Commit | Line | Data |
---|---|---|
9db5579b NP |
1 | /* |
2 | * Ram backed block device driver. | |
3 | * | |
4 | * Copyright (C) 2007 Nick Piggin | |
5 | * Copyright (C) 2007 Novell Inc. | |
6 | * | |
7 | * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright | |
8 | * of their respective owners. | |
9 | */ | |
10 | ||
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
13 | #include <linux/moduleparam.h> | |
14 | #include <linux/major.h> | |
15 | #include <linux/blkdev.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/highmem.h> | |
2a48fc0a | 18 | #include <linux/mutex.h> |
9db5579b | 19 | #include <linux/radix-tree.h> |
ff01bb48 | 20 | #include <linux/fs.h> |
5a0e3ad6 | 21 | #include <linux/slab.h> |
34c0fd54 DW |
22 | #ifdef CONFIG_BLK_DEV_RAM_DAX |
23 | #include <linux/pfn_t.h> | |
24 | #endif | |
9db5579b NP |
25 | |
26 | #include <asm/uaccess.h> | |
27 | ||
28 | #define SECTOR_SHIFT 9 | |
29 | #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT) | |
30 | #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT) | |
31 | ||
32 | /* | |
33 | * Each block ramdisk device has a radix_tree brd_pages of pages that stores | |
34 | * the pages containing the block device's contents. A brd page's ->index is | |
35 | * its offset in PAGE_SIZE units. This is similar to, but in no way connected | |
36 | * with, the kernel's pagecache or buffer cache (which sit above our block | |
37 | * device). | |
38 | */ | |
39 | struct brd_device { | |
40 | int brd_number; | |
9db5579b NP |
41 | |
42 | struct request_queue *brd_queue; | |
43 | struct gendisk *brd_disk; | |
44 | struct list_head brd_list; | |
45 | ||
46 | /* | |
47 | * Backing store of pages and lock to protect it. This is the contents | |
48 | * of the block device. | |
49 | */ | |
50 | spinlock_t brd_lock; | |
51 | struct radix_tree_root brd_pages; | |
52 | }; | |
53 | ||
54 | /* | |
55 | * Look up and return a brd's page for a given sector. | |
56 | */ | |
2a48fc0a | 57 | static DEFINE_MUTEX(brd_mutex); |
9db5579b NP |
58 | static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector) |
59 | { | |
60 | pgoff_t idx; | |
61 | struct page *page; | |
62 | ||
63 | /* | |
64 | * The page lifetime is protected by the fact that we have opened the | |
65 | * device node -- brd pages will never be deleted under us, so we | |
66 | * don't need any further locking or refcounting. | |
67 | * | |
68 | * This is strictly true for the radix-tree nodes as well (ie. we | |
69 | * don't actually need the rcu_read_lock()), however that is not a | |
70 | * documented feature of the radix-tree API so it is better to be | |
71 | * safe here (we don't have total exclusion from radix tree updates | |
72 | * here, only deletes). | |
73 | */ | |
74 | rcu_read_lock(); | |
75 | idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */ | |
76 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
77 | rcu_read_unlock(); | |
78 | ||
79 | BUG_ON(page && page->index != idx); | |
80 | ||
81 | return page; | |
82 | } | |
83 | ||
84 | /* | |
85 | * Look up and return a brd's page for a given sector. | |
86 | * If one does not exist, allocate an empty page, and insert that. Then | |
87 | * return it. | |
88 | */ | |
89 | static struct page *brd_insert_page(struct brd_device *brd, sector_t sector) | |
