Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Ram backed block device driver.
4 *
5 * Copyright (C) 2007 Nick Piggin
6 * Copyright (C) 2007 Novell Inc.
7 *
8 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
9 * of their respective owners.
10 */
11
12#include <linux/init.h>
13#include <linux/initrd.h>
14#include <linux/module.h>
15#include <linux/moduleparam.h>
16#include <linux/major.h>
17#include <linux/blkdev.h>
18#include <linux/bio.h>
19#include <linux/highmem.h>
20#include <linux/mutex.h>
21#include <linux/radix-tree.h>
22#include <linux/fs.h>
23#include <linux/slab.h>
24#include <linux/backing-dev.h>
25
26#include <linux/uaccess.h>
27
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 */
38struct brd_device {
39 int brd_number;
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 */
56static 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 */
87static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
88{
89 pgoff_t idx;
90 struct page *page;
91 gfp_t gfp_flags;
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.
100 */
101 gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM;
102 page = alloc_page(gfp_flags);
103 if (!page)
104 return NULL;
105
106 if (radix_tree_preload(GFP_NOIO)) {
107 __free_page(page);
108 return NULL;
109 }
110
111 spin_lock(&brd->brd_lock);
112 idx = sector >> PAGE_SECTORS_SHIFT;
113 page->index = idx;
114 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
115 __free_page(page);
116 page = radix_tree_lookup(&brd->brd_pages, idx);
117 BUG_ON(!page);
118 BUG_ON(page->index != idx);
119 }
120 spin_unlock(&brd->brd_lock);
121
122 radix_tree_preload_end();
123
124 return page;
125}
126
127/*
128 * Free all backing store pages and radix tree. This must only be called when
129 * there are no other users of the device.
130 */
131#define FREE_BATCH 16
132static void brd_free_pages(struct brd_device *brd)
133{
134 unsigned long pos = 0;
135 struct page *pages[FREE_BATCH];
136 int nr_pages;
137
138 do {
139 int i;
140
141 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
142 (void **)pages, pos, FREE_BATCH);
143
144 for (i = 0; i < nr_pages; i++) {
145 void *ret;
146
147 BUG_ON(pages[i]->index < pos);
148 pos = pages[i]->index;
149 ret = radix_tree_delete(&brd->brd_pages, pos);
150 BUG_ON(!ret || ret != pages[i]);
151 __free_page(pages[i]);
152 }
153
154 pos++;
155
156 /*
157 * It takes 3.4 seconds to remove 80GiB ramdisk.
158 * So, we need cond_resched to avoid stalling the CPU.
159 */
160 cond_resched();
161
162 /*
163 * This assumes radix_tree_gang_lookup always returns as
164 * many pages as possible. If the radix-tree code changes,
165 * so will this have to.
166 */
167 } while (nr_pages == FREE_BATCH);
168}
169
170/*
171 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
172 */
173static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
174{
175 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
176 size_t copy;
177
178 copy = min_t(size_t, n, PAGE_SIZE - offset);
179 if (!brd_insert_page(brd, sector))
180 return -ENOSPC;
181 if (copy < n) {
182 sector += copy >> SECTOR_SHIFT;
183 if (!brd_insert_page(brd, sector))
184 return -ENOSPC;
185 }
186 return 0;
187}
188
189/*
190 * Copy n bytes from src to the brd starting at sector. Does not sleep.
