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> | |
7db9cfd3 | 14 | #include <linux/device_cgroup.h> |
1da177e4 LT |
15 | #include <linux/highmem.h> |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/module.h> | |
18 | #include <linux/blkpg.h> | |
b502bd11 | 19 | #include <linux/magic.h> |
1da177e4 | 20 | #include <linux/buffer_head.h> |
ff01bb48 | 21 | #include <linux/swap.h> |
585d3bc0 | 22 | #include <linux/pagevec.h> |
811d736f | 23 | #include <linux/writeback.h> |
1da177e4 LT |
24 | #include <linux/mpage.h> |
25 | #include <linux/mount.h> | |
26 | #include <linux/uio.h> | |
27 | #include <linux/namei.h> | |
1368c4f2 | 28 | #include <linux/log2.h> |
ff01bb48 | 29 | #include <linux/cleancache.h> |
a27bb332 | 30 | #include <linux/aio.h> |
1da177e4 | 31 | #include <asm/uaccess.h> |
07f3f05c | 32 | #include "internal.h" |
1da177e4 LT |
33 | |
34 | struct bdev_inode { | |
35 | struct block_device bdev; | |
36 | struct inode vfs_inode; | |
37 | }; | |
38 | ||
4c54ac62 AB |
39 | static const struct address_space_operations def_blk_aops; |
40 | ||
1da177e4 LT |
41 | static inline struct bdev_inode *BDEV_I(struct inode *inode) |
42 | { | |
43 | return container_of(inode, struct bdev_inode, vfs_inode); | |
44 | } | |
45 | ||
46 | inline struct block_device *I_BDEV(struct inode *inode) | |
47 | { | |
48 | return &BDEV_I(inode)->bdev; | |
49 | } | |
1da177e4 LT |
50 | EXPORT_SYMBOL(I_BDEV); |
51 | ||
a5491e0c | 52 | /* |
f758eeab CH |
53 | * Move the inode from its current bdi to a new bdi. If the inode is dirty we |
54 | * need to move it onto the dirty list of @dst so that the inode is always on | |
55 | * the right list. | |
a5491e0c DC |
56 | */ |
57 | static void bdev_inode_switch_bdi(struct inode *inode, | |
58 | struct backing_dev_info *dst) | |
59 | { | |
f758eeab CH |
60 | struct backing_dev_info *old = inode->i_data.backing_dev_info; |
61 | ||
62 | if (unlikely(dst == old)) /* deadlock avoidance */ | |
63 | return; | |
64 | bdi_lock_two(&old->wb, &dst->wb); | |
250df6ed | 65 | spin_lock(&inode->i_lock); |
a5491e0c DC |
66 | inode->i_data.backing_dev_info = dst; |
67 | if (inode->i_state & I_DIRTY) | |
7ccf19a8 | 68 | list_move(&inode->i_wb_list, &dst->wb.b_dirty); |
250df6ed | 69 | spin_unlock(&inode->i_lock); |
f758eeab CH |
70 | spin_unlock(&old->wb.list_lock); |
71 | spin_unlock(&dst->wb.list_lock); | |
a5491e0c DC |
72 | } |
73 | ||
f9a14399 | 74 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
ff01bb48 | 75 | void kill_bdev(struct block_device *bdev) |
1da177e4 | 76 | { |
ff01bb48 AV |
77 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
78 | ||
79 | if (mapping->nrpages == 0) | |
f9a14399 | 80 | return; |
ff01bb48 | 81 | |
f9a14399 | 82 | invalidate_bh_lrus(); |
ff01bb48 | 83 | truncate_inode_pages(mapping, 0); |
1da177e4 | 84 | } |
ff01bb48 AV |
85 | EXPORT_SYMBOL(kill_bdev); |
86 | ||
87 | /* Invalidate clean unused buffers and pagecache. */ | |
88 | void invalidate_bdev(struct block_device *bdev) | |
89 | { | |
90 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
91 | ||
92 | if (mapping->nrpages == 0) | |
93 | return; | |
94 | ||
95 | invalidate_bh_lrus(); | |
96 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
97 | invalidate_mapping_pages(mapping, 0, -1); | |
98 | /* 99% of the time, we don't need to flush the cleancache on the bdev. | |
99 | * But, for the strange corners, lets be cautious | |
100 | */ | |
3167760f | 101 | cleancache_invalidate_inode(mapping); |
ff01bb48 AV |
102 | } |
103 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 LT |
104 | |
105 | int set_blocksize(struct block_device *bdev, int size) | |
106 | { | |
107 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 108 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
109 | return -EINVAL; |
110 | ||
111 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 112 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
113 | return -EINVAL; |
114 | ||
115 | /* Don't change the size if it is same as current */ | |
116 | if (bdev->bd_block_size != size) { | |
117 | sync_blockdev(bdev); | |
118 | bdev->bd_block_size = size; | |
119 | bdev->bd_inode->i_blkbits = blksize_bits(size); | |
120 | kill_bdev(bdev); | |
121 | } | |
122 | return 0; | |
123 | } | |
124 | ||
125 | EXPORT_SYMBOL(set_blocksize); | |
126 | ||
127 | int sb_set_blocksize(struct super_block *sb, int size) | |
128 | { | |
1da177e4 LT |
129 | if (set_blocksize(sb->s_bdev, size)) |
130 | return 0; | |
131 | /* If we get here, we know size is power of two | |
132 | * and it's value is between 512 and PAGE_SIZE */ | |
133 | sb->s_blocksize = size; | |
38885bd4 | 134 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
135 | return sb->s_blocksize; |
136 | } | |
137 | ||
138 | EXPORT_SYMBOL(sb_set_blocksize); | |
139 | ||
140 | int sb_min_blocksize(struct super_block *sb, int size) | |
141 | { | |
e1defc4f | 142 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
143 | if (size < minsize) |
144 | size = minsize; | |
145 | return sb_set_blocksize(sb, size); | |
146 | } | |
147 | ||
148 | EXPORT_SYMBOL(sb_min_blocksize); | |
149 | ||
150 | static int | |
151 | blkdev_get_block(struct inode *inode, sector_t iblock, | |
152 | struct buffer_head *bh, int create) | |
153 | { | |
1da177e4 LT |
154 | bh->b_bdev = I_BDEV(inode); |
155 | bh->b_blocknr = iblock; | |
156 | set_buffer_mapped(bh); | |
157 | return 0; | |
158 | } | |
159 | ||
b2e895db AM |
160 | static ssize_t |
161 | blkdev_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
162 | loff_t offset, unsigned long nr_segs) | |
163 | { | |
164 | struct file *file = iocb->ki_filp; | |
165 | struct inode *inode = file->f_mapping->host; | |
166 | ||
eafdc7d1 | 167 | return __blockdev_direct_IO(rw, iocb, inode, I_BDEV(inode), iov, offset, |
bbec0270 | 168 | nr_segs, blkdev_get_block, NULL, NULL, 0); |
b2e895db AM |
169 | } |
170 | ||
5cee5815 JK |
171 | int __sync_blockdev(struct block_device *bdev, int wait) |
172 | { | |
173 | if (!bdev) | |
174 | return 0; | |
175 | if (!wait) | |
176 | return filemap_flush(bdev->bd_inode->i_mapping); | |
177 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
178 | } | |
179 | ||
585d3bc0 NP |
180 | /* |
181 | * Write out and wait upon all the dirty data associated with a block | |
182 | * device via its mapping. Does not take the superblock lock. | |
183 | */ | |
184 | int sync_blockdev(struct block_device *bdev) | |
185 | { | |
5cee5815 | 186 | return __sync_blockdev(bdev, 1); |
585d3bc0 NP |
187 | } |
188 | EXPORT_SYMBOL(sync_blockdev); | |
189 | ||
190 | /* | |
191 | * Write out and wait upon all dirty data associated with this | |
192 | * device. Filesystem data as well as the underlying block | |
193 | * device. Takes the superblock lock. | |
194 | */ | |
195 | int fsync_bdev(struct block_device *bdev) | |
196 | { | |
197 | struct super_block *sb = get_super(bdev); | |
198 | if (sb) { | |
60b0680f | 199 | int res = sync_filesystem(sb); |
585d3bc0 NP |
200 | drop_super(sb); |
201 | return res; | |
202 | } | |
203 | return sync_blockdev(bdev); | |
204 | } | |
47e4491b | 205 | EXPORT_SYMBOL(fsync_bdev); |
585d3bc0 NP |
206 | |
207 | /** | |
208 | * freeze_bdev -- lock a filesystem and force it into a consistent state | |
209 | * @bdev: blockdevice to lock | |
210 | * | |
585d3bc0 NP |
211 | * If a superblock is found on this device, we take the s_umount semaphore |
212 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
213 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
214 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
215 | * freeze requests arrive simultaneously. It counts up in freeze_bdev() and | |
216 | * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze | |
217 | * actually. | |
218 | */ | |
219 | struct super_block *freeze_bdev(struct block_device *bdev) | |
220 | { | |
221 | struct super_block *sb; | |
