]>
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> | |
287f3ca5 | 12 | #include <linux/initrd.h> |
9db5579b NP |
13 | #include <linux/module.h> |
14 | #include <linux/moduleparam.h> | |
15 | #include <linux/major.h> | |
16 | #include <linux/blkdev.h> | |
17 | #include <linux/bio.h> | |
18 | #include <linux/highmem.h> | |
2a48fc0a | 19 | #include <linux/mutex.h> |
9db5579b | 20 | #include <linux/radix-tree.h> |
ff01bb48 | 21 | #include <linux/fs.h> |
5a0e3ad6 | 22 | #include <linux/slab.h> |
23c47d2a | 23 | #include <linux/backing-dev.h> |
9db5579b | 24 | |
7c0f6ba6 | 25 | #include <linux/uaccess.h> |
9db5579b NP |
26 | |
27 | #define SECTOR_SHIFT 9 | |
28 | #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT) | |
29 | #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT) | |
30 | ||
31 | /* | |
32 | * Each block ramdisk device has a radix_tree brd_pages of pages that stores | |
33 | * the pages containing the block device's contents. A brd page's ->index is | |
34 | * its offset in PAGE_SIZE units. This is similar to, but in no way connected | |
35 | * with, the kernel's pagecache or buffer cache (which sit above our block | |
36 | * device). | |
37 | */ | |
38 | struct brd_device { | |
39 | int brd_number; | |
9db5579b NP |
40 | |
41 | struct request_queue *brd_queue; | |
42 | struct gendisk *brd_disk; | |
43 | struct list_head brd_list; | |
44 | ||
45 | /* | |
46 | * Backing store of pages and lock to protect it. This is the contents | |
47 | * of the block device. | |
48 | */ | |
49 | spinlock_t brd_lock; | |
50 | struct radix_tree_root brd_pages; | |
51 | }; | |
52 | ||
53 | /* | |
54 | * Look up and return a brd's page for a given sector. | |
55 | */ | |
56 | static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector) | |
57 | { | |
58 | pgoff_t idx; | |
59 | struct page *page; | |
60 | ||
61 | /* | |
62 | * The page lifetime is protected by the fact that we have opened the | |
63 | * device node -- brd pages will never be deleted under us, so we | |
64 | * don't need any further locking or refcounting. | |
65 | * | |
66 | * This is strictly true for the radix-tree nodes as well (ie. we | |
67 | * don't actually need the rcu_read_lock()), however that is not a | |
68 | * documented feature of the radix-tree API so it is better to be | |
69 | * safe here (we don't have total exclusion from radix tree updates | |
70 | * here, only deletes). | |
71 | */ | |
72 | rcu_read_lock(); | |
73 | idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */ | |
74 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
75 | rcu_read_unlock(); | |
76 | ||
77 | BUG_ON(page && page->index != idx); | |
78 | ||
79 | return page; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Look up and return a brd's page for a given sector. | |
84 | * If one does not exist, allocate an empty page, and insert that. Then | |
85 | * return it. | |
86 | */ | |
87 | static struct page *brd_insert_page(struct brd_device *brd, sector_t sector) | |
88 | { | |
89 | pgoff_t idx; | |
90 | struct page *page; | |
75acb9cd | 91 | gfp_t gfp_flags; |
9db5579b NP |
92 | |
93 | page = brd_lookup_page(brd, sector); | |
94 | if (page) | |
95 | return page; | |
96 | ||
97 | /* | |
98 | * Must use NOIO because we don't want to recurse back into the | |
99 | * block or filesystem layers from page reclaim. | |
75acb9cd | 100 | * |
a7a97fc9 MW |
101 | * Cannot support DAX and highmem, because our ->direct_access |
102 | * routine for DAX must return memory that is always addressable. | |
103 | * If DAX was reworked to use pfns and kmap throughout, this | |
75acb9cd | 104 | * restriction might be able to be lifted. |
9db5579b | 105 | */ |
75acb9cd | 106 | gfp_flags = GFP_NOIO | __GFP_ZERO; |
