bd_holders)
874 bdev->bd_contains->bd_holder = NULL;
875 if (!--bdev->bd_holders)
876 bdev->bd_holder = NULL;
877 spin_unlock(&bdev_lock);
878}
879
880EXPORT_SYMBOL(bd_release);
881
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800882#ifdef CONFIG_SYSFS
883/*
884 * Functions for bd_claim_by_kobject / bd_release_from_kobject
885 *
886 * If a kobject is passed to bd_claim_by_kobject()
887 * and the kobject has a parent directory,
888 * following symlinks are created:
889 * o from the kobject to the claimed bdev
890 * o from "holders" directory of the bdev to the parent of the kobject
891 * bd_release_from_kobject() removes these symlinks.
892 *
893 * Example:
894 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
895 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
896 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
897 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
898 */
899
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700900static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800901{
902 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700903 return 0;
904 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800905}
906
907static void del_symlink(struct kobject *from, struct kobject *to)
908{
909 if (!from || !to)
910 return;
911 sysfs_remove_link(from, kobject_name(to));
912}
913
914/*
915 * 'struct bd_holder' contains pointers to kobjects symlinked by
916 * bd_claim_by_kobject.
917 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
918 */
919struct bd_holder {
920 struct list_head list; /* chain of holders of the bdev */
921 int count; /* references from the holder */
922 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
923 struct kobject *hdev; /* e.g. "/block/dm-0" */
924 struct kobject *hdir; /* e.g. "/block/sda/holders" */
925 struct kobject *sdev; /* e.g. "/block/sda" */
926};
927
928/*
929 * Get references of related kobjects at once.
930 * Returns 1 on success. 0 on failure.
931 *
932 * Should call bd_holder_release_dirs() after successful use.
933 */
934static int bd_holder_grab_dirs(struct block_device *bdev,
935 struct bd_holder *bo)
936{
937 if (!bdev || !bo)
938 return 0;
939
940 bo->sdir = kobject_get(bo->sdir);
941 if (!bo->sdir)
942 return 0;
943
944 bo->hdev = kobject_get(bo->sdir->parent);
945 if (!bo->hdev)
946 goto fail_put_sdir;
947
Tejun Heo0762b8b2008-08-25 19:56:12 +0900948 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800949 if (!bo->sdev)
950 goto fail_put_hdev;
951
Tejun Heo4c465012008-08-25 19:56:11 +0900952 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800953 if (!bo->hdir)
954 goto fail_put_sdev;
955
956 return 1;
957
958fail_put_sdev:
959 kobject_put(bo->sdev);
960fail_put_hdev:
961 kobject_put(bo->hdev);
962fail_put_sdir:
963 kobject_put(bo->sdir);
964
965 return 0;
966}
967
968/* Put references of related kobjects at once. */
969static void bd_holder_release_dirs(struct bd_holder *bo)
970{
971 kobject_put(bo->hdir);
972 kobject_put(bo->sdev);
973 kobject_put(bo->hdev);
974 kobject_put(bo->sdir);
975}
976
977static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
978{
979 struct bd_holder *bo;
980
981 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
982 if (!bo)
983 return NULL;
984
985 bo->count = 1;
986 bo->sdir = kobj;
987
988 return bo;
989}
990
991static void free_bd_holder(struct bd_holder *bo)
992{
993 kfree(bo);
994}
995
996/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500997 * find_bd_holder - find matching struct bd_holder from the block device
998 *
999 * @bdev: struct block device to be searched
1000 * @bo: target struct bd_holder
1001 *
1002 * Returns matching entry with @bo in @bdev->bd_holder_list.
1003 * If found, increment the reference count and return the pointer.
1004 * If not found, returns NULL.
1005 */
Andrew Morton36a561d2006-10-30 22:07:03 -08001006static struct bd_holder *find_bd_holder(struct block_device *bdev,
1007 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001008{
1009 struct bd_holder *tmp;
1010
1011 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1012 if (tmp->sdir == bo->sdir) {
1013 tmp->count++;
1014 return tmp;
1015 }
1016
1017 return NULL;
1018}
1019
1020/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001021 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1022 *
1023 * @bdev: block device to be bd_claimed
1024 * @bo: preallocated and initialized by alloc_bd_holder()
1025 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001026 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001027 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001028 * Returns 0 if symlinks are created.
