blob: ab3a456f66506a5a134f5280739e82c572b15f55 [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 Viroff01bb42011-09-16 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 Viroff01bb42011-09-16 02:31:11 -040029#include <linux/cleancache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <asm/uaccess.h>
David Howells07f3f05c2006-09-30 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
Peter Zijlstraf9a14392007-05-06 14:49:55 -070073/* Kill _all_ buffers and pagecache , dirty or not.. */
Al Viroff01bb42011-09-16 02:31:11 -040074void kill_bdev(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Al Viroff01bb42011-09-16 02:31:11 -040076 struct address_space *mapping = bdev->bd_inode->i_mapping;
77
78 if (mapping->nrpages == 0)
Peter Zijlstraf9a14392007-05-06 14:49:55 -070079 return;
Al Viroff01bb42011-09-16 02:31:11 -040080
Peter Zijlstraf9a14392007-05-06 14:49:55 -070081 invalidate_bh_lrus();
Al Viroff01bb42011-09-16 02:31:11 -040082 truncate_inode_pages(mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
Al Viroff01bb42011-09-16 02:31:11 -040084EXPORT_SYMBOL(kill_bdev);
85
86/* Invalidate clean unused buffers and pagecache. */
87void invalidate_bdev(struct block_device *bdev)
88{
89 struct address_space *mapping = bdev->bd_inode->i_mapping;
90
91 if (mapping->nrpages == 0)
92 return;
93
94 invalidate_bh_lrus();
95 lru_add_drain_all(); /* make sure all lru add caches are flushed */
96 invalidate_mapping_pages(mapping, 0, -1);
97 /* 99% of the time, we don't need to flush the cleancache on the bdev.
98 * But, for the strange corners, lets be cautious
99 */
Dan Magenheimer31677602011-09-21 11:56:28 -0400100 cleancache_invalidate_inode(mapping);
Al Viroff01bb42011-09-16 02:31:11 -0400101}
102EXPORT_SYMBOL(invalidate_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104int set_blocksize(struct block_device *bdev, int size)
105{
106 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -0700107 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 return -EINVAL;
109
110 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -0400111 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 return -EINVAL;
113
114 /* Don't change the size if it is same as current */
115 if (bdev->bd_block_size != size) {
116 sync_blockdev(bdev);
117 bdev->bd_block_size = size;
118 bdev->bd_inode->i_blkbits = blksize_bits(size);
119 kill_bdev(bdev);
120 }
121 return 0;
122}
123
124EXPORT_SYMBOL(set_blocksize);
125
126int sb_set_blocksize(struct super_block *sb, int size)
127{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 if (set_blocksize(sb->s_bdev, size))
129 return 0;
130 /* If we get here, we know size is power of two
131 * and it's value is between 512 and PAGE_SIZE */
132 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800133 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 return sb->s_blocksize;
135}
136
137EXPORT_SYMBOL(sb_set_blocksize);
138
139int sb_min_blocksize(struct super_block *sb, int size)
140{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400141 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 if (size < minsize)
143 size = minsize;
144 return sb_set_blocksize(sb, size);
145}
146
147EXPORT_SYMBOL(sb_min_blocksize);
148
149static int
150blkdev_get_block(struct inode *inode, sector_t iblock,
151 struct buffer_head *bh, int create)
152{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 bh->b_bdev = I_BDEV(inode);
154 bh->b_blocknr = iblock;
155 set_buffer_mapped(bh);
156 return 0;
157}
158
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800159static ssize_t
160blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
161 loff_t offset, unsigned long nr_segs)
162{
163 struct file *file = iocb->ki_filp;
164 struct inode *inode = file->f_mapping->host;
165
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200166 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800167 nr_segs, blkdev_get_block, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800168}
169
Jan Kara5cee5812009-04-27 16:43:51 +0200170int __sync_blockdev(struct block_device *bdev, int wait)
171{
172 if (!bdev)
173 return 0;
174 if (!wait)
175 return filemap_flush(bdev->bd_inode->i_mapping);
176 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
177}
178
Nick Piggin585d3bc2009-02-25 10:44:19 +0100179/*
180 * Write out and wait upon all the dirty data associated with a block
181 * device via its mapping. Does not take the superblock lock.
182 */
183int sync_blockdev(struct block_device *bdev)
184{
Jan Kara5cee5812009-04-27 16:43:51 +0200185 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100186}
187EXPORT_SYMBOL(sync_blockdev);
188
189/*
190 * Write out and wait upon all dirty data associated with this
191 * device. Filesystem data as well as the underlying block
192 * device. Takes the superblock lock.
193 */
194int fsync_bdev(struct block_device *bdev)
195{
196 struct super_block *sb = get_super(bdev);
197 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200198 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100199 drop_super(sb);
200 return res;
201 }
202 return sync_blockdev(bdev);
203}
Al Viro47e44912009-04-01 07:07:16 -0400204EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100205
206/**
207 * freeze_bdev -- lock a filesystem and force it into a consistent state
208 * @bdev: blockdevice to lock
209 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100210 * If a superblock is found on this device, we take the s_umount semaphore
211 * on it to make sure nobody unmounts until the snapshot creation is done.
212 * The reference counter (bd_fsfreeze_count) guarantees that only the last
213 * unfreeze process can unfreeze the frozen filesystem actually when multiple
214 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
215 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
216 * actually.
217 */
218struct super_block *freeze_bdev(struct block_device *bdev)
219{
220 struct super_block *sb;
221 int error = 0;
222
223 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200224 if (++bdev->bd_fsfreeze_count > 1) {
225 /*
226 * We don't even need to grab a reference - the first call
227 * to freeze_bdev grab an active reference and only the last
228 * thaw_bdev drops it.
