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>
30 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
31 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
34 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
35 * the pages containing the block device's contents. A brd page's ->index is
36 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
37 * with, the kernel's pagecache or buffer cache (which sit above our block
42 struct gendisk *brd_disk;
43 struct list_head brd_list;
46 * Backing store of pages and lock to protect it. This is the contents
47 * of the block device.
50 struct radix_tree_root brd_pages;
55 * Look up and return a brd's page for a given sector.
57 static struct page *brd_lookup_page(struct brd_device *brd, sector_t sector)
63 * The page lifetime is protected by the fact that we have opened the
64 * device node -- brd pages will never be deleted under us, so we
65 * don't need any further locking or refcounting.
67 * This is strictly true for the radix-tree nodes as well (ie. we
68 * don't actually need the rcu_read_lock()), however that is not a
69 * documented feature of the radix-tree API so it is better to be
70 * safe here (we don't have total exclusion from radix tree updates
71 * here, only deletes).
74 idx = sector >> PAGE_SECTORS_SHIFT; /* sector to page index */
75 page = radix_tree_lookup(&brd->brd_pages, idx);
78 BUG_ON(page && page->index != idx);
84 * Look up and return a brd's page for a given sector.
85 * If one does not exist, allocate an empty page, and insert that. Then
88 static struct page *brd_insert_page(struct brd_device *brd, sector_t sector)
94 page = brd_lookup_page(brd, sector);
99 * Must use NOIO because we don't want to recurse back into the
100 * block or filesystem layers from page reclaim.
102 gfp_flags = GFP_NOIO | __GFP_ZERO | __GFP_HIGHMEM;
103 page = alloc_page(gfp_flags);
107 if (radix_tree_preload(GFP_NOIO)) {
112 spin_lock(&brd->brd_lock);
113 idx = sector >> PAGE_SECTORS_SHIFT;
115 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
117 page = radix_tree_lookup(&brd->brd_pages, idx);
119 BUG_ON(page->index != idx);
123 spin_unlock(&brd->brd_lock);
125 radix_tree_preload_end();
131 * Free all backing store pages and radix tree. This must only be called when
132 * there are no other users of the device.
134 #define FREE_BATCH 16
135 static void brd_free_pages(struct brd_device *brd)
137 unsigned long pos = 0;
138 struct page *pages[FREE_BATCH];
144 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
145 (void **)pages, pos, FREE_BATCH);
147 for (i = 0; i < nr_pages; i++) {
150 BUG_ON(pages[i]->index < pos);
151 pos = pages[i]->index;
152 ret = radix_tree_delete(&brd->brd_pages, pos);
153 BUG_ON(!ret || ret != pages[i]);
154 __free_page(pages[i]);
160 * It takes 3.4 seconds to remove 80GiB ramdisk.
161 * So, we need cond_resched to avoid stalling the CPU.
166 * This assumes radix_tree_gang_lookup always returns as
167 * many pages as possible. If the radix-tree code changes,
168 * so will this have to.
170 } while (nr_pages == FREE_BATCH);
174 * copy_to_brd_setup must be called before copy_to_brd. It may sleep.
176 static int copy_to_brd_setup(struct brd_device *brd, sector_t sector, size_t n)
178 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
181 copy = min_t(size_t, n, PAGE_SIZE - offset);
182 if (!brd_insert_page(brd, sector))
185 sector += copy >> SECTOR_SHIFT;
186 if (!brd_insert_page(brd, sector))
193 * Copy n bytes from src to the brd starting at sector. Does not sleep.
195 static void copy_to_brd(struct brd_device *brd, const void *src,
196 sector_t sector, size_t n)
200 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
203 copy = min_t(size_t, n, PAGE_SIZE - offset);
204 page = brd_lookup_page(brd, sector);
207 dst = kmap_atomic(page);
208 memcpy(dst + offset, src, copy);
213 sector += copy >> SECTOR_SHIFT;
215 page = brd_lookup_page(brd, sector);
218 dst = kmap_atomic(page);
219 memcpy(dst, src, copy);
225 * Copy n bytes to dst from the brd starting at sector. Does not sleep.
227 static void copy_from_brd(void *dst, struct brd_device *brd,
228 sector_t sector, size_t n)
232 unsigned int offset = (sector & (PAGE_SECTORS-1)) << SECTOR_SHIFT;
235 copy = min_t(size_t, n, PAGE_SIZE - offset);
236 page = brd_lookup_page(brd, sector);
238 src = kmap_atomic(page);
239 memcpy(dst, src + offset, copy);
242 memset(dst, 0, copy);
246 sector += copy >> SECTOR_SHIFT;
248 page = brd_lookup_page(brd, sector);
250 src = kmap_atomic(page);
251 memcpy(dst, src, copy);
254 memset(dst, 0, copy);
259 * Process a single bvec of a bio.