1029 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001030 */
1031static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1032{
Johannes Weiner4e916722007-07-15 23:41:25 -07001033 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001034
1035 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001036 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001037
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001038 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001039 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001040
Johannes Weiner4e916722007-07-15 23:41:25 -07001041 err = add_symlink(bo->sdir, bo->sdev);
1042 if (err)
1043 return err;
1044
1045 err = add_symlink(bo->hdir, bo->hdev);
1046 if (err) {
1047 del_symlink(bo->sdir, bo->sdev);
1048 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001049 }
Johannes Weiner4e916722007-07-15 23:41:25 -07001050
1051 list_add_tail(&bo->list, &bdev->bd_holder_list);
1052 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001053}
1054
1055/**
1056 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1057 *
1058 * @bdev: block device to be bd_claimed
1059 * @kobj: holder's kobject
1060 *
1061 * If there is matching entry with @kobj in @bdev->bd_holder_list
1062 * and no other bd_claim() from the same kobject,
1063 * remove the struct bd_holder from the list, delete symlinks for it.
1064 *
1065 * Returns a pointer to the struct bd_holder when it's removed from the list
1066 * and ready to be freed.
1067 * Returns NULL if matching claim isn't found or there is other bd_claim()
1068 * by the same kobject.
1069 */
1070static struct bd_holder *del_bd_holder(struct block_device *bdev,
1071 struct kobject *kobj)
1072{
1073 struct bd_holder *bo;
1074
1075 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1076 if (bo->sdir == kobj) {
1077 bo->count--;
1078 BUG_ON(bo->count < 0);
1079 if (!bo->count) {
1080 list_del(&bo->list);
1081 del_symlink(bo->sdir, bo->sdev);
1082 del_symlink(bo->hdir, bo->hdev);
1083 bd_holder_release_dirs(bo);
1084 return bo;
1085 }
1086 break;
1087 }
1088 }
1089
1090 return NULL;
1091}
1092
1093/**
1094 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1095 *
1096 * @bdev: block device to be claimed
1097 * @holder: holder's signature
1098 * @kobj: holder's kobject
1099 *
1100 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1101 * the bdev and the holder's kobject.
1102 * Use bd_release_from_kobject() when relesing the claimed bdev.
1103 *
1104 * Returns 0 on success. (same as bd_claim())
1105 * Returns errno on failure.
1106 */
1107static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1108 struct kobject *kobj)
1109{
Johannes Weiner4e916722007-07-15 23:41:25 -07001110 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001111 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001112
1113 if (!kobj)
1114 return -EINVAL;
1115
1116 bo = alloc_bd_holder(kobj);
1117 if (!bo)
1118 return -ENOMEM;
1119
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001120 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001121
Johannes Weiner4e916722007-07-15 23:41:25 -07001122 err = bd_claim(bdev, holder);
1123 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -07001124 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001125
1126 found = find_bd_holder(bdev, bo);
1127 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -07001128 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001129
1130 err = add_bd_holder(bdev, bo);
1131 if (err)
1132 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001133 else
1134 bo = NULL;
1135fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001136 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -07001137 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -07001138 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001139}
1140
1141/**
1142 * bd_release_from_kobject - bd_release() with additional kobject signature
1143 *
1144 * @bdev: block device to be released
1145 * @kobj: holder's kobject
1146 *
1147 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1148 */
1149static void bd_release_from_kobject(struct block_device *bdev,
1150 struct kobject *kobj)
1151{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001152 if (!kobj)
1153 return;
1154
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001155 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001156 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001157 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001158 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001159}
1160
1161/**
1162 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1163 *
1164 * @bdev: block device to be claimed
1165 * @holder: holder's signature
1166 * @disk: holder's gendisk
1167 *
1168 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1169 */
1170int bd_claim_by_disk(struct block_device *bdev, void *holder,
1171 struct gendisk *disk)
1172{
1173 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1174}
1175EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1176
1177/**
1178 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1179 *
1180 * @bdev: block device to be claimed
1181 * @disk: holder's gendisk
1182 *
1183 * Call bd_release_from_kobject() and put @disk->slave_dir.
1184 */
1185void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1186{
1187 bd_release_from_kobject(bdev, disk->slave_dir);
1188 kobject_put(disk->slave_dir);
1189}
1190EXPORT_SYMBOL_GPL(bd_release_from_disk);
1191#endif
1192
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193/*
1194 * Tries to open block device by device number. Use it ONLY if you
1195 * really do not have anything better - i.e. when you are behind a
1196 * truly sucky interface and all you are given is a device number. _Never_
1197 * to be used for internal purposes. If you ever need it - reconsider
1198 * your API.