90 | { | |
91 | pgoff_t idx; | |
92 | struct page *page; | |
75acb9cd | 93 | gfp_t gfp_flags; |
9db5579b NP |
94 | |
95 | page = brd_lookup_page(brd, sector); | |
96 | if (page) | |
97 | return page; | |
98 | ||
99 | /* | |
100 | * Must use NOIO because we don't want to recurse back into the | |
101 | * block or filesystem layers from page reclaim. | |
75acb9cd | 102 | * |
a7a97fc9 MW |
103 | * Cannot support DAX and highmem, because our ->direct_access |
104 | * routine for DAX must return memory that is always addressable. | |
105 | * If DAX was reworked to use pfns and kmap throughout, this | |
75acb9cd | 106 | * restriction might be able to be lifted. |
9db5579b | 107 | */ |
75acb9cd | 108 | gfp_flags = GFP_NOIO | __GFP_ZERO; |
a7a97fc9 | 109 | #ifndef CONFIG_BLK_DEV_RAM_DAX |
75acb9cd NP |
110 | gfp_flags |= __GFP_HIGHMEM; |
111 | #endif | |
26defe34 | 112 | page = alloc_page(gfp_flags); |
9db5579b NP |
113 | if (!page) |
114 | return NULL; | |
115 | ||
116 | if (radix_tree_preload(GFP_NOIO)) { | |
117 | __free_page(page); | |
118 | return NULL; | |
119 | } | |
120 | ||
121 | spin_lock(&brd->brd_lock); | |
122 | idx = sector >> PAGE_SECTORS_SHIFT; | |
dfd20b2b | 123 | page->index = idx; |
9db5579b NP |
124 | if (radix_tree_insert(&brd->brd_pages, idx, page)) { |
125 | __free_page(page); | |
126 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
127 | BUG_ON(!page); | |
128 | BUG_ON(page->index != idx); | |
dfd20b2b | 129 | } |
9db5579b NP |
130 | spin_unlock(&brd->brd_lock); |
131 | ||
132 | radix_tree_preload_end(); | |
133 | ||
134 | return page; | |
135 | } | |
136 | ||
b7c33571 NP |
137 | static void brd_free_page(struct brd_device *brd, sector_t sector) |
138 | { | |
139 | struct page *page; | |
140 | pgoff_t idx; | |
141 | ||
142 | spin_lock(&brd->brd_lock); | |
143 | idx = sector >> PAGE_SECTORS_SHIFT; | |
144 | page = radix_tree_delete(&brd->brd_pages, idx); | |
145 | spin_unlock(&brd->brd_lock); | |
146 | if (page) | |
147 | __free_page(page); | |
148 | } | |
149 | ||
150 | static void brd_zero_page(struct brd_device *brd, sector_t sector) | |
151 | { | |
152 | struct page *page; | |
153 | ||
154 | page = brd_lookup_page(brd, sector); | |
155 | if (page) | |
156 | clear_highpage(page); | |
157 | } | |
158 | ||
9db5579b NP |
159 | /* |
160 | * Free all backing store pages and radix tree. This must only be called when | |
161 | * there are no other users of the device. | |
162 | */ | |
163 | #define FREE_BATCH 16 | |
164 | static void brd_free_pages(struct brd_device *brd) | |
165 | { | |
166 | unsigned long pos = 0; | |
167 | struct page *pages[FREE_BATCH]; | |
168 | int nr_pages; | |
169 | ||
170 | do { | |
171 | int i; | |
172 | ||
173 | nr_pages = radix_tree_gang_lookup(&brd->brd_pages, | |
174 | (void **)pages, pos, FREE_BATCH); | |
175 | ||
176 | for (i = 0; i < nr_pages; i++) { | |
177 | void *ret; | |
178 | ||
179 | BUG_ON(pages[i]->index < pos); | |
180 | pos = pages[i]->index; | |
181 | ret = radix_tree_delete(&brd->brd_pages, pos); | |
182 | BUG_ON(!ret || ret != pages[i]); | |
183 | __free_page(pages[i]); | |
184 | } | |
185 | ||
186 | pos++; | |
187 | ||
188 | /* | |
189 | * This assumes radix_tree_gang_lookup always returns as | |
