blob: 38e721b35d45388cb0febed8021a02277a4a2f1b [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>
Muthu Kumarb502bd12012-03-23 15:01:50 -070019#include <linux/magic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/buffer_head.h>
Al Viroff01bb482011-09-16 02:31:11 -040021#include <linux/swap.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010022#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010023#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/mpage.h>
25#include <linux/mount.h>
26#include <linux/uio.h>
27#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070028#include <linux/log2.h>
Al Viroff01bb482011-09-16 02:31:11 -040029#include <linux/cleancache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020031#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33struct bdev_inode {
34 struct block_device bdev;
35 struct inode vfs_inode;
36};
37
Adrian Bunk4c54ac62008-02-18 13:48:31 +010038static const struct address_space_operations def_blk_aops;
39
Linus Torvalds1da177e2005-04-16 15:20:36 -070040static inline struct bdev_inode *BDEV_I(struct inode *inode)
41{
42 return container_of(inode, struct bdev_inode, vfs_inode);
43}
44
45inline struct block_device *I_BDEV(struct inode *inode)
46{
47 return &BDEV_I(inode)->bdev;
48}
Linus Torvalds1da177e2005-04-16 15:20:36 -070049EXPORT_SYMBOL(I_BDEV);
50
Dave Chinnera5491e02010-10-21 11:49:26 +110051/*
Christoph Hellwigf758eea2011-04-21 18:19:44 -060052 * Move the inode from its current bdi to a new bdi. If the inode is dirty we
53 * need to move it onto the dirty list of @dst so that the inode is always on
54 * the right list.
Dave Chinnera5491e02010-10-21 11:49:26 +110055 */
56static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
58{
Christoph Hellwigf758eea2011-04-21 18:19:44 -060059 struct backing_dev_info *old = inode->i_data.backing_dev_info;
60
61 if (unlikely(dst == old)) /* deadlock avoidance */
62 return;
63 bdi_lock_two(&old->wb, &dst->wb);
Dave Chinner250df6e2011-03-22 22:23:36 +110064 spin_lock(&inode->i_lock);
Dave Chinnera5491e02010-10-21 11:49:26 +110065 inode->i_data.backing_dev_info = dst;
66 if (inode->i_state & I_DIRTY)
Nick Piggin7ccf19a2010-10-21 11:49:30 +110067 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Dave Chinner250df6e2011-03-22 22:23:36 +110068 spin_unlock(&inode->i_lock);
Christoph Hellwigf758eea2011-04-21 18:19:44 -060069 spin_unlock(&old->wb.list_lock);
70 spin_unlock(&dst->wb.list_lock);
Dave Chinnera5491e02010-10-21 11:49:26 +110071}
72
Jeff Moyer080399a2012-05-11 16:34:10 +020073sector_t blkdev_max_block(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
75 sector_t retval = ~((sector_t)0);
76 loff_t sz = i_size_read(bdev->bd_inode);
77
78 if (sz) {
79 unsigned int size = block_size(bdev);
80 unsigned int sizebits = blksize_bits(size);
81 retval = (sz >> sizebits);
82 }
83 return retval;
84}
85
Peter Zijlstraf9a14392007-05-06 14:49:55 -070086/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb482011-09-16 02:31:11 -040087void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Al Viroff01bb482011-09-16 02:31:11 -040089 struct address_space *mapping = bdev->bd_inode->i_mapping;
90
91 if (mapping->nrpages == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070092 return;
Al Viroff01bb482011-09-16 02:31:11 -040093
Peter Zijlstraf9a14392007-05-06 14:49:55 -070094 invalidate_bh_lrus();
Al Viroff01bb482011-09-16 02:31:11 -040095 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
Al Viroff01bb482011-09-16 02:31:11 -040097EXPORT_SYMBOL(kill_bdev);
98
99/* Invalidate clean unused buffers and pagecache. */
100void invalidate_bdev(struct block_device *bdev)
101{
102 struct address_space *mapping = bdev->bd_inode->i_mapping;
103
104 if (mapping->nrpages == 0)
105 return;
106
107 invalidate_bh_lrus();
108 lru_add_drain_all(); /* make sure all lru add caches are flushed */
109 invalidate_mapping_pages(mapping, 0, -1);
110 /* 99% of the time, we don't need to flush the cleancache on the bdev.
111 * But, for the strange corners, lets be cautious
112 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400113 cleancache_invalidate_inode(mapping);
Al Viroff01bb482011-09-16 02:31:11 -0400114}
115EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117int set_blocksize(struct block_device *bdev, int size)
118{
119 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700120 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 return -EINVAL;
122
123 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400124 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 return -EINVAL;
126
127 /* Don't change the size if it is same as current */
128 if (bdev->bd_block_size != size) {
129 sync_blockdev(bdev);
130 bdev->bd_block_size = size;
131 bdev->bd_inode->i_blkbits = blksize_bits(size);
132 kill_bdev(bdev);
133 }
134 return 0;
135}
136
137EXPORT_SYMBOL(set_blocksize);
138
139int sb_set_blocksize(struct super_block *sb, int size)
140{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 if (set_blocksize(sb->s_bdev, size))
142 return 0;
143 /* If we get here, we know size is power of two
144 * and it's value is between 512 and PAGE_SIZE */
145 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800146 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 return sb->s_blocksize;
148}
149
150EXPORT_SYMBOL(sb_set_blocksize);
151
152int sb_min_blocksize(struct super_block *sb, int size)
153{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400154 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 if (size < minsize)
156 size = minsize;
157 return sb_set_blocksize(sb, size);
158}
159
160EXPORT_SYMBOL(sb_min_blocksize);
161
162static int
163blkdev_get_block(struct inode *inode, sector_t iblock,
164 struct buffer_head *bh, int create)
165{
Jeff Moyer080399a2012-05-11 16:34:10 +0200166 if (iblock >= blkdev_max_block(I_BDEV(inode))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 if (create)
168 return -EIO;
169
170 /*
171 * for reads, we're just trying to fill a partial page.