229 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100230 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200231 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100232 mutex_unlock(&bdev->bd_fsfreeze_mutex);
233 return sb;
234 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100235
Christoph Hellwig45042302009-08-03 23:28:35 +0200236 sb = get_active_super(bdev);
237 if (!sb)
238 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400239 error = freeze_super(sb);
240 if (error) {
241 deactivate_super(sb);
242 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200243 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400244 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100245 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400246 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200247 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100248 sync_blockdev(bdev);
249 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200250 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100251}
252EXPORT_SYMBOL(freeze_bdev);
253
254/**
255 * thaw_bdev -- unlock filesystem
256 * @bdev: blockdevice to unlock
257 * @sb: associated superblock
258 *
259 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
260 */
261int thaw_bdev(struct block_device *bdev, struct super_block *sb)
262{
Christoph Hellwig45042302009-08-03 23:28:35 +0200263 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100264
265 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200266 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400267 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100268
Christoph Hellwig45042302009-08-03 23:28:35 +0200269 error = 0;
270 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400271 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100272
Christoph Hellwig45042302009-08-03 23:28:35 +0200273 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400274 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200275
Josef Bacik18e9e512010-03-23 10:34:56 -0400276 error = thaw_super(sb);
277 if (error) {
278 bdev->bd_fsfreeze_count++;
279 mutex_unlock(&bdev->bd_fsfreeze_mutex);
280 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100281 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400282out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100283 mutex_unlock(&bdev->bd_fsfreeze_mutex);
284 return 0;
285}
286EXPORT_SYMBOL(thaw_bdev);
287
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
289{
290 return block_write_full_page(page, blkdev_get_block, wbc);
291}
292
293static int blkdev_readpage(struct file * file, struct page * page)
294{
295 return block_read_full_page(page, blkdev_get_block);
296}
297
Nick Piggin6272b5a2007-10-16 01:25:04 -0700298static int blkdev_write_begin(struct file *file, struct address_space *mapping,
299 loff_t pos, unsigned len, unsigned flags,
300 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200302 return block_write_begin(mapping, pos, len, flags, pagep,
303 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
Nick Piggin6272b5a2007-10-16 01:25:04 -0700306static int blkdev_write_end(struct file *file, struct address_space *mapping,
307 loff_t pos, unsigned len, unsigned copied,
308 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700310 int ret;
311 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
312
313 unlock_page(page);
314 page_cache_release(page);
315
316 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317}
318
319/*
320 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800321 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 * so we compute the size by hand (just as in block_read/write above)
323 */
324static loff_t block_llseek(struct file *file, loff_t offset, int origin)
325{
326 struct inode *bd_inode = file->f_mapping->host;
327 loff_t size;
328 loff_t retval;
329
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800330 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 size = i_size_read(bd_inode);
332
Josef Bacik06222e42011-07-18 13:21:38 -0400333 retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 switch (origin) {
Josef Bacik06222e42011-07-18 13:21:38 -0400335 case SEEK_END:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 offset += size;
337 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400338 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 offset += file->f_pos;
Josef Bacik06222e42011-07-18 13:21:38 -0400340 case SEEK_SET:
341 break;
342 default:
343 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (offset >= 0 && offset <= size) {
346 if (offset != file->f_pos) {
347 file->f_pos = offset;
348 }
349 retval = offset;
350 }
Josef Bacik06222e42011-07-18 13:21:38 -0400351out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800352 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 return retval;
354}
355
Josef Bacik02c24a82011-07-16 20:44:56 -0400356int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400358 struct inode *bd_inode = filp->f_mapping->host;
359 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100360 int error;
Rafael J. Wysockida5aa862011-08-02 02:17:48 +0200361
362 error = filemap_write_and_wait_range(filp->f_mapping, start, end);
363 if (error)
364 return error;
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100365
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400366 /*
367 * There is no need to serialise calls to blkdev_issue_flush with
368 * i_mutex and doing so causes performance issues with concurrent
369 * O_SYNC writers to a block device.
370 */
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200371 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100372 if (error == -EOPNOTSUPP)
373 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400374
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100375 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700377EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379/*
380 * pseudo-fs
381 */
382
383static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800384static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386static struct inode *bdev_alloc_inode(struct super_block *sb)
387{
Christoph Lametere94b1762006-12-06 20:33:17 -0800388 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (!ei)
390 return NULL;
391 return &ei->vfs_inode;
392}
393
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100394static void bdev_i_callback(struct rcu_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100396 struct inode *inode = container_of(head, struct inode, i_rcu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 struct bdev_inode *bdi = BDEV_I(inode);
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 kmem_cache_free(bdev_cachep, bdi);
400}
401
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +1100402static void bdev_destroy_inode(struct inode *inode)
403{
404 call_rcu(&inode->i_rcu, bdev_i_callback);
405}
406
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700407static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408{
409 struct bdev_inode *ei = (struct bdev_inode *) foo;
410 struct block_device *bdev = &ei->bdev;
411
Christoph Lametera35afb82007-05-16 22:10:57 -0700412 memset(bdev, 0, sizeof(*bdev));
413 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700414 INIT_LIST_HEAD(&bdev->bd_inodes);
415 INIT_LIST_HEAD(&bdev->bd_list);
Tejun Heo49731ba2011-01-14 18:43:57 +0100416#ifdef CONFIG_SYSFS
417 INIT_LIST_HEAD(&bdev->bd_holder_disks);
418#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700419 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800420 /* Initialize mutex for freeze. */
421 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
424static inline void __bd_forget(struct inode *inode)
425{
426 list_del_init(&inode->i_devices);
427 inode->i_bdev = NULL;
428 inode->i_mapping = &inode->i_data;
429}
430
Al Virob57922d2010-06-07 14:34:48 -0400431static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
433 struct block_device *bdev = &BDEV_I(inode)->bdev;
434 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400435 truncate_inode_pages(&inode->i_data, 0);
436 invalidate_inode_buffers(inode); /* is it needed here? */
Jan Karadbd57682012-05-03 14:48:02 +0200437 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 spin_lock(&bdev_lock);
439 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
440 __bd_forget(list_entry(p, struct inode, i_devices));
441 }
442 list_del_init(&bdev->bd_list);
443 spin_unlock(&bdev_lock);
444}
445
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800446static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 .statfs = simple_statfs,
448 .alloc_inode = bdev_alloc_inode,
449 .destroy_inode = bdev_destroy_inode,
450 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400451 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452};
453
Al Viro51139ad2010-07-25 23:47:46 +0400454static struct dentry *bd_mount(struct file_system_type *fs_type,
455 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Muthu Kumarb502bd12012-03-23 15:01:50 -0700457 return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
460static struct file_system_type bd_type = {
461 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400462 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 .kill_sb = kill_anon_super,
464};
465
Al Virof47ec3f2011-11-21 21:15:42 -0500466static struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468void __init bdev_cache_init(void)
469{
470 int err;
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300471 static struct vfsmount *bd_mnt;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800474 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
475 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900476 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 err = register_filesystem(&bd_type);
478 if (err)
479 panic("Cannot register bdev pseudo-fs");
480 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 if (IS_ERR(bd_mnt))
482 panic("Cannot create bdev pseudo-fs");
Sergey Senozhatskyace85772012-01-10 02:43:59 +0300483 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484}
485
486/*
487 * Most likely _very_ bad one - but then it's hardly critical for small
488 * /dev and can be fixed when somebody will need really large one.
489 * Keep in mind that it will be fed through icache hash function too.
490 */
491static inline unsigned long hash(dev_t dev)
492{
493 return MAJOR(dev)+MINOR(dev);
494}
495
496static int bdev_test(struct inode *inode, void *data)
497{
498 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
499}
500
501static int bdev_set(struct inode *inode, void *data)
502{
503 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
504 return 0;
505}
506
507static LIST_HEAD(all_bdevs);
508
509struct block_device *bdget(dev_t dev)
510{
511 struct block_device *bdev;
512 struct inode *inode;
513
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800514 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 bdev_test, bdev_set, &dev);
516
517 if (!inode)
518 return NULL;
519
520 bdev = &BDEV_I(inode)->bdev;
521
522 if (inode->i_state & I_NEW) {
523 bdev->bd_contains = NULL;
Lachlan McIlroy782b94c2011-06-30 11:01:45 +1000524 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 bdev->bd_inode = inode;
526 bdev->bd_block_size = (1 << inode->i_blkbits);
527 bdev->bd_part_count = 0;
528 bdev->bd_invalidated = 0;
529 inode->i_mode = S_IFBLK;
530 inode->i_rdev = dev;
531 inode->i_bdev = bdev;
532 inode->i_data.a_ops = &def_blk_aops;
533 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
534 inode->i_data.backing_dev_info = &default_backing_dev_info;
535 spin_lock(&bdev_lock);
536 list_add(&bdev->bd_list, &all_bdevs);
537 spin_unlock(&bdev_lock);
538 unlock_new_inode(inode);
539 }
540 return bdev;
541}
542
543EXPORT_SYMBOL(bdget);
544
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200545/**
546 * bdgrab -- Grab a reference to an already referenced block device
547 * @bdev: Block device to grab a reference to.
