blob: 3598110d2cefcff5cc1b69319b8d1891f7f46a2f [file] [log] [blame]
Nick Piggin9db55792008-02-08 04:19:49 -08001/*
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>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020018#include <linux/mutex.h>
Nick Piggin9db55792008-02-08 04:19:49 -080019#include <linux/radix-tree.h>
Al Viroff01bb482011-09-16 02:31:11 -040020#include <linux/fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090021#include <linux/slab.h>
Nick Piggin9db55792008-02-08 04:19:49 -080022
23#include <asm/uaccess.h>
24
25#define SECTOR_SHIFT 9
26#define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
27#define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
28
29/*
30 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
31 * the pages containing the block device's contents. A brd page's ->index is
32 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
33 * with, the kernel's pagecache or buffer cache (which sit above our block
34 * device).
35 */
36struct brd_device {
37 int brd_number;
Nick Piggin9db55792008-02-08 04:19:49 -080038
39 struct request_queue *brd_queue;
40 struct gendisk *brd_disk;
41 struct list_head brd_list;
42
43 /*
44 * Backing store of pages and lock to protect it. This is the contents
45 * of the block device.
46 */
47 spinlock_t brd_lock;
48 struct radix_tree_root brd_pages;
49};
50
51/*
52 * Look up and return a brd's page for a given sector.
53 */
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020054static DEFINE_MUTEX(brd_mutex);
Nick Piggin9db55792008-02-08 04:19:49 -080055static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
56{
57 pgoff_t idx;
58 struct page *page;
59
60 /*
61 * The page lifetime is protected by the fact that we have opened the
62 * device node -- brd pages will never be deleted under us, so we
63 * don't need any further locking or refcounting.
64 *
65 * This is strictly true for the radix-tree nodes as well (ie. we
66 * don't actually need the rcu_read_lock()), however that is not a
67 * documented feature of the radix-tree API so it is better to be
68 * safe here (we don't have total exclusion from radix tree updates
69 * here, only deletes).
70 */
71 rcu_read_lock();
72 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
73 page = radix_tree_lookup(&brd->brd_pages, idx);
74 rcu_read_unlock();
75
76 BUG_ON(page && page->index != idx);
77
78 return page;
79}
80
81/*
82 * Look up and return a brd's page for a given sector.
83 * If one does not exist, allocate an empty page, and insert that. Then
84 * return it.
85 */
86static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
87{
88 pgoff_t idx;
89 struct page *page;
Nick Piggin75acb9c2008-02-08 04:19:50 -080090 gfp_t gfp_flags;
Nick Piggin9db55792008-02-08 04:19:49 -080091
92 page = brd_lookup_page(brd, sector);
93 if (page)
94 return page;
95
96 /*
97 * Must use NOIO because we don't want to recurse back into the
98 * block or filesystem layers from page reclaim.
Nick Piggin75acb9c2008-02-08 04:19:50 -080099 *
100 * Cannot support XIP and highmem, because our ->direct_access
101 * routine for XIP must return memory that is always addressable.
102 * If XIP was reworked to use pfns and kmap throughout, this
103 * restriction might be able to be lifted.