1199 */
Al Viroaeb5d722008-09-02 15:28:45 -04001200struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201{
1202 struct block_device *bdev = bdget(dev);
1203 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001205 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 return err ? ERR_PTR(err) : bdev;
1207}
1208
1209EXPORT_SYMBOL(open_by_devnum);
1210
Andrew Patterson0c002c22008-09-04 14:27:20 -06001211/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001212 * flush_disk - invalidates all buffer-cache entries on a disk
1213 *
1214 * @bdev: struct block device to be flushed
1215 *
1216 * Invalidates all buffer-cache entries on a disk. It should be called
1217 * when a disk has been changed -- either by a media change or online
1218 * resize.
1219 */
1220static void flush_disk(struct block_device *bdev)
1221{
1222 if (__invalidate_device(bdev)) {
1223 char name[BDEVNAME_SIZE] = "";
1224
1225 if (bdev->bd_disk)
1226 disk_name(bdev->bd_disk, 0, name);
1227 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1228 "resized disk %s\n", name);
1229 }
1230
1231 if (!bdev->bd_disk)
1232 return;
1233 if (disk_partitionable(bdev->bd_disk))
1234 bdev->bd_invalidated = 1;
1235}
1236
1237/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001238 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001239 * @disk: struct gendisk to check
1240 * @bdev: struct bdev to adjust.
1241 *
1242 * This routine checks to see if the bdev size does not match the disk size
1243 * and adjusts it if it differs.
1244 */
1245void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1246{
1247 loff_t disk_size, bdev_size;
1248
1249 disk_size = (loff_t)get_capacity(disk) << 9;
1250 bdev_size = i_size_read(bdev->bd_inode);
1251 if (disk_size != bdev_size) {
1252 char name[BDEVNAME_SIZE];
1253
1254 disk_name(disk, 0, name);
1255 printk(KERN_INFO
1256 "%s: detected capacity change from %lld to %lld\n",
1257 name, bdev_size, disk_size);
1258 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001259 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001260 }
1261}
1262EXPORT_SYMBOL(check_disk_size_change);
1263
1264/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001265 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001266 * @disk: struct gendisk to be revalidated
1267 *
1268 * This routine is a wrapper for lower-level driver's revalidate_disk
1269 * call-backs. It is used to do common pre and post operations needed
1270 * for all revalidate_disk operations.
1271 */
1272int revalidate_disk(struct gendisk *disk)
1273{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001274 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001275 int ret = 0;
1276
1277 if (disk->fops->revalidate_disk)
1278 ret = disk->fops->revalidate_disk(disk);
1279
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001280 bdev = bdget_disk(disk, 0);
1281 if (!bdev)
1282 return ret;
1283
1284 mutex_lock(&bdev->bd_mutex);
1285 check_disk_size_change(disk, bdev);
1286 mutex_unlock(&bdev->bd_mutex);
1287 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001288 return ret;
1289}
1290EXPORT_SYMBOL(revalidate_disk);
1291
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292/*
1293 * This routine checks whether a removable media has been changed,
1294 * and invalidates all buffer-cache-entries in that case. This
1295 * is a relatively slow routine, so we have to try to minimize using
1296 * it. Thus it is called only upon a 'mount' or 'open'. This
1297 * is the best way of combining speed and utility, I think.
1298 * People changing diskettes in the middle of an operation deserve
1299 * to lose :-)
1300 */
1301int check_disk_change(struct block_device *bdev)
1302{
1303 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001304 const struct block_device_operations *bdops = disk->fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306 if (!bdops->media_changed)
1307 return 0;
1308 if (!bdops->media_changed(bdev->bd_disk))
1309 return 0;
1310
Andrew Patterson56ade442008-09-04 14:27:40 -06001311 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312 if (bdops->revalidate_disk)
1313 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return 1;
1315}
1316
1317EXPORT_SYMBOL(check_disk_change);
1318
1319void bd_set_size(struct block_device *bdev, loff_t size)
1320{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001321 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322
1323 bdev->bd_inode->i_size = size;
1324 while (bsize < PAGE_CACHE_SIZE) {
1325 if (size & bsize)
1326 break;
1327 bsize <<= 1;
1328 }
1329 bdev->bd_block_size = bsize;
1330 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1331}
1332EXPORT_SYMBOL(bd_set_size);
1333
Al Viro9a1c3542008-02-22 20:40:24 -05001334static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001335
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001336/*
1337 * bd_mutex locking:
1338 *
1339 * mutex_lock(part->bd_mutex)
1340 * mutex_lock_nested(whole->bd_mutex, 1)
1341 */
1342
Al Viro572c4892007-10-08 13:24:05 -04001343static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001346 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001347 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001348 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
Al Viro572c4892007-10-08 13:24:05 -04001350 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001351 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001352 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001353 perm |= MAY_WRITE;
1354 /*
1355 * hooks: /n/, see "layering violations".