191 */
192static void copy_to_brd(struct brd_device *brd, const void *src,
193 sector_t sector, size_t n)
194{
195 struct page *page;
196 void *dst;
197 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
198 size_t copy;
199
200 copy = min_t(size_t, n, PAGE_SIZE - offset);
201 page = brd_lookup_page(brd, sector);
202 BUG_ON(!page);
203
204 dst = kmap_atomic(page);
205 memcpy(dst + offset, src, copy);
206 kunmap_atomic(dst);
207
208 if (copy < n) {
209 src += copy;
210 sector += copy >> SECTOR_SHIFT;
211 copy = n - copy;
212 page = brd_lookup_page(brd, sector);
213 BUG_ON(!page);
214
215 dst = kmap_atomic(page);
216 memcpy(dst, src, copy);
217 kunmap_atomic(dst);
218 }
219}
220
221/*
222 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
223 */
224static void copy_from_brd(void *dst, struct brd_device *brd,
225 sector_t sector, size_t n)
226{
227 struct page *page;
228 void *src;
229 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
230 size_t copy;
231
232 copy = min_t(size_t, n, PAGE_SIZE - offset);
233 page = brd_lookup_page(brd, sector);
234 if (page) {
235 src = kmap_atomic(page);
236 memcpy(dst, src + offset, copy);
237 kunmap_atomic(src);
238 } else
239 memset(dst, 0, copy);
240
241 if (copy < n) {
242 dst += copy;
243 sector += copy >> SECTOR_SHIFT;
244 copy = n - copy;
245 page = brd_lookup_page(brd, sector);
246 if (page) {
247 src = kmap_atomic(page);
248 memcpy(dst, src, copy);
249 kunmap_atomic(src);
250 } else
251 memset(dst, 0, copy);
252 }
253}
254
255/*
256 * Process a single bvec of a bio.
257 */
258static int brd_do_bvec(struct brd_device *brd, struct page *page,
259 unsigned int len, unsigned int off, unsigned int op,
260 sector_t sector)
261{
262 void *mem;
263 int err = 0;
264
265 if (op_is_write(op)) {
266 err = copy_to_brd_setup(brd, sector, len);
267 if (err)
268 goto out;
269 }
270
271 mem = kmap_atomic(page);
272 if (!op_is_write(op)) {
273 copy_from_brd(mem + off, brd, sector, len);
274 flush_dcache_page(page);
275 } else {
276 flush_dcache_page(page);
277 copy_to_brd(brd, mem + off, sector, len);
278 }
279 kunmap_atomic(mem);
280
281out:
282 return err;
283}
284
285static blk_qc_t brd_submit_bio(struct bio *bio)
286{
287 struct brd_device *brd = bio->bi_disk->private_data;
288 struct bio_vec bvec;
289 sector_t sector;
290 struct bvec_iter iter;
291
292 sector = bio->bi_iter.bi_sector;
293 if (bio_end_sector(bio) > get_capacity(bio->bi_disk))
294 goto io_error;
295
296 bio_for_each_segment(bvec, bio, iter) {
297 unsigned int len = bvec.bv_len;
298 int err;
299
300 /* Don't support un-aligned buffer */
301 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
302 (len & (SECTOR_SIZE - 1)));
303
304 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
305 bio_op(bio), sector);
306 if (err)
307 goto io_error;
308 sector += len >> SECTOR_SHIFT;
309 }
310
311 bio_endio(bio);
312 return BLK_QC_T_NONE;
313io_error:
314 bio_io_error(bio);
315 return BLK_QC_T_NONE;
316}
317
318static int brd_rw_page(struct block_device *bdev, sector_t sector,
319 struct page *page, unsigned int op)
320{
321 struct brd_device *brd = bdev->bd_disk->private_data;
322 int err;
323
324 if (PageTransHuge(page))
325 return -ENOTSUPP;
326 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
327 page_endio(page, op_is_write(op), err);
328 return err;
329}
330
331static const struct block_device_operations brd_fops = {
332 .owner = THIS_MODULE,
333 .submit_bio = brd_submit_bio,
334 .rw_page = brd_rw_page,
335};
336
337/*
338 * And now the modules code and kernel interface.
339 */
340static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
341module_param(rd_nr, int, 0444);
342MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
343
344unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
345module_param(rd_size, ulong, 0444);
346MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
347
348static int max_part = 1;
349module_param(max_part, int, 0444);
350MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
351
352MODULE_LICENSE("GPL");
353MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
354MODULE_ALIAS("rd");
355
356#ifndef MODULE
357/* Legacy boot options - nonmodular */
358static int __init ramdisk_size(char *str)
359{
360 rd_size = simple_strtol(str, NULL, 0);
361 return 1;
362}
363__setup("ramdisk_size=", ramdisk_size);
364#endif
365
366/*
367 * The device scheme is derived from loop.c. Keep them in synch where possible
368 * (should share code eventually).
