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);
236 aligned_sector += PAGE_SECTORS;
239 xa_unlock(&brd->brd_pages);
242 static void brd_submit_bio(struct bio *bio)
244 struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
245 sector_t sector = bio->bi_iter.bi_sector;
247 struct bvec_iter iter;
249 if (unlikely(op_is_discard(bio->bi_opf))) {
250 brd_do_discard(brd, sector, bio->bi_iter.bi_size);
255 bio_for_each_segment(bvec, bio, iter) {
256 unsigned int len = bvec.bv_len;
259 /* Don't support un-aligned buffer */
260 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
261 (len & (SECTOR_SIZE - 1)));
263 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
264 bio->bi_opf, sector);
266 if (err == -ENOMEM && bio->bi_opf & REQ_NOWAIT) {
267 bio_wouldblock_error(bio);
273 sector += len >> SECTOR_SHIFT;
279 static const struct block_device_operations brd_fops = {
280 .owner = THIS_MODULE,
281 .submit_bio = brd_submit_bio,
285 * And now the modules code and kernel interface.
287 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
288 module_param(rd_nr, int, 0444);
289 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
291 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
292 module_param(rd_size, ulong, 0444);
293 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
295 static int max_part = 1;
296 module_param(max_part, int, 0444);
297 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
299 MODULE_LICENSE("GPL");
300 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
304 /* Legacy boot options - nonmodular */
305 static int __init ramdisk_size(char *str)
307 rd_size = simple_strtol(str, NULL, 0);
310 __setup("ramdisk_size=", ramdisk_size);
314 * The device scheme is derived from loop.c. Keep them in synch where possible
315 * (should share code eventually).
317 static LIST_HEAD(brd_devices);
318 static struct dentry *brd_debugfs_dir;
320 static int brd_alloc(int i)
322 struct brd_device *brd;
323 struct gendisk *disk;
324 char buf[DISK_NAME_LEN];
326 struct queue_limits lim = {
328 * This is so fdisk will align partitions on 4k, because of
329 * direct_access API needing 4k alignment, returning a PFN
330 * (This is only a problem on very small devices <= 4M,
331 * otherwise fdisk will align on 1M. Regardless this call
334 .physical_block_size = PAGE_SIZE,
335 .max_hw_discard_sectors = UINT_MAX,
336 .max_discard_segments = 1,
337 .discard_granularity = PAGE_SIZE,
340 list_for_each_entry(brd, &brd_devices, brd_list)
341 if (brd->brd_number == i)
343 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
347 list_add_tail(&brd->brd_list, &brd_devices);
349 xa_init(&brd->brd_pages);
351 snprintf(buf, DISK_NAME_LEN, "ram%d", i);
352 if (!IS_ERR_OR_NULL(brd_debugfs_dir))
353 debugfs_create_u64(buf, 0444, brd_debugfs_dir,
356 disk = brd->brd_disk = blk_alloc_disk(&lim, NUMA_NO_NODE);
361 disk->major = RAMDISK_MAJOR;
362 disk->first_minor = i * max_part;
363 disk->minors = max_part;
364 disk->fops = &brd_fops;
365 disk->private_data = brd;
366 strscpy(disk->disk_name, buf, DISK_NAME_LEN);
367 set_capacity(disk, rd_size * 2);
369 /* Tell the block layer that this is not a rotational device */
370 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
371 blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, disk->queue);
372 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, disk->queue);
373 err = add_disk(disk);
375 goto out_cleanup_disk;
382 list_del(&brd->brd_list);
387 static void brd_probe(dev_t dev)
389 brd_alloc(MINOR(dev) / max_part);
392 static void brd_cleanup(void)
394 struct brd_device *brd, *next;
396 debugfs_remove_recursive(brd_debugfs_dir);
398 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
399 del_gendisk(brd->brd_disk);
400 put_disk(brd->brd_disk);
402 list_del(&brd->brd_list);
407 static inline void brd_check_and_reset_par(void)
409 if (unlikely(!max_part))
413 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
414 * otherwise, it is possiable to get same dev_t when adding partitions.
416 if ((1U << MINORBITS) % max_part != 0)
417 max_part = 1UL << fls(max_part);
419 if (max_part > DISK_MAX_PARTS) {
420 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
421 DISK_MAX_PARTS, DISK_MAX_PARTS);
422 max_part = DISK_MAX_PARTS;
426 static int __init brd_init(void)
430 brd_check_and_reset_par();
432 brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
434 for (i = 0; i < rd_nr; i++) {
441 * brd module now has a feature to instantiate underlying device
442 * structure on-demand, provided that there is an access dev node.
444 * (1) if rd_nr is specified, create that many upfront. else
445 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
446 * (2) User can further extend brd devices by create dev node themselves
447 * and have kernel automatically instantiate actual device
448 * on-demand. Example:
449 * mknod /path/devnod_name b 1 X # 1 is the rd major
450 * fdisk -l /path/devnod_name
451 * If (X / max_part) was not already created it will be created
455 if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe)) {
460 pr_info("brd: module loaded\n");
466 pr_info("brd: module NOT loaded !!!\n");
470 static void __exit brd_exit(void)
473 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
476 pr_info("brd: module unloaded\n");
479 module_init(brd_init);
480 module_exit(brd_exit);