26defe34 | 107 | page = alloc_page(gfp_flags); |
9db5579b NP |
108 | if (!page) |
109 | return NULL; | |
110 | ||
111 | if (radix_tree_preload(GFP_NOIO)) { | |
112 | __free_page(page); | |
113 | return NULL; | |
114 | } | |
115 | ||
116 | spin_lock(&brd->brd_lock); | |
117 | idx = sector >> PAGE_SECTORS_SHIFT; | |
dfd20b2b | 118 | page->index = idx; |
9db5579b NP |
119 | if (radix_tree_insert(&brd->brd_pages, idx, page)) { |
120 | __free_page(page); | |
121 | page = radix_tree_lookup(&brd->brd_pages, idx); | |
122 | BUG_ON(!page); | |
123 | BUG_ON(page->index != idx); | |
dfd20b2b | 124 | } |
9db5579b NP |
125 | spin_unlock(&brd->brd_lock); |
126 | ||
127 | radix_tree_preload_end(); | |
128 | ||
129 | return page; | |
130 | } | |
131 | ||
132 | /* | |
133 | * Free all backing store pages and radix tree. This must only be called when | |
134 | * there are no other users of the device. | |
135 | */ | |
136 | #define FREE_BATCH 16 | |
137 | static void brd_free_pages(struct brd_device *brd) | |
138 | { | |
139 | unsigned long pos = 0; | |
140 | struct page *pages[FREE_BATCH]; | |
141 | int nr_pages; | |
142 | ||
143 | do { | |
144 | int i; | |
145 | ||
146 | nr_pages = radix_tree_gang_lookup(&brd->brd_pages, | |
147 | (void **)pages, pos, FREE_BATCH); | |
148 | ||
149 | for (i = 0; i < nr_pages; i++) { | |
150 | void *ret; | |
151 | ||
152 | BUG_ON(pages[i]->index < pos); | |
153 | pos = pages[i]->index; | |
154 | ret = radix_tree_delete(&brd->brd_pages, pos); | |
155 | BUG_ON(!ret || ret != pages[i]); | |
156 | __free_page(pages[i]); | |
157 | } | |
158 | ||
159 | pos++; | |
160 | ||
161 | /* | |
162 | * This assumes radix_tree_gang_lookup always returns as | |
163 | * many pages as possible. If the radix-tree code changes, | |
164 | * so will this have to. | |
165 | */ | |
166 | } while (nr_pages == FREE_BATCH); | |
167 | } | |
168 | ||
169 | /* | |
170 | * copy_to_brd_setup must be called before copy_to_brd. It may sleep. | |
171 | */ | |
172 | static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n) | |
173 | { | |
174 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
175 | size_t copy; | |
176 | ||
177 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
178 | if (!brd_insert_page(brd, sector)) | |
96f8d8e0 | 179 | return -ENOSPC; |
9db5579b NP |
180 | if (copy < n) { |
181 | sector += copy >> SECTOR_SHIFT; | |
182 | if (!brd_insert_page(brd, sector)) | |
96f8d8e0 | 183 | return -ENOSPC; |
9db5579b NP |
184 | } |
185 | return 0; | |
186 | } | |
187 | ||
188 | /* | |
189 | * Copy n bytes from src to the brd starting at sector. Does not sleep. | |
190 | */ | |
191 | static void copy_to_brd(struct brd_device *brd, const void *src, | |
192 | sector_t sector, size_t n) | |
193 | { | |
194 | struct page *page; | |
195 | void *dst; | |
196 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
197 | size_t copy; | |
198 | ||
199 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
200 | page = brd_lookup_page(brd, sector); | |
201 | BUG_ON(!page); | |
202 | ||
cfd8005c | 203 | dst = kmap_atomic(page); |
9db5579b | 204 | memcpy(dst + offset, src, copy); |
cfd8005c | 205 | kunmap_atomic(dst); |
9db5579b NP |
206 | |
207 | if (copy < n) { | |
208 | src += copy; | |
209 | sector += copy >> SECTOR_SHIFT; | |
210 | copy = n - copy; | |
211 | page = brd_lookup_page(brd, sector); | |
212 | BUG_ON(!page); | |
213 | ||
cfd8005c | 214 | dst = kmap_atomic(page); |
9db5579b | 215 | memcpy(dst, src, copy); |
cfd8005c | 216 | kunmap_atomic(dst); |
9db5579b NP |
217 | } |
218 | } | |
219 | ||
220 | /* | |
221 | * Copy n bytes to dst from the brd starting at sector. Does not sleep. | |
222 | */ | |
223 | static void copy_from_brd(void *dst, struct brd_device *brd, | |
