Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Copyright (C) 1991, 1992 Linus Torvalds |
4 | * Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE | |
7b51e703 | 5 | * Copyright (C) 2016 - 2020 Christoph Hellwig |
1da177e4 LT |
6 | */ |
7 | ||
1da177e4 LT |
8 | #include <linux/init.h> |
9 | #include <linux/mm.h> | |
1da177e4 LT |
10 | #include <linux/slab.h> |
11 | #include <linux/kmod.h> | |
12 | #include <linux/major.h> | |
7db9cfd3 | 13 | #include <linux/device_cgroup.h> |
1da177e4 | 14 | #include <linux/blkdev.h> |
fe45e630 | 15 | #include <linux/blk-integrity.h> |
66114cad | 16 | #include <linux/backing-dev.h> |
1da177e4 LT |
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> |
811d736f | 22 | #include <linux/writeback.h> |
1da177e4 | 23 | #include <linux/mount.h> |
9030d16e | 24 | #include <linux/pseudo_fs.h> |
1da177e4 LT |
25 | #include <linux/uio.h> |
26 | #include <linux/namei.h> | |
15e3d2c5 | 27 | #include <linux/part_stat.h> |
7c0f6ba6 | 28 | #include <linux/uaccess.h> |
2d985f8c | 29 | #include <linux/stat.h> |
0dca4462 CH |
30 | #include "../fs/internal.h" |
31 | #include "blk.h" | |
1da177e4 | 32 | |
ed5cc702 JK |
33 | /* Should we allow writing to mounted block devices? */ |
34 | static bool bdev_allow_write_mounted = IS_ENABLED(CONFIG_BLK_DEV_WRITE_MOUNTED); | |
35 | ||
1da177e4 LT |
36 | struct bdev_inode { |
37 | struct block_device bdev; | |
38 | struct inode vfs_inode; | |
39 | }; | |
40 | ||
41 | static inline struct bdev_inode *BDEV_I(struct inode *inode) | |
42 | { | |
43 | return container_of(inode, struct bdev_inode, vfs_inode); | |
44 | } | |
45 | ||
ff5053f6 | 46 | struct block_device *I_BDEV(struct inode *inode) |
1da177e4 LT |
47 | { |
48 | return &BDEV_I(inode)->bdev; | |
49 | } | |
1da177e4 LT |
50 | EXPORT_SYMBOL(I_BDEV); |
51 | ||
f3a60882 CB |
52 | struct block_device *file_bdev(struct file *bdev_file) |
53 | { | |
7c09a4ed | 54 | return I_BDEV(bdev_file->f_mapping->host); |
f3a60882 CB |
55 | } |
56 | EXPORT_SYMBOL(file_bdev); | |
57 | ||
dbd3ca50 | 58 | static void bdev_write_inode(struct block_device *bdev) |
564f00f6 | 59 | { |
dbd3ca50 VG |
60 | struct inode *inode = bdev->bd_inode; |
61 | int ret; | |
62 | ||
564f00f6 CH |
63 | spin_lock(&inode->i_lock); |
64 | while (inode->i_state & I_DIRTY) { | |
65 | spin_unlock(&inode->i_lock); | |
dbd3ca50 | 66 | ret = write_inode_now(inode, true); |
5bf83e9a CH |
67 | if (ret) |
68 | pr_warn_ratelimited( | |
69 | "VFS: Dirty inode writeback failed for block device %pg (err=%d).\n", | |
70 | bdev, ret); | |
564f00f6 CH |
71 | spin_lock(&inode->i_lock); |
72 | } | |
73 | spin_unlock(&inode->i_lock); | |
74 | } | |
75 | ||
f9a14399 | 76 | /* Kill _all_ buffers and pagecache , dirty or not.. */ |
3373a346 | 77 | static void kill_bdev(struct block_device *bdev) |
1da177e4 | 78 | { |
ff01bb48 AV |
79 | struct address_space *mapping = bdev->bd_inode->i_mapping; |
80 | ||
7716506a | 81 | if (mapping_empty(mapping)) |
f9a14399 | 82 | return; |
ff01bb48 | 83 | |
f9a14399 | 84 | invalidate_bh_lrus(); |
ff01bb48 | 85 | truncate_inode_pages(mapping, 0); |
3373a346 | 86 | } |
ff01bb48 AV |
87 | |
88 | /* Invalidate clean unused buffers and pagecache. */ | |
89 | void invalidate_bdev(struct block_device *bdev) | |
90 | { | |
91 | struct address_space *mapping = bdev->bd_inode->i_mapping; | |
92 | ||
a5f6a6a9 AR |
93 | if (mapping->nrpages) { |
94 | invalidate_bh_lrus(); | |
95 | lru_add_drain_all(); /* make sure all lru add caches are flushed */ | |
96 | invalidate_mapping_pages(mapping, 0, -1); | |
97 | } | |
ff01bb48 AV |
98 | } |
99 | EXPORT_SYMBOL(invalidate_bdev); | |
1da177e4 | 100 | |
384d87ef JK |
101 | /* |
102 | * Drop all buffers & page cache for given bdev range. This function bails | |
103 | * with error if bdev has other exclusive owner (such as filesystem). | |
104 | */ | |
05bdb996 | 105 | int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode, |
384d87ef JK |
106 | loff_t lstart, loff_t lend) |
107 | { | |
384d87ef JK |
108 | /* |
109 | * If we don't hold exclusive handle for the device, upgrade to it | |
110 | * while we discard the buffer cache to avoid discarding buffers | |
111 | * under live filesystem. | |
112 | */ | |
05bdb996 | 113 | if (!(mode & BLK_OPEN_EXCL)) { |
0718afd4 | 114 | int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL); |
384d87ef | 115 | if (err) |
56887cff | 116 | goto invalidate; |
384d87ef | 117 | } |
37c3fc9a | 118 | |
384d87ef | 119 | truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend); |
05bdb996 | 120 | if (!(mode & BLK_OPEN_EXCL)) |
37c3fc9a | 121 | bd_abort_claiming(bdev, truncate_bdev_range); |
384d87ef | 122 | return 0; |
56887cff JK |
123 | |
124 | invalidate: | |
125 | /* | |
126 | * Someone else has handle exclusively open. Try invalidating instead. | |
127 | * The 'end' argument is inclusive so the rounding is safe. | |
128 | */ | |
129 | return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping, | |
130 | lstart >> PAGE_SHIFT, | |
131 | lend >> PAGE_SHIFT); | |
384d87ef | 132 | } |
384d87ef | 133 | |
04906b2f JK |
134 | static void set_init_blocksize(struct block_device *bdev) |
135 | { | |
8dc932d3 MM |
136 | unsigned int bsize = bdev_logical_block_size(bdev); |
137 | loff_t size = i_size_read(bdev->bd_inode); | |
138 | ||
139 | while (bsize < PAGE_SIZE) { | |
140 | if (size & bsize) | |
141 | break; | |
142 | bsize <<= 1; | |
143 | } | |
144 | bdev->bd_inode->i_blkbits = blksize_bits(bsize); | |
04906b2f JK |
145 | } |
146 | ||
1da177e4 LT |
147 | int set_blocksize(struct block_device *bdev, int size) |
148 | { | |
149 | /* Size must be a power of two, and between 512 and PAGE_SIZE */ | |
1368c4f2 | 150 | if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) |
1da177e4 LT |
151 | return -EINVAL; |
152 | ||
153 | /* Size cannot be smaller than the size supported by the device */ | |
e1defc4f | 154 | if (size < bdev_logical_block_size(bdev)) |
1da177e4 LT |
155 | return -EINVAL; |
156 | ||
157 | /* Don't change the size if it is same as current */ | |
6b7b181b | 158 | if (bdev->bd_inode->i_blkbits != blksize_bits(size)) { |
1da177e4 | 159 | sync_blockdev(bdev); |
1da177e4 LT |
160 | bdev->bd_inode->i_blkbits = blksize_bits(size); |