222 | int error = 0; | |
223 | ||
224 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a CH |
225 | if (++bdev->bd_fsfreeze_count > 1) { |
226 | /* | |
227 | * We don't even need to grab a reference - the first call | |
228 | * to freeze_bdev grab an active reference and only the last | |
229 | * thaw_bdev drops it. | |
230 | */ | |
585d3bc0 | 231 | sb = get_super(bdev); |
4504230a CH |
232 | drop_super(sb); |
233 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
234 | return sb; | |
235 | } | |
236 | ||
237 | sb = get_active_super(bdev); | |
238 | if (!sb) | |
239 | goto out; | |
18e9e510 JB |
240 | error = freeze_super(sb); |
241 | if (error) { | |
242 | deactivate_super(sb); | |
243 | bdev->bd_fsfreeze_count--; | |
585d3bc0 | 244 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
18e9e510 | 245 | return ERR_PTR(error); |
585d3bc0 | 246 | } |
18e9e510 | 247 | deactivate_super(sb); |
4504230a | 248 | out: |
585d3bc0 NP |
249 | sync_blockdev(bdev); |
250 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
4fadd7bb | 251 | return sb; /* thaw_bdev releases s->s_umount */ |
585d3bc0 NP |
252 | } |
253 | EXPORT_SYMBOL(freeze_bdev); | |
254 | ||
255 | /** | |
256 | * thaw_bdev -- unlock filesystem | |
257 | * @bdev: blockdevice to unlock | |
258 | * @sb: associated superblock | |
259 | * | |
260 | * Unlocks the filesystem and marks it writeable again after freeze_bdev(). | |
261 | */ | |
262 | int thaw_bdev(struct block_device *bdev, struct super_block *sb) | |
263 | { | |
4504230a | 264 | int error = -EINVAL; |
585d3bc0 NP |
265 | |
266 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
4504230a | 267 | if (!bdev->bd_fsfreeze_count) |
18e9e510 | 268 | goto out; |
4504230a CH |
269 | |
270 | error = 0; | |
271 | if (--bdev->bd_fsfreeze_count > 0) | |
18e9e510 | 272 | goto out; |
4504230a CH |
273 | |
274 | if (!sb) | |
18e9e510 | 275 | goto out; |
4504230a | 276 | |
18e9e510 JB |
277 | error = thaw_super(sb); |
278 | if (error) { | |
279 | bdev->bd_fsfreeze_count++; | |
280 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
281 | return error; | |
282 | } | |
283 | out: | |
585d3bc0 NP |
284 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
285 | return 0; | |
286 | } | |
287 | EXPORT_SYMBOL(thaw_bdev); | |
288 | ||
1da177e4 LT |
289 | static int blkdev_writepage(struct page *page, struct writeback_control *wbc) |
290 | { | |
291 | return block_write_full_page(page, blkdev_get_block, wbc); | |
292 | } | |
293 | ||
294 | static int blkdev_readpage(struct file * file, struct page * page) | |
295 | { | |
296 | return block_read_full_page(page, blkdev_get_block); | |
297 | } | |
298 | ||
6272b5a5 NP |
299 | static int blkdev_write_begin(struct file *file, struct address_space *mapping, |
300 | loff_t pos, unsigned len, unsigned flags, | |
301 | struct page **pagep, void **fsdata) | |
1da177e4 | 302 | { |
155130a4 CH |
303 | return block_write_begin(mapping, pos, len, flags, pagep, |
304 | blkdev_get_block); | |
1da177e4 LT |
305 | } |
306 | ||
6272b5a5 NP |
307 | static int blkdev_write_end(struct file *file, struct address_space *mapping, |
308 | loff_t pos, unsigned len, unsigned copied, | |
309 | struct page *page, void *fsdata) | |
1da177e4 | 310 | { |
6272b5a5 NP |
311 | int ret; |
312 | ret = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
313 | ||
314 | unlock_page(page); | |
315 | page_cache_release(page); | |
316 | ||
317 | return ret; | |
1da177e4 LT |
318 | } |
319 | ||
320 | /* | |
321 | * private llseek: | |
496ad9aa | 322 | * for a block special file file_inode(file)->i_size is zero |
1da177e4 LT |
323 | * so we compute the size by hand (just as in block_read/write above) |
324 | */ | |
965c8e59 | 325 | static loff_t block_llseek(struct file *file, loff_t offset, int whence) |
1da177e4 LT |
326 | { |
327 | struct inode *bd_inode = file->f_mapping->host; | |
1da177e4 LT |
328 | loff_t retval; |
329 | ||
1b1dcc1b | 330 | mutex_lock(&bd_inode->i_mutex); |
5d48f3a2 | 331 | retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode)); |
1b1dcc1b | 332 | mutex_unlock(&bd_inode->i_mutex); |
1da177e4 LT |
333 | return retval; |
334 | } | |
335 | ||
02c24a82 | 336 | int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync) |
1da177e4 | 337 | { |
b8af67e2 AB |
338 | struct inode *bd_inode = filp->f_mapping->host; |
339 | struct block_device *bdev = I_BDEV(bd_inode); | |
ab0a9735 | 340 | int error; |
da5aa861 RW |
341 | |
342 | error = filemap_write_and_wait_range(filp->f_mapping, start, end); | |
343 | if (error) | |
344 | return error; | |
ab0a9735 | 345 | |
b8af67e2 AB |
346 | /* |
347 | * There is no need to serialise calls to blkdev_issue_flush with | |
348 | * i_mutex and doing so causes performance issues with concurrent | |
349 | * O_SYNC writers to a block device. | |
350 | */ | |
dd3932ed | 351 | error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL); |
ab0a9735 CH |
352 | if (error == -EOPNOTSUPP) |
353 | error = 0; | |
b8af67e2 | 354 | |
ab0a9735 | 355 | return error; |
1da177e4 | 356 | } |
b1dd3b28 | 357 | EXPORT_SYMBOL(blkdev_fsync); |
1da177e4 LT |
358 | |
359 | /* | |
360 | * pseudo-fs | |
361 | */ | |
362 | ||
363 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock); | |
e18b890b | 364 | static struct kmem_cache * bdev_cachep __read_mostly; |
1da177e4 LT |
365 | |
366 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
367 | { | |
e94b1766 | 368 | struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL); |
1da177e4 LT |
369 | if (!ei) |
370 | return NULL; | |
371 | return &ei->vfs_inode; | |
372 | } | |
373 | ||
fa0d7e3d | 374 | static void bdev_i_callback(struct rcu_head *head) |
1da177e4 | 375 | { |
fa0d7e3d | 376 | struct inode *inode = container_of(head, struct inode, i_rcu); |
1da177e4 LT |
377 | struct bdev_inode *bdi = BDEV_I(inode); |
378 | ||
1da177e4 LT |
379 | kmem_cache_free(bdev_cachep, bdi); |
380 | } | |
381 | ||
fa0d7e3d NP |
382 | static void bdev_destroy_inode(struct inode *inode) |
383 | { | |
384 | call_rcu(&inode->i_rcu, bdev_i_callback); | |
385 | } | |
386 | ||
51cc5068 | 387 | static void init_once(void *foo) |
1da177e4 LT |
388 | { |
389 | struct bdev_inode *ei = (struct bdev_inode *) foo; | |
390 | struct block_device *bdev = &ei->bdev; | |
391 | ||
a35afb83 CL |
392 | memset(bdev, 0, sizeof(*bdev)); |
393 | mutex_init(&bdev->bd_mutex); | |
a35afb83 CL |
394 | INIT_LIST_HEAD(&bdev->bd_inodes); |
395 | INIT_LIST_HEAD(&bdev->bd_list); | |
49731baa TH |
396 | #ifdef CONFIG_SYSFS |
397 | INIT_LIST_HEAD(&bdev->bd_holder_disks); | |
398 | #endif | |
a35afb83 | 399 | inode_init_once(&ei->vfs_inode); |
fcccf502 TS |
400 | /* Initialize mutex for freeze. */ |
401 | mutex_init(&bdev->bd_fsfreeze_mutex); | |
1da177e4 LT |
402 | } |
403 | ||
404 | static inline void __bd_forget(struct inode *inode) | |
405 | { | |
406 | list_del_init(&inode->i_devices); | |
407 | inode->i_bdev = NULL; | |
408 | inode->i_mapping = &inode->i_data; | |
409 | } | |
410 | ||
b57922d9 | 411 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 LT |
412 | { |
413 | struct block_device *bdev = &BDEV_I(inode)->bdev; | |
414 | struct list_head *p; | |
b57922d9 AV |
415 | truncate_inode_pages(&inode->i_data, 0); |
416 | invalidate_inode_buffers(inode); /* is it needed here? */ | |
dbd5768f | 417 | clear_inode(inode); |
1da177e4 LT |
418 | spin_lock(&bdev_lock); |
419 | while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { | |
420 | __bd_forget(list_entry(p, struct inode, i_devices)); | |
421 | } | |
422 | list_del_init(&bdev->bd_list); | |
423 | spin_unlock(&bdev_lock); | |
424 | } | |
425 | ||
ee9b6d61 | 426 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
427 | .statfs = simple_statfs, |
428 | .alloc_inode = bdev_alloc_inode, | |
429 | .destroy_inode = bdev_destroy_inode, | |
430 | .drop_inode = generic_delete_inode, | |
