1 // SPDX-License-Identifier: GPL-2.0-only
3 * Ram backed block device driver.
5 * Copyright (C) 2007 Nick Piggin
6 * Copyright (C) 2007 Novell Inc.
8 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
9 * of their respective owners.
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/xarray.h>
24 #include <linux/slab.h>
25 #include <linux/backing-dev.h>
26 #include <linux/debugfs.h>
28 #include <linux/uaccess.h>
31 * Each block ramdisk device has a xarray brd_pages of pages that stores
32 * the pages containing the block device's contents.
36 struct gendisk
*brd_disk
;
37 struct list_head brd_list
;
40 * Backing store of pages. This is the contents of the block device.
42 struct xarray brd_pages
;
47 * Look up and return a brd's page for a given sector.
49 static struct page
*brd_lookup_page(struct brd_device
*brd
, sector_t sector
)
51 return xa_load(&brd
->brd_pages
, sector
>> PAGE_SECTORS_SHIFT
);
55 * Insert a new page for a given sector, if one does not already exist.
57 static int brd_insert_page(struct brd_device
*brd
, sector_t sector
, gfp_t gfp
)
59 pgoff_t idx
= sector
>> PAGE_SECTORS_SHIFT
;
63 page
= brd_lookup_page(brd
, sector
);
67 page
= alloc_page(gfp
| __GFP_ZERO
| __GFP_HIGHMEM
);
71 xa_lock(&brd
->brd_pages
);
72 ret
= __xa_insert(&brd
->brd_pages
, idx
, page
, gfp
);
75 xa_unlock(&brd
->brd_pages
);
86 * Free all backing store pages and xarray. This must only be called when
87 * there are no other users of the device.
89 static void brd_free_pages(struct brd_device
*brd
)
94 xa_for_each(&brd
->brd_pages
, idx
, page
) {
99 xa_destroy(&brd
->brd_pages
);
103 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
105 static int copy_to_brd_setup(struct brd_device
*brd
, sector_t sector
, size_t n
,
108 unsigned int offset
= (sector
& (PAGE_SECTORS
-1)) << SECTOR_SHIFT
;
112 copy
= min_t(size_t, n
, PAGE_SIZE
- offset
);
113 ret
= brd_insert_page(brd
, sector
, gfp
);
117 sector
+= copy
>> SECTOR_SHIFT
;
118 ret
= brd_insert_page(brd
, sector
, gfp
);
124 * Copy n bytes from src to the brd starting at sector. Does not sleep.
126 static void copy_to_brd(struct brd_device
*brd
, const void *src
,
127 sector_t sector
, size_t n
)
131 unsigned int offset
= (sector
& (PAGE_SECTORS
-1)) << SECTOR_SHIFT
;
134 copy
= min_t(size_t, n
, PAGE_SIZE
- offset
);
135 page
= brd_lookup_page(brd
, sector
);
138 dst
= kmap_atomic(page
);
139 memcpy(dst
+ offset
, src
, copy
);
144 sector
+= copy
>> SECTOR_SHIFT
;
146 page
= brd_lookup_page(brd
, sector
);
149 dst
= kmap_atomic(page
);
150 memcpy(dst
, src
, copy
);
156 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
158 static void copy_from_brd(void *dst
, struct brd_device
*brd
,
159 sector_t sector
, size_t n
)
163 unsigned int offset
= (sector
& (PAGE_SECTORS
-1)) << SECTOR_SHIFT
;
166 copy
= min_t(size_t, n
, PAGE_SIZE
- offset
);
167 page
= brd_lookup_page(brd
, sector
);
169 src
= kmap_atomic(page
);
170 memcpy(dst
, src
+ offset
, copy
);
173 memset(dst
, 0, copy
);
177 sector
+= copy
>> SECTOR_SHIFT
;
179 page
= brd_lookup_page(brd
, sector
);
181 src
= kmap_atomic(page
);
182 memcpy(dst
, src
, copy
);
185 memset(dst
, 0, copy
);
190 * Process a single bvec of a bio.
192 static int brd_do_bvec(struct brd_device
*brd
, struct page
*page
,
193 unsigned int len
, unsigned int off
, blk_opf_t opf
,
199 if (op_is_write(opf
)) {
201 * Must use NOIO because we don't want to recurse back into the
202 * block or filesystem layers from page reclaim.