548 */
549struct block_device *bdgrab(struct block_device *bdev)
550{
Al Viro7de9c6ee2010-10-23 11:11:40 -0400551 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200552 return bdev;
553}
554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555long nr_blockdev_pages(void)
556{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700557 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 long ret = 0;
559 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700560 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 ret += bdev->bd_inode->i_mapping->nrpages;
562 }
563 spin_unlock(&bdev_lock);
564 return ret;
565}
566
567void bdput(struct block_device *bdev)
568{
569 iput(bdev->bd_inode);
570}
571
572EXPORT_SYMBOL(bdput);
573
574static struct block_device *bd_acquire(struct inode *inode)
575{
576 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 spin_lock(&bdev_lock);
579 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700580 if (bdev) {
Al Viro7de9c6ee2010-10-23 11:11:40 -0400581 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 spin_unlock(&bdev_lock);
583 return bdev;
584 }
585 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700586
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 bdev = bdget(inode->i_rdev);
588 if (bdev) {
589 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700590 if (!inode->i_bdev) {
591 /*
Al Viro7de9c6ee2010-10-23 11:11:40 -0400592 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700593 * and it's released in clear_inode() of inode.
594 * So, we can access it via ->i_mapping always
595 * without igrab().
596 */
Al Viro7de9c6ee2010-10-23 11:11:40 -0400597 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700598 inode->i_bdev = bdev;
599 inode->i_mapping = bdev->bd_inode->i_mapping;
600 list_add(&inode->i_devices, &bdev->bd_inodes);
601 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 spin_unlock(&bdev_lock);
603 }
604 return bdev;
605}
606
Al Virof47ec3f2011-11-21 21:15:42 -0500607static inline int sb_is_blkdev_sb(struct super_block *sb)
608{
609 return sb == blockdev_superblock;
610}
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612/* Call when you free inode */
613
614void bd_forget(struct inode *inode)
615{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700616 struct block_device *bdev = NULL;
617
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700619 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800620 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700621 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700623 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700625
626 if (bdev)
627 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628}
629
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900630/**
631 * bd_may_claim - test whether a block device can be claimed
632 * @bdev: block device of interest
633 * @whole: whole block device containing @bdev, may equal @bdev
634 * @holder: holder trying to claim @bdev
635 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300636 * Test whether @bdev can be claimed by @holder.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900637 *
638 * CONTEXT:
639 * spin_lock(&bdev_lock).
640 *
641 * RETURNS:
642 * %true if @bdev can be claimed, %false otherwise.
643 */
644static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
645 void *holder)
646{
647 if (bdev->bd_holder == holder)
648 return true; /* already a holder */
649 else if (bdev->bd_holder != NULL)
650 return false; /* held by someone else */
651 else if (bdev->bd_contains == bdev)
652 return true; /* is a whole device which isn't held */
653
Tejun Heoe525fd82010-11-13 11:55:17 +0100654 else if (whole->bd_holder == bd_may_claim)
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900655 return true; /* is a partition of a device that is being partitioned */
656 else if (whole->bd_holder != NULL)
657 return false; /* is a partition of a held device */
658 else
659 return true; /* is a partition of an un-held device */
660}
661
662/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900663 * bd_prepare_to_claim - prepare to claim a block device
664 * @bdev: block device of interest
665 * @whole: the whole device containing @bdev, may equal @bdev
666 * @holder: holder trying to claim @bdev
667 *
668 * Prepare to claim @bdev. This function fails if @bdev is already
669 * claimed by another holder and waits if another claiming is in
670 * progress. This function doesn't actually claim. On successful
671 * return, the caller has ownership of bd_claiming and bd_holder[s].
672 *
673 * CONTEXT:
674 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
675 * it multiple times.
676 *
677 * RETURNS:
678 * 0 if @bdev can be claimed, -EBUSY otherwise.
679 */
680static int bd_prepare_to_claim(struct block_device *bdev,
681 struct block_device *whole, void *holder)
682{
683retry:
684 /* if someone else claimed, fail */
685 if (!bd_may_claim(bdev, whole, holder))
686 return -EBUSY;
687
Tejun Heoe75aa852010-08-04 17:59:39 +0200688 /* if claiming is already in progress, wait for it to finish */
689 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900690 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
691 DEFINE_WAIT(wait);
692
693 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
694 spin_unlock(&bdev_lock);
695 schedule();
696 finish_wait(wq, &wait);
697 spin_lock(&bdev_lock);
698 goto retry;
699 }
700
701 /* yay, all mine */
702 return 0;
703}
704
705/**
706 * bd_start_claiming - start claiming a block device
707 * @bdev: block device of interest
708 * @holder: holder trying to claim @bdev
709 *
710 * @bdev is about to be opened exclusively. Check @bdev can be opened
711 * exclusively and mark that an exclusive open is in progress. Each
712 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000713 * either bd_finish_claiming() or bd_abort_claiming() (which do not
714 * fail).
715 *
716 * This function is used to gain exclusive access to the block device
717 * without actually causing other exclusive open attempts to fail. It
718 * should be used when the open sequence itself requires exclusive
719 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900720 *
721 * CONTEXT:
722 * Might sleep.
723 *
724 * RETURNS:
725 * Pointer to the block device containing @bdev on success, ERR_PTR()
726 * value on failure.
727 */
728static struct block_device *bd_start_claiming(struct block_device *bdev,
729 void *holder)
730{
731 struct gendisk *disk;
732 struct block_device *whole;
733 int partno, err;
734
735 might_sleep();
736
737 /*
738 * @bdev might not have been initialized properly yet, look up
739 * and grab the outer block device the hard way.
740 */
741 disk = get_gendisk(bdev->bd_dev, &partno);
742 if (!disk)
743 return ERR_PTR(-ENXIO);
744
Tejun Heod4c208b2011-06-13 12:45:48 +0200745 /*
746 * Normally, @bdev should equal what's returned from bdget_disk()
747 * if partno is 0; however, some drivers (floppy) use multiple
748 * bdev's for the same physical device and @bdev may be one of the
749 * aliases. Keep @bdev if partno is 0. This means claimer
750 * tracking is broken for those devices but it has always been that
751 * way.