b57922d9 | 431 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
432 | }; |
433 | ||
51139ada AV |
434 | static struct dentry *bd_mount(struct file_system_type *fs_type, |
435 | int flags, const char *dev_name, void *data) | |
1da177e4 | 436 | { |
b502bd11 | 437 | return mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC); |
1da177e4 LT |
438 | } |
439 | ||
440 | static struct file_system_type bd_type = { | |
441 | .name = "bdev", | |
51139ada | 442 | .mount = bd_mount, |
1da177e4 LT |
443 | .kill_sb = kill_anon_super, |
444 | }; | |
445 | ||
f47ec3f2 | 446 | static struct super_block *blockdev_superblock __read_mostly; |
1da177e4 LT |
447 | |
448 | void __init bdev_cache_init(void) | |
449 | { | |
450 | int err; | |
ace8577a | 451 | static struct vfsmount *bd_mnt; |
c2acf7b9 | 452 | |
1da177e4 | 453 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 PJ |
454 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
455 | SLAB_MEM_SPREAD|SLAB_PANIC), | |
20c2df83 | 456 | init_once); |
1da177e4 LT |
457 | err = register_filesystem(&bd_type); |
458 | if (err) | |
459 | panic("Cannot register bdev pseudo-fs"); | |
460 | bd_mnt = kern_mount(&bd_type); | |
1da177e4 LT |
461 | if (IS_ERR(bd_mnt)) |
462 | panic("Cannot create bdev pseudo-fs"); | |
ace8577a | 463 | blockdev_superblock = bd_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
464 | } |
465 | ||
466 | /* | |
467 | * Most likely _very_ bad one - but then it's hardly critical for small | |
468 | * /dev and can be fixed when somebody will need really large one. | |
469 | * Keep in mind that it will be fed through icache hash function too. | |
470 | */ | |
471 | static inline unsigned long hash(dev_t dev) | |
472 | { | |
473 | return MAJOR(dev)+MINOR(dev); | |
474 | } | |
475 | ||
476 | static int bdev_test(struct inode *inode, void *data) | |
477 | { | |
478 | return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data; | |
479 | } | |
480 | ||
481 | static int bdev_set(struct inode *inode, void *data) | |
482 | { | |
483 | BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data; | |
484 | return 0; | |
485 | } | |
486 | ||
487 | static LIST_HEAD(all_bdevs); | |
488 | ||
489 | struct block_device *bdget(dev_t dev) | |
490 | { | |
491 | struct block_device *bdev; | |
492 | struct inode *inode; | |
493 | ||
c2acf7b9 | 494 | inode = iget5_locked(blockdev_superblock, hash(dev), |
1da177e4 LT |
495 | bdev_test, bdev_set, &dev); |
496 | ||
497 | if (!inode) | |
498 | return NULL; | |
499 | ||
500 | bdev = &BDEV_I(inode)->bdev; | |
501 | ||
502 | if (inode->i_state & I_NEW) { | |
503 | bdev->bd_contains = NULL; | |
782b94cd | 504 | bdev->bd_super = NULL; |
1da177e4 LT |
505 | bdev->bd_inode = inode; |
506 | bdev->bd_block_size = (1 << inode->i_blkbits); | |
507 | bdev->bd_part_count = 0; | |
508 | bdev->bd_invalidated = 0; | |
509 | inode->i_mode = S_IFBLK; | |
510 | inode->i_rdev = dev; | |
511 | inode->i_bdev = bdev; | |
512 | inode->i_data.a_ops = &def_blk_aops; | |
513 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
514 | inode->i_data.backing_dev_info = &default_backing_dev_info; | |
515 | spin_lock(&bdev_lock); | |
516 | list_add(&bdev->bd_list, &all_bdevs); | |
517 | spin_unlock(&bdev_lock); | |
518 | unlock_new_inode(inode); | |
519 | } | |
520 | return bdev; | |
521 | } | |
522 | ||
523 | EXPORT_SYMBOL(bdget); | |
524 | ||
dddac6a7 AJ |
525 | /** |
526 | * bdgrab -- Grab a reference to an already referenced block device | |
527 | * @bdev: Block device to grab a reference to. | |
528 | */ | |
529 | struct block_device *bdgrab(struct block_device *bdev) | |
530 | { | |
7de9c6ee | 531 | ihold(bdev->bd_inode); |
dddac6a7 AJ |
532 | return bdev; |
533 | } | |
c1681bf8 | 534 | EXPORT_SYMBOL(bdgrab); |
dddac6a7 | 535 | |
1da177e4 LT |
536 | long nr_blockdev_pages(void) |
537 | { | |
203a2935 | 538 | struct block_device *bdev; |
1da177e4 LT |
539 | long ret = 0; |
540 | spin_lock(&bdev_lock); | |
203a2935 | 541 | list_for_each_entry(bdev, &all_bdevs, bd_list) { |
1da177e4 LT |
542 | ret += bdev->bd_inode->i_mapping->nrpages; |
543 | } | |
544 | spin_unlock(&bdev_lock); | |
545 | return ret; | |
546 | } | |
547 | ||
548 | void bdput(struct block_device *bdev) | |
549 | { | |
550 | iput(bdev->bd_inode); | |
551 | } | |
552 | ||
553 | EXPORT_SYMBOL(bdput); | |
554 | ||
555 | static struct block_device *bd_acquire(struct inode *inode) | |
556 | { | |
557 | struct block_device *bdev; | |
09d967c6 | 558 | |
1da177e4 LT |
559 | spin_lock(&bdev_lock); |
560 | bdev = inode->i_bdev; | |
09d967c6 | 561 | if (bdev) { |
7de9c6ee | 562 | ihold(bdev->bd_inode); |
1da177e4 LT |
563 | spin_unlock(&bdev_lock); |
564 | return bdev; | |
565 | } | |
566 | spin_unlock(&bdev_lock); | |
09d967c6 | 567 | |
1da177e4 LT |
568 | bdev = bdget(inode->i_rdev); |
569 | if (bdev) { | |
570 | spin_lock(&bdev_lock); | |
09d967c6 OH |
571 | if (!inode->i_bdev) { |
572 | /* | |
7de9c6ee | 573 | * We take an additional reference to bd_inode, |
09d967c6 OH |
574 | * and it's released in clear_inode() of inode. |
575 | * So, we can access it via ->i_mapping always | |
576 | * without igrab(). | |
577 | */ | |
7de9c6ee | 578 | ihold(bdev->bd_inode); |
09d967c6 OH |
579 | inode->i_bdev = bdev; |
580 | inode->i_mapping = bdev->bd_inode->i_mapping; | |
581 | list_add(&inode->i_devices, &bdev->bd_inodes); | |
582 | } | |
1da177e4 LT |
583 | spin_unlock(&bdev_lock); |
584 | } | |
585 | return bdev; | |
586 | } | |
587 | ||
f47ec3f2 AV |
588 | static inline int sb_is_blkdev_sb(struct super_block *sb) |
589 | { | |
590 | return sb == blockdev_superblock; | |
591 | } | |
592 | ||
1da177e4 LT |
593 | /* Call when you free inode */ |
594 | ||
595 | void bd_forget(struct inode *inode) | |
596 | { | |
09d967c6 OH |
597 | struct block_device *bdev = NULL; |
598 | ||
1da177e4 | 599 | spin_lock(&bdev_lock); |
b4ea2eaa YH |
600 | if (!sb_is_blkdev_sb(inode->i_sb)) |
601 | bdev = inode->i_bdev; | |
602 | __bd_forget(inode); | |
1da177e4 | 603 | spin_unlock(&bdev_lock); |
09d967c6 OH |
604 | |
605 | if (bdev) | |
606 | iput(bdev->bd_inode); | |
1da177e4 LT |
607 | } |
608 | ||
1a3cbbc5 TH |
609 | /** |
610 | * bd_may_claim - test whether a block device can be claimed | |
611 | * @bdev: block device of interest | |
612 | * @whole: whole block device containing @bdev, may equal @bdev | |
613 | * @holder: holder trying to claim @bdev | |
614 | * | |
25985edc | 615 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 TH |
616 | * |
617 | * CONTEXT: | |
618 | * spin_lock(&bdev_lock). | |
619 | * | |
620 | * RETURNS: | |
621 | * %true if @bdev can be claimed, %false otherwise. | |
622 | */ | |
623 | static bool bd_may_claim(struct block_device *bdev, struct block_device *whole, | |
624 | void *holder) | |
1da177e4 | 625 | { |
1da177e4 | 626 | if (bdev->bd_holder == holder) |
1a3cbbc5 | 627 | return true; /* already a holder */ |
1da177e4 | 628 | else if (bdev->bd_holder != NULL) |
1a3cbbc5 | 629 | return false; /* held by someone else */ |
1da177e4 | 630 | else if (bdev->bd_contains == bdev) |
1a3cbbc5 | 631 | return true; /* is a whole device which isn't held */ |
1da177e4 | 632 | |
e525fd89 | 633 | else if (whole->bd_holder == bd_may_claim) |
1a3cbbc5 TH |
634 | return true; /* is a partition of a device that is being partitioned */ |
635 | else if (whole->bd_holder != NULL) | |
636 | return false; /* is a partition of a held device */ | |
1da177e4 | 637 | else |
1a3cbbc5 TH |
638 | return true; /* is a partition of an un-held device */ |
639 | } | |
640 | ||
6b4517a7 TH |
641 | /** |
642 | * bd_prepare_to_claim - prepare to claim a block device | |
643 | * @bdev: block device of interest | |
644 | * @whole: the whole device containing @bdev, may equal @bdev | |
645 | * @holder: holder trying to claim @bdev | |
646 | * | |
647 | * Prepare to claim @bdev. This function fails if @bdev is already | |