172 * return a hole, they will have to call get_block again
173 * before they can fill it, and they will get -EIO at that
174 * time
175 */
176 return 0;
177 }
178 bh->b_bdev = I_BDEV(inode);
179 bh->b_blocknr = iblock;
180 set_buffer_mapped(bh);
181 return 0;
182}
183
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800184static int
185blkdev_get_blocks(struct inode *inode, sector_t iblock,
186 struct buffer_head *bh, int create)
187{
Jeff Moyer080399a2012-05-11 16:34:10 +0200188 sector_t end_block = blkdev_max_block(I_BDEV(inode));
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800189 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
190
191 if ((iblock + max_blocks) > end_block) {
192 max_blocks = end_block - iblock;
193 if ((long)max_blocks <= 0) {
194 if (create)
195 return -EIO; /* write fully beyond EOF */
196 /*
197 * It is a read which is fully beyond EOF. We return
198 * a !buffer_mapped buffer
199 */
200 max_blocks = 0;
201 }
202 }
203
204 bh->b_bdev = I_BDEV(inode);
205 bh->b_blocknr = iblock;
206 bh->b_size = max_blocks << inode->i_blkbits;
207 if (max_blocks)
208 set_buffer_mapped(bh);
209 return 0;
210}
211
212static ssize_t
213blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
214 loff_t offset, unsigned long nr_segs)
215{
216 struct file *file = iocb->ki_filp;
217 struct inode *inode = file->f_mapping->host;
218
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200219 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
220 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800221}
222
Jan Kara5cee5812009-04-27 16:43:51 +0200223int __sync_blockdev(struct block_device *bdev, int wait)
224{
225 if (!bdev)
226 return 0;
227 if (!wait)
228 return filemap_flush(bdev->bd_inode->i_mapping);
229 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
230}
231
Nick Piggin585d3bc2009-02-25 10:44:19 +0100232/*
233 * Write out and wait upon all the dirty data associated with a block
234 * device via its mapping. Does not take the superblock lock.
235 */
236int sync_blockdev(struct block_device *bdev)
237{
Jan Kara5cee5812009-04-27 16:43:51 +0200238 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100239}
240EXPORT_SYMBOL(sync_blockdev);
241
242/*
243 * Write out and wait upon all dirty data associated with this
244 * device. Filesystem data as well as the underlying block
245 * device. Takes the superblock lock.
246 */
247int fsync_bdev(struct block_device *bdev)
248{
249 struct super_block *sb = get_super(bdev);
250 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200251 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100252 drop_super(sb);
253 return res;
254 }
255 return sync_blockdev(bdev);
256}
Al Viro47e44912009-04-01 07:07:16 -0400257EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100258
259/**
260 * freeze_bdev -- lock a filesystem and force it into a consistent state
261 * @bdev: blockdevice to lock
262 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100263 * If a superblock is found on this device, we take the s_umount semaphore
264 * on it to make sure nobody unmounts until the snapshot creation is done.
265 * The reference counter (bd_fsfreeze_count) guarantees that only the last
266 * unfreeze process can unfreeze the frozen filesystem actually when multiple
267 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
268 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
269 * actually.
270 */
271struct super_block *freeze_bdev(struct block_device *bdev)
272{
273 struct super_block *sb;
274 int error = 0;
275
276 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200277 if (++bdev->bd_fsfreeze_count > 1) {
278 /*
279 * We don't even need to grab a reference - the first call
280 * to freeze_bdev grab an active reference and only the last
281 * thaw_bdev drops it.
282 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100283 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200284 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100285 mutex_unlock(&bdev->bd_fsfreeze_mutex);
286 return sb;
287 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100288
Christoph Hellwig45042302009-08-03 23:28:35 +0200289 sb = get_active_super(bdev);
290 if (!sb)
291 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400292 error = freeze_super(sb);
293 if (error) {
294 deactivate_super(sb);
295 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200296 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400297 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100298 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400299 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200300 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100301 sync_blockdev(bdev);
302 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200303 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100304}
305EXPORT_SYMBOL(freeze_bdev);
306
307/**
308 * thaw_bdev -- unlock filesystem
309 * @bdev: blockdevice to unlock
310 * @sb: associated superblock
311 *
312 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
313 */
314int thaw_bdev(struct block_device *bdev, struct super_block *sb)
315{
Christoph Hellwig45042302009-08-03 23:28:35 +0200316 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100317
318 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200319 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400320 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100321
Christoph Hellwig45042302009-08-03 23:28:35 +0200322 error = 0;
323 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400324 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100325
Christoph Hellwig45042302009-08-03 23:28:35 +0200326 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400327 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200328
Josef Bacik18e9e512010-03-23 10:34:56 -0400329 error = thaw_super(sb);
330 if (error) {
331 bdev->bd_fsfreeze_count++;
332 mutex_unlock(&bdev->bd_fsfreeze_mutex);
333 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100334 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400335out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100336 mutex_unlock(&bdev->bd_fsfreeze_mutex);
337 return 0;
338}
339EXPORT_SYMBOL(thaw_bdev);
340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
342{
343 return block_write_full_page(page, blkdev_get_block, wbc);
344}
345
346static int blkdev_readpage(struct file * file, struct page * page)
347{
348 return block_read_full_page(page, blkdev_get_block);
349}
350
Nick Piggin6272b5a2007-10-16 01:25:04 -0700351static int blkdev_write_begin(struct file *file, struct address_space *mapping,
352 loff_t pos, unsigned len, unsigned flags,
353 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200355 return block_write_begin(mapping, pos, len, flags, pagep,
356 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357}
358
Nick Piggin6272b5a2007-10-16 01:25:04 -0700359static int blkdev_write_end(struct file *file, struct address_space *mapping,
360 loff_t pos, unsigned len, unsigned copied,
361 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700363 int ret;
364 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
365
366 unlock_page(page);
367 page_cache_release(page);
368
369 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
372/*
373 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800374 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 * so we compute the size by hand (just as in block_read/write above)
376 */
377static loff_t block_llseek(struct file *file, loff_t offset, int origin)
378{
379 struct inode *bd_inode = file->f_mapping->host;
380 loff_t size;
381 loff_t retval;
382
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800383 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 size = i_size_read(bd_inode);
385
Josef Bacik06222e42011-07-18 13:21:38 -0400386 retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 switch (origin) {
Josef Bacik06222e42011-07-18 13:21:38 -0400388 case SEEK_END:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 offset += size;
390 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400391 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400393 case SEEK_SET:
394 break;
395 default:
396 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 if (offset >= 0 && offset <= size) {
399 if (offset != file->f_pos) {
400 file->f_pos = offset;
401 }
402 retval = offset;
403 }
Josef Bacik06222e42011-07-18 13:21:38 -0400404out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800405 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 return retval;
407}
408
Josef Bacik02c24a82011-07-16 20:44:56 -0400409int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400411 struct inode *bd_inode = filp->f_mapping->host;
412 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100413 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200414
415 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
416 if (error)
417 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100418
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400419 /*
420 * There is no need to serialise calls to blkdev_issue_flush with
421 * i_mutex and doing so causes performance issues with concurrent
422 * O_SYNC writers to a block device.
