]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | ||
1da177e4 LT |
8 | #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> | |
1da177e4 | 14 | #include <linux/smp_lock.h> |
7db9cfd3 | 15 | #include <linux/device_cgroup.h> |
1da177e4 LT |
16 | #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> | |
585d3bc0 | 21 | #include <linux/pagevec.h> |
811d736f | 22 | #include <linux/writeback.h> |
1da177e4 LT |
23 | #include <linux/mpage.h> |
24 | #include <linux/mount.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/namei.h> | |
1368c4f2 | 27 | #include <linux/log2.h> |
2e1483c9 | 28 | #include <linux/kmemleak.h> |
1da177e4 | 29 | #include <asm/uaccess.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 LT |
31 | |
32 | struct bdev_inode { | |
33 | struct block_device bdev; | |
34 | struct inode vfs_inode; | |
35 | }; | |
36 | ||
4c54ac62 AB |
37 | static const struct address_space_operations def_blk_aops; |
38 | ||
1da177e4 LT |
39 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
40 | { | |
41 | return container_of(inode, struct bdev_inode, vfs_inode); | |
42 | } | |
43 | ||
44 | inline struct block_device *I_BDEV(struct inode *inode) | |
45 | { | |
46 | return &BDEV_I(inode)->bdev; | |
47 | } | |
48 | ||
49 | EXPORT_SYMBOL(I_BDEV); | |
50 | ||
a5491e0c DC |
51 | /* |
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 | */ | |
56 | static 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) | |
7ccf19a8 | 62 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); |
a5491e0c DC |
63 | spin_unlock(&inode_lock); |
64 | } | |
65 | ||
1da177e4 LT |
66 | static 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 | ||
f9a14399 | 79 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
1da177e4 LT |
80 | static void kill_bdev(struct block_device *bdev) |
81 | { | |
f9a14399 PZ |
82 | if (bdev->bd_inode->i_mapping->nrpages == 0) |
83 | return; | |
84 | invalidate_bh_lrus(); | |
1da177e4 LT |
85 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); |
86 | } | |
87 | ||
88 | int set_blocksize(struct block_device *bdev, int size) | |
89 | { | |
90 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 91 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
92 | return -EINVAL; |
93 | ||
94 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 95 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
96 | 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 | ||
108 | EXPORT_SYMBOL(set_blocksize); | |
109 | ||
110 | int sb_set_blocksize(struct super_block *sb, int size) | |
111 | { | |
1da177e4 LT |
112 | 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; | |
38885bd4 | 117 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
118 | return sb->s_blocksize; |
119 | } | |
120 | ||
121 | EXPORT_SYMBOL(sb_set_blocksize); | |
122 | ||
123 | int sb_min_blocksize(struct super_block *sb, int size) | |
124 | { | |
e1defc4f | 125 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
126 | if (size < minsize) |
127 | size = minsize; | |
128 | return sb_set_blocksize(sb, size); | |
129 | } | |
130 | ||
131 | EXPORT_SYMBOL(sb_min_blocksize); | |
132 | ||
133 | static int | |
134 | blkdev_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 | ||
b2e895db AM |
155 | static int |
156 | blkdev_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 | ||
183 | static ssize_t | |
184 | blkdev_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 | ||
eafdc7d1 CH |
190 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, |
191 | nr_segs, blkdev_get_blocks, NULL, NULL, 0); | |
b2e895db AM |
192 | } |
193 | ||
5cee5815 JK |
194 | int __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 | ||
585d3bc0 NP |
203 | /* |
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 | */ | |
207 | int sync_blockdev(struct block_device *bdev) | |
208 | { | |
5cee5815 | 209 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
210 | } |
211 | EXPORT_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 | */ | |
218 | int fsync_bdev(struct block_device *bdev) | |
219 | { | |
220 | struct super_block *sb = get_super(bdev); | |
221 | if (sb) { | |
60b0680f | 222 | int res = sync_filesystem(sb); |
585d3bc0 NP |
223 | drop_super(sb); |
224 | return res; | |
225 | } | |
226 | return sync_blockdev(bdev); | |
227 | } | |
47e4491b | 228 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
229 | |
230 | /** | |
231 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
232 | * @bdev: blockdevice to lock | |
233 | * | |
585d3bc0 NP |
234 | * 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 | */ | |
242 | struct 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); | |
4504230a CH |
248 | 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 | */ | |
585d3bc0 | 254 | sb = get_super(bdev); |
4504230a CH |
255 | drop_super(sb); |
256 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
257 | return sb; | |
258 | } | |
259 | ||
260 | sb = get_active_super(bdev); | |
261 | if (!sb) | |
262 | goto out; | |
18e9e510 JB |
263 | error = freeze_super(sb); |
264 | if (error) { | |
265 | deactivate_super(sb); | |
266 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 267 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 268 | return ERR_PTR(error); |
585d3bc0 | 269 | } |
18e9e510 | 270 | deactivate_super(sb); |
4504230a | 271 | out: |
585d3bc0 NP |
272 | sync_blockdev(bdev); |
273 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 274 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
275 | } |
276 | EXPORT_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 | */ | |
285 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
286 | { | |
4504230a | 287 | int error = -EINVAL; |
585d3bc0 NP |
288 | |
289 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 290 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 291 | goto out; |
4504230a CH |
292 | |
293 | error = 0; | |
294 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 295 | goto out; |
4504230a CH |
296 | |
297 | if (!sb) | |
