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> | |
18 | #include <linux/gfp.h> | |
19 | #include <linux/radix-tree.h> | |
20 | #include <linux/buffer_head.h> /* invalidate_bh_lrus() */ | |
21 | ||
22 | #include <asm/uaccess.h> | |
23 | ||
24 | #define SECTOR_SHIFT 9 | |
25 | #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT) | |
26 | #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT) | |
27 | ||
28 | /* | |
29 | * Each block ramdisk device has a radix_tree brd_pages of pages that stores | |
30 | * the pages containing the block device's contents. A brd page's ->index is | |
31 | * its offset in PAGE_SIZE units. This is similar to, but in no way connected | |
32 | * with, the kernel's pagecache or buffer cache (which sit above our block | |
33 | * device). | |
34 | */ | |
35 | struct brd_device { | |
36 | int brd_number; | |
37 | int brd_refcnt; | |
38 | loff_t brd_offset; | |
39 | loff_t brd_sizelimit; | |
40 | unsigned brd_blocksize; | |
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 | */ | |
57 | static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector) | |
58 | { | |
59 | pgoff_t idx; | |
60 | struct page *page; | |
61 | ||
62 | /* | |
63 | * The page lifetime is protected by the fact that we have opened the | |
64 | * device node -- brd pages will never be deleted under us, so we | |
65 | * don't need any further locking or refcounting. | |
66 | * | |
67 | * This is strictly true for the radix-tree nodes as well (ie. we | |
68 | * don't actually need the rcu_read_lock()), however that is not a | |
69 | * documented feature of the radix-tree API so it is better to be | |
70 | * safe here (we don't have total exclusion from radix tree updates | |
71 | * here, only deletes). | |
72 | */ | |
73 | rcu_read_lock(); | |
74 | idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */ | |
75 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
76 | rcu_read_unlock(); | |
77 | ||
78 | BUG_ON(page && page->index != idx); | |
79 | ||
80 | return page; | |
81 | } | |
82 | ||
83 | /* | |
84 | * Look up and return a brd's page for a given sector. | |
85 | * If one does not exist, allocate an empty page, and insert that. Then | |
86 | * return it. | |
87 | */ | |
88 | static struct page *brd_insert_page(struct brd_device *brd, sector_t sector) | |
89 | { | |
90 | pgoff_t idx; | |
91 | struct page *page; | |
75acb9cd | 92 | gfp_t gfp_flags; |
9db5579b NP |
93 | |
94 | page = brd_lookup_page(brd, sector); | |
95 | if (page) | |
96 | return page; | |
97 | ||
98 | /* | |
99 | * Must use NOIO because we don't want to recurse back into the | |
100 | * block or filesystem layers from page reclaim. | |
75acb9cd NP |
101 | * |
102 | * Cannot support XIP and highmem, because our ->direct_access | |
103 | * routine for XIP must return memory that is always addressable. | |
104 | * If XIP was reworked to use pfns and kmap throughout, this | |
105 | * restriction might be able to be lifted. | |
9db5579b | 106 | */ |
75acb9cd NP |
107 | gfp_flags = GFP_NOIO | __GFP_ZERO; |
108 | #ifndef CONFIG_BLK_DEV_XIP | |
109 | gfp_flags |= __GFP_HIGHMEM; | |
110 | #endif | |
26defe34 | 111 | page = alloc_page(gfp_flags); |
9db5579b NP |
112 | if (!page) |
113 | return NULL; | |
114 | ||
115 | if (radix_tree_preload(GFP_NOIO)) { | |
116 | __free_page(page); | |
117 | return NULL; | |
118 | } | |
119 | ||
120 | spin_lock(&brd->brd_lock); | |
121 | idx = sector >> PAGE_SECTORS_SHIFT; | |
122 | if (radix_tree_insert(&brd->brd_pages, idx, page)) { | |
123 | __free_page(page); | |
124 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
125 | BUG_ON(!page); | |
126 | BUG_ON(page->index != idx); | |
127 | } else | |
128 | page->index = idx; | |