161 | kill_bdev(bdev); | |
162 | } | |
163 | return 0; | |
164 | } | |
165 | ||
166 | EXPORT_SYMBOL(set_blocksize); | |
167 | ||
168 | int sb_set_blocksize(struct super_block *sb, int size) | |
169 | { | |
1da177e4 LT |
170 | if (set_blocksize(sb->s_bdev, size)) |
171 | return 0; | |
172 | /* If we get here, we know size is power of two | |
173 | * and it's value is between 512 and PAGE_SIZE */ | |
174 | sb->s_blocksize = size; | |
38885bd4 | 175 | sb->s_blocksize_bits = blksize_bits(size); |
1da177e4 LT |
176 | return sb->s_blocksize; |
177 | } | |
178 | ||
179 | EXPORT_SYMBOL(sb_set_blocksize); | |
180 | ||
181 | int sb_min_blocksize(struct super_block *sb, int size) | |
182 | { | |
e1defc4f | 183 | int minsize = bdev_logical_block_size(sb->s_bdev); |
1da177e4 LT |
184 | if (size < minsize) |
185 | size = minsize; | |
186 | return sb_set_blocksize(sb, size); | |
187 | } | |
188 | ||
189 | EXPORT_SYMBOL(sb_min_blocksize); | |
190 | ||
70164eb6 | 191 | int sync_blockdev_nowait(struct block_device *bdev) |
5cee5815 JK |
192 | { |
193 | if (!bdev) | |
194 | return 0; | |
70164eb6 | 195 | return filemap_flush(bdev->bd_inode->i_mapping); |
5cee5815 | 196 | } |
70164eb6 | 197 | EXPORT_SYMBOL_GPL(sync_blockdev_nowait); |
5cee5815 | 198 | |
585d3bc0 NP |
199 | /* |
200 | * Write out and wait upon all the dirty data associated with a block | |
201 | * device via its mapping. Does not take the superblock lock. | |
202 | */ | |
203 | int sync_blockdev(struct block_device *bdev) | |
204 | { | |
70164eb6 CH |
205 | if (!bdev) |
206 | return 0; | |
207 | return filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
585d3bc0 NP |
208 | } |
209 | EXPORT_SYMBOL(sync_blockdev); | |
210 | ||
97d6fb1b YM |
211 | int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend) |
212 | { | |
213 | return filemap_write_and_wait_range(bdev->bd_inode->i_mapping, | |
214 | lstart, lend); | |
215 | } | |
216 | EXPORT_SYMBOL(sync_blockdev_range); | |
217 | ||
585d3bc0 | 218 | /** |
982c3b30 | 219 | * bdev_freeze - lock a filesystem and force it into a consistent state |
585d3bc0 NP |
220 | * @bdev: blockdevice to lock |
221 | * | |
585d3bc0 NP |
222 | * If a superblock is found on this device, we take the s_umount semaphore |
223 | * on it to make sure nobody unmounts until the snapshot creation is done. | |
224 | * The reference counter (bd_fsfreeze_count) guarantees that only the last | |
225 | * unfreeze process can unfreeze the frozen filesystem actually when multiple | |
982c3b30 CB |
226 | * freeze requests arrive simultaneously. It counts up in bdev_freeze() and |
227 | * count down in bdev_thaw(). When it becomes 0, thaw_bdev() will unfreeze | |
585d3bc0 | 228 | * actually. |
982c3b30 CB |
229 | * |
230 | * Return: On success zero is returned, negative error code on failure. | |
585d3bc0 | 231 | */ |
982c3b30 | 232 | int bdev_freeze(struct block_device *bdev) |
585d3bc0 | 233 | { |
585d3bc0 NP |
234 | int error = 0; |
235 | ||
236 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
040f04bd | 237 | |
49ef8832 CB |
238 | if (atomic_inc_return(&bdev->bd_fsfreeze_count) > 1) { |
239 | mutex_unlock(&bdev->bd_fsfreeze_mutex); | |
240 | return 0; | |
585d3bc0 | 241 | } |
040f04bd | 242 | |
49ef8832 CB |
243 | mutex_lock(&bdev->bd_holder_lock); |
244 | if (bdev->bd_holder_ops && bdev->bd_holder_ops->freeze) { | |
245 | error = bdev->bd_holder_ops->freeze(bdev); | |
246 | lockdep_assert_not_held(&bdev->bd_holder_lock); | |
247 | } else { | |
248 | mutex_unlock(&bdev->bd_holder_lock); | |
249 | error = sync_blockdev(bdev); | |
585d3bc0 | 250 | } |
040f04bd | 251 | |
49ef8832 CB |
252 | if (error) |
253 | atomic_dec(&bdev->bd_fsfreeze_count); | |
254 | ||
585d3bc0 | 255 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
040f04bd | 256 | return error; |
585d3bc0 | 257 | } |
982c3b30 | 258 | EXPORT_SYMBOL(bdev_freeze); |
585d3bc0 NP |
259 | |
260 | /** | |
982c3b30 | 261 | * bdev_thaw - unlock filesystem |
585d3bc0 | 262 | * @bdev: blockdevice to unlock |
585d3bc0 | 263 | * |
982c3b30 CB |
264 | * Unlocks the filesystem and marks it writeable again after bdev_freeze(). |
265 | * | |
266 | * Return: On success zero is returned, negative error code on failure. | |
585d3bc0 | 267 | */ |
982c3b30 | 268 | int bdev_thaw(struct block_device *bdev) |
585d3bc0 | 269 | { |
49ef8832 | 270 | int error = -EINVAL, nr_freeze; |
585d3bc0 NP |
271 | |
272 | mutex_lock(&bdev->bd_fsfreeze_mutex); | |
49ef8832 CB |
273 | |
274 | /* | |
275 | * If this returns < 0 it means that @bd_fsfreeze_count was | |
276 | * already 0 and no decrement was performed. | |
277 | */ | |
278 | nr_freeze = atomic_dec_if_positive(&bdev->bd_fsfreeze_count); | |
279 | if (nr_freeze < 0) | |
18e9e510 | 280 | goto out; |
4504230a CH |
281 | |
282 | error = 0; | |
49ef8832 | 283 | if (nr_freeze > 0) |
18e9e510 | 284 | goto out; |
4504230a | 285 | |
49ef8832 CB |
286 | mutex_lock(&bdev->bd_holder_lock); |
287 | if (bdev->bd_holder_ops && bdev->bd_holder_ops->thaw) { | |
288 | error = bdev->bd_holder_ops->thaw(bdev); | |
289 | lockdep_assert_not_held(&bdev->bd_holder_lock); | |
290 | } else { | |
291 | mutex_unlock(&bdev->bd_holder_lock); | |
292 | } | |
4504230a | 293 | |
997198ba | 294 | if (error) |
49ef8832 | 295 | atomic_inc(&bdev->bd_fsfreeze_count); |
18e9e510 | 296 | out: |
585d3bc0 | 297 | mutex_unlock(&bdev->bd_fsfreeze_mutex); |
997198ba | 298 | return error; |
585d3bc0 | 299 | } |
982c3b30 | 300 | EXPORT_SYMBOL(bdev_thaw); |
585d3bc0 | 301 | |
1da177e4 LT |
302 | /* |
303 | * pseudo-fs | |
304 | */ | |
305 | ||
74e6464a | 306 | static __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock); |
68279f9c | 307 | static struct kmem_cache *bdev_cachep __ro_after_init; |
1da177e4 LT |
308 | |
309 | static struct inode *bdev_alloc_inode(struct super_block *sb) | |
310 | { | |
fd60b288 | 311 | struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL); |
2d2f6f1b | 312 | |
1da177e4 LT |
313 | if (!ei) |
314 | return NULL; | |
2d2f6f1b | 315 | memset(&ei->bdev, 0, sizeof(ei->bdev)); |
1da177e4 LT |
316 | return &ei->vfs_inode; |
317 | } | |
318 | ||
41149cb0 | 319 | static void bdev_free_inode(struct inode *inode) |
1da177e4 | 320 | { |