648 | * claimed by another holder and waits if another claiming is in | |
649 | * progress. This function doesn't actually claim. On successful | |
650 | * return, the caller has ownership of bd_claiming and bd_holder[s]. | |
651 | * | |
652 | * CONTEXT: | |
653 | * spin_lock(&bdev_lock). Might release bdev_lock, sleep and regrab | |
654 | * it multiple times. | |
655 | * | |
656 | * RETURNS: | |
657 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
658 | */ | |
659 | static int bd_prepare_to_claim(struct block_device *bdev, | |
660 | struct block_device *whole, void *holder) | |
661 | { | |
662 | retry: | |
663 | /* if someone else claimed, fail */ | |
664 | if (!bd_may_claim(bdev, whole, holder)) | |
665 | return -EBUSY; | |
666 | ||
e75aa858 TH |
667 | /* if claiming is already in progress, wait for it to finish */ |
668 | if (whole->bd_claiming) { | |
6b4517a7 TH |
669 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
670 | DEFINE_WAIT(wait); | |
671 | ||
672 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
673 | spin_unlock(&bdev_lock); | |
674 | schedule(); | |
675 | finish_wait(wq, &wait); | |
676 | spin_lock(&bdev_lock); | |
677 | goto retry; | |
678 | } | |
679 | ||
680 | /* yay, all mine */ | |
681 | return 0; | |
682 | } | |
683 | ||
684 | /** | |
685 | * bd_start_claiming - start claiming a block device | |
686 | * @bdev: block device of interest | |
687 | * @holder: holder trying to claim @bdev | |
688 | * | |
689 | * @bdev is about to be opened exclusively. Check @bdev can be opened | |
690 | * exclusively and mark that an exclusive open is in progress. Each | |
691 | * successful call to this function must be matched with a call to | |
b0018361 NP |
692 | * either bd_finish_claiming() or bd_abort_claiming() (which do not |
693 | * fail). | |
694 | * | |
695 | * This function is used to gain exclusive access to the block device | |
696 | * without actually causing other exclusive open attempts to fail. It | |
697 | * should be used when the open sequence itself requires exclusive | |
698 | * access but may subsequently fail. | |
6b4517a7 TH |
699 | * |
700 | * CONTEXT: | |
701 | * Might sleep. | |
702 | * | |
703 | * RETURNS: | |
704 | * Pointer to the block device containing @bdev on success, ERR_PTR() | |
705 | * value on failure. | |
706 | */ | |
707 | static struct block_device *bd_start_claiming(struct block_device *bdev, | |
708 | void *holder) | |
709 | { | |
710 | struct gendisk *disk; | |
711 | struct block_device *whole; | |
712 | int partno, err; | |
713 | ||
714 | might_sleep(); | |
715 | ||
716 | /* | |
717 | * @bdev might not have been initialized properly yet, look up | |
718 | * and grab the outer block device the hard way. | |
719 | */ | |
720 | disk = get_gendisk(bdev->bd_dev, &partno); | |
721 | if (!disk) | |
722 | return ERR_PTR(-ENXIO); | |
723 | ||
d4c208b8 TH |
724 | /* |
725 | * Normally, @bdev should equal what's returned from bdget_disk() | |
726 | * if partno is 0; however, some drivers (floppy) use multiple | |
727 | * bdev's for the same physical device and @bdev may be one of the | |
728 | * aliases. Keep @bdev if partno is 0. This means claimer | |
729 | * tracking is broken for those devices but it has always been that | |
730 | * way. | |
731 | */ | |
732 | if (partno) | |
733 | whole = bdget_disk(disk, 0); | |
734 | else | |
735 | whole = bdgrab(bdev); | |
736 | ||
cf342570 | 737 | module_put(disk->fops->owner); |
6b4517a7 TH |
738 | put_disk(disk); |
739 | if (!whole) | |
740 | return ERR_PTR(-ENOMEM); | |
741 | ||
742 | /* prepare to claim, if successful, mark claiming in progress */ | |
743 | spin_lock(&bdev_lock); | |
744 | ||
745 | err = bd_prepare_to_claim(bdev, whole, holder); | |
746 | if (err == 0) { | |
747 | whole->bd_claiming = holder; | |
748 | spin_unlock(&bdev_lock); | |
749 | return whole; | |
750 | } else { | |
751 | spin_unlock(&bdev_lock); | |
752 | bdput(whole); | |
753 | return ERR_PTR(err); | |
754 | } | |
755 | } | |
756 | ||
641dc636 | 757 | #ifdef CONFIG_SYSFS |
49731baa TH |
758 | struct bd_holder_disk { |
759 | struct list_head list; | |
760 | struct gendisk *disk; | |
761 | int refcnt; | |
762 | }; | |
763 | ||
764 | static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev, | |
765 | struct gendisk *disk) | |
766 | { | |
767 | struct bd_holder_disk *holder; | |
768 | ||
769 | list_for_each_entry(holder, &bdev->bd_holder_disks, list) | |
770 | if (holder->disk == disk) | |
771 | return holder; | |
772 | return NULL; | |
773 | } | |
774 | ||
4d7dd8fd | 775 | static int add_symlink(struct kobject *from, struct kobject *to) |
641dc636 | 776 | { |
4d7dd8fd | 777 | return sysfs_create_link(from, to, kobject_name(to)); |
641dc636 JN |
778 | } |
779 | ||
780 | static void del_symlink(struct kobject *from, struct kobject *to) | |
781 | { | |
641dc636 JN |
782 | sysfs_remove_link(from, kobject_name(to)); |
783 | } | |
784 | ||
df6c0cd9 | 785 | /** |
e09b457b TH |
786 | * bd_link_disk_holder - create symlinks between holding disk and slave bdev |
787 | * @bdev: the claimed slave bdev | |
788 | * @disk: the holding disk | |
df6c0cd9 | 789 | * |
49731baa TH |
790 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. |
791 | * | |
e09b457b | 792 | * This functions creates the following sysfs symlinks. |
641dc636 | 793 | * |
e09b457b TH |
794 | * - from "slaves" directory of the holder @disk to the claimed @bdev |
795 | * - from "holders" directory of the @bdev to the holder @disk | |
641dc636 | 796 | * |
e09b457b TH |
797 | * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is |
798 | * passed to bd_link_disk_holder(), then: | |
641dc636 | 799 | * |
e09b457b TH |
800 | * /sys/block/dm-0/slaves/sda --> /sys/block/sda |
801 | * /sys/block/sda/holders/dm-0 --> /sys/block/dm-0 | |
641dc636 | 802 | * |
e09b457b TH |
803 | * The caller must have claimed @bdev before calling this function and |
804 | * ensure that both @bdev and @disk are valid during the creation and | |
805 | * lifetime of these symlinks. | |
641dc636 | 806 | * |
e09b457b TH |
807 | * CONTEXT: |
808 | * Might sleep. | |
641dc636 | 809 | * |
e09b457b TH |
810 | * RETURNS: |
811 | * 0 on success, -errno on failure. | |
641dc636 | 812 | */ |
e09b457b | 813 | int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk) |
641dc636 | 814 | { |
49731baa | 815 | struct bd_holder_disk *holder; |
e09b457b | 816 | int ret = 0; |
641dc636 | 817 | |
2e7b651d | 818 | mutex_lock(&bdev->bd_mutex); |
df6c0cd9 | 819 | |
49731baa | 820 | WARN_ON_ONCE(!bdev->bd_holder); |
4e91672c | 821 | |
e09b457b TH |
822 | /* FIXME: remove the following once add_disk() handles errors */ |
823 | if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir)) | |
824 | goto out_unlock; | |
4e91672c | 825 | |
49731baa TH |
826 | holder = bd_find_holder_disk(bdev, disk); |
827 | if (holder) { | |
828 | holder->refcnt++; | |
e09b457b | 829 | goto out_unlock; |
49731baa | 830 | } |
641dc636 | 831 | |
49731baa TH |
832 | holder = kzalloc(sizeof(*holder), GFP_KERNEL); |
833 | if (!holder) { | |
834 | ret = -ENOMEM; | |
e09b457b TH |
835 | goto out_unlock; |
836 | } | |
641dc636 | 837 | |
49731baa TH |
838 | INIT_LIST_HEAD(&holder->list); |
839 | holder->disk = disk; | |
840 | holder->refcnt = 1; | |
841 | ||
842 | ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
843 | if (ret) | |
844 | goto out_free; | |
845 | ||
846 | ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj); | |
847 | if (ret) | |
848 | goto out_del; | |
e7407d16 TH |
849 | /* |
850 | * bdev could be deleted beneath us which would implicitly destroy | |
851 | * the holder directory. Hold on to it. | |
852 | */ | |
853 | kobject_get(bdev->bd_part->holder_dir); | |
49731baa TH |
854 | |
855 | list_add(&holder->list, &bdev->bd_holder_disks); | |
856 | goto out_unlock; | |
857 | ||
858 | out_del: | |
859 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