224 | sector_t sector, size_t n) | |
225 | { | |
226 | struct page *page; | |
227 | void *src; | |
228 | unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT; | |
229 | size_t copy; | |
230 | ||
231 | copy = min_t(size_t, n, PAGE_SIZE - offset); | |
232 | page = brd_lookup_page(brd, sector); | |
233 | if (page) { | |
cfd8005c | 234 | src = kmap_atomic(page); |
9db5579b | 235 | memcpy(dst, src + offset, copy); |
cfd8005c | 236 | kunmap_atomic(src); |
9db5579b NP |
237 | } else |
238 | memset(dst, 0, copy); | |
239 | ||
240 | if (copy < n) { | |
241 | dst += copy; | |
242 | sector += copy >> SECTOR_SHIFT; | |
243 | copy = n - copy; | |
244 | page = brd_lookup_page(brd, sector); | |
245 | if (page) { | |
cfd8005c | 246 | src = kmap_atomic(page); |
9db5579b | 247 | memcpy(dst, src, copy); |
cfd8005c | 248 | kunmap_atomic(src); |
9db5579b NP |
249 | } else |
250 | memset(dst, 0, copy); | |
251 | } | |
252 | } | |
253 | ||
254 | /* | |
255 | * Process a single bvec of a bio. | |
256 | */ | |
257 | static int brd_do_bvec(struct brd_device *brd, struct page *page, | |
c11f0c0b | 258 | unsigned int len, unsigned int off, bool is_write, |
9db5579b NP |
259 | sector_t sector) |
260 | { | |
261 | void *mem; | |
262 | int err = 0; | |
263 | ||
c11f0c0b | 264 | if (is_write) { |
9db5579b NP |
265 | err = copy_to_brd_setup(brd, sector, len); |
266 | if (err) | |
267 | goto out; | |
268 | } | |
269 | ||
cfd8005c | 270 | mem = kmap_atomic(page); |
c11f0c0b | 271 | if (!is_write) { |
9db5579b NP |
272 | copy_from_brd(mem + off, brd, sector, len); |
273 | flush_dcache_page(page); | |
c2572f2b NP |
274 | } else { |
275 | flush_dcache_page(page); | |
9db5579b | 276 | copy_to_brd(brd, mem + off, sector, len); |
c2572f2b | 277 | } |
cfd8005c | 278 | kunmap_atomic(mem); |
9db5579b NP |
279 | |
280 | out: | |
281 | return err; | |
282 | } | |
283 | ||
dece1635 | 284 | static blk_qc_t brd_make_request(struct request_queue *q, struct bio *bio) |
9db5579b | 285 | { |
74d46992 | 286 | struct brd_device *brd = bio->bi_disk->private_data; |
7988613b | 287 | struct bio_vec bvec; |
9db5579b | 288 | sector_t sector; |
7988613b | 289 | struct bvec_iter iter; |
9db5579b | 290 | |
4f024f37 | 291 | sector = bio->bi_iter.bi_sector; |
74d46992 | 292 | if (bio_end_sector(bio) > get_capacity(bio->bi_disk)) |
4246a0b6 | 293 | goto io_error; |
9db5579b | 294 | |
7988613b KO |
295 | bio_for_each_segment(bvec, bio, iter) { |
296 | unsigned int len = bvec.bv_len; | |
4246a0b6 CH |
297 | int err; |
298 | ||
c11f0c0b JA |
299 | err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset, |
300 | op_is_write(bio_op(bio)), sector); | |
9db5579b | 301 | if (err) |
4246a0b6 | 302 | goto io_error; |
9db5579b NP |
303 | sector += len >> SECTOR_SHIFT; |
304 | } | |
305 | ||
4246a0b6 | 306 | bio_endio(bio); |
dece1635 | 307 | return BLK_QC_T_NONE; |
4246a0b6 CH |
308 | io_error: |
309 | bio_io_error(bio); | |
dece1635 | 310 | return BLK_QC_T_NONE; |
9db5579b NP |
311 | } |
312 | ||
a72132c3 | 313 | static int brd_rw_page(struct block_device *bdev, sector_t sector, |
c11f0c0b | 314 | struct page *page, bool is_write) |
a72132c3 MW |
315 | { |
316 | struct brd_device *brd = bdev->bd_disk->private_data; | |
98cc093c HY |
317 | int err; |
318 | ||
319 | if (PageTransHuge(page)) | |
320 | return -ENOTSUPP; | |
321 | err = brd_do_bvec(brd, page, PAGE_SIZE, 0, is_write, sector); | |
c11f0c0b | 322 | page_endio(page, is_write, err); |
a72132c3 MW |
323 | return err; |
324 | } | |
325 | ||
83d5cde4 | 326 | static const struct block_device_operations brd_fops = { |
75acb9cd | 327 | .owner = THIS_MODULE, |
a72132c3 | 328 | .rw_page = brd_rw_page, |
9db5579b NP |
329 | }; |
330 | ||
331 | /* | |