369 */
370static LIST_HEAD(brd_devices);
371static DEFINE_MUTEX(brd_devices_mutex);
372
373static struct brd_device *brd_alloc(int i)
374{
375 struct brd_device *brd;
376 struct gendisk *disk;
377
378 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
379 if (!brd)
380 goto out;
381 brd->brd_number = i;
382 spin_lock_init(&brd->brd_lock);
383 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
384
385 brd->brd_queue = blk_alloc_queue(NUMA_NO_NODE);
386 if (!brd->brd_queue)
387 goto out_free_dev;
388
389 /* This is so fdisk will align partitions on 4k, because of
390 * direct_access API needing 4k alignment, returning a PFN
391 * (This is only a problem on very small devices <= 4M,
392 * otherwise fdisk will align on 1M. Regardless this call
393 * is harmless)
394 */
395 blk_queue_physical_block_size(brd->brd_queue, PAGE_SIZE);
396 disk = brd->brd_disk = alloc_disk(max_part);
397 if (!disk)
398 goto out_free_queue;
399 disk->major = RAMDISK_MAJOR;
400 disk->first_minor = i * max_part;
401 disk->fops = &brd_fops;
402 disk->private_data = brd;
403 disk->flags = GENHD_FL_EXT_DEVT;
404 sprintf(disk->disk_name, "ram%d", i);
405 set_capacity(disk, rd_size * 2);
406 brd->brd_queue->backing_dev_info->capabilities |= BDI_CAP_SYNCHRONOUS_IO;
407
408 /* Tell the block layer that this is not a rotational device */
409 blk_queue_flag_set(QUEUE_FLAG_NONROT, brd->brd_queue);
410 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, brd->brd_queue);
411
412 return brd;
413
414out_free_queue:
415 blk_cleanup_queue(brd->brd_queue);
416out_free_dev:
417 kfree(brd);
418out:
419 return NULL;
420}
421
422static void brd_free(struct brd_device *brd)
423{
424 put_disk(brd->brd_disk);
425 blk_cleanup_queue(brd->brd_queue);
426 brd_free_pages(brd);
427 kfree(brd);
428}
429
430static struct brd_device *brd_init_one(int i, bool *new)
431{
432 struct brd_device *brd;
433
434 *new = false;
435 list_for_each_entry(brd, &brd_devices, brd_list) {
436 if (brd->brd_number == i)
437 goto out;
438 }
439
440 brd = brd_alloc(i);
441 if (brd) {
442 brd->brd_disk->queue = brd->brd_queue;
443 add_disk(brd->brd_disk);
444 list_add_tail(&brd->brd_list, &brd_devices);
445 }
446 *new = true;
447out:
448 return brd;
449}
450
451static void brd_del_one(struct brd_device *brd)
452{
453 list_del(&brd->brd_list);
454 del_gendisk(brd->brd_disk);
455 brd_free(brd);
456}
457
458static struct kobject *brd_probe(dev_t dev, int *part, void *data)
459{
460 struct brd_device *brd;
461 struct kobject *kobj;
462 bool new;
463
464 mutex_lock(&brd_devices_mutex);
465 brd = brd_init_one(MINOR(dev) / max_part, &new);
466 kobj = brd ? get_disk_and_module(brd->brd_disk) : NULL;
467 mutex_unlock(&brd_devices_mutex);
468
469 if (new)
470 *part = 0;
471
472 return kobj;
473}
474
475static inline void brd_check_and_reset_par(void)
476{
477 if (unlikely(!max_part))
478 max_part = 1;
479
480 /*
481 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
482 * otherwise, it is possiable to get same dev_t when adding partitions.
483 */
484 if ((1U << MINORBITS) % max_part != 0)
485 max_part = 1UL << fls(max_part);
486
487 if (max_part > DISK_MAX_PARTS) {
488 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
489 DISK_MAX_PARTS, DISK_MAX_PARTS);
490 max_part = DISK_MAX_PARTS;
491 }
492}
493
494static int __init brd_init(void)
495{
496 struct brd_device *brd, *next;
497 int i;
498
499 /*
500 * brd module now has a feature to instantiate underlying device
501 * structure on-demand, provided that there is an access dev node.