1356 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001357 if (!for_part) {
1358 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1359 if (ret != 0) {
1360 bdput(bdev);
1361 return ret;
1362 }
Al Viro82666022008-08-01 05:32:04 -04001363 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001364
NeilBrownd3374822009-01-09 08:31:10 +11001365 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001366
Tejun Heo89f97492008-11-05 10:21:06 +01001367 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001368 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001369 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001370 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
NeilBrown6796bf52006-12-08 02:36:16 -08001372 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (!bdev->bd_openers) {
1374 bdev->bd_disk = disk;
1375 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001376 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001378
1379 ret = -ENXIO;
1380 bdev->bd_part = disk_get_part(disk, partno);
1381 if (!bdev->bd_part)
1382 goto out_clear;
1383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001385 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001386 if (ret == -ERESTARTSYS) {
1387 /* Lost a race with 'disk' being
1388 * deleted, try again.
1389 * See md.c
1390 */
1391 disk_put_part(bdev->bd_part);
1392 bdev->bd_part = NULL;
1393 module_put(disk->fops->owner);
1394 put_disk(disk);
1395 bdev->bd_disk = NULL;
1396 mutex_unlock(&bdev->bd_mutex);
1397 goto restart;
1398 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001400 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 }
1402 if (!bdev->bd_openers) {
1403 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1404 bdi = blk_get_backing_dev_info(bdev);
1405 if (bdi == NULL)
1406 bdi = &default_backing_dev_info;
Dave Chinnera5491e02010-10-21 11:49:26 +11001407 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 }
1409 if (bdev->bd_invalidated)
1410 rescan_partitions(disk, bdev);
1411 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 struct block_device *whole;
1413 whole = bdget_disk(disk, 0);
1414 ret = -ENOMEM;
1415 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001416 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001417 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001418 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001420 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001422 bdev_inode_switch_bdi(bdev->bd_inode,
1423 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001424 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001425 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001426 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001428 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 }
Tejun Heo89f97492008-11-05 10:21:06 +01001430 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 }
1432 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001433 module_put(disk->fops->owner);
Neil Brown960cc0f2009-10-26 08:59:17 +01001434 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001435 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 if (bdev->bd_contains == bdev) {
1437 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001438 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001440 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 }
1442 if (bdev->bd_invalidated)
1443 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 }
1445 }
1446 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001447 if (for_part)
1448 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001449 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 return 0;
1451
Tejun Heo0762b8b2008-08-25 19:56:12 +09001452 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001453 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001455 bdev->bd_part = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001456 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001458 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001460 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001461 mutex_unlock(&bdev->bd_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001462 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001463 if (disk)
1464 module_put(disk->fops->owner);
1465 put_disk(disk);
1466 bdput(bdev);
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 return ret;
1469}
1470
Al Viro572c4892007-10-08 13:24:05 -04001471int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472{
Al Viro572c4892007-10-08 13:24:05 -04001473 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001474}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475EXPORT_SYMBOL(blkdev_get);
1476
1477static int blkdev_open(struct inode * inode, struct file * filp)
1478{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001479 struct block_device *whole = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 struct block_device *bdev;
1481 int res;
1482
1483 /*
1484 * Preserve backwards compatibility and allow large file access
1485 * even if userspace doesn't ask for it explicitly. Some mkfs
1486 * binary needs it. We might want to drop this workaround
1487 * during an unstable branch.