Nick Piggin9db55792008-02-08 04:19:49 -0800104 */
Nick Piggin75acb9c2008-02-08 04:19:50 -0800105 gfp_flags = GFP_NOIO | __GFP_ZERO;
106#ifndef CONFIG_BLK_DEV_XIP
107 gfp_flags |= __GFP_HIGHMEM;
108#endif
Petr Tesarik26defe32008-04-22 05:36:52 +0200109 page = alloc_page(gfp_flags);
Nick Piggin9db55792008-02-08 04:19:49 -0800110 if (!page)
111 return NULL;
112
113 if (radix_tree_preload(GFP_NOIO)) {
114 __free_page(page);
115 return NULL;
116 }
117
118 spin_lock(&brd->brd_lock);
119 idx = sector >> PAGE_SECTORS_SHIFT;
Brian Behlendorfdfd20b22013-05-24 15:55:28 -0700120 page->index = idx;
Nick Piggin9db55792008-02-08 04:19:49 -0800121 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
122 __free_page(page);
123 page = radix_tree_lookup(&brd->brd_pages, idx);
124 BUG_ON(!page);
125 BUG_ON(page->index != idx);
Brian Behlendorfdfd20b22013-05-24 15:55:28 -0700126 }
Nick Piggin9db55792008-02-08 04:19:49 -0800127 spin_unlock(&brd->brd_lock);
128
129 radix_tree_preload_end();
130
131 return page;
132}
133
Nick Pigginb7c33572010-05-26 15:41:14 +0200134static void brd_free_page(struct brd_device *brd, sector_t sector)
135{
136 struct page *page;
137 pgoff_t idx;
138
139 spin_lock(&brd->brd_lock);
140 idx = sector >> PAGE_SECTORS_SHIFT;
141 page = radix_tree_delete(&brd->brd_pages, idx);
142 spin_unlock(&brd->brd_lock);
143 if (page)
144 __free_page(page);
145}
146
147static void brd_zero_page(struct brd_device *brd, sector_t sector)
148{
149 struct page *page;
150
151 page = brd_lookup_page(brd, sector);
152 if (page)
153 clear_highpage(page);
154}
155
Nick Piggin9db55792008-02-08 04:19:49 -0800156/*
157 * Free all backing store pages and radix tree. This must only be called when
158 * there are no other users of the device.
159 */
160#define FREE_BATCH 16
161static void brd_free_pages(struct brd_device *brd)
162{
163 unsigned long pos = 0;
164 struct page *pages[FREE_BATCH];
165 int nr_pages;
166
167 do {
168 int i;
169
170 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
171 (void **)pages, pos, FREE_BATCH);
172
173 for (i = 0; i < nr_pages; i++) {
174 void *ret;
175
176 BUG_ON(pages[i]->index < pos);
177 pos = pages[i]->index;
178 ret = radix_tree_delete(&brd->brd_pages, pos);
179 BUG_ON(!ret || ret != pages[i]);
180 __free_page(pages[i]);
181 }
182
183 pos++;
184
185 /*
186 * This assumes radix_tree_gang_lookup always returns as
187 * many pages as possible. If the radix-tree code changes,
188 * so will this have to.
189 */
190 } while (nr_pages == FREE_BATCH);
191}
192
193/*
194 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
195 */
196static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
197{
198 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
199 size_t copy;
200
201 copy = min_t(size_t, n, PAGE_SIZE - offset);
202 if (!brd_insert_page(brd, sector))
Matthew Wilcox96f8d8e2014-06-04 16:07:50 -0700203 return -ENOSPC;
Nick Piggin9db55792008-02-08 04:19:49 -0800204 if (copy < n) {
205 sector += copy >> SECTOR_SHIFT;
206 if (!brd_insert_page(brd, sector))
Matthew Wilcox96f8d8e2014-06-04 16:07:50 -0700207 return -ENOSPC;
Nick Piggin9db55792008-02-08 04:19:49 -0800208 }
209 return 0;
210}
211
Nick Pigginb7c33572010-05-26 15:41:14 +0200212static void discard_from_brd(struct brd_device *brd,
213 sector_t sector, size_t n)
214{
215 while (n >= PAGE_SIZE) {
216 /*
217 * Don't want to actually discard pages here because
218 * re-allocating the pages can result in writeback
219 * deadlocks under heavy load.
220 */
221 if (0)
222 brd_free_page(brd, sector);
223 else
224 brd_zero_page(brd, sector);
225 sector += PAGE_SIZE >> SECTOR_SHIFT;
226 n -= PAGE_SIZE;
227 }
228}
229
Nick Piggin9db55792008-02-08 04:19:49 -0800230/*
231 * Copy n bytes from src to the brd starting at sector. Does not sleep.