752 */
753 if (partno)
754 whole = bdget_disk(disk, 0);
755 else
756 whole = bdgrab(bdev);
757
Nick Piggincf342572010-05-26 01:50:21 +1000758 module_put(disk->fops->owner);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900759 put_disk(disk);
760 if (!whole)
761 return ERR_PTR(-ENOMEM);
762
763 /* prepare to claim, if successful, mark claiming in progress */
764 spin_lock(&bdev_lock);
765
766 err = bd_prepare_to_claim(bdev, whole, holder);
767 if (err == 0) {
768 whole->bd_claiming = holder;
769 spin_unlock(&bdev_lock);
770 return whole;
771 } else {
772 spin_unlock(&bdev_lock);
773 bdput(whole);
774 return ERR_PTR(err);
775 }
776}
777
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800778#ifdef CONFIG_SYSFS
Tejun Heo49731ba2011-01-14 18:43:57 +0100779struct bd_holder_disk {
780 struct list_head list;
781 struct gendisk *disk;
782 int refcnt;
783};
784
785static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
786 struct gendisk *disk)
787{
788 struct bd_holder_disk *holder;
789
790 list_for_each_entry(holder, &bdev->bd_holder_disks, list)
791 if (holder->disk == disk)
792 return holder;
793 return NULL;
794}
795
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700796static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800797{
Andrew Morton4d7dd8fd2006-09-29 01:58:56 -0700798 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800799}
800
801static void del_symlink(struct kobject *from, struct kobject *to)
802{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800803 sysfs_remove_link(from, kobject_name(to));
804}
805
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800806/**
Tejun Heoe09b4572010-11-13 11:55:17 +0100807 * bd_link_disk_holder - create symlinks between holding disk and slave bdev
808 * @bdev: the claimed slave bdev
809 * @disk: the holding disk
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500810 *
Tejun Heo49731ba2011-01-14 18:43:57 +0100811 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
812 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100813 * This functions creates the following sysfs symlinks.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500814 *
Tejun Heoe09b4572010-11-13 11:55:17 +0100815 * - from "slaves" directory of the holder @disk to the claimed @bdev
816 * - from "holders" directory of the @bdev to the holder @disk
817 *
818 * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
819 * passed to bd_link_disk_holder(), then:
820 *
821 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
822 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
823 *
824 * The caller must have claimed @bdev before calling this function and
825 * ensure that both @bdev and @disk are valid during the creation and
826 * lifetime of these symlinks.
827 *
828 * CONTEXT:
829 * Might sleep.
830 *
831 * RETURNS:
832 * 0 on success, -errno on failure.
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500833 */
Tejun Heoe09b4572010-11-13 11:55:17 +0100834int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500835{
Tejun Heo49731ba2011-01-14 18:43:57 +0100836 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100837 int ret = 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800838
Peter Zijlstra2e7b6512006-12-08 02:36:13 -0800839 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500840
Tejun Heo49731ba2011-01-14 18:43:57 +0100841 WARN_ON_ONCE(!bdev->bd_holder);
Johannes Weiner4e916722007-07-15 23:41:25 -0700842
Tejun Heoe09b4572010-11-13 11:55:17 +0100843 /* FIXME: remove the following once add_disk() handles errors */
844 if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
845 goto out_unlock;
Johannes Weiner4e916722007-07-15 23:41:25 -0700846
Tejun Heo49731ba2011-01-14 18:43:57 +0100847 holder = bd_find_holder_disk(bdev, disk);
848 if (holder) {
849 holder->refcnt++;
Tejun Heoe09b4572010-11-13 11:55:17 +0100850 goto out_unlock;
851 }
852
Tejun Heo49731ba2011-01-14 18:43:57 +0100853 holder = kzalloc(sizeof(*holder), GFP_KERNEL);
854 if (!holder) {
855 ret = -ENOMEM;
856 goto out_unlock;
857 }
858
859 INIT_LIST_HEAD(&holder->list);
860 holder->disk = disk;
861 holder->refcnt = 1;
862
863 ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
864 if (ret)
865 goto out_free;
866
867 ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
868 if (ret)
869 goto out_del;
Tejun Heoe7407d12011-02-24 09:56:32 +0100870 /*
871 * bdev could be deleted beneath us which would implicitly destroy
872 * the holder directory. Hold on to it.
873 */
874 kobject_get(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100875
876 list_add(&holder->list, &bdev->bd_holder_disks);
877 goto out_unlock;
878
879out_del:
880 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
881out_free:
882 kfree(holder);
Tejun Heoe09b4572010-11-13 11:55:17 +0100883out_unlock:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -0800884 mutex_unlock(&bdev->bd_mutex);
Tejun Heoe09b4572010-11-13 11:55:17 +0100885 return ret;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800886}
Tejun Heoe09b4572010-11-13 11:55:17 +0100887EXPORT_SYMBOL_GPL(bd_link_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800888
Tejun Heo49731ba2011-01-14 18:43:57 +0100889/**
890 * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
891 * @bdev: the calimed slave bdev
892 * @disk: the holding disk
893 *
894 * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
895 *
896 * CONTEXT:
897 * Might sleep.
898 */
899void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800900{
Tejun Heo49731ba2011-01-14 18:43:57 +0100901 struct bd_holder_disk *holder;
Tejun Heoe09b4572010-11-13 11:55:17 +0100902
Tejun Heo49731ba2011-01-14 18:43:57 +0100903 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800904
Tejun Heo49731ba2011-01-14 18:43:57 +0100905 holder = bd_find_holder_disk(bdev, disk);
906
907 if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
908 del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
909 del_symlink(bdev->bd_part->holder_dir,
910 &disk_to_dev(disk)->kobj);
Tejun Heoe7407d12011-02-24 09:56:32 +0100911 kobject_put(bdev->bd_part->holder_dir);
Tejun Heo49731ba2011-01-14 18:43:57 +0100912 list_del_init(&holder->list);
913 kfree(holder);
914 }
915
916 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800917}
Tejun Heo49731ba2011-01-14 18:43:57 +0100918EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800919#endif
920
Andrew Patterson0c002c22008-09-04 14:27:20 -0600921/**
Andrew Patterson56ade442008-09-04 14:27:40 -0600922 * flush_disk - invalidates all buffer-cache entries on a disk
923 *
924 * @bdev: struct block device to be flushed
Randy Dunlape6eb5ce2011-02-26 10:54:00 -0800925 * @kill_dirty: flag to guide handling of dirty inodes
Andrew Patterson56ade442008-09-04 14:27:40 -0600926 *
927 * Invalidates all buffer-cache entries on a disk. It should be called
928 * when a disk has been changed -- either by a media change or online
929 * resize.