15e3d2c5 CH |
321 | struct block_device *bdev = I_BDEV(inode); |
322 | ||
323 | free_percpu(bdev->bd_stats); | |
231926db | 324 | kfree(bdev->bd_meta_info); |
15e3d2c5 | 325 | |
889c05cc CH |
326 | if (!bdev_is_partition(bdev)) { |
327 | if (bdev->bd_disk && bdev->bd_disk->bdi) | |
328 | bdi_put(bdev->bd_disk->bdi); | |
340e8457 | 329 | kfree(bdev->bd_disk); |
889c05cc | 330 | } |
9451aa0a CH |
331 | |
332 | if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR) | |
333 | blk_free_ext_minor(MINOR(bdev->bd_dev)); | |
334 | ||
41149cb0 | 335 | kmem_cache_free(bdev_cachep, BDEV_I(inode)); |
fa0d7e3d NP |
336 | } |
337 | ||
e6cb5382 | 338 | static void init_once(void *data) |
1da177e4 | 339 | { |
e6cb5382 | 340 | struct bdev_inode *ei = data; |
1da177e4 | 341 | |
a35afb83 | 342 | inode_init_once(&ei->vfs_inode); |
1da177e4 LT |
343 | } |
344 | ||
b57922d9 | 345 | static void bdev_evict_inode(struct inode *inode) |
1da177e4 | 346 | { |
91b0abe3 | 347 | truncate_inode_pages_final(&inode->i_data); |
b57922d9 | 348 | invalidate_inode_buffers(inode); /* is it needed here? */ |
dbd5768f | 349 | clear_inode(inode); |
1da177e4 LT |
350 | } |
351 | ||
ee9b6d61 | 352 | static const struct super_operations bdev_sops = { |
1da177e4 LT |
353 | .statfs = simple_statfs, |
354 | .alloc_inode = bdev_alloc_inode, | |
41149cb0 | 355 | .free_inode = bdev_free_inode, |
1da177e4 | 356 | .drop_inode = generic_delete_inode, |
b57922d9 | 357 | .evict_inode = bdev_evict_inode, |
1da177e4 LT |
358 | }; |
359 | ||
9030d16e | 360 | static int bd_init_fs_context(struct fs_context *fc) |
1da177e4 | 361 | { |
9030d16e DH |
362 | struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); |
363 | if (!ctx) | |
364 | return -ENOMEM; | |
365 | fc->s_iflags |= SB_I_CGROUPWB; | |
366 | ctx->ops = &bdev_sops; | |
367 | return 0; | |
1da177e4 LT |
368 | } |
369 | ||
370 | static struct file_system_type bd_type = { | |
371 | .name = "bdev", | |
9030d16e | 372 | .init_fs_context = bd_init_fs_context, |
1da177e4 LT |
373 | .kill_sb = kill_anon_super, |
374 | }; | |
375 | ||
68279f9c | 376 | struct super_block *blockdev_superblock __ro_after_init; |
f3a60882 | 377 | struct vfsmount *blockdev_mnt __ro_after_init; |
a212b105 | 378 | EXPORT_SYMBOL_GPL(blockdev_superblock); |
1da177e4 LT |
379 | |
380 | void __init bdev_cache_init(void) | |
381 | { | |
382 | int err; | |
c2acf7b9 | 383 | |
1da177e4 | 384 | bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), |
fffb60f9 | 385 | 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| |
82c6515d | 386 | SLAB_ACCOUNT|SLAB_PANIC), |
20c2df83 | 387 | init_once); |
1da177e4 LT |
388 | err = register_filesystem(&bd_type); |
389 | if (err) | |
390 | panic("Cannot register bdev pseudo-fs"); | |
f3a60882 CB |
391 | blockdev_mnt = kern_mount(&bd_type); |
392 | if (IS_ERR(blockdev_mnt)) | |
1da177e4 | 393 | panic("Cannot create bdev pseudo-fs"); |
f3a60882 | 394 | blockdev_superblock = blockdev_mnt->mnt_sb; /* For writeback */ |
1da177e4 LT |
395 | } |
396 | ||
22ae8ce8 | 397 | struct block_device *bdev_alloc(struct gendisk *disk, u8 partno) |
1da177e4 LT |
398 | { |
399 | struct block_device *bdev; | |
400 | struct inode *inode; | |
401 | ||
22ae8ce8 | 402 | inode = new_inode(blockdev_superblock); |
1da177e4 LT |
403 | if (!inode) |
404 | return NULL; | |
22ae8ce8 CH |
405 | inode->i_mode = S_IFBLK; |
406 | inode->i_rdev = 0; | |
407 | inode->i_data.a_ops = &def_blk_aops; | |
408 | mapping_set_gfp_mask(&inode->i_data, GFP_USER); | |
409 | ||
410 | bdev = I_BDEV(inode); | |
e6cb5382 | 411 | mutex_init(&bdev->bd_fsfreeze_mutex); |
22ae8ce8 | 412 | spin_lock_init(&bdev->bd_size_lock); |
0718afd4 | 413 | mutex_init(&bdev->bd_holder_lock); |
22ae8ce8 | 414 | bdev->bd_partno = partno; |
22ae8ce8 | 415 | bdev->bd_inode = inode; |
17220ca5 | 416 | bdev->bd_queue = disk->queue; |
38c8e3df ML |
417 | if (partno) |
418 | bdev->bd_has_submit_bio = disk->part0->bd_has_submit_bio; | |
419 | else | |
420 | bdev->bd_has_submit_bio = false; | |
15e3d2c5 CH |
421 | bdev->bd_stats = alloc_percpu(struct disk_stats); |
422 | if (!bdev->bd_stats) { | |
423 | iput(inode); | |
424 | return NULL; | |
425 | } | |
06cc978d | 426 | bdev->bd_disk = disk; |
22ae8ce8 CH |
427 | return bdev; |
428 | } | |
1da177e4 | 429 | |
83794367 DLM |
430 | void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors) |
431 | { | |
432 | spin_lock(&bdev->bd_size_lock); | |
433 | i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); | |
434 | bdev->bd_nr_sectors = sectors; | |
435 | spin_unlock(&bdev->bd_size_lock); | |
436 | } | |
437 | ||
22ae8ce8 CH |
438 | void bdev_add(struct block_device *bdev, dev_t dev) |
439 | { | |
1898efcd CH |
440 | if (bdev_stable_writes(bdev)) |
441 | mapping_set_stable_writes(bdev->bd_inode->i_mapping); | |
22ae8ce8 CH |
442 | bdev->bd_dev = dev; |
443 | bdev->bd_inode->i_rdev = dev; | |
444 | bdev->bd_inode->i_ino = dev; | |
445 | insert_inode_hash(bdev->bd_inode); | |
446 | } | |
1da177e4 | 447 | |
1da177e4 LT |
448 | long nr_blockdev_pages(void) |
449 | { | |
1008fe6d | 450 | struct inode *inode; |
1da177e4 | 451 | long ret = 0; |
1008fe6d CH |
452 | |
453 | spin_lock(&blockdev_superblock->s_inode_list_lock); | |
454 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) | |
455 | ret += inode->i_mapping->nrpages; | |
456 | spin_unlock(&blockdev_superblock->s_inode_list_lock); | |
457 | ||
1da177e4 LT |
458 | return ret; |
459 | } | |
460 | ||
1a3cbbc5 TH |
461 | /** |
462 | * bd_may_claim - test whether a block device can be claimed | |
463 | * @bdev: block device of interest | |
1a3cbbc5 | 464 | * @holder: holder trying to claim @bdev |
0718afd4 | 465 | * @hops: holder ops |
1a3cbbc5 | 466 | * |
25985edc | 467 | * Test whether @bdev can be claimed by @holder. |
1a3cbbc5 | 468 | * |
1a3cbbc5 TH |
469 | * RETURNS: |
470 | * %true if @bdev can be claimed, %false otherwise. | |
471 | */ | |
0718afd4 CH |
472 | static bool bd_may_claim(struct block_device *bdev, void *holder, |
473 | const struct blk_holder_ops *hops) | |
ae5f855e CH |
474 | { |
475 | struct block_device *whole = bdev_whole(bdev); | |
476 | ||
74e6464a CH |
477 | lockdep_assert_held(&bdev_lock); |
478 | ||
ae5f855e CH |