232 */
233static void copy_to_brd(struct brd_device *brd, const void *src,
234 sector_t sector, size_t n)
235{
236 struct page *page;
237 void *dst;
238 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
239 size_t copy;
240
241 copy = min_t(size_t, n, PAGE_SIZE - offset);
242 page = brd_lookup_page(brd, sector);
243 BUG_ON(!page);
244
Cong Wangcfd80052011-11-25 23:14:18 +0800245 dst = kmap_atomic(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800246 memcpy(dst + offset, src, copy);
Cong Wangcfd80052011-11-25 23:14:18 +0800247 kunmap_atomic(dst);
Nick Piggin9db55792008-02-08 04:19:49 -0800248
249 if (copy < n) {
250 src += copy;
251 sector += copy >> SECTOR_SHIFT;
252 copy = n - copy;
253 page = brd_lookup_page(brd, sector);
254 BUG_ON(!page);
255
Cong Wangcfd80052011-11-25 23:14:18 +0800256 dst = kmap_atomic(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800257 memcpy(dst, src, copy);
Cong Wangcfd80052011-11-25 23:14:18 +0800258 kunmap_atomic(dst);
Nick Piggin9db55792008-02-08 04:19:49 -0800259 }
260}
261
262/*
263 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
264 */
265static void copy_from_brd(void *dst, struct brd_device *brd,
266 sector_t sector, size_t n)
267{
268 struct page *page;
269 void *src;
270 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
271 size_t copy;
272
273 copy = min_t(size_t, n, PAGE_SIZE - offset);
274 page = brd_lookup_page(brd, sector);
275 if (page) {
Cong Wangcfd80052011-11-25 23:14:18 +0800276 src = kmap_atomic(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800277 memcpy(dst, src + offset, copy);
Cong Wangcfd80052011-11-25 23:14:18 +0800278 kunmap_atomic(src);
Nick Piggin9db55792008-02-08 04:19:49 -0800279 } else
280 memset(dst, 0, copy);
281
282 if (copy < n) {
283 dst += copy;
284 sector += copy >> SECTOR_SHIFT;
285 copy = n - copy;
286 page = brd_lookup_page(brd, sector);
287 if (page) {
Cong Wangcfd80052011-11-25 23:14:18 +0800288 src = kmap_atomic(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800289 memcpy(dst, src, copy);
Cong Wangcfd80052011-11-25 23:14:18 +0800290 kunmap_atomic(src);
Nick Piggin9db55792008-02-08 04:19:49 -0800291 } else
292 memset(dst, 0, copy);
293 }
294}
295
296/*
297 * Process a single bvec of a bio.
298 */
299static int brd_do_bvec(struct brd_device *brd, struct page *page,
300 unsigned int len, unsigned int off, int rw,
301 sector_t sector)
302{
303 void *mem;
304 int err = 0;
305
306 if (rw != READ) {
307 err = copy_to_brd_setup(brd, sector, len);
308 if (err)
309 goto out;
310 }
311
Cong Wangcfd80052011-11-25 23:14:18 +0800312 mem = kmap_atomic(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800313 if (rw == READ) {
314 copy_from_brd(mem + off, brd, sector, len);
315 flush_dcache_page(page);
Nick Pigginc2572f22009-04-15 10:32:07 +0200316 } else {
317 flush_dcache_page(page);
Nick Piggin9db55792008-02-08 04:19:49 -0800318 copy_to_brd(brd, mem + off, sector, len);
Nick Pigginc2572f22009-04-15 10:32:07 +0200319 }
Cong Wangcfd80052011-11-25 23:14:18 +0800320 kunmap_atomic(mem);
Nick Piggin9db55792008-02-08 04:19:49 -0800321
322out:
323 return err;
324}
325
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200326static void brd_make_request(struct request_queue *q, struct bio *bio)
