blob: 06e8ff12b97c8fa74d4b2eb05c900801b3adb150 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/block_dev.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
6 */
7
Linus Torvalds1da177e2005-04-16 15:20:36 -07008#include <linux/init.h>
9#include <linux/mm.h>
10#include <linux/fcntl.h>
11#include <linux/slab.h>
12#include <linux/kmod.h>
13#include <linux/major.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/smp_lock.h>
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -070015#include <linux/device_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/highmem.h>
17#include <linux/blkdev.h>
18#include <linux/module.h>
19#include <linux/blkpg.h>
20#include <linux/buffer_head.h>
Nick Piggin585d3bc2009-02-25 10:44:19 +010021#include <linux/pagevec.h>
David Howells811d7362006-08-29 19:06:09 +010022#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/mpage.h>
24#include <linux/mount.h>
25#include <linux/uio.h>
26#include <linux/namei.h>
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070027#include <linux/log2.h>
Catalin Marinas2e1483c2009-06-11 13:24:13 +010028#include <linux/kmemleak.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
32struct bdev_inode {
33 struct block_device bdev;
34 struct inode vfs_inode;
35};
36
Adrian Bunk4c54ac62008-02-18 13:48:31 +010037static const struct address_space_operations def_blk_aops;
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039static inline struct bdev_inode *BDEV_I(struct inode *inode)
40{
41 return container_of(inode, struct bdev_inode, vfs_inode);
42}
43
44inline struct block_device *I_BDEV(struct inode *inode)
45{
46 return &BDEV_I(inode)->bdev;
47}
48
49EXPORT_SYMBOL(I_BDEV);
50
Dave Chinnera5491e02010-10-21 11:49:26 +110051/*
52 * move the inode from it's current bdi to the a new bdi. if the inode is dirty
53 * we need to move it onto the dirty list of @dst so that the inode is always
54 * on the right list.
55 */
56static void bdev_inode_switch_bdi(struct inode *inode,
57 struct backing_dev_info *dst)
58{
59 spin_lock(&inode_lock);
60 inode->i_data.backing_dev_info = dst;
61 if (inode->i_state & I_DIRTY)
Nick Piggin7ccf19a2010-10-21 11:49:30 +110062 list_move(&inode->i_wb_list, &dst->wb.b_dirty);
Dave Chinnera5491e02010-10-21 11:49:26 +110063 spin_unlock(&inode_lock);
64}
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066static sector_t max_block(struct block_device *bdev)
67{
68 sector_t retval = ~((sector_t)0);
69 loff_t sz = i_size_read(bdev->bd_inode);
70
71 if (sz) {
72 unsigned int size = block_size(bdev);
73 unsigned int sizebits = blksize_bits(size);
74 retval = (sz >> sizebits);
75 }
76 return retval;
77}
78
Peter Zijlstraf9a14392007-05-06 14:49:55 -070079/* Kill _all_ buffers and pagecache , dirty or not.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -070080static void kill_bdev(struct block_device *bdev)
81{
Peter Zijlstraf9a14392007-05-06 14:49:55 -070082 if (bdev->bd_inode->i_mapping->nrpages == 0)
83 return;
84 invalidate_bh_lrus();
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
86}
87
88int set_blocksize(struct block_device *bdev, int size)
89{
90 /* Size must be a power of two, and between 512 and PAGE_SIZE */
Vignesh Babu BM1368c4f2007-05-08 00:24:32 -070091 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 return -EINVAL;
93
94 /* Size cannot be smaller than the size supported by the device */
Martin K. Petersene1defc42009-05-22 17:17:49 -040095 if (size < bdev_logical_block_size(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 return -EINVAL;
97
98 /* Don't change the size if it is same as current */
99 if (bdev->bd_block_size != size) {
100 sync_blockdev(bdev);
101 bdev->bd_block_size = size;
102 bdev->bd_inode->i_blkbits = blksize_bits(size);
103 kill_bdev(bdev);
104 }
105 return 0;
106}
107
108EXPORT_SYMBOL(set_blocksize);
109
110int sb_set_blocksize(struct super_block *sb, int size)
111{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 if (set_blocksize(sb->s_bdev, size))
113 return 0;
114 /* If we get here, we know size is power of two
115 * and it's value is between 512 and PAGE_SIZE */
116 sb->s_blocksize = size;
Coywolf Qi Hunt38885bd2006-03-24 03:18:05 -0800117 sb->s_blocksize_bits = blksize_bits(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 return sb->s_blocksize;
119}
120
121EXPORT_SYMBOL(sb_set_blocksize);
122
123int sb_min_blocksize(struct super_block *sb, int size)
124{
Martin K. Petersene1defc42009-05-22 17:17:49 -0400125 int minsize = bdev_logical_block_size(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (size < minsize)
127 size = minsize;
128 return sb_set_blocksize(sb, size);
129}
130
131EXPORT_SYMBOL(sb_min_blocksize);
132
133static int
134blkdev_get_block(struct inode *inode, sector_t iblock,
135 struct buffer_head *bh, int create)
136{
137 if (iblock >= max_block(I_BDEV(inode))) {
138 if (create)
139 return -EIO;
140
141 /*
142 * for reads, we're just trying to fill a partial page.
143 * return a hole, they will have to call get_block again
144 * before they can fill it, and they will get -EIO at that
145 * time
146 */
147 return 0;
148 }
149 bh->b_bdev = I_BDEV(inode);
150 bh->b_blocknr = iblock;
151 set_buffer_mapped(bh);
152 return 0;
153}
154
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800155static int
156blkdev_get_blocks(struct inode *inode, sector_t iblock,
157 struct buffer_head *bh, int create)
158{
159 sector_t end_block = max_block(I_BDEV(inode));
160 unsigned long max_blocks = bh->b_size >> inode->i_blkbits;
161
162 if ((iblock + max_blocks) > end_block) {
163 max_blocks = end_block - iblock;
164 if ((long)max_blocks <= 0) {
165 if (create)
166 return -EIO; /* write fully beyond EOF */
167 /*
168 * It is a read which is fully beyond EOF. We return
169 * a !buffer_mapped buffer
170 */
171 max_blocks = 0;
172 }
173 }
174
175 bh->b_bdev = I_BDEV(inode);
176 bh->b_blocknr = iblock;
177 bh->b_size = max_blocks << inode->i_blkbits;
178 if (max_blocks)
179 set_buffer_mapped(bh);
180 return 0;
181}
182
183static ssize_t
184blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
185 loff_t offset, unsigned long nr_segs)
186{
187 struct file *file = iocb->ki_filp;
188 struct inode *inode = file->f_mapping->host;
189
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +0200190 return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset,
191 nr_segs, blkdev_get_blocks, NULL, NULL, 0);
Andrew Mortonb2e895d2007-02-03 01:14:01 -0800192}
193
Jan Kara5cee5812009-04-27 16:43:51 +0200194int __sync_blockdev(struct block_device *bdev, int wait)
195{
196 if (!bdev)
197 return 0;
198 if (!wait)
199 return filemap_flush(bdev->bd_inode->i_mapping);
200 return filemap_write_and_wait(bdev->bd_inode->i_mapping);
201}
202
Nick Piggin585d3bc2009-02-25 10:44:19 +0100203/*
204 * Write out and wait upon all the dirty data associated with a block
205 * device via its mapping. Does not take the superblock lock.
206 */
207int sync_blockdev(struct block_device *bdev)
208{
Jan Kara5cee5812009-04-27 16:43:51 +0200209 return __sync_blockdev(bdev, 1);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100210}
211EXPORT_SYMBOL(sync_blockdev);
212
213/*
214 * Write out and wait upon all dirty data associated with this
215 * device. Filesystem data as well as the underlying block
216 * device. Takes the superblock lock.
217 */
218int fsync_bdev(struct block_device *bdev)
219{
220 struct super_block *sb = get_super(bdev);
221 if (sb) {
Jan Kara60b06802009-04-27 16:43:53 +0200222 int res = sync_filesystem(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100223 drop_super(sb);
224 return res;
225 }
226 return sync_blockdev(bdev);
227}
Al Viro47e44912009-04-01 07:07:16 -0400228EXPORT_SYMBOL(fsync_bdev);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100229
230/**
231 * freeze_bdev -- lock a filesystem and force it into a consistent state
232 * @bdev: blockdevice to lock
233 *
Nick Piggin585d3bc2009-02-25 10:44:19 +0100234 * If a superblock is found on this device, we take the s_umount semaphore
235 * on it to make sure nobody unmounts until the snapshot creation is done.
236 * The reference counter (bd_fsfreeze_count) guarantees that only the last
237 * unfreeze process can unfreeze the frozen filesystem actually when multiple
238 * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
239 * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
240 * actually.