332 | * And now the modules code and kernel interface. | |
333 | */ | |
937af5ec | 334 | static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT; |
8892cbaf | 335 | module_param(rd_nr, int, S_IRUGO); |
9db5579b | 336 | MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices"); |
937af5ec | 337 | |
366f4aea JK |
338 | unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE; |
339 | module_param(rd_size, ulong, S_IRUGO); | |
9db5579b | 340 | MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes."); |
937af5ec BH |
341 | |
342 | static int max_part = 1; | |
8892cbaf | 343 | module_param(max_part, int, S_IRUGO); |
937af5ec BH |
344 | MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices"); |
345 | ||
9db5579b NP |
346 | MODULE_LICENSE("GPL"); |
347 | MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR); | |
efedf51c | 348 | MODULE_ALIAS("rd"); |
9db5579b NP |
349 | |
350 | #ifndef MODULE | |
351 | /* Legacy boot options - nonmodular */ | |
352 | static int __init ramdisk_size(char *str) | |
353 | { | |
354 | rd_size = simple_strtol(str, NULL, 0); | |
355 | return 1; | |
356 | } | |
1adbee50 | 357 | __setup("ramdisk_size=", ramdisk_size); |
9db5579b NP |
358 | #endif |
359 | ||
360 | /* | |
361 | * The device scheme is derived from loop.c. Keep them in synch where possible | |
362 | * (should share code eventually). | |
363 | */ | |
364 | static LIST_HEAD(brd_devices); | |
365 | static DEFINE_MUTEX(brd_devices_mutex); | |
366 | ||
367 | static struct brd_device *brd_alloc(int i) | |
368 | { | |
369 | struct brd_device *brd; | |
370 | struct gendisk *disk; | |
371 | ||
372 | brd = kzalloc(sizeof(*brd), GFP_KERNEL); | |
373 | if (!brd) | |
374 | goto out; | |
375 | brd->brd_number = i; | |
376 | spin_lock_init(&brd->brd_lock); | |
377 | INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC); | |
378 | ||
379 | brd->brd_queue = blk_alloc_queue(GFP_KERNEL); | |
380 | if (!brd->brd_queue) | |
381 | goto out_free_dev; | |
c8fa3173 | 382 | |
9db5579b | 383 | blk_queue_make_request(brd->brd_queue, brd_make_request); |
086fa5ff | 384 | blk_queue_max_hw_sectors(brd->brd_queue, 1024); |
9db5579b | 385 | |
c8fa3173 BH |
386 | /* This is so fdisk will align partitions on 4k, because of |
387 | * direct_access API needing 4k alignment, returning a PFN | |
388 | * (This is only a problem on very small devices <= 4M, | |
389 | * otherwise fdisk will align on 1M. Regardless this call | |
390 | * is harmless) | |
391 | */ | |
392 | blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE); | |
937af5ec | 393 | disk = brd->brd_disk = alloc_disk(max_part); |
9db5579b NP |
394 | if (!disk) |
395 | goto out_free_queue; | |
396 | disk->major = RAMDISK_MAJOR; | |
937af5ec | 397 | disk->first_minor = i * max_part; |
9db5579b NP |
398 | disk->fops = &brd_fops; |
399 | disk->private_data = brd; | |
400 | disk->queue = brd->brd_queue; | |
937af5ec | 401 | disk->flags = GENHD_FL_EXT_DEVT; |
9db5579b NP |
402 | sprintf(disk->disk_name, "ram%d", i); |
403 | set_capacity(disk, rd_size * 2); | |
23c47d2a | 404 | disk->queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO; |
9db5579b NP |
405 | |
406 | return brd; | |
407 | ||
408 | out_free_queue: | |
409 | blk_cleanup_queue(brd->brd_queue); | |
410 | out_free_dev: | |
411 | kfree(brd); | |
412 | out: | |
413 | return NULL; | |
414 | } | |
415 | ||
416 | static void brd_free(struct brd_device *brd) | |
417 | { | |
418 | put_disk(brd->brd_disk); | |
419 | blk_cleanup_queue(brd->brd_queue); | |
420 | brd_free_pages(brd); | |
421 | kfree(brd); | |
422 | } | |
423 | ||
937af5ec | 424 | static struct brd_device *brd_init_one(int i, bool *new) |
9db5579b NP |
425 | { |
426 | struct brd_device *brd; | |
427 | ||
937af5ec | 428 | *new = false; |
9db5579b NP |