479 | if (bdev->bd_holder) { |
480 | /* | |
481 | * The same holder can always re-claim. | |
482 | */ | |
0718afd4 CH |
483 | if (bdev->bd_holder == holder) { |
484 | if (WARN_ON_ONCE(bdev->bd_holder_ops != hops)) | |
485 | return false; | |
ae5f855e | 486 | return true; |
0718afd4 | 487 | } |
ae5f855e CH |
488 | return false; |
489 | } | |
490 | ||
491 | /* | |
492 | * If the whole devices holder is set to bd_may_claim, a partition on | |
493 | * the device is claimed, but not the whole device. | |
494 | */ | |
495 | if (whole != bdev && | |
496 | whole->bd_holder && whole->bd_holder != bd_may_claim) | |
497 | return false; | |
498 | return true; | |
1a3cbbc5 TH |
499 | } |
500 | ||
6b4517a7 | 501 | /** |
58e46ed9 | 502 | * bd_prepare_to_claim - claim a block device |
6b4517a7 | 503 | * @bdev: block device of interest |
6b4517a7 | 504 | * @holder: holder trying to claim @bdev |
0718afd4 | 505 | * @hops: holder ops. |
6b4517a7 | 506 | * |
58e46ed9 CH |
507 | * Claim @bdev. This function fails if @bdev is already claimed by another |
508 | * holder and waits if another claiming is in progress. return, the caller | |
509 | * has ownership of bd_claiming and bd_holder[s]. | |
6b4517a7 TH |
510 | * |
511 | * RETURNS: | |
512 | * 0 if @bdev can be claimed, -EBUSY otherwise. | |
513 | */ | |
0718afd4 CH |
514 | int bd_prepare_to_claim(struct block_device *bdev, void *holder, |
515 | const struct blk_holder_ops *hops) | |
6b4517a7 | 516 | { |
37c3fc9a CH |
517 | struct block_device *whole = bdev_whole(bdev); |
518 | ||
519 | if (WARN_ON_ONCE(!holder)) | |
520 | return -EINVAL; | |
6b4517a7 | 521 | retry: |
74e6464a | 522 | mutex_lock(&bdev_lock); |
6b4517a7 | 523 | /* if someone else claimed, fail */ |
0718afd4 | 524 | if (!bd_may_claim(bdev, holder, hops)) { |
74e6464a | 525 | mutex_unlock(&bdev_lock); |
6b4517a7 | 526 | return -EBUSY; |
58e46ed9 | 527 | } |
6b4517a7 | 528 | |
e75aa858 TH |
529 | /* if claiming is already in progress, wait for it to finish */ |
530 | if (whole->bd_claiming) { | |
6b4517a7 TH |
531 | wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); |
532 | DEFINE_WAIT(wait); | |
533 | ||
534 | prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); | |
74e6464a | 535 | mutex_unlock(&bdev_lock); |
6b4517a7 TH |
536 | schedule(); |
537 | finish_wait(wq, &wait); | |
6b4517a7 TH |
538 | goto retry; |
539 | } | |
540 | ||
541 | /* yay, all mine */ | |
58e46ed9 | 542 | whole->bd_claiming = holder; |
74e6464a | 543 | mutex_unlock(&bdev_lock); |
6b4517a7 TH |
544 | return 0; |
545 | } | |
ecbe6bc0 | 546 | EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */ |
6b4517a7 | 547 | |
89e524c0 JK |
548 | static void bd_clear_claiming(struct block_device *whole, void *holder) |
549 | { | |
550 | lockdep_assert_held(&bdev_lock); | |
551 | /* tell others that we're done */ | |
552 | BUG_ON(whole->bd_claiming != holder); | |
553 | whole->bd_claiming = NULL; | |
554 | wake_up_bit(&whole->bd_claiming, 0); | |
555 | } | |
556 | ||
557 | /** | |
558 | * bd_finish_claiming - finish claiming of a block device | |
559 | * @bdev: block device of interest | |
89e524c0 | 560 | * @holder: holder that has claimed @bdev |
017fb83e | 561 | * @hops: block device holder operations |
89e524c0 JK |
562 | * |
563 | * Finish exclusive open of a block device. Mark the device as exlusively | |
564 | * open by the holder and wake up all waiters for exclusive open to finish. | |
565 | */ | |
0718afd4 CH |
566 | static void bd_finish_claiming(struct block_device *bdev, void *holder, |
567 | const struct blk_holder_ops *hops) | |
89e524c0 | 568 | { |
37c3fc9a CH |
569 | struct block_device *whole = bdev_whole(bdev); |
570 | ||
74e6464a | 571 | mutex_lock(&bdev_lock); |
0718afd4 | 572 | BUG_ON(!bd_may_claim(bdev, holder, hops)); |
89e524c0 JK |
573 | /* |
574 | * Note that for a whole device bd_holders will be incremented twice, | |
575 | * and bd_holder will be set to bd_may_claim before being set to holder | |
576 | */ | |
577 | whole->bd_holders++; | |
578 | whole->bd_holder = bd_may_claim; | |
579 | bdev->bd_holders++; | |
0718afd4 | 580 | mutex_lock(&bdev->bd_holder_lock); |
89e524c0 | 581 | bdev->bd_holder = holder; |
0718afd4 CH |
582 | bdev->bd_holder_ops = hops; |
583 | mutex_unlock(&bdev->bd_holder_lock); | |
89e524c0 | 584 | bd_clear_claiming(whole, holder); |
74e6464a | 585 | mutex_unlock(&bdev_lock); |
89e524c0 | 586 | } |
89e524c0 JK |
587 | |
588 | /** | |
589 | * bd_abort_claiming - abort claiming of a block device | |
590 | * @bdev: block device of interest | |
89e524c0 JK |
591 | * @holder: holder that has claimed @bdev |
592 | * | |
593 | * Abort claiming of a block device when the exclusive open failed. This can be | |
594 | * also used when exclusive open is not actually desired and we just needed | |
595 | * to block other exclusive openers for a while. | |
596 | */ | |
37c3fc9a | 597 | void bd_abort_claiming(struct block_device *bdev, void *holder) |
89e524c0 | 598 | { |
74e6464a | 599 | mutex_lock(&bdev_lock); |
37c3fc9a | 600 | bd_clear_claiming(bdev_whole(bdev), holder); |
74e6464a | 601 | mutex_unlock(&bdev_lock); |
89e524c0 JK |
602 | } |
603 | EXPORT_SYMBOL(bd_abort_claiming); | |
6b4517a7 | 604 | |
2736e8ee | 605 | static void bd_end_claim(struct block_device *bdev, void *holder) |
0783b1a7 CH |
606 | { |
607 | struct block_device *whole = bdev_whole(bdev); | |
608 | bool unblock = false; | |
609 | ||
610 | /* | |
611 | * Release a claim on the device. The holder fields are protected with | |
612 | * bdev_lock. open_mutex is used to synchronize disk_holder unlinking. | |
613 | */ | |
74e6464a | 614 | mutex_lock(&bdev_lock); |
2736e8ee | 615 | WARN_ON_ONCE(bdev->bd_holder != holder); |
0783b1a7 CH |
616 | WARN_ON_ONCE(--bdev->bd_holders < 0); |
617 | WARN_ON_ONCE(--whole->bd_holders < 0); | |
618 | if (!bdev->bd_holders) { | |
0718afd4 | 619 | mutex_lock(&bdev->bd_holder_lock); |
0783b1a7 | 620 | bdev->bd_holder = NULL; |
0718afd4 CH |
621 | bdev->bd_holder_ops = NULL; |
622 | mutex_unlock(&bdev->bd_holder_lock); | |
0783b1a7 CH |
623 | if (bdev->bd_write_holder) |
624 | unblock = true; | |
625 | } | |
626 | if (!whole->bd_holders) | |
627 | whole->bd_holder = NULL; | |