241 */
242struct super_block *freeze_bdev(struct block_device *bdev)
243{
244 struct super_block *sb;
245 int error = 0;
246
247 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200248 if (++bdev->bd_fsfreeze_count > 1) {
249 /*
250 * We don't even need to grab a reference - the first call
251 * to freeze_bdev grab an active reference and only the last
252 * thaw_bdev drops it.
253 */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100254 sb = get_super(bdev);
Christoph Hellwig45042302009-08-03 23:28:35 +0200255 drop_super(sb);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100256 mutex_unlock(&bdev->bd_fsfreeze_mutex);
257 return sb;
258 }
Nick Piggin585d3bc2009-02-25 10:44:19 +0100259
Christoph Hellwig45042302009-08-03 23:28:35 +0200260 sb = get_active_super(bdev);
261 if (!sb)
262 goto out;
Josef Bacik18e9e512010-03-23 10:34:56 -0400263 error = freeze_super(sb);
264 if (error) {
265 deactivate_super(sb);
266 bdev->bd_fsfreeze_count--;
Christoph Hellwig45042302009-08-03 23:28:35 +0200267 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Josef Bacik18e9e512010-03-23 10:34:56 -0400268 return ERR_PTR(error);
Nick Piggin585d3bc2009-02-25 10:44:19 +0100269 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400270 deactivate_super(sb);
Christoph Hellwig45042302009-08-03 23:28:35 +0200271 out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100272 sync_blockdev(bdev);
273 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200274 return sb; /* thaw_bdev releases s->s_umount */
Nick Piggin585d3bc2009-02-25 10:44:19 +0100275}
276EXPORT_SYMBOL(freeze_bdev);
277
278/**
279 * thaw_bdev -- unlock filesystem
280 * @bdev: blockdevice to unlock
281 * @sb: associated superblock
282 *
283 * Unlocks the filesystem and marks it writeable again after freeze_bdev().
284 */
285int thaw_bdev(struct block_device *bdev, struct super_block *sb)
286{
Christoph Hellwig45042302009-08-03 23:28:35 +0200287 int error = -EINVAL;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100288
289 mutex_lock(&bdev->bd_fsfreeze_mutex);
Christoph Hellwig45042302009-08-03 23:28:35 +0200290 if (!bdev->bd_fsfreeze_count)
Josef Bacik18e9e512010-03-23 10:34:56 -0400291 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100292
Christoph Hellwig45042302009-08-03 23:28:35 +0200293 error = 0;
294 if (--bdev->bd_fsfreeze_count > 0)
Josef Bacik18e9e512010-03-23 10:34:56 -0400295 goto out;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100296
Christoph Hellwig45042302009-08-03 23:28:35 +0200297 if (!sb)
Josef Bacik18e9e512010-03-23 10:34:56 -0400298 goto out;
Christoph Hellwig45042302009-08-03 23:28:35 +0200299
Josef Bacik18e9e512010-03-23 10:34:56 -0400300 error = thaw_super(sb);
301 if (error) {
302 bdev->bd_fsfreeze_count++;
303 mutex_unlock(&bdev->bd_fsfreeze_mutex);
304 return error;
Nick Piggin585d3bc2009-02-25 10:44:19 +0100305 }
Josef Bacik18e9e512010-03-23 10:34:56 -0400306out:
Nick Piggin585d3bc2009-02-25 10:44:19 +0100307 mutex_unlock(&bdev->bd_fsfreeze_mutex);
308 return 0;
309}
310EXPORT_SYMBOL(thaw_bdev);
311
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
313{
314 return block_write_full_page(page, blkdev_get_block, wbc);
315}
316
317static int blkdev_readpage(struct file * file, struct page * page)
318{
319 return block_read_full_page(page, blkdev_get_block);
320}
321
Nick Piggin6272b5a2007-10-16 01:25:04 -0700322static int blkdev_write_begin(struct file *file, struct address_space *mapping,
323 loff_t pos, unsigned len, unsigned flags,
324 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325{
Christoph Hellwig155130a2010-06-04 11:29:58 +0200326 return block_write_begin(mapping, pos, len, flags, pagep,
327 blkdev_get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
Nick Piggin6272b5a2007-10-16 01:25:04 -0700330static int blkdev_write_end(struct file *file, struct address_space *mapping,
331 loff_t pos, unsigned len, unsigned copied,
332 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333{
Nick Piggin6272b5a2007-10-16 01:25:04 -0700334 int ret;
335 ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
336
337 unlock_page(page);
338 page_cache_release(page);
339
340 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341}
342
343/*
344 * private llseek:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800345 * for a block special file file->f_path.dentry->d_inode->i_size is zero
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 * so we compute the size by hand (just as in block_read/write above)
347 */
348static loff_t block_llseek(struct file *file, loff_t offset, int origin)
349{
350 struct inode *bd_inode = file->f_mapping->host;
351 loff_t size;
352 loff_t retval;
353
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800354 mutex_lock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 size = i_size_read(bd_inode);
356
357 switch (origin) {
358 case 2:
359 offset += size;
360 break;
361 case 1:
362 offset += file->f_pos;
363 }
364 retval = -EINVAL;
365 if (offset >= 0 && offset <= size) {
366 if (offset != file->f_pos) {
367 file->f_pos = offset;
368 }
369 retval = offset;
370 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800371 mutex_unlock(&bd_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 return retval;
373}
374
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200375int blkdev_fsync(struct file *filp, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400377 struct inode *bd_inode = filp->f_mapping->host;
378 struct block_device *bdev = I_BDEV(bd_inode);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100379 int error;
380
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400381 /*
382 * There is no need to serialise calls to blkdev_issue_flush with
383 * i_mutex and doing so causes performance issues with concurrent
384 * O_SYNC writers to a block device.
385 */
386 mutex_unlock(&bd_inode->i_mutex);
387
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200388 error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100389 if (error == -EOPNOTSUPP)
390 error = 0;
Anton Blanchardb8af67e2010-04-23 13:18:06 -0400391
392 mutex_lock(&bd_inode->i_mutex);
393
Christoph Hellwigab0a9732009-10-29 14:14:04 +0100394 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
Andrew Mortonb1dd3b22010-04-06 14:35:00 -0700396EXPORT_SYMBOL(blkdev_fsync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398/*
399 * pseudo-fs
400 */
401
402static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
Christoph Lametere18b8902006-12-06 20:33:20 -0800403static struct kmem_cache * bdev_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
405static struct inode *bdev_alloc_inode(struct super_block *sb)
406{
Christoph Lametere94b1762006-12-06 20:33:17 -0800407 struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 if (!ei)
409 return NULL;
410 return &ei->vfs_inode;
411}
412
413static void bdev_destroy_inode(struct inode *inode)
414{
415 struct bdev_inode *bdi = BDEV_I(inode);
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 kmem_cache_free(bdev_cachep, bdi);
418}
419
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700420static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421{
422 struct bdev_inode *ei = (struct bdev_inode *) foo;
423 struct block_device *bdev = &ei->bdev;
424
Christoph Lametera35afb82007-05-16 22:10:57 -0700425 memset(bdev, 0, sizeof(*bdev));
426 mutex_init(&bdev->bd_mutex);
Christoph Lametera35afb82007-05-16 22:10:57 -0700427 INIT_LIST_HEAD(&bdev->bd_inodes);
428 INIT_LIST_HEAD(&bdev->bd_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800429#ifdef CONFIG_SYSFS
Christoph Lametera35afb82007-05-16 22:10:57 -0700430 INIT_LIST_HEAD(&bdev->bd_holder_list);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800431#endif
Christoph Lametera35afb82007-05-16 22:10:57 -0700432 inode_init_once(&ei->vfs_inode);
Takashi Satofcccf502009-01-09 16:40:59 -0800433 /* Initialize mutex for freeze. */
434 mutex_init(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435}
436
437static inline void __bd_forget(struct inode *inode)
438{
439 list_del_init(&inode->i_devices);
440 inode->i_bdev = NULL;
441 inode->i_mapping = &inode->i_data;
442}
443
Al Virob57922d2010-06-07 14:34:48 -0400444static void bdev_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
446 struct block_device *bdev = &BDEV_I(inode)->bdev;
447 struct list_head *p;
Al Virob57922d2010-06-07 14:34:48 -0400448 truncate_inode_pages(&inode->i_data, 0);
449 invalidate_inode_buffers(inode); /* is it needed here? */
450 end_writeback(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 spin_lock(&bdev_lock);
452 while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) {
453 __bd_forget(list_entry(p, struct inode, i_devices));
454 }
455 list_del_init(&bdev->bd_list);
456 spin_unlock(&bdev_lock);
457}
458
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800459static const struct super_operations bdev_sops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 .statfs = simple_statfs,
461 .alloc_inode = bdev_alloc_inode,
462 .destroy_inode = bdev_destroy_inode,
463 .drop_inode = generic_delete_inode,
Al Virob57922d2010-06-07 14:34:48 -0400464 .evict_inode = bdev_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465};
466
Al Viro51139ad2010-07-25 23:47:46 +0400467static struct dentry *bd_mount(struct file_system_type *fs_type,
468 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469{
Al Viro51139ad2010-07-25 23:47:46 +0400470 return mount_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
473static struct file_system_type bd_type = {
474 .name = "bdev",
Al Viro51139ad2010-07-25 23:47:46 +0400475 .mount = bd_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 .kill_sb = kill_anon_super,
477};
478
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800479struct super_block *blockdev_superblock __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481void __init bdev_cache_init(void)
482{
483 int err;
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800484 struct vfsmount *bd_mnt;
485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
Paul Jacksonfffb60f2006-03-24 03:16:06 -0800487 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
488 SLAB_MEM_SPREAD|SLAB_PANIC),
Paul Mundt20c2df82007-07-20 10:11:58 +0900489 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 err = register_filesystem(&bd_type);
491 if (err)
492 panic("Cannot register bdev pseudo-fs");
493 bd_mnt = kern_mount(&bd_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 if (IS_ERR(bd_mnt))
495 panic("Cannot create bdev pseudo-fs");
Catalin Marinas2e1483c2009-06-11 13:24:13 +0100496 /*
497 * This vfsmount structure is only used to obtain the
498 * blockdev_superblock, so tell kmemleak not to report it.
