blob: 0e575d1304b461e67bfbbba6ef234c53838b8998 [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>
Al Viroff01bb42011-09-16 02:31:11 -040020#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010021#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010022#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/mpage.h>
24#include <linux/mount.h>
25#include <linux/uio.h>
26#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070027#include <linux/log2.h>
Al Viroff01bb42011-09-16 02:31:11 -040028#include <linux/cleancache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32struct bdev_inode {
33 struct block_device bdev;
34 struct inode vfs_inode;
35};
36
Adrian Bunk4c54ac62008-02-18 13:48:31 +010037static const struct address_space_operations def_blk_aops;
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039static inline struct bdev_inode *BDEV_I(struct inode *inode)
40{
41 return container_of(inode, struct bdev_inode, vfs_inode);
42}
43
44inline struct block_device *I_BDEV(struct inode *inode)
45{
46 return &BDEV_I(inode)->bdev;
47}
Linus Torvalds1da177e2005-04-16 15:20:36 -070048EXPORT_SYMBOL(I_BDEV);
49
Dave Chinnera5491e02010-10-21 11:49:26 +110050/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -060051 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
52 * need to move it onto the dirty list of @dst so that the inode is always on
53 * the right list.
Dave Chinnera5491e02010-10-21 11:49:26 +110054 */
55static void bdev_inode_switch_bdi(struct inode *inode,
56 struct backing_dev_info *dst)
57{
Christoph Hellwigf758eea2011-04-21 18:19:44 -060058 struct backing_dev_info *old = inode->i_data.backing_dev_info;
59
60 if (unlikely(dst == old)) /* deadlock avoidance */
61 return;
62 bdi_lock_two(&old->wb, &dst->wb);
Dave Chinner250df6e2011-03-22 22:23:36 +110063 spin_lock(&inode->i_lock);
Dave Chinnera5491e02010-10-21 11:49:26 +110064 inode->i_data.backing_dev_info = dst;
65 if (inode->i_state & I_DIRTY)
Nick Piggin7ccf19a2010-10-21 11:49:30 +110066 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Dave Chinner250df6e2011-03-22 22:23:36 +110067 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -060068 spin_unlock(&old->wb.list_lock);
69 spin_unlock(&dst->wb.list_lock);
Dave Chinnera5491e02010-10-21 11:49:26 +110070}
71
Linus Torvalds1da177e2005-04-16 15:20:36 -070072static sector_t max_block(struct block_device *bdev)
73{
74 sector_t retval = ~((sector_t)0);
75 loff_t sz = i_size_read(bdev->bd_inode);
76
77 if (sz) {
78 unsigned int size = block_size(bdev);
79 unsigned int sizebits = blksize_bits(size);
80 retval = (sz >> sizebits);
81 }
82 return retval;
83}
84
Peter Zijlstraf9a14392007-05-06 14:49:55 -070085/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040086void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087{
Al Viroff01bb42011-09-16 02:31:11 -040088 struct address_space *mapping = bdev->bd_inode->i_mapping;
89
90 if (mapping->nrpages == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070091 return;
Al Viroff01bb42011-09-16 02:31:11 -040092
Peter Zijlstraf9a14392007-05-06 14:49:55 -070093 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040094 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095}
Al Viroff01bb42011-09-16 02:31:11 -040096EXPORT_SYMBOL(kill_bdev);
97
98/* Invalidate clean unused buffers and pagecache. */
99void invalidate_bdev(struct block_device *bdev)
100{
101 struct address_space *mapping = bdev->bd_inode->i_mapping;
102
103 if (mapping->nrpages == 0)
104 return;
105
106 invalidate_bh_lrus();
107 lru_add_drain_all(); /* make sure all lru add caches are flushed */
108 invalidate_mapping_pages(mapping, 0, -1);
109 /* 99% of the time, we don't need to flush the cleancache on the bdev.
110 * But, for the strange corners, lets be cautious
111 */
112 cleancache_flush_inode(mapping);
113}
114EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116int set_blocksize(struct block_device *bdev, int size)
117{
118 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700119 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return -EINVAL;
121
122 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400123 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 return -EINVAL;
125
126 /* Don't change the size if it is same as current */
127 if (bdev->bd_block_size != size) {
128 sync_blockdev(bdev);
129 bdev->bd_block_size = size;
130 bdev->bd_inode->i_blkbits = blksize_bits(size);
131 kill_bdev(bdev);
132 }
133 return 0;
134}
135
136EXPORT_SYMBOL(set_blocksize);
137
138int sb_set_blocksize(struct super_block *sb, int size)
139{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 if (set_blocksize(sb->s_bdev, size))
141 return 0;
142 /* If we get here, we know size is power of two
143 * and it's value is between 512 and PAGE_SIZE */
144 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800145 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 return sb->s_blocksize;
147}
148
149EXPORT_SYMBOL(sb_set_blocksize);
150
151int sb_min_blocksize(struct super_block *sb, int size)
152{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400153 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 if (size < minsize)
155 size = minsize;
156 return sb_set_blocksize(sb, size);
157}
158
159EXPORT_SYMBOL(sb_min_blocksize);
160
161static int
162blkdev_get_block(struct inode *inode, sector_t iblock,
163 struct buffer_head *bh, int create)
164{
165 if (iblock >= max_block(I_BDEV(inode))) {
166 if (create)
167 return -EIO;
168
169 /*
170 * for reads, we're just trying to fill a partial page.