930 */
NeilBrown93b270f2011-02-24 17:25:47 +1100931static void flush_disk(struct block_device *bdev, bool kill_dirty)
Andrew Patterson56ade442008-09-04 14:27:40 -0600932{
NeilBrown93b270f2011-02-24 17:25:47 +1100933 if (__invalidate_device(bdev, kill_dirty)) {
Andrew Patterson56ade442008-09-04 14:27:40 -0600934 char name[BDEVNAME_SIZE] = "";
935
936 if (bdev->bd_disk)
937 disk_name(bdev->bd_disk, 0, name);
938 printk(KERN_WARNING "VFS: busy inodes on changed media or "
939 "resized disk %s\n", name);
940 }
941
942 if (!bdev->bd_disk)
943 return;
Tejun Heod27769e2011-08-23 20:01:04 +0200944 if (disk_part_scan_enabled(bdev->bd_disk))
Andrew Patterson56ade442008-09-04 14:27:40 -0600945 bdev->bd_invalidated = 1;
946}
947
948/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200949 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600950 * @disk: struct gendisk to check
951 * @bdev: struct bdev to adjust.
952 *
953 * This routine checks to see if the bdev size does not match the disk size
954 * and adjusts it if it differs.
955 */
956void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
957{
958 loff_t disk_size, bdev_size;
959
960 disk_size = (loff_t)get_capacity(disk) << 9;
961 bdev_size = i_size_read(bdev->bd_inode);
962 if (disk_size != bdev_size) {
963 char name[BDEVNAME_SIZE];
964
965 disk_name(disk, 0, name);
966 printk(KERN_INFO
967 "%s: detected capacity change from %lld to %lld\n",
968 name, bdev_size, disk_size);
969 i_size_write(bdev->bd_inode, disk_size);
NeilBrown93b270f2011-02-24 17:25:47 +1100970 flush_disk(bdev, false);
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600971 }
972}
973EXPORT_SYMBOL(check_disk_size_change);
974
975/**
Randy Dunlap57d1b532008-10-09 10:42:38 +0200976 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -0600977 * @disk: struct gendisk to be revalidated
978 *
979 * This routine is a wrapper for lower-level driver's revalidate_disk
980 * call-backs. It is used to do common pre and post operations needed
981 * for all revalidate_disk operations.
982 */
983int revalidate_disk(struct gendisk *disk)
984{
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600985 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -0600986 int ret = 0;
987
988 if (disk->fops->revalidate_disk)
989 ret = disk->fops->revalidate_disk(disk);
990
Andrew Pattersonc3279d12008-09-04 14:27:25 -0600991 bdev = bdget_disk(disk, 0);
992 if (!bdev)
993 return ret;
994
995 mutex_lock(&bdev->bd_mutex);
996 check_disk_size_change(disk, bdev);
997 mutex_unlock(&bdev->bd_mutex);
998 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -0600999 return ret;
1000}
1001EXPORT_SYMBOL(revalidate_disk);
1002
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003/*
1004 * This routine checks whether a removable media has been changed,
1005 * and invalidates all buffer-cache-entries in that case. This
1006 * is a relatively slow routine, so we have to try to minimize using
1007 * it. Thus it is called only upon a 'mount' or 'open'. This
1008 * is the best way of combining speed and utility, I think.
1009 * People changing diskettes in the middle of an operation deserve
1010 * to lose :-)
1011 */
1012int check_disk_change(struct block_device *bdev)
1013{
1014 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001015 const struct block_device_operations *bdops = disk->fops;
Tejun Heo77ea8872010-12-08 20:57:37 +01001016 unsigned int events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
Tejun Heo77ea8872010-12-08 20:57:37 +01001018 events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1019 DISK_EVENT_EJECT_REQUEST);
1020 if (!(events & DISK_EVENT_MEDIA_CHANGE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 return 0;
1022
NeilBrown93b270f2011-02-24 17:25:47 +11001023 flush_disk(bdev, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 if (bdops->revalidate_disk)
1025 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 return 1;
1027}
1028
1029EXPORT_SYMBOL(check_disk_change);
1030
1031void bd_set_size(struct block_device *bdev, loff_t size)
1032{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001033 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035 bdev->bd_inode->i_size = size;
1036 while (bsize < PAGE_CACHE_SIZE) {
1037 if (size & bsize)
1038 break;
1039 bsize <<= 1;
1040 }
1041 bdev->bd_block_size = bsize;
1042 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1043}
1044EXPORT_SYMBOL(bd_set_size);
1045
Al Viro9a1c3542008-02-22 20:40:24 -05001046static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001047
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001048/*
1049 * bd_mutex locking:
1050 *
1051 * mutex_lock(part->bd_mutex)
1052 * mutex_lock_nested(whole->bd_mutex, 1)
1053 */
1054
Al Viro572c4892007-10-08 13:24:05 -04001055static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 struct gendisk *disk;
Tejun Heo523e1d32011-10-19 14:31:07 +02001058 struct module *owner;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001059 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001060 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001061 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062
Al Viro572c4892007-10-08 13:24:05 -04001063 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001064 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001065 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001066 perm |= MAY_WRITE;
1067 /*
1068 * hooks: /n/, see "layering violations".
1069 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001070 if (!for_part) {
1071 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1072 if (ret != 0) {
1073 bdput(bdev);
1074 return ret;
1075 }
Al Viro82666022008-08-01 05:32:04 -04001076 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001077
NeilBrownd3374822009-01-09 08:31:10 +11001078 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001079
Tejun Heo89f97492008-11-05 10:21:06 +01001080 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001081 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001082 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001083 goto out;
Tejun Heo523e1d32011-10-19 14:31:07 +02001084 owner = disk->fops->owner;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085
Tejun Heo69e02c52011-03-09 19:54:27 +01001086 disk_block_events(disk);
NeilBrown6796bf52006-12-08 02:36:16 -08001087 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 if (!bdev->bd_openers) {
1089 bdev->bd_disk = disk;
Andi Kleen87192a22012-01-12 17:20:34 -08001090 bdev->bd_queue = disk->queue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001092 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001094
1095 ret = -ENXIO;
1096 bdev->bd_part = disk_get_part(disk, partno);
1097 if (!bdev->bd_part)
1098 goto out_clear;
1099
Tejun Heo1196f8b2011-04-21 20:54:45 +02001100 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001102 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001103 if (ret == -ERESTARTSYS) {
1104 /* Lost a race with 'disk' being
1105 * deleted, try again.
1106 * See md.c
1107 */
1108 disk_put_part(bdev->bd_part);
1109 bdev->bd_part = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001110 bdev->bd_disk = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001111 bdev->bd_queue = NULL;
NeilBrownd3374822009-01-09 08:31:10 +11001112 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001113 disk_unblock_events(disk);
Tejun Heo69e02c52011-03-09 19:54:27 +01001114 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001115 module_put(owner);
NeilBrownd3374822009-01-09 08:31:10 +11001116 goto restart;
1117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 }
Tejun Heo7e697232011-05-23 13:26:07 +02001119
1120 if (!ret && !bdev->bd_openers) {
1121 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1122 bdi = blk_get_backing_dev_info(bdev);
1123 if (bdi == NULL)
1124 bdi = &default_backing_dev_info;
1125 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
1126 }
1127
Tejun Heo1196f8b2011-04-21 20:54:45 +02001128 /*
1129 * If the device is invalidated, rescan partition
1130 * if open succeeded or failed with -ENOMEDIUM.