74e6464a | 628 | mutex_unlock(&bdev_lock); |
0783b1a7 CH |
629 | |
630 | /* | |
631 | * If this was the last claim, remove holder link and unblock evpoll if | |
632 | * it was a write holder. | |
633 | */ | |
634 | if (unblock) { | |
635 | disk_unblock_events(bdev->bd_disk); | |
636 | bdev->bd_write_holder = false; | |
637 | } | |
638 | } | |
639 | ||
c8276b95 CH |
640 | static void blkdev_flush_mapping(struct block_device *bdev) |
641 | { | |
642 | WARN_ON_ONCE(bdev->bd_holders); | |
643 | sync_blockdev(bdev); | |
644 | kill_bdev(bdev); | |
645 | bdev_write_inode(bdev); | |
646 | } | |
37be4124 | 647 | |
05bdb996 | 648 | static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode) |
a1548b67 | 649 | { |
142fe8f4 | 650 | struct gendisk *disk = bdev->bd_disk; |
af22fef3 | 651 | int ret; |
a1548b67 | 652 | |
362529d9 | 653 | if (disk->fops->open) { |
d32e2bf8 | 654 | ret = disk->fops->open(disk, mode); |
362529d9 CH |
655 | if (ret) { |
656 | /* avoid ghost partitions on a removed medium */ | |
657 | if (ret == -ENOMEDIUM && | |
658 | test_bit(GD_NEED_PART_SCAN, &disk->state)) | |
0384264e | 659 | bdev_disk_changed(disk, true); |
362529d9 CH |
660 | return ret; |
661 | } | |
d981cb5b | 662 | } |
a1548b67 | 663 | |
9acf381f | 664 | if (!atomic_read(&bdev->bd_openers)) |
362529d9 | 665 | set_init_blocksize(bdev); |
362529d9 | 666 | if (test_bit(GD_NEED_PART_SCAN, &disk->state)) |
0384264e | 667 | bdev_disk_changed(disk, false); |
9acf381f | 668 | atomic_inc(&bdev->bd_openers); |
483546c1 | 669 | return 0; |
362529d9 | 670 | } |
a1548b67 | 671 | |
ae220766 | 672 | static void blkdev_put_whole(struct block_device *bdev) |
c8276b95 | 673 | { |
9acf381f | 674 | if (atomic_dec_and_test(&bdev->bd_openers)) |
c8276b95 CH |
675 | blkdev_flush_mapping(bdev); |
676 | if (bdev->bd_disk->fops->release) | |
ae220766 | 677 | bdev->bd_disk->fops->release(bdev->bd_disk); |
a1548b67 CH |
678 | } |
679 | ||
05bdb996 | 680 | static int blkdev_get_part(struct block_device *part, blk_mode_t mode) |
1da177e4 | 681 | { |
362529d9 | 682 | struct gendisk *disk = part->bd_disk; |
362529d9 | 683 | int ret; |
5b642d8b | 684 | |
9d3b8813 | 685 | ret = blkdev_get_whole(bdev_whole(part), mode); |
a8698707 | 686 | if (ret) |
9d3b8813 | 687 | return ret; |
7e69723f | 688 | |
362529d9 CH |
689 | ret = -ENXIO; |
690 | if (!bdev_nr_sectors(part)) | |
691 | goto out_blkdev_put; | |
7e69723f | 692 | |
9d1c9287 CH |
693 | if (!atomic_read(&part->bd_openers)) { |
694 | disk->open_partitions++; | |
695 | set_init_blocksize(part); | |
696 | } | |
9acf381f | 697 | atomic_inc(&part->bd_openers); |
1da177e4 | 698 | return 0; |
5a023cdb | 699 | |
362529d9 | 700 | out_blkdev_put: |
ae220766 | 701 | blkdev_put_whole(bdev_whole(part)); |
362529d9 | 702 | return ret; |
1da177e4 | 703 | } |
5b56b6ed | 704 | |
a56aefca CB |
705 | int bdev_permission(dev_t dev, blk_mode_t mode, void *holder) |
706 | { | |
707 | int ret; | |
708 | ||
709 | ret = devcgroup_check_permission(DEVCG_DEV_BLOCK, | |
710 | MAJOR(dev), MINOR(dev), | |
711 | ((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) | | |
712 | ((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0)); | |
713 | if (ret) | |
714 | return ret; | |
715 | ||
716 | /* Blocking writes requires exclusive opener */ | |
717 | if (mode & BLK_OPEN_RESTRICT_WRITES && !holder) | |
718 | return -EINVAL; | |
719 | ||
ab838b3f CB |
720 | /* |
721 | * We're using error pointers to indicate to ->release() when we | |
722 | * failed to open that block device. Also this doesn't make sense. | |
723 | */ | |
724 | if (WARN_ON_ONCE(IS_ERR(holder))) | |
725 | return -EINVAL; | |
726 | ||
a56aefca CB |
727 | return 0; |
728 | } | |
729 | ||
ae220766 | 730 | static void blkdev_put_part(struct block_device *part) |
c8276b95 CH |
731 | { |
732 | struct block_device *whole = bdev_whole(part); | |
03e26279 | 733 | |
9d1c9287 CH |
734 | if (atomic_dec_and_test(&part->bd_openers)) { |
735 | blkdev_flush_mapping(part); | |
736 | whole->bd_disk->open_partitions--; | |
737 | } | |
ae220766 | 738 | blkdev_put_whole(whole); |
1da177e4 LT |
739 | } |
740 | ||
22ae8ce8 CH |
741 | struct block_device *blkdev_get_no_open(dev_t dev) |
742 | { | |
743 | struct block_device *bdev; | |
9d3b8813 | 744 | struct inode *inode; |
22ae8ce8 | 745 | |
9d3b8813 | 746 | inode = ilookup(blockdev_superblock, dev); |
fbdee71b | 747 | if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) { |
22ae8ce8 | 748 | blk_request_module(dev); |
9d3b8813 | 749 | inode = ilookup(blockdev_superblock, dev); |
fbdee71b CH |
750 | if (inode) |
751 | pr_warn_ratelimited( | |
451f0b6f | 752 | "block device autoloading is deprecated and will be removed.\n"); |
22ae8ce8 | 753 | } |
fbdee71b CH |
754 | if (!inode) |
755 | return NULL; | |
22ae8ce8 | 756 | |
9d3b8813 CH |
757 | /* switch from the inode reference to a device mode one: */ |
758 | bdev = &BDEV_I(inode)->bdev; | |
759 | if (!kobject_get_unless_zero(&bdev->bd_device.kobj)) | |
760 | bdev = NULL; | |
761 | iput(inode); | |
22ae8ce8 | 762 | return bdev; |
22ae8ce8 CH |
763 | } |
764 | ||
765 | void blkdev_put_no_open(struct block_device *bdev) | |
766 | { | |
9d3b8813 | 767 | put_device(&bdev->bd_device); |
22ae8ce8 | 768 | } |
ed5cc702 JK |
769 | |
770 | static bool bdev_writes_blocked(struct block_device *bdev) | |
771 | { | |
3ff56e28 | 772 | return bdev->bd_writers < 0; |
ed5cc702 JK |
773 | } |
774 | ||
775 | static void bdev_block_writes(struct block_device *bdev) | |
776 | { | |
3ff56e28 | 777 | bdev->bd_writers--; |
ed5cc702 JK |
778 | } |
779 | ||
780 | static void bdev_unblock_writes(struct block_device *bdev) | |
781 | { | |
3ff56e28 | 782 | bdev->bd_writers++; |
ed5cc702 JK |
783 | } |
784 | ||
785 | static bool bdev_may_open(struct block_device *bdev, blk_mode_t mode) | |
786 | { | |
787 | if (bdev_allow_write_mounted) | |
788 | return true; | |
789 | /* Writes blocked? */ | |
790 | if (mode & BLK_OPEN_WRITE && bdev_writes_blocked(bdev)) | |
791 | return false; | |
792 | if (mode & BLK_OPEN_RESTRICT_WRITES && bdev->bd_writers > 0) | |
793 | return false; | |
794 | return true; | |
795 | } | |
796 | ||
797 | static void bdev_claim_write_access(struct block_device *bdev, blk_mode_t mode) | |