171 * return a hole, they will have to call get_block again
172 * before they can fill it, and they will get -EIO at that
173 * time
174 */
175 return 0;
176 }
177 bh->b_bdev = I_BDEV(inode);
178 bh->b_blocknr = iblock;
179 set_buffer_mapped(bh);
180 return 0;
181}
182
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800183static int
184blkdev_get_blocks(struct inode *inode, sector_t iblock,
185 struct buffer_head *bh, int create)
186{
187 sector_t end_block = max_block(I_BDEV(inode));
188 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
189
190 if ((iblock + max_blocks) > end_block) {
191 max_blocks = end_block - iblock;
192 if ((long)max_blocks <= 0) {
193 if (create)
194 return -EIO; /* write fully beyond EOF */
195 /*
196 * It is a read which is fully beyond EOF. We return
197 * a !buffer_mapped buffer
198 */
199 max_blocks = 0;
200 }
201 }
202
203 bh->b_bdev = I_BDEV(inode);
204 bh->b_blocknr = iblock;
205 bh->b_size = max_blocks << inode->i_blkbits;
206 if (max_blocks)
207 set_buffer_mapped(bh);
208 return 0;
209}
210
211static ssize_t
212blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
213 loff_t offset, unsigned long nr_segs)
214{
215 struct file *file = iocb->ki_filp;
216 struct inode *inode = file->f_mapping->host;
217
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200218 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
219 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800220}
221
Jan Kara5cee5812009-04-27 16:43:51 +0200222int __sync_blockdev(struct block_device *bdev, int wait)
223{
224 if (!bdev)
225 return 0;
226 if (!wait)
227 return filemap_flush(bdev->bd_inode->i_mapping);
228 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
229}
230
Nick Piggin585d3bc2009-02-25 10:44:19 +0100231/*
232 * Write out and wait upon all the dirty data associated with a block
233 * device via its mapping. Does not take the superblock lock.
234 */
235int sync_blockdev(struct block_device *bdev)
236{
Jan Kara5cee5812009-04-27 16:43:51 +0200237 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100238}
239EXPORT_SYMBOL(sync_blockdev);
240
241/*
242 * Write out and wait upon all dirty data associated with this
243 * device. Filesystem data as well as the underlying block
244 * device. Takes the superblock lock.
245 */
246int fsync_bdev(struct block_device *bdev)
247{
248 struct super_block *sb = get_super(bdev);
249 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200250 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251 drop_super(sb);
252 return res;
253 }
254 return sync_blockdev(bdev);
255}
Al Viro47e44912009-04-01 07:07:16 -0400256EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100257
258/**
259 * freeze_bdev -- lock a filesystem and force it into a consistent state
260 * @bdev: blockdevice to lock
261 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100262 * If a superblock is found on this device, we take the s_umount semaphore
263 * on it to make sure nobody unmounts until the snapshot creation is done.
264 * The reference counter (bd_fsfreeze_count) guarantees that only the last
265 * unfreeze process can unfreeze the frozen filesystem actually when multiple
266 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
267 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
268 * actually.
269 */
270struct super_block *freeze_bdev(struct block_device *bdev)
271{
272 struct super_block *sb;
273 int error = 0;
274
275 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200276 if (++bdev->bd_fsfreeze_count > 1) {
277 /*
278 * We don't even need to grab a reference - the first call
279 * to freeze_bdev grab an active reference and only the last
280 * thaw_bdev drops it.
281 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100282 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200283 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100284 mutex_unlock(&bdev->bd_fsfreeze_mutex);
285 return sb;
286 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100287
Christoph Hellwig45042302009-08-03 23:28:35 +0200288 sb = get_active_super(bdev);
289 if (!sb)
290 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400291 error = freeze_super(sb);
292 if (error) {
293 deactivate_super(sb);
294 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200295 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400296 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100297 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400298 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200299 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100300 sync_blockdev(bdev);
301 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200302 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100303}
304EXPORT_SYMBOL(freeze_bdev);
305
306/**
307 * thaw_bdev -- unlock filesystem
308 * @bdev: blockdevice to unlock
309 * @sb: associated superblock
310 *
311 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
312 */
313int thaw_bdev(struct block_device *bdev, struct super_block *sb)
314{
Christoph Hellwig45042302009-08-03 23:28:35 +0200315 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100316
317 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200318 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400319 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100320
Christoph Hellwig45042302009-08-03 23:28:35 +0200321 error = 0;
322 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400323 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100324
Christoph Hellwig45042302009-08-03 23:28:35 +0200325 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400326 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200327
Josef Bacik18e9e512010-03-23 10:34:56 -0400328 error = thaw_super(sb);
329 if (error) {
330 bdev->bd_fsfreeze_count++;
331 mutex_unlock(&bdev->bd_fsfreeze_mutex);
332 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100333 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400334out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100335 mutex_unlock(&bdev->bd_fsfreeze_mutex);
336 return 0;
337}
338EXPORT_SYMBOL(thaw_bdev);
339
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
341{
342 return block_write_full_page(page, blkdev_get_block, wbc);
343}
344
345static int blkdev_readpage(struct file * file, struct page * page)
346{
347 return block_read_full_page(page, blkdev_get_block);
348}
349
Nick Piggin6272b5a2007-10-16 01:25:04 -0700350static int blkdev_write_begin(struct file *file, struct address_space *mapping,
351 loff_t pos, unsigned len, unsigned flags,
352 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200354 return block_write_begin(mapping, pos, len, flags, pagep,
355 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356}
357
Nick Piggin6272b5a2007-10-16 01:25:04 -0700358static int blkdev_write_end(struct file *file, struct address_space *mapping,
359 loff_t pos, unsigned len, unsigned copied,
360 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700362 int ret;
363 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
364
365 unlock_page(page);
366 page_cache_release(page);
367
368 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
371/*
372 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800373 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 * so we compute the size by hand (just as in block_read/write above)
375 */
376static loff_t block_llseek(struct file *file, loff_t offset, int origin)
377{
378 struct inode *bd_inode = file->f_mapping->host;
379 loff_t size;
380 loff_t retval;
381
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800382 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 size = i_size_read(bd_inode);
384
Josef Bacik06222e42011-07-18 13:21:38 -0400385 retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 switch (origin) {
Josef Bacik06222e42011-07-18 13:21:38 -0400387 case SEEK_END:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 offset += size;
389 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400390 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400392 case SEEK_SET:
393 break;
394 default:
395 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 if (offset >= 0 && offset <= size) {
398 if (offset != file->f_pos) {
399 file->f_pos = offset;
400 }
401 retval = offset;
402 }
Josef Bacik06222e42011-07-18 13:21:38 -0400403out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800404 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 return retval;
406}
407
Josef Bacik02c24a82011-07-16 20:44:56 -0400408int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400410 struct inode *bd_inode = filp->f_mapping->host;
411 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100412 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200413
414 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
415 if (error)
416 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100417
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400418 /*
419 * There is no need to serialise calls to blkdev_issue_flush with
420 * i_mutex and doing so causes performance issues with concurrent
421 * O_SYNC writers to a block device.