423 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200424 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100425 if (error == -EOPNOTSUPP)
426 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400427
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100428 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700430EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
432/*
433 * pseudo-fs
434 */
435
436static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800437static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438
439static struct inode *bdev_alloc_inode(struct super_block *sb)
440{
Christoph Lametere94b1762006-12-06 20:33:17 -0800441 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 if (!ei)
443 return NULL;
444 return &ei->vfs_inode;
445}
446
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100447static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100449 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 struct bdev_inode *bdi = BDEV_I(inode);
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 kmem_cache_free(bdev_cachep, bdi);
453}
454
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100455static void bdev_destroy_inode(struct inode *inode)
456{
457 call_rcu(&inode->i_rcu, bdev_i_callback);
458}
459
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700460static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
462 struct bdev_inode *ei = (struct bdev_inode *) foo;
463 struct block_device *bdev = &ei->bdev;
464
Christoph Lametera35afb82007-05-16 22:10:57 -0700465 memset(bdev, 0, sizeof(*bdev));
466 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700467 INIT_LIST_HEAD(&bdev->bd_inodes);
468 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100469#ifdef CONFIG_SYSFS
470 INIT_LIST_HEAD(&bdev->bd_holder_disks);
471#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700472 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800473 /* Initialize mutex for freeze. */
474 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477static inline void __bd_forget(struct inode *inode)
478{
479 list_del_init(&inode->i_devices);
480 inode->i_bdev = NULL;
481 inode->i_mapping = &inode->i_data;
482}
483
Al Virob57922d2010-06-07 14:34:48 -0400484static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
486 struct block_device *bdev = &BDEV_I(inode)->bdev;
487 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400488 truncate_inode_pages(&inode->i_data, 0);
489 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200490 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 spin_lock(&bdev_lock);
492 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
493 __bd_forget(list_entry(p, struct inode, i_devices));
494 }
495 list_del_init(&bdev->bd_list);
496 spin_unlock(&bdev_lock);
497}
498
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800499static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 .statfs = simple_statfs,
501 .alloc_inode = bdev_alloc_inode,
502 .destroy_inode = bdev_destroy_inode,
503 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400504 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505};
506
Al Viro51139ad2010-07-25 23:47:46 +0400507static struct dentry *bd_mount(struct file_system_type *fs_type,
508 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
Muthu Kumarb502bd12012-03-23 15:01:50 -0700510 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511}
512
513static struct file_system_type bd_type = {
514 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400515 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 .kill_sb = kill_anon_super,
517};
518
Al Virof47ec3f2011-11-21 21:15:42 -0500519static struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521void __init bdev_cache_init(void)
522{
523 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300524 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800527 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
528 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900529 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 err = register_filesystem(&bd_type);
531 if (err)
532 panic("Cannot register bdev pseudo-fs");
533 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (IS_ERR(bd_mnt))
535 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300536 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
539/*
540 * Most likely _very_ bad one - but then it's hardly critical for small
541 * /dev and can be fixed when somebody will need really large one.
542 * Keep in mind that it will be fed through icache hash function too.
543 */
544static inline unsigned long hash(dev_t dev)
545{
546 return MAJOR(dev)+MINOR(dev);
547}
548
549static int bdev_test(struct inode *inode, void *data)
550{
551 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
552}
553
554static int bdev_set(struct inode *inode, void *data)
555{
556 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
557 return 0;
558}
559
560static LIST_HEAD(all_bdevs);
561
562struct block_device *bdget(dev_t dev)
563{
564 struct block_device *bdev;
565 struct inode *inode;
566
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800567 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 bdev_test, bdev_set, &dev);
569
570 if (!inode)
571 return NULL;
572
573 bdev = &BDEV_I(inode)->bdev;
574
575 if (inode->i_state & I_NEW) {
576 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000577 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 bdev->bd_inode = inode;
579 bdev->bd_block_size = (1 << inode->i_blkbits);
580 bdev->bd_part_count = 0;
581 bdev->bd_invalidated = 0;
582 inode->i_mode = S_IFBLK;
583 inode->i_rdev = dev;
584 inode->i_bdev = bdev;
585 inode->i_data.a_ops = &def_blk_aops;
586 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
587 inode->i_data.backing_dev_info = &default_backing_dev_info;
588 spin_lock(&bdev_lock);
589 list_add(&bdev->bd_list, &all_bdevs);
590 spin_unlock(&bdev_lock);
591 unlock_new_inode(inode);
592 }
593 return bdev;
594}
595
596EXPORT_SYMBOL(bdget);
597
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200598/**
599 * bdgrab -- Grab a reference to an already referenced block device
600 * @bdev: Block device to grab a reference to.
601 */
602struct block_device *bdgrab(struct block_device *bdev)
603{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400604 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200605 return bdev;
606}
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608long nr_blockdev_pages(void)
609{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700610 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 long ret = 0;
612 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700613 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 ret += bdev->bd_inode->i_mapping->nrpages;
615 }
616 spin_unlock(&bdev_lock);
617 return ret;
618}
619
620void bdput(struct block_device *bdev)
621{
622 iput(bdev->bd_inode);
623}
624
625EXPORT_SYMBOL(bdput);
626
627static struct block_device *bd_acquire(struct inode *inode)
628{
629 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 spin_lock(&bdev_lock);
632 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700633 if (bdev) {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400634 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 spin_unlock(&bdev_lock);
636 return bdev;
637 }
638 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 bdev = bdget(inode->i_rdev);
641 if (bdev) {
642 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700643 if (!inode->i_bdev) {
644 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400645 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700646 * and it's released in clear_inode() of inode.
647 * So, we can access it via ->i_mapping always
648 * without igrab().