798 | { | |
799 | if (bdev_allow_write_mounted) | |
800 | return; | |
801 | ||
802 | /* Claim exclusive or shared write access. */ | |
803 | if (mode & BLK_OPEN_RESTRICT_WRITES) | |
804 | bdev_block_writes(bdev); | |
805 | else if (mode & BLK_OPEN_WRITE) | |
806 | bdev->bd_writers++; | |
807 | } | |
808 | ||
22650a99 CB |
809 | static inline bool bdev_unclaimed(const struct file *bdev_file) |
810 | { | |
811 | return bdev_file->private_data == BDEV_I(bdev_file->f_mapping->host); | |
812 | } | |
813 | ||
ab838b3f | 814 | static void bdev_yield_write_access(struct file *bdev_file) |
ed5cc702 | 815 | { |
321de651 CB |
816 | struct block_device *bdev; |
817 | ||
ed5cc702 JK |
818 | if (bdev_allow_write_mounted) |
819 | return; | |
820 | ||
22650a99 CB |
821 | if (bdev_unclaimed(bdev_file)) |
822 | return; | |
823 | ||
321de651 | 824 | bdev = file_bdev(bdev_file); |
ddd65e19 CB |
825 | |
826 | if (bdev_file->f_mode & FMODE_WRITE_RESTRICTED) | |
827 | bdev_unblock_writes(bdev); | |
828 | else if (bdev_file->f_mode & FMODE_WRITE) | |
829 | bdev->bd_writers--; | |
ed5cc702 JK |
830 | } |
831 | ||
d4d77629 | 832 | /** |
a56aefca CB |
833 | * bdev_open - open a block device |
834 | * @bdev: block device to open | |
05bdb996 | 835 | * @mode: open mode (BLK_OPEN_*) |
d4d77629 | 836 | * @holder: exclusive holder identifier |
0718afd4 | 837 | * @hops: holder operations |
a56aefca | 838 | * @bdev_file: file for the block device |
d4d77629 | 839 | * |
a56aefca CB |
840 | * Open the block device. If @holder is not %NULL, the block device is opened |
841 | * with exclusive access. Exclusive opens may nest for the same @holder. | |
d4d77629 TH |
842 | * |
843 | * CONTEXT: | |
844 | * Might sleep. | |
845 | * | |
846 | * RETURNS: | |
a56aefca | 847 | * zero on success, -errno on failure. |
d4d77629 | 848 | */ |
a56aefca CB |
849 | int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder, |
850 | const struct blk_holder_ops *hops, struct file *bdev_file) | |
1da177e4 | 851 | { |
cd34758c | 852 | bool unblock_events = true; |
a56aefca | 853 | struct gendisk *disk = bdev->bd_disk; |
5b56b6ed | 854 | int ret; |
e525fd89 | 855 | |
2736e8ee | 856 | if (holder) { |
05bdb996 | 857 | mode |= BLK_OPEN_EXCL; |
0718afd4 | 858 | ret = bd_prepare_to_claim(bdev, holder, hops); |
5b56b6ed | 859 | if (ret) |
ab838b3f | 860 | return ret; |
2736e8ee | 861 | } else { |
ab838b3f CB |
862 | if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) |
863 | return -EIO; | |
5b56b6ed CH |
864 | } |
865 | ||
866 | disk_block_events(disk); | |
867 | ||
a8698707 | 868 | mutex_lock(&disk->open_mutex); |
362529d9 | 869 | ret = -ENXIO; |
50b4aecf | 870 | if (!disk_live(disk)) |
362529d9 | 871 | goto abort_claiming; |
efcf5932 ML |
872 | if (!try_module_get(disk->fops->owner)) |
873 | goto abort_claiming; | |
ed5cc702 JK |
874 | ret = -EBUSY; |
875 | if (!bdev_may_open(bdev, mode)) | |
876 | goto abort_claiming; | |
362529d9 CH |
877 | if (bdev_is_partition(bdev)) |
878 | ret = blkdev_get_part(bdev, mode); | |
879 | else | |
880 | ret = blkdev_get_whole(bdev, mode); | |
22ae8ce8 | 881 | if (ret) |
efcf5932 | 882 | goto put_module; |
ed5cc702 | 883 | bdev_claim_write_access(bdev, mode); |
2736e8ee | 884 | if (holder) { |
0718afd4 | 885 | bd_finish_claiming(bdev, holder, hops); |
5b56b6ed CH |
886 | |
887 | /* | |
888 | * Block event polling for write claims if requested. Any write | |
889 | * holder makes the write_holder state stick until all are | |
890 | * released. This is good enough and tracking individual | |
891 | * writeable reference is too fragile given the way @mode is | |
892 | * used in blkdev_get/put(). | |
893 | */ | |
05bdb996 | 894 | if ((mode & BLK_OPEN_WRITE) && !bdev->bd_write_holder && |
1545e0b4 | 895 | (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) { |
5b56b6ed CH |
896 | bdev->bd_write_holder = true; |
897 | unblock_events = false; | |
898 | } | |
899 | } | |
a8698707 | 900 | mutex_unlock(&disk->open_mutex); |
5b56b6ed CH |
901 | |
902 | if (unblock_events) | |
903 | disk_unblock_events(disk); | |
a56aefca CB |
904 | |
905 | bdev_file->f_flags |= O_LARGEFILE; | |
906 | bdev_file->f_mode |= FMODE_BUF_RASYNC | FMODE_CAN_ODIRECT; | |
907 | if (bdev_nowait(bdev)) | |
908 | bdev_file->f_mode |= FMODE_NOWAIT; | |
ddd65e19 CB |
909 | if (mode & BLK_OPEN_RESTRICT_WRITES) |
910 | bdev_file->f_mode |= FMODE_WRITE_RESTRICTED; | |
7c09a4ed | 911 | bdev_file->f_mapping = bdev->bd_inode->i_mapping; |
a56aefca | 912 | bdev_file->f_wb_err = filemap_sample_wb_err(bdev_file->f_mapping); |
ab838b3f | 913 | bdev_file->private_data = holder; |
a56aefca CB |
914 | |
915 | return 0; | |
efcf5932 ML |
916 | put_module: |
917 | module_put(disk->fops->owner); | |
22ae8ce8 | 918 | abort_claiming: |
2736e8ee | 919 | if (holder) |
37c3fc9a | 920 | bd_abort_claiming(bdev, holder); |
a8698707 | 921 | mutex_unlock(&disk->open_mutex); |
22ae8ce8 | 922 | disk_unblock_events(disk); |
a56aefca | 923 | return ret; |
37be4124 | 924 | } |
e719b4d1 | 925 | |
f3a60882 CB |
926 | /* |
927 | * If BLK_OPEN_WRITE_IOCTL is set then this is a historical quirk | |
928 | * associated with the floppy driver where it has allowed ioctls if the | |
929 | * file was opened for writing, but does not allow reads or writes. | |
930 | * Make sure that this quirk is reflected in @f_flags. | |
931 | * | |
932 | * It can also happen if a block device is opened as O_RDWR | O_WRONLY. | |
933 | */ | |
934 | static unsigned blk_to_file_flags(blk_mode_t mode) | |
935 | { | |
936 | unsigned int flags = 0; | |
937 | ||
938 | if ((mode & (BLK_OPEN_READ | BLK_OPEN_WRITE)) == | |
939 | (BLK_OPEN_READ | BLK_OPEN_WRITE)) | |
940 | flags |= O_RDWR; | |
941 | else if (mode & BLK_OPEN_WRITE_IOCTL) | |
942 | flags |= O_RDWR | O_WRONLY; | |
943 | else if (mode & BLK_OPEN_WRITE) | |
944 | flags |= O_WRONLY; | |
945 | else if (mode & BLK_OPEN_READ) | |
946 | flags |= O_RDONLY; /* homeopathic, because O_RDONLY is 0 */ | |
947 | else | |
948 | WARN_ON_ONCE(true); | |
949 | ||
950 | if (mode & BLK_OPEN_NDELAY) | |
951 | flags |= O_NDELAY; | |
952 | ||
953 | return flags; | |
954 | } | |
955 | ||
956 | struct file *bdev_file_open_by_dev(dev_t dev, blk_mode_t mode, void *holder, | |