422 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200423 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100424 if (error == -EOPNOTSUPP)
425 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400426
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100427 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700429EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
431/*
432 * pseudo-fs
433 */
434
435static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800436static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438static struct inode *bdev_alloc_inode(struct super_block *sb)
439{
Christoph Lametere94b1762006-12-06 20:33:17 -0800440 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 if (!ei)
442 return NULL;
443 return &ei->vfs_inode;
444}
445
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100446static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100448 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 struct bdev_inode *bdi = BDEV_I(inode);
450
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 kmem_cache_free(bdev_cachep, bdi);
452}
453
Nick Pigginfa0d7e32011-01-07 17:49:49 +1100454static void bdev_destroy_inode(struct inode *inode)
455{
456 call_rcu(&inode->i_rcu, bdev_i_callback);
457}
458
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700459static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 struct bdev_inode *ei = (struct bdev_inode *) foo;
462 struct block_device *bdev = &ei->bdev;
463
Christoph Lametera35afb82007-05-16 22:10:57 -0700464 memset(bdev, 0, sizeof(*bdev));
465 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700466 INIT_LIST_HEAD(&bdev->bd_inodes);
467 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100468#ifdef CONFIG_SYSFS
469 INIT_LIST_HEAD(&bdev->bd_holder_disks);
470#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700471 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800472 /* Initialize mutex for freeze. */
473 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
476static inline void __bd_forget(struct inode *inode)
477{
478 list_del_init(&inode->i_devices);
479 inode->i_bdev = NULL;
480 inode->i_mapping = &inode->i_data;
481}
482
Al Virob57922d2010-06-07 14:34:48 -0400483static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 struct block_device *bdev = &BDEV_I(inode)->bdev;
486 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400487 truncate_inode_pages(&inode->i_data, 0);
488 invalidate_inode_buffers(inode); /* is it needed here? */
489 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 spin_lock(&bdev_lock);
491 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
492 __bd_forget(list_entry(p, struct inode, i_devices));
493 }
494 list_del_init(&bdev->bd_list);
495 spin_unlock(&bdev_lock);
496}
497
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800498static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 .statfs = simple_statfs,
500 .alloc_inode = bdev_alloc_inode,
501 .destroy_inode = bdev_destroy_inode,
502 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400503 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504};
505
Al Viro51139ad2010-07-25 23:47:46 +0400506static struct dentry *bd_mount(struct file_system_type *fs_type,
507 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
Al Viroc74a1cb2011-01-12 16:59:34 -0500509 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, 0x62646576);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510}
511
512static struct file_system_type bd_type = {
513 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400514 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 .kill_sb = kill_anon_super,
516};
517
Al Virof47ec3f2011-11-21 21:15:42 -0500518static struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
520void __init bdev_cache_init(void)
521{
522 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300523 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800524
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800526 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
527 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900528 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 err = register_filesystem(&bd_type);
530 if (err)
531 panic("Cannot register bdev pseudo-fs");
532 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 if (IS_ERR(bd_mnt))
534 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300535 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536}
537
538/*
539 * Most likely _very_ bad one - but then it's hardly critical for small
540 * /dev and can be fixed when somebody will need really large one.
541 * Keep in mind that it will be fed through icache hash function too.
542 */
543static inline unsigned long hash(dev_t dev)
544{
545 return MAJOR(dev)+MINOR(dev);
546}
547
548static int bdev_test(struct inode *inode, void *data)
549{
550 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
551}
552
553static int bdev_set(struct inode *inode, void *data)
554{
555 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
556 return 0;
557}
558
559static LIST_HEAD(all_bdevs);
560
561struct block_device *bdget(dev_t dev)
562{
563 struct block_device *bdev;
564 struct inode *inode;
565
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800566 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 bdev_test, bdev_set, &dev);
568
569 if (!inode)
570 return NULL;
571
572 bdev = &BDEV_I(inode)->bdev;
573
574 if (inode->i_state & I_NEW) {
575 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000576 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 bdev->bd_inode = inode;
578 bdev->bd_block_size = (1 << inode->i_blkbits);
579 bdev->bd_part_count = 0;
580 bdev->bd_invalidated = 0;
581 inode->i_mode = S_IFBLK;
582 inode->i_rdev = dev;
583 inode->i_bdev = bdev;
584 inode->i_data.a_ops = &def_blk_aops;
585 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
586 inode->i_data.backing_dev_info = &default_backing_dev_info;
587 spin_lock(&bdev_lock);
588 list_add(&bdev->bd_list, &all_bdevs);
589 spin_unlock(&bdev_lock);
590 unlock_new_inode(inode);
591 }
592 return bdev;
593}
594
595EXPORT_SYMBOL(bdget);
596
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200597/**
598 * bdgrab -- Grab a reference to an already referenced block device
599 * @bdev: Block device to grab a reference to.
600 */
601struct block_device *bdgrab(struct block_device *bdev)
602{
Al Viro7de9c6e2010-10-23 11:11:40 -0400603 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200604 return bdev;
605}
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607long nr_blockdev_pages(void)
608{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700609 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 long ret = 0;
611 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700612 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 ret += bdev->bd_inode->i_mapping->nrpages;
614 }
615 spin_unlock(&bdev_lock);
616 return ret;
617}
618
619void bdput(struct block_device *bdev)
620{
621 iput(bdev->bd_inode);
622}
623
624EXPORT_SYMBOL(bdput);
625
626static struct block_device *bd_acquire(struct inode *inode)
627{
628 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 spin_lock(&bdev_lock);
631 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700632 if (bdev) {
Al Viro7de9c6e2010-10-23 11:11:40 -0400633 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 spin_unlock(&bdev_lock);
635 return bdev;
636 }
637 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 bdev = bdget(inode->i_rdev);
640 if (bdev) {
641 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700642 if (!inode->i_bdev) {
643 /*
Al Viro7de9c6e2010-10-23 11:11:40 -0400644 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700645 * and it's released in clear_inode() of inode.