649 */
Al Viro7de9c6ee2010-10-23 11:11:40 -0400650 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700651 inode->i_bdev = bdev;
652 inode->i_mapping = bdev->bd_inode->i_mapping;
653 list_add(&inode->i_devices, &bdev->bd_inodes);
654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 spin_unlock(&bdev_lock);
656 }
657 return bdev;
658}
659
Al Virof47ec3f2011-11-21 21:15:42 -0500660static inline int sb_is_blkdev_sb(struct super_block *sb)
661{
662 return sb == blockdev_superblock;
663}
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665/* Call when you free inode */
666
667void bd_forget(struct inode *inode)
668{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700669 struct block_device *bdev = NULL;
670
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700672 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800673 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700674 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700676 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700678
679 if (bdev)
680 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900683/**
684 * bd_may_claim - test whether a block device can be claimed
685 * @bdev: block device of interest
686 * @whole: whole block device containing @bdev, may equal @bdev
687 * @holder: holder trying to claim @bdev
688 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300689 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900690 *
691 * CONTEXT:
692 * spin_lock(&bdev_lock).
693 *
694 * RETURNS:
695 * %true if @bdev can be claimed, %false otherwise.
696 */
697static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
698 void *holder)
699{
700 if (bdev->bd_holder == holder)
701 return true; /* already a holder */
702 else if (bdev->bd_holder != NULL)
703 return false; /* held by someone else */
704 else if (bdev->bd_contains == bdev)
705 return true; /* is a whole device which isn't held */
706
Tejun Heoe525fd82010-11-13 11:55:17 +0100707 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900708 return true; /* is a partition of a device that is being partitioned */
709 else if (whole->bd_holder != NULL)
710 return false; /* is a partition of a held device */
711 else
712 return true; /* is a partition of an un-held device */
713}
714
715/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900716 * bd_prepare_to_claim - prepare to claim a block device
717 * @bdev: block device of interest
718 * @whole: the whole device containing @bdev, may equal @bdev
719 * @holder: holder trying to claim @bdev
720 *
721 * Prepare to claim @bdev. This function fails if @bdev is already
722 * claimed by another holder and waits if another claiming is in
723 * progress. This function doesn't actually claim. On successful
724 * return, the caller has ownership of bd_claiming and bd_holder[s].
725 *
726 * CONTEXT:
727 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
728 * it multiple times.
729 *
730 * RETURNS:
731 * 0 if @bdev can be claimed, -EBUSY otherwise.
732 */
733static int bd_prepare_to_claim(struct block_device *bdev,
734 struct block_device *whole, void *holder)
735{
736retry:
737 /* if someone else claimed, fail */
738 if (!bd_may_claim(bdev, whole, holder))
739 return -EBUSY;
740
Tejun Heoe75aa852010-08-04 17:59:39 +0200741 /* if claiming is already in progress, wait for it to finish */
742 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900743 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
744 DEFINE_WAIT(wait);
745
746 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
747 spin_unlock(&bdev_lock);
748 schedule();
749 finish_wait(wq, &wait);
750 spin_lock(&bdev_lock);
751 goto retry;
752 }
753
754 /* yay, all mine */
755 return 0;
756}
757
758/**
759 * bd_start_claiming - start claiming a block device
760 * @bdev: block device of interest
761 * @holder: holder trying to claim @bdev
762 *
763 * @bdev is about to be opened exclusively. Check @bdev can be opened
764 * exclusively and mark that an exclusive open is in progress. Each
765 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000766 * either bd_finish_claiming() or bd_abort_claiming() (which do not
767 * fail).
768 *
769 * This function is used to gain exclusive access to the block device
770 * without actually causing other exclusive open attempts to fail. It
771 * should be used when the open sequence itself requires exclusive
772 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900773 *
774 * CONTEXT:
775 * Might sleep.
776 *
777 * RETURNS:
778 * Pointer to the block device containing @bdev on success, ERR_PTR()
779 * value on failure.
780 */
781static struct block_device *bd_start_claiming(struct block_device *bdev,
782 void *holder)
783{
784 struct gendisk *disk;
785 struct block_device *whole;
786 int partno, err;
787
788 might_sleep();
789
790 /*
791 * @bdev might not have been initialized properly yet, look up
792 * and grab the outer block device the hard way.
793 */
794 disk = get_gendisk(bdev->bd_dev, &partno);
795 if (!disk)
796 return ERR_PTR(-ENXIO);
797
Tejun Heod4c208b2011-06-13 12:45:48 +0200798 /*
799 * Normally, @bdev should equal what's returned from bdget_disk()
800 * if partno is 0; however, some drivers (floppy) use multiple
801 * bdev's for the same physical device and @bdev may be one of the
802 * aliases. Keep @bdev if partno is 0. This means claimer
803 * tracking is broken for those devices but it has always been that
804 * way.
805 */
806 if (partno)
807 whole = bdget_disk(disk, 0);
808 else
809 whole = bdgrab(bdev);
810
Nick Piggincf342572010-05-26 01:50:21 +1000811 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900812 put_disk(disk);
813 if (!whole)
814 return ERR_PTR(-ENOMEM);
815
816 /* prepare to claim, if successful, mark claiming in progress */
817 spin_lock(&bdev_lock);
818
819 err = bd_prepare_to_claim(bdev, whole, holder);
820 if (err == 0) {
821 whole->bd_claiming = holder;
822 spin_unlock(&bdev_lock);
823 return whole;
824 } else {
825 spin_unlock(&bdev_lock);
826 bdput(whole);
827 return ERR_PTR(err);
828 }
829}
830
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800831#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100832struct bd_holder_disk {
833 struct list_head list;
834 struct gendisk *disk;
835 int refcnt;
836};
837
838static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
839 struct gendisk *disk)
840{
841 struct bd_holder_disk *holder;
842
843 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
844 if (holder->disk == disk)
845 return holder;
846 return NULL;
847}
848
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700849static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800850{
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700851 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800852}
853
854static void del_symlink(struct kobject *from, struct kobject *to)
855{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800856 sysfs_remove_link(from, kobject_name(to));
857}
858
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800859/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100860 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
861 * @bdev: the claimed slave bdev
862 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500863 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100864 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
865 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100866 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500867 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100868 * - from "slaves" directory of the holder @disk to the claimed @bdev
869 * - from "holders" directory of the @bdev to the holder @disk
870 *
871 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
872 * passed to bd_link_disk_holder(), then:
873 *
874 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
875 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
876 *
877 * The caller must have claimed @bdev before calling this function and
878 * ensure that both @bdev and @disk are valid during the creation and
879 * lifetime of these symlinks.