499 */
500 kmemleak_not_leak(bd_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */
502}
503
504/*
505 * Most likely _very_ bad one - but then it's hardly critical for small
506 * /dev and can be fixed when somebody will need really large one.
507 * Keep in mind that it will be fed through icache hash function too.
508 */
509static inline unsigned long hash(dev_t dev)
510{
511 return MAJOR(dev)+MINOR(dev);
512}
513
514static int bdev_test(struct inode *inode, void *data)
515{
516 return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
517}
518
519static int bdev_set(struct inode *inode, void *data)
520{
521 BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
522 return 0;
523}
524
525static LIST_HEAD(all_bdevs);
526
527struct block_device *bdget(dev_t dev)
528{
529 struct block_device *bdev;
530 struct inode *inode;
531
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800532 inode = iget5_locked(blockdev_superblock, hash(dev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 bdev_test, bdev_set, &dev);
534
535 if (!inode)
536 return NULL;
537
538 bdev = &BDEV_I(inode)->bdev;
539
540 if (inode->i_state & I_NEW) {
541 bdev->bd_contains = NULL;
542 bdev->bd_inode = inode;
543 bdev->bd_block_size = (1 << inode->i_blkbits);
544 bdev->bd_part_count = 0;
545 bdev->bd_invalidated = 0;
546 inode->i_mode = S_IFBLK;
547 inode->i_rdev = dev;
548 inode->i_bdev = bdev;
549 inode->i_data.a_ops = &def_blk_aops;
550 mapping_set_gfp_mask(&inode->i_data, GFP_USER);
551 inode->i_data.backing_dev_info = &default_backing_dev_info;
552 spin_lock(&bdev_lock);
553 list_add(&bdev->bd_list, &all_bdevs);
554 spin_unlock(&bdev_lock);
555 unlock_new_inode(inode);
556 }
557 return bdev;
558}
559
560EXPORT_SYMBOL(bdget);
561
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200562/**
563 * bdgrab -- Grab a reference to an already referenced block device
564 * @bdev: Block device to grab a reference to.
565 */
566struct block_device *bdgrab(struct block_device *bdev)
567{
Al Viro7de9c6e2010-10-23 11:11:40 -0400568 ihold(bdev->bd_inode);
Alan Jenkinsdddac6a2009-07-29 21:07:55 +0200569 return bdev;
570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572long nr_blockdev_pages(void)
573{
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700574 struct block_device *bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 long ret = 0;
576 spin_lock(&bdev_lock);
Matthias Kaehlcke203a2932007-07-15 23:40:23 -0700577 list_for_each_entry(bdev, &all_bdevs, bd_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 ret += bdev->bd_inode->i_mapping->nrpages;
579 }
580 spin_unlock(&bdev_lock);
581 return ret;
582}
583
584void bdput(struct block_device *bdev)
585{
586 iput(bdev->bd_inode);
587}
588
589EXPORT_SYMBOL(bdput);
590
591static struct block_device *bd_acquire(struct inode *inode)
592{
593 struct block_device *bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 spin_lock(&bdev_lock);
596 bdev = inode->i_bdev;
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700597 if (bdev) {
Al Viro7de9c6e2010-10-23 11:11:40 -0400598 ihold(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 spin_unlock(&bdev_lock);
600 return bdev;
601 }
602 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 bdev = bdget(inode->i_rdev);
605 if (bdev) {
606 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700607 if (!inode->i_bdev) {
608 /*
Al Viro7de9c6e2010-10-23 11:11:40 -0400609 * We take an additional reference to bd_inode,
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700610 * and it's released in clear_inode() of inode.
611 * So, we can access it via ->i_mapping always
612 * without igrab().
613 */
Al Viro7de9c6e2010-10-23 11:11:40 -0400614 ihold(bdev->bd_inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700615 inode->i_bdev = bdev;
616 inode->i_mapping = bdev->bd_inode->i_mapping;
617 list_add(&inode->i_devices, &bdev->bd_inodes);
618 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 spin_unlock(&bdev_lock);
620 }
621 return bdev;
622}
623
624/* Call when you free inode */
625
626void bd_forget(struct inode *inode)
627{
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700628 struct block_device *bdev = NULL;
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 spin_lock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700631 if (inode->i_bdev) {
Denis ChengRqc2acf7b2008-12-01 14:34:56 -0800632 if (!sb_is_blkdev_sb(inode->i_sb))
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700633 bdev = inode->i_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 __bd_forget(inode);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 spin_unlock(&bdev_lock);
OGAWA Hirofumi09d967c2006-06-22 14:47:21 -0700637
638 if (bdev)
639 iput(bdev->bd_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
641
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900642/**
643 * bd_may_claim - test whether a block device can be claimed
644 * @bdev: block device of interest
645 * @whole: whole block device containing @bdev, may equal @bdev
646 * @holder: holder trying to claim @bdev
647 *
648 * Test whther @bdev can be claimed by @holder.
649 *
650 * CONTEXT:
651 * spin_lock(&bdev_lock).
652 *
653 * RETURNS:
654 * %true if @bdev can be claimed, %false otherwise.
655 */
656static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
657 void *holder)
658{
659 if (bdev->bd_holder == holder)
660 return true; /* already a holder */
661 else if (bdev->bd_holder != NULL)
662 return false; /* held by someone else */
663 else if (bdev->bd_contains == bdev)
664 return true; /* is a whole device which isn't held */
665
666 else if (whole->bd_holder == bd_claim)
667 return true; /* is a partition of a device that is being partitioned */
668 else if (whole->bd_holder != NULL)
669 return false; /* is a partition of a held device */
670 else
671 return true; /* is a partition of an un-held device */
672}
673
674/**
Tejun Heo6b4517a2010-04-07 18:53:59 +0900675 * bd_prepare_to_claim - prepare to claim a block device
676 * @bdev: block device of interest
677 * @whole: the whole device containing @bdev, may equal @bdev
678 * @holder: holder trying to claim @bdev
679 *
680 * Prepare to claim @bdev. This function fails if @bdev is already
681 * claimed by another holder and waits if another claiming is in
682 * progress. This function doesn't actually claim. On successful
683 * return, the caller has ownership of bd_claiming and bd_holder[s].
684 *
685 * CONTEXT:
686 * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab
687 * it multiple times.
688 *
689 * RETURNS:
690 * 0 if @bdev can be claimed, -EBUSY otherwise.
691 */
692static int bd_prepare_to_claim(struct block_device *bdev,
693 struct block_device *whole, void *holder)
694{
695retry:
696 /* if someone else claimed, fail */
697 if (!bd_may_claim(bdev, whole, holder))
698 return -EBUSY;
699
Tejun Heoe75aa852010-08-04 17:59:39 +0200700 /* if claiming is already in progress, wait for it to finish */
701 if (whole->bd_claiming) {
Tejun Heo6b4517a2010-04-07 18:53:59 +0900702 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
703 DEFINE_WAIT(wait);
704
705 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
706 spin_unlock(&bdev_lock);
707 schedule();
708 finish_wait(wq, &wait);
709 spin_lock(&bdev_lock);
710 goto retry;
711 }
712
713 /* yay, all mine */
714 return 0;
715}
716
717/**
718 * bd_start_claiming - start claiming a block device
719 * @bdev: block device of interest
720 * @holder: holder trying to claim @bdev
721 *
722 * @bdev is about to be opened exclusively. Check @bdev can be opened
723 * exclusively and mark that an exclusive open is in progress. Each
724 * successful call to this function must be matched with a call to
Nick Pigginb0018362010-05-26 01:51:19 +1000725 * either bd_finish_claiming() or bd_abort_claiming() (which do not
726 * fail).
