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 * Insert a new page for a given sector, if one does not already exist.
83 static int brd_insert_page(struct brd_device *brd, sector_t sector, gfp_t gfp)
89 page = brd_lookup_page(brd, sector);
93 page = alloc_page(gfp | __GFP_ZERO | __GFP_HIGHMEM);
97 if (radix_tree_maybe_preload(gfp)) {
102 spin_lock(&brd->brd_lock);
103 idx = sector >> PAGE_SECTORS_SHIFT;
105 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
107 page = radix_tree_lookup(&brd->brd_pages, idx);
110 else if (page->index != idx)
115 spin_unlock(&brd->brd_lock);
117 radix_tree_preload_end();
122 * Free all backing store pages and radix tree. This must only be called when
123 * there are no other users of the device.
125 #define FREE_BATCH 16
126 static void brd_free_pages(struct brd_device *brd)
128 unsigned long pos = 0;
129 struct page *pages[FREE_BATCH];
135 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
136 (void **)pages, pos, FREE_BATCH);
138 for (i = 0; i < nr_pages; i++) {
141 BUG_ON(pages[i]->index < pos);
142 pos = pages[i]->index;
143 ret = radix_tree_delete(&brd->brd_pages, pos);
144 BUG_ON(!ret || ret != pages[i]);
145 __free_page(pages[i]);
151 * It takes 3.4 seconds to remove 80GiB ramdisk.
152 * So, we need cond_resched to avoid stalling the CPU.
157 * This assumes radix_tree_gang_lookup always returns as
158 * many pages as possible. If the radix-tree code changes,
159 * so will this have to.
161 } while (nr_pages == FREE_BATCH);
165 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
167 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n,
170 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
174 copy = min_t(size_t, n, PAGE_SIZE - offset);
175 ret = brd_insert_page(brd, sector, gfp);
179 sector += copy >> SECTOR_SHIFT;
180 ret = brd_insert_page(brd, sector, gfp);
186 * Copy n bytes from src to the brd starting at sector. Does not sleep.
188 static void copy_to_brd(struct brd_device *brd, const void *src,
189 sector_t sector, size_t n)
193 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
196 copy = min_t(size_t, n, PAGE_SIZE - offset);
197 page = brd_lookup_page(brd, sector);
200 dst = kmap_atomic(page);
201 memcpy(dst + offset, src, copy);
206 sector += copy >> SECTOR_SHIFT;
208 page = brd_lookup_page(brd, sector);
211 dst = kmap_atomic(page);
212 memcpy(dst, src, copy);
218 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
220 static void copy_from_brd(void *dst, struct brd_device *brd,
221 sector_t sector, size_t n)
225 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
228 copy = min_t(size_t, n, PAGE_SIZE - offset);
229 page = brd_lookup_page(brd, sector);
231 src = kmap_atomic(page);
232 memcpy(dst, src + offset, copy);
235 memset(dst, 0, copy);
239 sector += copy >> SECTOR_SHIFT;
241 page = brd_lookup_page(brd, sector);
243 src = kmap_atomic(page);
244 memcpy(dst, src, copy);
247 memset(dst, 0, copy);
252 * Process a single bvec of a bio.
254 static int brd_do_bvec(struct brd_device *brd, struct page *page,
255 unsigned int len, unsigned int off, blk_opf_t opf,
261 if (op_is_write(opf)) {
263 * Must use NOIO because we don't want to recurse back into the
264 * block or filesystem layers from page reclaim.
266 gfp_t gfp = opf & REQ_NOWAIT ? GFP_NOWAIT : GFP_NOIO;
268 err = copy_to_brd_setup(brd, sector, len, gfp);
273 mem = kmap_atomic(page);
274 if (!op_is_write(opf)) {
275 copy_from_brd(mem + off, brd, sector, len);
276 flush_dcache_page(page);
278 flush_dcache_page(page);
279 copy_to_brd(brd, mem + off, sector, len);
287 static void brd_submit_bio(struct bio *bio)
289 struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
290 sector_t sector = bio->bi_iter.bi_sector;
292 struct bvec_iter iter;
294 bio_for_each_segment(bvec, bio, iter) {
295 unsigned int len = bvec.bv_len;
298 /* Don't support un-aligned buffer */
299 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
300 (len & (SECTOR_SIZE - 1)));
302 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
303 bio->bi_opf, sector);
305 if (err == -ENOMEM && bio->bi_opf & REQ_NOWAIT) {
306 bio_wouldblock_error(bio);
312 sector += len >> SECTOR_SHIFT;
318 static int brd_rw_page(struct block_device *bdev, sector_t sector,
319 struct page *page, enum req_op op)
321 struct brd_device *brd = bdev->bd_disk->private_data;
324 if (PageTransHuge(page))
326 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
327 page_endio(page, op_is_write(op), err);
331 static const struct block_device_operations brd_fops = {
332 .owner = THIS_MODULE,
333 .submit_bio = brd_submit_bio,
334 .rw_page = brd_rw_page,
338 * And now the modules code and kernel interface.