1131 * The latter is necessary to prevent ghost
1132 * partitions on a removed medium.
1133 */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001134 if (bdev->bd_invalidated) {
1135 if (!ret)
1136 rescan_partitions(disk, bdev);
1137 else if (ret == -ENOMEDIUM)
1138 invalidate_partitions(disk, bdev);
1139 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001140 if (ret)
1141 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 struct block_device *whole;
1144 whole = bdget_disk(disk, 0);
1145 ret = -ENOMEM;
1146 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001147 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001148 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001149 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001151 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001153 bdev_inode_switch_bdi(bdev->bd_inode,
1154 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001155 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001156 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001157 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001159 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
Tejun Heo89f97492008-11-05 10:21:06 +01001161 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 }
1163 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (bdev->bd_contains == bdev) {
Tejun Heo1196f8b2011-04-21 20:54:45 +02001165 ret = 0;
1166 if (bdev->bd_disk->fops->open)
Al Viro572c4892007-10-08 13:24:05 -04001167 ret = bdev->bd_disk->fops->open(bdev, mode);
Tejun Heo1196f8b2011-04-21 20:54:45 +02001168 /* the same as first opener case, read comment there */
Jun'ichi Nomurafe316bf2012-03-02 10:38:33 +01001169 if (bdev->bd_invalidated) {
1170 if (!ret)
1171 rescan_partitions(bdev->bd_disk, bdev);
1172 else if (ret == -ENOMEDIUM)
1173 invalidate_partitions(bdev->bd_disk, bdev);
1174 }
Tejun Heo1196f8b2011-04-21 20:54:45 +02001175 if (ret)
1176 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Tejun Heo69e02c52011-03-09 19:54:27 +01001178 /* only one opener holds refs to the module and disk */
Tejun Heo69e02c52011-03-09 19:54:27 +01001179 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001180 module_put(owner);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 }
1182 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001183 if (for_part)
1184 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001185 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001186 disk_unblock_events(disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 return 0;
1188
Tejun Heo0762b8b2008-08-25 19:56:12 +09001189 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001190 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001192 bdev->bd_part = NULL;
Andi Kleen87192a22012-01-12 17:20:34 -08001193 bdev->bd_queue = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001194 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001196 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001198 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001199 mutex_unlock(&bdev->bd_mutex);
Tejun Heo69e02c52011-03-09 19:54:27 +01001200 disk_unblock_events(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001201 put_disk(disk);
Tejun Heo523e1d32011-10-19 14:31:07 +02001202 module_put(owner);
Dan Carpenter4345cab2011-03-19 13:53:31 +01001203 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001204 bdput(bdev);
1205
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 return ret;
1207}
1208
Tejun Heod4d77622010-11-13 11:55:18 +01001209/**
1210 * blkdev_get - open a block device
1211 * @bdev: block_device to open
1212 * @mode: FMODE_* mask
1213 * @holder: exclusive holder identifier
1214 *
1215 * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is
1216 * open with exclusive access. Specifying %FMODE_EXCL with %NULL
1217 * @holder is invalid. Exclusive opens may nest for the same @holder.
1218 *
1219 * On success, the reference count of @bdev is unchanged. On failure,
1220 * @bdev is put.
1221 *
1222 * CONTEXT:
1223 * Might sleep.
1224 *
1225 * RETURNS:
1226 * 0 on success, -errno on failure.
1227 */
Tejun Heoe525fd82010-11-13 11:55:17 +01001228int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229{
Tejun Heoe525fd82010-11-13 11:55:17 +01001230 struct block_device *whole = NULL;
1231 int res;
1232
1233 WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1234
1235 if ((mode & FMODE_EXCL) && holder) {
1236 whole = bd_start_claiming(bdev, holder);
1237 if (IS_ERR(whole)) {
1238 bdput(bdev);
1239 return PTR_ERR(whole);
1240 }
1241 }
1242
1243 res = __blkdev_get(bdev, mode, 0);
1244
1245 if (whole) {
Tejun Heod4dc2102011-04-21 20:54:46 +02001246 struct gendisk *disk = whole->bd_disk;
1247
Tejun Heo6a027ef2010-11-13 11:55:17 +01001248 /* finish claiming */
Tejun Heo77ea8872010-12-08 20:57:37 +01001249 mutex_lock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001250 spin_lock(&bdev_lock);
1251
Tejun Heo77ea8872010-12-08 20:57:37 +01001252 if (!res) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001253 BUG_ON(!bd_may_claim(bdev, whole, holder));
1254 /*
1255 * Note that for a whole device bd_holders
1256 * will be incremented twice, and bd_holder
1257 * will be set to bd_may_claim before being
1258 * set to holder
1259 */
1260 whole->bd_holders++;
1261 whole->bd_holder = bd_may_claim;
1262 bdev->bd_holders++;
1263 bdev->bd_holder = holder;
1264 }
1265
1266 /* tell others that we're done */
1267 BUG_ON(whole->bd_claiming != holder);
1268 whole->bd_claiming = NULL;
1269 wake_up_bit(&whole->bd_claiming, 0);
1270
1271 spin_unlock(&bdev_lock);
Tejun Heo77ea8872010-12-08 20:57:37 +01001272
1273 /*
Tejun Heod4dc2102011-04-21 20:54:46 +02001274 * Block event polling for write claims if requested. Any
1275 * write holder makes the write_holder state stick until
1276 * all are released. This is good enough and tracking
1277 * individual writeable reference is too fragile given the
1278 * way @mode is used in blkdev_get/put().
Tejun Heo77ea8872010-12-08 20:57:37 +01001279 */
Tejun Heo4c49ff32011-06-01 08:27:41 +02001280 if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1281 (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
Tejun Heo77ea8872010-12-08 20:57:37 +01001282 bdev->bd_write_holder = true;
Tejun Heod4dc2102011-04-21 20:54:46 +02001283 disk_block_events(disk);
Tejun Heo77ea8872010-12-08 20:57:37 +01001284 }
1285
1286 mutex_unlock(&bdev->bd_mutex);
Tejun Heo6a027ef2010-11-13 11:55:17 +01001287 bdput(whole);
Tejun Heoe525fd82010-11-13 11:55:17 +01001288 }
1289
1290 return res;
NeilBrown37be4122006-12-08 02:36:16 -08001291}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292EXPORT_SYMBOL(blkdev_get);
1293
Tejun Heod4d77622010-11-13 11:55:18 +01001294/**
1295 * blkdev_get_by_path - open a block device by name
1296 * @path: path to the block device to open
1297 * @mode: FMODE_* mask
1298 * @holder: exclusive holder identifier
1299 *
1300 * Open the blockdevice described by the device file at @path. @mode
1301 * and @holder are identical to blkdev_get().
1302 *
1303 * On success, the returned block_device has reference count of one.
1304 *
1305 * CONTEXT:
1306 * Might sleep.