204 gfp_t gfp
= opf
& REQ_NOWAIT
? GFP_NOWAIT
: GFP_NOIO
;
206 err
= copy_to_brd_setup(brd
, sector
, len
, gfp
);
211 mem
= kmap_atomic(page
);
212 if (!op_is_write(opf
)) {
213 copy_from_brd(mem
+ off
, brd
, sector
, len
);
214 flush_dcache_page(page
);
216 flush_dcache_page(page
);
217 copy_to_brd(brd
, mem
+ off
, sector
, len
);
225 static void brd_do_discard(struct brd_device
*brd
, sector_t sector
, u32 size
)
227 sector_t aligned_sector
= (sector
+ PAGE_SECTORS
) & ~PAGE_SECTORS
;
230 size
-= (aligned_sector
- sector
) * SECTOR_SIZE
;
231 xa_lock(&brd
->brd_pages
);
232 while (size
>= PAGE_SIZE
&& aligned_sector
< rd_size
* 2) {
233 page
= __xa_erase(&brd
->brd_pages
, aligned_sector
>> PAGE_SECTORS_SHIFT
);
238 aligned_sector
+= PAGE_SECTORS
;
241 xa_unlock(&brd
->brd_pages
);
244 static void brd_submit_bio(struct bio
*bio
)
246 struct brd_device
*brd
= bio
->bi_bdev
->bd_disk
->private_data
;
247 sector_t sector
= bio
->bi_iter
.bi_sector
;
249 struct bvec_iter iter
;
251 if (unlikely(op_is_discard(bio
->bi_opf
))) {
252 brd_do_discard(brd
, sector
, bio
->bi_iter
.bi_size
);
257 bio_for_each_segment(bvec
, bio
, iter
) {
258 unsigned int len
= bvec
.bv_len
;
261 /* Don't support un-aligned buffer */
262 WARN_ON_ONCE((bvec
.bv_offset
& (SECTOR_SIZE
- 1)) ||
263 (len
& (SECTOR_SIZE
- 1)));
265 err
= brd_do_bvec(brd
, bvec
.bv_page
, len
, bvec
.bv_offset
,
266 bio
->bi_opf
, sector
);
268 if (err
== -ENOMEM
&& bio
->bi_opf
& REQ_NOWAIT
) {
269 bio_wouldblock_error(bio
);
275 sector
+= len
>> SECTOR_SHIFT
;
281 static const struct block_device_operations brd_fops
= {
282 .owner
= THIS_MODULE
,
283 .submit_bio
= brd_submit_bio
,
287 * And now the modules code and kernel interface.
289 static int rd_nr
= CONFIG_BLK_DEV_RAM_COUNT
;
290 module_param(rd_nr
, int, 0444);
291 MODULE_PARM_DESC(rd_nr
, "Maximum number of brd devices");
293 unsigned long rd_size
= CONFIG_BLK_DEV_RAM_SIZE
;
294 module_param(rd_size
, ulong
, 0444);
295 MODULE_PARM_DESC(rd_size
, "Size of each RAM disk in kbytes.");
297 static int max_part
= 1;
298 module_param(max_part
, int, 0444);
299 MODULE_PARM_DESC(max_part
, "Num Minors to reserve between devices");
301 MODULE_DESCRIPTION("Ram backed block device driver");
302 MODULE_LICENSE("GPL");
303 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR
);
307 /* Legacy boot options - nonmodular */
308 static int __init
ramdisk_size(char *str
)
310 rd_size
= simple_strtol(str
, NULL
, 0);
313 __setup("ramdisk_size=", ramdisk_size
);
317 * The device scheme is derived from loop.c. Keep them in synch where possible
318 * (should share code eventually).