18e9e510 | 298 | goto out; |
4504230a | 299 | |
18e9e510 JB |
300 | error = thaw_super(sb); |
301 | if (error) { | |
302 | bdev->bd_fsfreeze_count++; | |
303 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
304 | return error; | |
305 | } | |
306 | out: | |
585d3bc0 NP |
307 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
308 | return 0; | |
309 | } | |
310 | EXPORT_SYMBOL(thaw_bdev); | |
311 | ||
1da177e4 LT |
312 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
313 | { | |
314 | return block_write_full_page(page, blkdev_get_block, wbc); | |
315 | } | |
316 | ||
317 | static int blkdev_readpage(struct file * file, struct page * page) | |
318 | { | |
319 | return block_read_full_page(page, blkdev_get_block); | |
320 | } | |
321 | ||
6272b5a5 NP |
322 | static 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) | |
1da177e4 | 325 | { |
155130a4 CH |
326 | return block_write_begin(mapping, pos, len, flags, pagep, |
327 | blkdev_get_block); | |
1da177e4 LT |
328 | } |
329 | ||
6272b5a5 NP |
330 | static 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) | |
1da177e4 | 333 | { |
6272b5a5 NP |
334 | 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; | |
1da177e4 LT |
341 | } |
342 | ||
343 | /* | |
344 | * private llseek: | |
0f7fc9e4 | 345 | * for a block special file file->f_path.dentry->d_inode->i_size is zero |
1da177e4 LT |
346 | * so we compute the size by hand (just as in block_read/write above) |
347 | */ | |
348 | static 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 | ||
1b1dcc1b | 354 | mutex_lock(&bd_inode->i_mutex); |
1da177e4 LT |
355 | 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 | } | |
1b1dcc1b | 371 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
372 | return retval; |
373 | } | |
374 | ||
7ea80859 | 375 | int blkdev_fsync(struct file *filp, int datasync) |
1da177e4 | 376 | { |
b8af67e2 AB |
377 | struct inode *bd_inode = filp->f_mapping->host; |
378 | struct block_device *bdev = I_BDEV(bd_inode); | |
ab0a9735 CH |
379 | int error; |
380 | ||
b8af67e2 AB |
381 | /* |
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 | ||
dd3932ed | 388 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
389 | if (error == -EOPNOTSUPP) |
390 | error = 0; | |
b8af67e2 AB |
391 | |
392 | mutex_lock(&bd_inode->i_mutex); | |
393 | ||
ab0a9735 | 394 | return error; |
1da177e4 | 395 | } |
b1dd3b28 | 396 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 LT |
397 | |
398 | /* | |
399 | * pseudo-fs | |
400 | */ | |
401 | ||
402 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 403 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
404 | |
405 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
406 | { | |
e94b1766 | 407 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
408 | if (!ei) |
409 | return NULL; | |
410 | return &ei->vfs_inode; | |
411 | } | |
412 | ||
413 | static void bdev_destroy_inode(struct inode *inode) | |
414 | { | |
415 | struct bdev_inode *bdi = BDEV_I(inode); | |
416 | ||
1da177e4 LT |
417 | kmem_cache_free(bdev_cachep, bdi); |
418 | } | |
419 | ||
51cc5068 | 420 | static void init_once(void *foo) |
1da177e4 LT |
421 | { |
422 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
423 | struct block_device *bdev = &ei->bdev; | |
424 | ||
a35afb83 CL |
425 | memset(bdev, 0, sizeof(*bdev)); |
426 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
427 | INIT_LIST_HEAD(&bdev->bd_inodes); |
428 | INIT_LIST_HEAD(&bdev->bd_list); | |
a35afb83 | 429 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
430 | /* Initialize mutex for freeze. */ |
431 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
432 | } |
433 | ||
434 | static inline void __bd_forget(struct inode *inode) | |
435 | { | |
436 | list_del_init(&inode->i_devices); | |
437 | inode->i_bdev = NULL; | |
438 | inode->i_mapping = &inode->i_data; | |
439 | } | |
440 | ||
b57922d9 | 441 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
442 | { |
443 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
444 | struct list_head *p; | |
b57922d9 AV |
445 | truncate_inode_pages(&inode->i_data, 0); |
446 | invalidate_inode_buffers(inode); /* is it needed here? */ | |
447 | end_writeback(inode); | |
1da177e4 LT |
448 | spin_lock(&bdev_lock); |
449 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
450 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
451 | } | |
452 | list_del_init(&bdev->bd_list); | |
453 | spin_unlock(&bdev_lock); | |
454 | } | |
455 | ||
ee9b6d61 | 456 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
457 | .statfs = simple_statfs, |
458 | .alloc_inode = bdev_alloc_inode, | |
459 | .destroy_inode = bdev_destroy_inode, | |
460 | .drop_inode = generic_delete_inode, | |
b57922d9 | 461 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
462 | }; |
463 | ||
51139ada AV |
464 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
465 | int flags, const char *dev_name, void *data) | |
1da177e4 | 466 | { |
51139ada | 467 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, 0x62646576); |
1da177e4 LT |
468 | } |
469 | ||
470 | static struct file_system_type bd_type = { | |
471 | .name = "bdev", | |
51139ada | 472 | .mount = bd_mount, |
1da177e4 LT |
473 | .kill_sb = kill_anon_super, |
474 | }; | |
475 | ||
c2acf7b9 | 476 | struct super_block *blockdev_superblock __read_mostly; |
1da177e4 LT |
477 | |
478 | void __init bdev_cache_init(void) | |
479 | { | |
480 | int err; | |
c2acf7b9 DC |
481 | struct vfsmount *bd_mnt; |
482 | ||
1da177e4 | 483 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 PJ |
484 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
485 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
20c2df83 | 486 | init_once); |
1da177e4 LT |
487 | err = register_filesystem(&bd_type); |
488 | if (err) | |
489 | panic("Cannot register bdev pseudo-fs"); | |
490 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
491 | if (IS_ERR(bd_mnt)) |
492 | panic("Cannot create bdev pseudo-fs"); | |