129 | spin_unlock(&brd->brd_lock); | |
130 | ||
131 | radix_tree_preload_end(); | |
132 | ||
133 | return page; | |
134 | } | |
135 | ||
136 | /* | |
137 | * Free all backing store pages and radix tree. This must only be called when | |
138 | * there are no other users of the device. | |
139 | */ | |
140 | #define FREE_BATCH 16 | |
141 | static void brd_free_pages(struct brd_device *brd) | |
142 | { | |
143 | unsigned long pos = 0; | |
144 | struct page *pages[FREE_BATCH]; | |
145 | int nr_pages; | |
146 | ||
147 | do { | |
148 | int i; | |
149 | ||
150 | nr_pages = radix_tree_gang_lookup(&brd->brd_pages, | |
151 | (void **)pages, pos, FREE_BATCH); | |
152 | ||
153 | for (i = 0; i < nr_pages; i++) { | |
154 | void *ret; | |
155 | ||
156 | BUG_ON(pages[i]->index < pos); | |
157 | pos = pages[i]->index; | |
158 | ret = radix_tree_delete(&brd->brd_pages, pos); | |
159 | BUG_ON(!ret || ret != pages[i]); | |
160 | __free_page(pages[i]); | |
161 | } | |
162 | ||
163 | pos++; | |
164 | ||
165 | /* | |
166 | * This assumes radix_tree_gang_lookup always returns as | |
167 | * many pages as possible. If the radix-tree code changes, | |
168 | * so will this have to. | |
169 | */ | |
170 | } while (nr_pages == FREE_BATCH); | |
171 | } | |
172 | ||
173 | /* | |
174 | * copy_to_brd_setup must be called before copy_to_brd. It may sleep. | |
175 | */ | |
176 | static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n) | |
177 | { | |
178 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
179 | size_t copy; | |
180 | ||
181 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
182 | if (!brd_insert_page(brd, sector)) | |
183 | return -ENOMEM; | |
184 | if (copy < n) { | |
185 | sector += copy >> SECTOR_SHIFT; | |
186 | if (!brd_insert_page(brd, sector)) | |
187 | return -ENOMEM; | |
188 | } | |
189 | return 0; | |
190 | } | |
191 | ||
192 | /* | |
193 | * Copy n bytes from src to the brd starting at sector. Does not sleep. | |
194 | */ | |
195 | static void copy_to_brd(struct brd_device *brd, const void *src, | |
196 | sector_t sector, size_t n) | |
197 | { | |
198 | struct page *page; | |
199 | void *dst; | |
200 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
201 | size_t copy; | |
202 | ||
203 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
204 | page = brd_lookup_page(brd, sector); | |
205 | BUG_ON(!page); | |
206 | ||
207 | dst = kmap_atomic(page, KM_USER1); | |
208 | memcpy(dst + offset, src, copy); | |
209 | kunmap_atomic(dst, KM_USER1); | |
210 | ||
211 | if (copy < n) { | |
212 | src += copy; | |
213 | sector += copy >> SECTOR_SHIFT; | |
214 | copy = n - copy; | |
215 | page = brd_lookup_page(brd, sector); | |
216 | BUG_ON(!page); | |
217 | ||
218 | dst = kmap_atomic(page, KM_USER1); | |
219 | memcpy(dst, src, copy); | |
220 | kunmap_atomic(dst, KM_USER1); | |
221 | } | |
222 | } | |
223 | ||
224 | /* | |
225 | * Copy n bytes to dst from the brd starting at sector. Does not sleep. | |
226 | */ | |
227 | static void copy_from_brd(void *dst, struct brd_device *brd, | |
228 | sector_t sector, size_t n) | |
229 | { | |
230 | struct page *page; | |
231 | void *src; | |
232 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
233 | size_t copy; | |
234 | ||
235 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
236 | page = brd_lookup_page(brd, sector); | |
237 | if (page) { | |
238 | src = kmap_atomic(page, KM_USER1); | |
239 | memcpy(dst, src + offset, copy); | |
240 | kunmap_atomic(src, KM_USER1); | |
241 | } else | |
242 | memset(dst, 0, copy); | |
243 | ||
244 | if (copy < n) { | |
245 | dst += copy; | |
246 | sector += copy >> SECTOR_SHIFT; | |
247 | copy = n - copy; | |
248 | page = brd_lookup_page(brd, sector); | |
249 | if (page) { | |