320 static LIST_HEAD(brd_devices
);
321 static DEFINE_MUTEX(brd_devices_mutex
);
322 static struct dentry
*brd_debugfs_dir
;
324 static struct brd_device
*brd_find_or_alloc_device(int i
)
326 struct brd_device
*brd
;
328 mutex_lock(&brd_devices_mutex
);
329 list_for_each_entry(brd
, &brd_devices
, brd_list
) {
330 if (brd
->brd_number
== i
) {
331 mutex_unlock(&brd_devices_mutex
);
332 return ERR_PTR(-EEXIST
);
336 brd
= kzalloc(sizeof(*brd
), GFP_KERNEL
);
338 mutex_unlock(&brd_devices_mutex
);
339 return ERR_PTR(-ENOMEM
);
342 list_add_tail(&brd
->brd_list
, &brd_devices
);
343 mutex_unlock(&brd_devices_mutex
);
347 static void brd_free_device(struct brd_device
*brd
)
349 mutex_lock(&brd_devices_mutex
);
350 list_del(&brd
->brd_list
);
351 mutex_unlock(&brd_devices_mutex
);
355 static int brd_alloc(int i
)
357 struct brd_device
*brd
;
358 struct gendisk
*disk
;
359 char buf
[DISK_NAME_LEN
];
361 struct queue_limits lim
= {
363 * This is so fdisk will align partitions on 4k, because of
364 * direct_access API needing 4k alignment, returning a PFN
365 * (This is only a problem on very small devices <= 4M,
366 * otherwise fdisk will align on 1M. Regardless this call
369 .physical_block_size
= PAGE_SIZE
,
370 .max_hw_discard_sectors
= UINT_MAX
,
371 .max_discard_segments
= 1,
372 .discard_granularity
= PAGE_SIZE
,
373 .features
= BLK_FEAT_SYNCHRONOUS
|
377 brd
= brd_find_or_alloc_device(i
);
381 xa_init(&brd
->brd_pages
);
383 snprintf(buf
, DISK_NAME_LEN
, "ram%d", i
);
384 if (!IS_ERR_OR_NULL(brd_debugfs_dir
))
385 debugfs_create_u64(buf
, 0444, brd_debugfs_dir
,
388 disk
= brd
->brd_disk
= blk_alloc_disk(&lim
, NUMA_NO_NODE
);
393 disk
->major
= RAMDISK_MAJOR
;
394 disk
->first_minor
= i
* max_part
;
395 disk
->minors
= max_part
;
396 disk
->fops
= &brd_fops
;
397 disk
->private_data
= brd
;
398 strscpy(disk
->disk_name
, buf
, DISK_NAME_LEN
);
399 set_capacity(disk
, rd_size
* 2);
401 err
= add_disk(disk
);
403 goto out_cleanup_disk
;
410 brd_free_device(brd
);
414 static void brd_probe(dev_t dev
)
416 brd_alloc(MINOR(dev
) / max_part
);
419 static void brd_cleanup(void)
421 struct brd_device
*brd
, *next
;
423 debugfs_remove_recursive(brd_debugfs_dir
);
425 list_for_each_entry_safe(brd
, next
, &brd_devices
, brd_list
) {
426 del_gendisk(brd
->brd_disk
);
427 put_disk(brd
->brd_disk
);
429 brd_free_device(brd
);
433 static inline void brd_check_and_reset_par(void)
435 if (unlikely(!max_part
))
439 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
440 * otherwise, it is possiable to get same dev_t when adding partitions.
442 if ((1U << MINORBITS
) % max_part
!= 0)
443 max_part
= 1UL << fls(max_part
);
445 if (max_part
> DISK_MAX_PARTS
) {
446 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
447 DISK_MAX_PARTS
, DISK_MAX_PARTS
);
448 max_part
= DISK_MAX_PARTS
;
452 static int __init
brd_init(void)
457 * brd module now has a feature to instantiate underlying device
458 * structure on-demand, provided that there is an access dev node.
460 * (1) if rd_nr is specified, create that many upfront. else
461 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
462 * (2) User can further extend brd devices by create dev node themselves
463 * and have kernel automatically instantiate actual device
464 * on-demand. Example:
465 * mknod /path/devnod_name b 1 X # 1 is the rd major
466 * fdisk -l /path/devnod_name
467 * If (X / max_part) was not already created it will be created
471 brd_check_and_reset_par();
473 brd_debugfs_dir
= debugfs_create_dir("ramdisk_pages", NULL
);
475 if (__register_blkdev(RAMDISK_MAJOR
, "ramdisk", brd_probe
)) {
480 for (i
= 0; i
< rd_nr
; i
++)
483 pr_info("brd: module loaded\n");
489 pr_info("brd: module NOT loaded !!!\n");
493 static void __exit
brd_exit(void)
496 unregister_blkdev(RAMDISK_MAJOR
, "ramdisk");
499 pr_info("brd: module unloaded\n");
502 module_init(brd_init
);
503 module_exit(brd_exit
);