2e1483c9 CM |
493 | /* |
494 | * This vfsmount structure is only used to obtain the | |
495 | * blockdev_superblock, so tell kmemleak not to report it. | |
496 | */ | |
497 | kmemleak_not_leak(bd_mnt); | |
1da177e4 LT |
498 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
499 | } | |
500 | ||
501 | /* | |
502 | * Most likely _very_ bad one - but then it's hardly critical for small | |
503 | * /dev and can be fixed when somebody will need really large one. | |
504 | * Keep in mind that it will be fed through icache hash function too. | |
505 | */ | |
506 | static inline unsigned long hash(dev_t dev) | |
507 | { | |
508 | return MAJOR(dev)+MINOR(dev); | |
509 | } | |
510 | ||
511 | static int bdev_test(struct inode *inode, void *data) | |
512 | { | |
513 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
514 | } | |
515 | ||
516 | static int bdev_set(struct inode *inode, void *data) | |
517 | { | |
518 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
519 | return 0; | |
520 | } | |
521 | ||
522 | static LIST_HEAD(all_bdevs); | |
523 | ||
524 | struct block_device *bdget(dev_t dev) | |
525 | { | |
526 | struct block_device *bdev; | |
527 | struct inode *inode; | |
528 | ||
c2acf7b9 | 529 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
530 | bdev_test, bdev_set, &dev); |
531 | ||
532 | if (!inode) | |
533 | return NULL; | |
534 | ||
535 | bdev = &BDEV_I(inode)->bdev; | |
536 | ||
537 | if (inode->i_state & I_NEW) { | |
538 | bdev->bd_contains = NULL; | |
539 | bdev->bd_inode = inode; | |
540 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
541 | bdev->bd_part_count = 0; | |
542 | bdev->bd_invalidated = 0; | |
543 | inode->i_mode = S_IFBLK; | |
544 | inode->i_rdev = dev; | |
545 | inode->i_bdev = bdev; | |
546 | inode->i_data.a_ops = &def_blk_aops; | |
547 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
548 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
549 | spin_lock(&bdev_lock); | |
550 | list_add(&bdev->bd_list, &all_bdevs); | |
551 | spin_unlock(&bdev_lock); | |
552 | unlock_new_inode(inode); | |
553 | } | |
554 | return bdev; | |
555 | } | |
556 | ||
557 | EXPORT_SYMBOL(bdget); | |
558 | ||
dddac6a7 AJ |
559 | /** |
560 | * bdgrab -- Grab a reference to an already referenced block device | |
561 | * @bdev: Block device to grab a reference to. | |
562 | */ | |
563 | struct block_device *bdgrab(struct block_device *bdev) | |
564 | { | |
7de9c6ee | 565 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
566 | return bdev; |
567 | } | |
568 | ||
1da177e4 LT |
569 | long nr_blockdev_pages(void) |
570 | { | |
203a2935 | 571 | struct block_device *bdev; |
1da177e4 LT |
572 | long ret = 0; |
573 | spin_lock(&bdev_lock); | |
203a2935 | 574 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
575 | ret += bdev->bd_inode->i_mapping->nrpages; |
576 | } | |
577 | spin_unlock(&bdev_lock); | |
578 | return ret; | |
579 | } | |
580 | ||
581 | void bdput(struct block_device *bdev) | |
582 | { | |
583 | iput(bdev->bd_inode); | |
584 | } | |
585 | ||
586 | EXPORT_SYMBOL(bdput); | |
587 | ||
588 | static struct block_device *bd_acquire(struct inode *inode) | |
589 | { | |
590 | struct block_device *bdev; | |
09d967c6 | 591 | |
1da177e4 LT |
592 | spin_lock(&bdev_lock); |
593 | bdev = inode->i_bdev; | |
09d967c6 | 594 | if (bdev) { |
7de9c6ee | 595 | ihold(bdev->bd_inode); |
1da177e4 LT |
596 | spin_unlock(&bdev_lock); |
597 | return bdev; | |
598 | } | |
599 | spin_unlock(&bdev_lock); | |
09d967c6 | 600 | |
1da177e4 LT |
601 | bdev = bdget(inode->i_rdev); |
602 | if (bdev) { | |
603 | spin_lock(&bdev_lock); | |
09d967c6 OH |
604 | if (!inode->i_bdev) { |
605 | /* | |
7de9c6ee | 606 | * We take an additional reference to bd_inode, |
09d967c6 OH |
607 | * and it's released in clear_inode() of inode. |
608 | * So, we can access it via ->i_mapping always | |
609 | * without igrab(). | |
610 | */ | |
7de9c6ee | 611 | ihold(bdev->bd_inode); |
09d967c6 OH |
612 | inode->i_bdev = bdev; |
613 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
614 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
615 | } | |
1da177e4 LT |
616 | spin_unlock(&bdev_lock); |
617 | } | |
618 | return bdev; | |
619 | } | |
620 | ||
621 | /* Call when you free inode */ | |
622 | ||
623 | void bd_forget(struct inode *inode) | |
624 | { | |
09d967c6 OH |
625 | struct block_device *bdev = NULL; |
626 | ||
1da177e4 | 627 | spin_lock(&bdev_lock); |
09d967c6 | 628 | if (inode->i_bdev) { |
c2acf7b9 | 629 | if (!sb_is_blkdev_sb(inode->i_sb)) |
09d967c6 | 630 | bdev = inode->i_bdev; |
1da177e4 | 631 | __bd_forget(inode); |
09d967c6 | 632 | } |
1da177e4 | 633 | spin_unlock(&bdev_lock); |
09d967c6 OH |
634 | |
635 | if (bdev) | |
636 | iput(bdev->bd_inode); | |
1da177e4 LT |
637 | } |
638 | ||
1a3cbbc5 TH |
639 | /** |
640 | * bd_may_claim - test whether a block device can be claimed | |
641 | * @bdev: block device of interest | |
642 | * @whole: whole block device containing @bdev, may equal @bdev | |
643 | * @holder: holder trying to claim @bdev | |
644 | * | |
645 | * Test whther @bdev can be claimed by @holder. | |
646 | * | |
647 | * CONTEXT: | |
648 | * spin_lock(&bdev_lock). | |
649 | * | |
650 | * RETURNS: | |
651 | * %true if @bdev can be claimed, %false otherwise. | |
652 | */ | |
653 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
654 | void *holder) | |
1da177e4 | 655 | { |
1da177e4 | 656 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 657 | return true; /* already a holder */ |
1da177e4 | 658 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 659 | return false; /* held by someone else */ |
1da177e4 | 660 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 661 | return true; /* is a whole device which isn't held */ |
1da177e4 | 662 | |
e525fd89 | 663 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
664 | return true; /* is a partition of a device that is being partitioned */ |
665 | else if (whole->bd_holder != NULL) | |