190 | * many pages as possible. If the radix-tree code changes, | |
191 | * so will this have to. | |
192 | */ | |
193 | } while (nr_pages == FREE_BATCH); | |
194 | } | |
195 | ||
196 | /* | |
197 | * copy_to_brd_setup must be called before copy_to_brd. It may sleep. | |
198 | */ | |
199 | static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n) | |
200 | { | |
201 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
202 | size_t copy; | |
203 | ||
204 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
205 | if (!brd_insert_page(brd, sector)) | |
96f8d8e0 | 206 | return -ENOSPC; |
9db5579b NP |
207 | if (copy < n) { |
208 | sector += copy >> SECTOR_SHIFT; | |
209 | if (!brd_insert_page(brd, sector)) | |
96f8d8e0 | 210 | return -ENOSPC; |
9db5579b NP |
211 | } |
212 | return 0; | |
213 | } | |
214 | ||
b7c33571 NP |
215 | static void discard_from_brd(struct brd_device *brd, |
216 | sector_t sector, size_t n) | |
217 | { | |
218 | while (n >= PAGE_SIZE) { | |
219 | /* | |
220 | * Don't want to actually discard pages here because | |
221 | * re-allocating the pages can result in writeback | |
222 | * deadlocks under heavy load. | |
223 | */ | |
224 | if (0) | |
225 | brd_free_page(brd, sector); | |
226 | else | |
227 | brd_zero_page(brd, sector); | |
228 | sector += PAGE_SIZE >> SECTOR_SHIFT; | |
229 | n -= PAGE_SIZE; | |
230 | } | |
231 | } | |
232 | ||
9db5579b NP |
233 | /* |
234 | * Copy n bytes from src to the brd starting at sector. Does not sleep. | |
235 | */ | |
236 | static void copy_to_brd(struct brd_device *brd, const void *src, | |
237 | sector_t sector, size_t n) | |
238 | { | |
239 | struct page *page; | |
240 | void *dst; | |
241 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
242 | size_t copy; | |
243 | ||
244 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
245 | page = brd_lookup_page(brd, sector); | |
246 | BUG_ON(!page); | |
247 | ||
cfd8005c | 248 | dst = kmap_atomic(page); |
9db5579b | 249 | memcpy(dst + offset, src, copy); |
cfd8005c | 250 | kunmap_atomic(dst); |
9db5579b NP |
251 | |
252 | if (copy < n) { | |
253 | src += copy; | |
254 | sector += copy >> SECTOR_SHIFT; | |
255 | copy = n - copy; | |
256 | page = brd_lookup_page(brd, sector); | |
257 | BUG_ON(!page); | |
258 | ||
cfd8005c | 259 | dst = kmap_atomic(page); |
9db5579b | 260 | memcpy(dst, src, copy); |
cfd8005c | 261 | kunmap_atomic(dst); |
9db5579b NP |
262 | } |
263 | } | |
264 | ||
265 | /* | |
266 | * Copy n bytes to dst from the brd starting at sector. Does not sleep. | |
267 | */ | |
268 | static void copy_from_brd(void *dst, struct brd_device *brd, | |
269 | sector_t sector, size_t n) | |
270 | { | |
271 | struct page *page; | |
272 | void *src; | |
273 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
274 | size_t copy; | |
275 | ||
276 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
277 | page = brd_lookup_page(brd, sector); | |
278 | if (page) { | |
cfd8005c | 279 | src = kmap_atomic(page); |
9db5579b | 280 | memcpy(dst, src + offset, copy); |
cfd8005c | 281 | kunmap_atomic(src); |
9db5579b NP |
282 | } else |
283 | memset(dst, 0, copy); | |
284 | ||
285 | if (copy < n) { | |
286 | dst += copy; | |
287 | sector += copy >> SECTOR_SHIFT; | |
288 | copy = n - copy; | |
289 | page = brd_lookup_page(brd, sector); | |
