2 * Ram backed block device driver.
4 * Copyright (C) 2007 Nick Piggin
5 * Copyright (C) 2007 Novell Inc.
7 * Parts derived from drivers/block/rd.c, and drivers/block/loop.c, copyright
8 * of their respective owners.
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/moduleparam.h>
14 #include <linux/major.h>
15 #include <linux/blkdev.h>
16 #include <linux/bio.h>
17 #include <linux/highmem.h>
18 #include <linux/gfp.h>
19 #include <linux/radix-tree.h>
20 #include <linux/buffer_head.h> /* invalidate_bh_lrus() */
22 #include <asm/uaccess.h>
24 #define SECTOR_SHIFT 9
25 #define PAGE_SECTORS_SHIFT (PAGE_SHIFT - SECTOR_SHIFT)
26 #define PAGE_SECTORS (1 << PAGE_SECTORS_SHIFT)
29 * Each block ramdisk device has a radix_tree brd_pages of pages that stores
30 * the pages containing the block device's contents. A brd page's ->index is
31 * its offset in PAGE_SIZE units. This is similar to, but in no way connected
32 * with, the kernel's pagecache or buffer cache (which sit above our block
40 unsigned brd_blocksize;
42 struct request_queue *brd_queue;
43 struct gendisk *brd_disk;
44 struct list_head brd_list;
47 * Backing store of pages and lock to protect it. This is the contents
48 * of the block device.
51 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 * Cannot support XIP and highmem, because our ->direct_access
103 * routine for XIP must return memory that is always addressable.
104 * If XIP was reworked to use pfns and kmap throughout, this
105 * restriction might be able to be lifted.
107 gfp_flags = GFP_NOIO | __GFP_ZERO;
108 #ifndef CONFIG_BLK_DEV_XIP
109 gfp_flags |= __GFP_HIGHMEM;
111 page = alloc_page(gfp_flags);
115 if (radix_tree_preload(GFP_NOIO)) {
120 spin_lock(&brd->brd_lock);
121 idx = sector >> PAGE_SECTORS_SHIFT;
122 if (radix_tree_insert(&brd->brd_pages, idx, page)) {
124 page = radix_tree_lookup(&brd->brd_pages, idx);
126 BUG_ON(page->index != idx);
129 spin_unlock(&brd->brd_lock);
131 radix_tree_preload_end();
137 * Free all backing store pages and radix tree. This must only be called when
138 * there are no other users of the device.
140 #define FREE_BATCH 16
141 static void brd_free_pages(struct brd_device *brd)
143 unsigned long pos = 0;
144 struct page *pages[FREE_BATCH];
150 nr_pages = radix_tree_gang_lookup(&brd->brd_pages,
151 (void **)pages, pos, FREE_BATCH);
153 for (i = 0; i < nr_pages; i++) {
156 BUG_ON(pages[i]->index < pos);
157 pos = pages[i]->index;
158 ret = radix_tree_delete(&brd->brd_pages, pos);
159 BUG_ON(!ret || ret != pages[i]);
160 __free_page(pages[i]);
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, KM_USER1);
208 memcpy(dst + offset, src, copy);
209 kunmap_atomic(dst, KM_USER1);
213 sector += copy >> SECTOR_SHIFT;
215 page = brd_lookup_page(brd, sector);
218 dst = kmap_atomic(page, KM_USER1);
219 memcpy(dst, src, copy);
220 kunmap_atomic(dst, KM_USER1);
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, KM_USER1);
239 memcpy(dst, src + offset, copy);
240 kunmap_atomic(src, KM_USER1);
242 memset(dst, 0, copy);
246 sector += copy >> SECTOR_SHIFT;
248 page = brd_lookup_page(brd, sector);
250 src = kmap_atomic(page, KM_USER1);
251 memcpy(dst, src, copy);
252 kunmap_atomic(src, KM_USER1);
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, int rw,
269 err = copy_to_brd_setup(brd, sector, len);
274 mem = kmap_atomic(page, KM_USER0);
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);
282 kunmap_atomic(mem, KM_USER0);
288 static int brd_make_request(struct request_queue *q, struct bio *bio)
290 struct block_device *bdev = bio->bi_bdev;
291 struct brd_device *brd = bdev->bd_disk->private_data;
293 struct bio_vec *bvec;
298 sector = bio->bi_sector;
299 if (sector + (bio->bi_size >> SECTOR_SHIFT) >
300 get_capacity(bdev->bd_disk))
307 bio_for_each_segment(bvec, bio, i) {
308 unsigned int len = bvec->bv_len;
309 err = brd_do_bvec(brd, bvec->bv_page, len,
310 bvec->bv_offset, rw, sector);
313 sector += len >> SECTOR_SHIFT;
322 #ifdef CONFIG_BLK_DEV_XIP
323 static int brd_direct_access (struct block_device *bdev, sector_t sector,
324 void **kaddr, unsigned long *pfn)
326 struct brd_device *brd = bdev->bd_disk->private_data;
331 if (sector & (PAGE_SECTORS-1))
333 if (sector + PAGE_SECTORS > get_capacity(bdev->bd_disk))
335 page = brd_insert_page(brd, sector);
338 *kaddr = page_address(page);
339 *pfn = page_to_pfn(page);
345 static int brd_ioctl(struct block_device *bdev, fmode_t mode,
346 unsigned int cmd, unsigned long arg)
349 struct brd_device *brd = bdev->bd_disk->private_data;
351 if (cmd != BLKFLSBUF)
355 * ram device BLKFLSBUF has special semantics, we want to actually
356 * release and destroy the ramdisk data.
358 mutex_lock(&bdev->bd_mutex);
360 if (bdev->bd_openers <= 1) {
362 * Invalidate the cache first, so it isn't written
363 * back to the device.