666 | return false; /* is a partition of a held device */ | |
1da177e4 | 667 | else |
1a3cbbc5 TH |
668 | return true; /* is a partition of an un-held device */ |
669 | } | |
670 | ||
6b4517a7 TH |
671 | /** |
672 | * bd_prepare_to_claim - prepare to claim a block device | |
673 | * @bdev: block device of interest | |
674 | * @whole: the whole device containing @bdev, may equal @bdev | |
675 | * @holder: holder trying to claim @bdev | |
676 | * | |
677 | * Prepare to claim @bdev. This function fails if @bdev is already | |
678 | * claimed by another holder and waits if another claiming is in | |
679 | * progress. This function doesn't actually claim. On successful | |
680 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
681 | * | |
682 | * CONTEXT: | |
683 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
684 | * it multiple times. | |
685 | * | |
686 | * RETURNS: | |
687 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
688 | */ | |
689 | static int bd_prepare_to_claim(struct block_device *bdev, | |
690 | struct block_device *whole, void *holder) | |
691 | { | |
692 | retry: | |
693 | /* if someone else claimed, fail */ | |
694 | if (!bd_may_claim(bdev, whole, holder)) | |
695 | return -EBUSY; | |
696 | ||
e75aa858 TH |
697 | /* if claiming is already in progress, wait for it to finish */ |
698 | if (whole->bd_claiming) { | |
6b4517a7 TH |
699 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
700 | DEFINE_WAIT(wait); | |
701 | ||
702 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
703 | spin_unlock(&bdev_lock); | |
704 | schedule(); | |
705 | finish_wait(wq, &wait); | |
706 | spin_lock(&bdev_lock); | |
707 | goto retry; | |
708 | } | |
709 | ||
710 | /* yay, all mine */ | |
711 | return 0; | |
712 | } | |
713 | ||
714 | /** | |
715 | * bd_start_claiming - start claiming a block device | |
716 | * @bdev: block device of interest | |
717 | * @holder: holder trying to claim @bdev | |
718 | * | |
719 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
720 | * exclusively and mark that an exclusive open is in progress. Each | |
721 | * successful call to this function must be matched with a call to | |
b0018361 NP |
722 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
723 | * fail). | |
724 | * | |
725 | * This function is used to gain exclusive access to the block device | |
726 | * without actually causing other exclusive open attempts to fail. It | |
727 | * should be used when the open sequence itself requires exclusive | |
728 | * access but may subsequently fail. | |
6b4517a7 TH |
729 | * |
730 | * CONTEXT: | |
731 | * Might sleep. | |
732 | * | |
733 | * RETURNS: | |
734 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
735 | * value on failure. | |
736 | */ | |
737 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
738 | void *holder) | |
739 | { | |
740 | struct gendisk *disk; | |
741 | struct block_device *whole; | |
742 | int partno, err; | |
743 | ||
744 | might_sleep(); | |
745 | ||
746 | /* | |
747 | * @bdev might not have been initialized properly yet, look up | |
748 | * and grab the outer block device the hard way. | |
749 | */ | |
750 | disk = get_gendisk(bdev->bd_dev, &partno); | |
751 | if (!disk) | |
752 | return ERR_PTR(-ENXIO); | |
753 | ||
754 | whole = bdget_disk(disk, 0); | |
cf342570 | 755 | module_put(disk->fops->owner); |
6b4517a7 TH |
756 | put_disk(disk); |
757 | if (!whole) | |
758 | return ERR_PTR(-ENOMEM); | |
759 | ||
760 | /* prepare to claim, if successful, mark claiming in progress */ | |
761 | spin_lock(&bdev_lock); | |
762 | ||
763 | err = bd_prepare_to_claim(bdev, whole, holder); | |
764 | if (err == 0) { | |
765 | whole->bd_claiming = holder; | |
766 | spin_unlock(&bdev_lock); | |
767 | return whole; | |
768 | } else { | |
769 | spin_unlock(&bdev_lock); | |
770 | bdput(whole); | |
771 | return ERR_PTR(err); | |
772 | } | |
773 | } | |
774 | ||
641dc636 | 775 | #ifdef CONFIG_SYSFS |
4d7dd8fd | 776 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 777 | { |
4d7dd8fd | 778 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
779 | } |
780 | ||
781 | static void del_symlink(struct kobject *from, struct kobject *to) | |
782 | { | |
641dc636 JN |
783 | sysfs_remove_link(from, kobject_name(to)); |
784 | } | |
785 | ||
df6c0cd9 | 786 | /** |
e09b457b TH |
787 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
788 | * @bdev: the claimed slave bdev | |
789 | * @disk: the holding disk | |
df6c0cd9 | 790 | * |
e09b457b | 791 | * This functions creates the following sysfs symlinks. |
df6c0cd9 | 792 | * |
e09b457b TH |
793 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
794 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 795 | * |
e09b457b TH |
796 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
797 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 798 | * |
e09b457b TH |
799 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
800 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 801 | * |
e09b457b TH |
802 | * The caller must have claimed @bdev before calling this function and |
803 | * ensure that both @bdev and @disk are valid during the creation and | |
804 | * lifetime of these symlinks. | |
641dc636 | 805 | * |
e09b457b TH |
806 | * CONTEXT: |
807 | * Might sleep. | |
641dc636 | 808 | * |
e09b457b TH |
809 | * RETURNS: |
810 | * 0 on success, -errno on failure. | |
641dc636 | 811 | */ |
e09b457b | 812 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 813 | { |
e09b457b | 814 | int ret = 0; |
641dc636 | 815 | |
2e7b651d | 816 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 817 | |
e09b457b | 818 | WARN_ON_ONCE(!bdev->bd_holder || bdev->bd_holder_disk); |
4e91672c | 819 | |
e09b457b TH |
820 | /* FIXME: remove the following once add_disk() handles errors */ |
821 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
822 | goto out_unlock; | |
4e91672c | 823 | |
e09b457b TH |
824 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); |
825 | if (ret) | |