290 | if (page) { | |
cfd8005c | 291 | src = kmap_atomic(page); |
9db5579b | 292 | memcpy(dst, src, copy); |
cfd8005c | 293 | kunmap_atomic(src); |
9db5579b NP |
294 | } else |
295 | memset(dst, 0, copy); | |
296 | } | |
297 | } | |
298 | ||
299 | /* | |
300 | * Process a single bvec of a bio. | |
301 | */ | |
302 | static int brd_do_bvec(struct brd_device *brd, struct page *page, | |
303 | unsigned int len, unsigned int off, int rw, | |
304 | sector_t sector) | |
305 | { | |
306 | void *mem; | |
307 | int err = 0; | |
308 | ||
309 | if (rw != READ) { | |
310 | err = copy_to_brd_setup(brd, sector, len); | |
311 | if (err) | |
312 | goto out; | |
313 | } | |
314 | ||
cfd8005c | 315 | mem = kmap_atomic(page); |
9db5579b NP |
316 | if (rw == READ) { |
317 | copy_from_brd(mem + off, brd, sector, len); | |
318 | flush_dcache_page(page); | |
c2572f2b NP |
319 | } else { |
320 | flush_dcache_page(page); | |
9db5579b | 321 | copy_to_brd(brd, mem + off, sector, len); |
c2572f2b | 322 | } |
cfd8005c | 323 | kunmap_atomic(mem); |
9db5579b NP |
324 | |
325 | out: | |
326 | return err; | |
327 | } | |
328 | ||
dece1635 | 329 | static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio) |
9db5579b NP |
330 | { |
331 | struct block_device *bdev = bio->bi_bdev; | |
332 | struct brd_device *brd = bdev->bd_disk->private_data; | |
333 | int rw; | |
7988613b | 334 | struct bio_vec bvec; |
9db5579b | 335 | sector_t sector; |
7988613b | 336 | struct bvec_iter iter; |
9db5579b | 337 | |
4f024f37 | 338 | sector = bio->bi_iter.bi_sector; |
f73a1c7d | 339 | if (bio_end_sector(bio) > get_capacity(bdev->bd_disk)) |
4246a0b6 | 340 | goto io_error; |
9db5579b | 341 | |
95fe6c1a | 342 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
2dbe5495 | 343 | if (sector & ((PAGE_SIZE >> SECTOR_SHIFT) - 1) || |
5e4298be | 344 | bio->bi_iter.bi_size & ~PAGE_MASK) |
2dbe5495 | 345 | goto io_error; |
4f024f37 | 346 | discard_from_brd(brd, sector, bio->bi_iter.bi_size); |
b7c33571 NP |
347 | goto out; |
348 | } | |
349 | ||
70246286 | 350 | rw = bio_data_dir(bio); |
9db5579b | 351 | |
7988613b KO |
352 | bio_for_each_segment(bvec, bio, iter) { |
353 | unsigned int len = bvec.bv_len; | |
4246a0b6 CH |
354 | int err; |
355 | ||
7988613b KO |
356 | err = brd_do_bvec(brd, bvec.bv_page, len, |
357 | bvec.bv_offset, rw, sector); | |
9db5579b | 358 | if (err) |
4246a0b6 | 359 | goto io_error; |
9db5579b NP |
360 | sector += len >> SECTOR_SHIFT; |
361 | } | |
362 | ||
363 | out: | |
4246a0b6 | 364 | bio_endio(bio); |
dece1635 | 365 | return BLK_QC_T_NONE; |
4246a0b6 CH |
366 | io_error: |
367 | bio_io_error(bio); | |
dece1635 | 368 | return BLK_QC_T_NONE; |
9db5579b NP |
369 | } |
370 | ||
a72132c3 MW |
371 | static int brd_rw_page(struct block_device *bdev, sector_t sector, |
372 | struct page *page, int rw) | |
373 | { | |
374 | struct brd_device *brd = bdev->bd_disk->private_data; | |
09cbfeaf | 375 | int err = brd_do_bvec(brd, page, PAGE_SIZE, 0, rw, sector); |
a72132c3 MW |
376 | page_endio(page, rw & WRITE, err); |
377 | return err; | |
378 | } | |
379 | ||
a7a97fc9 | 380 | #ifdef CONFIG_BLK_DEV_RAM_DAX |
dd22f551 | 381 | static long brd_direct_access(struct block_device *bdev, sector_t sector, |
0a70bd43 | 382 | void __pmem **kaddr, pfn_t *pfn, long size) |