860 | out_free: | |
861 | kfree(holder); | |
e09b457b | 862 | out_unlock: |
b4cf1b72 | 863 | mutex_unlock(&bdev->bd_mutex); |
e09b457b | 864 | return ret; |
641dc636 | 865 | } |
e09b457b | 866 | EXPORT_SYMBOL_GPL(bd_link_disk_holder); |
641dc636 | 867 | |
49731baa TH |
868 | /** |
869 | * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder() | |
870 | * @bdev: the calimed slave bdev | |
871 | * @disk: the holding disk | |
872 | * | |
873 | * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT. | |
874 | * | |
875 | * CONTEXT: | |
876 | * Might sleep. | |
877 | */ | |
878 | void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk) | |
641dc636 | 879 | { |
49731baa | 880 | struct bd_holder_disk *holder; |
641dc636 | 881 | |
49731baa | 882 | mutex_lock(&bdev->bd_mutex); |
641dc636 | 883 | |
49731baa TH |
884 | holder = bd_find_holder_disk(bdev, disk); |
885 | ||
886 | if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) { | |
887 | del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj); | |
888 | del_symlink(bdev->bd_part->holder_dir, | |
889 | &disk_to_dev(disk)->kobj); | |
e7407d16 | 890 | kobject_put(bdev->bd_part->holder_dir); |
49731baa TH |
891 | list_del_init(&holder->list); |
892 | kfree(holder); | |
893 | } | |
894 | ||
895 | mutex_unlock(&bdev->bd_mutex); | |
1da177e4 | 896 | } |
49731baa | 897 | EXPORT_SYMBOL_GPL(bd_unlink_disk_holder); |
641dc636 | 898 | #endif |
1da177e4 | 899 | |
56ade44b AP |
900 | /** |
901 | * flush_disk - invalidates all buffer-cache entries on a disk | |
902 | * | |
903 | * @bdev: struct block device to be flushed | |
e6eb5ce1 | 904 | * @kill_dirty: flag to guide handling of dirty inodes |
56ade44b AP |
905 | * |
906 | * Invalidates all buffer-cache entries on a disk. It should be called | |
907 | * when a disk has been changed -- either by a media change or online | |
908 | * resize. | |
909 | */ | |
93b270f7 | 910 | static void flush_disk(struct block_device *bdev, bool kill_dirty) |
56ade44b | 911 | { |
93b270f7 | 912 | if (__invalidate_device(bdev, kill_dirty)) { |
56ade44b AP |
913 | char name[BDEVNAME_SIZE] = ""; |
914 | ||
915 | if (bdev->bd_disk) | |
916 | disk_name(bdev->bd_disk, 0, name); | |
917 | printk(KERN_WARNING "VFS: busy inodes on changed media or " | |
918 | "resized disk %s\n", name); | |
919 | } | |
920 | ||
921 | if (!bdev->bd_disk) | |
922 | return; | |
d27769ec | 923 | if (disk_part_scan_enabled(bdev->bd_disk)) |
56ade44b AP |
924 | bdev->bd_invalidated = 1; |
925 | } | |
926 | ||
c3279d14 | 927 | /** |
57d1b536 | 928 | * check_disk_size_change - checks for disk size change and adjusts bdev size. |
c3279d14 AP |
929 | * @disk: struct gendisk to check |
930 | * @bdev: struct bdev to adjust. | |
931 | * | |
932 | * This routine checks to see if the bdev size does not match the disk size | |
933 | * and adjusts it if it differs. | |
934 | */ | |
935 | void check_disk_size_change(struct gendisk *disk, struct block_device *bdev) | |
936 | { | |
937 | loff_t disk_size, bdev_size; | |
938 | ||
939 | disk_size = (loff_t)get_capacity(disk) << 9; | |
940 | bdev_size = i_size_read(bdev->bd_inode); | |
941 | if (disk_size != bdev_size) { | |
942 | char name[BDEVNAME_SIZE]; | |
943 | ||
944 | disk_name(disk, 0, name); | |
945 | printk(KERN_INFO | |
946 | "%s: detected capacity change from %lld to %lld\n", | |
947 | name, bdev_size, disk_size); | |
948 | i_size_write(bdev->bd_inode, disk_size); | |
93b270f7 | 949 | flush_disk(bdev, false); |
c3279d14 AP |
950 | } |
951 | } | |
952 | EXPORT_SYMBOL(check_disk_size_change); | |
953 | ||
0c002c2f | 954 | /** |
57d1b536 | 955 | * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back |
0c002c2f AP |
956 | * @disk: struct gendisk to be revalidated |
957 | * | |
958 | * This routine is a wrapper for lower-level driver's revalidate_disk | |
959 | * call-backs. It is used to do common pre and post operations needed | |
960 | * for all revalidate_disk operations. | |
961 | */ | |
962 | int revalidate_disk(struct gendisk *disk) | |
963 | { | |
c3279d14 | 964 | struct block_device *bdev; |
0c002c2f AP |
965 | int ret = 0; |
966 | ||
967 | if (disk->fops->revalidate_disk) | |
968 | ret = disk->fops->revalidate_disk(disk); | |
969 | ||
c3279d14 AP |
970 | bdev = bdget_disk(disk, 0); |
971 | if (!bdev) | |
972 | return ret; | |
973 | ||
974 | mutex_lock(&bdev->bd_mutex); | |
975 | check_disk_size_change(disk, bdev); | |
7630b661 | 976 | bdev->bd_invalidated = 0; |
c3279d14 AP |
977 | mutex_unlock(&bdev->bd_mutex); |
978 | bdput(bdev); | |
0c002c2f AP |
979 | return ret; |
980 | } | |
981 | EXPORT_SYMBOL(revalidate_disk); | |
982 | ||
1da177e4 LT |
983 | /* |
984 | * This routine checks whether a removable media has been changed, | |
985 | * and invalidates all buffer-cache-entries in that case. This | |
986 | * is a relatively slow routine, so we have to try to minimize using | |
987 | * it. Thus it is called only upon a 'mount' or 'open'. This | |
988 | * is the best way of combining speed and utility, I think. | |
989 | * People changing diskettes in the middle of an operation deserve | |
990 | * to lose :-) | |
991 | */ | |
992 | int check_disk_change(struct block_device *bdev) | |
993 | { | |
994 | struct gendisk *disk = bdev->bd_disk; | |
83d5cde4 | 995 | const struct block_device_operations *bdops = disk->fops; |
77ea887e | 996 | unsigned int events; |
1da177e4 | 997 | |
77ea887e TH |
998 | events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE | |
999 | DISK_EVENT_EJECT_REQUEST); | |
1000 | if (!(events & DISK_EVENT_MEDIA_CHANGE)) | |
1da177e4 LT |
1001 | return 0; |
1002 | ||
93b270f7 | 1003 | flush_disk(bdev, true); |
1da177e4 LT |
1004 | if (bdops->revalidate_disk) |
1005 | bdops->revalidate_disk(bdev->bd_disk); | |
1da177e4 LT |
1006 | return 1; |
1007 | } | |
1008 | ||
1009 | EXPORT_SYMBOL(check_disk_change); | |
1010 | ||
1011 | void bd_set_size(struct block_device *bdev, loff_t size) | |
1012 | { | |
e1defc4f | 1013 | unsigned bsize = bdev_logical_block_size(bdev); |
1da177e4 | 1014 | |
d646a02a GC |
1015 | mutex_lock(&bdev->bd_inode->i_mutex); |
1016 | i_size_write(bdev->bd_inode, size); | |
1017 | mutex_unlock(&bdev->bd_inode->i_mutex); | |
1da177e4 LT |
1018 | while (bsize < PAGE_CACHE_SIZE) { |
1019 | if (size & bsize) | |
1020 | break; | |
1021 | bsize <<= 1; | |
1022 | } | |
1023 | bdev->bd_block_size = bsize; | |
1024 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
1025 | } | |
1026 | EXPORT_SYMBOL(bd_set_size); | |
1027 | ||
4385bab1 | 1028 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part); |
37be4124 | 1029 | |
6d740cd5 PZ |
1030 | /* |
1031 | * bd_mutex locking: | |
1032 | * | |
1033 | * mutex_lock(part->bd_mutex) | |
1034 | * mutex_lock_nested(whole->bd_mutex, 1) | |
1035 | */ | |
1036 | ||
572c4892 | 1037 | static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part) |
1da177e4 | 1038 | { |
1da177e4 | 1039 | struct gendisk *disk; |
523e1d39 | 1040 | struct module *owner; |
7db9cfd3 | 1041 | int ret; |
cf771cb5 | 1042 | int partno; |
fe6e9c1f AV |
1043 | int perm = 0; |
1044 | ||
572c4892 | 1045 | if (mode & FMODE_READ) |
fe6e9c1f | 1046 | perm |= MAY_READ; |
572c4892 | 1047 | if (mode & FMODE_WRITE) |
fe6e9c1f AV |
1048 | perm |= MAY_WRITE; |
1049 | /* | |
1050 | * hooks: /n/, see "layering violations". | |
1051 | */ | |
b7300b78 CW |
1052 | if (!for_part) { |
1053 | ret = devcgroup_inode_permission(bdev->bd_inode, perm); | |
1054 | if (ret != 0) { | |
1055 | bdput(bdev); | |
1056 | return ret; | |
1057 | } | |
82666020 | 1058 | } |
7db9cfd3 | 1059 | |
d3374825 | 1060 | restart: |
0762b8bd | 1061 | |
89f97496 | 1062 | ret = -ENXIO; |
cf771cb5 | 1063 | disk = get_gendisk(bdev->bd_dev, &partno); |
0762b8bd | 1064 | if (!disk) |
6e9624b8 | 1065 | goto out; |
523e1d39 | 1066 | owner = disk->fops->owner; |
1da177e4 | 1067 | |
69e02c59 | 1068 | disk_block_events(disk); |