826 | goto out_unlock; | |
641dc636 | 827 | |
e09b457b TH |
828 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); |
829 | if (ret) { | |
830 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
831 | goto out_unlock; | |
832 | } | |
641dc636 | 833 | |
e09b457b TH |
834 | bdev->bd_holder_disk = disk; |
835 | out_unlock: | |
b4cf1b72 | 836 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 837 | return ret; |
641dc636 | 838 | } |
e09b457b | 839 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 840 | |
e525fd89 | 841 | static void bd_unlink_disk_holder(struct block_device *bdev) |
641dc636 | 842 | { |
e09b457b | 843 | struct gendisk *disk = bdev->bd_holder_disk; |
641dc636 | 844 | |
e09b457b TH |
845 | bdev->bd_holder_disk = NULL; |
846 | if (!disk) | |
847 | return; | |
848 | ||
849 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
850 | del_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
641dc636 | 851 | } |
e525fd89 TH |
852 | #else |
853 | static inline void bd_unlink_disk_holder(struct block_device *bdev) | |
854 | { } | |
641dc636 JN |
855 | #endif |
856 | ||
1da177e4 LT |
857 | /* |
858 | * Tries to open block device by device number. Use it ONLY if you | |
859 | * really do not have anything better - i.e. when you are behind a | |
860 | * truly sucky interface and all you are given is a device number. _Never_ | |
861 | * to be used for internal purposes. If you ever need it - reconsider | |
862 | * your API. | |
863 | */ | |
e525fd89 | 864 | struct block_device *open_by_devnum(dev_t dev, fmode_t mode, void *holder) |
1da177e4 LT |
865 | { |
866 | struct block_device *bdev = bdget(dev); | |
867 | int err = -ENOMEM; | |
1da177e4 | 868 | if (bdev) |
e525fd89 | 869 | err = blkdev_get(bdev, mode, holder); |
1da177e4 LT |
870 | return err ? ERR_PTR(err) : bdev; |
871 | } | |
872 | ||
873 | EXPORT_SYMBOL(open_by_devnum); | |
874 | ||
56ade44b AP |
875 | /** |
876 | * flush_disk - invalidates all buffer-cache entries on a disk | |
877 | * | |
878 | * @bdev: struct block device to be flushed | |
879 | * | |
880 | * Invalidates all buffer-cache entries on a disk. It should be called | |
881 | * when a disk has been changed -- either by a media change or online | |
882 | * resize. | |
883 | */ | |
884 | static void flush_disk(struct block_device *bdev) | |
885 | { | |
886 | if (__invalidate_device(bdev)) { | |
887 | char name[BDEVNAME_SIZE] = ""; | |
888 | ||
889 | if (bdev->bd_disk) | |
890 | disk_name(bdev->bd_disk, 0, name); | |
891 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
892 | "resized disk %s\n", name); | |
893 | } | |
894 | ||
895 | if (!bdev->bd_disk) | |
896 | return; | |
897 | if (disk_partitionable(bdev->bd_disk)) | |
898 | bdev->bd_invalidated = 1; | |
899 | } | |
900 | ||
c3279d14 | 901 | /** |
57d1b536 | 902 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
903 | * @disk: struct gendisk to check |
904 | * @bdev: struct bdev to adjust. | |
905 | * | |
906 | * This routine checks to see if the bdev size does not match the disk size | |
907 | * and adjusts it if it differs. | |
908 | */ | |
909 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
910 | { | |
911 | loff_t disk_size, bdev_size; | |
912 | ||
913 | disk_size = (loff_t)get_capacity(disk) << 9; | |
914 | bdev_size = i_size_read(bdev->bd_inode); | |
915 | if (disk_size != bdev_size) { | |
916 | char name[BDEVNAME_SIZE]; | |
917 | ||
918 | disk_name(disk, 0, name); | |
919 | printk(KERN_INFO | |
920 | "%s: detected capacity change from %lld to %lld\n", | |
921 | name, bdev_size, disk_size); | |
922 | i_size_write(bdev->bd_inode, disk_size); | |
608aeef1 | 923 | flush_disk(bdev); |
c3279d14 AP |
924 | } |
925 | } | |
926 | EXPORT_SYMBOL(check_disk_size_change); | |
927 | ||
0c002c2f | 928 | /** |
57d1b536 | 929 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
930 | * @disk: struct gendisk to be revalidated |
931 | * | |
932 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
933 | * call-backs. It is used to do common pre and post operations needed | |
934 | * for all revalidate_disk operations. | |
935 | */ | |
936 | int revalidate_disk(struct gendisk *disk) | |
937 | { | |
c3279d14 | 938 | struct block_device *bdev; |
0c002c2f AP |
939 | int ret = 0; |
940 | ||
941 | if (disk->fops->revalidate_disk) | |
942 | ret = disk->fops->revalidate_disk(disk); | |
943 | ||
c3279d14 AP |
944 | bdev = bdget_disk(disk, 0); |
945 | if (!bdev) | |
946 | return ret; | |
947 | ||
948 | mutex_lock(&bdev->bd_mutex); | |
949 | check_disk_size_change(disk, bdev); | |
950 | mutex_unlock(&bdev->bd_mutex); | |
951 | bdput(bdev); | |
0c002c2f AP |
952 | return ret; |
953 | } | |
954 | EXPORT_SYMBOL(revalidate_disk); | |
955 | ||
1da177e4 LT |
956 | /* |
957 | * This routine checks whether a removable media has been changed, | |
958 | * and invalidates all buffer-cache-entries in that case. This | |
959 | * is a relatively slow routine, so we have to try to minimize using | |
960 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
961 | * is the best way of combining speed and utility, I think. | |
962 | * People changing diskettes in the middle of an operation deserve | |
963 | * to lose :-) | |
964 | */ | |
965 | int check_disk_change(struct block_device *bdev) | |
966 | { | |
967 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 968 | const struct block_device_operations *bdops = disk->fops; |
1da177e4 LT |
969 | |
970 | if (!bdops->media_changed) | |
971 | return 0; | |
972 | if (!bdops->media_changed(bdev->bd_disk)) | |
973 | return 0; | |
974 | ||
56ade44b | 975 | flush_disk(bdev); |
1da177e4 LT |
976 | if (bdops->revalidate_disk) |
977 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
978 | return 1; |
979 | } | |
980 | ||
981 | EXPORT_SYMBOL(check_disk_change); | |
982 | ||
983 | void bd_set_size(struct block_device *bdev, loff_t size) | |
984 | { | |
e1defc4f | 985 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 LT |
986 | |
987 | bdev->bd_inode->i_size = size; | |