880 *
881 * CONTEXT:
882 * Might sleep.
883 *
884 * RETURNS:
885 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500886 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100887int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500888{
Tejun Heo49731ba2011-01-14 18:43:57 +0100889 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100890 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800891
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800892 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500893
Tejun Heo49731ba2011-01-14 18:43:57 +0100894 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700895
Tejun Heoe09b4572010-11-13 11:55:17 +0100896 /* FIXME: remove the following once add_disk() handles errors */
897 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
898 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700899
Tejun Heo49731ba2011-01-14 18:43:57 +0100900 holder = bd_find_holder_disk(bdev, disk);
901 if (holder) {
902 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100903 goto out_unlock;
904 }
905
Tejun Heo49731ba2011-01-14 18:43:57 +0100906 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
907 if (!holder) {
908 ret = -ENOMEM;
909 goto out_unlock;
910 }
911
912 INIT_LIST_HEAD(&holder->list);
913 holder->disk = disk;
914 holder->refcnt = 1;
915
916 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
917 if (ret)
918 goto out_free;
919
920 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
921 if (ret)
922 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100923 /*
924 * bdev could be deleted beneath us which would implicitly destroy
925 * the holder directory. Hold on to it.
926 */
927 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100928
929 list_add(&holder->list, &bdev->bd_holder_disks);
930 goto out_unlock;
931
932out_del:
933 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
934out_free:
935 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100936out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800937 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100938 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800939}
Tejun Heoe09b4572010-11-13 11:55:17 +0100940EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800941
Tejun Heo49731ba2011-01-14 18:43:57 +0100942/**
943 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
944 * @bdev: the calimed slave bdev
945 * @disk: the holding disk
946 *
947 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
948 *
949 * CONTEXT:
950 * Might sleep.
951 */
952void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800953{
Tejun Heo49731ba2011-01-14 18:43:57 +0100954 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100955
Tejun Heo49731ba2011-01-14 18:43:57 +0100956 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800957
Tejun Heo49731ba2011-01-14 18:43:57 +0100958 holder = bd_find_holder_disk(bdev, disk);
959
960 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
961 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
962 del_symlink(bdev->bd_part->holder_dir,
963 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +0100964 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100965 list_del_init(&holder->list);
966 kfree(holder);
967 }
968
969 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800970}
Tejun Heo49731ba2011-01-14 18:43:57 +0100971EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800972#endif
973
Andrew Patterson0c002c22008-09-04 14:27:20 -0600974/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600975 * flush_disk - invalidates all buffer-cache entries on a disk
976 *
977 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -0800978 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -0600979 *
980 * Invalidates all buffer-cache entries on a disk. It should be called
981 * when a disk has been changed -- either by a media change or online
982 * resize.
983 */
NeilBrown93b270f2011-02-24 17:25:47 +1100984static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -0600985{
NeilBrown93b270f2011-02-24 17:25:47 +1100986 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -0600987 char name[BDEVNAME_SIZE] = "";
988
989 if (bdev->bd_disk)
990 disk_name(bdev->bd_disk, 0, name);
991 printk(KERN_WARNING "VFS: busy inodes on changed media or "
992 "resized disk %s\n", name);
993 }
994
995 if (!bdev->bd_disk)
996 return;
Tejun Heod27769e2011-08-23 20:01:04 +0200997 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -0600998 bdev->bd_invalidated = 1;
999}
1000
1001/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001002 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001003 * @disk: struct gendisk to check
1004 * @bdev: struct bdev to adjust.
1005 *
1006 * This routine checks to see if the bdev size does not match the disk size
1007 * and adjusts it if it differs.
1008 */
1009void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1010{
1011 loff_t disk_size, bdev_size;
1012
1013 disk_size = (loff_t)get_capacity(disk) << 9;
1014 bdev_size = i_size_read(bdev->bd_inode);
1015 if (disk_size != bdev_size) {
1016 char name[BDEVNAME_SIZE];
1017
1018 disk_name(disk, 0, name);
1019 printk(KERN_INFO
1020 "%s: detected capacity change from %lld to %lld\n",
1021 name, bdev_size, disk_size);
1022 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +11001023 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001024 }
1025}
1026EXPORT_SYMBOL(check_disk_size_change);
1027
1028/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001029 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001030 * @disk: struct gendisk to be revalidated
1031 *
1032 * This routine is a wrapper for lower-level driver's revalidate_disk
1033 * call-backs. It is used to do common pre and post operations needed
1034 * for all revalidate_disk operations.
1035 */
1036int revalidate_disk(struct gendisk *disk)
1037{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001038 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001039 int ret = 0;
1040
1041 if (disk->fops->revalidate_disk)
1042 ret = disk->fops->revalidate_disk(disk);
1043
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001044 bdev = bdget_disk(disk, 0);
1045 if (!bdev)
1046 return ret;
1047
1048 mutex_lock(&bdev->bd_mutex);
1049 check_disk_size_change(disk, bdev);
1050 mutex_unlock(&bdev->bd_mutex);
1051 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001052 return ret;
1053}
1054EXPORT_SYMBOL(revalidate_disk);
1055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056/*
1057 * This routine checks whether a removable media has been changed,
1058 * and invalidates all buffer-cache-entries in that case. This
1059 * is a relatively slow routine, so we have to try to minimize using
1060 * it. Thus it is called only upon a 'mount' or 'open'. This
1061 * is the best way of combining speed and utility, I think.