727 *
728 * This function is used to gain exclusive access to the block device
729 * without actually causing other exclusive open attempts to fail. It
730 * should be used when the open sequence itself requires exclusive
731 * access but may subsequently fail.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900732 *
733 * CONTEXT:
734 * Might sleep.
735 *
736 * RETURNS:
737 * Pointer to the block device containing @bdev on success, ERR_PTR()
738 * value on failure.
739 */
740static struct block_device *bd_start_claiming(struct block_device *bdev,
741 void *holder)
742{
743 struct gendisk *disk;
744 struct block_device *whole;
745 int partno, err;
746
747 might_sleep();
748
749 /*
750 * @bdev might not have been initialized properly yet, look up
751 * and grab the outer block device the hard way.
752 */
753 disk = get_gendisk(bdev->bd_dev, &partno);
754 if (!disk)
755 return ERR_PTR(-ENXIO);
756
757 whole = bdget_disk(disk, 0);
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
778/* releases bdev_lock */
779static void __bd_abort_claiming(struct block_device *whole, void *holder)
780{
781 BUG_ON(whole->bd_claiming != holder);
782 whole->bd_claiming = NULL;
783 wake_up_bit(&whole->bd_claiming, 0);
784
785 spin_unlock(&bdev_lock);
786 bdput(whole);
787}
788
789/**
790 * bd_abort_claiming - abort claiming a block device
791 * @whole: whole block device returned by bd_start_claiming()
792 * @holder: holder trying to claim @bdev
793 *
794 * Abort a claiming block started by bd_start_claiming(). Note that
795 * @whole is not the block device to be claimed but the whole device
796 * returned by bd_start_claiming().
797 *
798 * CONTEXT:
799 * Grabs and releases bdev_lock.
800 */
801static void bd_abort_claiming(struct block_device *whole, void *holder)
802{
803 spin_lock(&bdev_lock);
804 __bd_abort_claiming(whole, holder); /* releases bdev_lock */
805}
806
Nick Pigginb0018362010-05-26 01:51:19 +1000807/* increment holders when we have a legitimate claim. requires bdev_lock */
808static void __bd_claim(struct block_device *bdev, struct block_device *whole,
809 void *holder)
810{
811 /* note that for a whole device bd_holders
812 * will be incremented twice, and bd_holder will
813 * be set to bd_claim before being set to holder
814 */
815 whole->bd_holders++;
816 whole->bd_holder = bd_claim;
817 bdev->bd_holders++;
818 bdev->bd_holder = holder;
819}
820
821/**
822 * bd_finish_claiming - finish claiming a block device
823 * @bdev: block device of interest (passed to bd_start_claiming())
824 * @whole: whole block device returned by bd_start_claiming()
825 * @holder: holder trying to claim @bdev
826 *
827 * Finish a claiming block started by bd_start_claiming().
828 *
829 * CONTEXT:
830 * Grabs and releases bdev_lock.
831 */
832static void bd_finish_claiming(struct block_device *bdev,
833 struct block_device *whole, void *holder)
834{
835 spin_lock(&bdev_lock);
Nick Pigginb0018362010-05-26 01:51:19 +1000836 BUG_ON(!bd_may_claim(bdev, whole, holder));
837 __bd_claim(bdev, whole, holder);
838 __bd_abort_claiming(whole, holder); /* not actually an abort */
839}
840
Tejun Heo6b4517a2010-04-07 18:53:59 +0900841/**
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900842 * bd_claim - claim a block device
843 * @bdev: block device to claim
844 * @holder: holder trying to claim @bdev
845 *
Nick Pigginb0018362010-05-26 01:51:19 +1000846 * Try to claim @bdev which must have been opened successfully.
Tejun Heo6b4517a2010-04-07 18:53:59 +0900847 *
848 * CONTEXT:
849 * Might sleep.
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900850 *
851 * RETURNS:
852 * 0 if successful, -EBUSY if @bdev is already claimed.
853 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854int bd_claim(struct block_device *bdev, void *holder)
855{
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900856 struct block_device *whole = bdev->bd_contains;
Tejun Heo6b4517a2010-04-07 18:53:59 +0900857 int res;
858
859 might_sleep();
Tejun Heo1a3cbbc2010-04-07 18:52:29 +0900860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 spin_lock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900862 res = bd_prepare_to_claim(bdev, whole, holder);
Nick Pigginb0018362010-05-26 01:51:19 +1000863 if (res == 0)
864 __bd_claim(bdev, whole, holder);
865 spin_unlock(&bdev_lock);
Tejun Heo6b4517a2010-04-07 18:53:59 +0900866
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return res;
868}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869EXPORT_SYMBOL(bd_claim);
870
871void bd_release(struct block_device *bdev)
872{
873 spin_lock(&bdev_lock);
874 if (!--bdev->bd_contains->bd_holders)
875 bdev->bd_contains->bd_holder = NULL;
876 if (!--bdev->bd_holders)
877 bdev->bd_holder = NULL;
878 spin_unlock(&bdev_lock);
879}
880
881EXPORT_SYMBOL(bd_release);
882
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800883#ifdef CONFIG_SYSFS
884/*
885 * Functions for bd_claim_by_kobject / bd_release_from_kobject
886 *
887 * If a kobject is passed to bd_claim_by_kobject()
888 * and the kobject has a parent directory,
889 * following symlinks are created:
890 * o from the kobject to the claimed bdev
891 * o from "holders" directory of the bdev to the parent of the kobject
892 * bd_release_from_kobject() removes these symlinks.
893 *
894 * Example:
895 * If /dev/dm-0 maps to /dev/sda, kobject corresponding to
896 * /sys/block/dm-0/slaves is passed to bd_claim_by_kobject(), then:
897 * /sys/block/dm-0/slaves/sda --> /sys/block/sda
898 * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
899 */
900
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700901static int add_symlink(struct kobject *from, struct kobject *to)
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800902{
903 if (!from || !to)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -0700904 return 0;
905 return sysfs_create_link(from, to, kobject_name(to));
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800906}
907
908static void del_symlink(struct kobject *from, struct kobject *to)
909{
910 if (!from || !to)
911 return;
912 sysfs_remove_link(from, kobject_name(to));
913}
914
915/*
916 * 'struct bd_holder' contains pointers to kobjects symlinked by
917 * bd_claim_by_kobject.
918 * It's connected to bd_holder_list which is protected by bdev->bd_sem.
919 */
920struct bd_holder {
921 struct list_head list; /* chain of holders of the bdev */
922 int count; /* references from the holder */
923 struct kobject *sdir; /* holder object, e.g. "/block/dm-0/slaves" */
924 struct kobject *hdev; /* e.g. "/block/dm-0" */
925 struct kobject *hdir; /* e.g. "/block/sda/holders" */
926 struct kobject *sdev; /* e.g. "/block/sda" */
927};
928
929/*
930 * Get references of related kobjects at once.
931 * Returns 1 on success. 0 on failure.
932 *
933 * Should call bd_holder_release_dirs() after successful use.
934 */
935static int bd_holder_grab_dirs(struct block_device *bdev,
936 struct bd_holder *bo)
937{
938 if (!bdev || !bo)
939 return 0;
940
941 bo->sdir = kobject_get(bo->sdir);
942 if (!bo->sdir)
943 return 0;
944
945 bo->hdev = kobject_get(bo->sdir->parent);
946 if (!bo->hdev)
947 goto fail_put_sdir;
948
Tejun Heo0762b8b2008-08-25 19:56:12 +0900949 bo->sdev = kobject_get(&part_to_dev(bdev->bd_part)->kobj);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800950 if (!bo->sdev)
951 goto fail_put_hdev;
952
Tejun Heo4c465012008-08-25 19:56:11 +0900953 bo->hdir = kobject_get(bdev->bd_part->holder_dir);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -0800954 if (!bo->hdir)
955 goto fail_put_sdev;
956
957 return 1;
958
959fail_put_sdev:
960 kobject_put(bo->sdev);
961fail_put_hdev:
962 kobject_put(bo->hdev);
963fail_put_sdir:
964 kobject_put(bo->sdir);
965
966 return 0;
967}
968
969/* Put references of related kobjects at once. */
970static void bd_holder_release_dirs(struct bd_holder *bo)
971{
972 kobject_put(bo->hdir);
973 kobject_put(bo->sdev);
974 kobject_put(bo->hdev);
975 kobject_put(bo->sdir);
976}
977
978static struct bd_holder *alloc_bd_holder(struct kobject *kobj)
979{
980 struct bd_holder *bo;
981
982 bo = kzalloc(sizeof(*bo), GFP_KERNEL);
983 if (!bo)
984 return NULL;
985
986 bo->count = 1;
987 bo->sdir = kobj;
988
989 return bo;
990}
991
992static void free_bd_holder(struct bd_holder *bo)
993{
994 kfree(bo);
995}
996
997/**
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -0500998 * find_bd_holder - find matching struct bd_holder from the block device
999 *
1000 * @bdev: struct block device to be searched
1001 * @bo: target struct bd_holder
1002 *
1003 * Returns matching entry with @bo in @bdev->bd_holder_list.