429 | list_for_each_entry(brd, &brd_devices, brd_list) { |
430 | if (brd->brd_number == i) | |
431 | goto out; | |
432 | } | |
433 | ||
434 | brd = brd_alloc(i); | |
435 | if (brd) { | |
436 | add_disk(brd->brd_disk); | |
437 | list_add_tail(&brd->brd_list, &brd_devices); | |
438 | } | |
937af5ec | 439 | *new = true; |
9db5579b NP |
440 | out: |
441 | return brd; | |
442 | } | |
443 | ||
444 | static void brd_del_one(struct brd_device *brd) | |
445 | { | |
446 | list_del(&brd->brd_list); | |
447 | del_gendisk(brd->brd_disk); | |
448 | brd_free(brd); | |
449 | } | |
450 | ||
451 | static struct kobject *brd_probe(dev_t dev, int *part, void *data) | |
452 | { | |
453 | struct brd_device *brd; | |
454 | struct kobject *kobj; | |
937af5ec | 455 | bool new; |
9db5579b NP |
456 | |
457 | mutex_lock(&brd_devices_mutex); | |
937af5ec | 458 | brd = brd_init_one(MINOR(dev) / max_part, &new); |
a207f593 | 459 | kobj = brd ? get_disk(brd->brd_disk) : NULL; |
9db5579b NP |
460 | mutex_unlock(&brd_devices_mutex); |
461 | ||
937af5ec BH |
462 | if (new) |
463 | *part = 0; | |
464 | ||
9db5579b NP |
465 | return kobj; |
466 | } | |
467 | ||
468 | static int __init brd_init(void) | |
469 | { | |
9db5579b | 470 | struct brd_device *brd, *next; |
937af5ec | 471 | int i; |
9db5579b NP |
472 | |
473 | /* | |
474 | * brd module now has a feature to instantiate underlying device | |
475 | * structure on-demand, provided that there is an access dev node. | |
9db5579b | 476 | * |
937af5ec BH |
477 | * (1) if rd_nr is specified, create that many upfront. else |
478 | * it defaults to CONFIG_BLK_DEV_RAM_COUNT | |
479 | * (2) User can further extend brd devices by create dev node themselves | |
480 | * and have kernel automatically instantiate actual device | |
481 | * on-demand. Example: | |
482 | * mknod /path/devnod_name b 1 X # 1 is the rd major | |
483 | * fdisk -l /path/devnod_name | |
484 | * If (X / max_part) was not already created it will be created | |
485 | * dynamically. | |
9db5579b | 486 | */ |
d7853d1f | 487 | |
9db5579b NP |
488 | if (register_blkdev(RAMDISK_MAJOR, "ramdisk")) |
489 | return -EIO; | |
490 | ||
937af5ec BH |
491 | if (unlikely(!max_part)) |
492 | max_part = 1; | |
493 | ||
494 | for (i = 0; i < rd_nr; i++) { | |
9db5579b NP |
495 | brd = brd_alloc(i); |
496 | if (!brd) | |
497 | goto out_free; | |
498 | list_add_tail(&brd->brd_list, &brd_devices); | |
499 | } | |
500 | ||
501 | /* point of no return */ | |
502 | ||
503 | list_for_each_entry(brd, &brd_devices, brd_list) | |
504 | add_disk(brd->brd_disk); | |
505 | ||
937af5ec | 506 | blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS, |
9db5579b NP |
507 | THIS_MODULE, brd_probe, NULL, NULL); |
508 | ||
937af5ec | 509 | pr_info("brd: module loaded\n"); |
9db5579b NP |
510 | return 0; |
511 | ||
512 | out_free: | |
513 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) { | |
514 | list_del(&brd->brd_list); | |
515 | brd_free(brd); | |
516 | } | |
c82f2966 | 517 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
9db5579b | 518 | |
937af5ec | 519 | pr_info("brd: module NOT loaded !!!\n"); |
9db5579b NP |
520 | return -ENOMEM; |
521 | } | |
522 | ||
523 | static void __exit brd_exit(void) | |
524 | { | |
9db5579b NP |
525 | struct brd_device *brd, *next; |
526 | ||
9db5579b NP |
527 | list_for_each_entry_safe(brd, next, &brd_devices, brd_list) |
528 | brd_del_one(brd); | |
529 | ||
937af5ec | 530 | blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS); |
9db5579b | 531 | unregister_blkdev(RAMDISK_MAJOR, "ramdisk"); |
937af5ec BH |
532 | |
533 | pr_info("brd: module unloaded\n"); | |
9db5579b NP |
534 | } |
535 | ||
536 | module_init(brd_init); | |
537 | module_exit(brd_exit); | |
538 |