250 | src = kmap_atomic(page, KM_USER1); | |
251 | memcpy(dst, src, copy); | |
252 | kunmap_atomic(src, KM_USER1); | |
253 | } else | |
254 | memset(dst, 0, copy); | |
255 | } | |
256 | } | |
257 | ||
258 | /* | |
259 | * Process a single bvec of a bio. | |
260 | */ | |
261 | static int brd_do_bvec(struct brd_device *brd, struct page *page, | |
262 | unsigned int len, unsigned int off, int rw, | |
263 | sector_t sector) | |
264 | { | |
265 | void *mem; | |
266 | int err = 0; | |
267 | ||
268 | if (rw != READ) { | |
269 | err = copy_to_brd_setup(brd, sector, len); | |
270 | if (err) | |
271 | goto out; | |
272 | } | |
273 | ||
274 | mem = kmap_atomic(page, KM_USER0); | |
275 | if (rw == READ) { | |
276 | copy_from_brd(mem + off, brd, sector, len); | |
277 | flush_dcache_page(page); | |
278 | } else | |
279 | copy_to_brd(brd, mem + off, sector, len); | |
280 | kunmap_atomic(mem, KM_USER0); | |
281 | ||
282 | out: | |
283 | return err; | |
284 | } | |
285 | ||
286 | static int brd_make_request(struct request_queue *q, struct bio *bio) | |
287 | { | |
288 | struct block_device *bdev = bio->bi_bdev; | |
289 | struct brd_device *brd = bdev->bd_disk->private_data; | |
290 | int rw; | |
291 | struct bio_vec *bvec; | |
292 | sector_t sector; | |
293 | int i; | |
294 | int err = -EIO; | |
295 | ||
296 | sector = bio->bi_sector; | |
297 | if (sector + (bio->bi_size >> SECTOR_SHIFT) > | |
298 | get_capacity(bdev->bd_disk)) | |
299 | goto out; | |
300 | ||
301 | rw = bio_rw(bio); | |
302 | if (rw == READA) | |
303 | rw = READ; | |
304 | ||
305 | bio_for_each_segment(bvec, bio, i) { | |
306 | unsigned int len = bvec->bv_len; | |
307 | err = brd_do_bvec(brd, bvec->bv_page, len, | |
308 | bvec->bv_offset, rw, sector); | |
309 | if (err) | |
310 | break; | |
311 | sector += len >> SECTOR_SHIFT; | |
312 | } | |
313 | ||
314 | out: | |
315 | bio_endio(bio, err); | |
316 | ||
317 | return 0; | |
318 | } | |
319 | ||
75acb9cd NP |
320 | #ifdef CONFIG_BLK_DEV_XIP |
321 | static int brd_direct_access (struct block_device *bdev, sector_t sector, | |
30afcb4b | 322 | void **kaddr, unsigned long *pfn) |
75acb9cd NP |
323 | { |
324 | struct brd_device *brd = bdev->bd_disk->private_data; | |
325 | struct page *page; | |
326 | ||
327 | if (!brd) | |
328 | return -ENODEV; | |
329 | if (sector & (PAGE_SECTORS-1)) | |
330 | return -EINVAL; | |
331 | if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk)) | |
332 | return -ERANGE; | |
333 | page = brd_insert_page(brd, sector); | |
334 | if (!page) | |
335 | return -ENOMEM; | |
30afcb4b JH |
336 | *kaddr = page_address(page); |
337 | *pfn = page_to_pfn(page); | |
75acb9cd NP |
338 | |
339 | return 0; | |
340 | } | |
341 | #endif | |
342 | ||
2b9ecd03 | 343 | static int brd_ioctl(struct block_device *bdev, fmode_t mode, |
9db5579b NP |
344 | unsigned int cmd, unsigned long arg) |
345 | { | |
346 | int error; | |
9db5579b NP |
347 | struct brd_device *brd = bdev->bd_disk->private_data; |
348 | ||
349 | if (cmd != BLKFLSBUF) | |
350 | return -ENOTTY; | |
351 | ||
352 | /* | |
353 | * ram device BLKFLSBUF has special semantics, we want to actually | |
354 | * release and destroy the ramdisk data. | |
355 | */ | |
356 | mutex_lock(&bdev->bd_mutex); | |
357 | error = -EBUSY; | |
358 | if (bdev->bd_openers <= 1) { | |
359 | /* | |
360 | * Invalidate the cache first, so it isn't written | |
361 | * back to the device. | |
362 | * | |
363 | * Another thread might instantiate more buffercache here, | |
364 | * but there is not much we can do to close that race. | |
365 | */ | |
366 | invalidate_bh_lrus(); | |
367 | truncate_inode_pages(bdev->bd_inode->i_mapping, 0); | |
368 | brd_free_pages(brd); | |
369 | error = 0; | |
370 | } | |
371 | mutex_unlock(&bdev->bd_mutex); | |