Nick Piggin9db55792008-02-08 04:19:49 -0800327{
328 struct block_device *bdev = bio->bi_bdev;
329 struct brd_device *brd = bdev->bd_disk->private_data;
330 int rw;
Kent Overstreet79886132013-11-23 17:19:00 -0800331 struct bio_vec bvec;
Nick Piggin9db55792008-02-08 04:19:49 -0800332 sector_t sector;
Kent Overstreet79886132013-11-23 17:19:00 -0800333 struct bvec_iter iter;
Nick Piggin9db55792008-02-08 04:19:49 -0800334 int err = -EIO;
335
Kent Overstreet4f024f32013-10-11 15:44:27 -0700336 sector = bio->bi_iter.bi_sector;
Kent Overstreetf73a1c72012-09-25 15:05:12 -0700337 if (bio_end_sector(bio) > get_capacity(bdev->bd_disk))
Nick Piggin9db55792008-02-08 04:19:49 -0800338 goto out;
339
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +0200340 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
Nick Pigginb7c33572010-05-26 15:41:14 +0200341 err = 0;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700342 discard_from_brd(brd, sector, bio->bi_iter.bi_size);
Nick Pigginb7c33572010-05-26 15:41:14 +0200343 goto out;
344 }
345
Nick Piggin9db55792008-02-08 04:19:49 -0800346 rw = bio_rw(bio);
347 if (rw == READA)
348 rw = READ;
349
Kent Overstreet79886132013-11-23 17:19:00 -0800350 bio_for_each_segment(bvec, bio, iter) {
351 unsigned int len = bvec.bv_len;
352 err = brd_do_bvec(brd, bvec.bv_page, len,
353 bvec.bv_offset, rw, sector);
Nick Piggin9db55792008-02-08 04:19:49 -0800354 if (err)
355 break;
356 sector += len >> SECTOR_SHIFT;
357 }
358
359out:
360 bio_endio(bio, err);
Nick Piggin9db55792008-02-08 04:19:49 -0800361}
362
Matthew Wilcoxa72132c2014-06-04 16:07:49 -0700363static int brd_rw_page(struct block_device *bdev, sector_t sector,
364 struct page *page, int rw)
365{
366 struct brd_device *brd = bdev->bd_disk->private_data;
367 int err = brd_do_bvec(brd, page, PAGE_CACHE_SIZE, 0, rw, sector);
368 page_endio(page, rw & WRITE, err);
369 return err;
370}
371
Nick Piggin75acb9c2008-02-08 04:19:50 -0800372#ifdef CONFIG_BLK_DEV_XIP
Nick Pigginb7c33572010-05-26 15:41:14 +0200373static int brd_direct_access(struct block_device *bdev, sector_t sector,
Jared Hulbert30afcb42008-04-28 02:13:02 -0700374 void **kaddr, unsigned long *pfn)
Nick Piggin75acb9c2008-02-08 04:19:50 -0800375{
376 struct brd_device *brd = bdev->bd_disk->private_data;
377 struct page *page;
378
379 if (!brd)
380 return -ENODEV;
381 if (sector & (PAGE_SECTORS-1))
382 return -EINVAL;
383 if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
384 return -ERANGE;
385 page = brd_insert_page(brd, sector);
386 if (!page)
Matthew Wilcox96f8d8e2014-06-04 16:07:50 -0700387 return -ENOSPC;
Jared Hulbert30afcb42008-04-28 02:13:02 -0700388 *kaddr = page_address(page);
389 *pfn = page_to_pfn(page);
Nick Piggin75acb9c2008-02-08 04:19:50 -0800390
391 return 0;
392}
393#endif
394
Al Viro2b9ecd02008-03-02 09:24:35 -0500395static int brd_ioctl(struct block_device *bdev, fmode_t mode,
Nick Piggin9db55792008-02-08 04:19:49 -0800396 unsigned int cmd, unsigned long arg)
397{
398 int error;
Nick Piggin9db55792008-02-08 04:19:49 -0800399 struct brd_device *brd = bdev->bd_disk->private_data;
400
401 if (cmd != BLKFLSBUF)
402 return -ENOTTY;
403
404 /*
405 * ram device BLKFLSBUF has special semantics, we want to actually
406 * release and destroy the ramdisk data.