6796bf54 | 1069 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
1da177e4 LT |
1070 | if (!bdev->bd_openers) { |
1071 | bdev->bd_disk = disk; | |
87192a2a | 1072 | bdev->bd_queue = disk->queue; |
1da177e4 | 1073 | bdev->bd_contains = bdev; |
cf771cb5 | 1074 | if (!partno) { |
1da177e4 | 1075 | struct backing_dev_info *bdi; |
89f97496 TH |
1076 | |
1077 | ret = -ENXIO; | |
1078 | bdev->bd_part = disk_get_part(disk, partno); | |
1079 | if (!bdev->bd_part) | |
1080 | goto out_clear; | |
1081 | ||
1196f8b8 | 1082 | ret = 0; |
1da177e4 | 1083 | if (disk->fops->open) { |
572c4892 | 1084 | ret = disk->fops->open(bdev, mode); |
d3374825 N |
1085 | if (ret == -ERESTARTSYS) { |
1086 | /* Lost a race with 'disk' being | |
1087 | * deleted, try again. | |
1088 | * See md.c | |
1089 | */ | |
1090 | disk_put_part(bdev->bd_part); | |
1091 | bdev->bd_part = NULL; | |
d3374825 | 1092 | bdev->bd_disk = NULL; |
87192a2a | 1093 | bdev->bd_queue = NULL; |
d3374825 | 1094 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1095 | disk_unblock_events(disk); |
69e02c59 | 1096 | put_disk(disk); |
523e1d39 | 1097 | module_put(owner); |
d3374825 N |
1098 | goto restart; |
1099 | } | |
1da177e4 | 1100 | } |
7e69723f | 1101 | |
de33127d | 1102 | if (!ret) { |
7e69723f TH |
1103 | bd_set_size(bdev,(loff_t)get_capacity(disk)<<9); |
1104 | bdi = blk_get_backing_dev_info(bdev); | |
1105 | if (bdi == NULL) | |
1106 | bdi = &default_backing_dev_info; | |
1107 | bdev_inode_switch_bdi(bdev->bd_inode, bdi); | |
1108 | } | |
1109 | ||
1196f8b8 TH |
1110 | /* |
1111 | * If the device is invalidated, rescan partition | |
1112 | * if open succeeded or failed with -ENOMEDIUM. | |
1113 | * The latter is necessary to prevent ghost | |
1114 | * partitions on a removed medium. | |
1115 | */ | |
fe316bf2 JN |
1116 | if (bdev->bd_invalidated) { |
1117 | if (!ret) | |
1118 | rescan_partitions(disk, bdev); | |
1119 | else if (ret == -ENOMEDIUM) | |
1120 | invalidate_partitions(disk, bdev); | |
1121 | } | |
1196f8b8 TH |
1122 | if (ret) |
1123 | goto out_clear; | |
1da177e4 | 1124 | } else { |
1da177e4 LT |
1125 | struct block_device *whole; |
1126 | whole = bdget_disk(disk, 0); | |
1127 | ret = -ENOMEM; | |
1128 | if (!whole) | |
0762b8bd | 1129 | goto out_clear; |
37be4124 | 1130 | BUG_ON(for_part); |
572c4892 | 1131 | ret = __blkdev_get(whole, mode, 1); |
1da177e4 | 1132 | if (ret) |
0762b8bd | 1133 | goto out_clear; |
1da177e4 | 1134 | bdev->bd_contains = whole; |
a5491e0c DC |
1135 | bdev_inode_switch_bdi(bdev->bd_inode, |
1136 | whole->bd_inode->i_data.backing_dev_info); | |
89f97496 | 1137 | bdev->bd_part = disk_get_part(disk, partno); |
e71bf0d0 | 1138 | if (!(disk->flags & GENHD_FL_UP) || |
89f97496 | 1139 | !bdev->bd_part || !bdev->bd_part->nr_sects) { |
1da177e4 | 1140 | ret = -ENXIO; |
0762b8bd | 1141 | goto out_clear; |
1da177e4 | 1142 | } |
89f97496 | 1143 | bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9); |
1da177e4 LT |
1144 | } |
1145 | } else { | |
1da177e4 | 1146 | if (bdev->bd_contains == bdev) { |
1196f8b8 TH |
1147 | ret = 0; |
1148 | if (bdev->bd_disk->fops->open) | |
572c4892 | 1149 | ret = bdev->bd_disk->fops->open(bdev, mode); |
1196f8b8 | 1150 | /* the same as first opener case, read comment there */ |
fe316bf2 JN |
1151 | if (bdev->bd_invalidated) { |
1152 | if (!ret) | |
1153 | rescan_partitions(bdev->bd_disk, bdev); | |
1154 | else if (ret == -ENOMEDIUM) | |
1155 | invalidate_partitions(bdev->bd_disk, bdev); | |
1156 | } | |
1196f8b8 TH |
1157 | if (ret) |
1158 | goto out_unlock_bdev; | |
1da177e4 | 1159 | } |
69e02c59 | 1160 | /* only one opener holds refs to the module and disk */ |
69e02c59 | 1161 | put_disk(disk); |
523e1d39 | 1162 | module_put(owner); |
1da177e4 LT |
1163 | } |
1164 | bdev->bd_openers++; | |
37be4124 N |
1165 | if (for_part) |
1166 | bdev->bd_part_count++; | |
c039e313 | 1167 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1168 | disk_unblock_events(disk); |
1da177e4 LT |
1169 | return 0; |
1170 | ||
0762b8bd | 1171 | out_clear: |
89f97496 | 1172 | disk_put_part(bdev->bd_part); |
1da177e4 | 1173 | bdev->bd_disk = NULL; |
0762b8bd | 1174 | bdev->bd_part = NULL; |
87192a2a | 1175 | bdev->bd_queue = NULL; |
a5491e0c | 1176 | bdev_inode_switch_bdi(bdev->bd_inode, &default_backing_dev_info); |
1da177e4 | 1177 | if (bdev != bdev->bd_contains) |
572c4892 | 1178 | __blkdev_put(bdev->bd_contains, mode, 1); |
1da177e4 | 1179 | bdev->bd_contains = NULL; |
0762b8bd | 1180 | out_unlock_bdev: |
c039e313 | 1181 | mutex_unlock(&bdev->bd_mutex); |
69e02c59 | 1182 | disk_unblock_events(disk); |
0762b8bd | 1183 | put_disk(disk); |
523e1d39 | 1184 | module_put(owner); |
4345caba | 1185 | out: |
0762b8bd TH |
1186 | bdput(bdev); |
1187 | ||
1da177e4 LT |
1188 | return ret; |
1189 | } | |
1190 | ||
d4d77629 TH |
1191 | /** |
1192 | * blkdev_get - open a block device | |
1193 | * @bdev: block_device to open | |
1194 | * @mode: FMODE_* mask | |
1195 | * @holder: exclusive holder identifier | |
1196 | * | |
1197 | * Open @bdev with @mode. If @mode includes %FMODE_EXCL, @bdev is | |
1198 | * open with exclusive access. Specifying %FMODE_EXCL with %NULL | |
1199 | * @holder is invalid. Exclusive opens may nest for the same @holder. | |
1200 | * | |
1201 | * On success, the reference count of @bdev is unchanged. On failure, | |
1202 | * @bdev is put. | |
1203 | * | |
1204 | * CONTEXT: | |
1205 | * Might sleep. | |
1206 | * | |
1207 | * RETURNS: | |
1208 | * 0 on success, -errno on failure. | |
1209 | */ | |
e525fd89 | 1210 | int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder) |
1da177e4 | 1211 | { |
e525fd89 TH |
1212 | struct block_device *whole = NULL; |
1213 | int res; | |
1214 | ||
1215 | WARN_ON_ONCE((mode & FMODE_EXCL) && !holder); | |
1216 | ||
1217 | if ((mode & FMODE_EXCL) && holder) { | |
1218 | whole = bd_start_claiming(bdev, holder); | |
1219 | if (IS_ERR(whole)) { | |
1220 | bdput(bdev); | |
1221 | return PTR_ERR(whole); | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | res = __blkdev_get(bdev, mode, 0); | |
1226 | ||
1227 | if (whole) { | |
d4dc210f TH |
1228 | struct gendisk *disk = whole->bd_disk; |
1229 | ||
6a027eff | 1230 | /* finish claiming */ |
77ea887e | 1231 | mutex_lock(&bdev->bd_mutex); |
6a027eff TH |
1232 | spin_lock(&bdev_lock); |
1233 | ||
77ea887e | 1234 | if (!res) { |
6a027eff TH |
1235 | BUG_ON(!bd_may_claim(bdev, whole, holder)); |
1236 | /* | |
1237 | * Note that for a whole device bd_holders | |
1238 | * will be incremented twice, and bd_holder | |
1239 | * will be set to bd_may_claim before being | |
1240 | * set to holder | |
1241 | */ | |
1242 | whole->bd_holders++; | |
1243 | whole->bd_holder = bd_may_claim; | |
1244 | bdev->bd_holders++; | |
1245 | bdev->bd_holder = holder; | |
1246 | } | |
1247 | ||
1248 | /* tell others that we're done */ | |
1249 | BUG_ON(whole->bd_claiming != holder); | |
1250 | whole->bd_claiming = NULL; | |
1251 | wake_up_bit(&whole->bd_claiming, 0); | |
1252 | ||
1253 | spin_unlock(&bdev_lock); | |
77ea887e TH |
1254 | |
1255 | /* | |
d4dc210f TH |
1256 | * Block event polling for write claims if requested. Any |
1257 | * write holder makes the write_holder state stick until | |
1258 | * all are released. This is good enough and tracking | |
1259 | * individual writeable reference is too fragile given the | |
1260 | * way @mode is used in blkdev_get/put(). | |
77ea887e | 1261 | */ |
4c49ff3f TH |
1262 | if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder && |
1263 | (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) { | |
77ea887e | 1264 | bdev->bd_write_holder = true; |
d4dc210f | 1265 | disk_block_events(disk); |
77ea887e TH |
1266 | } |
1267 | ||
1268 | mutex_unlock(&bdev->bd_mutex); | |
6a027eff | 1269 | bdput(whole); |
e525fd89 TH |
1270 | } |
1271 | ||
1272 | return res; | |
37be4124 | 1273 | } |