1062 * People changing diskettes in the middle of an operation deserve
1063 * to lose :-)
1064 */
1065int check_disk_change(struct block_device *bdev)
1066{
1067 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001068 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001069 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Tejun Heo77ea8872010-12-08 20:57:37 +01001071 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1072 DISK_EVENT_EJECT_REQUEST);
1073 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 return 0;
1075
NeilBrown93b270f2011-02-24 17:25:47 +11001076 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (bdops->revalidate_disk)
1078 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return 1;
1080}
1081
1082EXPORT_SYMBOL(check_disk_change);
1083
1084void bd_set_size(struct block_device *bdev, loff_t size)
1085{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001086 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 bdev->bd_inode->i_size = size;
1089 while (bsize < PAGE_CACHE_SIZE) {
1090 if (size & bsize)
1091 break;
1092 bsize <<= 1;
1093 }
1094 bdev->bd_block_size = bsize;
1095 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1096}
1097EXPORT_SYMBOL(bd_set_size);
1098
Al Viro9a1c3542008-02-22 20:40:24 -05001099static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001100
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001101/*
1102 * bd_mutex locking:
1103 *
1104 * mutex_lock(part->bd_mutex)
1105 * mutex_lock_nested(whole->bd_mutex, 1)
1106 */
1107
Al Viro572c4892007-10-08 13:24:05 -04001108static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001111 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001112 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001113 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001114 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
Al Viro572c4892007-10-08 13:24:05 -04001116 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001117 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001118 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001119 perm |= MAY_WRITE;
1120 /*
1121 * hooks: /n/, see "layering violations".
1122 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001123 if (!for_part) {
1124 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1125 if (ret != 0) {
1126 bdput(bdev);
1127 return ret;
1128 }
Al Viro82666022008-08-01 05:32:04 -04001129 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001130
NeilBrownd3374822009-01-09 08:31:10 +11001131 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001132
Tejun Heo89f97492008-11-05 10:21:06 +01001133 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001134 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001135 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001136 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001137 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Tejun Heo69e02c52011-03-09 19:54:27 +01001139 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001140 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (!bdev->bd_openers) {
1142 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001143 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001145 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001147
1148 ret = -ENXIO;
1149 bdev->bd_part = disk_get_part(disk, partno);
1150 if (!bdev->bd_part)
1151 goto out_clear;
1152
Tejun Heo1196f8b2011-04-21 20:54:45 +02001153 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001155 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001156 if (ret == -ERESTARTSYS) {
1157 /* Lost a race with 'disk' being
1158 * deleted, try again.
1159 * See md.c
1160 */
1161 disk_put_part(bdev->bd_part);
1162 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001163 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001164 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001165 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001166 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001167 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001168 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001169 goto restart;
1170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 }
Tejun Heo7e697232011-05-23 13:26:07 +02001172
1173 if (!ret && !bdev->bd_openers) {
1174 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1175 bdi = blk_get_backing_dev_info(bdev);
1176 if (bdi == NULL)
1177 bdi = &default_backing_dev_info;
1178 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1179 }
1180
Tejun Heo1196f8b2011-04-21 20:54:45 +02001181 /*
1182 * If the device is invalidated, rescan partition
1183 * if open succeeded or failed with -ENOMEDIUM.
1184 * The latter is necessary to prevent ghost
1185 * partitions on a removed medium.
1186 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001187 if (bdev->bd_invalidated) {
1188 if (!ret)
1189 rescan_partitions(disk, bdev);
1190 else if (ret == -ENOMEDIUM)
1191 invalidate_partitions(disk, bdev);
1192 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001193 if (ret)
1194 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 struct block_device *whole;
1197 whole = bdget_disk(disk, 0);
1198 ret = -ENOMEM;
1199 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001200 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001201 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001202 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001204 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001206 bdev_inode_switch_bdi(bdev->bd_inode,
1207 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001208 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001209 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001210 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001212 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Tejun Heo89f97492008-11-05 10:21:06 +01001214 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 }
1216 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001218 ret = 0;
1219 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001220 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001221 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001222 if (bdev->bd_invalidated) {
1223 if (!ret)
1224 rescan_partitions(bdev->bd_disk, bdev);
1225 else if (ret == -ENOMEDIUM)
1226 invalidate_partitions(bdev->bd_disk, bdev);
1227 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001228 if (ret)
1229 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001231 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001232 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001233 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 }
1235 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001236 if (for_part)
1237 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001238 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001239 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 return 0;
1241
Tejun Heo0762b8b2008-08-25 19:56:12 +09001242 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001243 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001245 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001246 bdev->bd_queue = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001247 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001249 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001251 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001252 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001253 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001254 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001255 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001256 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001257 bdput(bdev);
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 return ret;
1260}
1261
Tejun Heod4d77622010-11-13 11:55:18 +01001262/**
1263 * blkdev_get - open a block device
1264 * @bdev: block_device to open
1265 * @mode: FMODE_* mask
1266 * @holder: exclusive holder identifier
1267 *
1268 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1269 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1270 * @holder is invalid. Exclusive opens may nest for the same @holder.
1271 *
1272 * On success, the reference count of @bdev is unchanged. On failure,
1273 * @bdev is put.
1274 *
1275 * CONTEXT:
1276 * Might sleep.
1277 *
1278 * RETURNS:
1279 * 0 on success, -errno on failure.
1280 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001281int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282{
Tejun Heoe525fd82010-11-13 11:55:17 +01001283 struct block_device *whole = NULL;
1284 int res;
1285
1286 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1287
1288 if ((mode & FMODE_EXCL) && holder) {
1289 whole = bd_start_claiming(bdev, holder);
1290 if (IS_ERR(whole)) {
1291 bdput(bdev);
1292 return PTR_ERR(whole);
1293 }
1294 }
1295
1296 res = __blkdev_get(bdev, mode, 0);
1297
1298 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001299 struct gendisk *disk = whole->bd_disk;
1300
Tejun Heo6a027ef2010-11-13 11:55:17 +01001301 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001302 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001303 spin_lock(&bdev_lock);
1304
Tejun Heo77ea8872010-12-08 20:57:37 +01001305 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001306 BUG_ON(!bd_may_claim(bdev, whole, holder));
1307 /*
1308 * Note that for a whole device bd_holders
1309 * will be incremented twice, and bd_holder
1310 * will be set to bd_may_claim before being
1311 * set to holder
1312 */
1313 whole->bd_holders++;
1314 whole->bd_holder = bd_may_claim;
1315 bdev->bd_holders++;
1316 bdev->bd_holder = holder;
1317 }
1318
1319 /* tell others that we're done */
1320 BUG_ON(whole->bd_claiming != holder);
1321 whole->bd_claiming = NULL;
1322 wake_up_bit(&whole->bd_claiming, 0);
1323
1324 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001325
1326 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001327 * Block event polling for write claims if requested. Any
1328 * write holder makes the write_holder state stick until
1329 * all are released. This is good enough and tracking
1330 * individual writeable reference is too fragile given the
1331 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001332 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001333 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1334 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001335 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001336 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001337 }
1338
1339 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001340 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001341 }
1342
1343 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001344}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345EXPORT_SYMBOL(blkdev_get);
1346
Tejun Heod4d77622010-11-13 11:55:18 +01001347/**
1348 * blkdev_get_by_path - open a block device by name
1349 * @path: path to the block device to open
1350 * @mode: FMODE_* mask
1351 * @holder: exclusive holder identifier
1352 *
1353 * Open the blockdevice described by the device file at @path. @mode
1354 * and @holder are identical to blkdev_get().