957 | const struct blk_holder_ops *hops) | |
958 | { | |
959 | struct file *bdev_file; | |
a56aefca | 960 | struct block_device *bdev; |
f3a60882 | 961 | unsigned int flags; |
a56aefca | 962 | int ret; |
f3a60882 | 963 | |
a56aefca CB |
964 | ret = bdev_permission(dev, mode, holder); |
965 | if (ret) | |
966 | return ERR_PTR(ret); | |
967 | ||
968 | bdev = blkdev_get_no_open(dev); | |
969 | if (!bdev) | |
970 | return ERR_PTR(-ENXIO); | |
f3a60882 CB |
971 | |
972 | flags = blk_to_file_flags(mode); | |
a56aefca | 973 | bdev_file = alloc_file_pseudo_noaccount(bdev->bd_inode, |
f3a60882 CB |
974 | blockdev_mnt, "", flags | O_LARGEFILE, &def_blk_fops); |
975 | if (IS_ERR(bdev_file)) { | |
a56aefca | 976 | blkdev_put_no_open(bdev); |
f3a60882 CB |
977 | return bdev_file; |
978 | } | |
a56aefca | 979 | ihold(bdev->bd_inode); |
f3a60882 | 980 | |
a56aefca CB |
981 | ret = bdev_open(bdev, mode, holder, hops, bdev_file); |
982 | if (ret) { | |
ab838b3f CB |
983 | /* We failed to open the block device. Let ->release() know. */ |
984 | bdev_file->private_data = ERR_PTR(ret); | |
a56aefca CB |
985 | fput(bdev_file); |
986 | return ERR_PTR(ret); | |
987 | } | |
f3a60882 CB |
988 | return bdev_file; |
989 | } | |
990 | EXPORT_SYMBOL(bdev_file_open_by_dev); | |
991 | ||
992 | struct file *bdev_file_open_by_path(const char *path, blk_mode_t mode, | |
993 | void *holder, | |
994 | const struct blk_holder_ops *hops) | |
995 | { | |
7c09a4ed | 996 | struct file *file; |
f3a60882 CB |
997 | dev_t dev; |
998 | int error; | |
999 | ||
1000 | error = lookup_bdev(path, &dev); | |
1001 | if (error) | |
1002 | return ERR_PTR(error); | |
1003 | ||
7c09a4ed CB |
1004 | file = bdev_file_open_by_dev(dev, mode, holder, hops); |
1005 | if (!IS_ERR(file) && (mode & BLK_OPEN_WRITE)) { | |
1006 | if (bdev_read_only(file_bdev(file))) { | |
1007 | fput(file); | |
1008 | file = ERR_PTR(-EACCES); | |
f3a60882 CB |
1009 | } |
1010 | } | |
1011 | ||
7c09a4ed | 1012 | return file; |
f3a60882 CB |
1013 | } |
1014 | EXPORT_SYMBOL(bdev_file_open_by_path); | |
1015 | ||
22650a99 CB |
1016 | static inline void bd_yield_claim(struct file *bdev_file) |
1017 | { | |
1018 | struct block_device *bdev = file_bdev(bdev_file); | |
1019 | void *holder = bdev_file->private_data; | |
1020 | ||
1021 | lockdep_assert_held(&bdev->bd_disk->open_mutex); | |
1022 | ||
1023 | if (WARN_ON_ONCE(IS_ERR_OR_NULL(holder))) | |
1024 | return; | |
1025 | ||
1026 | if (!bdev_unclaimed(bdev_file)) | |
1027 | bd_end_claim(bdev, holder); | |
1028 | } | |
1029 | ||
7c09a4ed | 1030 | void bdev_release(struct file *bdev_file) |
2e7b651d | 1031 | { |
7c09a4ed | 1032 | struct block_device *bdev = file_bdev(bdev_file); |
ab838b3f | 1033 | void *holder = bdev_file->private_data; |
2e7b651d PZ |
1034 | struct gendisk *disk = bdev->bd_disk; |
1035 | ||
ab838b3f CB |
1036 | /* We failed to open that block device. */ |
1037 | if (IS_ERR(holder)) | |
1038 | goto put_no_open; | |
1039 | ||
b849dd84 DA |
1040 | /* |
1041 | * Sync early if it looks like we're the last one. If someone else | |
1042 | * opens the block device between now and the decrement of bd_openers | |
1043 | * then we did a sync that we didn't need to, but that's not the end | |
1044 | * of the world and we want to avoid long (could be several minute) | |
1045 | * syncs while holding the mutex. | |
1046 | */ | |
9acf381f | 1047 | if (atomic_read(&bdev->bd_openers) == 1) |
b849dd84 DA |
1048 | sync_blockdev(bdev); |
1049 | ||
a8698707 | 1050 | mutex_lock(&disk->open_mutex); |
ab838b3f | 1051 | bdev_yield_write_access(bdev_file); |
ed5cc702 | 1052 | |
ab838b3f | 1053 | if (holder) |
22650a99 | 1054 | bd_yield_claim(bdev_file); |
77ea887e | 1055 | |
85ef06d1 TH |
1056 | /* |
1057 | * Trigger event checking and tell drivers to flush MEDIA_CHANGE | |
1058 | * event. This is to ensure detection of media removal commanded | |
1059 | * from userland - e.g. eject(1). | |
1060 | */ | |
5b56b6ed | 1061 | disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE); |
85ef06d1 | 1062 | |
c8276b95 | 1063 | if (bdev_is_partition(bdev)) |
ae220766 | 1064 | blkdev_put_part(bdev); |
c8276b95 | 1065 | else |
ae220766 | 1066 | blkdev_put_whole(bdev); |
a8698707 CH |
1067 | mutex_unlock(&disk->open_mutex); |
1068 | ||
efcf5932 | 1069 | module_put(disk->fops->owner); |
ab838b3f | 1070 | put_no_open: |
22ae8ce8 | 1071 | blkdev_put_no_open(bdev); |
e719b4d1 | 1072 | } |
e719b4d1 | 1073 | |
22650a99 CB |
1074 | /** |
1075 | * bdev_fput - yield claim to the block device and put the file | |
1076 | * @bdev_file: open block device | |
1077 | * | |
1078 | * Yield claim on the block device and put the file. Ensure that the | |
1079 | * block device can be reclaimed before the file is closed which is a | |
1080 | * deferred operation. | |
1081 | */ | |
1082 | void bdev_fput(struct file *bdev_file) | |
1083 | { | |
1084 | if (WARN_ON_ONCE(bdev_file->f_op != &def_blk_fops)) | |
1085 | return; | |
1086 | ||
1087 | if (bdev_file->private_data) { | |
1088 | struct block_device *bdev = file_bdev(bdev_file); | |
1089 | struct gendisk *disk = bdev->bd_disk; | |
1090 | ||
1091 | mutex_lock(&disk->open_mutex); | |
1092 | bdev_yield_write_access(bdev_file); | |
1093 | bd_yield_claim(bdev_file); | |
1094 | /* | |
1095 | * Tell release we already gave up our hold on the | |
1096 | * device and if write restrictions are available that | |
1097 | * we already gave up write access to the device. | |
1098 | */ | |
1099 | bdev_file->private_data = BDEV_I(bdev_file->f_mapping->host); | |
1100 | mutex_unlock(&disk->open_mutex); | |
1101 | } | |
1102 | ||
1103 | fput(bdev_file); | |
1104 | } | |
1105 | EXPORT_SYMBOL(bdev_fput); | |
1106 | ||
1da177e4 | 1107 | /** |
0ba4566c MWO |
1108 | * lookup_bdev() - Look up a struct block_device by name. |
1109 | * @pathname: Name of the block device in the filesystem. | |
1110 | * @dev: Pointer to the block device's dev_t, if found. | |
1da177e4 | 1111 | * |
057178cf | 1112 | * Lookup the block device's dev_t at @pathname in the current |
0ba4566c | 1113 | * namespace if possible and return it in @dev. |
057178cf | 1114 | * |
0ba4566c MWO |
1115 | * Context: May sleep. |