372 | ||
373 | return error; | |
374 | } | |
375 | ||
376 | static struct block_device_operations brd_fops = { | |
75acb9cd | 377 | .owner = THIS_MODULE, |
2b9ecd03 | 378 | .locked_ioctl = brd_ioctl, |
75acb9cd NP |
379 | #ifdef CONFIG_BLK_DEV_XIP |
380 | .direct_access = brd_direct_access, | |
381 | #endif | |
9db5579b NP |
382 | }; |
383 | ||
384 | /* | |
385 | * And now the modules code and kernel interface. | |
386 | */ | |
387 | static int rd_nr; | |
388 | int rd_size = CONFIG_BLK_DEV_RAM_SIZE; | |
d7853d1f LV |
389 | static int max_part; |
390 | static int part_shift; | |
9db5579b NP |
391 | module_param(rd_nr, int, 0); |
392 | MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices"); | |
393 | module_param(rd_size, int, 0); | |
394 | MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes."); | |
d7853d1f LV |
395 | module_param(max_part, int, 0); |
396 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk"); | |
9db5579b NP |
397 | MODULE_LICENSE("GPL"); |
398 | MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR); | |
efedf51c | 399 | MODULE_ALIAS("rd"); |
9db5579b NP |
400 | |
401 | #ifndef MODULE | |
402 | /* Legacy boot options - nonmodular */ | |
403 | static int __init ramdisk_size(char *str) | |
404 | { | |
405 | rd_size = simple_strtol(str, NULL, 0); | |
406 | return 1; | |
407 | } | |
408 | static int __init ramdisk_size2(char *str) | |
409 | { | |
410 | return ramdisk_size(str); | |
411 | } | |
412 | __setup("ramdisk=", ramdisk_size); | |
413 | __setup("ramdisk_size=", ramdisk_size2); | |
414 | #endif | |
415 | ||
416 | /* | |
417 | * The device scheme is derived from loop.c. Keep them in synch where possible | |
418 | * (should share code eventually). | |
419 | */ | |
420 | static LIST_HEAD(brd_devices); | |
421 | static DEFINE_MUTEX(brd_devices_mutex); | |
422 | ||
423 | static struct brd_device *brd_alloc(int i) | |
424 | { | |
425 | struct brd_device *brd; | |
426 | struct gendisk *disk; | |
427 | ||
428 | brd = kzalloc(sizeof(*brd), GFP_KERNEL); | |
429 | if (!brd) | |
430 | goto out; | |
431 | brd->brd_number = i; | |
432 | spin_lock_init(&brd->brd_lock); | |
433 | INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC); | |
434 | ||
435 | brd->brd_queue = blk_alloc_queue(GFP_KERNEL); | |
436 | if (!brd->brd_queue) | |
437 | goto out_free_dev; | |
438 | blk_queue_make_request(brd->brd_queue, brd_make_request); | |
439 | blk_queue_max_sectors(brd->brd_queue, 1024); | |
440 | blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY); | |
441 | ||
d7853d1f | 442 | disk = brd->brd_disk = alloc_disk(1 << part_shift); |
9db5579b NP |
443 | if (!disk) |
444 | goto out_free_queue; | |
445 | disk->major = RAMDISK_MAJOR; | |
d7853d1f | 446 | disk->first_minor = i << part_shift; |
9db5579b NP |
447 | disk->fops = &brd_fops; |
448 | disk->private_data = brd; | |
449 | disk->queue = brd->brd_queue; | |
53978d0a | 450 | disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO; |
9db5579b NP |
451 | sprintf(disk->disk_name, "ram%d", i); |
452 | set_capacity(disk, rd_size * 2); | |
453 | ||
454 | return brd; | |
455 | ||
456 | out_free_queue: | |
457 | blk_cleanup_queue(brd->brd_queue); | |
458 | out_free_dev: | |
459 | kfree(brd); | |
460 | out: | |
461 | return NULL; | |
462 | } | |
463 | ||
464 | static void brd_free(struct brd_device *brd) | |
465 | { | |
466 | put_disk(brd->brd_disk); | |
467 | blk_cleanup_queue(brd->brd_queue); | |
468 | brd_free_pages(brd); | |
469 | kfree(brd); | |
470 | } | |
471 | ||
472 | static struct brd_device *brd_init_one(int i) | |
473 | { | |
474 | struct brd_device *brd; | |
475 | ||
476 | list_for_each_entry(brd, &brd_devices, brd_list) { | |
477 | if (brd->brd_number == i) | |
478 | goto out; | |