1307 *
1308 * RETURNS:
1309 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1310 */
1311struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1312 void *holder)
1313{
1314 struct block_device *bdev;
1315 int err;
1316
1317 bdev = lookup_bdev(path);
1318 if (IS_ERR(bdev))
1319 return bdev;
1320
1321 err = blkdev_get(bdev, mode, holder);
1322 if (err)
1323 return ERR_PTR(err);
1324
Chuck Ebberte51900f2011-02-16 18:11:53 -05001325 if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1326 blkdev_put(bdev, mode);
1327 return ERR_PTR(-EACCES);
1328 }
1329
Tejun Heod4d77622010-11-13 11:55:18 +01001330 return bdev;
1331}
1332EXPORT_SYMBOL(blkdev_get_by_path);
1333
1334/**
1335 * blkdev_get_by_dev - open a block device by device number
1336 * @dev: device number of block device to open
1337 * @mode: FMODE_* mask
1338 * @holder: exclusive holder identifier
1339 *
1340 * Open the blockdevice described by device number @dev. @mode and
1341 * @holder are identical to blkdev_get().
1342 *
1343 * Use it ONLY if you really do not have anything better - i.e. when
1344 * you are behind a truly sucky interface and all you are given is a
1345 * device number. _Never_ to be used for internal purposes. If you
1346 * ever need it - reconsider your API.
1347 *
1348 * On success, the returned block_device has reference count of one.
1349 *
1350 * CONTEXT:
1351 * Might sleep.
1352 *
1353 * RETURNS:
1354 * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1355 */
1356struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1357{
1358 struct block_device *bdev;
1359 int err;
1360
1361 bdev = bdget(dev);
1362 if (!bdev)
1363 return ERR_PTR(-ENOMEM);
1364
1365 err = blkdev_get(bdev, mode, holder);
1366 if (err)
1367 return ERR_PTR(err);
1368
1369 return bdev;
1370}
1371EXPORT_SYMBOL(blkdev_get_by_dev);
1372
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373static int blkdev_open(struct inode * inode, struct file * filp)
1374{
1375 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 /*
1378 * Preserve backwards compatibility and allow large file access
1379 * even if userspace doesn't ask for it explicitly. Some mkfs
1380 * binary needs it. We might want to drop this workaround
1381 * during an unstable branch.
1382 */
1383 filp->f_flags |= O_LARGEFILE;
1384
Al Viro572c4892007-10-08 13:24:05 -04001385 if (filp->f_flags & O_NDELAY)
1386 filp->f_mode |= FMODE_NDELAY;
1387 if (filp->f_flags & O_EXCL)
1388 filp->f_mode |= FMODE_EXCL;
1389 if ((filp->f_flags & O_ACCMODE) == 3)
1390 filp->f_mode |= FMODE_WRITE_IOCTL;
1391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001393 if (bdev == NULL)
1394 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395
Al Viro572c4892007-10-08 13:24:05 -04001396 filp->f_mapping = bdev->bd_inode->i_mapping;
1397
Tejun Heoe525fd82010-11-13 11:55:17 +01001398 return blkdev_get(bdev, filp->f_mode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399}
1400
Al Viro9a1c3542008-02-22 20:40:24 -05001401static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001402{
1403 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001404 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001405 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001406
NeilBrown6796bf52006-12-08 02:36:16 -08001407 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001408 if (for_part)
1409 bdev->bd_part_count--;
1410
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001411 if (!--bdev->bd_openers) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001412 WARN_ON_ONCE(bdev->bd_holders);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001413 sync_blockdev(bdev);
1414 kill_bdev(bdev);
NeilBrown94007752011-09-10 17:20:21 +10001415 /* ->release can cause the old bdi to disappear,
1416 * so must switch it out first
1417 */
1418 bdev_inode_switch_bdi(bdev->bd_inode,
1419 &default_backing_dev_info);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001420 }
1421 if (bdev->bd_contains == bdev) {
1422 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001423 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001424 }
1425 if (!bdev->bd_openers) {
1426 struct module *owner = disk->fops->owner;
1427
Tejun Heo0762b8b2008-08-25 19:56:12 +09001428 disk_put_part(bdev->bd_part);
1429 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001430 bdev->bd_disk = NULL;
NeilBrown37be4122006-12-08 02:36:16 -08001431 if (bdev != bdev->bd_contains)
1432 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001433 bdev->bd_contains = NULL;
Tejun Heo523e1d32011-10-19 14:31:07 +02001434
1435 put_disk(disk);
1436 module_put(owner);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001437 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001438 mutex_unlock(&bdev->bd_mutex);
1439 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001440 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001441 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001442 return ret;
1443}
1444
Al Viro9a1c3542008-02-22 20:40:24 -05001445int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001446{
Tejun Heo85ef06d2011-07-01 16:17:47 +02001447 mutex_lock(&bdev->bd_mutex);
1448
Tejun Heoe525fd82010-11-13 11:55:17 +01001449 if (mode & FMODE_EXCL) {
Tejun Heo6a027ef2010-11-13 11:55:17 +01001450 bool bdev_free;
1451
1452 /*
1453 * Release a claim on the device. The holder fields
1454 * are protected with bdev_lock. bd_mutex is to
1455 * synchronize disk_holder unlinking.
1456 */
Tejun Heo6a027ef2010-11-13 11:55:17 +01001457 spin_lock(&bdev_lock);
1458
1459 WARN_ON_ONCE(--bdev->bd_holders < 0);
1460 WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1461
1462 /* bd_contains might point to self, check in a separate step */
1463 if ((bdev_free = !bdev->bd_holders))
1464 bdev->bd_holder = NULL;
1465 if (!bdev->bd_contains->bd_holders)
1466 bdev->bd_contains->bd_holder = NULL;
1467
1468 spin_unlock(&bdev_lock);
1469
Tejun Heo77ea8872010-12-08 20:57:37 +01001470 /*
1471 * If this was the last claim, remove holder link and
1472 * unblock evpoll if it was a write holder.
1473 */
Tejun Heo85ef06d2011-07-01 16:17:47 +02001474 if (bdev_free && bdev->bd_write_holder) {
1475 disk_unblock_events(bdev->bd_disk);
1476 bdev->bd_write_holder = false;
Tejun Heo77ea8872010-12-08 20:57:37 +01001477 }
Tejun Heo69362172011-03-09 19:54:27 +01001478 }
Tejun Heo77ea8872010-12-08 20:57:37 +01001479
Tejun Heo85ef06d2011-07-01 16:17:47 +02001480 /*
1481 * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1482 * event. This is to ensure detection of media removal commanded
1483 * from userland - e.g. eject(1).
1484 */
1485 disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1486
1487 mutex_unlock(&bdev->bd_mutex);
1488
Al Viro9a1c3542008-02-22 20:40:24 -05001489 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001490}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001491EXPORT_SYMBOL(blkdev_put);
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493static int blkdev_close(struct inode * inode, struct file * filp)
1494{
1495 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
Tejun Heoe525fd82010-11-13 11:55:17 +01001496
Al Viro9a1c3542008-02-22 20:40:24 -05001497 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498}
1499
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001500static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501{
Al Viro56b26ad2008-09-19 03:17:36 -04001502 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1503 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001504
1505 /*
1506 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1507 * to updated it before every ioctl.