646 * So, we can access it via ->i_mapping always
647 * without igrab().
648 */
Al Viro7de9c6e2010-10-23 11:11:40 -0400649 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700650 inode->i_bdev = bdev;
651 inode->i_mapping = bdev->bd_inode->i_mapping;
652 list_add(&inode->i_devices, &bdev->bd_inodes);
653 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 spin_unlock(&bdev_lock);
655 }
656 return bdev;
657}
658
Al Virof47ec3f2011-11-21 21:15:42 -0500659static inline int sb_is_blkdev_sb(struct super_block *sb)
660{
661 return sb == blockdev_superblock;
662}
663
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664/* Call when you free inode */
665
666void bd_forget(struct inode *inode)
667{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700668 struct block_device *bdev = NULL;
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700671 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800672 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700673 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700675 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700677
678 if (bdev)
679 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680}
681
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900682/**
683 * bd_may_claim - test whether a block device can be claimed
684 * @bdev: block device of interest
685 * @whole: whole block device containing @bdev, may equal @bdev
686 * @holder: holder trying to claim @bdev
687 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300688 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900689 *
690 * CONTEXT:
691 * spin_lock(&bdev_lock).
692 *
693 * RETURNS:
694 * %true if @bdev can be claimed, %false otherwise.
695 */
696static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
697 void *holder)
698{
699 if (bdev->bd_holder == holder)
700 return true; /* already a holder */
701 else if (bdev->bd_holder != NULL)
702 return false; /* held by someone else */
703 else if (bdev->bd_contains == bdev)
704 return true; /* is a whole device which isn't held */
705
Tejun Heoe525fd82010-11-13 11:55:17 +0100706 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900707 return true; /* is a partition of a device that is being partitioned */
708 else if (whole->bd_holder != NULL)
709 return false; /* is a partition of a held device */
710 else
711 return true; /* is a partition of an un-held device */
712}
713
714/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900715 * bd_prepare_to_claim - prepare to claim a block device
716 * @bdev: block device of interest
717 * @whole: the whole device containing @bdev, may equal @bdev
718 * @holder: holder trying to claim @bdev
719 *
720 * Prepare to claim @bdev. This function fails if @bdev is already
721 * claimed by another holder and waits if another claiming is in
722 * progress. This function doesn't actually claim. On successful
723 * return, the caller has ownership of bd_claiming and bd_holder[s].
724 *
725 * CONTEXT:
726 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
727 * it multiple times.
728 *
729 * RETURNS:
730 * 0 if @bdev can be claimed, -EBUSY otherwise.
731 */
732static int bd_prepare_to_claim(struct block_device *bdev,
733 struct block_device *whole, void *holder)
734{
735retry:
736 /* if someone else claimed, fail */
737 if (!bd_may_claim(bdev, whole, holder))
738 return -EBUSY;
739
Tejun Heoe75aa852010-08-04 17:59:39 +0200740 /* if claiming is already in progress, wait for it to finish */
741 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900742 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
743 DEFINE_WAIT(wait);
744
745 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
746 spin_unlock(&bdev_lock);
747 schedule();
748 finish_wait(wq, &wait);
749 spin_lock(&bdev_lock);
750 goto retry;
751 }
752
753 /* yay, all mine */
754 return 0;
755}
756
757/**
758 * bd_start_claiming - start claiming a block device
759 * @bdev: block device of interest
760 * @holder: holder trying to claim @bdev
761 *
762 * @bdev is about to be opened exclusively. Check @bdev can be opened
763 * exclusively and mark that an exclusive open is in progress. Each
764 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000765 * either bd_finish_claiming() or bd_abort_claiming() (which do not
766 * fail).
767 *
768 * This function is used to gain exclusive access to the block device
769 * without actually causing other exclusive open attempts to fail. It
770 * should be used when the open sequence itself requires exclusive
771 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900772 *
773 * CONTEXT:
774 * Might sleep.
775 *
776 * RETURNS:
777 * Pointer to the block device containing @bdev on success, ERR_PTR()
778 * value on failure.
779 */
780static struct block_device *bd_start_claiming(struct block_device *bdev,
781 void *holder)
782{
783 struct gendisk *disk;
784 struct block_device *whole;
785 int partno, err;
786
787 might_sleep();
788
789 /*
790 * @bdev might not have been initialized properly yet, look up
791 * and grab the outer block device the hard way.
792 */
793 disk = get_gendisk(bdev->bd_dev, &partno);
794 if (!disk)
795 return ERR_PTR(-ENXIO);
796
Tejun Heod4c208b2011-06-13 12:45:48 +0200797 /*
798 * Normally, @bdev should equal what's returned from bdget_disk()
799 * if partno is 0; however, some drivers (floppy) use multiple
800 * bdev's for the same physical device and @bdev may be one of the
801 * aliases. Keep @bdev if partno is 0. This means claimer
802 * tracking is broken for those devices but it has always been that
803 * way.
804 */
805 if (partno)
806 whole = bdget_disk(disk, 0);
807 else
808 whole = bdgrab(bdev);
809
Nick Piggincf342572010-05-26 01:50:21 +1000810 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900811 put_disk(disk);
812 if (!whole)
813 return ERR_PTR(-ENOMEM);
814
815 /* prepare to claim, if successful, mark claiming in progress */
816 spin_lock(&bdev_lock);
817
818 err = bd_prepare_to_claim(bdev, whole, holder);
819 if (err == 0) {
820 whole->bd_claiming = holder;
821 spin_unlock(&bdev_lock);
822 return whole;
823 } else {
824 spin_unlock(&bdev_lock);
825 bdput(whole);
826 return ERR_PTR(err);
827 }
828}
829
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800830#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100831struct bd_holder_disk {
832 struct list_head list;
833 struct gendisk *disk;
834 int refcnt;
835};
836
837static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
838 struct gendisk *disk)
839{
840 struct bd_holder_disk *holder;
841
842 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
843 if (holder->disk == disk)
844 return holder;
845 return NULL;
846}
847
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700848static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800849{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700850 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800851}
852
853static void del_symlink(struct kobject *from, struct kobject *to)
854{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800855 sysfs_remove_link(from, kobject_name(to));
856}
857
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800858/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100859 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
860 * @bdev: the claimed slave bdev
861 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500862 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100863 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
864 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100865 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500866 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100867 * - from "slaves" directory of the holder @disk to the claimed @bdev
868 * - from "holders" directory of the @bdev to the holder @disk
869 *
870 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
871 * passed to bd_link_disk_holder(), then:
872 *
873 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
874 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
875 *
876 * The caller must have claimed @bdev before calling this function and
877 * ensure that both @bdev and @disk are valid during the creation and
878 * lifetime of these symlinks.