1488 */
1489 filp->f_flags |= O_LARGEFILE;
1490
Al Viro572c4892007-10-08 13:24:05 -04001491 if (filp->f_flags & O_NDELAY)
1492 filp->f_mode |= FMODE_NDELAY;
1493 if (filp->f_flags & O_EXCL)
1494 filp->f_mode |= FMODE_EXCL;
1495 if ((filp->f_flags & O_ACCMODE) == 3)
1496 filp->f_mode |= FMODE_WRITE_IOCTL;
1497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001499 if (bdev == NULL)
1500 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Tejun Heo6b4517a2010-04-07 18:53:59 +09001502 if (filp->f_mode & FMODE_EXCL) {
1503 whole = bd_start_claiming(bdev, filp);
1504 if (IS_ERR(whole)) {
1505 bdput(bdev);
1506 return PTR_ERR(whole);
1507 }
1508 }
1509
Al Viro572c4892007-10-08 13:24:05 -04001510 filp->f_mapping = bdev->bd_inode->i_mapping;
1511
1512 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Tejun Heo6b4517a2010-04-07 18:53:59 +09001514 if (whole) {
1515 if (res == 0)
Nick Pigginb0018362010-05-26 01:51:19 +10001516 bd_finish_claiming(bdev, whole, filp);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001517 else
1518 bd_abort_claiming(whole, filp);
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001519 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 return res;
1522}
1523
Al Viro9a1c3542008-02-22 20:40:24 -05001524static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001525{
1526 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001527 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001528 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001529
NeilBrown6796bf52006-12-08 02:36:16 -08001530 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001531 if (for_part)
1532 bdev->bd_part_count--;
1533
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001534 if (!--bdev->bd_openers) {
1535 sync_blockdev(bdev);
1536 kill_bdev(bdev);
1537 }
1538 if (bdev->bd_contains == bdev) {
1539 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001540 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001541 }
1542 if (!bdev->bd_openers) {
1543 struct module *owner = disk->fops->owner;
1544
1545 put_disk(disk);
1546 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001547 disk_put_part(bdev->bd_part);
1548 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001549 bdev->bd_disk = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001550 bdev_inode_switch_bdi(bdev->bd_inode,
1551 &default_backing_dev_info);
NeilBrown37be4122006-12-08 02:36:16 -08001552 if (bdev != bdev->bd_contains)
1553 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001554 bdev->bd_contains = NULL;
1555 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001556 mutex_unlock(&bdev->bd_mutex);
1557 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001558 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001559 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001560 return ret;
1561}
1562
Al Viro9a1c3542008-02-22 20:40:24 -05001563int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001564{
Al Viro9a1c3542008-02-22 20:40:24 -05001565 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001566}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001567EXPORT_SYMBOL(blkdev_put);
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569static int blkdev_close(struct inode * inode, struct file * filp)
1570{
1571 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1572 if (bdev->bd_holder == filp)
1573 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001574 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575}
1576
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001577static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578{
Al Viro56b26ad2008-09-19 03:17:36 -04001579 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1580 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001581
1582 /*
1583 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1584 * to updated it before every ioctl.
1585 */
Al Viro56b26ad2008-09-19 03:17:36 -04001586 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001587 mode |= FMODE_NDELAY;
1588 else
1589 mode &= ~FMODE_NDELAY;
1590
Al Viro56b26ad2008-09-19 03:17:36 -04001591 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592}
1593
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001594/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001595 * Write data to the block device. Only intended for the block device itself
1596 * and the raw driver which basically is a fake block device.
1597 *
1598 * Does not take i_mutex for the write and thus is not for general purpose
1599 * use.
1600 */
1601ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1602 unsigned long nr_segs, loff_t pos)
1603{
1604 struct file *file = iocb->ki_filp;
1605 ssize_t ret;
1606
1607 BUG_ON(iocb->ki_pos != pos);
1608
1609 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1610 if (ret > 0 || ret == -EIOCBQUEUED) {
1611 ssize_t err;
1612
1613 err = generic_write_sync(file, pos, ret);
1614 if (err < 0 && ret > 0)
1615 ret = err;
1616 }
1617 return ret;
1618}
1619EXPORT_SYMBOL_GPL(blkdev_aio_write);
1620
1621/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001622 * Try to release a page associated with block device when the system
1623 * is under memory pressure.