1004 * If found, increment the reference count and return the pointer.
1005 * If not found, returns NULL.
1006 */
Andrew Morton36a561d2006-10-30 22:07:03 -08001007static struct bd_holder *find_bd_holder(struct block_device *bdev,
1008 struct bd_holder *bo)
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001009{
1010 struct bd_holder *tmp;
1011
1012 list_for_each_entry(tmp, &bdev->bd_holder_list, list)
1013 if (tmp->sdir == bo->sdir) {
1014 tmp->count++;
1015 return tmp;
1016 }
1017
1018 return NULL;
1019}
1020
1021/**
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001022 * add_bd_holder - create sysfs symlinks for bd_claim() relationship
1023 *
1024 * @bdev: block device to be bd_claimed
1025 * @bo: preallocated and initialized by alloc_bd_holder()
1026 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001027 * Add @bo to @bdev->bd_holder_list, create symlinks.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001028 *
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001029 * Returns 0 if symlinks are created.
1030 * Returns -ve if something fails.
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001031 */
1032static int add_bd_holder(struct block_device *bdev, struct bd_holder *bo)
1033{
Johannes Weiner4e916722007-07-15 23:41:25 -07001034 int err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001035
1036 if (!bo)
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001037 return -EINVAL;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001038
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001039 if (!bd_holder_grab_dirs(bdev, bo))
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001040 return -EBUSY;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001041
Johannes Weiner4e916722007-07-15 23:41:25 -07001042 err = add_symlink(bo->sdir, bo->sdev);
1043 if (err)
1044 return err;
1045
1046 err = add_symlink(bo->hdir, bo->hdev);
1047 if (err) {
1048 del_symlink(bo->sdir, bo->sdev);
1049 return err;
Andrew Morton4d7dd8f2006-09-29 01:58:56 -07001050 }
Johannes Weiner4e916722007-07-15 23:41:25 -07001051
1052 list_add_tail(&bo->list, &bdev->bd_holder_list);
1053 return 0;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001054}
1055
1056/**
1057 * del_bd_holder - delete sysfs symlinks for bd_claim() relationship
1058 *
1059 * @bdev: block device to be bd_claimed
1060 * @kobj: holder's kobject
1061 *
1062 * If there is matching entry with @kobj in @bdev->bd_holder_list
1063 * and no other bd_claim() from the same kobject,
1064 * remove the struct bd_holder from the list, delete symlinks for it.
1065 *
1066 * Returns a pointer to the struct bd_holder when it's removed from the list
1067 * and ready to be freed.
1068 * Returns NULL if matching claim isn't found or there is other bd_claim()
1069 * by the same kobject.
1070 */
1071static struct bd_holder *del_bd_holder(struct block_device *bdev,
1072 struct kobject *kobj)
1073{
1074 struct bd_holder *bo;
1075
1076 list_for_each_entry(bo, &bdev->bd_holder_list, list) {
1077 if (bo->sdir == kobj) {
1078 bo->count--;
1079 BUG_ON(bo->count < 0);
1080 if (!bo->count) {
1081 list_del(&bo->list);
1082 del_symlink(bo->sdir, bo->sdev);
1083 del_symlink(bo->hdir, bo->hdev);
1084 bd_holder_release_dirs(bo);
1085 return bo;
1086 }
1087 break;
1088 }
1089 }
1090
1091 return NULL;
1092}
1093
1094/**
1095 * bd_claim_by_kobject - bd_claim() with additional kobject signature
1096 *
1097 * @bdev: block device to be claimed
1098 * @holder: holder's signature
1099 * @kobj: holder's kobject
1100 *
1101 * Do bd_claim() and if it succeeds, create sysfs symlinks between
1102 * the bdev and the holder's kobject.
1103 * Use bd_release_from_kobject() when relesing the claimed bdev.
1104 *
1105 * Returns 0 on success. (same as bd_claim())
1106 * Returns errno on failure.
1107 */
1108static int bd_claim_by_kobject(struct block_device *bdev, void *holder,
1109 struct kobject *kobj)
1110{
Johannes Weiner4e916722007-07-15 23:41:25 -07001111 int err;
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001112 struct bd_holder *bo, *found;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001113
1114 if (!kobj)
1115 return -EINVAL;
1116
1117 bo = alloc_bd_holder(kobj);
1118 if (!bo)
1119 return -ENOMEM;
1120
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001121 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomuradf6c0cd2006-10-30 16:23:56 -05001122
Johannes Weiner4e916722007-07-15 23:41:25 -07001123 err = bd_claim(bdev, holder);
1124 if (err)
Andrew Morton4210df22007-07-15 23:41:28 -07001125 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001126
1127 found = find_bd_holder(bdev, bo);
1128 if (found)
Andrew Morton4210df22007-07-15 23:41:28 -07001129 goto fail;
Johannes Weiner4e916722007-07-15 23:41:25 -07001130
1131 err = add_bd_holder(bdev, bo);
1132 if (err)
1133 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001134 else
1135 bo = NULL;
1136fail:
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001137 mutex_unlock(&bdev->bd_mutex);
Andrew Morton4210df22007-07-15 23:41:28 -07001138 free_bd_holder(bo);
Johannes Weiner4e916722007-07-15 23:41:25 -07001139 return err;
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001140}
1141
1142/**
1143 * bd_release_from_kobject - bd_release() with additional kobject signature
1144 *
1145 * @bdev: block device to be released
1146 * @kobj: holder's kobject
1147 *
1148 * Do bd_release() and remove sysfs symlinks created by bd_claim_by_kobject().
1149 */
1150static void bd_release_from_kobject(struct block_device *bdev,
1151 struct kobject *kobj)
1152{
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001153 if (!kobj)
1154 return;
1155
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001156 mutex_lock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001157 bd_release(bdev);
Andrew Morton4210df22007-07-15 23:41:28 -07001158 free_bd_holder(del_bd_holder(bdev, kobj));
Jun'ichi Nomurab4cf1b72006-03-27 01:18:00 -08001159 mutex_unlock(&bdev->bd_mutex);
Jun'ichi Nomura641dc632006-03-27 01:17:57 -08001160}
1161
1162/**
1163 * bd_claim_by_disk - wrapper function for bd_claim_by_kobject()
1164 *
1165 * @bdev: block device to be claimed
1166 * @holder: holder's signature
1167 * @disk: holder's gendisk
1168 *
1169 * Call bd_claim_by_kobject() with getting @disk->slave_dir.
1170 */
1171int bd_claim_by_disk(struct block_device *bdev, void *holder,
1172 struct gendisk *disk)
1173{
1174 return bd_claim_by_kobject(bdev, holder, kobject_get(disk->slave_dir));
1175}
1176EXPORT_SYMBOL_GPL(bd_claim_by_disk);
1177
1178/**
1179 * bd_release_from_disk - wrapper function for bd_release_from_kobject()
1180 *
1181 * @bdev: block device to be claimed
1182 * @disk: holder's gendisk
1183 *
1184 * Call bd_release_from_kobject() and put @disk->slave_dir.
1185 */
1186void bd_release_from_disk(struct block_device *bdev, struct gendisk *disk)
1187{
1188 bd_release_from_kobject(bdev, disk->slave_dir);
1189 kobject_put(disk->slave_dir);
1190}
1191EXPORT_SYMBOL_GPL(bd_release_from_disk);
1192#endif
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194/*
1195 * Tries to open block device by device number. Use it ONLY if you
1196 * really do not have anything better - i.e. when you are behind a
1197 * truly sucky interface and all you are given is a device number. _Never_
1198 * to be used for internal purposes. If you ever need it - reconsider
1199 * your API.