1116 | * Return: 0 if succeeded, negative errno otherwise. | |
1da177e4 | 1117 | */ |
4e7b5671 | 1118 | int lookup_bdev(const char *pathname, dev_t *dev) |
1da177e4 | 1119 | { |
1da177e4 | 1120 | struct inode *inode; |
421748ec | 1121 | struct path path; |
1da177e4 LT |
1122 | int error; |
1123 | ||
421748ec | 1124 | if (!pathname || !*pathname) |
4e7b5671 | 1125 | return -EINVAL; |
1da177e4 | 1126 | |
421748ec | 1127 | error = kern_path(pathname, LOOKUP_FOLLOW, &path); |
1da177e4 | 1128 | if (error) |
4e7b5671 | 1129 | return error; |
1da177e4 | 1130 | |
bb668734 | 1131 | inode = d_backing_inode(path.dentry); |
1da177e4 LT |
1132 | error = -ENOTBLK; |
1133 | if (!S_ISBLK(inode->i_mode)) | |
4e7b5671 | 1134 | goto out_path_put; |
1da177e4 | 1135 | error = -EACCES; |
a2982cc9 | 1136 | if (!may_open_dev(&path)) |
4e7b5671 CH |
1137 | goto out_path_put; |
1138 | ||
1139 | *dev = inode->i_rdev; | |
1140 | error = 0; | |
1141 | out_path_put: | |
421748ec | 1142 | path_put(&path); |
4e7b5671 | 1143 | return error; |
1da177e4 | 1144 | } |
d5686b44 | 1145 | EXPORT_SYMBOL(lookup_bdev); |
1da177e4 | 1146 | |
560e20e4 CH |
1147 | /** |
1148 | * bdev_mark_dead - mark a block device as dead | |
1149 | * @bdev: block device to operate on | |
1150 | * @surprise: indicate a surprise removal | |
1151 | * | |
1152 | * Tell the file system that this devices or media is dead. If @surprise is set | |
1153 | * to %true the device or media is already gone, if not we are preparing for an | |
1154 | * orderly removal. | |
1155 | * | |
d8530de5 CH |
1156 | * This calls into the file system, which then typicall syncs out all dirty data |
1157 | * and writes back inodes and then invalidates any cached data in the inodes on | |
1158 | * the file system. In addition we also invalidate the block device mapping. | |
560e20e4 CH |
1159 | */ |
1160 | void bdev_mark_dead(struct block_device *bdev, bool surprise) | |
b71e8a4c | 1161 | { |
d8530de5 CH |
1162 | mutex_lock(&bdev->bd_holder_lock); |
1163 | if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead) | |
1164 | bdev->bd_holder_ops->mark_dead(bdev, surprise); | |
fd146410 JK |
1165 | else { |
1166 | mutex_unlock(&bdev->bd_holder_lock); | |
d8530de5 | 1167 | sync_blockdev(bdev); |
fd146410 | 1168 | } |
b71e8a4c | 1169 | |
f98393a6 | 1170 | invalidate_bdev(bdev); |
b71e8a4c | 1171 | } |
560e20e4 | 1172 | /* |
6e57236e CH |
1173 | * New drivers should not use this directly. There are some drivers however |
1174 | * that needs this for historical reasons. For example, the DASD driver has | |
1175 | * historically had a shutdown to offline mode that doesn't actually remove the | |
1176 | * gendisk that otherwise looks a lot like a safe device removal. | |
560e20e4 CH |
1177 | */ |
1178 | EXPORT_SYMBOL_GPL(bdev_mark_dead); | |
5c0d6b60 | 1179 | |
1e03a36b | 1180 | void sync_bdevs(bool wait) |
5c0d6b60 JK |
1181 | { |
1182 | struct inode *inode, *old_inode = NULL; | |
1183 | ||
74278da9 | 1184 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1185 | list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { |
1186 | struct address_space *mapping = inode->i_mapping; | |
af309226 | 1187 | struct block_device *bdev; |
5c0d6b60 JK |
1188 | |
1189 | spin_lock(&inode->i_lock); | |
1190 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || | |
1191 | mapping->nrpages == 0) { | |
1192 | spin_unlock(&inode->i_lock); | |
1193 | continue; | |
1194 | } | |
1195 | __iget(inode); | |
1196 | spin_unlock(&inode->i_lock); | |
74278da9 | 1197 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1198 | /* |
1199 | * We hold a reference to 'inode' so it couldn't have been | |
1200 | * removed from s_inodes list while we dropped the | |
74278da9 | 1201 | * s_inode_list_lock We cannot iput the inode now as we can |
5c0d6b60 | 1202 | * be holding the last reference and we cannot iput it under |
74278da9 | 1203 | * s_inode_list_lock. So we keep the reference and iput it |
5c0d6b60 JK |
1204 | * later. |
1205 | */ | |
1206 | iput(old_inode); | |
1207 | old_inode = inode; | |
af309226 | 1208 | bdev = I_BDEV(inode); |
5c0d6b60 | 1209 | |
a8698707 | 1210 | mutex_lock(&bdev->bd_disk->open_mutex); |
9acf381f | 1211 | if (!atomic_read(&bdev->bd_openers)) { |
1e03a36b CH |
1212 | ; /* skip */ |
1213 | } else if (wait) { | |
1214 | /* | |
1215 | * We keep the error status of individual mapping so | |
1216 | * that applications can catch the writeback error using | |
1217 | * fsync(2). See filemap_fdatawait_keep_errors() for | |
1218 | * details. | |
1219 | */ | |
1220 | filemap_fdatawait_keep_errors(inode->i_mapping); | |
1221 | } else { | |
1222 | filemap_fdatawrite(inode->i_mapping); | |
1223 | } | |
a8698707 | 1224 | mutex_unlock(&bdev->bd_disk->open_mutex); |
5c0d6b60 | 1225 | |
74278da9 | 1226 | spin_lock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 | 1227 | } |
74278da9 | 1228 | spin_unlock(&blockdev_superblock->s_inode_list_lock); |
5c0d6b60 JK |
1229 | iput(old_inode); |
1230 | } | |
2d985f8c EB |
1231 | |
1232 | /* | |
1233 | * Handle STATX_DIOALIGN for block devices. | |
1234 | * | |
1235 | * Note that the inode passed to this is the inode of a block device node file, | |
1236 | * not the block device's internal inode. Therefore it is *not* valid to use | |
1237 | * I_BDEV() here; the block device has to be looked up by i_rdev instead. | |
1238 | */ | |
1239 | void bdev_statx_dioalign(struct inode *inode, struct kstat *stat) | |
1240 | { | |
1241 | struct block_device *bdev; | |
1242 | ||
1243 | bdev = blkdev_get_no_open(inode->i_rdev); | |
1244 | if (!bdev) | |
1245 | return; | |
1246 | ||
1247 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; | |
1248 | stat->dio_offset_align = bdev_logical_block_size(bdev); | |
1249 | stat->result_mask |= STATX_DIOALIGN; | |
1250 | ||
1251 | blkdev_put_no_open(bdev); | |
1252 | } | |
ed5cc702 JK |
1253 | |
1254 | static int __init setup_bdev_allow_write_mounted(char *str) | |
1255 | { | |
1256 | if (kstrtobool(str, &bdev_allow_write_mounted)) | |
1257 | pr_warn("Invalid option string for bdev_allow_write_mounted:" | |
1258 | " '%s'\n", str); | |
1259 | return 1; | |
1260 | } | |
1261 | __setup("bdev_allow_write_mounted=", setup_bdev_allow_write_mounted); |