75acb9cd NP |
383 | { |
384 | struct brd_device *brd = bdev->bd_disk->private_data; | |
385 | struct page *page; | |
386 | ||
387 | if (!brd) | |
388 | return -ENODEV; | |
75acb9cd NP |
389 | page = brd_insert_page(brd, sector); |
390 | if (!page) | |
96f8d8e0 | 391 | return -ENOSPC; |
e2e05394 | 392 | *kaddr = (void __pmem *)page_address(page); |
34c0fd54 | 393 | *pfn = page_to_pfn_t(page); |
75acb9cd | 394 | |
dd22f551 | 395 | return PAGE_SIZE; |
75acb9cd | 396 | } |
a7a97fc9 MW |
397 | #else |
398 | #define brd_direct_access NULL | |
75acb9cd NP |
399 | #endif |
400 | ||
2b9ecd03 | 401 | static int brd_ioctl(struct block_device *bdev, fmode_t mode, |
9db5579b NP |
402 | unsigned int cmd, unsigned long arg) |
403 | { | |
404 | int error; | |
9db5579b NP |
405 | struct brd_device *brd = bdev->bd_disk->private_data; |
406 | ||
407 | if (cmd != BLKFLSBUF) | |
408 | return -ENOTTY; | |
409 | ||
410 | /* | |
411 | * ram device BLKFLSBUF has special semantics, we want to actually | |
412 | * release and destroy the ramdisk data. | |
413 | */ | |
2a48fc0a | 414 | mutex_lock(&brd_mutex); |
9db5579b NP |
415 | mutex_lock(&bdev->bd_mutex); |
416 | error = -EBUSY; | |
417 | if (bdev->bd_openers <= 1) { | |
418 | /* | |
ff01bb48 AV |
419 | * Kill the cache first, so it isn't written back to the |
420 | * device. | |
9db5579b NP |
421 | * |
422 | * Another thread might instantiate more buffercache here, | |
423 | * but there is not much we can do to close that race. | |
424 | */ | |
ff01bb48 | 425 | kill_bdev(bdev); |
9db5579b NP |
426 | brd_free_pages(brd); |
427 | error = 0; | |
428 | } | |
429 | mutex_unlock(&bdev->bd_mutex); | |
2a48fc0a | 430 | mutex_unlock(&brd_mutex); |
9db5579b NP |
431 | |
432 | return error; | |
433 | } | |
434 | ||
83d5cde4 | 435 | static const struct block_device_operations brd_fops = { |
75acb9cd | 436 | .owner = THIS_MODULE, |
a72132c3 | 437 | .rw_page = brd_rw_page, |
8a6cfeb6 | 438 | .ioctl = brd_ioctl, |
75acb9cd | 439 | .direct_access = brd_direct_access, |
9db5579b NP |
440 | }; |
441 | ||
442 | /* | |
443 | * And now the modules code and kernel interface. | |
444 | */ | |
937af5ec | 445 | static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT; |
8892cbaf | 446 | module_param(rd_nr, int, S_IRUGO); |
9db5579b | 447 | MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices"); |
937af5ec BH |
448 | |
449 | int rd_size = CONFIG_BLK_DEV_RAM_SIZE; | |
8892cbaf | 450 | module_param(rd_size, int, S_IRUGO); |
9db5579b | 451 | MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes."); |
937af5ec BH |
452 | |
453 | static int max_part = 1; | |
8892cbaf | 454 | module_param(max_part, int, S_IRUGO); |
937af5ec BH |
455 | MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices"); |
456 | ||
9db5579b NP |
457 | MODULE_LICENSE("GPL"); |
458 | MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR); | |
efedf51c | 459 | MODULE_ALIAS("rd"); |
9db5579b NP |
460 | |
461 | #ifndef MODULE | |
462 | /* Legacy boot options - nonmodular */ | |
463 | static int __init ramdisk_size(char *str) | |
464 | { | |
465 | rd_size = simple_strtol(str, NULL, 0); | |
466 | return 1; | |
467 | } | |
1adbee50 | 468 | __setup("ramdisk_size=", ramdisk_size); |
9db5579b NP |
469 | #endif |
470 | ||
471 | /* | |