879 *
880 * CONTEXT:
881 * Might sleep.
882 *
883 * RETURNS:
884 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500885 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100886int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500887{
Tejun Heo49731ba2011-01-14 18:43:57 +0100888 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100889 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800890
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800891 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500892
Tejun Heo49731ba2011-01-14 18:43:57 +0100893 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700894
Tejun Heoe09b4572010-11-13 11:55:17 +0100895 /* FIXME: remove the following once add_disk() handles errors */
896 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
897 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700898
Tejun Heo49731ba2011-01-14 18:43:57 +0100899 holder = bd_find_holder_disk(bdev, disk);
900 if (holder) {
901 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100902 goto out_unlock;
903 }
904
Tejun Heo49731ba2011-01-14 18:43:57 +0100905 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
906 if (!holder) {
907 ret = -ENOMEM;
908 goto out_unlock;
909 }
910
911 INIT_LIST_HEAD(&holder->list);
912 holder->disk = disk;
913 holder->refcnt = 1;
914
915 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
916 if (ret)
917 goto out_free;
918
919 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
920 if (ret)
921 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100922 /*
923 * bdev could be deleted beneath us which would implicitly destroy
924 * the holder directory. Hold on to it.
925 */
926 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100927
928 list_add(&holder->list, &bdev->bd_holder_disks);
929 goto out_unlock;
930
931out_del:
932 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
933out_free:
934 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100935out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800936 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100937 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800938}
Tejun Heoe09b4572010-11-13 11:55:17 +0100939EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800940
Tejun Heo49731ba2011-01-14 18:43:57 +0100941/**
942 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
943 * @bdev: the calimed slave bdev
944 * @disk: the holding disk
945 *
946 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
947 *
948 * CONTEXT:
949 * Might sleep.
950 */
951void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800952{
Tejun Heo49731ba2011-01-14 18:43:57 +0100953 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100954
Tejun Heo49731ba2011-01-14 18:43:57 +0100955 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800956
Tejun Heo49731ba2011-01-14 18:43:57 +0100957 holder = bd_find_holder_disk(bdev, disk);
958
959 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
960 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
961 del_symlink(bdev->bd_part->holder_dir,
962 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +0100963 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100964 list_del_init(&holder->list);
965 kfree(holder);
966 }
967
968 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800969}
Tejun Heo49731ba2011-01-14 18:43:57 +0100970EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800971#endif
972
Andrew Patterson0c002c22008-09-04 14:27:20 -0600973/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600974 * flush_disk - invalidates all buffer-cache entries on a disk
975 *
976 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -0800977 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -0600978 *
979 * Invalidates all buffer-cache entries on a disk. It should be called
980 * when a disk has been changed -- either by a media change or online
981 * resize.
982 */
NeilBrown93b270f2011-02-24 17:25:47 +1100983static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -0600984{
NeilBrown93b270f2011-02-24 17:25:47 +1100985 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -0600986 char name[BDEVNAME_SIZE] = "";
987
988 if (bdev->bd_disk)
989 disk_name(bdev->bd_disk, 0, name);
990 printk(KERN_WARNING "VFS: busy inodes on changed media or "
991 "resized disk %s\n", name);
992 }
993
994 if (!bdev->bd_disk)
995 return;
Tejun Heod27769e2011-08-23 20:01:04 +0200996 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -0600997 bdev->bd_invalidated = 1;
998}
999
1000/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001001 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001002 * @disk: struct gendisk to check
1003 * @bdev: struct bdev to adjust.
1004 *
1005 * This routine checks to see if the bdev size does not match the disk size
1006 * and adjusts it if it differs.
1007 */
1008void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1009{
1010 loff_t disk_size, bdev_size;
1011
1012 disk_size = (loff_t)get_capacity(disk) << 9;
1013 bdev_size = i_size_read(bdev->bd_inode);
1014 if (disk_size != bdev_size) {
1015 char name[BDEVNAME_SIZE];
1016
1017 disk_name(disk, 0, name);
1018 printk(KERN_INFO
1019 "%s: detected capacity change from %lld to %lld\n",
1020 name, bdev_size, disk_size);
1021 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001022 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001023 }
1024}
1025EXPORT_SYMBOL(check_disk_size_change);
1026
1027/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001028 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001029 * @disk: struct gendisk to be revalidated
1030 *
1031 * This routine is a wrapper for lower-level driver's revalidate_disk
1032 * call-backs. It is used to do common pre and post operations needed
1033 * for all revalidate_disk operations.
1034 */
1035int revalidate_disk(struct gendisk *disk)
1036{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001037 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001038 int ret = 0;
1039
1040 if (disk->fops->revalidate_disk)
1041 ret = disk->fops->revalidate_disk(disk);
1042
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001043 bdev = bdget_disk(disk, 0);
1044 if (!bdev)
1045 return ret;
1046
1047 mutex_lock(&bdev->bd_mutex);
1048 check_disk_size_change(disk, bdev);
1049 mutex_unlock(&bdev->bd_mutex);
1050 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001051 return ret;
1052}
1053EXPORT_SYMBOL(revalidate_disk);
1054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055/*
1056 * This routine checks whether a removable media has been changed,
1057 * and invalidates all buffer-cache-entries in that case. This
1058 * is a relatively slow routine, so we have to try to minimize using
1059 * it. Thus it is called only upon a 'mount' or 'open'. This
1060 * is the best way of combining speed and utility, I think.