365 * Another thread might instantiate more buffercache here,
366 * but there is not much we can do to close that race.
368 invalidate_bh_lrus();
369 truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
373 mutex_unlock(&bdev->bd_mutex);
378 static struct block_device_operations brd_fops = {
379 .owner = THIS_MODULE,
380 .locked_ioctl = brd_ioctl,
381 #ifdef CONFIG_BLK_DEV_XIP
382 .direct_access = brd_direct_access,
387 * And now the modules code and kernel interface.
390 int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
392 static int part_shift;
393 module_param(rd_nr, int, 0);
394 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
395 module_param(rd_size, int, 0);
396 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
397 module_param(max_part, int, 0);
398 MODULE_PARM_DESC(max_part, "Maximum number of partitions per RAM disk");
399 MODULE_LICENSE("GPL");
400 MODULE_ALIAS_BLOCKDEV_MAJOR(RAMDISK_MAJOR);
404 /* Legacy boot options - nonmodular */
405 static int __init ramdisk_size(char *str)
407 rd_size = simple_strtol(str, NULL, 0);
410 static int __init ramdisk_size2(char *str)
412 return ramdisk_size(str);
414 __setup("ramdisk=", ramdisk_size);
415 __setup("ramdisk_size=", ramdisk_size2);
419 * The device scheme is derived from loop.c. Keep them in synch where possible
420 * (should share code eventually).
422 static LIST_HEAD(brd_devices);
423 static DEFINE_MUTEX(brd_devices_mutex);
425 static struct brd_device *brd_alloc(int i)
427 struct brd_device *brd;
428 struct gendisk *disk;
430 brd = kzalloc(sizeof(*brd), GFP_KERNEL);
434 spin_lock_init(&brd->brd_lock);
435 INIT_RADIX_TREE(&brd->brd_pages, GFP_ATOMIC);
437 brd->brd_queue = blk_alloc_queue(GFP_KERNEL);
440 blk_queue_make_request(brd->brd_queue, brd_make_request);
441 blk_queue_ordered(brd->brd_queue, QUEUE_ORDERED_TAG, NULL);
442 blk_queue_max_sectors(brd->brd_queue, 1024);
443 blk_queue_bounce_limit(brd->brd_queue, BLK_BOUNCE_ANY);
445 disk = brd->brd_disk = alloc_disk(1 << part_shift);
448 disk->major = RAMDISK_MAJOR;
449 disk->first_minor = i << part_shift;
450 disk->fops = &brd_fops;
451 disk->private_data = brd;
452 disk->queue = brd->brd_queue;
453 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
454 sprintf(disk->disk_name, "ram%d", i);
455 set_capacity(disk, rd_size * 2);
460 blk_cleanup_queue(brd->brd_queue);
467 static void brd_free(struct brd_device *brd)
469 put_disk(brd->brd_disk);
470 blk_cleanup_queue(brd->brd_queue);
475 static struct brd_device *brd_init_one(int i)
477 struct brd_device *brd;
479 list_for_each_entry(brd, &brd_devices, brd_list) {
480 if (brd->brd_number == i)
486 add_disk(brd->brd_disk);
487 list_add_tail(&brd->brd_list, &brd_devices);
493 static void brd_del_one(struct brd_device *brd)
495 list_del(&brd->brd_list);
496 del_gendisk(brd->brd_disk);
500 static struct kobject *brd_probe(dev_t dev, int *part, void *data)
502 struct brd_device *brd;
503 struct kobject *kobj;
505 mutex_lock(&brd_devices_mutex);
506 brd = brd_init_one(dev & MINORMASK);
507 kobj = brd ? get_disk(brd->brd_disk) : ERR_PTR(-ENOMEM);
508 mutex_unlock(&brd_devices_mutex);
514 static int __init brd_init(void)
518 struct brd_device *brd, *next;
521 * brd module now has a feature to instantiate underlying device
522 * structure on-demand, provided that there is an access dev node.
523 * However, this will not work well with user space tool that doesn't
524 * know about such "feature". In order to not break any existing
525 * tool, we do the following:
527 * (1) if rd_nr is specified, create that many upfront, and this
528 * also becomes a hard limit.
529 * (2) if rd_nr is not specified, create 1 rd device on module
530 * load, user can further extend brd device by create dev node
531 * themselves and have kernel automatically instantiate actual
537 part_shift = fls(max_part);
539 if (rd_nr > 1UL << (MINORBITS - part_shift))
546 nr = CONFIG_BLK_DEV_RAM_COUNT;
547 range = 1UL << (MINORBITS - part_shift);
550 if (register_blkdev(RAMDISK_MAJOR, "ramdisk"))
553 for (i = 0; i < nr; i++) {
557 list_add_tail(&brd->brd_list, &brd_devices);
560 /* point of no return */
562 list_for_each_entry(brd, &brd_devices, brd_list)
563 add_disk(brd->brd_disk);
565 blk_register_region(MKDEV(RAMDISK_MAJOR, 0), range,
566 THIS_MODULE, brd_probe, NULL, NULL);
568 printk(KERN_INFO "brd: module loaded\n");
572 list_for_each_entry_safe(brd, next, &brd_devices, brd_list) {
573 list_del(&brd->brd_list);
576 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
581 static void __exit brd_exit(void)
584 struct brd_device *brd, *next;
586 range = rd_nr ? rd_nr : 1UL << (MINORBITS - part_shift);
588 list_for_each_entry_safe(brd, next, &brd_devices, brd_list)
591 blk_unregister_region(MKDEV(RAMDISK_MAJOR, 0), range);
592 unregister_blkdev(RAMDISK_MAJOR, "ramdisk");
595 module_init(brd_init);
596 module_exit(brd_exit);