407 */
Arnd Bergmann2a48fc02010-06-02 14:28:52 +0200408 mutex_lock(&brd_mutex);
Nick Piggin9db55792008-02-08 04:19:49 -0800409 mutex_lock(&bdev->bd_mutex);
410 error = -EBUSY;
411 if (bdev->bd_openers <= 1) {
412 /*
Al Viroff01bb482011-09-16 02:31:11 -0400413 * Kill the cache first, so it isn't written back to the
414 * device.
Nick Piggin9db55792008-02-08 04:19:49 -0800415 *
416 * Another thread might instantiate more buffercache here,
417 * but there is not much we can do to close that race.
418 */
Al Viroff01bb482011-09-16 02:31:11 -0400419 kill_bdev(bdev);
Nick Piggin9db55792008-02-08 04:19:49 -0800420 brd_free_pages(brd);
421 error = 0;
422 }
423 mutex_unlock(&bdev->bd_mutex);
Arnd Bergmann2a48fc02010-06-02 14:28:52 +0200424 mutex_unlock(&brd_mutex);
Nick Piggin9db55792008-02-08 04:19:49 -0800425
426 return error;
427}
428
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700429static const struct block_device_operations brd_fops = {
Nick Piggin75acb9c2008-02-08 04:19:50 -0800430 .owner = THIS_MODULE,
Matthew Wilcoxa72132c2014-06-04 16:07:49 -0700431 .rw_page = brd_rw_page,
Arnd Bergmann8a6cfeb2010-07-08 10:18:46 +0200432 .ioctl = brd_ioctl,
Nick Piggin75acb9c2008-02-08 04:19:50 -0800433#ifdef CONFIG_BLK_DEV_XIP
434 .direct_access = brd_direct_access,
435#endif
Nick Piggin9db55792008-02-08 04:19:49 -0800436};
437
438/*
439 * And now the modules code and kernel interface.
440 */
441static int rd_nr;
442int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
Laurent Vivierd7853d12008-04-30 00:55:06 -0700443static int max_part;
444static int part_shift;
Dmitry Monakhovaeac3182014-08-18 12:49:08 +0400445static int part_show = 0;
Namhyung Kim8892cba2011-05-26 21:06:50 +0200446module_param(rd_nr, int, S_IRUGO);
Nick Piggin9db55792008-02-08 04:19:49 -0800447MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
Namhyung Kim8892cba2011-05-26 21:06:50 +0200448module_param(rd_size, int, S_IRUGO);
Nick Piggin9db55792008-02-08 04:19:49 -0800449MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
Namhyung Kim8892cba2011-05-26 21:06:50 +0200450module_param(max_part, int, S_IRUGO);
Laurent Vivierd7853d12008-04-30 00:55:06 -0700451MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
Dmitry Monakhovaeac3182014-08-18 12:49:08 +0400452module_param(part_show, int, S_IRUGO);
453MODULE_PARM_DESC(part_show, "Control RAM disk visibility in /proc/partitions");
Nick Piggin9db55792008-02-08 04:19:49 -0800454MODULE_LICENSE("GPL");
455MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
Nick Pigginefedf512008-06-04 17:18:42 +0200456MODULE_ALIAS("rd");
Nick Piggin9db55792008-02-08 04:19:49 -0800457
458#ifndef MODULE
459/* Legacy boot options - nonmodular */
460static int __init ramdisk_size(char *str)
461{
462 rd_size = simple_strtol(str, NULL, 0);
463 return 1;
464}
Robert P. J. Day1adbee52009-06-10 12:57:08 -0700465__setup("ramdisk_size=", ramdisk_size);
Nick Piggin9db55792008-02-08 04:19:49 -0800466#endif
467
468/*
469 * The device scheme is derived from loop.c. Keep them in synch where possible
470 * (should share code eventually).