1061 * People changing diskettes in the middle of an operation deserve
1062 * to lose :-)
1063 */
1064int check_disk_change(struct block_device *bdev)
1065{
1066 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001067 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001068 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069
Tejun Heo77ea8872010-12-08 20:57:37 +01001070 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1071 DISK_EVENT_EJECT_REQUEST);
1072 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 return 0;
1074
NeilBrown93b270f2011-02-24 17:25:47 +11001075 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (bdops->revalidate_disk)
1077 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 return 1;
1079}
1080
1081EXPORT_SYMBOL(check_disk_change);
1082
1083void bd_set_size(struct block_device *bdev, loff_t size)
1084{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001085 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 bdev->bd_inode->i_size = size;
1088 while (bsize < PAGE_CACHE_SIZE) {
1089 if (size & bsize)
1090 break;
1091 bsize <<= 1;
1092 }
1093 bdev->bd_block_size = bsize;
1094 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1095}
1096EXPORT_SYMBOL(bd_set_size);
1097
Al Viro9a1c3542008-02-22 20:40:24 -05001098static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001099
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001100/*
1101 * bd_mutex locking:
1102 *
1103 * mutex_lock(part->bd_mutex)
1104 * mutex_lock_nested(whole->bd_mutex, 1)
1105 */
1106
Al Viro572c4892007-10-08 13:24:05 -04001107static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001110 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001111 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001112 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001113 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Al Viro572c4892007-10-08 13:24:05 -04001115 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001116 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001117 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001118 perm |= MAY_WRITE;
1119 /*
1120 * hooks: /n/, see "layering violations".
1121 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001122 if (!for_part) {
1123 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1124 if (ret != 0) {
1125 bdput(bdev);
1126 return ret;
1127 }
Al Viro82666022008-08-01 05:32:04 -04001128 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001129
NeilBrownd3374822009-01-09 08:31:10 +11001130 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001131
Tejun Heo89f97492008-11-05 10:21:06 +01001132 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001133 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001134 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001135 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001136 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137
Tejun Heo69e02c52011-03-09 19:54:27 +01001138 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001139 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (!bdev->bd_openers) {
1141 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001142 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001144 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001146
1147 ret = -ENXIO;
1148 bdev->bd_part = disk_get_part(disk, partno);
1149 if (!bdev->bd_part)
1150 goto out_clear;
1151
Tejun Heo1196f8b2011-04-21 20:54:45 +02001152 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001154 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001155 if (ret == -ERESTARTSYS) {
1156 /* Lost a race with 'disk' being
1157 * deleted, try again.
1158 * See md.c
1159 */
1160 disk_put_part(bdev->bd_part);
1161 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001162 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001163 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001164 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001165 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001166 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001167 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001168 goto restart;
1169 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 }
Tejun Heo7e697232011-05-23 13:26:07 +02001171
1172 if (!ret && !bdev->bd_openers) {
1173 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1174 bdi = blk_get_backing_dev_info(bdev);
1175 if (bdi == NULL)
1176 bdi = &default_backing_dev_info;
1177 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1178 }
1179
Tejun Heo1196f8b2011-04-21 20:54:45 +02001180 /*
1181 * If the device is invalidated, rescan partition
1182 * if open succeeded or failed with -ENOMEDIUM.
1183 * The latter is necessary to prevent ghost
1184 * partitions on a removed medium.
1185 */
1186 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
1187 rescan_partitions(disk, bdev);
1188 if (ret)
1189 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 struct block_device *whole;
1192 whole = bdget_disk(disk, 0);
1193 ret = -ENOMEM;
1194 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001195 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001196 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001197 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001199 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001201 bdev_inode_switch_bdi(bdev->bd_inode,
1202 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001203 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001204 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001205 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001207 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
Tejun Heo89f97492008-11-05 10:21:06 +01001209 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 }
1211 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001213 ret = 0;
1214 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001215 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001216 /* the same as first opener case, read comment there */
1217 if (bdev->bd_invalidated && (!ret || ret == -ENOMEDIUM))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 rescan_partitions(bdev->bd_disk, bdev);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001219 if (ret)
1220 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001222 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001223 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001224 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 }
1226 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001227 if (for_part)
1228 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001229 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001230 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 return 0;
1232
Tejun Heo0762b8b2008-08-25 19:56:12 +09001233 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001234 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001236 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001237 bdev->bd_queue = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001238 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001240 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001242 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001243 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001244 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001245 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001246 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001247 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001248 bdput(bdev);
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 return ret;
1251}
1252
Tejun Heod4d77622010-11-13 11:55:18 +01001253/**
1254 * blkdev_get - open a block device
1255 * @bdev: block_device to open
1256 * @mode: FMODE_* mask
1257 * @holder: exclusive holder identifier
1258 *
1259 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1260 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1261 * @holder is invalid. Exclusive opens may nest for the same @holder.
1262 *
1263 * On success, the reference count of @bdev is unchanged. On failure,
1264 * @bdev is put.
1265 *
1266 * CONTEXT:
1267 * Might sleep.
1268 *
1269 * RETURNS:
1270 * 0 on success, -errno on failure.