1da177e4 LT |
1274 | EXPORT_SYMBOL(blkdev_get); |
1275 | ||
d4d77629 TH |
1276 | /** |
1277 | * blkdev_get_by_path - open a block device by name | |
1278 | * @path: path to the block device to open | |
1279 | * @mode: FMODE_* mask | |
1280 | * @holder: exclusive holder identifier | |
1281 | * | |
1282 | * Open the blockdevice described by the device file at @path. @mode | |
1283 | * and @holder are identical to blkdev_get(). | |
1284 | * | |
1285 | * On success, the returned block_device has reference count of one. | |
1286 | * | |
1287 | * CONTEXT: | |
1288 | * Might sleep. | |
1289 | * | |
1290 | * RETURNS: | |
1291 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1292 | */ | |
1293 | struct block_device *blkdev_get_by_path(const char *path, fmode_t mode, | |
1294 | void *holder) | |
1295 | { | |
1296 | struct block_device *bdev; | |
1297 | int err; | |
1298 | ||
1299 | bdev = lookup_bdev(path); | |
1300 | if (IS_ERR(bdev)) | |
1301 | return bdev; | |
1302 | ||
1303 | err = blkdev_get(bdev, mode, holder); | |
1304 | if (err) | |
1305 | return ERR_PTR(err); | |
1306 | ||
e51900f7 CE |
1307 | if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) { |
1308 | blkdev_put(bdev, mode); | |
1309 | return ERR_PTR(-EACCES); | |
1310 | } | |
1311 | ||
d4d77629 TH |
1312 | return bdev; |
1313 | } | |
1314 | EXPORT_SYMBOL(blkdev_get_by_path); | |
1315 | ||
1316 | /** | |
1317 | * blkdev_get_by_dev - open a block device by device number | |
1318 | * @dev: device number of block device to open | |
1319 | * @mode: FMODE_* mask | |
1320 | * @holder: exclusive holder identifier | |
1321 | * | |
1322 | * Open the blockdevice described by device number @dev. @mode and | |
1323 | * @holder are identical to blkdev_get(). | |
1324 | * | |
1325 | * Use it ONLY if you really do not have anything better - i.e. when | |
1326 | * you are behind a truly sucky interface and all you are given is a | |
1327 | * device number. _Never_ to be used for internal purposes. If you | |
1328 | * ever need it - reconsider your API. | |
1329 | * | |
1330 | * On success, the returned block_device has reference count of one. | |
1331 | * | |
1332 | * CONTEXT: | |
1333 | * Might sleep. | |
1334 | * | |
1335 | * RETURNS: | |
1336 | * Pointer to block_device on success, ERR_PTR(-errno) on failure. | |
1337 | */ | |
1338 | struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder) | |
1339 | { | |
1340 | struct block_device *bdev; | |
1341 | int err; | |
1342 | ||
1343 | bdev = bdget(dev); | |
1344 | if (!bdev) | |
1345 | return ERR_PTR(-ENOMEM); | |
1346 | ||
1347 | err = blkdev_get(bdev, mode, holder); | |
1348 | if (err) | |
1349 | return ERR_PTR(err); | |
1350 | ||
1351 | return bdev; | |
1352 | } | |
1353 | EXPORT_SYMBOL(blkdev_get_by_dev); | |
1354 | ||
1da177e4 LT |
1355 | static int blkdev_open(struct inode * inode, struct file * filp) |
1356 | { | |
1357 | struct block_device *bdev; | |
1da177e4 LT |
1358 | |
1359 | /* | |
1360 | * Preserve backwards compatibility and allow large file access | |
1361 | * even if userspace doesn't ask for it explicitly. Some mkfs | |
1362 | * binary needs it. We might want to drop this workaround | |
1363 | * during an unstable branch. | |
1364 | */ | |
1365 | filp->f_flags |= O_LARGEFILE; | |
1366 | ||
572c4892 AV |
1367 | if (filp->f_flags & O_NDELAY) |
1368 | filp->f_mode |= FMODE_NDELAY; | |
1369 | if (filp->f_flags & O_EXCL) | |
1370 | filp->f_mode |= FMODE_EXCL; | |
1371 | if ((filp->f_flags & O_ACCMODE) == 3) | |
1372 | filp->f_mode |= FMODE_WRITE_IOCTL; | |
1373 | ||
1da177e4 | 1374 | bdev = bd_acquire(inode); |
6a2aae06 PE |
1375 | if (bdev == NULL) |
1376 | return -ENOMEM; | |
1da177e4 | 1377 | |
572c4892 AV |
1378 | filp->f_mapping = bdev->bd_inode->i_mapping; |
1379 | ||
e525fd89 | 1380 | return blkdev_get(bdev, filp->f_mode, filp); |
1da177e4 LT |
1381 | } |
1382 | ||
4385bab1 | 1383 | static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part) |
2e7b651d | 1384 | { |
2e7b651d | 1385 | struct gendisk *disk = bdev->bd_disk; |
37be4124 | 1386 | struct block_device *victim = NULL; |
2e7b651d | 1387 | |
6796bf54 | 1388 | mutex_lock_nested(&bdev->bd_mutex, for_part); |
37be4124 N |
1389 | if (for_part) |
1390 | bdev->bd_part_count--; | |
1391 | ||
2e7b651d | 1392 | if (!--bdev->bd_openers) { |
6a027eff | 1393 | WARN_ON_ONCE(bdev->bd_holders); |
2e7b651d PZ |
1394 | sync_blockdev(bdev); |
1395 | kill_bdev(bdev); | |
94007751 N |
1396 | /* ->release can cause the old bdi to disappear, |
1397 | * so must switch it out first | |
1398 | */ | |
1399 | bdev_inode_switch_bdi(bdev->bd_inode, | |
1400 | &default_backing_dev_info); | |
2e7b651d PZ |
1401 | } |
1402 | if (bdev->bd_contains == bdev) { | |
1403 | if (disk->fops->release) | |
db2a144b | 1404 | disk->fops->release(disk, mode); |
2e7b651d PZ |
1405 | } |
1406 | if (!bdev->bd_openers) { | |
1407 | struct module *owner = disk->fops->owner; | |
1408 | ||
0762b8bd TH |
1409 | disk_put_part(bdev->bd_part); |
1410 | bdev->bd_part = NULL; | |
2e7b651d | 1411 | bdev->bd_disk = NULL; |
37be4124 N |
1412 | if (bdev != bdev->bd_contains) |
1413 | victim = bdev->bd_contains; | |
2e7b651d | 1414 | bdev->bd_contains = NULL; |
523e1d39 TH |
1415 | |
1416 | put_disk(disk); | |
1417 | module_put(owner); | |
2e7b651d | 1418 | } |
2e7b651d PZ |
1419 | mutex_unlock(&bdev->bd_mutex); |
1420 | bdput(bdev); | |
37be4124 | 1421 | if (victim) |
9a1c3542 | 1422 | __blkdev_put(victim, mode, 1); |
2e7b651d PZ |
1423 | } |
1424 | ||
4385bab1 | 1425 | void blkdev_put(struct block_device *bdev, fmode_t mode) |
37be4124 | 1426 | { |
85ef06d1 TH |
1427 | mutex_lock(&bdev->bd_mutex); |
1428 | ||
e525fd89 | 1429 | if (mode & FMODE_EXCL) { |
6a027eff TH |
1430 | bool bdev_free; |
1431 | ||
1432 | /* | |
1433 | * Release a claim on the device. The holder fields | |
1434 | * are protected with bdev_lock. bd_mutex is to | |
1435 | * synchronize disk_holder unlinking. | |
1436 | */ | |
6a027eff TH |
1437 | spin_lock(&bdev_lock); |
1438 | ||
1439 | WARN_ON_ONCE(--bdev->bd_holders < 0); | |
1440 | WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0); | |
1441 | ||
1442 | /* bd_contains might point to self, check in a separate step */ | |
1443 | if ((bdev_free = !bdev->bd_holders)) | |
1444 | bdev->bd_holder = NULL; | |
1445 | if (!bdev->bd_contains->bd_holders) | |
1446 | bdev->bd_contains->bd_holder = NULL; | |
1447 | ||
1448 | spin_unlock(&bdev_lock); | |
1449 | ||
77ea887e TH |
1450 | /* |
1451 | * If this was the last claim, remove holder link and | |
1452 | * unblock evpoll if it was a write holder. | |
1453 | */ | |
85ef06d1 TH |
1454 | if (bdev_free && bdev->bd_write_holder) { |
1455 | disk_unblock_events(bdev->bd_disk); | |
1456 | bdev->bd_write_holder = false; | |
77ea887e | 1457 | } |
6936217c | 1458 | } |
77ea887e | 1459 | |
85ef06d1 TH |
1460 | /* |
1461 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1462 | * event. This is to ensure detection of media removal commanded | |
1463 | * from userland - e.g. eject(1). | |
1464 | */ | |
1465 | disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE); | |
1466 | ||
1467 | mutex_unlock(&bdev->bd_mutex); | |
1468 | ||
4385bab1 | 1469 | __blkdev_put(bdev, mode, 0); |
37be4124 | 1470 | } |
2e7b651d PZ |
1471 | EXPORT_SYMBOL(blkdev_put); |
1472 | ||
1da177e4 LT |
1473 | static int blkdev_close(struct inode * inode, struct file * filp) |
1474 | { | |
1475 | struct block_device *bdev = I_BDEV(filp->f_mapping->host); | |
4385bab1 AV |
1476 | blkdev_put(bdev, filp->f_mode); |
1477 | return 0; | |
1da177e4 LT |
1478 | } |
1479 | ||
bb93e3a5 | 1480 | static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg) |
1da177e4 | 1481 | { |
56b26add AV |
1482 | struct block_device *bdev = I_BDEV(file->f_mapping->host); |
1483 | fmode_t mode = file->f_mode; | |
fd4ce1ac CH |
1484 | |
1485 | /* | |