479 | } | |
480 | ||
481 | brd = brd_alloc(i); | |
482 | if (brd) { | |
483 | add_disk(brd->brd_disk); | |
484 | list_add_tail(&brd->brd_list, &brd_devices); | |
485 | } | |
486 | out: | |
487 | return brd; | |
488 | } | |
489 | ||
490 | static void brd_del_one(struct brd_device *brd) | |
491 | { | |
492 | list_del(&brd->brd_list); | |
493 | del_gendisk(brd->brd_disk); | |
494 | brd_free(brd); | |
495 | } | |
496 | ||
497 | static struct kobject *brd_probe(dev_t dev, int *part, void *data) | |
498 | { | |
499 | struct brd_device *brd; | |
500 | struct kobject *kobj; | |
501 | ||
502 | mutex_lock(&brd_devices_mutex); | |
503 | brd = brd_init_one(dev & MINORMASK); | |
504 | kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM); | |
505 | mutex_unlock(&brd_devices_mutex); | |
506 | ||
507 | *part = 0; | |
508 | return kobj; | |
509 | } | |
510 | ||
511 | static int __init brd_init(void) | |
512 | { | |
513 | int i, nr; | |
514 | unsigned long range; | |
515 | struct brd_device *brd, *next; | |
516 | ||
517 | /* | |
518 | * brd module now has a feature to instantiate underlying device | |
519 | * structure on-demand, provided that there is an access dev node. | |
520 | * However, this will not work well with user space tool that doesn't | |
521 | * know about such "feature". In order to not break any existing | |
522 | * tool, we do the following: | |
523 | * | |
524 | * (1) if rd_nr is specified, create that many upfront, and this | |
525 | * also becomes a hard limit. | |
526 | * (2) if rd_nr is not specified, create 1 rd device on module | |
527 | * load, user can further extend brd device by create dev node | |
528 | * themselves and have kernel automatically instantiate actual | |
529 | * device on-demand. | |
530 | */ | |
d7853d1f LV |
531 | |
532 | part_shift = 0; | |
533 | if (max_part > 0) | |
534 | part_shift = fls(max_part); | |
535 | ||
536 | if (rd_nr > 1UL << (MINORBITS - part_shift)) | |
9db5579b NP |
537 | return -EINVAL; |
538 | ||
539 | if (rd_nr) { | |
540 | nr = rd_nr; | |
541 | range = rd_nr; | |
542 | } else { | |
543 | nr = CONFIG_BLK_DEV_RAM_COUNT; | |
d7853d1f | 544 | range = 1UL << (MINORBITS - part_shift); |
9db5579b NP |
545 | } |
546 | ||
547 | if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) | |
548 | return -EIO; | |
549 | ||
550 | for (i = 0; i < nr; i++) { | |
551 | brd = brd_alloc(i); | |
552 | if (!brd) | |
553 | goto out_free; | |
554 | list_add_tail(&brd->brd_list, &brd_devices); | |
555 | } | |
556 | ||
557 | /* point of no return */ | |
558 | ||
559 | list_for_each_entry(brd, &brd_devices, brd_list) | |
560 | add_disk(brd->brd_disk); | |
561 | ||
562 | blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range, | |
563 | THIS_MODULE, brd_probe, NULL, NULL); | |
564 | ||
565 | printk(KERN_INFO "brd: module loaded\n"); | |
566 | return 0; | |
567 | ||
568 | out_free: | |
569 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) { | |
570 | list_del(&brd->brd_list); | |
571 | brd_free(brd); | |
572 | } | |
c82f2966 | 573 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
9db5579b | 574 | |
9db5579b NP |
575 | return -ENOMEM; |
576 | } | |
577 | ||
578 | static void __exit brd_exit(void) | |
579 | { | |
580 | unsigned long range; | |
581 | struct brd_device *brd, *next; | |
582 | ||
d7853d1f | 583 | range = rd_nr ? rd_nr : 1UL << (MINORBITS - part_shift); |
9db5579b NP |
584 | |
585 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) | |
586 | brd_del_one(brd); | |
587 | ||
588 | blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range); | |
589 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); | |
590 | } | |
591 | ||
592 | module_init(brd_init); | |
593 | module_exit(brd_exit); | |
594 |