261 static int brd_do_bvec(struct brd_device *brd, struct page *page,
262 unsigned int len, unsigned int off, unsigned int op,
268 if (op_is_write(op)) {
269 err = copy_to_brd_setup(brd, sector, len);
274 mem = kmap_atomic(page);
275 if (!op_is_write(op)) {
276 copy_from_brd(mem + off, brd, sector, len);
277 flush_dcache_page(page);
279 flush_dcache_page(page);
280 copy_to_brd(brd, mem + off, sector, len);
288 static blk_qc_t brd_submit_bio(struct bio *bio)
290 struct brd_device *brd = bio->bi_bdev->bd_disk->private_data;
291 sector_t sector = bio->bi_iter.bi_sector;
293 struct bvec_iter iter;
295 bio_for_each_segment(bvec, bio, iter) {
296 unsigned int len = bvec.bv_len;
299 /* Don't support un-aligned buffer */
300 WARN_ON_ONCE((bvec.bv_offset & (SECTOR_SIZE - 1)) ||
301 (len & (SECTOR_SIZE - 1)));
303 err = brd_do_bvec(brd, bvec.bv_page, len, bvec.bv_offset,
304 bio_op(bio), sector);
307 sector += len >> SECTOR_SHIFT;
311 return BLK_QC_T_NONE;
314 return BLK_QC_T_NONE;
317 static int brd_rw_page(struct block_device *bdev, sector_t sector,
318 struct page *page, unsigned int op)
320 struct brd_device *brd = bdev->bd_disk->private_data;
323 if (PageTransHuge(page))
325 err = brd_do_bvec(brd, page, PAGE_SIZE, 0, op, sector);
326 page_endio(page, op_is_write(op), err);
330 static const struct block_device_operations brd_fops = {
331 .owner = THIS_MODULE,
332 .submit_bio = brd_submit_bio,
333 .rw_page = brd_rw_page,
337 * And now the modules code and kernel interface.
339 static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
340 module_param(rd_nr, int, 0444);
341 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
343 unsigned long rd_size = CONFIG_BLK_DEV_RAM_SIZE;
344 module_param(rd_size, ulong, 0444);
345 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
347 static int max_part = 1;
348 module_param(max_part, int, 0444);
349 MODULE_PARM_DESC(max_part, "Num Minors to reserve between devices");
351 MODULE_LICENSE("GPL");
352 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
356 /* Legacy boot options - nonmodular */
357 static int __init ramdisk_size(char *str)
359 rd_size = simple_strtol(str, NULL, 0);
362 __setup("ramdisk_size=", ramdisk_size);
366 * The device scheme is derived from loop.c. Keep them in synch where possible
367 * (should share code eventually).
369 static LIST_HEAD(brd_devices);
370 static DEFINE_MUTEX(brd_devices_mutex);
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];
379 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
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 disk->flags = GENHD_FL_EXT_DEVT;
401 strlcpy(disk->disk_name, buf, DISK_NAME_LEN);
402 set_capacity(disk, rd_size * 2);
405 * This is so fdisk will align partitions on 4k, because of
406 * direct_access API needing 4k alignment, returning a PFN
407 * (This is only a problem on very small devices <= 4M,
408 * otherwise fdisk will align on 1M. Regardless this call
411 blk_queue_physical_block_size(disk->queue, PAGE_SIZE);
413 /* Tell the block layer that this is not a rotational device */
414 blk_queue_flag_set(QUEUE_FLAG_NONROT, disk->queue);
415 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, disk->queue);
417 list_add_tail(&brd->brd_list, &brd_devices);
426 static void brd_probe(dev_t dev)
428 int i = MINOR(dev) / max_part;
429 struct brd_device *brd;
431 mutex_lock(&brd_devices_mutex);
432 list_for_each_entry(brd, &brd_devices, brd_list) {
433 if (brd->brd_number == i)
439 mutex_unlock(&brd_devices_mutex);
442 static void brd_del_one(struct brd_device *brd)
444 list_del(&brd->brd_list);
445 del_gendisk(brd->brd_disk);
446 blk_cleanup_disk(brd->brd_disk);
451 static inline void brd_check_and_reset_par(void)
453 if (unlikely(!max_part))
457 * make sure 'max_part' can be divided exactly by (1U << MINORBITS),
458 * otherwise, it is possiable to get same dev_t when adding partitions.
460 if ((1U << MINORBITS) % max_part != 0)
461 max_part = 1UL << fls(max_part);
463 if (max_part > DISK_MAX_PARTS) {
464 pr_info("brd: max_part can't be larger than %d, reset max_part = %d.\n",
465 DISK_MAX_PARTS, DISK_MAX_PARTS);
466 max_part = DISK_MAX_PARTS;
470 static int __init brd_init(void)
472 struct brd_device *brd, *next;
476 * brd module now has a feature to instantiate underlying device
477 * structure on-demand, provided that there is an access dev node.
479 * (1) if rd_nr is specified, create that many upfront. else
480 * it defaults to CONFIG_BLK_DEV_RAM_COUNT
481 * (2) User can further extend brd devices by create dev node themselves
482 * and have kernel automatically instantiate actual device
483 * on-demand. Example:
484 * mknod /path/devnod_name b 1 X # 1 is the rd major
485 * fdisk -l /path/devnod_name
486 * If (X / max_part) was not already created it will be created
490 if (__register_blkdev(RAMDISK_MAJOR, "ramdisk", brd_probe))
493 brd_check_and_reset_par();
495 brd_debugfs_dir = debugfs_create_dir("ramdisk_pages", NULL);
497 mutex_lock(&brd_devices_mutex);
498 for (i = 0; i < rd_nr; i++) {
504 mutex_unlock(&brd_devices_mutex);
506 pr_info("brd: module loaded\n");
510 debugfs_remove_recursive(brd_debugfs_dir);
512 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
514 mutex_unlock(&brd_devices_mutex);
515 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
517 pr_info("brd: module NOT loaded !!!\n");
521 static void __exit brd_exit(void)
523 struct brd_device *brd, *next;
525 debugfs_remove_recursive(brd_debugfs_dir);
527 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
530 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
532 pr_info("brd: module unloaded\n");
535 module_init(brd_init);
536 module_exit(brd_exit);