1200 */
Al Viroaeb5d722008-09-02 15:28:45 -04001201struct block_device *open_by_devnum(dev_t dev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
1203 struct block_device *bdev = bdget(dev);
1204 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (bdev)
Al Viro572c4892007-10-08 13:24:05 -04001206 err = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 return err ? ERR_PTR(err) : bdev;
1208}
1209
1210EXPORT_SYMBOL(open_by_devnum);
1211
Andrew Patterson0c002c22008-09-04 14:27:20 -06001212/**
Andrew Patterson56ade442008-09-04 14:27:40 -06001213 * flush_disk - invalidates all buffer-cache entries on a disk
1214 *
1215 * @bdev: struct block device to be flushed
1216 *
1217 * Invalidates all buffer-cache entries on a disk. It should be called
1218 * when a disk has been changed -- either by a media change or online
1219 * resize.
1220 */
1221static void flush_disk(struct block_device *bdev)
1222{
1223 if (__invalidate_device(bdev)) {
1224 char name[BDEVNAME_SIZE] = "";
1225
1226 if (bdev->bd_disk)
1227 disk_name(bdev->bd_disk, 0, name);
1228 printk(KERN_WARNING "VFS: busy inodes on changed media or "
1229 "resized disk %s\n", name);
1230 }
1231
1232 if (!bdev->bd_disk)
1233 return;
1234 if (disk_partitionable(bdev->bd_disk))
1235 bdev->bd_invalidated = 1;
1236}
1237
1238/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001239 * check_disk_size_change - checks for disk size change and adjusts bdev size.
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001240 * @disk: struct gendisk to check
1241 * @bdev: struct bdev to adjust.
1242 *
1243 * This routine checks to see if the bdev size does not match the disk size
1244 * and adjusts it if it differs.
1245 */
1246void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1247{
1248 loff_t disk_size, bdev_size;
1249
1250 disk_size = (loff_t)get_capacity(disk) << 9;
1251 bdev_size = i_size_read(bdev->bd_inode);
1252 if (disk_size != bdev_size) {
1253 char name[BDEVNAME_SIZE];
1254
1255 disk_name(disk, 0, name);
1256 printk(KERN_INFO
1257 "%s: detected capacity change from %lld to %lld\n",
1258 name, bdev_size, disk_size);
1259 i_size_write(bdev->bd_inode, disk_size);
Andrew Patterson608aeef2008-09-04 14:27:45 -06001260 flush_disk(bdev);
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001261 }
1262}
1263EXPORT_SYMBOL(check_disk_size_change);
1264
1265/**
Randy Dunlap57d1b532008-10-09 10:42:38 +02001266 * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
Andrew Patterson0c002c22008-09-04 14:27:20 -06001267 * @disk: struct gendisk to be revalidated
1268 *
1269 * This routine is a wrapper for lower-level driver's revalidate_disk
1270 * call-backs. It is used to do common pre and post operations needed
1271 * for all revalidate_disk operations.
1272 */
1273int revalidate_disk(struct gendisk *disk)
1274{
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001275 struct block_device *bdev;
Andrew Patterson0c002c22008-09-04 14:27:20 -06001276 int ret = 0;
1277
1278 if (disk->fops->revalidate_disk)
1279 ret = disk->fops->revalidate_disk(disk);
1280
Andrew Pattersonc3279d12008-09-04 14:27:25 -06001281 bdev = bdget_disk(disk, 0);
1282 if (!bdev)
1283 return ret;
1284
1285 mutex_lock(&bdev->bd_mutex);
1286 check_disk_size_change(disk, bdev);
1287 mutex_unlock(&bdev->bd_mutex);
1288 bdput(bdev);
Andrew Patterson0c002c22008-09-04 14:27:20 -06001289 return ret;
1290}
1291EXPORT_SYMBOL(revalidate_disk);
1292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293/*
1294 * This routine checks whether a removable media has been changed,
1295 * and invalidates all buffer-cache-entries in that case. This
1296 * is a relatively slow routine, so we have to try to minimize using
1297 * it. Thus it is called only upon a 'mount' or 'open'. This
1298 * is the best way of combining speed and utility, I think.
1299 * People changing diskettes in the middle of an operation deserve
1300 * to lose :-)
1301 */
1302int check_disk_change(struct block_device *bdev)
1303{
1304 struct gendisk *disk = bdev->bd_disk;
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001305 const struct block_device_operations *bdops = disk->fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 if (!bdops->media_changed)
1308 return 0;
1309 if (!bdops->media_changed(bdev->bd_disk))
1310 return 0;
1311
Andrew Patterson56ade442008-09-04 14:27:40 -06001312 flush_disk(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 if (bdops->revalidate_disk)
1314 bdops->revalidate_disk(bdev->bd_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 return 1;
1316}
1317
1318EXPORT_SYMBOL(check_disk_change);
1319
1320void bd_set_size(struct block_device *bdev, loff_t size)
1321{
Martin K. Petersene1defc42009-05-22 17:17:49 -04001322 unsigned bsize = bdev_logical_block_size(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
1324 bdev->bd_inode->i_size = size;
1325 while (bsize < PAGE_CACHE_SIZE) {
1326 if (size & bsize)
1327 break;
1328 bsize <<= 1;
1329 }
1330 bdev->bd_block_size = bsize;
1331 bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1332}
1333EXPORT_SYMBOL(bd_set_size);
1334
Al Viro9a1c3542008-02-22 20:40:24 -05001335static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001336
Peter Zijlstra6d740cd2007-02-20 13:58:18 -08001337/*
1338 * bd_mutex locking:
1339 *
1340 * mutex_lock(part->bd_mutex)
1341 * mutex_lock_nested(whole->bd_mutex, 1)
1342 */
1343
Al Viro572c4892007-10-08 13:24:05 -04001344static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 struct gendisk *disk;
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001347 int ret;
Tejun Heocf771cb2008-09-03 09:01:09 +02001348 int partno;
Al Virofe6e9c12008-06-23 08:30:55 -04001349 int perm = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Al Viro572c4892007-10-08 13:24:05 -04001351 if (mode & FMODE_READ)
Al Virofe6e9c12008-06-23 08:30:55 -04001352 perm |= MAY_READ;
Al Viro572c4892007-10-08 13:24:05 -04001353 if (mode & FMODE_WRITE)
Al Virofe6e9c12008-06-23 08:30:55 -04001354 perm |= MAY_WRITE;
1355 /*
1356 * hooks: /n/, see "layering violations".
1357 */
Chris Wrightb7300b72010-08-10 18:02:55 -07001358 if (!for_part) {
1359 ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1360 if (ret != 0) {
1361 bdput(bdev);
1362 return ret;
1363 }
Al Viro82666022008-08-01 05:32:04 -04001364 }
Pavel Emelyanov7db9cfd2008-06-05 22:46:27 -07001365
NeilBrownd3374822009-01-09 08:31:10 +11001366 restart:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001367
Tejun Heo89f97492008-11-05 10:21:06 +01001368 ret = -ENXIO;
Tejun Heocf771cb2008-09-03 09:01:09 +02001369 disk = get_gendisk(bdev->bd_dev, &partno);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001370 if (!disk)
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001371 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
NeilBrown6796bf52006-12-08 02:36:16 -08001373 mutex_lock_nested(&bdev->bd_mutex, for_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (!bdev->bd_openers) {
1375 bdev->bd_disk = disk;
1376 bdev->bd_contains = bdev;
Tejun Heocf771cb2008-09-03 09:01:09 +02001377 if (!partno) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 struct backing_dev_info *bdi;
Tejun Heo89f97492008-11-05 10:21:06 +01001379
1380 ret = -ENXIO;
1381 bdev->bd_part = disk_get_part(disk, partno);
1382 if (!bdev->bd_part)
1383 goto out_clear;
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 if (disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001386 ret = disk->fops->open(bdev, mode);
NeilBrownd3374822009-01-09 08:31:10 +11001387 if (ret == -ERESTARTSYS) {
1388 /* Lost a race with 'disk' being
1389 * deleted, try again.