1355 *
1356 * On success, the returned block_device has reference count of one.
1357 *
1358 * CONTEXT:
1359 * Might sleep.
1360 *
1361 * RETURNS:
1362 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1363 */
1364struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1365 void *holder)
1366{
1367 struct block_device *bdev;
1368 int err;
1369
1370 bdev = lookup_bdev(path);
1371 if (IS_ERR(bdev))
1372 return bdev;
1373
1374 err = blkdev_get(bdev, mode, holder);
1375 if (err)
1376 return ERR_PTR(err);
1377
Chuck Ebberte51900f2011-02-16 18:11:53 -05001378 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1379 blkdev_put(bdev, mode);
1380 return ERR_PTR(-EACCES);
1381 }
1382
Tejun Heod4d77622010-11-13 11:55:18 +01001383 return bdev;
1384}
1385EXPORT_SYMBOL(blkdev_get_by_path);
1386
1387/**
1388 * blkdev_get_by_dev - open a block device by device number
1389 * @dev: device number of block device to open
1390 * @mode: FMODE_* mask
1391 * @holder: exclusive holder identifier
1392 *
1393 * Open the blockdevice described by device number @dev. @mode and
1394 * @holder are identical to blkdev_get().
1395 *
1396 * Use it ONLY if you really do not have anything better - i.e. when
1397 * you are behind a truly sucky interface and all you are given is a
1398 * device number. _Never_ to be used for internal purposes. If you
1399 * ever need it - reconsider your API.
1400 *
1401 * On success, the returned block_device has reference count of one.
1402 *
1403 * CONTEXT:
1404 * Might sleep.
1405 *
1406 * RETURNS:
1407 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1408 */
1409struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1410{
1411 struct block_device *bdev;
1412 int err;
1413
1414 bdev = bdget(dev);
1415 if (!bdev)
1416 return ERR_PTR(-ENOMEM);
1417
1418 err = blkdev_get(bdev, mode, holder);
1419 if (err)
1420 return ERR_PTR(err);
1421
1422 return bdev;
1423}
1424EXPORT_SYMBOL(blkdev_get_by_dev);
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static int blkdev_open(struct inode * inode, struct file * filp)
1427{
1428 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /*
1431 * Preserve backwards compatibility and allow large file access
1432 * even if userspace doesn't ask for it explicitly. Some mkfs
1433 * binary needs it. We might want to drop this workaround
1434 * during an unstable branch.
1435 */
1436 filp->f_flags |= O_LARGEFILE;
1437
Al Viro572c4892007-10-08 13:24:05 -04001438 if (filp->f_flags & O_NDELAY)
1439 filp->f_mode |= FMODE_NDELAY;
1440 if (filp->f_flags & O_EXCL)
1441 filp->f_mode |= FMODE_EXCL;
1442 if ((filp->f_flags & O_ACCMODE) == 3)
1443 filp->f_mode |= FMODE_WRITE_IOCTL;
1444
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001446 if (bdev == NULL)
1447 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Al Viro572c4892007-10-08 13:24:05 -04001449 filp->f_mapping = bdev->bd_inode->i_mapping;
1450
Tejun Heoe525fd82010-11-13 11:55:17 +01001451 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
Al Viro9a1c3542008-02-22 20:40:24 -05001454static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001455{
1456 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001457 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001458 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001459
NeilBrown6796bf52006-12-08 02:36:16 -08001460 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001461 if (for_part)
1462 bdev->bd_part_count--;
1463
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001464 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001465 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001466 sync_blockdev(bdev);
1467 kill_bdev(bdev);
NeilBrown94007752011-09-10 17:20:21 +10001468 /* ->release can cause the old bdi to disappear,
1469 * so must switch it out first
1470 */
1471 bdev_inode_switch_bdi(bdev->bd_inode,
1472 &default_backing_dev_info);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001473 }
1474 if (bdev->bd_contains == bdev) {
1475 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001476 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001477 }
1478 if (!bdev->bd_openers) {
1479 struct module *owner = disk->fops->owner;
1480
Tejun Heo0762b8b2008-08-25 19:56:12 +09001481 disk_put_part(bdev->bd_part);
1482 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001483 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001484 if (bdev != bdev->bd_contains)
1485 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001486 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001487
1488 put_disk(disk);
1489 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001490 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001491 mutex_unlock(&bdev->bd_mutex);
1492 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001493 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001494 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001495 return ret;
1496}
1497
Al Viro9a1c3542008-02-22 20:40:24 -05001498int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001499{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001500 mutex_lock(&bdev->bd_mutex);
1501
Tejun Heoe525fd82010-11-13 11:55:17 +01001502 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001503 bool bdev_free;
1504
1505 /*
1506 * Release a claim on the device. The holder fields
1507 * are protected with bdev_lock. bd_mutex is to
1508 * synchronize disk_holder unlinking.
1509 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001510 spin_lock(&bdev_lock);
1511
1512 WARN_ON_ONCE(--bdev->bd_holders < 0);
1513 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1514
1515 /* bd_contains might point to self, check in a separate step */
1516 if ((bdev_free = !bdev->bd_holders))
1517 bdev->bd_holder = NULL;
1518 if (!bdev->bd_contains->bd_holders)
1519 bdev->bd_contains->bd_holder = NULL;
1520
1521 spin_unlock(&bdev_lock);
1522
Tejun Heo77ea8872010-12-08 20:57:37 +01001523 /*
1524 * If this was the last claim, remove holder link and
1525 * unblock evpoll if it was a write holder.