1271 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001272int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
Tejun Heoe525fd82010-11-13 11:55:17 +01001274 struct block_device *whole = NULL;
1275 int res;
1276
1277 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1278
1279 if ((mode & FMODE_EXCL) && holder) {
1280 whole = bd_start_claiming(bdev, holder);
1281 if (IS_ERR(whole)) {
1282 bdput(bdev);
1283 return PTR_ERR(whole);
1284 }
1285 }
1286
1287 res = __blkdev_get(bdev, mode, 0);
1288
1289 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001290 struct gendisk *disk = whole->bd_disk;
1291
Tejun Heo6a027ef2010-11-13 11:55:17 +01001292 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001293 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001294 spin_lock(&bdev_lock);
1295
Tejun Heo77ea8872010-12-08 20:57:37 +01001296 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001297 BUG_ON(!bd_may_claim(bdev, whole, holder));
1298 /*
1299 * Note that for a whole device bd_holders
1300 * will be incremented twice, and bd_holder
1301 * will be set to bd_may_claim before being
1302 * set to holder
1303 */
1304 whole->bd_holders++;
1305 whole->bd_holder = bd_may_claim;
1306 bdev->bd_holders++;
1307 bdev->bd_holder = holder;
1308 }
1309
1310 /* tell others that we're done */
1311 BUG_ON(whole->bd_claiming != holder);
1312 whole->bd_claiming = NULL;
1313 wake_up_bit(&whole->bd_claiming, 0);
1314
1315 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001316
1317 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001318 * Block event polling for write claims if requested. Any
1319 * write holder makes the write_holder state stick until
1320 * all are released. This is good enough and tracking
1321 * individual writeable reference is too fragile given the
1322 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001323 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001324 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1325 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001326 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001327 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001328 }
1329
1330 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001331 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001332 }
1333
1334 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001335}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336EXPORT_SYMBOL(blkdev_get);
1337
Tejun Heod4d77622010-11-13 11:55:18 +01001338/**
1339 * blkdev_get_by_path - open a block device by name
1340 * @path: path to the block device to open
1341 * @mode: FMODE_* mask
1342 * @holder: exclusive holder identifier
1343 *
1344 * Open the blockdevice described by the device file at @path. @mode
1345 * and @holder are identical to blkdev_get().
1346 *
1347 * On success, the returned block_device has reference count of one.
1348 *
1349 * CONTEXT:
1350 * Might sleep.
1351 *
1352 * RETURNS:
1353 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1354 */
1355struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1356 void *holder)
1357{
1358 struct block_device *bdev;
1359 int err;
1360
1361 bdev = lookup_bdev(path);
1362 if (IS_ERR(bdev))
1363 return bdev;
1364
1365 err = blkdev_get(bdev, mode, holder);
1366 if (err)
1367 return ERR_PTR(err);
1368
Chuck Ebberte51900f2011-02-16 18:11:53 -05001369 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1370 blkdev_put(bdev, mode);
1371 return ERR_PTR(-EACCES);
1372 }
1373
Tejun Heod4d77622010-11-13 11:55:18 +01001374 return bdev;
1375}
1376EXPORT_SYMBOL(blkdev_get_by_path);
1377
1378/**
1379 * blkdev_get_by_dev - open a block device by device number
1380 * @dev: device number of block device to open
1381 * @mode: FMODE_* mask
1382 * @holder: exclusive holder identifier
1383 *
1384 * Open the blockdevice described by device number @dev. @mode and
1385 * @holder are identical to blkdev_get().
1386 *
1387 * Use it ONLY if you really do not have anything better - i.e. when
1388 * you are behind a truly sucky interface and all you are given is a
1389 * device number. _Never_ to be used for internal purposes. If you
1390 * ever need it - reconsider your API.
1391 *
1392 * On success, the returned block_device has reference count of one.
1393 *
1394 * CONTEXT:
1395 * Might sleep.
1396 *
1397 * RETURNS:
1398 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1399 */
1400struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1401{
1402 struct block_device *bdev;
1403 int err;
1404
1405 bdev = bdget(dev);
1406 if (!bdev)
1407 return ERR_PTR(-ENOMEM);
1408
1409 err = blkdev_get(bdev, mode, holder);
1410 if (err)
1411 return ERR_PTR(err);
1412
1413 return bdev;
1414}
1415EXPORT_SYMBOL(blkdev_get_by_dev);
1416
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417static int blkdev_open(struct inode * inode, struct file * filp)
1418{
1419 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
1421 /*
1422 * Preserve backwards compatibility and allow large file access
1423 * even if userspace doesn't ask for it explicitly. Some mkfs
1424 * binary needs it. We might want to drop this workaround
1425 * during an unstable branch.
1426 */
1427 filp->f_flags |= O_LARGEFILE;
1428
Al Viro572c4892007-10-08 13:24:05 -04001429 if (filp->f_flags & O_NDELAY)
1430 filp->f_mode |= FMODE_NDELAY;
1431 if (filp->f_flags & O_EXCL)
1432 filp->f_mode |= FMODE_EXCL;
1433 if ((filp->f_flags & O_ACCMODE) == 3)
1434 filp->f_mode |= FMODE_WRITE_IOCTL;
1435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001437 if (bdev == NULL)
1438 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439
Al Viro572c4892007-10-08 13:24:05 -04001440 filp->f_mapping = bdev->bd_inode->i_mapping;
1441
Tejun Heoe525fd82010-11-13 11:55:17 +01001442 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443}
1444
Al Viro9a1c3542008-02-22 20:40:24 -05001445static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001446{
1447 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001448 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001449 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001450
NeilBrown6796bf52006-12-08 02:36:16 -08001451 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001452 if (for_part)
1453 bdev->bd_part_count--;
1454
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001455 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001456 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001457 sync_blockdev(bdev);
1458 kill_bdev(bdev);
NeilBrown94007752011-09-10 17:20:21 +10001459 /* ->release can cause the old bdi to disappear,
1460 * so must switch it out first
1461 */
1462 bdev_inode_switch_bdi(bdev->bd_inode,
1463 &default_backing_dev_info);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001464 }
1465 if (bdev->bd_contains == bdev) {
1466 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001467 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001468 }
1469 if (!bdev->bd_openers) {
1470 struct module *owner = disk->fops->owner;
1471
Tejun Heo0762b8b2008-08-25 19:56:12 +09001472 disk_put_part(bdev->bd_part);
1473 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001474 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001475 if (bdev != bdev->bd_contains)
1476 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001477 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001478
1479 put_disk(disk);
1480 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001481 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001482 mutex_unlock(&bdev->bd_mutex);
1483 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001484 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001485 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001486 return ret;
1487}
1488
Al Viro9a1c3542008-02-22 20:40:24 -05001489int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001490{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001491 mutex_lock(&bdev->bd_mutex);
1492
Tejun Heoe525fd82010-11-13 11:55:17 +01001493 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001494 bool bdev_free;
1495
1496 /*
1497 * Release a claim on the device. The holder fields
1498 * are protected with bdev_lock. bd_mutex is to
1499 * synchronize disk_holder unlinking.