1390 * See md.c
1391 */
1392 disk_put_part(bdev->bd_part);
1393 bdev->bd_part = NULL;
1394 module_put(disk->fops->owner);
1395 put_disk(disk);
1396 bdev->bd_disk = NULL;
1397 mutex_unlock(&bdev->bd_mutex);
1398 goto restart;
1399 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001401 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403 if (!bdev->bd_openers) {
1404 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1405 bdi = blk_get_backing_dev_info(bdev);
1406 if (bdi == NULL)
1407 bdi = &default_backing_dev_info;
Dave Chinnera5491e02010-10-21 11:49:26 +11001408 bdev_inode_switch_bdi(bdev->bd_inode, bdi);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 }
1410 if (bdev->bd_invalidated)
1411 rescan_partitions(disk, bdev);
1412 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 struct block_device *whole;
1414 whole = bdget_disk(disk, 0);
1415 ret = -ENOMEM;
1416 if (!whole)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001417 goto out_clear;
NeilBrown37be4122006-12-08 02:36:16 -08001418 BUG_ON(for_part);
Al Viro572c4892007-10-08 13:24:05 -04001419 ret = __blkdev_get(whole, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001421 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 bdev->bd_contains = whole;
Dave Chinnera5491e02010-10-21 11:49:26 +11001423 bdev_inode_switch_bdi(bdev->bd_inode,
1424 whole->bd_inode->i_data.backing_dev_info);
Tejun Heo89f97492008-11-05 10:21:06 +01001425 bdev->bd_part = disk_get_part(disk, partno);
Tejun Heoe71bf0d2008-09-03 09:03:02 +02001426 if (!(disk->flags & GENHD_FL_UP) ||
Tejun Heo89f97492008-11-05 10:21:06 +01001427 !bdev->bd_part || !bdev->bd_part->nr_sects) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 ret = -ENXIO;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001429 goto out_clear;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 }
Tejun Heo89f97492008-11-05 10:21:06 +01001431 bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 }
1433 } else {
Tejun Heo0762b8b2008-08-25 19:56:12 +09001434 module_put(disk->fops->owner);
Neil Brown960cc0f2009-10-26 08:59:17 +01001435 put_disk(disk);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001436 disk = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 if (bdev->bd_contains == bdev) {
1438 if (bdev->bd_disk->fops->open) {
Al Viro572c4892007-10-08 13:24:05 -04001439 ret = bdev->bd_disk->fops->open(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 if (ret)
Tejun Heo0762b8b2008-08-25 19:56:12 +09001441 goto out_unlock_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 }
1443 if (bdev->bd_invalidated)
1444 rescan_partitions(bdev->bd_disk, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 }
1446 }
1447 bdev->bd_openers++;
NeilBrown37be4122006-12-08 02:36:16 -08001448 if (for_part)
1449 bdev->bd_part_count++;
Arjan van de Venc039e312006-03-23 03:00:28 -08001450 mutex_unlock(&bdev->bd_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 return 0;
1452
Tejun Heo0762b8b2008-08-25 19:56:12 +09001453 out_clear:
Tejun Heo89f97492008-11-05 10:21:06 +01001454 disk_put_part(bdev->bd_part);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 bdev->bd_disk = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001456 bdev->bd_part = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001457 bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 if (bdev != bdev->bd_contains)
Al Viro572c4892007-10-08 13:24:05 -04001459 __blkdev_put(bdev->bd_contains, mode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 bdev->bd_contains = NULL;
Tejun Heo0762b8b2008-08-25 19:56:12 +09001461 out_unlock_bdev:
Arjan van de Venc039e312006-03-23 03:00:28 -08001462 mutex_unlock(&bdev->bd_mutex);
Arnd Bergmann6e9624b2010-08-07 18:25:34 +02001463 out:
Tejun Heo0762b8b2008-08-25 19:56:12 +09001464 if (disk)
1465 module_put(disk->fops->owner);
1466 put_disk(disk);
1467 bdput(bdev);
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 return ret;
1470}
1471
Al Viro572c4892007-10-08 13:24:05 -04001472int blkdev_get(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473{
Al Viro572c4892007-10-08 13:24:05 -04001474 return __blkdev_get(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001475}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476EXPORT_SYMBOL(blkdev_get);
1477
1478static int blkdev_open(struct inode * inode, struct file * filp)
1479{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001480 struct block_device *whole = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 struct block_device *bdev;
1482 int res;
1483
1484 /*
1485 * Preserve backwards compatibility and allow large file access
1486 * even if userspace doesn't ask for it explicitly. Some mkfs
1487 * binary needs it. We might want to drop this workaround
1488 * during an unstable branch.
1489 */
1490 filp->f_flags |= O_LARGEFILE;
1491
Al Viro572c4892007-10-08 13:24:05 -04001492 if (filp->f_flags & O_NDELAY)
1493 filp->f_mode |= FMODE_NDELAY;
1494 if (filp->f_flags & O_EXCL)
1495 filp->f_mode |= FMODE_EXCL;
1496 if ((filp->f_flags & O_ACCMODE) == 3)
1497 filp->f_mode |= FMODE_WRITE_IOCTL;
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 bdev = bd_acquire(inode);
Pavel Emelianov6a2aae02006-10-28 10:38:33 -07001500 if (bdev == NULL)
1501 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502
Tejun Heo6b4517a2010-04-07 18:53:59 +09001503 if (filp->f_mode & FMODE_EXCL) {
1504 whole = bd_start_claiming(bdev, filp);
1505 if (IS_ERR(whole)) {
1506 bdput(bdev);
1507 return PTR_ERR(whole);
1508 }
1509 }
1510
Al Viro572c4892007-10-08 13:24:05 -04001511 filp->f_mapping = bdev->bd_inode->i_mapping;
1512
1513 res = blkdev_get(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Tejun Heo6b4517a2010-04-07 18:53:59 +09001515 if (whole) {
1516 if (res == 0)
Nick Pigginb0018362010-05-26 01:51:19 +10001517 bd_finish_claiming(bdev, whole, filp);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001518 else
1519 bd_abort_claiming(whole, filp);
Christoph Hellwigebbefc02008-11-05 14:54:41 +01001520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 return res;
1523}
1524
Al Viro9a1c3542008-02-22 20:40:24 -05001525static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001526{
1527 int ret = 0;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001528 struct gendisk *disk = bdev->bd_disk;
NeilBrown37be4122006-12-08 02:36:16 -08001529 struct block_device *victim = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001530
NeilBrown6796bf52006-12-08 02:36:16 -08001531 mutex_lock_nested(&bdev->bd_mutex, for_part);
NeilBrown37be4122006-12-08 02:36:16 -08001532 if (for_part)
1533 bdev->bd_part_count--;
1534
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001535 if (!--bdev->bd_openers) {
1536 sync_blockdev(bdev);
1537 kill_bdev(bdev);
1538 }
1539 if (bdev->bd_contains == bdev) {
1540 if (disk->fops->release)
Al Viro9a1c3542008-02-22 20:40:24 -05001541 ret = disk->fops->release(disk, mode);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001542 }
1543 if (!bdev->bd_openers) {
1544 struct module *owner = disk->fops->owner;
1545
1546 put_disk(disk);
1547 module_put(owner);
Tejun Heo0762b8b2008-08-25 19:56:12 +09001548 disk_put_part(bdev->bd_part);
1549 bdev->bd_part = NULL;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001550 bdev->bd_disk = NULL;
Dave Chinnera5491e02010-10-21 11:49:26 +11001551 bdev_inode_switch_bdi(bdev->bd_inode,
1552 &default_backing_dev_info);
NeilBrown37be4122006-12-08 02:36:16 -08001553 if (bdev != bdev->bd_contains)
1554 victim = bdev->bd_contains;
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001555 bdev->bd_contains = NULL;
1556 }
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001557 mutex_unlock(&bdev->bd_mutex);
1558 bdput(bdev);
NeilBrown37be4122006-12-08 02:36:16 -08001559 if (victim)
Al Viro9a1c3542008-02-22 20:40:24 -05001560 __blkdev_put(victim, mode, 1);
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001561 return ret;
1562}
1563
Al Viro9a1c3542008-02-22 20:40:24 -05001564int blkdev_put(struct block_device *bdev, fmode_t mode)
NeilBrown37be4122006-12-08 02:36:16 -08001565{
Al Viro9a1c3542008-02-22 20:40:24 -05001566 return __blkdev_put(bdev, mode, 0);
NeilBrown37be4122006-12-08 02:36:16 -08001567}
Peter Zijlstra2e7b6512006-12-08 02:36:13 -08001568EXPORT_SYMBOL(blkdev_put);
1569
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570static int blkdev_close(struct inode * inode, struct file * filp)
1571{
1572 struct block_device *bdev = I_BDEV(filp->f_mapping->host);
1573 if (bdev->bd_holder == filp)
1574 bd_release(bdev);
Al Viro9a1c3542008-02-22 20:40:24 -05001575 return blkdev_put(bdev, filp->f_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576}
1577
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001578static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579{
Al Viro56b26ad2008-09-19 03:17:36 -04001580 struct block_device *bdev = I_BDEV(file->f_mapping->host);
1581 fmode_t mode = file->f_mode;
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001582
1583 /*
1584 * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1585 * to updated it before every ioctl.
1586 */
Al Viro56b26ad2008-09-19 03:17:36 -04001587 if (file->f_flags & O_NDELAY)
Christoph Hellwigfd4ce1a2008-11-05 14:58:42 +01001588 mode |= FMODE_NDELAY;
1589 else
1590 mode &= ~FMODE_NDELAY;
1591
Al Viro56b26ad2008-09-19 03:17:36 -04001592 return blkdev_ioctl(bdev, mode, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593}
1594
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001595/*
Christoph Hellwigeef99382009-08-20 17:43:41 +02001596 * Write data to the block device. Only intended for the block device itself
1597 * and the raw driver which basically is a fake block device.