472 | * The device scheme is derived from loop.c. Keep them in synch where possible | |
473 | * (should share code eventually). | |
474 | */ | |
475 | static LIST_HEAD(brd_devices); | |
476 | static DEFINE_MUTEX(brd_devices_mutex); | |
477 | ||
478 | static struct brd_device *brd_alloc(int i) | |
479 | { | |
480 | struct brd_device *brd; | |
481 | struct gendisk *disk; | |
482 | ||
483 | brd = kzalloc(sizeof(*brd), GFP_KERNEL); | |
484 | if (!brd) | |
485 | goto out; | |
486 | brd->brd_number = i; | |
487 | spin_lock_init(&brd->brd_lock); | |
488 | INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC); | |
489 | ||
490 | brd->brd_queue = blk_alloc_queue(GFP_KERNEL); | |
491 | if (!brd->brd_queue) | |
492 | goto out_free_dev; | |
c8fa3173 | 493 | |
9db5579b | 494 | blk_queue_make_request(brd->brd_queue, brd_make_request); |
086fa5ff | 495 | blk_queue_max_hw_sectors(brd->brd_queue, 1024); |
9db5579b NP |
496 | blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY); |
497 | ||
c8fa3173 BH |
498 | /* This is so fdisk will align partitions on 4k, because of |
499 | * direct_access API needing 4k alignment, returning a PFN | |
500 | * (This is only a problem on very small devices <= 4M, | |
501 | * otherwise fdisk will align on 1M. Regardless this call | |
502 | * is harmless) | |
503 | */ | |
504 | blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE); | |
505 | ||
b7c33571 | 506 | brd->brd_queue->limits.discard_granularity = PAGE_SIZE; |
2bb4cd5c | 507 | blk_queue_max_discard_sectors(brd->brd_queue, UINT_MAX); |
b7c33571 NP |
508 | brd->brd_queue->limits.discard_zeroes_data = 1; |
509 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue); | |
510 | ||
937af5ec | 511 | disk = brd->brd_disk = alloc_disk(max_part); |
9db5579b NP |
512 | if (!disk) |
513 | goto out_free_queue; | |
514 | disk->major = RAMDISK_MAJOR; | |
937af5ec | 515 | disk->first_minor = i * max_part; |
9db5579b NP |
516 | disk->fops = &brd_fops; |
517 | disk->private_data = brd; | |
518 | disk->queue = brd->brd_queue; | |
937af5ec | 519 | disk->flags = GENHD_FL_EXT_DEVT; |
9db5579b NP |
520 | sprintf(disk->disk_name, "ram%d", i); |
521 | set_capacity(disk, rd_size * 2); | |
522 | ||
523 | return brd; | |
524 | ||
525 | out_free_queue: | |
526 | blk_cleanup_queue(brd->brd_queue); | |
527 | out_free_dev: | |
528 | kfree(brd); | |
529 | out: | |
530 | return NULL; | |
531 | } | |
532 | ||
533 | static void brd_free(struct brd_device *brd) | |
534 | { | |
535 | put_disk(brd->brd_disk); | |
536 | blk_cleanup_queue(brd->brd_queue); | |
537 | brd_free_pages(brd); | |
538 | kfree(brd); | |
539 | } | |
540 | ||
937af5ec | 541 | static struct brd_device *brd_init_one(int i, bool *new) |
9db5579b NP |
542 | { |
543 | struct brd_device *brd; | |
544 | ||
937af5ec | 545 | *new = false; |
9db5579b NP |
546 | list_for_each_entry(brd, &brd_devices, brd_list) { |
547 | if (brd->brd_number == i) | |
548 | goto out; | |
549 | } | |
550 | ||
551 | brd = brd_alloc(i); | |
552 | if (brd) { | |
553 | add_disk(brd->brd_disk); | |
554 | list_add_tail(&brd->brd_list, &brd_devices); | |
555 | } | |
937af5ec | 556 | *new = true; |
9db5579b NP |
557 | out: |
558 | return brd; | |
559 | } | |
560 | ||
561 | static void brd_del_one(struct brd_device *brd) | |
562 | { | |