1500 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001501 spin_lock(&bdev_lock);
1502
1503 WARN_ON_ONCE(--bdev->bd_holders < 0);
1504 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1505
1506 /* bd_contains might point to self, check in a separate step */
1507 if ((bdev_free = !bdev->bd_holders))
1508 bdev->bd_holder = NULL;
1509 if (!bdev->bd_contains->bd_holders)
1510 bdev->bd_contains->bd_holder = NULL;
1511
1512 spin_unlock(&bdev_lock);
1513
Tejun Heo77ea8872010-12-08 20:57:37 +01001514 /*
1515 * If this was the last claim, remove holder link and
1516 * unblock evpoll if it was a write holder.
1517 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001518 if (bdev_free && bdev->bd_write_holder) {
1519 disk_unblock_events(bdev->bd_disk);
1520 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001521 }
Tejun Heo69362172011-03-09 19:54:27 +01001522 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001523
Tejun Heo85ef06d2011-07-01 16:17:47 +02001524 /*
1525 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1526 * event. This is to ensure detection of media removal commanded
1527 * from userland - e.g. eject(1).
1528 */
1529 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1530
1531 mutex_unlock(&bdev->bd_mutex);
1532
Al Viro9a1c3542008-02-22 20:40:24 -05001533 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001534}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535EXPORT_SYMBOL(blkdev_put);
1536
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537static int blkdev_close(struct inode * inode, struct file * filp)
1538{
1539 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Tejun Heoe525fd82010-11-13 11:55:17 +01001540
Al Viro9a1c3542008-02-22 20:40:24 -05001541 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
1543
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001544static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545{
Al Viro56b26ad2008-09-19 03:17:36 -04001546 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1547 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001548
1549 /*
1550 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1551 * to updated it before every ioctl.
1552 */
Al Viro56b26ad2008-09-19 03:17:36 -04001553 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001554 mode |= FMODE_NDELAY;
1555 else
1556 mode &= ~FMODE_NDELAY;
1557
Al Viro56b26ad2008-09-19 03:17:36 -04001558 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559}
1560
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001561/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001562 * Write data to the block device. Only intended for the block device itself
1563 * and the raw driver which basically is a fake block device.
1564 *
1565 * Does not take i_mutex for the write and thus is not for general purpose
1566 * use.
1567 */
1568ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1569 unsigned long nr_segs, loff_t pos)
1570{
1571 struct file *file = iocb->ki_filp;
1572 ssize_t ret;
1573
1574 BUG_ON(iocb->ki_pos != pos);
1575
1576 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1577 if (ret > 0 || ret == -EIOCBQUEUED) {
1578 ssize_t err;
1579
1580 err = generic_write_sync(file, pos, ret);
1581 if (err < 0 && ret > 0)
1582 ret = err;
1583 }
1584 return ret;
1585}
1586EXPORT_SYMBOL_GPL(blkdev_aio_write);
1587
1588/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001589 * Try to release a page associated with block device when the system
1590 * is under memory pressure.
1591 */
1592static int blkdev_releasepage(struct page *page, gfp_t wait)
1593{
1594 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1595
1596 if (super && super->s_op->bdev_try_to_free_page)
1597 return super->s_op->bdev_try_to_free_page(super, page, wait);
1598
1599 return try_to_free_buffers(page);
1600}
1601
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001602static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 .readpage = blkdev_readpage,
1604 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001605 .write_begin = blkdev_write_begin,
1606 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001608 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 .direct_IO = blkdev_direct_IO,
1610};
1611
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001612const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 .open = blkdev_open,
1614 .release = blkdev_close,
1615 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001616 .read = do_sync_read,
1617 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001619 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001621 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001622 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623#ifdef CONFIG_COMPAT
1624 .compat_ioctl = compat_blkdev_ioctl,
1625#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001626 .splice_read = generic_file_splice_read,
1627 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628};
1629
1630int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1631{
1632 int res;
1633 mm_segment_t old_fs = get_fs();
1634 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001635 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 set_fs(old_fs);
1637 return res;
1638}
1639
1640EXPORT_SYMBOL(ioctl_by_bdev);
1641
1642/**
1643 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001644 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001646 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 * namespace if possible and return it. Return ERR_PTR(error)
1648 * otherwise.
1649 */
Al Viro421748e2008-08-02 01:04:36 -04001650struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651{
1652 struct block_device *bdev;
1653 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001654 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 int error;
1656
Al Viro421748e2008-08-02 01:04:36 -04001657 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return ERR_PTR(-EINVAL);
1659
Al Viro421748e2008-08-02 01:04:36 -04001660 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 if (error)
1662 return ERR_PTR(error);
1663
Al Viro421748e2008-08-02 01:04:36 -04001664 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 error = -ENOTBLK;
1666 if (!S_ISBLK(inode->i_mode))
1667 goto fail;
1668 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001669 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 goto fail;
1671 error = -ENOMEM;
1672 bdev = bd_acquire(inode);
1673 if (!bdev)
1674 goto fail;
1675out:
Al Viro421748e2008-08-02 01:04:36 -04001676 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 return bdev;
1678fail:
1679 bdev = ERR_PTR(error);
1680 goto out;
1681}
Al Virod5686b42008-08-01 05:00:11 -04001682EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
NeilBrown93b270f2011-02-24 17:25:47 +11001684int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001685{
1686 struct super_block *sb = get_super(bdev);
1687 int res = 0;
1688
1689 if (sb) {
1690 /*
1691 * no need to lock the super, get_super holds the
1692 * read mutex so the filesystem cannot go away
1693 * under us (->put_super runs with the write lock
1694 * hold).
1695 */
1696 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001697 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001698 drop_super(sb);
1699 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001700 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001701 return res;
1702}
1703EXPORT_SYMBOL(__invalidate_device);