1598 *
1599 * Does not take i_mutex for the write and thus is not for general purpose
1600 * use.
1601 */
1602ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov,
1603 unsigned long nr_segs, loff_t pos)
1604{
1605 struct file *file = iocb->ki_filp;
1606 ssize_t ret;
1607
1608 BUG_ON(iocb->ki_pos != pos);
1609
1610 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
1611 if (ret > 0 || ret == -EIOCBQUEUED) {
1612 ssize_t err;
1613
1614 err = generic_write_sync(file, pos, ret);
1615 if (err < 0 && ret > 0)
1616 ret = err;
1617 }
1618 return ret;
1619}
1620EXPORT_SYMBOL_GPL(blkdev_aio_write);
1621
1622/*
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001623 * Try to release a page associated with block device when the system
1624 * is under memory pressure.
1625 */
1626static int blkdev_releasepage(struct page *page, gfp_t wait)
1627{
1628 struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
1629
1630 if (super && super->s_op->bdev_try_to_free_page)
1631 return super->s_op->bdev_try_to_free_page(super, page, wait);
1632
1633 return try_to_free_buffers(page);
1634}
1635
Adrian Bunk4c54ac62008-02-18 13:48:31 +01001636static const struct address_space_operations def_blk_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 .readpage = blkdev_readpage,
1638 .writepage = blkdev_writepage,
1639 .sync_page = block_sync_page,
Nick Piggin6272b5a2007-10-16 01:25:04 -07001640 .write_begin = blkdev_write_begin,
1641 .write_end = blkdev_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 .writepages = generic_writepages,
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001643 .releasepage = blkdev_releasepage,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 .direct_IO = blkdev_direct_IO,
1645};
1646
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001647const struct file_operations def_blk_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 .open = blkdev_open,
1649 .release = blkdev_close,
1650 .llseek = block_llseek,
Badari Pulavarty543ade12006-09-30 23:28:48 -07001651 .read = do_sync_read,
1652 .write = do_sync_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 .aio_read = generic_file_aio_read,
Christoph Hellwigeef99382009-08-20 17:43:41 +02001654 .aio_write = blkdev_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 .mmap = generic_file_mmap,
Andrew Mortonb1dd3b22010-04-06 14:35:00 -07001656 .fsync = blkdev_fsync,
Arnd Bergmannbb93e3a2005-06-23 00:10:15 -07001657 .unlocked_ioctl = block_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658#ifdef CONFIG_COMPAT
1659 .compat_ioctl = compat_blkdev_ioctl,
1660#endif
Jens Axboe7f9c51f2006-05-01 19:59:32 +02001661 .splice_read = generic_file_splice_read,
1662 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663};
1664
1665int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
1666{
1667 int res;
1668 mm_segment_t old_fs = get_fs();
1669 set_fs(KERNEL_DS);
Al Viro56b26ad2008-09-19 03:17:36 -04001670 res = blkdev_ioctl(bdev, 0, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 set_fs(old_fs);
1672 return res;
1673}
1674
1675EXPORT_SYMBOL(ioctl_by_bdev);
1676
1677/**
1678 * lookup_bdev - lookup a struct block_device by name
Randy Dunlap94e29592009-01-06 14:41:15 -08001679 * @pathname: special file representing the block device
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 *
Randy Dunlap57d1b532008-10-09 10:42:38 +02001681 * Get a reference to the blockdevice at @pathname in the current
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 * namespace if possible and return it. Return ERR_PTR(error)
1683 * otherwise.
1684 */
Al Viro421748e2008-08-02 01:04:36 -04001685struct block_device *lookup_bdev(const char *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686{
1687 struct block_device *bdev;
1688 struct inode *inode;
Al Viro421748e2008-08-02 01:04:36 -04001689 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 int error;
1691
Al Viro421748e2008-08-02 01:04:36 -04001692 if (!pathname || !*pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 return ERR_PTR(-EINVAL);
1694
Al Viro421748e2008-08-02 01:04:36 -04001695 error = kern_path(pathname, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (error)
1697 return ERR_PTR(error);
1698
Al Viro421748e2008-08-02 01:04:36 -04001699 inode = path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 error = -ENOTBLK;
1701 if (!S_ISBLK(inode->i_mode))
1702 goto fail;
1703 error = -EACCES;
Al Viro421748e2008-08-02 01:04:36 -04001704 if (path.mnt->mnt_flags & MNT_NODEV)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 goto fail;
1706 error = -ENOMEM;
1707 bdev = bd_acquire(inode);
1708 if (!bdev)
1709 goto fail;
1710out:
Al Viro421748e2008-08-02 01:04:36 -04001711 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 return bdev;
1713fail:
1714 bdev = ERR_PTR(error);
1715 goto out;
1716}
Al Virod5686b42008-08-01 05:00:11 -04001717EXPORT_SYMBOL(lookup_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
1719/**
Al Viro30c40d22008-02-22 19:50:45 -05001720 * open_bdev_exclusive - open a block device by name and set it up for use
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 *
1722 * @path: special file representing the block device
Al Viro30c40d22008-02-22 19:50:45 -05001723 * @mode: FMODE_... combination to pass be used
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 * @holder: owner for exclusion
1725 *
1726 * Open the blockdevice described by the special file at @path, claim it
1727 * for the @holder.
1728 */
Al Viro30c40d22008-02-22 19:50:45 -05001729struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730{
Tejun Heo6b4517a2010-04-07 18:53:59 +09001731 struct block_device *bdev, *whole;
1732 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734 bdev = lookup_bdev(path);
1735 if (IS_ERR(bdev))
1736 return bdev;
1737
Tejun Heo6b4517a2010-04-07 18:53:59 +09001738 whole = bd_start_claiming(bdev, holder);
1739 if (IS_ERR(whole)) {
1740 bdput(bdev);
1741 return whole;
1742 }
1743
Al Viro572c4892007-10-08 13:24:05 -04001744 error = blkdev_get(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 if (error)
Tejun Heo6b4517a2010-04-07 18:53:59 +09001746 goto out_abort_claiming;
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 error = -EACCES;
Al Viro30c40d22008-02-22 19:50:45 -05001749 if ((mode & FMODE_WRITE) && bdev_read_only(bdev))
Tejun Heo6b4517a2010-04-07 18:53:59 +09001750 goto out_blkdev_put;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
Nick Pigginb0018362010-05-26 01:51:19 +10001752 bd_finish_claiming(bdev, whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 return bdev;
Tejun Heo6b4517a2010-04-07 18:53:59 +09001754
1755out_blkdev_put:
Al Viro9a1c3542008-02-22 20:40:24 -05001756 blkdev_put(bdev, mode);
Tejun Heo6b4517a2010-04-07 18:53:59 +09001757out_abort_claiming:
1758 bd_abort_claiming(whole, holder);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 return ERR_PTR(error);
1760}
1761
Al Viro30c40d22008-02-22 19:50:45 -05001762EXPORT_SYMBOL(open_bdev_exclusive);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
1764/**
Al Viro30c40d22008-02-22 19:50:45 -05001765 * close_bdev_exclusive - close a blockdevice opened by open_bdev_exclusive()
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 *
1767 * @bdev: blockdevice to close
Al Viro30c40d22008-02-22 19:50:45 -05001768 * @mode: mode, must match that used to open.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 *
Al Viro30c40d22008-02-22 19:50:45 -05001770 * This is the counterpart to open_bdev_exclusive().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 */
Al Viro30c40d22008-02-22 19:50:45 -05001772void close_bdev_exclusive(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773{
1774 bd_release(bdev);
Al Viro30c40d22008-02-22 19:50:45 -05001775 blkdev_put(bdev, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776}
1777
Al Viro30c40d22008-02-22 19:50:45 -05001778EXPORT_SYMBOL(close_bdev_exclusive);
David Howellsb71e8a42006-08-29 19:06:11 +01001779
1780int __invalidate_device(struct block_device *bdev)
1781{
1782 struct super_block *sb = get_super(bdev);
1783 int res = 0;
1784
1785 if (sb) {
1786 /*
1787 * no need to lock the super, get_super holds the
1788 * read mutex so the filesystem cannot go away
1789 * under us (->put_super runs with the write lock
1790 * hold).
1791 */
1792 shrink_dcache_sb(sb);
1793 res = invalidate_inodes(sb);
1794 drop_super(sb);
1795 }
Peter Zijlstraf98393a2007-05-06 14:49:54 -07001796 invalidate_bdev(bdev);
David Howellsb71e8a42006-08-29 19:06:11 +01001797 return res;
1798}
1799EXPORT_SYMBOL(__invalidate_device);