1508 */
Al Viro56b26ad2008-09-19 03:17:36 -04001509 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001510 mode |= FMODE_NDELAY;
1511 else
1512 mode &= ~FMODE_NDELAY;
1513
Al Viro56b26ad2008-09-19 03:17:36 -04001514 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001517/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001518 * Write data to the block device. Only intended for the block device itself
1519 * and the raw driver which basically is a fake block device.
1520 *
1521 * Does not take i_mutex for the write and thus is not for general purpose
1522 * use.
1523 */
1524ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1525 unsigned long nr_segs, loff_t pos)
1526{
1527 struct file *file = iocb->ki_filp;
Jianpeng Ma53362a02012-08-02 09:50:39 +02001528 struct blk_plug plug;
Christoph Hellwigeef99382009-08-20 17:43:41 +02001529 ssize_t ret;
1530
1531 BUG_ON(iocb->ki_pos != pos);
1532
Jianpeng Ma53362a02012-08-02 09:50:39 +02001533 blk_start_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001534 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1535 if (ret > 0 || ret == -EIOCBQUEUED) {
1536 ssize_t err;
1537
1538 err = generic_write_sync(file, pos, ret);
1539 if (err < 0 && ret > 0)
1540 ret = err;
1541 }
Jianpeng Ma53362a02012-08-02 09:50:39 +02001542 blk_finish_plug(&plug);
Christoph Hellwigeef99382009-08-20 17:43:41 +02001543 return ret;
1544}
1545EXPORT_SYMBOL_GPL(blkdev_aio_write);
1546
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001547static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov,
1548 unsigned long nr_segs, loff_t pos)
1549{
1550 struct file *file = iocb->ki_filp;
1551 struct inode *bd_inode = file->f_mapping->host;
1552 loff_t size = i_size_read(bd_inode);
1553
1554 if (pos >= size)
1555 return 0;
1556
1557 size -= pos;
1558 if (size < INT_MAX)
1559 nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size);
1560 return generic_file_aio_read(iocb, iov, nr_segs, pos);
1561}
1562
Christoph Hellwigeef99382009-08-20 17:43:41 +02001563/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001564 * Try to release a page associated with block device when the system
1565 * is under memory pressure.
1566 */
1567static int blkdev_releasepage(struct page *page, gfp_t wait)
1568{
1569 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1570
1571 if (super && super->s_op->bdev_try_to_free_page)
1572 return super->s_op->bdev_try_to_free_page(super, page, wait);
1573
1574 return try_to_free_buffers(page);
1575}
1576
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001577static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 .readpage = blkdev_readpage,
1579 .writepage = blkdev_writepage,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001580 .write_begin = blkdev_write_begin,
1581 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001583 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 .direct_IO = blkdev_direct_IO,
1585};
1586
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001587const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 .open = blkdev_open,
1589 .release = blkdev_close,
1590 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001591 .read = do_sync_read,
1592 .write = do_sync_write,
Linus Torvalds684c9aa2012-12-07 16:48:39 -08001593 .aio_read = blkdev_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001594 .aio_write = blkdev_aio_write,
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001595 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001596 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001597 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598#ifdef CONFIG_COMPAT
1599 .compat_ioctl = compat_blkdev_ioctl,
1600#endif
Linus Torvalds1e8b3332012-11-29 10:49:50 -08001601 .splice_read = generic_file_splice_read,
1602 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603};
1604
1605int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1606{
1607 int res;
1608 mm_segment_t old_fs = get_fs();
1609 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001610 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 set_fs(old_fs);
1612 return res;
1613}
1614
1615EXPORT_SYMBOL(ioctl_by_bdev);
1616
1617/**
1618 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001619 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001621 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 * namespace if possible and return it. Return ERR_PTR(error)
1623 * otherwise.
1624 */
Al Viro421748e2008-08-02 01:04:36 -04001625struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626{
1627 struct block_device *bdev;
1628 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001629 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 int error;
1631
Al Viro421748e2008-08-02 01:04:36 -04001632 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 return ERR_PTR(-EINVAL);
1634
Al Viro421748e2008-08-02 01:04:36 -04001635 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (error)
1637 return ERR_PTR(error);
1638
Al Viro421748e2008-08-02 01:04:36 -04001639 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 error = -ENOTBLK;
1641 if (!S_ISBLK(inode->i_mode))
1642 goto fail;
1643 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001644 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 goto fail;
1646 error = -ENOMEM;
1647 bdev = bd_acquire(inode);
1648 if (!bdev)
1649 goto fail;
1650out:
Al Viro421748e2008-08-02 01:04:36 -04001651 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 return bdev;
1653fail:
1654 bdev = ERR_PTR(error);
1655 goto out;
1656}
Al Virod5686b42008-08-01 05:00:11 -04001657EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
NeilBrown93b270f2011-02-24 17:25:47 +11001659int __invalidate_device(struct block_device *bdev, bool kill_dirty)
David Howellsb71e8a42006-08-29 19:06:11 +01001660{
1661 struct super_block *sb = get_super(bdev);
1662 int res = 0;
1663
1664 if (sb) {
1665 /*
1666 * no need to lock the super, get_super holds the
1667 * read mutex so the filesystem cannot go away
1668 * under us (->put_super runs with the write lock
1669 * hold).
1670 */
1671 shrink_dcache_sb(sb);
NeilBrown93b270f2011-02-24 17:25:47 +11001672 res = invalidate_inodes(sb, kill_dirty);
David Howellsb71e8a42006-08-29 19:06:11 +01001673 drop_super(sb);
1674 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001675 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001676 return res;
1677}
1678EXPORT_SYMBOL(__invalidate_device);
Jan Kara5c0d6b62012-07-03 16:45:31 +02001679
1680void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
1681{
1682 struct inode *inode, *old_inode = NULL;
1683
1684 spin_lock(&inode_sb_list_lock);
1685 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
1686 struct address_space *mapping = inode->i_mapping;
1687
1688 spin_lock(&inode->i_lock);
1689 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
1690 mapping->nrpages == 0) {
1691 spin_unlock(&inode->i_lock);
1692 continue;
1693 }
1694 __iget(inode);
1695 spin_unlock(&inode->i_lock);
1696 spin_unlock(&inode_sb_list_lock);
1697 /*
1698 * We hold a reference to 'inode' so it couldn't have been
1699 * removed from s_inodes list while we dropped the
1700 * inode_sb_list_lock. We cannot iput the inode now as we can
1701 * be holding the last reference and we cannot iput it under
1702 * inode_sb_list_lock. So we keep the reference and iput it
1703 * later.
1704 */
1705 iput(old_inode);
1706 old_inode = inode;
1707
1708 func(I_BDEV(inode), arg);
1709
1710 spin_lock(&inode_sb_list_lock);
1711 }
1712 spin_unlock(&inode_sb_list_lock);
1713 iput(old_inode);
1714}