471 */
472static LIST_HEAD(brd_devices);
473static DEFINE_MUTEX(brd_devices_mutex);
474
475static struct brd_device *brd_alloc(int i)
476{
477 struct brd_device *brd;
478 struct gendisk *disk;
479
480 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
481 if (!brd)
482 goto out;
483 brd->brd_number = i;
484 spin_lock_init(&brd->brd_lock);
485 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
486
487 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
488 if (!brd->brd_queue)
489 goto out_free_dev;
490 blk_queue_make_request(brd->brd_queue, brd_make_request);
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500491 blk_queue_max_hw_sectors(brd->brd_queue, 1024);
Nick Piggin9db55792008-02-08 04:19:49 -0800492 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
493
Nick Pigginb7c33572010-05-26 15:41:14 +0200494 brd->brd_queue->limits.discard_granularity = PAGE_SIZE;
495 brd->brd_queue->limits.max_discard_sectors = UINT_MAX;
496 brd->brd_queue->limits.discard_zeroes_data = 1;
497 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, brd->brd_queue);
498
Laurent Vivierd7853d12008-04-30 00:55:06 -0700499 disk = brd->brd_disk = alloc_disk(1 << part_shift);
Nick Piggin9db55792008-02-08 04:19:49 -0800500 if (!disk)
501 goto out_free_queue;
502 disk->major = RAMDISK_MAJOR;
Laurent Vivierd7853d12008-04-30 00:55:06 -0700503 disk->first_minor = i << part_shift;
Nick Piggin9db55792008-02-08 04:19:49 -0800504 disk->fops = &brd_fops;
505 disk->private_data = brd;
506 disk->queue = brd->brd_queue;
Dmitry Monakhovaeac3182014-08-18 12:49:08 +0400507 if (!part_show)
508 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
Nick Piggin9db55792008-02-08 04:19:49 -0800509 sprintf(disk->disk_name, "ram%d", i);
510 set_capacity(disk, rd_size * 2);
511
512 return brd;
513
514out_free_queue:
515 blk_cleanup_queue(brd->brd_queue);
516out_free_dev:
517 kfree(brd);
518out:
519 return NULL;
520}
521
522static void brd_free(struct brd_device *brd)
523{
524 put_disk(brd->brd_disk);
525 blk_cleanup_queue(brd->brd_queue);
526 brd_free_pages(brd);
527 kfree(brd);
528}
529
530static struct brd_device *brd_init_one(int i)
531{
532 struct brd_device *brd;
533
534 list_for_each_entry(brd, &brd_devices, brd_list) {
535 if (brd->brd_number == i)
536 goto out;
537 }
538
539 brd = brd_alloc(i);
540 if (brd) {
541 add_disk(brd->brd_disk);
542 list_add_tail(&brd->brd_list, &brd_devices);
543 }
544out:
545 return brd;
546}
547
548static void brd_del_one(struct brd_device *brd)
549{
550 list_del(&brd->brd_list);
551 del_gendisk(brd->brd_disk);
552 brd_free(brd);
553}
554
555static struct kobject *brd_probe(dev_t dev, int *part, void *data)
556{
557 struct brd_device *brd;
558 struct kobject *kobj;
559
560 mutex_lock(&brd_devices_mutex);
Namhyung Kimaf465662011-05-26 21:06:50 +0200561 brd = brd_init_one(MINOR(dev) >> part_shift);
Mikulas Patockaa207f592013-10-14 12:13:24 -0400562 kobj = brd ? get_disk(brd->brd_disk) : NULL;
Nick Piggin9db55792008-02-08 04:19:49 -0800563 mutex_unlock(&brd_devices_mutex);
564
565 *part = 0;
566 return kobj;
567}
568
569static int __init brd_init(void)
570{
571 int i, nr;
572 unsigned long range;
573 struct brd_device *brd, *next;
574
575 /*
576 * brd module now has a feature to instantiate underlying device
577 * structure on-demand, provided that there is an access dev node.