502 *
503 * (1) if rd_nr is specified, create that many upfront. else
504 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
505 * (2) User can further extend brd devices by create dev node themselves
506 * and have kernel automatically instantiate actual device
507 * on-demand. Example:
508 * mknod /path/devnod_name b 1 X # 1 is the rd major
509 * fdisk -l /path/devnod_name
510 * If (X / max_part) was not already created it will be created
511 * dynamically.
512 */
513
514 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
515 return -EIO;
516
517 brd_check_and_reset_par();
518
519 for (i = 0; i < rd_nr; i++) {
520 brd = brd_alloc(i);
521 if (!brd)
522 goto out_free;
523 list_add_tail(&brd->brd_list, &brd_devices);
524 }
525
526 /* point of no return */
527
528 list_for_each_entry(brd, &brd_devices, brd_list) {
529 /*
530 * associate with queue just before adding disk for
531 * avoiding to mess up failure path
532 */
533 brd->brd_disk->queue = brd->brd_queue;
534 add_disk(brd->brd_disk);
535 }
536
537 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS,
538 THIS_MODULE, brd_probe, NULL, NULL);
539
540 pr_info("brd: module loaded\n");
541 return 0;
542
543out_free:
544 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
545 list_del(&brd->brd_list);
546 brd_free(brd);
547 }
548 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
549
550 pr_info("brd: module NOT loaded !!!\n");
551 return -ENOMEM;
552}
553
554static void __exit brd_exit(void)
555{
556 struct brd_device *brd, *next;
557
558 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
559 brd_del_one(brd);
560
561 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), 1UL << MINORBITS);
562 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
563
564 pr_info("brd: module unloaded\n");
565}
566
567module_init(brd_init);
568module_exit(brd_exit);
569
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Ram backed block device driver.
4 *
5 * Copyright (C) 2007 Nick Piggin
6 * Copyright (C) 2007 Novell Inc.
7 *
8 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
9 * of their respective owners.
10 */
11
12#include <linux/init.h>
13#include <linux/initrd.h>
14#include <linux/module.h>
15#include <linux/moduleparam.h>
16#include <linux/major.h>
17#include <linux/blkdev.h>
18#include <linux/bio.h>
19#include <linux/highmem.h>
20#include <linux/mutex.h>
21#include <linux/pagemap.h>
22#include <linux/radix-tree.h>
23#include <linux/fs.h>
24#include <linux/slab.h>
25#include <linux/backing-dev.h>
26#include <linux/debugfs.h>
27
28#include <linux/uaccess.h>
29
30/*
31 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
32 * the pages containing the block device's contents. A brd page's ->index is
33 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
34 * with, the kernel's pagecache or buffer cache (which sit above our block
35 * device).
36 */
37struct brd_device {
38 int brd_number;
39 struct gendisk *brd_disk;
40 struct list_head brd_list;
41
42 /*
43 * Backing store of pages and lock to protect it. This is the contents
44 * of the block device.
45 */
46 spinlock_t brd_lock;
47 struct radix_tree_root brd_pages;
48 u64 brd_nr_pages;
49};
50
51/*
52 * Look up and return a brd's page for a given sector.
53 */
54static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
55{
56 pgoff_t idx;
57 struct page *page;
58
59 /*
60 * The page lifetime is protected by the fact that we have opened the
61 * device node -- brd pages will never be deleted under us, so we
62 * don't need any further locking or refcounting.
63 *
64 * This is strictly true for the radix-tree nodes as well (ie. we
65 * don't actually need the rcu_read_lock()), however that is not a
66 * documented feature of the radix-tree API so it is better to be
67 * safe here (we don't have total exclusion from radix tree updates
68 * here, only deletes).