1624 */
1625static int blkdev_releasepage(struct page *page, gfp_t wait)
1626{
1627 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1628
1629 if (super && super->s_op->bdev_try_to_free_page)
1630 return super->s_op->bdev_try_to_free_page(super, page, wait);
1631
1632 return try_to_free_buffers(page);
1633}
1634
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001635static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 .readpage = blkdev_readpage,
1637 .writepage = blkdev_writepage,
1638 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001639 .write_begin = blkdev_write_begin,
1640 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001642 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643 .direct_IO = blkdev_direct_IO,
1644};
1645
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001646const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 .open = blkdev_open,
1648 .release = blkdev_close,
1649 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001650 .read = do_sync_read,
1651 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001653 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001655 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001656 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657#ifdef CONFIG_COMPAT
1658 .compat_ioctl = compat_blkdev_ioctl,
1659#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001660 .splice_read = generic_file_splice_read,
1661 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662};
1663
1664int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1665{
1666 int res;
1667 mm_segment_t old_fs = get_fs();
1668 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001669 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 set_fs(old_fs);
1671 return res;
1672}
1673
1674EXPORT_SYMBOL(ioctl_by_bdev);
1675
1676/**
1677 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001678 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001680 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 * namespace if possible and return it. Return ERR_PTR(error)
1682 * otherwise.
1683 */
Al Viro421748e2008-08-02 01:04:36 -04001684struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685{
1686 struct block_device *bdev;
1687 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001688 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 int error;
1690
Al Viro421748e2008-08-02 01:04:36 -04001691 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 return ERR_PTR(-EINVAL);
1693
Al Viro421748e2008-08-02 01:04:36 -04001694 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 if (error)
1696 return ERR_PTR(error);
1697
Al Viro421748e2008-08-02 01:04:36 -04001698 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 error = -ENOTBLK;
1700 if (!S_ISBLK(inode->i_mode))
1701 goto fail;
1702 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001703 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 goto fail;
1705 error = -ENOMEM;
1706 bdev = bd_acquire(inode);
1707 if (!bdev)
1708 goto fail;
1709out:
Al Viro421748e2008-08-02 01:04:36 -04001710 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 return bdev;
1712fail:
1713 bdev = ERR_PTR(error);
1714 goto out;
1715}
Al Virod5686b42008-08-01 05:00:11 -04001716EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718/**
Al Viro30c40d22008-02-22 19:50:45 -05001719 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 *
1721 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001722 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 * @holder: owner for exclusion
1724 *
1725 * Open the blockdevice described by the special file at @path, claim it
1726 * for the @holder.
1727 */
Al Viro30c40d22008-02-22 19:50:45 -05001728struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001730 struct block_device *bdev, *whole;
1731 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 bdev = lookup_bdev(path);
1734 if (IS_ERR(bdev))
1735 return bdev;
1736
Tejun Heo6b4517a2010-04-07 18:53:59 +09001737 whole = bd_start_claiming(bdev, holder);
1738 if (IS_ERR(whole)) {
1739 bdput(bdev);
1740 return whole;
1741 }
1742
Al Viro572c4892007-10-08 13:24:05 -04001743 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 if (error)
Tejun Heo6b4517a2010-04-07 18:53:59 +09001745 goto out_abort_claiming;
1746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001748 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Tejun Heo6b4517a2010-04-07 18:53:59 +09001749 goto out_blkdev_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750
Nick Pigginb0018362010-05-26 01:51:19 +10001751 bd_finish_claiming(bdev, whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 return bdev;
Tejun Heo6b4517a2010-04-07 18:53:59 +09001753
1754out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001755 blkdev_put(bdev, mode);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001756out_abort_claiming:
1757 bd_abort_claiming(whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 return ERR_PTR(error);
1759}
1760
Al Viro30c40d22008-02-22 19:50:45 -05001761EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762
1763/**
Al Viro30c40d22008-02-22 19:50:45 -05001764 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 *
1766 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001767 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 *
Al Viro30c40d22008-02-22 19:50:45 -05001769 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 */
Al Viro30c40d22008-02-22 19:50:45 -05001771void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772{
1773 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001774 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775}
1776
Al Viro30c40d22008-02-22 19:50:45 -05001777EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001778
1779int __invalidate_device(struct block_device *bdev)
1780{
1781 struct super_block *sb = get_super(bdev);
1782 int res = 0;
1783
1784 if (sb) {
1785 /*
1786 * no need to lock the super, get_super holds the
1787 * read mutex so the filesystem cannot go away
1788 * under us (->put_super runs with the write lock
1789 * hold).
1790 */
1791 shrink_dcache_sb(sb);
1792 res = invalidate_inodes(sb);
1793 drop_super(sb);
1794 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001795 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001796 return res;
1797}
1798EXPORT_SYMBOL(__invalidate_device);