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/radix-tree.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 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
39 struct gendisk *brd_disk;
40 struct list_head brd_list;
43 * Backing store of pages and lock to protect it. This is the contents
44 * of the block device.
47 struct radix_tree_root brd_pages;
52 * Look up and return a brd's page for a given sector.
54 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
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.
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).
71 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
72 page = radix_tree_lookup(&brd->brd_pages, idx);
75 BUG_ON(page && page->index != idx);
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
85 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
91 page = brd_lookup_page(brd, sector);
96 * Must use NOIO because we don't want to recurse back into the
97 * block or filesystem layers from page reclaim.
99 gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM;
100 page = alloc_page(gfp_flags);
104 if (radix_tree_preload(GFP_NOIO)) {
109 spin_lock(&brd->brd_lock);
110 idx = sector >> PAGE_SECTORS_SHIFT;
112 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
114 page = radix_tree_lookup(&brd->brd_pages, idx);
116 BUG_ON(page->index != idx);
120 spin_unlock(&brd->brd_lock);
122 radix_tree_preload_end();
128 * Free all backing store pages and radix tree. This must only be called when
129 * there are no other users of the device.
131 #define FREE_BATCH 16
132 static void brd_free_pages(struct brd_device *brd)
134 unsigned long pos = 0;
135 struct page *pages[FREE_BATCH];
141 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
142 (void **)pages, pos, FREE_BATCH);
144 for (i = 0; i < nr_pages; i++) {
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]);
157 * It takes 3.4 seconds to remove 80GiB ramdisk.
158 * So, we need cond_resched to avoid stalling the CPU.
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.
167 } while (nr_pages == FREE_BATCH);
171 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
173 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
175 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
178 copy = min_t(size_t, n, PAGE_SIZE - offset);
179 if (!brd_insert_page(brd, sector))
182 sector += copy >> SECTOR_SHIFT;
183 if (!brd_insert_page(brd, sector))
190 * Copy n bytes from src to the brd starting at sector. Does not sleep.
192 static void copy_to_brd(struct brd_device *brd, const void *src,
193 sector_t sector, size_t n)
197 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
200 copy = min_t(size_t, n, PAGE_SIZE - offset);
201 page = brd_lookup_page(brd, sector);
204 dst = kmap_atomic(page);
205 memcpy(dst + offset, src, copy);
210 sector += copy >> SECTOR_SHIFT;
212 page = brd_lookup_page(brd, sector);
215 dst = kmap_atomic(page);
216 memcpy(dst, src, copy);
222 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
224 static void copy_from_brd(void *dst, struct brd_device *brd,
225 sector_t sector, size_t n)
229 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
232 copy = min_t(size_t, n, PAGE_SIZE - offset);
233 page = brd_lookup_page(brd, sector);
235 src = kmap_atomic(page);
236 memcpy(dst, src + offset, copy);
239 memset(dst, 0, copy);
243 sector += copy >> SECTOR_SHIFT;
245 page = brd_lookup_page(brd, sector);
247 src = kmap_atomic(page);
248 memcpy(dst, src, copy);
251 memset(dst, 0, copy);
256 * Process a single bvec of a bio.
258 static int brd_do_bvec(struct brd_device *brd, struct page *page,
259 unsigned int len, unsigned int off, unsigned int op,
265 if (op_is_write(op)) {
266 err = copy_to_brd_setup(brd, sector, len);
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);
276 flush_dcache_page(page);
277 copy_to_brd(brd, mem + off, sector, len);
285 static void brd_submit_bio(struct bio *bio)
287 struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
288 sector_t sector = bio->bi_iter.bi_sector;
290 struct bvec_iter iter;
292 bio_for_each_segment(bvec, bio, iter) {
293 unsigned int len = bvec.bv_len;
296 /* Don't support un-aligned buffer */
297 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
298 (len & (SECTOR_SIZE - 1)));
300 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
301 bio_op(bio), sector);
306 sector += len >> SECTOR_SHIFT;
312 static int brd_rw_page(struct block_device *bdev, sector_t sector,
313 struct page *page, unsigned int op)
315 struct brd_device *brd = bdev->bd_disk->private_data;
318 if (PageTransHuge(page))
320 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
321 page_endio(page, op_is_write(op), err);
325 static const struct block_device_operations brd_fops = {
326 .owner = THIS_MODULE,
327 .submit_bio = brd_submit_bio,
328 .rw_page = brd_rw_page,
332 * And now the modules code and kernel interface.
334 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
335 module_param(rd_nr, int, 0444);
336 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
338 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
339 module_param(rd_size, ulong, 0444);
340 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
342 static int max_part = 1;
343 module_param(max_part, int, 0444);
344 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
346 MODULE_LICENSE("GPL");
347 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
351 /* Legacy boot options - nonmodular */
352 static int __init ramdisk_size(char *str)
354 rd_size = simple_strtol(str, NULL, 0);
357 __setup("ramdisk_size=", ramdisk_size);
361 * The device scheme is derived from loop.c. Keep them in synch where possible
362 * (should share code eventually).
364 static LIST_HEAD(brd_devices);
365 static struct dentry *brd_debugfs_dir;
367 static int brd_alloc(int i)
369 struct brd_device *brd;
370 struct gendisk *disk;
371 char buf[DISK_NAME_LEN];
374 list_for_each_entry(brd, &brd_devices, brd_list)
375 if (brd->brd_number == i)
377 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
381 list_add_tail(&brd->brd_list, &brd_devices);
383 spin_lock_init(&brd->brd_lock);
384 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
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,
391 disk = brd->brd_disk = blk_alloc_disk(NUMA_NO_NODE);
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 strlcpy(disk->disk_name, buf, DISK_NAME_LEN);
401 set_capacity(disk, rd_size * 2);
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
410 blk_queue_physical_block_size(disk->queue, PAGE_SIZE);
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);
417 goto out_cleanup_disk;
422 blk_cleanup_disk(disk);
424 list_del(&brd->brd_list);
429 static void brd_probe(dev_t dev)
431 brd_alloc(MINOR(dev) / max_part);
434 static void brd_cleanup(void)
436 struct brd_device *brd, *next;
438 debugfs_remove_recursive(brd_debugfs_dir);
440 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
441 del_gendisk(brd->brd_disk);
442 blk_cleanup_disk(brd->brd_disk);
444 list_del(&brd->brd_list);
449 static inline void brd_check_and_reset_par(void)
451 if (unlikely(!max_part))
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.
458 if ((1U << MINORBITS) % max_part != 0)
459 max_part = 1UL << fls(max_part);
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;
468 static int __init brd_init(void)
472 brd_check_and_reset_par();
474 brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
476 for (i = 0; i < rd_nr; i++) {
483 * brd module now has a feature to instantiate underlying device
484 * structure on-demand, provided that there is an access dev node.
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
497 if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe)) {
502 pr_info("brd: module loaded\n");
508 pr_info("brd: module NOT loaded !!!\n");
512 static void __exit brd_exit(void)
515 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
518 pr_info("brd: module unloaded\n");
521 module_init(brd_init);
522 module_exit(brd_exit);