563 | list_del(&brd->brd_list); | |
564 | del_gendisk(brd->brd_disk); | |
565 | brd_free(brd); | |
566 | } | |
567 | ||
568 | static struct kobject *brd_probe(dev_t dev, int *part, void *data) | |
569 | { | |
570 | struct brd_device *brd; | |
571 | struct kobject *kobj; | |
937af5ec | 572 | bool new; |
9db5579b NP |
573 | |
574 | mutex_lock(&brd_devices_mutex); | |
937af5ec | 575 | brd = brd_init_one(MINOR(dev) / max_part, &new); |
a207f593 | 576 | kobj = brd ? get_disk(brd->brd_disk) : NULL; |
9db5579b NP |
577 | mutex_unlock(&brd_devices_mutex); |
578 | ||
937af5ec BH |
579 | if (new) |
580 | *part = 0; | |
581 | ||
9db5579b NP |
582 | return kobj; |
583 | } | |
584 | ||
585 | static int __init brd_init(void) | |
586 | { | |
9db5579b | 587 | struct brd_device *brd, *next; |
937af5ec | 588 | int i; |
9db5579b NP |
589 | |
590 | /* | |
591 | * brd module now has a feature to instantiate underlying device | |
592 | * structure on-demand, provided that there is an access dev node. | |
9db5579b | 593 | * |
937af5ec BH |
594 | * (1) if rd_nr is specified, create that many upfront. else |
595 | * it defaults to CONFIG_BLK_DEV_RAM_COUNT | |
596 | * (2) User can further extend brd devices by create dev node themselves | |
597 | * and have kernel automatically instantiate actual device | |
598 | * on-demand. Example: | |
599 | * mknod /path/devnod_name b 1 X # 1 is the rd major | |
600 | * fdisk -l /path/devnod_name | |
601 | * If (X / max_part) was not already created it will be created | |
602 | * dynamically. | |
9db5579b | 603 | */ |
d7853d1f | 604 | |
9db5579b NP |
605 | if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) |
606 | return -EIO; | |
607 | ||
937af5ec BH |
608 | if (unlikely(!max_part)) |
609 | max_part = 1; | |
610 | ||
611 | for (i = 0; i < rd_nr; i++) { | |
9db5579b NP |
612 | brd = brd_alloc(i); |
613 | if (!brd) | |
614 | goto out_free; | |
615 | list_add_tail(&brd->brd_list, &brd_devices); | |
616 | } | |
617 | ||
618 | /* point of no return */ | |
619 | ||
620 | list_for_each_entry(brd, &brd_devices, brd_list) | |
621 | add_disk(brd->brd_disk); | |
622 | ||
937af5ec | 623 | blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS, |
9db5579b NP |
624 | THIS_MODULE, brd_probe, NULL, NULL); |
625 | ||
937af5ec | 626 | pr_info("brd: module loaded\n"); |
9db5579b NP |
627 | return 0; |
628 | ||
629 | out_free: | |
630 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) { | |
631 | list_del(&brd->brd_list); | |
632 | brd_free(brd); | |
633 | } | |
c82f2966 | 634 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
9db5579b | 635 | |
937af5ec | 636 | pr_info("brd: module NOT loaded !!!\n"); |
9db5579b NP |
637 | return -ENOMEM; |
638 | } | |
639 | ||
640 | static void __exit brd_exit(void) | |
641 | { | |
9db5579b NP |
642 | struct brd_device *brd, *next; |
643 | ||
9db5579b NP |
644 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) |
645 | brd_del_one(brd); | |
646 | ||
937af5ec | 647 | blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS); |
9db5579b | 648 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
937af5ec BH |
649 | |
650 | pr_info("brd: module unloaded\n"); | |
9db5579b NP |
651 | } |
652 | ||
653 | module_init(brd_init); | |
654 | module_exit(brd_exit); | |
655 |