69 */
70 rcu_read_lock();
71 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
72 page = radix_tree_lookup(&brd->brd_pages, idx);
73 rcu_read_unlock();
74
75 BUG_ON(page && page->index != idx);
76
77 return page;
78}
79
80/*
81 * Look up and return a brd's page for a given sector.
82 * If one does not exist, allocate an empty page, and insert that. Then
83 * return it.
84 */
85static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
86{
87 pgoff_t idx;
88 struct page *page;
89 gfp_t gfp_flags;
90
91 page = brd_lookup_page(brd, sector);
92 if (page)
93 return page;
94
95 /*
96 * Must use NOIO because we don't want to recurse back into the
97 * block or filesystem layers from page reclaim.
98 */
99 gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM;
100 page = alloc_page(gfp_flags);
101 if (!page)
102 return NULL;
103
104 if (radix_tree_preload(GFP_NOIO)) {
105 __free_page(page);
106 return NULL;
107 }
108
109 spin_lock(&brd->brd_lock);
110 idx = sector >> PAGE_SECTORS_SHIFT;
111 page->index = idx;
112 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
113 __free_page(page);
114 page = radix_tree_lookup(&brd->brd_pages, idx);
115 BUG_ON(!page);
116 BUG_ON(page->index != idx);
117 } else {
118 brd->brd_nr_pages++;
119 }
120 spin_unlock(&brd->brd_lock);
121
122 radix_tree_preload_end();
123
124 return page;
125}
126
127/*
128 * Free all backing store pages and radix tree. This must only be called when
129 * there are no other users of the device.
130 */
131#define FREE_BATCH 16
132static void brd_free_pages(struct brd_device *brd)
133{
134 unsigned long pos = 0;
135 struct page *pages[FREE_BATCH];
136 int nr_pages;
137
138 do {
139 int i;
140
141 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
142 (void **)pages, pos, FREE_BATCH);
143
144 for (i = 0; i < nr_pages; i++) {
145 void *ret;
146
147 BUG_ON(pages[i]->index < pos);
148 pos = pages[i]->index;
149 ret = radix_tree_delete(&brd->brd_pages, pos);
150 BUG_ON(!ret || ret != pages[i]);
151 __free_page(pages[i]);
152 }
153
154 pos++;
155
156 /*
157 * It takes 3.4 seconds to remove 80GiB ramdisk.
158 * So, we need cond_resched to avoid stalling the CPU.
159 */
160 cond_resched();
161
162 /*
163 * This assumes radix_tree_gang_lookup always returns as
164 * many pages as possible. If the radix-tree code changes,
165 * so will this have to.
166 */
167 } while (nr_pages == FREE_BATCH);
168}
169
170/*
171 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
172 */
173static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
174{
175 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
176 size_t copy;
177
178 copy = min_t(size_t, n, PAGE_SIZE - offset);
179 if (!brd_insert_page(brd, sector))
180 return -ENOSPC;
181 if (copy < n) {
182 sector += copy >> SECTOR_SHIFT;
183 if (!brd_insert_page(brd, sector))
184 return -ENOSPC;
185 }
186 return 0;
187}
188
189/*
190 * Copy n bytes from src to the brd starting at sector. Does not sleep.
191 */
192static void copy_to_brd(struct brd_device *brd, const void *src,
193 sector_t sector, size_t n)
194{
195 struct page *page;
196 void *dst;
197 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
198 size_t copy;
199
200 copy = min_t(size_t, n, PAGE_SIZE - offset);
201 page = brd_lookup_page(brd, sector);
202 BUG_ON(!page);
203
204 dst = kmap_atomic(page);
205 memcpy(dst + offset, src, copy);
206 kunmap_atomic(dst);
207
208 if (copy < n) {
209 src += copy;
210 sector += copy >> SECTOR_SHIFT;
211 copy = n - copy;
212 page = brd_lookup_page(brd, sector);
213 BUG_ON(!page);
214
215 dst = kmap_atomic(page);
216 memcpy(dst, src, copy);
217 kunmap_atomic(dst);
218 }
219}
220
221/*
222 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
223 */
224static void copy_from_brd(void *dst, struct brd_device *brd,
225 sector_t sector, size_t n)
226{
227 struct page *page;
228 void *src;
229 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
230 size_t copy;
231
232 copy = min_t(size_t, n, PAGE_SIZE - offset);
233 page = brd_lookup_page(brd, sector);
234 if (page) {
235 src = kmap_atomic(page);
236 memcpy(dst, src + offset, copy);
237 kunmap_atomic(src);
238 } else
239 memset(dst, 0, copy);
240
241 if (copy < n) {
242 dst += copy;
243 sector += copy >> SECTOR_SHIFT;
244 copy = n - copy;
245 page = brd_lookup_page(brd, sector);
246 if (page) {
247 src = kmap_atomic(page);
248 memcpy(dst, src, copy);
249 kunmap_atomic(src);
250 } else
251 memset(dst, 0, copy);
252 }
253}
254
255/*
256 * Process a single bvec of a bio.