988 | while (bsize < PAGE_CACHE_SIZE) { | |
989 | if (size & bsize) | |
990 | break; | |
991 | bsize <<= 1; | |
992 | } | |
993 | bdev->bd_block_size = bsize; | |
994 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
995 | } | |
996 | EXPORT_SYMBOL(bd_set_size); | |
997 | ||
9a1c3542 | 998 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 999 | |
6d740cd5 PZ |
1000 | /* |
1001 | * bd_mutex locking: | |
1002 | * | |
1003 | * mutex_lock(part->bd_mutex) | |
1004 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1005 | */ | |
1006 | ||
572c4892 | 1007 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1008 | { |
1da177e4 | 1009 | struct gendisk *disk; |
7db9cfd3 | 1010 | int ret; |
cf771cb5 | 1011 | int partno; |
fe6e9c1f AV |
1012 | int perm = 0; |
1013 | ||
572c4892 | 1014 | if (mode & FMODE_READ) |
fe6e9c1f | 1015 | perm |= MAY_READ; |
572c4892 | 1016 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1017 | perm |= MAY_WRITE; |
1018 | /* | |
1019 | * hooks: /n/, see "layering violations". | |
1020 | */ | |
b7300b78 CW |
1021 | if (!for_part) { |
1022 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1023 | if (ret != 0) { | |
1024 | bdput(bdev); | |
1025 | return ret; | |
1026 | } | |
82666020 | 1027 | } |
7db9cfd3 | 1028 | |
d3374825 | 1029 | restart: |
0762b8bd | 1030 | |
89f97496 | 1031 | ret = -ENXIO; |
cf771cb5 | 1032 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1033 | if (!disk) |
6e9624b8 | 1034 | goto out; |
1da177e4 | 1035 | |
6796bf54 | 1036 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1037 | if (!bdev->bd_openers) { |
1038 | bdev->bd_disk = disk; | |
1039 | bdev->bd_contains = bdev; | |
cf771cb5 | 1040 | if (!partno) { |
1da177e4 | 1041 | struct backing_dev_info *bdi; |
89f97496 TH |
1042 | |
1043 | ret = -ENXIO; | |
1044 | bdev->bd_part = disk_get_part(disk, partno); | |
1045 | if (!bdev->bd_part) | |
1046 | goto out_clear; | |
1047 | ||
1da177e4 | 1048 | if (disk->fops->open) { |
572c4892 | 1049 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1050 | if (ret == -ERESTARTSYS) { |
1051 | /* Lost a race with 'disk' being | |
1052 | * deleted, try again. | |
1053 | * See md.c | |
1054 | */ | |
1055 | disk_put_part(bdev->bd_part); | |
1056 | bdev->bd_part = NULL; | |
1057 | module_put(disk->fops->owner); | |
1058 | put_disk(disk); | |
1059 | bdev->bd_disk = NULL; | |
1060 | mutex_unlock(&bdev->bd_mutex); | |
1061 | goto restart; | |
1062 | } | |
1da177e4 | 1063 | if (ret) |
0762b8bd | 1064 | goto out_clear; |
1da177e4 LT |
1065 | } |
1066 | if (!bdev->bd_openers) { | |
1067 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); | |
1068 | bdi = blk_get_backing_dev_info(bdev); | |
1069 | if (bdi == NULL) | |
1070 | bdi = &default_backing_dev_info; | |
a5491e0c | 1071 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); |
1da177e4 LT |
1072 | } |
1073 | if (bdev->bd_invalidated) | |
1074 | rescan_partitions(disk, bdev); | |
1075 | } else { | |
1da177e4 LT |
1076 | struct block_device *whole; |
1077 | whole = bdget_disk(disk, 0); | |
1078 | ret = -ENOMEM; | |
1079 | if (!whole) | |
0762b8bd | 1080 | goto out_clear; |
37be4124 | 1081 | BUG_ON(for_part); |
572c4892 | 1082 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1083 | if (ret) |
0762b8bd | 1084 | goto out_clear; |
1da177e4 | 1085 | bdev->bd_contains = whole; |
a5491e0c DC |
1086 | bdev_inode_switch_bdi(bdev->bd_inode, |
1087 | whole->bd_inode->i_data.backing_dev_info); | |
89f97496 | 1088 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1089 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1090 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1091 | ret = -ENXIO; |
0762b8bd | 1092 | goto out_clear; |
1da177e4 | 1093 | } |
89f97496 | 1094 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 LT |
1095 | } |
1096 | } else { | |
0762b8bd | 1097 | module_put(disk->fops->owner); |
960cc0f4 | 1098 | put_disk(disk); |
0762b8bd | 1099 | disk = NULL; |
1da177e4 LT |
1100 | if (bdev->bd_contains == bdev) { |
1101 | if (bdev->bd_disk->fops->open) { | |
572c4892 | 1102 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1da177e4 | 1103 | if (ret) |
0762b8bd | 1104 | goto out_unlock_bdev; |
1da177e4 LT |
1105 | } |
1106 | if (bdev->bd_invalidated) | |
1107 | rescan_partitions(bdev->bd_disk, bdev); | |
1da177e4 LT |
1108 | } |
1109 | } | |
1110 | bdev->bd_openers++; | |
37be4124 N |
1111 | if (for_part) |
1112 | bdev->bd_part_count++; | |
c039e313 | 1113 | mutex_unlock(&bdev->bd_mutex); |
1da177e4 LT |
1114 | return 0; |
1115 | ||
0762b8bd | 1116 | out_clear: |
89f97496 | 1117 | disk_put_part(bdev->bd_part); |
1da177e4 | 1118 | bdev->bd_disk = NULL; |
0762b8bd | 1119 | bdev->bd_part = NULL; |
a5491e0c | 1120 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); |
1da177e4 | 1121 | if (bdev != bdev->bd_contains) |
572c4892 | 1122 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1123 | bdev->bd_contains = NULL; |
0762b8bd | 1124 | out_unlock_bdev: |
c039e313 | 1125 | mutex_unlock(&bdev->bd_mutex); |
6e9624b8 | 1126 | out: |
0762b8bd TH |
1127 | if (disk) |
1128 | module_put(disk->fops->owner); | |
1129 | put_disk(disk); | |
1130 | bdput(bdev); | |
1131 | ||
1da177e4 LT |
1132 | return ret; |
1133 | } | |
1134 | ||
e525fd89 | 1135 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1136 | { |
e525fd89 TH |
1137 | struct block_device *whole = NULL; |
1138 | int res; | |
1139 | ||
1140 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1141 | ||
1142 | if ((mode & FMODE_EXCL) && holder) { | |
1143 | whole = bd_start_claiming(bdev, holder); | |
1144 | if (IS_ERR(whole)) { | |
1145 | bdput(bdev); | |
1146 | return PTR_ERR(whole); | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | res = __blkdev_get(bdev, mode, 0); | |
1151 | ||
1152 | if (whole) { | |
6a027eff TH |
1153 | /* finish claiming */ |
1154 | spin_lock(&bdev_lock); | |
1155 | ||
1156 | if (res == 0) { | |
1157 | BUG_ON(!bd_may_claim(bdev, whole, holder)); | |
1158 | /* | |