340 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
341 module_param(rd_nr, int, 0444);
342 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
344 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
345 module_param(rd_size, ulong, 0444);
346 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
348 static int max_part = 1;
349 module_param(max_part, int, 0444);
350 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
352 MODULE_LICENSE("GPL");
353 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
357 /* Legacy boot options - nonmodular */
358 static int __init ramdisk_size(char *str)
360 rd_size = simple_strtol(str, NULL, 0);
363 __setup("ramdisk_size=", ramdisk_size);
367 * The device scheme is derived from loop.c. Keep them in synch where possible
368 * (should share code eventually).
370 static LIST_HEAD(brd_devices);
371 static struct dentry *brd_debugfs_dir;
373 static int brd_alloc(int i)
375 struct brd_device *brd;
376 struct gendisk *disk;
377 char buf[DISK_NAME_LEN];
380 list_for_each_entry(brd, &brd_devices, brd_list)
381 if (brd->brd_number == i)
383 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
387 list_add_tail(&brd->brd_list, &brd_devices);
389 spin_lock_init(&brd->brd_lock);
390 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
392 snprintf(buf, DISK_NAME_LEN, "ram%d", i);
393 if (!IS_ERR_OR_NULL(brd_debugfs_dir))
394 debugfs_create_u64(buf, 0444, brd_debugfs_dir,
397 disk = brd->brd_disk = blk_alloc_disk(NUMA_NO_NODE);
401 disk->major = RAMDISK_MAJOR;
402 disk->first_minor = i * max_part;
403 disk->minors = max_part;
404 disk->fops = &brd_fops;
405 disk->private_data = brd;
406 strscpy(disk->disk_name, buf, DISK_NAME_LEN);
407 set_capacity(disk, rd_size * 2);
410 * This is so fdisk will align partitions on 4k, because of
411 * direct_access API needing 4k alignment, returning a PFN
412 * (This is only a problem on very small devices <= 4M,
413 * otherwise fdisk will align on 1M. Regardless this call
416 blk_queue_physical_block_size(disk->queue, PAGE_SIZE);
418 /* Tell the block layer that this is not a rotational device */
419 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
420 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
421 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, disk->queue);
422 err = add_disk(disk);
424 goto out_cleanup_disk;
431 list_del(&brd->brd_list);
436 static void brd_probe(dev_t dev)
438 brd_alloc(MINOR(dev) / max_part);
441 static void brd_cleanup(void)
443 struct brd_device *brd, *next;
445 debugfs_remove_recursive(brd_debugfs_dir);
447 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
448 del_gendisk(brd->brd_disk);
449 put_disk(brd->brd_disk);
451 list_del(&brd->brd_list);
456 static inline void brd_check_and_reset_par(void)
458 if (unlikely(!max_part))
462 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
463 * otherwise, it is possiable to get same dev_t when adding partitions.
465 if ((1U << MINORBITS) % max_part != 0)
466 max_part = 1UL << fls(max_part);
468 if (max_part > DISK_MAX_PARTS) {
469 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
470 DISK_MAX_PARTS, DISK_MAX_PARTS);
471 max_part = DISK_MAX_PARTS;
475 static int __init brd_init(void)
479 brd_check_and_reset_par();
481 brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
483 for (i = 0; i < rd_nr; i++) {
490 * brd module now has a feature to instantiate underlying device
491 * structure on-demand, provided that there is an access dev node.
493 * (1) if rd_nr is specified, create that many upfront. else
494 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
495 * (2) User can further extend brd devices by create dev node themselves
496 * and have kernel automatically instantiate actual device
497 * on-demand. Example:
498 * mknod /path/devnod_name b 1 X # 1 is the rd major
499 * fdisk -l /path/devnod_name
500 * If (X / max_part) was not already created it will be created
504 if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe)) {
509 pr_info("brd: module loaded\n");
515 pr_info("brd: module NOT loaded !!!\n");
519 static void __exit brd_exit(void)
522 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
525 pr_info("brd: module unloaded\n");
528 module_init(brd_init);
529 module_exit(brd_exit);