1486 | * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have | |
1487 | * to updated it before every ioctl. | |
1488 | */ | |
56b26add | 1489 | if (file->f_flags & O_NDELAY) |
fd4ce1ac CH |
1490 | mode |= FMODE_NDELAY; |
1491 | else | |
1492 | mode &= ~FMODE_NDELAY; | |
1493 | ||
56b26add | 1494 | return blkdev_ioctl(bdev, mode, cmd, arg); |
1da177e4 LT |
1495 | } |
1496 | ||
eef99380 CH |
1497 | /* |
1498 | * Write data to the block device. Only intended for the block device itself | |
1499 | * and the raw driver which basically is a fake block device. | |
1500 | * | |
1501 | * Does not take i_mutex for the write and thus is not for general purpose | |
1502 | * use. | |
1503 | */ | |
1504 | ssize_t blkdev_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1505 | unsigned long nr_segs, loff_t pos) | |
1506 | { | |
1507 | struct file *file = iocb->ki_filp; | |
53362a05 | 1508 | struct blk_plug plug; |
eef99380 CH |
1509 | ssize_t ret; |
1510 | ||
1511 | BUG_ON(iocb->ki_pos != pos); | |
1512 | ||
53362a05 | 1513 | blk_start_plug(&plug); |
eef99380 CH |
1514 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1515 | if (ret > 0 || ret == -EIOCBQUEUED) { | |
1516 | ssize_t err; | |
1517 | ||
1518 | err = generic_write_sync(file, pos, ret); | |
1519 | if (err < 0 && ret > 0) | |
1520 | ret = err; | |
1521 | } | |
53362a05 | 1522 | blk_finish_plug(&plug); |
eef99380 CH |
1523 | return ret; |
1524 | } | |
1525 | EXPORT_SYMBOL_GPL(blkdev_aio_write); | |
1526 | ||
684c9aae LT |
1527 | static ssize_t blkdev_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1528 | unsigned long nr_segs, loff_t pos) | |
1529 | { | |
1530 | struct file *file = iocb->ki_filp; | |
1531 | struct inode *bd_inode = file->f_mapping->host; | |
1532 | loff_t size = i_size_read(bd_inode); | |
1533 | ||
1534 | if (pos >= size) | |
1535 | return 0; | |
1536 | ||
1537 | size -= pos; | |
6f8f5c26 | 1538 | if (size < iocb->ki_left) |
684c9aae LT |
1539 | nr_segs = iov_shorten((struct iovec *)iov, nr_segs, size); |
1540 | return generic_file_aio_read(iocb, iov, nr_segs, pos); | |
1541 | } | |
1542 | ||
87d8fe1e TT |
1543 | /* |
1544 | * Try to release a page associated with block device when the system | |
1545 | * is under memory pressure. | |
1546 | */ | |
1547 | static int blkdev_releasepage(struct page *page, gfp_t wait) | |
1548 | { | |
1549 | struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super; | |
1550 | ||
1551 | if (super && super->s_op->bdev_try_to_free_page) | |
1552 | return super->s_op->bdev_try_to_free_page(super, page, wait); | |
1553 | ||
1554 | return try_to_free_buffers(page); | |
1555 | } | |
1556 | ||
4c54ac62 | 1557 | static const struct address_space_operations def_blk_aops = { |
1da177e4 LT |
1558 | .readpage = blkdev_readpage, |
1559 | .writepage = blkdev_writepage, | |
6272b5a5 NP |
1560 | .write_begin = blkdev_write_begin, |
1561 | .write_end = blkdev_write_end, | |
1da177e4 | 1562 | .writepages = generic_writepages, |
87d8fe1e | 1563 | .releasepage = blkdev_releasepage, |
1da177e4 | 1564 | .direct_IO = blkdev_direct_IO, |
b4597226 | 1565 | .is_dirty_writeback = buffer_check_dirty_writeback, |
1da177e4 LT |
1566 | }; |
1567 | ||
4b6f5d20 | 1568 | const struct file_operations def_blk_fops = { |
1da177e4 LT |
1569 | .open = blkdev_open, |
1570 | .release = blkdev_close, | |
1571 | .llseek = block_llseek, | |
543ade1f BP |
1572 | .read = do_sync_read, |
1573 | .write = do_sync_write, | |
684c9aae | 1574 | .aio_read = blkdev_aio_read, |
eef99380 | 1575 | .aio_write = blkdev_aio_write, |
1e8b3332 | 1576 | .mmap = generic_file_mmap, |
b1dd3b28 | 1577 | .fsync = blkdev_fsync, |
bb93e3a5 | 1578 | .unlocked_ioctl = block_ioctl, |
1da177e4 LT |
1579 | #ifdef CONFIG_COMPAT |
1580 | .compat_ioctl = compat_blkdev_ioctl, | |
1581 | #endif | |
1e8b3332 LT |
1582 | .splice_read = generic_file_splice_read, |
1583 | .splice_write = generic_file_splice_write, | |
1da177e4 LT |
1584 | }; |
1585 | ||
1586 | int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg) | |
1587 | { | |
1588 | int res; | |
1589 | mm_segment_t old_fs = get_fs(); | |
1590 | set_fs(KERNEL_DS); | |
56b26add | 1591 | res = blkdev_ioctl(bdev, 0, cmd, arg); |
1da177e4 LT |
1592 | set_fs(old_fs); |
1593 | return res; | |
1594 | } | |
1595 | ||
1596 | EXPORT_SYMBOL(ioctl_by_bdev); | |
1597 | ||
1598 | /** | |
1599 | * lookup_bdev - lookup a struct block_device by name | |
94e2959e | 1600 | * @pathname: special file representing the block device |
1da177e4 | 1601 | * |
57d1b536 | 1602 | * Get a reference to the blockdevice at @pathname in the current |
1da177e4 LT |
1603 | * namespace if possible and return it. Return ERR_PTR(error) |
1604 | * otherwise. | |
1605 | */ | |
421748ec | 1606 | struct block_device *lookup_bdev(const char *pathname) |
1da177e4 LT |
1607 | { |
1608 | struct block_device *bdev; | |
1609 | struct inode *inode; | |
421748ec | 1610 | struct path path; |
1da177e4 LT |
1611 | int error; |
1612 | ||
421748ec | 1613 | if (!pathname || !*pathname) |
1da177e4 LT |
1614 | return ERR_PTR(-EINVAL); |
1615 | ||
421748ec | 1616 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 LT |
1617 | if (error) |
1618 | return ERR_PTR(error); | |
1619 | ||
421748ec | 1620 | inode = path.dentry->d_inode; |
1da177e4 LT |
1621 | error = -ENOTBLK; |
1622 | if (!S_ISBLK(inode->i_mode)) | |
1623 | goto fail; | |
1624 | error = -EACCES; | |
421748ec | 1625 | if (path.mnt->mnt_flags & MNT_NODEV) |
1da177e4 LT |
1626 | goto fail; |
1627 | error = -ENOMEM; | |
1628 | bdev = bd_acquire(inode); | |
1629 | if (!bdev) | |
1630 | goto fail; | |
1631 | out: | |
421748ec | 1632 | path_put(&path); |
1da177e4 LT |
1633 | return bdev; |
1634 | fail: | |
1635 | bdev = ERR_PTR(error); | |
1636 | goto out; | |
1637 | } | |
d5686b44 | 1638 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1639 | |
93b270f7 | 1640 | int __invalidate_device(struct block_device *bdev, bool kill_dirty) |
b71e8a4c DH |
1641 | { |
1642 | struct super_block *sb = get_super(bdev); | |
1643 | int res = 0; | |
1644 | ||
1645 | if (sb) { | |
1646 | /* | |
1647 | * no need to lock the super, get_super holds the | |
1648 | * read mutex so the filesystem cannot go away | |
1649 | * under us (->put_super runs with the write lock | |
1650 | * hold). | |
1651 | */ | |
1652 | shrink_dcache_sb(sb); | |
93b270f7 | 1653 | res = invalidate_inodes(sb, kill_dirty); |
b71e8a4c DH |
1654 | drop_super(sb); |
1655 | } | |
f98393a6 | 1656 | invalidate_bdev(bdev); |
b71e8a4c DH |
1657 | return res; |
1658 | } | |
1659 | EXPORT_SYMBOL(__invalidate_device); | |
5c0d6b60 JK |
1660 | |
1661 | void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg) | |
1662 | { | |
1663 | struct inode *inode, *old_inode = NULL; | |
1664 | ||
1665 | spin_lock(&inode_sb_list_lock); | |
1666 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { | |
1667 | struct address_space *mapping = inode->i_mapping; | |
1668 | ||
1669 | spin_lock(&inode->i_lock); | |
1670 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1671 | mapping->nrpages == 0) { | |
1672 | spin_unlock(&inode->i_lock); | |
1673 | continue; | |
1674 | } | |
1675 | __iget(inode); | |
1676 | spin_unlock(&inode->i_lock); | |
1677 | spin_unlock(&inode_sb_list_lock); | |
1678 | /* | |
1679 | * We hold a reference to 'inode' so it couldn't have been | |
1680 | * removed from s_inodes list while we dropped the | |
1681 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1682 | * be holding the last reference and we cannot iput it under | |
1683 | * inode_sb_list_lock. So we keep the reference and iput it | |
1684 | * later. | |
1685 | */ | |
1686 | iput(old_inode); | |
1687 | old_inode = inode; | |
1688 | ||
1689 | func(I_BDEV(inode), arg); | |
1690 | ||
1691 | spin_lock(&inode_sb_list_lock); | |
1692 | } | |
1693 | spin_unlock(&inode_sb_list_lock); | |
1694 | iput(old_inode); | |
1695 | } |