1159 | * Note that for a whole device bd_holders | |
1160 | * will be incremented twice, and bd_holder | |
1161 | * will be set to bd_may_claim before being | |
1162 | * set to holder | |
1163 | */ | |
1164 | whole->bd_holders++; | |
1165 | whole->bd_holder = bd_may_claim; | |
1166 | bdev->bd_holders++; | |
1167 | bdev->bd_holder = holder; | |
1168 | } | |
1169 | ||
1170 | /* tell others that we're done */ | |
1171 | BUG_ON(whole->bd_claiming != holder); | |
1172 | whole->bd_claiming = NULL; | |
1173 | wake_up_bit(&whole->bd_claiming, 0); | |
1174 | ||
1175 | spin_unlock(&bdev_lock); | |
1176 | bdput(whole); | |
e525fd89 TH |
1177 | } |
1178 | ||
1179 | return res; | |
37be4124 | 1180 | } |
1da177e4 LT |
1181 | EXPORT_SYMBOL(blkdev_get); |
1182 | ||
1183 | static int blkdev_open(struct inode * inode, struct file * filp) | |
1184 | { | |
1185 | struct block_device *bdev; | |
1da177e4 LT |
1186 | |
1187 | /* | |
1188 | * Preserve backwards compatibility and allow large file access | |
1189 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1190 | * binary needs it. We might want to drop this workaround | |
1191 | * during an unstable branch. | |
1192 | */ | |
1193 | filp->f_flags |= O_LARGEFILE; | |
1194 | ||
572c4892 AV |
1195 | if (filp->f_flags & O_NDELAY) |
1196 | filp->f_mode |= FMODE_NDELAY; | |
1197 | if (filp->f_flags & O_EXCL) | |
1198 | filp->f_mode |= FMODE_EXCL; | |
1199 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1200 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1201 | ||
1da177e4 | 1202 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1203 | if (bdev == NULL) |
1204 | return -ENOMEM; | |
1da177e4 | 1205 | |
572c4892 AV |
1206 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1207 | ||
e525fd89 | 1208 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1209 | } |
1210 | ||
9a1c3542 | 1211 | static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d PZ |
1212 | { |
1213 | int ret = 0; | |
2e7b651d | 1214 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1215 | struct block_device *victim = NULL; |
2e7b651d | 1216 | |
6796bf54 | 1217 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1218 | if (for_part) |
1219 | bdev->bd_part_count--; | |
1220 | ||
2e7b651d | 1221 | if (!--bdev->bd_openers) { |
6a027eff | 1222 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1223 | sync_blockdev(bdev); |
1224 | kill_bdev(bdev); | |
1225 | } | |
1226 | if (bdev->bd_contains == bdev) { | |
1227 | if (disk->fops->release) | |
9a1c3542 | 1228 | ret = disk->fops->release(disk, mode); |
2e7b651d PZ |
1229 | } |
1230 | if (!bdev->bd_openers) { | |
1231 | struct module *owner = disk->fops->owner; | |
1232 | ||
1233 | put_disk(disk); | |
1234 | module_put(owner); | |
0762b8bd TH |
1235 | disk_put_part(bdev->bd_part); |
1236 | bdev->bd_part = NULL; | |
2e7b651d | 1237 | bdev->bd_disk = NULL; |
a5491e0c DC |
1238 | bdev_inode_switch_bdi(bdev->bd_inode, |
1239 | &default_backing_dev_info); | |
37be4124 N |
1240 | if (bdev != bdev->bd_contains) |
1241 | victim = bdev->bd_contains; | |
2e7b651d PZ |
1242 | bdev->bd_contains = NULL; |
1243 | } | |
2e7b651d PZ |
1244 | mutex_unlock(&bdev->bd_mutex); |
1245 | bdput(bdev); | |
37be4124 | 1246 | if (victim) |
9a1c3542 | 1247 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1248 | return ret; |
1249 | } | |
1250 | ||
9a1c3542 | 1251 | int blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1252 | { |
e525fd89 | 1253 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1254 | bool bdev_free; |
1255 | ||
1256 | /* | |
1257 | * Release a claim on the device. The holder fields | |
1258 | * are protected with bdev_lock. bd_mutex is to | |
1259 | * synchronize disk_holder unlinking. | |
1260 | */ | |
e525fd89 | 1261 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1262 | spin_lock(&bdev_lock); |
1263 | ||
1264 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1265 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1266 | ||
1267 | /* bd_contains might point to self, check in a separate step */ | |
1268 | if ((bdev_free = !bdev->bd_holders)) | |
1269 | bdev->bd_holder = NULL; | |
1270 | if (!bdev->bd_contains->bd_holders) | |
1271 | bdev->bd_contains->bd_holder = NULL; | |
1272 | ||
1273 | spin_unlock(&bdev_lock); | |
1274 | ||
1275 | /* if this was the last claim, holder link should go too */ | |
1276 | if (bdev_free) | |
e525fd89 | 1277 | bd_unlink_disk_holder(bdev); |
6a027eff | 1278 | |
e525fd89 TH |
1279 | mutex_unlock(&bdev->bd_mutex); |
1280 | } | |
9a1c3542 | 1281 | return __blkdev_put(bdev, mode, 0); |
37be4124 | 1282 | } |
2e7b651d PZ |
1283 | EXPORT_SYMBOL(blkdev_put); |
1284 | ||
1da177e4 LT |
1285 | static int blkdev_close(struct inode * inode, struct file * filp) |
1286 | { | |
1287 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
e525fd89 | 1288 | |
9a1c3542 | 1289 | return blkdev_put(bdev, filp->f_mode); |
1da177e4 LT |
1290 | } |
1291 | ||
bb93e3a5 | 1292 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1293 | { |
56b26add AV |
1294 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1295 | fmode_t mode = file->f_mode; | |
fd4ce1ac CH |
1296 | |
1297 | /* | |
1298 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1299 | * to updated it before every ioctl. | |
1300 | */ | |
56b26add | 1301 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1302 | mode |= FMODE_NDELAY; |
1303 | else | |
1304 | mode &= ~FMODE_NDELAY; | |
1305 | ||
56b26add | 1306 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1307 | } |
1308 | ||
eef99380 CH |
1309 | /* |
1310 | * Write data to the block device. Only intended for the block device itself | |
1311 | * and the raw driver which basically is a fake block device. | |
1312 | * | |
1313 | * Does not take i_mutex for the write and thus is not for general purpose | |
1314 | * use. | |
1315 | */ | |
1316 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1317 | unsigned long nr_segs, loff_t pos) | |
1318 | { | |
1319 | struct file *file = iocb->ki_filp; | |