578 * However, this will not work well with user space tool that doesn't
579 * know about such "feature". In order to not break any existing
580 * tool, we do the following:
581 *
582 * (1) if rd_nr is specified, create that many upfront, and this
583 * also becomes a hard limit.
Namhyung Kim13868b72011-05-26 21:06:50 +0200584 * (2) if rd_nr is not specified, create CONFIG_BLK_DEV_RAM_COUNT
585 * (default 16) rd device on module load, user can further
586 * extend brd device by create dev node themselves and have
587 * kernel automatically instantiate actual device on-demand.
Nick Piggin9db55792008-02-08 04:19:49 -0800588 */
Laurent Vivierd7853d12008-04-30 00:55:06 -0700589
590 part_shift = 0;
Namhyung Kim8892cba2011-05-26 21:06:50 +0200591 if (max_part > 0) {
Laurent Vivierd7853d12008-04-30 00:55:06 -0700592 part_shift = fls(max_part);
593
Namhyung Kim8892cba2011-05-26 21:06:50 +0200594 /*
595 * Adjust max_part according to part_shift as it is exported
596 * to user space so that user can decide correct minor number
597 * if [s]he want to create more devices.
598 *
599 * Note that -1 is required because partition 0 is reserved
600 * for the whole disk.
601 */
602 max_part = (1UL << part_shift) - 1;
603 }
604
Namhyung Kim315980c2011-05-26 21:06:50 +0200605 if ((1UL << part_shift) > DISK_MAX_PARTS)
606 return -EINVAL;
607
Laurent Vivierd7853d12008-04-30 00:55:06 -0700608 if (rd_nr > 1UL << (MINORBITS - part_shift))
Nick Piggin9db55792008-02-08 04:19:49 -0800609 return -EINVAL;
610
611 if (rd_nr) {
612 nr = rd_nr;
Namhyung Kimaf465662011-05-26 21:06:50 +0200613 range = rd_nr << part_shift;
Nick Piggin9db55792008-02-08 04:19:49 -0800614 } else {
615 nr = CONFIG_BLK_DEV_RAM_COUNT;
Namhyung Kimaf465662011-05-26 21:06:50 +0200616 range = 1UL << MINORBITS;
Nick Piggin9db55792008-02-08 04:19:49 -0800617 }
618
619 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
620 return -EIO;
621
622 for (i = 0; i < nr; i++) {
623 brd = brd_alloc(i);
624 if (!brd)
625 goto out_free;
626 list_add_tail(&brd->brd_list, &brd_devices);
627 }
628
629 /* point of no return */
630
631 list_for_each_entry(brd, &brd_devices, brd_list)
632 add_disk(brd->brd_disk);
633
634 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
635 THIS_MODULE, brd_probe, NULL, NULL);
636
637 printk(KERN_INFO "brd: module loaded\n");
638 return 0;
639
640out_free:
641 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
642 list_del(&brd->brd_list);
643 brd_free(brd);
644 }
Akinobu Mitac82f2962008-08-20 14:09:09 -0700645 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
Nick Piggin9db55792008-02-08 04:19:49 -0800646
Nick Piggin9db55792008-02-08 04:19:49 -0800647 return -ENOMEM;
648}
649
650static void __exit brd_exit(void)
651{
652 unsigned long range;
653 struct brd_device *brd, *next;
654
Namhyung Kimaf465662011-05-26 21:06:50 +0200655 range = rd_nr ? rd_nr << part_shift : 1UL << MINORBITS;
Nick Piggin9db55792008-02-08 04:19:49 -0800656
657 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
658 brd_del_one(brd);
659
660 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
661 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
662}
663
664module_init(brd_init);
665module_exit(brd_exit);
666