257 */
258static int brd_do_bvec(struct brd_device *brd, struct page *page,
259 unsigned int len, unsigned int off, enum req_op op,
260 sector_t sector)
261{
262 void *mem;
263 int err = 0;
264
265 if (op_is_write(op)) {
266 err = copy_to_brd_setup(brd, sector, len);
267 if (err)
268 goto out;
269 }
270
271 mem = kmap_atomic(page);
272 if (!op_is_write(op)) {
273 copy_from_brd(mem + off, brd, sector, len);
274 flush_dcache_page(page);
275 } else {
276 flush_dcache_page(page);
277 copy_to_brd(brd, mem + off, sector, len);
278 }
279 kunmap_atomic(mem);
280
281out:
282 return err;
283}
284
285static void brd_submit_bio(struct bio *bio)
286{
287 struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
288 sector_t sector = bio->bi_iter.bi_sector;
289 struct bio_vec bvec;
290 struct bvec_iter iter;
291
292 bio_for_each_segment(bvec, bio, iter) {
293 unsigned int len = bvec.bv_len;
294 int err;
295
296 /* Don't support un-aligned buffer */
297 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
298 (len & (SECTOR_SIZE - 1)));
299
300 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
301 bio_op(bio), sector);
302 if (err) {
303 bio_io_error(bio);
304 return;
305 }
306 sector += len >> SECTOR_SHIFT;
307 }
308
309 bio_endio(bio);
310}
311
312static int brd_rw_page(struct block_device *bdev, sector_t sector,
313 struct page *page, enum req_op op)
314{
315 struct brd_device *brd = bdev->bd_disk->private_data;
316 int err;
317
318 if (PageTransHuge(page))
319 return -ENOTSUPP;
320 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
321 page_endio(page, op_is_write(op), err);
322 return err;
323}
324
325static const struct block_device_operations brd_fops = {
326 .owner = THIS_MODULE,
327 .submit_bio = brd_submit_bio,
328 .rw_page = brd_rw_page,
329};
330
331/*
332 * And now the modules code and kernel interface.