1320 | ssize_t ret; | |
1321 | ||
1322 | BUG_ON(iocb->ki_pos != pos); | |
1323 | ||
1324 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); | |
1325 | if (ret > 0 || ret == -EIOCBQUEUED) { | |
1326 | ssize_t err; | |
1327 | ||
1328 | err = generic_write_sync(file, pos, ret); | |
1329 | if (err < 0 && ret > 0) | |
1330 | ret = err; | |
1331 | } | |
1332 | return ret; | |
1333 | } | |
1334 | EXPORT_SYMBOL_GPL(blkdev_aio_write); | |
1335 | ||
87d8fe1e TT |
1336 | /* |
1337 | * Try to release a page associated with block device when the system | |
1338 | * is under memory pressure. | |
1339 | */ | |
1340 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1341 | { | |
1342 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1343 | ||
1344 | if (super && super->s_op->bdev_try_to_free_page) | |
1345 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1346 | ||
1347 | return try_to_free_buffers(page); | |
1348 | } | |
1349 | ||
4c54ac62 | 1350 | static const struct address_space_operations def_blk_aops = { |
1da177e4 LT |
1351 | .readpage = blkdev_readpage, |
1352 | .writepage = blkdev_writepage, | |
1353 | .sync_page = block_sync_page, | |
6272b5a5 NP |
1354 | .write_begin = blkdev_write_begin, |
1355 | .write_end = blkdev_write_end, | |
1da177e4 | 1356 | .writepages = generic_writepages, |
87d8fe1e | 1357 | .releasepage = blkdev_releasepage, |
1da177e4 LT |
1358 | .direct_IO = blkdev_direct_IO, |
1359 | }; | |
1360 | ||
4b6f5d20 | 1361 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1362 | .open = blkdev_open, |
1363 | .release = blkdev_close, | |
1364 | .llseek = block_llseek, | |
543ade1f BP |
1365 | .read = do_sync_read, |
1366 | .write = do_sync_write, | |
1da177e4 | 1367 | .aio_read = generic_file_aio_read, |
eef99380 | 1368 | .aio_write = blkdev_aio_write, |
1da177e4 | 1369 | .mmap = generic_file_mmap, |
b1dd3b28 | 1370 | .fsync = blkdev_fsync, |
bb93e3a5 | 1371 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1372 | #ifdef CONFIG_COMPAT |
1373 | .compat_ioctl = compat_blkdev_ioctl, | |
1374 | #endif | |
7f9c51f0 JA |
1375 | .splice_read = generic_file_splice_read, |
1376 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
1377 | }; |
1378 | ||
1379 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1380 | { | |
1381 | int res; | |
1382 | mm_segment_t old_fs = get_fs(); | |
1383 | set_fs(KERNEL_DS); | |
56b26add | 1384 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1385 | set_fs(old_fs); |
1386 | return res; | |
1387 | } | |
1388 | ||
1389 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1390 | ||
1391 | /** | |
1392 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1393 | * @pathname: special file representing the block device |
1da177e4 | 1394 | * |
57d1b536 | 1395 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1396 | * namespace if possible and return it. Return ERR_PTR(error) |
1397 | * otherwise. | |
1398 | */ | |
421748ec | 1399 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1400 | { |
1401 | struct block_device *bdev; | |
1402 | struct inode *inode; | |
421748ec | 1403 | struct path path; |
1da177e4 LT |
1404 | int error; |
1405 | ||
421748ec | 1406 | if (!pathname || !*pathname) |
1da177e4 LT |
1407 | return ERR_PTR(-EINVAL); |
1408 | ||
421748ec | 1409 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1410 | if (error) |
1411 | return ERR_PTR(error); | |
1412 | ||
421748ec | 1413 | inode = path.dentry->d_inode; |
1da177e4 LT |
1414 | error = -ENOTBLK; |
1415 | if (!S_ISBLK(inode->i_mode)) | |
1416 | goto fail; | |
1417 | error = -EACCES; | |
421748ec | 1418 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1419 | goto fail; |
1420 | error = -ENOMEM; | |
1421 | bdev = bd_acquire(inode); | |
1422 | if (!bdev) | |
1423 | goto fail; | |
1424 | out: | |
421748ec | 1425 | path_put(&path); |
1da177e4 LT |
1426 | return bdev; |
1427 | fail: | |
1428 | bdev = ERR_PTR(error); | |
1429 | goto out; | |
1430 | } | |
d5686b44 | 1431 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 LT |
1432 | |
1433 | /** | |
30c40d2c | 1434 | * open_bdev_exclusive - open a block device by name and set it up for use |
1da177e4 LT |
1435 | * |
1436 | * @path: special file representing the block device | |
30c40d2c | 1437 | * @mode: FMODE_... combination to pass be used |
1da177e4 LT |
1438 | * @holder: owner for exclusion |
1439 | * | |
1440 | * Open the blockdevice described by the special file at @path, claim it | |
1441 | * for the @holder. | |
1442 | */ | |
30c40d2c | 1443 | struct block_device *open_bdev_exclusive(const char *path, fmode_t mode, void *holder) |
1da177e4 | 1444 | { |
e525fd89 | 1445 | struct block_device *bdev; |
6b4517a7 | 1446 | int error; |
1da177e4 LT |
1447 | |
1448 | bdev = lookup_bdev(path); | |
1449 | if (IS_ERR(bdev)) | |
1450 | return bdev; | |
1451 | ||
e525fd89 | 1452 | error = blkdev_get(bdev, mode | FMODE_EXCL, holder); |
1da177e4 | 1453 | if (error) |
e525fd89 | 1454 | return ERR_PTR(error); |
6b4517a7 | 1455 | |
e525fd89 TH |
1456 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1457 | blkdev_put(bdev, mode); | |
1458 | return ERR_PTR(-EACCES); | |
1459 | } | |
1da177e4 LT |
1460 | |
1461 | return bdev; | |
1da177e4 LT |
1462 | } |
1463 | ||
30c40d2c | 1464 | EXPORT_SYMBOL(open_bdev_exclusive); |
1da177e4 | 1465 | |
b71e8a4c DH |
1466 | int __invalidate_device(struct block_device *bdev) |
1467 | { | |
1468 | struct super_block *sb = get_super(bdev); | |
1469 | int res = 0; | |
1470 | ||
1471 | if (sb) { | |
1472 | /* | |
1473 | * no need to lock the super, get_super holds the | |
1474 | * read mutex so the filesystem cannot go away | |
1475 | * under us (->put_super runs with the write lock | |
1476 | * hold). | |
1477 | */ | |
1478 | shrink_dcache_sb(sb); | |
1479 | res = invalidate_inodes(sb); | |
1480 | drop_super(sb); | |
1481 | } | |
f98393a6 | 1482 | invalidate_bdev(bdev); |
b71e8a4c DH |
1483 | return res; |
1484 | } | |
1485 | EXPORT_SYMBOL(__invalidate_device); |