1526 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001527 if (bdev_free && bdev->bd_write_holder) {
1528 disk_unblock_events(bdev->bd_disk);
1529 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001530 }
Tejun Heo69362172011-03-09 19:54:27 +01001531 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001532
Tejun Heo85ef06d2011-07-01 16:17:47 +02001533 /*
1534 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1535 * event. This is to ensure detection of media removal commanded
1536 * from userland - e.g. eject(1).
1537 */
1538 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1539
1540 mutex_unlock(&bdev->bd_mutex);
1541
Al Viro9a1c3542008-02-22 20:40:24 -05001542 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001543}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001544EXPORT_SYMBOL(blkdev_put);
1545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546static int blkdev_close(struct inode * inode, struct file * filp)
1547{
1548 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Tejun Heoe525fd82010-11-13 11:55:17 +01001549
Al Viro9a1c3542008-02-22 20:40:24 -05001550 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001553static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554{
Al Viro56b26ad2008-09-19 03:17:36 -04001555 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1556 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001557
1558 /*
1559 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1560 * to updated it before every ioctl.
1561 */
Al Viro56b26ad2008-09-19 03:17:36 -04001562 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001563 mode |= FMODE_NDELAY;
1564 else
1565 mode &= ~FMODE_NDELAY;
1566
Al Viro56b26ad2008-09-19 03:17:36 -04001567 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
1569
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001570/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001571 * Write data to the block device. Only intended for the block device itself
1572 * and the raw driver which basically is a fake block device.
1573 *
1574 * Does not take i_mutex for the write and thus is not for general purpose
1575 * use.
1576 */
1577ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1578 unsigned long nr_segs, loff_t pos)
1579{
1580 struct file *file = iocb->ki_filp;
Jianpeng Ma53362a02012-08-02 09:50:39 +02001581 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001582 ssize_t ret;
1583
1584 BUG_ON(iocb->ki_pos != pos);
1585
Jianpeng Ma53362a02012-08-02 09:50:39 +02001586 blk_start_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001587 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1588 if (ret > 0 || ret == -EIOCBQUEUED) {
1589 ssize_t err;
1590
1591 err = generic_write_sync(file, pos, ret);
1592 if (err < 0 && ret > 0)
1593 ret = err;
1594 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001595 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001596 return ret;
1597}
1598EXPORT_SYMBOL_GPL(blkdev_aio_write);
1599
1600/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001601 * Try to release a page associated with block device when the system
1602 * is under memory pressure.
1603 */
1604static int blkdev_releasepage(struct page *page, gfp_t wait)
1605{
1606 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1607
1608 if (super && super->s_op->bdev_try_to_free_page)
1609 return super->s_op->bdev_try_to_free_page(super, page, wait);
1610
1611 return try_to_free_buffers(page);
1612}
1613
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001614static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 .readpage = blkdev_readpage,
1616 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001617 .write_begin = blkdev_write_begin,
1618 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001620 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 .direct_IO = blkdev_direct_IO,
1622};
1623
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001624const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 .open = blkdev_open,
1626 .release = blkdev_close,
1627 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001628 .read = do_sync_read,
1629 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001631 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001633 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001634 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635#ifdef CONFIG_COMPAT
1636 .compat_ioctl = compat_blkdev_ioctl,
1637#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001638 .splice_read = generic_file_splice_read,
1639 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640};
1641
1642int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1643{
1644 int res;
1645 mm_segment_t old_fs = get_fs();
1646 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001647 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 set_fs(old_fs);
1649 return res;
1650}
1651
1652EXPORT_SYMBOL(ioctl_by_bdev);
1653
1654/**
1655 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001656 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001658 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 * namespace if possible and return it. Return ERR_PTR(error)
1660 * otherwise.
1661 */
Al Viro421748e2008-08-02 01:04:36 -04001662struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663{
1664 struct block_device *bdev;
1665 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001666 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 int error;
1668
Al Viro421748e2008-08-02 01:04:36 -04001669 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 return ERR_PTR(-EINVAL);
1671
Al Viro421748e2008-08-02 01:04:36 -04001672 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 if (error)
1674 return ERR_PTR(error);
1675
Al Viro421748e2008-08-02 01:04:36 -04001676 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 error = -ENOTBLK;
1678 if (!S_ISBLK(inode->i_mode))
1679 goto fail;
1680 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001681 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 goto fail;
1683 error = -ENOMEM;
1684 bdev = bd_acquire(inode);
1685 if (!bdev)
1686 goto fail;
1687out:
Al Viro421748e2008-08-02 01:04:36 -04001688 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 return bdev;
1690fail:
1691 bdev = ERR_PTR(error);
1692 goto out;
1693}
Al Virod5686b42008-08-01 05:00:11 -04001694EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695
NeilBrown93b270f2011-02-24 17:25:47 +11001696int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001697{
1698 struct super_block *sb = get_super(bdev);
1699 int res = 0;
1700
1701 if (sb) {
1702 /*
1703 * no need to lock the super, get_super holds the
1704 * read mutex so the filesystem cannot go away
1705 * under us (->put_super runs with the write lock
1706 * hold).
1707 */
1708 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001709 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001710 drop_super(sb);
1711 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001712 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001713 return res;
1714}
1715EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001716
1717void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1718{
1719 struct inode *inode, *old_inode = NULL;
1720
1721 spin_lock(&inode_sb_list_lock);
1722 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1723 struct address_space *mapping = inode->i_mapping;
1724
1725 spin_lock(&inode->i_lock);
1726 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1727 mapping->nrpages == 0) {
1728 spin_unlock(&inode->i_lock);
1729 continue;
1730 }
1731 __iget(inode);
1732 spin_unlock(&inode->i_lock);
1733 spin_unlock(&inode_sb_list_lock);
1734 /*
1735 * We hold a reference to 'inode' so it couldn't have been
1736 * removed from s_inodes list while we dropped the
1737 * inode_sb_list_lock. We cannot iput the inode now as we can
1738 * be holding the last reference and we cannot iput it under
1739 * inode_sb_list_lock. So we keep the reference and iput it
1740 * later.
1741 */
1742 iput(old_inode);
1743 old_inode = inode;
1744
1745 func(I_BDEV(inode), arg);
1746
1747 spin_lock(&inode_sb_list_lock);
1748 }
1749 spin_unlock(&inode_sb_list_lock);
1750 iput(old_inode);
1751}