333 */
334static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
335module_param(rd_nr, int, 0444);
336MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
337
338unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
339module_param(rd_size, ulong, 0444);
340MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
341
342static int max_part = 1;
343module_param(max_part, int, 0444);
344MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
345
346MODULE_LICENSE("GPL");
347MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
348MODULE_ALIAS("rd");
349
350#ifndef MODULE
351/* Legacy boot options - nonmodular */
352static int __init ramdisk_size(char *str)
353{
354 rd_size = simple_strtol(str, NULL, 0);
355 return 1;
356}
357__setup("ramdisk_size=", ramdisk_size);
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 */
364static LIST_HEAD(brd_devices);
365static struct dentry *brd_debugfs_dir;
366
367static int brd_alloc(int i)
368{
369 struct brd_device *brd;
370 struct gendisk *disk;
371 char buf[DISK_NAME_LEN];
372 int err = -ENOMEM;
373
374 list_for_each_entry(brd, &brd_devices, brd_list)
375 if (brd->brd_number == i)
376 return -EEXIST;
377 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
378 if (!brd)
379 return -ENOMEM;
380 brd->brd_number = i;
381 list_add_tail(&brd->brd_list, &brd_devices);
382
383 spin_lock_init(&brd->brd_lock);
384 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
385
386 snprintf(buf, DISK_NAME_LEN, "ram%d", i);
387 if (!IS_ERR_OR_NULL(brd_debugfs_dir))
388 debugfs_create_u64(buf, 0444, brd_debugfs_dir,
389 &brd->brd_nr_pages);
390
391 disk = brd->brd_disk = blk_alloc_disk(NUMA_NO_NODE);
392 if (!disk)
393 goto out_free_dev;
394
395 disk->major = RAMDISK_MAJOR;
396 disk->first_minor = i * max_part;
397 disk->minors = max_part;
398 disk->fops = &brd_fops;
399 disk->private_data = brd;
400 strscpy(disk->disk_name, buf, DISK_NAME_LEN);
401 set_capacity(disk, rd_size * 2);
402
403 /*
404 * This is so fdisk will align partitions on 4k, because of
405 * direct_access API needing 4k alignment, returning a PFN
406 * (This is only a problem on very small devices <= 4M,
407 * otherwise fdisk will align on 1M. Regardless this call
408 * is harmless)
409 */
410 blk_queue_physical_block_size(disk->queue, PAGE_SIZE);
411
412 /* Tell the block layer that this is not a rotational device */
413 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
414 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
415 err = add_disk(disk);
416 if (err)
417 goto out_cleanup_disk;
418
419 return 0;
420
421out_cleanup_disk:
422 put_disk(disk);
423out_free_dev:
424 list_del(&brd->brd_list);
425 kfree(brd);
426 return err;
427}
428
429static void brd_probe(dev_t dev)
430{
431 brd_alloc(MINOR(dev) / max_part);
432}
433
434static void brd_cleanup(void)
435{
436 struct brd_device *brd, *next;
437
438 debugfs_remove_recursive(brd_debugfs_dir);
439
440 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
441 del_gendisk(brd->brd_disk);
442 put_disk(brd->brd_disk);
443 brd_free_pages(brd);
444 list_del(&brd->brd_list);
445 kfree(brd);
446 }
447}
448
449static inline void brd_check_and_reset_par(void)
450{
451 if (unlikely(!max_part))
452 max_part = 1;
453
454 /*
455 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
456 * otherwise, it is possiable to get same dev_t when adding partitions.
457 */
458 if ((1U << MINORBITS) % max_part != 0)
459 max_part = 1UL << fls(max_part);
460
461 if (max_part > DISK_MAX_PARTS) {
462 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
463 DISK_MAX_PARTS, DISK_MAX_PARTS);
464 max_part = DISK_MAX_PARTS;
465 }
466}
467
468static int __init brd_init(void)
469{
470 int err, i;
471
472 brd_check_and_reset_par();
473
474 brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
475
476 for (i = 0; i < rd_nr; i++) {
477 err = brd_alloc(i);
478 if (err)
479 goto out_free;
480 }
481
482 /*
483 * brd module now has a feature to instantiate underlying device
484 * structure on-demand, provided that there is an access dev node.
485 *
486 * (1) if rd_nr is specified, create that many upfront. else
487 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
488 * (2) User can further extend brd devices by create dev node themselves
489 * and have kernel automatically instantiate actual device
490 * on-demand. Example:
491 * mknod /path/devnod_name b 1 X # 1 is the rd major
492 * fdisk -l /path/devnod_name
493 * If (X / max_part) was not already created it will be created
494 * dynamically.
495 */
496
497 if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe)) {
498 err = -EIO;
499 goto out_free;
500 }
501
502 pr_info("brd: module loaded\n");
503 return 0;
504
505out_free:
506 brd_cleanup();
507
508 pr_info("brd: module NOT loaded !!!\n");
509 return err;
510}
511
512static void __exit brd_exit(void)
513{
514
515 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
516 brd_cleanup();
517
518 pr_info("brd: module unloaded\n");
519}
520
521module_init(brd_init);
522module_exit(brd_exit);
523